Linux Audio

Check our new training course

Linux BSP upgrade and security maintenance

Need help to get security updates for your Linux BSP?
Loading...
v6.9.4
    1// SPDX-License-Identifier: GPL-2.0
    2/* Copyright(c) 2007 - 2018 Intel Corporation. */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
    3
    4#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
    5
    6#include <linux/module.h>
    7#include <linux/types.h>
    8#include <linux/init.h>
    9#include <linux/bitops.h>
   10#include <linux/vmalloc.h>
   11#include <linux/pagemap.h>
   12#include <linux/netdevice.h>
   13#include <linux/ipv6.h>
   14#include <linux/slab.h>
   15#include <net/checksum.h>
   16#include <net/ip6_checksum.h>
   17#include <net/pkt_sched.h>
   18#include <net/pkt_cls.h>
   19#include <linux/net_tstamp.h>
   20#include <linux/mii.h>
   21#include <linux/ethtool.h>
   22#include <linux/if.h>
   23#include <linux/if_vlan.h>
   24#include <linux/pci.h>
 
   25#include <linux/delay.h>
   26#include <linux/interrupt.h>
   27#include <linux/ip.h>
   28#include <linux/tcp.h>
   29#include <linux/sctp.h>
   30#include <linux/if_ether.h>
 
   31#include <linux/prefetch.h>
   32#include <linux/bpf.h>
   33#include <linux/bpf_trace.h>
   34#include <linux/pm_runtime.h>
   35#include <linux/etherdevice.h>
   36#ifdef CONFIG_IGB_DCA
   37#include <linux/dca.h>
   38#endif
   39#include <linux/i2c.h>
   40#include "igb.h"
   41
   42enum queue_mode {
   43	QUEUE_MODE_STRICT_PRIORITY,
   44	QUEUE_MODE_STREAM_RESERVATION,
   45};
   46
   47enum tx_queue_prio {
   48	TX_QUEUE_PRIO_HIGH,
   49	TX_QUEUE_PRIO_LOW,
   50};
   51
   52char igb_driver_name[] = "igb";
 
   53static const char igb_driver_string[] =
   54				"Intel(R) Gigabit Ethernet Network Driver";
   55static const char igb_copyright[] =
   56				"Copyright (c) 2007-2014 Intel Corporation.";
   57
   58static const struct e1000_info *igb_info_tbl[] = {
   59	[board_82575] = &e1000_82575_info,
   60};
   61
   62static const struct pci_device_id igb_pci_tbl[] = {
   63	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_1GBPS) },
   64	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_SGMII) },
   65	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) },
   66	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I211_COPPER), board_82575 },
   67	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 },
   68	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 },
   69	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 },
   70	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 },
   71	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER_FLASHLESS), board_82575 },
   72	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES_FLASHLESS), board_82575 },
   73	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
   74	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
   75	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
   76	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
   77	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
   78	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
   79	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 },
   80	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
   81	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
   82	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
   83	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 },
   84	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 },
   85	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 },
   86	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 },
   87	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
   88	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
   89	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
   90	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
   91	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
   92	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
   93	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
   94	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
   95	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
   96	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
   97	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
   98	/* required last entry */
   99	{0, }
  100};
  101
  102MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
  103
 
  104static int igb_setup_all_tx_resources(struct igb_adapter *);
  105static int igb_setup_all_rx_resources(struct igb_adapter *);
  106static void igb_free_all_tx_resources(struct igb_adapter *);
  107static void igb_free_all_rx_resources(struct igb_adapter *);
  108static void igb_setup_mrqc(struct igb_adapter *);
  109static int igb_probe(struct pci_dev *, const struct pci_device_id *);
  110static void igb_remove(struct pci_dev *pdev);
  111static void igb_init_queue_configuration(struct igb_adapter *adapter);
  112static int igb_sw_init(struct igb_adapter *);
  113int igb_open(struct net_device *);
  114int igb_close(struct net_device *);
  115static void igb_configure(struct igb_adapter *);
  116static void igb_configure_tx(struct igb_adapter *);
  117static void igb_configure_rx(struct igb_adapter *);
  118static void igb_clean_all_tx_rings(struct igb_adapter *);
  119static void igb_clean_all_rx_rings(struct igb_adapter *);
  120static void igb_clean_tx_ring(struct igb_ring *);
  121static void igb_clean_rx_ring(struct igb_ring *);
  122static void igb_set_rx_mode(struct net_device *);
  123static void igb_update_phy_info(struct timer_list *);
  124static void igb_watchdog(struct timer_list *);
  125static void igb_watchdog_task(struct work_struct *);
  126static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *);
  127static void igb_get_stats64(struct net_device *dev,
  128			    struct rtnl_link_stats64 *stats);
  129static int igb_change_mtu(struct net_device *, int);
  130static int igb_set_mac(struct net_device *, void *);
  131static void igb_set_uta(struct igb_adapter *adapter, bool set);
  132static irqreturn_t igb_intr(int irq, void *);
  133static irqreturn_t igb_intr_msi(int irq, void *);
  134static irqreturn_t igb_msix_other(int irq, void *);
  135static irqreturn_t igb_msix_ring(int irq, void *);
  136#ifdef CONFIG_IGB_DCA
  137static void igb_update_dca(struct igb_q_vector *);
  138static void igb_setup_dca(struct igb_adapter *);
  139#endif /* CONFIG_IGB_DCA */
  140static int igb_poll(struct napi_struct *, int);
  141static bool igb_clean_tx_irq(struct igb_q_vector *, int);
  142static int igb_clean_rx_irq(struct igb_q_vector *, int);
  143static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
  144static void igb_tx_timeout(struct net_device *, unsigned int txqueue);
  145static void igb_reset_task(struct work_struct *);
  146static void igb_vlan_mode(struct net_device *netdev,
  147			  netdev_features_t features);
  148static int igb_vlan_rx_add_vid(struct net_device *, __be16, u16);
  149static int igb_vlan_rx_kill_vid(struct net_device *, __be16, u16);
  150static void igb_restore_vlan(struct igb_adapter *);
  151static void igb_rar_set_index(struct igb_adapter *, u32);
  152static void igb_ping_all_vfs(struct igb_adapter *);
  153static void igb_msg_task(struct igb_adapter *);
  154static void igb_vmm_control(struct igb_adapter *);
  155static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
  156static void igb_flush_mac_table(struct igb_adapter *);
  157static int igb_available_rars(struct igb_adapter *, u8);
  158static void igb_set_default_mac_filter(struct igb_adapter *);
  159static int igb_uc_sync(struct net_device *, const unsigned char *);
  160static int igb_uc_unsync(struct net_device *, const unsigned char *);
  161static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
  162static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
  163static int igb_ndo_set_vf_vlan(struct net_device *netdev,
  164			       int vf, u16 vlan, u8 qos, __be16 vlan_proto);
  165static int igb_ndo_set_vf_bw(struct net_device *, int, int, int);
  166static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
  167				   bool setting);
  168static int igb_ndo_set_vf_trust(struct net_device *netdev, int vf,
  169				bool setting);
  170static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
  171				 struct ifla_vf_info *ivi);
  172static void igb_check_vf_rate_limit(struct igb_adapter *);
  173static void igb_nfc_filter_exit(struct igb_adapter *adapter);
  174static void igb_nfc_filter_restore(struct igb_adapter *adapter);
  175
  176#ifdef CONFIG_PCI_IOV
  177static int igb_vf_configure(struct igb_adapter *adapter, int vf);
  178static int igb_disable_sriov(struct pci_dev *dev, bool reinit);
  179#endif
  180
 
 
  181static int igb_suspend(struct device *);
 
  182static int igb_resume(struct device *);
 
  183static int igb_runtime_suspend(struct device *dev);
  184static int igb_runtime_resume(struct device *dev);
  185static int igb_runtime_idle(struct device *dev);
  186#ifdef CONFIG_PM
  187static const struct dev_pm_ops igb_pm_ops = {
  188	SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume)
  189	SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume,
  190			igb_runtime_idle)
  191};
  192#endif
  193static void igb_shutdown(struct pci_dev *);
  194static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs);
  195#ifdef CONFIG_IGB_DCA
  196static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
  197static struct notifier_block dca_notifier = {
  198	.notifier_call	= igb_notify_dca,
  199	.next		= NULL,
  200	.priority	= 0
  201};
  202#endif
 
 
 
 
  203#ifdef CONFIG_PCI_IOV
  204static unsigned int max_vfs;
  205module_param(max_vfs, uint, 0444);
  206MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate per physical function");
 
  207#endif /* CONFIG_PCI_IOV */
  208
  209static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
  210		     pci_channel_state_t);
  211static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
  212static void igb_io_resume(struct pci_dev *);
  213
  214static const struct pci_error_handlers igb_err_handler = {
  215	.error_detected = igb_io_error_detected,
  216	.slot_reset = igb_io_slot_reset,
  217	.resume = igb_io_resume,
  218};
  219
  220static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
  221
  222static struct pci_driver igb_driver = {
  223	.name     = igb_driver_name,
  224	.id_table = igb_pci_tbl,
  225	.probe    = igb_probe,
  226	.remove   = igb_remove,
  227#ifdef CONFIG_PM
  228	.driver.pm = &igb_pm_ops,
  229#endif
  230	.shutdown = igb_shutdown,
  231	.sriov_configure = igb_pci_sriov_configure,
  232	.err_handler = &igb_err_handler
  233};
  234
  235MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
  236MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
  237MODULE_LICENSE("GPL v2");
 
  238
  239#define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
  240static int debug = -1;
  241module_param(debug, int, 0);
  242MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  243
  244struct igb_reg_info {
  245	u32 ofs;
  246	char *name;
  247};
  248
  249static const struct igb_reg_info igb_reg_info_tbl[] = {
  250
  251	/* General Registers */
  252	{E1000_CTRL, "CTRL"},
  253	{E1000_STATUS, "STATUS"},
  254	{E1000_CTRL_EXT, "CTRL_EXT"},
  255
  256	/* Interrupt Registers */
  257	{E1000_ICR, "ICR"},
  258
  259	/* RX Registers */
  260	{E1000_RCTL, "RCTL"},
  261	{E1000_RDLEN(0), "RDLEN"},
  262	{E1000_RDH(0), "RDH"},
  263	{E1000_RDT(0), "RDT"},
  264	{E1000_RXDCTL(0), "RXDCTL"},
  265	{E1000_RDBAL(0), "RDBAL"},
  266	{E1000_RDBAH(0), "RDBAH"},
  267
  268	/* TX Registers */
  269	{E1000_TCTL, "TCTL"},
  270	{E1000_TDBAL(0), "TDBAL"},
  271	{E1000_TDBAH(0), "TDBAH"},
  272	{E1000_TDLEN(0), "TDLEN"},
  273	{E1000_TDH(0), "TDH"},
  274	{E1000_TDT(0), "TDT"},
  275	{E1000_TXDCTL(0), "TXDCTL"},
  276	{E1000_TDFH, "TDFH"},
  277	{E1000_TDFT, "TDFT"},
  278	{E1000_TDFHS, "TDFHS"},
  279	{E1000_TDFPC, "TDFPC"},
  280
  281	/* List Terminator */
  282	{}
  283};
  284
  285/* igb_regdump - register printout routine */
  286static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo)
  287{
  288	int n = 0;
  289	char rname[16];
  290	u32 regs[8];
  291
  292	switch (reginfo->ofs) {
  293	case E1000_RDLEN(0):
  294		for (n = 0; n < 4; n++)
  295			regs[n] = rd32(E1000_RDLEN(n));
  296		break;
  297	case E1000_RDH(0):
  298		for (n = 0; n < 4; n++)
  299			regs[n] = rd32(E1000_RDH(n));
  300		break;
  301	case E1000_RDT(0):
  302		for (n = 0; n < 4; n++)
  303			regs[n] = rd32(E1000_RDT(n));
  304		break;
  305	case E1000_RXDCTL(0):
  306		for (n = 0; n < 4; n++)
  307			regs[n] = rd32(E1000_RXDCTL(n));
  308		break;
  309	case E1000_RDBAL(0):
  310		for (n = 0; n < 4; n++)
  311			regs[n] = rd32(E1000_RDBAL(n));
  312		break;
  313	case E1000_RDBAH(0):
  314		for (n = 0; n < 4; n++)
  315			regs[n] = rd32(E1000_RDBAH(n));
  316		break;
  317	case E1000_TDBAL(0):
  318		for (n = 0; n < 4; n++)
  319			regs[n] = rd32(E1000_TDBAL(n));
  320		break;
  321	case E1000_TDBAH(0):
  322		for (n = 0; n < 4; n++)
  323			regs[n] = rd32(E1000_TDBAH(n));
  324		break;
  325	case E1000_TDLEN(0):
  326		for (n = 0; n < 4; n++)
  327			regs[n] = rd32(E1000_TDLEN(n));
  328		break;
  329	case E1000_TDH(0):
  330		for (n = 0; n < 4; n++)
  331			regs[n] = rd32(E1000_TDH(n));
  332		break;
  333	case E1000_TDT(0):
  334		for (n = 0; n < 4; n++)
  335			regs[n] = rd32(E1000_TDT(n));
  336		break;
  337	case E1000_TXDCTL(0):
  338		for (n = 0; n < 4; n++)
  339			regs[n] = rd32(E1000_TXDCTL(n));
  340		break;
  341	default:
  342		pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs));
  343		return;
  344	}
  345
  346	snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]");
  347	pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1],
  348		regs[2], regs[3]);
  349}
  350
  351/* igb_dump - Print registers, Tx-rings and Rx-rings */
  352static void igb_dump(struct igb_adapter *adapter)
  353{
  354	struct net_device *netdev = adapter->netdev;
  355	struct e1000_hw *hw = &adapter->hw;
  356	struct igb_reg_info *reginfo;
  357	struct igb_ring *tx_ring;
  358	union e1000_adv_tx_desc *tx_desc;
  359	struct my_u0 { __le64 a; __le64 b; } *u0;
  360	struct igb_ring *rx_ring;
  361	union e1000_adv_rx_desc *rx_desc;
  362	u32 staterr;
  363	u16 i, n;
  364
  365	if (!netif_msg_hw(adapter))
  366		return;
  367
  368	/* Print netdevice Info */
  369	if (netdev) {
  370		dev_info(&adapter->pdev->dev, "Net device Info\n");
  371		pr_info("Device Name     state            trans_start\n");
  372		pr_info("%-15s %016lX %016lX\n", netdev->name,
  373			netdev->state, dev_trans_start(netdev));
 
  374	}
  375
  376	/* Print Registers */
  377	dev_info(&adapter->pdev->dev, "Register Dump\n");
  378	pr_info(" Register Name   Value\n");
  379	for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
  380	     reginfo->name; reginfo++) {
  381		igb_regdump(hw, reginfo);
  382	}
  383
  384	/* Print TX Ring Summary */
  385	if (!netdev || !netif_running(netdev))
  386		goto exit;
  387
  388	dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
  389	pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
  390	for (n = 0; n < adapter->num_tx_queues; n++) {
  391		struct igb_tx_buffer *buffer_info;
  392		tx_ring = adapter->tx_ring[n];
  393		buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
  394		pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
  395			n, tx_ring->next_to_use, tx_ring->next_to_clean,
  396			(u64)dma_unmap_addr(buffer_info, dma),
  397			dma_unmap_len(buffer_info, len),
  398			buffer_info->next_to_watch,
  399			(u64)buffer_info->time_stamp);
  400	}
  401
  402	/* Print TX Rings */
  403	if (!netif_msg_tx_done(adapter))
  404		goto rx_ring_summary;
  405
  406	dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
  407
  408	/* Transmit Descriptor Formats
  409	 *
  410	 * Advanced Transmit Descriptor
  411	 *   +--------------------------------------------------------------+
  412	 * 0 |         Buffer Address [63:0]                                |
  413	 *   +--------------------------------------------------------------+
  414	 * 8 | PAYLEN  | PORTS  |CC|IDX | STA | DCMD  |DTYP|MAC|RSV| DTALEN |
  415	 *   +--------------------------------------------------------------+
  416	 *   63      46 45    40 39 38 36 35 32 31   24             15       0
  417	 */
  418
  419	for (n = 0; n < adapter->num_tx_queues; n++) {
  420		tx_ring = adapter->tx_ring[n];
  421		pr_info("------------------------------------\n");
  422		pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
  423		pr_info("------------------------------------\n");
  424		pr_info("T [desc]     [address 63:0  ] [PlPOCIStDDM Ln] [bi->dma       ] leng  ntw timestamp        bi->skb\n");
 
 
  425
  426		for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
  427			const char *next_desc;
  428			struct igb_tx_buffer *buffer_info;
  429			tx_desc = IGB_TX_DESC(tx_ring, i);
  430			buffer_info = &tx_ring->tx_buffer_info[i];
  431			u0 = (struct my_u0 *)tx_desc;
  432			if (i == tx_ring->next_to_use &&
  433			    i == tx_ring->next_to_clean)
  434				next_desc = " NTC/U";
  435			else if (i == tx_ring->next_to_use)
  436				next_desc = " NTU";
  437			else if (i == tx_ring->next_to_clean)
  438				next_desc = " NTC";
  439			else
  440				next_desc = "";
  441
  442			pr_info("T [0x%03X]    %016llX %016llX %016llX %04X  %p %016llX %p%s\n",
  443				i, le64_to_cpu(u0->a),
 
  444				le64_to_cpu(u0->b),
  445				(u64)dma_unmap_addr(buffer_info, dma),
  446				dma_unmap_len(buffer_info, len),
  447				buffer_info->next_to_watch,
  448				(u64)buffer_info->time_stamp,
  449				buffer_info->skb, next_desc);
  450
  451			if (netif_msg_pktdata(adapter) && buffer_info->skb)
  452				print_hex_dump(KERN_INFO, "",
  453					DUMP_PREFIX_ADDRESS,
  454					16, 1, buffer_info->skb->data,
  455					dma_unmap_len(buffer_info, len),
  456					true);
  457		}
  458	}
  459
  460	/* Print RX Rings Summary */
  461rx_ring_summary:
  462	dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
  463	pr_info("Queue [NTU] [NTC]\n");
  464	for (n = 0; n < adapter->num_rx_queues; n++) {
  465		rx_ring = adapter->rx_ring[n];
  466		pr_info(" %5d %5X %5X\n",
  467			n, rx_ring->next_to_use, rx_ring->next_to_clean);
  468	}
  469
  470	/* Print RX Rings */
  471	if (!netif_msg_rx_status(adapter))
  472		goto exit;
  473
  474	dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
  475
  476	/* Advanced Receive Descriptor (Read) Format
  477	 *    63                                           1        0
  478	 *    +-----------------------------------------------------+
  479	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
  480	 *    +----------------------------------------------+------+
  481	 *  8 |       Header Buffer Address [63:1]           |  DD  |
  482	 *    +-----------------------------------------------------+
  483	 *
  484	 *
  485	 * Advanced Receive Descriptor (Write-Back) Format
  486	 *
  487	 *   63       48 47    32 31  30      21 20 17 16   4 3     0
  488	 *   +------------------------------------------------------+
  489	 * 0 | Packet     IP     |SPH| HDR_LEN   | RSV|Packet|  RSS |
  490	 *   | Checksum   Ident  |   |           |    | Type | Type |
  491	 *   +------------------------------------------------------+
  492	 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
  493	 *   +------------------------------------------------------+
  494	 *   63       48 47    32 31            20 19               0
  495	 */
  496
  497	for (n = 0; n < adapter->num_rx_queues; n++) {
  498		rx_ring = adapter->rx_ring[n];
  499		pr_info("------------------------------------\n");
  500		pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
  501		pr_info("------------------------------------\n");
  502		pr_info("R  [desc]      [ PktBuf     A0] [  HeadBuf   DD] [bi->dma       ] [bi->skb] <-- Adv Rx Read format\n");
  503		pr_info("RWB[desc]      [PcsmIpSHl PtRs] [vl er S cks ln] ---------------- [bi->skb] <-- Adv Rx Write-Back format\n");
 
 
  504
  505		for (i = 0; i < rx_ring->count; i++) {
  506			const char *next_desc;
  507			struct igb_rx_buffer *buffer_info;
  508			buffer_info = &rx_ring->rx_buffer_info[i];
  509			rx_desc = IGB_RX_DESC(rx_ring, i);
  510			u0 = (struct my_u0 *)rx_desc;
  511			staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
  512
  513			if (i == rx_ring->next_to_use)
  514				next_desc = " NTU";
  515			else if (i == rx_ring->next_to_clean)
  516				next_desc = " NTC";
  517			else
  518				next_desc = "";
  519
  520			if (staterr & E1000_RXD_STAT_DD) {
  521				/* Descriptor Done */
  522				pr_info("%s[0x%03X]     %016llX %016llX ---------------- %s\n",
  523					"RWB", i,
  524					le64_to_cpu(u0->a),
  525					le64_to_cpu(u0->b),
  526					next_desc);
  527			} else {
  528				pr_info("%s[0x%03X]     %016llX %016llX %016llX %s\n",
  529					"R  ", i,
  530					le64_to_cpu(u0->a),
  531					le64_to_cpu(u0->b),
  532					(u64)buffer_info->dma,
  533					next_desc);
  534
  535				if (netif_msg_pktdata(adapter) &&
  536				    buffer_info->dma && buffer_info->page) {
  537					print_hex_dump(KERN_INFO, "",
  538					  DUMP_PREFIX_ADDRESS,
  539					  16, 1,
  540					  page_address(buffer_info->page) +
  541						      buffer_info->page_offset,
  542					  igb_rx_bufsz(rx_ring), true);
  543				}
  544			}
  545		}
  546	}
  547
  548exit:
  549	return;
  550}
  551
  552/**
  553 *  igb_get_i2c_data - Reads the I2C SDA data bit
  554 *  @data: opaque pointer to adapter struct
 
  555 *
  556 *  Returns the I2C data bit value
  557 **/
  558static int igb_get_i2c_data(void *data)
  559{
  560	struct igb_adapter *adapter = (struct igb_adapter *)data;
  561	struct e1000_hw *hw = &adapter->hw;
  562	s32 i2cctl = rd32(E1000_I2CPARAMS);
  563
  564	return !!(i2cctl & E1000_I2C_DATA_IN);
  565}
  566
  567/**
  568 *  igb_set_i2c_data - Sets the I2C data bit
  569 *  @data: pointer to hardware structure
  570 *  @state: I2C data value (0 or 1) to set
  571 *
  572 *  Sets the I2C data bit
  573 **/
  574static void igb_set_i2c_data(void *data, int state)
  575{
  576	struct igb_adapter *adapter = (struct igb_adapter *)data;
  577	struct e1000_hw *hw = &adapter->hw;
  578	s32 i2cctl = rd32(E1000_I2CPARAMS);
  579
  580	if (state) {
  581		i2cctl |= E1000_I2C_DATA_OUT | E1000_I2C_DATA_OE_N;
  582	} else {
  583		i2cctl &= ~E1000_I2C_DATA_OE_N;
  584		i2cctl &= ~E1000_I2C_DATA_OUT;
  585	}
  586
 
 
  587	wr32(E1000_I2CPARAMS, i2cctl);
  588	wrfl();
 
  589}
  590
  591/**
  592 *  igb_set_i2c_clk - Sets the I2C SCL clock
  593 *  @data: pointer to hardware structure
  594 *  @state: state to set clock
  595 *
  596 *  Sets the I2C clock line to state
  597 **/
  598static void igb_set_i2c_clk(void *data, int state)
  599{
  600	struct igb_adapter *adapter = (struct igb_adapter *)data;
  601	struct e1000_hw *hw = &adapter->hw;
  602	s32 i2cctl = rd32(E1000_I2CPARAMS);
  603
  604	if (state) {
  605		i2cctl |= E1000_I2C_CLK_OUT | E1000_I2C_CLK_OE_N;
 
  606	} else {
  607		i2cctl &= ~E1000_I2C_CLK_OUT;
  608		i2cctl &= ~E1000_I2C_CLK_OE_N;
  609	}
  610	wr32(E1000_I2CPARAMS, i2cctl);
  611	wrfl();
  612}
  613
  614/**
  615 *  igb_get_i2c_clk - Gets the I2C SCL clock state
  616 *  @data: pointer to hardware structure
  617 *
  618 *  Gets the I2C clock state
  619 **/
  620static int igb_get_i2c_clk(void *data)
  621{
  622	struct igb_adapter *adapter = (struct igb_adapter *)data;
  623	struct e1000_hw *hw = &adapter->hw;
  624	s32 i2cctl = rd32(E1000_I2CPARAMS);
  625
  626	return !!(i2cctl & E1000_I2C_CLK_IN);
  627}
  628
  629static const struct i2c_algo_bit_data igb_i2c_algo = {
  630	.setsda		= igb_set_i2c_data,
  631	.setscl		= igb_set_i2c_clk,
  632	.getsda		= igb_get_i2c_data,
  633	.getscl		= igb_get_i2c_clk,
  634	.udelay		= 5,
  635	.timeout	= 20,
  636};
  637
  638/**
  639 *  igb_get_hw_dev - return device
  640 *  @hw: pointer to hardware structure
  641 *
  642 *  used by hardware layer to print debugging information
  643 **/
  644struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
  645{
  646	struct igb_adapter *adapter = hw->back;
  647	return adapter->netdev;
  648}
  649
  650/**
  651 *  igb_init_module - Driver Registration Routine
  652 *
  653 *  igb_init_module is the first routine called when the driver is
  654 *  loaded. All it does is register with the PCI subsystem.
  655 **/
  656static int __init igb_init_module(void)
  657{
  658	int ret;
 
 
  659
  660	pr_info("%s\n", igb_driver_string);
  661	pr_info("%s\n", igb_copyright);
  662
  663#ifdef CONFIG_IGB_DCA
  664	dca_register_notify(&dca_notifier);
  665#endif
  666	ret = pci_register_driver(&igb_driver);
  667	return ret;
  668}
  669
  670module_init(igb_init_module);
  671
  672/**
  673 *  igb_exit_module - Driver Exit Cleanup Routine
  674 *
  675 *  igb_exit_module is called just before the driver is removed
  676 *  from memory.
  677 **/
  678static void __exit igb_exit_module(void)
  679{
  680#ifdef CONFIG_IGB_DCA
  681	dca_unregister_notify(&dca_notifier);
  682#endif
  683	pci_unregister_driver(&igb_driver);
  684}
  685
  686module_exit(igb_exit_module);
  687
  688#define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
  689/**
  690 *  igb_cache_ring_register - Descriptor ring to register mapping
  691 *  @adapter: board private structure to initialize
  692 *
  693 *  Once we know the feature-set enabled for the device, we'll cache
  694 *  the register offset the descriptor ring is assigned to.
  695 **/
  696static void igb_cache_ring_register(struct igb_adapter *adapter)
  697{
  698	int i = 0, j = 0;
  699	u32 rbase_offset = adapter->vfs_allocated_count;
  700
  701	switch (adapter->hw.mac.type) {
  702	case e1000_82576:
  703		/* The queues are allocated for virtualization such that VF 0
  704		 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
  705		 * In order to avoid collision we start at the first free queue
  706		 * and continue consuming queues in the same sequence
  707		 */
  708		if (adapter->vfs_allocated_count) {
  709			for (; i < adapter->rss_queues; i++)
  710				adapter->rx_ring[i]->reg_idx = rbase_offset +
  711							       Q_IDX_82576(i);
  712		}
  713		fallthrough;
  714	case e1000_82575:
  715	case e1000_82580:
  716	case e1000_i350:
  717	case e1000_i354:
  718	case e1000_i210:
  719	case e1000_i211:
  720	default:
  721		for (; i < adapter->num_rx_queues; i++)
  722			adapter->rx_ring[i]->reg_idx = rbase_offset + i;
  723		for (; j < adapter->num_tx_queues; j++)
  724			adapter->tx_ring[j]->reg_idx = rbase_offset + j;
  725		break;
  726	}
  727}
  728
  729u32 igb_rd32(struct e1000_hw *hw, u32 reg)
  730{
  731	struct igb_adapter *igb = container_of(hw, struct igb_adapter, hw);
  732	u8 __iomem *hw_addr = READ_ONCE(hw->hw_addr);
  733	u32 value = 0;
  734
  735	if (E1000_REMOVED(hw_addr))
  736		return ~value;
  737
  738	value = readl(&hw_addr[reg]);
  739
  740	/* reads should not return all F's */
  741	if (!(~value) && (!reg || !(~readl(hw_addr)))) {
  742		struct net_device *netdev = igb->netdev;
  743		hw->hw_addr = NULL;
  744		netdev_err(netdev, "PCIe link lost\n");
  745		WARN(pci_device_is_present(igb->pdev),
  746		     "igb: Failed to read reg 0x%x!\n", reg);
  747	}
  748
  749	return value;
  750}
  751
  752/**
  753 *  igb_write_ivar - configure ivar for given MSI-X vector
  754 *  @hw: pointer to the HW structure
  755 *  @msix_vector: vector number we are allocating to a given ring
  756 *  @index: row index of IVAR register to write within IVAR table
  757 *  @offset: column offset of in IVAR, should be multiple of 8
  758 *
  759 *  This function is intended to handle the writing of the IVAR register
  760 *  for adapters 82576 and newer.  The IVAR table consists of 2 columns,
  761 *  each containing an cause allocation for an Rx and Tx ring, and a
  762 *  variable number of rows depending on the number of queues supported.
  763 **/
  764static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
  765			   int index, int offset)
  766{
  767	u32 ivar = array_rd32(E1000_IVAR0, index);
  768
  769	/* clear any bits that are currently set */
  770	ivar &= ~((u32)0xFF << offset);
  771
  772	/* write vector and valid bit */
  773	ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
  774
  775	array_wr32(E1000_IVAR0, index, ivar);
  776}
  777
  778#define IGB_N0_QUEUE -1
  779static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
  780{
  781	struct igb_adapter *adapter = q_vector->adapter;
  782	struct e1000_hw *hw = &adapter->hw;
  783	int rx_queue = IGB_N0_QUEUE;
  784	int tx_queue = IGB_N0_QUEUE;
  785	u32 msixbm = 0;
  786
  787	if (q_vector->rx.ring)
  788		rx_queue = q_vector->rx.ring->reg_idx;
  789	if (q_vector->tx.ring)
  790		tx_queue = q_vector->tx.ring->reg_idx;
  791
  792	switch (hw->mac.type) {
  793	case e1000_82575:
  794		/* The 82575 assigns vectors using a bitmask, which matches the
  795		 * bitmask for the EICR/EIMS/EIMC registers.  To assign one
  796		 * or more queues to a vector, we write the appropriate bits
  797		 * into the MSIXBM register for that vector.
  798		 */
  799		if (rx_queue > IGB_N0_QUEUE)
  800			msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
  801		if (tx_queue > IGB_N0_QUEUE)
  802			msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
  803		if (!(adapter->flags & IGB_FLAG_HAS_MSIX) && msix_vector == 0)
  804			msixbm |= E1000_EIMS_OTHER;
  805		array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
  806		q_vector->eims_value = msixbm;
  807		break;
  808	case e1000_82576:
  809		/* 82576 uses a table that essentially consists of 2 columns
  810		 * with 8 rows.  The ordering is column-major so we use the
  811		 * lower 3 bits as the row index, and the 4th bit as the
  812		 * column offset.
  813		 */
  814		if (rx_queue > IGB_N0_QUEUE)
  815			igb_write_ivar(hw, msix_vector,
  816				       rx_queue & 0x7,
  817				       (rx_queue & 0x8) << 1);
  818		if (tx_queue > IGB_N0_QUEUE)
  819			igb_write_ivar(hw, msix_vector,
  820				       tx_queue & 0x7,
  821				       ((tx_queue & 0x8) << 1) + 8);
  822		q_vector->eims_value = BIT(msix_vector);
  823		break;
  824	case e1000_82580:
  825	case e1000_i350:
  826	case e1000_i354:
  827	case e1000_i210:
  828	case e1000_i211:
  829		/* On 82580 and newer adapters the scheme is similar to 82576
  830		 * however instead of ordering column-major we have things
  831		 * ordered row-major.  So we traverse the table by using
  832		 * bit 0 as the column offset, and the remaining bits as the
  833		 * row index.
  834		 */
  835		if (rx_queue > IGB_N0_QUEUE)
  836			igb_write_ivar(hw, msix_vector,
  837				       rx_queue >> 1,
  838				       (rx_queue & 0x1) << 4);
  839		if (tx_queue > IGB_N0_QUEUE)
  840			igb_write_ivar(hw, msix_vector,
  841				       tx_queue >> 1,
  842				       ((tx_queue & 0x1) << 4) + 8);
  843		q_vector->eims_value = BIT(msix_vector);
  844		break;
  845	default:
  846		BUG();
  847		break;
  848	}
  849
  850	/* add q_vector eims value to global eims_enable_mask */
  851	adapter->eims_enable_mask |= q_vector->eims_value;
  852
  853	/* configure q_vector to set itr on first interrupt */
  854	q_vector->set_itr = 1;
  855}
  856
  857/**
  858 *  igb_configure_msix - Configure MSI-X hardware
  859 *  @adapter: board private structure to initialize
  860 *
  861 *  igb_configure_msix sets up the hardware to properly
  862 *  generate MSI-X interrupts.
  863 **/
  864static void igb_configure_msix(struct igb_adapter *adapter)
  865{
  866	u32 tmp;
  867	int i, vector = 0;
  868	struct e1000_hw *hw = &adapter->hw;
  869
  870	adapter->eims_enable_mask = 0;
  871
  872	/* set vector for other causes, i.e. link changes */
  873	switch (hw->mac.type) {
  874	case e1000_82575:
  875		tmp = rd32(E1000_CTRL_EXT);
  876		/* enable MSI-X PBA support*/
  877		tmp |= E1000_CTRL_EXT_PBA_CLR;
  878
  879		/* Auto-Mask interrupts upon ICR read. */
  880		tmp |= E1000_CTRL_EXT_EIAME;
  881		tmp |= E1000_CTRL_EXT_IRCA;
  882
  883		wr32(E1000_CTRL_EXT, tmp);
  884
  885		/* enable msix_other interrupt */
  886		array_wr32(E1000_MSIXBM(0), vector++, E1000_EIMS_OTHER);
  887		adapter->eims_other = E1000_EIMS_OTHER;
  888
  889		break;
  890
  891	case e1000_82576:
  892	case e1000_82580:
  893	case e1000_i350:
  894	case e1000_i354:
  895	case e1000_i210:
  896	case e1000_i211:
  897		/* Turn on MSI-X capability first, or our settings
  898		 * won't stick.  And it will take days to debug.
  899		 */
  900		wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
  901		     E1000_GPIE_PBA | E1000_GPIE_EIAME |
  902		     E1000_GPIE_NSICR);
  903
  904		/* enable msix_other interrupt */
  905		adapter->eims_other = BIT(vector);
  906		tmp = (vector++ | E1000_IVAR_VALID) << 8;
  907
  908		wr32(E1000_IVAR_MISC, tmp);
  909		break;
  910	default:
  911		/* do nothing, since nothing else supports MSI-X */
  912		break;
  913	} /* switch (hw->mac.type) */
  914
  915	adapter->eims_enable_mask |= adapter->eims_other;
  916
  917	for (i = 0; i < adapter->num_q_vectors; i++)
  918		igb_assign_vector(adapter->q_vector[i], vector++);
  919
  920	wrfl();
  921}
  922
  923/**
  924 *  igb_request_msix - Initialize MSI-X interrupts
  925 *  @adapter: board private structure to initialize
  926 *
  927 *  igb_request_msix allocates MSI-X vectors and requests interrupts from the
  928 *  kernel.
  929 **/
  930static int igb_request_msix(struct igb_adapter *adapter)
  931{
  932	unsigned int num_q_vectors = adapter->num_q_vectors;
  933	struct net_device *netdev = adapter->netdev;
 
  934	int i, err = 0, vector = 0, free_vector = 0;
  935
  936	err = request_irq(adapter->msix_entries[vector].vector,
  937			  igb_msix_other, 0, netdev->name, adapter);
  938	if (err)
  939		goto err_out;
  940
  941	if (num_q_vectors > MAX_Q_VECTORS) {
  942		num_q_vectors = MAX_Q_VECTORS;
  943		dev_warn(&adapter->pdev->dev,
  944			 "The number of queue vectors (%d) is higher than max allowed (%d)\n",
  945			 adapter->num_q_vectors, MAX_Q_VECTORS);
  946	}
  947	for (i = 0; i < num_q_vectors; i++) {
  948		struct igb_q_vector *q_vector = adapter->q_vector[i];
  949
  950		vector++;
  951
  952		q_vector->itr_register = adapter->io_addr + E1000_EITR(vector);
  953
  954		if (q_vector->rx.ring && q_vector->tx.ring)
  955			sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
  956				q_vector->rx.ring->queue_index);
  957		else if (q_vector->tx.ring)
  958			sprintf(q_vector->name, "%s-tx-%u", netdev->name,
  959				q_vector->tx.ring->queue_index);
  960		else if (q_vector->rx.ring)
  961			sprintf(q_vector->name, "%s-rx-%u", netdev->name,
  962				q_vector->rx.ring->queue_index);
  963		else
  964			sprintf(q_vector->name, "%s-unused", netdev->name);
  965
  966		err = request_irq(adapter->msix_entries[vector].vector,
  967				  igb_msix_ring, 0, q_vector->name,
  968				  q_vector);
  969		if (err)
  970			goto err_free;
  971	}
  972
  973	igb_configure_msix(adapter);
  974	return 0;
  975
  976err_free:
  977	/* free already assigned IRQs */
  978	free_irq(adapter->msix_entries[free_vector++].vector, adapter);
  979
  980	vector--;
  981	for (i = 0; i < vector; i++) {
  982		free_irq(adapter->msix_entries[free_vector++].vector,
  983			 adapter->q_vector[i]);
  984	}
  985err_out:
  986	return err;
  987}
  988
  989/**
  990 *  igb_free_q_vector - Free memory allocated for specific interrupt vector
  991 *  @adapter: board private structure to initialize
  992 *  @v_idx: Index of vector to be freed
  993 *
  994 *  This function frees the memory allocated to the q_vector.
  995 **/
  996static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx)
  997{
  998	struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
  999
 1000	adapter->q_vector[v_idx] = NULL;
 1001
 1002	/* igb_get_stats64() might access the rings on this vector,
 1003	 * we must wait a grace period before freeing it.
 1004	 */
 1005	if (q_vector)
 1006		kfree_rcu(q_vector, rcu);
 1007}
 1008
 1009/**
 1010 *  igb_reset_q_vector - Reset config for interrupt vector
 1011 *  @adapter: board private structure to initialize
 1012 *  @v_idx: Index of vector to be reset
 1013 *
 1014 *  If NAPI is enabled it will delete any references to the
 1015 *  NAPI struct. This is preparation for igb_free_q_vector.
 1016 **/
 1017static void igb_reset_q_vector(struct igb_adapter *adapter, int v_idx)
 1018{
 1019	struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
 1020
 1021	/* Coming from igb_set_interrupt_capability, the vectors are not yet
 1022	 * allocated. So, q_vector is NULL so we should stop here.
 1023	 */
 1024	if (!q_vector)
 1025		return;
 1026
 1027	if (q_vector->tx.ring)
 1028		adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
 1029
 1030	if (q_vector->rx.ring)
 1031		adapter->rx_ring[q_vector->rx.ring->queue_index] = NULL;
 1032
 1033	netif_napi_del(&q_vector->napi);
 1034
 1035}
 1036
 1037static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
 1038{
 1039	int v_idx = adapter->num_q_vectors;
 1040
 1041	if (adapter->flags & IGB_FLAG_HAS_MSIX)
 1042		pci_disable_msix(adapter->pdev);
 1043	else if (adapter->flags & IGB_FLAG_HAS_MSI)
 1044		pci_disable_msi(adapter->pdev);
 1045
 1046	while (v_idx--)
 1047		igb_reset_q_vector(adapter, v_idx);
 1048}
 1049
 1050/**
 1051 *  igb_free_q_vectors - Free memory allocated for interrupt vectors
 1052 *  @adapter: board private structure to initialize
 1053 *
 1054 *  This function frees the memory allocated to the q_vectors.  In addition if
 1055 *  NAPI is enabled it will delete any references to the NAPI struct prior
 1056 *  to freeing the q_vector.
 1057 **/
 1058static void igb_free_q_vectors(struct igb_adapter *adapter)
 1059{
 1060	int v_idx = adapter->num_q_vectors;
 1061
 1062	adapter->num_tx_queues = 0;
 1063	adapter->num_rx_queues = 0;
 1064	adapter->num_q_vectors = 0;
 1065
 1066	while (v_idx--) {
 1067		igb_reset_q_vector(adapter, v_idx);
 1068		igb_free_q_vector(adapter, v_idx);
 1069	}
 1070}
 1071
 1072/**
 1073 *  igb_clear_interrupt_scheme - reset the device to a state of no interrupts
 1074 *  @adapter: board private structure to initialize
 1075 *
 1076 *  This function resets the device so that it has 0 Rx queues, Tx queues, and
 1077 *  MSI-X interrupts allocated.
 1078 */
 1079static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
 1080{
 1081	igb_free_q_vectors(adapter);
 1082	igb_reset_interrupt_capability(adapter);
 1083}
 1084
 1085/**
 1086 *  igb_set_interrupt_capability - set MSI or MSI-X if supported
 1087 *  @adapter: board private structure to initialize
 1088 *  @msix: boolean value of MSIX capability
 1089 *
 1090 *  Attempt to configure interrupts using the best available
 1091 *  capabilities of the hardware and kernel.
 1092 **/
 1093static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix)
 1094{
 1095	int err;
 1096	int numvecs, i;
 1097
 1098	if (!msix)
 1099		goto msi_only;
 1100	adapter->flags |= IGB_FLAG_HAS_MSIX;
 1101
 1102	/* Number of supported queues. */
 1103	adapter->num_rx_queues = adapter->rss_queues;
 1104	if (adapter->vfs_allocated_count)
 1105		adapter->num_tx_queues = 1;
 1106	else
 1107		adapter->num_tx_queues = adapter->rss_queues;
 1108
 1109	/* start with one vector for every Rx queue */
 1110	numvecs = adapter->num_rx_queues;
 1111
 1112	/* if Tx handler is separate add 1 for every Tx queue */
 1113	if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
 1114		numvecs += adapter->num_tx_queues;
 1115
 1116	/* store the number of vectors reserved for queues */
 1117	adapter->num_q_vectors = numvecs;
 1118
 1119	/* add 1 vector for link status interrupts */
 1120	numvecs++;
 1121	for (i = 0; i < numvecs; i++)
 1122		adapter->msix_entries[i].entry = i;
 1123
 1124	err = pci_enable_msix_range(adapter->pdev,
 1125				    adapter->msix_entries,
 1126				    numvecs,
 1127				    numvecs);
 1128	if (err > 0)
 1129		return;
 1130
 1131	igb_reset_interrupt_capability(adapter);
 1132
 1133	/* If we can't do MSI-X, try MSI */
 1134msi_only:
 1135	adapter->flags &= ~IGB_FLAG_HAS_MSIX;
 1136#ifdef CONFIG_PCI_IOV
 1137	/* disable SR-IOV for non MSI-X configurations */
 1138	if (adapter->vf_data) {
 1139		struct e1000_hw *hw = &adapter->hw;
 1140		/* disable iov and allow time for transactions to clear */
 1141		pci_disable_sriov(adapter->pdev);
 1142		msleep(500);
 1143
 1144		kfree(adapter->vf_mac_list);
 1145		adapter->vf_mac_list = NULL;
 1146		kfree(adapter->vf_data);
 1147		adapter->vf_data = NULL;
 1148		wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
 1149		wrfl();
 1150		msleep(100);
 1151		dev_info(&adapter->pdev->dev, "IOV Disabled\n");
 1152	}
 1153#endif
 1154	adapter->vfs_allocated_count = 0;
 1155	adapter->rss_queues = 1;
 1156	adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
 1157	adapter->num_rx_queues = 1;
 1158	adapter->num_tx_queues = 1;
 1159	adapter->num_q_vectors = 1;
 1160	if (!pci_enable_msi(adapter->pdev))
 1161		adapter->flags |= IGB_FLAG_HAS_MSI;
 1162}
 1163
 1164static void igb_add_ring(struct igb_ring *ring,
 1165			 struct igb_ring_container *head)
 1166{
 1167	head->ring = ring;
 1168	head->count++;
 1169}
 1170
 1171/**
 1172 *  igb_alloc_q_vector - Allocate memory for a single interrupt vector
 1173 *  @adapter: board private structure to initialize
 1174 *  @v_count: q_vectors allocated on adapter, used for ring interleaving
 1175 *  @v_idx: index of vector in adapter struct
 1176 *  @txr_count: total number of Tx rings to allocate
 1177 *  @txr_idx: index of first Tx ring to allocate
 1178 *  @rxr_count: total number of Rx rings to allocate
 1179 *  @rxr_idx: index of first Rx ring to allocate
 1180 *
 1181 *  We allocate one q_vector.  If allocation fails we return -ENOMEM.
 1182 **/
 1183static int igb_alloc_q_vector(struct igb_adapter *adapter,
 1184			      int v_count, int v_idx,
 1185			      int txr_count, int txr_idx,
 1186			      int rxr_count, int rxr_idx)
 1187{
 1188	struct igb_q_vector *q_vector;
 1189	struct igb_ring *ring;
 1190	int ring_count;
 1191	size_t size;
 1192
 1193	/* igb only supports 1 Tx and/or 1 Rx queue per vector */
 1194	if (txr_count > 1 || rxr_count > 1)
 1195		return -ENOMEM;
 1196
 1197	ring_count = txr_count + rxr_count;
 1198	size = kmalloc_size_roundup(struct_size(q_vector, ring, ring_count));
 
 1199
 1200	/* allocate q_vector and rings */
 1201	q_vector = adapter->q_vector[v_idx];
 1202	if (!q_vector) {
 1203		q_vector = kzalloc(size, GFP_KERNEL);
 1204	} else if (size > ksize(q_vector)) {
 1205		struct igb_q_vector *new_q_vector;
 1206
 1207		new_q_vector = kzalloc(size, GFP_KERNEL);
 1208		if (new_q_vector)
 1209			kfree_rcu(q_vector, rcu);
 1210		q_vector = new_q_vector;
 1211	} else {
 1212		memset(q_vector, 0, size);
 1213	}
 1214	if (!q_vector)
 1215		return -ENOMEM;
 1216
 1217	/* initialize NAPI */
 1218	netif_napi_add(adapter->netdev, &q_vector->napi, igb_poll);
 
 1219
 1220	/* tie q_vector and adapter together */
 1221	adapter->q_vector[v_idx] = q_vector;
 1222	q_vector->adapter = adapter;
 1223
 1224	/* initialize work limits */
 1225	q_vector->tx.work_limit = adapter->tx_work_limit;
 1226
 1227	/* initialize ITR configuration */
 1228	q_vector->itr_register = adapter->io_addr + E1000_EITR(0);
 1229	q_vector->itr_val = IGB_START_ITR;
 1230
 1231	/* initialize pointer to rings */
 1232	ring = q_vector->ring;
 1233
 1234	/* intialize ITR */
 1235	if (rxr_count) {
 1236		/* rx or rx/tx vector */
 1237		if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
 1238			q_vector->itr_val = adapter->rx_itr_setting;
 1239	} else {
 1240		/* tx only vector */
 1241		if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
 1242			q_vector->itr_val = adapter->tx_itr_setting;
 1243	}
 1244
 1245	if (txr_count) {
 1246		/* assign generic ring traits */
 1247		ring->dev = &adapter->pdev->dev;
 1248		ring->netdev = adapter->netdev;
 1249
 1250		/* configure backlink on ring */
 1251		ring->q_vector = q_vector;
 1252
 1253		/* update q_vector Tx values */
 1254		igb_add_ring(ring, &q_vector->tx);
 1255
 1256		/* For 82575, context index must be unique per ring. */
 1257		if (adapter->hw.mac.type == e1000_82575)
 1258			set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
 1259
 1260		/* apply Tx specific ring traits */
 1261		ring->count = adapter->tx_ring_count;
 1262		ring->queue_index = txr_idx;
 1263
 1264		ring->cbs_enable = false;
 1265		ring->idleslope = 0;
 1266		ring->sendslope = 0;
 1267		ring->hicredit = 0;
 1268		ring->locredit = 0;
 1269
 1270		u64_stats_init(&ring->tx_syncp);
 1271		u64_stats_init(&ring->tx_syncp2);
 1272
 1273		/* assign ring to adapter */
 1274		adapter->tx_ring[txr_idx] = ring;
 1275
 1276		/* push pointer to next ring */
 1277		ring++;
 1278	}
 1279
 1280	if (rxr_count) {
 1281		/* assign generic ring traits */
 1282		ring->dev = &adapter->pdev->dev;
 1283		ring->netdev = adapter->netdev;
 1284
 1285		/* configure backlink on ring */
 1286		ring->q_vector = q_vector;
 1287
 1288		/* update q_vector Rx values */
 1289		igb_add_ring(ring, &q_vector->rx);
 1290
 1291		/* set flag indicating ring supports SCTP checksum offload */
 1292		if (adapter->hw.mac.type >= e1000_82576)
 1293			set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
 1294
 1295		/* On i350, i354, i210, and i211, loopback VLAN packets
 
 1296		 * have the tag byte-swapped.
 1297		 */
 1298		if (adapter->hw.mac.type >= e1000_i350)
 1299			set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
 1300
 1301		/* apply Rx specific ring traits */
 1302		ring->count = adapter->rx_ring_count;
 1303		ring->queue_index = rxr_idx;
 1304
 1305		u64_stats_init(&ring->rx_syncp);
 1306
 1307		/* assign ring to adapter */
 1308		adapter->rx_ring[rxr_idx] = ring;
 1309	}
 1310
 1311	return 0;
 1312}
 1313
 1314
 1315/**
 1316 *  igb_alloc_q_vectors - Allocate memory for interrupt vectors
 1317 *  @adapter: board private structure to initialize
 1318 *
 1319 *  We allocate one q_vector per queue interrupt.  If allocation fails we
 1320 *  return -ENOMEM.
 1321 **/
 1322static int igb_alloc_q_vectors(struct igb_adapter *adapter)
 1323{
 1324	int q_vectors = adapter->num_q_vectors;
 1325	int rxr_remaining = adapter->num_rx_queues;
 1326	int txr_remaining = adapter->num_tx_queues;
 1327	int rxr_idx = 0, txr_idx = 0, v_idx = 0;
 1328	int err;
 1329
 1330	if (q_vectors >= (rxr_remaining + txr_remaining)) {
 1331		for (; rxr_remaining; v_idx++) {
 1332			err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
 1333						 0, 0, 1, rxr_idx);
 1334
 1335			if (err)
 1336				goto err_out;
 1337
 1338			/* update counts and index */
 1339			rxr_remaining--;
 1340			rxr_idx++;
 1341		}
 1342	}
 1343
 1344	for (; v_idx < q_vectors; v_idx++) {
 1345		int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
 1346		int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
 1347
 1348		err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
 1349					 tqpv, txr_idx, rqpv, rxr_idx);
 1350
 1351		if (err)
 1352			goto err_out;
 1353
 1354		/* update counts and index */
 1355		rxr_remaining -= rqpv;
 1356		txr_remaining -= tqpv;
 1357		rxr_idx++;
 1358		txr_idx++;
 1359	}
 1360
 1361	return 0;
 1362
 1363err_out:
 1364	adapter->num_tx_queues = 0;
 1365	adapter->num_rx_queues = 0;
 1366	adapter->num_q_vectors = 0;
 1367
 1368	while (v_idx--)
 1369		igb_free_q_vector(adapter, v_idx);
 1370
 1371	return -ENOMEM;
 1372}
 1373
 1374/**
 1375 *  igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
 1376 *  @adapter: board private structure to initialize
 1377 *  @msix: boolean value of MSIX capability
 1378 *
 1379 *  This function initializes the interrupts and allocates all of the queues.
 1380 **/
 1381static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
 1382{
 1383	struct pci_dev *pdev = adapter->pdev;
 1384	int err;
 1385
 1386	igb_set_interrupt_capability(adapter, msix);
 1387
 1388	err = igb_alloc_q_vectors(adapter);
 1389	if (err) {
 1390		dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
 1391		goto err_alloc_q_vectors;
 1392	}
 1393
 1394	igb_cache_ring_register(adapter);
 1395
 1396	return 0;
 1397
 1398err_alloc_q_vectors:
 1399	igb_reset_interrupt_capability(adapter);
 1400	return err;
 1401}
 1402
 1403/**
 1404 *  igb_request_irq - initialize interrupts
 1405 *  @adapter: board private structure to initialize
 1406 *
 1407 *  Attempts to configure interrupts using the best available
 1408 *  capabilities of the hardware and kernel.
 1409 **/
 1410static int igb_request_irq(struct igb_adapter *adapter)
 1411{
 1412	struct net_device *netdev = adapter->netdev;
 1413	struct pci_dev *pdev = adapter->pdev;
 1414	int err = 0;
 1415
 1416	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
 1417		err = igb_request_msix(adapter);
 1418		if (!err)
 1419			goto request_done;
 1420		/* fall back to MSI */
 1421		igb_free_all_tx_resources(adapter);
 1422		igb_free_all_rx_resources(adapter);
 1423
 1424		igb_clear_interrupt_scheme(adapter);
 1425		err = igb_init_interrupt_scheme(adapter, false);
 1426		if (err)
 1427			goto request_done;
 1428
 1429		igb_setup_all_tx_resources(adapter);
 1430		igb_setup_all_rx_resources(adapter);
 1431		igb_configure(adapter);
 1432	}
 1433
 1434	igb_assign_vector(adapter->q_vector[0], 0);
 1435
 1436	if (adapter->flags & IGB_FLAG_HAS_MSI) {
 1437		err = request_irq(pdev->irq, igb_intr_msi, 0,
 1438				  netdev->name, adapter);
 1439		if (!err)
 1440			goto request_done;
 1441
 1442		/* fall back to legacy interrupts */
 1443		igb_reset_interrupt_capability(adapter);
 1444		adapter->flags &= ~IGB_FLAG_HAS_MSI;
 1445	}
 1446
 1447	err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
 1448			  netdev->name, adapter);
 1449
 1450	if (err)
 1451		dev_err(&pdev->dev, "Error %d getting interrupt\n",
 1452			err);
 1453
 1454request_done:
 1455	return err;
 1456}
 1457
 1458static void igb_free_irq(struct igb_adapter *adapter)
 1459{
 1460	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
 1461		int vector = 0, i;
 1462
 1463		free_irq(adapter->msix_entries[vector++].vector, adapter);
 1464
 1465		for (i = 0; i < adapter->num_q_vectors; i++)
 1466			free_irq(adapter->msix_entries[vector++].vector,
 1467				 adapter->q_vector[i]);
 1468	} else {
 1469		free_irq(adapter->pdev->irq, adapter);
 1470	}
 1471}
 1472
 1473/**
 1474 *  igb_irq_disable - Mask off interrupt generation on the NIC
 1475 *  @adapter: board private structure
 1476 **/
 1477static void igb_irq_disable(struct igb_adapter *adapter)
 1478{
 1479	struct e1000_hw *hw = &adapter->hw;
 1480
 1481	/* we need to be careful when disabling interrupts.  The VFs are also
 1482	 * mapped into these registers and so clearing the bits can cause
 1483	 * issues on the VF drivers so we only need to clear what we set
 1484	 */
 1485	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
 1486		u32 regval = rd32(E1000_EIAM);
 1487
 1488		wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
 1489		wr32(E1000_EIMC, adapter->eims_enable_mask);
 1490		regval = rd32(E1000_EIAC);
 1491		wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
 1492	}
 1493
 1494	wr32(E1000_IAM, 0);
 1495	wr32(E1000_IMC, ~0);
 1496	wrfl();
 1497	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
 1498		int i;
 1499
 1500		for (i = 0; i < adapter->num_q_vectors; i++)
 1501			synchronize_irq(adapter->msix_entries[i].vector);
 1502	} else {
 1503		synchronize_irq(adapter->pdev->irq);
 1504	}
 1505}
 1506
 1507/**
 1508 *  igb_irq_enable - Enable default interrupt generation settings
 1509 *  @adapter: board private structure
 1510 **/
 1511static void igb_irq_enable(struct igb_adapter *adapter)
 1512{
 1513	struct e1000_hw *hw = &adapter->hw;
 1514
 1515	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
 1516		u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
 1517		u32 regval = rd32(E1000_EIAC);
 1518
 1519		wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
 1520		regval = rd32(E1000_EIAM);
 1521		wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
 1522		wr32(E1000_EIMS, adapter->eims_enable_mask);
 1523		if (adapter->vfs_allocated_count) {
 1524			wr32(E1000_MBVFIMR, 0xFF);
 1525			ims |= E1000_IMS_VMMB;
 1526		}
 1527		wr32(E1000_IMS, ims);
 1528	} else {
 1529		wr32(E1000_IMS, IMS_ENABLE_MASK |
 1530				E1000_IMS_DRSTA);
 1531		wr32(E1000_IAM, IMS_ENABLE_MASK |
 1532				E1000_IMS_DRSTA);
 1533	}
 1534}
 1535
 1536static void igb_update_mng_vlan(struct igb_adapter *adapter)
 1537{
 1538	struct e1000_hw *hw = &adapter->hw;
 1539	u16 pf_id = adapter->vfs_allocated_count;
 1540	u16 vid = adapter->hw.mng_cookie.vlan_id;
 1541	u16 old_vid = adapter->mng_vlan_id;
 1542
 1543	if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
 1544		/* add VID to filter table */
 1545		igb_vfta_set(hw, vid, pf_id, true, true);
 1546		adapter->mng_vlan_id = vid;
 1547	} else {
 1548		adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
 1549	}
 1550
 1551	if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
 1552	    (vid != old_vid) &&
 1553	    !test_bit(old_vid, adapter->active_vlans)) {
 1554		/* remove VID from filter table */
 1555		igb_vfta_set(hw, vid, pf_id, false, true);
 1556	}
 1557}
 1558
 1559/**
 1560 *  igb_release_hw_control - release control of the h/w to f/w
 1561 *  @adapter: address of board private structure
 1562 *
 1563 *  igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
 1564 *  For ASF and Pass Through versions of f/w this means that the
 1565 *  driver is no longer loaded.
 1566 **/
 1567static void igb_release_hw_control(struct igb_adapter *adapter)
 1568{
 1569	struct e1000_hw *hw = &adapter->hw;
 1570	u32 ctrl_ext;
 1571
 1572	/* Let firmware take over control of h/w */
 1573	ctrl_ext = rd32(E1000_CTRL_EXT);
 1574	wr32(E1000_CTRL_EXT,
 1575			ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
 1576}
 1577
 1578/**
 1579 *  igb_get_hw_control - get control of the h/w from f/w
 1580 *  @adapter: address of board private structure
 1581 *
 1582 *  igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
 1583 *  For ASF and Pass Through versions of f/w this means that
 1584 *  the driver is loaded.
 1585 **/
 1586static void igb_get_hw_control(struct igb_adapter *adapter)
 1587{
 1588	struct e1000_hw *hw = &adapter->hw;
 1589	u32 ctrl_ext;
 1590
 1591	/* Let firmware know the driver has taken over */
 1592	ctrl_ext = rd32(E1000_CTRL_EXT);
 1593	wr32(E1000_CTRL_EXT,
 1594			ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
 1595}
 1596
 1597static void enable_fqtss(struct igb_adapter *adapter, bool enable)
 1598{
 1599	struct net_device *netdev = adapter->netdev;
 1600	struct e1000_hw *hw = &adapter->hw;
 1601
 1602	WARN_ON(hw->mac.type != e1000_i210);
 1603
 1604	if (enable)
 1605		adapter->flags |= IGB_FLAG_FQTSS;
 1606	else
 1607		adapter->flags &= ~IGB_FLAG_FQTSS;
 1608
 1609	if (netif_running(netdev))
 1610		schedule_work(&adapter->reset_task);
 1611}
 1612
 1613static bool is_fqtss_enabled(struct igb_adapter *adapter)
 1614{
 1615	return (adapter->flags & IGB_FLAG_FQTSS) ? true : false;
 1616}
 1617
 1618static void set_tx_desc_fetch_prio(struct e1000_hw *hw, int queue,
 1619				   enum tx_queue_prio prio)
 1620{
 1621	u32 val;
 1622
 1623	WARN_ON(hw->mac.type != e1000_i210);
 1624	WARN_ON(queue < 0 || queue > 4);
 1625
 1626	val = rd32(E1000_I210_TXDCTL(queue));
 1627
 1628	if (prio == TX_QUEUE_PRIO_HIGH)
 1629		val |= E1000_TXDCTL_PRIORITY;
 1630	else
 1631		val &= ~E1000_TXDCTL_PRIORITY;
 1632
 1633	wr32(E1000_I210_TXDCTL(queue), val);
 1634}
 1635
 1636static void set_queue_mode(struct e1000_hw *hw, int queue, enum queue_mode mode)
 1637{
 1638	u32 val;
 1639
 1640	WARN_ON(hw->mac.type != e1000_i210);
 1641	WARN_ON(queue < 0 || queue > 1);
 1642
 1643	val = rd32(E1000_I210_TQAVCC(queue));
 1644
 1645	if (mode == QUEUE_MODE_STREAM_RESERVATION)
 1646		val |= E1000_TQAVCC_QUEUEMODE;
 1647	else
 1648		val &= ~E1000_TQAVCC_QUEUEMODE;
 1649
 1650	wr32(E1000_I210_TQAVCC(queue), val);
 1651}
 1652
 1653static bool is_any_cbs_enabled(struct igb_adapter *adapter)
 1654{
 1655	int i;
 1656
 1657	for (i = 0; i < adapter->num_tx_queues; i++) {
 1658		if (adapter->tx_ring[i]->cbs_enable)
 1659			return true;
 1660	}
 1661
 1662	return false;
 1663}
 1664
 1665static bool is_any_txtime_enabled(struct igb_adapter *adapter)
 1666{
 1667	int i;
 1668
 1669	for (i = 0; i < adapter->num_tx_queues; i++) {
 1670		if (adapter->tx_ring[i]->launchtime_enable)
 1671			return true;
 1672	}
 1673
 1674	return false;
 1675}
 1676
 1677/**
 1678 *  igb_config_tx_modes - Configure "Qav Tx mode" features on igb
 1679 *  @adapter: pointer to adapter struct
 1680 *  @queue: queue number
 1681 *
 1682 *  Configure CBS and Launchtime for a given hardware queue.
 1683 *  Parameters are retrieved from the correct Tx ring, so
 1684 *  igb_save_cbs_params() and igb_save_txtime_params() should be used
 1685 *  for setting those correctly prior to this function being called.
 1686 **/
 1687static void igb_config_tx_modes(struct igb_adapter *adapter, int queue)
 1688{
 1689	struct net_device *netdev = adapter->netdev;
 1690	struct e1000_hw *hw = &adapter->hw;
 1691	struct igb_ring *ring;
 1692	u32 tqavcc, tqavctrl;
 1693	u16 value;
 1694
 1695	WARN_ON(hw->mac.type != e1000_i210);
 1696	WARN_ON(queue < 0 || queue > 1);
 1697	ring = adapter->tx_ring[queue];
 1698
 1699	/* If any of the Qav features is enabled, configure queues as SR and
 1700	 * with HIGH PRIO. If none is, then configure them with LOW PRIO and
 1701	 * as SP.
 1702	 */
 1703	if (ring->cbs_enable || ring->launchtime_enable) {
 1704		set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_HIGH);
 1705		set_queue_mode(hw, queue, QUEUE_MODE_STREAM_RESERVATION);
 1706	} else {
 1707		set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_LOW);
 1708		set_queue_mode(hw, queue, QUEUE_MODE_STRICT_PRIORITY);
 1709	}
 1710
 1711	/* If CBS is enabled, set DataTranARB and config its parameters. */
 1712	if (ring->cbs_enable || queue == 0) {
 1713		/* i210 does not allow the queue 0 to be in the Strict
 1714		 * Priority mode while the Qav mode is enabled, so,
 1715		 * instead of disabling strict priority mode, we give
 1716		 * queue 0 the maximum of credits possible.
 1717		 *
 1718		 * See section 8.12.19 of the i210 datasheet, "Note:
 1719		 * Queue0 QueueMode must be set to 1b when
 1720		 * TransmitMode is set to Qav."
 1721		 */
 1722		if (queue == 0 && !ring->cbs_enable) {
 1723			/* max "linkspeed" idleslope in kbps */
 1724			ring->idleslope = 1000000;
 1725			ring->hicredit = ETH_FRAME_LEN;
 1726		}
 1727
 1728		/* Always set data transfer arbitration to credit-based
 1729		 * shaper algorithm on TQAVCTRL if CBS is enabled for any of
 1730		 * the queues.
 1731		 */
 1732		tqavctrl = rd32(E1000_I210_TQAVCTRL);
 1733		tqavctrl |= E1000_TQAVCTRL_DATATRANARB;
 1734		wr32(E1000_I210_TQAVCTRL, tqavctrl);
 1735
 1736		/* According to i210 datasheet section 7.2.7.7, we should set
 1737		 * the 'idleSlope' field from TQAVCC register following the
 1738		 * equation:
 1739		 *
 1740		 * For 100 Mbps link speed:
 1741		 *
 1742		 *     value = BW * 0x7735 * 0.2                          (E1)
 1743		 *
 1744		 * For 1000Mbps link speed:
 1745		 *
 1746		 *     value = BW * 0x7735 * 2                            (E2)
 1747		 *
 1748		 * E1 and E2 can be merged into one equation as shown below.
 1749		 * Note that 'link-speed' is in Mbps.
 1750		 *
 1751		 *     value = BW * 0x7735 * 2 * link-speed
 1752		 *                           --------------               (E3)
 1753		 *                                1000
 1754		 *
 1755		 * 'BW' is the percentage bandwidth out of full link speed
 1756		 * which can be found with the following equation. Note that
 1757		 * idleSlope here is the parameter from this function which
 1758		 * is in kbps.
 1759		 *
 1760		 *     BW =     idleSlope
 1761		 *          -----------------                             (E4)
 1762		 *          link-speed * 1000
 1763		 *
 1764		 * That said, we can come up with a generic equation to
 1765		 * calculate the value we should set it TQAVCC register by
 1766		 * replacing 'BW' in E3 by E4. The resulting equation is:
 1767		 *
 1768		 * value =     idleSlope     * 0x7735 * 2 * link-speed
 1769		 *         -----------------            --------------    (E5)
 1770		 *         link-speed * 1000                 1000
 1771		 *
 1772		 * 'link-speed' is present in both sides of the fraction so
 1773		 * it is canceled out. The final equation is the following:
 1774		 *
 1775		 *     value = idleSlope * 61034
 1776		 *             -----------------                          (E6)
 1777		 *                  1000000
 1778		 *
 1779		 * NOTE: For i210, given the above, we can see that idleslope
 1780		 *       is represented in 16.38431 kbps units by the value at
 1781		 *       the TQAVCC register (1Gbps / 61034), which reduces
 1782		 *       the granularity for idleslope increments.
 1783		 *       For instance, if you want to configure a 2576kbps
 1784		 *       idleslope, the value to be written on the register
 1785		 *       would have to be 157.23. If rounded down, you end
 1786		 *       up with less bandwidth available than originally
 1787		 *       required (~2572 kbps). If rounded up, you end up
 1788		 *       with a higher bandwidth (~2589 kbps). Below the
 1789		 *       approach we take is to always round up the
 1790		 *       calculated value, so the resulting bandwidth might
 1791		 *       be slightly higher for some configurations.
 1792		 */
 1793		value = DIV_ROUND_UP_ULL(ring->idleslope * 61034ULL, 1000000);
 1794
 1795		tqavcc = rd32(E1000_I210_TQAVCC(queue));
 1796		tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
 1797		tqavcc |= value;
 1798		wr32(E1000_I210_TQAVCC(queue), tqavcc);
 1799
 1800		wr32(E1000_I210_TQAVHC(queue),
 1801		     0x80000000 + ring->hicredit * 0x7735);
 1802	} else {
 1803
 1804		/* Set idleSlope to zero. */
 1805		tqavcc = rd32(E1000_I210_TQAVCC(queue));
 1806		tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
 1807		wr32(E1000_I210_TQAVCC(queue), tqavcc);
 1808
 1809		/* Set hiCredit to zero. */
 1810		wr32(E1000_I210_TQAVHC(queue), 0);
 1811
 1812		/* If CBS is not enabled for any queues anymore, then return to
 1813		 * the default state of Data Transmission Arbitration on
 1814		 * TQAVCTRL.
 1815		 */
 1816		if (!is_any_cbs_enabled(adapter)) {
 1817			tqavctrl = rd32(E1000_I210_TQAVCTRL);
 1818			tqavctrl &= ~E1000_TQAVCTRL_DATATRANARB;
 1819			wr32(E1000_I210_TQAVCTRL, tqavctrl);
 1820		}
 1821	}
 1822
 1823	/* If LaunchTime is enabled, set DataTranTIM. */
 1824	if (ring->launchtime_enable) {
 1825		/* Always set DataTranTIM on TQAVCTRL if LaunchTime is enabled
 1826		 * for any of the SR queues, and configure fetchtime delta.
 1827		 * XXX NOTE:
 1828		 *     - LaunchTime will be enabled for all SR queues.
 1829		 *     - A fixed offset can be added relative to the launch
 1830		 *       time of all packets if configured at reg LAUNCH_OS0.
 1831		 *       We are keeping it as 0 for now (default value).
 1832		 */
 1833		tqavctrl = rd32(E1000_I210_TQAVCTRL);
 1834		tqavctrl |= E1000_TQAVCTRL_DATATRANTIM |
 1835		       E1000_TQAVCTRL_FETCHTIME_DELTA;
 1836		wr32(E1000_I210_TQAVCTRL, tqavctrl);
 1837	} else {
 1838		/* If Launchtime is not enabled for any SR queues anymore,
 1839		 * then clear DataTranTIM on TQAVCTRL and clear fetchtime delta,
 1840		 * effectively disabling Launchtime.
 1841		 */
 1842		if (!is_any_txtime_enabled(adapter)) {
 1843			tqavctrl = rd32(E1000_I210_TQAVCTRL);
 1844			tqavctrl &= ~E1000_TQAVCTRL_DATATRANTIM;
 1845			tqavctrl &= ~E1000_TQAVCTRL_FETCHTIME_DELTA;
 1846			wr32(E1000_I210_TQAVCTRL, tqavctrl);
 1847		}
 1848	}
 1849
 1850	/* XXX: In i210 controller the sendSlope and loCredit parameters from
 1851	 * CBS are not configurable by software so we don't do any 'controller
 1852	 * configuration' in respect to these parameters.
 1853	 */
 1854
 1855	netdev_dbg(netdev, "Qav Tx mode: cbs %s, launchtime %s, queue %d idleslope %d sendslope %d hiCredit %d locredit %d\n",
 1856		   ring->cbs_enable ? "enabled" : "disabled",
 1857		   ring->launchtime_enable ? "enabled" : "disabled",
 1858		   queue,
 1859		   ring->idleslope, ring->sendslope,
 1860		   ring->hicredit, ring->locredit);
 1861}
 1862
 1863static int igb_save_txtime_params(struct igb_adapter *adapter, int queue,
 1864				  bool enable)
 1865{
 1866	struct igb_ring *ring;
 1867
 1868	if (queue < 0 || queue > adapter->num_tx_queues)
 1869		return -EINVAL;
 1870
 1871	ring = adapter->tx_ring[queue];
 1872	ring->launchtime_enable = enable;
 1873
 1874	return 0;
 1875}
 1876
 1877static int igb_save_cbs_params(struct igb_adapter *adapter, int queue,
 1878			       bool enable, int idleslope, int sendslope,
 1879			       int hicredit, int locredit)
 1880{
 1881	struct igb_ring *ring;
 1882
 1883	if (queue < 0 || queue > adapter->num_tx_queues)
 1884		return -EINVAL;
 1885
 1886	ring = adapter->tx_ring[queue];
 1887
 1888	ring->cbs_enable = enable;
 1889	ring->idleslope = idleslope;
 1890	ring->sendslope = sendslope;
 1891	ring->hicredit = hicredit;
 1892	ring->locredit = locredit;
 1893
 1894	return 0;
 1895}
 1896
 1897/**
 1898 *  igb_setup_tx_mode - Switch to/from Qav Tx mode when applicable
 1899 *  @adapter: pointer to adapter struct
 1900 *
 1901 *  Configure TQAVCTRL register switching the controller's Tx mode
 1902 *  if FQTSS mode is enabled or disabled. Additionally, will issue
 1903 *  a call to igb_config_tx_modes() per queue so any previously saved
 1904 *  Tx parameters are applied.
 1905 **/
 1906static void igb_setup_tx_mode(struct igb_adapter *adapter)
 1907{
 1908	struct net_device *netdev = adapter->netdev;
 1909	struct e1000_hw *hw = &adapter->hw;
 1910	u32 val;
 1911
 1912	/* Only i210 controller supports changing the transmission mode. */
 1913	if (hw->mac.type != e1000_i210)
 1914		return;
 1915
 1916	if (is_fqtss_enabled(adapter)) {
 1917		int i, max_queue;
 1918
 1919		/* Configure TQAVCTRL register: set transmit mode to 'Qav',
 1920		 * set data fetch arbitration to 'round robin', set SP_WAIT_SR
 1921		 * so SP queues wait for SR ones.
 1922		 */
 1923		val = rd32(E1000_I210_TQAVCTRL);
 1924		val |= E1000_TQAVCTRL_XMIT_MODE | E1000_TQAVCTRL_SP_WAIT_SR;
 1925		val &= ~E1000_TQAVCTRL_DATAFETCHARB;
 1926		wr32(E1000_I210_TQAVCTRL, val);
 1927
 1928		/* Configure Tx and Rx packet buffers sizes as described in
 1929		 * i210 datasheet section 7.2.7.7.
 1930		 */
 1931		val = rd32(E1000_TXPBS);
 1932		val &= ~I210_TXPBSIZE_MASK;
 1933		val |= I210_TXPBSIZE_PB0_6KB | I210_TXPBSIZE_PB1_6KB |
 1934			I210_TXPBSIZE_PB2_6KB | I210_TXPBSIZE_PB3_6KB;
 1935		wr32(E1000_TXPBS, val);
 1936
 1937		val = rd32(E1000_RXPBS);
 1938		val &= ~I210_RXPBSIZE_MASK;
 1939		val |= I210_RXPBSIZE_PB_30KB;
 1940		wr32(E1000_RXPBS, val);
 1941
 1942		/* Section 8.12.9 states that MAX_TPKT_SIZE from DTXMXPKTSZ
 1943		 * register should not exceed the buffer size programmed in
 1944		 * TXPBS. The smallest buffer size programmed in TXPBS is 4kB
 1945		 * so according to the datasheet we should set MAX_TPKT_SIZE to
 1946		 * 4kB / 64.
 1947		 *
 1948		 * However, when we do so, no frame from queue 2 and 3 are
 1949		 * transmitted.  It seems the MAX_TPKT_SIZE should not be great
 1950		 * or _equal_ to the buffer size programmed in TXPBS. For this
 1951		 * reason, we set MAX_ TPKT_SIZE to (4kB - 1) / 64.
 1952		 */
 1953		val = (4096 - 1) / 64;
 1954		wr32(E1000_I210_DTXMXPKTSZ, val);
 1955
 1956		/* Since FQTSS mode is enabled, apply any CBS configuration
 1957		 * previously set. If no previous CBS configuration has been
 1958		 * done, then the initial configuration is applied, which means
 1959		 * CBS is disabled.
 1960		 */
 1961		max_queue = (adapter->num_tx_queues < I210_SR_QUEUES_NUM) ?
 1962			    adapter->num_tx_queues : I210_SR_QUEUES_NUM;
 1963
 1964		for (i = 0; i < max_queue; i++) {
 1965			igb_config_tx_modes(adapter, i);
 1966		}
 1967	} else {
 1968		wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
 1969		wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
 1970		wr32(E1000_I210_DTXMXPKTSZ, I210_DTXMXPKTSZ_DEFAULT);
 1971
 1972		val = rd32(E1000_I210_TQAVCTRL);
 1973		/* According to Section 8.12.21, the other flags we've set when
 1974		 * enabling FQTSS are not relevant when disabling FQTSS so we
 1975		 * don't set they here.
 1976		 */
 1977		val &= ~E1000_TQAVCTRL_XMIT_MODE;
 1978		wr32(E1000_I210_TQAVCTRL, val);
 1979	}
 1980
 1981	netdev_dbg(netdev, "FQTSS %s\n", (is_fqtss_enabled(adapter)) ?
 1982		   "enabled" : "disabled");
 1983}
 1984
 1985/**
 1986 *  igb_configure - configure the hardware for RX and TX
 1987 *  @adapter: private board structure
 1988 **/
 1989static void igb_configure(struct igb_adapter *adapter)
 1990{
 1991	struct net_device *netdev = adapter->netdev;
 1992	int i;
 1993
 1994	igb_get_hw_control(adapter);
 1995	igb_set_rx_mode(netdev);
 1996	igb_setup_tx_mode(adapter);
 1997
 1998	igb_restore_vlan(adapter);
 1999
 2000	igb_setup_tctl(adapter);
 2001	igb_setup_mrqc(adapter);
 2002	igb_setup_rctl(adapter);
 2003
 2004	igb_nfc_filter_restore(adapter);
 2005	igb_configure_tx(adapter);
 2006	igb_configure_rx(adapter);
 2007
 2008	igb_rx_fifo_flush_82575(&adapter->hw);
 2009
 2010	/* call igb_desc_unused which always leaves
 2011	 * at least 1 descriptor unused to make sure
 2012	 * next_to_use != next_to_clean
 2013	 */
 2014	for (i = 0; i < adapter->num_rx_queues; i++) {
 2015		struct igb_ring *ring = adapter->rx_ring[i];
 2016		igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
 2017	}
 2018}
 2019
 2020/**
 2021 *  igb_power_up_link - Power up the phy/serdes link
 2022 *  @adapter: address of board private structure
 2023 **/
 2024void igb_power_up_link(struct igb_adapter *adapter)
 2025{
 2026	igb_reset_phy(&adapter->hw);
 2027
 2028	if (adapter->hw.phy.media_type == e1000_media_type_copper)
 2029		igb_power_up_phy_copper(&adapter->hw);
 2030	else
 2031		igb_power_up_serdes_link_82575(&adapter->hw);
 2032
 2033	igb_setup_link(&adapter->hw);
 2034}
 2035
 2036/**
 2037 *  igb_power_down_link - Power down the phy/serdes link
 2038 *  @adapter: address of board private structure
 2039 */
 2040static void igb_power_down_link(struct igb_adapter *adapter)
 2041{
 2042	if (adapter->hw.phy.media_type == e1000_media_type_copper)
 2043		igb_power_down_phy_copper_82575(&adapter->hw);
 2044	else
 2045		igb_shutdown_serdes_link_82575(&adapter->hw);
 2046}
 2047
 2048/**
 2049 * igb_check_swap_media -  Detect and switch function for Media Auto Sense
 2050 * @adapter: address of the board private structure
 2051 **/
 2052static void igb_check_swap_media(struct igb_adapter *adapter)
 2053{
 2054	struct e1000_hw *hw = &adapter->hw;
 2055	u32 ctrl_ext, connsw;
 2056	bool swap_now = false;
 2057
 2058	ctrl_ext = rd32(E1000_CTRL_EXT);
 2059	connsw = rd32(E1000_CONNSW);
 2060
 2061	/* need to live swap if current media is copper and we have fiber/serdes
 2062	 * to go to.
 2063	 */
 2064
 2065	if ((hw->phy.media_type == e1000_media_type_copper) &&
 2066	    (!(connsw & E1000_CONNSW_AUTOSENSE_EN))) {
 2067		swap_now = true;
 2068	} else if ((hw->phy.media_type != e1000_media_type_copper) &&
 2069		   !(connsw & E1000_CONNSW_SERDESD)) {
 2070		/* copper signal takes time to appear */
 2071		if (adapter->copper_tries < 4) {
 2072			adapter->copper_tries++;
 2073			connsw |= E1000_CONNSW_AUTOSENSE_CONF;
 2074			wr32(E1000_CONNSW, connsw);
 2075			return;
 2076		} else {
 2077			adapter->copper_tries = 0;
 2078			if ((connsw & E1000_CONNSW_PHYSD) &&
 2079			    (!(connsw & E1000_CONNSW_PHY_PDN))) {
 2080				swap_now = true;
 2081				connsw &= ~E1000_CONNSW_AUTOSENSE_CONF;
 2082				wr32(E1000_CONNSW, connsw);
 2083			}
 2084		}
 2085	}
 2086
 2087	if (!swap_now)
 2088		return;
 2089
 2090	switch (hw->phy.media_type) {
 2091	case e1000_media_type_copper:
 2092		netdev_info(adapter->netdev,
 2093			"MAS: changing media to fiber/serdes\n");
 2094		ctrl_ext |=
 2095			E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
 2096		adapter->flags |= IGB_FLAG_MEDIA_RESET;
 2097		adapter->copper_tries = 0;
 2098		break;
 2099	case e1000_media_type_internal_serdes:
 2100	case e1000_media_type_fiber:
 2101		netdev_info(adapter->netdev,
 2102			"MAS: changing media to copper\n");
 2103		ctrl_ext &=
 2104			~E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
 2105		adapter->flags |= IGB_FLAG_MEDIA_RESET;
 2106		break;
 2107	default:
 2108		/* shouldn't get here during regular operation */
 2109		netdev_err(adapter->netdev,
 2110			"AMS: Invalid media type found, returning\n");
 2111		break;
 2112	}
 2113	wr32(E1000_CTRL_EXT, ctrl_ext);
 2114}
 2115
 2116/**
 2117 *  igb_up - Open the interface and prepare it to handle traffic
 2118 *  @adapter: board private structure
 2119 **/
 2120int igb_up(struct igb_adapter *adapter)
 2121{
 2122	struct e1000_hw *hw = &adapter->hw;
 2123	int i;
 2124
 2125	/* hardware has been reset, we need to reload some things */
 2126	igb_configure(adapter);
 2127
 2128	clear_bit(__IGB_DOWN, &adapter->state);
 2129
 2130	for (i = 0; i < adapter->num_q_vectors; i++)
 2131		napi_enable(&(adapter->q_vector[i]->napi));
 2132
 2133	if (adapter->flags & IGB_FLAG_HAS_MSIX)
 2134		igb_configure_msix(adapter);
 2135	else
 2136		igb_assign_vector(adapter->q_vector[0], 0);
 2137
 2138	/* Clear any pending interrupts. */
 2139	rd32(E1000_TSICR);
 2140	rd32(E1000_ICR);
 2141	igb_irq_enable(adapter);
 2142
 2143	/* notify VFs that reset has been completed */
 2144	if (adapter->vfs_allocated_count) {
 2145		u32 reg_data = rd32(E1000_CTRL_EXT);
 2146
 2147		reg_data |= E1000_CTRL_EXT_PFRSTD;
 2148		wr32(E1000_CTRL_EXT, reg_data);
 2149	}
 2150
 2151	netif_tx_start_all_queues(adapter->netdev);
 2152
 2153	/* start the watchdog. */
 2154	hw->mac.get_link_status = 1;
 2155	schedule_work(&adapter->watchdog_task);
 2156
 2157	if ((adapter->flags & IGB_FLAG_EEE) &&
 2158	    (!hw->dev_spec._82575.eee_disable))
 2159		adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
 2160
 2161	return 0;
 2162}
 2163
 2164void igb_down(struct igb_adapter *adapter)
 2165{
 2166	struct net_device *netdev = adapter->netdev;
 2167	struct e1000_hw *hw = &adapter->hw;
 2168	u32 tctl, rctl;
 2169	int i;
 2170
 2171	/* signal that we're down so the interrupt handler does not
 2172	 * reschedule our watchdog timer
 2173	 */
 2174	set_bit(__IGB_DOWN, &adapter->state);
 2175
 2176	/* disable receives in the hardware */
 2177	rctl = rd32(E1000_RCTL);
 2178	wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
 2179	/* flush and sleep below */
 2180
 2181	igb_nfc_filter_exit(adapter);
 2182
 2183	netif_carrier_off(netdev);
 2184	netif_tx_stop_all_queues(netdev);
 2185
 2186	/* disable transmits in the hardware */
 2187	tctl = rd32(E1000_TCTL);
 2188	tctl &= ~E1000_TCTL_EN;
 2189	wr32(E1000_TCTL, tctl);
 2190	/* flush both disables and wait for them to finish */
 2191	wrfl();
 2192	usleep_range(10000, 11000);
 2193
 2194	igb_irq_disable(adapter);
 2195
 2196	adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
 2197
 2198	for (i = 0; i < adapter->num_q_vectors; i++) {
 2199		if (adapter->q_vector[i]) {
 2200			napi_synchronize(&adapter->q_vector[i]->napi);
 2201			napi_disable(&adapter->q_vector[i]->napi);
 2202		}
 2203	}
 2204
 
