Merge branch 'bjorn.button' into release
[linux-2.6] / drivers / net / igbvf / netdev.c
1 /*******************************************************************************
2
3   Intel(R) 82576 Virtual Function Linux driver
4   Copyright(c) 2009 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25
26 *******************************************************************************/
27
28 #include <linux/module.h>
29 #include <linux/types.h>
30 #include <linux/init.h>
31 #include <linux/pci.h>
32 #include <linux/vmalloc.h>
33 #include <linux/pagemap.h>
34 #include <linux/delay.h>
35 #include <linux/netdevice.h>
36 #include <linux/tcp.h>
37 #include <linux/ipv6.h>
38 #include <net/checksum.h>
39 #include <net/ip6_checksum.h>
40 #include <linux/mii.h>
41 #include <linux/ethtool.h>
42 #include <linux/if_vlan.h>
43 #include <linux/pm_qos_params.h>
44
45 #include "igbvf.h"
46
47 #define DRV_VERSION "1.0.0-k0"
48 char igbvf_driver_name[] = "igbvf";
49 const char igbvf_driver_version[] = DRV_VERSION;
50 static const char igbvf_driver_string[] =
51                                 "Intel(R) Virtual Function Network Driver";
52 static const char igbvf_copyright[] = "Copyright (c) 2009 Intel Corporation.";
53
54 static int igbvf_poll(struct napi_struct *napi, int budget);
55 static void igbvf_reset(struct igbvf_adapter *);
56 static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
57 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
58
59 static struct igbvf_info igbvf_vf_info = {
60         .mac                    = e1000_vfadapt,
61         .flags                  = FLAG_HAS_JUMBO_FRAMES
62                                   | FLAG_RX_CSUM_ENABLED,
63         .pba                    = 10,
64         .init_ops               = e1000_init_function_pointers_vf,
65 };
66
67 static 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  **/
74 static 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  **/
89 static 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);
100
101         netdev->last_rx = jiffies;
102 }
103
104 static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
105                                          u32 status_err, struct sk_buff *skb)
106 {
107         skb->ip_summed = CHECKSUM_NONE;
108
109         /* Ignore Checksum bit is set or checksum is disabled through ethtool */
110         if ((status_err & E1000_RXD_STAT_IXSM))
111                 return;
112         /* 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         }
119         /* It must be a TCP or UDP packet with a valid checksum */
120         if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
121                 skb->ip_summed = CHECKSUM_UNNECESSARY;
122
123         adapter->hw_csum_good++;
124 }
125
126 /**
127  * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
128  * @rx_ring: address of ring structure to repopulate
129  * @cleaned_count: number of buffers to repopulate
130  **/
131 static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
132                                    int cleaned_count)
133 {
134         struct igbvf_adapter *adapter = rx_ring->adapter;
135         struct net_device *netdev = adapter->netdev;
136         struct pci_dev *pdev = adapter->pdev;
137         union e1000_adv_rx_desc *rx_desc;
138         struct igbvf_buffer *buffer_info;
139         struct sk_buff *skb;
140         unsigned int i;
141         int bufsz;
142
143         i = rx_ring->next_to_use;
144         buffer_info = &rx_ring->buffer_info[i];
145
146         if (adapter->rx_ps_hdr_size)
147                 bufsz = adapter->rx_ps_hdr_size;
148         else
149                 bufsz = adapter->rx_buffer_len;
150         bufsz += NET_IP_ALIGN;
151
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 =
167                                 pci_map_page(pdev, buffer_info->page,
168                                              buffer_info->page_offset,
169                                              PAGE_SIZE / 2,
170                                              PCI_DMA_FROMDEVICE);
171                 }
172
173                 if (!buffer_info->skb) {
174                         skb = netdev_alloc_skb(netdev, bufsz);
175                         if (!skb) {
176                                 adapter->alloc_rx_buff_failed++;
177                                 goto no_buffers;
178                         }
179
180                         /* Make buffer alignment 2 beyond a 16 byte boundary
181                          * this will result in a 16 byte aligned IP header after
182                          * the 14 byte MAC header is removed
183                          */
184                         skb_reserve(skb, NET_IP_ALIGN);
185
186                         buffer_info->skb = skb;
187                         buffer_info->dma = pci_map_single(pdev, skb->data,
188                                                           bufsz,
189                                                           PCI_DMA_FROMDEVICE);
190                 }
191                 /* Refresh the desc even if buffer_addrs didn't change because
192                  * each write-back erases this info. */
193                 if (adapter->rx_ps_hdr_size) {
194                         rx_desc->read.pkt_addr =
195                              cpu_to_le64(buffer_info->page_dma);
196                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
197                 } else {
198                         rx_desc->read.pkt_addr =
199                              cpu_to_le64(buffer_info->dma);
200                         rx_desc->read.hdr_addr = 0;
201                 }
202
203                 i++;
204                 if (i == rx_ring->count)
205                         i = 0;
206                 buffer_info = &rx_ring->buffer_info[i];
207         }
208
209 no_buffers:
210         if (rx_ring->next_to_use != i) {
211                 rx_ring->next_to_use = i;
212                 if (i == 0)
213                         i = (rx_ring->count - 1);
214                 else
215                         i--;
216
217                 /* Force memory writes to complete before letting h/w
218                  * know there are new descriptors to fetch.  (Only
219                  * applicable for weak-ordered memory model archs,
220                  * such as IA-64). */
221                 wmb();
222                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
223         }
224 }
225
226 /**
227  * igbvf_clean_rx_irq - Send received data up the network stack; legacy
228  * @adapter: board private structure
229  *
230  * the return value indicates whether actual cleaning was done, there
231  * is no guarantee that everything was cleaned
232  **/
233 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
234                                int *work_done, int work_to_do)
235 {
236         struct igbvf_ring *rx_ring = adapter->rx_ring;
237         struct net_device *netdev = adapter->netdev;
238         struct pci_dev *pdev = adapter->pdev;
239         union e1000_adv_rx_desc *rx_desc, *next_rxd;
240         struct igbvf_buffer *buffer_info, *next_buffer;
241         struct sk_buff *skb;
242         bool cleaned = false;
243         int cleaned_count = 0;
244         unsigned int total_bytes = 0, total_packets = 0;
245         unsigned int i;
246         u32 length, hlen, staterr;
247
248         i = rx_ring->next_to_clean;
249         rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
250         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
251
252         while (staterr & E1000_RXD_STAT_DD) {
253                 if (*work_done >= work_to_do)
254                         break;
255                 (*work_done)++;
256
257                 buffer_info = &rx_ring->buffer_info[i];
258
259                 /* HW will not DMA in data larger than the given buffer, even
260                  * if it parses the (NFS, of course) header to be larger.  In
261                  * that case, it fills the header buffer and spills the rest
262                  * into the page.
263                  */
264                 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
265                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
266                 if (hlen > adapter->rx_ps_hdr_size)
267                         hlen = adapter->rx_ps_hdr_size;
268
269                 length = le16_to_cpu(rx_desc->wb.upper.length);
270                 cleaned = true;
271                 cleaned_count++;
272
273                 skb = buffer_info->skb;
274                 prefetch(skb->data - NET_IP_ALIGN);
275                 buffer_info->skb = NULL;
276                 if (!adapter->rx_ps_hdr_size) {
277                         pci_unmap_single(pdev, buffer_info->dma,
278                                          adapter->rx_buffer_len,
279                                          PCI_DMA_FROMDEVICE);
280                         buffer_info->dma = 0;
281                         skb_put(skb, length);
282                         goto send_up;
283                 }
284
285                 if (!skb_shinfo(skb)->nr_frags) {
286                         pci_unmap_single(pdev, buffer_info->dma,
287                                          adapter->rx_ps_hdr_size + NET_IP_ALIGN,
288                                          PCI_DMA_FROMDEVICE);
289                         skb_put(skb, hlen);
290                 }
291
292                 if (length) {
293                         pci_unmap_page(pdev, buffer_info->page_dma,
294                                        PAGE_SIZE / 2,
295                                        PCI_DMA_FROMDEVICE);
296                         buffer_info->page_dma = 0;
297
298                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
299                                            buffer_info->page,
300                                            buffer_info->page_offset,
301                                            length);
302
303                         if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
304                             (page_count(buffer_info->page) != 1))
305                                 buffer_info->page = NULL;
306                         else
307                                 get_page(buffer_info->page);
308
309                         skb->len += length;
310                         skb->data_len += length;
311                         skb->truesize += length;
312                 }
313 send_up:
314                 i++;
315                 if (i == rx_ring->count)
316                         i = 0;
317                 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
318                 prefetch(next_rxd);
319                 next_buffer = &rx_ring->buffer_info[i];
320
321                 if (!(staterr & E1000_RXD_STAT_EOP)) {
322                         buffer_info->skb = next_buffer->skb;
323                         buffer_info->dma = next_buffer->dma;
324                         next_buffer->skb = skb;
325                         next_buffer->dma = 0;
326                         goto next_desc;
327                 }
328
329                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
330                         dev_kfree_skb_irq(skb);
331                         goto next_desc;
332                 }
333
334                 total_bytes += skb->len;
335                 total_packets++;
336
337                 igbvf_rx_checksum_adv(adapter, staterr, skb);
338
339                 skb->protocol = eth_type_trans(skb, netdev);
340
341                 igbvf_receive_skb(adapter, netdev, skb, staterr,
342                                   rx_desc->wb.upper.vlan);
343
344                 netdev->last_rx = jiffies;
345
346 next_desc:
347                 rx_desc->wb.upper.status_error = 0;
348
349                 /* return some buffers to hardware, one at a time is too slow */
350                 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
351                         igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
352                         cleaned_count = 0;
353                 }
354
355                 /* use prefetched values */
356                 rx_desc = next_rxd;
357                 buffer_info = next_buffer;
358
359                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
360         }
361
362         rx_ring->next_to_clean = i;
363         cleaned_count = igbvf_desc_unused(rx_ring);
364
365         if (cleaned_count)
366                 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
367
368         adapter->total_rx_packets += total_packets;
369         adapter->total_rx_bytes += total_bytes;
370         adapter->net_stats.rx_bytes += total_bytes;
371         adapter->net_stats.rx_packets += total_packets;
372         return cleaned;
373 }
374
375 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
376                             struct igbvf_buffer *buffer_info)
377 {
378         buffer_info->dma = 0;
379         if (buffer_info->skb) {
380                 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
381                               DMA_TO_DEVICE);
382                 dev_kfree_skb_any(buffer_info->skb);
383                 buffer_info->skb = NULL;
384         }
385         buffer_info->time_stamp = 0;
386 }
387
388 static void igbvf_print_tx_hang(struct igbvf_adapter *adapter)
389 {
390         struct igbvf_ring *tx_ring = adapter->tx_ring;
