Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[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                  = 0,
62         .pba                    = 10,
63         .init_ops               = e1000_init_function_pointers_vf,
64 };
65
66 static const struct igbvf_info *igbvf_info_tbl[] = {
67         [board_vf]              = &igbvf_vf_info,
68 };
69
70 /**
71  * igbvf_desc_unused - calculate if we have unused descriptors
72  **/
73 static int igbvf_desc_unused(struct igbvf_ring *ring)
74 {
75         if (ring->next_to_clean > ring->next_to_use)
76                 return ring->next_to_clean - ring->next_to_use - 1;
77
78         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
79 }
80
81 /**
82  * igbvf_receive_skb - helper function to handle Rx indications
83  * @adapter: board private structure
84  * @status: descriptor status field as written by hardware
85  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
86  * @skb: pointer to sk_buff to be indicated to stack
87  **/
88 static void igbvf_receive_skb(struct igbvf_adapter *adapter,
89                               struct net_device *netdev,
90                               struct sk_buff *skb,
91                               u32 status, u16 vlan)
92 {
93         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
94                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
95                                          le16_to_cpu(vlan) &
96                                          E1000_RXD_SPC_VLAN_MASK);
97         else
98                 netif_receive_skb(skb);
99
100         netdev->last_rx = jiffies;
101 }
102
103 static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
104                                          u32 status_err, struct sk_buff *skb)
105 {
106         skb->ip_summed = CHECKSUM_NONE;
107
108         /* Ignore Checksum bit is set or checksum is disabled through ethtool */
109         if ((status_err & E1000_RXD_STAT_IXSM) ||
110             (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
111                 return;
112
113         /* TCP/UDP checksum error bit is set */
114         if (status_err &
115             (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
116                 /* let the stack verify checksum errors */
117                 adapter->hw_csum_err++;
118                 return;
119         }
120
121         /* It must be a TCP or UDP packet with a valid checksum */
122         if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
123                 skb->ip_summed = CHECKSUM_UNNECESSARY;
124
125         adapter->hw_csum_good++;
126 }
127
128 /**
129  * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
130  * @rx_ring: address of ring structure to repopulate
131  * @cleaned_count: number of buffers to repopulate
132  **/
133 static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
134                                    int cleaned_count)
135 {
136         struct igbvf_adapter *adapter = rx_ring->adapter;
137         struct net_device *netdev = adapter->netdev;
138         struct pci_dev *pdev = adapter->pdev;
139         union e1000_adv_rx_desc *rx_desc;
140         struct igbvf_buffer *buffer_info;
141         struct sk_buff *skb;
142         unsigned int i;
143         int bufsz;
144
145         i = rx_ring->next_to_use;
146         buffer_info = &rx_ring->buffer_info[i];
147
148         if (adapter->rx_ps_hdr_size)
149                 bufsz = adapter->rx_ps_hdr_size;
150         else
151                 bufsz = adapter->rx_buffer_len;
152         bufsz += NET_IP_ALIGN;
153
154         while (cleaned_count--) {
155                 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
156
157                 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
158                         if (!buffer_info->page) {
159                                 buffer_info->page = alloc_page(GFP_ATOMIC);
160                                 if (!buffer_info->page) {
161                                         adapter->alloc_rx_buff_failed++;
162                                         goto no_buffers;
163                                 }
164                                 buffer_info->page_offset = 0;
165                         } else {
166                                 buffer_info->page_offset ^= PAGE_SIZE / 2;
167                         }
168                         buffer_info->page_dma =
169                                 pci_map_page(pdev, buffer_info->page,
170                                              buffer_info->page_offset,
171                                              PAGE_SIZE / 2,
172                                              PCI_DMA_FROMDEVICE);
173                 }
174
175                 if (!buffer_info->skb) {
176                         skb = netdev_alloc_skb(netdev, bufsz);
177                         if (!skb) {
178                                 adapter->alloc_rx_buff_failed++;
179                                 goto no_buffers;
180                         }
181
182                         /* Make buffer alignment 2 beyond a 16 byte boundary
183                          * this will result in a 16 byte aligned IP header after
184                          * the 14 byte MAC header is removed
185                          */
186                         skb_reserve(skb, NET_IP_ALIGN);
187
188                         buffer_info->skb = skb;
189                         buffer_info->dma = pci_map_single(pdev, skb->data,
190                                                           bufsz,
191                                                           PCI_DMA_FROMDEVICE);
192                 }
193                 /* Refresh the desc even if buffer_addrs didn't change because
194                  * each write-back erases this info. */
195                 if (adapter->rx_ps_hdr_size) {
196                         rx_desc->read.pkt_addr =
197                              cpu_to_le64(buffer_info->page_dma);
198                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
199                 } else {
200                         rx_desc->read.pkt_addr =
201                              cpu_to_le64(buffer_info->dma);
202                         rx_desc->read.hdr_addr = 0;
203                 }
204
205                 i++;
206                 if (i == rx_ring->count)
207                         i = 0;
208                 buffer_info = &rx_ring->buffer_info[i];
209         }
210
211 no_buffers:
212         if (rx_ring->next_to_use != i) {
213                 rx_ring->next_to_use = i;
214                 if (i == 0)
215                         i = (rx_ring->count - 1);
216                 else
217                         i--;
218
219                 /* Force memory writes to complete before letting h/w
220                  * know there are new descriptors to fetch.  (Only
221                  * applicable for weak-ordered memory model archs,
222                  * such as IA-64). */
223                 wmb();
224                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
225         }
226 }
227
228 /**
229  * igbvf_clean_rx_irq - Send received data up the network stack; legacy
230  * @adapter: board private structure
231  *
232  * the return value indicates whether actual cleaning was done, there
233  * is no guarantee that everything was cleaned
234  **/
235 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
236                                int *work_done, int work_to_do)
237 {
238         struct igbvf_ring *rx_ring = adapter->rx_ring;
239         struct net_device *netdev = adapter->netdev;
240         struct pci_dev *pdev = adapter->pdev;
241         union e1000_adv_rx_desc *rx_desc, *next_rxd;
242         struct igbvf_buffer *buffer_info, *next_buffer;
243         struct sk_buff *skb;
244         bool cleaned = false;
245         int cleaned_count = 0;
246         unsigned int total_bytes = 0, total_packets = 0;
247         unsigned int i;
248         u32 length, hlen, staterr;
249
250         i = rx_ring->next_to_clean;
251         rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
252         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
253
254         while (staterr & E1000_RXD_STAT_DD) {
255                 if (*work_done >= work_to_do)
256                         break;
257                 (*work_done)++;
258
259                 buffer_info = &rx_ring->buffer_info[i];
260
261                 /* HW will not DMA in data larger than the given buffer, even
262                  * if it parses the (NFS, of course) header to be larger.  In
263                  * that case, it fills the header buffer and spills the rest
264                  * into the page.
265                  */
266                 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
267                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
268                 if (hlen > adapter->rx_ps_hdr_size)
269                         hlen = adapter->rx_ps_hdr_size;
270
271                 length = le16_to_cpu(rx_desc->wb.upper.length);
272                 cleaned = true;
273                 cleaned_count++;
274
275                 skb = buffer_info->skb;
276                 prefetch(skb->data - NET_IP_ALIGN);
277                 buffer_info->skb = NULL;
278                 if (!adapter->rx_ps_hdr_size) {
279                         pci_unmap_single(pdev, buffer_info->dma,
280                                          adapter->rx_buffer_len,
281                                          PCI_DMA_FROMDEVICE);
282                         buffer_info->dma = 0;
283                         skb_put(skb, length);
284                         goto send_up;
285                 }
286
287                 if (!skb_shinfo(skb)->nr_frags) {
288                         pci_unmap_single(pdev, buffer_info->dma,
289                                          adapter->rx_ps_hdr_size + NET_IP_ALIGN,
290                                          PCI_DMA_FROMDEVICE);
291                         skb_put(skb, hlen);
292                 }
293
294                 if (length) {
295                         pci_unmap_page(pdev, buffer_info->page_dma,
296                                        PAGE_SIZE / 2,
297                                        PCI_DMA_FROMDEVICE);
298                         buffer_info->page_dma = 0;
299
300                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
301                                            buffer_info->page,
302                                            buffer_info->page_offset,
303                                            length);
304
305                         if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
306                             (page_count(buffer_info->page) != 1))
307                                 buffer_info->page = NULL;
308                         else
309                                 get_page(buffer_info->page);
310
311                         skb->len += length;
312                         skb->data_len += length;
313                         skb->truesize += length;
314                 }
315 send_up:
316                 i++;
317                 if (i == rx_ring->count)
318                         i = 0;
319                 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
320                 prefetch(next_rxd);
321                 next_buffer = &rx_ring->buffer_info[i];
322
323                 if (!(staterr & E1000_RXD_STAT_EOP)) {
324                         buffer_info->skb = next_buffer->skb;
325                         buffer_info->dma = next_buffer->dma;
326                         next_buffer->skb = skb;
327                         next_buffer->dma = 0;
328                         goto next_desc;
329                 }
330
331                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
332                         dev_kfree_skb_irq(skb);
333                         goto next_desc;
334                 }
335
336                 total_bytes += skb->len;
337                 total_packets++;
338
339                 igbvf_rx_checksum_adv(adapter, staterr, skb);
340
341                 skb->protocol = eth_type_trans(skb, netdev);
342
343                 igbvf_receive_skb(adapter, netdev, skb, staterr,
344                                   rx_desc->wb.upper.vlan);
345
346                 netdev->last_rx = jiffies;
347
348 next_desc:
349                 rx_desc->wb.upper.status_error = 0;
350
351                 /* return some buffers to hardware, one at a time is too slow */
352                 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
353                         igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
354                         cleaned_count = 0;
355                 }
356
357                 /* use prefetched values */
358                 rx_desc = next_rxd;
359                 buffer_info = next_buffer;
360
361                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
362         }
363
364         rx_ring->next_to_clean = i;
365         cleaned_count = igbvf_desc_unused(rx_ring);
366
367         if (cleaned_count)
368                 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
369
370         adapter->total_rx_packets += total_packets;
371         adapter->total_rx_bytes += total_bytes;
372         adapter->net_stats.rx_bytes += total_bytes;
373         adapter->net_stats.rx_packets += total_packets;
374         return cleaned;
375 }
376
377 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
378                             struct igbvf_buffer *buffer_info)
379 {
380         buffer_info->dma = 0;
381         if (buffer_info->skb) {
382                 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
383                               DMA_TO_DEVICE);
