/drivers/net/igbvf/netdev.c
C | 2873 lines | 1898 code | 491 blank | 484 comment | 271 complexity | 8872c0746cce4e1c1ffb8dfb42e44d8f MD5 | raw file
Possible License(s): GPL-2.0, LGPL-2.0, AGPL-1.0
Large files files are truncated, but you can click here to view the full file
1/*******************************************************************************
2
3 Intel(R) 82576 Virtual Function Linux driver
4 Copyright(c) 2009 - 2010 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, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
21
22 Contact Information:
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25
26*******************************************************************************/
27
28#include <linux/module.h>
29#include <linux/types.h>
30#include <linux/init.h>
31#include <linux/pci.h>
32#include <linux/vmalloc.h>
33#include <linux/pagemap.h>
34#include <linux/delay.h>
35#include <linux/netdevice.h>
36#include <linux/tcp.h>
37#include <linux/ipv6.h>
38#include <linux/slab.h>
39#include <net/checksum.h>
40#include <net/ip6_checksum.h>
41#include <linux/mii.h>
42#include <linux/ethtool.h>
43#include <linux/if_vlan.h>
44#include <linux/prefetch.h>
45
46#include "igbvf.h"
47
48#define DRV_VERSION "1.0.8-k0"
49char igbvf_driver_name[] = "igbvf";
50const char igbvf_driver_version[] = DRV_VERSION;
51static const char igbvf_driver_string[] =
52 "Intel(R) Virtual Function Network Driver";
53static const char igbvf_copyright[] =
54 "Copyright (c) 2009 - 2010 Intel Corporation.";
55
56static int igbvf_poll(struct napi_struct *napi, int budget);
57static void igbvf_reset(struct igbvf_adapter *);
58static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
59static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
60
61static struct igbvf_info igbvf_vf_info = {
62 .mac = e1000_vfadapt,
63 .flags = 0,
64 .pba = 10,
65 .init_ops = e1000_init_function_pointers_vf,
66};
67
68static struct igbvf_info igbvf_i350_vf_info = {
69 .mac = e1000_vfadapt_i350,
70 .flags = 0,
71 .pba = 10,
72 .init_ops = e1000_init_function_pointers_vf,
73};
74
75static const struct igbvf_info *igbvf_info_tbl[] = {
76 [board_vf] = &igbvf_vf_info,
77 [board_i350_vf] = &igbvf_i350_vf_info,
78};
79
80/**
81 * igbvf_desc_unused - calculate if we have unused descriptors
82 **/
83static int igbvf_desc_unused(struct igbvf_ring *ring)
84{
85 if (ring->next_to_clean > ring->next_to_use)
86 return ring->next_to_clean - ring->next_to_use - 1;
87
88 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
89}
90
91/**
92 * igbvf_receive_skb - helper function to handle Rx indications
93 * @adapter: board private structure
94 * @status: descriptor status field as written by hardware
95 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
96 * @skb: pointer to sk_buff to be indicated to stack
97 **/
98static void igbvf_receive_skb(struct igbvf_adapter *adapter,
99 struct net_device *netdev,
100 struct sk_buff *skb,
101 u32 status, u16 vlan)
102{
103 if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
104 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
105 le16_to_cpu(vlan) &
106 E1000_RXD_SPC_VLAN_MASK);
107 else
108 netif_receive_skb(skb);
109}
110
111static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
112 u32 status_err, struct sk_buff *skb)
113{
114 skb_checksum_none_assert(skb);
115
116 /* Ignore Checksum bit is set or checksum is disabled through ethtool */
117 if ((status_err & E1000_RXD_STAT_IXSM) ||
118 (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
119 return;
120
121 /* TCP/UDP checksum error bit is set */
122 if (status_err &
123 (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
124 /* let the stack verify checksum errors */
125 adapter->hw_csum_err++;
126 return;
127 }
128
129 /* It must be a TCP or UDP packet with a valid checksum */
130 if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
131 skb->ip_summed = CHECKSUM_UNNECESSARY;
132
133 adapter->hw_csum_good++;
134}
135
136/**
137 * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
138 * @rx_ring: address of ring structure to repopulate
139 * @cleaned_count: number of buffers to repopulate
140 **/
141static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
142 int cleaned_count)
143{
144 struct igbvf_adapter *adapter = rx_ring->adapter;
145 struct net_device *netdev = adapter->netdev;
146 struct pci_dev *pdev = adapter->pdev;
147 union e1000_adv_rx_desc *rx_desc;
148 struct igbvf_buffer *buffer_info;
149 struct sk_buff *skb;
150 unsigned int i;
151 int bufsz;
152
153 i = rx_ring->next_to_use;
154 buffer_info = &rx_ring->buffer_info[i];
155
156 if (adapter->rx_ps_hdr_size)
157 bufsz = adapter->rx_ps_hdr_size;
158 else
159 bufsz = adapter->rx_buffer_len;
160
161 while (cleaned_count--) {
162 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
163
164 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
165 if (!buffer_info->page) {
166 buffer_info->page = alloc_page(GFP_ATOMIC);
167 if (!buffer_info->page) {
168 adapter->alloc_rx_buff_failed++;
169 goto no_buffers;
170 }
171 buffer_info->page_offset = 0;
172 } else {
173 buffer_info->page_offset ^= PAGE_SIZE / 2;
174 }
175 buffer_info->page_dma =
176 dma_map_page(&pdev->dev, buffer_info->page,
177 buffer_info->page_offset,
178 PAGE_SIZE / 2,
179 DMA_FROM_DEVICE);
180 }
181
182 if (!buffer_info->skb) {
183 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
184 if (!skb) {
185 adapter->alloc_rx_buff_failed++;
186 goto no_buffers;
187 }
188
189 buffer_info->skb = skb;
190 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
191 bufsz,
192 DMA_FROM_DEVICE);
193 }
194 /* Refresh the desc even if buffer_addrs didn't change because
195 * each write-back erases this info. */
196 if (adapter->rx_ps_hdr_size) {
197 rx_desc->read.pkt_addr =
198 cpu_to_le64(buffer_info->page_dma);
199 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
200 } else {
201 rx_desc->read.pkt_addr =
202 cpu_to_le64(buffer_info->dma);
203 rx_desc->read.hdr_addr = 0;
204 }
205
206 i++;
207 if (i == rx_ring->count)
208 i = 0;
209 buffer_info = &rx_ring->buffer_info[i];
210 }
211
212no_buffers:
213 if (rx_ring->next_to_use != i) {
214 rx_ring->next_to_use = i;
215 if (i == 0)
216 i = (rx_ring->count - 1);
217 else
218 i--;
219
220 /* Force memory writes to complete before letting h/w
221 * know there are new descriptors to fetch. (Only
222 * applicable for weak-ordered memory model archs,
223 * such as IA-64). */
224 wmb();
225 writel(i, adapter->hw.hw_addr + rx_ring->tail);
226 }
227}
228
229/**
230 * igbvf_clean_rx_irq - Send received data up the network stack; legacy
231 * @adapter: board private structure
232 *
233 * the return value indicates whether actual cleaning was done, there
234 * is no guarantee that everything was cleaned
235 **/
236static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
237 int *work_done, int work_to_do)
238{
239 struct igbvf_ring *rx_ring = adapter->rx_ring;
240 struct net_device *netdev = adapter->netdev;
241 struct pci_dev *pdev = adapter->pdev;
242 union e1000_adv_rx_desc *rx_desc, *next_rxd;
243 struct igbvf_buffer *buffer_info, *next_buffer;
244 struct sk_buff *skb;
245 bool cleaned = false;
246 int cleaned_count = 0;
247 unsigned int total_bytes = 0, total_packets = 0;
248 unsigned int i;
249 u32 length, hlen, staterr;
250
251 i = rx_ring->next_to_clean;
252 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
253 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
254
255 while (staterr & E1000_RXD_STAT_DD) {
256 if (*work_done >= work_to_do)
257 break;
258 (*work_done)++;
259 rmb(); /* read descriptor and rx_buffer_info after status DD */
260
261 buffer_info = &rx_ring->buffer_info[i];
262
263 /* HW will not DMA in data larger than the given buffer, even
264 * if it parses the (NFS, of course) header to be larger. In
265 * that case, it fills the header buffer and spills the rest
266 * into the page.
