d092eafde9bf042a1ea5346f6815625e9ff50eba
[safe/jmp/linux-2.6] / drivers / net / e1000e / netdev.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2008 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #include <linux/module.h>
30 #include <linux/types.h>
31 #include <linux/init.h>
32 #include <linux/pci.h>
33 #include <linux/vmalloc.h>
34 #include <linux/pagemap.h>
35 #include <linux/delay.h>
36 #include <linux/netdevice.h>
37 #include <linux/tcp.h>
38 #include <linux/ipv6.h>
39 #include <net/checksum.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/mii.h>
42 #include <linux/ethtool.h>
43 #include <linux/if_vlan.h>
44 #include <linux/cpu.h>
45 #include <linux/smp.h>
46 #include <linux/pm_qos_params.h>
47 #include <linux/aer.h>
48
49 #include "e1000.h"
50
51 #define DRV_VERSION "0.3.3.4-k2"
52 char e1000e_driver_name[] = "e1000e";
53 const char e1000e_driver_version[] = DRV_VERSION;
54
55 static const struct e1000_info *e1000_info_tbl[] = {
56         [board_82571]           = &e1000_82571_info,
57         [board_82572]           = &e1000_82572_info,
58         [board_82573]           = &e1000_82573_info,
59         [board_82574]           = &e1000_82574_info,
60         [board_82583]           = &e1000_82583_info,
61         [board_80003es2lan]     = &e1000_es2_info,
62         [board_ich8lan]         = &e1000_ich8_info,
63         [board_ich9lan]         = &e1000_ich9_info,
64         [board_ich10lan]        = &e1000_ich10_info,
65 };
66
67 #ifdef DEBUG
68 /**
69  * e1000_get_hw_dev_name - return device name string
70  * used by hardware layer to print debugging information
71  **/
72 char *e1000e_get_hw_dev_name(struct e1000_hw *hw)
73 {
74         return hw->adapter->netdev->name;
75 }
76 #endif
77
78 /**
79  * e1000_desc_unused - calculate if we have unused descriptors
80  **/
81 static int e1000_desc_unused(struct e1000_ring *ring)
82 {
83         if (ring->next_to_clean > ring->next_to_use)
84                 return ring->next_to_clean - ring->next_to_use - 1;
85
86         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
87 }
88
89 /**
90  * e1000_receive_skb - helper function to handle Rx indications
91  * @adapter: board private structure
92  * @status: descriptor status field as written by hardware
93  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
94  * @skb: pointer to sk_buff to be indicated to stack
95  **/
96 static void e1000_receive_skb(struct e1000_adapter *adapter,
97                               struct net_device *netdev,
98                               struct sk_buff *skb,
99                               u8 status, __le16 vlan)
100 {
101         skb->protocol = eth_type_trans(skb, netdev);
102
103         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
104                 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
105                                  le16_to_cpu(vlan), skb);
106         else
107                 napi_gro_receive(&adapter->napi, skb);
108 }
109
110 /**
111  * e1000_rx_checksum - Receive Checksum Offload for 82543
112  * @adapter:     board private structure
113  * @status_err:  receive descriptor status and error fields
114  * @csum:       receive descriptor csum field
115  * @sk_buff:     socket buffer with received data
116  **/
117 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
118                               u32 csum, struct sk_buff *skb)
119 {
120         u16 status = (u16)status_err;
121         u8 errors = (u8)(status_err >> 24);
122         skb->ip_summed = CHECKSUM_NONE;
123
124         /* Ignore Checksum bit is set */
125         if (status & E1000_RXD_STAT_IXSM)
126                 return;
127         /* TCP/UDP checksum error bit is set */
128         if (errors & E1000_RXD_ERR_TCPE) {
129                 /* let the stack verify checksum errors */
130                 adapter->hw_csum_err++;
131                 return;
132         }
133
134         /* TCP/UDP Checksum has not been calculated */
135         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
136                 return;
137
138         /* It must be a TCP or UDP packet with a valid checksum */
139         if (status & E1000_RXD_STAT_TCPCS) {
140                 /* TCP checksum is good */
141                 skb->ip_summed = CHECKSUM_UNNECESSARY;
142         } else {
143                 /*
144                  * IP fragment with UDP payload
145                  * Hardware complements the payload checksum, so we undo it
146                  * and then put the value in host order for further stack use.
147                  */
148                 __sum16 sum = (__force __sum16)htons(csum);
149                 skb->csum = csum_unfold(~sum);
150                 skb->ip_summed = CHECKSUM_COMPLETE;
151         }
152         adapter->hw_csum_good++;
153 }
154
155 /**
156  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
157  * @adapter: address of board private structure
158  **/
159 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
160                                    int cleaned_count)
161 {
162         struct net_device *netdev = adapter->netdev;
163         struct pci_dev *pdev = adapter->pdev;
164         struct e1000_ring *rx_ring = adapter->rx_ring;
165         struct e1000_rx_desc *rx_desc;
166         struct e1000_buffer *buffer_info;
167         struct sk_buff *skb;
168         unsigned int i;
169         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
170
171         i = rx_ring->next_to_use;
172         buffer_info = &rx_ring->buffer_info[i];
173
174         while (cleaned_count--) {
175                 skb = buffer_info->skb;
176                 if (skb) {
177                         skb_trim(skb, 0);
178                         goto map_skb;
179                 }
180
181                 skb = netdev_alloc_skb(netdev, bufsz);
182                 if (!skb) {
183                         /* Better luck next round */
184                         adapter->alloc_rx_buff_failed++;
185                         break;
186                 }
187
188                 /*
189                  * Make buffer alignment 2 beyond a 16 byte boundary
190                  * this will result in a 16 byte aligned IP header after
191                  * the 14 byte MAC header is removed
192                  */
193                 skb_reserve(skb, NET_IP_ALIGN);
194
195                 buffer_info->skb = skb;
196 map_skb:
197                 buffer_info->dma = pci_map_single(pdev, skb->data,
198                                                   adapter->rx_buffer_len,
199                                                   PCI_DMA_FROMDEVICE);
200                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
201                         dev_err(&pdev->dev, "RX DMA map failed\n");
202                         adapter->rx_dma_failed++;
203                         break;
204                 }
205
206                 rx_desc = E1000_RX_DESC(*rx_ring, i);
207                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
208
209                 i++;
210                 if (i == rx_ring->count)
211                         i = 0;
212                 buffer_info = &rx_ring->buffer_info[i];
213         }
214
215         if (rx_ring->next_to_use != i) {
216                 rx_ring->next_to_use = i;
217                 if (i-- == 0)
218                         i = (rx_ring->count - 1);
219
220                 /*
221                  * Force memory writes to complete before letting h/w
222                  * know there are new descriptors to fetch.  (Only
223                  * applicable for weak-ordered memory model archs,
224                  * such as IA-64).
225                  */
226                 wmb();
227                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
228         }
229 }
230
231 /**
232  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
233  * @adapter: address of board private structure
234  **/
235 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
236                                       int cleaned_count)
237 {
238         struct net_device *netdev = adapter->netdev;
239         struct pci_dev *pdev = adapter->pdev;
240         union e1000_rx_desc_packet_split *rx_desc;
241         struct e1000_ring *rx_ring = adapter->rx_ring;
242         struct e1000_buffer *buffer_info;
243         struct e1000_ps_page *ps_page;
244         struct sk_buff *skb;
245         unsigned int i, j;
246
247         i = rx_ring->next_to_use;
248         buffer_info = &rx_ring->buffer_info[i];
249
250         while (cleaned_count--) {
251                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
252
253                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
254                         ps_page = &buffer_info->ps_pages[j];
255                         if (j >= adapter->rx_ps_pages) {
256                                 /* all unused desc entries get hw null ptr */
257                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
258                                 continue;
259                         }
260                         if (!ps_page->page) {
261                                 ps_page->page = alloc_page(GFP_ATOMIC);
262                                 if (!ps_page->page) {
263                                         adapter->alloc_rx_buff_failed++;
264                                         goto no_buffers;
265                                 }
266                                 ps_page->dma = pci_map_page(pdev,
267                                                    ps_page->page,
268                                                    0, PAGE_SIZE,
269                                                    PCI_DMA_FROMDEVICE);
270                                 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
271                                         dev_err(&adapter->pdev->dev,
272                                           "RX DMA page map failed\n");
273                                         adapter->rx_dma_failed++;
274                                         goto no_buffers;
275                                 }
276                         }
277                         /*
278                          * Refresh the desc even if buffer_addrs
279                          * didn't change because each write-back
280                          * erases this info.
281                          */
282                         rx_desc->read.buffer_addr[j+1] =
283                              cpu_to_le64(ps_page->dma);
284                 }
285
286                 skb = netdev_alloc_skb(netdev,
287                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
288
289                 if (!skb) {
290                         adapter->alloc_rx_buff_failed++;
291                         break;
292                 }
293
294                 /*
295                  * Make buffer alignment 2 beyond a 16 byte boundary
296                  * this will result in a 16 byte aligned IP header after
297                  * the 14 byte MAC header is removed
298                  */
299                 skb_reserve(skb, NET_IP_ALIGN);
300
301                 buffer_info->skb = skb;
302                 buffer_info->dma = pci_map_single(pdev, skb->data,
303                                                   adapter->rx_ps_bsize0,
304                                                   PCI_DMA_FROMDEVICE);
305                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
306                         dev_err(&pdev->dev, "RX DMA map failed\n");
307                         adapter->rx_dma_failed++;
308                         /* cleanup skb */
309                         dev_kfree_skb_any(skb);
310                         buffer_info->skb = NULL;
311                         break;
312                 }
313
314                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
315
316                 i++;
317                 if (i == rx_ring->count)
318                         i = 0;
319                 buffer_info = &rx_ring->buffer_info[i];
320         }
321
322 no_buffers:
323         if (rx_ring->next_to_use != i) {
324                 rx_ring->next_to_use = i;
325
326                 if (!(i--))
327                         i = (rx_ring->count - 1);
328
329                 /*
330                  * Force memory writes to complete before letting h/w
331                  * know there are new descriptors to fetch.  (Only
332                  * applicable for weak-ordered memory model archs,
333                  * such as IA-64).
334                  */
335                 wmb();
336                 /*
337                  * Hardware increments by 16 bytes, but packet split
338                  * descriptors are 32 bytes...so we increment tail
339                  * twice as much.
340                  */
341                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
342         }
343 }
344
345 /**
346  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
347  * @adapter: address of board private structure
348  * @cleaned_count: number of buffers to allocate this pass
349  **/
350
351 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
352                                          int cleaned_count)
353 {
354         struct net_device *netdev = adapter->netdev;
355         struct pci_dev *pdev = adapter->pdev;
356         struct e1000_rx_desc *rx_desc;
357         struct e1000_ring *rx_ring = adapter->rx_ring;
358         struct e1000_buffer *buffer_info;
359         struct sk_buff *skb;
360         unsigned int i;
361         unsigned int bufsz = 256 -
362                              16 /* for skb_reserve */ -
363                              NET_IP_ALIGN;
364
365         i = rx_ring->next_to_use;
366         buffer_info = &rx_ring->buffer_info[i];
367
368         while (cleaned_count--) {
369                 skb = buffer_info->skb;
370                 if (skb) {
371                         skb_trim(skb, 0);
372                         goto check_page;
373                 }
374
375                 skb = netdev_alloc_skb(netdev, bufsz);
376                 if (unlikely(!skb)) {
377                         /* Better luck next round */
378                         adapter->alloc_rx_buff_failed++;
379                         break;
380                 }
381
382                 /* Make buffer alignment 2 beyond a 16 byte boundary
383                  * this will result in a 16 byte aligned IP header after
384                  * the 14 byte MAC header is removed
385                  */
386                 skb_reserve(skb, NET_IP_ALIGN);
387
388                 buffer_info->skb = skb;
389 check_page:
390                 /* allocate a new page if necessary */
391                 if (!buffer_info->page) {
392                         buffer_info->page = alloc_page(GFP_ATOMIC);
393                         if (unlikely(!buffer_info->page)) {
394                                 adapter->alloc_rx_buff_failed++;
395                                 break;
396                         }
397                 }
398
399                 if (!buffer_info->dma)
400                         buffer_info->dma = pci_map_page(pdev,
401                                                         buffer_info->page, 0,
402                                                         PAGE_SIZE,
403                                                         PCI_DMA_FROMDEVICE);
404
405                 rx_desc = E1000_RX_DESC(*rx_ring, i);
406                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
407
408                 if (unlikely(++i == rx_ring->count))
409                         i = 0;
410                 buffer_info = &rx_ring->buffer_info[i];
411         }
412
413         if (likely(rx_ring->next_to_use != i)) {
414                 rx_ring->next_to_use = i;
415                 if (unlikely(i-- == 0))
416                         i = (rx_ring->count - 1);
417
418                 /* Force memory writes to complete before letting h/w
419                  * know there are new descriptors to fetch.  (Only
420                  * applicable for weak-ordered memory model archs,
421                  * such as IA-64). */
422                 wmb();
423                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
424         }
425 }
426
427 /**
428  * e1000_clean_rx_irq - Send received data up the network stack; legacy
429  * @adapter: board private structure
430  *
431  * the return value indicates whether actual cleaning was done, there
432  * is no guarantee that everything was cleaned
433  **/
434 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
435                                int *work_done, int work_to_do)
436 {
437         struct net_device *netdev = adapter->netdev;
438         struct pci_dev *pdev = adapter->pdev;
439         struct e1000_ring *rx_ring = adapter->rx_ring;
440         struct e1000_rx_desc *rx_desc, *next_rxd;
441         struct e1000_buffer *buffer_info, *next_buffer;
442         u32 length;
443         unsigned int i;
444         int cleaned_count = 0;
445         bool cleaned = 0;
446         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
447
448         i = rx_ring->next_to_clean;
449         rx_desc = E1000_RX_DESC(*rx_ring, i);
450         buffer_info = &rx_ring->buffer_info[i];
451
452         while (rx_desc->status & E1000_RXD_STAT_DD) {
453                 struct sk_buff *skb;
454                 u8 status;
455
456                 if (*work_done >= work_to_do)
457                         break;
458                 (*work_done)++;
459
460                 status = rx_desc->status;
461                 skb = buffer_info->skb;
462                 buffer_info->skb = NULL;
463
464                 prefetch(skb->data - NET_IP_ALIGN);
465
466                 i++;
467                 if (i == rx_ring->count)
468                         i = 0;
469                 next_rxd = E1000_RX_DESC(*rx_ring, i);
470                 prefetch(next_rxd);
471
472                 next_buffer = &rx_ring->buffer_info[i];
473
474                 cleaned = 1;
475                 cleaned_count++;
476                 pci_unmap_single(pdev,
477                                  buffer_info->dma,
478                                  adapter->rx_buffer_len,
479                                  PCI_DMA_FROMDEVICE);
480                 buffer_info->dma = 0;
481
482                 length = le16_to_cpu(rx_desc->length);
483
484                 /* !EOP means multiple descriptors were used to store a single
485                  * packet, also make sure the frame isn't just CRC only */
486                 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
487                         /* All receives must fit into a single buffer */
488                         e_dbg("%s: Receive packet consumed multiple buffers\n",
489                               netdev->name);
490                         /* recycle */
491                         buffer_info->skb = skb;
492                         goto next_desc;
493                 }
494
495                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
496                         /* recycle */
497                         buffer_info->skb = skb;
498                         goto next_desc;
499                 }
500
501                 /* adjust length to remove Ethernet CRC */
502                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
503                         length -= 4;
504
505                 total_rx_bytes += length;
506                 total_rx_packets++;
507
508                 /*
509                  * code added for copybreak, this should improve
510                  * performance for small packets with large amounts
511                  * of reassembly being done in the stack
512                  */
513                 if (length < copybreak) {
514                         struct sk_buff *new_skb =
515                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
516                         if (new_skb) {
517                                 skb_reserve(new_skb, NET_IP_ALIGN);
518                                 skb_copy_to_linear_data_offset(new_skb,
519                                                                -NET_IP_ALIGN,
520                                                                (skb->data -
521                                                                 NET_IP_ALIGN),
522                                                                (length +
523                                                                 NET_IP_ALIGN));
524                                 /* save the skb in buffer_info as good */
525                                 buffer_info->skb = skb;
526                                 skb = new_skb;
527                         }
528                         /* else just continue with the old one */
529                 }
530                 /* end copybreak code */
531                 skb_put(skb, length);
532
533                 /* Receive Checksum Offload */
534                 e1000_rx_checksum(adapter,
535                                   (u32)(status) |
536                                   ((u32)(rx_desc->errors) << 24),
537                                   le16_to_cpu(rx_desc->csum), skb);
538
539                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
540
541 next_desc:
542                 rx_desc->status = 0;
543
544                 /* return some buffers to hardware, one at a time is too slow */
545                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
546                         adapter->alloc_rx_buf(adapter, cleaned_count);
547                         cleaned_count = 0;
548                 }
549
550                 /* use prefetched values */
551                 rx_desc = next_rxd;
552                 buffer_info = next_buffer;
553         }
554         rx_ring->next_to_clean = i;
555
556         cleaned_count = e1000_desc_unused(rx_ring);
557         if (cleaned_count)
558                 adapter->alloc_rx_buf(adapter, cleaned_count);
559
560         adapter->total_rx_bytes += total_rx_bytes;
561         adapter->total_rx_packets += total_rx_packets;
562         adapter->net_stats.rx_bytes += total_rx_bytes;
563         adapter->net_stats.rx_packets += total_rx_packets;
564         return cleaned;
565 }
566
567 static void e1000_put_txbuf(struct e1000_adapter *adapter,
568                              struct e1000_buffer *buffer_info)
569 {
570         buffer_info->dma = 0;
571         if (buffer_info->skb) {
572                 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
573                               DMA_TO_DEVICE);
574                 dev_kfree_skb_any(buffer_info->skb);
575                 buffer_info->skb = NULL;
576         }
577         buffer_info->time_stamp = 0;
578 }
579
580 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
581 {
582         struct e1000_ring *tx_ring = adapter->tx_ring;
583         unsigned int i = tx_ring->next_to_clean;
584         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
585         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
586
587         /* detected Tx unit hang */
588         e_err("Detected Tx Unit Hang:\n"
589               "  TDH                  <%x>\n"
590               "  TDT                  <%x>\n"
591               "  next_to_use          <%x>\n"
592               "  next_to_clean        <%x>\n"
593               "buffer_info[next_to_clean]:\n"
594               "  time_stamp           <%lx>\n"
595               "  next_to_watch        <%x>\n"
596               "  jiffies              <%lx>\n"
597               "  next_to_watch.status <%x>\n",
598               readl(adapter->hw.hw_addr + tx_ring->head),
599               readl(adapter->hw.hw_addr + tx_ring->tail),
600               tx_ring->next_to_use,
601               tx_ring->next_to_clean,
602               tx_ring->buffer_info[eop].time_stamp,
603               eop,
604               jiffies,
605               eop_desc->upper.fields.status);
606 }
607
608 /**
609  * e1000_clean_tx_irq - Reclaim resources after transmit completes
610  * @adapter: board private structure
611  *
612  * the return value indicates whether actual cleaning was done, there
613  * is no guarantee that everything was cleaned
614  **/
615 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
616 {
617         struct net_device *netdev = adapter->netdev;
618         struct e1000_hw *hw = &adapter->hw;
619         struct e1000_ring *tx_ring = adapter->tx_ring;
620         struct e1000_tx_desc *tx_desc, *eop_desc;
621         struct e1000_buffer *buffer_info;
622         unsigned int i, eop;
623         unsigned int count = 0;
624         bool cleaned = 0;
625         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
626
627         i = tx_ring->next_to_clean;
628         eop = tx_ring->buffer_info[i].next_to_watch;
629         eop_desc = E1000_TX_DESC(*tx_ring, eop);
630
631         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
632                 for (cleaned = 0; !cleaned; ) {
633                         tx_desc = E1000_TX_DESC(*tx_ring, i);
634                         buffer_info = &tx_ring->buffer_info[i];
635                         cleaned = (i == eop);
636
637                         if (cleaned) {
638                                 struct sk_buff *skb = buffer_info->skb;
639                                 unsigned int segs, bytecount;
640                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
641                                 /* multiply data chunks by size of headers */
642                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
643                                             skb->len;
644                                 total_tx_packets += segs;
645                                 total_tx_bytes += bytecount;
646                         }
647
648                         e1000_put_txbuf(adapter, buffer_info);
649                         tx_desc->upper.data = 0;
650
651                         i++;
652                         if (i == tx_ring->count)
653                                 i = 0;
654                 }
655
656                 eop = tx_ring->buffer_info[i].next_to_watch;
657                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
658 #define E1000_TX_WEIGHT 64
659                 /* weight of a sort for tx, to avoid endless transmit cleanup */
660                 if (count++ == E1000_TX_WEIGHT)
661                         break;
662         }
663
664         tx_ring->next_to_clean = i;
665
666 #define TX_WAKE_THRESHOLD 32
667         if (cleaned && netif_carrier_ok(netdev) &&
668                      e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
669                 /* Make sure that anybody stopping the queue after this
670                  * sees the new next_to_clean.
