igbvf: do not modify tx_queue_len on link speed change
[linux-2.6.git] / 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
101 static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
102                                          u32 status_err, struct sk_buff *skb)
103 {
104         skb->ip_summed = CHECKSUM_NONE;
105
106         /* Ignore Checksum bit is set or checksum is disabled through ethtool */
107         if ((status_err & E1000_RXD_STAT_IXSM) ||
108             (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
109                 return;
110
111         /* TCP/UDP checksum error bit is set */
112         if (status_err &
113             (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
114                 /* let the stack verify checksum errors */
115                 adapter->hw_csum_err++;
116                 return;
117         }
118
119         /* It must be a TCP or UDP packet with a valid checksum */
120         if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
121                 skb->ip_summed = CHECKSUM_UNNECESSARY;
122
123         adapter->hw_csum_good++;
124 }
125
126 /**
127  * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
128  * @rx_ring: address of ring structure to repopulate
129  * @cleaned_count: number of buffers to repopulate
130  **/
131 static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
132                                    int cleaned_count)
133 {
134         struct igbvf_adapter *adapter = rx_ring->adapter;
135         struct net_device *netdev = adapter->netdev;
136         struct pci_dev *pdev = adapter->pdev;
137         union e1000_adv_rx_desc *rx_desc;
138         struct igbvf_buffer *buffer_info;
139         struct sk_buff *skb;
140         unsigned int i;
141         int bufsz;
142
143         i = rx_ring->next_to_use;
144         buffer_info = &rx_ring->buffer_info[i];
145
146         if (adapter->rx_ps_hdr_size)
147                 bufsz = adapter->rx_ps_hdr_size;
148         else
149                 bufsz = adapter->rx_buffer_len;
150
151         while (cleaned_count--) {
152                 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
153
154                 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
155                         if (!buffer_info->page) {
156                                 buffer_info->page = alloc_page(GFP_ATOMIC);
157                                 if (!buffer_info->page) {
158                                         adapter->alloc_rx_buff_failed++;
159                                         goto no_buffers;
160                                 }
161                                 buffer_info->page_offset = 0;
162                         } else {
163                                 buffer_info->page_offset ^= PAGE_SIZE / 2;
164                         }
165                         buffer_info->page_dma =
166                                 pci_map_page(pdev, buffer_info->page,
167                                              buffer_info->page_offset,
168                                              PAGE_SIZE / 2,
169                                              PCI_DMA_FROMDEVICE);
170                 }
171
172                 if (!buffer_info->skb) {
173                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
174                         if (!skb) {
175                                 adapter->alloc_rx_buff_failed++;
176                                 goto no_buffers;
177                         }
178
179                         buffer_info->skb = skb;
180                         buffer_info->dma = pci_map_single(pdev, skb->data,
181                                                           bufsz,
182                                                           PCI_DMA_FROMDEVICE);
183                 }
184                 /* Refresh the desc even if buffer_addrs didn't change because
185                  * each write-back erases this info. */
186                 if (adapter->rx_ps_hdr_size) {
187                         rx_desc->read.pkt_addr =
188                              cpu_to_le64(buffer_info->page_dma);
189                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
190                 } else {
191                         rx_desc->read.pkt_addr =
192                              cpu_to_le64(buffer_info->dma);
193                         rx_desc->read.hdr_addr = 0;
194                 }
195
196                 i++;
197                 if (i == rx_ring->count)
198                         i = 0;
199                 buffer_info = &rx_ring->buffer_info[i];
200         }
201
202 no_buffers:
203         if (rx_ring->next_to_use != i) {
204                 rx_ring->next_to_use = i;
205                 if (i == 0)
206                         i = (rx_ring->count - 1);
207                 else
208                         i--;
209
210                 /* Force memory writes to complete before letting h/w
211                  * know there are new descriptors to fetch.  (Only
212                  * applicable for weak-ordered memory model archs,
213                  * such as IA-64). */
214                 wmb();
215                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
216         }
217 }
218
219 /**
220  * igbvf_clean_rx_irq - Send received data up the network stack; legacy
221  * @adapter: board private structure
222  *
223  * the return value indicates whether actual cleaning was done, there
224  * is no guarantee that everything was cleaned
225  **/
226 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
227                                int *work_done, int work_to_do)
228 {
229         struct igbvf_ring *rx_ring = adapter->rx_ring;
230         struct net_device *netdev = adapter->netdev;
231         struct pci_dev *pdev = adapter->pdev;
232         union e1000_adv_rx_desc *rx_desc, *next_rxd;
233         struct igbvf_buffer *buffer_info, *next_buffer;
234         struct sk_buff *skb;
235         bool cleaned = false;
236         int cleaned_count = 0;
237         unsigned int total_bytes = 0, total_packets = 0;
238         unsigned int i;
239         u32 length, hlen, staterr;
240
241         i = rx_ring->next_to_clean;
242         rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
243         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
244
245         while (staterr & E1000_RXD_STAT_DD) {
246                 if (*work_done >= work_to_do)
247                         break;
248                 (*work_done)++;
249
250                 buffer_info = &rx_ring->buffer_info[i];
251
252                 /* HW will not DMA in data larger than the given buffer, even
253                  * if it parses the (NFS, of course) header to be larger.  In
254                  * that case, it fills the header buffer and spills the rest
255                  * into the page.
256                  */
257                 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
258                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
259                 if (hlen > adapter->rx_ps_hdr_size)
260                         hlen = adapter->rx_ps_hdr_size;
261
262                 length = le16_to_cpu(rx_desc->wb.upper.length);
263                 cleaned = true;
264                 cleaned_count++;
265
266                 skb = buffer_info->skb;
267                 prefetch(skb->data - NET_IP_ALIGN);
268                 buffer_info->skb = NULL;
269                 if (!adapter->rx_ps_hdr_size) {
270                         pci_unmap_single(pdev, buffer_info->dma,
271                                          adapter->rx_buffer_len,
272                                          PCI_DMA_FROMDEVICE);
273                         buffer_info->dma = 0;
274                         skb_put(skb, length);
275                         goto send_up;
276                 }
277
278                 if (!skb_shinfo(skb)->nr_frags) {
279                         pci_unmap_single(pdev, buffer_info->dma,
280                                          adapter->rx_ps_hdr_size,
281                                          PCI_DMA_FROMDEVICE);
282                         skb_put(skb, hlen);
283                 }
284
285                 if (length) {
286                         pci_unmap_page(pdev, buffer_info->page_dma,
287                                        PAGE_SIZE / 2,
288                                        PCI_DMA_FROMDEVICE);
289                         buffer_info->page_dma = 0;
290
291                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
292                                            buffer_info->page,
293                                            buffer_info->page_offset,
294                                            length);
295
296                         if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
297                             (page_count(buffer_info->page) != 1))
298                                 buffer_info->page = NULL;
299                         else
300                                 get_page(buffer_info->page);
301
302                         skb->len += length;
303                         skb->data_len += length;
304                         skb->truesize += length;
305                 }
306 send_up:
307                 i++;
308                 if (i == rx_ring->count)
309                         i = 0;
310                 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
311                 prefetch(next_rxd);
312                 next_buffer = &rx_ring->buffer_info[i];
313
314                 if (!(staterr & E1000_RXD_STAT_EOP)) {
315                         buffer_info->skb = next_buffer->skb;
316                         buffer_info->dma = next_buffer->dma;
317                         next_buffer->skb = skb;
318                         next_buffer->dma = 0;
319                         goto next_desc;
320                 }
321
322                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
323                         dev_kfree_skb_irq(skb);
324                         goto next_desc;
325                 }
326
327                 total_bytes += skb->len;
328                 total_packets++;
329
330                 igbvf_rx_checksum_adv(adapter, staterr, skb);
331
332                 skb->protocol = eth_type_trans(skb, netdev);
333
334                 igbvf_receive_skb(adapter, netdev, skb, staterr,
335                                   rx_desc->wb.upper.vlan);
336
337 next_desc:
338                 rx_desc->wb.upper.status_error = 0;
339
340                 /* return some buffers to hardware, one at a time is too slow */
341                 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
342                         igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
343                         cleaned_count = 0;
344                 }
345
346                 /* use prefetched values */
347                 rx_desc = next_rxd;
348                 buffer_info = next_buffer;
349
350                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
351         }
352
353         rx_ring->next_to_clean = i;
354         cleaned_count = igbvf_desc_unused(rx_ring);
355
356         if (cleaned_count)
357                 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
358
359         adapter->total_rx_packets += total_packets;
360         adapter->total_rx_bytes += total_bytes;
361         adapter->net_stats.rx_bytes += total_bytes;
362         adapter->net_stats.rx_packets += total_packets;
363         return cleaned;
364 }
365
366 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
367                             struct igbvf_buffer *buffer_info)
368 {
369         if (buffer_info->dma) {
370                 if (buffer_info->mapped_as_page)
371                         pci_unmap_page(adapter->pdev,
372                                        buffer_info->dma,
373                                        buffer_info->length,
374                                        PCI_DMA_TODEVICE);
375                 else
376                         pci_unmap_single(adapter->pdev,
377                                          buffer_info->dma,
378                                          buffer_info->length,
379                                          PCI_DMA_TODEVICE);
380                 buffer_info->dma = 0;
381         }
382         if (buffer_info->skb) {
383                 dev_kfree_skb_any(buffer_info->skb);
384                 buffer_info->skb = NULL;
385         }
386         buffer_info->time_stamp = 0;
387 }
388
389 static void igbvf_print_tx_hang(struct igbvf_adapter *adapter)
390 {
391         struct igbvf_ring *tx_ring = adapter->tx_ring;
392         unsigned int i = tx_ring->next_to_clean;
393         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
394         union e1000_adv_tx_desc *eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
395
396         /* detected Tx unit hang */
397         dev_err(&adapter->pdev->dev,
398                 "Detected Tx Unit Hang:\n"
399                 "  TDH                  <%x>\n"
400                 "  TDT                  <%x>\n"
401                 "  next_to_use          <%x>\n"
402                 "  next_to_clean        <%x>\n"
403                 "buffer_info[next_to_clean]:\n"
404                 "  time_stamp           <%lx>\n"
405                 "  next_to_watch        <%x>\n"
406                 "  jiffies              <%lx>\n"
407                 "  next_to_watch.status <%x>\n",
408                 readl(adapter->hw.hw_addr + tx_ring->head),
409                 readl(adapter->hw.hw_addr + tx_ring->tail),
410                 tx_ring->next_to_use,
411                 tx_ring->next_to_clean,
412                 tx_ring->buffer_info[eop].time_stamp,
413                 eop,
414                 jiffies,
415                 eop_desc->wb.status);
416 }
417
418 /**
419  * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
420  * @adapter: board private structure
421  *
422  * Return 0 on success, negative on failure
423  **/
424 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
425                              struct igbvf_ring *tx_ring)
426 {
427         struct pci_dev *pdev = adapter->pdev;
428         int size;
429
430         size = sizeof(struct igbvf_buffer) * tx_ring->count;
431         tx_ring->buffer_info = vmalloc(size);
432         if (!tx_ring->buffer_info)
433                 goto err;
434         memset(tx_ring->buffer_info, 0, size);
435
436         /* round up to nearest 4K */
437         tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
438         tx_ring->size = ALIGN(tx_ring->size, 4096);
