1// SPDX-License-Identifier: GPL-2.0
   2/* Copyright(c) 2009 - 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/pci.h>
  10#include <linux/vmalloc.h>
  11#include <linux/pagemap.h>
  12#include <linux/delay.h>
  13#include <linux/netdevice.h>
  14#include <linux/tcp.h>
  15#include <linux/ipv6.h>
  16#include <linux/slab.h>
  17#include <net/checksum.h>
  18#include <net/ip6_checksum.h>
  19#include <linux/mii.h>
  20#include <linux/ethtool.h>
  21#include <linux/if_vlan.h>
  22#include <linux/prefetch.h>
  23#include <linux/sctp.h>
  24
  25#include "igbvf.h"
  26
  27char igbvf_driver_name[] = "igbvf";
  28static const char igbvf_driver_string[] =
  29		  "Intel(R) Gigabit Virtual Function Network Driver";
  30static const char igbvf_copyright[] =
  31		  "Copyright (c) 2009 - 2012 Intel Corporation.";
  32
  33#define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
  34static int debug = -1;
  35module_param(debug, int, 0);
  36MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  37
  38static int igbvf_poll(struct napi_struct *napi, int budget);
  39static void igbvf_reset(struct igbvf_adapter *);
  40static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
  41static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
  42
  43static struct igbvf_info igbvf_vf_info = {
  44	.mac		= e1000_vfadapt,
  45	.flags		= 0,
  46	.pba		= 10,
  47	.init_ops	= e1000_init_function_pointers_vf,
  48};
  49
  50static struct igbvf_info igbvf_i350_vf_info = {
  51	.mac		= e1000_vfadapt_i350,
  52	.flags		= 0,
  53	.pba		= 10,
  54	.init_ops	= e1000_init_function_pointers_vf,
  55};
  56
  57static const struct igbvf_info *igbvf_info_tbl[] = {
  58	[board_vf]	= &igbvf_vf_info,
  59	[board_i350_vf]	= &igbvf_i350_vf_info,
  60};
  61
  62/**
  63 * igbvf_desc_unused - calculate if we have unused descriptors
  64 * @ring: address of receive ring structure
  65 **/
  66static int igbvf_desc_unused(struct igbvf_ring *ring)
  67{
  68	if (ring->next_to_clean > ring->next_to_use)
  69		return ring->next_to_clean - ring->next_to_use - 1;
  70
  71	return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  72}
  73
  74/**
  75 * igbvf_receive_skb - helper function to handle Rx indications
  76 * @adapter: board private structure
  77 * @netdev: pointer to netdev struct
  78 * @skb: skb to indicate to stack
  79 * @status: descriptor status field as written by hardware
  80 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  81 * @skb: pointer to sk_buff to be indicated to stack
  82 **/
  83static void igbvf_receive_skb(struct igbvf_adapter *adapter,
  84			      struct net_device *netdev,
  85			      struct sk_buff *skb,
  86			      u32 status, __le16 vlan)
  87{
  88	u16 vid;
  89
  90	if (status & E1000_RXD_STAT_VP) {
  91		if ((adapter->flags & IGBVF_FLAG_RX_LB_VLAN_BSWAP) &&
  92		    (status & E1000_RXDEXT_STATERR_LB))
  93			vid = be16_to_cpu((__force __be16)vlan) & E1000_RXD_SPC_VLAN_MASK;
  94		else
  95			vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
  96		if (test_bit(vid, adapter->active_vlans))
  97			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
  98	}
  99
 100	napi_gro_receive(&adapter->rx_ring->napi, skb);
 101}
 102
 103static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
 104					 u32 status_err, struct sk_buff *skb)
 105{
 106	skb_checksum_none_assert(skb);
 107
 108	/* Ignore Checksum bit is set or checksum is disabled through ethtool */
 109	if ((status_err & E1000_RXD_STAT_IXSM) ||
 110	    (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
 111		return;
 112
 113	/* TCP/UDP checksum error bit is set */
 114	if (status_err &
 115	    (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
 116		/* let the stack verify checksum errors */
 117		adapter->hw_csum_err++;
 118		return;
 119	}
 120
 121	/* It must be a TCP or UDP packet with a valid checksum */
 122	if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
 123		skb->ip_summed = CHECKSUM_UNNECESSARY;
 124
 125	adapter->hw_csum_good++;
 126}
 127
 128/**
 129 * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
 130 * @rx_ring: address of ring structure to repopulate
 131 * @cleaned_count: number of buffers to repopulate
 132 **/
 133static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
 134				   int cleaned_count)
 135{
 136	struct igbvf_adapter *adapter = rx_ring->adapter;
 137	struct net_device *netdev = adapter->netdev;
 138	struct pci_dev *pdev = adapter->pdev;
 139	union e1000_adv_rx_desc *rx_desc;
 140	struct igbvf_buffer *buffer_info;
 141	struct sk_buff *skb;
 142	unsigned int i;
 143	int bufsz;
 144
 145	i = rx_ring->next_to_use;
 146	buffer_info = &rx_ring->buffer_info[i];
 147
 148	if (adapter->rx_ps_hdr_size)
 149		bufsz = adapter->rx_ps_hdr_size;
 150	else
 151		bufsz = adapter->rx_buffer_len;
 152
 153	while (cleaned_count--) {
 154		rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
 155
 156		if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
 157			if (!buffer_info->page) {
 158				buffer_info->page = alloc_page(GFP_ATOMIC);
 159				if (!buffer_info->page) {
 160					adapter->alloc_rx_buff_failed++;
 161					goto no_buffers;
 162				}
 163				buffer_info->page_offset = 0;
 164			} else {
 165				buffer_info->page_offset ^= PAGE_SIZE / 2;
 166			}
 167			buffer_info->page_dma =
 168				dma_map_page(&pdev->dev, buffer_info->page,
 169					     buffer_info->page_offset,
 170					     PAGE_SIZE / 2,
 171					     DMA_FROM_DEVICE);
 172			if (dma_mapping_error(&pdev->dev,
 173					      buffer_info->page_dma)) {
 174				__free_page(buffer_info->page);
 175				buffer_info->page = NULL;
 176				dev_err(&pdev->dev, "RX DMA map failed\n");
 177				break;
 178			}
 179		}
 180
 181		if (!buffer_info->skb) {
 182			skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 183			if (!skb) {
 184				adapter->alloc_rx_buff_failed++;
 185				goto no_buffers;
 186			}
 187
 188			buffer_info->skb = skb;
 189			buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 190							  bufsz,
 191							  DMA_FROM_DEVICE);
 192			if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 193				dev_kfree_skb(buffer_info->skb);
 194				buffer_info->skb = NULL;
 195				dev_err(&pdev->dev, "RX DMA map failed\n");
 196				goto no_buffers;
 197			}
 198		}
 199		/* Refresh the desc even if buffer_addrs didn't change because
 200		 * each write-back erases this info.
 201		 */
 202		if (adapter->rx_ps_hdr_size) {
 203			rx_desc->read.pkt_addr =
 204			     cpu_to_le64(buffer_info->page_dma);
 205			rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
 206		} else {
 207			rx_desc->read.pkt_addr = cpu_to_le64(buffer_info->dma);
 208			rx_desc->read.hdr_addr = 0;
 209		}
 210
 211		i++;
 212		if (i == rx_ring->count)
 213			i = 0;
 214		buffer_info = &rx_ring->buffer_info[i];
 215	}
 216
 217no_buffers:
 218	if (rx_ring->next_to_use != i) {
 219		rx_ring->next_to_use = i;
 220		if (i == 0)
 221			i = (rx_ring->count - 1);
 222		else
 223			i--;
 224
 225		/* Force memory writes to complete before letting h/w
 226		 * know there are new descriptors to fetch.  (Only
 227		 * applicable for weak-ordered memory model archs,
 228		 * such as IA-64).
 229		*/
 230		wmb();
 231		writel(i, adapter->hw.hw_addr + rx_ring->tail);
 232	}
 233}
 234
 235/**
 236 * igbvf_clean_rx_irq - Send received data up the network stack; legacy
 237 * @adapter: board private structure
 238 * @work_done: output parameter used to indicate completed work
 239 * @work_to_do: input parameter setting limit of work
 240 *
 241 * the return value indicates whether actual cleaning was done, there
 242 * is no guarantee that everything was cleaned
 243 **/
 244static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
 245			       int *work_done, int work_to_do)
 246{
 247	struct igbvf_ring *rx_ring = adapter->rx_ring;
 248	struct net_device *netdev = adapter->netdev;
 249	struct pci_dev *pdev = adapter->pdev;
 250	union e1000_adv_rx_desc *rx_desc, *next_rxd;
 251	struct igbvf_buffer *buffer_info, *next_buffer;
 252	struct sk_buff *skb;
 253	bool cleaned = false;
 254	int cleaned_count = 0;
 255	unsigned int total_bytes = 0, total_packets = 0;
 256	unsigned int i;
 257	u32 length, hlen, staterr;
 258
 259	i = rx_ring->next_to_clean;
 260	rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
 261	staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 262
 263	while (staterr & E1000_RXD_STAT_DD) {
 264		if (*work_done >= work_to_do)
 265			break;
 266		(*work_done)++;
 267		rmb(); /* read descriptor and rx_buffer_info after status DD */
 268
 269		buffer_info = &rx_ring->buffer_info[i];
 270
 271		/* HW will not DMA in data larger than the given buffer, even
 272		 * if it parses the (NFS, of course) header to be larger.  In
 273		 * that case, it fills the header buffer and spills the rest
 274		 * into the page.