384                 dev_kfree_skb_any(buffer_info->skb);
385                 buffer_info->skb = NULL;
386         }
387         buffer_info->time_stamp = 0;
388 }
389
390 static void igbvf_print_tx_hang(struct igbvf_adapter *adapter)
391 {
392         struct igbvf_ring *tx_ring = adapter->tx_ring;
393         unsigned int i = tx_ring->next_to_clean;
394         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
395         union e1000_adv_tx_desc *eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
396
397         /* detected Tx unit hang */
398         dev_err(&adapter->pdev->dev,
399                 "Detected Tx Unit Hang:\n"
400                 "  TDH                  <%x>\n"
401                 "  TDT                  <%x>\n"
402                 "  next_to_use          <%x>\n"
403                 "  next_to_clean        <%x>\n"
404                 "buffer_info[next_to_clean]:\n"
405                 "  time_stamp           <%lx>\n"
406                 "  next_to_watch        <%x>\n"
407                 "  jiffies              <%lx>\n"
408                 "  next_to_watch.status <%x>\n",
409                 readl(adapter->hw.hw_addr + tx_ring->head),
410                 readl(adapter->hw.hw_addr + tx_ring->tail),
411                 tx_ring->next_to_use,
412                 tx_ring->next_to_clean,
413                 tx_ring->buffer_info[eop].time_stamp,
414                 eop,
415                 jiffies,
416                 eop_desc->wb.status);
417 }
418
419 /**
420  * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
421  * @adapter: board private structure
422  *
423  * Return 0 on success, negative on failure
424  **/
425 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
426                              struct igbvf_ring *tx_ring)
427 {
428         struct pci_dev *pdev = adapter->pdev;
429         int size;
430
431         size = sizeof(struct igbvf_buffer) * tx_ring->count;
432         tx_ring->buffer_info = vmalloc(size);
433         if (!tx_ring->buffer_info)
434                 goto err;
435         memset(tx_ring->buffer_info, 0, size);
436
437         /* round up to nearest 4K */
438         tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
439         tx_ring->size = ALIGN(tx_ring->size, 4096);
440
441         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
442                                              &tx_ring->dma);
443
444         if (!tx_ring->desc)
445                 goto err;
446
447         tx_ring->adapter = adapter;
448         tx_ring->next_to_use = 0;
449         tx_ring->next_to_clean = 0;
450
451         return 0;
452 err:
453         vfree(tx_ring->buffer_info);
454         dev_err(&adapter->pdev->dev,
455                 "Unable to allocate memory for the transmit descriptor ring\n");
456         return -ENOMEM;
457 }
458
459 /**
460  * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
461  * @adapter: board private structure
462  *
463  * Returns 0 on success, negative on failure
464  **/
465 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
466                              struct igbvf_ring *rx_ring)
467 {
468         struct pci_dev *pdev = adapter->pdev;
469         int size, desc_len;
470
471         size = sizeof(struct igbvf_buffer) * rx_ring->count;
472         rx_ring->buffer_info = vmalloc(size);
473         if (!rx_ring->buffer_info)
474                 goto err;
475         memset(rx_ring->buffer_info, 0, size);
476
477         desc_len = sizeof(union e1000_adv_rx_desc);
478
479         /* Round up to nearest 4K */
480         rx_ring->size = rx_ring->count * desc_len;
481         rx_ring->size = ALIGN(rx_ring->size, 4096);
482
483         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
484                                              &rx_ring->dma);
485
486         if (!rx_ring->desc)
487                 goto err;
488
489         rx_ring->next_to_clean = 0;
490         rx_ring->next_to_use = 0;
491
492         rx_ring->adapter = adapter;
493
494         return 0;
495
496 err:
497         vfree(rx_ring->buffer_info);
498         rx_ring->buffer_info = NULL;
499         dev_err(&adapter->pdev->dev,
500                 "Unable to allocate memory for the receive descriptor ring\n");
501         return -ENOMEM;
502 }
503
504 /**
505  * igbvf_clean_tx_ring - Free Tx Buffers
506  * @tx_ring: ring to be cleaned
507  **/
508 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
509 {
510         struct igbvf_adapter *adapter = tx_ring->adapter;
511         struct igbvf_buffer *buffer_info;
512         unsigned long size;
513         unsigned int i;
514
515         if (!tx_ring->buffer_info)
516                 return;
517
518         /* Free all the Tx ring sk_buffs */
519         for (i = 0; i < tx_ring->count; i++) {
520                 buffer_info = &tx_ring->buffer_info[i];
521                 igbvf_put_txbuf(adapter, buffer_info);
522         }
523
524         size = sizeof(struct igbvf_buffer) * tx_ring->count;
525         memset(tx_ring->buffer_info, 0, size);
526
527         /* Zero out the descriptor ring */
528         memset(tx_ring->desc, 0, tx_ring->size);
529
530         tx_ring->next_to_use = 0;
531         tx_ring->next_to_clean = 0;
532
533         writel(0, adapter->hw.hw_addr + tx_ring->head);
534         writel(0, adapter->hw.hw_addr + tx_ring->tail);
535 }
536
537 /**
538  * igbvf_free_tx_resources - Free Tx Resources per Queue
539  * @tx_ring: ring to free resources from
540  *
541  * Free all transmit software resources
542  **/
543 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
544 {
545         struct pci_dev *pdev = tx_ring->adapter->pdev;
546
547         igbvf_clean_tx_ring(tx_ring);
548
549         vfree(tx_ring->buffer_info);
550         tx_ring->buffer_info = NULL;
551
552         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
553
554         tx_ring->desc = NULL;
555 }
556
557 /**
558  * igbvf_clean_rx_ring - Free Rx Buffers per Queue
559  * @adapter: board private structure
560  **/
561 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
562 {
563         struct igbvf_adapter *adapter = rx_ring->adapter;
564         struct igbvf_buffer *buffer_info;
565         struct pci_dev *pdev = adapter->pdev;
566         unsigned long size;
567         unsigned int i;
568
569         if (!rx_ring->buffer_info)
570                 return;
571
572         /* Free all the Rx ring sk_buffs */
573         for (i = 0; i < rx_ring->count; i++) {
574                 buffer_info = &rx_ring->buffer_info[i];
575                 if (buffer_info->dma) {
576                         if (adapter->rx_ps_hdr_size){
577                                 pci_unmap_single(pdev, buffer_info->dma,
578                                                  adapter->rx_ps_hdr_size,
579                                                  PCI_DMA_FROMDEVICE);
580                         } else {
581                                 pci_unmap_single(pdev, buffer_info->dma,
582                                                  adapter->rx_buffer_len,
583                                                  PCI_DMA_FROMDEVICE);
584                         }
585                         buffer_info->dma = 0;
586                 }
587
588                 if (buffer_info->skb) {
589                         dev_kfree_skb(buffer_info->skb);
590                         buffer_info->skb = NULL;
591                 }
592
593                 if (buffer_info->page) {
594                         if (buffer_info->page_dma)
595                                 pci_unmap_page(pdev, buffer_info->page_dma,
596                                                PAGE_SIZE / 2,
597                                                PCI_DMA_FROMDEVICE);
598                         put_page(buffer_info->page);
599                         buffer_info->page = NULL;
600                         buffer_info->page_dma = 0;
601                         buffer_info->page_offset = 0;
602                 }
603         }
604
605         size = sizeof(struct igbvf_buffer) * rx_ring->count;
606         memset(rx_ring->buffer_info, 0, size);
607
608         /* Zero out the descriptor ring */
609         memset(rx_ring->desc, 0, rx_ring->size);
610
611         rx_ring->next_to_clean = 0;
612         rx_ring->next_to_use = 0;
613
614         writel(0, adapter->hw.hw_addr + rx_ring->head);
615         writel(0, adapter->hw.hw_addr + rx_ring->tail);
616 }
617
618 /**
619  * igbvf_free_rx_resources - Free Rx Resources
620  * @rx_ring: ring to clean the resources from
621  *
622  * Free all receive software resources
623  **/
624
625 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
626 {
627         struct pci_dev *pdev = rx_ring->adapter->pdev;
628
629         igbvf_clean_rx_ring(rx_ring);
630
631         vfree(rx_ring->buffer_info);
632         rx_ring->buffer_info = NULL;
633
634         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
635                           rx_ring->dma);
636         rx_ring->desc = NULL;
637 }
638
639 /**
640  * igbvf_update_itr - update the dynamic ITR value based on statistics
641  * @adapter: pointer to adapter
642  * @itr_setting: current adapter->itr
643  * @packets: the number of packets during this measurement interval
644  * @bytes: the number of bytes during this measurement interval
645  *
646  *      Stores a new ITR value based on packets and byte
647  *      counts during the last interrupt.  The advantage of per interrupt
648  *      computation is faster updates and more accurate ITR for the current
649  *      traffic pattern.  Constants in this function were computed
650  *      based on theoretical maximum wire speed and thresholds were set based
651  *      on testing data as well as attempting to minimize response time
652  *      while increasing bulk throughput.  This functionality is controlled
653  *      by the InterruptThrottleRate module parameter.
654  **/
655 static unsigned int igbvf_update_itr(struct igbvf_adapter *adapter,
656                                      u16 itr_setting, int packets,
657                                      int bytes)
658 {
659         unsigned int retval = itr_setting;
660
661         if (packets == 0)
662                 goto update_itr_done;
663
664         switch (itr_setting) {
665         case lowest_latency:
666                 /* handle TSO and jumbo frames */
667                 if (bytes/packets > 8000)
668                         retval = bulk_latency;
669                 else if ((packets < 5) && (bytes > 512))
670                         retval = low_latency;
671                 break;
672         case low_latency:  /* 50 usec aka 20000 ints/s */
673                 if (bytes > 10000) {
674                         /* this if handles the TSO accounting */
675                         if (bytes/packets > 8000)
676                                 retval = bulk_latency;
677                         else if ((packets < 10) || ((bytes/packets) > 1200))
678                                 retval = bulk_latency;
679                         else if ((packets > 35))
680                                 retval = lowest_latency;
681                 } else if (bytes/packets > 2000) {
682                         retval = bulk_latency;
683                 } else if (packets <= 2 && bytes < 512) {
684                         retval = lowest_latency;
685                 }
686                 break;
687         case bulk_latency: /* 250 usec aka 4000 ints/s */
688                 if (bytes > 25000) {
689                         if (packets > 35)
690                                 retval = low_latency;
691                 } else if (bytes < 6000) {
692                         retval = low_latency;
693                 }
694                 break;
695         }
696
697 update_itr_done:
698         return retval;
699 }
700
701 static void igbvf_set_itr(struct igbvf_adapter *adapter)
702 {
703         struct e1000_hw *hw = &adapter->hw;
704         u16 current_itr;
705         u32 new_itr = adapter->itr;
706
707         adapter->tx_itr = igbvf_update_itr(adapter, adapter->tx_itr,
708                                            adapter->total_tx_packets,
709                                            adapter->total_tx_bytes);
710         /* conservative mode (itr 3) eliminates the lowest_latency setting */
711         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
712                 adapter->tx_itr = low_latency;
713
714         adapter->rx_itr = igbvf_update_itr(adapter, adapter->rx_itr,
715                                            adapter->total_rx_packets,
716                                            adapter->total_rx_bytes);
717         /* conservative mode (itr 3) eliminates the lowest_latency setting */
718         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
719                 adapter->rx_itr = low_latency;
720
721         current_itr = max(adapter->rx_itr, adapter->tx_itr);
722
723         switch (current_itr) {
724         /* counts and packets in update_itr are dependent on these numbers */
725         case lowest_latency:
726                 new_itr = 70000;
727                 break;
728         case low_latency:
729                 new_itr = 20000; /* aka hwitr = ~200 */
730                 break;
731         case bulk_latency:
732                 new_itr = 4000;
733                 break;
734         default:
735                 break;
736         }
737
738         if (new_itr != adapter->itr) {
739                 /*
740                  * this attempts to bias the interrupt rate towards Bulk
741                  * by adding intermediate steps when interrupt rate is
742                  * increasing
743                  */
744                 new_itr = new_itr > adapter->itr ?
