e1000e: add support for 82583 device id
[linux-2.6.git] / drivers / net / e1000e / netdev.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2008 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 "0.3.3.4-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 };
66
67 #ifdef DEBUG
68 /**
69  * e1000_get_hw_dev_name - return device name string
70  * used by hardware layer to print debugging information
71  **/
72 char *e1000e_get_hw_dev_name(struct e1000_hw *hw)
73 {
74         return hw->adapter->netdev->name;
75 }
76 #endif
77
78 /**
79  * e1000_desc_unused - calculate if we have unused descriptors
80  **/
81 static int e1000_desc_unused(struct e1000_ring *ring)
82 {
83         if (ring->next_to_clean > ring->next_to_use)
84                 return ring->next_to_clean - ring->next_to_use - 1;
85
86         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
87 }
88
89 /**
90  * e1000_receive_skb - helper function to handle Rx indications
91  * @adapter: board private structure
92  * @status: descriptor status field as written by hardware
93  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
94  * @skb: pointer to sk_buff to be indicated to stack
95  **/
96 static void e1000_receive_skb(struct e1000_adapter *adapter,
97                               struct net_device *netdev,
98                               struct sk_buff *skb,
99                               u8 status, __le16 vlan)
100 {
101         skb->protocol = eth_type_trans(skb, netdev);
102
103         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
104                 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
105                                  le16_to_cpu(vlan), skb);
106         else
107                 napi_gro_receive(&adapter->napi, skb);
108 }
109
110 /**
111  * e1000_rx_checksum - Receive Checksum Offload for 82543
112  * @adapter:     board private structure
113  * @status_err:  receive descriptor status and error fields
114  * @csum:       receive descriptor csum field
115  * @sk_buff:     socket buffer with received data
116  **/
117 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
118                               u32 csum, struct sk_buff *skb)
119 {
120         u16 status = (u16)status_err;
121         u8 errors = (u8)(status_err >> 24);
122         skb->ip_summed = CHECKSUM_NONE;
123
124         /* Ignore Checksum bit is set */
125         if (status & E1000_RXD_STAT_IXSM)
126                 return;
127         /* TCP/UDP checksum error bit is set */
128         if (errors & E1000_RXD_ERR_TCPE) {
129                 /* let the stack verify checksum errors */
130                 adapter->hw_csum_err++;
131                 return;
132         }
133
134         /* TCP/UDP Checksum has not been calculated */
135         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
136                 return;
137
138         /* It must be a TCP or UDP packet with a valid checksum */
139         if (status & E1000_RXD_STAT_TCPCS) {
140                 /* TCP checksum is good */
141                 skb->ip_summed = CHECKSUM_UNNECESSARY;
142         } else {
143                 /*
144                  * IP fragment with UDP payload
145                  * Hardware complements the payload checksum, so we undo it
146                  * and then put the value in host order for further stack use.
147                  */
148                 __sum16 sum = (__force __sum16)htons(csum);
149                 skb->csum = csum_unfold(~sum);
150                 skb->ip_summed = CHECKSUM_COMPLETE;
151         }
152         adapter->hw_csum_good++;
153 }
154
155 /**
156  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
157  * @adapter: address of board private structure
158  **/
159 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
160                                    int cleaned_count)
161 {
162         struct net_device *netdev = adapter->netdev;
163         struct pci_dev *pdev = adapter->pdev;
164         struct e1000_ring *rx_ring = adapter->rx_ring;
165         struct e1000_rx_desc *rx_desc;
166         struct e1000_buffer *buffer_info;
167         struct sk_buff *skb;
168         unsigned int i;
169         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
170
171         i = rx_ring->next_to_use;
172         buffer_info = &rx_ring->buffer_info[i];
173
174         while (cleaned_count--) {
175                 skb = buffer_info->skb;
176                 if (skb) {
177                         skb_trim(skb, 0);
178                         goto map_skb;
179                 }
180
181                 skb = netdev_alloc_skb(netdev, bufsz);
182                 if (!skb) {
183                         /* Better luck next round */
184                         adapter->alloc_rx_buff_failed++;
185                         break;
186                 }
187
188                 /*
189                  * Make buffer alignment 2 beyond a 16 byte boundary
190                  * this will result in a 16 byte aligned IP header after
191                  * the 14 byte MAC header is removed
192                  */
193                 skb_reserve(skb, NET_IP_ALIGN);
194
195                 buffer_info->skb = skb;
196 map_skb:
197                 buffer_info->dma = pci_map_single(pdev, skb->data,
198                                                   adapter->rx_buffer_len,
199                                                   PCI_DMA_FROMDEVICE);
200                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
201                         dev_err(&pdev->dev, "RX DMA map failed\n");
202                         adapter->rx_dma_failed++;
203                         break;
204                 }
205
206                 rx_desc = E1000_RX_DESC(*rx_ring, i);
207                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
208
209                 i++;
210                 if (i == rx_ring->count)
211                         i = 0;
212                 buffer_info = &rx_ring->buffer_info[i];
213         }
214
215         if (rx_ring->next_to_use != i) {
216                 rx_ring->next_to_use = i;
217                 if (i-- == 0)
218                         i = (rx_ring->count - 1);
219
220                 /*
221                  * Force memory writes to complete before letting h/w
222                  * know there are new descriptors to fetch.  (Only
223                  * applicable for weak-ordered memory model archs,
224                  * such as IA-64).
225                  */
226                 wmb();
227                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
228         }
229 }
230
231 /**
232  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
233  * @adapter: address of board private structure
234  **/
235 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
236                                       int cleaned_count)
237 {
238         struct net_device *netdev = adapter->netdev;
239         struct pci_dev *pdev = adapter->pdev;
240         union e1000_rx_desc_packet_split *rx_desc;
241         struct e1000_ring *rx_ring = adapter->rx_ring;
242         struct e1000_buffer *buffer_info;
243         struct e1000_ps_page *ps_page;
244         struct sk_buff *skb;
245         unsigned int i, j;
246
247         i = rx_ring->next_to_use;
248         buffer_info = &rx_ring->buffer_info[i];
249
250         while (cleaned_count--) {
251                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
252
253                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
254                         ps_page = &buffer_info->ps_pages[j];
255                         if (j >= adapter->rx_ps_pages) {
256                                 /* all unused desc entries get hw null ptr */
257                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
258                                 continue;
259                         }
260                         if (!ps_page->page) {
261                                 ps_page->page = alloc_page(GFP_ATOMIC);
262                                 if (!ps_page->page) {
263                                         adapter->alloc_rx_buff_failed++;
264                                         goto no_buffers;
265                                 }
266                                 ps_page->dma = pci_map_page(pdev,
267                                                    ps_page->page,
268                                                    0, PAGE_SIZE,
269                                                    PCI_DMA_FROMDEVICE);
270                                 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
271                                         dev_err(&adapter->pdev->dev,
272                                           "RX DMA page map failed\n");
273                                         adapter->rx_dma_failed++;
274                                         goto no_buffers;
275                                 }
276                         }
277                         /*
278                          * Refresh the desc even if buffer_addrs
279                          * didn't change because each write-back
280                          * erases this info.
281                          */
282                         rx_desc->read.buffer_addr[j+1] =
283                              cpu_to_le64(ps_page->dma);
284                 }
285
286                 skb = netdev_alloc_skb(netdev,
287                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
288
289                 if (!skb) {
290                         adapter->alloc_rx_buff_failed++;
291                         break;
292                 }
293
294                 /*
295                  * Make buffer alignment 2 beyond a 16 byte boundary
296                  * this will result in a 16 byte aligned IP header after
297                  * the 14 byte MAC header is removed
298                  */
299                 skb_reserve(skb, NET_IP_ALIGN);
300
301                 buffer_info->skb = skb;
302                 buffer_info->dma = pci_map_single(pdev, skb->data,
303                                                   adapter->rx_ps_bsize0,
304                                                   PCI_DMA_FROMDEVICE);
305                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
306                         dev_err(&pdev->dev, "RX DMA map failed\n");
307                         adapter->rx_dma_failed++;
308                         /* cleanup skb */
309                         dev_kfree_skb_any(skb);
310                         buffer_info->skb = NULL;
311                         break;
312                 }
313
314                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
315
316                 i++;
317                 if (i == rx_ring->count)
318                         i = 0;
319                 buffer_info = &rx_ring->buffer_info[i];
320         }
321
322 no_buffers:
323         if (rx_ring->next_to_use != i) {
324                 rx_ring->next_to_use = i;
325
326                 if (!(i--))
327                         i = (rx_ring->count - 1);
328
329                 /*
330                  * Force memory writes to complete before letting h/w
331                  * know there are new descriptors to fetch.  (Only
332                  * applicable for weak-ordered memory model archs,
333                  * such as IA-64).
334                  */
335                 wmb();
336                 /*
337                  * Hardware increments by 16 bytes, but packet split
338                  * descriptors are 32 bytes...so we increment tail
339                  * twice as much.
340                  */
341                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
342         }
343 }
344
345 /**
346  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
347  * @adapter: address of board private structure
348  * @cleaned_count: number of buffers to allocate this pass
349  **/
350
351 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
352                                          int cleaned_count)
353 {
354         struct net_device *netdev = adapter->netdev;
355         struct pci_dev *pdev = adapter->pdev;
356         struct e1000_rx_desc *rx_desc;
357         struct e1000_ring *rx_ring = adapter->rx_ring;
358         struct e1000_buffer *buffer_info;
359         struct sk_buff *skb;
360         unsigned int i;
361         unsigned int bufsz = 256 -
362                              16 /* for skb_reserve */ -
363                              NET_IP_ALIGN;
364
365         i = rx_ring->next_to_use;
366         buffer_info = &rx_ring->buffer_info[i];
367
368         while (cleaned_count--) {
369                 skb = buffer_info->skb;
370                 if (skb) {
371                         skb_trim(skb, 0);
372                         goto check_page;
373                 }
374
375                 skb = netdev_alloc_skb(netdev, bufsz);
376                 if (unlikely(!skb)) {
377                         /* Better luck next round */
378                         adapter->alloc_rx_buff_failed++;
379                         break;
380                 }
381
382                 /* Make buffer alignment 2 beyond a 16 byte boundary
383                  * this will result in a 16 byte aligned IP header after
384                  * the 14 byte MAC header is removed
385                  */
386                 skb_reserve(skb, NET_IP_ALIGN);
387
388                 buffer_info->skb = skb;
389 check_page:
390                 /* allocate a new page if necessary */
391                 if (!buffer_info->page) {
392                         buffer_info->page = alloc_page(GFP_ATOMIC);
393                         if (unlikely(!buffer_info->page)) {
394                                 adapter->alloc_rx_buff_failed++;
395                                 break;
396                         }
397                 }
398
399                 if (!buffer_info->dma)
400                         buffer_info->dma = pci_map_page(pdev,
401                                                         buffer_info->page, 0,
402                                                         PAGE_SIZE,
403                                                         PCI_DMA_FROMDEVICE);
404
405                 rx_desc = E1000_RX_DESC(*rx_ring, i);
406                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
407
408                 if (unlikely(++i == rx_ring->count))
409                         i = 0;
410                 buffer_info = &rx_ring->buffer_info[i];
411         }
412
413         if (likely(rx_ring->next_to_use != i)) {
414                 rx_ring->next_to_use = i;
415                 if (unlikely(i-- == 0))
416                         i = (rx_ring->count - 1);
417
418                 /* Force memory writes to complete before letting h/w
419                  * know there are new descriptors to fetch.  (Only
420                  * applicable for weak-ordered memory model archs,
421                  * such as IA-64). */
422                 wmb();
423                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
424         }
425 }
426
427 /**
428  * e1000_clean_rx_irq - Send received data up the network stack; legacy
429  * @adapter: board private structure
430  *
431  * the return value indicates whether actual cleaning was done, there
432  * is no guarantee that everything was cleaned
433  **/
434 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
435                                int *work_done, int work_to_do)
436 {
437         struct net_device *netdev = adapter->netdev;
438         struct pci_dev *pdev = adapter->pdev;
439         struct e1000_ring *rx_ring = adapter->rx_ring;
440         struct e1000_rx_desc *rx_desc, *next_rxd;
441         struct e1000_buffer *buffer_info, *next_buffer;
442         u32 length;
443         unsigned int i;
444         int cleaned_count = 0;
445         bool cleaned = 0;
446         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
447
448         i = rx_ring->next_to_clean;
449         rx_desc = E1000_RX_DESC(*rx_ring, i);
450         buffer_info = &rx_ring->buffer_info[i];
451
452         while (rx_desc->status & E1000_RXD_STAT_DD) {
453                 struct sk_buff *skb;
454                 u8 status;
455
456                 if (*work_done >= work_to_do)
457                         break;
458                 (*work_done)++;
459
460                 status = rx_desc->status;
461                 skb = buffer_info->skb;
462                 buffer_info->skb = NULL;
463
464                 prefetch(skb->data - NET_IP_ALIGN);
465
466                 i++;
467                 if (i == rx_ring->count)
468                         i = 0;
469                 next_rxd = E1000_RX_DESC(*rx_ring, i);
470                 prefetch(next_rxd);
471
472                 next_buffer = &rx_ring->buffer_info[i];
473
474                 cleaned = 1;
475                 cleaned_count++;
476                 pci_unmap_single(pdev,
477                                  buffer_info->dma,
478                                  adapter->rx_buffer_len,
479                                  PCI_DMA_FROMDEVICE);
480                 buffer_info->dma = 0;
481
482                 length = le16_to_cpu(rx_desc->length);
483
484                 /* !EOP means multiple descriptors were used to store a single
485                  * packet, also make sure the frame isn't just CRC only */
486                 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
487                         /* All receives must fit into a single buffer */
488                         e_dbg("%s: Receive packet consumed multiple buffers\n",
489                               netdev->name);
490                         /* recycle */
491                         buffer_info->skb = skb;
492                         goto next_desc;
493                 }
494
495                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
496                         /* recycle */
497                         buffer_info->skb = skb;
498                         goto next_desc;
499                 }
500
501                 /* adjust length to remove Ethernet CRC */
502                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
503                         length -= 4;
504
505                 total_rx_bytes += length;
506                 total_rx_packets++;
507
508                 /*
509                  * code added for copybreak, this should improve
510                  * performance for small packets with large amounts
511                  * of reassembly being done in the stack
512                  */
513                 if (length < copybreak) {
514                         struct sk_buff *new_skb =
515                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
516                         if (new_skb) {
517                                 skb_reserve(new_skb, NET_IP_ALIGN);
518                                 skb_copy_to_linear_data_offset(new_skb,
519                                                                -NET_IP_ALIGN,
520                                                                (skb->data -
521                                                                 NET_IP_ALIGN),
522                                                                (length +
523                                                                 NET_IP_ALIGN));
524                                 /* save the skb in buffer_info as good */
525                                 buffer_info->skb = skb;
526                                 skb = new_skb;
527                         }
528                         /* else just continue with the old one */
529                 }
530                 /* end copybreak code */
531                 skb_put(skb, length);
532
533                 /* Receive Checksum Offload */
534                 e1000_rx_checksum(adapter,
535                                   (u32)(status) |
536                                   ((u32)(rx_desc->errors) << 24),
537                                   le16_to_cpu(rx_desc->csum), skb);
538
539                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
540
541 next_desc:
542                 rx_desc->status = 0;
543
544                 /* return some buffers to hardware, one at a time is too slow */
545                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
546                         adapter->alloc_rx_buf(adapter, cleaned_count);
547                         cleaned_count = 0;
548                 }
549
550                 /* use prefetched values */
551                 rx_desc = next_rxd;
552                 buffer_info = next_buffer;
553         }
554         rx_ring->next_to_clean = i;
555
556         cleaned_count = e1000_desc_unused(rx_ring);
557         if (cleaned_count)
558                 adapter->alloc_rx_buf(adapter, cleaned_count);
559
560         adapter->total_rx_bytes += total_rx_bytes;
561         adapter->total_rx_packets += total_rx_packets;
562         adapter->net_stats.rx_bytes += total_rx_bytes;
563         adapter->net_stats.rx_packets += total_rx_packets;
564         return cleaned;
565 }
566
567 static void e1000_put_txbuf(struct e1000_adapter *adapter,
568                              struct e1000_buffer *buffer_info)
569 {
570         buffer_info->dma = 0;
571         if (buffer_info->skb) {
572                 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
573                               DMA_TO_DEVICE);
574                 dev_kfree_skb_any(buffer_info->skb);
575                 buffer_info->skb = NULL;
576         }
577 }
578
579 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
580 {
581         struct e1000_ring *tx_ring = adapter->tx_ring;
582         unsigned int i = tx_ring->next_to_clean;
583         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
584         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
585
586         /* detected Tx unit hang */
587         e_err("Detected Tx Unit Hang:\n"
588               "  TDH                  <%x>\n"
589               "  TDT                  <%x>\n"
590               "  next_to_use          <%x>\n"
591               "  next_to_clean        <%x>\n"
592               "buffer_info[next_to_clean]:\n"
593               "  time_stamp           <%lx>\n"
594               "  next_to_watch        <%x>\n"
595               "  jiffies              <%lx>\n"
596               "  next_to_watch.status <%x>\n",
597               readl(adapter->hw.hw_addr + tx_ring->head),
598               readl(adapter->hw.hw_addr + tx_ring->tail),
599               tx_ring->next_to_use,
600               tx_ring->next_to_clean,
601               tx_ring->buffer_info[eop].time_stamp,
602               eop,
603               jiffies,
604               eop_desc->upper.fields.status);
605 }
606
607 /**
608  * e1000_clean_tx_irq - Reclaim resources after transmit completes
609  * @adapter: board private structure
610  *
611  * the return value indicates whether actual cleaning was done, there
612  * is no guarantee that everything was cleaned
613  **/
614 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
615 {
616         struct net_device *netdev = adapter->netdev;
617         struct e1000_hw *hw = &adapter->hw;
618         struct e1000_ring *tx_ring = adapter->tx_ring;
619         struct e1000_tx_desc *tx_desc, *eop_desc;
620         struct e1000_buffer *buffer_info;
621         unsigned int i, eop;
622         unsigned int count = 0;
623         bool cleaned = 0;
624         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
625
626         i = tx_ring->next_to_clean;
627         eop = tx_ring->buffer_info[i].next_to_watch;
628         eop_desc = E1000_TX_DESC(*tx_ring, eop);
629
630         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
631                 for (cleaned = 0; !cleaned; ) {
632                         tx_desc = E1000_TX_DESC(*tx_ring, i);
633                         buffer_info = &tx_ring->buffer_info[i];
634                         cleaned = (i == eop);
635
636                         if (cleaned) {
637                                 struct sk_buff *skb = buffer_info->skb;
638                                 unsigned int segs, bytecount;
639                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
640                                 /* multiply data chunks by size of headers */
641                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
642                                             skb->len;
643                                 total_tx_packets += segs;
644                                 total_tx_bytes += bytecount;
645                         }
646
647                         e1000_put_txbuf(adapter, buffer_info);
648                         tx_desc->upper.data = 0;
649
650                         i++;
651                         if (i == tx_ring->count)
652                                 i = 0;
653                 }
654
655                 eop = tx_ring->buffer_info[i].next_to_watch;
656                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
657 #define E1000_TX_WEIGHT 64
658                 /* weight of a sort for tx, to avoid endless transmit cleanup */
659                 if (count++ == E1000_TX_WEIGHT)
660                         break;
661         }
662
663         tx_ring->next_to_clean = i;
664
665 #define TX_WAKE_THRESHOLD 32
666         if (cleaned && netif_carrier_ok(netdev) &&
667                      e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
668                 /* Make sure that anybody stopping the queue after this
669                  * sees the new next_to_clean.
