Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[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         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
1309 }
1310
1311 /**
1312  * igbvf_setup_srrctl - configure the receive control registers
1313  * @adapter: Board private structure
1314  **/
1315 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1316 {
1317         struct e1000_hw *hw = &adapter->hw;
1318         u32 srrctl = 0;
1319
1320         srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1321                     E1000_SRRCTL_BSIZEHDR_MASK |
1322                     E1000_SRRCTL_BSIZEPKT_MASK);
1323
1324         /* Enable queue drop to avoid head of line blocking */
1325         srrctl |= E1000_SRRCTL_DROP_EN;
1326
1327         /* Setup buffer sizes */
1328         srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1329                   E1000_SRRCTL_BSIZEPKT_SHIFT;
1330
1331         if (adapter->rx_buffer_len < 2048) {
1332                 adapter->rx_ps_hdr_size = 0;
1333                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1334         } else {
1335                 adapter->rx_ps_hdr_size = 128;
1336                 srrctl |= adapter->rx_ps_hdr_size <<
1337                           E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1338                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1339         }
1340
1341         ew32(SRRCTL(0), srrctl);
1342 }
1343
1344 /**
1345  * igbvf_configure_rx - Configure Receive Unit after Reset
1346  * @adapter: board private structure
1347  *
1348  * Configure the Rx unit of the MAC after a reset.
1349  **/
1350 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1351 {
1352         struct e1000_hw *hw = &adapter->hw;
1353         struct igbvf_ring *rx_ring = adapter->rx_ring;
1354         u64 rdba;
1355         u32 rdlen, rxdctl;
1356
1357         /* disable receives */
1358         rxdctl = er32(RXDCTL(0));
1359         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1360         msleep(10);
1361
1362         rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1363
1364         /*
1365          * Setup the HW Rx Head and Tail Descriptor Pointers and
1366          * the Base and Length of the Rx Descriptor Ring
1367          */
1368         rdba = rx_ring->dma;
1369         ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1370         ew32(RDBAH(0), (rdba >> 32));
1371         ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1372         rx_ring->head = E1000_RDH(0);
1373         rx_ring->tail = E1000_RDT(0);
1374         ew32(RDH(0), 0);
1375         ew32(RDT(0), 0);
1376
1377         rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1378         rxdctl &= 0xFFF00000;
1379         rxdctl |= IGBVF_RX_PTHRESH;
1380         rxdctl |= IGBVF_RX_HTHRESH << 8;
1381         rxdctl |= IGBVF_RX_WTHRESH << 16;
1382
1383         igbvf_set_rlpml(adapter);
1384
1385         /* enable receives */
1386         ew32(RXDCTL(0), rxdctl);
1387 }
1388
1389 /**
1390  * igbvf_set_multi - Multicast and Promiscuous mode set
1391  * @netdev: network interface device structure
1392  *
1393  * The set_multi entry point is called whenever the multicast address
1394  * list or the network interface flags are updated.  This routine is
1395  * responsible for configuring the hardware for proper multicast,
1396  * promiscuous mode, and all-multi behavior.
1397  **/
1398 static void igbvf_set_multi(struct net_device *netdev)
1399 {
1400         struct igbvf_adapter *adapter = netdev_priv(netdev);
1401         struct e1000_hw *hw = &adapter->hw;
1402         struct dev_mc_list *mc_ptr;
1403         u8  *mta_list = NULL;
1404         int i;
1405
1406         if (netdev->mc_count) {
1407                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
1408                 if (!mta_list) {
1409                         dev_err(&adapter->pdev->dev,
1410                                 "failed to allocate multicast filter list\n");
1411                         return;
1412                 }
1413         }
1414
1415         /* prepare a packed array of only addresses. */
1416         mc_ptr = netdev->mc_list;
1417
1418         for (i = 0; i < netdev->mc_count; i++) {
1419                 if (!mc_ptr)
1420                         break;
1421                 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
1422                        ETH_ALEN);
1423                 mc_ptr = mc_ptr->next;
1424         }
1425
1426         hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1427         kfree(mta_list);
1428 }
1429
1430 /**
1431  * igbvf_configure - configure the hardware for Rx and Tx
1432  * @adapter: private board structure
1433  **/
1434 static void igbvf_configure(struct igbvf_adapter *adapter)
1435 {
1436         igbvf_set_multi(adapter->netdev);
1437
1438         igbvf_restore_vlan(adapter);
1439
1440         igbvf_configure_tx(adapter);
1441         igbvf_setup_srrctl(adapter);
1442         igbvf_configure_rx(adapter);
1443         igbvf_alloc_rx_buffers(adapter->rx_ring,
1444                                igbvf_desc_unused(adapter->rx_ring));
1445 }
1446
1447 /* igbvf_reset - bring the hardware into a known good state
1448  *
1449  * This function boots the hardware and enables some settings that
1450  * require a configuration cycle of the hardware - those cannot be
1451  * set/changed during runtime. After reset the device needs to be
1452  * properly configured for Rx, Tx etc.
1453  */
1454 static void igbvf_reset(struct igbvf_adapter *adapter)
1455 {
1456         struct e1000_mac_info *mac = &adapter->hw.mac;
1457         struct net_device *netdev = adapter->netdev;
1458         struct e1000_hw *hw = &adapter->hw;
1459
1460         /* Allow time for pending master requests to run */
1461         if (mac->ops.reset_hw(hw))
1462                 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1463
1464         mac->ops.init_hw(hw);
1465
1466         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1467                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1468                        netdev->addr_len);
1469                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1470                        netdev->addr_len);
1471         }
1472
1473         adapter->last_reset = jiffies;
1474 }
1475
1476 int igbvf_up(struct igbvf_adapter *adapter)
1477 {
1478         struct e1000_hw *hw = &adapter->hw;
1479
1480         /* hardware has been reset, we need to reload some things */
1481         igbvf_configure(adapter);
1482
1483         clear_bit(__IGBVF_DOWN, &adapter->state);
1484
1485         napi_enable(&adapter->rx_ring->napi);
1486         if (adapter->msix_entries)
1487                 igbvf_configure_msix(adapter);
1488
1489         /* Clear any pending interrupts. */
1490         er32(EICR);
1491         igbvf_irq_enable(adapter);
1492
1493         /* start the watchdog */
1494         hw->mac.get_link_status = 1;
1495         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1496
1497
1498         return 0;
1499 }
1500
1501 void igbvf_down(struct igbvf_adapter *adapter)
1502 {
1503         struct net_device *netdev = adapter->netdev;
1504         struct e1000_hw *hw = &adapter->hw;
1505         u32 rxdctl, txdctl;
1506
1507         /*
1508          * signal that we're down so the interrupt handler does not
1509          * reschedule our watchdog timer
1510          */
1511         set_bit(__IGBVF_DOWN, &adapter->state);
1512
1513         /* disable receives in the hardware */
1514         rxdctl = er32(RXDCTL(0));
1515         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1516
1517         netif_stop_queue(netdev);
1518
1519         /* disable transmits in the hardware */
1520         txdctl = er32(TXDCTL(0));
1521         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1522
1523         /* flush both disables and wait for them to finish */
1524         e1e_flush();
1525         msleep(10);
1526
1527         napi_disable(&adapter->rx_ring->napi);
1528
1529         igbvf_irq_disable(adapter);
1530
1531         del_timer_sync(&adapter->watchdog_timer);
1532
1533         netdev->tx_queue_len = adapter->tx_queue_len;
1534         netif_carrier_off(netdev);
1535
1536         /* record the stats before reset*/
1537         igbvf_update_stats(adapter);
1538
1539         adapter->link_speed = 0;
1540         adapter->link_duplex = 0;
1541
1542         igbvf_reset(adapter);
1543         igbvf_clean_tx_ring(adapter->tx_ring);
1544         igbvf_clean_rx_ring(adapter->rx_ring);
1545 }
1546
1547 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1548 {
1549         might_sleep();
1550         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1551                 msleep(1);
1552         igbvf_down(adapter);
1553         igbvf_up(adapter);
1554         clear_bit(__IGBVF_RESETTING, &adapter->state);
1555 }
1556
1557 /**
1558  * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1559  * @adapter: board private structure to initialize
1560  *
1561  * igbvf_sw_init initializes the Adapter private data structure.
