blob: e01f44597a26db219581ea1588c323d563138c07 [file] [log] [blame]
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001/*******************************************************************************
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>
38#include <net/checksum.h>
39#include <net/ip6_checksum.h>
40#include <linux/mii.h>
41#include <linux/ethtool.h>
42#include <linux/if_vlan.h>
43#include <linux/pm_qos_params.h>
44
45#include "igbvf.h"
46
47#define DRV_VERSION "1.0.0-k0"
48char igbvf_driver_name[] = "igbvf";
49const char igbvf_driver_version[] = DRV_VERSION;
50static const char igbvf_driver_string[] =
51 "Intel(R) Virtual Function Network Driver";
52static const char igbvf_copyright[] = "Copyright (c) 2009 Intel Corporation.";
53
54static int igbvf_poll(struct napi_struct *napi, int budget);
Alexander Duyck2d165772009-04-09 22:49:20 +000055static void igbvf_reset(struct igbvf_adapter *);
56static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
57static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +000058
59static struct igbvf_info igbvf_vf_info = {
60 .mac = e1000_vfadapt,
Alexander Duyck0364d6f2009-05-06 10:25:01 +000061 .flags = 0,
Alexander Duyckd4e0fe02009-04-07 14:37:34 +000062 .pba = 10,
63 .init_ops = e1000_init_function_pointers_vf,
64};
65
66static const struct igbvf_info *igbvf_info_tbl[] = {
67 [board_vf] = &igbvf_vf_info,
68};
69
70/**
71 * igbvf_desc_unused - calculate if we have unused descriptors
72 **/
73static int igbvf_desc_unused(struct igbvf_ring *ring)
74{
75 if (ring->next_to_clean > ring->next_to_use)
76 return ring->next_to_clean - ring->next_to_use - 1;
77
78 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
79}
80
81/**
82 * igbvf_receive_skb - helper function to handle Rx indications
83 * @adapter: board private structure
84 * @status: descriptor status field as written by hardware
85 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
86 * @skb: pointer to sk_buff to be indicated to stack
87 **/
88static void igbvf_receive_skb(struct igbvf_adapter *adapter,
89 struct net_device *netdev,
90 struct sk_buff *skb,
91 u32 status, u16 vlan)
92{
93 if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
94 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
95 le16_to_cpu(vlan) &
96 E1000_RXD_SPC_VLAN_MASK);
97 else
98 netif_receive_skb(skb);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +000099}
100
101static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
102 u32 status_err, struct sk_buff *skb)
103{
104 skb->ip_summed = CHECKSUM_NONE;
105
106 /* Ignore Checksum bit is set or checksum is disabled through ethtool */
Alexander Duyck0364d6f2009-05-06 10:25:01 +0000107 if ((status_err & E1000_RXD_STAT_IXSM) ||
108 (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000109 return;
Alexander Duyck0364d6f2009-05-06 10:25:01 +0000110
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000111 /* TCP/UDP checksum error bit is set */
112 if (status_err &
113 (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
114 /* let the stack verify checksum errors */
115 adapter->hw_csum_err++;
116 return;
117 }
Alexander Duyck0364d6f2009-05-06 10:25:01 +0000118
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000119 /* It must be a TCP or UDP packet with a valid checksum */
120 if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
121 skb->ip_summed = CHECKSUM_UNNECESSARY;
122
123 adapter->hw_csum_good++;
124}
125
126/**
127 * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
128 * @rx_ring: address of ring structure to repopulate
129 * @cleaned_count: number of buffers to repopulate
130 **/
131static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
132 int cleaned_count)
133{
134 struct igbvf_adapter *adapter = rx_ring->adapter;
135 struct net_device *netdev = adapter->netdev;
136 struct pci_dev *pdev = adapter->pdev;
137 union e1000_adv_rx_desc *rx_desc;
138 struct igbvf_buffer *buffer_info;
139 struct sk_buff *skb;
140 unsigned int i;
141 int bufsz;
142
143 i = rx_ring->next_to_use;
144 buffer_info = &rx_ring->buffer_info[i];
145
146 if (adapter->rx_ps_hdr_size)
147 bufsz = adapter->rx_ps_hdr_size;
148 else
149 bufsz = adapter->rx_buffer_len;
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000150
151 while (cleaned_count--) {
152 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
153
154 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
155 if (!buffer_info->page) {
156 buffer_info->page = alloc_page(GFP_ATOMIC);
157 if (!buffer_info->page) {
158 adapter->alloc_rx_buff_failed++;
159 goto no_buffers;
160 }
161 buffer_info->page_offset = 0;
162 } else {
163 buffer_info->page_offset ^= PAGE_SIZE / 2;
164 }
165 buffer_info->page_dma =
166 pci_map_page(pdev, buffer_info->page,
167 buffer_info->page_offset,
168 PAGE_SIZE / 2,
169 PCI_DMA_FROMDEVICE);
170 }
171
172 if (!buffer_info->skb) {
Eric Dumazet89d71a62009-10-13 05:34:20 +0000173 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000174 if (!skb) {
175 adapter->alloc_rx_buff_failed++;
176 goto no_buffers;
177 }
178
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000179 buffer_info->skb = skb;
180 buffer_info->dma = pci_map_single(pdev, skb->data,
181 bufsz,
182 PCI_DMA_FROMDEVICE);
183 }
184 /* Refresh the desc even if buffer_addrs didn't change because
185 * each write-back erases this info. */
186 if (adapter->rx_ps_hdr_size) {
187 rx_desc->read.pkt_addr =
188 cpu_to_le64(buffer_info->page_dma);
189 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
190 } else {
191 rx_desc->read.pkt_addr =
192 cpu_to_le64(buffer_info->dma);
193 rx_desc->read.hdr_addr = 0;
194 }
195
196 i++;
197 if (i == rx_ring->count)
198 i = 0;
199 buffer_info = &rx_ring->buffer_info[i];
200 }
201
202no_buffers:
203 if (rx_ring->next_to_use != i) {
204 rx_ring->next_to_use = i;
205 if (i == 0)
206 i = (rx_ring->count - 1);
207 else
208 i--;
209
210 /* Force memory writes to complete before letting h/w
211 * know there are new descriptors to fetch. (Only
212 * applicable for weak-ordered memory model archs,
213 * such as IA-64). */
214 wmb();
215 writel(i, adapter->hw.hw_addr + rx_ring->tail);
216 }
217}
218
219/**
220 * igbvf_clean_rx_irq - Send received data up the network stack; legacy
221 * @adapter: board private structure
222 *
223 * the return value indicates whether actual cleaning was done, there
224 * is no guarantee that everything was cleaned
225 **/
226static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
227 int *work_done, int work_to_do)
228{
229 struct igbvf_ring *rx_ring = adapter->rx_ring;
230 struct net_device *netdev = adapter->netdev;
231 struct pci_dev *pdev = adapter->pdev;
232 union e1000_adv_rx_desc *rx_desc, *next_rxd;
233 struct igbvf_buffer *buffer_info, *next_buffer;
234 struct sk_buff *skb;
235 bool cleaned = false;
236 int cleaned_count = 0;
237 unsigned int total_bytes = 0, total_packets = 0;
238 unsigned int i;
239 u32 length, hlen, staterr;
240
241 i = rx_ring->next_to_clean;
242 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
243 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
244
245 while (staterr & E1000_RXD_STAT_DD) {
246 if (*work_done >= work_to_do)
247 break;
248 (*work_done)++;
249
250 buffer_info = &rx_ring->buffer_info[i];
251
252 /* HW will not DMA in data larger than the given buffer, even
253 * if it parses the (NFS, of course) header to be larger. In
254 * that case, it fills the header buffer and spills the rest
255 * into the page.
256 */
257 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
258 E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
259 if (hlen > adapter->rx_ps_hdr_size)
260 hlen = adapter->rx_ps_hdr_size;
261
262 length = le16_to_cpu(rx_desc->wb.upper.length);
263 cleaned = true;
264 cleaned_count++;
265
266 skb = buffer_info->skb;
267 prefetch(skb->data - NET_IP_ALIGN);
268 buffer_info->skb = NULL;
269 if (!adapter->rx_ps_hdr_size) {
270 pci_unmap_single(pdev, buffer_info->dma,
271 adapter->rx_buffer_len,
272 PCI_DMA_FROMDEVICE);
273 buffer_info->dma = 0;
274 skb_put(skb, length);
275 goto send_up;
276 }
277
278 if (!skb_shinfo(skb)->nr_frags) {
279 pci_unmap_single(pdev, buffer_info->dma,
Alexander Duyck92d947b2009-07-23 18:11:01 +0000280 adapter->rx_ps_hdr_size,
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000281 PCI_DMA_FROMDEVICE);
282 skb_put(skb, hlen);
283 }
284
285 if (length) {
286 pci_unmap_page(pdev, buffer_info->page_dma,
287 PAGE_SIZE / 2,
288 PCI_DMA_FROMDEVICE);
289 buffer_info->page_dma = 0;
290
291 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
292 buffer_info->page,
293 buffer_info->page_offset,
294 length);
295
296 if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
297 (page_count(buffer_info->page) != 1))
298 buffer_info->page = NULL;
299 else
300 get_page(buffer_info->page);
301
302 skb->len += length;
303 skb->data_len += length;
304 skb->truesize += length;
305 }
306send_up:
307 i++;
308 if (i == rx_ring->count)
309 i = 0;
310 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
311 prefetch(next_rxd);
312 next_buffer = &rx_ring->buffer_info[i];
313
314 if (!(staterr & E1000_RXD_STAT_EOP)) {
315 buffer_info->skb = next_buffer->skb;
316 buffer_info->dma = next_buffer->dma;
317 next_buffer->skb = skb;
318 next_buffer->dma = 0;
319 goto next_desc;
320 }
321
322 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
323 dev_kfree_skb_irq(skb);
324 goto next_desc;
325 }
326
327 total_bytes += skb->len;
328 total_packets++;
329
330 igbvf_rx_checksum_adv(adapter, staterr, skb);
331
332 skb->protocol = eth_type_trans(skb, netdev);
333
334 igbvf_receive_skb(adapter, netdev, skb, staterr,
335 rx_desc->wb.upper.vlan);
336
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000337next_desc:
338 rx_desc->wb.upper.status_error = 0;
339
340 /* return some buffers to hardware, one at a time is too slow */
341 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
342 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
343 cleaned_count = 0;
344 }
345
346 /* use prefetched values */
347 rx_desc = next_rxd;
348 buffer_info = next_buffer;
349
350 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
351 }
352
353 rx_ring->next_to_clean = i;
354 cleaned_count = igbvf_desc_unused(rx_ring);
355
356 if (cleaned_count)
357 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
358
359 adapter->total_rx_packets += total_packets;
360 adapter->total_rx_bytes += total_bytes;
361 adapter->net_stats.rx_bytes += total_bytes;
