projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
PM QOS update
[deliverable/linux.git] / drivers / net / igbvf / netdev.c
CommitLineData
d4e0fe01
AD		 1/*******************************************************************************
2
3 Intel(R) 82576 Virtual Function Linux driver
4 Copyright(c) 2009 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>
5a0e3ad6 38#include <linux/slab.h>
d4e0fe01
AD		 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/pm_qos_params.h>
45
46#include "igbvf.h"
47
48#define DRV_VERSION "1.0.0-k0"
49char igbvf_driver_name[] = "igbvf";
50const char igbvf_driver_version[] = DRV_VERSION;
ed77134b 51struct pm_qos_request_list *igbvf_driver_pm_qos_req;
d4e0fe01
AD		 52static const char igbvf_driver_string[] =
53 "Intel(R) Virtual Function Network Driver";
54static const char igbvf_copyright[] = "Copyright (c) 2009 Intel Corporation.";
55
56static int igbvf_poll(struct napi_struct *napi, int budget);
2d165771
AD		 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 *);
d4e0fe01
AD		 60
61static struct igbvf_info igbvf_vf_info = {
62 .mac = e1000_vfadapt,
0364d6fd 63 .flags = 0,
d4e0fe01
AD		 64 .pba = 10,
65 .init_ops = e1000_init_function_pointers_vf,
66};
67
68static const struct igbvf_info *igbvf_info_tbl[] = {
69 [board_vf] = &igbvf_vf_info,
70};
71
72/**
73 * igbvf_desc_unused - calculate if we have unused descriptors
74 **/
75static int igbvf_desc_unused(struct igbvf_ring *ring)
76{
77 if (ring->next_to_clean > ring->next_to_use)
78 return ring->next_to_clean - ring->next_to_use - 1;
79
80 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
81}
82
83/**
84 * igbvf_receive_skb - helper function to handle Rx indications
85 * @adapter: board private structure
86 * @status: descriptor status field as written by hardware
87 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
88 * @skb: pointer to sk_buff to be indicated to stack
89 **/
90static void igbvf_receive_skb(struct igbvf_adapter *adapter,
91 struct net_device *netdev,
92 struct sk_buff *skb,
93 u32 status, u16 vlan)
94{
95 if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
96 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
97 le16_to_cpu(vlan) &
98 E1000_RXD_SPC_VLAN_MASK);
99 else
100 netif_receive_skb(skb);
d4e0fe01
AD		 101}
102
103static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
104 u32 status_err, struct sk_buff *skb)
105{
106 skb->ip_summed = CHECKSUM_NONE;
107
108 /* Ignore Checksum bit is set or checksum is disabled through ethtool */
0364d6fd
AD		 109 if ((status_err & E1000_RXD_STAT_IXSM) ||
110 (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
d4e0fe01 111 return;
0364d6fd 112
d4e0fe01
AD		 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 }
0364d6fd 120
d4e0fe01
AD		 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;
d4e0fe01
AD		 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 pci_map_page(pdev, buffer_info->page,
169 buffer_info->page_offset,
170 PAGE_SIZE / 2,
171 PCI_DMA_FROMDEVICE);
172 }
173
174 if (!buffer_info->skb) {
89d71a66 175 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
d4e0fe01
AD		 176 if (!skb) {
177 adapter->alloc_rx_buff_failed++;
178 goto no_buffers;
179 }
180
d4e0fe01
AD		 181 buffer_info->skb = skb;
182 buffer_info->dma = pci_map_single(pdev, skb->data,
183 bufsz,
184 PCI_DMA_FROMDEVICE);
185 }
186 /* Refresh the desc even if buffer_addrs didn't change because
187 * each write-back erases this info. */
188 if (adapter->rx_ps_hdr_size) {
189 rx_desc->read.pkt_addr =
190 cpu_to_le64(buffer_info->page_dma);
191 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
192 } else {
193 rx_desc->read.pkt_addr =
194 cpu_to_le64(buffer_info->dma);
195 rx_desc->read.hdr_addr = 0;
196 }
197
198 i++;
199 if (i == rx_ring->count)
200 i = 0;
201 buffer_info = &rx_ring->buffer_info[i];
202 }
203
204no_buffers:
205 if (rx_ring->next_to_use != i) {
206 rx_ring->next_to_use = i;
207 if (i == 0)
208 i = (rx_ring->count - 1);
209 else
210 i--;
211
212 /* Force memory writes to complete before letting h/w
213 * know there are new descriptors to fetch. (Only
214 * applicable for weak-ordered memory model archs,
215 * such as IA-64). */
216 wmb();
217 writel(i, adapter->hw.hw_addr + rx_ring->tail);
218 }
219}
220
221/**
222 * igbvf_clean_rx_irq - Send received data up the network stack; legacy
223 * @adapter: board private structure
224 *
225 * the return value indicates whether actual cleaning was done, there
226 * is no guarantee that everything was cleaned
227 **/
228static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
229 int *work_done, int work_to_do)
230{
231 struct igbvf_ring *rx_ring = adapter->rx_ring;
232 struct net_device *netdev = adapter->netdev;
233 struct pci_dev *pdev = adapter->pdev;
234 union e1000_adv_rx_desc *rx_desc, *next_rxd;
235 struct igbvf_buffer *buffer_info, *next_buffer;
236 struct sk_buff *skb;
237 bool cleaned = false;
238 int cleaned_count = 0;
239 unsigned int total_bytes = 0, total_packets = 0;
240 unsigned int i;
241 u32 length, hlen, staterr;
242
243 i = rx_ring->next_to_clean;
244 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
245 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
246
247 while (staterr & E1000_RXD_STAT_DD) {
248 if (*work_done >= work_to_do)
249 break;
250 (*work_done)++;
251
252 buffer_info = &rx_ring->buffer_info[i];
253
254 /* HW will not DMA in data larger than the given buffer, even
255 * if it parses the (NFS, of course) header to be larger. In
256 * that case, it fills the header buffer and spills the rest
257 * into the page.
258 */
259 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
260 E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
261 if (hlen > adapter->rx_ps_hdr_size)
262 hlen = adapter->rx_ps_hdr_size;
263
264 length = le16_to_cpu(rx_desc->wb.upper.length);
265 cleaned = true;
266 cleaned_count++;
267
268 skb = buffer_info->skb;
269 prefetch(skb->data - NET_IP_ALIGN);
270 buffer_info->skb = NULL;
271 if (!adapter->rx_ps_hdr_size) {
272 pci_unmap_single(pdev, buffer_info->dma,
273 adapter->rx_buffer_len,
274 PCI_DMA_FROMDEVICE);
275 buffer_info->dma = 0;
276 skb_put(skb, length);
277 goto send_up;
278 }
279
280 if (!skb_shinfo(skb)->nr_frags) {
281 pci_unmap_single(pdev, buffer_info->dma,
92d947b7 282 adapter->rx_ps_hdr_size,
d4e0fe01
AD		 283 PCI_DMA_FROMDEVICE);
284 skb_put(skb, hlen);
285 }
286
287 if (length) {
288 pci_unmap_page(pdev, buffer_info->page_dma,
289 PAGE_SIZE / 2,
290 PCI_DMA_FROMDEVICE);
291 buffer_info->page_dma = 0;
292
293 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
294 buffer_info->page,
295 buffer_info->page_offset,
296 length);
297
298 if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
299 (page_count(buffer_info->page) != 1))
300 buffer_info->page = NULL;
301 else
302 get_page(buffer_info->page);
303
304 skb->len += length;
305 skb->data_len += length;
306 skb->truesize += length;
307 }
308send_up:
309 i++;
310 if (i == rx_ring->count)
311 i = 0;
312 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
313 prefetch(next_rxd);
314 next_buffer = &rx_ring->buffer_info[i];
315
316 if (!(staterr & E1000_RXD_STAT_EOP)) {
317 buffer_info->skb = next_buffer->skb;
318 buffer_info->dma = next_buffer->dma;
319 next_buffer->skb = skb;
320 next_buffer->dma = 0;
321 goto next_desc;
322 }
323
324 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
325 dev_kfree_skb_irq(skb);
326 goto next_desc;
327 }
328
329 total_bytes += skb->len;
330 total_packets++;
331
332 igbvf_rx_checksum_adv(adapter, staterr, skb);
333
334 skb->protocol = eth_type_trans(skb, netdev);
335
336 igbvf_receive_skb(adapter, netdev, skb, staterr,
337 rx_desc->wb.upper.vlan);
338
d4e0fe01
AD		 339next_desc:
340 rx_desc->wb.upper.status_error = 0;
341
342 /* return some buffers to hardware, one at a time is too slow */
343 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
344 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
345 cleaned_count = 0;
346 }
347
348 /* use prefetched values */
349 rx_desc = next_rxd;
350 buffer_info = next_buffer;
351
352 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
353 }
354
355 rx_ring->next_to_clean = i;
356 cleaned_count = igbvf_desc_unused(rx_ring);
357
358 if (cleaned_count)
359 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
360
361 adapter->total_rx_packets += total_packets;
362 adapter->total_rx_bytes += total_bytes;
363 adapter->net_stats.rx_bytes += total_bytes;
364 adapter->net_stats.rx_packets += total_packets;
365 return cleaned;
366}
367
368static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
369 struct igbvf_buffer *buffer_info)
370{
a7d5ca40
AD		 371 if (buffer_info->dma) {
372 if (buffer_info->mapped_as_page)
373 pci_unmap_page(adapter->pdev,
374 buffer_info->dma,
375 buffer_info->length,
376 PCI_DMA_TODEVICE);
377 else
378 pci_unmap_single(adapter->pdev,
379 buffer_info->dma,
380 buffer_info->length,
381 PCI_DMA_TODEVICE);
382 buffer_info->dma = 0;
383 }
d4e0fe01 384 if (buffer_info->skb) {
d4e0fe01
AD		 385 dev_kfree_skb_any(buffer_info->skb);
386 buffer_info->skb = NULL;
387 }
388 buffer_info->time_stamp = 0;
389}
390
391static void igbvf_print_tx_hang(struct igbvf_adapter *adapter)
392{
393 struct igbvf_ring *tx_ring = adapter->tx_ring;
394 unsigned int i = tx_ring->next_to_clean;
395 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
396 union e1000_adv_tx_desc *eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
397
398 /* detected Tx unit hang */
399 dev_err(&adapter->pdev->dev,
400 "Detected Tx Unit Hang:\n"
401 " TDH <%x>\n"
402 " TDT <%x>\n"
403 " next_to_use <%x>\n"
404 " next_to_clean <%x>\n"
405 "buffer_info[next_to_clean]:\n"
406 " time_stamp <%lx>\n"
407 " next_to_watch <%x>\n"
408 " jiffies <%lx>\n"
409 " next_to_watch.status <%x>\n",
410 readl(adapter->hw.hw_addr + tx_ring->head),
411 readl(adapter->hw.hw_addr + tx_ring->tail),
412 tx_ring->next_to_use,
413 tx_ring->next_to_clean,
414 tx_ring->buffer_info[eop].time_stamp,
415 eop,
416 jiffies,
417 eop_desc->wb.status);
418}
419
420/**
421 * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
422 * @adapter: board private structure
423 *
424 * Return 0 on success, negative on failure
425 **/
426int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
427 struct igbvf_ring *tx_ring)
428{
429 struct pci_dev *pdev = adapter->pdev;
430 int size;
431
432 size = sizeof(struct igbvf_buffer) * tx_ring->count;
433 tx_ring->buffer_info = vmalloc(size);
434 if (!tx_ring->buffer_info)
435 goto err;
436 memset(tx_ring->buffer_info, 0, size);
437
438 /* round up to nearest 4K */
439 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