1562  * Fields are initialized based on PCI device information and
1563  * OS network device settings (MTU size).
1564  **/
1565 static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1566 {
1567         struct net_device *netdev = adapter->netdev;
1568         s32 rc;
1569
1570         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1571         adapter->rx_ps_hdr_size = 0;
1572         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1573         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1574
1575         adapter->tx_int_delay = 8;
1576         adapter->tx_abs_int_delay = 32;
1577         adapter->rx_int_delay = 0;
1578         adapter->rx_abs_int_delay = 8;
1579         adapter->itr_setting = 3;
1580         adapter->itr = 20000;
1581
1582         /* Set various function pointers */
1583         adapter->ei->init_ops(&adapter->hw);
1584
1585         rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1586         if (rc)
1587                 return rc;
1588
1589         rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1590         if (rc)
1591                 return rc;
1592
1593         igbvf_set_interrupt_capability(adapter);
1594
1595         if (igbvf_alloc_queues(adapter))
1596                 return -ENOMEM;
1597
1598         spin_lock_init(&adapter->tx_queue_lock);
1599
1600         /* Explicitly disable IRQ since the NIC can be in any state. */
1601         igbvf_irq_disable(adapter);
1602
1603         spin_lock_init(&adapter->stats_lock);
1604
1605         set_bit(__IGBVF_DOWN, &adapter->state);
1606         return 0;
1607 }
1608
1609 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1610 {
1611         struct e1000_hw *hw = &adapter->hw;
1612
1613         adapter->stats.last_gprc = er32(VFGPRC);
1614         adapter->stats.last_gorc = er32(VFGORC);
1615         adapter->stats.last_gptc = er32(VFGPTC);
1616         adapter->stats.last_gotc = er32(VFGOTC);
1617         adapter->stats.last_mprc = er32(VFMPRC);
1618         adapter->stats.last_gotlbc = er32(VFGOTLBC);
1619         adapter->stats.last_gptlbc = er32(VFGPTLBC);
1620         adapter->stats.last_gorlbc = er32(VFGORLBC);
1621         adapter->stats.last_gprlbc = er32(VFGPRLBC);
1622
1623         adapter->stats.base_gprc = er32(VFGPRC);
1624         adapter->stats.base_gorc = er32(VFGORC);
1625         adapter->stats.base_gptc = er32(VFGPTC);
1626         adapter->stats.base_gotc = er32(VFGOTC);
1627         adapter->stats.base_mprc = er32(VFMPRC);
1628         adapter->stats.base_gotlbc = er32(VFGOTLBC);
1629         adapter->stats.base_gptlbc = er32(VFGPTLBC);
1630         adapter->stats.base_gorlbc = er32(VFGORLBC);
1631         adapter->stats.base_gprlbc = er32(VFGPRLBC);
1632 }
1633
1634 /**
1635  * igbvf_open - Called when a network interface is made active
1636  * @netdev: network interface device structure
1637  *
1638  * Returns 0 on success, negative value on failure
1639  *
1640  * The open entry point is called when a network interface is made
1641  * active by the system (IFF_UP).  At this point all resources needed
1642  * for transmit and receive operations are allocated, the interrupt
1643  * handler is registered with the OS, the watchdog timer is started,
1644  * and the stack is notified that the interface is ready.
1645  **/
1646 static int igbvf_open(struct net_device *netdev)
1647 {
1648         struct igbvf_adapter *adapter = netdev_priv(netdev);
1649         struct e1000_hw *hw = &adapter->hw;
1650         int err;
1651
1652         /* disallow open during test */
1653         if (test_bit(__IGBVF_TESTING, &adapter->state))
1654                 return -EBUSY;
1655
1656         /* allocate transmit descriptors */
1657         err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1658         if (err)
1659                 goto err_setup_tx;
1660
1661         /* allocate receive descriptors */
1662         err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1663         if (err)
1664                 goto err_setup_rx;
1665
1666         /*
1667          * before we allocate an interrupt, we must be ready to handle it.
1668          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1669          * as soon as we call pci_request_irq, so we have to setup our
1670          * clean_rx handler before we do so.
1671          */
1672         igbvf_configure(adapter);
1673
1674         err = igbvf_request_irq(adapter);
1675         if (err)
1676                 goto err_req_irq;
1677
1678         /* From here on the code is the same as igbvf_up() */
1679         clear_bit(__IGBVF_DOWN, &adapter->state);
1680
1681         napi_enable(&adapter->rx_ring->napi);
1682
1683         /* clear any pending interrupts */
1684         er32(EICR);
1685
1686         igbvf_irq_enable(adapter);
1687
1688         /* start the watchdog */
1689         hw->mac.get_link_status = 1;
1690         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1691
1692         return 0;
1693
1694 err_req_irq:
1695         igbvf_free_rx_resources(adapter->rx_ring);
1696 err_setup_rx:
1697         igbvf_free_tx_resources(adapter->tx_ring);
1698 err_setup_tx:
1699         igbvf_reset(adapter);
1700
1701         return err;
1702 }
1703
1704 /**
1705  * igbvf_close - Disables a network interface
1706  * @netdev: network interface device structure
1707  *
1708  * Returns 0, this is not allowed to fail
1709  *
1710  * The close entry point is called when an interface is de-activated
1711  * by the OS.  The hardware is still under the drivers control, but
1712  * needs to be disabled.  A global MAC reset is issued to stop the
1713  * hardware, and all transmit and receive resources are freed.
