17974121912d85499de931948047e80cd5c5ba19
[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;
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                (count < tx_ring->count)) {
633                 for (cleaned = 0; !cleaned; count++) {
634                         tx_desc = E1000_TX_DESC(*tx_ring, i);
635                         buffer_info = &tx_ring->buffer_info[i];
636                         cleaned = (i == eop);
637
638                         if (cleaned) {
639                                 struct sk_buff *skb = buffer_info->skb;
640                                 unsigned int segs, bytecount;
641                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
642                                 /* multiply data chunks by size of headers */
643                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
644                                             skb->len;
645                                 total_tx_packets += segs;
646                                 total_tx_bytes += bytecount;
647                         }
648
649                         e1000_put_txbuf(adapter, buffer_info);
650                         tx_desc->upper.data = 0;
651
652                         i++;
653                         if (i == tx_ring->count)
654                                 i = 0;
655                 }
656
657                 eop = tx_ring->buffer_info[i].next_to_watch;
658                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
659         }
660
661         tx_ring->next_to_clean = i;
662
663 #define TX_WAKE_THRESHOLD 32
664         if (cleaned && netif_carrier_ok(netdev) &&
665                      e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
666                 /* Make sure that anybody stopping the queue after this
667                  * sees the new next_to_clean.
668                  */
669                 smp_mb();
670
671                 if (netif_queue_stopped(netdev) &&
672                     !(test_bit(__E1000_DOWN, &adapter->state))) {
673                         netif_wake_queue(netdev);
674                         ++adapter->restart_queue;
675                 }
676         }
677
678         if (adapter->detect_tx_hung) {
679                 /* Detect a transmit hang in hardware, this serializes the
680                  * check with the clearing of time_stamp and movement of i */
681                 adapter->detect_tx_hung = 0;
682                 if (tx_ring->buffer_info[i].time_stamp &&
683                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
684                                + (adapter->tx_timeout_factor * HZ))
685                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
686                         e1000_print_tx_hang(adapter);
687                         netif_stop_queue(netdev);
688                 }
689         }
690         adapter->total_tx_bytes += total_tx_bytes;
691         adapter->total_tx_packets += total_tx_packets;
692         adapter->net_stats.tx_bytes += total_tx_bytes;
693         adapter->net_stats.tx_packets += total_tx_packets;
694         return (count < tx_ring->count);
695 }
696
697 /**
698  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
699  * @adapter: board private structure
700  *
701  * the return value indicates whether actual cleaning was done, there
702  * is no guarantee that everything was cleaned
703  **/
704 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
705                                   int *work_done, int work_to_do)
706 {
707         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
708         struct net_device *netdev = adapter->netdev;
709         struct pci_dev *pdev = adapter->pdev;
710         struct e1000_ring *rx_ring = adapter->rx_ring;
711         struct e1000_buffer *buffer_info, *next_buffer;
712         struct e1000_ps_page *ps_page;
713         struct sk_buff *skb;
714         unsigned int i, j;
715         u32 length, staterr;
716         int cleaned_count = 0;
717         bool cleaned = 0;
718         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
719
720         i = rx_ring->next_to_clean;
721         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
722         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
723         buffer_info = &rx_ring->buffer_info[i];
724
725         while (staterr & E1000_RXD_STAT_DD) {
726                 if (*work_done >= work_to_do)
727                         break;
728                 (*work_done)++;
729                 skb = buffer_info->skb;
730
731                 /* in the packet split case this is header only */
732                 prefetch(skb->data - NET_IP_ALIGN);
733
734                 i++;
735                 if (i == rx_ring->count)
736                         i = 0;
737                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
738                 prefetch(next_rxd);
739
740                 next_buffer = &rx_ring->buffer_info[i];
741
742                 cleaned = 1;
743                 cleaned_count++;
744                 pci_unmap_single(pdev, buffer_info->dma,
745                                  adapter->rx_ps_bsize0,
746                                  PCI_DMA_FROMDEVICE);
747                 buffer_info->dma = 0;
748
749                 if (!(staterr & E1000_RXD_STAT_EOP)) {
750                         e_dbg("%s: Packet Split buffers didn't pick up the "
751                               "full packet\n", netdev->name);
752                         dev_kfree_skb_irq(skb);
753                         goto next_desc;
754                 }
755
756                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
757                         dev_kfree_skb_irq(skb);
758                         goto next_desc;
759                 }
760
761                 length = le16_to_cpu(rx_desc->wb.middle.length0);
762
763                 if (!length) {
764                         e_dbg("%s: Last part of the packet spanning multiple "
765                               "descriptors\n", netdev->name);
766                         dev_kfree_skb_irq(skb);
767                         goto next_desc;
768                 }
769
770                 /* Good Receive */
771                 skb_put(skb, length);
772
773                 {
774                 /*
775                  * this looks ugly, but it seems compiler issues make it
776                  * more efficient than reusing j
777                  */
778                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
779
780                 /*
781                  * page alloc/put takes too long and effects small packet
782                  * throughput, so unsplit small packets and save the alloc/put
783                  * only valid in softirq (napi) context to call kmap_*
784                  */
785                 if (l1 && (l1 <= copybreak) &&
786                     ((length + l1) <= adapter->rx_ps_bsize0)) {
787                         u8 *vaddr;
788
789                         ps_page = &buffer_info->ps_pages[0];
790
791                         /*
792                          * there is no documentation about how to call
793                          * kmap_atomic, so we can't hold the mapping
794                          * very long
795                          */
796                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
797                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
798                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
799                         memcpy(skb_tail_pointer(skb), vaddr, l1);
800                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
801                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
802                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
803
804                         /* remove the CRC */
805                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
806                                 l1 -= 4;
807
808                         skb_put(skb, l1);
809                         goto copydone;
810                 } /* if */
811                 }
812
813                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
814                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
815                         if (!length)
816                                 break;
817
818                         ps_page = &buffer_info->ps_pages[j];
819                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