440 tx_ring->size = ALIGN(tx_ring->size, 4096);
441
442 tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
443 &tx_ring->dma);
444
445 if (!tx_ring->desc)
446 goto err;
447
448 tx_ring->adapter = adapter;
449 tx_ring->next_to_use = 0;
450 tx_ring->next_to_clean = 0;
451
452 return 0;
453err:
454 vfree(tx_ring->buffer_info);
455 dev_err(&adapter->pdev->dev,
456 "Unable to allocate memory for the transmit descriptor ring\n");
457 return -ENOMEM;
458}
459
460/**
461 * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
462 * @adapter: board private structure
463 *
464 * Returns 0 on success, negative on failure
465 **/
466int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
467 struct igbvf_ring *rx_ring)
468{
469 struct pci_dev *pdev = adapter->pdev;
470 int size, desc_len;
471
472 size = sizeof(struct igbvf_buffer) * rx_ring->count;
473 rx_ring->buffer_info = vmalloc(size);
474 if (!rx_ring->buffer_info)
475 goto err;
476 memset(rx_ring->buffer_info, 0, size);
477
478 desc_len = sizeof(union e1000_adv_rx_desc);
479
480 /* Round up to nearest 4K */
481 rx_ring->size = rx_ring->count * desc_len;
482 rx_ring->size = ALIGN(rx_ring->size, 4096);
483
484 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
485 &rx_ring->dma);
486
487 if (!rx_ring->desc)
488 goto err;
489
490 rx_ring->next_to_clean = 0;
491 rx_ring->next_to_use = 0;
492
493 rx_ring->adapter = adapter;
494
495 return 0;
496
497err:
498 vfree(rx_ring->buffer_info);
499 rx_ring->buffer_info = NULL;
500 dev_err(&adapter->pdev->dev,
501 "Unable to allocate memory for the receive descriptor ring\n");
502 return -ENOMEM;
503}
504
505/**
506 * igbvf_clean_tx_ring - Free Tx Buffers
507 * @tx_ring: ring to be cleaned
508 **/
509static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
510{
511 struct igbvf_adapter *adapter = tx_ring->adapter;
512 struct igbvf_buffer *buffer_info;
513 unsigned long size;
514 unsigned int i;
515
516 if (!tx_ring->buffer_info)
517 return;
518
519 /* Free all the Tx ring sk_buffs */
520 for (i = 0; i < tx_ring->count; i++) {
521 buffer_info = &tx_ring->buffer_info[i];
522 igbvf_put_txbuf(adapter, buffer_info);
523 }
524
525 size = sizeof(struct igbvf_buffer) * tx_ring->count;
526 memset(tx_ring->buffer_info, 0, size);
527
528 /* Zero out the descriptor ring */
529 memset(tx_ring->desc, 0, tx_ring->size);
530
531 tx_ring->next_to_use = 0;
532 tx_ring->next_to_clean = 0;
533
534 writel(0, adapter->hw.hw_addr + tx_ring->head);
535 writel(0, adapter->hw.hw_addr + tx_ring->tail);
536}
537
538/**
539 * igbvf_free_tx_resources - Free Tx Resources per Queue
540 * @tx_ring: ring to free resources from
541 *
542 * Free all transmit software resources
543 **/
544void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
545{
546 struct pci_dev *pdev = tx_ring->adapter->pdev;
547
548 igbvf_clean_tx_ring(tx_ring);
549
550 vfree(tx_ring->buffer_info);
551 tx_ring->buffer_info = NULL;
552
553 pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
554
555 tx_ring->desc = NULL;
556}
557
558/**
559 * igbvf_clean_rx_ring - Free Rx Buffers per Queue
560 * @adapter: board private structure
561 **/
562static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
563{
564 struct igbvf_adapter *adapter = rx_ring->adapter;
565 struct igbvf_buffer *buffer_info;
566 struct pci_dev *pdev = adapter->pdev;
567 unsigned long size;
568 unsigned int i;
569
570 if (!rx_ring->buffer_info)
571 return;
572
573 /* Free all the Rx ring sk_buffs */
574 for (i = 0; i < rx_ring->count; i++) {
575 buffer_info = &rx_ring->buffer_info[i];
576 if (buffer_info->dma) {
577 if (adapter->rx_ps_hdr_size){
578 pci_unmap_single(pdev, buffer_info->dma,
579 adapter->rx_ps_hdr_size,
580 PCI_DMA_FROMDEVICE);
581 } else {
582 pci_unmap_single(pdev, buffer_info->dma,
583 adapter->rx_buffer_len,
584 PCI_DMA_FROMDEVICE);
585 }
586 buffer_info->dma = 0;
587 }
588
589 if (buffer_info->skb) {
590 dev_kfree_skb(buffer_info->skb);
591 buffer_info->skb = NULL;
592 }
593
594 if (buffer_info->page) {
595 if (buffer_info->page_dma)
596 pci_unmap_page(pdev, buffer_info->page_dma,
597 PAGE_SIZE / 2,
598 PCI_DMA_FROMDEVICE);
599 put_page(buffer_info->page);
600 buffer_info->page = NULL;
601 buffer_info->page_dma = 0;
602 buffer_info->page_offset = 0;
603 }
604 }
605
606 size = sizeof(struct igbvf_buffer) * rx_ring->count;
607 memset(rx_ring->buffer_info, 0, size);
608
609 /* Zero out the descriptor ring */
610 memset(rx_ring->desc, 0, rx_ring->size);
611
612 rx_ring->next_to_clean = 0;
613 rx_ring->next_to_use = 0;
614
615 writel(0, adapter->hw.hw_addr + rx_ring->head);
616 writel(0, adapter->hw.hw_addr + rx_ring->tail);
617}
618
619/**
620 * igbvf_free_rx_resources - Free Rx Resources
621 * @rx_ring: ring to clean the resources from
622 *
623 * Free all receive software resources
624 **/
625
626void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
627{
628 struct pci_dev *pdev = rx_ring->adapter->pdev;
629
630 igbvf_clean_rx_ring(rx_ring);
631
632 vfree(rx_ring->buffer_info);
633 rx_ring->buffer_info = NULL;
634
635 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
636 rx_ring->dma);
637 rx_ring->desc = NULL;
638}
639
640/**
641 * igbvf_update_itr - update the dynamic ITR value based on statistics
642 * @adapter: pointer to adapter
643 * @itr_setting: current adapter->itr
644 * @packets: the number of packets during this measurement interval
645 * @bytes: the number of bytes during this measurement interval
646 *
647 * Stores a new ITR value based on packets and byte
648 * counts during the last interrupt. The advantage of per interrupt
649 * computation is faster updates and more accurate ITR for the current
650 * traffic pattern. Constants in this function were computed
651 * based on theoretical maximum wire speed and thresholds were set based
652 * on testing data as well as attempting to minimize response time
653 * while increasing bulk throughput. This functionality is controlled
654 * by the InterruptThrottleRate module parameter.
655 **/
656static unsigned int igbvf_update_itr(struct igbvf_adapter *adapter,
657 u16 itr_setting, int packets,
658 int bytes)
659{
660 unsigned int retval = itr_setting;
661
662 if (packets == 0)
663 goto update_itr_done;
664
665 switch (itr_setting) {
666 case lowest_latency:
667 /* handle TSO and jumbo frames */
668 if (bytes/packets > 8000)
669 retval = bulk_latency;
670 else if ((packets < 5) && (bytes > 512))
671 retval = low_latency;
672 break;
673 case low_latency: /* 50 usec aka 20000 ints/s */
674 if (bytes > 10000) {
675 /* this if handles the TSO accounting */
676 if (bytes/packets > 8000)
677 retval = bulk_latency;
678 else if ((packets < 10) || ((bytes/packets) > 1200))
679 retval = bulk_latency;
680 else if ((packets > 35))
681 retval = lowest_latency;
682 } else if (bytes/packets > 2000) {
683 retval = bulk_latency;
684 } else if (packets <= 2 && bytes < 512) {
685 retval = lowest_latency;
686 }
687 break;
688 case bulk_latency: /* 250 usec aka 4000 ints/s */
689 if (bytes > 25000) {
690 if (packets > 35)
691 retval = low_latency;
692 } else if (bytes < 6000) {
693 retval = low_latency;
694 }
695 break;
696 }
697
698update_itr_done:
699 return retval;
700}
701
702static void igbvf_set_itr(struct igbvf_adapter *adapter)
703{
704 struct e1000_hw *hw = &adapter->hw;
705 u16 current_itr;
706 u32 new_itr = adapter->itr;
707
708 adapter->tx_itr = igbvf_update_itr(adapter, adapter->tx_itr,
709 adapter->total_tx_packets,
710 adapter->total_tx_bytes);
711 /* conservative mode (itr 3) eliminates the lowest_latency setting */
712 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
713 adapter->tx_itr = low_latency;
714
715 adapter->rx_itr = igbvf_update_itr(adapter, adapter->rx_itr,
716 adapter->total_rx_packets,
717 adapter->total_rx_bytes);
718 /* conservative mode (itr 3) eliminates the lowest_latency setting */
719 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
720 adapter->rx_itr = low_latency;
721
722 current_itr = max(adapter->rx_itr, adapter->tx_itr);
723
724 switch (current_itr) {
725 /* counts and packets in update_itr are dependent on these numbers */
726 case lowest_latency:
727 new_itr = 70000;
728 break;
729 case low_latency:
730 new_itr = 20000; /* aka hwitr = ~200 */
731 break;
732 case bulk_latency:
733 new_itr = 4000;
734 break;
735 default:
736 break;
737 }
738
739 if (new_itr != adapter->itr) {
740 /*
741 * this attempts to bias the interrupt rate towards Bulk
742 * by adding intermediate steps when interrupt rate is
743 * increasing
744 */
745 new_itr = new_itr > adapter->itr ?
746 min(adapter->itr + (new_itr >> 2), new_itr) :
747 new_itr;
748 adapter->itr = new_itr;
749 adapter->rx_ring->itr_val = 1952;
750
751 if (adapter->msix_entries)
752 adapter->rx_ring->set_itr = 1;
753 else
754 ew32(ITR, 1952);
755 }
756}
757
758/**
759 * igbvf_clean_tx_irq - Reclaim resources after transmit completes
760 * @adapter: board private structure
761 * returns true if ring is completely cleaned
762 **/
763static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
764{
765 struct igbvf_adapter *adapter = tx_ring->adapter;
766 struct e1000_hw *hw = &adapter->hw;
767 struct net_device *netdev = adapter->netdev;
768 struct igbvf_buffer *buffer_info;
769 struct sk_buff *skb;
770 union e1000_adv_tx_desc *tx_desc, *eop_desc;
771 unsigned int total_bytes = 0, total_packets = 0;
772 unsigned int i, eop, count = 0;
773 bool cleaned = false;
774
775 i = tx_ring->next_to_clean;
776 eop = tx_ring->buffer_info[i].next_to_watch;
777 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
778
779 while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
780 (count < tx_ring->count)) {
781 for (cleaned = false; !cleaned; count++) {
782 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
783 buffer_info = &tx_ring->buffer_info[i];
784 cleaned = (i == eop);
785 skb = buffer_info->skb;
786
787 if (skb) {
788 unsigned int segs, bytecount;
789
790 /* gso_segs is currently only valid for tcp */
791 segs = skb_shinfo(skb)->gso_segs ?: 1;
792 /* multiply data chunks by size of headers */
793 bytecount = ((segs - 1) * skb_headlen(skb)) +
794 skb->len;
795 total_packets += segs;
796 total_bytes += bytecount;
797 }
798
799 igbvf_put_txbuf(adapter, buffer_info);
800 tx_desc->wb.status = 0;
801
802 i++;
803 if (i == tx_ring->count)
804 i = 0;
805 }
806 eop = tx_ring->buffer_info[i].next_to_watch;
807 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
808 }
809
810 tx_ring->next_to_clean = i;
811
812 if (unlikely(count &&
813 netif_carrier_ok(netdev) &&
814 igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
815 /* Make sure that anybody stopping the queue after this
816 * sees the new next_to_clean.