1714  **/
1715 static int igbvf_close(struct net_device *netdev)
1716 {
1717         struct igbvf_adapter *adapter = netdev_priv(netdev);
1718
1719         WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1720         igbvf_down(adapter);
1721
1722         igbvf_free_irq(adapter);
1723
1724         igbvf_free_tx_resources(adapter->tx_ring);
1725         igbvf_free_rx_resources(adapter->rx_ring);
1726
1727         return 0;
1728 }
1729 /**
1730  * igbvf_set_mac - Change the Ethernet Address of the NIC
1731  * @netdev: network interface device structure
1732  * @p: pointer to an address structure
1733  *
1734  * Returns 0 on success, negative on failure
1735  **/
1736 static int igbvf_set_mac(struct net_device *netdev, void *p)
1737 {
1738         struct igbvf_adapter *adapter = netdev_priv(netdev);
1739         struct e1000_hw *hw = &adapter->hw;
1740         struct sockaddr *addr = p;
1741
1742         if (!is_valid_ether_addr(addr->sa_data))
1743                 return -EADDRNOTAVAIL;
1744
1745         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1746
1747         hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1748
1749         if (memcmp(addr->sa_data, hw->mac.addr, 6))
1750                 return -EADDRNOTAVAIL;
1751
1752         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1753
1754         return 0;
1755 }
1756
1757 #define UPDATE_VF_COUNTER(reg, name)                                    \
1758         {                                                               \
1759                 u32 current_counter = er32(reg);                        \
1760                 if (current_counter < adapter->stats.last_##name)       \
1761                         adapter->stats.name += 0x100000000LL;           \
1762                 adapter->stats.last_##name = current_counter;           \
1763                 adapter->stats.name &= 0xFFFFFFFF00000000LL;            \
1764                 adapter->stats.name |= current_counter;                 \
1765         }
1766
1767 /**
1768  * igbvf_update_stats - Update the board statistics counters
1769  * @adapter: board private structure
1770 **/
1771 void igbvf_update_stats(struct igbvf_adapter *adapter)
1772 {
1773         struct e1000_hw *hw = &adapter->hw;
1774         struct pci_dev *pdev = adapter->pdev;
1775
1776         /*
1777          * Prevent stats update while adapter is being reset, link is down
1778          * or if the pci connection is down.
1779          */
1780         if (adapter->link_speed == 0)
1781                 return;
1782
1783         if (test_bit(__IGBVF_RESETTING, &adapter->state))
1784                 return;
1785
1786         if (pci_channel_offline(pdev))
1787                 return;
1788
1789         UPDATE_VF_COUNTER(VFGPRC, gprc);
1790         UPDATE_VF_COUNTER(VFGORC, gorc);
1791         UPDATE_VF_COUNTER(VFGPTC, gptc);
1792         UPDATE_VF_COUNTER(VFGOTC, gotc);
1793         UPDATE_VF_COUNTER(VFMPRC, mprc);
1794         UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1795         UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1796         UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1797         UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1798
1799         /* Fill out the OS statistics structure */
1800         adapter->net_stats.multicast = adapter->stats.mprc;
1801 }
1802
1803 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1804 {
1805         dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s\n",
1806                  adapter->link_speed,
1807                  ((adapter->link_duplex == FULL_DUPLEX) ?
1808                   "Full Duplex" : "Half Duplex"));
1809 }
1810
1811 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1812 {
1813         struct e1000_hw *hw = &adapter->hw;
1814         s32 ret_val = E1000_SUCCESS;
1815         bool link_active;
1816
1817         /* If interface is down, stay link down */
1818         if (test_bit(__IGBVF_DOWN, &adapter->state))
1819                 return false;
1820
1821         ret_val = hw->mac.ops.check_for_link(hw);
1822         link_active = !hw->mac.get_link_status;
1823
1824         /* if check for link returns error we will need to reset */
1825         if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1826                 schedule_work(&adapter->reset_task);
1827
1828         return link_active;
1829 }
1830
1831 /**
1832  * igbvf_watchdog - Timer Call-back
1833  * @data: pointer to adapter cast into an unsigned long
1834  **/
1835 static void igbvf_watchdog(unsigned long data)
1836 {
1837         struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1838
1839         /* Do the rest outside of interrupt context */
1840         schedule_work(&adapter->watchdog_task);
1841 }
1842
1843 static void igbvf_watchdog_task(struct work_struct *work)
1844 {
1845         struct igbvf_adapter *adapter = container_of(work,
1846                                                      struct igbvf_adapter,
1847                                                      watchdog_task);
1848         struct net_device *netdev = adapter->netdev;
1849         struct e1000_mac_info *mac = &adapter->hw.mac;
1850         struct igbvf_ring *tx_ring = adapter->tx_ring;
1851         struct e1000_hw *hw = &adapter->hw;
1852         u32 link;
1853         int tx_pending = 0;
1854
1855         link = igbvf_has_link(adapter);
1856
1857         if (link) {
1858                 if (!netif_carrier_ok(netdev)) {
1859                         bool txb2b = 1;
1860
1861                         mac->ops.get_link_up_info(&adapter->hw,
1862                                                   &adapter->link_speed,
1863                                                   &adapter->link_duplex);
1864                         igbvf_print_link_info(adapter);
1865
1866                         /*
1867                          * tweak tx_queue_len according to speed/duplex
1868                          * and adjust the timeout factor
1869                          */
1870                         netdev->tx_queue_len = adapter->tx_queue_len;
1871                         adapter->tx_timeout_factor = 1;
1872                         switch (adapter->link_speed) {
1873                         case SPEED_10:
1874                                 txb2b = 0;
1875                                 netdev->tx_queue_len = 10;
1876                                 adapter->tx_timeout_factor = 16;
1877                                 break;
1878                         case SPEED_100:
1879                                 txb2b = 0;
1880                                 netdev->tx_queue_len = 100;
1881                                 /* maybe add some timeout factor ? */
1882                                 break;
1883                         }
1884
1885                         netif_carrier_on(netdev);
1886                         netif_wake_queue(netdev);
1887                 }
1888         } else {
1889                 if (netif_carrier_ok(netdev)) {
1890                         adapter->link_speed = 0;
1891                         adapter->link_duplex = 0;
1892                         dev_info(&adapter->pdev->dev, "Link is Down\n");
1893                         netif_carrier_off(netdev);
1894                         netif_stop_queue(netdev);
1895                 }
1896         }
1897
1898         if (netif_carrier_ok(netdev)) {
1899                 igbvf_update_stats(adapter);
1900         } else {
1901                 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1902                               tx_ring->count);
1903                 if (tx_pending) {
1904                         /*
1905                          * We've lost link, so the controller stops DMA,
1906                          * but we've got queued Tx work that's never going
1907                          * to get done, so reset controller to flush Tx.
