e74eb3c606e07910c4f7e09f4b12ee7b95508043
[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_80003es2lan]     = &e1000_es2_info,
61         [board_ich8lan]         = &e1000_ich8_info,
62         [board_ich9lan]         = &e1000_ich9_info,
63         [board_ich10lan]        = &e1000_ich10_info,
64 };
65
66 #ifdef DEBUG
67 /**
68  * e1000_get_hw_dev_name - return device name string
69  * used by hardware layer to print debugging information
70  **/
71 char *e1000e_get_hw_dev_name(struct e1000_hw *hw)
72 {
73         return hw->adapter->netdev->name;
74 }
75 #endif
76
77 /**
78  * e1000_desc_unused - calculate if we have unused descriptors
79  **/
80 static int e1000_desc_unused(struct e1000_ring *ring)
81 {
82         if (ring->next_to_clean > ring->next_to_use)
83                 return ring->next_to_clean - ring->next_to_use - 1;
84
85         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
86 }
87
88 /**
89  * e1000_receive_skb - helper function to handle Rx indications
90  * @adapter: board private structure
91  * @status: descriptor status field as written by hardware
92  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
93  * @skb: pointer to sk_buff to be indicated to stack
94  **/
95 static void e1000_receive_skb(struct e1000_adapter *adapter,
96                               struct net_device *netdev,
97                               struct sk_buff *skb,
98                               u8 status, __le16 vlan)
99 {
100         skb->protocol = eth_type_trans(skb, netdev);
101
102         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
103                 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
104                                  le16_to_cpu(vlan), skb);
105         else
106                 napi_gro_receive(&adapter->napi, skb);
107 }
108
109 /**
110  * e1000_rx_checksum - Receive Checksum Offload for 82543
111  * @adapter:     board private structure
112  * @status_err:  receive descriptor status and error fields
113  * @csum:       receive descriptor csum field
114  * @sk_buff:     socket buffer with received data
115  **/
116 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
117                               u32 csum, struct sk_buff *skb)
118 {
119         u16 status = (u16)status_err;
120         u8 errors = (u8)(status_err >> 24);
121         skb->ip_summed = CHECKSUM_NONE;
122
123         /* Ignore Checksum bit is set */
124         if (status & E1000_RXD_STAT_IXSM)
125                 return;
126         /* TCP/UDP checksum error bit is set */
127         if (errors & E1000_RXD_ERR_TCPE) {
128                 /* let the stack verify checksum errors */
129                 adapter->hw_csum_err++;
130                 return;
131         }
132
133         /* TCP/UDP Checksum has not been calculated */
134         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
135                 return;
136
137         /* It must be a TCP or UDP packet with a valid checksum */
138         if (status & E1000_RXD_STAT_TCPCS) {
139                 /* TCP checksum is good */
140                 skb->ip_summed = CHECKSUM_UNNECESSARY;
141         } else {
142                 /*
143                  * IP fragment with UDP payload
144                  * Hardware complements the payload checksum, so we undo it
145                  * and then put the value in host order for further stack use.
146                  */
147                 __sum16 sum = (__force __sum16)htons(csum);
148                 skb->csum = csum_unfold(~sum);
149                 skb->ip_summed = CHECKSUM_COMPLETE;
150         }
151         adapter->hw_csum_good++;
152 }
153
154 /**
155  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
156  * @adapter: address of board private structure
157  **/
158 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
159                                    int cleaned_count)
160 {
161         struct net_device *netdev = adapter->netdev;
162         struct pci_dev *pdev = adapter->pdev;
163         struct e1000_ring *rx_ring = adapter->rx_ring;
164         struct e1000_rx_desc *rx_desc;
165         struct e1000_buffer *buffer_info;
166         struct sk_buff *skb;
167         unsigned int i;
168         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
169
170         i = rx_ring->next_to_use;
171         buffer_info = &rx_ring->buffer_info[i];
172
173         while (cleaned_count--) {
174                 skb = buffer_info->skb;
175                 if (skb) {
176                         skb_trim(skb, 0);
177                         goto map_skb;
178                 }
179
180                 skb = netdev_alloc_skb(netdev, bufsz);
181                 if (!skb) {
182                         /* Better luck next round */
183                         adapter->alloc_rx_buff_failed++;
184                         break;
185                 }
186
187                 /*
188                  * Make buffer alignment 2 beyond a 16 byte boundary
189                  * this will result in a 16 byte aligned IP header after
190                  * the 14 byte MAC header is removed
191                  */
192                 skb_reserve(skb, NET_IP_ALIGN);
193
194                 buffer_info->skb = skb;
195 map_skb:
196                 buffer_info->dma = pci_map_single(pdev, skb->data,
197                                                   adapter->rx_buffer_len,
198                                                   PCI_DMA_FROMDEVICE);
199                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
200                         dev_err(&pdev->dev, "RX DMA map failed\n");
201                         adapter->rx_dma_failed++;
202                         break;
203                 }
204
205                 rx_desc = E1000_RX_DESC(*rx_ring, i);
206                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
207
208                 i++;
209                 if (i == rx_ring->count)
210                         i = 0;
211                 buffer_info = &rx_ring->buffer_info[i];
212         }
213
214         if (rx_ring->next_to_use != i) {
215                 rx_ring->next_to_use = i;
216                 if (i-- == 0)
217                         i = (rx_ring->count - 1);
218
219                 /*
220                  * Force memory writes to complete before letting h/w
221                  * know there are new descriptors to fetch.  (Only
222                  * applicable for weak-ordered memory model archs,
223                  * such as IA-64).
224                  */
225                 wmb();
226                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
227         }
228 }
229
230 /**
231  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
232  * @adapter: address of board private structure
233  **/
234 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
235                                       int cleaned_count)
236 {
237         struct net_device *netdev = adapter->netdev;
238         struct pci_dev *pdev = adapter->pdev;
239         union e1000_rx_desc_packet_split *rx_desc;
240         struct e1000_ring *rx_ring = adapter->rx_ring;
241         struct e1000_buffer *buffer_info;
242         struct e1000_ps_page *ps_page;
243         struct sk_buff *skb;
244         unsigned int i, j;
245
246         i = rx_ring->next_to_use;
247         buffer_info = &rx_ring->buffer_info[i];
248
249         while (cleaned_count--) {
250                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
251
252                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
253                         ps_page = &buffer_info->ps_pages[j];
254                         if (j >= adapter->rx_ps_pages) {
255                                 /* all unused desc entries get hw null ptr */
256                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
257                                 continue;
258                         }
259                         if (!ps_page->page) {
260                                 ps_page->page = alloc_page(GFP_ATOMIC);
261                                 if (!ps_page->page) {
262                                         adapter->alloc_rx_buff_failed++;
263                                         goto no_buffers;
264                                 }
265                                 ps_page->dma = pci_map_page(pdev,
266                                                    ps_page->page,
267                                                    0, PAGE_SIZE,
268                                                    PCI_DMA_FROMDEVICE);
269                                 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
270                                         dev_err(&adapter->pdev->dev,
271                                           "RX DMA page map failed\n");
272                                         adapter->rx_dma_failed++;
273                                         goto no_buffers;
274                                 }
275                         }
276                         /*
277                          * Refresh the desc even if buffer_addrs
278                          * didn't change because each write-back
279                          * erases this info.
280                          */
281                         rx_desc->read.buffer_addr[j+1] =
282                              cpu_to_le64(ps_page->dma);
283                 }
284
285                 skb = netdev_alloc_skb(netdev,
286                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
287
288                 if (!skb) {
289                         adapter->alloc_rx_buff_failed++;
290                         break;
291                 }
292
293                 /*
294                  * Make buffer alignment 2 beyond a 16 byte boundary
295                  * this will result in a 16 byte aligned IP header after
296                  * the 14 byte MAC header is removed
297                  */
298                 skb_reserve(skb, NET_IP_ALIGN);
299
300                 buffer_info->skb = skb;
301                 buffer_info->dma = pci_map_single(pdev, skb->data,
302                                                   adapter->rx_ps_bsize0,
303                                                   PCI_DMA_FROMDEVICE);
304                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
305                         dev_err(&pdev->dev, "RX DMA map failed\n");
306                         adapter->rx_dma_failed++;
307                         /* cleanup skb */
308                         dev_kfree_skb_any(skb);
309                         buffer_info->skb = NULL;
310                         break;
311                 }
312
313                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
314
315                 i++;
316                 if (i == rx_ring->count)
317                         i = 0;
318                 buffer_info = &rx_ring->buffer_info[i];
319         }
320
321 no_buffers:
322         if (rx_ring->next_to_use != i) {
323                 rx_ring->next_to_use = i;
324
325                 if (!(i--))
326                         i = (rx_ring->count - 1);
327
328                 /*
329                  * Force memory writes to complete before letting h/w
330                  * know there are new descriptors to fetch.  (Only
331                  * applicable for weak-ordered memory model archs,
332                  * such as IA-64).
333                  */
334                 wmb();
335                 /*
336                  * Hardware increments by 16 bytes, but packet split
337                  * descriptors are 32 bytes...so we increment tail
338                  * twice as much.
339                  */
340                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
341         }
342 }
343
344 /**
345  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
346  * @adapter: address of board private structure
347  * @cleaned_count: number of buffers to allocate this pass
348  **/
349
350 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
351                                          int cleaned_count)
352 {
353         struct net_device *netdev = adapter->netdev;
354         struct pci_dev *pdev = adapter->pdev;
355         struct e1000_rx_desc *rx_desc;
356         struct e1000_ring *rx_ring = adapter->rx_ring;
357         struct e1000_buffer *buffer_info;
358         struct sk_buff *skb;
359         unsigned int i;
360         unsigned int bufsz = 256 -
361                              16 /* for skb_reserve */ -
362                              NET_IP_ALIGN;
363
364         i = rx_ring->next_to_use;
365         buffer_info = &rx_ring->buffer_info[i];
366
367         while (cleaned_count--) {
368                 skb = buffer_info->skb;
369                 if (skb) {
370                         skb_trim(skb, 0);
371                         goto check_page;
372                 }
373
374                 skb = netdev_alloc_skb(netdev, bufsz);
375                 if (unlikely(!skb)) {
376                         /* Better luck next round */
377                         adapter->alloc_rx_buff_failed++;
378                         break;
379                 }
380
381                 /* Make buffer alignment 2 beyond a 16 byte boundary
382                  * this will result in a 16 byte aligned IP header after
383                  * the 14 byte MAC header is removed
384                  */
385                 skb_reserve(skb, NET_IP_ALIGN);
386
387                 buffer_info->skb = skb;
388 check_page:
389                 /* allocate a new page if necessary */
390                 if (!buffer_info->page) {
391                         buffer_info->page = alloc_page(GFP_ATOMIC);
392                         if (unlikely(!buffer_info->page)) {
393                                 adapter->alloc_rx_buff_failed++;
394                                 break;
395                         }
396                 }
397
398                 if (!buffer_info->dma)
399                         buffer_info->dma = pci_map_page(pdev,
400                                                         buffer_info->page, 0,
401                                                         PAGE_SIZE,
402                                                         PCI_DMA_FROMDEVICE);
403
404                 rx_desc = E1000_RX_DESC(*rx_ring, i);
405                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
406
407                 if (unlikely(++i == rx_ring->count))
408                         i = 0;
409                 buffer_info = &rx_ring->buffer_info[i];
410         }
411
412         if (likely(rx_ring->next_to_use != i)) {
413                 rx_ring->next_to_use = i;
414                 if (unlikely(i-- == 0))
415                         i = (rx_ring->count - 1);
416
417                 /* Force memory writes to complete before letting h/w
418                  * know there are new descriptors to fetch.  (Only
419                  * applicable for weak-ordered memory model archs,
420                  * such as IA-64). */
421                 wmb();
422                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
423         }
424 }
425
426 /**
427  * e1000_clean_rx_irq - Send received data up the network stack; legacy
428  * @adapter: board private structure
429  *
430  * the return value indicates whether actual cleaning was done, there
431  * is no guarantee that everything was cleaned
432  **/
433 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
434                                int *work_done, int work_to_do)
435 {
436         struct net_device *netdev = adapter->netdev;
437         struct pci_dev *pdev = adapter->pdev;
438         struct e1000_ring *rx_ring = adapter->rx_ring;
439         struct e1000_rx_desc *rx_desc, *next_rxd;
440         struct e1000_buffer *buffer_info, *next_buffer;
441         u32 length;
442         unsigned int i;
443         int cleaned_count = 0;
444         bool cleaned = 0;
445         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
446
447         i = rx_ring->next_to_clean;
448         rx_desc = E1000_RX_DESC(*rx_ring, i);
449         buffer_info = &rx_ring->buffer_info[i];
450
451         while (rx_desc->status & E1000_RXD_STAT_DD) {
452                 struct sk_buff *skb;
453                 u8 status;
454
455                 if (*work_done >= work_to_do)
456                         break;
457                 (*work_done)++;
458
459                 status = rx_desc->status;
460                 skb = buffer_info->skb;
461                 buffer_info->skb = NULL;
462
463                 prefetch(skb->data - NET_IP_ALIGN);
464
465                 i++;
466                 if (i == rx_ring->count)
467                         i = 0;
468                 next_rxd = E1000_RX_DESC(*rx_ring, i);
469                 prefetch(next_rxd);
470
471                 next_buffer = &rx_ring->buffer_info[i];
472
473                 cleaned = 1;
474                 cleaned_count++;
475                 pci_unmap_single(pdev,
476                                  buffer_info->dma,
477                                  adapter->rx_buffer_len,
478                                  PCI_DMA_FROMDEVICE);
479                 buffer_info->dma = 0;
480
481                 length = le16_to_cpu(rx_desc->length);
482
483                 /* !EOP means multiple descriptors were used to store a single
484                  * packet, also make sure the frame isn't just CRC only */
485                 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
486                         /* All receives must fit into a single buffer */
487                         e_dbg("%s: Receive packet consumed multiple buffers\n",
488                               netdev->name);
489                         /* recycle */
490                         buffer_info->skb = skb;
491                         goto next_desc;
492                 }
493
494                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
495                         /* recycle */
496                         buffer_info->skb = skb;
497                         goto next_desc;
498                 }
499
500                 /* adjust length to remove Ethernet CRC */
501                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
502                         length -= 4;
503
504                 total_rx_bytes += length;
505                 total_rx_packets++;
506
507                 /*
508                  * code added for copybreak, this should improve
509                  * performance for small packets with large amounts
510                  * of reassembly being done in the stack
511                  */
512                 if (length < copybreak) {
513                         struct sk_buff *new_skb =
514                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
515                         if (new_skb) {
516                                 skb_reserve(new_skb, NET_IP_ALIGN);
517                                 skb_copy_to_linear_data_offset(new_skb,
518                                                                -NET_IP_ALIGN,
519                                                                (skb->data -
520                                                                 NET_IP_ALIGN),
521                                                                (length +
522                                                                 NET_IP_ALIGN));
523                                 /* save the skb in buffer_info as good */
524                                 buffer_info->skb = skb;
525                                 skb = new_skb;
526                         }
527                         /* else just continue with the old one */
528                 }
529                 /* end copybreak code */
530                 skb_put(skb, length);
531
532                 /* Receive Checksum Offload */
533                 e1000_rx_checksum(adapter,
534                                   (u32)(status) |
535                                   ((u32)(rx_desc->errors) << 24),
536                                   le16_to_cpu(rx_desc->csum), skb);
537
538                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
539
540 next_desc:
541                 rx_desc->status = 0;
542
543                 /* return some buffers to hardware, one at a time is too slow */
544                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
545                         adapter->alloc_rx_buf(adapter, cleaned_count);
546                         cleaned_count = 0;
547                 }
548
549                 /* use prefetched values */
550                 rx_desc = next_rxd;
551                 buffer_info = next_buffer;
552         }
553         rx_ring->next_to_clean = i;
554
555         cleaned_count = e1000_desc_unused(rx_ring);
556         if (cleaned_count)
557                 adapter->alloc_rx_buf(adapter, cleaned_count);
558
559         adapter->total_rx_bytes += total_rx_bytes;
560         adapter->total_rx_packets += total_rx_packets;
561         adapter->net_stats.rx_bytes += total_rx_bytes;
562         adapter->net_stats.rx_packets += total_rx_packets;
563         return cleaned;
564 }
565
566 static void e1000_put_txbuf(struct e1000_adapter *adapter,
567                              struct e1000_buffer *buffer_info)
568 {
569         buffer_info->dma = 0;
570         if (buffer_info->skb) {
571                 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
572                               DMA_TO_DEVICE);
573                 dev_kfree_skb_any(buffer_info->skb);
574                 buffer_info->skb = NULL;
575         }
576 }
577
578 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
579 {
580         struct e1000_ring *tx_ring = adapter->tx_ring;
581         unsigned int i = tx_ring->next_to_clean;
582         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
583         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
584
585         /* detected Tx unit hang */
586         e_err("Detected Tx Unit Hang:\n"
587               "  TDH                  <%x>\n"
588               "  TDT                  <%x>\n"
589               "  next_to_use          <%x>\n"
590               "  next_to_clean        <%x>\n"
591               "buffer_info[next_to_clean]:\n"
592               "  time_stamp           <%lx>\n"
593               "  next_to_watch        <%x>\n"
594               "  jiffies              <%lx>\n"
595               "  next_to_watch.status <%x>\n",
596               readl(adapter->hw.hw_addr + tx_ring->head),
597               readl(adapter->hw.hw_addr + tx_ring->tail),
598               tx_ring->next_to_use,
599               tx_ring->next_to_clean,
600               tx_ring->buffer_info[eop].time_stamp,
601               eop,
602               jiffies,
603               eop_desc->upper.fields.status);
604 }
605
606 /**
607  * e1000_clean_tx_irq - Reclaim resources after transmit completes
608  * @adapter: board private structure
609  *
610  * the return value indicates whether actual cleaning was done, there
611  * is no guarantee that everything was cleaned
612  **/
613 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
614 {
615         struct net_device *netdev = adapter->netdev;
616         struct e1000_hw *hw = &adapter->hw;
617         struct e1000_ring *tx_ring = adapter->tx_ring;
618         struct e1000_tx_desc *tx_desc, *eop_desc;
619         struct e1000_buffer *buffer_info;
620         unsigned int i, eop;
621         unsigned int count = 0;
622         bool cleaned = 0;
623         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
624
625         i = tx_ring->next_to_clean;
626         eop = tx_ring->buffer_info[i].next_to_watch;
627         eop_desc = E1000_TX_DESC(*tx_ring, eop);
628
629         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
630                 for (cleaned = 0; !cleaned; ) {
631                         tx_desc = E1000_TX_DESC(*tx_ring, i);
632                         buffer_info = &tx_ring->buffer_info[i];
633                         cleaned = (i == eop);
634
635                         if (cleaned) {
636                                 struct sk_buff *skb = buffer_info->skb;
637                                 unsigned int segs, bytecount;
638                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
639                                 /* multiply data chunks by size of headers */
640                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
641                                             skb->len;
642                                 total_tx_packets += segs;
643                                 total_tx_bytes += bytecount;
644                         }
645
646                         e1000_put_txbuf(adapter, buffer_info);
647                         tx_desc->upper.data = 0;
648
649                         i++;
650                         if (i == tx_ring->count)
651                                 i = 0;
652                 }
653
654                 eop = tx_ring->buffer_info[i].next_to_watch;
655                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
656 #define E1000_TX_WEIGHT 64
657                 /* weight of a sort for tx, to avoid endless transmit cleanup */
658                 if (count++ == E1000_TX_WEIGHT)
659                         break;
660         }
661
662         tx_ring->next_to_clean = i;
663
664 #define TX_WAKE_THRESHOLD 32
665         if (cleaned && netif_carrier_ok(netdev) &&
666                      e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
667                 /* Make sure that anybody stopping the queue after this
668                  * sees the new next_to_clean.