817 */
818 smp_mb();
819 if (netif_queue_stopped(netdev) &&
820 !(test_bit(__IGBVF_DOWN, &adapter->state))) {
821 netif_wake_queue(netdev);
822 ++adapter->restart_queue;
823 }
824 }
825
826 if (adapter->detect_tx_hung) {
827 /* Detect a transmit hang in hardware, this serializes the
828 * check with the clearing of time_stamp and movement of i */
829 adapter->detect_tx_hung = false;
830 if (tx_ring->buffer_info[i].time_stamp &&
831 time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
8e95a202
JP		 832 (adapter->tx_timeout_factor * HZ)) &&
833 !(er32(STATUS) & E1000_STATUS_TXOFF)) {
d4e0fe01
AD		 834
835 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
836 /* detected Tx unit hang */
837 igbvf_print_tx_hang(adapter);
838
839 netif_stop_queue(netdev);
840 }
841 }
842 adapter->net_stats.tx_bytes += total_bytes;
843 adapter->net_stats.tx_packets += total_packets;
844 return (count < tx_ring->count);
845}
846
847static irqreturn_t igbvf_msix_other(int irq, void *data)
848{
849 struct net_device *netdev = data;
850 struct igbvf_adapter *adapter = netdev_priv(netdev);
851 struct e1000_hw *hw = &adapter->hw;
852
853 adapter->int_counter1++;
854
855 netif_carrier_off(netdev);
856 hw->mac.get_link_status = 1;
857 if (!test_bit(__IGBVF_DOWN, &adapter->state))
858 mod_timer(&adapter->watchdog_timer, jiffies + 1);
859
860 ew32(EIMS, adapter->eims_other);
861
862 return IRQ_HANDLED;
863}
864
865static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
866{
867 struct net_device *netdev = data;
868 struct igbvf_adapter *adapter = netdev_priv(netdev);
869 struct e1000_hw *hw = &adapter->hw;
870 struct igbvf_ring *tx_ring = adapter->tx_ring;
871
872
873 adapter->total_tx_bytes = 0;
874 adapter->total_tx_packets = 0;
875
876 /* auto mask will automatically reenable the interrupt when we write
877 * EICS */
878 if (!igbvf_clean_tx_irq(tx_ring))
879 /* Ring was not completely cleaned, so fire another interrupt */
880 ew32(EICS, tx_ring->eims_value);
881 else
882 ew32(EIMS, tx_ring->eims_value);
883
884 return IRQ_HANDLED;
885}
886
887static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
888{
889 struct net_device *netdev = data;
890 struct igbvf_adapter *adapter = netdev_priv(netdev);
891
892 adapter->int_counter0++;
893
894 /* Write the ITR value calculated at the end of the
895 * previous interrupt.
896 */
897 if (adapter->rx_ring->set_itr) {
898 writel(adapter->rx_ring->itr_val,
899 adapter->hw.hw_addr + adapter->rx_ring->itr_register);
900 adapter->rx_ring->set_itr = 0;
901 }
902
903 if (napi_schedule_prep(&adapter->rx_ring->napi)) {
904 adapter->total_rx_bytes = 0;
905 adapter->total_rx_packets = 0;
906 __napi_schedule(&adapter->rx_ring->napi);
907 }
908
909 return IRQ_HANDLED;
910}
911
912#define IGBVF_NO_QUEUE -1
913
914static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
915 int tx_queue, int msix_vector)
916{
917 struct e1000_hw *hw = &adapter->hw;
918 u32 ivar, index;
919
920 /* 82576 uses a table-based method for assigning vectors.
921 Each queue has a single entry in the table to which we write
922 a vector number along with a "valid" bit. Sadly, the layout
923 of the table is somewhat counterintuitive. */
924 if (rx_queue > IGBVF_NO_QUEUE) {
925 index = (rx_queue >> 1);
926 ivar = array_er32(IVAR0, index);
927 if (rx_queue & 0x1) {
928 /* vector goes into third byte of register */
929 ivar = ivar & 0xFF00FFFF;
930 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
931 } else {
932 /* vector goes into low byte of register */
933 ivar = ivar & 0xFFFFFF00;
934 ivar |= msix_vector | E1000_IVAR_VALID;
935 }
936 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
937 array_ew32(IVAR0, index, ivar);
938 }
939 if (tx_queue > IGBVF_NO_QUEUE) {
940 index = (tx_queue >> 1);
941 ivar = array_er32(IVAR0, index);
942 if (tx_queue & 0x1) {
943 /* vector goes into high byte of register */
944 ivar = ivar & 0x00FFFFFF;
945 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
946 } else {
947 /* vector goes into second byte of register */
948 ivar = ivar & 0xFFFF00FF;
949 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
950 }
951 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
952 array_ew32(IVAR0, index, ivar);
953 }
954}
955
956/**
957 * igbvf_configure_msix - Configure MSI-X hardware
958 *
959 * igbvf_configure_msix sets up the hardware to properly
960 * generate MSI-X interrupts.
961 **/
962static void igbvf_configure_msix(struct igbvf_adapter *adapter)
963{
964 u32 tmp;
965 struct e1000_hw *hw = &adapter->hw;
966 struct igbvf_ring *tx_ring = adapter->tx_ring;
967 struct igbvf_ring *rx_ring = adapter->rx_ring;
968 int vector = 0;
969
970 adapter->eims_enable_mask = 0;
971
972 igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
973 adapter->eims_enable_mask |= tx_ring->eims_value;
974 if (tx_ring->itr_val)
975 writel(tx_ring->itr_val,
976 hw->hw_addr + tx_ring->itr_register);
977 else
978 writel(1952, hw->hw_addr + tx_ring->itr_register);
979
980 igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
981 adapter->eims_enable_mask |= rx_ring->eims_value;
982 if (rx_ring->itr_val)
983 writel(rx_ring->itr_val,
984 hw->hw_addr + rx_ring->itr_register);
985 else
986 writel(1952, hw->hw_addr + rx_ring->itr_register);
987
988 /* set vector for other causes, i.e. link changes */
989
990 tmp = (vector++ | E1000_IVAR_VALID);
991
992 ew32(IVAR_MISC, tmp);
993
994 adapter->eims_enable_mask = (1 << (vector)) - 1;
995 adapter->eims_other = 1 << (vector - 1);
996 e1e_flush();
997}
998
2d165771 999static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
d4e0fe01
AD		 1000{
1001 if (adapter->msix_entries) {
1002 pci_disable_msix(adapter->pdev);
1003 kfree(adapter->msix_entries);
1004 adapter->msix_entries = NULL;
1005 }
1006}
1007
1008/**
1009 * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1010 *
1011 * Attempt to configure interrupts using the best available
1012 * capabilities of the hardware and kernel.
1013 **/
2d165771 1014static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
d4e0fe01
AD		 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 = kcalloc(3, sizeof(struct msix_entry),
1021 GFP_KERNEL);
1022 if (adapter->msix_entries) {
1023 for (i = 0; i < 3; i++)
1024 adapter->msix_entries[i].entry = i;
1025
1026 err = pci_enable_msix(adapter->pdev,
1027 adapter->msix_entries, 3);
1028 }
1029
1030 if (err) {
1031 /* MSI-X failed */
1032 dev_err(&adapter->pdev->dev,
1033 "Failed to initialize MSI-X interrupts.\n");
1034 igbvf_reset_interrupt_capability(adapter);
1035 }
1036}
1037
1038/**
1039 * igbvf_request_msix - Initialize MSI-X interrupts
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,
a0607fd3 1058 igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
d4e0fe01
AD		 1059 netdev);
1060 if (err)
1061 goto out;
1062
1063 adapter->tx_ring->itr_register = E1000_EITR(vector);
1064 adapter->tx_ring->itr_val = 1952;
1065 vector++;
1066
1067 err = request_irq(adapter->msix_entries[vector].vector,
a0607fd3 1068 igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
d4e0fe01
AD		 1069 netdev);
1070 if (err)
1071 goto out;
1072
1073 adapter->rx_ring->itr_register = E1000_EITR(vector);
1074 adapter->rx_ring->itr_val = 1952;
1075 vector++;
1076
1077 err = request_irq(adapter->msix_entries[vector].vector,
a0607fd3 1078 igbvf_msix_other, 0, netdev->name, netdev);
d4e0fe01
AD		 1079 if (err)
1080 goto out;
1081
1082 igbvf_configure_msix(adapter);
1083 return 0;
1084out:
1085 return err;
1086}
1087
1088/**
1089 * igbvf_alloc_queues - Allocate memory for all rings
1090 * @adapter: board private structure to initialize
1091 **/
1092static int __devinit igbvf_alloc_queues(struct igbvf_adapter *adapter)
1093{
1094 struct net_device *netdev = adapter->netdev;
1095
1096 adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1097 if (!adapter->tx_ring)
1098 return -ENOMEM;
1099
1100 adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1101 if (!adapter->rx_ring) {
1102 kfree(adapter->tx_ring);
1103 return -ENOMEM;
1104 }
1105
1106 netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1107
1108 return 0;
1109}
1110
1111/**
1112 * igbvf_request_irq - initialize interrupts
1113 *
1114 * Attempts to configure interrupts using the best available
1115 * capabilities of the hardware and kernel.
