net: convert multicast list to list_head
[safe/jmp/linux-2.6] / drivers / net / e1000e / netdev.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 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   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30
31 #include <linux/module.h>
32 #include <linux/types.h>
33 #include <linux/init.h>
34 #include <linux/pci.h>
35 #include <linux/vmalloc.h>
36 #include <linux/pagemap.h>
37 #include <linux/delay.h>
38 #include <linux/netdevice.h>
39 #include <linux/tcp.h>
40 #include <linux/ipv6.h>
41 #include <net/checksum.h>
42 #include <net/ip6_checksum.h>
43 #include <linux/mii.h>
44 #include <linux/ethtool.h>
45 #include <linux/if_vlan.h>
46 #include <linux/cpu.h>
47 #include <linux/smp.h>
48 #include <linux/pm_qos_params.h>
49 #include <linux/pm_runtime.h>
50 #include <linux/aer.h>
51
52 #include "e1000.h"
53
54 #define DRV_VERSION "1.0.2-k2"
55 char e1000e_driver_name[] = "e1000e";
56 const char e1000e_driver_version[] = DRV_VERSION;
57
58 static const struct e1000_info *e1000_info_tbl[] = {
59         [board_82571]           = &e1000_82571_info,
60         [board_82572]           = &e1000_82572_info,
61         [board_82573]           = &e1000_82573_info,
62         [board_82574]           = &e1000_82574_info,
63         [board_82583]           = &e1000_82583_info,
64         [board_80003es2lan]     = &e1000_es2_info,
65         [board_ich8lan]         = &e1000_ich8_info,
66         [board_ich9lan]         = &e1000_ich9_info,
67         [board_ich10lan]        = &e1000_ich10_info,
68         [board_pchlan]          = &e1000_pch_info,
69 };
70
71 /**
72  * e1000_desc_unused - calculate if we have unused descriptors
73  **/
74 static int e1000_desc_unused(struct e1000_ring *ring)
75 {
76         if (ring->next_to_clean > ring->next_to_use)
77                 return ring->next_to_clean - ring->next_to_use - 1;
78
79         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
80 }
81
82 /**
83  * e1000_receive_skb - helper function to handle Rx indications
84  * @adapter: board private structure
85  * @status: descriptor status field as written by hardware
86  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
87  * @skb: pointer to sk_buff to be indicated to stack
88  **/
89 static void e1000_receive_skb(struct e1000_adapter *adapter,
90                               struct net_device *netdev,
91                               struct sk_buff *skb,
92                               u8 status, __le16 vlan)
93 {
94         skb->protocol = eth_type_trans(skb, netdev);
95
96         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
97                 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
98                                  le16_to_cpu(vlan), skb);
99         else
100                 napi_gro_receive(&adapter->napi, skb);
101 }
102
103 /**
104  * e1000_rx_checksum - Receive Checksum Offload for 82543
105  * @adapter:     board private structure
106  * @status_err:  receive descriptor status and error fields
107  * @csum:       receive descriptor csum field
108  * @sk_buff:     socket buffer with received data
109  **/
110 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
111                               u32 csum, struct sk_buff *skb)
112 {
113         u16 status = (u16)status_err;
114         u8 errors = (u8)(status_err >> 24);
115         skb->ip_summed = CHECKSUM_NONE;
116
117         /* Ignore Checksum bit is set */
118         if (status & E1000_RXD_STAT_IXSM)
119                 return;
120         /* TCP/UDP checksum error bit is set */
121         if (errors & E1000_RXD_ERR_TCPE) {
122                 /* let the stack verify checksum errors */
123                 adapter->hw_csum_err++;
124                 return;
125         }
126
127         /* TCP/UDP Checksum has not been calculated */
128         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
129                 return;
130
131         /* It must be a TCP or UDP packet with a valid checksum */
132         if (status & E1000_RXD_STAT_TCPCS) {
133                 /* TCP checksum is good */
134                 skb->ip_summed = CHECKSUM_UNNECESSARY;
135         } else {
136                 /*
137                  * IP fragment with UDP payload
138                  * Hardware complements the payload checksum, so we undo it
139                  * and then put the value in host order for further stack use.
140                  */
141                 __sum16 sum = (__force __sum16)htons(csum);
142                 skb->csum = csum_unfold(~sum);
143                 skb->ip_summed = CHECKSUM_COMPLETE;
144         }
145         adapter->hw_csum_good++;
146 }
147
148 /**
149  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
150  * @adapter: address of board private structure
151  **/
152 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
153                                    int cleaned_count)
154 {
155         struct net_device *netdev = adapter->netdev;
156         struct pci_dev *pdev = adapter->pdev;
157         struct e1000_ring *rx_ring = adapter->rx_ring;
158         struct e1000_rx_desc *rx_desc;
159         struct e1000_buffer *buffer_info;
160         struct sk_buff *skb;
161         unsigned int i;
162         unsigned int bufsz = adapter->rx_buffer_len;
163
164         i = rx_ring->next_to_use;
165         buffer_info = &rx_ring->buffer_info[i];
166
167         while (cleaned_count--) {
168                 skb = buffer_info->skb;
169                 if (skb) {
170                         skb_trim(skb, 0);
171                         goto map_skb;
172                 }
173
174                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
175                 if (!skb) {
176                         /* Better luck next round */
177                         adapter->alloc_rx_buff_failed++;
178                         break;
179                 }
180
181                 buffer_info->skb = skb;
182 map_skb:
183                 buffer_info->dma = pci_map_single(pdev, skb->data,
184                                                   adapter->rx_buffer_len,
185                                                   PCI_DMA_FROMDEVICE);
186                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
187                         dev_err(&pdev->dev, "RX DMA map failed\n");
188                         adapter->rx_dma_failed++;
189                         break;
190                 }
191
192                 rx_desc = E1000_RX_DESC(*rx_ring, i);
193                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
194
195                 i++;
196                 if (i == rx_ring->count)
197                         i = 0;
198                 buffer_info = &rx_ring->buffer_info[i];
199         }
200
201         if (rx_ring->next_to_use != i) {
202                 rx_ring->next_to_use = i;
203                 if (i-- == 0)
204                         i = (rx_ring->count - 1);
205
206                 /*
207                  * Force memory writes to complete before letting h/w
208                  * know there are new descriptors to fetch.  (Only
209                  * applicable for weak-ordered memory model archs,
210                  * such as IA-64).
211                  */
212                 wmb();
213                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
214         }
215 }
216
217 /**
218  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
219  * @adapter: address of board private structure
220  **/
221 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
222                                       int cleaned_count)
223 {
224         struct net_device *netdev = adapter->netdev;
225         struct pci_dev *pdev = adapter->pdev;
226         union e1000_rx_desc_packet_split *rx_desc;
227         struct e1000_ring *rx_ring = adapter->rx_ring;
228         struct e1000_buffer *buffer_info;
229         struct e1000_ps_page *ps_page;
230         struct sk_buff *skb;
231         unsigned int i, j;
232
233         i = rx_ring->next_to_use;
234         buffer_info = &rx_ring->buffer_info[i];
235
236         while (cleaned_count--) {
237                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
238
239                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
240                         ps_page = &buffer_info->ps_pages[j];
241                         if (j >= adapter->rx_ps_pages) {
242                                 /* all unused desc entries get hw null ptr */
243                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
244                                 continue;
245                         }
246                         if (!ps_page->page) {
247                                 ps_page->page = alloc_page(GFP_ATOMIC);
248                                 if (!ps_page->page) {
249                                         adapter->alloc_rx_buff_failed++;
250                                         goto no_buffers;
251                                 }
252                                 ps_page->dma = pci_map_page(pdev,
253                                                    ps_page->page,
254                                                    0, PAGE_SIZE,
255                                                    PCI_DMA_FROMDEVICE);
256                                 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
257                                         dev_err(&adapter->pdev->dev,
258                                           "RX DMA page map failed\n");
259                                         adapter->rx_dma_failed++;
260                                         goto no_buffers;
261                                 }
262                         }
263                         /*
264                          * Refresh the desc even if buffer_addrs
265                          * didn't change because each write-back
266                          * erases this info.
267                          */
268                         rx_desc->read.buffer_addr[j+1] =
269                              cpu_to_le64(ps_page->dma);
270                 }
271
272                 skb = netdev_alloc_skb_ip_align(netdev,
273                                                 adapter->rx_ps_bsize0);
274
275                 if (!skb) {
276                         adapter->alloc_rx_buff_failed++;
277                         break;
278                 }
279
280                 buffer_info->skb = skb;
281                 buffer_info->dma = pci_map_single(pdev, skb->data,
282                                                   adapter->rx_ps_bsize0,
283                                                   PCI_DMA_FROMDEVICE);
284                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
285                         dev_err(&pdev->dev, "RX DMA map failed\n");
286                         adapter->rx_dma_failed++;
287                         /* cleanup skb */
288                         dev_kfree_skb_any(skb);
289                         buffer_info->skb = NULL;
290                         break;
291                 }
292
293                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
294
295                 i++;
296                 if (i == rx_ring->count)
297                         i = 0;
298                 buffer_info = &rx_ring->buffer_info[i];
299         }
300
301 no_buffers:
302         if (rx_ring->next_to_use != i) {
303                 rx_ring->next_to_use = i;
304
305                 if (!(i--))
306                         i = (rx_ring->count - 1);
307
308                 /*
309                  * Force memory writes to complete before letting h/w
310                  * know there are new descriptors to fetch.  (Only
311                  * applicable for weak-ordered memory model archs,
312                  * such as IA-64).
313                  */
314                 wmb();
315                 /*
316                  * Hardware increments by 16 bytes, but packet split
317                  * descriptors are 32 bytes...so we increment tail
318                  * twice as much.
319                  */
320                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
321         }
322 }
323
324 /**
325  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
326  * @adapter: address of board private structure
327  * @cleaned_count: number of buffers to allocate this pass
328  **/
329
330 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
331                                          int cleaned_count)
332 {
333         struct net_device *netdev = adapter->netdev;
334         struct pci_dev *pdev = adapter->pdev;
335         struct e1000_rx_desc *rx_desc;
336         struct e1000_ring *rx_ring = adapter->rx_ring;
337         struct e1000_buffer *buffer_info;
338         struct sk_buff *skb;
339         unsigned int i;
340         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
341
342         i = rx_ring->next_to_use;
343         buffer_info = &rx_ring->buffer_info[i];
344
345         while (cleaned_count--) {
346                 skb = buffer_info->skb;
347                 if (skb) {
348                         skb_trim(skb, 0);
349                         goto check_page;
350                 }
351
352                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
353                 if (unlikely(!skb)) {
354                         /* Better luck next round */
355                         adapter->alloc_rx_buff_failed++;
356                         break;
357                 }
358
359                 buffer_info->skb = skb;
360 check_page:
361                 /* allocate a new page if necessary */
362                 if (!buffer_info->page) {
363                         buffer_info->page = alloc_page(GFP_ATOMIC);
364                         if (unlikely(!buffer_info->page)) {
365                                 adapter->alloc_rx_buff_failed++;
366                                 break;
367                         }
368                 }
369
370                 if (!buffer_info->dma)
371                         buffer_info->dma = pci_map_page(pdev,
372                                                         buffer_info->page, 0,
373                                                         PAGE_SIZE,
374                                                         PCI_DMA_FROMDEVICE);
375
376                 rx_desc = E1000_RX_DESC(*rx_ring, i);
377                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
378
379                 if (unlikely(++i == rx_ring->count))
380                         i = 0;
381                 buffer_info = &rx_ring->buffer_info[i];
382         }
383
384         if (likely(rx_ring->next_to_use != i)) {
385                 rx_ring->next_to_use = i;
386                 if (unlikely(i-- == 0))
387                         i = (rx_ring->count - 1);
388
389                 /* Force memory writes to complete before letting h/w
390                  * know there are new descriptors to fetch.  (Only
391                  * applicable for weak-ordered memory model archs,
392                  * such as IA-64). */
393                 wmb();
394                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
395         }
396 }
397
398 /**
399  * e1000_clean_rx_irq - Send received data up the network stack; legacy
400  * @adapter: board private structure
401  *
402  * the return value indicates whether actual cleaning was done, there
403  * is no guarantee that everything was cleaned
404  **/
405 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
406                                int *work_done, int work_to_do)
407 {
408         struct net_device *netdev = adapter->netdev;
409         struct pci_dev *pdev = adapter->pdev;
410         struct e1000_hw *hw = &adapter->hw;
411         struct e1000_ring *rx_ring = adapter->rx_ring;
412         struct e1000_rx_desc *rx_desc, *next_rxd;
413         struct e1000_buffer *buffer_info, *next_buffer;
414         u32 length;
415         unsigned int i;
416         int cleaned_count = 0;
417         bool cleaned = 0;
418         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
419
420         i = rx_ring->next_to_clean;
421         rx_desc = E1000_RX_DESC(*rx_ring, i);
422         buffer_info = &rx_ring->buffer_info[i];
423
424         while (rx_desc->status & E1000_RXD_STAT_DD) {
425                 struct sk_buff *skb;
426                 u8 status;
427
428                 if (*work_done >= work_to_do)
429                         break;
430                 (*work_done)++;
431
432                 status = rx_desc->status;
433                 skb = buffer_info->skb;
434                 buffer_info->skb = NULL;
435
436                 prefetch(skb->data - NET_IP_ALIGN);
437
438                 i++;
439                 if (i == rx_ring->count)
440                         i = 0;
441                 next_rxd = E1000_RX_DESC(*rx_ring, i);
442                 prefetch(next_rxd);
443
444                 next_buffer = &rx_ring->buffer_info[i];
445
446                 cleaned = 1;
447                 cleaned_count++;
448                 pci_unmap_single(pdev,
449                                  buffer_info->dma,
450                                  adapter->rx_buffer_len,
451                                  PCI_DMA_FROMDEVICE);
452                 buffer_info->dma = 0;
453
454                 length = le16_to_cpu(rx_desc->length);
455
456                 /*
457                  * !EOP means multiple descriptors were used to store a single
458                  * packet, if that's the case we need to toss it.  In fact, we
459                  * need to toss every packet with the EOP bit clear and the
460                  * next frame that _does_ have the EOP bit set, as it is by
461                  * definition only a frame fragment
462                  */
463                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
464                         adapter->flags2 |= FLAG2_IS_DISCARDING;
465
466                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
467                         /* All receives must fit into a single buffer */
468                         e_dbg("Receive packet consumed multiple buffers\n");
469                         /* recycle */
470                         buffer_info->skb = skb;
471                         if (status & E1000_RXD_STAT_EOP)
472                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
473                         goto next_desc;
474                 }
475
476                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
477                         /* recycle */
478                         buffer_info->skb = skb;
479                         goto next_desc;
480                 }
481
482                 /* adjust length to remove Ethernet CRC */
483                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
484                         length -= 4;
485
486                 total_rx_bytes += length;
487                 total_rx_packets++;
488
489                 /*
490                  * code added for copybreak, this should improve
491                  * performance for small packets with large amounts
492                  * of reassembly being done in the stack
493                  */
494                 if (length < copybreak) {
495                         struct sk_buff *new_skb =
496                             netdev_alloc_skb_ip_align(netdev, length);
497                         if (new_skb) {
498                                 skb_copy_to_linear_data_offset(new_skb,
499                                                                -NET_IP_ALIGN,
500                                                                (skb->data -
501                                                                 NET_IP_ALIGN),
502                                                                (length +
503                                                                 NET_IP_ALIGN));
504                                 /* save the skb in buffer_info as good */
505                                 buffer_info->skb = skb;
506                                 skb = new_skb;
507                         }
508                         /* else just continue with the old one */
509                 }
510                 /* end copybreak code */
511                 skb_put(skb, length);
512
513                 /* Receive Checksum Offload */
514                 e1000_rx_checksum(adapter,
515                                   (u32)(status) |
516                                   ((u32)(rx_desc->errors) << 24),
517                                   le16_to_cpu(rx_desc->csum), skb);
518
519                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
520
521 next_desc:
522                 rx_desc->status = 0;
523
524                 /* return some buffers to hardware, one at a time is too slow */
525                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
526                         adapter->alloc_rx_buf(adapter, cleaned_count);
527                         cleaned_count = 0;
528                 }
529
530                 /* use prefetched values */
531                 rx_desc = next_rxd;
532                 buffer_info = next_buffer;
533         }
534         rx_ring->next_to_clean = i;
535
536         cleaned_count = e1000_desc_unused(rx_ring);
537         if (cleaned_count)
538                 adapter->alloc_rx_buf(adapter, cleaned_count);
539
540         adapter->total_rx_bytes += total_rx_bytes;
541         adapter->total_rx_packets += total_rx_packets;
542         netdev->stats.rx_bytes += total_rx_bytes;
543         netdev->stats.rx_packets += total_rx_packets;
544         return cleaned;
545 }
546
547 static void e1000_put_txbuf(struct e1000_adapter *adapter,
548                              struct e1000_buffer *buffer_info)
549 {
550         if (buffer_info->dma) {
551                 if (buffer_info->mapped_as_page)
552                         pci_unmap_page(adapter->pdev, buffer_info->dma,
553                                        buffer_info->length, PCI_DMA_TODEVICE);
554                 else
555                         pci_unmap_single(adapter->pdev, buffer_info->dma,
556                                          buffer_info->length,
557                                          PCI_DMA_TODEVICE);
558                 buffer_info->dma = 0;
559         }
560         if (buffer_info->skb) {
561                 dev_kfree_skb_any(buffer_info->skb);
562                 buffer_info->skb = NULL;
563         }
564         buffer_info->time_stamp = 0;
565 }
566
567 static void e1000_print_hw_hang(struct work_struct *work)
568 {
569         struct e1000_adapter *adapter = container_of(work,
570                                                      struct e1000_adapter,
571                                                      print_hang_task);
572         struct e1000_ring *tx_ring = adapter->tx_ring;
573         unsigned int i = tx_ring->next_to_clean;
574         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
575         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
576         struct e1000_hw *hw = &adapter->hw;
577         u16 phy_status, phy_1000t_status, phy_ext_status;
578         u16 pci_status;
579
580         e1e_rphy(hw, PHY_STATUS, &phy_status);
581         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
582         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
583
584         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
585
586         /* detected Hardware unit hang */
587         e_err("Detected Hardware Unit Hang:\n"
588               "  TDH                  <%x>\n"
589               "  TDT                  <%x>\n"
590               "  next_to_use          <%x>\n"
591               "  next_to_clean        <%x>\n"
592               "buffer_info[next_to_clean]:\n"
593               "  time_stamp           <%lx>\n"
594               "  next_to_watch        <%x>\n"
595               "  jiffies              <%lx>\n"
596               "  next_to_watch.status <%x>\n"
597               "MAC Status             <%x>\n"
598               "PHY Status             <%x>\n"
599               "PHY 1000BASE-T Status  <%x>\n"
600               "PHY Extended Status    <%x>\n"
601               "PCI Status             <%x>\n",
602               readl(adapter->hw.hw_addr + tx_ring->head),
603               readl(adapter->hw.hw_addr + tx_ring->tail),
604               tx_ring->next_to_use,
605               tx_ring->next_to_clean,
606               tx_ring->buffer_info[eop].time_stamp,
607               eop,
608               jiffies,
609               eop_desc->upper.fields.status,
610               er32(STATUS),
611               phy_status,
612               phy_1000t_status,
613               phy_ext_status,
614               pci_status);
615 }
616
617 /**
618  * e1000_clean_tx_irq - Reclaim resources after transmit completes
619  * @adapter: board private structure
620  *
621  * the return value indicates whether actual cleaning was done, there
622  * is no guarantee that everything was cleaned
623  **/
624 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
625 {
626         struct net_device *netdev = adapter->netdev;
627         struct e1000_hw *hw = &adapter->hw;
628         struct e1000_ring *tx_ring = adapter->tx_ring;
629         struct e1000_tx_desc *tx_desc, *eop_desc;
630         struct e1000_buffer *buffer_info;
631         unsigned int i, eop;
632         unsigned int count = 0;
633         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
634
635         i = tx_ring->next_to_clean;
636         eop = tx_ring->buffer_info[i].next_to_watch;
637         eop_desc = E1000_TX_DESC(*tx_ring, eop);
638
639         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
640                (count < tx_ring->count)) {
641                 bool cleaned = false;
642                 for (; !cleaned; count++) {
643                         tx_desc = E1000_TX_DESC(*tx_ring, i);
644                         buffer_info = &tx_ring->buffer_info[i];
645                         cleaned = (i == eop);
646
647                         if (cleaned) {
648                                 struct sk_buff *skb = buffer_info->skb;
649                                 unsigned int segs, bytecount;
650                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
651                                 /* multiply data chunks by size of headers */
652                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
653                                             skb->len;
654                                 total_tx_packets += segs;
655                                 total_tx_bytes += bytecount;
656                         }
657
658                         e1000_put_txbuf(adapter, buffer_info);
659                         tx_desc->upper.data = 0;
660
661                         i++;
662                         if (i == tx_ring->count)
663                                 i = 0;
664                 }
665
666                 eop = tx_ring->buffer_info[i].next_to_watch;
667                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
668         }
669
670         tx_ring->next_to_clean = i;
671
672 #define TX_WAKE_THRESHOLD 32
673         if (count && netif_carrier_ok(netdev) &&
674             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
675                 /* Make sure that anybody stopping the queue after this
676                  * sees the new next_to_clean.
