projects/linux·blob master · refs

2991 lines · 77.5 KiB

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 hw->mac.get_link_status = 1;
859 if (!test_bit(__IGBVF_DOWN, &adapter->state))
860 mod_timer(&adapter->watchdog_timer, jiffies + 1);
861
862 ew32(EIMS, adapter->eims_other);
863
864 return IRQ_HANDLED;
865}
866
867static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
868{
869 struct net_device *netdev = data;
870 struct igbvf_adapter *adapter = netdev_priv(netdev);
871 struct e1000_hw *hw = &adapter->hw;
872 struct igbvf_ring *tx_ring = adapter->tx_ring;
873
874 if (tx_ring->set_itr) {
875 writel(tx_ring->itr_val,
876 adapter->hw.hw_addr + tx_ring->itr_register);
877 adapter->tx_ring->set_itr = 0;
878 }
879
880 adapter->total_tx_bytes = 0;
881 adapter->total_tx_packets = 0;
882
883 /* auto mask will automatically re-enable the interrupt when we write
884 * EICS
885 */
886 if (!igbvf_clean_tx_irq(tx_ring))
887 /* Ring was not completely cleaned, so fire another interrupt */
888 ew32(EICS, tx_ring->eims_value);
889 else
890 ew32(EIMS, tx_ring->eims_value);
891
892 return IRQ_HANDLED;
893}
894
895static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
896{
897 struct net_device *netdev = data;
898 struct igbvf_adapter *adapter = netdev_priv(netdev);
899
900 /* Write the ITR value calculated at the end of the
901 * previous interrupt.
902 */
903 if (adapter->rx_ring->set_itr) {
904 writel(adapter->rx_ring->itr_val,
905 adapter->hw.hw_addr + adapter->rx_ring->itr_register);
906 adapter->rx_ring->set_itr = 0;
907 }
908
909 if (napi_schedule_prep(&adapter->rx_ring->napi)) {
910 adapter->total_rx_bytes = 0;
911 adapter->total_rx_packets = 0;
912 __napi_schedule(&adapter->rx_ring->napi);
913 }
914
915 return IRQ_HANDLED;
916}
917
918#define IGBVF_NO_QUEUE -1
919
920static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
921 int tx_queue, int msix_vector)
922{
923 struct e1000_hw *hw = &adapter->hw;
924 u32 ivar, index;
925
926 /* 82576 uses a table-based method for assigning vectors.
927 * Each queue has a single entry in the table to which we write
928 * a vector number along with a "valid" bit. Sadly, the layout
929 * of the table is somewhat counterintuitive.
930 */
931 if (rx_queue > IGBVF_NO_QUEUE) {
932 index = (rx_queue >> 1);
933 ivar = array_er32(IVAR0, index);
934 if (rx_queue & 0x1) {
935 /* vector goes into third byte of register */
936 ivar = ivar & 0xFF00FFFF;
937 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
938 } else {
939 /* vector goes into low byte of register */
940 ivar = ivar & 0xFFFFFF00;
941 ivar |= msix_vector | E1000_IVAR_VALID;
942 }
943 adapter->rx_ring[rx_queue].eims_value = BIT(msix_vector);
944 array_ew32(IVAR0, index, ivar);
945 }
946 if (tx_queue > IGBVF_NO_QUEUE) {
947 index = (tx_queue >> 1);
948 ivar = array_er32(IVAR0, index);
949 if (tx_queue & 0x1) {
950 /* vector goes into high byte of register */
951 ivar = ivar & 0x00FFFFFF;
952 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
953 } else {
954 /* vector goes into second byte of register */
955 ivar = ivar & 0xFFFF00FF;
956 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
957 }
958 adapter->tx_ring[tx_queue].eims_value = BIT(msix_vector);
959 array_ew32(IVAR0, index, ivar);
960 }
961}
962
963/**
964 * igbvf_configure_msix - Configure MSI-X hardware
965 * @adapter: board private structure
966 *
967 * igbvf_configure_msix sets up the hardware to properly
968 * generate MSI-X interrupts.
969 **/
970static void igbvf_configure_msix(struct igbvf_adapter *adapter)
971{
972 u32 tmp;
973 struct e1000_hw *hw = &adapter->hw;
974 struct igbvf_ring *tx_ring = adapter->tx_ring;
975 struct igbvf_ring *rx_ring = adapter->rx_ring;
976 int vector = 0;
977
978 adapter->eims_enable_mask = 0;
979
980 igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
981 adapter->eims_enable_mask |= tx_ring->eims_value;
982 writel(tx_ring->itr_val, hw->hw_addr + tx_ring->itr_register);
983 igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
984 adapter->eims_enable_mask |= rx_ring->eims_value;
985 writel(rx_ring->itr_val, hw->hw_addr + rx_ring->itr_register);
986
987 /* set vector for other causes, i.e. link changes */
988
989 tmp = (vector++ | E1000_IVAR_VALID);
990
991 ew32(IVAR_MISC, tmp);
992
993 adapter->eims_enable_mask = GENMASK(vector - 1, 0);
994 adapter->eims_other = BIT(vector - 1);
995 e1e_flush();
996}
997
998static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
999{
1000 if (adapter->msix_entries) {
1001 pci_disable_msix(adapter->pdev);
1002 kfree(adapter->msix_entries);
1003 adapter->msix_entries = NULL;
1004 }
1005}
1006
1007/**
1008 * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1009 * @adapter: board private structure
1010 *
1011 * Attempt to configure interrupts using the best available
1012 * capabilities of the hardware and kernel.
1013 **/
1014static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1015{
1016 int err = -ENOMEM;
1017 int i;
1018
1019 /* we allocate 3 vectors, 1 for Tx, 1 for Rx, one for PF messages */
1020 adapter->msix_entries = kzalloc_objs(struct msix_entry, 3);
1021 if (adapter->msix_entries) {
1022 for (i = 0; i < 3; i++)
1023 adapter->msix_entries[i].entry = i;
1024
1025 err = pci_enable_msix_range(adapter->pdev,
1026 adapter->msix_entries, 3, 3);
1027 }
1028
1029 if (err < 0) {
1030 /* MSI-X failed */
1031 dev_err(&adapter->pdev->dev,
1032 "Failed to initialize MSI-X interrupts.\n");
1033 igbvf_reset_interrupt_capability(adapter);
1034 }
1035}
1036
1037/**
1038 * igbvf_request_msix - Initialize MSI-X interrupts
1039 * @adapter: board private structure
1040 *
1041 * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1042 * kernel.
1043 **/
1044static int igbvf_request_msix(struct igbvf_adapter *adapter)
1045{
1046 struct net_device *netdev = adapter->netdev;
1047 int err = 0, vector = 0;
1048
1049 if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1050 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1051 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1052 } else {
1053 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1054 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1055 }
1056
1057 err = request_irq(adapter->msix_entries[vector].vector,
1058 igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1059 netdev);
1060 if (err)
1061 goto out;
1062
1063 adapter->tx_ring->itr_register = E1000_EITR(vector);
1064 adapter->tx_ring->itr_val = adapter->current_itr;
1065 vector++;
1066
1067 err = request_irq(adapter->msix_entries[vector].vector,
1068 igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1069 netdev);
1070 if (err)
1071 goto free_irq_tx;
1072
1073 adapter->rx_ring->itr_register = E1000_EITR(vector);
1074 adapter->rx_ring->itr_val = adapter->current_itr;
1075 vector++;
1076
1077 err = request_irq(adapter->msix_entries[vector].vector,
1078 igbvf_msix_other, 0, netdev->name, netdev);
1079 if (err)
1080 goto free_irq_rx;
1081
1082 igbvf_configure_msix(adapter);
1083 return 0;
1084free_irq_rx:
1085 free_irq(adapter->msix_entries[--vector].vector, netdev);
1086free_irq_tx:
1087 free_irq(adapter->msix_entries[--vector].vector, netdev);
1088out:
1089 return err;
1090}
1091
1092/**
1093 * igbvf_alloc_queues - Allocate memory for all rings
1094 * @adapter: board private structure to initialize
1095 **/
1096static int igbvf_alloc_queues(struct igbvf_adapter *adapter)
1097{
1098 struct net_device *netdev = adapter->netdev;
1099
1100 adapter->tx_ring = kzalloc_obj(struct igbvf_ring);
1101 if (!adapter->tx_ring)
1102 return -ENOMEM;
1103
1104 adapter->rx_ring = kzalloc_obj(struct igbvf_ring);
1105 if (!adapter->rx_ring) {
1106 kfree(adapter->tx_ring);
1107 return -ENOMEM;
1108 }
1109
1110 netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll);
1111
1112 return 0;
1113}
1114
1115/**
1116 * igbvf_request_irq - initialize interrupts
1117 * @adapter: board private structure
1118 *
1119 * Attempts to configure interrupts using the best available
1120 * capabilities of the hardware and kernel.