391         unsigned int i = tx_ring->next_to_clean;
392         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
393         union e1000_adv_tx_desc *eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
394
395         /* detected Tx unit hang */
396         dev_err(&adapter->pdev->dev,
397                 "Detected Tx Unit Hang:\n"
398                 "  TDH                  <%x>\n"
399                 "  TDT                  <%x>\n"
400                 "  next_to_use          <%x>\n"
401                 "  next_to_clean        <%x>\n"
402                 "buffer_info[next_to_clean]:\n"
403                 "  time_stamp           <%lx>\n"
404                 "  next_to_watch        <%x>\n"
405                 "  jiffies              <%lx>\n"
406                 "  next_to_watch.status <%x>\n",
407                 readl(adapter->hw.hw_addr + tx_ring->head),
408                 readl(adapter->hw.hw_addr + tx_ring->tail),
409                 tx_ring->next_to_use,
410                 tx_ring->next_to_clean,
411                 tx_ring->buffer_info[eop].time_stamp,
412                 eop,
413                 jiffies,
414                 eop_desc->wb.status);
415 }
416
417 /**
418  * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
419  * @adapter: board private structure
420  *
421  * Return 0 on success, negative on failure
422  **/
423 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
424                              struct igbvf_ring *tx_ring)
425 {
426         struct pci_dev *pdev = adapter->pdev;
427         int size;
428
429         size = sizeof(struct igbvf_buffer) * tx_ring->count;
430         tx_ring->buffer_info = vmalloc(size);
431         if (!tx_ring->buffer_info)
432                 goto err;
433         memset(tx_ring->buffer_info, 0, size);
434
435         /* round up to nearest 4K */
436         tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
437         tx_ring->size = ALIGN(tx_ring->size, 4096);
438
439         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
440                                              &tx_ring->dma);
441
442         if (!tx_ring->desc)
443                 goto err;
444
445         tx_ring->adapter = adapter;
446         tx_ring->next_to_use = 0;
447         tx_ring->next_to_clean = 0;
448
449         return 0;
450 err:
451         vfree(tx_ring->buffer_info);
452         dev_err(&adapter->pdev->dev,
453                 "Unable to allocate memory for the transmit descriptor ring\n");
454         return -ENOMEM;
455 }
456
457 /**
458  * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
459  * @adapter: board private structure
460  *
461  * Returns 0 on success, negative on failure
462  **/
463 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
464                              struct igbvf_ring *rx_ring)
465 {
466         struct pci_dev *pdev = adapter->pdev;
467         int size, desc_len;
468
469         size = sizeof(struct igbvf_buffer) * rx_ring->count;
470         rx_ring->buffer_info = vmalloc(size);
471         if (!rx_ring->buffer_info)
472                 goto err;
473         memset(rx_ring->buffer_info, 0, size);
474
475         desc_len = sizeof(union e1000_adv_rx_desc);
476
477         /* Round up to nearest 4K */
478         rx_ring->size = rx_ring->count * desc_len;
479         rx_ring->size = ALIGN(rx_ring->size, 4096);
480
481         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
482                                              &rx_ring->dma);
483
484         if (!rx_ring->desc)
485                 goto err;
486
487         rx_ring->next_to_clean = 0;
488         rx_ring->next_to_use = 0;
489
490         rx_ring->adapter = adapter;
491
492         return 0;
493
494 err:
495         vfree(rx_ring->buffer_info);
496         rx_ring->buffer_info = NULL;
497         dev_err(&adapter->pdev->dev,
498                 "Unable to allocate memory for the receive descriptor ring\n");
499         return -ENOMEM;
500 }
501
502 /**
503  * igbvf_clean_tx_ring - Free Tx Buffers
504  * @tx_ring: ring to be cleaned
505  **/
506 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
507 {
508         struct igbvf_adapter *adapter = tx_ring->adapter;
509         struct igbvf_buffer *buffer_info;
510         unsigned long size;
511         unsigned int i;
512
513         if (!tx_ring->buffer_info)
514                 return;
515
516         /* Free all the Tx ring sk_buffs */
517         for (i = 0; i < tx_ring->count; i++) {
518                 buffer_info = &tx_ring->buffer_info[i];
519                 igbvf_put_txbuf(adapter, buffer_info);
520         }
521
522         size = sizeof(struct igbvf_buffer) * tx_ring->count;
523         memset(tx_ring->buffer_info, 0, size);
524
525         /* Zero out the descriptor ring */
526         memset(tx_ring->desc, 0, tx_ring->size);
527
528         tx_ring->next_to_use = 0;
529         tx_ring->next_to_clean = 0;
530
531         writel(0, adapter->hw.hw_addr + tx_ring->head);
532         writel(0, adapter->hw.hw_addr + tx_ring->tail);
533 }
534
535 /**
536  * igbvf_free_tx_resources - Free Tx Resources per Queue
537  * @tx_ring: ring to free resources from
538  *
539  * Free all transmit software resources
540  **/
541 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
542 {
543         struct pci_dev *pdev = tx_ring->adapter->pdev;
544
545         igbvf_clean_tx_ring(tx_ring);
546
547         vfree(tx_ring->buffer_info);
548         tx_ring->buffer_info = NULL;
549
550         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
551
552         tx_ring->desc = NULL;
553 }
554
555 /**
556  * igbvf_clean_rx_ring - Free Rx Buffers per Queue
557  * @adapter: board private structure
558  **/
559 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
560 {
561         struct igbvf_adapter *adapter = rx_ring->adapter;
562         struct igbvf_buffer *buffer_info;
563         struct pci_dev *pdev = adapter->pdev;
564         unsigned long size;
565         unsigned int i;
566
567         if (!rx_ring->buffer_info)
568                 return;
569
570         /* Free all the Rx ring sk_buffs */
571         for (i = 0; i < rx_ring->count; i++) {
572                 buffer_info = &rx_ring->buffer_info[i];
573                 if (buffer_info->dma) {
574                         if (adapter->rx_ps_hdr_size){
575                                 pci_unmap_single(pdev, buffer_info->dma,
576                                                  adapter->rx_ps_hdr_size,
577                                                  PCI_DMA_FROMDEVICE);
578                         } else {
579                                 pci_unmap_single(pdev, buffer_info->dma,
580                                                  adapter->rx_buffer_len,
581                                                  PCI_DMA_FROMDEVICE);
582                         }
583                         buffer_info->dma = 0;
584                 }
585
586                 if (buffer_info->skb) {
587                         dev_kfree_skb(buffer_info->skb);
588                         buffer_info->skb = NULL;
589                 }
590
591                 if (buffer_info->page) {
592                         if (buffer_info->page_dma)
593                                 pci_unmap_page(pdev, buffer_info->page_dma,
594                                                PAGE_SIZE / 2,
595                                                PCI_DMA_FROMDEVICE);
596                         put_page(buffer_info->page);
597                         buffer_info->page = NULL;
598                         buffer_info->page_dma = 0;
599                         buffer_info->page_offset = 0;
600                 }
601         }
602
603         size = sizeof(struct igbvf_buffer) * rx_ring->count;
604         memset(rx_ring->buffer_info, 0, size);
605
606         /* Zero out the descriptor ring */
607         memset(rx_ring->desc, 0, rx_ring->size);
608
609         rx_ring->next_to_clean = 0;
610         rx_ring->next_to_use = 0;
611
612         writel(0, adapter->hw.hw_addr + rx_ring->head);
613         writel(0, adapter->hw.hw_addr + rx_ring->tail);
614 }
615
616 /**
617  * igbvf_free_rx_resources - Free Rx Resources
618  * @rx_ring: ring to clean the resources from
619  *
620  * Free all receive software resources
621  **/
622
623 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
624 {
625         struct pci_dev *pdev = rx_ring->adapter->pdev;
626
627         igbvf_clean_rx_ring(rx_ring);
628
629         vfree(rx_ring->buffer_info);
630         rx_ring->buffer_info = NULL;
631
632         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
633                           rx_ring->dma);
634         rx_ring->desc = NULL;
635 }
636
637 /**
638  * igbvf_update_itr - update the dynamic ITR value based on statistics
639  * @adapter: pointer to adapter
640  * @itr_setting: current adapter->itr
641  * @packets: the number of packets during this measurement interval
642  * @bytes: the number of bytes during this measurement interval
643  *
644  *      Stores a new ITR value based on packets and byte
645  *      counts during the last interrupt.  The advantage of per interrupt
646  *      computation is faster updates and more accurate ITR for the current
647  *      traffic pattern.  Constants in this function were computed
648  *      based on theoretical maximum wire speed and thresholds were set based
649  *      on testing data as well as attempting to minimize response time
650  *      while increasing bulk throughput.  This functionality is controlled
651  *      by the InterruptThrottleRate module parameter.
652  **/
653 static unsigned int igbvf_update_itr(struct igbvf_adapter *adapter,
654                                      u16 itr_setting, int packets,
655                                      int bytes)
656 {
657         unsigned int retval = itr_setting;
658
659         if (packets == 0)
660                 goto update_itr_done;
661
662         switch (itr_setting) {
663         case lowest_latency:
664                 /* handle TSO and jumbo frames */
665                 if (bytes/packets > 8000)
666                         retval = bulk_latency;
667                 else if ((packets < 5) && (bytes > 512))
668                         retval = low_latency;
669                 break;
670         case low_latency:  /* 50 usec aka 20000 ints/s */
671                 if (bytes > 10000) {
672                         /* this if handles the TSO accounting */
673                         if (bytes/packets > 8000)
674                                 retval = bulk_latency;
675                         else if ((packets < 10) || ((bytes/packets) > 1200))
676                                 retval = bulk_latency;
677                         else if ((packets > 35))
678                                 retval = lowest_latency;
679                 } else if (bytes/packets > 2000) {
680                         retval = bulk_latency;
681                 } else if (packets <= 2 && bytes < 512) {
682                         retval = lowest_latency;
683                 }
684                 break;
685         case bulk_latency: /* 250 usec aka 4000 ints/s */
686                 if (bytes > 25000) {
687                         if (packets > 35)
688                                 retval = low_latency;
689                 } else if (bytes < 6000) {
690                         retval = low_latency;
691                 }
692                 break;
693         }
694
695 update_itr_done:
696         return retval;
697 }
698
699 static void igbvf_set_itr(struct igbvf_adapter *adapter)
700 {
701         struct e1000_hw *hw = &adapter->hw;
702         u16 current_itr;
703         u32 new_itr = adapter->itr;
704
705         adapter->tx_itr = igbvf_update_itr(adapter, adapter->tx_itr,
706                                            adapter->total_tx_packets,
707                                            adapter->total_tx_bytes);
708         /* conservative mode (itr 3) eliminates the lowest_latency setting */
709         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
710                 adapter->tx_itr = low_latency;
711
712         adapter->rx_itr = igbvf_update_itr(adapter, adapter->rx_itr,
713                                            adapter->total_rx_packets,
714                                            adapter->total_rx_bytes);
715         /* conservative mode (itr 3) eliminates the lowest_latency setting */
716         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
717                 adapter->rx_itr = low_latency;
718
719         current_itr = max(adapter->rx_itr, adapter->tx_itr);
720
721         switch (current_itr) {
722         /* counts and packets in update_itr are dependent on these numbers */
723         case lowest_latency:
724                 new_itr = 70000;
725                 break;
726         case low_latency:
727                 new_itr = 20000; /* aka hwitr = ~200 */
728                 break;
729         case bulk_latency:
730                 new_itr = 4000;
731                 break;
732         default:
733                 break;
734         }
735
736         if (new_itr != adapter->itr) {
737                 /*
738                  * this attempts to bias the interrupt rate towards Bulk
739                  * by adding intermediate steps when interrupt rate is
740                  * increasing
741                  */
742                 new_itr = new_itr > adapter->itr ?
