88d54d3efcef2ff78bd66bd2cd358c8809c22820
[linux-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 #include <linux/module.h>
30 #include <linux/types.h>
31 #include <linux/init.h>
32 #include <linux/pci.h>
33 #include <linux/vmalloc.h>
34 #include <linux/pagemap.h>
35 #include <linux/delay.h>
36 #include <linux/netdevice.h>
37 #include <linux/tcp.h>
38 #include <linux/ipv6.h>
39 #include <net/checksum.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/mii.h>
42 #include <linux/ethtool.h>
43 #include <linux/if_vlan.h>
44 #include <linux/cpu.h>
45 #include <linux/smp.h>
46 #include <linux/pm_qos_params.h>
47 #include <linux/aer.h>
48
49 #include "e1000.h"
50
51 #define DRV_VERSION "1.0.2-k2"
52 char e1000e_driver_name[] = "e1000e";
53 const char e1000e_driver_version[] = DRV_VERSION;
54
55 static const struct e1000_info *e1000_info_tbl[] = {
56         [board_82571]           = &e1000_82571_info,
57         [board_82572]           = &e1000_82572_info,
58         [board_82573]           = &e1000_82573_info,
59         [board_82574]           = &e1000_82574_info,
60         [board_82583]           = &e1000_82583_info,
61         [board_80003es2lan]     = &e1000_es2_info,
62         [board_ich8lan]         = &e1000_ich8_info,
63         [board_ich9lan]         = &e1000_ich9_info,
64         [board_ich10lan]        = &e1000_ich10_info,
65         [board_pchlan]          = &e1000_pch_info,
66 };
67
68 /**
69  * e1000_desc_unused - calculate if we have unused descriptors
70  **/
71 static int e1000_desc_unused(struct e1000_ring *ring)
72 {
73         if (ring->next_to_clean > ring->next_to_use)
74                 return ring->next_to_clean - ring->next_to_use - 1;
75
76         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
77 }
78
79 /**
80  * e1000_receive_skb - helper function to handle Rx indications
81  * @adapter: board private structure
82  * @status: descriptor status field as written by hardware
83  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
84  * @skb: pointer to sk_buff to be indicated to stack
85  **/
86 static void e1000_receive_skb(struct e1000_adapter *adapter,
87                               struct net_device *netdev,
88                               struct sk_buff *skb,
89                               u8 status, __le16 vlan)
90 {
91         skb->protocol = eth_type_trans(skb, netdev);
92
93         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
94                 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
95                                  le16_to_cpu(vlan), skb);
96         else
97                 napi_gro_receive(&adapter->napi, skb);
98 }
99
100 /**
101  * e1000_rx_checksum - Receive Checksum Offload for 82543
102  * @adapter:     board private structure
103  * @status_err:  receive descriptor status and error fields
104  * @csum:       receive descriptor csum field
105  * @sk_buff:     socket buffer with received data
106  **/
107 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
108                               u32 csum, struct sk_buff *skb)
109 {
110         u16 status = (u16)status_err;
111         u8 errors = (u8)(status_err >> 24);
112         skb->ip_summed = CHECKSUM_NONE;
113
114         /* Ignore Checksum bit is set */
115         if (status & E1000_RXD_STAT_IXSM)
116                 return;
117         /* TCP/UDP checksum error bit is set */
118         if (errors & E1000_RXD_ERR_TCPE) {
119                 /* let the stack verify checksum errors */
120                 adapter->hw_csum_err++;
121                 return;
122         }
123
124         /* TCP/UDP Checksum has not been calculated */
125         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
126                 return;
127
128         /* It must be a TCP or UDP packet with a valid checksum */
129         if (status & E1000_RXD_STAT_TCPCS) {
130                 /* TCP checksum is good */
131                 skb->ip_summed = CHECKSUM_UNNECESSARY;
132         } else {
133                 /*
134                  * IP fragment with UDP payload
135                  * Hardware complements the payload checksum, so we undo it
136                  * and then put the value in host order for further stack use.
137                  */
138                 __sum16 sum = (__force __sum16)htons(csum);
139                 skb->csum = csum_unfold(~sum);
140                 skb->ip_summed = CHECKSUM_COMPLETE;
141         }
142         adapter->hw_csum_good++;
143 }
144
145 /**
146  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
147  * @adapter: address of board private structure
148  **/
149 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
150                                    int cleaned_count)
151 {
152         struct net_device *netdev = adapter->netdev;
153         struct pci_dev *pdev = adapter->pdev;
154         struct e1000_ring *rx_ring = adapter->rx_ring;
155         struct e1000_rx_desc *rx_desc;
156         struct e1000_buffer *buffer_info;
157         struct sk_buff *skb;
158         unsigned int i;
159         unsigned int bufsz = adapter->rx_buffer_len;
160
161         i = rx_ring->next_to_use;
162         buffer_info = &rx_ring->buffer_info[i];
163
164         while (cleaned_count--) {
165                 skb = buffer_info->skb;
166                 if (skb) {
167                         skb_trim(skb, 0);
168                         goto map_skb;
169                 }
170
171                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
172                 if (!skb) {
173                         /* Better luck next round */
174                         adapter->alloc_rx_buff_failed++;
175                         break;
176                 }
177
178                 buffer_info->skb = skb;
179 map_skb:
180                 buffer_info->dma = pci_map_single(pdev, skb->data,
181                                                   adapter->rx_buffer_len,
182                                                   PCI_DMA_FROMDEVICE);
183                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
184                         dev_err(&pdev->dev, "RX DMA map failed\n");
185                         adapter->rx_dma_failed++;
186                         break;
187                 }
188
189                 rx_desc = E1000_RX_DESC(*rx_ring, i);
190                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
191
192                 i++;
193                 if (i == rx_ring->count)
194                         i = 0;
195                 buffer_info = &rx_ring->buffer_info[i];
196         }
197
198         if (rx_ring->next_to_use != i) {
199                 rx_ring->next_to_use = i;
200                 if (i-- == 0)
201                         i = (rx_ring->count - 1);
202
203                 /*
204                  * Force memory writes to complete before letting h/w
205                  * know there are new descriptors to fetch.  (Only
206                  * applicable for weak-ordered memory model archs,
207                  * such as IA-64).
208                  */
209                 wmb();
210                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
211         }
212 }
213
214 /**
215  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
216  * @adapter: address of board private structure
217  **/
218 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
219                                       int cleaned_count)
220 {
221         struct net_device *netdev = adapter->netdev;
222         struct pci_dev *pdev = adapter->pdev;
223         union e1000_rx_desc_packet_split *rx_desc;
224         struct e1000_ring *rx_ring = adapter->rx_ring;
225         struct e1000_buffer *buffer_info;
226         struct e1000_ps_page *ps_page;
227         struct sk_buff *skb;
228         unsigned int i, j;
229
230         i = rx_ring->next_to_use;
231         buffer_info = &rx_ring->buffer_info[i];
232
233         while (cleaned_count--) {
234                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
235
236                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
237                         ps_page = &buffer_info->ps_pages[j];
238                         if (j >= adapter->rx_ps_pages) {
239                                 /* all unused desc entries get hw null ptr */
240                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
241                                 continue;
242                         }
243                         if (!ps_page->page) {
244                                 ps_page->page = alloc_page(GFP_ATOMIC);
245                                 if (!ps_page->page) {
246                                         adapter->alloc_rx_buff_failed++;
247                                         goto no_buffers;
248                                 }
249                                 ps_page->dma = pci_map_page(pdev,
250                                                    ps_page->page,
251                                                    0, PAGE_SIZE,
252                                                    PCI_DMA_FROMDEVICE);
253                                 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
254                                         dev_err(&adapter->pdev->dev,
255                                           "RX DMA page map failed\n");
256                                         adapter->rx_dma_failed++;
257                                         goto no_buffers;
258                                 }
259                         }
260                         /*
261                          * Refresh the desc even if buffer_addrs
262                          * didn't change because each write-back
263                          * erases this info.
264                          */
265                         rx_desc->read.buffer_addr[j+1] =
266                              cpu_to_le64(ps_page->dma);
267                 }
268
269                 skb = netdev_alloc_skb_ip_align(netdev,
270                                                 adapter->rx_ps_bsize0);
271
272                 if (!skb) {
273                         adapter->alloc_rx_buff_failed++;
274                         break;
275                 }
276
277                 buffer_info->skb = skb;
278                 buffer_info->dma = pci_map_single(pdev, skb->data,
279                                                   adapter->rx_ps_bsize0,
280                                                   PCI_DMA_FROMDEVICE);
281                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
282                         dev_err(&pdev->dev, "RX DMA map failed\n");
283                         adapter->rx_dma_failed++;
284                         /* cleanup skb */
285                         dev_kfree_skb_any(skb);
286                         buffer_info->skb = NULL;
287                         break;
288                 }
289
290                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
291
292                 i++;
293                 if (i == rx_ring->count)
294                         i = 0;
295                 buffer_info = &rx_ring->buffer_info[i];
296         }
297
298 no_buffers:
299         if (rx_ring->next_to_use != i) {
300                 rx_ring->next_to_use = i;
301
302                 if (!(i--))
303                         i = (rx_ring->count - 1);
304
305                 /*
306                  * Force memory writes to complete before letting h/w
307                  * know there are new descriptors to fetch.  (Only
308                  * applicable for weak-ordered memory model archs,
309                  * such as IA-64).
310                  */
311                 wmb();
312                 /*
313                  * Hardware increments by 16 bytes, but packet split
314                  * descriptors are 32 bytes...so we increment tail
315                  * twice as much.
316                  */
317                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
318         }
319 }
320
321 /**
322  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
323  * @adapter: address of board private structure
324  * @cleaned_count: number of buffers to allocate this pass
325  **/
326
327 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
328                                          int cleaned_count)
329 {
330         struct net_device *netdev = adapter->netdev;
331         struct pci_dev *pdev = adapter->pdev;
332         struct e1000_rx_desc *rx_desc;
333         struct e1000_ring *rx_ring = adapter->rx_ring;
334         struct e1000_buffer *buffer_info;
335         struct sk_buff *skb;
336         unsigned int i;
337         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
338
339         i = rx_ring->next_to_use;
340         buffer_info = &rx_ring->buffer_info[i];
341
342         while (cleaned_count--) {
343                 skb = buffer_info->skb;
344                 if (skb) {
345                         skb_trim(skb, 0);
346                         goto check_page;
347                 }
348
349                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
350                 if (unlikely(!skb)) {
351                         /* Better luck next round */
352                         adapter->alloc_rx_buff_failed++;
353                         break;
354                 }
355
356                 buffer_info->skb = skb;
357 check_page:
358                 /* allocate a new page if necessary */
359                 if (!buffer_info->page) {
360                         buffer_info->page = alloc_page(GFP_ATOMIC);
361                         if (unlikely(!buffer_info->page)) {
362                                 adapter->alloc_rx_buff_failed++;
363                                 break;
364                         }
365                 }
366
367                 if (!buffer_info->dma)
368                         buffer_info->dma = pci_map_page(pdev,
369                                                         buffer_info->page, 0,
370                                                         PAGE_SIZE,
371                                                         PCI_DMA_FROMDEVICE);
372
373                 rx_desc = E1000_RX_DESC(*rx_ring, i);
374                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
375
376                 if (unlikely(++i == rx_ring->count))
377                         i = 0;
378                 buffer_info = &rx_ring->buffer_info[i];
379         }
380
381         if (likely(rx_ring->next_to_use != i)) {
382                 rx_ring->next_to_use = i;
383                 if (unlikely(i-- == 0))
384                         i = (rx_ring->count - 1);
385
386                 /* Force memory writes to complete before letting h/w
387                  * know there are new descriptors to fetch.  (Only
388                  * applicable for weak-ordered memory model archs,
389                  * such as IA-64). */
390                 wmb();
391                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
392         }
393 }
394
395 /**
396  * e1000_clean_rx_irq - Send received data up the network stack; legacy
397  * @adapter: board private structure
398  *
399  * the return value indicates whether actual cleaning was done, there
400  * is no guarantee that everything was cleaned
401  **/
402 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
403                                int *work_done, int work_to_do)
404 {
405         struct net_device *netdev = adapter->netdev;
406         struct pci_dev *pdev = adapter->pdev;
407         struct e1000_hw *hw = &adapter->hw;
408         struct e1000_ring *rx_ring = adapter->rx_ring;
409         struct e1000_rx_desc *rx_desc, *next_rxd;
410         struct e1000_buffer *buffer_info, *next_buffer;
411         u32 length;
412         unsigned int i;
413         int cleaned_count = 0;
414         bool cleaned = 0;
415         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
416
417         i = rx_ring->next_to_clean;
418         rx_desc = E1000_RX_DESC(*rx_ring, i);
419         buffer_info = &rx_ring->buffer_info[i];
420
421         while (rx_desc->status & E1000_RXD_STAT_DD) {
422                 struct sk_buff *skb;
423                 u8 status;
424
425                 if (*work_done >= work_to_do)
426                         break;
427                 (*work_done)++;
428
429                 status = rx_desc->status;
430                 skb = buffer_info->skb;
431                 buffer_info->skb = NULL;
432
433                 prefetch(skb->data - NET_IP_ALIGN);
434
435                 i++;
436                 if (i == rx_ring->count)
437                         i = 0;
438                 next_rxd = E1000_RX_DESC(*rx_ring, i);
439                 prefetch(next_rxd);
440
441                 next_buffer = &rx_ring->buffer_info[i];
442
443                 cleaned = 1;
444                 cleaned_count++;
445                 pci_unmap_single(pdev,
446                                  buffer_info->dma,
447                                  adapter->rx_buffer_len,
448                                  PCI_DMA_FROMDEVICE);
449                 buffer_info->dma = 0;
450
451                 length = le16_to_cpu(rx_desc->length);
452
453                 /*
454                  * !EOP means multiple descriptors were used to store a single
455                  * packet, if that's the case we need to toss it.  In fact, we
456                  * need to toss every packet with the EOP bit clear and the
457                  * next frame that _does_ have the EOP bit set, as it is by
458                  * definition only a frame fragment
459                  */
460                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
461                         adapter->flags2 |= FLAG2_IS_DISCARDING;
462
463                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
464                         /* All receives must fit into a single buffer */
465                         e_dbg("Receive packet consumed multiple buffers\n");
466                         /* recycle */
467                         buffer_info->skb = skb;
468                         if (status & E1000_RXD_STAT_EOP)
469                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
470                         goto next_desc;
471                 }
472
473                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
474                         /* recycle */
475                         buffer_info->skb = skb;
476                         goto next_desc;
477                 }
478
479                 /* adjust length to remove Ethernet CRC */
480                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
481                         length -= 4;
482
483                 total_rx_bytes += length;
484                 total_rx_packets++;
485
486                 /*
487                  * code added for copybreak, this should improve
488                  * performance for small packets with large amounts
489                  * of reassembly being done in the stack
490                  */
491                 if (length < copybreak) {
492                         struct sk_buff *new_skb =
493                             netdev_alloc_skb_ip_align(netdev, length);
494                         if (new_skb) {
495                                 skb_copy_to_linear_data_offset(new_skb,
496                                                                -NET_IP_ALIGN,
497                                                                (skb->data -
498                                                                 NET_IP_ALIGN),
499                                                                (length +
500                                                                 NET_IP_ALIGN));
501                                 /* save the skb in buffer_info as good */
502                                 buffer_info->skb = skb;
503                                 skb = new_skb;
504                         }
505                         /* else just continue with the old one */
506                 }
507                 /* end copybreak code */
508                 skb_put(skb, length);
509
510                 /* Receive Checksum Offload */
511                 e1000_rx_checksum(adapter,
512                                   (u32)(status) |
513                                   ((u32)(rx_desc->errors) << 24),
514                                   le16_to_cpu(rx_desc->csum), skb);
515
516                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
517
518 next_desc:
519                 rx_desc->status = 0;
520
521                 /* return some buffers to hardware, one at a time is too slow */
522                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
523                         adapter->alloc_rx_buf(adapter, cleaned_count);
524                         cleaned_count = 0;
525                 }
526
527                 /* use prefetched values */
528                 rx_desc = next_rxd;
529                 buffer_info = next_buffer;
530         }
531         rx_ring->next_to_clean = i;
532
533         cleaned_count = e1000_desc_unused(rx_ring);
534         if (cleaned_count)
535                 adapter->alloc_rx_buf(adapter, cleaned_count);
536
537         adapter->total_rx_bytes += total_rx_bytes;
538         adapter->total_rx_packets += total_rx_packets;
539         netdev->stats.rx_bytes += total_rx_bytes;
540         netdev->stats.rx_packets += total_rx_packets;
541         return cleaned;
542 }
543
544 static void e1000_put_txbuf(struct e1000_adapter *adapter,
545                              struct e1000_buffer *buffer_info)
546 {
547         if (buffer_info->dma) {
548                 if (buffer_info->mapped_as_page)
549                         pci_unmap_page(adapter->pdev, buffer_info->dma,
550                                        buffer_info->length, PCI_DMA_TODEVICE);
551                 else
552                         pci_unmap_single(adapter->pdev, buffer_info->dma,
553                                          buffer_info->length,
554                                          PCI_DMA_TODEVICE);
555                 buffer_info->dma = 0;
556         }
557         if (buffer_info->skb) {
558                 dev_kfree_skb_any(buffer_info->skb);
559                 buffer_info->skb = NULL;
560         }
561         buffer_info->time_stamp = 0;
562 }
563
564 static void e1000_print_hw_hang(struct work_struct *work)
565 {
566         struct e1000_adapter *adapter = container_of(work,
567                                                      struct e1000_adapter,
568                                                      print_hang_task);
569         struct e1000_ring *tx_ring = adapter->tx_ring;
570         unsigned int i = tx_ring->next_to_clean;
571         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
572         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
573         struct e1000_hw *hw = &adapter->hw;
574         u16 phy_status, phy_1000t_status, phy_ext_status;
575         u16 pci_status;
576
577         e1e_rphy(hw, PHY_STATUS, &phy_status);
578         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
579         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
580
581         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
582
583         /* detected Hardware unit hang */
584         e_err("Detected Hardware Unit Hang:\n"
585               "  TDH                  <%x>\n"
586               "  TDT                  <%x>\n"
587               "  next_to_use          <%x>\n"
588               "  next_to_clean        <%x>\n"
589               "buffer_info[next_to_clean]:\n"
590               "  time_stamp           <%lx>\n"
591               "  next_to_watch        <%x>\n"
592               "  jiffies              <%lx>\n"
593               "  next_to_watch.status <%x>\n"
594               "MAC Status             <%x>\n"
595               "PHY Status             <%x>\n"
596               "PHY 1000BASE-T Status  <%x>\n"
597               "PHY Extended Status    <%x>\n"
598               "PCI Status             <%x>\n",
599               readl(adapter->hw.hw_addr + tx_ring->head),
600               readl(adapter->hw.hw_addr + tx_ring->tail),
601               tx_ring->next_to_use,
602               tx_ring->next_to_clean,
603               tx_ring->buffer_info[eop].time_stamp,
604               eop,
605               jiffies,
606               eop_desc->upper.fields.status,
607               er32(STATUS),
608               phy_status,
609               phy_1000t_status,
610               phy_ext_status,
611               pci_status);
612 }
613
614 /**
615  * e1000_clean_tx_irq - Reclaim resources after transmit completes
616  * @adapter: board private structure
617  *
618  * the return value indicates whether actual cleaning was done, there
619  * is no guarantee that everything was cleaned
620  **/
621 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
622 {
623         struct net_device *netdev = adapter->netdev;
624         struct e1000_hw *hw = &adapter->hw;
625         struct e1000_ring *tx_ring = adapter->tx_ring;
626         struct e1000_tx_desc *tx_desc, *eop_desc;
627         struct e1000_buffer *buffer_info;
628         unsigned int i, eop;
629         unsigned int count = 0;
630         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
631
632         i = tx_ring->next_to_clean;
633         eop = tx_ring->buffer_info[i].next_to_watch;
634         eop_desc = E1000_TX_DESC(*tx_ring, eop);
635
636         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
637                (count < tx_ring->count)) {
638                 bool cleaned = false;
639                 for (; !cleaned; count++) {
640                         tx_desc = E1000_TX_DESC(*tx_ring, i);
641                         buffer_info = &tx_ring->buffer_info[i];
642                         cleaned = (i == eop);
643
644                         if (cleaned) {
645                                 struct sk_buff *skb = buffer_info->skb;
646                                 unsigned int segs, bytecount;
647                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
648                                 /* multiply data chunks by size of headers */
649                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
650                                             skb->len;
651                                 total_tx_packets += segs;
652                                 total_tx_bytes += bytecount;
653                         }
654
655                         e1000_put_txbuf(adapter, buffer_info);
656                         tx_desc->upper.data = 0;
657
658                         i++;
659                         if (i == tx_ring->count)
660                                 i = 0;
661                 }
662
663                 eop = tx_ring->buffer_info[i].next_to_watch;
664                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
665         }
666
667         tx_ring->next_to_clean = i;
668
669 #define TX_WAKE_THRESHOLD 32
670         if (count && netif_carrier_ok(netdev) &&
671             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
672                 /* Make sure that anybody stopping the queue after this
673                  * sees the new next_to_clean.