669                  */
670                 smp_mb();
671
672                 if (netif_queue_stopped(netdev) &&
673                     !(test_bit(__E1000_DOWN, &adapter->state))) {
674                         netif_wake_queue(netdev);
675                         ++adapter->restart_queue;
676                 }
677         }
678
679         if (adapter->detect_tx_hung) {
680                 /*
681                  * Detect a transmit hang in hardware, this serializes the
682                  * check with the clearing of time_stamp and movement of i
683                  */
684                 adapter->detect_tx_hung = 0;
685                 /*
686                  * read barrier to make sure that the ->dma member and time
687                  * stamp are updated fully
688                  */
689                 smp_rmb();
690                 if (tx_ring->buffer_info[eop].dma &&
691                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp
692                                + (adapter->tx_timeout_factor * HZ))
693                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
694                         e1000_print_tx_hang(adapter);
695                         netif_stop_queue(netdev);
696                 }
697         }
698         adapter->total_tx_bytes += total_tx_bytes;
699         adapter->total_tx_packets += total_tx_packets;
700         adapter->net_stats.tx_bytes += total_tx_bytes;
701         adapter->net_stats.tx_packets += total_tx_packets;
702         return cleaned;
703 }
704
705 /**
706  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
707  * @adapter: board private structure
708  *
709  * the return value indicates whether actual cleaning was done, there
710  * is no guarantee that everything was cleaned
711  **/
712 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
713                                   int *work_done, int work_to_do)
714 {
715         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
716         struct net_device *netdev = adapter->netdev;
717         struct pci_dev *pdev = adapter->pdev;
718         struct e1000_ring *rx_ring = adapter->rx_ring;
719         struct e1000_buffer *buffer_info, *next_buffer;
720         struct e1000_ps_page *ps_page;
721         struct sk_buff *skb;
722         unsigned int i, j;
723         u32 length, staterr;
724         int cleaned_count = 0;
725         bool cleaned = 0;
726         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
727
728         i = rx_ring->next_to_clean;
729         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
730         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
731         buffer_info = &rx_ring->buffer_info[i];
732
733         while (staterr & E1000_RXD_STAT_DD) {
734                 if (*work_done >= work_to_do)
735                         break;
736                 (*work_done)++;
737                 skb = buffer_info->skb;
738
739                 /* in the packet split case this is header only */
740                 prefetch(skb->data - NET_IP_ALIGN);
741
742                 i++;
743                 if (i == rx_ring->count)
744                         i = 0;
745                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
746                 prefetch(next_rxd);
747
748                 next_buffer = &rx_ring->buffer_info[i];
749
750                 cleaned = 1;
751                 cleaned_count++;
752                 pci_unmap_single(pdev, buffer_info->dma,
753                                  adapter->rx_ps_bsize0,
754                                  PCI_DMA_FROMDEVICE);
755                 buffer_info->dma = 0;
756
757                 if (!(staterr & E1000_RXD_STAT_EOP)) {
758                         e_dbg("%s: Packet Split buffers didn't pick up the "
759                               "full packet\n", netdev->name);
760                         dev_kfree_skb_irq(skb);
761                         goto next_desc;
762                 }
763
764                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
765                         dev_kfree_skb_irq(skb);
766                         goto next_desc;
767                 }
768
769                 length = le16_to_cpu(rx_desc->wb.middle.length0);
770
771                 if (!length) {
772                         e_dbg("%s: Last part of the packet spanning multiple "
773                               "descriptors\n", netdev->name);
774                         dev_kfree_skb_irq(skb);
775                         goto next_desc;
776                 }
777
778                 /* Good Receive */
779                 skb_put(skb, length);
780
781                 {
782                 /*
783                  * this looks ugly, but it seems compiler issues make it
784                  * more efficient than reusing j
785                  */
786                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
787
788                 /*
789                  * page alloc/put takes too long and effects small packet
790                  * throughput, so unsplit small packets and save the alloc/put
791                  * only valid in softirq (napi) context to call kmap_*
792                  */
793                 if (l1 && (l1 <= copybreak) &&
794                     ((length + l1) <= adapter->rx_ps_bsize0)) {
795                         u8 *vaddr;
796
797                         ps_page = &buffer_info->ps_pages[0];
798
799                         /*
800                          * there is no documentation about how to call
801                          * kmap_atomic, so we can't hold the mapping
802                          * very long
803                          */
804                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
805                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
806                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
807                         memcpy(skb_tail_pointer(skb), vaddr, l1);
808                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
809                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
810                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
811
812                         /* remove the CRC */
813                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
814                                 l1 -= 4;
815
816                         skb_put(skb, l1);
817                         goto copydone;
818                 } /* if */
819                 }
820
821                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
822                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
823                         if (!length)
824                                 break;
825
826                         ps_page = &buffer_info->ps_pages[j];
827                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
828                                        PCI_DMA_FROMDEVICE);
829                         ps_page->dma = 0;
830                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
831                         ps_page->page = NULL;
832                         skb->len += length;
833                         skb->data_len += length;
834                         skb->truesize += length;
835                 }
836
837                 /* strip the ethernet crc, problem is we're using pages now so
838                  * this whole operation can get a little cpu intensive
839                  */
840                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
841                         pskb_trim(skb, skb->len - 4);
842
843 copydone:
844                 total_rx_bytes += skb->len;
845                 total_rx_packets++;
846
847                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
848                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
849
850                 if (rx_desc->wb.upper.header_status &
851                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
852                         adapter->rx_hdr_split++;
853
854                 e1000_receive_skb(adapter, netdev, skb,
855                                   staterr, rx_desc->wb.middle.vlan);
856
857 next_desc:
858                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
859                 buffer_info->skb = NULL;
860
861                 /* return some buffers to hardware, one at a time is too slow */
862                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
863                         adapter->alloc_rx_buf(adapter, cleaned_count);
864                         cleaned_count = 0;
865                 }
866
867                 /* use prefetched values */
868                 rx_desc = next_rxd;
869                 buffer_info = next_buffer;
870
871                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
872         }
873         rx_ring->next_to_clean = i;
874
875         cleaned_count = e1000_desc_unused(rx_ring);
876         if (cleaned_count)
877                 adapter->alloc_rx_buf(adapter, cleaned_count);
878
879         adapter->total_rx_bytes += total_rx_bytes;
880         adapter->total_rx_packets += total_rx_packets;
881         adapter->net_stats.rx_bytes += total_rx_bytes;
882         adapter->net_stats.rx_packets += total_rx_packets;
883         return cleaned;
884 }
885
886 /**
887  * e1000_consume_page - helper function
888  **/
889 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
890                                u16 length)
891 {
892         bi->page = NULL;
893         skb->len += length;
894         skb->data_len += length;
895         skb->truesize += length;
896 }
897
898 /**
899  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
900  * @adapter: board private structure
901  *
902  * the return value indicates whether actual cleaning was done, there
903  * is no guarantee that everything was cleaned
904  **/
905
906 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
907                                      int *work_done, int work_to_do)
908 {
909         struct net_device *netdev = adapter->netdev;
910         struct pci_dev *pdev = adapter->pdev;
911         struct e1000_ring *rx_ring = adapter->rx_ring;
912         struct e1000_rx_desc *rx_desc, *next_rxd;
913         struct e1000_buffer *buffer_info, *next_buffer;
914         u32 length;
915         unsigned int i;
916         int cleaned_count = 0;
917         bool cleaned = false;
918         unsigned int total_rx_bytes=0, total_rx_packets=0;
919
920         i = rx_ring->next_to_clean;
921         rx_desc = E1000_RX_DESC(*rx_ring, i);
922         buffer_info = &rx_ring->buffer_info[i];
923
924         while (rx_desc->status & E1000_RXD_STAT_DD) {
925                 struct sk_buff *skb;
926                 u8 status;
927
928                 if (*work_done >= work_to_do)
929                         break;
930                 (*work_done)++;
931
932                 status = rx_desc->status;
933                 skb = buffer_info->skb;
934                 buffer_info->skb = NULL;
935
936                 ++i;
937                 if (i == rx_ring->count)
938                         i = 0;
939                 next_rxd = E1000_RX_DESC(*rx_ring, i);
940                 prefetch(next_rxd);
941
942                 next_buffer = &rx_ring->buffer_info[i];
943
944                 cleaned = true;
945                 cleaned_count++;
946                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
947                                PCI_DMA_FROMDEVICE);
948                 buffer_info->dma = 0;
949
950                 length = le16_to_cpu(rx_desc->length);
951
952                 /* errors is only valid for DD + EOP descriptors */
953                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
954                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
955                                 /* recycle both page and skb */
956                                 buffer_info->skb = skb;
957                                 /* an error means any chain goes out the window
958                                  * too */
959                                 if (rx_ring->rx_skb_top)
960                                         dev_kfree_skb(rx_ring->rx_skb_top);
961                                 rx_ring->rx_skb_top = NULL;
962                                 goto next_desc;
963                 }
964
965 #define rxtop rx_ring->rx_skb_top
966                 if (!(status & E1000_RXD_STAT_EOP)) {
967                         /* this descriptor is only the beginning (or middle) */
968                         if (!rxtop) {
969                                 /* this is the beginning of a chain */
970                                 rxtop = skb;
971                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
972                                                    0, length);
973                         } else {
974                                 /* this is the middle of a chain */
975                                 skb_fill_page_desc(rxtop,
976                                     skb_shinfo(rxtop)->nr_frags,
977                                     buffer_info->page, 0, length);
978                                 /* re-use the skb, only consumed the page */
979                                 buffer_info->skb = skb;
980                         }
981                         e1000_consume_page(buffer_info, rxtop, length);
982                         goto next_desc;
983                 } else {
984                         if (rxtop) {
985                                 /* end of the chain */
986                                 skb_fill_page_desc(rxtop,
987                                     skb_shinfo(rxtop)->nr_frags,
988                                     buffer_info->page, 0, length);
989                                 /* re-use the current skb, we only consumed the
990                                  * page */
991                                 buffer_info->skb = skb;
992                                 skb = rxtop;
993                                 rxtop = NULL;
994                                 e1000_consume_page(buffer_info, skb, length);
995                         } else {
996                                 /* no chain, got EOP, this buf is the packet
997                                  * copybreak to save the put_page/alloc_page */
998                                 if (length <= copybreak &&
999                                     skb_tailroom(skb) >= length) {
1000                                         u8 *vaddr;
1001                                         vaddr = kmap_atomic(buffer_info->page,
1002                                                            KM_SKB_DATA_SOFTIRQ);
1003                                         memcpy(skb_tail_pointer(skb), vaddr,
1004                                                length);
1005                                         kunmap_atomic(vaddr,
1006                                                       KM_SKB_DATA_SOFTIRQ);
1007                                         /* re-use the page, so don't erase
1008                                          * buffer_info->page */
1009                                         skb_put(skb, length);
1010                                 } else {
1011                                         skb_fill_page_desc(skb, 0,
1012                                                            buffer_info->page, 0,
1013                                                            length);
1014                                         e1000_consume_page(buffer_info, skb,
1015                                                            length);
1016                                 }
1017                         }
1018                 }
1019
1020                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1021                 e1000_rx_checksum(adapter,
1022                                   (u32)(status) |
1023                                   ((u32)(rx_desc->errors) << 24),
1024                                   le16_to_cpu(rx_desc->csum), skb);
1025
1026                 /* probably a little skewed due to removing CRC */
1027                 total_rx_bytes += skb->len;
1028                 total_rx_packets++;
1029
1030                 /* eth type trans needs skb->data to point to something */
1031                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1032                         e_err("pskb_may_pull failed.\n");
1033                         dev_kfree_skb(skb);
1034                         goto next_desc;
1035                 }
1036
1037                 e1000_receive_skb(adapter, netdev, skb, status,
1038                                   rx_desc->special);
1039
1040 next_desc:
1041                 rx_desc->status = 0;
1042
1043                 /* return some buffers to hardware, one at a time is too slow */
1044                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1045                         adapter->alloc_rx_buf(adapter, cleaned_count);
1046                         cleaned_count = 0;
1047                 }
1048
1049                 /* use prefetched values */
1050                 rx_desc = next_rxd;
1051                 buffer_info = next_buffer;
1052         }
1053         rx_ring->next_to_clean = i;
1054
1055         cleaned_count = e1000_desc_unused(rx_ring);
1056         if (cleaned_count)
1057                 adapter->alloc_rx_buf(adapter, cleaned_count);
1058
1059         adapter->total_rx_bytes += total_rx_bytes;
1060         adapter->total_rx_packets += total_rx_packets;
1061         adapter->net_stats.rx_bytes += total_rx_bytes;
1062         adapter->net_stats.rx_packets += total_rx_packets;
1063         return cleaned;
1064 }
1065
1066 /**
1067  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1068  * @adapter: board private structure
1069  **/
1070 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1071 {
1072         struct e1000_ring *rx_ring = adapter->rx_ring;
1073         struct e1000_buffer *buffer_info;
1074         struct e1000_ps_page *ps_page;
1075         struct pci_dev *pdev = adapter->pdev;
1076         unsigned int i, j;
1077
1078         /* Free all the Rx ring sk_buffs */
1079         for (i = 0; i < rx_ring->count; i++) {
1080                 buffer_info = &rx_ring->buffer_info[i];
1081                 if (buffer_info->dma) {
1082                         if (adapter->clean_rx == e1000_clean_rx_irq)
1083                                 pci_unmap_single(pdev, buffer_info->dma,
1084                                                  adapter->rx_buffer_len,
1085                                                  PCI_DMA_FROMDEVICE);
1086                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1087                                 pci_unmap_page(pdev, buffer_info->dma,
1088                                                PAGE_SIZE,
1089                                                PCI_DMA_FROMDEVICE);
1090                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1091                                 pci_unmap_single(pdev, buffer_info->dma,
1092                                                  adapter->rx_ps_bsize0,
1093                                                  PCI_DMA_FROMDEVICE);
1094                         buffer_info->dma = 0;
1095                 }
1096
1097                 if (buffer_info->page) {
1098                         put_page(buffer_info->page);
1099                         buffer_info->page = NULL;
1100                 }
1101
1102                 if (buffer_info->skb) {
1103                         dev_kfree_skb(buffer_info->skb);
1104                         buffer_info->skb = NULL;
1105                 }
1106
1107                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1108                         ps_page = &buffer_info->ps_pages[j];
1109                         if (!ps_page->page)
1110                                 break;
1111                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1112                                        PCI_DMA_FROMDEVICE);
1113                         ps_page->dma = 0;
1114                         put_page(ps_page->page);
1115                         ps_page->page = NULL;
1116                 }
1117         }
1118
1119         /* there also may be some cached data from a chained receive */
1120         if (rx_ring->rx_skb_top) {
1121                 dev_kfree_skb(rx_ring->rx_skb_top);
1122                 rx_ring->rx_skb_top = NULL;
1123         }
1124
1125         /* Zero out the descriptor ring */
1126         memset(rx_ring->desc, 0, rx_ring->size);
1127
1128         rx_ring->next_to_clean = 0;
1129         rx_ring->next_to_use = 0;
1130
1131         writel(0, adapter->hw.hw_addr + rx_ring->head);
1132         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1133 }
1134
1135 static void e1000e_downshift_workaround(struct work_struct *work)
1136 {
1137         struct e1000_adapter *adapter = container_of(work,
1138                                         struct e1000_adapter, downshift_task);
1139
1140         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1141 }
1142
1143 /**
1144  * e1000_intr_msi - Interrupt Handler
1145  * @irq: interrupt number
1146  * @data: pointer to a network interface device structure
1147  **/
1148 static irqreturn_t e1000_intr_msi(int irq, void *data)
1149 {
1150         struct net_device *netdev = data;
1151         struct e1000_adapter *adapter = netdev_priv(netdev);
1152         struct e1000_hw *hw = &adapter->hw;
1153         u32 icr = er32(ICR);
1154
1155         /*
1156          * read ICR disables interrupts using IAM
1157          */
1158
1159         if (icr & E1000_ICR_LSC) {
1160                 hw->mac.get_link_status = 1;
1161                 /*
1162                  * ICH8 workaround-- Call gig speed drop workaround on cable
1163                  * disconnect (LSC) before accessing any PHY registers
1164                  */
1165                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1166                     (!(er32(STATUS) & E1000_STATUS_LU)))
1167                         schedule_work(&adapter->downshift_task);
1168
1169                 /*
1170                  * 80003ES2LAN workaround-- For packet buffer work-around on
1171                  * link down event; disable receives here in the ISR and reset
1172                  * adapter in watchdog
1173                  */
1174                 if (netif_carrier_ok(netdev) &&
1175                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1176                         /* disable receives */
1177                         u32 rctl = er32(RCTL);
1178                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1179                         adapter->flags |= FLAG_RX_RESTART_NOW;
1180                 }
1181                 /* guard against interrupt when we're going down */
1182                 if (!test_bit(__E1000_DOWN, &adapter->state))
1183                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1184         }
1185
1186         if (napi_schedule_prep(&adapter->napi)) {
1187                 adapter->total_tx_bytes = 0;
1188                 adapter->total_tx_packets = 0;
1189                 adapter->total_rx_bytes = 0;
1190                 adapter->total_rx_packets = 0;
1191                 __napi_schedule(&adapter->napi);
1192         }
1193
1194         return IRQ_HANDLED;
1195 }
1196
1197 /**
1198  * e1000_intr - Interrupt Handler
1199  * @irq: interrupt number
1200  * @data: pointer to a network interface device structure
1201  **/
1202 static irqreturn_t e1000_intr(int irq, void *data)
1203 {
1204         struct net_device *netdev = data;
1205         struct e1000_adapter *adapter = netdev_priv(netdev);
1206         struct e1000_hw *hw = &adapter->hw;
1207         u32 rctl, icr = er32(ICR);
1208
1209         if (!icr)
1210                 return IRQ_NONE;  /* Not our interrupt */
1211
1212         /*
1213          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1214          * not set, then the adapter didn't send an interrupt
1215          */
1216         if (!(icr & E1000_ICR_INT_ASSERTED))
1217                 return IRQ_NONE;
1218
1219         /*
1220          * Interrupt Auto-Mask...upon reading ICR,
1221          * interrupts are masked.  No need for the
1222          * IMC write
1223          */
1224
1225         if (icr & E1000_ICR_LSC) {
1226                 hw->mac.get_link_status = 1;
1227                 /*
1228                  * ICH8 workaround-- Call gig speed drop workaround on cable
1229                  * disconnect (LSC) before accessing any PHY registers
1230                  */
1231                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1232                     (!(er32(STATUS) & E1000_STATUS_LU)))
1233                         schedule_work(&adapter->downshift_task);
1234
1235                 /*
1236                  * 80003ES2LAN workaround--
1237                  * For packet buffer work-around on link down event;
1238                  * disable receives here in the ISR and
1239                  * reset adapter in watchdog
1240                  */
1241                 if (netif_carrier_ok(netdev) &&
1242                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1243                         /* disable receives */
1244                         rctl = er32(RCTL);
1245                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1246                         adapter->flags |= FLAG_RX_RESTART_NOW;
1247                 }
1248                 /* guard against interrupt when we're going down */
1249                 if (!test_bit(__E1000_DOWN, &adapter->state))
1250                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1251         }
1252
1253         if (napi_schedule_prep(&adapter->napi)) {
1254                 adapter->total_tx_bytes = 0;
1255                 adapter->total_tx_packets = 0;
1256                 adapter->total_rx_bytes = 0;
1257                 adapter->total_rx_packets = 0;
1258                 __napi_schedule(&adapter->napi);
1259         }
1260
1261         return IRQ_HANDLED;
1262 }
1263
1264 static irqreturn_t e1000_msix_other(int irq, void *data)
1265 {
1266         struct net_device *netdev = data;
1267         struct e1000_adapter *adapter = netdev_priv(netdev);
1268         struct e1000_hw *hw = &adapter->hw;
1269         u32 icr = er32(ICR);
1270
1271         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1272                 ew32(IMS, E1000_IMS_OTHER);
1273                 return IRQ_NONE;
1274         }
1275
1276         if (icr & adapter->eiac_mask)
1277                 ew32(ICS, (icr & adapter->eiac_mask));
1278
1279         if (icr & E1000_ICR_OTHER) {
1280                 if (!(icr & E1000_ICR_LSC))
1281                         goto no_link_interrupt;
1282                 hw->mac.get_link_status = 1;
1283                 /* guard against interrupt when we're going down */
1284                 if (!test_bit(__E1000_DOWN, &adapter->state))
1285                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1286         }
1287
1288 no_link_interrupt:
1289         ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1290
1291         return IRQ_HANDLED;
1292 }
1293
1294
1295 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1296 {
1297         struct net_device *netdev = data;
1298         struct e1000_adapter *adapter = netdev_priv(netdev);
1299         struct e1000_hw *hw = &adapter->hw;
1300         struct e1000_ring *tx_ring = adapter->tx_ring;
1301
1302
1303         adapter->total_tx_bytes = 0;
1304         adapter->total_tx_packets = 0;
1305
1306         if (!e1000_clean_tx_irq(adapter))
1307                 /* Ring was not completely cleaned, so fire another interrupt */
1308                 ew32(ICS, tx_ring->ims_val);
1309
1310         return IRQ_HANDLED;
1311 }
1312
1313 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1314 {
1315         struct net_device *netdev = data;
1316         struct e1000_adapter *adapter = netdev_priv(netdev);
1317
1318         /* Write the ITR value calculated at the end of the
1319          * previous interrupt.
1320          */
1321         if (adapter->rx_ring->set_itr) {
1322                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1323                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1324                 adapter->rx_ring->set_itr = 0;
1325         }
1326
1327         if (napi_schedule_prep(&adapter->napi)) {
1328                 adapter->total_rx_bytes = 0;
1329                 adapter->total_rx_packets = 0;
1330                 __napi_schedule(&adapter->napi);
1331         }
1332         return IRQ_HANDLED;
1333 }
1334
1335 /**
1336  * e1000_configure_msix - Configure MSI-X hardware
1337  *
1338  * e1000_configure_msix sets up the hardware to properly
1339  * generate MSI-X interrupts.