743                              min(adapter->itr + (new_itr >> 2), new_itr) :
744                              new_itr;
745                 adapter->itr = new_itr;
746                 adapter->rx_ring->itr_val = 1952;
747
748                 if (adapter->msix_entries)
749                         adapter->rx_ring->set_itr = 1;
750                 else
751                         ew32(ITR, 1952);
752         }
753 }
754
755 /**
756  * igbvf_clean_tx_irq - Reclaim resources after transmit completes
757  * @adapter: board private structure
758  * returns true if ring is completely cleaned
759  **/
760 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
761 {
762         struct igbvf_adapter *adapter = tx_ring->adapter;
763         struct e1000_hw *hw = &adapter->hw;
764         struct net_device *netdev = adapter->netdev;
765         struct igbvf_buffer *buffer_info;
766         struct sk_buff *skb;
767         union e1000_adv_tx_desc *tx_desc, *eop_desc;
768         unsigned int total_bytes = 0, total_packets = 0;
769         unsigned int i, eop, count = 0;
770         bool cleaned = false;
771
772         i = tx_ring->next_to_clean;
773         eop = tx_ring->buffer_info[i].next_to_watch;
774         eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
775
776         while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
777                (count < tx_ring->count)) {
778                 for (cleaned = false; !cleaned; count++) {
779                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
780                         buffer_info = &tx_ring->buffer_info[i];
781                         cleaned = (i == eop);
782                         skb = buffer_info->skb;
783
784                         if (skb) {
785                                 unsigned int segs, bytecount;
786
787                                 /* gso_segs is currently only valid for tcp */
788                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
789                                 /* multiply data chunks by size of headers */
790                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
791                                             skb->len;
792                                 total_packets += segs;
793                                 total_bytes += bytecount;
794                         }
795
796                         igbvf_put_txbuf(adapter, buffer_info);
797                         tx_desc->wb.status = 0;
798
799                         i++;
800                         if (i == tx_ring->count)
801                                 i = 0;
802                 }
803                 eop = tx_ring->buffer_info[i].next_to_watch;
804                 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
805         }
806
807         tx_ring->next_to_clean = i;
808
809         if (unlikely(count &&
810                      netif_carrier_ok(netdev) &&
811                      igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
812                 /* Make sure that anybody stopping the queue after this
813                  * sees the new next_to_clean.
814                  */
815                 smp_mb();
816                 if (netif_queue_stopped(netdev) &&
817                     !(test_bit(__IGBVF_DOWN, &adapter->state))) {
818                         netif_wake_queue(netdev);
819                         ++adapter->restart_queue;
820                 }
821         }
822
823         if (adapter->detect_tx_hung) {
824                 /* Detect a transmit hang in hardware, this serializes the
825                  * check with the clearing of time_stamp and movement of i */
826                 adapter->detect_tx_hung = false;
827                 if (tx_ring->buffer_info[i].time_stamp &&
828                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
829                                (adapter->tx_timeout_factor * HZ))
830                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
831
832                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
833                         /* detected Tx unit hang */
834                         igbvf_print_tx_hang(adapter);
835
836                         netif_stop_queue(netdev);
837                 }
838         }
839         adapter->net_stats.tx_bytes += total_bytes;
840         adapter->net_stats.tx_packets += total_packets;
841         return (count < tx_ring->count);
842 }
843
844 static irqreturn_t igbvf_msix_other(int irq, void *data)
845 {
846         struct net_device *netdev = data;
847         struct igbvf_adapter *adapter = netdev_priv(netdev);
848         struct e1000_hw *hw = &adapter->hw;
849
850         adapter->int_counter1++;
851
852         netif_carrier_off(netdev);
853         hw->mac.get_link_status = 1;
854         if (!test_bit(__IGBVF_DOWN, &adapter->state))
855                 mod_timer(&adapter->watchdog_timer, jiffies + 1);
856
857         ew32(EIMS, adapter->eims_other);
858
859         return IRQ_HANDLED;
860 }
861
862 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
863 {
864         struct net_device *netdev = data;
865         struct igbvf_adapter *adapter = netdev_priv(netdev);
866         struct e1000_hw *hw = &adapter->hw;
867         struct igbvf_ring *tx_ring = adapter->tx_ring;
868
869
870         adapter->total_tx_bytes = 0;
871         adapter->total_tx_packets = 0;
872
873         /* auto mask will automatically reenable the interrupt when we write
874          * EICS */
875         if (!igbvf_clean_tx_irq(tx_ring))
876                 /* Ring was not completely cleaned, so fire another interrupt */
877                 ew32(EICS, tx_ring->eims_value);
878         else
879                 ew32(EIMS, tx_ring->eims_value);
880
881         return IRQ_HANDLED;
882 }
883
884 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
885 {
886         struct net_device *netdev = data;
887         struct igbvf_adapter *adapter = netdev_priv(netdev);
888
889         adapter->int_counter0++;
890
891         /* Write the ITR value calculated at the end of the
892          * previous interrupt.
893          */
894         if (adapter->rx_ring->set_itr) {
895                 writel(adapter->rx_ring->itr_val,
896                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
897                 adapter->rx_ring->set_itr = 0;
898         }
899
900         if (napi_schedule_prep(&adapter->rx_ring->napi)) {
901                 adapter->total_rx_bytes = 0;
902                 adapter->total_rx_packets = 0;
903                 __napi_schedule(&adapter->rx_ring->napi);
904         }
905
906         return IRQ_HANDLED;
907 }
908
909 #define IGBVF_NO_QUEUE -1
910
911 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
912                                 int tx_queue, int msix_vector)
913 {
914         struct e1000_hw *hw = &adapter->hw;
915         u32 ivar, index;
916
917         /* 82576 uses a table-based method for assigning vectors.
918            Each queue has a single entry in the table to which we write
919            a vector number along with a "valid" bit.  Sadly, the layout
920            of the table is somewhat counterintuitive. */
921         if (rx_queue > IGBVF_NO_QUEUE) {
922                 index = (rx_queue >> 1);
923                 ivar = array_er32(IVAR0, index);
924                 if (rx_queue & 0x1) {
925                         /* vector goes into third byte of register */
926                         ivar = ivar & 0xFF00FFFF;
927                         ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
928                 } else {
929                         /* vector goes into low byte of register */
930                         ivar = ivar & 0xFFFFFF00;
931                         ivar |= msix_vector | E1000_IVAR_VALID;
932                 }
933                 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
934                 array_ew32(IVAR0, index, ivar);
935         }
936         if (tx_queue > IGBVF_NO_QUEUE) {
937                 index = (tx_queue >> 1);
938                 ivar = array_er32(IVAR0, index);
939                 if (tx_queue & 0x1) {
940                         /* vector goes into high byte of register */
941                         ivar = ivar & 0x00FFFFFF;
942                         ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
943                 } else {
944                         /* vector goes into second byte of register */
945                         ivar = ivar & 0xFFFF00FF;
946                         ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
947                 }
948                 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
949                 array_ew32(IVAR0, index, ivar);
950         }
951 }
952
953 /**
954  * igbvf_configure_msix - Configure MSI-X hardware
955  *
956  * igbvf_configure_msix sets up the hardware to properly
957  * generate MSI-X interrupts.
958  **/
959 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
960 {
961         u32 tmp;
962         struct e1000_hw *hw = &adapter->hw;
963         struct igbvf_ring *tx_ring = adapter->tx_ring;
964         struct igbvf_ring *rx_ring = adapter->rx_ring;
965         int vector = 0;
966
967         adapter->eims_enable_mask = 0;
968
969         igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
970         adapter->eims_enable_mask |= tx_ring->eims_value;
971         if (tx_ring->itr_val)
972                 writel(tx_ring->itr_val,
973                        hw->hw_addr + tx_ring->itr_register);
974         else
975                 writel(1952, hw->hw_addr + tx_ring->itr_register);
976
977         igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
978         adapter->eims_enable_mask |= rx_ring->eims_value;
979         if (rx_ring->itr_val)
980                 writel(rx_ring->itr_val,
981                        hw->hw_addr + rx_ring->itr_register);
982         else
983                 writel(1952, hw->hw_addr + rx_ring->itr_register);
984
985         /* set vector for other causes, i.e. link changes */
986
987         tmp = (vector++ | E1000_IVAR_VALID);
988
989         ew32(IVAR_MISC, tmp);
990
991         adapter->eims_enable_mask = (1 << (vector)) - 1;
992         adapter->eims_other = 1 << (vector - 1);
993         e1e_flush();
994 }
995
996 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
997 {
998         if (adapter->msix_entries) {
999                 pci_disable_msix(adapter->pdev);
1000                 kfree(adapter->msix_entries);
1001                 adapter->msix_entries = NULL;
1002         }
1003 }
1004
1005 /**
1006  * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1007  *
1008  * Attempt to configure interrupts using the best available
1009  * capabilities of the hardware and kernel.
1010  **/
1011 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1012 {
1013         int err = -ENOMEM;
1014         int i;
1015
1016         /* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
1017         adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1018                                         GFP_KERNEL);
1019         if (adapter->msix_entries) {
1020                 for (i = 0; i < 3; i++)
1021                         adapter->msix_entries[i].entry = i;
1022
1023                 err = pci_enable_msix(adapter->pdev,
1024                                       adapter->msix_entries, 3);
1025         }
1026
1027         if (err) {
1028                 /* MSI-X failed */
1029                 dev_err(&adapter->pdev->dev,
1030                         "Failed to initialize MSI-X interrupts.\n");
1031                 igbvf_reset_interrupt_capability(adapter);
1032         }
1033 }
1034
1035 /**
1036  * igbvf_request_msix - Initialize MSI-X interrupts
1037  *
1038  * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1039  * kernel.
1040  **/
1041 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1042 {
1043         struct net_device *netdev = adapter->netdev;
1044         int err = 0, vector = 0;
1045
1046         if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1047                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1048                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1049         } else {
1050                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1051                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1052         }
1053
1054         err = request_irq(adapter->msix_entries[vector].vector,
1055                           &igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1056                           netdev);
1057         if (err)
1058                 goto out;
1059
1060         adapter->tx_ring->itr_register = E1000_EITR(vector);
1061         adapter->tx_ring->itr_val = 1952;
1062         vector++;
1063
1064         err = request_irq(adapter->msix_entries[vector].vector,
1065                           &igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1066                           netdev);
1067         if (err)
1068                 goto out;
1069
1070         adapter->rx_ring->itr_register = E1000_EITR(vector);
1071         adapter->rx_ring->itr_val = 1952;
1072         vector++;
1073
1074         err = request_irq(adapter->msix_entries[vector].vector,
1075                           &igbvf_msix_other, 0, netdev->name, netdev);
1076         if (err)
1077                 goto out;
1078
1079         igbvf_configure_msix(adapter);
1080         return 0;
1081 out:
1082         return err;
1083 }
1084
1085 /**
1086  * igbvf_alloc_queues - Allocate memory for all rings
1087  * @adapter: board private structure to initialize
1088  **/
1089 static int __devinit igbvf_alloc_queues(struct igbvf_adapter *adapter)
1090 {
1091         struct net_device *netdev = adapter->netdev;
1092
1093         adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1094         if (!adapter->tx_ring)
1095                 return -ENOMEM;
1096
1097         adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1098         if (!adapter->rx_ring) {
1099                 kfree(adapter->tx_ring);
1100                 return -ENOMEM;
1101         }
1102
1103         netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1104
1105         return 0;
1106 }
1107
1108 /**
1109  * igbvf_request_irq - initialize interrupts
1110  *
1111  * Attempts to configure interrupts using the best available
1112  * capabilities of the hardware and kernel.
