Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[safe/jmp/linux-2.6] / drivers / net / e1000e / netdev.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2009 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #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 256:
2354                 rctl |= E1000_RCTL_SZ_256;
2355                 rctl &= ~E1000_RCTL_BSEX;
2356                 break;
2357         case 512:
2358                 rctl |= E1000_RCTL_SZ_512;
2359                 rctl &= ~E1000_RCTL_BSEX;
2360                 break;
2361         case 1024:
2362                 rctl |= E1000_RCTL_SZ_1024;
2363                 rctl &= ~E1000_RCTL_BSEX;
2364                 break;
2365         case 2048:
2366         default:
2367                 rctl |= E1000_RCTL_SZ_2048;
2368                 rctl &= ~E1000_RCTL_BSEX;
2369                 break;
2370         case 4096:
2371                 rctl |= E1000_RCTL_SZ_4096;
2372                 break;
2373         case 8192:
2374                 rctl |= E1000_RCTL_SZ_8192;
2375                 break;
2376         case 16384:
2377                 rctl |= E1000_RCTL_SZ_16384;
2378                 break;
2379         }
2380
2381         /*
2382          * 82571 and greater support packet-split where the protocol
2383          * header is placed in skb->data and the packet data is
2384          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2385          * In the case of a non-split, skb->data is linearly filled,
2386          * followed by the page buffers.  Therefore, skb->data is
2387          * sized to hold the largest protocol header.
2388          *
2389          * allocations using alloc_page take too long for regular MTU
2390          * so only enable packet split for jumbo frames
2391          *
2392          * Using pages when the page size is greater than 16k wastes
2393          * a lot of memory, since we allocate 3 pages at all times
2394          * per packet.
2395          */
2396         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2397         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2398             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2399                 adapter->rx_ps_pages = pages;
2400         else
2401                 adapter->rx_ps_pages = 0;
2402
2403         if (adapter->rx_ps_pages) {
2404                 /* Configure extra packet-split registers */
2405                 rfctl = er32(RFCTL);
2406                 rfctl |= E1000_RFCTL_EXTEN;
2407                 /*
2408                  * disable packet split support for IPv6 extension headers,
2409                  * because some malformed IPv6 headers can hang the Rx
2410                  */
2411                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2412                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2413
2414                 ew32(RFCTL, rfctl);
2415
2416                 /* Enable Packet split descriptors */
2417                 rctl |= E1000_RCTL_DTYP_PS;
2418
2419                 psrctl |= adapter->rx_ps_bsize0 >>
2420                         E1000_PSRCTL_BSIZE0_SHIFT;
2421
2422                 switch (adapter->rx_ps_pages) {
2423                 case 3:
2424                         psrctl |= PAGE_SIZE <<
2425                                 E1000_PSRCTL_BSIZE3_SHIFT;
2426                 case 2:
2427                         psrctl |= PAGE_SIZE <<
2428                                 E1000_PSRCTL_BSIZE2_SHIFT;
2429                 case 1:
2430                         psrctl |= PAGE_SIZE >>
2431                                 E1000_PSRCTL_BSIZE1_SHIFT;
2432                         break;
2433                 }
2434
2435                 ew32(PSRCTL, psrctl);
2436         }
2437
2438         ew32(RCTL, rctl);
2439         /* just started the receive unit, no need to restart */
2440         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2441 }
2442
2443 /**
2444  * e1000_configure_rx - Configure Receive Unit after Reset
2445  * @adapter: board private structure
2446  *
2447  * Configure the Rx unit of the MAC after a reset.
2448  **/
2449 static void e1000_configure_rx(struct e1000_adapter *adapter)
2450 {
2451         struct e1000_hw *hw = &adapter->hw;
2452         struct e1000_ring *rx_ring = adapter->rx_ring;
2453         u64 rdba;
2454         u32 rdlen, rctl, rxcsum, ctrl_ext;
2455
2456         if (adapter->rx_ps_pages) {
2457                 /* this is a 32 byte descriptor */
2458                 rdlen = rx_ring->count *
2459                         sizeof(union e1000_rx_desc_packet_split);
2460                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2461                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2462         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2463                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2464                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2465                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2466         } else {
2467                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2468                 adapter->clean_rx = e1000_clean_rx_irq;
2469                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2470         }
2471
2472         /* disable receives while setting up the descriptors */
2473         rctl = er32(RCTL);
2474         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2475         e1e_flush();
2476         msleep(10);
2477
2478         /* set the Receive Delay Timer Register */
2479         ew32(RDTR, adapter->rx_int_delay);
2480
2481         /* irq moderation */
2482         ew32(RADV, adapter->rx_abs_int_delay);
2483         if (adapter->itr_setting != 0)
2484                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2485
2486         ctrl_ext = er32(CTRL_EXT);
2487         /* Auto-Mask interrupts upon ICR access */
2488         ctrl_ext |= E1000_CTRL_EXT_IAME;
2489         ew32(IAM, 0xffffffff);
2490         ew32(CTRL_EXT, ctrl_ext);
2491         e1e_flush();
2492
2493         /*
2494          * Setup the HW Rx Head and Tail Descriptor Pointers and
2495          * the Base and Length of the Rx Descriptor Ring
2496          */
2497         rdba = rx_ring->dma;
2498         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2499         ew32(RDBAH, (rdba >> 32));
2500         ew32(RDLEN, rdlen);
2501         ew32(RDH, 0);
2502         ew32(RDT, 0);
2503         rx_ring->head = E1000_RDH;
2504         rx_ring->tail = E1000_RDT;
2505
2506         /* Enable Receive Checksum Offload for TCP and UDP */
2507         rxcsum = er32(RXCSUM);
2508         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2509                 rxcsum |= E1000_RXCSUM_TUOFL;
2510
2511                 /*
2512                  * IPv4 payload checksum for UDP fragments must be
2513                  * used in conjunction with packet-split.
2514                  */
2515                 if (adapter->rx_ps_pages)
2516                         rxcsum |= E1000_RXCSUM_IPPCSE;
2517         } else {
2518                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2519                 /* no need to clear IPPCSE as it defaults to 0 */
2520         }
2521         ew32(RXCSUM, rxcsum);
2522
2523         /*
2524          * Enable early receives on supported devices, only takes effect when
2525          * packet size is equal or larger than the specified value (in 8 byte
2526          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2527          */
2528         if (adapter->flags & FLAG_HAS_ERT) {
2529                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2530                         u32 rxdctl = er32(RXDCTL(0));
2531                         ew32(RXDCTL(0), rxdctl | 0x3);
2532                         ew32(ERT, E1000_ERT_2048 | (1 << 13));
2533                         /*
2534                          * With jumbo frames and early-receive enabled,
2535                          * excessive C-state transition latencies result in
2536                          * dropped transactions.
2537                          */
2538                         pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2539                                                   adapter->netdev->name, 55);
2540                 } else {
2541                         pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2542                                                   adapter->netdev->name,
2543                                                   PM_QOS_DEFAULT_VALUE);
2544                 }
2545         }
2546
2547         /* Enable Receives */
2548         ew32(RCTL, rctl);
2549 }
2550
2551 /**
2552  *  e1000_update_mc_addr_list - Update Multicast addresses
2553  *  @hw: pointer to the HW structure
2554  *  @mc_addr_list: array of multicast addresses to program
2555  *  @mc_addr_count: number of multicast addresses to program
2556  *
2557  *  Updates the Multicast Table Array.
2558  *  The caller must have a packed mc_addr_list of multicast addresses.
2559  **/
2560 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2561                                       u32 mc_addr_count)
2562 {
2563         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
2564 }
2565
2566 /**
2567  * e1000_set_multi - Multicast and Promiscuous mode set
2568  * @netdev: network interface device structure
2569  *
2570  * The set_multi entry point is called whenever the multicast address
2571  * list or the network interface flags are updated.  This routine is
2572  * responsible for configuring the hardware for proper multicast,
2573  * promiscuous mode, and all-multi behavior.
2574  **/
2575 static void e1000_set_multi(struct net_device *netdev)
2576 {
2577         struct e1000_adapter *adapter = netdev_priv(netdev);
2578         struct e1000_hw *hw = &adapter->hw;
2579         struct dev_mc_list *mc_ptr;
2580         u8  *mta_list;
2581         u32 rctl;
2582         int i;
2583
2584         /* Check for Promiscuous and All Multicast modes */
2585
2586         rctl = er32(RCTL);
2587
2588         if (netdev->flags & IFF_PROMISC) {
2589                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2590                 rctl &= ~E1000_RCTL_VFE;
2591         } else {
2592                 if (netdev->flags & IFF_ALLMULTI) {
2593                         rctl |= E1000_RCTL_MPE;
2594                         rctl &= ~E1000_RCTL_UPE;
2595                 } else {
2596                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2597                 }
2598                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2599                         rctl |= E1000_RCTL_VFE;
2600         }
2601
2602         ew32(RCTL, rctl);
2603
2604         if (netdev->mc_count) {
2605                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2606                 if (!mta_list)
2607                         return;
2608
2609                 /* prepare a packed array of only addresses. */
2610                 mc_ptr = netdev->mc_list;
2611
2612                 for (i = 0; i < netdev->mc_count; i++) {
2613                         if (!mc_ptr)
2614                                 break;
2615                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2616                                ETH_ALEN);
2617                         mc_ptr = mc_ptr->next;
2618                 }
2619
2620                 e1000_update_mc_addr_list(hw, mta_list, i);
2621                 kfree(mta_list);
2622         } else {
2623                 /*
2624                  * if we're called from probe, we might not have
2625                  * anything to do here, so clear out the list
2626                  */
2627                 e1000_update_mc_addr_list(hw, NULL, 0);
2628         }
2629 }
2630
2631 /**
2632  * e1000_configure - configure the hardware for Rx and Tx
2633  * @adapter: private board structure
2634  **/
2635 static void e1000_configure(struct e1000_adapter *adapter)
2636 {
2637         e1000_set_multi(adapter->netdev);
2638
2639         e1000_restore_vlan(adapter);
2640         e1000_init_manageability(adapter);
2641
2642         e1000_configure_tx(adapter);
2643         e1000_setup_rctl(adapter);
2644         e1000_configure_rx(adapter);
2645         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2646 }
2647
2648 /**
2649  * e1000e_power_up_phy - restore link in case the phy was powered down
2650  * @adapter: address of board private structure
2651  *
2652  * The phy may be powered down to save power and turn off link when the
2653  * driver is unloaded and wake on lan is not enabled (among others)
2654  * *** this routine MUST be followed by a call to e1000e_reset ***
2655  **/
2656 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2657 {
2658         if (adapter->hw.phy.ops.power_up)
2659                 adapter->hw.phy.ops.power_up(&adapter->hw);
2660
2661         adapter->hw.mac.ops.setup_link(&adapter->hw);
2662 }
2663
2664 /**
2665  * e1000_power_down_phy - Power down the PHY
2666  *
2667  * Power down the PHY so no link is implied when interface is down.
2668  * The PHY cannot be powered down if management or WoL is active.
2669  */
2670 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2671 {
2672         /* WoL is enabled */
2673         if (adapter->wol)
2674                 return;
2675
2676         if (adapter->hw.phy.ops.power_down)
2677                 adapter->hw.phy.ops.power_down(&adapter->hw);
2678 }
2679
2680 /**
2681  * e1000e_reset - bring the hardware into a known good state
2682  *
2683  * This function boots the hardware and enables some settings that
2684  * require a configuration cycle of the hardware - those cannot be
2685  * set/changed during runtime. After reset the device needs to be
2686  * properly configured for Rx, Tx etc.
2687  */
2688 void e1000e_reset(struct e1000_adapter *adapter)
2689 {
2690         struct e1000_mac_info *mac = &adapter->hw.mac;
2691         struct e1000_fc_info *fc = &adapter->hw.fc;
2692         struct e1000_hw *hw = &adapter->hw;
2693         u32 tx_space, min_tx_space, min_rx_space;
2694         u32 pba = adapter->pba;
2695         u16 hwm;
2696
2697         /* reset Packet Buffer Allocation to default */
2698         ew32(PBA, pba);
2699
2700         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2701                 /*
2702                  * To maintain wire speed transmits, the Tx FIFO should be
2703                  * large enough to accommodate two full transmit packets,
2704                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2705                  * the Rx FIFO should be large enough to accommodate at least
2706                  * one full receive packet and is similarly rounded up and
2707                  * expressed in KB.