362 adapter->net_stats.rx_packets += total_packets;
363 return cleaned;
364}
365
366static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
367 struct igbvf_buffer *buffer_info)
368{
369 buffer_info->dma = 0;
370 if (buffer_info->skb) {
371 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
372 DMA_TO_DEVICE);
373 dev_kfree_skb_any(buffer_info->skb);
374 buffer_info->skb = NULL;
375 }
376 buffer_info->time_stamp = 0;
377}
378
379static void igbvf_print_tx_hang(struct igbvf_adapter *adapter)
380{
381 struct igbvf_ring *tx_ring = adapter->tx_ring;
382 unsigned int i = tx_ring->next_to_clean;
383 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
384 union e1000_adv_tx_desc *eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
385
386 /* detected Tx unit hang */
387 dev_err(&adapter->pdev->dev,
388 "Detected Tx Unit Hang:\n"
389 " TDH <%x>\n"
390 " TDT <%x>\n"
391 " next_to_use <%x>\n"
392 " next_to_clean <%x>\n"
393 "buffer_info[next_to_clean]:\n"
394 " time_stamp <%lx>\n"
395 " next_to_watch <%x>\n"
396 " jiffies <%lx>\n"
397 " next_to_watch.status <%x>\n",
398 readl(adapter->hw.hw_addr + tx_ring->head),
399 readl(adapter->hw.hw_addr + tx_ring->tail),
400 tx_ring->next_to_use,
401 tx_ring->next_to_clean,
402 tx_ring->buffer_info[eop].time_stamp,
403 eop,
404 jiffies,
405 eop_desc->wb.status);
406}
407
408/**
409 * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
410 * @adapter: board private structure
411 *
412 * Return 0 on success, negative on failure
413 **/
414int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
415 struct igbvf_ring *tx_ring)
416{
417 struct pci_dev *pdev = adapter->pdev;
418 int size;
419
420 size = sizeof(struct igbvf_buffer) * tx_ring->count;
421 tx_ring->buffer_info = vmalloc(size);
422 if (!tx_ring->buffer_info)
423 goto err;
424 memset(tx_ring->buffer_info, 0, size);
425
426 /* round up to nearest 4K */
427 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
428 tx_ring->size = ALIGN(tx_ring->size, 4096);
429
430 tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
431 &tx_ring->dma);
432
433 if (!tx_ring->desc)
434 goto err;
435
436 tx_ring->adapter = adapter;
437 tx_ring->next_to_use = 0;
438 tx_ring->next_to_clean = 0;
439
440 return 0;
441err:
442 vfree(tx_ring->buffer_info);
443 dev_err(&adapter->pdev->dev,
444 "Unable to allocate memory for the transmit descriptor ring\n");
445 return -ENOMEM;
446}
447
448/**
449 * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
450 * @adapter: board private structure
451 *
452 * Returns 0 on success, negative on failure
453 **/
454int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
455 struct igbvf_ring *rx_ring)
456{
457 struct pci_dev *pdev = adapter->pdev;
458 int size, desc_len;
459
460 size = sizeof(struct igbvf_buffer) * rx_ring->count;
461 rx_ring->buffer_info = vmalloc(size);
462 if (!rx_ring->buffer_info)
463 goto err;
464 memset(rx_ring->buffer_info, 0, size);
465
466 desc_len = sizeof(union e1000_adv_rx_desc);
467
468 /* Round up to nearest 4K */
469 rx_ring->size = rx_ring->count * desc_len;
470 rx_ring->size = ALIGN(rx_ring->size, 4096);
471
472 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
473 &rx_ring->dma);
474
475 if (!rx_ring->desc)
476 goto err;
477
478 rx_ring->next_to_clean = 0;
479 rx_ring->next_to_use = 0;
480
481 rx_ring->adapter = adapter;
482
483 return 0;
484
485err:
486 vfree(rx_ring->buffer_info);
487 rx_ring->buffer_info = NULL;
488 dev_err(&adapter->pdev->dev,
489 "Unable to allocate memory for the receive descriptor ring\n");
490 return -ENOMEM;
491}
492
493/**
494 * igbvf_clean_tx_ring - Free Tx Buffers
495 * @tx_ring: ring to be cleaned
496 **/
497static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
498{
499 struct igbvf_adapter *adapter = tx_ring->adapter;
500 struct igbvf_buffer *buffer_info;
501 unsigned long size;
502 unsigned int i;
503
504 if (!tx_ring->buffer_info)
505 return;
506
507 /* Free all the Tx ring sk_buffs */
508 for (i = 0; i < tx_ring->count; i++) {
509 buffer_info = &tx_ring->buffer_info[i];
510 igbvf_put_txbuf(adapter, buffer_info);
511 }
512
513 size = sizeof(struct igbvf_buffer) * tx_ring->count;
514 memset(tx_ring->buffer_info, 0, size);
515
516 /* Zero out the descriptor ring */
517 memset(tx_ring->desc, 0, tx_ring->size);
518
519 tx_ring->next_to_use = 0;
520 tx_ring->next_to_clean = 0;
521
522 writel(0, adapter->hw.hw_addr + tx_ring->head);
523 writel(0, adapter->hw.hw_addr + tx_ring->tail);
524}
525
526/**
527 * igbvf_free_tx_resources - Free Tx Resources per Queue
528 * @tx_ring: ring to free resources from
529 *
530 * Free all transmit software resources
531 **/
532void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
533{
534 struct pci_dev *pdev = tx_ring->adapter->pdev;
535
536 igbvf_clean_tx_ring(tx_ring);
537
538 vfree(tx_ring->buffer_info);
539 tx_ring->buffer_info = NULL;
540
541 pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
542
543 tx_ring->desc = NULL;
544}
545
546/**
547 * igbvf_clean_rx_ring - Free Rx Buffers per Queue
548 * @adapter: board private structure
549 **/
550static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
551{
552 struct igbvf_adapter *adapter = rx_ring->adapter;
553 struct igbvf_buffer *buffer_info;
554 struct pci_dev *pdev = adapter->pdev;
555 unsigned long size;
556 unsigned int i;
557
558 if (!rx_ring->buffer_info)
559 return;
560
561 /* Free all the Rx ring sk_buffs */
562 for (i = 0; i < rx_ring->count; i++) {
563 buffer_info = &rx_ring->buffer_info[i];
564 if (buffer_info->dma) {
565 if (adapter->rx_ps_hdr_size){
566 pci_unmap_single(pdev, buffer_info->dma,
567 adapter->rx_ps_hdr_size,
568 PCI_DMA_FROMDEVICE);
569 } else {
570 pci_unmap_single(pdev, buffer_info->dma,
571 adapter->rx_buffer_len,
572 PCI_DMA_FROMDEVICE);
573 }
574 buffer_info->dma = 0;
575 }
576
577 if (buffer_info->skb) {
578 dev_kfree_skb(buffer_info->skb);
579 buffer_info->skb = NULL;
580 }
581
582 if (buffer_info->page) {
583 if (buffer_info->page_dma)
584 pci_unmap_page(pdev, buffer_info->page_dma,
585 PAGE_SIZE / 2,
586 PCI_DMA_FROMDEVICE);
587 put_page(buffer_info->page);
588 buffer_info->page = NULL;
589 buffer_info->page_dma = 0;
590 buffer_info->page_offset = 0;
591 }
592 }
593
594 size = sizeof(struct igbvf_buffer) * rx_ring->count;
595 memset(rx_ring->buffer_info, 0, size);
596
597 /* Zero out the descriptor ring */
598 memset(rx_ring->desc, 0, rx_ring->size);
599
600 rx_ring->next_to_clean = 0;
601 rx_ring->next_to_use = 0;
602
603 writel(0, adapter->hw.hw_addr + rx_ring->head);
604 writel(0, adapter->hw.hw_addr + rx_ring->tail);
605}
606
607/**
608 * igbvf_free_rx_resources - Free Rx Resources
609 * @rx_ring: ring to clean the resources from
610 *
611 * Free all receive software resources
612 **/
613
614void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
615{
616 struct pci_dev *pdev = rx_ring->adapter->pdev;
617
618 igbvf_clean_rx_ring(rx_ring);
619
620 vfree(rx_ring->buffer_info);
621 rx_ring->buffer_info = NULL;
622
623 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
624 rx_ring->dma);
625 rx_ring->desc = NULL;
626}
627
628/**
629 * igbvf_update_itr - update the dynamic ITR value based on statistics
630 * @adapter: pointer to adapter
631 * @itr_setting: current adapter->itr
632 * @packets: the number of packets during this measurement interval
633 * @bytes: the number of bytes during this measurement interval
634 *
635 * Stores a new ITR value based on packets and byte
636 * counts during the last interrupt. The advantage of per interrupt
637 * computation is faster updates and more accurate ITR for the current
638 * traffic pattern. Constants in this function were computed
639 * based on theoretical maximum wire speed and thresholds were set based
640 * on testing data as well as attempting to minimize response time
641 * while increasing bulk throughput. This functionality is controlled
642 * by the InterruptThrottleRate module parameter.
643 **/
644static unsigned int igbvf_update_itr(struct igbvf_adapter *adapter,
645 u16 itr_setting, int packets,
646 int bytes)
647{
648 unsigned int retval = itr_setting;
649
650 if (packets == 0)
651 goto update_itr_done;
652
653 switch (itr_setting) {
654 case lowest_latency:
655 /* handle TSO and jumbo frames */
656 if (bytes/packets > 8000)
657 retval = bulk_latency;
658 else if ((packets < 5) && (bytes > 512))
659 retval = low_latency;
660 break;
661 case low_latency: /* 50 usec aka 20000 ints/s */
662 if (bytes > 10000) {
663 /* this if handles the TSO accounting */
664 if (bytes/packets > 8000)
665 retval = bulk_latency;
666 else if ((packets < 10) || ((bytes/packets) > 1200))
667 retval = bulk_latency;
668 else if ((packets > 35))
669 retval = lowest_latency;
670 } else if (bytes/packets > 2000) {
671 retval = bulk_latency;
672 } else if (packets <= 2 && bytes < 512) {
673 retval = lowest_latency;
674 }
675 break;
676 case bulk_latency: /* 250 usec aka 4000 ints/s */
677 if (bytes > 25000) {
678 if (packets > 35)
679 retval = low_latency;
680 } else if (bytes < 6000) {
681 retval = low_latency;
682 }
683 break;
684 }
685
686update_itr_done:
687 return retval;
688}
689
690static void igbvf_set_itr(struct igbvf_adapter *adapter)
691{
692 struct e1000_hw *hw = &adapter->hw;
693 u16 current_itr;
694 u32 new_itr = adapter->itr;
695
696 adapter->tx_itr = igbvf_update_itr(adapter, adapter->tx_itr,
697 adapter->total_tx_packets,
698 adapter->total_tx_bytes);
699 /* conservative mode (itr 3) eliminates the lowest_latency setting */
700 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
701 adapter->tx_itr = low_latency;
702
703 adapter->rx_itr = igbvf_update_itr(adapter, adapter->rx_itr,
704 adapter->total_rx_packets,
705 adapter->total_rx_bytes);
706 /* conservative mode (itr 3) eliminates the lowest_latency setting */
707 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
708 adapter->rx_itr = low_latency;
709
710 current_itr = max(adapter->rx_itr, adapter->tx_itr);
711
712 switch (current_itr) {
713 /* counts and packets in update_itr are dependent on these numbers */
714 case lowest_latency:
715 new_itr = 70000;
716 break;
717 case low_latency:
718 new_itr = 20000; /* aka hwitr = ~200 */
719 break;
720 case bulk_latency:
721 new_itr = 4000;
722 break;
723 default:
724 break;
725 }
726
727 if (new_itr != adapter->itr) {
728 /*
729 * this attempts to bias the interrupt rate towards Bulk
730 * by adding intermediate steps when interrupt rate is
731 * increasing
732 */
733 new_itr = new_itr > adapter->itr ?
