Loading...
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#include <linux/lockdep.h>
37#ifdef CONFIG_IGB_DCA
38#include <linux/dca.h>
39#endif
40#include <linux/i2c.h>
41#include "igb.h"
42
43enum queue_mode {
44 QUEUE_MODE_STRICT_PRIORITY,
45 QUEUE_MODE_STREAM_RESERVATION,
46};
47
48enum tx_queue_prio {
49 TX_QUEUE_PRIO_HIGH,
50 TX_QUEUE_PRIO_LOW,
51};
52
53char igb_driver_name[] = "igb";
54static const char igb_driver_string[] =
55 "Intel(R) Gigabit Ethernet Network Driver";
56static const char igb_copyright[] =
57 "Copyright (c) 2007-2014 Intel Corporation.";
58
59static const struct e1000_info *igb_info_tbl[] = {
60 [board_82575] = &e1000_82575_info,
61};
62
63static const struct pci_device_id igb_pci_tbl[] = {
64 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_1GBPS) },
65 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_SGMII) },
66 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) },
67 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I211_COPPER), board_82575 },
68 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 },
69 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 },
70 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 },
71 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 },
72 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER_FLASHLESS), board_82575 },
73 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES_FLASHLESS), board_82575 },
74 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
75 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
76 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
77 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
78 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
79 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
80 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 },
81 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
82 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
83 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
84 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 },
85 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 },
86 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 },
87 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 },
88 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
89 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
90 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
91 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
92 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
93 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
94 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
95 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
96 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
97 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
98 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
99 /* required last entry */
100 {0, }
101};
102
103MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
104
105static int igb_setup_all_tx_resources(struct igb_adapter *);
106static int igb_setup_all_rx_resources(struct igb_adapter *);
107static void igb_free_all_tx_resources(struct igb_adapter *);
108static void igb_free_all_rx_resources(struct igb_adapter *);
109static void igb_setup_mrqc(struct igb_adapter *);
110static void igb_init_queue_configuration(struct igb_adapter *adapter);
111static int igb_sw_init(struct igb_adapter *);
112int igb_open(struct net_device *);
113int igb_close(struct net_device *);
114static void igb_configure(struct igb_adapter *);
115static void igb_configure_tx(struct igb_adapter *);
116static void igb_configure_rx(struct igb_adapter *);
117static void igb_clean_all_tx_rings(struct igb_adapter *);
118static void igb_clean_all_rx_rings(struct igb_adapter *);
119static void igb_clean_tx_ring(struct igb_ring *);
120static void igb_clean_rx_ring(struct igb_ring *);
121static void igb_set_rx_mode(struct net_device *);
122static void igb_update_phy_info(struct timer_list *);
123static void igb_watchdog(struct timer_list *);
124static void igb_watchdog_task(struct work_struct *);
125static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *);
126static void igb_get_stats64(struct net_device *dev,
127 struct rtnl_link_stats64 *stats);
128static int igb_change_mtu(struct net_device *, int);
129static int igb_set_mac(struct net_device *, void *);
130static void igb_set_uta(struct igb_adapter *adapter, bool set);
131static irqreturn_t igb_intr(int irq, void *);
132static irqreturn_t igb_intr_msi(int irq, void *);
133static irqreturn_t igb_msix_other(int irq, void *);
134static irqreturn_t igb_msix_ring(int irq, void *);
135#ifdef CONFIG_IGB_DCA
136static void igb_update_dca(struct igb_q_vector *);
137static void igb_setup_dca(struct igb_adapter *);
138#endif /* CONFIG_IGB_DCA */
139static int igb_poll(struct napi_struct *, int);
140static bool igb_clean_tx_irq(struct igb_q_vector *, int);
141static int igb_clean_rx_irq(struct igb_q_vector *, int);
142static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
143static void igb_tx_timeout(struct net_device *, unsigned int txqueue);
144static void igb_reset_task(struct work_struct *);
145static void igb_vlan_mode(struct net_device *netdev,
146 netdev_features_t features);
147static int igb_vlan_rx_add_vid(struct net_device *, __be16, u16);
148static int igb_vlan_rx_kill_vid(struct net_device *, __be16, u16);
149static void igb_restore_vlan(struct igb_adapter *);
150static void igb_rar_set_index(struct igb_adapter *, u32);
151static void igb_ping_all_vfs(struct igb_adapter *);
152static void igb_msg_task(struct igb_adapter *);
153static void igb_vmm_control(struct igb_adapter *);
154static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
155static void igb_flush_mac_table(struct igb_adapter *);
156static int igb_available_rars(struct igb_adapter *, u8);
157static void igb_set_default_mac_filter(struct igb_adapter *);
158static int igb_uc_sync(struct net_device *, const unsigned char *);
159static int igb_uc_unsync(struct net_device *, const unsigned char *);
160static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
161static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
162static int igb_ndo_set_vf_vlan(struct net_device *netdev,
163 int vf, u16 vlan, u8 qos, __be16 vlan_proto);
164static int igb_ndo_set_vf_bw(struct net_device *, int, int, int);
165static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
166 bool setting);
167static int igb_ndo_set_vf_trust(struct net_device *netdev, int vf,
168 bool setting);
169static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
170 struct ifla_vf_info *ivi);
171static void igb_check_vf_rate_limit(struct igb_adapter *);
172static void igb_nfc_filter_exit(struct igb_adapter *adapter);
173static void igb_nfc_filter_restore(struct igb_adapter *adapter);
174
175#ifdef CONFIG_PCI_IOV
176static int igb_vf_configure(struct igb_adapter *adapter, int vf);
177static int igb_disable_sriov(struct pci_dev *dev, bool reinit);
178#endif
179
180#ifdef CONFIG_IGB_DCA
181static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
182static struct notifier_block dca_notifier = {
183 .notifier_call = igb_notify_dca,
184 .next = NULL,
185 .priority = 0
186};
187#endif
188#ifdef CONFIG_PCI_IOV
189static unsigned int max_vfs;
190module_param(max_vfs, uint, 0444);
191MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate per physical function");
192#endif /* CONFIG_PCI_IOV */
193
194static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
195 pci_channel_state_t);
196static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
197static void igb_io_resume(struct pci_dev *);
198
199static const struct pci_error_handlers igb_err_handler = {
200 .error_detected = igb_io_error_detected,
201 .slot_reset = igb_io_slot_reset,
202 .resume = igb_io_resume,
203};
204
205static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
206
207MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
208MODULE_LICENSE("GPL v2");
209
210#define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
211static int debug = -1;
212module_param(debug, int, 0);
213MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
214
215struct igb_reg_info {
216 u32 ofs;
217 char *name;
218};
219
220static const struct igb_reg_info igb_reg_info_tbl[] = {
221
222 /* General Registers */
223 {E1000_CTRL, "CTRL"},
224 {E1000_STATUS, "STATUS"},
225 {E1000_CTRL_EXT, "CTRL_EXT"},
226
227 /* Interrupt Registers */
228 {E1000_ICR, "ICR"},
229
230 /* RX Registers */
231 {E1000_RCTL, "RCTL"},
232 {E1000_RDLEN(0), "RDLEN"},
233 {E1000_RDH(0), "RDH"},
234 {E1000_RDT(0), "RDT"},
235 {E1000_RXDCTL(0), "RXDCTL"},
236 {E1000_RDBAL(0), "RDBAL"},
237 {E1000_RDBAH(0), "RDBAH"},
238
239 /* TX Registers */
240 {E1000_TCTL, "TCTL"},
241 {E1000_TDBAL(0), "TDBAL"},
242 {E1000_TDBAH(0), "TDBAH"},
243 {E1000_TDLEN(0), "TDLEN"},
244 {E1000_TDH(0), "TDH"},
245 {E1000_TDT(0), "TDT"},
246 {E1000_TXDCTL(0), "TXDCTL"},
247 {E1000_TDFH, "TDFH"},
248 {E1000_TDFT, "TDFT"},
249 {E1000_TDFHS, "TDFHS"},
250 {E1000_TDFPC, "TDFPC"},
251
252 /* List Terminator */
253 {}
254};
255
256/* igb_regdump - register printout routine */
257static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo)
258{
259 int n = 0;
260 char rname[16];
261 u32 regs[8];
262
263 switch (reginfo->ofs) {
264 case E1000_RDLEN(0):
265 for (n = 0; n < 4; n++)
266 regs[n] = rd32(E1000_RDLEN(n));
267 break;
268 case E1000_RDH(0):
269 for (n = 0; n < 4; n++)
270 regs[n] = rd32(E1000_RDH(n));
271 break;
272 case E1000_RDT(0):
273 for (n = 0; n < 4; n++)
274 regs[n] = rd32(E1000_RDT(n));
275 break;
276 case E1000_RXDCTL(0):
277 for (n = 0; n < 4; n++)
278 regs[n] = rd32(E1000_RXDCTL(n));
279 break;
280 case E1000_RDBAL(0):
281 for (n = 0; n < 4; n++)
282 regs[n] = rd32(E1000_RDBAL(n));
283 break;
284 case E1000_RDBAH(0):
285 for (n = 0; n < 4; n++)
286 regs[n] = rd32(E1000_RDBAH(n));
287 break;
288 case E1000_TDBAL(0):
289 for (n = 0; n < 4; n++)
290 regs[n] = rd32(E1000_TDBAL(n));
291 break;
292 case E1000_TDBAH(0):
293 for (n = 0; n < 4; n++)
294 regs[n] = rd32(E1000_TDBAH(n));
295 break;
296 case E1000_TDLEN(0):
297 for (n = 0; n < 4; n++)
298 regs[n] = rd32(E1000_TDLEN(n));
299 break;
300 case E1000_TDH(0):
301 for (n = 0; n < 4; n++)
302 regs[n] = rd32(E1000_TDH(n));
303 break;
304 case E1000_TDT(0):
305 for (n = 0; n < 4; n++)
306 regs[n] = rd32(E1000_TDT(n));
307 break;
308 case E1000_TXDCTL(0):
309 for (n = 0; n < 4; n++)
310 regs[n] = rd32(E1000_TXDCTL(n));
311 break;
312 default:
313 pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs));
314 return;
315 }
316
317 snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]");
318 pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1],
319 regs[2], regs[3]);
320}
321
322/* igb_dump - Print registers, Tx-rings and Rx-rings */
323static void igb_dump(struct igb_adapter *adapter)
324{
325 struct net_device *netdev = adapter->netdev;
326 struct e1000_hw *hw = &adapter->hw;
327 struct igb_reg_info *reginfo;
328 struct igb_ring *tx_ring;
329 union e1000_adv_tx_desc *tx_desc;
330 struct my_u0 { __le64 a; __le64 b; } *u0;
331 struct igb_ring *rx_ring;
332 union e1000_adv_rx_desc *rx_desc;
333 u32 staterr;
334 u16 i, n;
335
336 if (!netif_msg_hw(adapter))
337 return;
338
339 /* Print netdevice Info */
340 if (netdev) {
341 dev_info(&adapter->pdev->dev, "Net device Info\n");
342 pr_info("Device Name state trans_start\n");
343 pr_info("%-15s %016lX %016lX\n", netdev->name,
344 netdev->state, dev_trans_start(netdev));
345 }
346
347 /* Print Registers */
348 dev_info(&adapter->pdev->dev, "Register Dump\n");
349 pr_info(" Register Name Value\n");
350 for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
351 reginfo->name; reginfo++) {
352 igb_regdump(hw, reginfo);
353 }
354
355 /* Print TX Ring Summary */
356 if (!netdev || !netif_running(netdev))
357 goto exit;
358
359 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
360 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
361 for (n = 0; n < adapter->num_tx_queues; n++) {
362 struct igb_tx_buffer *buffer_info;
363 tx_ring = adapter->tx_ring[n];
364 buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
365 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
366 n, tx_ring->next_to_use, tx_ring->next_to_clean,
367 (u64)dma_unmap_addr(buffer_info, dma),
368 dma_unmap_len(buffer_info, len),
369 buffer_info->next_to_watch,
370 (u64)buffer_info->time_stamp);
371 }
372
373 /* Print TX Rings */
374 if (!netif_msg_tx_done(adapter))
375 goto rx_ring_summary;
376
377 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
378
379 /* Transmit Descriptor Formats
380 *
381 * Advanced Transmit Descriptor
382 * +--------------------------------------------------------------+
383 * 0 | Buffer Address [63:0] |
384 * +--------------------------------------------------------------+
385 * 8 | PAYLEN | PORTS |CC|IDX | STA | DCMD |DTYP|MAC|RSV| DTALEN |
386 * +--------------------------------------------------------------+
387 * 63 46 45 40 39 38 36 35 32 31 24 15 0
388 */
389
390 for (n = 0; n < adapter->num_tx_queues; n++) {
391 tx_ring = adapter->tx_ring[n];
392 pr_info("------------------------------------\n");
393 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
394 pr_info("------------------------------------\n");
395 pr_info("T [desc] [address 63:0 ] [PlPOCIStDDM Ln] [bi->dma ] leng ntw timestamp bi->skb\n");
396
397 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
398 const char *next_desc;
399 struct igb_tx_buffer *buffer_info;
400 tx_desc = IGB_TX_DESC(tx_ring, i);
401 buffer_info = &tx_ring->tx_buffer_info[i];
402 u0 = (struct my_u0 *)tx_desc;
403 if (i == tx_ring->next_to_use &&
404 i == tx_ring->next_to_clean)
405 next_desc = " NTC/U";
406 else if (i == tx_ring->next_to_use)
407 next_desc = " NTU";
408 else if (i == tx_ring->next_to_clean)
409 next_desc = " NTC";
410 else
411 next_desc = "";
412
413 pr_info("T [0x%03X] %016llX %016llX %016llX %04X %p %016llX %p%s\n",
414 i, le64_to_cpu(u0->a),
415 le64_to_cpu(u0->b),
416 (u64)dma_unmap_addr(buffer_info, dma),
417 dma_unmap_len(buffer_info, len),
418 buffer_info->next_to_watch,
419 (u64)buffer_info->time_stamp,
420 buffer_info->skb, next_desc);
421
422 if (netif_msg_pktdata(adapter) && buffer_info->skb)
423 print_hex_dump(KERN_INFO, "",
424 DUMP_PREFIX_ADDRESS,
425 16, 1, buffer_info->skb->data,
426 dma_unmap_len(buffer_info, len),
427 true);
428 }
429 }
430
431 /* Print RX Rings Summary */
432rx_ring_summary:
433 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
434 pr_info("Queue [NTU] [NTC]\n");
435 for (n = 0; n < adapter->num_rx_queues; n++) {
436 rx_ring = adapter->rx_ring[n];
437 pr_info(" %5d %5X %5X\n",
438 n, rx_ring->next_to_use, rx_ring->next_to_clean);
439 }
440
441 /* Print RX Rings */
442 if (!netif_msg_rx_status(adapter))
443 goto exit;
444
445 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
446
447 /* Advanced Receive Descriptor (Read) Format
448 * 63 1 0
449 * +-----------------------------------------------------+
450 * 0 | Packet Buffer Address [63:1] |A0/NSE|
451 * +----------------------------------------------+------+
452 * 8 | Header Buffer Address [63:1] | DD |
453 * +-----------------------------------------------------+
454 *
455 *
456 * Advanced Receive Descriptor (Write-Back) Format
457 *
458 * 63 48 47 32 31 30 21 20 17 16 4 3 0
459 * +------------------------------------------------------+
460 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
461 * | Checksum Ident | | | | Type | Type |
462 * +------------------------------------------------------+
463 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
464 * +------------------------------------------------------+
465 * 63 48 47 32 31 20 19 0
466 */
467
468 for (n = 0; n < adapter->num_rx_queues; n++) {
469 rx_ring = adapter->rx_ring[n];
470 pr_info("------------------------------------\n");
471 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
472 pr_info("------------------------------------\n");
473 pr_info("R [desc] [ PktBuf A0] [ HeadBuf DD] [bi->dma ] [bi->skb] <-- Adv Rx Read format\n");
474 pr_info("RWB[desc] [PcsmIpSHl PtRs] [vl er S cks ln] ---------------- [bi->skb] <-- Adv Rx Write-Back format\n");
475
476 for (i = 0; i < rx_ring->count; i++) {
477 const char *next_desc;
478 struct igb_rx_buffer *buffer_info;
479 buffer_info = &rx_ring->rx_buffer_info[i];
480 rx_desc = IGB_RX_DESC(rx_ring, i);
481 u0 = (struct my_u0 *)rx_desc;
482 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
483
484 if (i == rx_ring->next_to_use)
485 next_desc = " NTU";
486 else if (i == rx_ring->next_to_clean)
487 next_desc = " NTC";
488 else
489 next_desc = "";
490
491 if (staterr & E1000_RXD_STAT_DD) {
492 /* Descriptor Done */
493 pr_info("%s[0x%03X] %016llX %016llX ---------------- %s\n",
494 "RWB", i,
495 le64_to_cpu(u0->a),
496 le64_to_cpu(u0->b),
497 next_desc);
498 } else {
499 pr_info("%s[0x%03X] %016llX %016llX %016llX %s\n",
500 "R ", i,
501 le64_to_cpu(u0->a),
502 le64_to_cpu(u0->b),
503 (u64)buffer_info->dma,
504 next_desc);
505
506 if (netif_msg_pktdata(adapter) &&
507 buffer_info->dma && buffer_info->page) {
508 print_hex_dump(KERN_INFO, "",
509 DUMP_PREFIX_ADDRESS,
510 16, 1,
511 page_address(buffer_info->page) +
512 buffer_info->page_offset,
513 igb_rx_bufsz(rx_ring), true);
514 }
515 }
516 }
517 }
518
519exit:
520 return;
521}
522
523/**
524 * igb_get_i2c_data - Reads the I2C SDA data bit
525 * @data: opaque pointer to adapter struct
526 *
527 * Returns the I2C data bit value
528 **/
529static int igb_get_i2c_data(void *data)
530{
531 struct igb_adapter *adapter = (struct igb_adapter *)data;
532 struct e1000_hw *hw = &adapter->hw;
533 s32 i2cctl = rd32(E1000_I2CPARAMS);
534
535 return !!(i2cctl & E1000_I2C_DATA_IN);
536}
537
538/**
539 * igb_set_i2c_data - Sets the I2C data bit
540 * @data: pointer to hardware structure
541 * @state: I2C data value (0 or 1) to set
542 *
543 * Sets the I2C data bit
544 **/
545static void igb_set_i2c_data(void *data, int state)
546{
547 struct igb_adapter *adapter = (struct igb_adapter *)data;
548 struct e1000_hw *hw = &adapter->hw;
549 s32 i2cctl = rd32(E1000_I2CPARAMS);
550
551 if (state) {
552 i2cctl |= E1000_I2C_DATA_OUT | E1000_I2C_DATA_OE_N;
553 } else {
554 i2cctl &= ~E1000_I2C_DATA_OE_N;
555 i2cctl &= ~E1000_I2C_DATA_OUT;
556 }
557
558 wr32(E1000_I2CPARAMS, i2cctl);
559 wrfl();
560}
561
562/**
563 * igb_set_i2c_clk - Sets the I2C SCL clock
564 * @data: pointer to hardware structure
565 * @state: state to set clock
566 *
567 * Sets the I2C clock line to state
568 **/
569static void igb_set_i2c_clk(void *data, int state)
570{
571 struct igb_adapter *adapter = (struct igb_adapter *)data;
572 struct e1000_hw *hw = &adapter->hw;
573 s32 i2cctl = rd32(E1000_I2CPARAMS);
574
575 if (state) {
576 i2cctl |= E1000_I2C_CLK_OUT | E1000_I2C_CLK_OE_N;
577 } else {
578 i2cctl &= ~E1000_I2C_CLK_OUT;
579 i2cctl &= ~E1000_I2C_CLK_OE_N;
580 }
581 wr32(E1000_I2CPARAMS, i2cctl);
582 wrfl();
583}
584
585/**
586 * igb_get_i2c_clk - Gets the I2C SCL clock state
587 * @data: pointer to hardware structure
588 *
589 * Gets the I2C clock state
590 **/
591static int igb_get_i2c_clk(void *data)
592{
593 struct igb_adapter *adapter = (struct igb_adapter *)data;
594 struct e1000_hw *hw = &adapter->hw;
595 s32 i2cctl = rd32(E1000_I2CPARAMS);
596
597 return !!(i2cctl & E1000_I2C_CLK_IN);
598}
599
600static const struct i2c_algo_bit_data igb_i2c_algo = {
601 .setsda = igb_set_i2c_data,
602 .setscl = igb_set_i2c_clk,
603 .getsda = igb_get_i2c_data,
604 .getscl = igb_get_i2c_clk,
605 .udelay = 5,
606 .timeout = 20,
607};
608
609/**
610 * igb_get_hw_dev - return device
611 * @hw: pointer to hardware structure
612 *
613 * used by hardware layer to print debugging information
614 **/
615struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
616{
617 struct igb_adapter *adapter = hw->back;
618 return adapter->netdev;
619}
620
621static struct pci_driver igb_driver;
622
623/**
624 * igb_init_module - Driver Registration Routine
625 *
626 * igb_init_module is the first routine called when the driver is
627 * loaded. All it does is register with the PCI subsystem.
628 **/
629static int __init igb_init_module(void)
630{
631 int ret;
632
633 pr_info("%s\n", igb_driver_string);
634 pr_info("%s\n", igb_copyright);
635
636#ifdef CONFIG_IGB_DCA
637 dca_register_notify(&dca_notifier);
638#endif
639 ret = pci_register_driver(&igb_driver);
640#ifdef CONFIG_IGB_DCA
641 if (ret)
642 dca_unregister_notify(&dca_notifier);
643#endif
644 return ret;
645}
646
647module_init(igb_init_module);
648
649/**
650 * igb_exit_module - Driver Exit Cleanup Routine
651 *
652 * igb_exit_module is called just before the driver is removed
653 * from memory.
654 **/
655static void __exit igb_exit_module(void)
656{
657#ifdef CONFIG_IGB_DCA
658 dca_unregister_notify(&dca_notifier);
659#endif
660 pci_unregister_driver(&igb_driver);
661}
662
663module_exit(igb_exit_module);
664
665#define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
666/**
667 * igb_cache_ring_register - Descriptor ring to register mapping
668 * @adapter: board private structure to initialize
669 *
670 * Once we know the feature-set enabled for the device, we'll cache
671 * the register offset the descriptor ring is assigned to.
672 **/
673static void igb_cache_ring_register(struct igb_adapter *adapter)
674{
675 int i = 0, j = 0;
676 u32 rbase_offset = adapter->vfs_allocated_count;
677
678 switch (adapter->hw.mac.type) {
679 case e1000_82576:
680 /* The queues are allocated for virtualization such that VF 0
681 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
682 * In order to avoid collision we start at the first free queue
683 * and continue consuming queues in the same sequence
684 */
685 if (adapter->vfs_allocated_count) {
686 for (; i < adapter->rss_queues; i++)
687 adapter->rx_ring[i]->reg_idx = rbase_offset +
688 Q_IDX_82576(i);
689 }
690 fallthrough;
691 case e1000_82575:
692 case e1000_82580:
693 case e1000_i350:
694 case e1000_i354:
695 case e1000_i210:
696 case e1000_i211:
697 default:
698 for (; i < adapter->num_rx_queues; i++)
699 adapter->rx_ring[i]->reg_idx = rbase_offset + i;
700 for (; j < adapter->num_tx_queues; j++)
701 adapter->tx_ring[j]->reg_idx = rbase_offset + j;
702 break;
703 }
704}
705
706u32 igb_rd32(struct e1000_hw *hw, u32 reg)
707{
708 struct igb_adapter *igb = container_of(hw, struct igb_adapter, hw);
709 u8 __iomem *hw_addr = READ_ONCE(hw->hw_addr);
710 u32 value = 0;
711
712 if (E1000_REMOVED(hw_addr))
713 return ~value;
714
715 value = readl(&hw_addr[reg]);
716
717 /* reads should not return all F's */
718 if (!(~value) && (!reg || !(~readl(hw_addr)))) {
719 struct net_device *netdev = igb->netdev;
720 hw->hw_addr = NULL;
721 netdev_err(netdev, "PCIe link lost\n");
722 WARN(pci_device_is_present(igb->pdev),
723 "igb: Failed to read reg 0x%x!\n", reg);
724 }
725
726 return value;
727}
728
729/**
730 * igb_write_ivar - configure ivar for given MSI-X vector
731 * @hw: pointer to the HW structure
732 * @msix_vector: vector number we are allocating to a given ring
733 * @index: row index of IVAR register to write within IVAR table
734 * @offset: column offset of in IVAR, should be multiple of 8
735 *
736 * This function is intended to handle the writing of the IVAR register
737 * for adapters 82576 and newer. The IVAR table consists of 2 columns,
738 * each containing an cause allocation for an Rx and Tx ring, and a
739 * variable number of rows depending on the number of queues supported.
740 **/
741static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
742 int index, int offset)
743{
744 u32 ivar = array_rd32(E1000_IVAR0, index);
745
746 /* clear any bits that are currently set */
747 ivar &= ~((u32)0xFF << offset);
748
749 /* write vector and valid bit */
750 ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
751
752 array_wr32(E1000_IVAR0, index, ivar);
753}
754
755#define IGB_N0_QUEUE -1
756static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
757{
758 struct igb_adapter *adapter = q_vector->adapter;
759 struct e1000_hw *hw = &adapter->hw;
760 int rx_queue = IGB_N0_QUEUE;
761 int tx_queue = IGB_N0_QUEUE;
762 u32 msixbm = 0;
763
764 if (q_vector->rx.ring)
765 rx_queue = q_vector->rx.ring->reg_idx;
766 if (q_vector->tx.ring)
767 tx_queue = q_vector->tx.ring->reg_idx;
768
769 switch (hw->mac.type) {
770 case e1000_82575:
771 /* The 82575 assigns vectors using a bitmask, which matches the
772 * bitmask for the EICR/EIMS/EIMC registers. To assign one
773 * or more queues to a vector, we write the appropriate bits
774 * into the MSIXBM register for that vector.
775 */
776 if (rx_queue > IGB_N0_QUEUE)
777 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
778 if (tx_queue > IGB_N0_QUEUE)
779 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
780 if (!(adapter->flags & IGB_FLAG_HAS_MSIX) && msix_vector == 0)
781 msixbm |= E1000_EIMS_OTHER;
782 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
783 q_vector->eims_value = msixbm;
784 break;
785 case e1000_82576:
786 /* 82576 uses a table that essentially consists of 2 columns
787 * with 8 rows. The ordering is column-major so we use the
788 * lower 3 bits as the row index, and the 4th bit as the
789 * column offset.
790 */
791 if (rx_queue > IGB_N0_QUEUE)
792 igb_write_ivar(hw, msix_vector,
793 rx_queue & 0x7,
794 (rx_queue & 0x8) << 1);
795 if (tx_queue > IGB_N0_QUEUE)
796 igb_write_ivar(hw, msix_vector,
797 tx_queue & 0x7,
798 ((tx_queue & 0x8) << 1) + 8);
799 q_vector->eims_value = BIT(msix_vector);
800 break;
801 case e1000_82580:
802 case e1000_i350:
803 case e1000_i354:
804 case e1000_i210:
805 case e1000_i211:
806 /* On 82580 and newer adapters the scheme is similar to 82576
807 * however instead of ordering column-major we have things
808 * ordered row-major. So we traverse the table by using
809 * bit 0 as the column offset, and the remaining bits as the
810 * row index.
811 */
812 if (rx_queue > IGB_N0_QUEUE)
813 igb_write_ivar(hw, msix_vector,
814 rx_queue >> 1,
815 (rx_queue & 0x1) << 4);
816 if (tx_queue > IGB_N0_QUEUE)
817 igb_write_ivar(hw, msix_vector,
818 tx_queue >> 1,
819 ((tx_queue & 0x1) << 4) + 8);
820 q_vector->eims_value = BIT(msix_vector);
821 break;
822 default:
823 BUG();
824 break;
825 }
826
827 /* add q_vector eims value to global eims_enable_mask */
828 adapter->eims_enable_mask |= q_vector->eims_value;
829
830 /* configure q_vector to set itr on first interrupt */
831 q_vector->set_itr = 1;
832}
833
834/**
835 * igb_configure_msix - Configure MSI-X hardware
836 * @adapter: board private structure to initialize
837 *
838 * igb_configure_msix sets up the hardware to properly
839 * generate MSI-X interrupts.
840 **/
841static void igb_configure_msix(struct igb_adapter *adapter)
842{
843 u32 tmp;
844 int i, vector = 0;
845 struct e1000_hw *hw = &adapter->hw;
846
847 adapter->eims_enable_mask = 0;
848
849 /* set vector for other causes, i.e. link changes */
850 switch (hw->mac.type) {
851 case e1000_82575:
852 tmp = rd32(E1000_CTRL_EXT);
853 /* enable MSI-X PBA support*/
854 tmp |= E1000_CTRL_EXT_PBA_CLR;
855
856 /* Auto-Mask interrupts upon ICR read. */
857 tmp |= E1000_CTRL_EXT_EIAME;
858 tmp |= E1000_CTRL_EXT_IRCA;
859
860 wr32(E1000_CTRL_EXT, tmp);
861
862 /* enable msix_other interrupt */
863 array_wr32(E1000_MSIXBM(0), vector++, E1000_EIMS_OTHER);
864 adapter->eims_other = E1000_EIMS_OTHER;
865
866 break;
867
868 case e1000_82576:
869 case e1000_82580:
870 case e1000_i350:
871 case e1000_i354:
872 case e1000_i210:
873 case e1000_i211:
874 /* Turn on MSI-X capability first, or our settings
875 * won't stick. And it will take days to debug.
876 */
877 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
878 E1000_GPIE_PBA | E1000_GPIE_EIAME |
879 E1000_GPIE_NSICR);
880
881 /* enable msix_other interrupt */
882 adapter->eims_other = BIT(vector);
883 tmp = (vector++ | E1000_IVAR_VALID) << 8;
884
885 wr32(E1000_IVAR_MISC, tmp);
886 break;
887 default:
888 /* do nothing, since nothing else supports MSI-X */
889 break;
890 } /* switch (hw->mac.type) */
891
892 adapter->eims_enable_mask |= adapter->eims_other;
893
894 for (i = 0; i < adapter->num_q_vectors; i++)
895 igb_assign_vector(adapter->q_vector[i], vector++);
896
897 wrfl();
898}
899
900/**
901 * igb_request_msix - Initialize MSI-X interrupts
902 * @adapter: board private structure to initialize
903 *
904 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
905 * kernel.
906 **/
907static int igb_request_msix(struct igb_adapter *adapter)
908{
909 unsigned int num_q_vectors = adapter->num_q_vectors;
910 struct net_device *netdev = adapter->netdev;
911 int i, err = 0, vector = 0, free_vector = 0;
912
913 err = request_irq(adapter->msix_entries[vector].vector,
914 igb_msix_other, 0, netdev->name, adapter);
915 if (err)
916 goto err_out;
917
918 if (num_q_vectors > MAX_Q_VECTORS) {
919 num_q_vectors = MAX_Q_VECTORS;
920 dev_warn(&adapter->pdev->dev,
921 "The number of queue vectors (%d) is higher than max allowed (%d)\n",
922 adapter->num_q_vectors, MAX_Q_VECTORS);
923 }
924 for (i = 0; i < num_q_vectors; i++) {
925 struct igb_q_vector *q_vector = adapter->q_vector[i];
926
927 vector++;
928
929 q_vector->itr_register = adapter->io_addr + E1000_EITR(vector);
930
931 if (q_vector->rx.ring && q_vector->tx.ring)
932 sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
933 q_vector->rx.ring->queue_index);
934 else if (q_vector->tx.ring)
935 sprintf(q_vector->name, "%s-tx-%u", netdev->name,
936 q_vector->tx.ring->queue_index);
937 else if (q_vector->rx.ring)
938 sprintf(q_vector->name, "%s-rx-%u", netdev->name,
939 q_vector->rx.ring->queue_index);
940 else
941 sprintf(q_vector->name, "%s-unused", netdev->name);
942
943 err = request_irq(adapter->msix_entries[vector].vector,
944 igb_msix_ring, 0, q_vector->name,
945 q_vector);
946 if (err)
947 goto err_free;
948 }
949
950 igb_configure_msix(adapter);
951 return 0;
952
953err_free:
954 /* free already assigned IRQs */
955 free_irq(adapter->msix_entries[free_vector++].vector, adapter);
956
957 vector--;
958 for (i = 0; i < vector; i++) {
959 free_irq(adapter->msix_entries[free_vector++].vector,
960 adapter->q_vector[i]);
961 }
962err_out:
963 return err;
964}
965
966/**
967 * igb_free_q_vector - Free memory allocated for specific interrupt vector
968 * @adapter: board private structure to initialize
969 * @v_idx: Index of vector to be freed
970 *
971 * This function frees the memory allocated to the q_vector.
972 **/
973static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx)
974{
975 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
976
977 adapter->q_vector[v_idx] = NULL;
978
979 /* igb_get_stats64() might access the rings on this vector,
980 * we must wait a grace period before freeing it.
981 */
982 if (q_vector)
983 kfree_rcu(q_vector, rcu);
984}
985
986/**
987 * igb_reset_q_vector - Reset config for interrupt vector
988 * @adapter: board private structure to initialize
989 * @v_idx: Index of vector to be reset
990 *
991 * If NAPI is enabled it will delete any references to the
992 * NAPI struct. This is preparation for igb_free_q_vector.
993 **/
994static void igb_reset_q_vector(struct igb_adapter *adapter, int v_idx)
995{
996 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
997
998 /* Coming from igb_set_interrupt_capability, the vectors are not yet
999 * allocated. So, q_vector is NULL so we should stop here.
1000 */
1001 if (!q_vector)
1002 return;
1003
1004 if (q_vector->tx.ring)
1005 adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
1006
1007 if (q_vector->rx.ring)
1008 adapter->rx_ring[q_vector->rx.ring->queue_index] = NULL;
1009
1010 netif_napi_del(&q_vector->napi);
1011
1012}
1013
1014static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
1015{
1016 int v_idx = adapter->num_q_vectors;
1017
1018 if (adapter->flags & IGB_FLAG_HAS_MSIX)
1019 pci_disable_msix(adapter->pdev);
1020 else if (adapter->flags & IGB_FLAG_HAS_MSI)
1021 pci_disable_msi(adapter->pdev);
1022
1023 while (v_idx--)
1024 igb_reset_q_vector(adapter, v_idx);
1025}
1026
1027/**
1028 * igb_free_q_vectors - Free memory allocated for interrupt vectors
1029 * @adapter: board private structure to initialize
1030 *
1031 * This function frees the memory allocated to the q_vectors. In addition if
1032 * NAPI is enabled it will delete any references to the NAPI struct prior
1033 * to freeing the q_vector.
1034 **/
1035static void igb_free_q_vectors(struct igb_adapter *adapter)
1036{
1037 int v_idx = adapter->num_q_vectors;
1038
1039 adapter->num_tx_queues = 0;
1040 adapter->num_rx_queues = 0;
1041 adapter->num_q_vectors = 0;
1042
1043 while (v_idx--) {
1044 igb_reset_q_vector(adapter, v_idx);
1045 igb_free_q_vector(adapter, v_idx);
1046 }
1047}
1048
1049/**
1050 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1051 * @adapter: board private structure to initialize
1052 *
1053 * This function resets the device so that it has 0 Rx queues, Tx queues, and
1054 * MSI-X interrupts allocated.
1055 */
1056static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
1057{
1058 igb_free_q_vectors(adapter);
1059 igb_reset_interrupt_capability(adapter);
1060}
1061
1062/**
1063 * igb_set_interrupt_capability - set MSI or MSI-X if supported
1064 * @adapter: board private structure to initialize
1065 * @msix: boolean value of MSIX capability
1066 *
1067 * Attempt to configure interrupts using the best available
1068 * capabilities of the hardware and kernel.
1069 **/
1070static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix)
1071{
1072 int err;
1073 int numvecs, i;
1074
1075 if (!msix)
1076 goto msi_only;
1077 adapter->flags |= IGB_FLAG_HAS_MSIX;
1078
1079 /* Number of supported queues. */
1080 adapter->num_rx_queues = adapter->rss_queues;
1081 if (adapter->vfs_allocated_count)
1082 adapter->num_tx_queues = 1;
1083 else
1084 adapter->num_tx_queues = adapter->rss_queues;
1085
1086 /* start with one vector for every Rx queue */
1087 numvecs = adapter->num_rx_queues;
1088
1089 /* if Tx handler is separate add 1 for every Tx queue */
1090 if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1091 numvecs += adapter->num_tx_queues;
1092
1093 /* store the number of vectors reserved for queues */
1094 adapter->num_q_vectors = numvecs;
1095
1096 /* add 1 vector for link status interrupts */
1097 numvecs++;
1098 for (i = 0; i < numvecs; i++)
1099 adapter->msix_entries[i].entry = i;
1100
1101 err = pci_enable_msix_range(adapter->pdev,
1102 adapter->msix_entries,
1103 numvecs,
1104 numvecs);
1105 if (err > 0)
1106 return;
1107
1108 igb_reset_interrupt_capability(adapter);
1109
1110 /* If we can't do MSI-X, try MSI */
1111msi_only:
1112 adapter->flags &= ~IGB_FLAG_HAS_MSIX;
1113#ifdef CONFIG_PCI_IOV
1114 /* disable SR-IOV for non MSI-X configurations */
1115 if (adapter->vf_data) {
1116 struct e1000_hw *hw = &adapter->hw;
1117 /* disable iov and allow time for transactions to clear */
1118 pci_disable_sriov(adapter->pdev);
1119 msleep(500);
1120
1121 kfree(adapter->vf_mac_list);
1122 adapter->vf_mac_list = NULL;
1123 kfree(adapter->vf_data);
1124 adapter->vf_data = NULL;
1125 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1126 wrfl();
1127 msleep(100);
1128 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1129 }
1130#endif
1131 adapter->vfs_allocated_count = 0;
1132 adapter->rss_queues = 1;
1133 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1134 adapter->num_rx_queues = 1;
1135 adapter->num_tx_queues = 1;
1136 adapter->num_q_vectors = 1;
1137 if (!pci_enable_msi(adapter->pdev))
1138 adapter->flags |= IGB_FLAG_HAS_MSI;
1139}
1140
1141static void igb_add_ring(struct igb_ring *ring,
1142 struct igb_ring_container *head)
1143{
1144 head->ring = ring;
1145 head->count++;
1146}
1147
1148/**
1149 * igb_alloc_q_vector - Allocate memory for a single interrupt vector
1150 * @adapter: board private structure to initialize
1151 * @v_count: q_vectors allocated on adapter, used for ring interleaving
1152 * @v_idx: index of vector in adapter struct
1153 * @txr_count: total number of Tx rings to allocate
1154 * @txr_idx: index of first Tx ring to allocate
1155 * @rxr_count: total number of Rx rings to allocate
1156 * @rxr_idx: index of first Rx ring to allocate
1157 *
1158 * We allocate one q_vector. If allocation fails we return -ENOMEM.
1159 **/
1160static int igb_alloc_q_vector(struct igb_adapter *adapter,
1161 int v_count, int v_idx,
1162 int txr_count, int txr_idx,
1163 int rxr_count, int rxr_idx)
1164{
1165 struct igb_q_vector *q_vector;
1166 struct igb_ring *ring;
1167 int ring_count;
1168 size_t size;
1169
1170 /* igb only supports 1 Tx and/or 1 Rx queue per vector */
1171 if (txr_count > 1 || rxr_count > 1)
1172 return -ENOMEM;
1173
1174 ring_count = txr_count + rxr_count;
1175 size = kmalloc_size_roundup(struct_size(q_vector, ring, ring_count));
1176
1177 /* allocate q_vector and rings */
1178 q_vector = adapter->q_vector[v_idx];
1179 if (!q_vector) {
1180 q_vector = kzalloc(size, GFP_KERNEL);
1181 } else if (size > ksize(q_vector)) {
1182 struct igb_q_vector *new_q_vector;
1183
1184 new_q_vector = kzalloc(size, GFP_KERNEL);
1185 if (new_q_vector)
1186 kfree_rcu(q_vector, rcu);
1187 q_vector = new_q_vector;
1188 } else {
1189 memset(q_vector, 0, size);
1190 }
1191 if (!q_vector)
1192 return -ENOMEM;
1193
1194 /* initialize NAPI */
1195 netif_napi_add(adapter->netdev, &q_vector->napi, igb_poll);
1196
1197 /* tie q_vector and adapter together */
1198 adapter->q_vector[v_idx] = q_vector;
1199 q_vector->adapter = adapter;
1200
1201 /* initialize work limits */
1202 q_vector->tx.work_limit = adapter->tx_work_limit;
1203
1204 /* initialize ITR configuration */
1205 q_vector->itr_register = adapter->io_addr + E1000_EITR(0);
1206 q_vector->itr_val = IGB_START_ITR;
1207
1208 /* initialize pointer to rings */
1209 ring = q_vector->ring;
1210
1211 /* initialize ITR */
1212 if (rxr_count) {
1213 /* rx or rx/tx vector */
1214 if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
1215 q_vector->itr_val = adapter->rx_itr_setting;
1216 } else {
1217 /* tx only vector */
1218 if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
1219 q_vector->itr_val = adapter->tx_itr_setting;
1220 }
1221
1222 if (txr_count) {
1223 /* assign generic ring traits */
1224 ring->dev = &adapter->pdev->dev;
1225 ring->netdev = adapter->netdev;
1226
1227 /* configure backlink on ring */
1228 ring->q_vector = q_vector;
1229
1230 /* update q_vector Tx values */
1231 igb_add_ring(ring, &q_vector->tx);
1232
1233 /* For 82575, context index must be unique per ring. */
1234 if (adapter->hw.mac.type == e1000_82575)
1235 set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
1236
1237 /* apply Tx specific ring traits */
1238 ring->count = adapter->tx_ring_count;
1239 ring->queue_index = txr_idx;
1240
1241 ring->cbs_enable = false;
1242 ring->idleslope = 0;
1243 ring->sendslope = 0;
1244 ring->hicredit = 0;
1245 ring->locredit = 0;
1246
1247 u64_stats_init(&ring->tx_syncp);
1248 u64_stats_init(&ring->tx_syncp2);
1249
1250 /* assign ring to adapter */
1251 adapter->tx_ring[txr_idx] = ring;
1252
1253 /* push pointer to next ring */
1254 ring++;
1255 }
1256
1257 if (rxr_count) {
1258 /* assign generic ring traits */
1259 ring->dev = &adapter->pdev->dev;
1260 ring->netdev = adapter->netdev;
1261
1262 /* configure backlink on ring */
1263 ring->q_vector = q_vector;
1264
1265 /* update q_vector Rx values */
1266 igb_add_ring(ring, &q_vector->rx);
1267
1268 /* set flag indicating ring supports SCTP checksum offload */
1269 if (adapter->hw.mac.type >= e1000_82576)
1270 set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
1271
1272 /* On i350, i354, i210, and i211, loopback VLAN packets
1273 * have the tag byte-swapped.
1274 */
1275 if (adapter->hw.mac.type >= e1000_i350)
1276 set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
1277
1278 /* apply Rx specific ring traits */
1279 ring->count = adapter->rx_ring_count;
1280 ring->queue_index = rxr_idx;
1281
1282 u64_stats_init(&ring->rx_syncp);
1283
1284 /* assign ring to adapter */
1285 adapter->rx_ring[rxr_idx] = ring;
1286 }
1287
1288 return 0;
1289}
1290
1291
1292/**
1293 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
1294 * @adapter: board private structure to initialize
1295 *
1296 * We allocate one q_vector per queue interrupt. If allocation fails we
1297 * return -ENOMEM.
1298 **/
1299static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1300{
1301 int q_vectors = adapter->num_q_vectors;
1302 int rxr_remaining = adapter->num_rx_queues;
1303 int txr_remaining = adapter->num_tx_queues;
1304 int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1305 int err;
1306
1307 if (q_vectors >= (rxr_remaining + txr_remaining)) {
1308 for (; rxr_remaining; v_idx++) {
1309 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1310 0, 0, 1, rxr_idx);
1311
1312 if (err)
1313 goto err_out;
1314
1315 /* update counts and index */
1316 rxr_remaining--;
1317 rxr_idx++;
1318 }
1319 }
1320
1321 for (; v_idx < q_vectors; v_idx++) {
1322 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1323 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1324
1325 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1326 tqpv, txr_idx, rqpv, rxr_idx);
1327
1328 if (err)
1329 goto err_out;
1330
1331 /* update counts and index */
1332 rxr_remaining -= rqpv;
1333 txr_remaining -= tqpv;
1334 rxr_idx++;
1335 txr_idx++;
1336 }
1337
1338 return 0;
1339
1340err_out:
1341 adapter->num_tx_queues = 0;
1342 adapter->num_rx_queues = 0;
1343 adapter->num_q_vectors = 0;
1344
1345 while (v_idx--)
1346 igb_free_q_vector(adapter, v_idx);
1347
1348 return -ENOMEM;
1349}
1350
1351/**
1352 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1353 * @adapter: board private structure to initialize
1354 * @msix: boolean value of MSIX capability
1355 *
1356 * This function initializes the interrupts and allocates all of the queues.
1357 **/
1358static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
1359{
1360 struct pci_dev *pdev = adapter->pdev;
1361 int err;
1362
1363 igb_set_interrupt_capability(adapter, msix);
1364
1365 err = igb_alloc_q_vectors(adapter);
1366 if (err) {
1367 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1368 goto err_alloc_q_vectors;
1369 }
1370
1371 igb_cache_ring_register(adapter);
1372
1373 return 0;
1374
1375err_alloc_q_vectors:
1376 igb_reset_interrupt_capability(adapter);
1377 return err;
1378}
1379
1380/**
1381 * igb_request_irq - initialize interrupts
1382 * @adapter: board private structure to initialize
1383 *
1384 * Attempts to configure interrupts using the best available
1385 * capabilities of the hardware and kernel.
1386 **/
1387static int igb_request_irq(struct igb_adapter *adapter)
1388{
1389 struct net_device *netdev = adapter->netdev;
1390 struct pci_dev *pdev = adapter->pdev;
1391 int err = 0;
1392
1393 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1394 err = igb_request_msix(adapter);
1395 if (!err)
1396 goto request_done;
1397 /* fall back to MSI */
1398 igb_free_all_tx_resources(adapter);
1399 igb_free_all_rx_resources(adapter);
1400
1401 igb_clear_interrupt_scheme(adapter);
1402 err = igb_init_interrupt_scheme(adapter, false);
1403 if (err)
1404 goto request_done;
1405
1406 igb_setup_all_tx_resources(adapter);
1407 igb_setup_all_rx_resources(adapter);
1408 igb_configure(adapter);
1409 }
1410
1411 igb_assign_vector(adapter->q_vector[0], 0);
1412
1413 if (adapter->flags & IGB_FLAG_HAS_MSI) {
1414 err = request_irq(pdev->irq, igb_intr_msi, 0,
1415 netdev->name, adapter);
1416 if (!err)
1417 goto request_done;
1418
1419 /* fall back to legacy interrupts */
1420 igb_reset_interrupt_capability(adapter);
1421 adapter->flags &= ~IGB_FLAG_HAS_MSI;
1422 }
1423
1424 err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1425 netdev->name, adapter);
1426
1427 if (err)
1428 dev_err(&pdev->dev, "Error %d getting interrupt\n",
1429 err);
1430
1431request_done:
1432 return err;
1433}
1434
1435static void igb_free_irq(struct igb_adapter *adapter)
1436{
1437 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1438 int vector = 0, i;
1439
1440 free_irq(adapter->msix_entries[vector++].vector, adapter);
1441
1442 for (i = 0; i < adapter->num_q_vectors; i++)
1443 free_irq(adapter->msix_entries[vector++].vector,
1444 adapter->q_vector[i]);
1445 } else {
1446 free_irq(adapter->pdev->irq, adapter);
1447 }
1448}
1449
1450/**
1451 * igb_irq_disable - Mask off interrupt generation on the NIC
1452 * @adapter: board private structure
1453 **/
1454static void igb_irq_disable(struct igb_adapter *adapter)
1455{
1456 struct e1000_hw *hw = &adapter->hw;
1457
1458 /* we need to be careful when disabling interrupts. The VFs are also
1459 * mapped into these registers and so clearing the bits can cause
1460 * issues on the VF drivers so we only need to clear what we set
1461 */
1462 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1463 u32 regval = rd32(E1000_EIAM);
1464
1465 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1466 wr32(E1000_EIMC, adapter->eims_enable_mask);
1467 regval = rd32(E1000_EIAC);
1468 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1469 }
1470
1471 wr32(E1000_IAM, 0);
1472 wr32(E1000_IMC, ~0);
1473 wrfl();
1474 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1475 int i;
1476
1477 for (i = 0; i < adapter->num_q_vectors; i++)
1478 synchronize_irq(adapter->msix_entries[i].vector);
1479 } else {
1480 synchronize_irq(adapter->pdev->irq);
1481 }
1482}
1483
1484/**
1485 * igb_irq_enable - Enable default interrupt generation settings
1486 * @adapter: board private structure
1487 **/
1488static void igb_irq_enable(struct igb_adapter *adapter)
1489{
1490 struct e1000_hw *hw = &adapter->hw;
1491
1492 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1493 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1494 u32 regval = rd32(E1000_EIAC);
1495
1496 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1497 regval = rd32(E1000_EIAM);
1498 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1499 wr32(E1000_EIMS, adapter->eims_enable_mask);
1500 if (adapter->vfs_allocated_count) {
1501 wr32(E1000_MBVFIMR, 0xFF);
1502 ims |= E1000_IMS_VMMB;
1503 }
1504 wr32(E1000_IMS, ims);
1505 } else {
1506 wr32(E1000_IMS, IMS_ENABLE_MASK |
1507 E1000_IMS_DRSTA);
1508 wr32(E1000_IAM, IMS_ENABLE_MASK |
1509 E1000_IMS_DRSTA);
1510 }
1511}
1512
1513static void igb_update_mng_vlan(struct igb_adapter *adapter)
1514{
1515 struct e1000_hw *hw = &adapter->hw;
1516 u16 pf_id = adapter->vfs_allocated_count;
1517 u16 vid = adapter->hw.mng_cookie.vlan_id;
1518 u16 old_vid = adapter->mng_vlan_id;
1519
1520 if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1521 /* add VID to filter table */
1522 igb_vfta_set(hw, vid, pf_id, true, true);
1523 adapter->mng_vlan_id = vid;
1524 } else {
1525 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1526 }
1527
1528 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1529 (vid != old_vid) &&
1530 !test_bit(old_vid, adapter->active_vlans)) {
1531 /* remove VID from filter table */
1532 igb_vfta_set(hw, vid, pf_id, false, true);
1533 }
1534}
1535
1536/**
1537 * igb_release_hw_control - release control of the h/w to f/w
1538 * @adapter: address of board private structure
1539 *
1540 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1541 * For ASF and Pass Through versions of f/w this means that the
1542 * driver is no longer loaded.
1543 **/
1544static void igb_release_hw_control(struct igb_adapter *adapter)
1545{
1546 struct e1000_hw *hw = &adapter->hw;
1547 u32 ctrl_ext;
1548
1549 /* Let firmware take over control of h/w */
1550 ctrl_ext = rd32(E1000_CTRL_EXT);
1551 wr32(E1000_CTRL_EXT,
1552 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1553}
1554
1555/**
1556 * igb_get_hw_control - get control of the h/w from f/w
1557 * @adapter: address of board private structure
1558 *
1559 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1560 * For ASF and Pass Through versions of f/w this means that
1561 * the driver is loaded.
1562 **/
1563static void igb_get_hw_control(struct igb_adapter *adapter)
1564{
1565 struct e1000_hw *hw = &adapter->hw;
1566 u32 ctrl_ext;
1567
1568 /* Let firmware know the driver has taken over */
1569 ctrl_ext = rd32(E1000_CTRL_EXT);
1570 wr32(E1000_CTRL_EXT,
1571 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1572}
1573
1574static void enable_fqtss(struct igb_adapter *adapter, bool enable)
1575{
1576 struct net_device *netdev = adapter->netdev;
1577 struct e1000_hw *hw = &adapter->hw;
1578
1579 WARN_ON(hw->mac.type != e1000_i210);
1580
1581 if (enable)
1582 adapter->flags |= IGB_FLAG_FQTSS;
1583 else
1584 adapter->flags &= ~IGB_FLAG_FQTSS;
1585
1586 if (netif_running(netdev))
1587 schedule_work(&adapter->reset_task);
1588}
1589
1590static bool is_fqtss_enabled(struct igb_adapter *adapter)
1591{
1592 return (adapter->flags & IGB_FLAG_FQTSS) ? true : false;
1593}
1594
1595static void set_tx_desc_fetch_prio(struct e1000_hw *hw, int queue,
1596 enum tx_queue_prio prio)
1597{
1598 u32 val;
1599
1600 WARN_ON(hw->mac.type != e1000_i210);
1601 WARN_ON(queue < 0 || queue > 4);
1602
1603 val = rd32(E1000_I210_TXDCTL(queue));
1604
1605 if (prio == TX_QUEUE_PRIO_HIGH)
1606 val |= E1000_TXDCTL_PRIORITY;
1607 else
1608 val &= ~E1000_TXDCTL_PRIORITY;
1609
1610 wr32(E1000_I210_TXDCTL(queue), val);
1611}
1612
1613static void set_queue_mode(struct e1000_hw *hw, int queue, enum queue_mode mode)
1614{
1615 u32 val;
1616
1617 WARN_ON(hw->mac.type != e1000_i210);
1618 WARN_ON(queue < 0 || queue > 1);
1619
1620 val = rd32(E1000_I210_TQAVCC(queue));
1621
1622 if (mode == QUEUE_MODE_STREAM_RESERVATION)
1623 val |= E1000_TQAVCC_QUEUEMODE;
1624 else
1625 val &= ~E1000_TQAVCC_QUEUEMODE;
1626
1627 wr32(E1000_I210_TQAVCC(queue), val);
1628}
1629
1630static bool is_any_cbs_enabled(struct igb_adapter *adapter)
1631{
1632 int i;
1633
1634 for (i = 0; i < adapter->num_tx_queues; i++) {
1635 if (adapter->tx_ring[i]->cbs_enable)
1636 return true;
1637 }
1638
1639 return false;
1640}
1641
1642static bool is_any_txtime_enabled(struct igb_adapter *adapter)
1643{
1644 int i;
1645
1646 for (i = 0; i < adapter->num_tx_queues; i++) {
1647 if (adapter->tx_ring[i]->launchtime_enable)
1648 return true;
1649 }
1650
1651 return false;
1652}
1653
1654/**
1655 * igb_config_tx_modes - Configure "Qav Tx mode" features on igb
1656 * @adapter: pointer to adapter struct
1657 * @queue: queue number
1658 *
1659 * Configure CBS and Launchtime for a given hardware queue.
1660 * Parameters are retrieved from the correct Tx ring, so
1661 * igb_save_cbs_params() and igb_save_txtime_params() should be used
1662 * for setting those correctly prior to this function being called.
1663 **/
1664static void igb_config_tx_modes(struct igb_adapter *adapter, int queue)
1665{
1666 struct net_device *netdev = adapter->netdev;
1667 struct e1000_hw *hw = &adapter->hw;
1668 struct igb_ring *ring;
1669 u32 tqavcc, tqavctrl;
1670 u16 value;
1671
1672 WARN_ON(hw->mac.type != e1000_i210);
1673 WARN_ON(queue < 0 || queue > 1);
1674 ring = adapter->tx_ring[queue];
1675
1676 /* If any of the Qav features is enabled, configure queues as SR and
1677 * with HIGH PRIO. If none is, then configure them with LOW PRIO and
1678 * as SP.
1679 */
1680 if (ring->cbs_enable || ring->launchtime_enable) {
1681 set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_HIGH);
1682 set_queue_mode(hw, queue, QUEUE_MODE_STREAM_RESERVATION);
1683 } else {
1684 set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_LOW);
1685 set_queue_mode(hw, queue, QUEUE_MODE_STRICT_PRIORITY);
1686 }
1687
1688 /* If CBS is enabled, set DataTranARB and config its parameters. */
1689 if (ring->cbs_enable || queue == 0) {
1690 /* i210 does not allow the queue 0 to be in the Strict
1691 * Priority mode while the Qav mode is enabled, so,
1692 * instead of disabling strict priority mode, we give
1693 * queue 0 the maximum of credits possible.
1694 *
1695 * See section 8.12.19 of the i210 datasheet, "Note:
1696 * Queue0 QueueMode must be set to 1b when
1697 * TransmitMode is set to Qav."
1698 */
1699 if (queue == 0 && !ring->cbs_enable) {
1700 /* max "linkspeed" idleslope in kbps */
1701 ring->idleslope = 1000000;
1702 ring->hicredit = ETH_FRAME_LEN;
1703 }
1704
1705 /* Always set data transfer arbitration to credit-based
1706 * shaper algorithm on TQAVCTRL if CBS is enabled for any of
1707 * the queues.
1708 */
1709 tqavctrl = rd32(E1000_I210_TQAVCTRL);
1710 tqavctrl |= E1000_TQAVCTRL_DATATRANARB;
1711 wr32(E1000_I210_TQAVCTRL, tqavctrl);
1712
1713 /* According to i210 datasheet section 7.2.7.7, we should set
1714 * the 'idleSlope' field from TQAVCC register following the
1715 * equation:
1716 *
1717 * For 100 Mbps link speed:
1718 *
1719 * value = BW * 0x7735 * 0.2 (E1)
1720 *
1721 * For 1000Mbps link speed:
1722 *
1723 * value = BW * 0x7735 * 2 (E2)
1724 *
1725 * E1 and E2 can be merged into one equation as shown below.
1726 * Note that 'link-speed' is in Mbps.
1727 *
1728 * value = BW * 0x7735 * 2 * link-speed
1729 * -------------- (E3)
1730 * 1000
1731 *
1732 * 'BW' is the percentage bandwidth out of full link speed
1733 * which can be found with the following equation. Note that
1734 * idleSlope here is the parameter from this function which
1735 * is in kbps.
1736 *
1737 * BW = idleSlope
1738 * ----------------- (E4)
1739 * link-speed * 1000
1740 *
1741 * That said, we can come up with a generic equation to
1742 * calculate the value we should set it TQAVCC register by
1743 * replacing 'BW' in E3 by E4. The resulting equation is:
1744 *
1745 * value = idleSlope * 0x7735 * 2 * link-speed
1746 * ----------------- -------------- (E5)
1747 * link-speed * 1000 1000
1748 *
1749 * 'link-speed' is present in both sides of the fraction so
1750 * it is canceled out. The final equation is the following:
1751 *
1752 * value = idleSlope * 61034
1753 * ----------------- (E6)
1754 * 1000000
1755 *
1756 * NOTE: For i210, given the above, we can see that idleslope
1757 * is represented in 16.38431 kbps units by the value at
1758 * the TQAVCC register (1Gbps / 61034), which reduces
1759 * the granularity for idleslope increments.
1760 * For instance, if you want to configure a 2576kbps
1761 * idleslope, the value to be written on the register
1762 * would have to be 157.23. If rounded down, you end
1763 * up with less bandwidth available than originally
1764 * required (~2572 kbps). If rounded up, you end up
1765 * with a higher bandwidth (~2589 kbps). Below the
1766 * approach we take is to always round up the
1767 * calculated value, so the resulting bandwidth might
1768 * be slightly higher for some configurations.
1769 */
1770 value = DIV_ROUND_UP_ULL(ring->idleslope * 61034ULL, 1000000);
1771
1772 tqavcc = rd32(E1000_I210_TQAVCC(queue));
1773 tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
1774 tqavcc |= value;
1775 wr32(E1000_I210_TQAVCC(queue), tqavcc);
1776
1777 wr32(E1000_I210_TQAVHC(queue),
1778 0x80000000 + ring->hicredit * 0x7735);
1779 } else {
1780
1781 /* Set idleSlope to zero. */
1782 tqavcc = rd32(E1000_I210_TQAVCC(queue));
1783 tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
1784 wr32(E1000_I210_TQAVCC(queue), tqavcc);
1785
1786 /* Set hiCredit to zero. */
1787 wr32(E1000_I210_TQAVHC(queue), 0);
1788
1789 /* If CBS is not enabled for any queues anymore, then return to
1790 * the default state of Data Transmission Arbitration on
1791 * TQAVCTRL.
1792 */
1793 if (!is_any_cbs_enabled(adapter)) {
1794 tqavctrl = rd32(E1000_I210_TQAVCTRL);
1795 tqavctrl &= ~E1000_TQAVCTRL_DATATRANARB;
1796 wr32(E1000_I210_TQAVCTRL, tqavctrl);
1797 }
1798 }
1799
1800 /* If LaunchTime is enabled, set DataTranTIM. */
1801 if (ring->launchtime_enable) {
1802 /* Always set DataTranTIM on TQAVCTRL if LaunchTime is enabled
1803 * for any of the SR queues, and configure fetchtime delta.
1804 * XXX NOTE:
1805 * - LaunchTime will be enabled for all SR queues.
1806 * - A fixed offset can be added relative to the launch
1807 * time of all packets if configured at reg LAUNCH_OS0.
1808 * We are keeping it as 0 for now (default value).
1809 */
1810 tqavctrl = rd32(E1000_I210_TQAVCTRL);
1811 tqavctrl |= E1000_TQAVCTRL_DATATRANTIM |
1812 E1000_TQAVCTRL_FETCHTIME_DELTA;
1813 wr32(E1000_I210_TQAVCTRL, tqavctrl);
1814 } else {
1815 /* If Launchtime is not enabled for any SR queues anymore,
1816 * then clear DataTranTIM on TQAVCTRL and clear fetchtime delta,
1817 * effectively disabling Launchtime.
1818 */
1819 if (!is_any_txtime_enabled(adapter)) {
1820 tqavctrl = rd32(E1000_I210_TQAVCTRL);
1821 tqavctrl &= ~E1000_TQAVCTRL_DATATRANTIM;
1822 tqavctrl &= ~E1000_TQAVCTRL_FETCHTIME_DELTA;
1823 wr32(E1000_I210_TQAVCTRL, tqavctrl);
1824 }
1825 }
1826
1827 /* XXX: In i210 controller the sendSlope and loCredit parameters from
1828 * CBS are not configurable by software so we don't do any 'controller
1829 * configuration' in respect to these parameters.
1830 */
1831
1832 netdev_dbg(netdev, "Qav Tx mode: cbs %s, launchtime %s, queue %d idleslope %d sendslope %d hiCredit %d locredit %d\n",
1833 ring->cbs_enable ? "enabled" : "disabled",
1834 ring->launchtime_enable ? "enabled" : "disabled",
1835 queue,
1836 ring->idleslope, ring->sendslope,
1837 ring->hicredit, ring->locredit);
1838}
1839
1840static int igb_save_txtime_params(struct igb_adapter *adapter, int queue,
1841 bool enable)
1842{
1843 struct igb_ring *ring;
1844
1845 if (queue < 0 || queue > adapter->num_tx_queues)
1846 return -EINVAL;
1847
1848 ring = adapter->tx_ring[queue];
1849 ring->launchtime_enable = enable;
1850
1851 return 0;
1852}
1853
1854static int igb_save_cbs_params(struct igb_adapter *adapter, int queue,
1855 bool enable, int idleslope, int sendslope,
1856 int hicredit, int locredit)
1857{
1858 struct igb_ring *ring;
1859
1860 if (queue < 0 || queue > adapter->num_tx_queues)
1861 return -EINVAL;
1862
1863 ring = adapter->tx_ring[queue];
1864
1865 ring->cbs_enable = enable;
1866 ring->idleslope = idleslope;
1867 ring->sendslope = sendslope;
1868 ring->hicredit = hicredit;
1869 ring->locredit = locredit;
1870
1871 return 0;
1872}
1873
1874/**
1875 * igb_setup_tx_mode - Switch to/from Qav Tx mode when applicable
1876 * @adapter: pointer to adapter struct
1877 *
1878 * Configure TQAVCTRL register switching the controller's Tx mode
1879 * if FQTSS mode is enabled or disabled. Additionally, will issue
1880 * a call to igb_config_tx_modes() per queue so any previously saved
1881 * Tx parameters are applied.
1882 **/
1883static void igb_setup_tx_mode(struct igb_adapter *adapter)
1884{
1885 struct net_device *netdev = adapter->netdev;
1886 struct e1000_hw *hw = &adapter->hw;
1887 u32 val;
1888
1889 /* Only i210 controller supports changing the transmission mode. */
1890 if (hw->mac.type != e1000_i210)
1891 return;
1892
1893 if (is_fqtss_enabled(adapter)) {
1894 int i, max_queue;
1895
1896 /* Configure TQAVCTRL register: set transmit mode to 'Qav',
1897 * set data fetch arbitration to 'round robin', set SP_WAIT_SR
1898 * so SP queues wait for SR ones.
1899 */
1900 val = rd32(E1000_I210_TQAVCTRL);
1901 val |= E1000_TQAVCTRL_XMIT_MODE | E1000_TQAVCTRL_SP_WAIT_SR;
1902 val &= ~E1000_TQAVCTRL_DATAFETCHARB;
1903 wr32(E1000_I210_TQAVCTRL, val);
1904
1905 /* Configure Tx and Rx packet buffers sizes as described in
1906 * i210 datasheet section 7.2.7.7.
1907 */
1908 val = rd32(E1000_TXPBS);
1909 val &= ~I210_TXPBSIZE_MASK;
1910 val |= I210_TXPBSIZE_PB0_6KB | I210_TXPBSIZE_PB1_6KB |
1911 I210_TXPBSIZE_PB2_6KB | I210_TXPBSIZE_PB3_6KB;
1912 wr32(E1000_TXPBS, val);
1913
1914 val = rd32(E1000_RXPBS);
1915 val &= ~I210_RXPBSIZE_MASK;
1916 val |= I210_RXPBSIZE_PB_30KB;
1917 wr32(E1000_RXPBS, val);
1918
1919 /* Section 8.12.9 states that MAX_TPKT_SIZE from DTXMXPKTSZ
1920 * register should not exceed the buffer size programmed in
1921 * TXPBS. The smallest buffer size programmed in TXPBS is 4kB
1922 * so according to the datasheet we should set MAX_TPKT_SIZE to
1923 * 4kB / 64.
1924 *
1925 * However, when we do so, no frame from queue 2 and 3 are
1926 * transmitted. It seems the MAX_TPKT_SIZE should not be great
1927 * or _equal_ to the buffer size programmed in TXPBS. For this
1928 * reason, we set MAX_ TPKT_SIZE to (4kB - 1) / 64.
1929 */
1930 val = (4096 - 1) / 64;
1931 wr32(E1000_I210_DTXMXPKTSZ, val);
1932
1933 /* Since FQTSS mode is enabled, apply any CBS configuration
1934 * previously set. If no previous CBS configuration has been
1935 * done, then the initial configuration is applied, which means
1936 * CBS is disabled.
1937 */
1938 max_queue = (adapter->num_tx_queues < I210_SR_QUEUES_NUM) ?
1939 adapter->num_tx_queues : I210_SR_QUEUES_NUM;
1940
1941 for (i = 0; i < max_queue; i++) {
1942 igb_config_tx_modes(adapter, i);
1943 }
1944 } else {
1945 wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
1946 wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
1947 wr32(E1000_I210_DTXMXPKTSZ, I210_DTXMXPKTSZ_DEFAULT);
1948
1949 val = rd32(E1000_I210_TQAVCTRL);
1950 /* According to Section 8.12.21, the other flags we've set when
1951 * enabling FQTSS are not relevant when disabling FQTSS so we
1952 * don't set they here.
1953 */
1954 val &= ~E1000_TQAVCTRL_XMIT_MODE;
1955 wr32(E1000_I210_TQAVCTRL, val);
1956 }
1957
1958 netdev_dbg(netdev, "FQTSS %s\n", (is_fqtss_enabled(adapter)) ?
1959 "enabled" : "disabled");
1960}
1961
1962/**
1963 * igb_configure - configure the hardware for RX and TX
1964 * @adapter: private board structure
1965 **/
1966static void igb_configure(struct igb_adapter *adapter)
1967{
1968 struct net_device *netdev = adapter->netdev;
1969 int i;
1970
1971 igb_get_hw_control(adapter);
1972 igb_set_rx_mode(netdev);
1973 igb_setup_tx_mode(adapter);
1974
1975 igb_restore_vlan(adapter);
1976
1977 igb_setup_tctl(adapter);
1978 igb_setup_mrqc(adapter);
1979 igb_setup_rctl(adapter);
1980
1981 igb_nfc_filter_restore(adapter);
1982 igb_configure_tx(adapter);
1983 igb_configure_rx(adapter);
1984
1985 igb_rx_fifo_flush_82575(&adapter->hw);
1986
1987 /* call igb_desc_unused which always leaves
1988 * at least 1 descriptor unused to make sure
1989 * next_to_use != next_to_clean
1990 */
1991 for (i = 0; i < adapter->num_rx_queues; i++) {
1992 struct igb_ring *ring = adapter->rx_ring[i];
1993 igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
1994 }
1995}
1996
1997/**
1998 * igb_power_up_link - Power up the phy/serdes link
1999 * @adapter: address of board private structure
2000 **/
2001void igb_power_up_link(struct igb_adapter *adapter)
2002{
2003 igb_reset_phy(&adapter->hw);
2004
2005 if (adapter->hw.phy.media_type == e1000_media_type_copper)
2006 igb_power_up_phy_copper(&adapter->hw);
2007 else
2008 igb_power_up_serdes_link_82575(&adapter->hw);
2009
2010 igb_setup_link(&adapter->hw);
2011}
2012
2013/**
2014 * igb_power_down_link - Power down the phy/serdes link
2015 * @adapter: address of board private structure
2016 */
2017static void igb_power_down_link(struct igb_adapter *adapter)
2018{
2019 if (adapter->hw.phy.media_type == e1000_media_type_copper)
2020 igb_power_down_phy_copper_82575(&adapter->hw);
2021 else
2022 igb_shutdown_serdes_link_82575(&adapter->hw);
2023}
2024
2025/**
2026 * igb_check_swap_media - Detect and switch function for Media Auto Sense
2027 * @adapter: address of the board private structure
2028 **/
2029static void igb_check_swap_media(struct igb_adapter *adapter)
2030{
2031 struct e1000_hw *hw = &adapter->hw;
2032 u32 ctrl_ext, connsw;
2033 bool swap_now = false;
2034
2035 ctrl_ext = rd32(E1000_CTRL_EXT);
2036 connsw = rd32(E1000_CONNSW);
2037
2038 /* need to live swap if current media is copper and we have fiber/serdes
2039 * to go to.
2040 */
2041
2042 if ((hw->phy.media_type == e1000_media_type_copper) &&
2043 (!(connsw & E1000_CONNSW_AUTOSENSE_EN))) {
2044 swap_now = true;
2045 } else if ((hw->phy.media_type != e1000_media_type_copper) &&
2046 !(connsw & E1000_CONNSW_SERDESD)) {
2047 /* copper signal takes time to appear */
2048 if (adapter->copper_tries < 4) {
2049 adapter->copper_tries++;
2050 connsw |= E1000_CONNSW_AUTOSENSE_CONF;
2051 wr32(E1000_CONNSW, connsw);
2052 return;
2053 } else {
2054 adapter->copper_tries = 0;
2055 if ((connsw & E1000_CONNSW_PHYSD) &&
2056 (!(connsw & E1000_CONNSW_PHY_PDN))) {
2057 swap_now = true;
2058 connsw &= ~E1000_CONNSW_AUTOSENSE_CONF;
2059 wr32(E1000_CONNSW, connsw);
2060 }
2061 }
2062 }
2063
2064 if (!swap_now)
2065 return;
2066
2067 switch (hw->phy.media_type) {
2068 case e1000_media_type_copper:
2069 netdev_info(adapter->netdev,
2070 "MAS: changing media to fiber/serdes\n");
2071 ctrl_ext |=
2072 E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
2073 adapter->flags |= IGB_FLAG_MEDIA_RESET;
2074 adapter->copper_tries = 0;
2075 break;
2076 case e1000_media_type_internal_serdes:
2077 case e1000_media_type_fiber:
2078 netdev_info(adapter->netdev,
2079 "MAS: changing media to copper\n");
2080 ctrl_ext &=
2081 ~E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
2082 adapter->flags |= IGB_FLAG_MEDIA_RESET;
2083 break;
2084 default:
2085 /* shouldn't get here during regular operation */
2086 netdev_err(adapter->netdev,
2087 "AMS: Invalid media type found, returning\n");
2088 break;
2089 }
2090 wr32(E1000_CTRL_EXT, ctrl_ext);
2091}
2092
2093/**
2094 * igb_up - Open the interface and prepare it to handle traffic
2095 * @adapter: board private structure
2096 **/
2097int igb_up(struct igb_adapter *adapter)
2098{
2099 struct e1000_hw *hw = &adapter->hw;
2100 int i;
2101
2102 /* hardware has been reset, we need to reload some things */
2103 igb_configure(adapter);
2104
2105 clear_bit(__IGB_DOWN, &adapter->state);
2106
2107 for (i = 0; i < adapter->num_q_vectors; i++)
2108 napi_enable(&(adapter->q_vector[i]->napi));
2109
2110 if (adapter->flags & IGB_FLAG_HAS_MSIX)
2111 igb_configure_msix(adapter);
2112 else
2113 igb_assign_vector(adapter->q_vector[0], 0);
2114
2115 /* Clear any pending interrupts. */
2116 rd32(E1000_TSICR);
2117 rd32(E1000_ICR);
2118 igb_irq_enable(adapter);
2119
2120 /* notify VFs that reset has been completed */
2121 if (adapter->vfs_allocated_count) {
2122 u32 reg_data = rd32(E1000_CTRL_EXT);
2123
2124 reg_data |= E1000_CTRL_EXT_PFRSTD;
2125 wr32(E1000_CTRL_EXT, reg_data);
2126 }
2127
2128 netif_tx_start_all_queues(adapter->netdev);
2129
2130 /* start the watchdog. */
2131 hw->mac.get_link_status = 1;
2132 schedule_work(&adapter->watchdog_task);
2133
2134 if ((adapter->flags & IGB_FLAG_EEE) &&
2135 (!hw->dev_spec._82575.eee_disable))
2136 adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
2137
2138 return 0;
2139}
2140
2141void igb_down(struct igb_adapter *adapter)
2142{
2143 struct net_device *netdev = adapter->netdev;
2144 struct e1000_hw *hw = &adapter->hw;
2145 u32 tctl, rctl;
2146 int i;
2147
2148 /* signal that we're down so the interrupt handler does not
2149 * reschedule our watchdog timer
2150 */
2151 set_bit(__IGB_DOWN, &adapter->state);
2152
2153 /* disable receives in the hardware */
2154 rctl = rd32(E1000_RCTL);
2155 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
2156 /* flush and sleep below */
2157
2158 igb_nfc_filter_exit(adapter);
2159
2160 netif_carrier_off(netdev);
2161 netif_tx_stop_all_queues(netdev);
2162
2163 /* disable transmits in the hardware */
2164 tctl = rd32(E1000_TCTL);
2165 tctl &= ~E1000_TCTL_EN;
2166 wr32(E1000_TCTL, tctl);
2167 /* flush both disables and wait for them to finish */
2168 wrfl();
2169 usleep_range(10000, 11000);
2170
2171 igb_irq_disable(adapter);
2172
2173 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
2174
2175 for (i = 0; i < adapter->num_q_vectors; i++) {
2176 if (adapter->q_vector[i]) {
2177 napi_synchronize(&adapter->q_vector[i]->napi);
2178 napi_disable(&adapter->q_vector[i]->napi);
2179 }
2180 }
2181
2182 del_timer_sync(&adapter->watchdog_timer);
2183 del_timer_sync(&adapter->phy_info_timer);
2184
2185 /* record the stats before reset*/
2186 spin_lock(&adapter->stats64_lock);
2187 igb_update_stats(adapter);
2188 spin_unlock(&adapter->stats64_lock);
2189
2190 adapter->link_speed = 0;
2191 adapter->link_duplex = 0;
2192
2193 if (!pci_channel_offline(adapter->pdev))
2194 igb_reset(adapter);
2195
2196 /* clear VLAN promisc flag so VFTA will be updated if necessary */
2197 adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
2198
2199 igb_clean_all_tx_rings(adapter);
2200 igb_clean_all_rx_rings(adapter);
2201#ifdef CONFIG_IGB_DCA
2202
2203 /* since we reset the hardware DCA settings were cleared */
2204 igb_setup_dca(adapter);
2205#endif
2206}
2207
2208void igb_reinit_locked(struct igb_adapter *adapter)
2209{
2210 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
2211 usleep_range(1000, 2000);
2212 igb_down(adapter);
2213 igb_up(adapter);
2214 clear_bit(__IGB_RESETTING, &adapter->state);
2215}
2216
2217/** igb_enable_mas - Media Autosense re-enable after swap
2218 *
2219 * @adapter: adapter struct
2220 **/
2221static void igb_enable_mas(struct igb_adapter *adapter)
2222{
2223 struct e1000_hw *hw = &adapter->hw;
2224 u32 connsw = rd32(E1000_CONNSW);
2225
2226 /* configure for SerDes media detect */
2227 if ((hw->phy.media_type == e1000_media_type_copper) &&
2228 (!(connsw & E1000_CONNSW_SERDESD))) {
2229 connsw |= E1000_CONNSW_ENRGSRC;
2230 connsw |= E1000_CONNSW_AUTOSENSE_EN;
2231 wr32(E1000_CONNSW, connsw);
2232 wrfl();
2233 }
2234}
2235
2236#ifdef CONFIG_IGB_HWMON
2237/**
2238 * igb_set_i2c_bb - Init I2C interface
2239 * @hw: pointer to hardware structure
2240 **/
2241static void igb_set_i2c_bb(struct e1000_hw *hw)
2242{
2243 u32 ctrl_ext;
2244 s32 i2cctl;
2245
2246 ctrl_ext = rd32(E1000_CTRL_EXT);
2247 ctrl_ext |= E1000_CTRL_I2C_ENA;
2248 wr32(E1000_CTRL_EXT, ctrl_ext);
2249 wrfl();
2250
2251 i2cctl = rd32(E1000_I2CPARAMS);
2252 i2cctl |= E1000_I2CBB_EN
2253 | E1000_I2C_CLK_OE_N
2254 | E1000_I2C_DATA_OE_N;
2255 wr32(E1000_I2CPARAMS, i2cctl);
2256 wrfl();
2257}
2258#endif
2259
2260void igb_reset(struct igb_adapter *adapter)
2261{
2262 struct pci_dev *pdev = adapter->pdev;
2263 struct e1000_hw *hw = &adapter->hw;
2264 struct e1000_mac_info *mac = &hw->mac;
2265 struct e1000_fc_info *fc = &hw->fc;
2266 u32 pba, hwm;
2267
2268 /* Repartition Pba for greater than 9k mtu
2269 * To take effect CTRL.RST is required.