2708                  */
2709                 pba = er32(PBA);
2710                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2711                 tx_space = pba >> 16;
2712                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2713                 pba &= 0xffff;
2714                 /*
2715                  * the Tx fifo also stores 16 bytes of information about the tx
2716                  * but don't include ethernet FCS because hardware appends it
2717                  */
2718                 min_tx_space = (adapter->max_frame_size +
2719                                 sizeof(struct e1000_tx_desc) -
2720                                 ETH_FCS_LEN) * 2;
2721                 min_tx_space = ALIGN(min_tx_space, 1024);
2722                 min_tx_space >>= 10;
2723                 /* software strips receive CRC, so leave room for it */
2724                 min_rx_space = adapter->max_frame_size;
2725                 min_rx_space = ALIGN(min_rx_space, 1024);
2726                 min_rx_space >>= 10;
2727
2728                 /*
2729                  * If current Tx allocation is less than the min Tx FIFO size,
2730                  * and the min Tx FIFO size is less than the current Rx FIFO
2731                  * allocation, take space away from current Rx allocation
2732                  */
2733                 if ((tx_space < min_tx_space) &&
2734                     ((min_tx_space - tx_space) < pba)) {
2735                         pba -= min_tx_space - tx_space;
2736
2737                         /*
2738                          * if short on Rx space, Rx wins and must trump tx
2739                          * adjustment or use Early Receive if available
2740                          */
2741                         if ((pba < min_rx_space) &&
2742                             (!(adapter->flags & FLAG_HAS_ERT)))
2743                                 /* ERT enabled in e1000_configure_rx */
2744                                 pba = min_rx_space;
2745                 }
2746
2747                 ew32(PBA, pba);
2748         }
2749
2750
2751         /*
2752          * flow control settings
2753          *
2754          * The high water mark must be low enough to fit one full frame
2755          * (or the size used for early receive) above it in the Rx FIFO.
2756          * Set it to the lower of:
2757          * - 90% of the Rx FIFO size, and
2758          * - the full Rx FIFO size minus the early receive size (for parts
2759          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2760          * - the full Rx FIFO size minus one full frame
2761          */
2762         if (hw->mac.type == e1000_pchlan) {
2763                 /*
2764                  * Workaround PCH LOM adapter hangs with certain network
2765                  * loads.  If hangs persist, try disabling Tx flow control.
2766                  */
2767                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2768                         fc->high_water = 0x3500;
2769                         fc->low_water  = 0x1500;
2770                 } else {
2771                         fc->high_water = 0x5000;
2772                         fc->low_water  = 0x3000;
2773                 }
2774         } else {
2775                 if ((adapter->flags & FLAG_HAS_ERT) &&
2776                     (adapter->netdev->mtu > ETH_DATA_LEN))
2777                         hwm = min(((pba << 10) * 9 / 10),
2778                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
2779                 else
2780                         hwm = min(((pba << 10) * 9 / 10),
2781                                   ((pba << 10) - adapter->max_frame_size));
2782
2783                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
2784                 fc->low_water = fc->high_water - 8;
2785         }
2786
2787         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2788                 fc->pause_time = 0xFFFF;
2789         else
2790                 fc->pause_time = E1000_FC_PAUSE_TIME;
2791         fc->send_xon = 1;
2792         fc->current_mode = fc->requested_mode;
2793
2794         /* Allow time for pending master requests to run */
2795         mac->ops.reset_hw(hw);
2796
2797         /*
2798          * For parts with AMT enabled, let the firmware know
2799          * that the network interface is in control
2800          */
2801         if (adapter->flags & FLAG_HAS_AMT)
2802                 e1000_get_hw_control(adapter);
2803
2804         ew32(WUC, 0);
2805         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP)
2806                 e1e_wphy(&adapter->hw, BM_WUC, 0);
2807
2808         if (mac->ops.init_hw(hw))
2809                 e_err("Hardware Error\n");
2810
2811         /* additional part of the flow-control workaround above */
2812         if (hw->mac.type == e1000_pchlan)
2813                 ew32(FCRTV_PCH, 0x1000);
2814
2815         e1000_update_mng_vlan(adapter);
2816
2817         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2818         ew32(VET, ETH_P_8021Q);
2819
2820         e1000e_reset_adaptive(hw);
2821         e1000_get_phy_info(hw);
2822
2823         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
2824             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2825                 u16 phy_data = 0;
2826                 /*
2827                  * speed up time to link by disabling smart power down, ignore
2828                  * the return value of this function because there is nothing
2829                  * different we would do if it failed
2830                  */
2831                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2832                 phy_data &= ~IGP02E1000_PM_SPD;
2833                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2834         }
2835 }
2836
2837 int e1000e_up(struct e1000_adapter *adapter)
2838 {
2839         struct e1000_hw *hw = &adapter->hw;
2840
2841         /* DMA latency requirement to workaround early-receive/jumbo issue */
2842         if (adapter->flags & FLAG_HAS_ERT)
2843                 pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY,
2844                                        adapter->netdev->name,
2845                                        PM_QOS_DEFAULT_VALUE);
2846
2847         /* hardware has been reset, we need to reload some things */
2848         e1000_configure(adapter);
2849
2850         clear_bit(__E1000_DOWN, &adapter->state);
2851
2852         napi_enable(&adapter->napi);
2853         if (adapter->msix_entries)
2854                 e1000_configure_msix(adapter);
2855         e1000_irq_enable(adapter);
2856
2857         netif_wake_queue(adapter->netdev);
2858
2859         /* fire a link change interrupt to start the watchdog */
2860         ew32(ICS, E1000_ICS_LSC);
2861         return 0;
2862 }
2863
2864 void e1000e_down(struct e1000_adapter *adapter)
2865 {
2866         struct net_device *netdev = adapter->netdev;
2867         struct e1000_hw *hw = &adapter->hw;
2868         u32 tctl, rctl;
2869
2870         /*
2871          * signal that we're down so the interrupt handler does not
2872          * reschedule our watchdog timer
2873          */
2874         set_bit(__E1000_DOWN, &adapter->state);
2875
2876         /* disable receives in the hardware */
2877         rctl = er32(RCTL);
2878         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2879         /* flush and sleep below */
2880
2881         netif_stop_queue(netdev);
2882
2883         /* disable transmits in the hardware */
2884         tctl = er32(TCTL);
2885         tctl &= ~E1000_TCTL_EN;
2886         ew32(TCTL, tctl);
2887         /* flush both disables and wait for them to finish */
2888         e1e_flush();
2889         msleep(10);
2890
2891         napi_disable(&adapter->napi);
2892         e1000_irq_disable(adapter);
2893
2894         del_timer_sync(&adapter->watchdog_timer);
2895         del_timer_sync(&adapter->phy_info_timer);
2896
2897         netdev->tx_queue_len = adapter->tx_queue_len;
2898         netif_carrier_off(netdev);
2899         adapter->link_speed = 0;
2900         adapter->link_duplex = 0;
2901
2902         if (!pci_channel_offline(adapter->pdev))
2903                 e1000e_reset(adapter);
2904         e1000_clean_tx_ring(adapter);
2905         e1000_clean_rx_ring(adapter);
2906
2907         if (adapter->flags & FLAG_HAS_ERT)
2908                 pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY,
2909                                           adapter->netdev->name);
2910
2911         /*
2912          * TODO: for power management, we could drop the link and
2913          * pci_disable_device here.
2914          */
2915 }
2916
2917 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2918 {
2919         might_sleep();
2920         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2921                 msleep(1);
2922         e1000e_down(adapter);
2923         e1000e_up(adapter);
2924         clear_bit(__E1000_RESETTING, &adapter->state);
2925 }
2926
2927 /**
2928  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2929  * @adapter: board private structure to initialize
2930  *
2931  * e1000_sw_init initializes the Adapter private data structure.
2932  * Fields are initialized based on PCI device information and
2933  * OS network device settings (MTU size).
2934  **/
2935 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2936 {
2937         struct net_device *netdev = adapter->netdev;
2938
2939         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2940         adapter->rx_ps_bsize0 = 128;
2941         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2942         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2943
2944         e1000e_set_interrupt_capability(adapter);
2945
2946         if (e1000_alloc_queues(adapter))
2947                 return -ENOMEM;
2948
2949         /* Explicitly disable IRQ since the NIC can be in any state. */
2950         e1000_irq_disable(adapter);
2951
2952         set_bit(__E1000_DOWN, &adapter->state);
2953         return 0;
2954 }
2955
2956 /**
2957  * e1000_intr_msi_test - Interrupt Handler
2958  * @irq: interrupt number
2959  * @data: pointer to a network interface device structure
2960  **/
2961 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2962 {
2963         struct net_device *netdev = data;
2964         struct e1000_adapter *adapter = netdev_priv(netdev);
2965         struct e1000_hw *hw = &adapter->hw;
2966         u32 icr = er32(ICR);
2967
2968         e_dbg("icr is %08X\n", icr);
2969         if (icr & E1000_ICR_RXSEQ) {
2970                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2971                 wmb();
2972         }
2973
2974         return IRQ_HANDLED;
2975 }
2976
2977 /**
2978  * e1000_test_msi_interrupt - Returns 0 for successful test
2979  * @adapter: board private struct
2980  *
2981  * code flow taken from tg3.c
2982  **/
2983 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2984 {
2985         struct net_device *netdev = adapter->netdev;
2986         struct e1000_hw *hw = &adapter->hw;
2987         int err;
2988
2989         /* poll_enable hasn't been called yet, so don't need disable */
2990         /* clear any pending events */
2991         er32(ICR);
2992
2993         /* free the real vector and request a test handler */
2994         e1000_free_irq(adapter);
2995         e1000e_reset_interrupt_capability(adapter);
2996
2997         /* Assume that the test fails, if it succeeds then the test
2998          * MSI irq handler will unset this flag */
2999         adapter->flags |= FLAG_MSI_TEST_FAILED;
3000
3001         err = pci_enable_msi(adapter->pdev);
3002         if (err)
3003                 goto msi_test_failed;
3004
3005         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3006                           netdev->name, netdev);
3007         if (err) {
3008                 pci_disable_msi(adapter->pdev);
3009                 goto msi_test_failed;
3010         }
3011
3012         wmb();
3013
3014         e1000_irq_enable(adapter);
3015
3016         /* fire an unusual interrupt on the test handler */
3017         ew32(ICS, E1000_ICS_RXSEQ);
3018         e1e_flush();
3019         msleep(50);
3020
3021         e1000_irq_disable(adapter);
3022
3023         rmb();
3024
3025         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3026                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3027                 err = -EIO;
3028                 e_info("MSI interrupt test failed!\n");
3029         }
3030
3031         free_irq(adapter->pdev->irq, netdev);
3032         pci_disable_msi(adapter->pdev);
3033
3034         if (err == -EIO)
3035                 goto msi_test_failed;
3036
3037         /* okay so the test worked, restore settings */
3038         e_dbg("MSI interrupt test succeeded!\n");
3039 msi_test_failed:
3040         e1000e_set_interrupt_capability(adapter);
3041         e1000_request_irq(adapter);
3042         return err;
3043 }
3044
3045 /**
3046  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3047  * @adapter: board private struct
3048  *
3049  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3050  **/
3051 static int e1000_test_msi(struct e1000_adapter *adapter)
3052 {
3053         int err;
3054         u16 pci_cmd;
3055
3056         if (!(adapter->flags & FLAG_MSI_ENABLED))
3057                 return 0;
3058
3059         /* disable SERR in case the MSI write causes a master abort */
3060         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3061         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3062                               pci_cmd & ~PCI_COMMAND_SERR);
3063
3064         err = e1000_test_msi_interrupt(adapter);
3065
3066         /* restore previous setting of command word */
3067         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3068
3069         /* success ! */
3070         if (!err)
3071                 return 0;
3072
3073         /* EIO means MSI test failed */
3074         if (err != -EIO)
3075                 return err;
3076
3077         /* back to INTx mode */
3078         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3079
3080         e1000_free_irq(adapter);
3081
3082         err = e1000_request_irq(adapter);
3083
3084         return err;
3085 }
3086
3087 /**
3088  * e1000_open - Called when a network interface is made active
3089  * @netdev: network interface device structure
3090  *
3091  * Returns 0 on success, negative value on failure
3092  *
3093  * The open entry point is called when a network interface is made
3094  * active by the system (IFF_UP).  At this point all resources needed
3095  * for transmit and receive operations are allocated, the interrupt
3096  * handler is registered with the OS, the watchdog timer is started,
3097  * and the stack is notified that the interface is ready.
3098  **/
3099 static int e1000_open(struct net_device *netdev)
3100 {
3101         struct e1000_adapter *adapter = netdev_priv(netdev);
3102         struct e1000_hw *hw = &adapter->hw;
3103         int err;
3104
3105         /* disallow open during test */
3106         if (test_bit(__E1000_TESTING, &adapter->state))
3107                 return -EBUSY;
3108
3109         netif_carrier_off(netdev);
3110
3111         /* allocate transmit descriptors */
3112         err = e1000e_setup_tx_resources(adapter);
3113         if (err)
3114                 goto err_setup_tx;
3115
3116         /* allocate receive descriptors */
3117         err = e1000e_setup_rx_resources(adapter);
3118         if (err)
3119                 goto err_setup_rx;
3120
3121         e1000e_power_up_phy(adapter);
3122
3123         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3124         if ((adapter->hw.mng_cookie.status &
3125              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3126                 e1000_update_mng_vlan(adapter);
3127
3128         /*
3129          * If AMT is enabled, let the firmware know that the network
3130          * interface is now open
3131          */
3132         if (adapter->flags & FLAG_HAS_AMT)
3133                 e1000_get_hw_control(adapter);
3134
3135         /*
3136          * before we allocate an interrupt, we must be ready to handle it.