745                              min(adapter->itr + (new_itr >> 2), new_itr) :
746                              new_itr;
747                 adapter->itr = new_itr;
748                 adapter->rx_ring->itr_val = 1952;
749
750                 if (adapter->msix_entries)
751                         adapter->rx_ring->set_itr = 1;
752                 else
753                         ew32(ITR, 1952);
754         }
755 }
756
757 /**
758  * igbvf_clean_tx_irq - Reclaim resources after transmit completes
759  * @adapter: board private structure
760  * returns true if ring is completely cleaned
761  **/
762 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
763 {
764         struct igbvf_adapter *adapter = tx_ring->adapter;
765         struct e1000_hw *hw = &adapter->hw;
766         struct net_device *netdev = adapter->netdev;
767         struct igbvf_buffer *buffer_info;
768         struct sk_buff *skb;
769         union e1000_adv_tx_desc *tx_desc, *eop_desc;
770         unsigned int total_bytes = 0, total_packets = 0;
771         unsigned int i, eop, count = 0;
772         bool cleaned = false;
773
774         i = tx_ring->next_to_clean;
775         eop = tx_ring->buffer_info[i].next_to_watch;
776         eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
777
778         while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
779                (count < tx_ring->count)) {
780                 for (cleaned = false; !cleaned; count++) {
781                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
782                         buffer_info = &tx_ring->buffer_info[i];
783                         cleaned = (i == eop);
784                         skb = buffer_info->skb;
785
786                         if (skb) {
787                                 unsigned int segs, bytecount;
788
789                                 /* gso_segs is currently only valid for tcp */
790                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
791                                 /* multiply data chunks by size of headers */
792                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
793                                             skb->len;
794                                 total_packets += segs;
795                                 total_bytes += bytecount;
796                         }
797
798                         igbvf_put_txbuf(adapter, buffer_info);
799                         tx_desc->wb.status = 0;
800
801                         i++;
802                         if (i == tx_ring->count)
803                                 i = 0;
804                 }
805                 eop = tx_ring->buffer_info[i].next_to_watch;
806                 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
807         }
808
809         tx_ring->next_to_clean = i;
810
811         if (unlikely(count &&
812                      netif_carrier_ok(netdev) &&
813                      igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
814                 /* Make sure that anybody stopping the queue after this
815                  * sees the new next_to_clean.
816                  */
817                 smp_mb();
818                 if (netif_queue_stopped(netdev) &&
819                     !(test_bit(__IGBVF_DOWN, &adapter->state))) {
820                         netif_wake_queue(netdev);
821                         ++adapter->restart_queue;
822                 }
823         }
824
825         if (adapter->detect_tx_hung) {
826                 /* Detect a transmit hang in hardware, this serializes the
827                  * check with the clearing of time_stamp and movement of i */
828                 adapter->detect_tx_hung = false;
829                 if (tx_ring->buffer_info[i].time_stamp &&
830                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
831                                (adapter->tx_timeout_factor * HZ))
832                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
833
834                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
835                         /* detected Tx unit hang */
836                         igbvf_print_tx_hang(adapter);
837
838                         netif_stop_queue(netdev);
839                 }
840         }
841         adapter->net_stats.tx_bytes += total_bytes;
842         adapter->net_stats.tx_packets += total_packets;
843         return (count < tx_ring->count);
844 }
845
846 static irqreturn_t igbvf_msix_other(int irq, void *data)
847 {
848         struct net_device *netdev = data;
849         struct igbvf_adapter *adapter = netdev_priv(netdev);
850         struct e1000_hw *hw = &adapter->hw;
851
852         adapter->int_counter1++;
853
854         netif_carrier_off(netdev);
855         hw->mac.get_link_status = 1;
856         if (!test_bit(__IGBVF_DOWN, &adapter->state))
857                 mod_timer(&adapter->watchdog_timer, jiffies + 1);
858
859         ew32(EIMS, adapter->eims_other);
860
861         return IRQ_HANDLED;
862 }
863
864 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
865 {
866         struct net_device *netdev = data;
867         struct igbvf_adapter *adapter = netdev_priv(netdev);
868         struct e1000_hw *hw = &adapter->hw;
869         struct igbvf_ring *tx_ring = adapter->tx_ring;
870
871
872         adapter->total_tx_bytes = 0;
873         adapter->total_tx_packets = 0;
874
875         /* auto mask will automatically reenable the interrupt when we write
876          * EICS */
877         if (!igbvf_clean_tx_irq(tx_ring))
878                 /* Ring was not completely cleaned, so fire another interrupt */
879                 ew32(EICS, tx_ring->eims_value);
880         else
881                 ew32(EIMS, tx_ring->eims_value);
882
883         return IRQ_HANDLED;
884 }
885
886 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
887 {
888         struct net_device *netdev = data;
889         struct igbvf_adapter *adapter = netdev_priv(netdev);
890
891         adapter->int_counter0++;
892
893         /* Write the ITR value calculated at the end of the
894          * previous interrupt.
895          */
896         if (adapter->rx_ring->set_itr) {
897                 writel(adapter->rx_ring->itr_val,
898                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
899                 adapter->rx_ring->set_itr = 0;
900         }
901
902         if (napi_schedule_prep(&adapter->rx_ring->napi)) {
903                 adapter->total_rx_bytes = 0;
904                 adapter->total_rx_packets = 0;
905                 __napi_schedule(&adapter->rx_ring->napi);
906         }
907
908         return IRQ_HANDLED;
909 }
910
911 #define IGBVF_NO_QUEUE -1
912
913 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
914                                 int tx_queue, int msix_vector)
915 {
916         struct e1000_hw *hw = &adapter->hw;
917         u32 ivar, index;
918
919         /* 82576 uses a table-based method for assigning vectors.
920            Each queue has a single entry in the table to which we write
921            a vector number along with a "valid" bit.  Sadly, the layout
922            of the table is somewhat counterintuitive. */
923         if (rx_queue > IGBVF_NO_QUEUE) {
924                 index = (rx_queue >> 1);
925                 ivar = array_er32(IVAR0, index);
926                 if (rx_queue & 0x1) {
927                         /* vector goes into third byte of register */
928                         ivar = ivar & 0xFF00FFFF;
929                         ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
930                 } else {
931                         /* vector goes into low byte of register */
932                         ivar = ivar & 0xFFFFFF00;
933                         ivar |= msix_vector | E1000_IVAR_VALID;
934                 }
935                 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
936                 array_ew32(IVAR0, index, ivar);
937         }
938         if (tx_queue > IGBVF_NO_QUEUE) {
939                 index = (tx_queue >> 1);
940                 ivar = array_er32(IVAR0, index);
941                 if (tx_queue & 0x1) {
942                         /* vector goes into high byte of register */
943                         ivar = ivar & 0x00FFFFFF;
944                         ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
945                 } else {
946                         /* vector goes into second byte of register */
947                         ivar = ivar & 0xFFFF00FF;
948                         ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
949                 }
950                 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
951                 array_ew32(IVAR0, index, ivar);
952         }
953 }
954
955 /**
956  * igbvf_configure_msix - Configure MSI-X hardware
957  *
958  * igbvf_configure_msix sets up the hardware to properly
959  * generate MSI-X interrupts.
960  **/
961 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
962 {
963         u32 tmp;
964         struct e1000_hw *hw = &adapter->hw;
965         struct igbvf_ring *tx_ring = adapter->tx_ring;
966         struct igbvf_ring *rx_ring = adapter->rx_ring;
967         int vector = 0;
968
969         adapter->eims_enable_mask = 0;
970
971         igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
972         adapter->eims_enable_mask |= tx_ring->eims_value;
973         if (tx_ring->itr_val)
974                 writel(tx_ring->itr_val,
975                        hw->hw_addr + tx_ring->itr_register);
976         else
977                 writel(1952, hw->hw_addr + tx_ring->itr_register);
978
979         igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
980         adapter->eims_enable_mask |= rx_ring->eims_value;
981         if (rx_ring->itr_val)
982                 writel(rx_ring->itr_val,
983                        hw->hw_addr + rx_ring->itr_register);
984         else
985                 writel(1952, hw->hw_addr + rx_ring->itr_register);
986
987         /* set vector for other causes, i.e. link changes */
988
989         tmp = (vector++ | E1000_IVAR_VALID);
990
991         ew32(IVAR_MISC, tmp);
992
993         adapter->eims_enable_mask = (1 << (vector)) - 1;
994         adapter->eims_other = 1 << (vector - 1);
995         e1e_flush();
996 }
997
998 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
999 {
1000         if (adapter->msix_entries) {
1001                 pci_disable_msix(adapter->pdev);
1002                 kfree(adapter->msix_entries);
1003                 adapter->msix_entries = NULL;
1004         }
1005 }
1006
1007 /**
1008  * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1009  *
1010  * Attempt to configure interrupts using the best available
1011  * capabilities of the hardware and kernel.
1012  **/
1013 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1014 {
1015         int err = -ENOMEM;
1016         int i;
1017
1018         /* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
1019         adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1020                                         GFP_KERNEL);
1021         if (adapter->msix_entries) {
1022                 for (i = 0; i < 3; i++)
1023                         adapter->msix_entries[i].entry = i;
1024
1025                 err = pci_enable_msix(adapter->pdev,
1026                                       adapter->msix_entries, 3);
1027         }
1028
1029         if (err) {
1030                 /* MSI-X failed */
1031                 dev_err(&adapter->pdev->dev,
1032                         "Failed to initialize MSI-X interrupts.\n");
1033                 igbvf_reset_interrupt_capability(adapter);
1034         }
1035 }
1036
1037 /**
1038  * igbvf_request_msix - Initialize MSI-X interrupts
1039  *
1040  * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1041  * kernel.
1042  **/
1043 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1044 {
1045         struct net_device *netdev = adapter->netdev;
1046         int err = 0, vector = 0;
1047
1048         if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1049                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1050                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1051         } else {
1052                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1053                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1054         }
1055
1056         err = request_irq(adapter->msix_entries[vector].vector,
1057                           &igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1058                           netdev);
1059         if (err)
1060                 goto out;
1061
1062         adapter->tx_ring->itr_register = E1000_EITR(vector);
1063         adapter->tx_ring->itr_val = 1952;
1064         vector++;
1065
1066         err = request_irq(adapter->msix_entries[vector].vector,
1067                           &igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1068                           netdev);
1069         if (err)
1070                 goto out;
1071
1072         adapter->rx_ring->itr_register = E1000_EITR(vector);
1073         adapter->rx_ring->itr_val = 1952;
1074         vector++;
1075
1076         err = request_irq(adapter->msix_entries[vector].vector,
1077                           &igbvf_msix_other, 0, netdev->name, netdev);
1078         if (err)
1079                 goto out;
1080
1081         igbvf_configure_msix(adapter);
1082         return 0;
1083 out:
1084         return err;
1085 }
1086
1087 /**
1088  * igbvf_alloc_queues - Allocate memory for all rings
1089  * @adapter: board private structure to initialize
1090  **/
1091 static int __devinit igbvf_alloc_queues(struct igbvf_adapter *adapter)
1092 {
1093         struct net_device *netdev = adapter->netdev;
1094
1095         adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1096         if (!adapter->tx_ring)
1097                 return -ENOMEM;
1098
1099         adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1100         if (!adapter->rx_ring) {
1101                 kfree(adapter->tx_ring);
1102                 return -ENOMEM;
1103         }
1104
1105         netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1106
1107         return 0;
1108 }
1109
1110 /**
1111  * igbvf_request_irq - initialize interrupts
1112  *
1113  * Attempts to configure interrupts using the best available
1114  * capabilities of the hardware and kernel.