 2205	del_timer_sync(&adapter->watchdog_timer);
 2206	del_timer_sync(&adapter->phy_info_timer);
 2207
 
 
 2208	/* record the stats before reset*/
 2209	spin_lock(&adapter->stats64_lock);
 2210	igb_update_stats(adapter);
 2211	spin_unlock(&adapter->stats64_lock);
 2212
 2213	adapter->link_speed = 0;
 2214	adapter->link_duplex = 0;
 2215
 2216	if (!pci_channel_offline(adapter->pdev))
 2217		igb_reset(adapter);
 2218
 2219	/* clear VLAN promisc flag so VFTA will be updated if necessary */
 2220	adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
 2221
 2222	igb_clean_all_tx_rings(adapter);
 2223	igb_clean_all_rx_rings(adapter);
 2224#ifdef CONFIG_IGB_DCA
 2225
 2226	/* since we reset the hardware DCA settings were cleared */
 2227	igb_setup_dca(adapter);
 2228#endif
 2229}
 2230
 2231void igb_reinit_locked(struct igb_adapter *adapter)
 2232{
 
 2233	while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
 2234		usleep_range(1000, 2000);
 2235	igb_down(adapter);
 2236	igb_up(adapter);
 2237	clear_bit(__IGB_RESETTING, &adapter->state);
 2238}
 2239
 2240/** igb_enable_mas - Media Autosense re-enable after swap
 2241 *
 2242 * @adapter: adapter struct
 2243 **/
 2244static void igb_enable_mas(struct igb_adapter *adapter)
 2245{
 2246	struct e1000_hw *hw = &adapter->hw;
 2247	u32 connsw = rd32(E1000_CONNSW);
 
 
 
 
 
 2248
 2249	/* configure for SerDes media detect */
 2250	if ((hw->phy.media_type == e1000_media_type_copper) &&
 2251	    (!(connsw & E1000_CONNSW_SERDESD))) {
 2252		connsw |= E1000_CONNSW_ENRGSRC;
 2253		connsw |= E1000_CONNSW_AUTOSENSE_EN;
 2254		wr32(E1000_CONNSW, connsw);
 2255		wrfl();
 
 
 
 
 
 
 
 2256	}
 
 2257}
 2258
 2259#ifdef CONFIG_IGB_HWMON
 2260/**
 2261 *  igb_set_i2c_bb - Init I2C interface
 2262 *  @hw: pointer to hardware structure
 2263 **/
 2264static void igb_set_i2c_bb(struct e1000_hw *hw)
 2265{
 2266	u32 ctrl_ext;
 2267	s32 i2cctl;
 2268
 2269	ctrl_ext = rd32(E1000_CTRL_EXT);
 2270	ctrl_ext |= E1000_CTRL_I2C_ENA;
 2271	wr32(E1000_CTRL_EXT, ctrl_ext);
 2272	wrfl();
 2273
 2274	i2cctl = rd32(E1000_I2CPARAMS);
 2275	i2cctl |= E1000_I2CBB_EN
 2276		| E1000_I2C_CLK_OE_N
 2277		| E1000_I2C_DATA_OE_N;
 2278	wr32(E1000_I2CPARAMS, i2cctl);
 2279	wrfl();
 2280}
 2281#endif
 2282
 2283void igb_reset(struct igb_adapter *adapter)
 2284{
 2285	struct pci_dev *pdev = adapter->pdev;
 2286	struct e1000_hw *hw = &adapter->hw;
 2287	struct e1000_mac_info *mac = &hw->mac;
 2288	struct e1000_fc_info *fc = &hw->fc;
 2289	u32 pba, hwm;
 2290
 2291	/* Repartition Pba for greater than 9k mtu
 2292	 * To take effect CTRL.RST is required.
 2293	 */
 2294	switch (mac->type) {
 2295	case e1000_i350:
 2296	case e1000_i354:
 2297	case e1000_82580:
 2298		pba = rd32(E1000_RXPBS);
 2299		pba = igb_rxpbs_adjust_82580(pba);
 2300		break;
 2301	case e1000_82576:
 2302		pba = rd32(E1000_RXPBS);
 2303		pba &= E1000_RXPBS_SIZE_MASK_82576;
 2304		break;
 2305	case e1000_82575:
 2306	case e1000_i210:
 2307	case e1000_i211:
 2308	default:
 2309		pba = E1000_PBA_34K;
 2310		break;
 2311	}
 2312
 2313	if (mac->type == e1000_82575) {
 2314		u32 min_rx_space, min_tx_space, needed_tx_space;
 2315
 2316		/* write Rx PBA so that hardware can report correct Tx PBA */
 2317		wr32(E1000_PBA, pba);
 2318
 2319		/* To maintain wire speed transmits, the Tx FIFO should be
 2320		 * large enough to accommodate two full transmit packets,
 2321		 * rounded up to the next 1KB and expressed in KB.  Likewise,
 2322		 * the Rx FIFO should be large enough to accommodate at least
 2323		 * one full receive packet and is similarly rounded up and
 2324		 * expressed in KB.
 2325		 */
 2326		min_rx_space = DIV_ROUND_UP(MAX_JUMBO_FRAME_SIZE, 1024);
 2327
 2328		/* The Tx FIFO also stores 16 bytes of information about the Tx
 2329		 * but don't include Ethernet FCS because hardware appends it.
 2330		 * We only need to round down to the nearest 512 byte block
 2331		 * count since the value we care about is 2 frames, not 1.
 2332		 */
 2333		min_tx_space = adapter->max_frame_size;
 2334		min_tx_space += sizeof(union e1000_adv_tx_desc) - ETH_FCS_LEN;
 2335		min_tx_space = DIV_ROUND_UP(min_tx_space, 512);
 2336
 2337		/* upper 16 bits has Tx packet buffer allocation size in KB */
 2338		needed_tx_space = min_tx_space - (rd32(E1000_PBA) >> 16);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 2339
 2340		/* If current Tx allocation is less than the min Tx FIFO size,
 2341		 * and the min Tx FIFO size is less than the current Rx FIFO
 2342		 * allocation, take space away from current Rx allocation.
 2343		 */
 2344		if (needed_tx_space < pba) {
 2345			pba -= needed_tx_space;
 
 2346
 2347			/* if short on Rx space, Rx wins and must trump Tx
 2348			 * adjustment
 2349			 */
 2350			if (pba < min_rx_space)
 2351				pba = min_rx_space;
 2352		}
 2353
 2354		/* adjust PBA for jumbo frames */
 2355		wr32(E1000_PBA, pba);
 2356	}
 2357
 2358	/* flow control settings
 2359	 * The high water mark must be low enough to fit one full frame
 2360	 * after transmitting the pause frame.  As such we must have enough
 2361	 * space to allow for us to complete our current transmit and then
 2362	 * receive the frame that is in progress from the link partner.
 2363	 * Set it to:
 2364	 * - the full Rx FIFO size minus one full Tx plus one full Rx frame
 2365	 */
 2366	hwm = (pba << 10) - (adapter->max_frame_size + MAX_JUMBO_FRAME_SIZE);
 
 2367
 2368	fc->high_water = hwm & 0xFFFFFFF0;	/* 16-byte granularity */
 2369	fc->low_water = fc->high_water - 16;
 2370	fc->pause_time = 0xFFFF;
 2371	fc->send_xon = 1;
 2372	fc->current_mode = fc->requested_mode;
 2373
 2374	/* disable receive for all VFs and wait one second */
 2375	if (adapter->vfs_allocated_count) {
 2376		int i;
 2377
 2378		for (i = 0 ; i < adapter->vfs_allocated_count; i++)
 2379			adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
 2380
 2381		/* ping all the active vfs to let them know we are going down */
 2382		igb_ping_all_vfs(adapter);
 2383
 2384		/* disable transmits and receives */
 2385		wr32(E1000_VFRE, 0);
 2386		wr32(E1000_VFTE, 0);
 2387	}
 2388
 2389	/* Allow time for pending master requests to run */
 2390	hw->mac.ops.reset_hw(hw);
 2391	wr32(E1000_WUC, 0);
 2392
 2393	if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
 2394		/* need to resetup here after media swap */
 2395		adapter->ei.get_invariants(hw);
 2396		adapter->flags &= ~IGB_FLAG_MEDIA_RESET;
 2397	}
 2398	if ((mac->type == e1000_82575 || mac->type == e1000_i350) &&
 2399	    (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
 2400		igb_enable_mas(adapter);
 
 2401	}
 2402	if (hw->mac.ops.init_hw(hw))
 2403		dev_err(&pdev->dev, "Hardware Error\n");
 2404
 2405	/* RAR registers were cleared during init_hw, clear mac table */
 2406	igb_flush_mac_table(adapter);
 2407	__dev_uc_unsync(adapter->netdev, NULL);
 2408
 2409	/* Recover default RAR entry */
 2410	igb_set_default_mac_filter(adapter);
 2411
 2412	/* Flow control settings reset on hardware reset, so guarantee flow
 2413	 * control is off when forcing speed.
 2414	 */
 2415	if (!hw->mac.autoneg)
 2416		igb_force_mac_fc(hw);
 2417
 2418	igb_init_dmac(adapter, pba);
 2419#ifdef CONFIG_IGB_HWMON
 2420	/* Re-initialize the thermal sensor on i350 devices. */
 2421	if (!test_bit(__IGB_DOWN, &adapter->state)) {
 2422		if (mac->type == e1000_i350 && hw->bus.func == 0) {
 2423			/* If present, re-initialize the external thermal sensor
 2424			 * interface.
 2425			 */
 2426			if (adapter->ets)
 2427				igb_set_i2c_bb(hw);
 2428			mac->ops.init_thermal_sensor_thresh(hw);
 2429		}
 2430	}
 2431#endif
 2432	/* Re-establish EEE setting */
 2433	if (hw->phy.media_type == e1000_media_type_copper) {
 2434		switch (mac->type) {
 2435		case e1000_i350:
 2436		case e1000_i210:
 2437		case e1000_i211:
 2438			igb_set_eee_i350(hw, true, true);
 2439			break;
 2440		case e1000_i354:
 2441			igb_set_eee_i354(hw, true, true);
 2442			break;
 2443		default:
 2444			break;
 2445		}
 2446	}
 2447	if (!netif_running(adapter->netdev))
 2448		igb_power_down_link(adapter);
 2449
 2450	igb_update_mng_vlan(adapter);
 2451
 2452	/* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
 2453	wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
 2454
 2455	/* Re-enable PTP, where applicable. */
 2456	if (adapter->ptp_flags & IGB_PTP_ENABLED)
 2457		igb_ptp_reset(adapter);
 2458
 2459	igb_get_phy_info(hw);
 2460}
 2461
 2462static netdev_features_t igb_fix_features(struct net_device *netdev,
 2463	netdev_features_t features)
 2464{
 2465	/* Since there is no support for separate Rx/Tx vlan accel
 2466	 * enable/disable make sure Tx flag is always in same state as Rx.
 2467	 */
 2468	if (features & NETIF_F_HW_VLAN_CTAG_RX)
 2469		features |= NETIF_F_HW_VLAN_CTAG_TX;
 2470	else
 2471		features &= ~NETIF_F_HW_VLAN_CTAG_TX;
 2472
 2473	return features;
 2474}
 2475
 2476static int igb_set_features(struct net_device *netdev,
 2477	netdev_features_t features)
 2478{
 2479	netdev_features_t changed = netdev->features ^ features;
 2480	struct igb_adapter *adapter = netdev_priv(netdev);
 2481
 2482	if (changed & NETIF_F_HW_VLAN_CTAG_RX)
 2483		igb_vlan_mode(netdev, features);
 2484
 2485	if (!(changed & (NETIF_F_RXALL | NETIF_F_NTUPLE)))
 2486		return 0;
 2487
 2488	if (!(features & NETIF_F_NTUPLE)) {
 2489		struct hlist_node *node2;
 2490		struct igb_nfc_filter *rule;
 2491
 2492		spin_lock(&adapter->nfc_lock);
 2493		hlist_for_each_entry_safe(rule, node2,
 2494					  &adapter->nfc_filter_list, nfc_node) {
 2495			igb_erase_filter(adapter, rule);
 2496			hlist_del(&rule->nfc_node);
 2497			kfree(rule);
 2498		}
 2499		spin_unlock(&adapter->nfc_lock);
 2500		adapter->nfc_filter_count = 0;
 2501	}
 2502
 2503	netdev->features = features;
 2504
 2505	if (netif_running(netdev))
 2506		igb_reinit_locked(adapter);
 2507	else
 2508		igb_reset(adapter);
 2509
 2510	return 1;
 2511}
 2512
 2513static int igb_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 2514			   struct net_device *dev,
 2515			   const unsigned char *addr, u16 vid,
 2516			   u16 flags,
 2517			   struct netlink_ext_ack *extack)
 2518{
 2519	/* guarantee we can provide a unique filter for the unicast address */
 2520	if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
 2521		struct igb_adapter *adapter = netdev_priv(dev);
 2522		int vfn = adapter->vfs_allocated_count;
 2523
 2524		if (netdev_uc_count(dev) >= igb_available_rars(adapter, vfn))
 2525			return -ENOMEM;
 2526	}
 2527
 2528	return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
 2529}
 2530
 2531#define IGB_MAX_MAC_HDR_LEN	127
 2532#define IGB_MAX_NETWORK_HDR_LEN	511
 2533
 2534static netdev_features_t
 2535igb_features_check(struct sk_buff *skb, struct net_device *dev,
 2536		   netdev_features_t features)
 2537{
 2538	unsigned int network_hdr_len, mac_hdr_len;
 2539
 2540	/* Make certain the headers can be described by a context descriptor */
 2541	mac_hdr_len = skb_network_offset(skb);
 2542	if (unlikely(mac_hdr_len > IGB_MAX_MAC_HDR_LEN))
 2543		return features & ~(NETIF_F_HW_CSUM |
 2544				    NETIF_F_SCTP_CRC |
 2545				    NETIF_F_GSO_UDP_L4 |
 2546				    NETIF_F_HW_VLAN_CTAG_TX |
 2547				    NETIF_F_TSO |
 2548				    NETIF_F_TSO6);
 2549
 2550	network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
 2551	if (unlikely(network_hdr_len >  IGB_MAX_NETWORK_HDR_LEN))
 2552		return features & ~(NETIF_F_HW_CSUM |
 2553				    NETIF_F_SCTP_CRC |
 2554				    NETIF_F_GSO_UDP_L4 |
 2555				    NETIF_F_TSO |
 2556				    NETIF_F_TSO6);
 2557
 2558	/* We can only support IPV4 TSO in tunnels if we can mangle the
 2559	 * inner IP ID field, so strip TSO if MANGLEID is not supported.
 2560	 */
 2561	if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
 2562		features &= ~NETIF_F_TSO;
 2563
 2564	return features;
 2565}
 2566
 2567static void igb_offload_apply(struct igb_adapter *adapter, s32 queue)
 2568{
 2569	if (!is_fqtss_enabled(adapter)) {
 2570		enable_fqtss(adapter, true);
 2571		return;
 2572	}
 2573
 2574	igb_config_tx_modes(adapter, queue);
 2575
 2576	if (!is_any_cbs_enabled(adapter) && !is_any_txtime_enabled(adapter))
 2577		enable_fqtss(adapter, false);
 2578}
 2579
 2580static int igb_offload_cbs(struct igb_adapter *adapter,
 2581			   struct tc_cbs_qopt_offload *qopt)
 2582{
 2583	struct e1000_hw *hw = &adapter->hw;
 2584	int err;
 2585
 2586	/* CBS offloading is only supported by i210 controller. */
 2587	if (hw->mac.type != e1000_i210)
 2588		return -EOPNOTSUPP;
 2589
 2590	/* CBS offloading is only supported by queue 0 and queue 1. */
 2591	if (qopt->queue < 0 || qopt->queue > 1)
 2592		return -EINVAL;
 2593
 2594	err = igb_save_cbs_params(adapter, qopt->queue, qopt->enable,
 2595				  qopt->idleslope, qopt->sendslope,
 2596				  qopt->hicredit, qopt->locredit);
 2597	if (err)
 2598		return err;
 2599
 2600	igb_offload_apply(adapter, qopt->queue);
 2601
 2602	return 0;
 2603}
 2604
 2605#define ETHER_TYPE_FULL_MASK ((__force __be16)~0)
 2606#define VLAN_PRIO_FULL_MASK (0x07)
 2607
 2608static int igb_parse_cls_flower(struct igb_adapter *adapter,
 2609				struct flow_cls_offload *f,
 2610				int traffic_class,
 2611				struct igb_nfc_filter *input)
 2612{
 2613	struct flow_rule *rule = flow_cls_offload_flow_rule(f);
 2614	struct flow_dissector *dissector = rule->match.dissector;
 2615	struct netlink_ext_ack *extack = f->common.extack;
 2616
 2617	if (dissector->used_keys &
 2618	    ~(BIT_ULL(FLOW_DISSECTOR_KEY_BASIC) |
 2619	      BIT_ULL(FLOW_DISSECTOR_KEY_CONTROL) |
 2620	      BIT_ULL(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
 2621	      BIT_ULL(FLOW_DISSECTOR_KEY_VLAN))) {
 2622		NL_SET_ERR_MSG_MOD(extack,
 2623				   "Unsupported key used, only BASIC, CONTROL, ETH_ADDRS and VLAN are supported");
 2624		return -EOPNOTSUPP;
 2625	}
 2626
 2627	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
 2628		struct flow_match_eth_addrs match;
 2629
 2630		flow_rule_match_eth_addrs(rule, &match);
 2631		if (!is_zero_ether_addr(match.mask->dst)) {
 2632			if (!is_broadcast_ether_addr(match.mask->dst)) {
 2633				NL_SET_ERR_MSG_MOD(extack, "Only full masks are supported for destination MAC address");
 2634				return -EINVAL;
 2635			}
 2636
 2637			input->filter.match_flags |=
 2638				IGB_FILTER_FLAG_DST_MAC_ADDR;
 2639			ether_addr_copy(input->filter.dst_addr, match.key->dst);
 2640		}
 2641
 2642		if (!is_zero_ether_addr(match.mask->src)) {
 2643			if (!is_broadcast_ether_addr(match.mask->src)) {
 2644				NL_SET_ERR_MSG_MOD(extack, "Only full masks are supported for source MAC address");
 2645				return -EINVAL;
 2646			}
 2647
 2648			input->filter.match_flags |=
 2649				IGB_FILTER_FLAG_SRC_MAC_ADDR;
 2650			ether_addr_copy(input->filter.src_addr, match.key->src);
 2651		}
 2652	}
 2653
 2654	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
 2655		struct flow_match_basic match;
 2656
 2657		flow_rule_match_basic(rule, &match);
 2658		if (match.mask->n_proto) {
 2659			if (match.mask->n_proto != ETHER_TYPE_FULL_MASK) {
 2660				NL_SET_ERR_MSG_MOD(extack, "Only full mask is supported for EtherType filter");
 2661				return -EINVAL;
 2662			}
 2663
 2664			input->filter.match_flags |= IGB_FILTER_FLAG_ETHER_TYPE;
 2665			input->filter.etype = match.key->n_proto;
 2666		}
 2667	}
 2668
 2669	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN)) {
 2670		struct flow_match_vlan match;
 2671
 2672		flow_rule_match_vlan(rule, &match);
 2673		if (match.mask->vlan_priority) {
 2674			if (match.mask->vlan_priority != VLAN_PRIO_FULL_MASK) {
 2675				NL_SET_ERR_MSG_MOD(extack, "Only full mask is supported for VLAN priority");
 2676				return -EINVAL;
 2677			}
 2678
 2679			input->filter.match_flags |= IGB_FILTER_FLAG_VLAN_TCI;
 2680			input->filter.vlan_tci =
 2681				(__force __be16)match.key->vlan_priority;
 2682		}
 2683	}
 2684
 2685	input->action = traffic_class;
 2686	input->cookie = f->cookie;
 2687
 2688	return 0;
 2689}
 2690
 2691static int igb_configure_clsflower(struct igb_adapter *adapter,
 2692				   struct flow_cls_offload *cls_flower)
 2693{
 2694	struct netlink_ext_ack *extack = cls_flower->common.extack;
 2695	struct igb_nfc_filter *filter, *f;
 2696	int err, tc;
 2697
 2698	tc = tc_classid_to_hwtc(adapter->netdev, cls_flower->classid);
 2699	if (tc < 0) {
 2700		NL_SET_ERR_MSG_MOD(extack, "Invalid traffic class");
 2701		return -EINVAL;
 2702	}
 2703
 2704	filter = kzalloc(sizeof(*filter), GFP_KERNEL);
 2705	if (!filter)
 2706		return -ENOMEM;
 2707
 2708	err = igb_parse_cls_flower(adapter, cls_flower, tc, filter);
 2709	if (err < 0)
 2710		goto err_parse;
 2711
 2712	spin_lock(&adapter->nfc_lock);
 2713
 2714	hlist_for_each_entry(f, &adapter->nfc_filter_list, nfc_node) {
 2715		if (!memcmp(&f->filter, &filter->filter, sizeof(f->filter))) {
 2716			err = -EEXIST;
 2717			NL_SET_ERR_MSG_MOD(extack,
 2718					   "This filter is already set in ethtool");
 2719			goto err_locked;
 2720		}
 2721	}
 2722
 2723	hlist_for_each_entry(f, &adapter->cls_flower_list, nfc_node) {
 2724		if (!memcmp(&f->filter, &filter->filter, sizeof(f->filter))) {
 2725			err = -EEXIST;
 2726			NL_SET_ERR_MSG_MOD(extack,
 2727					   "This filter is already set in cls_flower");
 2728			goto err_locked;
 2729		}
 2730	}
 2731
 2732	err = igb_add_filter(adapter, filter);
 2733	if (err < 0) {
 2734		NL_SET_ERR_MSG_MOD(extack, "Could not add filter to the adapter");
 2735		goto err_locked;
 2736	}
 2737
 2738	hlist_add_head(&filter->nfc_node, &adapter->cls_flower_list);
 2739
 2740	spin_unlock(&adapter->nfc_lock);
 2741
 2742	return 0;
 2743
 2744err_locked:
 2745	spin_unlock(&adapter->nfc_lock);
 2746
 2747err_parse:
 2748	kfree(filter);
 2749
 2750	return err;
 2751}
 2752
 2753static int igb_delete_clsflower(struct igb_adapter *adapter,
 2754				struct flow_cls_offload *cls_flower)
 2755{
 2756	struct igb_nfc_filter *filter;
 2757	int err;
 2758
 2759	spin_lock(&adapter->nfc_lock);
 2760
 2761	hlist_for_each_entry(filter, &adapter->cls_flower_list, nfc_node)
 2762		if (filter->cookie == cls_flower->cookie)
 2763			break;
 2764
 2765	if (!filter) {
 2766		err = -ENOENT;
 2767		goto out;
 2768	}
 2769
 2770	err = igb_erase_filter(adapter, filter);
 2771	if (err < 0)
 2772		goto out;
 2773
 2774	hlist_del(&filter->nfc_node);
 2775	kfree(filter);
 2776
 2777out:
 2778	spin_unlock(&adapter->nfc_lock);
 2779
 2780	return err;
 2781}
 2782
 2783static int igb_setup_tc_cls_flower(struct igb_adapter *adapter,
 2784				   struct flow_cls_offload *cls_flower)
 2785{
 2786	switch (cls_flower->command) {
 2787	case FLOW_CLS_REPLACE:
 2788		return igb_configure_clsflower(adapter, cls_flower);
 2789	case FLOW_CLS_DESTROY:
 2790		return igb_delete_clsflower(adapter, cls_flower);
 2791	case FLOW_CLS_STATS:
 2792		return -EOPNOTSUPP;
 2793	default:
 2794		return -EOPNOTSUPP;
 2795	}
 2796}
 2797
 2798static int igb_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
 2799				 void *cb_priv)
 2800{
 2801	struct igb_adapter *adapter = cb_priv;
 2802
 2803	if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
 2804		return -EOPNOTSUPP;
 2805
 2806	switch (type) {
 2807	case TC_SETUP_CLSFLOWER:
 2808		return igb_setup_tc_cls_flower(adapter, type_data);
 2809
 2810	default:
 2811		return -EOPNOTSUPP;
 2812	}
 2813}
 2814
 2815static int igb_offload_txtime(struct igb_adapter *adapter,
 2816			      struct tc_etf_qopt_offload *qopt)
 2817{
 2818	struct e1000_hw *hw = &adapter->hw;
 2819	int err;
 2820
 2821	/* Launchtime offloading is only supported by i210 controller. */
 2822	if (hw->mac.type != e1000_i210)
 2823		return -EOPNOTSUPP;
 2824
 2825	/* Launchtime offloading is only supported by queues 0 and 1. */
 2826	if (qopt->queue < 0 || qopt->queue > 1)
 2827		return -EINVAL;
 2828
 2829	err = igb_save_txtime_params(adapter, qopt->queue, qopt->enable);
 2830	if (err)
 2831		return err;
 2832
 2833	igb_offload_apply(adapter, qopt->queue);
 2834
 2835	return 0;
 2836}
 2837
 2838static int igb_tc_query_caps(struct igb_adapter *adapter,
 2839			     struct tc_query_caps_base *base)
 2840{
 2841	switch (base->type) {
 2842	case TC_SETUP_QDISC_TAPRIO: {
 2843		struct tc_taprio_caps *caps = base->caps;
 2844
 2845		caps->broken_mqprio = true;
 2846
 2847		return 0;
 2848	}
 2849	default:
 2850		return -EOPNOTSUPP;
 2851	}
 2852}
 2853
 2854static LIST_HEAD(igb_block_cb_list);
 2855
 2856static int igb_setup_tc(struct net_device *dev, enum tc_setup_type type,
 2857			void *type_data)
 2858{
 2859	struct igb_adapter *adapter = netdev_priv(dev);
 2860
 2861	switch (type) {
 2862	case TC_QUERY_CAPS:
 2863		return igb_tc_query_caps(adapter, type_data);
 2864	case TC_SETUP_QDISC_CBS:
 2865		return igb_offload_cbs(adapter, type_data);
 2866	case TC_SETUP_BLOCK:
 2867		return flow_block_cb_setup_simple(type_data,
 2868						  &igb_block_cb_list,
 2869						  igb_setup_tc_block_cb,
 2870						  adapter, adapter, true);
 2871
 2872	case TC_SETUP_QDISC_ETF:
 2873		return igb_offload_txtime(adapter, type_data);
 2874
 2875	default:
 2876		return -EOPNOTSUPP;
 2877	}
 2878}
 2879
 2880static int igb_xdp_setup(struct net_device *dev, struct netdev_bpf *bpf)
 2881{
 2882	int i, frame_size = dev->mtu + IGB_ETH_PKT_HDR_PAD;
 2883	struct igb_adapter *adapter = netdev_priv(dev);
 2884	struct bpf_prog *prog = bpf->prog, *old_prog;
 2885	bool running = netif_running(dev);
 2886	bool need_reset;
 2887
 2888	/* verify igb ring attributes are sufficient for XDP */
 2889	for (i = 0; i < adapter->num_rx_queues; i++) {
 2890		struct igb_ring *ring = adapter->rx_ring[i];
 2891
 2892		if (frame_size > igb_rx_bufsz(ring)) {
 2893			NL_SET_ERR_MSG_MOD(bpf->extack,
 2894					   "The RX buffer size is too small for the frame size");
 2895			netdev_warn(dev, "XDP RX buffer size %d is too small for the frame size %d\n",
 2896				    igb_rx_bufsz(ring), frame_size);
 2897			return -EINVAL;
 2898		}
 2899	}
 2900
 2901	old_prog = xchg(&adapter->xdp_prog, prog);
 2902	need_reset = (!!prog != !!old_prog);
 2903
 2904	/* device is up and bpf is added/removed, must setup the RX queues */
 2905	if (need_reset && running) {
 2906		igb_close(dev);
 2907	} else {
 2908		for (i = 0; i < adapter->num_rx_queues; i++)
 2909			(void)xchg(&adapter->rx_ring[i]->xdp_prog,
 2910			    adapter->xdp_prog);
 2911	}
 2912
 2913	if (old_prog)
 2914		bpf_prog_put(old_prog);
 2915
 2916	/* bpf is just replaced, RXQ and MTU are already setup */
 2917	if (!need_reset) {
 2918		return 0;
 2919	} else {
 2920		if (prog)
 2921			xdp_features_set_redirect_target(dev, true);
 2922		else
 2923			xdp_features_clear_redirect_target(dev);
 2924	}
 2925
 2926	if (running)
 2927		igb_open(dev);
 2928
 2929	return 0;
 2930}
 2931
 2932static int igb_xdp(struct net_device *dev, struct netdev_bpf *xdp)
 2933{
 2934	switch (xdp->command) {
 2935	case XDP_SETUP_PROG:
 2936		return igb_xdp_setup(dev, xdp);
 2937	default:
 2938		return -EINVAL;
 2939	}
 2940}
 2941
 2942static void igb_xdp_ring_update_tail(struct igb_ring *ring)
 2943{
 2944	/* Force memory writes to complete before letting h/w know there
 2945	 * are new descriptors to fetch.
 2946	 */
 2947	wmb();
 2948	writel(ring->next_to_use, ring->tail);
 2949}
 2950
 2951static struct igb_ring *igb_xdp_tx_queue_mapping(struct igb_adapter *adapter)
 2952{
 2953	unsigned int r_idx = smp_processor_id();
 2954
 2955	if (r_idx >= adapter->num_tx_queues)
 2956		r_idx = r_idx % adapter->num_tx_queues;
 2957
 2958	return adapter->tx_ring[r_idx];
 2959}
 2960
 2961static int igb_xdp_xmit_back(struct igb_adapter *adapter, struct xdp_buff *xdp)
 2962{
 2963	struct xdp_frame *xdpf = xdp_convert_buff_to_frame(xdp);
 2964	int cpu = smp_processor_id();
 2965	struct igb_ring *tx_ring;
 2966	struct netdev_queue *nq;
 2967	u32 ret;
 2968
 2969	if (unlikely(!xdpf))
 2970		return IGB_XDP_CONSUMED;
 2971
 2972	/* During program transitions its possible adapter->xdp_prog is assigned
 2973	 * but ring has not been configured yet. In this case simply abort xmit.
 2974	 */
 2975	tx_ring = adapter->xdp_prog ? igb_xdp_tx_queue_mapping(adapter) : NULL;
 2976	if (unlikely(!tx_ring))
 2977		return IGB_XDP_CONSUMED;
 2978
 2979	nq = txring_txq(tx_ring);
 2980	__netif_tx_lock(nq, cpu);
 2981	/* Avoid transmit queue timeout since we share it with the slow path */
 2982	txq_trans_cond_update(nq);
 2983	ret = igb_xmit_xdp_ring(adapter, tx_ring, xdpf);
 2984	__netif_tx_unlock(nq);
 2985
 2986	return ret;
 2987}
 2988
 2989static int igb_xdp_xmit(struct net_device *dev, int n,
 2990			struct xdp_frame **frames, u32 flags)
 2991{
 2992	struct igb_adapter *adapter = netdev_priv(dev);
 2993	int cpu = smp_processor_id();
 2994	struct igb_ring *tx_ring;
 2995	struct netdev_queue *nq;
 2996	int nxmit = 0;
 2997	int i;
 2998
 2999	if (unlikely(test_bit(__IGB_DOWN, &adapter->state)))
 3000		return -ENETDOWN;
 3001
 3002	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
 3003		return -EINVAL;
 3004
 3005	/* During program transitions its possible adapter->xdp_prog is assigned
 3006	 * but ring has not been configured yet. In this case simply abort xmit.
 3007	 */
 3008	tx_ring = adapter->xdp_prog ? igb_xdp_tx_queue_mapping(adapter) : NULL;
 3009	if (unlikely(!tx_ring))
 3010		return -ENXIO;
 3011
 3012	nq = txring_txq(tx_ring);
 3013	__netif_tx_lock(nq, cpu);
 3014
 3015	/* Avoid transmit queue timeout since we share it with the slow path */
 3016	txq_trans_cond_update(nq);
 3017
 3018	for (i = 0; i < n; i++) {
 3019		struct xdp_frame *xdpf = frames[i];
 3020		int err;
 3021
 3022		err = igb_xmit_xdp_ring(adapter, tx_ring, xdpf);
 3023		if (err != IGB_XDP_TX)
 3024			break;
 3025		nxmit++;
 3026	}
 3027
 3028	__netif_tx_unlock(nq);
 3029
 3030	if (unlikely(flags & XDP_XMIT_FLUSH))
 3031		igb_xdp_ring_update_tail(tx_ring);
 3032
 3033	return nxmit;
 3034}
 3035
 3036static const struct net_device_ops igb_netdev_ops = {
 3037	.ndo_open		= igb_open,
 3038	.ndo_stop		= igb_close,
 3039	.ndo_start_xmit		= igb_xmit_frame,
 3040	.ndo_get_stats64	= igb_get_stats64,
 3041	.ndo_set_rx_mode	= igb_set_rx_mode,
 3042	.ndo_set_mac_address	= igb_set_mac,
 3043	.ndo_change_mtu		= igb_change_mtu,
 3044	.ndo_eth_ioctl		= igb_ioctl,
 3045	.ndo_tx_timeout		= igb_tx_timeout,
 3046	.ndo_validate_addr	= eth_validate_addr,
 3047	.ndo_vlan_rx_add_vid	= igb_vlan_rx_add_vid,
 3048	.ndo_vlan_rx_kill_vid	= igb_vlan_rx_kill_vid,
 3049	.ndo_set_vf_mac		= igb_ndo_set_vf_mac,
 3050	.ndo_set_vf_vlan	= igb_ndo_set_vf_vlan,
 3051	.ndo_set_vf_rate	= igb_ndo_set_vf_bw,
 3052	.ndo_set_vf_spoofchk	= igb_ndo_set_vf_spoofchk,
 3053	.ndo_set_vf_trust	= igb_ndo_set_vf_trust,
 3054	.ndo_get_vf_config	= igb_ndo_get_vf_config,
 
 
 
 3055	.ndo_fix_features	= igb_fix_features,
 3056	.ndo_set_features	= igb_set_features,
 3057	.ndo_fdb_add		= igb_ndo_fdb_add,
 3058	.ndo_features_check	= igb_features_check,
 3059	.ndo_setup_tc		= igb_setup_tc,
 3060	.ndo_bpf		= igb_xdp,
 3061	.ndo_xdp_xmit		= igb_xdp_xmit,
 3062};
 3063
 3064/**
 3065 * igb_set_fw_version - Configure version string for ethtool
 3066 * @adapter: adapter struct
 3067 **/
 3068void igb_set_fw_version(struct igb_adapter *adapter)
 3069{
 3070	struct e1000_hw *hw = &adapter->hw;
 3071	struct e1000_fw_version fw;
 3072
 3073	igb_get_fw_version(hw, &fw);
 3074
 3075	switch (hw->mac.type) {
 3076	case e1000_i210:
 3077	case e1000_i211:
 3078		if (!(igb_get_flash_presence_i210(hw))) {
 3079			snprintf(adapter->fw_version,
 3080				 sizeof(adapter->fw_version),
 3081				 "%2d.%2d-%d",
 3082				 fw.invm_major, fw.invm_minor,
 3083				 fw.invm_img_type);
 3084			break;
 3085		}
 3086		fallthrough;
 3087	default:
 3088		/* if option rom is valid, display its version too */
 3089		if (fw.or_valid) {
 3090			snprintf(adapter->fw_version,
 3091				 sizeof(adapter->fw_version),
 3092				 "%d.%d, 0x%08x, %d.%d.%d",
 3093				 fw.eep_major, fw.eep_minor, fw.etrack_id,
 3094				 fw.or_major, fw.or_build, fw.or_patch);
 3095		/* no option rom */
 3096		} else if (fw.etrack_id != 0X0000) {
 3097			snprintf(adapter->fw_version,
 3098				 sizeof(adapter->fw_version),
 3099				 "%d.%d, 0x%08x",
 3100				 fw.eep_major, fw.eep_minor, fw.etrack_id);
 3101		} else {
 3102			snprintf(adapter->fw_version,
 3103				 sizeof(adapter->fw_version),
 3104				 "%d.%d.%d",
 3105				 fw.eep_major, fw.eep_minor, fw.eep_build);
 3106		}
 3107		break;
 3108	}
 
 3109}
 3110
 3111/**
 3112 * igb_init_mas - init Media Autosense feature if enabled in the NVM
 3113 *
 3114 * @adapter: adapter struct
 3115 **/
 3116static void igb_init_mas(struct igb_adapter *adapter)
 3117{
 3118	struct e1000_hw *hw = &adapter->hw;
 3119	u16 eeprom_data;
 3120
 3121	hw->nvm.ops.read(hw, NVM_COMPAT, 1, &eeprom_data);
 3122	switch (hw->bus.func) {
 3123	case E1000_FUNC_0:
 3124		if (eeprom_data & IGB_MAS_ENABLE_0) {
 3125			adapter->flags |= IGB_FLAG_MAS_ENABLE;
 3126			netdev_info(adapter->netdev,
 3127				"MAS: Enabling Media Autosense for port %d\n",
 3128				hw->bus.func);
 3129		}
 3130		break;
 3131	case E1000_FUNC_1:
 3132		if (eeprom_data & IGB_MAS_ENABLE_1) {
 3133			adapter->flags |= IGB_FLAG_MAS_ENABLE;
 3134			netdev_info(adapter->netdev,
 3135				"MAS: Enabling Media Autosense for port %d\n",
 3136				hw->bus.func);
 3137		}
 3138		break;
 3139	case E1000_FUNC_2:
 3140		if (eeprom_data & IGB_MAS_ENABLE_2) {
 3141			adapter->flags |= IGB_FLAG_MAS_ENABLE;
 3142			netdev_info(adapter->netdev,
 3143				"MAS: Enabling Media Autosense for port %d\n",
 3144				hw->bus.func);
 3145		}
 3146		break;
 3147	case E1000_FUNC_3:
 3148		if (eeprom_data & IGB_MAS_ENABLE_3) {
 3149			adapter->flags |= IGB_FLAG_MAS_ENABLE;
 3150			netdev_info(adapter->netdev,
 3151				"MAS: Enabling Media Autosense for port %d\n",
 3152				hw->bus.func);
 3153		}
 3154		break;
 3155	default:
 3156		/* Shouldn't get here */
 3157		netdev_err(adapter->netdev,
 3158			"MAS: Invalid port configuration, returning\n");
 3159		break;
 3160	}
 3161}
 3162
 3163/**
 3164 *  igb_init_i2c - Init I2C interface
 3165 *  @adapter: pointer to adapter structure
 3166 **/
 3167static s32 igb_init_i2c(struct igb_adapter *adapter)
 3168{
 3169	s32 status = 0;
 3170
 3171	/* I2C interface supported on i350 devices */
 3172	if (adapter->hw.mac.type != e1000_i350)
 3173		return 0;
 3174
 3175	/* Initialize the i2c bus which is controlled by the registers.
 3176	 * This bus will use the i2c_algo_bit structure that implements
 3177	 * the protocol through toggling of the 4 bits in the register.
 3178	 */
 3179	adapter->i2c_adap.owner = THIS_MODULE;
 3180	adapter->i2c_algo = igb_i2c_algo;
 3181	adapter->i2c_algo.data = adapter;
 3182	adapter->i2c_adap.algo_data = &adapter->i2c_algo;
 3183	adapter->i2c_adap.dev.parent = &adapter->pdev->dev;
 3184	strscpy(adapter->i2c_adap.name, "igb BB",
 3185		sizeof(adapter->i2c_adap.name));
 3186	status = i2c_bit_add_bus(&adapter->i2c_adap);
 3187	return status;
 3188}
 3189
 3190/**
 3191 *  igb_probe - Device Initialization Routine
 3192 *  @pdev: PCI device information struct
 3193 *  @ent: entry in igb_pci_tbl
 3194 *
 3195 *  Returns 0 on success, negative on failure
 3196 *
 3197 *  igb_probe initializes an adapter identified by a pci_dev structure.
 3198 *  The OS initialization, configuring of the adapter private structure,
 3199 *  and a hardware reset occur.
 3200 **/
 3201static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 3202{
 3203	struct net_device *netdev;
 3204	struct igb_adapter *adapter;
 3205	struct e1000_hw *hw;
 3206	u16 eeprom_data = 0;
 3207	s32 ret_val;
 3208	static int global_quad_port_a; /* global quad port a indication */
 3209	const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
 
 3210	u8 part_str[E1000_PBANUM_LENGTH];
 3211	int err;
 3212
 3213	/* Catch broken hardware that put the wrong VF device ID in
 3214	 * the PCIe SR-IOV capability.
 3215	 */
 3216	if (pdev->is_virtfn) {
 3217		WARN(1, KERN_ERR "%s (%x:%x) should not be a VF!\n",
 3218			pci_name(pdev), pdev->vendor, pdev->device);
 3219		return -EINVAL;
 3220	}
 3221
 3222	err = pci_enable_device_mem(pdev);
 3223	if (err)
 3224		return err;
 3225
 
 3226	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
 3227	if (err) {
 3228		dev_err(&pdev->dev,
 3229			"No usable DMA configuration, aborting\n");
 3230		goto err_dma;
 
 
 
 
 
 3231	}
 3232
 3233	err = pci_request_mem_regions(pdev, igb_driver_name);
 
 
 3234	if (err)
 3235		goto err_pci_reg;
 3236
 
 
 3237	pci_set_master(pdev);
 3238	pci_save_state(pdev);
 3239
 3240	err = -ENOMEM;
 3241	netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
 3242				   IGB_MAX_TX_QUEUES);
 3243	if (!netdev)
 3244		goto err_alloc_etherdev;
 3245
 3246	SET_NETDEV_DEV(netdev, &pdev->dev);
 3247
 3248	pci_set_drvdata(pdev, netdev);
 3249	adapter = netdev_priv(netdev);
 3250	adapter->netdev = netdev;
 3251	adapter->pdev = pdev;
 3252	hw = &adapter->hw;
 3253	hw->back = adapter;
 3254	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
 3255
 3256	err = -EIO;
 3257	adapter->io_addr = pci_iomap(pdev, 0, 0);
 3258	if (!adapter->io_addr)
 3259		goto err_ioremap;
 3260	/* hw->hw_addr can be altered, we'll use adapter->io_addr for unmap */
 3261	hw->hw_addr = adapter->io_addr;
 3262
 3263	netdev->netdev_ops = &igb_netdev_ops;
 3264	igb_set_ethtool_ops(netdev);
 3265	netdev->watchdog_timeo = 5 * HZ;
 3266
 3267	strscpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
 3268
 3269	netdev->mem_start = pci_resource_start(pdev, 0);
 3270	netdev->mem_end = pci_resource_end(pdev, 0);
 3271
 3272	/* PCI config space info */
 3273	hw->vendor_id = pdev->vendor;
 3274	hw->device_id = pdev->device;
 3275	hw->revision_id = pdev->revision;
 3276	hw->subsystem_vendor_id = pdev->subsystem_vendor;
 3277	hw->subsystem_device_id = pdev->subsystem_device;
 3278
 3279	/* Copy the default MAC, PHY and NVM function pointers */
 3280	memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
 3281	memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
 3282	memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
 3283	/* Initialize skew-specific constants */
 3284	err = ei->get_invariants(hw);
 3285	if (err)
 3286		goto err_sw_init;
 3287
 3288	/* setup the private structure */
 3289	err = igb_sw_init(adapter);
 3290	if (err)
 3291		goto err_sw_init;
 3292
 3293	igb_get_bus_info_pcie(hw);
 3294
 3295	hw->phy.autoneg_wait_to_complete = false;
 3296
 3297	/* Copper options */
 3298	if (hw->phy.media_type == e1000_media_type_copper) {
 3299		hw->phy.mdix = AUTO_ALL_MODES;
 3300		hw->phy.disable_polarity_correction = false;
 3301		hw->phy.ms_type = e1000_ms_hw_default;
 3302	}
 3303
 3304	if (igb_check_reset_block(hw))
 3305		dev_info(&pdev->dev,
 3306			"PHY reset is blocked due to SOL/IDER session.\n");
 3307
 3308	/* features is initialized to 0 in allocation, it might have bits
 3309	 * set by igb_sw_init so we should use an or instead of an
 3310	 * assignment.
 3311	 */
 3312	netdev->features |= NETIF_F_SG |
 
 
 3313			    NETIF_F_TSO |
 3314			    NETIF_F_TSO6 |
 3315			    NETIF_F_RXHASH |
 3316			    NETIF_F_RXCSUM |
 3317			    NETIF_F_HW_CSUM;
 3318
 3319	if (hw->mac.type >= e1000_82576)
 3320		netdev->features |= NETIF_F_SCTP_CRC | NETIF_F_GSO_UDP_L4;
 3321
 3322	if (hw->mac.type >= e1000_i350)
 3323		netdev->features |= NETIF_F_HW_TC;
 3324
 3325#define IGB_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
 3326				  NETIF_F_GSO_GRE_CSUM | \
 3327				  NETIF_F_GSO_IPXIP4 | \
 3328				  NETIF_F_GSO_IPXIP6 | \
 3329				  NETIF_F_GSO_UDP_TUNNEL | \
 3330				  NETIF_F_GSO_UDP_TUNNEL_CSUM)
 3331
 3332	netdev->gso_partial_features = IGB_GSO_PARTIAL_FEATURES;
 3333	netdev->features |= NETIF_F_GSO_PARTIAL | IGB_GSO_PARTIAL_FEATURES;
 3334
 3335	/* copy netdev features into list of user selectable features */
 3336	netdev->hw_features |= netdev->features |
 3337			       NETIF_F_HW_VLAN_CTAG_RX |
 3338			       NETIF_F_HW_VLAN_CTAG_TX |
 3339			       NETIF_F_RXALL;
 3340
 3341	if (hw->mac.type >= e1000_i350)
 3342		netdev->hw_features |= NETIF_F_NTUPLE;
 3343
 3344	netdev->features |= NETIF_F_HIGHDMA;
 
 
 
 
 3345
 3346	netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
 3347	netdev->mpls_features |= NETIF_F_HW_CSUM;
 3348	netdev->hw_enc_features |= netdev->vlan_features;
 3349
 3350	/* set this bit last since it cannot be part of vlan_features */
 3351	netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
 3352			    NETIF_F_HW_VLAN_CTAG_RX |
 3353			    NETIF_F_HW_VLAN_CTAG_TX;
 3354
 3355	netdev->priv_flags |= IFF_SUPP_NOFCS;
 
 
 
 3356
 3357	netdev->priv_flags |= IFF_UNICAST_FLT;
 3358	netdev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT;
 3359
 3360	/* MTU range: 68 - 9216 */
 3361	netdev->min_mtu = ETH_MIN_MTU;
 3362	netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
 3363
 3364	adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
 3365
 3366	/* before reading the NVM, reset the controller to put the device in a
 3367	 * known good starting state
 3368	 */
 3369	hw->mac.ops.reset_hw(hw);
 3370
 3371	/* make sure the NVM is good , i211/i210 parts can have special NVM
 3372	 * that doesn't contain a checksum
 3373	 */
 3374	switch (hw->mac.type) {
 3375	case e1000_i210:
 3376	case e1000_i211:
 3377		if (igb_get_flash_presence_i210(hw)) {
 3378			if (hw->nvm.ops.validate(hw) < 0) {
 3379				dev_err(&pdev->dev,
 3380					"The NVM Checksum Is Not Valid\n");
 3381				err = -EIO;
 3382				goto err_eeprom;
 3383			}
 3384		}
 3385		break;
 3386	default:
 3387		if (hw->nvm.ops.validate(hw) < 0) {
 3388			dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
 3389			err = -EIO;
 3390			goto err_eeprom;
 3391		}
 3392		break;
 3393	}
 3394
 3395	if (eth_platform_get_mac_address(&pdev->dev, hw->mac.addr)) {
 3396		/* copy the MAC address out of the NVM */
 3397		if (hw->mac.ops.read_mac_addr(hw))
 3398			dev_err(&pdev->dev, "NVM Read Error\n");
 3399	}
 3400
 3401	eth_hw_addr_set(netdev, hw->mac.addr);
 3402
 3403	if (!is_valid_ether_addr(netdev->dev_addr)) {
 3404		dev_err(&pdev->dev, "Invalid MAC Address\n");
 3405		err = -EIO;
 3406		goto err_eeprom;
 3407	}
 3408
 3409	igb_set_default_mac_filter(adapter);
 3410
 3411	/* get firmware version for ethtool -i */
 3412	igb_set_fw_version(adapter);
 3413
 3414	/* configure RXPBSIZE and TXPBSIZE */
 3415	if (hw->mac.type == e1000_i210) {
 3416		wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
 3417		wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
 3418	}
 3419
 3420	timer_setup(&adapter->watchdog_timer, igb_watchdog, 0);
 3421	timer_setup(&adapter->phy_info_timer, igb_update_phy_info, 0);
 3422
 3423	INIT_WORK(&adapter->reset_task, igb_reset_task);
 3424	INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
 3425
 3426	/* Initialize link properties that are user-changeable */
 3427	adapter->fc_autoneg = true;
 3428	hw->mac.autoneg = true;
 3429	hw->phy.autoneg_advertised = 0x2f;
 3430
 3431	hw->fc.requested_mode = e1000_fc_default;
 3432	hw->fc.current_mode = e1000_fc_default;
 3433
 3434	igb_validate_mdi_setting(hw);
 3435
 3436	/* By default, support wake on port A */
 3437	if (hw->bus.func == 0)
 3438		adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
 3439
 3440	/* Check the NVM for wake support on non-port A ports */
 3441	if (hw->mac.type >= e1000_82580)
 3442		hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
 3443				 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
 3444				 &eeprom_data);
 3445	else if (hw->bus.func == 1)
 3446		hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
 3447
 3448	if (eeprom_data & IGB_EEPROM_APME)
 3449		adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
 3450
 3451	/* now that we have the eeprom settings, apply the special cases where
 3452	 * the eeprom may be wrong or the board simply won't support wake on
 3453	 * lan on a particular port
 3454	 */
 3455	switch (pdev->device) {
 3456	case E1000_DEV_ID_82575GB_QUAD_COPPER:
 3457		adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
 3458		break;
 3459	case E1000_DEV_ID_82575EB_FIBER_SERDES:
 3460	case E1000_DEV_ID_82576_FIBER:
 3461	case E1000_DEV_ID_82576_SERDES:
 3462		/* Wake events only supported on port A for dual fiber
 3463		 * regardless of eeprom setting
 3464		 */
 3465		if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
 3466			adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
 3467		break;
 3468	case E1000_DEV_ID_82576_QUAD_COPPER:
 3469	case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
 3470		/* if quad port adapter, disable WoL on all but port A */
 3471		if (global_quad_port_a != 0)
 3472			adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
 3473		else
 3474			adapter->flags |= IGB_FLAG_QUAD_PORT_A;
 3475		/* Reset for multiple quad port adapters */
 3476		if (++global_quad_port_a == 4)
 3477			global_quad_port_a = 0;
 3478		break;
 3479	default:
 3480		/* If the device can't wake, don't set software support */
 3481		if (!device_can_wakeup(&adapter->pdev->dev))
 3482			adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
 3483	}
 3484
 3485	/* initialize the wol settings based on the eeprom settings */
 3486	if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
 3487		adapter->wol |= E1000_WUFC_MAG;
 3488
 3489	/* Some vendors want WoL disabled by default, but still supported */
 3490	if ((hw->mac.type == e1000_i350) &&
 3491	    (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
 3492		adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
 3493		adapter->wol = 0;
 3494	}
 3495
 3496	/* Some vendors want the ability to Use the EEPROM setting as
 3497	 * enable/disable only, and not for capability
 3498	 */
 3499	if (((hw->mac.type == e1000_i350) ||
 3500	     (hw->mac.type == e1000_i354)) &&
 3501	    (pdev->subsystem_vendor == PCI_VENDOR_ID_DELL)) {
 3502		adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
 3503		adapter->wol = 0;
 3504	}
 3505	if (hw->mac.type == e1000_i350) {
 3506		if (((pdev->subsystem_device == 0x5001) ||
 3507		     (pdev->subsystem_device == 0x5002)) &&
 3508				(hw->bus.func == 0)) {
 3509			adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
 3510			adapter->wol = 0;
 3511		}
 3512		if (pdev->subsystem_device == 0x1F52)
 3513			adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
 3514	}
 3515
 3516	device_set_wakeup_enable(&adapter->pdev->dev,
 3517				 adapter->flags & IGB_FLAG_WOL_SUPPORTED);
 3518
 3519	/* reset the hardware with the new settings */
 3520	igb_reset(adapter);
 3521
 3522	/* Init the I2C interface */
 3523	err = igb_init_i2c(adapter);
 3524	if (err) {
 3525		dev_err(&pdev->dev, "failed to init i2c interface\n");
 3526		goto err_eeprom;
 3527	}
 3528
 3529	/* let the f/w know that the h/w is now under the control of the
 3530	 * driver.
 3531	 */
 3532	igb_get_hw_control(adapter);
 3533
 3534	strcpy(netdev->name, "eth%d");
 3535	err = register_netdev(netdev);
 3536	if (err)
 3537		goto err_register;
 3538
 3539	/* carrier off reporting is important to ethtool even BEFORE open */
 3540	netif_carrier_off(netdev);
 3541
 3542#ifdef CONFIG_IGB_DCA
 3543	if (dca_add_requester(&pdev->dev) == 0) {
 3544		adapter->flags |= IGB_FLAG_DCA_ENABLED;
 3545		dev_info(&pdev->dev, "DCA enabled\n");
 3546		igb_setup_dca(adapter);
 3547	}
 3548
 3549#endif
 3550#ifdef CONFIG_IGB_HWMON
 3551	/* Initialize the thermal sensor on i350 devices. */
 3552	if (hw->mac.type == e1000_i350 && hw->bus.func == 0) {
 3553		u16 ets_word;
 3554
 3555		/* Read the NVM to determine if this i350 device supports an
 3556		 * external thermal sensor.
 3557		 */
 3558		hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word);
 3559		if (ets_word != 0x0000 && ets_word != 0xFFFF)
 3560			adapter->ets = true;
 3561		else
 3562			adapter->ets = false;
 3563		/* Only enable I2C bit banging if an external thermal
 3564		 * sensor is supported.
 3565		 */
 3566		if (adapter->ets)
 3567			igb_set_i2c_bb(hw);
 3568		hw->mac.ops.init_thermal_sensor_thresh(hw);
 3569		if (igb_sysfs_init(adapter))
 3570			dev_err(&pdev->dev,
 3571				"failed to allocate sysfs resources\n");
 3572	} else {
 3573		adapter->ets = false;
 3574	}
 3575#endif
 3576	/* Check if Media Autosense is enabled */
 3577	adapter->ei = *ei;
 3578	if (hw->dev_spec._82575.mas_capable)
 3579		igb_init_mas(adapter);
 3580
 3581	/* do hw tstamp init after resetting */
 3582	igb_ptp_init(adapter);
 3583
 3584	dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
 3585	/* print bus type/speed/width info, not applicable to i354 */
 3586	if (hw->mac.type != e1000_i354) {
 3587		dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
 3588			 netdev->name,
 3589			 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
 3590			  (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
 3591			   "unknown"),
 3592			 ((hw->bus.width == e1000_bus_width_pcie_x4) ?
 3593			  "Width x4" :
 3594			  (hw->bus.width == e1000_bus_width_pcie_x2) ?
 3595			  "Width x2" :
 3596			  (hw->bus.width == e1000_bus_width_pcie_x1) ?
 3597			  "Width x1" : "unknown"), netdev->dev_addr);
 3598	}
 3599
 3600	if ((hw->mac.type == e1000_82576 &&
 3601	     rd32(E1000_EECD) & E1000_EECD_PRES) ||
 3602	    (hw->mac.type >= e1000_i210 ||
 3603	     igb_get_flash_presence_i210(hw))) {
 3604		ret_val = igb_read_part_string(hw, part_str,
 3605					       E1000_PBANUM_LENGTH);
 3606	} else {
 3607		ret_val = -E1000_ERR_INVM_VALUE_NOT_FOUND;
 3608	}
 3609
 3610	if (ret_val)
 3611		strcpy(part_str, "Unknown");
 3612	dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
 3613	dev_info(&pdev->dev,
 3614		"Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
 3615		(adapter->flags & IGB_FLAG_HAS_MSIX) ? "MSI-X" :
 3616		(adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
 3617		adapter->num_rx_queues, adapter->num_tx_queues);
 3618	if (hw->phy.media_type == e1000_media_type_copper) {
 3619		switch (hw->mac.type) {
 3620		case e1000_i350:
 3621		case e1000_i210:
 3622		case e1000_i211:
 3623			/* Enable EEE for internal copper PHY devices */
 3624			err = igb_set_eee_i350(hw, true, true);
 3625			if ((!err) &&
 3626			    (!hw->dev_spec._82575.eee_disable)) {
 3627				adapter->eee_advert =
 3628					MDIO_EEE_100TX | MDIO_EEE_1000T;
 3629				adapter->flags |= IGB_FLAG_EEE;
 3630			}
 3631			break;
 3632		case e1000_i354:
 3633			if ((rd32(E1000_CTRL_EXT) &
 3634			    E1000_CTRL_EXT_LINK_MODE_SGMII)) {
 3635				err = igb_set_eee_i354(hw, true, true);
 3636				if ((!err) &&
 3637					(!hw->dev_spec._82575.eee_disable)) {
 3638					adapter->eee_advert =
 3639					   MDIO_EEE_100TX | MDIO_EEE_1000T;
 3640					adapter->flags |= IGB_FLAG_EEE;
 3641				}
 3642			}
 3643			break;
 3644		default:
 3645			break;
 3646		}
 3647	}
 3648
 3649	dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_NO_DIRECT_COMPLETE);
 3650
 3651	pm_runtime_put_noidle(&pdev->dev);
 3652	return 0;
 3653
 3654err_register:
 3655	igb_release_hw_control(adapter);
 3656	memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap));
 3657err_eeprom:
 3658	if (!igb_check_reset_block(hw))
 3659		igb_reset_phy(hw);
 3660
 3661	if (hw->flash_address)
 3662		iounmap(hw->flash_address);
 3663err_sw_init:
 3664	kfree(adapter->mac_table);
 3665	kfree(adapter->shadow_vfta);
 3666	igb_clear_interrupt_scheme(adapter);
 3667#ifdef CONFIG_PCI_IOV
 3668	igb_disable_sriov(pdev, false);
 3669#endif
 3670	pci_iounmap(pdev, adapter->io_addr);
 3671err_ioremap:
 3672	free_netdev(netdev);
 3673err_alloc_etherdev:
 3674	pci_release_mem_regions(pdev);
 
 3675err_pci_reg:
 3676err_dma:
 3677	pci_disable_device(pdev);
 3678	return err;
 3679}
 3680
 3681#ifdef CONFIG_PCI_IOV
 3682static int igb_sriov_reinit(struct pci_dev *dev)
 3683{
 3684	struct net_device *netdev = pci_get_drvdata(dev);
 3685	struct igb_adapter *adapter = netdev_priv(netdev);
 3686	struct pci_dev *pdev = adapter->pdev;
 3687
 3688	rtnl_lock();
 3689
 3690	if (netif_running(netdev))
 3691		igb_close(netdev);
 3692	else
 3693		igb_reset(adapter);
 3694
 3695	igb_clear_interrupt_scheme(adapter);
 3696
 3697	igb_init_queue_configuration(adapter);
 3698
 3699	if (igb_init_interrupt_scheme(adapter, true)) {
 3700		rtnl_unlock();
 3701		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
 3702		return -ENOMEM;
 3703	}
 3704
 3705	if (netif_running(netdev))
 3706		igb_open(netdev);
 3707
 3708	rtnl_unlock();
 3709
 3710	return 0;
 3711}
 3712
 3713static int igb_disable_sriov(struct pci_dev *pdev, bool reinit)
 3714{
 3715	struct net_device *netdev = pci_get_drvdata(pdev);
 3716	struct igb_adapter *adapter = netdev_priv(netdev);
 3717	struct e1000_hw *hw = &adapter->hw;
 3718	unsigned long flags;
 3719
 3720	/* reclaim resources allocated to VFs */
 3721	if (adapter->vf_data) {
 3722		/* disable iov and allow time for transactions to clear */
 3723		if (pci_vfs_assigned(pdev)) {
 3724			dev_warn(&pdev->dev,
 3725				 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
 3726			return -EPERM;
 3727		} else {
 3728			pci_disable_sriov(pdev);
 3729			msleep(500);
 3730		}
 3731		spin_lock_irqsave(&adapter->vfs_lock, flags);
 3732		kfree(adapter->vf_mac_list);
 3733		adapter->vf_mac_list = NULL;
 3734		kfree(adapter->vf_data);
 3735		adapter->vf_data = NULL;
 3736		adapter->vfs_allocated_count = 0;
 3737		spin_unlock_irqrestore(&adapter->vfs_lock, flags);
 3738		wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
 3739		wrfl();
 3740		msleep(100);
 3741		dev_info(&pdev->dev, "IOV Disabled\n");
 3742
 3743		/* Re-enable DMA Coalescing flag since IOV is turned off */
 3744		adapter->flags |= IGB_FLAG_DMAC;
 3745	}
 3746
 3747	return reinit ? igb_sriov_reinit(pdev) : 0;
 3748}
 3749
 3750static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs, bool reinit)
 3751{
 3752	struct net_device *netdev = pci_get_drvdata(pdev);
 3753	struct igb_adapter *adapter = netdev_priv(netdev);
 3754	int old_vfs = pci_num_vf(pdev);
 3755	struct vf_mac_filter *mac_list;
 3756	int err = 0;
 3757	int num_vf_mac_filters, i;
 3758
 3759	if (!(adapter->flags & IGB_FLAG_HAS_MSIX) || num_vfs > 7) {
 3760		err = -EPERM;
 3761		goto out;
 3762	}
 3763	if (!num_vfs)
 3764		goto out;
 3765
 3766	if (old_vfs) {
 3767		dev_info(&pdev->dev, "%d pre-allocated VFs found - override max_vfs setting of %d\n",
 3768			 old_vfs, max_vfs);
 3769		adapter->vfs_allocated_count = old_vfs;
 3770	} else
 3771		adapter->vfs_allocated_count = num_vfs;
 3772
 3773	adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
 3774				sizeof(struct vf_data_storage), GFP_KERNEL);
 3775
 3776	/* if allocation failed then we do not support SR-IOV */
 3777	if (!adapter->vf_data) {
 3778		adapter->vfs_allocated_count = 0;
 
 
 3779		err = -ENOMEM;
 3780		goto out;
 3781	}
 3782
 3783	/* Due to the limited number of RAR entries calculate potential
 3784	 * number of MAC filters available for the VFs. Reserve entries
 3785	 * for PF default MAC, PF MAC filters and at least one RAR entry
 3786	 * for each VF for VF MAC.
 3787	 */
 3788	num_vf_mac_filters = adapter->hw.mac.rar_entry_count -
 3789			     (1 + IGB_PF_MAC_FILTERS_RESERVED +
 3790			      adapter->vfs_allocated_count);
 3791
 3792	adapter->vf_mac_list = kcalloc(num_vf_mac_filters,
 3793				       sizeof(struct vf_mac_filter),
 3794				       GFP_KERNEL);
 3795
 3796	mac_list = adapter->vf_mac_list;
 3797	INIT_LIST_HEAD(&adapter->vf_macs.l);
 3798
 3799	if (adapter->vf_mac_list) {
 3800		/* Initialize list of VF MAC filters */
 3801		for (i = 0; i < num_vf_mac_filters; i++) {
 3802			mac_list->vf = -1;
 3803			mac_list->free = true;
 3804			list_add(&mac_list->l, &adapter->vf_macs.l);
 3805			mac_list++;
 3806		}
 3807	} else {
 3808		/* If we could not allocate memory for the VF MAC filters
 3809		 * we can continue without this feature but warn user.
 3810		 */
 3811		dev_err(&pdev->dev,
 3812			"Unable to allocate memory for VF MAC filter list\n");
 3813	}
 3814
 3815	dev_info(&pdev->dev, "%d VFs allocated\n",
 3816		 adapter->vfs_allocated_count);
 3817	for (i = 0; i < adapter->vfs_allocated_count; i++)
 3818		igb_vf_configure(adapter, i);
 3819
 3820	/* DMA Coalescing is not supported in IOV mode. */
 3821	adapter->flags &= ~IGB_FLAG_DMAC;
 3822
 3823	if (reinit) {
 3824		err = igb_sriov_reinit(pdev);
 3825		if (err)
 3826			goto err_out;
 3827	}
 3828
 3829	/* only call pci_enable_sriov() if no VFs are allocated already */
 3830	if (!old_vfs) {
 3831		err = pci_enable_sriov(pdev, adapter->vfs_allocated_count);
 3832		if (err)
 3833			goto err_out;
 3834	}
 3835
 3836	goto out;
 3837
 3838err_out:
 3839	kfree(adapter->vf_mac_list);
 3840	adapter->vf_mac_list = NULL;
 3841	kfree(adapter->vf_data);
 3842	adapter->vf_data = NULL;
 3843	adapter->vfs_allocated_count = 0;
 3844out:
 3845	return err;
 3846}
 3847
 3848#endif
 3849/**
 3850 *  igb_remove_i2c - Cleanup  I2C interface
 3851 *  @adapter: pointer to adapter structure
 3852 **/
 3853static void igb_remove_i2c(struct igb_adapter *adapter)
 3854{
 3855	/* free the adapter bus structure */
 3856	i2c_del_adapter(&adapter->i2c_adap);
 3857}
 3858
 3859/**
 3860 *  igb_remove - Device Removal Routine
 3861 *  @pdev: PCI device information struct
 3862 *
 3863 *  igb_remove is called by the PCI subsystem to alert the driver
 3864 *  that it should release a PCI device.  The could be caused by a
 3865 *  Hot-Plug event, or because the driver is going to be removed from
 3866 *  memory.
 3867 **/
 3868static void igb_remove(struct pci_dev *pdev)
 3869{
 3870	struct net_device *netdev = pci_get_drvdata(pdev);
 3871	struct igb_adapter *adapter = netdev_priv(netdev);
 3872	struct e1000_hw *hw = &adapter->hw;
 3873
 3874	pm_runtime_get_noresume(&pdev->dev);
 3875#ifdef CONFIG_IGB_HWMON
 3876	igb_sysfs_exit(adapter);
 3877#endif
 3878	igb_remove_i2c(adapter);
 3879	igb_ptp_stop(adapter);
 3880	/* The watchdog timer may be rescheduled, so explicitly
 3881	 * disable watchdog from being rescheduled.
 3882	 */
 3883	set_bit(__IGB_DOWN, &adapter->state);
 3884	del_timer_sync(&adapter->watchdog_timer);
 3885	del_timer_sync(&adapter->phy_info_timer);
 3886
 3887	cancel_work_sync(&adapter->reset_task);
 3888	cancel_work_sync(&adapter->watchdog_task);
 3889
 3890#ifdef CONFIG_IGB_DCA
 3891	if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
 3892		dev_info(&pdev->dev, "DCA disabled\n");
 3893		dca_remove_requester(&pdev->dev);
 3894		adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
 3895		wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
 3896	}
 3897#endif
 3898
 3899	/* Release control of h/w to f/w.  If f/w is AMT enabled, this
 3900	 * would have already happened in close and is redundant.
 3901	 */
 3902	igb_release_hw_control(adapter);
 3903
 3904#ifdef CONFIG_PCI_IOV
 3905	igb_disable_sriov(pdev, false);
 3906#endif
 3907
 3908	unregister_netdev(netdev);
 3909
 3910	igb_clear_interrupt_scheme(adapter);
 3911
 3912	pci_iounmap(pdev, adapter->io_addr);
 
 
 
 
 3913	if (hw->flash_address)
 3914		iounmap(hw->flash_address);
 3915	pci_release_mem_regions(pdev);
 
 3916
 3917	kfree(adapter->mac_table);
 3918	kfree(adapter->shadow_vfta);
 3919	free_netdev(netdev);
 3920
 
 
 3921	pci_disable_device(pdev);
 3922}
 3923
 3924/**
 3925 *  igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
 3926 *  @adapter: board private structure to initialize
 3927 *
 3928 *  This function initializes the vf specific data storage and then attempts to
 3929 *  allocate the VFs.  The reason for ordering it this way is because it is much
 3930 *  mor expensive time wise to disable SR-IOV than it is to allocate and free
 3931 *  the memory for the VFs.
 3932 **/
 3933static void igb_probe_vfs(struct igb_adapter *adapter)
 3934{
 3935#ifdef CONFIG_PCI_IOV
 3936	struct pci_dev *pdev = adapter->pdev;
 3937	struct e1000_hw *hw = &adapter->hw;
 3938
 3939	/* Virtualization features not supported on i210 and 82580 family. */
 3940	if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211) ||
 3941	    (hw->mac.type == e1000_82580))
 3942		return;
 3943
 3944	/* Of the below we really only want the effect of getting
 3945	 * IGB_FLAG_HAS_MSIX set (if available), without which
 3946	 * igb_enable_sriov() has no effect.
 3947	 */
 3948	igb_set_interrupt_capability(adapter, true);
 3949	igb_reset_interrupt_capability(adapter);
 3950
 3951	pci_sriov_set_totalvfs(pdev, 7);
 3952	igb_enable_sriov(pdev, max_vfs, false);
 3953
 3954#endif /* CONFIG_PCI_IOV */
 3955}
 3956
 3957unsigned int igb_get_max_rss_queues(struct igb_adapter *adapter)
 3958{
 3959	struct e1000_hw *hw = &adapter->hw;
 3960	unsigned int max_rss_queues;
 3961
 3962	/* Determine the maximum number of RSS queues supported. */
 3963	switch (hw->mac.type) {
 3964	case e1000_i211:
 3965		max_rss_queues = IGB_MAX_RX_QUEUES_I211;
 3966		break;
 3967	case e1000_82575:
 3968	case e1000_i210:
 3969		max_rss_queues = IGB_MAX_RX_QUEUES_82575;
 3970		break;
 3971	case e1000_i350:
 3972		/* I350 cannot do RSS and SR-IOV at the same time */
 3973		if (!!adapter->vfs_allocated_count) {
 3974			max_rss_queues = 1;
 3975			break;
 3976		}
 3977		fallthrough;
 3978	case e1000_82576:
 3979		if (!!adapter->vfs_allocated_count) {
 3980			max_rss_queues = 2;
 3981			break;
 3982		}
 3983		fallthrough;
 3984	case e1000_82580:
 3985	case e1000_i354:
 3986	default:
 3987		max_rss_queues = IGB_MAX_RX_QUEUES;
 3988		break;
 3989	}
 3990
 3991	return max_rss_queues;
 3992}
 3993
 3994static void igb_init_queue_configuration(struct igb_adapter *adapter)
 3995{
 3996	u32 max_rss_queues;
 3997
 3998	max_rss_queues = igb_get_max_rss_queues(adapter);
 3999	adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
 4000
 4001	igb_set_flag_queue_pairs(adapter, max_rss_queues);
 4002}
 4003
 4004void igb_set_flag_queue_pairs(struct igb_adapter *adapter,
 4005			      const u32 max_rss_queues)
 4006{
 4007	struct e1000_hw *hw = &adapter->hw;
 4008
 4009	/* Determine if we need to pair queues. */
 4010	switch (hw->mac.type) {
 4011	case e1000_82575:
 4012	case e1000_i211:
 4013		/* Device supports enough interrupts without queue pairing. */
 4014		break;
 4015	case e1000_82576:
 
 
 
 
 
 
 
 
 4016	case e1000_82580:
 4017	case e1000_i350:
 4018	case e1000_i354:
 4019	case e1000_i210:
 4020	default:
 4021		/* If rss_queues > half of max_rss_queues, pair the queues in
 4022		 * order to conserve interrupts due to limited supply.
 4023		 */
 4024		if (adapter->rss_queues > (max_rss_queues / 2))
 4025			adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
 4026		else
 4027			adapter->flags &= ~IGB_FLAG_QUEUE_PAIRS;
 4028		break;
 4029	}
 4030}
 4031
 4032/**
 4033 *  igb_sw_init - Initialize general software structures (struct igb_adapter)
 4034 *  @adapter: board private structure to initialize
 4035 *
 4036 *  igb_sw_init initializes the Adapter private data structure.
 4037 *  Fields are initialized based on PCI device information and
 4038 *  OS network device settings (MTU size).
 4039 **/
 4040static int igb_sw_init(struct igb_adapter *adapter)
 4041{
 4042	struct e1000_hw *hw = &adapter->hw;
 4043	struct net_device *netdev = adapter->netdev;
 4044	struct pci_dev *pdev = adapter->pdev;
 4045
 4046	pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
 4047
 4048	/* set default ring sizes */
 4049	adapter->tx_ring_count = IGB_DEFAULT_TXD;
 4050	adapter->rx_ring_count = IGB_DEFAULT_RXD;
 4051
 4052	/* set default ITR values */
 4053	adapter->rx_itr_setting = IGB_DEFAULT_ITR;
 4054	adapter->tx_itr_setting = IGB_DEFAULT_ITR;
 4055
 4056	/* set default work limits */
 4057	adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
 4058
 4059	adapter->max_frame_size = netdev->mtu + IGB_ETH_PKT_HDR_PAD;
 
 4060	adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
 4061
 4062	spin_lock_init(&adapter->nfc_lock);
 4063	spin_lock_init(&adapter->stats64_lock);
 4064
 4065	/* init spinlock to avoid concurrency of VF resources */
 4066	spin_lock_init(&adapter->vfs_lock);
 4067#ifdef CONFIG_PCI_IOV
 4068	switch (hw->mac.type) {
 4069	case e1000_82576:
 4070	case e1000_i350:
 4071		if (max_vfs > 7) {
 4072			dev_warn(&pdev->dev,
 4073				 "Maximum of 7 VFs per PF, using max\n");
 4074			max_vfs = adapter->vfs_allocated_count = 7;
 4075		} else
 4076			adapter->vfs_allocated_count = max_vfs;
 4077		if (adapter->vfs_allocated_count)
 4078			dev_warn(&pdev->dev,
 4079				 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
 4080		break;
 4081	default:
 4082		break;
 4083	}
 4084#endif /* CONFIG_PCI_IOV */
 4085
 4086	/* Assume MSI-X interrupts, will be checked during IRQ allocation */
 4087	adapter->flags |= IGB_FLAG_HAS_MSIX;
 4088
 4089	adapter->mac_table = kcalloc(hw->mac.rar_entry_count,
 4090				     sizeof(struct igb_mac_addr),
 4091				     GFP_KERNEL);
 4092	if (!adapter->mac_table)
 4093		return -ENOMEM;
 4094
 4095	igb_probe_vfs(adapter);
 4096
 4097	igb_init_queue_configuration(adapter);
 4098
 4099	/* Setup and initialize a copy of the hw vlan table array */
 4100	adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32),
 4101				       GFP_KERNEL);
 4102	if (!adapter->shadow_vfta)
 4103		return -ENOMEM;
 4104
 4105	/* This call may decrease the number of queues */
 4106	if (igb_init_interrupt_scheme(adapter, true)) {
 4107		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
 4108		return -ENOMEM;
 4109	}
 4110
 
 
 4111	/* Explicitly disable IRQ since the NIC can be in any state. */
 4112	igb_irq_disable(adapter);
 4113
 4114	if (hw->mac.type >= e1000_i350)
 4115		adapter->flags &= ~IGB_FLAG_DMAC;
 4116
 4117	set_bit(__IGB_DOWN, &adapter->state);
 4118	return 0;
 4119}
 4120
 4121/**
 4122 *  __igb_open - Called when a network interface is made active
 4123 *  @netdev: network interface device structure
 4124 *  @resuming: indicates whether we are in a resume call
 4125 *
 4126 *  Returns 0 on success, negative value on failure
 4127 *
 4128 *  The open entry point is called when a network interface is made
 4129 *  active by the system (IFF_UP).  At this point all resources needed
 4130 *  for transmit and receive operations are allocated, the interrupt
 4131 *  handler is registered with the OS, the watchdog timer is started,
 4132 *  and the stack is notified that the interface is ready.
 4133 **/
 4134static int __igb_open(struct net_device *netdev, bool resuming)
 4135{
 4136	struct igb_adapter *adapter = netdev_priv(netdev);
 4137	struct e1000_hw *hw = &adapter->hw;
 4138	struct pci_dev *pdev = adapter->pdev;
 4139	int err;
 4140	int i;
 4141
 4142	/* disallow open during test */
 4143	if (test_bit(__IGB_TESTING, &adapter->state)) {
 4144		WARN_ON(resuming);
 4145		return -EBUSY;
 4146	}
 4147
 4148	if (!resuming)
 4149		pm_runtime_get_sync(&pdev->dev);
 4150
 4151	netif_carrier_off(netdev);
 4152
 4153	/* allocate transmit descriptors */
 4154	err = igb_setup_all_tx_resources(adapter);
 4155	if (err)
 4156		goto err_setup_tx;
 4157
 4158	/* allocate receive descriptors */
 4159	err = igb_setup_all_rx_resources(adapter);
 4160	if (err)
 4161		goto err_setup_rx;
 4162
 4163	igb_power_up_link(adapter);
 4164
 4165	/* before we allocate an interrupt, we must be ready to handle it.
 4166	 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
 4167	 * as soon as we call pci_request_irq, so we have to setup our
 4168	 * clean_rx handler before we do so.
 4169	 */
 4170	igb_configure(adapter);
 4171
 4172	err = igb_request_irq(adapter);
 4173	if (err)
 4174		goto err_req_irq;
 4175
 4176	/* Notify the stack of the actual queue counts. */
 4177	err = netif_set_real_num_tx_queues(adapter->netdev,
 4178					   adapter->num_tx_queues);
 4179	if (err)
 4180		goto err_set_queues;
 4181
 4182	err = netif_set_real_num_rx_queues(adapter->netdev,
 4183					   adapter->num_rx_queues);
 4184	if (err)
 4185		goto err_set_queues;
 4186
 4187	/* From here on the code is the same as igb_up() */
 4188	clear_bit(__IGB_DOWN, &adapter->state);
 4189
 4190	for (i = 0; i < adapter->num_q_vectors; i++)
 4191		napi_enable(&(adapter->q_vector[i]->napi));
 4192
 4193	/* Clear any pending interrupts. */
 4194	rd32(E1000_TSICR);
 4195	rd32(E1000_ICR);
 4196
 4197	igb_irq_enable(adapter);
 4198
 4199	/* notify VFs that reset has been completed */
 4200	if (adapter->vfs_allocated_count) {
 4201		u32 reg_data = rd32(E1000_CTRL_EXT);
 4202
 4203		reg_data |= E1000_CTRL_EXT_PFRSTD;
 4204		wr32(E1000_CTRL_EXT, reg_data);
 4205	}
 4206
 4207	netif_tx_start_all_queues(netdev);
 4208
 4209	if (!resuming)
 4210		pm_runtime_put(&pdev->dev);
 4211
 4212	/* start the watchdog. */
 4213	hw->mac.get_link_status = 1;
 4214	schedule_work(&adapter->watchdog_task);
 4215
 4216	return 0;
 4217
 4218err_set_queues:
 4219	igb_free_irq(adapter);
 4220err_req_irq:
 4221	igb_release_hw_control(adapter);
 4222	igb_power_down_link(adapter);
 4223	igb_free_all_rx_resources(adapter);
 4224err_setup_rx:
 4225	igb_free_all_tx_resources(adapter);
 4226err_setup_tx:
 4227	igb_reset(adapter);
 4228	if (!resuming)
 4229		pm_runtime_put(&pdev->dev);
 4230
 4231	return err;
 4232}
 4233
 4234int igb_open(struct net_device *netdev)
 4235{
 4236	return __igb_open(netdev, false);
 4237}
 4238
 4239/**
 4240 *  __igb_close - Disables a network interface
 4241 *  @netdev: network interface device structure
 4242 *  @suspending: indicates we are in a suspend call
 4243 *
 4244 *  Returns 0, this is not allowed to fail
 4245 *
 4246 *  The close entry point is called when an interface is de-activated
 4247 *  by the OS.  The hardware is still under the driver's control, but
 4248 *  needs to be disabled.  A global MAC reset is issued to stop the
 4249 *  hardware, and all transmit and receive resources are freed.
 4250 **/
 4251static int __igb_close(struct net_device *netdev, bool suspending)
 4252{
 4253	struct igb_adapter *adapter = netdev_priv(netdev);
 4254	struct pci_dev *pdev = adapter->pdev;
 4255
 4256	WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
 4257
 4258	if (!suspending)
 4259		pm_runtime_get_sync(&pdev->dev);
 4260
 4261	igb_down(adapter);
 4262	igb_free_irq(adapter);
 4263
 4264	igb_free_all_tx_resources(adapter);
 4265	igb_free_all_rx_resources(adapter);
 4266
 4267	if (!suspending)
 4268		pm_runtime_put_sync(&pdev->dev);
 4269	return 0;
 4270}
 4271
 4272int igb_close(struct net_device *netdev)
 4273{
 4274	if (netif_device_present(netdev) || netdev->dismantle)
 4275		return __igb_close(netdev, false);
 4276	return 0;
 4277}
 4278
 4279/**
 4280 *  igb_setup_tx_resources - allocate Tx resources (Descriptors)
 4281 *  @tx_ring: tx descriptor ring (for a specific queue) to setup
 4282 *
 4283 *  Return 0 on success, negative on failure
 4284 **/
 4285int igb_setup_tx_resources(struct igb_ring *tx_ring)
 4286{
 4287	struct device *dev = tx_ring->dev;
 4288	int size;
 4289
 4290	size = sizeof(struct igb_tx_buffer) * tx_ring->count;
 4291
 4292	tx_ring->tx_buffer_info = vmalloc(size);
 4293	if (!tx_ring->tx_buffer_info)
 4294		goto err;
 4295
 4296	/* round up to nearest 4K */
 4297	tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
 4298	tx_ring->size = ALIGN(tx_ring->size, 4096);
 4299
 4300	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
 4301					   &tx_ring->dma, GFP_KERNEL);
 4302	if (!tx_ring->desc)
 4303		goto err;
 4304
 4305	tx_ring->next_to_use = 0;
 4306	tx_ring->next_to_clean = 0;
 4307
 4308	return 0;
 4309
 4310err:
 4311	vfree(tx_ring->tx_buffer_info);
 4312	tx_ring->tx_buffer_info = NULL;
 4313	dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
 4314	return -ENOMEM;
 4315}
 4316
 4317/**
 4318 *  igb_setup_all_tx_resources - wrapper to allocate Tx resources
 4319 *				 (Descriptors) for all queues
 4320 *  @adapter: board private structure
 4321 *
 4322 *  Return 0 on success, negative on failure
 4323 **/
 4324static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
 4325{
 4326	struct pci_dev *pdev = adapter->pdev;
 4327	int i, err = 0;
 4328
 4329	for (i = 0; i < adapter->num_tx_queues; i++) {
 4330		err = igb_setup_tx_resources(adapter->tx_ring[i]);
 4331		if (err) {
 4332			dev_err(&pdev->dev,
 4333				"Allocation for Tx Queue %u failed\n", i);
 4334			for (i--; i >= 0; i--)
 4335				igb_free_tx_resources(adapter->tx_ring[i]);
 4336			break;
 4337		}
 4338	}
 4339
 4340	return err;
 4341}
 4342
 4343/**
 4344 *  igb_setup_tctl - configure the transmit control registers
 4345 *  @adapter: Board private structure
 4346 **/
 4347void igb_setup_tctl(struct igb_adapter *adapter)
 4348{
 4349	struct e1000_hw *hw = &adapter->hw;
 4350	u32 tctl;
 4351
 4352	/* disable queue 0 which is enabled by default on 82575 and 82576 */
 4353	wr32(E1000_TXDCTL(0), 0);
 4354
 4355	/* Program the Transmit Control Register */
 4356	tctl = rd32(E1000_TCTL);
 4357	tctl &= ~E1000_TCTL_CT;
 4358	tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
 4359		(E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
 4360
 4361	igb_config_collision_dist(hw);
 4362
 4363	/* Enable transmits */
 4364	tctl |= E1000_TCTL_EN;
 4365
 4366	wr32(E1000_TCTL, tctl);
 4367}
 4368
 4369/**
 4370 *  igb_configure_tx_ring - Configure transmit ring after Reset
 4371 *  @adapter: board private structure
 4372 *  @ring: tx ring to configure
 4373 *
 4374 *  Configure a transmit ring after a reset.
 4375 **/
 4376void igb_configure_tx_ring(struct igb_adapter *adapter,
 4377			   struct igb_ring *ring)
 4378{
 4379	struct e1000_hw *hw = &adapter->hw;
 4380	u32 txdctl = 0;
 4381	u64 tdba = ring->dma;
 4382	int reg_idx = ring->reg_idx;
 4383
 