671                  */
672                 smp_mb();
673
674                 if (netif_queue_stopped(netdev) &&
675                     !(test_bit(__E1000_DOWN, &adapter->state))) {
676                         netif_wake_queue(netdev);
677                         ++adapter->restart_queue;
678                 }
679         }
680
681         if (adapter->detect_tx_hung) {
682                 /* Detect a transmit hang in hardware, this serializes the
683                  * check with the clearing of time_stamp and movement of i */
684                 adapter->detect_tx_hung = 0;
685                 if (tx_ring->buffer_info[eop].time_stamp &&
686                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp
687                                + (adapter->tx_timeout_factor * HZ))
688                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
689                         e1000_print_tx_hang(adapter);
690                         netif_stop_queue(netdev);
691                 }
692         }
693         adapter->total_tx_bytes += total_tx_bytes;
694         adapter->total_tx_packets += total_tx_packets;
695         adapter->net_stats.tx_bytes += total_tx_bytes;
696         adapter->net_stats.tx_packets += total_tx_packets;
697         return cleaned;
698 }
699
700 /**
701  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
702  * @adapter: board private structure
703  *
704  * the return value indicates whether actual cleaning was done, there
705  * is no guarantee that everything was cleaned
706  **/
707 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
708                                   int *work_done, int work_to_do)
709 {
710         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
711         struct net_device *netdev = adapter->netdev;
712         struct pci_dev *pdev = adapter->pdev;
713         struct e1000_ring *rx_ring = adapter->rx_ring;
714         struct e1000_buffer *buffer_info, *next_buffer;
715         struct e1000_ps_page *ps_page;
716         struct sk_buff *skb;
717         unsigned int i, j;
718         u32 length, staterr;
719         int cleaned_count = 0;
720         bool cleaned = 0;
721         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
722
723         i = rx_ring->next_to_clean;
724         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
725         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
726         buffer_info = &rx_ring->buffer_info[i];
727
728         while (staterr & E1000_RXD_STAT_DD) {
729                 if (*work_done >= work_to_do)
730                         break;
731                 (*work_done)++;
732                 skb = buffer_info->skb;
733
734                 /* in the packet split case this is header only */
735                 prefetch(skb->data - NET_IP_ALIGN);
736
737                 i++;
738                 if (i == rx_ring->count)
739                         i = 0;
740                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
741                 prefetch(next_rxd);
742
743                 next_buffer = &rx_ring->buffer_info[i];
744
745                 cleaned = 1;
746                 cleaned_count++;
747                 pci_unmap_single(pdev, buffer_info->dma,
748                                  adapter->rx_ps_bsize0,
749                                  PCI_DMA_FROMDEVICE);
750                 buffer_info->dma = 0;
751
752                 if (!(staterr & E1000_RXD_STAT_EOP)) {
753                         e_dbg("%s: Packet Split buffers didn't pick up the "
754                               "full packet\n", netdev->name);
755                         dev_kfree_skb_irq(skb);
756                         goto next_desc;
757                 }
758
759                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
760                         dev_kfree_skb_irq(skb);
761                         goto next_desc;
762                 }
763
764                 length = le16_to_cpu(rx_desc->wb.middle.length0);
765
766                 if (!length) {
767                         e_dbg("%s: Last part of the packet spanning multiple "
768                               "descriptors\n", netdev->name);
769                         dev_kfree_skb_irq(skb);
770                         goto next_desc;
771                 }
772
773                 /* Good Receive */
774                 skb_put(skb, length);
775
776                 {
777                 /*
778                  * this looks ugly, but it seems compiler issues make it
779                  * more efficient than reusing j
780                  */
781                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
782
783                 /*
784                  * page alloc/put takes too long and effects small packet
785                  * throughput, so unsplit small packets and save the alloc/put
786                  * only valid in softirq (napi) context to call kmap_*
787                  */
788                 if (l1 && (l1 <= copybreak) &&
789                     ((length + l1) <= adapter->rx_ps_bsize0)) {
790                         u8 *vaddr;
791
792                         ps_page = &buffer_info->ps_pages[0];
793
794                         /*
795                          * there is no documentation about how to call
796                          * kmap_atomic, so we can't hold the mapping
797                          * very long
798                          */
799                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
800                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
801                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
802                         memcpy(skb_tail_pointer(skb), vaddr, l1);
803                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
804                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
805                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
806
807                         /* remove the CRC */
808                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
809                                 l1 -= 4;
810
811                         skb_put(skb, l1);
812                         goto copydone;
813                 } /* if */
814                 }
815
816                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
817                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
818                         if (!length)
819                                 break;
820
821                         ps_page = &buffer_info->ps_pages[j];
822                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
823                                        PCI_DMA_FROMDEVICE);
824                         ps_page->dma = 0;
825                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
826                         ps_page->page = NULL;
827                         skb->len += length;
828                         skb->data_len += length;
829                         skb->truesize += length;
830                 }
831
832                 /* strip the ethernet crc, problem is we're using pages now so
833                  * this whole operation can get a little cpu intensive
834                  */
835                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
836                         pskb_trim(skb, skb->len - 4);
837
838 copydone:
839                 total_rx_bytes += skb->len;
840                 total_rx_packets++;
841
842                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
843                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
844
845                 if (rx_desc->wb.upper.header_status &
846                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
847                         adapter->rx_hdr_split++;
848
849                 e1000_receive_skb(adapter, netdev, skb,
850                                   staterr, rx_desc->wb.middle.vlan);
851
852 next_desc:
853                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
854                 buffer_info->skb = NULL;
855
856                 /* return some buffers to hardware, one at a time is too slow */
857                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
858                         adapter->alloc_rx_buf(adapter, cleaned_count);
859                         cleaned_count = 0;
860                 }
861
862                 /* use prefetched values */
863                 rx_desc = next_rxd;
864                 buffer_info = next_buffer;
865
866                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
867         }
868         rx_ring->next_to_clean = i;
869
870         cleaned_count = e1000_desc_unused(rx_ring);
871         if (cleaned_count)
872                 adapter->alloc_rx_buf(adapter, cleaned_count);
873
874         adapter->total_rx_bytes += total_rx_bytes;
875         adapter->total_rx_packets += total_rx_packets;
876         adapter->net_stats.rx_bytes += total_rx_bytes;
877         adapter->net_stats.rx_packets += total_rx_packets;
878         return cleaned;
879 }
880
881 /**
882  * e1000_consume_page - helper function
883  **/
884 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
885                                u16 length)
886 {
887         bi->page = NULL;
888         skb->len += length;
889         skb->data_len += length;
890         skb->truesize += length;
891 }
892
893 /**
894  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
895  * @adapter: board private structure
896  *
897  * the return value indicates whether actual cleaning was done, there
898  * is no guarantee that everything was cleaned
899  **/
900
901 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
902                                      int *work_done, int work_to_do)
903 {
904         struct net_device *netdev = adapter->netdev;
905         struct pci_dev *pdev = adapter->pdev;
906         struct e1000_ring *rx_ring = adapter->rx_ring;
907         struct e1000_rx_desc *rx_desc, *next_rxd;
908         struct e1000_buffer *buffer_info, *next_buffer;
909         u32 length;
910         unsigned int i;
911         int cleaned_count = 0;
912         bool cleaned = false;
913         unsigned int total_rx_bytes=0, total_rx_packets=0;
914
915         i = rx_ring->next_to_clean;
916         rx_desc = E1000_RX_DESC(*rx_ring, i);
917         buffer_info = &rx_ring->buffer_info[i];
918
919         while (rx_desc->status & E1000_RXD_STAT_DD) {
920                 struct sk_buff *skb;
921                 u8 status;
922
923                 if (*work_done >= work_to_do)
924                         break;
925                 (*work_done)++;
926
927                 status = rx_desc->status;
928                 skb = buffer_info->skb;
929                 buffer_info->skb = NULL;
930
931                 ++i;
932                 if (i == rx_ring->count)
933                         i = 0;
934                 next_rxd = E1000_RX_DESC(*rx_ring, i);
935                 prefetch(next_rxd);
936
937                 next_buffer = &rx_ring->buffer_info[i];
938
939                 cleaned = true;
940                 cleaned_count++;
941                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
942                                PCI_DMA_FROMDEVICE);
943                 buffer_info->dma = 0;
944
945                 length = le16_to_cpu(rx_desc->length);
946
947                 /* errors is only valid for DD + EOP descriptors */
948                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
949                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
950                                 /* recycle both page and skb */
951                                 buffer_info->skb = skb;
952                                 /* an error means any chain goes out the window
953                                  * too */
954                                 if (rx_ring->rx_skb_top)
955                                         dev_kfree_skb(rx_ring->rx_skb_top);
956                                 rx_ring->rx_skb_top = NULL;
957                                 goto next_desc;
958                 }
959
960 #define rxtop rx_ring->rx_skb_top
961                 if (!(status & E1000_RXD_STAT_EOP)) {
962                         /* this descriptor is only the beginning (or middle) */
963                         if (!rxtop) {
964                                 /* this is the beginning of a chain */
965                                 rxtop = skb;
966                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
967                                                    0, length);
968                         } else {
969                                 /* this is the middle of a chain */
970                                 skb_fill_page_desc(rxtop,
971                                     skb_shinfo(rxtop)->nr_frags,
972                                     buffer_info->page, 0, length);
973                                 /* re-use the skb, only consumed the page */
974                                 buffer_info->skb = skb;
975                         }
976                         e1000_consume_page(buffer_info, rxtop, length);
977                         goto next_desc;
978                 } else {
979                         if (rxtop) {
980                                 /* end of the chain */
981                                 skb_fill_page_desc(rxtop,
982                                     skb_shinfo(rxtop)->nr_frags,
983                                     buffer_info->page, 0, length);
984                                 /* re-use the current skb, we only consumed the
985                                  * page */
986                                 buffer_info->skb = skb;
987                                 skb = rxtop;
988                                 rxtop = NULL;
989                                 e1000_consume_page(buffer_info, skb, length);
990                         } else {
991                                 /* no chain, got EOP, this buf is the packet
992                                  * copybreak to save the put_page/alloc_page */
993                                 if (length <= copybreak &&
994                                     skb_tailroom(skb) >= length) {
995                                         u8 *vaddr;
996                                         vaddr = kmap_atomic(buffer_info->page,
997                                                            KM_SKB_DATA_SOFTIRQ);
998                                         memcpy(skb_tail_pointer(skb), vaddr,
999                                                length);
1000                                         kunmap_atomic(vaddr,
1001                                                       KM_SKB_DATA_SOFTIRQ);
1002                                         /* re-use the page, so don't erase
1003                                          * buffer_info->page */
1004                                         skb_put(skb, length);
1005                                 } else {
1006                                         skb_fill_page_desc(skb, 0,
1007                                                            buffer_info->page, 0,
1008                                                            length);
1009                                         e1000_consume_page(buffer_info, skb,
1010                                                            length);
1011                                 }
1012                         }
1013                 }
1014
1015                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1016                 e1000_rx_checksum(adapter,
1017                                   (u32)(status) |
1018                                   ((u32)(rx_desc->errors) << 24),
1019                                   le16_to_cpu(rx_desc->csum), skb);
1020
1021                 /* probably a little skewed due to removing CRC */
1022                 total_rx_bytes += skb->len;
1023                 total_rx_packets++;
1024
1025                 /* eth type trans needs skb->data to point to something */
1026                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1027                         e_err("pskb_may_pull failed.\n");
1028                         dev_kfree_skb(skb);
1029                         goto next_desc;
1030                 }
1031
1032                 e1000_receive_skb(adapter, netdev, skb, status,
1033                                   rx_desc->special);
1034
1035 next_desc:
1036                 rx_desc->status = 0;
1037
1038                 /* return some buffers to hardware, one at a time is too slow */
1039                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1040                         adapter->alloc_rx_buf(adapter, cleaned_count);
1041                         cleaned_count = 0;
1042                 }
1043
1044                 /* use prefetched values */
1045                 rx_desc = next_rxd;
1046                 buffer_info = next_buffer;
1047         }
1048         rx_ring->next_to_clean = i;
1049
1050         cleaned_count = e1000_desc_unused(rx_ring);
1051         if (cleaned_count)
1052                 adapter->alloc_rx_buf(adapter, cleaned_count);
1053
1054         adapter->total_rx_bytes += total_rx_bytes;
1055         adapter->total_rx_packets += total_rx_packets;
1056         adapter->net_stats.rx_bytes += total_rx_bytes;
1057         adapter->net_stats.rx_packets += total_rx_packets;
1058         return cleaned;
1059 }
1060
1061 /**
1062  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1063  * @adapter: board private structure
1064  **/
1065 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1066 {
1067         struct e1000_ring *rx_ring = adapter->rx_ring;
1068         struct e1000_buffer *buffer_info;
1069         struct e1000_ps_page *ps_page;
1070         struct pci_dev *pdev = adapter->pdev;
1071         unsigned int i, j;
1072
1073         /* Free all the Rx ring sk_buffs */
1074         for (i = 0; i < rx_ring->count; i++) {
1075                 buffer_info = &rx_ring->buffer_info[i];
1076                 if (buffer_info->dma) {
1077                         if (adapter->clean_rx == e1000_clean_rx_irq)
1078                                 pci_unmap_single(pdev, buffer_info->dma,
1079                                                  adapter->rx_buffer_len,
1080                                                  PCI_DMA_FROMDEVICE);
1081                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1082                                 pci_unmap_page(pdev, buffer_info->dma,
1083                                                PAGE_SIZE,
1084                                                PCI_DMA_FROMDEVICE);
1085                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1086                                 pci_unmap_single(pdev, buffer_info->dma,
1087                                                  adapter->rx_ps_bsize0,
1088                                                  PCI_DMA_FROMDEVICE);
1089                         buffer_info->dma = 0;
1090                 }
1091
1092                 if (buffer_info->page) {
1093                         put_page(buffer_info->page);
1094                         buffer_info->page = NULL;
1095                 }
1096
1097                 if (buffer_info->skb) {
1098                         dev_kfree_skb(buffer_info->skb);
1099                         buffer_info->skb = NULL;
1100                 }
1101
1102                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1103                         ps_page = &buffer_info->ps_pages[j];
1104                         if (!ps_page->page)
1105                                 break;
1106                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1107                                        PCI_DMA_FROMDEVICE);
1108                         ps_page->dma = 0;
1109                         put_page(ps_page->page);
1110                         ps_page->page = NULL;
1111                 }
1112         }
1113
1114         /* there also may be some cached data from a chained receive */
1115         if (rx_ring->rx_skb_top) {
1116                 dev_kfree_skb(rx_ring->rx_skb_top);
1117                 rx_ring->rx_skb_top = NULL;
1118         }
1119
1120         /* Zero out the descriptor ring */
1121         memset(rx_ring->desc, 0, rx_ring->size);
1122
1123         rx_ring->next_to_clean = 0;
1124         rx_ring->next_to_use = 0;
1125
1126         writel(0, adapter->hw.hw_addr + rx_ring->head);
1127         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1128 }
1129
1130 static void e1000e_downshift_workaround(struct work_struct *work)
1131 {
1132         struct e1000_adapter *adapter = container_of(work,
1133                                         struct e1000_adapter, downshift_task);
1134
1135         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1136 }
1137
1138 /**
1139  * e1000_intr_msi - Interrupt Handler
1140  * @irq: interrupt number
1141  * @data: pointer to a network interface device structure
1142  **/
1143 static irqreturn_t e1000_intr_msi(int irq, void *data)
1144 {
1145         struct net_device *netdev = data;
1146         struct e1000_adapter *adapter = netdev_priv(netdev);
1147         struct e1000_hw *hw = &adapter->hw;
1148         u32 icr = er32(ICR);
1149
1150         /*
1151          * read ICR disables interrupts using IAM
1152          */
1153
1154         if (icr & E1000_ICR_LSC) {
1155                 hw->mac.get_link_status = 1;
1156                 /*
1157                  * ICH8 workaround-- Call gig speed drop workaround on cable
1158                  * disconnect (LSC) before accessing any PHY registers
1159                  */
1160                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1161                     (!(er32(STATUS) & E1000_STATUS_LU)))
1162                         schedule_work(&adapter->downshift_task);
1163
1164                 /*
1165                  * 80003ES2LAN workaround-- For packet buffer work-around on
1166                  * link down event; disable receives here in the ISR and reset
1167                  * adapter in watchdog
1168                  */
1169                 if (netif_carrier_ok(netdev) &&
1170                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1171                         /* disable receives */
1172                         u32 rctl = er32(RCTL);
1173                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1174                         adapter->flags |= FLAG_RX_RESTART_NOW;
1175                 }
1176                 /* guard against interrupt when we're going down */
1177                 if (!test_bit(__E1000_DOWN, &adapter->state))
1178                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1179         }
1180
1181         if (napi_schedule_prep(&adapter->napi)) {
1182                 adapter->total_tx_bytes = 0;
1183                 adapter->total_tx_packets = 0;
1184                 adapter->total_rx_bytes = 0;
1185                 adapter->total_rx_packets = 0;
1186                 __napi_schedule(&adapter->napi);
1187         }
1188
1189         return IRQ_HANDLED;
1190 }
1191
1192 /**
1193  * e1000_intr - Interrupt Handler
1194  * @irq: interrupt number
1195  * @data: pointer to a network interface device structure
1196  **/
1197 static irqreturn_t e1000_intr(int irq, void *data)
1198 {
1199         struct net_device *netdev = data;
1200         struct e1000_adapter *adapter = netdev_priv(netdev);
1201         struct e1000_hw *hw = &adapter->hw;
1202         u32 rctl, icr = er32(ICR);
1203
1204         if (!icr)
1205                 return IRQ_NONE;  /* Not our interrupt */
1206
1207         /*
1208          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1209          * not set, then the adapter didn't send an interrupt
1210          */
1211         if (!(icr & E1000_ICR_INT_ASSERTED))
1212                 return IRQ_NONE;
1213
1214         /*
1215          * Interrupt Auto-Mask...upon reading ICR,
1216          * interrupts are masked.  No need for the
1217          * IMC write
1218          */
1219
1220         if (icr & E1000_ICR_LSC) {
1221                 hw->mac.get_link_status = 1;
1222                 /*
1223                  * ICH8 workaround-- Call gig speed drop workaround on cable
1224                  * disconnect (LSC) before accessing any PHY registers
1225                  */
1226                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1227                     (!(er32(STATUS) & E1000_STATUS_LU)))
1228                         schedule_work(&adapter->downshift_task);
1229
1230                 /*
1231                  * 80003ES2LAN workaround--
1232                  * For packet buffer work-around on link down event;
1233                  * disable receives here in the ISR and
1234                  * reset adapter in watchdog
1235                  */
1236                 if (netif_carrier_ok(netdev) &&
1237                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1238                         /* disable receives */
1239                         rctl = er32(RCTL);
1240                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1241                         adapter->flags |= FLAG_RX_RESTART_NOW;
1242                 }
1243                 /* guard against interrupt when we're going down */
1244                 if (!test_bit(__E1000_DOWN, &adapter->state))
1245                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1246         }
1247
1248         if (napi_schedule_prep(&adapter->napi)) {
1249                 adapter->total_tx_bytes = 0;
1250                 adapter->total_tx_packets = 0;
1251                 adapter->total_rx_bytes = 0;
1252                 adapter->total_rx_packets = 0;
1253                 __napi_schedule(&adapter->napi);
1254         }
1255
1256         return IRQ_HANDLED;
1257 }
1258
1259 static irqreturn_t e1000_msix_other(int irq, void *data)
1260 {
1261         struct net_device *netdev = data;
1262         struct e1000_adapter *adapter = netdev_priv(netdev);
1263         struct e1000_hw *hw = &adapter->hw;
1264         u32 icr = er32(ICR);
1265
1266         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1267                 ew32(IMS, E1000_IMS_OTHER);
1268                 return IRQ_NONE;
1269         }
1270
1271         if (icr & adapter->eiac_mask)
1272                 ew32(ICS, (icr & adapter->eiac_mask));
1273
1274         if (icr & E1000_ICR_OTHER) {
1275                 if (!(icr & E1000_ICR_LSC))
1276                         goto no_link_interrupt;
1277                 hw->mac.get_link_status = 1;
1278                 /* guard against interrupt when we're going down */
1279                 if (!test_bit(__E1000_DOWN, &adapter->state))
1280                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1281         }
1282
1283 no_link_interrupt:
1284         ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1285
1286         return IRQ_HANDLED;
1287 }
1288
1289
1290 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1291 {
1292         struct net_device *netdev = data;
1293         struct e1000_adapter *adapter = netdev_priv(netdev);
1294         struct e1000_hw *hw = &adapter->hw;
1295         struct e1000_ring *tx_ring = adapter->tx_ring;
1296
1297
1298         adapter->total_tx_bytes = 0;
1299         adapter->total_tx_packets = 0;
1300
1301         if (!e1000_clean_tx_irq(adapter))
1302                 /* Ring was not completely cleaned, so fire another interrupt */
1303                 ew32(ICS, tx_ring->ims_val);
1304
1305         return IRQ_HANDLED;
1306 }
1307
1308 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1309 {
1310         struct net_device *netdev = data;
1311         struct e1000_adapter *adapter = netdev_priv(netdev);
1312
1313         /* Write the ITR value calculated at the end of the
1314          * previous interrupt.
