blob: 7012e3da3bdf351fa6ef928c2d4df4697e9ef47e [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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090038#include <linux/slab.h>
Alexander Duyckd4e0fe02009-04-07 14:37:34 +000039#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;
51static const char igbvf_driver_string[] =
52 "Intel(R) Virtual Function Network Driver";
53static const char igbvf_copyright[] = "Copyright (c) 2009 Intel Corporation.";
54
55static int igbvf_poll(struct napi_struct *napi, int budget);
Alexander Duyck2d165772009-04-09 22:49:20 +000056static void igbvf_reset(struct igbvf_adapter *);
57static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
58static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +000059
60static struct igbvf_info igbvf_vf_info = {
61 .mac = e1000_vfadapt,
Alexander Duyck0364d6f2009-05-06 10:25:01 +000062 .flags = 0,
Alexander Duyckd4e0fe02009-04-07 14:37:34 +000063 .pba = 10,
64 .init_ops = e1000_init_function_pointers_vf,
65};
66
67static const struct igbvf_info *igbvf_info_tbl[] = {
68 [board_vf] = &igbvf_vf_info,
69};
70
71/**
72 * igbvf_desc_unused - calculate if we have unused descriptors
73 **/
74static int igbvf_desc_unused(struct igbvf_ring *ring)
75{
76 if (ring->next_to_clean > ring->next_to_use)
77 return ring->next_to_clean - ring->next_to_use - 1;
78
79 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
80}
81
82/**
83 * igbvf_receive_skb - helper function to handle Rx indications
84 * @adapter: board private structure
85 * @status: descriptor status field as written by hardware
86 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
87 * @skb: pointer to sk_buff to be indicated to stack
88 **/
89static void igbvf_receive_skb(struct igbvf_adapter *adapter,
90 struct net_device *netdev,
91 struct sk_buff *skb,
92 u32 status, u16 vlan)
93{
94 if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
95 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
96 le16_to_cpu(vlan) &
97 E1000_RXD_SPC_VLAN_MASK);
98 else
99 netif_receive_skb(skb);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000100}
101
102static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
103 u32 status_err, struct sk_buff *skb)
104{
105 skb->ip_summed = CHECKSUM_NONE;
106
107 /* Ignore Checksum bit is set or checksum is disabled through ethtool */
Alexander Duyck0364d6f2009-05-06 10:25:01 +0000108 if ((status_err & E1000_RXD_STAT_IXSM) ||
109 (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000110 return;
Alexander Duyck0364d6f2009-05-06 10:25:01 +0000111
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000112 /* TCP/UDP checksum error bit is set */
113 if (status_err &
114 (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
115 /* let the stack verify checksum errors */
116 adapter->hw_csum_err++;
117 return;
118 }
Alexander Duyck0364d6f2009-05-06 10:25:01 +0000119
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000120 /* It must be a TCP or UDP packet with a valid checksum */
121 if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
122 skb->ip_summed = CHECKSUM_UNNECESSARY;
123
124 adapter->hw_csum_good++;
125}
126
127/**
128 * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
129 * @rx_ring: address of ring structure to repopulate
130 * @cleaned_count: number of buffers to repopulate
131 **/
132static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
133 int cleaned_count)
134{
135 struct igbvf_adapter *adapter = rx_ring->adapter;
136 struct net_device *netdev = adapter->netdev;
137 struct pci_dev *pdev = adapter->pdev;
138 union e1000_adv_rx_desc *rx_desc;
139 struct igbvf_buffer *buffer_info;
140 struct sk_buff *skb;
141 unsigned int i;
142 int bufsz;
143
144 i = rx_ring->next_to_use;
145 buffer_info = &rx_ring->buffer_info[i];
146
147 if (adapter->rx_ps_hdr_size)
148 bufsz = adapter->rx_ps_hdr_size;
149 else
150 bufsz = adapter->rx_buffer_len;
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000151
152 while (cleaned_count--) {
153 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
154
155 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
156 if (!buffer_info->page) {
157 buffer_info->page = alloc_page(GFP_ATOMIC);
158 if (!buffer_info->page) {
159 adapter->alloc_rx_buff_failed++;
160 goto no_buffers;
161 }
162 buffer_info->page_offset = 0;
163 } else {
164 buffer_info->page_offset ^= PAGE_SIZE / 2;
165 }
166 buffer_info->page_dma =
Nick Nunley123e9f12010-04-27 13:09:44 +0000167 dma_map_page(&pdev->dev, buffer_info->page,
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000168 buffer_info->page_offset,
169 PAGE_SIZE / 2,
Nick Nunley123e9f12010-04-27 13:09:44 +0000170 DMA_FROM_DEVICE);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000171 }
172
173 if (!buffer_info->skb) {
Eric Dumazet89d71a62009-10-13 05:34:20 +0000174 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000175 if (!skb) {
176 adapter->alloc_rx_buff_failed++;
177 goto no_buffers;
178 }
179
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000180 buffer_info->skb = skb;
Nick Nunley123e9f12010-04-27 13:09:44 +0000181 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000182 bufsz,
Nick Nunley123e9f12010-04-27 13:09:44 +0000183 DMA_FROM_DEVICE);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000184 }
185 /* Refresh the desc even if buffer_addrs didn't change because
186 * each write-back erases this info. */
187 if (adapter->rx_ps_hdr_size) {
188 rx_desc->read.pkt_addr =
189 cpu_to_le64(buffer_info->page_dma);
190 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
191 } else {
192 rx_desc->read.pkt_addr =
193 cpu_to_le64(buffer_info->dma);
194 rx_desc->read.hdr_addr = 0;
195 }
196
197 i++;
198 if (i == rx_ring->count)
199 i = 0;
200 buffer_info = &rx_ring->buffer_info[i];
201 }
202
203no_buffers:
204 if (rx_ring->next_to_use != i) {
205 rx_ring->next_to_use = i;
206 if (i == 0)
207 i = (rx_ring->count - 1);
208 else
209 i--;
210
211 /* Force memory writes to complete before letting h/w
212 * know there are new descriptors to fetch. (Only
213 * applicable for weak-ordered memory model archs,
214 * such as IA-64). */
215 wmb();
216 writel(i, adapter->hw.hw_addr + rx_ring->tail);
217 }
218}
219
220/**
221 * igbvf_clean_rx_irq - Send received data up the network stack; legacy
222 * @adapter: board private structure
223 *
224 * the return value indicates whether actual cleaning was done, there
225 * is no guarantee that everything was cleaned
226 **/
227static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
228 int *work_done, int work_to_do)
229{
230 struct igbvf_ring *rx_ring = adapter->rx_ring;
231 struct net_device *netdev = adapter->netdev;
232 struct pci_dev *pdev = adapter->pdev;
233 union e1000_adv_rx_desc *rx_desc, *next_rxd;
234 struct igbvf_buffer *buffer_info, *next_buffer;
235 struct sk_buff *skb;
236 bool cleaned = false;
237 int cleaned_count = 0;
238 unsigned int total_bytes = 0, total_packets = 0;
239 unsigned int i;
240 u32 length, hlen, staterr;
241
242 i = rx_ring->next_to_clean;
243 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
244 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
245
246 while (staterr & E1000_RXD_STAT_DD) {
247 if (*work_done >= work_to_do)
248 break;
249 (*work_done)++;
250
251 buffer_info = &rx_ring->buffer_info[i];
252
253 /* HW will not DMA in data larger than the given buffer, even
254 * if it parses the (NFS, of course) header to be larger. In
255 * that case, it fills the header buffer and spills the rest
256 * into the page.
257 */
258 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
259 E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
260 if (hlen > adapter->rx_ps_hdr_size)
261 hlen = adapter->rx_ps_hdr_size;
262
263 length = le16_to_cpu(rx_desc->wb.upper.length);
264 cleaned = true;
265 cleaned_count++;
266
267 skb = buffer_info->skb;
268 prefetch(skb->data - NET_IP_ALIGN);
269 buffer_info->skb = NULL;
270 if (!adapter->rx_ps_hdr_size) {
Nick Nunley123e9f12010-04-27 13:09:44 +0000271 dma_unmap_single(&pdev->dev, buffer_info->dma,
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000272 adapter->rx_buffer_len,
Nick Nunley123e9f12010-04-27 13:09:44 +0000273 DMA_FROM_DEVICE);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000274 buffer_info->dma = 0;
275 skb_put(skb, length);
276 goto send_up;
277 }
278
279 if (!skb_shinfo(skb)->nr_frags) {
Nick Nunley123e9f12010-04-27 13:09:44 +0000280 dma_unmap_single(&pdev->dev, buffer_info->dma,
Alexander Duyck92d947b2009-07-23 18:11:01 +0000281 adapter->rx_ps_hdr_size,
Nick Nunley123e9f12010-04-27 13:09:44 +0000282 DMA_FROM_DEVICE);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000283 skb_put(skb, hlen);
284 }
285
286 if (length) {
Nick Nunley123e9f12010-04-27 13:09:44 +0000287 dma_unmap_page(&pdev->dev, buffer_info->page_dma,
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000288 PAGE_SIZE / 2,
Nick Nunley123e9f12010-04-27 13:09:44 +0000289 DMA_FROM_DEVICE);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000290 buffer_info->page_dma = 0;
291
Koki Sanagiec857fd2010-04-27 01:01:39 +0000292 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000293 buffer_info->page,
294 buffer_info->page_offset,
295 length);
296
297 if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
298 (page_count(buffer_info->page) != 1))
299 buffer_info->page = NULL;
300 else
301 get_page(buffer_info->page);
302
303 skb->len += length;
304 skb->data_len += length;
305 skb->truesize += length;
306 }
307send_up:
308 i++;
309 if (i == rx_ring->count)
310 i = 0;
311 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
312 prefetch(next_rxd);
313 next_buffer = &rx_ring->buffer_info[i];
314
315 if (!(staterr & E1000_RXD_STAT_EOP)) {
316 buffer_info->skb = next_buffer->skb;
317 buffer_info->dma = next_buffer->dma;
318 next_buffer->skb = skb;
319 next_buffer->dma = 0;
320 goto next_desc;
321 }
322
323 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
324 dev_kfree_skb_irq(skb);
325 goto next_desc;
326 }
327
328 total_bytes += skb->len;
329 total_packets++;
330
331 igbvf_rx_checksum_adv(adapter, staterr, skb);
332
333 skb->protocol = eth_type_trans(skb, netdev);
334
335 igbvf_receive_skb(adapter, netdev, skb, staterr,
336 rx_desc->wb.upper.vlan);
337
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000338next_desc:
339 rx_desc->wb.upper.status_error = 0;
340
341 /* return some buffers to hardware, one at a time is too slow */
342 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
343 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
344 cleaned_count = 0;
345 }
346
347 /* use prefetched values */
348 rx_desc = next_rxd;
349 buffer_info = next_buffer;
350
351 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
352 }
353
354 rx_ring->next_to_clean = i;
355 cleaned_count = igbvf_desc_unused(rx_ring);
356
357 if (cleaned_count)
358 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
359
360 adapter->total_rx_packets += total_packets;
361 adapter->total_rx_bytes += total_bytes;
362 adapter->net_stats.rx_bytes += total_bytes;
363 adapter->net_stats.rx_packets += total_packets;
364 return cleaned;
365}
366
367static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
368 struct igbvf_buffer *buffer_info)
369{