3137          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3138          * as soon as we call pci_request_irq, so we have to setup our
3139          * clean_rx handler before we do so.
3140          */
3141         e1000_configure(adapter);
3142
3143         err = e1000_request_irq(adapter);
3144         if (err)
3145                 goto err_req_irq;
3146
3147         /*
3148          * Work around PCIe errata with MSI interrupts causing some chipsets to
3149          * ignore e1000e MSI messages, which means we need to test our MSI
3150          * interrupt now
3151          */
3152         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3153                 err = e1000_test_msi(adapter);
3154                 if (err) {
3155                         e_err("Interrupt allocation failed\n");
3156                         goto err_req_irq;
3157                 }
3158         }
3159
3160         /* From here on the code is the same as e1000e_up() */
3161         clear_bit(__E1000_DOWN, &adapter->state);
3162
3163         napi_enable(&adapter->napi);
3164
3165         e1000_irq_enable(adapter);
3166
3167         netif_start_queue(netdev);
3168
3169         /* fire a link status change interrupt to start the watchdog */
3170         ew32(ICS, E1000_ICS_LSC);
3171
3172         return 0;
3173
3174 err_req_irq:
3175         e1000_release_hw_control(adapter);
3176         e1000_power_down_phy(adapter);
3177         e1000e_free_rx_resources(adapter);
3178 err_setup_rx:
3179         e1000e_free_tx_resources(adapter);
3180 err_setup_tx:
3181         e1000e_reset(adapter);
3182
3183         return err;
3184 }
3185
3186 /**
3187  * e1000_close - Disables a network interface
3188  * @netdev: network interface device structure
3189  *
3190  * Returns 0, this is not allowed to fail
3191  *
3192  * The close entry point is called when an interface is de-activated
3193  * by the OS.  The hardware is still under the drivers control, but
3194  * needs to be disabled.  A global MAC reset is issued to stop the
3195  * hardware, and all transmit and receive resources are freed.
3196  **/
3197 static int e1000_close(struct net_device *netdev)
3198 {
3199         struct e1000_adapter *adapter = netdev_priv(netdev);
3200
3201         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3202         e1000e_down(adapter);
3203         e1000_power_down_phy(adapter);
3204         e1000_free_irq(adapter);
3205
3206         e1000e_free_tx_resources(adapter);
3207         e1000e_free_rx_resources(adapter);
3208
3209         /*
3210          * kill manageability vlan ID if supported, but not if a vlan with
3211          * the same ID is registered on the host OS (let 8021q kill it)
3212          */
3213         if ((adapter->hw.mng_cookie.status &
3214                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3215              !(adapter->vlgrp &&
3216                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3217                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3218
3219         /*
3220          * If AMT is enabled, let the firmware know that the network
3221          * interface is now closed
3222          */
3223         if (adapter->flags & FLAG_HAS_AMT)
3224                 e1000_release_hw_control(adapter);
3225
3226         return 0;
3227 }
3228 /**
3229  * e1000_set_mac - Change the Ethernet Address of the NIC
3230  * @netdev: network interface device structure
3231  * @p: pointer to an address structure
3232  *
3233  * Returns 0 on success, negative on failure
3234  **/
3235 static int e1000_set_mac(struct net_device *netdev, void *p)
3236 {
3237         struct e1000_adapter *adapter = netdev_priv(netdev);
3238         struct sockaddr *addr = p;
3239
3240         if (!is_valid_ether_addr(addr->sa_data))
3241                 return -EADDRNOTAVAIL;
3242
3243         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3244         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3245
3246         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3247
3248         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3249                 /* activate the work around */
3250                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3251
3252                 /*
3253                  * Hold a copy of the LAA in RAR[14] This is done so that
3254                  * between the time RAR[0] gets clobbered  and the time it
3255                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3256                  * of the RARs and no incoming packets directed to this port
3257                  * are dropped. Eventually the LAA will be in RAR[0] and
3258                  * RAR[14]
3259                  */
3260                 e1000e_rar_set(&adapter->hw,
3261                               adapter->hw.mac.addr,
3262                               adapter->hw.mac.rar_entry_count - 1);
3263         }
3264
3265         return 0;
3266 }
3267
3268 /**
3269  * e1000e_update_phy_task - work thread to update phy
3270  * @work: pointer to our work struct
3271  *
3272  * this worker thread exists because we must acquire a
3273  * semaphore to read the phy, which we could msleep while
3274  * waiting for it, and we can't msleep in a timer.
3275  **/
3276 static void e1000e_update_phy_task(struct work_struct *work)
3277 {
3278         struct e1000_adapter *adapter = container_of(work,
3279                                         struct e1000_adapter, update_phy_task);
3280         e1000_get_phy_info(&adapter->hw);
3281 }
3282
3283 /*
3284  * Need to wait a few seconds after link up to get diagnostic information from
3285  * the phy
3286  */
3287 static void e1000_update_phy_info(unsigned long data)
3288 {
3289         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3290         schedule_work(&adapter->update_phy_task);
3291 }
3292
3293 /**
3294  * e1000e_update_stats - Update the board statistics counters
3295  * @adapter: board private structure
3296  **/
3297 void e1000e_update_stats(struct e1000_adapter *adapter)
3298 {
3299         struct net_device *netdev = adapter->netdev;
3300         struct e1000_hw *hw = &adapter->hw;
3301         struct pci_dev *pdev = adapter->pdev;
3302         u16 phy_data;
3303
3304         /*
3305          * Prevent stats update while adapter is being reset, or if the pci
3306          * connection is down.
3307          */
3308         if (adapter->link_speed == 0)
3309                 return;
3310         if (pci_channel_offline(pdev))
3311                 return;
3312
3313         adapter->stats.crcerrs += er32(CRCERRS);
3314         adapter->stats.gprc += er32(GPRC);
3315         adapter->stats.gorc += er32(GORCL);
3316         er32(GORCH); /* Clear gorc */
3317         adapter->stats.bprc += er32(BPRC);
3318         adapter->stats.mprc += er32(MPRC);
3319         adapter->stats.roc += er32(ROC);
3320
3321         adapter->stats.mpc += er32(MPC);
3322         if ((hw->phy.type == e1000_phy_82578) ||
3323             (hw->phy.type == e1000_phy_82577)) {
3324                 e1e_rphy(hw, HV_SCC_UPPER, &phy_data);
3325                 if (!e1e_rphy(hw, HV_SCC_LOWER, &phy_data))
3326                         adapter->stats.scc += phy_data;
3327
3328                 e1e_rphy(hw, HV_ECOL_UPPER, &phy_data);
3329                 if (!e1e_rphy(hw, HV_ECOL_LOWER, &phy_data))
3330                         adapter->stats.ecol += phy_data;
3331
3332                 e1e_rphy(hw, HV_MCC_UPPER, &phy_data);
3333                 if (!e1e_rphy(hw, HV_MCC_LOWER, &phy_data))
3334                         adapter->stats.mcc += phy_data;
3335
3336                 e1e_rphy(hw, HV_LATECOL_UPPER, &phy_data);
3337                 if (!e1e_rphy(hw, HV_LATECOL_LOWER, &phy_data))
3338                         adapter->stats.latecol += phy_data;
3339
3340                 e1e_rphy(hw, HV_DC_UPPER, &phy_data);
3341                 if (!e1e_rphy(hw, HV_DC_LOWER, &phy_data))
3342                         adapter->stats.dc += phy_data;
3343         } else {
3344                 adapter->stats.scc += er32(SCC);
3345                 adapter->stats.ecol += er32(ECOL);
3346                 adapter->stats.mcc += er32(MCC);
3347                 adapter->stats.latecol += er32(LATECOL);
3348                 adapter->stats.dc += er32(DC);
3349         }
3350         adapter->stats.xonrxc += er32(XONRXC);
3351         adapter->stats.xontxc += er32(XONTXC);
3352         adapter->stats.xoffrxc += er32(XOFFRXC);
3353         adapter->stats.xofftxc += er32(XOFFTXC);
3354         adapter->stats.gptc += er32(GPTC);
3355         adapter->stats.gotc += er32(GOTCL);
3356         er32(GOTCH); /* Clear gotc */
3357         adapter->stats.rnbc += er32(RNBC);
3358         adapter->stats.ruc += er32(RUC);
3359
3360         adapter->stats.mptc += er32(MPTC);
3361         adapter->stats.bptc += er32(BPTC);
3362
3363         /* used for adaptive IFS */
3364
3365         hw->mac.tx_packet_delta = er32(TPT);
3366         adapter->stats.tpt += hw->mac.tx_packet_delta;
3367         if ((hw->phy.type == e1000_phy_82578) ||
3368             (hw->phy.type == e1000_phy_82577)) {
3369                 e1e_rphy(hw, HV_COLC_UPPER, &phy_data);
3370                 if (!e1e_rphy(hw, HV_COLC_LOWER, &phy_data))
3371                         hw->mac.collision_delta = phy_data;
3372         } else {
3373                 hw->mac.collision_delta = er32(COLC);
3374         }
3375         adapter->stats.colc += hw->mac.collision_delta;
3376
3377         adapter->stats.algnerrc += er32(ALGNERRC);
3378         adapter->stats.rxerrc += er32(RXERRC);
3379         if ((hw->phy.type == e1000_phy_82578) ||
3380             (hw->phy.type == e1000_phy_82577)) {
3381                 e1e_rphy(hw, HV_TNCRS_UPPER, &phy_data);
3382                 if (!e1e_rphy(hw, HV_TNCRS_LOWER, &phy_data))
3383                         adapter->stats.tncrs += phy_data;
3384         } else {
3385                 if ((hw->mac.type != e1000_82574) &&
3386                     (hw->mac.type != e1000_82583))
3387                         adapter->stats.tncrs += er32(TNCRS);
3388         }
3389         adapter->stats.cexterr += er32(CEXTERR);
3390         adapter->stats.tsctc += er32(TSCTC);
3391         adapter->stats.tsctfc += er32(TSCTFC);
3392
3393         /* Fill out the OS statistics structure */
3394         netdev->stats.multicast = adapter->stats.mprc;
3395         netdev->stats.collisions = adapter->stats.colc;
3396
3397         /* Rx Errors */
3398
3399         /*
3400          * RLEC on some newer hardware can be incorrect so build
3401          * our own version based on RUC and ROC
3402          */
3403         netdev->stats.rx_errors = adapter->stats.rxerrc +
3404                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3405                 adapter->stats.ruc + adapter->stats.roc +
3406                 adapter->stats.cexterr;
3407         netdev->stats.rx_length_errors = adapter->stats.ruc +
3408                                               adapter->stats.roc;
3409         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3410         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3411         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3412
3413         /* Tx Errors */
3414         netdev->stats.tx_errors = adapter->stats.ecol +
3415                                        adapter->stats.latecol;
3416         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3417         netdev->stats.tx_window_errors = adapter->stats.latecol;
3418         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3419
3420         /* Tx Dropped needs to be maintained elsewhere */
3421
3422         /* Management Stats */
3423         adapter->stats.mgptc += er32(MGTPTC);
3424         adapter->stats.mgprc += er32(MGTPRC);
3425         adapter->stats.mgpdc += er32(MGTPDC);
3426 }
3427
3428 /**
3429  * e1000_phy_read_status - Update the PHY register status snapshot
3430  * @adapter: board private structure
3431  **/
3432 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3433 {
3434         struct e1000_hw *hw = &adapter->hw;
3435         struct e1000_phy_regs *phy = &adapter->phy_regs;
3436         int ret_val;
3437
3438         if ((er32(STATUS) & E1000_STATUS_LU) &&
3439             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3440                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3441                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3442                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3443                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3444                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3445                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3446                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3447                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3448                 if (ret_val)
3449                         e_warn("Error reading PHY register\n");
3450         } else {
3451                 /*
3452                  * Do not read PHY registers if link is not up
3453                  * Set values to typical power-on defaults
3454                  */
3455                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3456                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3457                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3458                              BMSR_ERCAP);
3459                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3460                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3461                 phy->lpa = 0;
3462                 phy->expansion = EXPANSION_ENABLENPAGE;
3463                 phy->ctrl1000 = ADVERTISE_1000FULL;
3464                 phy->stat1000 = 0;
3465                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3466         }
3467 }
3468
3469 static void e1000_print_link_info(struct e1000_adapter *adapter)
3470 {
3471         struct e1000_hw *hw = &adapter->hw;
3472         u32 ctrl = er32(CTRL);
3473
3474         /* Link status message must follow this format for user tools */
3475         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3476                "Flow Control: %s\n",
3477                adapter->netdev->name,
3478                adapter->link_speed,
3479                (adapter->link_duplex == FULL_DUPLEX) ?