267 */
268 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
269 E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
270 if (hlen > adapter->rx_ps_hdr_size)
271 hlen = adapter->rx_ps_hdr_size;
272
273 length = le16_to_cpu(rx_desc->wb.upper.length);
274 cleaned = true;
275 cleaned_count++;
276
277 skb = buffer_info->skb;
278 prefetch(skb->data - NET_IP_ALIGN);
279 buffer_info->skb = NULL;
280 if (!adapter->rx_ps_hdr_size) {
281 dma_unmap_single(&pdev->dev, buffer_info->dma,
282 adapter->rx_buffer_len,
283 DMA_FROM_DEVICE);
284 buffer_info->dma = 0;
285 skb_put(skb, length);
286 goto send_up;
287 }
288
289 if (!skb_shinfo(skb)->nr_frags) {
290 dma_unmap_single(&pdev->dev, buffer_info->dma,
291 adapter->rx_ps_hdr_size,
292 DMA_FROM_DEVICE);
293 skb_put(skb, hlen);
294 }
295
296 if (length) {
297 dma_unmap_page(&pdev->dev, buffer_info->page_dma,
298 PAGE_SIZE / 2,
299 DMA_FROM_DEVICE);
300 buffer_info->page_dma = 0;
301
302 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
303 buffer_info->page,
304 buffer_info->page_offset,
305 length);
306
307 if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
308 (page_count(buffer_info->page) != 1))
309 buffer_info->page = NULL;
310 else
311 get_page(buffer_info->page);
312
313 skb->len += length;
314 skb->data_len += length;
315 skb->truesize += length;
316 }
317send_up:
318 i++;
319 if (i == rx_ring->count)
320 i = 0;
321 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
322 prefetch(next_rxd);
323 next_buffer = &rx_ring->buffer_info[i];
324
325 if (!(staterr & E1000_RXD_STAT_EOP)) {
326 buffer_info->skb = next_buffer->skb;
327 buffer_info->dma = next_buffer->dma;
328 next_buffer->skb = skb;
329 next_buffer->dma = 0;
330 goto next_desc;
331 }
332
333 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
334 dev_kfree_skb_irq(skb);
335 goto next_desc;
336 }
337
338 total_bytes += skb->len;
339 total_packets++;
340
341 igbvf_rx_checksum_adv(adapter, staterr, skb);
342
343 skb->protocol = eth_type_trans(skb, netdev);
344
345 igbvf_receive_skb(adapter, netdev, skb, staterr,
346 rx_desc->wb.upper.vlan);
347
348next_desc:
349 rx_desc->wb.upper.status_error = 0;
350
351 /* return some buffers to hardware, one at a time is too slow */
352 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
353 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
354 cleaned_count = 0;
355 }
356
357 /* use prefetched values */
358 rx_desc = next_rxd;
359 buffer_info = next_buffer;
360
361 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
362 }
363
364 rx_ring->next_to_clean = i;
365 cleaned_count = igbvf_desc_unused(rx_ring);
366
367 if (cleaned_count)
368 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
369
370 adapter->total_rx_packets += total_packets;
371 adapter->total_rx_bytes += total_bytes;
372 adapter->net_stats.rx_bytes += total_bytes;
373 adapter->net_stats.rx_packets += total_packets;
374 return cleaned;
375}
376
377static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
378 struct igbvf_buffer *buffer_info)
379{
380 if (buffer_info->dma) {
381 if (buffer_info->mapped_as_page)
382 dma_unmap_page(&adapter->pdev->dev,
383 buffer_info->dma,
384 buffer_info->length,
385 DMA_TO_DEVICE);
386 else
387 dma_unmap_single(&adapter->pdev->dev,
388 buffer_info->dma,
389 buffer_info->length,
390 DMA_TO_DEVICE);
391 buffer_info->dma = 0;
392 }
393 if (buffer_info->skb) {
394 dev_kfree_skb_any(buffer_info->skb);
395 buffer_info->skb = NULL;
396 }
397 buffer_info->time_stamp = 0;
398}
399
400/**
401 * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
402 * @adapter: board private structure
403 *
404 * Return 0 on success, negative on failure
405 **/
406int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
407 struct igbvf_ring *tx_ring)
408{
409 struct pci_dev *pdev = adapter->pdev;
410 int size;
411
412 size = sizeof(struct igbvf_buffer) * tx_ring->count;
413 tx_ring->buffer_info = vzalloc(size);
414 if (!tx_ring->buffer_info)
415 goto err;
416
417 /* round up to nearest 4K */
418 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
419 tx_ring->size = ALIGN(tx_ring->size, 4096);
420
421 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
422 &tx_ring->dma, GFP_KERNEL);
423
424 if (!tx_ring->desc)
425 goto err;
426
427 tx_ring->adapter = adapter;
428 tx_ring->next_to_use = 0;
429 tx_ring->next_to_clean = 0;
430
431 return 0;
432err:
433 vfree(tx_ring->buffer_info);
434 dev_err(&adapter->pdev->dev,
435 "Unable to allocate memory for the transmit descriptor ring\n");
436 return -ENOMEM;
437}
438
439/**
440 * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
441 * @adapter: board private structure
442 *
443 * Returns 0 on success, negative on failure
444 **/
445int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
446 struct igbvf_ring *rx_ring)
447{
448 struct pci_dev *pdev = adapter->pdev;
449 int size, desc_len;
450
451 size = sizeof(struct igbvf_buffer) * rx_ring->count;
452 rx_ring->buffer_info = vzalloc(size);
453 if (!rx_ring->buffer_info)
454 goto err;
455
456 desc_len = sizeof(union e1000_adv_rx_desc);
457
458 /* Round up to nearest 4K */
459 rx_ring->size = rx_ring->count * desc_len;
460 rx_ring->size = ALIGN(rx_ring->size, 4096);
461
462 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
463 &rx_ring->dma, GFP_KERNEL);
464
465 if (!rx_ring->desc)
466 goto err;
467
468 rx_ring->next_to_clean = 0;
469 rx_ring->next_to_use = 0;
470
471 rx_ring->adapter = adapter;
472
473 return 0;
474
475err:
476 vfree(rx_ring->buffer_info);
477 rx_ring->buffer_info = NULL;
478 dev_err(&adapter->pdev->dev,
479 "Unable to allocate memory for the receive descriptor ring\n");
480 return -ENOMEM;
481}
482
483/**
484 * igbvf_clean_tx_ring - Free Tx Buffers
485 * @tx_ring: ring to be cleaned
486 **/
487static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
488{
489 struct igbvf_adapter *adapter = tx_ring->adapter;
490 struct igbvf_buffer *buffer_info;
491 unsigned long size;
492 unsigned int i;
493
494 if (!tx_ring->buffer_info)
495 return;
496
497 /* Free all the Tx ring sk_buffs */
498 for (i = 0; i < tx_ring->count; i++) {
499 buffer_info = &tx_ring->buffer_info[i];
500 igbvf_put_txbuf(adapter, buffer_info);
501 }
502
503 size = sizeof(struct igbvf_buffer) * tx_ring->count;
504 memset(tx_ring->buffer_info, 0, size);
505