Alexander Duycka7d5ca42009-12-02 16:47:37 +0000370 if (buffer_info->dma) {
371 if (buffer_info->mapped_as_page)
Nick Nunley123e9f12010-04-27 13:09:44 +0000372 dma_unmap_page(&adapter->pdev->dev,
Alexander Duycka7d5ca42009-12-02 16:47:37 +0000373 buffer_info->dma,
374 buffer_info->length,
Nick Nunley123e9f12010-04-27 13:09:44 +0000375 DMA_TO_DEVICE);
Alexander Duycka7d5ca42009-12-02 16:47:37 +0000376 else
Nick Nunley123e9f12010-04-27 13:09:44 +0000377 dma_unmap_single(&adapter->pdev->dev,
Alexander Duycka7d5ca42009-12-02 16:47:37 +0000378 buffer_info->dma,
379 buffer_info->length,
Nick Nunley123e9f12010-04-27 13:09:44 +0000380 DMA_TO_DEVICE);
Alexander Duycka7d5ca42009-12-02 16:47:37 +0000381 buffer_info->dma = 0;
382 }
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000383 if (buffer_info->skb) {
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000384 dev_kfree_skb_any(buffer_info->skb);
385 buffer_info->skb = NULL;
386 }
387 buffer_info->time_stamp = 0;
388}
389
390static void igbvf_print_tx_hang(struct igbvf_adapter *adapter)
391{
392 struct igbvf_ring *tx_ring = adapter->tx_ring;
393 unsigned int i = tx_ring->next_to_clean;
394 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
395 union e1000_adv_tx_desc *eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
396
397 /* detected Tx unit hang */
398 dev_err(&adapter->pdev->dev,
399 "Detected Tx Unit Hang:\n"
400 " TDH <%x>\n"
401 " TDT <%x>\n"
402 " next_to_use <%x>\n"
403 " next_to_clean <%x>\n"
404 "buffer_info[next_to_clean]:\n"
405 " time_stamp <%lx>\n"
406 " next_to_watch <%x>\n"
407 " jiffies <%lx>\n"
408 " next_to_watch.status <%x>\n",
409 readl(adapter->hw.hw_addr + tx_ring->head),
410 readl(adapter->hw.hw_addr + tx_ring->tail),
411 tx_ring->next_to_use,
412 tx_ring->next_to_clean,
413 tx_ring->buffer_info[eop].time_stamp,
414 eop,
415 jiffies,
416 eop_desc->wb.status);
417}
418
419/**
420 * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
421 * @adapter: board private structure
422 *
423 * Return 0 on success, negative on failure
424 **/
425int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
426 struct igbvf_ring *tx_ring)
427{
428 struct pci_dev *pdev = adapter->pdev;
429 int size;
430
431 size = sizeof(struct igbvf_buffer) * tx_ring->count;
432 tx_ring->buffer_info = vmalloc(size);
433 if (!tx_ring->buffer_info)
434 goto err;
435 memset(tx_ring->buffer_info, 0, size);
436
437 /* round up to nearest 4K */
438 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
439 tx_ring->size = ALIGN(tx_ring->size, 4096);
440
Nick Nunley123e9f12010-04-27 13:09:44 +0000441 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
442 &tx_ring->dma, GFP_KERNEL);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000443
444 if (!tx_ring->desc)
445 goto err;
446
447 tx_ring->adapter = adapter;
448 tx_ring->next_to_use = 0;
449 tx_ring->next_to_clean = 0;
450
451 return 0;
452err:
453 vfree(tx_ring->buffer_info);
454 dev_err(&adapter->pdev->dev,
455 "Unable to allocate memory for the transmit descriptor ring\n");
456 return -ENOMEM;
457}
458
459/**
460 * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
461 * @adapter: board private structure
462 *
463 * Returns 0 on success, negative on failure
464 **/
465int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
466 struct igbvf_ring *rx_ring)
467{
468 struct pci_dev *pdev = adapter->pdev;
469 int size, desc_len;
470
471 size = sizeof(struct igbvf_buffer) * rx_ring->count;
472 rx_ring->buffer_info = vmalloc(size);
473 if (!rx_ring->buffer_info)
474 goto err;
475 memset(rx_ring->buffer_info, 0, size);
476
477 desc_len = sizeof(union e1000_adv_rx_desc);
478
479 /* Round up to nearest 4K */
480 rx_ring->size = rx_ring->count * desc_len;
481 rx_ring->size = ALIGN(rx_ring->size, 4096);
482
Nick Nunley123e9f12010-04-27 13:09:44 +0000483 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
484 &rx_ring->dma, GFP_KERNEL);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000485
486 if (!rx_ring->desc)
487 goto err;
488
489 rx_ring->next_to_clean = 0;
490 rx_ring->next_to_use = 0;
491
492 rx_ring->adapter = adapter;
493
494 return 0;
495
496err:
497 vfree(rx_ring->buffer_info);
498 rx_ring->buffer_info = NULL;
499 dev_err(&adapter->pdev->dev,
500 "Unable to allocate memory for the receive descriptor ring\n");
501 return -ENOMEM;
502}
503
504/**
505 * igbvf_clean_tx_ring - Free Tx Buffers
506 * @tx_ring: ring to be cleaned
507 **/
508static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
509{
510 struct igbvf_adapter *adapter = tx_ring->adapter;
511 struct igbvf_buffer *buffer_info;
512 unsigned long size;
513 unsigned int i;
514
515 if (!tx_ring->buffer_info)
516 return;
517
518 /* Free all the Tx ring sk_buffs */
519 for (i = 0; i < tx_ring->count; i++) {
520 buffer_info = &tx_ring->buffer_info[i];
521 igbvf_put_txbuf(adapter, buffer_info);
522 }
523
524 size = sizeof(struct igbvf_buffer) * tx_ring->count;
525 memset(tx_ring->buffer_info, 0, size);
526
527 /* Zero out the descriptor ring */
528 memset(tx_ring->desc, 0, tx_ring->size);
529
530 tx_ring->next_to_use = 0;
531 tx_ring->next_to_clean = 0;
532
533 writel(0, adapter->hw.hw_addr + tx_ring->head);
534 writel(0, adapter->hw.hw_addr + tx_ring->tail);
535}
536
537/**
538 * igbvf_free_tx_resources - Free Tx Resources per Queue
539 * @tx_ring: ring to free resources from
540 *
541 * Free all transmit software resources
542 **/
543void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
544{
545 struct pci_dev *pdev = tx_ring->adapter->pdev;
546
547 igbvf_clean_tx_ring(tx_ring);
548
549 vfree(tx_ring->buffer_info);
550 tx_ring->buffer_info = NULL;
551
Nick Nunley123e9f12010-04-27 13:09:44 +0000552 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
553 tx_ring->dma);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000554
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){
Nick Nunley123e9f12010-04-27 13:09:44 +0000578 dma_unmap_single(&pdev->dev, buffer_info->dma,
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000579 adapter->rx_ps_hdr_size,
Nick Nunley123e9f12010-04-27 13:09:44 +0000580 DMA_FROM_DEVICE);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000581 } else {
Nick Nunley123e9f12010-04-27 13:09:44 +0000582 dma_unmap_single(&pdev->dev, buffer_info->dma,
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000583 adapter->rx_buffer_len,
Nick Nunley123e9f12010-04-27 13:09:44 +0000584 DMA_FROM_DEVICE);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000585 }
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)
Nick Nunley123e9f12010-04-27 13:09:44 +0000596 dma_unmap_page(&pdev->dev,
597 buffer_info->page_dma,
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000598 PAGE_SIZE / 2,
Nick Nunley123e9f12010-04-27 13:09:44 +0000599 DMA_FROM_DEVICE);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000600 put_page(buffer_info->page);
601 buffer_info->page = NULL;
602 buffer_info->page_dma = 0;
603 buffer_info->page_offset = 0;
604 }
605 }
606
607 size = sizeof(struct igbvf_buffer) * rx_ring->count;
608 memset(rx_ring->buffer_info, 0, size);
609
610 /* Zero out the descriptor ring */
611 memset(rx_ring->desc, 0, rx_ring->size);
612
613 rx_ring->next_to_clean = 0;
614 rx_ring->next_to_use = 0;
615
616 writel(0, adapter->hw.hw_addr + rx_ring->head);
617 writel(0, adapter->hw.hw_addr + rx_ring->tail);
618}
619
620/**
621 * igbvf_free_rx_resources - Free Rx Resources
622 * @rx_ring: ring to clean the resources from
623 *
624 * Free all receive software resources
625 **/
626
627void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
628{
629 struct pci_dev *pdev = rx_ring->adapter->pdev;
630
631 igbvf_clean_rx_ring(rx_ring);
632
633 vfree(rx_ring->buffer_info);
634 rx_ring->buffer_info = NULL;
635
636 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
637 rx_ring->dma);
638 rx_ring->desc = NULL;
639}
640
641/**
642 * igbvf_update_itr - update the dynamic ITR value based on statistics
643 * @adapter: pointer to adapter
644 * @itr_setting: current adapter->itr
645 * @packets: the number of packets during this measurement interval
646 * @bytes: the number of bytes during this measurement interval
647 *
648 * Stores a new ITR value based on packets and byte
649 * counts during the last interrupt. The advantage of per interrupt
650 * computation is faster updates and more accurate ITR for the current
651 * traffic pattern. Constants in this function were computed
652 * based on theoretical maximum wire speed and thresholds were set based
653 * on testing data as well as attempting to minimize response time
654 * while increasing bulk throughput. This functionality is controlled
655 * by the InterruptThrottleRate module parameter.
656 **/
657static unsigned int igbvf_update_itr(struct igbvf_adapter *adapter,
658 u16 itr_setting, int packets,
659 int bytes)
660{
661 unsigned int retval = itr_setting;
662
663 if (packets == 0)
664 goto update_itr_done;
665
666 switch (itr_setting) {
667 case lowest_latency:
668 /* handle TSO and jumbo frames */
669 if (bytes/packets > 8000)
670 retval = bulk_latency;
671 else if ((packets < 5) && (bytes > 512))
672 retval = low_latency;
673 break;
674 case low_latency: /* 50 usec aka 20000 ints/s */
675 if (bytes > 10000) {
676 /* this if handles the TSO accounting */
677 if (bytes/packets > 8000)
678 retval = bulk_latency;
679 else if ((packets < 10) || ((bytes/packets) > 1200))
680 retval = bulk_latency;
681 else if ((packets > 35))
682 retval = lowest_latency;
683 } else if (bytes/packets > 2000) {
684 retval = bulk_latency;
685 } else if (packets <= 2 && bytes < 512) {
686 retval = lowest_latency;
687 }
688 break;
689 case bulk_latency: /* 250 usec aka 4000 ints/s */
690 if (bytes > 25000) {
691 if (packets > 35)
692 retval = low_latency;
693 } else if (bytes < 6000) {
694 retval = low_latency;
695 }
696 break;
697 }
698
699update_itr_done:
700 return retval;
701}
702
703static void igbvf_set_itr(struct igbvf_adapter *adapter)
704{
705 struct e1000_hw *hw = &adapter->hw;
706 u16 current_itr;
707 u32 new_itr = adapter->itr;
708
709 adapter->tx_itr = igbvf_update_itr(adapter, adapter->tx_itr,
710 adapter->total_tx_packets,
711 adapter->total_tx_bytes);
712 /* conservative mode (itr 3) eliminates the lowest_latency setting */
713 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
714 adapter->tx_itr = low_latency;
715
716 adapter->rx_itr = igbvf_update_itr(adapter, adapter->rx_itr,
717 adapter->total_rx_packets,
718 adapter->total_rx_bytes);
719 /* conservative mode (itr 3) eliminates the lowest_latency setting */
720 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
721 adapter->rx_itr = low_latency;
722
723 current_itr = max(adapter->rx_itr, adapter->tx_itr);
724
725 switch (current_itr) {
726 /* counts and packets in update_itr are dependent on these numbers */
727 case lowest_latency:
728 new_itr = 70000;
729 break;
730 case low_latency:
731 new_itr = 20000; /* aka hwitr = ~200 */
732 break;
733 case bulk_latency:
734 new_itr = 4000;
735 break;
736 default:
737 break;
738 }
739
740 if (new_itr != adapter->itr) {
741 /*
742 * this attempts to bias the interrupt rate towards Bulk
743 * by adding intermediate steps when interrupt rate is
744 * increasing
745 */
746 new_itr = new_itr > adapter->itr ?