 275		 */
 276		hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info)
 277		       & E1000_RXDADV_HDRBUFLEN_MASK) >>
 278		       E1000_RXDADV_HDRBUFLEN_SHIFT;
 279		if (hlen > adapter->rx_ps_hdr_size)
 280			hlen = adapter->rx_ps_hdr_size;
 281
 282		length = le16_to_cpu(rx_desc->wb.upper.length);
 283		cleaned = true;
 284		cleaned_count++;
 285
 286		skb = buffer_info->skb;
 287		prefetch(skb->data - NET_IP_ALIGN);
 288		buffer_info->skb = NULL;
 289		if (!adapter->rx_ps_hdr_size) {
 290			dma_unmap_single(&pdev->dev, buffer_info->dma,
 291					 adapter->rx_buffer_len,
 292					 DMA_FROM_DEVICE);
 293			buffer_info->dma = 0;
 294			skb_put(skb, length);
 295			goto send_up;
 296		}
 297
 298		if (!skb_shinfo(skb)->nr_frags) {
 299			dma_unmap_single(&pdev->dev, buffer_info->dma,
 300					 adapter->rx_ps_hdr_size,
 301					 DMA_FROM_DEVICE);
 302			buffer_info->dma = 0;
 303			skb_put(skb, hlen);
 304		}
 305
 306		if (length) {
 307			dma_unmap_page(&pdev->dev, buffer_info->page_dma,
 308				       PAGE_SIZE / 2,
 309				       DMA_FROM_DEVICE);
 310			buffer_info->page_dma = 0;
 311
 312			skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
 313					   buffer_info->page,
 314					   buffer_info->page_offset,
 315					   length);
 316
 317			if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
 318			    (page_count(buffer_info->page) != 1))
 319				buffer_info->page = NULL;
 320			else
 321				get_page(buffer_info->page);
 322
 323			skb->len += length;
 324			skb->data_len += length;
 325			skb->truesize += PAGE_SIZE / 2;
 326		}
 327send_up:
 328		i++;
 329		if (i == rx_ring->count)
 330			i = 0;
 331		next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
 332		prefetch(next_rxd);
 333		next_buffer = &rx_ring->buffer_info[i];
 334
 335		if (!(staterr & E1000_RXD_STAT_EOP)) {
 336			buffer_info->skb = next_buffer->skb;
 337			buffer_info->dma = next_buffer->dma;
 338			next_buffer->skb = skb;
 339			next_buffer->dma = 0;
 340			goto next_desc;
 341		}
 342
 343		if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 344			dev_kfree_skb_irq(skb);
 345			goto next_desc;
 346		}
 347
 348		total_bytes += skb->len;
 349		total_packets++;
 350
 351		igbvf_rx_checksum_adv(adapter, staterr, skb);
 352
 353		skb->protocol = eth_type_trans(skb, netdev);
 354
 355		igbvf_receive_skb(adapter, netdev, skb, staterr,
 356				  rx_desc->wb.upper.vlan);
 357
 358next_desc:
 359		rx_desc->wb.upper.status_error = 0;
 360
 361		/* return some buffers to hardware, one at a time is too slow */
 362		if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
 363			igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 364			cleaned_count = 0;
 365		}
 366
 367		/* use prefetched values */
 368		rx_desc = next_rxd;
 369		buffer_info = next_buffer;
 370
 371		staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 372	}
 373
 374	rx_ring->next_to_clean = i;
 375	cleaned_count = igbvf_desc_unused(rx_ring);
 376
 377	if (cleaned_count)
 378		igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 379
 380	adapter->total_rx_packets += total_packets;
 381	adapter->total_rx_bytes += total_bytes;
 382	netdev->stats.rx_bytes += total_bytes;
 383	netdev->stats.rx_packets += total_packets;
 384	return cleaned;
 385}
 386
 387static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
 388			    struct igbvf_buffer *buffer_info)
 389{
 390	if (buffer_info->dma) {
 391		if (buffer_info->mapped_as_page)
 392			dma_unmap_page(&adapter->pdev->dev,
 393				       buffer_info->dma,
 394				       buffer_info->length,
 395				       DMA_TO_DEVICE);
 396		else
 397			dma_unmap_single(&adapter->pdev->dev,
 398					 buffer_info->dma,
 399					 buffer_info->length,
 400					 DMA_TO_DEVICE);
 401		buffer_info->dma = 0;
 402	}
 403	if (buffer_info->skb) {
 404		dev_kfree_skb_any(buffer_info->skb);
 405		buffer_info->skb = NULL;
 406	}
 407	buffer_info->time_stamp = 0;
 408}
 409
 410/**
 411 * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
 412 * @adapter: board private structure
 413 * @tx_ring: ring being initialized
 414 *
 415 * Return 0 on success, negative on failure
 416 **/
 417int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
 418			     struct igbvf_ring *tx_ring)
 419{
 420	struct pci_dev *pdev = adapter->pdev;
 421	int size;
 422
 423	size = sizeof(struct igbvf_buffer) * tx_ring->count;
 424	tx_ring->buffer_info = vzalloc(size);
 425	if (!tx_ring->buffer_info)
 426		goto err;
 427
 428	/* round up to nearest 4K */
 429	tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
 430	tx_ring->size = ALIGN(tx_ring->size, 4096);
 431
 432	tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
 433					   &tx_ring->dma, GFP_KERNEL);
 434	if (!tx_ring->desc)
 435		goto err;
 436
 437	tx_ring->adapter = adapter;
 438	tx_ring->next_to_use = 0;
 439	tx_ring->next_to_clean = 0;
 440
 441	return 0;
 442err:
 443	vfree(tx_ring->buffer_info);
 444	dev_err(&adapter->pdev->dev,
 445		"Unable to allocate memory for the transmit descriptor ring\n");
 446	return -ENOMEM;
 447}
 448
 449/**
 450 * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
 451 * @adapter: board private structure
 452 * @rx_ring: ring being initialized
 453 *
 454 * Returns 0 on success, negative on failure
 455 **/
 456int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
 457			     struct igbvf_ring *rx_ring)
 458{
 459	struct pci_dev *pdev = adapter->pdev;
 460	int size, desc_len;
 461
 462	size = sizeof(struct igbvf_buffer) * rx_ring->count;
 463	rx_ring->buffer_info = vzalloc(size);
 464	if (!rx_ring->buffer_info)
 465		goto err;
 466
 467	desc_len = sizeof(union e1000_adv_rx_desc);
 468
 469	/* Round up to nearest 4K */
 470	rx_ring->size = rx_ring->count * desc_len;
 471	rx_ring->size = ALIGN(rx_ring->size, 4096);
 472
 473	rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
 474					   &rx_ring->dma, GFP_KERNEL);
 475	if (!rx_ring->desc)
 476		goto err;
 477
 478	rx_ring->next_to_clean = 0;
 479	rx_ring->next_to_use = 0;
 480
 481	rx_ring->adapter = adapter;
 482
 483	return 0;
 484
 485err:
 486	vfree(rx_ring->buffer_info);
 487	rx_ring->buffer_info = NULL;
 488	dev_err(&adapter->pdev->dev,
 489		"Unable to allocate memory for the receive descriptor ring\n");
 490	return -ENOMEM;
 491}
 492
 493/**
 494 * igbvf_clean_tx_ring - Free Tx Buffers
 495 * @tx_ring: ring to be cleaned
 496 **/
 497static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
 498{
 499	struct igbvf_adapter *adapter = tx_ring->adapter;
 500	struct igbvf_buffer *buffer_info;
 501	unsigned long size;
 502	unsigned int i;
 503
 504	if (!tx_ring->buffer_info)
 505		return;
 506
 507	/* Free all the Tx ring sk_buffs */
 508	for (i = 0; i < tx_ring->count; i++) {
 509		buffer_info = &tx_ring->buffer_info[i];
 510		igbvf_put_txbuf(adapter, buffer_info);
 511	}
 512
 513	size = sizeof(struct igbvf_buffer) * tx_ring->count;
 514	memset(tx_ring->buffer_info, 0, size);
 515
 516	/* Zero out the descriptor ring */
 517	memset(tx_ring->desc, 0, tx_ring->size);
 518
 519	tx_ring->next_to_use = 0;
 520	tx_ring->next_to_clean = 0;
 521
 522	writel(0, adapter->hw.hw_addr + tx_ring->head);
 523	writel(0, adapter->hw.hw_addr + tx_ring->tail);
 524}
 525
 526/**
 527 * igbvf_free_tx_resources - Free Tx Resources per Queue
 528 * @tx_ring: ring to free resources from
 529 *
 530 * Free all transmit software resources
 531 **/
 532void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
 533{
 534	struct pci_dev *pdev = tx_ring->adapter->pdev;
 535
 536	igbvf_clean_tx_ring(tx_ring);
 537
 538	vfree(tx_ring->buffer_info);
 539	tx_ring->buffer_info = NULL;
 540
 541	dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
 542			  tx_ring->dma);
 543
 544	tx_ring->desc = NULL;
 545}
 546
 547/**
 548 * igbvf_clean_rx_ring - Free Rx Buffers per Queue
 549 * @rx_ring: ring structure pointer to free buffers from
 550 **/
 551static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
 552{
 553	struct igbvf_adapter *adapter = rx_ring->adapter;
 554	struct igbvf_buffer *buffer_info;
 555	struct pci_dev *pdev = adapter->pdev;
 556	unsigned long size;
 557	unsigned int i;
 558
 559	if (!rx_ring->buffer_info)
 560		return;
 561
 562	/* Free all the Rx ring sk_buffs */
 563	for (i = 0; i < rx_ring->count; i++) {
 564		buffer_info = &rx_ring->buffer_info[i];
 565		if (buffer_info->dma) {
 566			if (adapter->rx_ps_hdr_size) {
 567				dma_unmap_single(&pdev->dev, buffer_info->dma,
 568						 adapter->rx_ps_hdr_size,
 569						 DMA_FROM_DEVICE);
 570			} else {
 571				dma_unmap_single(&pdev->dev, buffer_info->dma,
 572						 adapter->rx_buffer_len,
 573						 DMA_FROM_DEVICE);
 574			}
 575			buffer_info->dma = 0;
 576		}
 577
 578		if (buffer_info->skb) {
 579			dev_kfree_skb(buffer_info->skb);
 580			buffer_info->skb = NULL;
 581		}
 582
 583		if (buffer_info->page) {
 584			if (buffer_info->page_dma)
 585				dma_unmap_page(&pdev->dev,
 586					       buffer_info->page_dma,
 587					       PAGE_SIZE / 2,
 588					       DMA_FROM_DEVICE);
 589			put_page(buffer_info->page);
 590			buffer_info->page = NULL;
 591			buffer_info->page_dma = 0;
 592			buffer_info->page_offset = 0;
 593		}
 594	}
 595
 596	size = sizeof(struct igbvf_buffer) * rx_ring->count;
 597	memset(rx_ring->buffer_info, 0, size);
 598
 599	/* Zero out the descriptor ring */
 600	memset(rx_ring->desc, 0, rx_ring->size);
 601
 602	rx_ring->next_to_clean = 0;
 603	rx_ring->next_to_use = 0;
 604
 605	writel(0, adapter->hw.hw_addr + rx_ring->head);
 606	writel(0, adapter->hw.hw_addr + rx_ring->tail);
 607}
 608
 609/**
 610 * igbvf_free_rx_resources - Free Rx Resources
 611 * @rx_ring: ring to clean the resources from
 612 *
 613 * Free all receive software resources
 614 **/
 615
 616void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
 617{
 618	struct pci_dev *pdev = rx_ring->adapter->pdev;
 619
 620	igbvf_clean_rx_ring(rx_ring);
 621
 622	vfree(rx_ring->buffer_info);
 623	rx_ring->buffer_info = NULL;
 624
 625	dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
 626			  rx_ring->dma);
 627	rx_ring->desc = NULL;
 628}
 629
 630/**
 631 * igbvf_update_itr - update the dynamic ITR value based on statistics
 632 * @adapter: pointer to adapter
 633 * @itr_setting: current adapter->itr
 634 * @packets: the number of packets during this measurement interval
 635 * @bytes: the number of bytes during this measurement interval
 636 *
 637 * Stores a new ITR value based on packets and byte counts during the last
 638 * interrupt.  The advantage of per interrupt computation is faster updates
 639 * and more accurate ITR for the current traffic pattern.  Constants in this
 640 * function were computed based on theoretical maximum wire speed and thresholds
 641 * were set based on testing data as well as attempting to minimize response
 642 * time while increasing bulk throughput.
 643 **/
 644static enum latency_range igbvf_update_itr(struct igbvf_adapter *adapter,
 645					   enum latency_range itr_setting,
 646					   int packets, int bytes)
 647{
 648	enum latency_range retval = itr_setting;
 649
 650	if (packets == 0)
 651		goto update_itr_done;
 652
 653	switch (itr_setting) {
 654	case lowest_latency:
 655		/* handle TSO and jumbo frames */
 656		if (bytes/packets > 8000)
 657			retval = bulk_latency;
 658		else if ((packets < 5) && (bytes > 512))
 659			retval = low_latency;
 660		break;
 661	case low_latency:  /* 50 usec aka 20000 ints/s */
 662		if (bytes > 10000) {
 663			/* this if handles the TSO accounting */
 664			if (bytes/packets > 8000)
 665				retval = bulk_latency;
 666			else if ((packets < 10) || ((bytes/packets) > 1200))
 667				retval = bulk_latency;
 668			else if ((packets > 35))
 669				retval = lowest_latency;
 670		} else if (bytes/packets > 2000) {
 671			retval = bulk_latency;
 672		} else if (packets <= 2 && bytes < 512) {
 673			retval = lowest_latency;
 674		}
 675		break;
 676	case bulk_latency: /* 250 usec aka 4000 ints/s */
 677		if (bytes > 25000) {
 678			if (packets > 35)
 679				retval = low_latency;
 680		} else if (bytes < 6000) {
 681			retval = low_latency;
 682		}
 683		break;
 684	default:
 685		break;
 686	}
 687
 688update_itr_done:
 689	return retval;
 690}
 691
 692static int igbvf_range_to_itr(enum latency_range current_range)
 693{
 694	int new_itr;
 695
 696	switch (current_range) {
 697	/* counts and packets in update_itr are dependent on these numbers */
 698	case lowest_latency:
 699		new_itr = IGBVF_70K_ITR;
 700		break;
 701	case low_latency:
 702		new_itr = IGBVF_20K_ITR;
 703		break;
 704	case bulk_latency:
 705		new_itr = IGBVF_4K_ITR;
 706		break;
 707	default:
 708		new_itr = IGBVF_START_ITR;
 709		break;
 710	}
 711	return new_itr;
 712}
 713
 714static void igbvf_set_itr(struct igbvf_adapter *adapter)
 715{
 716	u32 new_itr;
 717
 718	adapter->tx_ring->itr_range =
 719			igbvf_update_itr(adapter,
 720					 adapter->tx_ring->itr_val,
 721					 adapter->total_tx_packets,
 722					 adapter->total_tx_bytes);
 723
 724	/* conservative mode (itr 3) eliminates the lowest_latency setting */
 725	if (adapter->requested_itr == 3 &&
 726	    adapter->tx_ring->itr_range == lowest_latency)
 727		adapter->tx_ring->itr_range = low_latency;
 728
 729	new_itr = igbvf_range_to_itr(adapter->tx_ring->itr_range);
 730
 731	if (new_itr != adapter->tx_ring->itr_val) {
 732		u32 current_itr = adapter->tx_ring->itr_val;
 733		/* this attempts to bias the interrupt rate towards Bulk
 734		 * by adding intermediate steps when interrupt rate is
 735		 * increasing
 736		 */
 737		new_itr = new_itr > current_itr ?