1115  **/
1116 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1117 {
1118         int err = -1;
1119
1120         /* igbvf supports msi-x only */
1121         if (adapter->msix_entries)
1122                 err = igbvf_request_msix(adapter);
1123
1124         if (!err)
1125                 return err;
1126
1127         dev_err(&adapter->pdev->dev,
1128                 "Unable to allocate interrupt, Error: %d\n", err);
1129
1130         return err;
1131 }
1132
1133 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1134 {
1135         struct net_device *netdev = adapter->netdev;
1136         int vector;
1137
1138         if (adapter->msix_entries) {
1139                 for (vector = 0; vector < 3; vector++)
1140                         free_irq(adapter->msix_entries[vector].vector, netdev);
1141         }
1142 }
1143
1144 /**
1145  * igbvf_irq_disable - Mask off interrupt generation on the NIC
1146  **/
1147 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1148 {
1149         struct e1000_hw *hw = &adapter->hw;
1150
1151         ew32(EIMC, ~0);
1152
1153         if (adapter->msix_entries)
1154                 ew32(EIAC, 0);
1155 }
1156
1157 /**
1158  * igbvf_irq_enable - Enable default interrupt generation settings
1159  **/
1160 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1161 {
1162         struct e1000_hw *hw = &adapter->hw;
1163
1164         ew32(EIAC, adapter->eims_enable_mask);
1165         ew32(EIAM, adapter->eims_enable_mask);
1166         ew32(EIMS, adapter->eims_enable_mask);
1167 }
1168
1169 /**
1170  * igbvf_poll - NAPI Rx polling callback
1171  * @napi: struct associated with this polling callback
1172  * @budget: amount of packets driver is allowed to process this poll
1173  **/
1174 static int igbvf_poll(struct napi_struct *napi, int budget)
1175 {
1176         struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1177         struct igbvf_adapter *adapter = rx_ring->adapter;
1178         struct e1000_hw *hw = &adapter->hw;
1179         int work_done = 0;
1180
1181         igbvf_clean_rx_irq(adapter, &work_done, budget);
1182
1183         /* If not enough Rx work done, exit the polling mode */
1184         if (work_done < budget) {
1185                 napi_complete(napi);
1186
1187                 if (adapter->itr_setting & 3)
1188                         igbvf_set_itr(adapter);
1189
1190                 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1191                         ew32(EIMS, adapter->rx_ring->eims_value);
1192         }
1193
1194         return work_done;
1195 }
1196
1197 /**
1198  * igbvf_set_rlpml - set receive large packet maximum length
1199  * @adapter: board private structure
1200  *
1201  * Configure the maximum size of packets that will be received
1202  */
1203 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1204 {
1205         int max_frame_size = adapter->max_frame_size;
1206         struct e1000_hw *hw = &adapter->hw;
1207
1208         if (adapter->vlgrp)
1209                 max_frame_size += VLAN_TAG_SIZE;
1210
1211         e1000_rlpml_set_vf(hw, max_frame_size);
1212 }
1213
1214 static void igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1215 {
1216         struct igbvf_adapter *adapter = netdev_priv(netdev);
1217         struct e1000_hw *hw = &adapter->hw;
1218
1219         if (hw->mac.ops.set_vfta(hw, vid, true))
1220                 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1221 }
1222
1223 static void igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1224 {
1225         struct igbvf_adapter *adapter = netdev_priv(netdev);
1226         struct e1000_hw *hw = &adapter->hw;
1227
1228         igbvf_irq_disable(adapter);
1229         vlan_group_set_device(adapter->vlgrp, vid, NULL);
1230
1231         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1232                 igbvf_irq_enable(adapter);
1233
1234         if (hw->mac.ops.set_vfta(hw, vid, false))
1235                 dev_err(&adapter->pdev->dev,
1236                         "Failed to remove vlan id %d\n", vid);
1237 }
1238
1239 static void igbvf_vlan_rx_register(struct net_device *netdev,
1240                                    struct vlan_group *grp)
1241 {
1242         struct igbvf_adapter *adapter = netdev_priv(netdev);
1243
1244         adapter->vlgrp = grp;
1245 }
1246
1247 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1248 {
1249         u16 vid;
1250
1251         if (!adapter->vlgrp)
1252                 return;
1253
1254         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1255                 if (!vlan_group_get_device(adapter->vlgrp, vid))
1256                         continue;
1257                 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1258         }
1259
1260         igbvf_set_rlpml(adapter);
1261 }
1262
1263 /**
1264  * igbvf_configure_tx - Configure Transmit Unit after Reset
1265  * @adapter: board private structure
1266  *
1267  * Configure the Tx unit of the MAC after a reset.
1268  **/
1269 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1270 {
1271         struct e1000_hw *hw = &adapter->hw;
1272         struct igbvf_ring *tx_ring = adapter->tx_ring;
1273         u64 tdba;
1274         u32 txdctl, dca_txctrl;
1275
1276         /* disable transmits */
1277         txdctl = er32(TXDCTL(0));
1278         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1279         msleep(10);
1280
1281         /* Setup the HW Tx Head and Tail descriptor pointers */
1282         ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1283         tdba = tx_ring->dma;
1284         ew32(TDBAL(0), (tdba & DMA_32BIT_MASK));
1285         ew32(TDBAH(0), (tdba >> 32));
1286         ew32(TDH(0), 0);
1287         ew32(TDT(0), 0);
1288         tx_ring->head = E1000_TDH(0);
1289         tx_ring->tail = E1000_TDT(0);
1290
1291         /* Turn off Relaxed Ordering on head write-backs.  The writebacks
1292          * MUST be delivered in order or it will completely screw up
1293          * our bookeeping.
1294          */
1295         dca_txctrl = er32(DCA_TXCTRL(0));
1296         dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1297         ew32(DCA_TXCTRL(0), dca_txctrl);
1298
1299         /* enable transmits */
1300         txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1301         ew32(TXDCTL(0), txdctl);
1302
1303         /* Setup Transmit Descriptor Settings for eop descriptor */
1304         adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1305
1306         /* enable Report Status bit */
1307         adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1308
1309         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
1310 }
1311
1312 /**
1313  * igbvf_setup_srrctl - configure the receive control registers
1314  * @adapter: Board private structure
1315  **/
1316 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1317 {
1318         struct e1000_hw *hw = &adapter->hw;
1319         u32 srrctl = 0;
1320
1321         srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1322                     E1000_SRRCTL_BSIZEHDR_MASK |
1323                     E1000_SRRCTL_BSIZEPKT_MASK);
1324
1325         /* Enable queue drop to avoid head of line blocking */
1326         srrctl |= E1000_SRRCTL_DROP_EN;
1327
1328         /* Setup buffer sizes */
1329         srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1330                   E1000_SRRCTL_BSIZEPKT_SHIFT;
1331
1332         if (adapter->rx_buffer_len < 2048) {
1333                 adapter->rx_ps_hdr_size = 0;
1334                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1335         } else {
1336                 adapter->rx_ps_hdr_size = 128;
1337                 srrctl |= adapter->rx_ps_hdr_size <<
1338                           E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1339                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1340         }
1341
1342         ew32(SRRCTL(0), srrctl);
1343 }
1344
1345 /**
1346  * igbvf_configure_rx - Configure Receive Unit after Reset
1347  * @adapter: board private structure
1348  *
1349  * Configure the Rx unit of the MAC after a reset.
1350  **/
1351 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1352 {
1353         struct e1000_hw *hw = &adapter->hw;
1354         struct igbvf_ring *rx_ring = adapter->rx_ring;
1355         u64 rdba;
1356         u32 rdlen, rxdctl;
1357
1358         /* disable receives */
1359         rxdctl = er32(RXDCTL(0));
1360         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1361         msleep(10);
1362
1363         rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1364
1365         /*
1366          * Setup the HW Rx Head and Tail Descriptor Pointers and
1367          * the Base and Length of the Rx Descriptor Ring
1368          */
1369         rdba = rx_ring->dma;
1370         ew32(RDBAL(0), (rdba & DMA_32BIT_MASK));
1371         ew32(RDBAH(0), (rdba >> 32));
1372         ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1373         rx_ring->head = E1000_RDH(0);
1374         rx_ring->tail = E1000_RDT(0);
1375         ew32(RDH(0), 0);
1376         ew32(RDT(0), 0);
1377
1378         rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1379         rxdctl &= 0xFFF00000;
1380         rxdctl |= IGBVF_RX_PTHRESH;
1381         rxdctl |= IGBVF_RX_HTHRESH << 8;
1382         rxdctl |= IGBVF_RX_WTHRESH << 16;
1383
1384         igbvf_set_rlpml(adapter);
1385
1386         /* enable receives */
1387         ew32(RXDCTL(0), rxdctl);
1388 }
1389
1390 /**
1391  * igbvf_set_multi - Multicast and Promiscuous mode set
1392  * @netdev: network interface device structure
1393  *
1394  * The set_multi entry point is called whenever the multicast address
1395  * list or the network interface flags are updated.  This routine is
1396  * responsible for configuring the hardware for proper multicast,
1397  * promiscuous mode, and all-multi behavior.
1398  **/
1399 static void igbvf_set_multi(struct net_device *netdev)
1400 {
1401         struct igbvf_adapter *adapter = netdev_priv(netdev);
1402         struct e1000_hw *hw = &adapter->hw;
1403         struct dev_mc_list *mc_ptr;
1404         u8  *mta_list = NULL;
1405         int i;
1406
1407         if (netdev->mc_count) {
1408                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
1409                 if (!mta_list) {
1410                         dev_err(&adapter->pdev->dev,
1411                                 "failed to allocate multicast filter list\n");
1412                         return;
1413                 }
1414         }
1415
1416         /* prepare a packed array of only addresses. */
1417         mc_ptr = netdev->mc_list;
1418
1419         for (i = 0; i < netdev->mc_count; i++) {
1420                 if (!mc_ptr)
1421                         break;
1422                 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
1423                        ETH_ALEN);
1424                 mc_ptr = mc_ptr->next;
1425         }
1426
1427         hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1428         kfree(mta_list);
1429 }
1430
1431 /**
1432  * igbvf_configure - configure the hardware for Rx and Tx
1433  * @adapter: private board structure
1434  **/
1435 static void igbvf_configure(struct igbvf_adapter *adapter)
1436 {
1437         igbvf_set_multi(adapter->netdev);
1438
1439         igbvf_restore_vlan(adapter);
1440
1441         igbvf_configure_tx(adapter);
1442         igbvf_setup_srrctl(adapter);
1443         igbvf_configure_rx(adapter);
1444         igbvf_alloc_rx_buffers(adapter->rx_ring,
1445                                igbvf_desc_unused(adapter->rx_ring));
1446 }
1447
1448 /* igbvf_reset - bring the hardware into a known good state
1449  *
1450  * This function boots the hardware and enables some settings that
1451  * require a configuration cycle of the hardware - those cannot be
1452  * set/changed during runtime. After reset the device needs to be
1453  * properly configured for Rx, Tx etc.