1908                          * (Do the reset outside of interrupt context).
1909                          */
1910                         adapter->tx_timeout_count++;
1911                         schedule_work(&adapter->reset_task);
1912                 }
1913         }
1914
1915         /* Cause software interrupt to ensure Rx ring is cleaned */
1916         ew32(EICS, adapter->rx_ring->eims_value);
1917
1918         /* Force detection of hung controller every watchdog period */
1919         adapter->detect_tx_hung = 1;
1920
1921         /* Reset the timer */
1922         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1923                 mod_timer(&adapter->watchdog_timer,
1924                           round_jiffies(jiffies + (2 * HZ)));
1925 }
1926
1927 #define IGBVF_TX_FLAGS_CSUM             0x00000001
1928 #define IGBVF_TX_FLAGS_VLAN             0x00000002
1929 #define IGBVF_TX_FLAGS_TSO              0x00000004
1930 #define IGBVF_TX_FLAGS_IPV4             0x00000008
1931 #define IGBVF_TX_FLAGS_VLAN_MASK        0xffff0000
1932 #define IGBVF_TX_FLAGS_VLAN_SHIFT       16
1933
1934 static int igbvf_tso(struct igbvf_adapter *adapter,
1935                      struct igbvf_ring *tx_ring,
1936                      struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1937 {
1938         struct e1000_adv_tx_context_desc *context_desc;
1939         unsigned int i;
1940         int err;
1941         struct igbvf_buffer *buffer_info;
1942         u32 info = 0, tu_cmd = 0;
1943         u32 mss_l4len_idx, l4len;
1944         *hdr_len = 0;
1945
1946         if (skb_header_cloned(skb)) {
1947                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1948                 if (err) {
1949                         dev_err(&adapter->pdev->dev,
1950                                 "igbvf_tso returning an error\n");
1951                         return err;
1952                 }
1953         }
1954
1955         l4len = tcp_hdrlen(skb);
1956         *hdr_len += l4len;
1957
1958         if (skb->protocol == htons(ETH_P_IP)) {
1959                 struct iphdr *iph = ip_hdr(skb);
1960                 iph->tot_len = 0;
1961                 iph->check = 0;
1962                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1963                                                          iph->daddr, 0,
1964                                                          IPPROTO_TCP,
1965                                                          0);
1966         } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
1967                 ipv6_hdr(skb)->payload_len = 0;
1968                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1969                                                        &ipv6_hdr(skb)->daddr,
1970                                                        0, IPPROTO_TCP, 0);
1971         }
1972
1973         i = tx_ring->next_to_use;
1974
1975         buffer_info = &tx_ring->buffer_info[i];
1976         context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1977         /* VLAN MACLEN IPLEN */
1978         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1979                 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1980         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1981         *hdr_len += skb_network_offset(skb);
1982         info |= (skb_transport_header(skb) - skb_network_header(skb));
1983         *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1984         context_desc->vlan_macip_lens = cpu_to_le32(info);
1985
1986         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1987         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1988
1989         if (skb->protocol == htons(ETH_P_IP))
1990                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1991         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1992
1993         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1994
1995         /* MSS L4LEN IDX */
1996         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1997         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1998
1999         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
2000         context_desc->seqnum_seed = 0;
2001
2002         buffer_info->time_stamp = jiffies;
2003         buffer_info->next_to_watch = i;
2004         buffer_info->dma = 0;
2005         i++;
2006         if (i == tx_ring->count)
2007                 i = 0;
2008
2009         tx_ring->next_to_use = i;
2010
2011         return true;
2012 }
2013
2014 static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
2015                                  struct igbvf_ring *tx_ring,
2016                                  struct sk_buff *skb, u32 tx_flags)
2017 {
2018         struct e1000_adv_tx_context_desc *context_desc;
2019         unsigned int i;
2020         struct igbvf_buffer *buffer_info;
2021         u32 info = 0, tu_cmd = 0;
2022
2023         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2024             (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
2025                 i = tx_ring->next_to_use;
2026                 buffer_info = &tx_ring->buffer_info[i];
2027                 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
2028
2029                 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2030                         info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
2031
2032                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2033                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2034                         info |= (skb_transport_header(skb) -
2035                                  skb_network_header(skb));
2036
2037
2038                 context_desc->vlan_macip_lens = cpu_to_le32(info);
2039
2040                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2041
2042                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2043                         switch (skb->protocol) {
2044                         case __constant_htons(ETH_P_IP):
2045                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2046                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2047                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2048                                 break;
2049                         case __constant_htons(ETH_P_IPV6):
2050                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2051                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2052                                 break;
2053                         default:
2054                                 break;
2055                         }
2056                 }
2057
2058                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2059                 context_desc->seqnum_seed = 0;
2060                 context_desc->mss_l4len_idx = 0;
2061
2062                 buffer_info->time_stamp = jiffies;
2063                 buffer_info->next_to_watch = i;
2064                 buffer_info->dma = 0;
2065                 i++;
2066                 if (i == tx_ring->count)
2067                         i = 0;
2068                 tx_ring->next_to_use = i;
2069
2070                 return true;
2071         }
2072
2073         return false;
2074 }
2075
2076 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2077 {
2078         struct igbvf_adapter *adapter = netdev_priv(netdev);
2079
2080         /* there is enough descriptors then we don't need to worry  */
2081         if (igbvf_desc_unused(adapter->tx_ring) >= size)
2082                 return 0;
2083
2084         netif_stop_queue(netdev);
2085
2086         smp_mb();
2087
2088         /* We need to check again just in case room has been made available */
2089         if (igbvf_desc_unused(adapter->tx_ring) < size)
2090                 return -EBUSY;
2091
2092         netif_wake_queue(netdev);
2093
2094         ++adapter->restart_queue;
2095         return 0;
2096 }
2097
2098 #define IGBVF_MAX_TXD_PWR       16
2099 #define IGBVF_MAX_DATA_PER_TXD  (1 << IGBVF_MAX_TXD_PWR)
2100
2101 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2102                                    struct igbvf_ring *tx_ring,
2103                                    struct sk_buff *skb,
2104                                    unsigned int first)
2105 {
2106         struct igbvf_buffer *buffer_info;
2107         struct pci_dev *pdev = adapter->pdev;
2108         unsigned int len = skb_headlen(skb);
2109         unsigned int count = 0, i;
2110         unsigned int f;
2111
2112         i = tx_ring->next_to_use;
2113
2114         buffer_info = &tx_ring->buffer_info[i];
2115         BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2116         buffer_info->length = len;
2117         /* set time_stamp *before* dma to help avoid a possible race */
2118         buffer_info->time_stamp = jiffies;
2119         buffer_info->next_to_watch = i;
2120         buffer_info->dma = pci_map_single(pdev, skb->data, len,
2121                                           PCI_DMA_TODEVICE);
2122         if (pci_dma_mapping_error(pdev, buffer_info->dma))
2123                 goto dma_error;
2124
2125