674                  */
675                 smp_mb();
676
677                 if (netif_queue_stopped(netdev) &&
678                     !(test_bit(__E1000_DOWN, &adapter->state))) {
679                         netif_wake_queue(netdev);
680                         ++adapter->restart_queue;
681                 }
682         }
683
684         if (adapter->detect_tx_hung) {
685                 /*
686                  * Detect a transmit hang in hardware, this serializes the
687                  * check with the clearing of time_stamp and movement of i
688                  */
689                 adapter->detect_tx_hung = 0;
690                 if (tx_ring->buffer_info[i].time_stamp &&
691                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
692                                + (adapter->tx_timeout_factor * HZ)) &&
693                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
694                         schedule_work(&adapter->print_hang_task);
695                         netif_stop_queue(netdev);
696                 }
697         }
698         adapter->total_tx_bytes += total_tx_bytes;
699         adapter->total_tx_packets += total_tx_packets;
700         netdev->stats.tx_bytes += total_tx_bytes;
701         netdev->stats.tx_packets += total_tx_packets;
702         return (count < tx_ring->count);
703 }
704
705 /**
706  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
707  * @adapter: board private structure
708  *
709  * the return value indicates whether actual cleaning was done, there
710  * is no guarantee that everything was cleaned
711  **/
712 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
713                                   int *work_done, int work_to_do)
714 {
715         struct e1000_hw *hw = &adapter->hw;
716         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
717         struct net_device *netdev = adapter->netdev;
718         struct pci_dev *pdev = adapter->pdev;
719         struct e1000_ring *rx_ring = adapter->rx_ring;
720         struct e1000_buffer *buffer_info, *next_buffer;
721         struct e1000_ps_page *ps_page;
722         struct sk_buff *skb;
723         unsigned int i, j;
724         u32 length, staterr;
725         int cleaned_count = 0;
726         bool cleaned = 0;
727         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
728
729         i = rx_ring->next_to_clean;
730         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
731         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
732         buffer_info = &rx_ring->buffer_info[i];
733
734         while (staterr & E1000_RXD_STAT_DD) {
735                 if (*work_done >= work_to_do)
736                         break;
737                 (*work_done)++;
738                 skb = buffer_info->skb;
739
740                 /* in the packet split case this is header only */
741                 prefetch(skb->data - NET_IP_ALIGN);
742
743                 i++;
744                 if (i == rx_ring->count)
745                         i = 0;
746                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
747                 prefetch(next_rxd);
748
749                 next_buffer = &rx_ring->buffer_info[i];
750
751                 cleaned = 1;
752                 cleaned_count++;
753                 pci_unmap_single(pdev, buffer_info->dma,
754                                  adapter->rx_ps_bsize0,
755                                  PCI_DMA_FROMDEVICE);
756                 buffer_info->dma = 0;
757
758                 /* see !EOP comment in other rx routine */
759                 if (!(staterr & E1000_RXD_STAT_EOP))
760                         adapter->flags2 |= FLAG2_IS_DISCARDING;
761
762                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
763                         e_dbg("Packet Split buffers didn't pick up the full "
764                               "packet\n");
765                         dev_kfree_skb_irq(skb);
766                         if (staterr & E1000_RXD_STAT_EOP)
767                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
768                         goto next_desc;
769                 }
770
771                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
772                         dev_kfree_skb_irq(skb);
773                         goto next_desc;
774                 }
775
776                 length = le16_to_cpu(rx_desc->wb.middle.length0);
777
778                 if (!length) {
779                         e_dbg("Last part of the packet spanning multiple "
780                               "descriptors\n");
781                         dev_kfree_skb_irq(skb);
782                         goto next_desc;
783                 }
784
785                 /* Good Receive */
786                 skb_put(skb, length);
787
788                 {
789                 /*
790                  * this looks ugly, but it seems compiler issues make it
791                  * more efficient than reusing j
792                  */
793                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
794
795                 /*
796                  * page alloc/put takes too long and effects small packet
797                  * throughput, so unsplit small packets and save the alloc/put
798                  * only valid in softirq (napi) context to call kmap_*
799                  */
800                 if (l1 && (l1 <= copybreak) &&
801                     ((length + l1) <= adapter->rx_ps_bsize0)) {
802                         u8 *vaddr;
803
804                         ps_page = &buffer_info->ps_pages[0];
805
806                         /*
807                          * there is no documentation about how to call
808                          * kmap_atomic, so we can't hold the mapping
809                          * very long
810                          */
811                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
812                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
813                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
814                         memcpy(skb_tail_pointer(skb), vaddr, l1);
815                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
816                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
817                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
818
819                         /* remove the CRC */
820                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
821                                 l1 -= 4;
822
823                         skb_put(skb, l1);
824                         goto copydone;
825                 } /* if */
826                 }
827
828                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
829                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
830                         if (!length)
831                                 break;
832
833                         ps_page = &buffer_info->ps_pages[j];
834                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
835                                        PCI_DMA_FROMDEVICE);
836                         ps_page->dma = 0;
837                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
838                         ps_page->page = NULL;
839                         skb->len += length;
840                         skb->data_len += length;
841                         skb->truesize += length;
842                 }
843
844                 /* strip the ethernet crc, problem is we're using pages now so
845                  * this whole operation can get a little cpu intensive
846                  */
847                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
848                         pskb_trim(skb, skb->len - 4);
849
850 copydone:
851                 total_rx_bytes += skb->len;
852                 total_rx_packets++;
853
854                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
855                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
856
857                 if (rx_desc->wb.upper.header_status &
858                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
859                         adapter->rx_hdr_split++;
860
861                 e1000_receive_skb(adapter, netdev, skb,
862                                   staterr, rx_desc->wb.middle.vlan);
863
864 next_desc:
865                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
866                 buffer_info->skb = NULL;
867
868                 /* return some buffers to hardware, one at a time is too slow */
869                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
870                         adapter->alloc_rx_buf(adapter, cleaned_count);
871                         cleaned_count = 0;
872                 }
873
874                 /* use prefetched values */
875                 rx_desc = next_rxd;
876                 buffer_info = next_buffer;
877
878                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
879         }
880         rx_ring->next_to_clean = i;
881
882         cleaned_count = e1000_desc_unused(rx_ring);
883         if (cleaned_count)
884                 adapter->alloc_rx_buf(adapter, cleaned_count);
885
886         adapter->total_rx_bytes += total_rx_bytes;
887         adapter->total_rx_packets += total_rx_packets;
888         netdev->stats.rx_bytes += total_rx_bytes;
889         netdev->stats.rx_packets += total_rx_packets;
890         return cleaned;
891 }
892
893 /**
894  * e1000_consume_page - helper function
895  **/
896 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
897                                u16 length)
898 {
899         bi->page = NULL;
900         skb->len += length;
901         skb->data_len += length;
902         skb->truesize += length;
903 }
904
905 /**
906  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
907  * @adapter: board private structure
908  *
909  * the return value indicates whether actual cleaning was done, there
910  * is no guarantee that everything was cleaned
911  **/
912
913 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
914                                      int *work_done, int work_to_do)
915 {
916         struct net_device *netdev = adapter->netdev;
917         struct pci_dev *pdev = adapter->pdev;
918         struct e1000_ring *rx_ring = adapter->rx_ring;
919         struct e1000_rx_desc *rx_desc, *next_rxd;
920         struct e1000_buffer *buffer_info, *next_buffer;
921         u32 length;
922         unsigned int i;
923         int cleaned_count = 0;
924         bool cleaned = false;
925         unsigned int total_rx_bytes=0, total_rx_packets=0;
926
927         i = rx_ring->next_to_clean;
928         rx_desc = E1000_RX_DESC(*rx_ring, i);
929         buffer_info = &rx_ring->buffer_info[i];
930
931         while (rx_desc->status & E1000_RXD_STAT_DD) {
932                 struct sk_buff *skb;
933                 u8 status;
934
935                 if (*work_done >= work_to_do)
936                         break;
937                 (*work_done)++;
938
939                 status = rx_desc->status;
940                 skb = buffer_info->skb;
941                 buffer_info->skb = NULL;
942
943                 ++i;
944                 if (i == rx_ring->count)
945                         i = 0;
946                 next_rxd = E1000_RX_DESC(*rx_ring, i);
947                 prefetch(next_rxd);
948
949                 next_buffer = &rx_ring->buffer_info[i];
950
951                 cleaned = true;
952                 cleaned_count++;
953                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
954                                PCI_DMA_FROMDEVICE);
955                 buffer_info->dma = 0;
956
957                 length = le16_to_cpu(rx_desc->length);
958
959                 /* errors is only valid for DD + EOP descriptors */
960                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
961                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
962                                 /* recycle both page and skb */
963                                 buffer_info->skb = skb;
964                                 /* an error means any chain goes out the window
965                                  * too */
966                                 if (rx_ring->rx_skb_top)
967                                         dev_kfree_skb(rx_ring->rx_skb_top);
968                                 rx_ring->rx_skb_top = NULL;
969                                 goto next_desc;
970                 }
971
972 #define rxtop rx_ring->rx_skb_top
973                 if (!(status & E1000_RXD_STAT_EOP)) {
974                         /* this descriptor is only the beginning (or middle) */
975                         if (!rxtop) {
976                                 /* this is the beginning of a chain */
977                                 rxtop = skb;
978                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
979                                                    0, length);
980                         } else {
981                                 /* this is the middle of a chain */
982                                 skb_fill_page_desc(rxtop,
983                                     skb_shinfo(rxtop)->nr_frags,
984                                     buffer_info->page, 0, length);
985                                 /* re-use the skb, only consumed the page */
986                                 buffer_info->skb = skb;
987                         }
988                         e1000_consume_page(buffer_info, rxtop, length);
989                         goto next_desc;
990                 } else {
991                         if (rxtop) {
992                                 /* end of the chain */
993                                 skb_fill_page_desc(rxtop,
994                                     skb_shinfo(rxtop)->nr_frags,
995                                     buffer_info->page, 0, length);
996                                 /* re-use the current skb, we only consumed the
997                                  * page */
998                                 buffer_info->skb = skb;
999                                 skb = rxtop;
1000                                 rxtop = NULL;
1001                                 e1000_consume_page(buffer_info, skb, length);
1002                         } else {
1003                                 /* no chain, got EOP, this buf is the packet
1004                                  * copybreak to save the put_page/alloc_page */
1005                                 if (length <= copybreak &&
1006                                     skb_tailroom(skb) >= length) {
1007                                         u8 *vaddr;
1008                                         vaddr = kmap_atomic(buffer_info->page,
1009                                                            KM_SKB_DATA_SOFTIRQ);
1010                                         memcpy(skb_tail_pointer(skb), vaddr,
1011                                                length);
1012                                         kunmap_atomic(vaddr,
1013                                                       KM_SKB_DATA_SOFTIRQ);
1014                                         /* re-use the page, so don't erase
1015                                          * buffer_info->page */
1016                                         skb_put(skb, length);
1017                                 } else {
1018                                         skb_fill_page_desc(skb, 0,
1019                                                            buffer_info->page, 0,
1020                                                            length);
1021                                         e1000_consume_page(buffer_info, skb,
1022                                                            length);
1023                                 }
1024                         }
1025                 }
1026
1027                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1028                 e1000_rx_checksum(adapter,
1029                                   (u32)(status) |
1030                                   ((u32)(rx_desc->errors) << 24),
1031                                   le16_to_cpu(rx_desc->csum), skb);
1032
1033                 /* probably a little skewed due to removing CRC */
1034                 total_rx_bytes += skb->len;
1035                 total_rx_packets++;
1036
1037                 /* eth type trans needs skb->data to point to something */
1038                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1039                         e_err("pskb_may_pull failed.\n");
1040                         dev_kfree_skb(skb);
1041                         goto next_desc;
1042                 }
1043
1044                 e1000_receive_skb(adapter, netdev, skb, status,
1045                                   rx_desc->special);
1046
1047 next_desc:
1048                 rx_desc->status = 0;
1049
1050                 /* return some buffers to hardware, one at a time is too slow */
1051                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1052                         adapter->alloc_rx_buf(adapter, cleaned_count);
1053                         cleaned_count = 0;
1054                 }
1055
1056                 /* use prefetched values */
1057                 rx_desc = next_rxd;
1058                 buffer_info = next_buffer;
1059         }
1060         rx_ring->next_to_clean = i;
1061
1062         cleaned_count = e1000_desc_unused(rx_ring);
1063         if (cleaned_count)
1064                 adapter->alloc_rx_buf(adapter, cleaned_count);
1065
1066         adapter->total_rx_bytes += total_rx_bytes;
1067         adapter->total_rx_packets += total_rx_packets;
1068         netdev->stats.rx_bytes += total_rx_bytes;
1069         netdev->stats.rx_packets += total_rx_packets;
1070         return cleaned;
1071 }
1072
1073 /**
1074  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1075  * @adapter: board private structure
1076  **/
1077 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1078 {
1079         struct e1000_ring *rx_ring = adapter->rx_ring;
1080         struct e1000_buffer *buffer_info;
1081         struct e1000_ps_page *ps_page;
1082         struct pci_dev *pdev = adapter->pdev;
1083         unsigned int i, j;
1084
1085         /* Free all the Rx ring sk_buffs */
1086         for (i = 0; i < rx_ring->count; i++) {
1087                 buffer_info = &rx_ring->buffer_info[i];
1088                 if (buffer_info->dma) {
1089                         if (adapter->clean_rx == e1000_clean_rx_irq)
1090                                 pci_unmap_single(pdev, buffer_info->dma,
1091                                                  adapter->rx_buffer_len,
1092                                                  PCI_DMA_FROMDEVICE);
1093                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1094                                 pci_unmap_page(pdev, buffer_info->dma,
1095                                                PAGE_SIZE,
1096                                                PCI_DMA_FROMDEVICE);
1097                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1098                                 pci_unmap_single(pdev, buffer_info->dma,
1099                                                  adapter->rx_ps_bsize0,
1100                                                  PCI_DMA_FROMDEVICE);
1101                         buffer_info->dma = 0;
1102                 }
1103
1104                 if (buffer_info->page) {
1105                         put_page(buffer_info->page);
1106                         buffer_info->page = NULL;
1107                 }
1108
1109                 if (buffer_info->skb) {
1110                         dev_kfree_skb(buffer_info->skb);
1111                         buffer_info->skb = NULL;
1112                 }
1113
1114                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1115                         ps_page = &buffer_info->ps_pages[j];
1116                         if (!ps_page->page)
1117                                 break;
1118                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1119                                        PCI_DMA_FROMDEVICE);
1120                         ps_page->dma = 0;
1121                         put_page(ps_page->page);
1122                         ps_page->page = NULL;
1123                 }
1124         }
1125
1126         /* there also may be some cached data from a chained receive */
1127         if (rx_ring->rx_skb_top) {
1128                 dev_kfree_skb(rx_ring->rx_skb_top);
1129                 rx_ring->rx_skb_top = NULL;
1130         }
1131
1132         /* Zero out the descriptor ring */
1133         memset(rx_ring->desc, 0, rx_ring->size);
1134
1135         rx_ring->next_to_clean = 0;
1136         rx_ring->next_to_use = 0;
1137         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1138
1139         writel(0, adapter->hw.hw_addr + rx_ring->head);
1140         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1141 }
1142
1143 static void e1000e_downshift_workaround(struct work_struct *work)
1144 {
1145         struct e1000_adapter *adapter = container_of(work,
1146                                         struct e1000_adapter, downshift_task);
1147
1148         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1149 }
1150
1151 /**
1152  * e1000_intr_msi - Interrupt Handler
1153  * @irq: interrupt number
1154  * @data: pointer to a network interface device structure
1155  **/
1156 static irqreturn_t e1000_intr_msi(int irq, void *data)
1157 {
1158         struct net_device *netdev = data;
1159         struct e1000_adapter *adapter = netdev_priv(netdev);
1160         struct e1000_hw *hw = &adapter->hw;
1161         u32 icr = er32(ICR);
1162
1163         /*
1164          * read ICR disables interrupts using IAM
1165          */
1166
1167         if (icr & E1000_ICR_LSC) {
1168                 hw->mac.get_link_status = 1;
1169                 /*
1170                  * ICH8 workaround-- Call gig speed drop workaround on cable
1171                  * disconnect (LSC) before accessing any PHY registers
1172                  */
1173                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1174                     (!(er32(STATUS) & E1000_STATUS_LU)))
1175                         schedule_work(&adapter->downshift_task);
1176
1177                 /*
1178                  * 80003ES2LAN workaround-- For packet buffer work-around on
1179                  * link down event; disable receives here in the ISR and reset
1180                  * adapter in watchdog
1181                  */
1182                 if (netif_carrier_ok(netdev) &&
1183                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1184                         /* disable receives */
1185                         u32 rctl = er32(RCTL);
1186                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1187                         adapter->flags |= FLAG_RX_RESTART_NOW;
1188                 }
1189                 /* guard against interrupt when we're going down */
1190                 if (!test_bit(__E1000_DOWN, &adapter->state))
1191                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1192         }
1193
1194         if (napi_schedule_prep(&adapter->napi)) {
1195                 adapter->total_tx_bytes = 0;
1196                 adapter->total_tx_packets = 0;
1197                 adapter->total_rx_bytes = 0;
1198                 adapter->total_rx_packets = 0;
1199                 __napi_schedule(&adapter->napi);
1200         }
1201
1202         return IRQ_HANDLED;
1203 }
1204
1205 /**
1206  * e1000_intr - Interrupt Handler
1207  * @irq: interrupt number
1208  * @data: pointer to a network interface device structure
1209  **/
1210 static irqreturn_t e1000_intr(int irq, void *data)
1211 {
1212         struct net_device *netdev = data;
1213         struct e1000_adapter *adapter = netdev_priv(netdev);
1214         struct e1000_hw *hw = &adapter->hw;
1215         u32 rctl, icr = er32(ICR);
1216
1217         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1218                 return IRQ_NONE;  /* Not our interrupt */
1219
1220         /*
1221          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1222          * not set, then the adapter didn't send an interrupt
1223          */
1224         if (!(icr & E1000_ICR_INT_ASSERTED))
1225                 return IRQ_NONE;
1226
1227         /*
1228          * Interrupt Auto-Mask...upon reading ICR,
1229          * interrupts are masked.  No need for the
1230          * IMC write
1231          */
1232
1233         if (icr & E1000_ICR_LSC) {
1234                 hw->mac.get_link_status = 1;
1235                 /*
1236                  * ICH8 workaround-- Call gig speed drop workaround on cable
1237                  * disconnect (LSC) before accessing any PHY registers
1238                  */
1239                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1240                     (!(er32(STATUS) & E1000_STATUS_LU)))
1241                         schedule_work(&adapter->downshift_task);
1242
1243                 /*
1244                  * 80003ES2LAN workaround--
1245                  * For packet buffer work-around on link down event;
1246                  * disable receives here in the ISR and
1247                  * reset adapter in watchdog
1248                  */
1249                 if (netif_carrier_ok(netdev) &&
1250                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1251                         /* disable receives */
1252                         rctl = er32(RCTL);
1253                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1254                         adapter->flags |= FLAG_RX_RESTART_NOW;
1255                 }
1256                 /* guard against interrupt when we're going down */
1257                 if (!test_bit(__E1000_DOWN, &adapter->state))
1258                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1259         }
1260
1261         if (napi_schedule_prep(&adapter->napi)) {
1262                 adapter->total_tx_bytes = 0;
1263                 adapter->total_tx_packets = 0;
1264                 adapter->total_rx_bytes = 0;
1265                 adapter->total_rx_packets = 0;
1266                 __napi_schedule(&adapter->napi);
1267         }
1268
1269         return IRQ_HANDLED;
1270 }
1271
1272 static irqreturn_t e1000_msix_other(int irq, void *data)
1273 {
1274         struct net_device *netdev = data;
1275         struct e1000_adapter *adapter = netdev_priv(netdev);
1276         struct e1000_hw *hw = &adapter->hw;
1277         u32 icr = er32(ICR);
1278
1279         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1280                 if (!test_bit(__E1000_DOWN, &adapter->state))
1281                         ew32(IMS, E1000_IMS_OTHER);
1282                 return IRQ_NONE;
1283         }
1284
1285         if (icr & adapter->eiac_mask)
1286                 ew32(ICS, (icr & adapter->eiac_mask));
1287
1288         if (icr & E1000_ICR_OTHER) {
1289                 if (!(icr & E1000_ICR_LSC))
1290                         goto no_link_interrupt;
1291                 hw->mac.get_link_status = 1;
1292                 /* guard against interrupt when we're going down */
1293                 if (!test_bit(__E1000_DOWN, &adapter->state))
1294                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1295         }
1296
1297 no_link_interrupt:
1298         if (!test_bit(__E1000_DOWN, &adapter->state))
1299                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1300
1301         return IRQ_HANDLED;
1302 }
1303
1304
1305 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1306 {
1307         struct net_device *netdev = data;
1308         struct e1000_adapter *adapter = netdev_priv(netdev);
1309         struct e1000_hw *hw = &adapter->hw;
1310         struct e1000_ring *tx_ring = adapter->tx_ring;
1311
1312
1313         adapter->total_tx_bytes = 0;
1314         adapter->total_tx_packets = 0;
1315
1316         if (!e1000_clean_tx_irq(adapter))
1317                 /* Ring was not completely cleaned, so fire another interrupt */
1318                 ew32(ICS, tx_ring->ims_val);
1319
1320         return IRQ_HANDLED;
1321 }
1322
1323 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1324 {
1325         struct net_device *netdev = data;
1326         struct e1000_adapter *adapter = netdev_priv(netdev);
1327
1328         /* Write the ITR value calculated at the end of the
1329          * previous interrupt.
1330          */
1331         if (adapter->rx_ring->set_itr) {
1332                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1333                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1334                 adapter->rx_ring->set_itr = 0;
1335         }
1336
1337         if (napi_schedule_prep(&adapter->napi)) {
1338                 adapter->total_rx_bytes = 0;
1339                 adapter->total_rx_packets = 0;
1340                 __napi_schedule(&adapter->napi);
1341         }
1342         return IRQ_HANDLED;
1343 }
1344
1345 /**
1346  * e1000_configure_msix - Configure MSI-X hardware
1347  *
1348  * e1000_configure_msix sets up the hardware to properly
1349  * generate MSI-X interrupts.