2270 */
2271 switch (mac->type) {
2272 case e1000_i350:
2273 case e1000_i354:
2274 case e1000_82580:
2275 pba = rd32(E1000_RXPBS);
2276 pba = igb_rxpbs_adjust_82580(pba);
2277 break;
2278 case e1000_82576:
2279 pba = rd32(E1000_RXPBS);
2280 pba &= E1000_RXPBS_SIZE_MASK_82576;
2281 break;
2282 case e1000_82575:
2283 case e1000_i210:
2284 case e1000_i211:
2285 default:
2286 pba = E1000_PBA_34K;
2287 break;
2288 }
2289
2290 if (mac->type == e1000_82575) {
2291 u32 min_rx_space, min_tx_space, needed_tx_space;
2292
2293 /* write Rx PBA so that hardware can report correct Tx PBA */
2294 wr32(E1000_PBA, pba);
2295
2296 /* To maintain wire speed transmits, the Tx FIFO should be
2297 * large enough to accommodate two full transmit packets,
2298 * rounded up to the next 1KB and expressed in KB. Likewise,
2299 * the Rx FIFO should be large enough to accommodate at least
2300 * one full receive packet and is similarly rounded up and
2301 * expressed in KB.
2302 */
2303 min_rx_space = DIV_ROUND_UP(MAX_JUMBO_FRAME_SIZE, 1024);
2304
2305 /* The Tx FIFO also stores 16 bytes of information about the Tx
2306 * but don't include Ethernet FCS because hardware appends it.
2307 * We only need to round down to the nearest 512 byte block
2308 * count since the value we care about is 2 frames, not 1.
2309 */
2310 min_tx_space = adapter->max_frame_size;
2311 min_tx_space += sizeof(union e1000_adv_tx_desc) - ETH_FCS_LEN;
2312 min_tx_space = DIV_ROUND_UP(min_tx_space, 512);
2313
2314 /* upper 16 bits has Tx packet buffer allocation size in KB */
2315 needed_tx_space = min_tx_space - (rd32(E1000_PBA) >> 16);
2316
2317 /* If current Tx allocation is less than the min Tx FIFO size,
2318 * and the min Tx FIFO size is less than the current Rx FIFO
2319 * allocation, take space away from current Rx allocation.
2320 */
2321 if (needed_tx_space < pba) {
2322 pba -= needed_tx_space;
2323
2324 /* if short on Rx space, Rx wins and must trump Tx
2325 * adjustment
2326 */
2327 if (pba < min_rx_space)
2328 pba = min_rx_space;
2329 }
2330
2331 /* adjust PBA for jumbo frames */
2332 wr32(E1000_PBA, pba);
2333 }
2334
2335 /* flow control settings
2336 * The high water mark must be low enough to fit one full frame
2337 * after transmitting the pause frame. As such we must have enough
2338 * space to allow for us to complete our current transmit and then
2339 * receive the frame that is in progress from the link partner.
2340 * Set it to:
2341 * - the full Rx FIFO size minus one full Tx plus one full Rx frame
2342 */
2343 hwm = (pba << 10) - (adapter->max_frame_size + MAX_JUMBO_FRAME_SIZE);
2344
2345 fc->high_water = hwm & 0xFFFFFFF0; /* 16-byte granularity */
2346 fc->low_water = fc->high_water - 16;
2347 fc->pause_time = 0xFFFF;
2348 fc->send_xon = 1;
2349 fc->current_mode = fc->requested_mode;
2350
2351 /* disable receive for all VFs and wait one second */
2352 if (adapter->vfs_allocated_count) {
2353 int i;
2354
2355 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
2356 adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
2357
2358 /* ping all the active vfs to let them know we are going down */
2359 igb_ping_all_vfs(adapter);
2360
2361 /* disable transmits and receives */
2362 wr32(E1000_VFRE, 0);
2363 wr32(E1000_VFTE, 0);
2364 }
2365
2366 /* Allow time for pending master requests to run */
2367 hw->mac.ops.reset_hw(hw);
2368 wr32(E1000_WUC, 0);
2369
2370 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
2371 /* need to resetup here after media swap */
2372 adapter->ei.get_invariants(hw);
2373 adapter->flags &= ~IGB_FLAG_MEDIA_RESET;
2374 }
2375 if ((mac->type == e1000_82575 || mac->type == e1000_i350) &&
2376 (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
2377 igb_enable_mas(adapter);
2378 }
2379 if (hw->mac.ops.init_hw(hw))
2380 dev_err(&pdev->dev, "Hardware Error\n");
2381
2382 /* RAR registers were cleared during init_hw, clear mac table */
2383 igb_flush_mac_table(adapter);
2384 __dev_uc_unsync(adapter->netdev, NULL);
2385
2386 /* Recover default RAR entry */
2387 igb_set_default_mac_filter(adapter);
2388
2389 /* Flow control settings reset on hardware reset, so guarantee flow
2390 * control is off when forcing speed.
2391 */
2392 if (!hw->mac.autoneg)
2393 igb_force_mac_fc(hw);
2394
2395 igb_init_dmac(adapter, pba);
2396#ifdef CONFIG_IGB_HWMON
2397 /* Re-initialize the thermal sensor on i350 devices. */
2398 if (!test_bit(__IGB_DOWN, &adapter->state)) {
2399 if (mac->type == e1000_i350 && hw->bus.func == 0) {
2400 /* If present, re-initialize the external thermal sensor
2401 * interface.
2402 */
2403 if (adapter->ets)
2404 igb_set_i2c_bb(hw);
2405 mac->ops.init_thermal_sensor_thresh(hw);
2406 }
2407 }
2408#endif
2409 /* Re-establish EEE setting */
2410 if (hw->phy.media_type == e1000_media_type_copper) {
2411 switch (mac->type) {
2412 case e1000_i350:
2413 case e1000_i210:
2414 case e1000_i211:
2415 igb_set_eee_i350(hw, true, true);
2416 break;
2417 case e1000_i354:
2418 igb_set_eee_i354(hw, true, true);
2419 break;
2420 default:
2421 break;
2422 }
2423 }
2424 if (!netif_running(adapter->netdev))
2425 igb_power_down_link(adapter);
2426
2427 igb_update_mng_vlan(adapter);
2428
2429 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2430 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
2431
2432 /* Re-enable PTP, where applicable. */
2433 if (adapter->ptp_flags & IGB_PTP_ENABLED)
2434 igb_ptp_reset(adapter);
2435
2436 igb_get_phy_info(hw);
2437}
2438
2439static netdev_features_t igb_fix_features(struct net_device *netdev,
2440 netdev_features_t features)
2441{
2442 /* Since there is no support for separate Rx/Tx vlan accel
2443 * enable/disable make sure Tx flag is always in same state as Rx.
2444 */
2445 if (features & NETIF_F_HW_VLAN_CTAG_RX)
2446 features |= NETIF_F_HW_VLAN_CTAG_TX;
2447 else
2448 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2449
2450 return features;
2451}
2452
2453static int igb_set_features(struct net_device *netdev,
2454 netdev_features_t features)
2455{
2456 netdev_features_t changed = netdev->features ^ features;
2457 struct igb_adapter *adapter = netdev_priv(netdev);
2458
2459 if (changed & NETIF_F_HW_VLAN_CTAG_RX)
2460 igb_vlan_mode(netdev, features);
2461
2462 if (!(changed & (NETIF_F_RXALL | NETIF_F_NTUPLE)))
2463 return 0;
2464
2465 if (!(features & NETIF_F_NTUPLE)) {
2466 struct hlist_node *node2;
2467 struct igb_nfc_filter *rule;
2468
2469 spin_lock(&adapter->nfc_lock);
2470 hlist_for_each_entry_safe(rule, node2,
2471 &adapter->nfc_filter_list, nfc_node) {
2472 igb_erase_filter(adapter, rule);
2473 hlist_del(&rule->nfc_node);
2474 kfree(rule);
2475 }
2476 spin_unlock(&adapter->nfc_lock);
2477 adapter->nfc_filter_count = 0;
2478 }
2479
2480 netdev->features = features;
2481
2482 if (netif_running(netdev))
2483 igb_reinit_locked(adapter);
2484 else
2485 igb_reset(adapter);
2486
2487 return 1;
2488}
2489
2490static int igb_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
2491 struct net_device *dev,
2492 const unsigned char *addr, u16 vid,
2493 u16 flags, bool *notified,
2494 struct netlink_ext_ack *extack)
2495{
2496 /* guarantee we can provide a unique filter for the unicast address */
2497 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
2498 struct igb_adapter *adapter = netdev_priv(dev);
2499 int vfn = adapter->vfs_allocated_count;
2500
2501 if (netdev_uc_count(dev) >= igb_available_rars(adapter, vfn))
2502 return -ENOMEM;
2503 }
2504
2505 return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
2506}
2507
2508#define IGB_MAX_MAC_HDR_LEN 127
2509#define IGB_MAX_NETWORK_HDR_LEN 511
2510
2511static netdev_features_t
2512igb_features_check(struct sk_buff *skb, struct net_device *dev,
2513 netdev_features_t features)
2514{
2515 unsigned int network_hdr_len, mac_hdr_len;
2516
2517 /* Make certain the headers can be described by a context descriptor */
2518 mac_hdr_len = skb_network_offset(skb);
2519 if (unlikely(mac_hdr_len > IGB_MAX_MAC_HDR_LEN))
2520 return features & ~(NETIF_F_HW_CSUM |
2521 NETIF_F_SCTP_CRC |
2522 NETIF_F_GSO_UDP_L4 |
2523 NETIF_F_HW_VLAN_CTAG_TX |
2524 NETIF_F_TSO |
2525 NETIF_F_TSO6);
2526
2527 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
2528 if (unlikely(network_hdr_len > IGB_MAX_NETWORK_HDR_LEN))
2529 return features & ~(NETIF_F_HW_CSUM |
2530 NETIF_F_SCTP_CRC |
2531 NETIF_F_GSO_UDP_L4 |
2532 NETIF_F_TSO |
2533 NETIF_F_TSO6);
2534
2535 /* We can only support IPV4 TSO in tunnels if we can mangle the
2536 * inner IP ID field, so strip TSO if MANGLEID is not supported.
2537 */
2538 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
2539 features &= ~NETIF_F_TSO;
2540
2541 return features;
2542}
2543
2544static void igb_offload_apply(struct igb_adapter *adapter, s32 queue)
2545{
2546 if (!is_fqtss_enabled(adapter)) {
2547 enable_fqtss(adapter, true);
2548 return;
2549 }
2550
2551 igb_config_tx_modes(adapter, queue);
2552
2553 if (!is_any_cbs_enabled(adapter) && !is_any_txtime_enabled(adapter))
2554 enable_fqtss(adapter, false);
2555}
2556
2557static int igb_offload_cbs(struct igb_adapter *adapter,
2558 struct tc_cbs_qopt_offload *qopt)
2559{
2560 struct e1000_hw *hw = &adapter->hw;
2561 int err;
2562
2563 /* CBS offloading is only supported by i210 controller. */
2564 if (hw->mac.type != e1000_i210)
2565 return -EOPNOTSUPP;
2566
2567 /* CBS offloading is only supported by queue 0 and queue 1. */
2568 if (qopt->queue < 0 || qopt->queue > 1)
2569 return -EINVAL;
2570
2571 err = igb_save_cbs_params(adapter, qopt->queue, qopt->enable,
2572 qopt->idleslope, qopt->sendslope,
2573 qopt->hicredit, qopt->locredit);
2574 if (err)
2575 return err;
2576
2577 igb_offload_apply(adapter, qopt->queue);
2578
2579 return 0;
2580}
2581
2582#define ETHER_TYPE_FULL_MASK ((__force __be16)~0)
2583#define VLAN_PRIO_FULL_MASK (0x07)
2584
2585static int igb_parse_cls_flower(struct igb_adapter *adapter,
2586 struct flow_cls_offload *f,
2587 int traffic_class,
2588 struct igb_nfc_filter *input)
2589{
2590 struct flow_rule *rule = flow_cls_offload_flow_rule(f);
2591 struct flow_dissector *dissector = rule->match.dissector;
2592 struct netlink_ext_ack *extack = f->common.extack;
2593
2594 if (dissector->used_keys &
2595 ~(BIT_ULL(FLOW_DISSECTOR_KEY_BASIC) |
2596 BIT_ULL(FLOW_DISSECTOR_KEY_CONTROL) |
2597 BIT_ULL(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
2598 BIT_ULL(FLOW_DISSECTOR_KEY_VLAN))) {
2599 NL_SET_ERR_MSG_MOD(extack,
2600 "Unsupported key used, only BASIC, CONTROL, ETH_ADDRS and VLAN are supported");
2601 return -EOPNOTSUPP;
2602 }
2603
2604 if (flow_rule_match_has_control_flags(rule, extack))
2605 return -EOPNOTSUPP;
2606
2607 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
2608 struct flow_match_eth_addrs match;
2609
2610 flow_rule_match_eth_addrs(rule, &match);
2611 if (!is_zero_ether_addr(match.mask->dst)) {
2612 if (!is_broadcast_ether_addr(match.mask->dst)) {
2613 NL_SET_ERR_MSG_MOD(extack, "Only full masks are supported for destination MAC address");
2614 return -EINVAL;
2615 }
2616
2617 input->filter.match_flags |=
2618 IGB_FILTER_FLAG_DST_MAC_ADDR;
2619 ether_addr_copy(input->filter.dst_addr, match.key->dst);
2620 }
2621
2622 if (!is_zero_ether_addr(match.mask->src)) {
2623 if (!is_broadcast_ether_addr(match.mask->src)) {
2624 NL_SET_ERR_MSG_MOD(extack, "Only full masks are supported for source MAC address");
2625 return -EINVAL;
2626 }
2627
2628 input->filter.match_flags |=
2629 IGB_FILTER_FLAG_SRC_MAC_ADDR;
2630 ether_addr_copy(input->filter.src_addr, match.key->src);
2631 }
2632 }
2633
2634 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
2635 struct flow_match_basic match;
2636
2637 flow_rule_match_basic(rule, &match);
2638 if (match.mask->n_proto) {
2639 if (match.mask->n_proto != ETHER_TYPE_FULL_MASK) {
2640 NL_SET_ERR_MSG_MOD(extack, "Only full mask is supported for EtherType filter");
2641 return -EINVAL;
2642 }
2643
2644 input->filter.match_flags |= IGB_FILTER_FLAG_ETHER_TYPE;
2645 input->filter.etype = match.key->n_proto;
2646 }
2647 }
2648
2649 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN)) {
2650 struct flow_match_vlan match;
2651
2652 flow_rule_match_vlan(rule, &match);
2653 if (match.mask->vlan_priority) {
2654 if (match.mask->vlan_priority != VLAN_PRIO_FULL_MASK) {
2655 NL_SET_ERR_MSG_MOD(extack, "Only full mask is supported for VLAN priority");
2656 return -EINVAL;
2657 }
2658
2659 input->filter.match_flags |= IGB_FILTER_FLAG_VLAN_TCI;
2660 input->filter.vlan_tci =
2661 (__force __be16)match.key->vlan_priority;
2662 }
2663 }
2664
2665 input->action = traffic_class;
2666 input->cookie = f->cookie;
2667
2668 return 0;
2669}
2670
2671static int igb_configure_clsflower(struct igb_adapter *adapter,
2672 struct flow_cls_offload *cls_flower)
2673{
2674 struct netlink_ext_ack *extack = cls_flower->common.extack;
2675 struct igb_nfc_filter *filter, *f;
2676 int err, tc;
2677
2678 tc = tc_classid_to_hwtc(adapter->netdev, cls_flower->classid);
2679 if (tc < 0) {
2680 NL_SET_ERR_MSG_MOD(extack, "Invalid traffic class");
2681 return -EINVAL;
2682 }
2683
2684 filter = kzalloc(sizeof(*filter), GFP_KERNEL);
2685 if (!filter)
2686 return -ENOMEM;
2687
2688 err = igb_parse_cls_flower(adapter, cls_flower, tc, filter);
2689 if (err < 0)
2690 goto err_parse;
2691
2692 spin_lock(&adapter->nfc_lock);
2693
2694 hlist_for_each_entry(f, &adapter->nfc_filter_list, nfc_node) {
2695 if (!memcmp(&f->filter, &filter->filter, sizeof(f->filter))) {
2696 err = -EEXIST;
2697 NL_SET_ERR_MSG_MOD(extack,
2698 "This filter is already set in ethtool");
2699 goto err_locked;
2700 }
2701 }
2702
2703 hlist_for_each_entry(f, &adapter->cls_flower_list, nfc_node) {
2704 if (!memcmp(&f->filter, &filter->filter, sizeof(f->filter))) {
2705 err = -EEXIST;
2706 NL_SET_ERR_MSG_MOD(extack,
2707 "This filter is already set in cls_flower");
2708 goto err_locked;
2709 }
2710 }
2711
2712 err = igb_add_filter(adapter, filter);
2713 if (err < 0) {
2714 NL_SET_ERR_MSG_MOD(extack, "Could not add filter to the adapter");
2715 goto err_locked;
2716 }
2717
2718 hlist_add_head(&filter->nfc_node, &adapter->cls_flower_list);
2719
2720 spin_unlock(&adapter->nfc_lock);
2721
2722 return 0;
2723
2724err_locked:
2725 spin_unlock(&adapter->nfc_lock);
2726
2727err_parse:
2728 kfree(filter);
2729
2730 return err;
2731}
2732
2733static int igb_delete_clsflower(struct igb_adapter *adapter,
2734 struct flow_cls_offload *cls_flower)
2735{
2736 struct igb_nfc_filter *filter;
2737 int err;
2738
2739 spin_lock(&adapter->nfc_lock);
2740
2741 hlist_for_each_entry(filter, &adapter->cls_flower_list, nfc_node)
2742 if (filter->cookie == cls_flower->cookie)
2743 break;
2744
2745 if (!filter) {
2746 err = -ENOENT;
2747 goto out;
2748 }
2749
2750 err = igb_erase_filter(adapter, filter);
2751 if (err < 0)
2752 goto out;
2753
2754 hlist_del(&filter->nfc_node);
2755 kfree(filter);
2756
2757out:
2758 spin_unlock(&adapter->nfc_lock);
2759
2760 return err;
2761}
2762
2763static int igb_setup_tc_cls_flower(struct igb_adapter *adapter,
2764 struct flow_cls_offload *cls_flower)
2765{
2766 switch (cls_flower->command) {
2767 case FLOW_CLS_REPLACE:
2768 return igb_configure_clsflower(adapter, cls_flower);
2769 case FLOW_CLS_DESTROY:
2770 return igb_delete_clsflower(adapter, cls_flower);
2771 case FLOW_CLS_STATS:
2772 return -EOPNOTSUPP;
2773 default:
2774 return -EOPNOTSUPP;
2775 }
2776}
2777
2778static int igb_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
2779 void *cb_priv)
2780{
2781 struct igb_adapter *adapter = cb_priv;
2782
2783 if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
2784 return -EOPNOTSUPP;
2785
2786 switch (type) {
2787 case TC_SETUP_CLSFLOWER:
2788 return igb_setup_tc_cls_flower(adapter, type_data);
2789
2790 default:
2791 return -EOPNOTSUPP;
2792 }
2793}
2794
2795static int igb_offload_txtime(struct igb_adapter *adapter,
2796 struct tc_etf_qopt_offload *qopt)
2797{
2798 struct e1000_hw *hw = &adapter->hw;
2799 int err;
2800
2801 /* Launchtime offloading is only supported by i210 controller. */
2802 if (hw->mac.type != e1000_i210)
2803 return -EOPNOTSUPP;
2804
2805 /* Launchtime offloading is only supported by queues 0 and 1. */
2806 if (qopt->queue < 0 || qopt->queue > 1)
2807 return -EINVAL;
2808
2809 err = igb_save_txtime_params(adapter, qopt->queue, qopt->enable);
2810 if (err)
2811 return err;
2812
2813 igb_offload_apply(adapter, qopt->queue);
2814
2815 return 0;
2816}
2817
2818static int igb_tc_query_caps(struct igb_adapter *adapter,
2819 struct tc_query_caps_base *base)
2820{
2821 switch (base->type) {
2822 case TC_SETUP_QDISC_TAPRIO: {
2823 struct tc_taprio_caps *caps = base->caps;
2824
2825 caps->broken_mqprio = true;
2826
2827 return 0;
2828 }
2829 default:
2830 return -EOPNOTSUPP;
2831 }
2832}
2833
2834static LIST_HEAD(igb_block_cb_list);
2835
2836static int igb_setup_tc(struct net_device *dev, enum tc_setup_type type,
2837 void *type_data)
2838{
2839 struct igb_adapter *adapter = netdev_priv(dev);
2840
2841 switch (type) {
2842 case TC_QUERY_CAPS:
2843 return igb_tc_query_caps(adapter, type_data);
2844 case TC_SETUP_QDISC_CBS:
2845 return igb_offload_cbs(adapter, type_data);
2846 case TC_SETUP_BLOCK:
2847 return flow_block_cb_setup_simple(type_data,
2848 &igb_block_cb_list,
2849 igb_setup_tc_block_cb,
2850 adapter, adapter, true);
2851
2852 case TC_SETUP_QDISC_ETF:
2853 return igb_offload_txtime(adapter, type_data);
2854
2855 default:
2856 return -EOPNOTSUPP;
2857 }
2858}
2859
2860static int igb_xdp_setup(struct net_device *dev, struct netdev_bpf *bpf)
2861{
2862 int i, frame_size = dev->mtu + IGB_ETH_PKT_HDR_PAD;
2863 struct igb_adapter *adapter = netdev_priv(dev);
2864 struct bpf_prog *prog = bpf->prog, *old_prog;
2865 bool running = netif_running(dev);
2866 bool need_reset;
2867
2868 /* verify igb ring attributes are sufficient for XDP */
2869 for (i = 0; i < adapter->num_rx_queues; i++) {
2870 struct igb_ring *ring = adapter->rx_ring[i];
2871
2872 if (frame_size > igb_rx_bufsz(ring)) {
2873 NL_SET_ERR_MSG_MOD(bpf->extack,
2874 "The RX buffer size is too small for the frame size");
2875 netdev_warn(dev, "XDP RX buffer size %d is too small for the frame size %d\n",
2876 igb_rx_bufsz(ring), frame_size);
2877 return -EINVAL;
2878 }
2879 }
2880
2881 old_prog = xchg(&adapter->xdp_prog, prog);
2882 need_reset = (!!prog != !!old_prog);
2883
2884 /* device is up and bpf is added/removed, must setup the RX queues */
2885 if (need_reset && running) {
2886 igb_close(dev);
2887 } else {
2888 for (i = 0; i < adapter->num_rx_queues; i++)
2889 (void)xchg(&adapter->rx_ring[i]->xdp_prog,
2890 adapter->xdp_prog);
2891 }
2892
2893 if (old_prog)
2894 bpf_prog_put(old_prog);
2895
2896 /* bpf is just replaced, RXQ and MTU are already setup */
2897 if (!need_reset) {
2898 return 0;
2899 } else {
2900 if (prog)
2901 xdp_features_set_redirect_target(dev, true);
2902 else
2903 xdp_features_clear_redirect_target(dev);
2904 }
2905
2906 if (running)
2907 igb_open(dev);
2908
2909 return 0;
2910}
2911
2912static int igb_xdp(struct net_device *dev, struct netdev_bpf *xdp)
2913{
2914 switch (xdp->command) {
2915 case XDP_SETUP_PROG:
2916 return igb_xdp_setup(dev, xdp);
2917 default:
2918 return -EINVAL;
2919 }
2920}
2921
2922/* This function assumes __netif_tx_lock is held by the caller. */
2923static void igb_xdp_ring_update_tail(struct igb_ring *ring)
2924{
2925 lockdep_assert_held(&txring_txq(ring)->_xmit_lock);
2926
2927 /* Force memory writes to complete before letting h/w know there
2928 * are new descriptors to fetch.
2929 */
2930 wmb();
2931 writel(ring->next_to_use, ring->tail);
2932}
2933
2934static struct igb_ring *igb_xdp_tx_queue_mapping(struct igb_adapter *adapter)
2935{
2936 unsigned int r_idx = smp_processor_id();
2937
2938 if (r_idx >= adapter->num_tx_queues)
2939 r_idx = r_idx % adapter->num_tx_queues;
2940
2941 return adapter->tx_ring[r_idx];
2942}
2943
2944static int igb_xdp_xmit_back(struct igb_adapter *adapter, struct xdp_buff *xdp)
2945{
2946 struct xdp_frame *xdpf = xdp_convert_buff_to_frame(xdp);
2947 int cpu = smp_processor_id();
2948 struct igb_ring *tx_ring;
2949 struct netdev_queue *nq;
2950 u32 ret;
2951
2952 if (unlikely(!xdpf))
2953 return IGB_XDP_CONSUMED;
2954
2955 /* During program transitions its possible adapter->xdp_prog is assigned
2956 * but ring has not been configured yet. In this case simply abort xmit.
2957 */
2958 tx_ring = adapter->xdp_prog ? igb_xdp_tx_queue_mapping(adapter) : NULL;
2959 if (unlikely(!tx_ring))
2960 return IGB_XDP_CONSUMED;
2961
2962 nq = txring_txq(tx_ring);
2963 __netif_tx_lock(nq, cpu);
2964 /* Avoid transmit queue timeout since we share it with the slow path */
2965 txq_trans_cond_update(nq);
2966 ret = igb_xmit_xdp_ring(adapter, tx_ring, xdpf);
2967 __netif_tx_unlock(nq);
2968
2969 return ret;
2970}
2971
2972static int igb_xdp_xmit(struct net_device *dev, int n,
2973 struct xdp_frame **frames, u32 flags)
2974{
2975 struct igb_adapter *adapter = netdev_priv(dev);
2976 int cpu = smp_processor_id();
2977 struct igb_ring *tx_ring;
2978 struct netdev_queue *nq;
2979 int nxmit = 0;
2980 int i;
2981
2982 if (unlikely(test_bit(__IGB_DOWN, &adapter->state)))
2983 return -ENETDOWN;
2984
2985 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
2986 return -EINVAL;
2987
2988 /* During program transitions its possible adapter->xdp_prog is assigned
2989 * but ring has not been configured yet. In this case simply abort xmit.
2990 */
2991 tx_ring = adapter->xdp_prog ? igb_xdp_tx_queue_mapping(adapter) : NULL;
2992 if (unlikely(!tx_ring))
2993 return -ENXIO;
2994
2995 nq = txring_txq(tx_ring);
2996 __netif_tx_lock(nq, cpu);
2997
2998 /* Avoid transmit queue timeout since we share it with the slow path */
2999 txq_trans_cond_update(nq);
3000
3001 for (i = 0; i < n; i++) {
3002 struct xdp_frame *xdpf = frames[i];
3003 int err;
3004
3005 err = igb_xmit_xdp_ring(adapter, tx_ring, xdpf);
3006 if (err != IGB_XDP_TX)
3007 break;
3008 nxmit++;
3009 }
3010
3011 if (unlikely(flags & XDP_XMIT_FLUSH))
3012 igb_xdp_ring_update_tail(tx_ring);
3013
3014 __netif_tx_unlock(nq);
3015
3016 return nxmit;
3017}
3018
3019static const struct net_device_ops igb_netdev_ops = {
3020 .ndo_open = igb_open,
3021 .ndo_stop = igb_close,
3022 .ndo_start_xmit = igb_xmit_frame,
3023 .ndo_get_stats64 = igb_get_stats64,
3024 .ndo_set_rx_mode = igb_set_rx_mode,
3025 .ndo_set_mac_address = igb_set_mac,
3026 .ndo_change_mtu = igb_change_mtu,
3027 .ndo_eth_ioctl = igb_ioctl,
3028 .ndo_tx_timeout = igb_tx_timeout,
3029 .ndo_validate_addr = eth_validate_addr,
3030 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
3031 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
3032 .ndo_set_vf_mac = igb_ndo_set_vf_mac,
3033 .ndo_set_vf_vlan = igb_ndo_set_vf_vlan,
3034 .ndo_set_vf_rate = igb_ndo_set_vf_bw,
3035 .ndo_set_vf_spoofchk = igb_ndo_set_vf_spoofchk,
3036 .ndo_set_vf_trust = igb_ndo_set_vf_trust,
3037 .ndo_get_vf_config = igb_ndo_get_vf_config,
3038 .ndo_fix_features = igb_fix_features,
3039 .ndo_set_features = igb_set_features,
3040 .ndo_fdb_add = igb_ndo_fdb_add,
3041 .ndo_features_check = igb_features_check,
3042 .ndo_setup_tc = igb_setup_tc,
3043 .ndo_bpf = igb_xdp,
3044 .ndo_xdp_xmit = igb_xdp_xmit,
3045};
3046
3047/**
3048 * igb_set_fw_version - Configure version string for ethtool
3049 * @adapter: adapter struct
3050 **/
3051void igb_set_fw_version(struct igb_adapter *adapter)
3052{
3053 struct e1000_hw *hw = &adapter->hw;
3054 struct e1000_fw_version fw;
3055
3056 igb_get_fw_version(hw, &fw);
3057
3058 switch (hw->mac.type) {
3059 case e1000_i210:
3060 case e1000_i211:
3061 if (!(igb_get_flash_presence_i210(hw))) {
3062 snprintf(adapter->fw_version,
3063 sizeof(adapter->fw_version),
3064 "%2d.%2d-%d",
3065 fw.invm_major, fw.invm_minor,
3066 fw.invm_img_type);
3067 break;
3068 }
3069 fallthrough;
3070 default:
3071 /* if option rom is valid, display its version too */
3072 if (fw.or_valid) {
3073 snprintf(adapter->fw_version,
3074 sizeof(adapter->fw_version),
3075 "%d.%d, 0x%08x, %d.%d.%d",
3076 fw.eep_major, fw.eep_minor, fw.etrack_id,
3077 fw.or_major, fw.or_build, fw.or_patch);
3078 /* no option rom */
3079 } else if (fw.etrack_id != 0X0000) {
3080 snprintf(adapter->fw_version,
3081 sizeof(adapter->fw_version),
3082 "%d.%d, 0x%08x",
3083 fw.eep_major, fw.eep_minor, fw.etrack_id);
3084 } else {
3085 snprintf(adapter->fw_version,
3086 sizeof(adapter->fw_version),
3087 "%d.%d.%d",
3088 fw.eep_major, fw.eep_minor, fw.eep_build);
3089 }
3090 break;
3091 }
3092}
3093
3094/**
3095 * igb_init_mas - init Media Autosense feature if enabled in the NVM
3096 *
3097 * @adapter: adapter struct
3098 **/
3099static void igb_init_mas(struct igb_adapter *adapter)
3100{
3101 struct e1000_hw *hw = &adapter->hw;
3102 u16 eeprom_data;
3103
3104 hw->nvm.ops.read(hw, NVM_COMPAT, 1, &eeprom_data);
3105 switch (hw->bus.func) {
3106 case E1000_FUNC_0:
3107 if (eeprom_data & IGB_MAS_ENABLE_0) {
3108 adapter->flags |= IGB_FLAG_MAS_ENABLE;
3109 netdev_info(adapter->netdev,
3110 "MAS: Enabling Media Autosense for port %d\n",
3111 hw->bus.func);
3112 }
3113 break;
3114 case E1000_FUNC_1:
3115 if (eeprom_data & IGB_MAS_ENABLE_1) {
3116 adapter->flags |= IGB_FLAG_MAS_ENABLE;
3117 netdev_info(adapter->netdev,
3118 "MAS: Enabling Media Autosense for port %d\n",
3119 hw->bus.func);
3120 }
3121 break;
3122 case E1000_FUNC_2:
3123 if (eeprom_data & IGB_MAS_ENABLE_2) {
3124 adapter->flags |= IGB_FLAG_MAS_ENABLE;
3125 netdev_info(adapter->netdev,
3126 "MAS: Enabling Media Autosense for port %d\n",
3127 hw->bus.func);
3128 }
3129 break;
3130 case E1000_FUNC_3:
3131 if (eeprom_data & IGB_MAS_ENABLE_3) {
3132 adapter->flags |= IGB_FLAG_MAS_ENABLE;
3133 netdev_info(adapter->netdev,
3134 "MAS: Enabling Media Autosense for port %d\n",
3135 hw->bus.func);
3136 }
3137 break;
3138 default:
3139 /* Shouldn't get here */
3140 netdev_err(adapter->netdev,
3141 "MAS: Invalid port configuration, returning\n");
3142 break;
3143 }
3144}
3145
3146/**
3147 * igb_init_i2c - Init I2C interface
3148 * @adapter: pointer to adapter structure
3149 **/
3150static s32 igb_init_i2c(struct igb_adapter *adapter)
3151{
3152 s32 status = 0;
3153
3154 /* I2C interface supported on i350 devices */
3155 if (adapter->hw.mac.type != e1000_i350)
3156 return 0;
3157
3158 /* Initialize the i2c bus which is controlled by the registers.
3159 * This bus will use the i2c_algo_bit structure that implements
3160 * the protocol through toggling of the 4 bits in the register.
3161 */
3162 adapter->i2c_adap.owner = THIS_MODULE;
3163 adapter->i2c_algo = igb_i2c_algo;
3164 adapter->i2c_algo.data = adapter;
3165 adapter->i2c_adap.algo_data = &adapter->i2c_algo;
3166 adapter->i2c_adap.dev.parent = &adapter->pdev->dev;
3167 strscpy(adapter->i2c_adap.name, "igb BB",
3168 sizeof(adapter->i2c_adap.name));
3169 status = i2c_bit_add_bus(&adapter->i2c_adap);
3170 return status;
3171}
3172
3173/**
3174 * igb_probe - Device Initialization Routine
3175 * @pdev: PCI device information struct
3176 * @ent: entry in igb_pci_tbl
3177 *
3178 * Returns 0 on success, negative on failure
3179 *
3180 * igb_probe initializes an adapter identified by a pci_dev structure.
3181 * The OS initialization, configuring of the adapter private structure,
3182 * and a hardware reset occur.
3183 **/
3184static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3185{
3186 struct net_device *netdev;
3187 struct igb_adapter *adapter;
3188 struct e1000_hw *hw;
3189 u16 eeprom_data = 0;
3190 s32 ret_val;
3191 static int global_quad_port_a; /* global quad port a indication */
3192 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
3193 u8 part_str[E1000_PBANUM_LENGTH];
3194 int err;
3195
3196 /* Catch broken hardware that put the wrong VF device ID in
3197 * the PCIe SR-IOV capability.
3198 */
3199 if (pdev->is_virtfn) {
3200 WARN(1, KERN_ERR "%s (%x:%x) should not be a VF!\n",
3201 pci_name(pdev), pdev->vendor, pdev->device);
3202 return -EINVAL;
3203 }
3204
3205 err = pci_enable_device_mem(pdev);
3206 if (err)
3207 return err;
3208
3209 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
3210 if (err) {
3211 dev_err(&pdev->dev,
3212 "No usable DMA configuration, aborting\n");
3213 goto err_dma;
3214 }
3215
3216 err = pci_request_mem_regions(pdev, igb_driver_name);
3217 if (err)
3218 goto err_pci_reg;
3219
3220 pci_set_master(pdev);
3221 pci_save_state(pdev);
3222
3223 err = -ENOMEM;
3224 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
3225 IGB_MAX_TX_QUEUES);
3226 if (!netdev)
3227 goto err_alloc_etherdev;
3228
3229 SET_NETDEV_DEV(netdev, &pdev->dev);
3230
3231 pci_set_drvdata(pdev, netdev);
3232 adapter = netdev_priv(netdev);
3233 adapter->netdev = netdev;
3234 adapter->pdev = pdev;
3235 hw = &adapter->hw;
3236 hw->back = adapter;
3237 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3238
3239 err = -EIO;
3240 adapter->io_addr = pci_iomap(pdev, 0, 0);
3241 if (!adapter->io_addr)
3242 goto err_ioremap;
3243 /* hw->hw_addr can be altered, we'll use adapter->io_addr for unmap */
3244 hw->hw_addr = adapter->io_addr;
3245
3246 netdev->netdev_ops = &igb_netdev_ops;
3247 igb_set_ethtool_ops(netdev);
3248 netdev->watchdog_timeo = 5 * HZ;
3249
3250 strscpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
3251
3252 netdev->mem_start = pci_resource_start(pdev, 0);
3253 netdev->mem_end = pci_resource_end(pdev, 0);
3254
3255 /* PCI config space info */
3256 hw->vendor_id = pdev->vendor;
3257 hw->device_id = pdev->device;
3258 hw->revision_id = pdev->revision;
3259 hw->subsystem_vendor_id = pdev->subsystem_vendor;
3260 hw->subsystem_device_id = pdev->subsystem_device;
3261
3262 /* Copy the default MAC, PHY and NVM function pointers */
3263 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
3264 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
3265 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
3266 /* Initialize skew-specific constants */
3267 err = ei->get_invariants(hw);
3268 if (err)
3269 goto err_sw_init;
3270
3271 /* setup the private structure */
3272 err = igb_sw_init(adapter);
3273 if (err)
3274 goto err_sw_init;
3275
3276 igb_get_bus_info_pcie(hw);
3277
3278 hw->phy.autoneg_wait_to_complete = false;
3279
3280 /* Copper options */
3281 if (hw->phy.media_type == e1000_media_type_copper) {
3282 hw->phy.mdix = AUTO_ALL_MODES;
3283 hw->phy.disable_polarity_correction = false;
3284 hw->phy.ms_type = e1000_ms_hw_default;
3285 }
3286
3287 if (igb_check_reset_block(hw))
3288 dev_info(&pdev->dev,
3289 "PHY reset is blocked due to SOL/IDER session.\n");
3290
3291 /* features is initialized to 0 in allocation, it might have bits
3292 * set by igb_sw_init so we should use an or instead of an
3293 * assignment.
3294 */
3295 netdev->features |= NETIF_F_SG |
3296 NETIF_F_TSO |
3297 NETIF_F_TSO6 |
3298 NETIF_F_RXHASH |
3299 NETIF_F_RXCSUM |
3300 NETIF_F_HW_CSUM;
3301
3302 if (hw->mac.type >= e1000_82576)
3303 netdev->features |= NETIF_F_SCTP_CRC | NETIF_F_GSO_UDP_L4;
3304
3305 if (hw->mac.type >= e1000_i350)
3306 netdev->features |= NETIF_F_HW_TC;
3307
3308#define IGB_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
3309 NETIF_F_GSO_GRE_CSUM | \
3310 NETIF_F_GSO_IPXIP4 | \
3311 NETIF_F_GSO_IPXIP6 | \
3312 NETIF_F_GSO_UDP_TUNNEL | \
3313 NETIF_F_GSO_UDP_TUNNEL_CSUM)
3314
3315 netdev->gso_partial_features = IGB_GSO_PARTIAL_FEATURES;
3316 netdev->features |= NETIF_F_GSO_PARTIAL | IGB_GSO_PARTIAL_FEATURES;
3317
3318 /* copy netdev features into list of user selectable features */
3319 netdev->hw_features |= netdev->features |
3320 NETIF_F_HW_VLAN_CTAG_RX |
3321 NETIF_F_HW_VLAN_CTAG_TX |
3322 NETIF_F_RXALL;
3323
3324 if (hw->mac.type >= e1000_i350)
3325 netdev->hw_features |= NETIF_F_NTUPLE;
3326
3327 netdev->features |= NETIF_F_HIGHDMA;
3328
3329 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
3330 netdev->mpls_features |= NETIF_F_HW_CSUM;
3331 netdev->hw_enc_features |= netdev->vlan_features;
3332
3333 /* set this bit last since it cannot be part of vlan_features */
3334 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
3335 NETIF_F_HW_VLAN_CTAG_RX |
3336 NETIF_F_HW_VLAN_CTAG_TX;
3337
3338 netdev->priv_flags |= IFF_SUPP_NOFCS;
3339
3340 netdev->priv_flags |= IFF_UNICAST_FLT;
3341 netdev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT;
3342
3343 /* MTU range: 68 - 9216 */
3344 netdev->min_mtu = ETH_MIN_MTU;
3345 netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
3346
3347 adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
3348
3349 /* before reading the NVM, reset the controller to put the device in a
3350 * known good starting state
3351 */
3352 hw->mac.ops.reset_hw(hw);
3353
3354 /* make sure the NVM is good , i211/i210 parts can have special NVM
3355 * that doesn't contain a checksum
3356 */
3357 switch (hw->mac.type) {
3358 case e1000_i210:
3359 case e1000_i211:
3360 if (igb_get_flash_presence_i210(hw)) {
3361 if (hw->nvm.ops.validate(hw) < 0) {
3362 dev_err(&pdev->dev,
3363 "The NVM Checksum Is Not Valid\n");
3364 err = -EIO;
3365 goto err_eeprom;
3366 }
3367 }
3368 break;
3369 default:
3370 if (hw->nvm.ops.validate(hw) < 0) {
3371 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
3372 err = -EIO;
3373 goto err_eeprom;
3374 }
3375 break;
3376 }
3377
3378 if (eth_platform_get_mac_address(&pdev->dev, hw->mac.addr)) {
3379 /* copy the MAC address out of the NVM */
3380 if (hw->mac.ops.read_mac_addr(hw))
3381 dev_err(&pdev->dev, "NVM Read Error\n");
3382 }
3383
3384 eth_hw_addr_set(netdev, hw->mac.addr);
3385
3386 if (!is_valid_ether_addr(netdev->dev_addr)) {
3387 dev_err(&pdev->dev, "Invalid MAC Address\n");
3388 err = -EIO;
3389 goto err_eeprom;
3390 }
3391
3392 igb_set_default_mac_filter(adapter);
3393
3394 /* get firmware version for ethtool -i */
3395 igb_set_fw_version(adapter);
3396
3397 /* configure RXPBSIZE and TXPBSIZE */
3398 if (hw->mac.type == e1000_i210) {
3399 wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
3400 wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
3401 }
3402
3403 timer_setup(&adapter->watchdog_timer, igb_watchdog, 0);
3404 timer_setup(&adapter->phy_info_timer, igb_update_phy_info, 0);
3405
3406 INIT_WORK(&adapter->reset_task, igb_reset_task);
3407 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
3408
3409 /* Initialize link properties that are user-changeable */
3410 adapter->fc_autoneg = true;
3411 hw->mac.autoneg = true;
3412 hw->phy.autoneg_advertised = 0x2f;
3413
3414 hw->fc.requested_mode = e1000_fc_default;
3415 hw->fc.current_mode = e1000_fc_default;
3416
3417 igb_validate_mdi_setting(hw);
3418
3419 /* By default, support wake on port A */
3420 if (hw->bus.func == 0)
3421 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3422
3423 /* Check the NVM for wake support on non-port A ports */
3424 if (hw->mac.type >= e1000_82580)
3425 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
3426 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
3427 &eeprom_data);
3428 else if (hw->bus.func == 1)
3429 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
3430
3431 if (eeprom_data & IGB_EEPROM_APME)
3432 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3433
3434 /* now that we have the eeprom settings, apply the special cases where
3435 * the eeprom may be wrong or the board simply won't support wake on
3436 * lan on a particular port
3437 */
3438 switch (pdev->device) {
3439 case E1000_DEV_ID_82575GB_QUAD_COPPER:
3440 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3441 break;
3442 case E1000_DEV_ID_82575EB_FIBER_SERDES:
3443 case E1000_DEV_ID_82576_FIBER:
3444 case E1000_DEV_ID_82576_SERDES:
3445 /* Wake events only supported on port A for dual fiber
3446 * regardless of eeprom setting
3447 */
3448 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
3449 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3450 break;
3451 case E1000_DEV_ID_82576_QUAD_COPPER:
3452 case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
3453 /* if quad port adapter, disable WoL on all but port A */
3454 if (global_quad_port_a != 0)
3455 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3456 else
3457 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
3458 /* Reset for multiple quad port adapters */
3459 if (++global_quad_port_a == 4)
3460 global_quad_port_a = 0;
3461 break;
3462 default:
3463 /* If the device can't wake, don't set software support */
3464 if (!device_can_wakeup(&adapter->pdev->dev))
3465 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3466 }
3467
3468 /* initialize the wol settings based on the eeprom settings */
3469 if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
3470 adapter->wol |= E1000_WUFC_MAG;
3471
3472 /* Some vendors want WoL disabled by default, but still supported */
3473 if ((hw->mac.type == e1000_i350) &&
3474 (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
3475 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3476 adapter->wol = 0;
3477 }
3478
3479 /* Some vendors want the ability to Use the EEPROM setting as
3480 * enable/disable only, and not for capability
3481 */
3482 if (((hw->mac.type == e1000_i350) ||
3483 (hw->mac.type == e1000_i354)) &&
3484 (pdev->subsystem_vendor == PCI_VENDOR_ID_DELL)) {
3485 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3486 adapter->wol = 0;
3487 }
3488 if (hw->mac.type == e1000_i350) {
3489 if (((pdev->subsystem_device == 0x5001) ||
3490 (pdev->subsystem_device == 0x5002)) &&
3491 (hw->bus.func == 0)) {
3492 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3493 adapter->wol = 0;
3494 }
3495 if (pdev->subsystem_device == 0x1F52)
3496 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3497 }
3498
3499 device_set_wakeup_enable(&adapter->pdev->dev,
3500 adapter->flags & IGB_FLAG_WOL_SUPPORTED);
3501
3502 /* reset the hardware with the new settings */
3503 igb_reset(adapter);
3504
3505 /* Init the I2C interface */
3506 err = igb_init_i2c(adapter);
3507 if (err) {
3508 dev_err(&pdev->dev, "failed to init i2c interface\n");
3509 goto err_eeprom;
3510 }
3511
3512 /* let the f/w know that the h/w is now under the control of the
3513 * driver.
3514 */
3515 igb_get_hw_control(adapter);
3516
3517 strcpy(netdev->name, "eth%d");
3518 err = register_netdev(netdev);
3519 if (err)
3520 goto err_register;
3521
3522 /* carrier off reporting is important to ethtool even BEFORE open */
3523 netif_carrier_off(netdev);
3524
3525#ifdef CONFIG_IGB_DCA
3526 if (dca_add_requester(&pdev->dev) == 0) {
3527 adapter->flags |= IGB_FLAG_DCA_ENABLED;
3528 dev_info(&pdev->dev, "DCA enabled\n");
3529 igb_setup_dca(adapter);
3530 }
3531
3532#endif
3533#ifdef CONFIG_IGB_HWMON
3534 /* Initialize the thermal sensor on i350 devices. */
3535 if (hw->mac.type == e1000_i350 && hw->bus.func == 0) {
3536 u16 ets_word;
3537
3538 /* Read the NVM to determine if this i350 device supports an
3539 * external thermal sensor.
3540 */
3541 hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word);
3542 if (ets_word != 0x0000 && ets_word != 0xFFFF)
3543 adapter->ets = true;
3544 else
3545 adapter->ets = false;
3546 /* Only enable I2C bit banging if an external thermal
3547 * sensor is supported.
3548 */
3549 if (adapter->ets)
3550 igb_set_i2c_bb(hw);
3551 hw->mac.ops.init_thermal_sensor_thresh(hw);
3552 if (igb_sysfs_init(adapter))
3553 dev_err(&pdev->dev,
3554 "failed to allocate sysfs resources\n");
3555 } else {
3556 adapter->ets = false;
3557 }
3558#endif
3559 /* Check if Media Autosense is enabled */
3560 adapter->ei = *ei;
3561 if (hw->dev_spec._82575.mas_capable)
3562 igb_init_mas(adapter);
3563
3564 /* do hw tstamp init after resetting */
3565 igb_ptp_init(adapter);
3566
3567 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
3568 /* print bus type/speed/width info, not applicable to i354 */
3569 if (hw->mac.type != e1000_i354) {
3570 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
3571 netdev->name,
3572 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
3573 (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
3574 "unknown"),
3575 ((hw->bus.width == e1000_bus_width_pcie_x4) ?
3576 "Width x4" :
3577 (hw->bus.width == e1000_bus_width_pcie_x2) ?
3578 "Width x2" :
3579 (hw->bus.width == e1000_bus_width_pcie_x1) ?
3580 "Width x1" : "unknown"), netdev->dev_addr);
3581 }
3582
3583 if ((hw->mac.type == e1000_82576 &&
3584 rd32(E1000_EECD) & E1000_EECD_PRES) ||
3585 (hw->mac.type >= e1000_i210 ||
3586 igb_get_flash_presence_i210(hw))) {
3587 ret_val = igb_read_part_string(hw, part_str,
3588 E1000_PBANUM_LENGTH);
3589 } else {
3590 ret_val = -E1000_ERR_INVM_VALUE_NOT_FOUND;
3591 }
3592
3593 if (ret_val)
3594 strcpy(part_str, "Unknown");
3595 dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
3596 dev_info(&pdev->dev,
3597 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
3598 (adapter->flags & IGB_FLAG_HAS_MSIX) ? "MSI-X" :
3599 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
3600 adapter->num_rx_queues, adapter->num_tx_queues);
3601 if (hw->phy.media_type == e1000_media_type_copper) {
3602 switch (hw->mac.type) {
3603 case e1000_i350:
3604 case e1000_i210:
3605 case e1000_i211:
3606 /* Enable EEE for internal copper PHY devices */
3607 err = igb_set_eee_i350(hw, true, true);
3608 if ((!err) &&
3609 (!hw->dev_spec._82575.eee_disable)) {
3610 adapter->eee_advert =
3611 MDIO_EEE_100TX | MDIO_EEE_1000T;
3612 adapter->flags |= IGB_FLAG_EEE;
3613 }
3614 break;
3615 case e1000_i354:
3616 if ((rd32(E1000_CTRL_EXT) &
3617 E1000_CTRL_EXT_LINK_MODE_SGMII)) {
3618 err = igb_set_eee_i354(hw, true, true);
3619 if ((!err) &&
3620 (!hw->dev_spec._82575.eee_disable)) {
3621 adapter->eee_advert =
3622 MDIO_EEE_100TX | MDIO_EEE_1000T;
3623 adapter->flags |= IGB_FLAG_EEE;
3624 }
3625 }
3626 break;
3627 default:
3628 break;
3629 }
3630 }
3631
3632 dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_NO_DIRECT_COMPLETE);
3633
3634 pm_runtime_put_noidle(&pdev->dev);
3635 return 0;
3636
3637err_register:
3638 igb_release_hw_control(adapter);
3639 memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap));
3640err_eeprom:
3641 if (!igb_check_reset_block(hw))
3642 igb_reset_phy(hw);
3643
3644 if (hw->flash_address)
3645 iounmap(hw->flash_address);
3646err_sw_init:
3647 kfree(adapter->mac_table);
3648 kfree(adapter->shadow_vfta);
3649 igb_clear_interrupt_scheme(adapter);
3650#ifdef CONFIG_PCI_IOV
3651 igb_disable_sriov(pdev, false);
3652#endif
3653 pci_iounmap(pdev, adapter->io_addr);
3654err_ioremap:
3655 free_netdev(netdev);
3656err_alloc_etherdev:
3657 pci_release_mem_regions(pdev);
3658err_pci_reg:
3659err_dma:
3660 pci_disable_device(pdev);
3661 return err;
3662}
3663
3664#ifdef CONFIG_PCI_IOV
3665static int igb_sriov_reinit(struct pci_dev *dev)
3666{
3667 struct net_device *netdev = pci_get_drvdata(dev);
3668 struct igb_adapter *adapter = netdev_priv(netdev);
3669 struct pci_dev *pdev = adapter->pdev;
3670
3671 rtnl_lock();
3672
3673 if (netif_running(netdev))
3674 igb_close(netdev);
3675 else
3676 igb_reset(adapter);
3677
3678 igb_clear_interrupt_scheme(adapter);
3679
3680 igb_init_queue_configuration(adapter);
3681
3682 if (igb_init_interrupt_scheme(adapter, true)) {
3683 rtnl_unlock();
3684 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
3685 return -ENOMEM;
3686 }
3687
3688 if (netif_running(netdev))
3689 igb_open(netdev);
3690
3691 rtnl_unlock();
3692
3693 return 0;
3694}
3695
3696static int igb_disable_sriov(struct pci_dev *pdev, bool reinit)
3697{
3698 struct net_device *netdev = pci_get_drvdata(pdev);
3699 struct igb_adapter *adapter = netdev_priv(netdev);
3700 struct e1000_hw *hw = &adapter->hw;
3701 unsigned long flags;
3702
3703 /* reclaim resources allocated to VFs */
3704 if (adapter->vf_data) {
3705 /* disable iov and allow time for transactions to clear */
3706 if (pci_vfs_assigned(pdev)) {
3707 dev_warn(&pdev->dev,
3708 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
3709 return -EPERM;
3710 } else {
3711 pci_disable_sriov(pdev);
3712 msleep(500);
3713 }
3714 spin_lock_irqsave(&adapter->vfs_lock, flags);
3715 kfree(adapter->vf_mac_list);
3716 adapter->vf_mac_list = NULL;
3717 kfree(adapter->vf_data);
3718 adapter->vf_data = NULL;
3719 adapter->vfs_allocated_count = 0;
3720 spin_unlock_irqrestore(&adapter->vfs_lock, flags);
3721 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
3722 wrfl();
3723 msleep(100);
3724 dev_info(&pdev->dev, "IOV Disabled\n");
3725
3726 /* Re-enable DMA Coalescing flag since IOV is turned off */
3727 adapter->flags |= IGB_FLAG_DMAC;
3728 }
3729
3730 return reinit ? igb_sriov_reinit(pdev) : 0;
3731}
3732
3733static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs, bool reinit)
3734{
3735 struct net_device *netdev = pci_get_drvdata(pdev);
3736 struct igb_adapter *adapter = netdev_priv(netdev);
3737 int old_vfs = pci_num_vf(pdev);
3738 struct vf_mac_filter *mac_list;
3739 int err = 0;
3740 int num_vf_mac_filters, i;
3741
3742 if (!(adapter->flags & IGB_FLAG_HAS_MSIX) || num_vfs > 7) {
3743 err = -EPERM;
3744 goto out;
3745 }
3746 if (!num_vfs)
3747 goto out;
3748
3749 if (old_vfs) {
3750 dev_info(&pdev->dev, "%d pre-allocated VFs found - override max_vfs setting of %d\n",
3751 old_vfs, max_vfs);
3752 adapter->vfs_allocated_count = old_vfs;
3753 } else
3754 adapter->vfs_allocated_count = num_vfs;
3755
3756 adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
3757 sizeof(struct vf_data_storage), GFP_KERNEL);
3758
3759 /* if allocation failed then we do not support SR-IOV */
3760 if (!adapter->vf_data) {
3761 adapter->vfs_allocated_count = 0;
3762 err = -ENOMEM;
3763 goto out;
3764 }
3765
3766 /* Due to the limited number of RAR entries calculate potential
3767 * number of MAC filters available for the VFs. Reserve entries
3768 * for PF default MAC, PF MAC filters and at least one RAR entry
3769 * for each VF for VF MAC.
3770 */
3771 num_vf_mac_filters = adapter->hw.mac.rar_entry_count -
3772 (1 + IGB_PF_MAC_FILTERS_RESERVED +
3773 adapter->vfs_allocated_count);
3774
3775 adapter->vf_mac_list = kcalloc(num_vf_mac_filters,
3776 sizeof(struct vf_mac_filter),
3777 GFP_KERNEL);
3778
3779 mac_list = adapter->vf_mac_list;
3780 INIT_LIST_HEAD(&adapter->vf_macs.l);
3781
3782 if (adapter->vf_mac_list) {
3783 /* Initialize list of VF MAC filters */
3784 for (i = 0; i < num_vf_mac_filters; i++) {
3785 mac_list->vf = -1;
3786 mac_list->free = true;
3787 list_add(&mac_list->l, &adapter->vf_macs.l);
3788 mac_list++;
3789 }
3790 } else {
3791 /* If we could not allocate memory for the VF MAC filters
3792 * we can continue without this feature but warn user.
3793 */
3794 dev_err(&pdev->dev,
3795 "Unable to allocate memory for VF MAC filter list\n");
3796 }
3797
3798 dev_info(&pdev->dev, "%d VFs allocated\n",
3799 adapter->vfs_allocated_count);
3800 for (i = 0; i < adapter->vfs_allocated_count; i++)
3801 igb_vf_configure(adapter, i);
3802
3803 /* DMA Coalescing is not supported in IOV mode. */
3804 adapter->flags &= ~IGB_FLAG_DMAC;
3805
3806 if (reinit) {
3807 err = igb_sriov_reinit(pdev);
3808 if (err)
3809 goto err_out;
3810 }
3811
3812 /* only call pci_enable_sriov() if no VFs are allocated already */
3813 if (!old_vfs) {
3814 err = pci_enable_sriov(pdev, adapter->vfs_allocated_count);
3815 if (err)
3816 goto err_out;
3817 }
3818
3819 goto out;
3820
3821err_out:
3822 kfree(adapter->vf_mac_list);
3823 adapter->vf_mac_list = NULL;
3824 kfree(adapter->vf_data);
3825 adapter->vf_data = NULL;
3826 adapter->vfs_allocated_count = 0;
3827out:
3828 return err;
3829}
3830
3831#endif
3832/**
3833 * igb_remove_i2c - Cleanup I2C interface
3834 * @adapter: pointer to adapter structure
3835 **/
3836static void igb_remove_i2c(struct igb_adapter *adapter)
3837{
3838 /* free the adapter bus structure */
3839 i2c_del_adapter(&adapter->i2c_adap);
3840}
3841
3842/**
3843 * igb_remove - Device Removal Routine
3844 * @pdev: PCI device information struct
3845 *
3846 * igb_remove is called by the PCI subsystem to alert the driver
3847 * that it should release a PCI device. The could be caused by a
3848 * Hot-Plug event, or because the driver is going to be removed from
3849 * memory.
3850 **/
3851static void igb_remove(struct pci_dev *pdev)
3852{
3853 struct net_device *netdev = pci_get_drvdata(pdev);
3854 struct igb_adapter *adapter = netdev_priv(netdev);
3855 struct e1000_hw *hw = &adapter->hw;
3856
3857 pm_runtime_get_noresume(&pdev->dev);
3858#ifdef CONFIG_IGB_HWMON
3859 igb_sysfs_exit(adapter);
3860#endif
3861 igb_remove_i2c(adapter);
3862 igb_ptp_stop(adapter);
3863 /* The watchdog timer may be rescheduled, so explicitly
3864 * disable watchdog from being rescheduled.
3865 */
3866 set_bit(__IGB_DOWN, &adapter->state);
3867 del_timer_sync(&adapter->watchdog_timer);
3868 del_timer_sync(&adapter->phy_info_timer);
3869
3870 cancel_work_sync(&adapter->reset_task);
3871 cancel_work_sync(&adapter->watchdog_task);
3872
3873#ifdef CONFIG_IGB_DCA
3874 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
3875 dev_info(&pdev->dev, "DCA disabled\n");
3876 dca_remove_requester(&pdev->dev);
3877 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
3878 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
3879 }
3880#endif
3881
3882 /* Release control of h/w to f/w. If f/w is AMT enabled, this
3883 * would have already happened in close and is redundant.
3884 */
3885 igb_release_hw_control(adapter);
3886
3887#ifdef CONFIG_PCI_IOV
3888 igb_disable_sriov(pdev, false);
3889#endif
3890
3891 unregister_netdev(netdev);
3892
3893 igb_clear_interrupt_scheme(adapter);
3894
3895 pci_iounmap(pdev, adapter->io_addr);
3896 if (hw->flash_address)
3897 iounmap(hw->flash_address);
3898 pci_release_mem_regions(pdev);
3899
3900 kfree(adapter->mac_table);
3901 kfree(adapter->shadow_vfta);
3902 free_netdev(netdev);
3903
3904 pci_disable_device(pdev);
3905}
3906
3907/**
3908 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
3909 * @adapter: board private structure to initialize
3910 *
3911 * This function initializes the vf specific data storage and then attempts to
3912 * allocate the VFs. The reason for ordering it this way is because it is much
3913 * more expensive time wise to disable SR-IOV than it is to allocate and free
3914 * the memory for the VFs.
3915 **/
3916static void igb_probe_vfs(struct igb_adapter *adapter)
3917{
3918#ifdef CONFIG_PCI_IOV
3919 struct pci_dev *pdev = adapter->pdev;
3920 struct e1000_hw *hw = &adapter->hw;
3921
3922 /* Virtualization features not supported on i210 and 82580 family. */
3923 if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211) ||
3924 (hw->mac.type == e1000_82580))
3925 return;
3926
3927 /* Of the below we really only want the effect of getting
3928 * IGB_FLAG_HAS_MSIX set (if available), without which
3929 * igb_enable_sriov() has no effect.
3930 */
3931 igb_set_interrupt_capability(adapter, true);
3932 igb_reset_interrupt_capability(adapter);
3933
3934 pci_sriov_set_totalvfs(pdev, 7);
3935 igb_enable_sriov(pdev, max_vfs, false);
3936
3937#endif /* CONFIG_PCI_IOV */
3938}
3939
3940unsigned int igb_get_max_rss_queues(struct igb_adapter *adapter)
3941{
3942 struct e1000_hw *hw = &adapter->hw;
3943 unsigned int max_rss_queues;
3944
3945 /* Determine the maximum number of RSS queues supported. */
3946 switch (hw->mac.type) {
3947 case e1000_i211:
3948 max_rss_queues = IGB_MAX_RX_QUEUES_I211;
3949 break;
3950 case e1000_82575:
3951 case e1000_i210:
3952 max_rss_queues = IGB_MAX_RX_QUEUES_82575;
3953 break;
3954 case e1000_i350:
3955 /* I350 cannot do RSS and SR-IOV at the same time */
3956 if (!!adapter->vfs_allocated_count) {
3957 max_rss_queues = 1;
3958 break;
3959 }
3960 fallthrough;
3961 case e1000_82576:
3962 if (!!adapter->vfs_allocated_count) {
3963 max_rss_queues = 2;
3964 break;
3965 }
3966 fallthrough;
3967 case e1000_82580:
3968 case e1000_i354:
3969 default:
3970 max_rss_queues = IGB_MAX_RX_QUEUES;
3971 break;
3972 }
3973
3974 return max_rss_queues;
3975}
3976
3977static void igb_init_queue_configuration(struct igb_adapter *adapter)
3978{
3979 u32 max_rss_queues;
3980
3981 max_rss_queues = igb_get_max_rss_queues(adapter);
3982 adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
3983
3984 igb_set_flag_queue_pairs(adapter, max_rss_queues);
3985}
3986
3987void igb_set_flag_queue_pairs(struct igb_adapter *adapter,
3988 const u32 max_rss_queues)
3989{
3990 struct e1000_hw *hw = &adapter->hw;
3991
3992 /* Determine if we need to pair queues. */
3993 switch (hw->mac.type) {
3994 case e1000_82575:
3995 case e1000_i211:
3996 /* Device supports enough interrupts without queue pairing. */
3997 break;
3998 case e1000_82576:
3999 case e1000_82580:
4000 case e1000_i350:
4001 case e1000_i354:
4002 case e1000_i210:
4003 default:
4004 /* If rss_queues > half of max_rss_queues, pair the queues in
4005 * order to conserve interrupts due to limited supply.
4006 */
4007 if (adapter->rss_queues > (max_rss_queues / 2))
4008 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
4009 else
4010 adapter->flags &= ~IGB_FLAG_QUEUE_PAIRS;
4011 break;
4012 }
4013}
4014
4015/**
4016 * igb_sw_init - Initialize general software structures (struct igb_adapter)
4017 * @adapter: board private structure to initialize
4018 *
4019 * igb_sw_init initializes the Adapter private data structure.
4020 * Fields are initialized based on PCI device information and
4021 * OS network device settings (MTU size).
4022 **/
4023static int igb_sw_init(struct igb_adapter *adapter)
4024{
4025 struct e1000_hw *hw = &adapter->hw;
4026 struct net_device *netdev = adapter->netdev;
4027 struct pci_dev *pdev = adapter->pdev;
4028
4029 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
4030
4031 /* set default ring sizes */
4032 adapter->tx_ring_count = IGB_DEFAULT_TXD;
4033 adapter->rx_ring_count = IGB_DEFAULT_RXD;
4034
4035 /* set default ITR values */
4036 adapter->rx_itr_setting = IGB_DEFAULT_ITR;
4037 adapter->tx_itr_setting = IGB_DEFAULT_ITR;
4038
4039 /* set default work limits */
4040 adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
4041
4042 adapter->max_frame_size = netdev->mtu + IGB_ETH_PKT_HDR_PAD;
4043 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
4044
4045 spin_lock_init(&adapter->nfc_lock);
4046 spin_lock_init(&adapter->stats64_lock);
4047
4048 /* init spinlock to avoid concurrency of VF resources */
4049 spin_lock_init(&adapter->vfs_lock);
4050#ifdef CONFIG_PCI_IOV
4051 switch (hw->mac.type) {
4052 case e1000_82576:
4053 case e1000_i350:
4054 if (max_vfs > 7) {
4055 dev_warn(&pdev->dev,
4056 "Maximum of 7 VFs per PF, using max\n");
4057 max_vfs = adapter->vfs_allocated_count = 7;
4058 } else
4059 adapter->vfs_allocated_count = max_vfs;
4060 if (adapter->vfs_allocated_count)
4061 dev_warn(&pdev->dev,
4062 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
4063 break;
4064 default:
4065 break;
4066 }
4067#endif /* CONFIG_PCI_IOV */
4068
4069 /* Assume MSI-X interrupts, will be checked during IRQ allocation */
4070 adapter->flags |= IGB_FLAG_HAS_MSIX;
4071
4072 adapter->mac_table = kcalloc(hw->mac.rar_entry_count,
4073 sizeof(struct igb_mac_addr),
4074 GFP_KERNEL);
4075 if (!adapter->mac_table)
4076 return -ENOMEM;
4077
4078 igb_probe_vfs(adapter);
4079
4080 igb_init_queue_configuration(adapter);
4081
4082 /* Setup and initialize a copy of the hw vlan table array */
4083 adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32),
4084 GFP_KERNEL);
4085 if (!adapter->shadow_vfta)
4086 return -ENOMEM;
4087
4088 /* This call may decrease the number of queues */
4089 if (igb_init_interrupt_scheme(adapter, true)) {
4090 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
4091 return -ENOMEM;
4092 }
4093
4094 /* Explicitly disable IRQ since the NIC can be in any state. */
4095 igb_irq_disable(adapter);
4096
4097 if (hw->mac.type >= e1000_i350)
4098 adapter->flags &= ~IGB_FLAG_DMAC;
4099
4100 set_bit(__IGB_DOWN, &adapter->state);
4101 return 0;
4102}
4103
4104/**
4105 * __igb_open - Called when a network interface is made active
4106 * @netdev: network interface device structure
4107 * @resuming: indicates whether we are in a resume call
4108 *
4109 * Returns 0 on success, negative value on failure
4110 *
4111 * The open entry point is called when a network interface is made
4112 * active by the system (IFF_UP). At this point all resources needed
4113 * for transmit and receive operations are allocated, the interrupt
4114 * handler is registered with the OS, the watchdog timer is started,
4115 * and the stack is notified that the interface is ready.
4116 **/
4117static int __igb_open(struct net_device *netdev, bool resuming)
4118{
4119 struct igb_adapter *adapter = netdev_priv(netdev);
4120 struct e1000_hw *hw = &adapter->hw;
4121 struct pci_dev *pdev = adapter->pdev;
4122 int err;
4123 int i;
4124
4125 /* disallow open during test */
4126 if (test_bit(__IGB_TESTING, &adapter->state)) {
4127 WARN_ON(resuming);
4128 return -EBUSY;
4129 }
4130
4131 if (!resuming)
4132 pm_runtime_get_sync(&pdev->dev);
4133
4134 netif_carrier_off(netdev);
4135
4136 /* allocate transmit descriptors */
4137 err = igb_setup_all_tx_resources(adapter);
4138 if (err)
4139 goto err_setup_tx;
4140
4141 /* allocate receive descriptors */
4142 err = igb_setup_all_rx_resources(adapter);
4143 if (err)
4144 goto err_setup_rx;
4145
4146 igb_power_up_link(adapter);
4147
4148 /* before we allocate an interrupt, we must be ready to handle it.
4149 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
4150 * as soon as we call pci_request_irq, so we have to setup our
4151 * clean_rx handler before we do so.
4152 */
4153 igb_configure(adapter);
4154
4155 err = igb_request_irq(adapter);
4156 if (err)
4157 goto err_req_irq;
4158
4159 /* Notify the stack of the actual queue counts. */
4160 err = netif_set_real_num_tx_queues(adapter->netdev,
4161 adapter->num_tx_queues);
4162 if (err)
4163 goto err_set_queues;
4164
4165 err = netif_set_real_num_rx_queues(adapter->netdev,
4166 adapter->num_rx_queues);
4167 if (err)
4168 goto err_set_queues;
4169
4170 /* From here on the code is the same as igb_up() */
4171 clear_bit(__IGB_DOWN, &adapter->state);
4172
4173 for (i = 0; i < adapter->num_q_vectors; i++)
4174 napi_enable(&(adapter->q_vector[i]->napi));
4175
4176 /* Clear any pending interrupts. */
4177 rd32(E1000_TSICR);
4178 rd32(E1000_ICR);
4179
4180 igb_irq_enable(adapter);
4181
4182 /* notify VFs that reset has been completed */
4183 if (adapter->vfs_allocated_count) {
4184 u32 reg_data = rd32(E1000_CTRL_EXT);
4185
4186 reg_data |= E1000_CTRL_EXT_PFRSTD;
4187 wr32(E1000_CTRL_EXT, reg_data);
4188 }
4189
4190 netif_tx_start_all_queues(netdev);
4191
4192 if (!resuming)
4193 pm_runtime_put(&pdev->dev);
4194
4195 /* start the watchdog. */
4196 hw->mac.get_link_status = 1;
4197 schedule_work(&adapter->watchdog_task);
4198
4199 return 0;
4200
4201err_set_queues:
4202 igb_free_irq(adapter);
4203err_req_irq:
4204 igb_release_hw_control(adapter);
4205 igb_power_down_link(adapter);
4206 igb_free_all_rx_resources(adapter);
4207err_setup_rx:
4208 igb_free_all_tx_resources(adapter);
4209err_setup_tx:
4210 igb_reset(adapter);
4211 if (!resuming)
4212 pm_runtime_put(&pdev->dev);
4213
4214 return err;
4215}
4216
4217int igb_open(struct net_device *netdev)
4218{
4219 return __igb_open(netdev, false);
4220}
4221
4222/**
4223 * __igb_close - Disables a network interface
4224 * @netdev: network interface device structure
4225 * @suspending: indicates we are in a suspend call
4226 *
4227 * Returns 0, this is not allowed to fail
4228 *
4229 * The close entry point is called when an interface is de-activated
4230 * by the OS. The hardware is still under the driver's control, but
4231 * needs to be disabled. A global MAC reset is issued to stop the
4232 * hardware, and all transmit and receive resources are freed.
4233 **/
4234static int __igb_close(struct net_device *netdev, bool suspending)
4235{
4236 struct igb_adapter *adapter = netdev_priv(netdev);
4237 struct pci_dev *pdev = adapter->pdev;
4238
4239 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
4240
4241 if (!suspending)
4242 pm_runtime_get_sync(&pdev->dev);
4243
4244 igb_down(adapter);
4245 igb_free_irq(adapter);
4246
4247 igb_free_all_tx_resources(adapter);
4248 igb_free_all_rx_resources(adapter);
4249
4250 if (!suspending)
4251 pm_runtime_put_sync(&pdev->dev);
4252 return 0;
4253}
4254
4255int igb_close(struct net_device *netdev)
4256{
4257 if (netif_device_present(netdev) || netdev->dismantle)
4258 return __igb_close(netdev, false);
4259 return 0;
4260}
4261
4262/**
4263 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
4264 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4265 *
4266 * Return 0 on success, negative on failure
4267 **/
4268int igb_setup_tx_resources(struct igb_ring *tx_ring)
4269{
4270 struct device *dev = tx_ring->dev;
4271 int size;
4272
4273 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
4274
4275 tx_ring->tx_buffer_info = vmalloc(size);
4276 if (!tx_ring->tx_buffer_info)
4277 goto err;
4278
4279 /* round up to nearest 4K */
4280 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
4281 tx_ring->size = ALIGN(tx_ring->size, 4096);
4282
4283 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4284 &tx_ring->dma, GFP_KERNEL);
4285 if (!tx_ring->desc)
4286 goto err;
4287
4288 tx_ring->next_to_use = 0;
4289 tx_ring->next_to_clean = 0;
4290
4291 return 0;
4292
4293err:
4294 vfree(tx_ring->tx_buffer_info);
4295 tx_ring->tx_buffer_info = NULL;
4296 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4297 return -ENOMEM;
4298}
4299
4300/**
4301 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
4302 * (Descriptors) for all queues
4303 * @adapter: board private structure
4304 *
4305 * Return 0 on success, negative on failure
4306 **/
4307static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
4308{
4309 struct pci_dev *pdev = adapter->pdev;
4310 int i, err = 0;
4311
4312 for (i = 0; i < adapter->num_tx_queues; i++) {
4313 err = igb_setup_tx_resources(adapter->tx_ring[i]);
4314 if (err) {
4315 dev_err(&pdev->dev,
4316 "Allocation for Tx Queue %u failed\n", i);
4317 for (i--; i >= 0; i--)
4318 igb_free_tx_resources(adapter->tx_ring[i]);
4319 break;
4320 }
4321 }
4322
4323 return err;
4324}
4325
4326/**
4327 * igb_setup_tctl - configure the transmit control registers
4328 * @adapter: Board private structure
4329 **/
4330void igb_setup_tctl(struct igb_adapter *adapter)
4331{
4332 struct e1000_hw *hw = &adapter->hw;
4333 u32 tctl;
4334
4335 /* disable queue 0 which is enabled by default on 82575 and 82576 */
4336 wr32(E1000_TXDCTL(0), 0);
4337
4338 /* Program the Transmit Control Register */
4339 tctl = rd32(E1000_TCTL);
4340 tctl &= ~E1000_TCTL_CT;
4341 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
4342 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
4343
4344 igb_config_collision_dist(hw);
4345
4346 /* Enable transmits */
4347 tctl |= E1000_TCTL_EN;
4348
4349 wr32(E1000_TCTL, tctl);
4350}
4351
4352/**
4353 * igb_configure_tx_ring - Configure transmit ring after Reset
4354 * @adapter: board private structure
4355 * @ring: tx ring to configure
4356 *
4357 * Configure a transmit ring after a reset.
4358 **/
4359void igb_configure_tx_ring(struct igb_adapter *adapter,
4360 struct igb_ring *ring)
4361{
4362 struct e1000_hw *hw = &adapter->hw;
4363 u32 txdctl = 0;
4364 u64 tdba = ring->dma;
4365 int reg_idx = ring->reg_idx;
4366
4367 wr32(E1000_TDLEN(reg_idx),
4368 ring->count * sizeof(union e1000_adv_tx_desc));
4369 wr32(E1000_TDBAL(reg_idx),
4370 tdba & 0x00000000ffffffffULL);
4371 wr32(E1000_TDBAH(reg_idx), tdba >> 32);
4372
4373 ring->tail = adapter->io_addr + E1000_TDT(reg_idx);
4374 wr32(E1000_TDH(reg_idx), 0);
4375 writel(0, ring->tail);
4376
4377 txdctl |= IGB_TX_PTHRESH;
4378 txdctl |= IGB_TX_HTHRESH << 8;
4379 txdctl |= IGB_TX_WTHRESH << 16;
4380
4381 /* reinitialize tx_buffer_info */
4382 memset(ring->tx_buffer_info, 0,
4383 sizeof(struct igb_tx_buffer) * ring->count);
4384
4385 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
4386 wr32(E1000_TXDCTL(reg_idx), txdctl);
4387}
4388
4389/**
4390 * igb_configure_tx - Configure transmit Unit after Reset
4391 * @adapter: board private structure
4392 *
4393 * Configure the Tx unit of the MAC after a reset.
4394 **/
4395static void igb_configure_tx(struct igb_adapter *adapter)
4396{
4397 struct e1000_hw *hw = &adapter->hw;
4398 int i;
4399
4400 /* disable the queues */
4401 for (i = 0; i < adapter->num_tx_queues; i++)
4402 wr32(E1000_TXDCTL(adapter->tx_ring[i]->reg_idx), 0);
4403
4404 wrfl();
4405 usleep_range(10000, 20000);
4406
4407 for (i = 0; i < adapter->num_tx_queues; i++)
4408 igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
4409}
4410
4411/**
4412 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
4413 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
4414 *
4415 * Returns 0 on success, negative on failure
4416 **/
4417int igb_setup_rx_resources(struct igb_ring *rx_ring)
4418{
4419 struct igb_adapter *adapter = netdev_priv(rx_ring->netdev);
4420 struct device *dev = rx_ring->dev;
4421 int size, res;
4422
4423 /* XDP RX-queue info */
4424 if (xdp_rxq_info_is_reg(&rx_ring->xdp_rxq))
4425 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
4426 res = xdp_rxq_info_reg(&rx_ring->xdp_rxq, rx_ring->netdev,
4427 rx_ring->queue_index, 0);
4428 if (res < 0) {
4429 dev_err(dev, "Failed to register xdp_rxq index %u\n",
4430 rx_ring->queue_index);
4431 return res;
4432 }
4433
4434 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
4435
4436 rx_ring->rx_buffer_info = vmalloc(size);
4437 if (!rx_ring->rx_buffer_info)
4438 goto err;
4439
4440 /* Round up to nearest 4K */
4441 rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
4442 rx_ring->size = ALIGN(rx_ring->size, 4096);
4443
4444 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
4445 &rx_ring->dma, GFP_KERNEL);
4446 if (!rx_ring->desc)
4447 goto err;
4448
4449 rx_ring->next_to_alloc = 0;
4450 rx_ring->next_to_clean = 0;
4451 rx_ring->next_to_use = 0;
4452
4453 rx_ring->xdp_prog = adapter->xdp_prog;
4454
4455 return 0;
4456
4457err:
4458 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
4459 vfree(rx_ring->rx_buffer_info);
4460 rx_ring->rx_buffer_info = NULL;
4461 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
4462 return -ENOMEM;
4463}
4464
4465/**
4466 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
4467 * (Descriptors) for all queues
4468 * @adapter: board private structure
4469 *
4470 * Return 0 on success, negative on failure
4471 **/
4472static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
4473{
4474 struct pci_dev *pdev = adapter->pdev;
4475 int i, err = 0;
4476
4477 for (i = 0; i < adapter->num_rx_queues; i++) {
4478 err = igb_setup_rx_resources(adapter->rx_ring[i]);
4479 if (err) {
4480 dev_err(&pdev->dev,
4481 "Allocation for Rx Queue %u failed\n", i);
4482 for (i--; i >= 0; i--)
4483 igb_free_rx_resources(adapter->rx_ring[i]);
4484 break;
4485 }
4486 }
4487
4488 return err;
4489}
4490
4491/**
4492 * igb_setup_mrqc - configure the multiple receive queue control registers
4493 * @adapter: Board private structure
4494 **/
4495static void igb_setup_mrqc(struct igb_adapter *adapter)
4496{
4497 struct e1000_hw *hw = &adapter->hw;
4498 u32 mrqc, rxcsum;
4499 u32 j, num_rx_queues;
4500 u32 rss_key[10];
4501
4502 netdev_rss_key_fill(rss_key, sizeof(rss_key));
4503 for (j = 0; j < 10; j++)
4504 wr32(E1000_RSSRK(j), rss_key[j]);
4505
4506 num_rx_queues = adapter->rss_queues;
4507
4508 switch (hw->mac.type) {
4509 case e1000_82576:
4510 /* 82576 supports 2 RSS queues for SR-IOV */
4511 if (adapter->vfs_allocated_count)
4512 num_rx_queues = 2;
4513 break;
4514 default:
4515 break;
4516 }
4517
4518 if (adapter->rss_indir_tbl_init != num_rx_queues) {
4519 for (j = 0; j < IGB_RETA_SIZE; j++)
4520 adapter->rss_indir_tbl[j] =
4521 (j * num_rx_queues) / IGB_RETA_SIZE;
4522 adapter->rss_indir_tbl_init = num_rx_queues;
4523 }
4524 igb_write_rss_indir_tbl(adapter);
4525
4526 /* Disable raw packet checksumming so that RSS hash is placed in
4527 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
4528 * offloads as they are enabled by default
4529 */
4530 rxcsum = rd32(E1000_RXCSUM);
4531 rxcsum |= E1000_RXCSUM_PCSD;
4532
4533 if (adapter->hw.mac.type >= e1000_82576)
4534 /* Enable Receive Checksum Offload for SCTP */
4535 rxcsum |= E1000_RXCSUM_CRCOFL;
4536
4537 /* Don't need to set TUOFL or IPOFL, they default to 1 */
4538 wr32(E1000_RXCSUM, rxcsum);
4539
4540 /* Generate RSS hash based on packet types, TCP/UDP
4541 * port numbers and/or IPv4/v6 src and dst addresses
4542 */
4543 mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
4544 E1000_MRQC_RSS_FIELD_IPV4_TCP |
4545 E1000_MRQC_RSS_FIELD_IPV6 |
4546 E1000_MRQC_RSS_FIELD_IPV6_TCP |
4547 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
4548
4549 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
4550 mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
4551 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
4552 mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
4553
4554 /* If VMDq is enabled then we set the appropriate mode for that, else
4555 * we default to RSS so that an RSS hash is calculated per packet even
4556 * if we are only using one queue
4557 */
4558 if (adapter->vfs_allocated_count) {
4559 if (hw->mac.type > e1000_82575) {
4560 /* Set the default pool for the PF's first queue */
4561 u32 vtctl = rd32(E1000_VT_CTL);
4562
4563 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
4564 E1000_VT_CTL_DISABLE_DEF_POOL);
4565 vtctl |= adapter->vfs_allocated_count <<
4566 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
4567 wr32(E1000_VT_CTL, vtctl);
4568 }
4569 if (adapter->rss_queues > 1)
4570 mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_MQ;
4571 else
4572 mrqc |= E1000_MRQC_ENABLE_VMDQ;
4573 } else {
4574 mrqc |= E1000_MRQC_ENABLE_RSS_MQ;
4575 }
4576 igb_vmm_control(adapter);
4577
4578 wr32(E1000_MRQC, mrqc);
4579}
4580
4581/**
4582 * igb_setup_rctl - configure the receive control registers
4583 * @adapter: Board private structure
4584 **/
4585void igb_setup_rctl(struct igb_adapter *adapter)
4586{
4587 struct e1000_hw *hw = &adapter->hw;
4588 u32 rctl;
4589
4590 rctl = rd32(E1000_RCTL);
4591
4592 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
4593 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
4594
4595 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
4596 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
4597
4598 /* enable stripping of CRC. It's unlikely this will break BMC
4599 * redirection as it did with e1000. Newer features require
4600 * that the HW strips the CRC.