506 /* Zero out the descriptor ring */
507 memset(tx_ring->desc, 0, tx_ring->size);
508
509 tx_ring->next_to_use = 0;
510 tx_ring->next_to_clean = 0;
511
512 writel(0, adapter->hw.hw_addr + tx_ring->head);
513 writel(0, adapter->hw.hw_addr + tx_ring->tail);
514}
515
516/**
517 * igbvf_free_tx_resources - Free Tx Resources per Queue
518 * @tx_ring: ring to free resources from
519 *
520 * Free all transmit software resources
521 **/
522void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
523{
524 struct pci_dev *pdev = tx_ring->adapter->pdev;
525
526 igbvf_clean_tx_ring(tx_ring);
527
528 vfree(tx_ring->buffer_info);
529 tx_ring->buffer_info = NULL;
530
531 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
532 tx_ring->dma);
533
534 tx_ring->desc = NULL;
535}
536
537/**
538 * igbvf_clean_rx_ring - Free Rx Buffers per Queue
539 * @adapter: board private structure
540 **/
541static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
542{
543 struct igbvf_adapter *adapter = rx_ring->adapter;
544 struct igbvf_buffer *buffer_info;
545 struct pci_dev *pdev = adapter->pdev;
546 unsigned long size;
547 unsigned int i;
548
549 if (!rx_ring->buffer_info)
550 return;
551
552 /* Free all the Rx ring sk_buffs */
553 for (i = 0; i < rx_ring->count; i++) {
554 buffer_info = &rx_ring->buffer_info[i];
555 if (buffer_info->dma) {
556 if (adapter->rx_ps_hdr_size){
557 dma_unmap_single(&pdev->dev, buffer_info->dma,
558 adapter->rx_ps_hdr_size,
559 DMA_FROM_DEVICE);
560 } else {
561 dma_unmap_single(&pdev->dev, buffer_info->dma,
562 adapter->rx_buffer_len,
563 DMA_FROM_DEVICE);
564 }
565 buffer_info->dma = 0;
566 }
567
568 if (buffer_info->skb) {
569 dev_kfree_skb(buffer_info->skb);
570 buffer_info->skb = NULL;
571 }
572
573 if (buffer_info->page) {
574 if (buffer_info->page_dma)
575 dma_unmap_page(&pdev->dev,
576 buffer_info->page_dma,
577 PAGE_SIZE / 2,
578 DMA_FROM_DEVICE);
579 put_page(buffer_info->page);
580 buffer_info->page = NULL;
581 buffer_info->page_dma = 0;
582 buffer_info->page_offset = 0;
583 }
584 }
585
586 size = sizeof(struct igbvf_buffer) * rx_ring->count;
587 memset(rx_ring->buffer_info, 0, size);
588
589 /* Zero out the descriptor ring */
590 memset(rx_ring->desc, 0, rx_ring->size);
591
592 rx_ring->next_to_clean = 0;
593 rx_ring->next_to_use = 0;
594
595 writel(0, adapter->hw.hw_addr + rx_ring->head);
596 writel(0, adapter->hw.hw_addr + rx_ring->tail);
597}
598
599/**
600 * igbvf_free_rx_resources - Free Rx Resources
601 * @rx_ring: ring to clean the resources from
602 *
603 * Free all receive software resources
604 **/
605
606void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
607{
608 struct pci_dev *pdev = rx_ring->adapter->pdev;
609
610 igbvf_clean_rx_ring(rx_ring);
611
612 vfree(rx_ring->buffer_info);
613 rx_ring->buffer_info = NULL;
614
615 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
616 rx_ring->dma);
617 rx_ring->desc = NULL;
618}
619
620/**
621 * igbvf_update_itr - update the dynamic ITR value based on statistics
622 * @adapter: pointer to adapter
623 * @itr_setting: current adapter->itr
624 * @packets: the number of packets during this measurement interval
625 * @bytes: the number of bytes during this measurement interval
626 *
627 * Stores a new ITR value based on packets and byte
628 * counts during the last interrupt. The advantage of per interrupt
629 * computation is faster updates and more accurate ITR for the current
630 * traffic pattern. Constants in this function were computed
631 * based on theoretical maximum wire speed and thresholds were set based
632 * on testing data as well as attempting to minimize response time
633 * while increasing bulk throughput. This functionality is controlled
634 * by the InterruptThrottleRate module parameter.
635 **/
636static unsigned int igbvf_update_itr(struct igbvf_adapter *adapter,
637 u16 itr_setting, int packets,
638 int bytes)
639{
640 unsigned int retval = itr_setting;
641
642 if (packets == 0)
643 goto update_itr_done;
644
645 switch (itr_setting) {
646 case lowest_latency:
647 /* handle TSO and jumbo frames */
648 if (bytes/packets > 8000)
649 retval = bulk_latency;
650 else if ((packets < 5) && (bytes > 512))
651 retval = low_latency;
652 break;
653 case low_latency: /* 50 usec aka 20000 ints/s */
654 if (bytes > 10000) {
655 /* this if handles the TSO accounting */
656 if (bytes/packets > 8000)
657 retval = bulk_latency;
658 else if ((packets < 10) || ((bytes/packets) > 1200))
659 retval = bulk_latency;
660 else if ((packets > 35))
661 retval = lowest_latency;
662 } else if (bytes/packets > 2000) {
663 retval = bulk_latency;
664 } else if (packets <= 2 && bytes < 512) {
665 retval = lowest_latency;
666 }
667 break;
668 case bulk_latency: /* 250 usec aka 4000 ints/s */
669 if (bytes > 25000) {
670 if (packets > 35)
671 retval = low_latency;
672 } else if (bytes < 6000) {
673 retval = low_latency;
674 }
675 break;
676 }
677
678update_itr_done:
679 return retval;
680}
681
682static void igbvf_set_itr(struct igbvf_adapter *adapter)
683{
684 struct e1000_hw *hw = &adapter->hw;
685 u16 current_itr;
686 u32 new_itr = adapter->itr;
687
688 adapter->tx_itr = igbvf_update_itr(adapter, adapter->tx_itr,
689 adapter->total_tx_packets,
690 adapter->total_tx_bytes);
691 /* conservative mode (itr 3) eliminates the lowest_latency setting */
692 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
693 adapter->tx_itr = low_latency;
694
695 adapter->rx_itr = igbvf_update_itr(adapter, adapter->rx_itr,
696 adapter->total_rx_packets,
697 adapter->total_rx_bytes);
698 /* conservative mode (itr 3) eliminates the lowest_latency setting */
699 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
700 adapter->rx_itr = low_latency;
701
702 current_itr = max(adapter->rx_itr, adapter->tx_itr);
703
704 switch (current_itr) {
705 /* counts and packets in update_itr are dependent on these numbers */
706 case lowest_latency:
707 new_itr = 70000;
708 break;
709 case low_latency:
710 new_itr = 20000; /* aka hwitr = ~200 */
711 break;
712 case bulk_latency:
713 new_itr = 4000;
714 break;
715 default:
716 break;
717 }
718
719 if (new_itr != adapter->itr) {
720 /*
721 * this attempts to bias the interrupt rate towards Bulk
722 * by adding intermediate steps when interrupt rate is
723 * increasing
724 */
725 new_itr = new_itr > adapter->itr ?