 738			  min(current_itr + (new_itr >> 2), new_itr) :
 739			  new_itr;
 740		adapter->tx_ring->itr_val = new_itr;
 741
 742		adapter->tx_ring->set_itr = 1;
 743	}
 744
 745	adapter->rx_ring->itr_range =
 746			igbvf_update_itr(adapter, adapter->rx_ring->itr_val,
 747					 adapter->total_rx_packets,
 748					 adapter->total_rx_bytes);
 749	if (adapter->requested_itr == 3 &&
 750	    adapter->rx_ring->itr_range == lowest_latency)
 751		adapter->rx_ring->itr_range = low_latency;
 752
 753	new_itr = igbvf_range_to_itr(adapter->rx_ring->itr_range);
 754
 755	if (new_itr != adapter->rx_ring->itr_val) {
 756		u32 current_itr = adapter->rx_ring->itr_val;
 757
 758		new_itr = new_itr > current_itr ?
 759			  min(current_itr + (new_itr >> 2), new_itr) :
 760			  new_itr;
 761		adapter->rx_ring->itr_val = new_itr;
 762
 763		adapter->rx_ring->set_itr = 1;
 764	}
 765}
 766
 767/**
 768 * igbvf_clean_tx_irq - Reclaim resources after transmit completes
 769 * @tx_ring: ring structure to clean descriptors from
 770 *
 771 * returns true if ring is completely cleaned
 772 **/
 773static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
 774{
 775	struct igbvf_adapter *adapter = tx_ring->adapter;
 776	struct net_device *netdev = adapter->netdev;
 777	struct igbvf_buffer *buffer_info;
 778	struct sk_buff *skb;
 779	union e1000_adv_tx_desc *tx_desc, *eop_desc;
 780	unsigned int total_bytes = 0, total_packets = 0;
 781	unsigned int i, count = 0;
 782	bool cleaned = false;
 783
 784	i = tx_ring->next_to_clean;
 785	buffer_info = &tx_ring->buffer_info[i];
 786	eop_desc = buffer_info->next_to_watch;
 787
 788	do {
 789		/* if next_to_watch is not set then there is no work pending */
 790		if (!eop_desc)
 791			break;
 792
 793		/* prevent any other reads prior to eop_desc */
 794		smp_rmb();
 795
 796		/* if DD is not set pending work has not been completed */
 797		if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
 798			break;
 799
 800		/* clear next_to_watch to prevent false hangs */
 801		buffer_info->next_to_watch = NULL;
 802
 803		for (cleaned = false; !cleaned; count++) {
 804			tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
 805			cleaned = (tx_desc == eop_desc);
 806			skb = buffer_info->skb;
 807
 808			if (skb) {
 809				unsigned int segs, bytecount;
 810
 811				/* gso_segs is currently only valid for tcp */
 812				segs = skb_shinfo(skb)->gso_segs ?: 1;
 813				/* multiply data chunks by size of headers */
 814				bytecount = ((segs - 1) * skb_headlen(skb)) +
 815					    skb->len;
 816				total_packets += segs;
 817				total_bytes += bytecount;
 818			}
 819
 820			igbvf_put_txbuf(adapter, buffer_info);
 821			tx_desc->wb.status = 0;
 822
 823			i++;
 824			if (i == tx_ring->count)
 825				i = 0;
 826
 827			buffer_info = &tx_ring->buffer_info[i];
 828		}
 829
 830		eop_desc = buffer_info->next_to_watch;
 831	} while (count < tx_ring->count);
 832
 833	tx_ring->next_to_clean = i;
 834
 835	if (unlikely(count && netif_carrier_ok(netdev) &&
 836	    igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
 837		/* Make sure that anybody stopping the queue after this
 838		 * sees the new next_to_clean.
 839		 */
 840		smp_mb();
 841		if (netif_queue_stopped(netdev) &&
 842		    !(test_bit(__IGBVF_DOWN, &adapter->state))) {
 843			netif_wake_queue(netdev);
 844			++adapter->restart_queue;
 845		}
 846	}
 847
 848	netdev->stats.tx_bytes += total_bytes;
 849	netdev->stats.tx_packets += total_packets;
 850	return count < tx_ring->count;
 851}
 852
 853static irqreturn_t igbvf_msix_other(int irq, void *data)
 854{
 855	struct net_device *netdev = data;
 856	struct igbvf_adapter *adapter = netdev_priv(netdev);
 857	struct e1000_hw *hw = &adapter->hw;
 858
 859	adapter->int_counter1++;
 860
 861	hw->mac.get_link_status = 1;
 862	if (!test_bit(__IGBVF_DOWN, &adapter->state))
 863		mod_timer(&adapter->watchdog_timer, jiffies + 1);
 864
 865	ew32(EIMS, adapter->eims_other);
 866
 867	return IRQ_HANDLED;
 868}
 869
 870static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
 871{
 872	struct net_device *netdev = data;
 873	struct igbvf_adapter *adapter = netdev_priv(netdev);
 874	struct e1000_hw *hw = &adapter->hw;
 875	struct igbvf_ring *tx_ring = adapter->tx_ring;
 876
 877	if (tx_ring->set_itr) {
 878		writel(tx_ring->itr_val,
 879		       adapter->hw.hw_addr + tx_ring->itr_register);
 880		adapter->tx_ring->set_itr = 0;
 881	}
 882
 883	adapter->total_tx_bytes = 0;
 884	adapter->total_tx_packets = 0;
 885
 886	/* auto mask will automatically re-enable the interrupt when we write
 887	 * EICS
 888	 */
 889	if (!igbvf_clean_tx_irq(tx_ring))
 890		/* Ring was not completely cleaned, so fire another interrupt */
 891		ew32(EICS, tx_ring->eims_value);
 892	else
 893		ew32(EIMS, tx_ring->eims_value);
 894
 895	return IRQ_HANDLED;
 896}
 897
 898static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
 899{
 900	struct net_device *netdev = data;
 901	struct igbvf_adapter *adapter = netdev_priv(netdev);
 902
 903	adapter->int_counter0++;
 904
 905	/* Write the ITR value calculated at the end of the
 906	 * previous interrupt.
 907	 */
 908	if (adapter->rx_ring->set_itr) {
 909		writel(adapter->rx_ring->itr_val,
 910		       adapter->hw.hw_addr + adapter->rx_ring->itr_register);
 911		adapter->rx_ring->set_itr = 0;
 912	}
 913
 914	if (napi_schedule_prep(&adapter->rx_ring->napi)) {
 915		adapter->total_rx_bytes = 0;
 916		adapter->total_rx_packets = 0;
 917		__napi_schedule(&adapter->rx_ring->napi);
 918	}
 919
 920	return IRQ_HANDLED;
 921}
 922
 923#define IGBVF_NO_QUEUE -1
 924
 925static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
 926				int tx_queue, int msix_vector)
 927{
 928	struct e1000_hw *hw = &adapter->hw;
 929	u32 ivar, index;
 930
 931	/* 82576 uses a table-based method for assigning vectors.
 932	 * Each queue has a single entry in the table to which we write
 933	 * a vector number along with a "valid" bit.  Sadly, the layout
 934	 * of the table is somewhat counterintuitive.
 935	 */
 936	if (rx_queue > IGBVF_NO_QUEUE) {
 937		index = (rx_queue >> 1);
 938		ivar = array_er32(IVAR0, index);
 939		if (rx_queue & 0x1) {
 940			/* vector goes into third byte of register */
 941			ivar = ivar & 0xFF00FFFF;
 942			ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
 943		} else {
 944			/* vector goes into low byte of register */
 945			ivar = ivar & 0xFFFFFF00;
 946			ivar |= msix_vector | E1000_IVAR_VALID;
 947		}
 948		adapter->rx_ring[rx_queue].eims_value = BIT(msix_vector);
 949		array_ew32(IVAR0, index, ivar);
 950	}
 951	if (tx_queue > IGBVF_NO_QUEUE) {
 952		index = (tx_queue >> 1);
 953		ivar = array_er32(IVAR0, index);
 954		if (tx_queue & 0x1) {
 955			/* vector goes into high byte of register */
 956			ivar = ivar & 0x00FFFFFF;
 957			ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
 958		} else {
 959			/* vector goes into second byte of register */
 960			ivar = ivar & 0xFFFF00FF;
 961			ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
 962		}
 963		adapter->tx_ring[tx_queue].eims_value = BIT(msix_vector);
 964		array_ew32(IVAR0, index, ivar);
 965	}
 966}
 967
 968/**
 969 * igbvf_configure_msix - Configure MSI-X hardware
 970 * @adapter: board private structure
 971 *
 972 * igbvf_configure_msix sets up the hardware to properly
 973 * generate MSI-X interrupts.
 974 **/
 975static void igbvf_configure_msix(struct igbvf_adapter *adapter)
 976{
 977	u32 tmp;
 978	struct e1000_hw *hw = &adapter->hw;
 979	struct igbvf_ring *tx_ring = adapter->tx_ring;
 980	struct igbvf_ring *rx_ring = adapter->rx_ring;
 981	int vector = 0;
 982
 983	adapter->eims_enable_mask = 0;
 984
 985	igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
 986	adapter->eims_enable_mask |= tx_ring->eims_value;
 987	writel(tx_ring->itr_val, hw->hw_addr + tx_ring->itr_register);
 988	igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
 989	adapter->eims_enable_mask |= rx_ring->eims_value;
 990	writel(rx_ring->itr_val, hw->hw_addr + rx_ring->itr_register);
 991
 992	/* set vector for other causes, i.e. link changes */
 993
 994	tmp = (vector++ | E1000_IVAR_VALID);
 995
 996	ew32(IVAR_MISC, tmp);
 997
 998	adapter->eims_enable_mask = GENMASK(vector - 1, 0);
 999	adapter->eims_other = BIT(vector - 1);
1000	e1e_flush();
1001}
1002
1003static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
1004{
1005	if (adapter->msix_entries) {
1006		pci_disable_msix(adapter->pdev);
1007		kfree(adapter->msix_entries);
1008		adapter->msix_entries = NULL;
1009	}
1010}
1011
1012/**
1013 * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1014 * @adapter: board private structure
1015 *
1016 * Attempt to configure interrupts using the best available
1017 * capabilities of the hardware and kernel.
1018 **/
1019static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1020{
1021	int err = -ENOMEM;
1022	int i;
1023
1024	/* we allocate 3 vectors, 1 for Tx, 1 for Rx, one for PF messages */
1025	adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1026					GFP_KERNEL);
1027	if (adapter->msix_entries) {
1028		for (i = 0; i < 3; i++)
1029			adapter->msix_entries[i].entry = i;
1030
1031		err = pci_enable_msix_range(adapter->pdev,
1032					    adapter->msix_entries, 3, 3);
1033	}
1034
1035	if (err < 0) {
1036		/* MSI-X failed */
1037		dev_err(&adapter->pdev->dev,
1038			"Failed to initialize MSI-X interrupts.\n");
1039		igbvf_reset_interrupt_capability(adapter);
1040	}
1041}
1042
1043/**
1044 * igbvf_request_msix - Initialize MSI-X interrupts
1045 * @adapter: board private structure
1046 *
1047 * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1048 * kernel.
1049 **/
1050static int igbvf_request_msix(struct igbvf_adapter *adapter)
1051{
1052	struct net_device *netdev = adapter->netdev;
1053	int err = 0, vector = 0;
1054
1055	if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1056		sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1057		sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1058	} else {
1059		memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1060		memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1061	}
1062
1063	err = request_irq(adapter->msix_entries[vector].vector,
1064			  igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1065			  netdev);
1066	if (err)
1067		goto out;
1068
1069	adapter->tx_ring->itr_register = E1000_EITR(vector);
1070	adapter->tx_ring->itr_val = adapter->current_itr;
1071	vector++;
1072
1073	err = request_irq(adapter->msix_entries[vector].vector,
1074			  igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1075			  netdev);
1076	if (err)
1077		goto out;
1078
1079	adapter->rx_ring->itr_register = E1000_EITR(vector);
1080	adapter->rx_ring->itr_val = adapter->current_itr;
1081	vector++;
1082
1083	err = request_irq(adapter->msix_entries[vector].vector,
1084			  igbvf_msix_other, 0, netdev->name, netdev);
1085	if (err)
1086		goto out;
1087
1088	igbvf_configure_msix(adapter);
1089	return 0;
1090out:
1091	return err;
1092}
1093
1094/**
1095 * igbvf_alloc_queues - Allocate memory for all rings
1096 * @adapter: board private structure to initialize
1097 **/
1098static int igbvf_alloc_queues(struct igbvf_adapter *adapter)
1099{
1100	struct net_device *netdev = adapter->netdev;
1101
1102	adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1103	if (!adapter->tx_ring)
1104		return -ENOMEM;
1105
1106	adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1107	if (!adapter->rx_ring) {
1108		kfree(adapter->tx_ring);
1109		return -ENOMEM;
1110	}
1111
1112	netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1113
1114	return 0;
1115}
1116
1117/**
1118 * igbvf_request_irq - initialize interrupts
1119 * @adapter: board private structure
1120 *
1121 * Attempts to configure interrupts using the best available
1122 * capabilities of the hardware and kernel.