2126         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2127                 struct skb_frag_struct *frag;
2128
2129                 i++;
2130                 if (i == tx_ring->count)
2131                         i = 0;
2132
2133                 frag = &skb_shinfo(skb)->frags[f];
2134                 len = frag->size;
2135
2136                 buffer_info = &tx_ring->buffer_info[i];
2137                 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2138                 buffer_info->length = len;
2139                 buffer_info->time_stamp = jiffies;
2140                 buffer_info->next_to_watch = i;
2141                 buffer_info->mapped_as_page = true;
2142                 buffer_info->dma = pci_map_page(pdev,
2143                                                 frag->page,
2144                                                 frag->page_offset,
2145                                                 len,
2146                                                 PCI_DMA_TODEVICE);
2147                 if (pci_dma_mapping_error(pdev, buffer_info->dma))
2148                         goto dma_error;
2149                 count++;
2150         }
2151
2152         tx_ring->buffer_info[i].skb = skb;
2153         tx_ring->buffer_info[first].next_to_watch = i;
2154
2155         return ++count;
2156
2157 dma_error:
2158         dev_err(&pdev->dev, "TX DMA map failed\n");
2159
2160         /* clear timestamp and dma mappings for failed buffer_info mapping */
2161         buffer_info->dma = 0;
2162         buffer_info->time_stamp = 0;
2163         buffer_info->length = 0;
2164         buffer_info->next_to_watch = 0;
2165         buffer_info->mapped_as_page = false;
2166         if (count)
2167                 count--;
2168
2169         /* clear timestamp and dma mappings for remaining portion of packet */
2170         while (count--) {
2171                 if (i==0)
2172                         i += tx_ring->count;
2173                 i--;
2174                 buffer_info = &tx_ring->buffer_info[i];
2175                 igbvf_put_txbuf(adapter, buffer_info);
2176         }
2177
2178         return 0;
2179 }
2180
2181 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2182                                       struct igbvf_ring *tx_ring,
2183                                       int tx_flags, int count, u32 paylen,
2184                                       u8 hdr_len)
2185 {
2186         union e1000_adv_tx_desc *tx_desc = NULL;
2187         struct igbvf_buffer *buffer_info;
2188         u32 olinfo_status = 0, cmd_type_len;
2189         unsigned int i;
2190
2191         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2192                         E1000_ADVTXD_DCMD_DEXT);
2193
2194         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2195                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2196
2197         if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2198                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2199
2200                 /* insert tcp checksum */
2201                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2202
2203                 /* insert ip checksum */
2204                 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2205                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2206
2207         } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2208                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2209         }
2210
2211         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2212
2213         i = tx_ring->next_to_use;
2214         while (count--) {
2215                 buffer_info = &tx_ring->buffer_info[i];
2216                 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2217                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2218                 tx_desc->read.cmd_type_len =
2219                          cpu_to_le32(cmd_type_len | buffer_info->length);
2220                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2221                 i++;
2222                 if (i == tx_ring->count)
2223                         i = 0;
2224         }
2225
2226         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2227         /* Force memory writes to complete before letting h/w
2228          * know there are new descriptors to fetch.  (Only
2229          * applicable for weak-ordered memory model archs,
2230          * such as IA-64). */
2231         wmb();
2232
2233         tx_ring->next_to_use = i;
2234         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2235         /* we need this if more than one processor can write to our tail
2236          * at a time, it syncronizes IO on IA64/Altix systems */
2237         mmiowb();
2238 }
2239
2240 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2241                                              struct net_device *netdev,
2242                                              struct igbvf_ring *tx_ring)
2243 {
2244         struct igbvf_adapter *adapter = netdev_priv(netdev);
2245         unsigned int first, tx_flags = 0;
2246         u8 hdr_len = 0;
2247         int count = 0;
2248         int tso = 0;
2249
2250         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2251                 dev_kfree_skb_any(skb);
2252                 return NETDEV_TX_OK;
2253         }
2254
2255         if (skb->len <= 0) {
2256                 dev_kfree_skb_any(skb);
2257                 return NETDEV_TX_OK;
2258         }
2259
2260         /*
2261          * need: count + 4 desc gap to keep tail from touching
2262          *       + 2 desc gap to keep tail from touching head,
2263          *       + 1 desc for skb->data,
2264          *       + 1 desc for context descriptor,
2265          * head, otherwise try next time
2266          */
2267         if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2268                 /* this is a hard error */
2269                 return NETDEV_TX_BUSY;
2270         }
2271
2272         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2273                 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2274                 tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2275         }
2276
2277         if (skb->protocol == htons(ETH_P_IP))
2278                 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2279
2280         first = tx_ring->next_to_use;
2281
2282         tso = skb_is_gso(skb) ?
2283                 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2284         if (unlikely(tso < 0)) {
2285                 dev_kfree_skb_any(skb);
2286                 return NETDEV_TX_OK;
2287         }
2288
2289         if (tso)
2290                 tx_flags |= IGBVF_TX_FLAGS_TSO;
2291         else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2292                  (skb->ip_summed == CHECKSUM_PARTIAL))
2293                 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2294
2295         /*
2296          * count reflects descriptors mapped, if 0 then mapping error
2297          * has occured and we need to rewind the descriptor queue
2298          */
2299         count = igbvf_tx_map_adv(adapter, tx_ring, skb, first);
2300
2301         if (count) {
2302                 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2303                                    skb->len, hdr_len);
2304                 /* Make sure there is space in the ring for the next send. */
2305                 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2306         } else {
2307                 dev_kfree_skb_any(skb);
2308                 tx_ring->buffer_info[first].time_stamp = 0;
2309                 tx_ring->next_to_use = first;
2310         }
2311
2312         return NETDEV_TX_OK;
2313 }
2314
2315 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2316                                     struct net_device *netdev)
2317 {
2318         struct igbvf_adapter *adapter = netdev_priv(netdev);
2319         struct igbvf_ring *tx_ring;
2320
2321         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2322                 dev_kfree_skb_any(skb);
2323                 return NETDEV_TX_OK;
2324         }
2325
2326         tx_ring = &adapter->tx_ring[0];
2327
2328         return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2329 }
2330
2331 /**
2332  * igbvf_tx_timeout - Respond to a Tx Hang
2333  * @netdev: network interface device structure
2334  **/
2335 static void igbvf_tx_timeout(struct net_device *netdev)
2336 {
2337         struct igbvf_adapter *adapter = netdev_priv(netdev);
2338
2339         /* Do the reset outside of interrupt context */
2340         adapter->tx_timeout_count++;
2341         schedule_work(&adapter->reset_task);
2342 }
2343
2344 static void igbvf_reset_task(struct work_struct *work)
2345 {
2346         struct igbvf_adapter *adapter;
2347         adapter = container_of(work, struct igbvf_adapter, reset_task);
2348
2349         igbvf_reinit_locked(adapter);
2350 }
2351
2352 /**
2353  * igbvf_get_stats - Get System Network Statistics
2354  * @netdev: network interface device structure
2355  *
2356  * Returns the address of the device statistics structure.