820                                        PCI_DMA_FROMDEVICE);
821                         ps_page->dma = 0;
822                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
823                         ps_page->page = NULL;
824                         skb->len += length;
825                         skb->data_len += length;
826                         skb->truesize += length;
827                 }
828
829                 /* strip the ethernet crc, problem is we're using pages now so
830                  * this whole operation can get a little cpu intensive
831                  */
832                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
833                         pskb_trim(skb, skb->len - 4);
834
835 copydone:
836                 total_rx_bytes += skb->len;
837                 total_rx_packets++;
838
839                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
840                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
841
842                 if (rx_desc->wb.upper.header_status &
843                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
844                         adapter->rx_hdr_split++;
845
846                 e1000_receive_skb(adapter, netdev, skb,
847                                   staterr, rx_desc->wb.middle.vlan);
848
849 next_desc:
850                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
851                 buffer_info->skb = NULL;
852
853                 /* return some buffers to hardware, one at a time is too slow */
854                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
855                         adapter->alloc_rx_buf(adapter, cleaned_count);
856                         cleaned_count = 0;
857                 }
858
859                 /* use prefetched values */
860                 rx_desc = next_rxd;
861                 buffer_info = next_buffer;
862
863                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
864         }
865         rx_ring->next_to_clean = i;
866
867         cleaned_count = e1000_desc_unused(rx_ring);
868         if (cleaned_count)
869                 adapter->alloc_rx_buf(adapter, cleaned_count);
870
871         adapter->total_rx_bytes += total_rx_bytes;
872         adapter->total_rx_packets += total_rx_packets;
873         adapter->net_stats.rx_bytes += total_rx_bytes;
874         adapter->net_stats.rx_packets += total_rx_packets;
875         return cleaned;
876 }
877
878 /**
879  * e1000_consume_page - helper function
880  **/
881 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
882                                u16 length)
883 {
884         bi->page = NULL;
885         skb->len += length;
886         skb->data_len += length;
887         skb->truesize += length;
888 }
889
890 /**
891  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
892  * @adapter: board private structure
893  *
894  * the return value indicates whether actual cleaning was done, there
895  * is no guarantee that everything was cleaned
896  **/
897
898 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
899                                      int *work_done, int work_to_do)
900 {
901         struct net_device *netdev = adapter->netdev;
902         struct pci_dev *pdev = adapter->pdev;
903         struct e1000_ring *rx_ring = adapter->rx_ring;
904         struct e1000_rx_desc *rx_desc, *next_rxd;
905         struct e1000_buffer *buffer_info, *next_buffer;
906         u32 length;
907         unsigned int i;
908         int cleaned_count = 0;
909         bool cleaned = false;
910         unsigned int total_rx_bytes=0, total_rx_packets=0;
911
912         i = rx_ring->next_to_clean;
913         rx_desc = E1000_RX_DESC(*rx_ring, i);
914         buffer_info = &rx_ring->buffer_info[i];
915
916         while (rx_desc->status & E1000_RXD_STAT_DD) {
917                 struct sk_buff *skb;
918                 u8 status;
919
920                 if (*work_done >= work_to_do)
921                         break;
922                 (*work_done)++;
923
924                 status = rx_desc->status;
925                 skb = buffer_info->skb;
926                 buffer_info->skb = NULL;
927
928                 ++i;
929                 if (i == rx_ring->count)
930                         i = 0;
931                 next_rxd = E1000_RX_DESC(*rx_ring, i);
932                 prefetch(next_rxd);
933
934                 next_buffer = &rx_ring->buffer_info[i];
935
936                 cleaned = true;
937                 cleaned_count++;
938                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
939                                PCI_DMA_FROMDEVICE);
940                 buffer_info->dma = 0;
941
942                 length = le16_to_cpu(rx_desc->length);
943
944                 /* errors is only valid for DD + EOP descriptors */
945                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
946                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
947                                 /* recycle both page and skb */
948                                 buffer_info->skb = skb;
949                                 /* an error means any chain goes out the window
950                                  * too */
951                                 if (rx_ring->rx_skb_top)
952                                         dev_kfree_skb(rx_ring->rx_skb_top);
953                                 rx_ring->rx_skb_top = NULL;
954                                 goto next_desc;
955                 }
956
957 #define rxtop rx_ring->rx_skb_top
958                 if (!(status & E1000_RXD_STAT_EOP)) {
959                         /* this descriptor is only the beginning (or middle) */
960                         if (!rxtop) {
961                                 /* this is the beginning of a chain */
962                                 rxtop = skb;
963                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
964                                                    0, length);
965                         } else {
966                                 /* this is the middle of a chain */
967                                 skb_fill_page_desc(rxtop,
968                                     skb_shinfo(rxtop)->nr_frags,
969                                     buffer_info->page, 0, length);
970                                 /* re-use the skb, only consumed the page */
971                                 buffer_info->skb = skb;
972                         }
973                         e1000_consume_page(buffer_info, rxtop, length);
974                         goto next_desc;
975                 } else {
976                         if (rxtop) {
977                                 /* end of the chain */
978                                 skb_fill_page_desc(rxtop,
979                                     skb_shinfo(rxtop)->nr_frags,
980                                     buffer_info->page, 0, length);
981                                 /* re-use the current skb, we only consumed the
982                                  * page */
983                                 buffer_info->skb = skb;
984                                 skb = rxtop;
985                                 rxtop = NULL;
986                                 e1000_consume_page(buffer_info, skb, length);
987                         } else {
988                                 /* no chain, got EOP, this buf is the packet
989                                  * copybreak to save the put_page/alloc_page */
990                                 if (length <= copybreak &&
991                                     skb_tailroom(skb) >= length) {
992                                         u8 *vaddr;
993                                         vaddr = kmap_atomic(buffer_info->page,
994                                                            KM_SKB_DATA_SOFTIRQ);
995                                         memcpy(skb_tail_pointer(skb), vaddr,
996                                                length);
997                                         kunmap_atomic(vaddr,
998                                                       KM_SKB_DATA_SOFTIRQ);
999                                         /* re-use the page, so don't erase
1000                                          * buffer_info->page */
1001                                         skb_put(skb, length);
1002                                 } else {
1003                                         skb_fill_page_desc(skb, 0,