1113  **/
1114 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1115 {
1116         int err = -1;
1117
1118         /* igbvf supports msi-x only */
1119         if (adapter->msix_entries)
1120                 err = igbvf_request_msix(adapter);
1121
1122         if (!err)
1123                 return err;
1124
1125         dev_err(&adapter->pdev->dev,
1126                 "Unable to allocate interrupt, Error: %d\n", err);
1127
1128         return err;
1129 }
1130
1131 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1132 {
1133         struct net_device *netdev = adapter->netdev;
1134         int vector;
1135
1136         if (adapter->msix_entries) {
1137                 for (vector = 0; vector < 3; vector++)
1138                         free_irq(adapter->msix_entries[vector].vector, netdev);
1139         }
1140 }
1141
1142 /**
1143  * igbvf_irq_disable - Mask off interrupt generation on the NIC
1144  **/
1145 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1146 {
1147         struct e1000_hw *hw = &adapter->hw;
1148
1149         ew32(EIMC, ~0);
1150
1151         if (adapter->msix_entries)
1152                 ew32(EIAC, 0);
1153 }
1154
1155 /**
1156  * igbvf_irq_enable - Enable default interrupt generation settings
1157  **/
1158 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1159 {
1160         struct e1000_hw *hw = &adapter->hw;
1161
1162         ew32(EIAC, adapter->eims_enable_mask);
1163         ew32(EIAM, adapter->eims_enable_mask);
1164         ew32(EIMS, adapter->eims_enable_mask);
1165 }
1166
1167 /**
1168  * igbvf_poll - NAPI Rx polling callback
1169  * @napi: struct associated with this polling callback
1170  * @budget: amount of packets driver is allowed to process this poll
1171  **/
1172 static int igbvf_poll(struct napi_struct *napi, int budget)
1173 {
1174         struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1175         struct igbvf_adapter *adapter = rx_ring->adapter;
1176         struct e1000_hw *hw = &adapter->hw;
1177         int work_done = 0;
1178
1179         igbvf_clean_rx_irq(adapter, &work_done, budget);
1180
1181         /* If not enough Rx work done, exit the polling mode */
1182         if (work_done < budget) {
1183                 napi_complete(napi);
1184
1185                 if (adapter->itr_setting & 3)
1186                         igbvf_set_itr(adapter);
1187
1188                 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1189                         ew32(EIMS, adapter->rx_ring->eims_value);
1190         }
1191
1192         return work_done;
1193 }
1194
1195 /**
1196  * igbvf_set_rlpml - set receive large packet maximum length
1197  * @adapter: board private structure
1198  *
1199  * Configure the maximum size of packets that will be received
1200  */
1201 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1202 {
1203         int max_frame_size = adapter->max_frame_size;
1204         struct e1000_hw *hw = &adapter->hw;
1205
1206         if (adapter->vlgrp)
1207                 max_frame_size += VLAN_TAG_SIZE;
1208
1209         e1000_rlpml_set_vf(hw, max_frame_size);
1210 }
1211
1212 static void igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1213 {
1214         struct igbvf_adapter *adapter = netdev_priv(netdev);
1215         struct e1000_hw *hw = &adapter->hw;
1216
1217         if (hw->mac.ops.set_vfta(hw, vid, true))
1218                 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1219 }
1220
1221 static void igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1222 {
1223         struct igbvf_adapter *adapter = netdev_priv(netdev);
1224         struct e1000_hw *hw = &adapter->hw;
1225
1226         igbvf_irq_disable(adapter);
1227         vlan_group_set_device(adapter->vlgrp, vid, NULL);
1228
1229         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1230                 igbvf_irq_enable(adapter);
1231
1232         if (hw->mac.ops.set_vfta(hw, vid, false))
1233                 dev_err(&adapter->pdev->dev,
1234                         "Failed to remove vlan id %d\n", vid);
1235 }
1236
1237 static void igbvf_vlan_rx_register(struct net_device *netdev,
1238                                    struct vlan_group *grp)
1239 {
1240         struct igbvf_adapter *adapter = netdev_priv(netdev);
1241
1242         adapter->vlgrp = grp;
1243 }
1244
1245 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1246 {
1247         u16 vid;
1248
1249         if (!adapter->vlgrp)
1250                 return;
1251
1252         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1253                 if (!vlan_group_get_device(adapter->vlgrp, vid))
1254                         continue;
1255                 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1256         }
1257
1258         igbvf_set_rlpml(adapter);
1259 }
1260
1261 /**
1262  * igbvf_configure_tx - Configure Transmit Unit after Reset
1263  * @adapter: board private structure
1264  *
1265  * Configure the Tx unit of the MAC after a reset.
1266  **/
1267 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1268 {
1269         struct e1000_hw *hw = &adapter->hw;
1270         struct igbvf_ring *tx_ring = adapter->tx_ring;
1271         u64 tdba;
1272         u32 txdctl, dca_txctrl;
1273
1274         /* disable transmits */
1275         txdctl = er32(TXDCTL(0));
1276         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1277         msleep(10);
1278
1279         /* Setup the HW Tx Head and Tail descriptor pointers */
1280         ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1281         tdba = tx_ring->dma;
1282         ew32(TDBAL(0), (tdba & DMA_32BIT_MASK));
1283         ew32(TDBAH(0), (tdba >> 32));
1284         ew32(TDH(0), 0);
1285         ew32(TDT(0), 0);
1286         tx_ring->head = E1000_TDH(0);
1287         tx_ring->tail = E1000_TDT(0);
1288
1289         /* Turn off Relaxed Ordering on head write-backs.  The writebacks
1290          * MUST be delivered in order or it will completely screw up
1291          * our bookeeping.
1292          */
1293         dca_txctrl = er32(DCA_TXCTRL(0));
1294         dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1295         ew32(DCA_TXCTRL(0), dca_txctrl);
1296
1297         /* enable transmits */
1298         txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1299         ew32(TXDCTL(0), txdctl);
1300
1301         /* Setup Transmit Descriptor Settings for eop descriptor */
1302         adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1303
1304         /* enable Report Status bit */
1305         adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1306
1307         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
1308 }
1309
1310 /**
1311  * igbvf_setup_srrctl - configure the receive control registers
1312  * @adapter: Board private structure
1313  **/
1314 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1315 {
1316         struct e1000_hw *hw = &adapter->hw;
1317         u32 srrctl = 0;
1318
1319         srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1320                     E1000_SRRCTL_BSIZEHDR_MASK |
1321                     E1000_SRRCTL_BSIZEPKT_MASK);
1322
1323         /* Enable queue drop to avoid head of line blocking */
1324         srrctl |= E1000_SRRCTL_DROP_EN;
1325
1326         /* Setup buffer sizes */
1327         srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1328                   E1000_SRRCTL_BSIZEPKT_SHIFT;
1329
1330         if (adapter->rx_buffer_len < 2048) {
1331                 adapter->rx_ps_hdr_size = 0;
1332                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1333         } else {
1334                 adapter->rx_ps_hdr_size = 128;
1335                 srrctl |= adapter->rx_ps_hdr_size <<
1336                           E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1337                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1338         }
1339
1340         ew32(SRRCTL(0), srrctl);
1341 }
1342
1343 /**
1344  * igbvf_configure_rx - Configure Receive Unit after Reset
1345  * @adapter: board private structure
1346  *
1347  * Configure the Rx unit of the MAC after a reset.
1348  **/
1349 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1350 {
1351         struct e1000_hw *hw = &adapter->hw;
1352         struct igbvf_ring *rx_ring = adapter->rx_ring;
1353         u64 rdba;
1354         u32 rdlen, rxdctl;
1355
1356         /* disable receives */
1357         rxdctl = er32(RXDCTL(0));
1358         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1359         msleep(10);
1360
1361         rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1362
1363         /*
1364          * Setup the HW Rx Head and Tail Descriptor Pointers and
1365          * the Base and Length of the Rx Descriptor Ring
1366          */
1367         rdba = rx_ring->dma;
1368         ew32(RDBAL(0), (rdba & DMA_32BIT_MASK));
1369         ew32(RDBAH(0), (rdba >> 32));
1370         ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1371         rx_ring->head = E1000_RDH(0);
1372         rx_ring->tail = E1000_RDT(0);
1373         ew32(RDH(0), 0);
1374         ew32(RDT(0), 0);
1375
1376         rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1377         rxdctl &= 0xFFF00000;
1378         rxdctl |= IGBVF_RX_PTHRESH;
1379         rxdctl |= IGBVF_RX_HTHRESH << 8;
1380         rxdctl |= IGBVF_RX_WTHRESH << 16;
1381
1382         igbvf_set_rlpml(adapter);
1383
1384         /* enable receives */
1385         ew32(RXDCTL(0), rxdctl);
1386 }
1387
1388 /**
1389  * igbvf_set_multi - Multicast and Promiscuous mode set
1390  * @netdev: network interface device structure
1391  *
1392  * The set_multi entry point is called whenever the multicast address
1393  * list or the network interface flags are updated.  This routine is
1394  * responsible for configuring the hardware for proper multicast,
1395  * promiscuous mode, and all-multi behavior.
1396  **/
1397 static void igbvf_set_multi(struct net_device *netdev)
1398 {
1399         struct igbvf_adapter *adapter = netdev_priv(netdev);
1400         struct e1000_hw *hw = &adapter->hw;
1401         struct dev_mc_list *mc_ptr;
1402         u8  *mta_list = NULL;
1403         int i;
1404
1405         if (netdev->mc_count) {
1406                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
1407                 if (!mta_list) {
1408                         dev_err(&adapter->pdev->dev,
1409                                 "failed to allocate multicast filter list\n");
1410                         return;
1411                 }
1412         }
1413
1414         /* prepare a packed array of only addresses. */
1415         mc_ptr = netdev->mc_list;
1416
1417         for (i = 0; i < netdev->mc_count; i++) {
1418                 if (!mc_ptr)
1419                         break;
1420                 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
1421                        ETH_ALEN);
1422                 mc_ptr = mc_ptr->next;
1423         }
1424
1425         hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1426         kfree(mta_list);
1427 }
1428
1429 /**
1430  * igbvf_configure - configure the hardware for Rx and Tx
1431  * @adapter: private board structure
1432  **/
1433 static void igbvf_configure(struct igbvf_adapter *adapter)
1434 {
1435         igbvf_set_multi(adapter->netdev);
1436
1437         igbvf_restore_vlan(adapter);
1438
1439         igbvf_configure_tx(adapter);
1440         igbvf_setup_srrctl(adapter);
1441         igbvf_configure_rx(adapter);
1442         igbvf_alloc_rx_buffers(adapter->rx_ring,
1443                                igbvf_desc_unused(adapter->rx_ring));
1444 }
1445
1446 /* igbvf_reset - bring the hardware into a known good state
1447  *
1448  * This function boots the hardware and enables some settings that
1449  * require a configuration cycle of the hardware - those cannot be
1450  * set/changed during runtime. After reset the device needs to be
1451  * properly configured for Rx, Tx etc.