2357  * The statistics are actually updated from the timer callback.
2358  **/
2359 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2360 {
2361         struct igbvf_adapter *adapter = netdev_priv(netdev);
2362
2363         /* only return the current stats */
2364         return &adapter->net_stats;
2365 }
2366
2367 /**
2368  * igbvf_change_mtu - Change the Maximum Transfer Unit
2369  * @netdev: network interface device structure
2370  * @new_mtu: new value for maximum frame size
2371  *
2372  * Returns 0 on success, negative on failure
2373  **/
2374 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2375 {
2376         struct igbvf_adapter *adapter = netdev_priv(netdev);
2377         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2378
2379         if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2380                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2381                 return -EINVAL;
2382         }
2383
2384 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2385         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2386                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2387                 return -EINVAL;
2388         }
2389
2390         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2391                 msleep(1);
2392         /* igbvf_down has a dependency on max_frame_size */
2393         adapter->max_frame_size = max_frame;
2394         if (netif_running(netdev))
2395                 igbvf_down(adapter);
2396
2397         /*
2398          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2399          * means we reserve 2 more, this pushes us to allocate from the next
2400          * larger slab size.
2401          * i.e. RXBUFFER_2048 --> size-4096 slab
2402          * However with the new *_jumbo_rx* routines, jumbo receives will use
2403          * fragmented skbs
2404          */
2405
2406         if (max_frame <= 1024)
2407                 adapter->rx_buffer_len = 1024;
2408         else if (max_frame <= 2048)
2409                 adapter->rx_buffer_len = 2048;
2410         else
2411 #if (PAGE_SIZE / 2) > 16384
2412                 adapter->rx_buffer_len = 16384;
2413 #else
2414                 adapter->rx_buffer_len = PAGE_SIZE / 2;
2415 #endif
2416
2417
2418         /* adjust allocation if LPE protects us, and we aren't using SBP */
2419         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2420              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2421                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2422                                          ETH_FCS_LEN;
2423
2424         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2425                  netdev->mtu, new_mtu);
2426         netdev->mtu = new_mtu;
2427
2428         if (netif_running(netdev))
2429                 igbvf_up(adapter);
2430         else
2431                 igbvf_reset(adapter);
2432
2433         clear_bit(__IGBVF_RESETTING, &adapter->state);
2434
2435         return 0;
2436 }
2437
2438 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2439 {
2440         switch (cmd) {
2441         default:
2442                 return -EOPNOTSUPP;
2443         }
2444 }
2445
2446 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2447 {
2448         struct net_device *netdev = pci_get_drvdata(pdev);
2449         struct igbvf_adapter *adapter = netdev_priv(netdev);
2450 #ifdef CONFIG_PM
2451         int retval = 0;
2452 #endif
2453
2454         netif_device_detach(netdev);
2455
2456         if (netif_running(netdev)) {
2457                 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2458                 igbvf_down(adapter);
2459                 igbvf_free_irq(adapter);
2460         }
2461
2462 #ifdef CONFIG_PM
2463         retval = pci_save_state(pdev);
2464         if (retval)
2465                 return retval;
2466 #endif
2467
2468         pci_disable_device(pdev);
2469
2470         return 0;
2471 }
2472
2473 #ifdef CONFIG_PM
2474 static int igbvf_resume(struct pci_dev *pdev)
2475 {
2476         struct net_device *netdev = pci_get_drvdata(pdev);
2477         struct igbvf_adapter *adapter = netdev_priv(netdev);
2478         u32 err;
2479
2480         pci_restore_state(pdev);
2481         err = pci_enable_device_mem(pdev);
2482         if (err) {
2483                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2484                 return err;
2485         }
2486
2487         pci_set_master(pdev);
2488
2489         if (netif_running(netdev)) {
2490                 err = igbvf_request_irq(adapter);
2491                 if (err)
2492                         return err;
2493         }
2494
2495         igbvf_reset(adapter);
2496
2497         if (netif_running(netdev))
2498                 igbvf_up(adapter);
2499
2500         netif_device_attach(netdev);
2501
2502         return 0;
2503 }
2504 #endif
2505
2506 static void igbvf_shutdown(struct pci_dev *pdev)
2507 {
2508         igbvf_suspend(pdev, PMSG_SUSPEND);
2509 }
2510
2511 #ifdef CONFIG_NET_POLL_CONTROLLER
2512 /*
2513  * Polling 'interrupt' - used by things like netconsole to send skbs
2514  * without having to re-enable interrupts. It's not called while
2515  * the interrupt routine is executing.
2516  */
2517 static void igbvf_netpoll(struct net_device *netdev)
2518 {
2519         struct igbvf_adapter *adapter = netdev_priv(netdev);
2520
2521         disable_irq(adapter->pdev->irq);
2522
2523         igbvf_clean_tx_irq(adapter->tx_ring);
2524
2525         enable_irq(adapter->pdev->irq);
2526 }
2527 #endif
2528
2529 /**
2530  * igbvf_io_error_detected - called when PCI error is detected
2531  * @pdev: Pointer to PCI device
2532  * @state: The current pci connection state
2533  *
2534  * This function is called after a PCI bus error affecting
2535  * this device has been detected.
2536  */
2537 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2538                                                 pci_channel_state_t state)
2539 {
2540         struct net_device *netdev = pci_get_drvdata(pdev);
2541         struct igbvf_adapter *adapter = netdev_priv(netdev);
2542
2543         netif_device_detach(netdev);
2544
2545         if (state == pci_channel_io_perm_failure)
2546                 return PCI_ERS_RESULT_DISCONNECT;
2547
2548         if (netif_running(netdev))
2549                 igbvf_down(adapter);
2550         pci_disable_device(pdev);
2551
2552         /* Request a slot slot reset. */
2553         return PCI_ERS_RESULT_NEED_RESET;
2554 }
2555
2556 /**
2557  * igbvf_io_slot_reset - called after the pci bus has been reset.
2558  * @pdev: Pointer to PCI device
2559  *
2560  * Restart the card from scratch, as if from a cold-boot. Implementation
2561  * resembles the first-half of the igbvf_resume routine.