1123 **/
1124static int igbvf_request_irq(struct igbvf_adapter *adapter)
1125{
1126	int err = -1;
1127
1128	/* igbvf supports msi-x only */
1129	if (adapter->msix_entries)
1130		err = igbvf_request_msix(adapter);
1131
1132	if (!err)
1133		return err;
1134
1135	dev_err(&adapter->pdev->dev,
1136		"Unable to allocate interrupt, Error: %d\n", err);
1137
1138	return err;
1139}
1140
1141static void igbvf_free_irq(struct igbvf_adapter *adapter)
1142{
1143	struct net_device *netdev = adapter->netdev;
1144	int vector;
1145
1146	if (adapter->msix_entries) {
1147		for (vector = 0; vector < 3; vector++)
1148			free_irq(adapter->msix_entries[vector].vector, netdev);
1149	}
1150}
1151
1152/**
1153 * igbvf_irq_disable - Mask off interrupt generation on the NIC
1154 * @adapter: board private structure
1155 **/
1156static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1157{
1158	struct e1000_hw *hw = &adapter->hw;
1159
1160	ew32(EIMC, ~0);
1161
1162	if (adapter->msix_entries)
1163		ew32(EIAC, 0);
1164}
1165
1166/**
1167 * igbvf_irq_enable - Enable default interrupt generation settings
1168 * @adapter: board private structure
1169 **/
1170static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1171{
1172	struct e1000_hw *hw = &adapter->hw;
1173
1174	ew32(EIAC, adapter->eims_enable_mask);
1175	ew32(EIAM, adapter->eims_enable_mask);
1176	ew32(EIMS, adapter->eims_enable_mask);
1177}
1178
1179/**
1180 * igbvf_poll - NAPI Rx polling callback
1181 * @napi: struct associated with this polling callback
1182 * @budget: amount of packets driver is allowed to process this poll
1183 **/
1184static int igbvf_poll(struct napi_struct *napi, int budget)
1185{
1186	struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1187	struct igbvf_adapter *adapter = rx_ring->adapter;
1188	struct e1000_hw *hw = &adapter->hw;
1189	int work_done = 0;
1190
1191	igbvf_clean_rx_irq(adapter, &work_done, budget);
1192
1193	if (work_done == budget)
1194		return budget;
1195
1196	/* Exit the polling mode, but don't re-enable interrupts if stack might
1197	 * poll us due to busy-polling
1198	 */
1199	if (likely(napi_complete_done(napi, work_done))) {
1200		if (adapter->requested_itr & 3)
1201			igbvf_set_itr(adapter);
1202
1203		if (!test_bit(__IGBVF_DOWN, &adapter->state))
1204			ew32(EIMS, adapter->rx_ring->eims_value);
1205	}
1206
1207	return work_done;
1208}
1209
1210/**
1211 * igbvf_set_rlpml - set receive large packet maximum length
1212 * @adapter: board private structure
1213 *
1214 * Configure the maximum size of packets that will be received
1215 */
1216static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1217{
1218	int max_frame_size;
1219	struct e1000_hw *hw = &adapter->hw;
1220
1221	max_frame_size = adapter->max_frame_size + VLAN_TAG_SIZE;
1222
1223	spin_lock_bh(&hw->mbx_lock);
1224
1225	e1000_rlpml_set_vf(hw, max_frame_size);
1226
1227	spin_unlock_bh(&hw->mbx_lock);
1228}
1229
1230static int igbvf_vlan_rx_add_vid(struct net_device *netdev,
1231				 __be16 proto, u16 vid)
1232{
1233	struct igbvf_adapter *adapter = netdev_priv(netdev);
1234	struct e1000_hw *hw = &adapter->hw;
1235
1236	spin_lock_bh(&hw->mbx_lock);
1237
1238	if (hw->mac.ops.set_vfta(hw, vid, true)) {
1239		dev_warn(&adapter->pdev->dev, "Vlan id %d\n is not added", vid);
1240		spin_unlock_bh(&hw->mbx_lock);
1241		return -EINVAL;
1242	}
1243
1244	spin_unlock_bh(&hw->mbx_lock);
1245
1246	set_bit(vid, adapter->active_vlans);
1247	return 0;
1248}
1249
1250static int igbvf_vlan_rx_kill_vid(struct net_device *netdev,
1251				  __be16 proto, u16 vid)
1252{
1253	struct igbvf_adapter *adapter = netdev_priv(netdev);
1254	struct e1000_hw *hw = &adapter->hw;
1255
1256	spin_lock_bh(&hw->mbx_lock);
1257
1258	if (hw->mac.ops.set_vfta(hw, vid, false)) {
1259		dev_err(&adapter->pdev->dev,
1260			"Failed to remove vlan id %d\n", vid);
1261		spin_unlock_bh(&hw->mbx_lock);
1262		return -EINVAL;
1263	}
1264
1265	spin_unlock_bh(&hw->mbx_lock);
1266
1267	clear_bit(vid, adapter->active_vlans);
1268	return 0;
1269}
1270
1271static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1272{
1273	u16 vid;
1274
1275	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1276		igbvf_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
1277}
1278
1279/**
1280 * igbvf_configure_tx - Configure Transmit Unit after Reset
1281 * @adapter: board private structure
1282 *
1283 * Configure the Tx unit of the MAC after a reset.
1284 **/
1285static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1286{
1287	struct e1000_hw *hw = &adapter->hw;
1288	struct igbvf_ring *tx_ring = adapter->tx_ring;
1289	u64 tdba;
1290	u32 txdctl, dca_txctrl;
1291
1292	/* disable transmits */
1293	txdctl = er32(TXDCTL(0));
1294	ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1295	e1e_flush();
1296	msleep(10);
1297
1298	/* Setup the HW Tx Head and Tail descriptor pointers */
1299	ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1300	tdba = tx_ring->dma;
1301	ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1302	ew32(TDBAH(0), (tdba >> 32));
1303	ew32(TDH(0), 0);
1304	ew32(TDT(0), 0);
1305	tx_ring->head = E1000_TDH(0);
1306	tx_ring->tail = E1000_TDT(0);
1307
1308	/* Turn off Relaxed Ordering on head write-backs.  The writebacks
1309	 * MUST be delivered in order or it will completely screw up
1310	 * our bookkeeping.
1311	 */
1312	dca_txctrl = er32(DCA_TXCTRL(0));
1313	dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1314	ew32(DCA_TXCTRL(0), dca_txctrl);
1315
1316	/* enable transmits */
1317	txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1318	ew32(TXDCTL(0), txdctl);
1319
1320	/* Setup Transmit Descriptor Settings for eop descriptor */
1321	adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1322
1323	/* enable Report Status bit */
1324	adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1325}
1326
1327/**
1328 * igbvf_setup_srrctl - configure the receive control registers
1329 * @adapter: Board private structure
1330 **/
1331static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1332{
1333	struct e1000_hw *hw = &adapter->hw;
1334	u32 srrctl = 0;
1335
1336	srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1337		    E1000_SRRCTL_BSIZEHDR_MASK |
1338		    E1000_SRRCTL_BSIZEPKT_MASK);
1339
1340	/* Enable queue drop to avoid head of line blocking */
1341	srrctl |= E1000_SRRCTL_DROP_EN;
1342
1343	/* Setup buffer sizes */
1344	srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1345		  E1000_SRRCTL_BSIZEPKT_SHIFT;
1346
1347	if (adapter->rx_buffer_len < 2048) {
1348		adapter->rx_ps_hdr_size = 0;
1349		srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1350	} else {
1351		adapter->rx_ps_hdr_size = 128;
1352		srrctl |= adapter->rx_ps_hdr_size <<
1353			  E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1354		srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1355	}
1356
1357	ew32(SRRCTL(0), srrctl);
1358}
1359
1360/**
1361 * igbvf_configure_rx - Configure Receive Unit after Reset
1362 * @adapter: board private structure
1363 *
1364 * Configure the Rx unit of the MAC after a reset.
1365 **/
1366static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1367{
1368	struct e1000_hw *hw = &adapter->hw;
1369	struct igbvf_ring *rx_ring = adapter->rx_ring;
1370	u64 rdba;
1371	u32 rxdctl;
1372
1373	/* disable receives */
1374	rxdctl = er32(RXDCTL(0));
1375	ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1376	e1e_flush();
1377	msleep(10);
1378
1379	/* Setup the HW Rx Head and Tail Descriptor Pointers and
1380	 * the Base and Length of the Rx Descriptor Ring
1381	 */
1382	rdba = rx_ring->dma;
1383	ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1384	ew32(RDBAH(0), (rdba >> 32));
1385	ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1386	rx_ring->head = E1000_RDH(0);
1387	rx_ring->tail = E1000_RDT(0);
1388	ew32(RDH(0), 0);
1389	ew32(RDT(0), 0);
1390
1391	rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1392	rxdctl &= 0xFFF00000;
1393	rxdctl |= IGBVF_RX_PTHRESH;
1394	rxdctl |= IGBVF_RX_HTHRESH << 8;
1395	rxdctl |= IGBVF_RX_WTHRESH << 16;
1396
1397	igbvf_set_rlpml(adapter);
1398
1399	/* enable receives */
1400	ew32(RXDCTL(0), rxdctl);
1401}
1402
1403/**
1404 * igbvf_set_multi - Multicast and Promiscuous mode set
1405 * @netdev: network interface device structure
1406 *
1407 * The set_multi entry point is called whenever the multicast address
1408 * list or the network interface flags are updated.  This routine is
1409 * responsible for configuring the hardware for proper multicast,
1410 * promiscuous mode, and all-multi behavior.
1411 **/
1412static void igbvf_set_multi(struct net_device *netdev)
1413{
1414	struct igbvf_adapter *adapter = netdev_priv(netdev);
1415	struct e1000_hw *hw = &adapter->hw;
1416	struct netdev_hw_addr *ha;
1417	u8  *mta_list = NULL;
1418	int i;
1419
1420	if (!netdev_mc_empty(netdev)) {
1421		mta_list = kmalloc_array(netdev_mc_count(netdev), ETH_ALEN,
1422					 GFP_ATOMIC);
1423		if (!mta_list)
1424			return;
1425	}
1426
1427	/* prepare a packed array of only addresses. */
1428	i = 0;
1429	netdev_for_each_mc_addr(ha, netdev)
1430		memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1431
1432	spin_lock_bh(&hw->mbx_lock);
1433
1434	hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1435
1436	spin_unlock_bh(&hw->mbx_lock);
1437	kfree(mta_list);
1438}
1439
1440/**
1441 * igbvf_set_uni - Configure unicast MAC filters
1442 * @netdev: network interface device structure
1443 *
1444 * This routine is responsible for configuring the hardware for proper
1445 * unicast filters.
1446 **/
1447static int igbvf_set_uni(struct net_device *netdev)
1448{
1449	struct igbvf_adapter *adapter = netdev_priv(netdev);
1450	struct e1000_hw *hw = &adapter->hw;
1451
1452	if (netdev_uc_count(netdev) > IGBVF_MAX_MAC_FILTERS) {
1453		pr_err("Too many unicast filters - No Space\n");
1454		return -ENOSPC;
1455	}
1456
1457	spin_lock_bh(&hw->mbx_lock);
1458
1459	/* Clear all unicast MAC filters */
1460	hw->mac.ops.set_uc_addr(hw, E1000_VF_MAC_FILTER_CLR, NULL);
1461
1462	spin_unlock_bh(&hw->mbx_lock);
1463
1464	if (!netdev_uc_empty(netdev)) {
1465		struct netdev_hw_addr *ha;
1466
1467		/* Add MAC filters one by one */
1468		netdev_for_each_uc_addr(ha, netdev) {
1469			spin_lock_bh(&hw->mbx_lock);
1470
1471			hw->mac.ops.set_uc_addr(hw, E1000_VF_MAC_FILTER_ADD,
1472						ha->addr);
1473
1474			spin_unlock_bh(&hw->mbx_lock);
1475			udelay(200);
1476		}
1477	}
1478
1479	return 0;
1480}
1481
1482static void igbvf_set_rx_mode(struct net_device *netdev)
1483{
1484	igbvf_set_multi(netdev);
1485	igbvf_set_uni(netdev);
1486}
1487
1488/**
1489 * igbvf_configure - configure the hardware for Rx and Tx
1490 * @adapter: private board structure
1491 **/
1492static void igbvf_configure(struct igbvf_adapter *adapter)
1493{
1494	igbvf_set_rx_mode(adapter->netdev);
1495
1496	igbvf_restore_vlan(adapter);
1497
1498	igbvf_configure_tx(adapter);
1499	igbvf_setup_srrctl(adapter);
1500	igbvf_configure_rx(adapter);
1501	igbvf_alloc_rx_buffers(adapter->rx_ring,
1502			       igbvf_desc_unused(adapter->rx_ring));
1503}
1504
1505/* igbvf_reset - bring the hardware into a known good state
1506 * @adapter: private board structure
1507 *
1508 * This function boots the hardware and enables some settings that
1509 * require a configuration cycle of the hardware - those cannot be
1510 * set/changed during runtime. After reset the device needs to be
1511 * properly configured for Rx, Tx etc.
