Loading...
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/bitfield.h>
7#include <linux/delay.h>
8#include <linux/ethtool.h>
9#include <linux/if_vlan.h>
10#include <linux/init.h>
11#include <linux/ipv6.h>
12#include <linux/mii.h>
13#include <linux/module.h>
14#include <linux/netdevice.h>
15#include <linux/pagemap.h>
16#include <linux/pci.h>
17#include <linux/prefetch.h>
18#include <linux/sctp.h>
19#include <linux/slab.h>
20#include <linux/tcp.h>
21#include <linux/types.h>
22#include <linux/vmalloc.h>
23#include <net/checksum.h>
24#include <net/ip6_checksum.h>
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_get_bits(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info,
277 E1000_RXDADV_HDRBUFLEN_MASK);
278 if (hlen > adapter->rx_ps_hdr_size)
279 hlen = adapter->rx_ps_hdr_size;
280
281 length = le16_to_cpu(rx_desc->wb.upper.length);
282 cleaned = true;
283 cleaned_count++;
284
285 skb = buffer_info->skb;
286 prefetch(skb->data - NET_IP_ALIGN);
287 buffer_info->skb = NULL;
288 if (!adapter->rx_ps_hdr_size) {
289 dma_unmap_single(&pdev->dev, buffer_info->dma,
290 adapter->rx_buffer_len,
291 DMA_FROM_DEVICE);
292 buffer_info->dma = 0;
293 skb_put(skb, length);
294 goto send_up;
295 }
296
297 if (!skb_shinfo(skb)->nr_frags) {
298 dma_unmap_single(&pdev->dev, buffer_info->dma,
299 adapter->rx_ps_hdr_size,
300 DMA_FROM_DEVICE);
301 buffer_info->dma = 0;
302 skb_put(skb, hlen);
303 }
304
305 if (length) {
306 dma_unmap_page(&pdev->dev, buffer_info->page_dma,
307 PAGE_SIZE / 2,
308 DMA_FROM_DEVICE);
309 buffer_info->page_dma = 0;
310
311 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
312 buffer_info->page,
313 buffer_info->page_offset,
314 length);
315
316 if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
317 (page_count(buffer_info->page) != 1))
318 buffer_info->page = NULL;
319 else
320 get_page(buffer_info->page);
321
322 skb->len += length;
323 skb->data_len += length;
324 skb->truesize += PAGE_SIZE / 2;
325 }
326send_up:
327 i++;
328 if (i == rx_ring->count)
329 i = 0;
330 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
331 prefetch(next_rxd);
332 next_buffer = &rx_ring->buffer_info[i];
333
334 if (!(staterr & E1000_RXD_STAT_EOP)) {
335 buffer_info->skb = next_buffer->skb;
336 buffer_info->dma = next_buffer->dma;
337 next_buffer->skb = skb;
338 next_buffer->dma = 0;
339 goto next_desc;
340 }
341
342 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
343 dev_kfree_skb_irq(skb);
344 goto next_desc;
345 }
346
347 total_bytes += skb->len;
348 total_packets++;
349
350 igbvf_rx_checksum_adv(adapter, staterr, skb);
351
352 skb->protocol = eth_type_trans(skb, netdev);
353
354 igbvf_receive_skb(adapter, netdev, skb, staterr,
355 rx_desc->wb.upper.vlan);
356
357next_desc:
358 rx_desc->wb.upper.status_error = 0;
359
360 /* return some buffers to hardware, one at a time is too slow */
361 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
362 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
363 cleaned_count = 0;
364 }
365
366 /* use prefetched values */
367 rx_desc = next_rxd;
368 buffer_info = next_buffer;
369
370 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
371 }
372
373 rx_ring->next_to_clean = i;
374 cleaned_count = igbvf_desc_unused(rx_ring);
375
376 if (cleaned_count)
377 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
378
379 adapter->total_rx_packets += total_packets;
380 adapter->total_rx_bytes += total_bytes;
381 netdev->stats.rx_bytes += total_bytes;
382 netdev->stats.rx_packets += total_packets;
383 return cleaned;
384}
385
386static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
387 struct igbvf_buffer *buffer_info)
388{
389 if (buffer_info->dma) {
390 if (buffer_info->mapped_as_page)
391 dma_unmap_page(&adapter->pdev->dev,
392 buffer_info->dma,
393 buffer_info->length,
394 DMA_TO_DEVICE);
395 else
396 dma_unmap_single(&adapter->pdev->dev,
397 buffer_info->dma,
398 buffer_info->length,
399 DMA_TO_DEVICE);
400 buffer_info->dma = 0;
401 }
402 if (buffer_info->skb) {
403 dev_kfree_skb_any(buffer_info->skb);
404 buffer_info->skb = NULL;
405 }
406 buffer_info->time_stamp = 0;
407}
408
409/**
410 * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
411 * @adapter: board private structure
412 * @tx_ring: ring being initialized
413 *
414 * Return 0 on success, negative on failure
415 **/
416int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
417 struct igbvf_ring *tx_ring)
418{
419 struct pci_dev *pdev = adapter->pdev;
420 int size;
421
422 size = sizeof(struct igbvf_buffer) * tx_ring->count;
423 tx_ring->buffer_info = vzalloc(size);
424 if (!tx_ring->buffer_info)
425 goto err;
426
427 /* round up to nearest 4K */
428 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
429 tx_ring->size = ALIGN(tx_ring->size, 4096);
430
431 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
432 &tx_ring->dma, GFP_KERNEL);
433 if (!tx_ring->desc)
434 goto err;
435
436 tx_ring->adapter = adapter;
437 tx_ring->next_to_use = 0;
438 tx_ring->next_to_clean = 0;
439
440 return 0;
441err:
442 vfree(tx_ring->buffer_info);
443 dev_err(&adapter->pdev->dev,
444 "Unable to allocate memory for the transmit descriptor ring\n");
445 return -ENOMEM;
446}
447
448/**
449 * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
450 * @adapter: board private structure
451 * @rx_ring: ring being initialized
452 *
453 * Returns 0 on success, negative on failure
454 **/
455int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
456 struct igbvf_ring *rx_ring)
457{
458 struct pci_dev *pdev = adapter->pdev;
459 int size, desc_len;
460
461 size = sizeof(struct igbvf_buffer) * rx_ring->count;
462 rx_ring->buffer_info = vzalloc(size);
463 if (!rx_ring->buffer_info)
464 goto err;
465
466 desc_len = sizeof(union e1000_adv_rx_desc);
467
468 /* Round up to nearest 4K */
469 rx_ring->size = rx_ring->count * desc_len;
470 rx_ring->size = ALIGN(rx_ring->size, 4096);
471
472 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
473 &rx_ring->dma, GFP_KERNEL);
474 if (!rx_ring->desc)
475 goto err;
476
477 rx_ring->next_to_clean = 0;
478 rx_ring->next_to_use = 0;
479
480 rx_ring->adapter = adapter;
481
482 return 0;
483
484err:
485 vfree(rx_ring->buffer_info);
486 rx_ring->buffer_info = NULL;
487 dev_err(&adapter->pdev->dev,
488 "Unable to allocate memory for the receive descriptor ring\n");
489 return -ENOMEM;
490}
491
492/**
493 * igbvf_clean_tx_ring - Free Tx Buffers
494 * @tx_ring: ring to be cleaned
495 **/
496static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
497{
498 struct igbvf_adapter *adapter = tx_ring->adapter;
499 struct igbvf_buffer *buffer_info;
500 unsigned long size;
501 unsigned int i;
502
503 if (!tx_ring->buffer_info)
504 return;
505
506 /* Free all the Tx ring sk_buffs */
507 for (i = 0; i < tx_ring->count; i++) {
508 buffer_info = &tx_ring->buffer_info[i];
509 igbvf_put_txbuf(adapter, buffer_info);
510 }
511
512 size = sizeof(struct igbvf_buffer) * tx_ring->count;
513 memset(tx_ring->buffer_info, 0, size);
514
515 /* Zero out the descriptor ring */
516 memset(tx_ring->desc, 0, tx_ring->size);
517
518 tx_ring->next_to_use = 0;
519 tx_ring->next_to_clean = 0;
520
521 writel(0, adapter->hw.hw_addr + tx_ring->head);
522 writel(0, adapter->hw.hw_addr + tx_ring->tail);
523}
524
525/**
526 * igbvf_free_tx_resources - Free Tx Resources per Queue
527 * @tx_ring: ring to free resources from
528 *
529 * Free all transmit software resources
530 **/
531void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
532{
533 struct pci_dev *pdev = tx_ring->adapter->pdev;
534
535 igbvf_clean_tx_ring(tx_ring);
536
537 vfree(tx_ring->buffer_info);
538 tx_ring->buffer_info = NULL;
539
540 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
541 tx_ring->dma);
542
543 tx_ring->desc = NULL;
544}
545
546/**
547 * igbvf_clean_rx_ring - Free Rx Buffers per Queue
548 * @rx_ring: ring structure pointer to free buffers from
549 **/
550static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
551{
552 struct igbvf_adapter *adapter = rx_ring->adapter;
553 struct igbvf_buffer *buffer_info;
554 struct pci_dev *pdev = adapter->pdev;
555 unsigned long size;
556 unsigned int i;
557
558 if (!rx_ring->buffer_info)
559 return;
560
561 /* Free all the Rx ring sk_buffs */
562 for (i = 0; i < rx_ring->count; i++) {
563 buffer_info = &rx_ring->buffer_info[i];
564 if (buffer_info->dma) {
565 if (adapter->rx_ps_hdr_size) {
566 dma_unmap_single(&pdev->dev, buffer_info->dma,
567 adapter->rx_ps_hdr_size,
568 DMA_FROM_DEVICE);
569 } else {
570 dma_unmap_single(&pdev->dev, buffer_info->dma,
571 adapter->rx_buffer_len,
572 DMA_FROM_DEVICE);
573 }
574 buffer_info->dma = 0;
575 }
576
577 if (buffer_info->skb) {
578 dev_kfree_skb(buffer_info->skb);
579 buffer_info->skb = NULL;
580 }
581
582 if (buffer_info->page) {
583 if (buffer_info->page_dma)
584 dma_unmap_page(&pdev->dev,
585 buffer_info->page_dma,
586 PAGE_SIZE / 2,
587 DMA_FROM_DEVICE);
588 put_page(buffer_info->page);
589 buffer_info->page = NULL;
590 buffer_info->page_dma = 0;
591 buffer_info->page_offset = 0;
592 }
593 }
594
595 size = sizeof(struct igbvf_buffer) * rx_ring->count;
596 memset(rx_ring->buffer_info, 0, size);
597
598 /* Zero out the descriptor ring */
599 memset(rx_ring->desc, 0, rx_ring->size);
600
601 rx_ring->next_to_clean = 0;
602 rx_ring->next_to_use = 0;
603
604 writel(0, adapter->hw.hw_addr + rx_ring->head);
605 writel(0, adapter->hw.hw_addr + rx_ring->tail);
606}
607
608/**
609 * igbvf_free_rx_resources - Free Rx Resources
610 * @rx_ring: ring to clean the resources from
611 *
612 * Free all receive software resources
613 **/
614
615void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
616{
617 struct pci_dev *pdev = rx_ring->adapter->pdev;
618
619 igbvf_clean_rx_ring(rx_ring);
620
621 vfree(rx_ring->buffer_info);
622 rx_ring->buffer_info = NULL;
623
624 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
625 rx_ring->dma);
626 rx_ring->desc = NULL;
627}
628
629/**
630 * igbvf_update_itr - update the dynamic ITR value based on statistics
631 * @adapter: pointer to adapter
632 * @itr_setting: current adapter->itr
633 * @packets: the number of packets during this measurement interval
634 * @bytes: the number of bytes during this measurement interval
635 *
636 * Stores a new ITR value based on packets and byte counts during the last
637 * interrupt. The advantage of per interrupt computation is faster updates
638 * and more accurate ITR for the current traffic pattern. Constants in this
639 * function were computed based on theoretical maximum wire speed and thresholds
640 * were set based on testing data as well as attempting to minimize response
641 * time while increasing bulk throughput.
642 **/
643static enum latency_range igbvf_update_itr(struct igbvf_adapter *adapter,
644 enum latency_range itr_setting,
645 int packets, int bytes)
646{
647 enum latency_range retval = itr_setting;
648
649 if (packets == 0)
650 goto update_itr_done;
651
652 switch (itr_setting) {
653 case lowest_latency:
654 /* handle TSO and jumbo frames */
655 if (bytes/packets > 8000)
656 retval = bulk_latency;
657 else if ((packets < 5) && (bytes > 512))
658 retval = low_latency;
659 break;
660 case low_latency: /* 50 usec aka 20000 ints/s */
661 if (bytes > 10000) {
662 /* this if handles the TSO accounting */
663 if (bytes/packets > 8000)
664 retval = bulk_latency;
665 else if ((packets < 10) || ((bytes/packets) > 1200))
666 retval = bulk_latency;
667 else if ((packets > 35))
668 retval = lowest_latency;
669 } else if (bytes/packets > 2000) {
670 retval = bulk_latency;
671 } else if (packets <= 2 && bytes < 512) {
672 retval = lowest_latency;
673 }
674 break;
675 case bulk_latency: /* 250 usec aka 4000 ints/s */
676 if (bytes > 25000) {
677 if (packets > 35)
678 retval = low_latency;
679 } else if (bytes < 6000) {
680 retval = low_latency;
681 }
682 break;
683 default:
684 break;
685 }
686
687update_itr_done:
688 return retval;
689}
690
691static int igbvf_range_to_itr(enum latency_range current_range)
692{
693 int new_itr;
694
695 switch (current_range) {
696 /* counts and packets in update_itr are dependent on these numbers */
697 case lowest_latency:
698 new_itr = IGBVF_70K_ITR;
699 break;
700 case low_latency:
701 new_itr = IGBVF_20K_ITR;
702 break;
703 case bulk_latency:
704 new_itr = IGBVF_4K_ITR;
705 break;
706 default:
707 new_itr = IGBVF_START_ITR;
708 break;
709 }
710 return new_itr;
711}
712
713static void igbvf_set_itr(struct igbvf_adapter *adapter)
714{
715 u32 new_itr;
716
717 adapter->tx_ring->itr_range =
718 igbvf_update_itr(adapter,
719 adapter->tx_ring->itr_val,
720 adapter->total_tx_packets,
721 adapter->total_tx_bytes);
722
723 /* conservative mode (itr 3) eliminates the lowest_latency setting */
724 if (adapter->requested_itr == 3 &&
725 adapter->tx_ring->itr_range == lowest_latency)
726 adapter->tx_ring->itr_range = low_latency;
727
728 new_itr = igbvf_range_to_itr(adapter->tx_ring->itr_range);
729
730 if (new_itr != adapter->tx_ring->itr_val) {
731 u32 current_itr = adapter->tx_ring->itr_val;
732 /* this attempts to bias the interrupt rate towards Bulk
733 * by adding intermediate steps when interrupt rate is
734 * increasing
735 */
736 new_itr = new_itr > current_itr ?
737 min(current_itr + (new_itr >> 2), new_itr) :
738 new_itr;
739 adapter->tx_ring->itr_val = new_itr;
740
741 adapter->tx_ring->set_itr = 1;
742 }
743
744 adapter->rx_ring->itr_range =
745 igbvf_update_itr(adapter, adapter->rx_ring->itr_val,
746 adapter->total_rx_packets,
747 adapter->total_rx_bytes);
748 if (adapter->requested_itr == 3 &&
749 adapter->rx_ring->itr_range == lowest_latency)
750 adapter->rx_ring->itr_range = low_latency;
751
752 new_itr = igbvf_range_to_itr(adapter->rx_ring->itr_range);
753
754 if (new_itr != adapter->rx_ring->itr_val) {
755 u32 current_itr = adapter->rx_ring->itr_val;
756
757 new_itr = new_itr > current_itr ?
758 min(current_itr + (new_itr >> 2), new_itr) :
759 new_itr;
760 adapter->rx_ring->itr_val = new_itr;
761
762 adapter->rx_ring->set_itr = 1;
763 }
764}
765
766/**
767 * igbvf_clean_tx_irq - Reclaim resources after transmit completes
768 * @tx_ring: ring structure to clean descriptors from
769 *
770 * returns true if ring is completely cleaned
771 **/
772static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
773{
774 struct igbvf_adapter *adapter = tx_ring->adapter;
775 struct net_device *netdev = adapter->netdev;
776 struct igbvf_buffer *buffer_info;
777 struct sk_buff *skb;
778 union e1000_adv_tx_desc *tx_desc, *eop_desc;
779 unsigned int total_bytes = 0, total_packets = 0;
780 unsigned int i, count = 0;
781 bool cleaned = false;
782
783 i = tx_ring->next_to_clean;
784 buffer_info = &tx_ring->buffer_info[i];
785 eop_desc = buffer_info->next_to_watch;
786
787 do {
788 /* if next_to_watch is not set then there is no work pending */
789 if (!eop_desc)
790 break;
791
792 /* prevent any other reads prior to eop_desc */
793 smp_rmb();
794
795 /* if DD is not set pending work has not been completed */
796 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
797 break;
798
799 /* clear next_to_watch to prevent false hangs */
800 buffer_info->next_to_watch = NULL;
801
802 for (cleaned = false; !cleaned; count++) {
803 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
804 cleaned = (tx_desc == eop_desc);
805 skb = buffer_info->skb;
806
807 if (skb) {
808 unsigned int segs, bytecount;
809
810 /* gso_segs is currently only valid for tcp */
811 segs = skb_shinfo(skb)->gso_segs ?: 1;
812 /* multiply data chunks by size of headers */
813 bytecount = ((segs - 1) * skb_headlen(skb)) +
814 skb->len;
815 total_packets += segs;
816 total_bytes += bytecount;
817 }
818
819 igbvf_put_txbuf(adapter, buffer_info);
820 tx_desc->wb.status = 0;
821
822 i++;
823 if (i == tx_ring->count)
824 i = 0;
825
826 buffer_info = &tx_ring->buffer_info[i];
827 }
828
829 eop_desc = buffer_info->next_to_watch;
830 } while (count < tx_ring->count);
831
832 tx_ring->next_to_clean = i;
833
834 if (unlikely(count && netif_carrier_ok(netdev) &&
835 igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
836 /* Make sure that anybody stopping the queue after this
837 * sees the new next_to_clean.
838 */
839 smp_mb();
840 if (netif_queue_stopped(netdev) &&
841 !(test_bit(__IGBVF_DOWN, &adapter->state))) {
842 netif_wake_queue(netdev);
843 ++adapter->restart_queue;
844 }
845 }
846
847 netdev->stats.tx_bytes += total_bytes;
848 netdev->stats.tx_packets += total_packets;
849 return count < tx_ring->count;
850}
851
852static irqreturn_t igbvf_msix_other(int irq, void *data)
853{
854 struct net_device *netdev = data;
855 struct igbvf_adapter *adapter = netdev_priv(netdev);
856 struct e1000_hw *hw = &adapter->hw;
857
858 adapter->int_counter1++;
859
860 hw->mac.get_link_status = 1;
861 if (!test_bit(__IGBVF_DOWN, &adapter->state))
862 mod_timer(&adapter->watchdog_timer, jiffies + 1);
863
864 ew32(EIMS, adapter->eims_other);
865
866 return IRQ_HANDLED;
867}
868
869static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
870{
871 struct net_device *netdev = data;
872 struct igbvf_adapter *adapter = netdev_priv(netdev);
873 struct e1000_hw *hw = &adapter->hw;
874 struct igbvf_ring *tx_ring = adapter->tx_ring;
875
876 if (tx_ring->set_itr) {
877 writel(tx_ring->itr_val,
878 adapter->hw.hw_addr + tx_ring->itr_register);
879 adapter->tx_ring->set_itr = 0;
880 }
881
882 adapter->total_tx_bytes = 0;
883 adapter->total_tx_packets = 0;
884
885 /* auto mask will automatically re-enable the interrupt when we write
886 * EICS
887 */
888 if (!igbvf_clean_tx_irq(tx_ring))
889 /* Ring was not completely cleaned, so fire another interrupt */
890 ew32(EICS, tx_ring->eims_value);
891 else
892 ew32(EIMS, tx_ring->eims_value);
893
894 return IRQ_HANDLED;
895}
896
897static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
898{
899 struct net_device *netdev = data;
900 struct igbvf_adapter *adapter = netdev_priv(netdev);
901
902 adapter->int_counter0++;
903
904 /* Write the ITR value calculated at the end of the
905 * previous interrupt.
906 */
907 if (adapter->rx_ring->set_itr) {
908 writel(adapter->rx_ring->itr_val,
909 adapter->hw.hw_addr + adapter->rx_ring->itr_register);
910 adapter->rx_ring->set_itr = 0;
911 }
912
913 if (napi_schedule_prep(&adapter->rx_ring->napi)) {
914 adapter->total_rx_bytes = 0;
915 adapter->total_rx_packets = 0;
916 __napi_schedule(&adapter->rx_ring->napi);
917 }
918
919 return IRQ_HANDLED;
920}
921
922#define IGBVF_NO_QUEUE -1
923
924static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
925 int tx_queue, int msix_vector)
926{
927 struct e1000_hw *hw = &adapter->hw;
928 u32 ivar, index;
929
930 /* 82576 uses a table-based method for assigning vectors.
931 * Each queue has a single entry in the table to which we write
932 * a vector number along with a "valid" bit. Sadly, the layout
933 * of the table is somewhat counterintuitive.
934 */
935 if (rx_queue > IGBVF_NO_QUEUE) {
936 index = (rx_queue >> 1);
937 ivar = array_er32(IVAR0, index);
938 if (rx_queue & 0x1) {
939 /* vector goes into third byte of register */
940 ivar = ivar & 0xFF00FFFF;
941 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
942 } else {
943 /* vector goes into low byte of register */
944 ivar = ivar & 0xFFFFFF00;
945 ivar |= msix_vector | E1000_IVAR_VALID;
946 }
947 adapter->rx_ring[rx_queue].eims_value = BIT(msix_vector);
948 array_ew32(IVAR0, index, ivar);
949 }
950 if (tx_queue > IGBVF_NO_QUEUE) {
951 index = (tx_queue >> 1);
952 ivar = array_er32(IVAR0, index);
953 if (tx_queue & 0x1) {
954 /* vector goes into high byte of register */
955 ivar = ivar & 0x00FFFFFF;
956 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
957 } else {
958 /* vector goes into second byte of register */
959 ivar = ivar & 0xFFFF00FF;
960 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
961 }
962 adapter->tx_ring[tx_queue].eims_value = BIT(msix_vector);
963 array_ew32(IVAR0, index, ivar);
964 }
965}
966
967/**
968 * igbvf_configure_msix - Configure MSI-X hardware
969 * @adapter: board private structure
970 *
971 * igbvf_configure_msix sets up the hardware to properly
972 * generate MSI-X interrupts.