439
440         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
441                                              &tx_ring->dma);
442
443         if (!tx_ring->desc)
444                 goto err;
445
446         tx_ring->adapter = adapter;
447         tx_ring->next_to_use = 0;
448         tx_ring->next_to_clean = 0;
449
450         return 0;
451 err:
452         vfree(tx_ring->buffer_info);
453         dev_err(&adapter->pdev->dev,
454                 "Unable to allocate memory for the transmit descriptor ring\n");
455         return -ENOMEM;
456 }
457
458 /**
459  * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
460  * @adapter: board private structure
461  *
462  * Returns 0 on success, negative on failure
463  **/
464 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
465                              struct igbvf_ring *rx_ring)
466 {
467         struct pci_dev *pdev = adapter->pdev;
468         int size, desc_len;
469
470         size = sizeof(struct igbvf_buffer) * rx_ring->count;
471         rx_ring->buffer_info = vmalloc(size);
472         if (!rx_ring->buffer_info)
473                 goto err;
474         memset(rx_ring->buffer_info, 0, size);
475
476         desc_len = sizeof(union e1000_adv_rx_desc);
477
478         /* Round up to nearest 4K */
479         rx_ring->size = rx_ring->count * desc_len;
480         rx_ring->size = ALIGN(rx_ring->size, 4096);
481
482         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
483                                              &rx_ring->dma);
484
485         if (!rx_ring->desc)
486                 goto err;
487
488         rx_ring->next_to_clean = 0;
489         rx_ring->next_to_use = 0;
490
491         rx_ring->adapter = adapter;
492
493         return 0;
494
495 err:
496         vfree(rx_ring->buffer_info);
497         rx_ring->buffer_info = NULL;
498         dev_err(&adapter->pdev->dev,
499                 "Unable to allocate memory for the receive descriptor ring\n");
500         return -ENOMEM;
501 }
502
503 /**
504  * igbvf_clean_tx_ring - Free Tx Buffers
505  * @tx_ring: ring to be cleaned
506  **/
507 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
508 {
509         struct igbvf_adapter *adapter = tx_ring->adapter;
510         struct igbvf_buffer *buffer_info;
511         unsigned long size;
512         unsigned int i;
513
514         if (!tx_ring->buffer_info)
515                 return;
516
517         /* Free all the Tx ring sk_buffs */
518         for (i = 0; i < tx_ring->count; i++) {
519                 buffer_info = &tx_ring->buffer_info[i];
520                 igbvf_put_txbuf(adapter, buffer_info);
521         }
522
523         size = sizeof(struct igbvf_buffer) * tx_ring->count;
524         memset(tx_ring->buffer_info, 0, size);
525
526         /* Zero out the descriptor ring */
527         memset(tx_ring->desc, 0, tx_ring->size);
528
529         tx_ring->next_to_use = 0;
530         tx_ring->next_to_clean = 0;
531
532         writel(0, adapter->hw.hw_addr + tx_ring->head);
533         writel(0, adapter->hw.hw_addr + tx_ring->tail);
534 }
535
536 /**
537  * igbvf_free_tx_resources - Free Tx Resources per Queue
538  * @tx_ring: ring to free resources from
539  *
540  * Free all transmit software resources
541  **/
542 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
543 {
544         struct pci_dev *pdev = tx_ring->adapter->pdev;
545
546         igbvf_clean_tx_ring(tx_ring);
547
548         vfree(tx_ring->buffer_info);
549         tx_ring->buffer_info = NULL;
550
551         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
552
553         tx_ring->desc = NULL;
554 }
555
556 /**
557  * igbvf_clean_rx_ring - Free Rx Buffers per Queue
558  * @adapter: board private structure
559  **/
560 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
561 {
562         struct igbvf_adapter *adapter = rx_ring->adapter;
563         struct igbvf_buffer *buffer_info;
564         struct pci_dev *pdev = adapter->pdev;
565         unsigned long size;
566         unsigned int i;
567
568         if (!rx_ring->buffer_info)
569                 return;
570
571         /* Free all the Rx ring sk_buffs */
572         for (i = 0; i < rx_ring->count; i++) {
573                 buffer_info = &rx_ring->buffer_info[i];
574                 if (buffer_info->dma) {
575                         if (adapter->rx_ps_hdr_size){
576                                 pci_unmap_single(pdev, buffer_info->dma,
577                                                  adapter->rx_ps_hdr_size,
578                                                  PCI_DMA_FROMDEVICE);
579                         } else {
580                                 pci_unmap_single(pdev, buffer_info->dma,
581                                                  adapter->rx_buffer_len,
582                                                  PCI_DMA_FROMDEVICE);
583                         }
584                         buffer_info->dma = 0;
585                 }
586
587                 if (buffer_info->skb) {
588                         dev_kfree_skb(buffer_info->skb);
589                         buffer_info->skb = NULL;
590                 }
591
592                 if (buffer_info->page) {
593                         if (buffer_info->page_dma)
594                                 pci_unmap_page(pdev, buffer_info->page_dma,
595                                                PAGE_SIZE / 2,
596                                                PCI_DMA_FROMDEVICE);
597                         put_page(buffer_info->page);
598                         buffer_info->page = NULL;
599                         buffer_info->page_dma = 0;
600                         buffer_info->page_offset = 0;
601                 }
602         }
603
604         size = sizeof(struct igbvf_buffer) * rx_ring->count;
605         memset(rx_ring->buffer_info, 0, size);
606
607         /* Zero out the descriptor ring */
608         memset(rx_ring->desc, 0, rx_ring->size);
609
610         rx_ring->next_to_clean = 0;
611         rx_ring->next_to_use = 0;
612
613         writel(0, adapter->hw.hw_addr + rx_ring->head);
614         writel(0, adapter->hw.hw_addr + rx_ring->tail);
615 }
616
617 /**
618  * igbvf_free_rx_resources - Free Rx Resources
619  * @rx_ring: ring to clean the resources from
620  *
621  * Free all receive software resources
622  **/
623
624 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
625 {
626         struct pci_dev *pdev = rx_ring->adapter->pdev;
627
628         igbvf_clean_rx_ring(rx_ring);
629
630         vfree(rx_ring->buffer_info);
631         rx_ring->buffer_info = NULL;
632
633         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
634                           rx_ring->dma);
635         rx_ring->desc = NULL;
636 }
637
638 /**
639  * igbvf_update_itr - update the dynamic ITR value based on statistics
640  * @adapter: pointer to adapter
641  * @itr_setting: current adapter->itr
642  * @packets: the number of packets during this measurement interval
643  * @bytes: the number of bytes during this measurement interval
644  *
645  *      Stores a new ITR value based on packets and byte
646  *      counts during the last interrupt.  The advantage of per interrupt
647  *      computation is faster updates and more accurate ITR for the current
648  *      traffic pattern.  Constants in this function were computed
649  *      based on theoretical maximum wire speed and thresholds were set based
650  *      on testing data as well as attempting to minimize response time
651  *      while increasing bulk throughput.  This functionality is controlled
652  *      by the InterruptThrottleRate module parameter.
653  **/
654 static unsigned int igbvf_update_itr(struct igbvf_adapter *adapter,
655                                      u16 itr_setting, int packets,
656                                      int bytes)
657 {
658         unsigned int retval = itr_setting;
659
660         if (packets == 0)
661                 goto update_itr_done;
662
663         switch (itr_setting) {
664         case lowest_latency:
665                 /* handle TSO and jumbo frames */
666                 if (bytes/packets > 8000)
667                         retval = bulk_latency;
668                 else if ((packets < 5) && (bytes > 512))
669                         retval = low_latency;
670                 break;
671         case low_latency:  /* 50 usec aka 20000 ints/s */
672                 if (bytes > 10000) {
673                         /* this if handles the TSO accounting */
674                         if (bytes/packets > 8000)
675                                 retval = bulk_latency;
676                         else if ((packets < 10) || ((bytes/packets) > 1200))
677                                 retval = bulk_latency;
678                         else if ((packets > 35))
679                                 retval = lowest_latency;
680                 } else if (bytes/packets > 2000) {
681                         retval = bulk_latency;
682                 } else if (packets <= 2 && bytes < 512) {
683                         retval = lowest_latency;
684                 }
685                 break;
686         case bulk_latency: /* 250 usec aka 4000 ints/s */
687                 if (bytes > 25000) {
688                         if (packets > 35)
689                                 retval = low_latency;
690                 } else if (bytes < 6000) {
691                         retval = low_latency;
692                 }
693                 break;
694         }
695
696 update_itr_done:
697         return retval;
698 }
699
700 static void igbvf_set_itr(struct igbvf_adapter *adapter)
701 {
702         struct e1000_hw *hw = &adapter->hw;
703         u16 current_itr;
704         u32 new_itr = adapter->itr;
705
706         adapter->tx_itr = igbvf_update_itr(adapter, adapter->tx_itr,
707                                            adapter->total_tx_packets,
708                                            adapter->total_tx_bytes);
709         /* conservative mode (itr 3) eliminates the lowest_latency setting */
710         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
711                 adapter->tx_itr = low_latency;
712
713         adapter->rx_itr = igbvf_update_itr(adapter, adapter->rx_itr,
714                                            adapter->total_rx_packets,
715                                            adapter->total_rx_bytes);
716         /* conservative mode (itr 3) eliminates the lowest_latency setting */
717         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
718                 adapter->rx_itr = low_latency;
719
720         current_itr = max(adapter->rx_itr, adapter->tx_itr);
721
722         switch (current_itr) {
723         /* counts and packets in update_itr are dependent on these numbers */
724         case lowest_latency:
725                 new_itr = 70000;
726                 break;
727         case low_latency:
728                 new_itr = 20000; /* aka hwitr = ~200 */
729                 break;
730         case bulk_latency:
731                 new_itr = 4000;
732                 break;
733         default:
734                 break;
735         }
736
737         if (new_itr != adapter->itr) {
738                 /*
739                  * this attempts to bias the interrupt rate towards Bulk
740                  * by adding intermediate steps when interrupt rate is
741                  * increasing
742                  */
743                 new_itr = new_itr > adapter->itr ?
744                              min(adapter->itr + (new_itr >> 2), new_itr) :
745                              new_itr;
746                 adapter->itr = new_itr;
747                 adapter->rx_ring->itr_val = 1952;
748
749                 if (adapter->msix_entries)
750                         adapter->rx_ring->set_itr = 1;
751                 else
752                         ew32(ITR, 1952);
753         }
754 }
755
756 /**
757  * igbvf_clean_tx_irq - Reclaim resources after transmit completes
758  * @adapter: board private structure
759  * returns true if ring is completely cleaned
760  **/
761 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
762 {
763         struct igbvf_adapter *adapter = tx_ring->adapter;
764         struct e1000_hw *hw = &adapter->hw;
765         struct net_device *netdev = adapter->netdev;
766         struct igbvf_buffer *buffer_info;
767         struct sk_buff *skb;
768         union e1000_adv_tx_desc *tx_desc, *eop_desc;
769         unsigned int total_bytes = 0, total_packets = 0;
770         unsigned int i, eop, count = 0;
771         bool cleaned = false;
772
773         i = tx_ring->next_to_clean;
774         eop = tx_ring->buffer_info[i].next_to_watch;
775         eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
776
777         while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
778                (count < tx_ring->count)) {
779                 for (cleaned = false; !cleaned; count++) {
780                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
781                         buffer_info = &tx_ring->buffer_info[i];
782                         cleaned = (i == eop);
783                         skb = buffer_info->skb;
784
785                         if (skb) {
786                                 unsigned int segs, bytecount;
787
788                                 /* gso_segs is currently only valid for tcp */
789                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
790                                 /* multiply data chunks by size of headers */
791                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
792                                             skb->len;
793                                 total_packets += segs;
794                                 total_bytes += bytecount;
795                         }
796
797                         igbvf_put_txbuf(adapter, buffer_info);
798                         tx_desc->wb.status = 0;
799
800                         i++;
801                         if (i == tx_ring->count)
802                                 i = 0;
803                 }
804                 eop = tx_ring->buffer_info[i].next_to_watch;
805                 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
806         }
807
808         tx_ring->next_to_clean = i;
809
810         if (unlikely(count &&
811                      netif_carrier_ok(netdev) &&
812                      igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
813                 /* Make sure that anybody stopping the queue after this
814                  * sees the new next_to_clean.