677                  */
678                 smp_mb();
679
680                 if (netif_queue_stopped(netdev) &&
681                     !(test_bit(__E1000_DOWN, &adapter->state))) {
682                         netif_wake_queue(netdev);
683                         ++adapter->restart_queue;
684                 }
685         }
686
687         if (adapter->detect_tx_hung) {
688                 /*
689                  * Detect a transmit hang in hardware, this serializes the
690                  * check with the clearing of time_stamp and movement of i
691                  */
692                 adapter->detect_tx_hung = 0;
693                 if (tx_ring->buffer_info[i].time_stamp &&
694                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
695                                + (adapter->tx_timeout_factor * HZ)) &&
696                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
697                         schedule_work(&adapter->print_hang_task);
698                         netif_stop_queue(netdev);
699                 }
700         }
701         adapter->total_tx_bytes += total_tx_bytes;
702         adapter->total_tx_packets += total_tx_packets;
703         netdev->stats.tx_bytes += total_tx_bytes;
704         netdev->stats.tx_packets += total_tx_packets;
705         return (count < tx_ring->count);
706 }
707
708 /**
709  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
710  * @adapter: board private structure
711  *
712  * the return value indicates whether actual cleaning was done, there
713  * is no guarantee that everything was cleaned
714  **/
715 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
716                                   int *work_done, int work_to_do)
717 {
718         struct e1000_hw *hw = &adapter->hw;
719         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
720         struct net_device *netdev = adapter->netdev;
721         struct pci_dev *pdev = adapter->pdev;
722         struct e1000_ring *rx_ring = adapter->rx_ring;
723         struct e1000_buffer *buffer_info, *next_buffer;
724         struct e1000_ps_page *ps_page;
725         struct sk_buff *skb;
726         unsigned int i, j;
727         u32 length, staterr;
728         int cleaned_count = 0;
729         bool cleaned = 0;
730         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
731
732         i = rx_ring->next_to_clean;
733         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
734         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
735         buffer_info = &rx_ring->buffer_info[i];
736
737         while (staterr & E1000_RXD_STAT_DD) {
738                 if (*work_done >= work_to_do)
739                         break;
740                 (*work_done)++;
741                 skb = buffer_info->skb;
742
743                 /* in the packet split case this is header only */
744                 prefetch(skb->data - NET_IP_ALIGN);
745
746                 i++;
747                 if (i == rx_ring->count)
748                         i = 0;
749                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
750                 prefetch(next_rxd);
751
752                 next_buffer = &rx_ring->buffer_info[i];
753
754                 cleaned = 1;
755                 cleaned_count++;
756                 pci_unmap_single(pdev, buffer_info->dma,
757                                  adapter->rx_ps_bsize0,
758                                  PCI_DMA_FROMDEVICE);
759                 buffer_info->dma = 0;
760
761                 /* see !EOP comment in other rx routine */
762                 if (!(staterr & E1000_RXD_STAT_EOP))
763                         adapter->flags2 |= FLAG2_IS_DISCARDING;
764
765                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
766                         e_dbg("Packet Split buffers didn't pick up the full "
767                               "packet\n");
768                         dev_kfree_skb_irq(skb);
769                         if (staterr & E1000_RXD_STAT_EOP)
770                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
771                         goto next_desc;
772                 }
773
774                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
775                         dev_kfree_skb_irq(skb);
776                         goto next_desc;
777                 }
778
779                 length = le16_to_cpu(rx_desc->wb.middle.length0);
780
781                 if (!length) {
782                         e_dbg("Last part of the packet spanning multiple "
783                               "descriptors\n");
784                         dev_kfree_skb_irq(skb);
785                         goto next_desc;
786                 }
787
788                 /* Good Receive */
789                 skb_put(skb, length);
790
791                 {
792                 /*
793                  * this looks ugly, but it seems compiler issues make it
794                  * more efficient than reusing j
795                  */
796                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
797
798                 /*
799                  * page alloc/put takes too long and effects small packet
800                  * throughput, so unsplit small packets and save the alloc/put
801                  * only valid in softirq (napi) context to call kmap_*
802                  */
803                 if (l1 && (l1 <= copybreak) &&
804                     ((length + l1) <= adapter->rx_ps_bsize0)) {
805                         u8 *vaddr;
806
807                         ps_page = &buffer_info->ps_pages[0];
808
809                         /*
810                          * there is no documentation about how to call
811                          * kmap_atomic, so we can't hold the mapping
812                          * very long
813                          */
814                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
815                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
816                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
817                         memcpy(skb_tail_pointer(skb), vaddr, l1);
818                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
819                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
820                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
821
822                         /* remove the CRC */
823                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
824                                 l1 -= 4;
825
826                         skb_put(skb, l1);
827                         goto copydone;
828                 } /* if */
829                 }
830
831                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
832                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
833                         if (!length)
834                                 break;
835
836                         ps_page = &buffer_info->ps_pages[j];
837                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
838                                        PCI_DMA_FROMDEVICE);
839                         ps_page->dma = 0;
840                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
841                         ps_page->page = NULL;
842                         skb->len += length;
843                         skb->data_len += length;
844                         skb->truesize += length;
845                 }
846
847                 /* strip the ethernet crc, problem is we're using pages now so
848                  * this whole operation can get a little cpu intensive
849                  */
850                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
851                         pskb_trim(skb, skb->len - 4);
852
853 copydone:
854                 total_rx_bytes += skb->len;
855                 total_rx_packets++;
856
857                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
858                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
859
860                 if (rx_desc->wb.upper.header_status &
861                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
862                         adapter->rx_hdr_split++;
863
864                 e1000_receive_skb(adapter, netdev, skb,
865                                   staterr, rx_desc->wb.middle.vlan);
866
867 next_desc:
868                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
869                 buffer_info->skb = NULL;
870
871                 /* return some buffers to hardware, one at a time is too slow */
872                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
873                         adapter->alloc_rx_buf(adapter, cleaned_count);
874                         cleaned_count = 0;
875                 }
876
877                 /* use prefetched values */
878                 rx_desc = next_rxd;
879                 buffer_info = next_buffer;
880
881                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
882         }
883         rx_ring->next_to_clean = i;
884
885         cleaned_count = e1000_desc_unused(rx_ring);
886         if (cleaned_count)
887                 adapter->alloc_rx_buf(adapter, cleaned_count);
888
889         adapter->total_rx_bytes += total_rx_bytes;
890         adapter->total_rx_packets += total_rx_packets;
891         netdev->stats.rx_bytes += total_rx_bytes;
892         netdev->stats.rx_packets += total_rx_packets;
893         return cleaned;
894 }
895
896 /**
897  * e1000_consume_page - helper function
898  **/
899 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
900                                u16 length)
901 {
902         bi->page = NULL;
903         skb->len += length;
904         skb->data_len += length;
905         skb->truesize += length;
906 }
907
908 /**
909  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
910  * @adapter: board private structure
911  *
912  * the return value indicates whether actual cleaning was done, there
913  * is no guarantee that everything was cleaned
914  **/
915
916 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
917                                      int *work_done, int work_to_do)
918 {
919         struct net_device *netdev = adapter->netdev;
920         struct pci_dev *pdev = adapter->pdev;
921         struct e1000_ring *rx_ring = adapter->rx_ring;
922         struct e1000_rx_desc *rx_desc, *next_rxd;
923         struct e1000_buffer *buffer_info, *next_buffer;
924         u32 length;
925         unsigned int i;
926         int cleaned_count = 0;
927         bool cleaned = false;
928         unsigned int total_rx_bytes=0, total_rx_packets=0;
929
930         i = rx_ring->next_to_clean;
931         rx_desc = E1000_RX_DESC(*rx_ring, i);
932         buffer_info = &rx_ring->buffer_info[i];
933
934         while (rx_desc->status & E1000_RXD_STAT_DD) {
935                 struct sk_buff *skb;
936                 u8 status;
937
938                 if (*work_done >= work_to_do)
939                         break;
940                 (*work_done)++;
941
942                 status = rx_desc->status;
943                 skb = buffer_info->skb;
944                 buffer_info->skb = NULL;
945
946                 ++i;
947                 if (i == rx_ring->count)
948                         i = 0;
949                 next_rxd = E1000_RX_DESC(*rx_ring, i);
950                 prefetch(next_rxd);
951
952                 next_buffer = &rx_ring->buffer_info[i];
953
954                 cleaned = true;
955                 cleaned_count++;
956                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
957                                PCI_DMA_FROMDEVICE);
958                 buffer_info->dma = 0;
959
960                 length = le16_to_cpu(rx_desc->length);
961
962                 /* errors is only valid for DD + EOP descriptors */
963                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
964                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
965                                 /* recycle both page and skb */
966                                 buffer_info->skb = skb;
967                                 /* an error means any chain goes out the window
968                                  * too */
969                                 if (rx_ring->rx_skb_top)
970                                         dev_kfree_skb(rx_ring->rx_skb_top);
971                                 rx_ring->rx_skb_top = NULL;
972                                 goto next_desc;
973                 }
974
975 #define rxtop rx_ring->rx_skb_top
976                 if (!(status & E1000_RXD_STAT_EOP)) {
977                         /* this descriptor is only the beginning (or middle) */
978                         if (!rxtop) {
979                                 /* this is the beginning of a chain */
980                                 rxtop = skb;
981                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
982                                                    0, length);
983                         } else {
984                                 /* this is the middle of a chain */
985                                 skb_fill_page_desc(rxtop,
986                                     skb_shinfo(rxtop)->nr_frags,
987                                     buffer_info->page, 0, length);
988                                 /* re-use the skb, only consumed the page */
989                                 buffer_info->skb = skb;
990                         }
991                         e1000_consume_page(buffer_info, rxtop, length);
992                         goto next_desc;
993                 } else {
994                         if (rxtop) {
995                                 /* end of the chain */
996                                 skb_fill_page_desc(rxtop,
997                                     skb_shinfo(rxtop)->nr_frags,
998                                     buffer_info->page, 0, length);
999                                 /* re-use the current skb, we only consumed the
1000                                  * page */
1001                                 buffer_info->skb = skb;
1002                                 skb = rxtop;
1003                                 rxtop = NULL;
1004                                 e1000_consume_page(buffer_info, skb, length);
1005                         } else {
1006                                 /* no chain, got EOP, this buf is the packet
1007                                  * copybreak to save the put_page/alloc_page */
1008                                 if (length <= copybreak &&
1009                                     skb_tailroom(skb) >= length) {
1010                                         u8 *vaddr;
1011                                         vaddr = kmap_atomic(buffer_info->page,
1012                                                            KM_SKB_DATA_SOFTIRQ);
1013                                         memcpy(skb_tail_pointer(skb), vaddr,
1014                                                length);
1015                                         kunmap_atomic(vaddr,
1016                                                       KM_SKB_DATA_SOFTIRQ);
1017                                         /* re-use the page, so don't erase
1018                                          * buffer_info->page */
1019                                         skb_put(skb, length);
1020                                 } else {
1021                                         skb_fill_page_desc(skb, 0,
1022                                                            buffer_info->page, 0,
1023                                                            length);
1024                                         e1000_consume_page(buffer_info, skb,
1025                                                            length);
1026                                 }
1027                         }
1028                 }
1029
1030                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1031                 e1000_rx_checksum(adapter,
1032                                   (u32)(status) |
1033                                   ((u32)(rx_desc->errors) << 24),
1034                                   le16_to_cpu(rx_desc->csum), skb);
1035
1036                 /* probably a little skewed due to removing CRC */
1037                 total_rx_bytes += skb->len;
1038                 total_rx_packets++;
1039
1040                 /* eth type trans needs skb->data to point to something */
1041                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1042                         e_err("pskb_may_pull failed.\n");
1043                         dev_kfree_skb(skb);
1044                         goto next_desc;
1045                 }
1046
1047                 e1000_receive_skb(adapter, netdev, skb, status,
1048                                   rx_desc->special);
1049
1050 next_desc:
1051                 rx_desc->status = 0;
1052
1053                 /* return some buffers to hardware, one at a time is too slow */
1054                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1055                         adapter->alloc_rx_buf(adapter, cleaned_count);
1056                         cleaned_count = 0;
1057                 }
1058
1059                 /* use prefetched values */
1060                 rx_desc = next_rxd;
1061                 buffer_info = next_buffer;
1062         }
1063         rx_ring->next_to_clean = i;
1064
1065         cleaned_count = e1000_desc_unused(rx_ring);
1066         if (cleaned_count)
1067                 adapter->alloc_rx_buf(adapter, cleaned_count);
1068
1069         adapter->total_rx_bytes += total_rx_bytes;
1070         adapter->total_rx_packets += total_rx_packets;
1071         netdev->stats.rx_bytes += total_rx_bytes;
1072         netdev->stats.rx_packets += total_rx_packets;
1073         return cleaned;
1074 }
1075
1076 /**
1077  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1078  * @adapter: board private structure
1079  **/
1080 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1081 {
1082         struct e1000_ring *rx_ring = adapter->rx_ring;
1083         struct e1000_buffer *buffer_info;
1084         struct e1000_ps_page *ps_page;
1085         struct pci_dev *pdev = adapter->pdev;
1086         unsigned int i, j;
1087
1088         /* Free all the Rx ring sk_buffs */
1089         for (i = 0; i < rx_ring->count; i++) {
1090                 buffer_info = &rx_ring->buffer_info[i];
1091                 if (buffer_info->dma) {
1092                         if (adapter->clean_rx == e1000_clean_rx_irq)
1093                                 pci_unmap_single(pdev, buffer_info->dma,
1094                                                  adapter->rx_buffer_len,
1095                                                  PCI_DMA_FROMDEVICE);
1096                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1097                                 pci_unmap_page(pdev, buffer_info->dma,
1098                                                PAGE_SIZE,
1099                                                PCI_DMA_FROMDEVICE);
1100                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1101                                 pci_unmap_single(pdev, buffer_info->dma,
1102                                                  adapter->rx_ps_bsize0,
1103                                                  PCI_DMA_FROMDEVICE);
1104                         buffer_info->dma = 0;
1105                 }
1106
1107                 if (buffer_info->page) {
1108                         put_page(buffer_info->page);
1109                         buffer_info->page = NULL;
1110                 }
1111
1112                 if (buffer_info->skb) {
1113                         dev_kfree_skb(buffer_info->skb);
1114                         buffer_info->skb = NULL;
1115                 }
1116
1117                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1118                         ps_page = &buffer_info->ps_pages[j];
1119                         if (!ps_page->page)
1120                                 break;
1121                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1122                                        PCI_DMA_FROMDEVICE);
1123                         ps_page->dma = 0;
1124                         put_page(ps_page->page);
1125                         ps_page->page = NULL;
1126                 }
1127         }
1128
1129         /* there also may be some cached data from a chained receive */
1130         if (rx_ring->rx_skb_top) {
1131                 dev_kfree_skb(rx_ring->rx_skb_top);
1132                 rx_ring->rx_skb_top = NULL;
1133         }
1134
1135         /* Zero out the descriptor ring */
1136         memset(rx_ring->desc, 0, rx_ring->size);
1137
1138         rx_ring->next_to_clean = 0;
1139         rx_ring->next_to_use = 0;
1140         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1141
1142         writel(0, adapter->hw.hw_addr + rx_ring->head);
1143         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1144 }
1145
1146 static void e1000e_downshift_workaround(struct work_struct *work)
1147 {
1148         struct e1000_adapter *adapter = container_of(work,
1149                                         struct e1000_adapter, downshift_task);
1150
1151         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1152 }
1153
1154 /**
1155  * e1000_intr_msi - Interrupt Handler
1156  * @irq: interrupt number
1157  * @data: pointer to a network interface device structure
1158  **/
1159 static irqreturn_t e1000_intr_msi(int irq, void *data)
1160 {
1161         struct net_device *netdev = data;
1162         struct e1000_adapter *adapter = netdev_priv(netdev);
1163         struct e1000_hw *hw = &adapter->hw;
1164         u32 icr = er32(ICR);
1165
1166         /*
1167          * read ICR disables interrupts using IAM
1168          */
1169
1170         if (icr & E1000_ICR_LSC) {
1171                 hw->mac.get_link_status = 1;
1172                 /*
1173                  * ICH8 workaround-- Call gig speed drop workaround on cable
1174                  * disconnect (LSC) before accessing any PHY registers
1175                  */
1176                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1177                     (!(er32(STATUS) & E1000_STATUS_LU)))
1178                         schedule_work(&adapter->downshift_task);
1179
1180                 /*
1181                  * 80003ES2LAN workaround-- For packet buffer work-around on
1182                  * link down event; disable receives here in the ISR and reset
1183                  * adapter in watchdog
1184                  */
1185                 if (netif_carrier_ok(netdev) &&
1186                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1187                         /* disable receives */
1188                         u32 rctl = er32(RCTL);
1189                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1190                         adapter->flags |= FLAG_RX_RESTART_NOW;
1191                 }
1192                 /* guard against interrupt when we're going down */
1193                 if (!test_bit(__E1000_DOWN, &adapter->state))
1194                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1195         }
1196
1197         if (napi_schedule_prep(&adapter->napi)) {
1198                 adapter->total_tx_bytes = 0;
1199                 adapter->total_tx_packets = 0;
1200                 adapter->total_rx_bytes = 0;
1201                 adapter->total_rx_packets = 0;
1202                 __napi_schedule(&adapter->napi);
1203         }
1204
1205         return IRQ_HANDLED;
1206 }
1207
1208 /**
1209  * e1000_intr - Interrupt Handler
1210  * @irq: interrupt number
1211  * @data: pointer to a network interface device structure
1212  **/
1213 static irqreturn_t e1000_intr(int irq, void *data)
1214 {
1215         struct net_device *netdev = data;
1216         struct e1000_adapter *adapter = netdev_priv(netdev);
1217         struct e1000_hw *hw = &adapter->hw;
1218         u32 rctl, icr = er32(ICR);
1219
1220         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1221                 return IRQ_NONE;  /* Not our interrupt */
1222
1223         /*
1224          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1225          * not set, then the adapter didn't send an interrupt
1226          */
1227         if (!(icr & E1000_ICR_INT_ASSERTED))
1228                 return IRQ_NONE;
1229
1230         /*
1231          * Interrupt Auto-Mask...upon reading ICR,
1232          * interrupts are masked.  No need for the
1233          * IMC write
1234          */
1235
1236         if (icr & E1000_ICR_LSC) {
1237                 hw->mac.get_link_status = 1;
1238                 /*
1239                  * ICH8 workaround-- Call gig speed drop workaround on cable
1240                  * disconnect (LSC) before accessing any PHY registers
1241                  */
1242                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1243                     (!(er32(STATUS) & E1000_STATUS_LU)))
1244                         schedule_work(&adapter->downshift_task);
1245
1246                 /*
1247                  * 80003ES2LAN workaround--
1248                  * For packet buffer work-around on link down event;
1249                  * disable receives here in the ISR and
1250                  * reset adapter in watchdog
1251                  */
1252                 if (netif_carrier_ok(netdev) &&
1253                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1254                         /* disable receives */
1255                         rctl = er32(RCTL);
1256                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1257                         adapter->flags |= FLAG_RX_RESTART_NOW;
1258                 }
1259                 /* guard against interrupt when we're going down */
1260                 if (!test_bit(__E1000_DOWN, &adapter->state))
1261                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1262         }
1263
1264         if (napi_schedule_prep(&adapter->napi)) {
1265                 adapter->total_tx_bytes = 0;
1266                 adapter->total_tx_packets = 0;
1267                 adapter->total_rx_bytes = 0;
1268                 adapter->total_rx_packets = 0;
1269                 __napi_schedule(&adapter->napi);
1270         }
1271
1272         return IRQ_HANDLED;
1273 }
1274
1275 static irqreturn_t e1000_msix_other(int irq, void *data)
1276 {
1277         struct net_device *netdev = data;
1278         struct e1000_adapter *adapter = netdev_priv(netdev);
1279         struct e1000_hw *hw = &adapter->hw;
1280         u32 icr = er32(ICR);
1281
1282         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1283                 if (!test_bit(__E1000_DOWN, &adapter->state))
1284                         ew32(IMS, E1000_IMS_OTHER);
1285                 return IRQ_NONE;
1286         }
1287
1288         if (icr & adapter->eiac_mask)
1289                 ew32(ICS, (icr & adapter->eiac_mask));
1290
1291         if (icr & E1000_ICR_OTHER) {
1292                 if (!(icr & E1000_ICR_LSC))
1293                         goto no_link_interrupt;
1294                 hw->mac.get_link_status = 1;
1295                 /* guard against interrupt when we're going down */
1296                 if (!test_bit(__E1000_DOWN, &adapter->state))
1297                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1298         }
1299
1300 no_link_interrupt:
1301         if (!test_bit(__E1000_DOWN, &adapter->state))
1302                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1303
1304         return IRQ_HANDLED;
1305 }
1306
1307
1308 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1309 {
1310         struct net_device *netdev = data;
1311         struct e1000_adapter *adapter = netdev_priv(netdev);
1312         struct e1000_hw *hw = &adapter->hw;
1313         struct e1000_ring *tx_ring = adapter->tx_ring;
1314
1315
1316         adapter->total_tx_bytes = 0;
1317         adapter->total_tx_packets = 0;
1318
1319         if (!e1000_clean_tx_irq(adapter))
1320                 /* Ring was not completely cleaned, so fire another interrupt */
1321                 ew32(ICS, tx_ring->ims_val);
1322
1323         return IRQ_HANDLED;
1324 }
1325
1326 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1327 {
1328         struct net_device *netdev = data;
1329         struct e1000_adapter *adapter = netdev_priv(netdev);
1330
1331         /* Write the ITR value calculated at the end of the
1332          * previous interrupt.
1333          */
1334         if (adapter->rx_ring->set_itr) {
1335                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1336                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1337                 adapter->rx_ring->set_itr = 0;
1338         }
1339
1340         if (napi_schedule_prep(&adapter->napi)) {
1341                 adapter->total_rx_bytes = 0;
1342                 adapter->total_rx_packets = 0;
1343                 __napi_schedule(&adapter->napi);
1344         }
1345         return IRQ_HANDLED;
1346 }
1347
1348 /**
1349  * e1000_configure_msix - Configure MSI-X hardware
1350  *
1351  * e1000_configure_msix sets up the hardware to properly
1352  * generate MSI-X interrupts.