973 **/
974static void igbvf_configure_msix(struct igbvf_adapter *adapter)
975{
976 u32 tmp;
977 struct e1000_hw *hw = &adapter->hw;
978 struct igbvf_ring *tx_ring = adapter->tx_ring;
979 struct igbvf_ring *rx_ring = adapter->rx_ring;
980 int vector = 0;
981
982 adapter->eims_enable_mask = 0;
983
984 igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
985 adapter->eims_enable_mask |= tx_ring->eims_value;
986 writel(tx_ring->itr_val, hw->hw_addr + tx_ring->itr_register);
987 igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
988 adapter->eims_enable_mask |= rx_ring->eims_value;
989 writel(rx_ring->itr_val, hw->hw_addr + rx_ring->itr_register);
990
991 /* set vector for other causes, i.e. link changes */
992
993 tmp = (vector++ | E1000_IVAR_VALID);
994
995 ew32(IVAR_MISC, tmp);
996
997 adapter->eims_enable_mask = GENMASK(vector - 1, 0);
998 adapter->eims_other = BIT(vector - 1);
999 e1e_flush();
1000}
1001
1002static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
1003{
1004 if (adapter->msix_entries) {
1005 pci_disable_msix(adapter->pdev);
1006 kfree(adapter->msix_entries);
1007 adapter->msix_entries = NULL;
1008 }
1009}
1010
1011/**
1012 * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1013 * @adapter: board private structure
1014 *
1015 * Attempt to configure interrupts using the best available
1016 * capabilities of the hardware and kernel.
1017 **/
1018static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1019{
1020 int err = -ENOMEM;
1021 int i;
1022
1023 /* we allocate 3 vectors, 1 for Tx, 1 for Rx, one for PF messages */
1024 adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1025 GFP_KERNEL);
1026 if (adapter->msix_entries) {
1027 for (i = 0; i < 3; i++)
1028 adapter->msix_entries[i].entry = i;
1029
1030 err = pci_enable_msix_range(adapter->pdev,
1031 adapter->msix_entries, 3, 3);
1032 }
1033
1034 if (err < 0) {
1035 /* MSI-X failed */
1036 dev_err(&adapter->pdev->dev,
1037 "Failed to initialize MSI-X interrupts.\n");
1038 igbvf_reset_interrupt_capability(adapter);
1039 }
1040}
1041
1042/**
1043 * igbvf_request_msix - Initialize MSI-X interrupts
1044 * @adapter: board private structure
1045 *
1046 * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1047 * kernel.
1048 **/
1049static int igbvf_request_msix(struct igbvf_adapter *adapter)
1050{
1051 struct net_device *netdev = adapter->netdev;
1052 int err = 0, vector = 0;
1053
1054 if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1055 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1056 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1057 } else {
1058 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1059 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1060 }
1061
1062 err = request_irq(adapter->msix_entries[vector].vector,
1063 igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1064 netdev);
1065 if (err)
1066 goto out;
1067
1068 adapter->tx_ring->itr_register = E1000_EITR(vector);
1069 adapter->tx_ring->itr_val = adapter->current_itr;
1070 vector++;
1071
1072 err = request_irq(adapter->msix_entries[vector].vector,
1073 igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1074 netdev);
1075 if (err)
1076 goto free_irq_tx;
1077
1078 adapter->rx_ring->itr_register = E1000_EITR(vector);
1079 adapter->rx_ring->itr_val = adapter->current_itr;
1080 vector++;
1081
1082 err = request_irq(adapter->msix_entries[vector].vector,
1083 igbvf_msix_other, 0, netdev->name, netdev);
1084 if (err)
1085 goto free_irq_rx;
1086
1087 igbvf_configure_msix(adapter);
1088 return 0;
1089free_irq_rx:
1090 free_irq(adapter->msix_entries[--vector].vector, netdev);
1091free_irq_tx:
1092 free_irq(adapter->msix_entries[--vector].vector, netdev);
1093out:
1094 return err;
1095}
1096
1097/**
1098 * igbvf_alloc_queues - Allocate memory for all rings
1099 * @adapter: board private structure to initialize
1100 **/
1101static int igbvf_alloc_queues(struct igbvf_adapter *adapter)
1102{
1103 struct net_device *netdev = adapter->netdev;
1104
1105 adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1106 if (!adapter->tx_ring)
1107 return -ENOMEM;
1108
1109 adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1110 if (!adapter->rx_ring) {
1111 kfree(adapter->tx_ring);
1112 return -ENOMEM;
1113 }
1114
1115 netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll);
1116
1117 return 0;
1118}
1119
1120/**
1121 * igbvf_request_irq - initialize interrupts
1122 * @adapter: board private structure
1123 *
1124 * Attempts to configure interrupts using the best available
1125 * capabilities of the hardware and kernel.
1126 **/
1127static int igbvf_request_irq(struct igbvf_adapter *adapter)
1128{
1129 int err = -1;
1130
1131 /* igbvf supports msi-x only */
1132 if (adapter->msix_entries)
1133 err = igbvf_request_msix(adapter);
1134
1135 if (!err)
1136 return err;
1137
1138 dev_err(&adapter->pdev->dev,
1139 "Unable to allocate interrupt, Error: %d\n", err);
1140
1141 return err;
1142}
1143
1144static void igbvf_free_irq(struct igbvf_adapter *adapter)
1145{
1146 struct net_device *netdev = adapter->netdev;
1147 int vector;
1148
1149 if (adapter->msix_entries) {
1150 for (vector = 0; vector < 3; vector++)
1151 free_irq(adapter->msix_entries[vector].vector, netdev);
1152 }
1153}
1154
1155/**
1156 * igbvf_irq_disable - Mask off interrupt generation on the NIC
1157 * @adapter: board private structure
1158 **/
1159static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1160{
1161 struct e1000_hw *hw = &adapter->hw;
1162
1163 ew32(EIMC, ~0);
1164
1165 if (adapter->msix_entries)
1166 ew32(EIAC, 0);
1167}
1168
1169/**
1170 * igbvf_irq_enable - Enable default interrupt generation settings
1171 * @adapter: board private structure
1172 **/
1173static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1174{
1175 struct e1000_hw *hw = &adapter->hw;
1176
1177 ew32(EIAC, adapter->eims_enable_mask);
1178 ew32(EIAM, adapter->eims_enable_mask);
1179 ew32(EIMS, adapter->eims_enable_mask);
1180}
1181
1182/**
1183 * igbvf_poll - NAPI Rx polling callback
1184 * @napi: struct associated with this polling callback
1185 * @budget: amount of packets driver is allowed to process this poll
1186 **/
1187static int igbvf_poll(struct napi_struct *napi, int budget)
1188{
1189 struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1190 struct igbvf_adapter *adapter = rx_ring->adapter;
1191 struct e1000_hw *hw = &adapter->hw;
1192 int work_done = 0;
1193
1194 igbvf_clean_rx_irq(adapter, &work_done, budget);
1195
1196 if (work_done == budget)
1197 return budget;
1198
1199 /* Exit the polling mode, but don't re-enable interrupts if stack might
1200 * poll us due to busy-polling
1201 */
1202 if (likely(napi_complete_done(napi, work_done))) {
1203 if (adapter->requested_itr & 3)
1204 igbvf_set_itr(adapter);
1205
1206 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1207 ew32(EIMS, adapter->rx_ring->eims_value);
1208 }
1209
1210 return work_done;
1211}
1212
1213/**
1214 * igbvf_set_rlpml - set receive large packet maximum length
1215 * @adapter: board private structure
1216 *
1217 * Configure the maximum size of packets that will be received
1218 */
1219static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1220{
1221 int max_frame_size;
1222 struct e1000_hw *hw = &adapter->hw;
1223
1224 max_frame_size = adapter->max_frame_size + VLAN_TAG_SIZE;
1225
1226 spin_lock_bh(&hw->mbx_lock);
1227
1228 e1000_rlpml_set_vf(hw, max_frame_size);
1229
1230 spin_unlock_bh(&hw->mbx_lock);
1231}
1232
1233static int igbvf_vlan_rx_add_vid(struct net_device *netdev,
1234 __be16 proto, u16 vid)
1235{
1236 struct igbvf_adapter *adapter = netdev_priv(netdev);
1237 struct e1000_hw *hw = &adapter->hw;
1238
1239 spin_lock_bh(&hw->mbx_lock);
1240
1241 if (hw->mac.ops.set_vfta(hw, vid, true)) {
1242 dev_warn(&adapter->pdev->dev, "Vlan id %d\n is not added", vid);
1243 spin_unlock_bh(&hw->mbx_lock);
1244 return -EINVAL;
1245 }
1246
1247 spin_unlock_bh(&hw->mbx_lock);
1248
1249 set_bit(vid, adapter->active_vlans);
1250 return 0;
1251}
1252
1253static int igbvf_vlan_rx_kill_vid(struct net_device *netdev,
1254 __be16 proto, u16 vid)
1255{
1256 struct igbvf_adapter *adapter = netdev_priv(netdev);
1257 struct e1000_hw *hw = &adapter->hw;
1258
1259 spin_lock_bh(&hw->mbx_lock);
1260
1261 if (hw->mac.ops.set_vfta(hw, vid, false)) {
1262 dev_err(&adapter->pdev->dev,
1263 "Failed to remove vlan id %d\n", vid);
1264 spin_unlock_bh(&hw->mbx_lock);
1265 return -EINVAL;
1266 }
1267
1268 spin_unlock_bh(&hw->mbx_lock);
1269
1270 clear_bit(vid, adapter->active_vlans);
1271 return 0;
1272}
1273
1274static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1275{
1276 u16 vid;
1277
1278 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1279 igbvf_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
1280}
1281
1282/**
1283 * igbvf_configure_tx - Configure Transmit Unit after Reset
1284 * @adapter: board private structure
1285 *
1286 * Configure the Tx unit of the MAC after a reset.
1287 **/
1288static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1289{
1290 struct e1000_hw *hw = &adapter->hw;
1291 struct igbvf_ring *tx_ring = adapter->tx_ring;
1292 u64 tdba;
1293 u32 txdctl, dca_txctrl;
1294
1295 /* disable transmits */
1296 txdctl = er32(TXDCTL(0));
1297 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1298 e1e_flush();
1299 msleep(10);
1300
1301 /* Setup the HW Tx Head and Tail descriptor pointers */
1302 ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1303 tdba = tx_ring->dma;
1304 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1305 ew32(TDBAH(0), (tdba >> 32));
1306 ew32(TDH(0), 0);
1307 ew32(TDT(0), 0);
1308 tx_ring->head = E1000_TDH(0);
1309 tx_ring->tail = E1000_TDT(0);
1310
1311 /* Turn off Relaxed Ordering on head write-backs. The writebacks
1312 * MUST be delivered in order or it will completely screw up
1313 * our bookkeeping.
1314 */
1315 dca_txctrl = er32(DCA_TXCTRL(0));
1316 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1317 ew32(DCA_TXCTRL(0), dca_txctrl);
1318
1319 /* enable transmits */
1320 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1321 ew32(TXDCTL(0), txdctl);
1322
1323 /* Setup Transmit Descriptor Settings for eop descriptor */
1324 adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1325
1326 /* enable Report Status bit */
1327 adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1328}
1329
1330/**
1331 * igbvf_setup_srrctl - configure the receive control registers
1332 * @adapter: Board private structure
1333 **/
1334static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1335{
1336 struct e1000_hw *hw = &adapter->hw;
1337 u32 srrctl = 0;
1338
1339 srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1340 E1000_SRRCTL_BSIZEHDR_MASK |
1341 E1000_SRRCTL_BSIZEPKT_MASK);
1342
1343 /* Enable queue drop to avoid head of line blocking */
1344 srrctl |= E1000_SRRCTL_DROP_EN;
1345
1346 /* Setup buffer sizes */
1347 srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1348 E1000_SRRCTL_BSIZEPKT_SHIFT;
1349
1350 if (adapter->rx_buffer_len < 2048) {
1351 adapter->rx_ps_hdr_size = 0;
1352 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1353 } else {
1354 adapter->rx_ps_hdr_size = 128;
1355 srrctl |= adapter->rx_ps_hdr_size <<
1356 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1357 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1358 }
1359
1360 ew32(SRRCTL(0), srrctl);
1361}
1362
1363/**
1364 * igbvf_configure_rx - Configure Receive Unit after Reset
1365 * @adapter: board private structure
1366 *
1367 * Configure the Rx unit of the MAC after a reset.
1368 **/
1369static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1370{
1371 struct e1000_hw *hw = &adapter->hw;
1372 struct igbvf_ring *rx_ring = adapter->rx_ring;
1373 u64 rdba;
1374 u32 rxdctl;
1375
1376 /* disable receives */
1377 rxdctl = er32(RXDCTL(0));
1378 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1379 e1e_flush();
1380 msleep(10);
1381
1382 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1383 * the Base and Length of the Rx Descriptor Ring
1384 */
1385 rdba = rx_ring->dma;
1386 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1387 ew32(RDBAH(0), (rdba >> 32));
1388 ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1389 rx_ring->head = E1000_RDH(0);
1390 rx_ring->tail = E1000_RDT(0);
1391 ew32(RDH(0), 0);
1392 ew32(RDT(0), 0);
1393
1394 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1395 rxdctl &= 0xFFF00000;
1396 rxdctl |= IGBVF_RX_PTHRESH;
1397 rxdctl |= IGBVF_RX_HTHRESH << 8;
1398 rxdctl |= IGBVF_RX_WTHRESH << 16;
1399
1400 igbvf_set_rlpml(adapter);
1401
1402 /* enable receives */
1403 ew32(RXDCTL(0), rxdctl);
1404}
1405
1406/**
1407 * igbvf_set_multi - Multicast and Promiscuous mode set
1408 * @netdev: network interface device structure
1409 *
1410 * The set_multi entry point is called whenever the multicast address
1411 * list or the network interface flags are updated. This routine is
1412 * responsible for configuring the hardware for proper multicast,
1413 * promiscuous mode, and all-multi behavior.
1414 **/
1415static void igbvf_set_multi(struct net_device *netdev)
1416{
1417 struct igbvf_adapter *adapter = netdev_priv(netdev);
1418 struct e1000_hw *hw = &adapter->hw;
1419 struct netdev_hw_addr *ha;
1420 u8 *mta_list = NULL;
1421 int i;
1422
1423 if (!netdev_mc_empty(netdev)) {
1424 mta_list = kmalloc_array(netdev_mc_count(netdev), ETH_ALEN,
1425 GFP_ATOMIC);
1426 if (!mta_list)
1427 return;
1428 }
1429
1430 /* prepare a packed array of only addresses. */
1431 i = 0;
1432 netdev_for_each_mc_addr(ha, netdev)
1433 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1434
1435 spin_lock_bh(&hw->mbx_lock);
1436
1437 hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1438
1439 spin_unlock_bh(&hw->mbx_lock);
1440 kfree(mta_list);
1441}
1442
1443/**
1444 * igbvf_set_uni - Configure unicast MAC filters
1445 * @netdev: network interface device structure
1446 *
1447 * This routine is responsible for configuring the hardware for proper
1448 * unicast filters.
1449 **/
1450static int igbvf_set_uni(struct net_device *netdev)
1451{
1452 struct igbvf_adapter *adapter = netdev_priv(netdev);
1453 struct e1000_hw *hw = &adapter->hw;
1454
1455 if (netdev_uc_count(netdev) > IGBVF_MAX_MAC_FILTERS) {
1456 pr_err("Too many unicast filters - No Space\n");
1457 return -ENOSPC;
1458 }
1459
1460 spin_lock_bh(&hw->mbx_lock);
1461
1462 /* Clear all unicast MAC filters */
1463 hw->mac.ops.set_uc_addr(hw, E1000_VF_MAC_FILTER_CLR, NULL);
1464
1465 spin_unlock_bh(&hw->mbx_lock);
1466
1467 if (!netdev_uc_empty(netdev)) {
1468 struct netdev_hw_addr *ha;
1469
1470 /* Add MAC filters one by one */
1471 netdev_for_each_uc_addr(ha, netdev) {
1472 spin_lock_bh(&hw->mbx_lock);
1473
1474 hw->mac.ops.set_uc_addr(hw, E1000_VF_MAC_FILTER_ADD,
1475 ha->addr);
1476
1477 spin_unlock_bh(&hw->mbx_lock);
1478 udelay(200);
1479 }
1480 }
1481
1482 return 0;
1483}
1484
1485static void igbvf_set_rx_mode(struct net_device *netdev)
1486{
1487 igbvf_set_multi(netdev);
1488 igbvf_set_uni(netdev);
1489}
1490
1491/**
1492 * igbvf_configure - configure the hardware for Rx and Tx
1493 * @adapter: private board structure
1494 **/
1495static void igbvf_configure(struct igbvf_adapter *adapter)
1496{
1497 igbvf_set_rx_mode(adapter->netdev);
1498
1499 igbvf_restore_vlan(adapter);
1500
1501 igbvf_configure_tx(adapter);
1502 igbvf_setup_srrctl(adapter);
1503 igbvf_configure_rx(adapter);
1504 igbvf_alloc_rx_buffers(adapter->rx_ring,
1505 igbvf_desc_unused(adapter->rx_ring));
1506}
1507
1508/* igbvf_reset - bring the hardware into a known good state
1509 * @adapter: private board structure
1510 *
1511 * This function boots the hardware and enables some settings that
1512 * require a configuration cycle of the hardware - those cannot be
1513 * set/changed during runtime. After reset the device needs to be
1514 * properly configured for Rx, Tx etc.
1515 */
1516static void igbvf_reset(struct igbvf_adapter *adapter)
1517{
1518 struct e1000_mac_info *mac = &adapter->hw.mac;
1519 struct net_device *netdev = adapter->netdev;
1520 struct e1000_hw *hw = &adapter->hw;
1521
1522 spin_lock_bh(&hw->mbx_lock);
1523
1524 /* Allow time for pending master requests to run */
1525 if (mac->ops.reset_hw(hw))
1526 dev_info(&adapter->pdev->dev, "PF still resetting\n");
1527
1528 mac->ops.init_hw(hw);
1529
1530 spin_unlock_bh(&hw->mbx_lock);
1531
1532 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1533 eth_hw_addr_set(netdev, adapter->hw.mac.addr);
1534 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1535 netdev->addr_len);
1536 }
1537
1538 adapter->last_reset = jiffies;
1539}
1540
1541int igbvf_up(struct igbvf_adapter *adapter)
1542{
1543 struct e1000_hw *hw = &adapter->hw;
1544
1545 /* hardware has been reset, we need to reload some things */
1546 igbvf_configure(adapter);
1547
1548 clear_bit(__IGBVF_DOWN, &adapter->state);
1549
1550 napi_enable(&adapter->rx_ring->napi);
1551 if (adapter->msix_entries)
1552 igbvf_configure_msix(adapter);
1553
1554 /* Clear any pending interrupts. */
1555 er32(EICR);
1556 igbvf_irq_enable(adapter);
1557
1558 /* start the watchdog */
1559 hw->mac.get_link_status = 1;
1560 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1561
1562 return 0;
1563}
1564
1565void igbvf_down(struct igbvf_adapter *adapter)
1566{
1567 struct net_device *netdev = adapter->netdev;
1568 struct e1000_hw *hw = &adapter->hw;
1569 u32 rxdctl, txdctl;
1570
1571 /* signal that we're down so the interrupt handler does not
1572 * reschedule our watchdog timer
1573 */
1574 set_bit(__IGBVF_DOWN, &adapter->state);
1575
1576 /* disable receives in the hardware */
1577 rxdctl = er32(RXDCTL(0));
1578 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1579
1580 netif_carrier_off(netdev);
1581 netif_stop_queue(netdev);
1582
1583 /* disable transmits in the hardware */
1584 txdctl = er32(TXDCTL(0));
1585 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1586
1587 /* flush both disables and wait for them to finish */
1588 e1e_flush();
1589 msleep(10);
1590
1591 napi_disable(&adapter->rx_ring->napi);
1592
1593 igbvf_irq_disable(adapter);
1594
1595 del_timer_sync(&adapter->watchdog_timer);
1596
1597 /* record the stats before reset*/
1598 igbvf_update_stats(adapter);
1599
1600 adapter->link_speed = 0;
1601 adapter->link_duplex = 0;
1602
1603 igbvf_reset(adapter);
1604 igbvf_clean_tx_ring(adapter->tx_ring);
1605 igbvf_clean_rx_ring(adapter->rx_ring);
1606}
1607
1608void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1609{
1610 might_sleep();
1611 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1612 usleep_range(1000, 2000);
1613 igbvf_down(adapter);
1614 igbvf_up(adapter);
1615 clear_bit(__IGBVF_RESETTING, &adapter->state);
1616}
1617
1618/**
1619 * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1620 * @adapter: board private structure to initialize
1621 *
1622 * igbvf_sw_init initializes the Adapter private data structure.
1623 * Fields are initialized based on PCI device information and
1624 * OS network device settings (MTU size).
1625 **/
1626static int igbvf_sw_init(struct igbvf_adapter *adapter)
1627{
1628 struct net_device *netdev = adapter->netdev;
1629 s32 rc;
1630
1631 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1632 adapter->rx_ps_hdr_size = 0;
1633 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1634 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1635
1636 adapter->tx_int_delay = 8;
1637 adapter->tx_abs_int_delay = 32;
1638 adapter->rx_int_delay = 0;
1639 adapter->rx_abs_int_delay = 8;
1640 adapter->requested_itr = 3;
1641 adapter->current_itr = IGBVF_START_ITR;
1642
1643 /* Set various function pointers */
1644 adapter->ei->init_ops(&adapter->hw);
1645
1646 rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1647 if (rc)
1648 return rc;
1649
1650 rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1651 if (rc)
1652 return rc;
1653
1654 igbvf_set_interrupt_capability(adapter);
1655
1656 if (igbvf_alloc_queues(adapter))
1657 return -ENOMEM;
1658
1659 spin_lock_init(&adapter->tx_queue_lock);
1660
1661 /* Explicitly disable IRQ since the NIC can be in any state. */
1662 igbvf_irq_disable(adapter);
1663
1664 spin_lock_init(&adapter->stats_lock);
1665 spin_lock_init(&adapter->hw.mbx_lock);
1666
1667 set_bit(__IGBVF_DOWN, &adapter->state);
1668 return 0;
1669}
1670
1671static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1672{
1673 struct e1000_hw *hw = &adapter->hw;
1674
1675 adapter->stats.last_gprc = er32(VFGPRC);
1676 adapter->stats.last_gorc = er32(VFGORC);
1677 adapter->stats.last_gptc = er32(VFGPTC);
1678 adapter->stats.last_gotc = er32(VFGOTC);
1679 adapter->stats.last_mprc = er32(VFMPRC);
1680 adapter->stats.last_gotlbc = er32(VFGOTLBC);
1681 adapter->stats.last_gptlbc = er32(VFGPTLBC);
1682 adapter->stats.last_gorlbc = er32(VFGORLBC);
1683 adapter->stats.last_gprlbc = er32(VFGPRLBC);
1684
1685 adapter->stats.base_gprc = er32(VFGPRC);
1686 adapter->stats.base_gorc = er32(VFGORC);
1687 adapter->stats.base_gptc = er32(VFGPTC);
1688 adapter->stats.base_gotc = er32(VFGOTC);
1689 adapter->stats.base_mprc = er32(VFMPRC);
1690 adapter->stats.base_gotlbc = er32(VFGOTLBC);
1691 adapter->stats.base_gptlbc = er32(VFGPTLBC);
1692 adapter->stats.base_gorlbc = er32(VFGORLBC);
1693 adapter->stats.base_gprlbc = er32(VFGPRLBC);
1694}
1695
1696/**
1697 * igbvf_open - Called when a network interface is made active
1698 * @netdev: network interface device structure
1699 *
1700 * Returns 0 on success, negative value on failure
1701 *
1702 * The open entry point is called when a network interface is made
1703 * active by the system (IFF_UP). At this point all resources needed
1704 * for transmit and receive operations are allocated, the interrupt
1705 * handler is registered with the OS, the watchdog timer is started,
1706 * and the stack is notified that the interface is ready.