1004                                                            buffer_info->page, 0,
1005                                                            length);
1006                                         e1000_consume_page(buffer_info, skb,
1007                                                            length);
1008                                 }
1009                         }
1010                 }
1011
1012                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1013                 e1000_rx_checksum(adapter,
1014                                   (u32)(status) |
1015                                   ((u32)(rx_desc->errors) << 24),
1016                                   le16_to_cpu(rx_desc->csum), skb);
1017
1018                 /* probably a little skewed due to removing CRC */
1019                 total_rx_bytes += skb->len;
1020                 total_rx_packets++;
1021
1022                 /* eth type trans needs skb->data to point to something */
1023                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1024                         e_err("pskb_may_pull failed.\n");
1025                         dev_kfree_skb(skb);
1026                         goto next_desc;
1027                 }
1028
1029                 e1000_receive_skb(adapter, netdev, skb, status,
1030                                   rx_desc->special);
1031
1032 next_desc:
1033                 rx_desc->status = 0;
1034
1035                 /* return some buffers to hardware, one at a time is too slow */
1036                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1037                         adapter->alloc_rx_buf(adapter, cleaned_count);
1038                         cleaned_count = 0;
1039                 }
1040
1041                 /* use prefetched values */
1042                 rx_desc = next_rxd;
1043                 buffer_info = next_buffer;
1044         }
1045         rx_ring->next_to_clean = i;
1046
1047         cleaned_count = e1000_desc_unused(rx_ring);
1048         if (cleaned_count)
1049                 adapter->alloc_rx_buf(adapter, cleaned_count);
1050
1051         adapter->total_rx_bytes += total_rx_bytes;
1052         adapter->total_rx_packets += total_rx_packets;
1053         adapter->net_stats.rx_bytes += total_rx_bytes;
1054         adapter->net_stats.rx_packets += total_rx_packets;
1055         return cleaned;
1056 }
1057
1058 /**
1059  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1060  * @adapter: board private structure
1061  **/
1062 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1063 {
1064         struct e1000_ring *rx_ring = adapter->rx_ring;
1065         struct e1000_buffer *buffer_info;
1066         struct e1000_ps_page *ps_page;
1067         struct pci_dev *pdev = adapter->pdev;
1068         unsigned int i, j;
1069
1070         /* Free all the Rx ring sk_buffs */
1071         for (i = 0; i < rx_ring->count; i++) {
1072                 buffer_info = &rx_ring->buffer_info[i];
1073                 if (buffer_info->dma) {
1074                         if (adapter->clean_rx == e1000_clean_rx_irq)
1075                                 pci_unmap_single(pdev, buffer_info->dma,
1076                                                  adapter->rx_buffer_len,
1077                                                  PCI_DMA_FROMDEVICE);
1078                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1079                                 pci_unmap_page(pdev, buffer_info->dma,
1080                                                PAGE_SIZE,
1081                                                PCI_DMA_FROMDEVICE);
1082                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1083                                 pci_unmap_single(pdev, buffer_info->dma,
1084                                                  adapter->rx_ps_bsize0,
1085                                                  PCI_DMA_FROMDEVICE);
1086                         buffer_info->dma = 0;
1087                 }
1088
1089                 if (buffer_info->page) {
1090                         put_page(buffer_info->page);
1091                         buffer_info->page = NULL;
1092                 }
1093
1094                 if (buffer_info->skb) {
1095                         dev_kfree_skb(buffer_info->skb);
1096                         buffer_info->skb = NULL;
1097                 }
1098
1099                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1100                         ps_page = &buffer_info->ps_pages[j];
1101                         if (!ps_page->page)
1102                                 break;
1103                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1104                                        PCI_DMA_FROMDEVICE);
1105                         ps_page->dma = 0;
1106                         put_page(ps_page->page);
1107                         ps_page->page = NULL;
1108                 }
1109         }
1110
1111         /* there also may be some cached data from a chained receive */
1112         if (rx_ring->rx_skb_top) {
1113                 dev_kfree_skb(rx_ring->rx_skb_top);
1114                 rx_ring->rx_skb_top = NULL;
1115         }
1116
1117         /* Zero out the descriptor ring */
1118         memset(rx_ring->desc, 0, rx_ring->size);
1119
1120         rx_ring->next_to_clean = 0;
1121         rx_ring->next_to_use = 0;
1122
1123         writel(0, adapter->hw.hw_addr + rx_ring->head);
1124         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1125 }
1126
1127 static void e1000e_downshift_workaround(struct work_struct *work)
1128 {
1129         struct e1000_adapter *adapter = container_of(work,
1130                                         struct e1000_adapter, downshift_task);
1131
1132         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1133 }
1134
1135 /**
1136  * e1000_intr_msi - Interrupt Handler
1137  * @irq: interrupt number
1138  * @data: pointer to a network interface device structure
1139  **/
1140 static irqreturn_t e1000_intr_msi(int irq, void *data)
1141 {
1142         struct net_device *netdev = data;
1143         struct e1000_adapter *adapter = netdev_priv(netdev);
1144         struct e1000_hw *hw = &adapter->hw;
1145         u32 icr = er32(ICR);
1146
1147         /*
1148          * read ICR disables interrupts using IAM
1149          */
1150
1151         if (icr & E1000_ICR_LSC) {
1152                 hw->mac.get_link_status = 1;
1153                 /*
1154                  * ICH8 workaround-- Call gig speed drop workaround on cable
1155                  * disconnect (LSC) before accessing any PHY registers
1156                  */
1157                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1158                     (!(er32(STATUS) & E1000_STATUS_LU)))
1159                         schedule_work(&adapter->downshift_task);
1160
1161                 /*
1162                  * 80003ES2LAN workaround-- For packet buffer work-around on
1163                  * link down event; disable receives here in the ISR and reset
1164                  * adapter in watchdog
1165                  */
1166                 if (netif_carrier_ok(netdev) &&
1167                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1168                         /* disable receives */
1169                         u32 rctl = er32(RCTL);
1170                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1171                         adapter->flags |= FLAG_RX_RESTART_NOW;
1172                 }
1173                 /* guard against interrupt when we're going down */
1174                 if (!test_bit(__E1000_DOWN, &adapter->state))
1175                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1176         }
1177
1178         if (napi_schedule_prep(&adapter->napi)) {
1179                 adapter->total_tx_bytes = 0;
1180                 adapter->total_tx_packets = 0;
1181                 adapter->total_rx_bytes = 0;
1182                 adapter->total_rx_packets = 0;
1183                 __napi_schedule(&adapter->napi);
1184         }
1185
1186         return IRQ_HANDLED;
1187 }
1188
1189 /**
1190  * e1000_intr - Interrupt Handler
1191  * @irq: interrupt number
1192  * @data: pointer to a network interface device structure
1193  **/
1194 static irqreturn_t e1000_intr(int irq, void *data)
1195 {
1196         struct net_device *netdev = data;
1197         struct e1000_adapter *adapter = netdev_priv(netdev);
1198         struct e1000_hw *hw = &adapter->hw;