726 min(adapter->itr + (new_itr >> 2), new_itr) :
727 new_itr;
728 adapter->itr = new_itr;
729 adapter->rx_ring->itr_val = 1952;
730
731 if (adapter->msix_entries)
732 adapter->rx_ring->set_itr = 1;
733 else
734 ew32(ITR, 1952);
735 }
736}
737
738/**
739 * igbvf_clean_tx_irq - Reclaim resources after transmit completes
740 * @adapter: board private structure
741 * returns true if ring is completely cleaned
742 **/
743static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
744{
745 struct igbvf_adapter *adapter = tx_ring->adapter;
746 struct net_device *netdev = adapter->netdev;
747 struct igbvf_buffer *buffer_info;
748 struct sk_buff *skb;
749 union e1000_adv_tx_desc *tx_desc, *eop_desc;
750 unsigned int total_bytes = 0, total_packets = 0;
751 unsigned int i, eop, count = 0;
752 bool cleaned = false;
753
754 i = tx_ring->next_to_clean;
755 eop = tx_ring->buffer_info[i].next_to_watch;
756 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
757
758 while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
759 (count < tx_ring->count)) {
760 rmb(); /* read buffer_info after eop_desc status */
761 for (cleaned = false; !cleaned; count++) {
762 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
763 buffer_info = &tx_ring->buffer_info[i];
764 cleaned = (i == eop);
765 skb = buffer_info->skb;
766
767 if (skb) {
768 unsigned int segs, bytecount;
769
770 /* gso_segs is currently only valid for tcp */
771 segs = skb_shinfo(skb)->gso_segs ?: 1;
772 /* multiply data chunks by size of headers */
773 bytecount = ((segs - 1) * skb_headlen(skb)) +
774 skb->len;
775 total_packets += segs;
776 total_bytes += bytecount;
777 }
778
779 igbvf_put_txbuf(adapter, buffer_info);
780 tx_desc->wb.status = 0;
781
782 i++;
783 if (i == tx_ring->count)
784 i = 0;
785 }
786 eop = tx_ring->buffer_info[i].next_to_watch;
787 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
788 }
789
790 tx_ring->next_to_clean = i;
791
792 if (unlikely(count &&
793 netif_carrier_ok(netdev) &&
794 igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
795 /* Make sure that anybody stopping the queue after this
796 * sees the new next_to_clean.
797 */
798 smp_mb();
799 if (netif_queue_stopped(netdev) &&
800 !(test_bit(__IGBVF_DOWN, &adapter->state))) {
801 netif_wake_queue(netdev);
802 ++adapter->restart_queue;
803 }
804 }
805
806 adapter->net_stats.tx_bytes += total_bytes;
807 adapter->net_stats.tx_packets += total_packets;
808 return count < tx_ring->count;
809}
810
811static irqreturn_t igbvf_msix_other(int irq, void *data)
812{
813 struct net_device *netdev = data;
814 struct igbvf_adapter *adapter = netdev_priv(netdev);
815 struct e1000_hw *hw = &adapter->hw;
816
817 adapter->int_counter1++;
818
819 netif_carrier_off(netdev);
820 hw->mac.get_link_status = 1;
821 if (!test_bit(__IGBVF_DOWN, &adapter->state))
822 mod_timer(&adapter->watchdog_timer, jiffies + 1);
823
824 ew32(EIMS, adapter->eims_other);
825
826 return IRQ_HANDLED;
827}
828
829static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
830{
831 struct net_device *netdev = data;
832 struct igbvf_adapter *adapter = netdev_priv(netdev);
833 struct e1000_hw *hw = &adapter->hw;
834 struct igbvf_ring *tx_ring = adapter->tx_ring;
835
836
837 adapter->total_tx_bytes = 0;
838 adapter->total_tx_packets = 0;
839
840 /* auto mask will automatically reenable the interrupt when we write
841 * EICS */
842 if (!igbvf_clean_tx_irq(tx_ring))
843 /* Ring was not completely cleaned, so fire another interrupt */
844 ew32(EICS, tx_ring->eims_value);
845 else
846 ew32(EIMS, tx_ring->eims_value);
847
848 return IRQ_HANDLED;
849}
850
851static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
852{
853 struct net_device *netdev = data;
854 struct igbvf_adapter *adapter = netdev_priv(netdev);
855
856 adapter->int_counter0++;
857
858 /* Write the ITR value calculated at the end of the
859 * previous interrupt.
860 */
861 if (adapter->rx_ring->set_itr) {
862 writel(adapter->rx_ring->itr_val,
863 adapter->hw.hw_addr + adapter->rx_ring->itr_register);
864 adapter->rx_ring->set_itr = 0;
865 }
866
867 if (napi_schedule_prep(&adapter->rx_ring->napi)) {
868 adapter->total_rx_bytes = 0;
869 adapter->total_rx_packets = 0;
870 __napi_schedule(&adapter->rx_ring->napi);
871 }
872
873 return IRQ_HANDLED;
874}
875
876#define IGBVF_NO_QUEUE -1
877
878static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
879 int tx_queue, int msix_vector)
880{
881 struct e1000_hw *hw = &adapter->hw;
882 u32 ivar, index;
883
884 /* 82576 uses a table-based method for assigning vectors.
885 Each queue has a single entry in the table to which we write
886 a vector number along with a "valid" bit. Sadly, the layout
887 of the table is somewhat counterintuitive. */
888 if (rx_queue > IGBVF_NO_QUEUE) {
889 index = (rx_queue >> 1);
890 ivar = array_er32(IVAR0, index);
891 if (rx_queue & 0x1) {
892 /* vector goes into third byte of register */
893 ivar = ivar & 0xFF00FFFF;
894 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
895 } else {
896 /* vector goes into low byte of register */
897 ivar = ivar & 0xFFFFFF00;
898 ivar |= msix_vector | E1000_IVAR_VALID;
899 }
900 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
901 array_ew32(IVAR0, index, ivar);
902 }
903 if (tx_queue > IGBVF_NO_QUEUE) {
904 index = (tx_queue >> 1);
905 ivar = array_er32(IVAR0, index);
906 if (tx_queue & 0x1) {
907 /* vector goes into high byte of register */
908 ivar = ivar & 0x00FFFFFF;
909 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
910 } else {
911 /* vector goes into second byte of register */
912 ivar = ivar & 0xFFFF00FF;
913 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
914 }
915 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
916 array_ew32(IVAR0, index, ivar);
917 }
918}
919
920/**
921 * igbvf_configure_msix - Configure MSI-X hardware
922 *
923 * igbvf_configure_msix sets up the hardware to properly
924 * generate MSI-X interrupts.