747 min(adapter->itr + (new_itr >> 2), new_itr) :
748 new_itr;
749 adapter->itr = new_itr;
750 adapter->rx_ring->itr_val = 1952;
751
752 if (adapter->msix_entries)
753 adapter->rx_ring->set_itr = 1;
754 else
755 ew32(ITR, 1952);
756 }
757}
758
759/**
760 * igbvf_clean_tx_irq - Reclaim resources after transmit completes
761 * @adapter: board private structure
762 * returns true if ring is completely cleaned
763 **/
764static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
765{
766 struct igbvf_adapter *adapter = tx_ring->adapter;
767 struct e1000_hw *hw = &adapter->hw;
768 struct net_device *netdev = adapter->netdev;
769 struct igbvf_buffer *buffer_info;
770 struct sk_buff *skb;
771 union e1000_adv_tx_desc *tx_desc, *eop_desc;
772 unsigned int total_bytes = 0, total_packets = 0;
773 unsigned int i, eop, count = 0;
774 bool cleaned = false;
775
776 i = tx_ring->next_to_clean;
777 eop = tx_ring->buffer_info[i].next_to_watch;
778 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
779
780 while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
781 (count < tx_ring->count)) {
782 for (cleaned = false; !cleaned; count++) {
783 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
784 buffer_info = &tx_ring->buffer_info[i];
785 cleaned = (i == eop);
786 skb = buffer_info->skb;
787
788 if (skb) {
789 unsigned int segs, bytecount;
790
791 /* gso_segs is currently only valid for tcp */
792 segs = skb_shinfo(skb)->gso_segs ?: 1;
793 /* multiply data chunks by size of headers */
794 bytecount = ((segs - 1) * skb_headlen(skb)) +
795 skb->len;
796 total_packets += segs;
797 total_bytes += bytecount;
798 }
799
800 igbvf_put_txbuf(adapter, buffer_info);
801 tx_desc->wb.status = 0;
802
803 i++;
804 if (i == tx_ring->count)
805 i = 0;
806 }
807 eop = tx_ring->buffer_info[i].next_to_watch;
808 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
809 }
810
811 tx_ring->next_to_clean = i;
812
813 if (unlikely(count &&
814 netif_carrier_ok(netdev) &&
815 igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
816 /* Make sure that anybody stopping the queue after this
817 * sees the new next_to_clean.
818 */
819 smp_mb();
820 if (netif_queue_stopped(netdev) &&
821 !(test_bit(__IGBVF_DOWN, &adapter->state))) {
822 netif_wake_queue(netdev);
823 ++adapter->restart_queue;
824 }
825 }
826
827 if (adapter->detect_tx_hung) {
828 /* Detect a transmit hang in hardware, this serializes the
829 * check with the clearing of time_stamp and movement of i */
830 adapter->detect_tx_hung = false;
831 if (tx_ring->buffer_info[i].time_stamp &&
832 time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
Joe Perches8e95a202009-12-03 07:58:21 +0000833 (adapter->tx_timeout_factor * HZ)) &&
834 !(er32(STATUS) & E1000_STATUS_TXOFF)) {
Alexander Duyckd4e0fe02009-04-07 14:37:34 +0000835
836 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
837 /* detected Tx unit hang */
838 igbvf_print_tx_hang(adapter);
839
840 netif_stop_queue(netdev);
841 }
842 }
843 adapter->net_stats.tx_bytes += total_bytes;
844 adapter->net_stats.tx_packets += total_packets;
845 return (count < tx_ring->count);
846}
847
848static irqreturn_t igbvf_msix_other(int irq, void *data)
849{
850 struct net_device *netdev = data;
851 struct igbvf_adapter *adapter = netdev_priv(netdev);
852 struct e1000_hw *hw = &adapter->hw;
853
854 adapter->int_counter1++;
855
856 netif_carrier_off(netdev);
857 hw->mac.get_link_status = 1;
858 if (!test_bit(__IGBVF_DOWN, &adapter->state))
859 mod_timer(&adapter->watchdog_timer, jiffies + 1);
860
861 ew32(EIMS, adapter->eims_other);
862
863 return IRQ_HANDLED;
864}
865
866static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
867{
868 struct net_device *netdev = data;
869 struct igbvf_adapter *adapter = netdev_priv(netdev);
870 struct e1000_hw *hw = &adapter->hw;
871 struct igbvf_ring *tx_ring = adapter->tx_ring;
872
873
874 adapter->total_tx_bytes = 0;
875 adapter->total_tx_packets = 0;
876
877 /* auto mask will automatically reenable the interrupt when we write
878 * EICS */
879 if (!igbvf_clean_tx_irq(tx_ring))
880 /* Ring was not completely cleaned, so fire another interrupt */
881 ew32(EICS, tx_ring->eims_value);
882 else
883 ew32(EIMS, tx_ring->eims_value);
884
885 return IRQ_HANDLED;
886}
887
888static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
889{
890 struct net_device *netdev = data;
891 struct igbvf_adapter *adapter = netdev_priv(netdev);
892
893 adapter->int_counter0++;
894
895 /* Write the ITR value calculated at the end of the
896 * previous interrupt.
897 */
898 if (adapter->rx_ring->set_itr) {
899 writel(adapter->rx_ring->itr_val,
900 adapter->hw.hw_addr + adapter->rx_ring->itr_register);
901 adapter->rx_ring->set_itr = 0;
902 }
903
904 if (napi_schedule_prep(&adapter->rx_ring->napi)) {
905 adapter->total_rx_bytes = 0;
906 adapter->total_rx_packets = 0;
907 __napi_schedule(&adapter->rx_ring->napi);
908 }
909
910 return IRQ_HANDLED;
911}
912
913#define IGBVF_NO_QUEUE -1
914
915static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
916 int tx_queue, int msix_vector)
917{
918 struct e1000_hw *hw = &adapter->hw;
919 u32 ivar, index;
920
921 /* 82576 uses a table-based method for assigning vectors.
922 Each queue has a single entry in the table to which we write
923 a vector number along with a "valid" bit. Sadly, the layout
924 of the table is somewhat counterintuitive. */
925 if (rx_queue > IGBVF_NO_QUEUE) {
926 index = (rx_queue >> 1);
927 ivar = array_er32(IVAR0, index);
928 if (rx_queue & 0x1) {
929 /* vector goes into third byte of register */
930 ivar = ivar & 0xFF00FFFF;
931 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
932 } else {
933 /* vector goes into low byte of register */
934 ivar = ivar & 0xFFFFFF00;
935 ivar |= msix_vector | E1000_IVAR_VALID;
936 }
937 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
938 array_ew32(IVAR0, index, ivar);
939 }
940 if (tx_queue > IGBVF_NO_QUEUE) {
941 index = (tx_queue >> 1);
942 ivar = array_er32(IVAR0, index);
943 if (tx_queue & 0x1) {
944 /* vector goes into high byte of register */
945 ivar = ivar & 0x00FFFFFF;
946 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
947 } else {
948 /* vector goes into second byte of register */
949 ivar = ivar & 0xFFFF00FF;
950 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
951 }
952 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
953 array_ew32(IVAR0, index, ivar);
954 }
955}
956
957/**
958 * igbvf_configure_msix - Configure MSI-X hardware
959 *
960 * igbvf_configure_msix sets up the hardware to properly
961 * generate MSI-X interrupts.
962 **/
963static void igbvf_configure_msix(struct igbvf_adapter *adapter)
964{
965 u32 tmp;
966 struct e1000_hw *hw = &adapter->hw;
967 struct igbvf_ring *tx_ring = adapter->tx_ring;
968 struct igbvf_ring *rx_ring = adapter->rx_ring;
969 int vector = 0;
970
971 adapter->eims_enable_mask = 0;
972
973 igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
974 adapter->eims_enable_mask |= tx_ring->eims_value;
975 if (tx_ring->itr_val)
976 writel(tx_ring->itr_val,
977 hw->hw_addr + tx_ring->itr_register);
978 else
979 writel(1952, hw->hw_addr + tx_ring->itr_register);
980
981 igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
982 adapter->eims_enable_mask |= rx_ring->eims_value;
983 if (rx_ring->itr_val)
984 writel(rx_ring->itr_val,
985 hw->hw_addr + rx_ring->itr_register);
986 else
987 writel(1952, hw->hw_addr + rx_ring->itr_register);
988
989 /* set vector for other causes, i.e. link changes */
990
991 tmp = (vector++ | E1000_IVAR_VALID);
992
993 ew32(IVAR_MISC, tmp);
994
995 adapter->eims_enable_mask = (1 << (vector)) - 1;
996 adapter->eims_other = 1 << (vector - 1);
997 e1e_flush();
998}
999
Alexander Duyck2d165772009-04-09 22:49:20 +00001000static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001001{
1002 if (adapter->msix_entries) {
1003 pci_disable_msix(adapter->pdev);
1004 kfree(adapter->msix_entries);
1005 adapter->msix_entries = NULL;
1006 }
1007}
1008
1009/**
1010 * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1011 *
1012 * Attempt to configure interrupts using the best available
1013 * capabilities of the hardware and kernel.
1014 **/
Alexander Duyck2d165772009-04-09 22:49:20 +00001015static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001016{
1017 int err = -ENOMEM;
1018 int i;
1019
1020 /* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
1021 adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1022 GFP_KERNEL);
1023 if (adapter->msix_entries) {
1024 for (i = 0; i < 3; i++)
1025 adapter->msix_entries[i].entry = i;
1026
1027 err = pci_enable_msix(adapter->pdev,
1028 adapter->msix_entries, 3);
1029 }
1030
1031 if (err) {
1032 /* MSI-X failed */
1033 dev_err(&adapter->pdev->dev,
1034 "Failed to initialize MSI-X interrupts.\n");
1035 igbvf_reset_interrupt_capability(adapter);
1036 }
1037}
1038
1039/**
1040 * igbvf_request_msix - Initialize MSI-X interrupts
1041 *
1042 * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1043 * kernel.
1044 **/
1045static int igbvf_request_msix(struct igbvf_adapter *adapter)
1046{
1047 struct net_device *netdev = adapter->netdev;
1048 int err = 0, vector = 0;
1049
1050 if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1051 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1052 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1053 } else {
1054 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1055 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1056 }
1057
1058 err = request_irq(adapter->msix_entries[vector].vector,
Joe Perchesa0607fd2009-11-18 23:29:17 -08001059 igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001060 netdev);
1061 if (err)
1062 goto out;
1063
1064 adapter->tx_ring->itr_register = E1000_EITR(vector);
1065 adapter->tx_ring->itr_val = 1952;
1066 vector++;
1067
1068 err = request_irq(adapter->msix_entries[vector].vector,
Joe Perchesa0607fd2009-11-18 23:29:17 -08001069 igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001070 netdev);
1071 if (err)
1072 goto out;
1073
1074 adapter->rx_ring->itr_register = E1000_EITR(vector);
1075 adapter->rx_ring->itr_val = 1952;
1076 vector++;
1077
1078 err = request_irq(adapter->msix_entries[vector].vector,
Joe Perchesa0607fd2009-11-18 23:29:17 -08001079 igbvf_msix_other, 0, netdev->name, netdev);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001080 if (err)
1081 goto out;
1082
1083 igbvf_configure_msix(adapter);
1084 return 0;
1085out:
1086 return err;
1087}
1088
1089/**
1090 * igbvf_alloc_queues - Allocate memory for all rings
1091 * @adapter: board private structure to initialize
1092 **/
1093static int __devinit igbvf_alloc_queues(struct igbvf_adapter *adapter)
1094{
1095 struct net_device *netdev = adapter->netdev;
1096
1097 adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1098 if (!adapter->tx_ring)
1099 return -ENOMEM;
1100
1101 adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1102 if (!adapter->rx_ring) {
1103 kfree(adapter->tx_ring);
1104 return -ENOMEM;
1105 }
1106
1107 netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1108
1109 return 0;
1110}
1111
1112/**
1113 * igbvf_request_irq - initialize interrupts
1114 *
1115 * Attempts to configure interrupts using the best available
1116 * capabilities of the hardware and kernel.