1199         u32 rctl, icr = er32(ICR);
1200
1201         if (!icr)
1202                 return IRQ_NONE;  /* Not our interrupt */
1203
1204         /*
1205          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1206          * not set, then the adapter didn't send an interrupt
1207          */
1208         if (!(icr & E1000_ICR_INT_ASSERTED))
1209                 return IRQ_NONE;
1210
1211         /*
1212          * Interrupt Auto-Mask...upon reading ICR,
1213          * interrupts are masked.  No need for the
1214          * IMC write
1215          */
1216
1217         if (icr & E1000_ICR_LSC) {
1218                 hw->mac.get_link_status = 1;
1219                 /*
1220                  * ICH8 workaround-- Call gig speed drop workaround on cable
1221                  * disconnect (LSC) before accessing any PHY registers
1222                  */
1223                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1224                     (!(er32(STATUS) & E1000_STATUS_LU)))
1225                         schedule_work(&adapter->downshift_task);
1226
1227                 /*
1228                  * 80003ES2LAN workaround--
1229                  * For packet buffer work-around on link down event;
1230                  * disable receives here in the ISR and
1231                  * reset adapter in watchdog
1232                  */
1233                 if (netif_carrier_ok(netdev) &&
1234                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1235                         /* disable receives */
1236                         rctl = er32(RCTL);
1237                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1238                         adapter->flags |= FLAG_RX_RESTART_NOW;
1239                 }
1240                 /* guard against interrupt when we're going down */
1241                 if (!test_bit(__E1000_DOWN, &adapter->state))
1242                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1243         }
1244
1245         if (napi_schedule_prep(&adapter->napi)) {
1246                 adapter->total_tx_bytes = 0;
1247                 adapter->total_tx_packets = 0;
1248                 adapter->total_rx_bytes = 0;
1249                 adapter->total_rx_packets = 0;
1250                 __napi_schedule(&adapter->napi);
1251         }
1252
1253         return IRQ_HANDLED;
1254 }
1255
1256 static irqreturn_t e1000_msix_other(int irq, void *data)
1257 {
1258         struct net_device *netdev = data;
1259         struct e1000_adapter *adapter = netdev_priv(netdev);
1260         struct e1000_hw *hw = &adapter->hw;
1261         u32 icr = er32(ICR);
1262
1263         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1264                 ew32(IMS, E1000_IMS_OTHER);
1265                 return IRQ_NONE;
1266         }
1267
1268         if (icr & adapter->eiac_mask)
1269                 ew32(ICS, (icr & adapter->eiac_mask));
1270
1271         if (icr & E1000_ICR_OTHER) {
1272                 if (!(icr & E1000_ICR_LSC))
1273                         goto no_link_interrupt;
1274                 hw->mac.get_link_status = 1;
1275                 /* guard against interrupt when we're going down */
1276                 if (!test_bit(__E1000_DOWN, &adapter->state))
1277                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1278         }
1279
1280 no_link_interrupt:
1281         ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1282
1283         return IRQ_HANDLED;
1284 }
1285
1286
1287 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1288 {
1289         struct net_device *netdev = data;
1290         struct e1000_adapter *adapter = netdev_priv(netdev);
1291         struct e1000_hw *hw = &adapter->hw;
1292         struct e1000_ring *tx_ring = adapter->tx_ring;
1293
1294
1295         adapter->total_tx_bytes = 0;
1296         adapter->total_tx_packets = 0;
1297
1298         if (!e1000_clean_tx_irq(adapter))
1299                 /* Ring was not completely cleaned, so fire another interrupt */
1300                 ew32(ICS, tx_ring->ims_val);
1301
1302         return IRQ_HANDLED;
1303 }
1304
1305 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1306 {
1307         struct net_device *netdev = data;
1308         struct e1000_adapter *adapter = netdev_priv(netdev);
1309
1310         /* Write the ITR value calculated at the end of the
1311          * previous interrupt.
1312          */
1313         if (adapter->rx_ring->set_itr) {
1314                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1315                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1316                 adapter->rx_ring->set_itr = 0;
1317         }
1318
1319         if (napi_schedule_prep(&adapter->napi)) {
1320                 adapter->total_rx_bytes = 0;
1321                 adapter->total_rx_packets = 0;
1322                 __napi_schedule(&adapter->napi);
1323         }
1324         return IRQ_HANDLED;
1325 }
1326
1327 /**
1328  * e1000_configure_msix - Configure MSI-X hardware
1329  *
1330  * e1000_configure_msix sets up the hardware to properly
1331  * generate MSI-X interrupts.
1332  **/
1333 static void e1000_configure_msix(struct e1000_adapter *adapter)
1334 {
1335         struct e1000_hw *hw = &adapter->hw;
1336         struct e1000_ring *rx_ring = adapter->rx_ring;
1337         struct e1000_ring *tx_ring = adapter->tx_ring;
1338         int vector = 0;
1339         u32 ctrl_ext, ivar = 0;
1340
1341         adapter->eiac_mask = 0;
1342
1343         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1344         if (hw->mac.type == e1000_82574) {
1345                 u32 rfctl = er32(RFCTL);
1346                 rfctl |= E1000_RFCTL_ACK_DIS;
1347                 ew32(RFCTL, rfctl);
1348         }
1349
1350 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1351         /* Configure Rx vector */
1352         rx_ring->ims_val = E1000_IMS_RXQ0;
1353         adapter->eiac_mask |= rx_ring->ims_val;
1354         if (rx_ring->itr_val)
1355                 writel(1000000000 / (rx_ring->itr_val * 256),
1356                        hw->hw_addr + rx_ring->itr_register);
1357         else
1358                 writel(1, hw->hw_addr + rx_ring->itr_register);
1359         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1360
1361         /* Configure Tx vector */
1362         tx_ring->ims_val = E1000_IMS_TXQ0;
1363         vector++;
1364         if (tx_ring->itr_val)
1365                 writel(1000000000 / (tx_ring->itr_val * 256),
1366                        hw->hw_addr + tx_ring->itr_register);
1367         else
1368                 writel(1, hw->hw_addr + tx_ring->itr_register);
1369         adapter->eiac_mask |= tx_ring->ims_val;
1370         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1371
1372         /* set vector for Other Causes, e.g. link changes */
1373         vector++;
1374         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1375         if (rx_ring->itr_val)
1376                 writel(1000000000 / (rx_ring->itr_val * 256),
1377                        hw->hw_addr + E1000_EITR_82574(vector));
1378         else
1379                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1380
1381         /* Cause Tx interrupts on every write back */
1382         ivar |= (1 << 31);
1383
1384         ew32(IVAR, ivar);
1385
1386         /* enable MSI-X PBA support */
1387         ctrl_ext = er32(CTRL_EXT);
1388         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1389
1390         /* Auto-Mask Other interrupts upon ICR read */
1391 #define E1000_EIAC_MASK_82574   0x01F00000
1392         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1393         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1394         ew32(CTRL_EXT, ctrl_ext);
1395         e1e_flush();
1396 }
1397
1398 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1399 {
1400         if (adapter->msix_entries) {
1401                 pci_disable_msix(adapter->pdev);
1402                 kfree(adapter->msix_entries);
1403                 adapter->msix_entries = NULL;
1404         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1405                 pci_disable_msi(adapter->pdev);
1406                 adapter->flags &= ~FLAG_MSI_ENABLED;
1407         }
1408
1409         return;
1410 }
1411
1412 /**
1413  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1414  *
1415  * Attempt to configure interrupts using the best available
1416  * capabilities of the hardware and kernel.