 
 
 
 
 4384	wr32(E1000_TDLEN(reg_idx),
 4385	     ring->count * sizeof(union e1000_adv_tx_desc));
 4386	wr32(E1000_TDBAL(reg_idx),
 4387	     tdba & 0x00000000ffffffffULL);
 4388	wr32(E1000_TDBAH(reg_idx), tdba >> 32);
 4389
 4390	ring->tail = adapter->io_addr + E1000_TDT(reg_idx);
 4391	wr32(E1000_TDH(reg_idx), 0);
 4392	writel(0, ring->tail);
 4393
 4394	txdctl |= IGB_TX_PTHRESH;
 4395	txdctl |= IGB_TX_HTHRESH << 8;
 4396	txdctl |= IGB_TX_WTHRESH << 16;
 4397
 4398	/* reinitialize tx_buffer_info */
 4399	memset(ring->tx_buffer_info, 0,
 4400	       sizeof(struct igb_tx_buffer) * ring->count);
 4401
 4402	txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
 4403	wr32(E1000_TXDCTL(reg_idx), txdctl);
 4404}
 4405
 4406/**
 4407 *  igb_configure_tx - Configure transmit Unit after Reset
 4408 *  @adapter: board private structure
 4409 *
 4410 *  Configure the Tx unit of the MAC after a reset.
 4411 **/
 4412static void igb_configure_tx(struct igb_adapter *adapter)
 4413{
 4414	struct e1000_hw *hw = &adapter->hw;
 4415	int i;
 4416
 4417	/* disable the queues */
 4418	for (i = 0; i < adapter->num_tx_queues; i++)
 4419		wr32(E1000_TXDCTL(adapter->tx_ring[i]->reg_idx), 0);
 4420
 4421	wrfl();
 4422	usleep_range(10000, 20000);
 4423
 4424	for (i = 0; i < adapter->num_tx_queues; i++)
 4425		igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
 4426}
 4427
 4428/**
 4429 *  igb_setup_rx_resources - allocate Rx resources (Descriptors)
 4430 *  @rx_ring: Rx descriptor ring (for a specific queue) to setup
 4431 *
 4432 *  Returns 0 on success, negative on failure
 4433 **/
 4434int igb_setup_rx_resources(struct igb_ring *rx_ring)
 4435{
 4436	struct igb_adapter *adapter = netdev_priv(rx_ring->netdev);
 4437	struct device *dev = rx_ring->dev;
 4438	int size, res;
 4439
 4440	/* XDP RX-queue info */
 4441	if (xdp_rxq_info_is_reg(&rx_ring->xdp_rxq))
 4442		xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 4443	res = xdp_rxq_info_reg(&rx_ring->xdp_rxq, rx_ring->netdev,
 4444			       rx_ring->queue_index, 0);
 4445	if (res < 0) {
 4446		dev_err(dev, "Failed to register xdp_rxq index %u\n",
 4447			rx_ring->queue_index);
 4448		return res;
 4449	}
 4450
 4451	size = sizeof(struct igb_rx_buffer) * rx_ring->count;
 4452
 4453	rx_ring->rx_buffer_info = vmalloc(size);
 4454	if (!rx_ring->rx_buffer_info)
 4455		goto err;
 4456
 4457	/* Round up to nearest 4K */
 4458	rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
 4459	rx_ring->size = ALIGN(rx_ring->size, 4096);
 4460
 4461	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
 4462					   &rx_ring->dma, GFP_KERNEL);
 4463	if (!rx_ring->desc)
 4464		goto err;
 4465
 4466	rx_ring->next_to_alloc = 0;
 4467	rx_ring->next_to_clean = 0;
 4468	rx_ring->next_to_use = 0;
 4469
 4470	rx_ring->xdp_prog = adapter->xdp_prog;
 4471
 4472	return 0;
 4473
 4474err:
 4475	xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 4476	vfree(rx_ring->rx_buffer_info);
 4477	rx_ring->rx_buffer_info = NULL;
 4478	dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
 4479	return -ENOMEM;
 4480}
 4481
 4482/**
 4483 *  igb_setup_all_rx_resources - wrapper to allocate Rx resources
 4484 *				 (Descriptors) for all queues
 4485 *  @adapter: board private structure
 4486 *
 4487 *  Return 0 on success, negative on failure
 4488 **/
 4489static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
 4490{
 4491	struct pci_dev *pdev = adapter->pdev;
 4492	int i, err = 0;
 4493
 4494	for (i = 0; i < adapter->num_rx_queues; i++) {
 4495		err = igb_setup_rx_resources(adapter->rx_ring[i]);
 4496		if (err) {
 4497			dev_err(&pdev->dev,
 4498				"Allocation for Rx Queue %u failed\n", i);
 4499			for (i--; i >= 0; i--)
 4500				igb_free_rx_resources(adapter->rx_ring[i]);
 4501			break;
 4502		}
 4503	}
 4504
 4505	return err;
 4506}
 4507
 4508/**
 4509 *  igb_setup_mrqc - configure the multiple receive queue control registers
 4510 *  @adapter: Board private structure
 4511 **/
 4512static void igb_setup_mrqc(struct igb_adapter *adapter)
 4513{
 4514	struct e1000_hw *hw = &adapter->hw;
 4515	u32 mrqc, rxcsum;
 4516	u32 j, num_rx_queues;
 4517	u32 rss_key[10];
 
 
 
 4518
 4519	netdev_rss_key_fill(rss_key, sizeof(rss_key));
 4520	for (j = 0; j < 10; j++)
 4521		wr32(E1000_RSSRK(j), rss_key[j]);
 4522
 4523	num_rx_queues = adapter->rss_queues;
 4524
 4525	switch (hw->mac.type) {
 4526	case e1000_82576:
 4527		/* 82576 supports 2 RSS queues for SR-IOV */
 4528		if (adapter->vfs_allocated_count)
 4529			num_rx_queues = 2;
 4530		break;
 4531	default:
 4532		break;
 4533	}
 4534
 4535	if (adapter->rss_indir_tbl_init != num_rx_queues) {
 4536		for (j = 0; j < IGB_RETA_SIZE; j++)
 4537			adapter->rss_indir_tbl[j] =
 4538			(j * num_rx_queues) / IGB_RETA_SIZE;
 4539		adapter->rss_indir_tbl_init = num_rx_queues;
 4540	}
 4541	igb_write_rss_indir_tbl(adapter);
 4542
 4543	/* Disable raw packet checksumming so that RSS hash is placed in
 4544	 * descriptor on writeback.  No need to enable TCP/UDP/IP checksum
 4545	 * offloads as they are enabled by default
 4546	 */
 4547	rxcsum = rd32(E1000_RXCSUM);
 4548	rxcsum |= E1000_RXCSUM_PCSD;
 4549
 4550	if (adapter->hw.mac.type >= e1000_82576)
 4551		/* Enable Receive Checksum Offload for SCTP */
 4552		rxcsum |= E1000_RXCSUM_CRCOFL;
 4553
 4554	/* Don't need to set TUOFL or IPOFL, they default to 1 */
 4555	wr32(E1000_RXCSUM, rxcsum);
 4556
 4557	/* Generate RSS hash based on packet types, TCP/UDP
 4558	 * port numbers and/or IPv4/v6 src and dst addresses
 4559	 */
 4560	mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
 4561	       E1000_MRQC_RSS_FIELD_IPV4_TCP |
 4562	       E1000_MRQC_RSS_FIELD_IPV6 |
 4563	       E1000_MRQC_RSS_FIELD_IPV6_TCP |
 4564	       E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
 4565
 4566	if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
 4567		mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
 4568	if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
 4569		mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
 4570
 4571	/* If VMDq is enabled then we set the appropriate mode for that, else
 4572	 * we default to RSS so that an RSS hash is calculated per packet even
 4573	 * if we are only using one queue
 4574	 */
 4575	if (adapter->vfs_allocated_count) {
 4576		if (hw->mac.type > e1000_82575) {
 4577			/* Set the default pool for the PF's first queue */
 4578			u32 vtctl = rd32(E1000_VT_CTL);
 4579
 4580			vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
 4581				   E1000_VT_CTL_DISABLE_DEF_POOL);
 4582			vtctl |= adapter->vfs_allocated_count <<
 4583				E1000_VT_CTL_DEFAULT_POOL_SHIFT;
 4584			wr32(E1000_VT_CTL, vtctl);
 4585		}
 4586		if (adapter->rss_queues > 1)
 4587			mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_MQ;
 4588		else
 4589			mrqc |= E1000_MRQC_ENABLE_VMDQ;
 4590	} else {
 4591		mrqc |= E1000_MRQC_ENABLE_RSS_MQ;
 
 4592	}
 4593	igb_vmm_control(adapter);
 4594
 4595	wr32(E1000_MRQC, mrqc);
 4596}
 4597
 4598/**
 4599 *  igb_setup_rctl - configure the receive control registers
 4600 *  @adapter: Board private structure
 4601 **/
 4602void igb_setup_rctl(struct igb_adapter *adapter)
 4603{
 4604	struct e1000_hw *hw = &adapter->hw;
 4605	u32 rctl;
 4606
 4607	rctl = rd32(E1000_RCTL);
 4608
 4609	rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
 4610	rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
 4611
 4612	rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
 4613		(hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
 4614
 4615	/* enable stripping of CRC. It's unlikely this will break BMC
 4616	 * redirection as it did with e1000. Newer features require
 4617	 * that the HW strips the CRC.
 4618	 */
 4619	rctl |= E1000_RCTL_SECRC;
 4620
 4621	/* disable store bad packets and clear size bits. */
 4622	rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
 4623
 4624	/* enable LPE to allow for reception of jumbo frames */
 4625	rctl |= E1000_RCTL_LPE;
 4626
 4627	/* disable queue 0 to prevent tail write w/o re-config */
 4628	wr32(E1000_RXDCTL(0), 0);
 4629
 4630	/* Attention!!!  For SR-IOV PF driver operations you must enable
 4631	 * queue drop for all VF and PF queues to prevent head of line blocking
 4632	 * if an un-trusted VF does not provide descriptors to hardware.
 4633	 */
 4634	if (adapter->vfs_allocated_count) {
 4635		/* set all queue drop enable bits */
 4636		wr32(E1000_QDE, ALL_QUEUES);
 4637	}
 4638
 4639	/* This is useful for sniffing bad packets. */
 4640	if (adapter->netdev->features & NETIF_F_RXALL) {
 4641		/* UPE and MPE will be handled by normal PROMISC logic
 4642		 * in e1000e_set_rx_mode
 4643		 */
 4644		rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
 4645			 E1000_RCTL_BAM | /* RX All Bcast Pkts */
 4646			 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
 4647
 4648		rctl &= ~(E1000_RCTL_DPF | /* Allow filtered pause */
 
 4649			  E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
 4650		/* Do not mess with E1000_CTRL_VME, it affects transmit as well,
 4651		 * and that breaks VLANs.
 4652		 */
 4653	}
 4654
 4655	wr32(E1000_RCTL, rctl);
 4656}
 4657
 4658static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
 4659				   int vfn)
 4660{
 4661	struct e1000_hw *hw = &adapter->hw;
 4662	u32 vmolr;
 4663
 4664	if (size > MAX_JUMBO_FRAME_SIZE)
 4665		size = MAX_JUMBO_FRAME_SIZE;
 
 
 
 
 4666
 4667	vmolr = rd32(E1000_VMOLR(vfn));
 4668	vmolr &= ~E1000_VMOLR_RLPML_MASK;
 4669	vmolr |= size | E1000_VMOLR_LPE;
 4670	wr32(E1000_VMOLR(vfn), vmolr);
 4671
 4672	return 0;
 4673}
 4674
 4675static inline void igb_set_vf_vlan_strip(struct igb_adapter *adapter,
 4676					 int vfn, bool enable)
 
 
 
 
 
 4677{
 
 4678	struct e1000_hw *hw = &adapter->hw;
 4679	u32 val, reg;
 4680
 4681	if (hw->mac.type < e1000_82576)
 4682		return;
 4683
 4684	if (hw->mac.type == e1000_i350)
 4685		reg = E1000_DVMOLR(vfn);
 4686	else
 4687		reg = E1000_VMOLR(vfn);
 
 
 
 
 
 
 4688
 4689	val = rd32(reg);
 4690	if (enable)
 4691		val |= E1000_VMOLR_STRVLAN;
 4692	else
 4693		val &= ~(E1000_VMOLR_STRVLAN);
 4694	wr32(reg, val);
 4695}
 4696
 4697static inline void igb_set_vmolr(struct igb_adapter *adapter,
 4698				 int vfn, bool aupe)
 4699{
 4700	struct e1000_hw *hw = &adapter->hw;
 4701	u32 vmolr;
 4702
 4703	/* This register exists only on 82576 and newer so if we are older then
 4704	 * we should exit and do nothing
 4705	 */
 4706	if (hw->mac.type < e1000_82576)
 4707		return;
 4708
 4709	vmolr = rd32(E1000_VMOLR(vfn));
 
 
 
 
 
 
 
 
 4710	if (aupe)
 4711		vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
 4712	else
 4713		vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
 4714
 4715	/* clear all bits that might not be set */
 4716	vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
 4717
 4718	if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
 4719		vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
 4720	/* for VMDq only allow the VFs and pool 0 to accept broadcast and
 4721	 * multicast packets
 4722	 */
 4723	if (vfn <= adapter->vfs_allocated_count)
 4724		vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
 4725
 4726	wr32(E1000_VMOLR(vfn), vmolr);
 4727}
 4728
 4729/**
 4730 *  igb_setup_srrctl - configure the split and replication receive control
 4731 *                     registers
 4732 *  @adapter: Board private structure
 4733 *  @ring: receive ring to be configured
 4734 **/
 4735void igb_setup_srrctl(struct igb_adapter *adapter, struct igb_ring *ring)
 4736{
 4737	struct e1000_hw *hw = &adapter->hw;
 4738	int reg_idx = ring->reg_idx;
 4739	u32 srrctl = 0;
 4740
 4741	srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
 4742	if (ring_uses_large_buffer(ring))
 4743		srrctl |= IGB_RXBUFFER_3072 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
 4744	else
 4745		srrctl |= IGB_RXBUFFER_2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
 4746	srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
 4747	if (hw->mac.type >= e1000_82580)
 4748		srrctl |= E1000_SRRCTL_TIMESTAMP;
 4749	/* Only set Drop Enable if VFs allocated, or we are supporting multiple
 4750	 * queues and rx flow control is disabled
 4751	 */
 4752	if (adapter->vfs_allocated_count ||
 4753	    (!(hw->fc.current_mode & e1000_fc_rx_pause) &&
 4754	     adapter->num_rx_queues > 1))
 4755		srrctl |= E1000_SRRCTL_DROP_EN;
 4756
 4757	wr32(E1000_SRRCTL(reg_idx), srrctl);
 4758}
 4759
 4760/**
 4761 *  igb_configure_rx_ring - Configure a receive ring after Reset
 4762 *  @adapter: board private structure
 4763 *  @ring: receive ring to be configured
 4764 *
 4765 *  Configure the Rx unit of the MAC after a reset.
 4766 **/
 4767void igb_configure_rx_ring(struct igb_adapter *adapter,
 4768			   struct igb_ring *ring)
 4769{
 4770	struct e1000_hw *hw = &adapter->hw;
 4771	union e1000_adv_rx_desc *rx_desc;
 4772	u64 rdba = ring->dma;
 4773	int reg_idx = ring->reg_idx;
 4774	u32 rxdctl = 0;
 4775
 4776	xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
 4777	WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
 4778					   MEM_TYPE_PAGE_SHARED, NULL));
 4779
 4780	/* disable the queue */
 4781	wr32(E1000_RXDCTL(reg_idx), 0);
 4782
 4783	/* Set DMA base address registers */
 4784	wr32(E1000_RDBAL(reg_idx),
 4785	     rdba & 0x00000000ffffffffULL);
 4786	wr32(E1000_RDBAH(reg_idx), rdba >> 32);
 4787	wr32(E1000_RDLEN(reg_idx),
 4788	     ring->count * sizeof(union e1000_adv_rx_desc));
 4789
 4790	/* initialize head and tail */
 4791	ring->tail = adapter->io_addr + E1000_RDT(reg_idx);
 4792	wr32(E1000_RDH(reg_idx), 0);
 4793	writel(0, ring->tail);
 4794
 4795	/* set descriptor configuration */
 4796	igb_setup_srrctl(adapter, ring);
 
 
 
 
 
 
 
 
 
 4797
 4798	/* set filtering for VMDQ pools */
 4799	igb_set_vmolr(adapter, reg_idx & 0x7, true);
 4800
 4801	rxdctl |= IGB_RX_PTHRESH;
 4802	rxdctl |= IGB_RX_HTHRESH << 8;
 4803	rxdctl |= IGB_RX_WTHRESH << 16;
 4804
 4805	/* initialize rx_buffer_info */
 4806	memset(ring->rx_buffer_info, 0,
 4807	       sizeof(struct igb_rx_buffer) * ring->count);
 4808
 4809	/* initialize Rx descriptor 0 */
 4810	rx_desc = IGB_RX_DESC(ring, 0);
 4811	rx_desc->wb.upper.length = 0;
 4812
 4813	/* enable receive descriptor fetching */
 4814	rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
 4815	wr32(E1000_RXDCTL(reg_idx), rxdctl);
 4816}
 4817
 4818static void igb_set_rx_buffer_len(struct igb_adapter *adapter,
 4819				  struct igb_ring *rx_ring)
 4820{
 4821#if (PAGE_SIZE < 8192)
 4822	struct e1000_hw *hw = &adapter->hw;
 4823#endif
 4824
 4825	/* set build_skb and buffer size flags */
 4826	clear_ring_build_skb_enabled(rx_ring);
 4827	clear_ring_uses_large_buffer(rx_ring);
 4828
 4829	if (adapter->flags & IGB_FLAG_RX_LEGACY)
 4830		return;
 4831
 4832	set_ring_build_skb_enabled(rx_ring);
 4833
 4834#if (PAGE_SIZE < 8192)
 4835	if (adapter->max_frame_size > IGB_MAX_FRAME_BUILD_SKB ||
 4836	    rd32(E1000_RCTL) & E1000_RCTL_SBP)
 4837		set_ring_uses_large_buffer(rx_ring);
 4838#endif
 4839}
 4840
 4841/**
 4842 *  igb_configure_rx - Configure receive Unit after Reset
 4843 *  @adapter: board private structure
 4844 *
 4845 *  Configure the Rx unit of the MAC after a reset.
 4846 **/
 4847static void igb_configure_rx(struct igb_adapter *adapter)
 4848{
 4849	int i;
 4850
 
 
 
 4851	/* set the correct pool for the PF default MAC address in entry 0 */
 4852	igb_set_default_mac_filter(adapter);
 
 4853
 4854	/* Setup the HW Rx Head and Tail Descriptor Pointers and
 4855	 * the Base and Length of the Rx Descriptor Ring
 4856	 */
 4857	for (i = 0; i < adapter->num_rx_queues; i++) {
 4858		struct igb_ring *rx_ring = adapter->rx_ring[i];
 4859
 4860		igb_set_rx_buffer_len(adapter, rx_ring);
 4861		igb_configure_rx_ring(adapter, rx_ring);
 4862	}
 4863}
 4864
 4865/**
 4866 *  igb_free_tx_resources - Free Tx Resources per Queue
 4867 *  @tx_ring: Tx descriptor ring for a specific queue
 4868 *
 4869 *  Free all transmit software resources
 4870 **/
 4871void igb_free_tx_resources(struct igb_ring *tx_ring)
 4872{
 4873	igb_clean_tx_ring(tx_ring);
 4874
 4875	vfree(tx_ring->tx_buffer_info);
 4876	tx_ring->tx_buffer_info = NULL;
 4877
 4878	/* if not set, then don't free */
 4879	if (!tx_ring->desc)
 4880		return;
 4881
 4882	dma_free_coherent(tx_ring->dev, tx_ring->size,
 4883			  tx_ring->desc, tx_ring->dma);
 4884
 4885	tx_ring->desc = NULL;
 4886}
 4887
 4888/**
 4889 *  igb_free_all_tx_resources - Free Tx Resources for All Queues
 4890 *  @adapter: board private structure
 4891 *
 4892 *  Free all transmit software resources
 4893 **/
 4894static void igb_free_all_tx_resources(struct igb_adapter *adapter)
 4895{
 4896	int i;
 4897
 4898	for (i = 0; i < adapter->num_tx_queues; i++)
 4899		if (adapter->tx_ring[i])
 4900			igb_free_tx_resources(adapter->tx_ring[i]);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 4901}
 4902
 4903/**
 4904 *  igb_clean_tx_ring - Free Tx Buffers
 4905 *  @tx_ring: ring to be cleaned
 4906 **/
 4907static void igb_clean_tx_ring(struct igb_ring *tx_ring)
 4908{
 4909	u16 i = tx_ring->next_to_clean;
 4910	struct igb_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
 4911
 4912	while (i != tx_ring->next_to_use) {
 4913		union e1000_adv_tx_desc *eop_desc, *tx_desc;
 4914
 4915		/* Free all the Tx ring sk_buffs or xdp frames */
 4916		if (tx_buffer->type == IGB_TYPE_SKB)
 4917			dev_kfree_skb_any(tx_buffer->skb);
 4918		else
 4919			xdp_return_frame(tx_buffer->xdpf);
 4920
 4921		/* unmap skb header data */
 4922		dma_unmap_single(tx_ring->dev,
 4923				 dma_unmap_addr(tx_buffer, dma),
 4924				 dma_unmap_len(tx_buffer, len),
 4925				 DMA_TO_DEVICE);
 4926
 4927		/* check for eop_desc to determine the end of the packet */
 4928		eop_desc = tx_buffer->next_to_watch;
 4929		tx_desc = IGB_TX_DESC(tx_ring, i);
 4930
 4931		/* unmap remaining buffers */
 4932		while (tx_desc != eop_desc) {
 4933			tx_buffer++;
 4934			tx_desc++;
 4935			i++;
 4936			if (unlikely(i == tx_ring->count)) {
 4937				i = 0;
 4938				tx_buffer = tx_ring->tx_buffer_info;
 4939				tx_desc = IGB_TX_DESC(tx_ring, 0);
 4940			}
 4941
 4942			/* unmap any remaining paged data */
 4943			if (dma_unmap_len(tx_buffer, len))
 4944				dma_unmap_page(tx_ring->dev,
 4945					       dma_unmap_addr(tx_buffer, dma),
 4946					       dma_unmap_len(tx_buffer, len),
 4947					       DMA_TO_DEVICE);
 4948		}
 4949
 4950		tx_buffer->next_to_watch = NULL;
 4951
 4952		/* move us one more past the eop_desc for start of next pkt */
 4953		tx_buffer++;
 4954		i++;
 4955		if (unlikely(i == tx_ring->count)) {
 4956			i = 0;
 4957			tx_buffer = tx_ring->tx_buffer_info;
 4958		}
 4959	}
 4960
 4961	/* reset BQL for queue */
 4962	netdev_tx_reset_queue(txring_txq(tx_ring));
 4963
 4964	/* reset next_to_use and next_to_clean */
 
 
 
 
 
 4965	tx_ring->next_to_use = 0;
 4966	tx_ring->next_to_clean = 0;
 4967}
 4968
 4969/**
 4970 *  igb_clean_all_tx_rings - Free Tx Buffers for all queues
 4971 *  @adapter: board private structure
 4972 **/
 4973static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
 4974{
 4975	int i;
 4976
 4977	for (i = 0; i < adapter->num_tx_queues; i++)
 4978		if (adapter->tx_ring[i])
 4979			igb_clean_tx_ring(adapter->tx_ring[i]);
 4980}
 4981
 4982/**
 4983 *  igb_free_rx_resources - Free Rx Resources
 4984 *  @rx_ring: ring to clean the resources from
 4985 *
 4986 *  Free all receive software resources
 4987 **/
 4988void igb_free_rx_resources(struct igb_ring *rx_ring)
 4989{
 4990	igb_clean_rx_ring(rx_ring);
 4991
 4992	rx_ring->xdp_prog = NULL;
 4993	xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 4994	vfree(rx_ring->rx_buffer_info);
 4995	rx_ring->rx_buffer_info = NULL;
 4996
 4997	/* if not set, then don't free */
 4998	if (!rx_ring->desc)
 4999		return;
 5000
 5001	dma_free_coherent(rx_ring->dev, rx_ring->size,
 5002			  rx_ring->desc, rx_ring->dma);
 5003
 5004	rx_ring->desc = NULL;
 5005}
 5006
 5007/**
 5008 *  igb_free_all_rx_resources - Free Rx Resources for All Queues
 5009 *  @adapter: board private structure
 5010 *
 5011 *  Free all receive software resources
 5012 **/
 5013static void igb_free_all_rx_resources(struct igb_adapter *adapter)
 5014{
 5015	int i;
 5016
 5017	for (i = 0; i < adapter->num_rx_queues; i++)
 5018		if (adapter->rx_ring[i])
 5019			igb_free_rx_resources(adapter->rx_ring[i]);
 5020}
 5021
 5022/**
 5023 *  igb_clean_rx_ring - Free Rx Buffers per Queue
 5024 *  @rx_ring: ring to free buffers from
 5025 **/
 5026static void igb_clean_rx_ring(struct igb_ring *rx_ring)
 5027{
 5028	u16 i = rx_ring->next_to_clean;
 
 5029
 5030	dev_kfree_skb(rx_ring->skb);
 
 5031	rx_ring->skb = NULL;
 5032
 
 
 
 5033	/* Free all the Rx ring sk_buffs */
 5034	while (i != rx_ring->next_to_alloc) {
 5035		struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
 5036
 5037		/* Invalidate cache lines that may have been written to by
 5038		 * device so that we avoid corrupting memory.
 5039		 */
 5040		dma_sync_single_range_for_cpu(rx_ring->dev,
 5041					      buffer_info->dma,
 5042					      buffer_info->page_offset,
 5043					      igb_rx_bufsz(rx_ring),
 5044					      DMA_FROM_DEVICE);
 5045
 5046		/* free resources associated with mapping */
 5047		dma_unmap_page_attrs(rx_ring->dev,
 5048				     buffer_info->dma,
 5049				     igb_rx_pg_size(rx_ring),
 5050				     DMA_FROM_DEVICE,
 5051				     IGB_RX_DMA_ATTR);
 5052		__page_frag_cache_drain(buffer_info->page,
 5053					buffer_info->pagecnt_bias);
 5054
 5055		i++;
 5056		if (i == rx_ring->count)
 5057			i = 0;
 
 
 
 
 5058	}
 5059
 
 
 
 
 
 
 5060	rx_ring->next_to_alloc = 0;
 5061	rx_ring->next_to_clean = 0;
 5062	rx_ring->next_to_use = 0;
 5063}
 5064
 5065/**
 5066 *  igb_clean_all_rx_rings - Free Rx Buffers for all queues
 5067 *  @adapter: board private structure
 5068 **/
 5069static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
 5070{
 5071	int i;
 5072
 5073	for (i = 0; i < adapter->num_rx_queues; i++)
 5074		if (adapter->rx_ring[i])
 5075			igb_clean_rx_ring(adapter->rx_ring[i]);
 5076}
 5077
 5078/**
 5079 *  igb_set_mac - Change the Ethernet Address of the NIC
 5080 *  @netdev: network interface device structure
 5081 *  @p: pointer to an address structure
 5082 *
 5083 *  Returns 0 on success, negative on failure
 5084 **/
 5085static int igb_set_mac(struct net_device *netdev, void *p)
 5086{
 5087	struct igb_adapter *adapter = netdev_priv(netdev);
 5088	struct e1000_hw *hw = &adapter->hw;
 5089	struct sockaddr *addr = p;
 5090
 5091	if (!is_valid_ether_addr(addr->sa_data))
 5092		return -EADDRNOTAVAIL;
 5093
 5094	eth_hw_addr_set(netdev, addr->sa_data);
 5095	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
 5096
 5097	/* set the correct pool for the new PF MAC address in entry 0 */
 5098	igb_set_default_mac_filter(adapter);
 
 5099
 5100	return 0;
 5101}
 5102
 5103/**
 5104 *  igb_write_mc_addr_list - write multicast addresses to MTA
 5105 *  @netdev: network interface device structure
 5106 *
 5107 *  Writes multicast address list to the MTA hash table.
 5108 *  Returns: -ENOMEM on failure
 5109 *           0 on no addresses written
 5110 *           X on writing X addresses to MTA
 5111 **/
 5112static int igb_write_mc_addr_list(struct net_device *netdev)
 5113{
 5114	struct igb_adapter *adapter = netdev_priv(netdev);
 5115	struct e1000_hw *hw = &adapter->hw;
 5116	struct netdev_hw_addr *ha;
 5117	u8  *mta_list;
 5118	int i;
 5119
 5120	if (netdev_mc_empty(netdev)) {
 5121		/* nothing to program, so clear mc list */
 5122		igb_update_mc_addr_list(hw, NULL, 0);
 5123		igb_restore_vf_multicasts(adapter);
 5124		return 0;
 5125	}
 5126
 5127	mta_list = kcalloc(netdev_mc_count(netdev), 6, GFP_ATOMIC);
 5128	if (!mta_list)
 5129		return -ENOMEM;
 5130
 5131	/* The shared function expects a packed array of only addresses. */
 5132	i = 0;
 5133	netdev_for_each_mc_addr(ha, netdev)
 5134		memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
 5135
 5136	igb_update_mc_addr_list(hw, mta_list, i);
 5137	kfree(mta_list);
 5138
 5139	return netdev_mc_count(netdev);
 5140}
 5141
 5142static int igb_vlan_promisc_enable(struct igb_adapter *adapter)
 5143{
 5144	struct e1000_hw *hw = &adapter->hw;
 5145	u32 i, pf_id;
 5146
 5147	switch (hw->mac.type) {
 5148	case e1000_i210:
 5149	case e1000_i211:
 5150	case e1000_i350:
 5151		/* VLAN filtering needed for VLAN prio filter */
 5152		if (adapter->netdev->features & NETIF_F_NTUPLE)
 5153			break;
 5154		fallthrough;
 5155	case e1000_82576:
 5156	case e1000_82580:
 5157	case e1000_i354:
 5158		/* VLAN filtering needed for pool filtering */
 5159		if (adapter->vfs_allocated_count)
 5160			break;
 5161		fallthrough;
 5162	default:
 5163		return 1;
 5164	}
 5165
 5166	/* We are already in VLAN promisc, nothing to do */
 5167	if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
 5168		return 0;
 5169
 5170	if (!adapter->vfs_allocated_count)
 5171		goto set_vfta;
 5172
 5173	/* Add PF to all active pools */
 5174	pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
 5175
 5176	for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
 5177		u32 vlvf = rd32(E1000_VLVF(i));
 5178
 5179		vlvf |= BIT(pf_id);
 5180		wr32(E1000_VLVF(i), vlvf);
 5181	}
 5182
 5183set_vfta:
 5184	/* Set all bits in the VLAN filter table array */
 5185	for (i = E1000_VLAN_FILTER_TBL_SIZE; i--;)
 5186		hw->mac.ops.write_vfta(hw, i, ~0U);
 5187
 5188	/* Set flag so we don't redo unnecessary work */
 5189	adapter->flags |= IGB_FLAG_VLAN_PROMISC;
 5190
 5191	return 0;
 5192}
 5193
 5194#define VFTA_BLOCK_SIZE 8
 5195static void igb_scrub_vfta(struct igb_adapter *adapter, u32 vfta_offset)
 5196{
 
 5197	struct e1000_hw *hw = &adapter->hw;
 5198	u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
 5199	u32 vid_start = vfta_offset * 32;
 5200	u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
 5201	u32 i, vid, word, bits, pf_id;
 5202
 5203	/* guarantee that we don't scrub out management VLAN */
 5204	vid = adapter->mng_vlan_id;
 5205	if (vid >= vid_start && vid < vid_end)
 5206		vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
 5207
 5208	if (!adapter->vfs_allocated_count)
 5209		goto set_vfta;
 5210
 5211	pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
 5212
 5213	for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
 5214		u32 vlvf = rd32(E1000_VLVF(i));
 5215
 5216		/* pull VLAN ID from VLVF */
 5217		vid = vlvf & VLAN_VID_MASK;
 5218
 5219		/* only concern ourselves with a certain range */
 5220		if (vid < vid_start || vid >= vid_end)
 5221			continue;
 5222
 5223		if (vlvf & E1000_VLVF_VLANID_ENABLE) {
 5224			/* record VLAN ID in VFTA */
 5225			vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
 5226
 5227			/* if PF is part of this then continue */
 5228			if (test_bit(vid, adapter->active_vlans))
 5229				continue;
 
 
 
 
 5230		}
 5231
 5232		/* remove PF from the pool */
 5233		bits = ~BIT(pf_id);
 5234		bits &= rd32(E1000_VLVF(i));
 5235		wr32(E1000_VLVF(i), bits);
 5236	}
 5237
 5238set_vfta:
 5239	/* extract values from active_vlans and write back to VFTA */
 5240	for (i = VFTA_BLOCK_SIZE; i--;) {
 5241		vid = (vfta_offset + i) * 32;
 5242		word = vid / BITS_PER_LONG;
 5243		bits = vid % BITS_PER_LONG;
 5244
 5245		vfta[i] |= adapter->active_vlans[word] >> bits;
 5246
 5247		hw->mac.ops.write_vfta(hw, vfta_offset + i, vfta[i]);
 5248	}
 5249}
 5250
 5251static void igb_vlan_promisc_disable(struct igb_adapter *adapter)
 5252{
 5253	u32 i;
 5254
 5255	/* We are not in VLAN promisc, nothing to do */
 5256	if (!(adapter->flags & IGB_FLAG_VLAN_PROMISC))
 5257		return;
 5258
 5259	/* Set flag so we don't redo unnecessary work */
 5260	adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
 5261
 5262	for (i = 0; i < E1000_VLAN_FILTER_TBL_SIZE; i += VFTA_BLOCK_SIZE)
 5263		igb_scrub_vfta(adapter, i);
 5264}
 5265
 5266/**
 5267 *  igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
 5268 *  @netdev: network interface device structure
 5269 *
 5270 *  The set_rx_mode entry point is called whenever the unicast or multicast
 5271 *  address lists or the network interface flags are updated.  This routine is
 5272 *  responsible for configuring the hardware for proper unicast, multicast,
 5273 *  promiscuous mode, and all-multi behavior.
 5274 **/
 5275static void igb_set_rx_mode(struct net_device *netdev)
 5276{
 5277	struct igb_adapter *adapter = netdev_priv(netdev);
 5278	struct e1000_hw *hw = &adapter->hw;
 5279	unsigned int vfn = adapter->vfs_allocated_count;
 5280	u32 rctl = 0, vmolr = 0, rlpml = MAX_JUMBO_FRAME_SIZE;
 5281	int count;
 5282
 5283	/* Check for Promiscuous and All Multicast modes */
 5284	if (netdev->flags & IFF_PROMISC) {
 5285		rctl |= E1000_RCTL_UPE | E1000_RCTL_MPE;
 5286		vmolr |= E1000_VMOLR_MPME;
 5287
 5288		/* enable use of UTA filter to force packets to default pool */
 5289		if (hw->mac.type == e1000_82576)
 5290			vmolr |= E1000_VMOLR_ROPE;
 
 
 
 
 
 
 5291	} else {
 5292		if (netdev->flags & IFF_ALLMULTI) {
 5293			rctl |= E1000_RCTL_MPE;
 5294			vmolr |= E1000_VMOLR_MPME;
 5295		} else {
 5296			/* Write addresses to the MTA, if the attempt fails
 5297			 * then we should just turn on promiscuous mode so
 5298			 * that we can at least receive multicast traffic
 5299			 */
 5300			count = igb_write_mc_addr_list(netdev);
 5301			if (count < 0) {
 5302				rctl |= E1000_RCTL_MPE;
 5303				vmolr |= E1000_VMOLR_MPME;
 5304			} else if (count) {
 5305				vmolr |= E1000_VMOLR_ROMPE;
 5306			}
 5307		}
 
 
 
 
 
 
 
 
 
 
 5308	}
 5309
 5310	/* Write addresses to available RAR registers, if there is not
 5311	 * sufficient space to store all the addresses then enable
 5312	 * unicast promiscuous mode
 5313	 */
 5314	if (__dev_uc_sync(netdev, igb_uc_sync, igb_uc_unsync)) {
 5315		rctl |= E1000_RCTL_UPE;
 5316		vmolr |= E1000_VMOLR_ROPE;
 5317	}
 5318
 5319	/* enable VLAN filtering by default */
 5320	rctl |= E1000_RCTL_VFE;
 5321
 5322	/* disable VLAN filtering for modes that require it */
 5323	if ((netdev->flags & IFF_PROMISC) ||
 5324	    (netdev->features & NETIF_F_RXALL)) {
 5325		/* if we fail to set all rules then just clear VFE */
 5326		if (igb_vlan_promisc_enable(adapter))
 5327			rctl &= ~E1000_RCTL_VFE;
 5328	} else {
 5329		igb_vlan_promisc_disable(adapter);
 5330	}
 5331
 5332	/* update state of unicast, multicast, and VLAN filtering modes */
 5333	rctl |= rd32(E1000_RCTL) & ~(E1000_RCTL_UPE | E1000_RCTL_MPE |
 5334				     E1000_RCTL_VFE);
 5335	wr32(E1000_RCTL, rctl);
 5336
 5337#if (PAGE_SIZE < 8192)
 5338	if (!adapter->vfs_allocated_count) {
 5339		if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
 5340			rlpml = IGB_MAX_FRAME_BUILD_SKB;
 5341	}
 5342#endif
 5343	wr32(E1000_RLPML, rlpml);
 5344
 5345	/* In order to support SR-IOV and eventually VMDq it is necessary to set
 5346	 * the VMOLR to enable the appropriate modes.  Without this workaround
 5347	 * we will have issues with VLAN tag stripping not being done for frames
 5348	 * that are only arriving because we are the default pool
 5349	 */
 5350	if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
 5351		return;
 5352
 5353	/* set UTA to appropriate mode */
 5354	igb_set_uta(adapter, !!(vmolr & E1000_VMOLR_ROPE));
 5355
 5356	vmolr |= rd32(E1000_VMOLR(vfn)) &
 5357		 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
 5358
 5359	/* enable Rx jumbo frames, restrict as needed to support build_skb */
 5360	vmolr &= ~E1000_VMOLR_RLPML_MASK;
 5361#if (PAGE_SIZE < 8192)
 5362	if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
 5363		vmolr |= IGB_MAX_FRAME_BUILD_SKB;
 5364	else
 5365#endif
 5366		vmolr |= MAX_JUMBO_FRAME_SIZE;
 5367	vmolr |= E1000_VMOLR_LPE;
 5368
 5369	wr32(E1000_VMOLR(vfn), vmolr);
 5370
 5371	igb_restore_vf_multicasts(adapter);
 5372}
 5373
 5374static void igb_check_wvbr(struct igb_adapter *adapter)
 5375{
 5376	struct e1000_hw *hw = &adapter->hw;
 5377	u32 wvbr = 0;
 5378
 5379	switch (hw->mac.type) {
 5380	case e1000_82576:
 5381	case e1000_i350:
 5382		wvbr = rd32(E1000_WVBR);
 5383		if (!wvbr)
 5384			return;
 5385		break;
 5386	default:
 5387		break;
 5388	}
 5389
 5390	adapter->wvbr |= wvbr;
 5391}
 5392
 5393#define IGB_STAGGERED_QUEUE_OFFSET 8
 5394
 5395static void igb_spoof_check(struct igb_adapter *adapter)
 5396{
 5397	int j;
 5398
 5399	if (!adapter->wvbr)
 5400		return;
 5401
 5402	for (j = 0; j < adapter->vfs_allocated_count; j++) {
 5403		if (adapter->wvbr & BIT(j) ||
 5404		    adapter->wvbr & BIT(j + IGB_STAGGERED_QUEUE_OFFSET)) {
 5405			dev_warn(&adapter->pdev->dev,
 5406				"Spoof event(s) detected on VF %d\n", j);
 5407			adapter->wvbr &=
 5408				~(BIT(j) |
 5409				  BIT(j + IGB_STAGGERED_QUEUE_OFFSET));
 5410		}
 5411	}
 5412}
 5413
 5414/* Need to wait a few seconds after link up to get diagnostic information from
 5415 * the phy
 5416 */
 5417static void igb_update_phy_info(struct timer_list *t)
 5418{
 5419	struct igb_adapter *adapter = from_timer(adapter, t, phy_info_timer);
 5420	igb_get_phy_info(&adapter->hw);
 5421}
 5422
 5423/**
 5424 *  igb_has_link - check shared code for link and determine up/down
 5425 *  @adapter: pointer to driver private info
 5426 **/
 5427bool igb_has_link(struct igb_adapter *adapter)
 5428{
 5429	struct e1000_hw *hw = &adapter->hw;
 5430	bool link_active = false;
 5431
 5432	/* get_link_status is set on LSC (link status) interrupt or
 5433	 * rx sequence error interrupt.  get_link_status will stay
 5434	 * false until the e1000_check_for_link establishes link
 5435	 * for copper adapters ONLY
 5436	 */
 5437	switch (hw->phy.media_type) {
 5438	case e1000_media_type_copper:
 5439		if (!hw->mac.get_link_status)
 5440			return true;
 5441		fallthrough;
 5442	case e1000_media_type_internal_serdes:
 5443		hw->mac.ops.check_for_link(hw);
 5444		link_active = !hw->mac.get_link_status;
 5445		break;
 5446	default:
 5447	case e1000_media_type_unknown:
 5448		break;
 5449	}
 5450
 5451	if (((hw->mac.type == e1000_i210) ||
 5452	     (hw->mac.type == e1000_i211)) &&
 5453	     (hw->phy.id == I210_I_PHY_ID)) {
 5454		if (!netif_carrier_ok(adapter->netdev)) {
 5455			adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
 5456		} else if (!(adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)) {
 5457			adapter->flags |= IGB_FLAG_NEED_LINK_UPDATE;
 5458			adapter->link_check_timeout = jiffies;
 5459		}
 5460	}
 5461
 5462	return link_active;
 5463}
 5464
 5465static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
 5466{
 5467	bool ret = false;
 5468	u32 ctrl_ext, thstat;
 5469
 5470	/* check for thermal sensor event on i350 copper only */
 5471	if (hw->mac.type == e1000_i350) {
 5472		thstat = rd32(E1000_THSTAT);
 5473		ctrl_ext = rd32(E1000_CTRL_EXT);
 5474
 5475		if ((hw->phy.media_type == e1000_media_type_copper) &&
 5476		    !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII))
 5477			ret = !!(thstat & event);
 5478	}
 5479
 5480	return ret;
 5481}
 5482
 5483/**
 5484 *  igb_check_lvmmc - check for malformed packets received
 5485 *  and indicated in LVMMC register
 5486 *  @adapter: pointer to adapter
 5487 **/
 5488static void igb_check_lvmmc(struct igb_adapter *adapter)
 5489{
 5490	struct e1000_hw *hw = &adapter->hw;
 5491	u32 lvmmc;
 5492
 5493	lvmmc = rd32(E1000_LVMMC);
 5494	if (lvmmc) {
 5495		if (unlikely(net_ratelimit())) {
 5496			netdev_warn(adapter->netdev,
 5497				    "malformed Tx packet detected and dropped, LVMMC:0x%08x\n",
 5498				    lvmmc);
 5499		}
 5500	}
 5501}
 5502
 5503/**
 5504 *  igb_watchdog - Timer Call-back
 5505 *  @t: pointer to timer_list containing our private info pointer
 5506 **/
 5507static void igb_watchdog(struct timer_list *t)
 5508{
 5509	struct igb_adapter *adapter = from_timer(adapter, t, watchdog_timer);
 5510	/* Do the rest outside of interrupt context */
 5511	schedule_work(&adapter->watchdog_task);
 5512}
 5513
 5514static void igb_watchdog_task(struct work_struct *work)
 5515{
 5516	struct igb_adapter *adapter = container_of(work,
 5517						   struct igb_adapter,
 5518						   watchdog_task);
 5519	struct e1000_hw *hw = &adapter->hw;
 5520	struct e1000_phy_info *phy = &hw->phy;
 5521	struct net_device *netdev = adapter->netdev;
 5522	u32 link;
 5523	int i;
 5524	u32 connsw;
 5525	u16 phy_data, retry_count = 20;
 5526
 5527	link = igb_has_link(adapter);
 5528
 5529	if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) {
 5530		if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
 5531			adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
 5532		else
 5533			link = false;
 5534	}
 5535
 5536	/* Force link down if we have fiber to swap to */
 5537	if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
 5538		if (hw->phy.media_type == e1000_media_type_copper) {
 5539			connsw = rd32(E1000_CONNSW);
 5540			if (!(connsw & E1000_CONNSW_AUTOSENSE_EN))
 5541				link = 0;
 5542		}
 5543	}
 5544	if (link) {
 5545		/* Perform a reset if the media type changed. */
 5546		if (hw->dev_spec._82575.media_changed) {
 5547			hw->dev_spec._82575.media_changed = false;
 5548			adapter->flags |= IGB_FLAG_MEDIA_RESET;
 5549			igb_reset(adapter);
 5550		}
 5551		/* Cancel scheduled suspend requests. */
 5552		pm_runtime_resume(netdev->dev.parent);
 5553
 5554		if (!netif_carrier_ok(netdev)) {
 5555			u32 ctrl;
 5556
 5557			hw->mac.ops.get_speed_and_duplex(hw,
 5558							 &adapter->link_speed,
 5559							 &adapter->link_duplex);
 5560
 5561			ctrl = rd32(E1000_CTRL);
 5562			/* Links status message must follow this format */
 5563			netdev_info(netdev,
 5564			       "igb: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
 5565			       netdev->name,
 5566			       adapter->link_speed,
 5567			       adapter->link_duplex == FULL_DUPLEX ?
 5568			       "Full" : "Half",
 5569			       (ctrl & E1000_CTRL_TFCE) &&
 5570			       (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
 5571			       (ctrl & E1000_CTRL_RFCE) ?  "RX" :
 5572			       (ctrl & E1000_CTRL_TFCE) ?  "TX" : "None");
 5573
 5574			/* disable EEE if enabled */
 5575			if ((adapter->flags & IGB_FLAG_EEE) &&
 5576				(adapter->link_duplex == HALF_DUPLEX)) {
 5577				dev_info(&adapter->pdev->dev,
 5578				"EEE Disabled: unsupported at half duplex. Re-enable using ethtool when at full duplex.\n");
 5579				adapter->hw.dev_spec._82575.eee_disable = true;
 5580				adapter->flags &= ~IGB_FLAG_EEE;
 5581			}
 5582
 5583			/* check if SmartSpeed worked */
 5584			igb_check_downshift(hw);
 5585			if (phy->speed_downgraded)
 5586				netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
 5587
 5588			/* check for thermal sensor event */
 5589			if (igb_thermal_sensor_event(hw,
 5590			    E1000_THSTAT_LINK_THROTTLE))
 5591				netdev_info(netdev, "The network adapter link speed was downshifted because it overheated\n");
 
 
 
 5592
 5593			/* adjust timeout factor according to speed/duplex */
 5594			adapter->tx_timeout_factor = 1;
 5595			switch (adapter->link_speed) {
 5596			case SPEED_10:
 5597				adapter->tx_timeout_factor = 14;
 5598				break;
 5599			case SPEED_100:
 5600				/* maybe add some timeout factor ? */
 5601				break;
 5602			}
 5603
 5604			if (adapter->link_speed != SPEED_1000 ||
 5605			    !hw->phy.ops.read_reg)
 5606				goto no_wait;
 5607
 5608			/* wait for Remote receiver status OK */
 5609retry_read_status:
 5610			if (!igb_read_phy_reg(hw, PHY_1000T_STATUS,
 5611					      &phy_data)) {
 5612				if (!(phy_data & SR_1000T_REMOTE_RX_STATUS) &&
 5613				    retry_count) {
 5614					msleep(100);
 5615					retry_count--;
 5616					goto retry_read_status;
 5617				} else if (!retry_count) {
 5618					dev_err(&adapter->pdev->dev, "exceed max 2 second\n");
 5619				}
 5620			} else {
 5621				dev_err(&adapter->pdev->dev, "read 1000Base-T Status Reg\n");
 5622			}
 5623no_wait:
 5624			netif_carrier_on(netdev);
 5625
 5626			igb_ping_all_vfs(adapter);
 5627			igb_check_vf_rate_limit(adapter);
 5628
 5629			/* link state has changed, schedule phy info update */
 5630			if (!test_bit(__IGB_DOWN, &adapter->state))
 5631				mod_timer(&adapter->phy_info_timer,
 5632					  round_jiffies(jiffies + 2 * HZ));
 5633		}
 5634	} else {
 5635		if (netif_carrier_ok(netdev)) {
 5636			adapter->link_speed = 0;
 5637			adapter->link_duplex = 0;
 5638
 5639			/* check for thermal sensor event */
 5640			if (igb_thermal_sensor_event(hw,
 5641			    E1000_THSTAT_PWR_DOWN)) {
 5642				netdev_err(netdev, "The network adapter was stopped because it overheated\n");
 
 5643			}
 5644
 5645			/* Links status message must follow this format */
 5646			netdev_info(netdev, "igb: %s NIC Link is Down\n",
 5647			       netdev->name);
 5648			netif_carrier_off(netdev);
 5649
 5650			igb_ping_all_vfs(adapter);
 5651
 5652			/* link state has changed, schedule phy info update */
 5653			if (!test_bit(__IGB_DOWN, &adapter->state))
 5654				mod_timer(&adapter->phy_info_timer,
 5655					  round_jiffies(jiffies + 2 * HZ));
 5656
 5657			/* link is down, time to check for alternate media */
 5658			if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
 5659				igb_check_swap_media(adapter);
 5660				if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
 5661					schedule_work(&adapter->reset_task);
 5662					/* return immediately */
 5663					return;
 5664				}
 5665			}
 5666			pm_schedule_suspend(netdev->dev.parent,
 5667					    MSEC_PER_SEC * 5);
 5668
 5669		/* also check for alternate media here */
 5670		} else if (!netif_carrier_ok(netdev) &&
 5671			   (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
 5672			igb_check_swap_media(adapter);
 5673			if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
 5674				schedule_work(&adapter->reset_task);
 5675				/* return immediately */
 5676				return;
 5677			}
 5678		}
 5679	}
 5680
 5681	spin_lock(&adapter->stats64_lock);
 5682	igb_update_stats(adapter);
 5683	spin_unlock(&adapter->stats64_lock);
 5684
 5685	for (i = 0; i < adapter->num_tx_queues; i++) {
 5686		struct igb_ring *tx_ring = adapter->tx_ring[i];
 5687		if (!netif_carrier_ok(netdev)) {
 5688			/* We've lost link, so the controller stops DMA,
 5689			 * but we've got queued Tx work that's never going
 5690			 * to get done, so reset controller to flush Tx.
 5691			 * (Do the reset outside of interrupt context).
 5692			 */
 5693			if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
 5694				adapter->tx_timeout_count++;
 5695				schedule_work(&adapter->reset_task);
 5696				/* return immediately since reset is imminent */
 5697				return;
 5698			}
 5699		}
 5700
 5701		/* Force detection of hung controller every watchdog period */
 5702		set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
 5703	}
 5704
 5705	/* Cause software interrupt to ensure Rx ring is cleaned */
 5706	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
 5707		u32 eics = 0;
 5708
 5709		for (i = 0; i < adapter->num_q_vectors; i++)
 5710			eics |= adapter->q_vector[i]->eims_value;
 5711		wr32(E1000_EICS, eics);
 5712	} else {
 5713		wr32(E1000_ICS, E1000_ICS_RXDMT0);
 5714	}
 5715
 5716	igb_spoof_check(adapter);
 5717	igb_ptp_rx_hang(adapter);
 5718	igb_ptp_tx_hang(adapter);
 5719
 5720	/* Check LVMMC register on i350/i354 only */
 5721	if ((adapter->hw.mac.type == e1000_i350) ||
 5722	    (adapter->hw.mac.type == e1000_i354))
 5723		igb_check_lvmmc(adapter);
 5724
 5725	/* Reset the timer */
 5726	if (!test_bit(__IGB_DOWN, &adapter->state)) {
 5727		if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)
 5728			mod_timer(&adapter->watchdog_timer,
 5729				  round_jiffies(jiffies +  HZ));
 5730		else
 5731			mod_timer(&adapter->watchdog_timer,
 5732				  round_jiffies(jiffies + 2 * HZ));
 5733	}
 5734}
 5735
 5736enum latency_range {
 5737	lowest_latency = 0,
 5738	low_latency = 1,
 5739	bulk_latency = 2,
 5740	latency_invalid = 255
 5741};
 5742
 5743/**
 5744 *  igb_update_ring_itr - update the dynamic ITR value based on packet size
 5745 *  @q_vector: pointer to q_vector
 5746 *
 5747 *  Stores a new ITR value based on strictly on packet size.  This
 5748 *  algorithm is less sophisticated than that used in igb_update_itr,
 5749 *  due to the difficulty of synchronizing statistics across multiple
 5750 *  receive rings.  The divisors and thresholds used by this function
 5751 *  were determined based on theoretical maximum wire speed and testing
 5752 *  data, in order to minimize response time while increasing bulk
 5753 *  throughput.
 5754 *  This functionality is controlled by ethtool's coalescing settings.
 5755 *  NOTE:  This function is called only when operating in a multiqueue
 5756 *         receive environment.
 5757 **/
 5758static void igb_update_ring_itr(struct igb_q_vector *q_vector)
 5759{
 5760	int new_val = q_vector->itr_val;
 5761	int avg_wire_size = 0;
 5762	struct igb_adapter *adapter = q_vector->adapter;
 5763	unsigned int packets;
 5764
 5765	/* For non-gigabit speeds, just fix the interrupt rate at 4000
 5766	 * ints/sec - ITR timer value of 120 ticks.
 5767	 */
 5768	if (adapter->link_speed != SPEED_1000) {
 5769		new_val = IGB_4K_ITR;
 5770		goto set_itr_val;
 5771	}
 5772
 5773	packets = q_vector->rx.total_packets;
 5774	if (packets)
 5775		avg_wire_size = q_vector->rx.total_bytes / packets;
 5776
 5777	packets = q_vector->tx.total_packets;
 5778	if (packets)
 5779		avg_wire_size = max_t(u32, avg_wire_size,
 5780				      q_vector->tx.total_bytes / packets);
 5781
 5782	/* if avg_wire_size isn't set no work was done */
 5783	if (!avg_wire_size)
 5784		goto clear_counts;
 5785
 5786	/* Add 24 bytes to size to account for CRC, preamble, and gap */
 5787	avg_wire_size += 24;
 5788
 5789	/* Don't starve jumbo frames */
 5790	avg_wire_size = min(avg_wire_size, 3000);
 5791
 5792	/* Give a little boost to mid-size frames */
 5793	if ((avg_wire_size > 300) && (avg_wire_size < 1200))
 5794		new_val = avg_wire_size / 3;
 5795	else
 5796		new_val = avg_wire_size / 2;
 5797
 5798	/* conservative mode (itr 3) eliminates the lowest_latency setting */
 5799	if (new_val < IGB_20K_ITR &&
 5800	    ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
 5801	     (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
 5802		new_val = IGB_20K_ITR;
 5803
 5804set_itr_val:
 5805	if (new_val != q_vector->itr_val) {
 5806		q_vector->itr_val = new_val;
 5807		q_vector->set_itr = 1;
 5808	}
 5809clear_counts:
 5810	q_vector->rx.total_bytes = 0;
 5811	q_vector->rx.total_packets = 0;
 5812	q_vector->tx.total_bytes = 0;
 5813	q_vector->tx.total_packets = 0;
 5814}
 5815
 5816/**
 5817 *  igb_update_itr - update the dynamic ITR value based on statistics
 5818 *  @q_vector: pointer to q_vector
 5819 *  @ring_container: ring info to update the itr for
 5820 *
 5821 *  Stores a new ITR value based on packets and byte
 5822 *  counts during the last interrupt.  The advantage of per interrupt
 5823 *  computation is faster updates and more accurate ITR for the current
 5824 *  traffic pattern.  Constants in this function were computed
 5825 *  based on theoretical maximum wire speed and thresholds were set based
 5826 *  on testing data as well as attempting to minimize response time
 5827 *  while increasing bulk throughput.
 5828 *  This functionality is controlled by ethtool's coalescing settings.
 5829 *  NOTE:  These calculations are only valid when operating in a single-
 5830 *         queue environment.
 5831 **/
 5832static void igb_update_itr(struct igb_q_vector *q_vector,
 5833			   struct igb_ring_container *ring_container)
 5834{
 5835	unsigned int packets = ring_container->total_packets;
 5836	unsigned int bytes = ring_container->total_bytes;
 5837	u8 itrval = ring_container->itr;
 5838
 5839	/* no packets, exit with status unchanged */
 5840	if (packets == 0)
 5841		return;
 5842
 5843	switch (itrval) {
 5844	case lowest_latency:
 5845		/* handle TSO and jumbo frames */
 5846		if (bytes/packets > 8000)
 5847			itrval = bulk_latency;
 5848		else if ((packets < 5) && (bytes > 512))
 5849			itrval = low_latency;
 5850		break;
 5851	case low_latency:  /* 50 usec aka 20000 ints/s */
 5852		if (bytes > 10000) {
 5853			/* this if handles the TSO accounting */
 5854			if (bytes/packets > 8000)
 5855				itrval = bulk_latency;
 5856			else if ((packets < 10) || ((bytes/packets) > 1200))
 5857				itrval = bulk_latency;
 5858			else if ((packets > 35))
 5859				itrval = lowest_latency;
 
 5860		} else if (bytes/packets > 2000) {
 5861			itrval = bulk_latency;
 5862		} else if (packets <= 2 && bytes < 512) {
 5863			itrval = lowest_latency;
 5864		}
 5865		break;
 5866	case bulk_latency: /* 250 usec aka 4000 ints/s */
 5867		if (bytes > 25000) {
 5868			if (packets > 35)
 5869				itrval = low_latency;
 5870		} else if (bytes < 1500) {
 5871			itrval = low_latency;
 5872		}
 5873		break;
 5874	}
 5875
 5876	/* clear work counters since we have the values we need */
 5877	ring_container->total_bytes = 0;
 5878	ring_container->total_packets = 0;
 5879
 5880	/* write updated itr to ring container */
 5881	ring_container->itr = itrval;
 5882}
 5883
 5884static void igb_set_itr(struct igb_q_vector *q_vector)
 5885{
 5886	struct igb_adapter *adapter = q_vector->adapter;
 5887	u32 new_itr = q_vector->itr_val;
 5888	u8 current_itr = 0;
 5889
 5890	/* for non-gigabit speeds, just fix the interrupt rate at 4000 */
 5891	if (adapter->link_speed != SPEED_1000) {
 5892		current_itr = 0;
 5893		new_itr = IGB_4K_ITR;
 5894		goto set_itr_now;
 5895	}
 5896
 5897	igb_update_itr(q_vector, &q_vector->tx);
 5898	igb_update_itr(q_vector, &q_vector->rx);
 5899
 5900	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
 5901
 5902	/* conservative mode (itr 3) eliminates the lowest_latency setting */
 5903	if (current_itr == lowest_latency &&
 5904	    ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
 5905	     (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
 5906		current_itr = low_latency;
 5907
 5908	switch (current_itr) {
 5909	/* counts and packets in update_itr are dependent on these numbers */
 5910	case lowest_latency:
 5911		new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
 5912		break;
 5913	case low_latency:
 5914		new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
 5915		break;
 5916	case bulk_latency:
 5917		new_itr = IGB_4K_ITR;  /* 4,000 ints/sec */
 5918		break;
 5919	default:
 5920		break;
 5921	}
 5922
 5923set_itr_now:
 5924	if (new_itr != q_vector->itr_val) {
 5925		/* this attempts to bias the interrupt rate towards Bulk
 5926		 * by adding intermediate steps when interrupt rate is
 5927		 * increasing
 5928		 */
 5929		new_itr = new_itr > q_vector->itr_val ?
 5930			  max((new_itr * q_vector->itr_val) /
 5931			  (new_itr + (q_vector->itr_val >> 2)),
 5932			  new_itr) : new_itr;
 5933		/* Don't write the value here; it resets the adapter's
 5934		 * internal timer, and causes us to delay far longer than
 5935		 * we should between interrupts.  Instead, we write the ITR
 5936		 * value at the beginning of the next interrupt so the timing
 5937		 * ends up being correct.
 5938		 */
 5939		q_vector->itr_val = new_itr;
 5940		q_vector->set_itr = 1;
 5941	}
 5942}
 5943
 5944static void igb_tx_ctxtdesc(struct igb_ring *tx_ring,
 5945			    struct igb_tx_buffer *first,
 5946			    u32 vlan_macip_lens, u32 type_tucmd,
 5947			    u32 mss_l4len_idx)
 5948{
 5949	struct e1000_adv_tx_context_desc *context_desc;
 5950	u16 i = tx_ring->next_to_use;
 5951	struct timespec64 ts;
 5952
 5953	context_desc = IGB_TX_CTXTDESC(tx_ring, i);
 5954
 5955	i++;
 5956	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
 5957
 5958	/* set bits to identify this as an advanced context descriptor */
 5959	type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
 5960
 5961	/* For 82575, context index must be unique per ring. */
 5962	if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
 5963		mss_l4len_idx |= tx_ring->reg_idx << 4;
 5964
 5965	context_desc->vlan_macip_lens	= cpu_to_le32(vlan_macip_lens);
 
 5966	context_desc->type_tucmd_mlhl	= cpu_to_le32(type_tucmd);
 5967	context_desc->mss_l4len_idx	= cpu_to_le32(mss_l4len_idx);
 5968
 5969	/* We assume there is always a valid tx time available. Invalid times
 5970	 * should have been handled by the upper layers.
 5971	 */
 5972	if (tx_ring->launchtime_enable) {
 5973		ts = ktime_to_timespec64(first->skb->tstamp);
 5974		skb_txtime_consumed(first->skb);
 5975		context_desc->seqnum_seed = cpu_to_le32(ts.tv_nsec / 32);
 5976	} else {
 5977		context_desc->seqnum_seed = 0;
 5978	}
 5979}
 5980
 5981static int igb_tso(struct igb_ring *tx_ring,
 5982		   struct igb_tx_buffer *first,
 5983		   u8 *hdr_len)
 5984{
 5985	u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
 5986	struct sk_buff *skb = first->skb;
 5987	union {
 5988		struct iphdr *v4;
 5989		struct ipv6hdr *v6;
 5990		unsigned char *hdr;
 5991	} ip;
 5992	union {
 5993		struct tcphdr *tcp;
 5994		struct udphdr *udp;
 5995		unsigned char *hdr;
 5996	} l4;
 5997	u32 paylen, l4_offset;
 5998	int err;
 5999
 6000	if (skb->ip_summed != CHECKSUM_PARTIAL)
 6001		return 0;
 6002
 6003	if (!skb_is_gso(skb))
 6004		return 0;
 6005
 6006	err = skb_cow_head(skb, 0);
 6007	if (err < 0)
 6008		return err;
 6009
 6010	ip.hdr = skb_network_header(skb);
 6011	l4.hdr = skb_checksum_start(skb);
 6012
 6013	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
 6014	type_tucmd = (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ?
 6015		      E1000_ADVTXD_TUCMD_L4T_UDP : E1000_ADVTXD_TUCMD_L4T_TCP;
 6016
 6017	/* initialize outer IP header fields */
 6018	if (ip.v4->version == 4) {
 6019		unsigned char *csum_start = skb_checksum_start(skb);
 6020		unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
 6021
 6022		/* IP header will have to cancel out any data that
 6023		 * is not a part of the outer IP header
 6024		 */
 6025		ip.v4->check = csum_fold(csum_partial(trans_start,
 6026						      csum_start - trans_start,
 6027						      0));
 6028		type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
 6029
 6030		ip.v4->tot_len = 0;
 6031		first->tx_flags |= IGB_TX_FLAGS_TSO |
 6032				   IGB_TX_FLAGS_CSUM |
 6033				   IGB_TX_FLAGS_IPV4;
 6034	} else {
 6035		ip.v6->payload_len = 0;
 
 
 
 6036		first->tx_flags |= IGB_TX_FLAGS_TSO |
 6037				   IGB_TX_FLAGS_CSUM;
 6038	}
 6039
 6040	/* determine offset of inner transport header */
 6041	l4_offset = l4.hdr - skb->data;
 6042
 6043	/* remove payload length from inner checksum */
 6044	paylen = skb->len - l4_offset;
 6045	if (type_tucmd & E1000_ADVTXD_TUCMD_L4T_TCP) {
 6046		/* compute length of segmentation header */
 6047		*hdr_len = (l4.tcp->doff * 4) + l4_offset;
 6048		csum_replace_by_diff(&l4.tcp->check,
 6049			(__force __wsum)htonl(paylen));
 6050	} else {
 6051		/* compute length of segmentation header */
 6052		*hdr_len = sizeof(*l4.udp) + l4_offset;
 6053		csum_replace_by_diff(&l4.udp->check,
 6054				     (__force __wsum)htonl(paylen));
 6055	}
 6056
 6057	/* update gso size and bytecount with header size */
 6058	first->gso_segs = skb_shinfo(skb)->gso_segs;
 6059	first->bytecount += (first->gso_segs - 1) * *hdr_len;
 6060
 6061	/* MSS L4LEN IDX */
 6062	mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT;
 6063	mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
 6064
 6065	/* VLAN MACLEN IPLEN */
 6066	vlan_macip_lens = l4.hdr - ip.hdr;
 6067	vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT;
 6068	vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
 6069
 6070	igb_tx_ctxtdesc(tx_ring, first, vlan_macip_lens,
 6071			type_tucmd, mss_l4len_idx);
 6072
 6073	return 1;
 6074}
 6075
 6076static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
 6077{
 6078	struct sk_buff *skb = first->skb;
 6079	u32 vlan_macip_lens = 0;
 
 6080	u32 type_tucmd = 0;
 6081
 6082	if (skb->ip_summed != CHECKSUM_PARTIAL) {
 6083csum_failed:
 6084		if (!(first->tx_flags & IGB_TX_FLAGS_VLAN) &&
 6085		    !tx_ring->launchtime_enable)
 6086			return;
 6087		goto no_csum;
 6088	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 6089
 6090	switch (skb->csum_offset) {
 6091	case offsetof(struct tcphdr, check):
 6092		type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
 6093		fallthrough;
 6094	case offsetof(struct udphdr, check):
 6095		break;
 6096	case offsetof(struct sctphdr, checksum):
 6097		/* validate that this is actually an SCTP request */
 6098		if (skb_csum_is_sctp(skb)) {
 6099			type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP;
 
 
 
 
 
 
 
 
 
 
 
 6100			break;
 6101		}
 6102		fallthrough;
 6103	default:
 6104		skb_checksum_help(skb);
 6105		goto csum_failed;
 6106	}
 6107
 6108	/* update TX checksum flag */
 6109	first->tx_flags |= IGB_TX_FLAGS_CSUM;
 6110	vlan_macip_lens = skb_checksum_start_offset(skb) -
 6111			  skb_network_offset(skb);
 6112no_csum:
 6113	vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
 6114	vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
 6115
 6116	igb_tx_ctxtdesc(tx_ring, first, vlan_macip_lens, type_tucmd, 0);
 6117}
 6118
 6119#define IGB_SET_FLAG(_input, _flag, _result) \
 6120	((_flag <= _result) ? \
 6121	 ((u32)(_input & _flag) * (_result / _flag)) : \
 6122	 ((u32)(_input & _flag) / (_flag / _result)))
 6123
 6124static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
 6125{
 6126	/* set type for advanced descriptor with frame checksum insertion */
 6127	u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
 6128		       E1000_ADVTXD_DCMD_DEXT |
 6129		       E1000_ADVTXD_DCMD_IFCS;
 6130
 6131	/* set HW vlan bit if vlan is present */
 6132	cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
 6133				 (E1000_ADVTXD_DCMD_VLE));
 6134
 6135	/* set segmentation bits for TSO */
 6136	cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
 6137				 (E1000_ADVTXD_DCMD_TSE));
 6138
 6139	/* set timestamp bit if present */
 6140	cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
 6141				 (E1000_ADVTXD_MAC_TSTAMP));
 6142
 6143	/* insert frame checksum */
 6144	cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
 6145
 6146	return cmd_type;
 6147}
 6148
 6149static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
 6150				 union e1000_adv_tx_desc *tx_desc,
 6151				 u32 tx_flags, unsigned int paylen)
 6152{
 6153	u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
 6154
 6155	/* 82575 requires a unique index per ring */
 6156	if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
 6157		olinfo_status |= tx_ring->reg_idx << 4;
 6158
 6159	/* insert L4 checksum */
 6160	olinfo_status |= IGB_SET_FLAG(tx_flags,
 6161				      IGB_TX_FLAGS_CSUM,
 6162				      (E1000_TXD_POPTS_TXSM << 8));
 6163
 6164	/* insert IPv4 checksum */
 6165	olinfo_status |= IGB_SET_FLAG(tx_flags,
 6166				      IGB_TX_FLAGS_IPV4,
 6167				      (E1000_TXD_POPTS_IXSM << 8));
 6168
 6169	tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
 6170}
 6171
 6172static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
 6173{
 6174	struct net_device *netdev = tx_ring->netdev;
 6175
 6176	netif_stop_subqueue(netdev, tx_ring->queue_index);
 6177
 6178	/* Herbert's original patch had:
 6179	 *  smp_mb__after_netif_stop_queue();
 6180	 * but since that doesn't exist yet, just open code it.
 6181	 */
 6182	smp_mb();
 6183
 6184	/* We need to check again in a case another CPU has just
 6185	 * made room available.
 6186	 */
 6187	if (igb_desc_unused(tx_ring) < size)
 6188		return -EBUSY;
 6189
 6190	/* A reprieve! */
 6191	netif_wake_subqueue(netdev, tx_ring->queue_index);
 6192
 6193	u64_stats_update_begin(&tx_ring->tx_syncp2);
 6194	tx_ring->tx_stats.restart_queue2++;
 6195	u64_stats_update_end(&tx_ring->tx_syncp2);
 6196
 6197	return 0;
 6198}
 6199
 6200static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
 6201{
 6202	if (igb_desc_unused(tx_ring) >= size)
 6203		return 0;
 6204	return __igb_maybe_stop_tx(tx_ring, size);
 6205}
 6206
 6207static int igb_tx_map(struct igb_ring *tx_ring,
 6208		      struct igb_tx_buffer *first,
 6209		      const u8 hdr_len)
 6210{
 6211	struct sk_buff *skb = first->skb;
 6212	struct igb_tx_buffer *tx_buffer;
 6213	union e1000_adv_tx_desc *tx_desc;
 6214	skb_frag_t *frag;
 6215	dma_addr_t dma;
 6216	unsigned int data_len, size;
 6217	u32 tx_flags = first->tx_flags;
 6218	u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
 6219	u16 i = tx_ring->next_to_use;
 6220
 6221	tx_desc = IGB_TX_DESC(tx_ring, i);
 6222
 6223	igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
 6224
 6225	size = skb_headlen(skb);
 6226	data_len = skb->data_len;
 6227
 6228	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
 6229
 6230	tx_buffer = first;
 6231
 6232	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
 6233		if (dma_mapping_error(tx_ring->dev, dma))
 6234			goto dma_error;
 6235
 6236		/* record length, and DMA address */
 6237		dma_unmap_len_set(tx_buffer, len, size);
 6238		dma_unmap_addr_set(tx_buffer, dma, dma);
 6239
 6240		tx_desc->read.buffer_addr = cpu_to_le64(dma);
 6241
 6242		while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
 6243			tx_desc->read.cmd_type_len =
 6244				cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
 6245
 6246			i++;
 6247			tx_desc++;
 6248			if (i == tx_ring->count) {
 6249				tx_desc = IGB_TX_DESC(tx_ring, 0);
 6250				i = 0;
 6251			}
 6252			tx_desc->read.olinfo_status = 0;
 6253
 6254			dma += IGB_MAX_DATA_PER_TXD;
 6255			size -= IGB_MAX_DATA_PER_TXD;
 6256
 6257			tx_desc->read.buffer_addr = cpu_to_le64(dma);
 6258		}
 6259
 6260		if (likely(!data_len))
 6261			break;
 6262
 6263		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
 6264
 6265		i++;
 6266		tx_desc++;
 6267		if (i == tx_ring->count) {
 6268			tx_desc = IGB_TX_DESC(tx_ring, 0);
 6269			i = 0;
 6270		}
 6271		tx_desc->read.olinfo_status = 0;
 6272
 6273		size = skb_frag_size(frag);
 6274		data_len -= size;
 6275
 6276		dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
 6277				       size, DMA_TO_DEVICE);
 6278
 6279		tx_buffer = &tx_ring->tx_buffer_info[i];
 6280	}
 6281
 6282	/* write last descriptor with RS and EOP bits */
 6283	cmd_type |= size | IGB_TXD_DCMD;
 6284	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
 6285
 6286	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
 6287
 6288	/* set the timestamp */
 6289	first->time_stamp = jiffies;
 6290
 6291	skb_tx_timestamp(skb);
 6292
 6293	/* Force memory writes to complete before letting h/w know there
 6294	 * are new descriptors to fetch.  (Only applicable for weak-ordered
 6295	 * memory model archs, such as IA-64).
 6296	 *
 6297	 * We also need this memory barrier to make certain all of the
 6298	 * status bits have been updated before next_to_watch is written.
 6299	 */
 6300	dma_wmb();
 6301
 6302	/* set next_to_watch value indicating a packet is present */
 6303	first->next_to_watch = tx_desc;
 6304
 6305	i++;
 6306	if (i == tx_ring->count)
 6307		i = 0;
 6308
 6309	tx_ring->next_to_use = i;
 6310
 6311	/* Make sure there is space in the ring for the next send. */
 6312	igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
 6313
 6314	if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
 6315		writel(i, tx_ring->tail);
 6316	}
 6317	return 0;
 
 
 6318
 6319dma_error:
 6320	dev_err(tx_ring->dev, "TX DMA map failed\n");
 6321	tx_buffer = &tx_ring->tx_buffer_info[i];
 6322
 6323	/* clear dma mappings for failed tx_buffer_info map */
 6324	while (tx_buffer != first) {
 6325		if (dma_unmap_len(tx_buffer, len))
 6326			dma_unmap_page(tx_ring->dev,
 6327				       dma_unmap_addr(tx_buffer, dma),
 6328				       dma_unmap_len(tx_buffer, len),
 6329				       DMA_TO_DEVICE);
 6330		dma_unmap_len_set(tx_buffer, len, 0);
 6331
 6332		if (i-- == 0)
 6333			i += tx_ring->count;
 6334		tx_buffer = &tx_ring->tx_buffer_info[i];
 
 
 
 
 
 
 6335	}
 6336
 6337	if (dma_unmap_len(tx_buffer, len))
 6338		dma_unmap_single(tx_ring->dev,
 6339				 dma_unmap_addr(tx_buffer, dma),
 6340				 dma_unmap_len(tx_buffer, len),
 6341				 DMA_TO_DEVICE);
 6342	dma_unmap_len_set(tx_buffer, len, 0);
 6343
 6344	dev_kfree_skb_any(tx_buffer->skb);
 6345	tx_buffer->skb = NULL;
 6346
 6347	tx_ring->next_to_use = i;
 6348
 6349	return -1;
 6350}
 6351
 6352int igb_xmit_xdp_ring(struct igb_adapter *adapter,
 6353		      struct igb_ring *tx_ring,
 6354		      struct xdp_frame *xdpf)
 6355{
 6356	struct skb_shared_info *sinfo = xdp_get_shared_info_from_frame(xdpf);
 6357	u8 nr_frags = unlikely(xdp_frame_has_frags(xdpf)) ? sinfo->nr_frags : 0;
 6358	u16 count, i, index = tx_ring->next_to_use;
 6359	struct igb_tx_buffer *tx_head = &tx_ring->tx_buffer_info[index];
 6360	struct igb_tx_buffer *tx_buffer = tx_head;
 6361	union e1000_adv_tx_desc *tx_desc = IGB_TX_DESC(tx_ring, index);
 6362	u32 len = xdpf->len, cmd_type, olinfo_status;
 6363	void *data = xdpf->data;
 6364
 6365	count = TXD_USE_COUNT(len);
 6366	for (i = 0; i < nr_frags; i++)
 6367		count += TXD_USE_COUNT(skb_frag_size(&sinfo->frags[i]));
 6368
 6369	if (igb_maybe_stop_tx(tx_ring, count + 3))
 6370		return IGB_XDP_CONSUMED;
 6371
 6372	i = 0;
 6373	/* record the location of the first descriptor for this packet */
 6374	tx_head->bytecount = xdp_get_frame_len(xdpf);
 6375	tx_head->type = IGB_TYPE_XDP;
 6376	tx_head->gso_segs = 1;
 6377	tx_head->xdpf = xdpf;
 6378
 6379	olinfo_status = tx_head->bytecount << E1000_ADVTXD_PAYLEN_SHIFT;
 6380	/* 82575 requires a unique index per ring */
 6381	if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
 6382		olinfo_status |= tx_ring->reg_idx << 4;
 6383	tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
 6384
 6385	for (;;) {
 6386		dma_addr_t dma;
 6387
 6388		dma = dma_map_single(tx_ring->dev, data, len, DMA_TO_DEVICE);
 6389		if (dma_mapping_error(tx_ring->dev, dma))
 6390			goto unmap;
 6391
 6392		/* record length, and DMA address */
 6393		dma_unmap_len_set(tx_buffer, len, len);
 6394		dma_unmap_addr_set(tx_buffer, dma, dma);
 6395
 6396		/* put descriptor type bits */
 6397		cmd_type = E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_DEXT |
 6398			   E1000_ADVTXD_DCMD_IFCS | len;
 6399
 6400		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
 6401		tx_desc->read.buffer_addr = cpu_to_le64(dma);
 6402
 6403		tx_buffer->protocol = 0;
 6404
 6405		if (++index == tx_ring->count)
 6406			index = 0;
 6407
 6408		if (i == nr_frags)
 6409			break;
 6410
 6411		tx_buffer = &tx_ring->tx_buffer_info[index];
 6412		tx_desc = IGB_TX_DESC(tx_ring, index);
 6413		tx_desc->read.olinfo_status = 0;
 6414
 6415		data = skb_frag_address(&sinfo->frags[i]);
 6416		len = skb_frag_size(&sinfo->frags[i]);
 6417		i++;
 6418	}
 6419	tx_desc->read.cmd_type_len |= cpu_to_le32(IGB_TXD_DCMD);
 6420
 6421	netdev_tx_sent_queue(txring_txq(tx_ring), tx_head->bytecount);
 6422	/* set the timestamp */
 6423	tx_head->time_stamp = jiffies;
 6424
 6425	/* Avoid any potential race with xdp_xmit and cleanup */
 6426	smp_wmb();
 6427
 6428	/* set next_to_watch value indicating a packet is present */
 6429	tx_head->next_to_watch = tx_desc;
 6430	tx_ring->next_to_use = index;
 6431
 6432	/* Make sure there is space in the ring for the next send. */
 6433	igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
 
 
 
 6434
 6435	if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more())
 6436		writel(index, tx_ring->tail);
 
 
 
 6437
 6438	return IGB_XDP_TX;
 
 6439
 6440unmap:
 6441	for (;;) {
 6442		tx_buffer = &tx_ring->tx_buffer_info[index];
 6443		if (dma_unmap_len(tx_buffer, len))
 6444			dma_unmap_page(tx_ring->dev,
 6445				       dma_unmap_addr(tx_buffer, dma),
 6446				       dma_unmap_len(tx_buffer, len),
 6447				       DMA_TO_DEVICE);
 6448		dma_unmap_len_set(tx_buffer, len, 0);
 6449		if (tx_buffer == tx_head)
 6450			break;
 6451
 6452		if (!index)
 6453			index += tx_ring->count;
 6454		index--;
 6455	}
 6456
 6457	return IGB_XDP_CONSUMED;
 
 
 
 
 6458}
 6459
 6460netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
 6461				struct igb_ring *tx_ring)
 6462{
 6463	struct igb_tx_buffer *first;
 6464	int tso;
 6465	u32 tx_flags = 0;
 6466	unsigned short f;
 6467	u16 count = TXD_USE_COUNT(skb_headlen(skb));
 6468	__be16 protocol = vlan_get_protocol(skb);
 6469	u8 hdr_len = 0;
 6470
 6471	/* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
 6472	 *       + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
 6473	 *       + 2 desc gap to keep tail from touching head,
 6474	 *       + 1 desc for context descriptor,
 6475	 * otherwise try next time
 6476	 */
 6477	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
 6478		count += TXD_USE_COUNT(skb_frag_size(
 6479						&skb_shinfo(skb)->frags[f]));
 
 
 
 
 6480
 6481	if (igb_maybe_stop_tx(tx_ring, count + 3)) {
 6482		/* this is a hard error */
 6483		return NETDEV_TX_BUSY;
 6484	}
 6485
 6486	/* record the location of the first descriptor for this packet */
 6487	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
 6488	first->type = IGB_TYPE_SKB;
 6489	first->skb = skb;
 6490	first->bytecount = skb->len;
 6491	first->gso_segs = 1;
 6492
 6493	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
 6494		struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
 6495
 6496		if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
 6497		    !test_and_set_bit_lock(__IGB_PTP_TX_IN_PROGRESS,
 6498					   &adapter->state)) {
 6499			skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
 6500			tx_flags |= IGB_TX_FLAGS_TSTAMP;
 6501
 6502			adapter->ptp_tx_skb = skb_get(skb);
 6503			adapter->ptp_tx_start = jiffies;
 6504			if (adapter->hw.mac.type == e1000_82576)
 6505				schedule_work(&adapter->ptp_tx_work);
 6506		} else {
 6507			adapter->tx_hwtstamp_skipped++;
 6508		}
 6509	}
 6510
 6511	if (skb_vlan_tag_present(skb)) {
 
 
 6512		tx_flags |= IGB_TX_FLAGS_VLAN;
 6513		tx_flags |= (skb_vlan_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
 6514	}
 6515
 6516	/* record initial flags and protocol */
 6517	first->tx_flags = tx_flags;
 6518	first->protocol = protocol;
 6519
 6520	tso = igb_tso(tx_ring, first, &hdr_len);
 6521	if (tso < 0)
 6522		goto out_drop;
 6523	else if (!tso)
 6524		igb_tx_csum(tx_ring, first);
 6525
 6526	if (igb_tx_map(tx_ring, first, hdr_len))
 6527		goto cleanup_tx_tstamp;
 
 
 6528
 6529	return NETDEV_TX_OK;
 6530
 6531out_drop:
 6532	dev_kfree_skb_any(first->skb);
 6533	first->skb = NULL;
 6534cleanup_tx_tstamp:
 6535	if (unlikely(tx_flags & IGB_TX_FLAGS_TSTAMP)) {
 6536		struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
 6537
 6538		dev_kfree_skb_any(adapter->ptp_tx_skb);
 6539		adapter->ptp_tx_skb = NULL;
 6540		if (adapter->hw.mac.type == e1000_82576)
 6541			cancel_work_sync(&adapter->ptp_tx_work);
 6542		clear_bit_unlock(__IGB_PTP_TX_IN_PROGRESS, &adapter->state);
 6543	}
 6544
 6545	return NETDEV_TX_OK;
 6546}
 6547
 6548static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
 6549						    struct sk_buff *skb)
 6550{
 6551	unsigned int r_idx = skb->queue_mapping;
 6552
 6553	if (r_idx >= adapter->num_tx_queues)
 6554		r_idx = r_idx % adapter->num_tx_queues;
 6555
 6556	return adapter->tx_ring[r_idx];
 6557}
 6558
 6559static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
 6560				  struct net_device *netdev)
 6561{
 6562	struct igb_adapter *adapter = netdev_priv(netdev);
 6563
 
 
 
 
 
 
 
 
 
 
 6564	/* The minimum packet size with TCTL.PSP set is 17 so pad the skb
 6565	 * in order to meet this minimum size requirement.
 6566	 */
 6567	if (skb_put_padto(skb, 17))
 6568		return NETDEV_TX_OK;
 
 
 