1117 **/
1118static int igbvf_request_irq(struct igbvf_adapter *adapter)
1119{
1120 int err = -1;
1121
1122 /* igbvf supports msi-x only */
1123 if (adapter->msix_entries)
1124 err = igbvf_request_msix(adapter);
1125
1126 if (!err)
1127 return err;
1128
1129 dev_err(&adapter->pdev->dev,
1130 "Unable to allocate interrupt, Error: %d\n", err);
1131
1132 return err;
1133}
1134
1135static void igbvf_free_irq(struct igbvf_adapter *adapter)
1136{
1137 struct net_device *netdev = adapter->netdev;
1138 int vector;
1139
1140 if (adapter->msix_entries) {
1141 for (vector = 0; vector < 3; vector++)
1142 free_irq(adapter->msix_entries[vector].vector, netdev);
1143 }
1144}
1145
1146/**
1147 * igbvf_irq_disable - Mask off interrupt generation on the NIC
1148 **/
1149static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1150{
1151 struct e1000_hw *hw = &adapter->hw;
1152
1153 ew32(EIMC, ~0);
1154
1155 if (adapter->msix_entries)
1156 ew32(EIAC, 0);
1157}
1158
1159/**
1160 * igbvf_irq_enable - Enable default interrupt generation settings
1161 **/
1162static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1163{
1164 struct e1000_hw *hw = &adapter->hw;
1165
1166 ew32(EIAC, adapter->eims_enable_mask);
1167 ew32(EIAM, adapter->eims_enable_mask);
1168 ew32(EIMS, adapter->eims_enable_mask);
1169}
1170
1171/**
1172 * igbvf_poll - NAPI Rx polling callback
1173 * @napi: struct associated with this polling callback
1174 * @budget: amount of packets driver is allowed to process this poll
1175 **/
1176static int igbvf_poll(struct napi_struct *napi, int budget)
1177{
1178 struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1179 struct igbvf_adapter *adapter = rx_ring->adapter;
1180 struct e1000_hw *hw = &adapter->hw;
1181 int work_done = 0;
1182
1183 igbvf_clean_rx_irq(adapter, &work_done, budget);
1184
1185 /* If not enough Rx work done, exit the polling mode */
1186 if (work_done < budget) {
1187 napi_complete(napi);
1188
1189 if (adapter->itr_setting & 3)
1190 igbvf_set_itr(adapter);
1191
1192 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1193 ew32(EIMS, adapter->rx_ring->eims_value);
1194 }
1195
1196 return work_done;
1197}
1198
1199/**
1200 * igbvf_set_rlpml - set receive large packet maximum length
1201 * @adapter: board private structure
1202 *
1203 * Configure the maximum size of packets that will be received
1204 */
1205static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1206{
1207 int max_frame_size = adapter->max_frame_size;
1208 struct e1000_hw *hw = &adapter->hw;
1209
1210 if (adapter->vlgrp)
1211 max_frame_size += VLAN_TAG_SIZE;
1212
1213 e1000_rlpml_set_vf(hw, max_frame_size);
1214}
1215
1216static void igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1217{
1218 struct igbvf_adapter *adapter = netdev_priv(netdev);
1219 struct e1000_hw *hw = &adapter->hw;
1220
1221 if (hw->mac.ops.set_vfta(hw, vid, true))
1222 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1223}
1224
1225static void igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1226{
1227 struct igbvf_adapter *adapter = netdev_priv(netdev);
1228 struct e1000_hw *hw = &adapter->hw;
1229
1230 igbvf_irq_disable(adapter);
1231 vlan_group_set_device(adapter->vlgrp, vid, NULL);
1232
1233 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1234 igbvf_irq_enable(adapter);
1235
1236 if (hw->mac.ops.set_vfta(hw, vid, false))
1237 dev_err(&adapter->pdev->dev,
1238 "Failed to remove vlan id %d\n", vid);
1239}
1240
1241static void igbvf_vlan_rx_register(struct net_device *netdev,
1242 struct vlan_group *grp)
1243{
1244 struct igbvf_adapter *adapter = netdev_priv(netdev);
1245
1246 adapter->vlgrp = grp;
1247}
1248
1249static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1250{
1251 u16 vid;
1252
1253 if (!adapter->vlgrp)
1254 return;
1255
1256 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1257 if (!vlan_group_get_device(adapter->vlgrp, vid))
1258 continue;
1259 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1260 }
1261
1262 igbvf_set_rlpml(adapter);
1263}
1264
1265/**
1266 * igbvf_configure_tx - Configure Transmit Unit after Reset
1267 * @adapter: board private structure
1268 *
1269 * Configure the Tx unit of the MAC after a reset.
1270 **/
1271static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1272{
1273 struct e1000_hw *hw = &adapter->hw;
1274 struct igbvf_ring *tx_ring = adapter->tx_ring;
1275 u64 tdba;
1276 u32 txdctl, dca_txctrl;
1277
1278 /* disable transmits */
1279 txdctl = er32(TXDCTL(0));
1280 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1281 msleep(10);
1282
1283 /* Setup the HW Tx Head and Tail descriptor pointers */
1284 ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1285 tdba = tx_ring->dma;
Andrew Morton8e20ce92009-06-18 16:49:17 -07001286 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001287 ew32(TDBAH(0), (tdba >> 32));
1288 ew32(TDH(0), 0);
1289 ew32(TDT(0), 0);
1290 tx_ring->head = E1000_TDH(0);
1291 tx_ring->tail = E1000_TDT(0);
1292
1293 /* Turn off Relaxed Ordering on head write-backs. The writebacks
1294 * MUST be delivered in order or it will completely screw up
1295 * our bookeeping.
1296 */
1297 dca_txctrl = er32(DCA_TXCTRL(0));
1298 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1299 ew32(DCA_TXCTRL(0), dca_txctrl);
1300
1301 /* enable transmits */
1302 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1303 ew32(TXDCTL(0), txdctl);
1304
1305 /* Setup Transmit Descriptor Settings for eop descriptor */
1306 adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1307
1308 /* enable Report Status bit */
1309 adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001310}
1311
1312/**
1313 * igbvf_setup_srrctl - configure the receive control registers
1314 * @adapter: Board private structure
1315 **/
1316static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1317{
1318 struct e1000_hw *hw = &adapter->hw;
1319 u32 srrctl = 0;
1320
1321 srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1322 E1000_SRRCTL_BSIZEHDR_MASK |
1323 E1000_SRRCTL_BSIZEPKT_MASK);
1324
1325 /* Enable queue drop to avoid head of line blocking */
1326 srrctl |= E1000_SRRCTL_DROP_EN;
1327
1328 /* Setup buffer sizes */
1329 srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1330 E1000_SRRCTL_BSIZEPKT_SHIFT;
1331
1332 if (adapter->rx_buffer_len < 2048) {
1333 adapter->rx_ps_hdr_size = 0;
1334 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1335 } else {
1336 adapter->rx_ps_hdr_size = 128;
1337 srrctl |= adapter->rx_ps_hdr_size <<
1338 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1339 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1340 }
1341
1342 ew32(SRRCTL(0), srrctl);
1343}
1344
1345/**
1346 * igbvf_configure_rx - Configure Receive Unit after Reset
1347 * @adapter: board private structure
1348 *
1349 * Configure the Rx unit of the MAC after a reset.
1350 **/
1351static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1352{
1353 struct e1000_hw *hw = &adapter->hw;
1354 struct igbvf_ring *rx_ring = adapter->rx_ring;
1355 u64 rdba;
1356 u32 rdlen, rxdctl;
1357
1358 /* disable receives */
1359 rxdctl = er32(RXDCTL(0));
1360 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1361 msleep(10);
1362
1363 rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1364
1365 /*
1366 * Setup the HW Rx Head and Tail Descriptor Pointers and
1367 * the Base and Length of the Rx Descriptor Ring
1368 */
1369 rdba = rx_ring->dma;
Andrew Morton8e20ce92009-06-18 16:49:17 -07001370 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001371 ew32(RDBAH(0), (rdba >> 32));
1372 ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1373 rx_ring->head = E1000_RDH(0);
1374 rx_ring->tail = E1000_RDT(0);
1375 ew32(RDH(0), 0);
1376 ew32(RDT(0), 0);
1377
1378 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1379 rxdctl &= 0xFFF00000;
1380 rxdctl |= IGBVF_RX_PTHRESH;
1381 rxdctl |= IGBVF_RX_HTHRESH << 8;
1382 rxdctl |= IGBVF_RX_WTHRESH << 16;
1383
1384 igbvf_set_rlpml(adapter);
1385
1386 /* enable receives */
1387 ew32(RXDCTL(0), rxdctl);
1388}
1389
1390/**
1391 * igbvf_set_multi - Multicast and Promiscuous mode set
1392 * @netdev: network interface device structure
1393 *
1394 * The set_multi entry point is called whenever the multicast address
1395 * list or the network interface flags are updated. This routine is
1396 * responsible for configuring the hardware for proper multicast,
1397 * promiscuous mode, and all-multi behavior.
1398 **/
1399static void igbvf_set_multi(struct net_device *netdev)
1400{
1401 struct igbvf_adapter *adapter = netdev_priv(netdev);
1402 struct e1000_hw *hw = &adapter->hw;
Jiri Pirko22bedad2010-04-01 21:22:57 +00001403 struct netdev_hw_addr *ha;
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001404 u8 *mta_list = NULL;
1405 int i;
1406
Jiri Pirko4cd24ea2010-02-08 04:30:35 +00001407 if (!netdev_mc_empty(netdev)) {
1408 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001409 if (!mta_list) {
1410 dev_err(&adapter->pdev->dev,
1411 "failed to allocate multicast filter list\n");
1412 return;
1413 }
1414 }
1415
1416 /* prepare a packed array of only addresses. */
Jiri Pirko48e2f182010-02-22 09:22:26 +00001417 i = 0;
Jiri Pirko22bedad2010-04-01 21:22:57 +00001418 netdev_for_each_mc_addr(ha, netdev)
1419 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001420
1421 hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1422 kfree(mta_list);
1423}
1424
1425/**
1426 * igbvf_configure - configure the hardware for Rx and Tx
1427 * @adapter: private board structure
1428 **/
1429static void igbvf_configure(struct igbvf_adapter *adapter)
1430{
1431 igbvf_set_multi(adapter->netdev);
1432
1433 igbvf_restore_vlan(adapter);
1434
1435 igbvf_configure_tx(adapter);
1436 igbvf_setup_srrctl(adapter);
1437 igbvf_configure_rx(adapter);
1438 igbvf_alloc_rx_buffers(adapter->rx_ring,
1439 igbvf_desc_unused(adapter->rx_ring));
1440}
1441
1442/* igbvf_reset - bring the hardware into a known good state
1443 *
1444 * This function boots the hardware and enables some settings that
1445 * require a configuration cycle of the hardware - those cannot be
1446 * set/changed during runtime. After reset the device needs to be
1447 * properly configured for Rx, Tx etc.