1417  **/
1418 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1419 {
1420         int err;
1421         int numvecs, i;
1422
1423
1424         switch (adapter->int_mode) {
1425         case E1000E_INT_MODE_MSIX:
1426                 if (adapter->flags & FLAG_HAS_MSIX) {
1427                         numvecs = 3; /* RxQ0, TxQ0 and other */
1428                         adapter->msix_entries = kcalloc(numvecs,
1429                                                       sizeof(struct msix_entry),
1430                                                       GFP_KERNEL);
1431                         if (adapter->msix_entries) {
1432                                 for (i = 0; i < numvecs; i++)
1433                                         adapter->msix_entries[i].entry = i;
1434
1435                                 err = pci_enable_msix(adapter->pdev,
1436                                                       adapter->msix_entries,
1437                                                       numvecs);
1438                                 if (err == 0)
1439                                         return;
1440                         }
1441                         /* MSI-X failed, so fall through and try MSI */
1442                         e_err("Failed to initialize MSI-X interrupts.  "
1443                               "Falling back to MSI interrupts.\n");
1444                         e1000e_reset_interrupt_capability(adapter);
1445                 }
1446                 adapter->int_mode = E1000E_INT_MODE_MSI;
1447                 /* Fall through */
1448         case E1000E_INT_MODE_MSI:
1449                 if (!pci_enable_msi(adapter->pdev)) {
1450                         adapter->flags |= FLAG_MSI_ENABLED;
1451                 } else {
1452                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1453                         e_err("Failed to initialize MSI interrupts.  Falling "
1454                               "back to legacy interrupts.\n");
1455                 }
1456                 /* Fall through */
1457         case E1000E_INT_MODE_LEGACY:
1458                 /* Don't do anything; this is the system default */
1459                 break;
1460         }
1461
1462         return;
1463 }
1464
1465 /**
1466  * e1000_request_msix - Initialize MSI-X interrupts
1467  *
1468  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1469  * kernel.
1470  **/
1471 static int e1000_request_msix(struct e1000_adapter *adapter)
1472 {
1473         struct net_device *netdev = adapter->netdev;
1474         int err = 0, vector = 0;
1475
1476         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1477                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1478         else
1479                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1480         err = request_irq(adapter->msix_entries[vector].vector,
1481                           &e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1482                           netdev);
1483         if (err)
1484                 goto out;
1485         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1486         adapter->rx_ring->itr_val = adapter->itr;
1487         vector++;
1488
1489         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1490                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1491         else
1492                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1493         err = request_irq(adapter->msix_entries[vector].vector,
1494                           &e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1495                           netdev);
1496         if (err)
1497                 goto out;
1498         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1499         adapter->tx_ring->itr_val = adapter->itr;
1500         vector++;
1501
1502         err = request_irq(adapter->msix_entries[vector].vector,
1503                           &e1000_msix_other, 0, netdev->name, netdev);
1504         if (err)
1505                 goto out;
1506
1507         e1000_configure_msix(adapter);
1508         return 0;
1509 out:
1510         return err;
1511 }
1512
1513 /**
1514  * e1000_request_irq - initialize interrupts
1515  *
1516  * Attempts to configure interrupts using the best available
1517  * capabilities of the hardware and kernel.
1518  **/
1519 static int e1000_request_irq(struct e1000_adapter *adapter)
1520 {
1521         struct net_device *netdev = adapter->netdev;
1522         int err;
1523
1524         if (adapter->msix_entries) {
1525                 err = e1000_request_msix(adapter);
1526                 if (!err)
1527                         return err;
1528                 /* fall back to MSI */
1529                 e1000e_reset_interrupt_capability(adapter);
1530                 adapter->int_mode = E1000E_INT_MODE_MSI;
1531                 e1000e_set_interrupt_capability(adapter);
1532         }
1533         if (adapter->flags & FLAG_MSI_ENABLED) {
1534                 err = request_irq(adapter->pdev->irq, &e1000_intr_msi, 0,
1535                                   netdev->name, netdev);
1536                 if (!err)
1537                         return err;
1538
1539                 /* fall back to legacy interrupt */
1540                 e1000e_reset_interrupt_capability(adapter);
1541                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1542         }
1543
1544         err = request_irq(adapter->pdev->irq, &e1000_intr, IRQF_SHARED,
1545                           netdev->name, netdev);
1546         if (err)
1547                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1548
1549         return err;
1550 }
1551
1552 static void e1000_free_irq(struct e1000_adapter *adapter)
1553 {
1554         struct net_device *netdev = adapter->netdev;
1555
1556         if (adapter->msix_entries) {
1557                 int vector = 0;
1558
1559                 free_irq(adapter->msix_entries[vector].vector, netdev);
1560                 vector++;
1561
1562                 free_irq(adapter->msix_entries[vector].vector, netdev);
1563                 vector++;
1564
1565                 /* Other Causes interrupt vector */
1566                 free_irq(adapter->msix_entries[vector].vector, netdev);
1567                 return;
1568         }
1569
1570         free_irq(adapter->pdev->irq, netdev);
1571 }
1572
1573 /**
1574  * e1000_irq_disable - Mask off interrupt generation on the NIC
1575  **/
1576 static void e1000_irq_disable(struct e1000_adapter *adapter)
1577 {
1578         struct e1000_hw *hw = &adapter->hw;
1579
1580         ew32(IMC, ~0);
1581         if (adapter->msix_entries)
1582                 ew32(EIAC_82574, 0);
1583         e1e_flush();
1584         synchronize_irq(adapter->pdev->irq);
1585 }
1586
1587 /**
1588  * e1000_irq_enable - Enable default interrupt generation settings
1589  **/
1590 static void e1000_irq_enable(struct e1000_adapter *adapter)
1591 {
1592         struct e1000_hw *hw = &adapter->hw;
1593
1594         if (adapter->msix_entries) {
1595                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1596                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1597         } else {
1598                 ew32(IMS, IMS_ENABLE_MASK);
1599         }
1600         e1e_flush();
1601 }
1602
1603 /**
1604  * e1000_get_hw_control - get control of the h/w from f/w
1605  * @adapter: address of board private structure
1606  *
1607  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1608  * For ASF and Pass Through versions of f/w this means that
1609  * the driver is loaded. For AMT version (only with 82573)
1610  * of the f/w this means that the network i/f is open.
1611  **/
1612 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1613 {
1614         struct e1000_hw *hw = &adapter->hw;
1615         u32 ctrl_ext;
1616         u32 swsm;
1617
1618         /* Let firmware know the driver has taken over */
1619         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1620                 swsm = er32(SWSM);
1621                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1622         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1623                 ctrl_ext = er32(CTRL_EXT);
1624                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1625         }
1626 }
1627
1628 /**
1629  * e1000_release_hw_control - release control of the h/w to f/w
1630  * @adapter: address of board private structure
1631  *
1632  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1633  * For ASF and Pass Through versions of f/w this means that the
1634  * driver is no longer loaded. For AMT version (only with 82573) i
1635  * of the f/w this means that the network i/f is closed.