815                  */
816                 smp_mb();
817                 if (netif_queue_stopped(netdev) &&
818                     !(test_bit(__IGBVF_DOWN, &adapter->state))) {
819                         netif_wake_queue(netdev);
820                         ++adapter->restart_queue;
821                 }
822         }
823
824         if (adapter->detect_tx_hung) {
825                 /* Detect a transmit hang in hardware, this serializes the
826                  * check with the clearing of time_stamp and movement of i */
827                 adapter->detect_tx_hung = false;
828                 if (tx_ring->buffer_info[i].time_stamp &&
829                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
830                                (adapter->tx_timeout_factor * HZ)) &&
831                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
832
833                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
834                         /* detected Tx unit hang */
835                         igbvf_print_tx_hang(adapter);
836
837                         netif_stop_queue(netdev);
838                 }
839         }
840         adapter->net_stats.tx_bytes += total_bytes;
841         adapter->net_stats.tx_packets += total_packets;
842         return (count < tx_ring->count);
843 }
844
845 static irqreturn_t igbvf_msix_other(int irq, void *data)
846 {
847         struct net_device *netdev = data;
848         struct igbvf_adapter *adapter = netdev_priv(netdev);
849         struct e1000_hw *hw = &adapter->hw;
850
851         adapter->int_counter1++;
852
853         netif_carrier_off(netdev);
854         hw->mac.get_link_status = 1;
855         if (!test_bit(__IGBVF_DOWN, &adapter->state))
856                 mod_timer(&adapter->watchdog_timer, jiffies + 1);
857
858         ew32(EIMS, adapter->eims_other);
859
860         return IRQ_HANDLED;
861 }
862
863 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
864 {
865         struct net_device *netdev = data;
866         struct igbvf_adapter *adapter = netdev_priv(netdev);
867         struct e1000_hw *hw = &adapter->hw;
868         struct igbvf_ring *tx_ring = adapter->tx_ring;
869
870
871         adapter->total_tx_bytes = 0;
872         adapter->total_tx_packets = 0;
873
874         /* auto mask will automatically reenable the interrupt when we write
875          * EICS */
876         if (!igbvf_clean_tx_irq(tx_ring))
877                 /* Ring was not completely cleaned, so fire another interrupt */
878                 ew32(EICS, tx_ring->eims_value);
879         else
880                 ew32(EIMS, tx_ring->eims_value);
881
882         return IRQ_HANDLED;
883 }
884
885 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
886 {
887         struct net_device *netdev = data;
888         struct igbvf_adapter *adapter = netdev_priv(netdev);
889
890         adapter->int_counter0++;
891
892         /* Write the ITR value calculated at the end of the
893          * previous interrupt.
894          */
895         if (adapter->rx_ring->set_itr) {
896                 writel(adapter->rx_ring->itr_val,
897                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
898                 adapter->rx_ring->set_itr = 0;
899         }
900
901         if (napi_schedule_prep(&adapter->rx_ring->napi)) {
902                 adapter->total_rx_bytes = 0;
903                 adapter->total_rx_packets = 0;
904                 __napi_schedule(&adapter->rx_ring->napi);
905         }
906
907         return IRQ_HANDLED;
908 }
909
910 #define IGBVF_NO_QUEUE -1
911
912 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
913                                 int tx_queue, int msix_vector)
914 {
915         struct e1000_hw *hw = &adapter->hw;
916         u32 ivar, index;
917
918         /* 82576 uses a table-based method for assigning vectors.
919            Each queue has a single entry in the table to which we write
920            a vector number along with a "valid" bit.  Sadly, the layout
921            of the table is somewhat counterintuitive. */
922         if (rx_queue > IGBVF_NO_QUEUE) {
923                 index = (rx_queue >> 1);
924                 ivar = array_er32(IVAR0, index);
925                 if (rx_queue & 0x1) {
926                         /* vector goes into third byte of register */
927                         ivar = ivar & 0xFF00FFFF;
928                         ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
929                 } else {
930                         /* vector goes into low byte of register */
931                         ivar = ivar & 0xFFFFFF00;
932                         ivar |= msix_vector | E1000_IVAR_VALID;
933                 }
934                 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
935                 array_ew32(IVAR0, index, ivar);
936         }
937         if (tx_queue > IGBVF_NO_QUEUE) {
938                 index = (tx_queue >> 1);
939                 ivar = array_er32(IVAR0, index);
940                 if (tx_queue & 0x1) {
941                         /* vector goes into high byte of register */
942                         ivar = ivar & 0x00FFFFFF;
943                         ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
944                 } else {
945                         /* vector goes into second byte of register */
946                         ivar = ivar & 0xFFFF00FF;
947                         ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
948                 }
949                 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
950                 array_ew32(IVAR0, index, ivar);
951         }
952 }
953
954 /**
955  * igbvf_configure_msix - Configure MSI-X hardware
956  *
957  * igbvf_configure_msix sets up the hardware to properly
958  * generate MSI-X interrupts.
959  **/
960 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
961 {
962         u32 tmp;
963         struct e1000_hw *hw = &adapter->hw;
964         struct igbvf_ring *tx_ring = adapter->tx_ring;
965         struct igbvf_ring *rx_ring = adapter->rx_ring;
966         int vector = 0;
967
968         adapter->eims_enable_mask = 0;
969
970         igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
971         adapter->eims_enable_mask |= tx_ring->eims_value;
972         if (tx_ring->itr_val)
973                 writel(tx_ring->itr_val,
974                        hw->hw_addr + tx_ring->itr_register);
975         else
976                 writel(1952, hw->hw_addr + tx_ring->itr_register);
977
978         igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
979         adapter->eims_enable_mask |= rx_ring->eims_value;
980         if (rx_ring->itr_val)
981                 writel(rx_ring->itr_val,
982                        hw->hw_addr + rx_ring->itr_register);
983         else
984                 writel(1952, hw->hw_addr + rx_ring->itr_register);
985
986         /* set vector for other causes, i.e. link changes */
987
988         tmp = (vector++ | E1000_IVAR_VALID);
989
990         ew32(IVAR_MISC, tmp);
991
992         adapter->eims_enable_mask = (1 << (vector)) - 1;
993         adapter->eims_other = 1 << (vector - 1);
994         e1e_flush();
995 }
996
997 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
998 {
999         if (adapter->msix_entries) {
1000                 pci_disable_msix(adapter->pdev);
1001                 kfree(adapter->msix_entries);
1002                 adapter->msix_entries = NULL;
1003         }
1004 }
1005
1006 /**
1007  * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1008  *
1009  * Attempt to configure interrupts using the best available
1010  * capabilities of the hardware and kernel.
1011  **/
1012 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1013 {
1014         int err = -ENOMEM;
1015         int i;
1016
1017         /* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
1018         adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1019                                         GFP_KERNEL);
1020         if (adapter->msix_entries) {
1021                 for (i = 0; i < 3; i++)
1022                         adapter->msix_entries[i].entry = i;
1023
1024                 err = pci_enable_msix(adapter->pdev,
1025                                       adapter->msix_entries, 3);
1026         }
1027
1028         if (err) {
1029                 /* MSI-X failed */
1030                 dev_err(&adapter->pdev->dev,
1031                         "Failed to initialize MSI-X interrupts.\n");
1032                 igbvf_reset_interrupt_capability(adapter);
1033         }
1034 }
1035
1036 /**
1037  * igbvf_request_msix - Initialize MSI-X interrupts
1038  *
1039  * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1040  * kernel.
1041  **/
1042 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1043 {
1044         struct net_device *netdev = adapter->netdev;
1045         int err = 0, vector = 0;
1046
1047         if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1048                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1049                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1050         } else {
1051                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1052                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1053         }
1054
1055         err = request_irq(adapter->msix_entries[vector].vector,
1056                           igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1057                           netdev);
1058         if (err)
1059                 goto out;
1060
1061         adapter->tx_ring->itr_register = E1000_EITR(vector);
1062         adapter->tx_ring->itr_val = 1952;
1063         vector++;
1064
1065         err = request_irq(adapter->msix_entries[vector].vector,
1066                           igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1067                           netdev);
1068         if (err)
1069                 goto out;
1070
1071         adapter->rx_ring->itr_register = E1000_EITR(vector);
1072         adapter->rx_ring->itr_val = 1952;
1073         vector++;
1074
1075         err = request_irq(adapter->msix_entries[vector].vector,
1076                           igbvf_msix_other, 0, netdev->name, netdev);
1077         if (err)
1078                 goto out;
1079
1080         igbvf_configure_msix(adapter);
1081         return 0;
1082 out:
1083         return err;
1084 }
1085
1086 /**
1087  * igbvf_alloc_queues - Allocate memory for all rings
1088  * @adapter: board private structure to initialize
1089  **/
1090 static int __devinit igbvf_alloc_queues(struct igbvf_adapter *adapter)
1091 {
1092         struct net_device *netdev = adapter->netdev;
1093
1094         adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1095         if (!adapter->tx_ring)
1096                 return -ENOMEM;
1097
1098         adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1099         if (!adapter->rx_ring) {
1100                 kfree(adapter->tx_ring);
1101                 return -ENOMEM;
1102         }
1103
1104         netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1105
1106         return 0;
1107 }
1108
1109 /**
1110  * igbvf_request_irq - initialize interrupts
1111  *
1112  * Attempts to configure interrupts using the best available
1113  * capabilities of the hardware and kernel.