925 **/
926static void igbvf_configure_msix(struct igbvf_adapter *adapter)
927{
928 u32 tmp;
929 struct e1000_hw *hw = &adapter->hw;
930 struct igbvf_ring *tx_ring = adapter->tx_ring;
931 struct igbvf_ring *rx_ring = adapter->rx_ring;
932 int vector = 0;
933
934 adapter->eims_enable_mask = 0;
935
936 igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
937 adapter->eims_enable_mask |= tx_ring->eims_value;
938 if (tx_ring->itr_val)
939 writel(tx_ring->itr_val,
940 hw->hw_addr + tx_ring->itr_register);
941 else
942 writel(1952, hw->hw_addr + tx_ring->itr_register);
943
944 igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
945 adapter->eims_enable_mask |= rx_ring->eims_value;
946 if (rx_ring->itr_val)
947 writel(rx_ring->itr_val,
948 hw->hw_addr + rx_ring->itr_register);
949 else
950 writel(1952, hw->hw_addr + rx_ring->itr_register);
951
952 /* set vector for other causes, i.e. link changes */
953
954 tmp = (vector++ | E1000_IVAR_VALID);
955
956 ew32(IVAR_MISC, tmp);
957
958 adapter->eims_enable_mask = (1 << (vector)) - 1;
959 adapter->eims_other = 1 << (vector - 1);
960 e1e_flush();
961}
962
963static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
964{
965 if (adapter->msix_entries) {
966 pci_disable_msix(adapter->pdev);
967 kfree(adapter->msix_entries);
968 adapter->msix_entries = NULL;
969 }
970}
971
972/**
973 * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
974 *
975 * Attempt to configure interrupts using the best available
976 * capabilities of the hardware and kernel.
977 **/
978static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
979{
980 int err = -ENOMEM;
981 int i;
982
983 /* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
984 adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
985 GFP_KERNEL);
986 if (adapter->msix_entries) {
987 for (i = 0; i < 3; i++)
988 adapter->msix_entries[i].entry = i;
989
990 err = pci_enable_msix(adapter->pdev,
991 adapter->msix_entries, 3);
992 }
993
994 if (err) {
995 /* MSI-X failed */
996 dev_err(&adapter->pdev->dev,
997 "Failed to initialize MSI-X interrupts.\n");
998 igbvf_reset_interrupt_capability(adapter);
999 }
1000}
1001
1002/**
1003 * igbvf_request_msix - Initialize MSI-X interrupts
1004 *
1005 * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1006 * kernel.
1007 **/
1008static int igbvf_request_msix(struct igbvf_adapter *adapter)
1009{
1010 struct net_device *netdev = adapter->netdev;
1011 int err = 0, vector = 0;
1012
1013 if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1014 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1015 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1016 } else {
1017 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1018 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1019 }
1020
1021 err = request_irq(adapter->msix_entries[vector].vector,
1022 igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1023 netdev);
1024 if (err)
1025 goto out;
1026
1027 adapter->tx_ring->itr_register = E1000_EITR(vector);
1028 adapter->tx_ring->itr_val = 1952;
1029 vector++;
1030
1031 err = request_irq(adapter->msix_entries[vector].vector,
1032 igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1033 netdev);
1034 if (err)
1035 goto out;
1036
1037 adapter->rx_ring->itr_register = E1000_EITR(vector);
1038 adapter->rx_ring->itr_val = 1952;
1039 vector++;
1040
1041 err = request_irq(adapter->msix_entries[vector].vector,
1042 igbvf_msix_other, 0, netdev->name, netdev);
1043 if (err)
1044 goto out;
1045
1046 igbvf_configure_msix(adapter);
1047 return 0;
1048out:
1049 return err;
1050}
1051
1052/**
1053 * igbvf_alloc_queues - Allocate memory for all rings
1054 * @adapter: board private structure to initialize
1055 **/
1056static int __devinit igbvf_alloc_queues(struct igbvf_adapter *adapter)
1057{
1058 struct net_device *netdev = adapter->netdev;
1059
1060 adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1061 if (!adapter->tx_ring)
1062 return -ENOMEM;
1063
1064 adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1065 if (!adapter->rx_ring) {
1066 kfree(adapter->tx_ring);
1067 return -ENOMEM;
1068 }
1069
1070 netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1071
1072 return 0;
1073}
1074
1075/**
1076 * igbvf_request_irq - initialize interrupts
1077 *
1078 * Attempts to configure interrupts using the best available
1079 * capabilities of the hardware and kernel.
1080 **/
1081static int igbvf_request_irq(struct igbvf_adapter *adapter)
1082{
1083 int err = -1;
1084
1085 /* igbvf supports msi-x only */
1086 if (adapter->msix_entries)
1087 err = igbvf_request_msix(adapter);
1088
1089 if (!err)
1090 return err;
1091
1092 dev_err(&adapter->pdev->dev,
1093 "Unable to allocate interrupt, Error: %d\n", err);
1094
1095 return err;
1096}
1097
1098static void igbvf_free_irq(struct igbvf_adapter *adapter)
1099{
1100 struct net_device *netdev = adapter->netdev;
1101 int vector;
1102
1103 if (adapter->msix_entries) {
1104 for (vector = 0; vector < 3; vector++)
1105 free_irq(adapter->msix_entries[vector].vector, netdev);
1106 }
1107}
1108
1109/**
1110 * igbvf_irq_disable - Mask off interrupt generation on the NIC
1111 **/
1112static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1113{
1114 struct e1000_hw *hw = &adapter->hw;
1115
1116 ew32(EIMC, ~0);
1117
1118 if (adapter->msix_entries)
1119 ew32(EIAC, 0);
1120}
1121
1122/**
1123 * igbvf_irq_enable - Enable default interrupt generation settings
1124 **/
1125static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1126{
1127 struct e1000_hw *hw = &adapter->hw;
1128
1129 ew32(EIAC, adapter->eims_enable_mask);
1130 ew32(EIAM, adapter->eims_enable_mask);
1131 ew32(EIMS, adapter->eims_enable_mask);
1132}
1133
1134/**
1135 * igbvf_poll - NAPI Rx polling callback
1136 * @napi: struct associated with this polling callback
1137 * @budget: amount of packets driver is allowed to process this poll
1138 **/
1139static int igbvf_poll(struct napi_struct *napi, int budget)
1140{
1141 struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1142 struct igbvf_adapter *adapter = rx_ring->adapter;
1143 struct e1000_hw *hw = &adapter->hw;
1144 int work_done = 0;
1145
1146 igbvf_clean_rx_irq(adapter, &work_done, budget);
1147
1148 /* If not enough Rx work done, exit the polling mode */
1149 if (work_done < budget) {
1150 napi_complete(napi);
1151
1152 if (adapter->itr_setting & 3)
1153 igbvf_set_itr(adapter);
1154
1155 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1156 ew32(EIMS, adapter->rx_ring->eims_value);
1157 }
1158
1159 return work_done;
1160}
1161
1162/**
1163 * igbvf_set_rlpml - set receive large packet maximum length
1164 * @adapter: board private structure
1165 *
1166 * Configure the maximum size of packets that will be received
1167 */
1168static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1169{
1170 int max_frame_size = adapter->max_frame_size;
1171 struct e1000_hw *hw = &adapter->hw;
1172
1173 if (adapter->vlgrp)
1174 max_frame_size += VLAN_TAG_SIZE;
1175
1176 e1000_rlpml_set_vf(hw, max_frame_size);
1177}
1178
1179static void igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1180{
1181 struct igbvf_adapter *adapter = netdev_priv(netdev);
1182 struct e1000_hw *hw = &adapter->hw;
1183
1184 if (hw->mac.ops.set_vfta(hw, vid, true))
1185 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1186}
1187
1188static void igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1189{
1190 struct igbvf_adapter *adapter = netdev_priv(netdev);
1191 struct e1000_hw *hw = &adapter->hw;
1192
1193 igbvf_irq_disable(adapter);
1194 vlan_group_set_device(adapter->vlgrp, vid, NULL);
1195
1196 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1197 igbvf_irq_enable(adapter);
1198
1199 if (hw->mac.ops.set_vfta(hw, vid, false))
1200 dev_err(&adapter->pdev->dev,
1201 "Failed to remove vlan id %d\n", vid);
1202}
1203
1204static void igbvf_vlan_rx_register(struct net_device *netdev,
1205 struct vlan_group *grp)
1206{
1207 struct igbvf_adapter *adapter = netdev_priv(netdev);
1208
1209 adapter->vlgrp = grp;
1210}
1211
1212static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1213{
1214 u16 vid;
1215
1216 if (!adapter->vlgrp)
1217 return;
1218
1219 for (vid = 0; vid < VLAN_N_VID; vid++) {
1220 if (!vlan_group_get_device(adapter->vlgrp, vid))
1221 continue;
1222 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1223 }
1224
1225 igbvf_set_rlpml(adapter);
1226}
1227
1228/**
1229 * igbvf_configure_tx - Configure Transmit Unit after Reset
1230 * @adapter: board private structure
1231 *
1232 * Configure the Tx unit of the MAC after a reset.