1353  **/
1354 static void e1000_configure_msix(struct e1000_adapter *adapter)
1355 {
1356         struct e1000_hw *hw = &adapter->hw;
1357         struct e1000_ring *rx_ring = adapter->rx_ring;
1358         struct e1000_ring *tx_ring = adapter->tx_ring;
1359         int vector = 0;
1360         u32 ctrl_ext, ivar = 0;
1361
1362         adapter->eiac_mask = 0;
1363
1364         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1365         if (hw->mac.type == e1000_82574) {
1366                 u32 rfctl = er32(RFCTL);
1367                 rfctl |= E1000_RFCTL_ACK_DIS;
1368                 ew32(RFCTL, rfctl);
1369         }
1370
1371 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1372         /* Configure Rx vector */
1373         rx_ring->ims_val = E1000_IMS_RXQ0;
1374         adapter->eiac_mask |= rx_ring->ims_val;
1375         if (rx_ring->itr_val)
1376                 writel(1000000000 / (rx_ring->itr_val * 256),
1377                        hw->hw_addr + rx_ring->itr_register);
1378         else
1379                 writel(1, hw->hw_addr + rx_ring->itr_register);
1380         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1381
1382         /* Configure Tx vector */
1383         tx_ring->ims_val = E1000_IMS_TXQ0;
1384         vector++;
1385         if (tx_ring->itr_val)
1386                 writel(1000000000 / (tx_ring->itr_val * 256),
1387                        hw->hw_addr + tx_ring->itr_register);
1388         else
1389                 writel(1, hw->hw_addr + tx_ring->itr_register);
1390         adapter->eiac_mask |= tx_ring->ims_val;
1391         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1392
1393         /* set vector for Other Causes, e.g. link changes */
1394         vector++;
1395         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1396         if (rx_ring->itr_val)
1397                 writel(1000000000 / (rx_ring->itr_val * 256),
1398                        hw->hw_addr + E1000_EITR_82574(vector));
1399         else
1400                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1401
1402         /* Cause Tx interrupts on every write back */
1403         ivar |= (1 << 31);
1404
1405         ew32(IVAR, ivar);
1406
1407         /* enable MSI-X PBA support */
1408         ctrl_ext = er32(CTRL_EXT);
1409         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1410
1411         /* Auto-Mask Other interrupts upon ICR read */
1412 #define E1000_EIAC_MASK_82574   0x01F00000
1413         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1414         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1415         ew32(CTRL_EXT, ctrl_ext);
1416         e1e_flush();
1417 }
1418
1419 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1420 {
1421         if (adapter->msix_entries) {
1422                 pci_disable_msix(adapter->pdev);
1423                 kfree(adapter->msix_entries);
1424                 adapter->msix_entries = NULL;
1425         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1426                 pci_disable_msi(adapter->pdev);
1427                 adapter->flags &= ~FLAG_MSI_ENABLED;
1428         }
1429
1430         return;
1431 }
1432
1433 /**
1434  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1435  *
1436  * Attempt to configure interrupts using the best available
1437  * capabilities of the hardware and kernel.
1438  **/
1439 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1440 {
1441         int err;
1442         int numvecs, i;
1443
1444
1445         switch (adapter->int_mode) {
1446         case E1000E_INT_MODE_MSIX:
1447                 if (adapter->flags & FLAG_HAS_MSIX) {
1448                         numvecs = 3; /* RxQ0, TxQ0 and other */
1449                         adapter->msix_entries = kcalloc(numvecs,
1450                                                       sizeof(struct msix_entry),
1451                                                       GFP_KERNEL);
1452                         if (adapter->msix_entries) {
1453                                 for (i = 0; i < numvecs; i++)
1454                                         adapter->msix_entries[i].entry = i;
1455
1456                                 err = pci_enable_msix(adapter->pdev,
1457                                                       adapter->msix_entries,
1458                                                       numvecs);
1459                                 if (err == 0)
1460                                         return;
1461                         }
1462                         /* MSI-X failed, so fall through and try MSI */
1463                         e_err("Failed to initialize MSI-X interrupts.  "
1464                               "Falling back to MSI interrupts.\n");
1465                         e1000e_reset_interrupt_capability(adapter);
1466                 }
1467                 adapter->int_mode = E1000E_INT_MODE_MSI;
1468                 /* Fall through */
1469         case E1000E_INT_MODE_MSI:
1470                 if (!pci_enable_msi(adapter->pdev)) {
1471                         adapter->flags |= FLAG_MSI_ENABLED;
1472                 } else {
1473                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1474                         e_err("Failed to initialize MSI interrupts.  Falling "
1475                               "back to legacy interrupts.\n");
1476                 }
1477                 /* Fall through */
1478         case E1000E_INT_MODE_LEGACY:
1479                 /* Don't do anything; this is the system default */
1480                 break;
1481         }
1482
1483         return;
1484 }
1485
1486 /**
1487  * e1000_request_msix - Initialize MSI-X interrupts
1488  *
1489  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1490  * kernel.
1491  **/
1492 static int e1000_request_msix(struct e1000_adapter *adapter)
1493 {
1494         struct net_device *netdev = adapter->netdev;
1495         int err = 0, vector = 0;
1496
1497         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1498                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1499         else
1500                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1501         err = request_irq(adapter->msix_entries[vector].vector,
1502                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1503                           netdev);
1504         if (err)
1505                 goto out;
1506         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1507         adapter->rx_ring->itr_val = adapter->itr;
1508         vector++;
1509
1510         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1511                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1512         else
1513                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1514         err = request_irq(adapter->msix_entries[vector].vector,
1515                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1516                           netdev);
1517         if (err)
1518                 goto out;
1519         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1520         adapter->tx_ring->itr_val = adapter->itr;
1521         vector++;
1522
1523         err = request_irq(adapter->msix_entries[vector].vector,
1524                           e1000_msix_other, 0, netdev->name, netdev);
1525         if (err)
1526                 goto out;
1527
1528         e1000_configure_msix(adapter);
1529         return 0;
1530 out:
1531         return err;
1532 }
1533
1534 /**
1535  * e1000_request_irq - initialize interrupts
1536  *
1537  * Attempts to configure interrupts using the best available
1538  * capabilities of the hardware and kernel.
1539  **/
1540 static int e1000_request_irq(struct e1000_adapter *adapter)
1541 {
1542         struct net_device *netdev = adapter->netdev;
1543         int err;
1544
1545         if (adapter->msix_entries) {
1546                 err = e1000_request_msix(adapter);
1547                 if (!err)
1548                         return err;
1549                 /* fall back to MSI */
1550                 e1000e_reset_interrupt_capability(adapter);
1551                 adapter->int_mode = E1000E_INT_MODE_MSI;
1552                 e1000e_set_interrupt_capability(adapter);
1553         }
1554         if (adapter->flags & FLAG_MSI_ENABLED) {
1555                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1556                                   netdev->name, netdev);
1557                 if (!err)
1558                         return err;
1559
1560                 /* fall back to legacy interrupt */
1561                 e1000e_reset_interrupt_capability(adapter);
1562                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1563         }
1564
1565         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1566                           netdev->name, netdev);
1567         if (err)
1568                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1569
1570         return err;
1571 }
1572
1573 static void e1000_free_irq(struct e1000_adapter *adapter)
1574 {
1575         struct net_device *netdev = adapter->netdev;
1576
1577         if (adapter->msix_entries) {
1578                 int vector = 0;
1579
1580                 free_irq(adapter->msix_entries[vector].vector, netdev);
1581                 vector++;
1582
1583                 free_irq(adapter->msix_entries[vector].vector, netdev);
1584                 vector++;
1585
1586                 /* Other Causes interrupt vector */
1587                 free_irq(adapter->msix_entries[vector].vector, netdev);
1588                 return;
1589         }
1590
1591         free_irq(adapter->pdev->irq, netdev);
1592 }
1593
1594 /**
1595  * e1000_irq_disable - Mask off interrupt generation on the NIC
1596  **/
1597 static void e1000_irq_disable(struct e1000_adapter *adapter)
1598 {
1599         struct e1000_hw *hw = &adapter->hw;
1600
1601         ew32(IMC, ~0);
1602         if (adapter->msix_entries)
1603                 ew32(EIAC_82574, 0);
1604         e1e_flush();
1605         synchronize_irq(adapter->pdev->irq);
1606 }
1607
1608 /**
1609  * e1000_irq_enable - Enable default interrupt generation settings
1610  **/
1611 static void e1000_irq_enable(struct e1000_adapter *adapter)
1612 {
1613         struct e1000_hw *hw = &adapter->hw;
1614
1615         if (adapter->msix_entries) {
1616                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1617                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1618         } else {
1619                 ew32(IMS, IMS_ENABLE_MASK);
1620         }
1621         e1e_flush();
1622 }
1623
1624 /**
1625  * e1000_get_hw_control - get control of the h/w from f/w
1626  * @adapter: address of board private structure
1627  *
1628  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1629  * For ASF and Pass Through versions of f/w this means that
1630  * the driver is loaded. For AMT version (only with 82573)
1631  * of the f/w this means that the network i/f is open.
1632  **/
1633 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1634 {
1635         struct e1000_hw *hw = &adapter->hw;
1636         u32 ctrl_ext;
1637         u32 swsm;
1638
1639         /* Let firmware know the driver has taken over */
1640         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1641                 swsm = er32(SWSM);
1642                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1643         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1644                 ctrl_ext = er32(CTRL_EXT);
1645                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1646         }
1647 }
1648
1649 /**
1650  * e1000_release_hw_control - release control of the h/w to f/w
1651  * @adapter: address of board private structure
1652  *
1653  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1654  * For ASF and Pass Through versions of f/w this means that the
1655  * driver is no longer loaded. For AMT version (only with 82573) i
1656  * of the f/w this means that the network i/f is closed.
1657  *
1658  **/
1659 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1660 {
1661         struct e1000_hw *hw = &adapter->hw;
1662         u32 ctrl_ext;
1663         u32 swsm;
1664
1665         /* Let firmware taken over control of h/w */
1666         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1667                 swsm = er32(SWSM);
1668                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1669         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1670                 ctrl_ext = er32(CTRL_EXT);
1671                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1672         }
1673 }
1674
1675 /**
1676  * @e1000_alloc_ring - allocate memory for a ring structure
1677  **/
1678 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1679                                 struct e1000_ring *ring)
1680 {
1681         struct pci_dev *pdev = adapter->pdev;
1682
1683         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1684                                         GFP_KERNEL);
1685         if (!ring->desc)
1686                 return -ENOMEM;
1687
1688         return 0;
1689 }
1690
1691 /**
1692  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1693  * @adapter: board private structure
1694  *
1695  * Return 0 on success, negative on failure
1696  **/
1697 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1698 {
1699         struct e1000_ring *tx_ring = adapter->tx_ring;
1700         int err = -ENOMEM, size;
1701
1702         size = sizeof(struct e1000_buffer) * tx_ring->count;
1703         tx_ring->buffer_info = vmalloc(size);
1704         if (!tx_ring->buffer_info)
1705                 goto err;
1706         memset(tx_ring->buffer_info, 0, size);
1707
1708         /* round up to nearest 4K */
1709         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1710         tx_ring->size = ALIGN(tx_ring->size, 4096);
1711
1712         err = e1000_alloc_ring_dma(adapter, tx_ring);
1713         if (err)
1714                 goto err;
1715
1716         tx_ring->next_to_use = 0;
1717         tx_ring->next_to_clean = 0;
1718
1719         return 0;
1720 err:
1721         vfree(tx_ring->buffer_info);
1722         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1723         return err;
1724 }
1725
1726 /**
1727  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1728  * @adapter: board private structure
1729  *
1730  * Returns 0 on success, negative on failure
1731  **/
1732 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1733 {
1734         struct e1000_ring *rx_ring = adapter->rx_ring;
1735         struct e1000_buffer *buffer_info;
1736         int i, size, desc_len, err = -ENOMEM;
1737
1738         size = sizeof(struct e1000_buffer) * rx_ring->count;
1739         rx_ring->buffer_info = vmalloc(size);
1740         if (!rx_ring->buffer_info)
1741                 goto err;
1742         memset(rx_ring->buffer_info, 0, size);
1743
1744         for (i = 0; i < rx_ring->count; i++) {
1745                 buffer_info = &rx_ring->buffer_info[i];
1746                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1747                                                 sizeof(struct e1000_ps_page),
1748                                                 GFP_KERNEL);
1749                 if (!buffer_info->ps_pages)
1750                         goto err_pages;
1751         }
1752
1753         desc_len = sizeof(union e1000_rx_desc_packet_split);
1754
1755         /* Round up to nearest 4K */
1756         rx_ring->size = rx_ring->count * desc_len;
1757         rx_ring->size = ALIGN(rx_ring->size, 4096);
1758
1759         err = e1000_alloc_ring_dma(adapter, rx_ring);
1760         if (err)
1761                 goto err_pages;
1762
1763         rx_ring->next_to_clean = 0;
1764         rx_ring->next_to_use = 0;
1765         rx_ring->rx_skb_top = NULL;
1766
1767         return 0;
1768
1769 err_pages:
1770         for (i = 0; i < rx_ring->count; i++) {
1771                 buffer_info = &rx_ring->buffer_info[i];
1772                 kfree(buffer_info->ps_pages);
1773         }
1774 err:
1775         vfree(rx_ring->buffer_info);
1776         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1777         return err;
1778 }
1779
1780 /**
1781  * e1000_clean_tx_ring - Free Tx Buffers
1782  * @adapter: board private structure
1783  **/
1784 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1785 {
1786         struct e1000_ring *tx_ring = adapter->tx_ring;
1787         struct e1000_buffer *buffer_info;
1788         unsigned long size;
1789         unsigned int i;
1790
1791         for (i = 0; i < tx_ring->count; i++) {
1792                 buffer_info = &tx_ring->buffer_info[i];
1793                 e1000_put_txbuf(adapter, buffer_info);
1794         }
1795
1796         size = sizeof(struct e1000_buffer) * tx_ring->count;
1797         memset(tx_ring->buffer_info, 0, size);
1798
1799         memset(tx_ring->desc, 0, tx_ring->size);
1800
1801         tx_ring->next_to_use = 0;
1802         tx_ring->next_to_clean = 0;
1803
1804         writel(0, adapter->hw.hw_addr + tx_ring->head);
1805         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1806 }
1807
1808 /**
1809  * e1000e_free_tx_resources - Free Tx Resources per Queue
1810  * @adapter: board private structure
1811  *
1812  * Free all transmit software resources
1813  **/
1814 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1815 {
1816         struct pci_dev *pdev = adapter->pdev;
1817         struct e1000_ring *tx_ring = adapter->tx_ring;
1818
1819         e1000_clean_tx_ring(adapter);
1820
1821         vfree(tx_ring->buffer_info);
1822         tx_ring->buffer_info = NULL;
1823
1824         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1825                           tx_ring->dma);
1826         tx_ring->desc = NULL;
1827 }
1828
1829 /**
1830  * e1000e_free_rx_resources - Free Rx Resources
1831  * @adapter: board private structure
1832  *
1833  * Free all receive software resources
1834  **/
1835
1836 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1837 {
1838         struct pci_dev *pdev = adapter->pdev;
1839         struct e1000_ring *rx_ring = adapter->rx_ring;
1840         int i;
1841
1842         e1000_clean_rx_ring(adapter);
1843
1844         for (i = 0; i < rx_ring->count; i++) {
1845                 kfree(rx_ring->buffer_info[i].ps_pages);
1846         }
1847
1848         vfree(rx_ring->buffer_info);
1849         rx_ring->buffer_info = NULL;
1850
1851         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1852                           rx_ring->dma);
1853         rx_ring->desc = NULL;
1854 }
1855
1856 /**
1857  * e1000_update_itr - update the dynamic ITR value based on statistics
1858  * @adapter: pointer to adapter
1859  * @itr_setting: current adapter->itr
1860  * @packets: the number of packets during this measurement interval
1861  * @bytes: the number of bytes during this measurement interval
1862  *
1863  *      Stores a new ITR value based on packets and byte
1864  *      counts during the last interrupt.  The advantage of per interrupt
1865  *      computation is faster updates and more accurate ITR for the current
1866  *      traffic pattern.  Constants in this function were computed
1867  *      based on theoretical maximum wire speed and thresholds were set based
1868  *      on testing data as well as attempting to minimize response time
1869  *      while increasing bulk throughput.  This functionality is controlled
1870  *      by the InterruptThrottleRate module parameter.
1871  **/
1872 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1873                                      u16 itr_setting, int packets,
1874                                      int bytes)
1875 {
1876         unsigned int retval = itr_setting;
1877
1878         if (packets == 0)
1879                 goto update_itr_done;
1880
1881         switch (itr_setting) {
1882         case lowest_latency:
1883                 /* handle TSO and jumbo frames */
1884                 if (bytes/packets > 8000)
1885                         retval = bulk_latency;
1886                 else if ((packets < 5) && (bytes > 512)) {
1887                         retval = low_latency;
1888                 }
1889                 break;
1890         case low_latency:  /* 50 usec aka 20000 ints/s */
1891                 if (bytes > 10000) {
1892                         /* this if handles the TSO accounting */
1893                         if (bytes/packets > 8000) {
1894                                 retval = bulk_latency;
1895                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1896                                 retval = bulk_latency;
1897                         } else if ((packets > 35)) {
1898                                 retval = lowest_latency;
1899                         }
1900                 } else if (bytes/packets > 2000) {
1901                         retval = bulk_latency;
1902                 } else if (packets <= 2 && bytes < 512) {
1903                         retval = lowest_latency;
1904                 }
1905                 break;
1906         case bulk_latency: /* 250 usec aka 4000 ints/s */
1907                 if (bytes > 25000) {
1908                         if (packets > 35) {
1909                                 retval = low_latency;
1910                         }
1911                 } else if (bytes < 6000) {
1912                         retval = low_latency;
1913                 }
1914                 break;
1915         }
1916
1917 update_itr_done:
1918         return retval;
1919 }
1920
1921 static void e1000_set_itr(struct e1000_adapter *adapter)
1922 {
1923         struct e1000_hw *hw = &adapter->hw;
1924         u16 current_itr;
1925         u32 new_itr = adapter->itr;
1926
1927         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1928         if (adapter->link_speed != SPEED_1000) {
1929                 current_itr = 0;
1930                 new_itr = 4000;
1931                 goto set_itr_now;
1932         }
1933
1934         adapter->tx_itr = e1000_update_itr(adapter,
1935                                     adapter->tx_itr,
1936                                     adapter->total_tx_packets,
1937                                     adapter->total_tx_bytes);
1938         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1939         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1940                 adapter->tx_itr = low_latency;
1941
1942         adapter->rx_itr = e1000_update_itr(adapter,
1943                                     adapter->rx_itr,
1944                                     adapter->total_rx_packets,
1945                                     adapter->total_rx_bytes);
1946         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1947         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1948                 adapter->rx_itr = low_latency;
1949
1950         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1951
1952         switch (current_itr) {
1953         /* counts and packets in update_itr are dependent on these numbers */
1954         case lowest_latency:
1955                 new_itr = 70000;
1956                 break;
1957         case low_latency:
1958                 new_itr = 20000; /* aka hwitr = ~200 */
1959                 break;
1960         case bulk_latency:
1961                 new_itr = 4000;
1962                 break;
1963         default:
1964                 break;
1965         }
1966
1967 set_itr_now:
1968         if (new_itr != adapter->itr) {
1969                 /*
1970                  * this attempts to bias the interrupt rate towards Bulk
1971                  * by adding intermediate steps when interrupt rate is
1972                  * increasing
1973                  */
1974                 new_itr = new_itr > adapter->itr ?
1975                              min(adapter->itr + (new_itr >> 2), new_itr) :
1976                              new_itr;
1977                 adapter->itr = new_itr;
1978                 adapter->rx_ring->itr_val = new_itr;
1979                 if (adapter->msix_entries)
1980                         adapter->rx_ring->set_itr = 1;
1981                 else
1982                         ew32(ITR, 1000000000 / (new_itr * 256));
1983         }
1984 }
1985
1986 /**
1987  * e1000_alloc_queues - Allocate memory for all rings
1988  * @adapter: board private structure to initialize
1989  **/
1990 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1991 {
1992         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1993         if (!adapter->tx_ring)
1994                 goto err;
1995
1996         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1997         if (!adapter->rx_ring)
1998                 goto err;
1999
2000         return 0;
2001 err:
2002         e_err("Unable to allocate memory for queues\n");
2003         kfree(adapter->rx_ring);
2004         kfree(adapter->tx_ring);
2005         return -ENOMEM;
2006 }
2007
2008 /**
2009  * e1000_clean - NAPI Rx polling callback
2010  * @napi: struct associated with this polling callback
2011  * @budget: amount of packets driver is allowed to process this poll
2012  **/
2013 static int e1000_clean(struct napi_struct *napi, int budget)
2014 {
2015         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2016         struct e1000_hw *hw = &adapter->hw;
2017         struct net_device *poll_dev = adapter->netdev;
2018         int tx_cleaned = 1, work_done = 0;
2019
2020         adapter = netdev_priv(poll_dev);
2021
2022         if (adapter->msix_entries &&
2023             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2024                 goto clean_rx;
2025
2026         tx_cleaned = e1000_clean_tx_irq(adapter);
2027
2028 clean_rx:
2029         adapter->clean_rx(adapter, &work_done, budget);
2030
2031         if (!tx_cleaned)
2032                 work_done = budget;
2033
2034         /* If budget not fully consumed, exit the polling mode */
2035         if (work_done < budget) {
2036                 if (adapter->itr_setting & 3)
2037                         e1000_set_itr(adapter);
2038                 napi_complete(napi);
2039                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2040                         if (adapter->msix_entries)
2041                                 ew32(IMS, adapter->rx_ring->ims_val);
2042                         else
2043                                 e1000_irq_enable(adapter);
2044                 }
2045         }
2046
2047         return work_done;
2048 }
2049
2050 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2051 {
2052         struct e1000_adapter *adapter = netdev_priv(netdev);
2053         struct e1000_hw *hw = &adapter->hw;
2054         u32 vfta, index;
2055
2056         /* don't update vlan cookie if already programmed */
2057         if ((adapter->hw.mng_cookie.status &
2058              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2059             (vid == adapter->mng_vlan_id))
2060                 return;
2061
2062         /* add VID to filter table */
2063         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2064                 index = (vid >> 5) & 0x7F;
2065                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2066                 vfta |= (1 << (vid & 0x1F));
2067                 hw->mac.ops.write_vfta(hw, index, vfta);
2068         }
2069 }
2070
2071 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2072 {
2073         struct e1000_adapter *adapter = netdev_priv(netdev);
2074         struct e1000_hw *hw = &adapter->hw;
2075         u32 vfta, index;
2076
2077         if (!test_bit(__E1000_DOWN, &adapter->state))
2078                 e1000_irq_disable(adapter);
2079         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2080
2081         if (!test_bit(__E1000_DOWN, &adapter->state))
2082                 e1000_irq_enable(adapter);
2083
2084         if ((adapter->hw.mng_cookie.status &
2085              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2086             (vid == adapter->mng_vlan_id)) {
2087                 /* release control to f/w */
2088                 e1000_release_hw_control(adapter);
2089                 return;
2090         }
2091
2092         /* remove VID from filter table */
2093         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2094                 index = (vid >> 5) & 0x7F;
2095                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2096                 vfta &= ~(1 << (vid & 0x1F));
2097                 hw->mac.ops.write_vfta(hw, index, vfta);
2098         }
2099 }
2100
2101 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2102 {
2103         struct net_device *netdev = adapter->netdev;
2104         u16 vid = adapter->hw.mng_cookie.vlan_id;
2105         u16 old_vid = adapter->mng_vlan_id;
2106
2107         if (!adapter->vlgrp)
2108                 return;
2109
2110         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2111                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2112                 if (adapter->hw.mng_cookie.status &
2113                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2114                         e1000_vlan_rx_add_vid(netdev, vid);
2115                         adapter->mng_vlan_id = vid;
2116                 }
2117
2118                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2119                                 (vid != old_vid) &&
2120                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2121                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2122         } else {
2123                 adapter->mng_vlan_id = vid;
2124         }
2125 }
2126
2127
2128 static void e1000_vlan_rx_register(struct net_device *netdev,
2129                                    struct vlan_group *grp)
2130 {
2131         struct e1000_adapter *adapter = netdev_priv(netdev);
2132         struct e1000_hw *hw = &adapter->hw;
2133         u32 ctrl, rctl;
2134
2135         if (!test_bit(__E1000_DOWN, &adapter->state))
2136                 e1000_irq_disable(adapter);
2137         adapter->vlgrp = grp;
2138
2139         if (grp) {
2140                 /* enable VLAN tag insert/strip */
2141                 ctrl = er32(CTRL);
2142                 ctrl |= E1000_CTRL_VME;
2143                 ew32(CTRL, ctrl);
2144
2145                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2146                         /* enable VLAN receive filtering */
2147                         rctl = er32(RCTL);
2148                         rctl &= ~E1000_RCTL_CFIEN;
2149                         ew32(RCTL, rctl);
2150                         e1000_update_mng_vlan(adapter);
2151                 }
2152         } else {
2153                 /* disable VLAN tag insert/strip */
2154                 ctrl = er32(CTRL);
2155                 ctrl &= ~E1000_CTRL_VME;
2156                 ew32(CTRL, ctrl);
2157
2158                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2159                         if (adapter->mng_vlan_id !=
2160                             (u16)E1000_MNG_VLAN_NONE) {
2161                                 e1000_vlan_rx_kill_vid(netdev,
2162                                                        adapter->mng_vlan_id);
2163                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2164                         }
2165                 }
2166         }
2167
2168         if (!test_bit(__E1000_DOWN, &adapter->state))
2169                 e1000_irq_enable(adapter);
2170 }
2171
2172 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2173 {
2174         u16 vid;
2175
2176         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2177
2178         if (!adapter->vlgrp)
2179                 return;
2180
2181         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2182                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2183                         continue;
2184                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2185         }
2186 }
2187
2188 static void e1000_init_manageability(struct e1000_adapter *adapter)
2189 {
2190         struct e1000_hw *hw = &adapter->hw;
2191         u32 manc, manc2h;
2192
2193         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2194                 return;
2195
2196         manc = er32(MANC);
2197
2198         /*
2199          * enable receiving management packets to the host. this will probably
2200          * generate destination unreachable messages from the host OS, but
2201          * the packets will be handled on SMBUS
2202          */
2203         manc |= E1000_MANC_EN_MNG2HOST;
2204         manc2h = er32(MANC2H);
2205 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2206 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2207         manc2h |= E1000_MNG2HOST_PORT_623;
2208         manc2h |= E1000_MNG2HOST_PORT_664;
2209         ew32(MANC2H, manc2h);
2210         ew32(MANC, manc);
2211 }
2212
2213 /**
2214  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2215  * @adapter: board private structure
2216  *
2217  * Configure the Tx unit of the MAC after a reset.