3480                                 "Full Duplex" : "Half Duplex",
3481                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3482                                 "RX/TX" :
3483                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3484                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3485 }
3486
3487 bool e1000_has_link(struct e1000_adapter *adapter)
3488 {
3489         struct e1000_hw *hw = &adapter->hw;
3490         bool link_active = 0;
3491         s32 ret_val = 0;
3492
3493         /*
3494          * get_link_status is set on LSC (link status) interrupt or
3495          * Rx sequence error interrupt.  get_link_status will stay
3496          * false until the check_for_link establishes link
3497          * for copper adapters ONLY
3498          */
3499         switch (hw->phy.media_type) {
3500         case e1000_media_type_copper:
3501                 if (hw->mac.get_link_status) {
3502                         ret_val = hw->mac.ops.check_for_link(hw);
3503                         link_active = !hw->mac.get_link_status;
3504                 } else {
3505                         link_active = 1;
3506                 }
3507                 break;
3508         case e1000_media_type_fiber:
3509                 ret_val = hw->mac.ops.check_for_link(hw);
3510                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3511                 break;
3512         case e1000_media_type_internal_serdes:
3513                 ret_val = hw->mac.ops.check_for_link(hw);
3514                 link_active = adapter->hw.mac.serdes_has_link;
3515                 break;
3516         default:
3517         case e1000_media_type_unknown:
3518                 break;
3519         }
3520
3521         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3522             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3523                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3524                 e_info("Gigabit has been disabled, downgrading speed\n");
3525         }
3526
3527         return link_active;
3528 }
3529
3530 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3531 {
3532         /* make sure the receive unit is started */
3533         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3534             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3535                 struct e1000_hw *hw = &adapter->hw;
3536                 u32 rctl = er32(RCTL);
3537                 ew32(RCTL, rctl | E1000_RCTL_EN);
3538                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3539         }
3540 }
3541
3542 /**
3543  * e1000_watchdog - Timer Call-back
3544  * @data: pointer to adapter cast into an unsigned long
3545  **/
3546 static void e1000_watchdog(unsigned long data)
3547 {
3548         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3549
3550         /* Do the rest outside of interrupt context */
3551         schedule_work(&adapter->watchdog_task);
3552
3553         /* TODO: make this use queue_delayed_work() */
3554 }
3555
3556 static void e1000_watchdog_task(struct work_struct *work)
3557 {
3558         struct e1000_adapter *adapter = container_of(work,
3559                                         struct e1000_adapter, watchdog_task);
3560         struct net_device *netdev = adapter->netdev;
3561         struct e1000_mac_info *mac = &adapter->hw.mac;
3562         struct e1000_phy_info *phy = &adapter->hw.phy;
3563         struct e1000_ring *tx_ring = adapter->tx_ring;
3564         struct e1000_hw *hw = &adapter->hw;
3565         u32 link, tctl;
3566         int tx_pending = 0;
3567
3568         link = e1000_has_link(adapter);
3569         if ((netif_carrier_ok(netdev)) && link) {
3570                 e1000e_enable_receives(adapter);
3571                 goto link_up;
3572         }
3573
3574         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3575             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3576                 e1000_update_mng_vlan(adapter);
3577
3578         if (link) {
3579                 if (!netif_carrier_ok(netdev)) {
3580                         bool txb2b = 1;
3581                         /* update snapshot of PHY registers on LSC */
3582                         e1000_phy_read_status(adapter);
3583                         mac->ops.get_link_up_info(&adapter->hw,
3584                                                    &adapter->link_speed,
3585                                                    &adapter->link_duplex);
3586                         e1000_print_link_info(adapter);
3587                         /*
3588                          * On supported PHYs, check for duplex mismatch only
3589                          * if link has autonegotiated at 10/100 half
3590                          */
3591                         if ((hw->phy.type == e1000_phy_igp_3 ||
3592                              hw->phy.type == e1000_phy_bm) &&
3593                             (hw->mac.autoneg == true) &&
3594                             (adapter->link_speed == SPEED_10 ||
3595                              adapter->link_speed == SPEED_100) &&
3596                             (adapter->link_duplex == HALF_DUPLEX)) {
3597                                 u16 autoneg_exp;
3598
3599                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3600
3601                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3602                                         e_info("Autonegotiated half duplex but"
3603                                                " link partner cannot autoneg. "
3604                                                " Try forcing full duplex if "
3605                                                "link gets many collisions.\n");
3606                         }
3607
3608                         /*
3609                          * tweak tx_queue_len according to speed/duplex
3610                          * and adjust the timeout factor
3611                          */
3612                         netdev->tx_queue_len = adapter->tx_queue_len;
3613                         adapter->tx_timeout_factor = 1;
3614                         switch (adapter->link_speed) {
3615                         case SPEED_10:
3616                                 txb2b = 0;
3617                                 netdev->tx_queue_len = 10;
3618                                 adapter->tx_timeout_factor = 16;
3619                                 break;
3620                         case SPEED_100:
3621                                 txb2b = 0;
3622                                 netdev->tx_queue_len = 100;
3623                                 adapter->tx_timeout_factor = 10;
3624                                 break;
3625                         }
3626
3627                         /*
3628                          * workaround: re-program speed mode bit after
3629                          * link-up event
3630                          */
3631                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3632                             !txb2b) {
3633                                 u32 tarc0;
3634                                 tarc0 = er32(TARC(0));
3635                                 tarc0 &= ~SPEED_MODE_BIT;
3636                                 ew32(TARC(0), tarc0);
3637                         }
3638
3639                         /*
3640                          * disable TSO for pcie and 10/100 speeds, to avoid
3641                          * some hardware issues
3642                          */
3643                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3644                                 switch (adapter->link_speed) {
3645                                 case SPEED_10:
3646                                 case SPEED_100:
3647                                         e_info("10/100 speed: disabling TSO\n");
3648                                         netdev->features &= ~NETIF_F_TSO;
3649                                         netdev->features &= ~NETIF_F_TSO6;
3650                                         break;
3651                                 case SPEED_1000:
3652                                         netdev->features |= NETIF_F_TSO;
3653                                         netdev->features |= NETIF_F_TSO6;
3654                                         break;
3655                                 default:
3656                                         /* oops */
3657                                         break;
3658                                 }
3659                         }
3660
3661                         /*
3662                          * enable transmits in the hardware, need to do this
3663                          * after setting TARC(0)
3664                          */
3665                         tctl = er32(TCTL);
3666                         tctl |= E1000_TCTL_EN;
3667                         ew32(TCTL, tctl);
3668
3669                         /*
3670                          * Perform any post-link-up configuration before
3671                          * reporting link up.
3672                          */
3673                         if (phy->ops.cfg_on_link_up)
3674                                 phy->ops.cfg_on_link_up(hw);
3675
3676                         netif_carrier_on(netdev);
3677
3678                         if (!test_bit(__E1000_DOWN, &adapter->state))
3679                                 mod_timer(&adapter->phy_info_timer,
3680                                           round_jiffies(jiffies + 2 * HZ));
3681                 }
3682         } else {
3683                 if (netif_carrier_ok(netdev)) {
3684                         adapter->link_speed = 0;
3685                         adapter->link_duplex = 0;
3686                         /* Link status message must follow this format */
3687                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
3688                                adapter->netdev->name);
3689                         netif_carrier_off(netdev);
3690                         if (!test_bit(__E1000_DOWN, &adapter->state))
3691                                 mod_timer(&adapter->phy_info_timer,
3692                                           round_jiffies(jiffies + 2 * HZ));
3693
3694                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3695                                 schedule_work(&adapter->reset_task);
3696                 }
3697         }
3698
3699 link_up:
3700         e1000e_update_stats(adapter);
3701
3702         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3703         adapter->tpt_old = adapter->stats.tpt;
3704         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3705         adapter->colc_old = adapter->stats.colc;
3706
3707         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3708         adapter->gorc_old = adapter->stats.gorc;
3709         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3710         adapter->gotc_old = adapter->stats.gotc;
3711
3712         e1000e_update_adaptive(&adapter->hw);
3713
3714         if (!netif_carrier_ok(netdev)) {
3715                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3716                                tx_ring->count);
3717                 if (tx_pending) {
3718                         /*
3719                          * We've lost link, so the controller stops DMA,
3720                          * but we've got queued Tx work that's never going
3721                          * to get done, so reset controller to flush Tx.
3722                          * (Do the reset outside of interrupt context).
3723                          */
3724                         adapter->tx_timeout_count++;
3725                         schedule_work(&adapter->reset_task);
3726                         /* return immediately since reset is imminent */
3727                         return;
3728                 }
3729         }
3730
3731         /* Cause software interrupt to ensure Rx ring is cleaned */
3732         if (adapter->msix_entries)
3733                 ew32(ICS, adapter->rx_ring->ims_val);
3734         else
3735                 ew32(ICS, E1000_ICS_RXDMT0);
3736
3737         /* Force detection of hung controller every watchdog period */
3738         adapter->detect_tx_hung = 1;
3739
3740         /*
3741          * With 82571 controllers, LAA may be overwritten due to controller
3742          * reset from the other port. Set the appropriate LAA in RAR[0]
3743          */
3744         if (e1000e_get_laa_state_82571(hw))
3745                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3746
3747         /* Reset the timer */
3748         if (!test_bit(__E1000_DOWN, &adapter->state))
3749                 mod_timer(&adapter->watchdog_timer,
3750                           round_jiffies(jiffies + 2 * HZ));
3751 }
3752
3753 #define E1000_TX_FLAGS_CSUM             0x00000001
3754 #define E1000_TX_FLAGS_VLAN             0x00000002
3755 #define E1000_TX_FLAGS_TSO              0x00000004
3756 #define E1000_TX_FLAGS_IPV4             0x00000008
3757 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3758 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3759
3760 static int e1000_tso(struct e1000_adapter *adapter,
3761                      struct sk_buff *skb)
3762 {
3763         struct e1000_ring *tx_ring = adapter->tx_ring;
3764         struct e1000_context_desc *context_desc;
3765         struct e1000_buffer *buffer_info;
3766         unsigned int i;
3767         u32 cmd_length = 0;
3768         u16 ipcse = 0, tucse, mss;
3769         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3770         int err;
3771
3772         if (!skb_is_gso(skb))
3773                 return 0;
3774
3775         if (skb_header_cloned(skb)) {
3776                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3777                 if (err)
3778                         return err;
3779         }
3780
3781         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3782         mss = skb_shinfo(skb)->gso_size;
3783         if (skb->protocol == htons(ETH_P_IP)) {
3784                 struct iphdr *iph = ip_hdr(skb);
3785                 iph->tot_len = 0;
3786                 iph->check = 0;
3787                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
3788                                                          0, IPPROTO_TCP, 0);
3789                 cmd_length = E1000_TXD_CMD_IP;
3790                 ipcse = skb_transport_offset(skb) - 1;
3791         } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3792                 ipv6_hdr(skb)->payload_len = 0;
3793                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3794                                                        &ipv6_hdr(skb)->daddr,
3795                                                        0, IPPROTO_TCP, 0);
3796                 ipcse = 0;
3797         }
3798         ipcss = skb_network_offset(skb);
3799         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3800         tucss = skb_transport_offset(skb);
3801         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3802         tucse = 0;
3803
3804         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3805                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3806
3807         i = tx_ring->next_to_use;
3808         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3809         buffer_info = &tx_ring->buffer_info[i];
3810
3811         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3812         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3813         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3814         context_desc->upper_setup.tcp_fields.tucss = tucss;
3815         context_desc->upper_setup.tcp_fields.tucso = tucso;
3816         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3817         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3818         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3819         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3820
3821         buffer_info->time_stamp = jiffies;
3822         buffer_info->next_to_watch = i;
3823
3824         i++;
3825         if (i == tx_ring->count)
3826                 i = 0;
3827         tx_ring->next_to_use = i;
3828
3829         return 1;
3830 }
3831
3832 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3833 {
3834         struct e1000_ring *tx_ring = adapter->tx_ring;
3835         struct e1000_context_desc *context_desc;
3836         struct e1000_buffer *buffer_info;
3837         unsigned int i;
3838         u8 css;
3839         u32 cmd_len = E1000_TXD_CMD_DEXT;
3840         __be16 protocol;