1340  **/
1341 static void e1000_configure_msix(struct e1000_adapter *adapter)
1342 {
1343         struct e1000_hw *hw = &adapter->hw;
1344         struct e1000_ring *rx_ring = adapter->rx_ring;
1345         struct e1000_ring *tx_ring = adapter->tx_ring;
1346         int vector = 0;
1347         u32 ctrl_ext, ivar = 0;
1348
1349         adapter->eiac_mask = 0;
1350
1351         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1352         if (hw->mac.type == e1000_82574) {
1353                 u32 rfctl = er32(RFCTL);
1354                 rfctl |= E1000_RFCTL_ACK_DIS;
1355                 ew32(RFCTL, rfctl);
1356         }
1357
1358 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1359         /* Configure Rx vector */
1360         rx_ring->ims_val = E1000_IMS_RXQ0;
1361         adapter->eiac_mask |= rx_ring->ims_val;
1362         if (rx_ring->itr_val)
1363                 writel(1000000000 / (rx_ring->itr_val * 256),
1364                        hw->hw_addr + rx_ring->itr_register);
1365         else
1366                 writel(1, hw->hw_addr + rx_ring->itr_register);
1367         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1368
1369         /* Configure Tx vector */
1370         tx_ring->ims_val = E1000_IMS_TXQ0;
1371         vector++;
1372         if (tx_ring->itr_val)
1373                 writel(1000000000 / (tx_ring->itr_val * 256),
1374                        hw->hw_addr + tx_ring->itr_register);
1375         else
1376                 writel(1, hw->hw_addr + tx_ring->itr_register);
1377         adapter->eiac_mask |= tx_ring->ims_val;
1378         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1379
1380         /* set vector for Other Causes, e.g. link changes */
1381         vector++;
1382         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1383         if (rx_ring->itr_val)
1384                 writel(1000000000 / (rx_ring->itr_val * 256),
1385                        hw->hw_addr + E1000_EITR_82574(vector));
1386         else
1387                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1388
1389         /* Cause Tx interrupts on every write back */
1390         ivar |= (1 << 31);
1391
1392         ew32(IVAR, ivar);
1393
1394         /* enable MSI-X PBA support */
1395         ctrl_ext = er32(CTRL_EXT);
1396         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1397
1398         /* Auto-Mask Other interrupts upon ICR read */
1399 #define E1000_EIAC_MASK_82574   0x01F00000
1400         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1401         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1402         ew32(CTRL_EXT, ctrl_ext);
1403         e1e_flush();
1404 }
1405
1406 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1407 {
1408         if (adapter->msix_entries) {
1409                 pci_disable_msix(adapter->pdev);
1410                 kfree(adapter->msix_entries);
1411                 adapter->msix_entries = NULL;
1412         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1413                 pci_disable_msi(adapter->pdev);
1414                 adapter->flags &= ~FLAG_MSI_ENABLED;
1415         }
1416
1417         return;
1418 }
1419
1420 /**
1421  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1422  *
1423  * Attempt to configure interrupts using the best available
1424  * capabilities of the hardware and kernel.
1425  **/
1426 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1427 {
1428         int err;
1429         int numvecs, i;
1430
1431
1432         switch (adapter->int_mode) {
1433         case E1000E_INT_MODE_MSIX:
1434                 if (adapter->flags & FLAG_HAS_MSIX) {
1435                         numvecs = 3; /* RxQ0, TxQ0 and other */
1436                         adapter->msix_entries = kcalloc(numvecs,
1437                                                       sizeof(struct msix_entry),
1438                                                       GFP_KERNEL);
1439                         if (adapter->msix_entries) {
1440                                 for (i = 0; i < numvecs; i++)
1441                                         adapter->msix_entries[i].entry = i;
1442
1443                                 err = pci_enable_msix(adapter->pdev,
1444                                                       adapter->msix_entries,
1445                                                       numvecs);
1446                                 if (err == 0)
1447                                         return;
1448                         }
1449                         /* MSI-X failed, so fall through and try MSI */
1450                         e_err("Failed to initialize MSI-X interrupts.  "
1451                               "Falling back to MSI interrupts.\n");
1452                         e1000e_reset_interrupt_capability(adapter);
1453                 }
1454                 adapter->int_mode = E1000E_INT_MODE_MSI;
1455                 /* Fall through */
1456         case E1000E_INT_MODE_MSI:
1457                 if (!pci_enable_msi(adapter->pdev)) {
1458                         adapter->flags |= FLAG_MSI_ENABLED;
1459                 } else {
1460                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1461                         e_err("Failed to initialize MSI interrupts.  Falling "
1462                               "back to legacy interrupts.\n");
1463                 }
1464                 /* Fall through */
1465         case E1000E_INT_MODE_LEGACY:
1466                 /* Don't do anything; this is the system default */
1467                 break;
1468         }
1469
1470         return;
1471 }
1472
1473 /**
1474  * e1000_request_msix - Initialize MSI-X interrupts
1475  *
1476  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1477  * kernel.
1478  **/
1479 static int e1000_request_msix(struct e1000_adapter *adapter)
1480 {
1481         struct net_device *netdev = adapter->netdev;
1482         int err = 0, vector = 0;
1483
1484         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1485                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1486         else
1487                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1488         err = request_irq(adapter->msix_entries[vector].vector,
1489                           &e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1490                           netdev);
1491         if (err)
1492                 goto out;
1493         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1494         adapter->rx_ring->itr_val = adapter->itr;
1495         vector++;
1496
1497         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1498                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1499         else
1500                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1501         err = request_irq(adapter->msix_entries[vector].vector,
1502                           &e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1503                           netdev);
1504         if (err)
1505                 goto out;
1506         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1507         adapter->tx_ring->itr_val = adapter->itr;
1508         vector++;
1509
1510         err = request_irq(adapter->msix_entries[vector].vector,
1511                           &e1000_msix_other, 0, netdev->name, netdev);
1512         if (err)
1513                 goto out;
1514
1515         e1000_configure_msix(adapter);
1516         return 0;
1517 out:
1518         return err;
1519 }
1520
1521 /**
1522  * e1000_request_irq - initialize interrupts
1523  *
1524  * Attempts to configure interrupts using the best available
1525  * capabilities of the hardware and kernel.
1526  **/
1527 static int e1000_request_irq(struct e1000_adapter *adapter)
1528 {
1529         struct net_device *netdev = adapter->netdev;
1530         int err;
1531
1532         if (adapter->msix_entries) {
1533                 err = e1000_request_msix(adapter);
1534                 if (!err)
1535                         return err;
1536                 /* fall back to MSI */
1537                 e1000e_reset_interrupt_capability(adapter);
1538                 adapter->int_mode = E1000E_INT_MODE_MSI;
1539                 e1000e_set_interrupt_capability(adapter);
1540         }
1541         if (adapter->flags & FLAG_MSI_ENABLED) {
1542                 err = request_irq(adapter->pdev->irq, &e1000_intr_msi, 0,
1543                                   netdev->name, netdev);
1544                 if (!err)
1545                         return err;
1546
1547                 /* fall back to legacy interrupt */
1548                 e1000e_reset_interrupt_capability(adapter);
1549                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1550         }
1551
1552         err = request_irq(adapter->pdev->irq, &e1000_intr, IRQF_SHARED,
1553                           netdev->name, netdev);
1554         if (err)
1555                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1556
1557         return err;
1558 }
1559
1560 static void e1000_free_irq(struct e1000_adapter *adapter)
1561 {
1562         struct net_device *netdev = adapter->netdev;
1563
1564         if (adapter->msix_entries) {
1565                 int vector = 0;
1566
1567                 free_irq(adapter->msix_entries[vector].vector, netdev);
1568                 vector++;
1569
1570                 free_irq(adapter->msix_entries[vector].vector, netdev);
1571                 vector++;
1572
1573                 /* Other Causes interrupt vector */
1574                 free_irq(adapter->msix_entries[vector].vector, netdev);
1575                 return;
1576         }
1577
1578         free_irq(adapter->pdev->irq, netdev);
1579 }
1580
1581 /**
1582  * e1000_irq_disable - Mask off interrupt generation on the NIC
1583  **/
1584 static void e1000_irq_disable(struct e1000_adapter *adapter)
1585 {
1586         struct e1000_hw *hw = &adapter->hw;
1587
1588         ew32(IMC, ~0);
1589         if (adapter->msix_entries)
1590                 ew32(EIAC_82574, 0);
1591         e1e_flush();
1592         synchronize_irq(adapter->pdev->irq);
1593 }
1594
1595 /**
1596  * e1000_irq_enable - Enable default interrupt generation settings
1597  **/
1598 static void e1000_irq_enable(struct e1000_adapter *adapter)
1599 {
1600         struct e1000_hw *hw = &adapter->hw;
1601
1602         if (adapter->msix_entries) {
1603                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1604                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1605         } else {
1606                 ew32(IMS, IMS_ENABLE_MASK);
1607         }
1608         e1e_flush();
1609 }
1610
1611 /**
1612  * e1000_get_hw_control - get control of the h/w from f/w
1613  * @adapter: address of board private structure
1614  *
1615  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1616  * For ASF and Pass Through versions of f/w this means that
1617  * the driver is loaded. For AMT version (only with 82573)
1618  * of the f/w this means that the network i/f is open.
1619  **/
1620 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1621 {
1622         struct e1000_hw *hw = &adapter->hw;
1623         u32 ctrl_ext;
1624         u32 swsm;
1625
1626         /* Let firmware know the driver has taken over */
1627         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1628                 swsm = er32(SWSM);
1629                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1630         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1631                 ctrl_ext = er32(CTRL_EXT);
1632                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1633         }
1634 }
1635
1636 /**
1637  * e1000_release_hw_control - release control of the h/w to f/w
1638  * @adapter: address of board private structure
1639  *
1640  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1641  * For ASF and Pass Through versions of f/w this means that the
1642  * driver is no longer loaded. For AMT version (only with 82573) i
1643  * of the f/w this means that the network i/f is closed.
1644  *
1645  **/
1646 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1647 {
1648         struct e1000_hw *hw = &adapter->hw;
1649         u32 ctrl_ext;
1650         u32 swsm;
1651
1652         /* Let firmware taken over control of h/w */
1653         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1654                 swsm = er32(SWSM);
1655                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1656         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1657                 ctrl_ext = er32(CTRL_EXT);
1658                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1659         }
1660 }
1661
1662 /**
1663  * @e1000_alloc_ring - allocate memory for a ring structure
1664  **/
1665 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1666                                 struct e1000_ring *ring)
1667 {
1668         struct pci_dev *pdev = adapter->pdev;
1669
1670         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1671                                         GFP_KERNEL);
1672         if (!ring->desc)
1673                 return -ENOMEM;
1674
1675         return 0;
1676 }
1677
1678 /**
1679  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1680  * @adapter: board private structure
1681  *
1682  * Return 0 on success, negative on failure
1683  **/
1684 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1685 {
1686         struct e1000_ring *tx_ring = adapter->tx_ring;
1687         int err = -ENOMEM, size;
1688
1689         size = sizeof(struct e1000_buffer) * tx_ring->count;
1690         tx_ring->buffer_info = vmalloc(size);
1691         if (!tx_ring->buffer_info)
1692                 goto err;
1693         memset(tx_ring->buffer_info, 0, size);
1694
1695         /* round up to nearest 4K */
1696         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1697         tx_ring->size = ALIGN(tx_ring->size, 4096);
1698
1699         err = e1000_alloc_ring_dma(adapter, tx_ring);
1700         if (err)
1701                 goto err;
1702
1703         tx_ring->next_to_use = 0;
1704         tx_ring->next_to_clean = 0;
1705
1706         return 0;
1707 err:
1708         vfree(tx_ring->buffer_info);
1709         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1710         return err;
1711 }
1712
1713 /**
1714  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1715  * @adapter: board private structure
1716  *
1717  * Returns 0 on success, negative on failure
1718  **/
1719 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1720 {
1721         struct e1000_ring *rx_ring = adapter->rx_ring;
1722         struct e1000_buffer *buffer_info;
1723         int i, size, desc_len, err = -ENOMEM;
1724
1725         size = sizeof(struct e1000_buffer) * rx_ring->count;
1726         rx_ring->buffer_info = vmalloc(size);
1727         if (!rx_ring->buffer_info)
1728                 goto err;
1729         memset(rx_ring->buffer_info, 0, size);
1730
1731         for (i = 0; i < rx_ring->count; i++) {
1732                 buffer_info = &rx_ring->buffer_info[i];
1733                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1734                                                 sizeof(struct e1000_ps_page),
1735                                                 GFP_KERNEL);
1736                 if (!buffer_info->ps_pages)
1737                         goto err_pages;
1738         }
1739
1740         desc_len = sizeof(union e1000_rx_desc_packet_split);
1741
1742         /* Round up to nearest 4K */
1743         rx_ring->size = rx_ring->count * desc_len;
1744         rx_ring->size = ALIGN(rx_ring->size, 4096);
1745
1746         err = e1000_alloc_ring_dma(adapter, rx_ring);
1747         if (err)
1748                 goto err_pages;
1749
1750         rx_ring->next_to_clean = 0;
1751         rx_ring->next_to_use = 0;
1752         rx_ring->rx_skb_top = NULL;
1753
1754         return 0;
1755
1756 err_pages:
1757         for (i = 0; i < rx_ring->count; i++) {
1758                 buffer_info = &rx_ring->buffer_info[i];
1759                 kfree(buffer_info->ps_pages);
1760         }
1761 err:
1762         vfree(rx_ring->buffer_info);
1763         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1764         return err;
1765 }
1766
1767 /**
1768  * e1000_clean_tx_ring - Free Tx Buffers
1769  * @adapter: board private structure
1770  **/
1771 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1772 {
1773         struct e1000_ring *tx_ring = adapter->tx_ring;
1774         struct e1000_buffer *buffer_info;
1775         unsigned long size;
1776         unsigned int i;
1777
1778         for (i = 0; i < tx_ring->count; i++) {
1779                 buffer_info = &tx_ring->buffer_info[i];
1780                 e1000_put_txbuf(adapter, buffer_info);
1781         }
1782
1783         size = sizeof(struct e1000_buffer) * tx_ring->count;
1784         memset(tx_ring->buffer_info, 0, size);
1785
1786         memset(tx_ring->desc, 0, tx_ring->size);
1787
1788         tx_ring->next_to_use = 0;
1789         tx_ring->next_to_clean = 0;
1790
1791         writel(0, adapter->hw.hw_addr + tx_ring->head);
1792         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1793 }
1794
1795 /**
1796  * e1000e_free_tx_resources - Free Tx Resources per Queue
1797  * @adapter: board private structure
1798  *
1799  * Free all transmit software resources
1800  **/
1801 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1802 {
1803         struct pci_dev *pdev = adapter->pdev;
1804         struct e1000_ring *tx_ring = adapter->tx_ring;
1805
1806         e1000_clean_tx_ring(adapter);
1807
1808         vfree(tx_ring->buffer_info);
1809         tx_ring->buffer_info = NULL;
1810
1811         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1812                           tx_ring->dma);
1813         tx_ring->desc = NULL;
1814 }
1815
1816 /**
1817  * e1000e_free_rx_resources - Free Rx Resources
1818  * @adapter: board private structure
1819  *
1820  * Free all receive software resources
1821  **/
1822
1823 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1824 {
1825         struct pci_dev *pdev = adapter->pdev;
1826         struct e1000_ring *rx_ring = adapter->rx_ring;
1827         int i;
1828
1829         e1000_clean_rx_ring(adapter);
1830
1831         for (i = 0; i < rx_ring->count; i++) {
1832                 kfree(rx_ring->buffer_info[i].ps_pages);
1833         }
1834
1835         vfree(rx_ring->buffer_info);
1836         rx_ring->buffer_info = NULL;
1837
1838         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1839                           rx_ring->dma);
1840         rx_ring->desc = NULL;
1841 }
1842
1843 /**
1844  * e1000_update_itr - update the dynamic ITR value based on statistics
1845  * @adapter: pointer to adapter
1846  * @itr_setting: current adapter->itr
1847  * @packets: the number of packets during this measurement interval
1848  * @bytes: the number of bytes during this measurement interval
1849  *
1850  *      Stores a new ITR value based on packets and byte
1851  *      counts during the last interrupt.  The advantage of per interrupt
1852  *      computation is faster updates and more accurate ITR for the current
1853  *      traffic pattern.  Constants in this function were computed
1854  *      based on theoretical maximum wire speed and thresholds were set based
1855  *      on testing data as well as attempting to minimize response time
1856  *      while increasing bulk throughput.  This functionality is controlled
1857  *      by the InterruptThrottleRate module parameter.
1858  **/
1859 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1860                                      u16 itr_setting, int packets,
1861                                      int bytes)
1862 {
1863         unsigned int retval = itr_setting;
1864
1865         if (packets == 0)
1866                 goto update_itr_done;
1867
1868         switch (itr_setting) {
1869         case lowest_latency:
1870                 /* handle TSO and jumbo frames */
1871                 if (bytes/packets > 8000)
1872                         retval = bulk_latency;
1873                 else if ((packets < 5) && (bytes > 512)) {
1874                         retval = low_latency;
1875                 }
1876                 break;
1877         case low_latency:  /* 50 usec aka 20000 ints/s */
1878                 if (bytes > 10000) {
1879                         /* this if handles the TSO accounting */
1880                         if (bytes/packets > 8000) {
1881                                 retval = bulk_latency;
1882                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1883                                 retval = bulk_latency;
1884                         } else if ((packets > 35)) {
1885                                 retval = lowest_latency;
1886                         }
1887                 } else if (bytes/packets > 2000) {
1888                         retval = bulk_latency;
1889                 } else if (packets <= 2 && bytes < 512) {
1890                         retval = lowest_latency;
1891                 }
1892                 break;
1893         case bulk_latency: /* 250 usec aka 4000 ints/s */
1894                 if (bytes > 25000) {
1895                         if (packets > 35) {
1896                                 retval = low_latency;
1897                         }
1898                 } else if (bytes < 6000) {
1899                         retval = low_latency;
1900                 }
1901                 break;
1902         }
1903
1904 update_itr_done:
1905         return retval;
1906 }
1907
1908 static void e1000_set_itr(struct e1000_adapter *adapter)
1909 {
1910         struct e1000_hw *hw = &adapter->hw;
1911         u16 current_itr;
1912         u32 new_itr = adapter->itr;
1913
1914         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1915         if (adapter->link_speed != SPEED_1000) {
1916                 current_itr = 0;
1917                 new_itr = 4000;
1918                 goto set_itr_now;
1919         }
1920
1921         adapter->tx_itr = e1000_update_itr(adapter,
1922                                     adapter->tx_itr,
1923                                     adapter->total_tx_packets,
1924                                     adapter->total_tx_bytes);
1925         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1926         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1927                 adapter->tx_itr = low_latency;
1928
1929         adapter->rx_itr = e1000_update_itr(adapter,
1930                                     adapter->rx_itr,
1931                                     adapter->total_rx_packets,
1932                                     adapter->total_rx_bytes);
1933         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1934         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1935                 adapter->rx_itr = low_latency;
1936
1937         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1938
1939         switch (current_itr) {
1940         /* counts and packets in update_itr are dependent on these numbers */
1941         case lowest_latency:
1942                 new_itr = 70000;
1943                 break;
1944         case low_latency:
1945                 new_itr = 20000; /* aka hwitr = ~200 */
1946                 break;
1947         case bulk_latency:
1948                 new_itr = 4000;
1949                 break;
1950         default:
1951                 break;
1952         }
1953
1954 set_itr_now:
1955         if (new_itr != adapter->itr) {
1956                 /*
1957                  * this attempts to bias the interrupt rate towards Bulk
1958                  * by adding intermediate steps when interrupt rate is
1959                  * increasing
1960                  */
1961                 new_itr = new_itr > adapter->itr ?