1121 **/
1122static int igbvf_request_irq(struct igbvf_adapter *adapter)
1123{
1124 int err = -1;
1125
1126 /* igbvf supports msi-x only */
1127 if (adapter->msix_entries)
1128 err = igbvf_request_msix(adapter);
1129
1130 if (!err)
1131 return err;
1132
1133 dev_err(&adapter->pdev->dev,
1134 "Unable to allocate interrupt, Error: %d\n", err);
1135
1136 return err;
1137}
1138
1139static void igbvf_free_irq(struct igbvf_adapter *adapter)
1140{
1141 struct net_device *netdev = adapter->netdev;
1142 int vector;
1143
1144 if (adapter->msix_entries) {
1145 for (vector = 0; vector < 3; vector++)
1146 free_irq(adapter->msix_entries[vector].vector, netdev);
1147 }
1148}
1149
1150/**
1151 * igbvf_irq_disable - Mask off interrupt generation on the NIC
1152 * @adapter: board private structure
1153 **/
1154static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1155{
1156 struct e1000_hw *hw = &adapter->hw;
1157
1158 ew32(EIMC, ~0);
1159
1160 if (adapter->msix_entries)
1161 ew32(EIAC, 0);
1162}
1163
1164/**
1165 * igbvf_irq_enable - Enable default interrupt generation settings
1166 * @adapter: board private structure
1167 **/
1168static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1169{
1170 struct e1000_hw *hw = &adapter->hw;
1171
1172 ew32(EIAC, adapter->eims_enable_mask);
1173 ew32(EIAM, adapter->eims_enable_mask);
1174 ew32(EIMS, adapter->eims_enable_mask);
1175}
1176
1177/**
1178 * igbvf_poll - NAPI Rx polling callback
1179 * @napi: struct associated with this polling callback
1180 * @budget: amount of packets driver is allowed to process this poll
1181 **/
1182static int igbvf_poll(struct napi_struct *napi, int budget)
1183{
1184 struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1185 struct igbvf_adapter *adapter = rx_ring->adapter;
1186 struct e1000_hw *hw = &adapter->hw;
1187 int work_done = 0;
1188
1189 igbvf_clean_rx_irq(adapter, &work_done, budget);
1190
1191 if (work_done == budget)
1192 return budget;
1193
1194 /* Exit the polling mode, but don't re-enable interrupts if stack might
1195 * poll us due to busy-polling
1196 */
1197 if (likely(napi_complete_done(napi, work_done))) {
1198 if (adapter->requested_itr & 3)
1199 igbvf_set_itr(adapter);
1200
1201 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1202 ew32(EIMS, adapter->rx_ring->eims_value);
1203 }
1204
1205 return work_done;
1206}
1207
1208/**
1209 * igbvf_set_rlpml - set receive large packet maximum length
1210 * @adapter: board private structure
1211 *
1212 * Configure the maximum size of packets that will be received
1213 */
1214static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1215{
1216 int max_frame_size;
1217 struct e1000_hw *hw = &adapter->hw;
1218
1219 max_frame_size = adapter->max_frame_size + VLAN_TAG_SIZE;
1220
1221 spin_lock_bh(&hw->mbx_lock);
1222
1223 e1000_rlpml_set_vf(hw, max_frame_size);
1224
1225 spin_unlock_bh(&hw->mbx_lock);
1226}
1227
1228static int igbvf_vlan_rx_add_vid(struct net_device *netdev,
1229 __be16 proto, u16 vid)
1230{
1231 struct igbvf_adapter *adapter = netdev_priv(netdev);
1232 struct e1000_hw *hw = &adapter->hw;
1233
1234 spin_lock_bh(&hw->mbx_lock);
1235
1236 if (hw->mac.ops.set_vfta(hw, vid, true)) {
1237 dev_warn(&adapter->pdev->dev, "Vlan id %d is not added\n", vid);
1238 spin_unlock_bh(&hw->mbx_lock);
1239 return -EINVAL;
1240 }
1241
1242 spin_unlock_bh(&hw->mbx_lock);
1243
1244 set_bit(vid, adapter->active_vlans);
1245 return 0;
1246}
1247
1248static int igbvf_vlan_rx_kill_vid(struct net_device *netdev,
1249 __be16 proto, u16 vid)
1250{
1251 struct igbvf_adapter *adapter = netdev_priv(netdev);
1252 struct e1000_hw *hw = &adapter->hw;
1253
1254 spin_lock_bh(&hw->mbx_lock);
1255
1256 if (hw->mac.ops.set_vfta(hw, vid, false)) {
1257 dev_err(&adapter->pdev->dev,
1258 "Failed to remove vlan id %d\n", vid);
1259 spin_unlock_bh(&hw->mbx_lock);
1260 return -EINVAL;
1261 }
1262
1263 spin_unlock_bh(&hw->mbx_lock);
1264
1265 clear_bit(vid, adapter->active_vlans);
1266 return 0;
1267}
1268
1269static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1270{
1271 u16 vid;
1272
1273 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1274 igbvf_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
1275}
1276
1277/**
1278 * igbvf_configure_tx - Configure Transmit Unit after Reset
1279 * @adapter: board private structure
1280 *
1281 * Configure the Tx unit of the MAC after a reset.
1282 **/
1283static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1284{
1285 struct e1000_hw *hw = &adapter->hw;
1286 struct igbvf_ring *tx_ring = adapter->tx_ring;
1287 u64 tdba;
1288 u32 txdctl, dca_txctrl;
1289
1290 /* disable transmits */
1291 txdctl = er32(TXDCTL(0));
1292 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1293 e1e_flush();
1294 msleep(10);
1295
1296 /* Setup the HW Tx Head and Tail descriptor pointers */
1297 ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1298 tdba = tx_ring->dma;
1299 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1300 ew32(TDBAH(0), (tdba >> 32));
1301 ew32(TDH(0), 0);
1302 ew32(TDT(0), 0);
1303 tx_ring->head = E1000_TDH(0);
1304 tx_ring->tail = E1000_TDT(0);
1305
1306 /* Turn off Relaxed Ordering on head write-backs. The writebacks
1307 * MUST be delivered in order or it will completely screw up
1308 * our bookkeeping.
1309 */
1310 dca_txctrl = er32(DCA_TXCTRL(0));
1311 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1312 ew32(DCA_TXCTRL(0), dca_txctrl);
1313
1314 /* enable transmits */
1315 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1316 ew32(TXDCTL(0), txdctl);
1317
1318 /* Setup Transmit Descriptor Settings for eop descriptor */
1319 adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1320
1321 /* enable Report Status bit */
1322 adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1323}
1324
1325/**
1326 * igbvf_setup_srrctl - configure the receive control registers
1327 * @adapter: Board private structure
1328 **/
1329static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1330{
1331 struct e1000_hw *hw = &adapter->hw;
1332 u32 srrctl = 0;
1333
1334 srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1335 E1000_SRRCTL_BSIZEHDR_MASK |
1336 E1000_SRRCTL_BSIZEPKT_MASK);
1337
1338 /* Enable queue drop to avoid head of line blocking */
1339 srrctl |= E1000_SRRCTL_DROP_EN;
1340
1341 /* Setup buffer sizes */
1342 srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1343 E1000_SRRCTL_BSIZEPKT_SHIFT;
1344
1345 if (adapter->rx_buffer_len < 2048) {
1346 adapter->rx_ps_hdr_size = 0;
1347 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1348 } else {
1349 adapter->rx_ps_hdr_size = 128;
1350 srrctl |= adapter->rx_ps_hdr_size <<
1351 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1352 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1353 }
1354
1355 ew32(SRRCTL(0), srrctl);
1356}
1357
1358/**
1359 * igbvf_configure_rx - Configure Receive Unit after Reset
1360 * @adapter: board private structure
1361 *
1362 * Configure the Rx unit of the MAC after a reset.
1363 **/
1364static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1365{
1366 struct e1000_hw *hw = &adapter->hw;
1367 struct igbvf_ring *rx_ring = adapter->rx_ring;
1368 u64 rdba;
1369 u32 rxdctl;
1370
1371 /* disable receives */
1372 rxdctl = er32(RXDCTL(0));
1373 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1374 e1e_flush();
1375 msleep(10);
1376
1377 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1378 * the Base and Length of the Rx Descriptor Ring
1379 */
1380 rdba = rx_ring->dma;
1381 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1382 ew32(RDBAH(0), (rdba >> 32));
1383 ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1384 rx_ring->head = E1000_RDH(0);
1385 rx_ring->tail = E1000_RDT(0);
1386 ew32(RDH(0), 0);
1387 ew32(RDT(0), 0);
1388
1389 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1390 rxdctl &= 0xFFF00000;
1391 rxdctl |= IGBVF_RX_PTHRESH;
1392 rxdctl |= IGBVF_RX_HTHRESH << 8;
1393 rxdctl |= IGBVF_RX_WTHRESH << 16;
1394
1395 igbvf_set_rlpml(adapter);
1396
1397 /* enable receives */
1398 ew32(RXDCTL(0), rxdctl);
1399}
1400
1401/**
1402 * igbvf_set_multi - Multicast and Promiscuous mode set
1403 * @netdev: network interface device structure
1404 *
1405 * The set_multi entry point is called whenever the multicast address
1406 * list or the network interface flags are updated. This routine is
1407 * responsible for configuring the hardware for proper multicast,
1408 * promiscuous mode, and all-multi behavior.