1315          */
1316         if (adapter->rx_ring->set_itr) {
1317                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1318                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1319                 adapter->rx_ring->set_itr = 0;
1320         }
1321
1322         if (napi_schedule_prep(&adapter->napi)) {
1323                 adapter->total_rx_bytes = 0;
1324                 adapter->total_rx_packets = 0;
1325                 __napi_schedule(&adapter->napi);
1326         }
1327         return IRQ_HANDLED;
1328 }
1329
1330 /**
1331  * e1000_configure_msix - Configure MSI-X hardware
1332  *
1333  * e1000_configure_msix sets up the hardware to properly
1334  * generate MSI-X interrupts.
1335  **/
1336 static void e1000_configure_msix(struct e1000_adapter *adapter)
1337 {
1338         struct e1000_hw *hw = &adapter->hw;
1339         struct e1000_ring *rx_ring = adapter->rx_ring;
1340         struct e1000_ring *tx_ring = adapter->tx_ring;
1341         int vector = 0;
1342         u32 ctrl_ext, ivar = 0;
1343
1344         adapter->eiac_mask = 0;
1345
1346         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1347         if (hw->mac.type == e1000_82574) {
1348                 u32 rfctl = er32(RFCTL);
1349                 rfctl |= E1000_RFCTL_ACK_DIS;
1350                 ew32(RFCTL, rfctl);
1351         }
1352
1353 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1354         /* Configure Rx vector */
1355         rx_ring->ims_val = E1000_IMS_RXQ0;
1356         adapter->eiac_mask |= rx_ring->ims_val;
1357         if (rx_ring->itr_val)
1358                 writel(1000000000 / (rx_ring->itr_val * 256),
1359                        hw->hw_addr + rx_ring->itr_register);
1360         else
1361                 writel(1, hw->hw_addr + rx_ring->itr_register);
1362         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1363
1364         /* Configure Tx vector */
1365         tx_ring->ims_val = E1000_IMS_TXQ0;
1366         vector++;
1367         if (tx_ring->itr_val)
1368                 writel(1000000000 / (tx_ring->itr_val * 256),
1369                        hw->hw_addr + tx_ring->itr_register);
1370         else
1371                 writel(1, hw->hw_addr + tx_ring->itr_register);
1372         adapter->eiac_mask |= tx_ring->ims_val;
1373         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1374
1375         /* set vector for Other Causes, e.g. link changes */
1376         vector++;
1377         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1378         if (rx_ring->itr_val)
1379                 writel(1000000000 / (rx_ring->itr_val * 256),
1380                        hw->hw_addr + E1000_EITR_82574(vector));
1381         else
1382                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1383
1384         /* Cause Tx interrupts on every write back */
1385         ivar |= (1 << 31);
1386
1387         ew32(IVAR, ivar);
1388
1389         /* enable MSI-X PBA support */
1390         ctrl_ext = er32(CTRL_EXT);
1391         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1392
1393         /* Auto-Mask Other interrupts upon ICR read */
1394 #define E1000_EIAC_MASK_82574   0x01F00000
1395         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1396         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1397         ew32(CTRL_EXT, ctrl_ext);
1398         e1e_flush();
1399 }
1400
1401 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1402 {
1403         if (adapter->msix_entries) {
1404                 pci_disable_msix(adapter->pdev);
1405                 kfree(adapter->msix_entries);
1406                 adapter->msix_entries = NULL;
1407         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1408                 pci_disable_msi(adapter->pdev);
1409                 adapter->flags &= ~FLAG_MSI_ENABLED;
1410         }
1411
1412         return;
1413 }
1414
1415 /**
1416  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1417  *
1418  * Attempt to configure interrupts using the best available
1419  * capabilities of the hardware and kernel.
1420  **/
1421 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1422 {
1423         int err;
1424         int numvecs, i;
1425
1426
1427         switch (adapter->int_mode) {
1428         case E1000E_INT_MODE_MSIX:
1429                 if (adapter->flags & FLAG_HAS_MSIX) {
1430                         numvecs = 3; /* RxQ0, TxQ0 and other */
1431                         adapter->msix_entries = kcalloc(numvecs,
1432                                                       sizeof(struct msix_entry),
1433                                                       GFP_KERNEL);
1434                         if (adapter->msix_entries) {
1435                                 for (i = 0; i < numvecs; i++)
1436                                         adapter->msix_entries[i].entry = i;
1437
1438                                 err = pci_enable_msix(adapter->pdev,
1439                                                       adapter->msix_entries,
1440                                                       numvecs);
1441                                 if (err == 0)
1442                                         return;
1443                         }
1444                         /* MSI-X failed, so fall through and try MSI */
1445                         e_err("Failed to initialize MSI-X interrupts.  "
1446                               "Falling back to MSI interrupts.\n");
1447                         e1000e_reset_interrupt_capability(adapter);
1448                 }
1449                 adapter->int_mode = E1000E_INT_MODE_MSI;
1450                 /* Fall through */
1451         case E1000E_INT_MODE_MSI:
1452                 if (!pci_enable_msi(adapter->pdev)) {
1453                         adapter->flags |= FLAG_MSI_ENABLED;
1454                 } else {
1455                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1456                         e_err("Failed to initialize MSI interrupts.  Falling "
1457                               "back to legacy interrupts.\n");
1458                 }
1459                 /* Fall through */
1460         case E1000E_INT_MODE_LEGACY:
1461                 /* Don't do anything; this is the system default */
1462                 break;
1463         }
1464
1465         return;
1466 }
1467
1468 /**
1469  * e1000_request_msix - Initialize MSI-X interrupts
1470  *
1471  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1472  * kernel.
1473  **/
1474 static int e1000_request_msix(struct e1000_adapter *adapter)
1475 {
1476         struct net_device *netdev = adapter->netdev;
1477         int err = 0, vector = 0;
1478
1479         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1480                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1481         else
1482                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1483         err = request_irq(adapter->msix_entries[vector].vector,
1484                           &e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1485                           netdev);
1486         if (err)
1487                 goto out;
1488         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1489         adapter->rx_ring->itr_val = adapter->itr;
1490         vector++;
1491
1492         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1493                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1494         else
1495                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1496         err = request_irq(adapter->msix_entries[vector].vector,
1497                           &e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1498                           netdev);
1499         if (err)
1500                 goto out;
1501         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1502         adapter->tx_ring->itr_val = adapter->itr;
1503         vector++;
1504
1505         err = request_irq(adapter->msix_entries[vector].vector,
1506                           &e1000_msix_other, 0, netdev->name, netdev);
1507         if (err)
1508                 goto out;
1509
1510         e1000_configure_msix(adapter);
1511         return 0;
1512 out:
1513         return err;
1514 }
1515
1516 /**
1517  * e1000_request_irq - initialize interrupts
1518  *
1519  * Attempts to configure interrupts using the best available
1520  * capabilities of the hardware and kernel.
1521  **/
1522 static int e1000_request_irq(struct e1000_adapter *adapter)
1523 {
1524         struct net_device *netdev = adapter->netdev;
1525         int err;
1526
1527         if (adapter->msix_entries) {
1528                 err = e1000_request_msix(adapter);
1529                 if (!err)
1530                         return err;
1531                 /* fall back to MSI */
1532                 e1000e_reset_interrupt_capability(adapter);
1533                 adapter->int_mode = E1000E_INT_MODE_MSI;
1534                 e1000e_set_interrupt_capability(adapter);
1535         }
1536         if (adapter->flags & FLAG_MSI_ENABLED) {
1537                 err = request_irq(adapter->pdev->irq, &e1000_intr_msi, 0,
1538                                   netdev->name, netdev);
1539                 if (!err)
1540                         return err;
1541
1542                 /* fall back to legacy interrupt */
1543                 e1000e_reset_interrupt_capability(adapter);
1544                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1545         }
1546
1547         err = request_irq(adapter->pdev->irq, &e1000_intr, IRQF_SHARED,
1548                           netdev->name, netdev);
1549         if (err)
1550                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1551
1552         return err;
1553 }
1554
1555 static void e1000_free_irq(struct e1000_adapter *adapter)
1556 {
1557         struct net_device *netdev = adapter->netdev;
1558
1559         if (adapter->msix_entries) {
1560                 int vector = 0;
1561
1562                 free_irq(adapter->msix_entries[vector].vector, netdev);
1563                 vector++;
1564
1565                 free_irq(adapter->msix_entries[vector].vector, netdev);
1566                 vector++;
1567
1568                 /* Other Causes interrupt vector */
1569                 free_irq(adapter->msix_entries[vector].vector, netdev);
1570                 return;
1571         }
1572
1573         free_irq(adapter->pdev->irq, netdev);
1574 }
1575
1576 /**
1577  * e1000_irq_disable - Mask off interrupt generation on the NIC
1578  **/
1579 static void e1000_irq_disable(struct e1000_adapter *adapter)
1580 {
1581         struct e1000_hw *hw = &adapter->hw;
1582
1583         ew32(IMC, ~0);
1584         if (adapter->msix_entries)
1585                 ew32(EIAC_82574, 0);
1586         e1e_flush();
1587         synchronize_irq(adapter->pdev->irq);
1588 }
1589
1590 /**
1591  * e1000_irq_enable - Enable default interrupt generation settings
1592  **/
1593 static void e1000_irq_enable(struct e1000_adapter *adapter)
1594 {
1595         struct e1000_hw *hw = &adapter->hw;
1596
1597         if (adapter->msix_entries) {
1598                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1599                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1600         } else {
1601                 ew32(IMS, IMS_ENABLE_MASK);
1602         }
1603         e1e_flush();
1604 }
1605
1606 /**
1607  * e1000_get_hw_control - get control of the h/w from f/w
1608  * @adapter: address of board private structure
1609  *
1610  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1611  * For ASF and Pass Through versions of f/w this means that
1612  * the driver is loaded. For AMT version (only with 82573)
1613  * of the f/w this means that the network i/f is open.
1614  **/
1615 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1616 {
1617         struct e1000_hw *hw = &adapter->hw;
1618         u32 ctrl_ext;
1619         u32 swsm;
1620
1621         /* Let firmware know the driver has taken over */
1622         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1623                 swsm = er32(SWSM);
1624                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1625         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1626                 ctrl_ext = er32(CTRL_EXT);
1627                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1628         }
1629 }
1630
1631 /**
1632  * e1000_release_hw_control - release control of the h/w to f/w
1633  * @adapter: address of board private structure
1634  *
1635  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1636  * For ASF and Pass Through versions of f/w this means that the
1637  * driver is no longer loaded. For AMT version (only with 82573) i
1638  * of the f/w this means that the network i/f is closed.
1639  *
1640  **/
1641 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1642 {
1643         struct e1000_hw *hw = &adapter->hw;
1644         u32 ctrl_ext;
1645         u32 swsm;
1646
1647         /* Let firmware taken over control of h/w */
1648         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1649                 swsm = er32(SWSM);
1650                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1651         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1652                 ctrl_ext = er32(CTRL_EXT);
1653                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1654         }
1655 }
1656
1657 /**
1658  * @e1000_alloc_ring - allocate memory for a ring structure
1659  **/
1660 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1661                                 struct e1000_ring *ring)
1662 {
1663         struct pci_dev *pdev = adapter->pdev;
1664
1665         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1666                                         GFP_KERNEL);
1667         if (!ring->desc)
1668                 return -ENOMEM;
1669
1670         return 0;
1671 }
1672
1673 /**
1674  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1675  * @adapter: board private structure
1676  *
1677  * Return 0 on success, negative on failure
1678  **/
1679 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1680 {
1681         struct e1000_ring *tx_ring = adapter->tx_ring;
1682         int err = -ENOMEM, size;
1683
1684         size = sizeof(struct e1000_buffer) * tx_ring->count;
1685         tx_ring->buffer_info = vmalloc(size);
1686         if (!tx_ring->buffer_info)
1687                 goto err;
1688         memset(tx_ring->buffer_info, 0, size);
1689
1690         /* round up to nearest 4K */
1691         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1692         tx_ring->size = ALIGN(tx_ring->size, 4096);
1693
1694         err = e1000_alloc_ring_dma(adapter, tx_ring);
1695         if (err)
1696                 goto err;
1697
1698         tx_ring->next_to_use = 0;
1699         tx_ring->next_to_clean = 0;
1700
1701         return 0;
1702 err:
1703         vfree(tx_ring->buffer_info);
1704         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1705         return err;
1706 }
1707
1708 /**
1709  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1710  * @adapter: board private structure
1711  *
1712  * Returns 0 on success, negative on failure
1713  **/
1714 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1715 {
1716         struct e1000_ring *rx_ring = adapter->rx_ring;
1717         struct e1000_buffer *buffer_info;
1718         int i, size, desc_len, err = -ENOMEM;
1719
1720         size = sizeof(struct e1000_buffer) * rx_ring->count;
1721         rx_ring->buffer_info = vmalloc(size);
1722         if (!rx_ring->buffer_info)
1723                 goto err;
1724         memset(rx_ring->buffer_info, 0, size);
1725
1726         for (i = 0; i < rx_ring->count; i++) {
1727                 buffer_info = &rx_ring->buffer_info[i];
1728                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1729                                                 sizeof(struct e1000_ps_page),
1730                                                 GFP_KERNEL);
1731                 if (!buffer_info->ps_pages)
1732                         goto err_pages;
1733         }
1734
1735         desc_len = sizeof(union e1000_rx_desc_packet_split);
1736
1737         /* Round up to nearest 4K */
1738         rx_ring->size = rx_ring->count * desc_len;
1739         rx_ring->size = ALIGN(rx_ring->size, 4096);
1740
1741         err = e1000_alloc_ring_dma(adapter, rx_ring);
1742         if (err)
1743                 goto err_pages;
1744
1745         rx_ring->next_to_clean = 0;
1746         rx_ring->next_to_use = 0;
1747         rx_ring->rx_skb_top = NULL;
1748
1749         return 0;
1750
1751 err_pages:
1752         for (i = 0; i < rx_ring->count; i++) {
1753                 buffer_info = &rx_ring->buffer_info[i];
1754                 kfree(buffer_info->ps_pages);
1755         }
1756 err:
1757         vfree(rx_ring->buffer_info);
1758         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1759         return err;
1760 }
1761
1762 /**
1763  * e1000_clean_tx_ring - Free Tx Buffers
1764  * @adapter: board private structure
1765  **/
1766 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1767 {
1768         struct e1000_ring *tx_ring = adapter->tx_ring;
1769         struct e1000_buffer *buffer_info;
1770         unsigned long size;
1771         unsigned int i;
1772
1773         for (i = 0; i < tx_ring->count; i++) {
1774                 buffer_info = &tx_ring->buffer_info[i];
1775                 e1000_put_txbuf(adapter, buffer_info);
1776         }
1777
1778         size = sizeof(struct e1000_buffer) * tx_ring->count;
1779         memset(tx_ring->buffer_info, 0, size);
1780
1781         memset(tx_ring->desc, 0, tx_ring->size);
1782
1783         tx_ring->next_to_use = 0;
1784         tx_ring->next_to_clean = 0;
1785
1786         writel(0, adapter->hw.hw_addr + tx_ring->head);
1787         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1788 }
1789
1790 /**
1791  * e1000e_free_tx_resources - Free Tx Resources per Queue
1792  * @adapter: board private structure
1793  *
1794  * Free all transmit software resources
1795  **/
1796 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1797 {
1798         struct pci_dev *pdev = adapter->pdev;
1799         struct e1000_ring *tx_ring = adapter->tx_ring;
1800
1801         e1000_clean_tx_ring(adapter);
1802
1803         vfree(tx_ring->buffer_info);
1804         tx_ring->buffer_info = NULL;
1805
1806         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1807                           tx_ring->dma);
1808         tx_ring->desc = NULL;
1809 }
1810
1811 /**
1812  * e1000e_free_rx_resources - Free Rx Resources
1813  * @adapter: board private structure
1814  *
1815  * Free all receive software resources
1816  **/
1817
1818 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1819 {
1820         struct pci_dev *pdev = adapter->pdev;
1821         struct e1000_ring *rx_ring = adapter->rx_ring;
1822         int i;
1823
1824         e1000_clean_rx_ring(adapter);
1825
1826         for (i = 0; i < rx_ring->count; i++) {
1827                 kfree(rx_ring->buffer_info[i].ps_pages);
1828         }
1829
1830         vfree(rx_ring->buffer_info);
1831         rx_ring->buffer_info = NULL;
1832
1833         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1834                           rx_ring->dma);
1835         rx_ring->desc = NULL;
1836 }
1837
1838 /**
1839  * e1000_update_itr - update the dynamic ITR value based on statistics
1840  * @adapter: pointer to adapter
1841  * @itr_setting: current adapter->itr
1842  * @packets: the number of packets during this measurement interval
1843  * @bytes: the number of bytes during this measurement interval
1844  *
1845  *      Stores a new ITR value based on packets and byte
1846  *      counts during the last interrupt.  The advantage of per interrupt
1847  *      computation is faster updates and more accurate ITR for the current
1848  *      traffic pattern.  Constants in this function were computed
1849  *      based on theoretical maximum wire speed and thresholds were set based
1850  *      on testing data as well as attempting to minimize response time
1851  *      while increasing bulk throughput.  This functionality is controlled
1852  *      by the InterruptThrottleRate module parameter.
1853  **/
1854 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1855                                      u16 itr_setting, int packets,
1856                                      int bytes)
1857 {
1858         unsigned int retval = itr_setting;
1859
1860         if (packets == 0)
1861                 goto update_itr_done;
1862
1863         switch (itr_setting) {
1864         case lowest_latency:
1865                 /* handle TSO and jumbo frames */
1866                 if (bytes/packets > 8000)
1867                         retval = bulk_latency;
1868                 else if ((packets < 5) && (bytes > 512)) {
1869                         retval = low_latency;
1870                 }
1871                 break;
1872         case low_latency:  /* 50 usec aka 20000 ints/s */
1873                 if (bytes > 10000) {
1874                         /* this if handles the TSO accounting */
1875                         if (bytes/packets > 8000) {
1876                                 retval = bulk_latency;
1877                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1878                                 retval = bulk_latency;
1879                         } else if ((packets > 35)) {
1880                                 retval = lowest_latency;
1881                         }
1882                 } else if (bytes/packets > 2000) {
1883                         retval = bulk_latency;
1884                 } else if (packets <= 2 && bytes < 512) {
1885                         retval = lowest_latency;
1886                 }
1887                 break;
1888         case bulk_latency: /* 250 usec aka 4000 ints/s */
1889                 if (bytes > 25000) {
1890                         if (packets > 35) {
1891                                 retval = low_latency;
1892                         }
1893                 } else if (bytes < 6000) {
1894                         retval = low_latency;
1895                 }
1896                 break;
1897         }
1898
1899 update_itr_done:
1900         return retval;
1901 }
1902
1903 static void e1000_set_itr(struct e1000_adapter *adapter)
1904 {
1905         struct e1000_hw *hw = &adapter->hw;
1906         u16 current_itr;
1907         u32 new_itr = adapter->itr;
1908
1909         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1910         if (adapter->link_speed != SPEED_1000) {
1911                 current_itr = 0;
1912                 new_itr = 4000;
1913                 goto set_itr_now;
1914         }
1915
1916         adapter->tx_itr = e1000_update_itr(adapter,
1917                                     adapter->tx_itr,
1918                                     adapter->total_tx_packets,
1919                                     adapter->total_tx_bytes);
1920         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1921         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1922                 adapter->tx_itr = low_latency;
1923
1924         adapter->rx_itr = e1000_update_itr(adapter,
1925                                     adapter->rx_itr,
1926                                     adapter->total_rx_packets,
1927                                     adapter->total_rx_bytes);
1928         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1929         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1930                 adapter->rx_itr = low_latency;
1931
1932         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1933
1934         switch (current_itr) {
1935         /* counts and packets in update_itr are dependent on these numbers */
1936         case lowest_latency:
1937                 new_itr = 70000;
1938                 break;
1939         case low_latency:
1940                 new_itr = 20000; /* aka hwitr = ~200 */
1941                 break;
1942         case bulk_latency:
1943                 new_itr = 4000;
1944                 break;
1945         default:
1946                 break;
1947         }
1948
1949 set_itr_now:
1950         if (new_itr != adapter->itr) {
1951                 /*
1952                  * this attempts to bias the interrupt rate towards Bulk
1953                  * by adding intermediate steps when interrupt rate is
1954                  * increasing
1955                  */
1956                 new_itr = new_itr > adapter->itr ?