1707 **/
1708static int igbvf_open(struct net_device *netdev)
1709{
1710 struct igbvf_adapter *adapter = netdev_priv(netdev);
1711 struct e1000_hw *hw = &adapter->hw;
1712 int err;
1713
1714 /* disallow open during test */
1715 if (test_bit(__IGBVF_TESTING, &adapter->state))
1716 return -EBUSY;
1717
1718 /* allocate transmit descriptors */
1719 err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1720 if (err)
1721 goto err_setup_tx;
1722
1723 /* allocate receive descriptors */
1724 err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1725 if (err)
1726 goto err_setup_rx;
1727
1728 /* before we allocate an interrupt, we must be ready to handle it.
1729 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1730 * as soon as we call pci_request_irq, so we have to setup our
1731 * clean_rx handler before we do so.
1732 */
1733 igbvf_configure(adapter);
1734
1735 err = igbvf_request_irq(adapter);
1736 if (err)
1737 goto err_req_irq;
1738
1739 /* From here on the code is the same as igbvf_up() */
1740 clear_bit(__IGBVF_DOWN, &adapter->state);
1741
1742 napi_enable(&adapter->rx_ring->napi);
1743
1744 /* clear any pending interrupts */
1745 er32(EICR);
1746
1747 igbvf_irq_enable(adapter);
1748
1749 /* start the watchdog */
1750 hw->mac.get_link_status = 1;
1751 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1752
1753 return 0;
1754
1755err_req_irq:
1756 igbvf_free_rx_resources(adapter->rx_ring);
1757err_setup_rx:
1758 igbvf_free_tx_resources(adapter->tx_ring);
1759err_setup_tx:
1760 igbvf_reset(adapter);
1761
1762 return err;
1763}
1764
1765/**
1766 * igbvf_close - Disables a network interface
1767 * @netdev: network interface device structure
1768 *
1769 * Returns 0, this is not allowed to fail
1770 *
1771 * The close entry point is called when an interface is de-activated
1772 * by the OS. The hardware is still under the drivers control, but
1773 * needs to be disabled. A global MAC reset is issued to stop the
1774 * hardware, and all transmit and receive resources are freed.
1775 **/
1776static int igbvf_close(struct net_device *netdev)
1777{
1778 struct igbvf_adapter *adapter = netdev_priv(netdev);
1779
1780 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1781 igbvf_down(adapter);
1782
1783 igbvf_free_irq(adapter);
1784
1785 igbvf_free_tx_resources(adapter->tx_ring);
1786 igbvf_free_rx_resources(adapter->rx_ring);
1787
1788 return 0;
1789}
1790
1791/**
1792 * igbvf_set_mac - Change the Ethernet Address of the NIC
1793 * @netdev: network interface device structure
1794 * @p: pointer to an address structure
1795 *
1796 * Returns 0 on success, negative on failure
1797 **/
1798static int igbvf_set_mac(struct net_device *netdev, void *p)
1799{
1800 struct igbvf_adapter *adapter = netdev_priv(netdev);
1801 struct e1000_hw *hw = &adapter->hw;
1802 struct sockaddr *addr = p;
1803
1804 if (!is_valid_ether_addr(addr->sa_data))
1805 return -EADDRNOTAVAIL;
1806
1807 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1808
1809 spin_lock_bh(&hw->mbx_lock);
1810
1811 hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1812
1813 spin_unlock_bh(&hw->mbx_lock);
1814
1815 if (!ether_addr_equal(addr->sa_data, hw->mac.addr))
1816 return -EADDRNOTAVAIL;
1817
1818 eth_hw_addr_set(netdev, addr->sa_data);
1819
1820 return 0;
1821}
1822
1823#define UPDATE_VF_COUNTER(reg, name) \
1824{ \
1825 u32 current_counter = er32(reg); \
1826 if (current_counter < adapter->stats.last_##name) \
1827 adapter->stats.name += 0x100000000LL; \
1828 adapter->stats.last_##name = current_counter; \
1829 adapter->stats.name &= 0xFFFFFFFF00000000LL; \
1830 adapter->stats.name |= current_counter; \
1831}
1832
1833/**
1834 * igbvf_update_stats - Update the board statistics counters
1835 * @adapter: board private structure
1836**/
1837void igbvf_update_stats(struct igbvf_adapter *adapter)
1838{
1839 struct e1000_hw *hw = &adapter->hw;
1840 struct pci_dev *pdev = adapter->pdev;
1841
1842 /* Prevent stats update while adapter is being reset, link is down
1843 * or if the pci connection is down.
1844 */
1845 if (adapter->link_speed == 0)
1846 return;
1847
1848 if (test_bit(__IGBVF_RESETTING, &adapter->state))
1849 return;
1850
1851 if (pci_channel_offline(pdev))
1852 return;
1853
1854 UPDATE_VF_COUNTER(VFGPRC, gprc);
1855 UPDATE_VF_COUNTER(VFGORC, gorc);
1856 UPDATE_VF_COUNTER(VFGPTC, gptc);
1857 UPDATE_VF_COUNTER(VFGOTC, gotc);
1858 UPDATE_VF_COUNTER(VFMPRC, mprc);
1859 UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1860 UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1861 UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1862 UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1863
1864 /* Fill out the OS statistics structure */
1865 adapter->netdev->stats.multicast = adapter->stats.mprc;
1866}
1867
1868static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1869{
1870 dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s Duplex\n",
1871 adapter->link_speed,
1872 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half");
1873}
1874
1875static bool igbvf_has_link(struct igbvf_adapter *adapter)
1876{
1877 struct e1000_hw *hw = &adapter->hw;
1878 s32 ret_val = E1000_SUCCESS;
1879 bool link_active;
1880
1881 /* If interface is down, stay link down */
1882 if (test_bit(__IGBVF_DOWN, &adapter->state))
1883 return false;
1884
1885 spin_lock_bh(&hw->mbx_lock);
1886
1887 ret_val = hw->mac.ops.check_for_link(hw);
1888
1889 spin_unlock_bh(&hw->mbx_lock);
1890
1891 link_active = !hw->mac.get_link_status;
1892
1893 /* if check for link returns error we will need to reset */
1894 if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1895 schedule_work(&adapter->reset_task);
1896
1897 return link_active;
1898}
1899
1900/**
1901 * igbvf_watchdog - Timer Call-back
1902 * @t: timer list pointer containing private struct
1903 **/
1904static void igbvf_watchdog(struct timer_list *t)
1905{
1906 struct igbvf_adapter *adapter = from_timer(adapter, t, watchdog_timer);
1907
1908 /* Do the rest outside of interrupt context */
1909 schedule_work(&adapter->watchdog_task);
1910}
1911
1912static void igbvf_watchdog_task(struct work_struct *work)
1913{
1914 struct igbvf_adapter *adapter = container_of(work,
1915 struct igbvf_adapter,
1916 watchdog_task);
1917 struct net_device *netdev = adapter->netdev;
1918 struct e1000_mac_info *mac = &adapter->hw.mac;
1919 struct igbvf_ring *tx_ring = adapter->tx_ring;
1920 struct e1000_hw *hw = &adapter->hw;
1921 u32 link;
1922 int tx_pending = 0;
1923
1924 link = igbvf_has_link(adapter);
1925
1926 if (link) {
1927 if (!netif_carrier_ok(netdev)) {
1928 mac->ops.get_link_up_info(&adapter->hw,
1929 &adapter->link_speed,
1930 &adapter->link_duplex);
1931 igbvf_print_link_info(adapter);
1932
1933 netif_carrier_on(netdev);
1934 netif_wake_queue(netdev);
1935 }
1936 } else {
1937 if (netif_carrier_ok(netdev)) {
1938 adapter->link_speed = 0;
1939 adapter->link_duplex = 0;
1940 dev_info(&adapter->pdev->dev, "Link is Down\n");
1941 netif_carrier_off(netdev);
1942 netif_stop_queue(netdev);
1943 }
1944 }
1945
1946 if (netif_carrier_ok(netdev)) {
1947 igbvf_update_stats(adapter);
1948 } else {
1949 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1950 tx_ring->count);
1951 if (tx_pending) {
1952 /* We've lost link, so the controller stops DMA,
1953 * but we've got queued Tx work that's never going
1954 * to get done, so reset controller to flush Tx.
1955 * (Do the reset outside of interrupt context).
1956 */
1957 adapter->tx_timeout_count++;
1958 schedule_work(&adapter->reset_task);
1959 }
1960 }
1961
1962 /* Cause software interrupt to ensure Rx ring is cleaned */
1963 ew32(EICS, adapter->rx_ring->eims_value);
1964
1965 /* Reset the timer */
1966 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1967 mod_timer(&adapter->watchdog_timer,
1968 round_jiffies(jiffies + (2 * HZ)));
1969}
1970
1971#define IGBVF_TX_FLAGS_CSUM 0x00000001
1972#define IGBVF_TX_FLAGS_VLAN 0x00000002
1973#define IGBVF_TX_FLAGS_TSO 0x00000004
1974#define IGBVF_TX_FLAGS_IPV4 0x00000008
1975#define IGBVF_TX_FLAGS_VLAN_MASK 0xffff0000
1976#define IGBVF_TX_FLAGS_VLAN_SHIFT 16
1977
1978static void igbvf_tx_ctxtdesc(struct igbvf_ring *tx_ring, u32 vlan_macip_lens,
1979 u32 type_tucmd, u32 mss_l4len_idx)
1980{
1981 struct e1000_adv_tx_context_desc *context_desc;
1982 struct igbvf_buffer *buffer_info;
1983 u16 i = tx_ring->next_to_use;
1984
1985 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1986 buffer_info = &tx_ring->buffer_info[i];
1987
1988 i++;
1989 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1990
1991 /* set bits to identify this as an advanced context descriptor */
1992 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
1993
1994 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
1995 context_desc->seqnum_seed = 0;
1996 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
1997 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1998
1999 buffer_info->time_stamp = jiffies;
2000 buffer_info->dma = 0;
2001}
2002
2003static int igbvf_tso(struct igbvf_ring *tx_ring,
2004 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2005{
2006 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
2007 union {
2008 struct iphdr *v4;
2009 struct ipv6hdr *v6;
2010 unsigned char *hdr;
2011 } ip;
2012 union {
2013 struct tcphdr *tcp;
2014 unsigned char *hdr;
2015 } l4;
2016 u32 paylen, l4_offset;
2017 int err;
2018
2019 if (skb->ip_summed != CHECKSUM_PARTIAL)
2020 return 0;
2021
2022 if (!skb_is_gso(skb))
2023 return 0;
2024
2025 err = skb_cow_head(skb, 0);
2026 if (err < 0)
2027 return err;
2028
2029 ip.hdr = skb_network_header(skb);
2030 l4.hdr = skb_checksum_start(skb);
2031
2032 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2033 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
2034
2035 /* initialize outer IP header fields */
2036 if (ip.v4->version == 4) {
2037 unsigned char *csum_start = skb_checksum_start(skb);
2038 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
2039
2040 /* IP header will have to cancel out any data that
2041 * is not a part of the outer IP header
2042 */
2043 ip.v4->check = csum_fold(csum_partial(trans_start,
2044 csum_start - trans_start,
2045 0));
2046 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
2047
2048 ip.v4->tot_len = 0;
2049 } else {
2050 ip.v6->payload_len = 0;
2051 }
2052
2053 /* determine offset of inner transport header */
2054 l4_offset = l4.hdr - skb->data;
2055
2056 /* compute length of segmentation header */
2057 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
2058
2059 /* remove payload length from inner checksum */
2060 paylen = skb->len - l4_offset;
2061 csum_replace_by_diff(&l4.tcp->check, (__force __wsum)htonl(paylen));
2062
2063 /* MSS L4LEN IDX */
2064 mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT;
2065 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
2066
2067 /* VLAN MACLEN IPLEN */
2068 vlan_macip_lens = l4.hdr - ip.hdr;
2069 vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT;
2070 vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK;
2071
2072 igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
2073
2074 return 1;
2075}
2076
2077static bool igbvf_tx_csum(struct igbvf_ring *tx_ring, struct sk_buff *skb,
2078 u32 tx_flags, __be16 protocol)
2079{
2080 u32 vlan_macip_lens = 0;
2081 u32 type_tucmd = 0;
2082
2083 if (skb->ip_summed != CHECKSUM_PARTIAL) {
2084csum_failed:
2085 if (!(tx_flags & IGBVF_TX_FLAGS_VLAN))
2086 return false;
2087 goto no_csum;
2088 }
2089
2090 switch (skb->csum_offset) {
2091 case offsetof(struct tcphdr, check):
2092 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
2093 fallthrough;
2094 case offsetof(struct udphdr, check):
2095 break;
2096 case offsetof(struct sctphdr, checksum):
2097 /* validate that this is actually an SCTP request */
2098 if (skb_csum_is_sctp(skb)) {
2099 type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP;
2100 break;
2101 }
2102 fallthrough;
2103 default:
2104 skb_checksum_help(skb);
2105 goto csum_failed;
2106 }
2107
2108 vlan_macip_lens = skb_checksum_start_offset(skb) -
2109 skb_network_offset(skb);
2110no_csum:
2111 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
2112 vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK;
2113
2114 igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0);
2115 return true;
2116}
2117
2118static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2119{
2120 struct igbvf_adapter *adapter = netdev_priv(netdev);
2121
2122 /* there is enough descriptors then we don't need to worry */
2123 if (igbvf_desc_unused(adapter->tx_ring) >= size)
2124 return 0;
2125
2126 netif_stop_queue(netdev);
2127
2128 /* Herbert's original patch had:
2129 * smp_mb__after_netif_stop_queue();
2130 * but since that doesn't exist yet, just open code it.
2131 */
2132 smp_mb();
2133
2134 /* We need to check again just in case room has been made available */
2135 if (igbvf_desc_unused(adapter->tx_ring) < size)
2136 return -EBUSY;
2137
2138 netif_wake_queue(netdev);
2139
2140 ++adapter->restart_queue;
2141 return 0;
2142}
2143
2144#define IGBVF_MAX_TXD_PWR 16
2145#define IGBVF_MAX_DATA_PER_TXD (1u << IGBVF_MAX_TXD_PWR)
2146
2147static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2148 struct igbvf_ring *tx_ring,
2149 struct sk_buff *skb)
2150{
2151 struct igbvf_buffer *buffer_info;
2152 struct pci_dev *pdev = adapter->pdev;
2153 unsigned int len = skb_headlen(skb);
2154 unsigned int count = 0, i;
2155 unsigned int f;
2156
2157 i = tx_ring->next_to_use;
2158
2159 buffer_info = &tx_ring->buffer_info[i];
2160 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2161 buffer_info->length = len;
2162 /* set time_stamp *before* dma to help avoid a possible race */
2163 buffer_info->time_stamp = jiffies;
2164 buffer_info->mapped_as_page = false;
2165 buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2166 DMA_TO_DEVICE);
2167 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2168 goto dma_error;
2169
2170 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2171 const skb_frag_t *frag;
2172
2173 count++;
2174 i++;
2175 if (i == tx_ring->count)
2176 i = 0;
2177
2178 frag = &skb_shinfo(skb)->frags[f];
2179 len = skb_frag_size(frag);
2180
2181 buffer_info = &tx_ring->buffer_info[i];
2182 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2183 buffer_info->length = len;
2184 buffer_info->time_stamp = jiffies;
2185 buffer_info->mapped_as_page = true;
2186 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag, 0, len,
2187 DMA_TO_DEVICE);
2188 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2189 goto dma_error;
2190 }
2191
2192 tx_ring->buffer_info[i].skb = skb;
2193
2194 return ++count;
2195
2196dma_error:
2197 dev_err(&pdev->dev, "TX DMA map failed\n");
2198
2199 /* clear timestamp and dma mappings for failed buffer_info mapping */
2200 buffer_info->dma = 0;
2201 buffer_info->time_stamp = 0;
2202 buffer_info->length = 0;
2203 buffer_info->mapped_as_page = false;
2204 if (count)
2205 count--;
2206
2207 /* clear timestamp and dma mappings for remaining portion of packet */
2208 while (count--) {
2209 if (i == 0)
2210 i += tx_ring->count;
2211 i--;
2212 buffer_info = &tx_ring->buffer_info[i];
2213 igbvf_put_txbuf(adapter, buffer_info);
2214 }
2215
2216 return 0;
2217}
2218
2219static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2220 struct igbvf_ring *tx_ring,
2221 int tx_flags, int count,
2222 unsigned int first, u32 paylen,
2223 u8 hdr_len)
2224{
2225 union e1000_adv_tx_desc *tx_desc = NULL;
2226 struct igbvf_buffer *buffer_info;
2227 u32 olinfo_status = 0, cmd_type_len;
2228 unsigned int i;
2229
2230 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2231 E1000_ADVTXD_DCMD_DEXT);
2232
2233 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2234 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2235
2236 if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2237 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2238
2239 /* insert tcp checksum */
2240 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2241
2242 /* insert ip checksum */
2243 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2244 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2245
2246 } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2247 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2248 }
2249
2250 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2251
2252 i = tx_ring->next_to_use;
2253 while (count--) {
2254 buffer_info = &tx_ring->buffer_info[i];
2255 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2256 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2257 tx_desc->read.cmd_type_len =
2258 cpu_to_le32(cmd_type_len | buffer_info->length);
2259 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2260 i++;
2261 if (i == tx_ring->count)
2262 i = 0;
2263 }
2264
2265 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2266 /* Force memory writes to complete before letting h/w
2267 * know there are new descriptors to fetch. (Only
2268 * applicable for weak-ordered memory model archs,
2269 * such as IA-64).
2270 */
2271 wmb();
2272
2273 tx_ring->buffer_info[first].next_to_watch = tx_desc;
2274 tx_ring->next_to_use = i;
2275 writel(i, adapter->hw.hw_addr + tx_ring->tail);
2276}
2277
2278static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2279 struct net_device *netdev,
2280 struct igbvf_ring *tx_ring)
2281{
2282 struct igbvf_adapter *adapter = netdev_priv(netdev);
2283 unsigned int first, tx_flags = 0;
2284 u8 hdr_len = 0;
2285 int count = 0;
2286 int tso = 0;
2287 __be16 protocol = vlan_get_protocol(skb);
2288
2289 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2290 dev_kfree_skb_any(skb);
2291 return NETDEV_TX_OK;
2292 }
2293
2294 if (skb->len <= 0) {
2295 dev_kfree_skb_any(skb);
2296 return NETDEV_TX_OK;
2297 }
2298
2299 /* need: count + 4 desc gap to keep tail from touching
2300 * + 2 desc gap to keep tail from touching head,
2301 * + 1 desc for skb->data,
2302 * + 1 desc for context descriptor,
2303 * head, otherwise try next time
2304 */
2305 if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2306 /* this is a hard error */
2307 return NETDEV_TX_BUSY;
2308 }
2309
2310 if (skb_vlan_tag_present(skb)) {
2311 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2312 tx_flags |= (skb_vlan_tag_get(skb) <<
2313 IGBVF_TX_FLAGS_VLAN_SHIFT);
2314 }
2315
2316 if (protocol == htons(ETH_P_IP))
2317 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2318
2319 first = tx_ring->next_to_use;
2320
2321 tso = igbvf_tso(tx_ring, skb, tx_flags, &hdr_len);
2322 if (unlikely(tso < 0)) {
2323 dev_kfree_skb_any(skb);
2324 return NETDEV_TX_OK;
2325 }
2326
2327 if (tso)
2328 tx_flags |= IGBVF_TX_FLAGS_TSO;
2329 else if (igbvf_tx_csum(tx_ring, skb, tx_flags, protocol) &&
2330 (skb->ip_summed == CHECKSUM_PARTIAL))
2331 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2332
2333 /* count reflects descriptors mapped, if 0 then mapping error
2334 * has occurred and we need to rewind the descriptor queue
2335 */
2336 count = igbvf_tx_map_adv(adapter, tx_ring, skb);
2337
2338 if (count) {
2339 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2340 first, skb->len, hdr_len);
2341 /* Make sure there is space in the ring for the next send. */
2342 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2343 } else {
2344 dev_kfree_skb_any(skb);
2345 tx_ring->buffer_info[first].time_stamp = 0;
2346 tx_ring->next_to_use = first;
2347 }
2348
2349 return NETDEV_TX_OK;
2350}
2351
2352static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2353 struct net_device *netdev)
2354{
2355 struct igbvf_adapter *adapter = netdev_priv(netdev);
2356 struct igbvf_ring *tx_ring;
2357
2358 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2359 dev_kfree_skb_any(skb);
2360 return NETDEV_TX_OK;
2361 }
2362
2363 tx_ring = &adapter->tx_ring[0];
2364
2365 return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2366}
2367
2368/**
2369 * igbvf_tx_timeout - Respond to a Tx Hang
2370 * @netdev: network interface device structure
2371 * @txqueue: queue timing out (unused)
2372 **/
2373static void igbvf_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
2374{
2375 struct igbvf_adapter *adapter = netdev_priv(netdev);
2376
2377 /* Do the reset outside of interrupt context */
2378 adapter->tx_timeout_count++;
2379 schedule_work(&adapter->reset_task);
2380}
2381
2382static void igbvf_reset_task(struct work_struct *work)
2383{
2384 struct igbvf_adapter *adapter;
2385
2386 adapter = container_of(work, struct igbvf_adapter, reset_task);
2387
2388 igbvf_reinit_locked(adapter);
2389}
2390
2391/**
2392 * igbvf_change_mtu - Change the Maximum Transfer Unit
2393 * @netdev: network interface device structure
2394 * @new_mtu: new value for maximum frame size
2395 *
2396 * Returns 0 on success, negative on failure
2397 **/
2398static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2399{
2400 struct igbvf_adapter *adapter = netdev_priv(netdev);
2401 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2402
2403 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2404 usleep_range(1000, 2000);
2405 /* igbvf_down has a dependency on max_frame_size */
2406 adapter->max_frame_size = max_frame;
2407 if (netif_running(netdev))
2408 igbvf_down(adapter);
2409
2410 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2411 * means we reserve 2 more, this pushes us to allocate from the next
2412 * larger slab size.