1233 **/
1234static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1235{
1236 struct e1000_hw *hw = &adapter->hw;
1237 struct igbvf_ring *tx_ring = adapter->tx_ring;
1238 u64 tdba;
1239 u32 txdctl, dca_txctrl;
1240
1241 /* disable transmits */
1242 txdctl = er32(TXDCTL(0));
1243 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1244 msleep(10);
1245
1246 /* Setup the HW Tx Head and Tail descriptor pointers */
1247 ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1248 tdba = tx_ring->dma;
1249 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1250 ew32(TDBAH(0), (tdba >> 32));
1251 ew32(TDH(0), 0);
1252 ew32(TDT(0), 0);
1253 tx_ring->head = E1000_TDH(0);
1254 tx_ring->tail = E1000_TDT(0);
1255
1256 /* Turn off Relaxed Ordering on head write-backs. The writebacks
1257 * MUST be delivered in order or it will completely screw up
1258 * our bookeeping.
1259 */
1260 dca_txctrl = er32(DCA_TXCTRL(0));
1261 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1262 ew32(DCA_TXCTRL(0), dca_txctrl);
1263
1264 /* enable transmits */
1265 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1266 ew32(TXDCTL(0), txdctl);
1267
1268 /* Setup Transmit Descriptor Settings for eop descriptor */
1269 adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1270
1271 /* enable Report Status bit */
1272 adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1273}
1274
1275/**
1276 * igbvf_setup_srrctl - configure the receive control registers
1277 * @adapter: Board private structure
1278 **/
1279static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1280{
1281 struct e1000_hw *hw = &adapter->hw;
1282 u32 srrctl = 0;
1283
1284 srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1285 E1000_SRRCTL_BSIZEHDR_MASK |
1286 E1000_SRRCTL_BSIZEPKT_MASK);
1287
1288 /* Enable queue drop to avoid head of line blocking */
1289 srrctl |= E1000_SRRCTL_DROP_EN;
1290
1291 /* Setup buffer sizes */
1292 srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1293 E1000_SRRCTL_BSIZEPKT_SHIFT;
1294
1295 if (adapter->rx_buffer_len < 2048) {
1296 adapter->rx_ps_hdr_size = 0;
1297 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1298 } else {
1299 adapter->rx_ps_hdr_size = 128;
1300 srrctl |= adapter->rx_ps_hdr_size <<
1301 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1302 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1303 }
1304
1305 ew32(SRRCTL(0), srrctl);
1306}
1307
1308/**
1309 * igbvf_configure_rx - Configure Receive Unit after Reset
1310 * @adapter: board private structure
1311 *
1312 * Configure the Rx unit of the MAC after a reset.
1313 **/
1314static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1315{
1316 struct e1000_hw *hw = &adapter->hw;
1317 struct igbvf_ring *rx_ring = adapter->rx_ring;
1318 u64 rdba;
1319 u32 rdlen, rxdctl;
1320
1321 /* disable receives */
1322 rxdctl = er32(RXDCTL(0));
1323 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1324 msleep(10);
1325
1326 rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1327
1328 /*
1329 * Setup the HW Rx Head and Tail Descriptor Pointers and
1330 * the Base and Length of the Rx Descriptor Ring
1331 */
1332 rdba = rx_ring->dma;
1333 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1334 ew32(RDBAH(0), (rdba >> 32));
1335 ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1336 rx_ring->head = E1000_RDH(0);
1337 rx_ring->tail = E1000_RDT(0);
1338 ew32(RDH(0), 0);
1339 ew32(RDT(0), 0);
1340
1341 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1342 rxdctl &= 0xFFF00000;
1343 rxdctl |= IGBVF_RX_PTHRESH;
1344 rxdctl |= IGBVF_RX_HTHRESH << 8;
1345 rxdctl |= IGBVF_RX_WTHRESH << 16;
1346
1347 igbvf_set_rlpml(adapter);
1348
1349 /* enable receives */
1350 ew32(RXDCTL(0), rxdctl);
1351}
1352
1353/**
1354 * igbvf_set_multi - Multicast and Promiscuous mode set
1355 * @netdev: network interface device structure
1356 *
1357 * The set_multi entry point is called whenever the multicast address
1358 * list or the network interface flags are updated. This routine is
1359 * responsible for configuring the hardware for proper multicast,
1360 * promiscuous mode, and all-multi behavior.