2218  **/
2219 static void e1000_configure_tx(struct e1000_adapter *adapter)
2220 {
2221         struct e1000_hw *hw = &adapter->hw;
2222         struct e1000_ring *tx_ring = adapter->tx_ring;
2223         u64 tdba;
2224         u32 tdlen, tctl, tipg, tarc;
2225         u32 ipgr1, ipgr2;
2226
2227         /* Setup the HW Tx Head and Tail descriptor pointers */
2228         tdba = tx_ring->dma;
2229         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2230         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2231         ew32(TDBAH, (tdba >> 32));
2232         ew32(TDLEN, tdlen);
2233         ew32(TDH, 0);
2234         ew32(TDT, 0);
2235         tx_ring->head = E1000_TDH;
2236         tx_ring->tail = E1000_TDT;
2237
2238         /* Set the default values for the Tx Inter Packet Gap timer */
2239         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2240         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2241         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2242
2243         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2244                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2245
2246         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2247         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2248         ew32(TIPG, tipg);
2249
2250         /* Set the Tx Interrupt Delay register */
2251         ew32(TIDV, adapter->tx_int_delay);
2252         /* Tx irq moderation */
2253         ew32(TADV, adapter->tx_abs_int_delay);
2254
2255         /* Program the Transmit Control Register */
2256         tctl = er32(TCTL);
2257         tctl &= ~E1000_TCTL_CT;
2258         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2259                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2260
2261         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2262                 tarc = er32(TARC(0));
2263                 /*
2264                  * set the speed mode bit, we'll clear it if we're not at
2265                  * gigabit link later
2266                  */
2267 #define SPEED_MODE_BIT (1 << 21)
2268                 tarc |= SPEED_MODE_BIT;
2269                 ew32(TARC(0), tarc);
2270         }
2271
2272         /* errata: program both queues to unweighted RR */
2273         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2274                 tarc = er32(TARC(0));
2275                 tarc |= 1;
2276                 ew32(TARC(0), tarc);
2277                 tarc = er32(TARC(1));
2278                 tarc |= 1;
2279                 ew32(TARC(1), tarc);
2280         }
2281
2282         /* Setup Transmit Descriptor Settings for eop descriptor */
2283         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2284
2285         /* only set IDE if we are delaying interrupts using the timers */
2286         if (adapter->tx_int_delay)
2287                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2288
2289         /* enable Report Status bit */
2290         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2291
2292         ew32(TCTL, tctl);
2293
2294         e1000e_config_collision_dist(hw);
2295
2296         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
2297 }
2298
2299 /**
2300  * e1000_setup_rctl - configure the receive control registers
2301  * @adapter: Board private structure
2302  **/
2303 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2304                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2305 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2306 {
2307         struct e1000_hw *hw = &adapter->hw;
2308         u32 rctl, rfctl;
2309         u32 psrctl = 0;
2310         u32 pages = 0;
2311
2312         /* Program MC offset vector base */
2313         rctl = er32(RCTL);
2314         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2315         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2316                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2317                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2318
2319         /* Do not Store bad packets */
2320         rctl &= ~E1000_RCTL_SBP;
2321
2322         /* Enable Long Packet receive */
2323         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2324                 rctl &= ~E1000_RCTL_LPE;
2325         else
2326                 rctl |= E1000_RCTL_LPE;
2327
2328         /* Some systems expect that the CRC is included in SMBUS traffic. The
2329          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2330          * host memory when this is enabled
2331          */
2332         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2333                 rctl |= E1000_RCTL_SECRC;
2334
2335         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2336         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2337                 u16 phy_data;
2338
2339                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2340                 phy_data &= 0xfff8;
2341                 phy_data |= (1 << 2);
2342                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2343
2344                 e1e_rphy(hw, 22, &phy_data);
2345                 phy_data &= 0x0fff;
2346                 phy_data |= (1 << 14);
2347                 e1e_wphy(hw, 0x10, 0x2823);
2348                 e1e_wphy(hw, 0x11, 0x0003);
2349                 e1e_wphy(hw, 22, phy_data);
2350         }
2351
2352         /* Setup buffer sizes */
2353         rctl &= ~E1000_RCTL_SZ_4096;
2354         rctl |= E1000_RCTL_BSEX;
2355         switch (adapter->rx_buffer_len) {
2356         case 2048:
2357         default:
2358                 rctl |= E1000_RCTL_SZ_2048;
2359                 rctl &= ~E1000_RCTL_BSEX;
2360                 break;
2361         case 4096:
2362                 rctl |= E1000_RCTL_SZ_4096;
2363                 break;
2364         case 8192:
2365                 rctl |= E1000_RCTL_SZ_8192;
2366                 break;
2367         case 16384:
2368                 rctl |= E1000_RCTL_SZ_16384;
2369                 break;
2370         }
2371
2372         /*
2373          * 82571 and greater support packet-split where the protocol
2374          * header is placed in skb->data and the packet data is
2375          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2376          * In the case of a non-split, skb->data is linearly filled,
2377          * followed by the page buffers.  Therefore, skb->data is
2378          * sized to hold the largest protocol header.
2379          *
2380          * allocations using alloc_page take too long for regular MTU
2381          * so only enable packet split for jumbo frames
2382          *
2383          * Using pages when the page size is greater than 16k wastes
2384          * a lot of memory, since we allocate 3 pages at all times
2385          * per packet.
2386          */
2387         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2388         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2389             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2390                 adapter->rx_ps_pages = pages;
2391         else
2392                 adapter->rx_ps_pages = 0;
2393
2394         if (adapter->rx_ps_pages) {
2395                 /* Configure extra packet-split registers */
2396                 rfctl = er32(RFCTL);
2397                 rfctl |= E1000_RFCTL_EXTEN;
2398                 /*
2399                  * disable packet split support for IPv6 extension headers,
2400                  * because some malformed IPv6 headers can hang the Rx
2401                  */
2402                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2403                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2404
2405                 ew32(RFCTL, rfctl);
2406
2407                 /* Enable Packet split descriptors */
2408                 rctl |= E1000_RCTL_DTYP_PS;
2409
2410                 psrctl |= adapter->rx_ps_bsize0 >>
2411                         E1000_PSRCTL_BSIZE0_SHIFT;
2412
2413                 switch (adapter->rx_ps_pages) {
2414                 case 3:
2415                         psrctl |= PAGE_SIZE <<
2416                                 E1000_PSRCTL_BSIZE3_SHIFT;
2417                 case 2:
2418                         psrctl |= PAGE_SIZE <<
2419                                 E1000_PSRCTL_BSIZE2_SHIFT;
2420                 case 1:
2421                         psrctl |= PAGE_SIZE >>
2422                                 E1000_PSRCTL_BSIZE1_SHIFT;
2423                         break;
2424                 }
2425
2426                 ew32(PSRCTL, psrctl);
2427         }
2428
2429         ew32(RCTL, rctl);
2430         /* just started the receive unit, no need to restart */
2431         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2432 }
2433
2434 /**
2435  * e1000_configure_rx - Configure Receive Unit after Reset
2436  * @adapter: board private structure
2437  *
2438  * Configure the Rx unit of the MAC after a reset.
2439  **/
2440 static void e1000_configure_rx(struct e1000_adapter *adapter)
2441 {
2442         struct e1000_hw *hw = &adapter->hw;
2443         struct e1000_ring *rx_ring = adapter->rx_ring;
2444         u64 rdba;
2445         u32 rdlen, rctl, rxcsum, ctrl_ext;
2446
2447         if (adapter->rx_ps_pages) {
2448                 /* this is a 32 byte descriptor */
2449                 rdlen = rx_ring->count *
2450                         sizeof(union e1000_rx_desc_packet_split);
2451                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2452                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2453         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2454                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2455                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2456                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2457         } else {
2458                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2459                 adapter->clean_rx = e1000_clean_rx_irq;
2460                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2461         }
2462
2463         /* disable receives while setting up the descriptors */
2464         rctl = er32(RCTL);
2465         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2466         e1e_flush();
2467         msleep(10);
2468
2469         /* set the Receive Delay Timer Register */
2470         ew32(RDTR, adapter->rx_int_delay);
2471
2472         /* irq moderation */
2473         ew32(RADV, adapter->rx_abs_int_delay);
2474         if (adapter->itr_setting != 0)
2475                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2476
2477         ctrl_ext = er32(CTRL_EXT);
2478         /* Auto-Mask interrupts upon ICR access */
2479         ctrl_ext |= E1000_CTRL_EXT_IAME;
2480         ew32(IAM, 0xffffffff);
2481         ew32(CTRL_EXT, ctrl_ext);
2482         e1e_flush();
2483
2484         /*
2485          * Setup the HW Rx Head and Tail Descriptor Pointers and
2486          * the Base and Length of the Rx Descriptor Ring
2487          */
2488         rdba = rx_ring->dma;
2489         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2490         ew32(RDBAH, (rdba >> 32));
2491         ew32(RDLEN, rdlen);
2492         ew32(RDH, 0);
2493         ew32(RDT, 0);
2494         rx_ring->head = E1000_RDH;
2495         rx_ring->tail = E1000_RDT;
2496
2497         /* Enable Receive Checksum Offload for TCP and UDP */
2498         rxcsum = er32(RXCSUM);
2499         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2500                 rxcsum |= E1000_RXCSUM_TUOFL;
2501
2502                 /*
2503                  * IPv4 payload checksum for UDP fragments must be
2504                  * used in conjunction with packet-split.
2505                  */
2506                 if (adapter->rx_ps_pages)
2507                         rxcsum |= E1000_RXCSUM_IPPCSE;
2508         } else {
2509                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2510                 /* no need to clear IPPCSE as it defaults to 0 */
2511         }
2512         ew32(RXCSUM, rxcsum);
2513
2514         /*
2515          * Enable early receives on supported devices, only takes effect when
2516          * packet size is equal or larger than the specified value (in 8 byte
2517          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2518          */
2519         if (adapter->flags & FLAG_HAS_ERT) {
2520                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2521                         u32 rxdctl = er32(RXDCTL(0));
2522                         ew32(RXDCTL(0), rxdctl | 0x3);
2523                         ew32(ERT, E1000_ERT_2048 | (1 << 13));
2524                         /*
2525                          * With jumbo frames and early-receive enabled,
2526                          * excessive C-state transition latencies result in
2527                          * dropped transactions.
2528                          */
2529                         pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2530                                                   adapter->netdev->name, 55);
2531                 } else {
2532                         pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2533                                                   adapter->netdev->name,
2534                                                   PM_QOS_DEFAULT_VALUE);
2535                 }
2536         }
2537
2538         /* Enable Receives */
2539         ew32(RCTL, rctl);
2540 }
2541
2542 /**
2543  *  e1000_update_mc_addr_list - Update Multicast addresses
2544  *  @hw: pointer to the HW structure
2545  *  @mc_addr_list: array of multicast addresses to program
2546  *  @mc_addr_count: number of multicast addresses to program
2547  *
2548  *  Updates the Multicast Table Array.
2549  *  The caller must have a packed mc_addr_list of multicast addresses.
2550  **/
2551 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2552                                       u32 mc_addr_count)
2553 {
2554         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
2555 }
2556
2557 /**
2558  * e1000_set_multi - Multicast and Promiscuous mode set
2559  * @netdev: network interface device structure
2560  *
2561  * The set_multi entry point is called whenever the multicast address
2562  * list or the network interface flags are updated.  This routine is
2563  * responsible for configuring the hardware for proper multicast,
2564  * promiscuous mode, and all-multi behavior.
2565  **/
2566 static void e1000_set_multi(struct net_device *netdev)
2567 {
2568         struct e1000_adapter *adapter = netdev_priv(netdev);
2569         struct e1000_hw *hw = &adapter->hw;
2570         struct netdev_hw_addr *ha;
2571         u8  *mta_list;
2572         u32 rctl;
2573         int i;
2574
2575         /* Check for Promiscuous and All Multicast modes */
2576
2577         rctl = er32(RCTL);
2578
2579         if (netdev->flags & IFF_PROMISC) {
2580                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2581                 rctl &= ~E1000_RCTL_VFE;
2582         } else {
2583                 if (netdev->flags & IFF_ALLMULTI) {
2584                         rctl |= E1000_RCTL_MPE;
2585                         rctl &= ~E1000_RCTL_UPE;
2586                 } else {
2587                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2588                 }
2589                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2590                         rctl |= E1000_RCTL_VFE;
2591         }
2592
2593         ew32(RCTL, rctl);
2594
2595         if (!netdev_mc_empty(netdev)) {
2596                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
2597                 if (!mta_list)
2598                         return;
2599
2600                 /* prepare a packed array of only addresses. */
2601                 i = 0;
2602                 netdev_for_each_mc_addr(ha, netdev)
2603                         memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
2604
2605                 e1000_update_mc_addr_list(hw, mta_list, i);
2606                 kfree(mta_list);
2607         } else {
2608                 /*
2609                  * if we're called from probe, we might not have
2610                  * anything to do here, so clear out the list
2611                  */
2612                 e1000_update_mc_addr_list(hw, NULL, 0);
2613         }
2614 }
2615
2616 /**
2617  * e1000_configure - configure the hardware for Rx and Tx
2618  * @adapter: private board structure
2619  **/
2620 static void e1000_configure(struct e1000_adapter *adapter)
2621 {
2622         e1000_set_multi(adapter->netdev);
2623
2624         e1000_restore_vlan(adapter);
2625         e1000_init_manageability(adapter);
2626
2627         e1000_configure_tx(adapter);
2628         e1000_setup_rctl(adapter);
2629         e1000_configure_rx(adapter);
2630         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2631 }
2632
2633 /**
2634  * e1000e_power_up_phy - restore link in case the phy was powered down
2635  * @adapter: address of board private structure
2636  *
2637  * The phy may be powered down to save power and turn off link when the
2638  * driver is unloaded and wake on lan is not enabled (among others)
2639  * *** this routine MUST be followed by a call to e1000e_reset ***
2640  **/
2641 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2642 {
2643         if (adapter->hw.phy.ops.power_up)
2644                 adapter->hw.phy.ops.power_up(&adapter->hw);
2645
2646         adapter->hw.mac.ops.setup_link(&adapter->hw);
2647 }
2648
2649 /**
2650  * e1000_power_down_phy - Power down the PHY
2651  *
2652  * Power down the PHY so no link is implied when interface is down.
2653  * The PHY cannot be powered down if management or WoL is active.
2654  */
2655 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2656 {
2657         /* WoL is enabled */
2658         if (adapter->wol)
2659                 return;
2660
2661         if (adapter->hw.phy.ops.power_down)
2662                 adapter->hw.phy.ops.power_down(&adapter->hw);
2663 }
2664
2665 /**
2666  * e1000e_reset - bring the hardware into a known good state
2667  *
2668  * This function boots the hardware and enables some settings that
2669  * require a configuration cycle of the hardware - those cannot be
2670  * set/changed during runtime. After reset the device needs to be
2671  * properly configured for Rx, Tx etc.
2672  */
2673 void e1000e_reset(struct e1000_adapter *adapter)
2674 {
2675         struct e1000_mac_info *mac = &adapter->hw.mac;
2676         struct e1000_fc_info *fc = &adapter->hw.fc;
2677         struct e1000_hw *hw = &adapter->hw;
2678         u32 tx_space, min_tx_space, min_rx_space;
2679         u32 pba = adapter->pba;
2680         u16 hwm;
2681
2682         /* reset Packet Buffer Allocation to default */
2683         ew32(PBA, pba);
2684
2685         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2686                 /*
2687                  * To maintain wire speed transmits, the Tx FIFO should be
2688                  * large enough to accommodate two full transmit packets,
2689                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2690                  * the Rx FIFO should be large enough to accommodate at least
2691                  * one full receive packet and is similarly rounded up and
2692                  * expressed in KB.
2693                  */
2694                 pba = er32(PBA);
2695                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2696                 tx_space = pba >> 16;
2697                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2698                 pba &= 0xffff;
2699                 /*
2700                  * the Tx fifo also stores 16 bytes of information about the tx
2701                  * but don't include ethernet FCS because hardware appends it
2702                  */
2703                 min_tx_space = (adapter->max_frame_size +
2704                                 sizeof(struct e1000_tx_desc) -
2705                                 ETH_FCS_LEN) * 2;
2706                 min_tx_space = ALIGN(min_tx_space, 1024);
2707                 min_tx_space >>= 10;
2708                 /* software strips receive CRC, so leave room for it */
2709                 min_rx_space = adapter->max_frame_size;
2710                 min_rx_space = ALIGN(min_rx_space, 1024);
2711                 min_rx_space >>= 10;
2712
2713                 /*
2714                  * If current Tx allocation is less than the min Tx FIFO size,
2715                  * and the min Tx FIFO size is less than the current Rx FIFO
2716                  * allocation, take space away from current Rx allocation
2717                  */
2718                 if ((tx_space < min_tx_space) &&
2719                     ((min_tx_space - tx_space) < pba)) {
2720                         pba -= min_tx_space - tx_space;
2721
2722                         /*
2723                          * if short on Rx space, Rx wins and must trump tx
2724                          * adjustment or use Early Receive if available
2725                          */
2726                         if ((pba < min_rx_space) &&
2727                             (!(adapter->flags & FLAG_HAS_ERT)))
2728                                 /* ERT enabled in e1000_configure_rx */
2729                                 pba = min_rx_space;
2730                 }
2731
2732                 ew32(PBA, pba);
2733         }
2734
2735
2736         /*
2737          * flow control settings
2738          *
2739          * The high water mark must be low enough to fit one full frame
2740          * (or the size used for early receive) above it in the Rx FIFO.
2741          * Set it to the lower of:
2742          * - 90% of the Rx FIFO size, and
2743          * - the full Rx FIFO size minus the early receive size (for parts
2744          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2745          * - the full Rx FIFO size minus one full frame
2746          */
2747         if (hw->mac.type == e1000_pchlan) {
2748                 /*
2749                  * Workaround PCH LOM adapter hangs with certain network
2750                  * loads.  If hangs persist, try disabling Tx flow control.