1957                              min(adapter->itr + (new_itr >> 2), new_itr) :
1958                              new_itr;
1959                 adapter->itr = new_itr;
1960                 adapter->rx_ring->itr_val = new_itr;
1961                 if (adapter->msix_entries)
1962                         adapter->rx_ring->set_itr = 1;
1963                 else
1964                         ew32(ITR, 1000000000 / (new_itr * 256));
1965         }
1966 }
1967
1968 /**
1969  * e1000_alloc_queues - Allocate memory for all rings
1970  * @adapter: board private structure to initialize
1971  **/
1972 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1973 {
1974         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1975         if (!adapter->tx_ring)
1976                 goto err;
1977
1978         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1979         if (!adapter->rx_ring)
1980                 goto err;
1981
1982         return 0;
1983 err:
1984         e_err("Unable to allocate memory for queues\n");
1985         kfree(adapter->rx_ring);
1986         kfree(adapter->tx_ring);
1987         return -ENOMEM;
1988 }
1989
1990 /**
1991  * e1000_clean - NAPI Rx polling callback
1992  * @napi: struct associated with this polling callback
1993  * @budget: amount of packets driver is allowed to process this poll
1994  **/
1995 static int e1000_clean(struct napi_struct *napi, int budget)
1996 {
1997         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1998         struct e1000_hw *hw = &adapter->hw;
1999         struct net_device *poll_dev = adapter->netdev;
2000         int tx_cleaned = 0, work_done = 0;
2001
2002         adapter = netdev_priv(poll_dev);
2003
2004         if (adapter->msix_entries &&
2005             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2006                 goto clean_rx;
2007
2008         tx_cleaned = e1000_clean_tx_irq(adapter);
2009
2010 clean_rx:
2011         adapter->clean_rx(adapter, &work_done, budget);
2012
2013         if (tx_cleaned)
2014                 work_done = budget;
2015
2016         /* If budget not fully consumed, exit the polling mode */
2017         if (work_done < budget) {
2018                 if (adapter->itr_setting & 3)
2019                         e1000_set_itr(adapter);
2020                 napi_complete(napi);
2021                 if (adapter->msix_entries)
2022                         ew32(IMS, adapter->rx_ring->ims_val);
2023                 else
2024                         e1000_irq_enable(adapter);
2025         }
2026
2027         return work_done;
2028 }
2029
2030 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2031 {
2032         struct e1000_adapter *adapter = netdev_priv(netdev);
2033         struct e1000_hw *hw = &adapter->hw;
2034         u32 vfta, index;
2035
2036         /* don't update vlan cookie if already programmed */
2037         if ((adapter->hw.mng_cookie.status &
2038              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2039             (vid == adapter->mng_vlan_id))
2040                 return;
2041         /* add VID to filter table */
2042         index = (vid >> 5) & 0x7F;
2043         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2044         vfta |= (1 << (vid & 0x1F));
2045         e1000e_write_vfta(hw, index, vfta);
2046 }
2047
2048 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2049 {
2050         struct e1000_adapter *adapter = netdev_priv(netdev);
2051         struct e1000_hw *hw = &adapter->hw;
2052         u32 vfta, index;
2053
2054         if (!test_bit(__E1000_DOWN, &adapter->state))
2055                 e1000_irq_disable(adapter);
2056         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2057
2058         if (!test_bit(__E1000_DOWN, &adapter->state))
2059                 e1000_irq_enable(adapter);
2060
2061         if ((adapter->hw.mng_cookie.status &
2062              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2063             (vid == adapter->mng_vlan_id)) {
2064                 /* release control to f/w */
2065                 e1000_release_hw_control(adapter);
2066                 return;
2067         }
2068
2069         /* remove VID from filter table */
2070         index = (vid >> 5) & 0x7F;
2071         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2072         vfta &= ~(1 << (vid & 0x1F));
2073         e1000e_write_vfta(hw, index, vfta);
2074 }
2075
2076 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2077 {
2078         struct net_device *netdev = adapter->netdev;
2079         u16 vid = adapter->hw.mng_cookie.vlan_id;
2080         u16 old_vid = adapter->mng_vlan_id;
2081
2082         if (!adapter->vlgrp)
2083                 return;
2084
2085         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2086                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2087                 if (adapter->hw.mng_cookie.status &
2088                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2089                         e1000_vlan_rx_add_vid(netdev, vid);
2090                         adapter->mng_vlan_id = vid;
2091                 }
2092
2093                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2094                                 (vid != old_vid) &&
2095                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2096                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2097         } else {
2098                 adapter->mng_vlan_id = vid;
2099         }
2100 }
2101
2102
2103 static void e1000_vlan_rx_register(struct net_device *netdev,
2104                                    struct vlan_group *grp)
2105 {
2106         struct e1000_adapter *adapter = netdev_priv(netdev);
2107         struct e1000_hw *hw = &adapter->hw;
2108         u32 ctrl, rctl;
2109
2110         if (!test_bit(__E1000_DOWN, &adapter->state))
2111                 e1000_irq_disable(adapter);
2112         adapter->vlgrp = grp;
2113
2114         if (grp) {
2115                 /* enable VLAN tag insert/strip */
2116                 ctrl = er32(CTRL);
2117                 ctrl |= E1000_CTRL_VME;
2118                 ew32(CTRL, ctrl);
2119
2120                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2121                         /* enable VLAN receive filtering */
2122                         rctl = er32(RCTL);
2123                         rctl &= ~E1000_RCTL_CFIEN;
2124                         ew32(RCTL, rctl);
2125                         e1000_update_mng_vlan(adapter);
2126                 }
2127         } else {
2128                 /* disable VLAN tag insert/strip */
2129                 ctrl = er32(CTRL);
2130                 ctrl &= ~E1000_CTRL_VME;
2131                 ew32(CTRL, ctrl);
2132
2133                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2134                         if (adapter->mng_vlan_id !=
2135                             (u16)E1000_MNG_VLAN_NONE) {
2136                                 e1000_vlan_rx_kill_vid(netdev,
2137                                                        adapter->mng_vlan_id);
2138                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2139                         }
2140                 }
2141         }
2142
2143         if (!test_bit(__E1000_DOWN, &adapter->state))
2144                 e1000_irq_enable(adapter);
2145 }
2146
2147 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2148 {
2149         u16 vid;
2150
2151         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2152
2153         if (!adapter->vlgrp)
2154                 return;
2155
2156         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2157                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2158                         continue;
2159                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2160         }
2161 }
2162
2163 static void e1000_init_manageability(struct e1000_adapter *adapter)
2164 {
2165         struct e1000_hw *hw = &adapter->hw;
2166         u32 manc, manc2h;
2167
2168         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2169                 return;
2170
2171         manc = er32(MANC);
2172
2173         /*
2174          * enable receiving management packets to the host. this will probably
2175          * generate destination unreachable messages from the host OS, but
2176          * the packets will be handled on SMBUS
2177          */
2178         manc |= E1000_MANC_EN_MNG2HOST;
2179         manc2h = er32(MANC2H);
2180 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2181 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2182         manc2h |= E1000_MNG2HOST_PORT_623;
2183         manc2h |= E1000_MNG2HOST_PORT_664;
2184         ew32(MANC2H, manc2h);
2185         ew32(MANC, manc);
2186 }
2187
2188 /**
2189  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2190  * @adapter: board private structure
2191  *
2192  * Configure the Tx unit of the MAC after a reset.
2193  **/
2194 static void e1000_configure_tx(struct e1000_adapter *adapter)
2195 {
2196         struct e1000_hw *hw = &adapter->hw;
2197         struct e1000_ring *tx_ring = adapter->tx_ring;
2198         u64 tdba;
2199         u32 tdlen, tctl, tipg, tarc;
2200         u32 ipgr1, ipgr2;
2201
2202         /* Setup the HW Tx Head and Tail descriptor pointers */
2203         tdba = tx_ring->dma;
2204         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2205         ew32(TDBAL, (tdba & DMA_32BIT_MASK));
2206         ew32(TDBAH, (tdba >> 32));
2207         ew32(TDLEN, tdlen);
2208         ew32(TDH, 0);
2209         ew32(TDT, 0);
2210         tx_ring->head = E1000_TDH;
2211         tx_ring->tail = E1000_TDT;
2212
2213         /* Set the default values for the Tx Inter Packet Gap timer */
2214         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2215         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2216         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2217
2218         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2219                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2220
2221         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2222         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2223         ew32(TIPG, tipg);
2224
2225         /* Set the Tx Interrupt Delay register */
2226         ew32(TIDV, adapter->tx_int_delay);
2227         /* Tx irq moderation */
2228         ew32(TADV, adapter->tx_abs_int_delay);
2229
2230         /* Program the Transmit Control Register */
2231         tctl = er32(TCTL);
2232         tctl &= ~E1000_TCTL_CT;
2233         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2234                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2235
2236         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2237                 tarc = er32(TARC(0));
2238                 /*
2239                  * set the speed mode bit, we'll clear it if we're not at
2240                  * gigabit link later
2241                  */
2242 #define SPEED_MODE_BIT (1 << 21)
2243                 tarc |= SPEED_MODE_BIT;
2244                 ew32(TARC(0), tarc);
2245         }
2246
2247         /* errata: program both queues to unweighted RR */
2248         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2249                 tarc = er32(TARC(0));
2250                 tarc |= 1;
2251                 ew32(TARC(0), tarc);
2252                 tarc = er32(TARC(1));
2253                 tarc |= 1;
2254                 ew32(TARC(1), tarc);
2255         }
2256
2257         e1000e_config_collision_dist(hw);
2258
2259         /* Setup Transmit Descriptor Settings for eop descriptor */
2260         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2261
2262         /* only set IDE if we are delaying interrupts using the timers */
2263         if (adapter->tx_int_delay)
2264                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2265
2266         /* enable Report Status bit */
2267         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2268
2269         ew32(TCTL, tctl);
2270
2271         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
2272 }
2273
2274 /**
2275  * e1000_setup_rctl - configure the receive control registers
2276  * @adapter: Board private structure
2277  **/
2278 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2279                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2280 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2281 {
2282         struct e1000_hw *hw = &adapter->hw;
2283         u32 rctl, rfctl;
2284         u32 psrctl = 0;
2285         u32 pages = 0;
2286
2287         /* Program MC offset vector base */
2288         rctl = er32(RCTL);
2289         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2290         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2291                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2292                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2293
2294         /* Do not Store bad packets */
2295         rctl &= ~E1000_RCTL_SBP;
2296
2297         /* Enable Long Packet receive */
2298         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2299                 rctl &= ~E1000_RCTL_LPE;
2300         else
2301                 rctl |= E1000_RCTL_LPE;
2302
2303         /* Some systems expect that the CRC is included in SMBUS traffic. The
2304          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2305          * host memory when this is enabled
2306          */
2307         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2308                 rctl |= E1000_RCTL_SECRC;
2309
2310         /* Setup buffer sizes */
2311         rctl &= ~E1000_RCTL_SZ_4096;
2312         rctl |= E1000_RCTL_BSEX;
2313         switch (adapter->rx_buffer_len) {
2314         case 256:
2315                 rctl |= E1000_RCTL_SZ_256;
2316                 rctl &= ~E1000_RCTL_BSEX;
2317                 break;
2318         case 512:
2319                 rctl |= E1000_RCTL_SZ_512;
2320                 rctl &= ~E1000_RCTL_BSEX;
2321                 break;
2322         case 1024:
2323                 rctl |= E1000_RCTL_SZ_1024;
2324                 rctl &= ~E1000_RCTL_BSEX;
2325                 break;
2326         case 2048:
2327         default:
2328                 rctl |= E1000_RCTL_SZ_2048;
2329                 rctl &= ~E1000_RCTL_BSEX;
2330                 break;
2331         case 4096:
2332                 rctl |= E1000_RCTL_SZ_4096;
2333                 break;
2334         case 8192:
2335                 rctl |= E1000_RCTL_SZ_8192;
2336                 break;
2337         case 16384:
2338                 rctl |= E1000_RCTL_SZ_16384;
2339                 break;
2340         }
2341
2342         /*
2343          * 82571 and greater support packet-split where the protocol
2344          * header is placed in skb->data and the packet data is
2345          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2346          * In the case of a non-split, skb->data is linearly filled,
2347          * followed by the page buffers.  Therefore, skb->data is
2348          * sized to hold the largest protocol header.
2349          *
2350          * allocations using alloc_page take too long for regular MTU
2351          * so only enable packet split for jumbo frames
2352          *
2353          * Using pages when the page size is greater than 16k wastes
2354          * a lot of memory, since we allocate 3 pages at all times
2355          * per packet.
2356          */
2357         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2358         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2359             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2360                 adapter->rx_ps_pages = pages;
2361         else
2362                 adapter->rx_ps_pages = 0;
2363
2364         if (adapter->rx_ps_pages) {
2365                 /* Configure extra packet-split registers */
2366                 rfctl = er32(RFCTL);
2367                 rfctl |= E1000_RFCTL_EXTEN;
2368                 /*
2369                  * disable packet split support for IPv6 extension headers,
2370                  * because some malformed IPv6 headers can hang the Rx
2371                  */
2372                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2373                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2374
2375                 ew32(RFCTL, rfctl);
2376
2377                 /* Enable Packet split descriptors */
2378                 rctl |= E1000_RCTL_DTYP_PS;
2379
2380                 psrctl |= adapter->rx_ps_bsize0 >>
2381                         E1000_PSRCTL_BSIZE0_SHIFT;
2382
2383                 switch (adapter->rx_ps_pages) {
2384                 case 3:
2385                         psrctl |= PAGE_SIZE <<
2386                                 E1000_PSRCTL_BSIZE3_SHIFT;
2387                 case 2:
2388                         psrctl |= PAGE_SIZE <<
2389                                 E1000_PSRCTL_BSIZE2_SHIFT;
2390                 case 1:
2391                         psrctl |= PAGE_SIZE >>
2392                                 E1000_PSRCTL_BSIZE1_SHIFT;
2393                         break;
2394                 }
2395
2396                 ew32(PSRCTL, psrctl);
2397         }
2398
2399         ew32(RCTL, rctl);
2400         /* just started the receive unit, no need to restart */
2401         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2402 }
2403
2404 /**
2405  * e1000_configure_rx - Configure Receive Unit after Reset
2406  * @adapter: board private structure
2407  *
2408  * Configure the Rx unit of the MAC after a reset.
2409  **/
2410 static void e1000_configure_rx(struct e1000_adapter *adapter)
2411 {
2412         struct e1000_hw *hw = &adapter->hw;
2413         struct e1000_ring *rx_ring = adapter->rx_ring;
2414         u64 rdba;
2415         u32 rdlen, rctl, rxcsum, ctrl_ext;
2416
2417         if (adapter->rx_ps_pages) {
2418                 /* this is a 32 byte descriptor */
2419                 rdlen = rx_ring->count *
2420                         sizeof(union e1000_rx_desc_packet_split);
2421                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2422                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2423         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2424                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2425                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2426                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2427         } else {
2428                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2429                 adapter->clean_rx = e1000_clean_rx_irq;
2430                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2431         }
2432
2433         /* disable receives while setting up the descriptors */
2434         rctl = er32(RCTL);
2435         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2436         e1e_flush();
2437         msleep(10);
2438
2439         /* set the Receive Delay Timer Register */
2440         ew32(RDTR, adapter->rx_int_delay);
2441
2442         /* irq moderation */
2443         ew32(RADV, adapter->rx_abs_int_delay);
2444         if (adapter->itr_setting != 0)
2445                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2446
2447         ctrl_ext = er32(CTRL_EXT);
2448         /* Reset delay timers after every interrupt */
2449         ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2450         /* Auto-Mask interrupts upon ICR access */
2451         ctrl_ext |= E1000_CTRL_EXT_IAME;
2452         ew32(IAM, 0xffffffff);
2453         ew32(CTRL_EXT, ctrl_ext);
2454         e1e_flush();
2455
2456         /*
2457          * Setup the HW Rx Head and Tail Descriptor Pointers and
2458          * the Base and Length of the Rx Descriptor Ring
2459          */
2460         rdba = rx_ring->dma;
2461         ew32(RDBAL, (rdba & DMA_32BIT_MASK));
2462         ew32(RDBAH, (rdba >> 32));
2463         ew32(RDLEN, rdlen);
2464         ew32(RDH, 0);
2465         ew32(RDT, 0);
2466         rx_ring->head = E1000_RDH;
2467         rx_ring->tail = E1000_RDT;
2468
2469         /* Enable Receive Checksum Offload for TCP and UDP */
2470         rxcsum = er32(RXCSUM);
2471         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2472                 rxcsum |= E1000_RXCSUM_TUOFL;
2473
2474                 /*
2475                  * IPv4 payload checksum for UDP fragments must be
2476                  * used in conjunction with packet-split.
2477                  */
2478                 if (adapter->rx_ps_pages)
2479                         rxcsum |= E1000_RXCSUM_IPPCSE;
2480         } else {
2481                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2482                 /* no need to clear IPPCSE as it defaults to 0 */
2483         }
2484         ew32(RXCSUM, rxcsum);
2485
2486         /*
2487          * Enable early receives on supported devices, only takes effect when
2488          * packet size is equal or larger than the specified value (in 8 byte
2489          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2490          */
2491         if ((adapter->flags & FLAG_HAS_ERT) &&
2492             (adapter->netdev->mtu > ETH_DATA_LEN)) {
2493                 u32 rxdctl = er32(RXDCTL(0));
2494                 ew32(RXDCTL(0), rxdctl | 0x3);
2495                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2496                 /*
2497                  * With jumbo frames and early-receive enabled, excessive
2498                  * C4->C2 latencies result in dropped transactions.
2499                  */
2500                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2501                                           e1000e_driver_name, 55);
2502         } else {
2503                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2504                                           e1000e_driver_name,
2505                                           PM_QOS_DEFAULT_VALUE);
2506         }
2507
2508         /* Enable Receives */
2509         ew32(RCTL, rctl);
2510 }
2511
2512 /**
2513  *  e1000_update_mc_addr_list - Update Multicast addresses
2514  *  @hw: pointer to the HW structure
2515  *  @mc_addr_list: array of multicast addresses to program
2516  *  @mc_addr_count: number of multicast addresses to program
2517  *  @rar_used_count: the first RAR register free to program
2518  *  @rar_count: total number of supported Receive Address Registers
2519  *
2520  *  Updates the Receive Address Registers and Multicast Table Array.
2521  *  The caller must have a packed mc_addr_list of multicast addresses.
2522  *  The parameter rar_count will usually be hw->mac.rar_entry_count
2523  *  unless there are workarounds that change this.  Currently no func pointer
2524  *  exists and all implementations are handled in the generic version of this
2525  *  function.
2526  **/
2527 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2528                                       u32 mc_addr_count, u32 rar_used_count,
2529                                       u32 rar_count)
2530 {
2531         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2532                                         rar_used_count, rar_count);
2533 }
2534
2535 /**
2536  * e1000_set_multi - Multicast and Promiscuous mode set
2537  * @netdev: network interface device structure
2538  *
2539  * The set_multi entry point is called whenever the multicast address
2540  * list or the network interface flags are updated.  This routine is
2541  * responsible for configuring the hardware for proper multicast,
2542  * promiscuous mode, and all-multi behavior.
2543  **/
2544 static void e1000_set_multi(struct net_device *netdev)
2545 {
2546         struct e1000_adapter *adapter = netdev_priv(netdev);
2547         struct e1000_hw *hw = &adapter->hw;
2548         struct e1000_mac_info *mac = &hw->mac;
2549         struct dev_mc_list *mc_ptr;
2550         u8  *mta_list;
2551         u32 rctl;
2552         int i;
2553
2554         /* Check for Promiscuous and All Multicast modes */
2555
2556         rctl = er32(RCTL);
2557
2558         if (netdev->flags & IFF_PROMISC) {
2559                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2560                 rctl &= ~E1000_RCTL_VFE;
2561         } else {
2562                 if (netdev->flags & IFF_ALLMULTI) {
2563                         rctl |= E1000_RCTL_MPE;
2564                         rctl &= ~E1000_RCTL_UPE;
2565                 } else {
2566                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2567                 }
2568                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2569                         rctl |= E1000_RCTL_VFE;
2570         }
2571
2572         ew32(RCTL, rctl);
2573
2574         if (netdev->mc_count) {
2575                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2576                 if (!mta_list)
2577                         return;
2578
2579                 /* prepare a packed array of only addresses. */
2580                 mc_ptr = netdev->mc_list;
2581
2582                 for (i = 0; i < netdev->mc_count; i++) {
2583                         if (!mc_ptr)
2584                                 break;
2585                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2586                                ETH_ALEN);
2587                         mc_ptr = mc_ptr->next;
2588                 }
2589
2590                 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2591                                           mac->rar_entry_count);
2592                 kfree(mta_list);
2593         } else {
2594                 /*
2595                  * if we're called from probe, we might not have
2596                  * anything to do here, so clear out the list
2597                  */
2598                 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2599         }
2600 }
2601
2602 /**
2603  * e1000_configure - configure the hardware for Rx and Tx
2604  * @adapter: private board structure
2605  **/
2606 static void e1000_configure(struct e1000_adapter *adapter)
2607 {
2608         e1000_set_multi(adapter->netdev);
2609
2610         e1000_restore_vlan(adapter);
2611         e1000_init_manageability(adapter);
2612
2613         e1000_configure_tx(adapter);
2614         e1000_setup_rctl(adapter);
2615         e1000_configure_rx(adapter);
2616         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2617 }
2618
2619 /**
2620  * e1000e_power_up_phy - restore link in case the phy was powered down
2621  * @adapter: address of board private structure
2622  *
2623  * The phy may be powered down to save power and turn off link when the
2624  * driver is unloaded and wake on lan is not enabled (among others)
2625  * *** this routine MUST be followed by a call to e1000e_reset ***
2626  **/
2627 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2628 {
2629         u16 mii_reg = 0;
2630
2631         /* Just clear the power down bit to wake the phy back up */
2632         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2633                 /*
2634                  * According to the manual, the phy will retain its
2635                  * settings across a power-down/up cycle
2636                  */
2637                 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2638                 mii_reg &= ~MII_CR_POWER_DOWN;
2639                 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2640         }
2641
2642         adapter->hw.mac.ops.setup_link(&adapter->hw);
2643 }
2644
2645 /**
2646  * e1000_power_down_phy - Power down the PHY
2647  *
2648  * Power down the PHY so no link is implied when interface is down
2649  * The PHY cannot be powered down is management or WoL is active
2650  */
2651 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2652 {
2653         struct e1000_hw *hw = &adapter->hw;
2654         u16 mii_reg;
2655
2656         /* WoL is enabled */
2657         if (adapter->wol)
2658                 return;
2659
2660         /* non-copper PHY? */
2661         if (adapter->hw.phy.media_type != e1000_media_type_copper)
2662                 return;
2663
2664         /* reset is blocked because of a SoL/IDER session */
2665         if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2666                 return;
2667
2668         /* manageability (AMT) is enabled */
2669         if (er32(MANC) & E1000_MANC_SMBUS_EN)
2670                 return;
2671
2672         /* power down the PHY */
2673         e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2674         mii_reg |= MII_CR_POWER_DOWN;
2675         e1e_wphy(hw, PHY_CONTROL, mii_reg);
2676         mdelay(1);
2677 }
2678
2679 /**
2680  * e1000e_reset - bring the hardware into a known good state
2681  *
2682  * This function boots the hardware and enables some settings that
2683  * require a configuration cycle of the hardware - those cannot be
2684  * set/changed during runtime. After reset the device needs to be
2685  * properly configured for Rx, Tx etc.