1409 **/
1410static void igbvf_set_multi(struct net_device *netdev)
1411{
1412 struct igbvf_adapter *adapter = netdev_priv(netdev);
1413 struct e1000_hw *hw = &adapter->hw;
1414 struct netdev_hw_addr *ha;
1415 u8 *mta_list = NULL;
1416 int i;
1417
1418 if (!netdev_mc_empty(netdev)) {
1419 mta_list = kmalloc_array(netdev_mc_count(netdev), ETH_ALEN,
1420 GFP_ATOMIC);
1421 if (!mta_list)
1422 return;
1423 }
1424
1425 /* prepare a packed array of only addresses. */
1426 i = 0;
1427 netdev_for_each_mc_addr(ha, netdev)
1428 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1429
1430 spin_lock_bh(&hw->mbx_lock);
1431
1432 hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1433
1434 spin_unlock_bh(&hw->mbx_lock);
1435 kfree(mta_list);
1436}
1437
1438/**
1439 * igbvf_set_uni - Configure unicast MAC filters
1440 * @netdev: network interface device structure
1441 *
1442 * This routine is responsible for configuring the hardware for proper
1443 * unicast filters.
1444 **/
1445static int igbvf_set_uni(struct net_device *netdev)
1446{
1447 struct igbvf_adapter *adapter = netdev_priv(netdev);
1448 struct e1000_hw *hw = &adapter->hw;
1449
1450 if (netdev_uc_count(netdev) > IGBVF_MAX_MAC_FILTERS) {
1451 pr_err("Too many unicast filters - No Space\n");
1452 return -ENOSPC;
1453 }
1454
1455 spin_lock_bh(&hw->mbx_lock);
1456
1457 /* Clear all unicast MAC filters */
1458 hw->mac.ops.set_uc_addr(hw, E1000_VF_MAC_FILTER_CLR, NULL);
1459
1460 spin_unlock_bh(&hw->mbx_lock);
1461
1462 if (!netdev_uc_empty(netdev)) {
1463 struct netdev_hw_addr *ha;
1464
1465 /* Add MAC filters one by one */
1466 netdev_for_each_uc_addr(ha, netdev) {
1467 spin_lock_bh(&hw->mbx_lock);
1468
1469 hw->mac.ops.set_uc_addr(hw, E1000_VF_MAC_FILTER_ADD,
1470 ha->addr);
1471
1472 spin_unlock_bh(&hw->mbx_lock);
1473 udelay(200);
1474 }
1475 }
1476
1477 return 0;
1478}
1479
1480static void igbvf_set_rx_mode(struct net_device *netdev)
1481{
1482 igbvf_set_multi(netdev);
1483 igbvf_set_uni(netdev);
1484}
1485
1486/**
1487 * igbvf_configure - configure the hardware for Rx and Tx
1488 * @adapter: private board structure
1489 **/
1490static void igbvf_configure(struct igbvf_adapter *adapter)
1491{
1492 igbvf_set_rx_mode(adapter->netdev);
1493
1494 igbvf_restore_vlan(adapter);
1495
1496 igbvf_configure_tx(adapter);
1497 igbvf_setup_srrctl(adapter);
1498 igbvf_configure_rx(adapter);
1499 igbvf_alloc_rx_buffers(adapter->rx_ring,
1500 igbvf_desc_unused(adapter->rx_ring));
1501}
1502
1503/* igbvf_reset - bring the hardware into a known good state
1504 * @adapter: private board structure
1505 *
1506 * This function boots the hardware and enables some settings that
1507 * require a configuration cycle of the hardware - those cannot be
1508 * set/changed during runtime. After reset the device needs to be
1509 * properly configured for Rx, Tx etc.
1510 */
1511static void igbvf_reset(struct igbvf_adapter *adapter)
1512{
1513 struct e1000_mac_info *mac = &adapter->hw.mac;
1514 struct net_device *netdev = adapter->netdev;
1515 struct e1000_hw *hw = &adapter->hw;
1516
1517 spin_lock_bh(&hw->mbx_lock);
1518
1519 /* Allow time for pending master requests to run */
1520 if (mac->ops.reset_hw(hw))
1521 dev_info(&adapter->pdev->dev, "PF still resetting\n");
1522
1523 mac->ops.init_hw(hw);
1524
1525 spin_unlock_bh(&hw->mbx_lock);
1526
1527 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1528 eth_hw_addr_set(netdev, adapter->hw.mac.addr);
1529 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1530 netdev->addr_len);
1531 }
1532
1533 adapter->last_reset = jiffies;
1534}
1535
1536int igbvf_up(struct igbvf_adapter *adapter)
1537{
1538 struct e1000_hw *hw = &adapter->hw;
1539
1540 /* hardware has been reset, we need to reload some things */
1541 igbvf_configure(adapter);
1542
1543 clear_bit(__IGBVF_DOWN, &adapter->state);
1544
1545 napi_enable(&adapter->rx_ring->napi);
1546 if (adapter->msix_entries)
1547 igbvf_configure_msix(adapter);
1548
1549 /* Clear any pending interrupts. */
1550 er32(EICR);
1551 igbvf_irq_enable(adapter);
1552
1553 /* start the watchdog */
1554 hw->mac.get_link_status = 1;
1555 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1556
1557 return 0;
1558}
1559
1560void igbvf_down(struct igbvf_adapter *adapter)
1561{
1562 struct net_device *netdev = adapter->netdev;
1563 struct e1000_hw *hw = &adapter->hw;
1564 u32 rxdctl, txdctl;
1565
1566 /* signal that we're down so the interrupt handler does not
1567 * reschedule our watchdog timer
1568 */
1569 set_bit(__IGBVF_DOWN, &adapter->state);
1570
1571 /* disable receives in the hardware */
1572 rxdctl = er32(RXDCTL(0));
1573 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1574
1575 netif_carrier_off(netdev);
1576 netif_stop_queue(netdev);
1577
1578 /* disable transmits in the hardware */
1579 txdctl = er32(TXDCTL(0));
1580 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1581
1582 /* flush both disables and wait for them to finish */
1583 e1e_flush();
1584 msleep(10);
1585
1586 napi_disable(&adapter->rx_ring->napi);
1587
1588 igbvf_irq_disable(adapter);
1589
1590 timer_delete_sync(&adapter->watchdog_timer);
1591
1592 /* record the stats before reset*/
1593 igbvf_update_stats(adapter);
1594
1595 adapter->link_speed = 0;
1596 adapter->link_duplex = 0;
1597
1598 igbvf_reset(adapter);
1599 igbvf_clean_tx_ring(adapter->tx_ring);
1600 igbvf_clean_rx_ring(adapter->rx_ring);
1601}
1602
1603void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1604{
1605 might_sleep();
1606 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1607 usleep_range(1000, 2000);
1608 igbvf_down(adapter);
1609 igbvf_up(adapter);
1610 clear_bit(__IGBVF_RESETTING, &adapter->state);
1611}
1612
1613/**
1614 * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1615 * @adapter: board private structure to initialize
1616 *
1617 * igbvf_sw_init initializes the Adapter private data structure.
1618 * Fields are initialized based on PCI device information and
1619 * OS network device settings (MTU size).
1620 **/
1621static int igbvf_sw_init(struct igbvf_adapter *adapter)
1622{
1623 struct net_device *netdev = adapter->netdev;
1624 s32 rc;
1625
1626 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1627 adapter->rx_ps_hdr_size = 0;
1628 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1629 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1630
1631 adapter->requested_itr = 3;
1632 adapter->current_itr = IGBVF_START_ITR;
1633
1634 /* Set various function pointers */
1635 adapter->ei->init_ops(&adapter->hw);
1636
1637 rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1638 if (rc)
1639 return rc;
1640
1641 rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1642 if (rc)
1643 return rc;
1644
1645 igbvf_set_interrupt_capability(adapter);
1646
1647 if (igbvf_alloc_queues(adapter))
1648 return -ENOMEM;
1649
1650 /* Explicitly disable IRQ since the NIC can be in any state. */
1651 igbvf_irq_disable(adapter);
1652
1653 spin_lock_init(&adapter->hw.mbx_lock);
1654
1655 set_bit(__IGBVF_DOWN, &adapter->state);
1656 return 0;
1657}
1658
1659static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1660{
1661 struct e1000_hw *hw = &adapter->hw;
1662
1663 adapter->stats.last_gprc = er32(VFGPRC);
1664 adapter->stats.last_gorc = er32(VFGORC);
1665 adapter->stats.last_gptc = er32(VFGPTC);
1666 adapter->stats.last_gotc = er32(VFGOTC);
1667 adapter->stats.last_mprc = er32(VFMPRC);
1668 adapter->stats.last_gotlbc = er32(VFGOTLBC);
1669 adapter->stats.last_gptlbc = er32(VFGPTLBC);
1670 adapter->stats.last_gorlbc = er32(VFGORLBC);
1671 adapter->stats.last_gprlbc = er32(VFGPRLBC);
1672
1673 adapter->stats.base_gprc = er32(VFGPRC);
1674 adapter->stats.base_gorc = er32(VFGORC);
1675 adapter->stats.base_gptc = er32(VFGPTC);
1676 adapter->stats.base_gotc = er32(VFGOTC);
1677 adapter->stats.base_mprc = er32(VFMPRC);
1678 adapter->stats.base_gotlbc = er32(VFGOTLBC);
1679 adapter->stats.base_gptlbc = er32(VFGPTLBC);
1680 adapter->stats.base_gorlbc = er32(VFGORLBC);
1681 adapter->stats.base_gprlbc = er32(VFGPRLBC);
1682}
1683
1684/**
1685 * igbvf_open - Called when a network interface is made active
1686 * @netdev: network interface device structure
1687 *
1688 * Returns 0 on success, negative value on failure
1689 *
1690 * The open entry point is called when a network interface is made
1691 * active by the system (IFF_UP). At this point all resources needed
1692 * for transmit and receive operations are allocated, the interrupt
1693 * handler is registered with the OS, the watchdog timer is started,
1694 * and the stack is notified that the interface is ready.