670                  */
671                 smp_mb();
672
673                 if (netif_queue_stopped(netdev) &&
674                     !(test_bit(__E1000_DOWN, &adapter->state))) {
675                         netif_wake_queue(netdev);
676                         ++adapter->restart_queue;
677                 }
678         }
679
680         if (adapter->detect_tx_hung) {
681                 /*
682                  * Detect a transmit hang in hardware, this serializes the
683                  * check with the clearing of time_stamp and movement of i
684                  */
685                 adapter->detect_tx_hung = 0;
686                 /*
687                  * read barrier to make sure that the ->dma member and time
688                  * stamp are updated fully
689                  */
690                 smp_rmb();
691                 if (tx_ring->buffer_info[eop].dma &&
692                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp
693                                + (adapter->tx_timeout_factor * HZ))
694                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
695                         e1000_print_tx_hang(adapter);
696                         netif_stop_queue(netdev);
697                 }
698         }
699         adapter->total_tx_bytes += total_tx_bytes;
700         adapter->total_tx_packets += total_tx_packets;
701         adapter->net_stats.tx_bytes += total_tx_bytes;
702         adapter->net_stats.tx_packets += total_tx_packets;
703         return cleaned;
704 }
705
706 /**
707  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
708  * @adapter: board private structure
709  *
710  * the return value indicates whether actual cleaning was done, there
711  * is no guarantee that everything was cleaned
712  **/
713 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
714                                   int *work_done, int work_to_do)
715 {
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                 if (!(staterr & E1000_RXD_STAT_EOP)) {
759                         e_dbg("%s: Packet Split buffers didn't pick up the "
760                               "full packet\n", netdev->name);
761                         dev_kfree_skb_irq(skb);
762                         goto next_desc;
763                 }
764
765                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
766                         dev_kfree_skb_irq(skb);
767                         goto next_desc;
768                 }
769
770                 length = le16_to_cpu(rx_desc->wb.middle.length0);
771
772                 if (!length) {
773                         e_dbg("%s: Last part of the packet spanning multiple "
774                               "descriptors\n", netdev->name);
775                         dev_kfree_skb_irq(skb);
776                         goto next_desc;
777                 }
778
779                 /* Good Receive */
780                 skb_put(skb, length);
781
782                 {
783                 /*
784                  * this looks ugly, but it seems compiler issues make it
785                  * more efficient than reusing j
786                  */
787                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
788
789                 /*
790                  * page alloc/put takes too long and effects small packet
791                  * throughput, so unsplit small packets and save the alloc/put
792                  * only valid in softirq (napi) context to call kmap_*
793                  */
794                 if (l1 && (l1 <= copybreak) &&
795                     ((length + l1) <= adapter->rx_ps_bsize0)) {
796                         u8 *vaddr;
797
798                         ps_page = &buffer_info->ps_pages[0];
799
800                         /*
801                          * there is no documentation about how to call
802                          * kmap_atomic, so we can't hold the mapping
803                          * very long
804                          */
805                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
806                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
807                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
808                         memcpy(skb_tail_pointer(skb), vaddr, l1);
809                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
810                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
811                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
812
813                         /* remove the CRC */
814                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
815                                 l1 -= 4;
816
817                         skb_put(skb, l1);
818                         goto copydone;
819                 } /* if */
820                 }
821
822                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
823                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
824                         if (!length)
825                                 break;
826
827                         ps_page = &buffer_info->ps_pages[j];
828                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
829                                        PCI_DMA_FROMDEVICE);
830                         ps_page->dma = 0;
831                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
832                         ps_page->page = NULL;
833                         skb->len += length;
834                         skb->data_len += length;
835                         skb->truesize += length;
836                 }
837
838                 /* strip the ethernet crc, problem is we're using pages now so
839                  * this whole operation can get a little cpu intensive
840                  */
841                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
842                         pskb_trim(skb, skb->len - 4);
843
844 copydone:
845                 total_rx_bytes += skb->len;
846                 total_rx_packets++;
847
848                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
849                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
850
851                 if (rx_desc->wb.upper.header_status &
852                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
853                         adapter->rx_hdr_split++;
854
855                 e1000_receive_skb(adapter, netdev, skb,
856                                   staterr, rx_desc->wb.middle.vlan);
857
858 next_desc:
859                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
860                 buffer_info->skb = NULL;
861
862                 /* return some buffers to hardware, one at a time is too slow */
863                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
864                         adapter->alloc_rx_buf(adapter, cleaned_count);
865                         cleaned_count = 0;
866                 }
867
868                 /* use prefetched values */
869                 rx_desc = next_rxd;
870                 buffer_info = next_buffer;
871
872                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
873         }
874         rx_ring->next_to_clean = i;
875
876         cleaned_count = e1000_desc_unused(rx_ring);
877         if (cleaned_count)
878                 adapter->alloc_rx_buf(adapter, cleaned_count);
879
880         adapter->total_rx_bytes += total_rx_bytes;
881         adapter->total_rx_packets += total_rx_packets;
882         adapter->net_stats.rx_bytes += total_rx_bytes;
883         adapter->net_stats.rx_packets += total_rx_packets;
884         return cleaned;
885 }
886
887 /**
888  * e1000_consume_page - helper function
889  **/
890 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
891                                u16 length)
892 {
893         bi->page = NULL;
894         skb->len += length;
895         skb->data_len += length;
896         skb->truesize += length;
897 }
898
899 /**
900  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
901  * @adapter: board private structure
902  *
903  * the return value indicates whether actual cleaning was done, there
904  * is no guarantee that everything was cleaned
905  **/
906
907 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
908                                      int *work_done, int work_to_do)
909 {
910         struct net_device *netdev = adapter->netdev;
911         struct pci_dev *pdev = adapter->pdev;
912         struct e1000_ring *rx_ring = adapter->rx_ring;
913         struct e1000_rx_desc *rx_desc, *next_rxd;
914         struct e1000_buffer *buffer_info, *next_buffer;
915         u32 length;
916         unsigned int i;
917         int cleaned_count = 0;
918         bool cleaned = false;
919         unsigned int total_rx_bytes=0, total_rx_packets=0;
920
921         i = rx_ring->next_to_clean;
922         rx_desc = E1000_RX_DESC(*rx_ring, i);
923         buffer_info = &rx_ring->buffer_info[i];
924
925         while (rx_desc->status & E1000_RXD_STAT_DD) {
926                 struct sk_buff *skb;
927                 u8 status;
928
929                 if (*work_done >= work_to_do)
930                         break;
931                 (*work_done)++;
932
933                 status = rx_desc->status;
934                 skb = buffer_info->skb;
935                 buffer_info->skb = NULL;
936
937                 ++i;
938                 if (i == rx_ring->count)
939                         i = 0;
940                 next_rxd = E1000_RX_DESC(*rx_ring, i);
941                 prefetch(next_rxd);
942
943                 next_buffer = &rx_ring->buffer_info[i];
944
945                 cleaned = true;
946                 cleaned_count++;
947                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
948                                PCI_DMA_FROMDEVICE);
949                 buffer_info->dma = 0;
950
951                 length = le16_to_cpu(rx_desc->length);
952
953                 /* errors is only valid for DD + EOP descriptors */
954                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
955                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
956                                 /* recycle both page and skb */
957                                 buffer_info->skb = skb;
958                                 /* an error means any chain goes out the window
959                                  * too */
960                                 if (rx_ring->rx_skb_top)
961                                         dev_kfree_skb(rx_ring->rx_skb_top);
962                                 rx_ring->rx_skb_top = NULL;
963                                 goto next_desc;
964                 }
965
966 #define rxtop rx_ring->rx_skb_top
967                 if (!(status & E1000_RXD_STAT_EOP)) {
968                         /* this descriptor is only the beginning (or middle) */
969                         if (!rxtop) {
970                                 /* this is the beginning of a chain */
971                                 rxtop = skb;
972                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
973                                                    0, length);
974                         } else {
975                                 /* this is the middle of a chain */
976                                 skb_fill_page_desc(rxtop,
977                                     skb_shinfo(rxtop)->nr_frags,
978                                     buffer_info->page, 0, length);
979                                 /* re-use the skb, only consumed the page */
980                                 buffer_info->skb = skb;
981                         }
982                         e1000_consume_page(buffer_info, rxtop, length);
983                         goto next_desc;
984                 } else {
985                         if (rxtop) {
986                                 /* end of the chain */
987                                 skb_fill_page_desc(rxtop,
988                                     skb_shinfo(rxtop)->nr_frags,
989                                     buffer_info->page, 0, length);
990                                 /* re-use the current skb, we only consumed the
991                                  * page */
992                                 buffer_info->skb = skb;
993                                 skb = rxtop;
994                                 rxtop = NULL;
995                                 e1000_consume_page(buffer_info, skb, length);
996                         } else {
997                                 /* no chain, got EOP, this buf is the packet
998                                  * copybreak to save the put_page/alloc_page */
999                                 if (length <= copybreak &&
1000                                     skb_tailroom(skb) >= length) {
1001                                         u8 *vaddr;
1002                                         vaddr = kmap_atomic(buffer_info->page,
1003                                                            KM_SKB_DATA_SOFTIRQ);
1004                                         memcpy(skb_tail_pointer(skb), vaddr,
1005                                                length);
1006                                         kunmap_atomic(vaddr,
1007                                                       KM_SKB_DATA_SOFTIRQ);
1008                                         /* re-use the page, so don't erase
1009                                          * buffer_info->page */
1010                                         skb_put(skb, length);
1011                                 } else {
1012                                         skb_fill_page_desc(skb, 0,
1013                                                            buffer_info->page, 0,
1014                                                            length);
1015                                         e1000_consume_page(buffer_info, skb,
1016                                                            length);
1017                                 }
1018                         }
1019                 }
1020
1021                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1022                 e1000_rx_checksum(adapter,
1023                                   (u32)(status) |
1024                                   ((u32)(rx_desc->errors) << 24),
1025                                   le16_to_cpu(rx_desc->csum), skb);
1026
1027                 /* probably a little skewed due to removing CRC */
1028                 total_rx_bytes += skb->len;
1029                 total_rx_packets++;
1030
1031                 /* eth type trans needs skb->data to point to something */
1032                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1033                         e_err("pskb_may_pull failed.\n");
1034                         dev_kfree_skb(skb);
1035                         goto next_desc;
1036                 }
1037
1038                 e1000_receive_skb(adapter, netdev, skb, status,
1039                                   rx_desc->special);
1040
1041 next_desc:
1042                 rx_desc->status = 0;
1043
1044                 /* return some buffers to hardware, one at a time is too slow */
1045                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1046                         adapter->alloc_rx_buf(adapter, cleaned_count);
1047                         cleaned_count = 0;
1048                 }
1049
1050                 /* use prefetched values */
1051                 rx_desc = next_rxd;
1052                 buffer_info = next_buffer;
1053         }
1054         rx_ring->next_to_clean = i;
1055
1056         cleaned_count = e1000_desc_unused(rx_ring);
1057         if (cleaned_count)
1058                 adapter->alloc_rx_buf(adapter, cleaned_count);
1059
1060         adapter->total_rx_bytes += total_rx_bytes;
1061         adapter->total_rx_packets += total_rx_packets;
1062         adapter->net_stats.rx_bytes += total_rx_bytes;
1063         adapter->net_stats.rx_packets += total_rx_packets;
1064         return cleaned;
1065 }
1066
1067 /**
1068  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1069  * @adapter: board private structure
1070  **/
1071 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1072 {
1073         struct e1000_ring *rx_ring = adapter->rx_ring;
1074         struct e1000_buffer *buffer_info;
1075         struct e1000_ps_page *ps_page;
1076         struct pci_dev *pdev = adapter->pdev;
1077         unsigned int i, j;
1078
1079         /* Free all the Rx ring sk_buffs */
1080         for (i = 0; i < rx_ring->count; i++) {
1081                 buffer_info = &rx_ring->buffer_info[i];
1082                 if (buffer_info->dma) {
1083                         if (adapter->clean_rx == e1000_clean_rx_irq)
1084                                 pci_unmap_single(pdev, buffer_info->dma,
1085                                                  adapter->rx_buffer_len,
1086                                                  PCI_DMA_FROMDEVICE);
1087                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1088                                 pci_unmap_page(pdev, buffer_info->dma,
1089                                                PAGE_SIZE,
1090                                                PCI_DMA_FROMDEVICE);
1091                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1092                                 pci_unmap_single(pdev, buffer_info->dma,
1093                                                  adapter->rx_ps_bsize0,
1094                                                  PCI_DMA_FROMDEVICE);
1095                         buffer_info->dma = 0;
1096                 }
1097
1098                 if (buffer_info->page) {
1099                         put_page(buffer_info->page);
1100                         buffer_info->page = NULL;
1101                 }
1102
1103                 if (buffer_info->skb) {
1104                         dev_kfree_skb(buffer_info->skb);
1105                         buffer_info->skb = NULL;
1106                 }
1107
1108                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1109                         ps_page = &buffer_info->ps_pages[j];
1110                         if (!ps_page->page)
1111                                 break;
1112                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1113                                        PCI_DMA_FROMDEVICE);
1114                         ps_page->dma = 0;
1115                         put_page(ps_page->page);
1116                         ps_page->page = NULL;
1117                 }
1118         }
1119
1120         /* there also may be some cached data from a chained receive */
1121         if (rx_ring->rx_skb_top) {
1122                 dev_kfree_skb(rx_ring->rx_skb_top);
1123                 rx_ring->rx_skb_top = NULL;
1124         }
1125
1126         /* Zero out the descriptor ring */
1127         memset(rx_ring->desc, 0, rx_ring->size);
1128
1129         rx_ring->next_to_clean = 0;
1130         rx_ring->next_to_use = 0;
1131
1132         writel(0, adapter->hw.hw_addr + rx_ring->head);
1133         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1134 }
1135
1136 static void e1000e_downshift_workaround(struct work_struct *work)
1137 {
1138         struct e1000_adapter *adapter = container_of(work,
1139                                         struct e1000_adapter, downshift_task);
1140
1141         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1142 }
1143
1144 /**
1145  * e1000_intr_msi - Interrupt Handler
1146  * @irq: interrupt number
1147  * @data: pointer to a network interface device structure
1148  **/
1149 static irqreturn_t e1000_intr_msi(int irq, void *data)
1150 {
1151         struct net_device *netdev = data;
1152         struct e1000_adapter *adapter = netdev_priv(netdev);
1153         struct e1000_hw *hw = &adapter->hw;
1154         u32 icr = er32(ICR);
1155
1156         /*
1157          * read ICR disables interrupts using IAM
1158          */
1159
1160         if (icr & E1000_ICR_LSC) {
1161                 hw->mac.get_link_status = 1;
1162                 /*
1163                  * ICH8 workaround-- Call gig speed drop workaround on cable
1164                  * disconnect (LSC) before accessing any PHY registers
1165                  */
1166                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1167                     (!(er32(STATUS) & E1000_STATUS_LU)))
1168                         schedule_work(&adapter->downshift_task);
1169
1170                 /*
1171                  * 80003ES2LAN workaround-- For packet buffer work-around on
1172                  * link down event; disable receives here in the ISR and reset
1173                  * adapter in watchdog
1174                  */
1175                 if (netif_carrier_ok(netdev) &&
1176                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1177                         /* disable receives */
1178                         u32 rctl = er32(RCTL);
1179                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1180                         adapter->flags |= FLAG_RX_RESTART_NOW;
1181                 }
1182                 /* guard against interrupt when we're going down */
1183                 if (!test_bit(__E1000_DOWN, &adapter->state))
1184                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1185         }
1186
1187         if (napi_schedule_prep(&adapter->napi)) {
1188                 adapter->total_tx_bytes = 0;
1189                 adapter->total_tx_packets = 0;
1190                 adapter->total_rx_bytes = 0;
1191                 adapter->total_rx_packets = 0;
1192                 __napi_schedule(&adapter->napi);
1193         }
1194
1195         return IRQ_HANDLED;
1196 }
1197
1198 /**
1199  * e1000_intr - Interrupt Handler
1200  * @irq: interrupt number
1201  * @data: pointer to a network interface device structure
1202  **/
1203 static irqreturn_t e1000_intr(int irq, void *data)
1204 {
1205         struct net_device *netdev = data;
1206         struct e1000_adapter *adapter = netdev_priv(netdev);
1207         struct e1000_hw *hw = &adapter->hw;
1208         u32 rctl, icr = er32(ICR);
1209
1210         if (!icr)
1211                 return IRQ_NONE;  /* Not our interrupt */
1212
1213         /*
1214          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1215          * not set, then the adapter didn't send an interrupt
1216          */
1217         if (!(icr & E1000_ICR_INT_ASSERTED))
1218                 return IRQ_NONE;
1219
1220         /*
1221          * Interrupt Auto-Mask...upon reading ICR,
1222          * interrupts are masked.  No need for the
1223          * IMC write
1224          */
1225
1226         if (icr & E1000_ICR_LSC) {
1227                 hw->mac.get_link_status = 1;
1228                 /*
1229                  * ICH8 workaround-- Call gig speed drop workaround on cable
1230                  * disconnect (LSC) before accessing any PHY registers
1231                  */
1232                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1233                     (!(er32(STATUS) & E1000_STATUS_LU)))
1234                         schedule_work(&adapter->downshift_task);
1235
1236                 /*
1237                  * 80003ES2LAN workaround--
1238                  * For packet buffer work-around on link down event;
1239                  * disable receives here in the ISR and
1240                  * reset adapter in watchdog
1241                  */
1242                 if (netif_carrier_ok(netdev) &&
1243                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1244                         /* disable receives */
1245                         rctl = er32(RCTL);
1246                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1247                         adapter->flags |= FLAG_RX_RESTART_NOW;
1248                 }
1249                 /* guard against interrupt when we're going down */
1250                 if (!test_bit(__E1000_DOWN, &adapter->state))
1251                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1252         }
1253
1254         if (napi_schedule_prep(&adapter->napi)) {
1255                 adapter->total_tx_bytes = 0;
1256                 adapter->total_tx_packets = 0;
1257                 adapter->total_rx_bytes = 0;
1258                 adapter->total_rx_packets = 0;
1259                 __napi_schedule(&adapter->napi);
1260         }
1261
1262         return IRQ_HANDLED;
1263 }
1264
1265 static irqreturn_t e1000_msix_other(int irq, void *data)
1266 {
1267         struct net_device *netdev = data;
1268         struct e1000_adapter *adapter = netdev_priv(netdev);
1269         struct e1000_hw *hw = &adapter->hw;
1270         u32 icr = er32(ICR);
1271
1272         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1273                 ew32(IMS, E1000_IMS_OTHER);
1274                 return IRQ_NONE;
1275         }
1276
1277         if (icr & adapter->eiac_mask)
1278                 ew32(ICS, (icr & adapter->eiac_mask));
1279
1280         if (icr & E1000_ICR_OTHER) {
1281                 if (!(icr & E1000_ICR_LSC))
1282                         goto no_link_interrupt;
1283                 hw->mac.get_link_status = 1;
1284                 /* guard against interrupt when we're going down */
1285                 if (!test_bit(__E1000_DOWN, &adapter->state))
1286                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1287         }
1288
1289 no_link_interrupt:
1290         ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1291
1292         return IRQ_HANDLED;
1293 }
1294
1295
1296 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1297 {
1298         struct net_device *netdev = data;
1299         struct e1000_adapter *adapter = netdev_priv(netdev);
1300         struct e1000_hw *hw = &adapter->hw;
1301         struct e1000_ring *tx_ring = adapter->tx_ring;
1302
1303
1304         adapter->total_tx_bytes = 0;
1305         adapter->total_tx_packets = 0;
1306
1307         if (!e1000_clean_tx_irq(adapter))
1308                 /* Ring was not completely cleaned, so fire another interrupt */
1309                 ew32(ICS, tx_ring->ims_val);
1310
1311         return IRQ_HANDLED;
1312 }
1313
1314 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1315 {
1316         struct net_device *netdev = data;
1317         struct e1000_adapter *adapter = netdev_priv(netdev);
1318
1319         /* Write the ITR value calculated at the end of the
1320          * previous interrupt.
1321          */
1322         if (adapter->rx_ring->set_itr) {
1323                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1324                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1325                 adapter->rx_ring->set_itr = 0;
1326         }
1327
1328         if (napi_schedule_prep(&adapter->napi)) {
1329                 adapter->total_rx_bytes = 0;
1330                 adapter->total_rx_packets = 0;
1331                 __napi_schedule(&adapter->napi);
1332         }
1333         return IRQ_HANDLED;
1334 }
1335
1336 /**
1337  * e1000_configure_msix - Configure MSI-X hardware
1338  *
1339  * e1000_configure_msix sets up the hardware to properly
1340  * generate MSI-X interrupts.