1116 **/
1117static int igbvf_request_irq(struct igbvf_adapter *adapter)
1118{
1119 int err = -1;
1120
1121 /* igbvf supports msi-x only */
1122 if (adapter->msix_entries)
1123 err = igbvf_request_msix(adapter);
1124
1125 if (!err)
1126 return err;
1127
1128 dev_err(&adapter->pdev->dev,
1129 "Unable to allocate interrupt, Error: %d\n", err);
1130
1131 return err;
1132}
1133
1134static void igbvf_free_irq(struct igbvf_adapter *adapter)
1135{
1136 struct net_device *netdev = adapter->netdev;
1137 int vector;
1138
1139 if (adapter->msix_entries) {
1140 for (vector = 0; vector < 3; vector++)
1141 free_irq(adapter->msix_entries[vector].vector, netdev);
1142 }
1143}
1144
1145/**
1146 * igbvf_irq_disable - Mask off interrupt generation on the NIC
1147 **/
1148static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1149{
1150 struct e1000_hw *hw = &adapter->hw;
1151
1152 ew32(EIMC, ~0);
1153
1154 if (adapter->msix_entries)
1155 ew32(EIAC, 0);
1156}
1157
1158/**
1159 * igbvf_irq_enable - Enable default interrupt generation settings
1160 **/
1161static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1162{
1163 struct e1000_hw *hw = &adapter->hw;
1164
1165 ew32(EIAC, adapter->eims_enable_mask);
1166 ew32(EIAM, adapter->eims_enable_mask);
1167 ew32(EIMS, adapter->eims_enable_mask);
1168}
1169
1170/**
1171 * igbvf_poll - NAPI Rx polling callback
1172 * @napi: struct associated with this polling callback
1173 * @budget: amount of packets driver is allowed to process this poll
1174 **/
1175static int igbvf_poll(struct napi_struct *napi, int budget)
1176{
1177 struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1178 struct igbvf_adapter *adapter = rx_ring->adapter;
1179 struct e1000_hw *hw = &adapter->hw;
1180 int work_done = 0;
1181
1182 igbvf_clean_rx_irq(adapter, &work_done, budget);
1183
1184 /* If not enough Rx work done, exit the polling mode */
1185 if (work_done < budget) {
1186 napi_complete(napi);
1187
1188 if (adapter->itr_setting & 3)
1189 igbvf_set_itr(adapter);
1190
1191 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1192 ew32(EIMS, adapter->rx_ring->eims_value);
1193 }
1194
1195 return work_done;
1196}
1197
1198/**
1199 * igbvf_set_rlpml - set receive large packet maximum length
1200 * @adapter: board private structure
1201 *
1202 * Configure the maximum size of packets that will be received
1203 */
1204static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1205{
1206 int max_frame_size = adapter->max_frame_size;
1207 struct e1000_hw *hw = &adapter->hw;
1208
1209 if (adapter->vlgrp)
1210 max_frame_size += VLAN_TAG_SIZE;
1211
1212 e1000_rlpml_set_vf(hw, max_frame_size);
1213}
1214
1215static void igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1216{
1217 struct igbvf_adapter *adapter = netdev_priv(netdev);
1218 struct e1000_hw *hw = &adapter->hw;
1219
1220 if (hw->mac.ops.set_vfta(hw, vid, true))
1221 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1222}
1223
1224static void igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1225{
1226 struct igbvf_adapter *adapter = netdev_priv(netdev);
1227 struct e1000_hw *hw = &adapter->hw;
1228
1229 igbvf_irq_disable(adapter);
1230 vlan_group_set_device(adapter->vlgrp, vid, NULL);
1231
1232 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1233 igbvf_irq_enable(adapter);
1234
1235 if (hw->mac.ops.set_vfta(hw, vid, false))
1236 dev_err(&adapter->pdev->dev,
1237 "Failed to remove vlan id %d\n", vid);
1238}
1239
1240static void igbvf_vlan_rx_register(struct net_device *netdev,
1241 struct vlan_group *grp)
1242{
1243 struct igbvf_adapter *adapter = netdev_priv(netdev);
1244
1245 adapter->vlgrp = grp;
1246}
1247
1248static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1249{
1250 u16 vid;
1251
1252 if (!adapter->vlgrp)
1253 return;
1254
1255 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1256 if (!vlan_group_get_device(adapter->vlgrp, vid))
1257 continue;
1258 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1259 }
1260
1261 igbvf_set_rlpml(adapter);
1262}
1263
1264/**
1265 * igbvf_configure_tx - Configure Transmit Unit after Reset
1266 * @adapter: board private structure
1267 *
1268 * Configure the Tx unit of the MAC after a reset.
1269 **/
1270static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1271{
1272 struct e1000_hw *hw = &adapter->hw;
1273 struct igbvf_ring *tx_ring = adapter->tx_ring;
1274 u64 tdba;
1275 u32 txdctl, dca_txctrl;
1276
1277 /* disable transmits */
1278 txdctl = er32(TXDCTL(0));
1279 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1280 msleep(10);
1281
1282 /* Setup the HW Tx Head and Tail descriptor pointers */
1283 ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1284 tdba = tx_ring->dma;
8e20ce94 1285 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
d4e0fe01
AD		 1286 ew32(TDBAH(0), (tdba >> 32));
1287 ew32(TDH(0), 0);
1288 ew32(TDT(0), 0);
1289 tx_ring->head = E1000_TDH(0);
1290 tx_ring->tail = E1000_TDT(0);
1291
1292 /* Turn off Relaxed Ordering on head write-backs. The writebacks
1293 * MUST be delivered in order or it will completely screw up
1294 * our bookeeping.
1295 */
1296 dca_txctrl = er32(DCA_TXCTRL(0));
1297 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1298 ew32(DCA_TXCTRL(0), dca_txctrl);
1299
1300 /* enable transmits */
1301 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1302 ew32(TXDCTL(0), txdctl);
1303
1304 /* Setup Transmit Descriptor Settings for eop descriptor */
1305 adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1306
1307 /* enable Report Status bit */
1308 adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
d4e0fe01
AD		 1309}
1310
1311/**
1312 * igbvf_setup_srrctl - configure the receive control registers
1313 * @adapter: Board private structure
1314 **/
1315static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1316{
1317 struct e1000_hw *hw = &adapter->hw;
1318 u32 srrctl = 0;
1319
1320 srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1321 E1000_SRRCTL_BSIZEHDR_MASK |
1322 E1000_SRRCTL_BSIZEPKT_MASK);
1323
1324 /* Enable queue drop to avoid head of line blocking */
1325 srrctl |= E1000_SRRCTL_DROP_EN;
1326
1327 /* Setup buffer sizes */
1328 srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1329 E1000_SRRCTL_BSIZEPKT_SHIFT;
1330
1331 if (adapter->rx_buffer_len < 2048) {
1332 adapter->rx_ps_hdr_size = 0;
1333 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1334 } else {
1335 adapter->rx_ps_hdr_size = 128;
1336 srrctl |= adapter->rx_ps_hdr_size <<
1337 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1338 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1339 }
1340
1341 ew32(SRRCTL(0), srrctl);
1342}
1343
1344/**
1345 * igbvf_configure_rx - Configure Receive Unit after Reset
1346 * @adapter: board private structure
1347 *
1348 * Configure the Rx unit of the MAC after a reset.
1349 **/
1350static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1351{
1352 struct e1000_hw *hw = &adapter->hw;
1353 struct igbvf_ring *rx_ring = adapter->rx_ring;
1354 u64 rdba;
1355 u32 rdlen, rxdctl;
1356
1357 /* disable receives */
1358 rxdctl = er32(RXDCTL(0));
1359 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1360 msleep(10);
1361
1362 rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1363
1364 /*
1365 * Setup the HW Rx Head and Tail Descriptor Pointers and
1366 * the Base and Length of the Rx Descriptor Ring
1367 */
1368 rdba = rx_ring->dma;
8e20ce94 1369 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
d4e0fe01
AD		 1370 ew32(RDBAH(0), (rdba >> 32));
1371 ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1372 rx_ring->head = E1000_RDH(0);
1373 rx_ring->tail = E1000_RDT(0);
1374 ew32(RDH(0), 0);
1375 ew32(RDT(0), 0);
1376
1377 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1378 rxdctl &= 0xFFF00000;
1379 rxdctl |= IGBVF_RX_PTHRESH;
1380 rxdctl |= IGBVF_RX_HTHRESH << 8;
1381 rxdctl |= IGBVF_RX_WTHRESH << 16;
1382
1383 igbvf_set_rlpml(adapter);
1384
1385 /* enable receives */
1386 ew32(RXDCTL(0), rxdctl);
1387}
1388
1389/**
1390 * igbvf_set_multi - Multicast and Promiscuous mode set
1391 * @netdev: network interface device structure
1392 *
1393 * The set_multi entry point is called whenever the multicast address
1394 * list or the network interface flags are updated. This routine is
1395 * responsible for configuring the hardware for proper multicast,
1396 * promiscuous mode, and all-multi behavior.
1397 **/
1398static void igbvf_set_multi(struct net_device *netdev)
1399{
1400 struct igbvf_adapter *adapter = netdev_priv(netdev);
1401 struct e1000_hw *hw = &adapter->hw;
1402 struct dev_mc_list *mc_ptr;
1403 u8 *mta_list = NULL;
1404 int i;
1405
4cd24eaf
JP		 1406 if (!netdev_mc_empty(netdev)) {
1407 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
d4e0fe01
AD		 1408 if (!mta_list) {
1409 dev_err(&adapter->pdev->dev,
1410 "failed to allocate multicast filter list\n");
1411 return;
1412 }
1413 }
1414
1415 /* prepare a packed array of only addresses. */
48e2f183
JP		 1416 i = 0;
1417 netdev_for_each_mc_addr(mc_ptr, netdev)
1418 memcpy(mta_list + (i++ * ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
d4e0fe01
AD		 1419
1420 hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1421 kfree(mta_list);
1422}
1423
1424/**
1425 * igbvf_configure - configure the hardware for Rx and Tx
1426 * @adapter: private board structure
1427 **/
1428static void igbvf_configure(struct igbvf_adapter *adapter)
1429{
1430 igbvf_set_multi(adapter->netdev);
1431
1432 igbvf_restore_vlan(adapter);
1433
1434 igbvf_configure_tx(adapter);
1435 igbvf_setup_srrctl(adapter);
1436 igbvf_configure_rx(adapter);
1437 igbvf_alloc_rx_buffers(adapter->rx_ring,
1438 igbvf_desc_unused(adapter->rx_ring));
1439}
1440
1441/* igbvf_reset - bring the hardware into a known good state
1442 *
1443 * This function boots the hardware and enables some settings that
1444 * require a configuration cycle of the hardware - those cannot be
1445 * set/changed during runtime. After reset the device needs to be
1446 * properly configured for Rx, Tx etc.
1447 */
2d165771 1448static void igbvf_reset(struct igbvf_adapter *adapter)
d4e0fe01
AD		 1449{
1450 struct e1000_mac_info *mac = &adapter->hw.mac;
1451 struct net_device *netdev = adapter->netdev;
1452 struct e1000_hw *hw = &adapter->hw;
1453
1454 /* Allow time for pending master requests to run */
1455 if (mac->ops.reset_hw(hw))
1456 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1457
1458 mac->ops.init_hw(hw);
1459
1460 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1461 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1462 netdev->addr_len);
1463 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1464 netdev->addr_len);
1465 }
72279093
AD		 1466
1467 adapter->last_reset = jiffies;
d4e0fe01
AD		 1468}
1469
1470int igbvf_up(struct igbvf_adapter *adapter)
1471{
1472 struct e1000_hw *hw = &adapter->hw;
1473
1474 /* hardware has been reset, we need to reload some things */
1475 igbvf_configure(adapter);
1476
1477 clear_bit(__IGBVF_DOWN, &adapter->state);
1478
1479 napi_enable(&adapter->rx_ring->napi);
1480 if (adapter->msix_entries)
1481 igbvf_configure_msix(adapter);
1482
1483 /* Clear any pending interrupts. */
1484 er32(EICR);
1485 igbvf_irq_enable(adapter);
1486
1487 /* start the watchdog */
1488 hw->mac.get_link_status = 1;
1489 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1490
1491
1492 return 0;
1493}
1494
1495void igbvf_down(struct igbvf_adapter *adapter)
1496{
1497 struct net_device *netdev = adapter->netdev;
1498 struct e1000_hw *hw = &adapter->hw;
1499 u32 rxdctl, txdctl;
1500
1501 /*
1502 * signal that we're down so the interrupt handler does not
1503 * reschedule our watchdog timer
1504 */
1505 set_bit(__IGBVF_DOWN, &adapter->state);
1506
1507 /* disable receives in the hardware */
1508 rxdctl = er32(RXDCTL(0));
1509 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1510
1511 netif_stop_queue(netdev);
1512
1513 /* disable transmits in the hardware */
1514 txdctl = er32(TXDCTL(0));
1515 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1516
1517 /* flush both disables and wait for them to finish */
1518 e1e_flush();
1519 msleep(10);
1520
1521 napi_disable(&adapter->rx_ring->napi);
1522
1523 igbvf_irq_disable(adapter);
1524
1525 del_timer_sync(&adapter->watchdog_timer);
1526
d4e0fe01
AD		 1527 netif_carrier_off(netdev);
1528
1529 /* record the stats before reset*/
1530 igbvf_update_stats(adapter);
1531
1532 adapter->link_speed = 0;
1533 adapter->link_duplex = 0;
1534
1535 igbvf_reset(adapter);
1536 igbvf_clean_tx_ring(adapter->tx_ring);
1537 igbvf_clean_rx_ring(adapter->rx_ring);
1538}
1539
1540void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1541{
1542 might_sleep();
1543 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1544 msleep(1);
1545 igbvf_down(adapter);
1546 igbvf_up(adapter);
1547 clear_bit(__IGBVF_RESETTING, &adapter->state);
1548}
1549
1550/**
1551 * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1552 * @adapter: board private structure to initialize
1553 *
1554 * igbvf_sw_init initializes the Adapter private data structure.