1350  **/
1351 static void e1000_configure_msix(struct e1000_adapter *adapter)
1352 {
1353         struct e1000_hw *hw = &adapter->hw;
1354         struct e1000_ring *rx_ring = adapter->rx_ring;
1355         struct e1000_ring *tx_ring = adapter->tx_ring;
1356         int vector = 0;
1357         u32 ctrl_ext, ivar = 0;
1358
1359         adapter->eiac_mask = 0;
1360
1361         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1362         if (hw->mac.type == e1000_82574) {
1363                 u32 rfctl = er32(RFCTL);
1364                 rfctl |= E1000_RFCTL_ACK_DIS;
1365                 ew32(RFCTL, rfctl);
1366         }
1367
1368 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1369         /* Configure Rx vector */
1370         rx_ring->ims_val = E1000_IMS_RXQ0;
1371         adapter->eiac_mask |= rx_ring->ims_val;
1372         if (rx_ring->itr_val)
1373                 writel(1000000000 / (rx_ring->itr_val * 256),
1374                        hw->hw_addr + rx_ring->itr_register);
1375         else
1376                 writel(1, hw->hw_addr + rx_ring->itr_register);
1377         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1378
1379         /* Configure Tx vector */
1380         tx_ring->ims_val = E1000_IMS_TXQ0;
1381         vector++;
1382         if (tx_ring->itr_val)
1383                 writel(1000000000 / (tx_ring->itr_val * 256),
1384                        hw->hw_addr + tx_ring->itr_register);
1385         else
1386                 writel(1, hw->hw_addr + tx_ring->itr_register);
1387         adapter->eiac_mask |= tx_ring->ims_val;
1388         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1389
1390         /* set vector for Other Causes, e.g. link changes */
1391         vector++;
1392         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1393         if (rx_ring->itr_val)
1394                 writel(1000000000 / (rx_ring->itr_val * 256),
1395                        hw->hw_addr + E1000_EITR_82574(vector));
1396         else
1397                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1398
1399         /* Cause Tx interrupts on every write back */
1400         ivar |= (1 << 31);
1401
1402         ew32(IVAR, ivar);
1403
1404         /* enable MSI-X PBA support */
1405         ctrl_ext = er32(CTRL_EXT);
1406         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1407
1408         /* Auto-Mask Other interrupts upon ICR read */
1409 #define E1000_EIAC_MASK_82574   0x01F00000
1410         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1411         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1412         ew32(CTRL_EXT, ctrl_ext);
1413         e1e_flush();
1414 }
1415
1416 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1417 {
1418         if (adapter->msix_entries) {
1419                 pci_disable_msix(adapter->pdev);
1420                 kfree(adapter->msix_entries);
1421                 adapter->msix_entries = NULL;
1422         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1423                 pci_disable_msi(adapter->pdev);
1424                 adapter->flags &= ~FLAG_MSI_ENABLED;
1425         }
1426
1427         return;
1428 }
1429
1430 /**
1431  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1432  *
1433  * Attempt to configure interrupts using the best available
1434  * capabilities of the hardware and kernel.
1435  **/
1436 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1437 {
1438         int err;
1439         int numvecs, i;
1440
1441
1442         switch (adapter->int_mode) {
1443         case E1000E_INT_MODE_MSIX:
1444                 if (adapter->flags & FLAG_HAS_MSIX) {
1445                         numvecs = 3; /* RxQ0, TxQ0 and other */
1446                         adapter->msix_entries = kcalloc(numvecs,
1447                                                       sizeof(struct msix_entry),
1448                                                       GFP_KERNEL);
1449                         if (adapter->msix_entries) {
1450                                 for (i = 0; i < numvecs; i++)
1451                                         adapter->msix_entries[i].entry = i;
1452
1453                                 err = pci_enable_msix(adapter->pdev,
1454                                                       adapter->msix_entries,
1455                                                       numvecs);
1456                                 if (err == 0)
1457                                         return;
1458                         }
1459                         /* MSI-X failed, so fall through and try MSI */
1460                         e_err("Failed to initialize MSI-X interrupts.  "
1461                               "Falling back to MSI interrupts.\n");
1462                         e1000e_reset_interrupt_capability(adapter);
1463                 }
1464                 adapter->int_mode = E1000E_INT_MODE_MSI;
1465                 /* Fall through */
1466         case E1000E_INT_MODE_MSI:
1467                 if (!pci_enable_msi(adapter->pdev)) {
1468                         adapter->flags |= FLAG_MSI_ENABLED;
1469                 } else {
1470                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1471                         e_err("Failed to initialize MSI interrupts.  Falling "
1472                               "back to legacy interrupts.\n");
1473                 }
1474                 /* Fall through */
1475         case E1000E_INT_MODE_LEGACY:
1476                 /* Don't do anything; this is the system default */
1477                 break;
1478         }
1479
1480         return;
1481 }
1482
1483 /**
1484  * e1000_request_msix - Initialize MSI-X interrupts
1485  *
1486  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1487  * kernel.
1488  **/
1489 static int e1000_request_msix(struct e1000_adapter *adapter)
1490 {
1491         struct net_device *netdev = adapter->netdev;
1492         int err = 0, vector = 0;
1493
1494         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1495                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1496         else
1497                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1498         err = request_irq(adapter->msix_entries[vector].vector,
1499                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1500                           netdev);
1501         if (err)
1502                 goto out;
1503         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1504         adapter->rx_ring->itr_val = adapter->itr;
1505         vector++;
1506
1507         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1508                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1509         else
1510                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1511         err = request_irq(adapter->msix_entries[vector].vector,
1512                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1513                           netdev);
1514         if (err)
1515                 goto out;
1516         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1517         adapter->tx_ring->itr_val = adapter->itr;
1518         vector++;
1519
1520         err = request_irq(adapter->msix_entries[vector].vector,
1521                           e1000_msix_other, 0, netdev->name, netdev);
1522         if (err)
1523                 goto out;
1524
1525         e1000_configure_msix(adapter);
1526         return 0;
1527 out:
1528         return err;
1529 }
1530
1531 /**
1532  * e1000_request_irq - initialize interrupts
1533  *
1534  * Attempts to configure interrupts using the best available
1535  * capabilities of the hardware and kernel.
1536  **/
1537 static int e1000_request_irq(struct e1000_adapter *adapter)
1538 {
1539         struct net_device *netdev = adapter->netdev;
1540         int err;
1541
1542         if (adapter->msix_entries) {
1543                 err = e1000_request_msix(adapter);
1544                 if (!err)
1545                         return err;
1546                 /* fall back to MSI */
1547                 e1000e_reset_interrupt_capability(adapter);
1548                 adapter->int_mode = E1000E_INT_MODE_MSI;
1549                 e1000e_set_interrupt_capability(adapter);
1550         }
1551         if (adapter->flags & FLAG_MSI_ENABLED) {
1552                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1553                                   netdev->name, netdev);
1554                 if (!err)
1555                         return err;
1556
1557                 /* fall back to legacy interrupt */
1558                 e1000e_reset_interrupt_capability(adapter);
1559                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1560         }
1561
1562         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1563                           netdev->name, netdev);
1564         if (err)
1565                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1566
1567         return err;
1568 }
1569
1570 static void e1000_free_irq(struct e1000_adapter *adapter)
1571 {
1572         struct net_device *netdev = adapter->netdev;
1573
1574         if (adapter->msix_entries) {
1575                 int vector = 0;
1576
1577                 free_irq(adapter->msix_entries[vector].vector, netdev);
1578                 vector++;
1579
1580                 free_irq(adapter->msix_entries[vector].vector, netdev);
1581                 vector++;
1582
1583                 /* Other Causes interrupt vector */
1584                 free_irq(adapter->msix_entries[vector].vector, netdev);
1585                 return;
1586         }
1587
1588         free_irq(adapter->pdev->irq, netdev);
1589 }
1590
1591 /**
1592  * e1000_irq_disable - Mask off interrupt generation on the NIC
1593  **/
1594 static void e1000_irq_disable(struct e1000_adapter *adapter)
1595 {
1596         struct e1000_hw *hw = &adapter->hw;
1597
1598         ew32(IMC, ~0);
1599         if (adapter->msix_entries)
1600                 ew32(EIAC_82574, 0);
1601         e1e_flush();
1602         synchronize_irq(adapter->pdev->irq);
1603 }
1604
1605 /**
1606  * e1000_irq_enable - Enable default interrupt generation settings
1607  **/
1608 static void e1000_irq_enable(struct e1000_adapter *adapter)
1609 {
1610         struct e1000_hw *hw = &adapter->hw;
1611
1612         if (adapter->msix_entries) {
1613                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1614                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1615         } else {
1616                 ew32(IMS, IMS_ENABLE_MASK);
1617         }
1618         e1e_flush();
1619 }
1620
1621 /**
1622  * e1000_get_hw_control - get control of the h/w from f/w
1623  * @adapter: address of board private structure
1624  *
1625  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1626  * For ASF and Pass Through versions of f/w this means that
1627  * the driver is loaded. For AMT version (only with 82573)
1628  * of the f/w this means that the network i/f is open.
1629  **/
1630 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1631 {
1632         struct e1000_hw *hw = &adapter->hw;
1633         u32 ctrl_ext;
1634         u32 swsm;
1635
1636         /* Let firmware know the driver has taken over */
1637         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1638                 swsm = er32(SWSM);
1639                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1640         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1641                 ctrl_ext = er32(CTRL_EXT);
1642                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1643         }
1644 }
1645
1646 /**
1647  * e1000_release_hw_control - release control of the h/w to f/w
1648  * @adapter: address of board private structure
1649  *
1650  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1651  * For ASF and Pass Through versions of f/w this means that the
1652  * driver is no longer loaded. For AMT version (only with 82573) i
1653  * of the f/w this means that the network i/f is closed.
1654  *
1655  **/
1656 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1657 {
1658         struct e1000_hw *hw = &adapter->hw;
1659         u32 ctrl_ext;
1660         u32 swsm;
1661
1662         /* Let firmware taken over control of h/w */
1663         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1664                 swsm = er32(SWSM);
1665                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1666         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1667                 ctrl_ext = er32(CTRL_EXT);
1668                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1669         }
1670 }
1671
1672 /**
1673  * @e1000_alloc_ring - allocate memory for a ring structure
1674  **/
1675 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1676                                 struct e1000_ring *ring)
1677 {
1678         struct pci_dev *pdev = adapter->pdev;
1679
1680         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1681                                         GFP_KERNEL);
1682         if (!ring->desc)
1683                 return -ENOMEM;
1684
1685         return 0;
1686 }
1687
1688 /**
1689  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1690  * @adapter: board private structure
1691  *
1692  * Return 0 on success, negative on failure
1693  **/
1694 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1695 {
1696         struct e1000_ring *tx_ring = adapter->tx_ring;
1697         int err = -ENOMEM, size;
1698
1699         size = sizeof(struct e1000_buffer) * tx_ring->count;
1700         tx_ring->buffer_info = vmalloc(size);
1701         if (!tx_ring->buffer_info)
1702                 goto err;
1703         memset(tx_ring->buffer_info, 0, size);
1704
1705         /* round up to nearest 4K */
1706         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1707         tx_ring->size = ALIGN(tx_ring->size, 4096);
1708
1709         err = e1000_alloc_ring_dma(adapter, tx_ring);
1710         if (err)
1711                 goto err;
1712
1713         tx_ring->next_to_use = 0;
1714         tx_ring->next_to_clean = 0;
1715
1716         return 0;
1717 err:
1718         vfree(tx_ring->buffer_info);
1719         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1720         return err;
1721 }
1722
1723 /**
1724  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1725  * @adapter: board private structure
1726  *
1727  * Returns 0 on success, negative on failure
1728  **/
1729 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1730 {
1731         struct e1000_ring *rx_ring = adapter->rx_ring;
1732         struct e1000_buffer *buffer_info;
1733         int i, size, desc_len, err = -ENOMEM;
1734
1735         size = sizeof(struct e1000_buffer) * rx_ring->count;
1736         rx_ring->buffer_info = vmalloc(size);
1737         if (!rx_ring->buffer_info)
1738                 goto err;
1739         memset(rx_ring->buffer_info, 0, size);
1740
1741         for (i = 0; i < rx_ring->count; i++) {
1742                 buffer_info = &rx_ring->buffer_info[i];
1743                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1744                                                 sizeof(struct e1000_ps_page),
1745                                                 GFP_KERNEL);
1746                 if (!buffer_info->ps_pages)
1747                         goto err_pages;
1748         }
1749
1750         desc_len = sizeof(union e1000_rx_desc_packet_split);
1751
1752         /* Round up to nearest 4K */
1753         rx_ring->size = rx_ring->count * desc_len;
1754         rx_ring->size = ALIGN(rx_ring->size, 4096);
1755
1756         err = e1000_alloc_ring_dma(adapter, rx_ring);
1757         if (err)
1758                 goto err_pages;
1759
1760         rx_ring->next_to_clean = 0;
1761         rx_ring->next_to_use = 0;
1762         rx_ring->rx_skb_top = NULL;
1763
1764         return 0;
1765
1766 err_pages:
1767         for (i = 0; i < rx_ring->count; i++) {
1768                 buffer_info = &rx_ring->buffer_info[i];
1769                 kfree(buffer_info->ps_pages);
1770         }
1771 err:
1772         vfree(rx_ring->buffer_info);
1773         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1774         return err;
1775 }
1776
1777 /**
1778  * e1000_clean_tx_ring - Free Tx Buffers
1779  * @adapter: board private structure
1780  **/
1781 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1782 {
1783         struct e1000_ring *tx_ring = adapter->tx_ring;
1784         struct e1000_buffer *buffer_info;
1785         unsigned long size;
1786         unsigned int i;
1787
1788         for (i = 0; i < tx_ring->count; i++) {
1789                 buffer_info = &tx_ring->buffer_info[i];
1790                 e1000_put_txbuf(adapter, buffer_info);
1791         }
1792
1793         size = sizeof(struct e1000_buffer) * tx_ring->count;
1794         memset(tx_ring->buffer_info, 0, size);
1795
1796         memset(tx_ring->desc, 0, tx_ring->size);
1797
1798         tx_ring->next_to_use = 0;
1799         tx_ring->next_to_clean = 0;
1800
1801         writel(0, adapter->hw.hw_addr + tx_ring->head);
1802         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1803 }
1804
1805 /**
1806  * e1000e_free_tx_resources - Free Tx Resources per Queue
1807  * @adapter: board private structure
1808  *
1809  * Free all transmit software resources
1810  **/
1811 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1812 {
1813         struct pci_dev *pdev = adapter->pdev;
1814         struct e1000_ring *tx_ring = adapter->tx_ring;
1815
1816         e1000_clean_tx_ring(adapter);
1817
1818         vfree(tx_ring->buffer_info);
1819         tx_ring->buffer_info = NULL;
1820
1821         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1822                           tx_ring->dma);
1823         tx_ring->desc = NULL;
1824 }
1825
1826 /**
1827  * e1000e_free_rx_resources - Free Rx Resources
1828  * @adapter: board private structure
1829  *
1830  * Free all receive software resources
1831  **/
1832
1833 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1834 {
1835         struct pci_dev *pdev = adapter->pdev;
1836         struct e1000_ring *rx_ring = adapter->rx_ring;
1837         int i;
1838
1839         e1000_clean_rx_ring(adapter);
1840
1841         for (i = 0; i < rx_ring->count; i++) {
1842                 kfree(rx_ring->buffer_info[i].ps_pages);
1843         }
1844
1845         vfree(rx_ring->buffer_info);
1846         rx_ring->buffer_info = NULL;
1847
1848         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1849                           rx_ring->dma);
1850         rx_ring->desc = NULL;
1851 }
1852
1853 /**
1854  * e1000_update_itr - update the dynamic ITR value based on statistics
1855  * @adapter: pointer to adapter
1856  * @itr_setting: current adapter->itr
1857  * @packets: the number of packets during this measurement interval
1858  * @bytes: the number of bytes during this measurement interval
1859  *
1860  *      Stores a new ITR value based on packets and byte
1861  *      counts during the last interrupt.  The advantage of per interrupt
1862  *      computation is faster updates and more accurate ITR for the current
1863  *      traffic pattern.  Constants in this function were computed
1864  *      based on theoretical maximum wire speed and thresholds were set based
1865  *      on testing data as well as attempting to minimize response time
1866  *      while increasing bulk throughput.  This functionality is controlled
1867  *      by the InterruptThrottleRate module parameter.
1868  **/
1869 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1870                                      u16 itr_setting, int packets,
1871                                      int bytes)
1872 {
1873         unsigned int retval = itr_setting;
1874
1875         if (packets == 0)
1876                 goto update_itr_done;
1877
1878         switch (itr_setting) {
1879         case lowest_latency:
1880                 /* handle TSO and jumbo frames */
1881                 if (bytes/packets > 8000)
1882                         retval = bulk_latency;
1883                 else if ((packets < 5) && (bytes > 512)) {
1884                         retval = low_latency;
1885                 }
1886                 break;
1887         case low_latency:  /* 50 usec aka 20000 ints/s */
1888                 if (bytes > 10000) {
1889                         /* this if handles the TSO accounting */
1890                         if (bytes/packets > 8000) {
1891                                 retval = bulk_latency;
1892                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1893                                 retval = bulk_latency;
1894                         } else if ((packets > 35)) {
1895                                 retval = lowest_latency;
1896                         }
1897                 } else if (bytes/packets > 2000) {
1898                         retval = bulk_latency;
1899                 } else if (packets <= 2 && bytes < 512) {
1900                         retval = lowest_latency;
1901                 }
1902                 break;
1903         case bulk_latency: /* 250 usec aka 4000 ints/s */
1904                 if (bytes > 25000) {
1905                         if (packets > 35) {
1906                                 retval = low_latency;
1907                         }
1908                 } else if (bytes < 6000) {
1909                         retval = low_latency;
1910                 }
1911                 break;
1912         }
1913
1914 update_itr_done:
1915         return retval;
1916 }
1917
1918 static void e1000_set_itr(struct e1000_adapter *adapter)
1919 {
1920         struct e1000_hw *hw = &adapter->hw;
1921         u16 current_itr;
1922         u32 new_itr = adapter->itr;
1923
1924         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1925         if (adapter->link_speed != SPEED_1000) {
1926                 current_itr = 0;
1927                 new_itr = 4000;
1928                 goto set_itr_now;
1929         }
1930
1931         adapter->tx_itr = e1000_update_itr(adapter,
1932                                     adapter->tx_itr,
1933                                     adapter->total_tx_packets,
1934                                     adapter->total_tx_bytes);
1935         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1936         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1937                 adapter->tx_itr = low_latency;
1938
1939         adapter->rx_itr = e1000_update_itr(adapter,
1940                                     adapter->rx_itr,
1941                                     adapter->total_rx_packets,
1942                                     adapter->total_rx_bytes);
1943         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1944         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1945                 adapter->rx_itr = low_latency;
1946
1947         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1948
1949         switch (current_itr) {
1950         /* counts and packets in update_itr are dependent on these numbers */
1951         case lowest_latency:
1952                 new_itr = 70000;
1953                 break;
1954         case low_latency:
1955                 new_itr = 20000; /* aka hwitr = ~200 */
1956                 break;
1957         case bulk_latency:
1958                 new_itr = 4000;
1959                 break;
1960         default:
1961                 break;
1962         }
1963
1964 set_itr_now:
1965         if (new_itr != adapter->itr) {
1966                 /*
1967                  * this attempts to bias the interrupt rate towards Bulk
1968                  * by adding intermediate steps when interrupt rate is
1969                  * increasing
1970                  */
1971                 new_itr = new_itr > adapter->itr ?