2751                  */
2752                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2753                         fc->high_water = 0x3500;
2754                         fc->low_water  = 0x1500;
2755                 } else {
2756                         fc->high_water = 0x5000;
2757                         fc->low_water  = 0x3000;
2758                 }
2759         } else {
2760                 if ((adapter->flags & FLAG_HAS_ERT) &&
2761                     (adapter->netdev->mtu > ETH_DATA_LEN))
2762                         hwm = min(((pba << 10) * 9 / 10),
2763                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
2764                 else
2765                         hwm = min(((pba << 10) * 9 / 10),
2766                                   ((pba << 10) - adapter->max_frame_size));
2767
2768                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
2769                 fc->low_water = fc->high_water - 8;
2770         }
2771
2772         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2773                 fc->pause_time = 0xFFFF;
2774         else
2775                 fc->pause_time = E1000_FC_PAUSE_TIME;
2776         fc->send_xon = 1;
2777         fc->current_mode = fc->requested_mode;
2778
2779         /* Allow time for pending master requests to run */
2780         mac->ops.reset_hw(hw);
2781
2782         /*
2783          * For parts with AMT enabled, let the firmware know
2784          * that the network interface is in control
2785          */
2786         if (adapter->flags & FLAG_HAS_AMT)
2787                 e1000_get_hw_control(adapter);
2788
2789         ew32(WUC, 0);
2790         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP)
2791                 e1e_wphy(&adapter->hw, BM_WUC, 0);
2792
2793         if (mac->ops.init_hw(hw))
2794                 e_err("Hardware Error\n");
2795
2796         /* additional part of the flow-control workaround above */
2797         if (hw->mac.type == e1000_pchlan)
2798                 ew32(FCRTV_PCH, 0x1000);
2799
2800         e1000_update_mng_vlan(adapter);
2801
2802         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2803         ew32(VET, ETH_P_8021Q);
2804
2805         e1000e_reset_adaptive(hw);
2806         e1000_get_phy_info(hw);
2807
2808         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
2809             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2810                 u16 phy_data = 0;
2811                 /*
2812                  * speed up time to link by disabling smart power down, ignore
2813                  * the return value of this function because there is nothing
2814                  * different we would do if it failed
2815                  */
2816                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2817                 phy_data &= ~IGP02E1000_PM_SPD;
2818                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2819         }
2820 }
2821
2822 int e1000e_up(struct e1000_adapter *adapter)
2823 {
2824         struct e1000_hw *hw = &adapter->hw;
2825
2826         /* DMA latency requirement to workaround early-receive/jumbo issue */
2827         if (adapter->flags & FLAG_HAS_ERT)
2828                 pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY,
2829                                        adapter->netdev->name,
2830                                        PM_QOS_DEFAULT_VALUE);
2831
2832         /* hardware has been reset, we need to reload some things */
2833         e1000_configure(adapter);
2834
2835         clear_bit(__E1000_DOWN, &adapter->state);
2836
2837         napi_enable(&adapter->napi);
2838         if (adapter->msix_entries)
2839                 e1000_configure_msix(adapter);
2840         e1000_irq_enable(adapter);
2841
2842         netif_wake_queue(adapter->netdev);
2843
2844         /* fire a link change interrupt to start the watchdog */
2845         ew32(ICS, E1000_ICS_LSC);
2846         return 0;
2847 }
2848
2849 void e1000e_down(struct e1000_adapter *adapter)
2850 {
2851         struct net_device *netdev = adapter->netdev;
2852         struct e1000_hw *hw = &adapter->hw;
2853         u32 tctl, rctl;
2854
2855         /*
2856          * signal that we're down so the interrupt handler does not
2857          * reschedule our watchdog timer
2858          */
2859         set_bit(__E1000_DOWN, &adapter->state);
2860
2861         /* disable receives in the hardware */
2862         rctl = er32(RCTL);
2863         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2864         /* flush and sleep below */
2865
2866         netif_stop_queue(netdev);
2867
2868         /* disable transmits in the hardware */
2869         tctl = er32(TCTL);
2870         tctl &= ~E1000_TCTL_EN;
2871         ew32(TCTL, tctl);
2872         /* flush both disables and wait for them to finish */
2873         e1e_flush();
2874         msleep(10);
2875
2876         napi_disable(&adapter->napi);
2877         e1000_irq_disable(adapter);
2878
2879         del_timer_sync(&adapter->watchdog_timer);
2880         del_timer_sync(&adapter->phy_info_timer);
2881
2882         netdev->tx_queue_len = adapter->tx_queue_len;
2883         netif_carrier_off(netdev);
2884         adapter->link_speed = 0;
2885         adapter->link_duplex = 0;
2886
2887         if (!pci_channel_offline(adapter->pdev))
2888                 e1000e_reset(adapter);
2889         e1000_clean_tx_ring(adapter);
2890         e1000_clean_rx_ring(adapter);
2891
2892         if (adapter->flags & FLAG_HAS_ERT)
2893                 pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY,
2894                                           adapter->netdev->name);
2895
2896         /*
2897          * TODO: for power management, we could drop the link and
2898          * pci_disable_device here.
2899          */
2900 }
2901
2902 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2903 {
2904         might_sleep();
2905         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2906                 msleep(1);
2907         e1000e_down(adapter);
2908         e1000e_up(adapter);
2909         clear_bit(__E1000_RESETTING, &adapter->state);
2910 }
2911
2912 /**
2913  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2914  * @adapter: board private structure to initialize
2915  *
2916  * e1000_sw_init initializes the Adapter private data structure.
2917  * Fields are initialized based on PCI device information and
2918  * OS network device settings (MTU size).
2919  **/
2920 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2921 {
2922         struct net_device *netdev = adapter->netdev;
2923
2924         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2925         adapter->rx_ps_bsize0 = 128;
2926         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2927         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2928
2929         e1000e_set_interrupt_capability(adapter);
2930
2931         if (e1000_alloc_queues(adapter))
2932                 return -ENOMEM;
2933
2934         /* Explicitly disable IRQ since the NIC can be in any state. */
2935         e1000_irq_disable(adapter);
2936
2937         set_bit(__E1000_DOWN, &adapter->state);
2938         return 0;
2939 }
2940
2941 /**
2942  * e1000_intr_msi_test - Interrupt Handler
2943  * @irq: interrupt number
2944  * @data: pointer to a network interface device structure
2945  **/
2946 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2947 {
2948         struct net_device *netdev = data;
2949         struct e1000_adapter *adapter = netdev_priv(netdev);
2950         struct e1000_hw *hw = &adapter->hw;
2951         u32 icr = er32(ICR);
2952
2953         e_dbg("icr is %08X\n", icr);
2954         if (icr & E1000_ICR_RXSEQ) {
2955                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2956                 wmb();
2957         }
2958
2959         return IRQ_HANDLED;
2960 }
2961
2962 /**
2963  * e1000_test_msi_interrupt - Returns 0 for successful test
2964  * @adapter: board private struct
2965  *
2966  * code flow taken from tg3.c
2967  **/
2968 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2969 {
2970         struct net_device *netdev = adapter->netdev;
2971         struct e1000_hw *hw = &adapter->hw;
2972         int err;
2973
2974         /* poll_enable hasn't been called yet, so don't need disable */
2975         /* clear any pending events */
2976         er32(ICR);
2977
2978         /* free the real vector and request a test handler */
2979         e1000_free_irq(adapter);
2980         e1000e_reset_interrupt_capability(adapter);
2981
2982         /* Assume that the test fails, if it succeeds then the test
2983          * MSI irq handler will unset this flag */
2984         adapter->flags |= FLAG_MSI_TEST_FAILED;
2985
2986         err = pci_enable_msi(adapter->pdev);
2987         if (err)
2988                 goto msi_test_failed;
2989
2990         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
2991                           netdev->name, netdev);
2992         if (err) {
2993                 pci_disable_msi(adapter->pdev);
2994                 goto msi_test_failed;
2995         }
2996
2997         wmb();
2998
2999         e1000_irq_enable(adapter);
3000
3001         /* fire an unusual interrupt on the test handler */
3002         ew32(ICS, E1000_ICS_RXSEQ);
3003         e1e_flush();
3004         msleep(50);
3005
3006         e1000_irq_disable(adapter);
3007
3008         rmb();
3009
3010         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3011                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3012                 err = -EIO;
3013                 e_info("MSI interrupt test failed!\n");
3014         }
3015
3016         free_irq(adapter->pdev->irq, netdev);
3017         pci_disable_msi(adapter->pdev);
3018
3019         if (err == -EIO)
3020                 goto msi_test_failed;
3021
3022         /* okay so the test worked, restore settings */
3023         e_dbg("MSI interrupt test succeeded!\n");
3024 msi_test_failed:
3025         e1000e_set_interrupt_capability(adapter);
3026         e1000_request_irq(adapter);
3027         return err;
3028 }
3029
3030 /**
3031  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3032  * @adapter: board private struct
3033  *
3034  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3035  **/
3036 static int e1000_test_msi(struct e1000_adapter *adapter)
3037 {
3038         int err;
3039         u16 pci_cmd;
3040
3041         if (!(adapter->flags & FLAG_MSI_ENABLED))
3042                 return 0;
3043
3044         /* disable SERR in case the MSI write causes a master abort */
3045         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3046         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3047                               pci_cmd & ~PCI_COMMAND_SERR);
3048
3049         err = e1000_test_msi_interrupt(adapter);
3050
3051         /* restore previous setting of command word */
3052         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3053
3054         /* success ! */
3055         if (!err)
3056                 return 0;
3057
3058         /* EIO means MSI test failed */
3059         if (err != -EIO)
3060                 return err;
3061
3062         /* back to INTx mode */
3063         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3064
3065         e1000_free_irq(adapter);
3066
3067         err = e1000_request_irq(adapter);
3068
3069         return err;
3070 }
3071
3072 /**
3073  * e1000_open - Called when a network interface is made active
3074  * @netdev: network interface device structure
3075  *
3076  * Returns 0 on success, negative value on failure
3077  *
3078  * The open entry point is called when a network interface is made
3079  * active by the system (IFF_UP).  At this point all resources needed
3080  * for transmit and receive operations are allocated, the interrupt
3081  * handler is registered with the OS, the watchdog timer is started,
3082  * and the stack is notified that the interface is ready.
3083  **/
3084 static int e1000_open(struct net_device *netdev)
3085 {
3086         struct e1000_adapter *adapter = netdev_priv(netdev);
3087         struct e1000_hw *hw = &adapter->hw;
3088         struct pci_dev *pdev = adapter->pdev;
3089         int err;
3090
3091         /* disallow open during test */
3092         if (test_bit(__E1000_TESTING, &adapter->state))
3093                 return -EBUSY;
3094
3095         pm_runtime_get_sync(&pdev->dev);
3096
3097         netif_carrier_off(netdev);
3098
3099         /* allocate transmit descriptors */
3100         err = e1000e_setup_tx_resources(adapter);
3101         if (err)
3102                 goto err_setup_tx;
3103
3104         /* allocate receive descriptors */
3105         err = e1000e_setup_rx_resources(adapter);
3106         if (err)
3107                 goto err_setup_rx;
3108
3109         e1000e_power_up_phy(adapter);
3110
3111         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3112         if ((adapter->hw.mng_cookie.status &
3113              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3114                 e1000_update_mng_vlan(adapter);
3115
3116         /*
3117          * If AMT is enabled, let the firmware know that the network
3118          * interface is now open
3119          */
3120         if (adapter->flags & FLAG_HAS_AMT)
3121                 e1000_get_hw_control(adapter);
3122
3123         /*
3124          * before we allocate an interrupt, we must be ready to handle it.
3125          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3126          * as soon as we call pci_request_irq, so we have to setup our
3127          * clean_rx handler before we do so.
3128          */
3129         e1000_configure(adapter);
3130
3131         err = e1000_request_irq(adapter);
3132         if (err)
3133                 goto err_req_irq;
3134
3135         /*
3136          * Work around PCIe errata with MSI interrupts causing some chipsets to
3137          * ignore e1000e MSI messages, which means we need to test our MSI
3138          * interrupt now
3139          */
3140         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3141                 err = e1000_test_msi(adapter);
3142                 if (err) {
3143                         e_err("Interrupt allocation failed\n");
3144                         goto err_req_irq;
3145                 }
3146         }
3147
3148         /* From here on the code is the same as e1000e_up() */
3149         clear_bit(__E1000_DOWN, &adapter->state);
3150
3151         napi_enable(&adapter->napi);
3152
3153         e1000_irq_enable(adapter);
3154
3155         netif_start_queue(netdev);
3156
3157         adapter->idle_check = true;
3158         pm_runtime_put(&pdev->dev);
3159
3160         /* fire a link status change interrupt to start the watchdog */
3161         ew32(ICS, E1000_ICS_LSC);
3162
3163         return 0;
3164
3165 err_req_irq:
3166         e1000_release_hw_control(adapter);
3167         e1000_power_down_phy(adapter);
3168         e1000e_free_rx_resources(adapter);
3169 err_setup_rx:
3170         e1000e_free_tx_resources(adapter);
3171 err_setup_tx:
3172         e1000e_reset(adapter);
3173         pm_runtime_put_sync(&pdev->dev);
3174
3175         return err;
3176 }
3177
3178 /**
3179  * e1000_close - Disables a network interface
3180  * @netdev: network interface device structure
3181  *
3182  * Returns 0, this is not allowed to fail
3183  *
3184  * The close entry point is called when an interface is de-activated
3185  * by the OS.  The hardware is still under the drivers control, but
3186  * needs to be disabled.  A global MAC reset is issued to stop the
3187  * hardware, and all transmit and receive resources are freed.
3188  **/
3189 static int e1000_close(struct net_device *netdev)
3190 {
3191         struct e1000_adapter *adapter = netdev_priv(netdev);
3192         struct pci_dev *pdev = adapter->pdev;
3193
3194         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3195
3196         pm_runtime_get_sync(&pdev->dev);
3197
3198         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3199                 e1000e_down(adapter);
3200                 e1000_free_irq(adapter);
3201         }
3202         e1000_power_down_phy(adapter);
3203
3204         e1000e_free_tx_resources(adapter);
3205         e1000e_free_rx_resources(adapter);
3206
3207         /*
3208          * kill manageability vlan ID if supported, but not if a vlan with
3209          * the same ID is registered on the host OS (let 8021q kill it)
3210          */
3211         if ((adapter->hw.mng_cookie.status &
3212                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3213              !(adapter->vlgrp &&
3214                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3215                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3216
3217         /*
3218          * If AMT is enabled, let the firmware know that the network
3219          * interface is now closed
3220          */
3221         if (adapter->flags & FLAG_HAS_AMT)
3222                 e1000_release_hw_control(adapter);
3223
3224         pm_runtime_put_sync(&pdev->dev);
3225
3226         return 0;
3227 }
3228 /**
3229  * e1000_set_mac - Change the Ethernet Address of the NIC
3230  * @netdev: network interface device structure
3231  * @p: pointer to an address structure
3232  *
3233  * Returns 0 on success, negative on failure
3234  **/
3235 static int e1000_set_mac(struct net_device *netdev, void *p)
3236 {
3237         struct e1000_adapter *adapter = netdev_priv(netdev);
3238         struct sockaddr *addr = p;
3239
3240         if (!is_valid_ether_addr(addr->sa_data))
3241                 return -EADDRNOTAVAIL;
3242
3243         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3244         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3245
3246         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3247
3248         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3249                 /* activate the work around */
3250                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3251
3252                 /*
3253                  * Hold a copy of the LAA in RAR[14] This is done so that
3254                  * between the time RAR[0] gets clobbered  and the time it
3255                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3256                  * of the RARs and no incoming packets directed to this port
3257                  * are dropped. Eventually the LAA will be in RAR[0] and
3258                  * RAR[14]
3259                  */
3260                 e1000e_rar_set(&adapter->hw,
3261                               adapter->hw.mac.addr,
3262                               adapter->hw.mac.rar_entry_count - 1);
3263         }
3264
3265         return 0;
3266 }
3267
3268 /**
3269  * e1000e_update_phy_task - work thread to update phy
3270  * @work: pointer to our work struct
3271  *
3272  * this worker thread exists because we must acquire a
3273  * semaphore to read the phy, which we could msleep while
3274  * waiting for it, and we can't msleep in a timer.
3275  **/
3276 static void e1000e_update_phy_task(struct work_struct *work)
3277 {
3278         struct e1000_adapter *adapter = container_of(work,
3279                                         struct e1000_adapter, update_phy_task);
3280         e1000_get_phy_info(&adapter->hw);
3281 }
3282
3283 /*
3284  * Need to wait a few seconds after link up to get diagnostic information from
3285  * the phy
3286  */
3287 static void e1000_update_phy_info(unsigned long data)
3288 {
3289         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3290         schedule_work(&adapter->update_phy_task);
3291 }
3292
3293 /**
3294  * e1000e_update_stats - Update the board statistics counters
3295  * @adapter: board private structure
3296  **/
3297 void e1000e_update_stats(struct e1000_adapter *adapter)
3298 {
3299         struct net_device *netdev = adapter->netdev;
3300         struct e1000_hw *hw = &adapter->hw;
3301         struct pci_dev *pdev = adapter->pdev;
3302         u16 phy_data;
3303
3304         /*
3305          * Prevent stats update while adapter is being reset, or if the pci
3306          * connection is down.
3307          */
3308         if (adapter->link_speed == 0)
3309                 return;
3310         if (pci_channel_offline(pdev))
3311                 return;
3312
3313         adapter->stats.crcerrs += er32(CRCERRS);
3314         adapter->stats.gprc += er32(GPRC);
3315         adapter->stats.gorc += er32(GORCL);
3316         er32(GORCH); /* Clear gorc */
3317         adapter->stats.bprc += er32(BPRC);
3318         adapter->stats.mprc += er32(MPRC);
3319         adapter->stats.roc += er32(ROC);
3320
3321         adapter->stats.mpc += er32(MPC);
3322         if ((hw->phy.type == e1000_phy_82578) ||
3323             (hw->phy.type == e1000_phy_82577)) {
3324                 e1e_rphy(hw, HV_SCC_UPPER, &phy_data);
3325                 if (!e1e_rphy(hw, HV_SCC_LOWER, &phy_data))
3326                         adapter->stats.scc += phy_data;
3327
3328                 e1e_rphy(hw, HV_ECOL_UPPER, &phy_data);
3329                 if (!e1e_rphy(hw, HV_ECOL_LOWER, &phy_data))
3330                         adapter->stats.ecol += phy_data;
3331
3332                 e1e_rphy(hw, HV_MCC_UPPER, &phy_data);
3333                 if (!e1e_rphy(hw, HV_MCC_LOWER, &phy_data))
3334                         adapter->stats.mcc += phy_data;
3335
3336                 e1e_rphy(hw, HV_LATECOL_UPPER, &phy_data);
3337                 if (!e1e_rphy(hw, HV_LATECOL_LOWER, &phy_data))
3338                         adapter->stats.latecol += phy_data;
3339
3340                 e1e_rphy(hw, HV_DC_UPPER, &phy_data);
3341                 if (!e1e_rphy(hw, HV_DC_LOWER, &phy_data))
3342                         adapter->stats.dc += phy_data;
3343         } else {
3344                 adapter->stats.scc += er32(SCC);
3345                 adapter->stats.ecol += er32(ECOL);
3346                 adapter->stats.mcc += er32(MCC);
3347                 adapter->stats.latecol += er32(LATECOL);
3348                 adapter->stats.dc += er32(DC);
3349         }
3350         adapter->stats.xonrxc += er32(XONRXC);
3351         adapter->stats.xontxc += er32(XONTXC);
3352         adapter->stats.xoffrxc += er32(XOFFRXC);
3353         adapter->stats.xofftxc += er32(XOFFTXC);
3354         adapter->stats.gptc += er32(GPTC);
3355         adapter->stats.gotc += er32(GOTCL);
3356         er32(GOTCH); /* Clear gotc */
3357         adapter->stats.rnbc += er32(RNBC);
3358         adapter->stats.ruc += er32(RUC);
3359
3360         adapter->stats.mptc += er32(MPTC);
3361         adapter->stats.bptc += er32(BPTC);
3362
3363         /* used for adaptive IFS */
3364
3365         hw->mac.tx_packet_delta = er32(TPT);
3366         adapter->stats.tpt += hw->mac.tx_packet_delta;
3367         if ((hw->phy.type == e1000_phy_82578) ||
3368             (hw->phy.type == e1000_phy_82577)) {
3369                 e1e_rphy(hw, HV_COLC_UPPER, &phy_data);
3370                 if (!e1e_rphy(hw, HV_COLC_LOWER, &phy_data))
3371                         hw->mac.collision_delta = phy_data;
3372         } else {
3373                 hw->mac.collision_delta = er32(COLC);
3374         }
3375         adapter->stats.colc += hw->mac.collision_delta;
3376
3377         adapter->stats.algnerrc += er32(ALGNERRC);
3378         adapter->stats.rxerrc += er32(RXERRC);
3379         if ((hw->phy.type == e1000_phy_82578) ||
3380             (hw->phy.type == e1000_phy_82577)) {
3381                 e1e_rphy(hw, HV_TNCRS_UPPER, &phy_data);
3382                 if (!e1e_rphy(hw, HV_TNCRS_LOWER, &phy_data))
3383                         adapter->stats.tncrs += phy_data;
3384         } else {
3385                 if ((hw->mac.type != e1000_82574) &&
3386                     (hw->mac.type != e1000_82583))
3387                         adapter->stats.tncrs += er32(TNCRS);
3388         }
3389         adapter->stats.cexterr += er32(CEXTERR);
3390         adapter->stats.tsctc += er32(TSCTC);
3391         adapter->stats.tsctfc += er32(TSCTFC);
3392
3393         /* Fill out the OS statistics structure */
3394         netdev->stats.multicast = adapter->stats.mprc;
3395         netdev->stats.collisions = adapter->stats.colc;
3396
3397         /* Rx Errors */
3398
3399         /*
3400          * RLEC on some newer hardware can be incorrect so build
3401          * our own version based on RUC and ROC
3402          */
3403         netdev->stats.rx_errors = adapter->stats.rxerrc +
3404                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3405                 adapter->stats.ruc + adapter->stats.roc +
3406                 adapter->stats.cexterr;
3407         netdev->stats.rx_length_errors = adapter->stats.ruc +
3408                                               adapter->stats.roc;
3409         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3410         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3411         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3412
3413         /* Tx Errors */
3414         netdev->stats.tx_errors = adapter->stats.ecol +
3415                                        adapter->stats.latecol;
3416         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3417         netdev->stats.tx_window_errors = adapter->stats.latecol;
3418         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3419
3420         /* Tx Dropped needs to be maintained elsewhere */
3421
3422         /* Management Stats */
3423         adapter->stats.mgptc += er32(MGTPTC);
3424         adapter->stats.mgprc += er32(MGTPRC);
3425         adapter->stats.mgpdc += er32(MGTPDC);
3426 }
3427
3428 /**
3429  * e1000_phy_read_status - Update the PHY register status snapshot
3430  * @adapter: board private structure
3431  **/
3432 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3433 {
3434         struct e1000_hw *hw = &adapter->hw;
3435         struct e1000_phy_regs *phy = &adapter->phy_regs;
3436         int ret_val;
3437
3438         if ((er32(STATUS) & E1000_STATUS_LU) &&
3439             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3440                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3441                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3442                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3443                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3444                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3445                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3446                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3447                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3448                 if (ret_val)
3449                         e_warn("Error reading PHY register\n");
3450         } else {
3451                 /*
3452                  * Do not read PHY registers if link is not up
3453                  * Set values to typical power-on defaults
3454                  */
3455                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3456                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3457                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3458                              BMSR_ERCAP);
3459                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3460                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3461                 phy->lpa = 0;
3462                 phy->expansion = EXPANSION_ENABLENPAGE;
3463                 phy->ctrl1000 = ADVERTISE_1000FULL;
3464                 phy->stat1000 = 0;
3465                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3466         }
3467 }
3468
3469 static void e1000_print_link_info(struct e1000_adapter *adapter)
3470 {
3471         struct e1000_hw *hw = &adapter->hw;
3472         u32 ctrl = er32(CTRL);
3473
3474         /* Link status message must follow this format for user tools */
3475         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3476                "Flow Control: %s\n",
3477                adapter->netdev->name,
3478                adapter->link_speed,
3479                (adapter->link_duplex == FULL_DUPLEX) ?