 
 6569
 6570	return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
 6571}
 6572
 6573/**
 6574 *  igb_tx_timeout - Respond to a Tx Hang
 6575 *  @netdev: network interface device structure
 6576 *  @txqueue: number of the Tx queue that hung (unused)
 6577 **/
 6578static void igb_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
 6579{
 6580	struct igb_adapter *adapter = netdev_priv(netdev);
 6581	struct e1000_hw *hw = &adapter->hw;
 6582
 6583	/* Do the reset outside of interrupt context */
 6584	adapter->tx_timeout_count++;
 6585
 6586	if (hw->mac.type >= e1000_82580)
 6587		hw->dev_spec._82575.global_device_reset = true;
 6588
 6589	schedule_work(&adapter->reset_task);
 6590	wr32(E1000_EICS,
 6591	     (adapter->eims_enable_mask & ~adapter->eims_other));
 6592}
 6593
 6594static void igb_reset_task(struct work_struct *work)
 6595{
 6596	struct igb_adapter *adapter;
 6597	adapter = container_of(work, struct igb_adapter, reset_task);
 6598
 6599	rtnl_lock();
 6600	/* If we're already down or resetting, just bail */
 6601	if (test_bit(__IGB_DOWN, &adapter->state) ||
 6602	    test_bit(__IGB_RESETTING, &adapter->state)) {
 6603		rtnl_unlock();
 6604		return;
 6605	}
 6606
 6607	igb_dump(adapter);
 6608	netdev_err(adapter->netdev, "Reset adapter\n");
 6609	igb_reinit_locked(adapter);
 6610	rtnl_unlock();
 6611}
 6612
 6613/**
 6614 *  igb_get_stats64 - Get System Network Statistics
 6615 *  @netdev: network interface device structure
 6616 *  @stats: rtnl_link_stats64 pointer
 6617 **/
 6618static void igb_get_stats64(struct net_device *netdev,
 6619			    struct rtnl_link_stats64 *stats)
 6620{
 6621	struct igb_adapter *adapter = netdev_priv(netdev);
 6622
 6623	spin_lock(&adapter->stats64_lock);
 6624	igb_update_stats(adapter);
 6625	memcpy(stats, &adapter->stats64, sizeof(*stats));
 6626	spin_unlock(&adapter->stats64_lock);
 
 
 6627}
 6628
 6629/**
 6630 *  igb_change_mtu - Change the Maximum Transfer Unit
 6631 *  @netdev: network interface device structure
 6632 *  @new_mtu: new value for maximum frame size
 6633 *
 6634 *  Returns 0 on success, negative on failure
 6635 **/
 6636static int igb_change_mtu(struct net_device *netdev, int new_mtu)
 6637{
 6638	struct igb_adapter *adapter = netdev_priv(netdev);
 6639	int max_frame = new_mtu + IGB_ETH_PKT_HDR_PAD;
 6640
 6641	if (adapter->xdp_prog) {
 6642		int i;
 6643
 6644		for (i = 0; i < adapter->num_rx_queues; i++) {
 6645			struct igb_ring *ring = adapter->rx_ring[i];
 
 
 6646
 6647			if (max_frame > igb_rx_bufsz(ring)) {
 6648				netdev_warn(adapter->netdev,
 6649					    "Requested MTU size is not supported with XDP. Max frame size is %d\n",
 6650					    max_frame);
 6651				return -EINVAL;
 6652			}
 6653		}
 6654	}
 6655
 6656	/* adjust max frame to be at least the size of a standard frame */
 6657	if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
 6658		max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
 6659
 6660	while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
 6661		usleep_range(1000, 2000);
 6662
 6663	/* igb_down has a dependency on max_frame_size */
 6664	adapter->max_frame_size = max_frame;
 6665
 6666	if (netif_running(netdev))
 6667		igb_down(adapter);
 6668
 6669	netdev_dbg(netdev, "changing MTU from %d to %d\n",
 6670		   netdev->mtu, new_mtu);
 6671	netdev->mtu = new_mtu;
 6672
 6673	if (netif_running(netdev))
 6674		igb_up(adapter);
 6675	else
 6676		igb_reset(adapter);
 6677
 6678	clear_bit(__IGB_RESETTING, &adapter->state);
 6679
 6680	return 0;
 6681}
 6682
 6683/**
 6684 *  igb_update_stats - Update the board statistics counters
 6685 *  @adapter: board private structure
 6686 **/
 6687void igb_update_stats(struct igb_adapter *adapter)
 
 6688{
 6689	struct rtnl_link_stats64 *net_stats = &adapter->stats64;
 6690	struct e1000_hw *hw = &adapter->hw;
 6691	struct pci_dev *pdev = adapter->pdev;
 6692	u32 reg, mpc;
 
 6693	int i;
 6694	u64 bytes, packets;
 6695	unsigned int start;
 6696	u64 _bytes, _packets;
 6697
 
 
 6698	/* Prevent stats update while adapter is being reset, or if the pci
 6699	 * connection is down.
 6700	 */
 6701	if (adapter->link_speed == 0)
 6702		return;
 6703	if (pci_channel_offline(pdev))
 6704		return;
 6705
 6706	bytes = 0;
 6707	packets = 0;
 6708
 6709	rcu_read_lock();
 6710	for (i = 0; i < adapter->num_rx_queues; i++) {
 6711		struct igb_ring *ring = adapter->rx_ring[i];
 6712		u32 rqdpc = rd32(E1000_RQDPC(i));
 6713		if (hw->mac.type >= e1000_i210)
 6714			wr32(E1000_RQDPC(i), 0);
 6715
 6716		if (rqdpc) {
 6717			ring->rx_stats.drops += rqdpc;
 6718			net_stats->rx_fifo_errors += rqdpc;
 6719		}
 6720
 6721		do {
 6722			start = u64_stats_fetch_begin(&ring->rx_syncp);
 6723			_bytes = ring->rx_stats.bytes;
 6724			_packets = ring->rx_stats.packets;
 6725		} while (u64_stats_fetch_retry(&ring->rx_syncp, start));
 6726		bytes += _bytes;
 6727		packets += _packets;
 6728	}
 6729
 6730	net_stats->rx_bytes = bytes;
 6731	net_stats->rx_packets = packets;
 6732
 6733	bytes = 0;
 6734	packets = 0;
 6735	for (i = 0; i < adapter->num_tx_queues; i++) {
 6736		struct igb_ring *ring = adapter->tx_ring[i];
 6737		do {
 6738			start = u64_stats_fetch_begin(&ring->tx_syncp);
 6739			_bytes = ring->tx_stats.bytes;
 6740			_packets = ring->tx_stats.packets;
 6741		} while (u64_stats_fetch_retry(&ring->tx_syncp, start));
 6742		bytes += _bytes;
 6743		packets += _packets;
 6744	}
 6745	net_stats->tx_bytes = bytes;
 6746	net_stats->tx_packets = packets;
 6747	rcu_read_unlock();
 6748
 6749	/* read stats registers */
 6750	adapter->stats.crcerrs += rd32(E1000_CRCERRS);
 6751	adapter->stats.gprc += rd32(E1000_GPRC);
 6752	adapter->stats.gorc += rd32(E1000_GORCL);
 6753	rd32(E1000_GORCH); /* clear GORCL */
 6754	adapter->stats.bprc += rd32(E1000_BPRC);
 6755	adapter->stats.mprc += rd32(E1000_MPRC);
 6756	adapter->stats.roc += rd32(E1000_ROC);
 6757
 6758	adapter->stats.prc64 += rd32(E1000_PRC64);
 6759	adapter->stats.prc127 += rd32(E1000_PRC127);
 6760	adapter->stats.prc255 += rd32(E1000_PRC255);
 6761	adapter->stats.prc511 += rd32(E1000_PRC511);
 6762	adapter->stats.prc1023 += rd32(E1000_PRC1023);
 6763	adapter->stats.prc1522 += rd32(E1000_PRC1522);
 6764	adapter->stats.symerrs += rd32(E1000_SYMERRS);
 6765	adapter->stats.sec += rd32(E1000_SEC);
 6766
 6767	mpc = rd32(E1000_MPC);
 6768	adapter->stats.mpc += mpc;
 6769	net_stats->rx_fifo_errors += mpc;
 6770	adapter->stats.scc += rd32(E1000_SCC);
 6771	adapter->stats.ecol += rd32(E1000_ECOL);
 6772	adapter->stats.mcc += rd32(E1000_MCC);
 6773	adapter->stats.latecol += rd32(E1000_LATECOL);
 6774	adapter->stats.dc += rd32(E1000_DC);
 6775	adapter->stats.rlec += rd32(E1000_RLEC);
 6776	adapter->stats.xonrxc += rd32(E1000_XONRXC);
 6777	adapter->stats.xontxc += rd32(E1000_XONTXC);
 6778	adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
 6779	adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
 6780	adapter->stats.fcruc += rd32(E1000_FCRUC);
 6781	adapter->stats.gptc += rd32(E1000_GPTC);
 6782	adapter->stats.gotc += rd32(E1000_GOTCL);
 6783	rd32(E1000_GOTCH); /* clear GOTCL */
 6784	adapter->stats.rnbc += rd32(E1000_RNBC);
 6785	adapter->stats.ruc += rd32(E1000_RUC);
 6786	adapter->stats.rfc += rd32(E1000_RFC);
 6787	adapter->stats.rjc += rd32(E1000_RJC);
 6788	adapter->stats.tor += rd32(E1000_TORH);
 6789	adapter->stats.tot += rd32(E1000_TOTH);
 6790	adapter->stats.tpr += rd32(E1000_TPR);
 6791
 6792	adapter->stats.ptc64 += rd32(E1000_PTC64);
 6793	adapter->stats.ptc127 += rd32(E1000_PTC127);
 6794	adapter->stats.ptc255 += rd32(E1000_PTC255);
 6795	adapter->stats.ptc511 += rd32(E1000_PTC511);
 6796	adapter->stats.ptc1023 += rd32(E1000_PTC1023);
 6797	adapter->stats.ptc1522 += rd32(E1000_PTC1522);
 6798
 6799	adapter->stats.mptc += rd32(E1000_MPTC);
 6800	adapter->stats.bptc += rd32(E1000_BPTC);
 6801
 6802	adapter->stats.tpt += rd32(E1000_TPT);
 6803	adapter->stats.colc += rd32(E1000_COLC);
 6804
 6805	adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
 6806	/* read internal phy specific stats */
 6807	reg = rd32(E1000_CTRL_EXT);
 6808	if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
 6809		adapter->stats.rxerrc += rd32(E1000_RXERRC);
 6810
 6811		/* this stat has invalid values on i210/i211 */
 6812		if ((hw->mac.type != e1000_i210) &&
 6813		    (hw->mac.type != e1000_i211))
 6814			adapter->stats.tncrs += rd32(E1000_TNCRS);
 6815	}
 6816
 6817	adapter->stats.tsctc += rd32(E1000_TSCTC);
 6818	adapter->stats.tsctfc += rd32(E1000_TSCTFC);
 6819
 6820	adapter->stats.iac += rd32(E1000_IAC);
 6821	adapter->stats.icrxoc += rd32(E1000_ICRXOC);
 6822	adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
 6823	adapter->stats.icrxatc += rd32(E1000_ICRXATC);
 6824	adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
 6825	adapter->stats.ictxatc += rd32(E1000_ICTXATC);
 6826	adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
 6827	adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
 6828	adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
 6829
 6830	/* Fill out the OS statistics structure */
 6831	net_stats->multicast = adapter->stats.mprc;
 6832	net_stats->collisions = adapter->stats.colc;
 6833
 6834	/* Rx Errors */
 6835
 6836	/* RLEC on some newer hardware can be incorrect so build
 6837	 * our own version based on RUC and ROC
 6838	 */
 6839	net_stats->rx_errors = adapter->stats.rxerrc +
 6840		adapter->stats.crcerrs + adapter->stats.algnerrc +
 6841		adapter->stats.ruc + adapter->stats.roc +
 6842		adapter->stats.cexterr;
 6843	net_stats->rx_length_errors = adapter->stats.ruc +
 6844				      adapter->stats.roc;
 6845	net_stats->rx_crc_errors = adapter->stats.crcerrs;
 6846	net_stats->rx_frame_errors = adapter->stats.algnerrc;
 6847	net_stats->rx_missed_errors = adapter->stats.mpc;
 6848
 6849	/* Tx Errors */
 6850	net_stats->tx_errors = adapter->stats.ecol +
 6851			       adapter->stats.latecol;
 6852	net_stats->tx_aborted_errors = adapter->stats.ecol;
 6853	net_stats->tx_window_errors = adapter->stats.latecol;
 6854	net_stats->tx_carrier_errors = adapter->stats.tncrs;
 6855
 6856	/* Tx Dropped needs to be maintained elsewhere */
 6857
 
 
 
 
 
 
 
 
 
 6858	/* Management Stats */
 6859	adapter->stats.mgptc += rd32(E1000_MGTPTC);
 6860	adapter->stats.mgprc += rd32(E1000_MGTPRC);
 6861	adapter->stats.mgpdc += rd32(E1000_MGTPDC);
 6862
 6863	/* OS2BMC Stats */
 6864	reg = rd32(E1000_MANC);
 6865	if (reg & E1000_MANC_EN_BMC2OS) {
 6866		adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
 6867		adapter->stats.o2bspc += rd32(E1000_O2BSPC);
 6868		adapter->stats.b2ospc += rd32(E1000_B2OSPC);
 6869		adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
 6870	}
 6871}
 6872
 6873static void igb_perout(struct igb_adapter *adapter, int tsintr_tt)
 6874{
 6875	int pin = ptp_find_pin(adapter->ptp_clock, PTP_PF_PEROUT, tsintr_tt);
 6876	struct e1000_hw *hw = &adapter->hw;
 6877	struct timespec64 ts;
 6878	u32 tsauxc;
 6879
 6880	if (pin < 0 || pin >= IGB_N_SDP)
 6881		return;
 6882
 6883	spin_lock(&adapter->tmreg_lock);
 6884
 6885	if (hw->mac.type == e1000_82580 ||
 6886	    hw->mac.type == e1000_i354 ||
 6887	    hw->mac.type == e1000_i350) {
 6888		s64 ns = timespec64_to_ns(&adapter->perout[tsintr_tt].period);
 6889		u32 systiml, systimh, level_mask, level, rem;
 6890		u64 systim, now;
 6891
 6892		/* read systim registers in sequence */
 6893		rd32(E1000_SYSTIMR);
 6894		systiml = rd32(E1000_SYSTIML);
 6895		systimh = rd32(E1000_SYSTIMH);
 6896		systim = (((u64)(systimh & 0xFF)) << 32) | ((u64)systiml);
 6897		now = timecounter_cyc2time(&adapter->tc, systim);
 6898
 6899		if (pin < 2) {
 6900			level_mask = (tsintr_tt == 1) ? 0x80000 : 0x40000;
 6901			level = (rd32(E1000_CTRL) & level_mask) ? 1 : 0;
 6902		} else {
 6903			level_mask = (tsintr_tt == 1) ? 0x80 : 0x40;
 6904			level = (rd32(E1000_CTRL_EXT) & level_mask) ? 1 : 0;
 6905		}
 6906
 6907		div_u64_rem(now, ns, &rem);
 6908		systim = systim + (ns - rem);
 6909
 6910		/* synchronize pin level with rising/falling edges */
 6911		div_u64_rem(now, ns << 1, &rem);
 6912		if (rem < ns) {
 6913			/* first half of period */
 6914			if (level == 0) {
 6915				/* output is already low, skip this period */
 6916				systim += ns;
 6917				pr_notice("igb: periodic output on %s missed falling edge\n",
 6918					  adapter->sdp_config[pin].name);
 6919			}
 6920		} else {
 6921			/* second half of period */
 6922			if (level == 1) {
 6923				/* output is already high, skip this period */
 6924				systim += ns;
 6925				pr_notice("igb: periodic output on %s missed rising edge\n",
 6926					  adapter->sdp_config[pin].name);
 6927			}
 6928		}
 6929
 6930		/* for this chip family tv_sec is the upper part of the binary value,
 6931		 * so not seconds
 6932		 */
 6933		ts.tv_nsec = (u32)systim;
 6934		ts.tv_sec  = ((u32)(systim >> 32)) & 0xFF;
 6935	} else {
 6936		ts = timespec64_add(adapter->perout[tsintr_tt].start,
 6937				    adapter->perout[tsintr_tt].period);
 6938	}
 6939
 6940	/* u32 conversion of tv_sec is safe until y2106 */
 6941	wr32((tsintr_tt == 1) ? E1000_TRGTTIML1 : E1000_TRGTTIML0, ts.tv_nsec);
 6942	wr32((tsintr_tt == 1) ? E1000_TRGTTIMH1 : E1000_TRGTTIMH0, (u32)ts.tv_sec);
 6943	tsauxc = rd32(E1000_TSAUXC);
 6944	tsauxc |= TSAUXC_EN_TT0;
 6945	wr32(E1000_TSAUXC, tsauxc);
 6946	adapter->perout[tsintr_tt].start = ts;
 6947
 6948	spin_unlock(&adapter->tmreg_lock);
 6949}
 6950
 6951static void igb_extts(struct igb_adapter *adapter, int tsintr_tt)
 6952{
 6953	int pin = ptp_find_pin(adapter->ptp_clock, PTP_PF_EXTTS, tsintr_tt);
 6954	int auxstmpl = (tsintr_tt == 1) ? E1000_AUXSTMPL1 : E1000_AUXSTMPL0;
 6955	int auxstmph = (tsintr_tt == 1) ? E1000_AUXSTMPH1 : E1000_AUXSTMPH0;
 6956	struct e1000_hw *hw = &adapter->hw;
 6957	struct ptp_clock_event event;
 6958	struct timespec64 ts;
 6959	unsigned long flags;
 6960
 6961	if (pin < 0 || pin >= IGB_N_SDP)
 6962		return;
 6963
 6964	if (hw->mac.type == e1000_82580 ||
 6965	    hw->mac.type == e1000_i354 ||
 6966	    hw->mac.type == e1000_i350) {
 6967		u64 ns = rd32(auxstmpl);
 6968
 6969		ns += ((u64)(rd32(auxstmph) & 0xFF)) << 32;
 6970		spin_lock_irqsave(&adapter->tmreg_lock, flags);
 6971		ns = timecounter_cyc2time(&adapter->tc, ns);
 6972		spin_unlock_irqrestore(&adapter->tmreg_lock, flags);
 6973		ts = ns_to_timespec64(ns);
 6974	} else {
 6975		ts.tv_nsec = rd32(auxstmpl);
 6976		ts.tv_sec  = rd32(auxstmph);
 6977	}
 6978
 6979	event.type = PTP_CLOCK_EXTTS;
 6980	event.index = tsintr_tt;
 6981	event.timestamp = ts.tv_sec * 1000000000ULL + ts.tv_nsec;
 6982	ptp_clock_event(adapter->ptp_clock, &event);
 6983}
 6984
 6985static void igb_tsync_interrupt(struct igb_adapter *adapter)
 6986{
 6987	struct e1000_hw *hw = &adapter->hw;
 6988	u32 tsicr = rd32(E1000_TSICR);
 6989	struct ptp_clock_event event;
 6990
 6991	if (tsicr & TSINTR_SYS_WRAP) {
 6992		event.type = PTP_CLOCK_PPS;
 6993		if (adapter->ptp_caps.pps)
 6994			ptp_clock_event(adapter->ptp_clock, &event);
 6995	}
 6996
 6997	if (tsicr & E1000_TSICR_TXTS) {
 6998		/* retrieve hardware timestamp */
 6999		schedule_work(&adapter->ptp_tx_work);
 7000	}
 7001
 7002	if (tsicr & TSINTR_TT0)
 7003		igb_perout(adapter, 0);
 7004
 7005	if (tsicr & TSINTR_TT1)
 7006		igb_perout(adapter, 1);
 7007
 7008	if (tsicr & TSINTR_AUTT0)
 7009		igb_extts(adapter, 0);
 7010
 7011	if (tsicr & TSINTR_AUTT1)
 7012		igb_extts(adapter, 1);
 7013}
 7014
 7015static irqreturn_t igb_msix_other(int irq, void *data)
 7016{
 7017	struct igb_adapter *adapter = data;
 7018	struct e1000_hw *hw = &adapter->hw;
 7019	u32 icr = rd32(E1000_ICR);
 7020	/* reading ICR causes bit 31 of EICR to be cleared */
 7021
 7022	if (icr & E1000_ICR_DRSTA)
 7023		schedule_work(&adapter->reset_task);
 7024
 7025	if (icr & E1000_ICR_DOUTSYNC) {
 7026		/* HW is reporting DMA is out of sync */
 7027		adapter->stats.doosync++;
 7028		/* The DMA Out of Sync is also indication of a spoof event
 7029		 * in IOV mode. Check the Wrong VM Behavior register to
 7030		 * see if it is really a spoof event.
 7031		 */
 7032		igb_check_wvbr(adapter);
 7033	}
 7034
 7035	/* Check for a mailbox event */
 7036	if (icr & E1000_ICR_VMMB)
 7037		igb_msg_task(adapter);
 7038
 7039	if (icr & E1000_ICR_LSC) {
 7040		hw->mac.get_link_status = 1;
 7041		/* guard against interrupt when we're going down */
 7042		if (!test_bit(__IGB_DOWN, &adapter->state))
 7043			mod_timer(&adapter->watchdog_timer, jiffies + 1);
 7044	}
 7045
 7046	if (icr & E1000_ICR_TS)
 7047		igb_tsync_interrupt(adapter);
 
 
 
 
 
 
 
 
 7048
 7049	wr32(E1000_EIMS, adapter->eims_other);
 7050
 7051	return IRQ_HANDLED;
 7052}
 7053
 7054static void igb_write_itr(struct igb_q_vector *q_vector)
 7055{
 7056	struct igb_adapter *adapter = q_vector->adapter;
 7057	u32 itr_val = q_vector->itr_val & 0x7FFC;
 7058
 7059	if (!q_vector->set_itr)
 7060		return;
 7061
 7062	if (!itr_val)
 7063		itr_val = 0x4;
 7064
 7065	if (adapter->hw.mac.type == e1000_82575)
 7066		itr_val |= itr_val << 16;
 7067	else
 7068		itr_val |= E1000_EITR_CNT_IGNR;
 7069
 7070	writel(itr_val, q_vector->itr_register);
 7071	q_vector->set_itr = 0;
 7072}
 7073
 7074static irqreturn_t igb_msix_ring(int irq, void *data)
 7075{
 7076	struct igb_q_vector *q_vector = data;
 7077
 7078	/* Write the ITR value calculated from the previous interrupt. */
 7079	igb_write_itr(q_vector);
 7080
 7081	napi_schedule(&q_vector->napi);
 7082
 7083	return IRQ_HANDLED;
 7084}
 7085
 7086#ifdef CONFIG_IGB_DCA
 7087static void igb_update_tx_dca(struct igb_adapter *adapter,
 7088			      struct igb_ring *tx_ring,
 7089			      int cpu)
 7090{
 7091	struct e1000_hw *hw = &adapter->hw;
 7092	u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
 7093
 7094	if (hw->mac.type != e1000_82575)
 7095		txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
 7096
 7097	/* We can enable relaxed ordering for reads, but not writes when
 7098	 * DCA is enabled.  This is due to a known issue in some chipsets
 7099	 * which will cause the DCA tag to be cleared.
 7100	 */
 7101	txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
 7102		  E1000_DCA_TXCTRL_DATA_RRO_EN |
 7103		  E1000_DCA_TXCTRL_DESC_DCA_EN;
 7104
 7105	wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
 7106}
 7107
 7108static void igb_update_rx_dca(struct igb_adapter *adapter,
 7109			      struct igb_ring *rx_ring,
 7110			      int cpu)
 7111{
 7112	struct e1000_hw *hw = &adapter->hw;
 7113	u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
 7114
 7115	if (hw->mac.type != e1000_82575)
 7116		rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
 7117
 7118	/* We can enable relaxed ordering for reads, but not writes when
 7119	 * DCA is enabled.  This is due to a known issue in some chipsets
 7120	 * which will cause the DCA tag to be cleared.
 7121	 */
 7122	rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
 7123		  E1000_DCA_RXCTRL_DESC_DCA_EN;
 7124
 7125	wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
 7126}
 7127
 7128static void igb_update_dca(struct igb_q_vector *q_vector)
 7129{
 7130	struct igb_adapter *adapter = q_vector->adapter;
 7131	int cpu = get_cpu();
 7132
 7133	if (q_vector->cpu == cpu)
 7134		goto out_no_update;
 7135
 7136	if (q_vector->tx.ring)
 7137		igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
 7138
 7139	if (q_vector->rx.ring)
 7140		igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
 7141
 7142	q_vector->cpu = cpu;
 7143out_no_update:
 7144	put_cpu();
 7145}
 7146
 7147static void igb_setup_dca(struct igb_adapter *adapter)
 7148{
 7149	struct e1000_hw *hw = &adapter->hw;
 7150	int i;
 7151
 7152	if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
 7153		return;
 7154
 7155	/* Always use CB2 mode, difference is masked in the CB driver. */
 7156	wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
 7157
 7158	for (i = 0; i < adapter->num_q_vectors; i++) {
 7159		adapter->q_vector[i]->cpu = -1;
 7160		igb_update_dca(adapter->q_vector[i]);
 7161	}
 7162}
 7163
 7164static int __igb_notify_dca(struct device *dev, void *data)
 7165{
 7166	struct net_device *netdev = dev_get_drvdata(dev);
 7167	struct igb_adapter *adapter = netdev_priv(netdev);
 7168	struct pci_dev *pdev = adapter->pdev;
 7169	struct e1000_hw *hw = &adapter->hw;
 7170	unsigned long event = *(unsigned long *)data;
 7171
 7172	switch (event) {
 7173	case DCA_PROVIDER_ADD:
 7174		/* if already enabled, don't do it again */
 7175		if (adapter->flags & IGB_FLAG_DCA_ENABLED)
 7176			break;
 7177		if (dca_add_requester(dev) == 0) {
 7178			adapter->flags |= IGB_FLAG_DCA_ENABLED;
 7179			dev_info(&pdev->dev, "DCA enabled\n");
 7180			igb_setup_dca(adapter);
 7181			break;
 7182		}
 7183		fallthrough; /* since DCA is disabled. */
 7184	case DCA_PROVIDER_REMOVE:
 7185		if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
 7186			/* without this a class_device is left
 7187			 * hanging around in the sysfs model
 7188			 */
 7189			dca_remove_requester(dev);
 7190			dev_info(&pdev->dev, "DCA disabled\n");
 7191			adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
 7192			wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
 7193		}
 7194		break;
 7195	}
 7196
 7197	return 0;
 7198}
 7199
 7200static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
 7201			  void *p)
 7202{
 7203	int ret_val;
 7204
 7205	ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
 7206					 __igb_notify_dca);
 7207
 7208	return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
 7209}
 7210#endif /* CONFIG_IGB_DCA */
 7211
 7212#ifdef CONFIG_PCI_IOV
 7213static int igb_vf_configure(struct igb_adapter *adapter, int vf)
 7214{
 7215	unsigned char mac_addr[ETH_ALEN];
 7216
 7217	eth_zero_addr(mac_addr);
 7218	igb_set_vf_mac(adapter, vf, mac_addr);
 7219
 7220	/* By default spoof check is enabled for all VFs */
 7221	adapter->vf_data[vf].spoofchk_enabled = true;
 7222
 7223	/* By default VFs are not trusted */
 7224	adapter->vf_data[vf].trusted = false;
 7225
 7226	return 0;
 7227}
 7228
 7229#endif
 7230static void igb_ping_all_vfs(struct igb_adapter *adapter)
 7231{
 7232	struct e1000_hw *hw = &adapter->hw;
 7233	u32 ping;
 7234	int i;
 7235
 7236	for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
 7237		ping = E1000_PF_CONTROL_MSG;
 7238		if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
 7239			ping |= E1000_VT_MSGTYPE_CTS;
 7240		igb_write_mbx(hw, &ping, 1, i);
 7241	}
 7242}
 7243
 7244static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
 7245{
 7246	struct e1000_hw *hw = &adapter->hw;
 7247	u32 vmolr = rd32(E1000_VMOLR(vf));
 7248	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
 7249
 7250	vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
 7251			    IGB_VF_FLAG_MULTI_PROMISC);
 7252	vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
 7253
 7254	if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
 7255		vmolr |= E1000_VMOLR_MPME;
 7256		vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
 7257		*msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
 7258	} else {
 7259		/* if we have hashes and we are clearing a multicast promisc
 7260		 * flag we need to write the hashes to the MTA as this step
 7261		 * was previously skipped
 7262		 */
 7263		if (vf_data->num_vf_mc_hashes > 30) {
 7264			vmolr |= E1000_VMOLR_MPME;
 7265		} else if (vf_data->num_vf_mc_hashes) {
 7266			int j;
 7267
 7268			vmolr |= E1000_VMOLR_ROMPE;
 7269			for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
 7270				igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
 7271		}
 7272	}
 7273
 7274	wr32(E1000_VMOLR(vf), vmolr);
 7275
 7276	/* there are flags left unprocessed, likely not supported */
 7277	if (*msgbuf & E1000_VT_MSGINFO_MASK)
 7278		return -EINVAL;
 7279
 7280	return 0;
 7281}
 7282
 7283static int igb_set_vf_multicasts(struct igb_adapter *adapter,
 7284				  u32 *msgbuf, u32 vf)
 7285{
 7286	int n = FIELD_GET(E1000_VT_MSGINFO_MASK, msgbuf[0]);
 7287	u16 *hash_list = (u16 *)&msgbuf[1];
 7288	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
 7289	int i;
 7290
 7291	/* salt away the number of multicast addresses assigned
 7292	 * to this VF for later use to restore when the PF multi cast
 7293	 * list changes
 7294	 */
 7295	vf_data->num_vf_mc_hashes = n;
 7296
 7297	/* only up to 30 hash values supported */
 7298	if (n > 30)
 7299		n = 30;
 7300
 7301	/* store the hashes for later use */
 7302	for (i = 0; i < n; i++)
 7303		vf_data->vf_mc_hashes[i] = hash_list[i];
 7304
 7305	/* Flush and reset the mta with the new values */
 7306	igb_set_rx_mode(adapter->netdev);
 7307
 7308	return 0;
 7309}
 7310
 7311static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
 7312{
 7313	struct e1000_hw *hw = &adapter->hw;
 7314	struct vf_data_storage *vf_data;
 7315	int i, j;
 7316
 7317	for (i = 0; i < adapter->vfs_allocated_count; i++) {
 7318		u32 vmolr = rd32(E1000_VMOLR(i));
 7319
 7320		vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
 7321
 7322		vf_data = &adapter->vf_data[i];
 7323
 7324		if ((vf_data->num_vf_mc_hashes > 30) ||
 7325		    (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
 7326			vmolr |= E1000_VMOLR_MPME;
 7327		} else if (vf_data->num_vf_mc_hashes) {
 7328			vmolr |= E1000_VMOLR_ROMPE;
 7329			for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
 7330				igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
 7331		}
 7332		wr32(E1000_VMOLR(i), vmolr);
 7333	}
 7334}
 7335
 7336static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
 7337{
 7338	struct e1000_hw *hw = &adapter->hw;
 7339	u32 pool_mask, vlvf_mask, i;
 
 7340
 7341	/* create mask for VF and other pools */
 7342	pool_mask = E1000_VLVF_POOLSEL_MASK;
 7343	vlvf_mask = BIT(E1000_VLVF_POOLSEL_SHIFT + vf);
 7344
 7345	/* drop PF from pool bits */
 7346	pool_mask &= ~BIT(E1000_VLVF_POOLSEL_SHIFT +
 7347			     adapter->vfs_allocated_count);
 7348
 7349	/* Find the vlan filter for this id */
 7350	for (i = E1000_VLVF_ARRAY_SIZE; i--;) {
 7351		u32 vlvf = rd32(E1000_VLVF(i));
 7352		u32 vfta_mask, vid, vfta;
 7353
 7354		/* remove the vf from the pool */
 7355		if (!(vlvf & vlvf_mask))
 7356			continue;
 7357
 7358		/* clear out bit from VLVF */
 7359		vlvf ^= vlvf_mask;
 7360
 7361		/* if other pools are present, just remove ourselves */
 7362		if (vlvf & pool_mask)
 7363			goto update_vlvfb;
 7364
 7365		/* if PF is present, leave VFTA */
 7366		if (vlvf & E1000_VLVF_POOLSEL_MASK)
 7367			goto update_vlvf;
 7368
 7369		vid = vlvf & E1000_VLVF_VLANID_MASK;
 7370		vfta_mask = BIT(vid % 32);
 7371
 7372		/* clear bit from VFTA */
 7373		vfta = adapter->shadow_vfta[vid / 32];
 7374		if (vfta & vfta_mask)
 7375			hw->mac.ops.write_vfta(hw, vid / 32, vfta ^ vfta_mask);
 7376update_vlvf:
 7377		/* clear pool selection enable */
 7378		if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
 7379			vlvf &= E1000_VLVF_POOLSEL_MASK;
 7380		else
 7381			vlvf = 0;
 7382update_vlvfb:
 7383		/* clear pool bits */
 7384		wr32(E1000_VLVF(i), vlvf);
 7385	}
 7386}
 7387
 7388static int igb_find_vlvf_entry(struct e1000_hw *hw, u32 vlan)
 7389{
 7390	u32 vlvf;
 7391	int idx;
 7392
 7393	/* short cut the special case */
 7394	if (vlan == 0)
 7395		return 0;
 
 
 
 
 7396
 7397	/* Search for the VLAN id in the VLVF entries */
 7398	for (idx = E1000_VLVF_ARRAY_SIZE; --idx;) {
 7399		vlvf = rd32(E1000_VLVF(idx));
 7400		if ((vlvf & VLAN_VID_MASK) == vlan)
 7401			break;
 7402	}
 7403
 7404	return idx;
 7405}
 7406
 7407static void igb_update_pf_vlvf(struct igb_adapter *adapter, u32 vid)
 7408{
 7409	struct e1000_hw *hw = &adapter->hw;
 7410	u32 bits, pf_id;
 7411	int idx;
 7412
 7413	idx = igb_find_vlvf_entry(hw, vid);
 7414	if (!idx)
 7415		return;
 7416
 7417	/* See if any other pools are set for this VLAN filter
 7418	 * entry other than the PF.
 7419	 */
 7420	pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
 7421	bits = ~BIT(pf_id) & E1000_VLVF_POOLSEL_MASK;
 7422	bits &= rd32(E1000_VLVF(idx));
 7423
 7424	/* Disable the filter so this falls into the default pool. */
 7425	if (!bits) {
 7426		if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
 7427			wr32(E1000_VLVF(idx), BIT(pf_id));
 7428		else
 7429			wr32(E1000_VLVF(idx), 0);
 7430	}
 7431}
 7432
 7433static s32 igb_set_vf_vlan(struct igb_adapter *adapter, u32 vid,
 7434			   bool add, u32 vf)
 7435{
 7436	int pf_id = adapter->vfs_allocated_count;
 7437	struct e1000_hw *hw = &adapter->hw;
 7438	int err;
 7439
 7440	/* If VLAN overlaps with one the PF is currently monitoring make
 7441	 * sure that we are able to allocate a VLVF entry.  This may be
 7442	 * redundant but it guarantees PF will maintain visibility to
 7443	 * the VLAN.
 7444	 */
 7445	if (add && test_bit(vid, adapter->active_vlans)) {
 7446		err = igb_vfta_set(hw, vid, pf_id, true, false);
 7447		if (err)
 7448			return err;
 7449	}
 7450
 7451	err = igb_vfta_set(hw, vid, vf, add, false);
 7452
 7453	if (add && !err)
 7454		return err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 7455
 7456	/* If we failed to add the VF VLAN or we are removing the VF VLAN
 7457	 * we may need to drop the PF pool bit in order to allow us to free
 7458	 * up the VLVF resources.
 7459	 */
 7460	if (test_bit(vid, adapter->active_vlans) ||
 7461	    (adapter->flags & IGB_FLAG_VLAN_PROMISC))
 7462		igb_update_pf_vlvf(adapter, vid);
 
 
 
 
 
 7463
 7464	return err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 7465}
 7466
 7467static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
 7468{
 7469	struct e1000_hw *hw = &adapter->hw;
 7470
 7471	if (vid)
 7472		wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
 7473	else
 7474		wr32(E1000_VMVIR(vf), 0);
 7475}
 7476
 7477static int igb_enable_port_vlan(struct igb_adapter *adapter, int vf,
 7478				u16 vlan, u8 qos)
 7479{
 7480	int err;
 7481
 7482	err = igb_set_vf_vlan(adapter, vlan, true, vf);
 7483	if (err)
 7484		return err;
 7485
 7486	igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
 7487	igb_set_vmolr(adapter, vf, !vlan);
 7488
 7489	/* revoke access to previous VLAN */
 7490	if (vlan != adapter->vf_data[vf].pf_vlan)
 7491		igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
 7492				false, vf);
 7493
 7494	adapter->vf_data[vf].pf_vlan = vlan;
 7495	adapter->vf_data[vf].pf_qos = qos;
 7496	igb_set_vf_vlan_strip(adapter, vf, true);
 7497	dev_info(&adapter->pdev->dev,
 7498		 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
 7499	if (test_bit(__IGB_DOWN, &adapter->state)) {
 7500		dev_warn(&adapter->pdev->dev,
 7501			 "The VF VLAN has been set, but the PF device is not up.\n");
 7502		dev_warn(&adapter->pdev->dev,
 7503			 "Bring the PF device up before attempting to use the VF device.\n");
 
 
 
 
 
 
 
 
 
 
 7504	}
 7505
 7506	return err;
 7507}
 7508
 7509static int igb_disable_port_vlan(struct igb_adapter *adapter, int vf)
 7510{
 7511	/* Restore tagless access via VLAN 0 */
 7512	igb_set_vf_vlan(adapter, 0, true, vf);
 7513
 7514	igb_set_vmvir(adapter, 0, vf);
 7515	igb_set_vmolr(adapter, vf, true);
 7516
 7517	/* Remove any PF assigned VLAN */
 7518	if (adapter->vf_data[vf].pf_vlan)
 7519		igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
 7520				false, vf);
 
 
 
 7521
 7522	adapter->vf_data[vf].pf_vlan = 0;
 7523	adapter->vf_data[vf].pf_qos = 0;
 7524	igb_set_vf_vlan_strip(adapter, vf, false);
 7525
 7526	return 0;
 7527}
 7528
 7529static int igb_ndo_set_vf_vlan(struct net_device *netdev, int vf,
 7530			       u16 vlan, u8 qos, __be16 vlan_proto)
 7531{
 7532	struct igb_adapter *adapter = netdev_priv(netdev);
 7533
 7534	if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
 7535		return -EINVAL;
 7536
 7537	if (vlan_proto != htons(ETH_P_8021Q))
 7538		return -EPROTONOSUPPORT;
 
 
 
 
 
 
 7539
 7540	return (vlan || qos) ? igb_enable_port_vlan(adapter, vf, vlan, qos) :
 7541			       igb_disable_port_vlan(adapter, vf);
 7542}
 7543
 7544static int igb_set_vf_vlan_msg(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
 7545{
 7546	int add = FIELD_GET(E1000_VT_MSGINFO_MASK, msgbuf[0]);
 7547	int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
 7548	int ret;
 7549
 7550	if (adapter->vf_data[vf].pf_vlan)
 7551		return -1;
 
 
 
 7552
 7553	/* VLAN 0 is a special case, don't allow it to be removed */
 7554	if (!vid && !add)
 7555		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 7556
 7557	ret = igb_set_vf_vlan(adapter, vid, !!add, vf);
 7558	if (!ret)
 7559		igb_set_vf_vlan_strip(adapter, vf, !!vid);
 7560	return ret;
 7561}
 7562
 7563static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
 7564{
 7565	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
 7566
 7567	/* clear flags - except flag that indicates PF has set the MAC */
 7568	vf_data->flags &= IGB_VF_FLAG_PF_SET_MAC;
 7569	vf_data->last_nack = jiffies;
 
 
 
 7570
 7571	/* reset vlans for device */
 7572	igb_clear_vf_vfta(adapter, vf);
 7573	igb_set_vf_vlan(adapter, vf_data->pf_vlan, true, vf);
 7574	igb_set_vmvir(adapter, vf_data->pf_vlan |
 7575			       (vf_data->pf_qos << VLAN_PRIO_SHIFT), vf);
 7576	igb_set_vmolr(adapter, vf, !vf_data->pf_vlan);
 7577	igb_set_vf_vlan_strip(adapter, vf, !!(vf_data->pf_vlan));
 
 7578
 7579	/* reset multicast table array for vf */
 7580	adapter->vf_data[vf].num_vf_mc_hashes = 0;
 7581
 7582	/* Flush and reset the mta with the new values */
 7583	igb_set_rx_mode(adapter->netdev);
 7584}
 7585
 7586static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
 7587{
 7588	unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
 7589
 7590	/* clear mac address as we were hotplug removed/added */
 7591	if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
 7592		eth_zero_addr(vf_mac);
 7593
 7594	/* process remaining reset events */
 7595	igb_vf_reset(adapter, vf);
 7596}
 7597
 7598static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
 7599{
 7600	struct e1000_hw *hw = &adapter->hw;
 7601	unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
 7602	u32 reg, msgbuf[3] = {};
 
 7603	u8 *addr = (u8 *)(&msgbuf[1]);
 7604
 7605	/* process all the same items cleared in a function level reset */
 7606	igb_vf_reset(adapter, vf);
 7607
 7608	/* set vf mac address */
 7609	igb_set_vf_mac(adapter, vf, vf_mac);
 7610
 7611	/* enable transmit and receive for vf */
 7612	reg = rd32(E1000_VFTE);
 7613	wr32(E1000_VFTE, reg | BIT(vf));
 7614	reg = rd32(E1000_VFRE);
 7615	wr32(E1000_VFRE, reg | BIT(vf));
 7616
 7617	adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
 7618
 7619	/* reply to reset with ack and vf mac address */
 7620	if (!is_zero_ether_addr(vf_mac)) {
 7621		msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
 7622		memcpy(addr, vf_mac, ETH_ALEN);
 7623	} else {
 7624		msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_NACK;
 7625	}
 7626	igb_write_mbx(hw, msgbuf, 3, vf);
 7627}
 7628
 7629static void igb_flush_mac_table(struct igb_adapter *adapter)
 7630{
 7631	struct e1000_hw *hw = &adapter->hw;
 7632	int i;
 7633
 7634	for (i = 0; i < hw->mac.rar_entry_count; i++) {
 7635		adapter->mac_table[i].state &= ~IGB_MAC_STATE_IN_USE;
 7636		eth_zero_addr(adapter->mac_table[i].addr);
 7637		adapter->mac_table[i].queue = 0;
 7638		igb_rar_set_index(adapter, i);
 7639	}
 7640}
 7641
 7642static int igb_available_rars(struct igb_adapter *adapter, u8 queue)
 7643{
 7644	struct e1000_hw *hw = &adapter->hw;
 7645	/* do not count rar entries reserved for VFs MAC addresses */
 7646	int rar_entries = hw->mac.rar_entry_count -
 7647			  adapter->vfs_allocated_count;
 7648	int i, count = 0;
 7649
 7650	for (i = 0; i < rar_entries; i++) {
 7651		/* do not count default entries */
 7652		if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT)
 7653			continue;
 7654
 7655		/* do not count "in use" entries for different queues */
 7656		if ((adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE) &&
 7657		    (adapter->mac_table[i].queue != queue))
 7658			continue;
 7659
 7660		count++;
 7661	}
 7662
 7663	return count;
 7664}
 7665
 7666/* Set default MAC address for the PF in the first RAR entry */
 7667static void igb_set_default_mac_filter(struct igb_adapter *adapter)
 7668{
 7669	struct igb_mac_addr *mac_table = &adapter->mac_table[0];
 7670
 7671	ether_addr_copy(mac_table->addr, adapter->hw.mac.addr);
 7672	mac_table->queue = adapter->vfs_allocated_count;
 7673	mac_table->state = IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE;
 7674
 7675	igb_rar_set_index(adapter, 0);
 7676}
 7677
 7678/* If the filter to be added and an already existing filter express
 7679 * the same address and address type, it should be possible to only
 7680 * override the other configurations, for example the queue to steer
 7681 * traffic.
 7682 */
 7683static bool igb_mac_entry_can_be_used(const struct igb_mac_addr *entry,
 7684				      const u8 *addr, const u8 flags)
 7685{
 7686	if (!(entry->state & IGB_MAC_STATE_IN_USE))
 7687		return true;
 7688
 7689	if ((entry->state & IGB_MAC_STATE_SRC_ADDR) !=
 7690	    (flags & IGB_MAC_STATE_SRC_ADDR))
 7691		return false;
 7692
 7693	if (!ether_addr_equal(addr, entry->addr))
 7694		return false;
 7695
 7696	return true;
 7697}
 7698
 7699/* Add a MAC filter for 'addr' directing matching traffic to 'queue',
 7700 * 'flags' is used to indicate what kind of match is made, match is by
 7701 * default for the destination address, if matching by source address
 7702 * is desired the flag IGB_MAC_STATE_SRC_ADDR can be used.
 7703 */
 7704static int igb_add_mac_filter_flags(struct igb_adapter *adapter,
 7705				    const u8 *addr, const u8 queue,
 7706				    const u8 flags)
 7707{
 7708	struct e1000_hw *hw = &adapter->hw;
 7709	int rar_entries = hw->mac.rar_entry_count -
 7710			  adapter->vfs_allocated_count;
 7711	int i;
 7712
 7713	if (is_zero_ether_addr(addr))
 7714		return -EINVAL;
 7715
 7716	/* Search for the first empty entry in the MAC table.
 7717	 * Do not touch entries at the end of the table reserved for the VF MAC
 7718	 * addresses.
 7719	 */
 7720	for (i = 0; i < rar_entries; i++) {
 7721		if (!igb_mac_entry_can_be_used(&adapter->mac_table[i],
 7722					       addr, flags))
 7723			continue;
 7724
 7725		ether_addr_copy(adapter->mac_table[i].addr, addr);
 7726		adapter->mac_table[i].queue = queue;
 7727		adapter->mac_table[i].state |= IGB_MAC_STATE_IN_USE | flags;
 7728
 7729		igb_rar_set_index(adapter, i);
 7730		return i;
 7731	}
 7732
 7733	return -ENOSPC;
 7734}
 7735
 7736static int igb_add_mac_filter(struct igb_adapter *adapter, const u8 *addr,
 7737			      const u8 queue)
 7738{
 7739	return igb_add_mac_filter_flags(adapter, addr, queue, 0);
 7740}
 7741
 7742/* Remove a MAC filter for 'addr' directing matching traffic to
 7743 * 'queue', 'flags' is used to indicate what kind of match need to be
 7744 * removed, match is by default for the destination address, if
 7745 * matching by source address is to be removed the flag
 7746 * IGB_MAC_STATE_SRC_ADDR can be used.
 7747 */
 7748static int igb_del_mac_filter_flags(struct igb_adapter *adapter,
 7749				    const u8 *addr, const u8 queue,
 7750				    const u8 flags)
 7751{
 7752	struct e1000_hw *hw = &adapter->hw;
 7753	int rar_entries = hw->mac.rar_entry_count -
 7754			  adapter->vfs_allocated_count;
 7755	int i;
 7756
 7757	if (is_zero_ether_addr(addr))
 7758		return -EINVAL;
 7759
 7760	/* Search for matching entry in the MAC table based on given address
 7761	 * and queue. Do not touch entries at the end of the table reserved
 7762	 * for the VF MAC addresses.
 7763	 */
 7764	for (i = 0; i < rar_entries; i++) {
 7765		if (!(adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE))
 7766			continue;
 7767		if ((adapter->mac_table[i].state & flags) != flags)
 7768			continue;
 7769		if (adapter->mac_table[i].queue != queue)
 7770			continue;
 7771		if (!ether_addr_equal(adapter->mac_table[i].addr, addr))
 7772			continue;
 7773
 7774		/* When a filter for the default address is "deleted",
 7775		 * we return it to its initial configuration
 7776		 */
 7777		if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT) {
 7778			adapter->mac_table[i].state =
 7779				IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE;
 7780			adapter->mac_table[i].queue =
 7781				adapter->vfs_allocated_count;
 7782		} else {
 7783			adapter->mac_table[i].state = 0;
 7784			adapter->mac_table[i].queue = 0;
 7785			eth_zero_addr(adapter->mac_table[i].addr);
 7786		}
 7787
 7788		igb_rar_set_index(adapter, i);
 7789		return 0;
 7790	}
 7791
 7792	return -ENOENT;
 7793}
 7794
 7795static int igb_del_mac_filter(struct igb_adapter *adapter, const u8 *addr,
 7796			      const u8 queue)
 7797{
 7798	return igb_del_mac_filter_flags(adapter, addr, queue, 0);
 7799}
 7800
 7801int igb_add_mac_steering_filter(struct igb_adapter *adapter,
 7802				const u8 *addr, u8 queue, u8 flags)
 7803{
 7804	struct e1000_hw *hw = &adapter->hw;
 7805
 7806	/* In theory, this should be supported on 82575 as well, but
 7807	 * that part wasn't easily accessible during development.
 7808	 */
 7809	if (hw->mac.type != e1000_i210)
 7810		return -EOPNOTSUPP;
 7811
 7812	return igb_add_mac_filter_flags(adapter, addr, queue,
 7813					IGB_MAC_STATE_QUEUE_STEERING | flags);
 7814}
 7815
 7816int igb_del_mac_steering_filter(struct igb_adapter *adapter,
 7817				const u8 *addr, u8 queue, u8 flags)
 7818{
 7819	return igb_del_mac_filter_flags(adapter, addr, queue,
 7820					IGB_MAC_STATE_QUEUE_STEERING | flags);
 7821}
 7822
 7823static int igb_uc_sync(struct net_device *netdev, const unsigned char *addr)
 7824{
 7825	struct igb_adapter *adapter = netdev_priv(netdev);
 7826	int ret;
 7827
 7828	ret = igb_add_mac_filter(adapter, addr, adapter->vfs_allocated_count);
 7829
 7830	return min_t(int, ret, 0);
 7831}
 7832
 7833static int igb_uc_unsync(struct net_device *netdev, const unsigned char *addr)
 7834{
 7835	struct igb_adapter *adapter = netdev_priv(netdev);
 7836
 7837	igb_del_mac_filter(adapter, addr, adapter->vfs_allocated_count);
 7838
 7839	return 0;
 7840}
 7841
 7842static int igb_set_vf_mac_filter(struct igb_adapter *adapter, const int vf,
 7843				 const u32 info, const u8 *addr)
 7844{
 7845	struct pci_dev *pdev = adapter->pdev;
 7846	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
 7847	struct vf_mac_filter *entry;
 7848	bool found = false;
 7849	int ret = 0;
 7850
 7851	if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) &&
 7852	    !vf_data->trusted) {
 7853		dev_warn(&pdev->dev,
 7854			 "VF %d requested MAC filter but is administratively denied\n",
 7855			  vf);
 7856		return -EINVAL;
 7857	}
 7858	if (!is_valid_ether_addr(addr)) {
 7859		dev_warn(&pdev->dev,
 7860			 "VF %d attempted to set invalid MAC filter\n",
 7861			  vf);
 7862		return -EINVAL;
 7863	}
 7864
 7865	switch (info) {
 7866	case E1000_VF_MAC_FILTER_CLR:
 7867		/* remove all unicast MAC filters related to the current VF */
 7868		list_for_each_entry(entry, &adapter->vf_macs.l, l) {
 7869			if (entry->vf == vf) {
 7870				entry->vf = -1;
 7871				entry->free = true;
 7872				igb_del_mac_filter(adapter, entry->vf_mac, vf);
 7873			}
 7874		}
 7875		break;
 7876	case E1000_VF_MAC_FILTER_ADD:
 7877		/* try to find empty slot in the list */
 7878		list_for_each_entry(entry, &adapter->vf_macs.l, l) {
 7879			if (entry->free) {
 7880				found = true;
 7881				break;
 7882			}
 7883		}
 7884
 7885		if (found) {
 7886			entry->free = false;
 7887			entry->vf = vf;
 7888			ether_addr_copy(entry->vf_mac, addr);
 7889
 7890			ret = igb_add_mac_filter(adapter, addr, vf);
 7891			ret = min_t(int, ret, 0);
 7892		} else {
 7893			ret = -ENOSPC;
 7894		}
 7895
 7896		if (ret == -ENOSPC)
 7897			dev_warn(&pdev->dev,
 7898				 "VF %d has requested MAC filter but there is no space for it\n",
 7899				 vf);
 7900		break;
 7901	default:
 7902		ret = -EINVAL;
 7903		break;
 7904	}
 7905
 7906	return ret;
 7907}
 7908
 7909static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
 7910{
 7911	struct pci_dev *pdev = adapter->pdev;
 7912	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
 7913	u32 info = msg[0] & E1000_VT_MSGINFO_MASK;
 7914
 7915	/* The VF MAC Address is stored in a packed array of bytes
 7916	 * starting at the second 32 bit word of the msg array
 7917	 */
 7918	unsigned char *addr = (unsigned char *)&msg[1];
 7919	int ret = 0;
 7920
 7921	if (!info) {
 7922		if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) &&
 7923		    !vf_data->trusted) {
 7924			dev_warn(&pdev->dev,
 7925				 "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
 7926				 vf);
 7927			return -EINVAL;
 7928		}
 7929
 7930		if (!is_valid_ether_addr(addr)) {
 7931			dev_warn(&pdev->dev,
 7932				 "VF %d attempted to set invalid MAC\n",
 7933				 vf);
 7934			return -EINVAL;
 7935		}
 7936
 7937		ret = igb_set_vf_mac(adapter, vf, addr);
 7938	} else {
 7939		ret = igb_set_vf_mac_filter(adapter, vf, info, addr);
 7940	}
 7941
 7942	return ret;
 7943}
 7944
 7945static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
 7946{
 7947	struct e1000_hw *hw = &adapter->hw;
 7948	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
 7949	u32 msg = E1000_VT_MSGTYPE_NACK;
 7950
 7951	/* if device isn't clear to send it shouldn't be reading either */
 7952	if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
 7953	    time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
 7954		igb_write_mbx(hw, &msg, 1, vf);
 7955		vf_data->last_nack = jiffies;
 7956	}
 7957}
 7958
 7959static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
 7960{
 7961	struct pci_dev *pdev = adapter->pdev;
 7962	u32 msgbuf[E1000_VFMAILBOX_SIZE];
 7963	struct e1000_hw *hw = &adapter->hw;
 7964	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
 7965	s32 retval;
 7966
 7967	retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf, false);
 7968
 7969	if (retval) {
 7970		/* if receive failed revoke VF CTS stats and restart init */
 7971		dev_err(&pdev->dev, "Error receiving message from VF\n");
 7972		vf_data->flags &= ~IGB_VF_FLAG_CTS;
 7973		if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
 7974			goto unlock;
 7975		goto out;
 7976	}
 7977
 7978	/* this is a message we already processed, do nothing */
 7979	if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
 7980		goto unlock;
 7981
 7982	/* until the vf completes a reset it should not be
 7983	 * allowed to start any configuration.
 7984	 */
 7985	if (msgbuf[0] == E1000_VF_RESET) {
 7986		/* unlocks mailbox */
 7987		igb_vf_reset_msg(adapter, vf);
 7988		return;
 7989	}
 7990
 7991	if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
 7992		if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
 7993			goto unlock;
 7994		retval = -1;
 7995		goto out;
 7996	}
 7997
 7998	switch ((msgbuf[0] & 0xFFFF)) {
 7999	case E1000_VF_SET_MAC_ADDR:
 8000		retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
 
 
 
 
 
 
 8001		break;
 8002	case E1000_VF_SET_PROMISC:
 8003		retval = igb_set_vf_promisc(adapter, msgbuf, vf);
 8004		break;
 8005	case E1000_VF_SET_MULTICAST:
 8006		retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
 8007		break;
 8008	case E1000_VF_SET_LPE:
 8009		retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
 8010		break;
 8011	case E1000_VF_SET_VLAN:
 8012		retval = -1;
 8013		if (vf_data->pf_vlan)
 8014			dev_warn(&pdev->dev,
 8015				 "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
 8016				 vf);
 8017		else
 8018			retval = igb_set_vf_vlan_msg(adapter, msgbuf, vf);
 8019		break;
 8020	default:
 8021		dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
 8022		retval = -1;
 8023		break;
 8024	}
 8025
 8026	msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
 8027out:
 8028	/* notify the VF of the results of what it sent us */
 8029	if (retval)
 8030		msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
 8031	else
 8032		msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
 8033
 8034	/* unlocks mailbox */
 8035	igb_write_mbx(hw, msgbuf, 1, vf);
 8036	return;
 8037
 8038unlock:
 8039	igb_unlock_mbx(hw, vf);
 8040}
 8041
 8042static void igb_msg_task(struct igb_adapter *adapter)
 8043{
 8044	struct e1000_hw *hw = &adapter->hw;
 8045	unsigned long flags;
 8046	u32 vf;
 8047
 8048	spin_lock_irqsave(&adapter->vfs_lock, flags);
 8049	for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
 8050		/* process any reset requests */
 8051		if (!igb_check_for_rst(hw, vf))
 8052			igb_vf_reset_event(adapter, vf);
 8053
 8054		/* process any messages pending */
 8055		if (!igb_check_for_msg(hw, vf))
 8056			igb_rcv_msg_from_vf(adapter, vf);
 8057
 8058		/* process any acks */
 8059		if (!igb_check_for_ack(hw, vf))
 8060			igb_rcv_ack_from_vf(adapter, vf);
 8061	}
 8062	spin_unlock_irqrestore(&adapter->vfs_lock, flags);
 8063}
 8064
 8065/**
 8066 *  igb_set_uta - Set unicast filter table address
 8067 *  @adapter: board private structure
 8068 *  @set: boolean indicating if we are setting or clearing bits
 8069 *
 8070 *  The unicast table address is a register array of 32-bit registers.
 8071 *  The table is meant to be used in a way similar to how the MTA is used
 8072 *  however due to certain limitations in the hardware it is necessary to
 8073 *  set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
 8074 *  enable bit to allow vlan tag stripping when promiscuous mode is enabled
 8075 **/
 8076static void igb_set_uta(struct igb_adapter *adapter, bool set)
 8077{
 8078	struct e1000_hw *hw = &adapter->hw;
 8079	u32 uta = set ? ~0 : 0;
 8080	int i;
 8081
 
 
 
 
 8082	/* we only need to do this if VMDq is enabled */
 8083	if (!adapter->vfs_allocated_count)
 8084		return;
 8085
 8086	for (i = hw->mac.uta_reg_count; i--;)
 8087		array_wr32(E1000_UTA, i, uta);
 8088}
 8089
 8090/**
 8091 *  igb_intr_msi - Interrupt Handler
 8092 *  @irq: interrupt number
 8093 *  @data: pointer to a network interface device structure
 8094 **/
 8095static irqreturn_t igb_intr_msi(int irq, void *data)
 8096{
 8097	struct igb_adapter *adapter = data;
 8098	struct igb_q_vector *q_vector = adapter->q_vector[0];
 8099	struct e1000_hw *hw = &adapter->hw;
 8100	/* read ICR disables interrupts using IAM */
 8101	u32 icr = rd32(E1000_ICR);
 8102
 8103	igb_write_itr(q_vector);
 8104
 8105	if (icr & E1000_ICR_DRSTA)
 8106		schedule_work(&adapter->reset_task);
 8107
 8108	if (icr & E1000_ICR_DOUTSYNC) {
 8109		/* HW is reporting DMA is out of sync */
 8110		adapter->stats.doosync++;
 8111	}
 8112
 8113	if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
 8114		hw->mac.get_link_status = 1;
 8115		if (!test_bit(__IGB_DOWN, &adapter->state))
 8116			mod_timer(&adapter->watchdog_timer, jiffies + 1);
 8117	}
 8118
 8119	if (icr & E1000_ICR_TS)
 8120		igb_tsync_interrupt(adapter);
 
 
 
 
 
 
 
 
 8121
 8122	napi_schedule(&q_vector->napi);
 8123
 8124	return IRQ_HANDLED;
 8125}
 8126
 8127/**
 8128 *  igb_intr - Legacy Interrupt Handler
 8129 *  @irq: interrupt number
 8130 *  @data: pointer to a network interface device structure
 8131 **/
 8132static irqreturn_t igb_intr(int irq, void *data)
 8133{
 8134	struct igb_adapter *adapter = data;
 8135	struct igb_q_vector *q_vector = adapter->q_vector[0];
 8136	struct e1000_hw *hw = &adapter->hw;
 8137	/* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
 8138	 * need for the IMC write
 8139	 */
 8140	u32 icr = rd32(E1000_ICR);
 8141
 8142	/* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
 8143	 * not set, then the adapter didn't send an interrupt
 8144	 */
 8145	if (!(icr & E1000_ICR_INT_ASSERTED))
 8146		return IRQ_NONE;
 8147
 8148	igb_write_itr(q_vector);
 8149
 8150	if (icr & E1000_ICR_DRSTA)
 8151		schedule_work(&adapter->reset_task);
 8152
 8153	if (icr & E1000_ICR_DOUTSYNC) {
 8154		/* HW is reporting DMA is out of sync */
 8155		adapter->stats.doosync++;
 8156	}
 8157
 8158	if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
 8159		hw->mac.get_link_status = 1;
 8160		/* guard against interrupt when we're going down */
 8161		if (!test_bit(__IGB_DOWN, &adapter->state))
 8162			mod_timer(&adapter->watchdog_timer, jiffies + 1);
 8163	}
 8164
 8165	if (icr & E1000_ICR_TS)
 8166		igb_tsync_interrupt(adapter);
 
 
 
 
 
 
 
 
 8167
 8168	napi_schedule(&q_vector->napi);
 8169
 8170	return IRQ_HANDLED;
 8171}
 8172
 8173static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
 8174{
 8175	struct igb_adapter *adapter = q_vector->adapter;
 8176	struct e1000_hw *hw = &adapter->hw;
 8177
 8178	if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
 8179	    (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
 8180		if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
 8181			igb_set_itr(q_vector);
 8182		else
 8183			igb_update_ring_itr(q_vector);
 8184	}
 8185
 8186	if (!test_bit(__IGB_DOWN, &adapter->state)) {
 8187		if (adapter->flags & IGB_FLAG_HAS_MSIX)
 8188			wr32(E1000_EIMS, q_vector->eims_value);
 8189		else
 8190			igb_irq_enable(adapter);
 8191	}
 8192}
 8193
 8194/**
 8195 *  igb_poll - NAPI Rx polling callback
 8196 *  @napi: napi polling structure
 8197 *  @budget: count of how many packets we should handle
 8198 **/
 8199static int igb_poll(struct napi_struct *napi, int budget)
 8200{
 8201	struct igb_q_vector *q_vector = container_of(napi,
 8202						     struct igb_q_vector,
 8203						     napi);
 8204	bool clean_complete = true;
 8205	int work_done = 0;
 8206
 8207#ifdef CONFIG_IGB_DCA
 8208	if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
 8209		igb_update_dca(q_vector);
 8210#endif
 8211	if (q_vector->tx.ring)
 8212		clean_complete = igb_clean_tx_irq(q_vector, budget);
 8213
 8214	if (q_vector->rx.ring) {
 8215		int cleaned = igb_clean_rx_irq(q_vector, budget);
 8216
 8217		work_done += cleaned;
 8218		if (cleaned >= budget)
 8219			clean_complete = false;
 8220	}
 8221
 8222	/* If all work not completed, return budget and keep polling */
 8223	if (!clean_complete)
 8224		return budget;
 8225
 8226	/* Exit the polling mode, but don't re-enable interrupts if stack might
 8227	 * poll us due to busy-polling
 8228	 */
 8229	if (likely(napi_complete_done(napi, work_done)))
 8230		igb_ring_irq_enable(q_vector);
 8231
 8232	return work_done;
 8233}
 8234
 8235/**
 8236 *  igb_clean_tx_irq - Reclaim resources after transmit completes
 8237 *  @q_vector: pointer to q_vector containing needed info
 8238 *  @napi_budget: Used to determine if we are in netpoll
 8239 *
 8240 *  returns true if ring is completely cleaned
 8241 **/
 8242static bool igb_clean_tx_irq(struct igb_q_vector *q_vector, int napi_budget)
 8243{
 8244	struct igb_adapter *adapter = q_vector->adapter;
 8245	struct igb_ring *tx_ring = q_vector->tx.ring;
 8246	struct igb_tx_buffer *tx_buffer;
 8247	union e1000_adv_tx_desc *tx_desc;
 8248	unsigned int total_bytes = 0, total_packets = 0;
 8249	unsigned int budget = q_vector->tx.work_limit;
 8250	unsigned int i = tx_ring->next_to_clean;
 8251
 8252	if (test_bit(__IGB_DOWN, &adapter->state))
 8253		return true;
 8254
 8255	tx_buffer = &tx_ring->tx_buffer_info[i];
 8256	tx_desc = IGB_TX_DESC(tx_ring, i);
 8257	i -= tx_ring->count;
 8258
 8259	do {
 8260		union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
 8261
 8262		/* if next_to_watch is not set then there is no work pending */
 8263		if (!eop_desc)
 8264			break;
 8265
 8266		/* prevent any other reads prior to eop_desc */
 8267		smp_rmb();
 8268
 8269		/* if DD is not set pending work has not been completed */
 8270		if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
 8271			break;
 8272
 8273		/* clear next_to_watch to prevent false hangs */
 8274		tx_buffer->next_to_watch = NULL;
 8275
 8276		/* update the statistics for this packet */
 8277		total_bytes += tx_buffer->bytecount;
 8278		total_packets += tx_buffer->gso_segs;
 8279
 8280		/* free the skb */
 8281		if (tx_buffer->type == IGB_TYPE_SKB)
 8282			napi_consume_skb(tx_buffer->skb, napi_budget);
 8283		else
 8284			xdp_return_frame(tx_buffer->xdpf);
 8285
 8286		/* unmap skb header data */
 8287		dma_unmap_single(tx_ring->dev,
 8288				 dma_unmap_addr(tx_buffer, dma),
 8289				 dma_unmap_len(tx_buffer, len),
 8290				 DMA_TO_DEVICE);
 8291
 8292		/* clear tx_buffer data */
 
 8293		dma_unmap_len_set(tx_buffer, len, 0);
 8294
 8295		/* clear last DMA location and unmap remaining buffers */
 8296		while (tx_desc != eop_desc) {
 8297			tx_buffer++;
 8298			tx_desc++;
 8299			i++;
 8300			if (unlikely(!i)) {
 8301				i -= tx_ring->count;
 8302				tx_buffer = tx_ring->tx_buffer_info;
 8303				tx_desc = IGB_TX_DESC(tx_ring, 0);
 8304			}
 8305
 8306			/* unmap any remaining paged data */
 8307			if (dma_unmap_len(tx_buffer, len)) {
 8308				dma_unmap_page(tx_ring->dev,
 8309					       dma_unmap_addr(tx_buffer, dma),
 8310					       dma_unmap_len(tx_buffer, len),
 8311					       DMA_TO_DEVICE);
 8312				dma_unmap_len_set(tx_buffer, len, 0);
 8313			}
 8314		}
 8315
 8316		/* move us one more past the eop_desc for start of next pkt */
 8317		tx_buffer++;
 8318		tx_desc++;
 8319		i++;
 8320		if (unlikely(!i)) {
 8321			i -= tx_ring->count;
 8322			tx_buffer = tx_ring->tx_buffer_info;
 8323			tx_desc = IGB_TX_DESC(tx_ring, 0);
 8324		}
 8325
 8326		/* issue prefetch for next Tx descriptor */
 8327		prefetch(tx_desc);
 8328
 8329		/* update budget accounting */
 8330		budget--;
 8331	} while (likely(budget));
 8332
 8333	netdev_tx_completed_queue(txring_txq(tx_ring),
 8334				  total_packets, total_bytes);
 8335	i += tx_ring->count;
 8336	tx_ring->next_to_clean = i;
 8337	u64_stats_update_begin(&tx_ring->tx_syncp);
 8338	tx_ring->tx_stats.bytes += total_bytes;
 8339	tx_ring->tx_stats.packets += total_packets;
 8340	u64_stats_update_end(&tx_ring->tx_syncp);
 8341	q_vector->tx.total_bytes += total_bytes;
 8342	q_vector->tx.total_packets += total_packets;
 8343
 8344	if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
 8345		struct e1000_hw *hw = &adapter->hw;
 8346
 8347		/* Detect a transmit hang in hardware, this serializes the
 8348		 * check with the clearing of time_stamp and movement of i
 8349		 */
 8350		clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
 8351		if (tx_buffer->next_to_watch &&
 8352		    time_after(jiffies, tx_buffer->time_stamp +
 8353			       (adapter->tx_timeout_factor * HZ)) &&
 8354		    !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
 8355
 8356			/* detected Tx unit hang */
 8357			dev_err(tx_ring->dev,
 8358				"Detected Tx Unit Hang\n"
 8359				"  Tx Queue             <%d>\n"
 8360				"  TDH                  <%x>\n"
 8361				"  TDT                  <%x>\n"
 8362				"  next_to_use          <%x>\n"
 8363				"  next_to_clean        <%x>\n"
 8364				"buffer_info[next_to_clean]\n"
 8365				"  time_stamp           <%lx>\n"
 8366				"  next_to_watch        <%p>\n"
 8367				"  jiffies              <%lx>\n"
 8368				"  desc.status          <%x>\n",
 8369				tx_ring->queue_index,
 8370				rd32(E1000_TDH(tx_ring->reg_idx)),
 8371				readl(tx_ring->tail),
 8372				tx_ring->next_to_use,
 8373				tx_ring->next_to_clean,
 8374				tx_buffer->time_stamp,
 8375				tx_buffer->next_to_watch,
 8376				jiffies,
 8377				tx_buffer->next_to_watch->wb.status);
 8378			netif_stop_subqueue(tx_ring->netdev,
 8379					    tx_ring->queue_index);
 8380
 8381			/* we are about to reset, no point in enabling stuff */
 8382			return true;
 8383		}
 8384	}
 8385
 8386#define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
 8387	if (unlikely(total_packets &&
 8388	    netif_carrier_ok(tx_ring->netdev) &&
 8389	    igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
 8390		/* Make sure that anybody stopping the queue after this
 8391		 * sees the new next_to_clean.
 8392		 */
 8393		smp_mb();
 8394		if (__netif_subqueue_stopped(tx_ring->netdev,
 8395					     tx_ring->queue_index) &&
 8396		    !(test_bit(__IGB_DOWN, &adapter->state))) {
 8397			netif_wake_subqueue(tx_ring->netdev,
 8398					    tx_ring->queue_index);
 8399
 8400			u64_stats_update_begin(&tx_ring->tx_syncp);
 8401			tx_ring->tx_stats.restart_queue++;
 8402			u64_stats_update_end(&tx_ring->tx_syncp);
 8403		}
 8404	}
 8405
 8406	return !!budget;
 8407}
 8408
 8409/**
 8410 *  igb_reuse_rx_page - page flip buffer and store it back on the ring
 8411 *  @rx_ring: rx descriptor ring to store buffers on
 8412 *  @old_buff: donor buffer to have page reused
 8413 *
 8414 *  Synchronizes page for reuse by the adapter
 8415 **/
 8416static void igb_reuse_rx_page(struct igb_ring *rx_ring,
 8417			      struct igb_rx_buffer *old_buff)
 8418{
 8419	struct igb_rx_buffer *new_buff;
 8420	u16 nta = rx_ring->next_to_alloc;
 8421
 8422	new_buff = &rx_ring->rx_buffer_info[nta];
 8423
 8424	/* update, and store next to alloc */
 8425	nta++;
 8426	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
 8427
 8428	/* Transfer page from old buffer to new buffer.
 8429	 * Move each member individually to avoid possible store
 8430	 * forwarding stalls.
 8431	 */
 8432	new_buff->dma		= old_buff->dma;
 8433	new_buff->page		= old_buff->page;
 8434	new_buff->page_offset	= old_buff->page_offset;
 8435	new_buff->pagecnt_bias	= old_buff->pagecnt_bias;
 8436}
 8437
 8438static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer,
 8439				  int rx_buf_pgcnt)
 
 8440{
 8441	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
 8442	struct page *page = rx_buffer->page;
 8443
 8444	/* avoid re-using remote and pfmemalloc pages */
 8445	if (!dev_page_is_reusable(page))
 8446		return false;
 8447
 8448#if (PAGE_SIZE < 8192)
 8449	/* if we are only owner of page we can reuse it */
 8450	if (unlikely((rx_buf_pgcnt - pagecnt_bias) > 1))
 8451		return false;
 
 
 
 
 
 
 
 
 
 8452#else
 8453#define IGB_LAST_OFFSET \
 8454	(SKB_WITH_OVERHEAD(PAGE_SIZE) - IGB_RXBUFFER_2048)
 8455
 8456	if (rx_buffer->page_offset > IGB_LAST_OFFSET)
 8457		return false;
 8458#endif
 8459
 8460	/* If we have drained the page fragment pool we need to update
 8461	 * the pagecnt_bias and page count so that we fully restock the
 8462	 * number of references the driver holds.
 8463	 */
 8464	if (unlikely(pagecnt_bias == 1)) {
 8465		page_ref_add(page, USHRT_MAX - 1);
 8466		rx_buffer->pagecnt_bias = USHRT_MAX;
 8467	}
 8468
 8469	return true;
 8470}
 8471
 8472/**
 8473 *  igb_add_rx_frag - Add contents of Rx buffer to sk_buff
 8474 *  @rx_ring: rx descriptor ring to transact packets on
 8475 *  @rx_buffer: buffer containing page to add
 
 8476 *  @skb: sk_buff to place the data into
 8477 *  @size: size of buffer to be added
 8478 *
 8479 *  This function will add the data contained in rx_buffer->page to the skb.
 