3841
3842         if (skb->ip_summed != CHECKSUM_PARTIAL)
3843                 return 0;
3844
3845         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
3846                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
3847         else
3848                 protocol = skb->protocol;
3849
3850         switch (protocol) {
3851         case cpu_to_be16(ETH_P_IP):
3852                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3853                         cmd_len |= E1000_TXD_CMD_TCP;
3854                 break;
3855         case cpu_to_be16(ETH_P_IPV6):
3856                 /* XXX not handling all IPV6 headers */
3857                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3858                         cmd_len |= E1000_TXD_CMD_TCP;
3859                 break;
3860         default:
3861                 if (unlikely(net_ratelimit()))
3862                         e_warn("checksum_partial proto=%x!\n",
3863                                be16_to_cpu(protocol));
3864                 break;
3865         }
3866
3867         css = skb_transport_offset(skb);
3868
3869         i = tx_ring->next_to_use;
3870         buffer_info = &tx_ring->buffer_info[i];
3871         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3872
3873         context_desc->lower_setup.ip_config = 0;
3874         context_desc->upper_setup.tcp_fields.tucss = css;
3875         context_desc->upper_setup.tcp_fields.tucso =
3876                                 css + skb->csum_offset;
3877         context_desc->upper_setup.tcp_fields.tucse = 0;
3878         context_desc->tcp_seg_setup.data = 0;
3879         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
3880
3881         buffer_info->time_stamp = jiffies;
3882         buffer_info->next_to_watch = i;
3883
3884         i++;
3885         if (i == tx_ring->count)
3886                 i = 0;
3887         tx_ring->next_to_use = i;
3888
3889         return 1;
3890 }
3891
3892 #define E1000_MAX_PER_TXD       8192
3893 #define E1000_MAX_TXD_PWR       12
3894
3895 static int e1000_tx_map(struct e1000_adapter *adapter,
3896                         struct sk_buff *skb, unsigned int first,
3897                         unsigned int max_per_txd, unsigned int nr_frags,
3898                         unsigned int mss)
3899 {
3900         struct e1000_ring *tx_ring = adapter->tx_ring;
3901         struct pci_dev *pdev = adapter->pdev;
3902         struct e1000_buffer *buffer_info;
3903         unsigned int len = skb_headlen(skb);
3904         unsigned int offset = 0, size, count = 0, i;
3905         unsigned int f;
3906
3907         i = tx_ring->next_to_use;
3908
3909         while (len) {
3910                 buffer_info = &tx_ring->buffer_info[i];
3911                 size = min(len, max_per_txd);
3912
3913                 buffer_info->length = size;
3914                 buffer_info->time_stamp = jiffies;
3915                 buffer_info->next_to_watch = i;
3916                 buffer_info->dma = pci_map_single(pdev, skb->data + offset,
3917                                                   size, PCI_DMA_TODEVICE);
3918                 buffer_info->mapped_as_page = false;
3919                 if (pci_dma_mapping_error(pdev, buffer_info->dma))
3920                         goto dma_error;
3921
3922                 len -= size;
3923                 offset += size;
3924                 count++;
3925
3926                 if (len) {
3927                         i++;
3928                         if (i == tx_ring->count)
3929                                 i = 0;
3930                 }
3931         }
3932
3933         for (f = 0; f < nr_frags; f++) {
3934                 struct skb_frag_struct *frag;
3935
3936                 frag = &skb_shinfo(skb)->frags[f];
3937                 len = frag->size;
3938                 offset = frag->page_offset;
3939
3940                 while (len) {
3941                         i++;
3942                         if (i == tx_ring->count)
3943                                 i = 0;
3944
3945                         buffer_info = &tx_ring->buffer_info[i];
3946                         size = min(len, max_per_txd);
3947
3948                         buffer_info->length = size;
3949                         buffer_info->time_stamp = jiffies;
3950                         buffer_info->next_to_watch = i;
3951                         buffer_info->dma = pci_map_page(pdev, frag->page,
3952                                                         offset, size,
3953                                                         PCI_DMA_TODEVICE);
3954                         buffer_info->mapped_as_page = true;
3955                         if (pci_dma_mapping_error(pdev, buffer_info->dma))
3956                                 goto dma_error;
3957
3958                         len -= size;
3959                         offset += size;
3960                         count++;
3961                 }
3962         }
3963
3964         tx_ring->buffer_info[i].skb = skb;
3965         tx_ring->buffer_info[first].next_to_watch = i;
3966
3967         return count;
3968
3969 dma_error:
3970         dev_err(&pdev->dev, "TX DMA map failed\n");
3971         buffer_info->dma = 0;
3972         if (count)
3973                 count--;
3974
3975         while (count--) {
3976                 if (i==0)
3977                         i += tx_ring->count;
3978                 i--;
3979                 buffer_info = &tx_ring->buffer_info[i];
3980                 e1000_put_txbuf(adapter, buffer_info);;
3981         }
3982
3983         return 0;
3984 }
3985
3986 static void e1000_tx_queue(struct e1000_adapter *adapter,
3987                            int tx_flags, int count)
3988 {
3989         struct e1000_ring *tx_ring = adapter->tx_ring;
3990         struct e1000_tx_desc *tx_desc = NULL;
3991         struct e1000_buffer *buffer_info;
3992         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3993         unsigned int i;
3994
3995         if (tx_flags & E1000_TX_FLAGS_TSO) {
3996                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3997                              E1000_TXD_CMD_TSE;
3998                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3999
4000                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4001                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4002         }
4003
4004         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4005                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4006                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4007         }
4008
4009         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4010                 txd_lower |= E1000_TXD_CMD_VLE;
4011                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4012         }
4013
4014         i = tx_ring->next_to_use;
4015
4016         while (count--) {
4017                 buffer_info = &tx_ring->buffer_info[i];
4018                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4019                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4020                 tx_desc->lower.data =
4021                         cpu_to_le32(txd_lower | buffer_info->length);
4022                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4023
4024                 i++;
4025                 if (i == tx_ring->count)
4026                         i = 0;
4027         }
4028
4029         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4030
4031         /*
4032          * Force memory writes to complete before letting h/w
4033          * know there are new descriptors to fetch.  (Only
4034          * applicable for weak-ordered memory model archs,
4035          * such as IA-64).
4036          */
4037         wmb();
4038
4039         tx_ring->next_to_use = i;
4040         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4041         /*
4042          * we need this if more than one processor can write to our tail
4043          * at a time, it synchronizes IO on IA64/Altix systems
4044          */
4045         mmiowb();
4046 }
4047
4048 #define MINIMUM_DHCP_PACKET_SIZE 282
4049 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4050                                     struct sk_buff *skb)
4051 {
4052         struct e1000_hw *hw =  &adapter->hw;
4053         u16 length, offset;
4054
4055         if (vlan_tx_tag_present(skb)) {
4056                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4057                     (adapter->hw.mng_cookie.status &
4058                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4059                         return 0;
4060         }
4061
4062         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4063                 return 0;
4064
4065         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4066                 return 0;
4067
4068         {
4069                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4070                 struct udphdr *udp;
4071
4072                 if (ip->protocol != IPPROTO_UDP)
4073                         return 0;
4074
4075                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4076                 if (ntohs(udp->dest) != 67)
4077                         return 0;
4078
4079                 offset = (u8 *)udp + 8 - skb->data;
4080                 length = skb->len - offset;
4081                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4082         }
4083
4084         return 0;
4085 }
4086
4087 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4088 {
4089         struct e1000_adapter *adapter = netdev_priv(netdev);
4090
4091         netif_stop_queue(netdev);
4092         /*
4093          * Herbert's original patch had:
4094          *  smp_mb__after_netif_stop_queue();
4095          * but since that doesn't exist yet, just open code it.
4096          */
4097         smp_mb();
4098
4099         /*
4100          * We need to check again in a case another CPU has just
4101          * made room available.
4102          */
4103         if (e1000_desc_unused(adapter->tx_ring) < size)
4104                 return -EBUSY;
4105
4106         /* A reprieve! */
4107         netif_start_queue(netdev);
4108         ++adapter->restart_queue;
4109         return 0;
4110 }
4111
4112 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4113 {
4114         struct e1000_adapter *adapter = netdev_priv(netdev);
4115
4116         if (e1000_desc_unused(adapter->tx_ring) >= size)
4117                 return 0;
4118         return __e1000_maybe_stop_tx(netdev, size);
4119 }
4120
4121 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4122 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4123                                     struct net_device *netdev)
4124 {
4125         struct e1000_adapter *adapter = netdev_priv(netdev);
4126         struct e1000_ring *tx_ring = adapter->tx_ring;
4127         unsigned int first;
4128         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4129         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4130         unsigned int tx_flags = 0;
4131         unsigned int len = skb->len - skb->data_len;
4132         unsigned int nr_frags;
4133         unsigned int mss;
4134         int count = 0;
4135         int tso;
4136         unsigned int f;
4137
4138         if (test_bit(__E1000_DOWN, &adapter->state)) {
4139                 dev_kfree_skb_any(skb);
4140                 return NETDEV_TX_OK;
4141         }
4142
4143         if (skb->len <= 0) {
4144                 dev_kfree_skb_any(skb);
4145                 return NETDEV_TX_OK;
4146         }
4147
4148         mss = skb_shinfo(skb)->gso_size;
4149         /*
4150          * The controller does a simple calculation to
4151          * make sure there is enough room in the FIFO before
4152          * initiating the DMA for each buffer.  The calc is:
4153          * 4 = ceil(buffer len/mss).  To make sure we don't
4154          * overrun the FIFO, adjust the max buffer len if mss
4155          * drops.
4156          */
4157         if (mss) {
4158                 u8 hdr_len;
4159                 max_per_txd = min(mss << 2, max_per_txd);
4160                 max_txd_pwr = fls(max_per_txd) - 1;
4161
4162                 /*
4163                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4164                  * points to just header, pull a few bytes of payload from
4165                  * frags into skb->data
4166                  */
4167                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4168                 /*
4169                  * we do this workaround for ES2LAN, but it is un-necessary,
4170                  * avoiding it could save a lot of cycles
4171                  */
4172                 if (skb->data_len && (hdr_len == len)) {
4173                         unsigned int pull_size;
4174
4175                         pull_size = min((unsigned int)4, skb->data_len);
4176                         if (!__pskb_pull_tail(skb, pull_size)) {
4177                                 e_err("__pskb_pull_tail failed.\n");
4178                                 dev_kfree_skb_any(skb);
4179                                 return NETDEV_TX_OK;
4180                         }
4181                         len = skb->len - skb->data_len;
4182                 }
4183         }
4184
4185         /* reserve a descriptor for the offload context */
4186         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4187                 count++;
4188         count++;
4189
4190         count += TXD_USE_COUNT(len, max_txd_pwr);
4191
4192         nr_frags = skb_shinfo(skb)->nr_frags;
4193         for (f = 0; f < nr_frags; f++)
4194                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4195                                        max_txd_pwr);
4196
4197         if (adapter->hw.mac.tx_pkt_filtering)
4198                 e1000_transfer_dhcp_info(adapter, skb);
4199
4200         /*
4201          * need: count + 2 desc gap to keep tail from touching
4202          * head, otherwise try next time
4203          */
4204         if (e1000_maybe_stop_tx(netdev, count + 2))
4205                 return NETDEV_TX_BUSY;
4206
4207         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4208                 tx_flags |= E1000_TX_FLAGS_VLAN;
4209                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4210         }
4211
4212         first = tx_ring->next_to_use;
4213
4214         tso = e1000_tso(adapter, skb);
4215         if (tso < 0) {
4216                 dev_kfree_skb_any(skb);
4217                 return NETDEV_TX_OK;
4218         }
4219
4220         if (tso)
4221                 tx_flags |= E1000_TX_FLAGS_TSO;
4222         else if (e1000_tx_csum(adapter, skb))
4223                 tx_flags |= E1000_TX_FLAGS_CSUM;
4224
4225         /*
4226          * Old method was to assume IPv4 packet by default if TSO was enabled.
4227          * 82571 hardware supports TSO capabilities for IPv6 as well...
4228          * no longer assume, we must.
4229          */
4230         if (skb->protocol == htons(ETH_P_IP))
4231                 tx_flags |= E1000_TX_FLAGS_IPV4;
4232
4233         /* if count is 0 then mapping error has occured */
4234         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4235         if (count) {
4236                 e1000_tx_queue(adapter, tx_flags, count);
4237                 /* Make sure there is space in the ring for the next send. */
4238                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4239
4240         } else {
4241                 dev_kfree_skb_any(skb);
4242                 tx_ring->buffer_info[first].time_stamp = 0;
4243                 tx_ring->next_to_use = first;
4244         }
4245
4246         return NETDEV_TX_OK;
4247 }
4248
4249 /**
4250  * e1000_tx_timeout - Respond to a Tx Hang
4251  * @netdev: network interface device structure
4252  **/
4253 static void e1000_tx_timeout(struct net_device *netdev)
4254 {
4255         struct e1000_adapter *adapter = netdev_priv(netdev);
4256
4257         /* Do the reset outside of interrupt context */
4258         adapter->tx_timeout_count++;
4259         schedule_work(&adapter->reset_task);
4260 }
4261
4262 static void e1000_reset_task(struct work_struct *work)
4263 {
4264         struct e1000_adapter *adapter;
4265         adapter = container_of(work, struct e1000_adapter, reset_task);
4266
4267         e1000e_reinit_locked(adapter);
4268 }
4269
4270 /**
4271  * e1000_get_stats - Get System Network Statistics
4272  * @netdev: network interface device structure
4273  *
4274  * Returns the address of the device statistics structure.