734 min(adapter->itr + (new_itr >> 2), new_itr) :
735 new_itr;
736 adapter->itr = new_itr;
737 adapter->rx_ring->itr_val = 1952;
738
739 if (adapter->msix_entries)
740 adapter->rx_ring->set_itr = 1;
741 else
742 ew32(ITR, 1952);
743 }
744}
745
746/**
747 * igbvf_clean_tx_irq - Reclaim resources after transmit completes
748 * @adapter: board private structure
749 * returns true if ring is completely cleaned
750 **/
751static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
752{
753 struct igbvf_adapter *adapter = tx_ring->adapter;
754 struct e1000_hw *hw = &adapter->hw;
755 struct net_device *netdev = adapter->netdev;
756 struct igbvf_buffer *buffer_info;
757 struct sk_buff *skb;
758 union e1000_adv_tx_desc *tx_desc, *eop_desc;
759 unsigned int total_bytes = 0, total_packets = 0;
760 unsigned int i, eop, count = 0;
761 bool cleaned = false;
762
763 i = tx_ring->next_to_clean;
764 eop = tx_ring->buffer_info[i].next_to_watch;
765 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
766
767 while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
768 (count < tx_ring->count)) {
769 for (cleaned = false; !cleaned; count++) {
770 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
771 buffer_info = &tx_ring->buffer_info[i];
772 cleaned = (i == eop);
773 skb = buffer_info->skb;
774
775 if (skb) {
776 unsigned int segs, bytecount;
777
778 /* gso_segs is currently only valid for tcp */
779 segs = skb_shinfo(skb)->gso_segs ?: 1;
780 /* multiply data chunks by size of headers */
781 bytecount = ((segs - 1) * skb_headlen(skb)) +
782 skb->len;
783 total_packets += segs;
784 total_bytes += bytecount;
785 }
786
787 igbvf_put_txbuf(adapter, buffer_info);
788 tx_desc->wb.status = 0;
789
790 i++;
791 if (i == tx_ring->count)
792 i = 0;
793 }
794 eop = tx_ring->buffer_info[i].next_to_watch;
795 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
796 }
797
798 tx_ring->next_to_clean = i;
799
800 if (unlikely(count &&
801 netif_carrier_ok(netdev) &&
802 igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
803 /* Make sure that anybody stopping the queue after this
804 * sees the new next_to_clean.
805 */
806 smp_mb();
807 if (netif_queue_stopped(netdev) &&
808 !(test_bit(__IGBVF_DOWN, &adapter->state))) {
809 netif_wake_queue(netdev);
810 ++adapter->restart_queue;
811 }
812 }
813
814 if (adapter->detect_tx_hung) {
815 /* Detect a transmit hang in hardware, this serializes the
816 * check with the clearing of time_stamp and movement of i */
817 adapter->detect_tx_hung = false;
818 if (tx_ring->buffer_info[i].time_stamp &&
819 time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
820 (adapter->tx_timeout_factor * HZ))
821 && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
822
823 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
824 /* detected Tx unit hang */
825 igbvf_print_tx_hang(adapter);
826
827 netif_stop_queue(netdev);
828 }
829 }
830 adapter->net_stats.tx_bytes += total_bytes;
831 adapter->net_stats.tx_packets += total_packets;
832 return (count < tx_ring->count);
833}
834
835static irqreturn_t igbvf_msix_other(int irq, void *data)
836{
837 struct net_device *netdev = data;
838 struct igbvf_adapter *adapter = netdev_priv(netdev);
839 struct e1000_hw *hw = &adapter->hw;
840
841 adapter->int_counter1++;
842
843 netif_carrier_off(netdev);
844 hw->mac.get_link_status = 1;
845 if (!test_bit(__IGBVF_DOWN, &adapter->state))
846 mod_timer(&adapter->watchdog_timer, jiffies + 1);
847
848 ew32(EIMS, adapter->eims_other);
849
850 return IRQ_HANDLED;
851}
852
853static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
854{
855 struct net_device *netdev = data;
856 struct igbvf_adapter *adapter = netdev_priv(netdev);
857 struct e1000_hw *hw = &adapter->hw;
858 struct igbvf_ring *tx_ring = adapter->tx_ring;
859
860
861 adapter->total_tx_bytes = 0;
862 adapter->total_tx_packets = 0;
863
864 /* auto mask will automatically reenable the interrupt when we write
865 * EICS */
866 if (!igbvf_clean_tx_irq(tx_ring))
867 /* Ring was not completely cleaned, so fire another interrupt */
868 ew32(EICS, tx_ring->eims_value);
869 else
870 ew32(EIMS, tx_ring->eims_value);
871
872 return IRQ_HANDLED;
873}
874
875static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
876{
877 struct net_device *netdev = data;
878 struct igbvf_adapter *adapter = netdev_priv(netdev);
879
880 adapter->int_counter0++;
881
882 /* Write the ITR value calculated at the end of the
883 * previous interrupt.
884 */
885 if (adapter->rx_ring->set_itr) {
886 writel(adapter->rx_ring->itr_val,
887 adapter->hw.hw_addr + adapter->rx_ring->itr_register);
888 adapter->rx_ring->set_itr = 0;
889 }
890
891 if (napi_schedule_prep(&adapter->rx_ring->napi)) {
892 adapter->total_rx_bytes = 0;
893 adapter->total_rx_packets = 0;
894 __napi_schedule(&adapter->rx_ring->napi);
895 }
896
897 return IRQ_HANDLED;
898}
899
900#define IGBVF_NO_QUEUE -1
901
902static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
903 int tx_queue, int msix_vector)
904{
905 struct e1000_hw *hw = &adapter->hw;
906 u32 ivar, index;
907
908 /* 82576 uses a table-based method for assigning vectors.
909 Each queue has a single entry in the table to which we write
910 a vector number along with a "valid" bit. Sadly, the layout
911 of the table is somewhat counterintuitive. */
912 if (rx_queue > IGBVF_NO_QUEUE) {
913 index = (rx_queue >> 1);
914 ivar = array_er32(IVAR0, index);
915 if (rx_queue & 0x1) {
916 /* vector goes into third byte of register */
917 ivar = ivar & 0xFF00FFFF;
918 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
919 } else {
920 /* vector goes into low byte of register */
921 ivar = ivar & 0xFFFFFF00;
922 ivar |= msix_vector | E1000_IVAR_VALID;
923 }
924 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
925 array_ew32(IVAR0, index, ivar);
926 }
927 if (tx_queue > IGBVF_NO_QUEUE) {
928 index = (tx_queue >> 1);
929 ivar = array_er32(IVAR0, index);
930 if (tx_queue & 0x1) {
931 /* vector goes into high byte of register */
932 ivar = ivar & 0x00FFFFFF;
933 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
934 } else {
935 /* vector goes into second byte of register */
936 ivar = ivar & 0xFFFF00FF;
937 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
938 }
939 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
940 array_ew32(IVAR0, index, ivar);
941 }
942}
943
944/**
945 * igbvf_configure_msix - Configure MSI-X hardware
946 *
947 * igbvf_configure_msix sets up the hardware to properly
948 * generate MSI-X interrupts.
949 **/
950static void igbvf_configure_msix(struct igbvf_adapter *adapter)
951{
952 u32 tmp;
953 struct e1000_hw *hw = &adapter->hw;
954 struct igbvf_ring *tx_ring = adapter->tx_ring;
955 struct igbvf_ring *rx_ring = adapter->rx_ring;
956 int vector = 0;
957
958 adapter->eims_enable_mask = 0;
959
960 igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
961 adapter->eims_enable_mask |= tx_ring->eims_value;
962 if (tx_ring->itr_val)
963 writel(tx_ring->itr_val,
964 hw->hw_addr + tx_ring->itr_register);
965 else
966 writel(1952, hw->hw_addr + tx_ring->itr_register);
967
968 igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
969 adapter->eims_enable_mask |= rx_ring->eims_value;
970 if (rx_ring->itr_val)
971 writel(rx_ring->itr_val,
972 hw->hw_addr + rx_ring->itr_register);
973 else
974 writel(1952, hw->hw_addr + rx_ring->itr_register);
975
976 /* set vector for other causes, i.e. link changes */
977
978 tmp = (vector++ | E1000_IVAR_VALID);
979
980 ew32(IVAR_MISC, tmp);
981
982 adapter->eims_enable_mask = (1 << (vector)) - 1;
983 adapter->eims_other = 1 << (vector - 1);
984 e1e_flush();
985}
986
Alexander Duyck2d165772009-04-09 22:49:20 +0000987static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000988{
989 if (adapter->msix_entries) {
990 pci_disable_msix(adapter->pdev);
991 kfree(adapter->msix_entries);
992 adapter->msix_entries = NULL;
993 }
994}
995
996/**
997 * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
998 *
999 * Attempt to configure interrupts using the best available
1000 * capabilities of the hardware and kernel.
1001 **/
Alexander Duyck2d165772009-04-09 22:49:20 +00001002static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001003{
1004 int err = -ENOMEM;
1005 int i;
1006
1007 /* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
1008 adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1009 GFP_KERNEL);
1010 if (adapter->msix_entries) {
1011 for (i = 0; i < 3; i++)
1012 adapter->msix_entries[i].entry = i;
1013
1014 err = pci_enable_msix(adapter->pdev,
1015 adapter->msix_entries, 3);
1016 }
1017
1018 if (err) {
1019 /* MSI-X failed */
1020 dev_err(&adapter->pdev->dev,
1021 "Failed to initialize MSI-X interrupts.\n");
1022 igbvf_reset_interrupt_capability(adapter);
1023 }
1024}
1025
1026/**
1027 * igbvf_request_msix - Initialize MSI-X interrupts
1028 *
1029 * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1030 * kernel.
1031 **/
1032static int igbvf_request_msix(struct igbvf_adapter *adapter)
1033{
1034 struct net_device *netdev = adapter->netdev;
1035 int err = 0, vector = 0;
1036
1037 if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1038 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1039 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1040 } else {
1041 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1042 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1043 }
1044
1045 err = request_irq(adapter->msix_entries[vector].vector,
Joe Perchesa0607fd2009-11-18 23:29:17 -08001046 igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001047 netdev);
1048 if (err)
1049 goto out;
1050
1051 adapter->tx_ring->itr_register = E1000_EITR(vector);
1052 adapter->tx_ring->itr_val = 1952;
1053 vector++;
1054
1055 err = request_irq(adapter->msix_entries[vector].vector,
Joe Perchesa0607fd2009-11-18 23:29:17 -08001056 igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001057 netdev);
1058 if (err)
1059 goto out;
1060
1061 adapter->rx_ring->itr_register = E1000_EITR(vector);
1062 adapter->rx_ring->itr_val = 1952;
1063 vector++;
1064
1065 err = request_irq(adapter->msix_entries[vector].vector,
Joe Perchesa0607fd2009-11-18 23:29:17 -08001066 igbvf_msix_other, 0, netdev->name, netdev);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001067 if (err)
1068 goto out;
1069
1070 igbvf_configure_msix(adapter);
1071 return 0;
1072out:
1073 return err;
1074}
1075
1076/**
1077 * igbvf_alloc_queues - Allocate memory for all rings
1078 * @adapter: board private structure to initialize
1079 **/
1080static int __devinit igbvf_alloc_queues(struct igbvf_adapter *adapter)
1081{
1082 struct net_device *netdev = adapter->netdev;
1083
1084 adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1085 if (!adapter->tx_ring)
1086 return -ENOMEM;
1087
1088 adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1089 if (!adapter->rx_ring) {
1090 kfree(adapter->tx_ring);
1091 return -ENOMEM;
1092 }
1093
1094 netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1095
1096 return 0;
1097}
1098
1099/**
1100 * igbvf_request_irq - initialize interrupts
1101 *
1102 * Attempts to configure interrupts using the best available
1103 * capabilities of the hardware and kernel.