1962                              min(adapter->itr + (new_itr >> 2), new_itr) :
1963                              new_itr;
1964                 adapter->itr = new_itr;
1965                 adapter->rx_ring->itr_val = new_itr;
1966                 if (adapter->msix_entries)
1967                         adapter->rx_ring->set_itr = 1;
1968                 else
1969                         ew32(ITR, 1000000000 / (new_itr * 256));
1970         }
1971 }
1972
1973 /**
1974  * e1000_alloc_queues - Allocate memory for all rings
1975  * @adapter: board private structure to initialize
1976  **/
1977 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1978 {
1979         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1980         if (!adapter->tx_ring)
1981                 goto err;
1982
1983         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1984         if (!adapter->rx_ring)
1985                 goto err;
1986
1987         return 0;
1988 err:
1989         e_err("Unable to allocate memory for queues\n");
1990         kfree(adapter->rx_ring);
1991         kfree(adapter->tx_ring);
1992         return -ENOMEM;
1993 }
1994
1995 /**
1996  * e1000_clean - NAPI Rx polling callback
1997  * @napi: struct associated with this polling callback
1998  * @budget: amount of packets driver is allowed to process this poll
1999  **/
2000 static int e1000_clean(struct napi_struct *napi, int budget)
2001 {
2002         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2003         struct e1000_hw *hw = &adapter->hw;
2004         struct net_device *poll_dev = adapter->netdev;
2005         int tx_cleaned = 0, work_done = 0;
2006
2007         adapter = netdev_priv(poll_dev);
2008
2009         if (adapter->msix_entries &&
2010             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2011                 goto clean_rx;
2012
2013         tx_cleaned = e1000_clean_tx_irq(adapter);
2014
2015 clean_rx:
2016         adapter->clean_rx(adapter, &work_done, budget);
2017
2018         if (tx_cleaned)
2019                 work_done = budget;
2020
2021         /* If budget not fully consumed, exit the polling mode */
2022         if (work_done < budget) {
2023                 if (adapter->itr_setting & 3)
2024                         e1000_set_itr(adapter);
2025                 napi_complete(napi);
2026                 if (adapter->msix_entries)
2027                         ew32(IMS, adapter->rx_ring->ims_val);
2028                 else
2029                         e1000_irq_enable(adapter);
2030         }
2031
2032         return work_done;
2033 }
2034
2035 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2036 {
2037         struct e1000_adapter *adapter = netdev_priv(netdev);
2038         struct e1000_hw *hw = &adapter->hw;
2039         u32 vfta, index;
2040
2041         /* don't update vlan cookie if already programmed */
2042         if ((adapter->hw.mng_cookie.status &
2043              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2044             (vid == adapter->mng_vlan_id))
2045                 return;
2046         /* add VID to filter table */
2047         index = (vid >> 5) & 0x7F;
2048         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2049         vfta |= (1 << (vid & 0x1F));
2050         e1000e_write_vfta(hw, index, vfta);
2051 }
2052
2053 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2054 {
2055         struct e1000_adapter *adapter = netdev_priv(netdev);
2056         struct e1000_hw *hw = &adapter->hw;
2057         u32 vfta, index;
2058
2059         if (!test_bit(__E1000_DOWN, &adapter->state))
2060                 e1000_irq_disable(adapter);
2061         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2062
2063         if (!test_bit(__E1000_DOWN, &adapter->state))
2064                 e1000_irq_enable(adapter);
2065
2066         if ((adapter->hw.mng_cookie.status &
2067              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2068             (vid == adapter->mng_vlan_id)) {
2069                 /* release control to f/w */
2070                 e1000_release_hw_control(adapter);
2071                 return;
2072         }
2073
2074         /* remove VID from filter table */
2075         index = (vid >> 5) & 0x7F;
2076         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2077         vfta &= ~(1 << (vid & 0x1F));
2078         e1000e_write_vfta(hw, index, vfta);
2079 }
2080
2081 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2082 {
2083         struct net_device *netdev = adapter->netdev;
2084         u16 vid = adapter->hw.mng_cookie.vlan_id;
2085         u16 old_vid = adapter->mng_vlan_id;
2086
2087         if (!adapter->vlgrp)
2088                 return;
2089
2090         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2091                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2092                 if (adapter->hw.mng_cookie.status &
2093                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2094                         e1000_vlan_rx_add_vid(netdev, vid);
2095                         adapter->mng_vlan_id = vid;
2096                 }
2097
2098                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2099                                 (vid != old_vid) &&
2100                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2101                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2102         } else {
2103                 adapter->mng_vlan_id = vid;
2104         }
2105 }
2106
2107
2108 static void e1000_vlan_rx_register(struct net_device *netdev,
2109                                    struct vlan_group *grp)
2110 {
2111         struct e1000_adapter *adapter = netdev_priv(netdev);
2112         struct e1000_hw *hw = &adapter->hw;
2113         u32 ctrl, rctl;
2114
2115         if (!test_bit(__E1000_DOWN, &adapter->state))
2116                 e1000_irq_disable(adapter);
2117         adapter->vlgrp = grp;
2118
2119         if (grp) {
2120                 /* enable VLAN tag insert/strip */
2121                 ctrl = er32(CTRL);
2122                 ctrl |= E1000_CTRL_VME;
2123                 ew32(CTRL, ctrl);
2124
2125                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2126                         /* enable VLAN receive filtering */
2127                         rctl = er32(RCTL);
2128                         rctl &= ~E1000_RCTL_CFIEN;
2129                         ew32(RCTL, rctl);
2130                         e1000_update_mng_vlan(adapter);
2131                 }
2132         } else {
2133                 /* disable VLAN tag insert/strip */
2134                 ctrl = er32(CTRL);
2135                 ctrl &= ~E1000_CTRL_VME;
2136                 ew32(CTRL, ctrl);
2137
2138                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2139                         if (adapter->mng_vlan_id !=
2140                             (u16)E1000_MNG_VLAN_NONE) {
2141                                 e1000_vlan_rx_kill_vid(netdev,
2142                                                        adapter->mng_vlan_id);
2143                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2144                         }
2145                 }
2146         }
2147
2148         if (!test_bit(__E1000_DOWN, &adapter->state))
2149                 e1000_irq_enable(adapter);
2150 }
2151
2152 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2153 {
2154         u16 vid;
2155
2156         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2157
2158         if (!adapter->vlgrp)
2159                 return;
2160
2161         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2162                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2163                         continue;
2164                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2165         }
2166 }
2167
2168 static void e1000_init_manageability(struct e1000_adapter *adapter)
2169 {
2170         struct e1000_hw *hw = &adapter->hw;
2171         u32 manc, manc2h;
2172
2173         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2174                 return;
2175
2176         manc = er32(MANC);
2177
2178         /*
2179          * enable receiving management packets to the host. this will probably
2180          * generate destination unreachable messages from the host OS, but
2181          * the packets will be handled on SMBUS
2182          */
2183         manc |= E1000_MANC_EN_MNG2HOST;
2184         manc2h = er32(MANC2H);
2185 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2186 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2187         manc2h |= E1000_MNG2HOST_PORT_623;
2188         manc2h |= E1000_MNG2HOST_PORT_664;
2189         ew32(MANC2H, manc2h);
2190         ew32(MANC, manc);
2191 }
2192
2193 /**
2194  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2195  * @adapter: board private structure
2196  *
2197  * Configure the Tx unit of the MAC after a reset.
2198  **/
2199 static void e1000_configure_tx(struct e1000_adapter *adapter)
2200 {
2201         struct e1000_hw *hw = &adapter->hw;
2202         struct e1000_ring *tx_ring = adapter->tx_ring;
2203         u64 tdba;
2204         u32 tdlen, tctl, tipg, tarc;
2205         u32 ipgr1, ipgr2;
2206
2207         /* Setup the HW Tx Head and Tail descriptor pointers */
2208         tdba = tx_ring->dma;
2209         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2210         ew32(TDBAL, (tdba & DMA_32BIT_MASK));
2211         ew32(TDBAH, (tdba >> 32));
2212         ew32(TDLEN, tdlen);
2213         ew32(TDH, 0);
2214         ew32(TDT, 0);
2215         tx_ring->head = E1000_TDH;
2216         tx_ring->tail = E1000_TDT;
2217
2218         /* Set the default values for the Tx Inter Packet Gap timer */
2219         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2220         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2221         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2222
2223         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2224                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2225
2226         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2227         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2228         ew32(TIPG, tipg);
2229
2230         /* Set the Tx Interrupt Delay register */
2231         ew32(TIDV, adapter->tx_int_delay);
2232         /* Tx irq moderation */
2233         ew32(TADV, adapter->tx_abs_int_delay);
2234
2235         /* Program the Transmit Control Register */
2236         tctl = er32(TCTL);
2237         tctl &= ~E1000_TCTL_CT;
2238         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2239                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2240
2241         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2242                 tarc = er32(TARC(0));
2243                 /*
2244                  * set the speed mode bit, we'll clear it if we're not at
2245                  * gigabit link later
2246                  */
2247 #define SPEED_MODE_BIT (1 << 21)
2248                 tarc |= SPEED_MODE_BIT;
2249                 ew32(TARC(0), tarc);
2250         }
2251
2252         /* errata: program both queues to unweighted RR */
2253         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2254                 tarc = er32(TARC(0));
2255                 tarc |= 1;
2256                 ew32(TARC(0), tarc);
2257                 tarc = er32(TARC(1));
2258                 tarc |= 1;
2259                 ew32(TARC(1), tarc);
2260         }
2261
2262         e1000e_config_collision_dist(hw);
2263
2264         /* Setup Transmit Descriptor Settings for eop descriptor */
2265         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2266
2267         /* only set IDE if we are delaying interrupts using the timers */
2268         if (adapter->tx_int_delay)
2269                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2270
2271         /* enable Report Status bit */
2272         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2273
2274         ew32(TCTL, tctl);
2275
2276         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
2277 }
2278
2279 /**
2280  * e1000_setup_rctl - configure the receive control registers
2281  * @adapter: Board private structure
2282  **/
2283 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2284                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2285 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2286 {
2287         struct e1000_hw *hw = &adapter->hw;
2288         u32 rctl, rfctl;
2289         u32 psrctl = 0;
2290         u32 pages = 0;
2291
2292         /* Program MC offset vector base */
2293         rctl = er32(RCTL);
2294         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2295         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2296                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2297                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2298
2299         /* Do not Store bad packets */
2300         rctl &= ~E1000_RCTL_SBP;
2301
2302         /* Enable Long Packet receive */
2303         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2304                 rctl &= ~E1000_RCTL_LPE;
2305         else
2306                 rctl |= E1000_RCTL_LPE;
2307
2308         /* Some systems expect that the CRC is included in SMBUS traffic. The
2309          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2310          * host memory when this is enabled
2311          */
2312         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2313                 rctl |= E1000_RCTL_SECRC;
2314
2315         /* Setup buffer sizes */
2316         rctl &= ~E1000_RCTL_SZ_4096;
2317         rctl |= E1000_RCTL_BSEX;
2318         switch (adapter->rx_buffer_len) {
2319         case 256:
2320                 rctl |= E1000_RCTL_SZ_256;
2321                 rctl &= ~E1000_RCTL_BSEX;
2322                 break;
2323         case 512:
2324                 rctl |= E1000_RCTL_SZ_512;
2325                 rctl &= ~E1000_RCTL_BSEX;
2326                 break;
2327         case 1024:
2328                 rctl |= E1000_RCTL_SZ_1024;
2329                 rctl &= ~E1000_RCTL_BSEX;
2330                 break;
2331         case 2048:
2332         default:
2333                 rctl |= E1000_RCTL_SZ_2048;
2334                 rctl &= ~E1000_RCTL_BSEX;
2335                 break;
2336         case 4096:
2337                 rctl |= E1000_RCTL_SZ_4096;
2338                 break;
2339         case 8192:
2340                 rctl |= E1000_RCTL_SZ_8192;
2341                 break;
2342         case 16384:
2343                 rctl |= E1000_RCTL_SZ_16384;
2344                 break;
2345         }
2346
2347         /*
2348          * 82571 and greater support packet-split where the protocol
2349          * header is placed in skb->data and the packet data is
2350          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2351          * In the case of a non-split, skb->data is linearly filled,
2352          * followed by the page buffers.  Therefore, skb->data is
2353          * sized to hold the largest protocol header.
2354          *
2355          * allocations using alloc_page take too long for regular MTU
2356          * so only enable packet split for jumbo frames
2357          *
2358          * Using pages when the page size is greater than 16k wastes
2359          * a lot of memory, since we allocate 3 pages at all times
2360          * per packet.
2361          */
2362         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2363         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2364             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2365                 adapter->rx_ps_pages = pages;
2366         else
2367                 adapter->rx_ps_pages = 0;
2368
2369         if (adapter->rx_ps_pages) {
2370                 /* Configure extra packet-split registers */
2371                 rfctl = er32(RFCTL);
2372                 rfctl |= E1000_RFCTL_EXTEN;
2373                 /*
2374                  * disable packet split support for IPv6 extension headers,
2375                  * because some malformed IPv6 headers can hang the Rx
2376                  */
2377                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2378                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2379
2380                 ew32(RFCTL, rfctl);
2381
2382                 /* Enable Packet split descriptors */
2383                 rctl |= E1000_RCTL_DTYP_PS;
2384
2385                 psrctl |= adapter->rx_ps_bsize0 >>
2386                         E1000_PSRCTL_BSIZE0_SHIFT;
2387
2388                 switch (adapter->rx_ps_pages) {
2389                 case 3:
2390                         psrctl |= PAGE_SIZE <<
2391                                 E1000_PSRCTL_BSIZE3_SHIFT;
2392                 case 2:
2393                         psrctl |= PAGE_SIZE <<
2394                                 E1000_PSRCTL_BSIZE2_SHIFT;
2395                 case 1:
2396                         psrctl |= PAGE_SIZE >>
2397                                 E1000_PSRCTL_BSIZE1_SHIFT;
2398                         break;
2399                 }
2400
2401                 ew32(PSRCTL, psrctl);
2402         }
2403
2404         ew32(RCTL, rctl);
2405         /* just started the receive unit, no need to restart */
2406         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2407 }
2408
2409 /**
2410  * e1000_configure_rx - Configure Receive Unit after Reset
2411  * @adapter: board private structure
2412  *
2413  * Configure the Rx unit of the MAC after a reset.
2414  **/
2415 static void e1000_configure_rx(struct e1000_adapter *adapter)
2416 {
2417         struct e1000_hw *hw = &adapter->hw;
2418         struct e1000_ring *rx_ring = adapter->rx_ring;
2419         u64 rdba;
2420         u32 rdlen, rctl, rxcsum, ctrl_ext;
2421
2422         if (adapter->rx_ps_pages) {
2423                 /* this is a 32 byte descriptor */
2424                 rdlen = rx_ring->count *
2425                         sizeof(union e1000_rx_desc_packet_split);
2426                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2427                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2428         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2429                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2430                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2431                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2432         } else {
2433                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2434                 adapter->clean_rx = e1000_clean_rx_irq;
2435                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2436         }
2437
2438         /* disable receives while setting up the descriptors */
2439         rctl = er32(RCTL);
2440         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2441         e1e_flush();
2442         msleep(10);
2443
2444         /* set the Receive Delay Timer Register */
2445         ew32(RDTR, adapter->rx_int_delay);
2446
2447         /* irq moderation */
2448         ew32(RADV, adapter->rx_abs_int_delay);
2449         if (adapter->itr_setting != 0)
2450                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2451
2452         ctrl_ext = er32(CTRL_EXT);
2453         /* Reset delay timers after every interrupt */
2454         ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2455         /* Auto-Mask interrupts upon ICR access */
2456         ctrl_ext |= E1000_CTRL_EXT_IAME;
2457         ew32(IAM, 0xffffffff);
2458         ew32(CTRL_EXT, ctrl_ext);
2459         e1e_flush();
2460
2461         /*
2462          * Setup the HW Rx Head and Tail Descriptor Pointers and
2463          * the Base and Length of the Rx Descriptor Ring
2464          */
2465         rdba = rx_ring->dma;
2466         ew32(RDBAL, (rdba & DMA_32BIT_MASK));
2467         ew32(RDBAH, (rdba >> 32));
2468         ew32(RDLEN, rdlen);
2469         ew32(RDH, 0);
2470         ew32(RDT, 0);
2471         rx_ring->head = E1000_RDH;
2472         rx_ring->tail = E1000_RDT;
2473
2474         /* Enable Receive Checksum Offload for TCP and UDP */
2475         rxcsum = er32(RXCSUM);
2476         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2477                 rxcsum |= E1000_RXCSUM_TUOFL;
2478
2479                 /*
2480                  * IPv4 payload checksum for UDP fragments must be
2481                  * used in conjunction with packet-split.
2482                  */
2483                 if (adapter->rx_ps_pages)
2484                         rxcsum |= E1000_RXCSUM_IPPCSE;
2485         } else {
2486                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2487                 /* no need to clear IPPCSE as it defaults to 0 */
2488         }
2489         ew32(RXCSUM, rxcsum);
2490
2491         /*
2492          * Enable early receives on supported devices, only takes effect when
2493          * packet size is equal or larger than the specified value (in 8 byte
2494          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2495          */
2496         if ((adapter->flags & FLAG_HAS_ERT) &&
2497             (adapter->netdev->mtu > ETH_DATA_LEN)) {
2498                 u32 rxdctl = er32(RXDCTL(0));
2499                 ew32(RXDCTL(0), rxdctl | 0x3);
2500                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2501                 /*
2502                  * With jumbo frames and early-receive enabled, excessive
2503                  * C4->C2 latencies result in dropped transactions.
2504                  */
2505                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2506                                           e1000e_driver_name, 55);
2507         } else {
2508                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2509                                           e1000e_driver_name,
2510                                           PM_QOS_DEFAULT_VALUE);
2511         }
2512
2513         /* Enable Receives */
2514         ew32(RCTL, rctl);
2515 }
2516
2517 /**
2518  *  e1000_update_mc_addr_list - Update Multicast addresses
2519  *  @hw: pointer to the HW structure
2520  *  @mc_addr_list: array of multicast addresses to program
2521  *  @mc_addr_count: number of multicast addresses to program
2522  *  @rar_used_count: the first RAR register free to program
2523  *  @rar_count: total number of supported Receive Address Registers
2524  *
2525  *  Updates the Receive Address Registers and Multicast Table Array.
2526  *  The caller must have a packed mc_addr_list of multicast addresses.
2527  *  The parameter rar_count will usually be hw->mac.rar_entry_count
2528  *  unless there are workarounds that change this.  Currently no func pointer
2529  *  exists and all implementations are handled in the generic version of this
2530  *  function.
2531  **/
2532 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2533                                       u32 mc_addr_count, u32 rar_used_count,
2534                                       u32 rar_count)
2535 {
2536         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2537                                         rar_used_count, rar_count);
2538 }
2539
2540 /**
2541  * e1000_set_multi - Multicast and Promiscuous mode set
2542  * @netdev: network interface device structure
2543  *
2544  * The set_multi entry point is called whenever the multicast address
2545  * list or the network interface flags are updated.  This routine is
2546  * responsible for configuring the hardware for proper multicast,
2547  * promiscuous mode, and all-multi behavior.
2548  **/
2549 static void e1000_set_multi(struct net_device *netdev)
2550 {
2551         struct e1000_adapter *adapter = netdev_priv(netdev);
2552         struct e1000_hw *hw = &adapter->hw;
2553         struct e1000_mac_info *mac = &hw->mac;
2554         struct dev_mc_list *mc_ptr;
2555         u8  *mta_list;
2556         u32 rctl;
2557         int i;
2558
2559         /* Check for Promiscuous and All Multicast modes */
2560
2561         rctl = er32(RCTL);
2562
2563         if (netdev->flags & IFF_PROMISC) {
2564                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2565                 rctl &= ~E1000_RCTL_VFE;
2566         } else {
2567                 if (netdev->flags & IFF_ALLMULTI) {
2568                         rctl |= E1000_RCTL_MPE;
2569                         rctl &= ~E1000_RCTL_UPE;
2570                 } else {
2571                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2572                 }
2573                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2574                         rctl |= E1000_RCTL_VFE;
2575         }
2576
2577         ew32(RCTL, rctl);
2578
2579         if (netdev->mc_count) {
2580                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2581                 if (!mta_list)
2582                         return;
2583
2584                 /* prepare a packed array of only addresses. */
2585                 mc_ptr = netdev->mc_list;
2586
2587                 for (i = 0; i < netdev->mc_count; i++) {
2588                         if (!mc_ptr)
2589                                 break;
2590                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2591                                ETH_ALEN);
2592                         mc_ptr = mc_ptr->next;
2593                 }
2594
2595                 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2596                                           mac->rar_entry_count);
2597                 kfree(mta_list);
2598         } else {
2599                 /*
2600                  * if we're called from probe, we might not have
2601                  * anything to do here, so clear out the list
2602                  */
2603                 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2604         }
2605 }
2606
2607 /**
2608  * e1000_configure - configure the hardware for Rx and Tx
2609  * @adapter: private board structure
2610  **/
2611 static void e1000_configure(struct e1000_adapter *adapter)
2612 {
2613         e1000_set_multi(adapter->netdev);
2614
2615         e1000_restore_vlan(adapter);
2616         e1000_init_manageability(adapter);
2617
2618         e1000_configure_tx(adapter);
2619         e1000_setup_rctl(adapter);
2620         e1000_configure_rx(adapter);
2621         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2622 }
2623
2624 /**
2625  * e1000e_power_up_phy - restore link in case the phy was powered down
2626  * @adapter: address of board private structure
2627  *
2628  * The phy may be powered down to save power and turn off link when the
2629  * driver is unloaded and wake on lan is not enabled (among others)
2630  * *** this routine MUST be followed by a call to e1000e_reset ***
2631  **/
2632 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2633 {
2634         u16 mii_reg = 0;
2635
2636         /* Just clear the power down bit to wake the phy back up */
2637         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2638                 /*
2639                  * According to the manual, the phy will retain its
2640                  * settings across a power-down/up cycle
2641                  */
2642                 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2643                 mii_reg &= ~MII_CR_POWER_DOWN;
2644                 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2645         }
2646
2647         adapter->hw.mac.ops.setup_link(&adapter->hw);
2648 }
2649
2650 /**
2651  * e1000_power_down_phy - Power down the PHY
2652  *
2653  * Power down the PHY so no link is implied when interface is down
2654  * The PHY cannot be powered down is management or WoL is active
2655  */
2656 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2657 {
2658         struct e1000_hw *hw = &adapter->hw;
2659         u16 mii_reg;
2660
2661         /* WoL is enabled */
2662         if (adapter->wol)
2663                 return;
2664
2665         /* non-copper PHY? */
2666         if (adapter->hw.phy.media_type != e1000_media_type_copper)
2667                 return;
2668
2669         /* reset is blocked because of a SoL/IDER session */
2670         if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2671                 return;
2672
2673         /* manageability (AMT) is enabled */
2674         if (er32(MANC) & E1000_MANC_SMBUS_EN)
2675                 return;
2676
2677         /* power down the PHY */
2678         e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2679         mii_reg |= MII_CR_POWER_DOWN;
2680         e1e_wphy(hw, PHY_CONTROL, mii_reg);
2681         mdelay(1);
2682 }
2683
2684 /**
2685  * e1000e_reset - bring the hardware into a known good state
2686  *
2687  * This function boots the hardware and enables some settings that
2688  * require a configuration cycle of the hardware - those cannot be
2689  * set/changed during runtime. After reset the device needs to be
2690  * properly configured for Rx, Tx etc.