1114  **/
1115 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1116 {
1117         int err = -1;
1118
1119         /* igbvf supports msi-x only */
1120         if (adapter->msix_entries)
1121                 err = igbvf_request_msix(adapter);
1122
1123         if (!err)
1124                 return err;
1125
1126         dev_err(&adapter->pdev->dev,
1127                 "Unable to allocate interrupt, Error: %d\n", err);
1128
1129         return err;
1130 }
1131
1132 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1133 {
1134         struct net_device *netdev = adapter->netdev;
1135         int vector;
1136
1137         if (adapter->msix_entries) {
1138                 for (vector = 0; vector < 3; vector++)
1139                         free_irq(adapter->msix_entries[vector].vector, netdev);
1140         }
1141 }
1142
1143 /**
1144  * igbvf_irq_disable - Mask off interrupt generation on the NIC
1145  **/
1146 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1147 {
1148         struct e1000_hw *hw = &adapter->hw;
1149
1150         ew32(EIMC, ~0);
1151
1152         if (adapter->msix_entries)
1153                 ew32(EIAC, 0);
1154 }
1155
1156 /**
1157  * igbvf_irq_enable - Enable default interrupt generation settings
1158  **/
1159 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1160 {
1161         struct e1000_hw *hw = &adapter->hw;
1162
1163         ew32(EIAC, adapter->eims_enable_mask);
1164         ew32(EIAM, adapter->eims_enable_mask);
1165         ew32(EIMS, adapter->eims_enable_mask);
1166 }
1167
1168 /**
1169  * igbvf_poll - NAPI Rx polling callback
1170  * @napi: struct associated with this polling callback
1171  * @budget: amount of packets driver is allowed to process this poll
1172  **/
1173 static int igbvf_poll(struct napi_struct *napi, int budget)
1174 {
1175         struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1176         struct igbvf_adapter *adapter = rx_ring->adapter;
1177         struct e1000_hw *hw = &adapter->hw;
1178         int work_done = 0;
1179
1180         igbvf_clean_rx_irq(adapter, &work_done, budget);
1181
1182         /* If not enough Rx work done, exit the polling mode */
1183         if (work_done < budget) {
1184                 napi_complete(napi);
1185
1186                 if (adapter->itr_setting & 3)
1187                         igbvf_set_itr(adapter);
1188
1189                 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1190                         ew32(EIMS, adapter->rx_ring->eims_value);
1191         }
1192
1193         return work_done;
1194 }
1195
1196 /**
1197  * igbvf_set_rlpml - set receive large packet maximum length
1198  * @adapter: board private structure
1199  *
1200  * Configure the maximum size of packets that will be received
1201  */
1202 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1203 {
1204         int max_frame_size = adapter->max_frame_size;
1205         struct e1000_hw *hw = &adapter->hw;
1206
1207         if (adapter->vlgrp)
1208                 max_frame_size += VLAN_TAG_SIZE;
1209
1210         e1000_rlpml_set_vf(hw, max_frame_size);
1211 }
1212
1213 static void igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1214 {
1215         struct igbvf_adapter *adapter = netdev_priv(netdev);
1216         struct e1000_hw *hw = &adapter->hw;
1217
1218         if (hw->mac.ops.set_vfta(hw, vid, true))
1219                 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1220 }
1221
1222 static void igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1223 {
1224         struct igbvf_adapter *adapter = netdev_priv(netdev);
1225         struct e1000_hw *hw = &adapter->hw;
1226
1227         igbvf_irq_disable(adapter);
1228         vlan_group_set_device(adapter->vlgrp, vid, NULL);
1229
1230         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1231                 igbvf_irq_enable(adapter);
1232
1233         if (hw->mac.ops.set_vfta(hw, vid, false))
1234                 dev_err(&adapter->pdev->dev,
1235                         "Failed to remove vlan id %d\n", vid);
1236 }
1237
1238 static void igbvf_vlan_rx_register(struct net_device *netdev,
1239                                    struct vlan_group *grp)
1240 {
1241         struct igbvf_adapter *adapter = netdev_priv(netdev);
1242
1243         adapter->vlgrp = grp;
1244 }
1245
1246 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1247 {
1248         u16 vid;
1249
1250         if (!adapter->vlgrp)
1251                 return;
1252
1253         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1254                 if (!vlan_group_get_device(adapter->vlgrp, vid))
1255                         continue;
1256                 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1257         }
1258
1259         igbvf_set_rlpml(adapter);
1260 }
1261
1262 /**
1263  * igbvf_configure_tx - Configure Transmit Unit after Reset
1264  * @adapter: board private structure
1265  *
1266  * Configure the Tx unit of the MAC after a reset.
1267  **/
1268 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1269 {
1270         struct e1000_hw *hw = &adapter->hw;
1271         struct igbvf_ring *tx_ring = adapter->tx_ring;
1272         u64 tdba;
1273         u32 txdctl, dca_txctrl;
1274
1275         /* disable transmits */
1276         txdctl = er32(TXDCTL(0));
1277         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1278         msleep(10);
1279
1280         /* Setup the HW Tx Head and Tail descriptor pointers */
1281         ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1282         tdba = tx_ring->dma;
1283         ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1284         ew32(TDBAH(0), (tdba >> 32));
1285         ew32(TDH(0), 0);
1286         ew32(TDT(0), 0);
1287         tx_ring->head = E1000_TDH(0);
1288         tx_ring->tail = E1000_TDT(0);
1289
1290         /* Turn off Relaxed Ordering on head write-backs.  The writebacks
1291          * MUST be delivered in order or it will completely screw up
1292          * our bookeeping.
1293          */
1294         dca_txctrl = er32(DCA_TXCTRL(0));
1295         dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1296         ew32(DCA_TXCTRL(0), dca_txctrl);
1297
1298         /* enable transmits */
1299         txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1300         ew32(TXDCTL(0), txdctl);
1301
1302         /* Setup Transmit Descriptor Settings for eop descriptor */
1303         adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1304
1305         /* enable Report Status bit */
1306         adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1307 }
1308
1309 /**
1310  * igbvf_setup_srrctl - configure the receive control registers
1311  * @adapter: Board private structure
1312  **/
1313 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1314 {
1315         struct e1000_hw *hw = &adapter->hw;
1316         u32 srrctl = 0;
1317
1318         srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1319                     E1000_SRRCTL_BSIZEHDR_MASK |
1320                     E1000_SRRCTL_BSIZEPKT_MASK);
1321
1322         /* Enable queue drop to avoid head of line blocking */
1323         srrctl |= E1000_SRRCTL_DROP_EN;
1324
1325         /* Setup buffer sizes */
1326         srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1327                   E1000_SRRCTL_BSIZEPKT_SHIFT;
1328
1329         if (adapter->rx_buffer_len < 2048) {
1330                 adapter->rx_ps_hdr_size = 0;
1331                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1332         } else {
1333                 adapter->rx_ps_hdr_size = 128;
1334                 srrctl |= adapter->rx_ps_hdr_size <<
1335                           E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1336                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1337         }
1338
1339         ew32(SRRCTL(0), srrctl);
1340 }
1341
1342 /**
1343  * igbvf_configure_rx - Configure Receive Unit after Reset
1344  * @adapter: board private structure
1345  *
1346  * Configure the Rx unit of the MAC after a reset.
1347  **/
1348 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1349 {
1350         struct e1000_hw *hw = &adapter->hw;
1351         struct igbvf_ring *rx_ring = adapter->rx_ring;
1352         u64 rdba;
1353         u32 rdlen, rxdctl;
1354
1355         /* disable receives */
1356         rxdctl = er32(RXDCTL(0));
1357         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1358         msleep(10);
1359
1360         rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1361
1362         /*
1363          * Setup the HW Rx Head and Tail Descriptor Pointers and
1364          * the Base and Length of the Rx Descriptor Ring
1365          */
1366         rdba = rx_ring->dma;
1367         ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1368         ew32(RDBAH(0), (rdba >> 32));
1369         ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1370         rx_ring->head = E1000_RDH(0);
1371         rx_ring->tail = E1000_RDT(0);
1372         ew32(RDH(0), 0);
1373         ew32(RDT(0), 0);
1374
1375         rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1376         rxdctl &= 0xFFF00000;
1377         rxdctl |= IGBVF_RX_PTHRESH;
1378         rxdctl |= IGBVF_RX_HTHRESH << 8;
1379         rxdctl |= IGBVF_RX_WTHRESH << 16;
1380
1381         igbvf_set_rlpml(adapter);
1382
1383         /* enable receives */
1384         ew32(RXDCTL(0), rxdctl);
1385 }
1386
1387 /**
1388  * igbvf_set_multi - Multicast and Promiscuous mode set
1389  * @netdev: network interface device structure
1390  *
1391  * The set_multi entry point is called whenever the multicast address
1392  * list or the network interface flags are updated.  This routine is
1393  * responsible for configuring the hardware for proper multicast,
1394  * promiscuous mode, and all-multi behavior.
1395  **/
1396 static void igbvf_set_multi(struct net_device *netdev)
1397 {
1398         struct igbvf_adapter *adapter = netdev_priv(netdev);
1399         struct e1000_hw *hw = &adapter->hw;
1400         struct dev_mc_list *mc_ptr;
1401         u8  *mta_list = NULL;
1402         int i;
1403
1404         if (!netdev_mc_empty(netdev)) {
1405                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
1406                 if (!mta_list) {
1407                         dev_err(&adapter->pdev->dev,
1408                                 "failed to allocate multicast filter list\n");
1409                         return;
1410                 }
1411         }
1412
1413         /* prepare a packed array of only addresses. */
1414         i = 0;
1415         netdev_for_each_mc_addr(mc_ptr, netdev)
1416                 memcpy(mta_list + (i++ * ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
1417
1418         hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1419         kfree(mta_list);
1420 }
1421
1422 /**
1423  * igbvf_configure - configure the hardware for Rx and Tx
1424  * @adapter: private board structure
1425  **/
1426 static void igbvf_configure(struct igbvf_adapter *adapter)
1427 {
1428         igbvf_set_multi(adapter->netdev);
1429
1430         igbvf_restore_vlan(adapter);
1431
1432         igbvf_configure_tx(adapter);
1433         igbvf_setup_srrctl(adapter);
1434         igbvf_configure_rx(adapter);
1435         igbvf_alloc_rx_buffers(adapter->rx_ring,
1436                                igbvf_desc_unused(adapter->rx_ring));
1437 }
1438
1439 /* igbvf_reset - bring the hardware into a known good state
1440  *
1441  * This function boots the hardware and enables some settings that
1442  * require a configuration cycle of the hardware - those cannot be
1443  * set/changed during runtime. After reset the device needs to be
1444  * properly configured for Rx, Tx etc.
1445  */
1446 static void igbvf_reset(struct igbvf_adapter *adapter)
1447 {
1448         struct e1000_mac_info *mac = &adapter->hw.mac;
1449         struct net_device *netdev = adapter->netdev;
1450         struct e1000_hw *hw = &adapter->hw;
1451
1452         /* Allow time for pending master requests to run */
1453         if (mac->ops.reset_hw(hw))
1454                 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1455
1456         mac->ops.init_hw(hw);
1457
1458         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1459                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1460                        netdev->addr_len);
1461                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1462                        netdev->addr_len);
1463         }
1464
1465         adapter->last_reset = jiffies;
1466 }
1467
1468 int igbvf_up(struct igbvf_adapter *adapter)
1469 {
1470         struct e1000_hw *hw = &adapter->hw;
1471
1472         /* hardware has been reset, we need to reload some things */
1473         igbvf_configure(adapter);
1474
1475         clear_bit(__IGBVF_DOWN, &adapter->state);
1476
1477         napi_enable(&adapter->rx_ring->napi);
1478         if (adapter->msix_entries)
1479                 igbvf_configure_msix(adapter);
1480
1481         /* Clear any pending interrupts. */
1482         er32(EICR);
1483         igbvf_irq_enable(adapter);
1484
1485         /* start the watchdog */
1486         hw->mac.get_link_status = 1;
1487         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1488
1489
1490         return 0;
1491 }
1492
1493 void igbvf_down(struct igbvf_adapter *adapter)
1494 {
1495         struct net_device *netdev = adapter->netdev;
1496         struct e1000_hw *hw = &adapter->hw;
1497         u32 rxdctl, txdctl;
1498
1499         /*
1500          * signal that we're down so the interrupt handler does not
1501          * reschedule our watchdog timer
1502          */
1503         set_bit(__IGBVF_DOWN, &adapter->state);
1504
1505         /* disable receives in the hardware */
1506         rxdctl = er32(RXDCTL(0));
1507         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1508
1509         netif_stop_queue(netdev);
1510
1511         /* disable transmits in the hardware */
1512         txdctl = er32(TXDCTL(0));
1513         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1514
1515         /* flush both disables and wait for them to finish */
1516         e1e_flush();
1517         msleep(10);
1518
1519         napi_disable(&adapter->rx_ring->napi);
1520
1521         igbvf_irq_disable(adapter);
1522
1523         del_timer_sync(&adapter->watchdog_timer);
1524
1525         netif_carrier_off(netdev);
1526
1527         /* record the stats before reset*/
1528         igbvf_update_stats(adapter);
1529
1530         adapter->link_speed = 0;
1531         adapter->link_duplex = 0;
1532
1533         igbvf_reset(adapter);
1534         igbvf_clean_tx_ring(adapter->tx_ring);
1535         igbvf_clean_rx_ring(adapter->rx_ring);
1536 }
1537
1538 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1539 {
1540         might_sleep();
1541         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1542                 msleep(1);
1543         igbvf_down(adapter);
1544         igbvf_up(adapter);
1545         clear_bit(__IGBVF_RESETTING, &adapter->state);
1546 }
1547
1548 /**
1549  * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1550  * @adapter: board private structure to initialize
1551  *
1552  * igbvf_sw_init initializes the Adapter private data structure.