3480                                 "Full Duplex" : "Half Duplex",
3481                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3482                                 "RX/TX" :
3483                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3484                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3485 }
3486
3487 bool e1000e_has_link(struct e1000_adapter *adapter)
3488 {
3489         struct e1000_hw *hw = &adapter->hw;
3490         bool link_active = 0;
3491         s32 ret_val = 0;
3492
3493         /*
3494          * get_link_status is set on LSC (link status) interrupt or
3495          * Rx sequence error interrupt.  get_link_status will stay
3496          * false until the check_for_link establishes link
3497          * for copper adapters ONLY
3498          */
3499         switch (hw->phy.media_type) {
3500         case e1000_media_type_copper:
3501                 if (hw->mac.get_link_status) {
3502                         ret_val = hw->mac.ops.check_for_link(hw);
3503                         link_active = !hw->mac.get_link_status;
3504                 } else {
3505                         link_active = 1;
3506                 }
3507                 break;
3508         case e1000_media_type_fiber:
3509                 ret_val = hw->mac.ops.check_for_link(hw);
3510                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3511                 break;
3512         case e1000_media_type_internal_serdes:
3513                 ret_val = hw->mac.ops.check_for_link(hw);
3514                 link_active = adapter->hw.mac.serdes_has_link;
3515                 break;
3516         default:
3517         case e1000_media_type_unknown:
3518                 break;
3519         }
3520
3521         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3522             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3523                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3524                 e_info("Gigabit has been disabled, downgrading speed\n");
3525         }
3526
3527         return link_active;
3528 }
3529
3530 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3531 {
3532         /* make sure the receive unit is started */
3533         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3534             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3535                 struct e1000_hw *hw = &adapter->hw;
3536                 u32 rctl = er32(RCTL);
3537                 ew32(RCTL, rctl | E1000_RCTL_EN);
3538                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3539         }
3540 }
3541
3542 /**
3543  * e1000_watchdog - Timer Call-back
3544  * @data: pointer to adapter cast into an unsigned long
3545  **/
3546 static void e1000_watchdog(unsigned long data)
3547 {
3548         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3549
3550         /* Do the rest outside of interrupt context */
3551         schedule_work(&adapter->watchdog_task);
3552
3553         /* TODO: make this use queue_delayed_work() */
3554 }
3555
3556 static void e1000_watchdog_task(struct work_struct *work)
3557 {
3558         struct e1000_adapter *adapter = container_of(work,
3559                                         struct e1000_adapter, watchdog_task);
3560         struct net_device *netdev = adapter->netdev;
3561         struct e1000_mac_info *mac = &adapter->hw.mac;
3562         struct e1000_phy_info *phy = &adapter->hw.phy;
3563         struct e1000_ring *tx_ring = adapter->tx_ring;
3564         struct e1000_hw *hw = &adapter->hw;
3565         u32 link, tctl;
3566         int tx_pending = 0;
3567
3568         link = e1000e_has_link(adapter);
3569         if ((netif_carrier_ok(netdev)) && link) {
3570                 /* Cancel scheduled suspend requests. */
3571                 pm_runtime_resume(netdev->dev.parent);
3572
3573                 e1000e_enable_receives(adapter);
3574                 goto link_up;
3575         }
3576
3577         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3578             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3579                 e1000_update_mng_vlan(adapter);
3580
3581         if (link) {
3582                 if (!netif_carrier_ok(netdev)) {
3583                         bool txb2b = 1;
3584
3585                         /* Cancel scheduled suspend requests. */
3586                         pm_runtime_resume(netdev->dev.parent);
3587
3588                         /* update snapshot of PHY registers on LSC */
3589                         e1000_phy_read_status(adapter);
3590                         mac->ops.get_link_up_info(&adapter->hw,
3591                                                    &adapter->link_speed,
3592                                                    &adapter->link_duplex);
3593                         e1000_print_link_info(adapter);
3594                         /*
3595                          * On supported PHYs, check for duplex mismatch only
3596                          * if link has autonegotiated at 10/100 half
3597                          */
3598                         if ((hw->phy.type == e1000_phy_igp_3 ||
3599                              hw->phy.type == e1000_phy_bm) &&
3600                             (hw->mac.autoneg == true) &&
3601                             (adapter->link_speed == SPEED_10 ||
3602                              adapter->link_speed == SPEED_100) &&
3603                             (adapter->link_duplex == HALF_DUPLEX)) {
3604                                 u16 autoneg_exp;
3605
3606                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3607
3608                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3609                                         e_info("Autonegotiated half duplex but"
3610                                                " link partner cannot autoneg. "
3611                                                " Try forcing full duplex if "
3612                                                "link gets many collisions.\n");
3613                         }
3614
3615                         /*
3616                          * tweak tx_queue_len according to speed/duplex
3617                          * and adjust the timeout factor
3618                          */
3619                         netdev->tx_queue_len = adapter->tx_queue_len;
3620                         adapter->tx_timeout_factor = 1;
3621                         switch (adapter->link_speed) {
3622                         case SPEED_10:
3623                                 txb2b = 0;
3624                                 netdev->tx_queue_len = 10;
3625                                 adapter->tx_timeout_factor = 16;
3626                                 break;
3627                         case SPEED_100:
3628                                 txb2b = 0;
3629                                 netdev->tx_queue_len = 100;
3630                                 adapter->tx_timeout_factor = 10;
3631                                 break;
3632                         }
3633
3634                         /*
3635                          * workaround: re-program speed mode bit after
3636                          * link-up event
3637                          */
3638                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3639                             !txb2b) {
3640                                 u32 tarc0;
3641                                 tarc0 = er32(TARC(0));
3642                                 tarc0 &= ~SPEED_MODE_BIT;
3643                                 ew32(TARC(0), tarc0);
3644                         }
3645
3646                         /*
3647                          * disable TSO for pcie and 10/100 speeds, to avoid
3648                          * some hardware issues
3649                          */
3650                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3651                                 switch (adapter->link_speed) {
3652                                 case SPEED_10:
3653                                 case SPEED_100:
3654                                         e_info("10/100 speed: disabling TSO\n");
3655                                         netdev->features &= ~NETIF_F_TSO;
3656                                         netdev->features &= ~NETIF_F_TSO6;
3657                                         break;
3658                                 case SPEED_1000:
3659                                         netdev->features |= NETIF_F_TSO;
3660                                         netdev->features |= NETIF_F_TSO6;
3661                                         break;
3662                                 default:
3663                                         /* oops */
3664                                         break;
3665                                 }
3666                         }
3667
3668                         /*
3669                          * enable transmits in the hardware, need to do this
3670                          * after setting TARC(0)
3671                          */
3672                         tctl = er32(TCTL);
3673                         tctl |= E1000_TCTL_EN;
3674                         ew32(TCTL, tctl);
3675
3676                         /*
3677                          * Perform any post-link-up configuration before
3678                          * reporting link up.
3679                          */
3680                         if (phy->ops.cfg_on_link_up)
3681                                 phy->ops.cfg_on_link_up(hw);
3682
3683                         netif_carrier_on(netdev);
3684
3685                         if (!test_bit(__E1000_DOWN, &adapter->state))
3686                                 mod_timer(&adapter->phy_info_timer,
3687                                           round_jiffies(jiffies + 2 * HZ));
3688                 }
3689         } else {
3690                 if (netif_carrier_ok(netdev)) {
3691                         adapter->link_speed = 0;
3692                         adapter->link_duplex = 0;
3693                         /* Link status message must follow this format */
3694                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
3695                                adapter->netdev->name);
3696                         netif_carrier_off(netdev);
3697                         if (!test_bit(__E1000_DOWN, &adapter->state))
3698                                 mod_timer(&adapter->phy_info_timer,
3699                                           round_jiffies(jiffies + 2 * HZ));
3700
3701                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3702                                 schedule_work(&adapter->reset_task);
3703                         else
3704                                 pm_schedule_suspend(netdev->dev.parent,
3705                                                         LINK_TIMEOUT);
3706                 }
3707         }
3708
3709 link_up:
3710         e1000e_update_stats(adapter);
3711
3712         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3713         adapter->tpt_old = adapter->stats.tpt;
3714         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3715         adapter->colc_old = adapter->stats.colc;
3716
3717         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3718         adapter->gorc_old = adapter->stats.gorc;
3719         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3720         adapter->gotc_old = adapter->stats.gotc;
3721
3722         e1000e_update_adaptive(&adapter->hw);
3723
3724         if (!netif_carrier_ok(netdev)) {
3725                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3726                                tx_ring->count);
3727                 if (tx_pending) {
3728                         /*
3729                          * We've lost link, so the controller stops DMA,
3730                          * but we've got queued Tx work that's never going
3731                          * to get done, so reset controller to flush Tx.
3732                          * (Do the reset outside of interrupt context).
3733                          */
3734                         adapter->tx_timeout_count++;
3735                         schedule_work(&adapter->reset_task);
3736                         /* return immediately since reset is imminent */
3737                         return;
3738                 }
3739         }
3740
3741         /* Cause software interrupt to ensure Rx ring is cleaned */
3742         if (adapter->msix_entries)
3743                 ew32(ICS, adapter->rx_ring->ims_val);
3744         else
3745                 ew32(ICS, E1000_ICS_RXDMT0);
3746
3747         /* Force detection of hung controller every watchdog period */
3748         adapter->detect_tx_hung = 1;
3749
3750         /*
3751          * With 82571 controllers, LAA may be overwritten due to controller
3752          * reset from the other port. Set the appropriate LAA in RAR[0]
3753          */
3754         if (e1000e_get_laa_state_82571(hw))
3755                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3756
3757         /* Reset the timer */
3758         if (!test_bit(__E1000_DOWN, &adapter->state))
3759                 mod_timer(&adapter->watchdog_timer,
3760                           round_jiffies(jiffies + 2 * HZ));
3761 }
3762
3763 #define E1000_TX_FLAGS_CSUM             0x00000001
3764 #define E1000_TX_FLAGS_VLAN             0x00000002
3765 #define E1000_TX_FLAGS_TSO              0x00000004
3766 #define E1000_TX_FLAGS_IPV4             0x00000008
3767 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3768 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3769
3770 static int e1000_tso(struct e1000_adapter *adapter,
3771                      struct sk_buff *skb)
3772 {
3773         struct e1000_ring *tx_ring = adapter->tx_ring;
3774         struct e1000_context_desc *context_desc;
3775         struct e1000_buffer *buffer_info;
3776         unsigned int i;
3777         u32 cmd_length = 0;
3778         u16 ipcse = 0, tucse, mss;
3779         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3780         int err;
3781
3782         if (!skb_is_gso(skb))
3783                 return 0;
3784
3785         if (skb_header_cloned(skb)) {
3786                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3787                 if (err)
3788                         return err;
3789         }
3790
3791         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3792         mss = skb_shinfo(skb)->gso_size;
3793         if (skb->protocol == htons(ETH_P_IP)) {
3794                 struct iphdr *iph = ip_hdr(skb);
3795                 iph->tot_len = 0;
3796                 iph->check = 0;
3797                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
3798                                                          0, IPPROTO_TCP, 0);
3799                 cmd_length = E1000_TXD_CMD_IP;
3800                 ipcse = skb_transport_offset(skb) - 1;
3801         } else if (skb_is_gso_v6(skb)) {
3802                 ipv6_hdr(skb)->payload_len = 0;
3803                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3804                                                        &ipv6_hdr(skb)->daddr,
3805                                                        0, IPPROTO_TCP, 0);
3806                 ipcse = 0;
3807         }
3808         ipcss = skb_network_offset(skb);
3809         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3810         tucss = skb_transport_offset(skb);
3811         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3812         tucse = 0;
3813
3814         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3815                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3816
3817         i = tx_ring->next_to_use;
3818         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3819         buffer_info = &tx_ring->buffer_info[i];
3820
3821         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3822         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3823         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3824         context_desc->upper_setup.tcp_fields.tucss = tucss;
3825         context_desc->upper_setup.tcp_fields.tucso = tucso;
3826         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3827         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3828         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3829         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3830
3831         buffer_info->time_stamp = jiffies;
3832         buffer_info->next_to_watch = i;
3833
3834         i++;
3835         if (i == tx_ring->count)
3836                 i = 0;
3837         tx_ring->next_to_use = i;
3838
3839         return 1;
3840 }
3841
3842 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3843 {
3844         struct e1000_ring *tx_ring = adapter->tx_ring;
3845         struct e1000_context_desc *context_desc;
3846         struct e1000_buffer *buffer_info;
3847         unsigned int i;
3848         u8 css;
3849         u32 cmd_len = E1000_TXD_CMD_DEXT;
3850         __be16 protocol;
3851
3852         if (skb->ip_summed != CHECKSUM_PARTIAL)
3853                 return 0;
3854
3855         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
3856                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
3857         else
3858                 protocol = skb->protocol;
3859
3860         switch (protocol) {
3861         case cpu_to_be16(ETH_P_IP):
3862                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3863                         cmd_len |= E1000_TXD_CMD_TCP;
3864                 break;
3865         case cpu_to_be16(ETH_P_IPV6):
3866                 /* XXX not handling all IPV6 headers */
3867                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3868                         cmd_len |= E1000_TXD_CMD_TCP;
3869                 break;
3870         default:
3871                 if (unlikely(net_ratelimit()))
3872                         e_warn("checksum_partial proto=%x!\n",
3873                                be16_to_cpu(protocol));
3874                 break;
3875         }
3876
3877         css = skb_transport_offset(skb);
3878
3879         i = tx_ring->next_to_use;
3880         buffer_info = &tx_ring->buffer_info[i];
3881         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3882
3883         context_desc->lower_setup.ip_config = 0;
3884         context_desc->upper_setup.tcp_fields.tucss = css;
3885         context_desc->upper_setup.tcp_fields.tucso =
3886                                 css + skb->csum_offset;
3887         context_desc->upper_setup.tcp_fields.tucse = 0;
3888         context_desc->tcp_seg_setup.data = 0;
3889         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
3890
3891         buffer_info->time_stamp = jiffies;
3892         buffer_info->next_to_watch = i;
3893
3894         i++;
3895         if (i == tx_ring->count)
3896                 i = 0;
3897         tx_ring->next_to_use = i;
3898
3899         return 1;
3900 }
3901
3902 #define E1000_MAX_PER_TXD       8192
3903 #define E1000_MAX_TXD_PWR       12
3904
3905 static int e1000_tx_map(struct e1000_adapter *adapter,
3906                         struct sk_buff *skb, unsigned int first,
3907                         unsigned int max_per_txd, unsigned int nr_frags,
3908                         unsigned int mss)
3909 {
3910         struct e1000_ring *tx_ring = adapter->tx_ring;
3911         struct pci_dev *pdev = adapter->pdev;
3912         struct e1000_buffer *buffer_info;
3913         unsigned int len = skb_headlen(skb);
3914         unsigned int offset = 0, size, count = 0, i;
3915         unsigned int f;
3916
3917         i = tx_ring->next_to_use;
3918
3919         while (len) {
3920                 buffer_info = &tx_ring->buffer_info[i];
3921                 size = min(len, max_per_txd);
3922
3923                 buffer_info->length = size;
3924                 buffer_info->time_stamp = jiffies;
3925                 buffer_info->next_to_watch = i;
3926                 buffer_info->dma = pci_map_single(pdev, skb->data + offset,
3927                                                   size, PCI_DMA_TODEVICE);
3928                 buffer_info->mapped_as_page = false;
3929                 if (pci_dma_mapping_error(pdev, buffer_info->dma))
3930                         goto dma_error;
3931
3932                 len -= size;
3933                 offset += size;
3934                 count++;
3935
3936                 if (len) {
3937                         i++;
3938                         if (i == tx_ring->count)
3939                                 i = 0;
3940                 }
3941         }
3942
3943         for (f = 0; f < nr_frags; f++) {
3944                 struct skb_frag_struct *frag;
3945
3946                 frag = &skb_shinfo(skb)->frags[f];
3947                 len = frag->size;
3948                 offset = frag->page_offset;
3949
3950                 while (len) {
3951                         i++;
3952                         if (i == tx_ring->count)
3953                                 i = 0;
3954
3955                         buffer_info = &tx_ring->buffer_info[i];
3956                         size = min(len, max_per_txd);
3957
3958                         buffer_info->length = size;
3959                         buffer_info->time_stamp = jiffies;
3960                         buffer_info->next_to_watch = i;
3961                         buffer_info->dma = pci_map_page(pdev, frag->page,
3962                                                         offset, size,
3963                                                         PCI_DMA_TODEVICE);
3964                         buffer_info->mapped_as_page = true;
3965                         if (pci_dma_mapping_error(pdev, buffer_info->dma))
3966                                 goto dma_error;
3967
3968                         len -= size;
3969                         offset += size;
3970                         count++;
3971                 }
3972         }
3973
3974         tx_ring->buffer_info[i].skb = skb;
3975         tx_ring->buffer_info[first].next_to_watch = i;
3976
3977         return count;
3978
3979 dma_error:
3980         dev_err(&pdev->dev, "TX DMA map failed\n");
3981         buffer_info->dma = 0;
3982         if (count)
3983                 count--;
3984
3985         while (count--) {
3986                 if (i==0)
3987                         i += tx_ring->count;
3988                 i--;
3989                 buffer_info = &tx_ring->buffer_info[i];
3990                 e1000_put_txbuf(adapter, buffer_info);;
3991         }
3992
3993         return 0;
3994 }
3995
3996 static void e1000_tx_queue(struct e1000_adapter *adapter,
3997                            int tx_flags, int count)
3998 {
3999         struct e1000_ring *tx_ring = adapter->tx_ring;
4000         struct e1000_tx_desc *tx_desc = NULL;
4001         struct e1000_buffer *buffer_info;
4002         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4003         unsigned int i;
4004
4005         if (tx_flags & E1000_TX_FLAGS_TSO) {
4006                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4007                              E1000_TXD_CMD_TSE;
4008                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4009
4010                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4011                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4012         }
4013
4014         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4015                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4016                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4017         }
4018
4019         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4020                 txd_lower |= E1000_TXD_CMD_VLE;
4021                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4022         }
4023
4024         i = tx_ring->next_to_use;
4025
4026         while (count--) {
4027                 buffer_info = &tx_ring->buffer_info[i];
4028                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4029                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4030                 tx_desc->lower.data =
4031                         cpu_to_le32(txd_lower | buffer_info->length);
4032                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4033
4034                 i++;
4035                 if (i == tx_ring->count)
4036                         i = 0;
4037         }
4038
4039         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4040
4041         /*
4042          * Force memory writes to complete before letting h/w
4043          * know there are new descriptors to fetch.  (Only
4044          * applicable for weak-ordered memory model archs,
4045          * such as IA-64).
4046          */
4047         wmb();
4048
4049         tx_ring->next_to_use = i;
4050         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4051         /*
4052          * we need this if more than one processor can write to our tail
4053          * at a time, it synchronizes IO on IA64/Altix systems
4054          */
4055         mmiowb();
4056 }
4057
4058 #define MINIMUM_DHCP_PACKET_SIZE 282
4059 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4060                                     struct sk_buff *skb)
4061 {
4062         struct e1000_hw *hw =  &adapter->hw;
4063         u16 length, offset;
4064
4065         if (vlan_tx_tag_present(skb)) {
4066                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4067                     (adapter->hw.mng_cookie.status &
4068                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4069                         return 0;
4070         }
4071
4072         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4073                 return 0;
4074
4075         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4076                 return 0;
4077
4078         {
4079                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4080                 struct udphdr *udp;
4081
4082                 if (ip->protocol != IPPROTO_UDP)
4083                         return 0;
4084
4085                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4086                 if (ntohs(udp->dest) != 67)
4087                         return 0;
4088
4089                 offset = (u8 *)udp + 8 - skb->data;
4090                 length = skb->len - offset;
4091                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4092         }
4093
4094         return 0;
4095 }
4096
4097 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4098 {
4099         struct e1000_adapter *adapter = netdev_priv(netdev);
4100
4101         netif_stop_queue(netdev);
4102         /*
4103          * Herbert's original patch had:
4104          *  smp_mb__after_netif_stop_queue();
4105          * but since that doesn't exist yet, just open code it.
4106          */
4107         smp_mb();
4108
4109         /*
4110          * We need to check again in a case another CPU has just
4111          * made room available.
4112          */
4113         if (e1000_desc_unused(adapter->tx_ring) < size)
4114                 return -EBUSY;
4115
4116         /* A reprieve! */
4117         netif_start_queue(netdev);
4118         ++adapter->restart_queue;
4119         return 0;
4120 }
4121
4122 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4123 {
4124         struct e1000_adapter *adapter = netdev_priv(netdev);
4125
4126         if (e1000_desc_unused(adapter->tx_ring) >= size)
4127                 return 0;
4128         return __e1000_maybe_stop_tx(netdev, size);
4129 }
4130
4131 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4132 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4133                                     struct net_device *netdev)
4134 {
4135         struct e1000_adapter *adapter = netdev_priv(netdev);
4136         struct e1000_ring *tx_ring = adapter->tx_ring;
4137         unsigned int first;
4138         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4139         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4140         unsigned int tx_flags = 0;
4141         unsigned int len = skb->len - skb->data_len;
4142         unsigned int nr_frags;
4143         unsigned int mss;
4144         int count = 0;
4145         int tso;
4146         unsigned int f;
4147
4148         if (test_bit(__E1000_DOWN, &adapter->state)) {
4149                 dev_kfree_skb_any(skb);
4150                 return NETDEV_TX_OK;
4151         }
4152
4153         if (skb->len <= 0) {
4154                 dev_kfree_skb_any(skb);
4155                 return NETDEV_TX_OK;
4156         }
4157
4158         mss = skb_shinfo(skb)->gso_size;
4159         /*
4160          * The controller does a simple calculation to
4161          * make sure there is enough room in the FIFO before
4162          * initiating the DMA for each buffer.  The calc is:
4163          * 4 = ceil(buffer len/mss).  To make sure we don't
4164          * overrun the FIFO, adjust the max buffer len if mss
4165          * drops.