1104 **/
1105static int igbvf_request_irq(struct igbvf_adapter *adapter)
1106{
1107 int err = -1;
1108
1109 /* igbvf supports msi-x only */
1110 if (adapter->msix_entries)
1111 err = igbvf_request_msix(adapter);
1112
1113 if (!err)
1114 return err;
1115
1116 dev_err(&adapter->pdev->dev,
1117 "Unable to allocate interrupt, Error: %d\n", err);
1118
1119 return err;
1120}
1121
1122static void igbvf_free_irq(struct igbvf_adapter *adapter)
1123{
1124 struct net_device *netdev = adapter->netdev;
1125 int vector;
1126
1127 if (adapter->msix_entries) {
1128 for (vector = 0; vector < 3; vector++)
1129 free_irq(adapter->msix_entries[vector].vector, netdev);
1130 }
1131}
1132
1133/**
1134 * igbvf_irq_disable - Mask off interrupt generation on the NIC
1135 **/
1136static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1137{
1138 struct e1000_hw *hw = &adapter->hw;
1139
1140 ew32(EIMC, ~0);
1141
1142 if (adapter->msix_entries)
1143 ew32(EIAC, 0);
1144}
1145
1146/**
1147 * igbvf_irq_enable - Enable default interrupt generation settings
1148 **/
1149static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1150{
1151 struct e1000_hw *hw = &adapter->hw;
1152
1153 ew32(EIAC, adapter->eims_enable_mask);
1154 ew32(EIAM, adapter->eims_enable_mask);
1155 ew32(EIMS, adapter->eims_enable_mask);
1156}
1157
1158/**
1159 * igbvf_poll - NAPI Rx polling callback
1160 * @napi: struct associated with this polling callback
1161 * @budget: amount of packets driver is allowed to process this poll
1162 **/
1163static int igbvf_poll(struct napi_struct *napi, int budget)
1164{
1165 struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1166 struct igbvf_adapter *adapter = rx_ring->adapter;
1167 struct e1000_hw *hw = &adapter->hw;
1168 int work_done = 0;
1169
1170 igbvf_clean_rx_irq(adapter, &work_done, budget);
1171
1172 /* If not enough Rx work done, exit the polling mode */
1173 if (work_done < budget) {
1174 napi_complete(napi);
1175
1176 if (adapter->itr_setting & 3)
1177 igbvf_set_itr(adapter);
1178
1179 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1180 ew32(EIMS, adapter->rx_ring->eims_value);
1181 }
1182
1183 return work_done;
1184}
1185
1186/**
1187 * igbvf_set_rlpml - set receive large packet maximum length
1188 * @adapter: board private structure
1189 *
1190 * Configure the maximum size of packets that will be received
1191 */
1192static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1193{
1194 int max_frame_size = adapter->max_frame_size;
1195 struct e1000_hw *hw = &adapter->hw;
1196
1197 if (adapter->vlgrp)
1198 max_frame_size += VLAN_TAG_SIZE;
1199
1200 e1000_rlpml_set_vf(hw, max_frame_size);
1201}
1202
1203static void igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1204{
1205 struct igbvf_adapter *adapter = netdev_priv(netdev);
1206 struct e1000_hw *hw = &adapter->hw;
1207
1208 if (hw->mac.ops.set_vfta(hw, vid, true))
1209 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1210}
1211
1212static void igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1213{
1214 struct igbvf_adapter *adapter = netdev_priv(netdev);
1215 struct e1000_hw *hw = &adapter->hw;
1216
1217 igbvf_irq_disable(adapter);
1218 vlan_group_set_device(adapter->vlgrp, vid, NULL);
1219
1220 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1221 igbvf_irq_enable(adapter);
1222
1223 if (hw->mac.ops.set_vfta(hw, vid, false))
1224 dev_err(&adapter->pdev->dev,
1225 "Failed to remove vlan id %d\n", vid);
1226}
1227
1228static void igbvf_vlan_rx_register(struct net_device *netdev,
1229 struct vlan_group *grp)
1230{
1231 struct igbvf_adapter *adapter = netdev_priv(netdev);
1232
1233 adapter->vlgrp = grp;
1234}
1235
1236static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1237{
1238 u16 vid;
1239
1240 if (!adapter->vlgrp)
1241 return;
1242
1243 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1244 if (!vlan_group_get_device(adapter->vlgrp, vid))
1245 continue;
1246 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1247 }
1248
1249 igbvf_set_rlpml(adapter);
1250}
1251
1252/**
1253 * igbvf_configure_tx - Configure Transmit Unit after Reset
1254 * @adapter: board private structure
1255 *
1256 * Configure the Tx unit of the MAC after a reset.
1257 **/
1258static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1259{
1260 struct e1000_hw *hw = &adapter->hw;
1261 struct igbvf_ring *tx_ring = adapter->tx_ring;
1262 u64 tdba;
1263 u32 txdctl, dca_txctrl;
1264
1265 /* disable transmits */
1266 txdctl = er32(TXDCTL(0));
1267 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1268 msleep(10);
1269
1270 /* Setup the HW Tx Head and Tail descriptor pointers */
1271 ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1272 tdba = tx_ring->dma;
Andrew Morton8e20ce92009-06-18 16:49:17 -07001273 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001274 ew32(TDBAH(0), (tdba >> 32));
1275 ew32(TDH(0), 0);
1276 ew32(TDT(0), 0);
1277 tx_ring->head = E1000_TDH(0);
1278 tx_ring->tail = E1000_TDT(0);
1279
1280 /* Turn off Relaxed Ordering on head write-backs. The writebacks
1281 * MUST be delivered in order or it will completely screw up
1282 * our bookeeping.
1283 */
1284 dca_txctrl = er32(DCA_TXCTRL(0));
1285 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1286 ew32(DCA_TXCTRL(0), dca_txctrl);
1287
1288 /* enable transmits */
1289 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1290 ew32(TXDCTL(0), txdctl);
1291
1292 /* Setup Transmit Descriptor Settings for eop descriptor */
1293 adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1294
1295 /* enable Report Status bit */
1296 adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1297
1298 adapter->tx_queue_len = adapter->netdev->tx_queue_len;
1299}
1300
1301/**
1302 * igbvf_setup_srrctl - configure the receive control registers
1303 * @adapter: Board private structure
1304 **/
1305static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1306{
1307 struct e1000_hw *hw = &adapter->hw;
1308 u32 srrctl = 0;
1309
1310 srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1311 E1000_SRRCTL_BSIZEHDR_MASK |
1312 E1000_SRRCTL_BSIZEPKT_MASK);
1313
1314 /* Enable queue drop to avoid head of line blocking */
1315 srrctl |= E1000_SRRCTL_DROP_EN;
1316
1317 /* Setup buffer sizes */
1318 srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1319 E1000_SRRCTL_BSIZEPKT_SHIFT;
1320
1321 if (adapter->rx_buffer_len < 2048) {
1322 adapter->rx_ps_hdr_size = 0;
1323 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1324 } else {
1325 adapter->rx_ps_hdr_size = 128;
1326 srrctl |= adapter->rx_ps_hdr_size <<
1327 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1328 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1329 }
1330
1331 ew32(SRRCTL(0), srrctl);
1332}
1333
1334/**
1335 * igbvf_configure_rx - Configure Receive Unit after Reset
1336 * @adapter: board private structure
1337 *
1338 * Configure the Rx unit of the MAC after a reset.
1339 **/
1340static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1341{
1342 struct e1000_hw *hw = &adapter->hw;
1343 struct igbvf_ring *rx_ring = adapter->rx_ring;
1344 u64 rdba;
1345 u32 rdlen, rxdctl;
1346
1347 /* disable receives */
1348 rxdctl = er32(RXDCTL(0));
1349 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1350 msleep(10);
1351
1352 rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1353
1354 /*
1355 * Setup the HW Rx Head and Tail Descriptor Pointers and
1356 * the Base and Length of the Rx Descriptor Ring
1357 */
1358 rdba = rx_ring->dma;
Andrew Morton8e20ce92009-06-18 16:49:17 -07001359 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001360 ew32(RDBAH(0), (rdba >> 32));
1361 ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1362 rx_ring->head = E1000_RDH(0);
1363 rx_ring->tail = E1000_RDT(0);
1364 ew32(RDH(0), 0);
1365 ew32(RDT(0), 0);
1366
1367 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1368 rxdctl &= 0xFFF00000;
1369 rxdctl |= IGBVF_RX_PTHRESH;
1370 rxdctl |= IGBVF_RX_HTHRESH << 8;
1371 rxdctl |= IGBVF_RX_WTHRESH << 16;
1372
1373 igbvf_set_rlpml(adapter);
1374
1375 /* enable receives */
1376 ew32(RXDCTL(0), rxdctl);
1377}
1378
1379/**
1380 * igbvf_set_multi - Multicast and Promiscuous mode set
1381 * @netdev: network interface device structure
1382 *
1383 * The set_multi entry point is called whenever the multicast address
1384 * list or the network interface flags are updated. This routine is
1385 * responsible for configuring the hardware for proper multicast,
1386 * promiscuous mode, and all-multi behavior.
1387 **/
1388static void igbvf_set_multi(struct net_device *netdev)
1389{
1390 struct igbvf_adapter *adapter = netdev_priv(netdev);
1391 struct e1000_hw *hw = &adapter->hw;
1392 struct dev_mc_list *mc_ptr;
1393 u8 *mta_list = NULL;
1394 int i;
1395
1396 if (netdev->mc_count) {
1397 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
1398 if (!mta_list) {
1399 dev_err(&adapter->pdev->dev,
1400 "failed to allocate multicast filter list\n");
1401 return;
1402 }
1403 }
1404
1405 /* prepare a packed array of only addresses. */
1406 mc_ptr = netdev->mc_list;
1407
1408 for (i = 0; i < netdev->mc_count; i++) {
1409 if (!mc_ptr)
1410 break;
1411 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
1412 ETH_ALEN);
1413 mc_ptr = mc_ptr->next;
1414 }
1415
1416 hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1417 kfree(mta_list);
1418}
1419
1420/**
1421 * igbvf_configure - configure the hardware for Rx and Tx
1422 * @adapter: private board structure
1423 **/
1424static void igbvf_configure(struct igbvf_adapter *adapter)
1425{
1426 igbvf_set_multi(adapter->netdev);
1427
1428 igbvf_restore_vlan(adapter);
1429
1430 igbvf_configure_tx(adapter);
1431 igbvf_setup_srrctl(adapter);
1432 igbvf_configure_rx(adapter);
1433 igbvf_alloc_rx_buffers(adapter->rx_ring,
1434 igbvf_desc_unused(adapter->rx_ring));
1435}
1436
1437/* igbvf_reset - bring the hardware into a known good state
1438 *
1439 * This function boots the hardware and enables some settings that
1440 * require a configuration cycle of the hardware - those cannot be
1441 * set/changed during runtime. After reset the device needs to be
1442 * properly configured for Rx, Tx etc.