1972                              min(adapter->itr + (new_itr >> 2), new_itr) :
1973                              new_itr;
1974                 adapter->itr = new_itr;
1975                 adapter->rx_ring->itr_val = new_itr;
1976                 if (adapter->msix_entries)
1977                         adapter->rx_ring->set_itr = 1;
1978                 else
1979                         ew32(ITR, 1000000000 / (new_itr * 256));
1980         }
1981 }
1982
1983 /**
1984  * e1000_alloc_queues - Allocate memory for all rings
1985  * @adapter: board private structure to initialize
1986  **/
1987 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1988 {
1989         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1990         if (!adapter->tx_ring)
1991                 goto err;
1992
1993         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1994         if (!adapter->rx_ring)
1995                 goto err;
1996
1997         return 0;
1998 err:
1999         e_err("Unable to allocate memory for queues\n");
2000         kfree(adapter->rx_ring);
2001         kfree(adapter->tx_ring);
2002         return -ENOMEM;
2003 }
2004
2005 /**
2006  * e1000_clean - NAPI Rx polling callback
2007  * @napi: struct associated with this polling callback
2008  * @budget: amount of packets driver is allowed to process this poll
2009  **/
2010 static int e1000_clean(struct napi_struct *napi, int budget)
2011 {
2012         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2013         struct e1000_hw *hw = &adapter->hw;
2014         struct net_device *poll_dev = adapter->netdev;
2015         int tx_cleaned = 1, work_done = 0;
2016
2017         adapter = netdev_priv(poll_dev);
2018
2019         if (adapter->msix_entries &&
2020             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2021                 goto clean_rx;
2022
2023         tx_cleaned = e1000_clean_tx_irq(adapter);
2024
2025 clean_rx:
2026         adapter->clean_rx(adapter, &work_done, budget);
2027
2028         if (!tx_cleaned)
2029                 work_done = budget;
2030
2031         /* If budget not fully consumed, exit the polling mode */
2032         if (work_done < budget) {
2033                 if (adapter->itr_setting & 3)
2034                         e1000_set_itr(adapter);
2035                 napi_complete(napi);
2036                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2037                         if (adapter->msix_entries)
2038                                 ew32(IMS, adapter->rx_ring->ims_val);
2039                         else
2040                                 e1000_irq_enable(adapter);
2041                 }
2042         }
2043
2044         return work_done;
2045 }
2046
2047 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2048 {
2049         struct e1000_adapter *adapter = netdev_priv(netdev);
2050         struct e1000_hw *hw = &adapter->hw;
2051         u32 vfta, index;
2052
2053         /* don't update vlan cookie if already programmed */
2054         if ((adapter->hw.mng_cookie.status &
2055              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2056             (vid == adapter->mng_vlan_id))
2057                 return;
2058
2059         /* add VID to filter table */
2060         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2061                 index = (vid >> 5) & 0x7F;
2062                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2063                 vfta |= (1 << (vid & 0x1F));
2064                 hw->mac.ops.write_vfta(hw, index, vfta);
2065         }
2066 }
2067
2068 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2069 {
2070         struct e1000_adapter *adapter = netdev_priv(netdev);
2071         struct e1000_hw *hw = &adapter->hw;
2072         u32 vfta, index;
2073
2074         if (!test_bit(__E1000_DOWN, &adapter->state))
2075                 e1000_irq_disable(adapter);
2076         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2077
2078         if (!test_bit(__E1000_DOWN, &adapter->state))
2079                 e1000_irq_enable(adapter);
2080
2081         if ((adapter->hw.mng_cookie.status &
2082              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2083             (vid == adapter->mng_vlan_id)) {
2084                 /* release control to f/w */
2085                 e1000_release_hw_control(adapter);
2086                 return;
2087         }
2088
2089         /* remove VID from filter table */
2090         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2091                 index = (vid >> 5) & 0x7F;
2092                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2093                 vfta &= ~(1 << (vid & 0x1F));
2094                 hw->mac.ops.write_vfta(hw, index, vfta);
2095         }
2096 }
2097
2098 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2099 {
2100         struct net_device *netdev = adapter->netdev;
2101         u16 vid = adapter->hw.mng_cookie.vlan_id;
2102         u16 old_vid = adapter->mng_vlan_id;
2103
2104         if (!adapter->vlgrp)
2105                 return;
2106
2107         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2108                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2109                 if (adapter->hw.mng_cookie.status &
2110                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2111                         e1000_vlan_rx_add_vid(netdev, vid);
2112                         adapter->mng_vlan_id = vid;
2113                 }
2114
2115                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2116                                 (vid != old_vid) &&
2117                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2118                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2119         } else {
2120                 adapter->mng_vlan_id = vid;
2121         }
2122 }
2123
2124
2125 static void e1000_vlan_rx_register(struct net_device *netdev,
2126                                    struct vlan_group *grp)
2127 {
2128         struct e1000_adapter *adapter = netdev_priv(netdev);
2129         struct e1000_hw *hw = &adapter->hw;
2130         u32 ctrl, rctl;
2131
2132         if (!test_bit(__E1000_DOWN, &adapter->state))
2133                 e1000_irq_disable(adapter);
2134         adapter->vlgrp = grp;
2135
2136         if (grp) {
2137                 /* enable VLAN tag insert/strip */
2138                 ctrl = er32(CTRL);
2139                 ctrl |= E1000_CTRL_VME;
2140                 ew32(CTRL, ctrl);
2141
2142                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2143                         /* enable VLAN receive filtering */
2144                         rctl = er32(RCTL);
2145                         rctl &= ~E1000_RCTL_CFIEN;
2146                         ew32(RCTL, rctl);
2147                         e1000_update_mng_vlan(adapter);
2148                 }
2149         } else {
2150                 /* disable VLAN tag insert/strip */
2151                 ctrl = er32(CTRL);
2152                 ctrl &= ~E1000_CTRL_VME;
2153                 ew32(CTRL, ctrl);
2154
2155                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2156                         if (adapter->mng_vlan_id !=
2157                             (u16)E1000_MNG_VLAN_NONE) {
2158                                 e1000_vlan_rx_kill_vid(netdev,
2159                                                        adapter->mng_vlan_id);
2160                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2161                         }
2162                 }
2163         }
2164
2165         if (!test_bit(__E1000_DOWN, &adapter->state))
2166                 e1000_irq_enable(adapter);
2167 }
2168
2169 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2170 {
2171         u16 vid;
2172
2173         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2174
2175         if (!adapter->vlgrp)
2176                 return;
2177
2178         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2179                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2180                         continue;
2181                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2182         }
2183 }
2184
2185 static void e1000_init_manageability(struct e1000_adapter *adapter)
2186 {
2187         struct e1000_hw *hw = &adapter->hw;
2188         u32 manc, manc2h;
2189
2190         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2191                 return;
2192
2193         manc = er32(MANC);
2194
2195         /*
2196          * enable receiving management packets to the host. this will probably
2197          * generate destination unreachable messages from the host OS, but
2198          * the packets will be handled on SMBUS
2199          */
2200         manc |= E1000_MANC_EN_MNG2HOST;
2201         manc2h = er32(MANC2H);
2202 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2203 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2204         manc2h |= E1000_MNG2HOST_PORT_623;
2205         manc2h |= E1000_MNG2HOST_PORT_664;
2206         ew32(MANC2H, manc2h);
2207         ew32(MANC, manc);
2208 }
2209
2210 /**
2211  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2212  * @adapter: board private structure
2213  *
2214  * Configure the Tx unit of the MAC after a reset.
2215  **/
2216 static void e1000_configure_tx(struct e1000_adapter *adapter)
2217 {
2218         struct e1000_hw *hw = &adapter->hw;
2219         struct e1000_ring *tx_ring = adapter->tx_ring;
2220         u64 tdba;
2221         u32 tdlen, tctl, tipg, tarc;
2222         u32 ipgr1, ipgr2;
2223
2224         /* Setup the HW Tx Head and Tail descriptor pointers */
2225         tdba = tx_ring->dma;
2226         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2227         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2228         ew32(TDBAH, (tdba >> 32));
2229         ew32(TDLEN, tdlen);
2230         ew32(TDH, 0);
2231         ew32(TDT, 0);
2232         tx_ring->head = E1000_TDH;
2233         tx_ring->tail = E1000_TDT;
2234
2235         /* Set the default values for the Tx Inter Packet Gap timer */
2236         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2237         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2238         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2239
2240         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2241                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2242
2243         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2244         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2245         ew32(TIPG, tipg);
2246
2247         /* Set the Tx Interrupt Delay register */
2248         ew32(TIDV, adapter->tx_int_delay);
2249         /* Tx irq moderation */
2250         ew32(TADV, adapter->tx_abs_int_delay);
2251
2252         /* Program the Transmit Control Register */
2253         tctl = er32(TCTL);
2254         tctl &= ~E1000_TCTL_CT;
2255         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2256                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2257
2258         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2259                 tarc = er32(TARC(0));
2260                 /*
2261                  * set the speed mode bit, we'll clear it if we're not at
2262                  * gigabit link later
2263                  */
2264 #define SPEED_MODE_BIT (1 << 21)
2265                 tarc |= SPEED_MODE_BIT;
2266                 ew32(TARC(0), tarc);
2267         }
2268
2269         /* errata: program both queues to unweighted RR */
2270         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2271                 tarc = er32(TARC(0));
2272                 tarc |= 1;
2273                 ew32(TARC(0), tarc);
2274                 tarc = er32(TARC(1));
2275                 tarc |= 1;
2276                 ew32(TARC(1), tarc);
2277         }
2278
2279         /* Setup Transmit Descriptor Settings for eop descriptor */
2280         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2281
2282         /* only set IDE if we are delaying interrupts using the timers */
2283         if (adapter->tx_int_delay)
2284                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2285
2286         /* enable Report Status bit */
2287         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2288
2289         ew32(TCTL, tctl);
2290
2291         e1000e_config_collision_dist(hw);
2292
2293         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
2294 }
2295
2296 /**
2297  * e1000_setup_rctl - configure the receive control registers
2298  * @adapter: Board private structure
2299  **/
2300 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2301                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2302 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2303 {
2304         struct e1000_hw *hw = &adapter->hw;
2305         u32 rctl, rfctl;
2306         u32 psrctl = 0;
2307         u32 pages = 0;
2308
2309         /* Program MC offset vector base */
2310         rctl = er32(RCTL);
2311         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2312         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2313                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2314                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2315
2316         /* Do not Store bad packets */
2317         rctl &= ~E1000_RCTL_SBP;
2318
2319         /* Enable Long Packet receive */
2320         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2321                 rctl &= ~E1000_RCTL_LPE;
2322         else
2323                 rctl |= E1000_RCTL_LPE;
2324
2325         /* Some systems expect that the CRC is included in SMBUS traffic. The
2326          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2327          * host memory when this is enabled
2328          */
2329         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2330                 rctl |= E1000_RCTL_SECRC;
2331
2332         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2333         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2334                 u16 phy_data;
2335
2336                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2337                 phy_data &= 0xfff8;
2338                 phy_data |= (1 << 2);
2339                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2340
2341                 e1e_rphy(hw, 22, &phy_data);
2342                 phy_data &= 0x0fff;
2343                 phy_data |= (1 << 14);
2344                 e1e_wphy(hw, 0x10, 0x2823);
2345                 e1e_wphy(hw, 0x11, 0x0003);
2346                 e1e_wphy(hw, 22, phy_data);
2347         }
2348
2349         /* Setup buffer sizes */
2350         rctl &= ~E1000_RCTL_SZ_4096;
2351         rctl |= E1000_RCTL_BSEX;
2352         switch (adapter->rx_buffer_len) {
2353         case 2048:
2354         default:
2355                 rctl |= E1000_RCTL_SZ_2048;
2356                 rctl &= ~E1000_RCTL_BSEX;
2357                 break;
2358         case 4096:
2359                 rctl |= E1000_RCTL_SZ_4096;
2360                 break;
2361         case 8192:
2362                 rctl |= E1000_RCTL_SZ_8192;
2363                 break;
2364         case 16384:
2365                 rctl |= E1000_RCTL_SZ_16384;
2366                 break;
2367         }
2368
2369         /*
2370          * 82571 and greater support packet-split where the protocol
2371          * header is placed in skb->data and the packet data is
2372          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2373          * In the case of a non-split, skb->data is linearly filled,
2374          * followed by the page buffers.  Therefore, skb->data is
2375          * sized to hold the largest protocol header.
2376          *
2377          * allocations using alloc_page take too long for regular MTU
2378          * so only enable packet split for jumbo frames
2379          *
2380          * Using pages when the page size is greater than 16k wastes
2381          * a lot of memory, since we allocate 3 pages at all times
2382          * per packet.
2383          */
2384         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2385         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2386             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2387                 adapter->rx_ps_pages = pages;
2388         else
2389                 adapter->rx_ps_pages = 0;
2390
2391         if (adapter->rx_ps_pages) {
2392                 /* Configure extra packet-split registers */
2393                 rfctl = er32(RFCTL);
2394                 rfctl |= E1000_RFCTL_EXTEN;
2395                 /*
2396                  * disable packet split support for IPv6 extension headers,
2397                  * because some malformed IPv6 headers can hang the Rx
2398                  */
2399                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2400                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2401
2402                 ew32(RFCTL, rfctl);
2403
2404                 /* Enable Packet split descriptors */
2405                 rctl |= E1000_RCTL_DTYP_PS;
2406
2407                 psrctl |= adapter->rx_ps_bsize0 >>
2408                         E1000_PSRCTL_BSIZE0_SHIFT;
2409
2410                 switch (adapter->rx_ps_pages) {
2411                 case 3:
2412                         psrctl |= PAGE_SIZE <<
2413                                 E1000_PSRCTL_BSIZE3_SHIFT;
2414                 case 2:
2415                         psrctl |= PAGE_SIZE <<
2416                                 E1000_PSRCTL_BSIZE2_SHIFT;
2417                 case 1:
2418                         psrctl |= PAGE_SIZE >>
2419                                 E1000_PSRCTL_BSIZE1_SHIFT;
2420                         break;
2421                 }
2422
2423                 ew32(PSRCTL, psrctl);
2424         }
2425
2426         ew32(RCTL, rctl);
2427         /* just started the receive unit, no need to restart */
2428         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2429 }
2430
2431 /**
2432  * e1000_configure_rx - Configure Receive Unit after Reset
2433  * @adapter: board private structure
2434  *
2435  * Configure the Rx unit of the MAC after a reset.
2436  **/
2437 static void e1000_configure_rx(struct e1000_adapter *adapter)
2438 {
2439         struct e1000_hw *hw = &adapter->hw;
2440         struct e1000_ring *rx_ring = adapter->rx_ring;
2441         u64 rdba;
2442         u32 rdlen, rctl, rxcsum, ctrl_ext;
2443
2444         if (adapter->rx_ps_pages) {
2445                 /* this is a 32 byte descriptor */
2446                 rdlen = rx_ring->count *
2447                         sizeof(union e1000_rx_desc_packet_split);
2448                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2449                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2450         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2451                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2452                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2453                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2454         } else {
2455                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2456                 adapter->clean_rx = e1000_clean_rx_irq;
2457                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2458         }
2459
2460         /* disable receives while setting up the descriptors */
2461         rctl = er32(RCTL);
2462         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2463         e1e_flush();
2464         msleep(10);
2465
2466         /* set the Receive Delay Timer Register */
2467         ew32(RDTR, adapter->rx_int_delay);
2468
2469         /* irq moderation */
2470         ew32(RADV, adapter->rx_abs_int_delay);
2471         if (adapter->itr_setting != 0)
2472                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2473
2474         ctrl_ext = er32(CTRL_EXT);
2475         /* Auto-Mask interrupts upon ICR access */
2476         ctrl_ext |= E1000_CTRL_EXT_IAME;
2477         ew32(IAM, 0xffffffff);
2478         ew32(CTRL_EXT, ctrl_ext);
2479         e1e_flush();
2480
2481         /*
2482          * Setup the HW Rx Head and Tail Descriptor Pointers and
2483          * the Base and Length of the Rx Descriptor Ring
2484          */
2485         rdba = rx_ring->dma;
2486         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2487         ew32(RDBAH, (rdba >> 32));
2488         ew32(RDLEN, rdlen);
2489         ew32(RDH, 0);
2490         ew32(RDT, 0);
2491         rx_ring->head = E1000_RDH;
2492         rx_ring->tail = E1000_RDT;
2493
2494         /* Enable Receive Checksum Offload for TCP and UDP */
2495         rxcsum = er32(RXCSUM);
2496         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2497                 rxcsum |= E1000_RXCSUM_TUOFL;
2498
2499                 /*
2500                  * IPv4 payload checksum for UDP fragments must be
2501                  * used in conjunction with packet-split.
2502                  */
2503                 if (adapter->rx_ps_pages)
2504                         rxcsum |= E1000_RXCSUM_IPPCSE;
2505         } else {
2506                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2507                 /* no need to clear IPPCSE as it defaults to 0 */
2508         }
2509         ew32(RXCSUM, rxcsum);
2510
2511         /*
2512          * Enable early receives on supported devices, only takes effect when
2513          * packet size is equal or larger than the specified value (in 8 byte
2514          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2515          */
2516         if (adapter->flags & FLAG_HAS_ERT) {
2517                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2518                         u32 rxdctl = er32(RXDCTL(0));
2519                         ew32(RXDCTL(0), rxdctl | 0x3);
2520                         ew32(ERT, E1000_ERT_2048 | (1 << 13));
2521                         /*
2522                          * With jumbo frames and early-receive enabled,
2523                          * excessive C-state transition latencies result in
2524                          * dropped transactions.
2525                          */
2526                         pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2527                                                   adapter->netdev->name, 55);
2528                 } else {
2529                         pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2530                                                   adapter->netdev->name,
2531                                                   PM_QOS_DEFAULT_VALUE);
2532                 }
2533         }
2534
2535         /* Enable Receives */
2536         ew32(RCTL, rctl);
2537 }
2538
2539 /**
2540  *  e1000_update_mc_addr_list - Update Multicast addresses
2541  *  @hw: pointer to the HW structure
2542  *  @mc_addr_list: array of multicast addresses to program
2543  *  @mc_addr_count: number of multicast addresses to program
2544  *
2545  *  Updates the Multicast Table Array.
2546  *  The caller must have a packed mc_addr_list of multicast addresses.
2547  **/
2548 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2549                                       u32 mc_addr_count)
2550 {
2551         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
2552 }
2553
2554 /**
2555  * e1000_set_multi - Multicast and Promiscuous mode set
2556  * @netdev: network interface device structure
2557  *
2558  * The set_multi entry point is called whenever the multicast address
2559  * list or the network interface flags are updated.  This routine is
2560  * responsible for configuring the hardware for proper multicast,
2561  * promiscuous mode, and all-multi behavior.
2562  **/
2563 static void e1000_set_multi(struct net_device *netdev)
2564 {
2565         struct e1000_adapter *adapter = netdev_priv(netdev);
2566         struct e1000_hw *hw = &adapter->hw;
2567         struct dev_mc_list *mc_ptr;
2568         u8  *mta_list;
2569         u32 rctl;
2570         int i;
2571
2572         /* Check for Promiscuous and All Multicast modes */
2573
2574         rctl = er32(RCTL);
2575
2576         if (netdev->flags & IFF_PROMISC) {
2577                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2578                 rctl &= ~E1000_RCTL_VFE;
2579         } else {
2580                 if (netdev->flags & IFF_ALLMULTI) {
2581                         rctl |= E1000_RCTL_MPE;
2582                         rctl &= ~E1000_RCTL_UPE;
2583                 } else {
2584                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2585                 }
2586                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2587                         rctl |= E1000_RCTL_VFE;
2588         }
2589
2590         ew32(RCTL, rctl);
2591
2592         if (!netdev_mc_empty(netdev)) {
2593                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
2594                 if (!mta_list)
2595                         return;
2596
2597                 /* prepare a packed array of only addresses. */
2598                 i = 0;
2599                 netdev_for_each_mc_addr(mc_ptr, netdev)
2600                         memcpy(mta_list + (i++ * ETH_ALEN),
2601                                mc_ptr->dmi_addr, ETH_ALEN);
2602
2603                 e1000_update_mc_addr_list(hw, mta_list, i);
2604                 kfree(mta_list);
2605         } else {
2606                 /*
2607                  * if we're called from probe, we might not have
2608                  * anything to do here, so clear out the list
2609                  */
2610                 e1000_update_mc_addr_list(hw, NULL, 0);
2611         }
2612 }
2613
2614 /**
2615  * e1000_configure - configure the hardware for Rx and Tx
2616  * @adapter: private board structure
2617  **/
2618 static void e1000_configure(struct e1000_adapter *adapter)
2619 {
2620         e1000_set_multi(adapter->netdev);
2621
2622         e1000_restore_vlan(adapter);
2623         e1000_init_manageability(adapter);
2624
2625         e1000_configure_tx(adapter);
2626         e1000_setup_rctl(adapter);
2627         e1000_configure_rx(adapter);
2628         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2629 }
2630
2631 /**
2632  * e1000e_power_up_phy - restore link in case the phy was powered down
2633  * @adapter: address of board private structure
2634  *
2635  * The phy may be powered down to save power and turn off link when the
2636  * driver is unloaded and wake on lan is not enabled (among others)
2637  * *** this routine MUST be followed by a call to e1000e_reset ***
2638  **/
2639 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2640 {
2641         if (adapter->hw.phy.ops.power_up)
2642                 adapter->hw.phy.ops.power_up(&adapter->hw);
2643
2644         adapter->hw.mac.ops.setup_link(&adapter->hw);
2645 }
2646
2647 /**
2648  * e1000_power_down_phy - Power down the PHY
2649  *
2650  * Power down the PHY so no link is implied when interface is down.
2651  * The PHY cannot be powered down if management or WoL is active.
2652  */
2653 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2654 {
2655         /* WoL is enabled */
2656         if (adapter->wol)
2657                 return;
2658
2659         if (adapter->hw.phy.ops.power_down)
2660                 adapter->hw.phy.ops.power_down(&adapter->hw);
2661 }
2662
2663 /**
2664  * e1000e_reset - bring the hardware into a known good state
2665  *
2666  * This function boots the hardware and enables some settings that
2667  * require a configuration cycle of the hardware - those cannot be
2668  * set/changed during runtime. After reset the device needs to be
2669  * properly configured for Rx, Tx etc.
2670  */
2671 void e1000e_reset(struct e1000_adapter *adapter)
2672 {
2673         struct e1000_mac_info *mac = &adapter->hw.mac;
2674         struct e1000_fc_info *fc = &adapter->hw.fc;
2675         struct e1000_hw *hw = &adapter->hw;
2676         u32 tx_space, min_tx_space, min_rx_space;
2677         u32 pba = adapter->pba;
2678         u16 hwm;
2679
2680         /* reset Packet Buffer Allocation to default */
2681         ew32(PBA, pba);
2682
2683         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2684                 /*
2685                  * To maintain wire speed transmits, the Tx FIFO should be
2686                  * large enough to accommodate two full transmit packets,
2687                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2688                  * the Rx FIFO should be large enough to accommodate at least
2689                  * one full receive packet and is similarly rounded up and
2690                  * expressed in KB.
2691                  */
2692                 pba = er32(PBA);
2693                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2694                 tx_space = pba >> 16;
2695                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2696                 pba &= 0xffff;
2697                 /*
2698                  * the Tx fifo also stores 16 bytes of information about the tx
2699                  * but don't include ethernet FCS because hardware appends it
2700                  */
2701                 min_tx_space = (adapter->max_frame_size +
2702                                 sizeof(struct e1000_tx_desc) -
2703                                 ETH_FCS_LEN) * 2;
2704                 min_tx_space = ALIGN(min_tx_space, 1024);
2705                 min_tx_space >>= 10;
2706                 /* software strips receive CRC, so leave room for it */
2707                 min_rx_space = adapter->max_frame_size;
2708                 min_rx_space = ALIGN(min_rx_space, 1024);
2709                 min_rx_space >>= 10;
2710
2711                 /*
2712                  * If current Tx allocation is less than the min Tx FIFO size,
2713                  * and the min Tx FIFO size is less than the current Rx FIFO
2714                  * allocation, take space away from current Rx allocation
2715                  */
2716                 if ((tx_space < min_tx_space) &&
2717                     ((min_tx_space - tx_space) < pba)) {
2718                         pba -= min_tx_space - tx_space;
2719
2720                         /*
2721                          * if short on Rx space, Rx wins and must trump tx
2722                          * adjustment or use Early Receive if available
2723                          */
2724                         if ((pba < min_rx_space) &&
2725                             (!(adapter->flags & FLAG_HAS_ERT)))
2726                                 /* ERT enabled in e1000_configure_rx */
2727                                 pba = min_rx_space;
2728                 }
2729
2730                 ew32(PBA, pba);
2731         }
2732
2733
2734         /*
2735          * flow control settings
2736          *
2737          * The high water mark must be low enough to fit one full frame
2738          * (or the size used for early receive) above it in the Rx FIFO.