1341  **/
1342 static void e1000_configure_msix(struct e1000_adapter *adapter)
1343 {
1344         struct e1000_hw *hw = &adapter->hw;
1345         struct e1000_ring *rx_ring = adapter->rx_ring;
1346         struct e1000_ring *tx_ring = adapter->tx_ring;
1347         int vector = 0;
1348         u32 ctrl_ext, ivar = 0;
1349
1350         adapter->eiac_mask = 0;
1351
1352         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1353         if (hw->mac.type == e1000_82574) {
1354                 u32 rfctl = er32(RFCTL);
1355                 rfctl |= E1000_RFCTL_ACK_DIS;
1356                 ew32(RFCTL, rfctl);
1357         }
1358
1359 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1360         /* Configure Rx vector */
1361         rx_ring->ims_val = E1000_IMS_RXQ0;
1362         adapter->eiac_mask |= rx_ring->ims_val;
1363         if (rx_ring->itr_val)
1364                 writel(1000000000 / (rx_ring->itr_val * 256),
1365                        hw->hw_addr + rx_ring->itr_register);
1366         else
1367                 writel(1, hw->hw_addr + rx_ring->itr_register);
1368         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1369
1370         /* Configure Tx vector */
1371         tx_ring->ims_val = E1000_IMS_TXQ0;
1372         vector++;
1373         if (tx_ring->itr_val)
1374                 writel(1000000000 / (tx_ring->itr_val * 256),
1375                        hw->hw_addr + tx_ring->itr_register);
1376         else
1377                 writel(1, hw->hw_addr + tx_ring->itr_register);
1378         adapter->eiac_mask |= tx_ring->ims_val;
1379         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1380
1381         /* set vector for Other Causes, e.g. link changes */
1382         vector++;
1383         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1384         if (rx_ring->itr_val)
1385                 writel(1000000000 / (rx_ring->itr_val * 256),
1386                        hw->hw_addr + E1000_EITR_82574(vector));
1387         else
1388                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1389
1390         /* Cause Tx interrupts on every write back */
1391         ivar |= (1 << 31);
1392
1393         ew32(IVAR, ivar);
1394
1395         /* enable MSI-X PBA support */
1396         ctrl_ext = er32(CTRL_EXT);
1397         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1398
1399         /* Auto-Mask Other interrupts upon ICR read */
1400 #define E1000_EIAC_MASK_82574   0x01F00000
1401         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1402         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1403         ew32(CTRL_EXT, ctrl_ext);
1404         e1e_flush();
1405 }
1406
1407 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1408 {
1409         if (adapter->msix_entries) {
1410                 pci_disable_msix(adapter->pdev);
1411                 kfree(adapter->msix_entries);
1412                 adapter->msix_entries = NULL;
1413         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1414                 pci_disable_msi(adapter->pdev);
1415                 adapter->flags &= ~FLAG_MSI_ENABLED;
1416         }
1417
1418         return;
1419 }
1420
1421 /**
1422  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1423  *
1424  * Attempt to configure interrupts using the best available
1425  * capabilities of the hardware and kernel.
1426  **/
1427 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1428 {
1429         int err;
1430         int numvecs, i;
1431
1432
1433         switch (adapter->int_mode) {
1434         case E1000E_INT_MODE_MSIX:
1435                 if (adapter->flags & FLAG_HAS_MSIX) {
1436                         numvecs = 3; /* RxQ0, TxQ0 and other */
1437                         adapter->msix_entries = kcalloc(numvecs,
1438                                                       sizeof(struct msix_entry),
1439                                                       GFP_KERNEL);
1440                         if (adapter->msix_entries) {
1441                                 for (i = 0; i < numvecs; i++)
1442                                         adapter->msix_entries[i].entry = i;
1443
1444                                 err = pci_enable_msix(adapter->pdev,
1445                                                       adapter->msix_entries,
1446                                                       numvecs);
1447                                 if (err == 0)
1448                                         return;
1449                         }
1450                         /* MSI-X failed, so fall through and try MSI */
1451                         e_err("Failed to initialize MSI-X interrupts.  "
1452                               "Falling back to MSI interrupts.\n");
1453                         e1000e_reset_interrupt_capability(adapter);
1454                 }
1455                 adapter->int_mode = E1000E_INT_MODE_MSI;
1456                 /* Fall through */
1457         case E1000E_INT_MODE_MSI:
1458                 if (!pci_enable_msi(adapter->pdev)) {
1459                         adapter->flags |= FLAG_MSI_ENABLED;
1460                 } else {
1461                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1462                         e_err("Failed to initialize MSI interrupts.  Falling "
1463                               "back to legacy interrupts.\n");
1464                 }
1465                 /* Fall through */
1466         case E1000E_INT_MODE_LEGACY:
1467                 /* Don't do anything; this is the system default */
1468                 break;
1469         }
1470
1471         return;
1472 }
1473
1474 /**
1475  * e1000_request_msix - Initialize MSI-X interrupts
1476  *
1477  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1478  * kernel.
1479  **/
1480 static int e1000_request_msix(struct e1000_adapter *adapter)
1481 {
1482         struct net_device *netdev = adapter->netdev;
1483         int err = 0, vector = 0;
1484
1485         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1486                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1487         else
1488                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1489         err = request_irq(adapter->msix_entries[vector].vector,
1490                           &e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1491                           netdev);
1492         if (err)
1493                 goto out;
1494         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1495         adapter->rx_ring->itr_val = adapter->itr;
1496         vector++;
1497
1498         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1499                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1500         else
1501                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1502         err = request_irq(adapter->msix_entries[vector].vector,
1503                           &e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1504                           netdev);
1505         if (err)
1506                 goto out;
1507         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1508         adapter->tx_ring->itr_val = adapter->itr;
1509         vector++;
1510
1511         err = request_irq(adapter->msix_entries[vector].vector,
1512                           &e1000_msix_other, 0, netdev->name, netdev);
1513         if (err)
1514                 goto out;
1515
1516         e1000_configure_msix(adapter);
1517         return 0;
1518 out:
1519         return err;
1520 }
1521
1522 /**
1523  * e1000_request_irq - initialize interrupts
1524  *
1525  * Attempts to configure interrupts using the best available
1526  * capabilities of the hardware and kernel.
1527  **/
1528 static int e1000_request_irq(struct e1000_adapter *adapter)
1529 {
1530         struct net_device *netdev = adapter->netdev;
1531         int err;
1532
1533         if (adapter->msix_entries) {
1534                 err = e1000_request_msix(adapter);
1535                 if (!err)
1536                         return err;
1537                 /* fall back to MSI */
1538                 e1000e_reset_interrupt_capability(adapter);
1539                 adapter->int_mode = E1000E_INT_MODE_MSI;
1540                 e1000e_set_interrupt_capability(adapter);
1541         }
1542         if (adapter->flags & FLAG_MSI_ENABLED) {
1543                 err = request_irq(adapter->pdev->irq, &e1000_intr_msi, 0,
1544                                   netdev->name, netdev);
1545                 if (!err)
1546                         return err;
1547
1548                 /* fall back to legacy interrupt */
1549                 e1000e_reset_interrupt_capability(adapter);
1550                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1551         }
1552
1553         err = request_irq(adapter->pdev->irq, &e1000_intr, IRQF_SHARED,
1554                           netdev->name, netdev);
1555         if (err)
1556                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1557
1558         return err;
1559 }
1560
1561 static void e1000_free_irq(struct e1000_adapter *adapter)
1562 {
1563         struct net_device *netdev = adapter->netdev;
1564
1565         if (adapter->msix_entries) {
1566                 int vector = 0;
1567
1568                 free_irq(adapter->msix_entries[vector].vector, netdev);
1569                 vector++;
1570
1571                 free_irq(adapter->msix_entries[vector].vector, netdev);
1572                 vector++;
1573
1574                 /* Other Causes interrupt vector */
1575                 free_irq(adapter->msix_entries[vector].vector, netdev);
1576                 return;
1577         }
1578
1579         free_irq(adapter->pdev->irq, netdev);
1580 }
1581
1582 /**
1583  * e1000_irq_disable - Mask off interrupt generation on the NIC
1584  **/
1585 static void e1000_irq_disable(struct e1000_adapter *adapter)
1586 {
1587         struct e1000_hw *hw = &adapter->hw;
1588
1589         ew32(IMC, ~0);
1590         if (adapter->msix_entries)
1591                 ew32(EIAC_82574, 0);
1592         e1e_flush();
1593         synchronize_irq(adapter->pdev->irq);
1594 }
1595
1596 /**
1597  * e1000_irq_enable - Enable default interrupt generation settings
1598  **/
1599 static void e1000_irq_enable(struct e1000_adapter *adapter)
1600 {
1601         struct e1000_hw *hw = &adapter->hw;
1602
1603         if (adapter->msix_entries) {
1604                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1605                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1606         } else {
1607                 ew32(IMS, IMS_ENABLE_MASK);
1608         }
1609         e1e_flush();
1610 }
1611
1612 /**
1613  * e1000_get_hw_control - get control of the h/w from f/w
1614  * @adapter: address of board private structure
1615  *
1616  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1617  * For ASF and Pass Through versions of f/w this means that
1618  * the driver is loaded. For AMT version (only with 82573)
1619  * of the f/w this means that the network i/f is open.
1620  **/
1621 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1622 {
1623         struct e1000_hw *hw = &adapter->hw;
1624         u32 ctrl_ext;
1625         u32 swsm;
1626
1627         /* Let firmware know the driver has taken over */
1628         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1629                 swsm = er32(SWSM);
1630                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1631         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1632                 ctrl_ext = er32(CTRL_EXT);
1633                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1634         }
1635 }
1636
1637 /**
1638  * e1000_release_hw_control - release control of the h/w to f/w
1639  * @adapter: address of board private structure
1640  *
1641  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1642  * For ASF and Pass Through versions of f/w this means that the
1643  * driver is no longer loaded. For AMT version (only with 82573) i
1644  * of the f/w this means that the network i/f is closed.
1645  *
1646  **/
1647 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1648 {
1649         struct e1000_hw *hw = &adapter->hw;
1650         u32 ctrl_ext;
1651         u32 swsm;
1652
1653         /* Let firmware taken over control of h/w */
1654         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1655                 swsm = er32(SWSM);
1656                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1657         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1658                 ctrl_ext = er32(CTRL_EXT);
1659                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1660         }
1661 }
1662
1663 /**
1664  * @e1000_alloc_ring - allocate memory for a ring structure
1665  **/
1666 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1667                                 struct e1000_ring *ring)
1668 {
1669         struct pci_dev *pdev = adapter->pdev;
1670
1671         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1672                                         GFP_KERNEL);
1673         if (!ring->desc)
1674                 return -ENOMEM;
1675
1676         return 0;
1677 }
1678
1679 /**
1680  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1681  * @adapter: board private structure
1682  *
1683  * Return 0 on success, negative on failure
1684  **/
1685 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1686 {
1687         struct e1000_ring *tx_ring = adapter->tx_ring;
1688         int err = -ENOMEM, size;
1689
1690         size = sizeof(struct e1000_buffer) * tx_ring->count;
1691         tx_ring->buffer_info = vmalloc(size);
1692         if (!tx_ring->buffer_info)
1693                 goto err;
1694         memset(tx_ring->buffer_info, 0, size);
1695
1696         /* round up to nearest 4K */
1697         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1698         tx_ring->size = ALIGN(tx_ring->size, 4096);
1699
1700         err = e1000_alloc_ring_dma(adapter, tx_ring);
1701         if (err)
1702                 goto err;
1703
1704         tx_ring->next_to_use = 0;
1705         tx_ring->next_to_clean = 0;
1706
1707         return 0;
1708 err:
1709         vfree(tx_ring->buffer_info);
1710         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1711         return err;
1712 }
1713
1714 /**
1715  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1716  * @adapter: board private structure
1717  *
1718  * Returns 0 on success, negative on failure
1719  **/
1720 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1721 {
1722         struct e1000_ring *rx_ring = adapter->rx_ring;
1723         struct e1000_buffer *buffer_info;
1724         int i, size, desc_len, err = -ENOMEM;
1725
1726         size = sizeof(struct e1000_buffer) * rx_ring->count;
1727         rx_ring->buffer_info = vmalloc(size);
1728         if (!rx_ring->buffer_info)
1729                 goto err;
1730         memset(rx_ring->buffer_info, 0, size);
1731
1732         for (i = 0; i < rx_ring->count; i++) {
1733                 buffer_info = &rx_ring->buffer_info[i];
1734                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1735                                                 sizeof(struct e1000_ps_page),
1736                                                 GFP_KERNEL);
1737                 if (!buffer_info->ps_pages)
1738                         goto err_pages;
1739         }
1740
1741         desc_len = sizeof(union e1000_rx_desc_packet_split);
1742
1743         /* Round up to nearest 4K */
1744         rx_ring->size = rx_ring->count * desc_len;
1745         rx_ring->size = ALIGN(rx_ring->size, 4096);
1746
1747         err = e1000_alloc_ring_dma(adapter, rx_ring);
1748         if (err)
1749                 goto err_pages;
1750
1751         rx_ring->next_to_clean = 0;
1752         rx_ring->next_to_use = 0;
1753         rx_ring->rx_skb_top = NULL;
1754
1755         return 0;
1756
1757 err_pages:
1758         for (i = 0; i < rx_ring->count; i++) {
1759                 buffer_info = &rx_ring->buffer_info[i];
1760                 kfree(buffer_info->ps_pages);
1761         }
1762 err:
1763         vfree(rx_ring->buffer_info);
1764         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1765         return err;
1766 }
1767
1768 /**
1769  * e1000_clean_tx_ring - Free Tx Buffers
1770  * @adapter: board private structure
1771  **/
1772 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1773 {
1774         struct e1000_ring *tx_ring = adapter->tx_ring;
1775         struct e1000_buffer *buffer_info;
1776         unsigned long size;
1777         unsigned int i;
1778
1779         for (i = 0; i < tx_ring->count; i++) {
1780                 buffer_info = &tx_ring->buffer_info[i];
1781                 e1000_put_txbuf(adapter, buffer_info);
1782         }
1783
1784         size = sizeof(struct e1000_buffer) * tx_ring->count;
1785         memset(tx_ring->buffer_info, 0, size);
1786
1787         memset(tx_ring->desc, 0, tx_ring->size);
1788
1789         tx_ring->next_to_use = 0;
1790         tx_ring->next_to_clean = 0;
1791
1792         writel(0, adapter->hw.hw_addr + tx_ring->head);
1793         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1794 }
1795
1796 /**
1797  * e1000e_free_tx_resources - Free Tx Resources per Queue
1798  * @adapter: board private structure
1799  *
1800  * Free all transmit software resources
1801  **/
1802 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1803 {
1804         struct pci_dev *pdev = adapter->pdev;
1805         struct e1000_ring *tx_ring = adapter->tx_ring;
1806
1807         e1000_clean_tx_ring(adapter);
1808
1809         vfree(tx_ring->buffer_info);
1810         tx_ring->buffer_info = NULL;
1811
1812         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1813                           tx_ring->dma);
1814         tx_ring->desc = NULL;
1815 }
1816
1817 /**
1818  * e1000e_free_rx_resources - Free Rx Resources
1819  * @adapter: board private structure
1820  *
1821  * Free all receive software resources
1822  **/
1823
1824 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1825 {
1826         struct pci_dev *pdev = adapter->pdev;
1827         struct e1000_ring *rx_ring = adapter->rx_ring;
1828         int i;
1829
1830         e1000_clean_rx_ring(adapter);
1831
1832         for (i = 0; i < rx_ring->count; i++) {
1833                 kfree(rx_ring->buffer_info[i].ps_pages);
1834         }
1835
1836         vfree(rx_ring->buffer_info);
1837         rx_ring->buffer_info = NULL;
1838
1839         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1840                           rx_ring->dma);
1841         rx_ring->desc = NULL;
1842 }
1843
1844 /**
1845  * e1000_update_itr - update the dynamic ITR value based on statistics
1846  * @adapter: pointer to adapter
1847  * @itr_setting: current adapter->itr
1848  * @packets: the number of packets during this measurement interval
1849  * @bytes: the number of bytes during this measurement interval
1850  *
1851  *      Stores a new ITR value based on packets and byte
1852  *      counts during the last interrupt.  The advantage of per interrupt
1853  *      computation is faster updates and more accurate ITR for the current
1854  *      traffic pattern.  Constants in this function were computed
1855  *      based on theoretical maximum wire speed and thresholds were set based
1856  *      on testing data as well as attempting to minimize response time
1857  *      while increasing bulk throughput.  This functionality is controlled
1858  *      by the InterruptThrottleRate module parameter.
1859  **/
1860 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1861                                      u16 itr_setting, int packets,
1862                                      int bytes)
1863 {
1864         unsigned int retval = itr_setting;
1865
1866         if (packets == 0)
1867                 goto update_itr_done;
1868
1869         switch (itr_setting) {
1870         case lowest_latency:
1871                 /* handle TSO and jumbo frames */
1872                 if (bytes/packets > 8000)
1873                         retval = bulk_latency;
1874                 else if ((packets < 5) && (bytes > 512)) {
1875                         retval = low_latency;
1876                 }
1877                 break;
1878         case low_latency:  /* 50 usec aka 20000 ints/s */
1879                 if (bytes > 10000) {
1880                         /* this if handles the TSO accounting */
1881                         if (bytes/packets > 8000) {
1882                                 retval = bulk_latency;
1883                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1884                                 retval = bulk_latency;
1885                         } else if ((packets > 35)) {
1886                                 retval = lowest_latency;
1887                         }
1888                 } else if (bytes/packets > 2000) {
1889                         retval = bulk_latency;
1890                 } else if (packets <= 2 && bytes < 512) {
1891                         retval = lowest_latency;
1892                 }
1893                 break;
1894         case bulk_latency: /* 250 usec aka 4000 ints/s */
1895                 if (bytes > 25000) {
1896                         if (packets > 35) {
1897                                 retval = low_latency;
1898                         }
1899                 } else if (bytes < 6000) {
1900                         retval = low_latency;
1901                 }
1902                 break;
1903         }
1904
1905 update_itr_done:
1906         return retval;
1907 }
1908
1909 static void e1000_set_itr(struct e1000_adapter *adapter)
1910 {
1911         struct e1000_hw *hw = &adapter->hw;
1912         u16 current_itr;
1913         u32 new_itr = adapter->itr;
1914
1915         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1916         if (adapter->link_speed != SPEED_1000) {
1917                 current_itr = 0;
1918                 new_itr = 4000;
1919                 goto set_itr_now;
1920         }
1921
1922         adapter->tx_itr = e1000_update_itr(adapter,
1923                                     adapter->tx_itr,
1924                                     adapter->total_tx_packets,
1925                                     adapter->total_tx_bytes);
1926         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1927         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1928                 adapter->tx_itr = low_latency;
1929
1930         adapter->rx_itr = e1000_update_itr(adapter,
1931                                     adapter->rx_itr,
1932                                     adapter->total_rx_packets,
1933                                     adapter->total_rx_bytes);
1934         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1935         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1936                 adapter->rx_itr = low_latency;
1937
1938         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1939
1940         switch (current_itr) {
1941         /* counts and packets in update_itr are dependent on these numbers */
1942         case lowest_latency:
1943                 new_itr = 70000;
1944                 break;
1945         case low_latency:
1946                 new_itr = 20000; /* aka hwitr = ~200 */
1947                 break;
1948         case bulk_latency:
1949                 new_itr = 4000;
1950                 break;
1951         default:
1952                 break;
1953         }
1954
1955 set_itr_now:
1956         if (new_itr != adapter->itr) {
1957                 /*
1958                  * this attempts to bias the interrupt rate towards Bulk
1959                  * by adding intermediate steps when interrupt rate is
1960                  * increasing
1961                  */
1962                 new_itr = new_itr > adapter->itr ?