1443 */
Alexander Duyck2d165772009-04-09 22:49:20 +00001444static void igbvf_reset(struct igbvf_adapter *adapter)
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001445{
1446 struct e1000_mac_info *mac = &adapter->hw.mac;
1447 struct net_device *netdev = adapter->netdev;
1448 struct e1000_hw *hw = &adapter->hw;
1449
1450 /* Allow time for pending master requests to run */
1451 if (mac->ops.reset_hw(hw))
1452 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1453
1454 mac->ops.init_hw(hw);
1455
1456 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1457 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1458 netdev->addr_len);
1459 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1460 netdev->addr_len);
1461 }
1462}
1463
1464int igbvf_up(struct igbvf_adapter *adapter)
1465{
1466 struct e1000_hw *hw = &adapter->hw;
1467
1468 /* hardware has been reset, we need to reload some things */
1469 igbvf_configure(adapter);
1470
1471 clear_bit(__IGBVF_DOWN, &adapter->state);
1472
1473 napi_enable(&adapter->rx_ring->napi);
1474 if (adapter->msix_entries)
1475 igbvf_configure_msix(adapter);
1476
1477 /* Clear any pending interrupts. */
1478 er32(EICR);
1479 igbvf_irq_enable(adapter);
1480
1481 /* start the watchdog */
1482 hw->mac.get_link_status = 1;
1483 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1484
1485
1486 return 0;
1487}
1488
1489void igbvf_down(struct igbvf_adapter *adapter)
1490{
1491 struct net_device *netdev = adapter->netdev;
1492 struct e1000_hw *hw = &adapter->hw;
1493 u32 rxdctl, txdctl;
1494
1495 /*
1496 * signal that we're down so the interrupt handler does not
1497 * reschedule our watchdog timer
1498 */
1499 set_bit(__IGBVF_DOWN, &adapter->state);
1500
1501 /* disable receives in the hardware */
1502 rxdctl = er32(RXDCTL(0));
1503 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1504
1505 netif_stop_queue(netdev);
1506
1507 /* disable transmits in the hardware */
1508 txdctl = er32(TXDCTL(0));
1509 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1510
1511 /* flush both disables and wait for them to finish */
1512 e1e_flush();
1513 msleep(10);
1514
1515 napi_disable(&adapter->rx_ring->napi);
1516
1517 igbvf_irq_disable(adapter);
1518
1519 del_timer_sync(&adapter->watchdog_timer);
1520
1521 netdev->tx_queue_len = adapter->tx_queue_len;
1522 netif_carrier_off(netdev);
1523
1524 /* record the stats before reset*/
1525 igbvf_update_stats(adapter);
1526
1527 adapter->link_speed = 0;
1528 adapter->link_duplex = 0;
1529
1530 igbvf_reset(adapter);
1531 igbvf_clean_tx_ring(adapter->tx_ring);
1532 igbvf_clean_rx_ring(adapter->rx_ring);
1533}
1534
1535void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1536{
1537 might_sleep();
1538 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1539 msleep(1);
1540 igbvf_down(adapter);
1541 igbvf_up(adapter);
1542 clear_bit(__IGBVF_RESETTING, &adapter->state);
1543}
1544
1545/**
1546 * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1547 * @adapter: board private structure to initialize
1548 *
1549 * igbvf_sw_init initializes the Adapter private data structure.
1550 * Fields are initialized based on PCI device information and
1551 * OS network device settings (MTU size).
1552 **/
1553static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1554{
1555 struct net_device *netdev = adapter->netdev;
1556 s32 rc;
1557
1558 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1559 adapter->rx_ps_hdr_size = 0;
1560 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1561 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1562
1563 adapter->tx_int_delay = 8;
1564 adapter->tx_abs_int_delay = 32;
1565 adapter->rx_int_delay = 0;
1566 adapter->rx_abs_int_delay = 8;
1567 adapter->itr_setting = 3;
1568 adapter->itr = 20000;
1569
1570 /* Set various function pointers */
1571 adapter->ei->init_ops(&adapter->hw);
1572
1573 rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1574 if (rc)
1575 return rc;
1576
1577 rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1578 if (rc)
1579 return rc;
1580
1581 igbvf_set_interrupt_capability(adapter);
1582
1583 if (igbvf_alloc_queues(adapter))
1584 return -ENOMEM;
1585
1586 spin_lock_init(&adapter->tx_queue_lock);
1587
1588 /* Explicitly disable IRQ since the NIC can be in any state. */
1589 igbvf_irq_disable(adapter);
1590
1591 spin_lock_init(&adapter->stats_lock);
1592
1593 set_bit(__IGBVF_DOWN, &adapter->state);
1594 return 0;
1595}
1596
1597static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1598{
1599 struct e1000_hw *hw = &adapter->hw;
1600
1601 adapter->stats.last_gprc = er32(VFGPRC);
1602 adapter->stats.last_gorc = er32(VFGORC);
1603 adapter->stats.last_gptc = er32(VFGPTC);
1604 adapter->stats.last_gotc = er32(VFGOTC);
1605 adapter->stats.last_mprc = er32(VFMPRC);
1606 adapter->stats.last_gotlbc = er32(VFGOTLBC);
1607 adapter->stats.last_gptlbc = er32(VFGPTLBC);
1608 adapter->stats.last_gorlbc = er32(VFGORLBC);
1609 adapter->stats.last_gprlbc = er32(VFGPRLBC);
1610
1611 adapter->stats.base_gprc = er32(VFGPRC);
1612 adapter->stats.base_gorc = er32(VFGORC);
1613 adapter->stats.base_gptc = er32(VFGPTC);
1614 adapter->stats.base_gotc = er32(VFGOTC);
1615 adapter->stats.base_mprc = er32(VFMPRC);
1616 adapter->stats.base_gotlbc = er32(VFGOTLBC);
1617 adapter->stats.base_gptlbc = er32(VFGPTLBC);
1618 adapter->stats.base_gorlbc = er32(VFGORLBC);
1619 adapter->stats.base_gprlbc = er32(VFGPRLBC);
1620}
1621
1622/**
1623 * igbvf_open - Called when a network interface is made active
1624 * @netdev: network interface device structure
1625 *
1626 * Returns 0 on success, negative value on failure
1627 *
1628 * The open entry point is called when a network interface is made
1629 * active by the system (IFF_UP). At this point all resources needed
1630 * for transmit and receive operations are allocated, the interrupt
1631 * handler is registered with the OS, the watchdog timer is started,
1632 * and the stack is notified that the interface is ready.
1633 **/
1634static int igbvf_open(struct net_device *netdev)
1635{
1636 struct igbvf_adapter *adapter = netdev_priv(netdev);
1637 struct e1000_hw *hw = &adapter->hw;
1638 int err;
1639
1640 /* disallow open during test */
1641 if (test_bit(__IGBVF_TESTING, &adapter->state))
1642 return -EBUSY;
1643
1644 /* allocate transmit descriptors */
1645 err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1646 if (err)
1647 goto err_setup_tx;
1648
1649 /* allocate receive descriptors */
1650 err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1651 if (err)
1652 goto err_setup_rx;
1653
1654 /*
1655 * before we allocate an interrupt, we must be ready to handle it.
1656 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1657 * as soon as we call pci_request_irq, so we have to setup our
1658 * clean_rx handler before we do so.
1659 */
1660 igbvf_configure(adapter);
1661
1662 err = igbvf_request_irq(adapter);
1663 if (err)
1664 goto err_req_irq;
1665
1666 /* From here on the code is the same as igbvf_up() */
1667 clear_bit(__IGBVF_DOWN, &adapter->state);
1668
1669 napi_enable(&adapter->rx_ring->napi);
1670
1671 /* clear any pending interrupts */
1672 er32(EICR);
1673
1674 igbvf_irq_enable(adapter);
1675
1676 /* start the watchdog */
1677 hw->mac.get_link_status = 1;
1678 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1679
1680 return 0;
1681
1682err_req_irq:
1683 igbvf_free_rx_resources(adapter->rx_ring);
1684err_setup_rx:
1685 igbvf_free_tx_resources(adapter->tx_ring);
1686err_setup_tx:
1687 igbvf_reset(adapter);
1688
1689 return err;
1690}
1691
1692/**
1693 * igbvf_close - Disables a network interface
1694 * @netdev: network interface device structure
1695 *
1696 * Returns 0, this is not allowed to fail
1697 *
1698 * The close entry point is called when an interface is de-activated
1699 * by the OS. The hardware is still under the drivers control, but
1700 * needs to be disabled. A global MAC reset is issued to stop the
1701 * hardware, and all transmit and receive resources are freed.
1702 **/
1703static int igbvf_close(struct net_device *netdev)
1704{
1705 struct igbvf_adapter *adapter = netdev_priv(netdev);
1706
1707 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1708 igbvf_down(adapter);
1709
1710 igbvf_free_irq(adapter);
1711
1712 igbvf_free_tx_resources(adapter->tx_ring);
1713 igbvf_free_rx_resources(adapter->rx_ring);
1714
1715 return 0;
1716}
1717/**
1718 * igbvf_set_mac - Change the Ethernet Address of the NIC
1719 * @netdev: network interface device structure
1720 * @p: pointer to an address structure
1721 *
1722 * Returns 0 on success, negative on failure
1723 **/
1724static int igbvf_set_mac(struct net_device *netdev, void *p)
1725{
1726 struct igbvf_adapter *adapter = netdev_priv(netdev);
1727 struct e1000_hw *hw = &adapter->hw;
1728 struct sockaddr *addr = p;
1729
1730 if (!is_valid_ether_addr(addr->sa_data))
1731 return -EADDRNOTAVAIL;
1732
1733 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1734
1735 hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1736
1737 if (memcmp(addr->sa_data, hw->mac.addr, 6))
1738 return -EADDRNOTAVAIL;
1739
1740 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1741
1742 return 0;
1743}
1744
1745#define UPDATE_VF_COUNTER(reg, name) \
1746 { \
1747 u32 current_counter = er32(reg); \
1748 if (current_counter < adapter->stats.last_##name) \
1749 adapter->stats.name += 0x100000000LL; \
1750 adapter->stats.last_##name = current_counter; \
1751 adapter->stats.name &= 0xFFFFFFFF00000000LL; \
1752 adapter->stats.name |= current_counter; \
1753 }
1754
1755/**
1756 * igbvf_update_stats - Update the board statistics counters
1757 * @adapter: board private structure
1758**/
1759void igbvf_update_stats(struct igbvf_adapter *adapter)
1760{
1761 struct e1000_hw *hw = &adapter->hw;
1762 struct pci_dev *pdev = adapter->pdev;
1763
1764 /*
1765 * Prevent stats update while adapter is being reset, link is down
1766 * or if the pci connection is down.
1767 */
1768 if (adapter->link_speed == 0)
1769 return;
1770
1771 if (test_bit(__IGBVF_RESETTING, &adapter->state))
1772 return;
1773
1774 if (pci_channel_offline(pdev))
1775 return;
1776
1777 UPDATE_VF_COUNTER(VFGPRC, gprc);
1778 UPDATE_VF_COUNTER(VFGORC, gorc);
1779 UPDATE_VF_COUNTER(VFGPTC, gptc);
1780 UPDATE_VF_COUNTER(VFGOTC, gotc);
1781 UPDATE_VF_COUNTER(VFMPRC, mprc);
1782 UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1783 UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1784 UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1785 UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1786
1787 /* Fill out the OS statistics structure */
1788 adapter->net_stats.multicast = adapter->stats.mprc;
1789}
1790
1791static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1792{
1793 dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s\n",
1794 adapter->link_speed,
1795 ((adapter->link_duplex == FULL_DUPLEX) ?