2739          * Set it to the lower of:
2740          * - 90% of the Rx FIFO size, and
2741          * - the full Rx FIFO size minus the early receive size (for parts
2742          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2743          * - the full Rx FIFO size minus one full frame
2744          */
2745         if (hw->mac.type == e1000_pchlan) {
2746                 /*
2747                  * Workaround PCH LOM adapter hangs with certain network
2748                  * loads.  If hangs persist, try disabling Tx flow control.
2749                  */
2750                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2751                         fc->high_water = 0x3500;
2752                         fc->low_water  = 0x1500;
2753                 } else {
2754                         fc->high_water = 0x5000;
2755                         fc->low_water  = 0x3000;
2756                 }
2757         } else {
2758                 if ((adapter->flags & FLAG_HAS_ERT) &&
2759                     (adapter->netdev->mtu > ETH_DATA_LEN))
2760                         hwm = min(((pba << 10) * 9 / 10),
2761                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
2762                 else
2763                         hwm = min(((pba << 10) * 9 / 10),
2764                                   ((pba << 10) - adapter->max_frame_size));
2765
2766                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
2767                 fc->low_water = fc->high_water - 8;
2768         }
2769
2770         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2771                 fc->pause_time = 0xFFFF;
2772         else
2773                 fc->pause_time = E1000_FC_PAUSE_TIME;
2774         fc->send_xon = 1;
2775         fc->current_mode = fc->requested_mode;
2776
2777         /* Allow time for pending master requests to run */
2778         mac->ops.reset_hw(hw);
2779
2780         /*
2781          * For parts with AMT enabled, let the firmware know
2782          * that the network interface is in control
2783          */
2784         if (adapter->flags & FLAG_HAS_AMT)
2785                 e1000_get_hw_control(adapter);
2786
2787         ew32(WUC, 0);
2788         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP)
2789                 e1e_wphy(&adapter->hw, BM_WUC, 0);
2790
2791         if (mac->ops.init_hw(hw))
2792                 e_err("Hardware Error\n");
2793
2794         /* additional part of the flow-control workaround above */
2795         if (hw->mac.type == e1000_pchlan)
2796                 ew32(FCRTV_PCH, 0x1000);
2797
2798         e1000_update_mng_vlan(adapter);
2799
2800         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2801         ew32(VET, ETH_P_8021Q);
2802
2803         e1000e_reset_adaptive(hw);
2804         e1000_get_phy_info(hw);
2805
2806         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
2807             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2808                 u16 phy_data = 0;
2809                 /*
2810                  * speed up time to link by disabling smart power down, ignore
2811                  * the return value of this function because there is nothing
2812                  * different we would do if it failed
2813                  */
2814                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2815                 phy_data &= ~IGP02E1000_PM_SPD;
2816                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2817         }
2818 }
2819
2820 int e1000e_up(struct e1000_adapter *adapter)
2821 {
2822         struct e1000_hw *hw = &adapter->hw;
2823
2824         /* DMA latency requirement to workaround early-receive/jumbo issue */
2825         if (adapter->flags & FLAG_HAS_ERT)
2826                 pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY,
2827                                        adapter->netdev->name,
2828                                        PM_QOS_DEFAULT_VALUE);
2829
2830         /* hardware has been reset, we need to reload some things */
2831         e1000_configure(adapter);
2832
2833         clear_bit(__E1000_DOWN, &adapter->state);
2834
2835         napi_enable(&adapter->napi);
2836         if (adapter->msix_entries)
2837                 e1000_configure_msix(adapter);
2838         e1000_irq_enable(adapter);
2839
2840         netif_wake_queue(adapter->netdev);
2841
2842         /* fire a link change interrupt to start the watchdog */
2843         ew32(ICS, E1000_ICS_LSC);
2844         return 0;
2845 }
2846
2847 void e1000e_down(struct e1000_adapter *adapter)
2848 {
2849         struct net_device *netdev = adapter->netdev;
2850         struct e1000_hw *hw = &adapter->hw;
2851         u32 tctl, rctl;
2852
2853         /*
2854          * signal that we're down so the interrupt handler does not
2855          * reschedule our watchdog timer
2856          */
2857         set_bit(__E1000_DOWN, &adapter->state);
2858
2859         /* disable receives in the hardware */
2860         rctl = er32(RCTL);
2861         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2862         /* flush and sleep below */
2863
2864         netif_stop_queue(netdev);
2865
2866         /* disable transmits in the hardware */
2867         tctl = er32(TCTL);
2868         tctl &= ~E1000_TCTL_EN;
2869         ew32(TCTL, tctl);
2870         /* flush both disables and wait for them to finish */
2871         e1e_flush();
2872         msleep(10);
2873
2874         napi_disable(&adapter->napi);
2875         e1000_irq_disable(adapter);
2876
2877         del_timer_sync(&adapter->watchdog_timer);
2878         del_timer_sync(&adapter->phy_info_timer);
2879
2880         netdev->tx_queue_len = adapter->tx_queue_len;
2881         netif_carrier_off(netdev);
2882         adapter->link_speed = 0;
2883         adapter->link_duplex = 0;
2884
2885         if (!pci_channel_offline(adapter->pdev))
2886                 e1000e_reset(adapter);
2887         e1000_clean_tx_ring(adapter);
2888         e1000_clean_rx_ring(adapter);
2889
2890         if (adapter->flags & FLAG_HAS_ERT)
2891                 pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY,
2892                                           adapter->netdev->name);
2893
2894         /*
2895          * TODO: for power management, we could drop the link and
2896          * pci_disable_device here.
2897          */
2898 }
2899
2900 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2901 {
2902         might_sleep();
2903         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2904                 msleep(1);
2905         e1000e_down(adapter);
2906         e1000e_up(adapter);
2907         clear_bit(__E1000_RESETTING, &adapter->state);
2908 }
2909
2910 /**
2911  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2912  * @adapter: board private structure to initialize
2913  *
2914  * e1000_sw_init initializes the Adapter private data structure.
2915  * Fields are initialized based on PCI device information and
2916  * OS network device settings (MTU size).
2917  **/
2918 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2919 {
2920         struct net_device *netdev = adapter->netdev;
2921
2922         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2923         adapter->rx_ps_bsize0 = 128;
2924         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2925         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2926
2927         e1000e_set_interrupt_capability(adapter);
2928
2929         if (e1000_alloc_queues(adapter))
2930                 return -ENOMEM;
2931
2932         /* Explicitly disable IRQ since the NIC can be in any state. */
2933         e1000_irq_disable(adapter);
2934
2935         set_bit(__E1000_DOWN, &adapter->state);
2936         return 0;
2937 }
2938
2939 /**
2940  * e1000_intr_msi_test - Interrupt Handler
2941  * @irq: interrupt number
2942  * @data: pointer to a network interface device structure
2943  **/
2944 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2945 {
2946         struct net_device *netdev = data;
2947         struct e1000_adapter *adapter = netdev_priv(netdev);
2948         struct e1000_hw *hw = &adapter->hw;
2949         u32 icr = er32(ICR);
2950
2951         e_dbg("icr is %08X\n", icr);
2952         if (icr & E1000_ICR_RXSEQ) {
2953                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2954                 wmb();
2955         }
2956
2957         return IRQ_HANDLED;
2958 }
2959
2960 /**
2961  * e1000_test_msi_interrupt - Returns 0 for successful test
2962  * @adapter: board private struct
2963  *
2964  * code flow taken from tg3.c
2965  **/
2966 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2967 {
2968         struct net_device *netdev = adapter->netdev;
2969         struct e1000_hw *hw = &adapter->hw;
2970         int err;
2971
2972         /* poll_enable hasn't been called yet, so don't need disable */
2973         /* clear any pending events */
2974         er32(ICR);
2975
2976         /* free the real vector and request a test handler */
2977         e1000_free_irq(adapter);
2978         e1000e_reset_interrupt_capability(adapter);
2979
2980         /* Assume that the test fails, if it succeeds then the test
2981          * MSI irq handler will unset this flag */
2982         adapter->flags |= FLAG_MSI_TEST_FAILED;
2983
2984         err = pci_enable_msi(adapter->pdev);
2985         if (err)
2986                 goto msi_test_failed;
2987
2988         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
2989                           netdev->name, netdev);
2990         if (err) {
2991                 pci_disable_msi(adapter->pdev);
2992                 goto msi_test_failed;
2993         }
2994
2995         wmb();
2996
2997         e1000_irq_enable(adapter);
2998
2999         /* fire an unusual interrupt on the test handler */
3000         ew32(ICS, E1000_ICS_RXSEQ);
3001         e1e_flush();
3002         msleep(50);
3003
3004         e1000_irq_disable(adapter);
3005
3006         rmb();
3007
3008         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3009                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3010                 err = -EIO;
3011                 e_info("MSI interrupt test failed!\n");
3012         }
3013
3014         free_irq(adapter->pdev->irq, netdev);
3015         pci_disable_msi(adapter->pdev);
3016
3017         if (err == -EIO)
3018                 goto msi_test_failed;
3019
3020         /* okay so the test worked, restore settings */
3021         e_dbg("MSI interrupt test succeeded!\n");
3022 msi_test_failed:
3023         e1000e_set_interrupt_capability(adapter);
3024         e1000_request_irq(adapter);
3025         return err;
3026 }
3027
3028 /**
3029  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3030  * @adapter: board private struct
3031  *
3032  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3033  **/
3034 static int e1000_test_msi(struct e1000_adapter *adapter)
3035 {
3036         int err;
3037         u16 pci_cmd;
3038
3039         if (!(adapter->flags & FLAG_MSI_ENABLED))
3040                 return 0;
3041
3042         /* disable SERR in case the MSI write causes a master abort */
3043         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3044         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3045                               pci_cmd & ~PCI_COMMAND_SERR);
3046
3047         err = e1000_test_msi_interrupt(adapter);
3048
3049         /* restore previous setting of command word */
3050         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3051
3052         /* success ! */
3053         if (!err)
3054                 return 0;
3055
3056         /* EIO means MSI test failed */
3057         if (err != -EIO)
3058                 return err;
3059
3060         /* back to INTx mode */
3061         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3062
3063         e1000_free_irq(adapter);
3064
3065         err = e1000_request_irq(adapter);
3066
3067         return err;
3068 }
3069
3070 /**
3071  * e1000_open - Called when a network interface is made active
3072  * @netdev: network interface device structure
3073  *
3074  * Returns 0 on success, negative value on failure
3075  *
3076  * The open entry point is called when a network interface is made
3077  * active by the system (IFF_UP).  At this point all resources needed
3078  * for transmit and receive operations are allocated, the interrupt
3079  * handler is registered with the OS, the watchdog timer is started,
3080  * and the stack is notified that the interface is ready.
3081  **/
3082 static int e1000_open(struct net_device *netdev)
3083 {
3084         struct e1000_adapter *adapter = netdev_priv(netdev);
3085         struct e1000_hw *hw = &adapter->hw;
3086         int err;
3087
3088         /* disallow open during test */
3089         if (test_bit(__E1000_TESTING, &adapter->state))
3090                 return -EBUSY;
3091
3092         netif_carrier_off(netdev);
3093
3094         /* allocate transmit descriptors */
3095         err = e1000e_setup_tx_resources(adapter);
3096         if (err)
3097                 goto err_setup_tx;
3098
3099         /* allocate receive descriptors */
3100         err = e1000e_setup_rx_resources(adapter);
3101         if (err)
3102                 goto err_setup_rx;
3103
3104         e1000e_power_up_phy(adapter);
3105
3106         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3107         if ((adapter->hw.mng_cookie.status &
3108              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3109                 e1000_update_mng_vlan(adapter);
3110
3111         /*
3112          * If AMT is enabled, let the firmware know that the network
3113          * interface is now open
3114          */
3115         if (adapter->flags & FLAG_HAS_AMT)
3116                 e1000_get_hw_control(adapter);
3117
3118         /*
3119          * before we allocate an interrupt, we must be ready to handle it.
3120          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3121          * as soon as we call pci_request_irq, so we have to setup our
3122          * clean_rx handler before we do so.
3123          */
3124         e1000_configure(adapter);
3125
3126         err = e1000_request_irq(adapter);
3127         if (err)
3128                 goto err_req_irq;
3129
3130         /*
3131          * Work around PCIe errata with MSI interrupts causing some chipsets to
3132          * ignore e1000e MSI messages, which means we need to test our MSI
3133          * interrupt now
3134          */
3135         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3136                 err = e1000_test_msi(adapter);
3137                 if (err) {
3138                         e_err("Interrupt allocation failed\n");
3139                         goto err_req_irq;
3140                 }
3141         }
3142
3143         /* From here on the code is the same as e1000e_up() */
3144         clear_bit(__E1000_DOWN, &adapter->state);
3145
3146         napi_enable(&adapter->napi);
3147
3148         e1000_irq_enable(adapter);
3149
3150         netif_start_queue(netdev);
3151
3152         /* fire a link status change interrupt to start the watchdog */
3153         ew32(ICS, E1000_ICS_LSC);
3154
3155         return 0;
3156
3157 err_req_irq:
3158         e1000_release_hw_control(adapter);
3159         e1000_power_down_phy(adapter);
3160         e1000e_free_rx_resources(adapter);
3161 err_setup_rx:
3162         e1000e_free_tx_resources(adapter);
3163 err_setup_tx:
3164         e1000e_reset(adapter);
3165
3166         return err;
3167 }
3168
3169 /**
3170  * e1000_close - Disables a network interface
3171  * @netdev: network interface device structure
3172  *
3173  * Returns 0, this is not allowed to fail
3174  *
3175  * The close entry point is called when an interface is de-activated
3176  * by the OS.  The hardware is still under the drivers control, but
3177  * needs to be disabled.  A global MAC reset is issued to stop the
3178  * hardware, and all transmit and receive resources are freed.
3179  **/
3180 static int e1000_close(struct net_device *netdev)
3181 {
3182         struct e1000_adapter *adapter = netdev_priv(netdev);
3183
3184         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3185         e1000e_down(adapter);
3186         e1000_power_down_phy(adapter);
3187         e1000_free_irq(adapter);
3188
3189         e1000e_free_tx_resources(adapter);
3190         e1000e_free_rx_resources(adapter);
3191
3192         /*
3193          * kill manageability vlan ID if supported, but not if a vlan with
3194          * the same ID is registered on the host OS (let 8021q kill it)
3195          */
3196         if ((adapter->hw.mng_cookie.status &
3197                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3198              !(adapter->vlgrp &&
3199                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3200                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3201
3202         /*
3203          * If AMT is enabled, let the firmware know that the network
3204          * interface is now closed
3205          */
3206         if (adapter->flags & FLAG_HAS_AMT)
3207                 e1000_release_hw_control(adapter);
3208
3209         return 0;
3210 }
3211 /**
3212  * e1000_set_mac - Change the Ethernet Address of the NIC
3213  * @netdev: network interface device structure
3214  * @p: pointer to an address structure
3215  *
3216  * Returns 0 on success, negative on failure
3217  **/
3218 static int e1000_set_mac(struct net_device *netdev, void *p)
3219 {
3220         struct e1000_adapter *adapter = netdev_priv(netdev);
3221         struct sockaddr *addr = p;
3222
3223         if (!is_valid_ether_addr(addr->sa_data))
3224                 return -EADDRNOTAVAIL;
3225
3226         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3227         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3228
3229         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3230
3231         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3232                 /* activate the work around */
3233                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3234
3235                 /*
3236                  * Hold a copy of the LAA in RAR[14] This is done so that
3237                  * between the time RAR[0] gets clobbered  and the time it
3238                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3239                  * of the RARs and no incoming packets directed to this port
3240                  * are dropped. Eventually the LAA will be in RAR[0] and
3241                  * RAR[14]
3242                  */
3243                 e1000e_rar_set(&adapter->hw,
3244                               adapter->hw.mac.addr,
3245                               adapter->hw.mac.rar_entry_count - 1);
3246         }
3247
3248         return 0;
3249 }
3250
3251 /**
3252  * e1000e_update_phy_task - work thread to update phy
3253  * @work: pointer to our work struct
3254  *
3255  * this worker thread exists because we must acquire a
3256  * semaphore to read the phy, which we could msleep while
3257  * waiting for it, and we can't msleep in a timer.
3258  **/
3259 static void e1000e_update_phy_task(struct work_struct *work)
3260 {
3261         struct e1000_adapter *adapter = container_of(work,
3262                                         struct e1000_adapter, update_phy_task);
3263         e1000_get_phy_info(&adapter->hw);
3264 }
3265
3266 /*
3267  * Need to wait a few seconds after link up to get diagnostic information from
3268  * the phy
3269  */
3270 static void e1000_update_phy_info(unsigned long data)
3271 {
3272         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3273         schedule_work(&adapter->update_phy_task);
3274 }
3275
3276 /**
3277  * e1000e_update_stats - Update the board statistics counters
3278  * @adapter: board private structure
3279  **/
3280 void e1000e_update_stats(struct e1000_adapter *adapter)
3281 {
3282         struct net_device *netdev = adapter->netdev;
3283         struct e1000_hw *hw = &adapter->hw;
3284         struct pci_dev *pdev = adapter->pdev;
3285         u16 phy_data;
3286
3287         /*
3288          * Prevent stats update while adapter is being reset, or if the pci
3289          * connection is down.
3290          */
3291         if (adapter->link_speed == 0)
3292                 return;
3293         if (pci_channel_offline(pdev))
3294                 return;
3295
3296         adapter->stats.crcerrs += er32(CRCERRS);
3297         adapter->stats.gprc += er32(GPRC);
3298         adapter->stats.gorc += er32(GORCL);
3299         er32(GORCH); /* Clear gorc */
3300         adapter->stats.bprc += er32(BPRC);
3301         adapter->stats.mprc += er32(MPRC);
3302         adapter->stats.roc += er32(ROC);
3303
3304         adapter->stats.mpc += er32(MPC);
3305         if ((hw->phy.type == e1000_phy_82578) ||
3306             (hw->phy.type == e1000_phy_82577)) {
3307                 e1e_rphy(hw, HV_SCC_UPPER, &phy_data);
3308                 if (!e1e_rphy(hw, HV_SCC_LOWER, &phy_data))
3309                         adapter->stats.scc += phy_data;
3310
3311                 e1e_rphy(hw, HV_ECOL_UPPER, &phy_data);
3312                 if (!e1e_rphy(hw, HV_ECOL_LOWER, &phy_data))
3313                         adapter->stats.ecol += phy_data;
3314
3315                 e1e_rphy(hw, HV_MCC_UPPER, &phy_data);
3316                 if (!e1e_rphy(hw, HV_MCC_LOWER, &phy_data))
3317                         adapter->stats.mcc += phy_data;
3318
3319                 e1e_rphy(hw, HV_LATECOL_UPPER, &phy_data);
3320                 if (!e1e_rphy(hw, HV_LATECOL_LOWER, &phy_data))
3321                         adapter->stats.latecol += phy_data;
3322
3323                 e1e_rphy(hw, HV_DC_UPPER, &phy_data);
3324                 if (!e1e_rphy(hw, HV_DC_LOWER, &phy_data))
3325                         adapter->stats.dc += phy_data;
3326         } else {
3327                 adapter->stats.scc += er32(SCC);
3328                 adapter->stats.ecol += er32(ECOL);
3329                 adapter->stats.mcc += er32(MCC);
3330                 adapter->stats.latecol += er32(LATECOL);
3331                 adapter->stats.dc += er32(DC);
3332         }
3333         adapter->stats.xonrxc += er32(XONRXC);
3334         adapter->stats.xontxc += er32(XONTXC);
3335         adapter->stats.xoffrxc += er32(XOFFRXC);
3336         adapter->stats.xofftxc += er32(XOFFTXC);
3337         adapter->stats.gptc += er32(GPTC);
3338         adapter->stats.gotc += er32(GOTCL);
3339         er32(GOTCH); /* Clear gotc */
3340         adapter->stats.rnbc += er32(RNBC);
3341         adapter->stats.ruc += er32(RUC);
3342
3343         adapter->stats.mptc += er32(MPTC);
3344         adapter->stats.bptc += er32(BPTC);
3345
3346         /* used for adaptive IFS */
3347
3348         hw->mac.tx_packet_delta = er32(TPT);
3349         adapter->stats.tpt += hw->mac.tx_packet_delta;
3350         if ((hw->phy.type == e1000_phy_82578) ||
3351             (hw->phy.type == e1000_phy_82577)) {
3352                 e1e_rphy(hw, HV_COLC_UPPER, &phy_data);
3353                 if (!e1e_rphy(hw, HV_COLC_LOWER, &phy_data))
3354                         hw->mac.collision_delta = phy_data;
3355         } else {
3356                 hw->mac.collision_delta = er32(COLC);
3357         }
3358         adapter->stats.colc += hw->mac.collision_delta;
3359
3360         adapter->stats.algnerrc += er32(ALGNERRC);
3361         adapter->stats.rxerrc += er32(RXERRC);
3362         if ((hw->phy.type == e1000_phy_82578) ||
3363             (hw->phy.type == e1000_phy_82577)) {
3364                 e1e_rphy(hw, HV_TNCRS_UPPER, &phy_data);
3365                 if (!e1e_rphy(hw, HV_TNCRS_LOWER, &phy_data))
3366                         adapter->stats.tncrs += phy_data;
3367         } else {
3368                 if ((hw->mac.type != e1000_82574) &&
3369                     (hw->mac.type != e1000_82583))
3370                         adapter->stats.tncrs += er32(TNCRS);
3371         }
3372         adapter->stats.cexterr += er32(CEXTERR);
3373         adapter->stats.tsctc += er32(TSCTC);
3374         adapter->stats.tsctfc += er32(TSCTFC);
3375
3376         /* Fill out the OS statistics structure */
3377         netdev->stats.multicast = adapter->stats.mprc;
3378         netdev->stats.collisions = adapter->stats.colc;
3379
3380         /* Rx Errors */
3381
3382         /*
3383          * RLEC on some newer hardware can be incorrect so build
3384          * our own version based on RUC and ROC
3385          */
3386         netdev->stats.rx_errors = adapter->stats.rxerrc +
3387                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3388                 adapter->stats.ruc + adapter->stats.roc +
3389                 adapter->stats.cexterr;
3390         netdev->stats.rx_length_errors = adapter->stats.ruc +
3391                                               adapter->stats.roc;
3392         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3393         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3394         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3395
3396         /* Tx Errors */
3397         netdev->stats.tx_errors = adapter->stats.ecol +
3398                                        adapter->stats.latecol;
3399         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3400         netdev->stats.tx_window_errors = adapter->stats.latecol;
3401         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3402
3403         /* Tx Dropped needs to be maintained elsewhere */
3404
3405         /* Management Stats */
3406         adapter->stats.mgptc += er32(MGTPTC);
3407         adapter->stats.mgprc += er32(MGTPRC);
3408         adapter->stats.mgpdc += er32(MGTPDC);
3409 }
3410
3411 /**
3412  * e1000_phy_read_status - Update the PHY register status snapshot
3413  * @adapter: board private structure
3414  **/
3415 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3416 {
3417         struct e1000_hw *hw = &adapter->hw;
3418         struct e1000_phy_regs *phy = &adapter->phy_regs;
3419         int ret_val;
3420
3421         if ((er32(STATUS) & E1000_STATUS_LU) &&
3422             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3423                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3424                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3425                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3426                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3427                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3428                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3429                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3430                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3431                 if (ret_val)
3432                         e_warn("Error reading PHY register\n");
3433         } else {
3434                 /*
3435                  * Do not read PHY registers if link is not up
3436                  * Set values to typical power-on defaults
3437                  */
3438                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3439                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3440                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3441                              BMSR_ERCAP);
3442                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3443                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3444                 phy->lpa = 0;
3445                 phy->expansion = EXPANSION_ENABLENPAGE;
3446                 phy->ctrl1000 = ADVERTISE_1000FULL;
3447                 phy->stat1000 = 0;
3448                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3449         }
3450 }
3451
3452 static void e1000_print_link_info(struct e1000_adapter *adapter)
3453 {
3454         struct e1000_hw *hw = &adapter->hw;
3455         u32 ctrl = er32(CTRL);
3456
3457         /* Link status message must follow this format for user tools */
3458         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3459                "Flow Control: %s\n",
3460                adapter->netdev->name,
3461                adapter->link_speed,
3462                (adapter->link_duplex == FULL_DUPLEX) ?