1555 * Fields are initialized based on PCI device information and
1556 * OS network device settings (MTU size).
1557 **/
1558static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1559{
1560 struct net_device *netdev = adapter->netdev;
1561 s32 rc;
1562
1563 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1564 adapter->rx_ps_hdr_size = 0;
1565 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1566 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1567
1568 adapter->tx_int_delay = 8;
1569 adapter->tx_abs_int_delay = 32;
1570 adapter->rx_int_delay = 0;
1571 adapter->rx_abs_int_delay = 8;
1572 adapter->itr_setting = 3;
1573 adapter->itr = 20000;
1574
1575 /* Set various function pointers */
1576 adapter->ei->init_ops(&adapter->hw);
1577
1578 rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1579 if (rc)
1580 return rc;
1581
1582 rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1583 if (rc)
1584 return rc;
1585
1586 igbvf_set_interrupt_capability(adapter);
1587
1588 if (igbvf_alloc_queues(adapter))
1589 return -ENOMEM;
1590
1591 spin_lock_init(&adapter->tx_queue_lock);
1592
1593 /* Explicitly disable IRQ since the NIC can be in any state. */
1594 igbvf_irq_disable(adapter);
1595
1596 spin_lock_init(&adapter->stats_lock);
1597
1598 set_bit(__IGBVF_DOWN, &adapter->state);
1599 return 0;
1600}
1601
1602static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1603{
1604 struct e1000_hw *hw = &adapter->hw;
1605
1606 adapter->stats.last_gprc = er32(VFGPRC);
1607 adapter->stats.last_gorc = er32(VFGORC);
1608 adapter->stats.last_gptc = er32(VFGPTC);
1609 adapter->stats.last_gotc = er32(VFGOTC);
1610 adapter->stats.last_mprc = er32(VFMPRC);
1611 adapter->stats.last_gotlbc = er32(VFGOTLBC);
1612 adapter->stats.last_gptlbc = er32(VFGPTLBC);
1613 adapter->stats.last_gorlbc = er32(VFGORLBC);
1614 adapter->stats.last_gprlbc = er32(VFGPRLBC);
1615
1616 adapter->stats.base_gprc = er32(VFGPRC);
1617 adapter->stats.base_gorc = er32(VFGORC);
1618 adapter->stats.base_gptc = er32(VFGPTC);
1619 adapter->stats.base_gotc = er32(VFGOTC);
1620 adapter->stats.base_mprc = er32(VFMPRC);
1621 adapter->stats.base_gotlbc = er32(VFGOTLBC);
1622 adapter->stats.base_gptlbc = er32(VFGPTLBC);
1623 adapter->stats.base_gorlbc = er32(VFGORLBC);
1624 adapter->stats.base_gprlbc = er32(VFGPRLBC);
1625}
1626
1627/**
1628 * igbvf_open - Called when a network interface is made active
1629 * @netdev: network interface device structure
1630 *
1631 * Returns 0 on success, negative value on failure
1632 *
1633 * The open entry point is called when a network interface is made
1634 * active by the system (IFF_UP). At this point all resources needed
1635 * for transmit and receive operations are allocated, the interrupt
1636 * handler is registered with the OS, the watchdog timer is started,
1637 * and the stack is notified that the interface is ready.
1638 **/
1639static int igbvf_open(struct net_device *netdev)
1640{
1641 struct igbvf_adapter *adapter = netdev_priv(netdev);
1642 struct e1000_hw *hw = &adapter->hw;
1643 int err;
1644
1645 /* disallow open during test */
1646 if (test_bit(__IGBVF_TESTING, &adapter->state))
1647 return -EBUSY;
1648
1649 /* allocate transmit descriptors */
1650 err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1651 if (err)
1652 goto err_setup_tx;
1653
1654 /* allocate receive descriptors */
1655 err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1656 if (err)
1657 goto err_setup_rx;
1658
1659 /*
1660 * before we allocate an interrupt, we must be ready to handle it.
1661 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1662 * as soon as we call pci_request_irq, so we have to setup our
1663 * clean_rx handler before we do so.
1664 */
1665 igbvf_configure(adapter);
1666
1667 err = igbvf_request_irq(adapter);
1668 if (err)
1669 goto err_req_irq;
1670
1671 /* From here on the code is the same as igbvf_up() */
1672 clear_bit(__IGBVF_DOWN, &adapter->state);
1673
1674 napi_enable(&adapter->rx_ring->napi);
1675
1676 /* clear any pending interrupts */
1677 er32(EICR);
1678
1679 igbvf_irq_enable(adapter);
1680
1681 /* start the watchdog */
1682 hw->mac.get_link_status = 1;
1683 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1684
1685 return 0;
1686
1687err_req_irq:
1688 igbvf_free_rx_resources(adapter->rx_ring);
1689err_setup_rx:
1690 igbvf_free_tx_resources(adapter->tx_ring);
1691err_setup_tx:
1692 igbvf_reset(adapter);
1693
1694 return err;
1695}
1696
1697/**
1698 * igbvf_close - Disables a network interface
1699 * @netdev: network interface device structure
1700 *
1701 * Returns 0, this is not allowed to fail
1702 *
1703 * The close entry point is called when an interface is de-activated
1704 * by the OS. The hardware is still under the drivers control, but
1705 * needs to be disabled. A global MAC reset is issued to stop the
1706 * hardware, and all transmit and receive resources are freed.
1707 **/
1708static int igbvf_close(struct net_device *netdev)
1709{
1710 struct igbvf_adapter *adapter = netdev_priv(netdev);
1711
1712 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1713 igbvf_down(adapter);
1714
1715 igbvf_free_irq(adapter);
1716
1717 igbvf_free_tx_resources(adapter->tx_ring);
1718 igbvf_free_rx_resources(adapter->rx_ring);
1719
1720 return 0;
1721}
1722/**
1723 * igbvf_set_mac - Change the Ethernet Address of the NIC
1724 * @netdev: network interface device structure
1725 * @p: pointer to an address structure
1726 *
1727 * Returns 0 on success, negative on failure
1728 **/
1729static int igbvf_set_mac(struct net_device *netdev, void *p)
1730{
1731 struct igbvf_adapter *adapter = netdev_priv(netdev);
1732 struct e1000_hw *hw = &adapter->hw;
1733 struct sockaddr *addr = p;
1734
1735 if (!is_valid_ether_addr(addr->sa_data))
1736 return -EADDRNOTAVAIL;
1737
1738 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1739
1740 hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1741
1742 if (memcmp(addr->sa_data, hw->mac.addr, 6))
1743 return -EADDRNOTAVAIL;
1744
1745 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1746
1747 return 0;
1748}
1749
1750#define UPDATE_VF_COUNTER(reg, name) \
1751 { \
1752 u32 current_counter = er32(reg); \
1753 if (current_counter < adapter->stats.last_##name) \
1754 adapter->stats.name += 0x100000000LL; \
1755 adapter->stats.last_##name = current_counter; \
1756 adapter->stats.name &= 0xFFFFFFFF00000000LL; \
1757 adapter->stats.name |= current_counter; \
1758 }
1759
1760/**
1761 * igbvf_update_stats - Update the board statistics counters
1762 * @adapter: board private structure
1763**/
1764void igbvf_update_stats(struct igbvf_adapter *adapter)
1765{
1766 struct e1000_hw *hw = &adapter->hw;
1767 struct pci_dev *pdev = adapter->pdev;
1768
1769 /*
1770 * Prevent stats update while adapter is being reset, link is down
1771 * or if the pci connection is down.
1772 */
1773 if (adapter->link_speed == 0)
1774 return;
1775
1776 if (test_bit(__IGBVF_RESETTING, &adapter->state))
1777 return;
1778
1779 if (pci_channel_offline(pdev))
1780 return;
1781
1782 UPDATE_VF_COUNTER(VFGPRC, gprc);
1783 UPDATE_VF_COUNTER(VFGORC, gorc);
1784 UPDATE_VF_COUNTER(VFGPTC, gptc);
1785 UPDATE_VF_COUNTER(VFGOTC, gotc);
1786 UPDATE_VF_COUNTER(VFMPRC, mprc);
1787 UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1788 UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1789 UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1790 UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1791
1792 /* Fill out the OS statistics structure */
1793 adapter->net_stats.multicast = adapter->stats.mprc;
1794}
1795
1796static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1797{
1798 dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s\n",
1799 adapter->link_speed,
1800 ((adapter->link_duplex == FULL_DUPLEX) ?
1801 "Full Duplex" : "Half Duplex"));
1802}
1803
1804static bool igbvf_has_link(struct igbvf_adapter *adapter)
1805{
1806 struct e1000_hw *hw = &adapter->hw;
1807 s32 ret_val = E1000_SUCCESS;
1808 bool link_active;
1809
72279093
AD		 1810 /* If interface is down, stay link down */
1811 if (test_bit(__IGBVF_DOWN, &adapter->state))
1812 return false;
1813
d4e0fe01
AD		 1814 ret_val = hw->mac.ops.check_for_link(hw);
1815 link_active = !hw->mac.get_link_status;
1816
1817 /* if check for link returns error we will need to reset */
72279093 1818 if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
d4e0fe01
AD		 1819 schedule_work(&adapter->reset_task);
1820
1821 return link_active;
1822}
1823
1824/**
1825 * igbvf_watchdog - Timer Call-back
1826 * @data: pointer to adapter cast into an unsigned long
1827 **/
1828static void igbvf_watchdog(unsigned long data)
1829{
1830 struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1831
1832 /* Do the rest outside of interrupt context */
1833 schedule_work(&adapter->watchdog_task);
1834}
1835
1836static void igbvf_watchdog_task(struct work_struct *work)
1837{
1838 struct igbvf_adapter *adapter = container_of(work,
1839 struct igbvf_adapter,
1840 watchdog_task);
1841 struct net_device *netdev = adapter->netdev;
1842 struct e1000_mac_info *mac = &adapter->hw.mac;
1843 struct igbvf_ring *tx_ring = adapter->tx_ring;
1844 struct e1000_hw *hw = &adapter->hw;
1845 u32 link;
1846 int tx_pending = 0;
1847
1848 link = igbvf_has_link(adapter);
1849
1850 if (link) {
1851 if (!netif_carrier_ok(netdev)) {
1852 bool txb2b = 1;
1853
1854 mac->ops.get_link_up_info(&adapter->hw,
1855 &adapter->link_speed,
1856 &adapter->link_duplex);
1857 igbvf_print_link_info(adapter);
1858
a08af745 1859 /* adjust timeout factor according to speed/duplex */
d4e0fe01
AD		 1860 adapter->tx_timeout_factor = 1;
1861 switch (adapter->link_speed) {
1862 case SPEED_10:
1863 txb2b = 0;
d4e0fe01
AD		 1864 adapter->tx_timeout_factor = 16;
1865 break;
1866 case SPEED_100:
1867 txb2b = 0;
d4e0fe01
AD		 1868 /* maybe add some timeout factor ? */
1869 break;
1870 }
1871
1872 netif_carrier_on(netdev);
1873 netif_wake_queue(netdev);
1874 }
1875 } else {
1876 if (netif_carrier_ok(netdev)) {
1877 adapter->link_speed = 0;
1878 adapter->link_duplex = 0;
1879 dev_info(&adapter->pdev->dev, "Link is Down\n");
1880 netif_carrier_off(netdev);
1881 netif_stop_queue(netdev);
1882 }
1883 }
1884
1885 if (netif_carrier_ok(netdev)) {
1886 igbvf_update_stats(adapter);
1887 } else {
1888 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1889 tx_ring->count);
1890 if (tx_pending) {
1891 /*
1892 * We've lost link, so the controller stops DMA,
1893 * but we've got queued Tx work that's never going
1894 * to get done, so reset controller to flush Tx.