1695 **/
1696static int igbvf_open(struct net_device *netdev)
1697{
1698 struct igbvf_adapter *adapter = netdev_priv(netdev);
1699 struct e1000_hw *hw = &adapter->hw;
1700 int err;
1701
1702 /* disallow open during test */
1703 if (test_bit(__IGBVF_TESTING, &adapter->state))
1704 return -EBUSY;
1705
1706 /* allocate transmit descriptors */
1707 err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1708 if (err)
1709 goto err_setup_tx;
1710
1711 /* allocate receive descriptors */
1712 err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1713 if (err)
1714 goto err_setup_rx;
1715
1716 /* before we allocate an interrupt, we must be ready to handle it.
1717 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1718 * as soon as we call pci_request_irq, so we have to setup our
1719 * clean_rx handler before we do so.
1720 */
1721 igbvf_configure(adapter);
1722
1723 err = igbvf_request_irq(adapter);
1724 if (err)
1725 goto err_req_irq;
1726
1727 /* From here on the code is the same as igbvf_up() */
1728 clear_bit(__IGBVF_DOWN, &adapter->state);
1729
1730 napi_enable(&adapter->rx_ring->napi);
1731
1732 /* clear any pending interrupts */
1733 er32(EICR);
1734
1735 igbvf_irq_enable(adapter);
1736
1737 /* start the watchdog */
1738 hw->mac.get_link_status = 1;
1739 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1740
1741 return 0;
1742
1743err_req_irq:
1744 igbvf_free_rx_resources(adapter->rx_ring);
1745err_setup_rx:
1746 igbvf_free_tx_resources(adapter->tx_ring);
1747err_setup_tx:
1748 igbvf_reset(adapter);
1749
1750 return err;
1751}
1752
1753/**
1754 * igbvf_close - Disables a network interface
1755 * @netdev: network interface device structure
1756 *
1757 * Returns 0, this is not allowed to fail
1758 *
1759 * The close entry point is called when an interface is de-activated
1760 * by the OS. The hardware is still under the drivers control, but
1761 * needs to be disabled. A global MAC reset is issued to stop the
1762 * hardware, and all transmit and receive resources are freed.
1763 **/
1764static int igbvf_close(struct net_device *netdev)
1765{
1766 struct igbvf_adapter *adapter = netdev_priv(netdev);
1767
1768 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1769 igbvf_down(adapter);
1770
1771 igbvf_free_irq(adapter);
1772
1773 igbvf_free_tx_resources(adapter->tx_ring);
1774 igbvf_free_rx_resources(adapter->rx_ring);
1775
1776 return 0;
1777}
1778
1779/**
1780 * igbvf_set_mac - Change the Ethernet Address of the NIC
1781 * @netdev: network interface device structure
1782 * @p: pointer to an address structure
1783 *
1784 * Returns 0 on success, negative on failure
1785 **/
1786static int igbvf_set_mac(struct net_device *netdev, void *p)
1787{
1788 struct igbvf_adapter *adapter = netdev_priv(netdev);
1789 struct e1000_hw *hw = &adapter->hw;
1790 struct sockaddr *addr = p;
1791
1792 if (!is_valid_ether_addr(addr->sa_data))
1793 return -EADDRNOTAVAIL;
1794
1795 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1796
1797 spin_lock_bh(&hw->mbx_lock);
1798
1799 hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1800
1801 spin_unlock_bh(&hw->mbx_lock);
1802
1803 if (!ether_addr_equal(addr->sa_data, hw->mac.addr))
1804 return -EADDRNOTAVAIL;
1805
1806 eth_hw_addr_set(netdev, addr->sa_data);
1807
1808 return 0;
1809}
1810
1811#define UPDATE_VF_COUNTER(reg, name) \
1812{ \
1813 u32 current_counter = er32(reg); \
1814 if (current_counter < adapter->stats.last_##name) \
1815 adapter->stats.name += 0x100000000LL; \
1816 adapter->stats.last_##name = current_counter; \
1817 adapter->stats.name &= 0xFFFFFFFF00000000LL; \
1818 adapter->stats.name |= current_counter; \
1819}
1820
1821/**
1822 * igbvf_update_stats - Update the board statistics counters
1823 * @adapter: board private structure
1824**/
1825void igbvf_update_stats(struct igbvf_adapter *adapter)
1826{
1827 struct e1000_hw *hw = &adapter->hw;
1828 struct pci_dev *pdev = adapter->pdev;
1829
1830 /* Prevent stats update while adapter is being reset, link is down
1831 * or if the pci connection is down.
1832 */
1833 if (adapter->link_speed == 0)
1834 return;
1835
1836 if (test_bit(__IGBVF_RESETTING, &adapter->state))
1837 return;
1838
1839 if (pci_channel_offline(pdev))
1840 return;
1841
1842 UPDATE_VF_COUNTER(VFGPRC, gprc);
1843 UPDATE_VF_COUNTER(VFGORC, gorc);
1844 UPDATE_VF_COUNTER(VFGPTC, gptc);
1845 UPDATE_VF_COUNTER(VFGOTC, gotc);
1846 UPDATE_VF_COUNTER(VFMPRC, mprc);
1847 UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1848 UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1849 UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1850 UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1851
1852 /* Fill out the OS statistics structure */
1853 adapter->netdev->stats.multicast = adapter->stats.mprc;
1854}
1855
1856static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1857{
1858 dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s Duplex\n",
1859 adapter->link_speed,
1860 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half");
1861}
1862
1863static bool igbvf_has_link(struct igbvf_adapter *adapter)
1864{
1865 struct e1000_hw *hw = &adapter->hw;
1866 s32 ret_val = E1000_SUCCESS;
1867 bool link_active;
1868
1869 /* If interface is down, stay link down */
1870 if (test_bit(__IGBVF_DOWN, &adapter->state))
1871 return false;
1872
1873 spin_lock_bh(&hw->mbx_lock);
1874
1875 ret_val = hw->mac.ops.check_for_link(hw);
1876
1877 spin_unlock_bh(&hw->mbx_lock);
1878
1879 link_active = !hw->mac.get_link_status;
1880
1881 /* if check for link returns error we will need to reset */
1882 if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1883 schedule_work(&adapter->reset_task);
1884
1885 return link_active;
1886}
1887
1888/**
1889 * igbvf_watchdog - Timer Call-back
1890 * @t: timer list pointer containing private struct
1891 **/
1892static void igbvf_watchdog(struct timer_list *t)
1893{
1894 struct igbvf_adapter *adapter = timer_container_of(adapter, t,
1895 watchdog_timer);
1896
1897 /* Do the rest outside of interrupt context */
1898 schedule_work(&adapter->watchdog_task);
1899}
1900
1901static void igbvf_watchdog_task(struct work_struct *work)
1902{
1903 struct igbvf_adapter *adapter = container_of(work,
1904 struct igbvf_adapter,
1905 watchdog_task);
1906 struct net_device *netdev = adapter->netdev;
1907 struct e1000_mac_info *mac = &adapter->hw.mac;
1908 struct igbvf_ring *tx_ring = adapter->tx_ring;
1909 struct e1000_hw *hw = &adapter->hw;
1910 u32 link;
1911 int tx_pending = 0;
1912
1913 link = igbvf_has_link(adapter);
1914
1915 if (link) {
1916 if (!netif_carrier_ok(netdev)) {
1917 mac->ops.get_link_up_info(&adapter->hw,
1918 &adapter->link_speed,
1919 &adapter->link_duplex);
1920 igbvf_print_link_info(adapter);
1921
1922 netif_carrier_on(netdev);
1923 netif_wake_queue(netdev);
1924 }
1925 } else {
1926 if (netif_carrier_ok(netdev)) {
1927 adapter->link_speed = 0;
1928 adapter->link_duplex = 0;
1929 dev_info(&adapter->pdev->dev, "Link is Down\n");
1930 netif_carrier_off(netdev);
1931 netif_stop_queue(netdev);
1932 }
1933 }
1934
1935 if (netif_carrier_ok(netdev)) {
1936 igbvf_update_stats(adapter);
1937 } else {
1938 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1939 tx_ring->count);
1940 if (tx_pending) {
1941 /* We've lost link, so the controller stops DMA,
1942 * but we've got queued Tx work that's never going
1943 * to get done, so reset controller to flush Tx.