2686  */
2687 void e1000e_reset(struct e1000_adapter *adapter)
2688 {
2689         struct e1000_mac_info *mac = &adapter->hw.mac;
2690         struct e1000_fc_info *fc = &adapter->hw.fc;
2691         struct e1000_hw *hw = &adapter->hw;
2692         u32 tx_space, min_tx_space, min_rx_space;
2693         u32 pba = adapter->pba;
2694         u16 hwm;
2695
2696         /* reset Packet Buffer Allocation to default */
2697         ew32(PBA, pba);
2698
2699         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2700                 /*
2701                  * To maintain wire speed transmits, the Tx FIFO should be
2702                  * large enough to accommodate two full transmit packets,
2703                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2704                  * the Rx FIFO should be large enough to accommodate at least
2705                  * one full receive packet and is similarly rounded up and
2706                  * expressed in KB.
2707                  */
2708                 pba = er32(PBA);
2709                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2710                 tx_space = pba >> 16;
2711                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2712                 pba &= 0xffff;
2713                 /*
2714                  * the Tx fifo also stores 16 bytes of information about the tx
2715                  * but don't include ethernet FCS because hardware appends it
2716                  */
2717                 min_tx_space = (adapter->max_frame_size +
2718                                 sizeof(struct e1000_tx_desc) -
2719                                 ETH_FCS_LEN) * 2;
2720                 min_tx_space = ALIGN(min_tx_space, 1024);
2721                 min_tx_space >>= 10;
2722                 /* software strips receive CRC, so leave room for it */
2723                 min_rx_space = adapter->max_frame_size;
2724                 min_rx_space = ALIGN(min_rx_space, 1024);
2725                 min_rx_space >>= 10;
2726
2727                 /*
2728                  * If current Tx allocation is less than the min Tx FIFO size,
2729                  * and the min Tx FIFO size is less than the current Rx FIFO
2730                  * allocation, take space away from current Rx allocation
2731                  */
2732                 if ((tx_space < min_tx_space) &&
2733                     ((min_tx_space - tx_space) < pba)) {
2734                         pba -= min_tx_space - tx_space;
2735
2736                         /*
2737                          * if short on Rx space, Rx wins and must trump tx
2738                          * adjustment or use Early Receive if available
2739                          */
2740                         if ((pba < min_rx_space) &&
2741                             (!(adapter->flags & FLAG_HAS_ERT)))
2742                                 /* ERT enabled in e1000_configure_rx */
2743                                 pba = min_rx_space;
2744                 }
2745
2746                 ew32(PBA, pba);
2747         }
2748
2749
2750         /*
2751          * flow control settings
2752          *
2753          * The high water mark must be low enough to fit one full frame
2754          * (or the size used for early receive) above it in the Rx FIFO.
2755          * Set it to the lower of:
2756          * - 90% of the Rx FIFO size, and
2757          * - the full Rx FIFO size minus the early receive size (for parts
2758          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2759          * - the full Rx FIFO size minus one full frame
2760          */
2761         if (adapter->flags & FLAG_HAS_ERT)
2762                 hwm = min(((pba << 10) * 9 / 10),
2763                           ((pba << 10) - (E1000_ERT_2048 << 3)));
2764         else
2765                 hwm = min(((pba << 10) * 9 / 10),
2766                           ((pba << 10) - adapter->max_frame_size));
2767
2768         fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
2769         fc->low_water = fc->high_water - 8;
2770
2771         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2772                 fc->pause_time = 0xFFFF;
2773         else
2774                 fc->pause_time = E1000_FC_PAUSE_TIME;
2775         fc->send_xon = 1;
2776         fc->current_mode = fc->requested_mode;
2777
2778         /* Allow time for pending master requests to run */
2779         mac->ops.reset_hw(hw);
2780
2781         /*
2782          * For parts with AMT enabled, let the firmware know
2783          * that the network interface is in control
2784          */
2785         if (adapter->flags & FLAG_HAS_AMT)
2786                 e1000_get_hw_control(adapter);
2787
2788         ew32(WUC, 0);
2789
2790         if (mac->ops.init_hw(hw))
2791                 e_err("Hardware Error\n");
2792
2793         e1000_update_mng_vlan(adapter);
2794
2795         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2796         ew32(VET, ETH_P_8021Q);
2797
2798         e1000e_reset_adaptive(hw);
2799         e1000_get_phy_info(hw);
2800
2801         if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2802                 u16 phy_data = 0;
2803                 /*
2804                  * speed up time to link by disabling smart power down, ignore
2805                  * the return value of this function because there is nothing
2806                  * different we would do if it failed
2807                  */
2808                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2809                 phy_data &= ~IGP02E1000_PM_SPD;
2810                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2811         }
2812 }
2813
2814 int e1000e_up(struct e1000_adapter *adapter)
2815 {
2816         struct e1000_hw *hw = &adapter->hw;
2817
2818         /* hardware has been reset, we need to reload some things */
2819         e1000_configure(adapter);
2820
2821         clear_bit(__E1000_DOWN, &adapter->state);
2822
2823         napi_enable(&adapter->napi);
2824         if (adapter->msix_entries)
2825                 e1000_configure_msix(adapter);
2826         e1000_irq_enable(adapter);
2827
2828         /* fire a link change interrupt to start the watchdog */
2829         ew32(ICS, E1000_ICS_LSC);
2830         return 0;
2831 }
2832
2833 void e1000e_down(struct e1000_adapter *adapter)
2834 {
2835         struct net_device *netdev = adapter->netdev;
2836         struct e1000_hw *hw = &adapter->hw;
2837         u32 tctl, rctl;
2838
2839         /*
2840          * signal that we're down so the interrupt handler does not
2841          * reschedule our watchdog timer
2842          */
2843         set_bit(__E1000_DOWN, &adapter->state);
2844
2845         /* disable receives in the hardware */
2846         rctl = er32(RCTL);
2847         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2848         /* flush and sleep below */
2849
2850         netif_tx_stop_all_queues(netdev);
2851
2852         /* disable transmits in the hardware */
2853         tctl = er32(TCTL);
2854         tctl &= ~E1000_TCTL_EN;
2855         ew32(TCTL, tctl);
2856         /* flush both disables and wait for them to finish */
2857         e1e_flush();
2858         msleep(10);
2859
2860         napi_disable(&adapter->napi);
2861         e1000_irq_disable(adapter);
2862
2863         del_timer_sync(&adapter->watchdog_timer);
2864         del_timer_sync(&adapter->phy_info_timer);
2865
2866         netdev->tx_queue_len = adapter->tx_queue_len;
2867         netif_carrier_off(netdev);
2868         adapter->link_speed = 0;
2869         adapter->link_duplex = 0;
2870
2871         if (!pci_channel_offline(adapter->pdev))
2872                 e1000e_reset(adapter);
2873         e1000_clean_tx_ring(adapter);
2874         e1000_clean_rx_ring(adapter);
2875
2876         /*
2877          * TODO: for power management, we could drop the link and
2878          * pci_disable_device here.
2879          */
2880 }
2881
2882 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2883 {
2884         might_sleep();
2885         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2886                 msleep(1);
2887         e1000e_down(adapter);
2888         e1000e_up(adapter);
2889         clear_bit(__E1000_RESETTING, &adapter->state);
2890 }
2891
2892 /**
2893  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2894  * @adapter: board private structure to initialize
2895  *
2896  * e1000_sw_init initializes the Adapter private data structure.
2897  * Fields are initialized based on PCI device information and
2898  * OS network device settings (MTU size).
2899  **/
2900 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2901 {
2902         struct net_device *netdev = adapter->netdev;
2903
2904         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2905         adapter->rx_ps_bsize0 = 128;
2906         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2907         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2908
2909         e1000e_set_interrupt_capability(adapter);
2910
2911         if (e1000_alloc_queues(adapter))
2912                 return -ENOMEM;
2913
2914         /* Explicitly disable IRQ since the NIC can be in any state. */
2915         e1000_irq_disable(adapter);
2916
2917         set_bit(__E1000_DOWN, &adapter->state);
2918         return 0;
2919 }
2920
2921 /**
2922  * e1000_intr_msi_test - Interrupt Handler
2923  * @irq: interrupt number
2924  * @data: pointer to a network interface device structure
2925  **/
2926 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2927 {
2928         struct net_device *netdev = data;
2929         struct e1000_adapter *adapter = netdev_priv(netdev);
2930         struct e1000_hw *hw = &adapter->hw;
2931         u32 icr = er32(ICR);
2932
2933         e_dbg("%s: icr is %08X\n", netdev->name, icr);
2934         if (icr & E1000_ICR_RXSEQ) {
2935                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2936                 wmb();
2937         }
2938
2939         return IRQ_HANDLED;
2940 }
2941
2942 /**
2943  * e1000_test_msi_interrupt - Returns 0 for successful test
2944  * @adapter: board private struct
2945  *
2946  * code flow taken from tg3.c
2947  **/
2948 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2949 {
2950         struct net_device *netdev = adapter->netdev;
2951         struct e1000_hw *hw = &adapter->hw;
2952         int err;
2953
2954         /* poll_enable hasn't been called yet, so don't need disable */
2955         /* clear any pending events */
2956         er32(ICR);
2957
2958         /* free the real vector and request a test handler */
2959         e1000_free_irq(adapter);
2960         e1000e_reset_interrupt_capability(adapter);
2961
2962         /* Assume that the test fails, if it succeeds then the test
2963          * MSI irq handler will unset this flag */
2964         adapter->flags |= FLAG_MSI_TEST_FAILED;
2965
2966         err = pci_enable_msi(adapter->pdev);
2967         if (err)
2968                 goto msi_test_failed;
2969
2970         err = request_irq(adapter->pdev->irq, &e1000_intr_msi_test, 0,
2971                           netdev->name, netdev);
2972         if (err) {
2973                 pci_disable_msi(adapter->pdev);
2974                 goto msi_test_failed;
2975         }
2976
2977         wmb();
2978
2979         e1000_irq_enable(adapter);
2980
2981         /* fire an unusual interrupt on the test handler */
2982         ew32(ICS, E1000_ICS_RXSEQ);
2983         e1e_flush();
2984         msleep(50);
2985
2986         e1000_irq_disable(adapter);
2987
2988         rmb();
2989
2990         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
2991                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2992                 err = -EIO;
2993                 e_info("MSI interrupt test failed!\n");
2994         }
2995
2996         free_irq(adapter->pdev->irq, netdev);
2997         pci_disable_msi(adapter->pdev);
2998
2999         if (err == -EIO)
3000                 goto msi_test_failed;
3001
3002         /* okay so the test worked, restore settings */
3003         e_dbg("%s: MSI interrupt test succeeded!\n", netdev->name);
3004 msi_test_failed:
3005         e1000e_set_interrupt_capability(adapter);
3006         e1000_request_irq(adapter);
3007         return err;
3008 }
3009
3010 /**
3011  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3012  * @adapter: board private struct
3013  *
3014  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3015  **/
3016 static int e1000_test_msi(struct e1000_adapter *adapter)
3017 {
3018         int err;
3019         u16 pci_cmd;
3020
3021         if (!(adapter->flags & FLAG_MSI_ENABLED))
3022                 return 0;
3023
3024         /* disable SERR in case the MSI write causes a master abort */
3025         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3026         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3027                               pci_cmd & ~PCI_COMMAND_SERR);
3028
3029         err = e1000_test_msi_interrupt(adapter);
3030
3031         /* restore previous setting of command word */
3032         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3033
3034         /* success ! */
3035         if (!err)
3036                 return 0;
3037
3038         /* EIO means MSI test failed */
3039         if (err != -EIO)
3040                 return err;
3041
3042         /* back to INTx mode */
3043         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3044
3045         e1000_free_irq(adapter);
3046
3047         err = e1000_request_irq(adapter);
3048
3049         return err;
3050 }
3051
3052 /**
3053  * e1000_open - Called when a network interface is made active
3054  * @netdev: network interface device structure
3055  *
3056  * Returns 0 on success, negative value on failure
3057  *
3058  * The open entry point is called when a network interface is made
3059  * active by the system (IFF_UP).  At this point all resources needed
3060  * for transmit and receive operations are allocated, the interrupt
3061  * handler is registered with the OS, the watchdog timer is started,
3062  * and the stack is notified that the interface is ready.
3063  **/
3064 static int e1000_open(struct net_device *netdev)
3065 {
3066         struct e1000_adapter *adapter = netdev_priv(netdev);
3067         struct e1000_hw *hw = &adapter->hw;
3068         int err;
3069
3070         /* disallow open during test */
3071         if (test_bit(__E1000_TESTING, &adapter->state))
3072                 return -EBUSY;
3073
3074         /* allocate transmit descriptors */
3075         err = e1000e_setup_tx_resources(adapter);
3076         if (err)
3077                 goto err_setup_tx;
3078
3079         /* allocate receive descriptors */
3080         err = e1000e_setup_rx_resources(adapter);
3081         if (err)
3082                 goto err_setup_rx;
3083
3084         e1000e_power_up_phy(adapter);
3085
3086         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3087         if ((adapter->hw.mng_cookie.status &
3088              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3089                 e1000_update_mng_vlan(adapter);
3090
3091         /*
3092          * If AMT is enabled, let the firmware know that the network
3093          * interface is now open
3094          */
3095         if (adapter->flags & FLAG_HAS_AMT)
3096                 e1000_get_hw_control(adapter);
3097
3098         /*
3099          * before we allocate an interrupt, we must be ready to handle it.
3100          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3101          * as soon as we call pci_request_irq, so we have to setup our
3102          * clean_rx handler before we do so.
3103          */
3104         e1000_configure(adapter);
3105
3106         err = e1000_request_irq(adapter);
3107         if (err)
3108                 goto err_req_irq;
3109
3110         /*
3111          * Work around PCIe errata with MSI interrupts causing some chipsets to
3112          * ignore e1000e MSI messages, which means we need to test our MSI
3113          * interrupt now
3114          */
3115         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3116                 err = e1000_test_msi(adapter);
3117                 if (err) {
3118                         e_err("Interrupt allocation failed\n");
3119                         goto err_req_irq;
3120                 }
3121         }
3122
3123         /* From here on the code is the same as e1000e_up() */
3124         clear_bit(__E1000_DOWN, &adapter->state);
3125
3126         napi_enable(&adapter->napi);
3127
3128         e1000_irq_enable(adapter);
3129
3130         netif_tx_start_all_queues(netdev);
3131
3132         /* fire a link status change interrupt to start the watchdog */
3133         ew32(ICS, E1000_ICS_LSC);
3134
3135         return 0;
3136
3137 err_req_irq:
3138         e1000_release_hw_control(adapter);
3139         e1000_power_down_phy(adapter);
3140         e1000e_free_rx_resources(adapter);
3141 err_setup_rx:
3142         e1000e_free_tx_resources(adapter);
3143 err_setup_tx:
3144         e1000e_reset(adapter);
3145
3146         return err;
3147 }
3148
3149 /**
3150  * e1000_close - Disables a network interface
3151  * @netdev: network interface device structure
3152  *
3153  * Returns 0, this is not allowed to fail
3154  *
3155  * The close entry point is called when an interface is de-activated
3156  * by the OS.  The hardware is still under the drivers control, but
3157  * needs to be disabled.  A global MAC reset is issued to stop the
3158  * hardware, and all transmit and receive resources are freed.
3159  **/
3160 static int e1000_close(struct net_device *netdev)
3161 {
3162         struct e1000_adapter *adapter = netdev_priv(netdev);
3163
3164         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3165         e1000e_down(adapter);
3166         e1000_power_down_phy(adapter);
3167         e1000_free_irq(adapter);
3168
3169         e1000e_free_tx_resources(adapter);
3170         e1000e_free_rx_resources(adapter);
3171
3172         /*
3173          * kill manageability vlan ID if supported, but not if a vlan with
3174          * the same ID is registered on the host OS (let 8021q kill it)
3175          */
3176         if ((adapter->hw.mng_cookie.status &
3177                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3178              !(adapter->vlgrp &&
3179                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3180                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3181
3182         /*
3183          * If AMT is enabled, let the firmware know that the network
3184          * interface is now closed
3185          */
3186         if (adapter->flags & FLAG_HAS_AMT)
3187                 e1000_release_hw_control(adapter);
3188
3189         return 0;
3190 }
3191 /**
3192  * e1000_set_mac - Change the Ethernet Address of the NIC
3193  * @netdev: network interface device structure
3194  * @p: pointer to an address structure
3195  *
3196  * Returns 0 on success, negative on failure
3197  **/
3198 static int e1000_set_mac(struct net_device *netdev, void *p)
3199 {
3200         struct e1000_adapter *adapter = netdev_priv(netdev);
3201         struct sockaddr *addr = p;
3202
3203         if (!is_valid_ether_addr(addr->sa_data))
3204                 return -EADDRNOTAVAIL;
3205
3206         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3207         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3208
3209         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3210
3211         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3212                 /* activate the work around */
3213                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3214
3215                 /*
3216                  * Hold a copy of the LAA in RAR[14] This is done so that
3217                  * between the time RAR[0] gets clobbered  and the time it
3218                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3219                  * of the RARs and no incoming packets directed to this port
3220                  * are dropped. Eventually the LAA will be in RAR[0] and
3221                  * RAR[14]
3222                  */
3223                 e1000e_rar_set(&adapter->hw,
3224                               adapter->hw.mac.addr,
3225                               adapter->hw.mac.rar_entry_count - 1);
3226         }
3227
3228         return 0;
3229 }
3230
3231 /**
3232  * e1000e_update_phy_task - work thread to update phy
3233  * @work: pointer to our work struct
3234  *
3235  * this worker thread exists because we must acquire a
3236  * semaphore to read the phy, which we could msleep while
3237  * waiting for it, and we can't msleep in a timer.
3238  **/
3239 static void e1000e_update_phy_task(struct work_struct *work)
3240 {
3241         struct e1000_adapter *adapter = container_of(work,
3242                                         struct e1000_adapter, update_phy_task);
3243         e1000_get_phy_info(&adapter->hw);
3244 }
3245
3246 /*
3247  * Need to wait a few seconds after link up to get diagnostic information from
3248  * the phy
3249  */
3250 static void e1000_update_phy_info(unsigned long data)
3251 {
3252         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3253         schedule_work(&adapter->update_phy_task);
3254 }
3255
3256 /**
3257  * e1000e_update_stats - Update the board statistics counters
3258  * @adapter: board private structure
3259  **/
3260 void e1000e_update_stats(struct e1000_adapter *adapter)
3261 {
3262         struct e1000_hw *hw = &adapter->hw;
3263         struct pci_dev *pdev = adapter->pdev;
3264
3265         /*
3266          * Prevent stats update while adapter is being reset, or if the pci
3267          * connection is down.