1452  */
1453 static void igbvf_reset(struct igbvf_adapter *adapter)
1454 {
1455         struct e1000_mac_info *mac = &adapter->hw.mac;
1456         struct net_device *netdev = adapter->netdev;
1457         struct e1000_hw *hw = &adapter->hw;
1458
1459         /* Allow time for pending master requests to run */
1460         if (mac->ops.reset_hw(hw))
1461                 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1462
1463         mac->ops.init_hw(hw);
1464
1465         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1466                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1467                        netdev->addr_len);
1468                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1469                        netdev->addr_len);
1470         }
1471 }
1472
1473 int igbvf_up(struct igbvf_adapter *adapter)
1474 {
1475         struct e1000_hw *hw = &adapter->hw;
1476
1477         /* hardware has been reset, we need to reload some things */
1478         igbvf_configure(adapter);
1479
1480         clear_bit(__IGBVF_DOWN, &adapter->state);
1481
1482         napi_enable(&adapter->rx_ring->napi);
1483         if (adapter->msix_entries)
1484                 igbvf_configure_msix(adapter);
1485
1486         /* Clear any pending interrupts. */
1487         er32(EICR);
1488         igbvf_irq_enable(adapter);
1489
1490         /* start the watchdog */
1491         hw->mac.get_link_status = 1;
1492         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1493
1494
1495         return 0;
1496 }
1497
1498 void igbvf_down(struct igbvf_adapter *adapter)
1499 {
1500         struct net_device *netdev = adapter->netdev;
1501         struct e1000_hw *hw = &adapter->hw;
1502         u32 rxdctl, txdctl;
1503
1504         /*
1505          * signal that we're down so the interrupt handler does not
1506          * reschedule our watchdog timer
1507          */
1508         set_bit(__IGBVF_DOWN, &adapter->state);
1509
1510         /* disable receives in the hardware */
1511         rxdctl = er32(RXDCTL(0));
1512         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1513
1514         netif_stop_queue(netdev);
1515
1516         /* disable transmits in the hardware */
1517         txdctl = er32(TXDCTL(0));
1518         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1519
1520         /* flush both disables and wait for them to finish */
1521         e1e_flush();
1522         msleep(10);
1523
1524         napi_disable(&adapter->rx_ring->napi);
1525
1526         igbvf_irq_disable(adapter);
1527
1528         del_timer_sync(&adapter->watchdog_timer);
1529
1530         netdev->tx_queue_len = adapter->tx_queue_len;
1531         netif_carrier_off(netdev);
1532
1533         /* record the stats before reset*/
1534         igbvf_update_stats(adapter);
1535
1536         adapter->link_speed = 0;
1537         adapter->link_duplex = 0;
1538
1539         igbvf_reset(adapter);
1540         igbvf_clean_tx_ring(adapter->tx_ring);
1541         igbvf_clean_rx_ring(adapter->rx_ring);
1542 }
1543
1544 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1545 {
1546         might_sleep();
1547         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1548                 msleep(1);
1549         igbvf_down(adapter);
1550         igbvf_up(adapter);
1551         clear_bit(__IGBVF_RESETTING, &adapter->state);
1552 }
1553
1554 /**
1555  * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1556  * @adapter: board private structure to initialize
1557  *
1558  * igbvf_sw_init initializes the Adapter private data structure.
1559  * Fields are initialized based on PCI device information and
1560  * OS network device settings (MTU size).
1561  **/
1562 static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1563 {
1564         struct net_device *netdev = adapter->netdev;
1565         s32 rc;
1566
1567         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1568         adapter->rx_ps_hdr_size = 0;
1569         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1570         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1571
1572         adapter->tx_int_delay = 8;
1573         adapter->tx_abs_int_delay = 32;
1574         adapter->rx_int_delay = 0;
1575         adapter->rx_abs_int_delay = 8;
1576         adapter->itr_setting = 3;
1577         adapter->itr = 20000;
1578
1579         /* Set various function pointers */
1580         adapter->ei->init_ops(&adapter->hw);
1581
1582         rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1583         if (rc)
1584                 return rc;
1585
1586         rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1587         if (rc)
1588                 return rc;
1589
1590         igbvf_set_interrupt_capability(adapter);
1591
1592         if (igbvf_alloc_queues(adapter))
1593                 return -ENOMEM;
1594
1595         spin_lock_init(&adapter->tx_queue_lock);
1596
1597         /* Explicitly disable IRQ since the NIC can be in any state. */
1598         igbvf_irq_disable(adapter);
1599
1600         spin_lock_init(&adapter->stats_lock);
1601
1602         set_bit(__IGBVF_DOWN, &adapter->state);
1603         return 0;
1604 }
1605
1606 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1607 {
1608         struct e1000_hw *hw = &adapter->hw;
1609
1610         adapter->stats.last_gprc = er32(VFGPRC);
1611         adapter->stats.last_gorc = er32(VFGORC);
1612         adapter->stats.last_gptc = er32(VFGPTC);
1613         adapter->stats.last_gotc = er32(VFGOTC);
1614         adapter->stats.last_mprc = er32(VFMPRC);
1615         adapter->stats.last_gotlbc = er32(VFGOTLBC);
1616         adapter->stats.last_gptlbc = er32(VFGPTLBC);
1617         adapter->stats.last_gorlbc = er32(VFGORLBC);
1618         adapter->stats.last_gprlbc = er32(VFGPRLBC);
1619
1620         adapter->stats.base_gprc = er32(VFGPRC);
1621         adapter->stats.base_gorc = er32(VFGORC);
1622         adapter->stats.base_gptc = er32(VFGPTC);
1623         adapter->stats.base_gotc = er32(VFGOTC);
1624         adapter->stats.base_mprc = er32(VFMPRC);
1625         adapter->stats.base_gotlbc = er32(VFGOTLBC);
1626         adapter->stats.base_gptlbc = er32(VFGPTLBC);
1627         adapter->stats.base_gorlbc = er32(VFGORLBC);
1628         adapter->stats.base_gprlbc = er32(VFGPRLBC);
1629 }
1630
1631 /**
1632  * igbvf_open - Called when a network interface is made active
1633  * @netdev: network interface device structure
1634  *
1635  * Returns 0 on success, negative value on failure
1636  *
1637  * The open entry point is called when a network interface is made
1638  * active by the system (IFF_UP).  At this point all resources needed
1639  * for transmit and receive operations are allocated, the interrupt
1640  * handler is registered with the OS, the watchdog timer is started,
1641  * and the stack is notified that the interface is ready.
1642  **/
1643 static int igbvf_open(struct net_device *netdev)
1644 {
1645         struct igbvf_adapter *adapter = netdev_priv(netdev);
1646         struct e1000_hw *hw = &adapter->hw;
1647         int err;
1648
1649         /* disallow open during test */
1650         if (test_bit(__IGBVF_TESTING, &adapter->state))
1651                 return -EBUSY;
1652
1653         /* allocate transmit descriptors */
1654         err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1655         if (err)
1656                 goto err_setup_tx;
1657
1658         /* allocate receive descriptors */
1659         err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1660         if (err)
1661                 goto err_setup_rx;
1662
1663         /*
1664          * before we allocate an interrupt, we must be ready to handle it.
1665          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1666          * as soon as we call pci_request_irq, so we have to setup our
1667          * clean_rx handler before we do so.
1668          */
1669         igbvf_configure(adapter);
1670
1671         err = igbvf_request_irq(adapter);
1672         if (err)
1673                 goto err_req_irq;
1674
1675         /* From here on the code is the same as igbvf_up() */
1676         clear_bit(__IGBVF_DOWN, &adapter->state);
1677
1678         napi_enable(&adapter->rx_ring->napi);
1679
1680         /* clear any pending interrupts */
1681         er32(EICR);
1682
1683         igbvf_irq_enable(adapter);
1684
1685         /* start the watchdog */
1686         hw->mac.get_link_status = 1;
1687         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1688
1689         return 0;
1690
1691 err_req_irq:
1692         igbvf_free_rx_resources(adapter->rx_ring);
1693 err_setup_rx:
1694         igbvf_free_tx_resources(adapter->tx_ring);
1695 err_setup_tx:
1696         igbvf_reset(adapter);
1697
1698         return err;
1699 }
1700
1701 /**
1702  * igbvf_close - Disables a network interface
1703  * @netdev: network interface device structure
1704  *
1705  * Returns 0, this is not allowed to fail
1706  *
1707  * The close entry point is called when an interface is de-activated
1708  * by the OS.  The hardware is still under the drivers control, but
1709  * needs to be disabled.  A global MAC reset is issued to stop the
1710  * hardware, and all transmit and receive resources are freed.
1711  **/
1712 static int igbvf_close(struct net_device *netdev)
1713 {
1714         struct igbvf_adapter *adapter = netdev_priv(netdev);
1715
1716         WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1717         igbvf_down(adapter);
1718
1719         igbvf_free_irq(adapter);
1720
1721         igbvf_free_tx_resources(adapter->tx_ring);
1722         igbvf_free_rx_resources(adapter->rx_ring);
1723
1724         return 0;
1725 }
1726 /**
1727  * igbvf_set_mac - Change the Ethernet Address of the NIC
1728  * @netdev: network interface device structure
1729  * @p: pointer to an address structure
1730  *
1731  * Returns 0 on success, negative on failure
1732  **/
1733 static int igbvf_set_mac(struct net_device *netdev, void *p)
1734 {
1735         struct igbvf_adapter *adapter = netdev_priv(netdev);
1736         struct e1000_hw *hw = &adapter->hw;
1737         struct sockaddr *addr = p;
1738
1739         if (!is_valid_ether_addr(addr->sa_data))
1740                 return -EADDRNOTAVAIL;
1741
1742         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1743
1744         hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1745
1746         if (memcmp(addr->sa_data, hw->mac.addr, 6))
1747                 return -EADDRNOTAVAIL;
1748
1749         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1750
1751         return 0;
1752 }
1753
1754 #define UPDATE_VF_COUNTER(reg, name)                                    \
1755         {                                                               \
1756                 u32 current_counter = er32(reg);                        \
1757                 if (current_counter < adapter->stats.last_##name)       \
1758                         adapter->stats.name += 0x100000000LL;           \
1759                 adapter->stats.last_##name = current_counter;           \
1760                 adapter->stats.name &= 0xFFFFFFFF00000000LL;            \
1761                 adapter->stats.name |= current_counter;                 \
1762         }
1763
1764 /**
1765  * igbvf_update_stats - Update the board statistics counters
1766  * @adapter: board private structure
1767 **/
1768 void igbvf_update_stats(struct igbvf_adapter *adapter)
1769 {
1770         struct e1000_hw *hw = &adapter->hw;
1771         struct pci_dev *pdev = adapter->pdev;
1772
1773         /*
1774          * Prevent stats update while adapter is being reset, link is down
1775          * or if the pci connection is down.
1776          */
1777         if (adapter->link_speed == 0)
1778                 return;
1779
1780         if (test_bit(__IGBVF_RESETTING, &adapter->state))
1781                 return;
1782
1783         if (pci_channel_offline(pdev))
1784                 return;
1785
1786         UPDATE_VF_COUNTER(VFGPRC, gprc);
1787         UPDATE_VF_COUNTER(VFGORC, gorc);
1788         UPDATE_VF_COUNTER(VFGPTC, gptc);
1789         UPDATE_VF_COUNTER(VFGOTC, gotc);
1790         UPDATE_VF_COUNTER(VFMPRC, mprc);
1791         UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1792         UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1793         UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1794         UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1795
1796         /* Fill out the OS statistics structure */
1797         adapter->net_stats.multicast = adapter->stats.mprc;
1798 }
1799
1800 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1801 {
1802         dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s\n",
1803                  adapter->link_speed,
1804                  ((adapter->link_duplex == FULL_DUPLEX) ?