4601 */
4602 rctl |= E1000_RCTL_SECRC;
4603
4604 /* disable store bad packets and clear size bits. */
4605 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
4606
4607 /* enable LPE to allow for reception of jumbo frames */
4608 rctl |= E1000_RCTL_LPE;
4609
4610 /* disable queue 0 to prevent tail write w/o re-config */
4611 wr32(E1000_RXDCTL(0), 0);
4612
4613 /* Attention!!! For SR-IOV PF driver operations you must enable
4614 * queue drop for all VF and PF queues to prevent head of line blocking
4615 * if an un-trusted VF does not provide descriptors to hardware.
4616 */
4617 if (adapter->vfs_allocated_count) {
4618 /* set all queue drop enable bits */
4619 wr32(E1000_QDE, ALL_QUEUES);
4620 }
4621
4622 /* This is useful for sniffing bad packets. */
4623 if (adapter->netdev->features & NETIF_F_RXALL) {
4624 /* UPE and MPE will be handled by normal PROMISC logic
4625 * in e1000e_set_rx_mode
4626 */
4627 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
4628 E1000_RCTL_BAM | /* RX All Bcast Pkts */
4629 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
4630
4631 rctl &= ~(E1000_RCTL_DPF | /* Allow filtered pause */
4632 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
4633 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
4634 * and that breaks VLANs.
4635 */
4636 }
4637
4638 wr32(E1000_RCTL, rctl);
4639}
4640
4641static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
4642 int vfn)
4643{
4644 struct e1000_hw *hw = &adapter->hw;
4645 u32 vmolr;
4646
4647 if (size > MAX_JUMBO_FRAME_SIZE)
4648 size = MAX_JUMBO_FRAME_SIZE;
4649
4650 vmolr = rd32(E1000_VMOLR(vfn));
4651 vmolr &= ~E1000_VMOLR_RLPML_MASK;
4652 vmolr |= size | E1000_VMOLR_LPE;
4653 wr32(E1000_VMOLR(vfn), vmolr);
4654
4655 return 0;
4656}
4657
4658static inline void igb_set_vf_vlan_strip(struct igb_adapter *adapter,
4659 int vfn, bool enable)
4660{
4661 struct e1000_hw *hw = &adapter->hw;
4662 u32 val, reg;
4663
4664 if (hw->mac.type < e1000_82576)
4665 return;
4666
4667 if (hw->mac.type == e1000_i350)
4668 reg = E1000_DVMOLR(vfn);
4669 else
4670 reg = E1000_VMOLR(vfn);
4671
4672 val = rd32(reg);
4673 if (enable)
4674 val |= E1000_VMOLR_STRVLAN;
4675 else
4676 val &= ~(E1000_VMOLR_STRVLAN);
4677 wr32(reg, val);
4678}
4679
4680static inline void igb_set_vmolr(struct igb_adapter *adapter,
4681 int vfn, bool aupe)
4682{
4683 struct e1000_hw *hw = &adapter->hw;
4684 u32 vmolr;
4685
4686 /* This register exists only on 82576 and newer so if we are older then
4687 * we should exit and do nothing
4688 */
4689 if (hw->mac.type < e1000_82576)
4690 return;
4691
4692 vmolr = rd32(E1000_VMOLR(vfn));
4693 if (aupe)
4694 vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
4695 else
4696 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
4697
4698 /* clear all bits that might not be set */
4699 vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
4700
4701 if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
4702 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
4703 /* for VMDq only allow the VFs and pool 0 to accept broadcast and
4704 * multicast packets
4705 */
4706 if (vfn <= adapter->vfs_allocated_count)
4707 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
4708
4709 wr32(E1000_VMOLR(vfn), vmolr);
4710}
4711
4712/**
4713 * igb_setup_srrctl - configure the split and replication receive control
4714 * registers
4715 * @adapter: Board private structure
4716 * @ring: receive ring to be configured
4717 **/
4718void igb_setup_srrctl(struct igb_adapter *adapter, struct igb_ring *ring)
4719{
4720 struct e1000_hw *hw = &adapter->hw;
4721 int reg_idx = ring->reg_idx;
4722 u32 srrctl = 0;
4723
4724 srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
4725 if (ring_uses_large_buffer(ring))
4726 srrctl |= IGB_RXBUFFER_3072 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
4727 else
4728 srrctl |= IGB_RXBUFFER_2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
4729 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
4730 if (hw->mac.type >= e1000_82580)
4731 srrctl |= E1000_SRRCTL_TIMESTAMP;
4732 /* Only set Drop Enable if VFs allocated, or we are supporting multiple
4733 * queues and rx flow control is disabled
4734 */
4735 if (adapter->vfs_allocated_count ||
4736 (!(hw->fc.current_mode & e1000_fc_rx_pause) &&
4737 adapter->num_rx_queues > 1))
4738 srrctl |= E1000_SRRCTL_DROP_EN;
4739
4740 wr32(E1000_SRRCTL(reg_idx), srrctl);
4741}
4742
4743/**
4744 * igb_configure_rx_ring - Configure a receive ring after Reset
4745 * @adapter: board private structure
4746 * @ring: receive ring to be configured
4747 *
4748 * Configure the Rx unit of the MAC after a reset.
4749 **/
4750void igb_configure_rx_ring(struct igb_adapter *adapter,
4751 struct igb_ring *ring)
4752{
4753 struct e1000_hw *hw = &adapter->hw;
4754 union e1000_adv_rx_desc *rx_desc;
4755 u64 rdba = ring->dma;
4756 int reg_idx = ring->reg_idx;
4757 u32 rxdctl = 0;
4758
4759 xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
4760 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4761 MEM_TYPE_PAGE_SHARED, NULL));
4762
4763 /* disable the queue */
4764 wr32(E1000_RXDCTL(reg_idx), 0);
4765
4766 /* Set DMA base address registers */
4767 wr32(E1000_RDBAL(reg_idx),
4768 rdba & 0x00000000ffffffffULL);
4769 wr32(E1000_RDBAH(reg_idx), rdba >> 32);
4770 wr32(E1000_RDLEN(reg_idx),
4771 ring->count * sizeof(union e1000_adv_rx_desc));
4772
4773 /* initialize head and tail */
4774 ring->tail = adapter->io_addr + E1000_RDT(reg_idx);
4775 wr32(E1000_RDH(reg_idx), 0);
4776 writel(0, ring->tail);
4777
4778 /* set descriptor configuration */
4779 igb_setup_srrctl(adapter, ring);
4780
4781 /* set filtering for VMDQ pools */
4782 igb_set_vmolr(adapter, reg_idx & 0x7, true);
4783
4784 rxdctl |= IGB_RX_PTHRESH;
4785 rxdctl |= IGB_RX_HTHRESH << 8;
4786 rxdctl |= IGB_RX_WTHRESH << 16;
4787
4788 /* initialize rx_buffer_info */
4789 memset(ring->rx_buffer_info, 0,
4790 sizeof(struct igb_rx_buffer) * ring->count);
4791
4792 /* initialize Rx descriptor 0 */
4793 rx_desc = IGB_RX_DESC(ring, 0);
4794 rx_desc->wb.upper.length = 0;
4795
4796 /* enable receive descriptor fetching */
4797 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
4798 wr32(E1000_RXDCTL(reg_idx), rxdctl);
4799}
4800
4801static void igb_set_rx_buffer_len(struct igb_adapter *adapter,
4802 struct igb_ring *rx_ring)
4803{
4804#if (PAGE_SIZE < 8192)
4805 struct e1000_hw *hw = &adapter->hw;
4806#endif
4807
4808 /* set build_skb and buffer size flags */
4809 clear_ring_build_skb_enabled(rx_ring);
4810 clear_ring_uses_large_buffer(rx_ring);
4811
4812 if (adapter->flags & IGB_FLAG_RX_LEGACY)
4813 return;
4814
4815 set_ring_build_skb_enabled(rx_ring);
4816
4817#if (PAGE_SIZE < 8192)
4818 if (adapter->max_frame_size > IGB_MAX_FRAME_BUILD_SKB ||
4819 IGB_2K_TOO_SMALL_WITH_PADDING ||
4820 rd32(E1000_RCTL) & E1000_RCTL_SBP)
4821 set_ring_uses_large_buffer(rx_ring);
4822#endif
4823}
4824
4825/**
4826 * igb_configure_rx - Configure receive Unit after Reset
4827 * @adapter: board private structure
4828 *
4829 * Configure the Rx unit of the MAC after a reset.
4830 **/
4831static void igb_configure_rx(struct igb_adapter *adapter)
4832{
4833 int i;
4834
4835 /* set the correct pool for the PF default MAC address in entry 0 */
4836 igb_set_default_mac_filter(adapter);
4837
4838 /* Setup the HW Rx Head and Tail Descriptor Pointers and
4839 * the Base and Length of the Rx Descriptor Ring
4840 */
4841 for (i = 0; i < adapter->num_rx_queues; i++) {
4842 struct igb_ring *rx_ring = adapter->rx_ring[i];
4843
4844 igb_set_rx_buffer_len(adapter, rx_ring);
4845 igb_configure_rx_ring(adapter, rx_ring);
4846 }
4847}
4848
4849/**
4850 * igb_free_tx_resources - Free Tx Resources per Queue
4851 * @tx_ring: Tx descriptor ring for a specific queue
4852 *
4853 * Free all transmit software resources
4854 **/
4855void igb_free_tx_resources(struct igb_ring *tx_ring)
4856{
4857 igb_clean_tx_ring(tx_ring);
4858
4859 vfree(tx_ring->tx_buffer_info);
4860 tx_ring->tx_buffer_info = NULL;
4861
4862 /* if not set, then don't free */
4863 if (!tx_ring->desc)
4864 return;
4865
4866 dma_free_coherent(tx_ring->dev, tx_ring->size,
4867 tx_ring->desc, tx_ring->dma);
4868
4869 tx_ring->desc = NULL;
4870}
4871
4872/**
4873 * igb_free_all_tx_resources - Free Tx Resources for All Queues
4874 * @adapter: board private structure
4875 *
4876 * Free all transmit software resources
4877 **/
4878static void igb_free_all_tx_resources(struct igb_adapter *adapter)
4879{
4880 int i;
4881
4882 for (i = 0; i < adapter->num_tx_queues; i++)
4883 if (adapter->tx_ring[i])
4884 igb_free_tx_resources(adapter->tx_ring[i]);
4885}
4886
4887/**
4888 * igb_clean_tx_ring - Free Tx Buffers
4889 * @tx_ring: ring to be cleaned
4890 **/
4891static void igb_clean_tx_ring(struct igb_ring *tx_ring)
4892{
4893 u16 i = tx_ring->next_to_clean;
4894 struct igb_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
4895
4896 while (i != tx_ring->next_to_use) {
4897 union e1000_adv_tx_desc *eop_desc, *tx_desc;
4898
4899 /* Free all the Tx ring sk_buffs or xdp frames */
4900 if (tx_buffer->type == IGB_TYPE_SKB)
4901 dev_kfree_skb_any(tx_buffer->skb);
4902 else
4903 xdp_return_frame(tx_buffer->xdpf);
4904
4905 /* unmap skb header data */
4906 dma_unmap_single(tx_ring->dev,
4907 dma_unmap_addr(tx_buffer, dma),
4908 dma_unmap_len(tx_buffer, len),
4909 DMA_TO_DEVICE);
4910
4911 /* check for eop_desc to determine the end of the packet */
4912 eop_desc = tx_buffer->next_to_watch;
4913 tx_desc = IGB_TX_DESC(tx_ring, i);
4914
4915 /* unmap remaining buffers */
4916 while (tx_desc != eop_desc) {
4917 tx_buffer++;
4918 tx_desc++;
4919 i++;
4920 if (unlikely(i == tx_ring->count)) {
4921 i = 0;
4922 tx_buffer = tx_ring->tx_buffer_info;
4923 tx_desc = IGB_TX_DESC(tx_ring, 0);
4924 }
4925
4926 /* unmap any remaining paged data */
4927 if (dma_unmap_len(tx_buffer, len))
4928 dma_unmap_page(tx_ring->dev,
4929 dma_unmap_addr(tx_buffer, dma),
4930 dma_unmap_len(tx_buffer, len),
4931 DMA_TO_DEVICE);
4932 }
4933
4934 tx_buffer->next_to_watch = NULL;
4935
4936 /* move us one more past the eop_desc for start of next pkt */
4937 tx_buffer++;
4938 i++;
4939 if (unlikely(i == tx_ring->count)) {
4940 i = 0;
4941 tx_buffer = tx_ring->tx_buffer_info;
4942 }
4943 }
4944
4945 /* reset BQL for queue */
4946 netdev_tx_reset_queue(txring_txq(tx_ring));
4947
4948 /* reset next_to_use and next_to_clean */
4949 tx_ring->next_to_use = 0;
4950 tx_ring->next_to_clean = 0;
4951}
4952
4953/**
4954 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
4955 * @adapter: board private structure
4956 **/
4957static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
4958{
4959 int i;
4960
4961 for (i = 0; i < adapter->num_tx_queues; i++)
4962 if (adapter->tx_ring[i])
4963 igb_clean_tx_ring(adapter->tx_ring[i]);
4964}
4965
4966/**
4967 * igb_free_rx_resources - Free Rx Resources
4968 * @rx_ring: ring to clean the resources from
4969 *
4970 * Free all receive software resources
4971 **/
4972void igb_free_rx_resources(struct igb_ring *rx_ring)
4973{
4974 igb_clean_rx_ring(rx_ring);
4975
4976 rx_ring->xdp_prog = NULL;
4977 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
4978 vfree(rx_ring->rx_buffer_info);
4979 rx_ring->rx_buffer_info = NULL;
4980
4981 /* if not set, then don't free */
4982 if (!rx_ring->desc)
4983 return;
4984
4985 dma_free_coherent(rx_ring->dev, rx_ring->size,
4986 rx_ring->desc, rx_ring->dma);
4987
4988 rx_ring->desc = NULL;
4989}
4990
4991/**
4992 * igb_free_all_rx_resources - Free Rx Resources for All Queues
4993 * @adapter: board private structure
4994 *
4995 * Free all receive software resources
4996 **/
4997static void igb_free_all_rx_resources(struct igb_adapter *adapter)
4998{
4999 int i;
5000
5001 for (i = 0; i < adapter->num_rx_queues; i++)
5002 if (adapter->rx_ring[i])
5003 igb_free_rx_resources(adapter->rx_ring[i]);
5004}
5005
5006/**
5007 * igb_clean_rx_ring - Free Rx Buffers per Queue
5008 * @rx_ring: ring to free buffers from
5009 **/
5010static void igb_clean_rx_ring(struct igb_ring *rx_ring)
5011{
5012 u16 i = rx_ring->next_to_clean;
5013
5014 dev_kfree_skb(rx_ring->skb);
5015 rx_ring->skb = NULL;
5016
5017 /* Free all the Rx ring sk_buffs */
5018 while (i != rx_ring->next_to_alloc) {
5019 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
5020
5021 /* Invalidate cache lines that may have been written to by
5022 * device so that we avoid corrupting memory.
5023 */
5024 dma_sync_single_range_for_cpu(rx_ring->dev,
5025 buffer_info->dma,
5026 buffer_info->page_offset,
5027 igb_rx_bufsz(rx_ring),
5028 DMA_FROM_DEVICE);
5029
5030 /* free resources associated with mapping */
5031 dma_unmap_page_attrs(rx_ring->dev,
5032 buffer_info->dma,
5033 igb_rx_pg_size(rx_ring),
5034 DMA_FROM_DEVICE,
5035 IGB_RX_DMA_ATTR);
5036 __page_frag_cache_drain(buffer_info->page,
5037 buffer_info->pagecnt_bias);
5038
5039 i++;
5040 if (i == rx_ring->count)
5041 i = 0;
5042 }
5043
5044 rx_ring->next_to_alloc = 0;
5045 rx_ring->next_to_clean = 0;
5046 rx_ring->next_to_use = 0;
5047}
5048
5049/**
5050 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
5051 * @adapter: board private structure
5052 **/
5053static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
5054{
5055 int i;
5056
5057 for (i = 0; i < adapter->num_rx_queues; i++)
5058 if (adapter->rx_ring[i])
5059 igb_clean_rx_ring(adapter->rx_ring[i]);
5060}
5061
5062/**
5063 * igb_set_mac - Change the Ethernet Address of the NIC
5064 * @netdev: network interface device structure
5065 * @p: pointer to an address structure
5066 *
5067 * Returns 0 on success, negative on failure
5068 **/
5069static int igb_set_mac(struct net_device *netdev, void *p)
5070{
5071 struct igb_adapter *adapter = netdev_priv(netdev);
5072 struct e1000_hw *hw = &adapter->hw;
5073 struct sockaddr *addr = p;
5074
5075 if (!is_valid_ether_addr(addr->sa_data))
5076 return -EADDRNOTAVAIL;
5077
5078 eth_hw_addr_set(netdev, addr->sa_data);
5079 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
5080
5081 /* set the correct pool for the new PF MAC address in entry 0 */
5082 igb_set_default_mac_filter(adapter);
5083
5084 return 0;
5085}
5086
5087/**
5088 * igb_write_mc_addr_list - write multicast addresses to MTA
5089 * @netdev: network interface device structure
5090 *
5091 * Writes multicast address list to the MTA hash table.
5092 * Returns: -ENOMEM on failure
5093 * 0 on no addresses written
5094 * X on writing X addresses to MTA
5095 **/
5096static int igb_write_mc_addr_list(struct net_device *netdev)
5097{
5098 struct igb_adapter *adapter = netdev_priv(netdev);
5099 struct e1000_hw *hw = &adapter->hw;
5100 struct netdev_hw_addr *ha;
5101 u8 *mta_list;
5102 int i;
5103
5104 if (netdev_mc_empty(netdev)) {
5105 /* nothing to program, so clear mc list */
5106 igb_update_mc_addr_list(hw, NULL, 0);
5107 igb_restore_vf_multicasts(adapter);
5108 return 0;
5109 }
5110
5111 mta_list = kcalloc(netdev_mc_count(netdev), 6, GFP_ATOMIC);
5112 if (!mta_list)
5113 return -ENOMEM;
5114
5115 /* The shared function expects a packed array of only addresses. */
5116 i = 0;
5117 netdev_for_each_mc_addr(ha, netdev)
5118 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
5119
5120 igb_update_mc_addr_list(hw, mta_list, i);
5121 kfree(mta_list);
5122
5123 return netdev_mc_count(netdev);
5124}
5125
5126static int igb_vlan_promisc_enable(struct igb_adapter *adapter)
5127{
5128 struct e1000_hw *hw = &adapter->hw;
5129 u32 i, pf_id;
5130
5131 switch (hw->mac.type) {
5132 case e1000_i210:
5133 case e1000_i211:
5134 case e1000_i350:
5135 /* VLAN filtering needed for VLAN prio filter */
5136 if (adapter->netdev->features & NETIF_F_NTUPLE)
5137 break;
5138 fallthrough;
5139 case e1000_82576:
5140 case e1000_82580:
5141 case e1000_i354:
5142 /* VLAN filtering needed for pool filtering */
5143 if (adapter->vfs_allocated_count)
5144 break;
5145 fallthrough;
5146 default:
5147 return 1;
5148 }
5149
5150 /* We are already in VLAN promisc, nothing to do */
5151 if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
5152 return 0;
5153
5154 if (!adapter->vfs_allocated_count)
5155 goto set_vfta;
5156
5157 /* Add PF to all active pools */
5158 pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
5159
5160 for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
5161 u32 vlvf = rd32(E1000_VLVF(i));
5162
5163 vlvf |= BIT(pf_id);
5164 wr32(E1000_VLVF(i), vlvf);
5165 }
5166
5167set_vfta:
5168 /* Set all bits in the VLAN filter table array */
5169 for (i = E1000_VLAN_FILTER_TBL_SIZE; i--;)
5170 hw->mac.ops.write_vfta(hw, i, ~0U);
5171
5172 /* Set flag so we don't redo unnecessary work */
5173 adapter->flags |= IGB_FLAG_VLAN_PROMISC;
5174
5175 return 0;
5176}
5177
5178#define VFTA_BLOCK_SIZE 8
5179static void igb_scrub_vfta(struct igb_adapter *adapter, u32 vfta_offset)
5180{
5181 struct e1000_hw *hw = &adapter->hw;
5182 u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
5183 u32 vid_start = vfta_offset * 32;
5184 u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
5185 u32 i, vid, word, bits, pf_id;
5186
5187 /* guarantee that we don't scrub out management VLAN */
5188 vid = adapter->mng_vlan_id;
5189 if (vid >= vid_start && vid < vid_end)
5190 vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
5191
5192 if (!adapter->vfs_allocated_count)
5193 goto set_vfta;
5194
5195 pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
5196
5197 for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
5198 u32 vlvf = rd32(E1000_VLVF(i));
5199
5200 /* pull VLAN ID from VLVF */
5201 vid = vlvf & VLAN_VID_MASK;
5202
5203 /* only concern ourselves with a certain range */
5204 if (vid < vid_start || vid >= vid_end)
5205 continue;
5206
5207 if (vlvf & E1000_VLVF_VLANID_ENABLE) {
5208 /* record VLAN ID in VFTA */
5209 vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
5210
5211 /* if PF is part of this then continue */
5212 if (test_bit(vid, adapter->active_vlans))
5213 continue;
5214 }
5215
5216 /* remove PF from the pool */
5217 bits = ~BIT(pf_id);
5218 bits &= rd32(E1000_VLVF(i));
5219 wr32(E1000_VLVF(i), bits);
5220 }
5221
5222set_vfta:
5223 /* extract values from active_vlans and write back to VFTA */
5224 for (i = VFTA_BLOCK_SIZE; i--;) {
5225 vid = (vfta_offset + i) * 32;
5226 word = vid / BITS_PER_LONG;
5227 bits = vid % BITS_PER_LONG;
5228
5229 vfta[i] |= adapter->active_vlans[word] >> bits;
5230
5231 hw->mac.ops.write_vfta(hw, vfta_offset + i, vfta[i]);
5232 }
5233}
5234
5235static void igb_vlan_promisc_disable(struct igb_adapter *adapter)
5236{
5237 u32 i;
5238
5239 /* We are not in VLAN promisc, nothing to do */
5240 if (!(adapter->flags & IGB_FLAG_VLAN_PROMISC))
5241 return;
5242
5243 /* Set flag so we don't redo unnecessary work */
5244 adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
5245
5246 for (i = 0; i < E1000_VLAN_FILTER_TBL_SIZE; i += VFTA_BLOCK_SIZE)
5247 igb_scrub_vfta(adapter, i);
5248}
5249
5250/**
5251 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
5252 * @netdev: network interface device structure
5253 *
5254 * The set_rx_mode entry point is called whenever the unicast or multicast
5255 * address lists or the network interface flags are updated. This routine is
5256 * responsible for configuring the hardware for proper unicast, multicast,
5257 * promiscuous mode, and all-multi behavior.
5258 **/
5259static void igb_set_rx_mode(struct net_device *netdev)
5260{
5261 struct igb_adapter *adapter = netdev_priv(netdev);
5262 struct e1000_hw *hw = &adapter->hw;
5263 unsigned int vfn = adapter->vfs_allocated_count;
5264 u32 rctl = 0, vmolr = 0, rlpml = MAX_JUMBO_FRAME_SIZE;
5265 int count;
5266
5267 /* Check for Promiscuous and All Multicast modes */
5268 if (netdev->flags & IFF_PROMISC) {
5269 rctl |= E1000_RCTL_UPE | E1000_RCTL_MPE;
5270 vmolr |= E1000_VMOLR_MPME;
5271
5272 /* enable use of UTA filter to force packets to default pool */
5273 if (hw->mac.type == e1000_82576)
5274 vmolr |= E1000_VMOLR_ROPE;
5275 } else {
5276 if (netdev->flags & IFF_ALLMULTI) {
5277 rctl |= E1000_RCTL_MPE;
5278 vmolr |= E1000_VMOLR_MPME;
5279 } else {
5280 /* Write addresses to the MTA, if the attempt fails
5281 * then we should just turn on promiscuous mode so
5282 * that we can at least receive multicast traffic
5283 */
5284 count = igb_write_mc_addr_list(netdev);
5285 if (count < 0) {
5286 rctl |= E1000_RCTL_MPE;
5287 vmolr |= E1000_VMOLR_MPME;
5288 } else if (count) {
5289 vmolr |= E1000_VMOLR_ROMPE;
5290 }
5291 }
5292 }
5293
5294 /* Write addresses to available RAR registers, if there is not
5295 * sufficient space to store all the addresses then enable
5296 * unicast promiscuous mode
5297 */
5298 if (__dev_uc_sync(netdev, igb_uc_sync, igb_uc_unsync)) {
5299 rctl |= E1000_RCTL_UPE;
5300 vmolr |= E1000_VMOLR_ROPE;
5301 }
5302
5303 /* enable VLAN filtering by default */
5304 rctl |= E1000_RCTL_VFE;
5305
5306 /* disable VLAN filtering for modes that require it */
5307 if ((netdev->flags & IFF_PROMISC) ||
5308 (netdev->features & NETIF_F_RXALL)) {
5309 /* if we fail to set all rules then just clear VFE */
5310 if (igb_vlan_promisc_enable(adapter))
5311 rctl &= ~E1000_RCTL_VFE;
5312 } else {
5313 igb_vlan_promisc_disable(adapter);
5314 }
5315
5316 /* update state of unicast, multicast, and VLAN filtering modes */
5317 rctl |= rd32(E1000_RCTL) & ~(E1000_RCTL_UPE | E1000_RCTL_MPE |
5318 E1000_RCTL_VFE);
5319 wr32(E1000_RCTL, rctl);
5320
5321#if (PAGE_SIZE < 8192)
5322 if (!adapter->vfs_allocated_count) {
5323 if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
5324 rlpml = IGB_MAX_FRAME_BUILD_SKB;
5325 }
5326#endif
5327 wr32(E1000_RLPML, rlpml);
5328
5329 /* In order to support SR-IOV and eventually VMDq it is necessary to set
5330 * the VMOLR to enable the appropriate modes. Without this workaround
5331 * we will have issues with VLAN tag stripping not being done for frames
5332 * that are only arriving because we are the default pool
5333 */
5334 if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
5335 return;
5336
5337 /* set UTA to appropriate mode */
5338 igb_set_uta(adapter, !!(vmolr & E1000_VMOLR_ROPE));
5339
5340 vmolr |= rd32(E1000_VMOLR(vfn)) &
5341 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
5342
5343 /* enable Rx jumbo frames, restrict as needed to support build_skb */
5344 vmolr &= ~E1000_VMOLR_RLPML_MASK;
5345#if (PAGE_SIZE < 8192)
5346 if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
5347 vmolr |= IGB_MAX_FRAME_BUILD_SKB;
5348 else
5349#endif
5350 vmolr |= MAX_JUMBO_FRAME_SIZE;
5351 vmolr |= E1000_VMOLR_LPE;
5352
5353 wr32(E1000_VMOLR(vfn), vmolr);
5354
5355 igb_restore_vf_multicasts(adapter);
5356}
5357
5358static void igb_check_wvbr(struct igb_adapter *adapter)
5359{
5360 struct e1000_hw *hw = &adapter->hw;
5361 u32 wvbr = 0;
5362
5363 switch (hw->mac.type) {
5364 case e1000_82576:
5365 case e1000_i350:
5366 wvbr = rd32(E1000_WVBR);
5367 if (!wvbr)
5368 return;
5369 break;
5370 default:
5371 break;
5372 }
5373
5374 adapter->wvbr |= wvbr;
5375}
5376
5377#define IGB_STAGGERED_QUEUE_OFFSET 8
5378
5379static void igb_spoof_check(struct igb_adapter *adapter)
5380{
5381 int j;
5382
5383 if (!adapter->wvbr)
5384 return;
5385
5386 for (j = 0; j < adapter->vfs_allocated_count; j++) {
5387 if (adapter->wvbr & BIT(j) ||
5388 adapter->wvbr & BIT(j + IGB_STAGGERED_QUEUE_OFFSET)) {
5389 dev_warn(&adapter->pdev->dev,
5390 "Spoof event(s) detected on VF %d\n", j);
5391 adapter->wvbr &=
5392 ~(BIT(j) |
5393 BIT(j + IGB_STAGGERED_QUEUE_OFFSET));
5394 }
5395 }
5396}
5397
5398/* Need to wait a few seconds after link up to get diagnostic information from
5399 * the phy
5400 */
5401static void igb_update_phy_info(struct timer_list *t)
5402{
5403 struct igb_adapter *adapter = from_timer(adapter, t, phy_info_timer);
5404 igb_get_phy_info(&adapter->hw);
5405}
5406
5407/**
5408 * igb_has_link - check shared code for link and determine up/down
5409 * @adapter: pointer to driver private info
5410 **/
5411bool igb_has_link(struct igb_adapter *adapter)
5412{
5413 struct e1000_hw *hw = &adapter->hw;
5414 bool link_active = false;
5415
5416 /* get_link_status is set on LSC (link status) interrupt or
5417 * rx sequence error interrupt. get_link_status will stay
5418 * false until the e1000_check_for_link establishes link
5419 * for copper adapters ONLY
5420 */
5421 switch (hw->phy.media_type) {
5422 case e1000_media_type_copper:
5423 if (!hw->mac.get_link_status)
5424 return true;
5425 fallthrough;
5426 case e1000_media_type_internal_serdes:
5427 hw->mac.ops.check_for_link(hw);
5428 link_active = !hw->mac.get_link_status;
5429 break;
5430 default:
5431 case e1000_media_type_unknown:
5432 break;
5433 }
5434
5435 if (((hw->mac.type == e1000_i210) ||
5436 (hw->mac.type == e1000_i211)) &&
5437 (hw->phy.id == I210_I_PHY_ID)) {
5438 if (!netif_carrier_ok(adapter->netdev)) {
5439 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
5440 } else if (!(adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)) {
5441 adapter->flags |= IGB_FLAG_NEED_LINK_UPDATE;
5442 adapter->link_check_timeout = jiffies;
5443 }
5444 }
5445
5446 return link_active;
5447}
5448
5449static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
5450{
5451 bool ret = false;
5452 u32 ctrl_ext, thstat;
5453
5454 /* check for thermal sensor event on i350 copper only */
5455 if (hw->mac.type == e1000_i350) {
5456 thstat = rd32(E1000_THSTAT);
5457 ctrl_ext = rd32(E1000_CTRL_EXT);
5458
5459 if ((hw->phy.media_type == e1000_media_type_copper) &&
5460 !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII))
5461 ret = !!(thstat & event);
5462 }
5463
5464 return ret;
5465}
5466
5467/**
5468 * igb_check_lvmmc - check for malformed packets received
5469 * and indicated in LVMMC register
5470 * @adapter: pointer to adapter
5471 **/
5472static void igb_check_lvmmc(struct igb_adapter *adapter)
5473{
5474 struct e1000_hw *hw = &adapter->hw;
5475 u32 lvmmc;
5476
5477 lvmmc = rd32(E1000_LVMMC);
5478 if (lvmmc) {
5479 if (unlikely(net_ratelimit())) {
5480 netdev_warn(adapter->netdev,
5481 "malformed Tx packet detected and dropped, LVMMC:0x%08x\n",
5482 lvmmc);
5483 }
5484 }
5485}
5486
5487/**
5488 * igb_watchdog - Timer Call-back
5489 * @t: pointer to timer_list containing our private info pointer
5490 **/
5491static void igb_watchdog(struct timer_list *t)
5492{
5493 struct igb_adapter *adapter = from_timer(adapter, t, watchdog_timer);
5494 /* Do the rest outside of interrupt context */
5495 schedule_work(&adapter->watchdog_task);
5496}
5497
5498static void igb_watchdog_task(struct work_struct *work)
5499{
5500 struct igb_adapter *adapter = container_of(work,
5501 struct igb_adapter,
5502 watchdog_task);
5503 struct e1000_hw *hw = &adapter->hw;
5504 struct e1000_phy_info *phy = &hw->phy;
5505 struct net_device *netdev = adapter->netdev;
5506 u32 link;
5507 int i;
5508 u32 connsw;
5509 u16 phy_data, retry_count = 20;
5510
5511 link = igb_has_link(adapter);
5512
5513 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) {
5514 if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
5515 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
5516 else
5517 link = false;
5518 }
5519
5520 /* Force link down if we have fiber to swap to */
5521 if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
5522 if (hw->phy.media_type == e1000_media_type_copper) {
5523 connsw = rd32(E1000_CONNSW);
5524 if (!(connsw & E1000_CONNSW_AUTOSENSE_EN))
5525 link = 0;
5526 }
5527 }
5528 if (link) {
5529 /* Perform a reset if the media type changed. */
5530 if (hw->dev_spec._82575.media_changed) {
5531 hw->dev_spec._82575.media_changed = false;
5532 adapter->flags |= IGB_FLAG_MEDIA_RESET;
5533 igb_reset(adapter);
5534 }
5535 /* Cancel scheduled suspend requests. */
5536 pm_runtime_resume(netdev->dev.parent);
5537
5538 if (!netif_carrier_ok(netdev)) {
5539 u32 ctrl;
5540
5541 hw->mac.ops.get_speed_and_duplex(hw,
5542 &adapter->link_speed,
5543 &adapter->link_duplex);
5544
5545 ctrl = rd32(E1000_CTRL);
5546 /* Links status message must follow this format */
5547 netdev_info(netdev,
5548 "igb: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
5549 netdev->name,
5550 adapter->link_speed,
5551 adapter->link_duplex == FULL_DUPLEX ?
5552 "Full" : "Half",
5553 (ctrl & E1000_CTRL_TFCE) &&
5554 (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
5555 (ctrl & E1000_CTRL_RFCE) ? "RX" :
5556 (ctrl & E1000_CTRL_TFCE) ? "TX" : "None");
5557
5558 /* disable EEE if enabled */
5559 if ((adapter->flags & IGB_FLAG_EEE) &&
5560 (adapter->link_duplex == HALF_DUPLEX)) {
5561 dev_info(&adapter->pdev->dev,
5562 "EEE Disabled: unsupported at half duplex. Re-enable using ethtool when at full duplex.\n");
5563 adapter->hw.dev_spec._82575.eee_disable = true;
5564 adapter->flags &= ~IGB_FLAG_EEE;
5565 }
5566
5567 /* check if SmartSpeed worked */
5568 igb_check_downshift(hw);
5569 if (phy->speed_downgraded)
5570 netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
5571
5572 /* check for thermal sensor event */
5573 if (igb_thermal_sensor_event(hw,
5574 E1000_THSTAT_LINK_THROTTLE))
5575 netdev_info(netdev, "The network adapter link speed was downshifted because it overheated\n");
5576
5577 /* adjust timeout factor according to speed/duplex */
5578 adapter->tx_timeout_factor = 1;
5579 switch (adapter->link_speed) {
5580 case SPEED_10:
5581 adapter->tx_timeout_factor = 14;
5582 break;
5583 case SPEED_100:
5584 /* maybe add some timeout factor ? */
5585 break;
5586 }
5587
5588 if (adapter->link_speed != SPEED_1000 ||
5589 !hw->phy.ops.read_reg)
5590 goto no_wait;
5591
5592 /* wait for Remote receiver status OK */
5593retry_read_status:
5594 if (!igb_read_phy_reg(hw, PHY_1000T_STATUS,
5595 &phy_data)) {
5596 if (!(phy_data & SR_1000T_REMOTE_RX_STATUS) &&
5597 retry_count) {
5598 msleep(100);
5599 retry_count--;
5600 goto retry_read_status;
5601 } else if (!retry_count) {
5602 dev_err(&adapter->pdev->dev, "exceed max 2 second\n");
5603 }
5604 } else {
5605 dev_err(&adapter->pdev->dev, "read 1000Base-T Status Reg\n");
5606 }
5607no_wait:
5608 netif_carrier_on(netdev);
5609
5610 igb_ping_all_vfs(adapter);
5611 igb_check_vf_rate_limit(adapter);
5612
5613 /* link state has changed, schedule phy info update */
5614 if (!test_bit(__IGB_DOWN, &adapter->state))
5615 mod_timer(&adapter->phy_info_timer,
5616 round_jiffies(jiffies + 2 * HZ));
5617 }
5618 } else {
5619 if (netif_carrier_ok(netdev)) {
5620 adapter->link_speed = 0;
5621 adapter->link_duplex = 0;
5622
5623 /* check for thermal sensor event */
5624 if (igb_thermal_sensor_event(hw,
5625 E1000_THSTAT_PWR_DOWN)) {
5626 netdev_err(netdev, "The network adapter was stopped because it overheated\n");
5627 }
5628
5629 /* Links status message must follow this format */
5630 netdev_info(netdev, "igb: %s NIC Link is Down\n",
5631 netdev->name);
5632 netif_carrier_off(netdev);
5633
5634 igb_ping_all_vfs(adapter);
5635
5636 /* link state has changed, schedule phy info update */
5637 if (!test_bit(__IGB_DOWN, &adapter->state))
5638 mod_timer(&adapter->phy_info_timer,
5639 round_jiffies(jiffies + 2 * HZ));
5640
5641 /* link is down, time to check for alternate media */
5642 if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
5643 igb_check_swap_media(adapter);
5644 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
5645 schedule_work(&adapter->reset_task);
5646 /* return immediately */
5647 return;
5648 }
5649 }
5650 pm_schedule_suspend(netdev->dev.parent,
5651 MSEC_PER_SEC * 5);
5652
5653 /* also check for alternate media here */
5654 } else if (!netif_carrier_ok(netdev) &&
5655 (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
5656 igb_check_swap_media(adapter);
5657 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
5658 schedule_work(&adapter->reset_task);
5659 /* return immediately */
5660 return;
5661 }
5662 }
5663 }
5664
5665 spin_lock(&adapter->stats64_lock);
5666 igb_update_stats(adapter);
5667 spin_unlock(&adapter->stats64_lock);
5668
5669 for (i = 0; i < adapter->num_tx_queues; i++) {
5670 struct igb_ring *tx_ring = adapter->tx_ring[i];
5671 if (!netif_carrier_ok(netdev)) {
5672 /* We've lost link, so the controller stops DMA,
5673 * but we've got queued Tx work that's never going
5674 * to get done, so reset controller to flush Tx.
5675 * (Do the reset outside of interrupt context).
5676 */
5677 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
5678 adapter->tx_timeout_count++;
5679 schedule_work(&adapter->reset_task);
5680 /* return immediately since reset is imminent */
5681 return;
5682 }
5683 }
5684
5685 /* Force detection of hung controller every watchdog period */
5686 set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
5687 }
5688
5689 /* Cause software interrupt to ensure Rx ring is cleaned */
5690 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
5691 u32 eics = 0;
5692
5693 for (i = 0; i < adapter->num_q_vectors; i++)
5694 eics |= adapter->q_vector[i]->eims_value;
5695 wr32(E1000_EICS, eics);
5696 } else {
5697 wr32(E1000_ICS, E1000_ICS_RXDMT0);
5698 }
5699
5700 igb_spoof_check(adapter);
5701 igb_ptp_rx_hang(adapter);
5702 igb_ptp_tx_hang(adapter);
5703
5704 /* Check LVMMC register on i350/i354 only */
5705 if ((adapter->hw.mac.type == e1000_i350) ||
5706 (adapter->hw.mac.type == e1000_i354))
5707 igb_check_lvmmc(adapter);
5708
5709 /* Reset the timer */
5710 if (!test_bit(__IGB_DOWN, &adapter->state)) {
5711 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)
5712 mod_timer(&adapter->watchdog_timer,
5713 round_jiffies(jiffies + HZ));
5714 else
5715 mod_timer(&adapter->watchdog_timer,
5716 round_jiffies(jiffies + 2 * HZ));
5717 }
5718}
5719
5720enum latency_range {
5721 lowest_latency = 0,
5722 low_latency = 1,
5723 bulk_latency = 2,
5724 latency_invalid = 255
5725};
5726
5727/**
5728 * igb_update_ring_itr - update the dynamic ITR value based on packet size
5729 * @q_vector: pointer to q_vector
5730 *
5731 * Stores a new ITR value based on strictly on packet size. This
5732 * algorithm is less sophisticated than that used in igb_update_itr,
5733 * due to the difficulty of synchronizing statistics across multiple
5734 * receive rings. The divisors and thresholds used by this function
5735 * were determined based on theoretical maximum wire speed and testing
5736 * data, in order to minimize response time while increasing bulk
5737 * throughput.
5738 * This functionality is controlled by ethtool's coalescing settings.
5739 * NOTE: This function is called only when operating in a multiqueue
5740 * receive environment.
5741 **/
5742static void igb_update_ring_itr(struct igb_q_vector *q_vector)
5743{
5744 int new_val = q_vector->itr_val;
5745 int avg_wire_size = 0;
5746 struct igb_adapter *adapter = q_vector->adapter;
5747 unsigned int packets;
5748
5749 /* For non-gigabit speeds, just fix the interrupt rate at 4000
5750 * ints/sec - ITR timer value of 120 ticks.
5751 */
5752 if (adapter->link_speed != SPEED_1000) {
5753 new_val = IGB_4K_ITR;
5754 goto set_itr_val;
5755 }
5756
5757 packets = q_vector->rx.total_packets;
5758 if (packets)
5759 avg_wire_size = q_vector->rx.total_bytes / packets;
5760
5761 packets = q_vector->tx.total_packets;
5762 if (packets)
5763 avg_wire_size = max_t(u32, avg_wire_size,
5764 q_vector->tx.total_bytes / packets);
5765
5766 /* if avg_wire_size isn't set no work was done */
5767 if (!avg_wire_size)
5768 goto clear_counts;
5769
5770 /* Add 24 bytes to size to account for CRC, preamble, and gap */
5771 avg_wire_size += 24;
5772
5773 /* Don't starve jumbo frames */
5774 avg_wire_size = min(avg_wire_size, 3000);
5775
5776 /* Give a little boost to mid-size frames */
5777 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
5778 new_val = avg_wire_size / 3;
5779 else
5780 new_val = avg_wire_size / 2;
5781
5782 /* conservative mode (itr 3) eliminates the lowest_latency setting */
5783 if (new_val < IGB_20K_ITR &&
5784 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
5785 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
5786 new_val = IGB_20K_ITR;
5787
5788set_itr_val:
5789 if (new_val != q_vector->itr_val) {
5790 q_vector->itr_val = new_val;
5791 q_vector->set_itr = 1;
5792 }
5793clear_counts:
5794 q_vector->rx.total_bytes = 0;
5795 q_vector->rx.total_packets = 0;
5796 q_vector->tx.total_bytes = 0;
5797 q_vector->tx.total_packets = 0;
5798}
5799
5800/**
5801 * igb_update_itr - update the dynamic ITR value based on statistics
5802 * @q_vector: pointer to q_vector
5803 * @ring_container: ring info to update the itr for
5804 *
5805 * Stores a new ITR value based on packets and byte
5806 * counts during the last interrupt. The advantage of per interrupt
5807 * computation is faster updates and more accurate ITR for the current
5808 * traffic pattern. Constants in this function were computed
5809 * based on theoretical maximum wire speed and thresholds were set based
5810 * on testing data as well as attempting to minimize response time
5811 * while increasing bulk throughput.
5812 * This functionality is controlled by ethtool's coalescing settings.
5813 * NOTE: These calculations are only valid when operating in a single-
5814 * queue environment.
5815 **/
5816static void igb_update_itr(struct igb_q_vector *q_vector,
5817 struct igb_ring_container *ring_container)
5818{
5819 unsigned int packets = ring_container->total_packets;
5820 unsigned int bytes = ring_container->total_bytes;
5821 u8 itrval = ring_container->itr;
5822
5823 /* no packets, exit with status unchanged */
5824 if (packets == 0)
5825 return;
5826
5827 switch (itrval) {
5828 case lowest_latency:
5829 /* handle TSO and jumbo frames */
5830 if (bytes/packets > 8000)
5831 itrval = bulk_latency;
5832 else if ((packets < 5) && (bytes > 512))
5833 itrval = low_latency;
5834 break;
5835 case low_latency: /* 50 usec aka 20000 ints/s */
5836 if (bytes > 10000) {
5837 /* this if handles the TSO accounting */
5838 if (bytes/packets > 8000)
5839 itrval = bulk_latency;
5840 else if ((packets < 10) || ((bytes/packets) > 1200))
5841 itrval = bulk_latency;
5842 else if ((packets > 35))
5843 itrval = lowest_latency;
5844 } else if (bytes/packets > 2000) {
5845 itrval = bulk_latency;
5846 } else if (packets <= 2 && bytes < 512) {
5847 itrval = lowest_latency;
5848 }
5849 break;
5850 case bulk_latency: /* 250 usec aka 4000 ints/s */
5851 if (bytes > 25000) {
5852 if (packets > 35)
5853 itrval = low_latency;
5854 } else if (bytes < 1500) {
5855 itrval = low_latency;
5856 }
5857 break;
5858 }
5859
5860 /* clear work counters since we have the values we need */
5861 ring_container->total_bytes = 0;
5862 ring_container->total_packets = 0;
5863
5864 /* write updated itr to ring container */
5865 ring_container->itr = itrval;
5866}
5867
5868static void igb_set_itr(struct igb_q_vector *q_vector)
5869{
5870 struct igb_adapter *adapter = q_vector->adapter;
5871 u32 new_itr = q_vector->itr_val;
5872 u8 current_itr = 0;
5873
5874 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
5875 if (adapter->link_speed != SPEED_1000) {
5876 current_itr = 0;
5877 new_itr = IGB_4K_ITR;
5878 goto set_itr_now;
5879 }
5880
5881 igb_update_itr(q_vector, &q_vector->tx);
5882 igb_update_itr(q_vector, &q_vector->rx);
5883
5884 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
5885
5886 /* conservative mode (itr 3) eliminates the lowest_latency setting */
5887 if (current_itr == lowest_latency &&
5888 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
5889 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
5890 current_itr = low_latency;
5891
5892 switch (current_itr) {
5893 /* counts and packets in update_itr are dependent on these numbers */
5894 case lowest_latency:
5895 new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
5896 break;
5897 case low_latency:
5898 new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
5899 break;
5900 case bulk_latency:
5901 new_itr = IGB_4K_ITR; /* 4,000 ints/sec */
5902 break;
5903 default:
5904 break;
5905 }
5906
5907set_itr_now:
5908 if (new_itr != q_vector->itr_val) {
5909 /* this attempts to bias the interrupt rate towards Bulk
5910 * by adding intermediate steps when interrupt rate is
5911 * increasing
5912 */
5913 new_itr = new_itr > q_vector->itr_val ?
5914 max((new_itr * q_vector->itr_val) /
5915 (new_itr + (q_vector->itr_val >> 2)),
5916 new_itr) : new_itr;
5917 /* Don't write the value here; it resets the adapter's
5918 * internal timer, and causes us to delay far longer than
5919 * we should between interrupts. Instead, we write the ITR
5920 * value at the beginning of the next interrupt so the timing
5921 * ends up being correct.
5922 */
5923 q_vector->itr_val = new_itr;
5924 q_vector->set_itr = 1;
5925 }
5926}
5927
5928static void igb_tx_ctxtdesc(struct igb_ring *tx_ring,
5929 struct igb_tx_buffer *first,
5930 u32 vlan_macip_lens, u32 type_tucmd,
5931 u32 mss_l4len_idx)
5932{
5933 struct e1000_adv_tx_context_desc *context_desc;
5934 u16 i = tx_ring->next_to_use;
5935 struct timespec64 ts;
5936
5937 context_desc = IGB_TX_CTXTDESC(tx_ring, i);
5938
5939 i++;
5940 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
5941
5942 /* set bits to identify this as an advanced context descriptor */
5943 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
5944
5945 /* For 82575, context index must be unique per ring. */
5946 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
5947 mss_l4len_idx |= tx_ring->reg_idx << 4;
5948
5949 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5950 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
5951 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
5952
5953 /* We assume there is always a valid tx time available. Invalid times
5954 * should have been handled by the upper layers.
5955 */
5956 if (tx_ring->launchtime_enable) {
5957 ts = ktime_to_timespec64(first->skb->tstamp);
5958 skb_txtime_consumed(first->skb);
5959 context_desc->seqnum_seed = cpu_to_le32(ts.tv_nsec / 32);
5960 } else {
5961 context_desc->seqnum_seed = 0;
5962 }
5963}
5964
5965static int igb_tso(struct igb_ring *tx_ring,
5966 struct igb_tx_buffer *first,
5967 u8 *hdr_len)
5968{
5969 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
5970 struct sk_buff *skb = first->skb;
5971 union {
5972 struct iphdr *v4;
5973 struct ipv6hdr *v6;
5974 unsigned char *hdr;
5975 } ip;
5976 union {
5977 struct tcphdr *tcp;
5978 struct udphdr *udp;
5979 unsigned char *hdr;
5980 } l4;
5981 u32 paylen, l4_offset;
5982 int err;
5983
5984 if (skb->ip_summed != CHECKSUM_PARTIAL)
5985 return 0;
5986
5987 if (!skb_is_gso(skb))
5988 return 0;
5989
5990 err = skb_cow_head(skb, 0);
5991 if (err < 0)
5992 return err;
5993
5994 ip.hdr = skb_network_header(skb);
5995 l4.hdr = skb_checksum_start(skb);
5996
5997 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5998 type_tucmd = (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ?
5999 E1000_ADVTXD_TUCMD_L4T_UDP : E1000_ADVTXD_TUCMD_L4T_TCP;
6000
6001 /* initialize outer IP header fields */
6002 if (ip.v4->version == 4) {
6003 unsigned char *csum_start = skb_checksum_start(skb);
6004 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
6005
6006 /* IP header will have to cancel out any data that
6007 * is not a part of the outer IP header
6008 */
6009 ip.v4->check = csum_fold(csum_partial(trans_start,
6010 csum_start - trans_start,
6011 0));
6012 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
6013
6014 ip.v4->tot_len = 0;
6015 first->tx_flags |= IGB_TX_FLAGS_TSO |
6016 IGB_TX_FLAGS_CSUM |
6017 IGB_TX_FLAGS_IPV4;
6018 } else {
6019 ip.v6->payload_len = 0;
6020 first->tx_flags |= IGB_TX_FLAGS_TSO |
6021 IGB_TX_FLAGS_CSUM;
6022 }
6023
6024 /* determine offset of inner transport header */
6025 l4_offset = l4.hdr - skb->data;
6026
6027 /* remove payload length from inner checksum */
6028 paylen = skb->len - l4_offset;
6029 if (type_tucmd & E1000_ADVTXD_TUCMD_L4T_TCP) {
6030 /* compute length of segmentation header */
6031 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
6032 csum_replace_by_diff(&l4.tcp->check,
6033 (__force __wsum)htonl(paylen));
6034 } else {
6035 /* compute length of segmentation header */
6036 *hdr_len = sizeof(*l4.udp) + l4_offset;
6037 csum_replace_by_diff(&l4.udp->check,
6038 (__force __wsum)htonl(paylen));
6039 }
6040
6041 /* update gso size and bytecount with header size */
6042 first->gso_segs = skb_shinfo(skb)->gso_segs;
6043 first->bytecount += (first->gso_segs - 1) * *hdr_len;
6044
6045 /* MSS L4LEN IDX */
6046 mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT;
6047 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
6048
6049 /* VLAN MACLEN IPLEN */
6050 vlan_macip_lens = l4.hdr - ip.hdr;
6051 vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT;
6052 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
6053
6054 igb_tx_ctxtdesc(tx_ring, first, vlan_macip_lens,
6055 type_tucmd, mss_l4len_idx);
6056
6057 return 1;
6058}
6059
6060static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
6061{
6062 struct sk_buff *skb = first->skb;
6063 u32 vlan_macip_lens = 0;
6064 u32 type_tucmd = 0;
6065
6066 if (skb->ip_summed != CHECKSUM_PARTIAL) {
6067csum_failed:
6068 if (!(first->tx_flags & IGB_TX_FLAGS_VLAN) &&
6069 !tx_ring->launchtime_enable)
6070 return;
6071 goto no_csum;
6072 }
6073
6074 switch (skb->csum_offset) {
6075 case offsetof(struct tcphdr, check):
6076 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
6077 fallthrough;
6078 case offsetof(struct udphdr, check):
6079 break;
6080 case offsetof(struct sctphdr, checksum):
6081 /* validate that this is actually an SCTP request */
6082 if (skb_csum_is_sctp(skb)) {
6083 type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP;
6084 break;
6085 }
6086 fallthrough;
6087 default:
6088 skb_checksum_help(skb);
6089 goto csum_failed;
6090 }
6091
6092 /* update TX checksum flag */
6093 first->tx_flags |= IGB_TX_FLAGS_CSUM;
6094 vlan_macip_lens = skb_checksum_start_offset(skb) -
6095 skb_network_offset(skb);
6096no_csum:
6097 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
6098 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
6099
6100 igb_tx_ctxtdesc(tx_ring, first, vlan_macip_lens, type_tucmd, 0);
6101}
6102
6103#define IGB_SET_FLAG(_input, _flag, _result) \
6104 ((_flag <= _result) ? \
6105 ((u32)(_input & _flag) * (_result / _flag)) : \
6106 ((u32)(_input & _flag) / (_flag / _result)))
6107
6108static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
6109{
6110 /* set type for advanced descriptor with frame checksum insertion */
6111 u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
6112 E1000_ADVTXD_DCMD_DEXT |
6113 E1000_ADVTXD_DCMD_IFCS;
6114
6115 /* set HW vlan bit if vlan is present */
6116 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
6117 (E1000_ADVTXD_DCMD_VLE));
6118
6119 /* set segmentation bits for TSO */
6120 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
6121 (E1000_ADVTXD_DCMD_TSE));
6122
6123 /* set timestamp bit if present */
6124 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
6125 (E1000_ADVTXD_MAC_TSTAMP));
6126
6127 /* insert frame checksum */
6128 cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
6129
6130 return cmd_type;
6131}
6132
6133static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
6134 union e1000_adv_tx_desc *tx_desc,
6135 u32 tx_flags, unsigned int paylen)
6136{
6137 u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
6138
6139 /* 82575 requires a unique index per ring */
6140 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
6141 olinfo_status |= tx_ring->reg_idx << 4;
6142
6143 /* insert L4 checksum */
6144 olinfo_status |= IGB_SET_FLAG(tx_flags,
6145 IGB_TX_FLAGS_CSUM,
6146 (E1000_TXD_POPTS_TXSM << 8));
6147
6148 /* insert IPv4 checksum */
6149 olinfo_status |= IGB_SET_FLAG(tx_flags,
6150 IGB_TX_FLAGS_IPV4,
6151 (E1000_TXD_POPTS_IXSM << 8));
6152
6153 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6154}
6155
6156static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
6157{
6158 struct net_device *netdev = tx_ring->netdev;
6159
6160 netif_stop_subqueue(netdev, tx_ring->queue_index);
6161
6162 /* Herbert's original patch had:
6163 * smp_mb__after_netif_stop_queue();
6164 * but since that doesn't exist yet, just open code it.
6165 */
6166 smp_mb();
6167
6168 /* We need to check again in a case another CPU has just
6169 * made room available.
6170 */
6171 if (igb_desc_unused(tx_ring) < size)
6172 return -EBUSY;
6173
6174 /* A reprieve! */
6175 netif_wake_subqueue(netdev, tx_ring->queue_index);
6176
6177 u64_stats_update_begin(&tx_ring->tx_syncp2);
6178 tx_ring->tx_stats.restart_queue2++;
6179 u64_stats_update_end(&tx_ring->tx_syncp2);
6180
6181 return 0;
6182}
6183
6184static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
6185{
6186 if (igb_desc_unused(tx_ring) >= size)
6187 return 0;
6188 return __igb_maybe_stop_tx(tx_ring, size);
6189}
6190
6191static int igb_tx_map(struct igb_ring *tx_ring,
6192 struct igb_tx_buffer *first,
6193 const u8 hdr_len)
6194{
6195 struct sk_buff *skb = first->skb;
6196 struct igb_tx_buffer *tx_buffer;
6197 union e1000_adv_tx_desc *tx_desc;
6198 skb_frag_t *frag;
6199 dma_addr_t dma;
6200 unsigned int data_len, size;
6201 u32 tx_flags = first->tx_flags;
6202 u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
6203 u16 i = tx_ring->next_to_use;
6204
6205 tx_desc = IGB_TX_DESC(tx_ring, i);
6206
6207 igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
6208
6209 size = skb_headlen(skb);
6210 data_len = skb->data_len;
6211
6212 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
6213
6214 tx_buffer = first;
6215
6216 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
6217 if (dma_mapping_error(tx_ring->dev, dma))
6218 goto dma_error;
6219
6220 /* record length, and DMA address */
6221 dma_unmap_len_set(tx_buffer, len, size);
6222 dma_unmap_addr_set(tx_buffer, dma, dma);
6223
6224 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6225
6226 while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
6227 tx_desc->read.cmd_type_len =
6228 cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
6229
6230 i++;
6231 tx_desc++;
6232 if (i == tx_ring->count) {
6233 tx_desc = IGB_TX_DESC(tx_ring, 0);
6234 i = 0;
6235 }
6236 tx_desc->read.olinfo_status = 0;
6237
6238 dma += IGB_MAX_DATA_PER_TXD;
6239 size -= IGB_MAX_DATA_PER_TXD;
6240
6241 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6242 }
6243
6244 if (likely(!data_len))
6245 break;
6246
6247 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
6248
6249 i++;
6250 tx_desc++;
6251 if (i == tx_ring->count) {
6252 tx_desc = IGB_TX_DESC(tx_ring, 0);
6253 i = 0;
6254 }
6255 tx_desc->read.olinfo_status = 0;
6256
6257 size = skb_frag_size(frag);
6258 data_len -= size;
6259
6260 dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
6261 size, DMA_TO_DEVICE);
6262
6263 tx_buffer = &tx_ring->tx_buffer_info[i];
6264 }
6265
6266 /* write last descriptor with RS and EOP bits */
6267 cmd_type |= size | IGB_TXD_DCMD;
6268 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
6269
6270 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
6271
6272 /* set the timestamp */
6273 first->time_stamp = jiffies;
6274
6275 skb_tx_timestamp(skb);
6276
6277 /* Force memory writes to complete before letting h/w know there
6278 * are new descriptors to fetch. (Only applicable for weak-ordered
6279 * memory model archs, such as IA-64).
6280 *
6281 * We also need this memory barrier to make certain all of the
6282 * status bits have been updated before next_to_watch is written.
6283 */
6284 dma_wmb();
6285
6286 /* set next_to_watch value indicating a packet is present */
6287 first->next_to_watch = tx_desc;
6288
6289 i++;
6290 if (i == tx_ring->count)
6291 i = 0;
6292
6293 tx_ring->next_to_use = i;
6294
6295 /* Make sure there is space in the ring for the next send. */
6296 igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
6297
6298 if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
6299 writel(i, tx_ring->tail);
6300 }
6301 return 0;
6302
6303dma_error:
6304 dev_err(tx_ring->dev, "TX DMA map failed\n");
6305 tx_buffer = &tx_ring->tx_buffer_info[i];
6306
6307 /* clear dma mappings for failed tx_buffer_info map */
6308 while (tx_buffer != first) {
6309 if (dma_unmap_len(tx_buffer, len))
6310 dma_unmap_page(tx_ring->dev,
6311 dma_unmap_addr(tx_buffer, dma),
6312 dma_unmap_len(tx_buffer, len),
6313 DMA_TO_DEVICE);
6314 dma_unmap_len_set(tx_buffer, len, 0);
6315
6316 if (i-- == 0)
6317 i += tx_ring->count;
6318 tx_buffer = &tx_ring->tx_buffer_info[i];
6319 }
6320
6321 if (dma_unmap_len(tx_buffer, len))
6322 dma_unmap_single(tx_ring->dev,
6323 dma_unmap_addr(tx_buffer, dma),
6324 dma_unmap_len(tx_buffer, len),
6325 DMA_TO_DEVICE);
6326 dma_unmap_len_set(tx_buffer, len, 0);
6327
6328 dev_kfree_skb_any(tx_buffer->skb);
6329 tx_buffer->skb = NULL;
6330
6331 tx_ring->next_to_use = i;
6332
6333 return -1;
6334}
6335
6336int igb_xmit_xdp_ring(struct igb_adapter *adapter,
6337 struct igb_ring *tx_ring,
6338 struct xdp_frame *xdpf)
6339{
6340 struct skb_shared_info *sinfo = xdp_get_shared_info_from_frame(xdpf);
6341 u8 nr_frags = unlikely(xdp_frame_has_frags(xdpf)) ? sinfo->nr_frags : 0;
6342 u16 count, i, index = tx_ring->next_to_use;
6343 struct igb_tx_buffer *tx_head = &tx_ring->tx_buffer_info[index];
6344 struct igb_tx_buffer *tx_buffer = tx_head;
6345 union e1000_adv_tx_desc *tx_desc = IGB_TX_DESC(tx_ring, index);
6346 u32 len = xdpf->len, cmd_type, olinfo_status;
6347 void *data = xdpf->data;
6348
6349 count = TXD_USE_COUNT(len);
6350 for (i = 0; i < nr_frags; i++)
6351 count += TXD_USE_COUNT(skb_frag_size(&sinfo->frags[i]));
6352
6353 if (igb_maybe_stop_tx(tx_ring, count + 3))
6354 return IGB_XDP_CONSUMED;
6355
6356 i = 0;
6357 /* record the location of the first descriptor for this packet */
6358 tx_head->bytecount = xdp_get_frame_len(xdpf);
6359 tx_head->type = IGB_TYPE_XDP;
6360 tx_head->gso_segs = 1;
6361 tx_head->xdpf = xdpf;
6362
6363 olinfo_status = tx_head->bytecount << E1000_ADVTXD_PAYLEN_SHIFT;
6364 /* 82575 requires a unique index per ring */
6365 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
6366 olinfo_status |= tx_ring->reg_idx << 4;
6367 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6368
6369 for (;;) {
6370 dma_addr_t dma;
6371
6372 dma = dma_map_single(tx_ring->dev, data, len, DMA_TO_DEVICE);
6373 if (dma_mapping_error(tx_ring->dev, dma))
6374 goto unmap;
6375
6376 /* record length, and DMA address */
6377 dma_unmap_len_set(tx_buffer, len, len);
6378 dma_unmap_addr_set(tx_buffer, dma, dma);
6379
6380 /* put descriptor type bits */
6381 cmd_type = E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_DEXT |
6382 E1000_ADVTXD_DCMD_IFCS | len;
6383
6384 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
6385 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6386
6387 tx_buffer->protocol = 0;
6388
6389 if (++index == tx_ring->count)
6390 index = 0;
6391
6392 if (i == nr_frags)
6393 break;
6394
6395 tx_buffer = &tx_ring->tx_buffer_info[index];
6396 tx_desc = IGB_TX_DESC(tx_ring, index);
6397 tx_desc->read.olinfo_status = 0;
6398
6399 data = skb_frag_address(&sinfo->frags[i]);
6400 len = skb_frag_size(&sinfo->frags[i]);
6401 i++;
6402 }
6403 tx_desc->read.cmd_type_len |= cpu_to_le32(IGB_TXD_DCMD);
6404
6405 netdev_tx_sent_queue(txring_txq(tx_ring), tx_head->bytecount);
6406 /* set the timestamp */
6407 tx_head->time_stamp = jiffies;
6408
6409 /* Avoid any potential race with xdp_xmit and cleanup */
6410 smp_wmb();
6411
6412 /* set next_to_watch value indicating a packet is present */
6413 tx_head->next_to_watch = tx_desc;
6414 tx_ring->next_to_use = index;
6415
6416 /* Make sure there is space in the ring for the next send. */
6417 igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
6418
6419 if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more())
6420 writel(index, tx_ring->tail);
6421
6422 return IGB_XDP_TX;
6423
6424unmap:
6425 for (;;) {
6426 tx_buffer = &tx_ring->tx_buffer_info[index];
6427 if (dma_unmap_len(tx_buffer, len))
6428 dma_unmap_page(tx_ring->dev,
6429 dma_unmap_addr(tx_buffer, dma),
6430 dma_unmap_len(tx_buffer, len),
6431 DMA_TO_DEVICE);
6432 dma_unmap_len_set(tx_buffer, len, 0);
6433 if (tx_buffer == tx_head)
6434 break;
6435
6436 if (!index)
6437 index += tx_ring->count;
6438 index--;
6439 }
6440
6441 return IGB_XDP_CONSUMED;
6442}
6443
6444netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
6445 struct igb_ring *tx_ring)
6446{
6447 struct igb_tx_buffer *first;
6448 int tso;
6449 u32 tx_flags = 0;
6450 unsigned short f;
6451 u16 count = TXD_USE_COUNT(skb_headlen(skb));
6452 __be16 protocol = vlan_get_protocol(skb);
6453 u8 hdr_len = 0;
6454
6455 /* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
6456 * + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
6457 * + 2 desc gap to keep tail from touching head,
6458 * + 1 desc for context descriptor,
6459 * otherwise try next time
6460 */
6461 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6462 count += TXD_USE_COUNT(skb_frag_size(
6463 &skb_shinfo(skb)->frags[f]));
6464
6465 if (igb_maybe_stop_tx(tx_ring, count + 3)) {
6466 /* this is a hard error */
6467 return NETDEV_TX_BUSY;
6468 }
6469
6470 /* record the location of the first descriptor for this packet */
6471 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6472 first->type = IGB_TYPE_SKB;
6473 first->skb = skb;
6474 first->bytecount = skb->len;
6475 first->gso_segs = 1;
6476
6477 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
6478 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
6479
6480 if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
6481 !test_and_set_bit_lock(__IGB_PTP_TX_IN_PROGRESS,
6482 &adapter->state)) {
6483 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
6484 tx_flags |= IGB_TX_FLAGS_TSTAMP;
6485
6486 adapter->ptp_tx_skb = skb_get(skb);
6487 adapter->ptp_tx_start = jiffies;
6488 if (adapter->hw.mac.type == e1000_82576)
6489 schedule_work(&adapter->ptp_tx_work);
6490 } else {
6491 adapter->tx_hwtstamp_skipped++;
6492 }
6493 }
6494
6495 if (skb_vlan_tag_present(skb)) {
6496 tx_flags |= IGB_TX_FLAGS_VLAN;
6497 tx_flags |= (skb_vlan_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
6498 }
6499
6500 /* record initial flags and protocol */
6501 first->tx_flags = tx_flags;
6502 first->protocol = protocol;
6503
6504 tso = igb_tso(tx_ring, first, &hdr_len);
6505 if (tso < 0)
6506 goto out_drop;
6507 else if (!tso)
6508 igb_tx_csum(tx_ring, first);
6509
6510 if (igb_tx_map(tx_ring, first, hdr_len))
6511 goto cleanup_tx_tstamp;
6512
6513 return NETDEV_TX_OK;
6514
6515out_drop:
6516 dev_kfree_skb_any(first->skb);
6517 first->skb = NULL;
6518cleanup_tx_tstamp:
6519 if (unlikely(tx_flags & IGB_TX_FLAGS_TSTAMP)) {
6520 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
6521
6522 dev_kfree_skb_any(adapter->ptp_tx_skb);
6523 adapter->ptp_tx_skb = NULL;
6524 if (adapter->hw.mac.type == e1000_82576)
6525 cancel_work_sync(&adapter->ptp_tx_work);
6526 clear_bit_unlock(__IGB_PTP_TX_IN_PROGRESS, &adapter->state);
6527 }
6528
6529 return NETDEV_TX_OK;
6530}
6531
6532static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
6533 struct sk_buff *skb)
6534{
6535 unsigned int r_idx = skb->queue_mapping;
6536
6537 if (r_idx >= adapter->num_tx_queues)
6538 r_idx = r_idx % adapter->num_tx_queues;
6539
6540 return adapter->tx_ring[r_idx];
6541}
6542
6543static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
6544 struct net_device *netdev)
6545{
6546 struct igb_adapter *adapter = netdev_priv(netdev);
6547
6548 /* The minimum packet size with TCTL.PSP set is 17 so pad the skb
6549 * in order to meet this minimum size requirement.
6550 */
6551 if (skb_put_padto(skb, 17))
6552 return NETDEV_TX_OK;
6553
6554 return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
6555}
6556
6557/**
6558 * igb_tx_timeout - Respond to a Tx Hang
6559 * @netdev: network interface device structure
6560 * @txqueue: number of the Tx queue that hung (unused)
6561 **/
6562static void igb_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
6563{
6564 struct igb_adapter *adapter = netdev_priv(netdev);
6565 struct e1000_hw *hw = &adapter->hw;
6566
6567 /* Do the reset outside of interrupt context */
6568 adapter->tx_timeout_count++;
6569
6570 if (hw->mac.type >= e1000_82580)
6571 hw->dev_spec._82575.global_device_reset = true;
6572
6573 schedule_work(&adapter->reset_task);
6574 wr32(E1000_EICS,
6575 (adapter->eims_enable_mask & ~adapter->eims_other));
6576}
6577
6578static void igb_reset_task(struct work_struct *work)
6579{
6580 struct igb_adapter *adapter;
6581 adapter = container_of(work, struct igb_adapter, reset_task);
6582
6583 rtnl_lock();
6584 /* If we're already down or resetting, just bail */
6585 if (test_bit(__IGB_DOWN, &adapter->state) ||
6586 test_bit(__IGB_RESETTING, &adapter->state)) {
6587 rtnl_unlock();
6588 return;
6589 }
6590
6591 igb_dump(adapter);
6592 netdev_err(adapter->netdev, "Reset adapter\n");
6593 igb_reinit_locked(adapter);
6594 rtnl_unlock();
6595}
6596
6597/**
6598 * igb_get_stats64 - Get System Network Statistics
6599 * @netdev: network interface device structure
6600 * @stats: rtnl_link_stats64 pointer
6601 **/
6602static void igb_get_stats64(struct net_device *netdev,
6603 struct rtnl_link_stats64 *stats)
6604{
6605 struct igb_adapter *adapter = netdev_priv(netdev);
6606
6607 spin_lock(&adapter->stats64_lock);
6608 igb_update_stats(adapter);
6609 memcpy(stats, &adapter->stats64, sizeof(*stats));
6610 spin_unlock(&adapter->stats64_lock);
6611}
6612
6613/**
6614 * igb_change_mtu - Change the Maximum Transfer Unit
6615 * @netdev: network interface device structure
6616 * @new_mtu: new value for maximum frame size
6617 *
6618 * Returns 0 on success, negative on failure
6619 **/
6620static int igb_change_mtu(struct net_device *netdev, int new_mtu)
6621{
6622 struct igb_adapter *adapter = netdev_priv(netdev);
6623 int max_frame = new_mtu + IGB_ETH_PKT_HDR_PAD;
6624
6625 if (adapter->xdp_prog) {
6626 int i;
6627
6628 for (i = 0; i < adapter->num_rx_queues; i++) {
6629 struct igb_ring *ring = adapter->rx_ring[i];
6630
6631 if (max_frame > igb_rx_bufsz(ring)) {
6632 netdev_warn(adapter->netdev,
6633 "Requested MTU size is not supported with XDP. Max frame size is %d\n",
6634 max_frame);
6635 return -EINVAL;
6636 }
6637 }
6638 }
6639
6640 /* adjust max frame to be at least the size of a standard frame */
6641 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
6642 max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
6643
6644 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
6645 usleep_range(1000, 2000);
6646
6647 /* igb_down has a dependency on max_frame_size */
6648 adapter->max_frame_size = max_frame;
6649
6650 if (netif_running(netdev))
6651 igb_down(adapter);
6652
6653 netdev_dbg(netdev, "changing MTU from %d to %d\n",
6654 netdev->mtu, new_mtu);
6655 WRITE_ONCE(netdev->mtu, new_mtu);
6656
6657 if (netif_running(netdev))
6658 igb_up(adapter);
6659 else
6660 igb_reset(adapter);
6661
6662 clear_bit(__IGB_RESETTING, &adapter->state);
6663
6664 return 0;
6665}
6666
6667/**
6668 * igb_update_stats - Update the board statistics counters
6669 * @adapter: board private structure
6670 **/
6671void igb_update_stats(struct igb_adapter *adapter)
6672{
6673 struct rtnl_link_stats64 *net_stats = &adapter->stats64;
6674 struct e1000_hw *hw = &adapter->hw;
6675 struct pci_dev *pdev = adapter->pdev;
6676 u32 reg, mpc;
6677 int i;
6678 u64 bytes, packets;
6679 unsigned int start;
6680 u64 _bytes, _packets;
6681
6682 /* Prevent stats update while adapter is being reset, or if the pci
6683 * connection is down.