1944 * (Do the reset outside of interrupt context).
1945 */
1946 adapter->tx_timeout_count++;
1947 schedule_work(&adapter->reset_task);
1948 }
1949 }
1950
1951 /* Cause software interrupt to ensure Rx ring is cleaned */
1952 ew32(EICS, adapter->rx_ring->eims_value);
1953
1954 /* Reset the timer */
1955 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1956 mod_timer(&adapter->watchdog_timer,
1957 round_jiffies(jiffies + (2 * HZ)));
1958}
1959
1960#define IGBVF_TX_FLAGS_CSUM 0x00000001
1961#define IGBVF_TX_FLAGS_VLAN 0x00000002
1962#define IGBVF_TX_FLAGS_TSO 0x00000004
1963#define IGBVF_TX_FLAGS_IPV4 0x00000008
1964#define IGBVF_TX_FLAGS_VLAN_MASK 0xffff0000
1965#define IGBVF_TX_FLAGS_VLAN_SHIFT 16
1966
1967static void igbvf_tx_ctxtdesc(struct igbvf_ring *tx_ring, u32 vlan_macip_lens,
1968 u32 type_tucmd, u32 mss_l4len_idx)
1969{
1970 struct e1000_adv_tx_context_desc *context_desc;
1971 struct igbvf_buffer *buffer_info;
1972 u16 i = tx_ring->next_to_use;
1973
1974 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1975 buffer_info = &tx_ring->buffer_info[i];
1976
1977 i++;
1978 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1979
1980 /* set bits to identify this as an advanced context descriptor */
1981 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
1982
1983 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
1984 context_desc->seqnum_seed = 0;
1985 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
1986 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1987
1988 buffer_info->time_stamp = jiffies;
1989 buffer_info->dma = 0;
1990}
1991
1992static int igbvf_tso(struct igbvf_ring *tx_ring,
1993 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1994{
1995 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
1996 union {
1997 struct iphdr *v4;
1998 struct ipv6hdr *v6;
1999 unsigned char *hdr;
2000 } ip;
2001 union {
2002 struct tcphdr *tcp;
2003 unsigned char *hdr;
2004 } l4;
2005 u32 paylen, l4_offset;
2006 int err;
2007
2008 if (skb->ip_summed != CHECKSUM_PARTIAL)
2009 return 0;
2010
2011 if (!skb_is_gso(skb))
2012 return 0;
2013
2014 err = skb_cow_head(skb, 0);
2015 if (err < 0)
2016 return err;
2017
2018 ip.hdr = skb_network_header(skb);
2019 l4.hdr = skb_checksum_start(skb);
2020
2021 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2022 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
2023
2024 /* initialize outer IP header fields */
2025 if (ip.v4->version == 4) {
2026 unsigned char *csum_start = skb_checksum_start(skb);
2027 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
2028
2029 /* IP header will have to cancel out any data that
2030 * is not a part of the outer IP header
2031 */
2032 ip.v4->check = csum_fold(csum_partial(trans_start,
2033 csum_start - trans_start,
2034 0));
2035 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
2036
2037 ip.v4->tot_len = 0;
2038 } else {
2039 ip.v6->payload_len = 0;
2040 }
2041
2042 /* determine offset of inner transport header */
2043 l4_offset = l4.hdr - skb->data;
2044
2045 /* compute length of segmentation header */
2046 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
2047
2048 /* remove payload length from inner checksum */
2049 paylen = skb->len - l4_offset;
2050 csum_replace_by_diff(&l4.tcp->check, (__force __wsum)htonl(paylen));
2051
2052 /* MSS L4LEN IDX */
2053 mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT;
2054 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
2055
2056 /* VLAN MACLEN IPLEN */
2057 vlan_macip_lens = l4.hdr - ip.hdr;
2058 vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT;
2059 vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK;
2060
2061 igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
2062
2063 return 1;
2064}
2065
2066static bool igbvf_tx_csum(struct igbvf_ring *tx_ring, struct sk_buff *skb,
2067 u32 tx_flags, __be16 protocol)
2068{
2069 u32 vlan_macip_lens = 0;
2070 u32 type_tucmd = 0;
2071
2072 if (skb->ip_summed != CHECKSUM_PARTIAL) {
2073csum_failed:
2074 if (!(tx_flags & IGBVF_TX_FLAGS_VLAN))
2075 return false;
2076 goto no_csum;
2077 }
2078
2079 switch (skb->csum_offset) {
2080 case offsetof(struct tcphdr, check):
2081 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
2082 fallthrough;
2083 case offsetof(struct udphdr, check):
2084 break;
2085 case offsetof(struct sctphdr, checksum):
2086 /* validate that this is actually an SCTP request */
2087 if (skb_csum_is_sctp(skb)) {
2088 type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP;
2089 break;
2090 }
2091 fallthrough;
2092 default:
2093 skb_checksum_help(skb);
2094 goto csum_failed;
2095 }
2096
2097 vlan_macip_lens = skb_checksum_start_offset(skb) -
2098 skb_network_offset(skb);
2099no_csum:
2100 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
2101 vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK;
2102
2103 igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0);
2104 return true;
2105}
2106
2107static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2108{
2109 struct igbvf_adapter *adapter = netdev_priv(netdev);
2110
2111 /* there is enough descriptors then we don't need to worry */
2112 if (igbvf_desc_unused(adapter->tx_ring) >= size)
2113 return 0;
2114
2115 netif_stop_queue(netdev);
2116
2117 /* Herbert's original patch had:
2118 * smp_mb__after_netif_stop_queue();
2119 * but since that doesn't exist yet, just open code it.
2120 */
2121 smp_mb();
2122
2123 /* We need to check again just in case room has been made available */
2124 if (igbvf_desc_unused(adapter->tx_ring) < size)
2125 return -EBUSY;
2126
2127 netif_wake_queue(netdev);
2128
2129 ++adapter->restart_queue;
2130 return 0;
2131}
2132
2133#define IGBVF_MAX_TXD_PWR 16
2134#define IGBVF_MAX_DATA_PER_TXD (1u << IGBVF_MAX_TXD_PWR)
2135
2136static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2137 struct igbvf_ring *tx_ring,
2138 struct sk_buff *skb)
2139{
2140 struct igbvf_buffer *buffer_info;
2141 struct pci_dev *pdev = adapter->pdev;
2142 unsigned int len = skb_headlen(skb);
2143 unsigned int count = 0, i;
2144 unsigned int f;
2145
2146 i = tx_ring->next_to_use;
2147
2148 buffer_info = &tx_ring->buffer_info[i];
2149 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2150 buffer_info->length = len;
2151 /* set time_stamp *before* dma to help avoid a possible race */
2152 buffer_info->time_stamp = jiffies;
2153 buffer_info->mapped_as_page = false;
2154 buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2155 DMA_TO_DEVICE);
2156 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2157 goto dma_error;
2158
2159 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2160 const skb_frag_t *frag;
2161
2162 count++;
2163 i++;
2164 if (i == tx_ring->count)
2165 i = 0;
2166
2167 frag = &skb_shinfo(skb)->frags[f];
2168 len = skb_frag_size(frag);
2169
2170 buffer_info = &tx_ring->buffer_info[i];
2171 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2172 buffer_info->length = len;
2173 buffer_info->time_stamp = jiffies;
2174 buffer_info->mapped_as_page = true;
2175 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag, 0, len,
2176 DMA_TO_DEVICE);
2177 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2178 goto dma_error;
2179 }
2180
2181 tx_ring->buffer_info[i].skb = skb;
2182
2183 return ++count;
2184
2185dma_error:
2186 dev_err(&pdev->dev, "TX DMA map failed\n");
2187
2188 /* clear timestamp and dma mappings for failed buffer_info mapping */
2189 buffer_info->dma = 0;
2190 buffer_info->time_stamp = 0;
2191 buffer_info->length = 0;
2192 buffer_info->mapped_as_page = false;
2193
2194 /* clear timestamp and dma mappings for remaining portion of packet */
2195 while (count--) {
2196 if (i == 0)
2197 i += tx_ring->count;
2198 i--;
2199 buffer_info = &tx_ring->buffer_info[i];
2200 igbvf_put_txbuf(adapter, buffer_info);
2201 }
2202
2203 return 0;
2204}
2205
2206static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2207 struct igbvf_ring *tx_ring,
2208 int tx_flags, int count,
2209 unsigned int first, u32 paylen,
2210 u8 hdr_len)
2211{
2212 union e1000_adv_tx_desc *tx_desc = NULL;
2213 struct igbvf_buffer *buffer_info;
2214 u32 olinfo_status = 0, cmd_type_len;
2215 unsigned int i;
2216
2217 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2218 E1000_ADVTXD_DCMD_DEXT);
2219
2220 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2221 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2222
2223 if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2224 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2225
2226 /* insert tcp checksum */
2227 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2228
2229 /* insert ip checksum */
2230 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2231 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2232
2233 } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2234 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2235 }
2236
2237 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2238
2239 i = tx_ring->next_to_use;
2240 while (count--) {
2241 buffer_info = &tx_ring->buffer_info[i];
2242 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2243 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2244 tx_desc->read.cmd_type_len =
2245 cpu_to_le32(cmd_type_len | buffer_info->length);
2246 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2247 i++;
2248 if (i == tx_ring->count)
2249 i = 0;
2250 }
2251
2252 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2253 /* Force memory writes to complete before letting h/w
2254 * know there are new descriptors to fetch. (Only
2255 * applicable for weak-ordered memory model archs,
2256 * such as IA-64).