1805                   "Full Duplex" : "Half Duplex"));
1806 }
1807
1808 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1809 {
1810         struct e1000_hw *hw = &adapter->hw;
1811         s32 ret_val = E1000_SUCCESS;
1812         bool link_active;
1813
1814         ret_val = hw->mac.ops.check_for_link(hw);
1815         link_active = !hw->mac.get_link_status;
1816
1817         /* if check for link returns error we will need to reset */
1818         if (ret_val)
1819                 schedule_work(&adapter->reset_task);
1820
1821         return link_active;
1822 }
1823
1824 /**
1825  * igbvf_watchdog - Timer Call-back
1826  * @data: pointer to adapter cast into an unsigned long
1827  **/
1828 static void igbvf_watchdog(unsigned long data)
1829 {
1830         struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1831
1832         /* Do the rest outside of interrupt context */
1833         schedule_work(&adapter->watchdog_task);
1834 }
1835
1836 static void igbvf_watchdog_task(struct work_struct *work)
1837 {
1838         struct igbvf_adapter *adapter = container_of(work,
1839                                                      struct igbvf_adapter,
1840                                                      watchdog_task);
1841         struct net_device *netdev = adapter->netdev;
1842         struct e1000_mac_info *mac = &adapter->hw.mac;
1843         struct igbvf_ring *tx_ring = adapter->tx_ring;
1844         struct e1000_hw *hw = &adapter->hw;
1845         u32 link;
1846         int tx_pending = 0;
1847
1848         link = igbvf_has_link(adapter);
1849
1850         if (link) {
1851                 if (!netif_carrier_ok(netdev)) {
1852                         bool txb2b = 1;
1853
1854                         mac->ops.get_link_up_info(&adapter->hw,
1855                                                   &adapter->link_speed,
1856                                                   &adapter->link_duplex);
1857                         igbvf_print_link_info(adapter);
1858
1859                         /*
1860                          * tweak tx_queue_len according to speed/duplex
1861                          * and adjust the timeout factor
1862                          */
1863                         netdev->tx_queue_len = adapter->tx_queue_len;
1864                         adapter->tx_timeout_factor = 1;
1865                         switch (adapter->link_speed) {
1866                         case SPEED_10:
1867                                 txb2b = 0;
1868                                 netdev->tx_queue_len = 10;
1869                                 adapter->tx_timeout_factor = 16;
1870                                 break;
1871                         case SPEED_100:
1872                                 txb2b = 0;
1873                                 netdev->tx_queue_len = 100;
1874                                 /* maybe add some timeout factor ? */
1875                                 break;
1876                         }
1877
1878                         netif_carrier_on(netdev);
1879                         netif_wake_queue(netdev);
1880                 }
1881         } else {
1882                 if (netif_carrier_ok(netdev)) {
1883                         adapter->link_speed = 0;
1884                         adapter->link_duplex = 0;
1885                         dev_info(&adapter->pdev->dev, "Link is Down\n");
1886                         netif_carrier_off(netdev);
1887                         netif_stop_queue(netdev);
1888                 }
1889         }
1890
1891         if (netif_carrier_ok(netdev)) {
1892                 igbvf_update_stats(adapter);
1893         } else {
1894                 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1895                               tx_ring->count);
1896                 if (tx_pending) {
1897                         /*
1898                          * We've lost link, so the controller stops DMA,
1899                          * but we've got queued Tx work that's never going
1900                          * to get done, so reset controller to flush Tx.
1901                          * (Do the reset outside of interrupt context).
1902                          */
1903                         adapter->tx_timeout_count++;
1904                         schedule_work(&adapter->reset_task);
1905                 }
1906         }
1907
1908         /* Cause software interrupt to ensure Rx ring is cleaned */
1909         ew32(EICS, adapter->rx_ring->eims_value);
1910
1911         /* Force detection of hung controller every watchdog period */
1912         adapter->detect_tx_hung = 1;
1913
1914         /* Reset the timer */
1915         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1916                 mod_timer(&adapter->watchdog_timer,
1917                           round_jiffies(jiffies + (2 * HZ)));
1918 }
1919
1920 #define IGBVF_TX_FLAGS_CSUM             0x00000001
1921 #define IGBVF_TX_FLAGS_VLAN             0x00000002
1922 #define IGBVF_TX_FLAGS_TSO              0x00000004
1923 #define IGBVF_TX_FLAGS_IPV4             0x00000008
1924 #define IGBVF_TX_FLAGS_VLAN_MASK        0xffff0000
1925 #define IGBVF_TX_FLAGS_VLAN_SHIFT       16
1926
1927 static int igbvf_tso(struct igbvf_adapter *adapter,
1928                      struct igbvf_ring *tx_ring,
1929                      struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1930 {
1931         struct e1000_adv_tx_context_desc *context_desc;
1932         unsigned int i;
1933         int err;
1934         struct igbvf_buffer *buffer_info;
1935         u32 info = 0, tu_cmd = 0;
1936         u32 mss_l4len_idx, l4len;
1937         *hdr_len = 0;
1938
1939         if (skb_header_cloned(skb)) {
1940                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1941                 if (err) {
1942                         dev_err(&adapter->pdev->dev,
1943                                 "igbvf_tso returning an error\n");
1944                         return err;
1945                 }
1946         }
1947
1948         l4len = tcp_hdrlen(skb);
1949         *hdr_len += l4len;
1950
1951         if (skb->protocol == htons(ETH_P_IP)) {
1952                 struct iphdr *iph = ip_hdr(skb);
1953                 iph->tot_len = 0;
1954                 iph->check = 0;
1955                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1956                                                          iph->daddr, 0,
1957                                                          IPPROTO_TCP,
1958                                                          0);
1959         } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
1960                 ipv6_hdr(skb)->payload_len = 0;
1961                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1962                                                        &ipv6_hdr(skb)->daddr,
1963                                                        0, IPPROTO_TCP, 0);
1964         }
1965
1966         i = tx_ring->next_to_use;
1967
1968         buffer_info = &tx_ring->buffer_info[i];
1969         context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1970         /* VLAN MACLEN IPLEN */
1971         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1972                 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1973         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1974         *hdr_len += skb_network_offset(skb);
1975         info |= (skb_transport_header(skb) - skb_network_header(skb));
1976         *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1977         context_desc->vlan_macip_lens = cpu_to_le32(info);
1978
1979         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1980         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1981
1982         if (skb->protocol == htons(ETH_P_IP))
1983                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1984         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1985
1986         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1987
1988         /* MSS L4LEN IDX */
1989         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1990         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1991
1992         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1993         context_desc->seqnum_seed = 0;
1994
1995         buffer_info->time_stamp = jiffies;
1996         buffer_info->next_to_watch = i;
1997         buffer_info->dma = 0;
1998         i++;
1999         if (i == tx_ring->count)
2000                 i = 0;
2001
2002         tx_ring->next_to_use = i;
2003
2004         return true;
2005 }
2006
2007 static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
2008                                  struct igbvf_ring *tx_ring,
2009                                  struct sk_buff *skb, u32 tx_flags)
2010 {
2011         struct e1000_adv_tx_context_desc *context_desc;
2012         unsigned int i;
2013         struct igbvf_buffer *buffer_info;
2014         u32 info = 0, tu_cmd = 0;
2015
2016         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2017             (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
2018                 i = tx_ring->next_to_use;
2019                 buffer_info = &tx_ring->buffer_info[i];
2020                 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
2021
2022                 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2023                         info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
2024
2025                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2026                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2027                         info |= (skb_transport_header(skb) -
2028                                  skb_network_header(skb));
2029
2030
2031                 context_desc->vlan_macip_lens = cpu_to_le32(info);
2032
2033                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2034
2035                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2036                         switch (skb->protocol) {
2037                         case __constant_htons(ETH_P_IP):
2038                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2039                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2040                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2041                                 break;
2042                         case __constant_htons(ETH_P_IPV6):
2043                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2044                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2045                                 break;
2046                         default:
2047                                 break;
2048                         }
2049                 }
2050
2051                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2052                 context_desc->seqnum_seed = 0;
2053                 context_desc->mss_l4len_idx = 0;
2054
2055                 buffer_info->time_stamp = jiffies;
2056                 buffer_info->next_to_watch = i;
2057                 buffer_info->dma = 0;
2058                 i++;
2059                 if (i == tx_ring->count)
2060                         i = 0;
2061                 tx_ring->next_to_use = i;
2062
2063                 return true;
2064         }
2065
2066         return false;
2067 }
2068
2069 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2070 {
2071         struct igbvf_adapter *adapter = netdev_priv(netdev);
2072
2073         /* there is enough descriptors then we don't need to worry  */
2074         if (igbvf_desc_unused(adapter->tx_ring) >= size)
2075                 return 0;
2076
2077         netif_stop_queue(netdev);
2078
2079         smp_mb();
2080
2081         /* We need to check again just in case room has been made available */
2082         if (igbvf_desc_unused(adapter->tx_ring) < size)
2083                 return -EBUSY;
2084
2085         netif_wake_queue(netdev);
2086
2087         ++adapter->restart_queue;
2088         return 0;
2089 }
2090
2091 #define IGBVF_MAX_TXD_PWR       16
2092 #define IGBVF_MAX_DATA_PER_TXD  (1 << IGBVF_MAX_TXD_PWR)
2093
2094 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2095                                    struct igbvf_ring *tx_ring,
2096                                    struct sk_buff *skb,
2097                                    unsigned int first)
2098 {
2099         struct igbvf_buffer *buffer_info;
2100         unsigned int len = skb_headlen(skb);
2101         unsigned int count = 0, i;
2102         unsigned int f;
2103         dma_addr_t *map;
2104
2105         i = tx_ring->next_to_use;
2106
2107         if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
2108                 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
2109                 return 0;
2110         }
2111
2112         map = skb_shinfo(skb)->dma_maps;
2113
2114         buffer_info = &tx_ring->buffer_info[i];
2115         BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2116         buffer_info->length = len;
2117         /* set time_stamp *before* dma to help avoid a possible race */
2118         buffer_info->time_stamp = jiffies;
2119         buffer_info->next_to_watch = i;
2120         buffer_info->dma = map[count];
2121         count++;
2122
2123         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2124                 struct skb_frag_struct *frag;
2125
2126                 i++;
2127                 if (i == tx_ring->count)
2128                         i = 0;
2129
2130                 frag = &skb_shinfo(skb)->frags[f];
2131                 len = frag->size;
2132
2133                 buffer_info = &tx_ring->buffer_info[i];
2134                 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2135                 buffer_info->length = len;
2136                 buffer_info->time_stamp = jiffies;
2137                 buffer_info->next_to_watch = i;
2138                 buffer_info->dma = map[count];
2139                 count++;
2140         }
2141
2142         tx_ring->buffer_info[i].skb = skb;
2143         tx_ring->buffer_info[first].next_to_watch = i;
2144
2145         return count;
2146 }
2147
2148 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2149                                       struct igbvf_ring *tx_ring,
2150                                       int tx_flags, int count, u32 paylen,
2151                                       u8 hdr_len)
2152 {
2153         union e1000_adv_tx_desc *tx_desc = NULL;
2154         struct igbvf_buffer *buffer_info;
2155         u32 olinfo_status = 0, cmd_type_len;
2156         unsigned int i;
2157
2158         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2159                         E1000_ADVTXD_DCMD_DEXT);
2160
2161         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2162                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2163
2164         if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2165                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2166
2167                 /* insert tcp checksum */
2168                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2169
2170                 /* insert ip checksum */
2171                 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2172                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2173
2174         } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2175                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2176         }
2177
2178         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2179
2180         i = tx_ring->next_to_use;
2181         while (count--) {
2182                 buffer_info = &tx_ring->buffer_info[i];
2183                 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2184                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2185                 tx_desc->read.cmd_type_len =
2186                          cpu_to_le32(cmd_type_len | buffer_info->length);
2187                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2188                 i++;
2189                 if (i == tx_ring->count)
2190                         i = 0;
2191         }
2192
2193         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2194         /* Force memory writes to complete before letting h/w
2195          * know there are new descriptors to fetch.  (Only
2196          * applicable for weak-ordered memory model archs,
2197          * such as IA-64). */
2198         wmb();
2199
2200         tx_ring->next_to_use = i;
2201         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2202         /* we need this if more than one processor can write to our tail
2203          * at a time, it syncronizes IO on IA64/Altix systems */
2204         mmiowb();
2205 }
2206
2207 static int igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2208                                    struct net_device *netdev,
2209                                    struct igbvf_ring *tx_ring)
2210 {
2211         struct igbvf_adapter *adapter = netdev_priv(netdev);
2212         unsigned int first, tx_flags = 0;
2213         u8 hdr_len = 0;
2214         int count = 0;
2215         int tso = 0;
2216
2217         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2218                 dev_kfree_skb_any(skb);
2219                 return NETDEV_TX_OK;
2220         }
2221
2222         if (skb->len <= 0) {
2223                 dev_kfree_skb_any(skb);
2224                 return NETDEV_TX_OK;
2225         }
2226
2227         /*
2228          * need: count + 4 desc gap to keep tail from touching
2229          *       + 2 desc gap to keep tail from touching head,
2230          *       + 1 desc for skb->data,
2231          *       + 1 desc for context descriptor,
2232          * head, otherwise try next time
2233          */
2234         if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2235                 /* this is a hard error */
2236                 return NETDEV_TX_BUSY;
2237         }
2238
2239         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2240                 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2241                 tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2242         }
2243
2244         if (skb->protocol == htons(ETH_P_IP))
2245                 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2246
2247         first = tx_ring->next_to_use;
2248
2249         tso = skb_is_gso(skb) ?