2413 * i.e. RXBUFFER_2048 --> size-4096 slab
2414 * However with the new *_jumbo_rx* routines, jumbo receives will use
2415 * fragmented skbs
2416 */
2417
2418 if (max_frame <= 1024)
2419 adapter->rx_buffer_len = 1024;
2420 else if (max_frame <= 2048)
2421 adapter->rx_buffer_len = 2048;
2422 else
2423#if (PAGE_SIZE / 2) > 16384
2424 adapter->rx_buffer_len = 16384;
2425#else
2426 adapter->rx_buffer_len = PAGE_SIZE / 2;
2427#endif
2428
2429 /* adjust allocation if LPE protects us, and we aren't using SBP */
2430 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2431 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2432 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2433 ETH_FCS_LEN;
2434
2435 netdev_dbg(netdev, "changing MTU from %d to %d\n",
2436 netdev->mtu, new_mtu);
2437 netdev->mtu = new_mtu;
2438
2439 if (netif_running(netdev))
2440 igbvf_up(adapter);
2441 else
2442 igbvf_reset(adapter);
2443
2444 clear_bit(__IGBVF_RESETTING, &adapter->state);
2445
2446 return 0;
2447}
2448
2449static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2450{
2451 switch (cmd) {
2452 default:
2453 return -EOPNOTSUPP;
2454 }
2455}
2456
2457static int igbvf_suspend(struct device *dev_d)
2458{
2459 struct net_device *netdev = dev_get_drvdata(dev_d);
2460 struct igbvf_adapter *adapter = netdev_priv(netdev);
2461
2462 netif_device_detach(netdev);
2463
2464 if (netif_running(netdev)) {
2465 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2466 igbvf_down(adapter);
2467 igbvf_free_irq(adapter);
2468 }
2469
2470 return 0;
2471}
2472
2473static int __maybe_unused igbvf_resume(struct device *dev_d)
2474{
2475 struct pci_dev *pdev = to_pci_dev(dev_d);
2476 struct net_device *netdev = pci_get_drvdata(pdev);
2477 struct igbvf_adapter *adapter = netdev_priv(netdev);
2478 u32 err;
2479
2480 pci_set_master(pdev);
2481
2482 if (netif_running(netdev)) {
2483 err = igbvf_request_irq(adapter);
2484 if (err)
2485 return err;
2486 }
2487
2488 igbvf_reset(adapter);
2489
2490 if (netif_running(netdev))
2491 igbvf_up(adapter);
2492
2493 netif_device_attach(netdev);
2494
2495 return 0;
2496}
2497
2498static void igbvf_shutdown(struct pci_dev *pdev)
2499{
2500 igbvf_suspend(&pdev->dev);
2501}
2502
2503#ifdef CONFIG_NET_POLL_CONTROLLER
2504/* Polling 'interrupt' - used by things like netconsole to send skbs
2505 * without having to re-enable interrupts. It's not called while
2506 * the interrupt routine is executing.
2507 */
2508static void igbvf_netpoll(struct net_device *netdev)
2509{
2510 struct igbvf_adapter *adapter = netdev_priv(netdev);
2511
2512 disable_irq(adapter->pdev->irq);
2513
2514 igbvf_clean_tx_irq(adapter->tx_ring);
2515
2516 enable_irq(adapter->pdev->irq);
2517}
2518#endif
2519
2520/**
2521 * igbvf_io_error_detected - called when PCI error is detected
2522 * @pdev: Pointer to PCI device
2523 * @state: The current pci connection state
2524 *
2525 * This function is called after a PCI bus error affecting
2526 * this device has been detected.
2527 */
2528static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2529 pci_channel_state_t state)
2530{
2531 struct net_device *netdev = pci_get_drvdata(pdev);
2532 struct igbvf_adapter *adapter = netdev_priv(netdev);
2533
2534 netif_device_detach(netdev);
2535
2536 if (state == pci_channel_io_perm_failure)
2537 return PCI_ERS_RESULT_DISCONNECT;
2538
2539 if (netif_running(netdev))
2540 igbvf_down(adapter);
2541 pci_disable_device(pdev);
2542
2543 /* Request a slot reset. */
2544 return PCI_ERS_RESULT_NEED_RESET;
2545}
2546
2547/**
2548 * igbvf_io_slot_reset - called after the pci bus has been reset.
2549 * @pdev: Pointer to PCI device
2550 *
2551 * Restart the card from scratch, as if from a cold-boot. Implementation
2552 * resembles the first-half of the igbvf_resume routine.
2553 */
2554static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2555{
2556 struct net_device *netdev = pci_get_drvdata(pdev);
2557 struct igbvf_adapter *adapter = netdev_priv(netdev);
2558
2559 if (pci_enable_device_mem(pdev)) {
2560 dev_err(&pdev->dev,
2561 "Cannot re-enable PCI device after reset.\n");
2562 return PCI_ERS_RESULT_DISCONNECT;
2563 }
2564 pci_set_master(pdev);
2565
2566 igbvf_reset(adapter);
2567
2568 return PCI_ERS_RESULT_RECOVERED;
2569}
2570
2571/**
2572 * igbvf_io_resume - called when traffic can start flowing again.
2573 * @pdev: Pointer to PCI device
2574 *
2575 * This callback is called when the error recovery driver tells us that
2576 * its OK to resume normal operation. Implementation resembles the
2577 * second-half of the igbvf_resume routine.
2578 */
2579static void igbvf_io_resume(struct pci_dev *pdev)
2580{
2581 struct net_device *netdev = pci_get_drvdata(pdev);
2582 struct igbvf_adapter *adapter = netdev_priv(netdev);
2583
2584 if (netif_running(netdev)) {
2585 if (igbvf_up(adapter)) {
2586 dev_err(&pdev->dev,
2587 "can't bring device back up after reset\n");
2588 return;
2589 }
2590 }
2591
2592 netif_device_attach(netdev);
2593}
2594
2595/**
2596 * igbvf_io_prepare - prepare device driver for PCI reset
2597 * @pdev: PCI device information struct
2598 */
2599static void igbvf_io_prepare(struct pci_dev *pdev)
2600{
2601 struct net_device *netdev = pci_get_drvdata(pdev);
2602 struct igbvf_adapter *adapter = netdev_priv(netdev);
2603
2604 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2605 usleep_range(1000, 2000);
2606 igbvf_down(adapter);
2607}
2608
2609/**
2610 * igbvf_io_reset_done - PCI reset done, device driver reset can begin
2611 * @pdev: PCI device information struct
2612 */
2613static void igbvf_io_reset_done(struct pci_dev *pdev)
2614{
2615 struct net_device *netdev = pci_get_drvdata(pdev);
2616 struct igbvf_adapter *adapter = netdev_priv(netdev);
2617
2618 igbvf_up(adapter);
2619 clear_bit(__IGBVF_RESETTING, &adapter->state);
2620}
2621
2622static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2623{
2624 struct e1000_hw *hw = &adapter->hw;
2625 struct net_device *netdev = adapter->netdev;
2626 struct pci_dev *pdev = adapter->pdev;
2627
2628 if (hw->mac.type == e1000_vfadapt_i350)
2629 dev_info(&pdev->dev, "Intel(R) I350 Virtual Function\n");
2630 else
2631 dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2632 dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2633}
2634
2635static int igbvf_set_features(struct net_device *netdev,
2636 netdev_features_t features)
2637{
2638 struct igbvf_adapter *adapter = netdev_priv(netdev);
2639
2640 if (features & NETIF_F_RXCSUM)
2641 adapter->flags &= ~IGBVF_FLAG_RX_CSUM_DISABLED;
2642 else
2643 adapter->flags |= IGBVF_FLAG_RX_CSUM_DISABLED;
2644
2645 return 0;
2646}
2647
2648#define IGBVF_MAX_MAC_HDR_LEN 127
2649#define IGBVF_MAX_NETWORK_HDR_LEN 511
2650
2651static netdev_features_t
2652igbvf_features_check(struct sk_buff *skb, struct net_device *dev,
2653 netdev_features_t features)
2654{
2655 unsigned int network_hdr_len, mac_hdr_len;
2656
2657 /* Make certain the headers can be described by a context descriptor */
2658 mac_hdr_len = skb_network_header(skb) - skb->data;
2659 if (unlikely(mac_hdr_len > IGBVF_MAX_MAC_HDR_LEN))
2660 return features & ~(NETIF_F_HW_CSUM |
2661 NETIF_F_SCTP_CRC |
2662 NETIF_F_HW_VLAN_CTAG_TX |
2663 NETIF_F_TSO |
2664 NETIF_F_TSO6);
2665
2666 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
2667 if (unlikely(network_hdr_len > IGBVF_MAX_NETWORK_HDR_LEN))
2668 return features & ~(NETIF_F_HW_CSUM |
2669 NETIF_F_SCTP_CRC |
2670 NETIF_F_TSO |
2671 NETIF_F_TSO6);
2672
2673 /* We can only support IPV4 TSO in tunnels if we can mangle the
2674 * inner IP ID field, so strip TSO if MANGLEID is not supported.
2675 */
2676 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
2677 features &= ~NETIF_F_TSO;
2678
2679 return features;
2680}
2681
2682static const struct net_device_ops igbvf_netdev_ops = {
2683 .ndo_open = igbvf_open,
2684 .ndo_stop = igbvf_close,
2685 .ndo_start_xmit = igbvf_xmit_frame,
2686 .ndo_set_rx_mode = igbvf_set_rx_mode,
2687 .ndo_set_mac_address = igbvf_set_mac,
2688 .ndo_change_mtu = igbvf_change_mtu,
2689 .ndo_eth_ioctl = igbvf_ioctl,
2690 .ndo_tx_timeout = igbvf_tx_timeout,
2691 .ndo_vlan_rx_add_vid = igbvf_vlan_rx_add_vid,
2692 .ndo_vlan_rx_kill_vid = igbvf_vlan_rx_kill_vid,
2693#ifdef CONFIG_NET_POLL_CONTROLLER
2694 .ndo_poll_controller = igbvf_netpoll,
2695#endif
2696 .ndo_set_features = igbvf_set_features,
2697 .ndo_features_check = igbvf_features_check,
2698};
2699
2700/**
2701 * igbvf_probe - Device Initialization Routine
2702 * @pdev: PCI device information struct
2703 * @ent: entry in igbvf_pci_tbl
2704 *
2705 * Returns 0 on success, negative on failure
2706 *
2707 * igbvf_probe initializes an adapter identified by a pci_dev structure.
2708 * The OS initialization, configuring of the adapter private structure,
2709 * and a hardware reset occur.
2710 **/
2711static int igbvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2712{
2713 struct net_device *netdev;
2714 struct igbvf_adapter *adapter;
2715 struct e1000_hw *hw;
2716 const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2717 static int cards_found;
2718 int err;
2719
2720 err = pci_enable_device_mem(pdev);
2721 if (err)
2722 return err;
2723
2724 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
2725 if (err) {
2726 dev_err(&pdev->dev,
2727 "No usable DMA configuration, aborting\n");
2728 goto err_dma;
2729 }
2730
2731 err = pci_request_regions(pdev, igbvf_driver_name);
2732 if (err)
2733 goto err_pci_reg;
2734
2735 pci_set_master(pdev);
2736
2737 err = -ENOMEM;
2738 netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2739 if (!netdev)
2740 goto err_alloc_etherdev;
2741
2742 SET_NETDEV_DEV(netdev, &pdev->dev);
2743
2744 pci_set_drvdata(pdev, netdev);
2745 adapter = netdev_priv(netdev);
2746 hw = &adapter->hw;
2747 adapter->netdev = netdev;
2748 adapter->pdev = pdev;
2749 adapter->ei = ei;
2750 adapter->pba = ei->pba;
2751 adapter->flags = ei->flags;
2752 adapter->hw.back = adapter;
2753 adapter->hw.mac.type = ei->mac;
2754 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2755
2756 /* PCI config space info */
2757
2758 hw->vendor_id = pdev->vendor;
2759 hw->device_id = pdev->device;
2760 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2761 hw->subsystem_device_id = pdev->subsystem_device;
2762 hw->revision_id = pdev->revision;
2763
2764 err = -EIO;
2765 adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2766 pci_resource_len(pdev, 0));
2767
2768 if (!adapter->hw.hw_addr)
2769 goto err_ioremap;
2770
2771 if (ei->get_variants) {
2772 err = ei->get_variants(adapter);
2773 if (err)
2774 goto err_get_variants;
2775 }
2776
2777 /* setup adapter struct */
2778 err = igbvf_sw_init(adapter);
2779 if (err)
2780 goto err_sw_init;
2781
2782 /* construct the net_device struct */
2783 netdev->netdev_ops = &igbvf_netdev_ops;
2784
2785 igbvf_set_ethtool_ops(netdev);
2786 netdev->watchdog_timeo = 5 * HZ;
2787 strscpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
2788
2789 adapter->bd_number = cards_found++;
2790
2791 netdev->hw_features = NETIF_F_SG |
2792 NETIF_F_TSO |
2793 NETIF_F_TSO6 |
2794 NETIF_F_RXCSUM |
2795 NETIF_F_HW_CSUM |
2796 NETIF_F_SCTP_CRC;
2797
2798#define IGBVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
2799 NETIF_F_GSO_GRE_CSUM | \
2800 NETIF_F_GSO_IPXIP4 | \
2801 NETIF_F_GSO_IPXIP6 | \
2802 NETIF_F_GSO_UDP_TUNNEL | \
2803 NETIF_F_GSO_UDP_TUNNEL_CSUM)
2804
2805 netdev->gso_partial_features = IGBVF_GSO_PARTIAL_FEATURES;
2806 netdev->hw_features |= NETIF_F_GSO_PARTIAL |
2807 IGBVF_GSO_PARTIAL_FEATURES;
2808
2809 netdev->features = netdev->hw_features | NETIF_F_HIGHDMA;
2810
2811 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
2812 netdev->mpls_features |= NETIF_F_HW_CSUM;
2813 netdev->hw_enc_features |= netdev->vlan_features;
2814
2815 /* set this bit last since it cannot be part of vlan_features */
2816 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
2817 NETIF_F_HW_VLAN_CTAG_RX |
2818 NETIF_F_HW_VLAN_CTAG_TX;
2819
2820 /* MTU range: 68 - 9216 */
2821 netdev->min_mtu = ETH_MIN_MTU;
2822 netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
2823
2824 spin_lock_bh(&hw->mbx_lock);
2825
2826 /*reset the controller to put the device in a known good state */
2827 err = hw->mac.ops.reset_hw(hw);
2828 if (err) {
2829 dev_info(&pdev->dev,
2830 "PF still in reset state. Is the PF interface up?\n");
2831 } else {
2832 err = hw->mac.ops.read_mac_addr(hw);
2833 if (err)
2834 dev_info(&pdev->dev, "Error reading MAC address.\n");
2835 else if (is_zero_ether_addr(adapter->hw.mac.addr))
2836 dev_info(&pdev->dev,
2837 "MAC address not assigned by administrator.\n");
2838 eth_hw_addr_set(netdev, adapter->hw.mac.addr);
2839 }
2840
2841 spin_unlock_bh(&hw->mbx_lock);
2842
2843 if (!is_valid_ether_addr(netdev->dev_addr)) {
2844 dev_info(&pdev->dev, "Assigning random MAC address.\n");
2845 eth_hw_addr_random(netdev);
2846 memcpy(adapter->hw.mac.addr, netdev->dev_addr,
2847 netdev->addr_len);
2848 }
2849
2850 timer_setup(&adapter->watchdog_timer, igbvf_watchdog, 0);
2851
2852 INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2853 INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2854
2855 /* ring size defaults */
2856 adapter->rx_ring->count = 1024;
2857 adapter->tx_ring->count = 1024;
2858
2859 /* reset the hardware with the new settings */
2860 igbvf_reset(adapter);
2861
2862 /* set hardware-specific flags */
2863 if (adapter->hw.mac.type == e1000_vfadapt_i350)
2864 adapter->flags |= IGBVF_FLAG_RX_LB_VLAN_BSWAP;
2865
2866 strcpy(netdev->name, "eth%d");
2867 err = register_netdev(netdev);
2868 if (err)
2869 goto err_hw_init;
2870
2871 /* tell the stack to leave us alone until igbvf_open() is called */
2872 netif_carrier_off(netdev);
2873 netif_stop_queue(netdev);
2874
2875 igbvf_print_device_info(adapter);
2876
2877 igbvf_initialize_last_counter_stats(adapter);
2878
2879 return 0;
2880
2881err_hw_init:
2882 netif_napi_del(&adapter->rx_ring->napi);
2883 kfree(adapter->tx_ring);
2884 kfree(adapter->rx_ring);
2885err_sw_init:
2886 igbvf_reset_interrupt_capability(adapter);
2887err_get_variants:
2888 iounmap(adapter->hw.hw_addr);
2889err_ioremap:
2890 free_netdev(netdev);
2891err_alloc_etherdev:
2892 pci_release_regions(pdev);
2893err_pci_reg:
2894err_dma:
2895 pci_disable_device(pdev);
2896 return err;
2897}
2898
2899/**
2900 * igbvf_remove - Device Removal Routine
2901 * @pdev: PCI device information struct
2902 *
2903 * igbvf_remove is called by the PCI subsystem to alert the driver
2904 * that it should release a PCI device. The could be caused by a
2905 * Hot-Plug event, or because the driver is going to be removed from
2906 * memory.
2907 **/
2908static void igbvf_remove(struct pci_dev *pdev)
2909{
2910 struct net_device *netdev = pci_get_drvdata(pdev);
2911 struct igbvf_adapter *adapter = netdev_priv(netdev);
2912 struct e1000_hw *hw = &adapter->hw;
2913
2914 /* The watchdog timer may be rescheduled, so explicitly
2915 * disable it from being rescheduled.
2916 */
2917 set_bit(__IGBVF_DOWN, &adapter->state);
2918 del_timer_sync(&adapter->watchdog_timer);
2919
2920 cancel_work_sync(&adapter->reset_task);
2921 cancel_work_sync(&adapter->watchdog_task);
2922
2923 unregister_netdev(netdev);
2924
2925 igbvf_reset_interrupt_capability(adapter);
2926
2927 /* it is important to delete the NAPI struct prior to freeing the
2928 * Rx ring so that you do not end up with null pointer refs
2929 */
2930 netif_napi_del(&adapter->rx_ring->napi);
2931 kfree(adapter->tx_ring);
2932 kfree(adapter->rx_ring);
2933
2934 iounmap(hw->hw_addr);
2935 if (hw->flash_address)
2936 iounmap(hw->flash_address);
2937 pci_release_regions(pdev);
2938
2939 free_netdev(netdev);
2940
2941 pci_disable_device(pdev);
2942}
2943
2944/* PCI Error Recovery (ERS) */
2945static const struct pci_error_handlers igbvf_err_handler = {
2946 .error_detected = igbvf_io_error_detected,
2947 .slot_reset = igbvf_io_slot_reset,
2948 .resume = igbvf_io_resume,
2949 .reset_prepare = igbvf_io_prepare,
2950 .reset_done = igbvf_io_reset_done,
2951};
2952
2953static const struct pci_device_id igbvf_pci_tbl[] = {
2954 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2955 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf },
2956 { } /* terminate list */
2957};
2958MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2959
2960static SIMPLE_DEV_PM_OPS(igbvf_pm_ops, igbvf_suspend, igbvf_resume);
2961
2962/* PCI Device API Driver */
2963static struct pci_driver igbvf_driver = {
2964 .name = igbvf_driver_name,
2965 .id_table = igbvf_pci_tbl,
2966 .probe = igbvf_probe,
2967 .remove = igbvf_remove,
2968 .driver.pm = &igbvf_pm_ops,
2969 .shutdown = igbvf_shutdown,
2970 .err_handler = &igbvf_err_handler
2971};
2972
2973/**
2974 * igbvf_init_module - Driver Registration Routine
2975 *
2976 * igbvf_init_module is the first routine called when the driver is
2977 * loaded. All it does is register with the PCI subsystem.
2978 **/
2979static int __init igbvf_init_module(void)
2980{
2981 int ret;
2982
2983 pr_info("%s\n", igbvf_driver_string);
2984 pr_info("%s\n", igbvf_copyright);
2985
2986 ret = pci_register_driver(&igbvf_driver);
2987
2988 return ret;
2989}
2990module_init(igbvf_init_module);
2991
2992/**
2993 * igbvf_exit_module - Driver Exit Cleanup Routine
2994 *
2995 * igbvf_exit_module is called just before the driver is removed
2996 * from memory.
2997 **/
2998static void __exit igbvf_exit_module(void)
2999{
3000 pci_unregister_driver(&igbvf_driver);
3001}
3002module_exit(igbvf_exit_module);
3003
3004MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
3005MODULE_DESCRIPTION("Intel(R) Gigabit Virtual Function Network Driver");
3006MODULE_LICENSE("GPL v2");
3007
3008/* netdev.c */
1/*******************************************************************************
2
3 Intel(R) 82576 Virtual Function Linux driver
4 Copyright(c) 2009 - 2012 Intel Corporation.
5
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
9
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 more details.
14
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, see <http://www.gnu.org/licenses/>.
17
18 The full GNU General Public License is included in this distribution in
19 the file called "COPYING".