1512 */
1513static void igbvf_reset(struct igbvf_adapter *adapter)
1514{
1515	struct e1000_mac_info *mac = &adapter->hw.mac;
1516	struct net_device *netdev = adapter->netdev;
1517	struct e1000_hw *hw = &adapter->hw;
1518
1519	spin_lock_bh(&hw->mbx_lock);
1520
1521	/* Allow time for pending master requests to run */
1522	if (mac->ops.reset_hw(hw))
1523		dev_warn(&adapter->pdev->dev, "PF still resetting\n");
1524
1525	mac->ops.init_hw(hw);
1526
1527	spin_unlock_bh(&hw->mbx_lock);
1528
1529	if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1530		memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1531		       netdev->addr_len);
1532		memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1533		       netdev->addr_len);
1534	}
1535
1536	adapter->last_reset = jiffies;
1537}
1538
1539int igbvf_up(struct igbvf_adapter *adapter)
1540{
1541	struct e1000_hw *hw = &adapter->hw;
1542
1543	/* hardware has been reset, we need to reload some things */
1544	igbvf_configure(adapter);
1545
1546	clear_bit(__IGBVF_DOWN, &adapter->state);
1547
1548	napi_enable(&adapter->rx_ring->napi);
1549	if (adapter->msix_entries)
1550		igbvf_configure_msix(adapter);
1551
1552	/* Clear any pending interrupts. */
1553	er32(EICR);
1554	igbvf_irq_enable(adapter);
1555
1556	/* start the watchdog */
1557	hw->mac.get_link_status = 1;
1558	mod_timer(&adapter->watchdog_timer, jiffies + 1);
1559
1560	return 0;
1561}
1562
1563void igbvf_down(struct igbvf_adapter *adapter)
1564{
1565	struct net_device *netdev = adapter->netdev;
1566	struct e1000_hw *hw = &adapter->hw;
1567	u32 rxdctl, txdctl;
1568
1569	/* signal that we're down so the interrupt handler does not
1570	 * reschedule our watchdog timer
1571	 */
1572	set_bit(__IGBVF_DOWN, &adapter->state);
1573
1574	/* disable receives in the hardware */
1575	rxdctl = er32(RXDCTL(0));
1576	ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1577
1578	netif_carrier_off(netdev);
1579	netif_stop_queue(netdev);
1580
1581	/* disable transmits in the hardware */
1582	txdctl = er32(TXDCTL(0));
1583	ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1584
1585	/* flush both disables and wait for them to finish */
1586	e1e_flush();
1587	msleep(10);
1588
1589	napi_disable(&adapter->rx_ring->napi);
1590
1591	igbvf_irq_disable(adapter);
1592
1593	del_timer_sync(&adapter->watchdog_timer);
1594
1595	/* record the stats before reset*/
1596	igbvf_update_stats(adapter);
1597
1598	adapter->link_speed = 0;
1599	adapter->link_duplex = 0;
1600
1601	igbvf_reset(adapter);
1602	igbvf_clean_tx_ring(adapter->tx_ring);
1603	igbvf_clean_rx_ring(adapter->rx_ring);
1604}
1605
1606void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1607{
1608	might_sleep();
1609	while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1610		usleep_range(1000, 2000);
1611	igbvf_down(adapter);
1612	igbvf_up(adapter);
1613	clear_bit(__IGBVF_RESETTING, &adapter->state);
1614}
1615
1616/**
1617 * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1618 * @adapter: board private structure to initialize
1619 *
1620 * igbvf_sw_init initializes the Adapter private data structure.
1621 * Fields are initialized based on PCI device information and
1622 * OS network device settings (MTU size).
1623 **/
1624static int igbvf_sw_init(struct igbvf_adapter *adapter)
1625{
1626	struct net_device *netdev = adapter->netdev;
1627	s32 rc;
1628
1629	adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1630	adapter->rx_ps_hdr_size = 0;
1631	adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1632	adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1633
1634	adapter->tx_int_delay = 8;
1635	adapter->tx_abs_int_delay = 32;
1636	adapter->rx_int_delay = 0;
1637	adapter->rx_abs_int_delay = 8;
1638	adapter->requested_itr = 3;
1639	adapter->current_itr = IGBVF_START_ITR;
1640
1641	/* Set various function pointers */
1642	adapter->ei->init_ops(&adapter->hw);
1643
1644	rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1645	if (rc)
1646		return rc;
1647
1648	rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1649	if (rc)
1650		return rc;
1651
1652	igbvf_set_interrupt_capability(adapter);
1653
1654	if (igbvf_alloc_queues(adapter))
1655		return -ENOMEM;
1656
1657	spin_lock_init(&adapter->tx_queue_lock);
1658
1659	/* Explicitly disable IRQ since the NIC can be in any state. */
1660	igbvf_irq_disable(adapter);
1661
1662	spin_lock_init(&adapter->stats_lock);
1663	spin_lock_init(&adapter->hw.mbx_lock);
1664
1665	set_bit(__IGBVF_DOWN, &adapter->state);
1666	return 0;
1667}
1668
1669static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1670{
1671	struct e1000_hw *hw = &adapter->hw;
1672
1673	adapter->stats.last_gprc = er32(VFGPRC);
1674	adapter->stats.last_gorc = er32(VFGORC);
1675	adapter->stats.last_gptc = er32(VFGPTC);
1676	adapter->stats.last_gotc = er32(VFGOTC);
1677	adapter->stats.last_mprc = er32(VFMPRC);
1678	adapter->stats.last_gotlbc = er32(VFGOTLBC);
1679	adapter->stats.last_gptlbc = er32(VFGPTLBC);
1680	adapter->stats.last_gorlbc = er32(VFGORLBC);
1681	adapter->stats.last_gprlbc = er32(VFGPRLBC);
1682
1683	adapter->stats.base_gprc = er32(VFGPRC);
1684	adapter->stats.base_gorc = er32(VFGORC);
1685	adapter->stats.base_gptc = er32(VFGPTC);
1686	adapter->stats.base_gotc = er32(VFGOTC);
1687	adapter->stats.base_mprc = er32(VFMPRC);
1688	adapter->stats.base_gotlbc = er32(VFGOTLBC);
1689	adapter->stats.base_gptlbc = er32(VFGPTLBC);
1690	adapter->stats.base_gorlbc = er32(VFGORLBC);
1691	adapter->stats.base_gprlbc = er32(VFGPRLBC);
1692}
1693
1694/**
1695 * igbvf_open - Called when a network interface is made active
1696 * @netdev: network interface device structure
1697 *
1698 * Returns 0 on success, negative value on failure
1699 *
1700 * The open entry point is called when a network interface is made
1701 * active by the system (IFF_UP).  At this point all resources needed
1702 * for transmit and receive operations are allocated, the interrupt
1703 * handler is registered with the OS, the watchdog timer is started,
1704 * and the stack is notified that the interface is ready.
1705 **/
1706static int igbvf_open(struct net_device *netdev)
1707{
1708	struct igbvf_adapter *adapter = netdev_priv(netdev);
1709	struct e1000_hw *hw = &adapter->hw;
1710	int err;
1711
1712	/* disallow open during test */
1713	if (test_bit(__IGBVF_TESTING, &adapter->state))
1714		return -EBUSY;
1715
1716	/* allocate transmit descriptors */
1717	err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1718	if (err)
1719		goto err_setup_tx;
1720
1721	/* allocate receive descriptors */
1722	err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1723	if (err)
1724		goto err_setup_rx;
1725
1726	/* before we allocate an interrupt, we must be ready to handle it.
1727	 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1728	 * as soon as we call pci_request_irq, so we have to setup our
1729	 * clean_rx handler before we do so.
1730	 */
1731	igbvf_configure(adapter);
1732
1733	err = igbvf_request_irq(adapter);
1734	if (err)
1735		goto err_req_irq;
1736
1737	/* From here on the code is the same as igbvf_up() */
1738	clear_bit(__IGBVF_DOWN, &adapter->state);
1739
1740	napi_enable(&adapter->rx_ring->napi);
1741
1742	/* clear any pending interrupts */
1743	er32(EICR);
1744
1745	igbvf_irq_enable(adapter);
1746
1747	/* start the watchdog */
1748	hw->mac.get_link_status = 1;
1749	mod_timer(&adapter->watchdog_timer, jiffies + 1);
1750
1751	return 0;
1752
1753err_req_irq:
1754	igbvf_free_rx_resources(adapter->rx_ring);
1755err_setup_rx:
1756	igbvf_free_tx_resources(adapter->tx_ring);
1757err_setup_tx:
1758	igbvf_reset(adapter);
1759
1760	return err;
1761}
1762
1763/**
1764 * igbvf_close - Disables a network interface
1765 * @netdev: network interface device structure
1766 *
1767 * Returns 0, this is not allowed to fail
1768 *
1769 * The close entry point is called when an interface is de-activated
1770 * by the OS.  The hardware is still under the drivers control, but
1771 * needs to be disabled.  A global MAC reset is issued to stop the
1772 * hardware, and all transmit and receive resources are freed.
1773 **/
1774static int igbvf_close(struct net_device *netdev)
1775{
1776	struct igbvf_adapter *adapter = netdev_priv(netdev);
1777
1778	WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1779	igbvf_down(adapter);
1780
1781	igbvf_free_irq(adapter);
1782
1783	igbvf_free_tx_resources(adapter->tx_ring);
1784	igbvf_free_rx_resources(adapter->rx_ring);
1785
1786	return 0;
1787}
1788
1789/**
1790 * igbvf_set_mac - Change the Ethernet Address of the NIC
1791 * @netdev: network interface device structure
1792 * @p: pointer to an address structure
1793 *
1794 * Returns 0 on success, negative on failure
1795 **/
1796static int igbvf_set_mac(struct net_device *netdev, void *p)
1797{
1798	struct igbvf_adapter *adapter = netdev_priv(netdev);
1799	struct e1000_hw *hw = &adapter->hw;
1800	struct sockaddr *addr = p;
1801
1802	if (!is_valid_ether_addr(addr->sa_data))
1803		return -EADDRNOTAVAIL;
1804
1805	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1806
1807	spin_lock_bh(&hw->mbx_lock);
1808
1809	hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1810
1811	spin_unlock_bh(&hw->mbx_lock);
1812
1813	if (!ether_addr_equal(addr->sa_data, hw->mac.addr))
1814		return -EADDRNOTAVAIL;
1815
1816	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1817
1818	return 0;
1819}
1820
1821#define UPDATE_VF_COUNTER(reg, name) \
1822{ \
1823	u32 current_counter = er32(reg); \
1824	if (current_counter < adapter->stats.last_##name) \
1825		adapter->stats.name += 0x100000000LL; \
1826	adapter->stats.last_##name = current_counter; \
1827	adapter->stats.name &= 0xFFFFFFFF00000000LL; \
1828	adapter->stats.name |= current_counter; \
1829}
1830
1831/**
1832 * igbvf_update_stats - Update the board statistics counters
1833 * @adapter: board private structure
1834**/
1835void igbvf_update_stats(struct igbvf_adapter *adapter)
1836{
1837	struct e1000_hw *hw = &adapter->hw;
1838	struct pci_dev *pdev = adapter->pdev;
1839
1840	/* Prevent stats update while adapter is being reset, link is down
1841	 * or if the pci connection is down.