1895 * (Do the reset outside of interrupt context).
1896 */
1897 adapter->tx_timeout_count++;
1898 schedule_work(&adapter->reset_task);
1899 }
1900 }
1901
1902 /* Cause software interrupt to ensure Rx ring is cleaned */
1903 ew32(EICS, adapter->rx_ring->eims_value);
1904
1905 /* Force detection of hung controller every watchdog period */
1906 adapter->detect_tx_hung = 1;
1907
1908 /* Reset the timer */
1909 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1910 mod_timer(&adapter->watchdog_timer,
1911 round_jiffies(jiffies + (2 * HZ)));
1912}
1913
1914#define IGBVF_TX_FLAGS_CSUM 0x00000001
1915#define IGBVF_TX_FLAGS_VLAN 0x00000002
1916#define IGBVF_TX_FLAGS_TSO 0x00000004
1917#define IGBVF_TX_FLAGS_IPV4 0x00000008
1918#define IGBVF_TX_FLAGS_VLAN_MASK 0xffff0000
1919#define IGBVF_TX_FLAGS_VLAN_SHIFT 16
1920
1921static int igbvf_tso(struct igbvf_adapter *adapter,
1922 struct igbvf_ring *tx_ring,
1923 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1924{
1925 struct e1000_adv_tx_context_desc *context_desc;
1926 unsigned int i;
1927 int err;
1928 struct igbvf_buffer *buffer_info;
1929 u32 info = 0, tu_cmd = 0;
1930 u32 mss_l4len_idx, l4len;
1931 *hdr_len = 0;
1932
1933 if (skb_header_cloned(skb)) {
1934 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1935 if (err) {
1936 dev_err(&adapter->pdev->dev,
1937 "igbvf_tso returning an error\n");
1938 return err;
1939 }
1940 }
1941
1942 l4len = tcp_hdrlen(skb);
1943 *hdr_len += l4len;
1944
1945 if (skb->protocol == htons(ETH_P_IP)) {
1946 struct iphdr *iph = ip_hdr(skb);
1947 iph->tot_len = 0;
1948 iph->check = 0;
1949 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1950 iph->daddr, 0,
1951 IPPROTO_TCP,
1952 0);
8e1e8a47 1953 } else if (skb_is_gso_v6(skb)) {
d4e0fe01
AD		 1954 ipv6_hdr(skb)->payload_len = 0;
1955 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1956 &ipv6_hdr(skb)->daddr,
1957 0, IPPROTO_TCP, 0);
1958 }
1959
1960 i = tx_ring->next_to_use;
1961
1962 buffer_info = &tx_ring->buffer_info[i];
1963 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1964 /* VLAN MACLEN IPLEN */
1965 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1966 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1967 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1968 *hdr_len += skb_network_offset(skb);
1969 info |= (skb_transport_header(skb) - skb_network_header(skb));
1970 *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1971 context_desc->vlan_macip_lens = cpu_to_le32(info);
1972
1973 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1974 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1975
1976 if (skb->protocol == htons(ETH_P_IP))
1977 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1978 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1979
1980 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1981
1982 /* MSS L4LEN IDX */
1983 mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1984 mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1985
1986 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1987 context_desc->seqnum_seed = 0;
1988
1989 buffer_info->time_stamp = jiffies;
1990 buffer_info->next_to_watch = i;
1991 buffer_info->dma = 0;
1992 i++;
1993 if (i == tx_ring->count)
1994 i = 0;
1995
1996 tx_ring->next_to_use = i;
1997
1998 return true;
1999}
2000
2001static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
2002 struct igbvf_ring *tx_ring,
2003 struct sk_buff *skb, u32 tx_flags)
2004{
2005 struct e1000_adv_tx_context_desc *context_desc;
2006 unsigned int i;
2007 struct igbvf_buffer *buffer_info;
2008 u32 info = 0, tu_cmd = 0;
2009
2010 if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2011 (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
2012 i = tx_ring->next_to_use;
2013 buffer_info = &tx_ring->buffer_info[i];
2014 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
2015
2016 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2017 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
2018
2019 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2020 if (skb->ip_summed == CHECKSUM_PARTIAL)
2021 info |= (skb_transport_header(skb) -
2022 skb_network_header(skb));
2023
2024
2025 context_desc->vlan_macip_lens = cpu_to_le32(info);
2026
2027 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2028
2029 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2030 switch (skb->protocol) {
2031 case __constant_htons(ETH_P_IP):
2032 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2033 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2034 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2035 break;
2036 case __constant_htons(ETH_P_IPV6):
2037 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2038 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2039 break;
2040 default:
2041 break;
2042 }
2043 }
2044
2045 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2046 context_desc->seqnum_seed = 0;
2047 context_desc->mss_l4len_idx = 0;
2048
2049 buffer_info->time_stamp = jiffies;
2050 buffer_info->next_to_watch = i;
2051 buffer_info->dma = 0;
2052 i++;
2053 if (i == tx_ring->count)
2054 i = 0;
2055 tx_ring->next_to_use = i;
2056
2057 return true;
2058 }
2059
2060 return false;
2061}
2062
2063static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2064{
2065 struct igbvf_adapter *adapter = netdev_priv(netdev);
2066
2067 /* there is enough descriptors then we don't need to worry */
2068 if (igbvf_desc_unused(adapter->tx_ring) >= size)
2069 return 0;
2070
2071 netif_stop_queue(netdev);
2072
2073 smp_mb();
2074
2075 /* We need to check again just in case room has been made available */
2076 if (igbvf_desc_unused(adapter->tx_ring) < size)
2077 return -EBUSY;
2078
2079 netif_wake_queue(netdev);
2080
2081 ++adapter->restart_queue;
2082 return 0;
2083}
2084
2085#define IGBVF_MAX_TXD_PWR 16
2086#define IGBVF_MAX_DATA_PER_TXD (1 << IGBVF_MAX_TXD_PWR)
2087
2088static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2089 struct igbvf_ring *tx_ring,
2090 struct sk_buff *skb,
2091 unsigned int first)
2092{
2093 struct igbvf_buffer *buffer_info;
a7d5ca40 2094 struct pci_dev *pdev = adapter->pdev;
d4e0fe01
AD		 2095 unsigned int len = skb_headlen(skb);
2096 unsigned int count = 0, i;
2097 unsigned int f;
d4e0fe01
AD		 2098
2099 i = tx_ring->next_to_use;
2100
d4e0fe01
AD		 2101 buffer_info = &tx_ring->buffer_info[i];
2102 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2103 buffer_info->length = len;
2104 /* set time_stamp *before* dma to help avoid a possible race */
2105 buffer_info->time_stamp = jiffies;
2106 buffer_info->next_to_watch = i;
ac26d7d6 2107 buffer_info->mapped_as_page = false;
a7d5ca40
AD		 2108 buffer_info->dma = pci_map_single(pdev, skb->data, len,
2109 PCI_DMA_TODEVICE);
2110 if (pci_dma_mapping_error(pdev, buffer_info->dma))
2111 goto dma_error;
2112
d4e0fe01
AD		 2113
2114 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2115 struct skb_frag_struct *frag;
2116
8581145f 2117 count++;
d4e0fe01
AD		 2118 i++;
2119 if (i == tx_ring->count)
2120 i = 0;
2121
2122 frag = &skb_shinfo(skb)->frags[f];
2123 len = frag->size;
2124
2125 buffer_info = &tx_ring->buffer_info[i];
2126 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2127 buffer_info->length = len;
2128 buffer_info->time_stamp = jiffies;
2129 buffer_info->next_to_watch = i;
a7d5ca40
AD		 2130 buffer_info->mapped_as_page = true;
2131 buffer_info->dma = pci_map_page(pdev,
2132 frag->page,
2133 frag->page_offset,
2134 len,
2135 PCI_DMA_TODEVICE);
2136 if (pci_dma_mapping_error(pdev, buffer_info->dma))
2137 goto dma_error;
d4e0fe01
AD		 2138 }
2139
2140 tx_ring->buffer_info[i].skb = skb;
2141 tx_ring->buffer_info[first].next_to_watch = i;
2142
a7d5ca40
AD		 2143 return ++count;
2144
2145dma_error:
2146 dev_err(&pdev->dev, "TX DMA map failed\n");
2147
2148 /* clear timestamp and dma mappings for failed buffer_info mapping */
2149 buffer_info->dma = 0;
2150 buffer_info->time_stamp = 0;
2151 buffer_info->length = 0;
2152 buffer_info->next_to_watch = 0;
2153 buffer_info->mapped_as_page = false;
c1fa347f
RK		 2154 if (count)
2155 count--;
a7d5ca40
AD		 2156
2157 /* clear timestamp and dma mappings for remaining portion of packet */
c1fa347f
RK		 2158 while (count--) {
2159 if (i==0)
a7d5ca40 2160 i += tx_ring->count;
c1fa347f 2161 i--;
a7d5ca40
AD		 2162 buffer_info = &tx_ring->buffer_info[i];
2163 igbvf_put_txbuf(adapter, buffer_info);
2164 }
2165
2166 return 0;
d4e0fe01
AD		 2167}
2168
2169static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2170 struct igbvf_ring *tx_ring,
2171 int tx_flags, int count, u32 paylen,
2172 u8 hdr_len)
2173{
2174 union e1000_adv_tx_desc *tx_desc = NULL;
2175 struct igbvf_buffer *buffer_info;
2176 u32 olinfo_status = 0, cmd_type_len;
2177 unsigned int i;
2178
2179 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2180 E1000_ADVTXD_DCMD_DEXT);
2181
2182 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2183 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2184
2185 if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2186 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2187
2188 /* insert tcp checksum */
2189 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2190
2191 /* insert ip checksum */
2192 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2193 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2194
2195 } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2196 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2197 }
2198
2199 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2200
2201 i = tx_ring->next_to_use;
2202 while (count--) {
2203 buffer_info = &tx_ring->buffer_info[i];
2204 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2205 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2206 tx_desc->read.cmd_type_len =
2207 cpu_to_le32(cmd_type_len | buffer_info->length);
2208 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2209 i++;
2210 if (i == tx_ring->count)
2211 i = 0;
2212 }
2213
2214 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2215 /* Force memory writes to complete before letting h/w
2216 * know there are new descriptors to fetch. (Only
2217 * applicable for weak-ordered memory model archs,
2218 * such as IA-64). */
2219 wmb();
2220
2221 tx_ring->next_to_use = i;
2222 writel(i, adapter->hw.hw_addr + tx_ring->tail);
2223 /* we need this if more than one processor can write to our tail
2224 * at a time, it syncronizes IO on IA64/Altix systems */
2225 mmiowb();
2226}
2227
3b29a56d
SH		 2228static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2229 struct net_device *netdev,
2230 struct igbvf_ring *tx_ring)
d4e0fe01
AD		 2231{
2232 struct igbvf_adapter *adapter = netdev_priv(netdev);
2233 unsigned int first, tx_flags = 0;
2234 u8 hdr_len = 0;
2235 int count = 0;
2236 int tso = 0;
2237
2238 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2239 dev_kfree_skb_any(skb);
2240 return NETDEV_TX_OK;
2241 }
2242
2243 if (skb->len <= 0) {
2244 dev_kfree_skb_any(skb);
2245 return NETDEV_TX_OK;
2246 }
2247
2248 /*
2249 * need: count + 4 desc gap to keep tail from touching
2250 * + 2 desc gap to keep tail from touching head,
2251 * + 1 desc for skb->data,
2252 * + 1 desc for context descriptor,
2253 * head, otherwise try next time
2254 */
2255 if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2256 /* this is a hard error */
2257 return NETDEV_TX_BUSY;
2258 }
2259
2260 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2261 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2262 tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2263 }
2264
2265 if (skb->protocol == htons(ETH_P_IP))
2266 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2267
2268 first = tx_ring->next_to_use;
2269
2270 tso = skb_is_gso(skb) ?