1448 */
Alexander Duyck2d165772009-04-09 22:49:20 +00001449static void igbvf_reset(struct igbvf_adapter *adapter)
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001450{
1451 struct e1000_mac_info *mac = &adapter->hw.mac;
1452 struct net_device *netdev = adapter->netdev;
1453 struct e1000_hw *hw = &adapter->hw;
1454
1455 /* Allow time for pending master requests to run */
1456 if (mac->ops.reset_hw(hw))
1457 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1458
1459 mac->ops.init_hw(hw);
1460
1461 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1462 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1463 netdev->addr_len);
1464 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1465 netdev->addr_len);
1466 }
Alexander Duyck72279092009-12-11 22:58:14 -08001467
1468 adapter->last_reset = jiffies;
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001469}
1470
1471int igbvf_up(struct igbvf_adapter *adapter)
1472{
1473 struct e1000_hw *hw = &adapter->hw;
1474
1475 /* hardware has been reset, we need to reload some things */
1476 igbvf_configure(adapter);
1477
1478 clear_bit(__IGBVF_DOWN, &adapter->state);
1479
1480 napi_enable(&adapter->rx_ring->napi);
1481 if (adapter->msix_entries)
1482 igbvf_configure_msix(adapter);
1483
1484 /* Clear any pending interrupts. */
1485 er32(EICR);
1486 igbvf_irq_enable(adapter);
1487
1488 /* start the watchdog */
1489 hw->mac.get_link_status = 1;
1490 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1491
1492
1493 return 0;
1494}
1495
1496void igbvf_down(struct igbvf_adapter *adapter)
1497{
1498 struct net_device *netdev = adapter->netdev;
1499 struct e1000_hw *hw = &adapter->hw;
1500 u32 rxdctl, txdctl;
1501
1502 /*
1503 * signal that we're down so the interrupt handler does not
1504 * reschedule our watchdog timer
1505 */
1506 set_bit(__IGBVF_DOWN, &adapter->state);
1507
1508 /* disable receives in the hardware */
1509 rxdctl = er32(RXDCTL(0));
1510 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1511
1512 netif_stop_queue(netdev);
1513
1514 /* disable transmits in the hardware */
1515 txdctl = er32(TXDCTL(0));
1516 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1517
1518 /* flush both disables and wait for them to finish */
1519 e1e_flush();
1520 msleep(10);
1521
1522 napi_disable(&adapter->rx_ring->napi);
1523
1524 igbvf_irq_disable(adapter);
1525
1526 del_timer_sync(&adapter->watchdog_timer);
1527
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001528 netif_carrier_off(netdev);
1529
1530 /* record the stats before reset*/
1531 igbvf_update_stats(adapter);
1532
1533 adapter->link_speed = 0;
1534 adapter->link_duplex = 0;
1535
1536 igbvf_reset(adapter);
1537 igbvf_clean_tx_ring(adapter->tx_ring);
1538 igbvf_clean_rx_ring(adapter->rx_ring);
1539}
1540
1541void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1542{
1543 might_sleep();
1544 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1545 msleep(1);
1546 igbvf_down(adapter);
1547 igbvf_up(adapter);
1548 clear_bit(__IGBVF_RESETTING, &adapter->state);
1549}
1550
1551/**
1552 * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1553 * @adapter: board private structure to initialize
1554 *
1555 * igbvf_sw_init initializes the Adapter private data structure.
1556 * Fields are initialized based on PCI device information and
1557 * OS network device settings (MTU size).
1558 **/
1559static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1560{
1561 struct net_device *netdev = adapter->netdev;
1562 s32 rc;
1563
1564 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1565 adapter->rx_ps_hdr_size = 0;
1566 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1567 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1568
1569 adapter->tx_int_delay = 8;
1570 adapter->tx_abs_int_delay = 32;
1571 adapter->rx_int_delay = 0;
1572 adapter->rx_abs_int_delay = 8;
1573 adapter->itr_setting = 3;
1574 adapter->itr = 20000;
1575
1576 /* Set various function pointers */
1577 adapter->ei->init_ops(&adapter->hw);
1578
1579 rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1580 if (rc)
1581 return rc;
1582
1583 rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1584 if (rc)
1585 return rc;
1586
1587 igbvf_set_interrupt_capability(adapter);
1588
1589 if (igbvf_alloc_queues(adapter))
1590 return -ENOMEM;
1591
1592 spin_lock_init(&adapter->tx_queue_lock);
1593
1594 /* Explicitly disable IRQ since the NIC can be in any state. */
1595 igbvf_irq_disable(adapter);
1596
1597 spin_lock_init(&adapter->stats_lock);
1598
1599 set_bit(__IGBVF_DOWN, &adapter->state);
1600 return 0;
1601}
1602
1603static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1604{
1605 struct e1000_hw *hw = &adapter->hw;
1606
1607 adapter->stats.last_gprc = er32(VFGPRC);
1608 adapter->stats.last_gorc = er32(VFGORC);
1609 adapter->stats.last_gptc = er32(VFGPTC);
1610 adapter->stats.last_gotc = er32(VFGOTC);
1611 adapter->stats.last_mprc = er32(VFMPRC);
1612 adapter->stats.last_gotlbc = er32(VFGOTLBC);
1613 adapter->stats.last_gptlbc = er32(VFGPTLBC);
1614 adapter->stats.last_gorlbc = er32(VFGORLBC);
1615 adapter->stats.last_gprlbc = er32(VFGPRLBC);
1616
1617 adapter->stats.base_gprc = er32(VFGPRC);
1618 adapter->stats.base_gorc = er32(VFGORC);
1619 adapter->stats.base_gptc = er32(VFGPTC);
1620 adapter->stats.base_gotc = er32(VFGOTC);
1621 adapter->stats.base_mprc = er32(VFMPRC);
1622 adapter->stats.base_gotlbc = er32(VFGOTLBC);
1623 adapter->stats.base_gptlbc = er32(VFGPTLBC);
1624 adapter->stats.base_gorlbc = er32(VFGORLBC);
1625 adapter->stats.base_gprlbc = er32(VFGPRLBC);
1626}
1627
1628/**
1629 * igbvf_open - Called when a network interface is made active
1630 * @netdev: network interface device structure
1631 *
1632 * Returns 0 on success, negative value on failure
1633 *
1634 * The open entry point is called when a network interface is made
1635 * active by the system (IFF_UP). At this point all resources needed
1636 * for transmit and receive operations are allocated, the interrupt
1637 * handler is registered with the OS, the watchdog timer is started,
1638 * and the stack is notified that the interface is ready.
1639 **/
1640static int igbvf_open(struct net_device *netdev)
1641{
1642 struct igbvf_adapter *adapter = netdev_priv(netdev);
1643 struct e1000_hw *hw = &adapter->hw;
1644 int err;
1645
1646 /* disallow open during test */
1647 if (test_bit(__IGBVF_TESTING, &adapter->state))
1648 return -EBUSY;
1649
1650 /* allocate transmit descriptors */
1651 err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1652 if (err)
1653 goto err_setup_tx;
1654
1655 /* allocate receive descriptors */
1656 err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1657 if (err)
1658 goto err_setup_rx;
1659
1660 /*
1661 * before we allocate an interrupt, we must be ready to handle it.
1662 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1663 * as soon as we call pci_request_irq, so we have to setup our
1664 * clean_rx handler before we do so.
1665 */
1666 igbvf_configure(adapter);
1667
1668 err = igbvf_request_irq(adapter);
1669 if (err)
1670 goto err_req_irq;
1671
1672 /* From here on the code is the same as igbvf_up() */
1673 clear_bit(__IGBVF_DOWN, &adapter->state);
1674
1675 napi_enable(&adapter->rx_ring->napi);
1676
1677 /* clear any pending interrupts */
1678 er32(EICR);
1679
1680 igbvf_irq_enable(adapter);
1681
1682 /* start the watchdog */
1683 hw->mac.get_link_status = 1;
1684 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1685
1686 return 0;
1687
1688err_req_irq:
1689 igbvf_free_rx_resources(adapter->rx_ring);
1690err_setup_rx:
1691 igbvf_free_tx_resources(adapter->tx_ring);
1692err_setup_tx:
1693 igbvf_reset(adapter);
1694
1695 return err;
1696}
1697
1698/**
1699 * igbvf_close - Disables a network interface
1700 * @netdev: network interface device structure
1701 *
1702 * Returns 0, this is not allowed to fail
1703 *
1704 * The close entry point is called when an interface is de-activated
1705 * by the OS. The hardware is still under the drivers control, but
1706 * needs to be disabled. A global MAC reset is issued to stop the
1707 * hardware, and all transmit and receive resources are freed.
1708 **/
1709static int igbvf_close(struct net_device *netdev)
1710{
1711 struct igbvf_adapter *adapter = netdev_priv(netdev);
1712
1713 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1714 igbvf_down(adapter);
1715
1716 igbvf_free_irq(adapter);
1717
1718 igbvf_free_tx_resources(adapter->tx_ring);
1719 igbvf_free_rx_resources(adapter->rx_ring);
1720
1721 return 0;
1722}
1723/**
1724 * igbvf_set_mac - Change the Ethernet Address of the NIC
1725 * @netdev: network interface device structure
1726 * @p: pointer to an address structure
1727 *
1728 * Returns 0 on success, negative on failure
1729 **/
1730static int igbvf_set_mac(struct net_device *netdev, void *p)
1731{
1732 struct igbvf_adapter *adapter = netdev_priv(netdev);
1733 struct e1000_hw *hw = &adapter->hw;
1734 struct sockaddr *addr = p;
1735
1736 if (!is_valid_ether_addr(addr->sa_data))
1737 return -EADDRNOTAVAIL;
1738
1739 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1740
1741 hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1742
1743 if (memcmp(addr->sa_data, hw->mac.addr, 6))
1744 return -EADDRNOTAVAIL;
1745
1746 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1747
1748 return 0;
1749}
1750
1751#define UPDATE_VF_COUNTER(reg, name) \
1752 { \
1753 u32 current_counter = er32(reg); \
1754 if (current_counter < adapter->stats.last_##name) \
1755 adapter->stats.name += 0x100000000LL; \
1756 adapter->stats.last_##name = current_counter; \
1757 adapter->stats.name &= 0xFFFFFFFF00000000LL; \
1758 adapter->stats.name |= current_counter; \
1759 }
1760
1761/**
1762 * igbvf_update_stats - Update the board statistics counters
1763 * @adapter: board private structure
1764**/
1765void igbvf_update_stats(struct igbvf_adapter *adapter)
1766{
1767 struct e1000_hw *hw = &adapter->hw;
1768 struct pci_dev *pdev = adapter->pdev;
1769
1770 /*
1771 * Prevent stats update while adapter is being reset, link is down
1772 * or if the pci connection is down.
1773 */
1774 if (adapter->link_speed == 0)
1775 return;
1776
1777 if (test_bit(__IGBVF_RESETTING, &adapter->state))
1778 return;
1779
1780 if (pci_channel_offline(pdev))
1781 return;
1782
1783 UPDATE_VF_COUNTER(VFGPRC, gprc);
1784 UPDATE_VF_COUNTER(VFGORC, gorc);
1785 UPDATE_VF_COUNTER(VFGPTC, gptc);
1786 UPDATE_VF_COUNTER(VFGOTC, gotc);
1787 UPDATE_VF_COUNTER(VFMPRC, mprc);
1788 UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1789 UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1790 UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1791 UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1792
1793 /* Fill out the OS statistics structure */
1794 adapter->net_stats.multicast = adapter->stats.mprc;
1795}
1796
1797static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1798{
1799 dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s\n",
1800 adapter->link_speed,
1801 ((adapter->link_duplex == FULL_DUPLEX) ?