1636  *
1637  **/
1638 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1639 {
1640         struct e1000_hw *hw = &adapter->hw;
1641         u32 ctrl_ext;
1642         u32 swsm;
1643
1644         /* Let firmware taken over control of h/w */
1645         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1646                 swsm = er32(SWSM);
1647                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1648         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1649                 ctrl_ext = er32(CTRL_EXT);
1650                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1651         }
1652 }
1653
1654 /**
1655  * @e1000_alloc_ring - allocate memory for a ring structure
1656  **/
1657 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1658                                 struct e1000_ring *ring)
1659 {
1660         struct pci_dev *pdev = adapter->pdev;
1661
1662         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1663                                         GFP_KERNEL);
1664         if (!ring->desc)
1665                 return -ENOMEM;
1666
1667         return 0;
1668 }
1669
1670 /**
1671  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1672  * @adapter: board private structure
1673  *
1674  * Return 0 on success, negative on failure
1675  **/
1676 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1677 {
1678         struct e1000_ring *tx_ring = adapter->tx_ring;
1679         int err = -ENOMEM, size;
1680
1681         size = sizeof(struct e1000_buffer) * tx_ring->count;
1682         tx_ring->buffer_info = vmalloc(size);
1683         if (!tx_ring->buffer_info)
1684                 goto err;
1685         memset(tx_ring->buffer_info, 0, size);
1686
1687         /* round up to nearest 4K */
1688         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1689         tx_ring->size = ALIGN(tx_ring->size, 4096);
1690
1691         err = e1000_alloc_ring_dma(adapter, tx_ring);
1692         if (err)
1693                 goto err;
1694
1695         tx_ring->next_to_use = 0;
1696         tx_ring->next_to_clean = 0;
1697
1698         return 0;
1699 err:
1700         vfree(tx_ring->buffer_info);
1701         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1702         return err;
1703 }
1704
1705 /**
1706  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1707  * @adapter: board private structure
1708  *
1709  * Returns 0 on success, negative on failure
1710  **/
1711 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1712 {
1713         struct e1000_ring *rx_ring = adapter->rx_ring;
1714         struct e1000_buffer *buffer_info;
1715         int i, size, desc_len, err = -ENOMEM;
1716
1717         size = sizeof(struct e1000_buffer) * rx_ring->count;
1718         rx_ring->buffer_info = vmalloc(size);
1719         if (!rx_ring->buffer_info)
1720                 goto err;
1721         memset(rx_ring->buffer_info, 0, size);
1722
1723         for (i = 0; i < rx_ring->count; i++) {
1724                 buffer_info = &rx_ring->buffer_info[i];
1725                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1726                                                 sizeof(struct e1000_ps_page),
1727                                                 GFP_KERNEL);
1728                 if (!buffer_info->ps_pages)
1729                         goto err_pages;
1730         }
1731
1732         desc_len = sizeof(union e1000_rx_desc_packet_split);
1733
1734         /* Round up to nearest 4K */
1735         rx_ring->size = rx_ring->count * desc_len;
1736         rx_ring->size = ALIGN(rx_ring->size, 4096);
1737
1738         err = e1000_alloc_ring_dma(adapter, rx_ring);
1739         if (err)
1740                 goto err_pages;
1741
1742         rx_ring->next_to_clean = 0;
1743         rx_ring->next_to_use = 0;
1744         rx_ring->rx_skb_top = NULL;
1745
1746         return 0;
1747
1748 err_pages:
1749         for (i = 0; i < rx_ring->count; i++) {
1750                 buffer_info = &rx_ring->buffer_info[i];
1751                 kfree(buffer_info->ps_pages);
1752         }
1753 err:
1754         vfree(rx_ring->buffer_info);
1755         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1756         return err;
1757 }
1758
1759 /**
1760  * e1000_clean_tx_ring - Free Tx Buffers
1761  * @adapter: board private structure
1762  **/
1763 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1764 {
1765         struct e1000_ring *tx_ring = adapter->tx_ring;
1766         struct e1000_buffer *buffer_info;
1767         unsigned long size;
1768         unsigned int i;
1769
1770         for (i = 0; i < tx_ring->count; i++) {
1771                 buffer_info = &tx_ring->buffer_info[i];
1772                 e1000_put_txbuf(adapter, buffer_info);
1773         }
1774
1775         size = sizeof(struct e1000_buffer) * tx_ring->count;
1776         memset(tx_ring->buffer_info, 0, size);
1777
1778         memset(tx_ring->desc, 0, tx_ring->size);
1779
1780         tx_ring->next_to_use = 0;
1781         tx_ring->next_to_clean = 0;
1782
1783         writel(0, adapter->hw.hw_addr + tx_ring->head);
1784         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1785 }
1786
1787 /**
1788  * e1000e_free_tx_resources - Free Tx Resources per Queue
1789  * @adapter: board private structure
1790  *
1791  * Free all transmit software resources
1792  **/
1793 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1794 {
1795         struct pci_dev *pdev = adapter->pdev;
1796         struct e1000_ring *tx_ring = adapter->tx_ring;
1797
1798         e1000_clean_tx_ring(adapter);
1799
1800         vfree(tx_ring->buffer_info);
1801         tx_ring->buffer_info = NULL;
1802
1803         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1804                           tx_ring->dma);
1805         tx_ring->desc = NULL;
1806 }
1807
1808 /**
1809  * e1000e_free_rx_resources - Free Rx Resources
1810  * @adapter: board private structure
1811  *
1812  * Free all receive software resources
1813  **/
1814
1815 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1816 {
1817         struct pci_dev *pdev = adapter->pdev;
1818         struct e1000_ring *rx_ring = adapter->rx_ring;
1819         int i;
1820
1821         e1000_clean_rx_ring(adapter);
1822
1823         for (i = 0; i < rx_ring->count; i++) {
1824                 kfree(rx_ring->buffer_info[i].ps_pages);
1825         }
1826
1827         vfree(rx_ring->buffer_info);
1828         rx_ring->buffer_info = NULL;
1829
1830         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1831                           rx_ring->dma);
1832         rx_ring->desc = NULL;
1833 }
1834
1835 /**
1836  * e1000_update_itr - update the dynamic ITR value based on statistics
1837  * @adapter: pointer to adapter
1838  * @itr_setting: current adapter->itr
1839  * @packets: the number of packets during this measurement interval
1840  * @bytes: the number of bytes during this measurement interval
1841  *
1842  *      Stores a new ITR value based on packets and byte
1843  *      counts during the last interrupt.  The advantage of per interrupt
1844  *      computation is faster updates and more accurate ITR for the current
1845  *      traffic pattern.  Constants in this function were computed
1846  *      based on theoretical maximum wire speed and thresholds were set based
1847  *      on testing data as well as attempting to minimize response time
1848  *      while increasing bulk throughput.  This functionality is controlled
1849  *      by the InterruptThrottleRate module parameter.