3463                                 "Full Duplex" : "Half Duplex",
3464                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3465                                 "RX/TX" :
3466                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3467                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3468 }
3469
3470 bool e1000e_has_link(struct e1000_adapter *adapter)
3471 {
3472         struct e1000_hw *hw = &adapter->hw;
3473         bool link_active = 0;
3474         s32 ret_val = 0;
3475
3476         /*
3477          * get_link_status is set on LSC (link status) interrupt or
3478          * Rx sequence error interrupt.  get_link_status will stay
3479          * false until the check_for_link establishes link
3480          * for copper adapters ONLY
3481          */
3482         switch (hw->phy.media_type) {
3483         case e1000_media_type_copper:
3484                 if (hw->mac.get_link_status) {
3485                         ret_val = hw->mac.ops.check_for_link(hw);
3486                         link_active = !hw->mac.get_link_status;
3487                 } else {
3488                         link_active = 1;
3489                 }
3490                 break;
3491         case e1000_media_type_fiber:
3492                 ret_val = hw->mac.ops.check_for_link(hw);
3493                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3494                 break;
3495         case e1000_media_type_internal_serdes:
3496                 ret_val = hw->mac.ops.check_for_link(hw);
3497                 link_active = adapter->hw.mac.serdes_has_link;
3498                 break;
3499         default:
3500         case e1000_media_type_unknown:
3501                 break;
3502         }
3503
3504         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3505             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3506                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3507                 e_info("Gigabit has been disabled, downgrading speed\n");
3508         }
3509
3510         return link_active;
3511 }
3512
3513 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3514 {
3515         /* make sure the receive unit is started */
3516         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3517             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3518                 struct e1000_hw *hw = &adapter->hw;
3519                 u32 rctl = er32(RCTL);
3520                 ew32(RCTL, rctl | E1000_RCTL_EN);
3521                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3522         }
3523 }
3524
3525 /**
3526  * e1000_watchdog - Timer Call-back
3527  * @data: pointer to adapter cast into an unsigned long
3528  **/
3529 static void e1000_watchdog(unsigned long data)
3530 {
3531         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3532
3533         /* Do the rest outside of interrupt context */
3534         schedule_work(&adapter->watchdog_task);
3535
3536         /* TODO: make this use queue_delayed_work() */
3537 }
3538
3539 static void e1000_watchdog_task(struct work_struct *work)
3540 {
3541         struct e1000_adapter *adapter = container_of(work,
3542                                         struct e1000_adapter, watchdog_task);
3543         struct net_device *netdev = adapter->netdev;
3544         struct e1000_mac_info *mac = &adapter->hw.mac;
3545         struct e1000_phy_info *phy = &adapter->hw.phy;
3546         struct e1000_ring *tx_ring = adapter->tx_ring;
3547         struct e1000_hw *hw = &adapter->hw;
3548         u32 link, tctl;
3549         int tx_pending = 0;
3550
3551         link = e1000e_has_link(adapter);
3552         if ((netif_carrier_ok(netdev)) && link) {
3553                 e1000e_enable_receives(adapter);
3554                 goto link_up;
3555         }
3556
3557         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3558             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3559                 e1000_update_mng_vlan(adapter);
3560
3561         if (link) {
3562                 if (!netif_carrier_ok(netdev)) {
3563                         bool txb2b = 1;
3564                         /* update snapshot of PHY registers on LSC */
3565                         e1000_phy_read_status(adapter);
3566                         mac->ops.get_link_up_info(&adapter->hw,
3567                                                    &adapter->link_speed,
3568                                                    &adapter->link_duplex);
3569                         e1000_print_link_info(adapter);
3570                         /*
3571                          * On supported PHYs, check for duplex mismatch only
3572                          * if link has autonegotiated at 10/100 half
3573                          */
3574                         if ((hw->phy.type == e1000_phy_igp_3 ||
3575                              hw->phy.type == e1000_phy_bm) &&
3576                             (hw->mac.autoneg == true) &&
3577                             (adapter->link_speed == SPEED_10 ||
3578                              adapter->link_speed == SPEED_100) &&
3579                             (adapter->link_duplex == HALF_DUPLEX)) {
3580                                 u16 autoneg_exp;
3581
3582                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3583
3584                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3585                                         e_info("Autonegotiated half duplex but"
3586                                                " link partner cannot autoneg. "
3587                                                " Try forcing full duplex if "
3588                                                "link gets many collisions.\n");
3589                         }
3590
3591                         /*
3592                          * tweak tx_queue_len according to speed/duplex
3593                          * and adjust the timeout factor
3594                          */
3595                         netdev->tx_queue_len = adapter->tx_queue_len;
3596                         adapter->tx_timeout_factor = 1;
3597                         switch (adapter->link_speed) {
3598                         case SPEED_10:
3599                                 txb2b = 0;
3600                                 netdev->tx_queue_len = 10;
3601                                 adapter->tx_timeout_factor = 16;
3602                                 break;
3603                         case SPEED_100:
3604                                 txb2b = 0;
3605                                 netdev->tx_queue_len = 100;
3606                                 adapter->tx_timeout_factor = 10;
3607                                 break;
3608                         }
3609
3610                         /*
3611                          * workaround: re-program speed mode bit after
3612                          * link-up event
3613                          */
3614                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3615                             !txb2b) {
3616                                 u32 tarc0;
3617                                 tarc0 = er32(TARC(0));
3618                                 tarc0 &= ~SPEED_MODE_BIT;
3619                                 ew32(TARC(0), tarc0);
3620                         }
3621
3622                         /*
3623                          * disable TSO for pcie and 10/100 speeds, to avoid
3624                          * some hardware issues
3625                          */
3626                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3627                                 switch (adapter->link_speed) {
3628                                 case SPEED_10:
3629                                 case SPEED_100:
3630                                         e_info("10/100 speed: disabling TSO\n");
3631                                         netdev->features &= ~NETIF_F_TSO;
3632                                         netdev->features &= ~NETIF_F_TSO6;
3633                                         break;
3634                                 case SPEED_1000:
3635                                         netdev->features |= NETIF_F_TSO;
3636                                         netdev->features |= NETIF_F_TSO6;
3637                                         break;
3638                                 default:
3639                                         /* oops */
3640                                         break;
3641                                 }
3642                         }
3643
3644                         /*
3645                          * enable transmits in the hardware, need to do this
3646                          * after setting TARC(0)
3647                          */
3648                         tctl = er32(TCTL);
3649                         tctl |= E1000_TCTL_EN;
3650                         ew32(TCTL, tctl);
3651
3652                         /*
3653                          * Perform any post-link-up configuration before
3654                          * reporting link up.
3655                          */
3656                         if (phy->ops.cfg_on_link_up)
3657                                 phy->ops.cfg_on_link_up(hw);
3658
3659                         netif_carrier_on(netdev);
3660
3661                         if (!test_bit(__E1000_DOWN, &adapter->state))
3662                                 mod_timer(&adapter->phy_info_timer,
3663                                           round_jiffies(jiffies + 2 * HZ));
3664                 }
3665         } else {
3666                 if (netif_carrier_ok(netdev)) {
3667                         adapter->link_speed = 0;
3668                         adapter->link_duplex = 0;
3669                         /* Link status message must follow this format */
3670                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
3671                                adapter->netdev->name);
3672                         netif_carrier_off(netdev);
3673                         if (!test_bit(__E1000_DOWN, &adapter->state))
3674                                 mod_timer(&adapter->phy_info_timer,
3675                                           round_jiffies(jiffies + 2 * HZ));
3676
3677                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3678                                 schedule_work(&adapter->reset_task);
3679                 }
3680         }
3681
3682 link_up:
3683         e1000e_update_stats(adapter);
3684
3685         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3686         adapter->tpt_old = adapter->stats.tpt;
3687         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3688         adapter->colc_old = adapter->stats.colc;
3689
3690         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3691         adapter->gorc_old = adapter->stats.gorc;
3692         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3693         adapter->gotc_old = adapter->stats.gotc;
3694
3695         e1000e_update_adaptive(&adapter->hw);
3696
3697         if (!netif_carrier_ok(netdev)) {
3698                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3699                                tx_ring->count);
3700                 if (tx_pending) {
3701                         /*
3702                          * We've lost link, so the controller stops DMA,
3703                          * but we've got queued Tx work that's never going
3704                          * to get done, so reset controller to flush Tx.
3705                          * (Do the reset outside of interrupt context).
3706                          */
3707                         adapter->tx_timeout_count++;
3708                         schedule_work(&adapter->reset_task);
3709                         /* return immediately since reset is imminent */
3710                         return;
3711                 }
3712         }
3713
3714         /* Cause software interrupt to ensure Rx ring is cleaned */
3715         if (adapter->msix_entries)
3716                 ew32(ICS, adapter->rx_ring->ims_val);
3717         else
3718                 ew32(ICS, E1000_ICS_RXDMT0);
3719
3720         /* Force detection of hung controller every watchdog period */
3721         adapter->detect_tx_hung = 1;
3722
3723         /*
3724          * With 82571 controllers, LAA may be overwritten due to controller
3725          * reset from the other port. Set the appropriate LAA in RAR[0]
3726          */
3727         if (e1000e_get_laa_state_82571(hw))
3728                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3729
3730         /* Reset the timer */
3731         if (!test_bit(__E1000_DOWN, &adapter->state))
3732                 mod_timer(&adapter->watchdog_timer,
3733                           round_jiffies(jiffies + 2 * HZ));
3734 }
3735
3736 #define E1000_TX_FLAGS_CSUM             0x00000001
3737 #define E1000_TX_FLAGS_VLAN             0x00000002
3738 #define E1000_TX_FLAGS_TSO              0x00000004
3739 #define E1000_TX_FLAGS_IPV4             0x00000008
3740 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3741 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3742
3743 static int e1000_tso(struct e1000_adapter *adapter,
3744                      struct sk_buff *skb)
3745 {
3746         struct e1000_ring *tx_ring = adapter->tx_ring;
3747         struct e1000_context_desc *context_desc;
3748         struct e1000_buffer *buffer_info;
3749         unsigned int i;
3750         u32 cmd_length = 0;
3751         u16 ipcse = 0, tucse, mss;
3752         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3753         int err;
3754
3755         if (!skb_is_gso(skb))
3756                 return 0;
3757
3758         if (skb_header_cloned(skb)) {
3759                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3760                 if (err)
3761                         return err;
3762         }
3763
3764         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3765         mss = skb_shinfo(skb)->gso_size;
3766         if (skb->protocol == htons(ETH_P_IP)) {
3767                 struct iphdr *iph = ip_hdr(skb);
3768                 iph->tot_len = 0;
3769                 iph->check = 0;
3770                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
3771                                                          0, IPPROTO_TCP, 0);
3772                 cmd_length = E1000_TXD_CMD_IP;
3773                 ipcse = skb_transport_offset(skb) - 1;
3774         } else if (skb_is_gso_v6(skb)) {
3775                 ipv6_hdr(skb)->payload_len = 0;
3776                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3777                                                        &ipv6_hdr(skb)->daddr,
3778                                                        0, IPPROTO_TCP, 0);
3779                 ipcse = 0;
3780         }
3781         ipcss = skb_network_offset(skb);
3782         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3783         tucss = skb_transport_offset(skb);
3784         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3785         tucse = 0;
3786
3787         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3788                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3789
3790         i = tx_ring->next_to_use;
3791         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3792         buffer_info = &tx_ring->buffer_info[i];
3793
3794         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3795         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3796         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3797         context_desc->upper_setup.tcp_fields.tucss = tucss;
3798         context_desc->upper_setup.tcp_fields.tucso = tucso;
3799         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3800         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3801         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3802         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3803
3804         buffer_info->time_stamp = jiffies;
3805         buffer_info->next_to_watch = i;
3806
3807         i++;
3808         if (i == tx_ring->count)
3809                 i = 0;
3810         tx_ring->next_to_use = i;
3811
3812         return 1;
3813 }
3814
3815 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3816 {
3817         struct e1000_ring *tx_ring = adapter->tx_ring;
3818         struct e1000_context_desc *context_desc;
3819         struct e1000_buffer *buffer_info;
3820         unsigned int i;
3821         u8 css;
3822         u32 cmd_len = E1000_TXD_CMD_DEXT;
3823         __be16 protocol;
3824
3825         if (skb->ip_summed != CHECKSUM_PARTIAL)
3826                 return 0;
3827
3828         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
3829                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
3830         else
3831                 protocol = skb->protocol;
3832
3833         switch (protocol) {
3834         case cpu_to_be16(ETH_P_IP):
3835                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3836                         cmd_len |= E1000_TXD_CMD_TCP;
3837                 break;
3838         case cpu_to_be16(ETH_P_IPV6):
3839                 /* XXX not handling all IPV6 headers */
3840                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3841                         cmd_len |= E1000_TXD_CMD_TCP;
3842                 break;
3843         default:
3844                 if (unlikely(net_ratelimit()))
3845                         e_warn("checksum_partial proto=%x!\n",
3846                                be16_to_cpu(protocol));
3847                 break;
3848         }
3849
3850         css = skb_transport_offset(skb);
3851
3852         i = tx_ring->next_to_use;
3853         buffer_info = &tx_ring->buffer_info[i];
3854         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3855
3856         context_desc->lower_setup.ip_config = 0;
3857         context_desc->upper_setup.tcp_fields.tucss = css;
3858         context_desc->upper_setup.tcp_fields.tucso =
3859                                 css + skb->csum_offset;
3860         context_desc->upper_setup.tcp_fields.tucse = 0;
3861         context_desc->tcp_seg_setup.data = 0;
3862         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
3863
3864         buffer_info->time_stamp = jiffies;
3865         buffer_info->next_to_watch = i;
3866
3867         i++;
3868         if (i == tx_ring->count)
3869                 i = 0;
3870         tx_ring->next_to_use = i;
3871
3872         return 1;
3873 }
3874
3875 #define E1000_MAX_PER_TXD       8192
3876 #define E1000_MAX_TXD_PWR       12
3877
3878 static int e1000_tx_map(struct e1000_adapter *adapter,
3879                         struct sk_buff *skb, unsigned int first,
3880                         unsigned int max_per_txd, unsigned int nr_frags,
3881                         unsigned int mss)
3882 {
3883         struct e1000_ring *tx_ring = adapter->tx_ring;
3884         struct pci_dev *pdev = adapter->pdev;
3885         struct e1000_buffer *buffer_info;
3886         unsigned int len = skb_headlen(skb);
3887         unsigned int offset = 0, size, count = 0, i;
3888         unsigned int f;
3889
3890         i = tx_ring->next_to_use;
3891
3892         while (len) {
3893                 buffer_info = &tx_ring->buffer_info[i];
3894                 size = min(len, max_per_txd);
3895
3896                 buffer_info->length = size;
3897                 buffer_info->time_stamp = jiffies;
3898                 buffer_info->next_to_watch = i;
3899                 buffer_info->dma = pci_map_single(pdev, skb->data + offset,
3900                                                   size, PCI_DMA_TODEVICE);
3901                 buffer_info->mapped_as_page = false;
3902                 if (pci_dma_mapping_error(pdev, buffer_info->dma))
3903                         goto dma_error;
3904
3905                 len -= size;
3906                 offset += size;
3907                 count++;
3908
3909                 if (len) {
3910                         i++;
3911                         if (i == tx_ring->count)
3912                                 i = 0;
3913                 }
3914         }
3915
3916         for (f = 0; f < nr_frags; f++) {
3917                 struct skb_frag_struct *frag;
3918
3919                 frag = &skb_shinfo(skb)->frags[f];
3920                 len = frag->size;
3921                 offset = frag->page_offset;
3922
3923                 while (len) {
3924                         i++;
3925                         if (i == tx_ring->count)
3926                                 i = 0;
3927
3928                         buffer_info = &tx_ring->buffer_info[i];
3929                         size = min(len, max_per_txd);
3930
3931                         buffer_info->length = size;
3932                         buffer_info->time_stamp = jiffies;
3933                         buffer_info->next_to_watch = i;
3934                         buffer_info->dma = pci_map_page(pdev, frag->page,
3935                                                         offset, size,
3936                                                         PCI_DMA_TODEVICE);
3937                         buffer_info->mapped_as_page = true;
3938                         if (pci_dma_mapping_error(pdev, buffer_info->dma))
3939                                 goto dma_error;
3940
3941                         len -= size;
3942                         offset += size;
3943                         count++;
3944                 }
3945         }
3946
3947         tx_ring->buffer_info[i].skb = skb;
3948         tx_ring->buffer_info[first].next_to_watch = i;
3949
3950         return count;
3951
3952 dma_error:
3953         dev_err(&pdev->dev, "TX DMA map failed\n");
3954         buffer_info->dma = 0;
3955         if (count)
3956                 count--;
3957
3958         while (count--) {
3959                 if (i==0)
3960                         i += tx_ring->count;
3961                 i--;
3962                 buffer_info = &tx_ring->buffer_info[i];
3963                 e1000_put_txbuf(adapter, buffer_info);;
3964         }
3965
3966         return 0;
3967 }
3968
3969 static void e1000_tx_queue(struct e1000_adapter *adapter,
3970                            int tx_flags, int count)
3971 {
3972         struct e1000_ring *tx_ring = adapter->tx_ring;
3973         struct e1000_tx_desc *tx_desc = NULL;
3974         struct e1000_buffer *buffer_info;
3975         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3976         unsigned int i;
3977
3978         if (tx_flags & E1000_TX_FLAGS_TSO) {
3979                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3980                              E1000_TXD_CMD_TSE;
3981                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3982
3983                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3984                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3985         }
3986
3987         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3988                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3989                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3990         }
3991
3992         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3993                 txd_lower |= E1000_TXD_CMD_VLE;
3994                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3995         }
3996
3997         i = tx_ring->next_to_use;
3998
3999         while (count--) {
4000                 buffer_info = &tx_ring->buffer_info[i];
4001                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4002                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4003                 tx_desc->lower.data =
4004                         cpu_to_le32(txd_lower | buffer_info->length);
4005                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4006
4007                 i++;
4008                 if (i == tx_ring->count)
4009                         i = 0;
4010         }
4011
4012         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4013
4014         /*
4015          * Force memory writes to complete before letting h/w
4016          * know there are new descriptors to fetch.  (Only
4017          * applicable for weak-ordered memory model archs,
4018          * such as IA-64).