3268          */
3269         if (adapter->link_speed == 0)
3270                 return;
3271         if (pci_channel_offline(pdev))
3272                 return;
3273
3274         adapter->stats.crcerrs += er32(CRCERRS);
3275         adapter->stats.gprc += er32(GPRC);
3276         adapter->stats.gorc += er32(GORCL);
3277         er32(GORCH); /* Clear gorc */
3278         adapter->stats.bprc += er32(BPRC);
3279         adapter->stats.mprc += er32(MPRC);
3280         adapter->stats.roc += er32(ROC);
3281
3282         adapter->stats.mpc += er32(MPC);
3283         adapter->stats.scc += er32(SCC);
3284         adapter->stats.ecol += er32(ECOL);
3285         adapter->stats.mcc += er32(MCC);
3286         adapter->stats.latecol += er32(LATECOL);
3287         adapter->stats.dc += er32(DC);
3288         adapter->stats.xonrxc += er32(XONRXC);
3289         adapter->stats.xontxc += er32(XONTXC);
3290         adapter->stats.xoffrxc += er32(XOFFRXC);
3291         adapter->stats.xofftxc += er32(XOFFTXC);
3292         adapter->stats.gptc += er32(GPTC);
3293         adapter->stats.gotc += er32(GOTCL);
3294         er32(GOTCH); /* Clear gotc */
3295         adapter->stats.rnbc += er32(RNBC);
3296         adapter->stats.ruc += er32(RUC);
3297
3298         adapter->stats.mptc += er32(MPTC);
3299         adapter->stats.bptc += er32(BPTC);
3300
3301         /* used for adaptive IFS */
3302
3303         hw->mac.tx_packet_delta = er32(TPT);
3304         adapter->stats.tpt += hw->mac.tx_packet_delta;
3305         hw->mac.collision_delta = er32(COLC);
3306         adapter->stats.colc += hw->mac.collision_delta;
3307
3308         adapter->stats.algnerrc += er32(ALGNERRC);
3309         adapter->stats.rxerrc += er32(RXERRC);
3310         if ((hw->mac.type != e1000_82574) && (hw->mac.type != e1000_82583))
3311                 adapter->stats.tncrs += er32(TNCRS);
3312         adapter->stats.cexterr += er32(CEXTERR);
3313         adapter->stats.tsctc += er32(TSCTC);
3314         adapter->stats.tsctfc += er32(TSCTFC);
3315
3316         /* Fill out the OS statistics structure */
3317         adapter->net_stats.multicast = adapter->stats.mprc;
3318         adapter->net_stats.collisions = adapter->stats.colc;
3319
3320         /* Rx Errors */
3321
3322         /*
3323          * RLEC on some newer hardware can be incorrect so build
3324          * our own version based on RUC and ROC
3325          */
3326         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3327                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3328                 adapter->stats.ruc + adapter->stats.roc +
3329                 adapter->stats.cexterr;
3330         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3331                                               adapter->stats.roc;
3332         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3333         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3334         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3335
3336         /* Tx Errors */
3337         adapter->net_stats.tx_errors = adapter->stats.ecol +
3338                                        adapter->stats.latecol;
3339         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3340         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3341         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3342
3343         /* Tx Dropped needs to be maintained elsewhere */
3344
3345         /* Management Stats */
3346         adapter->stats.mgptc += er32(MGTPTC);
3347         adapter->stats.mgprc += er32(MGTPRC);
3348         adapter->stats.mgpdc += er32(MGTPDC);
3349 }
3350
3351 /**
3352  * e1000_phy_read_status - Update the PHY register status snapshot
3353  * @adapter: board private structure
3354  **/
3355 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3356 {
3357         struct e1000_hw *hw = &adapter->hw;
3358         struct e1000_phy_regs *phy = &adapter->phy_regs;
3359         int ret_val;
3360
3361         if ((er32(STATUS) & E1000_STATUS_LU) &&
3362             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3363                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3364                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3365                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3366                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3367                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3368                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3369                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3370                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3371                 if (ret_val)
3372                         e_warn("Error reading PHY register\n");
3373         } else {
3374                 /*
3375                  * Do not read PHY registers if link is not up
3376                  * Set values to typical power-on defaults
3377                  */
3378                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3379                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3380                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3381                              BMSR_ERCAP);
3382                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3383                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3384                 phy->lpa = 0;
3385                 phy->expansion = EXPANSION_ENABLENPAGE;
3386                 phy->ctrl1000 = ADVERTISE_1000FULL;
3387                 phy->stat1000 = 0;
3388                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3389         }
3390 }
3391
3392 static void e1000_print_link_info(struct e1000_adapter *adapter)
3393 {
3394         struct e1000_hw *hw = &adapter->hw;
3395         u32 ctrl = er32(CTRL);
3396
3397         /* Link status message must follow this format for user tools */
3398         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3399                "Flow Control: %s\n",
3400                adapter->netdev->name,
3401                adapter->link_speed,
3402                (adapter->link_duplex == FULL_DUPLEX) ?
3403                                 "Full Duplex" : "Half Duplex",
3404                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3405                                 "RX/TX" :
3406                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3407                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3408 }
3409
3410 bool e1000_has_link(struct e1000_adapter *adapter)
3411 {
3412         struct e1000_hw *hw = &adapter->hw;
3413         bool link_active = 0;
3414         s32 ret_val = 0;
3415
3416         /*
3417          * get_link_status is set on LSC (link status) interrupt or
3418          * Rx sequence error interrupt.  get_link_status will stay
3419          * false until the check_for_link establishes link
3420          * for copper adapters ONLY
3421          */
3422         switch (hw->phy.media_type) {
3423         case e1000_media_type_copper:
3424                 if (hw->mac.get_link_status) {
3425                         ret_val = hw->mac.ops.check_for_link(hw);
3426                         link_active = !hw->mac.get_link_status;
3427                 } else {
3428                         link_active = 1;
3429                 }
3430                 break;
3431         case e1000_media_type_fiber:
3432                 ret_val = hw->mac.ops.check_for_link(hw);
3433                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3434                 break;
3435         case e1000_media_type_internal_serdes:
3436                 ret_val = hw->mac.ops.check_for_link(hw);
3437                 link_active = adapter->hw.mac.serdes_has_link;
3438                 break;
3439         default:
3440         case e1000_media_type_unknown:
3441                 break;
3442         }
3443
3444         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3445             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3446                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3447                 e_info("Gigabit has been disabled, downgrading speed\n");
3448         }
3449
3450         return link_active;
3451 }
3452
3453 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3454 {
3455         /* make sure the receive unit is started */
3456         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3457             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3458                 struct e1000_hw *hw = &adapter->hw;
3459                 u32 rctl = er32(RCTL);
3460                 ew32(RCTL, rctl | E1000_RCTL_EN);
3461                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3462         }
3463 }
3464
3465 /**
3466  * e1000_watchdog - Timer Call-back
3467  * @data: pointer to adapter cast into an unsigned long
3468  **/
3469 static void e1000_watchdog(unsigned long data)
3470 {
3471         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3472
3473         /* Do the rest outside of interrupt context */
3474         schedule_work(&adapter->watchdog_task);
3475
3476         /* TODO: make this use queue_delayed_work() */
3477 }
3478
3479 static void e1000_watchdog_task(struct work_struct *work)
3480 {
3481         struct e1000_adapter *adapter = container_of(work,
3482                                         struct e1000_adapter, watchdog_task);
3483         struct net_device *netdev = adapter->netdev;
3484         struct e1000_mac_info *mac = &adapter->hw.mac;
3485         struct e1000_phy_info *phy = &adapter->hw.phy;
3486         struct e1000_ring *tx_ring = adapter->tx_ring;
3487         struct e1000_hw *hw = &adapter->hw;
3488         u32 link, tctl;
3489         int tx_pending = 0;
3490
3491         link = e1000_has_link(adapter);
3492         if ((netif_carrier_ok(netdev)) && link) {
3493                 e1000e_enable_receives(adapter);
3494                 goto link_up;
3495         }
3496
3497         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3498             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3499                 e1000_update_mng_vlan(adapter);
3500
3501         if (link) {
3502                 if (!netif_carrier_ok(netdev)) {
3503                         bool txb2b = 1;
3504                         /* update snapshot of PHY registers on LSC */
3505                         e1000_phy_read_status(adapter);
3506                         mac->ops.get_link_up_info(&adapter->hw,
3507                                                    &adapter->link_speed,
3508                                                    &adapter->link_duplex);
3509                         e1000_print_link_info(adapter);
3510                         /*
3511                          * On supported PHYs, check for duplex mismatch only
3512                          * if link has autonegotiated at 10/100 half
3513                          */
3514                         if ((hw->phy.type == e1000_phy_igp_3 ||
3515                              hw->phy.type == e1000_phy_bm) &&
3516                             (hw->mac.autoneg == true) &&
3517                             (adapter->link_speed == SPEED_10 ||
3518                              adapter->link_speed == SPEED_100) &&
3519                             (adapter->link_duplex == HALF_DUPLEX)) {
3520                                 u16 autoneg_exp;
3521
3522                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3523
3524                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3525                                         e_info("Autonegotiated half duplex but"
3526                                                " link partner cannot autoneg. "
3527                                                " Try forcing full duplex if "
3528                                                "link gets many collisions.\n");
3529                         }
3530
3531                         /*
3532                          * tweak tx_queue_len according to speed/duplex
3533                          * and adjust the timeout factor
3534                          */
3535                         netdev->tx_queue_len = adapter->tx_queue_len;
3536                         adapter->tx_timeout_factor = 1;
3537                         switch (adapter->link_speed) {
3538                         case SPEED_10:
3539                                 txb2b = 0;
3540                                 netdev->tx_queue_len = 10;
3541                                 adapter->tx_timeout_factor = 16;
3542                                 break;
3543                         case SPEED_100:
3544                                 txb2b = 0;
3545                                 netdev->tx_queue_len = 100;
3546                                 /* maybe add some timeout factor ? */
3547                                 break;
3548                         }
3549
3550                         /*
3551                          * workaround: re-program speed mode bit after
3552                          * link-up event
3553                          */
3554                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3555                             !txb2b) {
3556                                 u32 tarc0;
3557                                 tarc0 = er32(TARC(0));
3558                                 tarc0 &= ~SPEED_MODE_BIT;
3559                                 ew32(TARC(0), tarc0);
3560                         }
3561
3562                         /*
3563                          * disable TSO for pcie and 10/100 speeds, to avoid
3564                          * some hardware issues
3565                          */
3566                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3567                                 switch (adapter->link_speed) {
3568                                 case SPEED_10:
3569                                 case SPEED_100:
3570                                         e_info("10/100 speed: disabling TSO\n");
3571                                         netdev->features &= ~NETIF_F_TSO;
3572                                         netdev->features &= ~NETIF_F_TSO6;
3573                                         break;
3574                                 case SPEED_1000:
3575                                         netdev->features |= NETIF_F_TSO;
3576                                         netdev->features |= NETIF_F_TSO6;
3577                                         break;
3578                                 default:
3579                                         /* oops */
3580                                         break;
3581                                 }
3582                         }
3583
3584                         /*
3585                          * enable transmits in the hardware, need to do this
3586                          * after setting TARC(0)
3587                          */
3588                         tctl = er32(TCTL);
3589                         tctl |= E1000_TCTL_EN;
3590                         ew32(TCTL, tctl);
3591
3592                         /*
3593                          * Perform any post-link-up configuration before
3594                          * reporting link up.
3595                          */
3596                         if (phy->ops.cfg_on_link_up)
3597                                 phy->ops.cfg_on_link_up(hw);
3598
3599                         netif_carrier_on(netdev);
3600                         netif_tx_wake_all_queues(netdev);
3601
3602                         if (!test_bit(__E1000_DOWN, &adapter->state))
3603                                 mod_timer(&adapter->phy_info_timer,
3604                                           round_jiffies(jiffies + 2 * HZ));
3605                 }
3606         } else {
3607                 if (netif_carrier_ok(netdev)) {
3608                         adapter->link_speed = 0;
3609                         adapter->link_duplex = 0;
3610                         /* Link status message must follow this format */
3611                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
3612                                adapter->netdev->name);
3613                         netif_carrier_off(netdev);
3614                         netif_tx_stop_all_queues(netdev);
3615                         if (!test_bit(__E1000_DOWN, &adapter->state))
3616                                 mod_timer(&adapter->phy_info_timer,
3617                                           round_jiffies(jiffies + 2 * HZ));
3618
3619                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3620                                 schedule_work(&adapter->reset_task);
3621                 }
3622         }
3623
3624 link_up:
3625         e1000e_update_stats(adapter);
3626
3627         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3628         adapter->tpt_old = adapter->stats.tpt;
3629         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3630         adapter->colc_old = adapter->stats.colc;
3631
3632         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3633         adapter->gorc_old = adapter->stats.gorc;
3634         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3635         adapter->gotc_old = adapter->stats.gotc;
3636
3637         e1000e_update_adaptive(&adapter->hw);
3638
3639         if (!netif_carrier_ok(netdev)) {
3640                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3641                                tx_ring->count);
3642                 if (tx_pending) {
3643                         /*
3644                          * We've lost link, so the controller stops DMA,
3645                          * but we've got queued Tx work that's never going
3646                          * to get done, so reset controller to flush Tx.
3647                          * (Do the reset outside of interrupt context).
3648                          */
3649                         adapter->tx_timeout_count++;
3650                         schedule_work(&adapter->reset_task);
3651                 }
3652         }
3653
3654         /* Cause software interrupt to ensure Rx ring is cleaned */
3655         if (adapter->msix_entries)
3656                 ew32(ICS, adapter->rx_ring->ims_val);
3657         else
3658                 ew32(ICS, E1000_ICS_RXDMT0);
3659
3660         /* Force detection of hung controller every watchdog period */
3661         adapter->detect_tx_hung = 1;
3662
3663         /*
3664          * With 82571 controllers, LAA may be overwritten due to controller
3665          * reset from the other port. Set the appropriate LAA in RAR[0]
3666          */
3667         if (e1000e_get_laa_state_82571(hw))
3668                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3669
3670         /* Reset the timer */
3671         if (!test_bit(__E1000_DOWN, &adapter->state))
3672                 mod_timer(&adapter->watchdog_timer,
3673                           round_jiffies(jiffies + 2 * HZ));
3674 }
3675
3676 #define E1000_TX_FLAGS_CSUM             0x00000001
3677 #define E1000_TX_FLAGS_VLAN             0x00000002
3678 #define E1000_TX_FLAGS_TSO              0x00000004
3679 #define E1000_TX_FLAGS_IPV4             0x00000008
3680 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3681 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3682
3683 static int e1000_tso(struct e1000_adapter *adapter,
3684                      struct sk_buff *skb)
3685 {
3686         struct e1000_ring *tx_ring = adapter->tx_ring;
3687         struct e1000_context_desc *context_desc;
3688         struct e1000_buffer *buffer_info;
3689         unsigned int i;
3690         u32 cmd_length = 0;
3691         u16 ipcse = 0, tucse, mss;
3692         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3693         int err;
3694
3695         if (skb_is_gso(skb)) {
3696                 if (skb_header_cloned(skb)) {
3697                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3698                         if (err)
3699                                 return err;
3700                 }
3701
3702                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3703                 mss = skb_shinfo(skb)->gso_size;
3704                 if (skb->protocol == htons(ETH_P_IP)) {
3705                         struct iphdr *iph = ip_hdr(skb);
3706                         iph->tot_len = 0;
3707                         iph->check = 0;
3708                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3709                                                                  iph->daddr, 0,
3710                                                                  IPPROTO_TCP,
3711                                                                  0);
3712                         cmd_length = E1000_TXD_CMD_IP;
3713                         ipcse = skb_transport_offset(skb) - 1;
3714                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3715                         ipv6_hdr(skb)->payload_len = 0;
3716                         tcp_hdr(skb)->check =
3717                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3718                                                  &ipv6_hdr(skb)->daddr,
3719                                                  0, IPPROTO_TCP, 0);
3720                         ipcse = 0;
3721                 }
3722                 ipcss = skb_network_offset(skb);
3723                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3724                 tucss = skb_transport_offset(skb);
3725                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3726                 tucse = 0;
3727
3728                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3729                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3730
3731                 i = tx_ring->next_to_use;
3732                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3733                 buffer_info = &tx_ring->buffer_info[i];
3734
3735                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3736                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3737                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3738                 context_desc->upper_setup.tcp_fields.tucss = tucss;
3739                 context_desc->upper_setup.tcp_fields.tucso = tucso;
3740                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3741                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3742                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3743                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3744
3745                 buffer_info->time_stamp = jiffies;
3746                 buffer_info->next_to_watch = i;
3747
3748                 i++;
3749                 if (i == tx_ring->count)
3750                         i = 0;
3751                 tx_ring->next_to_use = i;
3752
3753                 return 1;
3754         }
3755
3756         return 0;
3757 }
3758
3759 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3760 {
3761         struct e1000_ring *tx_ring = adapter->tx_ring;
3762         struct e1000_context_desc *context_desc;
3763         struct e1000_buffer *buffer_info;
3764         unsigned int i;
3765         u8 css;
3766         u32 cmd_len = E1000_TXD_CMD_DEXT;
3767         __be16 protocol;
3768
3769         if (skb->ip_summed != CHECKSUM_PARTIAL)
3770                 return 0;
3771
3772         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
3773                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
3774         else
3775                 protocol = skb->protocol;
3776
3777         switch (skb->protocol) {
3778         case cpu_to_be16(ETH_P_IP):
3779                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3780                         cmd_len |= E1000_TXD_CMD_TCP;
3781                 break;
3782         case cpu_to_be16(ETH_P_IPV6):
3783                 /* XXX not handling all IPV6 headers */
3784                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3785                         cmd_len |= E1000_TXD_CMD_TCP;
3786                 break;
3787         default:
3788                 if (unlikely(net_ratelimit()))
3789                         e_warn("checksum_partial proto=%x!\n",
3790                                be16_to_cpu(protocol));
3791                 break;
3792         }
3793
3794         css = skb_transport_offset(skb);
3795
3796         i = tx_ring->next_to_use;
3797         buffer_info = &tx_ring->buffer_info[i];
3798         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3799
3800         context_desc->lower_setup.ip_config = 0;
3801         context_desc->upper_setup.tcp_fields.tucss = css;
3802         context_desc->upper_setup.tcp_fields.tucso =
3803                                 css + skb->csum_offset;
3804         context_desc->upper_setup.tcp_fields.tucse = 0;
3805         context_desc->tcp_seg_setup.data = 0;
3806         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
3807
3808         buffer_info->time_stamp = jiffies;
3809         buffer_info->next_to_watch = i;
3810
3811         i++;
3812         if (i == tx_ring->count)
3813                 i = 0;
3814         tx_ring->next_to_use = i;
3815
3816         return 1;
3817 }
3818
3819 #define E1000_MAX_PER_TXD       8192
3820 #define E1000_MAX_TXD_PWR       12
3821
3822 static int e1000_tx_map(struct e1000_adapter *adapter,
3823                         struct sk_buff *skb, unsigned int first,
3824                         unsigned int max_per_txd, unsigned int nr_frags,
3825                         unsigned int mss)
3826 {
3827         struct e1000_ring *tx_ring = adapter->tx_ring;
3828         struct e1000_buffer *buffer_info;
3829         unsigned int len = skb_headlen(skb);
3830         unsigned int offset, size, count = 0, i;
3831         unsigned int f;
3832         dma_addr_t *map;
3833
3834         i = tx_ring->next_to_use;
3835
3836         if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
3837                 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3838                 adapter->tx_dma_failed++;
3839                 return 0;
3840         }
3841
3842         map = skb_shinfo(skb)->dma_maps;
3843         offset = 0;
3844
3845         while (len) {
3846                 buffer_info = &tx_ring->buffer_info[i];
3847                 size = min(len, max_per_txd);
3848
3849                 buffer_info->length = size;
3850                 buffer_info->time_stamp = jiffies;
3851                 buffer_info->next_to_watch = i;
3852                 buffer_info->dma = map[0] + offset;
3853                 count++;
3854
3855                 len -= size;
3856                 offset += size;
3857
3858                 if (len) {
3859                         i++;
3860                         if (i == tx_ring->count)
3861                                 i = 0;
3862                 }
3863         }
3864
3865         for (f = 0; f < nr_frags; f++) {
3866                 struct skb_frag_struct *frag;
3867
3868                 frag = &skb_shinfo(skb)->frags[f];
3869                 len = frag->size;
3870                 offset = 0;
3871
3872                 while (len) {
3873                         i++;
3874                         if (i == tx_ring->count)
3875                                 i = 0;
3876
3877                         buffer_info = &tx_ring->buffer_info[i];
3878                         size = min(len, max_per_txd);
3879
3880                         buffer_info->length = size;
3881                         buffer_info->time_stamp = jiffies;
3882                         buffer_info->next_to_watch = i;
3883                         buffer_info->dma = map[f + 1] + offset;
3884
3885                         len -= size;
3886                         offset += size;
3887                         count++;
3888                 }
3889         }
3890
3891         tx_ring->buffer_info[i].skb = skb;
3892         tx_ring->buffer_info[first].next_to_watch = i;
3893
3894         return count;
3895 }
3896
3897 static void e1000_tx_queue(struct e1000_adapter *adapter,
3898                            int tx_flags, int count)
3899 {
3900         struct e1000_ring *tx_ring = adapter->tx_ring;
3901         struct e1000_tx_desc *tx_desc = NULL;
3902         struct e1000_buffer *buffer_info;
3903         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3904         unsigned int i;
3905
3906         if (tx_flags & E1000_TX_FLAGS_TSO) {
3907                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3908                              E1000_TXD_CMD_TSE;
3909                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3910
3911                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3912                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3913         }
3914
3915         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3916                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3917                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3918         }
3919
3920         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3921                 txd_lower |= E1000_TXD_CMD_VLE;
3922                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3923         }
3924
3925         i = tx_ring->next_to_use;
3926
3927         while (count--) {
3928                 buffer_info = &tx_ring->buffer_info[i];
3929                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3930                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3931                 tx_desc->lower.data =
3932                         cpu_to_le32(txd_lower | buffer_info->length);
3933                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3934
3935                 i++;
3936                 if (i == tx_ring->count)
3937                         i = 0;
3938         }
3939
3940         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3941
3942         /*
3943          * Force memory writes to complete before letting h/w
3944          * know there are new descriptors to fetch.  (Only
3945          * applicable for weak-ordered memory model archs,
3946          * such as IA-64).