1553  * Fields are initialized based on PCI device information and
1554  * OS network device settings (MTU size).
1555  **/
1556 static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1557 {
1558         struct net_device *netdev = adapter->netdev;
1559         s32 rc;
1560
1561         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1562         adapter->rx_ps_hdr_size = 0;
1563         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1564         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1565
1566         adapter->tx_int_delay = 8;
1567         adapter->tx_abs_int_delay = 32;
1568         adapter->rx_int_delay = 0;
1569         adapter->rx_abs_int_delay = 8;
1570         adapter->itr_setting = 3;
1571         adapter->itr = 20000;
1572
1573         /* Set various function pointers */
1574         adapter->ei->init_ops(&adapter->hw);
1575
1576         rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1577         if (rc)
1578                 return rc;
1579
1580         rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1581         if (rc)
1582                 return rc;
1583
1584         igbvf_set_interrupt_capability(adapter);
1585
1586         if (igbvf_alloc_queues(adapter))
1587                 return -ENOMEM;
1588
1589         spin_lock_init(&adapter->tx_queue_lock);
1590
1591         /* Explicitly disable IRQ since the NIC can be in any state. */
1592         igbvf_irq_disable(adapter);
1593
1594         spin_lock_init(&adapter->stats_lock);
1595
1596         set_bit(__IGBVF_DOWN, &adapter->state);
1597         return 0;
1598 }
1599
1600 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1601 {
1602         struct e1000_hw *hw = &adapter->hw;
1603
1604         adapter->stats.last_gprc = er32(VFGPRC);
1605         adapter->stats.last_gorc = er32(VFGORC);
1606         adapter->stats.last_gptc = er32(VFGPTC);
1607         adapter->stats.last_gotc = er32(VFGOTC);
1608         adapter->stats.last_mprc = er32(VFMPRC);
1609         adapter->stats.last_gotlbc = er32(VFGOTLBC);
1610         adapter->stats.last_gptlbc = er32(VFGPTLBC);
1611         adapter->stats.last_gorlbc = er32(VFGORLBC);
1612         adapter->stats.last_gprlbc = er32(VFGPRLBC);
1613
1614         adapter->stats.base_gprc = er32(VFGPRC);
1615         adapter->stats.base_gorc = er32(VFGORC);
1616         adapter->stats.base_gptc = er32(VFGPTC);
1617         adapter->stats.base_gotc = er32(VFGOTC);
1618         adapter->stats.base_mprc = er32(VFMPRC);
1619         adapter->stats.base_gotlbc = er32(VFGOTLBC);
1620         adapter->stats.base_gptlbc = er32(VFGPTLBC);
1621         adapter->stats.base_gorlbc = er32(VFGORLBC);
1622         adapter->stats.base_gprlbc = er32(VFGPRLBC);
1623 }
1624
1625 /**
1626  * igbvf_open - Called when a network interface is made active
1627  * @netdev: network interface device structure
1628  *
1629  * Returns 0 on success, negative value on failure
1630  *
1631  * The open entry point is called when a network interface is made
1632  * active by the system (IFF_UP).  At this point all resources needed
1633  * for transmit and receive operations are allocated, the interrupt
1634  * handler is registered with the OS, the watchdog timer is started,
1635  * and the stack is notified that the interface is ready.
1636  **/
1637 static int igbvf_open(struct net_device *netdev)
1638 {
1639         struct igbvf_adapter *adapter = netdev_priv(netdev);
1640         struct e1000_hw *hw = &adapter->hw;
1641         int err;
1642
1643         /* disallow open during test */
1644         if (test_bit(__IGBVF_TESTING, &adapter->state))
1645                 return -EBUSY;
1646
1647         /* allocate transmit descriptors */
1648         err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1649         if (err)
1650                 goto err_setup_tx;
1651
1652         /* allocate receive descriptors */
1653         err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1654         if (err)
1655                 goto err_setup_rx;
1656
1657         /*
1658          * before we allocate an interrupt, we must be ready to handle it.
1659          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1660          * as soon as we call pci_request_irq, so we have to setup our
1661          * clean_rx handler before we do so.
1662          */
1663         igbvf_configure(adapter);
1664
1665         err = igbvf_request_irq(adapter);
1666         if (err)
1667                 goto err_req_irq;
1668
1669         /* From here on the code is the same as igbvf_up() */
1670         clear_bit(__IGBVF_DOWN, &adapter->state);
1671
1672         napi_enable(&adapter->rx_ring->napi);
1673
1674         /* clear any pending interrupts */
1675         er32(EICR);
1676
1677         igbvf_irq_enable(adapter);
1678
1679         /* start the watchdog */
1680         hw->mac.get_link_status = 1;
1681         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1682
1683         return 0;
1684
1685 err_req_irq:
1686         igbvf_free_rx_resources(adapter->rx_ring);
1687 err_setup_rx:
1688         igbvf_free_tx_resources(adapter->tx_ring);
1689 err_setup_tx:
1690         igbvf_reset(adapter);
1691
1692         return err;
1693 }
1694
1695 /**
1696  * igbvf_close - Disables a network interface
1697  * @netdev: network interface device structure
1698  *
1699  * Returns 0, this is not allowed to fail
1700  *
1701  * The close entry point is called when an interface is de-activated
1702  * by the OS.  The hardware is still under the drivers control, but
1703  * needs to be disabled.  A global MAC reset is issued to stop the
1704  * hardware, and all transmit and receive resources are freed.
1705  **/
1706 static int igbvf_close(struct net_device *netdev)
1707 {
1708         struct igbvf_adapter *adapter = netdev_priv(netdev);
1709
1710         WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1711         igbvf_down(adapter);
1712
1713         igbvf_free_irq(adapter);
1714
1715         igbvf_free_tx_resources(adapter->tx_ring);
1716         igbvf_free_rx_resources(adapter->rx_ring);
1717
1718         return 0;
1719 }
1720 /**
1721  * igbvf_set_mac - Change the Ethernet Address of the NIC
1722  * @netdev: network interface device structure
1723  * @p: pointer to an address structure
1724  *
1725  * Returns 0 on success, negative on failure
1726  **/
1727 static int igbvf_set_mac(struct net_device *netdev, void *p)
1728 {
1729         struct igbvf_adapter *adapter = netdev_priv(netdev);
1730         struct e1000_hw *hw = &adapter->hw;
1731         struct sockaddr *addr = p;
1732
1733         if (!is_valid_ether_addr(addr->sa_data))
1734                 return -EADDRNOTAVAIL;
1735
1736         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1737
1738         hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1739
1740         if (memcmp(addr->sa_data, hw->mac.addr, 6))
1741                 return -EADDRNOTAVAIL;
1742
1743         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1744
1745         return 0;
1746 }
1747
1748 #define UPDATE_VF_COUNTER(reg, name)                                    \
1749         {                                                               \
1750                 u32 current_counter = er32(reg);                        \
1751                 if (current_counter < adapter->stats.last_##name)       \
1752                         adapter->stats.name += 0x100000000LL;           \
1753                 adapter->stats.last_##name = current_counter;           \
1754                 adapter->stats.name &= 0xFFFFFFFF00000000LL;            \
1755                 adapter->stats.name |= current_counter;                 \
1756         }
1757
1758 /**
1759  * igbvf_update_stats - Update the board statistics counters
1760  * @adapter: board private structure
1761 **/
1762 void igbvf_update_stats(struct igbvf_adapter *adapter)
1763 {
1764         struct e1000_hw *hw = &adapter->hw;
1765         struct pci_dev *pdev = adapter->pdev;
1766
1767         /*
1768          * Prevent stats update while adapter is being reset, link is down
1769          * or if the pci connection is down.
1770          */
1771         if (adapter->link_speed == 0)
1772                 return;
1773
1774         if (test_bit(__IGBVF_RESETTING, &adapter->state))
1775                 return;
1776
1777         if (pci_channel_offline(pdev))
1778                 return;
1779
1780         UPDATE_VF_COUNTER(VFGPRC, gprc);
1781         UPDATE_VF_COUNTER(VFGORC, gorc);
1782         UPDATE_VF_COUNTER(VFGPTC, gptc);
1783         UPDATE_VF_COUNTER(VFGOTC, gotc);
1784         UPDATE_VF_COUNTER(VFMPRC, mprc);
1785         UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1786         UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1787         UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1788         UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1789
1790         /* Fill out the OS statistics structure */
1791         adapter->net_stats.multicast = adapter->stats.mprc;
1792 }
1793
1794 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1795 {
1796         dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s\n",
1797                  adapter->link_speed,
1798                  ((adapter->link_duplex == FULL_DUPLEX) ?
1799                   "Full Duplex" : "Half Duplex"));
1800 }
1801
1802 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1803 {
1804         struct e1000_hw *hw = &adapter->hw;
1805         s32 ret_val = E1000_SUCCESS;
1806         bool link_active;
1807
1808         /* If interface is down, stay link down */
1809         if (test_bit(__IGBVF_DOWN, &adapter->state))
1810                 return false;
1811
1812         ret_val = hw->mac.ops.check_for_link(hw);
1813         link_active = !hw->mac.get_link_status;
1814
1815         /* if check for link returns error we will need to reset */
1816         if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1817                 schedule_work(&adapter->reset_task);
1818
1819         return link_active;
1820 }
1821
1822 /**
1823  * igbvf_watchdog - Timer Call-back
1824  * @data: pointer to adapter cast into an unsigned long
1825  **/
1826 static void igbvf_watchdog(unsigned long data)
1827 {
1828         struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1829
1830         /* Do the rest outside of interrupt context */
1831         schedule_work(&adapter->watchdog_task);
1832 }
1833
1834 static void igbvf_watchdog_task(struct work_struct *work)
1835 {
1836         struct igbvf_adapter *adapter = container_of(work,
1837                                                      struct igbvf_adapter,
1838                                                      watchdog_task);
1839         struct net_device *netdev = adapter->netdev;
1840         struct e1000_mac_info *mac = &adapter->hw.mac;
1841         struct igbvf_ring *tx_ring = adapter->tx_ring;
1842         struct e1000_hw *hw = &adapter->hw;
1843         u32 link;
1844         int tx_pending = 0;
1845
1846         link = igbvf_has_link(adapter);
1847
1848         if (link) {
1849                 if (!netif_carrier_ok(netdev)) {
1850                         bool txb2b = 1;
1851
1852                         mac->ops.get_link_up_info(&adapter->hw,
1853                                                   &adapter->link_speed,
1854                                                   &adapter->link_duplex);
1855                         igbvf_print_link_info(adapter);
1856
1857                         /* adjust timeout factor according to speed/duplex */
1858                         adapter->tx_timeout_factor = 1;
1859                         switch (adapter->link_speed) {
1860                         case SPEED_10:
1861                                 txb2b = 0;
1862                                 adapter->tx_timeout_factor = 16;
1863                                 break;
1864                         case SPEED_100:
1865                                 txb2b = 0;
1866                                 /* maybe add some timeout factor ? */
1867                                 break;
1868                         }
1869
1870                         netif_carrier_on(netdev);
1871                         netif_wake_queue(netdev);
1872                 }
1873         } else {
1874                 if (netif_carrier_ok(netdev)) {
1875                         adapter->link_speed = 0;
1876                         adapter->link_duplex = 0;
1877                         dev_info(&adapter->pdev->dev, "Link is Down\n");
1878                         netif_carrier_off(netdev);
1879                         netif_stop_queue(netdev);
1880                 }
1881         }
1882
1883         if (netif_carrier_ok(netdev)) {
1884                 igbvf_update_stats(adapter);
1885         } else {
1886                 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1887                               tx_ring->count);
1888                 if (tx_pending) {
1889                         /*
1890                          * We've lost link, so the controller stops DMA,
1891                          * but we've got queued Tx work that's never going
1892                          * to get done, so reset controller to flush Tx.