6684 */
6685 if (adapter->link_speed == 0)
6686 return;
6687 if (pci_channel_offline(pdev))
6688 return;
6689
6690 bytes = 0;
6691 packets = 0;
6692
6693 rcu_read_lock();
6694 for (i = 0; i < adapter->num_rx_queues; i++) {
6695 struct igb_ring *ring = adapter->rx_ring[i];
6696 u32 rqdpc = rd32(E1000_RQDPC(i));
6697 if (hw->mac.type >= e1000_i210)
6698 wr32(E1000_RQDPC(i), 0);
6699
6700 if (rqdpc) {
6701 ring->rx_stats.drops += rqdpc;
6702 net_stats->rx_fifo_errors += rqdpc;
6703 }
6704
6705 do {
6706 start = u64_stats_fetch_begin(&ring->rx_syncp);
6707 _bytes = ring->rx_stats.bytes;
6708 _packets = ring->rx_stats.packets;
6709 } while (u64_stats_fetch_retry(&ring->rx_syncp, start));
6710 bytes += _bytes;
6711 packets += _packets;
6712 }
6713
6714 net_stats->rx_bytes = bytes;
6715 net_stats->rx_packets = packets;
6716
6717 bytes = 0;
6718 packets = 0;
6719 for (i = 0; i < adapter->num_tx_queues; i++) {
6720 struct igb_ring *ring = adapter->tx_ring[i];
6721 do {
6722 start = u64_stats_fetch_begin(&ring->tx_syncp);
6723 _bytes = ring->tx_stats.bytes;
6724 _packets = ring->tx_stats.packets;
6725 } while (u64_stats_fetch_retry(&ring->tx_syncp, start));
6726 bytes += _bytes;
6727 packets += _packets;
6728 }
6729 net_stats->tx_bytes = bytes;
6730 net_stats->tx_packets = packets;
6731 rcu_read_unlock();
6732
6733 /* read stats registers */
6734 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
6735 adapter->stats.gprc += rd32(E1000_GPRC);
6736 adapter->stats.gorc += rd32(E1000_GORCL);
6737 rd32(E1000_GORCH); /* clear GORCL */
6738 adapter->stats.bprc += rd32(E1000_BPRC);
6739 adapter->stats.mprc += rd32(E1000_MPRC);
6740 adapter->stats.roc += rd32(E1000_ROC);
6741
6742 adapter->stats.prc64 += rd32(E1000_PRC64);
6743 adapter->stats.prc127 += rd32(E1000_PRC127);
6744 adapter->stats.prc255 += rd32(E1000_PRC255);
6745 adapter->stats.prc511 += rd32(E1000_PRC511);
6746 adapter->stats.prc1023 += rd32(E1000_PRC1023);
6747 adapter->stats.prc1522 += rd32(E1000_PRC1522);
6748 adapter->stats.symerrs += rd32(E1000_SYMERRS);
6749 adapter->stats.sec += rd32(E1000_SEC);
6750
6751 mpc = rd32(E1000_MPC);
6752 adapter->stats.mpc += mpc;
6753 net_stats->rx_fifo_errors += mpc;
6754 adapter->stats.scc += rd32(E1000_SCC);
6755 adapter->stats.ecol += rd32(E1000_ECOL);
6756 adapter->stats.mcc += rd32(E1000_MCC);
6757 adapter->stats.latecol += rd32(E1000_LATECOL);
6758 adapter->stats.dc += rd32(E1000_DC);
6759 adapter->stats.rlec += rd32(E1000_RLEC);
6760 adapter->stats.xonrxc += rd32(E1000_XONRXC);
6761 adapter->stats.xontxc += rd32(E1000_XONTXC);
6762 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
6763 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
6764 adapter->stats.fcruc += rd32(E1000_FCRUC);
6765 adapter->stats.gptc += rd32(E1000_GPTC);
6766 adapter->stats.gotc += rd32(E1000_GOTCL);
6767 rd32(E1000_GOTCH); /* clear GOTCL */
6768 adapter->stats.rnbc += rd32(E1000_RNBC);
6769 adapter->stats.ruc += rd32(E1000_RUC);
6770 adapter->stats.rfc += rd32(E1000_RFC);
6771 adapter->stats.rjc += rd32(E1000_RJC);
6772 adapter->stats.tor += rd32(E1000_TORH);
6773 adapter->stats.tot += rd32(E1000_TOTH);
6774 adapter->stats.tpr += rd32(E1000_TPR);
6775
6776 adapter->stats.ptc64 += rd32(E1000_PTC64);
6777 adapter->stats.ptc127 += rd32(E1000_PTC127);
6778 adapter->stats.ptc255 += rd32(E1000_PTC255);
6779 adapter->stats.ptc511 += rd32(E1000_PTC511);
6780 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
6781 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
6782
6783 adapter->stats.mptc += rd32(E1000_MPTC);
6784 adapter->stats.bptc += rd32(E1000_BPTC);
6785
6786 adapter->stats.tpt += rd32(E1000_TPT);
6787 adapter->stats.colc += rd32(E1000_COLC);
6788
6789 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
6790 /* read internal phy specific stats */
6791 reg = rd32(E1000_CTRL_EXT);
6792 if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
6793 adapter->stats.rxerrc += rd32(E1000_RXERRC);
6794
6795 /* this stat has invalid values on i210/i211 */
6796 if ((hw->mac.type != e1000_i210) &&
6797 (hw->mac.type != e1000_i211))
6798 adapter->stats.tncrs += rd32(E1000_TNCRS);
6799 }
6800
6801 adapter->stats.tsctc += rd32(E1000_TSCTC);
6802 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
6803
6804 adapter->stats.iac += rd32(E1000_IAC);
6805 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
6806 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
6807 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
6808 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
6809 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
6810 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
6811 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
6812 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
6813
6814 /* Fill out the OS statistics structure */
6815 net_stats->multicast = adapter->stats.mprc;
6816 net_stats->collisions = adapter->stats.colc;
6817
6818 /* Rx Errors */
6819
6820 /* RLEC on some newer hardware can be incorrect so build
6821 * our own version based on RUC and ROC
6822 */
6823 net_stats->rx_errors = adapter->stats.rxerrc +
6824 adapter->stats.crcerrs + adapter->stats.algnerrc +
6825 adapter->stats.ruc + adapter->stats.roc +
6826 adapter->stats.cexterr;
6827 net_stats->rx_length_errors = adapter->stats.ruc +
6828 adapter->stats.roc;
6829 net_stats->rx_crc_errors = adapter->stats.crcerrs;
6830 net_stats->rx_frame_errors = adapter->stats.algnerrc;
6831 net_stats->rx_missed_errors = adapter->stats.mpc;
6832
6833 /* Tx Errors */
6834 net_stats->tx_errors = adapter->stats.ecol +
6835 adapter->stats.latecol;
6836 net_stats->tx_aborted_errors = adapter->stats.ecol;
6837 net_stats->tx_window_errors = adapter->stats.latecol;
6838 net_stats->tx_carrier_errors = adapter->stats.tncrs;
6839
6840 /* Tx Dropped needs to be maintained elsewhere */
6841
6842 /* Management Stats */
6843 adapter->stats.mgptc += rd32(E1000_MGTPTC);
6844 adapter->stats.mgprc += rd32(E1000_MGTPRC);
6845 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
6846
6847 /* OS2BMC Stats */
6848 reg = rd32(E1000_MANC);
6849 if (reg & E1000_MANC_EN_BMC2OS) {
6850 adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
6851 adapter->stats.o2bspc += rd32(E1000_O2BSPC);
6852 adapter->stats.b2ospc += rd32(E1000_B2OSPC);
6853 adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
6854 }
6855}
6856
6857static void igb_perout(struct igb_adapter *adapter, int tsintr_tt)
6858{
6859 int pin = ptp_find_pin(adapter->ptp_clock, PTP_PF_PEROUT, tsintr_tt);
6860 struct e1000_hw *hw = &adapter->hw;
6861 struct timespec64 ts;
6862 u32 tsauxc;
6863
6864 if (pin < 0 || pin >= IGB_N_SDP)
6865 return;
6866
6867 spin_lock(&adapter->tmreg_lock);
6868
6869 if (hw->mac.type == e1000_82580 ||
6870 hw->mac.type == e1000_i354 ||
6871 hw->mac.type == e1000_i350) {
6872 s64 ns = timespec64_to_ns(&adapter->perout[tsintr_tt].period);
6873 u32 systiml, systimh, level_mask, level, rem;
6874 u64 systim, now;
6875
6876 /* read systim registers in sequence */
6877 rd32(E1000_SYSTIMR);
6878 systiml = rd32(E1000_SYSTIML);
6879 systimh = rd32(E1000_SYSTIMH);
6880 systim = (((u64)(systimh & 0xFF)) << 32) | ((u64)systiml);
6881 now = timecounter_cyc2time(&adapter->tc, systim);
6882
6883 if (pin < 2) {
6884 level_mask = (tsintr_tt == 1) ? 0x80000 : 0x40000;
6885 level = (rd32(E1000_CTRL) & level_mask) ? 1 : 0;
6886 } else {
6887 level_mask = (tsintr_tt == 1) ? 0x80 : 0x40;
6888 level = (rd32(E1000_CTRL_EXT) & level_mask) ? 1 : 0;
6889 }
6890
6891 div_u64_rem(now, ns, &rem);
6892 systim = systim + (ns - rem);
6893
6894 /* synchronize pin level with rising/falling edges */
6895 div_u64_rem(now, ns << 1, &rem);
6896 if (rem < ns) {
6897 /* first half of period */
6898 if (level == 0) {
6899 /* output is already low, skip this period */
6900 systim += ns;
6901 pr_notice("igb: periodic output on %s missed falling edge\n",
6902 adapter->sdp_config[pin].name);
6903 }
6904 } else {
6905 /* second half of period */
6906 if (level == 1) {
6907 /* output is already high, skip this period */
6908 systim += ns;
6909 pr_notice("igb: periodic output on %s missed rising edge\n",
6910 adapter->sdp_config[pin].name);
6911 }
6912 }
6913
6914 /* for this chip family tv_sec is the upper part of the binary value,
6915 * so not seconds
6916 */
6917 ts.tv_nsec = (u32)systim;
6918 ts.tv_sec = ((u32)(systim >> 32)) & 0xFF;
6919 } else {
6920 ts = timespec64_add(adapter->perout[tsintr_tt].start,
6921 adapter->perout[tsintr_tt].period);
6922 }
6923
6924 /* u32 conversion of tv_sec is safe until y2106 */
6925 wr32((tsintr_tt == 1) ? E1000_TRGTTIML1 : E1000_TRGTTIML0, ts.tv_nsec);
6926 wr32((tsintr_tt == 1) ? E1000_TRGTTIMH1 : E1000_TRGTTIMH0, (u32)ts.tv_sec);
6927 tsauxc = rd32(E1000_TSAUXC);
6928 tsauxc |= TSAUXC_EN_TT0;
6929 wr32(E1000_TSAUXC, tsauxc);
6930 adapter->perout[tsintr_tt].start = ts;
6931
6932 spin_unlock(&adapter->tmreg_lock);
6933}
6934
6935static void igb_extts(struct igb_adapter *adapter, int tsintr_tt)
6936{
6937 int pin = ptp_find_pin(adapter->ptp_clock, PTP_PF_EXTTS, tsintr_tt);
6938 int auxstmpl = (tsintr_tt == 1) ? E1000_AUXSTMPL1 : E1000_AUXSTMPL0;
6939 int auxstmph = (tsintr_tt == 1) ? E1000_AUXSTMPH1 : E1000_AUXSTMPH0;
6940 struct e1000_hw *hw = &adapter->hw;
6941 struct ptp_clock_event event;
6942 struct timespec64 ts;
6943 unsigned long flags;
6944
6945 if (pin < 0 || pin >= IGB_N_SDP)
6946 return;
6947
6948 if (hw->mac.type == e1000_82580 ||
6949 hw->mac.type == e1000_i354 ||
6950 hw->mac.type == e1000_i350) {
6951 u64 ns = rd32(auxstmpl);
6952
6953 ns += ((u64)(rd32(auxstmph) & 0xFF)) << 32;
6954 spin_lock_irqsave(&adapter->tmreg_lock, flags);
6955 ns = timecounter_cyc2time(&adapter->tc, ns);
6956 spin_unlock_irqrestore(&adapter->tmreg_lock, flags);
6957 ts = ns_to_timespec64(ns);
6958 } else {
6959 ts.tv_nsec = rd32(auxstmpl);
6960 ts.tv_sec = rd32(auxstmph);
6961 }
6962
6963 event.type = PTP_CLOCK_EXTTS;
6964 event.index = tsintr_tt;
6965 event.timestamp = ts.tv_sec * 1000000000ULL + ts.tv_nsec;
6966 ptp_clock_event(adapter->ptp_clock, &event);
6967}
6968
6969static void igb_tsync_interrupt(struct igb_adapter *adapter)
6970{
6971 const u32 mask = (TSINTR_SYS_WRAP | E1000_TSICR_TXTS |
6972 TSINTR_TT0 | TSINTR_TT1 |
6973 TSINTR_AUTT0 | TSINTR_AUTT1);
6974 struct e1000_hw *hw = &adapter->hw;
6975 u32 tsicr = rd32(E1000_TSICR);
6976 struct ptp_clock_event event;
6977
6978 if (hw->mac.type == e1000_82580) {
6979 /* 82580 has a hardware bug that requires an explicit
6980 * write to clear the TimeSync interrupt cause.
6981 */
6982 wr32(E1000_TSICR, tsicr & mask);
6983 }
6984
6985 if (tsicr & TSINTR_SYS_WRAP) {
6986 event.type = PTP_CLOCK_PPS;
6987 if (adapter->ptp_caps.pps)
6988 ptp_clock_event(adapter->ptp_clock, &event);
6989 }
6990
6991 if (tsicr & E1000_TSICR_TXTS) {
6992 /* retrieve hardware timestamp */
6993 schedule_work(&adapter->ptp_tx_work);
6994 }
6995
6996 if (tsicr & TSINTR_TT0)
6997 igb_perout(adapter, 0);
6998
6999 if (tsicr & TSINTR_TT1)
7000 igb_perout(adapter, 1);
7001
7002 if (tsicr & TSINTR_AUTT0)
7003 igb_extts(adapter, 0);
7004
7005 if (tsicr & TSINTR_AUTT1)
7006 igb_extts(adapter, 1);
7007}
7008
7009static irqreturn_t igb_msix_other(int irq, void *data)
7010{
7011 struct igb_adapter *adapter = data;
7012 struct e1000_hw *hw = &adapter->hw;
7013 u32 icr = rd32(E1000_ICR);
7014 /* reading ICR causes bit 31 of EICR to be cleared */
7015
7016 if (icr & E1000_ICR_DRSTA)
7017 schedule_work(&adapter->reset_task);
7018
7019 if (icr & E1000_ICR_DOUTSYNC) {
7020 /* HW is reporting DMA is out of sync */
7021 adapter->stats.doosync++;
7022 /* The DMA Out of Sync is also indication of a spoof event
7023 * in IOV mode. Check the Wrong VM Behavior register to
7024 * see if it is really a spoof event.
7025 */
7026 igb_check_wvbr(adapter);
7027 }
7028
7029 /* Check for a mailbox event */
7030 if (icr & E1000_ICR_VMMB)
7031 igb_msg_task(adapter);
7032
7033 if (icr & E1000_ICR_LSC) {
7034 hw->mac.get_link_status = 1;
7035 /* guard against interrupt when we're going down */
7036 if (!test_bit(__IGB_DOWN, &adapter->state))
7037 mod_timer(&adapter->watchdog_timer, jiffies + 1);
7038 }
7039
7040 if (icr & E1000_ICR_TS)
7041 igb_tsync_interrupt(adapter);
7042
7043 wr32(E1000_EIMS, adapter->eims_other);
7044
7045 return IRQ_HANDLED;
7046}
7047
7048static void igb_write_itr(struct igb_q_vector *q_vector)
7049{
7050 struct igb_adapter *adapter = q_vector->adapter;
7051 u32 itr_val = q_vector->itr_val & 0x7FFC;
7052
7053 if (!q_vector->set_itr)
7054 return;
7055
7056 if (!itr_val)
7057 itr_val = 0x4;
7058
7059 if (adapter->hw.mac.type == e1000_82575)
7060 itr_val |= itr_val << 16;
7061 else
7062 itr_val |= E1000_EITR_CNT_IGNR;
7063
7064 writel(itr_val, q_vector->itr_register);
7065 q_vector->set_itr = 0;
7066}
7067
7068static irqreturn_t igb_msix_ring(int irq, void *data)
7069{
7070 struct igb_q_vector *q_vector = data;
7071
7072 /* Write the ITR value calculated from the previous interrupt. */
7073 igb_write_itr(q_vector);
7074
7075 napi_schedule(&q_vector->napi);
7076
7077 return IRQ_HANDLED;
7078}
7079
7080#ifdef CONFIG_IGB_DCA
7081static void igb_update_tx_dca(struct igb_adapter *adapter,
7082 struct igb_ring *tx_ring,
7083 int cpu)
7084{
7085 struct e1000_hw *hw = &adapter->hw;
7086 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
7087
7088 if (hw->mac.type != e1000_82575)
7089 txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
7090
7091 /* We can enable relaxed ordering for reads, but not writes when
7092 * DCA is enabled. This is due to a known issue in some chipsets
7093 * which will cause the DCA tag to be cleared.
7094 */
7095 txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
7096 E1000_DCA_TXCTRL_DATA_RRO_EN |
7097 E1000_DCA_TXCTRL_DESC_DCA_EN;
7098
7099 wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
7100}
7101
7102static void igb_update_rx_dca(struct igb_adapter *adapter,
7103 struct igb_ring *rx_ring,
7104 int cpu)
7105{
7106 struct e1000_hw *hw = &adapter->hw;
7107 u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
7108
7109 if (hw->mac.type != e1000_82575)
7110 rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
7111
7112 /* We can enable relaxed ordering for reads, but not writes when
7113 * DCA is enabled. This is due to a known issue in some chipsets
7114 * which will cause the DCA tag to be cleared.
7115 */
7116 rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
7117 E1000_DCA_RXCTRL_DESC_DCA_EN;
7118
7119 wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
7120}
7121
7122static void igb_update_dca(struct igb_q_vector *q_vector)
7123{
7124 struct igb_adapter *adapter = q_vector->adapter;
7125 int cpu = get_cpu();
7126
7127 if (q_vector->cpu == cpu)
7128 goto out_no_update;
7129
7130 if (q_vector->tx.ring)
7131 igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
7132
7133 if (q_vector->rx.ring)
7134 igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
7135
7136 q_vector->cpu = cpu;
7137out_no_update:
7138 put_cpu();
7139}
7140
7141static void igb_setup_dca(struct igb_adapter *adapter)
7142{
7143 struct e1000_hw *hw = &adapter->hw;
7144 int i;
7145
7146 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
7147 return;
7148
7149 /* Always use CB2 mode, difference is masked in the CB driver. */
7150 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
7151
7152 for (i = 0; i < adapter->num_q_vectors; i++) {
7153 adapter->q_vector[i]->cpu = -1;
7154 igb_update_dca(adapter->q_vector[i]);
7155 }
7156}
7157
7158static int __igb_notify_dca(struct device *dev, void *data)
7159{
7160 struct net_device *netdev = dev_get_drvdata(dev);
7161 struct igb_adapter *adapter = netdev_priv(netdev);
7162 struct pci_dev *pdev = adapter->pdev;
7163 struct e1000_hw *hw = &adapter->hw;
7164 unsigned long event = *(unsigned long *)data;
7165
7166 switch (event) {
7167 case DCA_PROVIDER_ADD:
7168 /* if already enabled, don't do it again */
7169 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
7170 break;
7171 if (dca_add_requester(dev) == 0) {
7172 adapter->flags |= IGB_FLAG_DCA_ENABLED;
7173 dev_info(&pdev->dev, "DCA enabled\n");
7174 igb_setup_dca(adapter);
7175 break;
7176 }
7177 fallthrough; /* since DCA is disabled. */
7178 case DCA_PROVIDER_REMOVE:
7179 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
7180 /* without this a class_device is left
7181 * hanging around in the sysfs model
7182 */
7183 dca_remove_requester(dev);
7184 dev_info(&pdev->dev, "DCA disabled\n");
7185 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
7186 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
7187 }
7188 break;
7189 }
7190
7191 return 0;
7192}
7193
7194static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
7195 void *p)
7196{
7197 int ret_val;
7198
7199 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
7200 __igb_notify_dca);
7201
7202 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7203}
7204#endif /* CONFIG_IGB_DCA */
7205
7206#ifdef CONFIG_PCI_IOV
7207static int igb_vf_configure(struct igb_adapter *adapter, int vf)
7208{
7209 unsigned char mac_addr[ETH_ALEN];
7210
7211 eth_zero_addr(mac_addr);
7212 igb_set_vf_mac(adapter, vf, mac_addr);
7213
7214 /* By default spoof check is enabled for all VFs */
7215 adapter->vf_data[vf].spoofchk_enabled = true;
7216
7217 /* By default VFs are not trusted */
7218 adapter->vf_data[vf].trusted = false;
7219
7220 return 0;
7221}
7222
7223#endif
7224static void igb_ping_all_vfs(struct igb_adapter *adapter)
7225{
7226 struct e1000_hw *hw = &adapter->hw;
7227 u32 ping;
7228 int i;
7229
7230 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
7231 ping = E1000_PF_CONTROL_MSG;
7232 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
7233 ping |= E1000_VT_MSGTYPE_CTS;
7234 igb_write_mbx(hw, &ping, 1, i);
7235 }
7236}
7237
7238static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
7239{
7240 struct e1000_hw *hw = &adapter->hw;
7241 u32 vmolr = rd32(E1000_VMOLR(vf));
7242 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7243
7244 vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
7245 IGB_VF_FLAG_MULTI_PROMISC);
7246 vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
7247
7248 if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
7249 vmolr |= E1000_VMOLR_MPME;
7250 vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
7251 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
7252 } else {
7253 /* if we have hashes and we are clearing a multicast promisc
7254 * flag we need to write the hashes to the MTA as this step
7255 * was previously skipped
7256 */
7257 if (vf_data->num_vf_mc_hashes > 30) {
7258 vmolr |= E1000_VMOLR_MPME;
7259 } else if (vf_data->num_vf_mc_hashes) {
7260 int j;
7261
7262 vmolr |= E1000_VMOLR_ROMPE;
7263 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
7264 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
7265 }
7266 }
7267
7268 wr32(E1000_VMOLR(vf), vmolr);
7269
7270 /* there are flags left unprocessed, likely not supported */
7271 if (*msgbuf & E1000_VT_MSGINFO_MASK)
7272 return -EINVAL;
7273
7274 return 0;
7275}
7276
7277static int igb_set_vf_multicasts(struct igb_adapter *adapter,
7278 u32 *msgbuf, u32 vf)
7279{
7280 int n = FIELD_GET(E1000_VT_MSGINFO_MASK, msgbuf[0]);
7281 u16 *hash_list = (u16 *)&msgbuf[1];
7282 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7283 int i;
7284
7285 /* salt away the number of multicast addresses assigned
7286 * to this VF for later use to restore when the PF multi cast
7287 * list changes
7288 */
7289 vf_data->num_vf_mc_hashes = n;
7290
7291 /* only up to 30 hash values supported */
7292 if (n > 30)
7293 n = 30;
7294
7295 /* store the hashes for later use */
7296 for (i = 0; i < n; i++)
7297 vf_data->vf_mc_hashes[i] = hash_list[i];
7298
7299 /* Flush and reset the mta with the new values */
7300 igb_set_rx_mode(adapter->netdev);
7301
7302 return 0;
7303}
7304
7305static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
7306{
7307 struct e1000_hw *hw = &adapter->hw;
7308 struct vf_data_storage *vf_data;
7309 int i, j;
7310
7311 for (i = 0; i < adapter->vfs_allocated_count; i++) {
7312 u32 vmolr = rd32(E1000_VMOLR(i));
7313
7314 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
7315
7316 vf_data = &adapter->vf_data[i];
7317
7318 if ((vf_data->num_vf_mc_hashes > 30) ||
7319 (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
7320 vmolr |= E1000_VMOLR_MPME;
7321 } else if (vf_data->num_vf_mc_hashes) {
7322 vmolr |= E1000_VMOLR_ROMPE;
7323 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
7324 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
7325 }
7326 wr32(E1000_VMOLR(i), vmolr);
7327 }
7328}
7329
7330static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
7331{
7332 struct e1000_hw *hw = &adapter->hw;
7333 u32 pool_mask, vlvf_mask, i;
7334
7335 /* create mask for VF and other pools */
7336 pool_mask = E1000_VLVF_POOLSEL_MASK;
7337 vlvf_mask = BIT(E1000_VLVF_POOLSEL_SHIFT + vf);
7338
7339 /* drop PF from pool bits */
7340 pool_mask &= ~BIT(E1000_VLVF_POOLSEL_SHIFT +
7341 adapter->vfs_allocated_count);
7342
7343 /* Find the vlan filter for this id */
7344 for (i = E1000_VLVF_ARRAY_SIZE; i--;) {
7345 u32 vlvf = rd32(E1000_VLVF(i));
7346 u32 vfta_mask, vid, vfta;
7347
7348 /* remove the vf from the pool */
7349 if (!(vlvf & vlvf_mask))
7350 continue;
7351
7352 /* clear out bit from VLVF */
7353 vlvf ^= vlvf_mask;
7354
7355 /* if other pools are present, just remove ourselves */
7356 if (vlvf & pool_mask)
7357 goto update_vlvfb;
7358
7359 /* if PF is present, leave VFTA */
7360 if (vlvf & E1000_VLVF_POOLSEL_MASK)
7361 goto update_vlvf;
7362
7363 vid = vlvf & E1000_VLVF_VLANID_MASK;
7364 vfta_mask = BIT(vid % 32);
7365
7366 /* clear bit from VFTA */
7367 vfta = adapter->shadow_vfta[vid / 32];
7368 if (vfta & vfta_mask)
7369 hw->mac.ops.write_vfta(hw, vid / 32, vfta ^ vfta_mask);
7370update_vlvf:
7371 /* clear pool selection enable */
7372 if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
7373 vlvf &= E1000_VLVF_POOLSEL_MASK;
7374 else
7375 vlvf = 0;
7376update_vlvfb:
7377 /* clear pool bits */
7378 wr32(E1000_VLVF(i), vlvf);
7379 }
7380}
7381
7382static int igb_find_vlvf_entry(struct e1000_hw *hw, u32 vlan)
7383{
7384 u32 vlvf;
7385 int idx;
7386
7387 /* short cut the special case */
7388 if (vlan == 0)
7389 return 0;
7390
7391 /* Search for the VLAN id in the VLVF entries */
7392 for (idx = E1000_VLVF_ARRAY_SIZE; --idx;) {
7393 vlvf = rd32(E1000_VLVF(idx));
7394 if ((vlvf & VLAN_VID_MASK) == vlan)
7395 break;
7396 }
7397
7398 return idx;
7399}
7400
7401static void igb_update_pf_vlvf(struct igb_adapter *adapter, u32 vid)
7402{
7403 struct e1000_hw *hw = &adapter->hw;
7404 u32 bits, pf_id;
7405 int idx;
7406
7407 idx = igb_find_vlvf_entry(hw, vid);
7408 if (!idx)
7409 return;
7410
7411 /* See if any other pools are set for this VLAN filter
7412 * entry other than the PF.
7413 */
7414 pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
7415 bits = ~BIT(pf_id) & E1000_VLVF_POOLSEL_MASK;
7416 bits &= rd32(E1000_VLVF(idx));
7417
7418 /* Disable the filter so this falls into the default pool. */
7419 if (!bits) {
7420 if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
7421 wr32(E1000_VLVF(idx), BIT(pf_id));
7422 else
7423 wr32(E1000_VLVF(idx), 0);
7424 }
7425}
7426
7427static s32 igb_set_vf_vlan(struct igb_adapter *adapter, u32 vid,
7428 bool add, u32 vf)
7429{
7430 int pf_id = adapter->vfs_allocated_count;
7431 struct e1000_hw *hw = &adapter->hw;
7432 int err;
7433
7434 /* If VLAN overlaps with one the PF is currently monitoring make
7435 * sure that we are able to allocate a VLVF entry. This may be
7436 * redundant but it guarantees PF will maintain visibility to
7437 * the VLAN.
7438 */
7439 if (add && test_bit(vid, adapter->active_vlans)) {
7440 err = igb_vfta_set(hw, vid, pf_id, true, false);
7441 if (err)
7442 return err;
7443 }
7444
7445 err = igb_vfta_set(hw, vid, vf, add, false);
7446
7447 if (add && !err)
7448 return err;
7449
7450 /* If we failed to add the VF VLAN or we are removing the VF VLAN
7451 * we may need to drop the PF pool bit in order to allow us to free
7452 * up the VLVF resources.
7453 */
7454 if (test_bit(vid, adapter->active_vlans) ||
7455 (adapter->flags & IGB_FLAG_VLAN_PROMISC))
7456 igb_update_pf_vlvf(adapter, vid);
7457
7458 return err;
7459}
7460
7461static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
7462{
7463 struct e1000_hw *hw = &adapter->hw;
7464
7465 if (vid)
7466 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
7467 else
7468 wr32(E1000_VMVIR(vf), 0);
7469}
7470
7471static int igb_enable_port_vlan(struct igb_adapter *adapter, int vf,
7472 u16 vlan, u8 qos)
7473{
7474 int err;
7475
7476 err = igb_set_vf_vlan(adapter, vlan, true, vf);
7477 if (err)
7478 return err;
7479
7480 igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
7481 igb_set_vmolr(adapter, vf, !vlan);
7482
7483 /* revoke access to previous VLAN */
7484 if (vlan != adapter->vf_data[vf].pf_vlan)
7485 igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
7486 false, vf);
7487
7488 adapter->vf_data[vf].pf_vlan = vlan;
7489 adapter->vf_data[vf].pf_qos = qos;
7490 igb_set_vf_vlan_strip(adapter, vf, true);
7491 dev_info(&adapter->pdev->dev,
7492 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
7493 if (test_bit(__IGB_DOWN, &adapter->state)) {
7494 dev_warn(&adapter->pdev->dev,
7495 "The VF VLAN has been set, but the PF device is not up.\n");
7496 dev_warn(&adapter->pdev->dev,
7497 "Bring the PF device up before attempting to use the VF device.\n");
7498 }
7499
7500 return err;
7501}
7502
7503static int igb_disable_port_vlan(struct igb_adapter *adapter, int vf)
7504{
7505 /* Restore tagless access via VLAN 0 */
7506 igb_set_vf_vlan(adapter, 0, true, vf);
7507
7508 igb_set_vmvir(adapter, 0, vf);
7509 igb_set_vmolr(adapter, vf, true);
7510
7511 /* Remove any PF assigned VLAN */
7512 if (adapter->vf_data[vf].pf_vlan)
7513 igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
7514 false, vf);
7515
7516 adapter->vf_data[vf].pf_vlan = 0;
7517 adapter->vf_data[vf].pf_qos = 0;
7518 igb_set_vf_vlan_strip(adapter, vf, false);
7519
7520 return 0;
7521}
7522
7523static int igb_ndo_set_vf_vlan(struct net_device *netdev, int vf,
7524 u16 vlan, u8 qos, __be16 vlan_proto)
7525{
7526 struct igb_adapter *adapter = netdev_priv(netdev);
7527
7528 if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
7529 return -EINVAL;
7530
7531 if (vlan_proto != htons(ETH_P_8021Q))
7532 return -EPROTONOSUPPORT;
7533
7534 return (vlan || qos) ? igb_enable_port_vlan(adapter, vf, vlan, qos) :
7535 igb_disable_port_vlan(adapter, vf);
7536}
7537
7538static int igb_set_vf_vlan_msg(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
7539{
7540 int add = FIELD_GET(E1000_VT_MSGINFO_MASK, msgbuf[0]);
7541 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
7542 int ret;
7543
7544 if (adapter->vf_data[vf].pf_vlan)
7545 return -1;
7546
7547 /* VLAN 0 is a special case, don't allow it to be removed */
7548 if (!vid && !add)
7549 return 0;
7550
7551 ret = igb_set_vf_vlan(adapter, vid, !!add, vf);
7552 if (!ret)
7553 igb_set_vf_vlan_strip(adapter, vf, !!vid);
7554 return ret;
7555}
7556
7557static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
7558{
7559 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7560
7561 /* clear flags - except flag that indicates PF has set the MAC */
7562 vf_data->flags &= IGB_VF_FLAG_PF_SET_MAC;
7563 vf_data->last_nack = jiffies;
7564
7565 /* reset vlans for device */
7566 igb_clear_vf_vfta(adapter, vf);
7567 igb_set_vf_vlan(adapter, vf_data->pf_vlan, true, vf);
7568 igb_set_vmvir(adapter, vf_data->pf_vlan |
7569 (vf_data->pf_qos << VLAN_PRIO_SHIFT), vf);
7570 igb_set_vmolr(adapter, vf, !vf_data->pf_vlan);
7571 igb_set_vf_vlan_strip(adapter, vf, !!(vf_data->pf_vlan));
7572
7573 /* reset multicast table array for vf */
7574 adapter->vf_data[vf].num_vf_mc_hashes = 0;
7575
7576 /* Flush and reset the mta with the new values */
7577 igb_set_rx_mode(adapter->netdev);
7578}
7579
7580static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
7581{
7582 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
7583
7584 /* clear mac address as we were hotplug removed/added */
7585 if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
7586 eth_zero_addr(vf_mac);
7587
7588 /* process remaining reset events */
7589 igb_vf_reset(adapter, vf);
7590}
7591
7592static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
7593{
7594 struct e1000_hw *hw = &adapter->hw;
7595 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
7596 u32 reg, msgbuf[3] = {};
7597 u8 *addr = (u8 *)(&msgbuf[1]);
7598
7599 /* process all the same items cleared in a function level reset */
7600 igb_vf_reset(adapter, vf);
7601
7602 /* set vf mac address */
7603 igb_set_vf_mac(adapter, vf, vf_mac);
7604
7605 /* enable transmit and receive for vf */
7606 reg = rd32(E1000_VFTE);
7607 wr32(E1000_VFTE, reg | BIT(vf));
7608 reg = rd32(E1000_VFRE);
7609 wr32(E1000_VFRE, reg | BIT(vf));
7610
7611 adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
7612
7613 /* reply to reset with ack and vf mac address */
7614 if (!is_zero_ether_addr(vf_mac)) {
7615 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
7616 memcpy(addr, vf_mac, ETH_ALEN);
7617 } else {
7618 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_NACK;
7619 }
7620 igb_write_mbx(hw, msgbuf, 3, vf);
7621}
7622
7623static void igb_flush_mac_table(struct igb_adapter *adapter)
7624{
7625 struct e1000_hw *hw = &adapter->hw;
7626 int i;
7627
7628 for (i = 0; i < hw->mac.rar_entry_count; i++) {
7629 adapter->mac_table[i].state &= ~IGB_MAC_STATE_IN_USE;
7630 eth_zero_addr(adapter->mac_table[i].addr);
7631 adapter->mac_table[i].queue = 0;
7632 igb_rar_set_index(adapter, i);
7633 }
7634}
7635
7636static int igb_available_rars(struct igb_adapter *adapter, u8 queue)
7637{
7638 struct e1000_hw *hw = &adapter->hw;
7639 /* do not count rar entries reserved for VFs MAC addresses */
7640 int rar_entries = hw->mac.rar_entry_count -
7641 adapter->vfs_allocated_count;
7642 int i, count = 0;
7643
7644 for (i = 0; i < rar_entries; i++) {
7645 /* do not count default entries */
7646 if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT)
7647 continue;
7648
7649 /* do not count "in use" entries for different queues */
7650 if ((adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE) &&
7651 (adapter->mac_table[i].queue != queue))
7652 continue;
7653
7654 count++;
7655 }
7656
7657 return count;
7658}
7659
7660/* Set default MAC address for the PF in the first RAR entry */
7661static void igb_set_default_mac_filter(struct igb_adapter *adapter)
7662{
7663 struct igb_mac_addr *mac_table = &adapter->mac_table[0];
7664
7665 ether_addr_copy(mac_table->addr, adapter->hw.mac.addr);
7666 mac_table->queue = adapter->vfs_allocated_count;
7667 mac_table->state = IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE;
7668
7669 igb_rar_set_index(adapter, 0);
7670}
7671
7672/* If the filter to be added and an already existing filter express
7673 * the same address and address type, it should be possible to only
7674 * override the other configurations, for example the queue to steer
7675 * traffic.
7676 */
7677static bool igb_mac_entry_can_be_used(const struct igb_mac_addr *entry,
7678 const u8 *addr, const u8 flags)
7679{
7680 if (!(entry->state & IGB_MAC_STATE_IN_USE))
7681 return true;
7682
7683 if ((entry->state & IGB_MAC_STATE_SRC_ADDR) !=
7684 (flags & IGB_MAC_STATE_SRC_ADDR))
7685 return false;
7686
7687 if (!ether_addr_equal(addr, entry->addr))
7688 return false;
7689
7690 return true;
7691}
7692
7693/* Add a MAC filter for 'addr' directing matching traffic to 'queue',
7694 * 'flags' is used to indicate what kind of match is made, match is by
7695 * default for the destination address, if matching by source address
7696 * is desired the flag IGB_MAC_STATE_SRC_ADDR can be used.
7697 */
7698static int igb_add_mac_filter_flags(struct igb_adapter *adapter,
7699 const u8 *addr, const u8 queue,
7700 const u8 flags)
7701{
7702 struct e1000_hw *hw = &adapter->hw;
7703 int rar_entries = hw->mac.rar_entry_count -
7704 adapter->vfs_allocated_count;
7705 int i;
7706
7707 if (is_zero_ether_addr(addr))
7708 return -EINVAL;
7709
7710 /* Search for the first empty entry in the MAC table.
7711 * Do not touch entries at the end of the table reserved for the VF MAC
7712 * addresses.
7713 */
7714 for (i = 0; i < rar_entries; i++) {
7715 if (!igb_mac_entry_can_be_used(&adapter->mac_table[i],
7716 addr, flags))
7717 continue;
7718
7719 ether_addr_copy(adapter->mac_table[i].addr, addr);
7720 adapter->mac_table[i].queue = queue;
7721 adapter->mac_table[i].state |= IGB_MAC_STATE_IN_USE | flags;
7722
7723 igb_rar_set_index(adapter, i);
7724 return i;
7725 }
7726
7727 return -ENOSPC;
7728}
7729
7730static int igb_add_mac_filter(struct igb_adapter *adapter, const u8 *addr,
7731 const u8 queue)
7732{
7733 return igb_add_mac_filter_flags(adapter, addr, queue, 0);
7734}
7735
7736/* Remove a MAC filter for 'addr' directing matching traffic to
7737 * 'queue', 'flags' is used to indicate what kind of match need to be
7738 * removed, match is by default for the destination address, if
7739 * matching by source address is to be removed the flag
7740 * IGB_MAC_STATE_SRC_ADDR can be used.
7741 */
7742static int igb_del_mac_filter_flags(struct igb_adapter *adapter,
7743 const u8 *addr, const u8 queue,
7744 const u8 flags)
7745{
7746 struct e1000_hw *hw = &adapter->hw;
7747 int rar_entries = hw->mac.rar_entry_count -
7748 adapter->vfs_allocated_count;
7749 int i;
7750
7751 if (is_zero_ether_addr(addr))
7752 return -EINVAL;
7753
7754 /* Search for matching entry in the MAC table based on given address
7755 * and queue. Do not touch entries at the end of the table reserved
7756 * for the VF MAC addresses.
7757 */
7758 for (i = 0; i < rar_entries; i++) {
7759 if (!(adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE))
7760 continue;
7761 if ((adapter->mac_table[i].state & flags) != flags)
7762 continue;
7763 if (adapter->mac_table[i].queue != queue)
7764 continue;
7765 if (!ether_addr_equal(adapter->mac_table[i].addr, addr))
7766 continue;
7767
7768 /* When a filter for the default address is "deleted",
7769 * we return it to its initial configuration
7770 */
7771 if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT) {
7772 adapter->mac_table[i].state =
7773 IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE;
7774 adapter->mac_table[i].queue =
7775 adapter->vfs_allocated_count;
7776 } else {
7777 adapter->mac_table[i].state = 0;
7778 adapter->mac_table[i].queue = 0;
7779 eth_zero_addr(adapter->mac_table[i].addr);
7780 }
7781
7782 igb_rar_set_index(adapter, i);
7783 return 0;
7784 }
7785
7786 return -ENOENT;
7787}
7788
7789static int igb_del_mac_filter(struct igb_adapter *adapter, const u8 *addr,
7790 const u8 queue)
7791{
7792 return igb_del_mac_filter_flags(adapter, addr, queue, 0);
7793}
7794
7795int igb_add_mac_steering_filter(struct igb_adapter *adapter,
7796 const u8 *addr, u8 queue, u8 flags)
7797{
7798 struct e1000_hw *hw = &adapter->hw;
7799
7800 /* In theory, this should be supported on 82575 as well, but
7801 * that part wasn't easily accessible during development.
7802 */
7803 if (hw->mac.type != e1000_i210)
7804 return -EOPNOTSUPP;
7805
7806 return igb_add_mac_filter_flags(adapter, addr, queue,
7807 IGB_MAC_STATE_QUEUE_STEERING | flags);
7808}
7809
7810int igb_del_mac_steering_filter(struct igb_adapter *adapter,
7811 const u8 *addr, u8 queue, u8 flags)
7812{
7813 return igb_del_mac_filter_flags(adapter, addr, queue,
7814 IGB_MAC_STATE_QUEUE_STEERING | flags);
7815}
7816
7817static int igb_uc_sync(struct net_device *netdev, const unsigned char *addr)
7818{
7819 struct igb_adapter *adapter = netdev_priv(netdev);
7820 int ret;
7821
7822 ret = igb_add_mac_filter(adapter, addr, adapter->vfs_allocated_count);
7823
7824 return min_t(int, ret, 0);
7825}
7826
7827static int igb_uc_unsync(struct net_device *netdev, const unsigned char *addr)
7828{
7829 struct igb_adapter *adapter = netdev_priv(netdev);
7830
7831 igb_del_mac_filter(adapter, addr, adapter->vfs_allocated_count);
7832
7833 return 0;
7834}
7835
7836static int igb_set_vf_mac_filter(struct igb_adapter *adapter, const int vf,
7837 const u32 info, const u8 *addr)
7838{
7839 struct pci_dev *pdev = adapter->pdev;
7840 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7841 struct vf_mac_filter *entry;
7842 bool found = false;
7843 int ret = 0;
7844
7845 if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) &&
7846 !vf_data->trusted) {
7847 dev_warn(&pdev->dev,
7848 "VF %d requested MAC filter but is administratively denied\n",
7849 vf);
7850 return -EINVAL;
7851 }
7852 if (!is_valid_ether_addr(addr)) {
7853 dev_warn(&pdev->dev,
7854 "VF %d attempted to set invalid MAC filter\n",
7855 vf);
7856 return -EINVAL;
7857 }
7858
7859 switch (info) {
7860 case E1000_VF_MAC_FILTER_CLR:
7861 /* remove all unicast MAC filters related to the current VF */
7862 list_for_each_entry(entry, &adapter->vf_macs.l, l) {
7863 if (entry->vf == vf) {
7864 entry->vf = -1;
7865 entry->free = true;
7866 igb_del_mac_filter(adapter, entry->vf_mac, vf);
7867 }
7868 }
7869 break;
7870 case E1000_VF_MAC_FILTER_ADD:
7871 /* try to find empty slot in the list */
7872 list_for_each_entry(entry, &adapter->vf_macs.l, l) {
7873 if (entry->free) {
7874 found = true;
7875 break;
7876 }
7877 }
7878
7879 if (found) {
7880 entry->free = false;
7881 entry->vf = vf;
7882 ether_addr_copy(entry->vf_mac, addr);
7883
7884 ret = igb_add_mac_filter(adapter, addr, vf);
7885 ret = min_t(int, ret, 0);
7886 } else {
7887 ret = -ENOSPC;
7888 }
7889
7890 if (ret == -ENOSPC)
7891 dev_warn(&pdev->dev,
7892 "VF %d has requested MAC filter but there is no space for it\n",
7893 vf);
7894 break;
7895 default:
7896 ret = -EINVAL;
7897 break;
7898 }
7899
7900 return ret;
7901}
7902
7903static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
7904{
7905 struct pci_dev *pdev = adapter->pdev;
7906 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7907 u32 info = msg[0] & E1000_VT_MSGINFO_MASK;
7908
7909 /* The VF MAC Address is stored in a packed array of bytes
7910 * starting at the second 32 bit word of the msg array
7911 */
7912 unsigned char *addr = (unsigned char *)&msg[1];
7913 int ret = 0;
7914
7915 if (!info) {
7916 if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) &&
7917 !vf_data->trusted) {
7918 dev_warn(&pdev->dev,
7919 "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
7920 vf);
7921 return -EINVAL;
7922 }
7923
7924 if (!is_valid_ether_addr(addr)) {
7925 dev_warn(&pdev->dev,
7926 "VF %d attempted to set invalid MAC\n",
7927 vf);
7928 return -EINVAL;
7929 }
7930
7931 ret = igb_set_vf_mac(adapter, vf, addr);
7932 } else {
7933 ret = igb_set_vf_mac_filter(adapter, vf, info, addr);
7934 }
7935
7936 return ret;
7937}
7938
7939static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
7940{
7941 struct e1000_hw *hw = &adapter->hw;
7942 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7943 u32 msg = E1000_VT_MSGTYPE_NACK;
7944
7945 /* if device isn't clear to send it shouldn't be reading either */
7946 if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
7947 time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
7948 igb_write_mbx(hw, &msg, 1, vf);
7949 vf_data->last_nack = jiffies;
7950 }
7951}
7952
7953static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
7954{
7955 struct pci_dev *pdev = adapter->pdev;
7956 u32 msgbuf[E1000_VFMAILBOX_SIZE];
7957 struct e1000_hw *hw = &adapter->hw;
7958 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7959 s32 retval;
7960
7961 retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf, false);
7962
7963 if (retval) {
7964 /* if receive failed revoke VF CTS stats and restart init */
7965 dev_err(&pdev->dev, "Error receiving message from VF\n");
7966 vf_data->flags &= ~IGB_VF_FLAG_CTS;
7967 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
7968 goto unlock;
7969 goto out;
7970 }
7971
7972 /* this is a message we already processed, do nothing */
7973 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
7974 goto unlock;
7975
7976 /* until the vf completes a reset it should not be
7977 * allowed to start any configuration.
7978 */
7979 if (msgbuf[0] == E1000_VF_RESET) {
7980 /* unlocks mailbox */
7981 igb_vf_reset_msg(adapter, vf);
7982 return;
7983 }
7984
7985 if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
7986 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
7987 goto unlock;
7988 retval = -1;
7989 goto out;
7990 }
7991
7992 switch ((msgbuf[0] & 0xFFFF)) {
7993 case E1000_VF_SET_MAC_ADDR:
7994 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
7995 break;
7996 case E1000_VF_SET_PROMISC:
7997 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
7998 break;
7999 case E1000_VF_SET_MULTICAST:
8000 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
8001 break;
8002 case E1000_VF_SET_LPE:
8003 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
8004 break;
8005 case E1000_VF_SET_VLAN:
8006 retval = -1;
8007 if (vf_data->pf_vlan)
8008 dev_warn(&pdev->dev,
8009 "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
8010 vf);
8011 else
8012 retval = igb_set_vf_vlan_msg(adapter, msgbuf, vf);
8013 break;
8014 default:
8015 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
8016 retval = -1;
8017 break;
8018 }
8019
8020 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
8021out:
8022 /* notify the VF of the results of what it sent us */
8023 if (retval)
8024 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
8025 else
8026 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
8027
8028 /* unlocks mailbox */
8029 igb_write_mbx(hw, msgbuf, 1, vf);
8030 return;
8031
8032unlock:
8033 igb_unlock_mbx(hw, vf);
8034}
8035
8036static void igb_msg_task(struct igb_adapter *adapter)
8037{
8038 struct e1000_hw *hw = &adapter->hw;
8039 unsigned long flags;
8040 u32 vf;
8041
8042 spin_lock_irqsave(&adapter->vfs_lock, flags);
8043 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
8044 /* process any reset requests */
8045 if (!igb_check_for_rst(hw, vf))
8046 igb_vf_reset_event(adapter, vf);
8047
8048 /* process any messages pending */
8049 if (!igb_check_for_msg(hw, vf))
8050 igb_rcv_msg_from_vf(adapter, vf);
8051
8052 /* process any acks */
8053 if (!igb_check_for_ack(hw, vf))
8054 igb_rcv_ack_from_vf(adapter, vf);
8055 }
8056 spin_unlock_irqrestore(&adapter->vfs_lock, flags);
8057}
8058
8059/**
8060 * igb_set_uta - Set unicast filter table address
8061 * @adapter: board private structure
8062 * @set: boolean indicating if we are setting or clearing bits
8063 *
8064 * The unicast table address is a register array of 32-bit registers.
8065 * The table is meant to be used in a way similar to how the MTA is used
8066 * however due to certain limitations in the hardware it is necessary to
8067 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
8068 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
8069 **/
8070static void igb_set_uta(struct igb_adapter *adapter, bool set)
8071{
8072 struct e1000_hw *hw = &adapter->hw;
8073 u32 uta = set ? ~0 : 0;
8074 int i;
8075
8076 /* we only need to do this if VMDq is enabled */
8077 if (!adapter->vfs_allocated_count)
8078 return;
8079
8080 for (i = hw->mac.uta_reg_count; i--;)
8081 array_wr32(E1000_UTA, i, uta);
8082}
8083
8084/**
8085 * igb_intr_msi - Interrupt Handler
8086 * @irq: interrupt number
8087 * @data: pointer to a network interface device structure
8088 **/
8089static irqreturn_t igb_intr_msi(int irq, void *data)
8090{
8091 struct igb_adapter *adapter = data;
8092 struct igb_q_vector *q_vector = adapter->q_vector[0];
8093 struct e1000_hw *hw = &adapter->hw;
8094 /* read ICR disables interrupts using IAM */
8095 u32 icr = rd32(E1000_ICR);
8096
8097 igb_write_itr(q_vector);
8098
8099 if (icr & E1000_ICR_DRSTA)
8100 schedule_work(&adapter->reset_task);
8101
8102 if (icr & E1000_ICR_DOUTSYNC) {
8103 /* HW is reporting DMA is out of sync */
8104 adapter->stats.doosync++;
8105 }
8106
8107 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
8108 hw->mac.get_link_status = 1;
8109 if (!test_bit(__IGB_DOWN, &adapter->state))
8110 mod_timer(&adapter->watchdog_timer, jiffies + 1);
8111 }
8112
8113 if (icr & E1000_ICR_TS)
8114 igb_tsync_interrupt(adapter);
8115
8116 napi_schedule(&q_vector->napi);
8117
8118 return IRQ_HANDLED;
8119}
8120
8121/**
8122 * igb_intr - Legacy Interrupt Handler
8123 * @irq: interrupt number
8124 * @data: pointer to a network interface device structure
8125 **/
8126static irqreturn_t igb_intr(int irq, void *data)
8127{
8128 struct igb_adapter *adapter = data;
8129 struct igb_q_vector *q_vector = adapter->q_vector[0];
8130 struct e1000_hw *hw = &adapter->hw;
8131 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
8132 * need for the IMC write
8133 */
8134 u32 icr = rd32(E1000_ICR);
8135
8136 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
8137 * not set, then the adapter didn't send an interrupt
8138 */
8139 if (!(icr & E1000_ICR_INT_ASSERTED))
8140 return IRQ_NONE;
8141
8142 igb_write_itr(q_vector);
8143
8144 if (icr & E1000_ICR_DRSTA)
8145 schedule_work(&adapter->reset_task);
8146
8147 if (icr & E1000_ICR_DOUTSYNC) {
8148 /* HW is reporting DMA is out of sync */
8149 adapter->stats.doosync++;
8150 }
8151
8152 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
8153 hw->mac.get_link_status = 1;
8154 /* guard against interrupt when we're going down */
8155 if (!test_bit(__IGB_DOWN, &adapter->state))
8156 mod_timer(&adapter->watchdog_timer, jiffies + 1);
8157 }
8158
8159 if (icr & E1000_ICR_TS)
8160 igb_tsync_interrupt(adapter);
8161
8162 napi_schedule(&q_vector->napi);
8163
8164 return IRQ_HANDLED;
8165}
8166
8167static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
8168{
8169 struct igb_adapter *adapter = q_vector->adapter;
8170 struct e1000_hw *hw = &adapter->hw;
8171
8172 if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
8173 (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
8174 if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
8175 igb_set_itr(q_vector);
8176 else
8177 igb_update_ring_itr(q_vector);
8178 }
8179
8180 if (!test_bit(__IGB_DOWN, &adapter->state)) {
8181 if (adapter->flags & IGB_FLAG_HAS_MSIX)
8182 wr32(E1000_EIMS, q_vector->eims_value);
8183 else
8184 igb_irq_enable(adapter);
8185 }
8186}
8187
8188/**
8189 * igb_poll - NAPI Rx polling callback
8190 * @napi: napi polling structure
8191 * @budget: count of how many packets we should handle
8192 **/
8193static int igb_poll(struct napi_struct *napi, int budget)
8194{
8195 struct igb_q_vector *q_vector = container_of(napi,
8196 struct igb_q_vector,
8197 napi);
8198 bool clean_complete = true;
8199 int work_done = 0;
8200
8201#ifdef CONFIG_IGB_DCA
8202 if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
8203 igb_update_dca(q_vector);
8204#endif
8205 if (q_vector->tx.ring)
8206 clean_complete = igb_clean_tx_irq(q_vector, budget);
8207
8208 if (q_vector->rx.ring) {
8209 int cleaned = igb_clean_rx_irq(q_vector, budget);
8210
8211 work_done += cleaned;
8212 if (cleaned >= budget)
8213 clean_complete = false;
8214 }
8215
8216 /* If all work not completed, return budget and keep polling */
8217 if (!clean_complete)
8218 return budget;
8219
8220 /* Exit the polling mode, but don't re-enable interrupts if stack might
8221 * poll us due to busy-polling
8222 */
8223 if (likely(napi_complete_done(napi, work_done)))
8224 igb_ring_irq_enable(q_vector);
8225
8226 return work_done;
8227}
8228
8229/**
8230 * igb_clean_tx_irq - Reclaim resources after transmit completes
8231 * @q_vector: pointer to q_vector containing needed info
8232 * @napi_budget: Used to determine if we are in netpoll
8233 *
8234 * returns true if ring is completely cleaned
8235 **/
8236static bool igb_clean_tx_irq(struct igb_q_vector *q_vector, int napi_budget)
8237{
8238 struct igb_adapter *adapter = q_vector->adapter;
8239 struct igb_ring *tx_ring = q_vector->tx.ring;
8240 struct igb_tx_buffer *tx_buffer;
8241 union e1000_adv_tx_desc *tx_desc;
8242 unsigned int total_bytes = 0, total_packets = 0;
8243 unsigned int budget = q_vector->tx.work_limit;
8244 unsigned int i = tx_ring->next_to_clean;
8245
8246 if (test_bit(__IGB_DOWN, &adapter->state))
8247 return true;
8248
8249 tx_buffer = &tx_ring->tx_buffer_info[i];
8250 tx_desc = IGB_TX_DESC(tx_ring, i);
8251 i -= tx_ring->count;
8252
8253 do {
8254 union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
8255
8256 /* if next_to_watch is not set then there is no work pending */
8257 if (!eop_desc)
8258 break;
8259
8260 /* prevent any other reads prior to eop_desc */
8261 smp_rmb();
8262
8263 /* if DD is not set pending work has not been completed */
8264 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
8265 break;
8266
8267 /* clear next_to_watch to prevent false hangs */
8268 tx_buffer->next_to_watch = NULL;
8269
8270 /* update the statistics for this packet */
8271 total_bytes += tx_buffer->bytecount;
8272 total_packets += tx_buffer->gso_segs;
8273
8274 /* free the skb */
8275 if (tx_buffer->type == IGB_TYPE_SKB)
8276 napi_consume_skb(tx_buffer->skb, napi_budget);
8277 else
8278 xdp_return_frame(tx_buffer->xdpf);
8279
8280 /* unmap skb header data */
8281 dma_unmap_single(tx_ring->dev,
8282 dma_unmap_addr(tx_buffer, dma),
8283 dma_unmap_len(tx_buffer, len),
8284 DMA_TO_DEVICE);
8285
8286 /* clear tx_buffer data */
8287 dma_unmap_len_set(tx_buffer, len, 0);
8288
8289 /* clear last DMA location and unmap remaining buffers */
8290 while (tx_desc != eop_desc) {
8291 tx_buffer++;
8292 tx_desc++;
8293 i++;
8294 if (unlikely(!i)) {
8295 i -= tx_ring->count;
8296 tx_buffer = tx_ring->tx_buffer_info;
8297 tx_desc = IGB_TX_DESC(tx_ring, 0);
8298 }
8299
8300 /* unmap any remaining paged data */
8301 if (dma_unmap_len(tx_buffer, len)) {
8302 dma_unmap_page(tx_ring->dev,
8303 dma_unmap_addr(tx_buffer, dma),
8304 dma_unmap_len(tx_buffer, len),
8305 DMA_TO_DEVICE);
8306 dma_unmap_len_set(tx_buffer, len, 0);
8307 }
8308 }
8309
8310 /* move us one more past the eop_desc for start of next pkt */
8311 tx_buffer++;
8312 tx_desc++;
8313 i++;
8314 if (unlikely(!i)) {
8315 i -= tx_ring->count;
8316 tx_buffer = tx_ring->tx_buffer_info;
8317 tx_desc = IGB_TX_DESC(tx_ring, 0);
8318 }
8319
8320 /* issue prefetch for next Tx descriptor */
8321 prefetch(tx_desc);
8322
8323 /* update budget accounting */
8324 budget--;
8325 } while (likely(budget));
8326
8327 netdev_tx_completed_queue(txring_txq(tx_ring),
8328 total_packets, total_bytes);
8329 i += tx_ring->count;
8330 tx_ring->next_to_clean = i;
8331 u64_stats_update_begin(&tx_ring->tx_syncp);
8332 tx_ring->tx_stats.bytes += total_bytes;
8333 tx_ring->tx_stats.packets += total_packets;
8334 u64_stats_update_end(&tx_ring->tx_syncp);
8335 q_vector->tx.total_bytes += total_bytes;
8336 q_vector->tx.total_packets += total_packets;
8337
8338 if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
8339 struct e1000_hw *hw = &adapter->hw;
8340
8341 /* Detect a transmit hang in hardware, this serializes the
8342 * check with the clearing of time_stamp and movement of i
8343 */
8344 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
8345 if (tx_buffer->next_to_watch &&
8346 time_after(jiffies, tx_buffer->time_stamp +
8347 (adapter->tx_timeout_factor * HZ)) &&
8348 !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
8349
8350 /* detected Tx unit hang */
8351 dev_err(tx_ring->dev,
8352 "Detected Tx Unit Hang\n"
8353 " Tx Queue <%d>\n"
8354 " TDH <%x>\n"
8355 " TDT <%x>\n"
8356 " next_to_use <%x>\n"
8357 " next_to_clean <%x>\n"
8358 "buffer_info[next_to_clean]\n"
8359 " time_stamp <%lx>\n"
8360 " next_to_watch <%p>\n"
8361 " jiffies <%lx>\n"
8362 " desc.status <%x>\n",
8363 tx_ring->queue_index,
8364 rd32(E1000_TDH(tx_ring->reg_idx)),
8365 readl(tx_ring->tail),
8366 tx_ring->next_to_use,
8367 tx_ring->next_to_clean,
8368 tx_buffer->time_stamp,
8369 tx_buffer->next_to_watch,
8370 jiffies,
8371 tx_buffer->next_to_watch->wb.status);
8372 netif_stop_subqueue(tx_ring->netdev,
8373 tx_ring->queue_index);
8374
8375 /* we are about to reset, no point in enabling stuff */
8376 return true;
8377 }
8378 }
8379
8380#define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
8381 if (unlikely(total_packets &&
8382 netif_carrier_ok(tx_ring->netdev) &&
8383 igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
8384 /* Make sure that anybody stopping the queue after this
8385 * sees the new next_to_clean.
8386 */
8387 smp_mb();
8388 if (__netif_subqueue_stopped(tx_ring->netdev,
8389 tx_ring->queue_index) &&
8390 !(test_bit(__IGB_DOWN, &adapter->state))) {
8391 netif_wake_subqueue(tx_ring->netdev,
8392 tx_ring->queue_index);
8393
8394 u64_stats_update_begin(&tx_ring->tx_syncp);
8395 tx_ring->tx_stats.restart_queue++;
8396 u64_stats_update_end(&tx_ring->tx_syncp);
8397 }
8398 }
8399
8400 return !!budget;
8401}
8402
8403/**
8404 * igb_reuse_rx_page - page flip buffer and store it back on the ring
8405 * @rx_ring: rx descriptor ring to store buffers on
8406 * @old_buff: donor buffer to have page reused
8407 *
8408 * Synchronizes page for reuse by the adapter
8409 **/
8410static void igb_reuse_rx_page(struct igb_ring *rx_ring,
8411 struct igb_rx_buffer *old_buff)
8412{
8413 struct igb_rx_buffer *new_buff;
8414 u16 nta = rx_ring->next_to_alloc;
8415
8416 new_buff = &rx_ring->rx_buffer_info[nta];
8417
8418 /* update, and store next to alloc */
8419 nta++;
8420 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
8421
8422 /* Transfer page from old buffer to new buffer.
8423 * Move each member individually to avoid possible store
8424 * forwarding stalls.
8425 */
8426 new_buff->dma = old_buff->dma;
8427 new_buff->page = old_buff->page;
8428 new_buff->page_offset = old_buff->page_offset;
8429 new_buff->pagecnt_bias = old_buff->pagecnt_bias;
8430}
8431
8432static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer,
8433 int rx_buf_pgcnt)
8434{
8435 unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
8436 struct page *page = rx_buffer->page;
8437
8438 /* avoid re-using remote and pfmemalloc pages */
8439 if (!dev_page_is_reusable(page))
8440 return false;
8441
8442#if (PAGE_SIZE < 8192)
8443 /* if we are only owner of page we can reuse it */
8444 if (unlikely((rx_buf_pgcnt - pagecnt_bias) > 1))
8445 return false;
8446#else
8447#define IGB_LAST_OFFSET \
8448 (SKB_WITH_OVERHEAD(PAGE_SIZE) - IGB_RXBUFFER_2048)
8449
8450 if (rx_buffer->page_offset > IGB_LAST_OFFSET)
8451 return false;
8452#endif
8453
8454 /* If we have drained the page fragment pool we need to update
8455 * the pagecnt_bias and page count so that we fully restock the
8456 * number of references the driver holds.
8457 */
8458 if (unlikely(pagecnt_bias == 1)) {
8459 page_ref_add(page, USHRT_MAX - 1);
8460 rx_buffer->pagecnt_bias = USHRT_MAX;
8461 }
8462
8463 return true;
8464}
8465
8466/**
8467 * igb_add_rx_frag - Add contents of Rx buffer to sk_buff
8468 * @rx_ring: rx descriptor ring to transact packets on
8469 * @rx_buffer: buffer containing page to add
8470 * @skb: sk_buff to place the data into
8471 * @size: size of buffer to be added
8472 *
8473 * This function will add the data contained in rx_buffer->page to the skb.
8474 **/
8475static void igb_add_rx_frag(struct igb_ring *rx_ring,
8476 struct igb_rx_buffer *rx_buffer,
8477 struct sk_buff *skb,
8478 unsigned int size)
8479{
8480#if (PAGE_SIZE < 8192)
8481 unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
8482#else
8483 unsigned int truesize = ring_uses_build_skb(rx_ring) ?
8484 SKB_DATA_ALIGN(IGB_SKB_PAD + size) :
8485 SKB_DATA_ALIGN(size);
8486#endif
8487 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
8488 rx_buffer->page_offset, size, truesize);
8489#if (PAGE_SIZE < 8192)
8490 rx_buffer->page_offset ^= truesize;
8491#else
8492 rx_buffer->page_offset += truesize;
8493#endif
8494}
8495
8496static struct sk_buff *igb_construct_skb(struct igb_ring *rx_ring,
8497 struct igb_rx_buffer *rx_buffer,
8498 struct xdp_buff *xdp,
8499 ktime_t timestamp)
8500{
8501#if (PAGE_SIZE < 8192)
8502 unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
8503#else
8504 unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
8505 xdp->data_hard_start);
8506#endif
8507 unsigned int size = xdp->data_end - xdp->data;
8508 unsigned int headlen;
8509 struct sk_buff *skb;
8510
8511 /* prefetch first cache line of first page */
8512 net_prefetch(xdp->data);
8513
8514 /* allocate a skb to store the frags */
8515 skb = napi_alloc_skb(&rx_ring->q_vector->napi, IGB_RX_HDR_LEN);
8516 if (unlikely(!skb))
8517 return NULL;
8518
8519 if (timestamp)
8520 skb_hwtstamps(skb)->hwtstamp = timestamp;
8521
8522 /* Determine available headroom for copy */
8523 headlen = size;
8524 if (headlen > IGB_RX_HDR_LEN)
8525 headlen = eth_get_headlen(skb->dev, xdp->data, IGB_RX_HDR_LEN);
8526
8527 /* align pull length to size of long to optimize memcpy performance */
8528 memcpy(__skb_put(skb, headlen), xdp->data, ALIGN(headlen, sizeof(long)));
8529
8530 /* update all of the pointers */
8531 size -= headlen;
8532 if (size) {
8533 skb_add_rx_frag(skb, 0, rx_buffer->page,
8534 (xdp->data + headlen) - page_address(rx_buffer->page),
8535 size, truesize);
8536#if (PAGE_SIZE < 8192)
8537 rx_buffer->page_offset ^= truesize;
8538#else
8539 rx_buffer->page_offset += truesize;
8540#endif
8541 } else {
8542 rx_buffer->pagecnt_bias++;
8543 }
8544
8545 return skb;
8546}
8547
8548static struct sk_buff *igb_build_skb(struct igb_ring *rx_ring,
8549 struct igb_rx_buffer *rx_buffer,
8550 struct xdp_buff *xdp,
8551 ktime_t timestamp)
8552{
8553#if (PAGE_SIZE < 8192)
8554 unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
8555#else
8556 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
8557 SKB_DATA_ALIGN(xdp->data_end -
8558 xdp->data_hard_start);
8559#endif
8560 unsigned int metasize = xdp->data - xdp->data_meta;
8561 struct sk_buff *skb;
8562
8563 /* prefetch first cache line of first page */
8564 net_prefetch(xdp->data_meta);
8565
8566 /* build an skb around the page buffer */
8567 skb = napi_build_skb(xdp->data_hard_start, truesize);
8568 if (unlikely(!skb))
8569 return NULL;
8570
8571 /* update pointers within the skb to store the data */
8572 skb_reserve(skb, xdp->data - xdp->data_hard_start);
8573 __skb_put(skb, xdp->data_end - xdp->data);
8574
8575 if (metasize)
8576 skb_metadata_set(skb, metasize);
8577
8578 if (timestamp)
8579 skb_hwtstamps(skb)->hwtstamp = timestamp;
8580
8581 /* update buffer offset */
8582#if (PAGE_SIZE < 8192)
8583 rx_buffer->page_offset ^= truesize;
8584#else
8585 rx_buffer->page_offset += truesize;
8586#endif
8587
8588 return skb;
8589}
8590
8591static struct sk_buff *igb_run_xdp(struct igb_adapter *adapter,
8592 struct igb_ring *rx_ring,
8593 struct xdp_buff *xdp)
8594{
8595 int err, result = IGB_XDP_PASS;
8596 struct bpf_prog *xdp_prog;
8597 u32 act;
8598
8599 xdp_prog = READ_ONCE(rx_ring->xdp_prog);
8600
8601 if (!xdp_prog)
8602 goto xdp_out;
8603
8604 prefetchw(xdp->data_hard_start); /* xdp_frame write */
8605
8606 act = bpf_prog_run_xdp(xdp_prog, xdp);
8607 switch (act) {
8608 case XDP_PASS:
8609 break;
8610 case XDP_TX:
8611 result = igb_xdp_xmit_back(adapter, xdp);
8612 if (result == IGB_XDP_CONSUMED)
8613 goto out_failure;
8614 break;
8615 case XDP_REDIRECT:
8616 err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
8617 if (err)
8618 goto out_failure;
8619 result = IGB_XDP_REDIR;
8620 break;
8621 default:
8622 bpf_warn_invalid_xdp_action(adapter->netdev, xdp_prog, act);
8623 fallthrough;
8624 case XDP_ABORTED:
8625out_failure:
8626 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
8627 fallthrough;
8628 case XDP_DROP:
8629 result = IGB_XDP_CONSUMED;
8630 break;
8631 }
8632xdp_out:
8633 return ERR_PTR(-result);
8634}
8635
8636static unsigned int igb_rx_frame_truesize(struct igb_ring *rx_ring,
8637 unsigned int size)
8638{
8639 unsigned int truesize;
8640
8641#if (PAGE_SIZE < 8192)
8642 truesize = igb_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
8643#else
8644 truesize = ring_uses_build_skb(rx_ring) ?
8645 SKB_DATA_ALIGN(IGB_SKB_PAD + size) +
8646 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
8647 SKB_DATA_ALIGN(size);
8648#endif
8649 return truesize;
8650}
8651
8652static void igb_rx_buffer_flip(struct igb_ring *rx_ring,
8653 struct igb_rx_buffer *rx_buffer,
8654 unsigned int size)
8655{
8656 unsigned int truesize = igb_rx_frame_truesize(rx_ring, size);
8657#if (PAGE_SIZE < 8192)
8658 rx_buffer->page_offset ^= truesize;
8659#else
8660 rx_buffer->page_offset += truesize;
8661#endif
8662}
8663
8664static inline void igb_rx_checksum(struct igb_ring *ring,
8665 union e1000_adv_rx_desc *rx_desc,
8666 struct sk_buff *skb)
8667{
8668 skb_checksum_none_assert(skb);
8669
8670 /* Ignore Checksum bit is set */
8671 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
8672 return;
8673
8674 /* Rx checksum disabled via ethtool */
8675 if (!(ring->netdev->features & NETIF_F_RXCSUM))
8676 return;
8677
8678 /* TCP/UDP checksum error bit is set */
8679 if (igb_test_staterr(rx_desc,
8680 E1000_RXDEXT_STATERR_TCPE |
8681 E1000_RXDEXT_STATERR_IPE)) {
8682 /* work around errata with sctp packets where the TCPE aka
8683 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
8684 * packets, (aka let the stack check the crc32c)
8685 */
8686 if (!((skb->len == 60) &&
8687 test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
8688 u64_stats_update_begin(&ring->rx_syncp);
8689 ring->rx_stats.csum_err++;
8690 u64_stats_update_end(&ring->rx_syncp);
8691 }
8692 /* let the stack verify checksum errors */
8693 return;
8694 }
8695 /* It must be a TCP or UDP packet with a valid checksum */
8696 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
8697 E1000_RXD_STAT_UDPCS))
8698 skb->ip_summed = CHECKSUM_UNNECESSARY;
8699
8700 dev_dbg(ring->dev, "cksum success: bits %08X\n",
8701 le32_to_cpu(rx_desc->wb.upper.status_error));
8702}
8703
8704static inline void igb_rx_hash(struct igb_ring *ring,
8705 union e1000_adv_rx_desc *rx_desc,
8706 struct sk_buff *skb)
8707{
8708 if (ring->netdev->features & NETIF_F_RXHASH)
8709 skb_set_hash(skb,
8710 le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
8711 PKT_HASH_TYPE_L3);
8712}
8713
8714/**
8715 * igb_is_non_eop - process handling of non-EOP buffers
8716 * @rx_ring: Rx ring being processed
8717 * @rx_desc: Rx descriptor for current buffer
8718 *
8719 * This function updates next to clean. If the buffer is an EOP buffer
8720 * this function exits returning false, otherwise it will place the
8721 * sk_buff in the next buffer to be chained and return true indicating
8722 * that this is in fact a non-EOP buffer.
8723 **/
8724static bool igb_is_non_eop(struct igb_ring *rx_ring,
8725 union e1000_adv_rx_desc *rx_desc)
8726{
8727 u32 ntc = rx_ring->next_to_clean + 1;
8728
8729 /* fetch, update, and store next to clean */
8730 ntc = (ntc < rx_ring->count) ? ntc : 0;
8731 rx_ring->next_to_clean = ntc;
8732
8733 prefetch(IGB_RX_DESC(rx_ring, ntc));
8734
8735 if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
8736 return false;
8737
8738 return true;
8739}
8740
8741/**
8742 * igb_cleanup_headers - Correct corrupted or empty headers
8743 * @rx_ring: rx descriptor ring packet is being transacted on
8744 * @rx_desc: pointer to the EOP Rx descriptor
8745 * @skb: pointer to current skb being fixed
8746 *
8747 * Address the case where we are pulling data in on pages only
8748 * and as such no data is present in the skb header.
8749 *
8750 * In addition if skb is not at least 60 bytes we need to pad it so that
8751 * it is large enough to qualify as a valid Ethernet frame.
8752 *
8753 * Returns true if an error was encountered and skb was freed.
8754 **/
8755static bool igb_cleanup_headers(struct igb_ring *rx_ring,
8756 union e1000_adv_rx_desc *rx_desc,
8757 struct sk_buff *skb)
8758{
8759 /* XDP packets use error pointer so abort at this point */
8760 if (IS_ERR(skb))
8761 return true;
8762
8763 if (unlikely((igb_test_staterr(rx_desc,
8764 E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
8765 struct net_device *netdev = rx_ring->netdev;
8766 if (!(netdev->features & NETIF_F_RXALL)) {
8767 dev_kfree_skb_any(skb);
8768 return true;
8769 }
8770 }
8771
8772 /* if eth_skb_pad returns an error the skb was freed */
8773 if (eth_skb_pad(skb))
8774 return true;
8775
8776 return false;
8777}
8778
8779/**
8780 * igb_process_skb_fields - Populate skb header fields from Rx descriptor
8781 * @rx_ring: rx descriptor ring packet is being transacted on
8782 * @rx_desc: pointer to the EOP Rx descriptor
8783 * @skb: pointer to current skb being populated
8784 *
8785 * This function checks the ring, descriptor, and packet information in
8786 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
8787 * other fields within the skb.
8788 **/
8789static void igb_process_skb_fields(struct igb_ring *rx_ring,
8790 union e1000_adv_rx_desc *rx_desc,
8791 struct sk_buff *skb)
8792{
8793 struct net_device *dev = rx_ring->netdev;
8794
8795 igb_rx_hash(rx_ring, rx_desc, skb);
8796
8797 igb_rx_checksum(rx_ring, rx_desc, skb);
8798
8799 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TS) &&
8800 !igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP))
8801 igb_ptp_rx_rgtstamp(rx_ring->q_vector, skb);
8802
8803 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
8804 igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
8805 u16 vid;
8806
8807 if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
8808 test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
8809 vid = be16_to_cpu((__force __be16)rx_desc->wb.upper.vlan);
8810 else
8811 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
8812
8813 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
8814 }
8815
8816 skb_record_rx_queue(skb, rx_ring->queue_index);
8817
8818 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
8819}
8820
8821static unsigned int igb_rx_offset(struct igb_ring *rx_ring)
8822{
8823 return ring_uses_build_skb(rx_ring) ? IGB_SKB_PAD : 0;
8824}
8825
8826static struct igb_rx_buffer *igb_get_rx_buffer(struct igb_ring *rx_ring,
8827 const unsigned int size, int *rx_buf_pgcnt)
8828{
8829 struct igb_rx_buffer *rx_buffer;
8830
8831 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
8832 *rx_buf_pgcnt =
8833#if (PAGE_SIZE < 8192)
8834 page_count(rx_buffer->page);
8835#else
8836 0;
8837#endif
8838 prefetchw(rx_buffer->page);
8839
8840 /* we are reusing so sync this buffer for CPU use */
8841 dma_sync_single_range_for_cpu(rx_ring->dev,
8842 rx_buffer->dma,
8843 rx_buffer->page_offset,
8844 size,
8845 DMA_FROM_DEVICE);
8846
8847 rx_buffer->pagecnt_bias--;
8848
8849 return rx_buffer;
8850}
8851
8852static void igb_put_rx_buffer(struct igb_ring *rx_ring,
8853 struct igb_rx_buffer *rx_buffer, int rx_buf_pgcnt)
8854{
8855 if (igb_can_reuse_rx_page(rx_buffer, rx_buf_pgcnt)) {
8856 /* hand second half of page back to the ring */
8857 igb_reuse_rx_page(rx_ring, rx_buffer);
8858 } else {
8859 /* We are not reusing the buffer so unmap it and free
8860 * any references we are holding to it
8861 */
8862 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
8863 igb_rx_pg_size(rx_ring), DMA_FROM_DEVICE,
8864 IGB_RX_DMA_ATTR);
8865 __page_frag_cache_drain(rx_buffer->page,
8866 rx_buffer->pagecnt_bias);
8867 }
8868
8869 /* clear contents of rx_buffer */
8870 rx_buffer->page = NULL;
8871}
8872
8873static int igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
8874{
8875 unsigned int total_bytes = 0, total_packets = 0;
8876 struct igb_adapter *adapter = q_vector->adapter;
8877 struct igb_ring *rx_ring = q_vector->rx.ring;
8878 u16 cleaned_count = igb_desc_unused(rx_ring);
8879 struct sk_buff *skb = rx_ring->skb;
8880 int cpu = smp_processor_id();
8881 unsigned int xdp_xmit = 0;
8882 struct netdev_queue *nq;
8883 struct xdp_buff xdp;
8884 u32 frame_sz = 0;
8885 int rx_buf_pgcnt;
8886
8887 /* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
8888#if (PAGE_SIZE < 8192)
8889 frame_sz = igb_rx_frame_truesize(rx_ring, 0);
8890#endif
8891 xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
8892
8893 while (likely(total_packets < budget)) {
8894 union e1000_adv_rx_desc *rx_desc;
8895 struct igb_rx_buffer *rx_buffer;
8896 ktime_t timestamp = 0;
8897 int pkt_offset = 0;
8898 unsigned int size;
8899 void *pktbuf;
8900
8901 /* return some buffers to hardware, one at a time is too slow */
8902 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
8903 igb_alloc_rx_buffers(rx_ring, cleaned_count);
8904 cleaned_count = 0;
8905 }
8906
8907 rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
8908 size = le16_to_cpu(rx_desc->wb.upper.length);
8909 if (!size)
8910 break;
8911
8912 /* This memory barrier is needed to keep us from reading
8913 * any other fields out of the rx_desc until we know the
8914 * descriptor has been written back
8915 */
8916 dma_rmb();
8917
8918 rx_buffer = igb_get_rx_buffer(rx_ring, size, &rx_buf_pgcnt);
8919 pktbuf = page_address(rx_buffer->page) + rx_buffer->page_offset;
8920
8921 /* pull rx packet timestamp if available and valid */
8922 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
8923 int ts_hdr_len;
8924
8925 ts_hdr_len = igb_ptp_rx_pktstamp(rx_ring->q_vector,
8926 pktbuf, ×tamp);
8927
8928 pkt_offset += ts_hdr_len;
8929 size -= ts_hdr_len;
8930 }
8931
8932 /* retrieve a buffer from the ring */
8933 if (!skb) {
8934 unsigned char *hard_start = pktbuf - igb_rx_offset(rx_ring);
8935 unsigned int offset = pkt_offset + igb_rx_offset(rx_ring);
8936
8937 xdp_prepare_buff(&xdp, hard_start, offset, size, true);
8938 xdp_buff_clear_frags_flag(&xdp);
8939#if (PAGE_SIZE > 4096)
8940 /* At larger PAGE_SIZE, frame_sz depend on len size */
8941 xdp.frame_sz = igb_rx_frame_truesize(rx_ring, size);
8942#endif
8943 skb = igb_run_xdp(adapter, rx_ring, &xdp);
8944 }
8945
8946 if (IS_ERR(skb)) {
8947 unsigned int xdp_res = -PTR_ERR(skb);
8948
8949 if (xdp_res & (IGB_XDP_TX | IGB_XDP_REDIR)) {
8950 xdp_xmit |= xdp_res;
8951 igb_rx_buffer_flip(rx_ring, rx_buffer, size);
8952 } else {
8953 rx_buffer->pagecnt_bias++;
8954 }
8955 total_packets++;
8956 total_bytes += size;
8957 } else if (skb)
8958 igb_add_rx_frag(rx_ring, rx_buffer, skb, size);
8959 else if (ring_uses_build_skb(rx_ring))
8960 skb = igb_build_skb(rx_ring, rx_buffer, &xdp,
8961 timestamp);
8962 else
8963 skb = igb_construct_skb(rx_ring, rx_buffer,
8964 &xdp, timestamp);
8965
8966 /* exit if we failed to retrieve a buffer */
8967 if (!skb) {
8968 rx_ring->rx_stats.alloc_failed++;
8969 rx_buffer->pagecnt_bias++;
8970 break;
8971 }
8972
8973 igb_put_rx_buffer(rx_ring, rx_buffer, rx_buf_pgcnt);
8974 cleaned_count++;
8975
8976 /* fetch next buffer in frame if non-eop */
8977 if (igb_is_non_eop(rx_ring, rx_desc))
8978 continue;
8979
8980 /* verify the packet layout is correct */
8981 if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
8982 skb = NULL;
8983 continue;
8984 }
8985
8986 /* probably a little skewed due to removing CRC */
8987 total_bytes += skb->len;
8988
8989 /* populate checksum, timestamp, VLAN, and protocol */
8990 igb_process_skb_fields(rx_ring, rx_desc, skb);
8991
8992 napi_gro_receive(&q_vector->napi, skb);
8993
8994 /* reset skb pointer */
8995 skb = NULL;
8996
8997 /* update budget accounting */
8998 total_packets++;
8999 }
9000
9001 /* place incomplete frames back on ring for completion */
9002 rx_ring->skb = skb;
9003
9004 if (xdp_xmit & IGB_XDP_REDIR)
9005 xdp_do_flush();
9006
9007 if (xdp_xmit & IGB_XDP_TX) {
9008 struct igb_ring *tx_ring = igb_xdp_tx_queue_mapping(adapter);
9009
9010 nq = txring_txq(tx_ring);
9011 __netif_tx_lock(nq, cpu);
9012 igb_xdp_ring_update_tail(tx_ring);
9013 __netif_tx_unlock(nq);
9014 }
9015
9016 u64_stats_update_begin(&rx_ring->rx_syncp);
9017 rx_ring->rx_stats.packets += total_packets;
9018 rx_ring->rx_stats.bytes += total_bytes;
9019 u64_stats_update_end(&rx_ring->rx_syncp);
9020 q_vector->rx.total_packets += total_packets;
9021 q_vector->rx.total_bytes += total_bytes;
9022
9023 if (cleaned_count)
9024 igb_alloc_rx_buffers(rx_ring, cleaned_count);
9025
9026 return total_packets;
9027}
9028
9029static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
9030 struct igb_rx_buffer *bi)
9031{
9032 struct page *page = bi->page;
9033 dma_addr_t dma;
9034
9035 /* since we are recycling buffers we should seldom need to alloc */
9036 if (likely(page))
9037 return true;
9038
9039 /* alloc new page for storage */
9040 page = dev_alloc_pages(igb_rx_pg_order(rx_ring));
9041 if (unlikely(!page)) {
9042 rx_ring->rx_stats.alloc_failed++;
9043 return false;
9044 }
9045
9046 /* map page for use */
9047 dma = dma_map_page_attrs(rx_ring->dev, page, 0,
9048 igb_rx_pg_size(rx_ring),
9049 DMA_FROM_DEVICE,
9050 IGB_RX_DMA_ATTR);
9051
9052 /* if mapping failed free memory back to system since
9053 * there isn't much point in holding memory we can't use
9054 */
9055 if (dma_mapping_error(rx_ring->dev, dma)) {
9056 __free_pages(page, igb_rx_pg_order(rx_ring));
9057
9058 rx_ring->rx_stats.alloc_failed++;
9059 return false;
9060 }
9061
9062 bi->dma = dma;
9063 bi->page = page;
9064 bi->page_offset = igb_rx_offset(rx_ring);
9065 page_ref_add(page, USHRT_MAX - 1);
9066 bi->pagecnt_bias = USHRT_MAX;
9067
9068 return true;
9069}
9070
9071/**
9072 * igb_alloc_rx_buffers - Replace used receive buffers
9073 * @rx_ring: rx descriptor ring to allocate new receive buffers
9074 * @cleaned_count: count of buffers to allocate
9075 **/
9076void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
9077{
9078 union e1000_adv_rx_desc *rx_desc;
9079 struct igb_rx_buffer *bi;
9080 u16 i = rx_ring->next_to_use;
9081 u16 bufsz;
9082
9083 /* nothing to do */
9084 if (!cleaned_count)
9085 return;
9086
9087 rx_desc = IGB_RX_DESC(rx_ring, i);
9088 bi = &rx_ring->rx_buffer_info[i];
9089 i -= rx_ring->count;
9090
9091 bufsz = igb_rx_bufsz(rx_ring);
9092
9093 do {
9094 if (!igb_alloc_mapped_page(rx_ring, bi))
9095 break;
9096
9097 /* sync the buffer for use by the device */
9098 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
9099 bi->page_offset, bufsz,
9100 DMA_FROM_DEVICE);
9101
9102 /* Refresh the desc even if buffer_addrs didn't change
9103 * because each write-back erases this info.