1796 "Full Duplex" : "Half Duplex"));
1797}
1798
1799static bool igbvf_has_link(struct igbvf_adapter *adapter)
1800{
1801 struct e1000_hw *hw = &adapter->hw;
1802 s32 ret_val = E1000_SUCCESS;
1803 bool link_active;
1804
1805 ret_val = hw->mac.ops.check_for_link(hw);
1806 link_active = !hw->mac.get_link_status;
1807
1808 /* if check for link returns error we will need to reset */
1809 if (ret_val)
1810 schedule_work(&adapter->reset_task);
1811
1812 return link_active;
1813}
1814
1815/**
1816 * igbvf_watchdog - Timer Call-back
1817 * @data: pointer to adapter cast into an unsigned long
1818 **/
1819static void igbvf_watchdog(unsigned long data)
1820{
1821 struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1822
1823 /* Do the rest outside of interrupt context */
1824 schedule_work(&adapter->watchdog_task);
1825}
1826
1827static void igbvf_watchdog_task(struct work_struct *work)
1828{
1829 struct igbvf_adapter *adapter = container_of(work,
1830 struct igbvf_adapter,
1831 watchdog_task);
1832 struct net_device *netdev = adapter->netdev;
1833 struct e1000_mac_info *mac = &adapter->hw.mac;
1834 struct igbvf_ring *tx_ring = adapter->tx_ring;
1835 struct e1000_hw *hw = &adapter->hw;
1836 u32 link;
1837 int tx_pending = 0;
1838
1839 link = igbvf_has_link(adapter);
1840
1841 if (link) {
1842 if (!netif_carrier_ok(netdev)) {
1843 bool txb2b = 1;
1844
1845 mac->ops.get_link_up_info(&adapter->hw,
1846 &adapter->link_speed,
1847 &adapter->link_duplex);
1848 igbvf_print_link_info(adapter);
1849
1850 /*
1851 * tweak tx_queue_len according to speed/duplex
1852 * and adjust the timeout factor
1853 */
1854 netdev->tx_queue_len = adapter->tx_queue_len;
1855 adapter->tx_timeout_factor = 1;
1856 switch (adapter->link_speed) {
1857 case SPEED_10:
1858 txb2b = 0;
1859 netdev->tx_queue_len = 10;
1860 adapter->tx_timeout_factor = 16;
1861 break;
1862 case SPEED_100:
1863 txb2b = 0;
1864 netdev->tx_queue_len = 100;
1865 /* maybe add some timeout factor ? */
1866 break;
1867 }
1868
1869 netif_carrier_on(netdev);
1870 netif_wake_queue(netdev);
1871 }
1872 } else {
1873 if (netif_carrier_ok(netdev)) {
1874 adapter->link_speed = 0;
1875 adapter->link_duplex = 0;
1876 dev_info(&adapter->pdev->dev, "Link is Down\n");
1877 netif_carrier_off(netdev);
1878 netif_stop_queue(netdev);
1879 }
1880 }
1881
1882 if (netif_carrier_ok(netdev)) {
1883 igbvf_update_stats(adapter);
1884 } else {
1885 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1886 tx_ring->count);
1887 if (tx_pending) {
1888 /*
1889 * We've lost link, so the controller stops DMA,
1890 * but we've got queued Tx work that's never going
1891 * to get done, so reset controller to flush Tx.
1892 * (Do the reset outside of interrupt context).
1893 */
1894 adapter->tx_timeout_count++;
1895 schedule_work(&adapter->reset_task);
1896 }
1897 }
1898
1899 /* Cause software interrupt to ensure Rx ring is cleaned */
1900 ew32(EICS, adapter->rx_ring->eims_value);
1901
1902 /* Force detection of hung controller every watchdog period */
1903 adapter->detect_tx_hung = 1;
1904
1905 /* Reset the timer */
1906 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1907 mod_timer(&adapter->watchdog_timer,
1908 round_jiffies(jiffies + (2 * HZ)));
1909}
1910
1911#define IGBVF_TX_FLAGS_CSUM 0x00000001
1912#define IGBVF_TX_FLAGS_VLAN 0x00000002
1913#define IGBVF_TX_FLAGS_TSO 0x00000004
1914#define IGBVF_TX_FLAGS_IPV4 0x00000008
1915#define IGBVF_TX_FLAGS_VLAN_MASK 0xffff0000
1916#define IGBVF_TX_FLAGS_VLAN_SHIFT 16
1917
1918static int igbvf_tso(struct igbvf_adapter *adapter,
1919 struct igbvf_ring *tx_ring,
1920 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1921{
1922 struct e1000_adv_tx_context_desc *context_desc;
1923 unsigned int i;
1924 int err;
1925 struct igbvf_buffer *buffer_info;
1926 u32 info = 0, tu_cmd = 0;
1927 u32 mss_l4len_idx, l4len;
1928 *hdr_len = 0;
1929
1930 if (skb_header_cloned(skb)) {
1931 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1932 if (err) {
1933 dev_err(&adapter->pdev->dev,
1934 "igbvf_tso returning an error\n");
1935 return err;
1936 }
1937 }
1938
1939 l4len = tcp_hdrlen(skb);
1940 *hdr_len += l4len;
1941
1942 if (skb->protocol == htons(ETH_P_IP)) {
1943 struct iphdr *iph = ip_hdr(skb);
1944 iph->tot_len = 0;
1945 iph->check = 0;
1946 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1947 iph->daddr, 0,
1948 IPPROTO_TCP,
1949 0);
1950 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
1951 ipv6_hdr(skb)->payload_len = 0;
1952 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1953 &ipv6_hdr(skb)->daddr,
1954 0, IPPROTO_TCP, 0);
1955 }
1956
1957 i = tx_ring->next_to_use;
1958
1959 buffer_info = &tx_ring->buffer_info[i];
1960 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1961 /* VLAN MACLEN IPLEN */
1962 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1963 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1964 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1965 *hdr_len += skb_network_offset(skb);
1966 info |= (skb_transport_header(skb) - skb_network_header(skb));
1967 *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1968 context_desc->vlan_macip_lens = cpu_to_le32(info);
1969
1970 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1971 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1972
1973 if (skb->protocol == htons(ETH_P_IP))
1974 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1975 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1976
1977 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1978
1979 /* MSS L4LEN IDX */
1980 mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1981 mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1982
1983 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1984 context_desc->seqnum_seed = 0;
1985
1986 buffer_info->time_stamp = jiffies;
1987 buffer_info->next_to_watch = i;
1988 buffer_info->dma = 0;
1989 i++;
1990 if (i == tx_ring->count)
1991 i = 0;
1992
1993 tx_ring->next_to_use = i;
1994
1995 return true;
1996}
1997
1998static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
1999 struct igbvf_ring *tx_ring,
2000 struct sk_buff *skb, u32 tx_flags)
2001{
2002 struct e1000_adv_tx_context_desc *context_desc;
2003 unsigned int i;
2004 struct igbvf_buffer *buffer_info;
2005 u32 info = 0, tu_cmd = 0;
2006
2007 if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2008 (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
2009 i = tx_ring->next_to_use;
2010 buffer_info = &tx_ring->buffer_info[i];
2011 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
2012
2013 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2014 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
2015
2016 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2017 if (skb->ip_summed == CHECKSUM_PARTIAL)
2018 info |= (skb_transport_header(skb) -
2019 skb_network_header(skb));
2020
2021
2022 context_desc->vlan_macip_lens = cpu_to_le32(info);
2023
2024 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2025
2026 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2027 switch (skb->protocol) {
2028 case __constant_htons(ETH_P_IP):
2029 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2030 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2031 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2032 break;
2033 case __constant_htons(ETH_P_IPV6):
2034 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2035 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2036 break;
2037 default:
2038 break;
2039 }
2040 }
2041
2042 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2043 context_desc->seqnum_seed = 0;
2044 context_desc->mss_l4len_idx = 0;
2045
2046 buffer_info->time_stamp = jiffies;
2047 buffer_info->next_to_watch = i;
2048 buffer_info->dma = 0;
2049 i++;
2050 if (i == tx_ring->count)
2051 i = 0;
2052 tx_ring->next_to_use = i;
2053
2054 return true;
2055 }
2056
2057 return false;
2058}
2059
2060static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2061{
2062 struct igbvf_adapter *adapter = netdev_priv(netdev);
2063
2064 /* there is enough descriptors then we don't need to worry */
2065 if (igbvf_desc_unused(adapter->tx_ring) >= size)
2066 return 0;
2067
2068 netif_stop_queue(netdev);
2069
2070 smp_mb();
2071
2072 /* We need to check again just in case room has been made available */
2073 if (igbvf_desc_unused(adapter->tx_ring) < size)
2074 return -EBUSY;
2075
2076 netif_wake_queue(netdev);
2077
2078 ++adapter->restart_queue;
2079 return 0;
2080}
2081
2082#define IGBVF_MAX_TXD_PWR 16
2083#define IGBVF_MAX_DATA_PER_TXD (1 << IGBVF_MAX_TXD_PWR)
2084
2085static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2086 struct igbvf_ring *tx_ring,
2087 struct sk_buff *skb,
2088 unsigned int first)
2089{
2090 struct igbvf_buffer *buffer_info;
2091 unsigned int len = skb_headlen(skb);
2092 unsigned int count = 0, i;
2093 unsigned int f;
2094 dma_addr_t *map;
2095
2096 i = tx_ring->next_to_use;
2097
2098 if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
2099 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
2100 return 0;
2101 }
2102
2103 map = skb_shinfo(skb)->dma_maps;
2104
2105 buffer_info = &tx_ring->buffer_info[i];
2106 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2107 buffer_info->length = len;
2108 /* set time_stamp *before* dma to help avoid a possible race */
2109 buffer_info->time_stamp = jiffies;
2110 buffer_info->next_to_watch = i;
Eric Dumazet042a53a2009-06-05 04:04:16 +00002111 buffer_info->dma = skb_shinfo(skb)->dma_head;
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002112
2113 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2114 struct skb_frag_struct *frag;
2115
2116 i++;
2117 if (i == tx_ring->count)
2118 i = 0;
2119
2120 frag = &skb_shinfo(skb)->frags[f];
2121 len = frag->size;
2122
2123 buffer_info = &tx_ring->buffer_info[i];
2124 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2125 buffer_info->length = len;
2126 buffer_info->time_stamp = jiffies;
2127 buffer_info->next_to_watch = i;
2128 buffer_info->dma = map[count];
2129 count++;
2130 }
2131
2132 tx_ring->buffer_info[i].skb = skb;
2133 tx_ring->buffer_info[first].next_to_watch = i;
2134
Eric Dumazet042a53a2009-06-05 04:04:16 +00002135 return count + 1;
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002136}
2137
2138static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2139 struct igbvf_ring *tx_ring,
2140 int tx_flags, int count, u32 paylen,
2141 u8 hdr_len)
2142{
2143 union e1000_adv_tx_desc *tx_desc = NULL;
2144 struct igbvf_buffer *buffer_info;
2145 u32 olinfo_status = 0, cmd_type_len;
2146 unsigned int i;
2147
2148 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2149 E1000_ADVTXD_DCMD_DEXT);
2150
2151 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2152 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2153
2154 if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2155 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2156
2157 /* insert tcp checksum */
2158 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2159
2160 /* insert ip checksum */
2161 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2162 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2163
2164 } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2165 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2166 }
2167
2168 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2169
2170 i = tx_ring->next_to_use;
2171 while (count--) {
2172 buffer_info = &tx_ring->buffer_info[i];
2173 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2174 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2175 tx_desc->read.cmd_type_len =
2176 cpu_to_le32(cmd_type_len | buffer_info->length);
2177 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2178 i++;
2179 if (i == tx_ring->count)
2180 i = 0;
2181 }
2182
2183 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2184 /* Force memory writes to complete before letting h/w
2185 * know there are new descriptors to fetch. (Only
2186 * applicable for weak-ordered memory model archs,
2187 * such as IA-64). */
2188 wmb();
2189
2190 tx_ring->next_to_use = i;
2191 writel(i, adapter->hw.hw_addr + tx_ring->tail);
2192 /* we need this if more than one processor can write to our tail
2193 * at a time, it syncronizes IO on IA64/Altix systems */
2194 mmiowb();
2195}
2196
Stephen Hemminger3b29a562009-08-31 19:50:55 +00002197static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2198 struct net_device *netdev,
2199 struct igbvf_ring *tx_ring)
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002200{
2201 struct igbvf_adapter *adapter = netdev_priv(netdev);
2202 unsigned int first, tx_flags = 0;
2203 u8 hdr_len = 0;
2204 int count = 0;
2205 int tso = 0;
2206
2207 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2208 dev_kfree_skb_any(skb);
2209 return NETDEV_TX_OK;
2210 }
2211
2212 if (skb->len <= 0) {
2213 dev_kfree_skb_any(skb);
2214 return NETDEV_TX_OK;
2215 }
2216
2217 /*
2218 * need: count + 4 desc gap to keep tail from touching
2219 * + 2 desc gap to keep tail from touching head,
2220 * + 1 desc for skb->data,
2221 * + 1 desc for context descriptor,
2222 * head, otherwise try next time
2223 */
2224 if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2225 /* this is a hard error */
2226 return NETDEV_TX_BUSY;
2227 }
2228
2229 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2230 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2231 tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2232 }
2233
2234 if (skb->protocol == htons(ETH_P_IP))
2235 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2236
2237 first = tx_ring->next_to_use;
2238
2239 tso = skb_is_gso(skb) ?