1893                          * (Do the reset outside of interrupt context).
1894                          */
1895                         adapter->tx_timeout_count++;
1896                         schedule_work(&adapter->reset_task);
1897                 }
1898         }
1899
1900         /* Cause software interrupt to ensure Rx ring is cleaned */
1901         ew32(EICS, adapter->rx_ring->eims_value);
1902
1903         /* Force detection of hung controller every watchdog period */
1904         adapter->detect_tx_hung = 1;
1905
1906         /* Reset the timer */
1907         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1908                 mod_timer(&adapter->watchdog_timer,
1909                           round_jiffies(jiffies + (2 * HZ)));
1910 }
1911
1912 #define IGBVF_TX_FLAGS_CSUM             0x00000001
1913 #define IGBVF_TX_FLAGS_VLAN             0x00000002
1914 #define IGBVF_TX_FLAGS_TSO              0x00000004
1915 #define IGBVF_TX_FLAGS_IPV4             0x00000008
1916 #define IGBVF_TX_FLAGS_VLAN_MASK        0xffff0000
1917 #define IGBVF_TX_FLAGS_VLAN_SHIFT       16
1918
1919 static int igbvf_tso(struct igbvf_adapter *adapter,
1920                      struct igbvf_ring *tx_ring,
1921                      struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1922 {
1923         struct e1000_adv_tx_context_desc *context_desc;
1924         unsigned int i;
1925         int err;
1926         struct igbvf_buffer *buffer_info;
1927         u32 info = 0, tu_cmd = 0;
1928         u32 mss_l4len_idx, l4len;
1929         *hdr_len = 0;
1930
1931         if (skb_header_cloned(skb)) {
1932                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1933                 if (err) {
1934                         dev_err(&adapter->pdev->dev,
1935                                 "igbvf_tso returning an error\n");
1936                         return err;
1937                 }
1938         }
1939
1940         l4len = tcp_hdrlen(skb);
1941         *hdr_len += l4len;
1942
1943         if (skb->protocol == htons(ETH_P_IP)) {
1944                 struct iphdr *iph = ip_hdr(skb);
1945                 iph->tot_len = 0;
1946                 iph->check = 0;
1947                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1948                                                          iph->daddr, 0,
1949                                                          IPPROTO_TCP,
1950                                                          0);
1951         } else if (skb_is_gso_v6(skb)) {
1952                 ipv6_hdr(skb)->payload_len = 0;
1953                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1954                                                        &ipv6_hdr(skb)->daddr,
1955                                                        0, IPPROTO_TCP, 0);
1956         }
1957
1958         i = tx_ring->next_to_use;
1959
1960         buffer_info = &tx_ring->buffer_info[i];
1961         context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1962         /* VLAN MACLEN IPLEN */
1963         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1964                 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1965         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1966         *hdr_len += skb_network_offset(skb);
1967         info |= (skb_transport_header(skb) - skb_network_header(skb));
1968         *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1969         context_desc->vlan_macip_lens = cpu_to_le32(info);
1970
1971         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1972         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1973
1974         if (skb->protocol == htons(ETH_P_IP))
1975                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1976         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1977
1978         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1979
1980         /* MSS L4LEN IDX */
1981         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1982         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1983
1984         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1985         context_desc->seqnum_seed = 0;
1986
1987         buffer_info->time_stamp = jiffies;
1988         buffer_info->next_to_watch = i;
1989         buffer_info->dma = 0;
1990         i++;
1991         if (i == tx_ring->count)
1992                 i = 0;
1993
1994         tx_ring->next_to_use = i;
1995
1996         return true;
1997 }
1998
1999 static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
2000                                  struct igbvf_ring *tx_ring,
2001                                  struct sk_buff *skb, u32 tx_flags)
2002 {
2003         struct e1000_adv_tx_context_desc *context_desc;
2004         unsigned int i;
2005         struct igbvf_buffer *buffer_info;
2006         u32 info = 0, tu_cmd = 0;
2007
2008         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2009             (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
2010                 i = tx_ring->next_to_use;
2011                 buffer_info = &tx_ring->buffer_info[i];
2012                 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
2013
2014                 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2015                         info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
2016
2017                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2018                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2019                         info |= (skb_transport_header(skb) -
2020                                  skb_network_header(skb));
2021
2022
2023                 context_desc->vlan_macip_lens = cpu_to_le32(info);
2024
2025                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2026
2027                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2028                         switch (skb->protocol) {
2029                         case __constant_htons(ETH_P_IP):
2030                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2031                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2032                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2033                                 break;
2034                         case __constant_htons(ETH_P_IPV6):
2035                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2036                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2037                                 break;
2038                         default:
2039                                 break;
2040                         }
2041                 }
2042
2043                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2044                 context_desc->seqnum_seed = 0;
2045                 context_desc->mss_l4len_idx = 0;
2046
2047                 buffer_info->time_stamp = jiffies;
2048                 buffer_info->next_to_watch = i;
2049                 buffer_info->dma = 0;
2050                 i++;
2051                 if (i == tx_ring->count)
2052                         i = 0;
2053                 tx_ring->next_to_use = i;
2054
2055                 return true;
2056         }
2057
2058         return false;
2059 }
2060
2061 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2062 {
2063         struct igbvf_adapter *adapter = netdev_priv(netdev);
2064
2065         /* there is enough descriptors then we don't need to worry  */
2066         if (igbvf_desc_unused(adapter->tx_ring) >= size)
2067                 return 0;
2068
2069         netif_stop_queue(netdev);
2070
2071         smp_mb();
2072
2073         /* We need to check again just in case room has been made available */
2074         if (igbvf_desc_unused(adapter->tx_ring) < size)
2075                 return -EBUSY;
2076
2077         netif_wake_queue(netdev);
2078
2079         ++adapter->restart_queue;
2080         return 0;
2081 }
2082
2083 #define IGBVF_MAX_TXD_PWR       16
2084 #define IGBVF_MAX_DATA_PER_TXD  (1 << IGBVF_MAX_TXD_PWR)
2085
2086 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2087                                    struct igbvf_ring *tx_ring,
2088                                    struct sk_buff *skb,
2089                                    unsigned int first)
2090 {
2091         struct igbvf_buffer *buffer_info;
2092         struct pci_dev *pdev = adapter->pdev;
2093         unsigned int len = skb_headlen(skb);
2094         unsigned int count = 0, i;
2095         unsigned int f;
2096
2097         i = tx_ring->next_to_use;
2098
2099         buffer_info = &tx_ring->buffer_info[i];
2100         BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2101         buffer_info->length = len;
2102         /* set time_stamp *before* dma to help avoid a possible race */
2103         buffer_info->time_stamp = jiffies;
2104         buffer_info->next_to_watch = i;
2105         buffer_info->mapped_as_page = false;
2106         buffer_info->dma = pci_map_single(pdev, skb->data, len,
2107                                           PCI_DMA_TODEVICE);
2108         if (pci_dma_mapping_error(pdev, buffer_info->dma))
2109                 goto dma_error;
2110
2111
2112         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2113                 struct skb_frag_struct *frag;
2114
2115                 count++;
2116                 i++;
2117                 if (i == tx_ring->count)
2118                         i = 0;
2119
2120                 frag = &skb_shinfo(skb)->frags[f];
2121                 len = frag->size;
2122
2123                 buffer_info = &tx_ring->buffer_info[i];
2124                 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2125                 buffer_info->length = len;
2126                 buffer_info->time_stamp = jiffies;
2127                 buffer_info->next_to_watch = i;
2128                 buffer_info->mapped_as_page = true;
2129                 buffer_info->dma = pci_map_page(pdev,
2130                                                 frag->page,
2131                                                 frag->page_offset,
2132                                                 len,
2133                                                 PCI_DMA_TODEVICE);
2134                 if (pci_dma_mapping_error(pdev, buffer_info->dma))
2135                         goto dma_error;
2136         }
2137
2138         tx_ring->buffer_info[i].skb = skb;
2139         tx_ring->buffer_info[first].next_to_watch = i;
2140
2141         return ++count;
2142
2143 dma_error:
2144         dev_err(&pdev->dev, "TX DMA map failed\n");
2145
2146         /* clear timestamp and dma mappings for failed buffer_info mapping */
2147         buffer_info->dma = 0;
2148         buffer_info->time_stamp = 0;
2149         buffer_info->length = 0;
2150         buffer_info->next_to_watch = 0;
2151         buffer_info->mapped_as_page = false;
2152         if (count)
2153                 count--;
2154
2155         /* clear timestamp and dma mappings for remaining portion of packet */
2156         while (count--) {
2157                 if (i==0)
2158                         i += tx_ring->count;
2159                 i--;
2160                 buffer_info = &tx_ring->buffer_info[i];
2161                 igbvf_put_txbuf(adapter, buffer_info);
2162         }
2163
2164         return 0;
2165 }
2166
2167 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2168                                       struct igbvf_ring *tx_ring,
2169                                       int tx_flags, int count, u32 paylen,
2170                                       u8 hdr_len)
2171 {
2172         union e1000_adv_tx_desc *tx_desc = NULL;
2173         struct igbvf_buffer *buffer_info;
2174         u32 olinfo_status = 0, cmd_type_len;
2175         unsigned int i;
2176
2177         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2178                         E1000_ADVTXD_DCMD_DEXT);
2179
2180         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2181                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2182
2183         if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2184                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2185
2186                 /* insert tcp checksum */
2187                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2188
2189                 /* insert ip checksum */
2190                 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2191                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2192
2193         } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2194                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2195         }
2196
2197         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2198
2199         i = tx_ring->next_to_use;
2200         while (count--) {
2201                 buffer_info = &tx_ring->buffer_info[i];
2202                 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2203                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2204                 tx_desc->read.cmd_type_len =
2205                          cpu_to_le32(cmd_type_len | buffer_info->length);
2206                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2207                 i++;
2208                 if (i == tx_ring->count)
2209                         i = 0;
2210         }
2211
2212         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2213         /* Force memory writes to complete before letting h/w
2214          * know there are new descriptors to fetch.  (Only
2215          * applicable for weak-ordered memory model archs,
2216          * such as IA-64). */
2217         wmb();
2218
2219         tx_ring->next_to_use = i;
2220         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2221         /* we need this if more than one processor can write to our tail
2222          * at a time, it syncronizes IO on IA64/Altix systems */
2223         mmiowb();
2224 }
2225
2226 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2227                                              struct net_device *netdev,
2228                                              struct igbvf_ring *tx_ring)
2229 {
2230         struct igbvf_adapter *adapter = netdev_priv(netdev);
2231         unsigned int first, tx_flags = 0;
2232         u8 hdr_len = 0;
2233         int count = 0;
2234         int tso = 0;
2235
2236         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2237                 dev_kfree_skb_any(skb);
2238                 return NETDEV_TX_OK;
2239         }
2240
2241         if (skb->len <= 0) {
2242                 dev_kfree_skb_any(skb);
2243                 return NETDEV_TX_OK;
2244         }
2245
2246         /*
2247          * need: count + 4 desc gap to keep tail from touching
2248          *       + 2 desc gap to keep tail from touching head,
2249          *       + 1 desc for skb->data,
2250          *       + 1 desc for context descriptor,
2251          * head, otherwise try next time
2252          */
2253         if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2254                 /* this is a hard error */
2255                 return NETDEV_TX_BUSY;
2256         }
2257
2258         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2259                 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2260                 tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2261         }
2262
2263         if (skb->protocol == htons(ETH_P_IP))
2264                 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2265
2266         first = tx_ring->next_to_use;
2267
2268         tso = skb_is_gso(skb) ?