1842	 */
1843	if (adapter->link_speed == 0)
1844		return;
1845
1846	if (test_bit(__IGBVF_RESETTING, &adapter->state))
1847		return;
1848
1849	if (pci_channel_offline(pdev))
1850		return;
1851
1852	UPDATE_VF_COUNTER(VFGPRC, gprc);
1853	UPDATE_VF_COUNTER(VFGORC, gorc);
1854	UPDATE_VF_COUNTER(VFGPTC, gptc);
1855	UPDATE_VF_COUNTER(VFGOTC, gotc);
1856	UPDATE_VF_COUNTER(VFMPRC, mprc);
1857	UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1858	UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1859	UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1860	UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1861
1862	/* Fill out the OS statistics structure */
1863	adapter->netdev->stats.multicast = adapter->stats.mprc;
1864}
1865
1866static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1867{
1868	dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s Duplex\n",
1869		 adapter->link_speed,
1870		 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half");
1871}
1872
1873static bool igbvf_has_link(struct igbvf_adapter *adapter)
1874{
1875	struct e1000_hw *hw = &adapter->hw;
1876	s32 ret_val = E1000_SUCCESS;
1877	bool link_active;
1878
1879	/* If interface is down, stay link down */
1880	if (test_bit(__IGBVF_DOWN, &adapter->state))
1881		return false;
1882
1883	spin_lock_bh(&hw->mbx_lock);
1884
1885	ret_val = hw->mac.ops.check_for_link(hw);
1886
1887	spin_unlock_bh(&hw->mbx_lock);
1888
1889	link_active = !hw->mac.get_link_status;
1890
1891	/* if check for link returns error we will need to reset */
1892	if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1893		schedule_work(&adapter->reset_task);
1894
1895	return link_active;
1896}
1897
1898/**
1899 * igbvf_watchdog - Timer Call-back
1900 * @t: timer list pointer containing private struct
1901 **/
1902static void igbvf_watchdog(struct timer_list *t)
1903{
1904	struct igbvf_adapter *adapter = from_timer(adapter, t, watchdog_timer);
1905
1906	/* Do the rest outside of interrupt context */
1907	schedule_work(&adapter->watchdog_task);
1908}
1909
1910static void igbvf_watchdog_task(struct work_struct *work)
1911{
1912	struct igbvf_adapter *adapter = container_of(work,
1913						     struct igbvf_adapter,
1914						     watchdog_task);
1915	struct net_device *netdev = adapter->netdev;
1916	struct e1000_mac_info *mac = &adapter->hw.mac;
1917	struct igbvf_ring *tx_ring = adapter->tx_ring;
1918	struct e1000_hw *hw = &adapter->hw;
1919	u32 link;
1920	int tx_pending = 0;
1921
1922	link = igbvf_has_link(adapter);
1923
1924	if (link) {
1925		if (!netif_carrier_ok(netdev)) {
1926			mac->ops.get_link_up_info(&adapter->hw,
1927						  &adapter->link_speed,
1928						  &adapter->link_duplex);
1929			igbvf_print_link_info(adapter);
1930
1931			netif_carrier_on(netdev);
1932			netif_wake_queue(netdev);
1933		}
1934	} else {
1935		if (netif_carrier_ok(netdev)) {
1936			adapter->link_speed = 0;
1937			adapter->link_duplex = 0;
1938			dev_info(&adapter->pdev->dev, "Link is Down\n");
1939			netif_carrier_off(netdev);
1940			netif_stop_queue(netdev);
1941		}
1942	}
1943
1944	if (netif_carrier_ok(netdev)) {
1945		igbvf_update_stats(adapter);
1946	} else {
1947		tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1948			      tx_ring->count);
1949		if (tx_pending) {
1950			/* We've lost link, so the controller stops DMA,
1951			 * but we've got queued Tx work that's never going
1952			 * to get done, so reset controller to flush Tx.
1953			 * (Do the reset outside of interrupt context).
1954			 */
1955			adapter->tx_timeout_count++;
1956			schedule_work(&adapter->reset_task);
1957		}
1958	}
1959
1960	/* Cause software interrupt to ensure Rx ring is cleaned */
1961	ew32(EICS, adapter->rx_ring->eims_value);
1962
1963	/* Reset the timer */
1964	if (!test_bit(__IGBVF_DOWN, &adapter->state))
1965		mod_timer(&adapter->watchdog_timer,
1966			  round_jiffies(jiffies + (2 * HZ)));
1967}
1968
1969#define IGBVF_TX_FLAGS_CSUM		0x00000001
1970#define IGBVF_TX_FLAGS_VLAN		0x00000002
1971#define IGBVF_TX_FLAGS_TSO		0x00000004
1972#define IGBVF_TX_FLAGS_IPV4		0x00000008
1973#define IGBVF_TX_FLAGS_VLAN_MASK	0xffff0000
1974#define IGBVF_TX_FLAGS_VLAN_SHIFT	16
1975
1976static void igbvf_tx_ctxtdesc(struct igbvf_ring *tx_ring, u32 vlan_macip_lens,
1977			      u32 type_tucmd, u32 mss_l4len_idx)
1978{
1979	struct e1000_adv_tx_context_desc *context_desc;
1980	struct igbvf_buffer *buffer_info;
1981	u16 i = tx_ring->next_to_use;
1982
1983	context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1984	buffer_info = &tx_ring->buffer_info[i];
1985
1986	i++;
1987	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1988
1989	/* set bits to identify this as an advanced context descriptor */
1990	type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
1991
1992	context_desc->vlan_macip_lens	= cpu_to_le32(vlan_macip_lens);
1993	context_desc->seqnum_seed	= 0;
1994	context_desc->type_tucmd_mlhl	= cpu_to_le32(type_tucmd);
1995	context_desc->mss_l4len_idx	= cpu_to_le32(mss_l4len_idx);
1996
1997	buffer_info->time_stamp = jiffies;
1998	buffer_info->dma = 0;
1999}
2000
2001static int igbvf_tso(struct igbvf_ring *tx_ring,
2002		     struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2003{
2004	u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
2005	union {
2006		struct iphdr *v4;
2007		struct ipv6hdr *v6;
2008		unsigned char *hdr;
2009	} ip;
2010	union {
2011		struct tcphdr *tcp;
2012		unsigned char *hdr;
2013	} l4;
2014	u32 paylen, l4_offset;
2015	int err;
2016
2017	if (skb->ip_summed != CHECKSUM_PARTIAL)
2018		return 0;
2019
2020	if (!skb_is_gso(skb))
2021		return 0;
2022
2023	err = skb_cow_head(skb, 0);
2024	if (err < 0)
2025		return err;
2026
2027	ip.hdr = skb_network_header(skb);
2028	l4.hdr = skb_checksum_start(skb);
2029
2030	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2031	type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
2032
2033	/* initialize outer IP header fields */
2034	if (ip.v4->version == 4) {
2035		unsigned char *csum_start = skb_checksum_start(skb);
2036		unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
2037
2038		/* IP header will have to cancel out any data that
2039		 * is not a part of the outer IP header
2040		 */
2041		ip.v4->check = csum_fold(csum_partial(trans_start,
2042						      csum_start - trans_start,
2043						      0));
2044		type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
2045
2046		ip.v4->tot_len = 0;
2047	} else {
2048		ip.v6->payload_len = 0;
2049	}
2050
2051	/* determine offset of inner transport header */
2052	l4_offset = l4.hdr - skb->data;
2053
2054	/* compute length of segmentation header */
2055	*hdr_len = (l4.tcp->doff * 4) + l4_offset;
2056
2057	/* remove payload length from inner checksum */
2058	paylen = skb->len - l4_offset;
2059	csum_replace_by_diff(&l4.tcp->check, (__force __wsum)htonl(paylen));
2060
2061	/* MSS L4LEN IDX */
2062	mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT;
2063	mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
2064
2065	/* VLAN MACLEN IPLEN */
2066	vlan_macip_lens = l4.hdr - ip.hdr;
2067	vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT;
2068	vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK;
2069
2070	igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
2071
2072	return 1;
2073}
2074
2075static bool igbvf_tx_csum(struct igbvf_ring *tx_ring, struct sk_buff *skb,
2076			  u32 tx_flags, __be16 protocol)
2077{
2078	u32 vlan_macip_lens = 0;
2079	u32 type_tucmd = 0;
2080
2081	if (skb->ip_summed != CHECKSUM_PARTIAL) {
2082csum_failed:
2083		if (!(tx_flags & IGBVF_TX_FLAGS_VLAN))
2084			return false;
2085		goto no_csum;
2086	}
2087
2088	switch (skb->csum_offset) {
2089	case offsetof(struct tcphdr, check):
2090		type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
2091		fallthrough;
2092	case offsetof(struct udphdr, check):
2093		break;
2094	case offsetof(struct sctphdr, checksum):
2095		/* validate that this is actually an SCTP request */
2096		if (skb_csum_is_sctp(skb)) {
2097			type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP;
2098			break;
2099		}
2100		fallthrough;
2101	default:
2102		skb_checksum_help(skb);
2103		goto csum_failed;
2104	}
2105
2106	vlan_macip_lens = skb_checksum_start_offset(skb) -
2107			  skb_network_offset(skb);
2108no_csum:
2109	vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
2110	vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK;
2111
2112	igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0);
2113	return true;
2114}
2115
2116static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2117{
2118	struct igbvf_adapter *adapter = netdev_priv(netdev);
2119
2120	/* there is enough descriptors then we don't need to worry  */
2121	if (igbvf_desc_unused(adapter->tx_ring) >= size)
2122		return 0;
2123
2124	netif_stop_queue(netdev);
2125
2126	/* Herbert's original patch had:
2127	 *  smp_mb__after_netif_stop_queue();
2128	 * but since that doesn't exist yet, just open code it.
2129	 */
2130	smp_mb();
2131
2132	/* We need to check again just in case room has been made available */
2133	if (igbvf_desc_unused(adapter->tx_ring) < size)
2134		return -EBUSY;
2135
2136	netif_wake_queue(netdev);
2137
2138	++adapter->restart_queue;
2139	return 0;
2140}
2141
2142#define IGBVF_MAX_TXD_PWR	16
2143#define IGBVF_MAX_DATA_PER_TXD	(1u << IGBVF_MAX_TXD_PWR)
2144
2145static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2146				   struct igbvf_ring *tx_ring,
2147				   struct sk_buff *skb)
2148{
2149	struct igbvf_buffer *buffer_info;
2150	struct pci_dev *pdev = adapter->pdev;
2151	unsigned int len = skb_headlen(skb);
2152	unsigned int count = 0, i;
2153	unsigned int f;
2154
2155	i = tx_ring->next_to_use;
2156
2157	buffer_info = &tx_ring->buffer_info[i];
2158	BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2159	buffer_info->length = len;
2160	/* set time_stamp *before* dma to help avoid a possible race */
2161	buffer_info->time_stamp = jiffies;
2162	buffer_info->mapped_as_page = false;
2163	buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2164					  DMA_TO_DEVICE);
2165	if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2166		goto dma_error;
2167
2168	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2169		const skb_frag_t *frag;
2170
2171		count++;
2172		i++;
2173		if (i == tx_ring->count)
2174			i = 0;
2175
2176		frag = &skb_shinfo(skb)->frags[f];
2177		len = skb_frag_size(frag);
2178
2179		buffer_info = &tx_ring->buffer_info[i];
2180		BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2181		buffer_info->length = len;
2182		buffer_info->time_stamp = jiffies;
2183		buffer_info->mapped_as_page = true;
2184		buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag, 0, len,
2185						    DMA_TO_DEVICE);
2186		if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2187			goto dma_error;
2188	}
2189
2190	tx_ring->buffer_info[i].skb = skb;
2191
2192	return ++count;
2193
2194dma_error:
2195	dev_err(&pdev->dev, "TX DMA map failed\n");
2196
2197	/* clear timestamp and dma mappings for failed buffer_info mapping */
2198	buffer_info->dma = 0;
2199	buffer_info->time_stamp = 0;
2200	buffer_info->length = 0;
2201	buffer_info->mapped_as_page = false;
2202	if (count)
2203		count--;
2204
2205	/* clear timestamp and dma mappings for remaining portion of packet */
2206	while (count--) {
2207		if (i == 0)
2208			i += tx_ring->count;
2209		i--;
2210		buffer_info = &tx_ring->buffer_info[i];
2211		igbvf_put_txbuf(adapter, buffer_info);
2212	}
2213
2214	return 0;
2215}
2216
2217static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2218				      struct igbvf_ring *tx_ring,
2219				      int tx_flags, int count,
2220				      unsigned int first, u32 paylen,
2221				      u8 hdr_len)
2222{
2223	union e1000_adv_tx_desc *tx_desc = NULL;
2224	struct igbvf_buffer *buffer_info;
2225	u32 olinfo_status = 0, cmd_type_len;
2226	unsigned int i;
2227
2228	cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2229			E1000_ADVTXD_DCMD_DEXT);
2230
2231	if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2232		cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2233
2234	if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2235		cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2236
2237		/* insert tcp checksum */
2238		olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2239
2240		/* insert ip checksum */
2241		if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2242			olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2243
2244	} else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2245		olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2246	}
2247
2248	olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2249
2250	i = tx_ring->next_to_use;
2251	while (count--) {
2252		buffer_info = &tx_ring->buffer_info[i];
2253		tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2254		tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2255		tx_desc->read.cmd_type_len =
2256			 cpu_to_le32(cmd_type_len | buffer_info->length);
2257		tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2258		i++;
2259		if (i == tx_ring->count)
2260			i = 0;
2261	}
2262
2263	tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2264	/* Force memory writes to complete before letting h/w
2265	 * know there are new descriptors to fetch.  (Only
2266	 * applicable for weak-ordered memory model archs,
2267	 * such as IA-64).