3947          */
3948         wmb();
3949
3950         tx_ring->next_to_use = i;
3951         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3952         /*
3953          * we need this if more than one processor can write to our tail
3954          * at a time, it synchronizes IO on IA64/Altix systems
3955          */
3956         mmiowb();
3957 }
3958
3959 #define MINIMUM_DHCP_PACKET_SIZE 282
3960 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3961                                     struct sk_buff *skb)
3962 {
3963         struct e1000_hw *hw =  &adapter->hw;
3964         u16 length, offset;
3965
3966         if (vlan_tx_tag_present(skb)) {
3967                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3968                     && (adapter->hw.mng_cookie.status &
3969                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3970                         return 0;
3971         }
3972
3973         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3974                 return 0;
3975
3976         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3977                 return 0;
3978
3979         {
3980                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3981                 struct udphdr *udp;
3982
3983                 if (ip->protocol != IPPROTO_UDP)
3984                         return 0;
3985
3986                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3987                 if (ntohs(udp->dest) != 67)
3988                         return 0;
3989
3990                 offset = (u8 *)udp + 8 - skb->data;
3991                 length = skb->len - offset;
3992                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
3993         }
3994
3995         return 0;
3996 }
3997
3998 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3999 {
4000         struct e1000_adapter *adapter = netdev_priv(netdev);
4001
4002         netif_stop_queue(netdev);
4003         /*
4004          * Herbert's original patch had:
4005          *  smp_mb__after_netif_stop_queue();
4006          * but since that doesn't exist yet, just open code it.
4007          */
4008         smp_mb();
4009
4010         /*
4011          * We need to check again in a case another CPU has just
4012          * made room available.
4013          */
4014         if (e1000_desc_unused(adapter->tx_ring) < size)
4015                 return -EBUSY;
4016
4017         /* A reprieve! */
4018         netif_start_queue(netdev);
4019         ++adapter->restart_queue;
4020         return 0;
4021 }
4022
4023 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4024 {
4025         struct e1000_adapter *adapter = netdev_priv(netdev);
4026
4027         if (e1000_desc_unused(adapter->tx_ring) >= size)
4028                 return 0;
4029         return __e1000_maybe_stop_tx(netdev, size);
4030 }
4031
4032 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4033 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4034 {
4035         struct e1000_adapter *adapter = netdev_priv(netdev);
4036         struct e1000_ring *tx_ring = adapter->tx_ring;
4037         unsigned int first;
4038         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4039         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4040         unsigned int tx_flags = 0;
4041         unsigned int len = skb->len - skb->data_len;
4042         unsigned int nr_frags;
4043         unsigned int mss;
4044         int count = 0;
4045         int tso;
4046         unsigned int f;
4047
4048         if (test_bit(__E1000_DOWN, &adapter->state)) {
4049                 dev_kfree_skb_any(skb);
4050                 return NETDEV_TX_OK;
4051         }
4052
4053         if (skb->len <= 0) {
4054                 dev_kfree_skb_any(skb);
4055                 return NETDEV_TX_OK;
4056         }
4057
4058         mss = skb_shinfo(skb)->gso_size;
4059         /*
4060          * The controller does a simple calculation to
4061          * make sure there is enough room in the FIFO before
4062          * initiating the DMA for each buffer.  The calc is:
4063          * 4 = ceil(buffer len/mss).  To make sure we don't
4064          * overrun the FIFO, adjust the max buffer len if mss
4065          * drops.
4066          */
4067         if (mss) {
4068                 u8 hdr_len;
4069                 max_per_txd = min(mss << 2, max_per_txd);
4070                 max_txd_pwr = fls(max_per_txd) - 1;
4071
4072                 /*
4073                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4074                  * points to just header, pull a few bytes of payload from
4075                  * frags into skb->data
4076                  */
4077                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4078                 /*
4079                  * we do this workaround for ES2LAN, but it is un-necessary,
4080                  * avoiding it could save a lot of cycles
4081                  */
4082                 if (skb->data_len && (hdr_len == len)) {
4083                         unsigned int pull_size;
4084
4085                         pull_size = min((unsigned int)4, skb->data_len);
4086                         if (!__pskb_pull_tail(skb, pull_size)) {
4087                                 e_err("__pskb_pull_tail failed.\n");
4088                                 dev_kfree_skb_any(skb);
4089                                 return NETDEV_TX_OK;
4090                         }
4091                         len = skb->len - skb->data_len;
4092                 }
4093         }
4094
4095         /* reserve a descriptor for the offload context */
4096         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4097                 count++;
4098         count++;
4099
4100         count += TXD_USE_COUNT(len, max_txd_pwr);
4101
4102         nr_frags = skb_shinfo(skb)->nr_frags;
4103         for (f = 0; f < nr_frags; f++)
4104                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4105                                        max_txd_pwr);
4106
4107         if (adapter->hw.mac.tx_pkt_filtering)
4108                 e1000_transfer_dhcp_info(adapter, skb);
4109
4110         /*
4111          * need: count + 2 desc gap to keep tail from touching
4112          * head, otherwise try next time
4113          */
4114         if (e1000_maybe_stop_tx(netdev, count + 2))
4115                 return NETDEV_TX_BUSY;
4116
4117         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4118                 tx_flags |= E1000_TX_FLAGS_VLAN;
4119                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4120         }
4121
4122         first = tx_ring->next_to_use;
4123
4124         tso = e1000_tso(adapter, skb);
4125         if (tso < 0) {
4126                 dev_kfree_skb_any(skb);
4127                 return NETDEV_TX_OK;
4128         }
4129
4130         if (tso)
4131                 tx_flags |= E1000_TX_FLAGS_TSO;
4132         else if (e1000_tx_csum(adapter, skb))
4133                 tx_flags |= E1000_TX_FLAGS_CSUM;
4134
4135         /*
4136          * Old method was to assume IPv4 packet by default if TSO was enabled.
4137          * 82571 hardware supports TSO capabilities for IPv6 as well...
4138          * no longer assume, we must.
4139          */
4140         if (skb->protocol == htons(ETH_P_IP))
4141                 tx_flags |= E1000_TX_FLAGS_IPV4;
4142
4143         /* if count is 0 then mapping error has occured */
4144         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4145         if (count) {
4146                 e1000_tx_queue(adapter, tx_flags, count);
4147                 netdev->trans_start = jiffies;
4148                 /* Make sure there is space in the ring for the next send. */
4149                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4150
4151         } else {
4152                 dev_kfree_skb_any(skb);
4153                 tx_ring->buffer_info[first].time_stamp = 0;
4154                 tx_ring->next_to_use = first;
4155         }
4156
4157         return NETDEV_TX_OK;
4158 }
4159
4160 /**
4161  * e1000_tx_timeout - Respond to a Tx Hang
4162  * @netdev: network interface device structure
4163  **/
4164 static void e1000_tx_timeout(struct net_device *netdev)
4165 {
4166         struct e1000_adapter *adapter = netdev_priv(netdev);
4167
4168         /* Do the reset outside of interrupt context */
4169         adapter->tx_timeout_count++;
4170         schedule_work(&adapter->reset_task);
4171 }
4172
4173 static void e1000_reset_task(struct work_struct *work)
4174 {
4175         struct e1000_adapter *adapter;
4176         adapter = container_of(work, struct e1000_adapter, reset_task);
4177
4178         e1000e_reinit_locked(adapter);
4179 }
4180
4181 /**
4182  * e1000_get_stats - Get System Network Statistics
4183  * @netdev: network interface device structure
4184  *
4185  * Returns the address of the device statistics structure.
4186  * The statistics are actually updated from the timer callback.
4187  **/
4188 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4189 {
4190         struct e1000_adapter *adapter = netdev_priv(netdev);
4191
4192         /* only return the current stats */
4193         return &adapter->net_stats;
4194 }
4195
4196 /**
4197  * e1000_change_mtu - Change the Maximum Transfer Unit
4198  * @netdev: network interface device structure
4199  * @new_mtu: new value for maximum frame size
4200  *
4201  * Returns 0 on success, negative on failure
4202  **/
4203 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4204 {
4205         struct e1000_adapter *adapter = netdev_priv(netdev);
4206         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4207
4208         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4209             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
4210                 e_err("Invalid MTU setting\n");
4211                 return -EINVAL;
4212         }
4213
4214         /* Jumbo frame size limits */
4215         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
4216                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4217                         e_err("Jumbo Frames not supported.\n");
4218                         return -EINVAL;
4219                 }
4220                 if (adapter->hw.phy.type == e1000_phy_ife) {
4221                         e_err("Jumbo Frames not supported.\n");
4222                         return -EINVAL;
4223                 }
4224         }
4225
4226 #define MAX_STD_JUMBO_FRAME_SIZE 9234
4227         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
4228                 e_err("MTU > 9216 not supported.\n");
4229                 return -EINVAL;
4230         }
4231
4232         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4233                 msleep(1);
4234         /* e1000e_down has a dependency on max_frame_size */
4235         adapter->max_frame_size = max_frame;
4236         if (netif_running(netdev))
4237                 e1000e_down(adapter);
4238
4239         /*
4240          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4241          * means we reserve 2 more, this pushes us to allocate from the next
4242          * larger slab size.
4243          * i.e. RXBUFFER_2048 --> size-4096 slab
4244          * However with the new *_jumbo_rx* routines, jumbo receives will use
4245          * fragmented skbs
4246          */
4247
4248         if (max_frame <= 256)
4249                 adapter->rx_buffer_len = 256;
4250         else if (max_frame <= 512)
4251                 adapter->rx_buffer_len = 512;
4252         else if (max_frame <= 1024)
4253                 adapter->rx_buffer_len = 1024;
4254         else if (max_frame <= 2048)
4255                 adapter->rx_buffer_len = 2048;
4256         else
4257                 adapter->rx_buffer_len = 4096;
4258
4259         /* adjust allocation if LPE protects us, and we aren't using SBP */
4260         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4261              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4262                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4263                                          + ETH_FCS_LEN;
4264
4265         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4266         netdev->mtu = new_mtu;
4267
4268         if (netif_running(netdev))
4269                 e1000e_up(adapter);
4270         else
4271                 e1000e_reset(adapter);
4272
4273         clear_bit(__E1000_RESETTING, &adapter->state);
4274
4275         return 0;
4276 }
4277
4278 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4279                            int cmd)
4280 {
4281         struct e1000_adapter *adapter = netdev_priv(netdev);
4282         struct mii_ioctl_data *data = if_mii(ifr);
4283
4284         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4285                 return -EOPNOTSUPP;
4286
4287         switch (cmd) {
4288         case SIOCGMIIPHY:
4289                 data->phy_id = adapter->hw.phy.addr;
4290                 break;
4291         case SIOCGMIIREG:
4292                 if (!capable(CAP_NET_ADMIN))
4293                         return -EPERM;
4294                 switch (data->reg_num & 0x1F) {
4295                 case MII_BMCR:
4296                         data->val_out = adapter->phy_regs.bmcr;
4297                         break;
4298                 case MII_BMSR:
4299                         data->val_out = adapter->phy_regs.bmsr;
4300                         break;
4301                 case MII_PHYSID1:
4302                         data->val_out = (adapter->hw.phy.id >> 16);
4303                         break;
4304                 case MII_PHYSID2:
4305                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4306                         break;
4307                 case MII_ADVERTISE:
4308                         data->val_out = adapter->phy_regs.advertise;
4309                         break;
4310                 case MII_LPA:
4311                         data->val_out = adapter->phy_regs.lpa;
4312                         break;
4313                 case MII_EXPANSION:
4314                         data->val_out = adapter->phy_regs.expansion;
4315                         break;
4316                 case MII_CTRL1000:
4317                         data->val_out = adapter->phy_regs.ctrl1000;
4318                         break;
4319                 case MII_STAT1000:
4320                         data->val_out = adapter->phy_regs.stat1000;
4321                         break;
4322                 case MII_ESTATUS:
4323                         data->val_out = adapter->phy_regs.estatus;
4324                         break;
4325                 default:
4326                         return -EIO;
4327                 }
4328                 break;
4329         case SIOCSMIIREG:
4330         default:
4331                 return -EOPNOTSUPP;
4332         }
4333         return 0;
4334 }
4335
4336 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4337 {
4338         switch (cmd) {
4339         case SIOCGMIIPHY:
4340         case SIOCGMIIREG:
4341         case SIOCSMIIREG:
4342                 return e1000_mii_ioctl(netdev, ifr, cmd);
4343         default:
4344                 return -EOPNOTSUPP;
4345         }
4346 }
4347
4348 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4349 {
4350         struct net_device *netdev = pci_get_drvdata(pdev);
4351         struct e1000_adapter *adapter = netdev_priv(netdev);
4352         struct e1000_hw *hw = &adapter->hw;
4353         u32 ctrl, ctrl_ext, rctl, status;
4354         u32 wufc = adapter->wol;
4355         int retval = 0;
4356
4357         netif_device_detach(netdev);
4358
4359         if (netif_running(netdev)) {
4360                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4361                 e1000e_down(adapter);
4362                 e1000_free_irq(adapter);
4363         }
4364         e1000e_reset_interrupt_capability(adapter);
4365
4366         retval = pci_save_state(pdev);
4367         if (retval)
4368                 return retval;
4369
4370         status = er32(STATUS);
4371         if (status & E1000_STATUS_LU)
4372                 wufc &= ~E1000_WUFC_LNKC;
4373
4374         if (wufc) {
4375                 e1000_setup_rctl(adapter);
4376                 e1000_set_multi(netdev);
4377
4378                 /* turn on all-multi mode if wake on multicast is enabled */
4379                 if (wufc & E1000_WUFC_MC) {
4380                         rctl = er32(RCTL);
4381                         rctl |= E1000_RCTL_MPE;
4382                         ew32(RCTL, rctl);
4383                 }
4384
4385                 ctrl = er32(CTRL);
4386                 /* advertise wake from D3Cold */
4387                 #define E1000_CTRL_ADVD3WUC 0x00100000
4388                 /* phy power management enable */
4389                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4390                 ctrl |= E1000_CTRL_ADVD3WUC |
4391                         E1000_CTRL_EN_PHY_PWR_MGMT;
4392                 ew32(CTRL, ctrl);
4393
4394                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4395                     adapter->hw.phy.media_type ==
4396                     e1000_media_type_internal_serdes) {
4397                         /* keep the laser running in D3 */
4398                         ctrl_ext = er32(CTRL_EXT);
4399                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4400                         ew32(CTRL_EXT, ctrl_ext);
4401                 }
4402
4403                 if (adapter->flags & FLAG_IS_ICH)
4404                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4405
4406                 /* Allow time for pending master requests to run */
4407                 e1000e_disable_pcie_master(&adapter->hw);
4408
4409                 ew32(WUC, E1000_WUC_PME_EN);
4410                 ew32(WUFC, wufc);
4411                 pci_enable_wake(pdev, PCI_D3hot, 1);
4412                 pci_enable_wake(pdev, PCI_D3cold, 1);
4413         } else {
4414                 ew32(WUC, 0);
4415                 ew32(WUFC, 0);
4416                 pci_enable_wake(pdev, PCI_D3hot, 0);
4417                 pci_enable_wake(pdev, PCI_D3cold, 0);
4418         }
4419
4420         /* make sure adapter isn't asleep if manageability is enabled */
4421         if (adapter->flags & FLAG_MNG_PT_ENABLED) {
4422                 pci_enable_wake(pdev, PCI_D3hot, 1);
4423                 pci_enable_wake(pdev, PCI_D3cold, 1);
4424         }
4425
4426         if (adapter->hw.phy.type == e1000_phy_igp_3)
4427                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4428
4429         /*
4430          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4431          * would have already happened in close and is redundant.
4432          */
4433         e1000_release_hw_control(adapter);
4434
4435         pci_disable_device(pdev);
4436
4437         /*
4438          * The pci-e switch on some quad port adapters will report a
4439          * correctable error when the MAC transitions from D0 to D3.  To
4440          * prevent this we need to mask off the correctable errors on the
4441          * downstream port of the pci-e switch.
4442          */
4443         if (adapter->flags & FLAG_IS_QUAD_PORT) {
4444                 struct pci_dev *us_dev = pdev->bus->self;
4445                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
4446                 u16 devctl;
4447
4448                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
4449                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
4450                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
4451
4452                 pci_set_power_state(pdev, pci_choose_state(pdev, state));
4453
4454                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
4455         } else {
4456                 pci_set_power_state(pdev, pci_choose_state(pdev, state));
4457         }
4458
4459         return 0;
4460 }
4461
4462 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4463 {
4464         int pos;
4465         u16 val;
4466
4467         /*
4468          * 82573 workaround - disable L1 ASPM on mobile chipsets
4469          *
4470          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4471          * resulting in lost data or garbage information on the pci-e link
4472          * level. This could result in (false) bad EEPROM checksum errors,
4473          * long ping times (up to 2s) or even a system freeze/hang.
4474          *
4475          * Unfortunately this feature saves about 1W power consumption when
4476          * active.
4477          */
4478         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4479         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4480         if (val & 0x2) {
4481                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4482                 val &= ~0x2;
4483                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4484         }
4485 }
4486
4487 #ifdef CONFIG_PM
4488 static int e1000_resume(struct pci_dev *pdev)
4489 {
4490         struct net_device *netdev = pci_get_drvdata(pdev);
4491         struct e1000_adapter *adapter = netdev_priv(netdev);
4492         struct e1000_hw *hw = &adapter->hw;
4493         u32 err;
4494
4495         pci_set_power_state(pdev, PCI_D0);
4496         pci_restore_state(pdev);
4497         e1000e_disable_l1aspm(pdev);
4498
4499         err = pci_enable_device_mem(pdev);
4500         if (err) {
4501                 dev_err(&pdev->dev,
4502                         "Cannot enable PCI device from suspend\n");
4503                 return err;
4504         }
4505
4506         /* AER (Advanced Error Reporting) hooks */
4507         err = pci_enable_pcie_error_reporting(pdev);
4508         if (err) {
4509                 dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
4510                                     "0x%x\n", err);
4511                 /* non-fatal, continue */
4512         }
4513
4514         pci_set_master(pdev);
4515
4516         pci_enable_wake(pdev, PCI_D3hot, 0);
4517         pci_enable_wake(pdev, PCI_D3cold, 0);
4518
4519         e1000e_set_interrupt_capability(adapter);
4520         if (netif_running(netdev)) {
4521                 err = e1000_request_irq(adapter);
4522                 if (err)
4523                         return err;
4524         }
4525
4526         e1000e_power_up_phy(adapter);
4527         e1000e_reset(adapter);
4528         ew32(WUS, ~0);
4529
4530         e1000_init_manageability(adapter);
4531
4532         if (netif_running(netdev))
4533                 e1000e_up(adapter);
4534
4535         netif_device_attach(netdev);
4536
4537         /*
4538          * If the controller has AMT, do not set DRV_LOAD until the interface
4539          * is up.  For all other cases, let the f/w know that the h/w is now
4540          * under the control of the driver.
4541          */
4542         if (!(adapter->flags & FLAG_HAS_AMT))
4543                 e1000_get_hw_control(adapter);
4544
4545         return 0;
4546 }
4547 #endif
4548
4549 static void e1000_shutdown(struct pci_dev *pdev)
4550 {
4551         e1000_suspend(pdev, PMSG_SUSPEND);
4552 }
4553
4554 #ifdef CONFIG_NET_POLL_CONTROLLER
4555 /*
4556  * Polling 'interrupt' - used by things like netconsole to send skbs
4557  * without having to re-enable interrupts. It's not called while
4558  * the interrupt routine is executing.
4559  */
4560 static void e1000_netpoll(struct net_device *netdev)
4561 {
4562         struct e1000_adapter *adapter = netdev_priv(netdev);
4563
4564         disable_irq(adapter->pdev->irq);
4565         e1000_intr(adapter->pdev->irq, netdev);
4566
4567         enable_irq(adapter->pdev->irq);
4568 }
4569 #endif
4570
4571 /**
4572  * e1000_io_error_detected - called when PCI error is detected
4573  * @pdev: Pointer to PCI device
4574  * @state: The current pci connection state
4575  *
4576  * This function is called after a PCI bus error affecting
4577  * this device has been detected.