1802 "Full Duplex" : "Half Duplex"));
1803}
1804
1805static bool igbvf_has_link(struct igbvf_adapter *adapter)
1806{
1807 struct e1000_hw *hw = &adapter->hw;
1808 s32 ret_val = E1000_SUCCESS;
1809 bool link_active;
1810
Alexander Duyck72279092009-12-11 22:58:14 -08001811 /* If interface is down, stay link down */
1812 if (test_bit(__IGBVF_DOWN, &adapter->state))
1813 return false;
1814
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001815 ret_val = hw->mac.ops.check_for_link(hw);
1816 link_active = !hw->mac.get_link_status;
1817
1818 /* if check for link returns error we will need to reset */
Alexander Duyck72279092009-12-11 22:58:14 -08001819 if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001820 schedule_work(&adapter->reset_task);
1821
1822 return link_active;
1823}
1824
1825/**
1826 * igbvf_watchdog - Timer Call-back
1827 * @data: pointer to adapter cast into an unsigned long
1828 **/
1829static void igbvf_watchdog(unsigned long data)
1830{
1831 struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1832
1833 /* Do the rest outside of interrupt context */
1834 schedule_work(&adapter->watchdog_task);
1835}
1836
1837static void igbvf_watchdog_task(struct work_struct *work)
1838{
1839 struct igbvf_adapter *adapter = container_of(work,
1840 struct igbvf_adapter,
1841 watchdog_task);
1842 struct net_device *netdev = adapter->netdev;
1843 struct e1000_mac_info *mac = &adapter->hw.mac;
1844 struct igbvf_ring *tx_ring = adapter->tx_ring;
1845 struct e1000_hw *hw = &adapter->hw;
1846 u32 link;
1847 int tx_pending = 0;
1848
1849 link = igbvf_has_link(adapter);
1850
1851 if (link) {
1852 if (!netif_carrier_ok(netdev)) {
1853 bool txb2b = 1;
1854
1855 mac->ops.get_link_up_info(&adapter->hw,
1856 &adapter->link_speed,
1857 &adapter->link_duplex);
1858 igbvf_print_link_info(adapter);
1859
Emil Tantilova08af742010-03-25 12:11:48 +00001860 /* adjust timeout factor according to speed/duplex */
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001861 adapter->tx_timeout_factor = 1;
1862 switch (adapter->link_speed) {
1863 case SPEED_10:
1864 txb2b = 0;
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001865 adapter->tx_timeout_factor = 16;
1866 break;
1867 case SPEED_100:
1868 txb2b = 0;
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001869 /* maybe add some timeout factor ? */
1870 break;
1871 }
1872
1873 netif_carrier_on(netdev);
1874 netif_wake_queue(netdev);
1875 }
1876 } else {
1877 if (netif_carrier_ok(netdev)) {
1878 adapter->link_speed = 0;
1879 adapter->link_duplex = 0;
1880 dev_info(&adapter->pdev->dev, "Link is Down\n");
1881 netif_carrier_off(netdev);
1882 netif_stop_queue(netdev);
1883 }
1884 }
1885
1886 if (netif_carrier_ok(netdev)) {
1887 igbvf_update_stats(adapter);
1888 } else {
1889 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1890 tx_ring->count);
1891 if (tx_pending) {
1892 /*
1893 * We've lost link, so the controller stops DMA,
1894 * but we've got queued Tx work that's never going
1895 * to get done, so reset controller to flush Tx.
1896 * (Do the reset outside of interrupt context).
1897 */
1898 adapter->tx_timeout_count++;
1899 schedule_work(&adapter->reset_task);
1900 }
1901 }
1902
1903 /* Cause software interrupt to ensure Rx ring is cleaned */
1904 ew32(EICS, adapter->rx_ring->eims_value);
1905
1906 /* Force detection of hung controller every watchdog period */
1907 adapter->detect_tx_hung = 1;
1908
1909 /* Reset the timer */
1910 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1911 mod_timer(&adapter->watchdog_timer,
1912 round_jiffies(jiffies + (2 * HZ)));
1913}
1914
1915#define IGBVF_TX_FLAGS_CSUM 0x00000001
1916#define IGBVF_TX_FLAGS_VLAN 0x00000002
1917#define IGBVF_TX_FLAGS_TSO 0x00000004
1918#define IGBVF_TX_FLAGS_IPV4 0x00000008
1919#define IGBVF_TX_FLAGS_VLAN_MASK 0xffff0000
1920#define IGBVF_TX_FLAGS_VLAN_SHIFT 16
1921
1922static int igbvf_tso(struct igbvf_adapter *adapter,
1923 struct igbvf_ring *tx_ring,
1924 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1925{
1926 struct e1000_adv_tx_context_desc *context_desc;
1927 unsigned int i;
1928 int err;
1929 struct igbvf_buffer *buffer_info;
1930 u32 info = 0, tu_cmd = 0;
1931 u32 mss_l4len_idx, l4len;
1932 *hdr_len = 0;
1933
1934 if (skb_header_cloned(skb)) {
1935 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1936 if (err) {
1937 dev_err(&adapter->pdev->dev,
1938 "igbvf_tso returning an error\n");
1939 return err;
1940 }
1941 }
1942
1943 l4len = tcp_hdrlen(skb);
1944 *hdr_len += l4len;
1945
1946 if (skb->protocol == htons(ETH_P_IP)) {
1947 struct iphdr *iph = ip_hdr(skb);
1948 iph->tot_len = 0;
1949 iph->check = 0;
1950 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1951 iph->daddr, 0,
1952 IPPROTO_TCP,
1953 0);
Sridhar Samudrala8e1e8a42010-01-23 02:02:21 -08001954 } else if (skb_is_gso_v6(skb)) {
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00001955 ipv6_hdr(skb)->payload_len = 0;
1956 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1957 &ipv6_hdr(skb)->daddr,
1958 0, IPPROTO_TCP, 0);
1959 }
1960
1961 i = tx_ring->next_to_use;
1962
1963 buffer_info = &tx_ring->buffer_info[i];
1964 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1965 /* VLAN MACLEN IPLEN */
1966 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1967 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1968 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1969 *hdr_len += skb_network_offset(skb);
1970 info |= (skb_transport_header(skb) - skb_network_header(skb));
1971 *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1972 context_desc->vlan_macip_lens = cpu_to_le32(info);
1973
1974 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1975 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1976
1977 if (skb->protocol == htons(ETH_P_IP))
1978 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1979 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1980
1981 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1982
1983 /* MSS L4LEN IDX */
1984 mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1985 mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1986
1987 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1988 context_desc->seqnum_seed = 0;
1989
1990 buffer_info->time_stamp = jiffies;
1991 buffer_info->next_to_watch = i;
1992 buffer_info->dma = 0;
1993 i++;
1994 if (i == tx_ring->count)
1995 i = 0;
1996
1997 tx_ring->next_to_use = i;
1998
1999 return true;
2000}
2001
2002static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
2003 struct igbvf_ring *tx_ring,
2004 struct sk_buff *skb, u32 tx_flags)
2005{
2006 struct e1000_adv_tx_context_desc *context_desc;
2007 unsigned int i;
2008 struct igbvf_buffer *buffer_info;
2009 u32 info = 0, tu_cmd = 0;
2010
2011 if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2012 (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
2013 i = tx_ring->next_to_use;
2014 buffer_info = &tx_ring->buffer_info[i];
2015 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
2016
2017 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2018 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
2019
2020 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2021 if (skb->ip_summed == CHECKSUM_PARTIAL)
2022 info |= (skb_transport_header(skb) -
2023 skb_network_header(skb));
2024
2025
2026 context_desc->vlan_macip_lens = cpu_to_le32(info);
2027
2028 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2029
2030 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2031 switch (skb->protocol) {
2032 case __constant_htons(ETH_P_IP):
2033 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2034 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2035 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2036 break;
2037 case __constant_htons(ETH_P_IPV6):
2038 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2039 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2040 break;
2041 default:
2042 break;
2043 }
2044 }
2045
2046 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2047 context_desc->seqnum_seed = 0;
2048 context_desc->mss_l4len_idx = 0;
2049
2050 buffer_info->time_stamp = jiffies;
2051 buffer_info->next_to_watch = i;
2052 buffer_info->dma = 0;
2053 i++;
2054 if (i == tx_ring->count)
2055 i = 0;
2056 tx_ring->next_to_use = i;
2057
2058 return true;
2059 }
2060
2061 return false;
2062}
2063
2064static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2065{
2066 struct igbvf_adapter *adapter = netdev_priv(netdev);
2067
2068 /* there is enough descriptors then we don't need to worry */
2069 if (igbvf_desc_unused(adapter->tx_ring) >= size)
2070 return 0;
2071
2072 netif_stop_queue(netdev);
2073
2074 smp_mb();
2075
2076 /* We need to check again just in case room has been made available */
2077 if (igbvf_desc_unused(adapter->tx_ring) < size)
2078 return -EBUSY;
2079
2080 netif_wake_queue(netdev);
2081
2082 ++adapter->restart_queue;
2083 return 0;
2084}
2085
2086#define IGBVF_MAX_TXD_PWR 16
2087#define IGBVF_MAX_DATA_PER_TXD (1 << IGBVF_MAX_TXD_PWR)
2088
2089static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2090 struct igbvf_ring *tx_ring,
2091 struct sk_buff *skb,
2092 unsigned int first)
2093{
2094 struct igbvf_buffer *buffer_info;
Alexander Duycka7d5ca42009-12-02 16:47:37 +00002095 struct pci_dev *pdev = adapter->pdev;
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002096 unsigned int len = skb_headlen(skb);
2097 unsigned int count = 0, i;
2098 unsigned int f;
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002099
2100 i = tx_ring->next_to_use;
2101
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002102 buffer_info = &tx_ring->buffer_info[i];
2103 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2104 buffer_info->length = len;
2105 /* set time_stamp *before* dma to help avoid a possible race */
2106 buffer_info->time_stamp = jiffies;
2107 buffer_info->next_to_watch = i;
Alexander Duyckac26d7d2010-01-27 15:30:39 +00002108 buffer_info->mapped_as_page = false;
Nick Nunley123e9f12010-04-27 13:09:44 +00002109 buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2110 DMA_TO_DEVICE);
2111 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
Alexander Duycka7d5ca42009-12-02 16:47:37 +00002112 goto dma_error;
2113
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002114
2115 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2116 struct skb_frag_struct *frag;
2117
Alexander Duyck85811452010-01-23 01:35:00 -08002118 count++;
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002119 i++;
2120 if (i == tx_ring->count)
2121 i = 0;
2122
2123 frag = &skb_shinfo(skb)->frags[f];
2124 len = frag->size;
2125
2126 buffer_info = &tx_ring->buffer_info[i];
2127 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2128 buffer_info->length = len;
2129 buffer_info->time_stamp = jiffies;
2130 buffer_info->next_to_watch = i;
Alexander Duycka7d5ca42009-12-02 16:47:37 +00002131 buffer_info->mapped_as_page = true;
Nick Nunley123e9f12010-04-27 13:09:44 +00002132 buffer_info->dma = dma_map_page(&pdev->dev,
Alexander Duycka7d5ca42009-12-02 16:47:37 +00002133 frag->page,
2134 frag->page_offset,
2135 len,
Nick Nunley123e9f12010-04-27 13:09:44 +00002136 DMA_TO_DEVICE);
2137 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
Alexander Duycka7d5ca42009-12-02 16:47:37 +00002138 goto dma_error;
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002139 }
2140
2141 tx_ring->buffer_info[i].skb = skb;
2142 tx_ring->buffer_info[first].next_to_watch = i;
2143
Alexander Duycka7d5ca42009-12-02 16:47:37 +00002144 return ++count;
2145
2146dma_error:
2147 dev_err(&pdev->dev, "TX DMA map failed\n");
2148
2149 /* clear timestamp and dma mappings for failed buffer_info mapping */
2150 buffer_info->dma = 0;
2151 buffer_info->time_stamp = 0;
2152 buffer_info->length = 0;
2153 buffer_info->next_to_watch = 0;
2154 buffer_info->mapped_as_page = false;
Roel Kluinc1fa3472010-01-19 14:21:45 +00002155 if (count)
2156 count--;
Alexander Duycka7d5ca42009-12-02 16:47:37 +00002157
2158 /* clear timestamp and dma mappings for remaining portion of packet */
Roel Kluinc1fa3472010-01-19 14:21:45 +00002159 while (count--) {
2160 if (i==0)
Alexander Duycka7d5ca42009-12-02 16:47:37 +00002161 i += tx_ring->count;
Roel Kluinc1fa3472010-01-19 14:21:45 +00002162 i--;
Alexander Duycka7d5ca42009-12-02 16:47:37 +00002163 buffer_info = &tx_ring->buffer_info[i];
2164 igbvf_put_txbuf(adapter, buffer_info);
2165 }
2166
2167 return 0;
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002168}
2169
2170static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2171 struct igbvf_ring *tx_ring,
2172 int tx_flags, int count, u32 paylen,
2173 u8 hdr_len)
2174{
2175 union e1000_adv_tx_desc *tx_desc = NULL;
2176 struct igbvf_buffer *buffer_info;
2177 u32 olinfo_status = 0, cmd_type_len;
2178 unsigned int i;
2179
2180 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2181 E1000_ADVTXD_DCMD_DEXT);
2182
2183 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2184 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2185
2186 if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2187 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2188
2189 /* insert tcp checksum */
2190 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2191
2192 /* insert ip checksum */
2193 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2194 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2195
2196 } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2197 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2198 }
2199
2200 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2201
2202 i = tx_ring->next_to_use;
2203 while (count--) {
2204 buffer_info = &tx_ring->buffer_info[i];
2205 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2206 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2207 tx_desc->read.cmd_type_len =
2208 cpu_to_le32(cmd_type_len | buffer_info->length);
2209 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2210 i++;
2211 if (i == tx_ring->count)
2212 i = 0;
2213 }
2214
2215 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2216 /* Force memory writes to complete before letting h/w
2217 * know there are new descriptors to fetch. (Only
2218 * applicable for weak-ordered memory model archs,
2219 * such as IA-64). */
2220 wmb();
2221
2222 tx_ring->next_to_use = i;
2223 writel(i, adapter->hw.hw_addr + tx_ring->tail);
2224 /* we need this if more than one processor can write to our tail
2225 * at a time, it syncronizes IO on IA64/Altix systems */
2226 mmiowb();
2227}
2228
Stephen Hemminger3b29a562009-08-31 19:50:55 +00002229static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2230 struct net_device *netdev,
2231 struct igbvf_ring *tx_ring)
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002232{
2233 struct igbvf_adapter *adapter = netdev_priv(netdev);
2234 unsigned int first, tx_flags = 0;
2235 u8 hdr_len = 0;
2236 int count = 0;
2237 int tso = 0;
2238
2239 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2240 dev_kfree_skb_any(skb);
2241 return NETDEV_TX_OK;
2242 }
2243
2244 if (skb->len <= 0) {
2245 dev_kfree_skb_any(skb);
2246 return NETDEV_TX_OK;
2247 }
2248
2249 /*
2250 * need: count + 4 desc gap to keep tail from touching
2251 * + 2 desc gap to keep tail from touching head,
2252 * + 1 desc for skb->data,
2253 * + 1 desc for context descriptor,
2254 * head, otherwise try next time
2255 */
2256 if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2257 /* this is a hard error */
2258 return NETDEV_TX_BUSY;
2259 }
2260
2261 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2262 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2263 tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2264 }
2265
2266 if (skb->protocol == htons(ETH_P_IP))
2267 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2268
2269 first = tx_ring->next_to_use;
2270
2271 tso = skb_is_gso(skb) ?