2271 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2272 if (unlikely(tso < 0)) {
2273 dev_kfree_skb_any(skb);
2274 return NETDEV_TX_OK;
2275 }
2276
2277 if (tso)
2278 tx_flags |= IGBVF_TX_FLAGS_TSO;
2279 else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2280 (skb->ip_summed == CHECKSUM_PARTIAL))
2281 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2282
2283 /*
2284 * count reflects descriptors mapped, if 0 then mapping error
2285 * has occured and we need to rewind the descriptor queue
2286 */
2287 count = igbvf_tx_map_adv(adapter, tx_ring, skb, first);
2288
2289 if (count) {
2290 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2291 skb->len, hdr_len);
d4e0fe01
AD		 2292 /* Make sure there is space in the ring for the next send. */
2293 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2294 } else {
2295 dev_kfree_skb_any(skb);
2296 tx_ring->buffer_info[first].time_stamp = 0;
2297 tx_ring->next_to_use = first;
2298 }
2299
2300 return NETDEV_TX_OK;
2301}
2302
3b29a56d
SH		 2303static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2304 struct net_device *netdev)
d4e0fe01
AD		 2305{
2306 struct igbvf_adapter *adapter = netdev_priv(netdev);
2307 struct igbvf_ring *tx_ring;
d4e0fe01
AD		 2308
2309 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2310 dev_kfree_skb_any(skb);
2311 return NETDEV_TX_OK;
2312 }
2313
2314 tx_ring = &adapter->tx_ring[0];
2315
3b29a56d 2316 return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
d4e0fe01
AD		 2317}
2318
2319/**
2320 * igbvf_tx_timeout - Respond to a Tx Hang
2321 * @netdev: network interface device structure
2322 **/
2323static void igbvf_tx_timeout(struct net_device *netdev)
2324{
2325 struct igbvf_adapter *adapter = netdev_priv(netdev);
2326
2327 /* Do the reset outside of interrupt context */
2328 adapter->tx_timeout_count++;
2329 schedule_work(&adapter->reset_task);
2330}
2331
2332static void igbvf_reset_task(struct work_struct *work)
2333{
2334 struct igbvf_adapter *adapter;
2335 adapter = container_of(work, struct igbvf_adapter, reset_task);
2336
2337 igbvf_reinit_locked(adapter);
2338}
2339
2340/**
2341 * igbvf_get_stats - Get System Network Statistics
2342 * @netdev: network interface device structure
2343 *
2344 * Returns the address of the device statistics structure.
2345 * The statistics are actually updated from the timer callback.
2346 **/
2347static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2348{
2349 struct igbvf_adapter *adapter = netdev_priv(netdev);
2350
2351 /* only return the current stats */
2352 return &adapter->net_stats;
2353}
2354
2355/**
2356 * igbvf_change_mtu - Change the Maximum Transfer Unit
2357 * @netdev: network interface device structure
2358 * @new_mtu: new value for maximum frame size
2359 *
2360 * Returns 0 on success, negative on failure
2361 **/
2362static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2363{
2364 struct igbvf_adapter *adapter = netdev_priv(netdev);
2365 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2366
2367 if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2368 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2369 return -EINVAL;
2370 }
2371
d4e0fe01
AD		 2372#define MAX_STD_JUMBO_FRAME_SIZE 9234
2373 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2374 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2375 return -EINVAL;
2376 }
2377
2378 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2379 msleep(1);
2380 /* igbvf_down has a dependency on max_frame_size */
2381 adapter->max_frame_size = max_frame;
2382 if (netif_running(netdev))
2383 igbvf_down(adapter);
2384
2385 /*
2386 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2387 * means we reserve 2 more, this pushes us to allocate from the next
2388 * larger slab size.
2389 * i.e. RXBUFFER_2048 --> size-4096 slab
2390 * However with the new *_jumbo_rx* routines, jumbo receives will use
2391 * fragmented skbs
2392 */
2393
2394 if (max_frame <= 1024)
2395 adapter->rx_buffer_len = 1024;
2396 else if (max_frame <= 2048)
2397 adapter->rx_buffer_len = 2048;
2398 else
2399#if (PAGE_SIZE / 2) > 16384
2400 adapter->rx_buffer_len = 16384;
2401#else
2402 adapter->rx_buffer_len = PAGE_SIZE / 2;
2403#endif
2404
2405
2406 /* adjust allocation if LPE protects us, and we aren't using SBP */
2407 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2408 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2409 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2410 ETH_FCS_LEN;
2411
2412 dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2413 netdev->mtu, new_mtu);
2414 netdev->mtu = new_mtu;
2415
2416 if (netif_running(netdev))
2417 igbvf_up(adapter);
2418 else
2419 igbvf_reset(adapter);
2420
2421 clear_bit(__IGBVF_RESETTING, &adapter->state);
2422
2423 return 0;
2424}
2425
2426static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2427{
2428 switch (cmd) {
2429 default:
2430 return -EOPNOTSUPP;
2431 }
2432}
2433
2434static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2435{
2436 struct net_device *netdev = pci_get_drvdata(pdev);
2437 struct igbvf_adapter *adapter = netdev_priv(netdev);
2438#ifdef CONFIG_PM
2439 int retval = 0;
2440#endif
2441
2442 netif_device_detach(netdev);
2443
2444 if (netif_running(netdev)) {
2445 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2446 igbvf_down(adapter);
2447 igbvf_free_irq(adapter);
2448 }
2449
2450#ifdef CONFIG_PM
2451 retval = pci_save_state(pdev);
2452 if (retval)
2453 return retval;
2454#endif
2455
2456 pci_disable_device(pdev);
2457
2458 return 0;
2459}
2460
2461#ifdef CONFIG_PM
2462static int igbvf_resume(struct pci_dev *pdev)
2463{
2464 struct net_device *netdev = pci_get_drvdata(pdev);
2465 struct igbvf_adapter *adapter = netdev_priv(netdev);
2466 u32 err;
2467
2468 pci_restore_state(pdev);
2469 err = pci_enable_device_mem(pdev);
2470 if (err) {
2471 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2472 return err;
2473 }
2474
2475 pci_set_master(pdev);
2476
2477 if (netif_running(netdev)) {
2478 err = igbvf_request_irq(adapter);
2479 if (err)
2480 return err;
2481 }
2482
2483 igbvf_reset(adapter);
2484
2485 if (netif_running(netdev))
2486 igbvf_up(adapter);
2487
2488 netif_device_attach(netdev);
2489
2490 return 0;
2491}
2492#endif
2493
2494static void igbvf_shutdown(struct pci_dev *pdev)
2495{
2496 igbvf_suspend(pdev, PMSG_SUSPEND);
2497}
2498
2499#ifdef CONFIG_NET_POLL_CONTROLLER
2500/*
2501 * Polling 'interrupt' - used by things like netconsole to send skbs
2502 * without having to re-enable interrupts. It's not called while
2503 * the interrupt routine is executing.
2504 */
2505static void igbvf_netpoll(struct net_device *netdev)
2506{
2507 struct igbvf_adapter *adapter = netdev_priv(netdev);
2508
2509 disable_irq(adapter->pdev->irq);
2510
2511 igbvf_clean_tx_irq(adapter->tx_ring);
2512
2513 enable_irq(adapter->pdev->irq);
2514}
2515#endif
2516
2517/**
2518 * igbvf_io_error_detected - called when PCI error is detected
2519 * @pdev: Pointer to PCI device
2520 * @state: The current pci connection state
2521 *
2522 * This function is called after a PCI bus error affecting
2523 * this device has been detected.
2524 */
2525static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2526 pci_channel_state_t state)
2527{
2528 struct net_device *netdev = pci_get_drvdata(pdev);
2529 struct igbvf_adapter *adapter = netdev_priv(netdev);
2530
2531 netif_device_detach(netdev);
2532
c06c430d
DN		 2533 if (state == pci_channel_io_perm_failure)
2534 return PCI_ERS_RESULT_DISCONNECT;
2535
d4e0fe01
AD		 2536 if (netif_running(netdev))
2537 igbvf_down(adapter);
2538 pci_disable_device(pdev);
2539
2540 /* Request a slot slot reset. */
2541 return PCI_ERS_RESULT_NEED_RESET;
2542}
2543
2544/**
2545 * igbvf_io_slot_reset - called after the pci bus has been reset.
2546 * @pdev: Pointer to PCI device
2547 *
2548 * Restart the card from scratch, as if from a cold-boot. Implementation
2549 * resembles the first-half of the igbvf_resume routine.
2550 */
2551static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2552{
2553 struct net_device *netdev = pci_get_drvdata(pdev);
2554 struct igbvf_adapter *adapter = netdev_priv(netdev);
2555
2556 if (pci_enable_device_mem(pdev)) {
2557 dev_err(&pdev->dev,
2558 "Cannot re-enable PCI device after reset.\n");
2559 return PCI_ERS_RESULT_DISCONNECT;
2560 }
2561 pci_set_master(pdev);
2562
2563 igbvf_reset(adapter);
2564
2565 return PCI_ERS_RESULT_RECOVERED;
2566}
2567
2568/**
2569 * igbvf_io_resume - called when traffic can start flowing again.
2570 * @pdev: Pointer to PCI device
2571 *
2572 * This callback is called when the error recovery driver tells us that
2573 * its OK to resume normal operation. Implementation resembles the
2574 * second-half of the igbvf_resume routine.
2575 */
2576static void igbvf_io_resume(struct pci_dev *pdev)
2577{
2578 struct net_device *netdev = pci_get_drvdata(pdev);
2579 struct igbvf_adapter *adapter = netdev_priv(netdev);
2580
2581 if (netif_running(netdev)) {
2582 if (igbvf_up(adapter)) {
2583 dev_err(&pdev->dev,
2584 "can't bring device back up after reset\n");
2585 return;
2586 }
2587 }
2588
2589 netif_device_attach(netdev);
2590}
2591
2592static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2593{
2594 struct e1000_hw *hw = &adapter->hw;
2595 struct net_device *netdev = adapter->netdev;
2596 struct pci_dev *pdev = adapter->pdev;
2597
2598 dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
753cdc33 2599 dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
d4e0fe01
AD		 2600 dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
2601}
2602
2603static const struct net_device_ops igbvf_netdev_ops = {
2604 .ndo_open = igbvf_open,
2605 .ndo_stop = igbvf_close,
2606 .ndo_start_xmit = igbvf_xmit_frame,
2607 .ndo_get_stats = igbvf_get_stats,
2608 .ndo_set_multicast_list = igbvf_set_multi,
2609 .ndo_set_mac_address = igbvf_set_mac,
2610 .ndo_change_mtu = igbvf_change_mtu,
2611 .ndo_do_ioctl = igbvf_ioctl,
2612 .ndo_tx_timeout = igbvf_tx_timeout,
2613 .ndo_vlan_rx_register = igbvf_vlan_rx_register,
2614 .ndo_vlan_rx_add_vid = igbvf_vlan_rx_add_vid,
2615 .ndo_vlan_rx_kill_vid = igbvf_vlan_rx_kill_vid,
2616#ifdef CONFIG_NET_POLL_CONTROLLER
2617 .ndo_poll_controller = igbvf_netpoll,
2618#endif
2619};
2620
2621/**
2622 * igbvf_probe - Device Initialization Routine
2623 * @pdev: PCI device information struct
2624 * @ent: entry in igbvf_pci_tbl
2625 *
2626 * Returns 0 on success, negative on failure
2627 *
2628 * igbvf_probe initializes an adapter identified by a pci_dev structure.