4019          */
4020         wmb();
4021
4022         tx_ring->next_to_use = i;
4023         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4024         /*
4025          * we need this if more than one processor can write to our tail
4026          * at a time, it synchronizes IO on IA64/Altix systems
4027          */
4028         mmiowb();
4029 }
4030
4031 #define MINIMUM_DHCP_PACKET_SIZE 282
4032 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4033                                     struct sk_buff *skb)
4034 {
4035         struct e1000_hw *hw =  &adapter->hw;
4036         u16 length, offset;
4037
4038         if (vlan_tx_tag_present(skb)) {
4039                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4040                     (adapter->hw.mng_cookie.status &
4041                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4042                         return 0;
4043         }
4044
4045         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4046                 return 0;
4047
4048         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4049                 return 0;
4050
4051         {
4052                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4053                 struct udphdr *udp;
4054
4055                 if (ip->protocol != IPPROTO_UDP)
4056                         return 0;
4057
4058                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4059                 if (ntohs(udp->dest) != 67)
4060                         return 0;
4061
4062                 offset = (u8 *)udp + 8 - skb->data;
4063                 length = skb->len - offset;
4064                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4065         }
4066
4067         return 0;
4068 }
4069
4070 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4071 {
4072         struct e1000_adapter *adapter = netdev_priv(netdev);
4073
4074         netif_stop_queue(netdev);
4075         /*
4076          * Herbert's original patch had:
4077          *  smp_mb__after_netif_stop_queue();
4078          * but since that doesn't exist yet, just open code it.
4079          */
4080         smp_mb();
4081
4082         /*
4083          * We need to check again in a case another CPU has just
4084          * made room available.
4085          */
4086         if (e1000_desc_unused(adapter->tx_ring) < size)
4087                 return -EBUSY;
4088
4089         /* A reprieve! */
4090         netif_start_queue(netdev);
4091         ++adapter->restart_queue;
4092         return 0;
4093 }
4094
4095 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4096 {
4097         struct e1000_adapter *adapter = netdev_priv(netdev);
4098
4099         if (e1000_desc_unused(adapter->tx_ring) >= size)
4100                 return 0;
4101         return __e1000_maybe_stop_tx(netdev, size);
4102 }
4103
4104 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4105 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4106                                     struct net_device *netdev)
4107 {
4108         struct e1000_adapter *adapter = netdev_priv(netdev);
4109         struct e1000_ring *tx_ring = adapter->tx_ring;
4110         unsigned int first;
4111         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4112         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4113         unsigned int tx_flags = 0;
4114         unsigned int len = skb->len - skb->data_len;
4115         unsigned int nr_frags;
4116         unsigned int mss;
4117         int count = 0;
4118         int tso;
4119         unsigned int f;
4120
4121         if (test_bit(__E1000_DOWN, &adapter->state)) {
4122                 dev_kfree_skb_any(skb);
4123                 return NETDEV_TX_OK;
4124         }
4125
4126         if (skb->len <= 0) {
4127                 dev_kfree_skb_any(skb);
4128                 return NETDEV_TX_OK;
4129         }
4130
4131         mss = skb_shinfo(skb)->gso_size;
4132         /*
4133          * The controller does a simple calculation to
4134          * make sure there is enough room in the FIFO before
4135          * initiating the DMA for each buffer.  The calc is:
4136          * 4 = ceil(buffer len/mss).  To make sure we don't
4137          * overrun the FIFO, adjust the max buffer len if mss
4138          * drops.
4139          */
4140         if (mss) {
4141                 u8 hdr_len;
4142                 max_per_txd = min(mss << 2, max_per_txd);
4143                 max_txd_pwr = fls(max_per_txd) - 1;
4144
4145                 /*
4146                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4147                  * points to just header, pull a few bytes of payload from
4148                  * frags into skb->data
4149                  */
4150                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4151                 /*
4152                  * we do this workaround for ES2LAN, but it is un-necessary,
4153                  * avoiding it could save a lot of cycles
4154                  */
4155                 if (skb->data_len && (hdr_len == len)) {
4156                         unsigned int pull_size;
4157
4158                         pull_size = min((unsigned int)4, skb->data_len);
4159                         if (!__pskb_pull_tail(skb, pull_size)) {
4160                                 e_err("__pskb_pull_tail failed.\n");
4161                                 dev_kfree_skb_any(skb);
4162                                 return NETDEV_TX_OK;
4163                         }
4164                         len = skb->len - skb->data_len;
4165                 }
4166         }
4167
4168         /* reserve a descriptor for the offload context */
4169         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4170                 count++;
4171         count++;
4172
4173         count += TXD_USE_COUNT(len, max_txd_pwr);
4174
4175         nr_frags = skb_shinfo(skb)->nr_frags;
4176         for (f = 0; f < nr_frags; f++)
4177                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4178                                        max_txd_pwr);
4179
4180         if (adapter->hw.mac.tx_pkt_filtering)
4181                 e1000_transfer_dhcp_info(adapter, skb);
4182
4183         /*
4184          * need: count + 2 desc gap to keep tail from touching
4185          * head, otherwise try next time
4186          */
4187         if (e1000_maybe_stop_tx(netdev, count + 2))
4188                 return NETDEV_TX_BUSY;
4189
4190         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4191                 tx_flags |= E1000_TX_FLAGS_VLAN;
4192                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4193         }
4194
4195         first = tx_ring->next_to_use;
4196
4197         tso = e1000_tso(adapter, skb);
4198         if (tso < 0) {
4199                 dev_kfree_skb_any(skb);
4200                 return NETDEV_TX_OK;
4201         }
4202
4203         if (tso)
4204                 tx_flags |= E1000_TX_FLAGS_TSO;
4205         else if (e1000_tx_csum(adapter, skb))
4206                 tx_flags |= E1000_TX_FLAGS_CSUM;
4207
4208         /*
4209          * Old method was to assume IPv4 packet by default if TSO was enabled.
4210          * 82571 hardware supports TSO capabilities for IPv6 as well...
4211          * no longer assume, we must.
4212          */
4213         if (skb->protocol == htons(ETH_P_IP))
4214                 tx_flags |= E1000_TX_FLAGS_IPV4;
4215
4216         /* if count is 0 then mapping error has occured */
4217         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4218         if (count) {
4219                 e1000_tx_queue(adapter, tx_flags, count);
4220                 /* Make sure there is space in the ring for the next send. */
4221                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4222
4223         } else {
4224                 dev_kfree_skb_any(skb);
4225                 tx_ring->buffer_info[first].time_stamp = 0;
4226                 tx_ring->next_to_use = first;
4227         }
4228
4229         return NETDEV_TX_OK;
4230 }
4231
4232 /**
4233  * e1000_tx_timeout - Respond to a Tx Hang
4234  * @netdev: network interface device structure
4235  **/
4236 static void e1000_tx_timeout(struct net_device *netdev)
4237 {
4238         struct e1000_adapter *adapter = netdev_priv(netdev);
4239
4240         /* Do the reset outside of interrupt context */
4241         adapter->tx_timeout_count++;
4242         schedule_work(&adapter->reset_task);
4243 }
4244
4245 static void e1000_reset_task(struct work_struct *work)
4246 {
4247         struct e1000_adapter *adapter;
4248         adapter = container_of(work, struct e1000_adapter, reset_task);
4249
4250         e1000e_reinit_locked(adapter);
4251 }
4252
4253 /**
4254  * e1000_get_stats - Get System Network Statistics
4255  * @netdev: network interface device structure
4256  *
4257  * Returns the address of the device statistics structure.
4258  * The statistics are actually updated from the timer callback.
4259  **/
4260 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4261 {
4262         /* only return the current stats */
4263         return &netdev->stats;
4264 }
4265
4266 /**
4267  * e1000_change_mtu - Change the Maximum Transfer Unit
4268  * @netdev: network interface device structure
4269  * @new_mtu: new value for maximum frame size
4270  *
4271  * Returns 0 on success, negative on failure
4272  **/
4273 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4274 {
4275         struct e1000_adapter *adapter = netdev_priv(netdev);
4276         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4277
4278         /* Jumbo frame support */
4279         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4280             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4281                 e_err("Jumbo Frames not supported.\n");
4282                 return -EINVAL;
4283         }
4284
4285         /* Supported frame sizes */
4286         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4287             (max_frame > adapter->max_hw_frame_size)) {
4288                 e_err("Unsupported MTU setting\n");
4289                 return -EINVAL;
4290         }
4291
4292         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4293                 msleep(1);
4294         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
4295         adapter->max_frame_size = max_frame;
4296         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4297         netdev->mtu = new_mtu;
4298         if (netif_running(netdev))
4299                 e1000e_down(adapter);
4300
4301         /*
4302          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4303          * means we reserve 2 more, this pushes us to allocate from the next
4304          * larger slab size.
4305          * i.e. RXBUFFER_2048 --> size-4096 slab
4306          * However with the new *_jumbo_rx* routines, jumbo receives will use
4307          * fragmented skbs
4308          */
4309
4310         if (max_frame <= 2048)
4311                 adapter->rx_buffer_len = 2048;
4312         else
4313                 adapter->rx_buffer_len = 4096;
4314
4315         /* adjust allocation if LPE protects us, and we aren't using SBP */
4316         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4317              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4318                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4319                                          + ETH_FCS_LEN;
4320
4321         if (netif_running(netdev))
4322                 e1000e_up(adapter);
4323         else
4324                 e1000e_reset(adapter);
4325
4326         clear_bit(__E1000_RESETTING, &adapter->state);
4327
4328         return 0;
4329 }
4330
4331 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4332                            int cmd)
4333 {
4334         struct e1000_adapter *adapter = netdev_priv(netdev);
4335         struct mii_ioctl_data *data = if_mii(ifr);
4336
4337         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4338                 return -EOPNOTSUPP;
4339
4340         switch (cmd) {
4341         case SIOCGMIIPHY:
4342                 data->phy_id = adapter->hw.phy.addr;
4343                 break;
4344         case SIOCGMIIREG:
4345                 e1000_phy_read_status(adapter);
4346
4347                 switch (data->reg_num & 0x1F) {
4348                 case MII_BMCR:
4349                         data->val_out = adapter->phy_regs.bmcr;
4350                         break;
4351                 case MII_BMSR:
4352                         data->val_out = adapter->phy_regs.bmsr;
4353                         break;
4354                 case MII_PHYSID1:
4355                         data->val_out = (adapter->hw.phy.id >> 16);
4356                         break;
4357                 case MII_PHYSID2:
4358                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4359                         break;
4360                 case MII_ADVERTISE:
4361                         data->val_out = adapter->phy_regs.advertise;
4362                         break;
4363                 case MII_LPA:
4364                         data->val_out = adapter->phy_regs.lpa;
4365                         break;
4366                 case MII_EXPANSION:
4367                         data->val_out = adapter->phy_regs.expansion;
4368                         break;
4369                 case MII_CTRL1000:
4370                         data->val_out = adapter->phy_regs.ctrl1000;
4371                         break;
4372                 case MII_STAT1000:
4373                         data->val_out = adapter->phy_regs.stat1000;
4374                         break;
4375                 case MII_ESTATUS:
4376                         data->val_out = adapter->phy_regs.estatus;
4377                         break;
4378                 default:
4379                         return -EIO;
4380                 }
4381                 break;
4382         case SIOCSMIIREG:
4383         default:
4384                 return -EOPNOTSUPP;
4385         }
4386         return 0;
4387 }
4388
4389 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4390 {
4391         switch (cmd) {
4392         case SIOCGMIIPHY:
4393         case SIOCGMIIREG:
4394         case SIOCSMIIREG:
4395                 return e1000_mii_ioctl(netdev, ifr, cmd);
4396         default:
4397                 return -EOPNOTSUPP;
4398         }
4399 }
4400
4401 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
4402 {
4403         struct e1000_hw *hw = &adapter->hw;
4404         u32 i, mac_reg;
4405         u16 phy_reg;
4406         int retval = 0;
4407
4408         /* copy MAC RARs to PHY RARs */
4409         for (i = 0; i < adapter->hw.mac.rar_entry_count; i++) {
4410                 mac_reg = er32(RAL(i));
4411                 e1e_wphy(hw, BM_RAR_L(i), (u16)(mac_reg & 0xFFFF));
4412                 e1e_wphy(hw, BM_RAR_M(i), (u16)((mac_reg >> 16) & 0xFFFF));
4413                 mac_reg = er32(RAH(i));
4414                 e1e_wphy(hw, BM_RAR_H(i), (u16)(mac_reg & 0xFFFF));
4415                 e1e_wphy(hw, BM_RAR_CTRL(i), (u16)((mac_reg >> 16) & 0xFFFF));
4416         }
4417
4418         /* copy MAC MTA to PHY MTA */
4419         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
4420                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
4421                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
4422                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
4423         }
4424
4425         /* configure PHY Rx Control register */
4426         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
4427         mac_reg = er32(RCTL);
4428         if (mac_reg & E1000_RCTL_UPE)
4429                 phy_reg |= BM_RCTL_UPE;
4430         if (mac_reg & E1000_RCTL_MPE)
4431                 phy_reg |= BM_RCTL_MPE;
4432         phy_reg &= ~(BM_RCTL_MO_MASK);
4433         if (mac_reg & E1000_RCTL_MO_3)
4434                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
4435                                 << BM_RCTL_MO_SHIFT);
4436         if (mac_reg & E1000_RCTL_BAM)
4437                 phy_reg |= BM_RCTL_BAM;
4438         if (mac_reg & E1000_RCTL_PMCF)
4439                 phy_reg |= BM_RCTL_PMCF;
4440         mac_reg = er32(CTRL);
4441         if (mac_reg & E1000_CTRL_RFCE)
4442                 phy_reg |= BM_RCTL_RFCE;
4443         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
4444
4445         /* enable PHY wakeup in MAC register */
4446         ew32(WUFC, wufc);
4447         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
4448
4449         /* configure and enable PHY wakeup in PHY registers */
4450         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
4451         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
4452
4453         /* activate PHY wakeup */
4454         retval = hw->phy.ops.acquire(hw);
4455         if (retval) {
4456                 e_err("Could not acquire PHY\n");
4457                 return retval;
4458         }
4459         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4460                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
4461         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
4462         if (retval) {
4463                 e_err("Could not read PHY page 769\n");
4464                 goto out;
4465         }
4466         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
4467         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
4468         if (retval)
4469                 e_err("Could not set PHY Host Wakeup bit\n");
4470 out:
4471         hw->phy.ops.release(hw);
4472
4473         return retval;
4474 }
4475
4476 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4477 {
4478         struct net_device *netdev = pci_get_drvdata(pdev);
4479         struct e1000_adapter *adapter = netdev_priv(netdev);
4480         struct e1000_hw *hw = &adapter->hw;
4481         u32 ctrl, ctrl_ext, rctl, status;
4482         u32 wufc = adapter->wol;
4483         int retval = 0;
4484
4485         netif_device_detach(netdev);
4486
4487         if (netif_running(netdev)) {
4488                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4489                 e1000e_down(adapter);
4490                 e1000_free_irq(adapter);
4491         }
4492         e1000e_reset_interrupt_capability(adapter);
4493
4494         retval = pci_save_state(pdev);
4495         if (retval)
4496                 return retval;
4497
4498         status = er32(STATUS);
4499         if (status & E1000_STATUS_LU)
4500                 wufc &= ~E1000_WUFC_LNKC;
4501
4502         if (wufc) {
4503                 e1000_setup_rctl(adapter);
4504                 e1000_set_multi(netdev);
4505
4506                 /* turn on all-multi mode if wake on multicast is enabled */
4507                 if (wufc & E1000_WUFC_MC) {
4508                         rctl = er32(RCTL);
4509                         rctl |= E1000_RCTL_MPE;
4510                         ew32(RCTL, rctl);
4511                 }
4512
4513                 ctrl = er32(CTRL);
4514                 /* advertise wake from D3Cold */
4515                 #define E1000_CTRL_ADVD3WUC 0x00100000
4516                 /* phy power management enable */
4517                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4518                 ctrl |= E1000_CTRL_ADVD3WUC;
4519                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
4520                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
4521                 ew32(CTRL, ctrl);
4522
4523                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4524                     adapter->hw.phy.media_type ==
4525                     e1000_media_type_internal_serdes) {
4526                         /* keep the laser running in D3 */
4527                         ctrl_ext = er32(CTRL_EXT);
4528                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
4529                         ew32(CTRL_EXT, ctrl_ext);
4530                 }
4531
4532                 if (adapter->flags & FLAG_IS_ICH)
4533                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4534
4535                 /* Allow time for pending master requests to run */
4536                 e1000e_disable_pcie_master(&adapter->hw);
4537
4538                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4539                         /* enable wakeup by the PHY */
4540                         retval = e1000_init_phy_wakeup(adapter, wufc);
4541                         if (retval)
4542                                 return retval;
4543                 } else {
4544                         /* enable wakeup by the MAC */
4545                         ew32(WUFC, wufc);
4546                         ew32(WUC, E1000_WUC_PME_EN);
4547                 }
4548         } else {
4549                 ew32(WUC, 0);
4550                 ew32(WUFC, 0);
4551         }
4552
4553         *enable_wake = !!wufc;
4554
4555         /* make sure adapter isn't asleep if manageability is enabled */
4556         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
4557             (hw->mac.ops.check_mng_mode(hw)))
4558                 *enable_wake = true;
4559
4560         if (adapter->hw.phy.type == e1000_phy_igp_3)
4561                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4562
4563         /*
4564          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4565          * would have already happened in close and is redundant.
4566          */
4567         e1000_release_hw_control(adapter);
4568
4569         pci_disable_device(pdev);
4570
4571         return 0;
4572 }
4573
4574 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
4575 {
4576         if (sleep && wake) {
4577                 pci_prepare_to_sleep(pdev);
4578                 return;
4579         }
4580
4581         pci_wake_from_d3(pdev, wake);
4582         pci_set_power_state(pdev, PCI_D3hot);
4583 }
4584
4585 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
4586                                     bool wake)
4587 {
4588         struct net_device *netdev = pci_get_drvdata(pdev);
4589         struct e1000_adapter *adapter = netdev_priv(netdev);
4590
4591         /*
4592          * The pci-e switch on some quad port adapters will report a
4593          * correctable error when the MAC transitions from D0 to D3.  To
4594          * prevent this we need to mask off the correctable errors on the
4595          * downstream port of the pci-e switch.
4596          */
4597         if (adapter->flags & FLAG_IS_QUAD_PORT) {
4598                 struct pci_dev *us_dev = pdev->bus->self;
4599                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
4600                 u16 devctl;
4601
4602                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
4603                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
4604                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
4605
4606                 e1000_power_off(pdev, sleep, wake);
4607
4608                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
4609         } else {
4610                 e1000_power_off(pdev, sleep, wake);
4611         }
4612 }
4613
4614 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4615 {
4616         int pos;
4617         u16 val;
4618
4619         /*
4620          * 82573 workaround - disable L1 ASPM on mobile chipsets
4621          *
4622          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4623          * resulting in lost data or garbage information on the pci-e link
4624          * level. This could result in (false) bad EEPROM checksum errors,
4625          * long ping times (up to 2s) or even a system freeze/hang.
4626          *
4627          * Unfortunately this feature saves about 1W power consumption when
4628          * active.
4629          */
4630         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4631         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4632         if (val & 0x2) {
4633                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4634                 val &= ~0x2;
4635                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4636         }
4637 }
4638
4639 #ifdef CONFIG_PM
4640 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4641 {
4642         int retval;
4643         bool wake;
4644
4645         retval = __e1000_shutdown(pdev, &wake);
4646         if (!retval)
4647                 e1000_complete_shutdown(pdev, true, wake);
4648
4649         return retval;
4650 }
4651
4652 static int e1000_resume(struct pci_dev *pdev)
4653 {
4654         struct net_device *netdev = pci_get_drvdata(pdev);
4655         struct e1000_adapter *adapter = netdev_priv(netdev);
4656         struct e1000_hw *hw = &adapter->hw;
4657         u32 err;
4658
4659         pci_set_power_state(pdev, PCI_D0);
4660         pci_restore_state(pdev);
4661         pci_save_state(pdev);
4662         e1000e_disable_l1aspm(pdev);
4663
4664         err = pci_enable_device_mem(pdev);
4665         if (err) {
4666                 dev_err(&pdev->dev,
4667                         "Cannot enable PCI device from suspend\n");
4668                 return err;
4669         }
4670
4671         pci_set_master(pdev);
4672
4673         pci_enable_wake(pdev, PCI_D3hot, 0);
4674         pci_enable_wake(pdev, PCI_D3cold, 0);
4675
4676         e1000e_set_interrupt_capability(adapter);
4677         if (netif_running(netdev)) {
4678                 err = e1000_request_irq(adapter);
4679                 if (err)
4680                         return err;
4681         }
4682
4683         e1000e_power_up_phy(adapter);
4684
4685         /* report the system wakeup cause from S3/S4 */
4686         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4687                 u16 phy_data;
4688
4689                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
4690                 if (phy_data) {
4691                         e_info("PHY Wakeup cause - %s\n",
4692                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
4693                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
4694                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
4695                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
4696                                 phy_data & E1000_WUS_LNKC ? "Link Status "
4697                                 " Change" : "other");
4698                 }
4699                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
4700         } else {
4701                 u32 wus = er32(WUS);
4702                 if (wus) {
4703                         e_info("MAC Wakeup cause - %s\n",
4704                                 wus & E1000_WUS_EX ? "Unicast Packet" :
4705                                 wus & E1000_WUS_MC ? "Multicast Packet" :
4706                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
4707                                 wus & E1000_WUS_MAG ? "Magic Packet" :
4708                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
4709                                 "other");
4710                 }
4711                 ew32(WUS, ~0);
4712         }
4713
4714         e1000e_reset(adapter);
4715
4716         e1000_init_manageability(adapter);
4717
4718         if (netif_running(netdev))
4719                 e1000e_up(adapter);
4720
4721         netif_device_attach(netdev);
4722
4723         /*
4724          * If the controller has AMT, do not set DRV_LOAD until the interface
4725          * is up.  For all other cases, let the f/w know that the h/w is now
4726          * under the control of the driver.
4727          */
4728         if (!(adapter->flags & FLAG_HAS_AMT))
4729                 e1000_get_hw_control(adapter);
4730
4731         return 0;
4732 }
4733 #endif
4734
4735 static void e1000_shutdown(struct pci_dev *pdev)
4736 {
4737         bool wake = false;
4738
4739         __e1000_shutdown(pdev, &wake);
4740
4741         if (system_state == SYSTEM_POWER_OFF)
4742                 e1000_complete_shutdown(pdev, false, wake);
4743 }
4744
4745 #ifdef CONFIG_NET_POLL_CONTROLLER
4746 /*
4747  * Polling 'interrupt' - used by things like netconsole to send skbs
4748  * without having to re-enable interrupts. It's not called while
4749  * the interrupt routine is executing.