2257 */
2258 wmb();
2259
2260 tx_ring->buffer_info[first].next_to_watch = tx_desc;
2261 tx_ring->next_to_use = i;
2262 writel(i, adapter->hw.hw_addr + tx_ring->tail);
2263}
2264
2265static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2266 struct net_device *netdev,
2267 struct igbvf_ring *tx_ring)
2268{
2269 struct igbvf_adapter *adapter = netdev_priv(netdev);
2270 unsigned int first, tx_flags = 0;
2271 u8 hdr_len = 0;
2272 int count = 0;
2273 int tso = 0;
2274 __be16 protocol = vlan_get_protocol(skb);
2275
2276 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2277 dev_kfree_skb_any(skb);
2278 return NETDEV_TX_OK;
2279 }
2280
2281 if (skb->len <= 0) {
2282 dev_kfree_skb_any(skb);
2283 return NETDEV_TX_OK;
2284 }
2285
2286 /* need: count + 4 desc gap to keep tail from touching
2287 * + 2 desc gap to keep tail from touching head,
2288 * + 1 desc for skb->data,
2289 * + 1 desc for context descriptor,
2290 * head, otherwise try next time
2291 */
2292 if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2293 /* this is a hard error */
2294 return NETDEV_TX_BUSY;
2295 }
2296
2297 if (skb_vlan_tag_present(skb)) {
2298 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2299 tx_flags |= (skb_vlan_tag_get(skb) <<
2300 IGBVF_TX_FLAGS_VLAN_SHIFT);
2301 }
2302
2303 if (protocol == htons(ETH_P_IP))
2304 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2305
2306 first = tx_ring->next_to_use;
2307
2308 tso = igbvf_tso(tx_ring, skb, tx_flags, &hdr_len);
2309 if (unlikely(tso < 0)) {
2310 dev_kfree_skb_any(skb);
2311 return NETDEV_TX_OK;
2312 }
2313
2314 if (tso)
2315 tx_flags |= IGBVF_TX_FLAGS_TSO;
2316 else if (igbvf_tx_csum(tx_ring, skb, tx_flags, protocol) &&
2317 (skb->ip_summed == CHECKSUM_PARTIAL))
2318 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2319
2320 /* count reflects descriptors mapped, if 0 then mapping error
2321 * has occurred and we need to rewind the descriptor queue
2322 */
2323 count = igbvf_tx_map_adv(adapter, tx_ring, skb);
2324
2325 if (count) {
2326 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2327 first, skb->len, hdr_len);
2328 /* Make sure there is space in the ring for the next send. */
2329 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2330 } else {
2331 dev_kfree_skb_any(skb);
2332 tx_ring->buffer_info[first].time_stamp = 0;
2333 tx_ring->next_to_use = first;
2334 }
2335
2336 return NETDEV_TX_OK;
2337}
2338
2339static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2340 struct net_device *netdev)
2341{
2342 struct igbvf_adapter *adapter = netdev_priv(netdev);
2343 struct igbvf_ring *tx_ring;
2344
2345 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2346 dev_kfree_skb_any(skb);
2347 return NETDEV_TX_OK;
2348 }
2349
2350 tx_ring = &adapter->tx_ring[0];
2351
2352 return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2353}
2354
2355/**
2356 * igbvf_tx_timeout - Respond to a Tx Hang
2357 * @netdev: network interface device structure
2358 * @txqueue: queue timing out (unused)
2359 **/
2360static void igbvf_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
2361{
2362 struct igbvf_adapter *adapter = netdev_priv(netdev);
2363
2364 /* Do the reset outside of interrupt context */
2365 adapter->tx_timeout_count++;
2366 schedule_work(&adapter->reset_task);
2367}
2368
2369static void igbvf_reset_task(struct work_struct *work)
2370{
2371 struct igbvf_adapter *adapter;
2372
2373 adapter = container_of(work, struct igbvf_adapter, reset_task);
2374
2375 igbvf_reinit_locked(adapter);
2376}
2377
2378/**
2379 * igbvf_change_mtu - Change the Maximum Transfer Unit
2380 * @netdev: network interface device structure
2381 * @new_mtu: new value for maximum frame size
2382 *
2383 * Returns 0 on success, negative on failure
2384 **/
2385static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2386{
2387 struct igbvf_adapter *adapter = netdev_priv(netdev);
2388 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2389
2390 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2391 usleep_range(1000, 2000);
2392 /* igbvf_down has a dependency on max_frame_size */
2393 adapter->max_frame_size = max_frame;
2394 if (netif_running(netdev))
2395 igbvf_down(adapter);
2396
2397 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2398 * means we reserve 2 more, this pushes us to allocate from the next
2399 * larger slab size.
2400 * i.e. RXBUFFER_2048 --> size-4096 slab
2401 * However with the new *_jumbo_rx* routines, jumbo receives will use
2402 * fragmented skbs
2403 */
2404
2405 if (max_frame <= 1024)
2406 adapter->rx_buffer_len = 1024;
2407 else if (max_frame <= 2048)
2408 adapter->rx_buffer_len = 2048;
2409 else
2410#if (PAGE_SIZE / 2) > 16384
2411 adapter->rx_buffer_len = 16384;
2412#else
2413 adapter->rx_buffer_len = PAGE_SIZE / 2;
2414#endif
2415
2416 /* adjust allocation if LPE protects us, and we aren't using SBP */
2417 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2418 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2419 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2420 ETH_FCS_LEN;
2421
2422 netdev_dbg(netdev, "changing MTU from %d to %d\n",
2423 netdev->mtu, new_mtu);
2424 WRITE_ONCE(netdev->mtu, new_mtu);
2425
2426 if (netif_running(netdev))
2427 igbvf_up(adapter);
2428 else
2429 igbvf_reset(adapter);
2430
2431 clear_bit(__IGBVF_RESETTING, &adapter->state);
2432
2433 return 0;
2434}
2435
2436static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2437{
2438 switch (cmd) {
2439 default:
2440 return -EOPNOTSUPP;
2441 }
2442}
2443
2444static int igbvf_suspend(struct device *dev_d)
2445{
2446 struct net_device *netdev = dev_get_drvdata(dev_d);
2447 struct igbvf_adapter *adapter = netdev_priv(netdev);
2448
2449 netif_device_detach(netdev);
2450
2451 if (netif_running(netdev)) {
2452 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2453 igbvf_down(adapter);
2454 igbvf_free_irq(adapter);
2455 }
2456
2457 return 0;
2458}
2459
2460static int igbvf_resume(struct device *dev_d)
2461{
2462 struct pci_dev *pdev = to_pci_dev(dev_d);
2463 struct net_device *netdev = pci_get_drvdata(pdev);
2464 struct igbvf_adapter *adapter = netdev_priv(netdev);
2465 u32 err;
2466
2467 pci_set_master(pdev);
2468
2469 if (netif_running(netdev)) {
2470 err = igbvf_request_irq(adapter);
2471 if (err)
2472 return err;
2473 }
2474
2475 igbvf_reset(adapter);
2476
2477 if (netif_running(netdev))
2478 igbvf_up(adapter);
2479
2480 netif_device_attach(netdev);
2481
2482 return 0;
2483}
2484
2485static void igbvf_shutdown(struct pci_dev *pdev)
2486{
2487 igbvf_suspend(&pdev->dev);
2488}
2489
2490#ifdef CONFIG_NET_POLL_CONTROLLER
2491/* Polling 'interrupt' - used by things like netconsole to send skbs
2492 * without having to re-enable interrupts. It's not called while
2493 * the interrupt routine is executing.
2494 */
2495static void igbvf_netpoll(struct net_device *netdev)
2496{
2497 struct igbvf_adapter *adapter = netdev_priv(netdev);
2498
2499 disable_irq(adapter->pdev->irq);
2500
2501 igbvf_clean_tx_irq(adapter->tx_ring);
2502
2503 enable_irq(adapter->pdev->irq);
2504}
2505#endif
2506
2507/**
2508 * igbvf_io_error_detected - called when PCI error is detected
2509 * @pdev: Pointer to PCI device
2510 * @state: The current pci connection state
2511 *
2512 * This function is called after a PCI bus error affecting
2513 * this device has been detected.