1963                              min(adapter->itr + (new_itr >> 2), new_itr) :
1964                              new_itr;
1965                 adapter->itr = new_itr;
1966                 adapter->rx_ring->itr_val = new_itr;
1967                 if (adapter->msix_entries)
1968                         adapter->rx_ring->set_itr = 1;
1969                 else
1970                         ew32(ITR, 1000000000 / (new_itr * 256));
1971         }
1972 }
1973
1974 /**
1975  * e1000_alloc_queues - Allocate memory for all rings
1976  * @adapter: board private structure to initialize
1977  **/
1978 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1979 {
1980         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1981         if (!adapter->tx_ring)
1982                 goto err;
1983
1984         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1985         if (!adapter->rx_ring)
1986                 goto err;
1987
1988         return 0;
1989 err:
1990         e_err("Unable to allocate memory for queues\n");
1991         kfree(adapter->rx_ring);
1992         kfree(adapter->tx_ring);
1993         return -ENOMEM;
1994 }
1995
1996 /**
1997  * e1000_clean - NAPI Rx polling callback
1998  * @napi: struct associated with this polling callback
1999  * @budget: amount of packets driver is allowed to process this poll
2000  **/
2001 static int e1000_clean(struct napi_struct *napi, int budget)
2002 {
2003         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2004         struct e1000_hw *hw = &adapter->hw;
2005         struct net_device *poll_dev = adapter->netdev;
2006         int tx_cleaned = 0, work_done = 0;
2007
2008         adapter = netdev_priv(poll_dev);
2009
2010         if (adapter->msix_entries &&
2011             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2012                 goto clean_rx;
2013
2014         tx_cleaned = e1000_clean_tx_irq(adapter);
2015
2016 clean_rx:
2017         adapter->clean_rx(adapter, &work_done, budget);
2018
2019         if (tx_cleaned)
2020                 work_done = budget;
2021
2022         /* If budget not fully consumed, exit the polling mode */
2023         if (work_done < budget) {
2024                 if (adapter->itr_setting & 3)
2025                         e1000_set_itr(adapter);
2026                 napi_complete(napi);
2027                 if (adapter->msix_entries)
2028                         ew32(IMS, adapter->rx_ring->ims_val);
2029                 else
2030                         e1000_irq_enable(adapter);
2031         }
2032
2033         return work_done;
2034 }
2035
2036 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2037 {
2038         struct e1000_adapter *adapter = netdev_priv(netdev);
2039         struct e1000_hw *hw = &adapter->hw;
2040         u32 vfta, index;
2041
2042         /* don't update vlan cookie if already programmed */
2043         if ((adapter->hw.mng_cookie.status &
2044              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2045             (vid == adapter->mng_vlan_id))
2046                 return;
2047         /* add VID to filter table */
2048         index = (vid >> 5) & 0x7F;
2049         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2050         vfta |= (1 << (vid & 0x1F));
2051         e1000e_write_vfta(hw, index, vfta);
2052 }
2053
2054 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2055 {
2056         struct e1000_adapter *adapter = netdev_priv(netdev);
2057         struct e1000_hw *hw = &adapter->hw;
2058         u32 vfta, index;
2059
2060         if (!test_bit(__E1000_DOWN, &adapter->state))
2061                 e1000_irq_disable(adapter);
2062         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2063
2064         if (!test_bit(__E1000_DOWN, &adapter->state))
2065                 e1000_irq_enable(adapter);
2066
2067         if ((adapter->hw.mng_cookie.status &
2068              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2069             (vid == adapter->mng_vlan_id)) {
2070                 /* release control to f/w */
2071                 e1000_release_hw_control(adapter);
2072                 return;
2073         }
2074
2075         /* remove VID from filter table */
2076         index = (vid >> 5) & 0x7F;
2077         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2078         vfta &= ~(1 << (vid & 0x1F));
2079         e1000e_write_vfta(hw, index, vfta);
2080 }
2081
2082 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2083 {
2084         struct net_device *netdev = adapter->netdev;
2085         u16 vid = adapter->hw.mng_cookie.vlan_id;
2086         u16 old_vid = adapter->mng_vlan_id;
2087
2088         if (!adapter->vlgrp)
2089                 return;
2090
2091         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2092                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2093                 if (adapter->hw.mng_cookie.status &
2094                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2095                         e1000_vlan_rx_add_vid(netdev, vid);
2096                         adapter->mng_vlan_id = vid;
2097                 }
2098
2099                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2100                                 (vid != old_vid) &&
2101                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2102                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2103         } else {
2104                 adapter->mng_vlan_id = vid;
2105         }
2106 }
2107
2108
2109 static void e1000_vlan_rx_register(struct net_device *netdev,
2110                                    struct vlan_group *grp)
2111 {
2112         struct e1000_adapter *adapter = netdev_priv(netdev);
2113         struct e1000_hw *hw = &adapter->hw;
2114         u32 ctrl, rctl;
2115
2116         if (!test_bit(__E1000_DOWN, &adapter->state))
2117                 e1000_irq_disable(adapter);
2118         adapter->vlgrp = grp;
2119
2120         if (grp) {
2121                 /* enable VLAN tag insert/strip */
2122                 ctrl = er32(CTRL);
2123                 ctrl |= E1000_CTRL_VME;
2124                 ew32(CTRL, ctrl);
2125
2126                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2127                         /* enable VLAN receive filtering */
2128                         rctl = er32(RCTL);
2129                         rctl &= ~E1000_RCTL_CFIEN;
2130                         ew32(RCTL, rctl);
2131                         e1000_update_mng_vlan(adapter);
2132                 }
2133         } else {
2134                 /* disable VLAN tag insert/strip */
2135                 ctrl = er32(CTRL);
2136                 ctrl &= ~E1000_CTRL_VME;
2137                 ew32(CTRL, ctrl);
2138
2139                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2140                         if (adapter->mng_vlan_id !=
2141                             (u16)E1000_MNG_VLAN_NONE) {
2142                                 e1000_vlan_rx_kill_vid(netdev,
2143                                                        adapter->mng_vlan_id);
2144                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2145                         }
2146                 }
2147         }
2148
2149         if (!test_bit(__E1000_DOWN, &adapter->state))
2150                 e1000_irq_enable(adapter);
2151 }
2152
2153 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2154 {
2155         u16 vid;
2156
2157         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2158
2159         if (!adapter->vlgrp)
2160                 return;
2161
2162         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2163                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2164                         continue;
2165                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2166         }
2167 }
2168
2169 static void e1000_init_manageability(struct e1000_adapter *adapter)
2170 {
2171         struct e1000_hw *hw = &adapter->hw;
2172         u32 manc, manc2h;
2173
2174         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2175                 return;
2176
2177         manc = er32(MANC);
2178
2179         /*
2180          * enable receiving management packets to the host. this will probably
2181          * generate destination unreachable messages from the host OS, but
2182          * the packets will be handled on SMBUS
2183          */
2184         manc |= E1000_MANC_EN_MNG2HOST;
2185         manc2h = er32(MANC2H);
2186 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2187 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2188         manc2h |= E1000_MNG2HOST_PORT_623;
2189         manc2h |= E1000_MNG2HOST_PORT_664;
2190         ew32(MANC2H, manc2h);
2191         ew32(MANC, manc);
2192 }
2193
2194 /**
2195  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2196  * @adapter: board private structure
2197  *
2198  * Configure the Tx unit of the MAC after a reset.
2199  **/
2200 static void e1000_configure_tx(struct e1000_adapter *adapter)
2201 {
2202         struct e1000_hw *hw = &adapter->hw;
2203         struct e1000_ring *tx_ring = adapter->tx_ring;
2204         u64 tdba;
2205         u32 tdlen, tctl, tipg, tarc;
2206         u32 ipgr1, ipgr2;
2207
2208         /* Setup the HW Tx Head and Tail descriptor pointers */
2209         tdba = tx_ring->dma;
2210         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2211         ew32(TDBAL, (tdba & DMA_32BIT_MASK));
2212         ew32(TDBAH, (tdba >> 32));
2213         ew32(TDLEN, tdlen);
2214         ew32(TDH, 0);
2215         ew32(TDT, 0);
2216         tx_ring->head = E1000_TDH;
2217         tx_ring->tail = E1000_TDT;
2218
2219         /* Set the default values for the Tx Inter Packet Gap timer */
2220         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2221         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2222         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2223
2224         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2225                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2226
2227         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2228         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2229         ew32(TIPG, tipg);
2230
2231         /* Set the Tx Interrupt Delay register */
2232         ew32(TIDV, adapter->tx_int_delay);
2233         /* Tx irq moderation */
2234         ew32(TADV, adapter->tx_abs_int_delay);
2235
2236         /* Program the Transmit Control Register */
2237         tctl = er32(TCTL);
2238         tctl &= ~E1000_TCTL_CT;
2239         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2240                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2241
2242         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2243                 tarc = er32(TARC(0));
2244                 /*
2245                  * set the speed mode bit, we'll clear it if we're not at
2246                  * gigabit link later
2247                  */
2248 #define SPEED_MODE_BIT (1 << 21)
2249                 tarc |= SPEED_MODE_BIT;
2250                 ew32(TARC(0), tarc);
2251         }
2252
2253         /* errata: program both queues to unweighted RR */
2254         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2255                 tarc = er32(TARC(0));
2256                 tarc |= 1;
2257                 ew32(TARC(0), tarc);
2258                 tarc = er32(TARC(1));
2259                 tarc |= 1;
2260                 ew32(TARC(1), tarc);
2261         }
2262
2263         e1000e_config_collision_dist(hw);
2264
2265         /* Setup Transmit Descriptor Settings for eop descriptor */
2266         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2267
2268         /* only set IDE if we are delaying interrupts using the timers */
2269         if (adapter->tx_int_delay)
2270                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2271
2272         /* enable Report Status bit */
2273         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2274
2275         ew32(TCTL, tctl);
2276
2277         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
2278 }
2279
2280 /**
2281  * e1000_setup_rctl - configure the receive control registers
2282  * @adapter: Board private structure
2283  **/
2284 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2285                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2286 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2287 {
2288         struct e1000_hw *hw = &adapter->hw;
2289         u32 rctl, rfctl;
2290         u32 psrctl = 0;
2291         u32 pages = 0;
2292
2293         /* Program MC offset vector base */
2294         rctl = er32(RCTL);
2295         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2296         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2297                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2298                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2299
2300         /* Do not Store bad packets */
2301         rctl &= ~E1000_RCTL_SBP;
2302
2303         /* Enable Long Packet receive */
2304         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2305                 rctl &= ~E1000_RCTL_LPE;
2306         else
2307                 rctl |= E1000_RCTL_LPE;
2308
2309         /* Some systems expect that the CRC is included in SMBUS traffic. The
2310          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2311          * host memory when this is enabled
2312          */
2313         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2314                 rctl |= E1000_RCTL_SECRC;
2315
2316         /* Setup buffer sizes */
2317         rctl &= ~E1000_RCTL_SZ_4096;
2318         rctl |= E1000_RCTL_BSEX;
2319         switch (adapter->rx_buffer_len) {
2320         case 256:
2321                 rctl |= E1000_RCTL_SZ_256;
2322                 rctl &= ~E1000_RCTL_BSEX;
2323                 break;
2324         case 512:
2325                 rctl |= E1000_RCTL_SZ_512;
2326                 rctl &= ~E1000_RCTL_BSEX;
2327                 break;
2328         case 1024:
2329                 rctl |= E1000_RCTL_SZ_1024;
2330                 rctl &= ~E1000_RCTL_BSEX;
2331                 break;
2332         case 2048:
2333         default:
2334                 rctl |= E1000_RCTL_SZ_2048;
2335                 rctl &= ~E1000_RCTL_BSEX;
2336                 break;
2337         case 4096:
2338                 rctl |= E1000_RCTL_SZ_4096;
2339                 break;
2340         case 8192:
2341                 rctl |= E1000_RCTL_SZ_8192;
2342                 break;
2343         case 16384:
2344                 rctl |= E1000_RCTL_SZ_16384;
2345                 break;
2346         }
2347
2348         /*
2349          * 82571 and greater support packet-split where the protocol
2350          * header is placed in skb->data and the packet data is
2351          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2352          * In the case of a non-split, skb->data is linearly filled,
2353          * followed by the page buffers.  Therefore, skb->data is
2354          * sized to hold the largest protocol header.
2355          *
2356          * allocations using alloc_page take too long for regular MTU
2357          * so only enable packet split for jumbo frames
2358          *
2359          * Using pages when the page size is greater than 16k wastes
2360          * a lot of memory, since we allocate 3 pages at all times
2361          * per packet.
2362          */
2363         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2364         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2365             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2366                 adapter->rx_ps_pages = pages;
2367         else
2368                 adapter->rx_ps_pages = 0;
2369
2370         if (adapter->rx_ps_pages) {
2371                 /* Configure extra packet-split registers */
2372                 rfctl = er32(RFCTL);
2373                 rfctl |= E1000_RFCTL_EXTEN;
2374                 /*
2375                  * disable packet split support for IPv6 extension headers,
2376                  * because some malformed IPv6 headers can hang the Rx
2377                  */
2378                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2379                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2380
2381                 ew32(RFCTL, rfctl);
2382
2383                 /* Enable Packet split descriptors */
2384                 rctl |= E1000_RCTL_DTYP_PS;
2385
2386                 psrctl |= adapter->rx_ps_bsize0 >>
2387                         E1000_PSRCTL_BSIZE0_SHIFT;
2388
2389                 switch (adapter->rx_ps_pages) {
2390                 case 3:
2391                         psrctl |= PAGE_SIZE <<
2392                                 E1000_PSRCTL_BSIZE3_SHIFT;
2393                 case 2:
2394                         psrctl |= PAGE_SIZE <<
2395                                 E1000_PSRCTL_BSIZE2_SHIFT;
2396                 case 1:
2397                         psrctl |= PAGE_SIZE >>
2398                                 E1000_PSRCTL_BSIZE1_SHIFT;
2399                         break;
2400                 }
2401
2402                 ew32(PSRCTL, psrctl);
2403         }
2404
2405         ew32(RCTL, rctl);
2406         /* just started the receive unit, no need to restart */
2407         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2408 }
2409
2410 /**
2411  * e1000_configure_rx - Configure Receive Unit after Reset
2412  * @adapter: board private structure
2413  *
2414  * Configure the Rx unit of the MAC after a reset.
2415  **/
2416 static void e1000_configure_rx(struct e1000_adapter *adapter)
2417 {
2418         struct e1000_hw *hw = &adapter->hw;
2419         struct e1000_ring *rx_ring = adapter->rx_ring;
2420         u64 rdba;
2421         u32 rdlen, rctl, rxcsum, ctrl_ext;
2422
2423         if (adapter->rx_ps_pages) {
2424                 /* this is a 32 byte descriptor */
2425                 rdlen = rx_ring->count *
2426                         sizeof(union e1000_rx_desc_packet_split);
2427                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2428                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2429         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2430                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2431                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2432                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2433         } else {
2434                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2435                 adapter->clean_rx = e1000_clean_rx_irq;
2436                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2437         }
2438
2439         /* disable receives while setting up the descriptors */
2440         rctl = er32(RCTL);
2441         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2442         e1e_flush();
2443         msleep(10);
2444
2445         /* set the Receive Delay Timer Register */
2446         ew32(RDTR, adapter->rx_int_delay);
2447
2448         /* irq moderation */
2449         ew32(RADV, adapter->rx_abs_int_delay);
2450         if (adapter->itr_setting != 0)
2451                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2452
2453         ctrl_ext = er32(CTRL_EXT);
2454         /* Reset delay timers after every interrupt */
2455         ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2456         /* Auto-Mask interrupts upon ICR access */
2457         ctrl_ext |= E1000_CTRL_EXT_IAME;
2458         ew32(IAM, 0xffffffff);
2459         ew32(CTRL_EXT, ctrl_ext);
2460         e1e_flush();
2461
2462         /*
2463          * Setup the HW Rx Head and Tail Descriptor Pointers and
2464          * the Base and Length of the Rx Descriptor Ring
2465          */
2466         rdba = rx_ring->dma;
2467         ew32(RDBAL, (rdba & DMA_32BIT_MASK));
2468         ew32(RDBAH, (rdba >> 32));
2469         ew32(RDLEN, rdlen);
2470         ew32(RDH, 0);
2471         ew32(RDT, 0);
2472         rx_ring->head = E1000_RDH;
2473         rx_ring->tail = E1000_RDT;
2474
2475         /* Enable Receive Checksum Offload for TCP and UDP */
2476         rxcsum = er32(RXCSUM);
2477         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2478                 rxcsum |= E1000_RXCSUM_TUOFL;
2479
2480                 /*
2481                  * IPv4 payload checksum for UDP fragments must be
2482                  * used in conjunction with packet-split.
2483                  */
2484                 if (adapter->rx_ps_pages)
2485                         rxcsum |= E1000_RXCSUM_IPPCSE;
2486         } else {
2487                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2488                 /* no need to clear IPPCSE as it defaults to 0 */
2489         }
2490         ew32(RXCSUM, rxcsum);
2491
2492         /*
2493          * Enable early receives on supported devices, only takes effect when
2494          * packet size is equal or larger than the specified value (in 8 byte
2495          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2496          */
2497         if ((adapter->flags & FLAG_HAS_ERT) &&
2498             (adapter->netdev->mtu > ETH_DATA_LEN)) {
2499                 u32 rxdctl = er32(RXDCTL(0));
2500                 ew32(RXDCTL(0), rxdctl | 0x3);
2501                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2502                 /*
2503                  * With jumbo frames and early-receive enabled, excessive
2504                  * C4->C2 latencies result in dropped transactions.
2505                  */
2506                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2507                                           e1000e_driver_name, 55);
2508         } else {
2509                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2510                                           e1000e_driver_name,
2511                                           PM_QOS_DEFAULT_VALUE);
2512         }
2513
2514         /* Enable Receives */
2515         ew32(RCTL, rctl);
2516 }
2517
2518 /**
2519  *  e1000_update_mc_addr_list - Update Multicast addresses
2520  *  @hw: pointer to the HW structure
2521  *  @mc_addr_list: array of multicast addresses to program
2522  *  @mc_addr_count: number of multicast addresses to program
2523  *  @rar_used_count: the first RAR register free to program
2524  *  @rar_count: total number of supported Receive Address Registers
2525  *
2526  *  Updates the Receive Address Registers and Multicast Table Array.
2527  *  The caller must have a packed mc_addr_list of multicast addresses.
2528  *  The parameter rar_count will usually be hw->mac.rar_entry_count
2529  *  unless there are workarounds that change this.  Currently no func pointer
2530  *  exists and all implementations are handled in the generic version of this
2531  *  function.
2532  **/
2533 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2534                                       u32 mc_addr_count, u32 rar_used_count,
2535                                       u32 rar_count)
2536 {
2537         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2538                                         rar_used_count, rar_count);
2539 }
2540
2541 /**
2542  * e1000_set_multi - Multicast and Promiscuous mode set
2543  * @netdev: network interface device structure
2544  *
2545  * The set_multi entry point is called whenever the multicast address
2546  * list or the network interface flags are updated.  This routine is
2547  * responsible for configuring the hardware for proper multicast,
2548  * promiscuous mode, and all-multi behavior.
2549  **/
2550 static void e1000_set_multi(struct net_device *netdev)
2551 {
2552         struct e1000_adapter *adapter = netdev_priv(netdev);
2553         struct e1000_hw *hw = &adapter->hw;
2554         struct e1000_mac_info *mac = &hw->mac;
2555         struct dev_mc_list *mc_ptr;
2556         u8  *mta_list;
2557         u32 rctl;
2558         int i;
2559
2560         /* Check for Promiscuous and All Multicast modes */
2561
2562         rctl = er32(RCTL);
2563
2564         if (netdev->flags & IFF_PROMISC) {
2565                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2566                 rctl &= ~E1000_RCTL_VFE;
2567         } else {
2568                 if (netdev->flags & IFF_ALLMULTI) {
2569                         rctl |= E1000_RCTL_MPE;
2570                         rctl &= ~E1000_RCTL_UPE;
2571                 } else {
2572                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2573                 }
2574                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2575                         rctl |= E1000_RCTL_VFE;
2576         }
2577
2578         ew32(RCTL, rctl);
2579
2580         if (netdev->mc_count) {
2581                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2582                 if (!mta_list)
2583                         return;
2584
2585                 /* prepare a packed array of only addresses. */
2586                 mc_ptr = netdev->mc_list;
2587
2588                 for (i = 0; i < netdev->mc_count; i++) {
2589                         if (!mc_ptr)
2590                                 break;
2591                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2592                                ETH_ALEN);
2593                         mc_ptr = mc_ptr->next;
2594                 }
2595
2596                 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2597                                           mac->rar_entry_count);
2598                 kfree(mta_list);
2599         } else {
2600                 /*
2601                  * if we're called from probe, we might not have
2602                  * anything to do here, so clear out the list
2603                  */
2604                 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2605         }
2606 }
2607
2608 /**
2609  * e1000_configure - configure the hardware for Rx and Tx
2610  * @adapter: private board structure
2611  **/
2612 static void e1000_configure(struct e1000_adapter *adapter)
2613 {
2614         e1000_set_multi(adapter->netdev);
2615
2616         e1000_restore_vlan(adapter);
2617         e1000_init_manageability(adapter);
2618
2619         e1000_configure_tx(adapter);
2620         e1000_setup_rctl(adapter);
2621         e1000_configure_rx(adapter);
2622         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2623 }
2624
2625 /**
2626  * e1000e_power_up_phy - restore link in case the phy was powered down
2627  * @adapter: address of board private structure
2628  *
2629  * The phy may be powered down to save power and turn off link when the
2630  * driver is unloaded and wake on lan is not enabled (among others)
2631  * *** this routine MUST be followed by a call to e1000e_reset ***
2632  **/
2633 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2634 {
2635         u16 mii_reg = 0;
2636
2637         /* Just clear the power down bit to wake the phy back up */
2638         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2639                 /*
2640                  * According to the manual, the phy will retain its
2641                  * settings across a power-down/up cycle
2642                  */
2643                 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2644                 mii_reg &= ~MII_CR_POWER_DOWN;
2645                 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2646         }
2647
2648         adapter->hw.mac.ops.setup_link(&adapter->hw);
2649 }
2650
2651 /**
2652  * e1000_power_down_phy - Power down the PHY
2653  *
2654  * Power down the PHY so no link is implied when interface is down
2655  * The PHY cannot be powered down is management or WoL is active
2656  */
2657 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2658 {
2659         struct e1000_hw *hw = &adapter->hw;
2660         u16 mii_reg;
2661
2662         /* WoL is enabled */
2663         if (adapter->wol)
2664                 return;
2665
2666         /* non-copper PHY? */
2667         if (adapter->hw.phy.media_type != e1000_media_type_copper)
2668                 return;
2669
2670         /* reset is blocked because of a SoL/IDER session */
2671         if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2672                 return;
2673
2674         /* manageability (AMT) is enabled */
2675         if (er32(MANC) & E1000_MANC_SMBUS_EN)
2676                 return;
2677
2678         /* power down the PHY */
2679         e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2680         mii_reg |= MII_CR_POWER_DOWN;
2681         e1e_wphy(hw, PHY_CONTROL, mii_reg);
2682         mdelay(1);
2683 }
2684
2685 /**
2686  * e1000e_reset - bring the hardware into a known good state
2687  *
2688  * This function boots the hardware and enables some settings that
2689  * require a configuration cycle of the hardware - those cannot be
2690  * set/changed during runtime. After reset the device needs to be
2691  * properly configured for Rx, Tx etc.