2691  */
2692 void e1000e_reset(struct e1000_adapter *adapter)
2693 {
2694         struct e1000_mac_info *mac = &adapter->hw.mac;
2695         struct e1000_fc_info *fc = &adapter->hw.fc;
2696         struct e1000_hw *hw = &adapter->hw;
2697         u32 tx_space, min_tx_space, min_rx_space;
2698         u32 pba = adapter->pba;
2699         u16 hwm;
2700
2701         /* reset Packet Buffer Allocation to default */
2702         ew32(PBA, pba);
2703
2704         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2705                 /*
2706                  * To maintain wire speed transmits, the Tx FIFO should be
2707                  * large enough to accommodate two full transmit packets,
2708                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2709                  * the Rx FIFO should be large enough to accommodate at least
2710                  * one full receive packet and is similarly rounded up and
2711                  * expressed in KB.
2712                  */
2713                 pba = er32(PBA);
2714                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2715                 tx_space = pba >> 16;
2716                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2717                 pba &= 0xffff;
2718                 /*
2719                  * the Tx fifo also stores 16 bytes of information about the tx
2720                  * but don't include ethernet FCS because hardware appends it
2721                  */
2722                 min_tx_space = (adapter->max_frame_size +
2723                                 sizeof(struct e1000_tx_desc) -
2724                                 ETH_FCS_LEN) * 2;
2725                 min_tx_space = ALIGN(min_tx_space, 1024);
2726                 min_tx_space >>= 10;
2727                 /* software strips receive CRC, so leave room for it */
2728                 min_rx_space = adapter->max_frame_size;
2729                 min_rx_space = ALIGN(min_rx_space, 1024);
2730                 min_rx_space >>= 10;
2731
2732                 /*
2733                  * If current Tx allocation is less than the min Tx FIFO size,
2734                  * and the min Tx FIFO size is less than the current Rx FIFO
2735                  * allocation, take space away from current Rx allocation
2736                  */
2737                 if ((tx_space < min_tx_space) &&
2738                     ((min_tx_space - tx_space) < pba)) {
2739                         pba -= min_tx_space - tx_space;
2740
2741                         /*
2742                          * if short on Rx space, Rx wins and must trump tx
2743                          * adjustment or use Early Receive if available
2744                          */
2745                         if ((pba < min_rx_space) &&
2746                             (!(adapter->flags & FLAG_HAS_ERT)))
2747                                 /* ERT enabled in e1000_configure_rx */
2748                                 pba = min_rx_space;
2749                 }
2750
2751                 ew32(PBA, pba);
2752         }
2753
2754
2755         /*
2756          * flow control settings
2757          *
2758          * The high water mark must be low enough to fit one full frame
2759          * (or the size used for early receive) above it in the Rx FIFO.
2760          * Set it to the lower of:
2761          * - 90% of the Rx FIFO size, and
2762          * - the full Rx FIFO size minus the early receive size (for parts
2763          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2764          * - the full Rx FIFO size minus one full frame
2765          */
2766         if (adapter->flags & FLAG_HAS_ERT)
2767                 hwm = min(((pba << 10) * 9 / 10),
2768                           ((pba << 10) - (E1000_ERT_2048 << 3)));
2769         else
2770                 hwm = min(((pba << 10) * 9 / 10),
2771                           ((pba << 10) - adapter->max_frame_size));
2772
2773         fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
2774         fc->low_water = fc->high_water - 8;
2775
2776         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2777                 fc->pause_time = 0xFFFF;
2778         else
2779                 fc->pause_time = E1000_FC_PAUSE_TIME;
2780         fc->send_xon = 1;
2781         fc->current_mode = fc->requested_mode;
2782
2783         /* Allow time for pending master requests to run */
2784         mac->ops.reset_hw(hw);
2785
2786         /*
2787          * For parts with AMT enabled, let the firmware know
2788          * that the network interface is in control
2789          */
2790         if (adapter->flags & FLAG_HAS_AMT)
2791                 e1000_get_hw_control(adapter);
2792
2793         ew32(WUC, 0);
2794
2795         if (mac->ops.init_hw(hw))
2796                 e_err("Hardware Error\n");
2797
2798         e1000_update_mng_vlan(adapter);
2799
2800         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2801         ew32(VET, ETH_P_8021Q);
2802
2803         e1000e_reset_adaptive(hw);
2804         e1000_get_phy_info(hw);
2805
2806         if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2807                 u16 phy_data = 0;
2808                 /*
2809                  * speed up time to link by disabling smart power down, ignore
2810                  * the return value of this function because there is nothing
2811                  * different we would do if it failed
2812                  */
2813                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2814                 phy_data &= ~IGP02E1000_PM_SPD;
2815                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2816         }
2817 }
2818
2819 int e1000e_up(struct e1000_adapter *adapter)
2820 {
2821         struct e1000_hw *hw = &adapter->hw;
2822
2823         /* hardware has been reset, we need to reload some things */
2824         e1000_configure(adapter);
2825
2826         clear_bit(__E1000_DOWN, &adapter->state);
2827
2828         napi_enable(&adapter->napi);
2829         if (adapter->msix_entries)
2830                 e1000_configure_msix(adapter);
2831         e1000_irq_enable(adapter);
2832
2833         /* fire a link change interrupt to start the watchdog */
2834         ew32(ICS, E1000_ICS_LSC);
2835         return 0;
2836 }
2837
2838 void e1000e_down(struct e1000_adapter *adapter)
2839 {
2840         struct net_device *netdev = adapter->netdev;
2841         struct e1000_hw *hw = &adapter->hw;
2842         u32 tctl, rctl;
2843
2844         /*
2845          * signal that we're down so the interrupt handler does not
2846          * reschedule our watchdog timer
2847          */
2848         set_bit(__E1000_DOWN, &adapter->state);
2849
2850         /* disable receives in the hardware */
2851         rctl = er32(RCTL);
2852         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2853         /* flush and sleep below */
2854
2855         netif_tx_stop_all_queues(netdev);
2856
2857         /* disable transmits in the hardware */
2858         tctl = er32(TCTL);
2859         tctl &= ~E1000_TCTL_EN;
2860         ew32(TCTL, tctl);
2861         /* flush both disables and wait for them to finish */
2862         e1e_flush();
2863         msleep(10);
2864
2865         napi_disable(&adapter->napi);
2866         e1000_irq_disable(adapter);
2867
2868         del_timer_sync(&adapter->watchdog_timer);
2869         del_timer_sync(&adapter->phy_info_timer);
2870
2871         netdev->tx_queue_len = adapter->tx_queue_len;
2872         netif_carrier_off(netdev);
2873         adapter->link_speed = 0;
2874         adapter->link_duplex = 0;
2875
2876         if (!pci_channel_offline(adapter->pdev))
2877                 e1000e_reset(adapter);
2878         e1000_clean_tx_ring(adapter);
2879         e1000_clean_rx_ring(adapter);
2880
2881         /*
2882          * TODO: for power management, we could drop the link and
2883          * pci_disable_device here.
2884          */
2885 }
2886
2887 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2888 {
2889         might_sleep();
2890         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2891                 msleep(1);
2892         e1000e_down(adapter);
2893         e1000e_up(adapter);
2894         clear_bit(__E1000_RESETTING, &adapter->state);
2895 }
2896
2897 /**
2898  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2899  * @adapter: board private structure to initialize
2900  *
2901  * e1000_sw_init initializes the Adapter private data structure.
2902  * Fields are initialized based on PCI device information and
2903  * OS network device settings (MTU size).
2904  **/
2905 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2906 {
2907         struct net_device *netdev = adapter->netdev;
2908
2909         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2910         adapter->rx_ps_bsize0 = 128;
2911         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2912         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2913
2914         e1000e_set_interrupt_capability(adapter);
2915
2916         if (e1000_alloc_queues(adapter))
2917                 return -ENOMEM;
2918
2919         /* Explicitly disable IRQ since the NIC can be in any state. */
2920         e1000_irq_disable(adapter);
2921
2922         set_bit(__E1000_DOWN, &adapter->state);
2923         return 0;
2924 }
2925
2926 /**
2927  * e1000_intr_msi_test - Interrupt Handler
2928  * @irq: interrupt number
2929  * @data: pointer to a network interface device structure
2930  **/
2931 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2932 {
2933         struct net_device *netdev = data;
2934         struct e1000_adapter *adapter = netdev_priv(netdev);
2935         struct e1000_hw *hw = &adapter->hw;
2936         u32 icr = er32(ICR);
2937
2938         e_dbg("%s: icr is %08X\n", netdev->name, icr);
2939         if (icr & E1000_ICR_RXSEQ) {
2940                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2941                 wmb();
2942         }
2943
2944         return IRQ_HANDLED;
2945 }
2946
2947 /**
2948  * e1000_test_msi_interrupt - Returns 0 for successful test
2949  * @adapter: board private struct
2950  *
2951  * code flow taken from tg3.c
2952  **/
2953 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2954 {
2955         struct net_device *netdev = adapter->netdev;
2956         struct e1000_hw *hw = &adapter->hw;
2957         int err;
2958
2959         /* poll_enable hasn't been called yet, so don't need disable */
2960         /* clear any pending events */
2961         er32(ICR);
2962
2963         /* free the real vector and request a test handler */
2964         e1000_free_irq(adapter);
2965         e1000e_reset_interrupt_capability(adapter);
2966
2967         /* Assume that the test fails, if it succeeds then the test
2968          * MSI irq handler will unset this flag */
2969         adapter->flags |= FLAG_MSI_TEST_FAILED;
2970
2971         err = pci_enable_msi(adapter->pdev);
2972         if (err)
2973                 goto msi_test_failed;
2974
2975         err = request_irq(adapter->pdev->irq, &e1000_intr_msi_test, 0,
2976                           netdev->name, netdev);
2977         if (err) {
2978                 pci_disable_msi(adapter->pdev);
2979                 goto msi_test_failed;
2980         }
2981
2982         wmb();
2983
2984         e1000_irq_enable(adapter);
2985
2986         /* fire an unusual interrupt on the test handler */
2987         ew32(ICS, E1000_ICS_RXSEQ);
2988         e1e_flush();
2989         msleep(50);
2990
2991         e1000_irq_disable(adapter);
2992
2993         rmb();
2994
2995         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
2996                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2997                 err = -EIO;
2998                 e_info("MSI interrupt test failed!\n");
2999         }
3000
3001         free_irq(adapter->pdev->irq, netdev);
3002         pci_disable_msi(adapter->pdev);
3003
3004         if (err == -EIO)
3005                 goto msi_test_failed;
3006
3007         /* okay so the test worked, restore settings */
3008         e_dbg("%s: MSI interrupt test succeeded!\n", netdev->name);
3009 msi_test_failed:
3010         e1000e_set_interrupt_capability(adapter);
3011         e1000_request_irq(adapter);
3012         return err;
3013 }
3014
3015 /**
3016  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3017  * @adapter: board private struct
3018  *
3019  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3020  **/
3021 static int e1000_test_msi(struct e1000_adapter *adapter)
3022 {
3023         int err;
3024         u16 pci_cmd;
3025
3026         if (!(adapter->flags & FLAG_MSI_ENABLED))
3027                 return 0;
3028
3029         /* disable SERR in case the MSI write causes a master abort */
3030         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3031         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3032                               pci_cmd & ~PCI_COMMAND_SERR);
3033
3034         err = e1000_test_msi_interrupt(adapter);
3035
3036         /* restore previous setting of command word */
3037         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3038
3039         /* success ! */
3040         if (!err)
3041                 return 0;
3042
3043         /* EIO means MSI test failed */
3044         if (err != -EIO)
3045                 return err;
3046
3047         /* back to INTx mode */
3048         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3049
3050         e1000_free_irq(adapter);
3051
3052         err = e1000_request_irq(adapter);
3053
3054         return err;
3055 }
3056
3057 /**
3058  * e1000_open - Called when a network interface is made active
3059  * @netdev: network interface device structure
3060  *
3061  * Returns 0 on success, negative value on failure
3062  *
3063  * The open entry point is called when a network interface is made
3064  * active by the system (IFF_UP).  At this point all resources needed
3065  * for transmit and receive operations are allocated, the interrupt
3066  * handler is registered with the OS, the watchdog timer is started,
3067  * and the stack is notified that the interface is ready.
3068  **/
3069 static int e1000_open(struct net_device *netdev)
3070 {
3071         struct e1000_adapter *adapter = netdev_priv(netdev);
3072         struct e1000_hw *hw = &adapter->hw;
3073         int err;
3074
3075         /* disallow open during test */
3076         if (test_bit(__E1000_TESTING, &adapter->state))
3077                 return -EBUSY;
3078
3079         /* allocate transmit descriptors */
3080         err = e1000e_setup_tx_resources(adapter);
3081         if (err)
3082                 goto err_setup_tx;
3083
3084         /* allocate receive descriptors */
3085         err = e1000e_setup_rx_resources(adapter);
3086         if (err)
3087                 goto err_setup_rx;
3088
3089         e1000e_power_up_phy(adapter);
3090
3091         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3092         if ((adapter->hw.mng_cookie.status &
3093              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3094                 e1000_update_mng_vlan(adapter);
3095
3096         /*
3097          * If AMT is enabled, let the firmware know that the network
3098          * interface is now open
3099          */
3100         if (adapter->flags & FLAG_HAS_AMT)
3101                 e1000_get_hw_control(adapter);
3102
3103         /*
3104          * before we allocate an interrupt, we must be ready to handle it.
3105          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3106          * as soon as we call pci_request_irq, so we have to setup our
3107          * clean_rx handler before we do so.
3108          */
3109         e1000_configure(adapter);
3110
3111         err = e1000_request_irq(adapter);
3112         if (err)
3113                 goto err_req_irq;
3114
3115         /*
3116          * Work around PCIe errata with MSI interrupts causing some chipsets to
3117          * ignore e1000e MSI messages, which means we need to test our MSI
3118          * interrupt now
3119          */
3120         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3121                 err = e1000_test_msi(adapter);
3122                 if (err) {
3123                         e_err("Interrupt allocation failed\n");
3124                         goto err_req_irq;
3125                 }
3126         }
3127
3128         /* From here on the code is the same as e1000e_up() */
3129         clear_bit(__E1000_DOWN, &adapter->state);
3130
3131         napi_enable(&adapter->napi);
3132
3133         e1000_irq_enable(adapter);
3134
3135         netif_tx_start_all_queues(netdev);
3136
3137         /* fire a link status change interrupt to start the watchdog */
3138         ew32(ICS, E1000_ICS_LSC);
3139
3140         return 0;
3141
3142 err_req_irq:
3143         e1000_release_hw_control(adapter);
3144         e1000_power_down_phy(adapter);
3145         e1000e_free_rx_resources(adapter);
3146 err_setup_rx:
3147         e1000e_free_tx_resources(adapter);
3148 err_setup_tx:
3149         e1000e_reset(adapter);
3150
3151         return err;
3152 }
3153
3154 /**
3155  * e1000_close - Disables a network interface
3156  * @netdev: network interface device structure
3157  *
3158  * Returns 0, this is not allowed to fail
3159  *
3160  * The close entry point is called when an interface is de-activated
3161  * by the OS.  The hardware is still under the drivers control, but
3162  * needs to be disabled.  A global MAC reset is issued to stop the
3163  * hardware, and all transmit and receive resources are freed.
3164  **/
3165 static int e1000_close(struct net_device *netdev)
3166 {
3167         struct e1000_adapter *adapter = netdev_priv(netdev);
3168
3169         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3170         e1000e_down(adapter);
3171         e1000_power_down_phy(adapter);
3172         e1000_free_irq(adapter);
3173
3174         e1000e_free_tx_resources(adapter);
3175         e1000e_free_rx_resources(adapter);
3176
3177         /*
3178          * kill manageability vlan ID if supported, but not if a vlan with
3179          * the same ID is registered on the host OS (let 8021q kill it)
3180          */
3181         if ((adapter->hw.mng_cookie.status &
3182                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3183              !(adapter->vlgrp &&
3184                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3185                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3186
3187         /*
3188          * If AMT is enabled, let the firmware know that the network
3189          * interface is now closed
3190          */
3191         if (adapter->flags & FLAG_HAS_AMT)
3192                 e1000_release_hw_control(adapter);
3193
3194         return 0;
3195 }
3196 /**
3197  * e1000_set_mac - Change the Ethernet Address of the NIC
3198  * @netdev: network interface device structure
3199  * @p: pointer to an address structure
3200  *
3201  * Returns 0 on success, negative on failure
3202  **/
3203 static int e1000_set_mac(struct net_device *netdev, void *p)
3204 {
3205         struct e1000_adapter *adapter = netdev_priv(netdev);
3206         struct sockaddr *addr = p;
3207
3208         if (!is_valid_ether_addr(addr->sa_data))
3209                 return -EADDRNOTAVAIL;
3210
3211         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3212         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3213
3214         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3215
3216         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3217                 /* activate the work around */
3218                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3219
3220                 /*
3221                  * Hold a copy of the LAA in RAR[14] This is done so that
3222                  * between the time RAR[0] gets clobbered  and the time it
3223                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3224                  * of the RARs and no incoming packets directed to this port
3225                  * are dropped. Eventually the LAA will be in RAR[0] and
3226                  * RAR[14]
3227                  */
3228                 e1000e_rar_set(&adapter->hw,
3229                               adapter->hw.mac.addr,
3230                               adapter->hw.mac.rar_entry_count - 1);
3231         }
3232
3233         return 0;
3234 }
3235
3236 /**
3237  * e1000e_update_phy_task - work thread to update phy
3238  * @work: pointer to our work struct
3239  *
3240  * this worker thread exists because we must acquire a
3241  * semaphore to read the phy, which we could msleep while
3242  * waiting for it, and we can't msleep in a timer.
3243  **/
3244 static void e1000e_update_phy_task(struct work_struct *work)
3245 {
3246         struct e1000_adapter *adapter = container_of(work,
3247                                         struct e1000_adapter, update_phy_task);
3248         e1000_get_phy_info(&adapter->hw);
3249 }
3250
3251 /*
3252  * Need to wait a few seconds after link up to get diagnostic information from
3253  * the phy
3254  */
3255 static void e1000_update_phy_info(unsigned long data)
3256 {
3257         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3258         schedule_work(&adapter->update_phy_task);
3259 }
3260
3261 /**
3262  * e1000e_update_stats - Update the board statistics counters
3263  * @adapter: board private structure
3264  **/
3265 void e1000e_update_stats(struct e1000_adapter *adapter)
3266 {
3267         struct e1000_hw *hw = &adapter->hw;
3268         struct pci_dev *pdev = adapter->pdev;
3269
3270         /*
3271          * Prevent stats update while adapter is being reset, or if the pci
3272          * connection is down.