2514 */
2515static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2516 pci_channel_state_t state)
2517{
2518 struct net_device *netdev = pci_get_drvdata(pdev);
2519 struct igbvf_adapter *adapter = netdev_priv(netdev);
2520
2521 netif_device_detach(netdev);
2522
2523 if (state == pci_channel_io_perm_failure)
2524 return PCI_ERS_RESULT_DISCONNECT;
2525
2526 if (netif_running(netdev))
2527 igbvf_down(adapter);
2528 pci_disable_device(pdev);
2529
2530 /* Request a slot reset. */
2531 return PCI_ERS_RESULT_NEED_RESET;
2532}
2533
2534/**
2535 * igbvf_io_slot_reset - called after the pci bus has been reset.
2536 * @pdev: Pointer to PCI device
2537 *
2538 * Restart the card from scratch, as if from a cold-boot. Implementation
2539 * resembles the first-half of the igbvf_resume routine.
2540 */
2541static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2542{
2543 struct net_device *netdev = pci_get_drvdata(pdev);
2544 struct igbvf_adapter *adapter = netdev_priv(netdev);
2545
2546 if (pci_enable_device_mem(pdev)) {
2547 dev_err(&pdev->dev,
2548 "Cannot re-enable PCI device after reset.\n");
2549 return PCI_ERS_RESULT_DISCONNECT;
2550 }
2551 pci_set_master(pdev);
2552
2553 igbvf_reset(adapter);
2554
2555 return PCI_ERS_RESULT_RECOVERED;
2556}
2557
2558/**
2559 * igbvf_io_resume - called when traffic can start flowing again.
2560 * @pdev: Pointer to PCI device
2561 *
2562 * This callback is called when the error recovery driver tells us that
2563 * its OK to resume normal operation. Implementation resembles the
2564 * second-half of the igbvf_resume routine.
2565 */
2566static void igbvf_io_resume(struct pci_dev *pdev)
2567{
2568 struct net_device *netdev = pci_get_drvdata(pdev);
2569 struct igbvf_adapter *adapter = netdev_priv(netdev);
2570
2571 if (netif_running(netdev)) {
2572 if (igbvf_up(adapter)) {
2573 dev_err(&pdev->dev,
2574 "can't bring device back up after reset\n");
2575 return;
2576 }
2577 }
2578
2579 netif_device_attach(netdev);
2580}
2581
2582/**
2583 * igbvf_io_prepare - prepare device driver for PCI reset
2584 * @pdev: PCI device information struct
2585 */
2586static void igbvf_io_prepare(struct pci_dev *pdev)
2587{
2588 struct net_device *netdev = pci_get_drvdata(pdev);
2589 struct igbvf_adapter *adapter = netdev_priv(netdev);
2590
2591 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2592 usleep_range(1000, 2000);
2593 igbvf_down(adapter);
2594}
2595
2596/**
2597 * igbvf_io_reset_done - PCI reset done, device driver reset can begin
2598 * @pdev: PCI device information struct
2599 */
2600static void igbvf_io_reset_done(struct pci_dev *pdev)
2601{
2602 struct net_device *netdev = pci_get_drvdata(pdev);
2603 struct igbvf_adapter *adapter = netdev_priv(netdev);
2604
2605 igbvf_up(adapter);
2606 clear_bit(__IGBVF_RESETTING, &adapter->state);
2607}
2608
2609static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2610{
2611 struct e1000_hw *hw = &adapter->hw;
2612 struct net_device *netdev = adapter->netdev;
2613 struct pci_dev *pdev = adapter->pdev;
2614
2615 if (hw->mac.type == e1000_vfadapt_i350)
2616 dev_info(&pdev->dev, "Intel(R) I350 Virtual Function\n");
2617 else
2618 dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2619 dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2620}
2621
2622static int igbvf_set_features(struct net_device *netdev,
2623 netdev_features_t features)
2624{
2625 struct igbvf_adapter *adapter = netdev_priv(netdev);
2626
2627 if (features & NETIF_F_RXCSUM)
2628 adapter->flags &= ~IGBVF_FLAG_RX_CSUM_DISABLED;
2629 else
2630 adapter->flags |= IGBVF_FLAG_RX_CSUM_DISABLED;
2631
2632 return 0;
2633}
2634
2635#define IGBVF_MAX_MAC_HDR_LEN 127
2636#define IGBVF_MAX_NETWORK_HDR_LEN 511
2637
2638static netdev_features_t
2639igbvf_features_check(struct sk_buff *skb, struct net_device *dev,
2640 netdev_features_t features)
2641{
2642 unsigned int network_hdr_len, mac_hdr_len;
2643
2644 /* Make certain the headers can be described by a context descriptor */
2645 mac_hdr_len = skb_network_offset(skb);
2646 if (unlikely(mac_hdr_len > IGBVF_MAX_MAC_HDR_LEN))
2647 return features & ~(NETIF_F_HW_CSUM |
2648 NETIF_F_SCTP_CRC |
2649 NETIF_F_HW_VLAN_CTAG_TX |
2650 NETIF_F_TSO |
2651 NETIF_F_TSO6);
2652
2653 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
2654 if (unlikely(network_hdr_len > IGBVF_MAX_NETWORK_HDR_LEN))
2655 return features & ~(NETIF_F_HW_CSUM |
2656 NETIF_F_SCTP_CRC |
2657 NETIF_F_TSO |
2658 NETIF_F_TSO6);
2659
2660 /* We can only support IPV4 TSO in tunnels if we can mangle the
2661 * inner IP ID field, so strip TSO if MANGLEID is not supported.
2662 */
2663 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
2664 features &= ~NETIF_F_TSO;
2665
2666 return features;
2667}
2668
2669static const struct net_device_ops igbvf_netdev_ops = {
2670 .ndo_open = igbvf_open,
2671 .ndo_stop = igbvf_close,
2672 .ndo_start_xmit = igbvf_xmit_frame,
2673 .ndo_set_rx_mode = igbvf_set_rx_mode,
2674 .ndo_set_mac_address = igbvf_set_mac,
2675 .ndo_change_mtu = igbvf_change_mtu,
2676 .ndo_eth_ioctl = igbvf_ioctl,
2677 .ndo_tx_timeout = igbvf_tx_timeout,
2678 .ndo_vlan_rx_add_vid = igbvf_vlan_rx_add_vid,
2679 .ndo_vlan_rx_kill_vid = igbvf_vlan_rx_kill_vid,
2680#ifdef CONFIG_NET_POLL_CONTROLLER
2681 .ndo_poll_controller = igbvf_netpoll,
2682#endif
2683 .ndo_set_features = igbvf_set_features,
2684 .ndo_features_check = igbvf_features_check,
2685};
2686
2687/**
2688 * igbvf_probe - Device Initialization Routine
2689 * @pdev: PCI device information struct
2690 * @ent: entry in igbvf_pci_tbl
2691 *
2692 * Returns 0 on success, negative on failure
2693 *
2694 * igbvf_probe initializes an adapter identified by a pci_dev structure.
2695 * The OS initialization, configuring of the adapter private structure,
2696 * and a hardware reset occur.