1850  **/
1851 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1852                                      u16 itr_setting, int packets,
1853                                      int bytes)
1854 {
1855         unsigned int retval = itr_setting;
1856
1857         if (packets == 0)
1858                 goto update_itr_done;
1859
1860         switch (itr_setting) {
1861         case lowest_latency:
1862                 /* handle TSO and jumbo frames */
1863                 if (bytes/packets > 8000)
1864                         retval = bulk_latency;
1865                 else if ((packets < 5) && (bytes > 512)) {
1866                         retval = low_latency;
1867                 }
1868                 break;
1869         case low_latency:  /* 50 usec aka 20000 ints/s */
1870                 if (bytes > 10000) {
1871                         /* this if handles the TSO accounting */
1872                         if (bytes/packets > 8000) {
1873                                 retval = bulk_latency;
1874                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1875                                 retval = bulk_latency;
1876                         } else if ((packets > 35)) {
1877                                 retval = lowest_latency;
1878                         }
1879                 } else if (bytes/packets > 2000) {
1880                         retval = bulk_latency;
1881                 } else if (packets <= 2 && bytes < 512) {
1882                         retval = lowest_latency;
1883                 }
1884                 break;
1885         case bulk_latency: /* 250 usec aka 4000 ints/s */
1886                 if (bytes > 25000) {
1887                         if (packets > 35) {
1888                                 retval = low_latency;
1889                         }
1890                 } else if (bytes < 6000) {
1891                         retval = low_latency;
1892                 }
1893                 break;
1894         }
1895
1896 update_itr_done:
1897         return retval;
1898 }
1899
1900 static void e1000_set_itr(struct e1000_adapter *adapter)
1901 {
1902         struct e1000_hw *hw = &adapter->hw;
1903         u16 current_itr;
1904         u32 new_itr = adapter->itr;
1905
1906         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1907         if (adapter->link_speed != SPEED_1000) {
1908                 current_itr = 0;
1909                 new_itr = 4000;
1910                 goto set_itr_now;
1911         }
1912
1913         adapter->tx_itr = e1000_update_itr(adapter,
1914                                     adapter->tx_itr,
1915                                     adapter->total_tx_packets,
1916                                     adapter->total_tx_bytes);
1917         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1918         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1919                 adapter->tx_itr = low_latency;
1920
1921         adapter->rx_itr = e1000_update_itr(adapter,
1922                                     adapter->rx_itr,
1923                                     adapter->total_rx_packets,
1924                                     adapter->total_rx_bytes);
1925         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1926         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1927                 adapter->rx_itr = low_latency;
1928
1929         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1930
1931         switch (current_itr) {
1932         /* counts and packets in update_itr are dependent on these numbers */
1933         case lowest_latency:
1934                 new_itr = 70000;
1935                 break;
1936         case low_latency:
1937                 new_itr = 20000; /* aka hwitr = ~200 */
1938                 break;
1939         case bulk_latency:
1940                 new_itr = 4000;
1941                 break;
1942         default:
1943                 break;
1944         }
1945
1946 set_itr_now:
1947         if (new_itr != adapter->itr) {
1948                 /*
1949                  * this attempts to bias the interrupt rate towards Bulk
1950                  * by adding intermediate steps when interrupt rate is
1951                  * increasing
1952                  */
1953                 new_itr = new_itr > adapter->itr ?
1954                              min(adapter->itr + (new_itr >> 2), new_itr) :
1955                              new_itr;
1956                 adapter->itr = new_itr;
1957                 adapter->rx_ring->itr_val = new_itr;
1958                 if (adapter->msix_entries)
1959                         adapter->rx_ring->set_itr = 1;
1960                 else
1961                         ew32(ITR, 1000000000 / (new_itr * 256));
1962         }
1963 }
1964
1965 /**
1966  * e1000_alloc_queues - Allocate memory for all rings
1967  * @adapter: board private structure to initialize
1968  **/
1969 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1970 {
1971         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1972         if (!adapter->tx_ring)
1973                 goto err;
1974
1975         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1976         if (!adapter->rx_ring)
1977                 goto err;
1978
1979         return 0;
1980 err:
1981         e_err("Unable to allocate memory for queues\n");
1982         kfree(adapter->rx_ring);
1983         kfree(adapter->tx_ring);
1984         return -ENOMEM;
1985 }
1986
1987 /**
1988  * e1000_clean - NAPI Rx polling callback
1989  * @napi: struct associated with this polling callback
1990  * @budget: amount of packets driver is allowed to process this poll
1991  **/
1992 static int e1000_clean(struct napi_struct *napi, int budget)
1993 {
1994         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1995         struct e1000_hw *hw = &adapter->hw;
1996         struct net_device *poll_dev = adapter->netdev;
1997         int tx_cleaned = 0, work_done = 0;
1998
1999         adapter = netdev_priv(poll_dev);
2000
2001         if (adapter->msix_entries &&
2002             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2003                 goto clean_rx;
2004
2005         tx_cleaned = e1000_clean_tx_irq(adapter);
2006
2007 clean_rx:
2008         adapter->clean_rx(adapter, &work_done, budget);
2009
2010         if (!tx_cleaned)
2011                 work_done = budget;
2012
2013         /* If budget not fully consumed, exit the polling mode */
2014         if (work_done < budget) {
2015                 if (adapter->itr_setting & 3)
2016                         e1000_set_itr(adapter);
2017                 napi_complete(napi);
2018                 if (adapter->msix_entries)
2019                         ew32(IMS, adapter->rx_ring->ims_val);
2020                 else
2021                         e1000_irq_enable(adapter);
2022         }
2023
2024         return work_done;
2025 }
2026
2027 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2028 {
2029         struct e1000_adapter *adapter = netdev_priv(netdev);
2030         struct e1000_hw *hw = &adapter->hw;
2031         u32 vfta, index;
2032
2033         /* don't update vlan cookie if already programmed */
2034         if ((adapter->hw.mng_cookie.status &
2035              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2036             (vid == adapter->mng_vlan_id))
2037                 return;
2038         /* add VID to filter table */
2039         index = (vid >> 5) & 0x7F;
2040         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2041         vfta |= (1 << (vid & 0x1F));
2042         e1000e_write_vfta(hw, index, vfta);
2043 }
2044
2045 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2046 {
2047         struct e1000_adapter *adapter = netdev_priv(netdev);
2048         struct e1000_hw *hw = &adapter->hw;
2049         u32 vfta, index;
2050
2051         if (!test_bit(__E1000_DOWN, &adapter->state))
2052                 e1000_irq_disable(adapter);
2053         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2054
2055         if (!test_bit(__E1000_DOWN, &adapter->state))
2056                 e1000_irq_enable(adapter);
2057
2058         if ((adapter->hw.mng_cookie.status &