4275  * The statistics are actually updated from the timer callback.
4276  **/
4277 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4278 {
4279         /* only return the current stats */
4280         return &netdev->stats;
4281 }
4282
4283 /**
4284  * e1000_change_mtu - Change the Maximum Transfer Unit
4285  * @netdev: network interface device structure
4286  * @new_mtu: new value for maximum frame size
4287  *
4288  * Returns 0 on success, negative on failure
4289  **/
4290 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4291 {
4292         struct e1000_adapter *adapter = netdev_priv(netdev);
4293         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4294
4295         /* Jumbo frame support */
4296         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4297             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4298                 e_err("Jumbo Frames not supported.\n");
4299                 return -EINVAL;
4300         }
4301
4302         /* Supported frame sizes */
4303         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4304             (max_frame > adapter->max_hw_frame_size)) {
4305                 e_err("Unsupported MTU setting\n");
4306                 return -EINVAL;
4307         }
4308
4309         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4310                 msleep(1);
4311         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
4312         adapter->max_frame_size = max_frame;
4313         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4314         netdev->mtu = new_mtu;
4315         if (netif_running(netdev))
4316                 e1000e_down(adapter);
4317
4318         /*
4319          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4320          * means we reserve 2 more, this pushes us to allocate from the next
4321          * larger slab size.
4322          * i.e. RXBUFFER_2048 --> size-4096 slab
4323          * However with the new *_jumbo_rx* routines, jumbo receives will use
4324          * fragmented skbs
4325          */
4326
4327         if (max_frame <= 256)
4328                 adapter->rx_buffer_len = 256;
4329         else if (max_frame <= 512)
4330                 adapter->rx_buffer_len = 512;
4331         else if (max_frame <= 1024)
4332                 adapter->rx_buffer_len = 1024;
4333         else if (max_frame <= 2048)
4334                 adapter->rx_buffer_len = 2048;
4335         else
4336                 adapter->rx_buffer_len = 4096;
4337
4338         /* adjust allocation if LPE protects us, and we aren't using SBP */
4339         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4340              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4341                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4342                                          + ETH_FCS_LEN;
4343
4344         if (netif_running(netdev))
4345                 e1000e_up(adapter);
4346         else
4347                 e1000e_reset(adapter);
4348
4349         clear_bit(__E1000_RESETTING, &adapter->state);
4350
4351         return 0;
4352 }
4353
4354 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4355                            int cmd)
4356 {
4357         struct e1000_adapter *adapter = netdev_priv(netdev);
4358         struct mii_ioctl_data *data = if_mii(ifr);
4359
4360         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4361                 return -EOPNOTSUPP;
4362
4363         switch (cmd) {
4364         case SIOCGMIIPHY:
4365                 data->phy_id = adapter->hw.phy.addr;
4366                 break;
4367         case SIOCGMIIREG:
4368                 e1000_phy_read_status(adapter);
4369
4370                 switch (data->reg_num & 0x1F) {
4371                 case MII_BMCR:
4372                         data->val_out = adapter->phy_regs.bmcr;
4373                         break;
4374                 case MII_BMSR:
4375                         data->val_out = adapter->phy_regs.bmsr;
4376                         break;
4377                 case MII_PHYSID1:
4378                         data->val_out = (adapter->hw.phy.id >> 16);
4379                         break;
4380                 case MII_PHYSID2:
4381                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4382                         break;
4383                 case MII_ADVERTISE:
4384                         data->val_out = adapter->phy_regs.advertise;
4385                         break;
4386                 case MII_LPA:
4387                         data->val_out = adapter->phy_regs.lpa;
4388                         break;
4389                 case MII_EXPANSION:
4390                         data->val_out = adapter->phy_regs.expansion;
4391                         break;
4392                 case MII_CTRL1000:
4393                         data->val_out = adapter->phy_regs.ctrl1000;
4394                         break;
4395                 case MII_STAT1000:
4396                         data->val_out = adapter->phy_regs.stat1000;
4397                         break;
4398                 case MII_ESTATUS:
4399                         data->val_out = adapter->phy_regs.estatus;
4400                         break;
4401                 default:
4402                         return -EIO;
4403                 }
4404                 break;
4405         case SIOCSMIIREG:
4406         default:
4407                 return -EOPNOTSUPP;
4408         }
4409         return 0;
4410 }
4411
4412 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4413 {
4414         switch (cmd) {
4415         case SIOCGMIIPHY:
4416         case SIOCGMIIREG:
4417         case SIOCSMIIREG:
4418                 return e1000_mii_ioctl(netdev, ifr, cmd);
4419         default:
4420                 return -EOPNOTSUPP;
4421         }
4422 }
4423
4424 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
4425 {
4426         struct e1000_hw *hw = &adapter->hw;
4427         u32 i, mac_reg;
4428         u16 phy_reg;
4429         int retval = 0;
4430
4431         /* copy MAC RARs to PHY RARs */
4432         for (i = 0; i < adapter->hw.mac.rar_entry_count; i++) {
4433                 mac_reg = er32(RAL(i));
4434                 e1e_wphy(hw, BM_RAR_L(i), (u16)(mac_reg & 0xFFFF));
4435                 e1e_wphy(hw, BM_RAR_M(i), (u16)((mac_reg >> 16) & 0xFFFF));
4436                 mac_reg = er32(RAH(i));
4437                 e1e_wphy(hw, BM_RAR_H(i), (u16)(mac_reg & 0xFFFF));
4438                 e1e_wphy(hw, BM_RAR_CTRL(i), (u16)((mac_reg >> 16) & 0xFFFF));
4439         }
4440
4441         /* copy MAC MTA to PHY MTA */
4442         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
4443                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
4444                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
4445                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
4446         }
4447
4448         /* configure PHY Rx Control register */
4449         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
4450         mac_reg = er32(RCTL);
4451         if (mac_reg & E1000_RCTL_UPE)
4452                 phy_reg |= BM_RCTL_UPE;
4453         if (mac_reg & E1000_RCTL_MPE)
4454                 phy_reg |= BM_RCTL_MPE;
4455         phy_reg &= ~(BM_RCTL_MO_MASK);
4456         if (mac_reg & E1000_RCTL_MO_3)
4457                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
4458                                 << BM_RCTL_MO_SHIFT);
4459         if (mac_reg & E1000_RCTL_BAM)
4460                 phy_reg |= BM_RCTL_BAM;
4461         if (mac_reg & E1000_RCTL_PMCF)
4462                 phy_reg |= BM_RCTL_PMCF;
4463         mac_reg = er32(CTRL);
4464         if (mac_reg & E1000_CTRL_RFCE)
4465                 phy_reg |= BM_RCTL_RFCE;
4466         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
4467
4468         /* enable PHY wakeup in MAC register */
4469         ew32(WUFC, wufc);
4470         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
4471
4472         /* configure and enable PHY wakeup in PHY registers */
4473         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
4474         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
4475
4476         /* activate PHY wakeup */
4477         retval = hw->phy.ops.acquire(hw);
4478         if (retval) {
4479                 e_err("Could not acquire PHY\n");
4480                 return retval;
4481         }
4482         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4483                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
4484         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
4485         if (retval) {
4486                 e_err("Could not read PHY page 769\n");
4487                 goto out;
4488         }
4489         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
4490         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
4491         if (retval)
4492                 e_err("Could not set PHY Host Wakeup bit\n");
4493 out:
4494         hw->phy.ops.release(hw);
4495
4496         return retval;
4497 }
4498
4499 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4500 {
4501         struct net_device *netdev = pci_get_drvdata(pdev);
4502         struct e1000_adapter *adapter = netdev_priv(netdev);
4503         struct e1000_hw *hw = &adapter->hw;
4504         u32 ctrl, ctrl_ext, rctl, status;
4505         u32 wufc = adapter->wol;
4506         int retval = 0;
4507
4508         netif_device_detach(netdev);
4509
4510         if (netif_running(netdev)) {
4511                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4512                 e1000e_down(adapter);
4513                 e1000_free_irq(adapter);
4514         }
4515         e1000e_reset_interrupt_capability(adapter);
4516
4517         retval = pci_save_state(pdev);
4518         if (retval)
4519                 return retval;
4520
4521         status = er32(STATUS);
4522         if (status & E1000_STATUS_LU)
4523                 wufc &= ~E1000_WUFC_LNKC;
4524
4525         if (wufc) {
4526                 e1000_setup_rctl(adapter);
4527                 e1000_set_multi(netdev);
4528
4529                 /* turn on all-multi mode if wake on multicast is enabled */
4530                 if (wufc & E1000_WUFC_MC) {
4531                         rctl = er32(RCTL);
4532                         rctl |= E1000_RCTL_MPE;
4533                         ew32(RCTL, rctl);
4534                 }
4535
4536                 ctrl = er32(CTRL);
4537                 /* advertise wake from D3Cold */
4538                 #define E1000_CTRL_ADVD3WUC 0x00100000
4539                 /* phy power management enable */
4540                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4541                 ctrl |= E1000_CTRL_ADVD3WUC;
4542                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
4543                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
4544                 ew32(CTRL, ctrl);
4545
4546                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4547                     adapter->hw.phy.media_type ==
4548                     e1000_media_type_internal_serdes) {
4549                         /* keep the laser running in D3 */
4550                         ctrl_ext = er32(CTRL_EXT);
4551                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
4552                         ew32(CTRL_EXT, ctrl_ext);
4553                 }
4554
4555                 if (adapter->flags & FLAG_IS_ICH)
4556                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4557
4558                 /* Allow time for pending master requests to run */
4559                 e1000e_disable_pcie_master(&adapter->hw);
4560
4561                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4562                         /* enable wakeup by the PHY */
4563                         retval = e1000_init_phy_wakeup(adapter, wufc);
4564                         if (retval)
4565                                 return retval;
4566                 } else {
4567                         /* enable wakeup by the MAC */
4568                         ew32(WUFC, wufc);
4569                         ew32(WUC, E1000_WUC_PME_EN);
4570                 }
4571         } else {
4572                 ew32(WUC, 0);
4573                 ew32(WUFC, 0);
4574         }
4575
4576         *enable_wake = !!wufc;
4577
4578         /* make sure adapter isn't asleep if manageability is enabled */
4579         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
4580             (hw->mac.ops.check_mng_mode(hw)))
4581                 *enable_wake = true;
4582
4583         if (adapter->hw.phy.type == e1000_phy_igp_3)
4584                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4585
4586         /*
4587          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4588          * would have already happened in close and is redundant.
4589          */
4590         e1000_release_hw_control(adapter);
4591
4592         pci_disable_device(pdev);
4593
4594         return 0;
4595 }
4596
4597 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
4598 {
4599         if (sleep && wake) {
4600                 pci_prepare_to_sleep(pdev);
4601                 return;
4602         }
4603
4604         pci_wake_from_d3(pdev, wake);
4605         pci_set_power_state(pdev, PCI_D3hot);
4606 }
4607
4608 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
4609                                     bool wake)
4610 {
4611         struct net_device *netdev = pci_get_drvdata(pdev);
4612         struct e1000_adapter *adapter = netdev_priv(netdev);
4613
4614         /*
4615          * The pci-e switch on some quad port adapters will report a
4616          * correctable error when the MAC transitions from D0 to D3.  To
4617          * prevent this we need to mask off the correctable errors on the
4618          * downstream port of the pci-e switch.
4619          */
4620         if (adapter->flags & FLAG_IS_QUAD_PORT) {
4621                 struct pci_dev *us_dev = pdev->bus->self;
4622                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
4623                 u16 devctl;
4624
4625                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
4626                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
4627                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
4628
4629                 e1000_power_off(pdev, sleep, wake);
4630
4631                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
4632         } else {
4633                 e1000_power_off(pdev, sleep, wake);
4634         }
4635 }
4636
4637 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4638 {
4639         int pos;
4640         u16 val;
4641
4642         /*
4643          * 82573 workaround - disable L1 ASPM on mobile chipsets
4644          *
4645          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4646          * resulting in lost data or garbage information on the pci-e link
4647          * level. This could result in (false) bad EEPROM checksum errors,
4648          * long ping times (up to 2s) or even a system freeze/hang.
4649          *
4650          * Unfortunately this feature saves about 1W power consumption when
4651          * active.