2692  */
2693 void e1000e_reset(struct e1000_adapter *adapter)
2694 {
2695         struct e1000_mac_info *mac = &adapter->hw.mac;
2696         struct e1000_fc_info *fc = &adapter->hw.fc;
2697         struct e1000_hw *hw = &adapter->hw;
2698         u32 tx_space, min_tx_space, min_rx_space;
2699         u32 pba = adapter->pba;
2700         u16 hwm;
2701
2702         /* reset Packet Buffer Allocation to default */
2703         ew32(PBA, pba);
2704
2705         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2706                 /*
2707                  * To maintain wire speed transmits, the Tx FIFO should be
2708                  * large enough to accommodate two full transmit packets,
2709                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2710                  * the Rx FIFO should be large enough to accommodate at least
2711                  * one full receive packet and is similarly rounded up and
2712                  * expressed in KB.
2713                  */
2714                 pba = er32(PBA);
2715                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2716                 tx_space = pba >> 16;
2717                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2718                 pba &= 0xffff;
2719                 /*
2720                  * the Tx fifo also stores 16 bytes of information about the tx
2721                  * but don't include ethernet FCS because hardware appends it
2722                  */
2723                 min_tx_space = (adapter->max_frame_size +
2724                                 sizeof(struct e1000_tx_desc) -
2725                                 ETH_FCS_LEN) * 2;
2726                 min_tx_space = ALIGN(min_tx_space, 1024);
2727                 min_tx_space >>= 10;
2728                 /* software strips receive CRC, so leave room for it */
2729                 min_rx_space = adapter->max_frame_size;
2730                 min_rx_space = ALIGN(min_rx_space, 1024);
2731                 min_rx_space >>= 10;
2732
2733                 /*
2734                  * If current Tx allocation is less than the min Tx FIFO size,
2735                  * and the min Tx FIFO size is less than the current Rx FIFO
2736                  * allocation, take space away from current Rx allocation
2737                  */
2738                 if ((tx_space < min_tx_space) &&
2739                     ((min_tx_space - tx_space) < pba)) {
2740                         pba -= min_tx_space - tx_space;
2741
2742                         /*
2743                          * if short on Rx space, Rx wins and must trump tx
2744                          * adjustment or use Early Receive if available
2745                          */
2746                         if ((pba < min_rx_space) &&
2747                             (!(adapter->flags & FLAG_HAS_ERT)))
2748                                 /* ERT enabled in e1000_configure_rx */
2749                                 pba = min_rx_space;
2750                 }
2751
2752                 ew32(PBA, pba);
2753         }
2754
2755
2756         /*
2757          * flow control settings
2758          *
2759          * The high water mark must be low enough to fit one full frame
2760          * (or the size used for early receive) above it in the Rx FIFO.
2761          * Set it to the lower of:
2762          * - 90% of the Rx FIFO size, and
2763          * - the full Rx FIFO size minus the early receive size (for parts
2764          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2765          * - the full Rx FIFO size minus one full frame
2766          */
2767         if (adapter->flags & FLAG_HAS_ERT)
2768                 hwm = min(((pba << 10) * 9 / 10),
2769                           ((pba << 10) - (E1000_ERT_2048 << 3)));
2770         else
2771                 hwm = min(((pba << 10) * 9 / 10),
2772                           ((pba << 10) - adapter->max_frame_size));
2773
2774         fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
2775         fc->low_water = fc->high_water - 8;
2776
2777         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2778                 fc->pause_time = 0xFFFF;
2779         else
2780                 fc->pause_time = E1000_FC_PAUSE_TIME;
2781         fc->send_xon = 1;
2782         fc->current_mode = fc->requested_mode;
2783
2784         /* Allow time for pending master requests to run */
2785         mac->ops.reset_hw(hw);
2786
2787         /*
2788          * For parts with AMT enabled, let the firmware know
2789          * that the network interface is in control
2790          */
2791         if (adapter->flags & FLAG_HAS_AMT)
2792                 e1000_get_hw_control(adapter);
2793
2794         ew32(WUC, 0);
2795
2796         if (mac->ops.init_hw(hw))
2797                 e_err("Hardware Error\n");
2798
2799         e1000_update_mng_vlan(adapter);
2800
2801         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2802         ew32(VET, ETH_P_8021Q);
2803
2804         e1000e_reset_adaptive(hw);
2805         e1000_get_phy_info(hw);
2806
2807         if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2808                 u16 phy_data = 0;
2809                 /*
2810                  * speed up time to link by disabling smart power down, ignore
2811                  * the return value of this function because there is nothing
2812                  * different we would do if it failed
2813                  */
2814                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2815                 phy_data &= ~IGP02E1000_PM_SPD;
2816                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2817         }
2818 }
2819
2820 int e1000e_up(struct e1000_adapter *adapter)
2821 {
2822         struct e1000_hw *hw = &adapter->hw;
2823
2824         /* hardware has been reset, we need to reload some things */
2825         e1000_configure(adapter);
2826
2827         clear_bit(__E1000_DOWN, &adapter->state);
2828
2829         napi_enable(&adapter->napi);
2830         if (adapter->msix_entries)
2831                 e1000_configure_msix(adapter);
2832         e1000_irq_enable(adapter);
2833
2834         /* fire a link change interrupt to start the watchdog */
2835         ew32(ICS, E1000_ICS_LSC);
2836         return 0;
2837 }
2838
2839 void e1000e_down(struct e1000_adapter *adapter)
2840 {
2841         struct net_device *netdev = adapter->netdev;
2842         struct e1000_hw *hw = &adapter->hw;
2843         u32 tctl, rctl;
2844
2845         /*
2846          * signal that we're down so the interrupt handler does not
2847          * reschedule our watchdog timer
2848          */
2849         set_bit(__E1000_DOWN, &adapter->state);
2850
2851         /* disable receives in the hardware */
2852         rctl = er32(RCTL);
2853         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2854         /* flush and sleep below */
2855
2856         netif_tx_stop_all_queues(netdev);
2857
2858         /* disable transmits in the hardware */
2859         tctl = er32(TCTL);
2860         tctl &= ~E1000_TCTL_EN;
2861         ew32(TCTL, tctl);
2862         /* flush both disables and wait for them to finish */
2863         e1e_flush();
2864         msleep(10);
2865
2866         napi_disable(&adapter->napi);
2867         e1000_irq_disable(adapter);
2868
2869         del_timer_sync(&adapter->watchdog_timer);
2870         del_timer_sync(&adapter->phy_info_timer);
2871
2872         netdev->tx_queue_len = adapter->tx_queue_len;
2873         netif_carrier_off(netdev);
2874         adapter->link_speed = 0;
2875         adapter->link_duplex = 0;
2876
2877         if (!pci_channel_offline(adapter->pdev))
2878                 e1000e_reset(adapter);
2879         e1000_clean_tx_ring(adapter);
2880         e1000_clean_rx_ring(adapter);
2881
2882         /*
2883          * TODO: for power management, we could drop the link and
2884          * pci_disable_device here.
2885          */
2886 }
2887
2888 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2889 {
2890         might_sleep();
2891         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2892                 msleep(1);
2893         e1000e_down(adapter);
2894         e1000e_up(adapter);
2895         clear_bit(__E1000_RESETTING, &adapter->state);
2896 }
2897
2898 /**
2899  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2900  * @adapter: board private structure to initialize
2901  *
2902  * e1000_sw_init initializes the Adapter private data structure.
2903  * Fields are initialized based on PCI device information and
2904  * OS network device settings (MTU size).
2905  **/
2906 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2907 {
2908         struct net_device *netdev = adapter->netdev;
2909
2910         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2911         adapter->rx_ps_bsize0 = 128;
2912         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2913         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2914
2915         e1000e_set_interrupt_capability(adapter);
2916
2917         if (e1000_alloc_queues(adapter))
2918                 return -ENOMEM;
2919
2920         /* Explicitly disable IRQ since the NIC can be in any state. */
2921         e1000_irq_disable(adapter);
2922
2923         set_bit(__E1000_DOWN, &adapter->state);
2924         return 0;
2925 }
2926
2927 /**
2928  * e1000_intr_msi_test - Interrupt Handler
2929  * @irq: interrupt number
2930  * @data: pointer to a network interface device structure
2931  **/
2932 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2933 {
2934         struct net_device *netdev = data;
2935         struct e1000_adapter *adapter = netdev_priv(netdev);
2936         struct e1000_hw *hw = &adapter->hw;
2937         u32 icr = er32(ICR);
2938
2939         e_dbg("%s: icr is %08X\n", netdev->name, icr);
2940         if (icr & E1000_ICR_RXSEQ) {
2941                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2942                 wmb();
2943         }
2944
2945         return IRQ_HANDLED;
2946 }
2947
2948 /**
2949  * e1000_test_msi_interrupt - Returns 0 for successful test
2950  * @adapter: board private struct
2951  *
2952  * code flow taken from tg3.c
2953  **/
2954 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2955 {
2956         struct net_device *netdev = adapter->netdev;
2957         struct e1000_hw *hw = &adapter->hw;
2958         int err;
2959
2960         /* poll_enable hasn't been called yet, so don't need disable */
2961         /* clear any pending events */
2962         er32(ICR);
2963
2964         /* free the real vector and request a test handler */
2965         e1000_free_irq(adapter);
2966         e1000e_reset_interrupt_capability(adapter);
2967
2968         /* Assume that the test fails, if it succeeds then the test
2969          * MSI irq handler will unset this flag */
2970         adapter->flags |= FLAG_MSI_TEST_FAILED;
2971
2972         err = pci_enable_msi(adapter->pdev);
2973         if (err)
2974                 goto msi_test_failed;
2975
2976         err = request_irq(adapter->pdev->irq, &e1000_intr_msi_test, 0,
2977                           netdev->name, netdev);
2978         if (err) {
2979                 pci_disable_msi(adapter->pdev);
2980                 goto msi_test_failed;
2981         }
2982
2983         wmb();
2984
2985         e1000_irq_enable(adapter);
2986
2987         /* fire an unusual interrupt on the test handler */
2988         ew32(ICS, E1000_ICS_RXSEQ);
2989         e1e_flush();
2990         msleep(50);
2991
2992         e1000_irq_disable(adapter);
2993
2994         rmb();
2995
2996         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
2997                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2998                 err = -EIO;
2999                 e_info("MSI interrupt test failed!\n");
3000         }
3001
3002         free_irq(adapter->pdev->irq, netdev);
3003         pci_disable_msi(adapter->pdev);
3004
3005         if (err == -EIO)
3006                 goto msi_test_failed;
3007
3008         /* okay so the test worked, restore settings */
3009         e_dbg("%s: MSI interrupt test succeeded!\n", netdev->name);
3010 msi_test_failed:
3011         e1000e_set_interrupt_capability(adapter);
3012         e1000_request_irq(adapter);
3013         return err;
3014 }
3015
3016 /**
3017  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3018  * @adapter: board private struct
3019  *
3020  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3021  **/
3022 static int e1000_test_msi(struct e1000_adapter *adapter)
3023 {
3024         int err;
3025         u16 pci_cmd;
3026
3027         if (!(adapter->flags & FLAG_MSI_ENABLED))
3028                 return 0;
3029
3030         /* disable SERR in case the MSI write causes a master abort */
3031         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3032         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3033                               pci_cmd & ~PCI_COMMAND_SERR);
3034
3035         err = e1000_test_msi_interrupt(adapter);
3036
3037         /* restore previous setting of command word */
3038         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3039
3040         /* success ! */
3041         if (!err)
3042                 return 0;
3043
3044         /* EIO means MSI test failed */
3045         if (err != -EIO)
3046                 return err;
3047
3048         /* back to INTx mode */
3049         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3050
3051         e1000_free_irq(adapter);
3052
3053         err = e1000_request_irq(adapter);
3054
3055         return err;
3056 }
3057
3058 /**
3059  * e1000_open - Called when a network interface is made active
3060  * @netdev: network interface device structure
3061  *
3062  * Returns 0 on success, negative value on failure
3063  *
3064  * The open entry point is called when a network interface is made
3065  * active by the system (IFF_UP).  At this point all resources needed
3066  * for transmit and receive operations are allocated, the interrupt
3067  * handler is registered with the OS, the watchdog timer is started,
3068  * and the stack is notified that the interface is ready.
3069  **/
3070 static int e1000_open(struct net_device *netdev)
3071 {
3072         struct e1000_adapter *adapter = netdev_priv(netdev);
3073         struct e1000_hw *hw = &adapter->hw;
3074         int err;
3075
3076         /* disallow open during test */
3077         if (test_bit(__E1000_TESTING, &adapter->state))
3078                 return -EBUSY;
3079
3080         /* allocate transmit descriptors */
3081         err = e1000e_setup_tx_resources(adapter);
3082         if (err)
3083                 goto err_setup_tx;
3084
3085         /* allocate receive descriptors */
3086         err = e1000e_setup_rx_resources(adapter);
3087         if (err)
3088                 goto err_setup_rx;
3089
3090         e1000e_power_up_phy(adapter);
3091
3092         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3093         if ((adapter->hw.mng_cookie.status &
3094              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3095                 e1000_update_mng_vlan(adapter);
3096
3097         /*
3098          * If AMT is enabled, let the firmware know that the network
3099          * interface is now open
3100          */
3101         if (adapter->flags & FLAG_HAS_AMT)
3102                 e1000_get_hw_control(adapter);
3103
3104         /*
3105          * before we allocate an interrupt, we must be ready to handle it.
3106          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3107          * as soon as we call pci_request_irq, so we have to setup our
3108          * clean_rx handler before we do so.
3109          */
3110         e1000_configure(adapter);
3111
3112         err = e1000_request_irq(adapter);
3113         if (err)
3114                 goto err_req_irq;
3115
3116         /*
3117          * Work around PCIe errata with MSI interrupts causing some chipsets to
3118          * ignore e1000e MSI messages, which means we need to test our MSI
3119          * interrupt now
3120          */
3121         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3122                 err = e1000_test_msi(adapter);
3123                 if (err) {
3124                         e_err("Interrupt allocation failed\n");
3125                         goto err_req_irq;
3126                 }
3127         }
3128
3129         /* From here on the code is the same as e1000e_up() */
3130         clear_bit(__E1000_DOWN, &adapter->state);
3131
3132         napi_enable(&adapter->napi);
3133
3134         e1000_irq_enable(adapter);
3135
3136         netif_tx_start_all_queues(netdev);
3137
3138         /* fire a link status change interrupt to start the watchdog */
3139         ew32(ICS, E1000_ICS_LSC);
3140
3141         return 0;
3142
3143 err_req_irq:
3144         e1000_release_hw_control(adapter);
3145         e1000_power_down_phy(adapter);
3146         e1000e_free_rx_resources(adapter);
3147 err_setup_rx:
3148         e1000e_free_tx_resources(adapter);
3149 err_setup_tx:
3150         e1000e_reset(adapter);
3151
3152         return err;
3153 }
3154
3155 /**
3156  * e1000_close - Disables a network interface
3157  * @netdev: network interface device structure
3158  *
3159  * Returns 0, this is not allowed to fail
3160  *
3161  * The close entry point is called when an interface is de-activated
3162  * by the OS.  The hardware is still under the drivers control, but
3163  * needs to be disabled.  A global MAC reset is issued to stop the
3164  * hardware, and all transmit and receive resources are freed.
3165  **/
3166 static int e1000_close(struct net_device *netdev)
3167 {
3168         struct e1000_adapter *adapter = netdev_priv(netdev);
3169
3170         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3171         e1000e_down(adapter);
3172         e1000_power_down_phy(adapter);
3173         e1000_free_irq(adapter);
3174
3175         e1000e_free_tx_resources(adapter);
3176         e1000e_free_rx_resources(adapter);
3177
3178         /*
3179          * kill manageability vlan ID if supported, but not if a vlan with
3180          * the same ID is registered on the host OS (let 8021q kill it)
3181          */
3182         if ((adapter->hw.mng_cookie.status &
3183                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3184              !(adapter->vlgrp &&
3185                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3186                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3187
3188         /*
3189          * If AMT is enabled, let the firmware know that the network
3190          * interface is now closed
3191          */
3192         if (adapter->flags & FLAG_HAS_AMT)
3193                 e1000_release_hw_control(adapter);
3194
3195         return 0;
3196 }
3197 /**
3198  * e1000_set_mac - Change the Ethernet Address of the NIC
3199  * @netdev: network interface device structure
3200  * @p: pointer to an address structure
3201  *
3202  * Returns 0 on success, negative on failure
3203  **/
3204 static int e1000_set_mac(struct net_device *netdev, void *p)
3205 {
3206         struct e1000_adapter *adapter = netdev_priv(netdev);
3207         struct sockaddr *addr = p;
3208
3209         if (!is_valid_ether_addr(addr->sa_data))
3210                 return -EADDRNOTAVAIL;
3211
3212         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3213         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3214
3215         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3216
3217         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3218                 /* activate the work around */
3219                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3220
3221                 /*
3222                  * Hold a copy of the LAA in RAR[14] This is done so that
3223                  * between the time RAR[0] gets clobbered  and the time it
3224                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3225                  * of the RARs and no incoming packets directed to this port
3226                  * are dropped. Eventually the LAA will be in RAR[0] and
3227                  * RAR[14]
3228                  */
3229                 e1000e_rar_set(&adapter->hw,
3230                               adapter->hw.mac.addr,
3231                               adapter->hw.mac.rar_entry_count - 1);
3232         }
3233
3234         return 0;
3235 }
3236
3237 /**
3238  * e1000e_update_phy_task - work thread to update phy
3239  * @work: pointer to our work struct
3240  *
3241  * this worker thread exists because we must acquire a
3242  * semaphore to read the phy, which we could msleep while
3243  * waiting for it, and we can't msleep in a timer.
3244  **/
3245 static void e1000e_update_phy_task(struct work_struct *work)
3246 {
3247         struct e1000_adapter *adapter = container_of(work,
3248                                         struct e1000_adapter, update_phy_task);
3249         e1000_get_phy_info(&adapter->hw);
3250 }
3251
3252 /*
3253  * Need to wait a few seconds after link up to get diagnostic information from
3254  * the phy
3255  */
3256 static void e1000_update_phy_info(unsigned long data)
3257 {
3258         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3259         schedule_work(&adapter->update_phy_task);
3260 }
3261
3262 /**
3263  * e1000e_update_stats - Update the board statistics counters
3264  * @adapter: board private structure
3265  **/
3266 void e1000e_update_stats(struct e1000_adapter *adapter)
3267 {
3268         struct e1000_hw *hw = &adapter->hw;
3269         struct pci_dev *pdev = adapter->pdev;
3270
3271         /*
3272          * Prevent stats update while adapter is being reset, or if the pci
3273          * connection is down.