1454  */
1455 static void igbvf_reset(struct igbvf_adapter *adapter)
1456 {
1457         struct e1000_mac_info *mac = &adapter->hw.mac;
1458         struct net_device *netdev = adapter->netdev;
1459         struct e1000_hw *hw = &adapter->hw;
1460
1461         /* Allow time for pending master requests to run */
1462         if (mac->ops.reset_hw(hw))
1463                 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1464
1465         mac->ops.init_hw(hw);
1466
1467         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1468                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1469                        netdev->addr_len);
1470                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1471                        netdev->addr_len);
1472         }
1473 }
1474
1475 int igbvf_up(struct igbvf_adapter *adapter)
1476 {
1477         struct e1000_hw *hw = &adapter->hw;
1478
1479         /* hardware has been reset, we need to reload some things */
1480         igbvf_configure(adapter);
1481
1482         clear_bit(__IGBVF_DOWN, &adapter->state);
1483
1484         napi_enable(&adapter->rx_ring->napi);
1485         if (adapter->msix_entries)
1486                 igbvf_configure_msix(adapter);
1487
1488         /* Clear any pending interrupts. */
1489         er32(EICR);
1490         igbvf_irq_enable(adapter);
1491
1492         /* start the watchdog */
1493         hw->mac.get_link_status = 1;
1494         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1495
1496
1497         return 0;
1498 }
1499
1500 void igbvf_down(struct igbvf_adapter *adapter)
1501 {
1502         struct net_device *netdev = adapter->netdev;
1503         struct e1000_hw *hw = &adapter->hw;
1504         u32 rxdctl, txdctl;
1505
1506         /*
1507          * signal that we're down so the interrupt handler does not
1508          * reschedule our watchdog timer
1509          */
1510         set_bit(__IGBVF_DOWN, &adapter->state);
1511
1512         /* disable receives in the hardware */
1513         rxdctl = er32(RXDCTL(0));
1514         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1515
1516         netif_stop_queue(netdev);
1517
1518         /* disable transmits in the hardware */
1519         txdctl = er32(TXDCTL(0));
1520         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1521
1522         /* flush both disables and wait for them to finish */
1523         e1e_flush();
1524         msleep(10);
1525
1526         napi_disable(&adapter->rx_ring->napi);
1527
1528         igbvf_irq_disable(adapter);
1529
1530         del_timer_sync(&adapter->watchdog_timer);
1531
1532         netdev->tx_queue_len = adapter->tx_queue_len;
1533         netif_carrier_off(netdev);
1534
1535         /* record the stats before reset*/
1536         igbvf_update_stats(adapter);
1537
1538         adapter->link_speed = 0;
1539         adapter->link_duplex = 0;
1540
1541         igbvf_reset(adapter);
1542         igbvf_clean_tx_ring(adapter->tx_ring);
1543         igbvf_clean_rx_ring(adapter->rx_ring);
1544 }
1545
1546 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1547 {
1548         might_sleep();
1549         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1550                 msleep(1);
1551         igbvf_down(adapter);
1552         igbvf_up(adapter);
1553         clear_bit(__IGBVF_RESETTING, &adapter->state);
1554 }
1555
1556 /**
1557  * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1558  * @adapter: board private structure to initialize
1559  *
1560  * igbvf_sw_init initializes the Adapter private data structure.
1561  * Fields are initialized based on PCI device information and
1562  * OS network device settings (MTU size).
1563  **/
1564 static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1565 {
1566         struct net_device *netdev = adapter->netdev;
1567         s32 rc;
1568
1569         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1570         adapter->rx_ps_hdr_size = 0;
1571         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1572         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1573
1574         adapter->tx_int_delay = 8;
1575         adapter->tx_abs_int_delay = 32;
1576         adapter->rx_int_delay = 0;
1577         adapter->rx_abs_int_delay = 8;
1578         adapter->itr_setting = 3;
1579         adapter->itr = 20000;
1580
1581         /* Set various function pointers */
1582         adapter->ei->init_ops(&adapter->hw);
1583
1584         rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1585         if (rc)
1586                 return rc;
1587
1588         rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1589         if (rc)
1590                 return rc;
1591
1592         igbvf_set_interrupt_capability(adapter);
1593
1594         if (igbvf_alloc_queues(adapter))
1595                 return -ENOMEM;
1596
1597         spin_lock_init(&adapter->tx_queue_lock);
1598
1599         /* Explicitly disable IRQ since the NIC can be in any state. */
1600         igbvf_irq_disable(adapter);
1601
1602         spin_lock_init(&adapter->stats_lock);
1603
1604         set_bit(__IGBVF_DOWN, &adapter->state);
1605         return 0;
1606 }
1607
1608 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1609 {
1610         struct e1000_hw *hw = &adapter->hw;
1611
1612         adapter->stats.last_gprc = er32(VFGPRC);
1613         adapter->stats.last_gorc = er32(VFGORC);
1614         adapter->stats.last_gptc = er32(VFGPTC);
1615         adapter->stats.last_gotc = er32(VFGOTC);
1616         adapter->stats.last_mprc = er32(VFMPRC);
1617         adapter->stats.last_gotlbc = er32(VFGOTLBC);
1618         adapter->stats.last_gptlbc = er32(VFGPTLBC);
1619         adapter->stats.last_gorlbc = er32(VFGORLBC);
1620         adapter->stats.last_gprlbc = er32(VFGPRLBC);
1621
1622         adapter->stats.base_gprc = er32(VFGPRC);
1623         adapter->stats.base_gorc = er32(VFGORC);
1624         adapter->stats.base_gptc = er32(VFGPTC);
1625         adapter->stats.base_gotc = er32(VFGOTC);
1626         adapter->stats.base_mprc = er32(VFMPRC);
1627         adapter->stats.base_gotlbc = er32(VFGOTLBC);
1628         adapter->stats.base_gptlbc = er32(VFGPTLBC);
1629         adapter->stats.base_gorlbc = er32(VFGORLBC);
1630         adapter->stats.base_gprlbc = er32(VFGPRLBC);
1631 }
1632
1633 /**
1634  * igbvf_open - Called when a network interface is made active
1635  * @netdev: network interface device structure
1636  *
1637  * Returns 0 on success, negative value on failure
1638  *
1639  * The open entry point is called when a network interface is made
1640  * active by the system (IFF_UP).  At this point all resources needed
1641  * for transmit and receive operations are allocated, the interrupt
1642  * handler is registered with the OS, the watchdog timer is started,
1643  * and the stack is notified that the interface is ready.
1644  **/
1645 static int igbvf_open(struct net_device *netdev)
1646 {
1647         struct igbvf_adapter *adapter = netdev_priv(netdev);
1648         struct e1000_hw *hw = &adapter->hw;
1649         int err;
1650
1651         /* disallow open during test */
1652         if (test_bit(__IGBVF_TESTING, &adapter->state))
1653                 return -EBUSY;
1654
1655         /* allocate transmit descriptors */
1656         err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1657         if (err)
1658                 goto err_setup_tx;
1659
1660         /* allocate receive descriptors */
1661         err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1662         if (err)
1663                 goto err_setup_rx;
1664
1665         /*
1666          * before we allocate an interrupt, we must be ready to handle it.
1667          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1668          * as soon as we call pci_request_irq, so we have to setup our
1669          * clean_rx handler before we do so.
1670          */
1671         igbvf_configure(adapter);
1672
1673         err = igbvf_request_irq(adapter);
1674         if (err)
1675                 goto err_req_irq;
1676
1677         /* From here on the code is the same as igbvf_up() */
1678         clear_bit(__IGBVF_DOWN, &adapter->state);
1679
1680         napi_enable(&adapter->rx_ring->napi);
1681
1682         /* clear any pending interrupts */
1683         er32(EICR);
1684
1685         igbvf_irq_enable(adapter);
1686
1687         /* start the watchdog */
1688         hw->mac.get_link_status = 1;
1689         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1690
1691         return 0;
1692
1693 err_req_irq:
1694         igbvf_free_rx_resources(adapter->rx_ring);
1695 err_setup_rx:
1696         igbvf_free_tx_resources(adapter->tx_ring);
1697 err_setup_tx:
1698         igbvf_reset(adapter);
1699
1700         return err;
1701 }
1702
1703 /**
1704  * igbvf_close - Disables a network interface
1705  * @netdev: network interface device structure
1706  *
1707  * Returns 0, this is not allowed to fail
1708  *
1709  * The close entry point is called when an interface is de-activated
1710  * by the OS.  The hardware is still under the drivers control, but
1711  * needs to be disabled.  A global MAC reset is issued to stop the
1712  * hardware, and all transmit and receive resources are freed.
1713  **/
1714 static int igbvf_close(struct net_device *netdev)
1715 {
1716         struct igbvf_adapter *adapter = netdev_priv(netdev);
1717
1718         WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1719         igbvf_down(adapter);
1720
1721         igbvf_free_irq(adapter);
1722
1723         igbvf_free_tx_resources(adapter->tx_ring);
1724         igbvf_free_rx_resources(adapter->rx_ring);
1725
1726         return 0;
1727 }
1728 /**
1729  * igbvf_set_mac - Change the Ethernet Address of the NIC
1730  * @netdev: network interface device structure
1731  * @p: pointer to an address structure
1732  *
1733  * Returns 0 on success, negative on failure
1734  **/
1735 static int igbvf_set_mac(struct net_device *netdev, void *p)
1736 {
1737         struct igbvf_adapter *adapter = netdev_priv(netdev);
1738         struct e1000_hw *hw = &adapter->hw;
1739         struct sockaddr *addr = p;
1740
1741         if (!is_valid_ether_addr(addr->sa_data))
1742                 return -EADDRNOTAVAIL;
1743
1744         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1745
1746         hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1747
1748         if (memcmp(addr->sa_data, hw->mac.addr, 6))
1749                 return -EADDRNOTAVAIL;
1750
1751         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1752
1753         return 0;
1754 }
1755
1756 #define UPDATE_VF_COUNTER(reg, name)                                    \
1757         {                                                               \
1758                 u32 current_counter = er32(reg);                        \
1759                 if (current_counter < adapter->stats.last_##name)       \
1760                         adapter->stats.name += 0x100000000LL;           \
1761                 adapter->stats.last_##name = current_counter;           \
1762                 adapter->stats.name &= 0xFFFFFFFF00000000LL;            \
1763                 adapter->stats.name |= current_counter;                 \
1764         }
1765
1766 /**
1767  * igbvf_update_stats - Update the board statistics counters
1768  * @adapter: board private structure
1769 **/
1770 void igbvf_update_stats(struct igbvf_adapter *adapter)
1771 {
1772         struct e1000_hw *hw = &adapter->hw;
1773         struct pci_dev *pdev = adapter->pdev;
1774
1775         /*
1776          * Prevent stats update while adapter is being reset, link is down
1777          * or if the pci connection is down.
1778          */
1779         if (adapter->link_speed == 0)
1780                 return;
1781
1782         if (test_bit(__IGBVF_RESETTING, &adapter->state))
1783                 return;
1784
1785         if (pci_channel_offline(pdev))
1786                 return;
1787
1788         UPDATE_VF_COUNTER(VFGPRC, gprc);
1789         UPDATE_VF_COUNTER(VFGORC, gorc);
1790         UPDATE_VF_COUNTER(VFGPTC, gptc);
1791         UPDATE_VF_COUNTER(VFGOTC, gotc);
1792         UPDATE_VF_COUNTER(VFMPRC, mprc);
1793         UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1794         UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1795         UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1796         UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1797
1798         /* Fill out the OS statistics structure */
1799         adapter->net_stats.multicast = adapter->stats.mprc;
1800 }
1801
1802 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1803 {
1804         dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s\n",
1805                  adapter->link_speed,
1806                  ((adapter->link_duplex == FULL_DUPLEX) ?