 
 
 
 
 
 8480 **/
 8481static void igb_add_rx_frag(struct igb_ring *rx_ring,
 8482			    struct igb_rx_buffer *rx_buffer,
 8483			    struct sk_buff *skb,
 8484			    unsigned int size)
 8485{
 
 
 8486#if (PAGE_SIZE < 8192)
 8487	unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
 8488#else
 8489	unsigned int truesize = ring_uses_build_skb(rx_ring) ?
 8490				SKB_DATA_ALIGN(IGB_SKB_PAD + size) :
 8491				SKB_DATA_ALIGN(size);
 8492#endif
 8493	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
 8494			rx_buffer->page_offset, size, truesize);
 8495#if (PAGE_SIZE < 8192)
 8496	rx_buffer->page_offset ^= truesize;
 8497#else
 8498	rx_buffer->page_offset += truesize;
 8499#endif
 8500}
 8501
 8502static struct sk_buff *igb_construct_skb(struct igb_ring *rx_ring,
 8503					 struct igb_rx_buffer *rx_buffer,
 8504					 struct xdp_buff *xdp,
 8505					 ktime_t timestamp)
 8506{
 8507#if (PAGE_SIZE < 8192)
 8508	unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
 8509#else
 8510	unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
 8511					       xdp->data_hard_start);
 8512#endif
 8513	unsigned int size = xdp->data_end - xdp->data;
 8514	unsigned int headlen;
 8515	struct sk_buff *skb;
 8516
 8517	/* prefetch first cache line of first page */
 8518	net_prefetch(xdp->data);
 8519
 8520	/* allocate a skb to store the frags */
 8521	skb = napi_alloc_skb(&rx_ring->q_vector->napi, IGB_RX_HDR_LEN);
 8522	if (unlikely(!skb))
 8523		return NULL;
 8524
 8525	if (timestamp)
 8526		skb_hwtstamps(skb)->hwtstamp = timestamp;
 8527
 8528	/* Determine available headroom for copy */
 8529	headlen = size;
 8530	if (headlen > IGB_RX_HDR_LEN)
 8531		headlen = eth_get_headlen(skb->dev, xdp->data, IGB_RX_HDR_LEN);
 8532
 8533	/* align pull length to size of long to optimize memcpy performance */
 8534	memcpy(__skb_put(skb, headlen), xdp->data, ALIGN(headlen, sizeof(long)));
 
 
 
 8535
 8536	/* update all of the pointers */
 8537	size -= headlen;
 8538	if (size) {
 8539		skb_add_rx_frag(skb, 0, rx_buffer->page,
 8540				(xdp->data + headlen) - page_address(rx_buffer->page),
 8541				size, truesize);
 8542#if (PAGE_SIZE < 8192)
 8543		rx_buffer->page_offset ^= truesize;
 8544#else
 8545		rx_buffer->page_offset += truesize;
 8546#endif
 8547	} else {
 8548		rx_buffer->pagecnt_bias++;
 8549	}
 8550
 8551	return skb;
 8552}
 
 8553
 8554static struct sk_buff *igb_build_skb(struct igb_ring *rx_ring,
 8555				     struct igb_rx_buffer *rx_buffer,
 8556				     struct xdp_buff *xdp,
 8557				     ktime_t timestamp)
 8558{
 8559#if (PAGE_SIZE < 8192)
 8560	unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
 8561#else
 8562	unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
 8563				SKB_DATA_ALIGN(xdp->data_end -
 8564					       xdp->data_hard_start);
 8565#endif
 8566	unsigned int metasize = xdp->data - xdp->data_meta;
 8567	struct sk_buff *skb;
 8568
 8569	/* prefetch first cache line of first page */
 8570	net_prefetch(xdp->data_meta);
 8571
 8572	/* build an skb around the page buffer */
 8573	skb = napi_build_skb(xdp->data_hard_start, truesize);
 8574	if (unlikely(!skb))
 8575		return NULL;
 8576
 8577	/* update pointers within the skb to store the data */
 8578	skb_reserve(skb, xdp->data - xdp->data_hard_start);
 8579	__skb_put(skb, xdp->data_end - xdp->data);
 
 
 
 8580
 8581	if (metasize)
 8582		skb_metadata_set(skb, metasize);
 8583
 8584	if (timestamp)
 8585		skb_hwtstamps(skb)->hwtstamp = timestamp;
 8586
 8587	/* update buffer offset */
 8588#if (PAGE_SIZE < 8192)
 8589	rx_buffer->page_offset ^= truesize;
 8590#else
 8591	rx_buffer->page_offset += truesize;
 
 
 
 8592#endif
 8593
 8594	return skb;
 8595}
 8596
 8597static struct sk_buff *igb_run_xdp(struct igb_adapter *adapter,
 8598				   struct igb_ring *rx_ring,
 8599				   struct xdp_buff *xdp)
 8600{
 8601	int err, result = IGB_XDP_PASS;
 8602	struct bpf_prog *xdp_prog;
 8603	u32 act;
 8604
 8605	xdp_prog = READ_ONCE(rx_ring->xdp_prog);
 8606
 8607	if (!xdp_prog)
 8608		goto xdp_out;
 
 
 
 
 8609
 8610	prefetchw(xdp->data_hard_start); /* xdp_frame write */
 
 
 
 
 
 8611
 8612	act = bpf_prog_run_xdp(xdp_prog, xdp);
 8613	switch (act) {
 8614	case XDP_PASS:
 8615		break;
 8616	case XDP_TX:
 8617		result = igb_xdp_xmit_back(adapter, xdp);
 8618		if (result == IGB_XDP_CONSUMED)
 8619			goto out_failure;
 8620		break;
 8621	case XDP_REDIRECT:
 8622		err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
 8623		if (err)
 8624			goto out_failure;
 8625		result = IGB_XDP_REDIR;
 8626		break;
 8627	default:
 8628		bpf_warn_invalid_xdp_action(adapter->netdev, xdp_prog, act);
 8629		fallthrough;
 8630	case XDP_ABORTED:
 8631out_failure:
 8632		trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
 8633		fallthrough;
 8634	case XDP_DROP:
 8635		result = IGB_XDP_CONSUMED;
 8636		break;
 8637	}
 8638xdp_out:
 8639	return ERR_PTR(-result);
 8640}
 8641
 8642static unsigned int igb_rx_frame_truesize(struct igb_ring *rx_ring,
 8643					  unsigned int size)
 8644{
 8645	unsigned int truesize;
 8646
 8647#if (PAGE_SIZE < 8192)
 8648	truesize = igb_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
 8649#else
 8650	truesize = ring_uses_build_skb(rx_ring) ?
 8651		SKB_DATA_ALIGN(IGB_SKB_PAD + size) +
 8652		SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
 8653		SKB_DATA_ALIGN(size);
 8654#endif
 8655	return truesize;
 8656}
 8657
 8658static void igb_rx_buffer_flip(struct igb_ring *rx_ring,
 8659			       struct igb_rx_buffer *rx_buffer,
 8660			       unsigned int size)
 8661{
 8662	unsigned int truesize = igb_rx_frame_truesize(rx_ring, size);
 8663#if (PAGE_SIZE < 8192)
 8664	rx_buffer->page_offset ^= truesize;
 8665#else
 8666	rx_buffer->page_offset += truesize;
 8667#endif
 8668}
 8669
 8670static inline void igb_rx_checksum(struct igb_ring *ring,
 8671				   union e1000_adv_rx_desc *rx_desc,
 8672				   struct sk_buff *skb)
 8673{
 8674	skb_checksum_none_assert(skb);
 8675
 8676	/* Ignore Checksum bit is set */
 8677	if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
 8678		return;
 8679
 8680	/* Rx checksum disabled via ethtool */
 8681	if (!(ring->netdev->features & NETIF_F_RXCSUM))
 8682		return;
 8683
 8684	/* TCP/UDP checksum error bit is set */
 8685	if (igb_test_staterr(rx_desc,
 8686			     E1000_RXDEXT_STATERR_TCPE |
 8687			     E1000_RXDEXT_STATERR_IPE)) {
 8688		/* work around errata with sctp packets where the TCPE aka
 8689		 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
 8690		 * packets, (aka let the stack check the crc32c)
 8691		 */
 8692		if (!((skb->len == 60) &&
 8693		      test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
 8694			u64_stats_update_begin(&ring->rx_syncp);
 8695			ring->rx_stats.csum_err++;
 8696			u64_stats_update_end(&ring->rx_syncp);
 8697		}
 8698		/* let the stack verify checksum errors */
 8699		return;
 8700	}
 8701	/* It must be a TCP or UDP packet with a valid checksum */
 8702	if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
 8703				      E1000_RXD_STAT_UDPCS))
 8704		skb->ip_summed = CHECKSUM_UNNECESSARY;
 8705
 8706	dev_dbg(ring->dev, "cksum success: bits %08X\n",
 8707		le32_to_cpu(rx_desc->wb.upper.status_error));
 8708}
 8709
 8710static inline void igb_rx_hash(struct igb_ring *ring,
 8711			       union e1000_adv_rx_desc *rx_desc,
 8712			       struct sk_buff *skb)
 8713{
 8714	if (ring->netdev->features & NETIF_F_RXHASH)
 8715		skb_set_hash(skb,
 8716			     le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
 8717			     PKT_HASH_TYPE_L3);
 8718}
 8719
 8720/**
 8721 *  igb_is_non_eop - process handling of non-EOP buffers
 8722 *  @rx_ring: Rx ring being processed
 8723 *  @rx_desc: Rx descriptor for current buffer
 
 8724 *
 8725 *  This function updates next to clean.  If the buffer is an EOP buffer
 8726 *  this function exits returning false, otherwise it will place the
 8727 *  sk_buff in the next buffer to be chained and return true indicating
 8728 *  that this is in fact a non-EOP buffer.
 8729 **/
 8730static bool igb_is_non_eop(struct igb_ring *rx_ring,
 8731			   union e1000_adv_rx_desc *rx_desc)
 8732{
 8733	u32 ntc = rx_ring->next_to_clean + 1;
 8734
 8735	/* fetch, update, and store next to clean */
 8736	ntc = (ntc < rx_ring->count) ? ntc : 0;
 8737	rx_ring->next_to_clean = ntc;
 8738
 8739	prefetch(IGB_RX_DESC(rx_ring, ntc));
 8740
 8741	if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
 8742		return false;
 8743
 8744	return true;
 8745}
 8746
 8747/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 8748 *  igb_cleanup_headers - Correct corrupted or empty headers
 8749 *  @rx_ring: rx descriptor ring packet is being transacted on
 8750 *  @rx_desc: pointer to the EOP Rx descriptor
 8751 *  @skb: pointer to current skb being fixed
 8752 *
 8753 *  Address the case where we are pulling data in on pages only
 8754 *  and as such no data is present in the skb header.
 8755 *
 8756 *  In addition if skb is not at least 60 bytes we need to pad it so that
 8757 *  it is large enough to qualify as a valid Ethernet frame.
 8758 *
 8759 *  Returns true if an error was encountered and skb was freed.
 8760 **/
 8761static bool igb_cleanup_headers(struct igb_ring *rx_ring,
 8762				union e1000_adv_rx_desc *rx_desc,
 8763				struct sk_buff *skb)
 8764{
 8765	/* XDP packets use error pointer so abort at this point */
 8766	if (IS_ERR(skb))
 8767		return true;
 8768
 8769	if (unlikely((igb_test_staterr(rx_desc,
 8770				       E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
 8771		struct net_device *netdev = rx_ring->netdev;
 8772		if (!(netdev->features & NETIF_F_RXALL)) {
 8773			dev_kfree_skb_any(skb);
 8774			return true;
 8775		}
 8776	}
 8777
 8778	/* if eth_skb_pad returns an error the skb was freed */
 8779	if (eth_skb_pad(skb))
 8780		return true;
 
 
 
 
 
 
 
 
 
 8781
 8782	return false;
 8783}
 8784
 8785/**
 8786 *  igb_process_skb_fields - Populate skb header fields from Rx descriptor
 8787 *  @rx_ring: rx descriptor ring packet is being transacted on
 8788 *  @rx_desc: pointer to the EOP Rx descriptor
 8789 *  @skb: pointer to current skb being populated
 8790 *
 8791 *  This function checks the ring, descriptor, and packet information in
 8792 *  order to populate the hash, checksum, VLAN, timestamp, protocol, and
 8793 *  other fields within the skb.
 8794 **/
 8795static void igb_process_skb_fields(struct igb_ring *rx_ring,
 8796				   union e1000_adv_rx_desc *rx_desc,
 8797				   struct sk_buff *skb)
 8798{
 8799	struct net_device *dev = rx_ring->netdev;
 8800
 8801	igb_rx_hash(rx_ring, rx_desc, skb);
 8802
 8803	igb_rx_checksum(rx_ring, rx_desc, skb);
 8804
 8805	if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TS) &&
 8806	    !igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP))
 8807		igb_ptp_rx_rgtstamp(rx_ring->q_vector, skb);
 8808
 8809	if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
 8810	    igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
 8811		u16 vid;
 8812
 8813		if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
 8814		    test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
 8815			vid = be16_to_cpu((__force __be16)rx_desc->wb.upper.vlan);
 8816		else
 8817			vid = le16_to_cpu(rx_desc->wb.upper.vlan);
 8818
 8819		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
 8820	}
 8821
 8822	skb_record_rx_queue(skb, rx_ring->queue_index);
 8823
 8824	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
 8825}
 8826
 8827static unsigned int igb_rx_offset(struct igb_ring *rx_ring)
 8828{
 8829	return ring_uses_build_skb(rx_ring) ? IGB_SKB_PAD : 0;
 8830}
 8831
 8832static struct igb_rx_buffer *igb_get_rx_buffer(struct igb_ring *rx_ring,
 8833					       const unsigned int size, int *rx_buf_pgcnt)
 8834{
 8835	struct igb_rx_buffer *rx_buffer;
 8836
 8837	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
 8838	*rx_buf_pgcnt =
 8839#if (PAGE_SIZE < 8192)
 8840		page_count(rx_buffer->page);
 8841#else
 8842		0;
 8843#endif
 8844	prefetchw(rx_buffer->page);
 8845
 8846	/* we are reusing so sync this buffer for CPU use */
 8847	dma_sync_single_range_for_cpu(rx_ring->dev,
 8848				      rx_buffer->dma,
 8849				      rx_buffer->page_offset,
 8850				      size,
 8851				      DMA_FROM_DEVICE);
 8852
 8853	rx_buffer->pagecnt_bias--;
 8854
 8855	return rx_buffer;
 8856}
 8857
 8858static void igb_put_rx_buffer(struct igb_ring *rx_ring,
 8859			      struct igb_rx_buffer *rx_buffer, int rx_buf_pgcnt)
 8860{
 8861	if (igb_can_reuse_rx_page(rx_buffer, rx_buf_pgcnt)) {
 8862		/* hand second half of page back to the ring */
 8863		igb_reuse_rx_page(rx_ring, rx_buffer);
 8864	} else {
 8865		/* We are not reusing the buffer so unmap it and free
 8866		 * any references we are holding to it
 8867		 */
 8868		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
 8869				     igb_rx_pg_size(rx_ring), DMA_FROM_DEVICE,
 8870				     IGB_RX_DMA_ATTR);
 8871		__page_frag_cache_drain(rx_buffer->page,
 8872					rx_buffer->pagecnt_bias);
 8873	}
 8874
 8875	/* clear contents of rx_buffer */
 8876	rx_buffer->page = NULL;
 8877}
 8878
 8879static int igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
 8880{
 8881	struct igb_adapter *adapter = q_vector->adapter;
 8882	struct igb_ring *rx_ring = q_vector->rx.ring;
 8883	struct sk_buff *skb = rx_ring->skb;
 8884	unsigned int total_bytes = 0, total_packets = 0;
 8885	u16 cleaned_count = igb_desc_unused(rx_ring);
 8886	unsigned int xdp_xmit = 0;
 8887	struct xdp_buff xdp;
 8888	u32 frame_sz = 0;
 8889	int rx_buf_pgcnt;
 8890
 8891	/* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
 8892#if (PAGE_SIZE < 8192)
 8893	frame_sz = igb_rx_frame_truesize(rx_ring, 0);
 8894#endif
 8895	xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
 8896
 8897	while (likely(total_packets < budget)) {
 8898		union e1000_adv_rx_desc *rx_desc;
 8899		struct igb_rx_buffer *rx_buffer;
 8900		ktime_t timestamp = 0;
 8901		int pkt_offset = 0;
 8902		unsigned int size;
 8903		void *pktbuf;
 8904
 8905		/* return some buffers to hardware, one at a time is too slow */
 8906		if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
 8907			igb_alloc_rx_buffers(rx_ring, cleaned_count);
 8908			cleaned_count = 0;
 8909		}
 8910
 8911		rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
 8912		size = le16_to_cpu(rx_desc->wb.upper.length);
 8913		if (!size)
 8914			break;
 8915
 8916		/* This memory barrier is needed to keep us from reading
 8917		 * any other fields out of the rx_desc until we know the
 8918		 * descriptor has been written back
 8919		 */
 8920		dma_rmb();
 8921
 8922		rx_buffer = igb_get_rx_buffer(rx_ring, size, &rx_buf_pgcnt);
 8923		pktbuf = page_address(rx_buffer->page) + rx_buffer->page_offset;
 8924
 8925		/* pull rx packet timestamp if available and valid */
 8926		if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
 8927			int ts_hdr_len;
 8928
 8929			ts_hdr_len = igb_ptp_rx_pktstamp(rx_ring->q_vector,
 8930							 pktbuf, &timestamp);
 8931
 8932			pkt_offset += ts_hdr_len;
 8933			size -= ts_hdr_len;
 8934		}
 8935
 8936		/* retrieve a buffer from the ring */
 8937		if (!skb) {
 8938			unsigned char *hard_start = pktbuf - igb_rx_offset(rx_ring);
 8939			unsigned int offset = pkt_offset + igb_rx_offset(rx_ring);
 8940
 8941			xdp_prepare_buff(&xdp, hard_start, offset, size, true);
 8942			xdp_buff_clear_frags_flag(&xdp);
 8943#if (PAGE_SIZE > 4096)
 8944			/* At larger PAGE_SIZE, frame_sz depend on len size */
 8945			xdp.frame_sz = igb_rx_frame_truesize(rx_ring, size);
 8946#endif
 8947			skb = igb_run_xdp(adapter, rx_ring, &xdp);
 8948		}
 8949
 8950		if (IS_ERR(skb)) {
 8951			unsigned int xdp_res = -PTR_ERR(skb);
 8952
 8953			if (xdp_res & (IGB_XDP_TX | IGB_XDP_REDIR)) {
 8954				xdp_xmit |= xdp_res;
 8955				igb_rx_buffer_flip(rx_ring, rx_buffer, size);
 8956			} else {
 8957				rx_buffer->pagecnt_bias++;
 8958			}
 8959			total_packets++;
 8960			total_bytes += size;
 8961		} else if (skb)
 8962			igb_add_rx_frag(rx_ring, rx_buffer, skb, size);
 8963		else if (ring_uses_build_skb(rx_ring))
 8964			skb = igb_build_skb(rx_ring, rx_buffer, &xdp,
 8965					    timestamp);
 8966		else
 8967			skb = igb_construct_skb(rx_ring, rx_buffer,
 8968						&xdp, timestamp);
 8969
 8970		/* exit if we failed to retrieve a buffer */
 8971		if (!skb) {
 8972			rx_ring->rx_stats.alloc_failed++;
 8973			rx_buffer->pagecnt_bias++;
 8974			break;
 8975		}
 8976
 8977		igb_put_rx_buffer(rx_ring, rx_buffer, rx_buf_pgcnt);
 8978		cleaned_count++;
 8979
 8980		/* fetch next buffer in frame if non-eop */
 8981		if (igb_is_non_eop(rx_ring, rx_desc))
 8982			continue;
 8983
 8984		/* verify the packet layout is correct */
 8985		if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
 8986			skb = NULL;
 8987			continue;
 8988		}
 8989
 8990		/* probably a little skewed due to removing CRC */
 8991		total_bytes += skb->len;
 8992
 8993		/* populate checksum, timestamp, VLAN, and protocol */
 8994		igb_process_skb_fields(rx_ring, rx_desc, skb);
 8995
 8996		napi_gro_receive(&q_vector->napi, skb);
 8997
 8998		/* reset skb pointer */
 8999		skb = NULL;
 9000
 9001		/* update budget accounting */
 9002		total_packets++;
 9003	}
 9004
 9005	/* place incomplete frames back on ring for completion */
 9006	rx_ring->skb = skb;
 9007
 9008	if (xdp_xmit & IGB_XDP_REDIR)
 9009		xdp_do_flush();
 9010
 9011	if (xdp_xmit & IGB_XDP_TX) {
 9012		struct igb_ring *tx_ring = igb_xdp_tx_queue_mapping(adapter);
 9013
 9014		igb_xdp_ring_update_tail(tx_ring);
 9015	}
 9016
 9017	u64_stats_update_begin(&rx_ring->rx_syncp);
 9018	rx_ring->rx_stats.packets += total_packets;
 9019	rx_ring->rx_stats.bytes += total_bytes;
 9020	u64_stats_update_end(&rx_ring->rx_syncp);
 9021	q_vector->rx.total_packets += total_packets;
 9022	q_vector->rx.total_bytes += total_bytes;
 9023
 9024	if (cleaned_count)
 9025		igb_alloc_rx_buffers(rx_ring, cleaned_count);
 9026
 9027	return total_packets;
 9028}
 9029
 9030static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
 9031				  struct igb_rx_buffer *bi)
 9032{
 9033	struct page *page = bi->page;
 9034	dma_addr_t dma;
 9035
 9036	/* since we are recycling buffers we should seldom need to alloc */
 9037	if (likely(page))
 9038		return true;
 9039
 9040	/* alloc new page for storage */
 9041	page = dev_alloc_pages(igb_rx_pg_order(rx_ring));
 9042	if (unlikely(!page)) {
 9043		rx_ring->rx_stats.alloc_failed++;
 9044		return false;
 9045	}
 9046
 9047	/* map page for use */
 9048	dma = dma_map_page_attrs(rx_ring->dev, page, 0,
 9049				 igb_rx_pg_size(rx_ring),
 9050				 DMA_FROM_DEVICE,
 9051				 IGB_RX_DMA_ATTR);
 9052
 9053	/* if mapping failed free memory back to system since
 9054	 * there isn't much point in holding memory we can't use
 9055	 */
 9056	if (dma_mapping_error(rx_ring->dev, dma)) {
 9057		__free_pages(page, igb_rx_pg_order(rx_ring));
 9058
 9059		rx_ring->rx_stats.alloc_failed++;
 9060		return false;
 9061	}
 9062
 9063	bi->dma = dma;
 9064	bi->page = page;
 9065	bi->page_offset = igb_rx_offset(rx_ring);
 9066	page_ref_add(page, USHRT_MAX - 1);
 9067	bi->pagecnt_bias = USHRT_MAX;
 9068
 9069	return true;
 9070}
 9071
 9072/**
 9073 *  igb_alloc_rx_buffers - Replace used receive buffers
 9074 *  @rx_ring: rx descriptor ring to allocate new receive buffers
 9075 *  @cleaned_count: count of buffers to allocate
 9076 **/
 9077void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
 9078{
 9079	union e1000_adv_rx_desc *rx_desc;
 9080	struct igb_rx_buffer *bi;
 9081	u16 i = rx_ring->next_to_use;
 9082	u16 bufsz;
 9083
 9084	/* nothing to do */
 9085	if (!cleaned_count)
 9086		return;
 9087
 9088	rx_desc = IGB_RX_DESC(rx_ring, i);
 9089	bi = &rx_ring->rx_buffer_info[i];
 9090	i -= rx_ring->count;
 9091
 9092	bufsz = igb_rx_bufsz(rx_ring);
 9093
 9094	do {
 9095		if (!igb_alloc_mapped_page(rx_ring, bi))
 9096			break;
 9097
 9098		/* sync the buffer for use by the device */
 9099		dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
 9100						 bi->page_offset, bufsz,
 9101						 DMA_FROM_DEVICE);
 9102
 9103		/* Refresh the desc even if buffer_addrs didn't change
 9104		 * because each write-back erases this info.
 9105		 */
 9106		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
 9107
 9108		rx_desc++;
 9109		bi++;
 9110		i++;
 9111		if (unlikely(!i)) {
 9112			rx_desc = IGB_RX_DESC(rx_ring, 0);
 9113			bi = rx_ring->rx_buffer_info;
 9114			i -= rx_ring->count;
 9115		}
 9116
 9117		/* clear the length for the next_to_use descriptor */
 9118		rx_desc->wb.upper.length = 0;
 9119
 9120		cleaned_count--;
 9121	} while (cleaned_count);
 9122
 9123	i += rx_ring->count;
 9124
 9125	if (rx_ring->next_to_use != i) {
 9126		/* record the next descriptor to use */
 9127		rx_ring->next_to_use = i;
 9128
 9129		/* update next to alloc since we have filled the ring */
 9130		rx_ring->next_to_alloc = i;
 9131
 9132		/* Force memory writes to complete before letting h/w
 9133		 * know there are new descriptors to fetch.  (Only
 9134		 * applicable for weak-ordered memory model archs,
 9135		 * such as IA-64).
 9136		 */
 9137		dma_wmb();
 9138		writel(i, rx_ring->tail);
 9139	}
 9140}
 9141
 9142/**
 9143 * igb_mii_ioctl -
 9144 * @netdev: pointer to netdev struct
 9145 * @ifr: interface structure
 9146 * @cmd: ioctl command to execute
 9147 **/
 9148static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
 9149{
 9150	struct igb_adapter *adapter = netdev_priv(netdev);
 9151	struct mii_ioctl_data *data = if_mii(ifr);
 9152
 9153	if (adapter->hw.phy.media_type != e1000_media_type_copper)
 9154		return -EOPNOTSUPP;
 9155
 9156	switch (cmd) {
 9157	case SIOCGMIIPHY:
 9158		data->phy_id = adapter->hw.phy.addr;
 9159		break;
 9160	case SIOCGMIIREG:
 9161		if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
 9162				     &data->val_out))
 9163			return -EIO;
 9164		break;
 9165	case SIOCSMIIREG:
 9166	default:
 9167		return -EOPNOTSUPP;
 9168	}
 9169	return 0;
 9170}
 9171
 9172/**
 9173 * igb_ioctl -
 9174 * @netdev: pointer to netdev struct
 9175 * @ifr: interface structure
 9176 * @cmd: ioctl command to execute
 9177 **/
 9178static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
 9179{
 9180	switch (cmd) {
 9181	case SIOCGMIIPHY:
 9182	case SIOCGMIIREG:
 9183	case SIOCSMIIREG:
 9184		return igb_mii_ioctl(netdev, ifr, cmd);
 9185	case SIOCGHWTSTAMP:
 9186		return igb_ptp_get_ts_config(netdev, ifr);
 9187	case SIOCSHWTSTAMP:
 9188		return igb_ptp_set_ts_config(netdev, ifr);
 9189	default:
 9190		return -EOPNOTSUPP;
 9191	}
 9192}
 9193
 9194void igb_read_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
 9195{
 9196	struct igb_adapter *adapter = hw->back;
 9197
 9198	pci_read_config_word(adapter->pdev, reg, value);
 9199}
 9200
 9201void igb_write_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
 9202{
 9203	struct igb_adapter *adapter = hw->back;
 9204
 9205	pci_write_config_word(adapter->pdev, reg, *value);
 9206}
 9207
 9208s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
 9209{
 9210	struct igb_adapter *adapter = hw->back;
 9211
 9212	if (pcie_capability_read_word(adapter->pdev, reg, value))
 9213		return -E1000_ERR_CONFIG;
 9214
 9215	return 0;
 9216}
 9217
 9218s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
 9219{
 9220	struct igb_adapter *adapter = hw->back;
 9221
 9222	if (pcie_capability_write_word(adapter->pdev, reg, *value))
 9223		return -E1000_ERR_CONFIG;
 9224
 9225	return 0;
 9226}
 9227
 9228static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
 9229{
 9230	struct igb_adapter *adapter = netdev_priv(netdev);
 9231	struct e1000_hw *hw = &adapter->hw;
 9232	u32 ctrl, rctl;
 9233	bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
 9234
 9235	if (enable) {
 9236		/* enable VLAN tag insert/strip */
 9237		ctrl = rd32(E1000_CTRL);
 9238		ctrl |= E1000_CTRL_VME;
 9239		wr32(E1000_CTRL, ctrl);
 9240
 9241		/* Disable CFI check */
 9242		rctl = rd32(E1000_RCTL);
 9243		rctl &= ~E1000_RCTL_CFIEN;
 9244		wr32(E1000_RCTL, rctl);
 9245	} else {
 9246		/* disable VLAN tag insert/strip */
 9247		ctrl = rd32(E1000_CTRL);
 9248		ctrl &= ~E1000_CTRL_VME;
 9249		wr32(E1000_CTRL, ctrl);
 9250	}
 9251
 9252	igb_set_vf_vlan_strip(adapter, adapter->vfs_allocated_count, enable);
 9253}
 9254
 9255static int igb_vlan_rx_add_vid(struct net_device *netdev,
 9256			       __be16 proto, u16 vid)
 9257{
 9258	struct igb_adapter *adapter = netdev_priv(netdev);
 9259	struct e1000_hw *hw = &adapter->hw;
 9260	int pf_id = adapter->vfs_allocated_count;
 9261
 
 
 
 9262	/* add the filter since PF can receive vlans w/o entry in vlvf */
 9263	if (!vid || !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
 9264		igb_vfta_set(hw, vid, pf_id, true, !!vid);
 9265
 9266	set_bit(vid, adapter->active_vlans);
 9267
 9268	return 0;
 9269}
 9270
 9271static int igb_vlan_rx_kill_vid(struct net_device *netdev,
 9272				__be16 proto, u16 vid)
 9273{
 9274	struct igb_adapter *adapter = netdev_priv(netdev);
 9275	int pf_id = adapter->vfs_allocated_count;
 9276	struct e1000_hw *hw = &adapter->hw;
 
 
 9277
 9278	/* remove VID from filter table */
 9279	if (vid && !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
 9280		igb_vfta_set(hw, vid, pf_id, false, true);
 
 
 
 9281
 9282	clear_bit(vid, adapter->active_vlans);
 9283
 9284	return 0;
 9285}
 9286
 9287static void igb_restore_vlan(struct igb_adapter *adapter)
 9288{
 9289	u16 vid = 1;
 9290
 9291	igb_vlan_mode(adapter->netdev, adapter->netdev->features);
 9292	igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
 9293
 9294	for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
 9295		igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
 9296}
 9297
 9298int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
 9299{
 9300	struct pci_dev *pdev = adapter->pdev;
 9301	struct e1000_mac_info *mac = &adapter->hw.mac;
 9302
 9303	mac->autoneg = 0;
 9304
 9305	/* Make sure dplx is at most 1 bit and lsb of speed is not set
 9306	 * for the switch() below to work
 9307	 */
 9308	if ((spd & 1) || (dplx & ~1))
 9309		goto err_inval;
 9310
 9311	/* Fiber NIC's only allow 1000 gbps Full duplex
 9312	 * and 100Mbps Full duplex for 100baseFx sfp
 9313	 */
 9314	if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
 9315		switch (spd + dplx) {
 9316		case SPEED_10 + DUPLEX_HALF:
 9317		case SPEED_10 + DUPLEX_FULL:
 9318		case SPEED_100 + DUPLEX_HALF:
 9319			goto err_inval;
 9320		default:
 9321			break;
 9322		}
 9323	}
 9324
 9325	switch (spd + dplx) {
 9326	case SPEED_10 + DUPLEX_HALF:
 9327		mac->forced_speed_duplex = ADVERTISE_10_HALF;
 9328		break;
 9329	case SPEED_10 + DUPLEX_FULL:
 9330		mac->forced_speed_duplex = ADVERTISE_10_FULL;
 9331		break;
 9332	case SPEED_100 + DUPLEX_HALF:
 9333		mac->forced_speed_duplex = ADVERTISE_100_HALF;
 9334		break;
 9335	case SPEED_100 + DUPLEX_FULL:
 9336		mac->forced_speed_duplex = ADVERTISE_100_FULL;
 9337		break;
 9338	case SPEED_1000 + DUPLEX_FULL:
 9339		mac->autoneg = 1;
 9340		adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
 9341		break;
 9342	case SPEED_1000 + DUPLEX_HALF: /* not supported */
 9343	default:
 9344		goto err_inval;
 9345	}
 9346
 9347	/* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
 9348	adapter->hw.phy.mdix = AUTO_ALL_MODES;
 9349
 9350	return 0;
 9351
 9352err_inval:
 9353	dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
 9354	return -EINVAL;
 9355}
 9356
 9357static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
 9358			  bool runtime)
 9359{
 9360	struct net_device *netdev = pci_get_drvdata(pdev);
 9361	struct igb_adapter *adapter = netdev_priv(netdev);
 9362	struct e1000_hw *hw = &adapter->hw;
 9363	u32 ctrl, rctl, status;
 9364	u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
 9365	bool wake;
 
 
 9366
 9367	rtnl_lock();
 9368	netif_device_detach(netdev);
 9369
 9370	if (netif_running(netdev))
 9371		__igb_close(netdev, true);
 9372
 9373	igb_ptp_suspend(adapter);
 9374
 9375	igb_clear_interrupt_scheme(adapter);
 9376	rtnl_unlock();
 
 
 
 
 
 9377
 9378	status = rd32(E1000_STATUS);
 9379	if (status & E1000_STATUS_LU)
 9380		wufc &= ~E1000_WUFC_LNKC;
 9381
 9382	if (wufc) {
 9383		igb_setup_rctl(adapter);
 9384		igb_set_rx_mode(netdev);
 9385
 9386		/* turn on all-multi mode if wake on multicast is enabled */
 9387		if (wufc & E1000_WUFC_MC) {
 9388			rctl = rd32(E1000_RCTL);
 9389			rctl |= E1000_RCTL_MPE;
 9390			wr32(E1000_RCTL, rctl);
 9391		}
 9392
 9393		ctrl = rd32(E1000_CTRL);
 
 
 
 
 9394		ctrl |= E1000_CTRL_ADVD3WUC;
 9395		wr32(E1000_CTRL, ctrl);
 9396
 9397		/* Allow time for pending master requests to run */
 9398		igb_disable_pcie_master(hw);
 9399
 9400		wr32(E1000_WUC, E1000_WUC_PME_EN);
 9401		wr32(E1000_WUFC, wufc);
 9402	} else {
 9403		wr32(E1000_WUC, 0);
 9404		wr32(E1000_WUFC, 0);
 9405	}
 9406
 9407	wake = wufc || adapter->en_mng_pt;
 9408	if (!wake)
 9409		igb_power_down_link(adapter);
 9410	else
 9411		igb_power_up_link(adapter);
 9412
 9413	if (enable_wake)
 9414		*enable_wake = wake;
 9415
 9416	/* Release control of h/w to f/w.  If f/w is AMT enabled, this
 9417	 * would have already happened in close and is redundant.
 9418	 */
 9419	igb_release_hw_control(adapter);
 9420
 9421	pci_disable_device(pdev);
 9422
 9423	return 0;
 9424}
 9425
 9426static void igb_deliver_wake_packet(struct net_device *netdev)
 
 
 9427{
 9428	struct igb_adapter *adapter = netdev_priv(netdev);
 9429	struct e1000_hw *hw = &adapter->hw;
 9430	struct sk_buff *skb;
 9431	u32 wupl;
 9432
 9433	wupl = rd32(E1000_WUPL) & E1000_WUPL_MASK;
 9434
 9435	/* WUPM stores only the first 128 bytes of the wake packet.
 9436	 * Read the packet only if we have the whole thing.
 9437	 */
 9438	if ((wupl == 0) || (wupl > E1000_WUPM_BYTES))
 9439		return;
 9440
 9441	skb = netdev_alloc_skb_ip_align(netdev, E1000_WUPM_BYTES);
 9442	if (!skb)
 9443		return;
 9444
 9445	skb_put(skb, wupl);
 9446
 9447	/* Ensure reads are 32-bit aligned */
 9448	wupl = roundup(wupl, 4);
 9449
 9450	memcpy_fromio(skb->data, hw->hw_addr + E1000_WUPM_REG(0), wupl);
 
 
 9451
 9452	skb->protocol = eth_type_trans(skb, netdev);
 9453	netif_rx(skb);
 9454}
 
 
 
 9455
 9456static int __maybe_unused igb_suspend(struct device *dev)
 9457{
 9458	return __igb_shutdown(to_pci_dev(dev), NULL, 0);
 9459}
 
 9460
 9461static int __maybe_unused __igb_resume(struct device *dev, bool rpm)
 9462{
 9463	struct pci_dev *pdev = to_pci_dev(dev);
 9464	struct net_device *netdev = pci_get_drvdata(pdev);
 9465	struct igb_adapter *adapter = netdev_priv(netdev);
 9466	struct e1000_hw *hw = &adapter->hw;
 9467	u32 err, val;
 9468
 9469	pci_set_power_state(pdev, PCI_D0);
 9470	pci_restore_state(pdev);
 9471	pci_save_state(pdev);
 9472
 9473	if (!pci_device_is_present(pdev))
 9474		return -ENODEV;
 9475	err = pci_enable_device_mem(pdev);
 9476	if (err) {
 9477		dev_err(&pdev->dev,
 9478			"igb: Cannot enable PCI device from suspend\n");
 9479		return err;
 9480	}
 9481	pci_set_master(pdev);
 9482
 9483	pci_enable_wake(pdev, PCI_D3hot, 0);
 9484	pci_enable_wake(pdev, PCI_D3cold, 0);
 9485
 9486	if (igb_init_interrupt_scheme(adapter, true)) {
 9487		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
 9488		return -ENOMEM;
 9489	}
 9490
 9491	igb_reset(adapter);
 9492
 9493	/* let the f/w know that the h/w is now under the control of the
 9494	 * driver.
 9495	 */
 9496	igb_get_hw_control(adapter);
 9497
 9498	val = rd32(E1000_WUS);
 9499	if (val & WAKE_PKT_WUS)
 9500		igb_deliver_wake_packet(netdev);
 9501
 9502	wr32(E1000_WUS, ~0);
 9503
 9504	if (!rpm)
 9505		rtnl_lock();
 9506	if (!err && netif_running(netdev))
 9507		err = __igb_open(netdev, true);
 9508
 9509	if (!err)
 9510		netif_device_attach(netdev);
 9511	if (!rpm)
 9512		rtnl_unlock();
 
 
 
 9513
 9514	return err;
 9515}
 9516
 9517static int __maybe_unused igb_resume(struct device *dev)
 9518{
 9519	return __igb_resume(dev, false);
 9520}
 9521
 9522static int __maybe_unused igb_runtime_idle(struct device *dev)
 
 9523{
 9524	struct net_device *netdev = dev_get_drvdata(dev);
 
 9525	struct igb_adapter *adapter = netdev_priv(netdev);
 9526
 9527	if (!igb_has_link(adapter))
 9528		pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
 9529
 9530	return -EBUSY;
 9531}
 9532
 9533static int __maybe_unused igb_runtime_suspend(struct device *dev)
 9534{
 9535	return __igb_shutdown(to_pci_dev(dev), NULL, 1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 9536}
 9537
 9538static int __maybe_unused igb_runtime_resume(struct device *dev)
 9539{
 9540	return __igb_resume(dev, true);
 9541}
 
 
 9542
 9543static void igb_shutdown(struct pci_dev *pdev)
 9544{
 9545	bool wake;
 9546
 9547	__igb_shutdown(pdev, &wake, 0);
 9548
 9549	if (system_state == SYSTEM_POWER_OFF) {
 9550		pci_wake_from_d3(pdev, wake);
 9551		pci_set_power_state(pdev, PCI_D3hot);
 9552	}
 9553}
 9554
 9555static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
 9556{
 9557#ifdef CONFIG_PCI_IOV
 9558	int err;
 
 
 
 
 9559
 9560	if (num_vfs == 0) {
 9561		return igb_disable_sriov(dev, true);
 9562	} else {
 9563		err = igb_enable_sriov(dev, num_vfs, true);
 9564		return err ? err : num_vfs;
 
 
 
 
 
 
 
 9565	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 9566#endif
 9567	return 0;
 9568}
 9569
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 9570/**
 9571 *  igb_io_error_detected - called when PCI error is detected
 9572 *  @pdev: Pointer to PCI device
 9573 *  @state: The current pci connection state
 9574 *
 9575 *  This function is called after a PCI bus error affecting
 9576 *  this device has been detected.
 9577 **/
 9578static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
 9579					      pci_channel_state_t state)
 9580{
 9581	struct net_device *netdev = pci_get_drvdata(pdev);
 9582	struct igb_adapter *adapter = netdev_priv(netdev);
 9583
 9584	if (state == pci_channel_io_normal) {
 9585		dev_warn(&pdev->dev, "Non-correctable non-fatal error reported.\n");
 9586		return PCI_ERS_RESULT_CAN_RECOVER;
 9587	}
 9588
 9589	netif_device_detach(netdev);
 9590
 9591	if (state == pci_channel_io_perm_failure)
 9592		return PCI_ERS_RESULT_DISCONNECT;
 9593
 9594	if (netif_running(netdev))
 9595		igb_down(adapter);
 9596	pci_disable_device(pdev);
 9597
 9598	/* Request a slot reset. */
 9599	return PCI_ERS_RESULT_NEED_RESET;
 9600}
 9601
 9602/**
 9603 *  igb_io_slot_reset - called after the pci bus has been reset.
 9604 *  @pdev: Pointer to PCI device
 9605 *
 9606 *  Restart the card from scratch, as if from a cold-boot. Implementation
 9607 *  resembles the first-half of the __igb_resume routine.
 9608 **/
 9609static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
 9610{
 9611	struct net_device *netdev = pci_get_drvdata(pdev);
 9612	struct igb_adapter *adapter = netdev_priv(netdev);
 9613	struct e1000_hw *hw = &adapter->hw;
 9614	pci_ers_result_t result;
 
 9615
 9616	if (pci_enable_device_mem(pdev)) {
 9617		dev_err(&pdev->dev,
 9618			"Cannot re-enable PCI device after reset.\n");
 9619		result = PCI_ERS_RESULT_DISCONNECT;
 9620	} else {
 9621		pci_set_master(pdev);
 9622		pci_restore_state(pdev);
 9623		pci_save_state(pdev);
 9624
 9625		pci_enable_wake(pdev, PCI_D3hot, 0);
 9626		pci_enable_wake(pdev, PCI_D3cold, 0);
 9627
 9628		/* In case of PCI error, adapter lose its HW address
 9629		 * so we should re-assign it here.
 9630		 */
 9631		hw->hw_addr = adapter->io_addr;
 9632
 9633		igb_reset(adapter);
 9634		wr32(E1000_WUS, ~0);
 9635		result = PCI_ERS_RESULT_RECOVERED;
 9636	}
 9637
 
 
 
 
 
 
 
 
 9638	return result;
 9639}
 9640
 9641/**
 9642 *  igb_io_resume - called when traffic can start flowing again.
 9643 *  @pdev: Pointer to PCI device
 9644 *
 9645 *  This callback is called when the error recovery driver tells us that
 9646 *  its OK to resume normal operation. Implementation resembles the
 9647 *  second-half of the __igb_resume routine.
 9648 */
 9649static void igb_io_resume(struct pci_dev *pdev)
 9650{
 9651	struct net_device *netdev = pci_get_drvdata(pdev);
 9652	struct igb_adapter *adapter = netdev_priv(netdev);
 9653
 9654	if (netif_running(netdev)) {
 9655		if (igb_up(adapter)) {
 9656			dev_err(&pdev->dev, "igb_up failed after reset\n");
 9657			return;
 9658		}
 9659	}
 9660
 9661	netif_device_attach(netdev);
 9662
 9663	/* let the f/w know that the h/w is now under the control of the
 9664	 * driver.
 9665	 */
 9666	igb_get_hw_control(adapter);
 9667}
 9668
 9669/**
 9670 *  igb_rar_set_index - Sync RAL[index] and RAH[index] registers with MAC table
 9671 *  @adapter: Pointer to adapter structure
 9672 *  @index: Index of the RAR entry which need to be synced with MAC table
 9673 **/
 9674static void igb_rar_set_index(struct igb_adapter *adapter, u32 index)
 9675{
 9676	struct e1000_hw *hw = &adapter->hw;
 9677	u32 rar_low, rar_high;
 9678	u8 *addr = adapter->mac_table[index].addr;
 9679
 9680	/* HW expects these to be in network order when they are plugged
 9681	 * into the registers which are little endian.  In order to guarantee
 9682	 * that ordering we need to do an leXX_to_cpup here in order to be
 9683	 * ready for the byteswap that occurs with writel
 9684	 */
 9685	rar_low = le32_to_cpup((__le32 *)(addr));
 9686	rar_high = le16_to_cpup((__le16 *)(addr + 4));
 
 9687
 9688	/* Indicate to hardware the Address is Valid. */
 9689	if (adapter->mac_table[index].state & IGB_MAC_STATE_IN_USE) {
 9690		if (is_valid_ether_addr(addr))
 9691			rar_high |= E1000_RAH_AV;
 9692
 9693		if (adapter->mac_table[index].state & IGB_MAC_STATE_SRC_ADDR)
 9694			rar_high |= E1000_RAH_ASEL_SRC_ADDR;
 9695
 9696		switch (hw->mac.type) {
 9697		case e1000_82575:
 9698		case e1000_i210:
 9699			if (adapter->mac_table[index].state &
 9700			    IGB_MAC_STATE_QUEUE_STEERING)
 9701				rar_high |= E1000_RAH_QSEL_ENABLE;
 9702
 9703			rar_high |= E1000_RAH_POOL_1 *
 9704				    adapter->mac_table[index].queue;
 9705			break;
 9706		default:
 9707			rar_high |= E1000_RAH_POOL_1 <<
 9708				    adapter->mac_table[index].queue;
 9709			break;
 9710		}
 9711	}
 9712
 9713	wr32(E1000_RAL(index), rar_low);
 9714	wrfl();
 9715	wr32(E1000_RAH(index), rar_high);
 9716	wrfl();
 9717}
 9718
 9719static int igb_set_vf_mac(struct igb_adapter *adapter,
 9720			  int vf, unsigned char *mac_addr)
 9721{
 9722	struct e1000_hw *hw = &adapter->hw;
 9723	/* VF MAC addresses start at end of receive addresses and moves
 9724	 * towards the first, as a result a collision should not be possible
 9725	 */
 9726	int rar_entry = hw->mac.rar_entry_count - (vf + 1);
 9727	unsigned char *vf_mac_addr = adapter->vf_data[vf].vf_mac_addresses;
 9728
 9729	ether_addr_copy(vf_mac_addr, mac_addr);
 9730	ether_addr_copy(adapter->mac_table[rar_entry].addr, mac_addr);
 9731	adapter->mac_table[rar_entry].queue = vf;
 9732	adapter->mac_table[rar_entry].state |= IGB_MAC_STATE_IN_USE;
 9733	igb_rar_set_index(adapter, rar_entry);
 9734
 9735	return 0;
 9736}
 9737
 9738static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
 9739{
 9740	struct igb_adapter *adapter = netdev_priv(netdev);
 9741
 9742	if (vf >= adapter->vfs_allocated_count)
 9743		return -EINVAL;
 9744
 9745	/* Setting the VF MAC to 0 reverts the IGB_VF_FLAG_PF_SET_MAC
 9746	 * flag and allows to overwrite the MAC via VF netdev.  This
 9747	 * is necessary to allow libvirt a way to restore the original
 9748	 * MAC after unbinding vfio-pci and reloading igbvf after shutting
 9749	 * down a VM.
 9750	 */
 9751	if (is_zero_ether_addr(mac)) {
 9752		adapter->vf_data[vf].flags &= ~IGB_VF_FLAG_PF_SET_MAC;
 9753		dev_info(&adapter->pdev->dev,
 9754			 "remove administratively set MAC on VF %d\n",
 9755			 vf);
 9756	} else if (is_valid_ether_addr(mac)) {
 9757		adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
 9758		dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n",
 9759			 mac, vf);
 9760		dev_info(&adapter->pdev->dev,
 9761			 "Reload the VF driver to make this change effective.");
 9762		/* Generate additional warning if PF is down */
 9763		if (test_bit(__IGB_DOWN, &adapter->state)) {
 9764			dev_warn(&adapter->pdev->dev,
 9765				 "The VF MAC address has been set, but the PF device is not up.\n");
 9766			dev_warn(&adapter->pdev->dev,
 9767				 "Bring the PF device up before attempting to use the VF device.\n");
 9768		}
 9769	} else {
 9770		return -EINVAL;
 
 
 
 
 
 
 
 
 
 9771	}
 9772	return igb_set_vf_mac(adapter, vf, mac);
 9773}
 9774
 9775static int igb_link_mbps(int internal_link_speed)
 9776{
 9777	switch (internal_link_speed) {
 9778	case SPEED_100:
 9779		return 100;
 9780	case SPEED_1000:
 9781		return 1000;
 9782	default:
 9783		return 0;
 9784	}
 9785}
 9786
 9787static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
 9788				  int link_speed)
 9789{
 9790	int rf_dec, rf_int;
 9791	u32 bcnrc_val;
 9792
 9793	if (tx_rate != 0) {
 9794		/* Calculate the rate factor values to set */
 9795		rf_int = link_speed / tx_rate;
 9796		rf_dec = (link_speed - (rf_int * tx_rate));
 9797		rf_dec = (rf_dec * BIT(E1000_RTTBCNRC_RF_INT_SHIFT)) /
 9798			 tx_rate;
 9799
 9800		bcnrc_val = E1000_RTTBCNRC_RS_ENA;
 9801		bcnrc_val |= FIELD_PREP(E1000_RTTBCNRC_RF_INT_MASK, rf_int);
 
 9802		bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
 9803	} else {
 9804		bcnrc_val = 0;
 9805	}
 9806
 9807	wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
 9808	/* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
 9809	 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
 9810	 */
 9811	wr32(E1000_RTTBCNRM, 0x14);
 9812	wr32(E1000_RTTBCNRC, bcnrc_val);
 9813}
 9814
 9815static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
 9816{
 9817	int actual_link_speed, i;
 9818	bool reset_rate = false;
 9819
 9820	/* VF TX rate limit was not set or not supported */
 9821	if ((adapter->vf_rate_link_speed == 0) ||
 9822	    (adapter->hw.mac.type != e1000_82576))
 9823		return;
 9824
 9825	actual_link_speed = igb_link_mbps(adapter->link_speed);
 9826	if (actual_link_speed != adapter->vf_rate_link_speed) {
 9827		reset_rate = true;
 9828		adapter->vf_rate_link_speed = 0;
 9829		dev_info(&adapter->pdev->dev,
 9830			 "Link speed has been changed. VF Transmit rate is disabled\n");
 9831	}
 9832
 9833	for (i = 0; i < adapter->vfs_allocated_count; i++) {
 9834		if (reset_rate)
 9835			adapter->vf_data[i].tx_rate = 0;
 9836
 9837		igb_set_vf_rate_limit(&adapter->hw, i,
 9838				      adapter->vf_data[i].tx_rate,
 9839				      actual_link_speed);
 9840	}
 9841}
 9842
 9843static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf,
 9844			     int min_tx_rate, int max_tx_rate)
 9845{
 9846	struct igb_adapter *adapter = netdev_priv(netdev);
 9847	struct e1000_hw *hw = &adapter->hw;
 9848	int actual_link_speed;
 9849
 9850	if (hw->mac.type != e1000_82576)
 9851		return -EOPNOTSUPP;
 9852
 9853	if (min_tx_rate)
 9854		return -EINVAL;
 9855
 9856	actual_link_speed = igb_link_mbps(adapter->link_speed);
 9857	if ((vf >= adapter->vfs_allocated_count) ||
 9858	    (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
 9859	    (max_tx_rate < 0) ||
 9860	    (max_tx_rate > actual_link_speed))
 9861		return -EINVAL;
 9862
 9863	adapter->vf_rate_link_speed = actual_link_speed;
 9864	adapter->vf_data[vf].tx_rate = (u16)max_tx_rate;
 9865	igb_set_vf_rate_limit(hw, vf, max_tx_rate, actual_link_speed);
 9866
 9867	return 0;
 9868}
 9869
 9870static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
 9871				   bool setting)
 9872{
 9873	struct igb_adapter *adapter = netdev_priv(netdev);
 9874	struct e1000_hw *hw = &adapter->hw;
 9875	u32 reg_val, reg_offset;
 9876
 9877	if (!adapter->vfs_allocated_count)
 9878		return -EOPNOTSUPP;
 9879
 9880	if (vf >= adapter->vfs_allocated_count)
 9881		return -EINVAL;
 9882
 9883	reg_offset = (hw->mac.type == e1000_82576) ? E1000_DTXSWC : E1000_TXSWC;
 9884	reg_val = rd32(reg_offset);
 9885	if (setting)
 9886		reg_val |= (BIT(vf) |
 9887			    BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
 9888	else
 9889		reg_val &= ~(BIT(vf) |
 9890			     BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
 9891	wr32(reg_offset, reg_val);
 9892
 9893	adapter->vf_data[vf].spoofchk_enabled = setting;
 9894	return 0;
 9895}
 9896
 9897static int igb_ndo_set_vf_trust(struct net_device *netdev, int vf, bool setting)
 9898{
 9899	struct igb_adapter *adapter = netdev_priv(netdev);
 9900
 9901	if (vf >= adapter->vfs_allocated_count)
 9902		return -EINVAL;
 9903	if (adapter->vf_data[vf].trusted == setting)
 9904		return 0;
 9905
 9906	adapter->vf_data[vf].trusted = setting;
 9907
 9908	dev_info(&adapter->pdev->dev, "VF %u is %strusted\n",
 9909		 vf, setting ? "" : "not ");
 9910	return 0;
 9911}
 9912
 9913static int igb_ndo_get_vf_config(struct net_device *netdev,
 9914				 int vf, struct ifla_vf_info *ivi)
 9915{
 9916	struct igb_adapter *adapter = netdev_priv(netdev);
 9917	if (vf >= adapter->vfs_allocated_count)
 9918		return -EINVAL;
 9919	ivi->vf = vf;
 9920	memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
 9921	ivi->max_tx_rate = adapter->vf_data[vf].tx_rate;
 9922	ivi->min_tx_rate = 0;
 9923	ivi->vlan = adapter->vf_data[vf].pf_vlan;
 9924	ivi->qos = adapter->vf_data[vf].pf_qos;
 9925	ivi->spoofchk = adapter->vf_data[vf].spoofchk_enabled;
 9926	ivi->trusted = adapter->vf_data[vf].trusted;
 9927	return 0;
 9928}
 9929
 9930static void igb_vmm_control(struct igb_adapter *adapter)
 9931{
 9932	struct e1000_hw *hw = &adapter->hw;
 9933	u32 reg;
 9934
 9935	switch (hw->mac.type) {
 9936	case e1000_82575:
 9937	case e1000_i210:
 9938	case e1000_i211:
 9939	case e1000_i354:
 9940	default:
 9941		/* replication is not supported for 82575 */
 9942		return;
 9943	case e1000_82576:
 9944		/* notify HW that the MAC is adding vlan tags */
 9945		reg = rd32(E1000_DTXCTL);
 9946		reg |= E1000_DTXCTL_VLAN_ADDED;
 9947		wr32(E1000_DTXCTL, reg);
 9948		fallthrough;
 9949	case e1000_82580:
 9950		/* enable replication vlan tag stripping */
 9951		reg = rd32(E1000_RPLOLR);
 9952		reg |= E1000_RPLOLR_STRVLAN;
 9953		wr32(E1000_RPLOLR, reg);
 9954		fallthrough;
 9955	case e1000_i350:
 9956		/* none of the above registers are supported by i350 */
 9957		break;
 9958	}
 9959
 9960	if (adapter->vfs_allocated_count) {
 9961		igb_vmdq_set_loopback_pf(hw, true);
 9962		igb_vmdq_set_replication_pf(hw, true);
 9963		igb_vmdq_set_anti_spoofing_pf(hw, true,
 9964					      adapter->vfs_allocated_count);
 9965	} else {
 9966		igb_vmdq_set_loopback_pf(hw, false);
 9967		igb_vmdq_set_replication_pf(hw, false);
 9968	}
 9969}
 9970
 9971static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
 9972{
 9973	struct e1000_hw *hw = &adapter->hw;
 9974	u32 dmac_thr;
 9975	u16 hwm;
 9976	u32 reg;
 9977
 9978	if (hw->mac.type > e1000_82580) {
 9979		if (adapter->flags & IGB_FLAG_DMAC) {
 
 
 9980			/* force threshold to 0. */
 9981			wr32(E1000_DMCTXTH, 0);
 9982
 9983			/* DMA Coalescing high water mark needs to be greater
 9984			 * than the Rx threshold. Set hwm to PBA - max frame
 9985			 * size in 16B units, capping it at PBA - 6KB.
 9986			 */
 9987			hwm = 64 * (pba - 6);
 
 
 9988			reg = rd32(E1000_FCRTC);
 9989			reg &= ~E1000_FCRTC_RTH_COAL_MASK;
 9990			reg |= FIELD_PREP(E1000_FCRTC_RTH_COAL_MASK, hwm);
 
 9991			wr32(E1000_FCRTC, reg);
 9992
 9993			/* Set the DMA Coalescing Rx threshold to PBA - 2 * max
 9994			 * frame size, capping it at PBA - 10KB.
 9995			 */
 9996			dmac_thr = pba - 10;
 
 
 9997			reg = rd32(E1000_DMACR);
 9998			reg &= ~E1000_DMACR_DMACTHR_MASK;
 9999			reg |= FIELD_PREP(E1000_DMACR_DMACTHR_MASK, dmac_thr);
 
10000
10001			/* transition to L0x or L1 if available..*/
10002			reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
10003
10004			/* watchdog timer= +-1000 usec in 32usec intervals */
10005			reg |= (1000 >> 5);
10006
10007			/* Disable BMC-to-OS Watchdog Enable */
10008			if (hw->mac.type != e1000_i354)
10009				reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
 
10010			wr32(E1000_DMACR, reg);
10011
10012			/* no lower threshold to disable
10013			 * coalescing(smart fifb)-UTRESH=0
10014			 */
10015			wr32(E1000_DMCRTRH, 0);
10016
10017			reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
10018
10019			wr32(E1000_DMCTLX, reg);
10020
10021			/* free space in tx packet buffer to wake from
10022			 * DMA coal
10023			 */
10024			wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
10025			     (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
10026		}
10027
10028		if (hw->mac.type >= e1000_i210 ||
10029		    (adapter->flags & IGB_FLAG_DMAC)) {
 
10030			reg = rd32(E1000_PCIEMISC);
10031			reg |= E1000_PCIEMISC_LX_DECISION;
10032			wr32(E1000_PCIEMISC, reg);
10033		} /* endif adapter->dmac is not disabled */
10034	} else if (hw->mac.type == e1000_82580) {
10035		u32 reg = rd32(E1000_PCIEMISC);
10036
10037		wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
10038		wr32(E1000_DMACR, 0);
10039	}
10040}
10041
10042/**
10043 *  igb_read_i2c_byte - Reads 8 bit word over I2C
10044 *  @hw: pointer to hardware structure
10045 *  @byte_offset: byte offset to read
10046 *  @dev_addr: device address
10047 *  @data: value read
10048 *
10049 *  Performs byte read operation over I2C interface at
10050 *  a specified device address.
10051 **/
10052s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
10053		      u8 dev_addr, u8 *data)
10054{
10055	struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
10056	struct i2c_client *this_client = adapter->i2c_client;
10057	s32 status;
10058	u16 swfw_mask = 0;
10059
10060	if (!this_client)
10061		return E1000_ERR_I2C;
10062
10063	swfw_mask = E1000_SWFW_PHY0_SM;
10064
10065	if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
 
10066		return E1000_ERR_SWFW_SYNC;
10067
10068	status = i2c_smbus_read_byte_data(this_client, byte_offset);
10069	hw->mac.ops.release_swfw_sync(hw, swfw_mask);
10070
10071	if (status < 0)
10072		return E1000_ERR_I2C;
10073	else {
10074		*data = status;
10075		return 0;
10076	}
10077}
10078
10079/**
10080 *  igb_write_i2c_byte - Writes 8 bit word over I2C
10081 *  @hw: pointer to hardware structure
10082 *  @byte_offset: byte offset to write
10083 *  @dev_addr: device address
10084 *  @data: value to write
10085 *
10086 *  Performs byte write operation over I2C interface at
10087 *  a specified device address.
10088 **/
10089s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
10090		       u8 dev_addr, u8 data)
10091{
10092	struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
10093	struct i2c_client *this_client = adapter->i2c_client;
10094	s32 status;
10095	u16 swfw_mask = E1000_SWFW_PHY0_SM;
10096
10097	if (!this_client)
10098		return E1000_ERR_I2C;
10099
10100	if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
10101		return E1000_ERR_SWFW_SYNC;
10102	status = i2c_smbus_write_byte_data(this_client, byte_offset, data);
10103	hw->mac.ops.release_swfw_sync(hw, swfw_mask);
10104
10105	if (status)
10106		return E1000_ERR_I2C;
10107	else
10108		return 0;
10109
10110}
10111
10112int igb_reinit_queues(struct igb_adapter *adapter)
10113{
10114	struct net_device *netdev = adapter->netdev;
10115	struct pci_dev *pdev = adapter->pdev;
10116	int err = 0;
10117
10118	if (netif_running(netdev))
10119		igb_close(netdev);
10120
10121	igb_reset_interrupt_capability(adapter);
10122
10123	if (igb_init_interrupt_scheme(adapter, true)) {
10124		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
10125		return -ENOMEM;
10126	}
10127
10128	if (netif_running(netdev))
10129		err = igb_open(netdev);
10130
10131	return err;
10132}
10133
10134static void igb_nfc_filter_exit(struct igb_adapter *adapter)
10135{
10136	struct igb_nfc_filter *rule;
10137
10138	spin_lock(&adapter->nfc_lock);
10139
10140	hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
10141		igb_erase_filter(adapter, rule);
10142
10143	hlist_for_each_entry(rule, &adapter->cls_flower_list, nfc_node)
10144		igb_erase_filter(adapter, rule);
10145
10146	spin_unlock(&adapter->nfc_lock);
10147}
10148
10149static void igb_nfc_filter_restore(struct igb_adapter *adapter)
10150{
10151	struct igb_nfc_filter *rule;
10152
10153	spin_lock(&adapter->nfc_lock);
10154
10155	hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
10156		igb_add_filter(adapter, rule);
10157
10158	spin_unlock(&adapter->nfc_lock);
10159}
10160/* igb_main.c */
v3.15
   1/*******************************************************************************
   2
   3  Intel(R) Gigabit Ethernet Linux driver
   4  Copyright(c) 2007-2014 Intel Corporation.
   5
   6  This program is free software; you can redistribute it and/or modify it
   7  under the terms and conditions of the GNU General Public License,
   8  version 2, as published by the Free Software Foundation.
   9
  10  This program is distributed in the hope it will be useful, but WITHOUT
  11  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13  more details.
  14
  15  You should have received a copy of the GNU General Public License along with
  16  this program; if not, see <http://www.gnu.org/licenses/>.
  17
  18  The full GNU General Public License is included in this distribution in
  19  the file called "COPYING".
  20
  21  Contact Information:
  22  e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  23  Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  24
  25*******************************************************************************/
  26
  27#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  28
  29#include <linux/module.h>
  30#include <linux/types.h>
  31#include <linux/init.h>
  32#include <linux/bitops.h>
  33#include <linux/vmalloc.h>
  34#include <linux/pagemap.h>
  35#include <linux/netdevice.h>
  36#include <linux/ipv6.h>
  37#include <linux/slab.h>
  38#include <net/checksum.h>
  39#include <net/ip6_checksum.h>
 
 
  40#include <linux/net_tstamp.h>
  41#include <linux/mii.h>
  42#include <linux/ethtool.h>
  43#include <linux/if.h>
  44#include <linux/if_vlan.h>
  45#include <linux/pci.h>
  46#include <linux/pci-aspm.h>
  47#include <linux/delay.h>
  48#include <linux/interrupt.h>
  49#include <linux/ip.h>
  50#include <linux/tcp.h>
  51#include <linux/sctp.h>
  52#include <linux/if_ether.h>
  53#include <linux/aer.h>
  54#include <linux/prefetch.h>
 
 
  55#include <linux/pm_runtime.h>
 
  56#ifdef CONFIG_IGB_DCA
  57#include <linux/dca.h>
  58#endif
  59#include <linux/i2c.h>
  60#include "igb.h"
  61
  62#define MAJ 5
  63#define MIN 0
  64#define BUILD 5
  65#define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
  66__stringify(BUILD) "-k"
 
 
 
 
 
  67char igb_driver_name[] = "igb";
  68char igb_driver_version[] = DRV_VERSION;
  69static const char igb_driver_string[] =
  70				"Intel(R) Gigabit Ethernet Network Driver";
  71static const char igb_copyright[] =
  72				"Copyright (c) 2007-2014 Intel Corporation.";
  73
  74static const struct e1000_info *igb_info_tbl[] = {
  75	[board_82575] = &e1000_82575_info,
  76};
  77
  78static DEFINE_PCI_DEVICE_TABLE(igb_pci_tbl) = {
  79	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_1GBPS) },
  80	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_SGMII) },
  81	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) },
  82	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I211_COPPER), board_82575 },
  83	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 },
  84	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 },
  85	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 },
  86	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 },
  87	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER_FLASHLESS), board_82575 },
  88	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES_FLASHLESS), board_82575 },
  89	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
  90	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
  91	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
  92	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
  93	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
  94	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
  95	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 },
  96	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
  97	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
  98	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
  99	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 },
 100	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 },
 101	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 },
 102	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 },
 103	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
 104	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
 105	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
 106	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
 107	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
 108	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
 109	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
 110	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
 111	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
 112	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
 113	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
 114	/* required last entry */
 115	{0, }
 116};
 117
 118MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
 119
 120void igb_reset(struct igb_adapter *);
 121static int igb_setup_all_tx_resources(struct igb_adapter *);
 122static int igb_setup_all_rx_resources(struct igb_adapter *);
 123static void igb_free_all_tx_resources(struct igb_adapter *);
 124static void igb_free_all_rx_resources(struct igb_adapter *);
 125static void igb_setup_mrqc(struct igb_adapter *);
 126static int igb_probe(struct pci_dev *, const struct pci_device_id *);
 127static void igb_remove(struct pci_dev *pdev);
 
 128static int igb_sw_init(struct igb_adapter *);
 129static int igb_open(struct net_device *);
 130static int igb_close(struct net_device *);
 131static void igb_configure(struct igb_adapter *);
 132static void igb_configure_tx(struct igb_adapter *);
 133static void igb_configure_rx(struct igb_adapter *);
 134static void igb_clean_all_tx_rings(struct igb_adapter *);
 135static void igb_clean_all_rx_rings(struct igb_adapter *);
 136static void igb_clean_tx_ring(struct igb_ring *);
 137static void igb_clean_rx_ring(struct igb_ring *);
 138static void igb_set_rx_mode(struct net_device *);
 139static void igb_update_phy_info(unsigned long);
 140static void igb_watchdog(unsigned long);
 141static void igb_watchdog_task(struct work_struct *);
 142static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *);
 143static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *dev,
 144						 struct rtnl_link_stats64 *stats);
 145static int igb_change_mtu(struct net_device *, int);
 146static int igb_set_mac(struct net_device *, void *);
 147static void igb_set_uta(struct igb_adapter *adapter);
 148static irqreturn_t igb_intr(int irq, void *);
 149static irqreturn_t igb_intr_msi(int irq, void *);
 150static irqreturn_t igb_msix_other(int irq, void *);
 151static irqreturn_t igb_msix_ring(int irq, void *);
 152#ifdef CONFIG_IGB_DCA
 153static void igb_update_dca(struct igb_q_vector *);
 154static void igb_setup_dca(struct igb_adapter *);
 155#endif /* CONFIG_IGB_DCA */
 156static int igb_poll(struct napi_struct *, int);
 157static bool igb_clean_tx_irq(struct igb_q_vector *);
 158static bool igb_clean_rx_irq(struct igb_q_vector *, int);
 159static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
 160static void igb_tx_timeout(struct net_device *);
 161static void igb_reset_task(struct work_struct *);
 162static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features);
 
 163static int igb_vlan_rx_add_vid(struct net_device *, __be16, u16);
 164static int igb_vlan_rx_kill_vid(struct net_device *, __be16, u16);
 165static void igb_restore_vlan(struct igb_adapter *);
 166static void igb_rar_set_qsel(struct igb_adapter *, u8 *, u32 , u8);
 167static void igb_ping_all_vfs(struct igb_adapter *);
 168static void igb_msg_task(struct igb_adapter *);
 169static void igb_vmm_control(struct igb_adapter *);
 170static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
 
 
 
 
 
 171static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
 172static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
 173static int igb_ndo_set_vf_vlan(struct net_device *netdev,
 174			       int vf, u16 vlan, u8 qos);
 175static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate);
 176static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
 177				   bool setting);
 
 
 178static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
 179				 struct ifla_vf_info *ivi);
 180static void igb_check_vf_rate_limit(struct igb_adapter *);
 
 
 181
 182#ifdef CONFIG_PCI_IOV
 183static int igb_vf_configure(struct igb_adapter *adapter, int vf);
 184static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs);
 185#endif
 186
 187#ifdef CONFIG_PM
 188#ifdef CONFIG_PM_SLEEP
 189static int igb_suspend(struct device *);
 190#endif
 191static int igb_resume(struct device *);
 192#ifdef CONFIG_PM_RUNTIME
 193static int igb_runtime_suspend(struct device *dev);
 194static int igb_runtime_resume(struct device *dev);
 195static int igb_runtime_idle(struct device *dev);
 196#endif
 197static const struct dev_pm_ops igb_pm_ops = {
 198	SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume)
 199	SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume,
 200			igb_runtime_idle)
 201};
 202#endif
 203static void igb_shutdown(struct pci_dev *);
 204static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs);
 205#ifdef CONFIG_IGB_DCA
 206static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
 207static struct notifier_block dca_notifier = {
 208	.notifier_call	= igb_notify_dca,
 209	.next		= NULL,
 210	.priority	= 0
 211};
 212#endif
 213#ifdef CONFIG_NET_POLL_CONTROLLER
 214/* for netdump / net console */
 215static void igb_netpoll(struct net_device *);
 216#endif
 217#ifdef CONFIG_PCI_IOV
 218static unsigned int max_vfs = 0;
 219module_param(max_vfs, uint, 0);
 220MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
 221                 "per physical function");
 222#endif /* CONFIG_PCI_IOV */
 223
 224static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
 225		     pci_channel_state_t);
 226static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
 227static void igb_io_resume(struct pci_dev *);
 228
 229static const struct pci_error_handlers igb_err_handler = {
 230	.error_detected = igb_io_error_detected,
 231	.slot_reset = igb_io_slot_reset,
 232	.resume = igb_io_resume,
 233};
 234
 235static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
 236
 237static struct pci_driver igb_driver = {
 238	.name     = igb_driver_name,
 239	.id_table = igb_pci_tbl,
 240	.probe    = igb_probe,
 241	.remove   = igb_remove,
 242#ifdef CONFIG_PM
 243	.driver.pm = &igb_pm_ops,
 244#endif
 245	.shutdown = igb_shutdown,
 246	.sriov_configure = igb_pci_sriov_configure,
 247	.err_handler = &igb_err_handler
 248};
 249
 250MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
 251MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
 252MODULE_LICENSE("GPL");
 253MODULE_VERSION(DRV_VERSION);
 254
 255#define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
 256static int debug = -1;
 257module_param(debug, int, 0);
 258MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
 259
 260struct igb_reg_info {
 261	u32 ofs;
 262	char *name;
 263};
 264
 265static const struct igb_reg_info igb_reg_info_tbl[] = {
 266
 267	/* General Registers */
 268	{E1000_CTRL, "CTRL"},
 269	{E1000_STATUS, "STATUS"},
 270	{E1000_CTRL_EXT, "CTRL_EXT"},
 271
 272	/* Interrupt Registers */
 273	{E1000_ICR, "ICR"},
 274
 275	/* RX Registers */
 276	{E1000_RCTL, "RCTL"},
 277	{E1000_RDLEN(0), "RDLEN"},
 278	{E1000_RDH(0), "RDH"},
 279	{E1000_RDT(0), "RDT"},
 280	{E1000_RXDCTL(0), "RXDCTL"},
 281	{E1000_RDBAL(0), "RDBAL"},
 282	{E1000_RDBAH(0), "RDBAH"},
 283
 284	/* TX Registers */
 285	{E1000_TCTL, "TCTL"},
 286	{E1000_TDBAL(0), "TDBAL"},
 287	{E1000_TDBAH(0), "TDBAH"},
 288	{E1000_TDLEN(0), "TDLEN"},
 289	{E1000_TDH(0), "TDH"},
 290	{E1000_TDT(0), "TDT"},
 291	{E1000_TXDCTL(0), "TXDCTL"},
 292	{E1000_TDFH, "TDFH"},
 293	{E1000_TDFT, "TDFT"},
 294	{E1000_TDFHS, "TDFHS"},
 295	{E1000_TDFPC, "TDFPC"},
 296
 297	/* List Terminator */
 298	{}
 299};
 300
 301/* igb_regdump - register printout routine */
 302static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo)
 303{
 304	int n = 0;
 305	char rname[16];
 306	u32 regs[8];
 307
 308	switch (reginfo->ofs) {
 309	case E1000_RDLEN(0):
 310		for (n = 0; n < 4; n++)
 311			regs[n] = rd32(E1000_RDLEN(n));
 312		break;
 313	case E1000_RDH(0):
 314		for (n = 0; n < 4; n++)
 315			regs[n] = rd32(E1000_RDH(n));
 316		break;
 317	case E1000_RDT(0):
 318		for (n = 0; n < 4; n++)
 319			regs[n] = rd32(E1000_RDT(n));
 320		break;
 321	case E1000_RXDCTL(0):
 322		for (n = 0; n < 4; n++)
 323			regs[n] = rd32(E1000_RXDCTL(n));
 324		break;
 325	case E1000_RDBAL(0):
 326		for (n = 0; n < 4; n++)
 327			regs[n] = rd32(E1000_RDBAL(n));
 328		break;
 329	case E1000_RDBAH(0):
 330		for (n = 0; n < 4; n++)
 331			regs[n] = rd32(E1000_RDBAH(n));
 332		break;
 333	case E1000_TDBAL(0):
 334		for (n = 0; n < 4; n++)
 335			regs[n] = rd32(E1000_RDBAL(n));
 336		break;
 337	case E1000_TDBAH(0):
 338		for (n = 0; n < 4; n++)
 339			regs[n] = rd32(E1000_TDBAH(n));
 340		break;
 341	case E1000_TDLEN(0):
 342		for (n = 0; n < 4; n++)
 343			regs[n] = rd32(E1000_TDLEN(n));
 344		break;
 345	case E1000_TDH(0):
 346		for (n = 0; n < 4; n++)
 347			regs[n] = rd32(E1000_TDH(n));
 348		break;
 349	case E1000_TDT(0):
 350		for (n = 0; n < 4; n++)
 351			regs[n] = rd32(E1000_TDT(n));
 352		break;
 353	case E1000_TXDCTL(0):
 354		for (n = 0; n < 4; n++)
 355			regs[n] = rd32(E1000_TXDCTL(n));
 356		break;
 357	default:
 358		pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs));
 359		return;
 360	}
 361
 362	snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]");
 363	pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1],
 364		regs[2], regs[3]);
 365}
 366
 367/* igb_dump - Print registers, Tx-rings and Rx-rings */
 368static void igb_dump(struct igb_adapter *adapter)
 369{
 370	struct net_device *netdev = adapter->netdev;
 371	struct e1000_hw *hw = &adapter->hw;
 372	struct igb_reg_info *reginfo;
 373	struct igb_ring *tx_ring;
 374	union e1000_adv_tx_desc *tx_desc;
 375	struct my_u0 { u64 a; u64 b; } *u0;
 376	struct igb_ring *rx_ring;
 377	union e1000_adv_rx_desc *rx_desc;
 378	u32 staterr;
 379	u16 i, n;
 380
 381	if (!netif_msg_hw(adapter))
 382		return;
 383
 384	/* Print netdevice Info */
 385	if (netdev) {
 386		dev_info(&adapter->pdev->dev, "Net device Info\n");
 387		pr_info("Device Name     state            trans_start      "
 388			"last_rx\n");
 389		pr_info("%-15s %016lX %016lX %016lX\n", netdev->name,
 390			netdev->state, netdev->trans_start, netdev->last_rx);
 391	}
 392
 393	/* Print Registers */
 394	dev_info(&adapter->pdev->dev, "Register Dump\n");
 395	pr_info(" Register Name   Value\n");
 396	for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
 397	     reginfo->name; reginfo++) {
 398		igb_regdump(hw, reginfo);
 399	}
 400
 401	/* Print TX Ring Summary */
 402	if (!netdev || !netif_running(netdev))
 403		goto exit;
 404
 405	dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
 406	pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
 407	for (n = 0; n < adapter->num_tx_queues; n++) {
 408		struct igb_tx_buffer *buffer_info;
 409		tx_ring = adapter->tx_ring[n];
 410		buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
 411		pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
 412			n, tx_ring->next_to_use, tx_ring->next_to_clean,
 413			(u64)dma_unmap_addr(buffer_info, dma),
 414			dma_unmap_len(buffer_info, len),
 415			buffer_info->next_to_watch,
 416			(u64)buffer_info->time_stamp);
 417	}
 418
 419	/* Print TX Rings */
 420	if (!netif_msg_tx_done(adapter))
 421		goto rx_ring_summary;
 422
 423	dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
 424
 425	/* Transmit Descriptor Formats
 426	 *
 427	 * Advanced Transmit Descriptor
 428	 *   +--------------------------------------------------------------+
 429	 * 0 |         Buffer Address [63:0]                                |
 430	 *   +--------------------------------------------------------------+
 431	 * 8 | PAYLEN  | PORTS  |CC|IDX | STA | DCMD  |DTYP|MAC|RSV| DTALEN |
 432	 *   +--------------------------------------------------------------+
 433	 *   63      46 45    40 39 38 36 35 32 31   24             15       0
 434	 */
 435
 436	for (n = 0; n < adapter->num_tx_queues; n++) {
 437		tx_ring = adapter->tx_ring[n];
 438		pr_info("------------------------------------\n");
 439		pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
 440		pr_info("------------------------------------\n");
 441		pr_info("T [desc]     [address 63:0  ] [PlPOCIStDDM Ln] "
 442			"[bi->dma       ] leng  ntw timestamp        "
 443			"bi->skb\n");
 444
 445		for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 446			const char *next_desc;
 447			struct igb_tx_buffer *buffer_info;
 448			tx_desc = IGB_TX_DESC(tx_ring, i);
 449			buffer_info = &tx_ring->tx_buffer_info[i];
 450			u0 = (struct my_u0 *)tx_desc;
 451			if (i == tx_ring->next_to_use &&
 452			    i == tx_ring->next_to_clean)
 453				next_desc = " NTC/U";
 454			else if (i == tx_ring->next_to_use)
 455				next_desc = " NTU";
 456			else if (i == tx_ring->next_to_clean)
 457				next_desc = " NTC";
 458			else
 459				next_desc = "";
 460
 461			pr_info("T [0x%03X]    %016llX %016llX %016llX"
 462				" %04X  %p %016llX %p%s\n", i,
 463				le64_to_cpu(u0->a),
 464				le64_to_cpu(u0->b),
 465				(u64)dma_unmap_addr(buffer_info, dma),
 466				dma_unmap_len(buffer_info, len),
 467				buffer_info->next_to_watch,
 468				(u64)buffer_info->time_stamp,
 469				buffer_info->skb, next_desc);
 470
 471			if (netif_msg_pktdata(adapter) && buffer_info->skb)
 472				print_hex_dump(KERN_INFO, "",
 473					DUMP_PREFIX_ADDRESS,
 474					16, 1, buffer_info->skb->data,
 475					dma_unmap_len(buffer_info, len),
 476					true);
 477		}
 478	}
 479
 480	/* Print RX Rings Summary */
 481rx_ring_summary:
 482	dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
 483	pr_info("Queue [NTU] [NTC]\n");
 484	for (n = 0; n < adapter->num_rx_queues; n++) {
 485		rx_ring = adapter->rx_ring[n];
 486		pr_info(" %5d %5X %5X\n",
 487			n, rx_ring->next_to_use, rx_ring->next_to_clean);
 488	}
 489
 490	/* Print RX Rings */
 491	if (!netif_msg_rx_status(adapter))
 492		goto exit;
 493
 494	dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
 495
 496	/* Advanced Receive Descriptor (Read) Format
 497	 *    63                                           1        0
 498	 *    +-----------------------------------------------------+
 499	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
 500	 *    +----------------------------------------------+------+
 501	 *  8 |       Header Buffer Address [63:1]           |  DD  |
 502	 *    +-----------------------------------------------------+
 503	 *
 504	 *
 505	 * Advanced Receive Descriptor (Write-Back) Format
 506	 *
 507	 *   63       48 47    32 31  30      21 20 17 16   4 3     0
 508	 *   +------------------------------------------------------+
 509	 * 0 | Packet     IP     |SPH| HDR_LEN   | RSV|Packet|  RSS |
 510	 *   | Checksum   Ident  |   |           |    | Type | Type |
 511	 *   +------------------------------------------------------+
 512	 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 513	 *   +------------------------------------------------------+
 514	 *   63       48 47    32 31            20 19               0
 515	 */
 516
 517	for (n = 0; n < adapter->num_rx_queues; n++) {
 518		rx_ring = adapter->rx_ring[n];
 519		pr_info("------------------------------------\n");
 520		pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
 521		pr_info("------------------------------------\n");
 522		pr_info("R  [desc]      [ PktBuf     A0] [  HeadBuf   DD] "
 523			"[bi->dma       ] [bi->skb] <-- Adv Rx Read format\n");
 524		pr_info("RWB[desc]      [PcsmIpSHl PtRs] [vl er S cks ln] -----"
 525			"----------- [bi->skb] <-- Adv Rx Write-Back format\n");
 526
 527		for (i = 0; i < rx_ring->count; i++) {
 528			const char *next_desc;
 529			struct igb_rx_buffer *buffer_info;
 530			buffer_info = &rx_ring->rx_buffer_info[i];
 531			rx_desc = IGB_RX_DESC(rx_ring, i);
 532			u0 = (struct my_u0 *)rx_desc;
 533			staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 534
 535			if (i == rx_ring->next_to_use)
 536				next_desc = " NTU";
 537			else if (i == rx_ring->next_to_clean)
 538				next_desc = " NTC";
 539			else
 540				next_desc = "";
 541
 542			if (staterr & E1000_RXD_STAT_DD) {
 543				/* Descriptor Done */
 544				pr_info("%s[0x%03X]     %016llX %016llX ---------------- %s\n",
 545					"RWB", i,
 546					le64_to_cpu(u0->a),
 547					le64_to_cpu(u0->b),
 548					next_desc);
 549			} else {
 550				pr_info("%s[0x%03X]     %016llX %016llX %016llX %s\n",
 551					"R  ", i,
 552					le64_to_cpu(u0->a),
 553					le64_to_cpu(u0->b),
 554					(u64)buffer_info->dma,
 555					next_desc);
 556
 557				if (netif_msg_pktdata(adapter) &&
 558				    buffer_info->dma && buffer_info->page) {
 559					print_hex_dump(KERN_INFO, "",
 560					  DUMP_PREFIX_ADDRESS,
 561					  16, 1,
 562					  page_address(buffer_info->page) +
 563						      buffer_info->page_offset,
 564					  IGB_RX_BUFSZ, true);
 565				}
 566			}
 567		}
 568	}
 569
 570exit:
 571	return;
 572}
 573
 574/**
 575 *  igb_get_i2c_data - Reads the I2C SDA data bit
 576 *  @hw: pointer to hardware structure
 577 *  @i2cctl: Current value of I2CCTL register
 578 *
 579 *  Returns the I2C data bit value
 580 **/
 581static int igb_get_i2c_data(void *data)
 582{
 583	struct igb_adapter *adapter = (struct igb_adapter *)data;
 584	struct e1000_hw *hw = &adapter->hw;
 585	s32 i2cctl = rd32(E1000_I2CPARAMS);
 586
 587	return ((i2cctl & E1000_I2C_DATA_IN) != 0);
 588}
 589
 590/**
 591 *  igb_set_i2c_data - Sets the I2C data bit
 592 *  @data: pointer to hardware structure
 593 *  @state: I2C data value (0 or 1) to set
 594 *
 595 *  Sets the I2C data bit
 596 **/
 597static void igb_set_i2c_data(void *data, int state)
 598{
 599	struct igb_adapter *adapter = (struct igb_adapter *)data;
 600	struct e1000_hw *hw = &adapter->hw;
 601	s32 i2cctl = rd32(E1000_I2CPARAMS);
 602
 603	if (state)
 604		i2cctl |= E1000_I2C_DATA_OUT;
 605	else
 
 606		i2cctl &= ~E1000_I2C_DATA_OUT;
 
 607
 608	i2cctl &= ~E1000_I2C_DATA_OE_N;
 609	i2cctl |= E1000_I2C_CLK_OE_N;
 610	wr32(E1000_I2CPARAMS, i2cctl);
 611	wrfl();
 612
 613}
 614
 615/**
 616 *  igb_set_i2c_clk - Sets the I2C SCL clock
 617 *  @data: pointer to hardware structure
 618 *  @state: state to set clock
 619 *
 620 *  Sets the I2C clock line to state
 621 **/
 622static void igb_set_i2c_clk(void *data, int state)
 623{
 624	struct igb_adapter *adapter = (struct igb_adapter *)data;
 625	struct e1000_hw *hw = &adapter->hw;
 626	s32 i2cctl = rd32(E1000_I2CPARAMS);
 627
 628	if (state) {
 629		i2cctl |= E1000_I2C_CLK_OUT;
 630		i2cctl &= ~E1000_I2C_CLK_OE_N;
 631	} else {
 632		i2cctl &= ~E1000_I2C_CLK_OUT;
 633		i2cctl &= ~E1000_I2C_CLK_OE_N;
 634	}
 635	wr32(E1000_I2CPARAMS, i2cctl);
 636	wrfl();
 637}
 638
 639/**
 640 *  igb_get_i2c_clk - Gets the I2C SCL clock state
 641 *  @data: pointer to hardware structure
 642 *
 643 *  Gets the I2C clock state
 644 **/
 645static int igb_get_i2c_clk(void *data)
 646{
 647	struct igb_adapter *adapter = (struct igb_adapter *)data;
 648	struct e1000_hw *hw = &adapter->hw;
 649	s32 i2cctl = rd32(E1000_I2CPARAMS);
 650
 651	return ((i2cctl & E1000_I2C_CLK_IN) != 0);
 652}
 653
 654static const struct i2c_algo_bit_data igb_i2c_algo = {
 655	.setsda		= igb_set_i2c_data,
 656	.setscl		= igb_set_i2c_clk,
 657	.getsda		= igb_get_i2c_data,
 658	.getscl		= igb_get_i2c_clk,
 659	.udelay		= 5,
 660	.timeout	= 20,
 661};
 662
 663/**
 664 *  igb_get_hw_dev - return device
 665 *  @hw: pointer to hardware structure
 666 *
 667 *  used by hardware layer to print debugging information
 668 **/
 669struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
 670{
 671	struct igb_adapter *adapter = hw->back;
 672	return adapter->netdev;
 673}
 674
 675/**
 676 *  igb_init_module - Driver Registration Routine
 677 *
 678 *  igb_init_module is the first routine called when the driver is
 679 *  loaded. All it does is register with the PCI subsystem.
 680 **/
 681static int __init igb_init_module(void)
 682{
 683	int ret;
 684	pr_info("%s - version %s\n",
 685	       igb_driver_string, igb_driver_version);
 686
 
 687	pr_info("%s\n", igb_copyright);
 688
 689#ifdef CONFIG_IGB_DCA
 690	dca_register_notify(&dca_notifier);
 691#endif
 692	ret = pci_register_driver(&igb_driver);
 693	return ret;
 694}
 695
 696module_init(igb_init_module);
 697
 698/**
 699 *  igb_exit_module - Driver Exit Cleanup Routine
 700 *
 701 *  igb_exit_module is called just before the driver is removed
 702 *  from memory.
 703 **/
 704static void __exit igb_exit_module(void)
 705{
 706#ifdef CONFIG_IGB_DCA
 707	dca_unregister_notify(&dca_notifier);
 708#endif
 709	pci_unregister_driver(&igb_driver);
 710}
 711
 712module_exit(igb_exit_module);
 713
 714#define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
 715/**
 716 *  igb_cache_ring_register - Descriptor ring to register mapping
 717 *  @adapter: board private structure to initialize
 718 *
 719 *  Once we know the feature-set enabled for the device, we'll cache
 720 *  the register offset the descriptor ring is assigned to.
 721 **/
 722static void igb_cache_ring_register(struct igb_adapter *adapter)
 723{
 724	int i = 0, j = 0;
 725	u32 rbase_offset = adapter->vfs_allocated_count;
 726
 727	switch (adapter->hw.mac.type) {
 728	case e1000_82576:
 729		/* The queues are allocated for virtualization such that VF 0
 730		 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
 731		 * In order to avoid collision we start at the first free queue
 732		 * and continue consuming queues in the same sequence
 733		 */
 734		if (adapter->vfs_allocated_count) {
 735			for (; i < adapter->rss_queues; i++)
 736				adapter->rx_ring[i]->reg_idx = rbase_offset +
 737							       Q_IDX_82576(i);
 738		}
 
 739	case e1000_82575:
 740	case e1000_82580:
 741	case e1000_i350:
 742	case e1000_i354:
 743	case e1000_i210:
 744	case e1000_i211:
 745	default:
 746		for (; i < adapter->num_rx_queues; i++)
 747			adapter->rx_ring[i]->reg_idx = rbase_offset + i;
 748		for (; j < adapter->num_tx_queues; j++)
 749			adapter->tx_ring[j]->reg_idx = rbase_offset + j;
 750		break;
 751	}
 752}
 753
 754u32 igb_rd32(struct e1000_hw *hw, u32 reg)
 755{
 756	struct igb_adapter *igb = container_of(hw, struct igb_adapter, hw);
 757	u8 __iomem *hw_addr = ACCESS_ONCE(hw->hw_addr);
 758	u32 value = 0;
 759
 760	if (E1000_REMOVED(hw_addr))
 761		return ~value;
 762
 763	value = readl(&hw_addr[reg]);
 764
 765	/* reads should not return all F's */
 766	if (!(~value) && (!reg || !(~readl(hw_addr)))) {
 767		struct net_device *netdev = igb->netdev;
 768		hw->hw_addr = NULL;
 769		netif_device_detach(netdev);
 770		netdev_err(netdev, "PCIe link lost, device now detached\n");
 
 771	}
 772
 773	return value;
 774}
 775
 776/**
 777 *  igb_write_ivar - configure ivar for given MSI-X vector
 778 *  @hw: pointer to the HW structure
 779 *  @msix_vector: vector number we are allocating to a given ring
 780 *  @index: row index of IVAR register to write within IVAR table
 781 *  @offset: column offset of in IVAR, should be multiple of 8
 782 *
 783 *  This function is intended to handle the writing of the IVAR register
 784 *  for adapters 82576 and newer.  The IVAR table consists of 2 columns,
 785 *  each containing an cause allocation for an Rx and Tx ring, and a
 786 *  variable number of rows depending on the number of queues supported.
 787 **/
 788static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
 789			   int index, int offset)
 790{
 791	u32 ivar = array_rd32(E1000_IVAR0, index);
 792
 793	/* clear any bits that are currently set */
 794	ivar &= ~((u32)0xFF << offset);
 795
 796	/* write vector and valid bit */
 797	ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
 798
 799	array_wr32(E1000_IVAR0, index, ivar);
 800}
 801
 802#define IGB_N0_QUEUE -1
 803static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
 804{
 805	struct igb_adapter *adapter = q_vector->adapter;
 806	struct e1000_hw *hw = &adapter->hw;
 807	int rx_queue = IGB_N0_QUEUE;
 808	int tx_queue = IGB_N0_QUEUE;
 809	u32 msixbm = 0;
 810
 811	if (q_vector->rx.ring)
 812		rx_queue = q_vector->rx.ring->reg_idx;
 813	if (q_vector->tx.ring)
 814		tx_queue = q_vector->tx.ring->reg_idx;
 815
 816	switch (hw->mac.type) {
 817	case e1000_82575:
 818		/* The 82575 assigns vectors using a bitmask, which matches the
 819		 * bitmask for the EICR/EIMS/EIMC registers.  To assign one
 820		 * or more queues to a vector, we write the appropriate bits
 821		 * into the MSIXBM register for that vector.
 822		 */
 823		if (rx_queue > IGB_N0_QUEUE)
 824			msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
 825		if (tx_queue > IGB_N0_QUEUE)
 826			msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
 827		if (!(adapter->flags & IGB_FLAG_HAS_MSIX) && msix_vector == 0)
 828			msixbm |= E1000_EIMS_OTHER;
 829		array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
 830		q_vector->eims_value = msixbm;
 831		break;
 832	case e1000_82576:
 833		/* 82576 uses a table that essentially consists of 2 columns
 834		 * with 8 rows.  The ordering is column-major so we use the
 835		 * lower 3 bits as the row index, and the 4th bit as the
 836		 * column offset.
 837		 */
 838		if (rx_queue > IGB_N0_QUEUE)
 839			igb_write_ivar(hw, msix_vector,
 840				       rx_queue & 0x7,
 841				       (rx_queue & 0x8) << 1);
 842		if (tx_queue > IGB_N0_QUEUE)
 843			igb_write_ivar(hw, msix_vector,
 844				       tx_queue & 0x7,
 845				       ((tx_queue & 0x8) << 1) + 8);
 846		q_vector->eims_value = 1 << msix_vector;
 847		break;
 848	case e1000_82580:
 849	case e1000_i350:
 850	case e1000_i354:
 851	case e1000_i210:
 852	case e1000_i211:
 853		/* On 82580 and newer adapters the scheme is similar to 82576
 854		 * however instead of ordering column-major we have things
 855		 * ordered row-major.  So we traverse the table by using
 856		 * bit 0 as the column offset, and the remaining bits as the
 857		 * row index.
 858		 */
 859		if (rx_queue > IGB_N0_QUEUE)
 860			igb_write_ivar(hw, msix_vector,
 861				       rx_queue >> 1,
 862				       (rx_queue & 0x1) << 4);
 863		if (tx_queue > IGB_N0_QUEUE)
 864			igb_write_ivar(hw, msix_vector,
 865				       tx_queue >> 1,
 866				       ((tx_queue & 0x1) << 4) + 8);
 867		q_vector->eims_value = 1 << msix_vector;
 868		break;
 869	default:
 870		BUG();
 871		break;
 872	}
 873
 874	/* add q_vector eims value to global eims_enable_mask */
 875	adapter->eims_enable_mask |= q_vector->eims_value;
 876
 877	/* configure q_vector to set itr on first interrupt */
 878	q_vector->set_itr = 1;
 879}
 880
 881/**
 882 *  igb_configure_msix - Configure MSI-X hardware
 883 *  @adapter: board private structure to initialize
 884 *
 885 *  igb_configure_msix sets up the hardware to properly
 886 *  generate MSI-X interrupts.
 887 **/
 888static void igb_configure_msix(struct igb_adapter *adapter)
 889{
 890	u32 tmp;
 891	int i, vector = 0;
 892	struct e1000_hw *hw = &adapter->hw;
 893
 894	adapter->eims_enable_mask = 0;
 895
 896	/* set vector for other causes, i.e. link changes */
 897	switch (hw->mac.type) {
 898	case e1000_82575:
 899		tmp = rd32(E1000_CTRL_EXT);
 900		/* enable MSI-X PBA support*/
 901		tmp |= E1000_CTRL_EXT_PBA_CLR;
 902
 903		/* Auto-Mask interrupts upon ICR read. */
 904		tmp |= E1000_CTRL_EXT_EIAME;
 905		tmp |= E1000_CTRL_EXT_IRCA;
 906
 907		wr32(E1000_CTRL_EXT, tmp);
 908
 909		/* enable msix_other interrupt */
 910		array_wr32(E1000_MSIXBM(0), vector++, E1000_EIMS_OTHER);
 911		adapter->eims_other = E1000_EIMS_OTHER;
 912
 913		break;
 914
 915	case e1000_82576:
 916	case e1000_82580:
 917	case e1000_i350:
 918	case e1000_i354:
 919	case e1000_i210:
 920	case e1000_i211:
 921		/* Turn on MSI-X capability first, or our settings
 922		 * won't stick.  And it will take days to debug.
 923		 */
 924		wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
 925		     E1000_GPIE_PBA | E1000_GPIE_EIAME |
 926		     E1000_GPIE_NSICR);
 927
 928		/* enable msix_other interrupt */
 929		adapter->eims_other = 1 << vector;
 930		tmp = (vector++ | E1000_IVAR_VALID) << 8;
 931
 932		wr32(E1000_IVAR_MISC, tmp);
 933		break;
 934	default:
 935		/* do nothing, since nothing else supports MSI-X */
 936		break;
 937	} /* switch (hw->mac.type) */
 938
 939	adapter->eims_enable_mask |= adapter->eims_other;
 940
 941	for (i = 0; i < adapter->num_q_vectors; i++)
 942		igb_assign_vector(adapter->q_vector[i], vector++);
 943
 944	wrfl();
 945}
 946
 947/**
 948 *  igb_request_msix - Initialize MSI-X interrupts
 949 *  @adapter: board private structure to initialize
 950 *
 951 *  igb_request_msix allocates MSI-X vectors and requests interrupts from the
 952 *  kernel.
 953 **/
 954static int igb_request_msix(struct igb_adapter *adapter)
 955{
 