4652          */
4653         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4654         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4655         if (val & 0x2) {
4656                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4657                 val &= ~0x2;
4658                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4659         }
4660 }
4661
4662 #ifdef CONFIG_PM
4663 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4664 {
4665         int retval;
4666         bool wake;
4667
4668         retval = __e1000_shutdown(pdev, &wake);
4669         if (!retval)
4670                 e1000_complete_shutdown(pdev, true, wake);
4671
4672         return retval;
4673 }
4674
4675 static int e1000_resume(struct pci_dev *pdev)
4676 {
4677         struct net_device *netdev = pci_get_drvdata(pdev);
4678         struct e1000_adapter *adapter = netdev_priv(netdev);
4679         struct e1000_hw *hw = &adapter->hw;
4680         u32 err;
4681
4682         pci_set_power_state(pdev, PCI_D0);
4683         pci_restore_state(pdev);
4684         pci_save_state(pdev);
4685         e1000e_disable_l1aspm(pdev);
4686
4687         err = pci_enable_device_mem(pdev);
4688         if (err) {
4689                 dev_err(&pdev->dev,
4690                         "Cannot enable PCI device from suspend\n");
4691                 return err;
4692         }
4693
4694         pci_set_master(pdev);
4695
4696         pci_enable_wake(pdev, PCI_D3hot, 0);
4697         pci_enable_wake(pdev, PCI_D3cold, 0);
4698
4699         e1000e_set_interrupt_capability(adapter);
4700         if (netif_running(netdev)) {
4701                 err = e1000_request_irq(adapter);
4702                 if (err)
4703                         return err;
4704         }
4705
4706         e1000e_power_up_phy(adapter);
4707
4708         /* report the system wakeup cause from S3/S4 */
4709         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4710                 u16 phy_data;
4711
4712                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
4713                 if (phy_data) {
4714                         e_info("PHY Wakeup cause - %s\n",
4715                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
4716                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
4717                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
4718                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
4719                                 phy_data & E1000_WUS_LNKC ? "Link Status "
4720                                 " Change" : "other");
4721                 }
4722                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
4723         } else {
4724                 u32 wus = er32(WUS);
4725                 if (wus) {
4726                         e_info("MAC Wakeup cause - %s\n",
4727                                 wus & E1000_WUS_EX ? "Unicast Packet" :
4728                                 wus & E1000_WUS_MC ? "Multicast Packet" :
4729                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
4730                                 wus & E1000_WUS_MAG ? "Magic Packet" :
4731                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
4732                                 "other");
4733                 }
4734                 ew32(WUS, ~0);
4735         }
4736
4737         e1000e_reset(adapter);
4738
4739         e1000_init_manageability(adapter);
4740
4741         if (netif_running(netdev))
4742                 e1000e_up(adapter);
4743
4744         netif_device_attach(netdev);
4745
4746         /*
4747          * If the controller has AMT, do not set DRV_LOAD until the interface
4748          * is up.  For all other cases, let the f/w know that the h/w is now
4749          * under the control of the driver.
4750          */
4751         if (!(adapter->flags & FLAG_HAS_AMT))
4752                 e1000_get_hw_control(adapter);
4753
4754         return 0;
4755 }
4756 #endif
4757
4758 static void e1000_shutdown(struct pci_dev *pdev)
4759 {
4760         bool wake = false;
4761
4762         __e1000_shutdown(pdev, &wake);
4763
4764         if (system_state == SYSTEM_POWER_OFF)
4765                 e1000_complete_shutdown(pdev, false, wake);
4766 }
4767
4768 #ifdef CONFIG_NET_POLL_CONTROLLER
4769 /*
4770  * Polling 'interrupt' - used by things like netconsole to send skbs
4771  * without having to re-enable interrupts. It's not called while
4772  * the interrupt routine is executing.
4773  */
4774 static void e1000_netpoll(struct net_device *netdev)
4775 {
4776         struct e1000_adapter *adapter = netdev_priv(netdev);
4777
4778         disable_irq(adapter->pdev->irq);
4779         e1000_intr(adapter->pdev->irq, netdev);
4780
4781         enable_irq(adapter->pdev->irq);
4782 }
4783 #endif
4784
4785 /**
4786  * e1000_io_error_detected - called when PCI error is detected
4787  * @pdev: Pointer to PCI device
4788  * @state: The current pci connection state
4789  *
4790  * This function is called after a PCI bus error affecting
4791  * this device has been detected.
4792  */
4793 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4794                                                 pci_channel_state_t state)
4795 {
4796         struct net_device *netdev = pci_get_drvdata(pdev);
4797         struct e1000_adapter *adapter = netdev_priv(netdev);
4798
4799         netif_device_detach(netdev);
4800
4801         if (state == pci_channel_io_perm_failure)
4802                 return PCI_ERS_RESULT_DISCONNECT;
4803
4804         if (netif_running(netdev))
4805                 e1000e_down(adapter);
4806         pci_disable_device(pdev);
4807
4808         /* Request a slot slot reset. */
4809         return PCI_ERS_RESULT_NEED_RESET;
4810 }
4811
4812 /**
4813  * e1000_io_slot_reset - called after the pci bus has been reset.
4814  * @pdev: Pointer to PCI device
4815  *
4816  * Restart the card from scratch, as if from a cold-boot. Implementation
4817  * resembles the first-half of the e1000_resume routine.
4818  */
4819 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4820 {
4821         struct net_device *netdev = pci_get_drvdata(pdev);
4822         struct e1000_adapter *adapter = netdev_priv(netdev);
4823         struct e1000_hw *hw = &adapter->hw;
4824         int err;
4825         pci_ers_result_t result;
4826
4827         e1000e_disable_l1aspm(pdev);
4828         err = pci_enable_device_mem(pdev);
4829         if (err) {
4830                 dev_err(&pdev->dev,
4831                         "Cannot re-enable PCI device after reset.\n");
4832                 result = PCI_ERS_RESULT_DISCONNECT;
4833         } else {
4834                 pci_set_master(pdev);
4835                 pci_restore_state(pdev);
4836                 pci_save_state(pdev);
4837
4838                 pci_enable_wake(pdev, PCI_D3hot, 0);
4839                 pci_enable_wake(pdev, PCI_D3cold, 0);
4840
4841                 e1000e_reset(adapter);
4842                 ew32(WUS, ~0);
4843                 result = PCI_ERS_RESULT_RECOVERED;
4844         }
4845
4846         pci_cleanup_aer_uncorrect_error_status(pdev);
4847
4848         return result;
4849 }
4850
4851 /**
4852  * e1000_io_resume - called when traffic can start flowing again.
4853  * @pdev: Pointer to PCI device
4854  *
4855  * This callback is called when the error recovery driver tells us that
4856  * its OK to resume normal operation. Implementation resembles the
4857  * second-half of the e1000_resume routine.
4858  */
4859 static void e1000_io_resume(struct pci_dev *pdev)
4860 {
4861         struct net_device *netdev = pci_get_drvdata(pdev);
4862         struct e1000_adapter *adapter = netdev_priv(netdev);
4863
4864         e1000_init_manageability(adapter);
4865
4866         if (netif_running(netdev)) {
4867                 if (e1000e_up(adapter)) {
4868                         dev_err(&pdev->dev,
4869                                 "can't bring device back up after reset\n");
4870                         return;
4871                 }
4872         }
4873
4874         netif_device_attach(netdev);
4875
4876         /*
4877          * If the controller has AMT, do not set DRV_LOAD until the interface
4878          * is up.  For all other cases, let the f/w know that the h/w is now
4879          * under the control of the driver.
4880          */
4881         if (!(adapter->flags & FLAG_HAS_AMT))
4882                 e1000_get_hw_control(adapter);
4883
4884 }
4885
4886 static void e1000_print_device_info(struct e1000_adapter *adapter)
4887 {
4888         struct e1000_hw *hw = &adapter->hw;
4889         struct net_device *netdev = adapter->netdev;
4890         u32 pba_num;
4891
4892         /* print bus type/speed/width info */
4893         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
4894                /* bus width */
4895                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4896                 "Width x1"),
4897                /* MAC address */
4898                netdev->dev_addr);
4899         e_info("Intel(R) PRO/%s Network Connection\n",
4900                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4901         e1000e_read_pba_num(hw, &pba_num);
4902         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4903                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4904 }
4905
4906 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4907 {
4908         struct e1000_hw *hw = &adapter->hw;
4909         int ret_val;
4910         u16 buf = 0;
4911
4912         if (hw->mac.type != e1000_82573)
4913                 return;
4914
4915         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4916         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
4917                 /* Deep Smart Power Down (DSPD) */
4918                 dev_warn(&adapter->pdev->dev,
4919                          "Warning: detected DSPD enabled in EEPROM\n");
4920         }
4921
4922         ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4923         if (!ret_val && (le16_to_cpu(buf) & (3 << 2))) {
4924                 /* ASPM enable */
4925                 dev_warn(&adapter->pdev->dev,
4926                          "Warning: detected ASPM enabled in EEPROM\n");
4927         }
4928 }
4929
4930 static const struct net_device_ops e1000e_netdev_ops = {
4931         .ndo_open               = e1000_open,
4932         .ndo_stop               = e1000_close,
4933         .ndo_start_xmit         = e1000_xmit_frame,
4934         .ndo_get_stats          = e1000_get_stats,
4935         .ndo_set_multicast_list = e1000_set_multi,
4936         .ndo_set_mac_address    = e1000_set_mac,
4937         .ndo_change_mtu         = e1000_change_mtu,
4938         .ndo_do_ioctl           = e1000_ioctl,
4939         .ndo_tx_timeout         = e1000_tx_timeout,
4940         .ndo_validate_addr      = eth_validate_addr,
4941
4942         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
4943         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
4944         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
4945 #ifdef CONFIG_NET_POLL_CONTROLLER
4946         .ndo_poll_controller    = e1000_netpoll,
4947 #endif
4948 };
4949
4950 /**
4951  * e1000_probe - Device Initialization Routine
4952  * @pdev: PCI device information struct
4953  * @ent: entry in e1000_pci_tbl
4954  *
4955  * Returns 0 on success, negative on failure
4956  *
4957  * e1000_probe initializes an adapter identified by a pci_dev structure.
4958  * The OS initialization, configuring of the adapter private structure,
4959  * and a hardware reset occur.