1361 **/
1362static void igbvf_set_multi(struct net_device *netdev)
1363{
1364 struct igbvf_adapter *adapter = netdev_priv(netdev);
1365 struct e1000_hw *hw = &adapter->hw;
1366 struct netdev_hw_addr *ha;
1367 u8 *mta_list = NULL;
1368 int i;
1369
1370 if (!netdev_mc_empty(netdev)) {
1371 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
1372 if (!mta_list) {
1373 dev_err(&adapter->pdev->dev,
1374 "failed to allocate multicast filter list\n");
1375 return;
1376 }
1377 }
1378
1379 /* prepare a packed array of only addresses. */
1380 i = 0;
1381 netdev_for_each_mc_addr(ha, netdev)
1382 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1383
1384 hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1385 kfree(mta_list);
1386}
1387
1388/**
1389 * igbvf_configure - configure the hardware for Rx and Tx
1390 * @adapter: private board structure
1391 **/
1392static void igbvf_configure(struct igbvf_adapter *adapter)
1393{
1394 igbvf_set_multi(adapter->netdev);
1395
1396 igbvf_restore_vlan(adapter);
1397
1398 igbvf_configure_tx(adapter);
1399 igbvf_setup_srrctl(adapter);
1400 igbvf_configure_rx(adapter);
1401 igbvf_alloc_rx_buffers(adapter->rx_ring,
1402 igbvf_desc_unused(adapter->rx_ring));
1403}
1404
1405/* igbvf_reset - bring the hardware into a known good state
1406 *
1407 * This function boots the hardware and enables some settings that
1408 * require a configuration cycle of the hardware - those cannot be
1409 * set/changed during runtime. After reset the device needs to be
1410 * properly configured for Rx, Tx etc.
1411 */
1412static void igbvf_reset(struct igbvf_adapter *adapter)
1413{
1414 struct e1000_mac_info *mac = &adapter->hw.mac;
1415 struct net_device *netdev = adapter->netdev;
1416 struct e1000_hw *hw = &adapter->hw;
1417
1418 /* Allow time for pending master requests to run */
1419 if (mac->ops.reset_hw(hw))
1420 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1421
1422 mac->ops.init_hw(hw);
1423
1424 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1425 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1426 netdev->addr_len);
1427 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1428 netdev->addr_len);
1429 }
1430
1431 adapter->last_reset = jiffies;
1432}
1433
1434int igbvf_up(struct igbvf_adapter *adapter)
1435{
1436 struct e1000_hw *hw = &adapter->hw;
1437
1438 /* hardware has been reset, we need to reload some things */
1439 igbvf_configure(adapter);
1440
1441 clear_bit(__IGBVF_DOWN, &adapter->state);
1442
1443 napi_enable(&adapter->rx_ring->napi);
1444 if (adapter->msix_entries)
1445 igbvf_configure_msix(adapter);
1446
1447 /* Clear any pending interrupts. */
1448 er32(EICR);
1449 igbvf_irq_enable(adapter);
1450
1451 /* start the watchdog */
1452 hw->mac.get_link_status = 1;
1453 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1454
1455
1456 return 0;
1457}
1458
1459void igbvf_down(struct igbvf_adapter *adapter)
1460{
1461 struct net_device *netdev = adapter->netdev;
1462 struct e1000_hw *hw = &adapter->hw;
1463 u32 rxdctl, txdctl;
1464
1465 /*
1466 * signal that we're down so the interrupt handler does not
1467 * reschedule our watchdog timer
1468 */
1469 set_bit(__IGBVF_DOWN, &adapter->state);
1470
1471 /* disable receives in the hardware */
1472 rxdctl = er32(RXDCTL(0));
1473 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1474
1475 netif_stop_queue(netdev);
1476
1477 /* disable transmits in the hardware */
1478 txdctl = er32(TXDCTL(0));
1479 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1480
1481 /* flush both disables and wait for them to finish */
1482 e1e_flush();
1483 msleep(10);
1484
1485 napi_disable(&adapter->rx_ring->napi);
1486
1487 igbvf_irq_disable(adapter);
1488
1489 del_timer_sync(&adapter->watchdog_timer);
1490
1491 netif_carrier_off(netdev);
1492
1493 /* record the stats before reset*/
1494 igbvf_update_stats(adapter);
1495
1496 adapter->link_speed = 0;
1497 adapter->link_duplex = 0;
1498
1499 igbvf_reset(adapter);
1500 igbvf_clean_tx_ring(adapter->tx_ring);
1501 igbvf_clean_rx_ring(adapter->rx_ring);
1502}
1503
1504void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1505{
1506 might_sleep();
1507 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1508 msleep(1);
1509 igbvf_down(adapter);
1510 igbvf_up(adapter);
1511 clear_bit(__IGBVF_RESETTING, &adapter->state);
1512}
1513
1514/**
1515 * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1516 * @adapter: board private structure to initialize
1517 *
1518 * igbvf_sw_init initializes the Adapter private data structure.
1519 * Fields are initialized based on PCI device information and
1520 * OS network device settings (MTU size).
1521 **/
1522static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1523{
1524 struct net_device *netdev = adapter->netdev;
1525 s32 rc;
1526
1527 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1528 adapter->rx_ps_hdr_size = 0;
1529 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1530 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1531
1532 adapter->tx_int_delay = 8;
1533 adapter->tx_abs_int_delay = 32;
1534 adapter->rx_int_delay = 0;
1535 adapter->rx_abs_int_delay = 8;
1536 adapter->itr_setting = 3;
1537 adapter->itr = 20000;
1538
1539 /* Set various function pointers */
1540 adapter->ei->init_ops(&adapter->hw);
1541
1542 rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1543 if (rc)
1544 return rc;
1545
1546 rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1547 if (rc)
1548 return rc;
1549
1550 igbvf_set_interrupt_capability(adapter);
1551
1552 if (igbvf_alloc_queues(adapter))
1553 return -ENOMEM;
1554
1555 spin_lock_init(&adapter->tx_queue_lock);
1556
1557 /* Explicitly disable IRQ since the NIC can be in any state. */
1558 igbvf_irq_disable(adapter);
1559
1560 spin_lock_init(&adapter->stats_lock);
1561
1562 set_bit(__IGBVF_DOWN, &adapter->state);
1563 return 0;
1564}
1565
1566static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1567{
1568 struct e1000_hw *hw = &adapter->hw;
1569
1570 adapter->stats.last_gprc = er32(VFGPRC);
1571 adapter->stats.last_gorc = er32(VFGORC);
1572 adapter->stats.last_gptc = er32(VFGPTC);
1573 adapter->stats.last_gotc = er32(VFGOTC);
1574 adapter->stats.last_mprc = er32(VFMPRC);
1575 adapter->stats.last_gotlbc = er32(VFGOTLBC);
1576 adapter->stats.last_gptlbc = er32(VFGPTLBC);
1577 adapter->stats.last_gorlbc = er32(VFGORLBC);
1578 adapter->stats.last_gprlbc = er32(VFGPRLBC);
1579
1580 adapter->stats.base_gprc = er32(VFGPRC);
1581 adapter->stats.base_gorc = er32(VFGORC);
1582 adapter->stats.base_gptc = er32(VFGPTC);
1583 adapter->stats.base_gotc = er32(VFGOTC);
1584 adapter->stats.base_mprc = er32(VFMPRC);
1585 adapter->stats.base_gotlbc = er32(VFGOTLBC);
1586 adapter->stats.base_gptlbc = er32(VFGPTLBC);
1587 adapter->stats.base_gorlbc = er32(VFGORLBC);
1588 adapter->stats.base_gprlbc = er32(VFGPRLBC);
1589}
1590
1591/**
1592 * igbvf_open - Called when a network interface is made active
1593 * @netdev: network interface device structure
1594 *
1595 * Returns 0 on success, negative value on failure
1596 *
1597 * The open entry point is called when a network interface is made
1598 * active by the system (IFF_UP). At this point all resources needed
1599 * for transmit and receive operations are allocated, the interrupt
1600 * handler is registered with the OS, the watchdog timer is started,
1601 * and the stack is notified that the interface is ready.