2240 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2241 if (unlikely(tso < 0)) {
2242 dev_kfree_skb_any(skb);
2243 return NETDEV_TX_OK;
2244 }
2245
2246 if (tso)
2247 tx_flags |= IGBVF_TX_FLAGS_TSO;
2248 else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2249 (skb->ip_summed == CHECKSUM_PARTIAL))
2250 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2251
2252 /*
2253 * count reflects descriptors mapped, if 0 then mapping error
2254 * has occured and we need to rewind the descriptor queue
2255 */
2256 count = igbvf_tx_map_adv(adapter, tx_ring, skb, first);
2257
2258 if (count) {
2259 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2260 skb->len, hdr_len);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002261 /* Make sure there is space in the ring for the next send. */
2262 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2263 } else {
2264 dev_kfree_skb_any(skb);
2265 tx_ring->buffer_info[first].time_stamp = 0;
2266 tx_ring->next_to_use = first;
2267 }
2268
2269 return NETDEV_TX_OK;
2270}
2271
Stephen Hemminger3b29a562009-08-31 19:50:55 +00002272static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2273 struct net_device *netdev)
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002274{
2275 struct igbvf_adapter *adapter = netdev_priv(netdev);
2276 struct igbvf_ring *tx_ring;
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002277
2278 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2279 dev_kfree_skb_any(skb);
2280 return NETDEV_TX_OK;
2281 }
2282
2283 tx_ring = &adapter->tx_ring[0];
2284
Stephen Hemminger3b29a562009-08-31 19:50:55 +00002285 return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002286}
2287
2288/**
2289 * igbvf_tx_timeout - Respond to a Tx Hang
2290 * @netdev: network interface device structure
2291 **/
2292static void igbvf_tx_timeout(struct net_device *netdev)
2293{
2294 struct igbvf_adapter *adapter = netdev_priv(netdev);
2295
2296 /* Do the reset outside of interrupt context */
2297 adapter->tx_timeout_count++;
2298 schedule_work(&adapter->reset_task);
2299}
2300
2301static void igbvf_reset_task(struct work_struct *work)
2302{
2303 struct igbvf_adapter *adapter;
2304 adapter = container_of(work, struct igbvf_adapter, reset_task);
2305
2306 igbvf_reinit_locked(adapter);
2307}
2308
2309/**
2310 * igbvf_get_stats - Get System Network Statistics
2311 * @netdev: network interface device structure
2312 *
2313 * Returns the address of the device statistics structure.
2314 * The statistics are actually updated from the timer callback.
2315 **/
2316static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2317{
2318 struct igbvf_adapter *adapter = netdev_priv(netdev);
2319
2320 /* only return the current stats */
2321 return &adapter->net_stats;
2322}
2323
2324/**
2325 * igbvf_change_mtu - Change the Maximum Transfer Unit
2326 * @netdev: network interface device structure
2327 * @new_mtu: new value for maximum frame size
2328 *
2329 * Returns 0 on success, negative on failure
2330 **/
2331static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2332{
2333 struct igbvf_adapter *adapter = netdev_priv(netdev);
2334 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2335
2336 if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2337 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2338 return -EINVAL;
2339 }
2340
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002341#define MAX_STD_JUMBO_FRAME_SIZE 9234
2342 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2343 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2344 return -EINVAL;
2345 }
2346
2347 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2348 msleep(1);
2349 /* igbvf_down has a dependency on max_frame_size */
2350 adapter->max_frame_size = max_frame;
2351 if (netif_running(netdev))
2352 igbvf_down(adapter);
2353
2354 /*
2355 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2356 * means we reserve 2 more, this pushes us to allocate from the next
2357 * larger slab size.
2358 * i.e. RXBUFFER_2048 --> size-4096 slab
2359 * However with the new *_jumbo_rx* routines, jumbo receives will use
2360 * fragmented skbs
2361 */
2362
2363 if (max_frame <= 1024)
2364 adapter->rx_buffer_len = 1024;
2365 else if (max_frame <= 2048)
2366 adapter->rx_buffer_len = 2048;
2367 else
2368#if (PAGE_SIZE / 2) > 16384
2369 adapter->rx_buffer_len = 16384;
2370#else
2371 adapter->rx_buffer_len = PAGE_SIZE / 2;
2372#endif
2373
2374
2375 /* adjust allocation if LPE protects us, and we aren't using SBP */
2376 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2377 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2378 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2379 ETH_FCS_LEN;
2380
2381 dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2382 netdev->mtu, new_mtu);
2383 netdev->mtu = new_mtu;
2384
2385 if (netif_running(netdev))
2386 igbvf_up(adapter);
2387 else
2388 igbvf_reset(adapter);
2389
2390 clear_bit(__IGBVF_RESETTING, &adapter->state);
2391
2392 return 0;
2393}
2394
2395static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2396{
2397 switch (cmd) {
2398 default:
2399 return -EOPNOTSUPP;
2400 }
2401}
2402
2403static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2404{
2405 struct net_device *netdev = pci_get_drvdata(pdev);
2406 struct igbvf_adapter *adapter = netdev_priv(netdev);
2407#ifdef CONFIG_PM
2408 int retval = 0;
2409#endif
2410
2411 netif_device_detach(netdev);
2412
2413 if (netif_running(netdev)) {
2414 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2415 igbvf_down(adapter);
2416 igbvf_free_irq(adapter);
2417 }
2418
2419#ifdef CONFIG_PM
2420 retval = pci_save_state(pdev);
2421 if (retval)
2422 return retval;
2423#endif
2424
2425 pci_disable_device(pdev);
2426
2427 return 0;
2428}
2429
2430#ifdef CONFIG_PM
2431static int igbvf_resume(struct pci_dev *pdev)
2432{
2433 struct net_device *netdev = pci_get_drvdata(pdev);
2434 struct igbvf_adapter *adapter = netdev_priv(netdev);
2435 u32 err;
2436
2437 pci_restore_state(pdev);
2438 err = pci_enable_device_mem(pdev);
2439 if (err) {
2440 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2441 return err;
2442 }
2443
2444 pci_set_master(pdev);
2445
2446 if (netif_running(netdev)) {
2447 err = igbvf_request_irq(adapter);
2448 if (err)
2449 return err;
2450 }
2451
2452 igbvf_reset(adapter);
2453
2454 if (netif_running(netdev))
2455 igbvf_up(adapter);
2456
2457 netif_device_attach(netdev);
2458
2459 return 0;
2460}
2461#endif
2462
2463static void igbvf_shutdown(struct pci_dev *pdev)
2464{
2465 igbvf_suspend(pdev, PMSG_SUSPEND);
2466}
2467
2468#ifdef CONFIG_NET_POLL_CONTROLLER
2469/*
2470 * Polling 'interrupt' - used by things like netconsole to send skbs
2471 * without having to re-enable interrupts. It's not called while
2472 * the interrupt routine is executing.
2473 */
2474static void igbvf_netpoll(struct net_device *netdev)
2475{
2476 struct igbvf_adapter *adapter = netdev_priv(netdev);
2477
2478 disable_irq(adapter->pdev->irq);
2479
2480 igbvf_clean_tx_irq(adapter->tx_ring);
2481
2482 enable_irq(adapter->pdev->irq);
2483}
2484#endif
2485
2486/**
2487 * igbvf_io_error_detected - called when PCI error is detected
2488 * @pdev: Pointer to PCI device
2489 * @state: The current pci connection state
2490 *
2491 * This function is called after a PCI bus error affecting
2492 * this device has been detected.
2493 */
2494static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2495 pci_channel_state_t state)
2496{
2497 struct net_device *netdev = pci_get_drvdata(pdev);
2498 struct igbvf_adapter *adapter = netdev_priv(netdev);
2499
2500 netif_device_detach(netdev);
2501
Dean Nelsonc06c4302009-07-31 09:13:33 +00002502 if (state == pci_channel_io_perm_failure)
2503 return PCI_ERS_RESULT_DISCONNECT;
2504
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002505 if (netif_running(netdev))
2506 igbvf_down(adapter);
2507 pci_disable_device(pdev);
2508
2509 /* Request a slot slot reset. */
2510 return PCI_ERS_RESULT_NEED_RESET;
2511}
2512
2513/**
2514 * igbvf_io_slot_reset - called after the pci bus has been reset.
2515 * @pdev: Pointer to PCI device
2516 *
2517 * Restart the card from scratch, as if from a cold-boot. Implementation
2518 * resembles the first-half of the igbvf_resume routine.
2519 */
2520static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2521{
2522 struct net_device *netdev = pci_get_drvdata(pdev);
2523 struct igbvf_adapter *adapter = netdev_priv(netdev);
2524
2525 if (pci_enable_device_mem(pdev)) {
2526 dev_err(&pdev->dev,
2527 "Cannot re-enable PCI device after reset.\n");
2528 return PCI_ERS_RESULT_DISCONNECT;
2529 }
2530 pci_set_master(pdev);
2531
2532 igbvf_reset(adapter);
2533
2534 return PCI_ERS_RESULT_RECOVERED;
2535}
2536
2537/**
2538 * igbvf_io_resume - called when traffic can start flowing again.
2539 * @pdev: Pointer to PCI device
2540 *
2541 * This callback is called when the error recovery driver tells us that
2542 * its OK to resume normal operation. Implementation resembles the
2543 * second-half of the igbvf_resume routine.
2544 */
2545static void igbvf_io_resume(struct pci_dev *pdev)
2546{
2547 struct net_device *netdev = pci_get_drvdata(pdev);
2548 struct igbvf_adapter *adapter = netdev_priv(netdev);
2549
2550 if (netif_running(netdev)) {
2551 if (igbvf_up(adapter)) {
2552 dev_err(&pdev->dev,
2553 "can't bring device back up after reset\n");
2554 return;
2555 }
2556 }
2557
2558 netif_device_attach(netdev);
2559}
2560
2561static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2562{
2563 struct e1000_hw *hw = &adapter->hw;
2564 struct net_device *netdev = adapter->netdev;
2565 struct pci_dev *pdev = adapter->pdev;
2566
2567 dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2568 dev_info(&pdev->dev, "Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
2569 /* MAC address */
2570 netdev->dev_addr[0], netdev->dev_addr[1],
2571 netdev->dev_addr[2], netdev->dev_addr[3],
2572 netdev->dev_addr[4], netdev->dev_addr[5]);
2573 dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
2574}
2575
2576static const struct net_device_ops igbvf_netdev_ops = {
2577 .ndo_open = igbvf_open,
2578 .ndo_stop = igbvf_close,
2579 .ndo_start_xmit = igbvf_xmit_frame,
2580 .ndo_get_stats = igbvf_get_stats,
2581 .ndo_set_multicast_list = igbvf_set_multi,
2582 .ndo_set_mac_address = igbvf_set_mac,
2583 .ndo_change_mtu = igbvf_change_mtu,
2584 .ndo_do_ioctl = igbvf_ioctl,
2585 .ndo_tx_timeout = igbvf_tx_timeout,
2586 .ndo_vlan_rx_register = igbvf_vlan_rx_register,
2587 .ndo_vlan_rx_add_vid = igbvf_vlan_rx_add_vid,
2588 .ndo_vlan_rx_kill_vid = igbvf_vlan_rx_kill_vid,
2589#ifdef CONFIG_NET_POLL_CONTROLLER
2590 .ndo_poll_controller = igbvf_netpoll,
2591#endif
2592};
2593
2594/**
2595 * igbvf_probe - Device Initialization Routine
2596 * @pdev: PCI device information struct
2597 * @ent: entry in igbvf_pci_tbl
2598 *
2599 * Returns 0 on success, negative on failure
2600 *
2601 * igbvf_probe initializes an adapter identified by a pci_dev structure.