1807                   "Full Duplex" : "Half Duplex"));
1808 }
1809
1810 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1811 {
1812         struct e1000_hw *hw = &adapter->hw;
1813         s32 ret_val = E1000_SUCCESS;
1814         bool link_active;
1815
1816         ret_val = hw->mac.ops.check_for_link(hw);
1817         link_active = !hw->mac.get_link_status;
1818
1819         /* if check for link returns error we will need to reset */
1820         if (ret_val)
1821                 schedule_work(&adapter->reset_task);
1822
1823         return link_active;
1824 }
1825
1826 /**
1827  * igbvf_watchdog - Timer Call-back
1828  * @data: pointer to adapter cast into an unsigned long
1829  **/
1830 static void igbvf_watchdog(unsigned long data)
1831 {
1832         struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1833
1834         /* Do the rest outside of interrupt context */
1835         schedule_work(&adapter->watchdog_task);
1836 }
1837
1838 static void igbvf_watchdog_task(struct work_struct *work)
1839 {
1840         struct igbvf_adapter *adapter = container_of(work,
1841                                                      struct igbvf_adapter,
1842                                                      watchdog_task);
1843         struct net_device *netdev = adapter->netdev;
1844         struct e1000_mac_info *mac = &adapter->hw.mac;
1845         struct igbvf_ring *tx_ring = adapter->tx_ring;
1846         struct e1000_hw *hw = &adapter->hw;
1847         u32 link;
1848         int tx_pending = 0;
1849
1850         link = igbvf_has_link(adapter);
1851
1852         if (link) {
1853                 if (!netif_carrier_ok(netdev)) {
1854                         bool txb2b = 1;
1855
1856                         mac->ops.get_link_up_info(&adapter->hw,
1857                                                   &adapter->link_speed,
1858                                                   &adapter->link_duplex);
1859                         igbvf_print_link_info(adapter);
1860
1861                         /*
1862                          * tweak tx_queue_len according to speed/duplex
1863                          * and adjust the timeout factor
1864                          */
1865                         netdev->tx_queue_len = adapter->tx_queue_len;
1866                         adapter->tx_timeout_factor = 1;
1867                         switch (adapter->link_speed) {
1868                         case SPEED_10:
1869                                 txb2b = 0;
1870                                 netdev->tx_queue_len = 10;
1871                                 adapter->tx_timeout_factor = 16;
1872                                 break;
1873                         case SPEED_100:
1874                                 txb2b = 0;
1875                                 netdev->tx_queue_len = 100;
1876                                 /* maybe add some timeout factor ? */
1877                                 break;
1878                         }
1879
1880                         netif_carrier_on(netdev);
1881                         netif_wake_queue(netdev);
1882                 }
1883         } else {
1884                 if (netif_carrier_ok(netdev)) {
1885                         adapter->link_speed = 0;
1886                         adapter->link_duplex = 0;
1887                         dev_info(&adapter->pdev->dev, "Link is Down\n");
1888                         netif_carrier_off(netdev);
1889                         netif_stop_queue(netdev);
1890                 }
1891         }
1892
1893         if (netif_carrier_ok(netdev)) {
1894                 igbvf_update_stats(adapter);
1895         } else {
1896                 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1897                               tx_ring->count);
1898                 if (tx_pending) {
1899                         /*
1900                          * We've lost link, so the controller stops DMA,
1901                          * but we've got queued Tx work that's never going
1902                          * to get done, so reset controller to flush Tx.
1903                          * (Do the reset outside of interrupt context).
1904                          */
1905                         adapter->tx_timeout_count++;
1906                         schedule_work(&adapter->reset_task);
1907                 }
1908         }
1909
1910         /* Cause software interrupt to ensure Rx ring is cleaned */
1911         ew32(EICS, adapter->rx_ring->eims_value);
1912
1913         /* Force detection of hung controller every watchdog period */
1914         adapter->detect_tx_hung = 1;
1915
1916         /* Reset the timer */
1917         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1918                 mod_timer(&adapter->watchdog_timer,
1919                           round_jiffies(jiffies + (2 * HZ)));
1920 }
1921
1922 #define IGBVF_TX_FLAGS_CSUM             0x00000001
1923 #define IGBVF_TX_FLAGS_VLAN             0x00000002
1924 #define IGBVF_TX_FLAGS_TSO              0x00000004
1925 #define IGBVF_TX_FLAGS_IPV4             0x00000008
1926 #define IGBVF_TX_FLAGS_VLAN_MASK        0xffff0000
1927 #define IGBVF_TX_FLAGS_VLAN_SHIFT       16
1928
1929 static int igbvf_tso(struct igbvf_adapter *adapter,
1930                      struct igbvf_ring *tx_ring,
1931                      struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1932 {
1933         struct e1000_adv_tx_context_desc *context_desc;
1934         unsigned int i;
1935         int err;
1936         struct igbvf_buffer *buffer_info;
1937         u32 info = 0, tu_cmd = 0;
1938         u32 mss_l4len_idx, l4len;
1939         *hdr_len = 0;
1940
1941         if (skb_header_cloned(skb)) {
1942                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1943                 if (err) {
1944                         dev_err(&adapter->pdev->dev,
1945                                 "igbvf_tso returning an error\n");
1946                         return err;
1947                 }
1948         }
1949
1950         l4len = tcp_hdrlen(skb);
1951         *hdr_len += l4len;
1952
1953         if (skb->protocol == htons(ETH_P_IP)) {
1954                 struct iphdr *iph = ip_hdr(skb);
1955                 iph->tot_len = 0;
1956                 iph->check = 0;
1957                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1958                                                          iph->daddr, 0,
1959                                                          IPPROTO_TCP,
1960                                                          0);
1961         } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
1962                 ipv6_hdr(skb)->payload_len = 0;
1963                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1964                                                        &ipv6_hdr(skb)->daddr,
1965                                                        0, IPPROTO_TCP, 0);
1966         }
1967
1968         i = tx_ring->next_to_use;
1969
1970         buffer_info = &tx_ring->buffer_info[i];
1971         context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1972         /* VLAN MACLEN IPLEN */
1973         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1974                 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1975         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1976         *hdr_len += skb_network_offset(skb);
1977         info |= (skb_transport_header(skb) - skb_network_header(skb));
1978         *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1979         context_desc->vlan_macip_lens = cpu_to_le32(info);
1980
1981         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1982         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1983
1984         if (skb->protocol == htons(ETH_P_IP))
1985                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1986         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1987
1988         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1989
1990         /* MSS L4LEN IDX */
1991         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1992         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1993
1994         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1995         context_desc->seqnum_seed = 0;
1996
1997         buffer_info->time_stamp = jiffies;
1998         buffer_info->next_to_watch = i;
1999         buffer_info->dma = 0;
2000         i++;
2001         if (i == tx_ring->count)
2002                 i = 0;
2003
2004         tx_ring->next_to_use = i;
2005
2006         return true;
2007 }
2008
2009 static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
2010                                  struct igbvf_ring *tx_ring,
2011                                  struct sk_buff *skb, u32 tx_flags)
2012 {
2013         struct e1000_adv_tx_context_desc *context_desc;
2014         unsigned int i;
2015         struct igbvf_buffer *buffer_info;
2016         u32 info = 0, tu_cmd = 0;
2017
2018         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2019             (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
2020                 i = tx_ring->next_to_use;
2021                 buffer_info = &tx_ring->buffer_info[i];
2022                 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
2023
2024                 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2025                         info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
2026
2027                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2028                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2029                         info |= (skb_transport_header(skb) -
2030                                  skb_network_header(skb));
2031
2032
2033                 context_desc->vlan_macip_lens = cpu_to_le32(info);
2034
2035                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2036
2037                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2038                         switch (skb->protocol) {
2039                         case __constant_htons(ETH_P_IP):
2040                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2041                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2042                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2043                                 break;
2044                         case __constant_htons(ETH_P_IPV6):
2045                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2046                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2047                                 break;
2048                         default:
2049                                 break;
2050                         }
2051                 }
2052
2053                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2054                 context_desc->seqnum_seed = 0;
2055                 context_desc->mss_l4len_idx = 0;
2056
2057                 buffer_info->time_stamp = jiffies;
2058                 buffer_info->next_to_watch = i;
2059                 buffer_info->dma = 0;
2060                 i++;
2061                 if (i == tx_ring->count)
2062                         i = 0;
2063                 tx_ring->next_to_use = i;
2064
2065                 return true;
2066         }
2067
2068         return false;
2069 }
2070
2071 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2072 {
2073         struct igbvf_adapter *adapter = netdev_priv(netdev);
2074
2075         /* there is enough descriptors then we don't need to worry  */
2076         if (igbvf_desc_unused(adapter->tx_ring) >= size)
2077                 return 0;
2078
2079         netif_stop_queue(netdev);
2080
2081         smp_mb();
2082
2083         /* We need to check again just in case room has been made available */
2084         if (igbvf_desc_unused(adapter->tx_ring) < size)
2085                 return -EBUSY;
2086
2087         netif_wake_queue(netdev);
2088
2089         ++adapter->restart_queue;
2090         return 0;
2091 }
2092
2093 #define IGBVF_MAX_TXD_PWR       16
2094 #define IGBVF_MAX_DATA_PER_TXD  (1 << IGBVF_MAX_TXD_PWR)
2095
2096 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2097                                    struct igbvf_ring *tx_ring,
2098                                    struct sk_buff *skb,
2099                                    unsigned int first)
2100 {
2101         struct igbvf_buffer *buffer_info;
2102         unsigned int len = skb_headlen(skb);
2103         unsigned int count = 0, i;
2104         unsigned int f;
2105         dma_addr_t *map;
2106
2107         i = tx_ring->next_to_use;
2108
2109         if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
2110                 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
2111                 return 0;
2112         }
2113
2114         map = skb_shinfo(skb)->dma_maps;
2115
2116         buffer_info = &tx_ring->buffer_info[i];
2117         BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2118         buffer_info->length = len;
2119         /* set time_stamp *before* dma to help avoid a possible race */
2120         buffer_info->time_stamp = jiffies;
2121         buffer_info->next_to_watch = i;
2122         buffer_info->dma = map[count];
2123         count++;
2124
2125         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2126                 struct skb_frag_struct *frag;
2127
2128                 i++;
2129                 if (i == tx_ring->count)
2130                         i = 0;
2131
2132                 frag = &skb_shinfo(skb)->frags[f];
2133                 len = frag->size;
2134
2135                 buffer_info = &tx_ring->buffer_info[i];
2136                 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2137                 buffer_info->length = len;
2138                 buffer_info->time_stamp = jiffies;
2139                 buffer_info->next_to_watch = i;
2140                 buffer_info->dma = map[count];
2141                 count++;
2142         }
2143
2144         tx_ring->buffer_info[i].skb = skb;
2145         tx_ring->buffer_info[first].next_to_watch = i;
2146
2147         return count;
2148 }
2149
2150 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2151                                       struct igbvf_ring *tx_ring,
2152                                       int tx_flags, int count, u32 paylen,
2153                                       u8 hdr_len)
2154 {
2155         union e1000_adv_tx_desc *tx_desc = NULL;
2156         struct igbvf_buffer *buffer_info;
2157         u32 olinfo_status = 0, cmd_type_len;
2158         unsigned int i;
2159
2160         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2161                         E1000_ADVTXD_DCMD_DEXT);
2162
2163         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2164                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2165
2166         if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2167                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2168
2169                 /* insert tcp checksum */
2170                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2171
2172                 /* insert ip checksum */
2173                 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2174                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2175
2176         } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2177                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2178         }
2179
2180         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2181
2182         i = tx_ring->next_to_use;
2183         while (count--) {
2184                 buffer_info = &tx_ring->buffer_info[i];
2185                 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2186                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2187                 tx_desc->read.cmd_type_len =
2188                          cpu_to_le32(cmd_type_len | buffer_info->length);
2189                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2190                 i++;
2191                 if (i == tx_ring->count)
2192                         i = 0;
2193         }
2194
2195         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2196         /* Force memory writes to complete before letting h/w
2197          * know there are new descriptors to fetch.  (Only
2198          * applicable for weak-ordered memory model archs,
2199          * such as IA-64). */
2200         wmb();
2201
2202         tx_ring->next_to_use = i;
2203         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2204         /* we need this if more than one processor can write to our tail
2205          * at a time, it syncronizes IO on IA64/Altix systems */
2206         mmiowb();
2207 }
2208
2209 static int igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2210                                    struct net_device *netdev,
2211                                    struct igbvf_ring *tx_ring)
2212 {
2213         struct igbvf_adapter *adapter = netdev_priv(netdev);
2214         unsigned int first, tx_flags = 0;
2215         u8 hdr_len = 0;
2216         int count = 0;
2217         int tso = 0;
2218
2219         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2220                 dev_kfree_skb_any(skb);
2221                 return NETDEV_TX_OK;
2222         }
2223
2224         if (skb->len <= 0) {
2225                 dev_kfree_skb_any(skb);
2226                 return NETDEV_TX_OK;
2227         }
2228
2229         /*
2230          * need: count + 4 desc gap to keep tail from touching
2231          *       + 2 desc gap to keep tail from touching head,
2232          *       + 1 desc for skb->data,
2233          *       + 1 desc for context descriptor,
2234          * head, otherwise try next time
2235          */
2236         if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2237                 /* this is a hard error */
2238                 return NETDEV_TX_BUSY;
2239         }
2240
2241         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2242                 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2243                 tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2244         }
2245
2246         if (skb->protocol == htons(ETH_P_IP))
2247                 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2248
2249         first = tx_ring->next_to_use;
2250
2251         tso = skb_is_gso(skb) ?