20
21 Contact Information:
22 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
23 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
24
25*******************************************************************************/
26
27#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
28
29#include <linux/module.h>
30#include <linux/types.h>
31#include <linux/init.h>
32#include <linux/pci.h>
33#include <linux/vmalloc.h>
34#include <linux/pagemap.h>
35#include <linux/delay.h>
36#include <linux/netdevice.h>
37#include <linux/tcp.h>
38#include <linux/ipv6.h>
39#include <linux/slab.h>
40#include <net/checksum.h>
41#include <net/ip6_checksum.h>
42#include <linux/mii.h>
43#include <linux/ethtool.h>
44#include <linux/if_vlan.h>
45#include <linux/prefetch.h>
46#include <linux/sctp.h>
47
48#include "igbvf.h"
49
50#define DRV_VERSION "2.4.0-k"
51char igbvf_driver_name[] = "igbvf";
52const char igbvf_driver_version[] = DRV_VERSION;
53static const char igbvf_driver_string[] =
54 "Intel(R) Gigabit Virtual Function Network Driver";
55static const char igbvf_copyright[] =
56 "Copyright (c) 2009 - 2012 Intel Corporation.";
57
58#define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
59static int debug = -1;
60module_param(debug, int, 0);
61MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
62
63static int igbvf_poll(struct napi_struct *napi, int budget);
64static void igbvf_reset(struct igbvf_adapter *);
65static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
66static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
67
68static struct igbvf_info igbvf_vf_info = {
69 .mac = e1000_vfadapt,
70 .flags = 0,
71 .pba = 10,
72 .init_ops = e1000_init_function_pointers_vf,
73};
74
75static struct igbvf_info igbvf_i350_vf_info = {
76 .mac = e1000_vfadapt_i350,
77 .flags = 0,
78 .pba = 10,
79 .init_ops = e1000_init_function_pointers_vf,
80};
81
82static const struct igbvf_info *igbvf_info_tbl[] = {
83 [board_vf] = &igbvf_vf_info,
84 [board_i350_vf] = &igbvf_i350_vf_info,
85};
86
87/**
88 * igbvf_desc_unused - calculate if we have unused descriptors
89 * @rx_ring: address of receive ring structure
90 **/
91static int igbvf_desc_unused(struct igbvf_ring *ring)
92{
93 if (ring->next_to_clean > ring->next_to_use)
94 return ring->next_to_clean - ring->next_to_use - 1;
95
96 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
97}
98
99/**
100 * igbvf_receive_skb - helper function to handle Rx indications
101 * @adapter: board private structure
102 * @status: descriptor status field as written by hardware
103 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
104 * @skb: pointer to sk_buff to be indicated to stack
105 **/
106static void igbvf_receive_skb(struct igbvf_adapter *adapter,
107 struct net_device *netdev,
108 struct sk_buff *skb,
109 u32 status, u16 vlan)
110{
111 u16 vid;
112
113 if (status & E1000_RXD_STAT_VP) {
114 if ((adapter->flags & IGBVF_FLAG_RX_LB_VLAN_BSWAP) &&
115 (status & E1000_RXDEXT_STATERR_LB))
116 vid = be16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
117 else
118 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
119 if (test_bit(vid, adapter->active_vlans))
120 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
121 }
122
123 napi_gro_receive(&adapter->rx_ring->napi, skb);
124}
125
126static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
127 u32 status_err, struct sk_buff *skb)
128{
129 skb_checksum_none_assert(skb);
130
131 /* Ignore Checksum bit is set or checksum is disabled through ethtool */
132 if ((status_err & E1000_RXD_STAT_IXSM) ||
133 (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
134 return;
135
136 /* TCP/UDP checksum error bit is set */
137 if (status_err &
138 (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
139 /* let the stack verify checksum errors */
140 adapter->hw_csum_err++;
141 return;
142 }
143
144 /* It must be a TCP or UDP packet with a valid checksum */
145 if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
146 skb->ip_summed = CHECKSUM_UNNECESSARY;
147
148 adapter->hw_csum_good++;
149}
150
151/**
152 * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
153 * @rx_ring: address of ring structure to repopulate
154 * @cleaned_count: number of buffers to repopulate
155 **/
156static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
157 int cleaned_count)
158{
159 struct igbvf_adapter *adapter = rx_ring->adapter;
160 struct net_device *netdev = adapter->netdev;
161 struct pci_dev *pdev = adapter->pdev;
162 union e1000_adv_rx_desc *rx_desc;
163 struct igbvf_buffer *buffer_info;
164 struct sk_buff *skb;
165 unsigned int i;
166 int bufsz;
167
168 i = rx_ring->next_to_use;
169 buffer_info = &rx_ring->buffer_info[i];
170
171 if (adapter->rx_ps_hdr_size)
172 bufsz = adapter->rx_ps_hdr_size;
173 else
174 bufsz = adapter->rx_buffer_len;
175
176 while (cleaned_count--) {
177 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
178
179 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
180 if (!buffer_info->page) {
181 buffer_info->page = alloc_page(GFP_ATOMIC);
182 if (!buffer_info->page) {
183 adapter->alloc_rx_buff_failed++;
184 goto no_buffers;
185 }
186 buffer_info->page_offset = 0;
187 } else {
188 buffer_info->page_offset ^= PAGE_SIZE / 2;
189 }
190 buffer_info->page_dma =
191 dma_map_page(&pdev->dev, buffer_info->page,
192 buffer_info->page_offset,
193 PAGE_SIZE / 2,
194 DMA_FROM_DEVICE);
195 if (dma_mapping_error(&pdev->dev,
196 buffer_info->page_dma)) {
197 __free_page(buffer_info->page);
198 buffer_info->page = NULL;
199 dev_err(&pdev->dev, "RX DMA map failed\n");
200 break;
201 }
202 }
203
204 if (!buffer_info->skb) {
205 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
206 if (!skb) {
207 adapter->alloc_rx_buff_failed++;
208 goto no_buffers;
209 }
210
211 buffer_info->skb = skb;
212 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
213 bufsz,
214 DMA_FROM_DEVICE);
215 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
216 dev_kfree_skb(buffer_info->skb);
217 buffer_info->skb = NULL;
218 dev_err(&pdev->dev, "RX DMA map failed\n");
219 goto no_buffers;
220 }
221 }
222 /* Refresh the desc even if buffer_addrs didn't change because
223 * each write-back erases this info.
224 */
225 if (adapter->rx_ps_hdr_size) {
226 rx_desc->read.pkt_addr =
227 cpu_to_le64(buffer_info->page_dma);
228 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
229 } else {
230 rx_desc->read.pkt_addr = cpu_to_le64(buffer_info->dma);
231 rx_desc->read.hdr_addr = 0;
232 }
233
234 i++;
235 if (i == rx_ring->count)
236 i = 0;
237 buffer_info = &rx_ring->buffer_info[i];
238 }
239
240no_buffers:
241 if (rx_ring->next_to_use != i) {
242 rx_ring->next_to_use = i;
243 if (i == 0)
244 i = (rx_ring->count - 1);
245 else
246 i--;
247
248 /* Force memory writes to complete before letting h/w
249 * know there are new descriptors to fetch. (Only
250 * applicable for weak-ordered memory model archs,
251 * such as IA-64).
252 */
253 wmb();
254 writel(i, adapter->hw.hw_addr + rx_ring->tail);
255 }
256}
257
258/**
259 * igbvf_clean_rx_irq - Send received data up the network stack; legacy
260 * @adapter: board private structure
261 *
262 * the return value indicates whether actual cleaning was done, there
263 * is no guarantee that everything was cleaned
264 **/
265static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
266 int *work_done, int work_to_do)
267{
268 struct igbvf_ring *rx_ring = adapter->rx_ring;
269 struct net_device *netdev = adapter->netdev;
270 struct pci_dev *pdev = adapter->pdev;
271 union e1000_adv_rx_desc *rx_desc, *next_rxd;
272 struct igbvf_buffer *buffer_info, *next_buffer;
273 struct sk_buff *skb;
274 bool cleaned = false;
275 int cleaned_count = 0;
276 unsigned int total_bytes = 0, total_packets = 0;
277 unsigned int i;
278 u32 length, hlen, staterr;
279
280 i = rx_ring->next_to_clean;
281 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
282 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
283
284 while (staterr & E1000_RXD_STAT_DD) {
285 if (*work_done >= work_to_do)
286 break;
287 (*work_done)++;
288 rmb(); /* read descriptor and rx_buffer_info after status DD */
289
290 buffer_info = &rx_ring->buffer_info[i];
291
292 /* HW will not DMA in data larger than the given buffer, even
293 * if it parses the (NFS, of course) header to be larger. In
294 * that case, it fills the header buffer and spills the rest
295 * into the page.
296 */
297 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info)
298 & E1000_RXDADV_HDRBUFLEN_MASK) >>
299 E1000_RXDADV_HDRBUFLEN_SHIFT;
300 if (hlen > adapter->rx_ps_hdr_size)
301 hlen = adapter->rx_ps_hdr_size;
302
303 length = le16_to_cpu(rx_desc->wb.upper.length);
304 cleaned = true;
305 cleaned_count++;
306
307 skb = buffer_info->skb;
308 prefetch(skb->data - NET_IP_ALIGN);
309 buffer_info->skb = NULL;
310 if (!adapter->rx_ps_hdr_size) {
311 dma_unmap_single(&pdev->dev, buffer_info->dma,
312 adapter->rx_buffer_len,
313 DMA_FROM_DEVICE);
314 buffer_info->dma = 0;
315 skb_put(skb, length);
316 goto send_up;
317 }
318
319 if (!skb_shinfo(skb)->nr_frags) {
320 dma_unmap_single(&pdev->dev, buffer_info->dma,
321 adapter->rx_ps_hdr_size,
322 DMA_FROM_DEVICE);
323 buffer_info->dma = 0;
324 skb_put(skb, hlen);
325 }
326
327 if (length) {
328 dma_unmap_page(&pdev->dev, buffer_info->page_dma,
329 PAGE_SIZE / 2,
330 DMA_FROM_DEVICE);
331 buffer_info->page_dma = 0;
332
333 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
334 buffer_info->page,
335 buffer_info->page_offset,
336 length);
337
338 if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
339 (page_count(buffer_info->page) != 1))
340 buffer_info->page = NULL;
341 else
342 get_page(buffer_info->page);
343
344 skb->len += length;
345 skb->data_len += length;
346 skb->truesize += PAGE_SIZE / 2;
347 }
348send_up:
349 i++;
350 if (i == rx_ring->count)
351 i = 0;
352 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
353 prefetch(next_rxd);
354 next_buffer = &rx_ring->buffer_info[i];
355
356 if (!(staterr & E1000_RXD_STAT_EOP)) {
357 buffer_info->skb = next_buffer->skb;
358 buffer_info->dma = next_buffer->dma;
359 next_buffer->skb = skb;
360 next_buffer->dma = 0;
361 goto next_desc;
362 }
363
364 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
365 dev_kfree_skb_irq(skb);
366 goto next_desc;
367 }
368
369 total_bytes += skb->len;
370 total_packets++;
371
372 igbvf_rx_checksum_adv(adapter, staterr, skb);
373
374 skb->protocol = eth_type_trans(skb, netdev);
375
376 igbvf_receive_skb(adapter, netdev, skb, staterr,
377 rx_desc->wb.upper.vlan);
378
379next_desc:
380 rx_desc->wb.upper.status_error = 0;
381
382 /* return some buffers to hardware, one at a time is too slow */
383 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
384 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
385 cleaned_count = 0;
386 }
387
388 /* use prefetched values */
389 rx_desc = next_rxd;
390 buffer_info = next_buffer;
391
392 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
393 }
394
395 rx_ring->next_to_clean = i;
396 cleaned_count = igbvf_desc_unused(rx_ring);
397
398 if (cleaned_count)
399 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
400
401 adapter->total_rx_packets += total_packets;
402 adapter->total_rx_bytes += total_bytes;
403 adapter->net_stats.rx_bytes += total_bytes;
404 adapter->net_stats.rx_packets += total_packets;
405 return cleaned;
406}
407
408static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
409 struct igbvf_buffer *buffer_info)
410{
411 if (buffer_info->dma) {
412 if (buffer_info->mapped_as_page)
413 dma_unmap_page(&adapter->pdev->dev,
414 buffer_info->dma,
415 buffer_info->length,
416 DMA_TO_DEVICE);
417 else
418 dma_unmap_single(&adapter->pdev->dev,
419 buffer_info->dma,
420 buffer_info->length,
421 DMA_TO_DEVICE);
422 buffer_info->dma = 0;
423 }
424 if (buffer_info->skb) {
425 dev_kfree_skb_any(buffer_info->skb);
426 buffer_info->skb = NULL;
427 }
428 buffer_info->time_stamp = 0;
429}
430
431/**
432 * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
433 * @adapter: board private structure
434 *
435 * Return 0 on success, negative on failure
436 **/
437int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
438 struct igbvf_ring *tx_ring)
439{
440 struct pci_dev *pdev = adapter->pdev;
441 int size;
442
443 size = sizeof(struct igbvf_buffer) * tx_ring->count;
444 tx_ring->buffer_info = vzalloc(size);
445 if (!tx_ring->buffer_info)
446 goto err;
447
448 /* round up to nearest 4K */
449 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
450 tx_ring->size = ALIGN(tx_ring->size, 4096);
451
452 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
453 &tx_ring->dma, GFP_KERNEL);
454 if (!tx_ring->desc)
455 goto err;
456
457 tx_ring->adapter = adapter;
458 tx_ring->next_to_use = 0;
459 tx_ring->next_to_clean = 0;
460
461 return 0;
462err:
463 vfree(tx_ring->buffer_info);
464 dev_err(&adapter->pdev->dev,
465 "Unable to allocate memory for the transmit descriptor ring\n");
466 return -ENOMEM;
467}
468
469/**
470 * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
471 * @adapter: board private structure
472 *
473 * Returns 0 on success, negative on failure
474 **/
475int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
476 struct igbvf_ring *rx_ring)
477{
478 struct pci_dev *pdev = adapter->pdev;
479 int size, desc_len;
480
481 size = sizeof(struct igbvf_buffer) * rx_ring->count;
482 rx_ring->buffer_info = vzalloc(size);
483 if (!rx_ring->buffer_info)
484 goto err;
485
486 desc_len = sizeof(union e1000_adv_rx_desc);
487
488 /* Round up to nearest 4K */
489 rx_ring->size = rx_ring->count * desc_len;
490 rx_ring->size = ALIGN(rx_ring->size, 4096);
491
492 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
493 &rx_ring->dma, GFP_KERNEL);
494 if (!rx_ring->desc)
495 goto err;
496
497 rx_ring->next_to_clean = 0;
498 rx_ring->next_to_use = 0;
499
500 rx_ring->adapter = adapter;
501
502 return 0;
503
504err:
505 vfree(rx_ring->buffer_info);
506 rx_ring->buffer_info = NULL;
507 dev_err(&adapter->pdev->dev,
508 "Unable to allocate memory for the receive descriptor ring\n");
509 return -ENOMEM;
510}
511
512/**
513 * igbvf_clean_tx_ring - Free Tx Buffers
514 * @tx_ring: ring to be cleaned
515 **/
516static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
517{
518 struct igbvf_adapter *adapter = tx_ring->adapter;
519 struct igbvf_buffer *buffer_info;
520 unsigned long size;
521 unsigned int i;
522
523 if (!tx_ring->buffer_info)
524 return;
525
526 /* Free all the Tx ring sk_buffs */
527 for (i = 0; i < tx_ring->count; i++) {
528 buffer_info = &tx_ring->buffer_info[i];
529 igbvf_put_txbuf(adapter, buffer_info);
530 }
531
532 size = sizeof(struct igbvf_buffer) * tx_ring->count;
533 memset(tx_ring->buffer_info, 0, size);
534
535 /* Zero out the descriptor ring */
536 memset(tx_ring->desc, 0, tx_ring->size);
537
538 tx_ring->next_to_use = 0;
539 tx_ring->next_to_clean = 0;
540
541 writel(0, adapter->hw.hw_addr + tx_ring->head);
542 writel(0, adapter->hw.hw_addr + tx_ring->tail);
543}
544
545/**
546 * igbvf_free_tx_resources - Free Tx Resources per Queue
547 * @tx_ring: ring to free resources from
548 *
549 * Free all transmit software resources
550 **/
551void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
552{
553 struct pci_dev *pdev = tx_ring->adapter->pdev;
554
555 igbvf_clean_tx_ring(tx_ring);
556
557 vfree(tx_ring->buffer_info);
558 tx_ring->buffer_info = NULL;
559
560 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
561 tx_ring->dma);
562
563 tx_ring->desc = NULL;
564}
565
566/**
567 * igbvf_clean_rx_ring - Free Rx Buffers per Queue
568 * @adapter: board private structure
569 **/
570static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
571{
572 struct igbvf_adapter *adapter = rx_ring->adapter;
573 struct igbvf_buffer *buffer_info;
574 struct pci_dev *pdev = adapter->pdev;
575 unsigned long size;
576 unsigned int i;
577
578 if (!rx_ring->buffer_info)
579 return;
580
581 /* Free all the Rx ring sk_buffs */
582 for (i = 0; i < rx_ring->count; i++) {
583 buffer_info = &rx_ring->buffer_info[i];
584 if (buffer_info->dma) {
585 if (adapter->rx_ps_hdr_size) {
586 dma_unmap_single(&pdev->dev, buffer_info->dma,
587 adapter->rx_ps_hdr_size,
588 DMA_FROM_DEVICE);
589 } else {
590 dma_unmap_single(&pdev->dev, buffer_info->dma,
591 adapter->rx_buffer_len,
592 DMA_FROM_DEVICE);
593 }
594 buffer_info->dma = 0;
595 }
596
597 if (buffer_info->skb) {
598 dev_kfree_skb(buffer_info->skb);
599 buffer_info->skb = NULL;
600 }
601
602 if (buffer_info->page) {
603 if (buffer_info->page_dma)
604 dma_unmap_page(&pdev->dev,
605 buffer_info->page_dma,
606 PAGE_SIZE / 2,
607 DMA_FROM_DEVICE);
608 put_page(buffer_info->page);
609 buffer_info->page = NULL;
610 buffer_info->page_dma = 0;
611 buffer_info->page_offset = 0;
612 }
613 }
614
615 size = sizeof(struct igbvf_buffer) * rx_ring->count;
616 memset(rx_ring->buffer_info, 0, size);
617
618 /* Zero out the descriptor ring */
619 memset(rx_ring->desc, 0, rx_ring->size);
620
621 rx_ring->next_to_clean = 0;
622 rx_ring->next_to_use = 0;
623
624 writel(0, adapter->hw.hw_addr + rx_ring->head);
625 writel(0, adapter->hw.hw_addr + rx_ring->tail);
626}
627
628/**
629 * igbvf_free_rx_resources - Free Rx Resources
630 * @rx_ring: ring to clean the resources from
631 *
632 * Free all receive software resources
633 **/
634
635void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
636{
637 struct pci_dev *pdev = rx_ring->adapter->pdev;
638
639 igbvf_clean_rx_ring(rx_ring);
640
641 vfree(rx_ring->buffer_info);
642 rx_ring->buffer_info = NULL;
643
644 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
645 rx_ring->dma);
646 rx_ring->desc = NULL;
647}
648
649/**
650 * igbvf_update_itr - update the dynamic ITR value based on statistics
651 * @adapter: pointer to adapter
652 * @itr_setting: current adapter->itr
653 * @packets: the number of packets during this measurement interval
654 * @bytes: the number of bytes during this measurement interval
655 *
656 * Stores a new ITR value based on packets and byte counts during the last
657 * interrupt. The advantage of per interrupt computation is faster updates
658 * and more accurate ITR for the current traffic pattern. Constants in this
659 * function were computed based on theoretical maximum wire speed and thresholds
660 * were set based on testing data as well as attempting to minimize response
661 * time while increasing bulk throughput.
662 **/
663static enum latency_range igbvf_update_itr(struct igbvf_adapter *adapter,
664 enum latency_range itr_setting,
665 int packets, int bytes)
666{
667 enum latency_range retval = itr_setting;
668
669 if (packets == 0)
670 goto update_itr_done;
671
672 switch (itr_setting) {
673 case lowest_latency:
674 /* handle TSO and jumbo frames */
675 if (bytes/packets > 8000)
676 retval = bulk_latency;
677 else if ((packets < 5) && (bytes > 512))
678 retval = low_latency;
679 break;
680 case low_latency: /* 50 usec aka 20000 ints/s */
681 if (bytes > 10000) {
682 /* this if handles the TSO accounting */
683 if (bytes/packets > 8000)
684 retval = bulk_latency;
685 else if ((packets < 10) || ((bytes/packets) > 1200))
686 retval = bulk_latency;
687 else if ((packets > 35))
688 retval = lowest_latency;
689 } else if (bytes/packets > 2000) {
690 retval = bulk_latency;
691 } else if (packets <= 2 && bytes < 512) {
692 retval = lowest_latency;
693 }
694 break;
695 case bulk_latency: /* 250 usec aka 4000 ints/s */
696 if (bytes > 25000) {
697 if (packets > 35)
698 retval = low_latency;
699 } else if (bytes < 6000) {
700 retval = low_latency;
701 }
702 break;
703 default:
704 break;
705 }
706
707update_itr_done:
708 return retval;
709}
710
711static int igbvf_range_to_itr(enum latency_range current_range)
712{
713 int new_itr;
714
715 switch (current_range) {
716 /* counts and packets in update_itr are dependent on these numbers */
717 case lowest_latency:
718 new_itr = IGBVF_70K_ITR;
719 break;
720 case low_latency:
721 new_itr = IGBVF_20K_ITR;
722 break;
723 case bulk_latency:
724 new_itr = IGBVF_4K_ITR;
725 break;
726 default:
727 new_itr = IGBVF_START_ITR;
728 break;
729 }
730 return new_itr;
731}
732
733static void igbvf_set_itr(struct igbvf_adapter *adapter)
734{
735 u32 new_itr;
736
737 adapter->tx_ring->itr_range =
738 igbvf_update_itr(adapter,
739 adapter->tx_ring->itr_val,
740 adapter->total_tx_packets,
741 adapter->total_tx_bytes);
742
743 /* conservative mode (itr 3) eliminates the lowest_latency setting */
744 if (adapter->requested_itr == 3 &&
745 adapter->tx_ring->itr_range == lowest_latency)
746 adapter->tx_ring->itr_range = low_latency;
747
748 new_itr = igbvf_range_to_itr(adapter->tx_ring->itr_range);
749
750 if (new_itr != adapter->tx_ring->itr_val) {
751 u32 current_itr = adapter->tx_ring->itr_val;
752 /* this attempts to bias the interrupt rate towards Bulk
753 * by adding intermediate steps when interrupt rate is
754 * increasing
755 */
756 new_itr = new_itr > current_itr ?
757 min(current_itr + (new_itr >> 2), new_itr) :
758 new_itr;
759 adapter->tx_ring->itr_val = new_itr;
760
761 adapter->tx_ring->set_itr = 1;
762 }
763
764 adapter->rx_ring->itr_range =
765 igbvf_update_itr(adapter, adapter->rx_ring->itr_val,
766 adapter->total_rx_packets,
767 adapter->total_rx_bytes);
768 if (adapter->requested_itr == 3 &&
769 adapter->rx_ring->itr_range == lowest_latency)
770 adapter->rx_ring->itr_range = low_latency;
771
772 new_itr = igbvf_range_to_itr(adapter->rx_ring->itr_range);
773
774 if (new_itr != adapter->rx_ring->itr_val) {
775 u32 current_itr = adapter->rx_ring->itr_val;
776
777 new_itr = new_itr > current_itr ?
778 min(current_itr + (new_itr >> 2), new_itr) :
779 new_itr;
780 adapter->rx_ring->itr_val = new_itr;
781
782 adapter->rx_ring->set_itr = 1;
783 }
784}
785
786/**
787 * igbvf_clean_tx_irq - Reclaim resources after transmit completes
788 * @adapter: board private structure
789 *
790 * returns true if ring is completely cleaned
791 **/
792static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
793{
794 struct igbvf_adapter *adapter = tx_ring->adapter;
795 struct net_device *netdev = adapter->netdev;
796 struct igbvf_buffer *buffer_info;
797 struct sk_buff *skb;
798 union e1000_adv_tx_desc *tx_desc, *eop_desc;
799 unsigned int total_bytes = 0, total_packets = 0;
800 unsigned int i, count = 0;
801 bool cleaned = false;
802
803 i = tx_ring->next_to_clean;
804 buffer_info = &tx_ring->buffer_info[i];
805 eop_desc = buffer_info->next_to_watch;
806
807 do {
808 /* if next_to_watch is not set then there is no work pending */
809 if (!eop_desc)
810 break;
811
812 /* prevent any other reads prior to eop_desc */
813 read_barrier_depends();
814
815 /* if DD is not set pending work has not been completed */
816 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
817 break;
818
819 /* clear next_to_watch to prevent false hangs */
820 buffer_info->next_to_watch = NULL;
821
822 for (cleaned = false; !cleaned; count++) {
823 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
824 cleaned = (tx_desc == eop_desc);
825 skb = buffer_info->skb;
826
827 if (skb) {
828 unsigned int segs, bytecount;
829
830 /* gso_segs is currently only valid for tcp */
831 segs = skb_shinfo(skb)->gso_segs ?: 1;
832 /* multiply data chunks by size of headers */
833 bytecount = ((segs - 1) * skb_headlen(skb)) +
834 skb->len;
835 total_packets += segs;
836 total_bytes += bytecount;
837 }
838
839 igbvf_put_txbuf(adapter, buffer_info);
840 tx_desc->wb.status = 0;
841
842 i++;
843 if (i == tx_ring->count)
844 i = 0;
845
846 buffer_info = &tx_ring->buffer_info[i];
847 }
848
849 eop_desc = buffer_info->next_to_watch;
850 } while (count < tx_ring->count);
851
852 tx_ring->next_to_clean = i;
853
854 if (unlikely(count && netif_carrier_ok(netdev) &&
855 igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
856 /* Make sure that anybody stopping the queue after this
857 * sees the new next_to_clean.