9104 */
9105 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
9106
9107 rx_desc++;
9108 bi++;
9109 i++;
9110 if (unlikely(!i)) {
9111 rx_desc = IGB_RX_DESC(rx_ring, 0);
9112 bi = rx_ring->rx_buffer_info;
9113 i -= rx_ring->count;
9114 }
9115
9116 /* clear the length for the next_to_use descriptor */
9117 rx_desc->wb.upper.length = 0;
9118
9119 cleaned_count--;
9120 } while (cleaned_count);
9121
9122 i += rx_ring->count;
9123
9124 if (rx_ring->next_to_use != i) {
9125 /* record the next descriptor to use */
9126 rx_ring->next_to_use = i;
9127
9128 /* update next to alloc since we have filled the ring */
9129 rx_ring->next_to_alloc = i;
9130
9131 /* Force memory writes to complete before letting h/w
9132 * know there are new descriptors to fetch. (Only
9133 * applicable for weak-ordered memory model archs,
9134 * such as IA-64).
9135 */
9136 dma_wmb();
9137 writel(i, rx_ring->tail);
9138 }
9139}
9140
9141/**
9142 * igb_mii_ioctl -
9143 * @netdev: pointer to netdev struct
9144 * @ifr: interface structure
9145 * @cmd: ioctl command to execute
9146 **/
9147static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
9148{
9149 struct igb_adapter *adapter = netdev_priv(netdev);
9150 struct mii_ioctl_data *data = if_mii(ifr);
9151
9152 if (adapter->hw.phy.media_type != e1000_media_type_copper)
9153 return -EOPNOTSUPP;
9154
9155 switch (cmd) {
9156 case SIOCGMIIPHY:
9157 data->phy_id = adapter->hw.phy.addr;
9158 break;
9159 case SIOCGMIIREG:
9160 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
9161 &data->val_out))
9162 return -EIO;
9163 break;
9164 case SIOCSMIIREG:
9165 if (igb_write_phy_reg(&adapter->hw, data->reg_num & 0x1F,
9166 data->val_in))
9167 return -EIO;
9168 break;
9169 default:
9170 return -EOPNOTSUPP;
9171 }
9172 return 0;
9173}
9174
9175/**
9176 * igb_ioctl -
9177 * @netdev: pointer to netdev struct
9178 * @ifr: interface structure
9179 * @cmd: ioctl command to execute
9180 **/
9181static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
9182{
9183 switch (cmd) {
9184 case SIOCGMIIPHY:
9185 case SIOCGMIIREG:
9186 case SIOCSMIIREG:
9187 return igb_mii_ioctl(netdev, ifr, cmd);
9188 case SIOCGHWTSTAMP:
9189 return igb_ptp_get_ts_config(netdev, ifr);
9190 case SIOCSHWTSTAMP:
9191 return igb_ptp_set_ts_config(netdev, ifr);
9192 default:
9193 return -EOPNOTSUPP;
9194 }
9195}
9196
9197void igb_read_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
9198{
9199 struct igb_adapter *adapter = hw->back;
9200
9201 pci_read_config_word(adapter->pdev, reg, value);
9202}
9203
9204void igb_write_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
9205{
9206 struct igb_adapter *adapter = hw->back;
9207
9208 pci_write_config_word(adapter->pdev, reg, *value);
9209}
9210
9211s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
9212{
9213 struct igb_adapter *adapter = hw->back;
9214
9215 if (pcie_capability_read_word(adapter->pdev, reg, value))
9216 return -E1000_ERR_CONFIG;
9217
9218 return 0;
9219}
9220
9221s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
9222{
9223 struct igb_adapter *adapter = hw->back;
9224
9225 if (pcie_capability_write_word(adapter->pdev, reg, *value))
9226 return -E1000_ERR_CONFIG;
9227
9228 return 0;
9229}
9230
9231static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
9232{
9233 struct igb_adapter *adapter = netdev_priv(netdev);
9234 struct e1000_hw *hw = &adapter->hw;
9235 u32 ctrl, rctl;
9236 bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
9237
9238 if (enable) {
9239 /* enable VLAN tag insert/strip */
9240 ctrl = rd32(E1000_CTRL);
9241 ctrl |= E1000_CTRL_VME;
9242 wr32(E1000_CTRL, ctrl);
9243
9244 /* Disable CFI check */
9245 rctl = rd32(E1000_RCTL);
9246 rctl &= ~E1000_RCTL_CFIEN;
9247 wr32(E1000_RCTL, rctl);
9248 } else {
9249 /* disable VLAN tag insert/strip */
9250 ctrl = rd32(E1000_CTRL);
9251 ctrl &= ~E1000_CTRL_VME;
9252 wr32(E1000_CTRL, ctrl);
9253 }
9254
9255 igb_set_vf_vlan_strip(adapter, adapter->vfs_allocated_count, enable);
9256}
9257
9258static int igb_vlan_rx_add_vid(struct net_device *netdev,
9259 __be16 proto, u16 vid)
9260{
9261 struct igb_adapter *adapter = netdev_priv(netdev);
9262 struct e1000_hw *hw = &adapter->hw;
9263 int pf_id = adapter->vfs_allocated_count;
9264
9265 /* add the filter since PF can receive vlans w/o entry in vlvf */
9266 if (!vid || !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
9267 igb_vfta_set(hw, vid, pf_id, true, !!vid);
9268
9269 set_bit(vid, adapter->active_vlans);
9270
9271 return 0;
9272}
9273
9274static int igb_vlan_rx_kill_vid(struct net_device *netdev,
9275 __be16 proto, u16 vid)
9276{
9277 struct igb_adapter *adapter = netdev_priv(netdev);
9278 int pf_id = adapter->vfs_allocated_count;
9279 struct e1000_hw *hw = &adapter->hw;
9280
9281 /* remove VID from filter table */
9282 if (vid && !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
9283 igb_vfta_set(hw, vid, pf_id, false, true);
9284
9285 clear_bit(vid, adapter->active_vlans);
9286
9287 return 0;
9288}
9289
9290static void igb_restore_vlan(struct igb_adapter *adapter)
9291{
9292 u16 vid = 1;
9293
9294 igb_vlan_mode(adapter->netdev, adapter->netdev->features);
9295 igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
9296
9297 for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
9298 igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
9299}
9300
9301int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
9302{
9303 struct pci_dev *pdev = adapter->pdev;
9304 struct e1000_mac_info *mac = &adapter->hw.mac;
9305
9306 mac->autoneg = 0;
9307
9308 /* Make sure dplx is at most 1 bit and lsb of speed is not set
9309 * for the switch() below to work
9310 */
9311 if ((spd & 1) || (dplx & ~1))
9312 goto err_inval;
9313
9314 /* Fiber NIC's only allow 1000 gbps Full duplex
9315 * and 100Mbps Full duplex for 100baseFx sfp
9316 */
9317 if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
9318 switch (spd + dplx) {
9319 case SPEED_10 + DUPLEX_HALF:
9320 case SPEED_10 + DUPLEX_FULL:
9321 case SPEED_100 + DUPLEX_HALF:
9322 goto err_inval;
9323 default:
9324 break;
9325 }
9326 }
9327
9328 switch (spd + dplx) {
9329 case SPEED_10 + DUPLEX_HALF:
9330 mac->forced_speed_duplex = ADVERTISE_10_HALF;
9331 break;
9332 case SPEED_10 + DUPLEX_FULL:
9333 mac->forced_speed_duplex = ADVERTISE_10_FULL;
9334 break;
9335 case SPEED_100 + DUPLEX_HALF:
9336 mac->forced_speed_duplex = ADVERTISE_100_HALF;
9337 break;
9338 case SPEED_100 + DUPLEX_FULL:
9339 mac->forced_speed_duplex = ADVERTISE_100_FULL;
9340 break;
9341 case SPEED_1000 + DUPLEX_FULL:
9342 mac->autoneg = 1;
9343 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
9344 break;
9345 case SPEED_1000 + DUPLEX_HALF: /* not supported */
9346 default:
9347 goto err_inval;
9348 }
9349
9350 /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
9351 adapter->hw.phy.mdix = AUTO_ALL_MODES;
9352
9353 return 0;
9354
9355err_inval:
9356 dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
9357 return -EINVAL;
9358}
9359
9360static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
9361 bool runtime)
9362{
9363 struct net_device *netdev = pci_get_drvdata(pdev);
9364 struct igb_adapter *adapter = netdev_priv(netdev);
9365 struct e1000_hw *hw = &adapter->hw;
9366 u32 ctrl, rctl, status;
9367 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
9368 bool wake;
9369
9370 rtnl_lock();
9371 netif_device_detach(netdev);
9372
9373 if (netif_running(netdev))
9374 __igb_close(netdev, true);
9375
9376 igb_ptp_suspend(adapter);
9377
9378 igb_clear_interrupt_scheme(adapter);
9379 rtnl_unlock();
9380
9381 status = rd32(E1000_STATUS);
9382 if (status & E1000_STATUS_LU)
9383 wufc &= ~E1000_WUFC_LNKC;
9384
9385 if (wufc) {
9386 igb_setup_rctl(adapter);
9387 igb_set_rx_mode(netdev);
9388
9389 /* turn on all-multi mode if wake on multicast is enabled */
9390 if (wufc & E1000_WUFC_MC) {
9391 rctl = rd32(E1000_RCTL);
9392 rctl |= E1000_RCTL_MPE;
9393 wr32(E1000_RCTL, rctl);
9394 }
9395
9396 ctrl = rd32(E1000_CTRL);
9397 ctrl |= E1000_CTRL_ADVD3WUC;
9398 wr32(E1000_CTRL, ctrl);
9399
9400 /* Allow time for pending master requests to run */
9401 igb_disable_pcie_master(hw);
9402
9403 wr32(E1000_WUC, E1000_WUC_PME_EN);
9404 wr32(E1000_WUFC, wufc);
9405 } else {
9406 wr32(E1000_WUC, 0);
9407 wr32(E1000_WUFC, 0);
9408 }
9409
9410 wake = wufc || adapter->en_mng_pt;
9411 if (!wake)
9412 igb_power_down_link(adapter);
9413 else
9414 igb_power_up_link(adapter);
9415
9416 if (enable_wake)
9417 *enable_wake = wake;
9418
9419 /* Release control of h/w to f/w. If f/w is AMT enabled, this
9420 * would have already happened in close and is redundant.
9421 */
9422 igb_release_hw_control(adapter);
9423
9424 pci_disable_device(pdev);
9425
9426 return 0;
9427}
9428
9429static void igb_deliver_wake_packet(struct net_device *netdev)
9430{
9431 struct igb_adapter *adapter = netdev_priv(netdev);
9432 struct e1000_hw *hw = &adapter->hw;
9433 struct sk_buff *skb;
9434 u32 wupl;
9435
9436 wupl = rd32(E1000_WUPL) & E1000_WUPL_MASK;
9437
9438 /* WUPM stores only the first 128 bytes of the wake packet.
9439 * Read the packet only if we have the whole thing.
9440 */
9441 if ((wupl == 0) || (wupl > E1000_WUPM_BYTES))
9442 return;
9443
9444 skb = netdev_alloc_skb_ip_align(netdev, E1000_WUPM_BYTES);
9445 if (!skb)
9446 return;
9447
9448 skb_put(skb, wupl);
9449
9450 /* Ensure reads are 32-bit aligned */
9451 wupl = roundup(wupl, 4);
9452
9453 memcpy_fromio(skb->data, hw->hw_addr + E1000_WUPM_REG(0), wupl);
9454
9455 skb->protocol = eth_type_trans(skb, netdev);
9456 netif_rx(skb);
9457}
9458
9459static int igb_suspend(struct device *dev)
9460{
9461 return __igb_shutdown(to_pci_dev(dev), NULL, 0);
9462}
9463
9464static int __igb_resume(struct device *dev, bool rpm)
9465{
9466 struct pci_dev *pdev = to_pci_dev(dev);
9467 struct net_device *netdev = pci_get_drvdata(pdev);
9468 struct igb_adapter *adapter = netdev_priv(netdev);
9469 struct e1000_hw *hw = &adapter->hw;
9470 u32 err, val;
9471
9472 pci_set_power_state(pdev, PCI_D0);
9473 pci_restore_state(pdev);
9474 pci_save_state(pdev);
9475
9476 if (!pci_device_is_present(pdev))
9477 return -ENODEV;
9478 err = pci_enable_device_mem(pdev);
9479 if (err) {
9480 dev_err(&pdev->dev,
9481 "igb: Cannot enable PCI device from suspend\n");
9482 return err;
9483 }
9484 pci_set_master(pdev);
9485
9486 pci_enable_wake(pdev, PCI_D3hot, 0);
9487 pci_enable_wake(pdev, PCI_D3cold, 0);
9488
9489 if (igb_init_interrupt_scheme(adapter, true)) {
9490 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
9491 return -ENOMEM;
9492 }
9493
9494 igb_reset(adapter);
9495
9496 /* let the f/w know that the h/w is now under the control of the
9497 * driver.
9498 */
9499 igb_get_hw_control(adapter);
9500
9501 val = rd32(E1000_WUS);
9502 if (val & WAKE_PKT_WUS)
9503 igb_deliver_wake_packet(netdev);
9504
9505 wr32(E1000_WUS, ~0);
9506
9507 if (!rpm)
9508 rtnl_lock();
9509 if (!err && netif_running(netdev))
9510 err = __igb_open(netdev, true);
9511
9512 if (!err)
9513 netif_device_attach(netdev);
9514 if (!rpm)
9515 rtnl_unlock();
9516
9517 return err;
9518}
9519
9520static int igb_resume(struct device *dev)
9521{
9522 return __igb_resume(dev, false);
9523}
9524
9525static int igb_runtime_idle(struct device *dev)
9526{
9527 struct net_device *netdev = dev_get_drvdata(dev);
9528 struct igb_adapter *adapter = netdev_priv(netdev);
9529
9530 if (!igb_has_link(adapter))
9531 pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
9532
9533 return -EBUSY;
9534}
9535
9536static int igb_runtime_suspend(struct device *dev)
9537{
9538 return __igb_shutdown(to_pci_dev(dev), NULL, 1);
9539}
9540
9541static int igb_runtime_resume(struct device *dev)
9542{
9543 return __igb_resume(dev, true);
9544}
9545
9546static void igb_shutdown(struct pci_dev *pdev)
9547{
9548 bool wake;
9549
9550 __igb_shutdown(pdev, &wake, 0);
9551
9552 if (system_state == SYSTEM_POWER_OFF) {
9553 pci_wake_from_d3(pdev, wake);
9554 pci_set_power_state(pdev, PCI_D3hot);
9555 }
9556}
9557
9558static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
9559{
9560#ifdef CONFIG_PCI_IOV
9561 int err;
9562
9563 if (num_vfs == 0) {
9564 return igb_disable_sriov(dev, true);
9565 } else {
9566 err = igb_enable_sriov(dev, num_vfs, true);
9567 return err ? err : num_vfs;
9568 }
9569#endif
9570 return 0;
9571}
9572
9573/**
9574 * igb_io_error_detected - called when PCI error is detected
9575 * @pdev: Pointer to PCI device
9576 * @state: The current pci connection state
9577 *
9578 * This function is called after a PCI bus error affecting
9579 * this device has been detected.
9580 **/
9581static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
9582 pci_channel_state_t state)
9583{
9584 struct net_device *netdev = pci_get_drvdata(pdev);
9585 struct igb_adapter *adapter = netdev_priv(netdev);
9586
9587 if (state == pci_channel_io_normal) {
9588 dev_warn(&pdev->dev, "Non-correctable non-fatal error reported.\n");
9589 return PCI_ERS_RESULT_CAN_RECOVER;
9590 }
9591
9592 netif_device_detach(netdev);
9593
9594 if (state == pci_channel_io_perm_failure)
9595 return PCI_ERS_RESULT_DISCONNECT;
9596
9597 if (netif_running(netdev))
9598 igb_down(adapter);
9599 pci_disable_device(pdev);
9600
9601 /* Request a slot reset. */
9602 return PCI_ERS_RESULT_NEED_RESET;
9603}
9604
9605/**
9606 * igb_io_slot_reset - called after the pci bus has been reset.
9607 * @pdev: Pointer to PCI device
9608 *
9609 * Restart the card from scratch, as if from a cold-boot. Implementation
9610 * resembles the first-half of the __igb_resume routine.
9611 **/
9612static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
9613{
9614 struct net_device *netdev = pci_get_drvdata(pdev);
9615 struct igb_adapter *adapter = netdev_priv(netdev);
9616 struct e1000_hw *hw = &adapter->hw;
9617 pci_ers_result_t result;
9618
9619 if (pci_enable_device_mem(pdev)) {
9620 dev_err(&pdev->dev,
9621 "Cannot re-enable PCI device after reset.\n");
9622 result = PCI_ERS_RESULT_DISCONNECT;
9623 } else {
9624 pci_set_master(pdev);
9625 pci_restore_state(pdev);
9626 pci_save_state(pdev);
9627
9628 pci_enable_wake(pdev, PCI_D3hot, 0);
9629 pci_enable_wake(pdev, PCI_D3cold, 0);
9630
9631 /* In case of PCI error, adapter lose its HW address
9632 * so we should re-assign it here.
9633 */
9634 hw->hw_addr = adapter->io_addr;
9635
9636 igb_reset(adapter);
9637 wr32(E1000_WUS, ~0);
9638 result = PCI_ERS_RESULT_RECOVERED;
9639 }
9640
9641 return result;
9642}
9643
9644/**
9645 * igb_io_resume - called when traffic can start flowing again.
9646 * @pdev: Pointer to PCI device
9647 *
9648 * This callback is called when the error recovery driver tells us that
9649 * its OK to resume normal operation. Implementation resembles the
9650 * second-half of the __igb_resume routine.
9651 */
9652static void igb_io_resume(struct pci_dev *pdev)
9653{
9654 struct net_device *netdev = pci_get_drvdata(pdev);
9655 struct igb_adapter *adapter = netdev_priv(netdev);
9656
9657 if (netif_running(netdev)) {
9658 if (!test_bit(__IGB_DOWN, &adapter->state)) {
9659 dev_dbg(&pdev->dev, "Resuming from non-fatal error, do nothing.\n");
9660 return;
9661 }
9662 if (igb_up(adapter)) {
9663 dev_err(&pdev->dev, "igb_up failed after reset\n");
9664 return;
9665 }
9666 }
9667
9668 netif_device_attach(netdev);
9669
9670 /* let the f/w know that the h/w is now under the control of the
9671 * driver.
9672 */
9673 igb_get_hw_control(adapter);
9674}
9675
9676/**
9677 * igb_rar_set_index - Sync RAL[index] and RAH[index] registers with MAC table
9678 * @adapter: Pointer to adapter structure
9679 * @index: Index of the RAR entry which need to be synced with MAC table
9680 **/
9681static void igb_rar_set_index(struct igb_adapter *adapter, u32 index)
9682{
9683 struct e1000_hw *hw = &adapter->hw;
9684 u32 rar_low, rar_high;
9685 u8 *addr = adapter->mac_table[index].addr;
9686
9687 /* HW expects these to be in network order when they are plugged
9688 * into the registers which are little endian. In order to guarantee
9689 * that ordering we need to do an leXX_to_cpup here in order to be
9690 * ready for the byteswap that occurs with writel
9691 */
9692 rar_low = le32_to_cpup((__le32 *)(addr));
9693 rar_high = le16_to_cpup((__le16 *)(addr + 4));
9694
9695 /* Indicate to hardware the Address is Valid. */
9696 if (adapter->mac_table[index].state & IGB_MAC_STATE_IN_USE) {
9697 if (is_valid_ether_addr(addr))
9698 rar_high |= E1000_RAH_AV;
9699
9700 if (adapter->mac_table[index].state & IGB_MAC_STATE_SRC_ADDR)
9701 rar_high |= E1000_RAH_ASEL_SRC_ADDR;
9702
9703 switch (hw->mac.type) {
9704 case e1000_82575:
9705 case e1000_i210:
9706 if (adapter->mac_table[index].state &
9707 IGB_MAC_STATE_QUEUE_STEERING)
9708 rar_high |= E1000_RAH_QSEL_ENABLE;
9709
9710 rar_high |= E1000_RAH_POOL_1 *
9711 adapter->mac_table[index].queue;
9712 break;
9713 default:
9714 rar_high |= E1000_RAH_POOL_1 <<
9715 adapter->mac_table[index].queue;
9716 break;
9717 }
9718 }
9719
9720 wr32(E1000_RAL(index), rar_low);
9721 wrfl();
9722 wr32(E1000_RAH(index), rar_high);
9723 wrfl();
9724}
9725
9726static int igb_set_vf_mac(struct igb_adapter *adapter,
9727 int vf, unsigned char *mac_addr)
9728{
9729 struct e1000_hw *hw = &adapter->hw;
9730 /* VF MAC addresses start at end of receive addresses and moves
9731 * towards the first, as a result a collision should not be possible
9732 */
9733 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
9734 unsigned char *vf_mac_addr = adapter->vf_data[vf].vf_mac_addresses;
9735
9736 ether_addr_copy(vf_mac_addr, mac_addr);
9737 ether_addr_copy(adapter->mac_table[rar_entry].addr, mac_addr);
9738 adapter->mac_table[rar_entry].queue = vf;
9739 adapter->mac_table[rar_entry].state |= IGB_MAC_STATE_IN_USE;
9740 igb_rar_set_index(adapter, rar_entry);
9741
9742 return 0;
9743}
9744
9745static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
9746{
9747 struct igb_adapter *adapter = netdev_priv(netdev);
9748
9749 if (vf >= adapter->vfs_allocated_count)
9750 return -EINVAL;
9751
9752 /* Setting the VF MAC to 0 reverts the IGB_VF_FLAG_PF_SET_MAC
9753 * flag and allows to overwrite the MAC via VF netdev. This
9754 * is necessary to allow libvirt a way to restore the original
9755 * MAC after unbinding vfio-pci and reloading igbvf after shutting
9756 * down a VM.
9757 */
9758 if (is_zero_ether_addr(mac)) {
9759 adapter->vf_data[vf].flags &= ~IGB_VF_FLAG_PF_SET_MAC;
9760 dev_info(&adapter->pdev->dev,
9761 "remove administratively set MAC on VF %d\n",
9762 vf);
9763 } else if (is_valid_ether_addr(mac)) {
9764 adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
9765 dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n",
9766 mac, vf);
9767 dev_info(&adapter->pdev->dev,
9768 "Reload the VF driver to make this change effective.");
9769 /* Generate additional warning if PF is down */
9770 if (test_bit(__IGB_DOWN, &adapter->state)) {
9771 dev_warn(&adapter->pdev->dev,
9772 "The VF MAC address has been set, but the PF device is not up.\n");
9773 dev_warn(&adapter->pdev->dev,
9774 "Bring the PF device up before attempting to use the VF device.\n");
9775 }
9776 } else {
9777 return -EINVAL;
9778 }
9779 return igb_set_vf_mac(adapter, vf, mac);
9780}
9781
9782static int igb_link_mbps(int internal_link_speed)
9783{
9784 switch (internal_link_speed) {
9785 case SPEED_100:
9786 return 100;
9787 case SPEED_1000:
9788 return 1000;
9789 default:
9790 return 0;
9791 }
9792}
9793
9794static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
9795 int link_speed)
9796{
9797 int rf_dec, rf_int;
9798 u32 bcnrc_val;
9799
9800 if (tx_rate != 0) {
9801 /* Calculate the rate factor values to set */
9802 rf_int = link_speed / tx_rate;
9803 rf_dec = (link_speed - (rf_int * tx_rate));
9804 rf_dec = (rf_dec * BIT(E1000_RTTBCNRC_RF_INT_SHIFT)) /
9805 tx_rate;
9806
9807 bcnrc_val = E1000_RTTBCNRC_RS_ENA;
9808 bcnrc_val |= FIELD_PREP(E1000_RTTBCNRC_RF_INT_MASK, rf_int);
9809 bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
9810 } else {
9811 bcnrc_val = 0;
9812 }
9813
9814 wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
9815 /* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
9816 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
9817 */
9818 wr32(E1000_RTTBCNRM, 0x14);
9819 wr32(E1000_RTTBCNRC, bcnrc_val);
9820}
9821
9822static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
9823{
9824 int actual_link_speed, i;
9825 bool reset_rate = false;
9826
9827 /* VF TX rate limit was not set or not supported */
9828 if ((adapter->vf_rate_link_speed == 0) ||
9829 (adapter->hw.mac.type != e1000_82576))
9830 return;
9831
9832 actual_link_speed = igb_link_mbps(adapter->link_speed);
9833 if (actual_link_speed != adapter->vf_rate_link_speed) {
9834 reset_rate = true;
9835 adapter->vf_rate_link_speed = 0;
9836 dev_info(&adapter->pdev->dev,
9837 "Link speed has been changed. VF Transmit rate is disabled\n");
9838 }
9839
9840 for (i = 0; i < adapter->vfs_allocated_count; i++) {
9841 if (reset_rate)
9842 adapter->vf_data[i].tx_rate = 0;
9843
9844 igb_set_vf_rate_limit(&adapter->hw, i,
9845 adapter->vf_data[i].tx_rate,
9846 actual_link_speed);
9847 }
9848}
9849
9850static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf,
9851 int min_tx_rate, int max_tx_rate)
9852{
9853 struct igb_adapter *adapter = netdev_priv(netdev);
9854 struct e1000_hw *hw = &adapter->hw;
9855 int actual_link_speed;
9856
9857 if (hw->mac.type != e1000_82576)
9858 return -EOPNOTSUPP;
9859
9860 if (min_tx_rate)
9861 return -EINVAL;
9862
9863 actual_link_speed = igb_link_mbps(adapter->link_speed);
9864 if ((vf >= adapter->vfs_allocated_count) ||
9865 (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
9866 (max_tx_rate < 0) ||
9867 (max_tx_rate > actual_link_speed))
9868 return -EINVAL;
9869
9870 adapter->vf_rate_link_speed = actual_link_speed;
9871 adapter->vf_data[vf].tx_rate = (u16)max_tx_rate;
9872 igb_set_vf_rate_limit(hw, vf, max_tx_rate, actual_link_speed);
9873
9874 return 0;
9875}
9876
9877static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
9878 bool setting)
9879{
9880 struct igb_adapter *adapter = netdev_priv(netdev);
9881 struct e1000_hw *hw = &adapter->hw;
9882 u32 reg_val, reg_offset;
9883
9884 if (!adapter->vfs_allocated_count)
9885 return -EOPNOTSUPP;
9886
9887 if (vf >= adapter->vfs_allocated_count)
9888 return -EINVAL;
9889
9890 reg_offset = (hw->mac.type == e1000_82576) ? E1000_DTXSWC : E1000_TXSWC;
9891 reg_val = rd32(reg_offset);
9892 if (setting)
9893 reg_val |= (BIT(vf) |
9894 BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
9895 else
9896 reg_val &= ~(BIT(vf) |
9897 BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
9898 wr32(reg_offset, reg_val);
9899
9900 adapter->vf_data[vf].spoofchk_enabled = setting;
9901 return 0;
9902}
9903
9904static int igb_ndo_set_vf_trust(struct net_device *netdev, int vf, bool setting)
9905{
9906 struct igb_adapter *adapter = netdev_priv(netdev);
9907
9908 if (vf >= adapter->vfs_allocated_count)
9909 return -EINVAL;
9910 if (adapter->vf_data[vf].trusted == setting)
9911 return 0;
9912
9913 adapter->vf_data[vf].trusted = setting;
9914
9915 dev_info(&adapter->pdev->dev, "VF %u is %strusted\n",
9916 vf, setting ? "" : "not ");
9917 return 0;
9918}
9919
9920static int igb_ndo_get_vf_config(struct net_device *netdev,
9921 int vf, struct ifla_vf_info *ivi)
9922{
9923 struct igb_adapter *adapter = netdev_priv(netdev);
9924 if (vf >= adapter->vfs_allocated_count)
9925 return -EINVAL;
9926 ivi->vf = vf;
9927 memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
9928 ivi->max_tx_rate = adapter->vf_data[vf].tx_rate;
9929 ivi->min_tx_rate = 0;
9930 ivi->vlan = adapter->vf_data[vf].pf_vlan;
9931 ivi->qos = adapter->vf_data[vf].pf_qos;
9932 ivi->spoofchk = adapter->vf_data[vf].spoofchk_enabled;
9933 ivi->trusted = adapter->vf_data[vf].trusted;
9934 return 0;
9935}
9936
9937static void igb_vmm_control(struct igb_adapter *adapter)
9938{
9939 struct e1000_hw *hw = &adapter->hw;
9940 u32 reg;
9941
9942 switch (hw->mac.type) {
9943 case e1000_82575:
9944 case e1000_i210:
9945 case e1000_i211:
9946 case e1000_i354:
9947 default:
9948 /* replication is not supported for 82575 */
9949 return;
9950 case e1000_82576:
9951 /* notify HW that the MAC is adding vlan tags */
9952 reg = rd32(E1000_DTXCTL);
9953 reg |= E1000_DTXCTL_VLAN_ADDED;
9954 wr32(E1000_DTXCTL, reg);
9955 fallthrough;
9956 case e1000_82580:
9957 /* enable replication vlan tag stripping */
9958 reg = rd32(E1000_RPLOLR);
9959 reg |= E1000_RPLOLR_STRVLAN;
9960 wr32(E1000_RPLOLR, reg);
9961 fallthrough;
9962 case e1000_i350:
9963 /* none of the above registers are supported by i350 */
9964 break;
9965 }
9966
9967 if (adapter->vfs_allocated_count) {
9968 igb_vmdq_set_loopback_pf(hw, true);
9969 igb_vmdq_set_replication_pf(hw, true);
9970 igb_vmdq_set_anti_spoofing_pf(hw, true,
9971 adapter->vfs_allocated_count);
9972 } else {
9973 igb_vmdq_set_loopback_pf(hw, false);
9974 igb_vmdq_set_replication_pf(hw, false);
9975 }
9976}
9977
9978static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
9979{
9980 struct e1000_hw *hw = &adapter->hw;
9981 u32 dmac_thr;
9982 u16 hwm;
9983 u32 reg;
9984
9985 if (hw->mac.type > e1000_82580) {
9986 if (adapter->flags & IGB_FLAG_DMAC) {
9987 /* force threshold to 0. */
9988 wr32(E1000_DMCTXTH, 0);
9989
9990 /* DMA Coalescing high water mark needs to be greater
9991 * than the Rx threshold. Set hwm to PBA - max frame
9992 * size in 16B units, capping it at PBA - 6KB.
9993 */
9994 hwm = 64 * (pba - 6);
9995 reg = rd32(E1000_FCRTC);
9996 reg &= ~E1000_FCRTC_RTH_COAL_MASK;
9997 reg |= FIELD_PREP(E1000_FCRTC_RTH_COAL_MASK, hwm);
9998 wr32(E1000_FCRTC, reg);
9999
10000 /* Set the DMA Coalescing Rx threshold to PBA - 2 * max
10001 * frame size, capping it at PBA - 10KB.
10002 */
10003 dmac_thr = pba - 10;
10004 reg = rd32(E1000_DMACR);
10005 reg &= ~E1000_DMACR_DMACTHR_MASK;
10006 reg |= FIELD_PREP(E1000_DMACR_DMACTHR_MASK, dmac_thr);
10007
10008 /* transition to L0x or L1 if available..*/
10009 reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
10010
10011 /* watchdog timer= +-1000 usec in 32usec intervals */
10012 reg |= (1000 >> 5);
10013
10014 /* Disable BMC-to-OS Watchdog Enable */
10015 if (hw->mac.type != e1000_i354)
10016 reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
10017 wr32(E1000_DMACR, reg);
10018
10019 /* no lower threshold to disable
10020 * coalescing(smart fifb)-UTRESH=0
10021 */
10022 wr32(E1000_DMCRTRH, 0);
10023
10024 reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
10025
10026 wr32(E1000_DMCTLX, reg);
10027
10028 /* free space in tx packet buffer to wake from
10029 * DMA coal
10030 */
10031 wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
10032 (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
10033 }
10034
10035 if (hw->mac.type >= e1000_i210 ||
10036 (adapter->flags & IGB_FLAG_DMAC)) {
10037 reg = rd32(E1000_PCIEMISC);
10038 reg |= E1000_PCIEMISC_LX_DECISION;
10039 wr32(E1000_PCIEMISC, reg);
10040 } /* endif adapter->dmac is not disabled */
10041 } else if (hw->mac.type == e1000_82580) {
10042 u32 reg = rd32(E1000_PCIEMISC);
10043
10044 wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
10045 wr32(E1000_DMACR, 0);
10046 }
10047}
10048
10049/**
10050 * igb_read_i2c_byte - Reads 8 bit word over I2C
10051 * @hw: pointer to hardware structure
10052 * @byte_offset: byte offset to read
10053 * @dev_addr: device address
10054 * @data: value read
10055 *
10056 * Performs byte read operation over I2C interface at
10057 * a specified device address.
10058 **/
10059s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
10060 u8 dev_addr, u8 *data)
10061{
10062 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
10063 struct i2c_client *this_client = adapter->i2c_client;
10064 s32 status;
10065 u16 swfw_mask = 0;
10066
10067 if (!this_client)
10068 return E1000_ERR_I2C;
10069
10070 swfw_mask = E1000_SWFW_PHY0_SM;
10071
10072 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
10073 return E1000_ERR_SWFW_SYNC;
10074
10075 status = i2c_smbus_read_byte_data(this_client, byte_offset);
10076 hw->mac.ops.release_swfw_sync(hw, swfw_mask);
10077
10078 if (status < 0)
10079 return E1000_ERR_I2C;
10080 else {
10081 *data = status;
10082 return 0;
10083 }
10084}
10085
10086/**
10087 * igb_write_i2c_byte - Writes 8 bit word over I2C
10088 * @hw: pointer to hardware structure
10089 * @byte_offset: byte offset to write
10090 * @dev_addr: device address
10091 * @data: value to write
10092 *
10093 * Performs byte write operation over I2C interface at
10094 * a specified device address.
10095 **/
10096s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
10097 u8 dev_addr, u8 data)
10098{
10099 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
10100 struct i2c_client *this_client = adapter->i2c_client;
10101 s32 status;
10102 u16 swfw_mask = E1000_SWFW_PHY0_SM;
10103
10104 if (!this_client)
10105 return E1000_ERR_I2C;
10106
10107 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
10108 return E1000_ERR_SWFW_SYNC;
10109 status = i2c_smbus_write_byte_data(this_client, byte_offset, data);
10110 hw->mac.ops.release_swfw_sync(hw, swfw_mask);
10111
10112 if (status)
10113 return E1000_ERR_I2C;
10114 else
10115 return 0;
10116
10117}
10118
10119int igb_reinit_queues(struct igb_adapter *adapter)
10120{
10121 struct net_device *netdev = adapter->netdev;
10122 struct pci_dev *pdev = adapter->pdev;
10123 int err = 0;
10124
10125 if (netif_running(netdev))
10126 igb_close(netdev);
10127
10128 igb_reset_interrupt_capability(adapter);
10129
10130 if (igb_init_interrupt_scheme(adapter, true)) {
10131 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
10132 return -ENOMEM;
10133 }
10134
10135 if (netif_running(netdev))
10136 err = igb_open(netdev);
10137
10138 return err;
10139}
10140
10141static void igb_nfc_filter_exit(struct igb_adapter *adapter)
10142{
10143 struct igb_nfc_filter *rule;
10144
10145 spin_lock(&adapter->nfc_lock);
10146
10147 hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
10148 igb_erase_filter(adapter, rule);
10149
10150 hlist_for_each_entry(rule, &adapter->cls_flower_list, nfc_node)
10151 igb_erase_filter(adapter, rule);
10152
10153 spin_unlock(&adapter->nfc_lock);
10154}
10155
10156static void igb_nfc_filter_restore(struct igb_adapter *adapter)
10157{
10158 struct igb_nfc_filter *rule;
10159
10160 spin_lock(&adapter->nfc_lock);
10161
10162 hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
10163 igb_add_filter(adapter, rule);
10164
10165 spin_unlock(&adapter->nfc_lock);
10166}
10167
10168static _DEFINE_DEV_PM_OPS(igb_pm_ops, igb_suspend, igb_resume,
10169 igb_runtime_suspend, igb_runtime_resume,
10170 igb_runtime_idle);
10171
10172static struct pci_driver igb_driver = {
10173 .name = igb_driver_name,
10174 .id_table = igb_pci_tbl,
10175 .probe = igb_probe,
10176 .remove = igb_remove,
10177 .driver.pm = pm_ptr(&igb_pm_ops),
10178 .shutdown = igb_shutdown,
10179 .sriov_configure = igb_pci_sriov_configure,
10180 .err_handler = &igb_err_handler
10181};
10182
10183/* igb_main.c */
1/* Intel(R) Gigabit Ethernet Linux driver
2 * Copyright(c) 2007-2014 Intel Corporation.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
11 * more details.
12 *
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, see <http://www.gnu.org/licenses/>.
15 *
16 * The full GNU General Public License is included in this distribution in
17 * the file called "COPYING".
18 *
19 * Contact Information:
20 * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
21 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
22 */
23
24#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
25
26#include <linux/module.h>
27#include <linux/types.h>
28#include <linux/init.h>
29#include <linux/bitops.h>
30#include <linux/vmalloc.h>
31#include <linux/pagemap.h>
32#include <linux/netdevice.h>
33#include <linux/ipv6.h>
34#include <linux/slab.h>
35#include <net/checksum.h>
36#include <net/ip6_checksum.h>
37#include <linux/net_tstamp.h>
38#include <linux/mii.h>
39#include <linux/ethtool.h>
40#include <linux/if.h>
41#include <linux/if_vlan.h>
42#include <linux/pci.h>
43#include <linux/pci-aspm.h>
44#include <linux/delay.h>
45#include <linux/interrupt.h>
46#include <linux/ip.h>
47#include <linux/tcp.h>
48#include <linux/sctp.h>
49#include <linux/if_ether.h>
50#include <linux/aer.h>
51#include <linux/prefetch.h>
52#include <linux/pm_runtime.h>
53#ifdef CONFIG_IGB_DCA
54#include <linux/dca.h>
55#endif
56#include <linux/i2c.h>
57#include "igb.h"
58
59#define MAJ 5
60#define MIN 3
61#define BUILD 0
62#define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
63__stringify(BUILD) "-k"
64char igb_driver_name[] = "igb";
65char igb_driver_version[] = DRV_VERSION;
66static const char igb_driver_string[] =
67 "Intel(R) Gigabit Ethernet Network Driver";
68static const char igb_copyright[] =
69 "Copyright (c) 2007-2014 Intel Corporation.";
70
71static const struct e1000_info *igb_info_tbl[] = {
72 [board_82575] = &e1000_82575_info,
73};
74
75static const struct pci_device_id igb_pci_tbl[] = {
76 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_1GBPS) },
77 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_SGMII) },
78 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) },
79 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I211_COPPER), board_82575 },
80 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 },
81 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 },
82 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 },
83 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 },
84 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER_FLASHLESS), board_82575 },
85 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES_FLASHLESS), board_82575 },
86 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
87 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
88 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
89 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
90 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
91 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
92 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 },
93 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
94 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
95 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
96 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 },
97 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 },
98 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 },
99 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 },
100 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
101 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
102 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
103 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
104 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
105 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
106 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
107 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
108 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
109 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
110 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
111 /* required last entry */
112 {0, }
113};
114
115MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
116
117static int igb_setup_all_tx_resources(struct igb_adapter *);
118static int igb_setup_all_rx_resources(struct igb_adapter *);
119static void igb_free_all_tx_resources(struct igb_adapter *);
120static void igb_free_all_rx_resources(struct igb_adapter *);
121static void igb_setup_mrqc(struct igb_adapter *);
122static int igb_probe(struct pci_dev *, const struct pci_device_id *);
123static void igb_remove(struct pci_dev *pdev);
124static int igb_sw_init(struct igb_adapter *);
125int igb_open(struct net_device *);
126int igb_close(struct net_device *);
127static void igb_configure(struct igb_adapter *);
128static void igb_configure_tx(struct igb_adapter *);
129static void igb_configure_rx(struct igb_adapter *);
130static void igb_clean_all_tx_rings(struct igb_adapter *);
131static void igb_clean_all_rx_rings(struct igb_adapter *);
132static void igb_clean_tx_ring(struct igb_ring *);
133static void igb_clean_rx_ring(struct igb_ring *);
134static void igb_set_rx_mode(struct net_device *);
135static void igb_update_phy_info(unsigned long);
136static void igb_watchdog(unsigned long);
137static void igb_watchdog_task(struct work_struct *);
138static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *);
139static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *dev,
140 struct rtnl_link_stats64 *stats);
141static int igb_change_mtu(struct net_device *, int);
142static int igb_set_mac(struct net_device *, void *);
143static void igb_set_uta(struct igb_adapter *adapter, bool set);
144static irqreturn_t igb_intr(int irq, void *);
145static irqreturn_t igb_intr_msi(int irq, void *);
146static irqreturn_t igb_msix_other(int irq, void *);
147static irqreturn_t igb_msix_ring(int irq, void *);
148#ifdef CONFIG_IGB_DCA
149static void igb_update_dca(struct igb_q_vector *);
150static void igb_setup_dca(struct igb_adapter *);
151#endif /* CONFIG_IGB_DCA */
152static int igb_poll(struct napi_struct *, int);
153static bool igb_clean_tx_irq(struct igb_q_vector *);
154static int igb_clean_rx_irq(struct igb_q_vector *, int);
155static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
156static void igb_tx_timeout(struct net_device *);
157static void igb_reset_task(struct work_struct *);
158static void igb_vlan_mode(struct net_device *netdev,
159 netdev_features_t features);
160static int igb_vlan_rx_add_vid(struct net_device *, __be16, u16);
161static int igb_vlan_rx_kill_vid(struct net_device *, __be16, u16);
162static void igb_restore_vlan(struct igb_adapter *);
163static void igb_rar_set_qsel(struct igb_adapter *, u8 *, u32 , u8);
164static void igb_ping_all_vfs(struct igb_adapter *);
165static void igb_msg_task(struct igb_adapter *);
166static void igb_vmm_control(struct igb_adapter *);
167static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
168static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
169static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
170static int igb_ndo_set_vf_vlan(struct net_device *netdev,
171 int vf, u16 vlan, u8 qos);
172static int igb_ndo_set_vf_bw(struct net_device *, int, int, int);
173static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
174 bool setting);
175static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
176 struct ifla_vf_info *ivi);
177static void igb_check_vf_rate_limit(struct igb_adapter *);
178
179#ifdef CONFIG_PCI_IOV
180static int igb_vf_configure(struct igb_adapter *adapter, int vf);
181static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs);
182static int igb_disable_sriov(struct pci_dev *dev);
183static int igb_pci_disable_sriov(struct pci_dev *dev);
184#endif
185
186#ifdef CONFIG_PM
187#ifdef CONFIG_PM_SLEEP
188static int igb_suspend(struct device *);
189#endif
190static int igb_resume(struct device *);
191static int igb_runtime_suspend(struct device *dev);
192static int igb_runtime_resume(struct device *dev);
193static int igb_runtime_idle(struct device *dev);
194static const struct dev_pm_ops igb_pm_ops = {
195 SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume)
196 SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume,
197 igb_runtime_idle)
198};
199#endif
200static void igb_shutdown(struct pci_dev *);
201static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs);
202#ifdef CONFIG_IGB_DCA
203static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
204static struct notifier_block dca_notifier = {
205 .notifier_call = igb_notify_dca,
206 .next = NULL,
207 .priority = 0
208};
209#endif
210#ifdef CONFIG_NET_POLL_CONTROLLER
211/* for netdump / net console */
212static void igb_netpoll(struct net_device *);
213#endif
214#ifdef CONFIG_PCI_IOV
215static unsigned int max_vfs;
216module_param(max_vfs, uint, 0);
217MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate per physical function");
218#endif /* CONFIG_PCI_IOV */
219
220static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
221 pci_channel_state_t);
222static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
223static void igb_io_resume(struct pci_dev *);
224
225static const struct pci_error_handlers igb_err_handler = {
226 .error_detected = igb_io_error_detected,
227 .slot_reset = igb_io_slot_reset,
228 .resume = igb_io_resume,
229};
230
231static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
232
233static struct pci_driver igb_driver = {
234 .name = igb_driver_name,
235 .id_table = igb_pci_tbl,
236 .probe = igb_probe,
237 .remove = igb_remove,
238#ifdef CONFIG_PM
239 .driver.pm = &igb_pm_ops,
240#endif
241 .shutdown = igb_shutdown,
242 .sriov_configure = igb_pci_sriov_configure,
243 .err_handler = &igb_err_handler
244};
245
246MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
247MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
248MODULE_LICENSE("GPL");
249MODULE_VERSION(DRV_VERSION);
250
251#define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
252static int debug = -1;
253module_param(debug, int, 0);
254MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
255
256struct igb_reg_info {
257 u32 ofs;
258 char *name;
259};
260
261static const struct igb_reg_info igb_reg_info_tbl[] = {
262
263 /* General Registers */
264 {E1000_CTRL, "CTRL"},
265 {E1000_STATUS, "STATUS"},
266 {E1000_CTRL_EXT, "CTRL_EXT"},
267
268 /* Interrupt Registers */
269 {E1000_ICR, "ICR"},
270
271 /* RX Registers */
272 {E1000_RCTL, "RCTL"},
273 {E1000_RDLEN(0), "RDLEN"},
274 {E1000_RDH(0), "RDH"},
275 {E1000_RDT(0), "RDT"},
276 {E1000_RXDCTL(0), "RXDCTL"},
277 {E1000_RDBAL(0), "RDBAL"},
278 {E1000_RDBAH(0), "RDBAH"},
279
280 /* TX Registers */
281 {E1000_TCTL, "TCTL"},
282 {E1000_TDBAL(0), "TDBAL"},
283 {E1000_TDBAH(0), "TDBAH"},
284 {E1000_TDLEN(0), "TDLEN"},
285 {E1000_TDH(0), "TDH"},
286 {E1000_TDT(0), "TDT"},
287 {E1000_TXDCTL(0), "TXDCTL"},
288 {E1000_TDFH, "TDFH"},
289 {E1000_TDFT, "TDFT"},
290 {E1000_TDFHS, "TDFHS"},
291 {E1000_TDFPC, "TDFPC"},
292
293 /* List Terminator */
294 {}
295};
296
297/* igb_regdump - register printout routine */
298static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo)
299{
300 int n = 0;
301 char rname[16];
302 u32 regs[8];
303
304 switch (reginfo->ofs) {
305 case E1000_RDLEN(0):
306 for (n = 0; n < 4; n++)
307 regs[n] = rd32(E1000_RDLEN(n));
308 break;
309 case E1000_RDH(0):
310 for (n = 0; n < 4; n++)
311 regs[n] = rd32(E1000_RDH(n));
312 break;
313 case E1000_RDT(0):
314 for (n = 0; n < 4; n++)
315 regs[n] = rd32(E1000_RDT(n));
316 break;
317 case E1000_RXDCTL(0):
318 for (n = 0; n < 4; n++)
319 regs[n] = rd32(E1000_RXDCTL(n));
320 break;
321 case E1000_RDBAL(0):
322 for (n = 0; n < 4; n++)
323 regs[n] = rd32(E1000_RDBAL(n));
324 break;
325 case E1000_RDBAH(0):
326 for (n = 0; n < 4; n++)
327 regs[n] = rd32(E1000_RDBAH(n));
328 break;
329 case E1000_TDBAL(0):
330 for (n = 0; n < 4; n++)
331 regs[n] = rd32(E1000_RDBAL(n));
332 break;
333 case E1000_TDBAH(0):
334 for (n = 0; n < 4; n++)
335 regs[n] = rd32(E1000_TDBAH(n));
336 break;
337 case E1000_TDLEN(0):
338 for (n = 0; n < 4; n++)
339 regs[n] = rd32(E1000_TDLEN(n));
340 break;
341 case E1000_TDH(0):
342 for (n = 0; n < 4; n++)
343 regs[n] = rd32(E1000_TDH(n));
344 break;
345 case E1000_TDT(0):
346 for (n = 0; n < 4; n++)
347 regs[n] = rd32(E1000_TDT(n));
348 break;
349 case E1000_TXDCTL(0):
350 for (n = 0; n < 4; n++)
351 regs[n] = rd32(E1000_TXDCTL(n));
352 break;
353 default:
354 pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs));
355 return;
356 }
357
358 snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]");
359 pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1],
360 regs[2], regs[3]);
361}
362
363/* igb_dump - Print registers, Tx-rings and Rx-rings */
364static void igb_dump(struct igb_adapter *adapter)
365{
366 struct net_device *netdev = adapter->netdev;
367 struct e1000_hw *hw = &adapter->hw;
368 struct igb_reg_info *reginfo;
369 struct igb_ring *tx_ring;
370 union e1000_adv_tx_desc *tx_desc;
371 struct my_u0 { u64 a; u64 b; } *u0;
372 struct igb_ring *rx_ring;
373 union e1000_adv_rx_desc *rx_desc;
374 u32 staterr;
375 u16 i, n;
376
377 if (!netif_msg_hw(adapter))
378 return;
379
380 /* Print netdevice Info */
381 if (netdev) {
382 dev_info(&adapter->pdev->dev, "Net device Info\n");
383 pr_info("Device Name state trans_start last_rx\n");
384 pr_info("%-15s %016lX %016lX %016lX\n", netdev->name,
385 netdev->state, netdev->trans_start, netdev->last_rx);
386 }
387
388 /* Print Registers */
389 dev_info(&adapter->pdev->dev, "Register Dump\n");
390 pr_info(" Register Name Value\n");
391 for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
392 reginfo->name; reginfo++) {
393 igb_regdump(hw, reginfo);
394 }
395
396 /* Print TX Ring Summary */
397 if (!netdev || !netif_running(netdev))
398 goto exit;
399
400 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
401 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
402 for (n = 0; n < adapter->num_tx_queues; n++) {
403 struct igb_tx_buffer *buffer_info;
404 tx_ring = adapter->tx_ring[n];
405 buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
406 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
407 n, tx_ring->next_to_use, tx_ring->next_to_clean,
408 (u64)dma_unmap_addr(buffer_info, dma),
409 dma_unmap_len(buffer_info, len),
410 buffer_info->next_to_watch,
411 (u64)buffer_info->time_stamp);
412 }
413
414 /* Print TX Rings */
415 if (!netif_msg_tx_done(adapter))
416 goto rx_ring_summary;
417
418 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
419
420 /* Transmit Descriptor Formats
421 *
422 * Advanced Transmit Descriptor
423 * +--------------------------------------------------------------+
424 * 0 | Buffer Address [63:0] |
425 * +--------------------------------------------------------------+
426 * 8 | PAYLEN | PORTS |CC|IDX | STA | DCMD |DTYP|MAC|RSV| DTALEN |
427 * +--------------------------------------------------------------+
428 * 63 46 45 40 39 38 36 35 32 31 24 15 0
429 */
430
431 for (n = 0; n < adapter->num_tx_queues; n++) {
432 tx_ring = adapter->tx_ring[n];
433 pr_info("------------------------------------\n");
434 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
435 pr_info("------------------------------------\n");
436 pr_info("T [desc] [address 63:0 ] [PlPOCIStDDM Ln] [bi->dma ] leng ntw timestamp bi->skb\n");
437
438 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
439 const char *next_desc;
440 struct igb_tx_buffer *buffer_info;
441 tx_desc = IGB_TX_DESC(tx_ring, i);
442 buffer_info = &tx_ring->tx_buffer_info[i];
443 u0 = (struct my_u0 *)tx_desc;
444 if (i == tx_ring->next_to_use &&
445 i == tx_ring->next_to_clean)
446 next_desc = " NTC/U";
447 else if (i == tx_ring->next_to_use)
448 next_desc = " NTU";
449 else if (i == tx_ring->next_to_clean)
450 next_desc = " NTC";
451 else
452 next_desc = "";
453
454 pr_info("T [0x%03X] %016llX %016llX %016llX %04X %p %016llX %p%s\n",
455 i, le64_to_cpu(u0->a),
456 le64_to_cpu(u0->b),
457 (u64)dma_unmap_addr(buffer_info, dma),
458 dma_unmap_len(buffer_info, len),
459 buffer_info->next_to_watch,
460 (u64)buffer_info->time_stamp,
461 buffer_info->skb, next_desc);
462
463 if (netif_msg_pktdata(adapter) && buffer_info->skb)
464 print_hex_dump(KERN_INFO, "",
465 DUMP_PREFIX_ADDRESS,
466 16, 1, buffer_info->skb->data,
467 dma_unmap_len(buffer_info, len),
468 true);
469 }
470 }
471
472 /* Print RX Rings Summary */
473rx_ring_summary:
474 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
475 pr_info("Queue [NTU] [NTC]\n");
476 for (n = 0; n < adapter->num_rx_queues; n++) {
477 rx_ring = adapter->rx_ring[n];
478 pr_info(" %5d %5X %5X\n",
479 n, rx_ring->next_to_use, rx_ring->next_to_clean);
480 }
481
482 /* Print RX Rings */
483 if (!netif_msg_rx_status(adapter))
484 goto exit;
485
486 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
487
488 /* Advanced Receive Descriptor (Read) Format
489 * 63 1 0
490 * +-----------------------------------------------------+
491 * 0 | Packet Buffer Address [63:1] |A0/NSE|
492 * +----------------------------------------------+------+
493 * 8 | Header Buffer Address [63:1] | DD |
494 * +-----------------------------------------------------+
495 *
496 *
497 * Advanced Receive Descriptor (Write-Back) Format
498 *
499 * 63 48 47 32 31 30 21 20 17 16 4 3 0
500 * +------------------------------------------------------+
501 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
502 * | Checksum Ident | | | | Type | Type |
503 * +------------------------------------------------------+
504 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
505 * +------------------------------------------------------+
506 * 63 48 47 32 31 20 19 0
507 */
508
509 for (n = 0; n < adapter->num_rx_queues; n++) {
510 rx_ring = adapter->rx_ring[n];
511 pr_info("------------------------------------\n");
512 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
513 pr_info("------------------------------------\n");
514 pr_info("R [desc] [ PktBuf A0] [ HeadBuf DD] [bi->dma ] [bi->skb] <-- Adv Rx Read format\n");
515 pr_info("RWB[desc] [PcsmIpSHl PtRs] [vl er S cks ln] ---------------- [bi->skb] <-- Adv Rx Write-Back format\n");
516
517 for (i = 0; i < rx_ring->count; i++) {
518 const char *next_desc;
519 struct igb_rx_buffer *buffer_info;
520 buffer_info = &rx_ring->rx_buffer_info[i];
521 rx_desc = IGB_RX_DESC(rx_ring, i);
522 u0 = (struct my_u0 *)rx_desc;
523 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
524
525 if (i == rx_ring->next_to_use)
526 next_desc = " NTU";
527 else if (i == rx_ring->next_to_clean)
528 next_desc = " NTC";
529 else
530 next_desc = "";
531
532 if (staterr & E1000_RXD_STAT_DD) {
533 /* Descriptor Done */
534 pr_info("%s[0x%03X] %016llX %016llX ---------------- %s\n",
535 "RWB", i,
536 le64_to_cpu(u0->a),
537 le64_to_cpu(u0->b),
538 next_desc);
539 } else {
540 pr_info("%s[0x%03X] %016llX %016llX %016llX %s\n",
541 "R ", i,
542 le64_to_cpu(u0->a),
543 le64_to_cpu(u0->b),
544 (u64)buffer_info->dma,
545 next_desc);
546
547 if (netif_msg_pktdata(adapter) &&
548 buffer_info->dma && buffer_info->page) {
549 print_hex_dump(KERN_INFO, "",
550 DUMP_PREFIX_ADDRESS,
551 16, 1,
552 page_address(buffer_info->page) +
553 buffer_info->page_offset,
554 IGB_RX_BUFSZ, true);
555 }
556 }
557 }
558 }
559
560exit:
561 return;
562}
563
564/**
565 * igb_get_i2c_data - Reads the I2C SDA data bit
566 * @hw: pointer to hardware structure
567 * @i2cctl: Current value of I2CCTL register
568 *
569 * Returns the I2C data bit value
570 **/
571static int igb_get_i2c_data(void *data)
572{
573 struct igb_adapter *adapter = (struct igb_adapter *)data;
574 struct e1000_hw *hw = &adapter->hw;
575 s32 i2cctl = rd32(E1000_I2CPARAMS);
576
577 return !!(i2cctl & E1000_I2C_DATA_IN);
578}
579
580/**
581 * igb_set_i2c_data - Sets the I2C data bit
582 * @data: pointer to hardware structure
583 * @state: I2C data value (0 or 1) to set
584 *
585 * Sets the I2C data bit
586 **/
587static void igb_set_i2c_data(void *data, int state)
588{
589 struct igb_adapter *adapter = (struct igb_adapter *)data;
590 struct e1000_hw *hw = &adapter->hw;
591 s32 i2cctl = rd32(E1000_I2CPARAMS);
592
593 if (state)
594 i2cctl |= E1000_I2C_DATA_OUT;
595 else
596 i2cctl &= ~E1000_I2C_DATA_OUT;
597
598 i2cctl &= ~E1000_I2C_DATA_OE_N;
599 i2cctl |= E1000_I2C_CLK_OE_N;
600 wr32(E1000_I2CPARAMS, i2cctl);
601 wrfl();
602
603}
604
605/**
606 * igb_set_i2c_clk - Sets the I2C SCL clock
607 * @data: pointer to hardware structure
608 * @state: state to set clock
609 *
610 * Sets the I2C clock line to state
611 **/
612static void igb_set_i2c_clk(void *data, int state)
613{
614 struct igb_adapter *adapter = (struct igb_adapter *)data;
615 struct e1000_hw *hw = &adapter->hw;
616 s32 i2cctl = rd32(E1000_I2CPARAMS);
617
618 if (state) {
619 i2cctl |= E1000_I2C_CLK_OUT;
620 i2cctl &= ~E1000_I2C_CLK_OE_N;
621 } else {
622 i2cctl &= ~E1000_I2C_CLK_OUT;
623 i2cctl &= ~E1000_I2C_CLK_OE_N;
624 }
625 wr32(E1000_I2CPARAMS, i2cctl);
626 wrfl();
627}
628
629/**
630 * igb_get_i2c_clk - Gets the I2C SCL clock state
631 * @data: pointer to hardware structure
632 *
633 * Gets the I2C clock state
634 **/
635static int igb_get_i2c_clk(void *data)
636{
637 struct igb_adapter *adapter = (struct igb_adapter *)data;
638 struct e1000_hw *hw = &adapter->hw;
639 s32 i2cctl = rd32(E1000_I2CPARAMS);
640
641 return !!(i2cctl & E1000_I2C_CLK_IN);
642}
643
644static const struct i2c_algo_bit_data igb_i2c_algo = {
645 .setsda = igb_set_i2c_data,
646 .setscl = igb_set_i2c_clk,
647 .getsda = igb_get_i2c_data,
648 .getscl = igb_get_i2c_clk,
649 .udelay = 5,
650 .timeout = 20,
651};
652
653/**
654 * igb_get_hw_dev - return device
655 * @hw: pointer to hardware structure
656 *
657 * used by hardware layer to print debugging information
658 **/
659struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
660{
661 struct igb_adapter *adapter = hw->back;
662 return adapter->netdev;
663}
664
665/**
666 * igb_init_module - Driver Registration Routine
667 *
668 * igb_init_module is the first routine called when the driver is
669 * loaded. All it does is register with the PCI subsystem.
670 **/
671static int __init igb_init_module(void)
672{
673 int ret;
674
675 pr_info("%s - version %s\n",
676 igb_driver_string, igb_driver_version);
677 pr_info("%s\n", igb_copyright);
678
679#ifdef CONFIG_IGB_DCA
680 dca_register_notify(&dca_notifier);
681#endif
682 ret = pci_register_driver(&igb_driver);
683 return ret;
684}
685
686module_init(igb_init_module);
687
688/**
689 * igb_exit_module - Driver Exit Cleanup Routine
690 *
691 * igb_exit_module is called just before the driver is removed
692 * from memory.
693 **/
694static void __exit igb_exit_module(void)
695{
696#ifdef CONFIG_IGB_DCA
697 dca_unregister_notify(&dca_notifier);
698#endif
699 pci_unregister_driver(&igb_driver);
700}
701
702module_exit(igb_exit_module);
703
704#define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
705/**
706 * igb_cache_ring_register - Descriptor ring to register mapping
707 * @adapter: board private structure to initialize
708 *
709 * Once we know the feature-set enabled for the device, we'll cache
710 * the register offset the descriptor ring is assigned to.
711 **/
712static void igb_cache_ring_register(struct igb_adapter *adapter)
713{
714 int i = 0, j = 0;
715 u32 rbase_offset = adapter->vfs_allocated_count;
716
717 switch (adapter->hw.mac.type) {
718 case e1000_82576:
719 /* The queues are allocated for virtualization such that VF 0
720 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
721 * In order to avoid collision we start at the first free queue
722 * and continue consuming queues in the same sequence
723 */
724 if (adapter->vfs_allocated_count) {
725 for (; i < adapter->rss_queues; i++)
726 adapter->rx_ring[i]->reg_idx = rbase_offset +
727 Q_IDX_82576(i);
728 }
729 /* Fall through */
730 case e1000_82575:
731 case e1000_82580:
732 case e1000_i350:
733 case e1000_i354:
734 case e1000_i210:
735 case e1000_i211:
736 /* Fall through */
737 default:
738 for (; i < adapter->num_rx_queues; i++)
739 adapter->rx_ring[i]->reg_idx = rbase_offset + i;
740 for (; j < adapter->num_tx_queues; j++)
741 adapter->tx_ring[j]->reg_idx = rbase_offset + j;
742 break;
743 }
744}
745
746u32 igb_rd32(struct e1000_hw *hw, u32 reg)
747{
748 struct igb_adapter *igb = container_of(hw, struct igb_adapter, hw);
749 u8 __iomem *hw_addr = ACCESS_ONCE(hw->hw_addr);
750 u32 value = 0;
751
752 if (E1000_REMOVED(hw_addr))
753 return ~value;
754
755 value = readl(&hw_addr[reg]);
756
757 /* reads should not return all F's */
758 if (!(~value) && (!reg || !(~readl(hw_addr)))) {
759 struct net_device *netdev = igb->netdev;
760 hw->hw_addr = NULL;
761 netif_device_detach(netdev);
762 netdev_err(netdev, "PCIe link lost, device now detached\n");
763 }
764
765 return value;
766}
767
768/**
769 * igb_write_ivar - configure ivar for given MSI-X vector
770 * @hw: pointer to the HW structure
771 * @msix_vector: vector number we are allocating to a given ring
772 * @index: row index of IVAR register to write within IVAR table
773 * @offset: column offset of in IVAR, should be multiple of 8
774 *
775 * This function is intended to handle the writing of the IVAR register
776 * for adapters 82576 and newer. The IVAR table consists of 2 columns,
777 * each containing an cause allocation for an Rx and Tx ring, and a
778 * variable number of rows depending on the number of queues supported.
779 **/
780static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
781 int index, int offset)
782{
783 u32 ivar = array_rd32(E1000_IVAR0, index);
784
785 /* clear any bits that are currently set */
786 ivar &= ~((u32)0xFF << offset);
787
788 /* write vector and valid bit */
789 ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
790
791 array_wr32(E1000_IVAR0, index, ivar);
792}
793
794#define IGB_N0_QUEUE -1
795static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
796{
797 struct igb_adapter *adapter = q_vector->adapter;
798 struct e1000_hw *hw = &adapter->hw;
799 int rx_queue = IGB_N0_QUEUE;
800 int tx_queue = IGB_N0_QUEUE;
801 u32 msixbm = 0;
802
803 if (q_vector->rx.ring)
804 rx_queue = q_vector->rx.ring->reg_idx;
805 if (q_vector->tx.ring)
806 tx_queue = q_vector->tx.ring->reg_idx;
807
808 switch (hw->mac.type) {
809 case e1000_82575:
810 /* The 82575 assigns vectors using a bitmask, which matches the
811 * bitmask for the EICR/EIMS/EIMC registers. To assign one
812 * or more queues to a vector, we write the appropriate bits
813 * into the MSIXBM register for that vector.
814 */
815 if (rx_queue > IGB_N0_QUEUE)
816 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
817 if (tx_queue > IGB_N0_QUEUE)
818 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
819 if (!(adapter->flags & IGB_FLAG_HAS_MSIX) && msix_vector == 0)
820 msixbm |= E1000_EIMS_OTHER;
821 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
822 q_vector->eims_value = msixbm;
823 break;
824 case e1000_82576:
825 /* 82576 uses a table that essentially consists of 2 columns
826 * with 8 rows. The ordering is column-major so we use the
827 * lower 3 bits as the row index, and the 4th bit as the
828 * column offset.
829 */
830 if (rx_queue > IGB_N0_QUEUE)
831 igb_write_ivar(hw, msix_vector,
832 rx_queue & 0x7,
833 (rx_queue & 0x8) << 1);
834 if (tx_queue > IGB_N0_QUEUE)
835 igb_write_ivar(hw, msix_vector,
836 tx_queue & 0x7,
837 ((tx_queue & 0x8) << 1) + 8);
838 q_vector->eims_value = 1 << msix_vector;
839 break;
840 case e1000_82580:
841 case e1000_i350:
842 case e1000_i354:
843 case e1000_i210:
844 case e1000_i211:
845 /* On 82580 and newer adapters the scheme is similar to 82576
846 * however instead of ordering column-major we have things
847 * ordered row-major. So we traverse the table by using
848 * bit 0 as the column offset, and the remaining bits as the
849 * row index.
850 */
851 if (rx_queue > IGB_N0_QUEUE)
852 igb_write_ivar(hw, msix_vector,
853 rx_queue >> 1,
854 (rx_queue & 0x1) << 4);
855 if (tx_queue > IGB_N0_QUEUE)
856 igb_write_ivar(hw, msix_vector,
857 tx_queue >> 1,
858 ((tx_queue & 0x1) << 4) + 8);
859 q_vector->eims_value = 1 << msix_vector;
860 break;
861 default:
862 BUG();
863 break;
864 }
865
866 /* add q_vector eims value to global eims_enable_mask */
867 adapter->eims_enable_mask |= q_vector->eims_value;
868
869 /* configure q_vector to set itr on first interrupt */
870 q_vector->set_itr = 1;
871}
872
873/**
874 * igb_configure_msix - Configure MSI-X hardware
875 * @adapter: board private structure to initialize
876 *
877 * igb_configure_msix sets up the hardware to properly
878 * generate MSI-X interrupts.
879 **/
880static void igb_configure_msix(struct igb_adapter *adapter)
881{
882 u32 tmp;
883 int i, vector = 0;
884 struct e1000_hw *hw = &adapter->hw;
885
886 adapter->eims_enable_mask = 0;
887
888 /* set vector for other causes, i.e. link changes */
889 switch (hw->mac.type) {
890 case e1000_82575:
891 tmp = rd32(E1000_CTRL_EXT);
892 /* enable MSI-X PBA support*/
893 tmp |= E1000_CTRL_EXT_PBA_CLR;
894
895 /* Auto-Mask interrupts upon ICR read. */
896 tmp |= E1000_CTRL_EXT_EIAME;
897 tmp |= E1000_CTRL_EXT_IRCA;
898
899 wr32(E1000_CTRL_EXT, tmp);
900
901 /* enable msix_other interrupt */
902 array_wr32(E1000_MSIXBM(0), vector++, E1000_EIMS_OTHER);
903 adapter->eims_other = E1000_EIMS_OTHER;
904
905 break;
906
907 case e1000_82576:
908 case e1000_82580:
909 case e1000_i350:
910 case e1000_i354:
911 case e1000_i210:
912 case e1000_i211:
913 /* Turn on MSI-X capability first, or our settings
914 * won't stick. And it will take days to debug.
915 */
916 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
917 E1000_GPIE_PBA | E1000_GPIE_EIAME |
918 E1000_GPIE_NSICR);
919
920 /* enable msix_other interrupt */
921 adapter->eims_other = 1 << vector;
922 tmp = (vector++ | E1000_IVAR_VALID) << 8;
923
924 wr32(E1000_IVAR_MISC, tmp);
925 break;
926 default:
927 /* do nothing, since nothing else supports MSI-X */
928 break;
929 } /* switch (hw->mac.type) */
930
931 adapter->eims_enable_mask |= adapter->eims_other;
932
933 for (i = 0; i < adapter->num_q_vectors; i++)
934 igb_assign_vector(adapter->q_vector[i], vector++);
935
936 wrfl();
937}
938
939/**
940 * igb_request_msix - Initialize MSI-X interrupts
941 * @adapter: board private structure to initialize
942 *
943 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
944 * kernel.
945 **/
946static int igb_request_msix(struct igb_adapter *adapter)
947{
948 struct net_device *netdev = adapter->netdev;
949 int i, err = 0, vector = 0, free_vector = 0;
950
951 err = request_irq(adapter->msix_entries[vector].vector,
952 igb_msix_other, 0, netdev->name, adapter);
953 if (err)
954 goto err_out;
955
956 for (i = 0; i < adapter->num_q_vectors; i++) {
957 struct igb_q_vector *q_vector = adapter->q_vector[i];
958
959 vector++;
960
961 q_vector->itr_register = adapter->io_addr + E1000_EITR(vector);
962
963 if (q_vector->rx.ring && q_vector->tx.ring)
964 sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
965 q_vector->rx.ring->queue_index);
966 else if (q_vector->tx.ring)
967 sprintf(q_vector->name, "%s-tx-%u", netdev->name,
968 q_vector->tx.ring->queue_index);
969 else if (q_vector->rx.ring)
970 sprintf(q_vector->name, "%s-rx-%u", netdev->name,
971 q_vector->rx.ring->queue_index);
972 else
973 sprintf(q_vector->name, "%s-unused", netdev->name);
974
975 err = request_irq(adapter->msix_entries[vector].vector,
976 igb_msix_ring, 0, q_vector->name,
977 q_vector);
978 if (err)
979 goto err_free;
980 }
981
982 igb_configure_msix(adapter);
983 return 0;
984
985err_free:
986 /* free already assigned IRQs */
987 free_irq(adapter->msix_entries[free_vector++].vector, adapter);
988
989 vector--;
990 for (i = 0; i < vector; i++) {
991 free_irq(adapter->msix_entries[free_vector++].vector,
992 adapter->q_vector[i]);
993 }
994err_out:
995 return err;
996}
997
998/**
999 * igb_free_q_vector - Free memory allocated for specific interrupt vector
1000 * @adapter: board private structure to initialize
1001 * @v_idx: Index of vector to be freed
1002 *
1003 * This function frees the memory allocated to the q_vector.
1004 **/
1005static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx)
1006{
1007 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1008
1009 adapter->q_vector[v_idx] = NULL;
1010
1011 /* igb_get_stats64() might access the rings on this vector,
1012 * we must wait a grace period before freeing it.
1013 */
1014 if (q_vector)
1015 kfree_rcu(q_vector, rcu);
1016}
1017
1018/**
1019 * igb_reset_q_vector - Reset config for interrupt vector
1020 * @adapter: board private structure to initialize
1021 * @v_idx: Index of vector to be reset
1022 *
1023 * If NAPI is enabled it will delete any references to the
1024 * NAPI struct. This is preparation for igb_free_q_vector.
1025 **/
1026static void igb_reset_q_vector(struct igb_adapter *adapter, int v_idx)
1027{
1028 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1029
1030 /* Coming from igb_set_interrupt_capability, the vectors are not yet
1031 * allocated. So, q_vector is NULL so we should stop here.
1032 */
1033 if (!q_vector)
1034 return;
1035
1036 if (q_vector->tx.ring)
1037 adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
1038
1039 if (q_vector->rx.ring)
1040 adapter->rx_ring[q_vector->rx.ring->queue_index] = NULL;
1041
1042 netif_napi_del(&q_vector->napi);
1043
1044}
1045
1046static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
1047{
1048 int v_idx = adapter->num_q_vectors;
1049
1050 if (adapter->flags & IGB_FLAG_HAS_MSIX)
1051 pci_disable_msix(adapter->pdev);
1052 else if (adapter->flags & IGB_FLAG_HAS_MSI)
1053 pci_disable_msi(adapter->pdev);
1054
1055 while (v_idx--)
1056 igb_reset_q_vector(adapter, v_idx);
1057}
1058
1059/**
1060 * igb_free_q_vectors - Free memory allocated for interrupt vectors
1061 * @adapter: board private structure to initialize
1062 *
1063 * This function frees the memory allocated to the q_vectors. In addition if
1064 * NAPI is enabled it will delete any references to the NAPI struct prior
1065 * to freeing the q_vector.
1066 **/
1067static void igb_free_q_vectors(struct igb_adapter *adapter)
1068{
1069 int v_idx = adapter->num_q_vectors;
1070
1071 adapter->num_tx_queues = 0;
1072 adapter->num_rx_queues = 0;
1073 adapter->num_q_vectors = 0;
1074
1075 while (v_idx--) {
1076 igb_reset_q_vector(adapter, v_idx);
1077 igb_free_q_vector(adapter, v_idx);
1078 }
1079}
1080
1081/**
1082 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1083 * @adapter: board private structure to initialize
1084 *
1085 * This function resets the device so that it has 0 Rx queues, Tx queues, and
1086 * MSI-X interrupts allocated.
1087 */
1088static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
1089{
1090 igb_free_q_vectors(adapter);
1091 igb_reset_interrupt_capability(adapter);
1092}
1093
1094/**
1095 * igb_set_interrupt_capability - set MSI or MSI-X if supported
1096 * @adapter: board private structure to initialize
1097 * @msix: boolean value of MSIX capability
1098 *
1099 * Attempt to configure interrupts using the best available
1100 * capabilities of the hardware and kernel.
1101 **/
1102static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix)
1103{
1104 int err;
1105 int numvecs, i;
1106
1107 if (!msix)
1108 goto msi_only;
1109 adapter->flags |= IGB_FLAG_HAS_MSIX;
1110
1111 /* Number of supported queues. */
1112 adapter->num_rx_queues = adapter->rss_queues;
1113 if (adapter->vfs_allocated_count)
1114 adapter->num_tx_queues = 1;
1115 else
1116 adapter->num_tx_queues = adapter->rss_queues;
1117
1118 /* start with one vector for every Rx queue */
1119 numvecs = adapter->num_rx_queues;
1120
1121 /* if Tx handler is separate add 1 for every Tx queue */
1122 if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1123 numvecs += adapter->num_tx_queues;
1124
1125 /* store the number of vectors reserved for queues */
1126 adapter->num_q_vectors = numvecs;
1127
1128 /* add 1 vector for link status interrupts */
1129 numvecs++;
1130 for (i = 0; i < numvecs; i++)
1131 adapter->msix_entries[i].entry = i;
1132
1133 err = pci_enable_msix_range(adapter->pdev,
1134 adapter->msix_entries,
1135 numvecs,
1136 numvecs);
1137 if (err > 0)
1138 return;
1139
1140 igb_reset_interrupt_capability(adapter);
1141
1142 /* If we can't do MSI-X, try MSI */
1143msi_only:
1144 adapter->flags &= ~IGB_FLAG_HAS_MSIX;
1145#ifdef CONFIG_PCI_IOV
1146 /* disable SR-IOV for non MSI-X configurations */
1147 if (adapter->vf_data) {
1148 struct e1000_hw *hw = &adapter->hw;
1149 /* disable iov and allow time for transactions to clear */
1150 pci_disable_sriov(adapter->pdev);
1151 msleep(500);
1152
1153 kfree(adapter->vf_data);
1154 adapter->vf_data = NULL;
1155 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1156 wrfl();
1157 msleep(100);
1158 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1159 }
1160#endif
1161 adapter->vfs_allocated_count = 0;
1162 adapter->rss_queues = 1;
1163 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1164 adapter->num_rx_queues = 1;
1165 adapter->num_tx_queues = 1;
1166 adapter->num_q_vectors = 1;
1167 if (!pci_enable_msi(adapter->pdev))
1168 adapter->flags |= IGB_FLAG_HAS_MSI;
1169}
1170
1171static void igb_add_ring(struct igb_ring *ring,
1172 struct igb_ring_container *head)
1173{
1174 head->ring = ring;
1175 head->count++;
1176}
1177
1178/**
1179 * igb_alloc_q_vector - Allocate memory for a single interrupt vector
1180 * @adapter: board private structure to initialize
1181 * @v_count: q_vectors allocated on adapter, used for ring interleaving
1182 * @v_idx: index of vector in adapter struct
1183 * @txr_count: total number of Tx rings to allocate
1184 * @txr_idx: index of first Tx ring to allocate
1185 * @rxr_count: total number of Rx rings to allocate
1186 * @rxr_idx: index of first Rx ring to allocate
1187 *
1188 * We allocate one q_vector. If allocation fails we return -ENOMEM.