2252                 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2253         if (unlikely(tso < 0)) {
2254                 dev_kfree_skb_any(skb);
2255                 return NETDEV_TX_OK;
2256         }
2257
2258         if (tso)
2259                 tx_flags |= IGBVF_TX_FLAGS_TSO;
2260         else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2261                  (skb->ip_summed == CHECKSUM_PARTIAL))
2262                 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2263
2264         /*
2265          * count reflects descriptors mapped, if 0 then mapping error
2266          * has occured and we need to rewind the descriptor queue
2267          */
2268         count = igbvf_tx_map_adv(adapter, tx_ring, skb, first);
2269
2270         if (count) {
2271                 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2272                                    skb->len, hdr_len);
2273                 /* Make sure there is space in the ring for the next send. */
2274                 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2275         } else {
2276                 dev_kfree_skb_any(skb);
2277                 tx_ring->buffer_info[first].time_stamp = 0;
2278                 tx_ring->next_to_use = first;
2279         }
2280
2281         return NETDEV_TX_OK;
2282 }
2283
2284 static int igbvf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
2285 {
2286         struct igbvf_adapter *adapter = netdev_priv(netdev);
2287         struct igbvf_ring *tx_ring;
2288         int retval;
2289
2290         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2291                 dev_kfree_skb_any(skb);
2292                 return NETDEV_TX_OK;
2293         }
2294
2295         tx_ring = &adapter->tx_ring[0];
2296
2297         retval = igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2298
2299         return retval;
2300 }
2301
2302 /**
2303  * igbvf_tx_timeout - Respond to a Tx Hang
2304  * @netdev: network interface device structure
2305  **/
2306 static void igbvf_tx_timeout(struct net_device *netdev)
2307 {
2308         struct igbvf_adapter *adapter = netdev_priv(netdev);
2309
2310         /* Do the reset outside of interrupt context */
2311         adapter->tx_timeout_count++;
2312         schedule_work(&adapter->reset_task);
2313 }
2314
2315 static void igbvf_reset_task(struct work_struct *work)
2316 {
2317         struct igbvf_adapter *adapter;
2318         adapter = container_of(work, struct igbvf_adapter, reset_task);
2319
2320         igbvf_reinit_locked(adapter);
2321 }
2322
2323 /**
2324  * igbvf_get_stats - Get System Network Statistics
2325  * @netdev: network interface device structure
2326  *
2327  * Returns the address of the device statistics structure.
2328  * The statistics are actually updated from the timer callback.
2329  **/
2330 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2331 {
2332         struct igbvf_adapter *adapter = netdev_priv(netdev);
2333
2334         /* only return the current stats */
2335         return &adapter->net_stats;
2336 }
2337
2338 /**
2339  * igbvf_change_mtu - Change the Maximum Transfer Unit
2340  * @netdev: network interface device structure
2341  * @new_mtu: new value for maximum frame size
2342  *
2343  * Returns 0 on success, negative on failure
2344  **/
2345 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2346 {
2347         struct igbvf_adapter *adapter = netdev_priv(netdev);
2348         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2349
2350         if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2351                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2352                 return -EINVAL;
2353         }
2354
2355 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2356         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2357                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2358                 return -EINVAL;
2359         }
2360
2361         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2362                 msleep(1);
2363         /* igbvf_down has a dependency on max_frame_size */
2364         adapter->max_frame_size = max_frame;
2365         if (netif_running(netdev))
2366                 igbvf_down(adapter);
2367
2368         /*
2369          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2370          * means we reserve 2 more, this pushes us to allocate from the next
2371          * larger slab size.
2372          * i.e. RXBUFFER_2048 --> size-4096 slab
2373          * However with the new *_jumbo_rx* routines, jumbo receives will use
2374          * fragmented skbs
2375          */
2376
2377         if (max_frame <= 1024)
2378                 adapter->rx_buffer_len = 1024;
2379         else if (max_frame <= 2048)
2380                 adapter->rx_buffer_len = 2048;
2381         else
2382 #if (PAGE_SIZE / 2) > 16384
2383                 adapter->rx_buffer_len = 16384;
2384 #else
2385                 adapter->rx_buffer_len = PAGE_SIZE / 2;
2386 #endif
2387
2388
2389         /* adjust allocation if LPE protects us, and we aren't using SBP */
2390         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2391              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2392                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2393                                          ETH_FCS_LEN;
2394
2395         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2396                  netdev->mtu, new_mtu);
2397         netdev->mtu = new_mtu;
2398
2399         if (netif_running(netdev))
2400                 igbvf_up(adapter);
2401         else
2402                 igbvf_reset(adapter);
2403
2404         clear_bit(__IGBVF_RESETTING, &adapter->state);
2405
2406         return 0;
2407 }
2408
2409 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2410 {
2411         switch (cmd) {
2412         default:
2413                 return -EOPNOTSUPP;
2414         }
2415 }
2416
2417 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2418 {
2419         struct net_device *netdev = pci_get_drvdata(pdev);
2420         struct igbvf_adapter *adapter = netdev_priv(netdev);
2421 #ifdef CONFIG_PM
2422         int retval = 0;
2423 #endif
2424
2425         netif_device_detach(netdev);
2426
2427         if (netif_running(netdev)) {
2428                 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2429                 igbvf_down(adapter);
2430                 igbvf_free_irq(adapter);
2431         }
2432
2433 #ifdef CONFIG_PM
2434         retval = pci_save_state(pdev);
2435         if (retval)
2436                 return retval;
2437 #endif
2438
2439         pci_disable_device(pdev);
2440
2441         return 0;
2442 }
2443
2444 #ifdef CONFIG_PM
2445 static int igbvf_resume(struct pci_dev *pdev)
2446 {
2447         struct net_device *netdev = pci_get_drvdata(pdev);
2448         struct igbvf_adapter *adapter = netdev_priv(netdev);
2449         u32 err;
2450
2451         pci_restore_state(pdev);
2452         err = pci_enable_device_mem(pdev);
2453         if (err) {
2454                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2455                 return err;
2456         }
2457
2458         pci_set_master(pdev);
2459
2460         if (netif_running(netdev)) {
2461                 err = igbvf_request_irq(adapter);
2462                 if (err)
2463                         return err;
2464         }
2465
2466         igbvf_reset(adapter);
2467
2468         if (netif_running(netdev))
2469                 igbvf_up(adapter);
2470
2471         netif_device_attach(netdev);
2472
2473         return 0;
2474 }
2475 #endif
2476
2477 static void igbvf_shutdown(struct pci_dev *pdev)
2478 {
2479         igbvf_suspend(pdev, PMSG_SUSPEND);
2480 }
2481
2482 #ifdef CONFIG_NET_POLL_CONTROLLER
2483 /*
2484  * Polling 'interrupt' - used by things like netconsole to send skbs
2485  * without having to re-enable interrupts. It's not called while
2486  * the interrupt routine is executing.
2487  */
2488 static void igbvf_netpoll(struct net_device *netdev)
2489 {
2490         struct igbvf_adapter *adapter = netdev_priv(netdev);
2491
2492         disable_irq(adapter->pdev->irq);
2493
2494         igbvf_clean_tx_irq(adapter->tx_ring);
2495
2496         enable_irq(adapter->pdev->irq);
2497 }
2498 #endif
2499
2500 /**
2501  * igbvf_io_error_detected - called when PCI error is detected
2502  * @pdev: Pointer to PCI device
2503  * @state: The current pci connection state
2504  *
2505  * This function is called after a PCI bus error affecting
2506  * this device has been detected.
2507  */
2508 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2509                                                 pci_channel_state_t state)
2510 {
2511         struct net_device *netdev = pci_get_drvdata(pdev);
2512         struct igbvf_adapter *adapter = netdev_priv(netdev);
2513
2514         netif_device_detach(netdev);
2515
2516         if (netif_running(netdev))
2517                 igbvf_down(adapter);
2518         pci_disable_device(pdev);
2519
2520         /* Request a slot slot reset. */
2521         return PCI_ERS_RESULT_NEED_RESET;
2522 }
2523
2524 /**
2525  * igbvf_io_slot_reset - called after the pci bus has been reset.
2526  * @pdev: Pointer to PCI device
2527  *
2528  * Restart the card from scratch, as if from a cold-boot. Implementation
2529  * resembles the first-half of the igbvf_resume routine.
2530  */
2531 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2532 {
2533         struct net_device *netdev = pci_get_drvdata(pdev);
2534         struct igbvf_adapter *adapter = netdev_priv(netdev);
2535
2536         if (pci_enable_device_mem(pdev)) {
2537                 dev_err(&pdev->dev,
2538                         "Cannot re-enable PCI device after reset.\n");
2539                 return PCI_ERS_RESULT_DISCONNECT;
2540         }
2541         pci_set_master(pdev);
2542
2543         igbvf_reset(adapter);
2544
2545         return PCI_ERS_RESULT_RECOVERED;
2546 }
2547
2548 /**
2549  * igbvf_io_resume - called when traffic can start flowing again.
2550  * @pdev: Pointer to PCI device
2551  *
2552  * This callback is called when the error recovery driver tells us that
2553  * its OK to resume normal operation. Implementation resembles the
2554  * second-half of the igbvf_resume routine.