2268	 */
2269	wmb();
2270
2271	tx_ring->buffer_info[first].next_to_watch = tx_desc;
2272	tx_ring->next_to_use = i;
2273	writel(i, adapter->hw.hw_addr + tx_ring->tail);
2274}
2275
2276static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2277					     struct net_device *netdev,
2278					     struct igbvf_ring *tx_ring)
2279{
2280	struct igbvf_adapter *adapter = netdev_priv(netdev);
2281	unsigned int first, tx_flags = 0;
2282	u8 hdr_len = 0;
2283	int count = 0;
2284	int tso = 0;
2285	__be16 protocol = vlan_get_protocol(skb);
2286
2287	if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2288		dev_kfree_skb_any(skb);
2289		return NETDEV_TX_OK;
2290	}
2291
2292	if (skb->len <= 0) {
2293		dev_kfree_skb_any(skb);
2294		return NETDEV_TX_OK;
2295	}
2296
2297	/* need: count + 4 desc gap to keep tail from touching
2298	 *       + 2 desc gap to keep tail from touching head,
2299	 *       + 1 desc for skb->data,
2300	 *       + 1 desc for context descriptor,
2301	 * head, otherwise try next time
2302	 */
2303	if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2304		/* this is a hard error */
2305		return NETDEV_TX_BUSY;
2306	}
2307
2308	if (skb_vlan_tag_present(skb)) {
2309		tx_flags |= IGBVF_TX_FLAGS_VLAN;
2310		tx_flags |= (skb_vlan_tag_get(skb) <<
2311			     IGBVF_TX_FLAGS_VLAN_SHIFT);
2312	}
2313
2314	if (protocol == htons(ETH_P_IP))
2315		tx_flags |= IGBVF_TX_FLAGS_IPV4;
2316
2317	first = tx_ring->next_to_use;
2318
2319	tso = igbvf_tso(tx_ring, skb, tx_flags, &hdr_len);
2320	if (unlikely(tso < 0)) {
2321		dev_kfree_skb_any(skb);
2322		return NETDEV_TX_OK;
2323	}
2324
2325	if (tso)
2326		tx_flags |= IGBVF_TX_FLAGS_TSO;
2327	else if (igbvf_tx_csum(tx_ring, skb, tx_flags, protocol) &&
2328		 (skb->ip_summed == CHECKSUM_PARTIAL))
2329		tx_flags |= IGBVF_TX_FLAGS_CSUM;
2330
2331	/* count reflects descriptors mapped, if 0 then mapping error
2332	 * has occurred and we need to rewind the descriptor queue
2333	 */
2334	count = igbvf_tx_map_adv(adapter, tx_ring, skb);
2335
2336	if (count) {
2337		igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2338				   first, skb->len, hdr_len);
2339		/* Make sure there is space in the ring for the next send. */
2340		igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2341	} else {
2342		dev_kfree_skb_any(skb);
2343		tx_ring->buffer_info[first].time_stamp = 0;
2344		tx_ring->next_to_use = first;
2345	}
2346
2347	return NETDEV_TX_OK;
2348}
2349
2350static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2351				    struct net_device *netdev)
2352{
2353	struct igbvf_adapter *adapter = netdev_priv(netdev);
2354	struct igbvf_ring *tx_ring;
2355
2356	if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2357		dev_kfree_skb_any(skb);
2358		return NETDEV_TX_OK;
2359	}
2360
2361	tx_ring = &adapter->tx_ring[0];
2362
2363	return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2364}
2365
2366/**
2367 * igbvf_tx_timeout - Respond to a Tx Hang
2368 * @netdev: network interface device structure
2369 * @txqueue: queue timing out (unused)
2370 **/
2371static void igbvf_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
2372{
2373	struct igbvf_adapter *adapter = netdev_priv(netdev);
2374
2375	/* Do the reset outside of interrupt context */
2376	adapter->tx_timeout_count++;
2377	schedule_work(&adapter->reset_task);
2378}
2379
2380static void igbvf_reset_task(struct work_struct *work)
2381{
2382	struct igbvf_adapter *adapter;
2383
2384	adapter = container_of(work, struct igbvf_adapter, reset_task);
2385
2386	igbvf_reinit_locked(adapter);
2387}
2388
2389/**
2390 * igbvf_change_mtu - Change the Maximum Transfer Unit
2391 * @netdev: network interface device structure
2392 * @new_mtu: new value for maximum frame size
2393 *
2394 * Returns 0 on success, negative on failure
2395 **/
2396static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2397{
2398	struct igbvf_adapter *adapter = netdev_priv(netdev);
2399	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2400
2401	while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2402		usleep_range(1000, 2000);
2403	/* igbvf_down has a dependency on max_frame_size */
2404	adapter->max_frame_size = max_frame;
2405	if (netif_running(netdev))
2406		igbvf_down(adapter);
2407
2408	/* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2409	 * means we reserve 2 more, this pushes us to allocate from the next
2410	 * larger slab size.
2411	 * i.e. RXBUFFER_2048 --> size-4096 slab
2412	 * However with the new *_jumbo_rx* routines, jumbo receives will use
2413	 * fragmented skbs
2414	 */
2415
2416	if (max_frame <= 1024)
2417		adapter->rx_buffer_len = 1024;
2418	else if (max_frame <= 2048)
2419		adapter->rx_buffer_len = 2048;
2420	else
2421#if (PAGE_SIZE / 2) > 16384
2422		adapter->rx_buffer_len = 16384;
2423#else
2424		adapter->rx_buffer_len = PAGE_SIZE / 2;
2425#endif
2426
2427	/* adjust allocation if LPE protects us, and we aren't using SBP */
2428	if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2429	    (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2430		adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2431					 ETH_FCS_LEN;
2432
2433	netdev_dbg(netdev, "changing MTU from %d to %d\n",
2434		   netdev->mtu, new_mtu);
2435	netdev->mtu = new_mtu;
2436
2437	if (netif_running(netdev))
2438		igbvf_up(adapter);
2439	else
2440		igbvf_reset(adapter);
2441
2442	clear_bit(__IGBVF_RESETTING, &adapter->state);
2443
2444	return 0;
2445}
2446
2447static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2448{
2449	switch (cmd) {
2450	default:
2451		return -EOPNOTSUPP;
2452	}
2453}
2454
2455static int igbvf_suspend(struct device *dev_d)
2456{
2457	struct net_device *netdev = dev_get_drvdata(dev_d);
2458	struct igbvf_adapter *adapter = netdev_priv(netdev);
2459
2460	netif_device_detach(netdev);
2461
2462	if (netif_running(netdev)) {
2463		WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2464		igbvf_down(adapter);
2465		igbvf_free_irq(adapter);
2466	}
2467
2468	return 0;
2469}
2470
2471static int __maybe_unused igbvf_resume(struct device *dev_d)
2472{
2473	struct pci_dev *pdev = to_pci_dev(dev_d);
2474	struct net_device *netdev = pci_get_drvdata(pdev);
2475	struct igbvf_adapter *adapter = netdev_priv(netdev);
2476	u32 err;
2477
2478	pci_set_master(pdev);
2479
2480	if (netif_running(netdev)) {
2481		err = igbvf_request_irq(adapter);
2482		if (err)
2483			return err;
2484	}
2485
2486	igbvf_reset(adapter);
2487
2488	if (netif_running(netdev))
2489		igbvf_up(adapter);
2490
2491	netif_device_attach(netdev);
2492
2493	return 0;
2494}
2495
2496static void igbvf_shutdown(struct pci_dev *pdev)
2497{
2498	igbvf_suspend(&pdev->dev);
2499}
2500
2501#ifdef CONFIG_NET_POLL_CONTROLLER
2502/* Polling 'interrupt' - used by things like netconsole to send skbs
2503 * without having to re-enable interrupts. It's not called while
2504 * the interrupt routine is executing.
2505 */
2506static void igbvf_netpoll(struct net_device *netdev)
2507{
2508	struct igbvf_adapter *adapter = netdev_priv(netdev);
2509
2510	disable_irq(adapter->pdev->irq);
2511
2512	igbvf_clean_tx_irq(adapter->tx_ring);
2513
2514	enable_irq(adapter->pdev->irq);
2515}
2516#endif
2517
2518/**
2519 * igbvf_io_error_detected - called when PCI error is detected
2520 * @pdev: Pointer to PCI device
2521 * @state: The current pci connection state
2522 *
2523 * This function is called after a PCI bus error affecting
2524 * this device has been detected.
2525 */
2526static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2527						pci_channel_state_t state)
2528{
2529	struct net_device *netdev = pci_get_drvdata(pdev);
2530	struct igbvf_adapter *adapter = netdev_priv(netdev);
2531
2532	netif_device_detach(netdev);
2533
2534	if (state == pci_channel_io_perm_failure)
2535		return PCI_ERS_RESULT_DISCONNECT;
2536
2537	if (netif_running(netdev))
2538		igbvf_down(adapter);
2539	pci_disable_device(pdev);
2540
2541	/* Request a slot slot reset. */
2542	return PCI_ERS_RESULT_NEED_RESET;
2543}
2544
2545/**
2546 * igbvf_io_slot_reset - called after the pci bus has been reset.
2547 * @pdev: Pointer to PCI device
2548 *
2549 * Restart the card from scratch, as if from a cold-boot. Implementation
2550 * resembles the first-half of the igbvf_resume routine.
2551 */
2552static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2553{
2554	struct net_device *netdev = pci_get_drvdata(pdev);
2555	struct igbvf_adapter *adapter = netdev_priv(netdev);
2556
2557	if (pci_enable_device_mem(pdev)) {
2558		dev_err(&pdev->dev,
2559			"Cannot re-enable PCI device after reset.\n");
2560		return PCI_ERS_RESULT_DISCONNECT;
2561	}
2562	pci_set_master(pdev);
2563
2564	igbvf_reset(adapter);
2565
2566	return PCI_ERS_RESULT_RECOVERED;
2567}
2568
2569/**
2570 * igbvf_io_resume - called when traffic can start flowing again.
2571 * @pdev: Pointer to PCI device
2572 *
2573 * This callback is called when the error recovery driver tells us that
2574 * its OK to resume normal operation. Implementation resembles the
2575 * second-half of the igbvf_resume routine.
2576 */
2577static void igbvf_io_resume(struct pci_dev *pdev)
2578{
2579	struct net_device *netdev = pci_get_drvdata(pdev);
2580	struct igbvf_adapter *adapter = netdev_priv(netdev);
2581
2582	if (netif_running(netdev)) {
2583		if (igbvf_up(adapter)) {
2584			dev_err(&pdev->dev,
2585				"can't bring device back up after reset\n");
2586			return;
2587		}
2588	}
2589
2590	netif_device_attach(netdev);
2591}
2592
2593static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2594{
2595	struct e1000_hw *hw = &adapter->hw;
2596	struct net_device *netdev = adapter->netdev;
2597	struct pci_dev *pdev = adapter->pdev;
2598
2599	if (hw->mac.type == e1000_vfadapt_i350)
2600		dev_info(&pdev->dev, "Intel(R) I350 Virtual Function\n");
2601	else
2602		dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2603	dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2604}
2605
2606static int igbvf_set_features(struct net_device *netdev,
2607			      netdev_features_t features)
2608{
2609	struct igbvf_adapter *adapter = netdev_priv(netdev);
2610
2611	if (features & NETIF_F_RXCSUM)
2612		adapter->flags &= ~IGBVF_FLAG_RX_CSUM_DISABLED;
2613	else
2614		adapter->flags |= IGBVF_FLAG_RX_CSUM_DISABLED;
2615
2616	return 0;
2617}
2618
2619#define IGBVF_MAX_MAC_HDR_LEN		127
2620#define IGBVF_MAX_NETWORK_HDR_LEN	511
2621
2622static netdev_features_t
2623igbvf_features_check(struct sk_buff *skb, struct net_device *dev,
2624		     netdev_features_t features)
2625{
2626	unsigned int network_hdr_len, mac_hdr_len;
2627
2628	/* Make certain the headers can be described by a context descriptor */
2629	mac_hdr_len = skb_network_header(skb) - skb->data;
2630	if (unlikely(mac_hdr_len > IGBVF_MAX_MAC_HDR_LEN))
2631		return features & ~(NETIF_F_HW_CSUM |
2632				    NETIF_F_SCTP_CRC |
2633				    NETIF_F_HW_VLAN_CTAG_TX |
2634				    NETIF_F_TSO |
2635				    NETIF_F_TSO6);
2636
2637	network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
2638	if (unlikely(network_hdr_len >  IGBVF_MAX_NETWORK_HDR_LEN))
2639		return features & ~(NETIF_F_HW_CSUM |
2640				    NETIF_F_SCTP_CRC |
2641				    NETIF_F_TSO |
2642				    NETIF_F_TSO6);
2643
2644	/* We can only support IPV4 TSO in tunnels if we can mangle the
2645	 * inner IP ID field, so strip TSO if MANGLEID is not supported.