2602 * The OS initialization, configuring of the adapter private structure,
2603 * and a hardware reset occur.
2604 **/
2605static int __devinit igbvf_probe(struct pci_dev *pdev,
2606 const struct pci_device_id *ent)
2607{
2608 struct net_device *netdev;
2609 struct igbvf_adapter *adapter;
2610 struct e1000_hw *hw;
2611 const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2612
2613 static int cards_found;
2614 int err, pci_using_dac;
2615
2616 err = pci_enable_device_mem(pdev);
2617 if (err)
2618 return err;
2619
2620 pci_using_dac = 0;
Andrew Morton8e20ce92009-06-18 16:49:17 -07002621 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002622 if (!err) {
Andrew Morton8e20ce92009-06-18 16:49:17 -07002623 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002624 if (!err)
2625 pci_using_dac = 1;
2626 } else {
Andrew Morton8e20ce92009-06-18 16:49:17 -07002627 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002628 if (err) {
Andrew Morton8e20ce92009-06-18 16:49:17 -07002629 err = pci_set_consistent_dma_mask(pdev,
2630 DMA_BIT_MASK(32));
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002631 if (err) {
2632 dev_err(&pdev->dev, "No usable DMA "
2633 "configuration, aborting\n");
2634 goto err_dma;
2635 }
2636 }
2637 }
2638
2639 err = pci_request_regions(pdev, igbvf_driver_name);
2640 if (err)
2641 goto err_pci_reg;
2642
2643 pci_set_master(pdev);
2644
2645 err = -ENOMEM;
2646 netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2647 if (!netdev)
2648 goto err_alloc_etherdev;
2649
2650 SET_NETDEV_DEV(netdev, &pdev->dev);
2651
2652 pci_set_drvdata(pdev, netdev);
2653 adapter = netdev_priv(netdev);
2654 hw = &adapter->hw;
2655 adapter->netdev = netdev;
2656 adapter->pdev = pdev;
2657 adapter->ei = ei;
2658 adapter->pba = ei->pba;
2659 adapter->flags = ei->flags;
2660 adapter->hw.back = adapter;
2661 adapter->hw.mac.type = ei->mac;
2662 adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
2663
2664 /* PCI config space info */
2665
2666 hw->vendor_id = pdev->vendor;
2667 hw->device_id = pdev->device;
2668 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2669 hw->subsystem_device_id = pdev->subsystem_device;
2670
2671 pci_read_config_byte(pdev, PCI_REVISION_ID, &hw->revision_id);
2672
2673 err = -EIO;
2674 adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2675 pci_resource_len(pdev, 0));
2676
2677 if (!adapter->hw.hw_addr)
2678 goto err_ioremap;
2679
2680 if (ei->get_variants) {
2681 err = ei->get_variants(adapter);
2682 if (err)
2683 goto err_ioremap;
2684 }
2685
2686 /* setup adapter struct */
2687 err = igbvf_sw_init(adapter);
2688 if (err)
2689 goto err_sw_init;
2690
2691 /* construct the net_device struct */
2692 netdev->netdev_ops = &igbvf_netdev_ops;
2693
2694 igbvf_set_ethtool_ops(netdev);
2695 netdev->watchdog_timeo = 5 * HZ;
2696 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2697
2698 adapter->bd_number = cards_found++;
2699
2700 netdev->features = NETIF_F_SG |
2701 NETIF_F_IP_CSUM |
2702 NETIF_F_HW_VLAN_TX |
2703 NETIF_F_HW_VLAN_RX |
2704 NETIF_F_HW_VLAN_FILTER;
2705
2706 netdev->features |= NETIF_F_IPV6_CSUM;
2707 netdev->features |= NETIF_F_TSO;
2708 netdev->features |= NETIF_F_TSO6;
2709
2710 if (pci_using_dac)
2711 netdev->features |= NETIF_F_HIGHDMA;
2712
2713 netdev->vlan_features |= NETIF_F_TSO;
2714 netdev->vlan_features |= NETIF_F_TSO6;
2715 netdev->vlan_features |= NETIF_F_IP_CSUM;
2716 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2717 netdev->vlan_features |= NETIF_F_SG;
2718
2719 /*reset the controller to put the device in a known good state */
2720 err = hw->mac.ops.reset_hw(hw);
2721 if (err) {
2722 dev_info(&pdev->dev,
2723 "PF still in reset state, assigning new address\n");
2724 random_ether_addr(hw->mac.addr);
2725 } else {
2726 err = hw->mac.ops.read_mac_addr(hw);
2727 if (err) {
2728 dev_err(&pdev->dev, "Error reading MAC address\n");
2729 goto err_hw_init;
2730 }
2731 }
2732
2733 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
2734 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
2735
2736 if (!is_valid_ether_addr(netdev->perm_addr)) {
2737 dev_err(&pdev->dev, "Invalid MAC Address: "
2738 "%02x:%02x:%02x:%02x:%02x:%02x\n",
2739 netdev->dev_addr[0], netdev->dev_addr[1],
2740 netdev->dev_addr[2], netdev->dev_addr[3],
2741 netdev->dev_addr[4], netdev->dev_addr[5]);
2742 err = -EIO;
2743 goto err_hw_init;
2744 }
2745
2746 setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2747 (unsigned long) adapter);
2748
2749 INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2750 INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2751
2752 /* ring size defaults */
2753 adapter->rx_ring->count = 1024;
2754 adapter->tx_ring->count = 1024;
2755
2756 /* reset the hardware with the new settings */
2757 igbvf_reset(adapter);
2758
2759 /* tell the stack to leave us alone until igbvf_open() is called */
2760 netif_carrier_off(netdev);
2761 netif_stop_queue(netdev);
2762
2763 strcpy(netdev->name, "eth%d");
2764 err = register_netdev(netdev);
2765 if (err)
2766 goto err_hw_init;
2767
2768 igbvf_print_device_info(adapter);
2769
2770 igbvf_initialize_last_counter_stats(adapter);
2771
2772 return 0;
2773
2774err_hw_init:
2775 kfree(adapter->tx_ring);
2776 kfree(adapter->rx_ring);
2777err_sw_init:
2778 igbvf_reset_interrupt_capability(adapter);
2779 iounmap(adapter->hw.hw_addr);
2780err_ioremap:
2781 free_netdev(netdev);
2782err_alloc_etherdev:
2783 pci_release_regions(pdev);
2784err_pci_reg:
2785err_dma:
2786 pci_disable_device(pdev);
2787 return err;
2788}
2789
2790/**
2791 * igbvf_remove - Device Removal Routine
2792 * @pdev: PCI device information struct
2793 *
2794 * igbvf_remove is called by the PCI subsystem to alert the driver
2795 * that it should release a PCI device. The could be caused by a
2796 * Hot-Plug event, or because the driver is going to be removed from
2797 * memory.
2798 **/
2799static void __devexit igbvf_remove(struct pci_dev *pdev)
2800{
2801 struct net_device *netdev = pci_get_drvdata(pdev);
2802 struct igbvf_adapter *adapter = netdev_priv(netdev);
2803 struct e1000_hw *hw = &adapter->hw;
2804
2805 /*
2806 * flush_scheduled work may reschedule our watchdog task, so
2807 * explicitly disable watchdog tasks from being rescheduled
2808 */
2809 set_bit(__IGBVF_DOWN, &adapter->state);
2810 del_timer_sync(&adapter->watchdog_timer);
2811
2812 flush_scheduled_work();
2813
2814 unregister_netdev(netdev);
2815
2816 igbvf_reset_interrupt_capability(adapter);
2817
2818 /*
2819 * it is important to delete the napi struct prior to freeing the
2820 * rx ring so that you do not end up with null pointer refs
2821 */
2822 netif_napi_del(&adapter->rx_ring->napi);
2823 kfree(adapter->tx_ring);
2824 kfree(adapter->rx_ring);
2825
2826 iounmap(hw->hw_addr);
2827 if (hw->flash_address)
2828 iounmap(hw->flash_address);
2829 pci_release_regions(pdev);
2830
2831 free_netdev(netdev);
2832
2833 pci_disable_device(pdev);
2834}
2835
2836/* PCI Error Recovery (ERS) */
2837static struct pci_error_handlers igbvf_err_handler = {
2838 .error_detected = igbvf_io_error_detected,
2839 .slot_reset = igbvf_io_slot_reset,
2840 .resume = igbvf_io_resume,
2841};
2842
2843static struct pci_device_id igbvf_pci_tbl[] = {
2844 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2845 { } /* terminate list */
2846};
2847MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2848
2849/* PCI Device API Driver */
2850static struct pci_driver igbvf_driver = {
2851 .name = igbvf_driver_name,
2852 .id_table = igbvf_pci_tbl,
2853 .probe = igbvf_probe,
2854 .remove = __devexit_p(igbvf_remove),
2855#ifdef CONFIG_PM
2856 /* Power Management Hooks */
2857 .suspend = igbvf_suspend,
2858 .resume = igbvf_resume,
2859#endif
2860 .shutdown = igbvf_shutdown,
2861 .err_handler = &igbvf_err_handler
2862};
2863
2864/**
2865 * igbvf_init_module - Driver Registration Routine
2866 *
2867 * igbvf_init_module is the first routine called when the driver is
2868 * loaded. All it does is register with the PCI subsystem.
2869 **/
2870static int __init igbvf_init_module(void)
2871{
2872 int ret;
2873 printk(KERN_INFO "%s - version %s\n",
2874 igbvf_driver_string, igbvf_driver_version);
2875 printk(KERN_INFO "%s\n", igbvf_copyright);
2876
2877 ret = pci_register_driver(&igbvf_driver);
2878 pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, igbvf_driver_name,
2879 PM_QOS_DEFAULT_VALUE);
2880
2881 return ret;
2882}
2883module_init(igbvf_init_module);
2884
2885/**
2886 * igbvf_exit_module - Driver Exit Cleanup Routine
2887 *
2888 * igbvf_exit_module is called just before the driver is removed
2889 * from memory.
2890 **/
2891static void __exit igbvf_exit_module(void)
2892{
2893 pci_unregister_driver(&igbvf_driver);
2894 pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, igbvf_driver_name);
2895}
2896module_exit(igbvf_exit_module);
2897
2898
2899MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2900MODULE_DESCRIPTION("Intel(R) 82576 Virtual Function Network Driver");
2901MODULE_LICENSE("GPL");
2902MODULE_VERSION(DRV_VERSION);
2903
2904/* netdev.c */