1602 **/
1603static int igbvf_open(struct net_device *netdev)
1604{
1605 struct igbvf_adapter *adapter = netdev_priv(netdev);
1606 struct e1000_hw *hw = &adapter->hw;
1607 int err;
1608
1609 /* disallow open during test */
1610 if (test_bit(__IGBVF_TESTING, &adapter->state))
1611 return -EBUSY;
1612
1613 /* allocate transmit descriptors */
1614 err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1615 if (err)
1616 goto err_setup_tx;
1617
1618 /* allocate receive descriptors */
1619 err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1620 if (err)
1621 goto err_setup_rx;
1622
1623 /*
1624 * before we allocate an interrupt, we must be ready to handle it.
1625 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1626 * as soon as we call pci_request_irq, so we have to setup our
1627 * clean_rx handler before we do so.
1628 */
1629 igbvf_configure(adapter);
1630
1631 err = igbvf_request_irq(adapter);
1632 if (err)
1633 goto err_req_irq;
1634
1635 /* From here on the code is the same as igbvf_up() */
1636 clear_bit(__IGBVF_DOWN, &adapter->state);
1637
1638 napi_enable(&adapter->rx_ring->napi);
1639
1640 /* clear any pending interrupts */
1641 er32(EICR);
1642
1643 igbvf_irq_enable(adapter);
1644
1645 /* start the watchdog */
1646 hw->mac.get_link_status = 1;
1647 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1648
1649 return 0;
1650
1651err_req_irq:
1652 igbvf_free_rx_resources(adapter->rx_ring);
1653err_setup_rx:
1654 igbvf_free_tx_resources(adapter->tx_ring);
1655err_setup_tx:
1656 igbvf_reset(adapter);
1657
1658 return err;
1659}
1660
1661/**
1662 * igbvf_close - Disables a network interface
1663 * @netdev: network interface device structure
1664 *
1665 * Returns 0, this is not allowed to fail
1666 *
1667 * The close entry point is called when an interface is de-activated
1668 * by the OS. The hardware is still under the drivers control, but
1669 * needs to be disabled. A global MAC reset is issued to stop the
1670 * hardware, and all transmit and receive resources are freed.
1671 **/
1672static int igbvf_close(struct net_device *netdev)
1673{
1674 struct igbvf_adapter *adapter = netdev_priv(netdev);
1675
1676 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1677 igbvf_down(adapter);
1678
1679 igbvf_free_irq(adapter);
1680
1681 igbvf_free_tx_resources(adapter->tx_ring);
1682 igbvf_free_rx_resources(adapter->rx_ring);
1683
1684 return 0;
1685}
1686/**
1687 * igbvf_set_mac - Change the Ethernet Address of the NIC
1688 * @netdev: network interface device structure
1689 * @p: pointer to an address structure
1690 *
1691 * Returns 0 on success, negative on failure
1692 **/
1693static int igbvf_set_mac(struct net_device *netdev, void *p)
1694{
1695 struct igbvf_adapter *adapter = netdev_priv(netdev);
1696 struct e1000_hw *hw = &adapter->hw;
1697 struct sockaddr *addr = p;
1698
1699 if (!is_valid_ether_addr(addr->sa_data))
1700 return -EADDRNOTAVAIL;
1701
1702 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1703
1704 hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1705
1706 if (memcmp(addr->sa_data, hw->mac.addr, 6))
1707 return -EADDRNOTAVAIL;
1708
1709 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1710
1711 return 0;
1712}
1713
1714#define UPDATE_VF_COUNTER(reg, name) \
1715 { \
1716 u32 current_counter = er32(reg); \
1717 if (current_counter < adapter->stats.last_##name) \
1718 adapter->stats.name += 0x100000000LL; \
1719 adapter->stats.last_##name = current_counter; \
1720 adapter->stats.name &= 0xFFFFFFFF00000000LL; \
1721 adapter->stats.name |= current_counter; \
1722 }
1723
1724/**
1725 * igbvf_update_stats - Update the board statistics counters
1726 * @adapter: board private structure
1727**/
1728void igbvf_update_stats(struct igbvf_adapter *adapter)
1729{
1730 struct e1000_hw *hw = &adapter->hw;
1731 struct pci_dev *pdev = adapter->pdev;
1732
1733 /*
1734 * Prevent stats update while adapter is being reset, link is down
1735 * or if the pci connection is down.
1736 */
1737 if (adapter->link_speed == 0)
1738 return;
1739
1740 if (test_bit(__IGBVF_RESETTING, &adapter->state))
1741 return;
1742
1743 if (pci_channel_offline(pdev))
1744 return;
1745
1746 UPDATE_VF_COUNTER(VFGPRC, gprc);
1747 UPDATE_VF_COUNTER(VFGORC, gorc);
1748 UPDATE_VF_COUNTER(VFGPTC, gptc);
1749 UPDATE_VF_COUNTER(VFGOTC, gotc);
1750 UPDATE_VF_COUNTER(VFMPRC, mprc);
1751 UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1752 UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1753 UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1754 UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1755
1756 /* Fill out the OS statistics structure */
1757 adapter->net_stats.multicast = adapter->stats.mprc;
1758}
1759
1760static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1761{
1762 dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s\n",
1763 adapter->link_speed,
1764 ((adapter->link_duplex == FULL_DUPLEX) ?
1765 "Full Duplex" : "Half Duplex"));
1766}
1767
1768static bool igbvf_has_link(struct igbvf_adapter *adapter)
1769{
1770 struct e1000_hw *hw = &adapter->hw;
1771 s32 ret_val = E1000_SUCCESS;
1772 bool link_active;
1773
1774 /* If interface is down, stay link down */
1775 if (test_bit(__IGBVF_DOWN, &adapter->state))
1776 return false;
1777
1778 ret_val = hw->mac.ops.check_for_link(hw);
1779 link_active = !hw->mac.get_link_status;
1780
1781 /* if check for link returns error we will need to reset */
1782 if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1783 schedule_work(&adapter->reset_task);
1784
1785 return link_active;
1786}
1787
1788/**
1789 * igbvf_watchdog - Timer Call-back
1790 * @data: pointer to adapter cast into an unsigned long
1791 **/
1792static void igbvf_watchdog(unsigned long data)
1793{
1794 struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1795
1796 /* Do the rest outside of interrupt context */
1797 schedule_work(&adapter->watchdog_task);
1798}
1799
1800static void igbvf_watchdog_task(struct work_struct *work)
1801{
1802 struct igbvf_adapter *adapter = container_of(work,
1803 struct igbvf_adapter,
1804 watchdog_task);
1805 struct net_device *netdev = adapter->netdev;
1806 struct e1000_mac_info *mac = &adapter->hw.mac;
1807 struct igbvf_ring *tx_ring = adapter->tx_ring;
1808 struct e1000_hw *hw = &adapter->hw;
1809 u32 link;
1810 int tx_pending = 0;
1811
1812 link = igbvf_has_link(adapter);
1813
1814 if (link) {
1815 if (!netif_carrier_ok(netdev)) {
1816 mac->ops.get_link_up_info(&adapter->hw,
1817 &adapter->link_speed,
1818 &…Large files files are truncated, but you can click here to view the full file