1189 **/
1190static int igb_alloc_q_vector(struct igb_adapter *adapter,
1191 int v_count, int v_idx,
1192 int txr_count, int txr_idx,
1193 int rxr_count, int rxr_idx)
1194{
1195 struct igb_q_vector *q_vector;
1196 struct igb_ring *ring;
1197 int ring_count, size;
1198
1199 /* igb only supports 1 Tx and/or 1 Rx queue per vector */
1200 if (txr_count > 1 || rxr_count > 1)
1201 return -ENOMEM;
1202
1203 ring_count = txr_count + rxr_count;
1204 size = sizeof(struct igb_q_vector) +
1205 (sizeof(struct igb_ring) * ring_count);
1206
1207 /* allocate q_vector and rings */
1208 q_vector = adapter->q_vector[v_idx];
1209 if (!q_vector) {
1210 q_vector = kzalloc(size, GFP_KERNEL);
1211 } else if (size > ksize(q_vector)) {
1212 kfree_rcu(q_vector, rcu);
1213 q_vector = kzalloc(size, GFP_KERNEL);
1214 } else {
1215 memset(q_vector, 0, size);
1216 }
1217 if (!q_vector)
1218 return -ENOMEM;
1219
1220 /* initialize NAPI */
1221 netif_napi_add(adapter->netdev, &q_vector->napi,
1222 igb_poll, 64);
1223
1224 /* tie q_vector and adapter together */
1225 adapter->q_vector[v_idx] = q_vector;
1226 q_vector->adapter = adapter;
1227
1228 /* initialize work limits */
1229 q_vector->tx.work_limit = adapter->tx_work_limit;
1230
1231 /* initialize ITR configuration */
1232 q_vector->itr_register = adapter->io_addr + E1000_EITR(0);
1233 q_vector->itr_val = IGB_START_ITR;
1234
1235 /* initialize pointer to rings */
1236 ring = q_vector->ring;
1237
1238 /* intialize ITR */
1239 if (rxr_count) {
1240 /* rx or rx/tx vector */
1241 if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
1242 q_vector->itr_val = adapter->rx_itr_setting;
1243 } else {
1244 /* tx only vector */
1245 if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
1246 q_vector->itr_val = adapter->tx_itr_setting;
1247 }
1248
1249 if (txr_count) {
1250 /* assign generic ring traits */
1251 ring->dev = &adapter->pdev->dev;
1252 ring->netdev = adapter->netdev;
1253
1254 /* configure backlink on ring */
1255 ring->q_vector = q_vector;
1256
1257 /* update q_vector Tx values */
1258 igb_add_ring(ring, &q_vector->tx);
1259
1260 /* For 82575, context index must be unique per ring. */
1261 if (adapter->hw.mac.type == e1000_82575)
1262 set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
1263
1264 /* apply Tx specific ring traits */
1265 ring->count = adapter->tx_ring_count;
1266 ring->queue_index = txr_idx;
1267
1268 u64_stats_init(&ring->tx_syncp);
1269 u64_stats_init(&ring->tx_syncp2);
1270
1271 /* assign ring to adapter */
1272 adapter->tx_ring[txr_idx] = ring;
1273
1274 /* push pointer to next ring */
1275 ring++;
1276 }
1277
1278 if (rxr_count) {
1279 /* assign generic ring traits */
1280 ring->dev = &adapter->pdev->dev;
1281 ring->netdev = adapter->netdev;
1282
1283 /* configure backlink on ring */
1284 ring->q_vector = q_vector;
1285
1286 /* update q_vector Rx values */
1287 igb_add_ring(ring, &q_vector->rx);
1288
1289 /* set flag indicating ring supports SCTP checksum offload */
1290 if (adapter->hw.mac.type >= e1000_82576)
1291 set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
1292
1293 /* On i350, i354, i210, and i211, loopback VLAN packets
1294 * have the tag byte-swapped.
1295 */
1296 if (adapter->hw.mac.type >= e1000_i350)
1297 set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
1298
1299 /* apply Rx specific ring traits */
1300 ring->count = adapter->rx_ring_count;
1301 ring->queue_index = rxr_idx;
1302
1303 u64_stats_init(&ring->rx_syncp);
1304
1305 /* assign ring to adapter */
1306 adapter->rx_ring[rxr_idx] = ring;
1307 }
1308
1309 return 0;
1310}
1311
1312
1313/**
1314 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
1315 * @adapter: board private structure to initialize
1316 *
1317 * We allocate one q_vector per queue interrupt. If allocation fails we
1318 * return -ENOMEM.
1319 **/
1320static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1321{
1322 int q_vectors = adapter->num_q_vectors;
1323 int rxr_remaining = adapter->num_rx_queues;
1324 int txr_remaining = adapter->num_tx_queues;
1325 int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1326 int err;
1327
1328 if (q_vectors >= (rxr_remaining + txr_remaining)) {
1329 for (; rxr_remaining; v_idx++) {
1330 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1331 0, 0, 1, rxr_idx);
1332
1333 if (err)
1334 goto err_out;
1335
1336 /* update counts and index */
1337 rxr_remaining--;
1338 rxr_idx++;
1339 }
1340 }
1341
1342 for (; v_idx < q_vectors; v_idx++) {
1343 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1344 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1345
1346 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1347 tqpv, txr_idx, rqpv, rxr_idx);
1348
1349 if (err)
1350 goto err_out;
1351
1352 /* update counts and index */
1353 rxr_remaining -= rqpv;
1354 txr_remaining -= tqpv;
1355 rxr_idx++;
1356 txr_idx++;
1357 }
1358
1359 return 0;
1360
1361err_out:
1362 adapter->num_tx_queues = 0;
1363 adapter->num_rx_queues = 0;
1364 adapter->num_q_vectors = 0;
1365
1366 while (v_idx--)
1367 igb_free_q_vector(adapter, v_idx);
1368
1369 return -ENOMEM;
1370}
1371
1372/**
1373 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1374 * @adapter: board private structure to initialize
1375 * @msix: boolean value of MSIX capability
1376 *
1377 * This function initializes the interrupts and allocates all of the queues.
1378 **/
1379static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
1380{
1381 struct pci_dev *pdev = adapter->pdev;
1382 int err;
1383
1384 igb_set_interrupt_capability(adapter, msix);
1385
1386 err = igb_alloc_q_vectors(adapter);
1387 if (err) {
1388 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1389 goto err_alloc_q_vectors;
1390 }
1391
1392 igb_cache_ring_register(adapter);
1393
1394 return 0;
1395
1396err_alloc_q_vectors:
1397 igb_reset_interrupt_capability(adapter);
1398 return err;
1399}
1400
1401/**
1402 * igb_request_irq - initialize interrupts
1403 * @adapter: board private structure to initialize
1404 *
1405 * Attempts to configure interrupts using the best available
1406 * capabilities of the hardware and kernel.
1407 **/
1408static int igb_request_irq(struct igb_adapter *adapter)
1409{
1410 struct net_device *netdev = adapter->netdev;
1411 struct pci_dev *pdev = adapter->pdev;
1412 int err = 0;
1413
1414 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1415 err = igb_request_msix(adapter);
1416 if (!err)
1417 goto request_done;
1418 /* fall back to MSI */
1419 igb_free_all_tx_resources(adapter);
1420 igb_free_all_rx_resources(adapter);
1421
1422 igb_clear_interrupt_scheme(adapter);
1423 err = igb_init_interrupt_scheme(adapter, false);
1424 if (err)
1425 goto request_done;
1426
1427 igb_setup_all_tx_resources(adapter);
1428 igb_setup_all_rx_resources(adapter);
1429 igb_configure(adapter);
1430 }
1431
1432 igb_assign_vector(adapter->q_vector[0], 0);
1433
1434 if (adapter->flags & IGB_FLAG_HAS_MSI) {
1435 err = request_irq(pdev->irq, igb_intr_msi, 0,
1436 netdev->name, adapter);
1437 if (!err)
1438 goto request_done;
1439
1440 /* fall back to legacy interrupts */
1441 igb_reset_interrupt_capability(adapter);
1442 adapter->flags &= ~IGB_FLAG_HAS_MSI;
1443 }
1444
1445 err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1446 netdev->name, adapter);
1447
1448 if (err)
1449 dev_err(&pdev->dev, "Error %d getting interrupt\n",
1450 err);
1451
1452request_done:
1453 return err;
1454}
1455
1456static void igb_free_irq(struct igb_adapter *adapter)
1457{
1458 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1459 int vector = 0, i;
1460
1461 free_irq(adapter->msix_entries[vector++].vector, adapter);
1462
1463 for (i = 0; i < adapter->num_q_vectors; i++)
1464 free_irq(adapter->msix_entries[vector++].vector,
1465 adapter->q_vector[i]);
1466 } else {
1467 free_irq(adapter->pdev->irq, adapter);
1468 }
1469}
1470
1471/**
1472 * igb_irq_disable - Mask off interrupt generation on the NIC
1473 * @adapter: board private structure
1474 **/
1475static void igb_irq_disable(struct igb_adapter *adapter)
1476{
1477 struct e1000_hw *hw = &adapter->hw;
1478
1479 /* we need to be careful when disabling interrupts. The VFs are also
1480 * mapped into these registers and so clearing the bits can cause
1481 * issues on the VF drivers so we only need to clear what we set
1482 */
1483 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1484 u32 regval = rd32(E1000_EIAM);
1485
1486 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1487 wr32(E1000_EIMC, adapter->eims_enable_mask);
1488 regval = rd32(E1000_EIAC);
1489 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1490 }
1491
1492 wr32(E1000_IAM, 0);
1493 wr32(E1000_IMC, ~0);
1494 wrfl();
1495 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1496 int i;
1497
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
1517 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1518 regval = rd32(E1000_EIAM);
1519 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1520 wr32(E1000_EIMS, adapter->eims_enable_mask);
1521 if (adapter->vfs_allocated_count) {
1522 wr32(E1000_MBVFIMR, 0xFF);
1523 ims |= E1000_IMS_VMMB;
1524 }
1525 wr32(E1000_IMS, ims);
1526 } else {
1527 wr32(E1000_IMS, IMS_ENABLE_MASK |
1528 E1000_IMS_DRSTA);
1529 wr32(E1000_IAM, IMS_ENABLE_MASK |
1530 E1000_IMS_DRSTA);
1531 }
1532}
1533
1534static void igb_update_mng_vlan(struct igb_adapter *adapter)
1535{
1536 struct e1000_hw *hw = &adapter->hw;
1537 u16 pf_id = adapter->vfs_allocated_count;
1538 u16 vid = adapter->hw.mng_cookie.vlan_id;
1539 u16 old_vid = adapter->mng_vlan_id;
1540
1541 if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1542 /* add VID to filter table */
1543 igb_vfta_set(hw, vid, pf_id, true, true);
1544 adapter->mng_vlan_id = vid;
1545 } else {
1546 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1547 }
1548
1549 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1550 (vid != old_vid) &&
1551 !test_bit(old_vid, adapter->active_vlans)) {
1552 /* remove VID from filter table */
1553 igb_vfta_set(hw, vid, pf_id, false, true);
1554 }
1555}
1556
1557/**
1558 * igb_release_hw_control - release control of the h/w to f/w
1559 * @adapter: address of board private structure
1560 *
1561 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1562 * For ASF and Pass Through versions of f/w this means that the
1563 * driver is no longer loaded.
1564 **/
1565static void igb_release_hw_control(struct igb_adapter *adapter)
1566{
1567 struct e1000_hw *hw = &adapter->hw;
1568 u32 ctrl_ext;
1569
1570 /* Let firmware take over control of h/w */
1571 ctrl_ext = rd32(E1000_CTRL_EXT);
1572 wr32(E1000_CTRL_EXT,
1573 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1574}
1575
1576/**
1577 * igb_get_hw_control - get control of the h/w from f/w
1578 * @adapter: address of board private structure
1579 *
1580 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1581 * For ASF and Pass Through versions of f/w this means that
1582 * the driver is loaded.
1583 **/
1584static void igb_get_hw_control(struct igb_adapter *adapter)
1585{
1586 struct e1000_hw *hw = &adapter->hw;
1587 u32 ctrl_ext;
1588
1589 /* Let firmware know the driver has taken over */
1590 ctrl_ext = rd32(E1000_CTRL_EXT);
1591 wr32(E1000_CTRL_EXT,
1592 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1593}
1594
1595/**
1596 * igb_configure - configure the hardware for RX and TX
1597 * @adapter: private board structure
1598 **/
1599static void igb_configure(struct igb_adapter *adapter)
1600{
1601 struct net_device *netdev = adapter->netdev;
1602 int i;
1603
1604 igb_get_hw_control(adapter);
1605 igb_set_rx_mode(netdev);
1606
1607 igb_restore_vlan(adapter);
1608
1609 igb_setup_tctl(adapter);
1610 igb_setup_mrqc(adapter);
1611 igb_setup_rctl(adapter);
1612
1613 igb_configure_tx(adapter);
1614 igb_configure_rx(adapter);
1615
1616 igb_rx_fifo_flush_82575(&adapter->hw);
1617
1618 /* call igb_desc_unused which always leaves
1619 * at least 1 descriptor unused to make sure
1620 * next_to_use != next_to_clean
1621 */
1622 for (i = 0; i < adapter->num_rx_queues; i++) {
1623 struct igb_ring *ring = adapter->rx_ring[i];
1624 igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
1625 }
1626}
1627
1628/**
1629 * igb_power_up_link - Power up the phy/serdes link
1630 * @adapter: address of board private structure
1631 **/
1632void igb_power_up_link(struct igb_adapter *adapter)
1633{
1634 igb_reset_phy(&adapter->hw);
1635
1636 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1637 igb_power_up_phy_copper(&adapter->hw);
1638 else
1639 igb_power_up_serdes_link_82575(&adapter->hw);
1640
1641 igb_setup_link(&adapter->hw);
1642}
1643
1644/**
1645 * igb_power_down_link - Power down the phy/serdes link
1646 * @adapter: address of board private structure
1647 */
1648static void igb_power_down_link(struct igb_adapter *adapter)
1649{
1650 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1651 igb_power_down_phy_copper_82575(&adapter->hw);
1652 else
1653 igb_shutdown_serdes_link_82575(&adapter->hw);
1654}
1655
1656/**
1657 * Detect and switch function for Media Auto Sense
1658 * @adapter: address of the board private structure
1659 **/
1660static void igb_check_swap_media(struct igb_adapter *adapter)
1661{
1662 struct e1000_hw *hw = &adapter->hw;
1663 u32 ctrl_ext, connsw;
1664 bool swap_now = false;
1665
1666 ctrl_ext = rd32(E1000_CTRL_EXT);
1667 connsw = rd32(E1000_CONNSW);
1668
1669 /* need to live swap if current media is copper and we have fiber/serdes
1670 * to go to.
1671 */
1672
1673 if ((hw->phy.media_type == e1000_media_type_copper) &&
1674 (!(connsw & E1000_CONNSW_AUTOSENSE_EN))) {
1675 swap_now = true;
1676 } else if (!(connsw & E1000_CONNSW_SERDESD)) {
1677 /* copper signal takes time to appear */
1678 if (adapter->copper_tries < 4) {
1679 adapter->copper_tries++;
1680 connsw |= E1000_CONNSW_AUTOSENSE_CONF;
1681 wr32(E1000_CONNSW, connsw);
1682 return;
1683 } else {
1684 adapter->copper_tries = 0;
1685 if ((connsw & E1000_CONNSW_PHYSD) &&
1686 (!(connsw & E1000_CONNSW_PHY_PDN))) {
1687 swap_now = true;
1688 connsw &= ~E1000_CONNSW_AUTOSENSE_CONF;
1689 wr32(E1000_CONNSW, connsw);
1690 }
1691 }
1692 }
1693
1694 if (!swap_now)
1695 return;
1696
1697 switch (hw->phy.media_type) {
1698 case e1000_media_type_copper:
1699 netdev_info(adapter->netdev,
1700 "MAS: changing media to fiber/serdes\n");
1701 ctrl_ext |=
1702 E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
1703 adapter->flags |= IGB_FLAG_MEDIA_RESET;
1704 adapter->copper_tries = 0;
1705 break;
1706 case e1000_media_type_internal_serdes:
1707 case e1000_media_type_fiber:
1708 netdev_info(adapter->netdev,
1709 "MAS: changing media to copper\n");
1710 ctrl_ext &=
1711 ~E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
1712 adapter->flags |= IGB_FLAG_MEDIA_RESET;
1713 break;
1714 default:
1715 /* shouldn't get here during regular operation */
1716 netdev_err(adapter->netdev,
1717 "AMS: Invalid media type found, returning\n");
1718 break;
1719 }
1720 wr32(E1000_CTRL_EXT, ctrl_ext);
1721}
1722
1723/**
1724 * igb_up - Open the interface and prepare it to handle traffic
1725 * @adapter: board private structure
1726 **/
1727int igb_up(struct igb_adapter *adapter)
1728{
1729 struct e1000_hw *hw = &adapter->hw;
1730 int i;
1731
1732 /* hardware has been reset, we need to reload some things */
1733 igb_configure(adapter);
1734
1735 clear_bit(__IGB_DOWN, &adapter->state);
1736
1737 for (i = 0; i < adapter->num_q_vectors; i++)
1738 napi_enable(&(adapter->q_vector[i]->napi));
1739
1740 if (adapter->flags & IGB_FLAG_HAS_MSIX)
1741 igb_configure_msix(adapter);
1742 else
1743 igb_assign_vector(adapter->q_vector[0], 0);
1744
1745 /* Clear any pending interrupts. */
1746 rd32(E1000_ICR);
1747 igb_irq_enable(adapter);
1748
1749 /* notify VFs that reset has been completed */
1750 if (adapter->vfs_allocated_count) {
1751 u32 reg_data = rd32(E1000_CTRL_EXT);
1752
1753 reg_data |= E1000_CTRL_EXT_PFRSTD;
1754 wr32(E1000_CTRL_EXT, reg_data);
1755 }
1756
1757 netif_tx_start_all_queues(adapter->netdev);
1758
1759 /* start the watchdog. */
1760 hw->mac.get_link_status = 1;
1761 schedule_work(&adapter->watchdog_task);
1762
1763 if ((adapter->flags & IGB_FLAG_EEE) &&
1764 (!hw->dev_spec._82575.eee_disable))
1765 adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
1766
1767 return 0;
1768}
1769
1770void igb_down(struct igb_adapter *adapter)
1771{
1772 struct net_device *netdev = adapter->netdev;
1773 struct e1000_hw *hw = &adapter->hw;
1774 u32 tctl, rctl;
1775 int i;
1776
1777 /* signal that we're down so the interrupt handler does not
1778 * reschedule our watchdog timer
1779 */
1780 set_bit(__IGB_DOWN, &adapter->state);
1781
1782 /* disable receives in the hardware */
1783 rctl = rd32(E1000_RCTL);
1784 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1785 /* flush and sleep below */
1786
1787 netif_carrier_off(netdev);
1788 netif_tx_stop_all_queues(netdev);
1789
1790 /* disable transmits in the hardware */
1791 tctl = rd32(E1000_TCTL);
1792 tctl &= ~E1000_TCTL_EN;
1793 wr32(E1000_TCTL, tctl);
1794 /* flush both disables and wait for them to finish */
1795 wrfl();
1796 usleep_range(10000, 11000);
1797
1798 igb_irq_disable(adapter);
1799
1800 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
1801
1802 for (i = 0; i < adapter->num_q_vectors; i++) {
1803 if (adapter->q_vector[i]) {
1804 napi_synchronize(&adapter->q_vector[i]->napi);
1805 napi_disable(&adapter->q_vector[i]->napi);
1806 }
1807 }
1808
1809 del_timer_sync(&adapter->watchdog_timer);
1810 del_timer_sync(&adapter->phy_info_timer);
1811
1812 /* record the stats before reset*/
1813 spin_lock(&adapter->stats64_lock);
1814 igb_update_stats(adapter, &adapter->stats64);
1815 spin_unlock(&adapter->stats64_lock);
1816
1817 adapter->link_speed = 0;
1818 adapter->link_duplex = 0;
1819
1820 if (!pci_channel_offline(adapter->pdev))
1821 igb_reset(adapter);
1822
1823 /* clear VLAN promisc flag so VFTA will be updated if necessary */
1824 adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
1825
1826 igb_clean_all_tx_rings(adapter);
1827 igb_clean_all_rx_rings(adapter);
1828#ifdef CONFIG_IGB_DCA
1829
1830 /* since we reset the hardware DCA settings were cleared */
1831 igb_setup_dca(adapter);
1832#endif
1833}
1834
1835void igb_reinit_locked(struct igb_adapter *adapter)
1836{
1837 WARN_ON(in_interrupt());
1838 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1839 usleep_range(1000, 2000);
1840 igb_down(adapter);
1841 igb_up(adapter);
1842 clear_bit(__IGB_RESETTING, &adapter->state);
1843}
1844
1845/** igb_enable_mas - Media Autosense re-enable after swap
1846 *
1847 * @adapter: adapter struct
1848 **/
1849static void igb_enable_mas(struct igb_adapter *adapter)
1850{
1851 struct e1000_hw *hw = &adapter->hw;
1852 u32 connsw = rd32(E1000_CONNSW);
1853
1854 /* configure for SerDes media detect */
1855 if ((hw->phy.media_type == e1000_media_type_copper) &&
1856 (!(connsw & E1000_CONNSW_SERDESD))) {
1857 connsw |= E1000_CONNSW_ENRGSRC;
1858 connsw |= E1000_CONNSW_AUTOSENSE_EN;
1859 wr32(E1000_CONNSW, connsw);
1860 wrfl();
1861 }
1862}
1863
1864void igb_reset(struct igb_adapter *adapter)
1865{
1866 struct pci_dev *pdev = adapter->pdev;
1867 struct e1000_hw *hw = &adapter->hw;
1868 struct e1000_mac_info *mac = &hw->mac;
1869 struct e1000_fc_info *fc = &hw->fc;
1870 u32 pba, hwm;
1871
1872 /* Repartition Pba for greater than 9k mtu
1873 * To take effect CTRL.RST is required.
1874 */
1875 switch (mac->type) {
1876 case e1000_i350:
1877 case e1000_i354:
1878 case e1000_82580:
1879 pba = rd32(E1000_RXPBS);
1880 pba = igb_rxpbs_adjust_82580(pba);
1881 break;
1882 case e1000_82576:
1883 pba = rd32(E1000_RXPBS);
1884 pba &= E1000_RXPBS_SIZE_MASK_82576;
1885 break;
1886 case e1000_82575:
1887 case e1000_i210:
1888 case e1000_i211:
1889 default:
1890 pba = E1000_PBA_34K;
1891 break;
1892 }
1893
1894 if (mac->type == e1000_82575) {
1895 u32 min_rx_space, min_tx_space, needed_tx_space;
1896
1897 /* write Rx PBA so that hardware can report correct Tx PBA */
1898 wr32(E1000_PBA, pba);
1899
1900 /* To maintain wire speed transmits, the Tx FIFO should be
1901 * large enough to accommodate two full transmit packets,
1902 * rounded up to the next 1KB and expressed in KB. Likewise,
1903 * the Rx FIFO should be large enough to accommodate at least
1904 * one full receive packet and is similarly rounded up and
1905 * expressed in KB.
1906 */
1907 min_rx_space = DIV_ROUND_UP(MAX_JUMBO_FRAME_SIZE, 1024);
1908
1909 /* The Tx FIFO also stores 16 bytes of information about the Tx
1910 * but don't include Ethernet FCS because hardware appends it.
1911 * We only need to round down to the nearest 512 byte block
1912 * count since the value we care about is 2 frames, not 1.
1913 */
1914 min_tx_space = adapter->max_frame_size;
1915 min_tx_space += sizeof(union e1000_adv_tx_desc) - ETH_FCS_LEN;
1916 min_tx_space = DIV_ROUND_UP(min_tx_space, 512);
1917
1918 /* upper 16 bits has Tx packet buffer allocation size in KB */
1919 needed_tx_space = min_tx_space - (rd32(E1000_PBA) >> 16);
1920
1921 /* If current Tx allocation is less than the min Tx FIFO size,
1922 * and the min Tx FIFO size is less than the current Rx FIFO
1923 * allocation, take space away from current Rx allocation.
1924 */
1925 if (needed_tx_space < pba) {
1926 pba -= needed_tx_space;
1927
1928 /* if short on Rx space, Rx wins and must trump Tx
1929 * adjustment
1930 */
1931 if (pba < min_rx_space)
1932 pba = min_rx_space;
1933 }
1934
1935 /* adjust PBA for jumbo frames */
1936 wr32(E1000_PBA, pba);
1937 }
1938
1939 /* flow control settings
1940 * The high water mark must be low enough to fit one full frame
1941 * after transmitting the pause frame. As such we must have enough
1942 * space to allow for us to complete our current transmit and then
1943 * receive the frame that is in progress from the link partner.
1944 * Set it to:
1945 * - the full Rx FIFO size minus one full Tx plus one full Rx frame
1946 */
1947 hwm = (pba << 10) - (adapter->max_frame_size + MAX_JUMBO_FRAME_SIZE);
1948
1949 fc->high_water = hwm & 0xFFFFFFF0; /* 16-byte granularity */
1950 fc->low_water = fc->high_water - 16;
1951 fc->pause_time = 0xFFFF;
1952 fc->send_xon = 1;
1953 fc->current_mode = fc->requested_mode;
1954
1955 /* disable receive for all VFs and wait one second */
1956 if (adapter->vfs_allocated_count) {
1957 int i;
1958
1959 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1960 adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
1961
1962 /* ping all the active vfs to let them know we are going down */
1963 igb_ping_all_vfs(adapter);
1964
1965 /* disable transmits and receives */
1966 wr32(E1000_VFRE, 0);
1967 wr32(E1000_VFTE, 0);
1968 }
1969
1970 /* Allow time for pending master requests to run */
1971 hw->mac.ops.reset_hw(hw);
1972 wr32(E1000_WUC, 0);
1973
1974 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
1975 /* need to resetup here after media swap */
1976 adapter->ei.get_invariants(hw);
1977 adapter->flags &= ~IGB_FLAG_MEDIA_RESET;
1978 }
1979 if ((mac->type == e1000_82575) &&
1980 (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
1981 igb_enable_mas(adapter);
1982 }
1983 if (hw->mac.ops.init_hw(hw))
1984 dev_err(&pdev->dev, "Hardware Error\n");
1985
1986 /* Flow control settings reset on hardware reset, so guarantee flow
1987 * control is off when forcing speed.
1988 */
1989 if (!hw->mac.autoneg)
1990 igb_force_mac_fc(hw);
1991
1992 igb_init_dmac(adapter, pba);
1993#ifdef CONFIG_IGB_HWMON
1994 /* Re-initialize the thermal sensor on i350 devices. */
1995 if (!test_bit(__IGB_DOWN, &adapter->state)) {
1996 if (mac->type == e1000_i350 && hw->bus.func == 0) {
1997 /* If present, re-initialize the external thermal sensor
1998 * interface.
1999 */
2000 if (adapter->ets)
2001 mac->ops.init_thermal_sensor_thresh(hw);
2002 }
2003 }
2004#endif
2005 /* Re-establish EEE setting */
2006 if (hw->phy.media_type == e1000_media_type_copper) {
2007 switch (mac->type) {
2008 case e1000_i350:
2009 case e1000_i210:
2010 case e1000_i211:
2011 igb_set_eee_i350(hw, true, true);
2012 break;
2013 case e1000_i354:
2014 igb_set_eee_i354(hw, true, true);
2015 break;
2016 default:
2017 break;
2018 }
2019 }
2020 if (!netif_running(adapter->netdev))
2021 igb_power_down_link(adapter);
2022
2023 igb_update_mng_vlan(adapter);
2024
2025 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2026 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
2027
2028 /* Re-enable PTP, where applicable. */
2029 igb_ptp_reset(adapter);
2030
2031 igb_get_phy_info(hw);
2032}
2033
2034static netdev_features_t igb_fix_features(struct net_device *netdev,
2035 netdev_features_t features)
2036{
2037 /* Since there is no support for separate Rx/Tx vlan accel
2038 * enable/disable make sure Tx flag is always in same state as Rx.
2039 */
2040 if (features & NETIF_F_HW_VLAN_CTAG_RX)
2041 features |= NETIF_F_HW_VLAN_CTAG_TX;
2042 else
2043 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2044
2045 return features;
2046}
2047
2048static int igb_set_features(struct net_device *netdev,
2049 netdev_features_t features)
2050{
2051 netdev_features_t changed = netdev->features ^ features;
2052 struct igb_adapter *adapter = netdev_priv(netdev);
2053
2054 if (changed & NETIF_F_HW_VLAN_CTAG_RX)
2055 igb_vlan_mode(netdev, features);
2056
2057 if (!(changed & (NETIF_F_RXALL | NETIF_F_NTUPLE)))
2058 return 0;
2059
2060 netdev->features = features;
2061
2062 if (netif_running(netdev))
2063 igb_reinit_locked(adapter);
2064 else
2065 igb_reset(adapter);
2066
2067 return 0;
2068}
2069
2070static int igb_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
2071 struct net_device *dev,
2072 const unsigned char *addr, u16 vid,
2073 u16 flags)
2074{
2075 /* guarantee we can provide a unique filter for the unicast address */
2076 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
2077 struct igb_adapter *adapter = netdev_priv(dev);
2078 struct e1000_hw *hw = &adapter->hw;
2079 int vfn = adapter->vfs_allocated_count;
2080 int rar_entries = hw->mac.rar_entry_count - (vfn + 1);
2081
2082 if (netdev_uc_count(dev) >= rar_entries)
2083 return -ENOMEM;
2084 }
2085
2086 return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
2087}
2088
2089static const struct net_device_ops igb_netdev_ops = {
2090 .ndo_open = igb_open,
2091 .ndo_stop = igb_close,
2092 .ndo_start_xmit = igb_xmit_frame,
2093 .ndo_get_stats64 = igb_get_stats64,
2094 .ndo_set_rx_mode = igb_set_rx_mode,
2095 .ndo_set_mac_address = igb_set_mac,
2096 .ndo_change_mtu = igb_change_mtu,
2097 .ndo_do_ioctl = igb_ioctl,
2098 .ndo_tx_timeout = igb_tx_timeout,
2099 .ndo_validate_addr = eth_validate_addr,
2100 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
2101 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
2102 .ndo_set_vf_mac = igb_ndo_set_vf_mac,
2103 .ndo_set_vf_vlan = igb_ndo_set_vf_vlan,
2104 .ndo_set_vf_rate = igb_ndo_set_vf_bw,
2105 .ndo_set_vf_spoofchk = igb_ndo_set_vf_spoofchk,
2106 .ndo_get_vf_config = igb_ndo_get_vf_config,
2107#ifdef CONFIG_NET_POLL_CONTROLLER
2108 .ndo_poll_controller = igb_netpoll,
2109#endif
2110 .ndo_fix_features = igb_fix_features,
2111 .ndo_set_features = igb_set_features,
2112 .ndo_fdb_add = igb_ndo_fdb_add,
2113 .ndo_features_check = passthru_features_check,
2114};
2115
2116/**
2117 * igb_set_fw_version - Configure version string for ethtool
2118 * @adapter: adapter struct
2119 **/
2120void igb_set_fw_version(struct igb_adapter *adapter)
2121{
2122 struct e1000_hw *hw = &adapter->hw;
2123 struct e1000_fw_version fw;
2124
2125 igb_get_fw_version(hw, &fw);
2126
2127 switch (hw->mac.type) {
2128 case e1000_i210:
2129 case e1000_i211:
2130 if (!(igb_get_flash_presence_i210(hw))) {
2131 snprintf(adapter->fw_version,
2132 sizeof(adapter->fw_version),
2133 "%2d.%2d-%d",
2134 fw.invm_major, fw.invm_minor,
2135 fw.invm_img_type);
2136 break;
2137 }
2138 /* fall through */
2139 default:
2140 /* if option is rom valid, display its version too */
2141 if (fw.or_valid) {
2142 snprintf(adapter->fw_version,
2143 sizeof(adapter->fw_version),
2144 "%d.%d, 0x%08x, %d.%d.%d",
2145 fw.eep_major, fw.eep_minor, fw.etrack_id,
2146 fw.or_major, fw.or_build, fw.or_patch);
2147 /* no option rom */
2148 } else if (fw.etrack_id != 0X0000) {
2149 snprintf(adapter->fw_version,
2150 sizeof(adapter->fw_version),
2151 "%d.%d, 0x%08x",
2152 fw.eep_major, fw.eep_minor, fw.etrack_id);
2153 } else {
2154 snprintf(adapter->fw_version,
2155 sizeof(adapter->fw_version),
2156 "%d.%d.%d",
2157 fw.eep_major, fw.eep_minor, fw.eep_build);
2158 }
2159 break;
2160 }
2161}
2162
2163/**
2164 * igb_init_mas - init Media Autosense feature if enabled in the NVM
2165 *
2166 * @adapter: adapter struct
2167 **/
2168static void igb_init_mas(struct igb_adapter *adapter)
2169{
2170 struct e1000_hw *hw = &adapter->hw;
2171 u16 eeprom_data;
2172
2173 hw->nvm.ops.read(hw, NVM_COMPAT, 1, &eeprom_data);
2174 switch (hw->bus.func) {
2175 case E1000_FUNC_0:
2176 if (eeprom_data & IGB_MAS_ENABLE_0) {
2177 adapter->flags |= IGB_FLAG_MAS_ENABLE;
2178 netdev_info(adapter->netdev,
2179 "MAS: Enabling Media Autosense for port %d\n",
2180 hw->bus.func);
2181 }
2182 break;
2183 case E1000_FUNC_1:
2184 if (eeprom_data & IGB_MAS_ENABLE_1) {
2185 adapter->flags |= IGB_FLAG_MAS_ENABLE;
2186 netdev_info(adapter->netdev,
2187 "MAS: Enabling Media Autosense for port %d\n",
2188 hw->bus.func);
2189 }
2190 break;
2191 case E1000_FUNC_2:
2192 if (eeprom_data & IGB_MAS_ENABLE_2) {
2193 adapter->flags |= IGB_FLAG_MAS_ENABLE;
2194 netdev_info(adapter->netdev,
2195 "MAS: Enabling Media Autosense for port %d\n",
2196 hw->bus.func);
2197 }
2198 break;
2199 case E1000_FUNC_3:
2200 if (eeprom_data & IGB_MAS_ENABLE_3) {
2201 adapter->flags |= IGB_FLAG_MAS_ENABLE;
2202 netdev_info(adapter->netdev,
2203 "MAS: Enabling Media Autosense for port %d\n",
2204 hw->bus.func);
2205 }
2206 break;
2207 default:
2208 /* Shouldn't get here */
2209 netdev_err(adapter->netdev,
2210 "MAS: Invalid port configuration, returning\n");
2211 break;
2212 }
2213}
2214
2215/**
2216 * igb_init_i2c - Init I2C interface
2217 * @adapter: pointer to adapter structure
2218 **/
2219static s32 igb_init_i2c(struct igb_adapter *adapter)
2220{
2221 s32 status = 0;
2222
2223 /* I2C interface supported on i350 devices */
2224 if (adapter->hw.mac.type != e1000_i350)
2225 return 0;
2226
2227 /* Initialize the i2c bus which is controlled by the registers.
2228 * This bus will use the i2c_algo_bit structue that implements
2229 * the protocol through toggling of the 4 bits in the register.
2230 */
2231 adapter->i2c_adap.owner = THIS_MODULE;
2232 adapter->i2c_algo = igb_i2c_algo;
2233 adapter->i2c_algo.data = adapter;
2234 adapter->i2c_adap.algo_data = &adapter->i2c_algo;
2235 adapter->i2c_adap.dev.parent = &adapter->pdev->dev;
2236 strlcpy(adapter->i2c_adap.name, "igb BB",
2237 sizeof(adapter->i2c_adap.name));
2238 status = i2c_bit_add_bus(&adapter->i2c_adap);
2239 return status;
2240}
2241
2242/**
2243 * igb_probe - Device Initialization Routine
2244 * @pdev: PCI device information struct
2245 * @ent: entry in igb_pci_tbl
2246 *
2247 * Returns 0 on success, negative on failure
2248 *
2249 * igb_probe initializes an adapter identified by a pci_dev structure.
2250 * The OS initialization, configuring of the adapter private structure,
2251 * and a hardware reset occur.
2252 **/
2253static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2254{
2255 struct net_device *netdev;
2256 struct igb_adapter *adapter;
2257 struct e1000_hw *hw;
2258 u16 eeprom_data = 0;
2259 s32 ret_val;
2260 static int global_quad_port_a; /* global quad port a indication */
2261 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
2262 int err, pci_using_dac;
2263 u8 part_str[E1000_PBANUM_LENGTH];
2264
2265 /* Catch broken hardware that put the wrong VF device ID in
2266 * the PCIe SR-IOV capability.
2267 */
2268 if (pdev->is_virtfn) {
2269 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
2270 pci_name(pdev), pdev->vendor, pdev->device);
2271 return -EINVAL;
2272 }
2273
2274 err = pci_enable_device_mem(pdev);
2275 if (err)
2276 return err;
2277
2278 pci_using_dac = 0;
2279 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
2280 if (!err) {
2281 pci_using_dac = 1;
2282 } else {
2283 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
2284 if (err) {
2285 dev_err(&pdev->dev,
2286 "No usable DMA configuration, aborting\n");
2287 goto err_dma;
2288 }
2289 }
2290
2291 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
2292 IORESOURCE_MEM),
2293 igb_driver_name);
2294 if (err)
2295 goto err_pci_reg;
2296
2297 pci_enable_pcie_error_reporting(pdev);
2298
2299 pci_set_master(pdev);
2300 pci_save_state(pdev);
2301
2302 err = -ENOMEM;
2303 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
2304 IGB_MAX_TX_QUEUES);
2305 if (!netdev)
2306 goto err_alloc_etherdev;
2307
2308 SET_NETDEV_DEV(netdev, &pdev->dev);
2309
2310 pci_set_drvdata(pdev, netdev);
2311 adapter = netdev_priv(netdev);
2312 adapter->netdev = netdev;
2313 adapter->pdev = pdev;
2314 hw = &adapter->hw;
2315 hw->back = adapter;
2316 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2317
2318 err = -EIO;
2319 adapter->io_addr = pci_iomap(pdev, 0, 0);
2320 if (!adapter->io_addr)
2321 goto err_ioremap;
2322 /* hw->hw_addr can be altered, we'll use adapter->io_addr for unmap */
2323 hw->hw_addr = adapter->io_addr;
2324
2325 netdev->netdev_ops = &igb_netdev_ops;
2326 igb_set_ethtool_ops(netdev);
2327 netdev->watchdog_timeo = 5 * HZ;
2328
2329 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2330
2331 netdev->mem_start = pci_resource_start(pdev, 0);
2332 netdev->mem_end = pci_resource_end(pdev, 0);
2333
2334 /* PCI config space info */
2335 hw->vendor_id = pdev->vendor;
2336 hw->device_id = pdev->device;
2337 hw->revision_id = pdev->revision;
2338 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2339 hw->subsystem_device_id = pdev->subsystem_device;
2340
2341 /* Copy the default MAC, PHY and NVM function pointers */
2342 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
2343 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
2344 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
2345 /* Initialize skew-specific constants */
2346 err = ei->get_invariants(hw);
2347 if (err)
2348 goto err_sw_init;
2349
2350 /* setup the private structure */
2351 err = igb_sw_init(adapter);
2352 if (err)
2353 goto err_sw_init;
2354
2355 igb_get_bus_info_pcie(hw);
2356
2357 hw->phy.autoneg_wait_to_complete = false;
2358
2359 /* Copper options */
2360 if (hw->phy.media_type == e1000_media_type_copper) {
2361 hw->phy.mdix = AUTO_ALL_MODES;
2362 hw->phy.disable_polarity_correction = false;
2363 hw->phy.ms_type = e1000_ms_hw_default;
2364 }
2365
2366 if (igb_check_reset_block(hw))
2367 dev_info(&pdev->dev,
2368 "PHY reset is blocked due to SOL/IDER session.\n");
2369
2370 /* features is initialized to 0 in allocation, it might have bits
2371 * set by igb_sw_init so we should use an or instead of an
2372 * assignment.
2373 */
2374 netdev->features |= NETIF_F_SG |
2375 NETIF_F_TSO |
2376 NETIF_F_TSO6 |
2377 NETIF_F_RXHASH |
2378 NETIF_F_RXCSUM |
2379 NETIF_F_HW_CSUM |
2380 NETIF_F_HW_VLAN_CTAG_RX |
2381 NETIF_F_HW_VLAN_CTAG_TX;
2382
2383 if (hw->mac.type >= e1000_82576)
2384 netdev->features |= NETIF_F_SCTP_CRC;
2385
2386 /* copy netdev features into list of user selectable features */
2387 netdev->hw_features |= netdev->features;
2388 netdev->hw_features |= NETIF_F_RXALL;
2389
2390 if (hw->mac.type >= e1000_i350)
2391 netdev->hw_features |= NETIF_F_NTUPLE;
2392
2393 /* set this bit last since it cannot be part of hw_features */
2394 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
2395
2396 netdev->vlan_features |= NETIF_F_SG |
2397 NETIF_F_TSO |
2398 NETIF_F_TSO6 |
2399 NETIF_F_HW_CSUM |
2400 NETIF_F_SCTP_CRC;
2401
2402 netdev->mpls_features |= NETIF_F_HW_CSUM;
2403 netdev->hw_enc_features |= NETIF_F_HW_CSUM;
2404
2405 netdev->priv_flags |= IFF_SUPP_NOFCS;
2406
2407 if (pci_using_dac) {
2408 netdev->features |= NETIF_F_HIGHDMA;
2409 netdev->vlan_features |= NETIF_F_HIGHDMA;
2410 }
2411
2412 netdev->priv_flags |= IFF_UNICAST_FLT;
2413
2414 adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
2415
2416 /* before reading the NVM, reset the controller to put the device in a
2417 * known good starting state
2418 */
2419 hw->mac.ops.reset_hw(hw);
2420
2421 /* make sure the NVM is good , i211/i210 parts can have special NVM
2422 * that doesn't contain a checksum
2423 */
2424 switch (hw->mac.type) {
2425 case e1000_i210:
2426 case e1000_i211:
2427 if (igb_get_flash_presence_i210(hw)) {
2428 if (hw->nvm.ops.validate(hw) < 0) {
2429 dev_err(&pdev->dev,
2430 "The NVM Checksum Is Not Valid\n");
2431 err = -EIO;
2432 goto err_eeprom;
2433 }
2434 }
2435 break;
2436 default:
2437 if (hw->nvm.ops.validate(hw) < 0) {
2438 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
2439 err = -EIO;
2440 goto err_eeprom;
2441 }
2442 break;
2443 }
2444
2445 /* copy the MAC address out of the NVM */
2446 if (hw->mac.ops.read_mac_addr(hw))
2447 dev_err(&pdev->dev, "NVM Read Error\n");
2448
2449 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2450
2451 if (!is_valid_ether_addr(netdev->dev_addr)) {
2452 dev_err(&pdev->dev, "Invalid MAC Address\n");
2453 err = -EIO;
2454 goto err_eeprom;
2455 }
2456
2457 /* get firmware version for ethtool -i */
2458 igb_set_fw_version(adapter);
2459
2460 /* configure RXPBSIZE and TXPBSIZE */
2461 if (hw->mac.type == e1000_i210) {
2462 wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
2463 wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
2464 }
2465
2466 setup_timer(&adapter->watchdog_timer, igb_watchdog,
2467 (unsigned long) adapter);
2468 setup_timer(&adapter->phy_info_timer, igb_update_phy_info,
2469 (unsigned long) adapter);
2470
2471 INIT_WORK(&adapter->reset_task, igb_reset_task);
2472 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
2473
2474 /* Initialize link properties that are user-changeable */
2475 adapter->fc_autoneg = true;
2476 hw->mac.autoneg = true;
2477 hw->phy.autoneg_advertised = 0x2f;
2478
2479 hw->fc.requested_mode = e1000_fc_default;
2480 hw->fc.current_mode = e1000_fc_default;
2481
2482 igb_validate_mdi_setting(hw);
2483
2484 /* By default, support wake on port A */
2485 if (hw->bus.func == 0)
2486 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2487
2488 /* Check the NVM for wake support on non-port A ports */
2489 if (hw->mac.type >= e1000_82580)
2490 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
2491 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
2492 &eeprom_data);
2493 else if (hw->bus.func == 1)
2494 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
2495
2496 if (eeprom_data & IGB_EEPROM_APME)
2497 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2498
2499 /* now that we have the eeprom settings, apply the special cases where
2500 * the eeprom may be wrong or the board simply won't support wake on
2501 * lan on a particular port
2502 */
2503 switch (pdev->device) {
2504 case E1000_DEV_ID_82575GB_QUAD_COPPER:
2505 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2506 break;
2507 case E1000_DEV_ID_82575EB_FIBER_SERDES:
2508 case E1000_DEV_ID_82576_FIBER:
2509 case E1000_DEV_ID_82576_SERDES:
2510 /* Wake events only supported on port A for dual fiber
2511 * regardless of eeprom setting
2512 */
2513 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
2514 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2515 break;
2516 case E1000_DEV_ID_82576_QUAD_COPPER:
2517 case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
2518 /* if quad port adapter, disable WoL on all but port A */
2519 if (global_quad_port_a != 0)
2520 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2521 else
2522 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
2523 /* Reset for multiple quad port adapters */
2524 if (++global_quad_port_a == 4)
2525 global_quad_port_a = 0;
2526 break;
2527 default:
2528 /* If the device can't wake, don't set software support */
2529 if (!device_can_wakeup(&adapter->pdev->dev))
2530 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2531 }
2532
2533 /* initialize the wol settings based on the eeprom settings */
2534 if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
2535 adapter->wol |= E1000_WUFC_MAG;
2536
2537 /* Some vendors want WoL disabled by default, but still supported */
2538 if ((hw->mac.type == e1000_i350) &&
2539 (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
2540 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2541 adapter->wol = 0;
2542 }
2543
2544 /* Some vendors want the ability to Use the EEPROM setting as
2545 * enable/disable only, and not for capability
2546 */
2547 if (((hw->mac.type == e1000_i350) ||
2548 (hw->mac.type == e1000_i354)) &&
2549 (pdev->subsystem_vendor == PCI_VENDOR_ID_DELL)) {
2550 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2551 adapter->wol = 0;
2552 }
2553 if (hw->mac.type == e1000_i350) {
2554 if (((pdev->subsystem_device == 0x5001) ||
2555 (pdev->subsystem_device == 0x5002)) &&
2556 (hw->bus.func == 0)) {
2557 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2558 adapter->wol = 0;
2559 }
2560 if (pdev->subsystem_device == 0x1F52)
2561 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2562 }
2563
2564 device_set_wakeup_enable(&adapter->pdev->dev,
2565 adapter->flags & IGB_FLAG_WOL_SUPPORTED);
2566
2567 /* reset the hardware with the new settings */
2568 igb_reset(adapter);
2569
2570 /* Init the I2C interface */
2571 err = igb_init_i2c(adapter);
2572 if (err) {
2573 dev_err(&pdev->dev, "failed to init i2c interface\n");
2574 goto err_eeprom;
2575 }
2576
2577 /* let the f/w know that the h/w is now under the control of the
2578 * driver.
2579 */
2580 igb_get_hw_control(adapter);
2581
2582 strcpy(netdev->name, "eth%d");
2583 err = register_netdev(netdev);
2584 if (err)
2585 goto err_register;
2586
2587 /* carrier off reporting is important to ethtool even BEFORE open */
2588 netif_carrier_off(netdev);
2589
2590#ifdef CONFIG_IGB_DCA
2591 if (dca_add_requester(&pdev->dev) == 0) {
2592 adapter->flags |= IGB_FLAG_DCA_ENABLED;
2593 dev_info(&pdev->dev, "DCA enabled\n");
2594 igb_setup_dca(adapter);
2595 }
2596
2597#endif
2598#ifdef CONFIG_IGB_HWMON
2599 /* Initialize the thermal sensor on i350 devices. */
2600 if (hw->mac.type == e1000_i350 && hw->bus.func == 0) {
2601 u16 ets_word;
2602
2603 /* Read the NVM to determine if this i350 device supports an
2604 * external thermal sensor.
2605 */
2606 hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word);
2607 if (ets_word != 0x0000 && ets_word != 0xFFFF)
2608 adapter->ets = true;
2609 else
2610 adapter->ets = false;
2611 if (igb_sysfs_init(adapter))
2612 dev_err(&pdev->dev,
2613 "failed to allocate sysfs resources\n");
2614 } else {
2615 adapter->ets = false;
2616 }
2617#endif
2618 /* Check if Media Autosense is enabled */
2619 adapter->ei = *ei;
2620 if (hw->dev_spec._82575.mas_capable)
2621 igb_init_mas(adapter);
2622
2623 /* do hw tstamp init after resetting */
2624 igb_ptp_init(adapter);
2625
2626 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
2627 /* print bus type/speed/width info, not applicable to i354 */
2628 if (hw->mac.type != e1000_i354) {
2629 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
2630 netdev->name,
2631 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
2632 (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
2633 "unknown"),
2634 ((hw->bus.width == e1000_bus_width_pcie_x4) ?
2635 "Width x4" :
2636 (hw->bus.width == e1000_bus_width_pcie_x2) ?
2637 "Width x2" :
2638 (hw->bus.width == e1000_bus_width_pcie_x1) ?
2639 "Width x1" : "unknown"), netdev->dev_addr);
2640 }
2641
2642 if ((hw->mac.type >= e1000_i210 ||
2643 igb_get_flash_presence_i210(hw))) {
2644 ret_val = igb_read_part_string(hw, part_str,
2645 E1000_PBANUM_LENGTH);
2646 } else {
2647 ret_val = -E1000_ERR_INVM_VALUE_NOT_FOUND;
2648 }
2649
2650 if (ret_val)
2651 strcpy(part_str, "Unknown");
2652 dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
2653 dev_info(&pdev->dev,
2654 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
2655 (adapter->flags & IGB_FLAG_HAS_MSIX) ? "MSI-X" :
2656 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
2657 adapter->num_rx_queues, adapter->num_tx_queues);
2658 if (hw->phy.media_type == e1000_media_type_copper) {
2659 switch (hw->mac.type) {
2660 case e1000_i350:
2661 case e1000_i210:
2662 case e1000_i211:
2663 /* Enable EEE for internal copper PHY devices */
2664 err = igb_set_eee_i350(hw, true, true);
2665 if ((!err) &&
2666 (!hw->dev_spec._82575.eee_disable)) {
2667 adapter->eee_advert =
2668 MDIO_EEE_100TX | MDIO_EEE_1000T;
2669 adapter->flags |= IGB_FLAG_EEE;
2670 }
2671 break;
2672 case e1000_i354:
2673 if ((rd32(E1000_CTRL_EXT) &
2674 E1000_CTRL_EXT_LINK_MODE_SGMII)) {
2675 err = igb_set_eee_i354(hw, true, true);
2676 if ((!err) &&
2677 (!hw->dev_spec._82575.eee_disable)) {
2678 adapter->eee_advert =
2679 MDIO_EEE_100TX | MDIO_EEE_1000T;
2680 adapter->flags |= IGB_FLAG_EEE;
2681 }
2682 }
2683 break;
2684 default:
2685 break;
2686 }
2687 }
2688 pm_runtime_put_noidle(&pdev->dev);
2689 return 0;
2690
2691err_register:
2692 igb_release_hw_control(adapter);
2693 memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap));
2694err_eeprom:
2695 if (!igb_check_reset_block(hw))
2696 igb_reset_phy(hw);
2697
2698 if (hw->flash_address)
2699 iounmap(hw->flash_address);
2700err_sw_init:
2701 kfree(adapter->shadow_vfta);
2702 igb_clear_interrupt_scheme(adapter);
2703#ifdef CONFIG_PCI_IOV
2704 igb_disable_sriov(pdev);
2705#endif
2706 pci_iounmap(pdev, adapter->io_addr);
2707err_ioremap:
2708 free_netdev(netdev);
2709err_alloc_etherdev:
2710 pci_release_selected_regions(pdev,
2711 pci_select_bars(pdev, IORESOURCE_MEM));
2712err_pci_reg:
2713err_dma:
2714 pci_disable_device(pdev);
2715 return err;
2716}
2717
2718#ifdef CONFIG_PCI_IOV
2719static int igb_disable_sriov(struct pci_dev *pdev)
2720{
2721 struct net_device *netdev = pci_get_drvdata(pdev);
2722 struct igb_adapter *adapter = netdev_priv(netdev);
2723 struct e1000_hw *hw = &adapter->hw;
2724
2725 /* reclaim resources allocated to VFs */
2726 if (adapter->vf_data) {
2727 /* disable iov and allow time for transactions to clear */
2728 if (pci_vfs_assigned(pdev)) {
2729 dev_warn(&pdev->dev,
2730 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
2731 return -EPERM;
2732 } else {
2733 pci_disable_sriov(pdev);
2734 msleep(500);
2735 }
2736
2737 kfree(adapter->vf_data);
2738 adapter->vf_data = NULL;
2739 adapter->vfs_allocated_count = 0;
2740 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
2741 wrfl();
2742 msleep(100);
2743 dev_info(&pdev->dev, "IOV Disabled\n");
2744
2745 /* Re-enable DMA Coalescing flag since IOV is turned off */
2746 adapter->flags |= IGB_FLAG_DMAC;
2747 }
2748
2749 return 0;
2750}
2751
2752static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs)
2753{
2754 struct net_device *netdev = pci_get_drvdata(pdev);
2755 struct igb_adapter *adapter = netdev_priv(netdev);
2756 int old_vfs = pci_num_vf(pdev);
2757 int err = 0;
2758 int i;
2759
2760 if (!(adapter->flags & IGB_FLAG_HAS_MSIX) || num_vfs > 7) {
2761 err = -EPERM;
2762 goto out;
2763 }
2764 if (!num_vfs)
2765 goto out;
2766
2767 if (old_vfs) {
2768 dev_info(&pdev->dev, "%d pre-allocated VFs found - override max_vfs setting of %d\n",
2769 old_vfs, max_vfs);
2770 adapter->vfs_allocated_count = old_vfs;
2771 } else
2772 adapter->vfs_allocated_count = num_vfs;
2773
2774 adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
2775 sizeof(struct vf_data_storage), GFP_KERNEL);
2776
2777 /* if allocation failed then we do not support SR-IOV */
2778 if (!adapter->vf_data) {
2779 adapter->vfs_allocated_count = 0;
2780 dev_err(&pdev->dev,
2781 "Unable to allocate memory for VF Data Storage\n");
2782 err = -ENOMEM;
2783 goto out;
2784 }
2785
2786 /* only call pci_enable_sriov() if no VFs are allocated already */
2787 if (!old_vfs) {
2788 err = pci_enable_sriov(pdev, adapter->vfs_allocated_count);
2789 if (err)
2790 goto err_out;
2791 }
2792 dev_info(&pdev->dev, "%d VFs allocated\n",
2793 adapter->vfs_allocated_count);
2794 for (i = 0; i < adapter->vfs_allocated_count; i++)
2795 igb_vf_configure(adapter, i);
2796
2797 /* DMA Coalescing is not supported in IOV mode. */
2798 adapter->flags &= ~IGB_FLAG_DMAC;
2799 goto out;
2800
2801err_out:
2802 kfree(adapter->vf_data);
2803 adapter->vf_data = NULL;
2804 adapter->vfs_allocated_count = 0;
2805out:
2806 return err;
2807}
2808
2809#endif
2810/**
2811 * igb_remove_i2c - Cleanup I2C interface
2812 * @adapter: pointer to adapter structure
2813 **/
2814static void igb_remove_i2c(struct igb_adapter *adapter)
2815{
2816 /* free the adapter bus structure */
2817 i2c_del_adapter(&adapter->i2c_adap);
2818}
2819
2820/**
2821 * igb_remove - Device Removal Routine
2822 * @pdev: PCI device information struct
2823 *
2824 * igb_remove is called by the PCI subsystem to alert the driver
2825 * that it should release a PCI device. The could be caused by a
2826 * Hot-Plug event, or because the driver is going to be removed from
2827 * memory.
2828 **/
2829static void igb_remove(struct pci_dev *pdev)
2830{
2831 struct net_device *netdev = pci_get_drvdata(pdev);
2832 struct igb_adapter *adapter = netdev_priv(netdev);
2833 struct e1000_hw *hw = &adapter->hw;
2834
2835 pm_runtime_get_noresume(&pdev->dev);
2836#ifdef CONFIG_IGB_HWMON
2837 igb_sysfs_exit(adapter);
2838#endif
2839 igb_remove_i2c(adapter);
2840 igb_ptp_stop(adapter);
2841 /* The watchdog timer may be rescheduled, so explicitly
2842 * disable watchdog from being rescheduled.
2843 */
2844 set_bit(__IGB_DOWN, &adapter->state);
2845 del_timer_sync(&adapter->watchdog_timer);
2846 del_timer_sync(&adapter->phy_info_timer);
2847
2848 cancel_work_sync(&adapter->reset_task);
2849 cancel_work_sync(&adapter->watchdog_task);
2850
2851#ifdef CONFIG_IGB_DCA
2852 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
2853 dev_info(&pdev->dev, "DCA disabled\n");
2854 dca_remove_requester(&pdev->dev);
2855 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
2856 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
2857 }
2858#endif
2859
2860 /* Release control of h/w to f/w. If f/w is AMT enabled, this
2861 * would have already happened in close and is redundant.
2862 */
2863 igb_release_hw_control(adapter);
2864
2865#ifdef CONFIG_PCI_IOV
2866 igb_disable_sriov(pdev);
2867#endif
2868
2869 unregister_netdev(netdev);
2870
2871 igb_clear_interrupt_scheme(adapter);
2872
2873 pci_iounmap(pdev, adapter->io_addr);
2874 if (hw->flash_address)
2875 iounmap(hw->flash_address);
2876 pci_release_selected_regions(pdev,
2877 pci_select_bars(pdev, IORESOURCE_MEM));
2878
2879 kfree(adapter->shadow_vfta);
2880 free_netdev(netdev);
2881
2882 pci_disable_pcie_error_reporting(pdev);
2883
2884 pci_disable_device(pdev);
2885}
2886
2887/**
2888 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
2889 * @adapter: board private structure to initialize
2890 *
2891 * This function initializes the vf specific data storage and then attempts to
2892 * allocate the VFs. The reason for ordering it this way is because it is much
2893 * mor expensive time wise to disable SR-IOV than it is to allocate and free
2894 * the memory for the VFs.
2895 **/
2896static void igb_probe_vfs(struct igb_adapter *adapter)
2897{
2898#ifdef CONFIG_PCI_IOV
2899 struct pci_dev *pdev = adapter->pdev;
2900 struct e1000_hw *hw = &adapter->hw;
2901
2902 /* Virtualization features not supported on i210 family. */
2903 if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
2904 return;
2905
2906 /* Of the below we really only want the effect of getting
2907 * IGB_FLAG_HAS_MSIX set (if available), without which
2908 * igb_enable_sriov() has no effect.
2909 */
2910 igb_set_interrupt_capability(adapter, true);
2911 igb_reset_interrupt_capability(adapter);
2912
2913 pci_sriov_set_totalvfs(pdev, 7);
2914 igb_enable_sriov(pdev, max_vfs);
2915
2916#endif /* CONFIG_PCI_IOV */
2917}
2918
2919static void igb_init_queue_configuration(struct igb_adapter *adapter)
2920{
2921 struct e1000_hw *hw = &adapter->hw;
2922 u32 max_rss_queues;
2923
2924 /* Determine the maximum number of RSS queues supported. */
2925 switch (hw->mac.type) {
2926 case e1000_i211:
2927 max_rss_queues = IGB_MAX_RX_QUEUES_I211;
2928 break;
2929 case e1000_82575:
2930 case e1000_i210:
2931 max_rss_queues = IGB_MAX_RX_QUEUES_82575;
2932 break;
2933 case e1000_i350:
2934 /* I350 cannot do RSS and SR-IOV at the same time */
2935 if (!!adapter->vfs_allocated_count) {
2936 max_rss_queues = 1;
2937 break;
2938 }
2939 /* fall through */
2940 case e1000_82576:
2941 if (!!adapter->vfs_allocated_count) {
2942 max_rss_queues = 2;
2943 break;
2944 }
2945 /* fall through */
2946 case e1000_82580:
2947 case e1000_i354:
2948 default:
2949 max_rss_queues = IGB_MAX_RX_QUEUES;
2950 break;
2951 }
2952
2953 adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
2954
2955 igb_set_flag_queue_pairs(adapter, max_rss_queues);
2956}
2957
2958void igb_set_flag_queue_pairs(struct igb_adapter *adapter,
2959 const u32 max_rss_queues)
2960{
2961 struct e1000_hw *hw = &adapter->hw;
2962
2963 /* Determine if we need to pair queues. */
2964 switch (hw->mac.type) {
2965 case e1000_82575:
2966 case e1000_i211:
2967 /* Device supports enough interrupts without queue pairing. */
2968 break;
2969 case e1000_82576:
2970 case e1000_82580:
2971 case e1000_i350:
2972 case e1000_i354:
2973 case e1000_i210:
2974 default:
2975 /* If rss_queues > half of max_rss_queues, pair the queues in
2976 * order to conserve interrupts due to limited supply.
2977 */
2978 if (adapter->rss_queues > (max_rss_queues / 2))
2979 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2980 else
2981 adapter->flags &= ~IGB_FLAG_QUEUE_PAIRS;
2982 break;
2983 }
2984}
2985
2986/**
2987 * igb_sw_init - Initialize general software structures (struct igb_adapter)
2988 * @adapter: board private structure to initialize
2989 *
2990 * igb_sw_init initializes the Adapter private data structure.
2991 * Fields are initialized based on PCI device information and
2992 * OS network device settings (MTU size).
2993 **/
2994static int igb_sw_init(struct igb_adapter *adapter)
2995{
2996 struct e1000_hw *hw = &adapter->hw;
2997 struct net_device *netdev = adapter->netdev;
2998 struct pci_dev *pdev = adapter->pdev;
2999
3000 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
3001
3002 /* set default ring sizes */
3003 adapter->tx_ring_count = IGB_DEFAULT_TXD;
3004 adapter->rx_ring_count = IGB_DEFAULT_RXD;
3005
3006 /* set default ITR values */
3007 adapter->rx_itr_setting = IGB_DEFAULT_ITR;
3008 adapter->tx_itr_setting = IGB_DEFAULT_ITR;
3009
3010 /* set default work limits */
3011 adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
3012
3013 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
3014 VLAN_HLEN;
3015 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3016
3017 spin_lock_init(&adapter->stats64_lock);
3018#ifdef CONFIG_PCI_IOV
3019 switch (hw->mac.type) {
3020 case e1000_82576:
3021 case e1000_i350:
3022 if (max_vfs > 7) {
3023 dev_warn(&pdev->dev,
3024 "Maximum of 7 VFs per PF, using max\n");
3025 max_vfs = adapter->vfs_allocated_count = 7;
3026 } else
3027 adapter->vfs_allocated_count = max_vfs;
3028 if (adapter->vfs_allocated_count)
3029 dev_warn(&pdev->dev,
3030 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
3031 break;
3032 default:
3033 break;
3034 }
3035#endif /* CONFIG_PCI_IOV */
3036
3037 /* Assume MSI-X interrupts, will be checked during IRQ allocation */
3038 adapter->flags |= IGB_FLAG_HAS_MSIX;
3039
3040 igb_probe_vfs(adapter);
3041
3042 igb_init_queue_configuration(adapter);
3043
3044 /* Setup and initialize a copy of the hw vlan table array */
3045 adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32),
3046 GFP_ATOMIC);
3047
3048 /* This call may decrease the number of queues */
3049 if (igb_init_interrupt_scheme(adapter, true)) {
3050 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
3051 return -ENOMEM;
3052 }
3053
3054 /* Explicitly disable IRQ since the NIC can be in any state. */
3055 igb_irq_disable(adapter);
3056
3057 if (hw->mac.type >= e1000_i350)
3058 adapter->flags &= ~IGB_FLAG_DMAC;
3059
3060 set_bit(__IGB_DOWN, &adapter->state);
3061 return 0;
3062}
3063
3064/**
3065 * igb_open - Called when a network interface is made active
3066 * @netdev: network interface device structure
3067 *
3068 * Returns 0 on success, negative value on failure
3069 *
3070 * The open entry point is called when a network interface is made
3071 * active by the system (IFF_UP). At this point all resources needed
3072 * for transmit and receive operations are allocated, the interrupt
3073 * handler is registered with the OS, the watchdog timer is started,
3074 * and the stack is notified that the interface is ready.
3075 **/
3076static int __igb_open(struct net_device *netdev, bool resuming)
3077{
3078 struct igb_adapter *adapter = netdev_priv(netdev);
3079 struct e1000_hw *hw = &adapter->hw;
3080 struct pci_dev *pdev = adapter->pdev;
3081 int err;
3082 int i;
3083
3084 /* disallow open during test */
3085 if (test_bit(__IGB_TESTING, &adapter->state)) {
3086 WARN_ON(resuming);
3087 return -EBUSY;
3088 }
3089
3090 if (!resuming)
3091 pm_runtime_get_sync(&pdev->dev);
3092
3093 netif_carrier_off(netdev);
3094
3095 /* allocate transmit descriptors */
3096 err = igb_setup_all_tx_resources(adapter);
3097 if (err)
3098 goto err_setup_tx;
3099
3100 /* allocate receive descriptors */
3101 err = igb_setup_all_rx_resources(adapter);
3102 if (err)
3103 goto err_setup_rx;
3104
3105 igb_power_up_link(adapter);
3106
3107 /* before we allocate an interrupt, we must be ready to handle it.
3108 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3109 * as soon as we call pci_request_irq, so we have to setup our
3110 * clean_rx handler before we do so.
3111 */
3112 igb_configure(adapter);
3113
3114 err = igb_request_irq(adapter);
3115 if (err)
3116 goto err_req_irq;
3117
3118 /* Notify the stack of the actual queue counts. */
3119 err = netif_set_real_num_tx_queues(adapter->netdev,
3120 adapter->num_tx_queues);
3121 if (err)
3122 goto err_set_queues;
3123
3124 err = netif_set_real_num_rx_queues(adapter->netdev,
3125 adapter->num_rx_queues);
3126 if (err)
3127 goto err_set_queues;
3128
3129 /* From here on the code is the same as igb_up() */
3130 clear_bit(__IGB_DOWN, &adapter->state);
3131
3132 for (i = 0; i < adapter->num_q_vectors; i++)
3133 napi_enable(&(adapter->q_vector[i]->napi));
3134
3135 /* Clear any pending interrupts. */
3136 rd32(E1000_ICR);
3137
3138 igb_irq_enable(adapter);
3139
3140 /* notify VFs that reset has been completed */
3141 if (adapter->vfs_allocated_count) {
3142 u32 reg_data = rd32(E1000_CTRL_EXT);
3143
3144 reg_data |= E1000_CTRL_EXT_PFRSTD;
3145 wr32(E1000_CTRL_EXT, reg_data);
3146 }
3147
3148 netif_tx_start_all_queues(netdev);
3149
3150 if (!resuming)
3151 pm_runtime_put(&pdev->dev);
3152
3153 /* start the watchdog. */
3154 hw->mac.get_link_status = 1;
3155 schedule_work(&adapter->watchdog_task);
3156
3157 return 0;
3158
3159err_set_queues:
3160 igb_free_irq(adapter);
3161err_req_irq:
3162 igb_release_hw_control(adapter);
3163 igb_power_down_link(adapter);
3164 igb_free_all_rx_resources(adapter);
3165err_setup_rx:
3166 igb_free_all_tx_resources(adapter);
3167err_setup_tx:
3168 igb_reset(adapter);
3169 if (!resuming)
3170 pm_runtime_put(&pdev->dev);
3171
3172 return err;
3173}
3174
3175int igb_open(struct net_device *netdev)
3176{
3177 return __igb_open(netdev, false);
3178}
3179
3180/**
3181 * igb_close - Disables a network interface
3182 * @netdev: network interface device structure
3183 *
3184 * Returns 0, this is not allowed to fail
3185 *
3186 * The close entry point is called when an interface is de-activated
3187 * by the OS. The hardware is still under the driver's control, but
3188 * needs to be disabled. A global MAC reset is issued to stop the
3189 * hardware, and all transmit and receive resources are freed.
3190 **/
3191static int __igb_close(struct net_device *netdev, bool suspending)
3192{
3193 struct igb_adapter *adapter = netdev_priv(netdev);
3194 struct pci_dev *pdev = adapter->pdev;
3195
3196 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
3197
3198 if (!suspending)
3199 pm_runtime_get_sync(&pdev->dev);
3200
3201 igb_down(adapter);
3202 igb_free_irq(adapter);
3203
3204 igb_free_all_tx_resources(adapter);
3205 igb_free_all_rx_resources(adapter);
3206
3207 if (!suspending)
3208 pm_runtime_put_sync(&pdev->dev);
3209 return 0;
3210}
3211
3212int igb_close(struct net_device *netdev)
3213{
3214 return __igb_close(netdev, false);
3215}
3216
3217/**
3218 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
3219 * @tx_ring: tx descriptor ring (for a specific queue) to setup
3220 *
3221 * Return 0 on success, negative on failure
3222 **/
3223int igb_setup_tx_resources(struct igb_ring *tx_ring)
3224{
3225 struct device *dev = tx_ring->dev;
3226 int size;
3227
3228 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3229
3230 tx_ring->tx_buffer_info = vzalloc(size);
3231 if (!tx_ring->tx_buffer_info)
3232 goto err;
3233
3234 /* round up to nearest 4K */
3235 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
3236 tx_ring->size = ALIGN(tx_ring->size, 4096);
3237
3238 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
3239 &tx_ring->dma, GFP_KERNEL);
3240 if (!tx_ring->desc)
3241 goto err;
3242
3243 tx_ring->next_to_use = 0;
3244 tx_ring->next_to_clean = 0;
3245
3246 return 0;
3247
3248err:
3249 vfree(tx_ring->tx_buffer_info);
3250 tx_ring->tx_buffer_info = NULL;
3251 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
3252 return -ENOMEM;
3253}
3254
3255/**
3256 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
3257 * (Descriptors) for all queues
3258 * @adapter: board private structure
3259 *
3260 * Return 0 on success, negative on failure
3261 **/
3262static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
3263{
3264 struct pci_dev *pdev = adapter->pdev;
3265 int i, err = 0;
3266
3267 for (i = 0; i < adapter->num_tx_queues; i++) {
3268 err = igb_setup_tx_resources(adapter->tx_ring[i]);
3269 if (err) {
3270 dev_err(&pdev->dev,
3271 "Allocation for Tx Queue %u failed\n", i);
3272 for (i--; i >= 0; i--)
3273 igb_free_tx_resources(adapter->tx_ring[i]);
3274 break;
3275 }
3276 }
3277
3278 return err;
3279}
3280
3281/**
3282 * igb_setup_tctl - configure the transmit control registers
3283 * @adapter: Board private structure
3284 **/
3285void igb_setup_tctl(struct igb_adapter *adapter)
3286{
3287 struct e1000_hw *hw = &adapter->hw;
3288 u32 tctl;
3289
3290 /* disable queue 0 which is enabled by default on 82575 and 82576 */
3291 wr32(E1000_TXDCTL(0), 0);
3292
3293 /* Program the Transmit Control Register */
3294 tctl = rd32(E1000_TCTL);
3295 tctl &= ~E1000_TCTL_CT;
3296 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
3297 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
3298
3299 igb_config_collision_dist(hw);
3300
3301 /* Enable transmits */
3302 tctl |= E1000_TCTL_EN;
3303
3304 wr32(E1000_TCTL, tctl);
3305}
3306
3307/**
3308 * igb_configure_tx_ring - Configure transmit ring after Reset
3309 * @adapter: board private structure
3310 * @ring: tx ring to configure
3311 *
3312 * Configure a transmit ring after a reset.
3313 **/
3314void igb_configure_tx_ring(struct igb_adapter *adapter,
3315 struct igb_ring *ring)
3316{
3317 struct e1000_hw *hw = &adapter->hw;
3318 u32 txdctl = 0;
3319 u64 tdba = ring->dma;
3320 int reg_idx = ring->reg_idx;
3321
3322 /* disable the queue */
3323 wr32(E1000_TXDCTL(reg_idx), 0);
3324 wrfl();
3325 mdelay(10);
3326
3327 wr32(E1000_TDLEN(reg_idx),
3328 ring->count * sizeof(union e1000_adv_tx_desc));
3329 wr32(E1000_TDBAL(reg_idx),
3330 tdba & 0x00000000ffffffffULL);
3331 wr32(E1000_TDBAH(reg_idx), tdba >> 32);
3332
3333 ring->tail = hw->hw_addr + E1000_TDT(reg_idx);
3334 wr32(E1000_TDH(reg_idx), 0);
3335 writel(0, ring->tail);
3336
3337 txdctl |= IGB_TX_PTHRESH;
3338 txdctl |= IGB_TX_HTHRESH << 8;
3339 txdctl |= IGB_TX_WTHRESH << 16;
3340
3341 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
3342 wr32(E1000_TXDCTL(reg_idx), txdctl);
3343}
3344
3345/**
3346 * igb_configure_tx - Configure transmit Unit after Reset
3347 * @adapter: board private structure
3348 *
3349 * Configure the Tx unit of the MAC after a reset.
3350 **/
3351static void igb_configure_tx(struct igb_adapter *adapter)
3352{
3353 int i;
3354
3355 for (i = 0; i < adapter->num_tx_queues; i++)
3356 igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
3357}
3358
3359/**
3360 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
3361 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3362 *
3363 * Returns 0 on success, negative on failure
3364 **/
3365int igb_setup_rx_resources(struct igb_ring *rx_ring)
3366{
3367 struct device *dev = rx_ring->dev;
3368 int size;
3369
3370 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3371
3372 rx_ring->rx_buffer_info = vzalloc(size);
3373 if (!rx_ring->rx_buffer_info)
3374 goto err;
3375
3376 /* Round up to nearest 4K */
3377 rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
3378 rx_ring->size = ALIGN(rx_ring->size, 4096);
3379
3380 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
3381 &rx_ring->dma, GFP_KERNEL);
3382 if (!rx_ring->desc)
3383 goto err;
3384
3385 rx_ring->next_to_alloc = 0;
3386 rx_ring->next_to_clean = 0;
3387 rx_ring->next_to_use = 0;
3388
3389 return 0;
3390
3391err:
3392 vfree(rx_ring->rx_buffer_info);
3393 rx_ring->rx_buffer_info = NULL;
3394 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
3395 return -ENOMEM;
3396}
3397
3398/**
3399 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
3400 * (Descriptors) for all queues
3401 * @adapter: board private structure
3402 *
3403 * Return 0 on success, negative on failure
3404 **/
3405static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
3406{
3407 struct pci_dev *pdev = adapter->pdev;
3408 int i, err = 0;
3409
3410 for (i = 0; i < adapter->num_rx_queues; i++) {
3411 err = igb_setup_rx_resources(adapter->rx_ring[i]);
3412 if (err) {
3413 dev_err(&pdev->dev,
3414 "Allocation for Rx Queue %u failed\n", i);
3415 for (i--; i >= 0; i--)
3416 igb_free_rx_resources(adapter->rx_ring[i]);
3417 break;
3418 }
3419 }
3420
3421 return err;
3422}
3423
3424/**
3425 * igb_setup_mrqc - configure the multiple receive queue control registers
3426 * @adapter: Board private structure
3427 **/
3428static void igb_setup_mrqc(struct igb_adapter *adapter)
3429{
3430 struct e1000_hw *hw = &adapter->hw;
3431 u32 mrqc, rxcsum;
3432 u32 j, num_rx_queues;
3433 u32 rss_key[10];
3434
3435 netdev_rss_key_fill(rss_key, sizeof(rss_key));
3436 for (j = 0; j < 10; j++)
3437 wr32(E1000_RSSRK(j), rss_key[j]);
3438
3439 num_rx_queues = adapter->rss_queues;
3440
3441 switch (hw->mac.type) {
3442 case e1000_82576:
3443 /* 82576 supports 2 RSS queues for SR-IOV */
3444 if (adapter->vfs_allocated_count)
3445 num_rx_queues = 2;
3446 break;
3447 default:
3448 break;
3449 }
3450
3451 if (adapter->rss_indir_tbl_init != num_rx_queues) {
3452 for (j = 0; j < IGB_RETA_SIZE; j++)
3453 adapter->rss_indir_tbl[j] =
3454 (j * num_rx_queues) / IGB_RETA_SIZE;
3455 adapter->rss_indir_tbl_init = num_rx_queues;
3456 }
3457 igb_write_rss_indir_tbl(adapter);
3458
3459 /* Disable raw packet checksumming so that RSS hash is placed in
3460 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
3461 * offloads as they are enabled by default
3462 */
3463 rxcsum = rd32(E1000_RXCSUM);
3464 rxcsum |= E1000_RXCSUM_PCSD;
3465
3466 if (adapter->hw.mac.type >= e1000_82576)
3467 /* Enable Receive Checksum Offload for SCTP */
3468 rxcsum |= E1000_RXCSUM_CRCOFL;
3469
3470 /* Don't need to set TUOFL or IPOFL, they default to 1 */
3471 wr32(E1000_RXCSUM, rxcsum);
3472
3473 /* Generate RSS hash based on packet types, TCP/UDP
3474 * port numbers and/or IPv4/v6 src and dst addresses
3475 */
3476 mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
3477 E1000_MRQC_RSS_FIELD_IPV4_TCP |
3478 E1000_MRQC_RSS_FIELD_IPV6 |
3479 E1000_MRQC_RSS_FIELD_IPV6_TCP |
3480 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
3481
3482 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
3483 mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
3484 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
3485 mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
3486
3487 /* If VMDq is enabled then we set the appropriate mode for that, else
3488 * we default to RSS so that an RSS hash is calculated per packet even
3489 * if we are only using one queue
3490 */
3491 if (adapter->vfs_allocated_count) {
3492 if (hw->mac.type > e1000_82575) {
3493 /* Set the default pool for the PF's first queue */
3494 u32 vtctl = rd32(E1000_VT_CTL);
3495
3496 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
3497 E1000_VT_CTL_DISABLE_DEF_POOL);
3498 vtctl |= adapter->vfs_allocated_count <<
3499 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
3500 wr32(E1000_VT_CTL, vtctl);
3501 }
3502 if (adapter->rss_queues > 1)
3503 mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_MQ;
3504 else
3505 mrqc |= E1000_MRQC_ENABLE_VMDQ;
3506 } else {
3507 if (hw->mac.type != e1000_i211)
3508 mrqc |= E1000_MRQC_ENABLE_RSS_MQ;
3509 }
3510 igb_vmm_control(adapter);
3511
3512 wr32(E1000_MRQC, mrqc);
3513}
3514
3515/**
3516 * igb_setup_rctl - configure the receive control registers
3517 * @adapter: Board private structure
3518 **/
3519void igb_setup_rctl(struct igb_adapter *adapter)
3520{
3521 struct e1000_hw *hw = &adapter->hw;
3522 u32 rctl;
3523
3524 rctl = rd32(E1000_RCTL);
3525
3526 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3527 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
3528
3529 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
3530 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3531
3532 /* enable stripping of CRC. It's unlikely this will break BMC
3533 * redirection as it did with e1000. Newer features require
3534 * that the HW strips the CRC.
3535 */
3536 rctl |= E1000_RCTL_SECRC;
3537
3538 /* disable store bad packets and clear size bits. */
3539 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
3540
3541 /* enable LPE to allow for reception of jumbo frames */
3542 rctl |= E1000_RCTL_LPE;
3543
3544 /* disable queue 0 to prevent tail write w/o re-config */
3545 wr32(E1000_RXDCTL(0), 0);
3546
3547 /* Attention!!! For SR-IOV PF driver operations you must enable
3548 * queue drop for all VF and PF queues to prevent head of line blocking
3549 * if an un-trusted VF does not provide descriptors to hardware.