3274          */
3275         if (adapter->link_speed == 0)
3276                 return;
3277         if (pci_channel_offline(pdev))
3278                 return;
3279
3280         adapter->stats.crcerrs += er32(CRCERRS);
3281         adapter->stats.gprc += er32(GPRC);
3282         adapter->stats.gorc += er32(GORCL);
3283         er32(GORCH); /* Clear gorc */
3284         adapter->stats.bprc += er32(BPRC);
3285         adapter->stats.mprc += er32(MPRC);
3286         adapter->stats.roc += er32(ROC);
3287
3288         adapter->stats.mpc += er32(MPC);
3289         adapter->stats.scc += er32(SCC);
3290         adapter->stats.ecol += er32(ECOL);
3291         adapter->stats.mcc += er32(MCC);
3292         adapter->stats.latecol += er32(LATECOL);
3293         adapter->stats.dc += er32(DC);
3294         adapter->stats.xonrxc += er32(XONRXC);
3295         adapter->stats.xontxc += er32(XONTXC);
3296         adapter->stats.xoffrxc += er32(XOFFRXC);
3297         adapter->stats.xofftxc += er32(XOFFTXC);
3298         adapter->stats.gptc += er32(GPTC);
3299         adapter->stats.gotc += er32(GOTCL);
3300         er32(GOTCH); /* Clear gotc */
3301         adapter->stats.rnbc += er32(RNBC);
3302         adapter->stats.ruc += er32(RUC);
3303
3304         adapter->stats.mptc += er32(MPTC);
3305         adapter->stats.bptc += er32(BPTC);
3306
3307         /* used for adaptive IFS */
3308
3309         hw->mac.tx_packet_delta = er32(TPT);
3310         adapter->stats.tpt += hw->mac.tx_packet_delta;
3311         hw->mac.collision_delta = er32(COLC);
3312         adapter->stats.colc += hw->mac.collision_delta;
3313
3314         adapter->stats.algnerrc += er32(ALGNERRC);
3315         adapter->stats.rxerrc += er32(RXERRC);
3316         if ((hw->mac.type != e1000_82574) && (hw->mac.type != e1000_82583))
3317                 adapter->stats.tncrs += er32(TNCRS);
3318         adapter->stats.cexterr += er32(CEXTERR);
3319         adapter->stats.tsctc += er32(TSCTC);
3320         adapter->stats.tsctfc += er32(TSCTFC);
3321
3322         /* Fill out the OS statistics structure */
3323         adapter->net_stats.multicast = adapter->stats.mprc;
3324         adapter->net_stats.collisions = adapter->stats.colc;
3325
3326         /* Rx Errors */
3327
3328         /*
3329          * RLEC on some newer hardware can be incorrect so build
3330          * our own version based on RUC and ROC
3331          */
3332         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3333                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3334                 adapter->stats.ruc + adapter->stats.roc +
3335                 adapter->stats.cexterr;
3336         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3337                                               adapter->stats.roc;
3338         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3339         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3340         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3341
3342         /* Tx Errors */
3343         adapter->net_stats.tx_errors = adapter->stats.ecol +
3344                                        adapter->stats.latecol;
3345         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3346         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3347         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3348
3349         /* Tx Dropped needs to be maintained elsewhere */
3350
3351         /* Management Stats */
3352         adapter->stats.mgptc += er32(MGTPTC);
3353         adapter->stats.mgprc += er32(MGTPRC);
3354         adapter->stats.mgpdc += er32(MGTPDC);
3355 }
3356
3357 /**
3358  * e1000_phy_read_status - Update the PHY register status snapshot
3359  * @adapter: board private structure
3360  **/
3361 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3362 {
3363         struct e1000_hw *hw = &adapter->hw;
3364         struct e1000_phy_regs *phy = &adapter->phy_regs;
3365         int ret_val;
3366
3367         if ((er32(STATUS) & E1000_STATUS_LU) &&
3368             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3369                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3370                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3371                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3372                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3373                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3374                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3375                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3376                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3377                 if (ret_val)
3378                         e_warn("Error reading PHY register\n");
3379         } else {
3380                 /*
3381                  * Do not read PHY registers if link is not up
3382                  * Set values to typical power-on defaults
3383                  */
3384                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3385                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3386                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3387                              BMSR_ERCAP);
3388                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3389                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3390                 phy->lpa = 0;
3391                 phy->expansion = EXPANSION_ENABLENPAGE;
3392                 phy->ctrl1000 = ADVERTISE_1000FULL;
3393                 phy->stat1000 = 0;
3394                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3395         }
3396 }
3397
3398 static void e1000_print_link_info(struct e1000_adapter *adapter)
3399 {
3400         struct e1000_hw *hw = &adapter->hw;
3401         u32 ctrl = er32(CTRL);
3402
3403         /* Link status message must follow this format for user tools */
3404         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3405                "Flow Control: %s\n",
3406                adapter->netdev->name,
3407                adapter->link_speed,
3408                (adapter->link_duplex == FULL_DUPLEX) ?
3409                                 "Full Duplex" : "Half Duplex",
3410                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3411                                 "RX/TX" :
3412                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3413                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3414 }
3415
3416 bool e1000_has_link(struct e1000_adapter *adapter)
3417 {
3418         struct e1000_hw *hw = &adapter->hw;
3419         bool link_active = 0;
3420         s32 ret_val = 0;
3421
3422         /*
3423          * get_link_status is set on LSC (link status) interrupt or
3424          * Rx sequence error interrupt.  get_link_status will stay
3425          * false until the check_for_link establishes link
3426          * for copper adapters ONLY
3427          */
3428         switch (hw->phy.media_type) {
3429         case e1000_media_type_copper:
3430                 if (hw->mac.get_link_status) {
3431                         ret_val = hw->mac.ops.check_for_link(hw);
3432                         link_active = !hw->mac.get_link_status;
3433                 } else {
3434                         link_active = 1;
3435                 }
3436                 break;
3437         case e1000_media_type_fiber:
3438                 ret_val = hw->mac.ops.check_for_link(hw);
3439                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3440                 break;
3441         case e1000_media_type_internal_serdes:
3442                 ret_val = hw->mac.ops.check_for_link(hw);
3443                 link_active = adapter->hw.mac.serdes_has_link;
3444                 break;
3445         default:
3446         case e1000_media_type_unknown:
3447                 break;
3448         }
3449
3450         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3451             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3452                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3453                 e_info("Gigabit has been disabled, downgrading speed\n");
3454         }
3455
3456         return link_active;
3457 }
3458
3459 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3460 {
3461         /* make sure the receive unit is started */
3462         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3463             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3464                 struct e1000_hw *hw = &adapter->hw;
3465                 u32 rctl = er32(RCTL);
3466                 ew32(RCTL, rctl | E1000_RCTL_EN);
3467                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3468         }
3469 }
3470
3471 /**
3472  * e1000_watchdog - Timer Call-back
3473  * @data: pointer to adapter cast into an unsigned long
3474  **/
3475 static void e1000_watchdog(unsigned long data)
3476 {
3477         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3478
3479         /* Do the rest outside of interrupt context */
3480         schedule_work(&adapter->watchdog_task);
3481
3482         /* TODO: make this use queue_delayed_work() */
3483 }
3484
3485 static void e1000_watchdog_task(struct work_struct *work)
3486 {
3487         struct e1000_adapter *adapter = container_of(work,
3488                                         struct e1000_adapter, watchdog_task);
3489         struct net_device *netdev = adapter->netdev;
3490         struct e1000_mac_info *mac = &adapter->hw.mac;
3491         struct e1000_phy_info *phy = &adapter->hw.phy;
3492         struct e1000_ring *tx_ring = adapter->tx_ring;
3493         struct e1000_hw *hw = &adapter->hw;
3494         u32 link, tctl;
3495         int tx_pending = 0;
3496
3497         link = e1000_has_link(adapter);
3498         if ((netif_carrier_ok(netdev)) && link) {
3499                 e1000e_enable_receives(adapter);
3500                 goto link_up;
3501         }
3502
3503         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3504             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3505                 e1000_update_mng_vlan(adapter);
3506
3507         if (link) {
3508                 if (!netif_carrier_ok(netdev)) {
3509                         bool txb2b = 1;
3510                         /* update snapshot of PHY registers on LSC */
3511                         e1000_phy_read_status(adapter);
3512                         mac->ops.get_link_up_info(&adapter->hw,
3513                                                    &adapter->link_speed,
3514                                                    &adapter->link_duplex);
3515                         e1000_print_link_info(adapter);
3516                         /*
3517                          * On supported PHYs, check for duplex mismatch only
3518                          * if link has autonegotiated at 10/100 half
3519                          */
3520                         if ((hw->phy.type == e1000_phy_igp_3 ||
3521                              hw->phy.type == e1000_phy_bm) &&
3522                             (hw->mac.autoneg == true) &&
3523                             (adapter->link_speed == SPEED_10 ||
3524                              adapter->link_speed == SPEED_100) &&
3525                             (adapter->link_duplex == HALF_DUPLEX)) {
3526                                 u16 autoneg_exp;
3527
3528                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3529
3530                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3531                                         e_info("Autonegotiated half duplex but"
3532                                                " link partner cannot autoneg. "
3533                                                " Try forcing full duplex if "
3534                                                "link gets many collisions.\n");
3535                         }
3536
3537                         /*
3538                          * tweak tx_queue_len according to speed/duplex
3539                          * and adjust the timeout factor
3540                          */
3541                         netdev->tx_queue_len = adapter->tx_queue_len;
3542                         adapter->tx_timeout_factor = 1;
3543                         switch (adapter->link_speed) {
3544                         case SPEED_10:
3545                                 txb2b = 0;
3546                                 netdev->tx_queue_len = 10;
3547                                 adapter->tx_timeout_factor = 16;
3548                                 break;
3549                         case SPEED_100:
3550                                 txb2b = 0;
3551                                 netdev->tx_queue_len = 100;
3552                                 /* maybe add some timeout factor ? */
3553                                 break;
3554                         }
3555
3556                         /*
3557                          * workaround: re-program speed mode bit after
3558                          * link-up event
3559                          */
3560                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3561                             !txb2b) {
3562                                 u32 tarc0;
3563                                 tarc0 = er32(TARC(0));
3564                                 tarc0 &= ~SPEED_MODE_BIT;
3565                                 ew32(TARC(0), tarc0);
3566                         }
3567
3568                         /*
3569                          * disable TSO for pcie and 10/100 speeds, to avoid
3570                          * some hardware issues
3571                          */
3572                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3573                                 switch (adapter->link_speed) {
3574                                 case SPEED_10:
3575                                 case SPEED_100:
3576                                         e_info("10/100 speed: disabling TSO\n");
3577                                         netdev->features &= ~NETIF_F_TSO;
3578                                         netdev->features &= ~NETIF_F_TSO6;
3579                                         break;
3580                                 case SPEED_1000:
3581                                         netdev->features |= NETIF_F_TSO;
3582                                         netdev->features |= NETIF_F_TSO6;
3583                                         break;
3584                                 default:
3585                                         /* oops */
3586                                         break;
3587                                 }
3588                         }
3589
3590                         /*
3591                          * enable transmits in the hardware, need to do this
3592                          * after setting TARC(0)
3593                          */
3594                         tctl = er32(TCTL);
3595                         tctl |= E1000_TCTL_EN;
3596                         ew32(TCTL, tctl);
3597
3598                         /*
3599                          * Perform any post-link-up configuration before
3600                          * reporting link up.
3601                          */
3602                         if (phy->ops.cfg_on_link_up)
3603                                 phy->ops.cfg_on_link_up(hw);
3604
3605                         netif_carrier_on(netdev);
3606                         netif_tx_wake_all_queues(netdev);
3607
3608                         if (!test_bit(__E1000_DOWN, &adapter->state))
3609                                 mod_timer(&adapter->phy_info_timer,
3610                                           round_jiffies(jiffies + 2 * HZ));
3611                 }
3612         } else {
3613                 if (netif_carrier_ok(netdev)) {
3614                         adapter->link_speed = 0;
3615                         adapter->link_duplex = 0;
3616                         /* Link status message must follow this format */
3617                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
3618                                adapter->netdev->name);
3619                         netif_carrier_off(netdev);
3620                         netif_tx_stop_all_queues(netdev);
3621                         if (!test_bit(__E1000_DOWN, &adapter->state))
3622                                 mod_timer(&adapter->phy_info_timer,
3623                                           round_jiffies(jiffies + 2 * HZ));
3624
3625                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3626                                 schedule_work(&adapter->reset_task);
3627                 }
3628         }
3629
3630 link_up:
3631         e1000e_update_stats(adapter);
3632
3633         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3634         adapter->tpt_old = adapter->stats.tpt;
3635         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3636         adapter->colc_old = adapter->stats.colc;
3637
3638         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3639         adapter->gorc_old = adapter->stats.gorc;
3640         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3641         adapter->gotc_old = adapter->stats.gotc;
3642
3643         e1000e_update_adaptive(&adapter->hw);
3644
3645         if (!netif_carrier_ok(netdev)) {
3646                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3647                                tx_ring->count);
3648                 if (tx_pending) {
3649                         /*
3650                          * We've lost link, so the controller stops DMA,
3651                          * but we've got queued Tx work that's never going
3652                          * to get done, so reset controller to flush Tx.
3653                          * (Do the reset outside of interrupt context).
3654                          */
3655                         adapter->tx_timeout_count++;
3656                         schedule_work(&adapter->reset_task);
3657                 }
3658         }
3659
3660         /* Cause software interrupt to ensure Rx ring is cleaned */
3661         if (adapter->msix_entries)
3662                 ew32(ICS, adapter->rx_ring->ims_val);
3663         else
3664                 ew32(ICS, E1000_ICS_RXDMT0);
3665
3666         /* Force detection of hung controller every watchdog period */
3667         adapter->detect_tx_hung = 1;
3668
3669         /*
3670          * With 82571 controllers, LAA may be overwritten due to controller
3671          * reset from the other port. Set the appropriate LAA in RAR[0]
3672          */
3673         if (e1000e_get_laa_state_82571(hw))
3674                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3675
3676         /* Reset the timer */
3677         if (!test_bit(__E1000_DOWN, &adapter->state))
3678                 mod_timer(&adapter->watchdog_timer,
3679                           round_jiffies(jiffies + 2 * HZ));
3680 }
3681
3682 #define E1000_TX_FLAGS_CSUM             0x00000001
3683 #define E1000_TX_FLAGS_VLAN             0x00000002
3684 #define E1000_TX_FLAGS_TSO              0x00000004
3685 #define E1000_TX_FLAGS_IPV4             0x00000008
3686 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3687 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3688
3689 static int e1000_tso(struct e1000_adapter *adapter,
3690                      struct sk_buff *skb)
3691 {
3692         struct e1000_ring *tx_ring = adapter->tx_ring;
3693         struct e1000_context_desc *context_desc;
3694         struct e1000_buffer *buffer_info;
3695         unsigned int i;
3696         u32 cmd_length = 0;
3697         u16 ipcse = 0, tucse, mss;
3698         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3699         int err;
3700
3701         if (skb_is_gso(skb)) {
3702                 if (skb_header_cloned(skb)) {
3703                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3704                         if (err)
3705                                 return err;
3706                 }
3707
3708                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3709                 mss = skb_shinfo(skb)->gso_size;
3710                 if (skb->protocol == htons(ETH_P_IP)) {
3711                         struct iphdr *iph = ip_hdr(skb);
3712                         iph->tot_len = 0;
3713                         iph->check = 0;
3714                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3715                                                                  iph->daddr, 0,
3716                                                                  IPPROTO_TCP,
3717                                                                  0);
3718                         cmd_length = E1000_TXD_CMD_IP;
3719                         ipcse = skb_transport_offset(skb) - 1;
3720                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3721                         ipv6_hdr(skb)->payload_len = 0;
3722                         tcp_hdr(skb)->check =
3723                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3724                                                  &ipv6_hdr(skb)->daddr,
3725                                                  0, IPPROTO_TCP, 0);
3726                         ipcse = 0;
3727                 }
3728                 ipcss = skb_network_offset(skb);
3729                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3730                 tucss = skb_transport_offset(skb);
3731                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3732                 tucse = 0;
3733
3734                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3735                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3736
3737                 i = tx_ring->next_to_use;
3738                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3739                 buffer_info = &tx_ring->buffer_info[i];
3740
3741                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3742                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3743                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3744                 context_desc->upper_setup.tcp_fields.tucss = tucss;
3745                 context_desc->upper_setup.tcp_fields.tucso = tucso;
3746                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3747                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3748                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3749                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3750
3751                 buffer_info->time_stamp = jiffies;
3752                 buffer_info->next_to_watch = i;
3753
3754                 i++;
3755                 if (i == tx_ring->count)
3756                         i = 0;
3757                 tx_ring->next_to_use = i;
3758
3759                 return 1;
3760         }
3761
3762         return 0;
3763 }
3764
3765 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3766 {
3767         struct e1000_ring *tx_ring = adapter->tx_ring;
3768         struct e1000_context_desc *context_desc;
3769         struct e1000_buffer *buffer_info;
3770         unsigned int i;
3771         u8 css;
3772         u32 cmd_len = E1000_TXD_CMD_DEXT;
3773
3774         if (skb->ip_summed != CHECKSUM_PARTIAL)
3775                 return 0;
3776
3777         switch (skb->protocol) {
3778         case cpu_to_be16(ETH_P_IP):
3779                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3780                         cmd_len |= E1000_TXD_CMD_TCP;
3781                 break;
3782         case cpu_to_be16(ETH_P_IPV6):
3783                 /* XXX not handling all IPV6 headers */
3784                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3785                         cmd_len |= E1000_TXD_CMD_TCP;
3786                 break;
3787         default:
3788                 if (unlikely(net_ratelimit()))
3789                         e_warn("checksum_partial proto=%x!\n", skb->protocol);
3790                 break;
3791         }
3792
3793         css = skb_transport_offset(skb);
3794
3795         i = tx_ring->next_to_use;
3796         buffer_info = &tx_ring->buffer_info[i];
3797         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3798
3799         context_desc->lower_setup.ip_config = 0;
3800         context_desc->upper_setup.tcp_fields.tucss = css;
3801         context_desc->upper_setup.tcp_fields.tucso =
3802                                 css + skb->csum_offset;
3803         context_desc->upper_setup.tcp_fields.tucse = 0;
3804         context_desc->tcp_seg_setup.data = 0;
3805         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
3806
3807         buffer_info->time_stamp = jiffies;
3808         buffer_info->next_to_watch = i;
3809
3810         i++;
3811         if (i == tx_ring->count)
3812                 i = 0;
3813         tx_ring->next_to_use = i;
3814
3815         return 1;
3816 }
3817
3818 #define E1000_MAX_PER_TXD       8192
3819 #define E1000_MAX_TXD_PWR       12
3820
3821 static int e1000_tx_map(struct e1000_adapter *adapter,
3822                         struct sk_buff *skb, unsigned int first,
3823                         unsigned int max_per_txd, unsigned int nr_frags,
3824                         unsigned int mss)
3825 {
3826         struct e1000_ring *tx_ring = adapter->tx_ring;
3827         unsigned int len = skb_headlen(skb);
3828         unsigned int offset, size, count = 0, i;
3829         unsigned int f;
3830         dma_addr_t map;
3831
3832         i = tx_ring->next_to_use;
3833
3834         if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
3835                 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3836                 adapter->tx_dma_failed++;
3837                 dev_kfree_skb(skb);
3838                 return -2;
3839         }
3840
3841         map = skb_shinfo(skb)->dma_maps[0];
3842         offset = 0;
3843
3844         while (len) {
3845                 struct e1000_buffer *buffer_info = &tx_ring->buffer_info[i];
3846                 size = min(len, max_per_txd);
3847
3848                 buffer_info->length = size;
3849                 /* set time_stamp *before* dma to help avoid a possible race */
3850                 buffer_info->time_stamp = jiffies;
3851                 buffer_info->dma = map + offset;
3852                 buffer_info->next_to_watch = i;
3853
3854                 len -= size;
3855                 offset += size;
3856                 count++;
3857                 i++;
3858                 if (i == tx_ring->count)
3859                         i = 0;
3860         }
3861
3862         for (f = 0; f < nr_frags; f++) {
3863                 struct skb_frag_struct *frag;
3864
3865                 frag = &skb_shinfo(skb)->frags[f];
3866                 len = frag->size;
3867                 map = skb_shinfo(skb)->dma_maps[f + 1];
3868                 offset = 0;
3869
3870                 while (len) {
3871                         struct e1000_buffer *buffer_info;
3872                         buffer_info = &tx_ring->buffer_info[i];
3873                         size = min(len, max_per_txd);
3874
3875                         buffer_info->length = size;
3876                         buffer_info->time_stamp = jiffies;
3877                         buffer_info->dma = map + offset;
3878                         buffer_info->next_to_watch = i;
3879
3880                         len -= size;
3881                         offset += size;
3882                         count++;
3883
3884                         i++;
3885                         if (i == tx_ring->count)
3886                                 i = 0;
3887                 }
3888         }
3889
3890         if (i == 0)
3891                 i = tx_ring->count - 1;
3892         else
3893                 i--;
3894
3895         tx_ring->buffer_info[i].skb = skb;
3896         tx_ring->buffer_info[first].next_to_watch = i;
3897         smp_wmb();
3898
3899         return count;
3900 }
3901
3902 static void e1000_tx_queue(struct e1000_adapter *adapter,
3903                            int tx_flags, int count)
3904 {
3905         struct e1000_ring *tx_ring = adapter->tx_ring;
3906         struct e1000_tx_desc *tx_desc = NULL;
3907         struct e1000_buffer *buffer_info;
3908         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3909         unsigned int i;
3910
3911         if (tx_flags & E1000_TX_FLAGS_TSO) {
3912                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3913                              E1000_TXD_CMD_TSE;
3914                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3915
3916                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3917                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3918         }
3919
3920         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3921                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3922                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3923         }
3924
3925         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3926                 txd_lower |= E1000_TXD_CMD_VLE;
3927                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3928         }
3929
3930         i = tx_ring->next_to_use;
3931
3932         while (count--) {
3933                 buffer_info = &tx_ring->buffer_info[i];
3934                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3935                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3936                 tx_desc->lower.data =
3937                         cpu_to_le32(txd_lower | buffer_info->length);
3938                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3939
3940                 i++;
3941                 if (i == tx_ring->count)
3942                         i = 0;
3943         }
3944
3945         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3946
3947         /*
3948          * Force memory writes to complete before letting h/w
3949          * know there are new descriptors to fetch.  (Only
3950          * applicable for weak-ordered memory model archs,
3951          * such as IA-64).