858 */
859 smp_mb();
860 if (netif_queue_stopped(netdev) &&
861 !(test_bit(__IGBVF_DOWN, &adapter->state))) {
862 netif_wake_queue(netdev);
863 ++adapter->restart_queue;
864 }
865 }
866
867 adapter->net_stats.tx_bytes += total_bytes;
868 adapter->net_stats.tx_packets += total_packets;
869 return count < tx_ring->count;
870}
871
872static irqreturn_t igbvf_msix_other(int irq, void *data)
873{
874 struct net_device *netdev = data;
875 struct igbvf_adapter *adapter = netdev_priv(netdev);
876 struct e1000_hw *hw = &adapter->hw;
877
878 adapter->int_counter1++;
879
880 hw->mac.get_link_status = 1;
881 if (!test_bit(__IGBVF_DOWN, &adapter->state))
882 mod_timer(&adapter->watchdog_timer, jiffies + 1);
883
884 ew32(EIMS, adapter->eims_other);
885
886 return IRQ_HANDLED;
887}
888
889static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
890{
891 struct net_device *netdev = data;
892 struct igbvf_adapter *adapter = netdev_priv(netdev);
893 struct e1000_hw *hw = &adapter->hw;
894 struct igbvf_ring *tx_ring = adapter->tx_ring;
895
896 if (tx_ring->set_itr) {
897 writel(tx_ring->itr_val,
898 adapter->hw.hw_addr + tx_ring->itr_register);
899 adapter->tx_ring->set_itr = 0;
900 }
901
902 adapter->total_tx_bytes = 0;
903 adapter->total_tx_packets = 0;
904
905 /* auto mask will automatically re-enable the interrupt when we write
906 * EICS
907 */
908 if (!igbvf_clean_tx_irq(tx_ring))
909 /* Ring was not completely cleaned, so fire another interrupt */
910 ew32(EICS, tx_ring->eims_value);
911 else
912 ew32(EIMS, tx_ring->eims_value);
913
914 return IRQ_HANDLED;
915}
916
917static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
918{
919 struct net_device *netdev = data;
920 struct igbvf_adapter *adapter = netdev_priv(netdev);
921
922 adapter->int_counter0++;
923
924 /* Write the ITR value calculated at the end of the
925 * previous interrupt.
926 */
927 if (adapter->rx_ring->set_itr) {
928 writel(adapter->rx_ring->itr_val,
929 adapter->hw.hw_addr + adapter->rx_ring->itr_register);
930 adapter->rx_ring->set_itr = 0;
931 }
932
933 if (napi_schedule_prep(&adapter->rx_ring->napi)) {
934 adapter->total_rx_bytes = 0;
935 adapter->total_rx_packets = 0;
936 __napi_schedule(&adapter->rx_ring->napi);
937 }
938
939 return IRQ_HANDLED;
940}
941
942#define IGBVF_NO_QUEUE -1
943
944static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
945 int tx_queue, int msix_vector)
946{
947 struct e1000_hw *hw = &adapter->hw;
948 u32 ivar, index;
949
950 /* 82576 uses a table-based method for assigning vectors.
951 * Each queue has a single entry in the table to which we write
952 * a vector number along with a "valid" bit. Sadly, the layout
953 * of the table is somewhat counterintuitive.
954 */
955 if (rx_queue > IGBVF_NO_QUEUE) {
956 index = (rx_queue >> 1);
957 ivar = array_er32(IVAR0, index);
958 if (rx_queue & 0x1) {
959 /* vector goes into third byte of register */
960 ivar = ivar & 0xFF00FFFF;
961 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
962 } else {
963 /* vector goes into low byte of register */
964 ivar = ivar & 0xFFFFFF00;
965 ivar |= msix_vector | E1000_IVAR_VALID;
966 }
967 adapter->rx_ring[rx_queue].eims_value = BIT(msix_vector);
968 array_ew32(IVAR0, index, ivar);
969 }
970 if (tx_queue > IGBVF_NO_QUEUE) {
971 index = (tx_queue >> 1);
972 ivar = array_er32(IVAR0, index);
973 if (tx_queue & 0x1) {
974 /* vector goes into high byte of register */
975 ivar = ivar & 0x00FFFFFF;
976 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
977 } else {
978 /* vector goes into second byte of register */
979 ivar = ivar & 0xFFFF00FF;
980 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
981 }
982 adapter->tx_ring[tx_queue].eims_value = BIT(msix_vector);
983 array_ew32(IVAR0, index, ivar);
984 }
985}
986
987/**
988 * igbvf_configure_msix - Configure MSI-X hardware
989 * @adapter: board private structure
990 *
991 * igbvf_configure_msix sets up the hardware to properly
992 * generate MSI-X interrupts.
993 **/
994static void igbvf_configure_msix(struct igbvf_adapter *adapter)
995{
996 u32 tmp;
997 struct e1000_hw *hw = &adapter->hw;
998 struct igbvf_ring *tx_ring = adapter->tx_ring;
999 struct igbvf_ring *rx_ring = adapter->rx_ring;
1000 int vector = 0;
1001
1002 adapter->eims_enable_mask = 0;
1003
1004 igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
1005 adapter->eims_enable_mask |= tx_ring->eims_value;
1006 writel(tx_ring->itr_val, hw->hw_addr + tx_ring->itr_register);
1007 igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
1008 adapter->eims_enable_mask |= rx_ring->eims_value;
1009 writel(rx_ring->itr_val, hw->hw_addr + rx_ring->itr_register);
1010
1011 /* set vector for other causes, i.e. link changes */
1012
1013 tmp = (vector++ | E1000_IVAR_VALID);
1014
1015 ew32(IVAR_MISC, tmp);
1016
1017 adapter->eims_enable_mask = GENMASK(vector - 1, 0);
1018 adapter->eims_other = BIT(vector - 1);
1019 e1e_flush();
1020}
1021
1022static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
1023{
1024 if (adapter->msix_entries) {
1025 pci_disable_msix(adapter->pdev);
1026 kfree(adapter->msix_entries);
1027 adapter->msix_entries = NULL;
1028 }
1029}
1030
1031/**
1032 * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1033 * @adapter: board private structure
1034 *
1035 * Attempt to configure interrupts using the best available
1036 * capabilities of the hardware and kernel.
1037 **/
1038static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1039{
1040 int err = -ENOMEM;
1041 int i;
1042
1043 /* we allocate 3 vectors, 1 for Tx, 1 for Rx, one for PF messages */
1044 adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1045 GFP_KERNEL);
1046 if (adapter->msix_entries) {
1047 for (i = 0; i < 3; i++)
1048 adapter->msix_entries[i].entry = i;
1049
1050 err = pci_enable_msix_range(adapter->pdev,
1051 adapter->msix_entries, 3, 3);
1052 }
1053
1054 if (err < 0) {
1055 /* MSI-X failed */
1056 dev_err(&adapter->pdev->dev,
1057 "Failed to initialize MSI-X interrupts.\n");
1058 igbvf_reset_interrupt_capability(adapter);
1059 }
1060}
1061
1062/**
1063 * igbvf_request_msix - Initialize MSI-X interrupts
1064 * @adapter: board private structure
1065 *
1066 * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1067 * kernel.
1068 **/
1069static int igbvf_request_msix(struct igbvf_adapter *adapter)
1070{
1071 struct net_device *netdev = adapter->netdev;
1072 int err = 0, vector = 0;
1073
1074 if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1075 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1076 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1077 } else {
1078 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1079 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1080 }
1081
1082 err = request_irq(adapter->msix_entries[vector].vector,
1083 igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1084 netdev);
1085 if (err)
1086 goto out;
1087
1088 adapter->tx_ring->itr_register = E1000_EITR(vector);
1089 adapter->tx_ring->itr_val = adapter->current_itr;
1090 vector++;
1091
1092 err = request_irq(adapter->msix_entries[vector].vector,
1093 igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1094 netdev);
1095 if (err)
1096 goto out;
1097
1098 adapter->rx_ring->itr_register = E1000_EITR(vector);
1099 adapter->rx_ring->itr_val = adapter->current_itr;
1100 vector++;
1101
1102 err = request_irq(adapter->msix_entries[vector].vector,
1103 igbvf_msix_other, 0, netdev->name, netdev);
1104 if (err)
1105 goto out;
1106
1107 igbvf_configure_msix(adapter);
1108 return 0;
1109out:
1110 return err;
1111}
1112
1113/**
1114 * igbvf_alloc_queues - Allocate memory for all rings
1115 * @adapter: board private structure to initialize
1116 **/
1117static int igbvf_alloc_queues(struct igbvf_adapter *adapter)
1118{
1119 struct net_device *netdev = adapter->netdev;
1120
1121 adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1122 if (!adapter->tx_ring)
1123 return -ENOMEM;
1124
1125 adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1126 if (!adapter->rx_ring) {
1127 kfree(adapter->tx_ring);
1128 return -ENOMEM;
1129 }
1130
1131 netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1132
1133 return 0;
1134}
1135
1136/**
1137 * igbvf_request_irq - initialize interrupts
1138 * @adapter: board private structure
1139 *
1140 * Attempts to configure interrupts using the best available
1141 * capabilities of the hardware and kernel.
1142 **/
1143static int igbvf_request_irq(struct igbvf_adapter *adapter)
1144{
1145 int err = -1;
1146
1147 /* igbvf supports msi-x only */
1148 if (adapter->msix_entries)
1149 err = igbvf_request_msix(adapter);
1150
1151 if (!err)
1152 return err;
1153
1154 dev_err(&adapter->pdev->dev,
1155 "Unable to allocate interrupt, Error: %d\n", err);
1156
1157 return err;
1158}
1159
1160static void igbvf_free_irq(struct igbvf_adapter *adapter)
1161{
1162 struct net_device *netdev = adapter->netdev;
1163 int vector;
1164
1165 if (adapter->msix_entries) {
1166 for (vector = 0; vector < 3; vector++)
1167 free_irq(adapter->msix_entries[vector].vector, netdev);
1168 }
1169}
1170
1171/**
1172 * igbvf_irq_disable - Mask off interrupt generation on the NIC
1173 * @adapter: board private structure
1174 **/
1175static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1176{
1177 struct e1000_hw *hw = &adapter->hw;
1178
1179 ew32(EIMC, ~0);
1180
1181 if (adapter->msix_entries)
1182 ew32(EIAC, 0);
1183}
1184
1185/**
1186 * igbvf_irq_enable - Enable default interrupt generation settings
1187 * @adapter: board private structure
1188 **/
1189static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1190{
1191 struct e1000_hw *hw = &adapter->hw;
1192
1193 ew32(EIAC, adapter->eims_enable_mask);
1194 ew32(EIAM, adapter->eims_enable_mask);
1195 ew32(EIMS, adapter->eims_enable_mask);
1196}
1197
1198/**
1199 * igbvf_poll - NAPI Rx polling callback
1200 * @napi: struct associated with this polling callback
1201 * @budget: amount of packets driver is allowed to process this poll
1202 **/
1203static int igbvf_poll(struct napi_struct *napi, int budget)
1204{
1205 struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1206 struct igbvf_adapter *adapter = rx_ring->adapter;
1207 struct e1000_hw *hw = &adapter->hw;
1208 int work_done = 0;
1209
1210 igbvf_clean_rx_irq(adapter, &work_done, budget);
1211
1212 /* If not enough Rx work done, exit the polling mode */
1213 if (work_done < budget) {
1214 napi_complete_done(napi, work_done);
1215
1216 if (adapter->requested_itr & 3)
1217 igbvf_set_itr(adapter);
1218
1219 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1220 ew32(EIMS, adapter->rx_ring->eims_value);
1221 }
1222
1223 return work_done;
1224}
1225
1226/**
1227 * igbvf_set_rlpml - set receive large packet maximum length
1228 * @adapter: board private structure
1229 *
1230 * Configure the maximum size of packets that will be received
1231 */
1232static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1233{
1234 int max_frame_size;
1235 struct e1000_hw *hw = &adapter->hw;
1236
1237 max_frame_size = adapter->max_frame_size + VLAN_TAG_SIZE;
1238 e1000_rlpml_set_vf(hw, max_frame_size);
1239}
1240
1241static int igbvf_vlan_rx_add_vid(struct net_device *netdev,
1242 __be16 proto, u16 vid)
1243{
1244 struct igbvf_adapter *adapter = netdev_priv(netdev);
1245 struct e1000_hw *hw = &adapter->hw;
1246
1247 if (hw->mac.ops.set_vfta(hw, vid, true)) {
1248 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1249 return -EINVAL;
1250 }
1251 set_bit(vid, adapter->active_vlans);
1252 return 0;
1253}
1254
1255static int igbvf_vlan_rx_kill_vid(struct net_device *netdev,
1256 __be16 proto, u16 vid)
1257{
1258 struct igbvf_adapter *adapter = netdev_priv(netdev);
1259 struct e1000_hw *hw = &adapter->hw;
1260
1261 if (hw->mac.ops.set_vfta(hw, vid, false)) {
1262 dev_err(&adapter->pdev->dev,
1263 "Failed to remove vlan id %d\n", vid);
1264 return -EINVAL;
1265 }
1266 clear_bit(vid, adapter->active_vlans);
1267 return 0;
1268}
1269
1270static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1271{
1272 u16 vid;
1273
1274 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1275 igbvf_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
1276}
1277
1278/**
1279 * igbvf_configure_tx - Configure Transmit Unit after Reset
1280 * @adapter: board private structure
1281 *
1282 * Configure the Tx unit of the MAC after a reset.
1283 **/
1284static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1285{
1286 struct e1000_hw *hw = &adapter->hw;
1287 struct igbvf_ring *tx_ring = adapter->tx_ring;
1288 u64 tdba;
1289 u32 txdctl, dca_txctrl;
1290
1291 /* disable transmits */
1292 txdctl = er32(TXDCTL(0));
1293 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1294 e1e_flush();
1295 msleep(10);
1296
1297 /* Setup the HW Tx Head and Tail descriptor pointers */
1298 ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1299 tdba = tx_ring->dma;
1300 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1301 ew32(TDBAH(0), (tdba >> 32));
1302 ew32(TDH(0), 0);
1303 ew32(TDT(0), 0);
1304 tx_ring->head = E1000_TDH(0);
1305 tx_ring->tail = E1000_TDT(0);
1306
1307 /* Turn off Relaxed Ordering on head write-backs. The writebacks
1308 * MUST be delivered in order or it will completely screw up
1309 * our bookkeeping.
1310 */
1311 dca_txctrl = er32(DCA_TXCTRL(0));
1312 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1313 ew32(DCA_TXCTRL(0), dca_txctrl);
1314
1315 /* enable transmits */
1316 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1317 ew32(TXDCTL(0), txdctl);
1318
1319 /* Setup Transmit Descriptor Settings for eop descriptor */
1320 adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1321
1322 /* enable Report Status bit */
1323 adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1324}
1325
1326/**
1327 * igbvf_setup_srrctl - configure the receive control registers
1328 * @adapter: Board private structure
1329 **/
1330static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1331{
1332 struct e1000_hw *hw = &adapter->hw;
1333 u32 srrctl = 0;
1334
1335 srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1336 E1000_SRRCTL_BSIZEHDR_MASK |
1337 E1000_SRRCTL_BSIZEPKT_MASK);
1338
1339 /* Enable queue drop to avoid head of line blocking */
1340 srrctl |= E1000_SRRCTL_DROP_EN;
1341
1342 /* Setup buffer sizes */
1343 srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1344 E1000_SRRCTL_BSIZEPKT_SHIFT;
1345
1346 if (adapter->rx_buffer_len < 2048) {
1347 adapter->rx_ps_hdr_size = 0;
1348 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1349 } else {
1350 adapter->rx_ps_hdr_size = 128;
1351 srrctl |= adapter->rx_ps_hdr_size <<
1352 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1353 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1354 }
1355
1356 ew32(SRRCTL(0), srrctl);
1357}
1358
1359/**
1360 * igbvf_configure_rx - Configure Receive Unit after Reset
1361 * @adapter: board private structure
1362 *
1363 * Configure the Rx unit of the MAC after a reset.
1364 **/
1365static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1366{
1367 struct e1000_hw *hw = &adapter->hw;
1368 struct igbvf_ring *rx_ring = adapter->rx_ring;
1369 u64 rdba;
1370 u32 rxdctl;
1371
1372 /* disable receives */
1373 rxdctl = er32(RXDCTL(0));
1374 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1375 e1e_flush();
1376 msleep(10);
1377
1378 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1379 * the Base and Length of the Rx Descriptor Ring
1380 */
1381 rdba = rx_ring->dma;
1382 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1383 ew32(RDBAH(0), (rdba >> 32));
1384 ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1385 rx_ring->head = E1000_RDH(0);
1386 rx_ring->tail = E1000_RDT(0);
1387 ew32(RDH(0), 0);
1388 ew32(RDT(0), 0);
1389
1390 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1391 rxdctl &= 0xFFF00000;
1392 rxdctl |= IGBVF_RX_PTHRESH;
1393 rxdctl |= IGBVF_RX_HTHRESH << 8;
1394 rxdctl |= IGBVF_RX_WTHRESH << 16;
1395
1396 igbvf_set_rlpml(adapter);
1397
1398 /* enable receives */
1399 ew32(RXDCTL(0), rxdctl);
1400}
1401
1402/**
1403 * igbvf_set_multi - Multicast and Promiscuous mode set
1404 * @netdev: network interface device structure
1405 *
1406 * The set_multi entry point is called whenever the multicast address
1407 * list or the network interface flags are updated. This routine is
1408 * responsible for configuring the hardware for proper multicast,
1409 * promiscuous mode, and all-multi behavior.
1410 **/
1411static void igbvf_set_multi(struct net_device *netdev)
1412{
1413 struct igbvf_adapter *adapter = netdev_priv(netdev);
1414 struct e1000_hw *hw = &adapter->hw;
1415 struct netdev_hw_addr *ha;
1416 u8 *mta_list = NULL;
1417 int i;
1418
1419 if (!netdev_mc_empty(netdev)) {
1420 mta_list = kmalloc_array(netdev_mc_count(netdev), ETH_ALEN,
1421 GFP_ATOMIC);
1422 if (!mta_list)
1423 return;
1424 }
1425
1426 /* prepare a packed array of only addresses. */
1427 i = 0;
1428 netdev_for_each_mc_addr(ha, netdev)
1429 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1430
1431 hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1432 kfree(mta_list);
1433}
1434
1435/**
1436 * igbvf_configure - configure the hardware for Rx and Tx
1437 * @adapter: private board structure
1438 **/
1439static void igbvf_configure(struct igbvf_adapter *adapter)
1440{
1441 igbvf_set_multi(adapter->netdev);
1442
1443 igbvf_restore_vlan(adapter);
1444
1445 igbvf_configure_tx(adapter);
1446 igbvf_setup_srrctl(adapter);
1447 igbvf_configure_rx(adapter);
1448 igbvf_alloc_rx_buffers(adapter->rx_ring,
1449 igbvf_desc_unused(adapter->rx_ring));
1450}
1451
1452/* igbvf_reset - bring the hardware into a known good state
1453 * @adapter: private board structure
1454 *
1455 * This function boots the hardware and enables some settings that
1456 * require a configuration cycle of the hardware - those cannot be
1457 * set/changed during runtime. After reset the device needs to be
1458 * properly configured for Rx, Tx etc.
1459 */
1460static void igbvf_reset(struct igbvf_adapter *adapter)
1461{
1462 struct e1000_mac_info *mac = &adapter->hw.mac;
1463 struct net_device *netdev = adapter->netdev;
1464 struct e1000_hw *hw = &adapter->hw;
1465
1466 /* Allow time for pending master requests to run */
1467 if (mac->ops.reset_hw(hw))
1468 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1469
1470 mac->ops.init_hw(hw);
1471
1472 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1473 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1474 netdev->addr_len);
1475 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1476 netdev->addr_len);
1477 }
1478
1479 adapter->last_reset = jiffies;
1480}
1481
1482int igbvf_up(struct igbvf_adapter *adapter)
1483{
1484 struct e1000_hw *hw = &adapter->hw;
1485
1486 /* hardware has been reset, we need to reload some things */
1487 igbvf_configure(adapter);
1488
1489 clear_bit(__IGBVF_DOWN, &adapter->state);
1490
1491 napi_enable(&adapter->rx_ring->napi);
1492 if (adapter->msix_entries)
1493 igbvf_configure_msix(adapter);
1494
1495 /* Clear any pending interrupts. */
1496 er32(EICR);
1497 igbvf_irq_enable(adapter);
1498
1499 /* start the watchdog */
1500 hw->mac.get_link_status = 1;
1501 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1502
1503 return 0;
1504}
1505
1506void igbvf_down(struct igbvf_adapter *adapter)
1507{
1508 struct net_device *netdev = adapter->netdev;
1509 struct e1000_hw *hw = &adapter->hw;
1510 u32 rxdctl, txdctl;
1511
1512 /* signal that we're down so the interrupt handler does not
1513 * reschedule our watchdog timer
1514 */
1515 set_bit(__IGBVF_DOWN, &adapter->state);
1516
1517 /* disable receives in the hardware */
1518 rxdctl = er32(RXDCTL(0));
1519 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1520
1521 netif_carrier_off(netdev);
1522 netif_stop_queue(netdev);
1523
1524 /* disable transmits in the hardware */
1525 txdctl = er32(TXDCTL(0));
1526 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1527
1528 /* flush both disables and wait for them to finish */
1529 e1e_flush();
1530 msleep(10);
1531
1532 napi_disable(&adapter->rx_ring->napi);
1533
1534 igbvf_irq_disable(adapter);
1535
1536 del_timer_sync(&adapter->watchdog_timer);
1537
1538 /* record the stats before reset*/
1539 igbvf_update_stats(adapter);
1540
1541 adapter->link_speed = 0;
1542 adapter->link_duplex = 0;
1543
1544 igbvf_reset(adapter);
1545 igbvf_clean_tx_ring(adapter->tx_ring);
1546 igbvf_clean_rx_ring(adapter->rx_ring);
1547}
1548
1549void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1550{
1551 might_sleep();
1552 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1553 usleep_range(1000, 2000);
1554 igbvf_down(adapter);
1555 igbvf_up(adapter);
1556 clear_bit(__IGBVF_RESETTING, &adapter->state);
1557}
1558
1559/**
1560 * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1561 * @adapter: board private structure to initialize
1562 *
1563 * igbvf_sw_init initializes the Adapter private data structure.