3550 */
3551 if (adapter->vfs_allocated_count) {
3552 /* set all queue drop enable bits */
3553 wr32(E1000_QDE, ALL_QUEUES);
3554 }
3555
3556 /* This is useful for sniffing bad packets. */
3557 if (adapter->netdev->features & NETIF_F_RXALL) {
3558 /* UPE and MPE will be handled by normal PROMISC logic
3559 * in e1000e_set_rx_mode
3560 */
3561 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3562 E1000_RCTL_BAM | /* RX All Bcast Pkts */
3563 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3564
3565 rctl &= ~(E1000_RCTL_DPF | /* Allow filtered pause */
3566 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3567 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3568 * and that breaks VLANs.
3569 */
3570 }
3571
3572 wr32(E1000_RCTL, rctl);
3573}
3574
3575static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
3576 int vfn)
3577{
3578 struct e1000_hw *hw = &adapter->hw;
3579 u32 vmolr;
3580
3581 if (size > MAX_JUMBO_FRAME_SIZE)
3582 size = MAX_JUMBO_FRAME_SIZE;
3583
3584 vmolr = rd32(E1000_VMOLR(vfn));
3585 vmolr &= ~E1000_VMOLR_RLPML_MASK;
3586 vmolr |= size | E1000_VMOLR_LPE;
3587 wr32(E1000_VMOLR(vfn), vmolr);
3588
3589 return 0;
3590}
3591
3592static inline void igb_set_vf_vlan_strip(struct igb_adapter *adapter,
3593 int vfn, bool enable)
3594{
3595 struct e1000_hw *hw = &adapter->hw;
3596 u32 val, reg;
3597
3598 if (hw->mac.type < e1000_82576)
3599 return;
3600
3601 if (hw->mac.type == e1000_i350)
3602 reg = E1000_DVMOLR(vfn);
3603 else
3604 reg = E1000_VMOLR(vfn);
3605
3606 val = rd32(reg);
3607 if (enable)
3608 val |= E1000_VMOLR_STRVLAN;
3609 else
3610 val &= ~(E1000_VMOLR_STRVLAN);
3611 wr32(reg, val);
3612}
3613
3614static inline void igb_set_vmolr(struct igb_adapter *adapter,
3615 int vfn, bool aupe)
3616{
3617 struct e1000_hw *hw = &adapter->hw;
3618 u32 vmolr;
3619
3620 /* This register exists only on 82576 and newer so if we are older then
3621 * we should exit and do nothing
3622 */
3623 if (hw->mac.type < e1000_82576)
3624 return;
3625
3626 vmolr = rd32(E1000_VMOLR(vfn));
3627 if (aupe)
3628 vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
3629 else
3630 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
3631
3632 /* clear all bits that might not be set */
3633 vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
3634
3635 if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
3636 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
3637 /* for VMDq only allow the VFs and pool 0 to accept broadcast and
3638 * multicast packets
3639 */
3640 if (vfn <= adapter->vfs_allocated_count)
3641 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
3642
3643 wr32(E1000_VMOLR(vfn), vmolr);
3644}
3645
3646/**
3647 * igb_configure_rx_ring - Configure a receive ring after Reset
3648 * @adapter: board private structure
3649 * @ring: receive ring to be configured
3650 *
3651 * Configure the Rx unit of the MAC after a reset.
3652 **/
3653void igb_configure_rx_ring(struct igb_adapter *adapter,
3654 struct igb_ring *ring)
3655{
3656 struct e1000_hw *hw = &adapter->hw;
3657 u64 rdba = ring->dma;
3658 int reg_idx = ring->reg_idx;
3659 u32 srrctl = 0, rxdctl = 0;
3660
3661 /* disable the queue */
3662 wr32(E1000_RXDCTL(reg_idx), 0);
3663
3664 /* Set DMA base address registers */
3665 wr32(E1000_RDBAL(reg_idx),
3666 rdba & 0x00000000ffffffffULL);
3667 wr32(E1000_RDBAH(reg_idx), rdba >> 32);
3668 wr32(E1000_RDLEN(reg_idx),
3669 ring->count * sizeof(union e1000_adv_rx_desc));
3670
3671 /* initialize head and tail */
3672 ring->tail = hw->hw_addr + E1000_RDT(reg_idx);
3673 wr32(E1000_RDH(reg_idx), 0);
3674 writel(0, ring->tail);
3675
3676 /* set descriptor configuration */
3677 srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
3678 srrctl |= IGB_RX_BUFSZ >> E1000_SRRCTL_BSIZEPKT_SHIFT;
3679 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
3680 if (hw->mac.type >= e1000_82580)
3681 srrctl |= E1000_SRRCTL_TIMESTAMP;
3682 /* Only set Drop Enable if we are supporting multiple queues */
3683 if (adapter->vfs_allocated_count || adapter->num_rx_queues > 1)
3684 srrctl |= E1000_SRRCTL_DROP_EN;
3685
3686 wr32(E1000_SRRCTL(reg_idx), srrctl);
3687
3688 /* set filtering for VMDQ pools */
3689 igb_set_vmolr(adapter, reg_idx & 0x7, true);
3690
3691 rxdctl |= IGB_RX_PTHRESH;
3692 rxdctl |= IGB_RX_HTHRESH << 8;
3693 rxdctl |= IGB_RX_WTHRESH << 16;
3694
3695 /* enable receive descriptor fetching */
3696 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
3697 wr32(E1000_RXDCTL(reg_idx), rxdctl);
3698}
3699
3700/**
3701 * igb_configure_rx - Configure receive Unit after Reset
3702 * @adapter: board private structure
3703 *
3704 * Configure the Rx unit of the MAC after a reset.
3705 **/
3706static void igb_configure_rx(struct igb_adapter *adapter)
3707{
3708 int i;
3709
3710 /* set the correct pool for the PF default MAC address in entry 0 */
3711 igb_rar_set_qsel(adapter, adapter->hw.mac.addr, 0,
3712 adapter->vfs_allocated_count);
3713
3714 /* Setup the HW Rx Head and Tail Descriptor Pointers and
3715 * the Base and Length of the Rx Descriptor Ring
3716 */
3717 for (i = 0; i < adapter->num_rx_queues; i++)
3718 igb_configure_rx_ring(adapter, adapter->rx_ring[i]);
3719}
3720
3721/**
3722 * igb_free_tx_resources - Free Tx Resources per Queue
3723 * @tx_ring: Tx descriptor ring for a specific queue
3724 *
3725 * Free all transmit software resources
3726 **/
3727void igb_free_tx_resources(struct igb_ring *tx_ring)
3728{
3729 igb_clean_tx_ring(tx_ring);
3730
3731 vfree(tx_ring->tx_buffer_info);
3732 tx_ring->tx_buffer_info = NULL;
3733
3734 /* if not set, then don't free */
3735 if (!tx_ring->desc)
3736 return;
3737
3738 dma_free_coherent(tx_ring->dev, tx_ring->size,
3739 tx_ring->desc, tx_ring->dma);
3740
3741 tx_ring->desc = NULL;
3742}
3743
3744/**
3745 * igb_free_all_tx_resources - Free Tx Resources for All Queues
3746 * @adapter: board private structure
3747 *
3748 * Free all transmit software resources
3749 **/
3750static void igb_free_all_tx_resources(struct igb_adapter *adapter)
3751{
3752 int i;
3753
3754 for (i = 0; i < adapter->num_tx_queues; i++)
3755 if (adapter->tx_ring[i])
3756 igb_free_tx_resources(adapter->tx_ring[i]);
3757}
3758
3759void igb_unmap_and_free_tx_resource(struct igb_ring *ring,
3760 struct igb_tx_buffer *tx_buffer)
3761{
3762 if (tx_buffer->skb) {
3763 dev_kfree_skb_any(tx_buffer->skb);
3764 if (dma_unmap_len(tx_buffer, len))
3765 dma_unmap_single(ring->dev,
3766 dma_unmap_addr(tx_buffer, dma),
3767 dma_unmap_len(tx_buffer, len),
3768 DMA_TO_DEVICE);
3769 } else if (dma_unmap_len(tx_buffer, len)) {
3770 dma_unmap_page(ring->dev,
3771 dma_unmap_addr(tx_buffer, dma),
3772 dma_unmap_len(tx_buffer, len),
3773 DMA_TO_DEVICE);
3774 }
3775 tx_buffer->next_to_watch = NULL;
3776 tx_buffer->skb = NULL;
3777 dma_unmap_len_set(tx_buffer, len, 0);
3778 /* buffer_info must be completely set up in the transmit path */
3779}
3780
3781/**
3782 * igb_clean_tx_ring - Free Tx Buffers
3783 * @tx_ring: ring to be cleaned
3784 **/
3785static void igb_clean_tx_ring(struct igb_ring *tx_ring)
3786{
3787 struct igb_tx_buffer *buffer_info;
3788 unsigned long size;
3789 u16 i;
3790
3791 if (!tx_ring->tx_buffer_info)
3792 return;
3793 /* Free all the Tx ring sk_buffs */
3794
3795 for (i = 0; i < tx_ring->count; i++) {
3796 buffer_info = &tx_ring->tx_buffer_info[i];
3797 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
3798 }
3799
3800 netdev_tx_reset_queue(txring_txq(tx_ring));
3801
3802 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3803 memset(tx_ring->tx_buffer_info, 0, size);
3804
3805 /* Zero out the descriptor ring */
3806 memset(tx_ring->desc, 0, tx_ring->size);
3807
3808 tx_ring->next_to_use = 0;
3809 tx_ring->next_to_clean = 0;
3810}
3811
3812/**
3813 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
3814 * @adapter: board private structure
3815 **/
3816static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
3817{
3818 int i;
3819
3820 for (i = 0; i < adapter->num_tx_queues; i++)
3821 if (adapter->tx_ring[i])
3822 igb_clean_tx_ring(adapter->tx_ring[i]);
3823}
3824
3825/**
3826 * igb_free_rx_resources - Free Rx Resources
3827 * @rx_ring: ring to clean the resources from
3828 *
3829 * Free all receive software resources
3830 **/
3831void igb_free_rx_resources(struct igb_ring *rx_ring)
3832{
3833 igb_clean_rx_ring(rx_ring);
3834
3835 vfree(rx_ring->rx_buffer_info);
3836 rx_ring->rx_buffer_info = NULL;
3837
3838 /* if not set, then don't free */
3839 if (!rx_ring->desc)
3840 return;
3841
3842 dma_free_coherent(rx_ring->dev, rx_ring->size,
3843 rx_ring->desc, rx_ring->dma);
3844
3845 rx_ring->desc = NULL;
3846}
3847
3848/**
3849 * igb_free_all_rx_resources - Free Rx Resources for All Queues
3850 * @adapter: board private structure
3851 *
3852 * Free all receive software resources
3853 **/
3854static void igb_free_all_rx_resources(struct igb_adapter *adapter)
3855{
3856 int i;
3857
3858 for (i = 0; i < adapter->num_rx_queues; i++)
3859 if (adapter->rx_ring[i])
3860 igb_free_rx_resources(adapter->rx_ring[i]);
3861}
3862
3863/**
3864 * igb_clean_rx_ring - Free Rx Buffers per Queue
3865 * @rx_ring: ring to free buffers from
3866 **/
3867static void igb_clean_rx_ring(struct igb_ring *rx_ring)
3868{
3869 unsigned long size;
3870 u16 i;
3871
3872 if (rx_ring->skb)
3873 dev_kfree_skb(rx_ring->skb);
3874 rx_ring->skb = NULL;
3875
3876 if (!rx_ring->rx_buffer_info)
3877 return;
3878
3879 /* Free all the Rx ring sk_buffs */
3880 for (i = 0; i < rx_ring->count; i++) {
3881 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
3882
3883 if (!buffer_info->page)
3884 continue;
3885
3886 dma_unmap_page(rx_ring->dev,
3887 buffer_info->dma,
3888 PAGE_SIZE,
3889 DMA_FROM_DEVICE);
3890 __free_page(buffer_info->page);
3891
3892 buffer_info->page = NULL;
3893 }
3894
3895 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3896 memset(rx_ring->rx_buffer_info, 0, size);
3897
3898 /* Zero out the descriptor ring */
3899 memset(rx_ring->desc, 0, rx_ring->size);
3900
3901 rx_ring->next_to_alloc = 0;
3902 rx_ring->next_to_clean = 0;
3903 rx_ring->next_to_use = 0;
3904}
3905
3906/**
3907 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
3908 * @adapter: board private structure
3909 **/
3910static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
3911{
3912 int i;
3913
3914 for (i = 0; i < adapter->num_rx_queues; i++)
3915 if (adapter->rx_ring[i])
3916 igb_clean_rx_ring(adapter->rx_ring[i]);
3917}
3918
3919/**
3920 * igb_set_mac - Change the Ethernet Address of the NIC
3921 * @netdev: network interface device structure
3922 * @p: pointer to an address structure
3923 *
3924 * Returns 0 on success, negative on failure
3925 **/
3926static int igb_set_mac(struct net_device *netdev, void *p)
3927{
3928 struct igb_adapter *adapter = netdev_priv(netdev);
3929 struct e1000_hw *hw = &adapter->hw;
3930 struct sockaddr *addr = p;
3931
3932 if (!is_valid_ether_addr(addr->sa_data))
3933 return -EADDRNOTAVAIL;
3934
3935 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3936 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3937
3938 /* set the correct pool for the new PF MAC address in entry 0 */
3939 igb_rar_set_qsel(adapter, hw->mac.addr, 0,
3940 adapter->vfs_allocated_count);
3941
3942 return 0;
3943}
3944
3945/**
3946 * igb_write_mc_addr_list - write multicast addresses to MTA
3947 * @netdev: network interface device structure
3948 *
3949 * Writes multicast address list to the MTA hash table.
3950 * Returns: -ENOMEM on failure
3951 * 0 on no addresses written
3952 * X on writing X addresses to MTA
3953 **/
3954static int igb_write_mc_addr_list(struct net_device *netdev)
3955{
3956 struct igb_adapter *adapter = netdev_priv(netdev);
3957 struct e1000_hw *hw = &adapter->hw;
3958 struct netdev_hw_addr *ha;
3959 u8 *mta_list;
3960 int i;
3961
3962 if (netdev_mc_empty(netdev)) {
3963 /* nothing to program, so clear mc list */
3964 igb_update_mc_addr_list(hw, NULL, 0);
3965 igb_restore_vf_multicasts(adapter);
3966 return 0;
3967 }
3968
3969 mta_list = kzalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3970 if (!mta_list)
3971 return -ENOMEM;
3972
3973 /* The shared function expects a packed array of only addresses. */
3974 i = 0;
3975 netdev_for_each_mc_addr(ha, netdev)
3976 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3977
3978 igb_update_mc_addr_list(hw, mta_list, i);
3979 kfree(mta_list);
3980
3981 return netdev_mc_count(netdev);
3982}
3983
3984/**
3985 * igb_write_uc_addr_list - write unicast addresses to RAR table
3986 * @netdev: network interface device structure
3987 *
3988 * Writes unicast address list to the RAR table.
3989 * Returns: -ENOMEM on failure/insufficient address space
3990 * 0 on no addresses written
3991 * X on writing X addresses to the RAR table
3992 **/
3993static int igb_write_uc_addr_list(struct net_device *netdev)
3994{
3995 struct igb_adapter *adapter = netdev_priv(netdev);
3996 struct e1000_hw *hw = &adapter->hw;
3997 unsigned int vfn = adapter->vfs_allocated_count;
3998 unsigned int rar_entries = hw->mac.rar_entry_count - (vfn + 1);
3999 int count = 0;
4000
4001 /* return ENOMEM indicating insufficient memory for addresses */
4002 if (netdev_uc_count(netdev) > rar_entries)
4003 return -ENOMEM;
4004
4005 if (!netdev_uc_empty(netdev) && rar_entries) {
4006 struct netdev_hw_addr *ha;
4007
4008 netdev_for_each_uc_addr(ha, netdev) {
4009 if (!rar_entries)
4010 break;
4011 igb_rar_set_qsel(adapter, ha->addr,
4012 rar_entries--,
4013 vfn);
4014 count++;
4015 }
4016 }
4017 /* write the addresses in reverse order to avoid write combining */
4018 for (; rar_entries > 0 ; rar_entries--) {
4019 wr32(E1000_RAH(rar_entries), 0);
4020 wr32(E1000_RAL(rar_entries), 0);
4021 }
4022 wrfl();
4023
4024 return count;
4025}
4026
4027static int igb_vlan_promisc_enable(struct igb_adapter *adapter)
4028{
4029 struct e1000_hw *hw = &adapter->hw;
4030 u32 i, pf_id;
4031
4032 switch (hw->mac.type) {
4033 case e1000_i210:
4034 case e1000_i211:
4035 case e1000_i350:
4036 /* VLAN filtering needed for VLAN prio filter */
4037 if (adapter->netdev->features & NETIF_F_NTUPLE)
4038 break;
4039 /* fall through */
4040 case e1000_82576:
4041 case e1000_82580:
4042 case e1000_i354:
4043 /* VLAN filtering needed for pool filtering */
4044 if (adapter->vfs_allocated_count)
4045 break;
4046 /* fall through */
4047 default:
4048 return 1;
4049 }
4050
4051 /* We are already in VLAN promisc, nothing to do */
4052 if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
4053 return 0;
4054
4055 if (!adapter->vfs_allocated_count)
4056 goto set_vfta;
4057
4058 /* Add PF to all active pools */
4059 pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
4060
4061 for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
4062 u32 vlvf = rd32(E1000_VLVF(i));
4063
4064 vlvf |= 1 << pf_id;
4065 wr32(E1000_VLVF(i), vlvf);
4066 }
4067
4068set_vfta:
4069 /* Set all bits in the VLAN filter table array */
4070 for (i = E1000_VLAN_FILTER_TBL_SIZE; i--;)
4071 hw->mac.ops.write_vfta(hw, i, ~0U);
4072
4073 /* Set flag so we don't redo unnecessary work */
4074 adapter->flags |= IGB_FLAG_VLAN_PROMISC;
4075
4076 return 0;
4077}
4078
4079#define VFTA_BLOCK_SIZE 8
4080static void igb_scrub_vfta(struct igb_adapter *adapter, u32 vfta_offset)
4081{
4082 struct e1000_hw *hw = &adapter->hw;
4083 u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4084 u32 vid_start = vfta_offset * 32;
4085 u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4086 u32 i, vid, word, bits, pf_id;
4087
4088 /* guarantee that we don't scrub out management VLAN */
4089 vid = adapter->mng_vlan_id;
4090 if (vid >= vid_start && vid < vid_end)
4091 vfta[(vid - vid_start) / 32] |= 1 << (vid % 32);
4092
4093 if (!adapter->vfs_allocated_count)
4094 goto set_vfta;
4095
4096 pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
4097
4098 for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
4099 u32 vlvf = rd32(E1000_VLVF(i));
4100
4101 /* pull VLAN ID from VLVF */
4102 vid = vlvf & VLAN_VID_MASK;
4103
4104 /* only concern ourselves with a certain range */
4105 if (vid < vid_start || vid >= vid_end)
4106 continue;
4107
4108 if (vlvf & E1000_VLVF_VLANID_ENABLE) {
4109 /* record VLAN ID in VFTA */
4110 vfta[(vid - vid_start) / 32] |= 1 << (vid % 32);
4111
4112 /* if PF is part of this then continue */
4113 if (test_bit(vid, adapter->active_vlans))
4114 continue;
4115 }
4116
4117 /* remove PF from the pool */
4118 bits = ~(1 << pf_id);
4119 bits &= rd32(E1000_VLVF(i));
4120 wr32(E1000_VLVF(i), bits);
4121 }
4122
4123set_vfta:
4124 /* extract values from active_vlans and write back to VFTA */
4125 for (i = VFTA_BLOCK_SIZE; i--;) {
4126 vid = (vfta_offset + i) * 32;
4127 word = vid / BITS_PER_LONG;
4128 bits = vid % BITS_PER_LONG;
4129
4130 vfta[i] |= adapter->active_vlans[word] >> bits;
4131
4132 hw->mac.ops.write_vfta(hw, vfta_offset + i, vfta[i]);
4133 }
4134}
4135
4136static void igb_vlan_promisc_disable(struct igb_adapter *adapter)
4137{
4138 u32 i;
4139
4140 /* We are not in VLAN promisc, nothing to do */
4141 if (!(adapter->flags & IGB_FLAG_VLAN_PROMISC))
4142 return;
4143
4144 /* Set flag so we don't redo unnecessary work */
4145 adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
4146
4147 for (i = 0; i < E1000_VLAN_FILTER_TBL_SIZE; i += VFTA_BLOCK_SIZE)
4148 igb_scrub_vfta(adapter, i);
4149}
4150
4151/**
4152 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
4153 * @netdev: network interface device structure
4154 *
4155 * The set_rx_mode entry point is called whenever the unicast or multicast
4156 * address lists or the network interface flags are updated. This routine is
4157 * responsible for configuring the hardware for proper unicast, multicast,
4158 * promiscuous mode, and all-multi behavior.
4159 **/
4160static void igb_set_rx_mode(struct net_device *netdev)
4161{
4162 struct igb_adapter *adapter = netdev_priv(netdev);
4163 struct e1000_hw *hw = &adapter->hw;
4164 unsigned int vfn = adapter->vfs_allocated_count;
4165 u32 rctl = 0, vmolr = 0;
4166 int count;
4167
4168 /* Check for Promiscuous and All Multicast modes */
4169 if (netdev->flags & IFF_PROMISC) {
4170 rctl |= E1000_RCTL_UPE | E1000_RCTL_MPE;
4171 vmolr |= E1000_VMOLR_MPME;
4172
4173 /* enable use of UTA filter to force packets to default pool */
4174 if (hw->mac.type == e1000_82576)
4175 vmolr |= E1000_VMOLR_ROPE;
4176 } else {
4177 if (netdev->flags & IFF_ALLMULTI) {
4178 rctl |= E1000_RCTL_MPE;
4179 vmolr |= E1000_VMOLR_MPME;
4180 } else {
4181 /* Write addresses to the MTA, if the attempt fails
4182 * then we should just turn on promiscuous mode so
4183 * that we can at least receive multicast traffic
4184 */
4185 count = igb_write_mc_addr_list(netdev);
4186 if (count < 0) {
4187 rctl |= E1000_RCTL_MPE;
4188 vmolr |= E1000_VMOLR_MPME;
4189 } else if (count) {
4190 vmolr |= E1000_VMOLR_ROMPE;
4191 }
4192 }
4193 }
4194
4195 /* Write addresses to available RAR registers, if there is not
4196 * sufficient space to store all the addresses then enable
4197 * unicast promiscuous mode
4198 */
4199 count = igb_write_uc_addr_list(netdev);
4200 if (count < 0) {
4201 rctl |= E1000_RCTL_UPE;
4202 vmolr |= E1000_VMOLR_ROPE;
4203 }
4204
4205 /* enable VLAN filtering by default */
4206 rctl |= E1000_RCTL_VFE;
4207
4208 /* disable VLAN filtering for modes that require it */
4209 if ((netdev->flags & IFF_PROMISC) ||
4210 (netdev->features & NETIF_F_RXALL)) {
4211 /* if we fail to set all rules then just clear VFE */
4212 if (igb_vlan_promisc_enable(adapter))
4213 rctl &= ~E1000_RCTL_VFE;
4214 } else {
4215 igb_vlan_promisc_disable(adapter);
4216 }
4217
4218 /* update state of unicast, multicast, and VLAN filtering modes */
4219 rctl |= rd32(E1000_RCTL) & ~(E1000_RCTL_UPE | E1000_RCTL_MPE |
4220 E1000_RCTL_VFE);
4221 wr32(E1000_RCTL, rctl);
4222
4223 /* In order to support SR-IOV and eventually VMDq it is necessary to set
4224 * the VMOLR to enable the appropriate modes. Without this workaround
4225 * we will have issues with VLAN tag stripping not being done for frames
4226 * that are only arriving because we are the default pool
4227 */
4228 if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
4229 return;
4230
4231 /* set UTA to appropriate mode */
4232 igb_set_uta(adapter, !!(vmolr & E1000_VMOLR_ROPE));
4233
4234 vmolr |= rd32(E1000_VMOLR(vfn)) &
4235 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
4236
4237 /* enable Rx jumbo frames, no need for restriction */
4238 vmolr &= ~E1000_VMOLR_RLPML_MASK;
4239 vmolr |= MAX_JUMBO_FRAME_SIZE | E1000_VMOLR_LPE;
4240
4241 wr32(E1000_VMOLR(vfn), vmolr);
4242 wr32(E1000_RLPML, MAX_JUMBO_FRAME_SIZE);
4243
4244 igb_restore_vf_multicasts(adapter);
4245}
4246
4247static void igb_check_wvbr(struct igb_adapter *adapter)
4248{
4249 struct e1000_hw *hw = &adapter->hw;
4250 u32 wvbr = 0;
4251
4252 switch (hw->mac.type) {
4253 case e1000_82576:
4254 case e1000_i350:
4255 wvbr = rd32(E1000_WVBR);
4256 if (!wvbr)
4257 return;
4258 break;
4259 default:
4260 break;
4261 }
4262
4263 adapter->wvbr |= wvbr;
4264}
4265
4266#define IGB_STAGGERED_QUEUE_OFFSET 8
4267
4268static void igb_spoof_check(struct igb_adapter *adapter)
4269{
4270 int j;
4271
4272 if (!adapter->wvbr)
4273 return;
4274
4275 for (j = 0; j < adapter->vfs_allocated_count; j++) {
4276 if (adapter->wvbr & (1 << j) ||
4277 adapter->wvbr & (1 << (j + IGB_STAGGERED_QUEUE_OFFSET))) {
4278 dev_warn(&adapter->pdev->dev,
4279 "Spoof event(s) detected on VF %d\n", j);
4280 adapter->wvbr &=
4281 ~((1 << j) |
4282 (1 << (j + IGB_STAGGERED_QUEUE_OFFSET)));
4283 }
4284 }
4285}
4286
4287/* Need to wait a few seconds after link up to get diagnostic information from
4288 * the phy
4289 */
4290static void igb_update_phy_info(unsigned long data)
4291{
4292 struct igb_adapter *adapter = (struct igb_adapter *) data;
4293 igb_get_phy_info(&adapter->hw);
4294}
4295
4296/**
4297 * igb_has_link - check shared code for link and determine up/down
4298 * @adapter: pointer to driver private info
4299 **/
4300bool igb_has_link(struct igb_adapter *adapter)
4301{
4302 struct e1000_hw *hw = &adapter->hw;
4303 bool link_active = false;
4304
4305 /* get_link_status is set on LSC (link status) interrupt or
4306 * rx sequence error interrupt. get_link_status will stay
4307 * false until the e1000_check_for_link establishes link
4308 * for copper adapters ONLY
4309 */
4310 switch (hw->phy.media_type) {
4311 case e1000_media_type_copper:
4312 if (!hw->mac.get_link_status)
4313 return true;
4314 case e1000_media_type_internal_serdes:
4315 hw->mac.ops.check_for_link(hw);
4316 link_active = !hw->mac.get_link_status;
4317 break;
4318 default:
4319 case e1000_media_type_unknown:
4320 break;
4321 }
4322
4323 if (((hw->mac.type == e1000_i210) ||
4324 (hw->mac.type == e1000_i211)) &&
4325 (hw->phy.id == I210_I_PHY_ID)) {
4326 if (!netif_carrier_ok(adapter->netdev)) {
4327 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
4328 } else if (!(adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)) {
4329 adapter->flags |= IGB_FLAG_NEED_LINK_UPDATE;
4330 adapter->link_check_timeout = jiffies;
4331 }
4332 }
4333
4334 return link_active;
4335}
4336
4337static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
4338{
4339 bool ret = false;
4340 u32 ctrl_ext, thstat;
4341
4342 /* check for thermal sensor event on i350 copper only */
4343 if (hw->mac.type == e1000_i350) {
4344 thstat = rd32(E1000_THSTAT);
4345 ctrl_ext = rd32(E1000_CTRL_EXT);
4346
4347 if ((hw->phy.media_type == e1000_media_type_copper) &&
4348 !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII))
4349 ret = !!(thstat & event);
4350 }
4351
4352 return ret;
4353}
4354
4355/**
4356 * igb_check_lvmmc - check for malformed packets received
4357 * and indicated in LVMMC register
4358 * @adapter: pointer to adapter
4359 **/
4360static void igb_check_lvmmc(struct igb_adapter *adapter)
4361{
4362 struct e1000_hw *hw = &adapter->hw;
4363 u32 lvmmc;
4364
4365 lvmmc = rd32(E1000_LVMMC);
4366 if (lvmmc) {
4367 if (unlikely(net_ratelimit())) {
4368 netdev_warn(adapter->netdev,
4369 "malformed Tx packet detected and dropped, LVMMC:0x%08x\n",
4370 lvmmc);
4371 }
4372 }
4373}
4374
4375/**
4376 * igb_watchdog - Timer Call-back
4377 * @data: pointer to adapter cast into an unsigned long
4378 **/
4379static void igb_watchdog(unsigned long data)
4380{
4381 struct igb_adapter *adapter = (struct igb_adapter *)data;
4382 /* Do the rest outside of interrupt context */
4383 schedule_work(&adapter->watchdog_task);
4384}
4385
4386static void igb_watchdog_task(struct work_struct *work)
4387{
4388 struct igb_adapter *adapter = container_of(work,
4389 struct igb_adapter,
4390 watchdog_task);
4391 struct e1000_hw *hw = &adapter->hw;
4392 struct e1000_phy_info *phy = &hw->phy;
4393 struct net_device *netdev = adapter->netdev;
4394 u32 link;
4395 int i;
4396 u32 connsw;
4397 u16 phy_data, retry_count = 20;
4398
4399 link = igb_has_link(adapter);
4400
4401 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) {
4402 if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
4403 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
4404 else
4405 link = false;
4406 }
4407
4408 /* Force link down if we have fiber to swap to */
4409 if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
4410 if (hw->phy.media_type == e1000_media_type_copper) {
4411 connsw = rd32(E1000_CONNSW);
4412 if (!(connsw & E1000_CONNSW_AUTOSENSE_EN))
4413 link = 0;
4414 }
4415 }
4416 if (link) {
4417 /* Perform a reset if the media type changed. */
4418 if (hw->dev_spec._82575.media_changed) {
4419 hw->dev_spec._82575.media_changed = false;
4420 adapter->flags |= IGB_FLAG_MEDIA_RESET;
4421 igb_reset(adapter);
4422 }
4423 /* Cancel scheduled suspend requests. */
4424 pm_runtime_resume(netdev->dev.parent);
4425
4426 if (!netif_carrier_ok(netdev)) {
4427 u32 ctrl;
4428
4429 hw->mac.ops.get_speed_and_duplex(hw,
4430 &adapter->link_speed,
4431 &adapter->link_duplex);
4432
4433 ctrl = rd32(E1000_CTRL);
4434 /* Links status message must follow this format */
4435 netdev_info(netdev,
4436 "igb: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4437 netdev->name,
4438 adapter->link_speed,
4439 adapter->link_duplex == FULL_DUPLEX ?
4440 "Full" : "Half",
4441 (ctrl & E1000_CTRL_TFCE) &&
4442 (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
4443 (ctrl & E1000_CTRL_RFCE) ? "RX" :
4444 (ctrl & E1000_CTRL_TFCE) ? "TX" : "None");
4445
4446 /* disable EEE if enabled */
4447 if ((adapter->flags & IGB_FLAG_EEE) &&
4448 (adapter->link_duplex == HALF_DUPLEX)) {
4449 dev_info(&adapter->pdev->dev,
4450 "EEE Disabled: unsupported at half duplex. Re-enable using ethtool when at full duplex.\n");
4451 adapter->hw.dev_spec._82575.eee_disable = true;
4452 adapter->flags &= ~IGB_FLAG_EEE;
4453 }
4454
4455 /* check if SmartSpeed worked */
4456 igb_check_downshift(hw);
4457 if (phy->speed_downgraded)
4458 netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
4459
4460 /* check for thermal sensor event */
4461 if (igb_thermal_sensor_event(hw,
4462 E1000_THSTAT_LINK_THROTTLE))
4463 netdev_info(netdev, "The network adapter link speed was downshifted because it overheated\n");
4464
4465 /* adjust timeout factor according to speed/duplex */
4466 adapter->tx_timeout_factor = 1;
4467 switch (adapter->link_speed) {
4468 case SPEED_10:
4469 adapter->tx_timeout_factor = 14;
4470 break;
4471 case SPEED_100:
4472 /* maybe add some timeout factor ? */
4473 break;
4474 }
4475
4476 if (adapter->link_speed != SPEED_1000)
4477 goto no_wait;
4478
4479 /* wait for Remote receiver status OK */
4480retry_read_status:
4481 if (!igb_read_phy_reg(hw, PHY_1000T_STATUS,
4482 &phy_data)) {
4483 if (!(phy_data & SR_1000T_REMOTE_RX_STATUS) &&
4484 retry_count) {
4485 msleep(100);
4486 retry_count--;
4487 goto retry_read_status;
4488 } else if (!retry_count) {
4489 dev_err(&adapter->pdev->dev, "exceed max 2 second\n");
4490 }
4491 } else {
4492 dev_err(&adapter->pdev->dev, "read 1000Base-T Status Reg\n");
4493 }
4494no_wait:
4495 netif_carrier_on(netdev);
4496
4497 igb_ping_all_vfs(adapter);
4498 igb_check_vf_rate_limit(adapter);
4499
4500 /* link state has changed, schedule phy info update */
4501 if (!test_bit(__IGB_DOWN, &adapter->state))
4502 mod_timer(&adapter->phy_info_timer,
4503 round_jiffies(jiffies + 2 * HZ));
4504 }
4505 } else {
4506 if (netif_carrier_ok(netdev)) {
4507 adapter->link_speed = 0;
4508 adapter->link_duplex = 0;
4509
4510 /* check for thermal sensor event */
4511 if (igb_thermal_sensor_event(hw,
4512 E1000_THSTAT_PWR_DOWN)) {
4513 netdev_err(netdev, "The network adapter was stopped because it overheated\n");
4514 }
4515
4516 /* Links status message must follow this format */
4517 netdev_info(netdev, "igb: %s NIC Link is Down\n",
4518 netdev->name);
4519 netif_carrier_off(netdev);
4520
4521 igb_ping_all_vfs(adapter);
4522
4523 /* link state has changed, schedule phy info update */
4524 if (!test_bit(__IGB_DOWN, &adapter->state))
4525 mod_timer(&adapter->phy_info_timer,
4526 round_jiffies(jiffies + 2 * HZ));
4527
4528 /* link is down, time to check for alternate media */
4529 if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
4530 igb_check_swap_media(adapter);
4531 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
4532 schedule_work(&adapter->reset_task);
4533 /* return immediately */
4534 return;
4535 }
4536 }
4537 pm_schedule_suspend(netdev->dev.parent,
4538 MSEC_PER_SEC * 5);
4539
4540 /* also check for alternate media here */
4541 } else if (!netif_carrier_ok(netdev) &&
4542 (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
4543 igb_check_swap_media(adapter);
4544 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
4545 schedule_work(&adapter->reset_task);
4546 /* return immediately */
4547 return;
4548 }
4549 }
4550 }
4551
4552 spin_lock(&adapter->stats64_lock);
4553 igb_update_stats(adapter, &adapter->stats64);
4554 spin_unlock(&adapter->stats64_lock);
4555
4556 for (i = 0; i < adapter->num_tx_queues; i++) {
4557 struct igb_ring *tx_ring = adapter->tx_ring[i];
4558 if (!netif_carrier_ok(netdev)) {
4559 /* We've lost link, so the controller stops DMA,
4560 * but we've got queued Tx work that's never going
4561 * to get done, so reset controller to flush Tx.
4562 * (Do the reset outside of interrupt context).
4563 */
4564 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
4565 adapter->tx_timeout_count++;
4566 schedule_work(&adapter->reset_task);
4567 /* return immediately since reset is imminent */
4568 return;
4569 }
4570 }
4571
4572 /* Force detection of hung controller every watchdog period */
4573 set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
4574 }
4575
4576 /* Cause software interrupt to ensure Rx ring is cleaned */
4577 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
4578 u32 eics = 0;
4579
4580 for (i = 0; i < adapter->num_q_vectors; i++)
4581 eics |= adapter->q_vector[i]->eims_value;
4582 wr32(E1000_EICS, eics);
4583 } else {
4584 wr32(E1000_ICS, E1000_ICS_RXDMT0);
4585 }
4586
4587 igb_spoof_check(adapter);
4588 igb_ptp_rx_hang(adapter);
4589
4590 /* Check LVMMC register on i350/i354 only */
4591 if ((adapter->hw.mac.type == e1000_i350) ||
4592 (adapter->hw.mac.type == e1000_i354))
4593 igb_check_lvmmc(adapter);
4594
4595 /* Reset the timer */
4596 if (!test_bit(__IGB_DOWN, &adapter->state)) {
4597 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)
4598 mod_timer(&adapter->watchdog_timer,
4599 round_jiffies(jiffies + HZ));
4600 else
4601 mod_timer(&adapter->watchdog_timer,
4602 round_jiffies(jiffies + 2 * HZ));
4603 }
4604}
4605
4606enum latency_range {
4607 lowest_latency = 0,
4608 low_latency = 1,
4609 bulk_latency = 2,
4610 latency_invalid = 255
4611};
4612
4613/**
4614 * igb_update_ring_itr - update the dynamic ITR value based on packet size
4615 * @q_vector: pointer to q_vector
4616 *
4617 * Stores a new ITR value based on strictly on packet size. This
4618 * algorithm is less sophisticated than that used in igb_update_itr,
4619 * due to the difficulty of synchronizing statistics across multiple
4620 * receive rings. The divisors and thresholds used by this function
4621 * were determined based on theoretical maximum wire speed and testing
4622 * data, in order to minimize response time while increasing bulk
4623 * throughput.
4624 * This functionality is controlled by ethtool's coalescing settings.
4625 * NOTE: This function is called only when operating in a multiqueue
4626 * receive environment.
4627 **/
4628static void igb_update_ring_itr(struct igb_q_vector *q_vector)
4629{
4630 int new_val = q_vector->itr_val;
4631 int avg_wire_size = 0;
4632 struct igb_adapter *adapter = q_vector->adapter;
4633 unsigned int packets;
4634
4635 /* For non-gigabit speeds, just fix the interrupt rate at 4000
4636 * ints/sec - ITR timer value of 120 ticks.
4637 */
4638 if (adapter->link_speed != SPEED_1000) {
4639 new_val = IGB_4K_ITR;
4640 goto set_itr_val;
4641 }
4642
4643 packets = q_vector->rx.total_packets;
4644 if (packets)
4645 avg_wire_size = q_vector->rx.total_bytes / packets;
4646
4647 packets = q_vector->tx.total_packets;
4648 if (packets)
4649 avg_wire_size = max_t(u32, avg_wire_size,
4650 q_vector->tx.total_bytes / packets);
4651
4652 /* if avg_wire_size isn't set no work was done */
4653 if (!avg_wire_size)
4654 goto clear_counts;
4655
4656 /* Add 24 bytes to size to account for CRC, preamble, and gap */
4657 avg_wire_size += 24;
4658
4659 /* Don't starve jumbo frames */
4660 avg_wire_size = min(avg_wire_size, 3000);
4661
4662 /* Give a little boost to mid-size frames */
4663 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
4664 new_val = avg_wire_size / 3;
4665 else
4666 new_val = avg_wire_size / 2;
4667
4668 /* conservative mode (itr 3) eliminates the lowest_latency setting */
4669 if (new_val < IGB_20K_ITR &&
4670 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4671 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4672 new_val = IGB_20K_ITR;
4673
4674set_itr_val:
4675 if (new_val != q_vector->itr_val) {
4676 q_vector->itr_val = new_val;
4677 q_vector->set_itr = 1;
4678 }
4679clear_counts:
4680 q_vector->rx.total_bytes = 0;
4681 q_vector->rx.total_packets = 0;
4682 q_vector->tx.total_bytes = 0;
4683 q_vector->tx.total_packets = 0;
4684}
4685
4686/**
4687 * igb_update_itr - update the dynamic ITR value based on statistics
4688 * @q_vector: pointer to q_vector
4689 * @ring_container: ring info to update the itr for
4690 *
4691 * Stores a new ITR value based on packets and byte
4692 * counts during the last interrupt. The advantage of per interrupt
4693 * computation is faster updates and more accurate ITR for the current
4694 * traffic pattern. Constants in this function were computed
4695 * based on theoretical maximum wire speed and thresholds were set based
4696 * on testing data as well as attempting to minimize response time
4697 * while increasing bulk throughput.
4698 * This functionality is controlled by ethtool's coalescing settings.
4699 * NOTE: These calculations are only valid when operating in a single-
4700 * queue environment.
4701 **/
4702static void igb_update_itr(struct igb_q_vector *q_vector,
4703 struct igb_ring_container *ring_container)
4704{
4705 unsigned int packets = ring_container->total_packets;
4706 unsigned int bytes = ring_container->total_bytes;
4707 u8 itrval = ring_container->itr;
4708
4709 /* no packets, exit with status unchanged */
4710 if (packets == 0)
4711 return;
4712
4713 switch (itrval) {
4714 case lowest_latency:
4715 /* handle TSO and jumbo frames */
4716 if (bytes/packets > 8000)
4717 itrval = bulk_latency;
4718 else if ((packets < 5) && (bytes > 512))
4719 itrval = low_latency;
4720 break;
4721 case low_latency: /* 50 usec aka 20000 ints/s */
4722 if (bytes > 10000) {
4723 /* this if handles the TSO accounting */
4724 if (bytes/packets > 8000)
4725 itrval = bulk_latency;
4726 else if ((packets < 10) || ((bytes/packets) > 1200))
4727 itrval = bulk_latency;
4728 else if ((packets > 35))
4729 itrval = lowest_latency;
4730 } else if (bytes/packets > 2000) {
4731 itrval = bulk_latency;
4732 } else if (packets <= 2 && bytes < 512) {
4733 itrval = lowest_latency;
4734 }
4735 break;
4736 case bulk_latency: /* 250 usec aka 4000 ints/s */
4737 if (bytes > 25000) {
4738 if (packets > 35)
4739 itrval = low_latency;
4740 } else if (bytes < 1500) {
4741 itrval = low_latency;
4742 }
4743 break;
4744 }
4745
4746 /* clear work counters since we have the values we need */
4747 ring_container->total_bytes = 0;
4748 ring_container->total_packets = 0;
4749
4750 /* write updated itr to ring container */
4751 ring_container->itr = itrval;
4752}
4753
4754static void igb_set_itr(struct igb_q_vector *q_vector)
4755{
4756 struct igb_adapter *adapter = q_vector->adapter;
4757 u32 new_itr = q_vector->itr_val;
4758 u8 current_itr = 0;
4759
4760 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
4761 if (adapter->link_speed != SPEED_1000) {
4762 current_itr = 0;
4763 new_itr = IGB_4K_ITR;
4764 goto set_itr_now;
4765 }
4766
4767 igb_update_itr(q_vector, &q_vector->tx);
4768 igb_update_itr(q_vector, &q_vector->rx);
4769
4770 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
4771
4772 /* conservative mode (itr 3) eliminates the lowest_latency setting */
4773 if (current_itr == lowest_latency &&
4774 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4775 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4776 current_itr = low_latency;
4777
4778 switch (current_itr) {
4779 /* counts and packets in update_itr are dependent on these numbers */
4780 case lowest_latency:
4781 new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
4782 break;
4783 case low_latency:
4784 new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
4785 break;
4786 case bulk_latency:
4787 new_itr = IGB_4K_ITR; /* 4,000 ints/sec */
4788 break;
4789 default:
4790 break;
4791 }
4792
4793set_itr_now:
4794 if (new_itr != q_vector->itr_val) {
4795 /* this attempts to bias the interrupt rate towards Bulk
4796 * by adding intermediate steps when interrupt rate is
4797 * increasing
4798 */
4799 new_itr = new_itr > q_vector->itr_val ?
4800 max((new_itr * q_vector->itr_val) /
4801 (new_itr + (q_vector->itr_val >> 2)),
4802 new_itr) : new_itr;
4803 /* Don't write the value here; it resets the adapter's
4804 * internal timer, and causes us to delay far longer than
4805 * we should between interrupts. Instead, we write the ITR
4806 * value at the beginning of the next interrupt so the timing
4807 * ends up being correct.
4808 */
4809 q_vector->itr_val = new_itr;
4810 q_vector->set_itr = 1;
4811 }
4812}
4813
4814static void igb_tx_ctxtdesc(struct igb_ring *tx_ring, u32 vlan_macip_lens,
4815 u32 type_tucmd, u32 mss_l4len_idx)
4816{
4817 struct e1000_adv_tx_context_desc *context_desc;
4818 u16 i = tx_ring->next_to_use;
4819
4820 context_desc = IGB_TX_CTXTDESC(tx_ring, i);
4821
4822 i++;
4823 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
4824
4825 /* set bits to identify this as an advanced context descriptor */
4826 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
4827
4828 /* For 82575, context index must be unique per ring. */
4829 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4830 mss_l4len_idx |= tx_ring->reg_idx << 4;
4831
4832 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4833 context_desc->seqnum_seed = 0;
4834 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
4835 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
4836}
4837
4838static int igb_tso(struct igb_ring *tx_ring,
4839 struct igb_tx_buffer *first,
4840 u8 *hdr_len)
4841{
4842 struct sk_buff *skb = first->skb;
4843 u32 vlan_macip_lens, type_tucmd;
4844 u32 mss_l4len_idx, l4len;
4845 int err;
4846
4847 if (skb->ip_summed != CHECKSUM_PARTIAL)
4848 return 0;
4849
4850 if (!skb_is_gso(skb))
4851 return 0;
4852
4853 err = skb_cow_head(skb, 0);
4854 if (err < 0)
4855 return err;
4856
4857 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4858 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
4859
4860 if (first->protocol == htons(ETH_P_IP)) {
4861 struct iphdr *iph = ip_hdr(skb);
4862 iph->tot_len = 0;
4863 iph->check = 0;
4864 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
4865 iph->daddr, 0,
4866 IPPROTO_TCP,
4867 0);
4868 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4869 first->tx_flags |= IGB_TX_FLAGS_TSO |
4870 IGB_TX_FLAGS_CSUM |
4871 IGB_TX_FLAGS_IPV4;
4872 } else if (skb_is_gso_v6(skb)) {
4873 ipv6_hdr(skb)->payload_len = 0;
4874 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4875 &ipv6_hdr(skb)->daddr,
4876 0, IPPROTO_TCP, 0);
4877 first->tx_flags |= IGB_TX_FLAGS_TSO |
4878 IGB_TX_FLAGS_CSUM;
4879 }
4880
4881 /* compute header lengths */
4882 l4len = tcp_hdrlen(skb);
4883 *hdr_len = skb_transport_offset(skb) + l4len;
4884
4885 /* update gso size and bytecount with header size */
4886 first->gso_segs = skb_shinfo(skb)->gso_segs;
4887 first->bytecount += (first->gso_segs - 1) * *hdr_len;
4888
4889 /* MSS L4LEN IDX */
4890 mss_l4len_idx = l4len << E1000_ADVTXD_L4LEN_SHIFT;
4891 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
4892
4893 /* VLAN MACLEN IPLEN */
4894 vlan_macip_lens = skb_network_header_len(skb);
4895 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4896 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4897
4898 igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4899
4900 return 1;
4901}
4902
4903static inline bool igb_ipv6_csum_is_sctp(struct sk_buff *skb)
4904{
4905 unsigned int offset = 0;
4906
4907 ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
4908
4909 return offset == skb_checksum_start_offset(skb);
4910}
4911
4912static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
4913{
4914 struct sk_buff *skb = first->skb;
4915 u32 vlan_macip_lens = 0;
4916 u32 type_tucmd = 0;
4917
4918 if (skb->ip_summed != CHECKSUM_PARTIAL) {
4919csum_failed:
4920 if (!(first->tx_flags & IGB_TX_FLAGS_VLAN))
4921 return;
4922 goto no_csum;
4923 }
4924
4925 switch (skb->csum_offset) {
4926 case offsetof(struct tcphdr, check):
4927 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
4928 /* fall through */
4929 case offsetof(struct udphdr, check):
4930 break;
4931 case offsetof(struct sctphdr, checksum):
4932 /* validate that this is actually an SCTP request */
4933 if (((first->protocol == htons(ETH_P_IP)) &&
4934 (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
4935 ((first->protocol == htons(ETH_P_IPV6)) &&
4936 igb_ipv6_csum_is_sctp(skb))) {
4937 type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP;
4938 break;
4939 }
4940 default:
4941 skb_checksum_help(skb);
4942 goto csum_failed;
4943 }
4944
4945 /* update TX checksum flag */
4946 first->tx_flags |= IGB_TX_FLAGS_CSUM;
4947 vlan_macip_lens = skb_checksum_start_offset(skb) -
4948 skb_network_offset(skb);
4949no_csum:
4950 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4951 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4952
4953 igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0);
4954}
4955
4956#define IGB_SET_FLAG(_input, _flag, _result) \
4957 ((_flag <= _result) ? \
4958 ((u32)(_input & _flag) * (_result / _flag)) : \
4959 ((u32)(_input & _flag) / (_flag / _result)))
4960
4961static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
4962{
4963 /* set type for advanced descriptor with frame checksum insertion */
4964 u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
4965 E1000_ADVTXD_DCMD_DEXT |
4966 E1000_ADVTXD_DCMD_IFCS;
4967
4968 /* set HW vlan bit if vlan is present */
4969 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
4970 (E1000_ADVTXD_DCMD_VLE));
4971
4972 /* set segmentation bits for TSO */
4973 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
4974 (E1000_ADVTXD_DCMD_TSE));
4975
4976 /* set timestamp bit if present */
4977 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
4978 (E1000_ADVTXD_MAC_TSTAMP));
4979
4980 /* insert frame checksum */
4981 cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
4982
4983 return cmd_type;
4984}
4985
4986static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
4987 union e1000_adv_tx_desc *tx_desc,
4988 u32 tx_flags, unsigned int paylen)
4989{
4990 u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
4991
4992 /* 82575 requires a unique index per ring */
4993 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4994 olinfo_status |= tx_ring->reg_idx << 4;
4995
4996 /* insert L4 checksum */
4997 olinfo_status |= IGB_SET_FLAG(tx_flags,
4998 IGB_TX_FLAGS_CSUM,
4999 (E1000_TXD_POPTS_TXSM << 8));
5000
5001 /* insert IPv4 checksum */
5002 olinfo_status |= IGB_SET_FLAG(tx_flags,
5003 IGB_TX_FLAGS_IPV4,
5004 (E1000_TXD_POPTS_IXSM << 8));
5005
5006 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
5007}
5008
5009static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
5010{
5011 struct net_device *netdev = tx_ring->netdev;
5012
5013 netif_stop_subqueue(netdev, tx_ring->queue_index);
5014
5015 /* Herbert's original patch had:
5016 * smp_mb__after_netif_stop_queue();
5017 * but since that doesn't exist yet, just open code it.
5018 */
5019 smp_mb();
5020
5021 /* We need to check again in a case another CPU has just
5022 * made room available.
5023 */
5024 if (igb_desc_unused(tx_ring) < size)
5025 return -EBUSY;
5026
5027 /* A reprieve! */
5028 netif_wake_subqueue(netdev, tx_ring->queue_index);
5029
5030 u64_stats_update_begin(&tx_ring->tx_syncp2);
5031 tx_ring->tx_stats.restart_queue2++;
5032 u64_stats_update_end(&tx_ring->tx_syncp2);
5033
5034 return 0;
5035}
5036
5037static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
5038{
5039 if (igb_desc_unused(tx_ring) >= size)
5040 return 0;
5041 return __igb_maybe_stop_tx(tx_ring, size);
5042}
5043
5044static void igb_tx_map(struct igb_ring *tx_ring,
5045 struct igb_tx_buffer *first,
5046 const u8 hdr_len)
5047{
5048 struct sk_buff *skb = first->skb;
5049 struct igb_tx_buffer *tx_buffer;
5050 union e1000_adv_tx_desc *tx_desc;
5051 struct skb_frag_struct *frag;
5052 dma_addr_t dma;
5053 unsigned int data_len, size;
5054 u32 tx_flags = first->tx_flags;
5055 u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
5056 u16 i = tx_ring->next_to_use;
5057
5058 tx_desc = IGB_TX_DESC(tx_ring, i);
5059
5060 igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
5061
5062 size = skb_headlen(skb);
5063 data_len = skb->data_len;
5064
5065 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
5066
5067 tx_buffer = first;
5068
5069 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
5070 if (dma_mapping_error(tx_ring->dev, dma))
5071 goto dma_error;
5072
5073 /* record length, and DMA address */
5074 dma_unmap_len_set(tx_buffer, len, size);
5075 dma_unmap_addr_set(tx_buffer, dma, dma);
5076
5077 tx_desc->read.buffer_addr = cpu_to_le64(dma);
5078
5079 while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
5080 tx_desc->read.cmd_type_len =
5081 cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
5082
5083 i++;
5084 tx_desc++;
5085 if (i == tx_ring->count) {
5086 tx_desc = IGB_TX_DESC(tx_ring, 0);
5087 i = 0;
5088 }
5089 tx_desc->read.olinfo_status = 0;
5090
5091 dma += IGB_MAX_DATA_PER_TXD;
5092 size -= IGB_MAX_DATA_PER_TXD;
5093
5094 tx_desc->read.buffer_addr = cpu_to_le64(dma);
5095 }
5096
5097 if (likely(!data_len))
5098 break;
5099
5100 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
5101
5102 i++;
5103 tx_desc++;
5104 if (i == tx_ring->count) {
5105 tx_desc = IGB_TX_DESC(tx_ring, 0);
5106 i = 0;
5107 }
5108 tx_desc->read.olinfo_status = 0;
5109
5110 size = skb_frag_size(frag);
5111 data_len -= size;
5112
5113 dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
5114 size, DMA_TO_DEVICE);
5115
5116 tx_buffer = &tx_ring->tx_buffer_info[i];
5117 }
5118
5119 /* write last descriptor with RS and EOP bits */
5120 cmd_type |= size | IGB_TXD_DCMD;
5121 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
5122
5123 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
5124
5125 /* set the timestamp */
5126 first->time_stamp = jiffies;
5127
5128 /* Force memory writes to complete before letting h/w know there
5129 * are new descriptors to fetch. (Only applicable for weak-ordered
5130 * memory model archs, such as IA-64).
5131 *
5132 * We also need this memory barrier to make certain all of the
5133 * status bits have been updated before next_to_watch is written.
5134 */
5135 wmb();
5136
5137 /* set next_to_watch value indicating a packet is present */
5138 first->next_to_watch = tx_desc;
5139
5140 i++;
5141 if (i == tx_ring->count)
5142 i = 0;
5143
5144 tx_ring->next_to_use = i;
5145
5146 /* Make sure there is space in the ring for the next send. */
5147 igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
5148
5149 if (netif_xmit_stopped(txring_txq(tx_ring)) || !skb->xmit_more) {
5150 writel(i, tx_ring->tail);
5151
5152 /* we need this if more than one processor can write to our tail
5153 * at a time, it synchronizes IO on IA64/Altix systems
5154 */
5155 mmiowb();
5156 }
5157 return;
5158
5159dma_error:
5160 dev_err(tx_ring->dev, "TX DMA map failed\n");
5161
5162 /* clear dma mappings for failed tx_buffer_info map */
5163 for (;;) {
5164 tx_buffer = &tx_ring->tx_buffer_info[i];
5165 igb_unmap_and_free_tx_resource(tx_ring, tx_buffer);
5166 if (tx_buffer == first)
5167 break;
5168 if (i == 0)
5169 i = tx_ring->count;
5170 i--;
5171 }
5172
5173 tx_ring->next_to_use = i;
5174}
5175
5176netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
5177 struct igb_ring *tx_ring)
5178{
5179 struct igb_tx_buffer *first;
5180 int tso;
5181 u32 tx_flags = 0;
5182 unsigned short f;
5183 u16 count = TXD_USE_COUNT(skb_headlen(skb));
5184 __be16 protocol = vlan_get_protocol(skb);
5185 u8 hdr_len = 0;
5186
5187 /* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
5188 * + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
5189 * + 2 desc gap to keep tail from touching head,
5190 * + 1 desc for context descriptor,
5191 * otherwise try next time
5192 */
5193 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
5194 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
5195
5196 if (igb_maybe_stop_tx(tx_ring, count + 3)) {
5197 /* this is a hard error */
5198 return NETDEV_TX_BUSY;
5199 }
5200
5201 /* record the location of the first descriptor for this packet */
5202 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
5203 first->skb = skb;
5204 first->bytecount = skb->len;
5205 first->gso_segs = 1;
5206
5207 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
5208 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
5209
5210 if (!test_and_set_bit_lock(__IGB_PTP_TX_IN_PROGRESS,
5211 &adapter->state)) {
5212 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
5213 tx_flags |= IGB_TX_FLAGS_TSTAMP;
5214
5215 adapter->ptp_tx_skb = skb_get(skb);
5216 adapter->ptp_tx_start = jiffies;
5217 if (adapter->hw.mac.type == e1000_82576)
5218 schedule_work(&adapter->ptp_tx_work);
5219 }
5220 }
5221
5222 skb_tx_timestamp(skb);
5223
5224 if (skb_vlan_tag_present(skb)) {
5225 tx_flags |= IGB_TX_FLAGS_VLAN;
5226 tx_flags |= (skb_vlan_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
5227 }
5228
5229 /* record initial flags and protocol */
5230 first->tx_flags = tx_flags;
5231 first->protocol = protocol;
5232
5233 tso = igb_tso(tx_ring, first, &hdr_len);
5234 if (tso < 0)
5235 goto out_drop;
5236 else if (!tso)
5237 igb_tx_csum(tx_ring, first);
5238
5239 igb_tx_map(tx_ring, first, hdr_len);
5240
5241 return NETDEV_TX_OK;
5242
5243out_drop:
5244 igb_unmap_and_free_tx_resource(tx_ring, first);
5245
5246 return NETDEV_TX_OK;
5247}
5248
5249static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
5250 struct sk_buff *skb)
5251{
5252 unsigned int r_idx = skb->queue_mapping;
5253
5254 if (r_idx >= adapter->num_tx_queues)
5255 r_idx = r_idx % adapter->num_tx_queues;
5256
5257 return adapter->tx_ring[r_idx];
5258}
5259
5260static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
5261 struct net_device *netdev)
5262{
5263 struct igb_adapter *adapter = netdev_priv(netdev);
5264
5265 /* The minimum packet size with TCTL.PSP set is 17 so pad the skb
5266 * in order to meet this minimum size requirement.
5267 */
5268 if (skb_put_padto(skb, 17))
5269 return NETDEV_TX_OK;
5270
5271 return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
5272}
5273
5274/**
5275 * igb_tx_timeout - Respond to a Tx Hang
5276 * @netdev: network interface device structure
5277 **/
5278static void igb_tx_timeout(struct net_device *netdev)
5279{
5280 struct igb_adapter *adapter = netdev_priv(netdev);
5281 struct e1000_hw *hw = &adapter->hw;
5282
5283 /* Do the reset outside of interrupt context */
5284 adapter->tx_timeout_count++;
5285
5286 if (hw->mac.type >= e1000_82580)
5287 hw->dev_spec._82575.global_device_reset = true;
5288
5289 schedule_work(&adapter->reset_task);
5290 wr32(E1000_EICS,
5291 (adapter->eims_enable_mask & ~adapter->eims_other));
5292}
5293
5294static void igb_reset_task(struct work_struct *work)
5295{
5296 struct igb_adapter *adapter;
5297 adapter = container_of(work, struct igb_adapter, reset_task);
5298
5299 igb_dump(adapter);
5300 netdev_err(adapter->netdev, "Reset adapter\n");
5301 igb_reinit_locked(adapter);
5302}
5303
5304/**
5305 * igb_get_stats64 - Get System Network Statistics
5306 * @netdev: network interface device structure
5307 * @stats: rtnl_link_stats64 pointer
5308 **/
5309static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *netdev,
5310 struct rtnl_link_stats64 *stats)
5311{
5312 struct igb_adapter *adapter = netdev_priv(netdev);
5313
5314 spin_lock(&adapter->stats64_lock);
5315 igb_update_stats(adapter, &adapter->stats64);
5316 memcpy(stats, &adapter->stats64, sizeof(*stats));
5317 spin_unlock(&adapter->stats64_lock);
5318
5319 return stats;
5320}
5321
5322/**
5323 * igb_change_mtu - Change the Maximum Transfer Unit
5324 * @netdev: network interface device structure
5325 * @new_mtu: new value for maximum frame size
5326 *
5327 * Returns 0 on success, negative on failure
5328 **/
5329static int igb_change_mtu(struct net_device *netdev, int new_mtu)
5330{
5331 struct igb_adapter *adapter = netdev_priv(netdev);
5332 struct pci_dev *pdev = adapter->pdev;
5333 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
5334
5335 if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
5336 dev_err(&pdev->dev, "Invalid MTU setting\n");
5337 return -EINVAL;
5338 }
5339
5340#define MAX_STD_JUMBO_FRAME_SIZE 9238
5341 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
5342 dev_err(&pdev->dev, "MTU > 9216 not supported.\n");
5343 return -EINVAL;
5344 }
5345
5346 /* adjust max frame to be at least the size of a standard frame */
5347 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
5348 max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
5349
5350 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
5351 usleep_range(1000, 2000);
5352
5353 /* igb_down has a dependency on max_frame_size */
5354 adapter->max_frame_size = max_frame;
5355
5356 if (netif_running(netdev))
5357 igb_down(adapter);
5358
5359 dev_info(&pdev->dev, "changing MTU from %d to %d\n",
5360 netdev->mtu, new_mtu);
5361 netdev->mtu = new_mtu;
5362
5363 if (netif_running(netdev))
5364 igb_up(adapter);
5365 else
5366 igb_reset(adapter);
5367
5368 clear_bit(__IGB_RESETTING, &adapter->state);
5369
5370 return 0;
5371}
5372
5373/**
5374 * igb_update_stats - Update the board statistics counters
5375 * @adapter: board private structure
5376 **/
5377void igb_update_stats(struct igb_adapter *adapter,
5378 struct rtnl_link_stats64 *net_stats)
5379{
5380 struct e1000_hw *hw = &adapter->hw;
5381 struct pci_dev *pdev = adapter->pdev;
5382 u32 reg, mpc;
5383 int i;
5384 u64 bytes, packets;
5385 unsigned int start;
5386 u64 _bytes, _packets;
5387
5388 /* Prevent stats update while adapter is being reset, or if the pci
5389 * connection is down.
5390 */
5391 if (adapter->link_speed == 0)
5392 return;
5393 if (pci_channel_offline(pdev))
5394 return;
5395
5396 bytes = 0;
5397 packets = 0;
5398
5399 rcu_read_lock();
5400 for (i = 0; i < adapter->num_rx_queues; i++) {
5401 struct igb_ring *ring = adapter->rx_ring[i];
5402 u32 rqdpc = rd32(E1000_RQDPC(i));
5403 if (hw->mac.type >= e1000_i210)
5404 wr32(E1000_RQDPC(i), 0);
5405
5406 if (rqdpc) {
5407 ring->rx_stats.drops += rqdpc;
5408 net_stats->rx_fifo_errors += rqdpc;
5409 }
5410
5411 do {
5412 start = u64_stats_fetch_begin_irq(&ring->rx_syncp);
5413 _bytes = ring->rx_stats.bytes;
5414 _packets = ring->rx_stats.packets;
5415 } while (u64_stats_fetch_retry_irq(&ring->rx_syncp, start));
5416 bytes += _bytes;
5417 packets += _packets;
5418 }
5419
5420 net_stats->rx_bytes = bytes;
5421 net_stats->rx_packets = packets;
5422
5423 bytes = 0;
5424 packets = 0;
5425 for (i = 0; i < adapter->num_tx_queues; i++) {
5426 struct igb_ring *ring = adapter->tx_ring[i];
5427 do {
5428 start = u64_stats_fetch_begin_irq(&ring->tx_syncp);
5429 _bytes = ring->tx_stats.bytes;
5430 _packets = ring->tx_stats.packets;
5431 } while (u64_stats_fetch_retry_irq(&ring->tx_syncp, start));
5432 bytes += _bytes;
5433 packets += _packets;
5434 }
5435 net_stats->tx_bytes = bytes;
5436 net_stats->tx_packets = packets;
5437 rcu_read_unlock();
5438
5439 /* read stats registers */
5440 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
5441 adapter->stats.gprc += rd32(E1000_GPRC);
5442 adapter->stats.gorc += rd32(E1000_GORCL);
5443 rd32(E1000_GORCH); /* clear GORCL */
5444 adapter->stats.bprc += rd32(E1000_BPRC);
5445 adapter->stats.mprc += rd32(E1000_MPRC);
5446 adapter->stats.roc += rd32(E1000_ROC);
5447
5448 adapter->stats.prc64 += rd32(E1000_PRC64);
5449 adapter->stats.prc127 += rd32(E1000_PRC127);
5450 adapter->stats.prc255 += rd32(E1000_PRC255);
5451 adapter->stats.prc511 += rd32(E1000_PRC511);
5452 adapter->stats.prc1023 += rd32(E1000_PRC1023);
5453 adapter->stats.prc1522 += rd32(E1000_PRC1522);
5454 adapter->stats.symerrs += rd32(E1000_SYMERRS);
5455 adapter->stats.sec += rd32(E1000_SEC);
5456
5457 mpc = rd32(E1000_MPC);
5458 adapter->stats.mpc += mpc;
5459 net_stats->rx_fifo_errors += mpc;
5460 adapter->stats.scc += rd32(E1000_SCC);
5461 adapter->stats.ecol += rd32(E1000_ECOL);
5462 adapter->stats.mcc += rd32(E1000_MCC);
5463 adapter->stats.latecol += rd32(E1000_LATECOL);
5464 adapter->stats.dc += rd32(E1000_DC);
5465 adapter->stats.rlec += rd32(E1000_RLEC);
5466 adapter->stats.xonrxc += rd32(E1000_XONRXC);
5467 adapter->stats.xontxc += rd32(E1000_XONTXC);
5468 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
5469 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
5470 adapter->stats.fcruc += rd32(E1000_FCRUC);
5471 adapter->stats.gptc += rd32(E1000_GPTC);
5472 adapter->stats.gotc += rd32(E1000_GOTCL);
5473 rd32(E1000_GOTCH); /* clear GOTCL */
5474 adapter->stats.rnbc += rd32(E1000_RNBC);
5475 adapter->stats.ruc += rd32(E1000_RUC);
5476 adapter->stats.rfc += rd32(E1000_RFC);
5477 adapter->stats.rjc += rd32(E1000_RJC);
5478 adapter->stats.tor += rd32(E1000_TORH);
5479 adapter->stats.tot += rd32(E1000_TOTH);
5480 adapter->stats.tpr += rd32(E1000_TPR);
5481
5482 adapter->stats.ptc64 += rd32(E1000_PTC64);
5483 adapter->stats.ptc127 += rd32(E1000_PTC127);
5484 adapter->stats.ptc255 += rd32(E1000_PTC255);
5485 adapter->stats.ptc511 += rd32(E1000_PTC511);
5486 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
5487 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
5488
5489 adapter->stats.mptc += rd32(E1000_MPTC);
5490 adapter->stats.bptc += rd32(E1000_BPTC);
5491
5492 adapter->stats.tpt += rd32(E1000_TPT);
5493 adapter->stats.colc += rd32(E1000_COLC);
5494
5495 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
5496 /* read internal phy specific stats */
5497 reg = rd32(E1000_CTRL_EXT);
5498 if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
5499 adapter->stats.rxerrc += rd32(E1000_RXERRC);
5500
5501 /* this stat has invalid values on i210/i211 */
5502 if ((hw->mac.type != e1000_i210) &&
5503 (hw->mac.type != e1000_i211))
5504 adapter->stats.tncrs += rd32(E1000_TNCRS);
5505 }
5506
5507 adapter->stats.tsctc += rd32(E1000_TSCTC);
5508 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
5509
5510 adapter->stats.iac += rd32(E1000_IAC);
5511 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
5512 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
5513 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
5514 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
5515 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
5516 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
5517 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
5518 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
5519
5520 /* Fill out the OS statistics structure */
5521 net_stats->multicast = adapter->stats.mprc;
5522 net_stats->collisions = adapter->stats.colc;
5523
5524 /* Rx Errors */
5525
5526 /* RLEC on some newer hardware can be incorrect so build
5527 * our own version based on RUC and ROC
5528 */
5529 net_stats->rx_errors = adapter->stats.rxerrc +
5530 adapter->stats.crcerrs + adapter->stats.algnerrc +
5531 adapter->stats.ruc + adapter->stats.roc +
5532 adapter->stats.cexterr;
5533 net_stats->rx_length_errors = adapter->stats.ruc +
5534 adapter->stats.roc;
5535 net_stats->rx_crc_errors = adapter->stats.crcerrs;
5536 net_stats->rx_frame_errors = adapter->stats.algnerrc;
5537 net_stats->rx_missed_errors = adapter->stats.mpc;
5538
5539 /* Tx Errors */
5540 net_stats->tx_errors = adapter->stats.ecol +
5541 adapter->stats.latecol;
5542 net_stats->tx_aborted_errors = adapter->stats.ecol;
5543 net_stats->tx_window_errors = adapter->stats.latecol;
5544 net_stats->tx_carrier_errors = adapter->stats.tncrs;
5545
5546 /* Tx Dropped needs to be maintained elsewhere */
5547
5548 /* Management Stats */
5549 adapter->stats.mgptc += rd32(E1000_MGTPTC);
5550 adapter->stats.mgprc += rd32(E1000_MGTPRC);
5551 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
5552
5553 /* OS2BMC Stats */
5554 reg = rd32(E1000_MANC);
5555 if (reg & E1000_MANC_EN_BMC2OS) {
5556 adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
5557 adapter->stats.o2bspc += rd32(E1000_O2BSPC);
5558 adapter->stats.b2ospc += rd32(E1000_B2OSPC);
5559 adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
5560 }
5561}
5562
5563static void igb_tsync_interrupt(struct igb_adapter *adapter)
5564{
5565 struct e1000_hw *hw = &adapter->hw;
5566 struct ptp_clock_event event;
5567 struct timespec64 ts;
5568 u32 ack = 0, tsauxc, sec, nsec, tsicr = rd32(E1000_TSICR);
5569
5570 if (tsicr & TSINTR_SYS_WRAP) {
5571 event.type = PTP_CLOCK_PPS;
5572 if (adapter->ptp_caps.pps)
5573 ptp_clock_event(adapter->ptp_clock, &event);
5574 else
5575 dev_err(&adapter->pdev->dev, "unexpected SYS WRAP");
5576 ack |= TSINTR_SYS_WRAP;
5577 }
5578
5579 if (tsicr & E1000_TSICR_TXTS) {
5580 /* retrieve hardware timestamp */
5581 schedule_work(&adapter->ptp_tx_work);
5582 ack |= E1000_TSICR_TXTS;
5583 }
5584
5585 if (tsicr & TSINTR_TT0) {
5586 spin_lock(&adapter->tmreg_lock);
5587 ts = timespec64_add(adapter->perout[0].start,
5588 adapter->perout[0].period);
5589 /* u32 conversion of tv_sec is safe until y2106 */
5590 wr32(E1000_TRGTTIML0, ts.tv_nsec);
5591 wr32(E1000_TRGTTIMH0, (u32)ts.tv_sec);
5592 tsauxc = rd32(E1000_TSAUXC);
5593 tsauxc |= TSAUXC_EN_TT0;
5594 wr32(E1000_TSAUXC, tsauxc);
5595 adapter->perout[0].start = ts;
5596 spin_unlock(&adapter->tmreg_lock);
5597 ack |= TSINTR_TT0;
5598 }
5599
5600 if (tsicr & TSINTR_TT1) {
5601 spin_lock(&adapter->tmreg_lock);
5602 ts = timespec64_add(adapter->perout[1].start,
5603 adapter->perout[1].period);
5604 wr32(E1000_TRGTTIML1, ts.tv_nsec);
5605 wr32(E1000_TRGTTIMH1, (u32)ts.tv_sec);
5606 tsauxc = rd32(E1000_TSAUXC);
5607 tsauxc |= TSAUXC_EN_TT1;
5608 wr32(E1000_TSAUXC, tsauxc);
5609 adapter->perout[1].start = ts;
5610 spin_unlock(&adapter->tmreg_lock);
5611 ack |= TSINTR_TT1;
5612 }
5613
5614 if (tsicr & TSINTR_AUTT0) {
5615 nsec = rd32(E1000_AUXSTMPL0);
5616 sec = rd32(E1000_AUXSTMPH0);
5617 event.type = PTP_CLOCK_EXTTS;
5618 event.index = 0;
5619 event.timestamp = sec * 1000000000ULL + nsec;
5620 ptp_clock_event(adapter->ptp_clock, &event);
5621 ack |= TSINTR_AUTT0;
5622 }
5623
5624 if (tsicr & TSINTR_AUTT1) {
5625 nsec = rd32(E1000_AUXSTMPL1);
5626 sec = rd32(E1000_AUXSTMPH1);
5627 event.type = PTP_CLOCK_EXTTS;
5628 event.index = 1;
5629 event.timestamp = sec * 1000000000ULL + nsec;
5630 ptp_clock_event(adapter->ptp_clock, &event);
5631 ack |= TSINTR_AUTT1;
5632 }
5633
5634 /* acknowledge the interrupts */
5635 wr32(E1000_TSICR, ack);
5636}
5637
5638static irqreturn_t igb_msix_other(int irq, void *data)
5639{
5640 struct igb_adapter *adapter = data;
5641 struct e1000_hw *hw = &adapter->hw;
5642 u32 icr = rd32(E1000_ICR);
5643 /* reading ICR causes bit 31 of EICR to be cleared */
5644
5645 if (icr & E1000_ICR_DRSTA)
5646 schedule_work(&adapter->reset_task);
5647
5648 if (icr & E1000_ICR_DOUTSYNC) {
5649 /* HW is reporting DMA is out of sync */
5650 adapter->stats.doosync++;
5651 /* The DMA Out of Sync is also indication of a spoof event
5652 * in IOV mode. Check the Wrong VM Behavior register to
5653 * see if it is really a spoof event.
5654 */
5655 igb_check_wvbr(adapter);
5656 }
5657
5658 /* Check for a mailbox event */
5659 if (icr & E1000_ICR_VMMB)
5660 igb_msg_task(adapter);
5661
5662 if (icr & E1000_ICR_LSC) {
5663 hw->mac.get_link_status = 1;
5664 /* guard against interrupt when we're going down */
5665 if (!test_bit(__IGB_DOWN, &adapter->state))
5666 mod_timer(&adapter->watchdog_timer, jiffies + 1);
5667 }
5668
5669 if (icr & E1000_ICR_TS)
5670 igb_tsync_interrupt(adapter);
5671
5672 wr32(E1000_EIMS, adapter->eims_other);
5673
5674 return IRQ_HANDLED;
5675}
5676
5677static void igb_write_itr(struct igb_q_vector *q_vector)
5678{
5679 struct igb_adapter *adapter = q_vector->adapter;
5680 u32 itr_val = q_vector->itr_val & 0x7FFC;
5681
5682 if (!q_vector->set_itr)
5683 return;
5684
5685 if (!itr_val)
5686 itr_val = 0x4;
5687
5688 if (adapter->hw.mac.type == e1000_82575)
5689 itr_val |= itr_val << 16;
5690 else
5691 itr_val |= E1000_EITR_CNT_IGNR;
5692
5693 writel(itr_val, q_vector->itr_register);
5694 q_vector->set_itr = 0;
5695}
5696
5697static irqreturn_t igb_msix_ring(int irq, void *data)
5698{
5699 struct igb_q_vector *q_vector = data;
5700
5701 /* Write the ITR value calculated from the previous interrupt. */
5702 igb_write_itr(q_vector);
5703
5704 napi_schedule(&q_vector->napi);
5705
5706 return IRQ_HANDLED;
5707}
5708
5709#ifdef CONFIG_IGB_DCA
5710static void igb_update_tx_dca(struct igb_adapter *adapter,
5711 struct igb_ring *tx_ring,
5712 int cpu)
5713{
5714 struct e1000_hw *hw = &adapter->hw;
5715 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
5716
5717 if (hw->mac.type != e1000_82575)
5718 txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
5719
5720 /* We can enable relaxed ordering for reads, but not writes when
5721 * DCA is enabled. This is due to a known issue in some chipsets
5722 * which will cause the DCA tag to be cleared.
5723 */
5724 txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
5725 E1000_DCA_TXCTRL_DATA_RRO_EN |
5726 E1000_DCA_TXCTRL_DESC_DCA_EN;
5727
5728 wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
5729}
5730
5731static void igb_update_rx_dca(struct igb_adapter *adapter,
5732 struct igb_ring *rx_ring,
5733 int cpu)
5734{
5735 struct e1000_hw *hw = &adapter->hw;
5736 u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
5737
5738 if (hw->mac.type != e1000_82575)
5739 rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
5740
5741 /* We can enable relaxed ordering for reads, but not writes when
5742 * DCA is enabled. This is due to a known issue in some chipsets
5743 * which will cause the DCA tag to be cleared.