2562  */
2563 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2564 {
2565         struct net_device *netdev = pci_get_drvdata(pdev);
2566         struct igbvf_adapter *adapter = netdev_priv(netdev);
2567
2568         if (pci_enable_device_mem(pdev)) {
2569                 dev_err(&pdev->dev,
2570                         "Cannot re-enable PCI device after reset.\n");
2571                 return PCI_ERS_RESULT_DISCONNECT;
2572         }
2573         pci_set_master(pdev);
2574
2575         igbvf_reset(adapter);
2576
2577         return PCI_ERS_RESULT_RECOVERED;
2578 }
2579
2580 /**
2581  * igbvf_io_resume - called when traffic can start flowing again.
2582  * @pdev: Pointer to PCI device
2583  *
2584  * This callback is called when the error recovery driver tells us that
2585  * its OK to resume normal operation. Implementation resembles the
2586  * second-half of the igbvf_resume routine.
2587  */
2588 static void igbvf_io_resume(struct pci_dev *pdev)
2589 {
2590         struct net_device *netdev = pci_get_drvdata(pdev);
2591         struct igbvf_adapter *adapter = netdev_priv(netdev);
2592
2593         if (netif_running(netdev)) {
2594                 if (igbvf_up(adapter)) {
2595                         dev_err(&pdev->dev,
2596                                 "can't bring device back up after reset\n");
2597                         return;
2598                 }
2599         }
2600
2601         netif_device_attach(netdev);
2602 }
2603
2604 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2605 {
2606         struct e1000_hw *hw = &adapter->hw;
2607         struct net_device *netdev = adapter->netdev;
2608         struct pci_dev *pdev = adapter->pdev;
2609
2610         dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2611         dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2612         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
2613 }
2614
2615 static const struct net_device_ops igbvf_netdev_ops = {
2616         .ndo_open                       = igbvf_open,
2617         .ndo_stop                       = igbvf_close,
2618         .ndo_start_xmit                 = igbvf_xmit_frame,
2619         .ndo_get_stats                  = igbvf_get_stats,
2620         .ndo_set_multicast_list         = igbvf_set_multi,
2621         .ndo_set_mac_address            = igbvf_set_mac,
2622         .ndo_change_mtu                 = igbvf_change_mtu,
2623         .ndo_do_ioctl                   = igbvf_ioctl,
2624         .ndo_tx_timeout                 = igbvf_tx_timeout,
2625         .ndo_vlan_rx_register           = igbvf_vlan_rx_register,
2626         .ndo_vlan_rx_add_vid            = igbvf_vlan_rx_add_vid,
2627         .ndo_vlan_rx_kill_vid           = igbvf_vlan_rx_kill_vid,
2628 #ifdef CONFIG_NET_POLL_CONTROLLER
2629         .ndo_poll_controller            = igbvf_netpoll,
2630 #endif
2631 };
2632
2633 /**
2634  * igbvf_probe - Device Initialization Routine
2635  * @pdev: PCI device information struct
2636  * @ent: entry in igbvf_pci_tbl
2637  *
2638  * Returns 0 on success, negative on failure
2639  *
2640  * igbvf_probe initializes an adapter identified by a pci_dev structure.
2641  * The OS initialization, configuring of the adapter private structure,
2642  * and a hardware reset occur.
2643  **/
2644 static int __devinit igbvf_probe(struct pci_dev *pdev,
2645                                  const struct pci_device_id *ent)
2646 {
2647         struct net_device *netdev;
2648         struct igbvf_adapter *adapter;
2649         struct e1000_hw *hw;
2650         const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2651
2652         static int cards_found;
2653         int err, pci_using_dac;
2654
2655         err = pci_enable_device_mem(pdev);
2656         if (err)
2657                 return err;
2658
2659         pci_using_dac = 0;
2660         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2661         if (!err) {
2662                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2663                 if (!err)
2664                         pci_using_dac = 1;
2665         } else {
2666                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2667                 if (err) {
2668                         err = pci_set_consistent_dma_mask(pdev,
2669                                                           DMA_BIT_MASK(32));
2670                         if (err) {
2671                                 dev_err(&pdev->dev, "No usable DMA "
2672                                         "configuration, aborting\n");
2673                                 goto err_dma;
2674                         }
2675                 }
2676         }
2677
2678         err = pci_request_regions(pdev, igbvf_driver_name);
2679         if (err)
2680                 goto err_pci_reg;
2681
2682         pci_set_master(pdev);
2683
2684         err = -ENOMEM;
2685         netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2686         if (!netdev)
2687                 goto err_alloc_etherdev;
2688
2689         SET_NETDEV_DEV(netdev, &pdev->dev);
2690
2691         pci_set_drvdata(pdev, netdev);
2692         adapter = netdev_priv(netdev);
2693         hw = &adapter->hw;
2694         adapter->netdev = netdev;
2695         adapter->pdev = pdev;
2696         adapter->ei = ei;
2697         adapter->pba = ei->pba;
2698         adapter->flags = ei->flags;
2699         adapter->hw.back = adapter;
2700         adapter->hw.mac.type = ei->mac;
2701         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
2702
2703         /* PCI config space info */
2704
2705         hw->vendor_id = pdev->vendor;
2706         hw->device_id = pdev->device;
2707         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2708         hw->subsystem_device_id = pdev->subsystem_device;
2709
2710         pci_read_config_byte(pdev, PCI_REVISION_ID, &hw->revision_id);
2711
2712         err = -EIO;
2713         adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2714                                       pci_resource_len(pdev, 0));
2715
2716         if (!adapter->hw.hw_addr)
2717                 goto err_ioremap;
2718
2719         if (ei->get_variants) {
2720                 err = ei->get_variants(adapter);
2721                 if (err)
2722                         goto err_ioremap;
2723         }
2724
2725         /* setup adapter struct */
2726         err = igbvf_sw_init(adapter);
2727         if (err)
2728                 goto err_sw_init;
2729
2730         /* construct the net_device struct */
2731         netdev->netdev_ops = &igbvf_netdev_ops;
2732
2733         igbvf_set_ethtool_ops(netdev);
2734         netdev->watchdog_timeo = 5 * HZ;
2735         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2736
2737         adapter->bd_number = cards_found++;
2738
2739         netdev->features = NETIF_F_SG |
2740                            NETIF_F_IP_CSUM |
2741                            NETIF_F_HW_VLAN_TX |
2742                            NETIF_F_HW_VLAN_RX |
2743                            NETIF_F_HW_VLAN_FILTER;
2744
2745         netdev->features |= NETIF_F_IPV6_CSUM;
2746         netdev->features |= NETIF_F_TSO;
2747         netdev->features |= NETIF_F_TSO6;
2748
2749         if (pci_using_dac)
2750                 netdev->features |= NETIF_F_HIGHDMA;
2751
2752         netdev->vlan_features |= NETIF_F_TSO;
2753         netdev->vlan_features |= NETIF_F_TSO6;
2754         netdev->vlan_features |= NETIF_F_IP_CSUM;
2755         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2756         netdev->vlan_features |= NETIF_F_SG;
2757
2758         /*reset the controller to put the device in a known good state */
2759         err = hw->mac.ops.reset_hw(hw);
2760         if (err) {
2761                 dev_info(&pdev->dev,
2762                          "PF still in reset state, assigning new address."