1564 * Fields are initialized based on PCI device information and
1565 * OS network device settings (MTU size).
1566 **/
1567static int igbvf_sw_init(struct igbvf_adapter *adapter)
1568{
1569 struct net_device *netdev = adapter->netdev;
1570 s32 rc;
1571
1572 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1573 adapter->rx_ps_hdr_size = 0;
1574 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1575 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1576
1577 adapter->tx_int_delay = 8;
1578 adapter->tx_abs_int_delay = 32;
1579 adapter->rx_int_delay = 0;
1580 adapter->rx_abs_int_delay = 8;
1581 adapter->requested_itr = 3;
1582 adapter->current_itr = IGBVF_START_ITR;
1583
1584 /* Set various function pointers */
1585 adapter->ei->init_ops(&adapter->hw);
1586
1587 rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1588 if (rc)
1589 return rc;
1590
1591 rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1592 if (rc)
1593 return rc;
1594
1595 igbvf_set_interrupt_capability(adapter);
1596
1597 if (igbvf_alloc_queues(adapter))
1598 return -ENOMEM;
1599
1600 spin_lock_init(&adapter->tx_queue_lock);
1601
1602 /* Explicitly disable IRQ since the NIC can be in any state. */
1603 igbvf_irq_disable(adapter);
1604
1605 spin_lock_init(&adapter->stats_lock);
1606
1607 set_bit(__IGBVF_DOWN, &adapter->state);
1608 return 0;
1609}
1610
1611static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1612{
1613 struct e1000_hw *hw = &adapter->hw;
1614
1615 adapter->stats.last_gprc = er32(VFGPRC);
1616 adapter->stats.last_gorc = er32(VFGORC);
1617 adapter->stats.last_gptc = er32(VFGPTC);
1618 adapter->stats.last_gotc = er32(VFGOTC);
1619 adapter->stats.last_mprc = er32(VFMPRC);
1620 adapter->stats.last_gotlbc = er32(VFGOTLBC);
1621 adapter->stats.last_gptlbc = er32(VFGPTLBC);
1622 adapter->stats.last_gorlbc = er32(VFGORLBC);
1623 adapter->stats.last_gprlbc = er32(VFGPRLBC);
1624
1625 adapter->stats.base_gprc = er32(VFGPRC);
1626 adapter->stats.base_gorc = er32(VFGORC);
1627 adapter->stats.base_gptc = er32(VFGPTC);
1628 adapter->stats.base_gotc = er32(VFGOTC);
1629 adapter->stats.base_mprc = er32(VFMPRC);
1630 adapter->stats.base_gotlbc = er32(VFGOTLBC);
1631 adapter->stats.base_gptlbc = er32(VFGPTLBC);
1632 adapter->stats.base_gorlbc = er32(VFGORLBC);
1633 adapter->stats.base_gprlbc = er32(VFGPRLBC);
1634}
1635
1636/**
1637 * igbvf_open - Called when a network interface is made active
1638 * @netdev: network interface device structure
1639 *
1640 * Returns 0 on success, negative value on failure
1641 *
1642 * The open entry point is called when a network interface is made
1643 * active by the system (IFF_UP). At this point all resources needed
1644 * for transmit and receive operations are allocated, the interrupt
1645 * handler is registered with the OS, the watchdog timer is started,
1646 * and the stack is notified that the interface is ready.
1647 **/
1648static int igbvf_open(struct net_device *netdev)
1649{
1650 struct igbvf_adapter *adapter = netdev_priv(netdev);
1651 struct e1000_hw *hw = &adapter->hw;
1652 int err;
1653
1654 /* disallow open during test */
1655 if (test_bit(__IGBVF_TESTING, &adapter->state))
1656 return -EBUSY;
1657
1658 /* allocate transmit descriptors */
1659 err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1660 if (err)
1661 goto err_setup_tx;
1662
1663 /* allocate receive descriptors */
1664 err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1665 if (err)
1666 goto err_setup_rx;
1667
1668 /* before we allocate an interrupt, we must be ready to handle it.
1669 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1670 * as soon as we call pci_request_irq, so we have to setup our
1671 * clean_rx handler before we do so.
1672 */
1673 igbvf_configure(adapter);
1674
1675 err = igbvf_request_irq(adapter);
1676 if (err)
1677 goto err_req_irq;
1678
1679 /* From here on the code is the same as igbvf_up() */
1680 clear_bit(__IGBVF_DOWN, &adapter->state);
1681
1682 napi_enable(&adapter->rx_ring->napi);
1683
1684 /* clear any pending interrupts */
1685 er32(EICR);
1686
1687 igbvf_irq_enable(adapter);
1688
1689 /* start the watchdog */
1690 hw->mac.get_link_status = 1;
1691 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1692
1693 return 0;
1694
1695err_req_irq:
1696 igbvf_free_rx_resources(adapter->rx_ring);
1697err_setup_rx:
1698 igbvf_free_tx_resources(adapter->tx_ring);
1699err_setup_tx:
1700 igbvf_reset(adapter);
1701
1702 return err;
1703}
1704
1705/**
1706 * igbvf_close - Disables a network interface
1707 * @netdev: network interface device structure
1708 *
1709 * Returns 0, this is not allowed to fail
1710 *
1711 * The close entry point is called when an interface is de-activated
1712 * by the OS. The hardware is still under the drivers control, but
1713 * needs to be disabled. A global MAC reset is issued to stop the
1714 * hardware, and all transmit and receive resources are freed.
1715 **/
1716static int igbvf_close(struct net_device *netdev)
1717{
1718 struct igbvf_adapter *adapter = netdev_priv(netdev);
1719
1720 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1721 igbvf_down(adapter);
1722
1723 igbvf_free_irq(adapter);
1724
1725 igbvf_free_tx_resources(adapter->tx_ring);
1726 igbvf_free_rx_resources(adapter->rx_ring);
1727
1728 return 0;
1729}
1730
1731/**
1732 * igbvf_set_mac - Change the Ethernet Address of the NIC
1733 * @netdev: network interface device structure
1734 * @p: pointer to an address structure
1735 *
1736 * Returns 0 on success, negative on failure
1737 **/
1738static int igbvf_set_mac(struct net_device *netdev, void *p)
1739{
1740 struct igbvf_adapter *adapter = netdev_priv(netdev);
1741 struct e1000_hw *hw = &adapter->hw;
1742 struct sockaddr *addr = p;
1743
1744 if (!is_valid_ether_addr(addr->sa_data))
1745 return -EADDRNOTAVAIL;
1746
1747 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1748
1749 hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1750
1751 if (!ether_addr_equal(addr->sa_data, hw->mac.addr))
1752 return -EADDRNOTAVAIL;
1753
1754 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1755
1756 return 0;
1757}
1758
1759#define UPDATE_VF_COUNTER(reg, name) \
1760{ \
1761 u32 current_counter = er32(reg); \
1762 if (current_counter < adapter->stats.last_##name) \
1763 adapter->stats.name += 0x100000000LL; \
1764 adapter->stats.last_##name = current_counter; \
1765 adapter->stats.name &= 0xFFFFFFFF00000000LL; \
1766 adapter->stats.name |= current_counter; \
1767}
1768
1769/**
1770 * igbvf_update_stats - Update the board statistics counters
1771 * @adapter: board private structure
1772**/
1773void igbvf_update_stats(struct igbvf_adapter *adapter)
1774{
1775 struct e1000_hw *hw = &adapter->hw;
1776 struct pci_dev *pdev = adapter->pdev;
1777
1778 /* Prevent stats update while adapter is being reset, link is down
1779 * or if the pci connection is down.
1780 */
1781 if (adapter->link_speed == 0)
1782 return;
1783
1784 if (test_bit(__IGBVF_RESETTING, &adapter->state))
1785 return;
1786
1787 if (pci_channel_offline(pdev))
1788 return;
1789
1790 UPDATE_VF_COUNTER(VFGPRC, gprc);
1791 UPDATE_VF_COUNTER(VFGORC, gorc);
1792 UPDATE_VF_COUNTER(VFGPTC, gptc);
1793 UPDATE_VF_COUNTER(VFGOTC, gotc);
1794 UPDATE_VF_COUNTER(VFMPRC, mprc);
1795 UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1796 UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1797 UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1798 UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1799
1800 /* Fill out the OS statistics structure */
1801 adapter->net_stats.multicast = adapter->stats.mprc;
1802}
1803
1804static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1805{
1806 dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s Duplex\n",
1807 adapter->link_speed,
1808 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half");
1809}
1810
1811static bool igbvf_has_link(struct igbvf_adapter *adapter)
1812{
1813 struct e1000_hw *hw = &adapter->hw;
1814 s32 ret_val = E1000_SUCCESS;
1815 bool link_active;
1816
1817 /* If interface is down, stay link down */
1818 if (test_bit(__IGBVF_DOWN, &adapter->state))
1819 return false;
1820
1821 ret_val = hw->mac.ops.check_for_link(hw);
1822 link_active = !hw->mac.get_link_status;
1823
1824 /* if check for link returns error we will need to reset */
1825 if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1826 schedule_work(&adapter->reset_task);
1827
1828 return link_active;
1829}
1830
1831/**
1832 * igbvf_watchdog - Timer Call-back
1833 * @data: pointer to adapter cast into an unsigned long
1834 **/
1835static void igbvf_watchdog(unsigned long data)
1836{
1837 struct igbvf_adapter *adapter = (struct igbvf_adapter *)data;
1838
1839 /* Do the rest outside of interrupt context */
1840 schedule_work(&adapter->watchdog_task);
1841}
1842
1843static void igbvf_watchdog_task(struct work_struct *work)
1844{
1845 struct igbvf_adapter *adapter = container_of(work,
1846 struct igbvf_adapter,
1847 watchdog_task);
1848 struct net_device *netdev = adapter->netdev;
1849 struct e1000_mac_info *mac = &adapter->hw.mac;
1850 struct igbvf_ring *tx_ring = adapter->tx_ring;
1851 struct e1000_hw *hw = &adapter->hw;
1852 u32 link;
1853 int tx_pending = 0;
1854
1855 link = igbvf_has_link(adapter);
1856
1857 if (link) {
1858 if (!netif_carrier_ok(netdev)) {
1859 mac->ops.get_link_up_info(&adapter->hw,
1860 &adapter->link_speed,
1861 &adapter->link_duplex);
1862 igbvf_print_link_info(adapter);
1863
1864 netif_carrier_on(netdev);
1865 netif_wake_queue(netdev);
1866 }
1867 } else {
1868 if (netif_carrier_ok(netdev)) {
1869 adapter->link_speed = 0;
1870 adapter->link_duplex = 0;
1871 dev_info(&adapter->pdev->dev, "Link is Down\n");
1872 netif_carrier_off(netdev);
1873 netif_stop_queue(netdev);
1874 }
1875 }
1876
1877 if (netif_carrier_ok(netdev)) {
1878 igbvf_update_stats(adapter);
1879 } else {
1880 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1881 tx_ring->count);
1882 if (tx_pending) {
1883 /* We've lost link, so the controller stops DMA,
1884 * but we've got queued Tx work that's never going
1885 * to get done, so reset controller to flush Tx.
1886 * (Do the reset outside of interrupt context).
1887 */
1888 adapter->tx_timeout_count++;
1889 schedule_work(&adapter->reset_task);
1890 }
1891 }
1892
1893 /* Cause software interrupt to ensure Rx ring is cleaned */
1894 ew32(EICS, adapter->rx_ring->eims_value);
1895
1896 /* Reset the timer */
1897 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1898 mod_timer(&adapter->watchdog_timer,
1899 round_jiffies(jiffies + (2 * HZ)));
1900}
1901
1902#define IGBVF_TX_FLAGS_CSUM 0x00000001
1903#define IGBVF_TX_FLAGS_VLAN 0x00000002
1904#define IGBVF_TX_FLAGS_TSO 0x00000004
1905#define IGBVF_TX_FLAGS_IPV4 0x00000008
1906#define IGBVF_TX_FLAGS_VLAN_MASK 0xffff0000
1907#define IGBVF_TX_FLAGS_VLAN_SHIFT 16
1908
1909static void igbvf_tx_ctxtdesc(struct igbvf_ring *tx_ring, u32 vlan_macip_lens,
1910 u32 type_tucmd, u32 mss_l4len_idx)
1911{
1912 struct e1000_adv_tx_context_desc *context_desc;
1913 struct igbvf_buffer *buffer_info;
1914 u16 i = tx_ring->next_to_use;
1915
1916 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1917 buffer_info = &tx_ring->buffer_info[i];
1918
1919 i++;
1920 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1921
1922 /* set bits to identify this as an advanced context descriptor */
1923 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
1924
1925 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
1926 context_desc->seqnum_seed = 0;
1927 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
1928 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1929
1930 buffer_info->time_stamp = jiffies;
1931 buffer_info->dma = 0;
1932}
1933
1934static int igbvf_tso(struct igbvf_ring *tx_ring,
1935 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1936{
1937 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
1938 union {
1939 struct iphdr *v4;
1940 struct ipv6hdr *v6;
1941 unsigned char *hdr;
1942 } ip;
1943 union {
1944 struct tcphdr *tcp;
1945 unsigned char *hdr;
1946 } l4;
1947 u32 paylen, l4_offset;
1948 int err;
1949
1950 if (skb->ip_summed != CHECKSUM_PARTIAL)
1951 return 0;
1952
1953 if (!skb_is_gso(skb))
1954 return 0;
1955
1956 err = skb_cow_head(skb, 0);
1957 if (err < 0)
1958 return err;
1959
1960 ip.hdr = skb_network_header(skb);
1961 l4.hdr = skb_checksum_start(skb);
1962
1963 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1964 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
1965
1966 /* initialize outer IP header fields */
1967 if (ip.v4->version == 4) {
1968 unsigned char *csum_start = skb_checksum_start(skb);
1969 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
1970
1971 /* IP header will have to cancel out any data that
1972 * is not a part of the outer IP header
1973 */
1974 ip.v4->check = csum_fold(csum_partial(trans_start,
1975 csum_start - trans_start,
1976 0));
1977 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
1978
1979 ip.v4->tot_len = 0;
1980 } else {
1981 ip.v6->payload_len = 0;
1982 }
1983
1984 /* determine offset of inner transport header */
1985 l4_offset = l4.hdr - skb->data;
1986
1987 /* compute length of segmentation header */
1988 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
1989
1990 /* remove payload length from inner checksum */
1991 paylen = skb->len - l4_offset;
1992 csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
1993
1994 /* MSS L4LEN IDX */
1995 mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT;
1996 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
1997
1998 /* VLAN MACLEN IPLEN */
1999 vlan_macip_lens = l4.hdr - ip.hdr;
2000 vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT;
2001 vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK;
2002
2003 igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
2004
2005 return 1;
2006}
2007
2008static inline bool igbvf_ipv6_csum_is_sctp(struct sk_buff *skb)
2009{
2010 unsigned int offset = 0;
2011
2012 ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
2013
2014 return offset == skb_checksum_start_offset(skb);
2015}
2016
2017static bool igbvf_tx_csum(struct igbvf_ring *tx_ring, struct sk_buff *skb,
2018 u32 tx_flags, __be16 protocol)
2019{
2020 u32 vlan_macip_lens = 0;
2021 u32 type_tucmd = 0;
2022
2023 if (skb->ip_summed != CHECKSUM_PARTIAL) {
2024csum_failed:
2025 if (!(tx_flags & IGBVF_TX_FLAGS_VLAN))
2026 return false;
2027 goto no_csum;
2028 }
2029
2030 switch (skb->csum_offset) {
2031 case offsetof(struct tcphdr, check):
2032 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
2033 /* fall through */
2034 case offsetof(struct udphdr, check):
2035 break;
2036 case offsetof(struct sctphdr, checksum):
2037 /* validate that this is actually an SCTP request */
2038 if (((protocol == htons(ETH_P_IP)) &&
2039 (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
2040 ((protocol == htons(ETH_P_IPV6)) &&
2041 igbvf_ipv6_csum_is_sctp(skb))) {
2042 type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP;
2043 break;
2044 }
2045 default:
2046 skb_checksum_help(skb);
2047 goto csum_failed;
2048 }
2049
2050 vlan_macip_lens = skb_checksum_start_offset(skb) -
2051 skb_network_offset(skb);
2052no_csum:
2053 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
2054 vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK;
2055
2056 igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0);
2057 return true;
2058}
2059
2060static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2061{
2062 struct igbvf_adapter *adapter = netdev_priv(netdev);
2063
2064 /* there is enough descriptors then we don't need to worry */
2065 if (igbvf_desc_unused(adapter->tx_ring) >= size)
2066 return 0;
2067
2068 netif_stop_queue(netdev);
2069
2070 /* Herbert's original patch had:
2071 * smp_mb__after_netif_stop_queue();
2072 * but since that doesn't exist yet, just open code it.
2073 */
2074 smp_mb();
2075
2076 /* We need to check again just in case room has been made available */
2077 if (igbvf_desc_unused(adapter->tx_ring) < size)
2078 return -EBUSY;
2079
2080 netif_wake_queue(netdev);
2081
2082 ++adapter->restart_queue;
2083 return 0;
2084}
2085
2086#define IGBVF_MAX_TXD_PWR 16
2087#define IGBVF_MAX_DATA_PER_TXD (1u << IGBVF_MAX_TXD_PWR)
2088
2089static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2090 struct igbvf_ring *tx_ring,
2091 struct sk_buff *skb)
2092{
2093 struct igbvf_buffer *buffer_info;
2094 struct pci_dev *pdev = adapter->pdev;
2095 unsigned int len = skb_headlen(skb);
2096 unsigned int count = 0, i;
2097 unsigned int f;
2098
2099 i = tx_ring->next_to_use;
2100
2101 buffer_info = &tx_ring->buffer_info[i];
2102 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2103 buffer_info->length = len;
2104 /* set time_stamp *before* dma to help avoid a possible race */
2105 buffer_info->time_stamp = jiffies;
2106 buffer_info->mapped_as_page = false;
2107 buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2108 DMA_TO_DEVICE);
2109 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2110 goto dma_error;
2111
2112 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2113 const struct skb_frag_struct *frag;
2114
2115 count++;
2116 i++;
2117 if (i == tx_ring->count)
2118 i = 0;
2119
2120 frag = &skb_shinfo(skb)->frags[f];
2121 len = skb_frag_size(frag);
2122
2123 buffer_info = &tx_ring->buffer_info[i];
2124 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2125 buffer_info->length = len;
2126 buffer_info->time_stamp = jiffies;
2127 buffer_info->mapped_as_page = true;
2128 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag, 0, len,
2129 DMA_TO_DEVICE);
2130 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2131 goto dma_error;
2132 }
2133
2134 tx_ring->buffer_info[i].skb = skb;
2135
2136 return ++count;
2137
2138dma_error:
2139 dev_err(&pdev->dev, "TX DMA map failed\n");
2140
2141 /* clear timestamp and dma mappings for failed buffer_info mapping */
2142 buffer_info->dma = 0;
2143 buffer_info->time_stamp = 0;
2144 buffer_info->length = 0;
2145 buffer_info->mapped_as_page = false;
2146 if (count)
2147 count--;
2148
2149 /* clear timestamp and dma mappings for remaining portion of packet */
2150 while (count--) {
2151 if (i == 0)
2152 i += tx_ring->count;
2153 i--;
2154 buffer_info = &tx_ring->buffer_info[i];
2155 igbvf_put_txbuf(adapter, buffer_info);
2156 }
2157
2158 return 0;
2159}
2160
2161static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2162 struct igbvf_ring *tx_ring,
2163 int tx_flags, int count,
2164 unsigned int first, u32 paylen,
2165 u8 hdr_len)
2166{
2167 union e1000_adv_tx_desc *tx_desc = NULL;
2168 struct igbvf_buffer *buffer_info;
2169 u32 olinfo_status = 0, cmd_type_len;
2170 unsigned int i;
2171
2172 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2173 E1000_ADVTXD_DCMD_DEXT);
2174
2175 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2176 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2177
2178 if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2179 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2180
2181 /* insert tcp checksum */
2182 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2183
2184 /* insert ip checksum */
2185 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2186 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2187
2188 } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2189 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2190 }
2191
2192 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2193
2194 i = tx_ring->next_to_use;
2195 while (count--) {
2196 buffer_info = &tx_ring->buffer_info[i];
2197 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2198 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2199 tx_desc->read.cmd_type_len =
2200 cpu_to_le32(cmd_type_len | buffer_info->length);
2201 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2202 i++;
2203 if (i == tx_ring->count)
2204 i = 0;
2205 }
2206
2207 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2208 /* Force memory writes to complete before letting h/w
2209 * know there are new descriptors to fetch. (Only
2210 * applicable for weak-ordered memory model archs,
2211 * such as IA-64).
2212 */
2213 wmb();
2214
2215 tx_ring->buffer_info[first].next_to_watch = tx_desc;
2216 tx_ring->next_to_use = i;
2217 writel(i, adapter->hw.hw_addr + tx_ring->tail);
2218 /* we need this if more than one processor can write to our tail
2219 * at a time, it synchronizes IO on IA64/Altix systems
2220 */
2221 mmiowb();
2222}
2223
2224static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2225 struct net_device *netdev,
2226 struct igbvf_ring *tx_ring)
2227{
2228 struct igbvf_adapter *adapter = netdev_priv(netdev);
2229 unsigned int first, tx_flags = 0;
2230 u8 hdr_len = 0;
2231 int count = 0;
2232 int tso = 0;
2233 __be16 protocol = vlan_get_protocol(skb);
2234
2235 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2236 dev_kfree_skb_any(skb);
2237 return NETDEV_TX_OK;
2238 }
2239
2240 if (skb->len <= 0) {
2241 dev_kfree_skb_any(skb);
2242 return NETDEV_TX_OK;
2243 }
2244
2245 /* need: count + 4 desc gap to keep tail from touching
2246 * + 2 desc gap to keep tail from touching head,
2247 * + 1 desc for skb->data,
2248 * + 1 desc for context descriptor,
2249 * head, otherwise try next time
2250 */
2251 if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2252 /* this is a hard error */
2253 return NETDEV_TX_BUSY;
2254 }
2255
2256 if (skb_vlan_tag_present(skb)) {
2257 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2258 tx_flags |= (skb_vlan_tag_get(skb) <<
2259 IGBVF_TX_FLAGS_VLAN_SHIFT);
2260 }
2261
2262 if (protocol == htons(ETH_P_IP))
2263 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2264
2265 first = tx_ring->next_to_use;
2266
2267 tso = igbvf_tso(tx_ring, skb, tx_flags, &hdr_len);
2268 if (unlikely(tso < 0)) {
2269 dev_kfree_skb_any(skb);
2270 return NETDEV_TX_OK;
2271 }
2272
2273 if (tso)
2274 tx_flags |= IGBVF_TX_FLAGS_TSO;
2275 else if (igbvf_tx_csum(tx_ring, skb, tx_flags, protocol) &&
2276 (skb->ip_summed == CHECKSUM_PARTIAL))
2277 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2278
2279 /* count reflects descriptors mapped, if 0 then mapping error
2280 * has occurred and we need to rewind the descriptor queue
2281 */
2282 count = igbvf_tx_map_adv(adapter, tx_ring, skb);
2283
2284 if (count) {
2285 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2286 first, skb->len, hdr_len);
2287 /* Make sure there is space in the ring for the next send. */
2288 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2289 } else {
2290 dev_kfree_skb_any(skb);
2291 tx_ring->buffer_info[first].time_stamp = 0;
2292 tx_ring->next_to_use = first;
2293 }
2294
2295 return NETDEV_TX_OK;
2296}
2297
2298static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2299 struct net_device *netdev)
2300{
2301 struct igbvf_adapter *adapter = netdev_priv(netdev);
2302 struct igbvf_ring *tx_ring;
2303
2304 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2305 dev_kfree_skb_any(skb);
2306 return NETDEV_TX_OK;
2307 }
2308
2309 tx_ring = &adapter->tx_ring[0];
2310
2311 return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2312}
2313
2314/**
2315 * igbvf_tx_timeout - Respond to a Tx Hang
2316 * @netdev: network interface device structure
2317 **/
2318static void igbvf_tx_timeout(struct net_device *netdev)
2319{
2320 struct igbvf_adapter *adapter = netdev_priv(netdev);
2321
2322 /* Do the reset outside of interrupt context */
2323 adapter->tx_timeout_count++;
2324 schedule_work(&adapter->reset_task);
2325}
2326
2327static void igbvf_reset_task(struct work_struct *work)
2328{
2329 struct igbvf_adapter *adapter;
2330
2331 adapter = container_of(work, struct igbvf_adapter, reset_task);
2332
2333 igbvf_reinit_locked(adapter);
2334}
2335
2336/**
2337 * igbvf_get_stats - Get System Network Statistics
2338 * @netdev: network interface device structure
2339 *
2340 * Returns the address of the device statistics structure.