2272 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2273 if (unlikely(tso < 0)) {
2274 dev_kfree_skb_any(skb);
2275 return NETDEV_TX_OK;
2276 }
2277
2278 if (tso)
2279 tx_flags |= IGBVF_TX_FLAGS_TSO;
2280 else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2281 (skb->ip_summed == CHECKSUM_PARTIAL))
2282 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2283
2284 /*
2285 * count reflects descriptors mapped, if 0 then mapping error
2286 * has occured and we need to rewind the descriptor queue
2287 */
2288 count = igbvf_tx_map_adv(adapter, tx_ring, skb, first);
2289
2290 if (count) {
2291 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2292 skb->len, hdr_len);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002293 /* Make sure there is space in the ring for the next send. */
2294 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2295 } else {
2296 dev_kfree_skb_any(skb);
2297 tx_ring->buffer_info[first].time_stamp = 0;
2298 tx_ring->next_to_use = first;
2299 }
2300
2301 return NETDEV_TX_OK;
2302}
2303
Stephen Hemminger3b29a562009-08-31 19:50:55 +00002304static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2305 struct net_device *netdev)
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002306{
2307 struct igbvf_adapter *adapter = netdev_priv(netdev);
2308 struct igbvf_ring *tx_ring;
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002309
2310 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2311 dev_kfree_skb_any(skb);
2312 return NETDEV_TX_OK;
2313 }
2314
2315 tx_ring = &adapter->tx_ring[0];
2316
Stephen Hemminger3b29a562009-08-31 19:50:55 +00002317 return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002318}
2319
2320/**
2321 * igbvf_tx_timeout - Respond to a Tx Hang
2322 * @netdev: network interface device structure
2323 **/
2324static void igbvf_tx_timeout(struct net_device *netdev)
2325{
2326 struct igbvf_adapter *adapter = netdev_priv(netdev);
2327
2328 /* Do the reset outside of interrupt context */
2329 adapter->tx_timeout_count++;
2330 schedule_work(&adapter->reset_task);
2331}
2332
2333static void igbvf_reset_task(struct work_struct *work)
2334{
2335 struct igbvf_adapter *adapter;
2336 adapter = container_of(work, struct igbvf_adapter, reset_task);
2337
2338 igbvf_reinit_locked(adapter);
2339}
2340
2341/**
2342 * igbvf_get_stats - Get System Network Statistics
2343 * @netdev: network interface device structure
2344 *
2345 * Returns the address of the device statistics structure.
2346 * The statistics are actually updated from the timer callback.
2347 **/
2348static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2349{
2350 struct igbvf_adapter *adapter = netdev_priv(netdev);
2351
2352 /* only return the current stats */
2353 return &adapter->net_stats;
2354}
2355
2356/**
2357 * igbvf_change_mtu - Change the Maximum Transfer Unit
2358 * @netdev: network interface device structure
2359 * @new_mtu: new value for maximum frame size
2360 *
2361 * Returns 0 on success, negative on failure
2362 **/
2363static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2364{
2365 struct igbvf_adapter *adapter = netdev_priv(netdev);
2366 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2367
2368 if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2369 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2370 return -EINVAL;
2371 }
2372
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002373#define MAX_STD_JUMBO_FRAME_SIZE 9234
2374 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2375 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2376 return -EINVAL;
2377 }
2378
2379 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2380 msleep(1);
2381 /* igbvf_down has a dependency on max_frame_size */
2382 adapter->max_frame_size = max_frame;
2383 if (netif_running(netdev))
2384 igbvf_down(adapter);
2385
2386 /*
2387 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2388 * means we reserve 2 more, this pushes us to allocate from the next
2389 * larger slab size.
2390 * i.e. RXBUFFER_2048 --> size-4096 slab
2391 * However with the new *_jumbo_rx* routines, jumbo receives will use
2392 * fragmented skbs
2393 */
2394
2395 if (max_frame <= 1024)
2396 adapter->rx_buffer_len = 1024;
2397 else if (max_frame <= 2048)
2398 adapter->rx_buffer_len = 2048;
2399 else
2400#if (PAGE_SIZE / 2) > 16384
2401 adapter->rx_buffer_len = 16384;
2402#else
2403 adapter->rx_buffer_len = PAGE_SIZE / 2;
2404#endif
2405
2406
2407 /* adjust allocation if LPE protects us, and we aren't using SBP */
2408 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2409 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2410 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2411 ETH_FCS_LEN;
2412
2413 dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2414 netdev->mtu, new_mtu);
2415 netdev->mtu = new_mtu;
2416
2417 if (netif_running(netdev))
2418 igbvf_up(adapter);
2419 else
2420 igbvf_reset(adapter);
2421
2422 clear_bit(__IGBVF_RESETTING, &adapter->state);
2423
2424 return 0;
2425}
2426
2427static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2428{
2429 switch (cmd) {
2430 default:
2431 return -EOPNOTSUPP;
2432 }
2433}
2434
2435static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2436{
2437 struct net_device *netdev = pci_get_drvdata(pdev);
2438 struct igbvf_adapter *adapter = netdev_priv(netdev);
2439#ifdef CONFIG_PM
2440 int retval = 0;
2441#endif
2442
2443 netif_device_detach(netdev);
2444
2445 if (netif_running(netdev)) {
2446 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2447 igbvf_down(adapter);
2448 igbvf_free_irq(adapter);
2449 }
2450
2451#ifdef CONFIG_PM
2452 retval = pci_save_state(pdev);
2453 if (retval)
2454 return retval;
2455#endif
2456
2457 pci_disable_device(pdev);
2458
2459 return 0;
2460}
2461
2462#ifdef CONFIG_PM
2463static int igbvf_resume(struct pci_dev *pdev)
2464{
2465 struct net_device *netdev = pci_get_drvdata(pdev);
2466 struct igbvf_adapter *adapter = netdev_priv(netdev);
2467 u32 err;
2468
2469 pci_restore_state(pdev);
2470 err = pci_enable_device_mem(pdev);
2471 if (err) {
2472 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2473 return err;
2474 }
2475
2476 pci_set_master(pdev);
2477
2478 if (netif_running(netdev)) {
2479 err = igbvf_request_irq(adapter);
2480 if (err)
2481 return err;
2482 }
2483
2484 igbvf_reset(adapter);
2485
2486 if (netif_running(netdev))
2487 igbvf_up(adapter);
2488
2489 netif_device_attach(netdev);
2490
2491 return 0;
2492}
2493#endif
2494
2495static void igbvf_shutdown(struct pci_dev *pdev)
2496{
2497 igbvf_suspend(pdev, PMSG_SUSPEND);
2498}
2499
2500#ifdef CONFIG_NET_POLL_CONTROLLER
2501/*
2502 * Polling 'interrupt' - used by things like netconsole to send skbs
2503 * without having to re-enable interrupts. It's not called while
2504 * the interrupt routine is executing.
2505 */
2506static void igbvf_netpoll(struct net_device *netdev)
2507{
2508 struct igbvf_adapter *adapter = netdev_priv(netdev);
2509
2510 disable_irq(adapter->pdev->irq);
2511
2512 igbvf_clean_tx_irq(adapter->tx_ring);
2513
2514 enable_irq(adapter->pdev->irq);
2515}
2516#endif
2517
2518/**
2519 * igbvf_io_error_detected - called when PCI error is detected
2520 * @pdev: Pointer to PCI device
2521 * @state: The current pci connection state
2522 *
2523 * This function is called after a PCI bus error affecting
2524 * this device has been detected.
2525 */
2526static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2527 pci_channel_state_t state)
2528{
2529 struct net_device *netdev = pci_get_drvdata(pdev);
2530 struct igbvf_adapter *adapter = netdev_priv(netdev);
2531
2532 netif_device_detach(netdev);
2533
Dean Nelsonc06c4302009-07-31 09:13:33 +00002534 if (state == pci_channel_io_perm_failure)
2535 return PCI_ERS_RESULT_DISCONNECT;
2536
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002537 if (netif_running(netdev))
2538 igbvf_down(adapter);
2539 pci_disable_device(pdev);
2540
2541 /* Request a slot slot reset. */
2542 return PCI_ERS_RESULT_NEED_RESET;
2543}
2544
2545/**
2546 * igbvf_io_slot_reset - called after the pci bus has been reset.
2547 * @pdev: Pointer to PCI device
2548 *
2549 * Restart the card from scratch, as if from a cold-boot. Implementation
2550 * resembles the first-half of the igbvf_resume routine.
2551 */
2552static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2553{
2554 struct net_device *netdev = pci_get_drvdata(pdev);
2555 struct igbvf_adapter *adapter = netdev_priv(netdev);
2556
2557 if (pci_enable_device_mem(pdev)) {
2558 dev_err(&pdev->dev,
2559 "Cannot re-enable PCI device after reset.\n");
2560 return PCI_ERS_RESULT_DISCONNECT;
2561 }
2562 pci_set_master(pdev);
2563
2564 igbvf_reset(adapter);
2565
2566 return PCI_ERS_RESULT_RECOVERED;
2567}
2568
2569/**
2570 * igbvf_io_resume - called when traffic can start flowing again.
2571 * @pdev: Pointer to PCI device
2572 *
2573 * This callback is called when the error recovery driver tells us that
2574 * its OK to resume normal operation. Implementation resembles the
2575 * second-half of the igbvf_resume routine.