3273          */
3274         if (adapter->link_speed == 0)
3275                 return;
3276         if (pci_channel_offline(pdev))
3277                 return;
3278
3279         adapter->stats.crcerrs += er32(CRCERRS);
3280         adapter->stats.gprc += er32(GPRC);
3281         adapter->stats.gorc += er32(GORCL);
3282         er32(GORCH); /* Clear gorc */
3283         adapter->stats.bprc += er32(BPRC);
3284         adapter->stats.mprc += er32(MPRC);
3285         adapter->stats.roc += er32(ROC);
3286
3287         adapter->stats.mpc += er32(MPC);
3288         adapter->stats.scc += er32(SCC);
3289         adapter->stats.ecol += er32(ECOL);
3290         adapter->stats.mcc += er32(MCC);
3291         adapter->stats.latecol += er32(LATECOL);
3292         adapter->stats.dc += er32(DC);
3293         adapter->stats.xonrxc += er32(XONRXC);
3294         adapter->stats.xontxc += er32(XONTXC);
3295         adapter->stats.xoffrxc += er32(XOFFRXC);
3296         adapter->stats.xofftxc += er32(XOFFTXC);
3297         adapter->stats.gptc += er32(GPTC);
3298         adapter->stats.gotc += er32(GOTCL);
3299         er32(GOTCH); /* Clear gotc */
3300         adapter->stats.rnbc += er32(RNBC);
3301         adapter->stats.ruc += er32(RUC);
3302
3303         adapter->stats.mptc += er32(MPTC);
3304         adapter->stats.bptc += er32(BPTC);
3305
3306         /* used for adaptive IFS */
3307
3308         hw->mac.tx_packet_delta = er32(TPT);
3309         adapter->stats.tpt += hw->mac.tx_packet_delta;
3310         hw->mac.collision_delta = er32(COLC);
3311         adapter->stats.colc += hw->mac.collision_delta;
3312
3313         adapter->stats.algnerrc += er32(ALGNERRC);
3314         adapter->stats.rxerrc += er32(RXERRC);
3315         if (hw->mac.type != e1000_82574)
3316                 adapter->stats.tncrs += er32(TNCRS);
3317         adapter->stats.cexterr += er32(CEXTERR);
3318         adapter->stats.tsctc += er32(TSCTC);
3319         adapter->stats.tsctfc += er32(TSCTFC);
3320
3321         /* Fill out the OS statistics structure */
3322         adapter->net_stats.multicast = adapter->stats.mprc;
3323         adapter->net_stats.collisions = adapter->stats.colc;
3324
3325         /* Rx Errors */
3326
3327         /*
3328          * RLEC on some newer hardware can be incorrect so build
3329          * our own version based on RUC and ROC
3330          */
3331         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3332                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3333                 adapter->stats.ruc + adapter->stats.roc +
3334                 adapter->stats.cexterr;
3335         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3336                                               adapter->stats.roc;
3337         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3338         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3339         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3340
3341         /* Tx Errors */
3342         adapter->net_stats.tx_errors = adapter->stats.ecol +
3343                                        adapter->stats.latecol;
3344         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3345         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3346         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3347
3348         /* Tx Dropped needs to be maintained elsewhere */
3349
3350         /* Management Stats */
3351         adapter->stats.mgptc += er32(MGTPTC);
3352         adapter->stats.mgprc += er32(MGTPRC);
3353         adapter->stats.mgpdc += er32(MGTPDC);
3354 }
3355
3356 /**
3357  * e1000_phy_read_status - Update the PHY register status snapshot
3358  * @adapter: board private structure
3359  **/
3360 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3361 {
3362         struct e1000_hw *hw = &adapter->hw;
3363         struct e1000_phy_regs *phy = &adapter->phy_regs;
3364         int ret_val;
3365
3366         if ((er32(STATUS) & E1000_STATUS_LU) &&
3367             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3368                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3369                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3370                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3371                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3372                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3373                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3374                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3375                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3376                 if (ret_val)
3377                         e_warn("Error reading PHY register\n");
3378         } else {
3379                 /*
3380                  * Do not read PHY registers if link is not up
3381                  * Set values to typical power-on defaults
3382                  */
3383                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3384                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3385                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3386                              BMSR_ERCAP);
3387                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3388                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3389                 phy->lpa = 0;
3390                 phy->expansion = EXPANSION_ENABLENPAGE;
3391                 phy->ctrl1000 = ADVERTISE_1000FULL;
3392                 phy->stat1000 = 0;
3393                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3394         }
3395 }
3396
3397 static void e1000_print_link_info(struct e1000_adapter *adapter)
3398 {
3399         struct e1000_hw *hw = &adapter->hw;
3400         u32 ctrl = er32(CTRL);
3401
3402         /* Link status message must follow this format for user tools */
3403         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3404                "Flow Control: %s\n",
3405                adapter->netdev->name,
3406                adapter->link_speed,
3407                (adapter->link_duplex == FULL_DUPLEX) ?
3408                                 "Full Duplex" : "Half Duplex",
3409                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3410                                 "RX/TX" :
3411                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3412                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3413 }
3414
3415 bool e1000_has_link(struct e1000_adapter *adapter)
3416 {
3417         struct e1000_hw *hw = &adapter->hw;
3418         bool link_active = 0;
3419         s32 ret_val = 0;
3420
3421         /*
3422          * get_link_status is set on LSC (link status) interrupt or
3423          * Rx sequence error interrupt.  get_link_status will stay
3424          * false until the check_for_link establishes link
3425          * for copper adapters ONLY
3426          */
3427         switch (hw->phy.media_type) {
3428         case e1000_media_type_copper:
3429                 if (hw->mac.get_link_status) {
3430                         ret_val = hw->mac.ops.check_for_link(hw);
3431                         link_active = !hw->mac.get_link_status;
3432                 } else {
3433                         link_active = 1;
3434                 }
3435                 break;
3436         case e1000_media_type_fiber:
3437                 ret_val = hw->mac.ops.check_for_link(hw);
3438                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3439                 break;
3440         case e1000_media_type_internal_serdes:
3441                 ret_val = hw->mac.ops.check_for_link(hw);
3442                 link_active = adapter->hw.mac.serdes_has_link;
3443                 break;
3444         default:
3445         case e1000_media_type_unknown:
3446                 break;
3447         }
3448
3449         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3450             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3451                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3452                 e_info("Gigabit has been disabled, downgrading speed\n");
3453         }
3454
3455         return link_active;
3456 }
3457
3458 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3459 {
3460         /* make sure the receive unit is started */
3461         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3462             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3463                 struct e1000_hw *hw = &adapter->hw;
3464                 u32 rctl = er32(RCTL);
3465                 ew32(RCTL, rctl | E1000_RCTL_EN);
3466                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3467         }
3468 }
3469
3470 /**
3471  * e1000_watchdog - Timer Call-back
3472  * @data: pointer to adapter cast into an unsigned long
3473  **/
3474 static void e1000_watchdog(unsigned long data)
3475 {
3476         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3477
3478         /* Do the rest outside of interrupt context */
3479         schedule_work(&adapter->watchdog_task);
3480
3481         /* TODO: make this use queue_delayed_work() */
3482 }
3483
3484 static void e1000_watchdog_task(struct work_struct *work)
3485 {
3486         struct e1000_adapter *adapter = container_of(work,
3487                                         struct e1000_adapter, watchdog_task);
3488         struct net_device *netdev = adapter->netdev;
3489         struct e1000_mac_info *mac = &adapter->hw.mac;
3490         struct e1000_phy_info *phy = &adapter->hw.phy;
3491         struct e1000_ring *tx_ring = adapter->tx_ring;
3492         struct e1000_hw *hw = &adapter->hw;
3493         u32 link, tctl;
3494         int tx_pending = 0;
3495
3496         link = e1000_has_link(adapter);
3497         if ((netif_carrier_ok(netdev)) && link) {
3498                 e1000e_enable_receives(adapter);
3499                 goto link_up;
3500         }
3501
3502         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3503             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3504                 e1000_update_mng_vlan(adapter);
3505
3506         if (link) {
3507                 if (!netif_carrier_ok(netdev)) {
3508                         bool txb2b = 1;
3509                         /* update snapshot of PHY registers on LSC */
3510                         e1000_phy_read_status(adapter);
3511                         mac->ops.get_link_up_info(&adapter->hw,
3512                                                    &adapter->link_speed,
3513                                                    &adapter->link_duplex);
3514                         e1000_print_link_info(adapter);
3515                         /*
3516                          * On supported PHYs, check for duplex mismatch only
3517                          * if link has autonegotiated at 10/100 half
3518                          */
3519                         if ((hw->phy.type == e1000_phy_igp_3 ||
3520                              hw->phy.type == e1000_phy_bm) &&
3521                             (hw->mac.autoneg == true) &&
3522                             (adapter->link_speed == SPEED_10 ||
3523                              adapter->link_speed == SPEED_100) &&
3524                             (adapter->link_duplex == HALF_DUPLEX)) {
3525                                 u16 autoneg_exp;
3526
3527                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3528
3529                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3530                                         e_info("Autonegotiated half duplex but"
3531                                                " link partner cannot autoneg. "
3532                                                " Try forcing full duplex if "
3533                                                "link gets many collisions.\n");
3534                         }
3535
3536                         /*
3537                          * tweak tx_queue_len according to speed/duplex
3538                          * and adjust the timeout factor
3539                          */
3540                         netdev->tx_queue_len = adapter->tx_queue_len;
3541                         adapter->tx_timeout_factor = 1;
3542                         switch (adapter->link_speed) {
3543                         case SPEED_10:
3544                                 txb2b = 0;
3545                                 netdev->tx_queue_len = 10;
3546                                 adapter->tx_timeout_factor = 16;
3547                                 break;
3548                         case SPEED_100:
3549                                 txb2b = 0;
3550                                 netdev->tx_queue_len = 100;
3551                                 /* maybe add some timeout factor ? */
3552                                 break;
3553                         }
3554
3555                         /*
3556                          * workaround: re-program speed mode bit after
3557                          * link-up event
3558                          */
3559                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3560                             !txb2b) {
3561                                 u32 tarc0;
3562                                 tarc0 = er32(TARC(0));
3563                                 tarc0 &= ~SPEED_MODE_BIT;
3564                                 ew32(TARC(0), tarc0);
3565                         }
3566
3567                         /*
3568                          * disable TSO for pcie and 10/100 speeds, to avoid
3569                          * some hardware issues
3570                          */
3571                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3572                                 switch (adapter->link_speed) {
3573                                 case SPEED_10:
3574                                 case SPEED_100:
3575                                         e_info("10/100 speed: disabling TSO\n");
3576                                         netdev->features &= ~NETIF_F_TSO;
3577                                         netdev->features &= ~NETIF_F_TSO6;
3578                                         break;
3579                                 case SPEED_1000:
3580                                         netdev->features |= NETIF_F_TSO;
3581                                         netdev->features |= NETIF_F_TSO6;
3582                                         break;
3583                                 default:
3584                                         /* oops */
3585                                         break;
3586                                 }
3587                         }
3588
3589                         /*
3590                          * enable transmits in the hardware, need to do this
3591                          * after setting TARC(0)
3592                          */
3593                         tctl = er32(TCTL);
3594                         tctl |= E1000_TCTL_EN;
3595                         ew32(TCTL, tctl);
3596
3597                         /*
3598                          * Perform any post-link-up configuration before
3599                          * reporting link up.
3600                          */
3601                         if (phy->ops.cfg_on_link_up)
3602                                 phy->ops.cfg_on_link_up(hw);
3603
3604                         netif_carrier_on(netdev);
3605                         netif_tx_wake_all_queues(netdev);
3606
3607                         if (!test_bit(__E1000_DOWN, &adapter->state))
3608                                 mod_timer(&adapter->phy_info_timer,
3609                                           round_jiffies(jiffies + 2 * HZ));
3610                 }
3611         } else {
3612                 if (netif_carrier_ok(netdev)) {
3613                         adapter->link_speed = 0;
3614                         adapter->link_duplex = 0;
3615                         /* Link status message must follow this format */
3616                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
3617                                adapter->netdev->name);
3618                         netif_carrier_off(netdev);
3619                         netif_tx_stop_all_queues(netdev);
3620                         if (!test_bit(__E1000_DOWN, &adapter->state))
3621                                 mod_timer(&adapter->phy_info_timer,
3622                                           round_jiffies(jiffies + 2 * HZ));
3623
3624                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3625                                 schedule_work(&adapter->reset_task);
3626                 }
3627         }
3628
3629 link_up:
3630         e1000e_update_stats(adapter);
3631
3632         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3633         adapter->tpt_old = adapter->stats.tpt;
3634         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3635         adapter->colc_old = adapter->stats.colc;
3636
3637         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3638         adapter->gorc_old = adapter->stats.gorc;
3639         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3640         adapter->gotc_old = adapter->stats.gotc;
3641
3642         e1000e_update_adaptive(&adapter->hw);
3643
3644         if (!netif_carrier_ok(netdev)) {
3645                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3646                                tx_ring->count);
3647                 if (tx_pending) {
3648                         /*
3649                          * We've lost link, so the controller stops DMA,
3650                          * but we've got queued Tx work that's never going
3651                          * to get done, so reset controller to flush Tx.
3652                          * (Do the reset outside of interrupt context).
3653                          */
3654                         adapter->tx_timeout_count++;
3655                         schedule_work(&adapter->reset_task);
3656                 }
3657         }
3658
3659         /* Cause software interrupt to ensure Rx ring is cleaned */
3660         if (adapter->msix_entries)
3661                 ew32(ICS, adapter->rx_ring->ims_val);
3662         else
3663                 ew32(ICS, E1000_ICS_RXDMT0);
3664
3665         /* Force detection of hung controller every watchdog period */
3666         adapter->detect_tx_hung = 1;
3667
3668         /*
3669          * With 82571 controllers, LAA may be overwritten due to controller
3670          * reset from the other port. Set the appropriate LAA in RAR[0]
3671          */
3672         if (e1000e_get_laa_state_82571(hw))
3673                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3674
3675         /* Reset the timer */
3676         if (!test_bit(__E1000_DOWN, &adapter->state))
3677                 mod_timer(&adapter->watchdog_timer,
3678                           round_jiffies(jiffies + 2 * HZ));
3679 }
3680
3681 #define E1000_TX_FLAGS_CSUM             0x00000001
3682 #define E1000_TX_FLAGS_VLAN             0x00000002
3683 #define E1000_TX_FLAGS_TSO              0x00000004
3684 #define E1000_TX_FLAGS_IPV4             0x00000008
3685 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3686 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3687
3688 static int e1000_tso(struct e1000_adapter *adapter,
3689                      struct sk_buff *skb)
3690 {
3691         struct e1000_ring *tx_ring = adapter->tx_ring;
3692         struct e1000_context_desc *context_desc;
3693         struct e1000_buffer *buffer_info;
3694         unsigned int i;
3695         u32 cmd_length = 0;
3696         u16 ipcse = 0, tucse, mss;
3697         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3698         int err;
3699
3700         if (skb_is_gso(skb)) {
3701                 if (skb_header_cloned(skb)) {
3702                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3703                         if (err)
3704                                 return err;
3705                 }
3706
3707                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3708                 mss = skb_shinfo(skb)->gso_size;
3709                 if (skb->protocol == htons(ETH_P_IP)) {
3710                         struct iphdr *iph = ip_hdr(skb);
3711                         iph->tot_len = 0;
3712                         iph->check = 0;
3713                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3714                                                                  iph->daddr, 0,
3715                                                                  IPPROTO_TCP,
3716                                                                  0);
3717                         cmd_length = E1000_TXD_CMD_IP;
3718                         ipcse = skb_transport_offset(skb) - 1;
3719                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3720                         ipv6_hdr(skb)->payload_len = 0;
3721                         tcp_hdr(skb)->check =
3722                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3723                                                  &ipv6_hdr(skb)->daddr,
3724                                                  0, IPPROTO_TCP, 0);
3725                         ipcse = 0;
3726                 }
3727                 ipcss = skb_network_offset(skb);
3728                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3729                 tucss = skb_transport_offset(skb);
3730                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3731                 tucse = 0;
3732
3733                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3734                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3735
3736                 i = tx_ring->next_to_use;
3737                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3738                 buffer_info = &tx_ring->buffer_info[i];
3739
3740                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3741                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3742                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3743                 context_desc->upper_setup.tcp_fields.tucss = tucss;
3744                 context_desc->upper_setup.tcp_fields.tucso = tucso;
3745                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3746                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3747                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3748                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3749
3750                 buffer_info->time_stamp = jiffies;
3751                 buffer_info->next_to_watch = i;
3752
3753                 i++;
3754                 if (i == tx_ring->count)
3755                         i = 0;
3756                 tx_ring->next_to_use = i;
3757
3758                 return 1;
3759         }
3760
3761         return 0;
3762 }
3763
3764 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3765 {
3766         struct e1000_ring *tx_ring = adapter->tx_ring;
3767         struct e1000_context_desc *context_desc;
3768         struct e1000_buffer *buffer_info;
3769         unsigned int i;
3770         u8 css;
3771         u32 cmd_len = E1000_TXD_CMD_DEXT;
3772
3773         if (skb->ip_summed != CHECKSUM_PARTIAL)
3774                 return 0;
3775
3776         switch (skb->protocol) {
3777         case cpu_to_be16(ETH_P_IP):
3778                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3779                         cmd_len |= E1000_TXD_CMD_TCP;
3780                 break;
3781         case cpu_to_be16(ETH_P_IPV6):
3782                 /* XXX not handling all IPV6 headers */
3783                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3784                         cmd_len |= E1000_TXD_CMD_TCP;
3785                 break;
3786         default:
3787                 if (unlikely(net_ratelimit()))
3788                         e_warn("checksum_partial proto=%x!\n", skb->protocol);
3789                 break;
3790         }
3791
3792         css = skb_transport_offset(skb);
3793
3794         i = tx_ring->next_to_use;
3795         buffer_info = &tx_ring->buffer_info[i];
3796         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3797
3798         context_desc->lower_setup.ip_config = 0;
3799         context_desc->upper_setup.tcp_fields.tucss = css;
3800         context_desc->upper_setup.tcp_fields.tucso =
3801                                 css + skb->csum_offset;
3802         context_desc->upper_setup.tcp_fields.tucse = 0;
3803         context_desc->tcp_seg_setup.data = 0;
3804         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
3805
3806         buffer_info->time_stamp = jiffies;
3807         buffer_info->next_to_watch = i;
3808
3809         i++;
3810         if (i == tx_ring->count)
3811                 i = 0;
3812         tx_ring->next_to_use = i;
3813
3814         return 1;
3815 }
3816
3817 #define E1000_MAX_PER_TXD       8192
3818 #define E1000_MAX_TXD_PWR       12
3819
3820 static int e1000_tx_map(struct e1000_adapter *adapter,
3821                         struct sk_buff *skb, unsigned int first,
3822                         unsigned int max_per_txd, unsigned int nr_frags,
3823                         unsigned int mss)
3824 {
3825         struct e1000_ring *tx_ring = adapter->tx_ring;
3826         unsigned int len = skb_headlen(skb);
3827         unsigned int offset, size, count = 0, i;
3828         unsigned int f;
3829         dma_addr_t map;
3830
3831         i = tx_ring->next_to_use;
3832
3833         if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
3834                 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3835                 adapter->tx_dma_failed++;
3836                 dev_kfree_skb(skb);
3837                 return -2;
3838         }
3839
3840         map = skb_shinfo(skb)->dma_maps[0];
3841         offset = 0;
3842
3843         while (len) {
3844                 struct e1000_buffer *buffer_info = &tx_ring->buffer_info[i];
3845                 size = min(len, max_per_txd);
3846
3847                 buffer_info->length = size;
3848                 /* set time_stamp *before* dma to help avoid a possible race */
3849                 buffer_info->time_stamp = jiffies;
3850                 buffer_info->dma = map + offset;
3851                 buffer_info->next_to_watch = i;
3852
3853                 len -= size;
3854                 offset += size;
3855                 count++;
3856                 i++;
3857                 if (i == tx_ring->count)
3858                         i = 0;
3859         }
3860
3861         for (f = 0; f < nr_frags; f++) {
3862                 struct skb_frag_struct *frag;
3863
3864                 frag = &skb_shinfo(skb)->frags[f];
3865                 len = frag->size;
3866                 map = skb_shinfo(skb)->dma_maps[f + 1];
3867                 offset = 0;
3868
3869                 while (len) {
3870                         struct e1000_buffer *buffer_info;
3871                         buffer_info = &tx_ring->buffer_info[i];
3872                         size = min(len, max_per_txd);
3873
3874                         buffer_info->length = size;
3875                         buffer_info->time_stamp = jiffies;
3876                         buffer_info->dma = map + offset;
3877                         buffer_info->next_to_watch = i;
3878
3879                         len -= size;
3880                         offset += size;
3881                         count++;
3882
3883                         i++;
3884                         if (i == tx_ring->count)
3885                                 i = 0;
3886                 }
3887         }
3888
3889         if (i == 0)
3890                 i = tx_ring->count - 1;
3891         else
3892                 i--;
3893
3894         tx_ring->buffer_info[i].skb = skb;
3895         tx_ring->buffer_info[first].next_to_watch = i;
3896         smp_wmb();
3897
3898         return count;
3899 }
3900
3901 static void e1000_tx_queue(struct e1000_adapter *adapter,
3902                            int tx_flags, int count)
3903 {
3904         struct e1000_ring *tx_ring = adapter->tx_ring;
3905         struct e1000_tx_desc *tx_desc = NULL;
3906         struct e1000_buffer *buffer_info;
3907         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3908         unsigned int i;
3909
3910         if (tx_flags & E1000_TX_FLAGS_TSO) {
3911                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3912                              E1000_TXD_CMD_TSE;
3913                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3914
3915                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3916                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3917         }
3918
3919         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3920                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3921                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3922         }
3923
3924         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3925                 txd_lower |= E1000_TXD_CMD_VLE;
3926                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3927         }
3928
3929         i = tx_ring->next_to_use;
3930
3931         while (count--) {
3932                 buffer_info = &tx_ring->buffer_info[i];
3933                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3934                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3935                 tx_desc->lower.data =
3936                         cpu_to_le32(txd_lower | buffer_info->length);
3937                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3938
3939                 i++;
3940                 if (i == tx_ring->count)
3941                         i = 0;
3942         }
3943
3944         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3945
3946         /*
3947          * Force memory writes to complete before letting h/w
3948          * know there are new descriptors to fetch.  (Only
3949          * applicable for weak-ordered memory model archs,
3950          * such as IA-64).