2269                 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2270         if (unlikely(tso < 0)) {
2271                 dev_kfree_skb_any(skb);
2272                 return NETDEV_TX_OK;
2273         }
2274
2275         if (tso)
2276                 tx_flags |= IGBVF_TX_FLAGS_TSO;
2277         else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2278                  (skb->ip_summed == CHECKSUM_PARTIAL))
2279                 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2280
2281         /*
2282          * count reflects descriptors mapped, if 0 then mapping error
2283          * has occured and we need to rewind the descriptor queue
2284          */
2285         count = igbvf_tx_map_adv(adapter, tx_ring, skb, first);
2286
2287         if (count) {
2288                 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2289                                    skb->len, hdr_len);
2290                 /* Make sure there is space in the ring for the next send. */
2291                 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2292         } else {
2293                 dev_kfree_skb_any(skb);
2294                 tx_ring->buffer_info[first].time_stamp = 0;
2295                 tx_ring->next_to_use = first;
2296         }
2297
2298         return NETDEV_TX_OK;
2299 }
2300
2301 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2302                                     struct net_device *netdev)
2303 {
2304         struct igbvf_adapter *adapter = netdev_priv(netdev);
2305         struct igbvf_ring *tx_ring;
2306
2307         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2308                 dev_kfree_skb_any(skb);
2309                 return NETDEV_TX_OK;
2310         }
2311
2312         tx_ring = &adapter->tx_ring[0];
2313
2314         return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2315 }
2316
2317 /**
2318  * igbvf_tx_timeout - Respond to a Tx Hang
2319  * @netdev: network interface device structure
2320  **/
2321 static void igbvf_tx_timeout(struct net_device *netdev)
2322 {
2323         struct igbvf_adapter *adapter = netdev_priv(netdev);
2324
2325         /* Do the reset outside of interrupt context */
2326         adapter->tx_timeout_count++;
2327         schedule_work(&adapter->reset_task);
2328 }
2329
2330 static void igbvf_reset_task(struct work_struct *work)
2331 {
2332         struct igbvf_adapter *adapter;
2333         adapter = container_of(work, struct igbvf_adapter, reset_task);
2334
2335         igbvf_reinit_locked(adapter);
2336 }
2337
2338 /**
2339  * igbvf_get_stats - Get System Network Statistics
2340  * @netdev: network interface device structure
2341  *
2342  * Returns the address of the device statistics structure.
2343  * The statistics are actually updated from the timer callback.
2344  **/
2345 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2346 {
2347         struct igbvf_adapter *adapter = netdev_priv(netdev);
2348
2349         /* only return the current stats */
2350         return &adapter->net_stats;
2351 }
2352
2353 /**
2354  * igbvf_change_mtu - Change the Maximum Transfer Unit
2355  * @netdev: network interface device structure
2356  * @new_mtu: new value for maximum frame size
2357  *
2358  * Returns 0 on success, negative on failure
2359  **/
2360 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2361 {
2362         struct igbvf_adapter *adapter = netdev_priv(netdev);
2363         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2364
2365         if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2366                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2367                 return -EINVAL;
2368         }
2369
2370 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2371         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2372                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2373                 return -EINVAL;
2374         }
2375
2376         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2377                 msleep(1);
2378         /* igbvf_down has a dependency on max_frame_size */
2379         adapter->max_frame_size = max_frame;
2380         if (netif_running(netdev))
2381                 igbvf_down(adapter);
2382
2383         /*
2384          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2385          * means we reserve 2 more, this pushes us to allocate from the next
2386          * larger slab size.
2387          * i.e. RXBUFFER_2048 --> size-4096 slab
2388          * However with the new *_jumbo_rx* routines, jumbo receives will use
2389          * fragmented skbs
2390          */
2391
2392         if (max_frame <= 1024)
2393                 adapter->rx_buffer_len = 1024;
2394         else if (max_frame <= 2048)
2395                 adapter->rx_buffer_len = 2048;
2396         else
2397 #if (PAGE_SIZE / 2) > 16384
2398                 adapter->rx_buffer_len = 16384;
2399 #else
2400                 adapter->rx_buffer_len = PAGE_SIZE / 2;
2401 #endif
2402
2403
2404         /* adjust allocation if LPE protects us, and we aren't using SBP */
2405         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2406              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2407                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2408                                          ETH_FCS_LEN;
2409
2410         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2411                  netdev->mtu, new_mtu);
2412         netdev->mtu = new_mtu;
2413
2414         if (netif_running(netdev))
2415                 igbvf_up(adapter);
2416         else
2417                 igbvf_reset(adapter);
2418
2419         clear_bit(__IGBVF_RESETTING, &adapter->state);
2420
2421         return 0;
2422 }
2423
2424 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2425 {
2426         switch (cmd) {
2427         default:
2428                 return -EOPNOTSUPP;
2429         }
2430 }
2431
2432 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2433 {
2434         struct net_device *netdev = pci_get_drvdata(pdev);
2435         struct igbvf_adapter *adapter = netdev_priv(netdev);
2436 #ifdef CONFIG_PM
2437         int retval = 0;
2438 #endif
2439
2440         netif_device_detach(netdev);
2441
2442         if (netif_running(netdev)) {
2443                 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2444                 igbvf_down(adapter);
2445                 igbvf_free_irq(adapter);
2446         }
2447
2448 #ifdef CONFIG_PM
2449         retval = pci_save_state(pdev);
2450         if (retval)
2451                 return retval;
2452 #endif
2453
2454         pci_disable_device(pdev);
2455
2456         return 0;
2457 }
2458
2459 #ifdef CONFIG_PM
2460 static int igbvf_resume(struct pci_dev *pdev)
2461 {
2462         struct net_device *netdev = pci_get_drvdata(pdev);
2463         struct igbvf_adapter *adapter = netdev_priv(netdev);
2464         u32 err;
2465
2466         pci_restore_state(pdev);
2467         err = pci_enable_device_mem(pdev);
2468         if (err) {
2469                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2470                 return err;
2471         }
2472
2473         pci_set_master(pdev);
2474
2475         if (netif_running(netdev)) {
2476                 err = igbvf_request_irq(adapter);
2477                 if (err)
2478                         return err;
2479         }
2480
2481         igbvf_reset(adapter);
2482
2483         if (netif_running(netdev))
2484                 igbvf_up(adapter);
2485
2486         netif_device_attach(netdev);
2487
2488         return 0;
2489 }
2490 #endif
2491
2492 static void igbvf_shutdown(struct pci_dev *pdev)
2493 {
2494         igbvf_suspend(pdev, PMSG_SUSPEND);
2495 }
2496
2497 #ifdef CONFIG_NET_POLL_CONTROLLER
2498 /*
2499  * Polling 'interrupt' - used by things like netconsole to send skbs
2500  * without having to re-enable interrupts. It's not called while
2501  * the interrupt routine is executing.
2502  */
2503 static void igbvf_netpoll(struct net_device *netdev)
2504 {
2505         struct igbvf_adapter *adapter = netdev_priv(netdev);
2506
2507         disable_irq(adapter->pdev->irq);
2508
2509         igbvf_clean_tx_irq(adapter->tx_ring);
2510
2511         enable_irq(adapter->pdev->irq);
2512 }
2513 #endif
2514
2515 /**
2516  * igbvf_io_error_detected - called when PCI error is detected
2517  * @pdev: Pointer to PCI device
2518  * @state: The current pci connection state
2519  *
2520  * This function is called after a PCI bus error affecting
2521  * this device has been detected.
2522  */
2523 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2524                                                 pci_channel_state_t state)
2525 {
2526         struct net_device *netdev = pci_get_drvdata(pdev);
2527         struct igbvf_adapter *adapter = netdev_priv(netdev);
2528
2529         netif_device_detach(netdev);
2530
2531         if (state == pci_channel_io_perm_failure)
2532                 return PCI_ERS_RESULT_DISCONNECT;
2533
2534         if (netif_running(netdev))
2535                 igbvf_down(adapter);
2536         pci_disable_device(pdev);
2537
2538         /* Request a slot slot reset. */
2539         return PCI_ERS_RESULT_NEED_RESET;
2540 }
2541
2542 /**
2543  * igbvf_io_slot_reset - called after the pci bus has been reset.
2544  * @pdev: Pointer to PCI device
2545  *
2546  * Restart the card from scratch, as if from a cold-boot. Implementation
2547  * resembles the first-half of the igbvf_resume routine.
2548  */
2549 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2550 {
2551         struct net_device *netdev = pci_get_drvdata(pdev);
2552         struct igbvf_adapter *adapter = netdev_priv(netdev);
2553
2554         if (pci_enable_device_mem(pdev)) {
2555                 dev_err(&pdev->dev,
2556                         "Cannot re-enable PCI device after reset.\n");
2557                 return PCI_ERS_RESULT_DISCONNECT;
2558         }
2559         pci_set_master(pdev);
2560
2561         igbvf_reset(adapter);
2562
2563         return PCI_ERS_RESULT_RECOVERED;
2564 }
2565
2566 /**
2567  * igbvf_io_resume - called when traffic can start flowing again.
2568  * @pdev: Pointer to PCI device
2569  *
2570  * This callback is called when the error recovery driver tells us that
2571  * its OK to resume normal operation. Implementation resembles the
2572  * second-half of the igbvf_resume routine.