2763                          " Is the PF interface up?\n");
2764                 random_ether_addr(hw->mac.addr);
2765         } else {
2766                 err = hw->mac.ops.read_mac_addr(hw);
2767                 if (err) {
2768                         dev_err(&pdev->dev, "Error reading MAC address\n");
2769                         goto err_hw_init;
2770                 }
2771         }
2772
2773         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
2774         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
2775
2776         if (!is_valid_ether_addr(netdev->perm_addr)) {
2777                 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
2778                         netdev->dev_addr);
2779                 err = -EIO;
2780                 goto err_hw_init;
2781         }
2782
2783         setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2784                     (unsigned long) adapter);
2785
2786         INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2787         INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2788
2789         /* ring size defaults */
2790         adapter->rx_ring->count = 1024;
2791         adapter->tx_ring->count = 1024;
2792
2793         /* reset the hardware with the new settings */
2794         igbvf_reset(adapter);
2795
2796         /* tell the stack to leave us alone until igbvf_open() is called */
2797         netif_carrier_off(netdev);
2798         netif_stop_queue(netdev);
2799
2800         strcpy(netdev->name, "eth%d");
2801         err = register_netdev(netdev);
2802         if (err)
2803                 goto err_hw_init;
2804
2805         igbvf_print_device_info(adapter);
2806
2807         igbvf_initialize_last_counter_stats(adapter);
2808
2809         return 0;
2810
2811 err_hw_init:
2812         kfree(adapter->tx_ring);
2813         kfree(adapter->rx_ring);
2814 err_sw_init:
2815         igbvf_reset_interrupt_capability(adapter);
2816         iounmap(adapter->hw.hw_addr);
2817 err_ioremap:
2818         free_netdev(netdev);
2819 err_alloc_etherdev:
2820         pci_release_regions(pdev);
2821 err_pci_reg:
2822 err_dma:
2823         pci_disable_device(pdev);
2824         return err;
2825 }
2826
2827 /**
2828  * igbvf_remove - Device Removal Routine
2829  * @pdev: PCI device information struct
2830  *
2831  * igbvf_remove is called by the PCI subsystem to alert the driver
2832  * that it should release a PCI device.  The could be caused by a
2833  * Hot-Plug event, or because the driver is going to be removed from
2834  * memory.
2835  **/
2836 static void __devexit igbvf_remove(struct pci_dev *pdev)
2837 {
2838         struct net_device *netdev = pci_get_drvdata(pdev);
2839         struct igbvf_adapter *adapter = netdev_priv(netdev);
2840         struct e1000_hw *hw = &adapter->hw;
2841
2842         /*
2843          * flush_scheduled work may reschedule our watchdog task, so
2844          * explicitly disable watchdog tasks from being rescheduled
2845          */
2846         set_bit(__IGBVF_DOWN, &adapter->state);
2847         del_timer_sync(&adapter->watchdog_timer);
2848
2849         flush_scheduled_work();
2850
2851         unregister_netdev(netdev);
2852
2853         igbvf_reset_interrupt_capability(adapter);
2854
2855         /*
2856          * it is important to delete the napi struct prior to freeing the
2857          * rx ring so that you do not end up with null pointer refs
2858          */
2859         netif_napi_del(&adapter->rx_ring->napi);
2860         kfree(adapter->tx_ring);
2861         kfree(adapter->rx_ring);
2862
2863         iounmap(hw->hw_addr);
2864         if (hw->flash_address)
2865                 iounmap(hw->flash_address);
2866         pci_release_regions(pdev);
2867
2868         free_netdev(netdev);
2869
2870         pci_disable_device(pdev);
2871 }
2872
2873 /* PCI Error Recovery (ERS) */
2874 static struct pci_error_handlers igbvf_err_handler = {
2875         .error_detected = igbvf_io_error_detected,
2876         .slot_reset = igbvf_io_slot_reset,
2877         .resume = igbvf_io_resume,
2878 };
2879
2880 static DEFINE_PCI_DEVICE_TABLE(igbvf_pci_tbl) = {
2881         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2882         { } /* terminate list */
2883 };
2884 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2885
2886 /* PCI Device API Driver */
2887 static struct pci_driver igbvf_driver = {
2888         .name     = igbvf_driver_name,
2889         .id_table = igbvf_pci_tbl,
2890         .probe    = igbvf_probe,
2891         .remove   = __devexit_p(igbvf_remove),
2892 #ifdef CONFIG_PM
2893         /* Power Management Hooks */
2894         .suspend  = igbvf_suspend,
2895         .resume   = igbvf_resume,
2896 #endif
2897         .shutdown = igbvf_shutdown,
2898         .err_handler = &igbvf_err_handler
2899 };
2900
2901 /**
2902  * igbvf_init_module - Driver Registration Routine
2903  *
2904  * igbvf_init_module is the first routine called when the driver is
2905  * loaded. All it does is register with the PCI subsystem.
2906  **/
2907 static int __init igbvf_init_module(void)
2908 {
2909         int ret;
2910         printk(KERN_INFO "%s - version %s\n",
2911                igbvf_driver_string, igbvf_driver_version);
2912         printk(KERN_INFO "%s\n", igbvf_copyright);
2913
2914         ret = pci_register_driver(&igbvf_driver);
2915         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, igbvf_driver_name,
2916                                PM_QOS_DEFAULT_VALUE);
2917
2918         return ret;
2919 }
2920 module_init(igbvf_init_module);
2921
2922 /**
2923  * igbvf_exit_module - Driver Exit Cleanup Routine
2924  *
2925  * igbvf_exit_module is called just before the driver is removed
2926  * from memory.
2927  **/
2928 static void __exit igbvf_exit_module(void)
2929 {
2930         pci_unregister_driver(&igbvf_driver);
2931         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, igbvf_driver_name);
2932 }
2933 module_exit(igbvf_exit_module);
2934
2935
2936 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2937 MODULE_DESCRIPTION("Intel(R) 82576 Virtual Function Network Driver");
2938 MODULE_LICENSE("GPL");
2939 MODULE_VERSION(DRV_VERSION);
2940
2941 /* netdev.c */