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