2555  */
2556 static void igbvf_io_resume(struct pci_dev *pdev)
2557 {
2558         struct net_device *netdev = pci_get_drvdata(pdev);
2559         struct igbvf_adapter *adapter = netdev_priv(netdev);
2560
2561         if (netif_running(netdev)) {
2562                 if (igbvf_up(adapter)) {
2563                         dev_err(&pdev->dev,
2564                                 "can't bring device back up after reset\n");
2565                         return;
2566                 }
2567         }
2568
2569         netif_device_attach(netdev);
2570 }
2571
2572 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2573 {
2574         struct e1000_hw *hw = &adapter->hw;
2575         struct net_device *netdev = adapter->netdev;
2576         struct pci_dev *pdev = adapter->pdev;
2577
2578         dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2579         dev_info(&pdev->dev, "Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
2580                  /* MAC address */
2581                  netdev->dev_addr[0], netdev->dev_addr[1],
2582                  netdev->dev_addr[2], netdev->dev_addr[3],
2583                  netdev->dev_addr[4], netdev->dev_addr[5]);
2584         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
2585 }
2586
2587 static const struct net_device_ops igbvf_netdev_ops = {
2588         .ndo_open                       = igbvf_open,
2589         .ndo_stop                       = igbvf_close,
2590         .ndo_start_xmit                 = igbvf_xmit_frame,
2591         .ndo_get_stats                  = igbvf_get_stats,
2592         .ndo_set_multicast_list         = igbvf_set_multi,
2593         .ndo_set_mac_address            = igbvf_set_mac,
2594         .ndo_change_mtu                 = igbvf_change_mtu,
2595         .ndo_do_ioctl                   = igbvf_ioctl,
2596         .ndo_tx_timeout                 = igbvf_tx_timeout,
2597         .ndo_vlan_rx_register           = igbvf_vlan_rx_register,
2598         .ndo_vlan_rx_add_vid            = igbvf_vlan_rx_add_vid,
2599         .ndo_vlan_rx_kill_vid           = igbvf_vlan_rx_kill_vid,
2600 #ifdef CONFIG_NET_POLL_CONTROLLER
2601         .ndo_poll_controller            = igbvf_netpoll,
2602 #endif
2603 };
2604
2605 /**
2606  * igbvf_probe - Device Initialization Routine
2607  * @pdev: PCI device information struct
2608  * @ent: entry in igbvf_pci_tbl
2609  *
2610  * Returns 0 on success, negative on failure
2611  *
2612  * igbvf_probe initializes an adapter identified by a pci_dev structure.
2613  * The OS initialization, configuring of the adapter private structure,
2614  * and a hardware reset occur.
2615  **/
2616 static int __devinit igbvf_probe(struct pci_dev *pdev,
2617                                  const struct pci_device_id *ent)
2618 {
2619         struct net_device *netdev;
2620         struct igbvf_adapter *adapter;
2621         struct e1000_hw *hw;
2622         const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2623
2624         static int cards_found;
2625         int err, pci_using_dac;
2626
2627         err = pci_enable_device_mem(pdev);
2628         if (err)
2629                 return err;
2630
2631         pci_using_dac = 0;
2632         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
2633         if (!err) {
2634                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
2635                 if (!err)
2636                         pci_using_dac = 1;
2637         } else {
2638                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
2639                 if (err) {
2640                         err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
2641                         if (err) {
2642                                 dev_err(&pdev->dev, "No usable DMA "
2643                                         "configuration, aborting\n");
2644                                 goto err_dma;
2645                         }
2646                 }
2647         }
2648
2649         err = pci_request_regions(pdev, igbvf_driver_name);
2650         if (err)
2651                 goto err_pci_reg;
2652
2653         pci_set_master(pdev);
2654
2655         err = -ENOMEM;
2656         netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2657         if (!netdev)
2658                 goto err_alloc_etherdev;
2659
2660         SET_NETDEV_DEV(netdev, &pdev->dev);
2661
2662         pci_set_drvdata(pdev, netdev);
2663         adapter = netdev_priv(netdev);
2664         hw = &adapter->hw;
2665         adapter->netdev = netdev;
2666         adapter->pdev = pdev;
2667         adapter->ei = ei;
2668         adapter->pba = ei->pba;
2669         adapter->flags = ei->flags;
2670         adapter->hw.back = adapter;
2671         adapter->hw.mac.type = ei->mac;
2672         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
2673
2674         /* PCI config space info */
2675
2676         hw->vendor_id = pdev->vendor;
2677         hw->device_id = pdev->device;
2678         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2679         hw->subsystem_device_id = pdev->subsystem_device;
2680
2681         pci_read_config_byte(pdev, PCI_REVISION_ID, &hw->revision_id);
2682
2683         err = -EIO;
2684         adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2685                                       pci_resource_len(pdev, 0));
2686
2687         if (!adapter->hw.hw_addr)
2688                 goto err_ioremap;
2689
2690         if (ei->get_variants) {
2691                 err = ei->get_variants(adapter);
2692                 if (err)
2693                         goto err_ioremap;
2694         }
2695
2696         /* setup adapter struct */
2697         err = igbvf_sw_init(adapter);
2698         if (err)
2699                 goto err_sw_init;
2700
2701         /* construct the net_device struct */
2702         netdev->netdev_ops = &igbvf_netdev_ops;
2703
2704         igbvf_set_ethtool_ops(netdev);
2705         netdev->watchdog_timeo = 5 * HZ;
2706         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2707
2708         adapter->bd_number = cards_found++;
2709
2710         netdev->features = NETIF_F_SG |
2711                            NETIF_F_IP_CSUM |
2712                            NETIF_F_HW_VLAN_TX |
2713                            NETIF_F_HW_VLAN_RX |
2714                            NETIF_F_HW_VLAN_FILTER;
2715
2716         netdev->features |= NETIF_F_IPV6_CSUM;
2717         netdev->features |= NETIF_F_TSO;
2718         netdev->features |= NETIF_F_TSO6;
2719
2720         if (pci_using_dac)
2721                 netdev->features |= NETIF_F_HIGHDMA;
2722
2723         netdev->vlan_features |= NETIF_F_TSO;
2724         netdev->vlan_features |= NETIF_F_TSO6;
2725         netdev->vlan_features |= NETIF_F_IP_CSUM;
2726         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2727         netdev->vlan_features |= NETIF_F_SG;
2728
2729         /*reset the controller to put the device in a known good state */
2730         err = hw->mac.ops.reset_hw(hw);
2731         if (err) {
2732                 dev_info(&pdev->dev,
2733                          "PF still in reset state, assigning new address\n");
2734                 random_ether_addr(hw->mac.addr);
2735         } else {
2736                 err = hw->mac.ops.read_mac_addr(hw);
2737                 if (err) {
2738                         dev_err(&pdev->dev, "Error reading MAC address\n");
2739                         goto err_hw_init;
2740                 }
2741         }
2742
2743         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
2744         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
2745
2746         if (!is_valid_ether_addr(netdev->perm_addr)) {
2747                 dev_err(&pdev->dev, "Invalid MAC Address: "
2748                         "%02x:%02x:%02x:%02x:%02x:%02x\n",
2749                         netdev->dev_addr[0], netdev->dev_addr[1],
2750                         netdev->dev_addr[2], netdev->dev_addr[3],
2751                         netdev->dev_addr[4], netdev->dev_addr[5]);
2752                 err = -EIO;
2753                 goto err_hw_init;
2754         }
2755
2756         setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2757                     (unsigned long) adapter);
2758
2759         INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2760         INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2761
2762         /* ring size defaults */
2763         adapter->rx_ring->count = 1024;
2764         adapter->tx_ring->count = 1024;
2765
2766         /* reset the hardware with the new settings */
2767         igbvf_reset(adapter);
2768
2769         /* tell the stack to leave us alone until igbvf_open() is called */
2770         netif_carrier_off(netdev);
2771         netif_stop_queue(netdev);
2772
2773         strcpy(netdev->name, "eth%d");
2774         err = register_netdev(netdev);
2775         if (err)
2776                 goto err_hw_init;
2777
2778         igbvf_print_device_info(adapter);
2779
2780         igbvf_initialize_last_counter_stats(adapter);
2781
2782         return 0;
2783
2784 err_hw_init:
2785         kfree(adapter->tx_ring);
2786         kfree(adapter->rx_ring);
2787 err_sw_init:
2788         igbvf_reset_interrupt_capability(adapter);
2789         iounmap(adapter->hw.hw_addr);
2790 err_ioremap:
2791         free_netdev(netdev);
2792 err_alloc_etherdev:
2793         pci_release_regions(pdev);
2794 err_pci_reg:
2795 err_dma:
2796         pci_disable_device(pdev);
2797         return err;
2798 }
2799
2800 /**
2801  * igbvf_remove - Device Removal Routine
2802  * @pdev: PCI device information struct
2803  *
2804  * igbvf_remove is called by the PCI subsystem to alert the driver
2805  * that it should release a PCI device.  The could be caused by a
2806  * Hot-Plug event, or because the driver is going to be removed from
2807  * memory.
2808  **/
2809 static void __devexit igbvf_remove(struct pci_dev *pdev)
2810 {
2811         struct net_device *netdev = pci_get_drvdata(pdev);
2812         struct igbvf_adapter *adapter = netdev_priv(netdev);
2813         struct e1000_hw *hw = &adapter->hw;
2814
2815         /*
2816          * flush_scheduled work may reschedule our watchdog task, so
2817          * explicitly disable watchdog tasks from being rescheduled
2818          */
2819         set_bit(__IGBVF_DOWN, &adapter->state);
2820         del_timer_sync(&adapter->watchdog_timer);
2821
2822         flush_scheduled_work();
2823
2824         unregister_netdev(netdev);
2825
2826         igbvf_reset_interrupt_capability(adapter);
2827
2828         /*
2829          * it is important to delete the napi struct prior to freeing the
2830          * rx ring so that you do not end up with null pointer refs
2831          */
2832         netif_napi_del(&adapter->rx_ring->napi);
2833         kfree(adapter->tx_ring);
2834         kfree(adapter->rx_ring);
2835
2836         iounmap(hw->hw_addr);
2837         if (hw->flash_address)
2838                 iounmap(hw->flash_address);
2839         pci_release_regions(pdev);
2840
2841         free_netdev(netdev);
2842
2843         pci_disable_device(pdev);
2844 }
2845
2846 /* PCI Error Recovery (ERS) */
2847 static struct pci_error_handlers igbvf_err_handler = {
2848         .error_detected = igbvf_io_error_detected,
2849         .slot_reset = igbvf_io_slot_reset,
2850         .resume = igbvf_io_resume,
2851 };
2852
2853 static struct pci_device_id igbvf_pci_tbl[] = {
2854         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2855         { } /* terminate list */
2856 };
2857 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2858
2859 /* PCI Device API Driver */
2860 static struct pci_driver igbvf_driver = {
2861         .name     = igbvf_driver_name,
2862         .id_table = igbvf_pci_tbl,
2863         .probe    = igbvf_probe,
2864         .remove   = __devexit_p(igbvf_remove),
2865 #ifdef CONFIG_PM
2866         /* Power Management Hooks */
2867         .suspend  = igbvf_suspend,
2868         .resume   = igbvf_resume,
2869 #endif
2870         .shutdown = igbvf_shutdown,
2871         .err_handler = &igbvf_err_handler
2872 };
2873
2874 /**
2875  * igbvf_init_module - Driver Registration Routine
2876  *
2877  * igbvf_init_module is the first routine called when the driver is
2878  * loaded. All it does is register with the PCI subsystem.
2879  **/
2880 static int __init igbvf_init_module(void)
2881 {
2882         int ret;
2883         printk(KERN_INFO "%s - version %s\n",
2884                igbvf_driver_string, igbvf_driver_version);
2885         printk(KERN_INFO "%s\n", igbvf_copyright);
2886
2887         ret = pci_register_driver(&igbvf_driver);
2888         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, igbvf_driver_name,
2889                                PM_QOS_DEFAULT_VALUE);
2890
2891         return ret;
2892 }
2893 module_init(igbvf_init_module);
2894
2895 /**
2896  * igbvf_exit_module - Driver Exit Cleanup Routine
2897  *
2898  * igbvf_exit_module is called just before the driver is removed
2899  * from memory.
2900  **/
2901 static void __exit igbvf_exit_module(void)
2902 {
2903         pci_unregister_driver(&igbvf_driver);
2904         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, igbvf_driver_name);
2905 }
2906 module_exit(igbvf_exit_module);
2907
2908
2909 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2910 MODULE_DESCRIPTION("Intel(R) 82576 Virtual Function Network Driver");
2911 MODULE_LICENSE("GPL");
2912 MODULE_VERSION(DRV_VERSION);
2913
2914 /* netdev.c */