 956	struct net_device *netdev = adapter->netdev;
 957	struct e1000_hw *hw = &adapter->hw;
 958	int i, err = 0, vector = 0, free_vector = 0;
 959
 960	err = request_irq(adapter->msix_entries[vector].vector,
 961			  igb_msix_other, 0, netdev->name, adapter);
 962	if (err)
 963		goto err_out;
 964
 965	for (i = 0; i < adapter->num_q_vectors; i++) {
 
 
 
 
 
 
 966		struct igb_q_vector *q_vector = adapter->q_vector[i];
 967
 968		vector++;
 969
 970		q_vector->itr_register = hw->hw_addr + E1000_EITR(vector);
 971
 972		if (q_vector->rx.ring && q_vector->tx.ring)
 973			sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
 974				q_vector->rx.ring->queue_index);
 975		else if (q_vector->tx.ring)
 976			sprintf(q_vector->name, "%s-tx-%u", netdev->name,
 977				q_vector->tx.ring->queue_index);
 978		else if (q_vector->rx.ring)
 979			sprintf(q_vector->name, "%s-rx-%u", netdev->name,
 980				q_vector->rx.ring->queue_index);
 981		else
 982			sprintf(q_vector->name, "%s-unused", netdev->name);
 983
 984		err = request_irq(adapter->msix_entries[vector].vector,
 985				  igb_msix_ring, 0, q_vector->name,
 986				  q_vector);
 987		if (err)
 988			goto err_free;
 989	}
 990
 991	igb_configure_msix(adapter);
 992	return 0;
 993
 994err_free:
 995	/* free already assigned IRQs */
 996	free_irq(adapter->msix_entries[free_vector++].vector, adapter);
 997
 998	vector--;
 999	for (i = 0; i < vector; i++) {
1000		free_irq(adapter->msix_entries[free_vector++].vector,
1001			 adapter->q_vector[i]);
1002	}
1003err_out:
1004	return err;
1005}
1006
1007/**
1008 *  igb_free_q_vector - Free memory allocated for specific interrupt vector
1009 *  @adapter: board private structure to initialize
1010 *  @v_idx: Index of vector to be freed
1011 *
1012 *  This function frees the memory allocated to the q_vector.
1013 **/
1014static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx)
1015{
1016	struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1017
1018	adapter->q_vector[v_idx] = NULL;
1019
1020	/* igb_get_stats64() might access the rings on this vector,
1021	 * we must wait a grace period before freeing it.
1022	 */
1023	kfree_rcu(q_vector, rcu);
 
1024}
1025
1026/**
1027 *  igb_reset_q_vector - Reset config for interrupt vector
1028 *  @adapter: board private structure to initialize
1029 *  @v_idx: Index of vector to be reset
1030 *
1031 *  If NAPI is enabled it will delete any references to the
1032 *  NAPI struct. This is preparation for igb_free_q_vector.
1033 **/
1034static void igb_reset_q_vector(struct igb_adapter *adapter, int v_idx)
1035{
1036	struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1037
1038	/* Coming from igb_set_interrupt_capability, the vectors are not yet
1039	 * allocated. So, q_vector is NULL so we should stop here.
1040	 */
1041	if (!q_vector)
1042		return;
1043
1044	if (q_vector->tx.ring)
1045		adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
1046
1047	if (q_vector->rx.ring)
1048		adapter->tx_ring[q_vector->rx.ring->queue_index] = NULL;
1049
1050	netif_napi_del(&q_vector->napi);
1051
1052}
1053
1054static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
1055{
1056	int v_idx = adapter->num_q_vectors;
1057
1058	if (adapter->flags & IGB_FLAG_HAS_MSIX)
1059		pci_disable_msix(adapter->pdev);
1060	else if (adapter->flags & IGB_FLAG_HAS_MSI)
1061		pci_disable_msi(adapter->pdev);
1062
1063	while (v_idx--)
1064		igb_reset_q_vector(adapter, v_idx);
1065}
1066
1067/**
1068 *  igb_free_q_vectors - Free memory allocated for interrupt vectors
1069 *  @adapter: board private structure to initialize
1070 *
1071 *  This function frees the memory allocated to the q_vectors.  In addition if
1072 *  NAPI is enabled it will delete any references to the NAPI struct prior
1073 *  to freeing the q_vector.
1074 **/
1075static void igb_free_q_vectors(struct igb_adapter *adapter)
1076{
1077	int v_idx = adapter->num_q_vectors;
1078
1079	adapter->num_tx_queues = 0;
1080	adapter->num_rx_queues = 0;
1081	adapter->num_q_vectors = 0;
1082
1083	while (v_idx--) {
1084		igb_reset_q_vector(adapter, v_idx);
1085		igb_free_q_vector(adapter, v_idx);
1086	}
1087}
1088
1089/**
1090 *  igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1091 *  @adapter: board private structure to initialize
1092 *
1093 *  This function resets the device so that it has 0 Rx queues, Tx queues, and
1094 *  MSI-X interrupts allocated.
1095 */
1096static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
1097{
1098	igb_free_q_vectors(adapter);
1099	igb_reset_interrupt_capability(adapter);
1100}
1101
1102/**
1103 *  igb_set_interrupt_capability - set MSI or MSI-X if supported
1104 *  @adapter: board private structure to initialize
1105 *  @msix: boolean value of MSIX capability
1106 *
1107 *  Attempt to configure interrupts using the best available
1108 *  capabilities of the hardware and kernel.
1109 **/
1110static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix)
1111{
1112	int err;
1113	int numvecs, i;
1114
1115	if (!msix)
1116		goto msi_only;
1117	adapter->flags |= IGB_FLAG_HAS_MSIX;
1118
1119	/* Number of supported queues. */
1120	adapter->num_rx_queues = adapter->rss_queues;
1121	if (adapter->vfs_allocated_count)
1122		adapter->num_tx_queues = 1;
1123	else
1124		adapter->num_tx_queues = adapter->rss_queues;
1125
1126	/* start with one vector for every Rx queue */
1127	numvecs = adapter->num_rx_queues;
1128
1129	/* if Tx handler is separate add 1 for every Tx queue */
1130	if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1131		numvecs += adapter->num_tx_queues;
1132
1133	/* store the number of vectors reserved for queues */
1134	adapter->num_q_vectors = numvecs;
1135
1136	/* add 1 vector for link status interrupts */
1137	numvecs++;
1138	for (i = 0; i < numvecs; i++)
1139		adapter->msix_entries[i].entry = i;
1140
1141	err = pci_enable_msix_range(adapter->pdev,
1142				    adapter->msix_entries,
1143				    numvecs,
1144				    numvecs);
1145	if (err > 0)
1146		return;
1147
1148	igb_reset_interrupt_capability(adapter);
1149
1150	/* If we can't do MSI-X, try MSI */
1151msi_only:
1152	adapter->flags &= ~IGB_FLAG_HAS_MSIX;
1153#ifdef CONFIG_PCI_IOV
1154	/* disable SR-IOV for non MSI-X configurations */
1155	if (adapter->vf_data) {
1156		struct e1000_hw *hw = &adapter->hw;
1157		/* disable iov and allow time for transactions to clear */
1158		pci_disable_sriov(adapter->pdev);
1159		msleep(500);
1160
 
 
1161		kfree(adapter->vf_data);
1162		adapter->vf_data = NULL;
1163		wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1164		wrfl();
1165		msleep(100);
1166		dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1167	}
1168#endif
1169	adapter->vfs_allocated_count = 0;
1170	adapter->rss_queues = 1;
1171	adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1172	adapter->num_rx_queues = 1;
1173	adapter->num_tx_queues = 1;
1174	adapter->num_q_vectors = 1;
1175	if (!pci_enable_msi(adapter->pdev))
1176		adapter->flags |= IGB_FLAG_HAS_MSI;
1177}
1178
1179static void igb_add_ring(struct igb_ring *ring,
1180			 struct igb_ring_container *head)
1181{
1182	head->ring = ring;
1183	head->count++;
1184}
1185
1186/**
1187 *  igb_alloc_q_vector - Allocate memory for a single interrupt vector
1188 *  @adapter: board private structure to initialize
1189 *  @v_count: q_vectors allocated on adapter, used for ring interleaving
1190 *  @v_idx: index of vector in adapter struct
1191 *  @txr_count: total number of Tx rings to allocate
1192 *  @txr_idx: index of first Tx ring to allocate
1193 *  @rxr_count: total number of Rx rings to allocate
1194 *  @rxr_idx: index of first Rx ring to allocate
1195 *
1196 *  We allocate one q_vector.  If allocation fails we return -ENOMEM.
1197 **/
1198static int igb_alloc_q_vector(struct igb_adapter *adapter,
1199			      int v_count, int v_idx,
1200			      int txr_count, int txr_idx,
1201			      int rxr_count, int rxr_idx)
1202{
1203	struct igb_q_vector *q_vector;
1204	struct igb_ring *ring;
1205	int ring_count, size;
 
1206
1207	/* igb only supports 1 Tx and/or 1 Rx queue per vector */
1208	if (txr_count > 1 || rxr_count > 1)
1209		return -ENOMEM;
1210
1211	ring_count = txr_count + rxr_count;
1212	size = sizeof(struct igb_q_vector) +
1213	       (sizeof(struct igb_ring) * ring_count);
1214
1215	/* allocate q_vector and rings */
1216	q_vector = adapter->q_vector[v_idx];
1217	if (!q_vector)
1218		q_vector = kzalloc(size, GFP_KERNEL);
 
 
 
 
 
 
 
 
 
 
1219	if (!q_vector)
1220		return -ENOMEM;
1221
1222	/* initialize NAPI */
1223	netif_napi_add(adapter->netdev, &q_vector->napi,
1224		       igb_poll, 64);
1225
1226	/* tie q_vector and adapter together */
1227	adapter->q_vector[v_idx] = q_vector;
1228	q_vector->adapter = adapter;
1229
1230	/* initialize work limits */
1231	q_vector->tx.work_limit = adapter->tx_work_limit;
1232
1233	/* initialize ITR configuration */
1234	q_vector->itr_register = adapter->hw.hw_addr + E1000_EITR(0);
1235	q_vector->itr_val = IGB_START_ITR;
1236
1237	/* initialize pointer to rings */
1238	ring = q_vector->ring;
1239
1240	/* intialize ITR */
1241	if (rxr_count) {
1242		/* rx or rx/tx vector */
1243		if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
1244			q_vector->itr_val = adapter->rx_itr_setting;
1245	} else {
1246		/* tx only vector */
1247		if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
1248			q_vector->itr_val = adapter->tx_itr_setting;
1249	}
1250
1251	if (txr_count) {
1252		/* assign generic ring traits */
1253		ring->dev = &adapter->pdev->dev;
1254		ring->netdev = adapter->netdev;
1255
1256		/* configure backlink on ring */
1257		ring->q_vector = q_vector;
1258
1259		/* update q_vector Tx values */
1260		igb_add_ring(ring, &q_vector->tx);
1261
1262		/* For 82575, context index must be unique per ring. */
1263		if (adapter->hw.mac.type == e1000_82575)
1264			set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
1265
1266		/* apply Tx specific ring traits */
1267		ring->count = adapter->tx_ring_count;
1268		ring->queue_index = txr_idx;
1269
 
 
 
 
 
 
1270		u64_stats_init(&ring->tx_syncp);
1271		u64_stats_init(&ring->tx_syncp2);
1272
1273		/* assign ring to adapter */
1274		adapter->tx_ring[txr_idx] = ring;
1275
1276		/* push pointer to next ring */
1277		ring++;
1278	}
1279
1280	if (rxr_count) {
1281		/* assign generic ring traits */
1282		ring->dev = &adapter->pdev->dev;
1283		ring->netdev = adapter->netdev;
1284
1285		/* configure backlink on ring */
1286		ring->q_vector = q_vector;
1287
1288		/* update q_vector Rx values */
1289		igb_add_ring(ring, &q_vector->rx);
1290
1291		/* set flag indicating ring supports SCTP checksum offload */
1292		if (adapter->hw.mac.type >= e1000_82576)
1293			set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
1294
1295		/*
1296		 * On i350, i354, i210, and i211, loopback VLAN packets
1297		 * have the tag byte-swapped.
1298		 */
1299		if (adapter->hw.mac.type >= e1000_i350)
1300			set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
1301
1302		/* apply Rx specific ring traits */
1303		ring->count = adapter->rx_ring_count;
1304		ring->queue_index = rxr_idx;
1305
1306		u64_stats_init(&ring->rx_syncp);
1307
1308		/* assign ring to adapter */
1309		adapter->rx_ring[rxr_idx] = ring;
1310	}
1311
1312	return 0;
1313}
1314
1315
1316/**
1317 *  igb_alloc_q_vectors - Allocate memory for interrupt vectors
1318 *  @adapter: board private structure to initialize
1319 *
1320 *  We allocate one q_vector per queue interrupt.  If allocation fails we
1321 *  return -ENOMEM.
1322 **/
1323static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1324{
1325	int q_vectors = adapter->num_q_vectors;
1326	int rxr_remaining = adapter->num_rx_queues;
1327	int txr_remaining = adapter->num_tx_queues;
1328	int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1329	int err;
1330
1331	if (q_vectors >= (rxr_remaining + txr_remaining)) {
1332		for (; rxr_remaining; v_idx++) {
1333			err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1334						 0, 0, 1, rxr_idx);
1335
1336			if (err)
1337				goto err_out;
1338
1339			/* update counts and index */
1340			rxr_remaining--;
1341			rxr_idx++;
1342		}
1343	}
1344
1345	for (; v_idx < q_vectors; v_idx++) {
1346		int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1347		int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
 
1348		err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1349					 tqpv, txr_idx, rqpv, rxr_idx);
1350
1351		if (err)
1352			goto err_out;
1353
1354		/* update counts and index */
1355		rxr_remaining -= rqpv;
1356		txr_remaining -= tqpv;
1357		rxr_idx++;
1358		txr_idx++;
1359	}
1360
1361	return 0;
1362
1363err_out:
1364	adapter->num_tx_queues = 0;
1365	adapter->num_rx_queues = 0;
1366	adapter->num_q_vectors = 0;
1367
1368	while (v_idx--)
1369		igb_free_q_vector(adapter, v_idx);
1370
1371	return -ENOMEM;
1372}
1373
1374/**
1375 *  igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1376 *  @adapter: board private structure to initialize
1377 *  @msix: boolean value of MSIX capability
1378 *
1379 *  This function initializes the interrupts and allocates all of the queues.
1380 **/
1381static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
1382{
1383	struct pci_dev *pdev = adapter->pdev;
1384	int err;
1385
1386	igb_set_interrupt_capability(adapter, msix);
1387
1388	err = igb_alloc_q_vectors(adapter);
1389	if (err) {
1390		dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1391		goto err_alloc_q_vectors;
1392	}
1393
1394	igb_cache_ring_register(adapter);
1395
1396	return 0;
1397
1398err_alloc_q_vectors:
1399	igb_reset_interrupt_capability(adapter);
1400	return err;
1401}
1402
1403/**
1404 *  igb_request_irq - initialize interrupts
1405 *  @adapter: board private structure to initialize
1406 *
1407 *  Attempts to configure interrupts using the best available
1408 *  capabilities of the hardware and kernel.
1409 **/
1410static int igb_request_irq(struct igb_adapter *adapter)
1411{
1412	struct net_device *netdev = adapter->netdev;
1413	struct pci_dev *pdev = adapter->pdev;
1414	int err = 0;
1415
1416	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1417		err = igb_request_msix(adapter);
1418		if (!err)
1419			goto request_done;
1420		/* fall back to MSI */
1421		igb_free_all_tx_resources(adapter);
1422		igb_free_all_rx_resources(adapter);
1423
1424		igb_clear_interrupt_scheme(adapter);
1425		err = igb_init_interrupt_scheme(adapter, false);
1426		if (err)
1427			goto request_done;
1428
1429		igb_setup_all_tx_resources(adapter);
1430		igb_setup_all_rx_resources(adapter);
1431		igb_configure(adapter);
1432	}
1433
1434	igb_assign_vector(adapter->q_vector[0], 0);
1435
1436	if (adapter->flags & IGB_FLAG_HAS_MSI) {
1437		err = request_irq(pdev->irq, igb_intr_msi, 0,
1438				  netdev->name, adapter);
1439		if (!err)
1440			goto request_done;
1441
1442		/* fall back to legacy interrupts */
1443		igb_reset_interrupt_capability(adapter);
1444		adapter->flags &= ~IGB_FLAG_HAS_MSI;
1445	}
1446
1447	err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1448			  netdev->name, adapter);
1449
1450	if (err)
1451		dev_err(&pdev->dev, "Error %d getting interrupt\n",
1452			err);
1453
1454request_done:
1455	return err;
1456}
1457
1458static void igb_free_irq(struct igb_adapter *adapter)
1459{
1460	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1461		int vector = 0, i;
1462
1463		free_irq(adapter->msix_entries[vector++].vector, adapter);
1464
1465		for (i = 0; i < adapter->num_q_vectors; i++)
1466			free_irq(adapter->msix_entries[vector++].vector,
1467				 adapter->q_vector[i]);
1468	} else {
1469		free_irq(adapter->pdev->irq, adapter);
1470	}
1471}
1472
1473/**
1474 *  igb_irq_disable - Mask off interrupt generation on the NIC
1475 *  @adapter: board private structure
1476 **/
1477static void igb_irq_disable(struct igb_adapter *adapter)
1478{
1479	struct e1000_hw *hw = &adapter->hw;
1480
1481	/* we need to be careful when disabling interrupts.  The VFs are also
1482	 * mapped into these registers and so clearing the bits can cause
1483	 * issues on the VF drivers so we only need to clear what we set
1484	 */
1485	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1486		u32 regval = rd32(E1000_EIAM);
 
1487		wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1488		wr32(E1000_EIMC, adapter->eims_enable_mask);
1489		regval = rd32(E1000_EIAC);
1490		wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1491	}
1492
1493	wr32(E1000_IAM, 0);
1494	wr32(E1000_IMC, ~0);
1495	wrfl();
1496	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1497		int i;
 
1498		for (i = 0; i < adapter->num_q_vectors; i++)
1499			synchronize_irq(adapter->msix_entries[i].vector);
1500	} else {
1501		synchronize_irq(adapter->pdev->irq);
1502	}
1503}
1504
1505/**
1506 *  igb_irq_enable - Enable default interrupt generation settings
1507 *  @adapter: board private structure
1508 **/
1509static void igb_irq_enable(struct igb_adapter *adapter)
1510{
1511	struct e1000_hw *hw = &adapter->hw;
1512
1513	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1514		u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1515		u32 regval = rd32(E1000_EIAC);
 
1516		wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1517		regval = rd32(E1000_EIAM);
1518		wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1519		wr32(E1000_EIMS, adapter->eims_enable_mask);
1520		if (adapter->vfs_allocated_count) {
1521			wr32(E1000_MBVFIMR, 0xFF);
1522			ims |= E1000_IMS_VMMB;
1523		}
1524		wr32(E1000_IMS, ims);
1525	} else {
1526		wr32(E1000_IMS, IMS_ENABLE_MASK |
1527				E1000_IMS_DRSTA);
1528		wr32(E1000_IAM, IMS_ENABLE_MASK |
1529				E1000_IMS_DRSTA);
1530	}
1531}
1532
1533static void igb_update_mng_vlan(struct igb_adapter *adapter)
1534{
1535	struct e1000_hw *hw = &adapter->hw;
 
1536	u16 vid = adapter->hw.mng_cookie.vlan_id;
1537	u16 old_vid = adapter->mng_vlan_id;
1538
1539	if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1540		/* add VID to filter table */
1541		igb_vfta_set(hw, vid, true);
1542		adapter->mng_vlan_id = vid;
1543	} else {
1544		adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1545	}
1546
1547	if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1548	    (vid != old_vid) &&
1549	    !test_bit(old_vid, adapter->active_vlans)) {
1550		/* remove VID from filter table */
1551		igb_vfta_set(hw, old_vid, false);
1552	}
1553}
1554
1555/**
1556 *  igb_release_hw_control - release control of the h/w to f/w
1557 *  @adapter: address of board private structure
1558 *
1559 *  igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1560 *  For ASF and Pass Through versions of f/w this means that the
1561 *  driver is no longer loaded.
1562 **/
1563static void igb_release_hw_control(struct igb_adapter *adapter)
1564{
1565	struct e1000_hw *hw = &adapter->hw;
1566	u32 ctrl_ext;
1567
1568	/* Let firmware take over control of h/w */
1569	ctrl_ext = rd32(E1000_CTRL_EXT);
1570	wr32(E1000_CTRL_EXT,
1571			ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1572}
1573
1574/**
1575 *  igb_get_hw_control - get control of the h/w from f/w
1576 *  @adapter: address of board private structure
1577 *
1578 *  igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1579 *  For ASF and Pass Through versions of f/w this means that
1580 *  the driver is loaded.
1581 **/
1582static void igb_get_hw_control(struct igb_adapter *adapter)
1583{
1584	struct e1000_hw *hw = &adapter->hw;
1585	u32 ctrl_ext;
1586
1587	/* Let firmware know the driver has taken over */
1588	ctrl_ext = rd32(E1000_CTRL_EXT);
1589	wr32(E1000_CTRL_EXT,
1590			ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1591}
1592
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1593/**
1594 *  igb_configure - configure the hardware for RX and TX
1595 *  @adapter: private board structure
1596 **/
1597static void igb_configure(struct igb_adapter *adapter)
1598{
1599	struct net_device *netdev = adapter->netdev;
1600	int i;
1601
1602	igb_get_hw_control(adapter);
1603	igb_set_rx_mode(netdev);
 
1604
1605	igb_restore_vlan(adapter);
1606
1607	igb_setup_tctl(adapter);
1608	igb_setup_mrqc(adapter);
1609	igb_setup_rctl(adapter);
1610
 
1611	igb_configure_tx(adapter);
1612	igb_configure_rx(adapter);
1613
1614	igb_rx_fifo_flush_82575(&adapter->hw);
1615
1616	/* call igb_desc_unused which always leaves
1617	 * at least 1 descriptor unused to make sure
1618	 * next_to_use != next_to_clean
1619	 */
1620	for (i = 0; i < adapter->num_rx_queues; i++) {
1621		struct igb_ring *ring = adapter->rx_ring[i];
1622		igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
1623	}
1624}
1625
1626/**
1627 *  igb_power_up_link - Power up the phy/serdes link
1628 *  @adapter: address of board private structure
1629 **/
1630void igb_power_up_link(struct igb_adapter *adapter)
1631{
1632	igb_reset_phy(&adapter->hw);
1633
1634	if (adapter->hw.phy.media_type == e1000_media_type_copper)
1635		igb_power_up_phy_copper(&adapter->hw);
1636	else
1637		igb_power_up_serdes_link_82575(&adapter->hw);
 
 
1638}
1639
1640/**
1641 *  igb_power_down_link - Power down the phy/serdes link
1642 *  @adapter: address of board private structure
1643 */
1644static void igb_power_down_link(struct igb_adapter *adapter)
1645{
1646	if (adapter->hw.phy.media_type == e1000_media_type_copper)
1647		igb_power_down_phy_copper_82575(&adapter->hw);
1648	else
1649		igb_shutdown_serdes_link_82575(&adapter->hw);
1650}
1651
1652/**
1653 * Detect and switch function for Media Auto Sense
1654 * @adapter: address of the board private structure
1655 **/
1656static void igb_check_swap_media(struct igb_adapter *adapter)
1657{
1658	struct e1000_hw *hw = &adapter->hw;
1659	u32 ctrl_ext, connsw;
1660	bool swap_now = false;
1661
1662	ctrl_ext = rd32(E1000_CTRL_EXT);
1663	connsw = rd32(E1000_CONNSW);
1664
1665	/* need to live swap if current media is copper and we have fiber/serdes
1666	 * to go to.
1667	 */
1668
1669	if ((hw->phy.media_type == e1000_media_type_copper) &&
1670	    (!(connsw & E1000_CONNSW_AUTOSENSE_EN))) {
1671		swap_now = true;
1672	} else if (!(connsw & E1000_CONNSW_SERDESD)) {
 
1673		/* copper signal takes time to appear */
1674		if (adapter->copper_tries < 4) {
1675			adapter->copper_tries++;
1676			connsw |= E1000_CONNSW_AUTOSENSE_CONF;
1677			wr32(E1000_CONNSW, connsw);
1678			return;
1679		} else {
1680			adapter->copper_tries = 0;
1681			if ((connsw & E1000_CONNSW_PHYSD) &&
1682			    (!(connsw & E1000_CONNSW_PHY_PDN))) {
1683				swap_now = true;
1684				connsw &= ~E1000_CONNSW_AUTOSENSE_CONF;
1685				wr32(E1000_CONNSW, connsw);
1686			}
1687		}
1688	}
1689
1690	if (!swap_now)
1691		return;
1692
1693	switch (hw->phy.media_type) {
1694	case e1000_media_type_copper:
1695		netdev_info(adapter->netdev,
1696			"MAS: changing media to fiber/serdes\n");
1697		ctrl_ext |=
1698			E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
1699		adapter->flags |= IGB_FLAG_MEDIA_RESET;
1700		adapter->copper_tries = 0;
1701		break;
1702	case e1000_media_type_internal_serdes:
1703	case e1000_media_type_fiber:
1704		netdev_info(adapter->netdev,
1705			"MAS: changing media to copper\n");
1706		ctrl_ext &=
1707			~E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
1708		adapter->flags |= IGB_FLAG_MEDIA_RESET;
1709		break;
1710	default:
1711		/* shouldn't get here during regular operation */
1712		netdev_err(adapter->netdev,
1713			"AMS: Invalid media type found, returning\n");
1714		break;
1715	}
1716	wr32(E1000_CTRL_EXT, ctrl_ext);
1717}
1718
1719/**
1720 *  igb_up - Open the interface and prepare it to handle traffic
1721 *  @adapter: board private structure
1722 **/
1723int igb_up(struct igb_adapter *adapter)
1724{
1725	struct e1000_hw *hw = &adapter->hw;
1726	int i;
1727
1728	/* hardware has been reset, we need to reload some things */
1729	igb_configure(adapter);
1730
1731	clear_bit(__IGB_DOWN, &adapter->state);
1732
1733	for (i = 0; i < adapter->num_q_vectors; i++)
1734		napi_enable(&(adapter->q_vector[i]->napi));
1735
1736	if (adapter->flags & IGB_FLAG_HAS_MSIX)
1737		igb_configure_msix(adapter);
1738	else
1739		igb_assign_vector(adapter->q_vector[0], 0);
1740
1741	/* Clear any pending interrupts. */
 
1742	rd32(E1000_ICR);
1743	igb_irq_enable(adapter);
1744
1745	/* notify VFs that reset has been completed */
1746	if (adapter->vfs_allocated_count) {
1747		u32 reg_data = rd32(E1000_CTRL_EXT);
 
1748		reg_data |= E1000_CTRL_EXT_PFRSTD;
1749		wr32(E1000_CTRL_EXT, reg_data);
1750	}
1751
1752	netif_tx_start_all_queues(adapter->netdev);
1753
1754	/* start the watchdog. */
1755	hw->mac.get_link_status = 1;
1756	schedule_work(&adapter->watchdog_task);
1757
1758	if ((adapter->flags & IGB_FLAG_EEE) &&
1759	    (!hw->dev_spec._82575.eee_disable))
1760		adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
1761
1762	return 0;
1763}
1764
1765void igb_down(struct igb_adapter *adapter)
1766{
1767	struct net_device *netdev = adapter->netdev;
1768	struct e1000_hw *hw = &adapter->hw;
1769	u32 tctl, rctl;
1770	int i;
1771
1772	/* signal that we're down so the interrupt handler does not
1773	 * reschedule our watchdog timer
1774	 */
1775	set_bit(__IGB_DOWN, &adapter->state);
1776
1777	/* disable receives in the hardware */
1778	rctl = rd32(E1000_RCTL);
1779	wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1780	/* flush and sleep below */
1781
 
 
 
1782	netif_tx_stop_all_queues(netdev);
1783
1784	/* disable transmits in the hardware */
1785	tctl = rd32(E1000_TCTL);
1786	tctl &= ~E1000_TCTL_EN;
1787	wr32(E1000_TCTL, tctl);
1788	/* flush both disables and wait for them to finish */
1789	wrfl();
1790	msleep(10);
1791
1792	igb_irq_disable(adapter);
1793
1794	adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
1795
1796	for (i = 0; i < adapter->num_q_vectors; i++) {
1797		napi_synchronize(&(adapter->q_vector[i]->napi));
1798		napi_disable(&(adapter->q_vector[i]->napi));
 
 
1799	}
1800
1801
1802	del_timer_sync(&adapter->watchdog_timer);
1803	del_timer_sync(&adapter->phy_info_timer);
1804
1805	netif_carrier_off(netdev);
1806
1807	/* record the stats before reset*/
1808	spin_lock(&adapter->stats64_lock);
1809	igb_update_stats(adapter, &adapter->stats64);
1810	spin_unlock(&adapter->stats64_lock);
1811
1812	adapter->link_speed = 0;
1813	adapter->link_duplex = 0;
1814
1815	if (!pci_channel_offline(adapter->pdev))
1816		igb_reset(adapter);
 
 
 
 
1817	igb_clean_all_tx_rings(adapter);
1818	igb_clean_all_rx_rings(adapter);
1819#ifdef CONFIG_IGB_DCA
1820
1821	/* since we reset the hardware DCA settings were cleared */
1822	igb_setup_dca(adapter);
1823#endif
1824}
1825
1826void igb_reinit_locked(struct igb_adapter *adapter)
1827{
1828	WARN_ON(in_interrupt());
1829	while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1830		msleep(1);
1831	igb_down(adapter);
1832	igb_up(adapter);
1833	clear_bit(__IGB_RESETTING, &adapter->state);
1834}
1835
1836/** igb_enable_mas - Media Autosense re-enable after swap
1837 *
1838 * @adapter: adapter struct
1839 **/
1840static s32 igb_enable_mas(struct igb_adapter *adapter)
1841{
1842	struct e1000_hw *hw = &adapter->hw;
1843	u32 connsw;
1844	s32 ret_val = 0;
1845
1846	connsw = rd32(E1000_CONNSW);
1847	if (!(hw->phy.media_type == e1000_media_type_copper))
1848		return ret_val;
1849
1850	/* configure for SerDes media detect */
1851	if (!(connsw & E1000_CONNSW_SERDESD)) {
 
1852		connsw |= E1000_CONNSW_ENRGSRC;
1853		connsw |= E1000_CONNSW_AUTOSENSE_EN;
1854		wr32(E1000_CONNSW, connsw);
1855		wrfl();
1856	} else if (connsw & E1000_CONNSW_SERDESD) {
1857		/* already SerDes, no need to enable anything */
1858		return ret_val;
1859	} else {
1860		netdev_info(adapter->netdev,
1861			"MAS: Unable to configure feature, disabling..\n");
1862		adapter->flags &= ~IGB_FLAG_MAS_ENABLE;
1863	}
1864	return ret_val;
1865}
1866
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1867void igb_reset(struct igb_adapter *adapter)
1868{
1869	struct pci_dev *pdev = adapter->pdev;
1870	struct e1000_hw *hw = &adapter->hw;
1871	struct e1000_mac_info *mac = &hw->mac;
1872	struct e1000_fc_info *fc = &hw->fc;
1873	u32 pba = 0, tx_space, min_tx_space, min_rx_space, hwm;
1874
1875	/* Repartition Pba for greater than 9k mtu
1876	 * To take effect CTRL.RST is required.
1877	 */
1878	switch (mac->type) {
1879	case e1000_i350:
1880	case e1000_i354:
1881	case e1000_82580:
1882		pba = rd32(E1000_RXPBS);
1883		pba = igb_rxpbs_adjust_82580(pba);
1884		break;
1885	case e1000_82576:
1886		pba = rd32(E1000_RXPBS);
1887		pba &= E1000_RXPBS_SIZE_MASK_82576;
1888		break;
1889	case e1000_82575:
1890	case e1000_i210:
1891	case e1000_i211:
1892	default:
1893		pba = E1000_PBA_34K;
1894		break;
1895	}
1896
1897	if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1898	    (mac->type < e1000_82576)) {
1899		/* adjust PBA for jumbo frames */
 
1900		wr32(E1000_PBA, pba);
1901
1902		/* To maintain wire speed transmits, the Tx FIFO should be
1903		 * large enough to accommodate two full transmit packets,
1904		 * rounded up to the next 1KB and expressed in KB.  Likewise,
1905		 * the Rx FIFO should be large enough to accommodate at least
1906		 * one full receive packet and is similarly rounded up and
1907		 * expressed in KB.
1908		 */
1909		pba = rd32(E1000_PBA);
 
 
 
 
 
 
 
 
 
 
1910		/* upper 16 bits has Tx packet buffer allocation size in KB */
1911		tx_space = pba >> 16;
1912		/* lower 16 bits has Rx packet buffer allocation size in KB */
1913		pba &= 0xffff;
1914		/* the Tx fifo also stores 16 bytes of information about the Tx
1915		 * but don't include ethernet FCS because hardware appends it
1916		 */
1917		min_tx_space = (adapter->max_frame_size +
1918				sizeof(union e1000_adv_tx_desc) -
1919				ETH_FCS_LEN) * 2;
1920		min_tx_space = ALIGN(min_tx_space, 1024);
1921		min_tx_space >>= 10;
1922		/* software strips receive CRC, so leave room for it */
1923		min_rx_space = adapter->max_frame_size;
1924		min_rx_space = ALIGN(min_rx_space, 1024);
1925		min_rx_space >>= 10;
1926
1927		/* If current Tx allocation is less than the min Tx FIFO size,
1928		 * and the min Tx FIFO size is less than the current Rx FIFO
1929		 * allocation, take space away from current Rx allocation
1930		 */
1931		if (tx_space < min_tx_space &&
1932		    ((min_tx_space - tx_space) < pba)) {
1933			pba = pba - (min_tx_space - tx_space);
1934
1935			/* if short on Rx space, Rx wins and must trump Tx
1936			 * adjustment
1937			 */
1938			if (pba < min_rx_space)
1939				pba = min_rx_space;
1940		}
 
 
1941		wr32(E1000_PBA, pba);
1942	}
1943
1944	/* flow control settings */
1945	/* The high water mark must be low enough to fit one full frame
1946	 * (or the size used for early receive) above it in the Rx FIFO.
1947	 * Set it to the lower of:
1948	 * - 90% of the Rx FIFO size, or
1949	 * - the full Rx FIFO size minus one full frame
 
1950	 */
1951	hwm = min(((pba << 10) * 9 / 10),
1952			((pba << 10) - 2 * adapter->max_frame_size));
1953
1954	fc->high_water = hwm & 0xFFFFFFF0;	/* 16-byte granularity */
1955	fc->low_water = fc->high_water - 16;
1956	fc->pause_time = 0xFFFF;
1957	fc->send_xon = 1;
1958	fc->current_mode = fc->requested_mode;
1959
1960	/* disable receive for all VFs and wait one second */
1961	if (adapter->vfs_allocated_count) {
1962		int i;
 
1963		for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1964			adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
1965
1966		/* ping all the active vfs to let them know we are going down */
1967		igb_ping_all_vfs(adapter);
1968
1969		/* disable transmits and receives */
1970		wr32(E1000_VFRE, 0);
1971		wr32(E1000_VFTE, 0);
1972	}
1973
1974	/* Allow time for pending master requests to run */
1975	hw->mac.ops.reset_hw(hw);
1976	wr32(E1000_WUC, 0);
1977
1978	if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
1979		/* need to resetup here after media swap */
1980		adapter->ei.get_invariants(hw);
1981		adapter->flags &= ~IGB_FLAG_MEDIA_RESET;
1982	}
1983	if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
1984		if (igb_enable_mas(adapter))
1985			dev_err(&pdev->dev,
1986				"Error enabling Media Auto Sense\n");
1987	}
1988	if (hw->mac.ops.init_hw(hw))
1989		dev_err(&pdev->dev, "Hardware Error\n");
1990
 
 
 
 
 
 
 
1991	/* Flow control settings reset on hardware reset, so guarantee flow
1992	 * control is off when forcing speed.
1993	 */
1994	if (!hw->mac.autoneg)
1995		igb_force_mac_fc(hw);
1996
1997	igb_init_dmac(adapter, pba);
1998#ifdef CONFIG_IGB_HWMON
1999	/* Re-initialize the thermal sensor on i350 devices. */
2000	if (!test_bit(__IGB_DOWN, &adapter->state)) {
2001		if (mac->type == e1000_i350 && hw->bus.func == 0) {
2002			/* If present, re-initialize the external thermal sensor
2003			 * interface.
2004			 */
2005			if (adapter->ets)
2006				mac->ops.init_thermal_sensor_thresh(hw);
 
2007		}
2008	}
2009#endif
2010	/* Re-establish EEE setting */
2011	if (hw->phy.media_type == e1000_media_type_copper) {
2012		switch (mac->type) {
2013		case e1000_i350:
2014		case e1000_i210:
2015		case e1000_i211:
2016			igb_set_eee_i350(hw);
2017			break;
2018		case e1000_i354:
2019			igb_set_eee_i354(hw);
2020			break;
2021		default:
2022			break;
2023		}
2024	}
2025	if (!netif_running(adapter->netdev))
2026		igb_power_down_link(adapter);
2027
2028	igb_update_mng_vlan(adapter);
2029
2030	/* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2031	wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
2032
2033	/* Re-enable PTP, where applicable. */
2034	igb_ptp_reset(adapter);
 
2035
2036	igb_get_phy_info(hw);
2037}
2038
2039static netdev_features_t igb_fix_features(struct net_device *netdev,
2040	netdev_features_t features)
2041{
2042	/* Since there is no support for separate Rx/Tx vlan accel
2043	 * enable/disable make sure Tx flag is always in same state as Rx.
2044	 */
2045	if (features & NETIF_F_HW_VLAN_CTAG_RX)
2046		features |= NETIF_F_HW_VLAN_CTAG_TX;
2047	else
2048		features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2049
2050	return features;
2051}
2052
2053static int igb_set_features(struct net_device *netdev,
2054	netdev_features_t features)
2055{
2056	netdev_features_t changed = netdev->features ^ features;
2057	struct igb_adapter *adapter = netdev_priv(netdev);
2058
2059	if (changed & NETIF_F_HW_VLAN_CTAG_RX)
2060		igb_vlan_mode(netdev, features);
2061
2062	if (!(changed & NETIF_F_RXALL))
2063		return 0;
2064
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2065	netdev->features = features;
2066
2067	if (netif_running(netdev))
2068		igb_reinit_locked(adapter);
2069	else
2070		igb_reset(adapter);
2071
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2072	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2073}
2074
2075static const struct net_device_ops igb_netdev_ops = {
2076	.ndo_open		= igb_open,
2077	.ndo_stop		= igb_close,
2078	.ndo_start_xmit		= igb_xmit_frame,
2079	.ndo_get_stats64	= igb_get_stats64,
2080	.ndo_set_rx_mode	= igb_set_rx_mode,
2081	.ndo_set_mac_address	= igb_set_mac,
2082	.ndo_change_mtu		= igb_change_mtu,
2083	.ndo_do_ioctl		= igb_ioctl,
2084	.ndo_tx_timeout		= igb_tx_timeout,
2085	.ndo_validate_addr	= eth_validate_addr,
2086	.ndo_vlan_rx_add_vid	= igb_vlan_rx_add_vid,
2087	.ndo_vlan_rx_kill_vid	= igb_vlan_rx_kill_vid,
2088	.ndo_set_vf_mac		= igb_ndo_set_vf_mac,
2089	.ndo_set_vf_vlan	= igb_ndo_set_vf_vlan,
2090	.ndo_set_vf_tx_rate	= igb_ndo_set_vf_bw,
2091	.ndo_set_vf_spoofchk	= igb_ndo_set_vf_spoofchk,
 
2092	.ndo_get_vf_config	= igb_ndo_get_vf_config,
2093#ifdef CONFIG_NET_POLL_CONTROLLER
2094	.ndo_poll_controller	= igb_netpoll,
2095#endif
2096	.ndo_fix_features	= igb_fix_features,
2097	.ndo_set_features	= igb_set_features,
 
 
 
 
 
2098};
2099
2100/**
2101 * igb_set_fw_version - Configure version string for ethtool
2102 * @adapter: adapter struct
2103 **/
2104void igb_set_fw_version(struct igb_adapter *adapter)
2105{
2106	struct e1000_hw *hw = &adapter->hw;
2107	struct e1000_fw_version fw;
2108
2109	igb_get_fw_version(hw, &fw);
2110
2111	switch (hw->mac.type) {
2112	case e1000_i210:
2113	case e1000_i211:
2114		if (!(igb_get_flash_presence_i210(hw))) {
2115			snprintf(adapter->fw_version,
2116				 sizeof(adapter->fw_version),
2117				 "%2d.%2d-%d",
2118				 fw.invm_major, fw.invm_minor,
2119				 fw.invm_img_type);
2120			break;
2121		}
2122		/* fall through */
2123	default:
2124		/* if option is rom valid, display its version too */
2125		if (fw.or_valid) {
2126			snprintf(adapter->fw_version,
2127				 sizeof(adapter->fw_version),
2128				 "%d.%d, 0x%08x, %d.%d.%d",
2129				 fw.eep_major, fw.eep_minor, fw.etrack_id,
2130				 fw.or_major, fw.or_build, fw.or_patch);
2131		/* no option rom */
2132		} else if (fw.etrack_id != 0X0000) {
2133			snprintf(adapter->fw_version,
2134			    sizeof(adapter->fw_version),
2135			    "%d.%d, 0x%08x",
2136			    fw.eep_major, fw.eep_minor, fw.etrack_id);
2137		} else {
2138		snprintf(adapter->fw_version,
2139		    sizeof(adapter->fw_version),
2140		    "%d.%d.%d",
2141		    fw.eep_major, fw.eep_minor, fw.eep_build);
2142		}
2143		break;
2144	}
2145	return;
2146}
2147
2148/**
2149 * igb_init_mas - init Media Autosense feature if enabled in the NVM
2150 *
2151 * @adapter: adapter struct
2152 **/
2153static void igb_init_mas(struct igb_adapter *adapter)
2154{
2155	struct e1000_hw *hw = &adapter->hw;
2156	u16 eeprom_data;
2157
2158	hw->nvm.ops.read(hw, NVM_COMPAT, 1, &eeprom_data);
2159	switch (hw->bus.func) {
2160	case E1000_FUNC_0:
2161		if (eeprom_data & IGB_MAS_ENABLE_0) {
2162			adapter->flags |= IGB_FLAG_MAS_ENABLE;
2163			netdev_info(adapter->netdev,
2164				"MAS: Enabling Media Autosense for port %d\n",
2165				hw->bus.func);
2166		}
2167		break;
2168	case E1000_FUNC_1:
2169		if (eeprom_data & IGB_MAS_ENABLE_1) {
2170			adapter->flags |= IGB_FLAG_MAS_ENABLE;
2171			netdev_info(adapter->netdev,
2172				"MAS: Enabling Media Autosense for port %d\n",
2173				hw->bus.func);
2174		}
2175		break;
2176	case E1000_FUNC_2:
2177		if (eeprom_data & IGB_MAS_ENABLE_2) {
2178			adapter->flags |= IGB_FLAG_MAS_ENABLE;
2179			netdev_info(adapter->netdev,
2180				"MAS: Enabling Media Autosense for port %d\n",
2181				hw->bus.func);
2182		}
2183		break;
2184	case E1000_FUNC_3:
2185		if (eeprom_data & IGB_MAS_ENABLE_3) {
2186			adapter->flags |= IGB_FLAG_MAS_ENABLE;
2187			netdev_info(adapter->netdev,
2188				"MAS: Enabling Media Autosense for port %d\n",
2189				hw->bus.func);
2190		}
2191		break;
2192	default:
2193		/* Shouldn't get here */
2194		netdev_err(adapter->netdev,
2195			"MAS: Invalid port configuration, returning\n");
2196		break;
2197	}
2198}
2199
2200/**
2201 *  igb_init_i2c - Init I2C interface
2202 *  @adapter: pointer to adapter structure
2203 **/
2204static s32 igb_init_i2c(struct igb_adapter *adapter)
2205{
2206	s32 status = E1000_SUCCESS;
2207
2208	/* I2C interface supported on i350 devices */
2209	if (adapter->hw.mac.type != e1000_i350)
2210		return E1000_SUCCESS;
2211
2212	/* Initialize the i2c bus which is controlled by the registers.
2213	 * This bus will use the i2c_algo_bit structue that implements
2214	 * the protocol through toggling of the 4 bits in the register.
2215	 */
2216	adapter->i2c_adap.owner = THIS_MODULE;
2217	adapter->i2c_algo = igb_i2c_algo;
2218	adapter->i2c_algo.data = adapter;
2219	adapter->i2c_adap.algo_data = &adapter->i2c_algo;
2220	adapter->i2c_adap.dev.parent = &adapter->pdev->dev;
2221	strlcpy(adapter->i2c_adap.name, "igb BB",
2222		sizeof(adapter->i2c_adap.name));
2223	status = i2c_bit_add_bus(&adapter->i2c_adap);
2224	return status;
2225}
2226
2227/**
2228 *  igb_probe - Device Initialization Routine
2229 *  @pdev: PCI device information struct
2230 *  @ent: entry in igb_pci_tbl
2231 *
2232 *  Returns 0 on success, negative on failure
2233 *
2234 *  igb_probe initializes an adapter identified by a pci_dev structure.
2235 *  The OS initialization, configuring of the adapter private structure,
2236 *  and a hardware reset occur.
2237 **/
2238static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2239{
2240	struct net_device *netdev;
2241	struct igb_adapter *adapter;
2242	struct e1000_hw *hw;
2243	u16 eeprom_data = 0;
2244	s32 ret_val;
2245	static int global_quad_port_a; /* global quad port a indication */
2246	const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
2247	int err, pci_using_dac;
2248	u8 part_str[E1000_PBANUM_LENGTH];
 
2249
2250	/* Catch broken hardware that put the wrong VF device ID in
2251	 * the PCIe SR-IOV capability.
2252	 */
2253	if (pdev->is_virtfn) {
2254		WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
2255			pci_name(pdev), pdev->vendor, pdev->device);
2256		return -EINVAL;
2257	}
2258
2259	err = pci_enable_device_mem(pdev);
2260	if (err)
2261		return err;
2262
2263	pci_using_dac = 0;
2264	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
2265	if (!err) {
2266		pci_using_dac = 1;
2267	} else {
2268		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
2269		if (err) {
2270			dev_err(&pdev->dev,
2271				"No usable DMA configuration, aborting\n");
2272			goto err_dma;
2273		}
2274	}
2275
2276	err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
2277					   IORESOURCE_MEM),
2278					   igb_driver_name);
2279	if (err)
2280		goto err_pci_reg;
2281
2282	pci_enable_pcie_error_reporting(pdev);
2283
2284	pci_set_master(pdev);
2285	pci_save_state(pdev);
2286
2287	err = -ENOMEM;
2288	netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
2289				   IGB_MAX_TX_QUEUES);
2290	if (!netdev)
2291		goto err_alloc_etherdev;
2292
2293	SET_NETDEV_DEV(netdev, &pdev->dev);
2294
2295	pci_set_drvdata(pdev, netdev);
2296	adapter = netdev_priv(netdev);
2297	adapter->netdev = netdev;
2298	adapter->pdev = pdev;
2299	hw = &adapter->hw;
2300	hw->back = adapter;
2301	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2302
2303	err = -EIO;
2304	hw->hw_addr = pci_iomap(pdev, 0, 0);
2305	if (!hw->hw_addr)
2306		goto err_ioremap;
 
 
2307
2308	netdev->netdev_ops = &igb_netdev_ops;
2309	igb_set_ethtool_ops(netdev);
2310	netdev->watchdog_timeo = 5 * HZ;
2311
2312	strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2313
2314	netdev->mem_start = pci_resource_start(pdev, 0);
2315	netdev->mem_end = pci_resource_end(pdev, 0);
2316
2317	/* PCI config space info */
2318	hw->vendor_id = pdev->vendor;
2319	hw->device_id = pdev->device;
2320	hw->revision_id = pdev->revision;
2321	hw->subsystem_vendor_id = pdev->subsystem_vendor;
2322	hw->subsystem_device_id = pdev->subsystem_device;
2323
2324	/* Copy the default MAC, PHY and NVM function pointers */
2325	memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
2326	memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
2327	memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
2328	/* Initialize skew-specific constants */
2329	err = ei->get_invariants(hw);
2330	if (err)
2331		goto err_sw_init;
2332
2333	/* setup the private structure */
2334	err = igb_sw_init(adapter);
2335	if (err)
2336		goto err_sw_init;
2337
2338	igb_get_bus_info_pcie(hw);
2339
2340	hw->phy.autoneg_wait_to_complete = false;
2341
2342	/* Copper options */
2343	if (hw->phy.media_type == e1000_media_type_copper) {
2344		hw->phy.mdix = AUTO_ALL_MODES;
2345		hw->phy.disable_polarity_correction = false;
2346		hw->phy.ms_type = e1000_ms_hw_default;
2347	}
2348
2349	if (igb_check_reset_block(hw))
2350		dev_info(&pdev->dev,
2351			"PHY reset is blocked due to SOL/IDER session.\n");
2352
2353	/* features is initialized to 0 in allocation, it might have bits
2354	 * set by igb_sw_init so we should use an or instead of an
2355	 * assignment.
2356	 */
2357	netdev->features |= NETIF_F_SG |
2358			    NETIF_F_IP_CSUM |
2359			    NETIF_F_IPV6_CSUM |
2360			    NETIF_F_TSO |
2361			    NETIF_F_TSO6 |
2362			    NETIF_F_RXHASH |
2363			    NETIF_F_RXCSUM |
2364			    NETIF_F_HW_VLAN_CTAG_RX |
2365			    NETIF_F_HW_VLAN_CTAG_TX;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2366
2367	/* copy netdev features into list of user selectable features */
2368	netdev->hw_features |= netdev->features;
2369	netdev->hw_features |= NETIF_F_RXALL;
 
 
2370
2371	/* set this bit last since it cannot be part of hw_features */
2372	netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
2373
2374	netdev->vlan_features |= NETIF_F_TSO |
2375				 NETIF_F_TSO6 |
2376				 NETIF_F_IP_CSUM |
2377				 NETIF_F_IPV6_CSUM |
2378				 NETIF_F_SG;
2379
2380	netdev->priv_flags |= IFF_SUPP_NOFCS;
 
 
2381
2382	if (pci_using_dac) {
2383		netdev->features |= NETIF_F_HIGHDMA;
2384		netdev->vlan_features |= NETIF_F_HIGHDMA;
2385	}
2386
2387	if (hw->mac.type >= e1000_82576) {
2388		netdev->hw_features |= NETIF_F_SCTP_CSUM;
2389		netdev->features |= NETIF_F_SCTP_CSUM;
2390	}
2391
2392	netdev->priv_flags |= IFF_UNICAST_FLT;
 
 
 
 
 
2393
2394	adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
2395
2396	/* before reading the NVM, reset the controller to put the device in a
2397	 * known good starting state
2398	 */
2399	hw->mac.ops.reset_hw(hw);
2400
2401	/* make sure the NVM is good , i211/i210 parts can have special NVM
2402	 * that doesn't contain a checksum
2403	 */
2404	switch (hw->mac.type) {
2405	case e1000_i210:
2406	case e1000_i211:
2407		if (igb_get_flash_presence_i210(hw)) {
2408			if (hw->nvm.ops.validate(hw) < 0) {
2409				dev_err(&pdev->dev,
2410					"The NVM Checksum Is Not Valid\n");
2411				err = -EIO;
2412				goto err_eeprom;
2413			}
2414		}
2415		break;
2416	default:
2417		if (hw->nvm.ops.validate(hw) < 0) {
2418			dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
2419			err = -EIO;
2420			goto err_eeprom;
2421		}
2422		break;
2423	}
2424
2425	/* copy the MAC address out of the NVM */
2426	if (hw->mac.ops.read_mac_addr(hw))
2427		dev_err(&pdev->dev, "NVM Read Error\n");
 
 
2428
2429	memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2430
2431	if (!is_valid_ether_addr(netdev->dev_addr)) {
2432		dev_err(&pdev->dev, "Invalid MAC Address\n");
2433		err = -EIO;
2434		goto err_eeprom;
2435	}
2436
 
 
2437	/* get firmware version for ethtool -i */
2438	igb_set_fw_version(adapter);
2439
2440	setup_timer(&adapter->watchdog_timer, igb_watchdog,
2441		    (unsigned long) adapter);
2442	setup_timer(&adapter->phy_info_timer, igb_update_phy_info,
2443		    (unsigned long) adapter);
 
 
 
 
2444
2445	INIT_WORK(&adapter->reset_task, igb_reset_task);
2446	INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
2447
2448	/* Initialize link properties that are user-changeable */
2449	adapter->fc_autoneg = true;
2450	hw->mac.autoneg = true;
2451	hw->phy.autoneg_advertised = 0x2f;
2452
2453	hw->fc.requested_mode = e1000_fc_default;
2454	hw->fc.current_mode = e1000_fc_default;
2455
2456	igb_validate_mdi_setting(hw);
2457
2458	/* By default, support wake on port A */
2459	if (hw->bus.func == 0)
2460		adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2461
2462	/* Check the NVM for wake support on non-port A ports */
2463	if (hw->mac.type >= e1000_82580)
2464		hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
2465				 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
2466				 &eeprom_data);
2467	else if (hw->bus.func == 1)
2468		hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
2469
2470	if (eeprom_data & IGB_EEPROM_APME)
2471		adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2472
2473	/* now that we have the eeprom settings, apply the special cases where
2474	 * the eeprom may be wrong or the board simply won't support wake on
2475	 * lan on a particular port
2476	 */
2477	switch (pdev->device) {
2478	case E1000_DEV_ID_82575GB_QUAD_COPPER:
2479		adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2480		break;
2481	case E1000_DEV_ID_82575EB_FIBER_SERDES:
2482	case E1000_DEV_ID_82576_FIBER:
2483	case E1000_DEV_ID_82576_SERDES:
2484		/* Wake events only supported on port A for dual fiber
2485		 * regardless of eeprom setting
2486		 */
2487		if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
2488			adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2489		break;
2490	case E1000_DEV_ID_82576_QUAD_COPPER:
2491	case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
2492		/* if quad port adapter, disable WoL on all but port A */
2493		if (global_quad_port_a != 0)
2494			adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2495		else
2496			adapter->flags |= IGB_FLAG_QUAD_PORT_A;
2497		/* Reset for multiple quad port adapters */
2498		if (++global_quad_port_a == 4)
2499			global_quad_port_a = 0;
2500		break;
2501	default:
2502		/* If the device can't wake, don't set software support */
2503		if (!device_can_wakeup(&adapter->pdev->dev))
2504			adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2505	}
2506
2507	/* initialize the wol settings based on the eeprom settings */
2508	if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
2509		adapter->wol |= E1000_WUFC_MAG;
2510
2511	/* Some vendors want WoL disabled by default, but still supported */
2512	if ((hw->mac.type == e1000_i350) &&
2513	    (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
2514		adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2515		adapter->wol = 0;
2516	}
2517
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2518	device_set_wakeup_enable(&adapter->pdev->dev,
2519				 adapter->flags & IGB_FLAG_WOL_SUPPORTED);
2520
2521	/* reset the hardware with the new settings */
2522	igb_reset(adapter);
2523
2524	/* Init the I2C interface */
2525	err = igb_init_i2c(adapter);
2526	if (err) {
2527		dev_err(&pdev->dev, "failed to init i2c interface\n");
2528		goto err_eeprom;
2529	}
2530
2531	/* let the f/w know that the h/w is now under the control of the
2532	 * driver. */
 
2533	igb_get_hw_control(adapter);
2534
2535	strcpy(netdev->name, "eth%d");
2536	err = register_netdev(netdev);
2537	if (err)
2538		goto err_register;
2539
2540	/* carrier off reporting is important to ethtool even BEFORE open */
2541	netif_carrier_off(netdev);
2542
2543#ifdef CONFIG_IGB_DCA
2544	if (dca_add_requester(&pdev->dev) == 0) {
2545		adapter->flags |= IGB_FLAG_DCA_ENABLED;
2546		dev_info(&pdev->dev, "DCA enabled\n");
2547		igb_setup_dca(adapter);
2548	}
2549
2550#endif
2551#ifdef CONFIG_IGB_HWMON
2552	/* Initialize the thermal sensor on i350 devices. */
2553	if (hw->mac.type == e1000_i350 && hw->bus.func == 0) {
2554		u16 ets_word;
2555
2556		/* Read the NVM to determine if this i350 device supports an
2557		 * external thermal sensor.
2558		 */
2559		hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word);
2560		if (ets_word != 0x0000 && ets_word != 0xFFFF)
2561			adapter->ets = true;
2562		else
2563			adapter->ets = false;
 
 
 
 
 
 
2564		if (igb_sysfs_init(adapter))
2565			dev_err(&pdev->dev,
2566				"failed to allocate sysfs resources\n");
2567	} else {
2568		adapter->ets = false;
2569	}
2570#endif
2571	/* Check if Media Autosense is enabled */
2572	adapter->ei = *ei;
2573	if (hw->dev_spec._82575.mas_capable)
2574		igb_init_mas(adapter);
2575
2576	/* do hw tstamp init after resetting */
2577	igb_ptp_init(adapter);
2578
2579	dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
2580	/* print bus type/speed/width info, not applicable to i354 */
2581	if (hw->mac.type != e1000_i354) {
2582		dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
2583			 netdev->name,
2584			 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
2585			  (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
2586			   "unknown"),
2587			 ((hw->bus.width == e1000_bus_width_pcie_x4) ?
2588			  "Width x4" :
2589			  (hw->bus.width == e1000_bus_width_pcie_x2) ?
2590			  "Width x2" :
2591			  (hw->bus.width == e1000_bus_width_pcie_x1) ?
2592			  "Width x1" : "unknown"), netdev->dev_addr);
2593	}
2594
2595	if ((hw->mac.type >= e1000_i210 ||
 
 
2596	     igb_get_flash_presence_i210(hw))) {
2597		ret_val = igb_read_part_string(hw, part_str,
2598					       E1000_PBANUM_LENGTH);
2599	} else {
2600		ret_val = -E1000_ERR_INVM_VALUE_NOT_FOUND;
2601	}
2602
2603	if (ret_val)
2604		strcpy(part_str, "Unknown");
2605	dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
2606	dev_info(&pdev->dev,
2607		"Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
2608		(adapter->flags & IGB_FLAG_HAS_MSIX) ? "MSI-X" :
2609		(adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
2610		adapter->num_rx_queues, adapter->num_tx_queues);
2611	if (hw->phy.media_type == e1000_media_type_copper) {
2612		switch (hw->mac.type) {
2613		case e1000_i350:
2614		case e1000_i210:
2615		case e1000_i211:
2616			/* Enable EEE for internal copper PHY devices */
2617			err = igb_set_eee_i350(hw);
2618			if ((!err) &&
2619			    (!hw->dev_spec._82575.eee_disable)) {
2620				adapter->eee_advert =
2621					MDIO_EEE_100TX | MDIO_EEE_1000T;
2622				adapter->flags |= IGB_FLAG_EEE;
2623			}
2624			break;
2625		case e1000_i354:
2626			if ((rd32(E1000_CTRL_EXT) &
2627			    E1000_CTRL_EXT_LINK_MODE_SGMII)) {
2628				err = igb_set_eee_i354(hw);
2629				if ((!err) &&
2630					(!hw->dev_spec._82575.eee_disable)) {
2631					adapter->eee_advert =
2632					   MDIO_EEE_100TX | MDIO_EEE_1000T;
2633					adapter->flags |= IGB_FLAG_EEE;
2634				}
2635			}
2636			break;
2637		default:
2638			break;
2639		}
2640	}
 
 
 
2641	pm_runtime_put_noidle(&pdev->dev);
2642	return 0;
2643
2644err_register:
2645	igb_release_hw_control(adapter);
2646	memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap));
2647err_eeprom:
2648	if (!igb_check_reset_block(hw))
2649		igb_reset_phy(hw);
2650
2651	if (hw->flash_address)
2652		iounmap(hw->flash_address);
2653err_sw_init:
 
 
2654	igb_clear_interrupt_scheme(adapter);
2655	pci_iounmap(pdev, hw->hw_addr);
 
 
 
2656err_ioremap:
2657	free_netdev(netdev);
2658err_alloc_etherdev:
2659	pci_release_selected_regions(pdev,
2660				     pci_select_bars(pdev, IORESOURCE_MEM));
2661err_pci_reg:
2662err_dma:
2663	pci_disable_device(pdev);
2664	return err;
2665}
2666
2667#ifdef CONFIG_PCI_IOV
2668static int igb_disable_sriov(struct pci_dev *pdev)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2669{
2670	struct net_device *netdev = pci_get_drvdata(pdev);
2671	struct igb_adapter *adapter = netdev_priv(netdev);
2672	struct e1000_hw *hw = &adapter->hw;
 
2673
2674	/* reclaim resources allocated to VFs */
2675	if (adapter->vf_data) {
2676		/* disable iov and allow time for transactions to clear */
2677		if (pci_vfs_assigned(pdev)) {
2678			dev_warn(&pdev->dev,
2679				 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
2680			return -EPERM;
2681		} else {
2682			pci_disable_sriov(pdev);
2683			msleep(500);
2684		}
2685
 
 
2686		kfree(adapter->vf_data);
2687		adapter->vf_data = NULL;
2688		adapter->vfs_allocated_count = 0;
 
2689		wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
2690		wrfl();
2691		msleep(100);
2692		dev_info(&pdev->dev, "IOV Disabled\n");
2693
2694		/* Re-enable DMA Coalescing flag since IOV is turned off */
2695		adapter->flags |= IGB_FLAG_DMAC;
2696	}
2697
2698	return 0;
2699}
2700
2701static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs)
2702{
2703	struct net_device *netdev = pci_get_drvdata(pdev);
2704	struct igb_adapter *adapter = netdev_priv(netdev);
2705	int old_vfs = pci_num_vf(pdev);
 
2706	int err = 0;
2707	int i;
2708
2709	if (!(adapter->flags & IGB_FLAG_HAS_MSIX) || num_vfs > 7) {
2710		err = -EPERM;
2711		goto out;
2712	}
2713	if (!num_vfs)
2714		goto out;
2715
2716	if (old_vfs) {
2717		dev_info(&pdev->dev, "%d pre-allocated VFs found - override max_vfs setting of %d\n",
2718			 old_vfs, max_vfs);
2719		adapter->vfs_allocated_count = old_vfs;
2720	} else
2721		adapter->vfs_allocated_count = num_vfs;
2722
2723	adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
2724				sizeof(struct vf_data_storage), GFP_KERNEL);
2725
2726	/* if allocation failed then we do not support SR-IOV */
2727	if (!adapter->vf_data) {
2728		adapter->vfs_allocated_count = 0;
2729		dev_err(&pdev->dev,
2730			"Unable to allocate memory for VF Data Storage\n");
2731		err = -ENOMEM;
2732		goto out;
2733	}
2734
2735	/* only call pci_enable_sriov() if no VFs are allocated already */
2736	if (!old_vfs) {
2737		err = pci_enable_sriov(pdev, adapter->vfs_allocated_count);
2738		if (err)
2739			goto err_out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2740	}
 
2741	dev_info(&pdev->dev, "%d VFs allocated\n",
2742		 adapter->vfs_allocated_count);
2743	for (i = 0; i < adapter->vfs_allocated_count; i++)
2744		igb_vf_configure(adapter, i);
2745
2746	/* DMA Coalescing is not supported in IOV mode. */
2747	adapter->flags &= ~IGB_FLAG_DMAC;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2748	goto out;
2749
2750err_out:
 
 
2751	kfree(adapter->vf_data);
2752	adapter->vf_data = NULL;
2753	adapter->vfs_allocated_count = 0;
2754out:
2755	return err;
2756}
2757
2758#endif
2759/**
2760 *  igb_remove_i2c - Cleanup  I2C interface
2761 *  @adapter: pointer to adapter structure
2762 **/
2763static void igb_remove_i2c(struct igb_adapter *adapter)
2764{
2765	/* free the adapter bus structure */
2766	i2c_del_adapter(&adapter->i2c_adap);
2767}
2768
2769/**
2770 *  igb_remove - Device Removal Routine
2771 *  @pdev: PCI device information struct
2772 *
2773 *  igb_remove is called by the PCI subsystem to alert the driver
2774 *  that it should release a PCI device.  The could be caused by a
2775 *  Hot-Plug event, or because the driver is going to be removed from
2776 *  memory.
2777 **/
2778static void igb_remove(struct pci_dev *pdev)
2779{
2780	struct net_device *netdev = pci_get_drvdata(pdev);
2781	struct igb_adapter *adapter = netdev_priv(netdev);
2782	struct e1000_hw *hw = &adapter->hw;
2783
2784	pm_runtime_get_noresume(&pdev->dev);
2785#ifdef CONFIG_IGB_HWMON
2786	igb_sysfs_exit(adapter);
2787#endif
2788	igb_remove_i2c(adapter);
2789	igb_ptp_stop(adapter);
2790	/* The watchdog timer may be rescheduled, so explicitly
2791	 * disable watchdog from being rescheduled.
2792	 */
2793	set_bit(__IGB_DOWN, &adapter->state);
2794	del_timer_sync(&adapter->watchdog_timer);
2795	del_timer_sync(&adapter->phy_info_timer);
2796
2797	cancel_work_sync(&adapter->reset_task);
2798	cancel_work_sync(&adapter->watchdog_task);
2799
2800#ifdef CONFIG_IGB_DCA
2801	if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
2802		dev_info(&pdev->dev, "DCA disabled\n");
2803		dca_remove_requester(&pdev->dev);
2804		adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
2805		wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
2806	}
2807#endif
2808
2809	/* Release control of h/w to f/w.  If f/w is AMT enabled, this
2810	 * would have already happened in close and is redundant.
2811	 */
2812	igb_release_hw_control(adapter);
2813
 
 
 
 
2814	unregister_netdev(netdev);
2815
2816	igb_clear_interrupt_scheme(adapter);
2817
2818#ifdef CONFIG_PCI_IOV
2819	igb_disable_sriov(pdev);
2820#endif
2821
2822	pci_iounmap(pdev, hw->hw_addr);
2823	if (hw->flash_address)
2824		iounmap(hw->flash_address);
2825	pci_release_selected_regions(pdev,
2826				     pci_select_bars(pdev, IORESOURCE_MEM));
2827
 
2828	kfree(adapter->shadow_vfta);
2829	free_netdev(netdev);
2830
2831	pci_disable_pcie_error_reporting(pdev);
2832
2833	pci_disable_device(pdev);
2834}
2835
2836/**
2837 *  igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
2838 *  @adapter: board private structure to initialize
2839 *
2840 *  This function initializes the vf specific data storage and then attempts to
2841 *  allocate the VFs.  The reason for ordering it this way is because it is much
2842 *  mor expensive time wise to disable SR-IOV than it is to allocate and free
2843 *  the memory for the VFs.
2844 **/
2845static void igb_probe_vfs(struct igb_adapter *adapter)
2846{
2847#ifdef CONFIG_PCI_IOV
2848	struct pci_dev *pdev = adapter->pdev;
2849	struct e1000_hw *hw = &adapter->hw;
2850
2851	/* Virtualization features not supported on i210 family. */
2852	if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
 
2853		return;
2854
 
 
 
 
 
 
 
2855	pci_sriov_set_totalvfs(pdev, 7);
2856	igb_pci_enable_sriov(pdev, max_vfs);
2857
2858#endif /* CONFIG_PCI_IOV */
2859}
2860
2861static void igb_init_queue_configuration(struct igb_adapter *adapter)
2862{
2863	struct e1000_hw *hw = &adapter->hw;
2864	u32 max_rss_queues;
2865
2866	/* Determine the maximum number of RSS queues supported. */
2867	switch (hw->mac.type) {
2868	case e1000_i211:
2869		max_rss_queues = IGB_MAX_RX_QUEUES_I211;
2870		break;
2871	case e1000_82575:
2872	case e1000_i210:
2873		max_rss_queues = IGB_MAX_RX_QUEUES_82575;
2874		break;
2875	case e1000_i350:
2876		/* I350 cannot do RSS and SR-IOV at the same time */
2877		if (!!adapter->vfs_allocated_count) {
2878			max_rss_queues = 1;
2879			break;
2880		}
2881		/* fall through */
2882	case e1000_82576:
2883		if (!!adapter->vfs_allocated_count) {
2884			max_rss_queues = 2;
2885			break;
2886		}
2887		/* fall through */
2888	case e1000_82580:
2889	case e1000_i354:
2890	default:
2891		max_rss_queues = IGB_MAX_RX_QUEUES;
2892		break;
2893	}
2894
 
 
 
 
 
 
 
 
2895	adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
2896
 
 
 
 
 
 
 
 
2897	/* Determine if we need to pair queues. */
2898	switch (hw->mac.type) {
2899	case e1000_82575:
2900	case e1000_i211:
2901		/* Device supports enough interrupts without queue pairing. */
2902		break;
2903	case e1000_82576:
2904		/* If VFs are going to be allocated with RSS queues then we
2905		 * should pair the queues in order to conserve interrupts due
2906		 * to limited supply.
2907		 */
2908		if ((adapter->rss_queues > 1) &&
2909		    (adapter->vfs_allocated_count > 6))
2910			adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2911		/* fall through */
2912	case e1000_82580:
2913	case e1000_i350:
2914	case e1000_i354:
2915	case e1000_i210:
2916	default:
2917		/* If rss_queues > half of max_rss_queues, pair the queues in
2918		 * order to conserve interrupts due to limited supply.
2919		 */
2920		if (adapter->rss_queues > (max_rss_queues / 2))
2921			adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
 
 
2922		break;
2923	}
2924}
2925
2926/**
2927 *  igb_sw_init - Initialize general software structures (struct igb_adapter)
2928 *  @adapter: board private structure to initialize
2929 *
2930 *  igb_sw_init initializes the Adapter private data structure.
2931 *  Fields are initialized based on PCI device information and
2932 *  OS network device settings (MTU size).
2933 **/
2934static int igb_sw_init(struct igb_adapter *adapter)
2935{
2936	struct e1000_hw *hw = &adapter->hw;
2937	struct net_device *netdev = adapter->netdev;
2938	struct pci_dev *pdev = adapter->pdev;
2939
2940	pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
2941
2942	/* set default ring sizes */
2943	adapter->tx_ring_count = IGB_DEFAULT_TXD;
2944	adapter->rx_ring_count = IGB_DEFAULT_RXD;
2945
2946	/* set default ITR values */
2947	adapter->rx_itr_setting = IGB_DEFAULT_ITR;
2948	adapter->tx_itr_setting = IGB_DEFAULT_ITR;
2949
2950	/* set default work limits */
2951	adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
2952
2953	adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
2954				  VLAN_HLEN;
2955	adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2956
 
2957	spin_lock_init(&adapter->stats64_lock);
 
 
 
2958#ifdef CONFIG_PCI_IOV
2959	switch (hw->mac.type) {
2960	case e1000_82576:
2961	case e1000_i350:
2962		if (max_vfs > 7) {
2963			dev_warn(&pdev->dev,
2964				 "Maximum of 7 VFs per PF, using max\n");
2965			max_vfs = adapter->vfs_allocated_count = 7;
2966		} else
2967			adapter->vfs_allocated_count = max_vfs;
2968		if (adapter->vfs_allocated_count)
2969			dev_warn(&pdev->dev,
2970				 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
2971		break;
2972	default:
2973		break;
2974	}
2975#endif /* CONFIG_PCI_IOV */
2976
 
 
 
 
 
 
 
 
 
 
 
2977	igb_init_queue_configuration(adapter);
2978
2979	/* Setup and initialize a copy of the hw vlan table array */
2980	adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32),
2981				       GFP_ATOMIC);
 
 
2982
2983	/* This call may decrease the number of queues */
2984	if (igb_init_interrupt_scheme(adapter, true)) {
2985		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
2986		return -ENOMEM;
2987	}
2988
2989	igb_probe_vfs(adapter);
2990
2991	/* Explicitly disable IRQ since the NIC can be in any state. */
2992	igb_irq_disable(adapter);
2993
2994	if (hw->mac.type >= e1000_i350)
2995		adapter->flags &= ~IGB_FLAG_DMAC;
2996
2997	set_bit(__IGB_DOWN, &adapter->state);
2998	return 0;
2999}
3000
3001/**
3002 *  igb_open - Called when a network interface is made active
3003 *  @netdev: network interface device structure
 
3004 *
3005 *  Returns 0 on success, negative value on failure
3006 *
3007 *  The open entry point is called when a network interface is made
3008 *  active by the system (IFF_UP).  At this point all resources needed
3009 *  for transmit and receive operations are allocated, the interrupt
3010 *  handler is registered with the OS, the watchdog timer is started,
3011 *  and the stack is notified that the interface is ready.
3012 **/
3013static int __igb_open(struct net_device *netdev, bool resuming)
3014{
3015	struct igb_adapter *adapter = netdev_priv(netdev);
3016	struct e1000_hw *hw = &adapter->hw;
3017	struct pci_dev *pdev = adapter->pdev;
3018	int err;
3019	int i;
3020
3021	/* disallow open during test */
3022	if (test_bit(__IGB_TESTING, &adapter->state)) {
3023		WARN_ON(resuming);
3024		return -EBUSY;
3025	}
3026
3027	if (!resuming)
3028		pm_runtime_get_sync(&pdev->dev);
3029
3030	netif_carrier_off(netdev);
3031
3032	/* allocate transmit descriptors */
3033	err = igb_setup_all_tx_resources(adapter);
3034	if (err)
3035		goto err_setup_tx;
3036
3037	/* allocate receive descriptors */
3038	err = igb_setup_all_rx_resources(adapter);
3039	if (err)
3040		goto err_setup_rx;
3041
3042	igb_power_up_link(adapter);
3043
3044	/* before we allocate an interrupt, we must be ready to handle it.
3045	 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3046	 * as soon as we call pci_request_irq, so we have to setup our
3047	 * clean_rx handler before we do so.
3048	 */
3049	igb_configure(adapter);
3050
3051	err = igb_request_irq(adapter);
3052	if (err)
3053		goto err_req_irq;
3054
3055	/* Notify the stack of the actual queue counts. */
3056	err = netif_set_real_num_tx_queues(adapter->netdev,
3057					   adapter->num_tx_queues);
3058	if (err)
3059		goto err_set_queues;
3060
3061	err = netif_set_real_num_rx_queues(adapter->netdev,
3062					   adapter->num_rx_queues);
3063	if (err)
3064		goto err_set_queues;
3065
3066	/* From here on the code is the same as igb_up() */
3067	clear_bit(__IGB_DOWN, &adapter->state);
3068
3069	for (i = 0; i < adapter->num_q_vectors; i++)
3070		napi_enable(&(adapter->q_vector[i]->napi));
3071
3072	/* Clear any pending interrupts. */
 
3073	rd32(E1000_ICR);
3074
3075	igb_irq_enable(adapter);
3076
3077	/* notify VFs that reset has been completed */
3078	if (adapter->vfs_allocated_count) {
3079		u32 reg_data = rd32(E1000_CTRL_EXT);
 
3080		reg_data |= E1000_CTRL_EXT_PFRSTD;
3081		wr32(E1000_CTRL_EXT, reg_data);
3082	}
3083
3084	netif_tx_start_all_queues(netdev);
3085
3086	if (!resuming)
3087		pm_runtime_put(&pdev->dev);
3088
3089	/* start the watchdog. */
3090	hw->mac.get_link_status = 1;
3091	schedule_work(&adapter->watchdog_task);
3092
3093	return 0;
3094
3095err_set_queues:
3096	igb_free_irq(adapter);
3097err_req_irq:
3098	igb_release_hw_control(adapter);
3099	igb_power_down_link(adapter);
3100	igb_free_all_rx_resources(adapter);
3101err_setup_rx:
3102	igb_free_all_tx_resources(adapter);
3103err_setup_tx:
3104	igb_reset(adapter);
3105	if (!resuming)
3106		pm_runtime_put(&pdev->dev);
3107
3108	return err;
3109}
3110
3111static int igb_open(struct net_device *netdev)
3112{
3113	return __igb_open(netdev, false);
3114}
3115
3116/**
3117 *  igb_close - Disables a network interface
3118 *  @netdev: network interface device structure
 
3119 *
3120 *  Returns 0, this is not allowed to fail
3121 *
3122 *  The close entry point is called when an interface is de-activated
3123 *  by the OS.  The hardware is still under the driver's control, but
3124 *  needs to be disabled.  A global MAC reset is issued to stop the
3125 *  hardware, and all transmit and receive resources are freed.
3126 **/
3127static int __igb_close(struct net_device *netdev, bool suspending)
3128{
3129	struct igb_adapter *adapter = netdev_priv(netdev);
3130	struct pci_dev *pdev = adapter->pdev;
3131
3132	WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
3133
3134	if (!suspending)
3135		pm_runtime_get_sync(&pdev->dev);
3136
3137	igb_down(adapter);
3138	igb_free_irq(adapter);
3139
3140	igb_free_all_tx_resources(adapter);
3141	igb_free_all_rx_resources(adapter);
3142
3143	if (!suspending)
3144		pm_runtime_put_sync(&pdev->dev);
3145	return 0;
3146}
3147
3148static int igb_close(struct net_device *netdev)
3149{
3150	return __igb_close(netdev, false);
 
 
3151}
3152
3153/**
3154 *  igb_setup_tx_resources - allocate Tx resources (Descriptors)
3155 *  @tx_ring: tx descriptor ring (for a specific queue) to setup
3156 *
3157 *  Return 0 on success, negative on failure
3158 **/
3159int igb_setup_tx_resources(struct igb_ring *tx_ring)
3160{
3161	struct device *dev = tx_ring->dev;
3162	int size;
3163
3164	size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3165
3166	tx_ring->tx_buffer_info = vzalloc(size);
3167	if (!tx_ring->tx_buffer_info)
3168		goto err;
3169
3170	/* round up to nearest 4K */
3171	tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
3172	tx_ring->size = ALIGN(tx_ring->size, 4096);
3173
3174	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
3175					   &tx_ring->dma, GFP_KERNEL);
3176	if (!tx_ring->desc)
3177		goto err;
3178
3179	tx_ring->next_to_use = 0;
3180	tx_ring->next_to_clean = 0;
3181
3182	return 0;
3183
3184err:
3185	vfree(tx_ring->tx_buffer_info);
3186	tx_ring->tx_buffer_info = NULL;
3187	dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
3188	return -ENOMEM;
3189}
3190
3191/**
3192 *  igb_setup_all_tx_resources - wrapper to allocate Tx resources
3193 *				 (Descriptors) for all queues
3194 *  @adapter: board private structure
3195 *
3196 *  Return 0 on success, negative on failure
3197 **/
3198static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
3199{
3200	struct pci_dev *pdev = adapter->pdev;
3201	int i, err = 0;
3202
3203	for (i = 0; i < adapter->num_tx_queues; i++) {
3204		err = igb_setup_tx_resources(adapter->tx_ring[i]);
3205		if (err) {
3206			dev_err(&pdev->dev,
3207				"Allocation for Tx Queue %u failed\n", i);
3208			for (i--; i >= 0; i--)
3209				igb_free_tx_resources(adapter->tx_ring[i]);
3210			break;
3211		}
3212	}
3213
3214	return err;
3215}
3216
3217/**
3218 *  igb_setup_tctl - configure the transmit control registers
3219 *  @adapter: Board private structure
3220 **/
3221void igb_setup_tctl(struct igb_adapter *adapter)
3222{
3223	struct e1000_hw *hw = &adapter->hw;
3224	u32 tctl;
3225
3226	/* disable queue 0 which is enabled by default on 82575 and 82576 */
3227	wr32(E1000_TXDCTL(0), 0);
3228
3229	/* Program the Transmit Control Register */
3230	tctl = rd32(E1000_TCTL);
3231	tctl &= ~E1000_TCTL_CT;
3232	tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
3233		(E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
3234
3235	igb_config_collision_dist(hw);
3236
3237	/* Enable transmits */
3238	tctl |= E1000_TCTL_EN;
3239
3240	wr32(E1000_TCTL, tctl);
3241}
3242
3243/**
3244 *  igb_configure_tx_ring - Configure transmit ring after Reset
3245 *  @adapter: board private structure
3246 *  @ring: tx ring to configure
3247 *
3248 *  Configure a transmit ring after a reset.
3249 **/
3250void igb_configure_tx_ring(struct igb_adapter *adapter,
3251                           struct igb_ring *ring)
3252{
3253	struct e1000_hw *hw = &adapter->hw;
3254	u32 txdctl = 0;
3255	u64 tdba = ring->dma;
3256	int reg_idx = ring->reg_idx;
3257
3258	/* disable the queue */
3259	wr32(E1000_TXDCTL(reg_idx), 0);
3260	wrfl();
3261	mdelay(10);
3262
3263	wr32(E1000_TDLEN(reg_idx),
3264	     ring->count * sizeof(union e1000_adv_tx_desc));
3265	wr32(E1000_TDBAL(reg_idx),
3266	     tdba & 0x00000000ffffffffULL);
3267	wr32(E1000_TDBAH(reg_idx), tdba >> 32);
3268
3269	ring->tail = hw->hw_addr + E1000_TDT(reg_idx);
3270	wr32(E1000_TDH(reg_idx), 0);
3271	writel(0, ring->tail);
3272
3273	txdctl |= IGB_TX_PTHRESH;
3274	txdctl |= IGB_TX_HTHRESH << 8;
3275	txdctl |= IGB_TX_WTHRESH << 16;
3276
 
 
 
 
3277	txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
3278	wr32(E1000_TXDCTL(reg_idx), txdctl);
3279}
3280
3281/**
3282 *  igb_configure_tx - Configure transmit Unit after Reset
3283 *  @adapter: board private structure
3284 *
3285 *  Configure the Tx unit of the MAC after a reset.
3286 **/
3287static void igb_configure_tx(struct igb_adapter *adapter)
3288{
 
3289	int i;
3290
 
 
 
 
 
 
 
3291	for (i = 0; i < adapter->num_tx_queues; i++)
3292		igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
3293}
3294
3295/**
3296 *  igb_setup_rx_resources - allocate Rx resources (Descriptors)
3297 *  @rx_ring: Rx descriptor ring (for a specific queue) to setup
3298 *
3299 *  Returns 0 on success, negative on failure
3300 **/
3301int igb_setup_rx_resources(struct igb_ring *rx_ring)
3302{
 
3303	struct device *dev = rx_ring->dev;
3304	int size;
 
 
 
 
 
 
 
 
 
 
 
3305
3306	size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3307
3308	rx_ring->rx_buffer_info = vzalloc(size);
3309	if (!rx_ring->rx_buffer_info)
3310		goto err;
3311
3312	/* Round up to nearest 4K */
3313	rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
3314	rx_ring->size = ALIGN(rx_ring->size, 4096);
3315
3316	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
3317					   &rx_ring->dma, GFP_KERNEL);
3318	if (!rx_ring->desc)
3319		goto err;
3320
3321	rx_ring->next_to_alloc = 0;
3322	rx_ring->next_to_clean = 0;
3323	rx_ring->next_to_use = 0;
3324
 