4960  **/
4961 static int __devinit e1000_probe(struct pci_dev *pdev,
4962                                  const struct pci_device_id *ent)
4963 {
4964         struct net_device *netdev;
4965         struct e1000_adapter *adapter;
4966         struct e1000_hw *hw;
4967         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4968         resource_size_t mmio_start, mmio_len;
4969         resource_size_t flash_start, flash_len;
4970
4971         static int cards_found;
4972         int i, err, pci_using_dac;
4973         u16 eeprom_data = 0;
4974         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4975
4976         e1000e_disable_l1aspm(pdev);
4977
4978         err = pci_enable_device_mem(pdev);
4979         if (err)
4980                 return err;
4981
4982         pci_using_dac = 0;
4983         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
4984         if (!err) {
4985                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
4986                 if (!err)
4987                         pci_using_dac = 1;
4988         } else {
4989                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
4990                 if (err) {
4991                         err = pci_set_consistent_dma_mask(pdev,
4992                                                           DMA_BIT_MASK(32));
4993                         if (err) {
4994                                 dev_err(&pdev->dev, "No usable DMA "
4995                                         "configuration, aborting\n");
4996                                 goto err_dma;
4997                         }
4998                 }
4999         }
5000
5001         err = pci_request_selected_regions_exclusive(pdev,
5002                                           pci_select_bars(pdev, IORESOURCE_MEM),
5003                                           e1000e_driver_name);
5004         if (err)
5005                 goto err_pci_reg;
5006
5007         /* AER (Advanced Error Reporting) hooks */
5008         pci_enable_pcie_error_reporting(pdev);
5009
5010         pci_set_master(pdev);
5011         /* PCI config space info */
5012         err = pci_save_state(pdev);
5013         if (err)
5014                 goto err_alloc_etherdev;
5015
5016         err = -ENOMEM;
5017         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5018         if (!netdev)
5019                 goto err_alloc_etherdev;
5020
5021         SET_NETDEV_DEV(netdev, &pdev->dev);
5022
5023         pci_set_drvdata(pdev, netdev);
5024         adapter = netdev_priv(netdev);
5025         hw = &adapter->hw;
5026         adapter->netdev = netdev;
5027         adapter->pdev = pdev;
5028         adapter->ei = ei;
5029         adapter->pba = ei->pba;
5030         adapter->flags = ei->flags;
5031         adapter->flags2 = ei->flags2;
5032         adapter->hw.adapter = adapter;
5033         adapter->hw.mac.type = ei->mac;
5034         adapter->max_hw_frame_size = ei->max_hw_frame_size;
5035         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5036
5037         mmio_start = pci_resource_start(pdev, 0);
5038         mmio_len = pci_resource_len(pdev, 0);
5039
5040         err = -EIO;
5041         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5042         if (!adapter->hw.hw_addr)
5043                 goto err_ioremap;
5044
5045         if ((adapter->flags & FLAG_HAS_FLASH) &&
5046             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5047                 flash_start = pci_resource_start(pdev, 1);
5048                 flash_len = pci_resource_len(pdev, 1);
5049                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5050                 if (!adapter->hw.flash_address)
5051                         goto err_flashmap;
5052         }
5053
5054         /* construct the net_device struct */
5055         netdev->netdev_ops              = &e1000e_netdev_ops;
5056         e1000e_set_ethtool_ops(netdev);
5057         netdev->watchdog_timeo          = 5 * HZ;
5058         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5059         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5060
5061         netdev->mem_start = mmio_start;
5062         netdev->mem_end = mmio_start + mmio_len;
5063
5064         adapter->bd_number = cards_found++;
5065
5066         e1000e_check_options(adapter);
5067
5068         /* setup adapter struct */
5069         err = e1000_sw_init(adapter);
5070         if (err)
5071                 goto err_sw_init;
5072
5073         err = -EIO;
5074
5075         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5076         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5077         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5078
5079         err = ei->get_variants(adapter);
5080         if (err)
5081                 goto err_hw_init;
5082
5083         if ((adapter->flags & FLAG_IS_ICH) &&
5084             (adapter->flags & FLAG_READ_ONLY_NVM))
5085                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5086
5087         hw->mac.ops.get_bus_info(&adapter->hw);
5088
5089         adapter->hw.phy.autoneg_wait_to_complete = 0;
5090
5091         /* Copper options */
5092         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5093                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5094                 adapter->hw.phy.disable_polarity_correction = 0;
5095                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5096         }
5097
5098         if (e1000_check_reset_block(&adapter->hw))
5099                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5100
5101         netdev->features = NETIF_F_SG |
5102                            NETIF_F_HW_CSUM |
5103                            NETIF_F_HW_VLAN_TX |
5104                            NETIF_F_HW_VLAN_RX;
5105
5106         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5107                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5108
5109         netdev->features |= NETIF_F_TSO;
5110         netdev->features |= NETIF_F_TSO6;
5111
5112         netdev->vlan_features |= NETIF_F_TSO;
5113         netdev->vlan_features |= NETIF_F_TSO6;
5114         netdev->vlan_features |= NETIF_F_HW_CSUM;
5115         netdev->vlan_features |= NETIF_F_SG;
5116
5117         if (pci_using_dac)
5118                 netdev->features |= NETIF_F_HIGHDMA;
5119
5120         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5121                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5122
5123         /*
5124          * before reading the NVM, reset the controller to
5125          * put the device in a known good starting state
5126          */
5127         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5128
5129         /*
5130          * systems with ASPM and others may see the checksum fail on the first
5131          * attempt. Let's give it a few tries
5132          */
5133         for (i = 0;; i++) {
5134                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5135                         break;
5136                 if (i == 2) {
5137                         e_err("The NVM Checksum Is Not Valid\n");
5138                         err = -EIO;
5139                         goto err_eeprom;
5140                 }
5141         }
5142
5143         e1000_eeprom_checks(adapter);
5144
5145         /* copy the MAC address */
5146         if (e1000e_read_mac_addr(&adapter->hw))
5147                 e_err("NVM Read Error while reading MAC address\n");
5148
5149         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5150         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5151
5152         if (!is_valid_ether_addr(netdev->perm_addr)) {
5153                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5154                 err = -EIO;
5155                 goto err_eeprom;
5156         }
5157
5158         init_timer(&adapter->watchdog_timer);
5159         adapter->watchdog_timer.function = &e1000_watchdog;
5160         adapter->watchdog_timer.data = (unsigned long) adapter;
5161
5162         init_timer(&adapter->phy_info_timer);
5163         adapter->phy_info_timer.function = &e1000_update_phy_info;
5164         adapter->phy_info_timer.data = (unsigned long) adapter;
5165
5166         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5167         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5168         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5169         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5170         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5171
5172         /* Initialize link parameters. User can change them with ethtool */
5173         adapter->hw.mac.autoneg = 1;
5174         adapter->fc_autoneg = 1;
5175         adapter->hw.fc.requested_mode = e1000_fc_default;
5176         adapter->hw.fc.current_mode = e1000_fc_default;
5177         adapter->hw.phy.autoneg_advertised = 0x2f;
5178
5179         /* ring size defaults */
5180         adapter->rx_ring->count = 256;
5181         adapter->tx_ring->count = 256;
5182
5183         /*
5184          * Initial Wake on LAN setting - If APM wake is enabled in
5185          * the EEPROM, enable the ACPI Magic Packet filter
5186          */
5187         if (adapter->flags & FLAG_APME_IN_WUC) {
5188                 /* APME bit in EEPROM is mapped to WUC.APME */
5189                 eeprom_data = er32(WUC);
5190                 eeprom_apme_mask = E1000_WUC_APME;
5191                 if (eeprom_data & E1000_WUC_PHY_WAKE)
5192                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5193         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5194                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5195                     (adapter->hw.bus.func == 1))
5196                         e1000_read_nvm(&adapter->hw,
5197                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5198                 else
5199                         e1000_read_nvm(&adapter->hw,
5200                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5201         }
5202
5203         /* fetch WoL from EEPROM */
5204         if (eeprom_data & eeprom_apme_mask)
5205                 adapter->eeprom_wol |= E1000_WUFC_MAG;
5206
5207         /*
5208          * now that we have the eeprom settings, apply the special cases
5209          * where the eeprom may be wrong or the board simply won't support
5210          * wake on lan on a particular port
5211          */
5212         if (!(adapter->flags & FLAG_HAS_WOL))
5213                 adapter->eeprom_wol = 0;
5214
5215         /* initialize the wol settings based on the eeprom settings */
5216         adapter->wol = adapter->eeprom_wol;
5217         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5218
5219         /* save off EEPROM version number */
5220         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5221
5222         /* reset the hardware with the new settings */
5223         e1000e_reset(adapter);
5224
5225         /*
5226          * If the controller has AMT, do not set DRV_LOAD until the interface
5227          * is up.  For all other cases, let the f/w know that the h/w is now
5228          * under the control of the driver.
5229          */
5230         if (!(adapter->flags & FLAG_HAS_AMT))
5231                 e1000_get_hw_control(adapter);
5232
5233         strcpy(netdev->name, "eth%d");
5234         err = register_netdev(netdev);
5235         if (err)
5236                 goto err_register;
5237
5238         /* carrier off reporting is important to ethtool even BEFORE open */
5239         netif_carrier_off(netdev);
5240
5241         e1000_print_device_info(adapter);
5242
5243         return 0;
5244
5245 err_register:
5246         if (!(adapter->flags & FLAG_HAS_AMT))
5247                 e1000_release_hw_control(adapter);
5248 err_eeprom:
5249         if (!e1000_check_reset_block(&adapter->hw))
5250                 e1000_phy_hw_reset(&adapter->hw);
5251 err_hw_init:
5252
5253         kfree(adapter->tx_ring);
5254         kfree(adapter->rx_ring);
5255 err_sw_init:
5256         if (adapter->hw.flash_address)
5257                 iounmap(adapter->hw.flash_address);
5258         e1000e_reset_interrupt_capability(adapter);
5259 err_flashmap:
5260         iounmap(adapter->hw.hw_addr);
5261 err_ioremap:
5262         free_netdev(netdev);
5263 err_alloc_etherdev:
5264         pci_release_selected_regions(pdev,
5265                                      pci_select_bars(pdev, IORESOURCE_MEM));
5266 err_pci_reg:
5267 err_dma:
5268         pci_disable_device(pdev);
5269         return err;
5270 }
5271
5272 /**
5273  * e1000_remove - Device Removal Routine
5274  * @pdev: PCI device information struct
5275  *
5276  * e1000_remove is called by the PCI subsystem to alert the driver
5277  * that it should release a PCI device.  The could be caused by a
5278  * Hot-Plug event, or because the driver is going to be removed from
5279  * memory.
5280  **/
5281 static void __devexit e1000_remove(struct pci_dev *pdev)
5282 {
5283         struct net_device *netdev = pci_get_drvdata(pdev);
5284         struct e1000_adapter *adapter = netdev_priv(netdev);
5285
5286         /*
5287          * flush_scheduled work may reschedule our watchdog task, so
5288          * explicitly disable watchdog tasks from being rescheduled
5289          */
5290         set_bit(__E1000_DOWN, &adapter->state);
5291         del_timer_sync(&adapter->watchdog_timer);
5292         del_timer_sync(&adapter->phy_info_timer);
5293
5294         cancel_work_sync(&adapter->reset_task);
5295         cancel_work_sync(&adapter->watchdog_task);
5296         cancel_work_sync(&adapter->downshift_task);
5297         cancel_work_sync(&adapter->update_phy_task);
5298         cancel_work_sync(&adapter->print_hang_task);
5299         flush_scheduled_work();
5300
5301         if (!(netdev->flags & IFF_UP))
5302                 e1000_power_down_phy(adapter);
5303
5304         unregister_netdev(netdev);
5305
5306         /*
5307          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5308          * would have already happened in close and is redundant.
5309          */
5310         e1000_release_hw_control(adapter);
5311
5312         e1000e_reset_interrupt_capability(adapter);
5313         kfree(adapter->tx_ring);
5314         kfree(adapter->rx_ring);
5315
5316         iounmap(adapter->hw.hw_addr);
5317         if (adapter->hw.flash_address)
5318                 iounmap(adapter->hw.flash_address);
5319         pci_release_selected_regions(pdev,
5320                                      pci_select_bars(pdev, IORESOURCE_MEM));
5321
5322         free_netdev(netdev);
5323
5324         /* AER disable */
5325         pci_disable_pcie_error_reporting(pdev);
5326
5327         pci_disable_device(pdev);
5328 }
5329
5330 /* PCI Error Recovery (ERS) */
5331 static struct pci_error_handlers e1000_err_handler = {
5332         .error_detected = e1000_io_error_detected,
5333         .slot_reset = e1000_io_slot_reset,
5334         .resume = e1000_io_resume,
5335 };
5336
5337 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
5338         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5339         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5340         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5341         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5342         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5343         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5344         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5345         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5346         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5347
5348         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5349         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5350         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5351         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5352
5353         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5354         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5355         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5356
5357         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5358         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
5359         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
5360
5361         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5362           board_80003es2lan },
5363         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5364           board_80003es2lan },
5365         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5366           board_80003es2lan },
5367         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5368           board_80003es2lan },
5369
5370         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5371         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5372         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5373         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5374         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5375         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5376         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5377         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
5378
5379         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5380         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5381         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5382         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5383         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5384         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5385         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5386         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5387         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5388
5389         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5390         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5391         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5392
5393         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5394         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5395
5396         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
5397         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
5398         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
5399         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
5400
5401         { }     /* terminate list */
5402 };
5403 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5404
5405 /* PCI Device API Driver */
5406 static struct pci_driver e1000_driver = {
5407         .name     = e1000e_driver_name,
5408         .id_table = e1000_pci_tbl,
5409         .probe    = e1000_probe,
5410         .remove   = __devexit_p(e1000_remove),
5411 #ifdef CONFIG_PM
5412         /* Power Management Hooks */
5413         .suspend  = e1000_suspend,
5414         .resume   = e1000_resume,
5415 #endif
5416         .shutdown = e1000_shutdown,
5417         .err_handler = &e1000_err_handler
5418 };
5419
5420 /**
5421  * e1000_init_module - Driver Registration Routine
5422  *
5423  * e1000_init_module is the first routine called when the driver is
5424  * loaded. All it does is register with the PCI subsystem.
5425  **/
5426 static int __init e1000_init_module(void)
5427 {
5428         int ret;
5429         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
5430                e1000e_driver_name, e1000e_driver_version);
5431         printk(KERN_INFO "%s: Copyright (c) 1999 - 2009 Intel Corporation.\n",
5432                e1000e_driver_name);
5433         ret = pci_register_driver(&e1000_driver);
5434
5435         return ret;
5436 }
5437 module_init(e1000_init_module);
5438
5439 /**
5440  * e1000_exit_module - Driver Exit Cleanup Routine
5441  *
5442  * e1000_exit_module is called just before the driver is removed
5443  * from memory.
5444  **/
5445 static void __exit e1000_exit_module(void)
5446 {
5447         pci_unregister_driver(&e1000_driver);
5448 }
5449 module_exit(e1000_exit_module);
5450
5451
5452 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5453 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5454 MODULE_LICENSE("GPL");
5455 MODULE_VERSION(DRV_VERSION);
5456
5457 /* e1000_main.c */