4578  */
4579 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4580                                                 pci_channel_state_t state)
4581 {
4582         struct net_device *netdev = pci_get_drvdata(pdev);
4583         struct e1000_adapter *adapter = netdev_priv(netdev);
4584
4585         netif_device_detach(netdev);
4586
4587         if (netif_running(netdev))
4588                 e1000e_down(adapter);
4589         pci_disable_device(pdev);
4590
4591         /* Request a slot slot reset. */
4592         return PCI_ERS_RESULT_NEED_RESET;
4593 }
4594
4595 /**
4596  * e1000_io_slot_reset - called after the pci bus has been reset.
4597  * @pdev: Pointer to PCI device
4598  *
4599  * Restart the card from scratch, as if from a cold-boot. Implementation
4600  * resembles the first-half of the e1000_resume routine.
4601  */
4602 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4603 {
4604         struct net_device *netdev = pci_get_drvdata(pdev);
4605         struct e1000_adapter *adapter = netdev_priv(netdev);
4606         struct e1000_hw *hw = &adapter->hw;
4607         int err;
4608         pci_ers_result_t result;
4609
4610         e1000e_disable_l1aspm(pdev);
4611         err = pci_enable_device_mem(pdev);
4612         if (err) {
4613                 dev_err(&pdev->dev,
4614                         "Cannot re-enable PCI device after reset.\n");
4615                 result = PCI_ERS_RESULT_DISCONNECT;
4616         } else {
4617                 pci_set_master(pdev);
4618                 pci_restore_state(pdev);
4619
4620                 pci_enable_wake(pdev, PCI_D3hot, 0);
4621                 pci_enable_wake(pdev, PCI_D3cold, 0);
4622
4623                 e1000e_reset(adapter);
4624                 ew32(WUS, ~0);
4625                 result = PCI_ERS_RESULT_RECOVERED;
4626         }
4627
4628         pci_cleanup_aer_uncorrect_error_status(pdev);
4629
4630         return result;
4631 }
4632
4633 /**
4634  * e1000_io_resume - called when traffic can start flowing again.
4635  * @pdev: Pointer to PCI device
4636  *
4637  * This callback is called when the error recovery driver tells us that
4638  * its OK to resume normal operation. Implementation resembles the
4639  * second-half of the e1000_resume routine.
4640  */
4641 static void e1000_io_resume(struct pci_dev *pdev)
4642 {
4643         struct net_device *netdev = pci_get_drvdata(pdev);
4644         struct e1000_adapter *adapter = netdev_priv(netdev);
4645
4646         e1000_init_manageability(adapter);
4647
4648         if (netif_running(netdev)) {
4649                 if (e1000e_up(adapter)) {
4650                         dev_err(&pdev->dev,
4651                                 "can't bring device back up after reset\n");
4652                         return;
4653                 }
4654         }
4655
4656         netif_device_attach(netdev);
4657
4658         /*
4659          * If the controller has AMT, do not set DRV_LOAD until the interface
4660          * is up.  For all other cases, let the f/w know that the h/w is now
4661          * under the control of the driver.
4662          */
4663         if (!(adapter->flags & FLAG_HAS_AMT))
4664                 e1000_get_hw_control(adapter);
4665
4666 }
4667
4668 static void e1000_print_device_info(struct e1000_adapter *adapter)
4669 {
4670         struct e1000_hw *hw = &adapter->hw;
4671         struct net_device *netdev = adapter->netdev;
4672         u32 pba_num;
4673
4674         /* print bus type/speed/width info */
4675         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
4676                /* bus width */
4677                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4678                 "Width x1"),
4679                /* MAC address */
4680                netdev->dev_addr);
4681         e_info("Intel(R) PRO/%s Network Connection\n",
4682                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4683         e1000e_read_pba_num(hw, &pba_num);
4684         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4685                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4686 }
4687
4688 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4689 {
4690         struct e1000_hw *hw = &adapter->hw;
4691         int ret_val;
4692         u16 buf = 0;
4693
4694         if (hw->mac.type != e1000_82573)
4695                 return;
4696
4697         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4698         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
4699                 /* Deep Smart Power Down (DSPD) */
4700                 dev_warn(&adapter->pdev->dev,
4701                          "Warning: detected DSPD enabled in EEPROM\n");
4702         }
4703
4704         ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4705         if (!ret_val && (le16_to_cpu(buf) & (3 << 2))) {
4706                 /* ASPM enable */
4707                 dev_warn(&adapter->pdev->dev,
4708                          "Warning: detected ASPM enabled in EEPROM\n");
4709         }
4710 }
4711
4712 static const struct net_device_ops e1000e_netdev_ops = {
4713         .ndo_open               = e1000_open,
4714         .ndo_stop               = e1000_close,
4715         .ndo_start_xmit         = e1000_xmit_frame,
4716         .ndo_get_stats          = e1000_get_stats,
4717         .ndo_set_multicast_list = e1000_set_multi,
4718         .ndo_set_mac_address    = e1000_set_mac,
4719         .ndo_change_mtu         = e1000_change_mtu,
4720         .ndo_do_ioctl           = e1000_ioctl,
4721         .ndo_tx_timeout         = e1000_tx_timeout,
4722         .ndo_validate_addr      = eth_validate_addr,
4723
4724         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
4725         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
4726         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
4727 #ifdef CONFIG_NET_POLL_CONTROLLER
4728         .ndo_poll_controller    = e1000_netpoll,
4729 #endif
4730 };
4731
4732 /**
4733  * e1000_probe - Device Initialization Routine
4734  * @pdev: PCI device information struct
4735  * @ent: entry in e1000_pci_tbl
4736  *
4737  * Returns 0 on success, negative on failure
4738  *
4739  * e1000_probe initializes an adapter identified by a pci_dev structure.
4740  * The OS initialization, configuring of the adapter private structure,
4741  * and a hardware reset occur.
4742  **/
4743 static int __devinit e1000_probe(struct pci_dev *pdev,
4744                                  const struct pci_device_id *ent)
4745 {
4746         struct net_device *netdev;
4747         struct e1000_adapter *adapter;
4748         struct e1000_hw *hw;
4749         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4750         resource_size_t mmio_start, mmio_len;
4751         resource_size_t flash_start, flash_len;
4752
4753         static int cards_found;
4754         int i, err, pci_using_dac;
4755         u16 eeprom_data = 0;
4756         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4757
4758         e1000e_disable_l1aspm(pdev);
4759
4760         err = pci_enable_device_mem(pdev);
4761         if (err)
4762                 return err;
4763
4764         pci_using_dac = 0;
4765         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
4766         if (!err) {
4767                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
4768                 if (!err)
4769                         pci_using_dac = 1;
4770         } else {
4771                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
4772                 if (err) {
4773                         err = pci_set_consistent_dma_mask(pdev,
4774                                                           DMA_32BIT_MASK);
4775                         if (err) {
4776                                 dev_err(&pdev->dev, "No usable DMA "
4777                                         "configuration, aborting\n");
4778                                 goto err_dma;
4779                         }
4780                 }
4781         }
4782
4783         err = pci_request_selected_regions_exclusive(pdev,
4784                                           pci_select_bars(pdev, IORESOURCE_MEM),
4785                                           e1000e_driver_name);
4786         if (err)
4787                 goto err_pci_reg;
4788
4789         pci_set_master(pdev);
4790         /* PCI config space info */
4791         err = pci_save_state(pdev);
4792         if (err)
4793                 goto err_alloc_etherdev;
4794
4795         err = -ENOMEM;
4796         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4797         if (!netdev)
4798                 goto err_alloc_etherdev;
4799
4800         SET_NETDEV_DEV(netdev, &pdev->dev);
4801
4802         pci_set_drvdata(pdev, netdev);
4803         adapter = netdev_priv(netdev);
4804         hw = &adapter->hw;
4805         adapter->netdev = netdev;
4806         adapter->pdev = pdev;
4807         adapter->ei = ei;
4808         adapter->pba = ei->pba;
4809         adapter->flags = ei->flags;
4810         adapter->flags2 = ei->flags2;
4811         adapter->hw.adapter = adapter;
4812         adapter->hw.mac.type = ei->mac;
4813         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4814
4815         mmio_start = pci_resource_start(pdev, 0);
4816         mmio_len = pci_resource_len(pdev, 0);
4817
4818         err = -EIO;
4819         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4820         if (!adapter->hw.hw_addr)
4821                 goto err_ioremap;
4822
4823         if ((adapter->flags & FLAG_HAS_FLASH) &&
4824             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
4825                 flash_start = pci_resource_start(pdev, 1);
4826                 flash_len = pci_resource_len(pdev, 1);
4827                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
4828                 if (!adapter->hw.flash_address)
4829                         goto err_flashmap;
4830         }
4831
4832         /* construct the net_device struct */
4833         netdev->netdev_ops              = &e1000e_netdev_ops;
4834         e1000e_set_ethtool_ops(netdev);
4835         netdev->watchdog_timeo          = 5 * HZ;
4836         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
4837         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
4838
4839         netdev->mem_start = mmio_start;
4840         netdev->mem_end = mmio_start + mmio_len;
4841
4842         adapter->bd_number = cards_found++;
4843
4844         e1000e_check_options(adapter);
4845
4846         /* setup adapter struct */
4847         err = e1000_sw_init(adapter);
4848         if (err)
4849                 goto err_sw_init;
4850
4851         err = -EIO;
4852
4853         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4854         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
4855         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4856
4857         err = ei->get_variants(adapter);
4858         if (err)
4859                 goto err_hw_init;
4860
4861         if ((adapter->flags & FLAG_IS_ICH) &&
4862             (adapter->flags & FLAG_READ_ONLY_NVM))
4863                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
4864
4865         hw->mac.ops.get_bus_info(&adapter->hw);
4866
4867         adapter->hw.phy.autoneg_wait_to_complete = 0;
4868
4869         /* Copper options */
4870         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
4871                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
4872                 adapter->hw.phy.disable_polarity_correction = 0;
4873                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
4874         }
4875
4876         if (e1000_check_reset_block(&adapter->hw))
4877                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
4878
4879         netdev->features = NETIF_F_SG |
4880                            NETIF_F_HW_CSUM |
4881                            NETIF_F_HW_VLAN_TX |
4882                            NETIF_F_HW_VLAN_RX;
4883
4884         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
4885                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
4886
4887         netdev->features |= NETIF_F_TSO;
4888         netdev->features |= NETIF_F_TSO6;
4889
4890         netdev->vlan_features |= NETIF_F_TSO;
4891         netdev->vlan_features |= NETIF_F_TSO6;
4892         netdev->vlan_features |= NETIF_F_HW_CSUM;
4893         netdev->vlan_features |= NETIF_F_SG;
4894
4895         if (pci_using_dac)
4896                 netdev->features |= NETIF_F_HIGHDMA;
4897
4898         if (e1000e_enable_mng_pass_thru(&adapter->hw))
4899                 adapter->flags |= FLAG_MNG_PT_ENABLED;
4900
4901         /*
4902          * before reading the NVM, reset the controller to
4903          * put the device in a known good starting state
4904          */
4905         adapter->hw.mac.ops.reset_hw(&adapter->hw);
4906
4907         /*
4908          * systems with ASPM and others may see the checksum fail on the first
4909          * attempt. Let's give it a few tries
4910          */
4911         for (i = 0;; i++) {
4912                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
4913                         break;
4914                 if (i == 2) {
4915                         e_err("The NVM Checksum Is Not Valid\n");
4916                         err = -EIO;
4917                         goto err_eeprom;
4918                 }
4919         }
4920
4921         e1000_eeprom_checks(adapter);
4922
4923         /* copy the MAC address out of the NVM */
4924         if (e1000e_read_mac_addr(&adapter->hw))
4925                 e_err("NVM Read Error while reading MAC address\n");
4926
4927         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
4928         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
4929
4930         if (!is_valid_ether_addr(netdev->perm_addr)) {
4931                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
4932                 err = -EIO;
4933                 goto err_eeprom;
4934         }
4935
4936         init_timer(&adapter->watchdog_timer);
4937         adapter->watchdog_timer.function = &e1000_watchdog;
4938         adapter->watchdog_timer.data = (unsigned long) adapter;
4939
4940         init_timer(&adapter->phy_info_timer);
4941         adapter->phy_info_timer.function = &e1000_update_phy_info;
4942         adapter->phy_info_timer.data = (unsigned long) adapter;
4943
4944         INIT_WORK(&adapter->reset_task, e1000_reset_task);
4945         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
4946         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
4947         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
4948
4949         /* Initialize link parameters. User can change them with ethtool */
4950         adapter->hw.mac.autoneg = 1;
4951         adapter->fc_autoneg = 1;
4952         adapter->hw.fc.requested_mode = e1000_fc_default;
4953         adapter->hw.fc.current_mode = e1000_fc_default;
4954         adapter->hw.phy.autoneg_advertised = 0x2f;
4955
4956         /* ring size defaults */
4957         adapter->rx_ring->count = 256;
4958         adapter->tx_ring->count = 256;
4959
4960         /*
4961          * Initial Wake on LAN setting - If APM wake is enabled in
4962          * the EEPROM, enable the ACPI Magic Packet filter
4963          */
4964         if (adapter->flags & FLAG_APME_IN_WUC) {
4965                 /* APME bit in EEPROM is mapped to WUC.APME */
4966                 eeprom_data = er32(WUC);
4967                 eeprom_apme_mask = E1000_WUC_APME;
4968         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
4969                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
4970                     (adapter->hw.bus.func == 1))
4971                         e1000_read_nvm(&adapter->hw,
4972                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
4973                 else
4974                         e1000_read_nvm(&adapter->hw,
4975                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
4976         }
4977
4978         /* fetch WoL from EEPROM */
4979         if (eeprom_data & eeprom_apme_mask)
4980                 adapter->eeprom_wol |= E1000_WUFC_MAG;
4981
4982         /*
4983          * now that we have the eeprom settings, apply the special cases
4984          * where the eeprom may be wrong or the board simply won't support
4985          * wake on lan on a particular port
4986          */
4987         if (!(adapter->flags & FLAG_HAS_WOL))
4988                 adapter->eeprom_wol = 0;
4989
4990         /* initialize the wol settings based on the eeprom settings */
4991         adapter->wol = adapter->eeprom_wol;
4992         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
4993
4994         /* save off EEPROM version number */
4995         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
4996
4997         /* reset the hardware with the new settings */
4998         e1000e_reset(adapter);
4999
5000         /*
5001          * If the controller has AMT, do not set DRV_LOAD until the interface
5002          * is up.  For all other cases, let the f/w know that the h/w is now
5003          * under the control of the driver.
5004          */
5005         if (!(adapter->flags & FLAG_HAS_AMT))
5006                 e1000_get_hw_control(adapter);
5007
5008         /* tell the stack to leave us alone until e1000_open() is called */
5009         netif_carrier_off(netdev);
5010         netif_tx_stop_all_queues(netdev);
5011
5012         strcpy(netdev->name, "eth%d");
5013         err = register_netdev(netdev);
5014         if (err)
5015                 goto err_register;
5016
5017         e1000_print_device_info(adapter);
5018
5019         return 0;
5020
5021 err_register:
5022         if (!(adapter->flags & FLAG_HAS_AMT))
5023                 e1000_release_hw_control(adapter);
5024 err_eeprom:
5025         if (!e1000_check_reset_block(&adapter->hw))
5026                 e1000_phy_hw_reset(&adapter->hw);
5027 err_hw_init:
5028
5029         kfree(adapter->tx_ring);
5030         kfree(adapter->rx_ring);
5031 err_sw_init:
5032         if (adapter->hw.flash_address)
5033                 iounmap(adapter->hw.flash_address);
5034         e1000e_reset_interrupt_capability(adapter);
5035 err_flashmap:
5036         iounmap(adapter->hw.hw_addr);
5037 err_ioremap:
5038         free_netdev(netdev);
5039 err_alloc_etherdev:
5040         pci_release_selected_regions(pdev,
5041                                      pci_select_bars(pdev, IORESOURCE_MEM));
5042 err_pci_reg:
5043 err_dma:
5044         pci_disable_device(pdev);
5045         return err;
5046 }
5047
5048 /**
5049  * e1000_remove - Device Removal Routine
5050  * @pdev: PCI device information struct
5051  *
5052  * e1000_remove is called by the PCI subsystem to alert the driver
5053  * that it should release a PCI device.  The could be caused by a
5054  * Hot-Plug event, or because the driver is going to be removed from
5055  * memory.
5056  **/
5057 static void __devexit e1000_remove(struct pci_dev *pdev)
5058 {
5059         struct net_device *netdev = pci_get_drvdata(pdev);
5060         struct e1000_adapter *adapter = netdev_priv(netdev);
5061         int err;
5062
5063         /*
5064          * flush_scheduled work may reschedule our watchdog task, so
5065          * explicitly disable watchdog tasks from being rescheduled
5066          */
5067         set_bit(__E1000_DOWN, &adapter->state);
5068         del_timer_sync(&adapter->watchdog_timer);
5069         del_timer_sync(&adapter->phy_info_timer);
5070
5071         flush_scheduled_work();
5072
5073         /*
5074          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5075          * would have already happened in close and is redundant.
5076          */
5077         e1000_release_hw_control(adapter);
5078
5079         unregister_netdev(netdev);
5080
5081         if (!e1000_check_reset_block(&adapter->hw))
5082                 e1000_phy_hw_reset(&adapter->hw);
5083
5084         e1000e_reset_interrupt_capability(adapter);
5085         kfree(adapter->tx_ring);
5086         kfree(adapter->rx_ring);
5087
5088         iounmap(adapter->hw.hw_addr);
5089         if (adapter->hw.flash_address)
5090                 iounmap(adapter->hw.flash_address);
5091         pci_release_selected_regions(pdev,
5092                                      pci_select_bars(pdev, IORESOURCE_MEM));
5093
5094         free_netdev(netdev);
5095
5096         /* AER disable */
5097         err = pci_disable_pcie_error_reporting(pdev);
5098         if (err)
5099                 dev_err(&pdev->dev,
5100                         "pci_disable_pcie_error_reporting failed 0x%x\n", err);
5101
5102         pci_disable_device(pdev);
5103 }
5104
5105 /* PCI Error Recovery (ERS) */
5106 static struct pci_error_handlers e1000_err_handler = {
5107         .error_detected = e1000_io_error_detected,
5108         .slot_reset = e1000_io_slot_reset,
5109         .resume = e1000_io_resume,
5110 };
5111
5112 static struct pci_device_id e1000_pci_tbl[] = {
5113         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5114         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5115         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5116         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5117         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5118         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5119         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5120         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5121         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5122
5123         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5124         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5125         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5126         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5127
5128         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5129         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5130         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5131
5132         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5133         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
5134
5135         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5136           board_80003es2lan },
5137         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5138           board_80003es2lan },
5139         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5140           board_80003es2lan },
5141         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5142           board_80003es2lan },
5143
5144         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5145         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5146         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5147         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5148         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5149         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5150         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5151
5152         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5153         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5154         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5155         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5156         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5157         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5158         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5159         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5160         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5161
5162         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5163         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5164         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5165
5166         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5167         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5168
5169         { }     /* terminate list */
5170 };
5171 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5172
5173 /* PCI Device API Driver */
5174 static struct pci_driver e1000_driver = {
5175         .name     = e1000e_driver_name,
5176         .id_table = e1000_pci_tbl,
5177         .probe    = e1000_probe,
5178         .remove   = __devexit_p(e1000_remove),
5179 #ifdef CONFIG_PM
5180         /* Power Management Hooks */
5181         .suspend  = e1000_suspend,
5182         .resume   = e1000_resume,
5183 #endif
5184         .shutdown = e1000_shutdown,
5185         .err_handler = &e1000_err_handler
5186 };
5187
5188 /**
5189  * e1000_init_module - Driver Registration Routine
5190  *
5191  * e1000_init_module is the first routine called when the driver is
5192  * loaded. All it does is register with the PCI subsystem.
5193  **/
5194 static int __init e1000_init_module(void)
5195 {
5196         int ret;
5197         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
5198                e1000e_driver_name, e1000e_driver_version);
5199         printk(KERN_INFO "%s: Copyright (c) 1999-2008 Intel Corporation.\n",
5200                e1000e_driver_name);
5201         ret = pci_register_driver(&e1000_driver);
5202         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name,
5203                                PM_QOS_DEFAULT_VALUE);
5204                                 
5205         return ret;
5206 }
5207 module_init(e1000_init_module);
5208
5209 /**
5210  * e1000_exit_module - Driver Exit Cleanup Routine
5211  *
5212  * e1000_exit_module is called just before the driver is removed
5213  * from memory.
5214  **/
5215 static void __exit e1000_exit_module(void)
5216 {
5217         pci_unregister_driver(&e1000_driver);
5218         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name);
5219 }
5220 module_exit(e1000_exit_module);
5221
5222
5223 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5224 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5225 MODULE_LICENSE("GPL");
5226 MODULE_VERSION(DRV_VERSION);
5227
5228 /* e1000_main.c */