2341 * The statistics are actually updated from the timer callback.
2342 **/
2343static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2344{
2345 struct igbvf_adapter *adapter = netdev_priv(netdev);
2346
2347 /* only return the current stats */
2348 return &adapter->net_stats;
2349}
2350
2351/**
2352 * igbvf_change_mtu - Change the Maximum Transfer Unit
2353 * @netdev: network interface device structure
2354 * @new_mtu: new value for maximum frame size
2355 *
2356 * Returns 0 on success, negative on failure
2357 **/
2358static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2359{
2360 struct igbvf_adapter *adapter = netdev_priv(netdev);
2361 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2362
2363 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2364 usleep_range(1000, 2000);
2365 /* igbvf_down has a dependency on max_frame_size */
2366 adapter->max_frame_size = max_frame;
2367 if (netif_running(netdev))
2368 igbvf_down(adapter);
2369
2370 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2371 * means we reserve 2 more, this pushes us to allocate from the next
2372 * larger slab size.
2373 * i.e. RXBUFFER_2048 --> size-4096 slab
2374 * However with the new *_jumbo_rx* routines, jumbo receives will use
2375 * fragmented skbs
2376 */
2377
2378 if (max_frame <= 1024)
2379 adapter->rx_buffer_len = 1024;
2380 else if (max_frame <= 2048)
2381 adapter->rx_buffer_len = 2048;
2382 else
2383#if (PAGE_SIZE / 2) > 16384
2384 adapter->rx_buffer_len = 16384;
2385#else
2386 adapter->rx_buffer_len = PAGE_SIZE / 2;
2387#endif
2388
2389 /* adjust allocation if LPE protects us, and we aren't using SBP */
2390 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2391 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2392 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2393 ETH_FCS_LEN;
2394
2395 dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2396 netdev->mtu, new_mtu);
2397 netdev->mtu = new_mtu;
2398
2399 if (netif_running(netdev))
2400 igbvf_up(adapter);
2401 else
2402 igbvf_reset(adapter);
2403
2404 clear_bit(__IGBVF_RESETTING, &adapter->state);
2405
2406 return 0;
2407}
2408
2409static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2410{
2411 switch (cmd) {
2412 default:
2413 return -EOPNOTSUPP;
2414 }
2415}
2416
2417static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2418{
2419 struct net_device *netdev = pci_get_drvdata(pdev);
2420 struct igbvf_adapter *adapter = netdev_priv(netdev);
2421#ifdef CONFIG_PM
2422 int retval = 0;
2423#endif
2424
2425 netif_device_detach(netdev);
2426
2427 if (netif_running(netdev)) {
2428 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2429 igbvf_down(adapter);
2430 igbvf_free_irq(adapter);
2431 }
2432
2433#ifdef CONFIG_PM
2434 retval = pci_save_state(pdev);
2435 if (retval)
2436 return retval;
2437#endif
2438
2439 pci_disable_device(pdev);
2440
2441 return 0;
2442}
2443
2444#ifdef CONFIG_PM
2445static int igbvf_resume(struct pci_dev *pdev)
2446{
2447 struct net_device *netdev = pci_get_drvdata(pdev);
2448 struct igbvf_adapter *adapter = netdev_priv(netdev);
2449 u32 err;
2450
2451 pci_restore_state(pdev);
2452 err = pci_enable_device_mem(pdev);
2453 if (err) {
2454 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2455 return err;
2456 }
2457
2458 pci_set_master(pdev);
2459
2460 if (netif_running(netdev)) {
2461 err = igbvf_request_irq(adapter);
2462 if (err)
2463 return err;
2464 }
2465
2466 igbvf_reset(adapter);
2467
2468 if (netif_running(netdev))
2469 igbvf_up(adapter);
2470
2471 netif_device_attach(netdev);
2472
2473 return 0;
2474}
2475#endif
2476
2477static void igbvf_shutdown(struct pci_dev *pdev)
2478{
2479 igbvf_suspend(pdev, PMSG_SUSPEND);
2480}
2481
2482#ifdef CONFIG_NET_POLL_CONTROLLER
2483/* Polling 'interrupt' - used by things like netconsole to send skbs
2484 * without having to re-enable interrupts. It's not called while
2485 * the interrupt routine is executing.
2486 */
2487static void igbvf_netpoll(struct net_device *netdev)
2488{
2489 struct igbvf_adapter *adapter = netdev_priv(netdev);
2490
2491 disable_irq(adapter->pdev->irq);
2492
2493 igbvf_clean_tx_irq(adapter->tx_ring);
2494
2495 enable_irq(adapter->pdev->irq);
2496}
2497#endif
2498
2499/**
2500 * igbvf_io_error_detected - called when PCI error is detected
2501 * @pdev: Pointer to PCI device
2502 * @state: The current pci connection state
2503 *
2504 * This function is called after a PCI bus error affecting
2505 * this device has been detected.
2506 */
2507static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2508 pci_channel_state_t state)
2509{
2510 struct net_device *netdev = pci_get_drvdata(pdev);
2511 struct igbvf_adapter *adapter = netdev_priv(netdev);
2512
2513 netif_device_detach(netdev);
2514
2515 if (state == pci_channel_io_perm_failure)
2516 return PCI_ERS_RESULT_DISCONNECT;
2517
2518 if (netif_running(netdev))
2519 igbvf_down(adapter);
2520 pci_disable_device(pdev);
2521
2522 /* Request a slot slot reset. */
2523 return PCI_ERS_RESULT_NEED_RESET;
2524}
2525
2526/**
2527 * igbvf_io_slot_reset - called after the pci bus has been reset.
2528 * @pdev: Pointer to PCI device
2529 *
2530 * Restart the card from scratch, as if from a cold-boot. Implementation
2531 * resembles the first-half of the igbvf_resume routine.
2532 */
2533static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2534{
2535 struct net_device *netdev = pci_get_drvdata(pdev);
2536 struct igbvf_adapter *adapter = netdev_priv(netdev);
2537
2538 if (pci_enable_device_mem(pdev)) {
2539 dev_err(&pdev->dev,
2540 "Cannot re-enable PCI device after reset.\n");
2541 return PCI_ERS_RESULT_DISCONNECT;
2542 }
2543 pci_set_master(pdev);
2544
2545 igbvf_reset(adapter);
2546
2547 return PCI_ERS_RESULT_RECOVERED;
2548}
2549
2550/**
2551 * igbvf_io_resume - called when traffic can start flowing again.
2552 * @pdev: Pointer to PCI device
2553 *
2554 * This callback is called when the error recovery driver tells us that
2555 * its OK to resume normal operation. Implementation resembles the
2556 * second-half of the igbvf_resume routine.
2557 */
2558static void igbvf_io_resume(struct pci_dev *pdev)
2559{
2560 struct net_device *netdev = pci_get_drvdata(pdev);
2561 struct igbvf_adapter *adapter = netdev_priv(netdev);
2562
2563 if (netif_running(netdev)) {
2564 if (igbvf_up(adapter)) {
2565 dev_err(&pdev->dev,
2566 "can't bring device back up after reset\n");
2567 return;
2568 }
2569 }
2570
2571 netif_device_attach(netdev);
2572}
2573
2574static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2575{
2576 struct e1000_hw *hw = &adapter->hw;
2577 struct net_device *netdev = adapter->netdev;
2578 struct pci_dev *pdev = adapter->pdev;
2579
2580 if (hw->mac.type == e1000_vfadapt_i350)
2581 dev_info(&pdev->dev, "Intel(R) I350 Virtual Function\n");
2582 else
2583 dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2584 dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2585}
2586
2587static int igbvf_set_features(struct net_device *netdev,
2588 netdev_features_t features)
2589{
2590 struct igbvf_adapter *adapter = netdev_priv(netdev);
2591
2592 if (features & NETIF_F_RXCSUM)
2593 adapter->flags &= ~IGBVF_FLAG_RX_CSUM_DISABLED;
2594 else
2595 adapter->flags |= IGBVF_FLAG_RX_CSUM_DISABLED;
2596
2597 return 0;
2598}
2599
2600#define IGBVF_MAX_MAC_HDR_LEN 127
2601#define IGBVF_MAX_NETWORK_HDR_LEN 511
2602
2603static netdev_features_t
2604igbvf_features_check(struct sk_buff *skb, struct net_device *dev,
2605 netdev_features_t features)
2606{
2607 unsigned int network_hdr_len, mac_hdr_len;
2608
2609 /* Make certain the headers can be described by a context descriptor */
2610 mac_hdr_len = skb_network_header(skb) - skb->data;
2611 if (unlikely(mac_hdr_len > IGBVF_MAX_MAC_HDR_LEN))
2612 return features & ~(NETIF_F_HW_CSUM |
2613 NETIF_F_SCTP_CRC |
2614 NETIF_F_HW_VLAN_CTAG_TX |
2615 NETIF_F_TSO |
2616 NETIF_F_TSO6);
2617
2618 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
2619 if (unlikely(network_hdr_len > IGBVF_MAX_NETWORK_HDR_LEN))
2620 return features & ~(NETIF_F_HW_CSUM |
2621 NETIF_F_SCTP_CRC |
2622 NETIF_F_TSO |
2623 NETIF_F_TSO6);
2624
2625 /* We can only support IPV4 TSO in tunnels if we can mangle the
2626 * inner IP ID field, so strip TSO if MANGLEID is not supported.
2627 */
2628 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
2629 features &= ~NETIF_F_TSO;
2630
2631 return features;
2632}
2633
2634static const struct net_device_ops igbvf_netdev_ops = {
2635 .ndo_open = igbvf_open,
2636 .ndo_stop = igbvf_close,
2637 .ndo_start_xmit = igbvf_xmit_frame,
2638 .ndo_get_stats = igbvf_get_stats,
2639 .ndo_set_rx_mode = igbvf_set_multi,
2640 .ndo_set_mac_address = igbvf_set_mac,
2641 .ndo_change_mtu = igbvf_change_mtu,
2642 .ndo_do_ioctl = igbvf_ioctl,
2643 .ndo_tx_timeout = igbvf_tx_timeout,
2644 .ndo_vlan_rx_add_vid = igbvf_vlan_rx_add_vid,
2645 .ndo_vlan_rx_kill_vid = igbvf_vlan_rx_kill_vid,
2646#ifdef CONFIG_NET_POLL_CONTROLLER
2647 .ndo_poll_controller = igbvf_netpoll,
2648#endif
2649 .ndo_set_features = igbvf_set_features,
2650 .ndo_features_check = igbvf_features_check,
2651};
2652
2653/**
2654 * igbvf_probe - Device Initialization Routine
2655 * @pdev: PCI device information struct
2656 * @ent: entry in igbvf_pci_tbl
2657 *
2658 * Returns 0 on success, negative on failure
2659 *
2660 * igbvf_probe initializes an adapter identified by a pci_dev structure.
2661 * The OS initialization, configuring of the adapter private structure,
2662 * and a hardware reset occur.
2663 **/
2664static int igbvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2665{
2666 struct net_device *netdev;
2667 struct igbvf_adapter *adapter;
2668 struct e1000_hw *hw;
2669 const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2670
2671 static int cards_found;
2672 int err, pci_using_dac;
2673
2674 err = pci_enable_device_mem(pdev);
2675 if (err)
2676 return err;
2677
2678 pci_using_dac = 0;
2679 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
2680 if (!err) {
2681 pci_using_dac = 1;
2682 } else {
2683 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
2684 if (err) {
2685 dev_err(&pdev->dev,
2686 "No usable DMA configuration, aborting\n");
2687 goto err_dma;
2688 }
2689 }
2690
2691 err = pci_request_regions(pdev, igbvf_driver_name);
2692 if (err)
2693 goto err_pci_reg;
2694
2695 pci_set_master(pdev);
2696
2697 err = -ENOMEM;
2698 netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2699 if (!netdev)
2700 goto err_alloc_etherdev;
2701
2702 SET_NETDEV_DEV(netdev, &pdev->dev);
2703
2704 pci_set_drvdata(pdev, netdev);
2705 adapter = netdev_priv(netdev);
2706 hw = &adapter->hw;
2707 adapter->netdev = netdev;
2708 adapter->pdev = pdev;
2709 adapter->ei = ei;
2710 adapter->pba = ei->pba;
2711 adapter->flags = ei->flags;
2712 adapter->hw.back = adapter;
2713 adapter->hw.mac.type = ei->mac;
2714 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2715
2716 /* PCI config space info */
2717
2718 hw->vendor_id = pdev->vendor;
2719 hw->device_id = pdev->device;
2720 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2721 hw->subsystem_device_id = pdev->subsystem_device;
2722 hw->revision_id = pdev->revision;
2723
2724 err = -EIO;
2725 adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2726 pci_resource_len(pdev, 0));
2727
2728 if (!adapter->hw.hw_addr)
2729 goto err_ioremap;
2730
2731 if (ei->get_variants) {
2732 err = ei->get_variants(adapter);
2733 if (err)
2734 goto err_get_variants;
2735 }
2736
2737 /* setup adapter struct */
2738 err = igbvf_sw_init(adapter);
2739 if (err)
2740 goto err_sw_init;
2741
2742 /* construct the net_device struct */
2743 netdev->netdev_ops = &igbvf_netdev_ops;
2744
2745 igbvf_set_ethtool_ops(netdev);
2746 netdev->watchdog_timeo = 5 * HZ;
2747 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2748
2749 adapter->bd_number = cards_found++;
2750
2751 netdev->hw_features = NETIF_F_SG |
2752 NETIF_F_TSO |
2753 NETIF_F_TSO6 |
2754 NETIF_F_RXCSUM |
2755 NETIF_F_HW_CSUM |
2756 NETIF_F_SCTP_CRC;
2757
2758#define IGBVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
2759 NETIF_F_GSO_GRE_CSUM | \
2760 NETIF_F_GSO_IPXIP4 | \
2761 NETIF_F_GSO_IPXIP6 | \
2762 NETIF_F_GSO_UDP_TUNNEL | \
2763 NETIF_F_GSO_UDP_TUNNEL_CSUM)
2764
2765 netdev->gso_partial_features = IGBVF_GSO_PARTIAL_FEATURES;
2766 netdev->hw_features |= NETIF_F_GSO_PARTIAL |
2767 IGBVF_GSO_PARTIAL_FEATURES;
2768
2769 netdev->features = netdev->hw_features;
2770
2771 if (pci_using_dac)
2772 netdev->features |= NETIF_F_HIGHDMA;
2773
2774 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
2775 netdev->mpls_features |= NETIF_F_HW_CSUM;
2776 netdev->hw_enc_features |= netdev->vlan_features;
2777
2778 /* set this bit last since it cannot be part of vlan_features */
2779 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
2780 NETIF_F_HW_VLAN_CTAG_RX |
2781 NETIF_F_HW_VLAN_CTAG_TX;
2782
2783 /* MTU range: 68 - 9216 */
2784 netdev->min_mtu = ETH_MIN_MTU;
2785 netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
2786
2787 /*reset the controller to put the device in a known good state */
2788 err = hw->mac.ops.reset_hw(hw);
2789 if (err) {
2790 dev_info(&pdev->dev,
2791 "PF still in reset state. Is the PF interface up?\n");
2792 } else {
2793 err = hw->mac.ops.read_mac_addr(hw);
2794 if (err)
2795 dev_info(&pdev->dev, "Error reading MAC address.\n");
2796 else if (is_zero_ether_addr(adapter->hw.mac.addr))
2797 dev_info(&pdev->dev,
2798 "MAC address not assigned by administrator.\n");
2799 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
2800 netdev->addr_len);
2801 }
2802
2803 if (!is_valid_ether_addr(netdev->dev_addr)) {
2804 dev_info(&pdev->dev, "Assigning random MAC address.\n");
2805 eth_hw_addr_random(netdev);
2806 memcpy(adapter->hw.mac.addr, netdev->dev_addr,
2807 netdev->addr_len);
2808 }
2809
2810 setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2811 (unsigned long)adapter);
2812
2813 INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2814 INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2815
2816 /* ring size defaults */
2817 adapter->rx_ring->count = 1024;
2818 adapter->tx_ring->count = 1024;
2819
2820 /* reset the hardware with the new settings */
2821 igbvf_reset(adapter);
2822
2823 /* set hardware-specific flags */
2824 if (adapter->hw.mac.type == e1000_vfadapt_i350)
2825 adapter->flags |= IGBVF_FLAG_RX_LB_VLAN_BSWAP;
2826
2827 strcpy(netdev->name, "eth%d");
2828 err = register_netdev(netdev);
2829 if (err)
2830 goto err_hw_init;
2831
2832 /* tell the stack to leave us alone until igbvf_open() is called */
2833 netif_carrier_off(netdev);
2834 netif_stop_queue(netdev);
2835
2836 igbvf_print_device_info(adapter);
2837
2838 igbvf_initialize_last_counter_stats(adapter);
2839
2840 return 0;
2841
2842err_hw_init:
2843 kfree(adapter->tx_ring);
2844 kfree(adapter->rx_ring);
2845err_sw_init:
2846 igbvf_reset_interrupt_capability(adapter);
2847err_get_variants:
2848 iounmap(adapter->hw.hw_addr);
2849err_ioremap:
2850 free_netdev(netdev);
2851err_alloc_etherdev:
2852 pci_release_regions(pdev);
2853err_pci_reg:
2854err_dma:
2855 pci_disable_device(pdev);
2856 return err;
2857}
2858
2859/**
2860 * igbvf_remove - Device Removal Routine
2861 * @pdev: PCI device information struct
2862 *
2863 * igbvf_remove is called by the PCI subsystem to alert the driver
2864 * that it should release a PCI device. The could be caused by a
2865 * Hot-Plug event, or because the driver is going to be removed from
2866 * memory.
2867 **/
2868static void igbvf_remove(struct pci_dev *pdev)
2869{
2870 struct net_device *netdev = pci_get_drvdata(pdev);
2871 struct igbvf_adapter *adapter = netdev_priv(netdev);
2872 struct e1000_hw *hw = &adapter->hw;
2873
2874 /* The watchdog timer may be rescheduled, so explicitly
2875 * disable it from being rescheduled.
2876 */
2877 set_bit(__IGBVF_DOWN, &adapter->state);
2878 del_timer_sync(&adapter->watchdog_timer);
2879
2880 cancel_work_sync(&adapter->reset_task);
2881 cancel_work_sync(&adapter->watchdog_task);
2882
2883 unregister_netdev(netdev);
2884
2885 igbvf_reset_interrupt_capability(adapter);
2886
2887 /* it is important to delete the NAPI struct prior to freeing the
2888 * Rx ring so that you do not end up with null pointer refs
2889 */
2890 netif_napi_del(&adapter->rx_ring->napi);
2891 kfree(adapter->tx_ring);
2892 kfree(adapter->rx_ring);
2893
2894 iounmap(hw->hw_addr);
2895 if (hw->flash_address)
2896 iounmap(hw->flash_address);
2897 pci_release_regions(pdev);
2898
2899 free_netdev(netdev);
2900
2901 pci_disable_device(pdev);
2902}
2903
2904/* PCI Error Recovery (ERS) */
2905static const struct pci_error_handlers igbvf_err_handler = {
2906 .error_detected = igbvf_io_error_detected,
2907 .slot_reset = igbvf_io_slot_reset,
2908 .resume = igbvf_io_resume,
2909};
2910
2911static const struct pci_device_id igbvf_pci_tbl[] = {
2912 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2913 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf },
2914 { } /* terminate list */
2915};
2916MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2917
2918/* PCI Device API Driver */
2919static struct pci_driver igbvf_driver = {
2920 .name = igbvf_driver_name,
2921 .id_table = igbvf_pci_tbl,
2922 .probe = igbvf_probe,
2923 .remove = igbvf_remove,
2924#ifdef CONFIG_PM
2925 /* Power Management Hooks */
2926 .suspend = igbvf_suspend,
2927 .resume = igbvf_resume,
2928#endif
2929 .shutdown = igbvf_shutdown,
2930 .err_handler = &igbvf_err_handler
2931};
2932
2933/**
2934 * igbvf_init_module - Driver Registration Routine
2935 *
2936 * igbvf_init_module is the first routine called when the driver is
2937 * loaded. All it does is register with the PCI subsystem.
2938 **/
2939static int __init igbvf_init_module(void)
2940{
2941 int ret;
2942
2943 pr_info("%s - version %s\n", igbvf_driver_string, igbvf_driver_version);
2944 pr_info("%s\n", igbvf_copyright);
2945
2946 ret = pci_register_driver(&igbvf_driver);
2947
2948 return ret;
2949}
2950module_init(igbvf_init_module);
2951
2952/**
2953 * igbvf_exit_module - Driver Exit Cleanup Routine
2954 *
2955 * igbvf_exit_module is called just before the driver is removed
2956 * from memory.
2957 **/
2958static void __exit igbvf_exit_module(void)
2959{
2960 pci_unregister_driver(&igbvf_driver);
2961}
2962module_exit(igbvf_exit_module);
2963
2964MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2965MODULE_DESCRIPTION("Intel(R) Gigabit Virtual Function Network Driver");
2966MODULE_LICENSE("GPL");
2967MODULE_VERSION(DRV_VERSION);
2968
2969/* netdev.c */