2646	 */
2647	if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
2648		features &= ~NETIF_F_TSO;
2649
2650	return features;
2651}
2652
2653static const struct net_device_ops igbvf_netdev_ops = {
2654	.ndo_open		= igbvf_open,
2655	.ndo_stop		= igbvf_close,
2656	.ndo_start_xmit		= igbvf_xmit_frame,
2657	.ndo_set_rx_mode	= igbvf_set_rx_mode,
2658	.ndo_set_mac_address	= igbvf_set_mac,
2659	.ndo_change_mtu		= igbvf_change_mtu,
2660	.ndo_do_ioctl		= igbvf_ioctl,
2661	.ndo_tx_timeout		= igbvf_tx_timeout,
2662	.ndo_vlan_rx_add_vid	= igbvf_vlan_rx_add_vid,
2663	.ndo_vlan_rx_kill_vid	= igbvf_vlan_rx_kill_vid,
2664#ifdef CONFIG_NET_POLL_CONTROLLER
2665	.ndo_poll_controller	= igbvf_netpoll,
2666#endif
2667	.ndo_set_features	= igbvf_set_features,
2668	.ndo_features_check	= igbvf_features_check,
2669};
2670
2671/**
2672 * igbvf_probe - Device Initialization Routine
2673 * @pdev: PCI device information struct
2674 * @ent: entry in igbvf_pci_tbl
2675 *
2676 * Returns 0 on success, negative on failure
2677 *
2678 * igbvf_probe initializes an adapter identified by a pci_dev structure.
2679 * The OS initialization, configuring of the adapter private structure,
2680 * and a hardware reset occur.
2681 **/
2682static int igbvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2683{
2684	struct net_device *netdev;
2685	struct igbvf_adapter *adapter;
2686	struct e1000_hw *hw;
2687	const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2688
2689	static int cards_found;
2690	int err, pci_using_dac;
2691
2692	err = pci_enable_device_mem(pdev);
2693	if (err)
2694		return err;
2695
2696	pci_using_dac = 0;
2697	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
2698	if (!err) {
2699		pci_using_dac = 1;
2700	} else {
2701		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
2702		if (err) {
2703			dev_err(&pdev->dev,
2704				"No usable DMA configuration, aborting\n");
2705			goto err_dma;
2706		}
2707	}
2708
2709	err = pci_request_regions(pdev, igbvf_driver_name);
2710	if (err)
2711		goto err_pci_reg;
2712
2713	pci_set_master(pdev);
2714
2715	err = -ENOMEM;
2716	netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2717	if (!netdev)
2718		goto err_alloc_etherdev;
2719
2720	SET_NETDEV_DEV(netdev, &pdev->dev);
2721
2722	pci_set_drvdata(pdev, netdev);
2723	adapter = netdev_priv(netdev);
2724	hw = &adapter->hw;
2725	adapter->netdev = netdev;
2726	adapter->pdev = pdev;
2727	adapter->ei = ei;
2728	adapter->pba = ei->pba;
2729	adapter->flags = ei->flags;
2730	adapter->hw.back = adapter;
2731	adapter->hw.mac.type = ei->mac;
2732	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2733
2734	/* PCI config space info */
2735
2736	hw->vendor_id = pdev->vendor;
2737	hw->device_id = pdev->device;
2738	hw->subsystem_vendor_id = pdev->subsystem_vendor;
2739	hw->subsystem_device_id = pdev->subsystem_device;
2740	hw->revision_id = pdev->revision;
2741
2742	err = -EIO;
2743	adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2744				      pci_resource_len(pdev, 0));
2745
2746	if (!adapter->hw.hw_addr)
2747		goto err_ioremap;
2748
2749	if (ei->get_variants) {
2750		err = ei->get_variants(adapter);
2751		if (err)
2752			goto err_get_variants;
2753	}
2754
2755	/* setup adapter struct */
2756	err = igbvf_sw_init(adapter);
2757	if (err)
2758		goto err_sw_init;
2759
2760	/* construct the net_device struct */
2761	netdev->netdev_ops = &igbvf_netdev_ops;
2762
2763	igbvf_set_ethtool_ops(netdev);
2764	netdev->watchdog_timeo = 5 * HZ;
2765	strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2766
2767	adapter->bd_number = cards_found++;
2768
2769	netdev->hw_features = NETIF_F_SG |
2770			      NETIF_F_TSO |
2771			      NETIF_F_TSO6 |
2772			      NETIF_F_RXCSUM |
2773			      NETIF_F_HW_CSUM |
2774			      NETIF_F_SCTP_CRC;
2775
2776#define IGBVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
2777				    NETIF_F_GSO_GRE_CSUM | \
2778				    NETIF_F_GSO_IPXIP4 | \
2779				    NETIF_F_GSO_IPXIP6 | \
2780				    NETIF_F_GSO_UDP_TUNNEL | \
2781				    NETIF_F_GSO_UDP_TUNNEL_CSUM)
2782
2783	netdev->gso_partial_features = IGBVF_GSO_PARTIAL_FEATURES;
2784	netdev->hw_features |= NETIF_F_GSO_PARTIAL |
2785			       IGBVF_GSO_PARTIAL_FEATURES;
2786
2787	netdev->features = netdev->hw_features;
2788
2789	if (pci_using_dac)
2790		netdev->features |= NETIF_F_HIGHDMA;
2791
2792	netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
2793	netdev->mpls_features |= NETIF_F_HW_CSUM;
2794	netdev->hw_enc_features |= netdev->vlan_features;
2795
2796	/* set this bit last since it cannot be part of vlan_features */
2797	netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
2798			    NETIF_F_HW_VLAN_CTAG_RX |
2799			    NETIF_F_HW_VLAN_CTAG_TX;
2800
2801	/* MTU range: 68 - 9216 */
2802	netdev->min_mtu = ETH_MIN_MTU;
2803	netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
2804
2805	spin_lock_bh(&hw->mbx_lock);
2806
2807	/*reset the controller to put the device in a known good state */
2808	err = hw->mac.ops.reset_hw(hw);
2809	if (err) {
2810		dev_info(&pdev->dev,
2811			 "PF still in reset state. Is the PF interface up?\n");
2812	} else {
2813		err = hw->mac.ops.read_mac_addr(hw);
2814		if (err)
2815			dev_info(&pdev->dev, "Error reading MAC address.\n");
2816		else if (is_zero_ether_addr(adapter->hw.mac.addr))
2817			dev_info(&pdev->dev,
2818				 "MAC address not assigned by administrator.\n");
2819		memcpy(netdev->dev_addr, adapter->hw.mac.addr,
2820		       netdev->addr_len);
2821	}
2822
2823	spin_unlock_bh(&hw->mbx_lock);
2824
2825	if (!is_valid_ether_addr(netdev->dev_addr)) {
2826		dev_info(&pdev->dev, "Assigning random MAC address.\n");
2827		eth_hw_addr_random(netdev);
2828		memcpy(adapter->hw.mac.addr, netdev->dev_addr,
2829		       netdev->addr_len);
2830	}
2831
2832	timer_setup(&adapter->watchdog_timer, igbvf_watchdog, 0);
2833
2834	INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2835	INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2836
2837	/* ring size defaults */
2838	adapter->rx_ring->count = 1024;
2839	adapter->tx_ring->count = 1024;
2840
2841	/* reset the hardware with the new settings */
2842	igbvf_reset(adapter);
2843
2844	/* set hardware-specific flags */
2845	if (adapter->hw.mac.type == e1000_vfadapt_i350)
2846		adapter->flags |= IGBVF_FLAG_RX_LB_VLAN_BSWAP;
2847
2848	strcpy(netdev->name, "eth%d");
2849	err = register_netdev(netdev);
2850	if (err)
2851		goto err_hw_init;
2852
2853	/* tell the stack to leave us alone until igbvf_open() is called */
2854	netif_carrier_off(netdev);
2855	netif_stop_queue(netdev);
2856
2857	igbvf_print_device_info(adapter);
2858
2859	igbvf_initialize_last_counter_stats(adapter);
2860
2861	return 0;
2862
2863err_hw_init:
2864	kfree(adapter->tx_ring);
2865	kfree(adapter->rx_ring);
2866err_sw_init:
2867	igbvf_reset_interrupt_capability(adapter);
2868err_get_variants:
2869	iounmap(adapter->hw.hw_addr);
2870err_ioremap:
2871	free_netdev(netdev);
2872err_alloc_etherdev:
2873	pci_release_regions(pdev);
2874err_pci_reg:
2875err_dma:
2876	pci_disable_device(pdev);
2877	return err;
2878}
2879
2880/**
2881 * igbvf_remove - Device Removal Routine
2882 * @pdev: PCI device information struct
2883 *
2884 * igbvf_remove is called by the PCI subsystem to alert the driver
2885 * that it should release a PCI device.  The could be caused by a
2886 * Hot-Plug event, or because the driver is going to be removed from
2887 * memory.
2888 **/
2889static void igbvf_remove(struct pci_dev *pdev)
2890{
2891	struct net_device *netdev = pci_get_drvdata(pdev);
2892	struct igbvf_adapter *adapter = netdev_priv(netdev);
2893	struct e1000_hw *hw = &adapter->hw;
2894
2895	/* The watchdog timer may be rescheduled, so explicitly
2896	 * disable it from being rescheduled.
2897	 */
2898	set_bit(__IGBVF_DOWN, &adapter->state);
2899	del_timer_sync(&adapter->watchdog_timer);
2900
2901	cancel_work_sync(&adapter->reset_task);
2902	cancel_work_sync(&adapter->watchdog_task);
2903
2904	unregister_netdev(netdev);
2905
2906	igbvf_reset_interrupt_capability(adapter);
2907
2908	/* it is important to delete the NAPI struct prior to freeing the
2909	 * Rx ring so that you do not end up with null pointer refs
2910	 */
2911	netif_napi_del(&adapter->rx_ring->napi);
2912	kfree(adapter->tx_ring);
2913	kfree(adapter->rx_ring);
2914
2915	iounmap(hw->hw_addr);
2916	if (hw->flash_address)
2917		iounmap(hw->flash_address);
2918	pci_release_regions(pdev);
2919
2920	free_netdev(netdev);
2921
2922	pci_disable_device(pdev);
2923}
2924
2925/* PCI Error Recovery (ERS) */
2926static const struct pci_error_handlers igbvf_err_handler = {
2927	.error_detected = igbvf_io_error_detected,
2928	.slot_reset = igbvf_io_slot_reset,
2929	.resume = igbvf_io_resume,
2930};
2931
2932static const struct pci_device_id igbvf_pci_tbl[] = {
2933	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2934	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf },
2935	{ } /* terminate list */
2936};
2937MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2938
2939static SIMPLE_DEV_PM_OPS(igbvf_pm_ops, igbvf_suspend, igbvf_resume);
2940
2941/* PCI Device API Driver */
2942static struct pci_driver igbvf_driver = {
2943	.name		= igbvf_driver_name,
2944	.id_table	= igbvf_pci_tbl,
2945	.probe		= igbvf_probe,
2946	.remove		= igbvf_remove,
2947	.driver.pm	= &igbvf_pm_ops,
2948	.shutdown	= igbvf_shutdown,
2949	.err_handler	= &igbvf_err_handler
2950};
2951
2952/**
2953 * igbvf_init_module - Driver Registration Routine
2954 *
2955 * igbvf_init_module is the first routine called when the driver is
2956 * loaded. All it does is register with the PCI subsystem.
2957 **/
2958static int __init igbvf_init_module(void)
2959{
2960	int ret;
2961
2962	pr_info("%s\n", igbvf_driver_string);
2963	pr_info("%s\n", igbvf_copyright);
2964
2965	ret = pci_register_driver(&igbvf_driver);
2966
2967	return ret;
2968}
2969module_init(igbvf_init_module);
2970
2971/**
2972 * igbvf_exit_module - Driver Exit Cleanup Routine
2973 *
2974 * igbvf_exit_module is called just before the driver is removed
2975 * from memory.
2976 **/
2977static void __exit igbvf_exit_module(void)
2978{
2979	pci_unregister_driver(&igbvf_driver);
2980}
2981module_exit(igbvf_exit_module);
2982
2983MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2984MODULE_DESCRIPTION("Intel(R) Gigabit Virtual Function Network Driver");
2985MODULE_LICENSE("GPL v2");
2986
2987/* netdev.c */