2629 * The OS initialization, configuring of the adapter private structure,
2630 * and a hardware reset occur.
2631 **/
2632static int __devinit igbvf_probe(struct pci_dev *pdev,
2633 const struct pci_device_id *ent)
2634{
2635 struct net_device *netdev;
2636 struct igbvf_adapter *adapter;
2637 struct e1000_hw *hw;
2638 const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2639
2640 static int cards_found;
2641 int err, pci_using_dac;
2642
2643 err = pci_enable_device_mem(pdev);
2644 if (err)
2645 return err;
2646
2647 pci_using_dac = 0;
8e20ce94 2648 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
d4e0fe01 2649 if (!err) {
8e20ce94 2650 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
d4e0fe01
AD		 2651 if (!err)
2652 pci_using_dac = 1;
2653 } else {
8e20ce94 2654 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
d4e0fe01 2655 if (err) {
8e20ce94
AM		 2656 err = pci_set_consistent_dma_mask(pdev,
2657 DMA_BIT_MASK(32));
d4e0fe01
AD		 2658 if (err) {
2659 dev_err(&pdev->dev, "No usable DMA "
2660 "configuration, aborting\n");
2661 goto err_dma;
2662 }
2663 }
2664 }
2665
2666 err = pci_request_regions(pdev, igbvf_driver_name);
2667 if (err)
2668 goto err_pci_reg;
2669
2670 pci_set_master(pdev);
2671
2672 err = -ENOMEM;
2673 netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2674 if (!netdev)
2675 goto err_alloc_etherdev;
2676
2677 SET_NETDEV_DEV(netdev, &pdev->dev);
2678
2679 pci_set_drvdata(pdev, netdev);
2680 adapter = netdev_priv(netdev);
2681 hw = &adapter->hw;
2682 adapter->netdev = netdev;
2683 adapter->pdev = pdev;
2684 adapter->ei = ei;
2685 adapter->pba = ei->pba;
2686 adapter->flags = ei->flags;
2687 adapter->hw.back = adapter;
2688 adapter->hw.mac.type = ei->mac;
2689 adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
2690
2691 /* PCI config space info */
2692
2693 hw->vendor_id = pdev->vendor;
2694 hw->device_id = pdev->device;
2695 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2696 hw->subsystem_device_id = pdev->subsystem_device;
2697
2698 pci_read_config_byte(pdev, PCI_REVISION_ID, &hw->revision_id);
2699
2700 err = -EIO;
2701 adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2702 pci_resource_len(pdev, 0));
2703
2704 if (!adapter->hw.hw_addr)
2705 goto err_ioremap;
2706
2707 if (ei->get_variants) {
2708 err = ei->get_variants(adapter);
2709 if (err)
2710 goto err_ioremap;
2711 }
2712
2713 /* setup adapter struct */
2714 err = igbvf_sw_init(adapter);
2715 if (err)
2716 goto err_sw_init;
2717
2718 /* construct the net_device struct */
2719 netdev->netdev_ops = &igbvf_netdev_ops;
2720
2721 igbvf_set_ethtool_ops(netdev);
2722 netdev->watchdog_timeo = 5 * HZ;
2723 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2724
2725 adapter->bd_number = cards_found++;
2726
2727 netdev->features = NETIF_F_SG |
2728 NETIF_F_IP_CSUM |
2729 NETIF_F_HW_VLAN_TX |
2730 NETIF_F_HW_VLAN_RX |
2731 NETIF_F_HW_VLAN_FILTER;
2732
2733 netdev->features |= NETIF_F_IPV6_CSUM;
2734 netdev->features |= NETIF_F_TSO;
2735 netdev->features |= NETIF_F_TSO6;
2736
2737 if (pci_using_dac)
2738 netdev->features |= NETIF_F_HIGHDMA;
2739
2740 netdev->vlan_features |= NETIF_F_TSO;
2741 netdev->vlan_features |= NETIF_F_TSO6;
2742 netdev->vlan_features |= NETIF_F_IP_CSUM;
2743 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2744 netdev->vlan_features |= NETIF_F_SG;
2745
2746 /*reset the controller to put the device in a known good state */
2747 err = hw->mac.ops.reset_hw(hw);
2748 if (err) {
2749 dev_info(&pdev->dev,
1242b6f3
WM		 2750 "PF still in reset state, assigning new address."
2751 " Is the PF interface up?\n");
d4e0fe01
AD		 2752 random_ether_addr(hw->mac.addr);
2753 } else {
2754 err = hw->mac.ops.read_mac_addr(hw);
2755 if (err) {
2756 dev_err(&pdev->dev, "Error reading MAC address\n");
2757 goto err_hw_init;
2758 }
2759 }
2760
2761 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
2762 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
2763
2764 if (!is_valid_ether_addr(netdev->perm_addr)) {
753cdc33
HS		 2765 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
2766 netdev->dev_addr);
d4e0fe01
AD		 2767 err = -EIO;
2768 goto err_hw_init;
2769 }
2770
2771 setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2772 (unsigned long) adapter);
2773
2774 INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2775 INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2776
2777 /* ring size defaults */
2778 adapter->rx_ring->count = 1024;
2779 adapter->tx_ring->count = 1024;
2780
2781 /* reset the hardware with the new settings */
2782 igbvf_reset(adapter);
2783
2784 /* tell the stack to leave us alone until igbvf_open() is called */
2785 netif_carrier_off(netdev);
2786 netif_stop_queue(netdev);
2787
2788 strcpy(netdev->name, "eth%d");
2789 err = register_netdev(netdev);
2790 if (err)
2791 goto err_hw_init;
2792
2793 igbvf_print_device_info(adapter);
2794
2795 igbvf_initialize_last_counter_stats(adapter);
2796
2797 return 0;
2798
2799err_hw_init:
2800 kfree(adapter->tx_ring);
2801 kfree(adapter->rx_ring);
2802err_sw_init:
2803 igbvf_reset_interrupt_capability(adapter);
2804 iounmap(adapter->hw.hw_addr);
2805err_ioremap:
2806 free_netdev(netdev);
2807err_alloc_etherdev:
2808 pci_release_regions(pdev);
2809err_pci_reg:
2810err_dma:
2811 pci_disable_device(pdev);
2812 return err;
2813}
2814
2815/**
2816 * igbvf_remove - Device Removal Routine
2817 * @pdev: PCI device information struct
2818 *
2819 * igbvf_remove is called by the PCI subsystem to alert the driver
2820 * that it should release a PCI device. The could be caused by a
2821 * Hot-Plug event, or because the driver is going to be removed from
2822 * memory.
2823 **/
2824static void __devexit igbvf_remove(struct pci_dev *pdev)
2825{
2826 struct net_device *netdev = pci_get_drvdata(pdev);
2827 struct igbvf_adapter *adapter = netdev_priv(netdev);
2828 struct e1000_hw *hw = &adapter->hw;
2829
2830 /*
2831 * flush_scheduled work may reschedule our watchdog task, so
2832 * explicitly disable watchdog tasks from being rescheduled
2833 */
2834 set_bit(__IGBVF_DOWN, &adapter->state);
2835 del_timer_sync(&adapter->watchdog_timer);
2836
2837 flush_scheduled_work();
2838
2839 unregister_netdev(netdev);
2840
2841 igbvf_reset_interrupt_capability(adapter);
2842
2843 /*
2844 * it is important to delete the napi struct prior to freeing the
2845 * rx ring so that you do not end up with null pointer refs
2846 */
2847 netif_napi_del(&adapter->rx_ring->napi);
2848 kfree(adapter->tx_ring);
2849 kfree(adapter->rx_ring);
2850
2851 iounmap(hw->hw_addr);
2852 if (hw->flash_address)
2853 iounmap(hw->flash_address);
2854 pci_release_regions(pdev);
2855
2856 free_netdev(netdev);
2857
2858 pci_disable_device(pdev);
2859}
2860
2861/* PCI Error Recovery (ERS) */
2862static struct pci_error_handlers igbvf_err_handler = {
2863 .error_detected = igbvf_io_error_detected,
2864 .slot_reset = igbvf_io_slot_reset,
2865 .resume = igbvf_io_resume,
2866};
2867
a3aa1884 2868static DEFINE_PCI_DEVICE_TABLE(igbvf_pci_tbl) = {
d4e0fe01
AD		 2869 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2870 { } /* terminate list */
2871};
2872MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2873
2874/* PCI Device API Driver */
2875static struct pci_driver igbvf_driver = {
2876 .name = igbvf_driver_name,
2877 .id_table = igbvf_pci_tbl,
2878 .probe = igbvf_probe,
2879 .remove = __devexit_p(igbvf_remove),
2880#ifdef CONFIG_PM
2881 /* Power Management Hooks */
2882 .suspend = igbvf_suspend,
2883 .resume = igbvf_resume,
2884#endif
2885 .shutdown = igbvf_shutdown,
2886 .err_handler = &igbvf_err_handler
2887};
2888
2889/**
2890 * igbvf_init_module - Driver Registration Routine
2891 *
2892 * igbvf_init_module is the first routine called when the driver is
2893 * loaded. All it does is register with the PCI subsystem.
2894 **/
2895static int __init igbvf_init_module(void)
2896{
2897 int ret;
2898 printk(KERN_INFO "%s - version %s\n",
2899 igbvf_driver_string, igbvf_driver_version);
2900 printk(KERN_INFO "%s\n", igbvf_copyright);
2901
2902 ret = pci_register_driver(&igbvf_driver);
ed77134b 2903 igbvf_driver_pm_qos_req = pm_qos_add_request(PM_QOS_CPU_DMA_LATENCY,
d4e0fe01
AD		 2904 PM_QOS_DEFAULT_VALUE);
2905
2906 return ret;
2907}
2908module_init(igbvf_init_module);
2909
2910/**
2911 * igbvf_exit_module - Driver Exit Cleanup Routine
2912 *
2913 * igbvf_exit_module is called just before the driver is removed
2914 * from memory.
2915 **/
2916static void __exit igbvf_exit_module(void)
2917{
2918 pci_unregister_driver(&igbvf_driver);
ed77134b
MG		 2919 pm_qos_remove_request(igbvf_driver_pm_qos_req);
2920 igbvf_driver_pm_qos_req = NULL;
d4e0fe01
AD		 2921}
2922module_exit(igbvf_exit_module);
2923
2924
2925MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2926MODULE_DESCRIPTION("Intel(R) 82576 Virtual Function Network Driver");
2927MODULE_LICENSE("GPL");
2928MODULE_VERSION(DRV_VERSION);
2929
2930/* netdev.c */
Linux kernel - Deliverables
This page took 0.257097 seconds and 5 git commands to generate.