2573  */
2574 static void igbvf_io_resume(struct pci_dev *pdev)
2575 {
2576         struct net_device *netdev = pci_get_drvdata(pdev);
2577         struct igbvf_adapter *adapter = netdev_priv(netdev);
2578
2579         if (netif_running(netdev)) {
2580                 if (igbvf_up(adapter)) {
2581                         dev_err(&pdev->dev,
2582                                 "can't bring device back up after reset\n");
2583                         return;
2584                 }
2585         }
2586
2587         netif_device_attach(netdev);
2588 }
2589
2590 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2591 {
2592         struct e1000_hw *hw = &adapter->hw;
2593         struct net_device *netdev = adapter->netdev;
2594         struct pci_dev *pdev = adapter->pdev;
2595
2596         dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2597         dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2598         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
2599 }
2600
2601 static const struct net_device_ops igbvf_netdev_ops = {
2602         .ndo_open                       = igbvf_open,
2603         .ndo_stop                       = igbvf_close,
2604         .ndo_start_xmit                 = igbvf_xmit_frame,
2605         .ndo_get_stats                  = igbvf_get_stats,
2606         .ndo_set_multicast_list         = igbvf_set_multi,
2607         .ndo_set_mac_address            = igbvf_set_mac,
2608         .ndo_change_mtu                 = igbvf_change_mtu,
2609         .ndo_do_ioctl                   = igbvf_ioctl,
2610         .ndo_tx_timeout                 = igbvf_tx_timeout,
2611         .ndo_vlan_rx_register           = igbvf_vlan_rx_register,
2612         .ndo_vlan_rx_add_vid            = igbvf_vlan_rx_add_vid,
2613         .ndo_vlan_rx_kill_vid           = igbvf_vlan_rx_kill_vid,
2614 #ifdef CONFIG_NET_POLL_CONTROLLER
2615         .ndo_poll_controller            = igbvf_netpoll,
2616 #endif
2617 };
2618
2619 /**
2620  * igbvf_probe - Device Initialization Routine
2621  * @pdev: PCI device information struct
2622  * @ent: entry in igbvf_pci_tbl
2623  *
2624  * Returns 0 on success, negative on failure
2625  *
2626  * igbvf_probe initializes an adapter identified by a pci_dev structure.
2627  * The OS initialization, configuring of the adapter private structure,
2628  * and a hardware reset occur.
2629  **/
2630 static int __devinit igbvf_probe(struct pci_dev *pdev,
2631                                  const struct pci_device_id *ent)
2632 {
2633         struct net_device *netdev;
2634         struct igbvf_adapter *adapter;
2635         struct e1000_hw *hw;
2636         const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2637
2638         static int cards_found;
2639         int err, pci_using_dac;
2640
2641         err = pci_enable_device_mem(pdev);
2642         if (err)
2643                 return err;
2644
2645         pci_using_dac = 0;
2646         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2647         if (!err) {
2648                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2649                 if (!err)
2650                         pci_using_dac = 1;
2651         } else {
2652                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2653                 if (err) {
2654                         err = pci_set_consistent_dma_mask(pdev,
2655                                                           DMA_BIT_MASK(32));
2656                         if (err) {
2657                                 dev_err(&pdev->dev, "No usable DMA "
2658                                         "configuration, aborting\n");
2659                                 goto err_dma;
2660                         }
2661                 }
2662         }
2663
2664         err = pci_request_regions(pdev, igbvf_driver_name);
2665         if (err)
2666                 goto err_pci_reg;
2667
2668         pci_set_master(pdev);
2669
2670         err = -ENOMEM;
2671         netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2672         if (!netdev)
2673                 goto err_alloc_etherdev;
2674
2675         SET_NETDEV_DEV(netdev, &pdev->dev);
2676
2677         pci_set_drvdata(pdev, netdev);
2678         adapter = netdev_priv(netdev);
2679         hw = &adapter->hw;
2680         adapter->netdev = netdev;
2681         adapter->pdev = pdev;
2682         adapter->ei = ei;
2683         adapter->pba = ei->pba;
2684         adapter->flags = ei->flags;
2685         adapter->hw.back = adapter;
2686         adapter->hw.mac.type = ei->mac;
2687         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
2688
2689         /* PCI config space info */
2690
2691         hw->vendor_id = pdev->vendor;
2692         hw->device_id = pdev->device;
2693         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2694         hw->subsystem_device_id = pdev->subsystem_device;
2695
2696         pci_read_config_byte(pdev, PCI_REVISION_ID, &hw->revision_id);
2697
2698         err = -EIO;
2699         adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2700                                       pci_resource_len(pdev, 0));
2701
2702         if (!adapter->hw.hw_addr)
2703                 goto err_ioremap;
2704
2705         if (ei->get_variants) {
2706                 err = ei->get_variants(adapter);
2707                 if (err)
2708                         goto err_ioremap;
2709         }
2710
2711         /* setup adapter struct */
2712         err = igbvf_sw_init(adapter);
2713         if (err)
2714                 goto err_sw_init;
2715
2716         /* construct the net_device struct */
2717         netdev->netdev_ops = &igbvf_netdev_ops;
2718
2719         igbvf_set_ethtool_ops(netdev);
2720         netdev->watchdog_timeo = 5 * HZ;
2721         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2722
2723         adapter->bd_number = cards_found++;
2724
2725         netdev->features = NETIF_F_SG |
2726                            NETIF_F_IP_CSUM |
2727                            NETIF_F_HW_VLAN_TX |
2728                            NETIF_F_HW_VLAN_RX |
2729                            NETIF_F_HW_VLAN_FILTER;
2730
2731         netdev->features |= NETIF_F_IPV6_CSUM;
2732         netdev->features |= NETIF_F_TSO;
2733         netdev->features |= NETIF_F_TSO6;
2734
2735         if (pci_using_dac)
2736                 netdev->features |= NETIF_F_HIGHDMA;
2737
2738         netdev->vlan_features |= NETIF_F_TSO;
2739         netdev->vlan_features |= NETIF_F_TSO6;
2740         netdev->vlan_features |= NETIF_F_IP_CSUM;
2741         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2742         netdev->vlan_features |= NETIF_F_SG;
2743
2744         /*reset the controller to put the device in a known good state */
2745         err = hw->mac.ops.reset_hw(hw);
2746         if (err) {
2747                 dev_info(&pdev->dev,
2748                          "PF still in reset state, assigning new address."
2749                          " Is the PF interface up?\n");
2750                 random_ether_addr(hw->mac.addr);
2751         } else {
2752                 err = hw->mac.ops.read_mac_addr(hw);
2753                 if (err) {
2754                         dev_err(&pdev->dev, "Error reading MAC address\n");
2755                         goto err_hw_init;
2756                 }
2757         }
2758
2759         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
2760         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
2761
2762         if (!is_valid_ether_addr(netdev->perm_addr)) {
2763                 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
2764                         netdev->dev_addr);
2765                 err = -EIO;
2766                 goto err_hw_init;
2767         }
2768
2769         setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2770                     (unsigned long) adapter);
2771
2772         INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2773         INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2774
2775         /* ring size defaults */
2776         adapter->rx_ring->count = 1024;
2777         adapter->tx_ring->count = 1024;
2778
2779         /* reset the hardware with the new settings */
2780         igbvf_reset(adapter);
2781
2782         /* tell the stack to leave us alone until igbvf_open() is called */
2783         netif_carrier_off(netdev);
2784         netif_stop_queue(netdev);
2785
2786         strcpy(netdev->name, "eth%d");
2787         err = register_netdev(netdev);
2788         if (err)
2789                 goto err_hw_init;
2790
2791         igbvf_print_device_info(adapter);
2792
2793         igbvf_initialize_last_counter_stats(adapter);
2794
2795         return 0;
2796
2797 err_hw_init:
2798         kfree(adapter->tx_ring);
2799         kfree(adapter->rx_ring);
2800 err_sw_init:
2801         igbvf_reset_interrupt_capability(adapter);
2802         iounmap(adapter->hw.hw_addr);
2803 err_ioremap:
2804         free_netdev(netdev);
2805 err_alloc_etherdev:
2806         pci_release_regions(pdev);
2807 err_pci_reg:
2808 err_dma:
2809         pci_disable_device(pdev);
2810         return err;
2811 }
2812
2813 /**
2814  * igbvf_remove - Device Removal Routine
2815  * @pdev: PCI device information struct
2816  *
2817  * igbvf_remove is called by the PCI subsystem to alert the driver
2818  * that it should release a PCI device.  The could be caused by a
2819  * Hot-Plug event, or because the driver is going to be removed from
2820  * memory.
2821  **/
2822 static void __devexit igbvf_remove(struct pci_dev *pdev)
2823 {
2824         struct net_device *netdev = pci_get_drvdata(pdev);
2825         struct igbvf_adapter *adapter = netdev_priv(netdev);
2826         struct e1000_hw *hw = &adapter->hw;
2827
2828         /*
2829          * flush_scheduled work may reschedule our watchdog task, so
2830          * explicitly disable watchdog tasks from being rescheduled
2831          */
2832         set_bit(__IGBVF_DOWN, &adapter->state);
2833         del_timer_sync(&adapter->watchdog_timer);
2834
2835         flush_scheduled_work();
2836
2837         unregister_netdev(netdev);
2838
2839         igbvf_reset_interrupt_capability(adapter);
2840
2841         /*
2842          * it is important to delete the napi struct prior to freeing the
2843          * rx ring so that you do not end up with null pointer refs
2844          */
2845         netif_napi_del(&adapter->rx_ring->napi);
2846         kfree(adapter->tx_ring);
2847         kfree(adapter->rx_ring);
2848
2849         iounmap(hw->hw_addr);
2850         if (hw->flash_address)
2851                 iounmap(hw->flash_address);
2852         pci_release_regions(pdev);
2853
2854         free_netdev(netdev);
2855
2856         pci_disable_device(pdev);
2857 }
2858
2859 /* PCI Error Recovery (ERS) */
2860 static struct pci_error_handlers igbvf_err_handler = {
2861         .error_detected = igbvf_io_error_detected,
2862         .slot_reset = igbvf_io_slot_reset,
2863         .resume = igbvf_io_resume,
2864 };
2865
2866 static DEFINE_PCI_DEVICE_TABLE(igbvf_pci_tbl) = {
2867         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2868         { } /* terminate list */
2869 };
2870 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2871
2872 /* PCI Device API Driver */
2873 static struct pci_driver igbvf_driver = {
2874         .name     = igbvf_driver_name,
2875         .id_table = igbvf_pci_tbl,
2876         .probe    = igbvf_probe,
2877         .remove   = __devexit_p(igbvf_remove),
2878 #ifdef CONFIG_PM
2879         /* Power Management Hooks */
2880         .suspend  = igbvf_suspend,
2881         .resume   = igbvf_resume,
2882 #endif
2883         .shutdown = igbvf_shutdown,
2884         .err_handler = &igbvf_err_handler
2885 };
2886
2887 /**
2888  * igbvf_init_module - Driver Registration Routine
2889  *
2890  * igbvf_init_module is the first routine called when the driver is
2891  * loaded. All it does is register with the PCI subsystem.
2892  **/
2893 static int __init igbvf_init_module(void)
2894 {
2895         int ret;
2896         printk(KERN_INFO "%s - version %s\n",
2897                igbvf_driver_string, igbvf_driver_version);
2898         printk(KERN_INFO "%s\n", igbvf_copyright);
2899
2900         ret = pci_register_driver(&igbvf_driver);
2901         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, igbvf_driver_name,
2902                                PM_QOS_DEFAULT_VALUE);
2903
2904         return ret;
2905 }
2906 module_init(igbvf_init_module);
2907
2908 /**
2909  * igbvf_exit_module - Driver Exit Cleanup Routine
2910  *
2911  * igbvf_exit_module is called just before the driver is removed
2912  * from memory.
2913  **/
2914 static void __exit igbvf_exit_module(void)
2915 {
2916         pci_unregister_driver(&igbvf_driver);
2917         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, igbvf_driver_name);
2918 }
2919 module_exit(igbvf_exit_module);
2920
2921
2922 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2923 MODULE_DESCRIPTION("Intel(R) 82576 Virtual Function Network Driver");
2924 MODULE_LICENSE("GPL");
2925 MODULE_VERSION(DRV_VERSION);
2926
2927 /* netdev.c */