4750  */
4751 static void e1000_netpoll(struct net_device *netdev)
4752 {
4753         struct e1000_adapter *adapter = netdev_priv(netdev);
4754
4755         disable_irq(adapter->pdev->irq);
4756         e1000_intr(adapter->pdev->irq, netdev);
4757
4758         enable_irq(adapter->pdev->irq);
4759 }
4760 #endif
4761
4762 /**
4763  * e1000_io_error_detected - called when PCI error is detected
4764  * @pdev: Pointer to PCI device
4765  * @state: The current pci connection state
4766  *
4767  * This function is called after a PCI bus error affecting
4768  * this device has been detected.
4769  */
4770 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4771                                                 pci_channel_state_t state)
4772 {
4773         struct net_device *netdev = pci_get_drvdata(pdev);
4774         struct e1000_adapter *adapter = netdev_priv(netdev);
4775
4776         netif_device_detach(netdev);
4777
4778         if (state == pci_channel_io_perm_failure)
4779                 return PCI_ERS_RESULT_DISCONNECT;
4780
4781         if (netif_running(netdev))
4782                 e1000e_down(adapter);
4783         pci_disable_device(pdev);
4784
4785         /* Request a slot slot reset. */
4786         return PCI_ERS_RESULT_NEED_RESET;
4787 }
4788
4789 /**
4790  * e1000_io_slot_reset - called after the pci bus has been reset.
4791  * @pdev: Pointer to PCI device
4792  *
4793  * Restart the card from scratch, as if from a cold-boot. Implementation
4794  * resembles the first-half of the e1000_resume routine.
4795  */
4796 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4797 {
4798         struct net_device *netdev = pci_get_drvdata(pdev);
4799         struct e1000_adapter *adapter = netdev_priv(netdev);
4800         struct e1000_hw *hw = &adapter->hw;
4801         int err;
4802         pci_ers_result_t result;
4803
4804         e1000e_disable_l1aspm(pdev);
4805         err = pci_enable_device_mem(pdev);
4806         if (err) {
4807                 dev_err(&pdev->dev,
4808                         "Cannot re-enable PCI device after reset.\n");
4809                 result = PCI_ERS_RESULT_DISCONNECT;
4810         } else {
4811                 pci_set_master(pdev);
4812                 pci_restore_state(pdev);
4813                 pci_save_state(pdev);
4814
4815                 pci_enable_wake(pdev, PCI_D3hot, 0);
4816                 pci_enable_wake(pdev, PCI_D3cold, 0);
4817
4818                 e1000e_reset(adapter);
4819                 ew32(WUS, ~0);
4820                 result = PCI_ERS_RESULT_RECOVERED;
4821         }
4822
4823         pci_cleanup_aer_uncorrect_error_status(pdev);
4824
4825         return result;
4826 }
4827
4828 /**
4829  * e1000_io_resume - called when traffic can start flowing again.
4830  * @pdev: Pointer to PCI device
4831  *
4832  * This callback is called when the error recovery driver tells us that
4833  * its OK to resume normal operation. Implementation resembles the
4834  * second-half of the e1000_resume routine.
4835  */
4836 static void e1000_io_resume(struct pci_dev *pdev)
4837 {
4838         struct net_device *netdev = pci_get_drvdata(pdev);
4839         struct e1000_adapter *adapter = netdev_priv(netdev);
4840
4841         e1000_init_manageability(adapter);
4842
4843         if (netif_running(netdev)) {
4844                 if (e1000e_up(adapter)) {
4845                         dev_err(&pdev->dev,
4846                                 "can't bring device back up after reset\n");
4847                         return;
4848                 }
4849         }
4850
4851         netif_device_attach(netdev);
4852
4853         /*
4854          * If the controller has AMT, do not set DRV_LOAD until the interface
4855          * is up.  For all other cases, let the f/w know that the h/w is now
4856          * under the control of the driver.
4857          */
4858         if (!(adapter->flags & FLAG_HAS_AMT))
4859                 e1000_get_hw_control(adapter);
4860
4861 }
4862
4863 static void e1000_print_device_info(struct e1000_adapter *adapter)
4864 {
4865         struct e1000_hw *hw = &adapter->hw;
4866         struct net_device *netdev = adapter->netdev;
4867         u32 pba_num;
4868
4869         /* print bus type/speed/width info */
4870         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
4871                /* bus width */
4872                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4873                 "Width x1"),
4874                /* MAC address */
4875                netdev->dev_addr);
4876         e_info("Intel(R) PRO/%s Network Connection\n",
4877                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4878         e1000e_read_pba_num(hw, &pba_num);
4879         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4880                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4881 }
4882
4883 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4884 {
4885         struct e1000_hw *hw = &adapter->hw;
4886         int ret_val;
4887         u16 buf = 0;
4888
4889         if (hw->mac.type != e1000_82573)
4890                 return;
4891
4892         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4893         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
4894                 /* Deep Smart Power Down (DSPD) */
4895                 dev_warn(&adapter->pdev->dev,
4896                          "Warning: detected DSPD enabled in EEPROM\n");
4897         }
4898
4899         ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4900         if (!ret_val && (le16_to_cpu(buf) & (3 << 2))) {
4901                 /* ASPM enable */
4902                 dev_warn(&adapter->pdev->dev,
4903                          "Warning: detected ASPM enabled in EEPROM\n");
4904         }
4905 }
4906
4907 static const struct net_device_ops e1000e_netdev_ops = {
4908         .ndo_open               = e1000_open,
4909         .ndo_stop               = e1000_close,
4910         .ndo_start_xmit         = e1000_xmit_frame,
4911         .ndo_get_stats          = e1000_get_stats,
4912         .ndo_set_multicast_list = e1000_set_multi,
4913         .ndo_set_mac_address    = e1000_set_mac,
4914         .ndo_change_mtu         = e1000_change_mtu,
4915         .ndo_do_ioctl           = e1000_ioctl,
4916         .ndo_tx_timeout         = e1000_tx_timeout,
4917         .ndo_validate_addr      = eth_validate_addr,
4918
4919         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
4920         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
4921         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
4922 #ifdef CONFIG_NET_POLL_CONTROLLER
4923         .ndo_poll_controller    = e1000_netpoll,
4924 #endif
4925 };
4926
4927 /**
4928  * e1000_probe - Device Initialization Routine
4929  * @pdev: PCI device information struct
4930  * @ent: entry in e1000_pci_tbl
4931  *
4932  * Returns 0 on success, negative on failure
4933  *
4934  * e1000_probe initializes an adapter identified by a pci_dev structure.
4935  * The OS initialization, configuring of the adapter private structure,
4936  * and a hardware reset occur.
4937  **/
4938 static int __devinit e1000_probe(struct pci_dev *pdev,
4939                                  const struct pci_device_id *ent)
4940 {
4941         struct net_device *netdev;
4942         struct e1000_adapter *adapter;
4943         struct e1000_hw *hw;
4944         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4945         resource_size_t mmio_start, mmio_len;
4946         resource_size_t flash_start, flash_len;
4947
4948         static int cards_found;
4949         int i, err, pci_using_dac;
4950         u16 eeprom_data = 0;
4951         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4952
4953         e1000e_disable_l1aspm(pdev);
4954
4955         err = pci_enable_device_mem(pdev);
4956         if (err)
4957                 return err;
4958
4959         pci_using_dac = 0;
4960         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
4961         if (!err) {
4962                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
4963                 if (!err)
4964                         pci_using_dac = 1;
4965         } else {
4966                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
4967                 if (err) {
4968                         err = pci_set_consistent_dma_mask(pdev,
4969                                                           DMA_BIT_MASK(32));
4970                         if (err) {
4971                                 dev_err(&pdev->dev, "No usable DMA "
4972                                         "configuration, aborting\n");
4973                                 goto err_dma;
4974                         }
4975                 }
4976         }
4977
4978         err = pci_request_selected_regions_exclusive(pdev,
4979                                           pci_select_bars(pdev, IORESOURCE_MEM),
4980                                           e1000e_driver_name);
4981         if (err)
4982                 goto err_pci_reg;
4983
4984         /* AER (Advanced Error Reporting) hooks */
4985         pci_enable_pcie_error_reporting(pdev);
4986
4987         pci_set_master(pdev);
4988         /* PCI config space info */
4989         err = pci_save_state(pdev);
4990         if (err)
4991                 goto err_alloc_etherdev;
4992
4993         err = -ENOMEM;
4994         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4995         if (!netdev)
4996                 goto err_alloc_etherdev;
4997
4998         SET_NETDEV_DEV(netdev, &pdev->dev);
4999
5000         pci_set_drvdata(pdev, netdev);
5001         adapter = netdev_priv(netdev);
5002         hw = &adapter->hw;
5003         adapter->netdev = netdev;
5004         adapter->pdev = pdev;
5005         adapter->ei = ei;
5006         adapter->pba = ei->pba;
5007         adapter->flags = ei->flags;
5008         adapter->flags2 = ei->flags2;
5009         adapter->hw.adapter = adapter;
5010         adapter->hw.mac.type = ei->mac;
5011         adapter->max_hw_frame_size = ei->max_hw_frame_size;
5012         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5013
5014         mmio_start = pci_resource_start(pdev, 0);
5015         mmio_len = pci_resource_len(pdev, 0);
5016
5017         err = -EIO;
5018         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5019         if (!adapter->hw.hw_addr)
5020                 goto err_ioremap;
5021
5022         if ((adapter->flags & FLAG_HAS_FLASH) &&
5023             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5024                 flash_start = pci_resource_start(pdev, 1);
5025                 flash_len = pci_resource_len(pdev, 1);
5026                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5027                 if (!adapter->hw.flash_address)
5028                         goto err_flashmap;
5029         }
5030
5031         /* construct the net_device struct */
5032         netdev->netdev_ops              = &e1000e_netdev_ops;
5033         e1000e_set_ethtool_ops(netdev);
5034         netdev->watchdog_timeo          = 5 * HZ;
5035         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5036         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5037
5038         netdev->mem_start = mmio_start;
5039         netdev->mem_end = mmio_start + mmio_len;
5040
5041         adapter->bd_number = cards_found++;
5042
5043         e1000e_check_options(adapter);
5044
5045         /* setup adapter struct */
5046         err = e1000_sw_init(adapter);
5047         if (err)
5048                 goto err_sw_init;
5049
5050         err = -EIO;
5051
5052         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5053         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5054         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5055
5056         err = ei->get_variants(adapter);
5057         if (err)
5058                 goto err_hw_init;
5059
5060         if ((adapter->flags & FLAG_IS_ICH) &&
5061             (adapter->flags & FLAG_READ_ONLY_NVM))
5062                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5063
5064         hw->mac.ops.get_bus_info(&adapter->hw);
5065
5066         adapter->hw.phy.autoneg_wait_to_complete = 0;
5067
5068         /* Copper options */
5069         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5070                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5071                 adapter->hw.phy.disable_polarity_correction = 0;
5072                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5073         }
5074
5075         if (e1000_check_reset_block(&adapter->hw))
5076                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5077
5078         netdev->features = NETIF_F_SG |
5079                            NETIF_F_HW_CSUM |
5080                            NETIF_F_HW_VLAN_TX |
5081                            NETIF_F_HW_VLAN_RX;
5082
5083         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5084                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5085
5086         netdev->features |= NETIF_F_TSO;
5087         netdev->features |= NETIF_F_TSO6;
5088
5089         netdev->vlan_features |= NETIF_F_TSO;
5090         netdev->vlan_features |= NETIF_F_TSO6;
5091         netdev->vlan_features |= NETIF_F_HW_CSUM;
5092         netdev->vlan_features |= NETIF_F_SG;
5093
5094         if (pci_using_dac)
5095                 netdev->features |= NETIF_F_HIGHDMA;
5096
5097         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5098                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5099
5100         /*
5101          * before reading the NVM, reset the controller to
5102          * put the device in a known good starting state
5103          */
5104         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5105
5106         /*
5107          * systems with ASPM and others may see the checksum fail on the first
5108          * attempt. Let's give it a few tries
5109          */
5110         for (i = 0;; i++) {
5111                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5112                         break;
5113                 if (i == 2) {
5114                         e_err("The NVM Checksum Is Not Valid\n");
5115                         err = -EIO;
5116                         goto err_eeprom;
5117                 }
5118         }
5119
5120         e1000_eeprom_checks(adapter);
5121
5122         /* copy the MAC address */
5123         if (e1000e_read_mac_addr(&adapter->hw))
5124                 e_err("NVM Read Error while reading MAC address\n");
5125
5126         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5127         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5128
5129         if (!is_valid_ether_addr(netdev->perm_addr)) {
5130                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5131                 err = -EIO;
5132                 goto err_eeprom;
5133         }
5134
5135         init_timer(&adapter->watchdog_timer);
5136         adapter->watchdog_timer.function = &e1000_watchdog;
5137         adapter->watchdog_timer.data = (unsigned long) adapter;
5138
5139         init_timer(&adapter->phy_info_timer);
5140         adapter->phy_info_timer.function = &e1000_update_phy_info;
5141         adapter->phy_info_timer.data = (unsigned long) adapter;
5142
5143         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5144         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5145         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5146         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5147         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5148
5149         /* Initialize link parameters. User can change them with ethtool */
5150         adapter->hw.mac.autoneg = 1;
5151         adapter->fc_autoneg = 1;
5152         adapter->hw.fc.requested_mode = e1000_fc_default;
5153         adapter->hw.fc.current_mode = e1000_fc_default;
5154         adapter->hw.phy.autoneg_advertised = 0x2f;
5155
5156         /* ring size defaults */
5157         adapter->rx_ring->count = 256;
5158         adapter->tx_ring->count = 256;
5159
5160         /*
5161          * Initial Wake on LAN setting - If APM wake is enabled in
5162          * the EEPROM, enable the ACPI Magic Packet filter
5163          */
5164         if (adapter->flags & FLAG_APME_IN_WUC) {
5165                 /* APME bit in EEPROM is mapped to WUC.APME */
5166                 eeprom_data = er32(WUC);
5167                 eeprom_apme_mask = E1000_WUC_APME;
5168                 if (eeprom_data & E1000_WUC_PHY_WAKE)
5169                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5170         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5171                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5172                     (adapter->hw.bus.func == 1))
5173                         e1000_read_nvm(&adapter->hw,
5174                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5175                 else
5176                         e1000_read_nvm(&adapter->hw,
5177                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5178         }
5179
5180         /* fetch WoL from EEPROM */
5181         if (eeprom_data & eeprom_apme_mask)
5182                 adapter->eeprom_wol |= E1000_WUFC_MAG;
5183
5184         /*
5185          * now that we have the eeprom settings, apply the special cases
5186          * where the eeprom may be wrong or the board simply won't support
5187          * wake on lan on a particular port
5188          */
5189         if (!(adapter->flags & FLAG_HAS_WOL))
5190                 adapter->eeprom_wol = 0;
5191
5192         /* initialize the wol settings based on the eeprom settings */
5193         adapter->wol = adapter->eeprom_wol;
5194         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5195
5196         /* save off EEPROM version number */
5197         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5198
5199         /* reset the hardware with the new settings */
5200         e1000e_reset(adapter);
5201
5202         /*
5203          * If the controller has AMT, do not set DRV_LOAD until the interface
5204          * is up.  For all other cases, let the f/w know that the h/w is now
5205          * under the control of the driver.
5206          */
5207         if (!(adapter->flags & FLAG_HAS_AMT))
5208                 e1000_get_hw_control(adapter);
5209
5210         strcpy(netdev->name, "eth%d");
5211         err = register_netdev(netdev);
5212         if (err)
5213                 goto err_register;
5214
5215         /* carrier off reporting is important to ethtool even BEFORE open */
5216         netif_carrier_off(netdev);
5217
5218         e1000_print_device_info(adapter);
5219
5220         return 0;
5221
5222 err_register:
5223         if (!(adapter->flags & FLAG_HAS_AMT))
5224                 e1000_release_hw_control(adapter);
5225 err_eeprom:
5226         if (!e1000_check_reset_block(&adapter->hw))
5227                 e1000_phy_hw_reset(&adapter->hw);
5228 err_hw_init:
5229
5230         kfree(adapter->tx_ring);
5231         kfree(adapter->rx_ring);
5232 err_sw_init:
5233         if (adapter->hw.flash_address)
5234                 iounmap(adapter->hw.flash_address);
5235         e1000e_reset_interrupt_capability(adapter);
5236 err_flashmap:
5237         iounmap(adapter->hw.hw_addr);
5238 err_ioremap:
5239         free_netdev(netdev);
5240 err_alloc_etherdev:
5241         pci_release_selected_regions(pdev,
5242                                      pci_select_bars(pdev, IORESOURCE_MEM));
5243 err_pci_reg:
5244 err_dma:
5245         pci_disable_device(pdev);
5246         return err;
5247 }
5248
5249 /**
5250  * e1000_remove - Device Removal Routine
5251  * @pdev: PCI device information struct
5252  *
5253  * e1000_remove is called by the PCI subsystem to alert the driver
5254  * that it should release a PCI device.  The could be caused by a
5255  * Hot-Plug event, or because the driver is going to be removed from
5256  * memory.
5257  **/
5258 static void __devexit e1000_remove(struct pci_dev *pdev)
5259 {
5260         struct net_device *netdev = pci_get_drvdata(pdev);
5261         struct e1000_adapter *adapter = netdev_priv(netdev);
5262
5263         /*
5264          * flush_scheduled work may reschedule our watchdog task, so
5265          * explicitly disable watchdog tasks from being rescheduled
5266          */
5267         set_bit(__E1000_DOWN, &adapter->state);
5268         del_timer_sync(&adapter->watchdog_timer);
5269         del_timer_sync(&adapter->phy_info_timer);
5270
5271         cancel_work_sync(&adapter->reset_task);
5272         cancel_work_sync(&adapter->watchdog_task);
5273         cancel_work_sync(&adapter->downshift_task);
5274         cancel_work_sync(&adapter->update_phy_task);
5275         cancel_work_sync(&adapter->print_hang_task);
5276         flush_scheduled_work();
5277
5278         if (!(netdev->flags & IFF_UP))
5279                 e1000_power_down_phy(adapter);
5280
5281         unregister_netdev(netdev);
5282
5283         /*
5284          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5285          * would have already happened in close and is redundant.
5286          */
5287         e1000_release_hw_control(adapter);
5288
5289         e1000e_reset_interrupt_capability(adapter);
5290         kfree(adapter->tx_ring);
5291         kfree(adapter->rx_ring);
5292
5293         iounmap(adapter->hw.hw_addr);
5294         if (adapter->hw.flash_address)
5295                 iounmap(adapter->hw.flash_address);
5296         pci_release_selected_regions(pdev,
5297                                      pci_select_bars(pdev, IORESOURCE_MEM));
5298
5299         free_netdev(netdev);
5300
5301         /* AER disable */
5302         pci_disable_pcie_error_reporting(pdev);
5303
5304         pci_disable_device(pdev);
5305 }
5306
5307 /* PCI Error Recovery (ERS) */
5308 static struct pci_error_handlers e1000_err_handler = {
5309         .error_detected = e1000_io_error_detected,
5310         .slot_reset = e1000_io_slot_reset,
5311         .resume = e1000_io_resume,
5312 };
5313
5314 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
5315         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5316         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5317         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5318         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5319         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5320         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5321         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5322         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5323         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5324
5325         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5326         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5327         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5328         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5329
5330         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5331         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5332         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5333
5334         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5335         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
5336         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
5337
5338         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5339           board_80003es2lan },
5340         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5341           board_80003es2lan },
5342         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5343           board_80003es2lan },
5344         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5345           board_80003es2lan },
5346
5347         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5348         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5349         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5350         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5351         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5352         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5353         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5354         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
5355
5356         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5357         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5358         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5359         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5360         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5361         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5362         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5363         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5364         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5365
5366         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5367         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5368         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5369
5370         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5371         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5372
5373         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
5374         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
5375         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
5376         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
5377
5378         { }     /* terminate list */
5379 };
5380 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5381
5382 /* PCI Device API Driver */
5383 static struct pci_driver e1000_driver = {
5384         .name     = e1000e_driver_name,
5385         .id_table = e1000_pci_tbl,
5386         .probe    = e1000_probe,
5387         .remove   = __devexit_p(e1000_remove),
5388 #ifdef CONFIG_PM
5389         /* Power Management Hooks */
5390         .suspend  = e1000_suspend,
5391         .resume   = e1000_resume,
5392 #endif
5393         .shutdown = e1000_shutdown,
5394         .err_handler = &e1000_err_handler
5395 };
5396
5397 /**
5398  * e1000_init_module - Driver Registration Routine
5399  *
5400  * e1000_init_module is the first routine called when the driver is
5401  * loaded. All it does is register with the PCI subsystem.
5402  **/
5403 static int __init e1000_init_module(void)
5404 {
5405         int ret;
5406         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
5407                e1000e_driver_name, e1000e_driver_version);
5408         printk(KERN_INFO "%s: Copyright (c) 1999 - 2009 Intel Corporation.\n",
5409                e1000e_driver_name);
5410         ret = pci_register_driver(&e1000_driver);
5411
5412         return ret;
5413 }
5414 module_init(e1000_init_module);
5415
5416 /**
5417  * e1000_exit_module - Driver Exit Cleanup Routine
5418  *
5419  * e1000_exit_module is called just before the driver is removed
5420  * from memory.
5421  **/
5422 static void __exit e1000_exit_module(void)
5423 {
5424         pci_unregister_driver(&e1000_driver);
5425 }
5426 module_exit(e1000_exit_module);
5427
5428
5429 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5430 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5431 MODULE_LICENSE("GPL");
5432 MODULE_VERSION(DRV_VERSION);
5433
5434 /* e1000_main.c */