2576 */
2577static void igbvf_io_resume(struct pci_dev *pdev)
2578{
2579 struct net_device *netdev = pci_get_drvdata(pdev);
2580 struct igbvf_adapter *adapter = netdev_priv(netdev);
2581
2582 if (netif_running(netdev)) {
2583 if (igbvf_up(adapter)) {
2584 dev_err(&pdev->dev,
2585 "can't bring device back up after reset\n");
2586 return;
2587 }
2588 }
2589
2590 netif_device_attach(netdev);
2591}
2592
2593static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2594{
2595 struct e1000_hw *hw = &adapter->hw;
2596 struct net_device *netdev = adapter->netdev;
2597 struct pci_dev *pdev = adapter->pdev;
2598
2599 dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
H Hartley Sweeten753cdc32009-12-29 20:02:29 -08002600 dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002601 dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
2602}
2603
2604static const struct net_device_ops igbvf_netdev_ops = {
2605 .ndo_open = igbvf_open,
2606 .ndo_stop = igbvf_close,
2607 .ndo_start_xmit = igbvf_xmit_frame,
2608 .ndo_get_stats = igbvf_get_stats,
2609 .ndo_set_multicast_list = igbvf_set_multi,
2610 .ndo_set_mac_address = igbvf_set_mac,
2611 .ndo_change_mtu = igbvf_change_mtu,
2612 .ndo_do_ioctl = igbvf_ioctl,
2613 .ndo_tx_timeout = igbvf_tx_timeout,
2614 .ndo_vlan_rx_register = igbvf_vlan_rx_register,
2615 .ndo_vlan_rx_add_vid = igbvf_vlan_rx_add_vid,
2616 .ndo_vlan_rx_kill_vid = igbvf_vlan_rx_kill_vid,
2617#ifdef CONFIG_NET_POLL_CONTROLLER
2618 .ndo_poll_controller = igbvf_netpoll,
2619#endif
2620};
2621
2622/**
2623 * igbvf_probe - Device Initialization Routine
2624 * @pdev: PCI device information struct
2625 * @ent: entry in igbvf_pci_tbl
2626 *
2627 * Returns 0 on success, negative on failure
2628 *
2629 * igbvf_probe initializes an adapter identified by a pci_dev structure.
2630 * The OS initialization, configuring of the adapter private structure,
2631 * and a hardware reset occur.
2632 **/
2633static int __devinit igbvf_probe(struct pci_dev *pdev,
2634 const struct pci_device_id *ent)
2635{
2636 struct net_device *netdev;
2637 struct igbvf_adapter *adapter;
2638 struct e1000_hw *hw;
2639 const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2640
2641 static int cards_found;
2642 int err, pci_using_dac;
2643
2644 err = pci_enable_device_mem(pdev);
2645 if (err)
2646 return err;
2647
2648 pci_using_dac = 0;
Nick Nunley123e9f12010-04-27 13:09:44 +00002649 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002650 if (!err) {
Nick Nunley123e9f12010-04-27 13:09:44 +00002651 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002652 if (!err)
2653 pci_using_dac = 1;
2654 } else {
Nick Nunley123e9f12010-04-27 13:09:44 +00002655 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002656 if (err) {
Nick Nunley123e9f12010-04-27 13:09:44 +00002657 err = dma_set_coherent_mask(&pdev->dev,
2658 DMA_BIT_MASK(32));
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002659 if (err) {
2660 dev_err(&pdev->dev, "No usable DMA "
2661 "configuration, aborting\n");
2662 goto err_dma;
2663 }
2664 }
2665 }
2666
2667 err = pci_request_regions(pdev, igbvf_driver_name);
2668 if (err)
2669 goto err_pci_reg;
2670
2671 pci_set_master(pdev);
2672
2673 err = -ENOMEM;
2674 netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2675 if (!netdev)
2676 goto err_alloc_etherdev;
2677
2678 SET_NETDEV_DEV(netdev, &pdev->dev);
2679
2680 pci_set_drvdata(pdev, netdev);
2681 adapter = netdev_priv(netdev);
2682 hw = &adapter->hw;
2683 adapter->netdev = netdev;
2684 adapter->pdev = pdev;
2685 adapter->ei = ei;
2686 adapter->pba = ei->pba;
2687 adapter->flags = ei->flags;
2688 adapter->hw.back = adapter;
2689 adapter->hw.mac.type = ei->mac;
2690 adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
2691
2692 /* PCI config space info */
2693
2694 hw->vendor_id = pdev->vendor;
2695 hw->device_id = pdev->device;
2696 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2697 hw->subsystem_device_id = pdev->subsystem_device;
2698
2699 pci_read_config_byte(pdev, PCI_REVISION_ID, &hw->revision_id);
2700
2701 err = -EIO;
2702 adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2703 pci_resource_len(pdev, 0));
2704
2705 if (!adapter->hw.hw_addr)
2706 goto err_ioremap;
2707
2708 if (ei->get_variants) {
2709 err = ei->get_variants(adapter);
2710 if (err)
2711 goto err_ioremap;
2712 }
2713
2714 /* setup adapter struct */
2715 err = igbvf_sw_init(adapter);
2716 if (err)
2717 goto err_sw_init;
2718
2719 /* construct the net_device struct */
2720 netdev->netdev_ops = &igbvf_netdev_ops;
2721
2722 igbvf_set_ethtool_ops(netdev);
2723 netdev->watchdog_timeo = 5 * HZ;
2724 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2725
2726 adapter->bd_number = cards_found++;
2727
2728 netdev->features = NETIF_F_SG |
2729 NETIF_F_IP_CSUM |
2730 NETIF_F_HW_VLAN_TX |
2731 NETIF_F_HW_VLAN_RX |
2732 NETIF_F_HW_VLAN_FILTER;
2733
2734 netdev->features |= NETIF_F_IPV6_CSUM;
2735 netdev->features |= NETIF_F_TSO;
2736 netdev->features |= NETIF_F_TSO6;
2737
2738 if (pci_using_dac)
2739 netdev->features |= NETIF_F_HIGHDMA;
2740
2741 netdev->vlan_features |= NETIF_F_TSO;
2742 netdev->vlan_features |= NETIF_F_TSO6;
2743 netdev->vlan_features |= NETIF_F_IP_CSUM;
2744 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2745 netdev->vlan_features |= NETIF_F_SG;
2746
2747 /*reset the controller to put the device in a known good state */
2748 err = hw->mac.ops.reset_hw(hw);
2749 if (err) {
2750 dev_info(&pdev->dev,
Williams, Mitch A1242b6f2009-12-23 13:22:43 +00002751 "PF still in reset state, assigning new address."
2752 " Is the PF interface up?\n");
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002753 random_ether_addr(hw->mac.addr);
2754 } else {
2755 err = hw->mac.ops.read_mac_addr(hw);
2756 if (err) {
2757 dev_err(&pdev->dev, "Error reading MAC address\n");
2758 goto err_hw_init;
2759 }
2760 }
2761
2762 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
2763 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
2764
2765 if (!is_valid_ether_addr(netdev->perm_addr)) {
H Hartley Sweeten753cdc32009-12-29 20:02:29 -08002766 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
2767 netdev->dev_addr);
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002768 err = -EIO;
2769 goto err_hw_init;
2770 }
2771
2772 setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2773 (unsigned long) adapter);
2774
2775 INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2776 INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2777
2778 /* ring size defaults */
2779 adapter->rx_ring->count = 1024;
2780 adapter->tx_ring->count = 1024;
2781
2782 /* reset the hardware with the new settings */
2783 igbvf_reset(adapter);
2784
2785 /* tell the stack to leave us alone until igbvf_open() is called */
2786 netif_carrier_off(netdev);
2787 netif_stop_queue(netdev);
2788
2789 strcpy(netdev->name, "eth%d");
2790 err = register_netdev(netdev);
2791 if (err)
2792 goto err_hw_init;
2793
2794 igbvf_print_device_info(adapter);
2795
2796 igbvf_initialize_last_counter_stats(adapter);
2797
2798 return 0;
2799
2800err_hw_init:
2801 kfree(adapter->tx_ring);
2802 kfree(adapter->rx_ring);
2803err_sw_init:
2804 igbvf_reset_interrupt_capability(adapter);
2805 iounmap(adapter->hw.hw_addr);
2806err_ioremap:
2807 free_netdev(netdev);
2808err_alloc_etherdev:
2809 pci_release_regions(pdev);
2810err_pci_reg:
2811err_dma:
2812 pci_disable_device(pdev);
2813 return err;
2814}
2815
2816/**
2817 * igbvf_remove - Device Removal Routine
2818 * @pdev: PCI device information struct
2819 *
2820 * igbvf_remove is called by the PCI subsystem to alert the driver
2821 * that it should release a PCI device. The could be caused by a
2822 * Hot-Plug event, or because the driver is going to be removed from
2823 * memory.
2824 **/
2825static void __devexit igbvf_remove(struct pci_dev *pdev)
2826{
2827 struct net_device *netdev = pci_get_drvdata(pdev);
2828 struct igbvf_adapter *adapter = netdev_priv(netdev);
2829 struct e1000_hw *hw = &adapter->hw;
2830
2831 /*
2832 * flush_scheduled work may reschedule our watchdog task, so
2833 * explicitly disable watchdog tasks from being rescheduled
2834 */
2835 set_bit(__IGBVF_DOWN, &adapter->state);
2836 del_timer_sync(&adapter->watchdog_timer);
2837
2838 flush_scheduled_work();
2839
2840 unregister_netdev(netdev);
2841
2842 igbvf_reset_interrupt_capability(adapter);
2843
2844 /*
2845 * it is important to delete the napi struct prior to freeing the
2846 * rx ring so that you do not end up with null pointer refs
2847 */
2848 netif_napi_del(&adapter->rx_ring->napi);
2849 kfree(adapter->tx_ring);
2850 kfree(adapter->rx_ring);
2851
2852 iounmap(hw->hw_addr);
2853 if (hw->flash_address)
2854 iounmap(hw->flash_address);
2855 pci_release_regions(pdev);
2856
2857 free_netdev(netdev);
2858
2859 pci_disable_device(pdev);
2860}
2861
2862/* PCI Error Recovery (ERS) */
2863static struct pci_error_handlers igbvf_err_handler = {
2864 .error_detected = igbvf_io_error_detected,
2865 .slot_reset = igbvf_io_slot_reset,
2866 .resume = igbvf_io_resume,
2867};
2868
Alexey Dobriyana3aa1882010-01-07 11:58:11 +00002869static DEFINE_PCI_DEVICE_TABLE(igbvf_pci_tbl) = {
Alexander Duyckd4e0fe02009-04-07 14:37:34 +00002870 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2871 { } /* terminate list */
2872};
2873MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2874
2875/* PCI Device API Driver */
2876static struct pci_driver igbvf_driver = {
2877 .name = igbvf_driver_name,
2878 .id_table = igbvf_pci_tbl,
2879 .probe = igbvf_probe,
2880 .remove = __devexit_p(igbvf_remove),
2881#ifdef CONFIG_PM
2882 /* Power Management Hooks */
2883 .suspend = igbvf_suspend,
2884 .resume = igbvf_resume,
2885#endif
2886 .shutdown = igbvf_shutdown,
2887 .err_handler = &igbvf_err_handler
2888};
2889
2890/**
2891 * igbvf_init_module - Driver Registration Routine
2892 *
2893 * igbvf_init_module is the first routine called when the driver is
2894 * loaded. All it does is register with the PCI subsystem.
2895 **/
2896static int __init igbvf_init_module(void)
2897{
2898 int ret;
2899 printk(KERN_INFO "%s - version %s\n",
2900 igbvf_driver_string, igbvf_driver_version);
2901 printk(KERN_INFO "%s\n", igbvf_copyright);
2902
2903 ret = pci_register_driver(&igbvf_driver);
2904 pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, igbvf_driver_name,
2905 PM_QOS_DEFAULT_VALUE);
2906
2907 return ret;
2908}
2909module_init(igbvf_init_module);
2910
2911/**
2912 * igbvf_exit_module - Driver Exit Cleanup Routine
2913 *
2914 * igbvf_exit_module is called just before the driver is removed
2915 * from memory.
2916 **/
2917static void __exit igbvf_exit_module(void)
2918{
2919 pci_unregister_driver(&igbvf_driver);
2920 pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, igbvf_driver_name);
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 */