2697 **/
2698static int igbvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2699{
2700 struct net_device *netdev;
2701 struct igbvf_adapter *adapter;
2702 struct e1000_hw *hw;
2703 const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2704 int err;
2705
2706 err = pci_enable_device_mem(pdev);
2707 if (err)
2708 return err;
2709
2710 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
2711 if (err) {
2712 dev_err(&pdev->dev,
2713 "No usable DMA configuration, aborting\n");
2714 goto err_dma;
2715 }
2716
2717 err = pci_request_regions(pdev, igbvf_driver_name);
2718 if (err)
2719 goto err_pci_reg;
2720
2721 pci_set_master(pdev);
2722
2723 err = -ENOMEM;
2724 netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2725 if (!netdev)
2726 goto err_alloc_etherdev;
2727
2728 SET_NETDEV_DEV(netdev, &pdev->dev);
2729
2730 pci_set_drvdata(pdev, netdev);
2731 adapter = netdev_priv(netdev);
2732 hw = &adapter->hw;
2733 adapter->netdev = netdev;
2734 adapter->pdev = pdev;
2735 adapter->ei = ei;
2736 adapter->pba = ei->pba;
2737 adapter->flags = ei->flags;
2738 adapter->hw.back = adapter;
2739 adapter->hw.mac.type = ei->mac;
2740 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2741
2742 /* PCI config space info */
2743
2744 hw->vendor_id = pdev->vendor;
2745 hw->device_id = pdev->device;
2746 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2747 hw->subsystem_device_id = pdev->subsystem_device;
2748 hw->revision_id = pdev->revision;
2749
2750 err = -EIO;
2751 adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2752 pci_resource_len(pdev, 0));
2753
2754 if (!adapter->hw.hw_addr)
2755 goto err_ioremap;
2756
2757 if (ei->get_variants) {
2758 err = ei->get_variants(adapter);
2759 if (err)
2760 goto err_get_variants;
2761 }
2762
2763 /* setup adapter struct */
2764 err = igbvf_sw_init(adapter);
2765 if (err)
2766 goto err_sw_init;
2767
2768 /* construct the net_device struct */
2769 netdev->netdev_ops = &igbvf_netdev_ops;
2770
2771 igbvf_set_ethtool_ops(netdev);
2772 netdev->watchdog_timeo = 5 * HZ;
2773 strscpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
2774
2775 netdev->hw_features = NETIF_F_SG |
2776 NETIF_F_TSO |
2777 NETIF_F_TSO6 |
2778 NETIF_F_RXCSUM |
2779 NETIF_F_HW_CSUM |
2780 NETIF_F_SCTP_CRC;
2781
2782#define IGBVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
2783 NETIF_F_GSO_GRE_CSUM | \
2784 NETIF_F_GSO_IPXIP4 | \
2785 NETIF_F_GSO_IPXIP6 | \
2786 NETIF_F_GSO_UDP_TUNNEL | \
2787 NETIF_F_GSO_UDP_TUNNEL_CSUM)
2788
2789 netdev->gso_partial_features = IGBVF_GSO_PARTIAL_FEATURES;
2790 netdev->hw_features |= NETIF_F_GSO_PARTIAL |
2791 IGBVF_GSO_PARTIAL_FEATURES;
2792
2793 netdev->features = netdev->hw_features | NETIF_F_HIGHDMA;
2794
2795 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
2796 netdev->mpls_features |= NETIF_F_HW_CSUM;
2797 netdev->hw_enc_features |= netdev->vlan_features;
2798
2799 /* set this bit last since it cannot be part of vlan_features */
2800 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
2801 NETIF_F_HW_VLAN_CTAG_RX |
2802 NETIF_F_HW_VLAN_CTAG_TX;
2803
2804 /* MTU range: 68 - 9216 */
2805 netdev->min_mtu = ETH_MIN_MTU;
2806 netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
2807
2808 spin_lock_bh(&hw->mbx_lock);
2809
2810 /*reset the controller to put the device in a known good state */
2811 err = hw->mac.ops.reset_hw(hw);
2812 if (err) {
2813 dev_info(&pdev->dev,
2814 "PF still in reset state. Is the PF interface up?\n");
2815 } else {
2816 err = hw->mac.ops.read_mac_addr(hw);
2817 if (err)
2818 dev_info(&pdev->dev, "Error reading MAC address.\n");
2819 else if (is_zero_ether_addr(adapter->hw.mac.addr))
2820 dev_info(&pdev->dev,
2821 "MAC address not assigned by administrator.\n");
2822 eth_hw_addr_set(netdev, adapter->hw.mac.addr);
2823 }
2824
2825 spin_unlock_bh(&hw->mbx_lock);
2826
2827 if (!is_valid_ether_addr(netdev->dev_addr)) {
2828 dev_info(&pdev->dev, "Assigning random MAC address.\n");
2829 eth_hw_addr_random(netdev);
2830 memcpy(adapter->hw.mac.addr, netdev->dev_addr,
2831 netdev->addr_len);
2832 }
2833
2834 timer_setup(&adapter->watchdog_timer, igbvf_watchdog, 0);
2835
2836 INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2837 INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2838
2839 /* ring size defaults */
2840 adapter->rx_ring->count = 1024;
2841 adapter->tx_ring->count = 1024;
2842
2843 /* reset the hardware with the new settings */
2844 igbvf_reset(adapter);
2845
2846 /* set hardware-specific flags */
2847 if (adapter->hw.mac.type == e1000_vfadapt_i350)
2848 adapter->flags |= IGBVF_FLAG_RX_LB_VLAN_BSWAP;
2849
2850 strcpy(netdev->name, "eth%d");
2851 err = register_netdev(netdev);
2852 if (err)
2853 goto err_hw_init;
2854
2855 /* tell the stack to leave us alone until igbvf_open() is called */
2856 netif_carrier_off(netdev);
2857 netif_stop_queue(netdev);
2858
2859 igbvf_print_device_info(adapter);
2860
2861 igbvf_initialize_last_counter_stats(adapter);
2862
2863 return 0;
2864
2865err_hw_init:
2866 netif_napi_del(&adapter->rx_ring->napi);
2867 kfree(adapter->tx_ring);
2868 kfree(adapter->rx_ring);
2869err_sw_init:
2870 igbvf_reset_interrupt_capability(adapter);
2871err_get_variants:
2872 iounmap(adapter->hw.hw_addr);
2873err_ioremap:
2874 free_netdev(netdev);
2875err_alloc_etherdev:
2876 pci_release_regions(pdev);
2877err_pci_reg:
2878err_dma:
2879 pci_disable_device(pdev);
2880 return err;
2881}
2882
2883/**
2884 * igbvf_remove - Device Removal Routine
2885 * @pdev: PCI device information struct
2886 *
2887 * igbvf_remove is called by the PCI subsystem to alert the driver
2888 * that it should release a PCI device. The could be caused by a
2889 * Hot-Plug event, or because the driver is going to be removed from
2890 * memory.
2891 **/
2892static void igbvf_remove(struct pci_dev *pdev)
2893{
2894 struct net_device *netdev = pci_get_drvdata(pdev);
2895 struct igbvf_adapter *adapter = netdev_priv(netdev);
2896 struct e1000_hw *hw = &adapter->hw;
2897
2898 /* The watchdog timer may be rescheduled, so explicitly
2899 * disable it from being rescheduled.
2900 */
2901 set_bit(__IGBVF_DOWN, &adapter->state);
2902 timer_delete_sync(&adapter->watchdog_timer);
2903
2904 cancel_work_sync(&adapter->reset_task);
2905 cancel_work_sync(&adapter->watchdog_task);
2906
2907 unregister_netdev(netdev);
2908
2909 igbvf_reset_interrupt_capability(adapter);
2910
2911 /* it is important to delete the NAPI struct prior to freeing the
2912 * Rx ring so that you do not end up with null pointer refs
2913 */
2914 netif_napi_del(&adapter->rx_ring->napi);
2915 kfree(adapter->tx_ring);
2916 kfree(adapter->rx_ring);
2917
2918 iounmap(hw->hw_addr);
2919 if (hw->flash_address)
2920 iounmap(hw->flash_address);
2921 pci_release_regions(pdev);
2922
2923 free_netdev(netdev);
2924
2925 pci_disable_device(pdev);
2926}
2927
2928/* PCI Error Recovery (ERS) */
2929static const struct pci_error_handlers igbvf_err_handler = {
2930 .error_detected = igbvf_io_error_detected,
2931 .slot_reset = igbvf_io_slot_reset,
2932 .resume = igbvf_io_resume,
2933 .reset_prepare = igbvf_io_prepare,
2934 .reset_done = igbvf_io_reset_done,
2935};
2936
2937static const struct pci_device_id igbvf_pci_tbl[] = {
2938 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), .driver_data = board_vf },
2939 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), .driver_data = board_i350_vf },
2940 { } /* terminate list */
2941};
2942MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2943
2944static DEFINE_SIMPLE_DEV_PM_OPS(igbvf_pm_ops, igbvf_suspend, igbvf_resume);
2945
2946/* PCI Device API Driver */
2947static struct pci_driver igbvf_driver = {
2948 .name = igbvf_driver_name,
2949 .id_table = igbvf_pci_tbl,
2950 .probe = igbvf_probe,
2951 .remove = igbvf_remove,
2952 .driver.pm = pm_sleep_ptr(&igbvf_pm_ops),
2953 .shutdown = igbvf_shutdown,
2954 .err_handler = &igbvf_err_handler
2955};
2956
2957/**
2958 * igbvf_init_module - Driver Registration Routine
2959 *
2960 * igbvf_init_module is the first routine called when the driver is
2961 * loaded. All it does is register with the PCI subsystem.
2962 **/
2963static int __init igbvf_init_module(void)
2964{
2965 int ret;
2966
2967 pr_info("%s\n", igbvf_driver_string);
2968 pr_info("%s\n", igbvf_copyright);
2969
2970 ret = pci_register_driver(&igbvf_driver);
2971
2972 return ret;
2973}
2974module_init(igbvf_init_module);
2975
2976/**
2977 * igbvf_exit_module - Driver Exit Cleanup Routine
2978 *
2979 * igbvf_exit_module is called just before the driver is removed
2980 * from memory.
2981 **/
2982static void __exit igbvf_exit_module(void)
2983{
2984 pci_unregister_driver(&igbvf_driver);
2985}
2986module_exit(igbvf_exit_module);
2987
2988MODULE_DESCRIPTION("Intel(R) Gigabit Virtual Function Network Driver");
2989MODULE_LICENSE("GPL v2");
2990
2991/* netdev.c */