4166          */
4167         if (mss) {
4168                 u8 hdr_len;
4169                 max_per_txd = min(mss << 2, max_per_txd);
4170                 max_txd_pwr = fls(max_per_txd) - 1;
4171
4172                 /*
4173                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4174                  * points to just header, pull a few bytes of payload from
4175                  * frags into skb->data
4176                  */
4177                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4178                 /*
4179                  * we do this workaround for ES2LAN, but it is un-necessary,
4180                  * avoiding it could save a lot of cycles
4181                  */
4182                 if (skb->data_len && (hdr_len == len)) {
4183                         unsigned int pull_size;
4184
4185                         pull_size = min((unsigned int)4, skb->data_len);
4186                         if (!__pskb_pull_tail(skb, pull_size)) {
4187                                 e_err("__pskb_pull_tail failed.\n");
4188                                 dev_kfree_skb_any(skb);
4189                                 return NETDEV_TX_OK;
4190                         }
4191                         len = skb->len - skb->data_len;
4192                 }
4193         }
4194
4195         /* reserve a descriptor for the offload context */
4196         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4197                 count++;
4198         count++;
4199
4200         count += TXD_USE_COUNT(len, max_txd_pwr);
4201
4202         nr_frags = skb_shinfo(skb)->nr_frags;
4203         for (f = 0; f < nr_frags; f++)
4204                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4205                                        max_txd_pwr);
4206
4207         if (adapter->hw.mac.tx_pkt_filtering)
4208                 e1000_transfer_dhcp_info(adapter, skb);
4209
4210         /*
4211          * need: count + 2 desc gap to keep tail from touching
4212          * head, otherwise try next time
4213          */
4214         if (e1000_maybe_stop_tx(netdev, count + 2))
4215                 return NETDEV_TX_BUSY;
4216
4217         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4218                 tx_flags |= E1000_TX_FLAGS_VLAN;
4219                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4220         }
4221
4222         first = tx_ring->next_to_use;
4223
4224         tso = e1000_tso(adapter, skb);
4225         if (tso < 0) {
4226                 dev_kfree_skb_any(skb);
4227                 return NETDEV_TX_OK;
4228         }
4229
4230         if (tso)
4231                 tx_flags |= E1000_TX_FLAGS_TSO;
4232         else if (e1000_tx_csum(adapter, skb))
4233                 tx_flags |= E1000_TX_FLAGS_CSUM;
4234
4235         /*
4236          * Old method was to assume IPv4 packet by default if TSO was enabled.
4237          * 82571 hardware supports TSO capabilities for IPv6 as well...
4238          * no longer assume, we must.
4239          */
4240         if (skb->protocol == htons(ETH_P_IP))
4241                 tx_flags |= E1000_TX_FLAGS_IPV4;
4242
4243         /* if count is 0 then mapping error has occured */
4244         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4245         if (count) {
4246                 e1000_tx_queue(adapter, tx_flags, count);
4247                 /* Make sure there is space in the ring for the next send. */
4248                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4249
4250         } else {
4251                 dev_kfree_skb_any(skb);
4252                 tx_ring->buffer_info[first].time_stamp = 0;
4253                 tx_ring->next_to_use = first;
4254         }
4255
4256         return NETDEV_TX_OK;
4257 }
4258
4259 /**
4260  * e1000_tx_timeout - Respond to a Tx Hang
4261  * @netdev: network interface device structure
4262  **/
4263 static void e1000_tx_timeout(struct net_device *netdev)
4264 {
4265         struct e1000_adapter *adapter = netdev_priv(netdev);
4266
4267         /* Do the reset outside of interrupt context */
4268         adapter->tx_timeout_count++;
4269         schedule_work(&adapter->reset_task);
4270 }
4271
4272 static void e1000_reset_task(struct work_struct *work)
4273 {
4274         struct e1000_adapter *adapter;
4275         adapter = container_of(work, struct e1000_adapter, reset_task);
4276
4277         e1000e_reinit_locked(adapter);
4278 }
4279
4280 /**
4281  * e1000_get_stats - Get System Network Statistics
4282  * @netdev: network interface device structure
4283  *
4284  * Returns the address of the device statistics structure.
4285  * The statistics are actually updated from the timer callback.
4286  **/
4287 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4288 {
4289         /* only return the current stats */
4290         return &netdev->stats;
4291 }
4292
4293 /**
4294  * e1000_change_mtu - Change the Maximum Transfer Unit
4295  * @netdev: network interface device structure
4296  * @new_mtu: new value for maximum frame size
4297  *
4298  * Returns 0 on success, negative on failure
4299  **/
4300 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4301 {
4302         struct e1000_adapter *adapter = netdev_priv(netdev);
4303         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4304
4305         /* Jumbo frame support */
4306         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4307             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4308                 e_err("Jumbo Frames not supported.\n");
4309                 return -EINVAL;
4310         }
4311
4312         /* Supported frame sizes */
4313         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4314             (max_frame > adapter->max_hw_frame_size)) {
4315                 e_err("Unsupported MTU setting\n");
4316                 return -EINVAL;
4317         }
4318
4319         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4320                 msleep(1);
4321         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
4322         adapter->max_frame_size = max_frame;
4323         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4324         netdev->mtu = new_mtu;
4325         if (netif_running(netdev))
4326                 e1000e_down(adapter);
4327
4328         /*
4329          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4330          * means we reserve 2 more, this pushes us to allocate from the next
4331          * larger slab size.
4332          * i.e. RXBUFFER_2048 --> size-4096 slab
4333          * However with the new *_jumbo_rx* routines, jumbo receives will use
4334          * fragmented skbs
4335          */
4336
4337         if (max_frame <= 2048)
4338                 adapter->rx_buffer_len = 2048;
4339         else
4340                 adapter->rx_buffer_len = 4096;
4341
4342         /* adjust allocation if LPE protects us, and we aren't using SBP */
4343         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4344              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4345                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4346                                          + ETH_FCS_LEN;
4347
4348         if (netif_running(netdev))
4349                 e1000e_up(adapter);
4350         else
4351                 e1000e_reset(adapter);
4352
4353         clear_bit(__E1000_RESETTING, &adapter->state);
4354
4355         return 0;
4356 }
4357
4358 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4359                            int cmd)
4360 {
4361         struct e1000_adapter *adapter = netdev_priv(netdev);
4362         struct mii_ioctl_data *data = if_mii(ifr);
4363
4364         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4365                 return -EOPNOTSUPP;
4366
4367         switch (cmd) {
4368         case SIOCGMIIPHY:
4369                 data->phy_id = adapter->hw.phy.addr;
4370                 break;
4371         case SIOCGMIIREG:
4372                 e1000_phy_read_status(adapter);
4373
4374                 switch (data->reg_num & 0x1F) {
4375                 case MII_BMCR:
4376                         data->val_out = adapter->phy_regs.bmcr;
4377                         break;
4378                 case MII_BMSR:
4379                         data->val_out = adapter->phy_regs.bmsr;
4380                         break;
4381                 case MII_PHYSID1:
4382                         data->val_out = (adapter->hw.phy.id >> 16);
4383                         break;
4384                 case MII_PHYSID2:
4385                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4386                         break;
4387                 case MII_ADVERTISE:
4388                         data->val_out = adapter->phy_regs.advertise;
4389                         break;
4390                 case MII_LPA:
4391                         data->val_out = adapter->phy_regs.lpa;
4392                         break;
4393                 case MII_EXPANSION:
4394                         data->val_out = adapter->phy_regs.expansion;
4395                         break;
4396                 case MII_CTRL1000:
4397                         data->val_out = adapter->phy_regs.ctrl1000;
4398                         break;
4399                 case MII_STAT1000:
4400                         data->val_out = adapter->phy_regs.stat1000;
4401                         break;
4402                 case MII_ESTATUS:
4403                         data->val_out = adapter->phy_regs.estatus;
4404                         break;
4405                 default:
4406                         return -EIO;
4407                 }
4408                 break;
4409         case SIOCSMIIREG:
4410         default:
4411                 return -EOPNOTSUPP;
4412         }
4413         return 0;
4414 }
4415
4416 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4417 {
4418         switch (cmd) {
4419         case SIOCGMIIPHY:
4420         case SIOCGMIIREG:
4421         case SIOCSMIIREG:
4422                 return e1000_mii_ioctl(netdev, ifr, cmd);
4423         default:
4424                 return -EOPNOTSUPP;
4425         }
4426 }
4427
4428 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
4429 {
4430         struct e1000_hw *hw = &adapter->hw;
4431         u32 i, mac_reg;
4432         u16 phy_reg;
4433         int retval = 0;
4434
4435         /* copy MAC RARs to PHY RARs */
4436         for (i = 0; i < adapter->hw.mac.rar_entry_count; i++) {
4437                 mac_reg = er32(RAL(i));
4438                 e1e_wphy(hw, BM_RAR_L(i), (u16)(mac_reg & 0xFFFF));
4439                 e1e_wphy(hw, BM_RAR_M(i), (u16)((mac_reg >> 16) & 0xFFFF));
4440                 mac_reg = er32(RAH(i));
4441                 e1e_wphy(hw, BM_RAR_H(i), (u16)(mac_reg & 0xFFFF));
4442                 e1e_wphy(hw, BM_RAR_CTRL(i), (u16)((mac_reg >> 16) & 0xFFFF));
4443         }
4444
4445         /* copy MAC MTA to PHY MTA */
4446         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
4447                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
4448                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
4449                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
4450         }
4451
4452         /* configure PHY Rx Control register */
4453         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
4454         mac_reg = er32(RCTL);
4455         if (mac_reg & E1000_RCTL_UPE)
4456                 phy_reg |= BM_RCTL_UPE;
4457         if (mac_reg & E1000_RCTL_MPE)
4458                 phy_reg |= BM_RCTL_MPE;
4459         phy_reg &= ~(BM_RCTL_MO_MASK);
4460         if (mac_reg & E1000_RCTL_MO_3)
4461                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
4462                                 << BM_RCTL_MO_SHIFT);
4463         if (mac_reg & E1000_RCTL_BAM)
4464                 phy_reg |= BM_RCTL_BAM;
4465         if (mac_reg & E1000_RCTL_PMCF)
4466                 phy_reg |= BM_RCTL_PMCF;
4467         mac_reg = er32(CTRL);
4468         if (mac_reg & E1000_CTRL_RFCE)
4469                 phy_reg |= BM_RCTL_RFCE;
4470         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
4471
4472         /* enable PHY wakeup in MAC register */
4473         ew32(WUFC, wufc);
4474         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
4475
4476         /* configure and enable PHY wakeup in PHY registers */
4477         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
4478         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
4479
4480         /* activate PHY wakeup */
4481         retval = hw->phy.ops.acquire(hw);
4482         if (retval) {
4483                 e_err("Could not acquire PHY\n");
4484                 return retval;
4485         }
4486         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4487                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
4488         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
4489         if (retval) {
4490                 e_err("Could not read PHY page 769\n");
4491                 goto out;
4492         }
4493         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
4494         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
4495         if (retval)
4496                 e_err("Could not set PHY Host Wakeup bit\n");
4497 out:
4498         hw->phy.ops.release(hw);
4499
4500         return retval;
4501 }
4502
4503 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
4504                             bool runtime)
4505 {
4506         struct net_device *netdev = pci_get_drvdata(pdev);
4507         struct e1000_adapter *adapter = netdev_priv(netdev);
4508         struct e1000_hw *hw = &adapter->hw;
4509         u32 ctrl, ctrl_ext, rctl, status;
4510         /* Runtime suspend should only enable wakeup for link changes */
4511         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
4512         int retval = 0;
4513
4514         netif_device_detach(netdev);
4515
4516         if (netif_running(netdev)) {
4517                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4518                 e1000e_down(adapter);
4519                 e1000_free_irq(adapter);
4520         }
4521         e1000e_reset_interrupt_capability(adapter);
4522
4523         retval = pci_save_state(pdev);
4524         if (retval)
4525                 return retval;
4526
4527         status = er32(STATUS);
4528         if (status & E1000_STATUS_LU)
4529                 wufc &= ~E1000_WUFC_LNKC;
4530
4531         if (wufc) {
4532                 e1000_setup_rctl(adapter);
4533                 e1000_set_multi(netdev);
4534
4535                 /* turn on all-multi mode if wake on multicast is enabled */
4536                 if (wufc & E1000_WUFC_MC) {
4537                         rctl = er32(RCTL);
4538                         rctl |= E1000_RCTL_MPE;
4539                         ew32(RCTL, rctl);
4540                 }
4541
4542                 ctrl = er32(CTRL);
4543                 /* advertise wake from D3Cold */
4544                 #define E1000_CTRL_ADVD3WUC 0x00100000
4545                 /* phy power management enable */
4546                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4547                 ctrl |= E1000_CTRL_ADVD3WUC;
4548                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
4549                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
4550                 ew32(CTRL, ctrl);
4551
4552                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4553                     adapter->hw.phy.media_type ==
4554                     e1000_media_type_internal_serdes) {
4555                         /* keep the laser running in D3 */
4556                         ctrl_ext = er32(CTRL_EXT);
4557                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
4558                         ew32(CTRL_EXT, ctrl_ext);
4559                 }
4560
4561                 if (adapter->flags & FLAG_IS_ICH)
4562                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4563
4564                 /* Allow time for pending master requests to run */
4565                 e1000e_disable_pcie_master(&adapter->hw);
4566
4567                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4568                         /* enable wakeup by the PHY */
4569                         retval = e1000_init_phy_wakeup(adapter, wufc);
4570                         if (retval)
4571                                 return retval;
4572                 } else {
4573                         /* enable wakeup by the MAC */
4574                         ew32(WUFC, wufc);
4575                         ew32(WUC, E1000_WUC_PME_EN);
4576                 }
4577         } else {
4578                 ew32(WUC, 0);
4579                 ew32(WUFC, 0);
4580         }
4581
4582         *enable_wake = !!wufc;
4583
4584         /* make sure adapter isn't asleep if manageability is enabled */
4585         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
4586             (hw->mac.ops.check_mng_mode(hw)))
4587                 *enable_wake = true;
4588
4589         if (adapter->hw.phy.type == e1000_phy_igp_3)
4590                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4591
4592         /*
4593          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4594          * would have already happened in close and is redundant.
4595          */
4596         e1000_release_hw_control(adapter);
4597
4598         pci_disable_device(pdev);
4599
4600         return 0;
4601 }
4602
4603 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
4604 {
4605         if (sleep && wake) {
4606                 pci_prepare_to_sleep(pdev);
4607                 return;
4608         }
4609
4610         pci_wake_from_d3(pdev, wake);
4611         pci_set_power_state(pdev, PCI_D3hot);
4612 }
4613
4614 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
4615                                     bool wake)
4616 {
4617         struct net_device *netdev = pci_get_drvdata(pdev);
4618         struct e1000_adapter *adapter = netdev_priv(netdev);
4619
4620         /*
4621          * The pci-e switch on some quad port adapters will report a
4622          * correctable error when the MAC transitions from D0 to D3.  To
4623          * prevent this we need to mask off the correctable errors on the
4624          * downstream port of the pci-e switch.
4625          */
4626         if (adapter->flags & FLAG_IS_QUAD_PORT) {
4627                 struct pci_dev *us_dev = pdev->bus->self;
4628                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
4629                 u16 devctl;
4630
4631                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
4632                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
4633                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
4634
4635                 e1000_power_off(pdev, sleep, wake);
4636
4637                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
4638         } else {
4639                 e1000_power_off(pdev, sleep, wake);
4640         }
4641 }
4642
4643 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4644 {
4645         int pos;
4646         u16 val;
4647
4648         /*
4649          * 82573 workaround - disable L1 ASPM on mobile chipsets
4650          *
4651          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4652          * resulting in lost data or garbage information on the pci-e link
4653          * level. This could result in (false) bad EEPROM checksum errors,
4654          * long ping times (up to 2s) or even a system freeze/hang.
4655          *
4656          * Unfortunately this feature saves about 1W power consumption when
4657          * active.
4658          */
4659         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4660         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4661         if (val & 0x2) {
4662                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4663                 val &= ~0x2;
4664                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4665         }
4666 }
4667
4668 #ifdef CONFIG_PM_OPS
4669 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
4670 {
4671         return !!adapter->tx_ring->buffer_info;
4672 }
4673
4674 static int __e1000_resume(struct pci_dev *pdev)
4675 {
4676         struct net_device *netdev = pci_get_drvdata(pdev);
4677         struct e1000_adapter *adapter = netdev_priv(netdev);
4678         struct e1000_hw *hw = &adapter->hw;
4679         u32 err;
4680
4681         e1000e_disable_l1aspm(pdev);
4682
4683         e1000e_set_interrupt_capability(adapter);
4684         if (netif_running(netdev)) {
4685                 err = e1000_request_irq(adapter);
4686                 if (err)
4687                         return err;
4688         }
4689
4690         e1000e_power_up_phy(adapter);
4691
4692         /* report the system wakeup cause from S3/S4 */
4693         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4694                 u16 phy_data;
4695
4696                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
4697                 if (phy_data) {
4698                         e_info("PHY Wakeup cause - %s\n",
4699                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
4700                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
4701                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
4702                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
4703                                 phy_data & E1000_WUS_LNKC ? "Link Status "
4704                                 " Change" : "other");
4705                 }
4706                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
4707         } else {
4708                 u32 wus = er32(WUS);
4709                 if (wus) {
4710                         e_info("MAC Wakeup cause - %s\n",
4711                                 wus & E1000_WUS_EX ? "Unicast Packet" :
4712                                 wus & E1000_WUS_MC ? "Multicast Packet" :
4713                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
4714                                 wus & E1000_WUS_MAG ? "Magic Packet" :
4715                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
4716                                 "other");
4717                 }
4718                 ew32(WUS, ~0);
4719         }
4720
4721         e1000e_reset(adapter);
4722
4723         e1000_init_manageability(adapter);
4724
4725         if (netif_running(netdev))
4726                 e1000e_up(adapter);
4727
4728         netif_device_attach(netdev);
4729
4730         /*
4731          * If the controller has AMT, do not set DRV_LOAD until the interface
4732          * is up.  For all other cases, let the f/w know that the h/w is now
4733          * under the control of the driver.
4734          */
4735         if (!(adapter->flags & FLAG_HAS_AMT))
4736                 e1000_get_hw_control(adapter);
4737
4738         return 0;
4739 }
4740
4741 #ifdef CONFIG_PM_SLEEP
4742 static int e1000_suspend(struct device *dev)
4743 {
4744         struct pci_dev *pdev = to_pci_dev(dev);
4745         int retval;
4746         bool wake;
4747
4748         retval = __e1000_shutdown(pdev, &wake, false);
4749         if (!retval)
4750                 e1000_complete_shutdown(pdev, true, wake);
4751
4752         return retval;
4753 }
4754
4755 static int e1000_resume(struct device *dev)
4756 {
4757         struct pci_dev *pdev = to_pci_dev(dev);
4758         struct net_device *netdev = pci_get_drvdata(pdev);
4759         struct e1000_adapter *adapter = netdev_priv(netdev);
4760
4761         if (e1000e_pm_ready(adapter))
4762                 adapter->idle_check = true;
4763
4764         return __e1000_resume(pdev);
4765 }
4766 #endif /* CONFIG_PM_SLEEP */
4767
4768 #ifdef CONFIG_PM_RUNTIME
4769 static int e1000_runtime_suspend(struct device *dev)
4770 {
4771         struct pci_dev *pdev = to_pci_dev(dev);
4772         struct net_device *netdev = pci_get_drvdata(pdev);
4773         struct e1000_adapter *adapter = netdev_priv(netdev);
4774
4775         if (e1000e_pm_ready(adapter)) {
4776                 bool wake;
4777
4778                 __e1000_shutdown(pdev, &wake, true);
4779         }
4780
4781         return 0;
4782 }
4783
4784 static int e1000_idle(struct device *dev)
4785 {
4786         struct pci_dev *pdev = to_pci_dev(dev);
4787         struct net_device *netdev = pci_get_drvdata(pdev);
4788         struct e1000_adapter *adapter = netdev_priv(netdev);
4789
4790         if (!e1000e_pm_ready(adapter))
4791                 return 0;
4792
4793         if (adapter->idle_check) {
4794                 adapter->idle_check = false;
4795                 if (!e1000e_has_link(adapter))
4796                         pm_schedule_suspend(dev, MSEC_PER_SEC);
4797         }
4798
4799         return -EBUSY;
4800 }
4801
4802 static int e1000_runtime_resume(struct device *dev)
4803 {
4804         struct pci_dev *pdev = to_pci_dev(dev);
4805         struct net_device *netdev = pci_get_drvdata(pdev);
4806         struct e1000_adapter *adapter = netdev_priv(netdev);
4807
4808         if (!e1000e_pm_ready(adapter))
4809                 return 0;
4810
4811         adapter->idle_check = !dev->power.runtime_auto;
4812         return __e1000_resume(pdev);
4813 }
4814 #endif /* CONFIG_PM_RUNTIME */
4815 #endif /* CONFIG_PM_OPS */
4816
4817 static void e1000_shutdown(struct pci_dev *pdev)
4818 {
4819         bool wake = false;
4820
4821         __e1000_shutdown(pdev, &wake, false);
4822
4823         if (system_state == SYSTEM_POWER_OFF)
4824                 e1000_complete_shutdown(pdev, false, wake);
4825 }
4826
4827 #ifdef CONFIG_NET_POLL_CONTROLLER
4828 /*
4829  * Polling 'interrupt' - used by things like netconsole to send skbs
4830  * without having to re-enable interrupts. It's not called while
4831  * the interrupt routine is executing.