3951          */
3952         wmb();
3953
3954         tx_ring->next_to_use = i;
3955         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3956         /*
3957          * we need this if more than one processor can write to our tail
3958          * at a time, it synchronizes IO on IA64/Altix systems
3959          */
3960         mmiowb();
3961 }
3962
3963 #define MINIMUM_DHCP_PACKET_SIZE 282
3964 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3965                                     struct sk_buff *skb)
3966 {
3967         struct e1000_hw *hw =  &adapter->hw;
3968         u16 length, offset;
3969
3970         if (vlan_tx_tag_present(skb)) {
3971                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3972                     && (adapter->hw.mng_cookie.status &
3973                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3974                         return 0;
3975         }
3976
3977         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3978                 return 0;
3979
3980         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3981                 return 0;
3982
3983         {
3984                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3985                 struct udphdr *udp;
3986
3987                 if (ip->protocol != IPPROTO_UDP)
3988                         return 0;
3989
3990                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3991                 if (ntohs(udp->dest) != 67)
3992                         return 0;
3993
3994                 offset = (u8 *)udp + 8 - skb->data;
3995                 length = skb->len - offset;
3996                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
3997         }
3998
3999         return 0;
4000 }
4001
4002 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4003 {
4004         struct e1000_adapter *adapter = netdev_priv(netdev);
4005
4006         netif_stop_queue(netdev);
4007         /*
4008          * Herbert's original patch had:
4009          *  smp_mb__after_netif_stop_queue();
4010          * but since that doesn't exist yet, just open code it.
4011          */
4012         smp_mb();
4013
4014         /*
4015          * We need to check again in a case another CPU has just
4016          * made room available.
4017          */
4018         if (e1000_desc_unused(adapter->tx_ring) < size)
4019                 return -EBUSY;
4020
4021         /* A reprieve! */
4022         netif_start_queue(netdev);
4023         ++adapter->restart_queue;
4024         return 0;
4025 }
4026
4027 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4028 {
4029         struct e1000_adapter *adapter = netdev_priv(netdev);
4030
4031         if (e1000_desc_unused(adapter->tx_ring) >= size)
4032                 return 0;
4033         return __e1000_maybe_stop_tx(netdev, size);
4034 }
4035
4036 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4037 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4038 {
4039         struct e1000_adapter *adapter = netdev_priv(netdev);
4040         struct e1000_ring *tx_ring = adapter->tx_ring;
4041         unsigned int first;
4042         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4043         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4044         unsigned int tx_flags = 0;
4045         unsigned int len = skb->len - skb->data_len;
4046         unsigned int nr_frags;
4047         unsigned int mss;
4048         int count = 0;
4049         int tso;
4050         unsigned int f;
4051
4052         if (test_bit(__E1000_DOWN, &adapter->state)) {
4053                 dev_kfree_skb_any(skb);
4054                 return NETDEV_TX_OK;
4055         }
4056
4057         if (skb->len <= 0) {
4058                 dev_kfree_skb_any(skb);
4059                 return NETDEV_TX_OK;
4060         }
4061
4062         mss = skb_shinfo(skb)->gso_size;
4063         /*
4064          * The controller does a simple calculation to
4065          * make sure there is enough room in the FIFO before
4066          * initiating the DMA for each buffer.  The calc is:
4067          * 4 = ceil(buffer len/mss).  To make sure we don't
4068          * overrun the FIFO, adjust the max buffer len if mss
4069          * drops.
4070          */
4071         if (mss) {
4072                 u8 hdr_len;
4073                 max_per_txd = min(mss << 2, max_per_txd);
4074                 max_txd_pwr = fls(max_per_txd) - 1;
4075
4076                 /*
4077                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4078                  * points to just header, pull a few bytes of payload from
4079                  * frags into skb->data
4080                  */
4081                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4082                 /*
4083                  * we do this workaround for ES2LAN, but it is un-necessary,
4084                  * avoiding it could save a lot of cycles
4085                  */
4086                 if (skb->data_len && (hdr_len == len)) {
4087                         unsigned int pull_size;
4088
4089                         pull_size = min((unsigned int)4, skb->data_len);
4090                         if (!__pskb_pull_tail(skb, pull_size)) {
4091                                 e_err("__pskb_pull_tail failed.\n");
4092                                 dev_kfree_skb_any(skb);
4093                                 return NETDEV_TX_OK;
4094                         }
4095                         len = skb->len - skb->data_len;
4096                 }
4097         }
4098
4099         /* reserve a descriptor for the offload context */
4100         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4101                 count++;
4102         count++;
4103
4104         count += TXD_USE_COUNT(len, max_txd_pwr);
4105
4106         nr_frags = skb_shinfo(skb)->nr_frags;
4107         for (f = 0; f < nr_frags; f++)
4108                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4109                                        max_txd_pwr);
4110
4111         if (adapter->hw.mac.tx_pkt_filtering)
4112                 e1000_transfer_dhcp_info(adapter, skb);
4113
4114         /*
4115          * need: count + 2 desc gap to keep tail from touching
4116          * head, otherwise try next time
4117          */
4118         if (e1000_maybe_stop_tx(netdev, count + 2))
4119                 return NETDEV_TX_BUSY;
4120
4121         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4122                 tx_flags |= E1000_TX_FLAGS_VLAN;
4123                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4124         }
4125
4126         first = tx_ring->next_to_use;
4127
4128         tso = e1000_tso(adapter, skb);
4129         if (tso < 0) {
4130                 dev_kfree_skb_any(skb);
4131                 return NETDEV_TX_OK;
4132         }
4133
4134         if (tso)
4135                 tx_flags |= E1000_TX_FLAGS_TSO;
4136         else if (e1000_tx_csum(adapter, skb))
4137                 tx_flags |= E1000_TX_FLAGS_CSUM;
4138
4139         /*
4140          * Old method was to assume IPv4 packet by default if TSO was enabled.
4141          * 82571 hardware supports TSO capabilities for IPv6 as well...
4142          * no longer assume, we must.
4143          */
4144         if (skb->protocol == htons(ETH_P_IP))
4145                 tx_flags |= E1000_TX_FLAGS_IPV4;
4146
4147         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4148         if (count < 0) {
4149                 /* handle pci_map_single() error in e1000_tx_map */
4150                 dev_kfree_skb_any(skb);
4151                 return NETDEV_TX_OK;
4152         }
4153
4154         e1000_tx_queue(adapter, tx_flags, count);
4155
4156         netdev->trans_start = jiffies;
4157
4158         /* Make sure there is space in the ring for the next send. */
4159         e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4160
4161         return NETDEV_TX_OK;
4162 }
4163
4164 /**
4165  * e1000_tx_timeout - Respond to a Tx Hang
4166  * @netdev: network interface device structure
4167  **/
4168 static void e1000_tx_timeout(struct net_device *netdev)
4169 {
4170         struct e1000_adapter *adapter = netdev_priv(netdev);
4171
4172         /* Do the reset outside of interrupt context */
4173         adapter->tx_timeout_count++;
4174         schedule_work(&adapter->reset_task);
4175 }
4176
4177 static void e1000_reset_task(struct work_struct *work)
4178 {
4179         struct e1000_adapter *adapter;
4180         adapter = container_of(work, struct e1000_adapter, reset_task);
4181
4182         e1000e_reinit_locked(adapter);
4183 }
4184
4185 /**
4186  * e1000_get_stats - Get System Network Statistics
4187  * @netdev: network interface device structure
4188  *
4189  * Returns the address of the device statistics structure.
4190  * The statistics are actually updated from the timer callback.
4191  **/
4192 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4193 {
4194         struct e1000_adapter *adapter = netdev_priv(netdev);
4195
4196         /* only return the current stats */
4197         return &adapter->net_stats;
4198 }
4199
4200 /**
4201  * e1000_change_mtu - Change the Maximum Transfer Unit
4202  * @netdev: network interface device structure
4203  * @new_mtu: new value for maximum frame size
4204  *
4205  * Returns 0 on success, negative on failure
4206  **/
4207 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4208 {
4209         struct e1000_adapter *adapter = netdev_priv(netdev);
4210         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4211
4212         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4213             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
4214                 e_err("Invalid MTU setting\n");
4215                 return -EINVAL;
4216         }
4217
4218         /* Jumbo frame size limits */
4219         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
4220                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4221                         e_err("Jumbo Frames not supported.\n");
4222                         return -EINVAL;
4223                 }
4224                 if (adapter->hw.phy.type == e1000_phy_ife) {
4225                         e_err("Jumbo Frames not supported.\n");
4226                         return -EINVAL;
4227                 }
4228         }
4229
4230 #define MAX_STD_JUMBO_FRAME_SIZE 9234
4231         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
4232                 e_err("MTU > 9216 not supported.\n");
4233                 return -EINVAL;
4234         }
4235
4236         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4237                 msleep(1);
4238         /* e1000e_down has a dependency on max_frame_size */
4239         adapter->max_frame_size = max_frame;
4240         if (netif_running(netdev))
4241                 e1000e_down(adapter);
4242
4243         /*
4244          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4245          * means we reserve 2 more, this pushes us to allocate from the next
4246          * larger slab size.
4247          * i.e. RXBUFFER_2048 --> size-4096 slab
4248          * However with the new *_jumbo_rx* routines, jumbo receives will use
4249          * fragmented skbs
4250          */
4251
4252         if (max_frame <= 256)
4253                 adapter->rx_buffer_len = 256;
4254         else if (max_frame <= 512)
4255                 adapter->rx_buffer_len = 512;
4256         else if (max_frame <= 1024)
4257                 adapter->rx_buffer_len = 1024;
4258         else if (max_frame <= 2048)
4259                 adapter->rx_buffer_len = 2048;
4260         else
4261                 adapter->rx_buffer_len = 4096;
4262
4263         /* adjust allocation if LPE protects us, and we aren't using SBP */
4264         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4265              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4266                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4267                                          + ETH_FCS_LEN;
4268
4269         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4270         netdev->mtu = new_mtu;
4271
4272         if (netif_running(netdev))
4273                 e1000e_up(adapter);
4274         else
4275                 e1000e_reset(adapter);
4276
4277         clear_bit(__E1000_RESETTING, &adapter->state);
4278
4279         return 0;
4280 }
4281
4282 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4283                            int cmd)
4284 {
4285         struct e1000_adapter *adapter = netdev_priv(netdev);
4286         struct mii_ioctl_data *data = if_mii(ifr);
4287
4288         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4289                 return -EOPNOTSUPP;
4290
4291         switch (cmd) {
4292         case SIOCGMIIPHY:
4293                 data->phy_id = adapter->hw.phy.addr;
4294                 break;
4295         case SIOCGMIIREG:
4296                 if (!capable(CAP_NET_ADMIN))
4297                         return -EPERM;
4298                 switch (data->reg_num & 0x1F) {
4299                 case MII_BMCR:
4300                         data->val_out = adapter->phy_regs.bmcr;
4301                         break;
4302                 case MII_BMSR:
4303                         data->val_out = adapter->phy_regs.bmsr;
4304                         break;
4305                 case MII_PHYSID1:
4306                         data->val_out = (adapter->hw.phy.id >> 16);
4307                         break;
4308                 case MII_PHYSID2:
4309                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4310                         break;
4311                 case MII_ADVERTISE:
4312                         data->val_out = adapter->phy_regs.advertise;
4313                         break;
4314                 case MII_LPA:
4315                         data->val_out = adapter->phy_regs.lpa;
4316                         break;
4317                 case MII_EXPANSION:
4318                         data->val_out = adapter->phy_regs.expansion;
4319                         break;
4320                 case MII_CTRL1000:
4321                         data->val_out = adapter->phy_regs.ctrl1000;
4322                         break;
4323                 case MII_STAT1000:
4324                         data->val_out = adapter->phy_regs.stat1000;
4325                         break;
4326                 case MII_ESTATUS:
4327                         data->val_out = adapter->phy_regs.estatus;
4328                         break;
4329                 default:
4330                         return -EIO;
4331                 }
4332                 break;
4333         case SIOCSMIIREG:
4334         default:
4335                 return -EOPNOTSUPP;
4336         }
4337         return 0;
4338 }
4339
4340 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4341 {
4342         switch (cmd) {
4343         case SIOCGMIIPHY:
4344         case SIOCGMIIREG:
4345         case SIOCSMIIREG:
4346                 return e1000_mii_ioctl(netdev, ifr, cmd);
4347         default:
4348                 return -EOPNOTSUPP;
4349         }
4350 }
4351
4352 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4353 {
4354         struct net_device *netdev = pci_get_drvdata(pdev);
4355         struct e1000_adapter *adapter = netdev_priv(netdev);
4356         struct e1000_hw *hw = &adapter->hw;
4357         u32 ctrl, ctrl_ext, rctl, status;
4358         u32 wufc = adapter->wol;
4359         int retval = 0;
4360
4361         netif_device_detach(netdev);
4362
4363         if (netif_running(netdev)) {
4364                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4365                 e1000e_down(adapter);
4366                 e1000_free_irq(adapter);
4367         }
4368         e1000e_reset_interrupt_capability(adapter);
4369
4370         retval = pci_save_state(pdev);
4371         if (retval)
4372                 return retval;
4373
4374         status = er32(STATUS);
4375         if (status & E1000_STATUS_LU)
4376                 wufc &= ~E1000_WUFC_LNKC;
4377
4378         if (wufc) {
4379                 e1000_setup_rctl(adapter);
4380                 e1000_set_multi(netdev);
4381
4382                 /* turn on all-multi mode if wake on multicast is enabled */
4383                 if (wufc & E1000_WUFC_MC) {
4384                         rctl = er32(RCTL);
4385                         rctl |= E1000_RCTL_MPE;
4386                         ew32(RCTL, rctl);
4387                 }
4388
4389                 ctrl = er32(CTRL);
4390                 /* advertise wake from D3Cold */
4391                 #define E1000_CTRL_ADVD3WUC 0x00100000
4392                 /* phy power management enable */
4393                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4394                 ctrl |= E1000_CTRL_ADVD3WUC |
4395                         E1000_CTRL_EN_PHY_PWR_MGMT;
4396                 ew32(CTRL, ctrl);
4397
4398                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4399                     adapter->hw.phy.media_type ==
4400                     e1000_media_type_internal_serdes) {
4401                         /* keep the laser running in D3 */
4402                         ctrl_ext = er32(CTRL_EXT);
4403                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4404                         ew32(CTRL_EXT, ctrl_ext);
4405                 }
4406
4407                 if (adapter->flags & FLAG_IS_ICH)
4408                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4409
4410                 /* Allow time for pending master requests to run */
4411                 e1000e_disable_pcie_master(&adapter->hw);
4412
4413                 ew32(WUC, E1000_WUC_PME_EN);
4414                 ew32(WUFC, wufc);
4415                 pci_enable_wake(pdev, PCI_D3hot, 1);
4416                 pci_enable_wake(pdev, PCI_D3cold, 1);
4417         } else {
4418                 ew32(WUC, 0);
4419                 ew32(WUFC, 0);
4420                 pci_enable_wake(pdev, PCI_D3hot, 0);
4421                 pci_enable_wake(pdev, PCI_D3cold, 0);
4422         }
4423
4424         /* make sure adapter isn't asleep if manageability is enabled */
4425         if (adapter->flags & FLAG_MNG_PT_ENABLED) {
4426                 pci_enable_wake(pdev, PCI_D3hot, 1);
4427                 pci_enable_wake(pdev, PCI_D3cold, 1);
4428         }
4429
4430         if (adapter->hw.phy.type == e1000_phy_igp_3)
4431                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4432
4433         /*
4434          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4435          * would have already happened in close and is redundant.
4436          */
4437         e1000_release_hw_control(adapter);
4438
4439         pci_disable_device(pdev);
4440
4441         /*
4442          * The pci-e switch on some quad port adapters will report a
4443          * correctable error when the MAC transitions from D0 to D3.  To
4444          * prevent this we need to mask off the correctable errors on the
4445          * downstream port of the pci-e switch.
4446          */
4447         if (adapter->flags & FLAG_IS_QUAD_PORT) {
4448                 struct pci_dev *us_dev = pdev->bus->self;
4449                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
4450                 u16 devctl;
4451
4452                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
4453                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
4454                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
4455
4456                 pci_set_power_state(pdev, pci_choose_state(pdev, state));
4457
4458                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
4459         } else {
4460                 pci_set_power_state(pdev, pci_choose_state(pdev, state));
4461         }
4462
4463         return 0;
4464 }
4465
4466 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4467 {
4468         int pos;
4469         u16 val;
4470
4471         /*
4472          * 82573 workaround - disable L1 ASPM on mobile chipsets
4473          *
4474          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4475          * resulting in lost data or garbage information on the pci-e link
4476          * level. This could result in (false) bad EEPROM checksum errors,
4477          * long ping times (up to 2s) or even a system freeze/hang.
4478          *
4479          * Unfortunately this feature saves about 1W power consumption when
4480          * active.
4481          */
4482         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4483         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4484         if (val & 0x2) {
4485                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4486                 val &= ~0x2;
4487                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4488         }
4489 }
4490
4491 #ifdef CONFIG_PM
4492 static int e1000_resume(struct pci_dev *pdev)
4493 {
4494         struct net_device *netdev = pci_get_drvdata(pdev);
4495         struct e1000_adapter *adapter = netdev_priv(netdev);
4496         struct e1000_hw *hw = &adapter->hw;
4497         u32 err;
4498
4499         pci_set_power_state(pdev, PCI_D0);
4500         pci_restore_state(pdev);
4501         e1000e_disable_l1aspm(pdev);
4502
4503         err = pci_enable_device_mem(pdev);
4504         if (err) {
4505                 dev_err(&pdev->dev,
4506                         "Cannot enable PCI device from suspend\n");
4507                 return err;
4508         }
4509
4510         /* AER (Advanced Error Reporting) hooks */
4511         err = pci_enable_pcie_error_reporting(pdev);
4512         if (err) {
4513                 dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
4514                                     "0x%x\n", err);
4515                 /* non-fatal, continue */
4516         }
4517
4518         pci_set_master(pdev);
4519
4520         pci_enable_wake(pdev, PCI_D3hot, 0);
4521         pci_enable_wake(pdev, PCI_D3cold, 0);
4522
4523         e1000e_set_interrupt_capability(adapter);
4524         if (netif_running(netdev)) {
4525                 err = e1000_request_irq(adapter);
4526                 if (err)
4527                         return err;
4528         }
4529
4530         e1000e_power_up_phy(adapter);
4531         e1000e_reset(adapter);
4532         ew32(WUS, ~0);
4533
4534         e1000_init_manageability(adapter);
4535
4536         if (netif_running(netdev))
4537                 e1000e_up(adapter);
4538
4539         netif_device_attach(netdev);
4540
4541         /*
4542          * If the controller has AMT, do not set DRV_LOAD until the interface
4543          * is up.  For all other cases, let the f/w know that the h/w is now
4544          * under the control of the driver.
4545          */
4546         if (!(adapter->flags & FLAG_HAS_AMT))
4547                 e1000_get_hw_control(adapter);
4548
4549         return 0;
4550 }
4551 #endif
4552
4553 static void e1000_shutdown(struct pci_dev *pdev)
4554 {
4555         e1000_suspend(pdev, PMSG_SUSPEND);
4556 }
4557
4558 #ifdef CONFIG_NET_POLL_CONTROLLER
4559 /*
4560  * Polling 'interrupt' - used by things like netconsole to send skbs
4561  * without having to re-enable interrupts. It's not called while
4562  * the interrupt routine is executing.
4563  */
4564 static void e1000_netpoll(struct net_device *netdev)
4565 {
4566         struct e1000_adapter *adapter = netdev_priv(netdev);
4567
4568         disable_irq(adapter->pdev->irq);
4569         e1000_intr(adapter->pdev->irq, netdev);
4570
4571         enable_irq(adapter->pdev->irq);
4572 }
4573 #endif
4574
4575 /**
4576  * e1000_io_error_detected - called when PCI error is detected
4577  * @pdev: Pointer to PCI device
4578  * @state: The current pci connection state
4579  *
4580  * This function is called after a PCI bus error affecting
4581  * this device has been detected.