5744 */
5745 rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
5746 E1000_DCA_RXCTRL_DESC_DCA_EN;
5747
5748 wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
5749}
5750
5751static void igb_update_dca(struct igb_q_vector *q_vector)
5752{
5753 struct igb_adapter *adapter = q_vector->adapter;
5754 int cpu = get_cpu();
5755
5756 if (q_vector->cpu == cpu)
5757 goto out_no_update;
5758
5759 if (q_vector->tx.ring)
5760 igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
5761
5762 if (q_vector->rx.ring)
5763 igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
5764
5765 q_vector->cpu = cpu;
5766out_no_update:
5767 put_cpu();
5768}
5769
5770static void igb_setup_dca(struct igb_adapter *adapter)
5771{
5772 struct e1000_hw *hw = &adapter->hw;
5773 int i;
5774
5775 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
5776 return;
5777
5778 /* Always use CB2 mode, difference is masked in the CB driver. */
5779 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
5780
5781 for (i = 0; i < adapter->num_q_vectors; i++) {
5782 adapter->q_vector[i]->cpu = -1;
5783 igb_update_dca(adapter->q_vector[i]);
5784 }
5785}
5786
5787static int __igb_notify_dca(struct device *dev, void *data)
5788{
5789 struct net_device *netdev = dev_get_drvdata(dev);
5790 struct igb_adapter *adapter = netdev_priv(netdev);
5791 struct pci_dev *pdev = adapter->pdev;
5792 struct e1000_hw *hw = &adapter->hw;
5793 unsigned long event = *(unsigned long *)data;
5794
5795 switch (event) {
5796 case DCA_PROVIDER_ADD:
5797 /* if already enabled, don't do it again */
5798 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
5799 break;
5800 if (dca_add_requester(dev) == 0) {
5801 adapter->flags |= IGB_FLAG_DCA_ENABLED;
5802 dev_info(&pdev->dev, "DCA enabled\n");
5803 igb_setup_dca(adapter);
5804 break;
5805 }
5806 /* Fall Through since DCA is disabled. */
5807 case DCA_PROVIDER_REMOVE:
5808 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
5809 /* without this a class_device is left
5810 * hanging around in the sysfs model
5811 */
5812 dca_remove_requester(dev);
5813 dev_info(&pdev->dev, "DCA disabled\n");
5814 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
5815 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
5816 }
5817 break;
5818 }
5819
5820 return 0;
5821}
5822
5823static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
5824 void *p)
5825{
5826 int ret_val;
5827
5828 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
5829 __igb_notify_dca);
5830
5831 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
5832}
5833#endif /* CONFIG_IGB_DCA */
5834
5835#ifdef CONFIG_PCI_IOV
5836static int igb_vf_configure(struct igb_adapter *adapter, int vf)
5837{
5838 unsigned char mac_addr[ETH_ALEN];
5839
5840 eth_zero_addr(mac_addr);
5841 igb_set_vf_mac(adapter, vf, mac_addr);
5842
5843 /* By default spoof check is enabled for all VFs */
5844 adapter->vf_data[vf].spoofchk_enabled = true;
5845
5846 return 0;
5847}
5848
5849#endif
5850static void igb_ping_all_vfs(struct igb_adapter *adapter)
5851{
5852 struct e1000_hw *hw = &adapter->hw;
5853 u32 ping;
5854 int i;
5855
5856 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
5857 ping = E1000_PF_CONTROL_MSG;
5858 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
5859 ping |= E1000_VT_MSGTYPE_CTS;
5860 igb_write_mbx(hw, &ping, 1, i);
5861 }
5862}
5863
5864static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5865{
5866 struct e1000_hw *hw = &adapter->hw;
5867 u32 vmolr = rd32(E1000_VMOLR(vf));
5868 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5869
5870 vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
5871 IGB_VF_FLAG_MULTI_PROMISC);
5872 vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5873
5874 if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
5875 vmolr |= E1000_VMOLR_MPME;
5876 vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
5877 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
5878 } else {
5879 /* if we have hashes and we are clearing a multicast promisc
5880 * flag we need to write the hashes to the MTA as this step
5881 * was previously skipped
5882 */
5883 if (vf_data->num_vf_mc_hashes > 30) {
5884 vmolr |= E1000_VMOLR_MPME;
5885 } else if (vf_data->num_vf_mc_hashes) {
5886 int j;
5887
5888 vmolr |= E1000_VMOLR_ROMPE;
5889 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5890 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5891 }
5892 }
5893
5894 wr32(E1000_VMOLR(vf), vmolr);
5895
5896 /* there are flags left unprocessed, likely not supported */
5897 if (*msgbuf & E1000_VT_MSGINFO_MASK)
5898 return -EINVAL;
5899
5900 return 0;
5901}
5902
5903static int igb_set_vf_multicasts(struct igb_adapter *adapter,
5904 u32 *msgbuf, u32 vf)
5905{
5906 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5907 u16 *hash_list = (u16 *)&msgbuf[1];
5908 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5909 int i;
5910
5911 /* salt away the number of multicast addresses assigned
5912 * to this VF for later use to restore when the PF multi cast
5913 * list changes
5914 */
5915 vf_data->num_vf_mc_hashes = n;
5916
5917 /* only up to 30 hash values supported */
5918 if (n > 30)
5919 n = 30;
5920
5921 /* store the hashes for later use */
5922 for (i = 0; i < n; i++)
5923 vf_data->vf_mc_hashes[i] = hash_list[i];
5924
5925 /* Flush and reset the mta with the new values */
5926 igb_set_rx_mode(adapter->netdev);
5927
5928 return 0;
5929}
5930
5931static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
5932{
5933 struct e1000_hw *hw = &adapter->hw;
5934 struct vf_data_storage *vf_data;
5935 int i, j;
5936
5937 for (i = 0; i < adapter->vfs_allocated_count; i++) {
5938 u32 vmolr = rd32(E1000_VMOLR(i));
5939
5940 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5941
5942 vf_data = &adapter->vf_data[i];
5943
5944 if ((vf_data->num_vf_mc_hashes > 30) ||
5945 (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
5946 vmolr |= E1000_VMOLR_MPME;
5947 } else if (vf_data->num_vf_mc_hashes) {
5948 vmolr |= E1000_VMOLR_ROMPE;
5949 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5950 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5951 }
5952 wr32(E1000_VMOLR(i), vmolr);
5953 }
5954}
5955
5956static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
5957{
5958 struct e1000_hw *hw = &adapter->hw;
5959 u32 pool_mask, vlvf_mask, i;
5960
5961 /* create mask for VF and other pools */
5962 pool_mask = E1000_VLVF_POOLSEL_MASK;
5963 vlvf_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5964
5965 /* drop PF from pool bits */
5966 pool_mask &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT +
5967 adapter->vfs_allocated_count));
5968
5969 /* Find the vlan filter for this id */
5970 for (i = E1000_VLVF_ARRAY_SIZE; i--;) {
5971 u32 vlvf = rd32(E1000_VLVF(i));
5972 u32 vfta_mask, vid, vfta;
5973
5974 /* remove the vf from the pool */
5975 if (!(vlvf & vlvf_mask))
5976 continue;
5977
5978 /* clear out bit from VLVF */
5979 vlvf ^= vlvf_mask;
5980
5981 /* if other pools are present, just remove ourselves */
5982 if (vlvf & pool_mask)
5983 goto update_vlvfb;
5984
5985 /* if PF is present, leave VFTA */
5986 if (vlvf & E1000_VLVF_POOLSEL_MASK)
5987 goto update_vlvf;
5988
5989 vid = vlvf & E1000_VLVF_VLANID_MASK;
5990 vfta_mask = 1 << (vid % 32);
5991
5992 /* clear bit from VFTA */
5993 vfta = adapter->shadow_vfta[vid / 32];
5994 if (vfta & vfta_mask)
5995 hw->mac.ops.write_vfta(hw, vid / 32, vfta ^ vfta_mask);
5996update_vlvf:
5997 /* clear pool selection enable */
5998 if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
5999 vlvf &= E1000_VLVF_POOLSEL_MASK;
6000 else
6001 vlvf = 0;
6002update_vlvfb:
6003 /* clear pool bits */
6004 wr32(E1000_VLVF(i), vlvf);
6005 }
6006}
6007
6008static int igb_find_vlvf_entry(struct e1000_hw *hw, u32 vlan)
6009{
6010 u32 vlvf;
6011 int idx;
6012
6013 /* short cut the special case */
6014 if (vlan == 0)
6015 return 0;
6016
6017 /* Search for the VLAN id in the VLVF entries */
6018 for (idx = E1000_VLVF_ARRAY_SIZE; --idx;) {
6019 vlvf = rd32(E1000_VLVF(idx));
6020 if ((vlvf & VLAN_VID_MASK) == vlan)
6021 break;
6022 }
6023
6024 return idx;
6025}
6026
6027void igb_update_pf_vlvf(struct igb_adapter *adapter, u32 vid)
6028{
6029 struct e1000_hw *hw = &adapter->hw;
6030 u32 bits, pf_id;
6031 int idx;
6032
6033 idx = igb_find_vlvf_entry(hw, vid);
6034 if (!idx)
6035 return;
6036
6037 /* See if any other pools are set for this VLAN filter
6038 * entry other than the PF.
6039 */
6040 pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
6041 bits = ~(1 << pf_id) & E1000_VLVF_POOLSEL_MASK;
6042 bits &= rd32(E1000_VLVF(idx));
6043
6044 /* Disable the filter so this falls into the default pool. */
6045 if (!bits) {
6046 if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
6047 wr32(E1000_VLVF(idx), 1 << pf_id);
6048 else
6049 wr32(E1000_VLVF(idx), 0);
6050 }
6051}
6052
6053static s32 igb_set_vf_vlan(struct igb_adapter *adapter, u32 vid,
6054 bool add, u32 vf)
6055{
6056 int pf_id = adapter->vfs_allocated_count;
6057 struct e1000_hw *hw = &adapter->hw;
6058 int err;
6059
6060 /* If VLAN overlaps with one the PF is currently monitoring make
6061 * sure that we are able to allocate a VLVF entry. This may be
6062 * redundant but it guarantees PF will maintain visibility to
6063 * the VLAN.
6064 */
6065 if (add && test_bit(vid, adapter->active_vlans)) {
6066 err = igb_vfta_set(hw, vid, pf_id, true, false);
6067 if (err)
6068 return err;
6069 }
6070
6071 err = igb_vfta_set(hw, vid, vf, add, false);
6072
6073 if (add && !err)
6074 return err;
6075
6076 /* If we failed to add the VF VLAN or we are removing the VF VLAN
6077 * we may need to drop the PF pool bit in order to allow us to free
6078 * up the VLVF resources.
6079 */
6080 if (test_bit(vid, adapter->active_vlans) ||
6081 (adapter->flags & IGB_FLAG_VLAN_PROMISC))
6082 igb_update_pf_vlvf(adapter, vid);
6083
6084 return err;
6085}
6086
6087static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
6088{
6089 struct e1000_hw *hw = &adapter->hw;
6090
6091 if (vid)
6092 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
6093 else
6094 wr32(E1000_VMVIR(vf), 0);
6095}
6096
6097static int igb_enable_port_vlan(struct igb_adapter *adapter, int vf,
6098 u16 vlan, u8 qos)
6099{
6100 int err;
6101
6102 err = igb_set_vf_vlan(adapter, vlan, true, vf);
6103 if (err)
6104 return err;
6105
6106 igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
6107 igb_set_vmolr(adapter, vf, !vlan);
6108
6109 /* revoke access to previous VLAN */
6110 if (vlan != adapter->vf_data[vf].pf_vlan)
6111 igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
6112 false, vf);
6113
6114 adapter->vf_data[vf].pf_vlan = vlan;
6115 adapter->vf_data[vf].pf_qos = qos;
6116 igb_set_vf_vlan_strip(adapter, vf, true);
6117 dev_info(&adapter->pdev->dev,
6118 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
6119 if (test_bit(__IGB_DOWN, &adapter->state)) {
6120 dev_warn(&adapter->pdev->dev,
6121 "The VF VLAN has been set, but the PF device is not up.\n");
6122 dev_warn(&adapter->pdev->dev,
6123 "Bring the PF device up before attempting to use the VF device.\n");
6124 }
6125
6126 return err;
6127}
6128
6129static int igb_disable_port_vlan(struct igb_adapter *adapter, int vf)
6130{
6131 /* Restore tagless access via VLAN 0 */
6132 igb_set_vf_vlan(adapter, 0, true, vf);
6133
6134 igb_set_vmvir(adapter, 0, vf);
6135 igb_set_vmolr(adapter, vf, true);
6136
6137 /* Remove any PF assigned VLAN */
6138 if (adapter->vf_data[vf].pf_vlan)
6139 igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
6140 false, vf);
6141
6142 adapter->vf_data[vf].pf_vlan = 0;
6143 adapter->vf_data[vf].pf_qos = 0;
6144 igb_set_vf_vlan_strip(adapter, vf, false);
6145
6146 return 0;
6147}
6148
6149static int igb_ndo_set_vf_vlan(struct net_device *netdev,
6150 int vf, u16 vlan, u8 qos)
6151{
6152 struct igb_adapter *adapter = netdev_priv(netdev);
6153
6154 if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
6155 return -EINVAL;
6156
6157 return (vlan || qos) ? igb_enable_port_vlan(adapter, vf, vlan, qos) :
6158 igb_disable_port_vlan(adapter, vf);
6159}
6160
6161static int igb_set_vf_vlan_msg(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
6162{
6163 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
6164 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
6165 int ret;
6166
6167 if (adapter->vf_data[vf].pf_vlan)
6168 return -1;
6169
6170 /* VLAN 0 is a special case, don't allow it to be removed */
6171 if (!vid && !add)
6172 return 0;
6173
6174 ret = igb_set_vf_vlan(adapter, vid, !!add, vf);
6175 if (!ret)
6176 igb_set_vf_vlan_strip(adapter, vf, !!vid);
6177 return ret;
6178}
6179
6180static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
6181{
6182 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6183
6184 /* clear flags - except flag that indicates PF has set the MAC */
6185 vf_data->flags &= IGB_VF_FLAG_PF_SET_MAC;
6186 vf_data->last_nack = jiffies;
6187
6188 /* reset vlans for device */
6189 igb_clear_vf_vfta(adapter, vf);
6190 igb_set_vf_vlan(adapter, vf_data->pf_vlan, true, vf);
6191 igb_set_vmvir(adapter, vf_data->pf_vlan |
6192 (vf_data->pf_qos << VLAN_PRIO_SHIFT), vf);
6193 igb_set_vmolr(adapter, vf, !vf_data->pf_vlan);
6194 igb_set_vf_vlan_strip(adapter, vf, !!(vf_data->pf_vlan));
6195
6196 /* reset multicast table array for vf */
6197 adapter->vf_data[vf].num_vf_mc_hashes = 0;
6198
6199 /* Flush and reset the mta with the new values */
6200 igb_set_rx_mode(adapter->netdev);
6201}
6202
6203static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
6204{
6205 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
6206
6207 /* clear mac address as we were hotplug removed/added */
6208 if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
6209 eth_zero_addr(vf_mac);
6210
6211 /* process remaining reset events */
6212 igb_vf_reset(adapter, vf);
6213}
6214
6215static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
6216{
6217 struct e1000_hw *hw = &adapter->hw;
6218 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
6219 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
6220 u32 reg, msgbuf[3];
6221 u8 *addr = (u8 *)(&msgbuf[1]);
6222
6223 /* process all the same items cleared in a function level reset */
6224 igb_vf_reset(adapter, vf);
6225
6226 /* set vf mac address */
6227 igb_rar_set_qsel(adapter, vf_mac, rar_entry, vf);
6228
6229 /* enable transmit and receive for vf */
6230 reg = rd32(E1000_VFTE);
6231 wr32(E1000_VFTE, reg | (1 << vf));
6232 reg = rd32(E1000_VFRE);
6233 wr32(E1000_VFRE, reg | (1 << vf));
6234
6235 adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
6236
6237 /* reply to reset with ack and vf mac address */
6238 if (!is_zero_ether_addr(vf_mac)) {
6239 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
6240 memcpy(addr, vf_mac, ETH_ALEN);
6241 } else {
6242 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_NACK;
6243 }
6244 igb_write_mbx(hw, msgbuf, 3, vf);
6245}
6246
6247static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
6248{
6249 /* The VF MAC Address is stored in a packed array of bytes
6250 * starting at the second 32 bit word of the msg array
6251 */
6252 unsigned char *addr = (char *)&msg[1];
6253 int err = -1;
6254
6255 if (is_valid_ether_addr(addr))
6256 err = igb_set_vf_mac(adapter, vf, addr);
6257
6258 return err;
6259}
6260
6261static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
6262{
6263 struct e1000_hw *hw = &adapter->hw;
6264 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6265 u32 msg = E1000_VT_MSGTYPE_NACK;
6266
6267 /* if device isn't clear to send it shouldn't be reading either */
6268 if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
6269 time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
6270 igb_write_mbx(hw, &msg, 1, vf);
6271 vf_data->last_nack = jiffies;
6272 }
6273}
6274
6275static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
6276{
6277 struct pci_dev *pdev = adapter->pdev;
6278 u32 msgbuf[E1000_VFMAILBOX_SIZE];
6279 struct e1000_hw *hw = &adapter->hw;
6280 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6281 s32 retval;
6282
6283 retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf);
6284
6285 if (retval) {
6286 /* if receive failed revoke VF CTS stats and restart init */
6287 dev_err(&pdev->dev, "Error receiving message from VF\n");
6288 vf_data->flags &= ~IGB_VF_FLAG_CTS;
6289 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
6290 return;
6291 goto out;
6292 }
6293
6294 /* this is a message we already processed, do nothing */
6295 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
6296 return;
6297
6298 /* until the vf completes a reset it should not be
6299 * allowed to start any configuration.
6300 */
6301 if (msgbuf[0] == E1000_VF_RESET) {
6302 igb_vf_reset_msg(adapter, vf);
6303 return;
6304 }
6305
6306 if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
6307 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
6308 return;
6309 retval = -1;
6310 goto out;
6311 }
6312
6313 switch ((msgbuf[0] & 0xFFFF)) {
6314 case E1000_VF_SET_MAC_ADDR:
6315 retval = -EINVAL;
6316 if (!(vf_data->flags & IGB_VF_FLAG_PF_SET_MAC))
6317 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
6318 else
6319 dev_warn(&pdev->dev,
6320 "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
6321 vf);
6322 break;
6323 case E1000_VF_SET_PROMISC:
6324 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
6325 break;
6326 case E1000_VF_SET_MULTICAST:
6327 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
6328 break;
6329 case E1000_VF_SET_LPE:
6330 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
6331 break;
6332 case E1000_VF_SET_VLAN:
6333 retval = -1;
6334 if (vf_data->pf_vlan)
6335 dev_warn(&pdev->dev,
6336 "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
6337 vf);
6338 else
6339 retval = igb_set_vf_vlan_msg(adapter, msgbuf, vf);
6340 break;
6341 default:
6342 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
6343 retval = -1;
6344 break;
6345 }
6346
6347 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
6348out:
6349 /* notify the VF of the results of what it sent us */
6350 if (retval)
6351 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
6352 else
6353 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
6354
6355 igb_write_mbx(hw, msgbuf, 1, vf);
6356}
6357
6358static void igb_msg_task(struct igb_adapter *adapter)
6359{
6360 struct e1000_hw *hw = &adapter->hw;
6361 u32 vf;
6362
6363 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
6364 /* process any reset requests */
6365 if (!igb_check_for_rst(hw, vf))
6366 igb_vf_reset_event(adapter, vf);
6367
6368 /* process any messages pending */
6369 if (!igb_check_for_msg(hw, vf))
6370 igb_rcv_msg_from_vf(adapter, vf);
6371
6372 /* process any acks */
6373 if (!igb_check_for_ack(hw, vf))
6374 igb_rcv_ack_from_vf(adapter, vf);
6375 }
6376}
6377
6378/**
6379 * igb_set_uta - Set unicast filter table address
6380 * @adapter: board private structure
6381 * @set: boolean indicating if we are setting or clearing bits
6382 *
6383 * The unicast table address is a register array of 32-bit registers.
6384 * The table is meant to be used in a way similar to how the MTA is used
6385 * however due to certain limitations in the hardware it is necessary to
6386 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
6387 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
6388 **/
6389static void igb_set_uta(struct igb_adapter *adapter, bool set)
6390{
6391 struct e1000_hw *hw = &adapter->hw;
6392 u32 uta = set ? ~0 : 0;
6393 int i;
6394
6395 /* we only need to do this if VMDq is enabled */
6396 if (!adapter->vfs_allocated_count)
6397 return;
6398
6399 for (i = hw->mac.uta_reg_count; i--;)
6400 array_wr32(E1000_UTA, i, uta);
6401}
6402
6403/**
6404 * igb_intr_msi - Interrupt Handler
6405 * @irq: interrupt number
6406 * @data: pointer to a network interface device structure
6407 **/
6408static irqreturn_t igb_intr_msi(int irq, void *data)
6409{
6410 struct igb_adapter *adapter = data;
6411 struct igb_q_vector *q_vector = adapter->q_vector[0];
6412 struct e1000_hw *hw = &adapter->hw;
6413 /* read ICR disables interrupts using IAM */
6414 u32 icr = rd32(E1000_ICR);
6415
6416 igb_write_itr(q_vector);
6417
6418 if (icr & E1000_ICR_DRSTA)
6419 schedule_work(&adapter->reset_task);
6420
6421 if (icr & E1000_ICR_DOUTSYNC) {
6422 /* HW is reporting DMA is out of sync */
6423 adapter->stats.doosync++;
6424 }
6425
6426 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
6427 hw->mac.get_link_status = 1;
6428 if (!test_bit(__IGB_DOWN, &adapter->state))
6429 mod_timer(&adapter->watchdog_timer, jiffies + 1);
6430 }
6431
6432 if (icr & E1000_ICR_TS)
6433 igb_tsync_interrupt(adapter);
6434
6435 napi_schedule(&q_vector->napi);
6436
6437 return IRQ_HANDLED;
6438}
6439
6440/**
6441 * igb_intr - Legacy Interrupt Handler
6442 * @irq: interrupt number
6443 * @data: pointer to a network interface device structure
6444 **/
6445static irqreturn_t igb_intr(int irq, void *data)
6446{
6447 struct igb_adapter *adapter = data;
6448 struct igb_q_vector *q_vector = adapter->q_vector[0];
6449 struct e1000_hw *hw = &adapter->hw;
6450 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
6451 * need for the IMC write
6452 */
6453 u32 icr = rd32(E1000_ICR);
6454
6455 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
6456 * not set, then the adapter didn't send an interrupt
6457 */
6458 if (!(icr & E1000_ICR_INT_ASSERTED))
6459 return IRQ_NONE;
6460
6461 igb_write_itr(q_vector);
6462
6463 if (icr & E1000_ICR_DRSTA)
6464 schedule_work(&adapter->reset_task);
6465
6466 if (icr & E1000_ICR_DOUTSYNC) {
6467 /* HW is reporting DMA is out of sync */
6468 adapter->stats.doosync++;
6469 }
6470
6471 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
6472 hw->mac.get_link_status = 1;
6473 /* guard against interrupt when we're going down */
6474 if (!test_bit(__IGB_DOWN, &adapter->state))
6475 mod_timer(&adapter->watchdog_timer, jiffies + 1);
6476 }
6477
6478 if (icr & E1000_ICR_TS)
6479 igb_tsync_interrupt(adapter);
6480
6481 napi_schedule(&q_vector->napi);
6482
6483 return IRQ_HANDLED;
6484}
6485
6486static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
6487{
6488 struct igb_adapter *adapter = q_vector->adapter;
6489 struct e1000_hw *hw = &adapter->hw;
6490
6491 if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
6492 (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
6493 if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
6494 igb_set_itr(q_vector);
6495 else
6496 igb_update_ring_itr(q_vector);
6497 }
6498
6499 if (!test_bit(__IGB_DOWN, &adapter->state)) {
6500 if (adapter->flags & IGB_FLAG_HAS_MSIX)
6501 wr32(E1000_EIMS, q_vector->eims_value);
6502 else
6503 igb_irq_enable(adapter);
6504 }
6505}
6506
6507/**
6508 * igb_poll - NAPI Rx polling callback
6509 * @napi: napi polling structure
6510 * @budget: count of how many packets we should handle
6511 **/
6512static int igb_poll(struct napi_struct *napi, int budget)
6513{
6514 struct igb_q_vector *q_vector = container_of(napi,
6515 struct igb_q_vector,
6516 napi);
6517 bool clean_complete = true;
6518 int work_done = 0;
6519
6520#ifdef CONFIG_IGB_DCA
6521 if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
6522 igb_update_dca(q_vector);
6523#endif
6524 if (q_vector->tx.ring)
6525 clean_complete = igb_clean_tx_irq(q_vector);
6526
6527 if (q_vector->rx.ring) {
6528 int cleaned = igb_clean_rx_irq(q_vector, budget);
6529
6530 work_done += cleaned;
6531 clean_complete &= (cleaned < budget);
6532 }
6533
6534 /* If all work not completed, return budget and keep polling */
6535 if (!clean_complete)
6536 return budget;
6537
6538 /* If not enough Rx work done, exit the polling mode */
6539 napi_complete_done(napi, work_done);
6540 igb_ring_irq_enable(q_vector);
6541
6542 return 0;
6543}
6544
6545/**
6546 * igb_clean_tx_irq - Reclaim resources after transmit completes
6547 * @q_vector: pointer to q_vector containing needed info
6548 *
6549 * returns true if ring is completely cleaned
6550 **/
6551static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
6552{
6553 struct igb_adapter *adapter = q_vector->adapter;
6554 struct igb_ring *tx_ring = q_vector->tx.ring;
6555 struct igb_tx_buffer *tx_buffer;
6556 union e1000_adv_tx_desc *tx_desc;
6557 unsigned int total_bytes = 0, total_packets = 0;
6558 unsigned int budget = q_vector->tx.work_limit;
6559 unsigned int i = tx_ring->next_to_clean;
6560
6561 if (test_bit(__IGB_DOWN, &adapter->state))
6562 return true;
6563
6564 tx_buffer = &tx_ring->tx_buffer_info[i];
6565 tx_desc = IGB_TX_DESC(tx_ring, i);
6566 i -= tx_ring->count;
6567
6568 do {
6569 union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
6570
6571 /* if next_to_watch is not set then there is no work pending */
6572 if (!eop_desc)
6573 break;
6574
6575 /* prevent any other reads prior to eop_desc */
6576 read_barrier_depends();
6577
6578 /* if DD is not set pending work has not been completed */
6579 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
6580 break;
6581
6582 /* clear next_to_watch to prevent false hangs */
6583 tx_buffer->next_to_watch = NULL;
6584
6585 /* update the statistics for this packet */
6586 total_bytes += tx_buffer->bytecount;
6587 total_packets += tx_buffer->gso_segs;
6588
6589 /* free the skb */
6590 dev_consume_skb_any(tx_buffer->skb);
6591
6592 /* unmap skb header data */
6593 dma_unmap_single(tx_ring->dev,
6594 dma_unmap_addr(tx_buffer, dma),
6595 dma_unmap_len(tx_buffer, len),
6596 DMA_TO_DEVICE);
6597
6598 /* clear tx_buffer data */
6599 tx_buffer->skb = NULL;
6600 dma_unmap_len_set(tx_buffer, len, 0);
6601
6602 /* clear last DMA location and unmap remaining buffers */
6603 while (tx_desc != eop_desc) {
6604 tx_buffer++;
6605 tx_desc++;
6606 i++;
6607 if (unlikely(!i)) {
6608 i -= tx_ring->count;
6609 tx_buffer = tx_ring->tx_buffer_info;
6610 tx_desc = IGB_TX_DESC(tx_ring, 0);
6611 }
6612
6613 /* unmap any remaining paged data */
6614 if (dma_unmap_len(tx_buffer, len)) {
6615 dma_unmap_page(tx_ring->dev,
6616 dma_unmap_addr(tx_buffer, dma),
6617 dma_unmap_len(tx_buffer, len),
6618 DMA_TO_DEVICE);
6619 dma_unmap_len_set(tx_buffer, len, 0);
6620 }
6621 }
6622
6623 /* move us one more past the eop_desc for start of next pkt */
6624 tx_buffer++;
6625 tx_desc++;
6626 i++;
6627 if (unlikely(!i)) {
6628 i -= tx_ring->count;
6629 tx_buffer = tx_ring->tx_buffer_info;
6630 tx_desc = IGB_TX_DESC(tx_ring, 0);
6631 }
6632
6633 /* issue prefetch for next Tx descriptor */
6634 prefetch(tx_desc);
6635
6636 /* update budget accounting */
6637 budget--;
6638 } while (likely(budget));
6639
6640 netdev_tx_completed_queue(txring_txq(tx_ring),
6641 total_packets, total_bytes);
6642 i += tx_ring->count;
6643 tx_ring->next_to_clean = i;
6644 u64_stats_update_begin(&tx_ring->tx_syncp);
6645 tx_ring->tx_stats.bytes += total_bytes;
6646 tx_ring->tx_stats.packets += total_packets;
6647 u64_stats_update_end(&tx_ring->tx_syncp);
6648 q_vector->tx.total_bytes += total_bytes;
6649 q_vector->tx.total_packets += total_packets;
6650
6651 if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
6652 struct e1000_hw *hw = &adapter->hw;
6653
6654 /* Detect a transmit hang in hardware, this serializes the
6655 * check with the clearing of time_stamp and movement of i
6656 */
6657 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
6658 if (tx_buffer->next_to_watch &&
6659 time_after(jiffies, tx_buffer->time_stamp +
6660 (adapter->tx_timeout_factor * HZ)) &&
6661 !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
6662
6663 /* detected Tx unit hang */
6664 dev_err(tx_ring->dev,
6665 "Detected Tx Unit Hang\n"
6666 " Tx Queue <%d>\n"
6667 " TDH <%x>\n"
6668 " TDT <%x>\n"
6669 " next_to_use <%x>\n"
6670 " next_to_clean <%x>\n"
6671 "buffer_info[next_to_clean]\n"
6672 " time_stamp <%lx>\n"
6673 " next_to_watch <%p>\n"
6674 " jiffies <%lx>\n"
6675 " desc.status <%x>\n",
6676 tx_ring->queue_index,
6677 rd32(E1000_TDH(tx_ring->reg_idx)),
6678 readl(tx_ring->tail),
6679 tx_ring->next_to_use,
6680 tx_ring->next_to_clean,
6681 tx_buffer->time_stamp,
6682 tx_buffer->next_to_watch,
6683 jiffies,
6684 tx_buffer->next_to_watch->wb.status);
6685 netif_stop_subqueue(tx_ring->netdev,
6686 tx_ring->queue_index);
6687
6688 /* we are about to reset, no point in enabling stuff */
6689 return true;
6690 }
6691 }
6692
6693#define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
6694 if (unlikely(total_packets &&
6695 netif_carrier_ok(tx_ring->netdev) &&
6696 igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
6697 /* Make sure that anybody stopping the queue after this
6698 * sees the new next_to_clean.
6699 */
6700 smp_mb();
6701 if (__netif_subqueue_stopped(tx_ring->netdev,
6702 tx_ring->queue_index) &&
6703 !(test_bit(__IGB_DOWN, &adapter->state))) {
6704 netif_wake_subqueue(tx_ring->netdev,
6705 tx_ring->queue_index);
6706
6707 u64_stats_update_begin(&tx_ring->tx_syncp);
6708 tx_ring->tx_stats.restart_queue++;
6709 u64_stats_update_end(&tx_ring->tx_syncp);
6710 }
6711 }
6712
6713 return !!budget;
6714}
6715
6716/**
6717 * igb_reuse_rx_page - page flip buffer and store it back on the ring
6718 * @rx_ring: rx descriptor ring to store buffers on
6719 * @old_buff: donor buffer to have page reused
6720 *
6721 * Synchronizes page for reuse by the adapter
6722 **/
6723static void igb_reuse_rx_page(struct igb_ring *rx_ring,
6724 struct igb_rx_buffer *old_buff)
6725{
6726 struct igb_rx_buffer *new_buff;
6727 u16 nta = rx_ring->next_to_alloc;
6728
6729 new_buff = &rx_ring->rx_buffer_info[nta];
6730
6731 /* update, and store next to alloc */
6732 nta++;
6733 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
6734
6735 /* transfer page from old buffer to new buffer */
6736 *new_buff = *old_buff;
6737
6738 /* sync the buffer for use by the device */
6739 dma_sync_single_range_for_device(rx_ring->dev, old_buff->dma,
6740 old_buff->page_offset,
6741 IGB_RX_BUFSZ,
6742 DMA_FROM_DEVICE);
6743}
6744
6745static inline bool igb_page_is_reserved(struct page *page)
6746{
6747 return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
6748}
6749
6750static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer,
6751 struct page *page,
6752 unsigned int truesize)
6753{
6754 /* avoid re-using remote pages */
6755 if (unlikely(igb_page_is_reserved(page)))
6756 return false;
6757
6758#if (PAGE_SIZE < 8192)
6759 /* if we are only owner of page we can reuse it */
6760 if (unlikely(page_count(page) != 1))
6761 return false;
6762
6763 /* flip page offset to other buffer */
6764 rx_buffer->page_offset ^= IGB_RX_BUFSZ;
6765#else
6766 /* move offset up to the next cache line */
6767 rx_buffer->page_offset += truesize;
6768
6769 if (rx_buffer->page_offset > (PAGE_SIZE - IGB_RX_BUFSZ))
6770 return false;
6771#endif
6772
6773 /* Even if we own the page, we are not allowed to use atomic_set()
6774 * This would break get_page_unless_zero() users.
6775 */
6776 page_ref_inc(page);
6777
6778 return true;
6779}
6780
6781/**
6782 * igb_add_rx_frag - Add contents of Rx buffer to sk_buff
6783 * @rx_ring: rx descriptor ring to transact packets on
6784 * @rx_buffer: buffer containing page to add
6785 * @rx_desc: descriptor containing length of buffer written by hardware
6786 * @skb: sk_buff to place the data into
6787 *
6788 * This function will add the data contained in rx_buffer->page to the skb.
6789 * This is done either through a direct copy if the data in the buffer is
6790 * less than the skb header size, otherwise it will just attach the page as
6791 * a frag to the skb.
6792 *
6793 * The function will then update the page offset if necessary and return
6794 * true if the buffer can be reused by the adapter.
6795 **/
6796static bool igb_add_rx_frag(struct igb_ring *rx_ring,
6797 struct igb_rx_buffer *rx_buffer,
6798 union e1000_adv_rx_desc *rx_desc,
6799 struct sk_buff *skb)
6800{
6801 struct page *page = rx_buffer->page;
6802 unsigned char *va = page_address(page) + rx_buffer->page_offset;
6803 unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
6804#if (PAGE_SIZE < 8192)
6805 unsigned int truesize = IGB_RX_BUFSZ;
6806#else
6807 unsigned int truesize = SKB_DATA_ALIGN(size);
6808#endif
6809 unsigned int pull_len;
6810
6811 if (unlikely(skb_is_nonlinear(skb)))
6812 goto add_tail_frag;
6813
6814 if (unlikely(igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP))) {
6815 igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
6816 va += IGB_TS_HDR_LEN;
6817 size -= IGB_TS_HDR_LEN;
6818 }
6819
6820 if (likely(size <= IGB_RX_HDR_LEN)) {
6821 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
6822
6823 /* page is not reserved, we can reuse buffer as-is */
6824 if (likely(!igb_page_is_reserved(page)))
6825 return true;
6826
6827 /* this page cannot be reused so discard it */
6828 __free_page(page);
6829 return false;
6830 }
6831
6832 /* we need the header to contain the greater of either ETH_HLEN or
6833 * 60 bytes if the skb->len is less than 60 for skb_pad.
6834 */
6835 pull_len = eth_get_headlen(va, IGB_RX_HDR_LEN);
6836
6837 /* align pull length to size of long to optimize memcpy performance */
6838 memcpy(__skb_put(skb, pull_len), va, ALIGN(pull_len, sizeof(long)));
6839
6840 /* update all of the pointers */
6841 va += pull_len;
6842 size -= pull_len;
6843
6844add_tail_frag:
6845 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
6846 (unsigned long)va & ~PAGE_MASK, size, truesize);
6847
6848 return igb_can_reuse_rx_page(rx_buffer, page, truesize);
6849}
6850
6851static struct sk_buff *igb_fetch_rx_buffer(struct igb_ring *rx_ring,
6852 union e1000_adv_rx_desc *rx_desc,
6853 struct sk_buff *skb)
6854{
6855 struct igb_rx_buffer *rx_buffer;
6856 struct page *page;
6857
6858 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
6859 page = rx_buffer->page;
6860 prefetchw(page);
6861
6862 if (likely(!skb)) {
6863 void *page_addr = page_address(page) +
6864 rx_buffer->page_offset;
6865
6866 /* prefetch first cache line of first page */
6867 prefetch(page_addr);
6868#if L1_CACHE_BYTES < 128
6869 prefetch(page_addr + L1_CACHE_BYTES);
6870#endif
6871
6872 /* allocate a skb to store the frags */
6873 skb = napi_alloc_skb(&rx_ring->q_vector->napi, IGB_RX_HDR_LEN);
6874 if (unlikely(!skb)) {
6875 rx_ring->rx_stats.alloc_failed++;
6876 return NULL;
6877 }
6878
6879 /* we will be copying header into skb->data in
6880 * pskb_may_pull so it is in our interest to prefetch
6881 * it now to avoid a possible cache miss
6882 */
6883 prefetchw(skb->data);
6884 }
6885
6886 /* we are reusing so sync this buffer for CPU use */
6887 dma_sync_single_range_for_cpu(rx_ring->dev,
6888 rx_buffer->dma,
6889 rx_buffer->page_offset,
6890 IGB_RX_BUFSZ,
6891 DMA_FROM_DEVICE);
6892
6893 /* pull page into skb */
6894 if (igb_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
6895 /* hand second half of page back to the ring */
6896 igb_reuse_rx_page(rx_ring, rx_buffer);
6897 } else {
6898 /* we are not reusing the buffer so unmap it */
6899 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
6900 PAGE_SIZE, DMA_FROM_DEVICE);
6901 }
6902
6903 /* clear contents of rx_buffer */
6904 rx_buffer->page = NULL;
6905
6906 return skb;
6907}
6908
6909static inline void igb_rx_checksum(struct igb_ring *ring,
6910 union e1000_adv_rx_desc *rx_desc,
6911 struct sk_buff *skb)
6912{
6913 skb_checksum_none_assert(skb);
6914
6915 /* Ignore Checksum bit is set */
6916 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
6917 return;
6918
6919 /* Rx checksum disabled via ethtool */
6920 if (!(ring->netdev->features & NETIF_F_RXCSUM))
6921 return;
6922
6923 /* TCP/UDP checksum error bit is set */
6924 if (igb_test_staterr(rx_desc,
6925 E1000_RXDEXT_STATERR_TCPE |
6926 E1000_RXDEXT_STATERR_IPE)) {
6927 /* work around errata with sctp packets where the TCPE aka
6928 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
6929 * packets, (aka let the stack check the crc32c)
6930 */
6931 if (!((skb->len == 60) &&
6932 test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
6933 u64_stats_update_begin(&ring->rx_syncp);
6934 ring->rx_stats.csum_err++;
6935 u64_stats_update_end(&ring->rx_syncp);
6936 }
6937 /* let the stack verify checksum errors */
6938 return;
6939 }
6940 /* It must be a TCP or UDP packet with a valid checksum */
6941 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
6942 E1000_RXD_STAT_UDPCS))
6943 skb->ip_summed = CHECKSUM_UNNECESSARY;
6944
6945 dev_dbg(ring->dev, "cksum success: bits %08X\n",
6946 le32_to_cpu(rx_desc->wb.upper.status_error));
6947}
6948
6949static inline void igb_rx_hash(struct igb_ring *ring,
6950 union e1000_adv_rx_desc *rx_desc,
6951 struct sk_buff *skb)
6952{
6953 if (ring->netdev->features & NETIF_F_RXHASH)
6954 skb_set_hash(skb,
6955 le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
6956 PKT_HASH_TYPE_L3);
6957}
6958
6959/**
6960 * igb_is_non_eop - process handling of non-EOP buffers
6961 * @rx_ring: Rx ring being processed
6962 * @rx_desc: Rx descriptor for current buffer
6963 * @skb: current socket buffer containing buffer in progress
6964 *
6965 * This function updates next to clean. If the buffer is an EOP buffer
6966 * this function exits returning false, otherwise it will place the
6967 * sk_buff in the next buffer to be chained and return true indicating
6968 * that this is in fact a non-EOP buffer.
6969 **/
6970static bool igb_is_non_eop(struct igb_ring *rx_ring,
6971 union e1000_adv_rx_desc *rx_desc)
6972{
6973 u32 ntc = rx_ring->next_to_clean + 1;
6974
6975 /* fetch, update, and store next to clean */
6976 ntc = (ntc < rx_ring->count) ? ntc : 0;
6977 rx_ring->next_to_clean = ntc;
6978
6979 prefetch(IGB_RX_DESC(rx_ring, ntc));
6980
6981 if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
6982 return false;
6983
6984 return true;
6985}
6986
6987/**
6988 * igb_cleanup_headers - Correct corrupted or empty headers
6989 * @rx_ring: rx descriptor ring packet is being transacted on
6990 * @rx_desc: pointer to the EOP Rx descriptor
6991 * @skb: pointer to current skb being fixed
6992 *
6993 * Address the case where we are pulling data in on pages only
6994 * and as such no data is present in the skb header.
6995 *
6996 * In addition if skb is not at least 60 bytes we need to pad it so that
6997 * it is large enough to qualify as a valid Ethernet frame.
6998 *
6999 * Returns true if an error was encountered and skb was freed.
7000 **/
7001static bool igb_cleanup_headers(struct igb_ring *rx_ring,
7002 union e1000_adv_rx_desc *rx_desc,
7003 struct sk_buff *skb)
7004{
7005 if (unlikely((igb_test_staterr(rx_desc,
7006 E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
7007 struct net_device *netdev = rx_ring->netdev;
7008 if (!(netdev->features & NETIF_F_RXALL)) {
7009 dev_kfree_skb_any(skb);
7010 return true;
7011 }
7012 }
7013
7014 /* if eth_skb_pad returns an error the skb was freed */
7015 if (eth_skb_pad(skb))
7016 return true;
7017
7018 return false;
7019}
7020
7021/**
7022 * igb_process_skb_fields - Populate skb header fields from Rx descriptor
7023 * @rx_ring: rx descriptor ring packet is being transacted on
7024 * @rx_desc: pointer to the EOP Rx descriptor
7025 * @skb: pointer to current skb being populated
7026 *
7027 * This function checks the ring, descriptor, and packet information in
7028 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
7029 * other fields within the skb.
7030 **/
7031static void igb_process_skb_fields(struct igb_ring *rx_ring,
7032 union e1000_adv_rx_desc *rx_desc,
7033 struct sk_buff *skb)
7034{
7035 struct net_device *dev = rx_ring->netdev;
7036
7037 igb_rx_hash(rx_ring, rx_desc, skb);
7038
7039 igb_rx_checksum(rx_ring, rx_desc, skb);
7040
7041 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TS) &&
7042 !igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP))
7043 igb_ptp_rx_rgtstamp(rx_ring->q_vector, skb);
7044
7045 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
7046 igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
7047 u16 vid;
7048
7049 if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
7050 test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
7051 vid = be16_to_cpu(rx_desc->wb.upper.vlan);
7052 else
7053 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
7054
7055 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
7056 }
7057
7058 skb_record_rx_queue(skb, rx_ring->queue_index);
7059
7060 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
7061}
7062
7063static int igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
7064{
7065 struct igb_ring *rx_ring = q_vector->rx.ring;
7066 struct sk_buff *skb = rx_ring->skb;
7067 unsigned int total_bytes = 0, total_packets = 0;
7068 u16 cleaned_count = igb_desc_unused(rx_ring);
7069
7070 while (likely(total_packets < budget)) {
7071 union e1000_adv_rx_desc *rx_desc;
7072
7073 /* return some buffers to hardware, one at a time is too slow */
7074 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
7075 igb_alloc_rx_buffers(rx_ring, cleaned_count);
7076 cleaned_count = 0;
7077 }
7078
7079 rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
7080
7081 if (!rx_desc->wb.upper.status_error)
7082 break;
7083
7084 /* This memory barrier is needed to keep us from reading
7085 * any other fields out of the rx_desc until we know the
7086 * descriptor has been written back
7087 */
7088 dma_rmb();
7089
7090 /* retrieve a buffer from the ring */
7091 skb = igb_fetch_rx_buffer(rx_ring, rx_desc, skb);
7092
7093 /* exit if we failed to retrieve a buffer */
7094 if (!skb)
7095 break;
7096
7097 cleaned_count++;
7098
7099 /* fetch next buffer in frame if non-eop */
7100 if (igb_is_non_eop(rx_ring, rx_desc))
7101 continue;
7102
7103 /* verify the packet layout is correct */
7104 if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
7105 skb = NULL;
7106 continue;
7107 }
7108
7109 /* probably a little skewed due to removing CRC */
7110 total_bytes += skb->len;
7111
7112 /* populate checksum, timestamp, VLAN, and protocol */
7113 igb_process_skb_fields(rx_ring, rx_desc, skb);
7114
7115 napi_gro_receive(&q_vector->napi, skb);
7116
7117 /* reset skb pointer */
7118 skb = NULL;
7119
7120 /* update budget accounting */
7121 total_packets++;
7122 }
7123
7124 /* place incomplete frames back on ring for completion */
7125 rx_ring->skb = skb;
7126
7127 u64_stats_update_begin(&rx_ring->rx_syncp);
7128 rx_ring->rx_stats.packets += total_packets;
7129 rx_ring->rx_stats.bytes += total_bytes;
7130 u64_stats_update_end(&rx_ring->rx_syncp);
7131 q_vector->rx.total_packets += total_packets;
7132 q_vector->rx.total_bytes += total_bytes;
7133
7134 if (cleaned_count)
7135 igb_alloc_rx_buffers(rx_ring, cleaned_count);
7136
7137 return total_packets;
7138}
7139
7140static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
7141 struct igb_rx_buffer *bi)
7142{
7143 struct page *page = bi->page;
7144 dma_addr_t dma;
7145
7146 /* since we are recycling buffers we should seldom need to alloc */
7147 if (likely(page))
7148 return true;
7149
7150 /* alloc new page for storage */
7151 page = dev_alloc_page();
7152 if (unlikely(!page)) {
7153 rx_ring->rx_stats.alloc_failed++;
7154 return false;
7155 }
7156
7157 /* map page for use */
7158 dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE, DMA_FROM_DEVICE);
7159
7160 /* if mapping failed free memory back to system since
7161 * there isn't much point in holding memory we can't use
7162 */
7163 if (dma_mapping_error(rx_ring->dev, dma)) {
7164 __free_page(page);
7165
7166 rx_ring->rx_stats.alloc_failed++;
7167 return false;
7168 }
7169
7170 bi->dma = dma;
7171 bi->page = page;
7172 bi->page_offset = 0;
7173
7174 return true;
7175}
7176
7177/**
7178 * igb_alloc_rx_buffers - Replace used receive buffers; packet split
7179 * @adapter: address of board private structure
7180 **/
7181void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
7182{
7183 union e1000_adv_rx_desc *rx_desc;
7184 struct igb_rx_buffer *bi;
7185 u16 i = rx_ring->next_to_use;
7186
7187 /* nothing to do */
7188 if (!cleaned_count)
7189 return;
7190
7191 rx_desc = IGB_RX_DESC(rx_ring, i);
7192 bi = &rx_ring->rx_buffer_info[i];
7193 i -= rx_ring->count;
7194
7195 do {
7196 if (!igb_alloc_mapped_page(rx_ring, bi))
7197 break;
7198
7199 /* Refresh the desc even if buffer_addrs didn't change
7200 * because each write-back erases this info.
7201 */
7202 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
7203
7204 rx_desc++;
7205 bi++;
7206 i++;
7207 if (unlikely(!i)) {
7208 rx_desc = IGB_RX_DESC(rx_ring, 0);
7209 bi = rx_ring->rx_buffer_info;
7210 i -= rx_ring->count;
7211 }
7212
7213 /* clear the status bits for the next_to_use descriptor */
7214 rx_desc->wb.upper.status_error = 0;
7215
7216 cleaned_count--;
7217 } while (cleaned_count);
7218
7219 i += rx_ring->count;
7220
7221 if (rx_ring->next_to_use != i) {
7222 /* record the next descriptor to use */
7223 rx_ring->next_to_use = i;
7224
7225 /* update next to alloc since we have filled the ring */
7226 rx_ring->next_to_alloc = i;
7227
7228 /* Force memory writes to complete before letting h/w
7229 * know there are new descriptors to fetch. (Only
7230 * applicable for weak-ordered memory model archs,
7231 * such as IA-64).
7232 */
7233 wmb();
7234 writel(i, rx_ring->tail);
7235 }
7236}
7237
7238/**
7239 * igb_mii_ioctl -
7240 * @netdev:
7241 * @ifreq:
7242 * @cmd:
7243 **/
7244static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
7245{
7246 struct igb_adapter *adapter = netdev_priv(netdev);
7247 struct mii_ioctl_data *data = if_mii(ifr);
7248
7249 if (adapter->hw.phy.media_type != e1000_media_type_copper)
7250 return -EOPNOTSUPP;
7251
7252 switch (cmd) {
7253 case SIOCGMIIPHY:
7254 data->phy_id = adapter->hw.phy.addr;
7255 break;
7256 case SIOCGMIIREG:
7257 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
7258 &data->val_out))
7259 return -EIO;
7260 break;
7261 case SIOCSMIIREG:
7262 default:
7263 return -EOPNOTSUPP;
7264 }
7265 return 0;
7266}
7267
7268/**
7269 * igb_ioctl -
7270 * @netdev:
7271 * @ifreq:
7272 * @cmd:
7273 **/
7274static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
7275{
7276 switch (cmd) {
7277 case SIOCGMIIPHY:
7278 case SIOCGMIIREG:
7279 case SIOCSMIIREG:
7280 return igb_mii_ioctl(netdev, ifr, cmd);
7281 case SIOCGHWTSTAMP:
7282 return igb_ptp_get_ts_config(netdev, ifr);
7283 case SIOCSHWTSTAMP:
7284 return igb_ptp_set_ts_config(netdev, ifr);
7285 default:
7286 return -EOPNOTSUPP;
7287 }
7288}
7289
7290void igb_read_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
7291{
7292 struct igb_adapter *adapter = hw->back;
7293
7294 pci_read_config_word(adapter->pdev, reg, value);
7295}
7296
7297void igb_write_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
7298{
7299 struct igb_adapter *adapter = hw->back;
7300
7301 pci_write_config_word(adapter->pdev, reg, *value);
7302}
7303
7304s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
7305{
7306 struct igb_adapter *adapter = hw->back;
7307
7308 if (pcie_capability_read_word(adapter->pdev, reg, value))
7309 return -E1000_ERR_CONFIG;
7310
7311 return 0;
7312}
7313
7314s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
7315{
7316 struct igb_adapter *adapter = hw->back;
7317
7318 if (pcie_capability_write_word(adapter->pdev, reg, *value))
7319 return -E1000_ERR_CONFIG;
7320
7321 return 0;
7322}
7323
7324static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
7325{
7326 struct igb_adapter *adapter = netdev_priv(netdev);
7327 struct e1000_hw *hw = &adapter->hw;
7328 u32 ctrl, rctl;
7329 bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
7330
7331 if (enable) {
7332 /* enable VLAN tag insert/strip */
7333 ctrl = rd32(E1000_CTRL);
7334 ctrl |= E1000_CTRL_VME;
7335 wr32(E1000_CTRL, ctrl);
7336
7337 /* Disable CFI check */
7338 rctl = rd32(E1000_RCTL);
7339 rctl &= ~E1000_RCTL_CFIEN;
7340 wr32(E1000_RCTL, rctl);
7341 } else {
7342 /* disable VLAN tag insert/strip */
7343 ctrl = rd32(E1000_CTRL);
7344 ctrl &= ~E1000_CTRL_VME;
7345 wr32(E1000_CTRL, ctrl);
7346 }
7347
7348 igb_set_vf_vlan_strip(adapter, adapter->vfs_allocated_count, enable);
7349}
7350
7351static int igb_vlan_rx_add_vid(struct net_device *netdev,
7352 __be16 proto, u16 vid)
7353{
7354 struct igb_adapter *adapter = netdev_priv(netdev);
7355 struct e1000_hw *hw = &adapter->hw;
7356 int pf_id = adapter->vfs_allocated_count;
7357
7358 /* add the filter since PF can receive vlans w/o entry in vlvf */
7359 if (!vid || !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
7360 igb_vfta_set(hw, vid, pf_id, true, !!vid);
7361
7362 set_bit(vid, adapter->active_vlans);
7363
7364 return 0;
7365}
7366
7367static int igb_vlan_rx_kill_vid(struct net_device *netdev,
7368 __be16 proto, u16 vid)
7369{
7370 struct igb_adapter *adapter = netdev_priv(netdev);
7371 int pf_id = adapter->vfs_allocated_count;
7372 struct e1000_hw *hw = &adapter->hw;
7373
7374 /* remove VID from filter table */
7375 if (vid && !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
7376 igb_vfta_set(hw, vid, pf_id, false, true);
7377
7378 clear_bit(vid, adapter->active_vlans);
7379
7380 return 0;
7381}
7382
7383static void igb_restore_vlan(struct igb_adapter *adapter)
7384{
7385 u16 vid = 1;
7386
7387 igb_vlan_mode(adapter->netdev, adapter->netdev->features);
7388 igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
7389
7390 for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
7391 igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
7392}
7393
7394int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
7395{
7396 struct pci_dev *pdev = adapter->pdev;
7397 struct e1000_mac_info *mac = &adapter->hw.mac;
7398
7399 mac->autoneg = 0;
7400
7401 /* Make sure dplx is at most 1 bit and lsb of speed is not set
7402 * for the switch() below to work
7403 */
7404 if ((spd & 1) || (dplx & ~1))
7405 goto err_inval;
7406
7407 /* Fiber NIC's only allow 1000 gbps Full duplex
7408 * and 100Mbps Full duplex for 100baseFx sfp
7409 */
7410 if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
7411 switch (spd + dplx) {
7412 case SPEED_10 + DUPLEX_HALF:
7413 case SPEED_10 + DUPLEX_FULL:
7414 case SPEED_100 + DUPLEX_HALF:
7415 goto err_inval;
7416 default:
7417 break;
7418 }
7419 }
7420
7421 switch (spd + dplx) {
7422 case SPEED_10 + DUPLEX_HALF:
7423 mac->forced_speed_duplex = ADVERTISE_10_HALF;
7424 break;
7425 case SPEED_10 + DUPLEX_FULL:
7426 mac->forced_speed_duplex = ADVERTISE_10_FULL;
7427 break;
7428 case SPEED_100 + DUPLEX_HALF:
7429 mac->forced_speed_duplex = ADVERTISE_100_HALF;
7430 break;
7431 case SPEED_100 + DUPLEX_FULL:
7432 mac->forced_speed_duplex = ADVERTISE_100_FULL;
7433 break;
7434 case SPEED_1000 + DUPLEX_FULL:
7435 mac->autoneg = 1;
7436 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
7437 break;
7438 case SPEED_1000 + DUPLEX_HALF: /* not supported */
7439 default:
7440 goto err_inval;
7441 }
7442
7443 /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
7444 adapter->hw.phy.mdix = AUTO_ALL_MODES;
7445
7446 return 0;
7447
7448err_inval:
7449 dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
7450 return -EINVAL;
7451}
7452
7453static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
7454 bool runtime)
7455{
7456 struct net_device *netdev = pci_get_drvdata(pdev);
7457 struct igb_adapter *adapter = netdev_priv(netdev);
7458 struct e1000_hw *hw = &adapter->hw;
7459 u32 ctrl, rctl, status;
7460 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
7461#ifdef CONFIG_PM
7462 int retval = 0;
7463#endif
7464
7465 netif_device_detach(netdev);
7466
7467 if (netif_running(netdev))
7468 __igb_close(netdev, true);
7469
7470 igb_clear_interrupt_scheme(adapter);
7471
7472#ifdef CONFIG_PM
7473 retval = pci_save_state(pdev);
7474 if (retval)
7475 return retval;
7476#endif
7477
7478 status = rd32(E1000_STATUS);
7479 if (status & E1000_STATUS_LU)
7480 wufc &= ~E1000_WUFC_LNKC;
7481
7482 if (wufc) {
7483 igb_setup_rctl(adapter);
7484 igb_set_rx_mode(netdev);
7485
7486 /* turn on all-multi mode if wake on multicast is enabled */
7487 if (wufc & E1000_WUFC_MC) {
7488 rctl = rd32(E1000_RCTL);
7489 rctl |= E1000_RCTL_MPE;
7490 wr32(E1000_RCTL, rctl);
7491 }
7492
7493 ctrl = rd32(E1000_CTRL);
7494 /* advertise wake from D3Cold */
7495 #define E1000_CTRL_ADVD3WUC 0x00100000
7496 /* phy power management enable */
7497 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
7498 ctrl |= E1000_CTRL_ADVD3WUC;
7499 wr32(E1000_CTRL, ctrl);
7500
7501 /* Allow time for pending master requests to run */
7502 igb_disable_pcie_master(hw);
7503
7504 wr32(E1000_WUC, E1000_WUC_PME_EN);
7505 wr32(E1000_WUFC, wufc);
7506 } else {
7507 wr32(E1000_WUC, 0);
7508 wr32(E1000_WUFC, 0);
7509 }
7510
7511 *enable_wake = wufc || adapter->en_mng_pt;
7512 if (!*enable_wake)
7513 igb_power_down_link(adapter);
7514 else
7515 igb_power_up_link(adapter);
7516
7517 /* Release control of h/w to f/w. If f/w is AMT enabled, this
7518 * would have already happened in close and is redundant.
7519 */
7520 igb_release_hw_control(adapter);
7521
7522 pci_disable_device(pdev);
7523
7524 return 0;
7525}
7526
7527#ifdef CONFIG_PM
7528#ifdef CONFIG_PM_SLEEP
7529static int igb_suspend(struct device *dev)
7530{
7531 int retval;
7532 bool wake;
7533 struct pci_dev *pdev = to_pci_dev(dev);
7534
7535 retval = __igb_shutdown(pdev, &wake, 0);
7536 if (retval)
7537 return retval;
7538
7539 if (wake) {
7540 pci_prepare_to_sleep(pdev);
7541 } else {
7542 pci_wake_from_d3(pdev, false);
7543 pci_set_power_state(pdev, PCI_D3hot);
7544 }
7545
7546 return 0;
7547}
7548#endif /* CONFIG_PM_SLEEP */
7549
7550static int igb_resume(struct device *dev)
7551{
7552 struct pci_dev *pdev = to_pci_dev(dev);
7553 struct net_device *netdev = pci_get_drvdata(pdev);
7554 struct igb_adapter *adapter = netdev_priv(netdev);
7555 struct e1000_hw *hw = &adapter->hw;
7556 u32 err;
7557
7558 pci_set_power_state(pdev, PCI_D0);
7559 pci_restore_state(pdev);
7560 pci_save_state(pdev);
7561
7562 if (!pci_device_is_present(pdev))
7563 return -ENODEV;
7564 err = pci_enable_device_mem(pdev);
7565 if (err) {
7566 dev_err(&pdev->dev,
7567 "igb: Cannot enable PCI device from suspend\n");
7568 return err;
7569 }
7570 pci_set_master(pdev);
7571
7572 pci_enable_wake(pdev, PCI_D3hot, 0);
7573 pci_enable_wake(pdev, PCI_D3cold, 0);
7574
7575 if (igb_init_interrupt_scheme(adapter, true)) {
7576 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
7577 rtnl_unlock();
7578 return -ENOMEM;
7579 }
7580
7581 igb_reset(adapter);
7582
7583 /* let the f/w know that the h/w is now under the control of the
7584 * driver.
7585 */
7586 igb_get_hw_control(adapter);
7587
7588 wr32(E1000_WUS, ~0);
7589
7590 if (netdev->flags & IFF_UP) {
7591 rtnl_lock();
7592 err = __igb_open(netdev, true);
7593 rtnl_unlock();
7594 if (err)
7595 return err;
7596 }
7597
7598 netif_device_attach(netdev);
7599 return 0;
7600}
7601
7602static int igb_runtime_idle(struct device *dev)
7603{
7604 struct pci_dev *pdev = to_pci_dev(dev);
7605 struct net_device *netdev = pci_get_drvdata(pdev);
7606 struct igb_adapter *adapter = netdev_priv(netdev);
7607
7608 if (!igb_has_link(adapter))
7609 pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
7610
7611 return -EBUSY;
7612}
7613
7614static int igb_runtime_suspend(struct device *dev)
7615{
7616 struct pci_dev *pdev = to_pci_dev(dev);
7617 int retval;
7618 bool wake;
7619
7620 retval = __igb_shutdown(pdev, &wake, 1);
7621 if (retval)
7622 return retval;
7623
7624 if (wake) {
7625 pci_prepare_to_sleep(pdev);
7626 } else {
7627 pci_wake_from_d3(pdev, false);
7628 pci_set_power_state(pdev, PCI_D3hot);
7629 }
7630
7631 return 0;
7632}
7633
7634static int igb_runtime_resume(struct device *dev)
7635{
7636 return igb_resume(dev);
7637}
7638#endif /* CONFIG_PM */
7639
7640static void igb_shutdown(struct pci_dev *pdev)
7641{
7642 bool wake;
7643
7644 __igb_shutdown(pdev, &wake, 0);
7645
7646 if (system_state == SYSTEM_POWER_OFF) {
7647 pci_wake_from_d3(pdev, wake);
7648 pci_set_power_state(pdev, PCI_D3hot);
7649 }
7650}
7651
7652#ifdef CONFIG_PCI_IOV
7653static int igb_sriov_reinit(struct pci_dev *dev)
7654{
7655 struct net_device *netdev = pci_get_drvdata(dev);
7656 struct igb_adapter *adapter = netdev_priv(netdev);
7657 struct pci_dev *pdev = adapter->pdev;
7658
7659 rtnl_lock();
7660
7661 if (netif_running(netdev))
7662 igb_close(netdev);
7663 else
7664 igb_reset(adapter);
7665
7666 igb_clear_interrupt_scheme(adapter);
7667
7668 igb_init_queue_configuration(adapter);
7669
7670 if (igb_init_interrupt_scheme(adapter, true)) {
7671 rtnl_unlock();
7672 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
7673 return -ENOMEM;
7674 }
7675
7676 if (netif_running(netdev))
7677 igb_open(netdev);
7678
7679 rtnl_unlock();
7680
7681 return 0;
7682}
7683
7684static int igb_pci_disable_sriov(struct pci_dev *dev)
7685{
7686 int err = igb_disable_sriov(dev);
7687
7688 if (!err)
7689 err = igb_sriov_reinit(dev);
7690
7691 return err;
7692}
7693
7694static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs)
7695{
7696 int err = igb_enable_sriov(dev, num_vfs);
7697
7698 if (err)
7699 goto out;
7700
7701 err = igb_sriov_reinit(dev);
7702 if (!err)
7703 return num_vfs;
7704
7705out:
7706 return err;
7707}
7708
7709#endif
7710static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
7711{
7712#ifdef CONFIG_PCI_IOV
7713 if (num_vfs == 0)
7714 return igb_pci_disable_sriov(dev);
7715 else
7716 return igb_pci_enable_sriov(dev, num_vfs);
7717#endif
7718 return 0;
7719}
7720
7721#ifdef CONFIG_NET_POLL_CONTROLLER
7722/* Polling 'interrupt' - used by things like netconsole to send skbs
7723 * without having to re-enable interrupts. It's not called while
7724 * the interrupt routine is executing.
7725 */
7726static void igb_netpoll(struct net_device *netdev)
7727{
7728 struct igb_adapter *adapter = netdev_priv(netdev);
7729 struct e1000_hw *hw = &adapter->hw;
7730 struct igb_q_vector *q_vector;
7731 int i;
7732
7733 for (i = 0; i < adapter->num_q_vectors; i++) {
7734 q_vector = adapter->q_vector[i];
7735 if (adapter->flags & IGB_FLAG_HAS_MSIX)
7736 wr32(E1000_EIMC, q_vector->eims_value);
7737 else
7738 igb_irq_disable(adapter);
7739 napi_schedule(&q_vector->napi);
7740 }
7741}
7742#endif /* CONFIG_NET_POLL_CONTROLLER */
7743
7744/**
7745 * igb_io_error_detected - called when PCI error is detected
7746 * @pdev: Pointer to PCI device
7747 * @state: The current pci connection state
7748 *
7749 * This function is called after a PCI bus error affecting
7750 * this device has been detected.
7751 **/
7752static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
7753 pci_channel_state_t state)
7754{
7755 struct net_device *netdev = pci_get_drvdata(pdev);
7756 struct igb_adapter *adapter = netdev_priv(netdev);
7757
7758 netif_device_detach(netdev);
7759
7760 if (state == pci_channel_io_perm_failure)
7761 return PCI_ERS_RESULT_DISCONNECT;
7762
7763 if (netif_running(netdev))
7764 igb_down(adapter);
7765 pci_disable_device(pdev);
7766
7767 /* Request a slot slot reset. */
7768 return PCI_ERS_RESULT_NEED_RESET;
7769}
7770
7771/**
7772 * igb_io_slot_reset - called after the pci bus has been reset.
7773 * @pdev: Pointer to PCI device
7774 *
7775 * Restart the card from scratch, as if from a cold-boot. Implementation
7776 * resembles the first-half of the igb_resume routine.
7777 **/
7778static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
7779{
7780 struct net_device *netdev = pci_get_drvdata(pdev);
7781 struct igb_adapter *adapter = netdev_priv(netdev);
7782 struct e1000_hw *hw = &adapter->hw;
7783 pci_ers_result_t result;
7784 int err;
7785
7786 if (pci_enable_device_mem(pdev)) {
7787 dev_err(&pdev->dev,
7788 "Cannot re-enable PCI device after reset.\n");
7789 result = PCI_ERS_RESULT_DISCONNECT;
7790 } else {
7791 pci_set_master(pdev);
7792 pci_restore_state(pdev);
7793 pci_save_state(pdev);
7794
7795 pci_enable_wake(pdev, PCI_D3hot, 0);
7796 pci_enable_wake(pdev, PCI_D3cold, 0);
7797
7798 igb_reset(adapter);
7799 wr32(E1000_WUS, ~0);
7800 result = PCI_ERS_RESULT_RECOVERED;
7801 }
7802
7803 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7804 if (err) {
7805 dev_err(&pdev->dev,
7806 "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n",
7807 err);
7808 /* non-fatal, continue */
7809 }
7810
7811 return result;
7812}
7813
7814/**
7815 * igb_io_resume - called when traffic can start flowing again.
7816 * @pdev: Pointer to PCI device
7817 *
7818 * This callback is called when the error recovery driver tells us that
7819 * its OK to resume normal operation. Implementation resembles the
7820 * second-half of the igb_resume routine.
7821 */
7822static void igb_io_resume(struct pci_dev *pdev)
7823{
7824 struct net_device *netdev = pci_get_drvdata(pdev);
7825 struct igb_adapter *adapter = netdev_priv(netdev);
7826
7827 if (netif_running(netdev)) {
7828 if (igb_up(adapter)) {
7829 dev_err(&pdev->dev, "igb_up failed after reset\n");
7830 return;
7831 }
7832 }
7833
7834 netif_device_attach(netdev);
7835
7836 /* let the f/w know that the h/w is now under the control of the
7837 * driver.
7838 */
7839 igb_get_hw_control(adapter);
7840}
7841
7842static void igb_rar_set_qsel(struct igb_adapter *adapter, u8 *addr, u32 index,
7843 u8 qsel)
7844{
7845 struct e1000_hw *hw = &adapter->hw;
7846 u32 rar_low, rar_high;
7847
7848 /* HW expects these in little endian so we reverse the byte order
7849 * from network order (big endian) to CPU endian
7850 */
7851 rar_low = le32_to_cpup((__be32 *)(addr));
7852 rar_high = le16_to_cpup((__be16 *)(addr + 4));
7853
7854 /* Indicate to hardware the Address is Valid. */
7855 rar_high |= E1000_RAH_AV;
7856
7857 if (hw->mac.type == e1000_82575)
7858 rar_high |= E1000_RAH_POOL_1 * qsel;
7859 else
7860 rar_high |= E1000_RAH_POOL_1 << qsel;
7861
7862 wr32(E1000_RAL(index), rar_low);
7863 wrfl();
7864 wr32(E1000_RAH(index), rar_high);
7865 wrfl();
7866}
7867
7868static int igb_set_vf_mac(struct igb_adapter *adapter,
7869 int vf, unsigned char *mac_addr)
7870{
7871 struct e1000_hw *hw = &adapter->hw;
7872 /* VF MAC addresses start at end of receive addresses and moves
7873 * towards the first, as a result a collision should not be possible
7874 */
7875 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
7876
7877 memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
7878
7879 igb_rar_set_qsel(adapter, mac_addr, rar_entry, vf);
7880
7881 return 0;
7882}
7883
7884static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
7885{
7886 struct igb_adapter *adapter = netdev_priv(netdev);
7887 if (!is_valid_ether_addr(mac) || (vf >= adapter->vfs_allocated_count))
7888 return -EINVAL;
7889 adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
7890 dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n", mac, vf);
7891 dev_info(&adapter->pdev->dev,
7892 "Reload the VF driver to make this change effective.");
7893 if (test_bit(__IGB_DOWN, &adapter->state)) {
7894 dev_warn(&adapter->pdev->dev,
7895 "The VF MAC address has been set, but the PF device is not up.\n");
7896 dev_warn(&adapter->pdev->dev,
7897 "Bring the PF device up before attempting to use the VF device.\n");
7898 }
7899 return igb_set_vf_mac(adapter, vf, mac);
7900}
7901
7902static int igb_link_mbps(int internal_link_speed)
7903{
7904 switch (internal_link_speed) {
7905 case SPEED_100:
7906 return 100;
7907 case SPEED_1000:
7908 return 1000;
7909 default:
7910 return 0;
7911 }
7912}
7913
7914static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
7915 int link_speed)
7916{
7917 int rf_dec, rf_int;
7918 u32 bcnrc_val;
7919
7920 if (tx_rate != 0) {
7921 /* Calculate the rate factor values to set */
7922 rf_int = link_speed / tx_rate;
7923 rf_dec = (link_speed - (rf_int * tx_rate));
7924 rf_dec = (rf_dec * (1 << E1000_RTTBCNRC_RF_INT_SHIFT)) /
7925 tx_rate;
7926
7927 bcnrc_val = E1000_RTTBCNRC_RS_ENA;
7928 bcnrc_val |= ((rf_int << E1000_RTTBCNRC_RF_INT_SHIFT) &
7929 E1000_RTTBCNRC_RF_INT_MASK);
7930 bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
7931 } else {
7932 bcnrc_val = 0;
7933 }
7934
7935 wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
7936 /* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
7937 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
7938 */
7939 wr32(E1000_RTTBCNRM, 0x14);
7940 wr32(E1000_RTTBCNRC, bcnrc_val);
7941}
7942
7943static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
7944{
7945 int actual_link_speed, i;
7946 bool reset_rate = false;
7947
7948 /* VF TX rate limit was not set or not supported */
7949 if ((adapter->vf_rate_link_speed == 0) ||
7950 (adapter->hw.mac.type != e1000_82576))
7951 return;
7952
7953 actual_link_speed = igb_link_mbps(adapter->link_speed);
7954 if (actual_link_speed != adapter->vf_rate_link_speed) {
7955 reset_rate = true;
7956 adapter->vf_rate_link_speed = 0;
7957 dev_info(&adapter->pdev->dev,
7958 "Link speed has been changed. VF Transmit rate is disabled\n");
7959 }
7960
7961 for (i = 0; i < adapter->vfs_allocated_count; i++) {
7962 if (reset_rate)
7963 adapter->vf_data[i].tx_rate = 0;
7964
7965 igb_set_vf_rate_limit(&adapter->hw, i,
7966 adapter->vf_data[i].tx_rate,
7967 actual_link_speed);
7968 }
7969}
7970
7971static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf,
7972 int min_tx_rate, int max_tx_rate)
7973{
7974 struct igb_adapter *adapter = netdev_priv(netdev);
7975 struct e1000_hw *hw = &adapter->hw;
7976 int actual_link_speed;
7977
7978 if (hw->mac.type != e1000_82576)
7979 return -EOPNOTSUPP;
7980
7981 if (min_tx_rate)
7982 return -EINVAL;
7983
7984 actual_link_speed = igb_link_mbps(adapter->link_speed);
7985 if ((vf >= adapter->vfs_allocated_count) ||
7986 (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
7987 (max_tx_rate < 0) ||
7988 (max_tx_rate > actual_link_speed))
7989 return -EINVAL;
7990
7991 adapter->vf_rate_link_speed = actual_link_speed;
7992 adapter->vf_data[vf].tx_rate = (u16)max_tx_rate;
7993 igb_set_vf_rate_limit(hw, vf, max_tx_rate, actual_link_speed);
7994
7995 return 0;
7996}
7997
7998static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
7999 bool setting)
8000{
8001 struct igb_adapter *adapter = netdev_priv(netdev);
8002 struct e1000_hw *hw = &adapter->hw;
8003 u32 reg_val, reg_offset;
8004
8005 if (!adapter->vfs_allocated_count)
8006 return -EOPNOTSUPP;
8007
8008 if (vf >= adapter->vfs_allocated_count)
8009 return -EINVAL;
8010
8011 reg_offset = (hw->mac.type == e1000_82576) ? E1000_DTXSWC : E1000_TXSWC;
8012 reg_val = rd32(reg_offset);
8013 if (setting)
8014 reg_val |= ((1 << vf) |
8015 (1 << (vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT)));
8016 else
8017 reg_val &= ~((1 << vf) |
8018 (1 << (vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT)));
8019 wr32(reg_offset, reg_val);
8020
8021 adapter->vf_data[vf].spoofchk_enabled = setting;
8022 return 0;
8023}
8024
8025static int igb_ndo_get_vf_config(struct net_device *netdev,
8026 int vf, struct ifla_vf_info *ivi)
8027{
8028 struct igb_adapter *adapter = netdev_priv(netdev);
8029 if (vf >= adapter->vfs_allocated_count)
8030 return -EINVAL;
8031 ivi->vf = vf;
8032 memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
8033 ivi->max_tx_rate = adapter->vf_data[vf].tx_rate;
8034 ivi->min_tx_rate = 0;
8035 ivi->vlan = adapter->vf_data[vf].pf_vlan;
8036 ivi->qos = adapter->vf_data[vf].pf_qos;
8037 ivi->spoofchk = adapter->vf_data[vf].spoofchk_enabled;
8038 return 0;
8039}
8040
8041static void igb_vmm_control(struct igb_adapter *adapter)
8042{
8043 struct e1000_hw *hw = &adapter->hw;
8044 u32 reg;
8045
8046 switch (hw->mac.type) {
8047 case e1000_82575:
8048 case e1000_i210:
8049 case e1000_i211:
8050 case e1000_i354:
8051 default:
8052 /* replication is not supported for 82575 */
8053 return;
8054 case e1000_82576:
8055 /* notify HW that the MAC is adding vlan tags */
8056 reg = rd32(E1000_DTXCTL);
8057 reg |= E1000_DTXCTL_VLAN_ADDED;
8058 wr32(E1000_DTXCTL, reg);
8059 /* Fall through */
8060 case e1000_82580:
8061 /* enable replication vlan tag stripping */
8062 reg = rd32(E1000_RPLOLR);
8063 reg |= E1000_RPLOLR_STRVLAN;
8064 wr32(E1000_RPLOLR, reg);
8065 /* Fall through */
8066 case e1000_i350:
8067 /* none of the above registers are supported by i350 */
8068 break;
8069 }
8070
8071 if (adapter->vfs_allocated_count) {
8072 igb_vmdq_set_loopback_pf(hw, true);
8073 igb_vmdq_set_replication_pf(hw, true);
8074 igb_vmdq_set_anti_spoofing_pf(hw, true,
8075 adapter->vfs_allocated_count);
8076 } else {
8077 igb_vmdq_set_loopback_pf(hw, false);
8078 igb_vmdq_set_replication_pf(hw, false);
8079 }
8080}
8081
8082static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
8083{
8084 struct e1000_hw *hw = &adapter->hw;
8085 u32 dmac_thr;
8086 u16 hwm;
8087
8088 if (hw->mac.type > e1000_82580) {
8089 if (adapter->flags & IGB_FLAG_DMAC) {
8090 u32 reg;
8091
8092 /* force threshold to 0. */
8093 wr32(E1000_DMCTXTH, 0);
8094
8095 /* DMA Coalescing high water mark needs to be greater
8096 * than the Rx threshold. Set hwm to PBA - max frame
8097 * size in 16B units, capping it at PBA - 6KB.
8098 */
8099 hwm = 64 * (pba - 6);
8100 reg = rd32(E1000_FCRTC);
8101 reg &= ~E1000_FCRTC_RTH_COAL_MASK;
8102 reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
8103 & E1000_FCRTC_RTH_COAL_MASK);
8104 wr32(E1000_FCRTC, reg);
8105
8106 /* Set the DMA Coalescing Rx threshold to PBA - 2 * max
8107 * frame size, capping it at PBA - 10KB.
8108 */
8109 dmac_thr = pba - 10;
8110 reg = rd32(E1000_DMACR);
8111 reg &= ~E1000_DMACR_DMACTHR_MASK;
8112 reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
8113 & E1000_DMACR_DMACTHR_MASK);
8114
8115 /* transition to L0x or L1 if available..*/
8116 reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
8117
8118 /* watchdog timer= +-1000 usec in 32usec intervals */
8119 reg |= (1000 >> 5);
8120
8121 /* Disable BMC-to-OS Watchdog Enable */
8122 if (hw->mac.type != e1000_i354)
8123 reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
8124
8125 wr32(E1000_DMACR, reg);
8126
8127 /* no lower threshold to disable
8128 * coalescing(smart fifb)-UTRESH=0
8129 */
8130 wr32(E1000_DMCRTRH, 0);
8131
8132 reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
8133
8134 wr32(E1000_DMCTLX, reg);
8135
8136 /* free space in tx packet buffer to wake from
8137 * DMA coal
8138 */
8139 wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
8140 (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
8141
8142 /* make low power state decision controlled
8143 * by DMA coal
8144 */
8145 reg = rd32(E1000_PCIEMISC);
8146 reg &= ~E1000_PCIEMISC_LX_DECISION;
8147 wr32(E1000_PCIEMISC, reg);
8148 } /* endif adapter->dmac is not disabled */
8149 } else if (hw->mac.type == e1000_82580) {
8150 u32 reg = rd32(E1000_PCIEMISC);
8151
8152 wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
8153 wr32(E1000_DMACR, 0);
8154 }
8155}
8156
8157/**
8158 * igb_read_i2c_byte - Reads 8 bit word over I2C
8159 * @hw: pointer to hardware structure
8160 * @byte_offset: byte offset to read
8161 * @dev_addr: device address
8162 * @data: value read
8163 *
8164 * Performs byte read operation over I2C interface at
8165 * a specified device address.
8166 **/
8167s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
8168 u8 dev_addr, u8 *data)
8169{
8170 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
8171 struct i2c_client *this_client = adapter->i2c_client;
8172 s32 status;
8173 u16 swfw_mask = 0;
8174
8175 if (!this_client)
8176 return E1000_ERR_I2C;
8177
8178 swfw_mask = E1000_SWFW_PHY0_SM;
8179
8180 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
8181 return E1000_ERR_SWFW_SYNC;
8182
8183 status = i2c_smbus_read_byte_data(this_client, byte_offset);
8184 hw->mac.ops.release_swfw_sync(hw, swfw_mask);
8185
8186 if (status < 0)
8187 return E1000_ERR_I2C;
8188 else {
8189 *data = status;
8190 return 0;
8191 }
8192}
8193
8194/**
8195 * igb_write_i2c_byte - Writes 8 bit word over I2C
8196 * @hw: pointer to hardware structure
8197 * @byte_offset: byte offset to write
8198 * @dev_addr: device address
8199 * @data: value to write
8200 *
8201 * Performs byte write operation over I2C interface at
8202 * a specified device address.
8203 **/
8204s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
8205 u8 dev_addr, u8 data)
8206{
8207 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
8208 struct i2c_client *this_client = adapter->i2c_client;
8209 s32 status;
8210 u16 swfw_mask = E1000_SWFW_PHY0_SM;
8211
8212 if (!this_client)
8213 return E1000_ERR_I2C;
8214
8215 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
8216 return E1000_ERR_SWFW_SYNC;
8217 status = i2c_smbus_write_byte_data(this_client, byte_offset, data);
8218 hw->mac.ops.release_swfw_sync(hw, swfw_mask);
8219
8220 if (status)
8221 return E1000_ERR_I2C;
8222 else
8223 return 0;
8224
8225}
8226
8227int igb_reinit_queues(struct igb_adapter *adapter)
8228{
8229 struct net_device *netdev = adapter->netdev;
8230 struct pci_dev *pdev = adapter->pdev;
8231 int err = 0;
8232
8233 if (netif_running(netdev))
8234 igb_close(netdev);
8235
8236 igb_reset_interrupt_capability(adapter);
8237
8238 if (igb_init_interrupt_scheme(adapter, true)) {
8239 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
8240 return -ENOMEM;
8241 }
8242
8243 if (netif_running(netdev))
8244 err = igb_open(netdev);
8245
8246 return err;
8247}
8248/* igb_main.c */