3952          */
3953         wmb();
3954
3955         tx_ring->next_to_use = i;
3956         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3957         /*
3958          * we need this if more than one processor can write to our tail
3959          * at a time, it synchronizes IO on IA64/Altix systems
3960          */
3961         mmiowb();
3962 }
3963
3964 #define MINIMUM_DHCP_PACKET_SIZE 282
3965 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3966                                     struct sk_buff *skb)
3967 {
3968         struct e1000_hw *hw =  &adapter->hw;
3969         u16 length, offset;
3970
3971         if (vlan_tx_tag_present(skb)) {
3972                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3973                     && (adapter->hw.mng_cookie.status &
3974                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3975                         return 0;
3976         }
3977
3978         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3979                 return 0;
3980
3981         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3982                 return 0;
3983
3984         {
3985                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3986                 struct udphdr *udp;
3987
3988                 if (ip->protocol != IPPROTO_UDP)
3989                         return 0;
3990
3991                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3992                 if (ntohs(udp->dest) != 67)
3993                         return 0;
3994
3995                 offset = (u8 *)udp + 8 - skb->data;
3996                 length = skb->len - offset;
3997                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
3998         }
3999
4000         return 0;
4001 }
4002
4003 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4004 {
4005         struct e1000_adapter *adapter = netdev_priv(netdev);
4006
4007         netif_stop_queue(netdev);
4008         /*
4009          * Herbert's original patch had:
4010          *  smp_mb__after_netif_stop_queue();
4011          * but since that doesn't exist yet, just open code it.
4012          */
4013         smp_mb();
4014
4015         /*
4016          * We need to check again in a case another CPU has just
4017          * made room available.
4018          */
4019         if (e1000_desc_unused(adapter->tx_ring) < size)
4020                 return -EBUSY;
4021
4022         /* A reprieve! */
4023         netif_start_queue(netdev);
4024         ++adapter->restart_queue;
4025         return 0;
4026 }
4027
4028 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4029 {
4030         struct e1000_adapter *adapter = netdev_priv(netdev);
4031
4032         if (e1000_desc_unused(adapter->tx_ring) >= size)
4033                 return 0;
4034         return __e1000_maybe_stop_tx(netdev, size);
4035 }
4036
4037 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4038 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4039 {
4040         struct e1000_adapter *adapter = netdev_priv(netdev);
4041         struct e1000_ring *tx_ring = adapter->tx_ring;
4042         unsigned int first;
4043         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4044         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4045         unsigned int tx_flags = 0;
4046         unsigned int len = skb->len - skb->data_len;
4047         unsigned int nr_frags;
4048         unsigned int mss;
4049         int count = 0;
4050         int tso;
4051         unsigned int f;
4052
4053         if (test_bit(__E1000_DOWN, &adapter->state)) {
4054                 dev_kfree_skb_any(skb);
4055                 return NETDEV_TX_OK;
4056         }
4057
4058         if (skb->len <= 0) {
4059                 dev_kfree_skb_any(skb);
4060                 return NETDEV_TX_OK;
4061         }
4062
4063         mss = skb_shinfo(skb)->gso_size;
4064         /*
4065          * The controller does a simple calculation to
4066          * make sure there is enough room in the FIFO before
4067          * initiating the DMA for each buffer.  The calc is:
4068          * 4 = ceil(buffer len/mss).  To make sure we don't
4069          * overrun the FIFO, adjust the max buffer len if mss
4070          * drops.
4071          */
4072         if (mss) {
4073                 u8 hdr_len;
4074                 max_per_txd = min(mss << 2, max_per_txd);
4075                 max_txd_pwr = fls(max_per_txd) - 1;
4076
4077                 /*
4078                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4079                  * points to just header, pull a few bytes of payload from
4080                  * frags into skb->data
4081                  */
4082                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4083                 /*
4084                  * we do this workaround for ES2LAN, but it is un-necessary,
4085                  * avoiding it could save a lot of cycles
4086                  */
4087                 if (skb->data_len && (hdr_len == len)) {
4088                         unsigned int pull_size;
4089
4090                         pull_size = min((unsigned int)4, skb->data_len);
4091                         if (!__pskb_pull_tail(skb, pull_size)) {
4092                                 e_err("__pskb_pull_tail failed.\n");
4093                                 dev_kfree_skb_any(skb);
4094                                 return NETDEV_TX_OK;
4095                         }
4096                         len = skb->len - skb->data_len;
4097                 }
4098         }
4099
4100         /* reserve a descriptor for the offload context */
4101         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4102                 count++;
4103         count++;
4104
4105         count += TXD_USE_COUNT(len, max_txd_pwr);
4106
4107         nr_frags = skb_shinfo(skb)->nr_frags;
4108         for (f = 0; f < nr_frags; f++)
4109                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4110                                        max_txd_pwr);
4111
4112         if (adapter->hw.mac.tx_pkt_filtering)
4113                 e1000_transfer_dhcp_info(adapter, skb);
4114
4115         /*
4116          * need: count + 2 desc gap to keep tail from touching
4117          * head, otherwise try next time
4118          */
4119         if (e1000_maybe_stop_tx(netdev, count + 2))
4120                 return NETDEV_TX_BUSY;
4121
4122         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4123                 tx_flags |= E1000_TX_FLAGS_VLAN;
4124                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4125         }
4126
4127         first = tx_ring->next_to_use;
4128
4129         tso = e1000_tso(adapter, skb);
4130         if (tso < 0) {
4131                 dev_kfree_skb_any(skb);
4132                 return NETDEV_TX_OK;
4133         }
4134
4135         if (tso)
4136                 tx_flags |= E1000_TX_FLAGS_TSO;
4137         else if (e1000_tx_csum(adapter, skb))
4138                 tx_flags |= E1000_TX_FLAGS_CSUM;
4139
4140         /*
4141          * Old method was to assume IPv4 packet by default if TSO was enabled.
4142          * 82571 hardware supports TSO capabilities for IPv6 as well...
4143          * no longer assume, we must.
4144          */
4145         if (skb->protocol == htons(ETH_P_IP))
4146                 tx_flags |= E1000_TX_FLAGS_IPV4;
4147
4148         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4149         if (count < 0) {
4150                 /* handle pci_map_single() error in e1000_tx_map */
4151                 dev_kfree_skb_any(skb);
4152                 return NETDEV_TX_OK;
4153         }
4154
4155         e1000_tx_queue(adapter, tx_flags, count);
4156
4157         netdev->trans_start = jiffies;
4158
4159         /* Make sure there is space in the ring for the next send. */
4160         e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4161
4162         return NETDEV_TX_OK;
4163 }
4164
4165 /**
4166  * e1000_tx_timeout - Respond to a Tx Hang
4167  * @netdev: network interface device structure
4168  **/
4169 static void e1000_tx_timeout(struct net_device *netdev)
4170 {
4171         struct e1000_adapter *adapter = netdev_priv(netdev);
4172
4173         /* Do the reset outside of interrupt context */
4174         adapter->tx_timeout_count++;
4175         schedule_work(&adapter->reset_task);
4176 }
4177
4178 static void e1000_reset_task(struct work_struct *work)
4179 {
4180         struct e1000_adapter *adapter;
4181         adapter = container_of(work, struct e1000_adapter, reset_task);
4182
4183         e1000e_reinit_locked(adapter);
4184 }
4185
4186 /**
4187  * e1000_get_stats - Get System Network Statistics
4188  * @netdev: network interface device structure
4189  *
4190  * Returns the address of the device statistics structure.
4191  * The statistics are actually updated from the timer callback.
4192  **/
4193 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4194 {
4195         struct e1000_adapter *adapter = netdev_priv(netdev);
4196
4197         /* only return the current stats */
4198         return &adapter->net_stats;
4199 }
4200
4201 /**
4202  * e1000_change_mtu - Change the Maximum Transfer Unit
4203  * @netdev: network interface device structure
4204  * @new_mtu: new value for maximum frame size
4205  *
4206  * Returns 0 on success, negative on failure
4207  **/
4208 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4209 {
4210         struct e1000_adapter *adapter = netdev_priv(netdev);
4211         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4212
4213         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4214             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
4215                 e_err("Invalid MTU setting\n");
4216                 return -EINVAL;
4217         }
4218
4219         /* Jumbo frame size limits */
4220         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
4221                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4222                         e_err("Jumbo Frames not supported.\n");
4223                         return -EINVAL;
4224                 }
4225                 if (adapter->hw.phy.type == e1000_phy_ife) {
4226                         e_err("Jumbo Frames not supported.\n");
4227                         return -EINVAL;
4228                 }
4229         }
4230
4231 #define MAX_STD_JUMBO_FRAME_SIZE 9234
4232         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
4233                 e_err("MTU > 9216 not supported.\n");
4234                 return -EINVAL;
4235         }
4236
4237         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4238                 msleep(1);
4239         /* e1000e_down has a dependency on max_frame_size */
4240         adapter->max_frame_size = max_frame;
4241         if (netif_running(netdev))
4242                 e1000e_down(adapter);
4243
4244         /*
4245          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4246          * means we reserve 2 more, this pushes us to allocate from the next
4247          * larger slab size.
4248          * i.e. RXBUFFER_2048 --> size-4096 slab
4249          * However with the new *_jumbo_rx* routines, jumbo receives will use
4250          * fragmented skbs
4251          */
4252
4253         if (max_frame <= 256)
4254                 adapter->rx_buffer_len = 256;
4255         else if (max_frame <= 512)
4256                 adapter->rx_buffer_len = 512;
4257         else if (max_frame <= 1024)
4258                 adapter->rx_buffer_len = 1024;
4259         else if (max_frame <= 2048)
4260                 adapter->rx_buffer_len = 2048;
4261         else
4262                 adapter->rx_buffer_len = 4096;
4263
4264         /* adjust allocation if LPE protects us, and we aren't using SBP */
4265         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4266              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4267                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4268                                          + ETH_FCS_LEN;
4269
4270         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4271         netdev->mtu = new_mtu;
4272
4273         if (netif_running(netdev))
4274                 e1000e_up(adapter);
4275         else
4276                 e1000e_reset(adapter);
4277
4278         clear_bit(__E1000_RESETTING, &adapter->state);
4279
4280         return 0;
4281 }
4282
4283 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4284                            int cmd)
4285 {
4286         struct e1000_adapter *adapter = netdev_priv(netdev);
4287         struct mii_ioctl_data *data = if_mii(ifr);
4288
4289         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4290                 return -EOPNOTSUPP;
4291
4292         switch (cmd) {
4293         case SIOCGMIIPHY:
4294                 data->phy_id = adapter->hw.phy.addr;
4295                 break;
4296         case SIOCGMIIREG:
4297                 if (!capable(CAP_NET_ADMIN))
4298                         return -EPERM;
4299                 switch (data->reg_num & 0x1F) {
4300                 case MII_BMCR:
4301                         data->val_out = adapter->phy_regs.bmcr;
4302                         break;
4303                 case MII_BMSR:
4304                         data->val_out = adapter->phy_regs.bmsr;
4305                         break;
4306                 case MII_PHYSID1:
4307                         data->val_out = (adapter->hw.phy.id >> 16);
4308                         break;
4309                 case MII_PHYSID2:
4310                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4311                         break;
4312                 case MII_ADVERTISE:
4313                         data->val_out = adapter->phy_regs.advertise;
4314                         break;
4315                 case MII_LPA:
4316                         data->val_out = adapter->phy_regs.lpa;
4317                         break;
4318                 case MII_EXPANSION:
4319                         data->val_out = adapter->phy_regs.expansion;
4320                         break;
4321                 case MII_CTRL1000:
4322                         data->val_out = adapter->phy_regs.ctrl1000;
4323                         break;
4324                 case MII_STAT1000:
4325                         data->val_out = adapter->phy_regs.stat1000;
4326                         break;
4327                 case MII_ESTATUS:
4328                         data->val_out = adapter->phy_regs.estatus;
4329                         break;
4330                 default:
4331                         return -EIO;
4332                 }
4333                 break;
4334         case SIOCSMIIREG:
4335         default:
4336                 return -EOPNOTSUPP;
4337         }
4338         return 0;
4339 }
4340
4341 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4342 {
4343         switch (cmd) {
4344         case SIOCGMIIPHY:
4345         case SIOCGMIIREG:
4346         case SIOCSMIIREG:
4347                 return e1000_mii_ioctl(netdev, ifr, cmd);
4348         default:
4349                 return -EOPNOTSUPP;
4350         }
4351 }
4352
4353 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4354 {
4355         struct net_device *netdev = pci_get_drvdata(pdev);
4356         struct e1000_adapter *adapter = netdev_priv(netdev);
4357         struct e1000_hw *hw = &adapter->hw;
4358         u32 ctrl, ctrl_ext, rctl, status;
4359         u32 wufc = adapter->wol;
4360         int retval = 0;
4361
4362         netif_device_detach(netdev);
4363
4364         if (netif_running(netdev)) {
4365                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4366                 e1000e_down(adapter);
4367                 e1000_free_irq(adapter);
4368         }
4369         e1000e_reset_interrupt_capability(adapter);
4370
4371         retval = pci_save_state(pdev);
4372         if (retval)
4373                 return retval;
4374
4375         status = er32(STATUS);
4376         if (status & E1000_STATUS_LU)
4377                 wufc &= ~E1000_WUFC_LNKC;
4378
4379         if (wufc) {
4380                 e1000_setup_rctl(adapter);
4381                 e1000_set_multi(netdev);
4382
4383                 /* turn on all-multi mode if wake on multicast is enabled */
4384                 if (wufc & E1000_WUFC_MC) {
4385                         rctl = er32(RCTL);
4386                         rctl |= E1000_RCTL_MPE;
4387                         ew32(RCTL, rctl);
4388                 }
4389
4390                 ctrl = er32(CTRL);
4391                 /* advertise wake from D3Cold */
4392                 #define E1000_CTRL_ADVD3WUC 0x00100000
4393                 /* phy power management enable */
4394                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4395                 ctrl |= E1000_CTRL_ADVD3WUC |
4396                         E1000_CTRL_EN_PHY_PWR_MGMT;
4397                 ew32(CTRL, ctrl);
4398
4399                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4400                     adapter->hw.phy.media_type ==
4401                     e1000_media_type_internal_serdes) {
4402                         /* keep the laser running in D3 */
4403                         ctrl_ext = er32(CTRL_EXT);
4404                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4405                         ew32(CTRL_EXT, ctrl_ext);
4406                 }
4407
4408                 if (adapter->flags & FLAG_IS_ICH)
4409                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4410
4411                 /* Allow time for pending master requests to run */
4412                 e1000e_disable_pcie_master(&adapter->hw);
4413
4414                 ew32(WUC, E1000_WUC_PME_EN);
4415                 ew32(WUFC, wufc);
4416                 pci_enable_wake(pdev, PCI_D3hot, 1);
4417                 pci_enable_wake(pdev, PCI_D3cold, 1);
4418         } else {
4419                 ew32(WUC, 0);
4420                 ew32(WUFC, 0);
4421                 pci_enable_wake(pdev, PCI_D3hot, 0);
4422                 pci_enable_wake(pdev, PCI_D3cold, 0);
4423         }
4424
4425         /* make sure adapter isn't asleep if manageability is enabled */
4426         if (adapter->flags & FLAG_MNG_PT_ENABLED) {
4427                 pci_enable_wake(pdev, PCI_D3hot, 1);
4428                 pci_enable_wake(pdev, PCI_D3cold, 1);
4429         }
4430
4431         if (adapter->hw.phy.type == e1000_phy_igp_3)
4432                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4433
4434         /*
4435          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4436          * would have already happened in close and is redundant.
4437          */
4438         e1000_release_hw_control(adapter);
4439
4440         pci_disable_device(pdev);
4441
4442         /*
4443          * The pci-e switch on some quad port adapters will report a
4444          * correctable error when the MAC transitions from D0 to D3.  To
4445          * prevent this we need to mask off the correctable errors on the
4446          * downstream port of the pci-e switch.
4447          */
4448         if (adapter->flags & FLAG_IS_QUAD_PORT) {
4449                 struct pci_dev *us_dev = pdev->bus->self;
4450                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
4451                 u16 devctl;
4452
4453                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
4454                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
4455                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
4456
4457                 pci_set_power_state(pdev, pci_choose_state(pdev, state));
4458
4459                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
4460         } else {
4461                 pci_set_power_state(pdev, pci_choose_state(pdev, state));
4462         }
4463
4464         return 0;
4465 }
4466
4467 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4468 {
4469         int pos;
4470         u16 val;
4471
4472         /*
4473          * 82573 workaround - disable L1 ASPM on mobile chipsets
4474          *
4475          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4476          * resulting in lost data or garbage information on the pci-e link
4477          * level. This could result in (false) bad EEPROM checksum errors,
4478          * long ping times (up to 2s) or even a system freeze/hang.
4479          *
4480          * Unfortunately this feature saves about 1W power consumption when
4481          * active.
4482          */
4483         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4484         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4485         if (val & 0x2) {
4486                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4487                 val &= ~0x2;
4488                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4489         }
4490 }
4491
4492 #ifdef CONFIG_PM
4493 static int e1000_resume(struct pci_dev *pdev)
4494 {
4495         struct net_device *netdev = pci_get_drvdata(pdev);
4496         struct e1000_adapter *adapter = netdev_priv(netdev);
4497         struct e1000_hw *hw = &adapter->hw;
4498         u32 err;
4499
4500         pci_set_power_state(pdev, PCI_D0);
4501         pci_restore_state(pdev);
4502         e1000e_disable_l1aspm(pdev);
4503
4504         err = pci_enable_device_mem(pdev);
4505         if (err) {
4506                 dev_err(&pdev->dev,
4507                         "Cannot enable PCI device from suspend\n");
4508                 return err;
4509         }
4510
4511         /* AER (Advanced Error Reporting) hooks */
4512         err = pci_enable_pcie_error_reporting(pdev);
4513         if (err) {
4514                 dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
4515                                     "0x%x\n", err);
4516                 /* non-fatal, continue */
4517         }
4518
4519         pci_set_master(pdev);
4520
4521         pci_enable_wake(pdev, PCI_D3hot, 0);
4522         pci_enable_wake(pdev, PCI_D3cold, 0);
4523
4524         e1000e_set_interrupt_capability(adapter);
4525         if (netif_running(netdev)) {
4526                 err = e1000_request_irq(adapter);
4527                 if (err)
4528                         return err;
4529         }
4530
4531         e1000e_power_up_phy(adapter);
4532         e1000e_reset(adapter);
4533         ew32(WUS, ~0);
4534
4535         e1000_init_manageability(adapter);
4536
4537         if (netif_running(netdev))
4538                 e1000e_up(adapter);
4539
4540         netif_device_attach(netdev);
4541
4542         /*
4543          * If the controller has AMT, do not set DRV_LOAD until the interface
4544          * is up.  For all other cases, let the f/w know that the h/w is now
4545          * under the control of the driver.
4546          */
4547         if (!(adapter->flags & FLAG_HAS_AMT))
4548                 e1000_get_hw_control(adapter);
4549
4550         return 0;
4551 }
4552 #endif
4553
4554 static void e1000_shutdown(struct pci_dev *pdev)
4555 {
4556         e1000_suspend(pdev, PMSG_SUSPEND);
4557 }
4558
4559 #ifdef CONFIG_NET_POLL_CONTROLLER
4560 /*
4561  * Polling 'interrupt' - used by things like netconsole to send skbs
4562  * without having to re-enable interrupts. It's not called while
4563  * the interrupt routine is executing.
4564  */
4565 static void e1000_netpoll(struct net_device *netdev)
4566 {
4567         struct e1000_adapter *adapter = netdev_priv(netdev);
4568
4569         disable_irq(adapter->pdev->irq);
4570         e1000_intr(adapter->pdev->irq, netdev);
4571
4572         enable_irq(adapter->pdev->irq);
4573 }
4574 #endif
4575
4576 /**
4577  * e1000_io_error_detected - called when PCI error is detected
4578  * @pdev: Pointer to PCI device
4579  * @state: The current pci connection state
4580  *
4581  * This function is called after a PCI bus error affecting
4582  * this device has been detected.