4832  */
4833 static void e1000_netpoll(struct net_device *netdev)
4834 {
4835         struct e1000_adapter *adapter = netdev_priv(netdev);
4836
4837         disable_irq(adapter->pdev->irq);
4838         e1000_intr(adapter->pdev->irq, netdev);
4839
4840         enable_irq(adapter->pdev->irq);
4841 }
4842 #endif
4843
4844 /**
4845  * e1000_io_error_detected - called when PCI error is detected
4846  * @pdev: Pointer to PCI device
4847  * @state: The current pci connection state
4848  *
4849  * This function is called after a PCI bus error affecting
4850  * this device has been detected.
4851  */
4852 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4853                                                 pci_channel_state_t state)
4854 {
4855         struct net_device *netdev = pci_get_drvdata(pdev);
4856         struct e1000_adapter *adapter = netdev_priv(netdev);
4857
4858         netif_device_detach(netdev);
4859
4860         if (state == pci_channel_io_perm_failure)
4861                 return PCI_ERS_RESULT_DISCONNECT;
4862
4863         if (netif_running(netdev))
4864                 e1000e_down(adapter);
4865         pci_disable_device(pdev);
4866
4867         /* Request a slot slot reset. */
4868         return PCI_ERS_RESULT_NEED_RESET;
4869 }
4870
4871 /**
4872  * e1000_io_slot_reset - called after the pci bus has been reset.
4873  * @pdev: Pointer to PCI device
4874  *
4875  * Restart the card from scratch, as if from a cold-boot. Implementation
4876  * resembles the first-half of the e1000_resume routine.
4877  */
4878 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4879 {
4880         struct net_device *netdev = pci_get_drvdata(pdev);
4881         struct e1000_adapter *adapter = netdev_priv(netdev);
4882         struct e1000_hw *hw = &adapter->hw;
4883         int err;
4884         pci_ers_result_t result;
4885
4886         e1000e_disable_l1aspm(pdev);
4887         err = pci_enable_device_mem(pdev);
4888         if (err) {
4889                 dev_err(&pdev->dev,
4890                         "Cannot re-enable PCI device after reset.\n");
4891                 result = PCI_ERS_RESULT_DISCONNECT;
4892         } else {
4893                 pci_set_master(pdev);
4894                 pdev->state_saved = true;
4895                 pci_restore_state(pdev);
4896
4897                 pci_enable_wake(pdev, PCI_D3hot, 0);
4898                 pci_enable_wake(pdev, PCI_D3cold, 0);
4899
4900                 e1000e_reset(adapter);
4901                 ew32(WUS, ~0);
4902                 result = PCI_ERS_RESULT_RECOVERED;
4903         }
4904
4905         pci_cleanup_aer_uncorrect_error_status(pdev);
4906
4907         return result;
4908 }
4909
4910 /**
4911  * e1000_io_resume - called when traffic can start flowing again.
4912  * @pdev: Pointer to PCI device
4913  *
4914  * This callback is called when the error recovery driver tells us that
4915  * its OK to resume normal operation. Implementation resembles the
4916  * second-half of the e1000_resume routine.
4917  */
4918 static void e1000_io_resume(struct pci_dev *pdev)
4919 {
4920         struct net_device *netdev = pci_get_drvdata(pdev);
4921         struct e1000_adapter *adapter = netdev_priv(netdev);
4922
4923         e1000_init_manageability(adapter);
4924
4925         if (netif_running(netdev)) {
4926                 if (e1000e_up(adapter)) {
4927                         dev_err(&pdev->dev,
4928                                 "can't bring device back up after reset\n");
4929                         return;
4930                 }
4931         }
4932
4933         netif_device_attach(netdev);
4934
4935         /*
4936          * If the controller has AMT, do not set DRV_LOAD until the interface
4937          * is up.  For all other cases, let the f/w know that the h/w is now
4938          * under the control of the driver.
4939          */
4940         if (!(adapter->flags & FLAG_HAS_AMT))
4941                 e1000_get_hw_control(adapter);
4942
4943 }
4944
4945 static void e1000_print_device_info(struct e1000_adapter *adapter)
4946 {
4947         struct e1000_hw *hw = &adapter->hw;
4948         struct net_device *netdev = adapter->netdev;
4949         u32 pba_num;
4950
4951         /* print bus type/speed/width info */
4952         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
4953                /* bus width */
4954                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4955                 "Width x1"),
4956                /* MAC address */
4957                netdev->dev_addr);
4958         e_info("Intel(R) PRO/%s Network Connection\n",
4959                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4960         e1000e_read_pba_num(hw, &pba_num);
4961         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4962                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4963 }
4964
4965 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4966 {
4967         struct e1000_hw *hw = &adapter->hw;
4968         int ret_val;
4969         u16 buf = 0;
4970
4971         if (hw->mac.type != e1000_82573)
4972                 return;
4973
4974         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4975         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
4976                 /* Deep Smart Power Down (DSPD) */
4977                 dev_warn(&adapter->pdev->dev,
4978                          "Warning: detected DSPD enabled in EEPROM\n");
4979         }
4980
4981         ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4982         if (!ret_val && (le16_to_cpu(buf) & (3 << 2))) {
4983                 /* ASPM enable */
4984                 dev_warn(&adapter->pdev->dev,
4985                          "Warning: detected ASPM enabled in EEPROM\n");
4986         }
4987 }
4988
4989 static const struct net_device_ops e1000e_netdev_ops = {
4990         .ndo_open               = e1000_open,
4991         .ndo_stop               = e1000_close,
4992         .ndo_start_xmit         = e1000_xmit_frame,
4993         .ndo_get_stats          = e1000_get_stats,
4994         .ndo_set_multicast_list = e1000_set_multi,
4995         .ndo_set_mac_address    = e1000_set_mac,
4996         .ndo_change_mtu         = e1000_change_mtu,
4997         .ndo_do_ioctl           = e1000_ioctl,
4998         .ndo_tx_timeout         = e1000_tx_timeout,
4999         .ndo_validate_addr      = eth_validate_addr,
5000
5001         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
5002         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
5003         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
5004 #ifdef CONFIG_NET_POLL_CONTROLLER
5005         .ndo_poll_controller    = e1000_netpoll,
5006 #endif
5007 };
5008
5009 /**
5010  * e1000_probe - Device Initialization Routine
5011  * @pdev: PCI device information struct
5012  * @ent: entry in e1000_pci_tbl
5013  *
5014  * Returns 0 on success, negative on failure
5015  *
5016  * e1000_probe initializes an adapter identified by a pci_dev structure.
5017  * The OS initialization, configuring of the adapter private structure,
5018  * and a hardware reset occur.
5019  **/
5020 static int __devinit e1000_probe(struct pci_dev *pdev,
5021                                  const struct pci_device_id *ent)
5022 {
5023         struct net_device *netdev;
5024         struct e1000_adapter *adapter;
5025         struct e1000_hw *hw;
5026         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5027         resource_size_t mmio_start, mmio_len;
5028         resource_size_t flash_start, flash_len;
5029
5030         static int cards_found;
5031         int i, err, pci_using_dac;
5032         u16 eeprom_data = 0;
5033         u16 eeprom_apme_mask = E1000_EEPROM_APME;
5034
5035         e1000e_disable_l1aspm(pdev);
5036
5037         err = pci_enable_device_mem(pdev);
5038         if (err)
5039                 return err;
5040
5041         pci_using_dac = 0;
5042         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
5043         if (!err) {
5044                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
5045                 if (!err)
5046                         pci_using_dac = 1;
5047         } else {
5048                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
5049                 if (err) {
5050                         err = pci_set_consistent_dma_mask(pdev,
5051                                                           DMA_BIT_MASK(32));
5052                         if (err) {
5053                                 dev_err(&pdev->dev, "No usable DMA "
5054                                         "configuration, aborting\n");
5055                                 goto err_dma;
5056                         }
5057                 }
5058         }
5059
5060         err = pci_request_selected_regions_exclusive(pdev,
5061                                           pci_select_bars(pdev, IORESOURCE_MEM),
5062                                           e1000e_driver_name);
5063         if (err)
5064                 goto err_pci_reg;
5065
5066         /* AER (Advanced Error Reporting) hooks */
5067         pci_enable_pcie_error_reporting(pdev);
5068
5069         pci_set_master(pdev);
5070         /* PCI config space info */
5071         err = pci_save_state(pdev);
5072         if (err)
5073                 goto err_alloc_etherdev;
5074
5075         err = -ENOMEM;
5076         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5077         if (!netdev)
5078                 goto err_alloc_etherdev;
5079
5080         SET_NETDEV_DEV(netdev, &pdev->dev);
5081
5082         pci_set_drvdata(pdev, netdev);
5083         adapter = netdev_priv(netdev);
5084         hw = &adapter->hw;
5085         adapter->netdev = netdev;
5086         adapter->pdev = pdev;
5087         adapter->ei = ei;
5088         adapter->pba = ei->pba;
5089         adapter->flags = ei->flags;
5090         adapter->flags2 = ei->flags2;
5091         adapter->hw.adapter = adapter;
5092         adapter->hw.mac.type = ei->mac;
5093         adapter->max_hw_frame_size = ei->max_hw_frame_size;
5094         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5095
5096         mmio_start = pci_resource_start(pdev, 0);
5097         mmio_len = pci_resource_len(pdev, 0);
5098
5099         err = -EIO;
5100         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5101         if (!adapter->hw.hw_addr)
5102                 goto err_ioremap;
5103
5104         if ((adapter->flags & FLAG_HAS_FLASH) &&
5105             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5106                 flash_start = pci_resource_start(pdev, 1);
5107                 flash_len = pci_resource_len(pdev, 1);
5108                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5109                 if (!adapter->hw.flash_address)
5110                         goto err_flashmap;
5111         }
5112
5113         /* construct the net_device struct */
5114         netdev->netdev_ops              = &e1000e_netdev_ops;
5115         e1000e_set_ethtool_ops(netdev);
5116         netdev->watchdog_timeo          = 5 * HZ;
5117         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5118         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5119
5120         netdev->mem_start = mmio_start;
5121         netdev->mem_end = mmio_start + mmio_len;
5122
5123         adapter->bd_number = cards_found++;
5124
5125         e1000e_check_options(adapter);
5126
5127         /* setup adapter struct */
5128         err = e1000_sw_init(adapter);
5129         if (err)
5130                 goto err_sw_init;
5131
5132         err = -EIO;
5133
5134         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5135         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5136         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5137
5138         err = ei->get_variants(adapter);
5139         if (err)
5140                 goto err_hw_init;
5141
5142         if ((adapter->flags & FLAG_IS_ICH) &&
5143             (adapter->flags & FLAG_READ_ONLY_NVM))
5144                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5145
5146         hw->mac.ops.get_bus_info(&adapter->hw);
5147
5148         adapter->hw.phy.autoneg_wait_to_complete = 0;
5149
5150         /* Copper options */
5151         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5152                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5153                 adapter->hw.phy.disable_polarity_correction = 0;
5154                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5155         }
5156
5157         if (e1000_check_reset_block(&adapter->hw))
5158                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5159
5160         netdev->features = NETIF_F_SG |
5161                            NETIF_F_HW_CSUM |
5162                            NETIF_F_HW_VLAN_TX |
5163                            NETIF_F_HW_VLAN_RX;
5164
5165         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5166                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5167
5168         netdev->features |= NETIF_F_TSO;
5169         netdev->features |= NETIF_F_TSO6;
5170
5171         netdev->vlan_features |= NETIF_F_TSO;
5172         netdev->vlan_features |= NETIF_F_TSO6;
5173         netdev->vlan_features |= NETIF_F_HW_CSUM;
5174         netdev->vlan_features |= NETIF_F_SG;
5175
5176         if (pci_using_dac)
5177                 netdev->features |= NETIF_F_HIGHDMA;
5178
5179         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5180                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5181
5182         /*
5183          * before reading the NVM, reset the controller to
5184          * put the device in a known good starting state
5185          */
5186         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5187
5188         /*
5189          * systems with ASPM and others may see the checksum fail on the first
5190          * attempt. Let's give it a few tries
5191          */
5192         for (i = 0;; i++) {
5193                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5194                         break;
5195                 if (i == 2) {
5196                         e_err("The NVM Checksum Is Not Valid\n");
5197                         err = -EIO;
5198                         goto err_eeprom;
5199                 }
5200         }
5201
5202         e1000_eeprom_checks(adapter);
5203
5204         /* copy the MAC address */
5205         if (e1000e_read_mac_addr(&adapter->hw))
5206                 e_err("NVM Read Error while reading MAC address\n");
5207
5208         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5209         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5210
5211         if (!is_valid_ether_addr(netdev->perm_addr)) {
5212                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5213                 err = -EIO;
5214                 goto err_eeprom;
5215         }
5216
5217         init_timer(&adapter->watchdog_timer);
5218         adapter->watchdog_timer.function = &e1000_watchdog;
5219         adapter->watchdog_timer.data = (unsigned long) adapter;
5220
5221         init_timer(&adapter->phy_info_timer);
5222         adapter->phy_info_timer.function = &e1000_update_phy_info;
5223         adapter->phy_info_timer.data = (unsigned long) adapter;
5224
5225         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5226         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5227         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5228         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5229         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5230
5231         /* Initialize link parameters. User can change them with ethtool */
5232         adapter->hw.mac.autoneg = 1;
5233         adapter->fc_autoneg = 1;
5234         adapter->hw.fc.requested_mode = e1000_fc_default;
5235         adapter->hw.fc.current_mode = e1000_fc_default;
5236         adapter->hw.phy.autoneg_advertised = 0x2f;
5237
5238         /* ring size defaults */
5239         adapter->rx_ring->count = 256;
5240         adapter->tx_ring->count = 256;
5241
5242         /*
5243          * Initial Wake on LAN setting - If APM wake is enabled in
5244          * the EEPROM, enable the ACPI Magic Packet filter
5245          */
5246         if (adapter->flags & FLAG_APME_IN_WUC) {
5247                 /* APME bit in EEPROM is mapped to WUC.APME */
5248                 eeprom_data = er32(WUC);
5249                 eeprom_apme_mask = E1000_WUC_APME;
5250                 if (eeprom_data & E1000_WUC_PHY_WAKE)
5251                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5252         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5253                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5254                     (adapter->hw.bus.func == 1))
5255                         e1000_read_nvm(&adapter->hw,
5256                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5257                 else
5258                         e1000_read_nvm(&adapter->hw,
5259                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5260         }
5261
5262         /* fetch WoL from EEPROM */
5263         if (eeprom_data & eeprom_apme_mask)
5264                 adapter->eeprom_wol |= E1000_WUFC_MAG;
5265
5266         /*
5267          * now that we have the eeprom settings, apply the special cases
5268          * where the eeprom may be wrong or the board simply won't support
5269          * wake on lan on a particular port
5270          */
5271         if (!(adapter->flags & FLAG_HAS_WOL))
5272                 adapter->eeprom_wol = 0;
5273
5274         /* initialize the wol settings based on the eeprom settings */
5275         adapter->wol = adapter->eeprom_wol;
5276         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5277
5278         /* save off EEPROM version number */
5279         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5280
5281         /* reset the hardware with the new settings */
5282         e1000e_reset(adapter);
5283
5284         /*
5285          * If the controller has AMT, do not set DRV_LOAD until the interface
5286          * is up.  For all other cases, let the f/w know that the h/w is now
5287          * under the control of the driver.
5288          */
5289         if (!(adapter->flags & FLAG_HAS_AMT))
5290                 e1000_get_hw_control(adapter);
5291
5292         strcpy(netdev->name, "eth%d");
5293         err = register_netdev(netdev);
5294         if (err)
5295                 goto err_register;
5296
5297         /* carrier off reporting is important to ethtool even BEFORE open */
5298         netif_carrier_off(netdev);
5299
5300         e1000_print_device_info(adapter);
5301
5302         if (pci_dev_run_wake(pdev)) {
5303                 pm_runtime_set_active(&pdev->dev);
5304                 pm_runtime_enable(&pdev->dev);
5305         }
5306         pm_schedule_suspend(&pdev->dev, MSEC_PER_SEC);
5307
5308         return 0;
5309
5310 err_register:
5311         if (!(adapter->flags & FLAG_HAS_AMT))
5312                 e1000_release_hw_control(adapter);
5313 err_eeprom:
5314         if (!e1000_check_reset_block(&adapter->hw))
5315                 e1000_phy_hw_reset(&adapter->hw);
5316 err_hw_init:
5317
5318         kfree(adapter->tx_ring);
5319         kfree(adapter->rx_ring);
5320 err_sw_init:
5321         if (adapter->hw.flash_address)
5322                 iounmap(adapter->hw.flash_address);
5323         e1000e_reset_interrupt_capability(adapter);
5324 err_flashmap:
5325         iounmap(adapter->hw.hw_addr);
5326 err_ioremap:
5327         free_netdev(netdev);
5328 err_alloc_etherdev:
5329         pci_release_selected_regions(pdev,
5330                                      pci_select_bars(pdev, IORESOURCE_MEM));
5331 err_pci_reg:
5332 err_dma:
5333         pci_disable_device(pdev);
5334         return err;
5335 }
5336
5337 /**
5338  * e1000_remove - Device Removal Routine
5339  * @pdev: PCI device information struct
5340  *
5341  * e1000_remove is called by the PCI subsystem to alert the driver
5342  * that it should release a PCI device.  The could be caused by a
5343  * Hot-Plug event, or because the driver is going to be removed from
5344  * memory.
5345  **/
5346 static void __devexit e1000_remove(struct pci_dev *pdev)
5347 {
5348         struct net_device *netdev = pci_get_drvdata(pdev);
5349         struct e1000_adapter *adapter = netdev_priv(netdev);
5350         bool down = test_bit(__E1000_DOWN, &adapter->state);
5351
5352         pm_runtime_get_sync(&pdev->dev);
5353
5354         /*
5355          * flush_scheduled work may reschedule our watchdog task, so
5356          * explicitly disable watchdog tasks from being rescheduled
5357          */
5358         if (!down)
5359                 set_bit(__E1000_DOWN, &adapter->state);
5360         del_timer_sync(&adapter->watchdog_timer);
5361         del_timer_sync(&adapter->phy_info_timer);
5362
5363         cancel_work_sync(&adapter->reset_task);
5364         cancel_work_sync(&adapter->watchdog_task);
5365         cancel_work_sync(&adapter->downshift_task);
5366         cancel_work_sync(&adapter->update_phy_task);
5367         cancel_work_sync(&adapter->print_hang_task);
5368         flush_scheduled_work();
5369
5370         if (!(netdev->flags & IFF_UP))
5371                 e1000_power_down_phy(adapter);
5372
5373         /* Don't lie to e1000_close() down the road. */
5374         if (!down)
5375                 clear_bit(__E1000_DOWN, &adapter->state);
5376         unregister_netdev(netdev);
5377
5378         if (pci_dev_run_wake(pdev)) {
5379                 pm_runtime_disable(&pdev->dev);
5380                 pm_runtime_set_suspended(&pdev->dev);
5381         }
5382         pm_runtime_put_noidle(&pdev->dev);
5383
5384         /*
5385          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5386          * would have already happened in close and is redundant.
5387          */
5388         e1000_release_hw_control(adapter);
5389
5390         e1000e_reset_interrupt_capability(adapter);
5391         kfree(adapter->tx_ring);
5392         kfree(adapter->rx_ring);
5393
5394         iounmap(adapter->hw.hw_addr);
5395         if (adapter->hw.flash_address)
5396                 iounmap(adapter->hw.flash_address);
5397         pci_release_selected_regions(pdev,
5398                                      pci_select_bars(pdev, IORESOURCE_MEM));
5399
5400         free_netdev(netdev);
5401
5402         /* AER disable */
5403         pci_disable_pcie_error_reporting(pdev);
5404
5405         pci_disable_device(pdev);
5406 }
5407
5408 /* PCI Error Recovery (ERS) */
5409 static struct pci_error_handlers e1000_err_handler = {
5410         .error_detected = e1000_io_error_detected,
5411         .slot_reset = e1000_io_slot_reset,
5412         .resume = e1000_io_resume,
5413 };
5414
5415 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
5416         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5417         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5418         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5419         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5420         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5421         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5422         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5423         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5424         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5425
5426         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5427         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5428         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5429         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5430
5431         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5432         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5433         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5434
5435         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5436         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
5437         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
5438
5439         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5440           board_80003es2lan },
5441         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5442           board_80003es2lan },
5443         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5444           board_80003es2lan },
5445         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5446           board_80003es2lan },
5447
5448         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5449         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5450         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5451         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5452         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5453         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5454         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5455         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
5456
5457         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5458         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5459         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5460         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5461         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5462         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5463         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5464         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5465         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5466
5467         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5468         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5469         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5470
5471         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5472         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5473
5474         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
5475         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
5476         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
5477         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
5478
5479         { }     /* terminate list */
5480 };
5481 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5482
5483 #ifdef CONFIG_PM_OPS
5484 static const struct dev_pm_ops e1000_pm_ops = {
5485         SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
5486         SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
5487                                 e1000_runtime_resume, e1000_idle)
5488 };
5489 #endif
5490
5491 /* PCI Device API Driver */
5492 static struct pci_driver e1000_driver = {
5493         .name     = e1000e_driver_name,
5494         .id_table = e1000_pci_tbl,
5495         .probe    = e1000_probe,
5496         .remove   = __devexit_p(e1000_remove),
5497 #ifdef CONFIG_PM_OPS
5498         .driver.pm = &e1000_pm_ops,
5499 #endif
5500         .shutdown = e1000_shutdown,
5501         .err_handler = &e1000_err_handler
5502 };
5503
5504 /**
5505  * e1000_init_module - Driver Registration Routine
5506  *
5507  * e1000_init_module is the first routine called when the driver is
5508  * loaded. All it does is register with the PCI subsystem.
5509  **/
5510 static int __init e1000_init_module(void)
5511 {
5512         int ret;
5513         pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
5514                 e1000e_driver_version);
5515         pr_info("Copyright (c) 1999 - 2009 Intel Corporation.\n");
5516         ret = pci_register_driver(&e1000_driver);
5517
5518         return ret;
5519 }
5520 module_init(e1000_init_module);
5521
5522 /**
5523  * e1000_exit_module - Driver Exit Cleanup Routine
5524  *
5525  * e1000_exit_module is called just before the driver is removed
5526  * from memory.
5527  **/
5528 static void __exit e1000_exit_module(void)
5529 {
5530         pci_unregister_driver(&e1000_driver);
5531 }
5532 module_exit(e1000_exit_module);
5533
5534
5535 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5536 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5537 MODULE_LICENSE("GPL");
5538 MODULE_VERSION(DRV_VERSION);
5539
5540 /* e1000_main.c */