4582  */
4583 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4584                                                 pci_channel_state_t state)
4585 {
4586         struct net_device *netdev = pci_get_drvdata(pdev);
4587         struct e1000_adapter *adapter = netdev_priv(netdev);
4588
4589         netif_device_detach(netdev);
4590
4591         if (netif_running(netdev))
4592                 e1000e_down(adapter);
4593         pci_disable_device(pdev);
4594
4595         /* Request a slot slot reset. */
4596         return PCI_ERS_RESULT_NEED_RESET;
4597 }
4598
4599 /**
4600  * e1000_io_slot_reset - called after the pci bus has been reset.
4601  * @pdev: Pointer to PCI device
4602  *
4603  * Restart the card from scratch, as if from a cold-boot. Implementation
4604  * resembles the first-half of the e1000_resume routine.
4605  */
4606 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4607 {
4608         struct net_device *netdev = pci_get_drvdata(pdev);
4609         struct e1000_adapter *adapter = netdev_priv(netdev);
4610         struct e1000_hw *hw = &adapter->hw;
4611         int err;
4612         pci_ers_result_t result;
4613
4614         e1000e_disable_l1aspm(pdev);
4615         err = pci_enable_device_mem(pdev);
4616         if (err) {
4617                 dev_err(&pdev->dev,
4618                         "Cannot re-enable PCI device after reset.\n");
4619                 result = PCI_ERS_RESULT_DISCONNECT;
4620         } else {
4621                 pci_set_master(pdev);
4622                 pci_restore_state(pdev);
4623
4624                 pci_enable_wake(pdev, PCI_D3hot, 0);
4625                 pci_enable_wake(pdev, PCI_D3cold, 0);
4626
4627                 e1000e_reset(adapter);
4628                 ew32(WUS, ~0);
4629                 result = PCI_ERS_RESULT_RECOVERED;
4630         }
4631
4632         pci_cleanup_aer_uncorrect_error_status(pdev);
4633
4634         return result;
4635 }
4636
4637 /**
4638  * e1000_io_resume - called when traffic can start flowing again.
4639  * @pdev: Pointer to PCI device
4640  *
4641  * This callback is called when the error recovery driver tells us that
4642  * its OK to resume normal operation. Implementation resembles the
4643  * second-half of the e1000_resume routine.
4644  */
4645 static void e1000_io_resume(struct pci_dev *pdev)
4646 {
4647         struct net_device *netdev = pci_get_drvdata(pdev);
4648         struct e1000_adapter *adapter = netdev_priv(netdev);
4649
4650         e1000_init_manageability(adapter);
4651
4652         if (netif_running(netdev)) {
4653                 if (e1000e_up(adapter)) {
4654                         dev_err(&pdev->dev,
4655                                 "can't bring device back up after reset\n");
4656                         return;
4657                 }
4658         }
4659
4660         netif_device_attach(netdev);
4661
4662         /*
4663          * If the controller has AMT, do not set DRV_LOAD until the interface
4664          * is up.  For all other cases, let the f/w know that the h/w is now
4665          * under the control of the driver.
4666          */
4667         if (!(adapter->flags & FLAG_HAS_AMT))
4668                 e1000_get_hw_control(adapter);
4669
4670 }
4671
4672 static void e1000_print_device_info(struct e1000_adapter *adapter)
4673 {
4674         struct e1000_hw *hw = &adapter->hw;
4675         struct net_device *netdev = adapter->netdev;
4676         u32 pba_num;
4677
4678         /* print bus type/speed/width info */
4679         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
4680                /* bus width */
4681                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4682                 "Width x1"),
4683                /* MAC address */
4684                netdev->dev_addr);
4685         e_info("Intel(R) PRO/%s Network Connection\n",
4686                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4687         e1000e_read_pba_num(hw, &pba_num);
4688         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4689                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4690 }
4691
4692 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4693 {
4694         struct e1000_hw *hw = &adapter->hw;
4695         int ret_val;
4696         u16 buf = 0;
4697
4698         if (hw->mac.type != e1000_82573)
4699                 return;
4700
4701         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4702         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
4703                 /* Deep Smart Power Down (DSPD) */
4704                 dev_warn(&adapter->pdev->dev,
4705                          "Warning: detected DSPD enabled in EEPROM\n");
4706         }
4707
4708         ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4709         if (!ret_val && (le16_to_cpu(buf) & (3 << 2))) {
4710                 /* ASPM enable */
4711                 dev_warn(&adapter->pdev->dev,
4712                          "Warning: detected ASPM enabled in EEPROM\n");
4713         }
4714 }
4715
4716 static const struct net_device_ops e1000e_netdev_ops = {
4717         .ndo_open               = e1000_open,
4718         .ndo_stop               = e1000_close,
4719         .ndo_start_xmit         = e1000_xmit_frame,
4720         .ndo_get_stats          = e1000_get_stats,
4721         .ndo_set_multicast_list = e1000_set_multi,
4722         .ndo_set_mac_address    = e1000_set_mac,
4723         .ndo_change_mtu         = e1000_change_mtu,
4724         .ndo_do_ioctl           = e1000_ioctl,
4725         .ndo_tx_timeout         = e1000_tx_timeout,
4726         .ndo_validate_addr      = eth_validate_addr,
4727
4728         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
4729         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
4730         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
4731 #ifdef CONFIG_NET_POLL_CONTROLLER
4732         .ndo_poll_controller    = e1000_netpoll,
4733 #endif
4734 };
4735
4736 /**
4737  * e1000_probe - Device Initialization Routine
4738  * @pdev: PCI device information struct
4739  * @ent: entry in e1000_pci_tbl
4740  *
4741  * Returns 0 on success, negative on failure
4742  *
4743  * e1000_probe initializes an adapter identified by a pci_dev structure.
4744  * The OS initialization, configuring of the adapter private structure,
4745  * and a hardware reset occur.
4746  **/
4747 static int __devinit e1000_probe(struct pci_dev *pdev,
4748                                  const struct pci_device_id *ent)
4749 {
4750         struct net_device *netdev;
4751         struct e1000_adapter *adapter;
4752         struct e1000_hw *hw;
4753         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4754         resource_size_t mmio_start, mmio_len;
4755         resource_size_t flash_start, flash_len;
4756
4757         static int cards_found;
4758         int i, err, pci_using_dac;
4759         u16 eeprom_data = 0;
4760         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4761
4762         e1000e_disable_l1aspm(pdev);
4763
4764         err = pci_enable_device_mem(pdev);
4765         if (err)
4766                 return err;
4767
4768         pci_using_dac = 0;
4769         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
4770         if (!err) {
4771                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
4772                 if (!err)
4773                         pci_using_dac = 1;
4774         } else {
4775                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
4776                 if (err) {
4777                         err = pci_set_consistent_dma_mask(pdev,
4778                                                           DMA_32BIT_MASK);
4779                         if (err) {
4780                                 dev_err(&pdev->dev, "No usable DMA "
4781                                         "configuration, aborting\n");
4782                                 goto err_dma;
4783                         }
4784                 }
4785         }
4786
4787         err = pci_request_selected_regions_exclusive(pdev,
4788                                           pci_select_bars(pdev, IORESOURCE_MEM),
4789                                           e1000e_driver_name);
4790         if (err)
4791                 goto err_pci_reg;
4792
4793         pci_set_master(pdev);
4794         /* PCI config space info */
4795         err = pci_save_state(pdev);
4796         if (err)
4797                 goto err_alloc_etherdev;
4798
4799         err = -ENOMEM;
4800         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4801         if (!netdev)
4802                 goto err_alloc_etherdev;
4803
4804         SET_NETDEV_DEV(netdev, &pdev->dev);
4805
4806         pci_set_drvdata(pdev, netdev);
4807         adapter = netdev_priv(netdev);
4808         hw = &adapter->hw;
4809         adapter->netdev = netdev;
4810         adapter->pdev = pdev;
4811         adapter->ei = ei;
4812         adapter->pba = ei->pba;
4813         adapter->flags = ei->flags;
4814         adapter->flags2 = ei->flags2;
4815         adapter->hw.adapter = adapter;
4816         adapter->hw.mac.type = ei->mac;
4817         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4818
4819         mmio_start = pci_resource_start(pdev, 0);
4820         mmio_len = pci_resource_len(pdev, 0);
4821
4822         err = -EIO;
4823         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4824         if (!adapter->hw.hw_addr)
4825                 goto err_ioremap;
4826
4827         if ((adapter->flags & FLAG_HAS_FLASH) &&
4828             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
4829                 flash_start = pci_resource_start(pdev, 1);
4830                 flash_len = pci_resource_len(pdev, 1);
4831                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
4832                 if (!adapter->hw.flash_address)
4833                         goto err_flashmap;
4834         }
4835
4836         /* construct the net_device struct */
4837         netdev->netdev_ops              = &e1000e_netdev_ops;
4838         e1000e_set_ethtool_ops(netdev);
4839         netdev->watchdog_timeo          = 5 * HZ;
4840         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
4841         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
4842
4843         netdev->mem_start = mmio_start;
4844         netdev->mem_end = mmio_start + mmio_len;
4845
4846         adapter->bd_number = cards_found++;
4847
4848         e1000e_check_options(adapter);
4849
4850         /* setup adapter struct */
4851         err = e1000_sw_init(adapter);
4852         if (err)
4853                 goto err_sw_init;
4854
4855         err = -EIO;
4856
4857         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4858         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
4859         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4860
4861         err = ei->get_variants(adapter);
4862         if (err)
4863                 goto err_hw_init;
4864
4865         if ((adapter->flags & FLAG_IS_ICH) &&
4866             (adapter->flags & FLAG_READ_ONLY_NVM))
4867                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
4868
4869         hw->mac.ops.get_bus_info(&adapter->hw);
4870
4871         adapter->hw.phy.autoneg_wait_to_complete = 0;
4872
4873         /* Copper options */
4874         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
4875                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
4876                 adapter->hw.phy.disable_polarity_correction = 0;
4877                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
4878         }
4879
4880         if (e1000_check_reset_block(&adapter->hw))
4881                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
4882
4883         netdev->features = NETIF_F_SG |
4884                            NETIF_F_HW_CSUM |
4885                            NETIF_F_HW_VLAN_TX |
4886                            NETIF_F_HW_VLAN_RX;
4887
4888         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
4889                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
4890
4891         netdev->features |= NETIF_F_TSO;
4892         netdev->features |= NETIF_F_TSO6;
4893
4894         netdev->vlan_features |= NETIF_F_TSO;
4895         netdev->vlan_features |= NETIF_F_TSO6;
4896         netdev->vlan_features |= NETIF_F_HW_CSUM;
4897         netdev->vlan_features |= NETIF_F_SG;
4898
4899         if (pci_using_dac)
4900                 netdev->features |= NETIF_F_HIGHDMA;
4901
4902         if (e1000e_enable_mng_pass_thru(&adapter->hw))
4903                 adapter->flags |= FLAG_MNG_PT_ENABLED;
4904
4905         /*
4906          * before reading the NVM, reset the controller to
4907          * put the device in a known good starting state
4908          */
4909         adapter->hw.mac.ops.reset_hw(&adapter->hw);
4910
4911         /*
4912          * systems with ASPM and others may see the checksum fail on the first
4913          * attempt. Let's give it a few tries
4914          */
4915         for (i = 0;; i++) {
4916                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
4917                         break;
4918                 if (i == 2) {
4919                         e_err("The NVM Checksum Is Not Valid\n");
4920                         err = -EIO;
4921                         goto err_eeprom;
4922                 }
4923         }
4924
4925         e1000_eeprom_checks(adapter);
4926
4927         /* copy the MAC address out of the NVM */
4928         if (e1000e_read_mac_addr(&adapter->hw))
4929                 e_err("NVM Read Error while reading MAC address\n");
4930
4931         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
4932         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
4933
4934         if (!is_valid_ether_addr(netdev->perm_addr)) {
4935                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
4936                 err = -EIO;
4937                 goto err_eeprom;
4938         }
4939
4940         init_timer(&adapter->watchdog_timer);
4941         adapter->watchdog_timer.function = &e1000_watchdog;
4942         adapter->watchdog_timer.data = (unsigned long) adapter;
4943
4944         init_timer(&adapter->phy_info_timer);
4945         adapter->phy_info_timer.function = &e1000_update_phy_info;
4946         adapter->phy_info_timer.data = (unsigned long) adapter;
4947
4948         INIT_WORK(&adapter->reset_task, e1000_reset_task);
4949         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
4950         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
4951         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
4952
4953         /* Initialize link parameters. User can change them with ethtool */
4954         adapter->hw.mac.autoneg = 1;
4955         adapter->fc_autoneg = 1;
4956         adapter->hw.fc.requested_mode = e1000_fc_default;
4957         adapter->hw.fc.current_mode = e1000_fc_default;
4958         adapter->hw.phy.autoneg_advertised = 0x2f;
4959
4960         /* ring size defaults */
4961         adapter->rx_ring->count = 256;
4962         adapter->tx_ring->count = 256;
4963
4964         /*
4965          * Initial Wake on LAN setting - If APM wake is enabled in
4966          * the EEPROM, enable the ACPI Magic Packet filter
4967          */
4968         if (adapter->flags & FLAG_APME_IN_WUC) {
4969                 /* APME bit in EEPROM is mapped to WUC.APME */
4970                 eeprom_data = er32(WUC);
4971                 eeprom_apme_mask = E1000_WUC_APME;
4972         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
4973                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
4974                     (adapter->hw.bus.func == 1))
4975                         e1000_read_nvm(&adapter->hw,
4976                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
4977                 else
4978                         e1000_read_nvm(&adapter->hw,
4979                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
4980         }
4981
4982         /* fetch WoL from EEPROM */
4983         if (eeprom_data & eeprom_apme_mask)
4984                 adapter->eeprom_wol |= E1000_WUFC_MAG;
4985
4986         /*
4987          * now that we have the eeprom settings, apply the special cases
4988          * where the eeprom may be wrong or the board simply won't support
4989          * wake on lan on a particular port
4990          */
4991         if (!(adapter->flags & FLAG_HAS_WOL))
4992                 adapter->eeprom_wol = 0;
4993
4994         /* initialize the wol settings based on the eeprom settings */
4995         adapter->wol = adapter->eeprom_wol;
4996         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
4997
4998         /* save off EEPROM version number */
4999         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5000
5001         /* reset the hardware with the new settings */
5002         e1000e_reset(adapter);
5003
5004         /*
5005          * If the controller has AMT, do not set DRV_LOAD until the interface
5006          * is up.  For all other cases, let the f/w know that the h/w is now
5007          * under the control of the driver.
5008          */
5009         if (!(adapter->flags & FLAG_HAS_AMT))
5010                 e1000_get_hw_control(adapter);
5011
5012         /* tell the stack to leave us alone until e1000_open() is called */
5013         netif_carrier_off(netdev);
5014         netif_tx_stop_all_queues(netdev);
5015
5016         strcpy(netdev->name, "eth%d");
5017         err = register_netdev(netdev);
5018         if (err)
5019                 goto err_register;
5020
5021         e1000_print_device_info(adapter);
5022
5023         return 0;
5024
5025 err_register:
5026         if (!(adapter->flags & FLAG_HAS_AMT))
5027                 e1000_release_hw_control(adapter);
5028 err_eeprom:
5029         if (!e1000_check_reset_block(&adapter->hw))
5030                 e1000_phy_hw_reset(&adapter->hw);
5031 err_hw_init:
5032
5033         kfree(adapter->tx_ring);
5034         kfree(adapter->rx_ring);
5035 err_sw_init:
5036         if (adapter->hw.flash_address)
5037                 iounmap(adapter->hw.flash_address);
5038         e1000e_reset_interrupt_capability(adapter);
5039 err_flashmap:
5040         iounmap(adapter->hw.hw_addr);
5041 err_ioremap:
5042         free_netdev(netdev);
5043 err_alloc_etherdev:
5044         pci_release_selected_regions(pdev,
5045                                      pci_select_bars(pdev, IORESOURCE_MEM));
5046 err_pci_reg:
5047 err_dma:
5048         pci_disable_device(pdev);
5049         return err;
5050 }
5051
5052 /**
5053  * e1000_remove - Device Removal Routine
5054  * @pdev: PCI device information struct
5055  *
5056  * e1000_remove is called by the PCI subsystem to alert the driver
5057  * that it should release a PCI device.  The could be caused by a
5058  * Hot-Plug event, or because the driver is going to be removed from
5059  * memory.
5060  **/
5061 static void __devexit e1000_remove(struct pci_dev *pdev)
5062 {
5063         struct net_device *netdev = pci_get_drvdata(pdev);
5064         struct e1000_adapter *adapter = netdev_priv(netdev);
5065         int err;
5066
5067         /*
5068          * flush_scheduled work may reschedule our watchdog task, so
5069          * explicitly disable watchdog tasks from being rescheduled
5070          */
5071         set_bit(__E1000_DOWN, &adapter->state);
5072         del_timer_sync(&adapter->watchdog_timer);
5073         del_timer_sync(&adapter->phy_info_timer);
5074
5075         flush_scheduled_work();
5076
5077         /*
5078          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5079          * would have already happened in close and is redundant.
5080          */
5081         e1000_release_hw_control(adapter);
5082
5083         unregister_netdev(netdev);
5084
5085         if (!e1000_check_reset_block(&adapter->hw))
5086                 e1000_phy_hw_reset(&adapter->hw);
5087
5088         e1000e_reset_interrupt_capability(adapter);
5089         kfree(adapter->tx_ring);
5090         kfree(adapter->rx_ring);
5091
5092         iounmap(adapter->hw.hw_addr);
5093         if (adapter->hw.flash_address)
5094                 iounmap(adapter->hw.flash_address);
5095         pci_release_selected_regions(pdev,
5096                                      pci_select_bars(pdev, IORESOURCE_MEM));
5097
5098         free_netdev(netdev);
5099
5100         /* AER disable */
5101         err = pci_disable_pcie_error_reporting(pdev);
5102         if (err)
5103                 dev_err(&pdev->dev,
5104                         "pci_disable_pcie_error_reporting failed 0x%x\n", err);
5105
5106         pci_disable_device(pdev);
5107 }
5108
5109 /* PCI Error Recovery (ERS) */
5110 static struct pci_error_handlers e1000_err_handler = {
5111         .error_detected = e1000_io_error_detected,
5112         .slot_reset = e1000_io_slot_reset,
5113         .resume = e1000_io_resume,
5114 };
5115
5116 static struct pci_device_id e1000_pci_tbl[] = {
5117         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5118         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5119         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5120         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5121         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5122         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5123         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5124         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5125         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5126
5127         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5128         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5129         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5130         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5131
5132         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5133         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5134         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5135
5136         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5137
5138         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5139           board_80003es2lan },
5140         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5141           board_80003es2lan },
5142         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5143           board_80003es2lan },
5144         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5145           board_80003es2lan },
5146
5147         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5148         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5149         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5150         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5151         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5152         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5153         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5154
5155         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5156         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5157         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5158         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5159         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5160         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5161         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5162         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5163         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5164
5165         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5166         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5167         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5168
5169         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5170         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5171
5172         { }     /* terminate list */
5173 };
5174 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5175
5176 /* PCI Device API Driver */
5177 static struct pci_driver e1000_driver = {
5178         .name     = e1000e_driver_name,
5179         .id_table = e1000_pci_tbl,
5180         .probe    = e1000_probe,
5181         .remove   = __devexit_p(e1000_remove),
5182 #ifdef CONFIG_PM
5183         /* Power Management Hooks */
5184         .suspend  = e1000_suspend,
5185         .resume   = e1000_resume,
5186 #endif
5187         .shutdown = e1000_shutdown,
5188         .err_handler = &e1000_err_handler
5189 };
5190
5191 /**
5192  * e1000_init_module - Driver Registration Routine
5193  *
5194  * e1000_init_module is the first routine called when the driver is
5195  * loaded. All it does is register with the PCI subsystem.
5196  **/
5197 static int __init e1000_init_module(void)
5198 {
5199         int ret;
5200         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
5201                e1000e_driver_name, e1000e_driver_version);
5202         printk(KERN_INFO "%s: Copyright (c) 1999-2008 Intel Corporation.\n",
5203                e1000e_driver_name);
5204         ret = pci_register_driver(&e1000_driver);
5205         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name,
5206                                PM_QOS_DEFAULT_VALUE);
5207                                 
5208         return ret;
5209 }
5210 module_init(e1000_init_module);
5211
5212 /**
5213  * e1000_exit_module - Driver Exit Cleanup Routine
5214  *
5215  * e1000_exit_module is called just before the driver is removed
5216  * from memory.
5217  **/
5218 static void __exit e1000_exit_module(void)
5219 {
5220         pci_unregister_driver(&e1000_driver);
5221         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name);
5222 }
5223 module_exit(e1000_exit_module);
5224
5225
5226 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5227 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5228 MODULE_LICENSE("GPL");
5229 MODULE_VERSION(DRV_VERSION);
5230
5231 /* e1000_main.c */