2250                 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2251         if (unlikely(tso < 0)) {
2252                 dev_kfree_skb_any(skb);
2253                 return NETDEV_TX_OK;
2254         }
2255
2256         if (tso)
2257                 tx_flags |= IGBVF_TX_FLAGS_TSO;
2258         else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2259                  (skb->ip_summed == CHECKSUM_PARTIAL))
2260                 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2261
2262         /*
2263          * count reflects descriptors mapped, if 0 then mapping error
2264          * has occured and we need to rewind the descriptor queue
2265          */
2266         count = igbvf_tx_map_adv(adapter, tx_ring, skb, first);
2267
2268         if (count) {
2269                 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2270                                    skb->len, hdr_len);
2271                 netdev->trans_start = jiffies;
2272                 /* Make sure there is space in the ring for the next send. */
2273                 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2274         } else {
2275                 dev_kfree_skb_any(skb);
2276                 tx_ring->buffer_info[first].time_stamp = 0;
2277                 tx_ring->next_to_use = first;
2278         }
2279
2280         return NETDEV_TX_OK;
2281 }
2282
2283 static int igbvf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
2284 {
2285         struct igbvf_adapter *adapter = netdev_priv(netdev);
2286         struct igbvf_ring *tx_ring;
2287         int retval;
2288
2289         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2290                 dev_kfree_skb_any(skb);
2291                 return NETDEV_TX_OK;
2292         }
2293
2294         tx_ring = &adapter->tx_ring[0];
2295
2296         retval = igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2297
2298         return retval;
2299 }
2300
2301 /**
2302  * igbvf_tx_timeout - Respond to a Tx Hang
2303  * @netdev: network interface device structure
2304  **/
2305 static void igbvf_tx_timeout(struct net_device *netdev)
2306 {
2307         struct igbvf_adapter *adapter = netdev_priv(netdev);
2308
2309         /* Do the reset outside of interrupt context */
2310         adapter->tx_timeout_count++;
2311         schedule_work(&adapter->reset_task);
2312 }
2313
2314 static void igbvf_reset_task(struct work_struct *work)
2315 {
2316         struct igbvf_adapter *adapter;
2317         adapter = container_of(work, struct igbvf_adapter, reset_task);
2318
2319         igbvf_reinit_locked(adapter);
2320 }
2321
2322 /**
2323  * igbvf_get_stats - Get System Network Statistics
2324  * @netdev: network interface device structure
2325  *
2326  * Returns the address of the device statistics structure.
2327  * The statistics are actually updated from the timer callback.
2328  **/
2329 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2330 {
2331         struct igbvf_adapter *adapter = netdev_priv(netdev);
2332
2333         /* only return the current stats */
2334         return &adapter->net_stats;
2335 }
2336
2337 /**
2338  * igbvf_change_mtu - Change the Maximum Transfer Unit
2339  * @netdev: network interface device structure
2340  * @new_mtu: new value for maximum frame size
2341  *
2342  * Returns 0 on success, negative on failure
2343  **/
2344 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2345 {
2346         struct igbvf_adapter *adapter = netdev_priv(netdev);
2347         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2348
2349         if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2350                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2351                 return -EINVAL;
2352         }
2353
2354         /* Jumbo frame size limits */
2355         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
2356                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
2357                         dev_err(&adapter->pdev->dev,
2358                                 "Jumbo Frames not supported.\n");
2359                         return -EINVAL;
2360                 }
2361         }
2362
2363 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2364         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2365                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2366                 return -EINVAL;
2367         }
2368
2369         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2370                 msleep(1);
2371         /* igbvf_down has a dependency on max_frame_size */
2372         adapter->max_frame_size = max_frame;
2373         if (netif_running(netdev))
2374                 igbvf_down(adapter);
2375
2376         /*
2377          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2378          * means we reserve 2 more, this pushes us to allocate from the next
2379          * larger slab size.
2380          * i.e. RXBUFFER_2048 --> size-4096 slab
2381          * However with the new *_jumbo_rx* routines, jumbo receives will use
2382          * fragmented skbs
2383          */
2384
2385         if (max_frame <= 1024)
2386                 adapter->rx_buffer_len = 1024;
2387         else if (max_frame <= 2048)
2388                 adapter->rx_buffer_len = 2048;
2389         else
2390 #if (PAGE_SIZE / 2) > 16384
2391                 adapter->rx_buffer_len = 16384;
2392 #else
2393                 adapter->rx_buffer_len = PAGE_SIZE / 2;
2394 #endif
2395
2396
2397         /* adjust allocation if LPE protects us, and we aren't using SBP */
2398         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2399              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2400                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2401                                          ETH_FCS_LEN;
2402
2403         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2404                  netdev->mtu, new_mtu);
2405         netdev->mtu = new_mtu;
2406
2407         if (netif_running(netdev))
2408                 igbvf_up(adapter);
2409         else
2410                 igbvf_reset(adapter);
2411
2412         clear_bit(__IGBVF_RESETTING, &adapter->state);
2413
2414         return 0;
2415 }
2416
2417 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2418 {
2419         switch (cmd) {
2420         default:
2421                 return -EOPNOTSUPP;
2422         }
2423 }
2424
2425 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2426 {
2427         struct net_device *netdev = pci_get_drvdata(pdev);
2428         struct igbvf_adapter *adapter = netdev_priv(netdev);
2429 #ifdef CONFIG_PM
2430         int retval = 0;
2431 #endif
2432
2433         netif_device_detach(netdev);
2434
2435         if (netif_running(netdev)) {
2436                 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2437                 igbvf_down(adapter);
2438                 igbvf_free_irq(adapter);
2439         }
2440
2441 #ifdef CONFIG_PM
2442         retval = pci_save_state(pdev);
2443         if (retval)
2444                 return retval;
2445 #endif
2446
2447         pci_disable_device(pdev);
2448
2449         return 0;
2450 }
2451
2452 #ifdef CONFIG_PM
2453 static int igbvf_resume(struct pci_dev *pdev)
2454 {
2455         struct net_device *netdev = pci_get_drvdata(pdev);
2456         struct igbvf_adapter *adapter = netdev_priv(netdev);
2457         u32 err;
2458
2459         pci_restore_state(pdev);
2460         err = pci_enable_device_mem(pdev);
2461         if (err) {
2462                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2463                 return err;
2464         }
2465
2466         pci_set_master(pdev);
2467
2468         if (netif_running(netdev)) {
2469                 err = igbvf_request_irq(adapter);
2470                 if (err)
2471                         return err;
2472         }
2473
2474         igbvf_reset(adapter);
2475
2476         if (netif_running(netdev))
2477                 igbvf_up(adapter);
2478
2479         netif_device_attach(netdev);
2480
2481         return 0;
2482 }
2483 #endif
2484
2485 static void igbvf_shutdown(struct pci_dev *pdev)
2486 {
2487         igbvf_suspend(pdev, PMSG_SUSPEND);
2488 }
2489
2490 #ifdef CONFIG_NET_POLL_CONTROLLER
2491 /*
2492  * Polling 'interrupt' - used by things like netconsole to send skbs
2493  * without having to re-enable interrupts. It's not called while
2494  * the interrupt routine is executing.
2495  */
2496 static void igbvf_netpoll(struct net_device *netdev)
2497 {
2498         struct igbvf_adapter *adapter = netdev_priv(netdev);
2499
2500         disable_irq(adapter->pdev->irq);
2501
2502         igbvf_clean_tx_irq(adapter->tx_ring);
2503
2504         enable_irq(adapter->pdev->irq);
2505 }
2506 #endif
2507
2508 /**
2509  * igbvf_io_error_detected - called when PCI error is detected
2510  * @pdev: Pointer to PCI device
2511  * @state: The current pci connection state
2512  *
2513  * This function is called after a PCI bus error affecting
2514  * this device has been detected.
2515  */
2516 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2517                                                 pci_channel_state_t state)
2518 {
2519         struct net_device *netdev = pci_get_drvdata(pdev);
2520         struct igbvf_adapter *adapter = netdev_priv(netdev);
2521
2522         netif_device_detach(netdev);
2523
2524         if (netif_running(netdev))
2525                 igbvf_down(adapter);
2526         pci_disable_device(pdev);
2527
2528         /* Request a slot slot reset. */
2529         return PCI_ERS_RESULT_NEED_RESET;
2530 }
2531
2532 /**
2533  * igbvf_io_slot_reset - called after the pci bus has been reset.
2534  * @pdev: Pointer to PCI device
2535  *
2536  * Restart the card from scratch, as if from a cold-boot. Implementation
2537  * resembles the first-half of the igbvf_resume routine.
2538  */
2539 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2540 {
2541         struct net_device *netdev = pci_get_drvdata(pdev);
2542         struct igbvf_adapter *adapter = netdev_priv(netdev);
2543
2544         if (pci_enable_device_mem(pdev)) {
2545                 dev_err(&pdev->dev,
2546                         "Cannot re-enable PCI device after reset.\n");
2547                 return PCI_ERS_RESULT_DISCONNECT;
2548         }
2549         pci_set_master(pdev);
2550
2551         igbvf_reset(adapter);
2552
2553         return PCI_ERS_RESULT_RECOVERED;
2554 }
2555
2556 /**
2557  * igbvf_io_resume - called when traffic can start flowing again.
2558  * @pdev: Pointer to PCI device
2559  *
2560  * This callback is called when the error recovery driver tells us that
2561  * its OK to resume normal operation. Implementation resembles the
2562  * second-half of the igbvf_resume routine.