 
3325	return 0;
3326
3327err:
 
3328	vfree(rx_ring->rx_buffer_info);
3329	rx_ring->rx_buffer_info = NULL;
3330	dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
3331	return -ENOMEM;
3332}
3333
3334/**
3335 *  igb_setup_all_rx_resources - wrapper to allocate Rx resources
3336 *				 (Descriptors) for all queues
3337 *  @adapter: board private structure
3338 *
3339 *  Return 0 on success, negative on failure
3340 **/
3341static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
3342{
3343	struct pci_dev *pdev = adapter->pdev;
3344	int i, err = 0;
3345
3346	for (i = 0; i < adapter->num_rx_queues; i++) {
3347		err = igb_setup_rx_resources(adapter->rx_ring[i]);
3348		if (err) {
3349			dev_err(&pdev->dev,
3350				"Allocation for Rx Queue %u failed\n", i);
3351			for (i--; i >= 0; i--)
3352				igb_free_rx_resources(adapter->rx_ring[i]);
3353			break;
3354		}
3355	}
3356
3357	return err;
3358}
3359
3360/**
3361 *  igb_setup_mrqc - configure the multiple receive queue control registers
3362 *  @adapter: Board private structure
3363 **/
3364static void igb_setup_mrqc(struct igb_adapter *adapter)
3365{
3366	struct e1000_hw *hw = &adapter->hw;
3367	u32 mrqc, rxcsum;
3368	u32 j, num_rx_queues;
3369	static const u32 rsskey[10] = { 0xDA565A6D, 0xC20E5B25, 0x3D256741,
3370					0xB08FA343, 0xCB2BCAD0, 0xB4307BAE,
3371					0xA32DCB77, 0x0CF23080, 0x3BB7426A,
3372					0xFA01ACBE };
3373
3374	/* Fill out hash function seeds */
3375	for (j = 0; j < 10; j++)
3376		wr32(E1000_RSSRK(j), rsskey[j]);
3377
3378	num_rx_queues = adapter->rss_queues;
3379
3380	switch (hw->mac.type) {
3381	case e1000_82576:
3382		/* 82576 supports 2 RSS queues for SR-IOV */
3383		if (adapter->vfs_allocated_count)
3384			num_rx_queues = 2;
3385		break;
3386	default:
3387		break;
3388	}
3389
3390	if (adapter->rss_indir_tbl_init != num_rx_queues) {
3391		for (j = 0; j < IGB_RETA_SIZE; j++)
3392			adapter->rss_indir_tbl[j] = (j * num_rx_queues) / IGB_RETA_SIZE;
 
3393		adapter->rss_indir_tbl_init = num_rx_queues;
3394	}
3395	igb_write_rss_indir_tbl(adapter);
3396
3397	/* Disable raw packet checksumming so that RSS hash is placed in
3398	 * descriptor on writeback.  No need to enable TCP/UDP/IP checksum
3399	 * offloads as they are enabled by default
3400	 */
3401	rxcsum = rd32(E1000_RXCSUM);
3402	rxcsum |= E1000_RXCSUM_PCSD;
3403
3404	if (adapter->hw.mac.type >= e1000_82576)
3405		/* Enable Receive Checksum Offload for SCTP */
3406		rxcsum |= E1000_RXCSUM_CRCOFL;
3407
3408	/* Don't need to set TUOFL or IPOFL, they default to 1 */
3409	wr32(E1000_RXCSUM, rxcsum);
3410
3411	/* Generate RSS hash based on packet types, TCP/UDP
3412	 * port numbers and/or IPv4/v6 src and dst addresses
3413	 */
3414	mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
3415	       E1000_MRQC_RSS_FIELD_IPV4_TCP |
3416	       E1000_MRQC_RSS_FIELD_IPV6 |
3417	       E1000_MRQC_RSS_FIELD_IPV6_TCP |
3418	       E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
3419
3420	if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
3421		mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
3422	if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
3423		mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
3424
3425	/* If VMDq is enabled then we set the appropriate mode for that, else
3426	 * we default to RSS so that an RSS hash is calculated per packet even
3427	 * if we are only using one queue
3428	 */
3429	if (adapter->vfs_allocated_count) {
3430		if (hw->mac.type > e1000_82575) {
3431			/* Set the default pool for the PF's first queue */
3432			u32 vtctl = rd32(E1000_VT_CTL);
 
3433			vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
3434				   E1000_VT_CTL_DISABLE_DEF_POOL);
3435			vtctl |= adapter->vfs_allocated_count <<
3436				E1000_VT_CTL_DEFAULT_POOL_SHIFT;
3437			wr32(E1000_VT_CTL, vtctl);
3438		}
3439		if (adapter->rss_queues > 1)
3440			mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
3441		else
3442			mrqc |= E1000_MRQC_ENABLE_VMDQ;
3443	} else {
3444		if (hw->mac.type != e1000_i211)
3445			mrqc |= E1000_MRQC_ENABLE_RSS_4Q;
3446	}
3447	igb_vmm_control(adapter);
3448
3449	wr32(E1000_MRQC, mrqc);
3450}
3451
3452/**
3453 *  igb_setup_rctl - configure the receive control registers
3454 *  @adapter: Board private structure
3455 **/
3456void igb_setup_rctl(struct igb_adapter *adapter)
3457{
3458	struct e1000_hw *hw = &adapter->hw;
3459	u32 rctl;
3460
3461	rctl = rd32(E1000_RCTL);
3462
3463	rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3464	rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
3465
3466	rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
3467		(hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3468
3469	/* enable stripping of CRC. It's unlikely this will break BMC
3470	 * redirection as it did with e1000. Newer features require
3471	 * that the HW strips the CRC.
3472	 */
3473	rctl |= E1000_RCTL_SECRC;
3474
3475	/* disable store bad packets and clear size bits. */
3476	rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
3477
3478	/* enable LPE to prevent packets larger than max_frame_size */
3479	rctl |= E1000_RCTL_LPE;
3480
3481	/* disable queue 0 to prevent tail write w/o re-config */
3482	wr32(E1000_RXDCTL(0), 0);
3483
3484	/* Attention!!!  For SR-IOV PF driver operations you must enable
3485	 * queue drop for all VF and PF queues to prevent head of line blocking
3486	 * if an un-trusted VF does not provide descriptors to hardware.
3487	 */
3488	if (adapter->vfs_allocated_count) {
3489		/* set all queue drop enable bits */
3490		wr32(E1000_QDE, ALL_QUEUES);
3491	}
3492
3493	/* This is useful for sniffing bad packets. */
3494	if (adapter->netdev->features & NETIF_F_RXALL) {
3495		/* UPE and MPE will be handled by normal PROMISC logic
3496		 * in e1000e_set_rx_mode
3497		 */
3498		rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3499			 E1000_RCTL_BAM | /* RX All Bcast Pkts */
3500			 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3501
3502		rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3503			  E1000_RCTL_DPF | /* Allow filtered pause */
3504			  E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3505		/* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3506		 * and that breaks VLANs.
3507		 */
3508	}
3509
3510	wr32(E1000_RCTL, rctl);
3511}
3512
3513static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
3514                                   int vfn)
3515{
3516	struct e1000_hw *hw = &adapter->hw;
3517	u32 vmolr;
3518
3519	/* if it isn't the PF check to see if VFs are enabled and
3520	 * increase the size to support vlan tags
3521	 */
3522	if (vfn < adapter->vfs_allocated_count &&
3523	    adapter->vf_data[vfn].vlans_enabled)
3524		size += VLAN_TAG_SIZE;
3525
3526	vmolr = rd32(E1000_VMOLR(vfn));
3527	vmolr &= ~E1000_VMOLR_RLPML_MASK;
3528	vmolr |= size | E1000_VMOLR_LPE;
3529	wr32(E1000_VMOLR(vfn), vmolr);
3530
3531	return 0;
3532}
3533
3534/**
3535 *  igb_rlpml_set - set maximum receive packet size
3536 *  @adapter: board private structure
3537 *
3538 *  Configure maximum receivable packet size.
3539 **/
3540static void igb_rlpml_set(struct igb_adapter *adapter)
3541{
3542	u32 max_frame_size = adapter->max_frame_size;
3543	struct e1000_hw *hw = &adapter->hw;
3544	u16 pf_id = adapter->vfs_allocated_count;
 
 
 
3545
3546	if (pf_id) {
3547		igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
3548		/* If we're in VMDQ or SR-IOV mode, then set global RLPML
3549		 * to our max jumbo frame size, in case we need to enable
3550		 * jumbo frames on one of the rings later.
3551		 * This will not pass over-length frames into the default
3552		 * queue because it's gated by the VMOLR.RLPML.
3553		 */
3554		max_frame_size = MAX_JUMBO_FRAME_SIZE;
3555	}
3556
3557	wr32(E1000_RLPML, max_frame_size);
 
 
 
 
 
3558}
3559
3560static inline void igb_set_vmolr(struct igb_adapter *adapter,
3561				 int vfn, bool aupe)
3562{
3563	struct e1000_hw *hw = &adapter->hw;
3564	u32 vmolr;
3565
3566	/* This register exists only on 82576 and newer so if we are older then
3567	 * we should exit and do nothing
3568	 */
3569	if (hw->mac.type < e1000_82576)
3570		return;
3571
3572	vmolr = rd32(E1000_VMOLR(vfn));
3573	vmolr |= E1000_VMOLR_STRVLAN; /* Strip vlan tags */
3574	if (hw->mac.type == e1000_i350) {
3575		u32 dvmolr;
3576
3577		dvmolr = rd32(E1000_DVMOLR(vfn));
3578		dvmolr |= E1000_DVMOLR_STRVLAN;
3579		wr32(E1000_DVMOLR(vfn), dvmolr);
3580	}
3581	if (aupe)
3582		vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
3583	else
3584		vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
3585
3586	/* clear all bits that might not be set */
3587	vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
3588
3589	if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
3590		vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
3591	/* for VMDq only allow the VFs and pool 0 to accept broadcast and
3592	 * multicast packets
3593	 */
3594	if (vfn <= adapter->vfs_allocated_count)
3595		vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
3596
3597	wr32(E1000_VMOLR(vfn), vmolr);
3598}
3599
3600/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3601 *  igb_configure_rx_ring - Configure a receive ring after Reset
3602 *  @adapter: board private structure
3603 *  @ring: receive ring to be configured
3604 *
3605 *  Configure the Rx unit of the MAC after a reset.
3606 **/
3607void igb_configure_rx_ring(struct igb_adapter *adapter,
3608			   struct igb_ring *ring)
3609{
3610	struct e1000_hw *hw = &adapter->hw;
 
3611	u64 rdba = ring->dma;
3612	int reg_idx = ring->reg_idx;
3613	u32 srrctl = 0, rxdctl = 0;
 
 
 
 
3614
3615	/* disable the queue */
3616	wr32(E1000_RXDCTL(reg_idx), 0);
3617
3618	/* Set DMA base address registers */
3619	wr32(E1000_RDBAL(reg_idx),
3620	     rdba & 0x00000000ffffffffULL);
3621	wr32(E1000_RDBAH(reg_idx), rdba >> 32);
3622	wr32(E1000_RDLEN(reg_idx),
3623	     ring->count * sizeof(union e1000_adv_rx_desc));
3624
3625	/* initialize head and tail */
3626	ring->tail = hw->hw_addr + E1000_RDT(reg_idx);
3627	wr32(E1000_RDH(reg_idx), 0);
3628	writel(0, ring->tail);
3629
3630	/* set descriptor configuration */
3631	srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
3632	srrctl |= IGB_RX_BUFSZ >> E1000_SRRCTL_BSIZEPKT_SHIFT;
3633	srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
3634	if (hw->mac.type >= e1000_82580)
3635		srrctl |= E1000_SRRCTL_TIMESTAMP;
3636	/* Only set Drop Enable if we are supporting multiple queues */
3637	if (adapter->vfs_allocated_count || adapter->num_rx_queues > 1)
3638		srrctl |= E1000_SRRCTL_DROP_EN;
3639
3640	wr32(E1000_SRRCTL(reg_idx), srrctl);
3641
3642	/* set filtering for VMDQ pools */
3643	igb_set_vmolr(adapter, reg_idx & 0x7, true);
3644
3645	rxdctl |= IGB_RX_PTHRESH;
3646	rxdctl |= IGB_RX_HTHRESH << 8;
3647	rxdctl |= IGB_RX_WTHRESH << 16;
3648
 
 
 
 
 
 
 
 
3649	/* enable receive descriptor fetching */
3650	rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
3651	wr32(E1000_RXDCTL(reg_idx), rxdctl);
3652}
3653
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3654/**
3655 *  igb_configure_rx - Configure receive Unit after Reset
3656 *  @adapter: board private structure
3657 *
3658 *  Configure the Rx unit of the MAC after a reset.
3659 **/
3660static void igb_configure_rx(struct igb_adapter *adapter)
3661{
3662	int i;
3663
3664	/* set UTA to appropriate mode */
3665	igb_set_uta(adapter);
3666
3667	/* set the correct pool for the PF default MAC address in entry 0 */
3668	igb_rar_set_qsel(adapter, adapter->hw.mac.addr, 0,
3669			 adapter->vfs_allocated_count);
3670
3671	/* Setup the HW Rx Head and Tail Descriptor Pointers and
3672	 * the Base and Length of the Rx Descriptor Ring
3673	 */
3674	for (i = 0; i < adapter->num_rx_queues; i++)
3675		igb_configure_rx_ring(adapter, adapter->rx_ring[i]);
 
 
 
 
3676}
3677
3678/**
3679 *  igb_free_tx_resources - Free Tx Resources per Queue
3680 *  @tx_ring: Tx descriptor ring for a specific queue
3681 *
3682 *  Free all transmit software resources
3683 **/
3684void igb_free_tx_resources(struct igb_ring *tx_ring)
3685{
3686	igb_clean_tx_ring(tx_ring);
3687
3688	vfree(tx_ring->tx_buffer_info);
3689	tx_ring->tx_buffer_info = NULL;
3690
3691	/* if not set, then don't free */
3692	if (!tx_ring->desc)
3693		return;
3694
3695	dma_free_coherent(tx_ring->dev, tx_ring->size,
3696			  tx_ring->desc, tx_ring->dma);
3697
3698	tx_ring->desc = NULL;
3699}
3700
3701/**
3702 *  igb_free_all_tx_resources - Free Tx Resources for All Queues
3703 *  @adapter: board private structure
3704 *
3705 *  Free all transmit software resources
3706 **/
3707static void igb_free_all_tx_resources(struct igb_adapter *adapter)
3708{
3709	int i;
3710
3711	for (i = 0; i < adapter->num_tx_queues; i++)
3712		igb_free_tx_resources(adapter->tx_ring[i]);
3713}
3714
3715void igb_unmap_and_free_tx_resource(struct igb_ring *ring,
3716				    struct igb_tx_buffer *tx_buffer)
3717{
3718	if (tx_buffer->skb) {
3719		dev_kfree_skb_any(tx_buffer->skb);
3720		if (dma_unmap_len(tx_buffer, len))
3721			dma_unmap_single(ring->dev,
3722					 dma_unmap_addr(tx_buffer, dma),
3723					 dma_unmap_len(tx_buffer, len),
3724					 DMA_TO_DEVICE);
3725	} else if (dma_unmap_len(tx_buffer, len)) {
3726		dma_unmap_page(ring->dev,
3727			       dma_unmap_addr(tx_buffer, dma),
3728			       dma_unmap_len(tx_buffer, len),
3729			       DMA_TO_DEVICE);
3730	}
3731	tx_buffer->next_to_watch = NULL;
3732	tx_buffer->skb = NULL;
3733	dma_unmap_len_set(tx_buffer, len, 0);
3734	/* buffer_info must be completely set up in the transmit path */
3735}
3736
3737/**
3738 *  igb_clean_tx_ring - Free Tx Buffers
3739 *  @tx_ring: ring to be cleaned
3740 **/
3741static void igb_clean_tx_ring(struct igb_ring *tx_ring)
3742{
3743	struct igb_tx_buffer *buffer_info;
3744	unsigned long size;
3745	u16 i;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3746
3747	if (!tx_ring->tx_buffer_info)
3748		return;
3749	/* Free all the Tx ring sk_buffs */
 
 
 
 
 
 
3750
3751	for (i = 0; i < tx_ring->count; i++) {
3752		buffer_info = &tx_ring->tx_buffer_info[i];
3753		igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
 
 
 
 
3754	}
3755
 
3756	netdev_tx_reset_queue(txring_txq(tx_ring));
3757
3758	size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3759	memset(tx_ring->tx_buffer_info, 0, size);
3760
3761	/* Zero out the descriptor ring */
3762	memset(tx_ring->desc, 0, tx_ring->size);
3763
3764	tx_ring->next_to_use = 0;
3765	tx_ring->next_to_clean = 0;
3766}
3767
3768/**
3769 *  igb_clean_all_tx_rings - Free Tx Buffers for all queues
3770 *  @adapter: board private structure
3771 **/
3772static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
3773{
3774	int i;
3775
3776	for (i = 0; i < adapter->num_tx_queues; i++)
3777		igb_clean_tx_ring(adapter->tx_ring[i]);
 
3778}
3779
3780/**
3781 *  igb_free_rx_resources - Free Rx Resources
3782 *  @rx_ring: ring to clean the resources from
3783 *
3784 *  Free all receive software resources
3785 **/
3786void igb_free_rx_resources(struct igb_ring *rx_ring)
3787{
3788	igb_clean_rx_ring(rx_ring);
3789
 
 
3790	vfree(rx_ring->rx_buffer_info);
3791	rx_ring->rx_buffer_info = NULL;
3792
3793	/* if not set, then don't free */
3794	if (!rx_ring->desc)
3795		return;
3796
3797	dma_free_coherent(rx_ring->dev, rx_ring->size,
3798			  rx_ring->desc, rx_ring->dma);
3799
3800	rx_ring->desc = NULL;
3801}
3802
3803/**
3804 *  igb_free_all_rx_resources - Free Rx Resources for All Queues
3805 *  @adapter: board private structure
3806 *
3807 *  Free all receive software resources
3808 **/
3809static void igb_free_all_rx_resources(struct igb_adapter *adapter)
3810{
3811	int i;
3812
3813	for (i = 0; i < adapter->num_rx_queues; i++)
3814		igb_free_rx_resources(adapter->rx_ring[i]);
 
3815}
3816
3817/**
3818 *  igb_clean_rx_ring - Free Rx Buffers per Queue
3819 *  @rx_ring: ring to free buffers from
3820 **/
3821static void igb_clean_rx_ring(struct igb_ring *rx_ring)
3822{
3823	unsigned long size;
3824	u16 i;
3825
3826	if (rx_ring->skb)
3827		dev_kfree_skb(rx_ring->skb);
3828	rx_ring->skb = NULL;
3829
3830	if (!rx_ring->rx_buffer_info)
3831		return;
3832
3833	/* Free all the Rx ring sk_buffs */
3834	for (i = 0; i < rx_ring->count; i++) {
3835		struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
3836
3837		if (!buffer_info->page)
3838			continue;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3839
3840		dma_unmap_page(rx_ring->dev,
3841			       buffer_info->dma,
3842			       PAGE_SIZE,
3843			       DMA_FROM_DEVICE);
3844		__free_page(buffer_info->page);
3845
3846		buffer_info->page = NULL;
3847	}
3848
3849	size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3850	memset(rx_ring->rx_buffer_info, 0, size);
3851
3852	/* Zero out the descriptor ring */
3853	memset(rx_ring->desc, 0, rx_ring->size);
3854
3855	rx_ring->next_to_alloc = 0;
3856	rx_ring->next_to_clean = 0;
3857	rx_ring->next_to_use = 0;
3858}
3859
3860/**
3861 *  igb_clean_all_rx_rings - Free Rx Buffers for all queues
3862 *  @adapter: board private structure
3863 **/
3864static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
3865{
3866	int i;
3867
3868	for (i = 0; i < adapter->num_rx_queues; i++)
3869		igb_clean_rx_ring(adapter->rx_ring[i]);
 
3870}
3871
3872/**
3873 *  igb_set_mac - Change the Ethernet Address of the NIC
3874 *  @netdev: network interface device structure
3875 *  @p: pointer to an address structure
3876 *
3877 *  Returns 0 on success, negative on failure
3878 **/
3879static int igb_set_mac(struct net_device *netdev, void *p)
3880{
3881	struct igb_adapter *adapter = netdev_priv(netdev);
3882	struct e1000_hw *hw = &adapter->hw;
3883	struct sockaddr *addr = p;
3884
3885	if (!is_valid_ether_addr(addr->sa_data))
3886		return -EADDRNOTAVAIL;
3887
3888	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3889	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3890
3891	/* set the correct pool for the new PF MAC address in entry 0 */
3892	igb_rar_set_qsel(adapter, hw->mac.addr, 0,
3893			 adapter->vfs_allocated_count);
3894
3895	return 0;
3896}
3897
3898/**
3899 *  igb_write_mc_addr_list - write multicast addresses to MTA
3900 *  @netdev: network interface device structure
3901 *
3902 *  Writes multicast address list to the MTA hash table.
3903 *  Returns: -ENOMEM on failure
3904 *           0 on no addresses written
3905 *           X on writing X addresses to MTA
3906 **/
3907static int igb_write_mc_addr_list(struct net_device *netdev)
3908{
3909	struct igb_adapter *adapter = netdev_priv(netdev);
3910	struct e1000_hw *hw = &adapter->hw;
3911	struct netdev_hw_addr *ha;
3912	u8  *mta_list;
3913	int i;
3914
3915	if (netdev_mc_empty(netdev)) {
3916		/* nothing to program, so clear mc list */
3917		igb_update_mc_addr_list(hw, NULL, 0);
3918		igb_restore_vf_multicasts(adapter);
3919		return 0;
3920	}
3921
3922	mta_list = kzalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3923	if (!mta_list)
3924		return -ENOMEM;
3925
3926	/* The shared function expects a packed array of only addresses. */
3927	i = 0;
3928	netdev_for_each_mc_addr(ha, netdev)
3929		memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3930
3931	igb_update_mc_addr_list(hw, mta_list, i);
3932	kfree(mta_list);
3933
3934	return netdev_mc_count(netdev);
3935}
3936
3937/**
3938 *  igb_write_uc_addr_list - write unicast addresses to RAR table
3939 *  @netdev: network interface device structure
3940 *
3941 *  Writes unicast address list to the RAR table.
3942 *  Returns: -ENOMEM on failure/insufficient address space
3943 *           0 on no addresses written
3944 *           X on writing X addresses to the RAR table
3945 **/
3946static int igb_write_uc_addr_list(struct net_device *netdev)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3947{
3948	struct igb_adapter *adapter = netdev_priv(netdev);
3949	struct e1000_hw *hw = &adapter->hw;
3950	unsigned int vfn = adapter->vfs_allocated_count;
3951	unsigned int rar_entries = hw->mac.rar_entry_count - (vfn + 1);
3952	int count = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3953
3954	/* return ENOMEM indicating insufficient memory for addresses */
3955	if (netdev_uc_count(netdev) > rar_entries)
3956		return -ENOMEM;
3957
3958	if (!netdev_uc_empty(netdev) && rar_entries) {
3959		struct netdev_hw_addr *ha;
 
3960
3961		netdev_for_each_uc_addr(ha, netdev) {
3962			if (!rar_entries)
3963				break;
3964			igb_rar_set_qsel(adapter, ha->addr,
3965					 rar_entries--,
3966					 vfn);
3967			count++;
3968		}
 
 
 
 
 
3969	}
3970	/* write the addresses in reverse order to avoid write combining */
3971	for (; rar_entries > 0 ; rar_entries--) {
3972		wr32(E1000_RAH(rar_entries), 0);
3973		wr32(E1000_RAL(rar_entries), 0);
 
 
 
 
 
 
 
3974	}
3975	wrfl();
 
 
 
 
 
 
 
 
 
 
 
3976
3977	return count;
 
3978}
3979
3980/**
3981 *  igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
3982 *  @netdev: network interface device structure
3983 *
3984 *  The set_rx_mode entry point is called whenever the unicast or multicast
3985 *  address lists or the network interface flags are updated.  This routine is
3986 *  responsible for configuring the hardware for proper unicast, multicast,
3987 *  promiscuous mode, and all-multi behavior.
3988 **/
3989static void igb_set_rx_mode(struct net_device *netdev)
3990{
3991	struct igb_adapter *adapter = netdev_priv(netdev);
3992	struct e1000_hw *hw = &adapter->hw;
3993	unsigned int vfn = adapter->vfs_allocated_count;
3994	u32 rctl, vmolr = 0;
3995	int count;
3996
3997	/* Check for Promiscuous and All Multicast modes */
3998	rctl = rd32(E1000_RCTL);
 
 
3999
4000	/* clear the effected bits */
4001	rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE | E1000_RCTL_VFE);
4002
4003	if (netdev->flags & IFF_PROMISC) {
4004		/* retain VLAN HW filtering if in VT mode */
4005		if (adapter->vfs_allocated_count)
4006			rctl |= E1000_RCTL_VFE;
4007		rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
4008		vmolr |= (E1000_VMOLR_ROPE | E1000_VMOLR_MPME);
4009	} else {
4010		if (netdev->flags & IFF_ALLMULTI) {
4011			rctl |= E1000_RCTL_MPE;
4012			vmolr |= E1000_VMOLR_MPME;
4013		} else {
4014			/* Write addresses to the MTA, if the attempt fails
4015			 * then we should just turn on promiscuous mode so
4016			 * that we can at least receive multicast traffic
4017			 */
4018			count = igb_write_mc_addr_list(netdev);
4019			if (count < 0) {
4020				rctl |= E1000_RCTL_MPE;
4021				vmolr |= E1000_VMOLR_MPME;
4022			} else if (count) {
4023				vmolr |= E1000_VMOLR_ROMPE;
4024			}
4025		}
4026		/* Write addresses to available RAR registers, if there is not
4027		 * sufficient space to store all the addresses then enable
4028		 * unicast promiscuous mode
4029		 */
4030		count = igb_write_uc_addr_list(netdev);
4031		if (count < 0) {
4032			rctl |= E1000_RCTL_UPE;
4033			vmolr |= E1000_VMOLR_ROPE;
4034		}
4035		rctl |= E1000_RCTL_VFE;
4036	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4037	wr32(E1000_RCTL, rctl);
4038
 
 
 
 
 
 
 
 
4039	/* In order to support SR-IOV and eventually VMDq it is necessary to set
4040	 * the VMOLR to enable the appropriate modes.  Without this workaround
4041	 * we will have issues with VLAN tag stripping not being done for frames
4042	 * that are only arriving because we are the default pool
4043	 */
4044	if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
4045		return;
4046
 
 
 
4047	vmolr |= rd32(E1000_VMOLR(vfn)) &
4048		 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
 
 
 
 
 
 
 
 
 
 
 
4049	wr32(E1000_VMOLR(vfn), vmolr);
 
4050	igb_restore_vf_multicasts(adapter);
4051}
4052
4053static void igb_check_wvbr(struct igb_adapter *adapter)
4054{
4055	struct e1000_hw *hw = &adapter->hw;
4056	u32 wvbr = 0;
4057
4058	switch (hw->mac.type) {
4059	case e1000_82576:
4060	case e1000_i350:
4061		if (!(wvbr = rd32(E1000_WVBR)))
 
4062			return;
4063		break;
4064	default:
4065		break;
4066	}
4067
4068	adapter->wvbr |= wvbr;
4069}
4070
4071#define IGB_STAGGERED_QUEUE_OFFSET 8
4072
4073static void igb_spoof_check(struct igb_adapter *adapter)
4074{
4075	int j;
4076
4077	if (!adapter->wvbr)
4078		return;
4079
4080	for(j = 0; j < adapter->vfs_allocated_count; j++) {
4081		if (adapter->wvbr & (1 << j) ||
4082		    adapter->wvbr & (1 << (j + IGB_STAGGERED_QUEUE_OFFSET))) {
4083			dev_warn(&adapter->pdev->dev,
4084				"Spoof event(s) detected on VF %d\n", j);
4085			adapter->wvbr &=
4086				~((1 << j) |
4087				  (1 << (j + IGB_STAGGERED_QUEUE_OFFSET)));
4088		}
4089	}
4090}
4091
4092/* Need to wait a few seconds after link up to get diagnostic information from
4093 * the phy
4094 */
4095static void igb_update_phy_info(unsigned long data)
4096{
4097	struct igb_adapter *adapter = (struct igb_adapter *) data;
4098	igb_get_phy_info(&adapter->hw);
4099}
4100
4101/**
4102 *  igb_has_link - check shared code for link and determine up/down
4103 *  @adapter: pointer to driver private info
4104 **/
4105bool igb_has_link(struct igb_adapter *adapter)
4106{
4107	struct e1000_hw *hw = &adapter->hw;
4108	bool link_active = false;
4109
4110	/* get_link_status is set on LSC (link status) interrupt or
4111	 * rx sequence error interrupt.  get_link_status will stay
4112	 * false until the e1000_check_for_link establishes link
4113	 * for copper adapters ONLY
4114	 */
4115	switch (hw->phy.media_type) {
4116	case e1000_media_type_copper:
4117		if (!hw->mac.get_link_status)
4118			return true;
 
4119	case e1000_media_type_internal_serdes:
4120		hw->mac.ops.check_for_link(hw);
4121		link_active = !hw->mac.get_link_status;
4122		break;
4123	default:
4124	case e1000_media_type_unknown:
4125		break;
4126	}
4127
4128	if (((hw->mac.type == e1000_i210) ||
4129	     (hw->mac.type == e1000_i211)) &&
4130	     (hw->phy.id == I210_I_PHY_ID)) {
4131		if (!netif_carrier_ok(adapter->netdev)) {
4132			adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
4133		} else if (!(adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)) {
4134			adapter->flags |= IGB_FLAG_NEED_LINK_UPDATE;
4135			adapter->link_check_timeout = jiffies;
4136		}
4137	}
4138
4139	return link_active;
4140}
4141
4142static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
4143{
4144	bool ret = false;
4145	u32 ctrl_ext, thstat;
4146
4147	/* check for thermal sensor event on i350 copper only */
4148	if (hw->mac.type == e1000_i350) {
4149		thstat = rd32(E1000_THSTAT);
4150		ctrl_ext = rd32(E1000_CTRL_EXT);
4151
4152		if ((hw->phy.media_type == e1000_media_type_copper) &&
4153		    !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII))
4154			ret = !!(thstat & event);
4155	}
4156
4157	return ret;
4158}
4159
4160/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4161 *  igb_watchdog - Timer Call-back
4162 *  @data: pointer to adapter cast into an unsigned long
4163 **/
4164static void igb_watchdog(unsigned long data)
4165{
4166	struct igb_adapter *adapter = (struct igb_adapter *)data;
4167	/* Do the rest outside of interrupt context */
4168	schedule_work(&adapter->watchdog_task);
4169}
4170
4171static void igb_watchdog_task(struct work_struct *work)
4172{
4173	struct igb_adapter *adapter = container_of(work,
4174						   struct igb_adapter,
4175						   watchdog_task);
4176	struct e1000_hw *hw = &adapter->hw;
4177	struct e1000_phy_info *phy = &hw->phy;
4178	struct net_device *netdev = adapter->netdev;
4179	u32 link;
4180	int i;
4181	u32 connsw;
 
4182
4183	link = igb_has_link(adapter);
4184
4185	if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) {
4186		if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
4187			adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
4188		else
4189			link = false;
4190	}
4191
4192	/* Force link down if we have fiber to swap to */
4193	if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
4194		if (hw->phy.media_type == e1000_media_type_copper) {
4195			connsw = rd32(E1000_CONNSW);
4196			if (!(connsw & E1000_CONNSW_AUTOSENSE_EN))
4197				link = 0;
4198		}
4199	}
4200	if (link) {
4201		/* Perform a reset if the media type changed. */
4202		if (hw->dev_spec._82575.media_changed) {
4203			hw->dev_spec._82575.media_changed = false;
4204			adapter->flags |= IGB_FLAG_MEDIA_RESET;
4205			igb_reset(adapter);
4206		}
4207		/* Cancel scheduled suspend requests. */
4208		pm_runtime_resume(netdev->dev.parent);
4209
4210		if (!netif_carrier_ok(netdev)) {
4211			u32 ctrl;
 
4212			hw->mac.ops.get_speed_and_duplex(hw,
4213							 &adapter->link_speed,
4214							 &adapter->link_duplex);
4215
4216			ctrl = rd32(E1000_CTRL);
4217			/* Links status message must follow this format */
4218			printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s "
4219			       "Duplex, Flow Control: %s\n",
4220			       netdev->name,
4221			       adapter->link_speed,
4222			       adapter->link_duplex == FULL_DUPLEX ?
4223			       "Full" : "Half",
4224			       (ctrl & E1000_CTRL_TFCE) &&
4225			       (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
4226			       (ctrl & E1000_CTRL_RFCE) ?  "RX" :
4227			       (ctrl & E1000_CTRL_TFCE) ?  "TX" : "None");
4228
4229			/* disable EEE if enabled */
4230			if ((adapter->flags & IGB_FLAG_EEE) &&
4231				(adapter->link_duplex == HALF_DUPLEX)) {
4232				dev_info(&adapter->pdev->dev,
4233				"EEE Disabled: unsupported at half duplex. Re-enable using ethtool when at full duplex.\n");
4234				adapter->hw.dev_spec._82575.eee_disable = true;
4235				adapter->flags &= ~IGB_FLAG_EEE;
4236			}
4237
4238			/* check if SmartSpeed worked */
4239			igb_check_downshift(hw);
4240			if (phy->speed_downgraded)
4241				netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
4242
4243			/* check for thermal sensor event */
4244			if (igb_thermal_sensor_event(hw,
4245			    E1000_THSTAT_LINK_THROTTLE)) {
4246				netdev_info(netdev, "The network adapter link "
4247					    "speed was downshifted because it "
4248					    "overheated\n");
4249			}
4250
4251			/* adjust timeout factor according to speed/duplex */
4252			adapter->tx_timeout_factor = 1;
4253			switch (adapter->link_speed) {
4254			case SPEED_10:
4255				adapter->tx_timeout_factor = 14;
4256				break;
4257			case SPEED_100:
4258				/* maybe add some timeout factor ? */
4259				break;
4260			}
4261
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4262			netif_carrier_on(netdev);
4263
4264			igb_ping_all_vfs(adapter);
4265			igb_check_vf_rate_limit(adapter);
4266
4267			/* link state has changed, schedule phy info update */
4268			if (!test_bit(__IGB_DOWN, &adapter->state))
4269				mod_timer(&adapter->phy_info_timer,
4270					  round_jiffies(jiffies + 2 * HZ));
4271		}
4272	} else {
4273		if (netif_carrier_ok(netdev)) {
4274			adapter->link_speed = 0;
4275			adapter->link_duplex = 0;
4276
4277			/* check for thermal sensor event */
4278			if (igb_thermal_sensor_event(hw,
4279			    E1000_THSTAT_PWR_DOWN)) {
4280				netdev_err(netdev, "The network adapter was "
4281					   "stopped because it overheated\n");
4282			}
4283
4284			/* Links status message must follow this format */
4285			printk(KERN_INFO "igb: %s NIC Link is Down\n",
4286			       netdev->name);
4287			netif_carrier_off(netdev);
4288
4289			igb_ping_all_vfs(adapter);
4290
4291			/* link state has changed, schedule phy info update */
4292			if (!test_bit(__IGB_DOWN, &adapter->state))
4293				mod_timer(&adapter->phy_info_timer,
4294					  round_jiffies(jiffies + 2 * HZ));
4295
4296			/* link is down, time to check for alternate media */
4297			if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
4298				igb_check_swap_media(adapter);
4299				if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
4300					schedule_work(&adapter->reset_task);
4301					/* return immediately */
4302					return;
4303				}
4304			}
4305			pm_schedule_suspend(netdev->dev.parent,
4306					    MSEC_PER_SEC * 5);
4307
4308		/* also check for alternate media here */
4309		} else if (!netif_carrier_ok(netdev) &&
4310			   (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
4311			igb_check_swap_media(adapter);
4312			if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
4313				schedule_work(&adapter->reset_task);
4314				/* return immediately */
4315				return;
4316			}
4317		}
4318	}
4319
4320	spin_lock(&adapter->stats64_lock);
4321	igb_update_stats(adapter, &adapter->stats64);
4322	spin_unlock(&adapter->stats64_lock);
4323
4324	for (i = 0; i < adapter->num_tx_queues; i++) {
4325		struct igb_ring *tx_ring = adapter->tx_ring[i];
4326		if (!netif_carrier_ok(netdev)) {
4327			/* We've lost link, so the controller stops DMA,
4328			 * but we've got queued Tx work that's never going
4329			 * to get done, so reset controller to flush Tx.
4330			 * (Do the reset outside of interrupt context).
4331			 */
4332			if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
4333				adapter->tx_timeout_count++;
4334				schedule_work(&adapter->reset_task);
4335				/* return immediately since reset is imminent */
4336				return;
4337			}
4338		}
4339
4340		/* Force detection of hung controller every watchdog period */
4341		set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
4342	}
4343
4344	/* Cause software interrupt to ensure Rx ring is cleaned */
4345	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
4346		u32 eics = 0;
 
4347		for (i = 0; i < adapter->num_q_vectors; i++)
4348			eics |= adapter->q_vector[i]->eims_value;
4349		wr32(E1000_EICS, eics);
4350	} else {
4351		wr32(E1000_ICS, E1000_ICS_RXDMT0);
4352	}
4353
4354	igb_spoof_check(adapter);
4355	igb_ptp_rx_hang(adapter);
 
 
 
 
 
 
4356
4357	/* Reset the timer */
4358	if (!test_bit(__IGB_DOWN, &adapter->state)) {
4359		if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)
4360			mod_timer(&adapter->watchdog_timer,
4361				  round_jiffies(jiffies +  HZ));
4362		else
4363			mod_timer(&adapter->watchdog_timer,
4364				  round_jiffies(jiffies + 2 * HZ));
4365	}
4366}
4367
4368enum latency_range {
4369	lowest_latency = 0,
4370	low_latency = 1,
4371	bulk_latency = 2,
4372	latency_invalid = 255
4373};
4374
4375/**
4376 *  igb_update_ring_itr - update the dynamic ITR value based on packet size
4377 *  @q_vector: pointer to q_vector
4378 *
4379 *  Stores a new ITR value based on strictly on packet size.  This
4380 *  algorithm is less sophisticated than that used in igb_update_itr,
4381 *  due to the difficulty of synchronizing statistics across multiple
4382 *  receive rings.  The divisors and thresholds used by this function
4383 *  were determined based on theoretical maximum wire speed and testing
4384 *  data, in order to minimize response time while increasing bulk
4385 *  throughput.
4386 *  This functionality is controlled by ethtool's coalescing settings.
4387 *  NOTE:  This function is called only when operating in a multiqueue
4388 *         receive environment.
4389 **/
4390static void igb_update_ring_itr(struct igb_q_vector *q_vector)
4391{
4392	int new_val = q_vector->itr_val;
4393	int avg_wire_size = 0;
4394	struct igb_adapter *adapter = q_vector->adapter;
4395	unsigned int packets;
4396
4397	/* For non-gigabit speeds, just fix the interrupt rate at 4000
4398	 * ints/sec - ITR timer value of 120 ticks.
4399	 */
4400	if (adapter->link_speed != SPEED_1000) {
4401		new_val = IGB_4K_ITR;
4402		goto set_itr_val;
4403	}
4404
4405	packets = q_vector->rx.total_packets;
4406	if (packets)
4407		avg_wire_size = q_vector->rx.total_bytes / packets;
4408
4409	packets = q_vector->tx.total_packets;
4410	if (packets)
4411		avg_wire_size = max_t(u32, avg_wire_size,
4412				      q_vector->tx.total_bytes / packets);
4413
4414	/* if avg_wire_size isn't set no work was done */
4415	if (!avg_wire_size)
4416		goto clear_counts;
4417
4418	/* Add 24 bytes to size to account for CRC, preamble, and gap */
4419	avg_wire_size += 24;
4420
4421	/* Don't starve jumbo frames */
4422	avg_wire_size = min(avg_wire_size, 3000);
4423
4424	/* Give a little boost to mid-size frames */
4425	if ((avg_wire_size > 300) && (avg_wire_size < 1200))
4426		new_val = avg_wire_size / 3;
4427	else
4428		new_val = avg_wire_size / 2;
4429
4430	/* conservative mode (itr 3) eliminates the lowest_latency setting */
4431	if (new_val < IGB_20K_ITR &&
4432	    ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4433	     (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4434		new_val = IGB_20K_ITR;
4435
4436set_itr_val:
4437	if (new_val != q_vector->itr_val) {
4438		q_vector->itr_val = new_val;
4439		q_vector->set_itr = 1;
4440	}
4441clear_counts:
4442	q_vector->rx.total_bytes = 0;
4443	q_vector->rx.total_packets = 0;
4444	q_vector->tx.total_bytes = 0;
4445	q_vector->tx.total_packets = 0;
4446}
4447
4448/**
4449 *  igb_update_itr - update the dynamic ITR value based on statistics
4450 *  @q_vector: pointer to q_vector
4451 *  @ring_container: ring info to update the itr for
4452 *
4453 *  Stores a new ITR value based on packets and byte
4454 *  counts during the last interrupt.  The advantage of per interrupt
4455 *  computation is faster updates and more accurate ITR for the current
4456 *  traffic pattern.  Constants in this function were computed
4457 *  based on theoretical maximum wire speed and thresholds were set based
4458 *  on testing data as well as attempting to minimize response time
4459 *  while increasing bulk throughput.
4460 *  This functionality is controlled by ethtool's coalescing settings.
4461 *  NOTE:  These calculations are only valid when operating in a single-
4462 *         queue environment.
4463 **/
4464static void igb_update_itr(struct igb_q_vector *q_vector,
4465			   struct igb_ring_container *ring_container)
4466{
4467	unsigned int packets = ring_container->total_packets;
4468	unsigned int bytes = ring_container->total_bytes;
4469	u8 itrval = ring_container->itr;
4470
4471	/* no packets, exit with status unchanged */
4472	if (packets == 0)
4473		return;
4474
4475	switch (itrval) {
4476	case lowest_latency:
4477		/* handle TSO and jumbo frames */
4478		if (bytes/packets > 8000)
4479			itrval = bulk_latency;
4480		else if ((packets < 5) && (bytes > 512))
4481			itrval = low_latency;
4482		break;
4483	case low_latency:  /* 50 usec aka 20000 ints/s */
4484		if (bytes > 10000) {
4485			/* this if handles the TSO accounting */
4486			if (bytes/packets > 8000) {
4487				itrval = bulk_latency;
4488			} else if ((packets < 10) || ((bytes/packets) > 1200)) {
4489				itrval = bulk_latency;
4490			} else if ((packets > 35)) {
4491				itrval = lowest_latency;
4492			}
4493		} else if (bytes/packets > 2000) {
4494			itrval = bulk_latency;
4495		} else if (packets <= 2 && bytes < 512) {
4496			itrval = lowest_latency;
4497		}
4498		break;
4499	case bulk_latency: /* 250 usec aka 4000 ints/s */
4500		if (bytes > 25000) {
4501			if (packets > 35)
4502				itrval = low_latency;
4503		} else if (bytes < 1500) {
4504			itrval = low_latency;
4505		}
4506		break;
4507	}
4508
4509	/* clear work counters since we have the values we need */
4510	ring_container->total_bytes = 0;
4511	ring_container->total_packets = 0;
4512
4513	/* write updated itr to ring container */
4514	ring_container->itr = itrval;
4515}
4516
4517static void igb_set_itr(struct igb_q_vector *q_vector)
4518{
4519	struct igb_adapter *adapter = q_vector->adapter;
4520	u32 new_itr = q_vector->itr_val;
4521	u8 current_itr = 0;
4522
4523	/* for non-gigabit speeds, just fix the interrupt rate at 4000 */
4524	if (adapter->link_speed != SPEED_1000) {
4525		current_itr = 0;
4526		new_itr = IGB_4K_ITR;
4527		goto set_itr_now;
4528	}
4529
4530	igb_update_itr(q_vector, &q_vector->tx);
4531	igb_update_itr(q_vector, &q_vector->rx);
4532
4533	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
4534
4535	/* conservative mode (itr 3) eliminates the lowest_latency setting */
4536	if (current_itr == lowest_latency &&
4537	    ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4538	     (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4539		current_itr = low_latency;
4540
4541	switch (current_itr) {
4542	/* counts and packets in update_itr are dependent on these numbers */
4543	case lowest_latency:
4544		new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
4545		break;
4546	case low_latency:
4547		new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
4548		break;
4549	case bulk_latency:
4550		new_itr = IGB_4K_ITR;  /* 4,000 ints/sec */
4551		break;
4552	default:
4553		break;
4554	}
4555
4556set_itr_now:
4557	if (new_itr != q_vector->itr_val) {
4558		/* this attempts to bias the interrupt rate towards Bulk
4559		 * by adding intermediate steps when interrupt rate is
4560		 * increasing
4561		 */
4562		new_itr = new_itr > q_vector->itr_val ?
4563			  max((new_itr * q_vector->itr_val) /
4564			  (new_itr + (q_vector->itr_val >> 2)),
4565			  new_itr) : new_itr;
4566		/* Don't write the value here; it resets the adapter's
4567		 * internal timer, and causes us to delay far longer than
4568		 * we should between interrupts.  Instead, we write the ITR
4569		 * value at the beginning of the next interrupt so the timing
4570		 * ends up being correct.
4571		 */
4572		q_vector->itr_val = new_itr;
4573		q_vector->set_itr = 1;
4574	}
4575}
4576
4577static void igb_tx_ctxtdesc(struct igb_ring *tx_ring, u32 vlan_macip_lens,
4578			    u32 type_tucmd, u32 mss_l4len_idx)
 
 
4579{
4580	struct e1000_adv_tx_context_desc *context_desc;
4581	u16 i = tx_ring->next_to_use;
 
4582
4583	context_desc = IGB_TX_CTXTDESC(tx_ring, i);
4584
4585	i++;
4586	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
4587
4588	/* set bits to identify this as an advanced context descriptor */
4589	type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
4590
4591	/* For 82575, context index must be unique per ring. */
4592	if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4593		mss_l4len_idx |= tx_ring->reg_idx << 4;
4594
4595	context_desc->vlan_macip_lens	= cpu_to_le32(vlan_macip_lens);
4596	context_desc->seqnum_seed	= 0;
4597	context_desc->type_tucmd_mlhl	= cpu_to_le32(type_tucmd);
4598	context_desc->mss_l4len_idx	= cpu_to_le32(mss_l4len_idx);
 
 
 
 
 
 
 
 
 
 
 
4599}
4600
4601static int igb_tso(struct igb_ring *tx_ring,
4602		   struct igb_tx_buffer *first,
4603		   u8 *hdr_len)
4604{
 
4605	struct sk_buff *skb = first->skb;
4606	u32 vlan_macip_lens, type_tucmd;
4607	u32 mss_l4len_idx, l4len;
 
 
 
 
 
 
 
 
 
4608	int err;
4609
4610	if (skb->ip_summed != CHECKSUM_PARTIAL)
4611		return 0;
4612
4613	if (!skb_is_gso(skb))
4614		return 0;
4615
4616	err = skb_cow_head(skb, 0);
4617	if (err < 0)
4618		return err;
4619
 
 
 
4620	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4621	type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
 
4622
4623	if (first->protocol == htons(ETH_P_IP)) {
4624		struct iphdr *iph = ip_hdr(skb);
4625		iph->tot_len = 0;
4626		iph->check = 0;
4627		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
4628							 iph->daddr, 0,
4629							 IPPROTO_TCP,
4630							 0);
 
 
 
4631		type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
 
 
4632		first->tx_flags |= IGB_TX_FLAGS_TSO |
4633				   IGB_TX_FLAGS_CSUM |
4634				   IGB_TX_FLAGS_IPV4;
4635	} else if (skb_is_gso_v6(skb)) {
4636		ipv6_hdr(skb)->payload_len = 0;
4637		tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4638						       &ipv6_hdr(skb)->daddr,
4639						       0, IPPROTO_TCP, 0);
4640		first->tx_flags |= IGB_TX_FLAGS_TSO |
4641				   IGB_TX_FLAGS_CSUM;
4642	}
4643
4644	/* compute header lengths */
4645	l4len = tcp_hdrlen(skb);
4646	*hdr_len = skb_transport_offset(skb) + l4len;
 
 
 
 
 
 
 
 
 
 
 
 
 
4647
4648	/* update gso size and bytecount with header size */
4649	first->gso_segs = skb_shinfo(skb)->gso_segs;
4650	first->bytecount += (first->gso_segs - 1) * *hdr_len;
4651
4652	/* MSS L4LEN IDX */
4653	mss_l4len_idx = l4len << E1000_ADVTXD_L4LEN_SHIFT;
4654	mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
4655
4656	/* VLAN MACLEN IPLEN */
4657	vlan_macip_lens = skb_network_header_len(skb);
4658	vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4659	vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4660
4661	igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
 
4662
4663	return 1;
4664}
4665
4666static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
4667{
4668	struct sk_buff *skb = first->skb;
4669	u32 vlan_macip_lens = 0;
4670	u32 mss_l4len_idx = 0;
4671	u32 type_tucmd = 0;
4672
4673	if (skb->ip_summed != CHECKSUM_PARTIAL) {
4674		if (!(first->tx_flags & IGB_TX_FLAGS_VLAN))
 
 
4675			return;
4676	} else {
4677		u8 l4_hdr = 0;
4678		switch (first->protocol) {
4679		case htons(ETH_P_IP):
4680			vlan_macip_lens |= skb_network_header_len(skb);
4681			type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4682			l4_hdr = ip_hdr(skb)->protocol;
4683			break;
4684		case htons(ETH_P_IPV6):
4685			vlan_macip_lens |= skb_network_header_len(skb);
4686			l4_hdr = ipv6_hdr(skb)->nexthdr;
4687			break;
4688		default:
4689			if (unlikely(net_ratelimit())) {
4690				dev_warn(tx_ring->dev,
4691					 "partial checksum but proto=%x!\n",
4692					 first->protocol);
4693			}
4694			break;
4695		}
4696
4697		switch (l4_hdr) {
4698		case IPPROTO_TCP:
4699			type_tucmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
4700			mss_l4len_idx = tcp_hdrlen(skb) <<
4701					E1000_ADVTXD_L4LEN_SHIFT;
4702			break;
4703		case IPPROTO_SCTP:
4704			type_tucmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
4705			mss_l4len_idx = sizeof(struct sctphdr) <<
4706					E1000_ADVTXD_L4LEN_SHIFT;
4707			break;
4708		case IPPROTO_UDP:
4709			mss_l4len_idx = sizeof(struct udphdr) <<
4710					E1000_ADVTXD_L4LEN_SHIFT;
4711			break;
4712		default:
4713			if (unlikely(net_ratelimit())) {
4714				dev_warn(tx_ring->dev,
4715					 "partial checksum but l4 proto=%x!\n",
4716					 l4_hdr);
4717			}
4718			break;
4719		}
4720
4721		/* update TX checksum flag */
4722		first->tx_flags |= IGB_TX_FLAGS_CSUM;
 
4723	}
4724
 
 
 
 
 
4725	vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4726	vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4727
4728	igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4729}
4730
4731#define IGB_SET_FLAG(_input, _flag, _result) \
4732	((_flag <= _result) ? \
4733	 ((u32)(_input & _flag) * (_result / _flag)) : \
4734	 ((u32)(_input & _flag) / (_flag / _result)))
4735
4736static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
4737{
4738	/* set type for advanced descriptor with frame checksum insertion */
4739	u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
4740		       E1000_ADVTXD_DCMD_DEXT |
4741		       E1000_ADVTXD_DCMD_IFCS;
4742
4743	/* set HW vlan bit if vlan is present */
4744	cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
4745				 (E1000_ADVTXD_DCMD_VLE));
4746
4747	/* set segmentation bits for TSO */
4748	cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
4749				 (E1000_ADVTXD_DCMD_TSE));
4750
4751	/* set timestamp bit if present */
4752	cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
4753				 (E1000_ADVTXD_MAC_TSTAMP));
4754
4755	/* insert frame checksum */
4756	cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
4757
4758	return cmd_type;
4759}
4760
4761static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
4762				 union e1000_adv_tx_desc *tx_desc,
4763				 u32 tx_flags, unsigned int paylen)
4764{
4765	u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
4766
4767	/* 82575 requires a unique index per ring */
4768	if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4769		olinfo_status |= tx_ring->reg_idx << 4;
4770
4771	/* insert L4 checksum */
4772	olinfo_status |= IGB_SET_FLAG(tx_flags,
4773				      IGB_TX_FLAGS_CSUM,
4774				      (E1000_TXD_POPTS_TXSM << 8));
4775
4776	/* insert IPv4 checksum */
4777	olinfo_status |= IGB_SET_FLAG(tx_flags,
4778				      IGB_TX_FLAGS_IPV4,
4779				      (E1000_TXD_POPTS_IXSM << 8));
4780
4781	tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
4782}
4783
4784static void igb_tx_map(struct igb_ring *tx_ring,
4785		       struct igb_tx_buffer *first,
4786		       const u8 hdr_len)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4787{
4788	struct sk_buff *skb = first->skb;
4789	struct igb_tx_buffer *tx_buffer;
4790	union e1000_adv_tx_desc *tx_desc;
4791	struct skb_frag_struct *frag;
4792	dma_addr_t dma;
4793	unsigned int data_len, size;
4794	u32 tx_flags = first->tx_flags;
4795	u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
4796	u16 i = tx_ring->next_to_use;
4797
4798	tx_desc = IGB_TX_DESC(tx_ring, i);
4799
4800	igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
4801
4802	size = skb_headlen(skb);
4803	data_len = skb->data_len;
4804
4805	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
4806
4807	tx_buffer = first;
4808
4809	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
4810		if (dma_mapping_error(tx_ring->dev, dma))
4811			goto dma_error;
4812
4813		/* record length, and DMA address */
4814		dma_unmap_len_set(tx_buffer, len, size);
4815		dma_unmap_addr_set(tx_buffer, dma, dma);
4816
4817		tx_desc->read.buffer_addr = cpu_to_le64(dma);
4818
4819		while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
4820			tx_desc->read.cmd_type_len =
4821				cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
4822
4823			i++;
4824			tx_desc++;
4825			if (i == tx_ring->count) {
4826				tx_desc = IGB_TX_DESC(tx_ring, 0);
4827				i = 0;
4828			}
4829			tx_desc->read.olinfo_status = 0;
4830
4831			dma += IGB_MAX_DATA_PER_TXD;
4832			size -= IGB_MAX_DATA_PER_TXD;
4833
4834			tx_desc->read.buffer_addr = cpu_to_le64(dma);
4835		}
4836
4837		if (likely(!data_len))
4838			break;
4839
4840		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
4841
4842		i++;
4843		tx_desc++;
4844		if (i == tx_ring->count) {
4845			tx_desc = IGB_TX_DESC(tx_ring, 0);
4846			i = 0;
4847		}
4848		tx_desc->read.olinfo_status = 0;
4849
4850		size = skb_frag_size(frag);
4851		data_len -= size;
4852
4853		dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
4854				       size, DMA_TO_DEVICE);
4855
4856		tx_buffer = &tx_ring->tx_buffer_info[i];
4857	}
4858
4859	/* write last descriptor with RS and EOP bits */
4860	cmd_type |= size | IGB_TXD_DCMD;
4861	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
4862
4863	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
4864
4865	/* set the timestamp */
4866	first->time_stamp = jiffies;
4867
 
 
4868	/* Force memory writes to complete before letting h/w know there
4869	 * are new descriptors to fetch.  (Only applicable for weak-ordered
4870	 * memory model archs, such as IA-64).
4871	 *
4872	 * We also need this memory barrier to make certain all of the
4873	 * status bits have been updated before next_to_watch is written.
4874	 */
4875	wmb();
4876
4877	/* set next_to_watch value indicating a packet is present */
4878	first->next_to_watch = tx_desc;
4879
4880	i++;
4881	if (i == tx_ring->count)
4882		i = 0;
4883
4884	tx_ring->next_to_use = i;
4885
4886	writel(i, tx_ring->tail);
 
4887
4888	/* we need this if more than one processor can write to our tail
4889	 * at a time, it synchronizes IO on IA64/Altix systems
4890	 */
4891	mmiowb();
4892
4893	return;
4894
4895dma_error:
4896	dev_err(tx_ring->dev, "TX DMA map failed\n");
 
4897
4898	/* clear dma mappings for failed tx_buffer_info map */
4899	for (;;) {
 
 
 
 
 
 
 
 
 
4900		tx_buffer = &tx_ring->tx_buffer_info[i];
4901		igb_unmap_and_free_tx_resource(tx_ring, tx_buffer);
4902		if (tx_buffer == first)
4903			break;
4904		if (i == 0)
4905			i = tx_ring->count;
4906		i--;
4907	}
4908
 
 
 
 
 
 
 
 
 
 
4909	tx_ring->next_to_use = i;
 
 
4910}
4911
4912static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4913{
4914	struct net_device *netdev = tx_ring->netdev;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4915
4916	netif_stop_subqueue(netdev, tx_ring->queue_index);
 
 
4917
4918	/* Herbert's original patch had:
4919	 *  smp_mb__after_netif_stop_queue();
4920	 * but since that doesn't exist yet, just open code it.
4921	 */
4922	smp_mb();
4923
4924	/* We need to check again in a case another CPU has just
4925	 * made room available.
4926	 */
4927	if (igb_desc_unused(tx_ring) < size)
4928		return -EBUSY;
4929
4930	/* A reprieve! */
4931	netif_wake_subqueue(netdev, tx_ring->queue_index);
4932
4933	u64_stats_update_begin(&tx_ring->tx_syncp2);
4934	tx_ring->tx_stats.restart_queue2++;
4935	u64_stats_update_end(&tx_ring->tx_syncp2);
 
 
 
 
 
 
 
 
4936
4937	return 0;
4938}
 
 
4939
4940static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4941{
4942	if (igb_desc_unused(tx_ring) >= size)
4943		return 0;
4944	return __igb_maybe_stop_tx(tx_ring, size);
4945}
4946
4947netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
4948				struct igb_ring *tx_ring)
4949{
4950	struct igb_tx_buffer *first;
4951	int tso;
4952	u32 tx_flags = 0;
 
4953	u16 count = TXD_USE_COUNT(skb_headlen(skb));
4954	__be16 protocol = vlan_get_protocol(skb);
4955	u8 hdr_len = 0;
4956
4957	/* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
4958	 *       + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
4959	 *       + 2 desc gap to keep tail from touching head,
4960	 *       + 1 desc for context descriptor,
4961	 * otherwise try next time
4962	 */
4963	if (NETDEV_FRAG_PAGE_MAX_SIZE > IGB_MAX_DATA_PER_TXD) {
4964		unsigned short f;
4965		for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
4966			count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
4967	} else {
4968		count += skb_shinfo(skb)->nr_frags;
4969	}
4970
4971	if (igb_maybe_stop_tx(tx_ring, count + 3)) {
4972		/* this is a hard error */
4973		return NETDEV_TX_BUSY;
4974	}
4975
4976	/* record the location of the first descriptor for this packet */
4977	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
 
4978	first->skb = skb;
4979	first->bytecount = skb->len;
4980	first->gso_segs = 1;
4981
4982	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
4983		struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
4984
4985		if (!test_and_set_bit_lock(__IGB_PTP_TX_IN_PROGRESS,
 
4986					   &adapter->state)) {
4987			skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
4988			tx_flags |= IGB_TX_FLAGS_TSTAMP;
4989
4990			adapter->ptp_tx_skb = skb_get(skb);
4991			adapter->ptp_tx_start = jiffies;
4992			if (adapter->hw.mac.type == e1000_82576)
4993				schedule_work(&adapter->ptp_tx_work);
 
 
4994		}
4995	}
4996
4997	skb_tx_timestamp(skb);
4998
4999	if (vlan_tx_tag_present(skb)) {
5000		tx_flags |= IGB_TX_FLAGS_VLAN;
5001		tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
5002	}
5003
5004	/* record initial flags and protocol */
5005	first->tx_flags = tx_flags;
5006	first->protocol = protocol;
5007
5008	tso = igb_tso(tx_ring, first, &hdr_len);
5009	if (tso < 0)
5010		goto out_drop;
5011	else if (!tso)
5012		igb_tx_csum(tx_ring, first);
5013
5014	igb_tx_map(tx_ring, first, hdr_len);
5015
5016	/* Make sure there is space in the ring for the next send. */
5017	igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
5018
5019	return NETDEV_TX_OK;
5020
5021out_drop:
5022	igb_unmap_and_free_tx_resource(tx_ring, first);
 
 
 
 
 
 
 
 
 
 
 
5023
5024	return NETDEV_TX_OK;
5025}
5026
5027static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
5028						    struct sk_buff *skb)
5029{
5030	unsigned int r_idx = skb->queue_mapping;
5031
5032	if (r_idx >= adapter->num_tx_queues)
5033		r_idx = r_idx % adapter->num_tx_queues;
5034
5035	return adapter->tx_ring[r_idx];
5036}
5037
5038static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
5039				  struct net_device *netdev)
5040{
5041	struct igb_adapter *adapter = netdev_priv(netdev);
5042
5043	if (test_bit(__IGB_DOWN, &adapter->state)) {
5044		dev_kfree_skb_any(skb);
5045		return NETDEV_TX_OK;
5046	}
5047
5048	if (skb->len <= 0) {
5049		dev_kfree_skb_any(skb);
5050		return NETDEV_TX_OK;
5051	}
5052
5053	/* The minimum packet size with TCTL.PSP set is 17 so pad the skb
5054	 * in order to meet this minimum size requirement.
5055	 */
5056	if (unlikely(skb->len < 17)) {
5057		if (skb_pad(skb, 17 - skb->len))
5058			return NETDEV_TX_OK;
5059		skb->len = 17;
5060		skb_set_tail_pointer(skb, 17);
5061	}
5062
5063	return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
5064}
5065
5066/**
5067 *  igb_tx_timeout - Respond to a Tx Hang
5068 *  @netdev: network interface device structure
 
5069 **/
5070static void igb_tx_timeout(struct net_device *netdev)
5071{
5072	struct igb_adapter *adapter = netdev_priv(netdev);
5073	struct e1000_hw *hw = &adapter->hw;
5074
5075	/* Do the reset outside of interrupt context */
5076	adapter->tx_timeout_count++;
5077
5078	if (hw->mac.type >= e1000_82580)
5079		hw->dev_spec._82575.global_device_reset = true;
5080
5081	schedule_work(&adapter->reset_task);
5082	wr32(E1000_EICS,
5083	     (adapter->eims_enable_mask & ~adapter->eims_other));
5084}
5085
5086static void igb_reset_task(struct work_struct *work)
5087{
5088	struct igb_adapter *adapter;
5089	adapter = container_of(work, struct igb_adapter, reset_task);
5090
 
 
 
 
 
 
 
 
5091	igb_dump(adapter);
5092	netdev_err(adapter->netdev, "Reset adapter\n");
5093	igb_reinit_locked(adapter);
 
5094}
5095
5096/**
5097 *  igb_get_stats64 - Get System Network Statistics
5098 *  @netdev: network interface device structure
5099 *  @stats: rtnl_link_stats64 pointer
5100 **/
5101static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *netdev,
5102						struct rtnl_link_stats64 *stats)
5103{
5104	struct igb_adapter *adapter = netdev_priv(netdev);
5105
5106	spin_lock(&adapter->stats64_lock);
5107	igb_update_stats(adapter, &adapter->stats64);
5108	memcpy(stats, &adapter->stats64, sizeof(*stats));
5109	spin_unlock(&adapter->stats64_lock);
5110
5111	return stats;
5112}
5113
5114/**
5115 *  igb_change_mtu - Change the Maximum Transfer Unit
5116 *  @netdev: network interface device structure
5117 *  @new_mtu: new value for maximum frame size
5118 *
5119 *  Returns 0 on success, negative on failure
5120 **/
5121static int igb_change_mtu(struct net_device *netdev, int new_mtu)
5122{
5123	struct igb_adapter *adapter = netdev_priv(netdev);
5124	struct pci_dev *pdev = adapter->pdev;
5125	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
 
 
5126
5127	if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
5128		dev_err(&pdev->dev, "Invalid MTU setting\n");
5129		return -EINVAL;
5130	}
5131
5132#define MAX_STD_JUMBO_FRAME_SIZE 9238
5133	if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
5134		dev_err(&pdev->dev, "MTU > 9216 not supported.\n");
5135		return -EINVAL;
 
 
 
5136	}
5137
5138	/* adjust max frame to be at least the size of a standard frame */
5139	if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
5140		max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
5141
5142	while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
5143		msleep(1);
5144
5145	/* igb_down has a dependency on max_frame_size */
5146	adapter->max_frame_size = max_frame;
5147
5148	if (netif_running(netdev))
5149		igb_down(adapter);
5150
5151	dev_info(&pdev->dev, "changing MTU from %d to %d\n",
5152		 netdev->mtu, new_mtu);
5153	netdev->mtu = new_mtu;
5154
5155	if (netif_running(netdev))
5156		igb_up(adapter);
5157	else
5158		igb_reset(adapter);
5159
5160	clear_bit(__IGB_RESETTING, &adapter->state);
5161
5162	return 0;
5163}
5164
5165/**
5166 *  igb_update_stats - Update the board statistics counters
5167 *  @adapter: board private structure
5168 **/
5169void igb_update_stats(struct igb_adapter *adapter,
5170		      struct rtnl_link_stats64 *net_stats)
5171{
 
5172	struct e1000_hw *hw = &adapter->hw;
5173	struct pci_dev *pdev = adapter->pdev;
5174	u32 reg, mpc;
5175	u16 phy_tmp;
5176	int i;
5177	u64 bytes, packets;
5178	unsigned int start;
5179	u64 _bytes, _packets;
5180
5181#define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
5182
5183	/* Prevent stats update while adapter is being reset, or if the pci
5184	 * connection is down.
5185	 */
5186	if (adapter->link_speed == 0)
5187		return;
5188	if (pci_channel_offline(pdev))
5189		return;
5190
5191	bytes = 0;
5192	packets = 0;
5193
5194	rcu_read_lock();
5195	for (i = 0; i < adapter->num_rx_queues; i++) {
5196		struct igb_ring *ring = adapter->rx_ring[i];
5197		u32 rqdpc = rd32(E1000_RQDPC(i));
5198		if (hw->mac.type >= e1000_i210)
5199			wr32(E1000_RQDPC(i), 0);
5200
5201		if (rqdpc) {
5202			ring->rx_stats.drops += rqdpc;
5203			net_stats->rx_fifo_errors += rqdpc;
5204		}
5205
5206		do {
5207			start = u64_stats_fetch_begin_irq(&ring->rx_syncp);
5208			_bytes = ring->rx_stats.bytes;
5209			_packets = ring->rx_stats.packets;
5210		} while (u64_stats_fetch_retry_irq(&ring->rx_syncp, start));
5211		bytes += _bytes;
5212		packets += _packets;
5213	}
5214
5215	net_stats->rx_bytes = bytes;
5216	net_stats->rx_packets = packets;
5217
5218	bytes = 0;
5219	packets = 0;
5220	for (i = 0; i < adapter->num_tx_queues; i++) {
5221		struct igb_ring *ring = adapter->tx_ring[i];
5222		do {
5223			start = u64_stats_fetch_begin_irq(&ring->tx_syncp);
5224			_bytes = ring->tx_stats.bytes;
5225			_packets = ring->tx_stats.packets;
5226		} while (u64_stats_fetch_retry_irq(&ring->tx_syncp, start));
5227		bytes += _bytes;
5228		packets += _packets;
5229	}
5230	net_stats->tx_bytes = bytes;
5231	net_stats->tx_packets = packets;
5232	rcu_read_unlock();
5233
5234	/* read stats registers */
5235	adapter->stats.crcerrs += rd32(E1000_CRCERRS);
5236	adapter->stats.gprc += rd32(E1000_GPRC);
5237	adapter->stats.gorc += rd32(E1000_GORCL);
5238	rd32(E1000_GORCH); /* clear GORCL */
5239	adapter->stats.bprc += rd32(E1000_BPRC);
5240	adapter->stats.mprc += rd32(E1000_MPRC);
5241	adapter->stats.roc += rd32(E1000_ROC);
5242
5243	adapter->stats.prc64 += rd32(E1000_PRC64);
5244	adapter->stats.prc127 += rd32(E1000_PRC127);
5245	adapter->stats.prc255 += rd32(E1000_PRC255);
5246	adapter->stats.prc511 += rd32(E1000_PRC511);
5247	adapter->stats.prc1023 += rd32(E1000_PRC1023);
5248	adapter->stats.prc1522 += rd32(E1000_PRC1522);
5249	adapter->stats.symerrs += rd32(E1000_SYMERRS);
5250	adapter->stats.sec += rd32(E1000_SEC);
5251
5252	mpc = rd32(E1000_MPC);
5253	adapter->stats.mpc += mpc;
5254	net_stats->rx_fifo_errors += mpc;
5255	adapter->stats.scc += rd32(E1000_SCC);
5256	adapter->stats.ecol += rd32(E1000_ECOL);
5257	adapter->stats.mcc += rd32(E1000_MCC);
5258	adapter->stats.latecol += rd32(E1000_LATECOL);
5259	adapter->stats.dc += rd32(E1000_DC);
5260	adapter->stats.rlec += rd32(E1000_RLEC);
5261	adapter->stats.xonrxc += rd32(E1000_XONRXC);
5262	adapter->stats.xontxc += rd32(E1000_XONTXC);
5263	adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
5264	adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
5265	adapter->stats.fcruc += rd32(E1000_FCRUC);
5266	adapter->stats.gptc += rd32(E1000_GPTC);
5267	adapter->stats.gotc += rd32(E1000_GOTCL);
5268	rd32(E1000_GOTCH); /* clear GOTCL */
5269	adapter->stats.rnbc += rd32(E1000_RNBC);
5270	adapter->stats.ruc += rd32(E1000_RUC);
5271	adapter->stats.rfc += rd32(E1000_RFC);
5272	adapter->stats.rjc += rd32(E1000_RJC);
5273	adapter->stats.tor += rd32(E1000_TORH);
5274	adapter->stats.tot += rd32(E1000_TOTH);
5275	adapter->stats.tpr += rd32(E1000_TPR);
5276
5277	adapter->stats.ptc64 += rd32(E1000_PTC64);
5278	adapter->stats.ptc127 += rd32(E1000_PTC127);
5279	adapter->stats.ptc255 += rd32(E1000_PTC255);
5280	adapter->stats.ptc511 += rd32(E1000_PTC511);
5281	adapter->stats.ptc1023 += rd32(E1000_PTC1023);
5282	adapter->stats.ptc1522 += rd32(E1000_PTC1522);
5283
5284	adapter->stats.mptc += rd32(E1000_MPTC);
5285	adapter->stats.bptc += rd32(E1000_BPTC);
5286
5287	adapter->stats.tpt += rd32(E1000_TPT);
5288	adapter->stats.colc += rd32(E1000_COLC);
5289
5290	adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
5291	/* read internal phy specific stats */
5292	reg = rd32(E1000_CTRL_EXT);
5293	if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
5294		adapter->stats.rxerrc += rd32(E1000_RXERRC);
5295
5296		/* this stat has invalid values on i210/i211 */
5297		if ((hw->mac.type != e1000_i210) &&
5298		    (hw->mac.type != e1000_i211))
5299			adapter->stats.tncrs += rd32(E1000_TNCRS);
5300	}
5301
5302	adapter->stats.tsctc += rd32(E1000_TSCTC);
5303	adapter->stats.tsctfc += rd32(E1000_TSCTFC);
5304
5305	adapter->stats.iac += rd32(E1000_IAC);
5306	adapter->stats.icrxoc += rd32(E1000_ICRXOC);
5307	adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
5308	adapter->stats.icrxatc += rd32(E1000_ICRXATC);
5309	adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
5310	adapter->stats.ictxatc += rd32(E1000_ICTXATC);
5311	adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
5312	adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
5313	adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
5314
5315	/* Fill out the OS statistics structure */
5316	net_stats->multicast = adapter->stats.mprc;
5317	net_stats->collisions = adapter->stats.colc;
5318
5319	/* Rx Errors */
5320
5321	/* RLEC on some newer hardware can be incorrect so build
5322	 * our own version based on RUC and ROC
5323	 */
5324	net_stats->rx_errors = adapter->stats.rxerrc +
5325		adapter->stats.crcerrs + adapter->stats.algnerrc +
5326		adapter->stats.ruc + adapter->stats.roc +
5327		adapter->stats.cexterr;
5328	net_stats->rx_length_errors = adapter->stats.ruc +
5329				      adapter->stats.roc;
5330	net_stats->rx_crc_errors = adapter->stats.crcerrs;
5331	net_stats->rx_frame_errors = adapter->stats.algnerrc;
5332	net_stats->rx_missed_errors = adapter->stats.mpc;
5333
5334	/* Tx Errors */
5335	net_stats->tx_errors = adapter->stats.ecol +
5336			       adapter->stats.latecol;
5337	net_stats->tx_aborted_errors = adapter->stats.ecol;
5338	net_stats->tx_window_errors = adapter->stats.latecol;
5339	net_stats->tx_carrier_errors = adapter->stats.tncrs;
5340
5341	/* Tx Dropped needs to be maintained elsewhere */
5342
5343	/* Phy Stats */
5344	if (hw->phy.media_type == e1000_media_type_copper) {
5345		if ((adapter->link_speed == SPEED_1000) &&
5346		   (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
5347			phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
5348			adapter->phy_stats.idle_errors += phy_tmp;
5349		}
5350	}
5351
5352	/* Management Stats */
5353	adapter->stats.mgptc += rd32(E1000_MGTPTC);
5354	adapter->stats.mgprc += rd32(E1000_MGTPRC);
5355	adapter->stats.mgpdc += rd32(E1000_MGTPDC);
5356
5357	/* OS2BMC Stats */
5358	reg = rd32(E1000_MANC);
5359	if (reg & E1000_MANC_EN_BMC2OS) {
5360		adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
5361		adapter->stats.o2bspc += rd32(E1000_O2BSPC);
5362		adapter->stats.b2ospc += rd32(E1000_B2OSPC);
5363		adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
5364	}
5365}
5366
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5367static irqreturn_t igb_msix_other(int irq, void *data)
5368{
5369	struct igb_adapter *adapter = data;
5370	struct e1000_hw *hw = &adapter->hw;
5371	u32 icr = rd32(E1000_ICR);
5372	/* reading ICR causes bit 31 of EICR to be cleared */
5373
5374	if (icr & E1000_ICR_DRSTA)
5375		schedule_work(&adapter->reset_task);
5376
5377	if (icr & E1000_ICR_DOUTSYNC) {
5378		/* HW is reporting DMA is out of sync */
5379		adapter->stats.doosync++;
5380		/* The DMA Out of Sync is also indication of a spoof event
5381		 * in IOV mode. Check the Wrong VM Behavior register to
5382		 * see if it is really a spoof event.
5383		 */
5384		igb_check_wvbr(adapter);
5385	}
5386
5387	/* Check for a mailbox event */
5388	if (icr & E1000_ICR_VMMB)
5389		igb_msg_task(adapter);
5390
5391	if (icr & E1000_ICR_LSC) {
5392		hw->mac.get_link_status = 1;
5393		/* guard against interrupt when we're going down */
5394		if (!test_bit(__IGB_DOWN, &adapter->state))
5395			mod_timer(&adapter->watchdog_timer, jiffies + 1);
5396	}
5397
5398	if (icr & E1000_ICR_TS) {
5399		u32 tsicr = rd32(E1000_TSICR);
5400
5401		if (tsicr & E1000_TSICR_TXTS) {
5402			/* acknowledge the interrupt */
5403			wr32(E1000_TSICR, E1000_TSICR_TXTS);
5404			/* retrieve hardware timestamp */
5405			schedule_work(&adapter->ptp_tx_work);
5406		}
5407	}
5408
5409	wr32(E1000_EIMS, adapter->eims_other);
5410
5411	return IRQ_HANDLED;
5412}
5413
5414static void igb_write_itr(struct igb_q_vector *q_vector)
5415{
5416	struct igb_adapter *adapter = q_vector->adapter;
5417	u32 itr_val = q_vector->itr_val & 0x7FFC;
5418
5419	if (!q_vector->set_itr)
5420		return;
5421
5422	if (!itr_val)
5423		itr_val = 0x4;
5424
5425	if (adapter->hw.mac.type == e1000_82575)
5426		itr_val |= itr_val << 16;
5427	else
5428		itr_val |= E1000_EITR_CNT_IGNR;
5429
5430	writel(itr_val, q_vector->itr_register);
5431	q_vector->set_itr = 0;
5432}
5433
5434static irqreturn_t igb_msix_ring(int irq, void *data)
5435{
5436	struct igb_q_vector *q_vector = data;
5437
5438	/* Write the ITR value calculated from the previous interrupt. */
5439	igb_write_itr(q_vector);
5440
5441	napi_schedule(&q_vector->napi);
5442
5443	return IRQ_HANDLED;
5444}
5445
5446#ifdef CONFIG_IGB_DCA
5447static void igb_update_tx_dca(struct igb_adapter *adapter,
5448			      struct igb_ring *tx_ring,
5449			      int cpu)
5450{
5451	struct e1000_hw *hw = &adapter->hw;
5452	u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
5453
5454	if (hw->mac.type != e1000_82575)
5455		txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
5456
5457	/* We can enable relaxed ordering for reads, but not writes when
5458	 * DCA is enabled.  This is due to a known issue in some chipsets
5459	 * which will cause the DCA tag to be cleared.
5460	 */
5461	txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
5462		  E1000_DCA_TXCTRL_DATA_RRO_EN |
5463		  E1000_DCA_TXCTRL_DESC_DCA_EN;
5464
5465	wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
5466}
5467
5468static void igb_update_rx_dca(struct igb_adapter *adapter,
5469			      struct igb_ring *rx_ring,
5470			      int cpu)
5471{
5472	struct e1000_hw *hw = &adapter->hw;
5473	u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
5474
5475	if (hw->mac.type != e1000_82575)
5476		rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
5477
5478	/* We can enable relaxed ordering for reads, but not writes when
5479	 * DCA is enabled.  This is due to a known issue in some chipsets
5480	 * which will cause the DCA tag to be cleared.
5481	 */
5482	rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
5483		  E1000_DCA_RXCTRL_DESC_DCA_EN;
5484
5485	wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
5486}
5487
5488static void igb_update_dca(struct igb_q_vector *q_vector)
5489{
5490	struct igb_adapter *adapter = q_vector->adapter;
5491	int cpu = get_cpu();
5492
5493	if (q_vector->cpu == cpu)
5494		goto out_no_update;
5495
5496	if (q_vector->tx.ring)
5497		igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
5498
5499	if (q_vector->rx.ring)
5500		igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
5501
5502	q_vector->cpu = cpu;
5503out_no_update:
5504	put_cpu();
5505}
5506
5507static void igb_setup_dca(struct igb_adapter *adapter)
5508{
5509	struct e1000_hw *hw = &adapter->hw;
5510	int i;
5511
5512	if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
5513		return;
5514
5515	/* Always use CB2 mode, difference is masked in the CB driver. */
5516	wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
5517
5518	for (i = 0; i < adapter->num_q_vectors; i++) {
5519		adapter->q_vector[i]->cpu = -1;
5520		igb_update_dca(adapter->q_vector[i]);
5521	}
5522}
5523
5524static int __igb_notify_dca(struct device *dev, void *data)
5525{
5526	struct net_device *netdev = dev_get_drvdata(dev);
5527	struct igb_adapter *adapter = netdev_priv(netdev);
5528	struct pci_dev *pdev = adapter->pdev;
5529	struct e1000_hw *hw = &adapter->hw;
5530	unsigned long event = *(unsigned long *)data;
5531
5532	switch (event) {
5533	case DCA_PROVIDER_ADD:
5534		/* if already enabled, don't do it again */
5535		if (adapter->flags & IGB_FLAG_DCA_ENABLED)
5536			break;
5537		if (dca_add_requester(dev) == 0) {
5538			adapter->flags |= IGB_FLAG_DCA_ENABLED;
5539			dev_info(&pdev->dev, "DCA enabled\n");
5540			igb_setup_dca(adapter);
5541			break;
5542		}
5543		/* Fall Through since DCA is disabled. */
5544	case DCA_PROVIDER_REMOVE:
5545		if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
5546			/* without this a class_device is left
5547			 * hanging around in the sysfs model
5548			 */
5549			dca_remove_requester(dev);
5550			dev_info(&pdev->dev, "DCA disabled\n");
5551			adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
5552			wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
5553		}
5554		break;
5555	}
5556
5557	return 0;
5558}
5559
5560static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
5561			  void *p)
5562{
5563	int ret_val;
5564
5565	ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
5566					 __igb_notify_dca);
5567
5568	return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
5569}
5570#endif /* CONFIG_IGB_DCA */
5571
5572#ifdef CONFIG_PCI_IOV
5573static int igb_vf_configure(struct igb_adapter *adapter, int vf)
5574{
5575	unsigned char mac_addr[ETH_ALEN];
5576
5577	eth_zero_addr(mac_addr);
5578	igb_set_vf_mac(adapter, vf, mac_addr);
5579
5580	/* By default spoof check is enabled for all VFs */
5581	adapter->vf_data[vf].spoofchk_enabled = true;
5582
 
 
 
5583	return 0;
5584}
5585
5586#endif
5587static void igb_ping_all_vfs(struct igb_adapter *adapter)
5588{
5589	struct e1000_hw *hw = &adapter->hw;
5590	u32 ping;
5591	int i;
5592
5593	for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
5594		ping = E1000_PF_CONTROL_MSG;
5595		if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
5596			ping |= E1000_VT_MSGTYPE_CTS;
5597		igb_write_mbx(hw, &ping, 1, i);
5598	}
5599}
5600
5601static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5602{
5603	struct e1000_hw *hw = &adapter->hw;
5604	u32 vmolr = rd32(E1000_VMOLR(vf));
5605	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5606
5607	vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
5608			    IGB_VF_FLAG_MULTI_PROMISC);
5609	vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5610
5611	if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
5612		vmolr |= E1000_VMOLR_MPME;
5613		vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
5614		*msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
5615	} else {
5616		/* if we have hashes and we are clearing a multicast promisc
5617		 * flag we need to write the hashes to the MTA as this step
5618		 * was previously skipped
5619		 */
5620		if (vf_data->num_vf_mc_hashes > 30) {
5621			vmolr |= E1000_VMOLR_MPME;
5622		} else if (vf_data->num_vf_mc_hashes) {
5623			int j;
 
5624			vmolr |= E1000_VMOLR_ROMPE;
5625			for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5626				igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5627		}
5628	}
5629
5630	wr32(E1000_VMOLR(vf), vmolr);
5631
5632	/* there are flags left unprocessed, likely not supported */
5633	if (*msgbuf & E1000_VT_MSGINFO_MASK)
5634		return -EINVAL;
5635
5636	return 0;
5637}
5638
5639static int igb_set_vf_multicasts(struct igb_adapter *adapter,
5640				  u32 *msgbuf, u32 vf)
5641{
5642	int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5643	u16 *hash_list = (u16 *)&msgbuf[1];
5644	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5645	int i;
5646
5647	/* salt away the number of multicast addresses assigned
5648	 * to this VF for later use to restore when the PF multi cast
5649	 * list changes
5650	 */
5651	vf_data->num_vf_mc_hashes = n;
5652
5653	/* only up to 30 hash values supported */
5654	if (n > 30)
5655		n = 30;
5656
5657	/* store the hashes for later use */
5658	for (i = 0; i < n; i++)
5659		vf_data->vf_mc_hashes[i] = hash_list[i];
5660
5661	/* Flush and reset the mta with the new values */
5662	igb_set_rx_mode(adapter->netdev);
5663
5664	return 0;
5665}
5666
5667static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
5668{
5669	struct e1000_hw *hw = &adapter->hw;
5670	struct vf_data_storage *vf_data;
5671	int i, j;
5672
5673	for (i = 0; i < adapter->vfs_allocated_count; i++) {
5674		u32 vmolr = rd32(E1000_VMOLR(i));
 
5675		vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5676
5677		vf_data = &adapter->vf_data[i];
5678
5679		if ((vf_data->num_vf_mc_hashes > 30) ||
5680		    (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
5681			vmolr |= E1000_VMOLR_MPME;
5682		} else if (vf_data->num_vf_mc_hashes) {
5683			vmolr |= E1000_VMOLR_ROMPE;
5684			for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5685				igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5686		}
5687		wr32(E1000_VMOLR(i), vmolr);
5688	}
5689}
5690
5691static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
5692{
5693	struct e1000_hw *hw = &adapter->hw;
5694	u32 pool_mask, reg, vid;
5695	int i;
5696
5697	pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
 