4583  */
4584 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4585                                                 pci_channel_state_t state)
4586 {
4587         struct net_device *netdev = pci_get_drvdata(pdev);
4588         struct e1000_adapter *adapter = netdev_priv(netdev);
4589
4590         netif_device_detach(netdev);
4591
4592         if (netif_running(netdev))
4593                 e1000e_down(adapter);
4594         pci_disable_device(pdev);
4595
4596         /* Request a slot slot reset. */
4597         return PCI_ERS_RESULT_NEED_RESET;
4598 }
4599
4600 /**
4601  * e1000_io_slot_reset - called after the pci bus has been reset.
4602  * @pdev: Pointer to PCI device
4603  *
4604  * Restart the card from scratch, as if from a cold-boot. Implementation
4605  * resembles the first-half of the e1000_resume routine.
4606  */
4607 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4608 {
4609         struct net_device *netdev = pci_get_drvdata(pdev);
4610         struct e1000_adapter *adapter = netdev_priv(netdev);
4611         struct e1000_hw *hw = &adapter->hw;
4612         int err;
4613         pci_ers_result_t result;
4614
4615         e1000e_disable_l1aspm(pdev);
4616         err = pci_enable_device_mem(pdev);
4617         if (err) {
4618                 dev_err(&pdev->dev,
4619                         "Cannot re-enable PCI device after reset.\n");
4620                 result = PCI_ERS_RESULT_DISCONNECT;
4621         } else {
4622                 pci_set_master(pdev);
4623                 pci_restore_state(pdev);
4624
4625                 pci_enable_wake(pdev, PCI_D3hot, 0);
4626                 pci_enable_wake(pdev, PCI_D3cold, 0);
4627
4628                 e1000e_reset(adapter);
4629                 ew32(WUS, ~0);
4630                 result = PCI_ERS_RESULT_RECOVERED;
4631         }
4632
4633         pci_cleanup_aer_uncorrect_error_status(pdev);
4634
4635         return result;
4636 }
4637
4638 /**
4639  * e1000_io_resume - called when traffic can start flowing again.
4640  * @pdev: Pointer to PCI device
4641  *
4642  * This callback is called when the error recovery driver tells us that
4643  * its OK to resume normal operation. Implementation resembles the
4644  * second-half of the e1000_resume routine.
4645  */
4646 static void e1000_io_resume(struct pci_dev *pdev)
4647 {
4648         struct net_device *netdev = pci_get_drvdata(pdev);
4649         struct e1000_adapter *adapter = netdev_priv(netdev);
4650
4651         e1000_init_manageability(adapter);
4652
4653         if (netif_running(netdev)) {
4654                 if (e1000e_up(adapter)) {
4655                         dev_err(&pdev->dev,
4656                                 "can't bring device back up after reset\n");
4657                         return;
4658                 }
4659         }
4660
4661         netif_device_attach(netdev);
4662
4663         /*
4664          * If the controller has AMT, do not set DRV_LOAD until the interface
4665          * is up.  For all other cases, let the f/w know that the h/w is now
4666          * under the control of the driver.
4667          */
4668         if (!(adapter->flags & FLAG_HAS_AMT))
4669                 e1000_get_hw_control(adapter);
4670
4671 }
4672
4673 static void e1000_print_device_info(struct e1000_adapter *adapter)
4674 {
4675         struct e1000_hw *hw = &adapter->hw;
4676         struct net_device *netdev = adapter->netdev;
4677         u32 pba_num;
4678
4679         /* print bus type/speed/width info */
4680         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
4681                /* bus width */
4682                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4683                 "Width x1"),
4684                /* MAC address */
4685                netdev->dev_addr);
4686         e_info("Intel(R) PRO/%s Network Connection\n",
4687                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4688         e1000e_read_pba_num(hw, &pba_num);
4689         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4690                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4691 }
4692
4693 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4694 {
4695         struct e1000_hw *hw = &adapter->hw;
4696         int ret_val;
4697         u16 buf = 0;
4698
4699         if (hw->mac.type != e1000_82573)
4700                 return;
4701
4702         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4703         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
4704                 /* Deep Smart Power Down (DSPD) */
4705                 dev_warn(&adapter->pdev->dev,
4706                          "Warning: detected DSPD enabled in EEPROM\n");
4707         }
4708
4709         ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4710         if (!ret_val && (le16_to_cpu(buf) & (3 << 2))) {
4711                 /* ASPM enable */
4712                 dev_warn(&adapter->pdev->dev,
4713                          "Warning: detected ASPM enabled in EEPROM\n");
4714         }
4715 }
4716
4717 static const struct net_device_ops e1000e_netdev_ops = {
4718         .ndo_open               = e1000_open,
4719         .ndo_stop               = e1000_close,
4720         .ndo_start_xmit         = e1000_xmit_frame,
4721         .ndo_get_stats          = e1000_get_stats,
4722         .ndo_set_multicast_list = e1000_set_multi,
4723         .ndo_set_mac_address    = e1000_set_mac,
4724         .ndo_change_mtu         = e1000_change_mtu,
4725         .ndo_do_ioctl           = e1000_ioctl,
4726         .ndo_tx_timeout         = e1000_tx_timeout,
4727         .ndo_validate_addr      = eth_validate_addr,
4728
4729         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
4730         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
4731         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
4732 #ifdef CONFIG_NET_POLL_CONTROLLER
4733         .ndo_poll_controller    = e1000_netpoll,
4734 #endif
4735 };
4736
4737 /**
4738  * e1000_probe - Device Initialization Routine
4739  * @pdev: PCI device information struct
4740  * @ent: entry in e1000_pci_tbl
4741  *
4742  * Returns 0 on success, negative on failure
4743  *
4744  * e1000_probe initializes an adapter identified by a pci_dev structure.
4745  * The OS initialization, configuring of the adapter private structure,
4746  * and a hardware reset occur.
4747  **/
4748 static int __devinit e1000_probe(struct pci_dev *pdev,
4749                                  const struct pci_device_id *ent)
4750 {
4751         struct net_device *netdev;
4752         struct e1000_adapter *adapter;
4753         struct e1000_hw *hw;
4754         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4755         resource_size_t mmio_start, mmio_len;
4756         resource_size_t flash_start, flash_len;
4757
4758         static int cards_found;
4759         int i, err, pci_using_dac;
4760         u16 eeprom_data = 0;
4761         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4762
4763         e1000e_disable_l1aspm(pdev);
4764
4765         err = pci_enable_device_mem(pdev);
4766         if (err)
4767                 return err;
4768
4769         pci_using_dac = 0;
4770         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
4771         if (!err) {
4772                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
4773                 if (!err)
4774                         pci_using_dac = 1;
4775         } else {
4776                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
4777                 if (err) {
4778                         err = pci_set_consistent_dma_mask(pdev,
4779                                                           DMA_32BIT_MASK);
4780                         if (err) {
4781                                 dev_err(&pdev->dev, "No usable DMA "
4782                                         "configuration, aborting\n");
4783                                 goto err_dma;
4784                         }
4785                 }
4786         }
4787
4788         err = pci_request_selected_regions_exclusive(pdev,
4789                                           pci_select_bars(pdev, IORESOURCE_MEM),
4790                                           e1000e_driver_name);
4791         if (err)
4792                 goto err_pci_reg;
4793
4794         pci_set_master(pdev);
4795         /* PCI config space info */
4796         err = pci_save_state(pdev);
4797         if (err)
4798                 goto err_alloc_etherdev;
4799
4800         err = -ENOMEM;
4801         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4802         if (!netdev)
4803                 goto err_alloc_etherdev;
4804
4805         SET_NETDEV_DEV(netdev, &pdev->dev);
4806
4807         pci_set_drvdata(pdev, netdev);
4808         adapter = netdev_priv(netdev);
4809         hw = &adapter->hw;
4810         adapter->netdev = netdev;
4811         adapter->pdev = pdev;
4812         adapter->ei = ei;
4813         adapter->pba = ei->pba;
4814         adapter->flags = ei->flags;
4815         adapter->flags2 = ei->flags2;
4816         adapter->hw.adapter = adapter;
4817         adapter->hw.mac.type = ei->mac;
4818         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4819
4820         mmio_start = pci_resource_start(pdev, 0);
4821         mmio_len = pci_resource_len(pdev, 0);
4822
4823         err = -EIO;
4824         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4825         if (!adapter->hw.hw_addr)
4826                 goto err_ioremap;
4827
4828         if ((adapter->flags & FLAG_HAS_FLASH) &&
4829             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
4830                 flash_start = pci_resource_start(pdev, 1);
4831                 flash_len = pci_resource_len(pdev, 1);
4832                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
4833                 if (!adapter->hw.flash_address)
4834                         goto err_flashmap;
4835         }
4836
4837         /* construct the net_device struct */
4838         netdev->netdev_ops              = &e1000e_netdev_ops;
4839         e1000e_set_ethtool_ops(netdev);
4840         netdev->watchdog_timeo          = 5 * HZ;
4841         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
4842         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
4843
4844         netdev->mem_start = mmio_start;
4845         netdev->mem_end = mmio_start + mmio_len;
4846
4847         adapter->bd_number = cards_found++;
4848
4849         e1000e_check_options(adapter);
4850
4851         /* setup adapter struct */
4852         err = e1000_sw_init(adapter);
4853         if (err)
4854                 goto err_sw_init;
4855
4856         err = -EIO;
4857
4858         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4859         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
4860         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4861
4862         err = ei->get_variants(adapter);
4863         if (err)
4864                 goto err_hw_init;
4865
4866         if ((adapter->flags & FLAG_IS_ICH) &&
4867             (adapter->flags & FLAG_READ_ONLY_NVM))
4868                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
4869
4870         hw->mac.ops.get_bus_info(&adapter->hw);
4871
4872         adapter->hw.phy.autoneg_wait_to_complete = 0;
4873
4874         /* Copper options */
4875         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
4876                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
4877                 adapter->hw.phy.disable_polarity_correction = 0;
4878                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
4879         }
4880
4881         if (e1000_check_reset_block(&adapter->hw))
4882                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
4883
4884         netdev->features = NETIF_F_SG |
4885                            NETIF_F_HW_CSUM |
4886                            NETIF_F_HW_VLAN_TX |
4887                            NETIF_F_HW_VLAN_RX;
4888
4889         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
4890                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
4891
4892         netdev->features |= NETIF_F_TSO;
4893         netdev->features |= NETIF_F_TSO6;
4894
4895         netdev->vlan_features |= NETIF_F_TSO;
4896         netdev->vlan_features |= NETIF_F_TSO6;
4897         netdev->vlan_features |= NETIF_F_HW_CSUM;
4898         netdev->vlan_features |= NETIF_F_SG;
4899
4900         if (pci_using_dac)
4901                 netdev->features |= NETIF_F_HIGHDMA;
4902
4903         if (e1000e_enable_mng_pass_thru(&adapter->hw))
4904                 adapter->flags |= FLAG_MNG_PT_ENABLED;
4905
4906         /*
4907          * before reading the NVM, reset the controller to
4908          * put the device in a known good starting state
4909          */
4910         adapter->hw.mac.ops.reset_hw(&adapter->hw);
4911
4912         /*
4913          * systems with ASPM and others may see the checksum fail on the first
4914          * attempt. Let's give it a few tries
4915          */
4916         for (i = 0;; i++) {
4917                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
4918                         break;
4919                 if (i == 2) {
4920                         e_err("The NVM Checksum Is Not Valid\n");
4921                         err = -EIO;
4922                         goto err_eeprom;
4923                 }
4924         }
4925
4926         e1000_eeprom_checks(adapter);
4927
4928         /* copy the MAC address out of the NVM */
4929         if (e1000e_read_mac_addr(&adapter->hw))
4930                 e_err("NVM Read Error while reading MAC address\n");
4931
4932         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
4933         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
4934
4935         if (!is_valid_ether_addr(netdev->perm_addr)) {
4936                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
4937                 err = -EIO;
4938                 goto err_eeprom;
4939         }
4940
4941         init_timer(&adapter->watchdog_timer);
4942         adapter->watchdog_timer.function = &e1000_watchdog;
4943         adapter->watchdog_timer.data = (unsigned long) adapter;
4944
4945         init_timer(&adapter->phy_info_timer);
4946         adapter->phy_info_timer.function = &e1000_update_phy_info;
4947         adapter->phy_info_timer.data = (unsigned long) adapter;
4948
4949         INIT_WORK(&adapter->reset_task, e1000_reset_task);
4950         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
4951         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
4952         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
4953
4954         /* Initialize link parameters. User can change them with ethtool */
4955         adapter->hw.mac.autoneg = 1;
4956         adapter->fc_autoneg = 1;
4957         adapter->hw.fc.requested_mode = e1000_fc_default;
4958         adapter->hw.fc.current_mode = e1000_fc_default;
4959         adapter->hw.phy.autoneg_advertised = 0x2f;
4960
4961         /* ring size defaults */
4962         adapter->rx_ring->count = 256;
4963         adapter->tx_ring->count = 256;
4964
4965         /*
4966          * Initial Wake on LAN setting - If APM wake is enabled in
4967          * the EEPROM, enable the ACPI Magic Packet filter
4968          */
4969         if (adapter->flags & FLAG_APME_IN_WUC) {
4970                 /* APME bit in EEPROM is mapped to WUC.APME */
4971                 eeprom_data = er32(WUC);
4972                 eeprom_apme_mask = E1000_WUC_APME;
4973         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
4974                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
4975                     (adapter->hw.bus.func == 1))
4976                         e1000_read_nvm(&adapter->hw,
4977                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
4978                 else
4979                         e1000_read_nvm(&adapter->hw,
4980                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
4981         }
4982
4983         /* fetch WoL from EEPROM */
4984         if (eeprom_data & eeprom_apme_mask)
4985                 adapter->eeprom_wol |= E1000_WUFC_MAG;
4986
4987         /*
4988          * now that we have the eeprom settings, apply the special cases
4989          * where the eeprom may be wrong or the board simply won't support
4990          * wake on lan on a particular port
4991          */
4992         if (!(adapter->flags & FLAG_HAS_WOL))
4993                 adapter->eeprom_wol = 0;
4994
4995         /* initialize the wol settings based on the eeprom settings */
4996         adapter->wol = adapter->eeprom_wol;
4997         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
4998
4999         /* save off EEPROM version number */
5000         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5001
5002         /* reset the hardware with the new settings */
5003         e1000e_reset(adapter);
5004
5005         /*
5006          * If the controller has AMT, do not set DRV_LOAD until the interface
5007          * is up.  For all other cases, let the f/w know that the h/w is now
5008          * under the control of the driver.
5009          */
5010         if (!(adapter->flags & FLAG_HAS_AMT))
5011                 e1000_get_hw_control(adapter);
5012
5013         /* tell the stack to leave us alone until e1000_open() is called */
5014         netif_carrier_off(netdev);
5015         netif_tx_stop_all_queues(netdev);
5016
5017         strcpy(netdev->name, "eth%d");
5018         err = register_netdev(netdev);
5019         if (err)
5020                 goto err_register;
5021
5022         e1000_print_device_info(adapter);
5023
5024         return 0;
5025
5026 err_register:
5027         if (!(adapter->flags & FLAG_HAS_AMT))
5028                 e1000_release_hw_control(adapter);
5029 err_eeprom:
5030         if (!e1000_check_reset_block(&adapter->hw))
5031                 e1000_phy_hw_reset(&adapter->hw);
5032 err_hw_init:
5033
5034         kfree(adapter->tx_ring);
5035         kfree(adapter->rx_ring);
5036 err_sw_init:
5037         if (adapter->hw.flash_address)
5038                 iounmap(adapter->hw.flash_address);
5039         e1000e_reset_interrupt_capability(adapter);
5040 err_flashmap:
5041         iounmap(adapter->hw.hw_addr);
5042 err_ioremap:
5043         free_netdev(netdev);
5044 err_alloc_etherdev:
5045         pci_release_selected_regions(pdev,
5046                                      pci_select_bars(pdev, IORESOURCE_MEM));
5047 err_pci_reg:
5048 err_dma:
5049         pci_disable_device(pdev);
5050         return err;
5051 }
5052
5053 /**
5054  * e1000_remove - Device Removal Routine
5055  * @pdev: PCI device information struct
5056  *
5057  * e1000_remove is called by the PCI subsystem to alert the driver
5058  * that it should release a PCI device.  The could be caused by a
5059  * Hot-Plug event, or because the driver is going to be removed from
5060  * memory.
5061  **/
5062 static void __devexit e1000_remove(struct pci_dev *pdev)
5063 {
5064         struct net_device *netdev = pci_get_drvdata(pdev);
5065         struct e1000_adapter *adapter = netdev_priv(netdev);
5066         int err;
5067
5068         /*
5069          * flush_scheduled work may reschedule our watchdog task, so
5070          * explicitly disable watchdog tasks from being rescheduled
5071          */
5072         set_bit(__E1000_DOWN, &adapter->state);
5073         del_timer_sync(&adapter->watchdog_timer);
5074         del_timer_sync(&adapter->phy_info_timer);
5075
5076         flush_scheduled_work();
5077
5078         /*
5079          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5080          * would have already happened in close and is redundant.
5081          */
5082         e1000_release_hw_control(adapter);
5083
5084         unregister_netdev(netdev);
5085
5086         if (!e1000_check_reset_block(&adapter->hw))
5087                 e1000_phy_hw_reset(&adapter->hw);
5088
5089         e1000e_reset_interrupt_capability(adapter);
5090         kfree(adapter->tx_ring);
5091         kfree(adapter->rx_ring);
5092
5093         iounmap(adapter->hw.hw_addr);
5094         if (adapter->hw.flash_address)
5095                 iounmap(adapter->hw.flash_address);
5096         pci_release_selected_regions(pdev,
5097                                      pci_select_bars(pdev, IORESOURCE_MEM));
5098
5099         free_netdev(netdev);
5100
5101         /* AER disable */
5102         err = pci_disable_pcie_error_reporting(pdev);
5103         if (err)
5104                 dev_err(&pdev->dev,
5105                         "pci_disable_pcie_error_reporting failed 0x%x\n", err);
5106
5107         pci_disable_device(pdev);
5108 }
5109
5110 /* PCI Error Recovery (ERS) */
5111 static struct pci_error_handlers e1000_err_handler = {
5112         .error_detected = e1000_io_error_detected,
5113         .slot_reset = e1000_io_slot_reset,
5114         .resume = e1000_io_resume,
5115 };
5116
5117 static struct pci_device_id e1000_pci_tbl[] = {
5118         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5119         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5120         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5121         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5122         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5123         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5124         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5125         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5126         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5127
5128         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5129         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5130         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5131         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5132
5133         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5134         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5135         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5136
5137         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5138         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
5139
5140         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5141           board_80003es2lan },
5142         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5143           board_80003es2lan },
5144         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5145           board_80003es2lan },
5146         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5147           board_80003es2lan },
5148
5149         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5150         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5151         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5152         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5153         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5154         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5155         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5156
5157         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5158         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5159         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5160         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5161         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5162         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5163         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5164         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5165         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5166
5167         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5168         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5169         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5170
5171         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5172         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5173
5174         { }     /* terminate list */
5175 };
5176 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5177
5178 /* PCI Device API Driver */
5179 static struct pci_driver e1000_driver = {
5180         .name     = e1000e_driver_name,
5181         .id_table = e1000_pci_tbl,
5182         .probe    = e1000_probe,
5183         .remove   = __devexit_p(e1000_remove),
5184 #ifdef CONFIG_PM
5185         /* Power Management Hooks */
5186         .suspend  = e1000_suspend,
5187         .resume   = e1000_resume,
5188 #endif
5189         .shutdown = e1000_shutdown,
5190         .err_handler = &e1000_err_handler
5191 };
5192
5193 /**
5194  * e1000_init_module - Driver Registration Routine
5195  *
5196  * e1000_init_module is the first routine called when the driver is
5197  * loaded. All it does is register with the PCI subsystem.
5198  **/
5199 static int __init e1000_init_module(void)
5200 {
5201         int ret;
5202         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
5203                e1000e_driver_name, e1000e_driver_version);
5204         printk(KERN_INFO "%s: Copyright (c) 1999-2008 Intel Corporation.\n",
5205                e1000e_driver_name);
5206         ret = pci_register_driver(&e1000_driver);
5207         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name,
5208                                PM_QOS_DEFAULT_VALUE);
5209                                 
5210         return ret;
5211 }
5212 module_init(e1000_init_module);
5213
5214 /**
5215  * e1000_exit_module - Driver Exit Cleanup Routine
5216  *
5217  * e1000_exit_module is called just before the driver is removed
5218  * from memory.
5219  **/
5220 static void __exit e1000_exit_module(void)
5221 {
5222         pci_unregister_driver(&e1000_driver);
5223         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name);
5224 }
5225 module_exit(e1000_exit_module);
5226
5227
5228 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5229 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5230 MODULE_LICENSE("GPL");
5231 MODULE_VERSION(DRV_VERSION);
5232
5233 /* e1000_main.c */