2563  */
2564 static void igbvf_io_resume(struct pci_dev *pdev)
2565 {
2566         struct net_device *netdev = pci_get_drvdata(pdev);
2567         struct igbvf_adapter *adapter = netdev_priv(netdev);
2568
2569         if (netif_running(netdev)) {
2570                 if (igbvf_up(adapter)) {
2571                         dev_err(&pdev->dev,
2572                                 "can't bring device back up after reset\n");
2573                         return;
2574                 }
2575         }
2576
2577         netif_device_attach(netdev);
2578 }
2579
2580 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2581 {
2582         struct e1000_hw *hw = &adapter->hw;
2583         struct net_device *netdev = adapter->netdev;
2584         struct pci_dev *pdev = adapter->pdev;
2585
2586         dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2587         dev_info(&pdev->dev, "Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
2588                  /* MAC address */
2589                  netdev->dev_addr[0], netdev->dev_addr[1],
2590                  netdev->dev_addr[2], netdev->dev_addr[3],
2591                  netdev->dev_addr[4], netdev->dev_addr[5]);
2592         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
2593 }
2594
2595 static const struct net_device_ops igbvf_netdev_ops = {
2596         .ndo_open                       = igbvf_open,
2597         .ndo_stop                       = igbvf_close,
2598         .ndo_start_xmit                 = igbvf_xmit_frame,
2599         .ndo_get_stats                  = igbvf_get_stats,
2600         .ndo_set_multicast_list         = igbvf_set_multi,
2601         .ndo_set_mac_address            = igbvf_set_mac,
2602         .ndo_change_mtu                 = igbvf_change_mtu,
2603         .ndo_do_ioctl                   = igbvf_ioctl,
2604         .ndo_tx_timeout                 = igbvf_tx_timeout,
2605         .ndo_vlan_rx_register           = igbvf_vlan_rx_register,
2606         .ndo_vlan_rx_add_vid            = igbvf_vlan_rx_add_vid,
2607         .ndo_vlan_rx_kill_vid           = igbvf_vlan_rx_kill_vid,
2608 #ifdef CONFIG_NET_POLL_CONTROLLER
2609         .ndo_poll_controller            = igbvf_netpoll,
2610 #endif
2611 };
2612
2613 /**
2614  * igbvf_probe - Device Initialization Routine
2615  * @pdev: PCI device information struct
2616  * @ent: entry in igbvf_pci_tbl
2617  *
2618  * Returns 0 on success, negative on failure
2619  *
2620  * igbvf_probe initializes an adapter identified by a pci_dev structure.
2621  * The OS initialization, configuring of the adapter private structure,
2622  * and a hardware reset occur.
2623  **/
2624 static int __devinit igbvf_probe(struct pci_dev *pdev,
2625                                  const struct pci_device_id *ent)
2626 {
2627         struct net_device *netdev;
2628         struct igbvf_adapter *adapter;
2629         struct e1000_hw *hw;
2630         const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2631
2632         static int cards_found;
2633         int err, pci_using_dac;
2634
2635         err = pci_enable_device_mem(pdev);
2636         if (err)
2637                 return err;
2638
2639         pci_using_dac = 0;
2640         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
2641         if (!err) {
2642                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
2643                 if (!err)
2644                         pci_using_dac = 1;
2645         } else {
2646                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
2647                 if (err) {
2648                         err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
2649                         if (err) {
2650                                 dev_err(&pdev->dev, "No usable DMA "
2651                                         "configuration, aborting\n");
2652                                 goto err_dma;
2653                         }
2654                 }
2655         }
2656
2657         err = pci_request_regions(pdev, igbvf_driver_name);
2658         if (err)
2659                 goto err_pci_reg;
2660
2661         pci_set_master(pdev);
2662
2663         err = -ENOMEM;
2664         netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2665         if (!netdev)
2666                 goto err_alloc_etherdev;
2667
2668         SET_NETDEV_DEV(netdev, &pdev->dev);
2669
2670         pci_set_drvdata(pdev, netdev);
2671         adapter = netdev_priv(netdev);
2672         hw = &adapter->hw;
2673         adapter->netdev = netdev;
2674         adapter->pdev = pdev;
2675         adapter->ei = ei;
2676         adapter->pba = ei->pba;
2677         adapter->flags = ei->flags;
2678         adapter->hw.back = adapter;
2679         adapter->hw.mac.type = ei->mac;
2680         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
2681
2682         /* PCI config space info */
2683
2684         hw->vendor_id = pdev->vendor;
2685         hw->device_id = pdev->device;
2686         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2687         hw->subsystem_device_id = pdev->subsystem_device;
2688
2689         pci_read_config_byte(pdev, PCI_REVISION_ID, &hw->revision_id);
2690
2691         err = -EIO;
2692         adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2693                                       pci_resource_len(pdev, 0));
2694
2695         if (!adapter->hw.hw_addr)
2696                 goto err_ioremap;
2697
2698         if (ei->get_variants) {
2699                 err = ei->get_variants(adapter);
2700                 if (err)
2701                         goto err_ioremap;
2702         }
2703
2704         /* setup adapter struct */
2705         err = igbvf_sw_init(adapter);
2706         if (err)
2707                 goto err_sw_init;
2708
2709         /* construct the net_device struct */
2710         netdev->netdev_ops = &igbvf_netdev_ops;
2711
2712         igbvf_set_ethtool_ops(netdev);
2713         netdev->watchdog_timeo = 5 * HZ;
2714         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2715
2716         adapter->bd_number = cards_found++;
2717
2718         netdev->features = NETIF_F_SG |
2719                            NETIF_F_IP_CSUM |
2720                            NETIF_F_HW_VLAN_TX |
2721                            NETIF_F_HW_VLAN_RX |
2722                            NETIF_F_HW_VLAN_FILTER;
2723
2724         netdev->features |= NETIF_F_IPV6_CSUM;
2725         netdev->features |= NETIF_F_TSO;
2726         netdev->features |= NETIF_F_TSO6;
2727
2728         if (pci_using_dac)
2729                 netdev->features |= NETIF_F_HIGHDMA;
2730
2731         netdev->vlan_features |= NETIF_F_TSO;
2732         netdev->vlan_features |= NETIF_F_TSO6;
2733         netdev->vlan_features |= NETIF_F_IP_CSUM;
2734         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2735         netdev->vlan_features |= NETIF_F_SG;
2736
2737         /*reset the controller to put the device in a known good state */
2738         err = hw->mac.ops.reset_hw(hw);
2739         if (err) {
2740                 dev_info(&pdev->dev,
2741                          "PF still in reset state, assigning new address\n");
2742                 random_ether_addr(hw->mac.addr);
2743         } else {
2744                 err = hw->mac.ops.read_mac_addr(hw);
2745                 if (err) {
2746                         dev_err(&pdev->dev, "Error reading MAC address\n");
2747                         goto err_hw_init;
2748                 }
2749         }
2750
2751         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
2752         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
2753
2754         if (!is_valid_ether_addr(netdev->perm_addr)) {
2755                 dev_err(&pdev->dev, "Invalid MAC Address: "
2756                         "%02x:%02x:%02x:%02x:%02x:%02x\n",
2757                         netdev->dev_addr[0], netdev->dev_addr[1],
2758                         netdev->dev_addr[2], netdev->dev_addr[3],
2759                         netdev->dev_addr[4], netdev->dev_addr[5]);
2760                 err = -EIO;
2761                 goto err_hw_init;
2762         }
2763
2764         setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2765                     (unsigned long) adapter);
2766
2767         INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2768         INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2769
2770         /* ring size defaults */
2771         adapter->rx_ring->count = 1024;
2772         adapter->tx_ring->count = 1024;
2773
2774         /* reset the hardware with the new settings */
2775         igbvf_reset(adapter);
2776
2777         /* tell the stack to leave us alone until igbvf_open() is called */
2778         netif_carrier_off(netdev);
2779         netif_stop_queue(netdev);
2780
2781         strcpy(netdev->name, "eth%d");
2782         err = register_netdev(netdev);
2783         if (err)
2784                 goto err_hw_init;
2785
2786         igbvf_print_device_info(adapter);
2787
2788         igbvf_initialize_last_counter_stats(adapter);
2789
2790         return 0;
2791
2792 err_hw_init:
2793         kfree(adapter->tx_ring);
2794         kfree(adapter->rx_ring);
2795 err_sw_init:
2796         igbvf_reset_interrupt_capability(adapter);
2797         iounmap(adapter->hw.hw_addr);
2798 err_ioremap:
2799         free_netdev(netdev);
2800 err_alloc_etherdev:
2801         pci_release_regions(pdev);
2802 err_pci_reg:
2803 err_dma:
2804         pci_disable_device(pdev);
2805         return err;
2806 }
2807
2808 /**
2809  * igbvf_remove - Device Removal Routine
2810  * @pdev: PCI device information struct
2811  *
2812  * igbvf_remove is called by the PCI subsystem to alert the driver
2813  * that it should release a PCI device.  The could be caused by a
2814  * Hot-Plug event, or because the driver is going to be removed from
2815  * memory.
2816  **/
2817 static void __devexit igbvf_remove(struct pci_dev *pdev)
2818 {
2819         struct net_device *netdev = pci_get_drvdata(pdev);
2820         struct igbvf_adapter *adapter = netdev_priv(netdev);
2821         struct e1000_hw *hw = &adapter->hw;
2822
2823         /*
2824          * flush_scheduled work may reschedule our watchdog task, so
2825          * explicitly disable watchdog tasks from being rescheduled
2826          */
2827         set_bit(__IGBVF_DOWN, &adapter->state);
2828         del_timer_sync(&adapter->watchdog_timer);
2829
2830         flush_scheduled_work();
2831
2832         unregister_netdev(netdev);
2833
2834         igbvf_reset_interrupt_capability(adapter);
2835
2836         /*
2837          * it is important to delete the napi struct prior to freeing the
2838          * rx ring so that you do not end up with null pointer refs
2839          */
2840         netif_napi_del(&adapter->rx_ring->napi);
2841         kfree(adapter->tx_ring);
2842         kfree(adapter->rx_ring);
2843
2844         iounmap(hw->hw_addr);
2845         if (hw->flash_address)
2846                 iounmap(hw->flash_address);
2847         pci_release_regions(pdev);
2848
2849         free_netdev(netdev);
2850
2851         pci_disable_device(pdev);
2852 }
2853
2854 /* PCI Error Recovery (ERS) */
2855 static struct pci_error_handlers igbvf_err_handler = {
2856         .error_detected = igbvf_io_error_detected,
2857         .slot_reset = igbvf_io_slot_reset,
2858         .resume = igbvf_io_resume,
2859 };
2860
2861 static struct pci_device_id igbvf_pci_tbl[] = {
2862         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2863         { } /* terminate list */
2864 };
2865 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2866
2867 /* PCI Device API Driver */
2868 static struct pci_driver igbvf_driver = {
2869         .name     = igbvf_driver_name,
2870         .id_table = igbvf_pci_tbl,
2871         .probe    = igbvf_probe,
2872         .remove   = __devexit_p(igbvf_remove),
2873 #ifdef CONFIG_PM
2874         /* Power Management Hooks */
2875         .suspend  = igbvf_suspend,
2876         .resume   = igbvf_resume,
2877 #endif
2878         .shutdown = igbvf_shutdown,
2879         .err_handler = &igbvf_err_handler
2880 };
2881
2882 /**
2883  * igbvf_init_module - Driver Registration Routine
2884  *
2885  * igbvf_init_module is the first routine called when the driver is
2886  * loaded. All it does is register with the PCI subsystem.
2887  **/
2888 static int __init igbvf_init_module(void)
2889 {
2890         int ret;
2891         printk(KERN_INFO "%s - version %s\n",
2892                igbvf_driver_string, igbvf_driver_version);
2893         printk(KERN_INFO "%s\n", igbvf_copyright);
2894
2895         ret = pci_register_driver(&igbvf_driver);
2896         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, igbvf_driver_name,
2897                                PM_QOS_DEFAULT_VALUE);
2898
2899         return ret;
2900 }
2901 module_init(igbvf_init_module);
2902
2903 /**
2904  * igbvf_exit_module - Driver Exit Cleanup Routine
2905  *
2906  * igbvf_exit_module is called just before the driver is removed
2907  * from memory.
2908  **/
2909 static void __exit igbvf_exit_module(void)
2910 {
2911         pci_unregister_driver(&igbvf_driver);
2912         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, igbvf_driver_name);
2913 }
2914 module_exit(igbvf_exit_module);
2915
2916
2917 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2918 MODULE_DESCRIPTION("Intel(R) 82576 Virtual Function Network Driver");
2919 MODULE_LICENSE("GPL");
2920 MODULE_VERSION(DRV_VERSION);
2921
2922 /* netdev.c */