 
 
 
 
 
5698
5699	/* Find the vlan filter for this id */
5700	for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5701		reg = rd32(E1000_VLVF(i));
 
5702
5703		/* remove the vf from the pool */
5704		reg &= ~pool_mask;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5705
5706		/* if pool is empty then remove entry from vfta */
5707		if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
5708		    (reg & E1000_VLVF_VLANID_ENABLE)) {
5709			reg = 0;
5710			vid = reg & E1000_VLVF_VLANID_MASK;
5711			igb_vfta_set(hw, vid, false);
5712		}
5713
5714		wr32(E1000_VLVF(i), reg);
 
 
 
 
5715	}
5716
5717	adapter->vf_data[vf].vlans_enabled = 0;
5718}
5719
5720static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
5721{
5722	struct e1000_hw *hw = &adapter->hw;
5723	u32 reg, i;
 
5724
5725	/* The vlvf table only exists on 82576 hardware and newer */
5726	if (hw->mac.type < e1000_82576)
5727		return -1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5728
5729	/* we only need to do this if VMDq is enabled */
5730	if (!adapter->vfs_allocated_count)
5731		return -1;
 
 
 
5732
5733	/* Find the vlan filter for this id */
5734	for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5735		reg = rd32(E1000_VLVF(i));
5736		if ((reg & E1000_VLVF_VLANID_ENABLE) &&
5737		    vid == (reg & E1000_VLVF_VLANID_MASK))
5738			break;
 
 
 
5739	}
5740
5741	if (add) {
5742		if (i == E1000_VLVF_ARRAY_SIZE) {
5743			/* Did not find a matching VLAN ID entry that was
5744			 * enabled.  Search for a free filter entry, i.e.
5745			 * one without the enable bit set
5746			 */
5747			for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5748				reg = rd32(E1000_VLVF(i));
5749				if (!(reg & E1000_VLVF_VLANID_ENABLE))
5750					break;
5751			}
5752		}
5753		if (i < E1000_VLVF_ARRAY_SIZE) {
5754			/* Found an enabled/available entry */
5755			reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5756
5757			/* if !enabled we need to set this up in vfta */
5758			if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
5759				/* add VID to filter table */
5760				igb_vfta_set(hw, vid, true);
5761				reg |= E1000_VLVF_VLANID_ENABLE;
5762			}
5763			reg &= ~E1000_VLVF_VLANID_MASK;
5764			reg |= vid;
5765			wr32(E1000_VLVF(i), reg);
5766
5767			/* do not modify RLPML for PF devices */
5768			if (vf >= adapter->vfs_allocated_count)
5769				return 0;
5770
5771			if (!adapter->vf_data[vf].vlans_enabled) {
5772				u32 size;
5773				reg = rd32(E1000_VMOLR(vf));
5774				size = reg & E1000_VMOLR_RLPML_MASK;
5775				size += 4;
5776				reg &= ~E1000_VMOLR_RLPML_MASK;
5777				reg |= size;
5778				wr32(E1000_VMOLR(vf), reg);
5779			}
5780
5781			adapter->vf_data[vf].vlans_enabled++;
5782		}
5783	} else {
5784		if (i < E1000_VLVF_ARRAY_SIZE) {
5785			/* remove vf from the pool */
5786			reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
5787			/* if pool is empty then remove entry from vfta */
5788			if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
5789				reg = 0;
5790				igb_vfta_set(hw, vid, false);
5791			}
5792			wr32(E1000_VLVF(i), reg);
5793
5794			/* do not modify RLPML for PF devices */
5795			if (vf >= adapter->vfs_allocated_count)
5796				return 0;
5797
5798			adapter->vf_data[vf].vlans_enabled--;
5799			if (!adapter->vf_data[vf].vlans_enabled) {
5800				u32 size;
5801				reg = rd32(E1000_VMOLR(vf));
5802				size = reg & E1000_VMOLR_RLPML_MASK;
5803				size -= 4;
5804				reg &= ~E1000_VMOLR_RLPML_MASK;
5805				reg |= size;
5806				wr32(E1000_VMOLR(vf), reg);
5807			}
5808		}
5809	}
5810	return 0;
5811}
5812
5813static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
5814{
5815	struct e1000_hw *hw = &adapter->hw;
5816
5817	if (vid)
5818		wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
5819	else
5820		wr32(E1000_VMVIR(vf), 0);
5821}
5822
5823static int igb_ndo_set_vf_vlan(struct net_device *netdev,
5824			       int vf, u16 vlan, u8 qos)
5825{
5826	int err = 0;
5827	struct igb_adapter *adapter = netdev_priv(netdev);
 
 
 
 
 
 
5828
5829	if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
5830		return -EINVAL;
5831	if (vlan || qos) {
5832		err = igb_vlvf_set(adapter, vlan, !!vlan, vf);
5833		if (err)
5834			goto out;
5835		igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
5836		igb_set_vmolr(adapter, vf, !vlan);
5837		adapter->vf_data[vf].pf_vlan = vlan;
5838		adapter->vf_data[vf].pf_qos = qos;
5839		dev_info(&adapter->pdev->dev,
5840			 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
5841		if (test_bit(__IGB_DOWN, &adapter->state)) {
5842			dev_warn(&adapter->pdev->dev,
5843				 "The VF VLAN has been set, but the PF device is not up.\n");
5844			dev_warn(&adapter->pdev->dev,
5845				 "Bring the PF device up before attempting to use the VF device.\n");
5846		}
5847	} else {
5848		igb_vlvf_set(adapter, adapter->vf_data[vf].pf_vlan,
5849			     false, vf);
5850		igb_set_vmvir(adapter, vlan, vf);
5851		igb_set_vmolr(adapter, vf, true);
5852		adapter->vf_data[vf].pf_vlan = 0;
5853		adapter->vf_data[vf].pf_qos = 0;
5854	}
5855out:
5856	return err;
5857}
5858
5859static int igb_find_vlvf_entry(struct igb_adapter *adapter, int vid)
5860{
5861	struct e1000_hw *hw = &adapter->hw;
5862	int i;
5863	u32 reg;
 
 
5864
5865	/* Find the vlan filter for this id */
5866	for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5867		reg = rd32(E1000_VLVF(i));
5868		if ((reg & E1000_VLVF_VLANID_ENABLE) &&
5869		    vid == (reg & E1000_VLVF_VLANID_MASK))
5870			break;
5871	}
5872
5873	if (i >= E1000_VLVF_ARRAY_SIZE)
5874		i = -1;
 
5875
5876	return i;
5877}
5878
5879static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
 
5880{
5881	struct e1000_hw *hw = &adapter->hw;
5882	int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5883	int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
5884	int err = 0;
5885
5886	/* If in promiscuous mode we need to make sure the PF also has
5887	 * the VLAN filter set.
5888	 */
5889	if (add && (adapter->netdev->flags & IFF_PROMISC))
5890		err = igb_vlvf_set(adapter, vid, add,
5891				   adapter->vfs_allocated_count);
5892	if (err)
5893		goto out;
5894
5895	err = igb_vlvf_set(adapter, vid, add, vf);
 
 
5896
5897	if (err)
5898		goto out;
 
 
 
5899
5900	/* Go through all the checks to see if the VLAN filter should
5901	 * be wiped completely.
5902	 */
5903	if (!add && (adapter->netdev->flags & IFF_PROMISC)) {
5904		u32 vlvf, bits;
5905
5906		int regndx = igb_find_vlvf_entry(adapter, vid);
5907		if (regndx < 0)
5908			goto out;
5909		/* See if any other pools are set for this VLAN filter
5910		 * entry other than the PF.
5911		 */
5912		vlvf = bits = rd32(E1000_VLVF(regndx));
5913		bits &= 1 << (E1000_VLVF_POOLSEL_SHIFT +
5914			      adapter->vfs_allocated_count);
5915		/* If the filter was removed then ensure PF pool bit
5916		 * is cleared if the PF only added itself to the pool
5917		 * because the PF is in promiscuous mode.
5918		 */
5919		if ((vlvf & VLAN_VID_MASK) == vid &&
5920		    !test_bit(vid, adapter->active_vlans) &&
5921		    !bits)
5922			igb_vlvf_set(adapter, vid, add,
5923				     adapter->vfs_allocated_count);
5924	}
5925
5926out:
5927	return err;
 
 
5928}
5929
5930static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
5931{
 
 
5932	/* clear flags - except flag that indicates PF has set the MAC */
5933	adapter->vf_data[vf].flags &= IGB_VF_FLAG_PF_SET_MAC;
5934	adapter->vf_data[vf].last_nack = jiffies;
5935
5936	/* reset offloads to defaults */
5937	igb_set_vmolr(adapter, vf, true);
5938
5939	/* reset vlans for device */
5940	igb_clear_vf_vfta(adapter, vf);
5941	if (adapter->vf_data[vf].pf_vlan)
5942		igb_ndo_set_vf_vlan(adapter->netdev, vf,
5943				    adapter->vf_data[vf].pf_vlan,
5944				    adapter->vf_data[vf].pf_qos);
5945	else
5946		igb_clear_vf_vfta(adapter, vf);
5947
5948	/* reset multicast table array for vf */
5949	adapter->vf_data[vf].num_vf_mc_hashes = 0;
5950
5951	/* Flush and reset the mta with the new values */
5952	igb_set_rx_mode(adapter->netdev);
5953}
5954
5955static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
5956{
5957	unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5958
5959	/* clear mac address as we were hotplug removed/added */
5960	if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
5961		eth_zero_addr(vf_mac);
5962
5963	/* process remaining reset events */
5964	igb_vf_reset(adapter, vf);
5965}
5966
5967static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
5968{
5969	struct e1000_hw *hw = &adapter->hw;
5970	unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5971	int rar_entry = hw->mac.rar_entry_count - (vf + 1);
5972	u32 reg, msgbuf[3];
5973	u8 *addr = (u8 *)(&msgbuf[1]);
5974
5975	/* process all the same items cleared in a function level reset */
5976	igb_vf_reset(adapter, vf);
5977
5978	/* set vf mac address */
5979	igb_rar_set_qsel(adapter, vf_mac, rar_entry, vf);
5980
5981	/* enable transmit and receive for vf */
5982	reg = rd32(E1000_VFTE);
5983	wr32(E1000_VFTE, reg | (1 << vf));
5984	reg = rd32(E1000_VFRE);
5985	wr32(E1000_VFRE, reg | (1 << vf));
5986
5987	adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
5988
5989	/* reply to reset with ack and vf mac address */
5990	msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
5991	memcpy(addr, vf_mac, ETH_ALEN);
 
 
 
 
5992	igb_write_mbx(hw, msgbuf, 3, vf);
5993}
5994
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5995static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
5996{
 
 
 
 
5997	/* The VF MAC Address is stored in a packed array of bytes
5998	 * starting at the second 32 bit word of the msg array
5999	 */
6000	unsigned char *addr = (char *)&msg[1];
6001	int err = -1;
6002
6003	if (is_valid_ether_addr(addr))
6004		err = igb_set_vf_mac(adapter, vf, addr);
 
 
 
 
 
 
6005
6006	return err;
 
 
 
 
 
 
 
 
 
 
 
 
6007}
6008
6009static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
6010{
6011	struct e1000_hw *hw = &adapter->hw;
6012	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6013	u32 msg = E1000_VT_MSGTYPE_NACK;
6014
6015	/* if device isn't clear to send it shouldn't be reading either */
6016	if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
6017	    time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
6018		igb_write_mbx(hw, &msg, 1, vf);
6019		vf_data->last_nack = jiffies;
6020	}
6021}
6022
6023static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
6024{
6025	struct pci_dev *pdev = adapter->pdev;
6026	u32 msgbuf[E1000_VFMAILBOX_SIZE];
6027	struct e1000_hw *hw = &adapter->hw;
6028	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6029	s32 retval;
6030
6031	retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf);
6032
6033	if (retval) {
6034		/* if receive failed revoke VF CTS stats and restart init */
6035		dev_err(&pdev->dev, "Error receiving message from VF\n");
6036		vf_data->flags &= ~IGB_VF_FLAG_CTS;
6037		if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
6038			return;
6039		goto out;
6040	}
6041
6042	/* this is a message we already processed, do nothing */
6043	if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
6044		return;
6045
6046	/* until the vf completes a reset it should not be
6047	 * allowed to start any configuration.
6048	 */
6049	if (msgbuf[0] == E1000_VF_RESET) {
 
6050		igb_vf_reset_msg(adapter, vf);
6051		return;
6052	}
6053
6054	if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
6055		if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
6056			return;
6057		retval = -1;
6058		goto out;
6059	}
6060
6061	switch ((msgbuf[0] & 0xFFFF)) {
6062	case E1000_VF_SET_MAC_ADDR:
6063		retval = -EINVAL;
6064		if (!(vf_data->flags & IGB_VF_FLAG_PF_SET_MAC))
6065			retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
6066		else
6067			dev_warn(&pdev->dev,
6068				 "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
6069				 vf);
6070		break;
6071	case E1000_VF_SET_PROMISC:
6072		retval = igb_set_vf_promisc(adapter, msgbuf, vf);
6073		break;
6074	case E1000_VF_SET_MULTICAST:
6075		retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
6076		break;
6077	case E1000_VF_SET_LPE:
6078		retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
6079		break;
6080	case E1000_VF_SET_VLAN:
6081		retval = -1;
6082		if (vf_data->pf_vlan)
6083			dev_warn(&pdev->dev,
6084				 "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
6085				 vf);
6086		else
6087			retval = igb_set_vf_vlan(adapter, msgbuf, vf);
6088		break;
6089	default:
6090		dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
6091		retval = -1;
6092		break;
6093	}
6094
6095	msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
6096out:
6097	/* notify the VF of the results of what it sent us */
6098	if (retval)
6099		msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
6100	else
6101		msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
6102
 
6103	igb_write_mbx(hw, msgbuf, 1, vf);
 
 
 
 
6104}
6105
6106static void igb_msg_task(struct igb_adapter *adapter)
6107{
6108	struct e1000_hw *hw = &adapter->hw;
 
6109	u32 vf;
6110
 
6111	for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
6112		/* process any reset requests */
6113		if (!igb_check_for_rst(hw, vf))
6114			igb_vf_reset_event(adapter, vf);
6115
6116		/* process any messages pending */
6117		if (!igb_check_for_msg(hw, vf))
6118			igb_rcv_msg_from_vf(adapter, vf);
6119
6120		/* process any acks */
6121		if (!igb_check_for_ack(hw, vf))
6122			igb_rcv_ack_from_vf(adapter, vf);
6123	}
 
6124}
6125
6126/**
6127 *  igb_set_uta - Set unicast filter table address
6128 *  @adapter: board private structure
 
6129 *
6130 *  The unicast table address is a register array of 32-bit registers.
6131 *  The table is meant to be used in a way similar to how the MTA is used
6132 *  however due to certain limitations in the hardware it is necessary to
6133 *  set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
6134 *  enable bit to allow vlan tag stripping when promiscuous mode is enabled
6135 **/
6136static void igb_set_uta(struct igb_adapter *adapter)
6137{
6138	struct e1000_hw *hw = &adapter->hw;
 
6139	int i;
6140
6141	/* The UTA table only exists on 82576 hardware and newer */
6142	if (hw->mac.type < e1000_82576)
6143		return;
6144
6145	/* we only need to do this if VMDq is enabled */
6146	if (!adapter->vfs_allocated_count)
6147		return;
6148
6149	for (i = 0; i < hw->mac.uta_reg_count; i++)
6150		array_wr32(E1000_UTA, i, ~0);
6151}
6152
6153/**
6154 *  igb_intr_msi - Interrupt Handler
6155 *  @irq: interrupt number
6156 *  @data: pointer to a network interface device structure
6157 **/
6158static irqreturn_t igb_intr_msi(int irq, void *data)
6159{
6160	struct igb_adapter *adapter = data;
6161	struct igb_q_vector *q_vector = adapter->q_vector[0];
6162	struct e1000_hw *hw = &adapter->hw;
6163	/* read ICR disables interrupts using IAM */
6164	u32 icr = rd32(E1000_ICR);
6165
6166	igb_write_itr(q_vector);
6167
6168	if (icr & E1000_ICR_DRSTA)
6169		schedule_work(&adapter->reset_task);
6170
6171	if (icr & E1000_ICR_DOUTSYNC) {
6172		/* HW is reporting DMA is out of sync */
6173		adapter->stats.doosync++;
6174	}
6175
6176	if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
6177		hw->mac.get_link_status = 1;
6178		if (!test_bit(__IGB_DOWN, &adapter->state))
6179			mod_timer(&adapter->watchdog_timer, jiffies + 1);
6180	}
6181
6182	if (icr & E1000_ICR_TS) {
6183		u32 tsicr = rd32(E1000_TSICR);
6184
6185		if (tsicr & E1000_TSICR_TXTS) {
6186			/* acknowledge the interrupt */
6187			wr32(E1000_TSICR, E1000_TSICR_TXTS);
6188			/* retrieve hardware timestamp */
6189			schedule_work(&adapter->ptp_tx_work);
6190		}
6191	}
6192
6193	napi_schedule(&q_vector->napi);
6194
6195	return IRQ_HANDLED;
6196}
6197
6198/**
6199 *  igb_intr - Legacy Interrupt Handler
6200 *  @irq: interrupt number
6201 *  @data: pointer to a network interface device structure
6202 **/
6203static irqreturn_t igb_intr(int irq, void *data)
6204{
6205	struct igb_adapter *adapter = data;
6206	struct igb_q_vector *q_vector = adapter->q_vector[0];
6207	struct e1000_hw *hw = &adapter->hw;
6208	/* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
6209	 * need for the IMC write
6210	 */
6211	u32 icr = rd32(E1000_ICR);
6212
6213	/* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
6214	 * not set, then the adapter didn't send an interrupt
6215	 */
6216	if (!(icr & E1000_ICR_INT_ASSERTED))
6217		return IRQ_NONE;
6218
6219	igb_write_itr(q_vector);
6220
6221	if (icr & E1000_ICR_DRSTA)
6222		schedule_work(&adapter->reset_task);
6223
6224	if (icr & E1000_ICR_DOUTSYNC) {
6225		/* HW is reporting DMA is out of sync */
6226		adapter->stats.doosync++;
6227	}
6228
6229	if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
6230		hw->mac.get_link_status = 1;
6231		/* guard against interrupt when we're going down */
6232		if (!test_bit(__IGB_DOWN, &adapter->state))
6233			mod_timer(&adapter->watchdog_timer, jiffies + 1);
6234	}
6235
6236	if (icr & E1000_ICR_TS) {
6237		u32 tsicr = rd32(E1000_TSICR);
6238
6239		if (tsicr & E1000_TSICR_TXTS) {
6240			/* acknowledge the interrupt */
6241			wr32(E1000_TSICR, E1000_TSICR_TXTS);
6242			/* retrieve hardware timestamp */
6243			schedule_work(&adapter->ptp_tx_work);
6244		}
6245	}
6246
6247	napi_schedule(&q_vector->napi);
6248
6249	return IRQ_HANDLED;
6250}
6251
6252static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
6253{
6254	struct igb_adapter *adapter = q_vector->adapter;
6255	struct e1000_hw *hw = &adapter->hw;
6256
6257	if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
6258	    (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
6259		if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
6260			igb_set_itr(q_vector);
6261		else
6262			igb_update_ring_itr(q_vector);
6263	}
6264
6265	if (!test_bit(__IGB_DOWN, &adapter->state)) {
6266		if (adapter->flags & IGB_FLAG_HAS_MSIX)
6267			wr32(E1000_EIMS, q_vector->eims_value);
6268		else
6269			igb_irq_enable(adapter);
6270	}
6271}
6272
6273/**
6274 *  igb_poll - NAPI Rx polling callback
6275 *  @napi: napi polling structure
6276 *  @budget: count of how many packets we should handle
6277 **/
6278static int igb_poll(struct napi_struct *napi, int budget)
6279{
6280	struct igb_q_vector *q_vector = container_of(napi,
6281						     struct igb_q_vector,
6282						     napi);
6283	bool clean_complete = true;
 
6284
6285#ifdef CONFIG_IGB_DCA
6286	if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
6287		igb_update_dca(q_vector);
6288#endif
6289	if (q_vector->tx.ring)
6290		clean_complete = igb_clean_tx_irq(q_vector);
 
 
 
6291
6292	if (q_vector->rx.ring)
6293		clean_complete &= igb_clean_rx_irq(q_vector, budget);
 
 
6294
6295	/* If all work not completed, return budget and keep polling */
6296	if (!clean_complete)
6297		return budget;
6298
6299	/* If not enough Rx work done, exit the polling mode */
6300	napi_complete(napi);
6301	igb_ring_irq_enable(q_vector);
 
 
6302
6303	return 0;
6304}
6305
6306/**
6307 *  igb_clean_tx_irq - Reclaim resources after transmit completes
6308 *  @q_vector: pointer to q_vector containing needed info
 
6309 *
6310 *  returns true if ring is completely cleaned
6311 **/
6312static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
6313{
6314	struct igb_adapter *adapter = q_vector->adapter;
6315	struct igb_ring *tx_ring = q_vector->tx.ring;
6316	struct igb_tx_buffer *tx_buffer;
6317	union e1000_adv_tx_desc *tx_desc;
6318	unsigned int total_bytes = 0, total_packets = 0;
6319	unsigned int budget = q_vector->tx.work_limit;
6320	unsigned int i = tx_ring->next_to_clean;
6321
6322	if (test_bit(__IGB_DOWN, &adapter->state))
6323		return true;
6324
6325	tx_buffer = &tx_ring->tx_buffer_info[i];
6326	tx_desc = IGB_TX_DESC(tx_ring, i);
6327	i -= tx_ring->count;
6328
6329	do {
6330		union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
6331
6332		/* if next_to_watch is not set then there is no work pending */
6333		if (!eop_desc)
6334			break;
6335
6336		/* prevent any other reads prior to eop_desc */
6337		read_barrier_depends();
6338
6339		/* if DD is not set pending work has not been completed */
6340		if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
6341			break;
6342
6343		/* clear next_to_watch to prevent false hangs */
6344		tx_buffer->next_to_watch = NULL;
6345
6346		/* update the statistics for this packet */
6347		total_bytes += tx_buffer->bytecount;
6348		total_packets += tx_buffer->gso_segs;
6349
6350		/* free the skb */
6351		dev_kfree_skb_any(tx_buffer->skb);
 
 
 
6352
6353		/* unmap skb header data */
6354		dma_unmap_single(tx_ring->dev,
6355				 dma_unmap_addr(tx_buffer, dma),
6356				 dma_unmap_len(tx_buffer, len),
6357				 DMA_TO_DEVICE);
6358
6359		/* clear tx_buffer data */
6360		tx_buffer->skb = NULL;
6361		dma_unmap_len_set(tx_buffer, len, 0);
6362
6363		/* clear last DMA location and unmap remaining buffers */
6364		while (tx_desc != eop_desc) {
6365			tx_buffer++;
6366			tx_desc++;
6367			i++;
6368			if (unlikely(!i)) {
6369				i -= tx_ring->count;
6370				tx_buffer = tx_ring->tx_buffer_info;
6371				tx_desc = IGB_TX_DESC(tx_ring, 0);
6372			}
6373
6374			/* unmap any remaining paged data */
6375			if (dma_unmap_len(tx_buffer, len)) {
6376				dma_unmap_page(tx_ring->dev,
6377					       dma_unmap_addr(tx_buffer, dma),
6378					       dma_unmap_len(tx_buffer, len),
6379					       DMA_TO_DEVICE);
6380				dma_unmap_len_set(tx_buffer, len, 0);
6381			}
6382		}
6383
6384		/* move us one more past the eop_desc for start of next pkt */
6385		tx_buffer++;
6386		tx_desc++;
6387		i++;
6388		if (unlikely(!i)) {
6389			i -= tx_ring->count;
6390			tx_buffer = tx_ring->tx_buffer_info;
6391			tx_desc = IGB_TX_DESC(tx_ring, 0);
6392		}
6393
6394		/* issue prefetch for next Tx descriptor */
6395		prefetch(tx_desc);
6396
6397		/* update budget accounting */
6398		budget--;
6399	} while (likely(budget));
6400
6401	netdev_tx_completed_queue(txring_txq(tx_ring),
6402				  total_packets, total_bytes);
6403	i += tx_ring->count;
6404	tx_ring->next_to_clean = i;
6405	u64_stats_update_begin(&tx_ring->tx_syncp);
6406	tx_ring->tx_stats.bytes += total_bytes;
6407	tx_ring->tx_stats.packets += total_packets;
6408	u64_stats_update_end(&tx_ring->tx_syncp);
6409	q_vector->tx.total_bytes += total_bytes;
6410	q_vector->tx.total_packets += total_packets;
6411
6412	if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
6413		struct e1000_hw *hw = &adapter->hw;
6414
6415		/* Detect a transmit hang in hardware, this serializes the
6416		 * check with the clearing of time_stamp and movement of i
6417		 */
6418		clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
6419		if (tx_buffer->next_to_watch &&
6420		    time_after(jiffies, tx_buffer->time_stamp +
6421			       (adapter->tx_timeout_factor * HZ)) &&
6422		    !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
6423
6424			/* detected Tx unit hang */
6425			dev_err(tx_ring->dev,
6426				"Detected Tx Unit Hang\n"
6427				"  Tx Queue             <%d>\n"
6428				"  TDH                  <%x>\n"
6429				"  TDT                  <%x>\n"
6430				"  next_to_use          <%x>\n"
6431				"  next_to_clean        <%x>\n"
6432				"buffer_info[next_to_clean]\n"
6433				"  time_stamp           <%lx>\n"
6434				"  next_to_watch        <%p>\n"
6435				"  jiffies              <%lx>\n"
6436				"  desc.status          <%x>\n",
6437				tx_ring->queue_index,
6438				rd32(E1000_TDH(tx_ring->reg_idx)),
6439				readl(tx_ring->tail),
6440				tx_ring->next_to_use,
6441				tx_ring->next_to_clean,
6442				tx_buffer->time_stamp,
6443				tx_buffer->next_to_watch,
6444				jiffies,
6445				tx_buffer->next_to_watch->wb.status);
6446			netif_stop_subqueue(tx_ring->netdev,
6447					    tx_ring->queue_index);
6448
6449			/* we are about to reset, no point in enabling stuff */
6450			return true;
6451		}
6452	}
6453
6454#define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
6455	if (unlikely(total_packets &&
6456	    netif_carrier_ok(tx_ring->netdev) &&
6457	    igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
6458		/* Make sure that anybody stopping the queue after this
6459		 * sees the new next_to_clean.
6460		 */
6461		smp_mb();
6462		if (__netif_subqueue_stopped(tx_ring->netdev,
6463					     tx_ring->queue_index) &&
6464		    !(test_bit(__IGB_DOWN, &adapter->state))) {
6465			netif_wake_subqueue(tx_ring->netdev,
6466					    tx_ring->queue_index);
6467
6468			u64_stats_update_begin(&tx_ring->tx_syncp);
6469			tx_ring->tx_stats.restart_queue++;
6470			u64_stats_update_end(&tx_ring->tx_syncp);
6471		}
6472	}
6473
6474	return !!budget;
6475}
6476
6477/**
6478 *  igb_reuse_rx_page - page flip buffer and store it back on the ring
6479 *  @rx_ring: rx descriptor ring to store buffers on
6480 *  @old_buff: donor buffer to have page reused
6481 *
6482 *  Synchronizes page for reuse by the adapter
6483 **/
6484static void igb_reuse_rx_page(struct igb_ring *rx_ring,
6485			      struct igb_rx_buffer *old_buff)
6486{
6487	struct igb_rx_buffer *new_buff;
6488	u16 nta = rx_ring->next_to_alloc;
6489
6490	new_buff = &rx_ring->rx_buffer_info[nta];
6491
6492	/* update, and store next to alloc */
6493	nta++;
6494	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
6495
6496	/* transfer page from old buffer to new buffer */
6497	memcpy(new_buff, old_buff, sizeof(struct igb_rx_buffer));
6498
6499	/* sync the buffer for use by the device */
6500	dma_sync_single_range_for_device(rx_ring->dev, old_buff->dma,
6501					 old_buff->page_offset,
6502					 IGB_RX_BUFSZ,
6503					 DMA_FROM_DEVICE);
6504}
6505
6506static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer,
6507				  struct page *page,
6508				  unsigned int truesize)
6509{
6510	/* avoid re-using remote pages */
6511	if (unlikely(page_to_nid(page) != numa_node_id()))
 
 
 
6512		return false;
6513
6514#if (PAGE_SIZE < 8192)
6515	/* if we are only owner of page we can reuse it */
6516	if (unlikely(page_count(page) != 1))
6517		return false;
6518
6519	/* flip page offset to other buffer */
6520	rx_buffer->page_offset ^= IGB_RX_BUFSZ;
6521
6522	/* since we are the only owner of the page and we need to
6523	 * increment it, just set the value to 2 in order to avoid
6524	 * an unnecessary locked operation
6525	 */
6526	atomic_set(&page->_count, 2);
6527#else
6528	/* move offset up to the next cache line */
6529	rx_buffer->page_offset += truesize;
6530
6531	if (rx_buffer->page_offset > (PAGE_SIZE - IGB_RX_BUFSZ))
6532		return false;
 
6533
6534	/* bump ref count on page before it is given to the stack */
6535	get_page(page);
6536#endif
 
 
 
 
 
6537
6538	return true;
6539}
6540
6541/**
6542 *  igb_add_rx_frag - Add contents of Rx buffer to sk_buff
6543 *  @rx_ring: rx descriptor ring to transact packets on
6544 *  @rx_buffer: buffer containing page to add
6545 *  @rx_desc: descriptor containing length of buffer written by hardware
6546 *  @skb: sk_buff to place the data into
 
6547 *
6548 *  This function will add the data contained in rx_buffer->page to the skb.
6549 *  This is done either through a direct copy if the data in the buffer is
6550 *  less than the skb header size, otherwise it will just attach the page as
6551 *  a frag to the skb.
6552 *
6553 *  The function will then update the page offset if necessary and return
6554 *  true if the buffer can be reused by the adapter.
6555 **/
6556static bool igb_add_rx_frag(struct igb_ring *rx_ring,
6557			    struct igb_rx_buffer *rx_buffer,
6558			    union e1000_adv_rx_desc *rx_desc,
6559			    struct sk_buff *skb)
6560{
6561	struct page *page = rx_buffer->page;
6562	unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
6563#if (PAGE_SIZE < 8192)
6564	unsigned int truesize = IGB_RX_BUFSZ;
6565#else
6566	unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
 
 
6567#endif
 
 
 
 
 
 
 
 
6568
6569	if ((size <= IGB_RX_HDR_LEN) && !skb_is_nonlinear(skb)) {
6570		unsigned char *va = page_address(page) + rx_buffer->page_offset;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6571
6572		if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6573			igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
6574			va += IGB_TS_HDR_LEN;
6575			size -= IGB_TS_HDR_LEN;
6576		}
6577
6578		memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
 
 
 
 
 
 
 
 
 
 
 
 
 
6579
6580		/* we can reuse buffer as-is, just make sure it is local */
6581		if (likely(page_to_nid(page) == numa_node_id()))
6582			return true;
6583
6584		/* this page cannot be reused so discard it */
6585		put_page(page);
6586		return false;
6587	}
 
 
 
 
 
 
 
 
 
 
6588
6589	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
6590			rx_buffer->page_offset, size, truesize);
6591
6592	return igb_can_reuse_rx_page(rx_buffer, page, truesize);
6593}
 
 
6594
6595static struct sk_buff *igb_fetch_rx_buffer(struct igb_ring *rx_ring,
6596					   union e1000_adv_rx_desc *rx_desc,
6597					   struct sk_buff *skb)
6598{
6599	struct igb_rx_buffer *rx_buffer;
6600	struct page *page;
6601
6602	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
 
6603
6604	page = rx_buffer->page;
6605	prefetchw(page);
6606
6607	if (likely(!skb)) {
6608		void *page_addr = page_address(page) +
6609				  rx_buffer->page_offset;
6610
6611		/* prefetch first cache line of first page */
6612		prefetch(page_addr);
6613#if L1_CACHE_BYTES < 128
6614		prefetch(page_addr + L1_CACHE_BYTES);
6615#endif
6616
6617		/* allocate a skb to store the frags */
6618		skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
6619						IGB_RX_HDR_LEN);
6620		if (unlikely(!skb)) {
6621			rx_ring->rx_stats.alloc_failed++;
6622			return NULL;
6623		}
 
 
 
 
 
6624
6625		/* we will be copying header into skb->data in
6626		 * pskb_may_pull so it is in our interest to prefetch
6627		 * it now to avoid a possible cache miss
6628		 */
6629		prefetchw(skb->data);
6630	}
6631
6632	/* we are reusing so sync this buffer for CPU use */
6633	dma_sync_single_range_for_cpu(rx_ring->dev,
6634				      rx_buffer->dma,
6635				      rx_buffer->page_offset,
6636				      IGB_RX_BUFSZ,
6637				      DMA_FROM_DEVICE);
6638
6639	/* pull page into skb */
6640	if (igb_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
6641		/* hand second half of page back to the ring */
6642		igb_reuse_rx_page(rx_ring, rx_buffer);
6643	} else {
6644		/* we are not reusing the buffer so unmap it */
6645		dma_unmap_page(rx_ring->dev, rx_buffer->dma,
6646			       PAGE_SIZE, DMA_FROM_DEVICE);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6647	}
 
 
 
 
 
 
 
 
6648
6649	/* clear contents of rx_buffer */
6650	rx_buffer->page = NULL;
 
 
 
 
 
 
 
 
6651
6652	return skb;
 
 
 
 
 
 
 
 
 
6653}
6654
6655static inline void igb_rx_checksum(struct igb_ring *ring,
6656				   union e1000_adv_rx_desc *rx_desc,
6657				   struct sk_buff *skb)
6658{
6659	skb_checksum_none_assert(skb);
6660
6661	/* Ignore Checksum bit is set */
6662	if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
6663		return;
6664
6665	/* Rx checksum disabled via ethtool */
6666	if (!(ring->netdev->features & NETIF_F_RXCSUM))
6667		return;
6668
6669	/* TCP/UDP checksum error bit is set */
6670	if (igb_test_staterr(rx_desc,
6671			     E1000_RXDEXT_STATERR_TCPE |
6672			     E1000_RXDEXT_STATERR_IPE)) {
6673		/* work around errata with sctp packets where the TCPE aka
6674		 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
6675		 * packets, (aka let the stack check the crc32c)
6676		 */
6677		if (!((skb->len == 60) &&
6678		      test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
6679			u64_stats_update_begin(&ring->rx_syncp);
6680			ring->rx_stats.csum_err++;
6681			u64_stats_update_end(&ring->rx_syncp);
6682		}
6683		/* let the stack verify checksum errors */
6684		return;
6685	}
6686	/* It must be a TCP or UDP packet with a valid checksum */
6687	if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
6688				      E1000_RXD_STAT_UDPCS))
6689		skb->ip_summed = CHECKSUM_UNNECESSARY;
6690
6691	dev_dbg(ring->dev, "cksum success: bits %08X\n",
6692		le32_to_cpu(rx_desc->wb.upper.status_error));
6693}
6694
6695static inline void igb_rx_hash(struct igb_ring *ring,
6696			       union e1000_adv_rx_desc *rx_desc,
6697			       struct sk_buff *skb)
6698{
6699	if (ring->netdev->features & NETIF_F_RXHASH)
6700		skb_set_hash(skb,
6701			     le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
6702			     PKT_HASH_TYPE_L3);
6703}
6704
6705/**
6706 *  igb_is_non_eop - process handling of non-EOP buffers
6707 *  @rx_ring: Rx ring being processed
6708 *  @rx_desc: Rx descriptor for current buffer
6709 *  @skb: current socket buffer containing buffer in progress
6710 *
6711 *  This function updates next to clean.  If the buffer is an EOP buffer
6712 *  this function exits returning false, otherwise it will place the
6713 *  sk_buff in the next buffer to be chained and return true indicating
6714 *  that this is in fact a non-EOP buffer.
6715 **/
6716static bool igb_is_non_eop(struct igb_ring *rx_ring,
6717			   union e1000_adv_rx_desc *rx_desc)
6718{
6719	u32 ntc = rx_ring->next_to_clean + 1;
6720
6721	/* fetch, update, and store next to clean */
6722	ntc = (ntc < rx_ring->count) ? ntc : 0;
6723	rx_ring->next_to_clean = ntc;
6724
6725	prefetch(IGB_RX_DESC(rx_ring, ntc));
6726
6727	if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
6728		return false;
6729
6730	return true;
6731}
6732
6733/**
6734 *  igb_get_headlen - determine size of header for LRO/GRO
6735 *  @data: pointer to the start of the headers
6736 *  @max_len: total length of section to find headers in
6737 *
6738 *  This function is meant to determine the length of headers that will
6739 *  be recognized by hardware for LRO, and GRO offloads.  The main
6740 *  motivation of doing this is to only perform one pull for IPv4 TCP
6741 *  packets so that we can do basic things like calculating the gso_size
6742 *  based on the average data per packet.
6743 **/
6744static unsigned int igb_get_headlen(unsigned char *data,
6745				    unsigned int max_len)
6746{
6747	union {
6748		unsigned char *network;
6749		/* l2 headers */
6750		struct ethhdr *eth;
6751		struct vlan_hdr *vlan;
6752		/* l3 headers */
6753		struct iphdr *ipv4;
6754		struct ipv6hdr *ipv6;
6755	} hdr;
6756	__be16 protocol;
6757	u8 nexthdr = 0;	/* default to not TCP */
6758	u8 hlen;
6759
6760	/* this should never happen, but better safe than sorry */
6761	if (max_len < ETH_HLEN)
6762		return max_len;
6763
6764	/* initialize network frame pointer */
6765	hdr.network = data;
6766
6767	/* set first protocol and move network header forward */
6768	protocol = hdr.eth->h_proto;
6769	hdr.network += ETH_HLEN;
6770
6771	/* handle any vlan tag if present */
6772	if (protocol == htons(ETH_P_8021Q)) {
6773		if ((hdr.network - data) > (max_len - VLAN_HLEN))
6774			return max_len;
6775
6776		protocol = hdr.vlan->h_vlan_encapsulated_proto;
6777		hdr.network += VLAN_HLEN;
6778	}
6779
6780	/* handle L3 protocols */
6781	if (protocol == htons(ETH_P_IP)) {
6782		if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
6783			return max_len;
6784
6785		/* access ihl as a u8 to avoid unaligned access on ia64 */
6786		hlen = (hdr.network[0] & 0x0F) << 2;
6787
6788		/* verify hlen meets minimum size requirements */
6789		if (hlen < sizeof(struct iphdr))
6790			return hdr.network - data;
6791
6792		/* record next protocol if header is present */
6793		if (!(hdr.ipv4->frag_off & htons(IP_OFFSET)))
6794			nexthdr = hdr.ipv4->protocol;
6795	} else if (protocol == htons(ETH_P_IPV6)) {
6796		if ((hdr.network - data) > (max_len - sizeof(struct ipv6hdr)))
6797			return max_len;
6798
6799		/* record next protocol */
6800		nexthdr = hdr.ipv6->nexthdr;
6801		hlen = sizeof(struct ipv6hdr);
6802	} else {
6803		return hdr.network - data;
6804	}
6805
6806	/* relocate pointer to start of L4 header */
6807	hdr.network += hlen;
6808
6809	/* finally sort out TCP */
6810	if (nexthdr == IPPROTO_TCP) {
6811		if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
6812			return max_len;
6813
6814		/* access doff as a u8 to avoid unaligned access on ia64 */
6815		hlen = (hdr.network[12] & 0xF0) >> 2;
6816
6817		/* verify hlen meets minimum size requirements */
6818		if (hlen < sizeof(struct tcphdr))
6819			return hdr.network - data;
6820
6821		hdr.network += hlen;
6822	} else if (nexthdr == IPPROTO_UDP) {
6823		if ((hdr.network - data) > (max_len - sizeof(struct udphdr)))
6824			return max_len;
6825
6826		hdr.network += sizeof(struct udphdr);
6827	}
6828
6829	/* If everything has gone correctly hdr.network should be the
6830	 * data section of the packet and will be the end of the header.
6831	 * If not then it probably represents the end of the last recognized
6832	 * header.
6833	 */
6834	if ((hdr.network - data) < max_len)
6835		return hdr.network - data;
6836	else
6837		return max_len;
6838}
6839
6840/**
6841 *  igb_pull_tail - igb specific version of skb_pull_tail
6842 *  @rx_ring: rx descriptor ring packet is being transacted on
6843 *  @rx_desc: pointer to the EOP Rx descriptor
6844 *  @skb: pointer to current skb being adjusted
6845 *
6846 *  This function is an igb specific version of __pskb_pull_tail.  The
6847 *  main difference between this version and the original function is that
6848 *  this function can make several assumptions about the state of things
6849 *  that allow for significant optimizations versus the standard function.
6850 *  As a result we can do things like drop a frag and maintain an accurate
6851 *  truesize for the skb.
6852 */
6853static void igb_pull_tail(struct igb_ring *rx_ring,
6854			  union e1000_adv_rx_desc *rx_desc,
6855			  struct sk_buff *skb)
6856{
6857	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
6858	unsigned char *va;
6859	unsigned int pull_len;
6860
6861	/* it is valid to use page_address instead of kmap since we are
6862	 * working with pages allocated out of the lomem pool per
6863	 * alloc_page(GFP_ATOMIC)
6864	 */
6865	va = skb_frag_address(frag);
6866
6867	if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6868		/* retrieve timestamp from buffer */
6869		igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
6870
6871		/* update pointers to remove timestamp header */
6872		skb_frag_size_sub(frag, IGB_TS_HDR_LEN);
6873		frag->page_offset += IGB_TS_HDR_LEN;
6874		skb->data_len -= IGB_TS_HDR_LEN;
6875		skb->len -= IGB_TS_HDR_LEN;
6876
6877		/* move va to start of packet data */
6878		va += IGB_TS_HDR_LEN;
6879	}
6880
6881	/* we need the header to contain the greater of either ETH_HLEN or
6882	 * 60 bytes if the skb->len is less than 60 for skb_pad.
6883	 */
6884	pull_len = igb_get_headlen(va, IGB_RX_HDR_LEN);
6885
6886	/* align pull length to size of long to optimize memcpy performance */
6887	skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
6888
6889	/* update all of the pointers */
6890	skb_frag_size_sub(frag, pull_len);
6891	frag->page_offset += pull_len;
6892	skb->data_len -= pull_len;
6893	skb->tail += pull_len;
6894}
6895
6896/**
6897 *  igb_cleanup_headers - Correct corrupted or empty headers
6898 *  @rx_ring: rx descriptor ring packet is being transacted on
6899 *  @rx_desc: pointer to the EOP Rx descriptor
6900 *  @skb: pointer to current skb being fixed
6901 *
6902 *  Address the case where we are pulling data in on pages only
6903 *  and as such no data is present in the skb header.
6904 *
6905 *  In addition if skb is not at least 60 bytes we need to pad it so that
6906 *  it is large enough to qualify as a valid Ethernet frame.
6907 *
6908 *  Returns true if an error was encountered and skb was freed.
6909 **/
6910static bool igb_cleanup_headers(struct igb_ring *rx_ring,
6911				union e1000_adv_rx_desc *rx_desc,
6912				struct sk_buff *skb)
6913{
 
 
 
 
6914	if (unlikely((igb_test_staterr(rx_desc,
6915				       E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
6916		struct net_device *netdev = rx_ring->netdev;
6917		if (!(netdev->features & NETIF_F_RXALL)) {
6918			dev_kfree_skb_any(skb);
6919			return true;
6920		}
6921	}
6922
6923	/* place header in linear portion of buffer */
6924	if (skb_is_nonlinear(skb))
6925		igb_pull_tail(rx_ring, rx_desc, skb);
6926
6927	/* if skb_pad returns an error the skb was freed */
6928	if (unlikely(skb->len < 60)) {
6929		int pad_len = 60 - skb->len;
6930
6931		if (skb_pad(skb, pad_len))
6932			return true;
6933		__skb_put(skb, pad_len);
6934	}
6935
6936	return false;
6937}
6938
6939/**
6940 *  igb_process_skb_fields - Populate skb header fields from Rx descriptor
6941 *  @rx_ring: rx descriptor ring packet is being transacted on
6942 *  @rx_desc: pointer to the EOP Rx descriptor
6943 *  @skb: pointer to current skb being populated
6944 *
6945 *  This function checks the ring, descriptor, and packet information in
6946 *  order to populate the hash, checksum, VLAN, timestamp, protocol, and
6947 *  other fields within the skb.
6948 **/
6949static void igb_process_skb_fields(struct igb_ring *rx_ring,
6950				   union e1000_adv_rx_desc *rx_desc,
6951				   struct sk_buff *skb)
6952{
6953	struct net_device *dev = rx_ring->netdev;
6954
6955	igb_rx_hash(rx_ring, rx_desc, skb);
6956
6957	igb_rx_checksum(rx_ring, rx_desc, skb);
6958
6959	if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TS) &&
6960	    !igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP))
6961		igb_ptp_rx_rgtstamp(rx_ring->q_vector, skb);
6962
6963	if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
6964	    igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
6965		u16 vid;
 
6966		if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
6967		    test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
6968			vid = be16_to_cpu(rx_desc->wb.upper.vlan);
6969		else
6970			vid = le16_to_cpu(rx_desc->wb.upper.vlan);
6971
6972		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
6973	}
6974
6975	skb_record_rx_queue(skb, rx_ring->queue_index);
6976
6977	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
6978}
6979
6980static bool igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
6981{
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6982	struct igb_ring *rx_ring = q_vector->rx.ring;
6983	struct sk_buff *skb = rx_ring->skb;
6984	unsigned int total_bytes = 0, total_packets = 0;
6985	u16 cleaned_count = igb_desc_unused(rx_ring);
 
 
 
 
 
 
 
 
 
 
6986
6987	while (likely(total_packets < budget)) {
6988		union e1000_adv_rx_desc *rx_desc;
 
 
 
 
 
6989
6990		/* return some buffers to hardware, one at a time is too slow */
6991		if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
6992			igb_alloc_rx_buffers(rx_ring, cleaned_count);
6993			cleaned_count = 0;
6994		}
6995
6996		rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
6997
6998		if (!igb_test_staterr(rx_desc, E1000_RXD_STAT_DD))
6999			break;
7000
7001		/* This memory barrier is needed to keep us from reading
7002		 * any other fields out of the rx_desc until we know the
7003		 * RXD_STAT_DD bit is set
7004		 */
7005		rmb();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7006
7007		/* retrieve a buffer from the ring */
7008		skb = igb_fetch_rx_buffer(rx_ring, rx_desc, skb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7009
7010		/* exit if we failed to retrieve a buffer */
7011		if (!skb)
 
 
7012			break;
 
7013
 
7014		cleaned_count++;
7015
7016		/* fetch next buffer in frame if non-eop */
7017		if (igb_is_non_eop(rx_ring, rx_desc))
7018			continue;
7019
7020		/* verify the packet layout is correct */
7021		if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
7022			skb = NULL;
7023			continue;
7024		}
7025
7026		/* probably a little skewed due to removing CRC */
7027		total_bytes += skb->len;
7028
7029		/* populate checksum, timestamp, VLAN, and protocol */
7030		igb_process_skb_fields(rx_ring, rx_desc, skb);
7031
7032		napi_gro_receive(&q_vector->napi, skb);
7033
7034		/* reset skb pointer */
7035		skb = NULL;
7036
7037		/* update budget accounting */
7038		total_packets++;
7039	}
7040
7041	/* place incomplete frames back on ring for completion */
7042	rx_ring->skb = skb;
7043
 
 
 
 
 
 
 
 
 
7044	u64_stats_update_begin(&rx_ring->rx_syncp);
7045	rx_ring->rx_stats.packets += total_packets;
7046	rx_ring->rx_stats.bytes += total_bytes;
7047	u64_stats_update_end(&rx_ring->rx_syncp);
7048	q_vector->rx.total_packets += total_packets;
7049	q_vector->rx.total_bytes += total_bytes;
7050
7051	if (cleaned_count)
7052		igb_alloc_rx_buffers(rx_ring, cleaned_count);
7053
7054	return (total_packets < budget);
7055}
7056
7057static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
7058				  struct igb_rx_buffer *bi)
7059{
7060	struct page *page = bi->page;
7061	dma_addr_t dma;
7062
7063	/* since we are recycling buffers we should seldom need to alloc */
7064	if (likely(page))
7065		return true;
7066
7067	/* alloc new page for storage */
7068	page = __skb_alloc_page(GFP_ATOMIC | __GFP_COLD, NULL);
7069	if (unlikely(!page)) {
7070		rx_ring->rx_stats.alloc_failed++;
7071		return false;
7072	}
7073
7074	/* map page for use */
7075	dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE, DMA_FROM_DEVICE);
 
 
 
7076
7077	/* if mapping failed free memory back to system since
7078	 * there isn't much point in holding memory we can't use
7079	 */
7080	if (dma_mapping_error(rx_ring->dev, dma)) {
7081		__free_page(page);
7082
7083		rx_ring->rx_stats.alloc_failed++;
7084		return false;
7085	}
7086
7087	bi->dma = dma;
7088	bi->page = page;
7089	bi->page_offset = 0;
 
 
7090
7091	return true;
7092}
7093
7094/**
7095 *  igb_alloc_rx_buffers - Replace used receive buffers; packet split
7096 *  @adapter: address of board private structure
 
7097 **/
7098void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
7099{
7100	union e1000_adv_rx_desc *rx_desc;
7101	struct igb_rx_buffer *bi;
7102	u16 i = rx_ring->next_to_use;
 
7103
7104	/* nothing to do */
7105	if (!cleaned_count)
7106		return;
7107
7108	rx_desc = IGB_RX_DESC(rx_ring, i);
7109	bi = &rx_ring->rx_buffer_info[i];
7110	i -= rx_ring->count;
7111
 
 
7112	do {
7113		if (!igb_alloc_mapped_page(rx_ring, bi))
7114			break;
7115
 
 
 
 
 
7116		/* Refresh the desc even if buffer_addrs didn't change
7117		 * because each write-back erases this info.
7118		 */
7119		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
7120
7121		rx_desc++;
7122		bi++;
7123		i++;
7124		if (unlikely(!i)) {
7125			rx_desc = IGB_RX_DESC(rx_ring, 0);
7126			bi = rx_ring->rx_buffer_info;
7127			i -= rx_ring->count;
7128		}
7129
7130		/* clear the hdr_addr for the next_to_use descriptor */
7131		rx_desc->read.hdr_addr = 0;
7132
7133		cleaned_count--;
7134	} while (cleaned_count);
7135
7136	i += rx_ring->count;
7137
7138	if (rx_ring->next_to_use != i) {
7139		/* record the next descriptor to use */
7140		rx_ring->next_to_use = i;
7141
7142		/* update next to alloc since we have filled the ring */
7143		rx_ring->next_to_alloc = i;
7144
7145		/* Force memory writes to complete before letting h/w
7146		 * know there are new descriptors to fetch.  (Only
7147		 * applicable for weak-ordered memory model archs,
7148		 * such as IA-64).
7149		 */
7150		wmb();
7151		writel(i, rx_ring->tail);
7152	}
7153}
7154
7155/**
7156 * igb_mii_ioctl -
7157 * @netdev:
7158 * @ifreq:
7159 * @cmd:
7160 **/
7161static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
7162{
7163	struct igb_adapter *adapter = netdev_priv(netdev);
7164	struct mii_ioctl_data *data = if_mii(ifr);
7165
7166	if (adapter->hw.phy.media_type != e1000_media_type_copper)
7167		return -EOPNOTSUPP;
7168
7169	switch (cmd) {
7170	case SIOCGMIIPHY:
7171		data->phy_id = adapter->hw.phy.addr;
7172		break;
7173	case SIOCGMIIREG:
7174		if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
7175		                     &data->val_out))
7176			return -EIO;
7177		break;
7178	case SIOCSMIIREG:
7179	default:
7180		return -EOPNOTSUPP;
7181	}
7182	return 0;
7183}
7184
7185/**
7186 * igb_ioctl -
7187 * @netdev:
7188 * @ifreq:
7189 * @cmd:
7190 **/
7191static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
7192{
7193	switch (cmd) {
7194	case SIOCGMIIPHY:
7195	case SIOCGMIIREG:
7196	case SIOCSMIIREG:
7197		return igb_mii_ioctl(netdev, ifr, cmd);
7198	case SIOCGHWTSTAMP:
7199		return igb_ptp_get_ts_config(netdev, ifr);
7200	case SIOCSHWTSTAMP:
7201		return igb_ptp_set_ts_config(netdev, ifr);
7202	default:
7203		return -EOPNOTSUPP;
7204	}
7205}
7206
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7207s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
7208{
7209	struct igb_adapter *adapter = hw->back;
7210
7211	if (pcie_capability_read_word(adapter->pdev, reg, value))
7212		return -E1000_ERR_CONFIG;
7213
7214	return 0;
7215}
7216
7217s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
7218{
7219	struct igb_adapter *adapter = hw->back;
7220
7221	if (pcie_capability_write_word(adapter->pdev, reg, *value))
7222		return -E1000_ERR_CONFIG;
7223
7224	return 0;
7225}
7226
7227static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
7228{
7229	struct igb_adapter *adapter = netdev_priv(netdev);
7230	struct e1000_hw *hw = &adapter->hw;
7231	u32 ctrl, rctl;
7232	bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
7233
7234	if (enable) {
7235		/* enable VLAN tag insert/strip */
7236		ctrl = rd32(E1000_CTRL);
7237		ctrl |= E1000_CTRL_VME;
7238		wr32(E1000_CTRL, ctrl);
7239
7240		/* Disable CFI check */
7241		rctl = rd32(E1000_RCTL);
7242		rctl &= ~E1000_RCTL_CFIEN;
7243		wr32(E1000_RCTL, rctl);
7244	} else {
7245		/* disable VLAN tag insert/strip */
7246		ctrl = rd32(E1000_CTRL);
7247		ctrl &= ~E1000_CTRL_VME;
7248		wr32(E1000_CTRL, ctrl);
7249	}
7250
7251	igb_rlpml_set(adapter);
7252}
7253
7254static int igb_vlan_rx_add_vid(struct net_device *netdev,
7255			       __be16 proto, u16 vid)
7256{
7257	struct igb_adapter *adapter = netdev_priv(netdev);
7258	struct e1000_hw *hw = &adapter->hw;
7259	int pf_id = adapter->vfs_allocated_count;
7260
7261	/* attempt to add filter to vlvf array */
7262	igb_vlvf_set(adapter, vid, true, pf_id);
7263
7264	/* add the filter since PF can receive vlans w/o entry in vlvf */
7265	igb_vfta_set(hw, vid, true);
 
7266
7267	set_bit(vid, adapter->active_vlans);
7268
7269	return 0;
7270}
7271
7272static int igb_vlan_rx_kill_vid(struct net_device *netdev,
7273				__be16 proto, u16 vid)
7274{
7275	struct igb_adapter *adapter = netdev_priv(netdev);
 
7276	struct e1000_hw *hw = &adapter->hw;
7277	int pf_id = adapter->vfs_allocated_count;
7278	s32 err;
7279
7280	/* remove vlan from VLVF table array */
7281	err = igb_vlvf_set(adapter, vid, false, pf_id);
7282
7283	/* if vid was not present in VLVF just remove it from table */
7284	if (err)
7285		igb_vfta_set(hw, vid, false);
7286
7287	clear_bit(vid, adapter->active_vlans);
7288
7289	return 0;
7290}
7291
7292static void igb_restore_vlan(struct igb_adapter *adapter)
7293{
7294	u16 vid;
7295
7296	igb_vlan_mode(adapter->netdev, adapter->netdev->features);
 
7297
7298	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
7299		igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
7300}
7301
7302int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
7303{
7304	struct pci_dev *pdev = adapter->pdev;
7305	struct e1000_mac_info *mac = &adapter->hw.mac;
7306
7307	mac->autoneg = 0;
7308
7309	/* Make sure dplx is at most 1 bit and lsb of speed is not set
7310	 * for the switch() below to work
7311	 */
7312	if ((spd & 1) || (dplx & ~1))
7313		goto err_inval;
7314
7315	/* Fiber NIC's only allow 1000 gbps Full duplex
7316	 * and 100Mbps Full duplex for 100baseFx sfp
7317	 */
7318	if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
7319		switch (spd + dplx) {
7320		case SPEED_10 + DUPLEX_HALF:
7321		case SPEED_10 + DUPLEX_FULL:
7322		case SPEED_100 + DUPLEX_HALF:
7323			goto err_inval;
7324		default:
7325			break;
7326		}
7327	}
7328
7329	switch (spd + dplx) {
7330	case SPEED_10 + DUPLEX_HALF:
7331		mac->forced_speed_duplex = ADVERTISE_10_HALF;
7332		break;
7333	case SPEED_10 + DUPLEX_FULL:
7334		mac->forced_speed_duplex = ADVERTISE_10_FULL;
7335		break;
7336	case SPEED_100 + DUPLEX_HALF:
7337		mac->forced_speed_duplex = ADVERTISE_100_HALF;
7338		break;
7339	case SPEED_100 + DUPLEX_FULL:
7340		mac->forced_speed_duplex = ADVERTISE_100_FULL;
7341		break;
7342	case SPEED_1000 + DUPLEX_FULL:
7343		mac->autoneg = 1;
7344		adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
7345		break;
7346	case SPEED_1000 + DUPLEX_HALF: /* not supported */
7347	default:
7348		goto err_inval;
7349	}
7350
7351	/* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
7352	adapter->hw.phy.mdix = AUTO_ALL_MODES;
7353
7354	return 0;
7355
7356err_inval:
7357	dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
7358	return -EINVAL;
7359}
7360
7361static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
7362			  bool runtime)
7363{
7364	struct net_device *netdev = pci_get_drvdata(pdev);
7365	struct igb_adapter *adapter = netdev_priv(netdev);
7366	struct e1000_hw *hw = &adapter->hw;
7367	u32 ctrl, rctl, status;
7368	u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
7369#ifdef CONFIG_PM
7370	int retval = 0;
7371#endif
7372
 
7373	netif_device_detach(netdev);
7374
7375	if (netif_running(netdev))
7376		__igb_close(netdev, true);
7377
 
 
7378	igb_clear_interrupt_scheme(adapter);
7379
7380#ifdef CONFIG_PM
7381	retval = pci_save_state(pdev);
7382	if (retval)
7383		return retval;
7384#endif
7385
7386	status = rd32(E1000_STATUS);
7387	if (status & E1000_STATUS_LU)
7388		wufc &= ~E1000_WUFC_LNKC;
7389
7390	if (wufc) {
7391		igb_setup_rctl(adapter);
7392		igb_set_rx_mode(netdev);
7393
7394		/* turn on all-multi mode if wake on multicast is enabled */
7395		if (wufc & E1000_WUFC_MC) {
7396			rctl = rd32(E1000_RCTL);
7397			rctl |= E1000_RCTL_MPE;
7398			wr32(E1000_RCTL, rctl);
7399		}
7400
7401		ctrl = rd32(E1000_CTRL);
7402		/* advertise wake from D3Cold */
7403		#define E1000_CTRL_ADVD3WUC 0x00100000
7404		/* phy power management enable */
7405		#define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
7406		ctrl |= E1000_CTRL_ADVD3WUC;
7407		wr32(E1000_CTRL, ctrl);
7408
7409		/* Allow time for pending master requests to run */
7410		igb_disable_pcie_master(hw);
7411
7412		wr32(E1000_WUC, E1000_WUC_PME_EN);
7413		wr32(E1000_WUFC, wufc);
7414	} else {
7415		wr32(E1000_WUC, 0);
7416		wr32(E1000_WUFC, 0);
7417	}
7418
7419	*enable_wake = wufc || adapter->en_mng_pt;
7420	if (!*enable_wake)
7421		igb_power_down_link(adapter);
7422	else
7423		igb_power_up_link(adapter);
7424
 
 
 
7425	/* Release control of h/w to f/w.  If f/w is AMT enabled, this
7426	 * would have already happened in close and is redundant.
7427	 */
7428	igb_release_hw_control(adapter);
7429
7430	pci_disable_device(pdev);
7431
7432	return 0;
7433}
7434
7435#ifdef CONFIG_PM
7436#ifdef CONFIG_PM_SLEEP
7437static int igb_suspend(struct device *dev)
7438{
7439	int retval;
7440	bool wake;
7441	struct pci_dev *pdev = to_pci_dev(dev);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7442
7443	retval = __igb_shutdown(pdev, &wake, 0);
7444	if (retval)
7445		return retval;
7446
7447	if (wake) {
7448		pci_prepare_to_sleep(pdev);
7449	} else {
7450		pci_wake_from_d3(pdev, false);
7451		pci_set_power_state(pdev, PCI_D3hot);
7452	}
7453
7454	return 0;
 
 
7455}
7456#endif /* CONFIG_PM_SLEEP */
7457
7458static int igb_resume(struct device *dev)
7459{
7460	struct pci_dev *pdev = to_pci_dev(dev);
7461	struct net_device *netdev = pci_get_drvdata(pdev);
7462	struct igb_adapter *adapter = netdev_priv(netdev);
7463	struct e1000_hw *hw = &adapter->hw;
7464	u32 err;
7465
7466	pci_set_power_state(pdev, PCI_D0);
7467	pci_restore_state(pdev);
7468	pci_save_state(pdev);
7469
 
 
7470	err = pci_enable_device_mem(pdev);
7471	if (err) {
7472		dev_err(&pdev->dev,
7473			"igb: Cannot enable PCI device from suspend\n");
7474		return err;
7475	}
7476	pci_set_master(pdev);
7477
7478	pci_enable_wake(pdev, PCI_D3hot, 0);
7479	pci_enable_wake(pdev, PCI_D3cold, 0);
7480
7481	if (igb_init_interrupt_scheme(adapter, true)) {
7482		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
7483		return -ENOMEM;
7484	}
7485
7486	igb_reset(adapter);
7487
7488	/* let the f/w know that the h/w is now under the control of the
7489	 * driver.
7490	 */
7491	igb_get_hw_control(adapter);
7492
 
 
 
 
7493	wr32(E1000_WUS, ~0);
7494
7495	if (netdev->flags & IFF_UP) {
7496		rtnl_lock();
 
7497		err = __igb_open(netdev, true);
 
 
 
 
7498		rtnl_unlock();
7499		if (err)
7500			return err;
7501	}
7502
7503	netif_device_attach(netdev);
7504	return 0;
 
 
 
 
7505}
7506
7507#ifdef CONFIG_PM_RUNTIME
7508static int igb_runtime_idle(struct device *dev)
7509{
7510	struct pci_dev *pdev = to_pci_dev(dev);
7511	struct net_device *netdev = pci_get_drvdata(pdev);
7512	struct igb_adapter *adapter = netdev_priv(netdev);
7513
7514	if (!igb_has_link(adapter))
7515		pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
7516
7517	return -EBUSY;
7518}
7519
7520static int igb_runtime_suspend(struct device *dev)
7521{
7522	struct pci_dev *pdev = to_pci_dev(dev);
7523	int retval;
7524	bool wake;
7525
7526	retval = __igb_shutdown(pdev, &wake, 1);
7527	if (retval)
7528		return retval;
7529
7530	if (wake) {
7531		pci_prepare_to_sleep(pdev);
7532	} else {
7533		pci_wake_from_d3(pdev, false);
7534		pci_set_power_state(pdev, PCI_D3hot);
7535	}
7536
7537	return 0;
7538}
7539
7540static int igb_runtime_resume(struct device *dev)
7541{
7542	return igb_resume(dev);
7543}
7544#endif /* CONFIG_PM_RUNTIME */
7545#endif
7546
7547static void igb_shutdown(struct pci_dev *pdev)
7548{
7549	bool wake;
7550
7551	__igb_shutdown(pdev, &wake, 0);
7552
7553	if (system_state == SYSTEM_POWER_OFF) {
7554		pci_wake_from_d3(pdev, wake);
7555		pci_set_power_state(pdev, PCI_D3hot);
7556	}
7557}
7558
 
 
7559#ifdef CONFIG_PCI_IOV
7560static int igb_sriov_reinit(struct pci_dev *dev)
7561{
7562	struct net_device *netdev = pci_get_drvdata(dev);
7563	struct igb_adapter *adapter = netdev_priv(netdev);
7564	struct pci_dev *pdev = adapter->pdev;
7565
7566	rtnl_lock();
7567
7568	if (netif_running(netdev))
7569		igb_close(netdev);
7570
7571	igb_clear_interrupt_scheme(adapter);
7572
7573	igb_init_queue_configuration(adapter);
7574
7575	if (igb_init_interrupt_scheme(adapter, true)) {
7576		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
7577		return -ENOMEM;
7578	}
7579
7580	if (netif_running(netdev))
7581		igb_open(netdev);
7582
7583	rtnl_unlock();
7584
7585	return 0;
7586}
7587
7588static int igb_pci_disable_sriov(struct pci_dev *dev)
7589{
7590	int err = igb_disable_sriov(dev);
7591
7592	if (!err)
7593		err = igb_sriov_reinit(dev);
7594
7595	return err;
7596}
7597
7598static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs)
7599{
7600	int err = igb_enable_sriov(dev, num_vfs);
7601
7602	if (err)
7603		goto out;
7604
7605	err = igb_sriov_reinit(dev);
7606	if (!err)
7607		return num_vfs;
7608
7609out:
7610	return err;
7611}
7612
7613#endif
7614static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
7615{
7616#ifdef CONFIG_PCI_IOV
7617	if (num_vfs == 0)
7618		return igb_pci_disable_sriov(dev);
7619	else
7620		return igb_pci_enable_sriov(dev, num_vfs);
7621#endif
7622	return 0;
7623}
7624
7625#ifdef CONFIG_NET_POLL_CONTROLLER
7626/* Polling 'interrupt' - used by things like netconsole to send skbs
7627 * without having to re-enable interrupts. It's not called while
7628 * the interrupt routine is executing.
7629 */
7630static void igb_netpoll(struct net_device *netdev)
7631{
7632	struct igb_adapter *adapter = netdev_priv(netdev);
7633	struct e1000_hw *hw = &adapter->hw;
7634	struct igb_q_vector *q_vector;
7635	int i;
7636
7637	for (i = 0; i < adapter->num_q_vectors; i++) {
7638		q_vector = adapter->q_vector[i];
7639		if (adapter->flags & IGB_FLAG_HAS_MSIX)
7640			wr32(E1000_EIMC, q_vector->eims_value);
7641		else
7642			igb_irq_disable(adapter);
7643		napi_schedule(&q_vector->napi);
7644	}
7645}
7646#endif /* CONFIG_NET_POLL_CONTROLLER */
7647
7648/**
7649 *  igb_io_error_detected - called when PCI error is detected
7650 *  @pdev: Pointer to PCI device
7651 *  @state: The current pci connection state
7652 *
7653 *  This function is called after a PCI bus error affecting
7654 *  this device has been detected.
7655 **/
7656static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
7657					      pci_channel_state_t state)
7658{
7659	struct net_device *netdev = pci_get_drvdata(pdev);
7660	struct igb_adapter *adapter = netdev_priv(netdev);
7661
 
 
 
 
 
7662	netif_device_detach(netdev);
7663
7664	if (state == pci_channel_io_perm_failure)
7665		return PCI_ERS_RESULT_DISCONNECT;
7666
7667	if (netif_running(netdev))
7668		igb_down(adapter);
7669	pci_disable_device(pdev);
7670
7671	/* Request a slot slot reset. */
7672	return PCI_ERS_RESULT_NEED_RESET;
7673}
7674
7675/**
7676 *  igb_io_slot_reset - called after the pci bus has been reset.
7677 *  @pdev: Pointer to PCI device
7678 *
7679 *  Restart the card from scratch, as if from a cold-boot. Implementation
7680 *  resembles the first-half of the igb_resume routine.
7681 **/
7682static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
7683{
7684	struct net_device *netdev = pci_get_drvdata(pdev);
7685	struct igb_adapter *adapter = netdev_priv(netdev);
7686	struct e1000_hw *hw = &adapter->hw;
7687	pci_ers_result_t result;
7688	int err;
7689
7690	if (pci_enable_device_mem(pdev)) {
7691		dev_err(&pdev->dev,
7692			"Cannot re-enable PCI device after reset.\n");
7693		result = PCI_ERS_RESULT_DISCONNECT;
7694	} else {
7695		pci_set_master(pdev);
7696		pci_restore_state(pdev);
7697		pci_save_state(pdev);
7698
7699		pci_enable_wake(pdev, PCI_D3hot, 0);
7700		pci_enable_wake(pdev, PCI_D3cold, 0);
7701
 
 
 
 
 
7702		igb_reset(adapter);
7703		wr32(E1000_WUS, ~0);
7704		result = PCI_ERS_RESULT_RECOVERED;
7705	}
7706
7707	err = pci_cleanup_aer_uncorrect_error_status(pdev);
7708	if (err) {
7709		dev_err(&pdev->dev,
7710			"pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n",
7711			err);
7712		/* non-fatal, continue */
7713	}
7714
7715	return result;
7716}
7717
7718/**
7719 *  igb_io_resume - called when traffic can start flowing again.
7720 *  @pdev: Pointer to PCI device
7721 *
7722 *  This callback is called when the error recovery driver tells us that
7723 *  its OK to resume normal operation. Implementation resembles the
7724 *  second-half of the igb_resume routine.
7725 */
7726static void igb_io_resume(struct pci_dev *pdev)
7727{
7728	struct net_device *netdev = pci_get_drvdata(pdev);
7729	struct igb_adapter *adapter = netdev_priv(netdev);
7730
7731	if (netif_running(netdev)) {
7732		if (igb_up(adapter)) {
7733			dev_err(&pdev->dev, "igb_up failed after reset\n");
7734			return;
7735		}
7736	}
7737
7738	netif_device_attach(netdev);
7739
7740	/* let the f/w know that the h/w is now under the control of the
7741	 * driver.
7742	 */
7743	igb_get_hw_control(adapter);
7744}
7745
7746static void igb_rar_set_qsel(struct igb_adapter *adapter, u8 *addr, u32 index,
7747			     u8 qsel)
 
 
 
 
7748{
 
7749	u32 rar_low, rar_high;
7750	struct e1000_hw *hw = &adapter->hw;
7751
7752	/* HW expects these in little endian so we reverse the byte order
7753	 * from network order (big endian) to little endian
 
 
7754	 */
7755	rar_low = ((u32) addr[0] | ((u32) addr[1] << 8) |
7756		   ((u32) addr[2] << 16) | ((u32) addr[3] << 24));
7757	rar_high = ((u32) addr[4] | ((u32) addr[5] << 8));
7758
7759	/* Indicate to hardware the Address is Valid. */
7760	rar_high |= E1000_RAH_AV;
 
 
 
 
 
 
 
 
 
 
 
 
7761
7762	if (hw->mac.type == e1000_82575)
7763		rar_high |= E1000_RAH_POOL_1 * qsel;
7764	else
7765		rar_high |= E1000_RAH_POOL_1 << qsel;
 
 
 
 
 
7766
7767	wr32(E1000_RAL(index), rar_low);
7768	wrfl();
7769	wr32(E1000_RAH(index), rar_high);
7770	wrfl();
7771}
7772
7773static int igb_set_vf_mac(struct igb_adapter *adapter,
7774			  int vf, unsigned char *mac_addr)
7775{
7776	struct e1000_hw *hw = &adapter->hw;
7777	/* VF MAC addresses start at end of receive addresses and moves
7778	 * towards the first, as a result a collision should not be possible
7779	 */
7780	int rar_entry = hw->mac.rar_entry_count - (vf + 1);
 
7781
7782	memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
7783
7784	igb_rar_set_qsel(adapter, mac_addr, rar_entry, vf);
 
 
7785
7786	return 0;
7787}
7788
7789static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
7790{
7791	struct igb_adapter *adapter = netdev_priv(netdev);
7792	if (!is_valid_ether_addr(mac) || (vf >= adapter->vfs_allocated_count))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7793		return -EINVAL;
7794	adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
7795	dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n", mac, vf);
7796	dev_info(&adapter->pdev->dev,
7797		 "Reload the VF driver to make this change effective.");
7798	if (test_bit(__IGB_DOWN, &adapter->state)) {
7799		dev_warn(&adapter->pdev->dev,
7800			 "The VF MAC address has been set, but the PF device is not up.\n");
7801		dev_warn(&adapter->pdev->dev,
7802			 "Bring the PF device up before attempting to use the VF device.\n");
7803	}
7804	return igb_set_vf_mac(adapter, vf, mac);
7805}
7806
7807static int igb_link_mbps(int internal_link_speed)
7808{
7809	switch (internal_link_speed) {
7810	case SPEED_100:
7811		return 100;
7812	case SPEED_1000:
7813		return 1000;
7814	default:
7815		return 0;
7816	}
7817}
7818
7819static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
7820				  int link_speed)
7821{
7822	int rf_dec, rf_int;
7823	u32 bcnrc_val;
7824
7825	if (tx_rate != 0) {
7826		/* Calculate the rate factor values to set */
7827		rf_int = link_speed / tx_rate;
7828		rf_dec = (link_speed - (rf_int * tx_rate));
7829		rf_dec = (rf_dec * (1 << E1000_RTTBCNRC_RF_INT_SHIFT)) /
7830			 tx_rate;
7831
7832		bcnrc_val = E1000_RTTBCNRC_RS_ENA;
7833		bcnrc_val |= ((rf_int << E1000_RTTBCNRC_RF_INT_SHIFT) &
7834			      E1000_RTTBCNRC_RF_INT_MASK);
7835		bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
7836	} else {
7837		bcnrc_val = 0;
7838	}
7839
7840	wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
7841	/* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
7842	 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
7843	 */
7844	wr32(E1000_RTTBCNRM, 0x14);
7845	wr32(E1000_RTTBCNRC, bcnrc_val);
7846}
7847
7848static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
7849{
7850	int actual_link_speed, i;
7851	bool reset_rate = false;
7852
7853	/* VF TX rate limit was not set or not supported */
7854	if ((adapter->vf_rate_link_speed == 0) ||
7855	    (adapter->hw.mac.type != e1000_82576))
7856		return;
7857
7858	actual_link_speed = igb_link_mbps(adapter->link_speed);
7859	if (actual_link_speed != adapter->vf_rate_link_speed) {
7860		reset_rate = true;
7861		adapter->vf_rate_link_speed = 0;
7862		dev_info(&adapter->pdev->dev,
7863			 "Link speed has been changed. VF Transmit rate is disabled\n");
7864	}
7865
7866	for (i = 0; i < adapter->vfs_allocated_count; i++) {
7867		if (reset_rate)
7868			adapter->vf_data[i].tx_rate = 0;
7869
7870		igb_set_vf_rate_limit(&adapter->hw, i,
7871				      adapter->vf_data[i].tx_rate,
7872				      actual_link_speed);
7873	}
7874}
7875
7876static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate)
 
7877{
7878	struct igb_adapter *adapter = netdev_priv(netdev);
7879	struct e1000_hw *hw = &adapter->hw;
7880	int actual_link_speed;
7881
7882	if (hw->mac.type != e1000_82576)
7883		return -EOPNOTSUPP;
7884
 
 
 
7885	actual_link_speed = igb_link_mbps(adapter->link_speed);
7886	if ((vf >= adapter->vfs_allocated_count) ||
7887	    (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
7888	    (tx_rate < 0) || (tx_rate > actual_link_speed))
 
7889		return -EINVAL;
7890
7891	adapter->vf_rate_link_speed = actual_link_speed;
7892	adapter->vf_data[vf].tx_rate = (u16)tx_rate;
7893	igb_set_vf_rate_limit(hw, vf, tx_rate, actual_link_speed);
7894
7895	return 0;
7896}
7897
7898static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
7899				   bool setting)
7900{
7901	struct igb_adapter *adapter = netdev_priv(netdev);
7902	struct e1000_hw *hw = &adapter->hw;
7903	u32 reg_val, reg_offset;
7904
7905	if (!adapter->vfs_allocated_count)
7906		return -EOPNOTSUPP;
7907
7908	if (vf >= adapter->vfs_allocated_count)
7909		return -EINVAL;
7910
7911	reg_offset = (hw->mac.type == e1000_82576) ? E1000_DTXSWC : E1000_TXSWC;
7912	reg_val = rd32(reg_offset);
7913	if (setting)
7914		reg_val |= ((1 << vf) |
7915			    (1 << (vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT)));
7916	else
7917		reg_val &= ~((1 << vf) |
7918			     (1 << (vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT)));
7919	wr32(reg_offset, reg_val);
7920
7921	adapter->vf_data[vf].spoofchk_enabled = setting;
7922	return E1000_SUCCESS;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7923}
7924
7925static int igb_ndo_get_vf_config(struct net_device *netdev,
7926				 int vf, struct ifla_vf_info *ivi)
7927{
7928	struct igb_adapter *adapter = netdev_priv(netdev);
7929	if (vf >= adapter->vfs_allocated_count)
7930		return -EINVAL;
7931	ivi->vf = vf;
7932	memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
7933	ivi->tx_rate = adapter->vf_data[vf].tx_rate;
 
7934	ivi->vlan = adapter->vf_data[vf].pf_vlan;
7935	ivi->qos = adapter->vf_data[vf].pf_qos;
7936	ivi->spoofchk = adapter->vf_data[vf].spoofchk_enabled;
 
7937	return 0;
7938}
7939
7940static void igb_vmm_control(struct igb_adapter *adapter)
7941{
7942	struct e1000_hw *hw = &adapter->hw;
7943	u32 reg;
7944
7945	switch (hw->mac.type) {
7946	case e1000_82575:
7947	case e1000_i210:
7948	case e1000_i211:
7949	case e1000_i354:
7950	default:
7951		/* replication is not supported for 82575 */
7952		return;
7953	case e1000_82576:
7954		/* notify HW that the MAC is adding vlan tags */
7955		reg = rd32(E1000_DTXCTL);
7956		reg |= E1000_DTXCTL_VLAN_ADDED;
7957		wr32(E1000_DTXCTL, reg);
 
7958	case e1000_82580:
7959		/* enable replication vlan tag stripping */
7960		reg = rd32(E1000_RPLOLR);
7961		reg |= E1000_RPLOLR_STRVLAN;
7962		wr32(E1000_RPLOLR, reg);
 
7963	case e1000_i350:
7964		/* none of the above registers are supported by i350 */
7965		break;
7966	}
7967
7968	if (adapter->vfs_allocated_count) {
7969		igb_vmdq_set_loopback_pf(hw, true);
7970		igb_vmdq_set_replication_pf(hw, true);
7971		igb_vmdq_set_anti_spoofing_pf(hw, true,
7972					      adapter->vfs_allocated_count);
7973	} else {
7974		igb_vmdq_set_loopback_pf(hw, false);
7975		igb_vmdq_set_replication_pf(hw, false);
7976	}
7977}
7978
7979static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
7980{
7981	struct e1000_hw *hw = &adapter->hw;
7982	u32 dmac_thr;
7983	u16 hwm;
 
7984
7985	if (hw->mac.type > e1000_82580) {
7986		if (adapter->flags & IGB_FLAG_DMAC) {
7987			u32 reg;
7988
7989			/* force threshold to 0. */
7990			wr32(E1000_DMCTXTH, 0);
7991
7992			/* DMA Coalescing high water mark needs to be greater
7993			 * than the Rx threshold. Set hwm to PBA - max frame
7994			 * size in 16B units, capping it at PBA - 6KB.
7995			 */
7996			hwm = 64 * pba - adapter->max_frame_size / 16;
7997			if (hwm < 64 * (pba - 6))
7998				hwm = 64 * (pba - 6);
7999			reg = rd32(E1000_FCRTC);
8000			reg &= ~E1000_FCRTC_RTH_COAL_MASK;
8001			reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
8002				& E1000_FCRTC_RTH_COAL_MASK);
8003			wr32(E1000_FCRTC, reg);
8004
8005			/* Set the DMA Coalescing Rx threshold to PBA - 2 * max
8006			 * frame size, capping it at PBA - 10KB.
8007			 */
8008			dmac_thr = pba - adapter->max_frame_size / 512;
8009			if (dmac_thr < pba - 10)
8010				dmac_thr = pba - 10;
8011			reg = rd32(E1000_DMACR);
8012			reg &= ~E1000_DMACR_DMACTHR_MASK;
8013			reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
8014				& E1000_DMACR_DMACTHR_MASK);
8015
8016			/* transition to L0x or L1 if available..*/
8017			reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
8018
8019			/* watchdog timer= +-1000 usec in 32usec intervals */
8020			reg |= (1000 >> 5);
8021
8022			/* Disable BMC-to-OS Watchdog Enable */
8023			if (hw->mac.type != e1000_i354)
8024				reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
8025
8026			wr32(E1000_DMACR, reg);
8027
8028			/* no lower threshold to disable
8029			 * coalescing(smart fifb)-UTRESH=0
8030			 */
8031			wr32(E1000_DMCRTRH, 0);
8032
8033			reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
8034
8035			wr32(E1000_DMCTLX, reg);
8036
8037			/* free space in tx packet buffer to wake from
8038			 * DMA coal
8039			 */
8040			wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
8041			     (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
 
8042
8043			/* make low power state decision controlled
8044			 * by DMA coal
8045			 */
8046			reg = rd32(E1000_PCIEMISC);
8047			reg &= ~E1000_PCIEMISC_LX_DECISION;
8048			wr32(E1000_PCIEMISC, reg);
8049		} /* endif adapter->dmac is not disabled */
8050	} else if (hw->mac.type == e1000_82580) {
8051		u32 reg = rd32(E1000_PCIEMISC);
 
8052		wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
8053		wr32(E1000_DMACR, 0);
8054	}
8055}
8056
8057/**
8058 *  igb_read_i2c_byte - Reads 8 bit word over I2C
8059 *  @hw: pointer to hardware structure
8060 *  @byte_offset: byte offset to read
8061 *  @dev_addr: device address
8062 *  @data: value read
8063 *
8064 *  Performs byte read operation over I2C interface at
8065 *  a specified device address.
8066 **/
8067s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
8068		      u8 dev_addr, u8 *data)
8069{
8070	struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
8071	struct i2c_client *this_client = adapter->i2c_client;
8072	s32 status;
8073	u16 swfw_mask = 0;
8074
8075	if (!this_client)
8076		return E1000_ERR_I2C;
8077
8078	swfw_mask = E1000_SWFW_PHY0_SM;
8079
8080	if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask)
8081	    != E1000_SUCCESS)
8082		return E1000_ERR_SWFW_SYNC;
8083
8084	status = i2c_smbus_read_byte_data(this_client, byte_offset);
8085	hw->mac.ops.release_swfw_sync(hw, swfw_mask);
8086
8087	if (status < 0)
8088		return E1000_ERR_I2C;
8089	else {
8090		*data = status;
8091		return E1000_SUCCESS;
8092	}
8093}
8094
8095/**
8096 *  igb_write_i2c_byte - Writes 8 bit word over I2C
8097 *  @hw: pointer to hardware structure
8098 *  @byte_offset: byte offset to write
8099 *  @dev_addr: device address
8100 *  @data: value to write
8101 *
8102 *  Performs byte write operation over I2C interface at
8103 *  a specified device address.
8104 **/
8105s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
8106		       u8 dev_addr, u8 data)
8107{
8108	struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
8109	struct i2c_client *this_client = adapter->i2c_client;
8110	s32 status;
8111	u16 swfw_mask = E1000_SWFW_PHY0_SM;
8112
8113	if (!this_client)
8114		return E1000_ERR_I2C;
8115
8116	if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask) != E1000_SUCCESS)
8117		return E1000_ERR_SWFW_SYNC;
8118	status = i2c_smbus_write_byte_data(this_client, byte_offset, data);
8119	hw->mac.ops.release_swfw_sync(hw, swfw_mask);
8120
8121	if (status)
8122		return E1000_ERR_I2C;
8123	else
8124		return E1000_SUCCESS;
8125
8126}
8127
8128int igb_reinit_queues(struct igb_adapter *adapter)
8129{
8130	struct net_device *netdev = adapter->netdev;
8131	struct pci_dev *pdev = adapter->pdev;
8132	int err = 0;
8133
8134	if (netif_running(netdev))
8135		igb_close(netdev);
8136
8137	igb_reset_interrupt_capability(adapter);
8138
8139	if (igb_init_interrupt_scheme(adapter, true)) {
8140		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
8141		return -ENOMEM;
8142	}
8143
8144	if (netif_running(netdev))
8145		err = igb_open(netdev);
8146
8147	return err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8148}
8149/* igb_main.c */