2059              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2060             (vid == adapter->mng_vlan_id)) {
2061                 /* release control to f/w */
2062                 e1000_release_hw_control(adapter);
2063                 return;
2064         }
2065
2066         /* remove VID from filter table */
2067         index = (vid >> 5) & 0x7F;
2068         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2069         vfta &= ~(1 << (vid & 0x1F));
2070         e1000e_write_vfta(hw, index, vfta);
2071 }
2072
2073 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2074 {
2075         struct net_device *netdev = adapter->netdev;
2076         u16 vid = adapter->hw.mng_cookie.vlan_id;
2077         u16 old_vid = adapter->mng_vlan_id;
2078
2079         if (!adapter->vlgrp)
2080                 return;
2081
2082         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2083                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2084                 if (adapter->hw.mng_cookie.status &
2085                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2086                         e1000_vlan_rx_add_vid(netdev, vid);
2087                         adapter->mng_vlan_id = vid;
2088                 }
2089
2090                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2091                                 (vid != old_vid) &&
2092                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2093                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2094         } else {
2095                 adapter->mng_vlan_id = vid;
2096         }
2097 }
2098
2099
2100 static void e1000_vlan_rx_register(struct net_device *netdev,
2101                                    struct vlan_group *grp)
2102 {
2103         struct e1000_adapter *adapter = netdev_priv(netdev);
2104         struct e1000_hw *hw = &adapter->hw;
2105         u32 ctrl, rctl;
2106
2107         if (!test_bit(__E1000_DOWN, &adapter->state))
2108                 e1000_irq_disable(adapter);
2109         adapter->vlgrp = grp;
2110
2111         if (grp) {
2112                 /* enable VLAN tag insert/strip */
2113                 ctrl = er32(CTRL);
2114                 ctrl |= E1000_CTRL_VME;
2115                 ew32(CTRL, ctrl);
2116
2117                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2118                         /* enable VLAN receive filtering */
2119                         rctl = er32(RCTL);
2120                         rctl &= ~E1000_RCTL_CFIEN;
2121                         ew32(RCTL, rctl);
2122                         e1000_update_mng_vlan(adapter);
2123                 }
2124         } else {
2125                 /* disable VLAN tag insert/strip */
2126                 ctrl = er32(CTRL);
2127                 ctrl &= ~E1000_CTRL_VME;
2128                 ew32(CTRL, ctrl);
2129
2130                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2131                         if (adapter->mng_vlan_id !=
2132                             (u16)E1000_MNG_VLAN_NONE) {
2133                                 e1000_vlan_rx_kill_vid(netdev,
2134                                                        adapter->mng_vlan_id);
2135                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2136                         }
2137                 }
2138         }
2139
2140         if (!test_bit(__E1000_DOWN, &adapter->state))
2141                 e1000_irq_enable(adapter);
2142 }
2143
2144 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2145 {
2146         u16 vid;
2147
2148         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2149
2150         if (!adapter->vlgrp)
2151                 return;
2152
2153         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2154                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2155                         continue;
2156                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2157         }
2158 }
2159
2160 static void e1000_init_manageability(struct e1000_adapter *adapter)
2161 {
2162         struct e1000_hw *hw = &adapter->hw;
2163         u32 manc, manc2h;
2164
2165         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2166                 return;
2167
2168         manc = er32(MANC);
2169
2170         /*
2171          * enable receiving management packets to the host. this will probably
2172          * generate destination unreachable messages from the host OS, but
2173          * the packets will be handled on SMBUS
2174          */
2175         manc |= E1000_MANC_EN_MNG2HOST;
2176         manc2h = er32(MANC2H);
2177 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2178 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2179         manc2h |= E1000_MNG2HOST_PORT_623;
2180         manc2h |= E1000_MNG2HOST_PORT_664;
2181         ew32(MANC2H, manc2h);
2182         ew32(MANC, manc);
2183 }
2184
2185 /**
2186  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2187  * @adapter: board private structure
2188  *
2189  * Configure the Tx unit of the MAC after a reset.
2190  **/
2191 static void e1000_configure_tx(struct e1000_adapter *adapter)
2192 {
2193         struct e1000_hw *hw = &adapter->hw;
2194         struct e1000_ring *tx_ring = adapter->tx_ring;
2195         u64 tdba;
2196         u32 tdlen, tctl, tipg, tarc;
2197         u32 ipgr1, ipgr2;
2198
2199         /* Setup the HW Tx Head and Tail descriptor pointers */
2200         tdba = tx_ring->dma;
2201         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2202         ew32(TDBAL, (tdba & DMA_32BIT_MASK));
2203         ew32(TDBAH, (tdba >> 32));
2204         ew32(TDLEN, tdlen);
2205         ew32(TDH, 0);
2206         ew32(TDT, 0);
2207         tx_ring->head = E1000_TDH;
2208         tx_ring->tail = E1000_TDT;
2209
2210         /* Set the default values for the Tx Inter Packet Gap timer */
2211         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2212         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2213         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2214
2215         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2216                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2217
2218         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2219         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2220         ew32(TIPG, tipg);
2221
2222         /* Set the Tx Interrupt Delay register */
2223         ew32(TIDV, adapter->tx_int_delay);
2224         /* Tx irq moderation */
2225         ew32(TADV, adapter->tx_abs_int_delay);
2226
2227         /* Program the Transmit Control Register */
2228         tctl = er32(TCTL);
2229         tctl &= ~E1000_TCTL_CT;
2230         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2231                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2232
2233         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2234                 tarc = er32(TARC(0));
2235                 /*
2236                  * set the speed mode bit, we'll clear it if we're not at
2237                  * gigabit link later
2238                  */
2239 #define SPEED_MODE_BIT (1 << 21)
2240                 tarc |= SPEED_MODE_BIT;
2241                 ew32(TARC(0), tarc);
2242         }
2243
2244         /* errata: program both queues to unweighted RR */
2245         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2246                 tarc = er32(TARC(0));
2247                 tarc |= 1;
2248                 ew32(TARC(0), tarc);
2249                 tarc = er32(TARC(1));
2250                 tarc |= 1;
2251                 ew32(TARC(1), tarc);
2252         }
2253
2254         e1000e_config_collision_dist(hw);
2255
2256         /* Setup Transmit Descriptor Settings for eop descriptor */
2257         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2258
2259         /* only set IDE if we are delaying interrupts using the timers */
2260         if (adapter->tx_int_delay)
2261                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2262
2263         /* enable Report Status bit */
2264         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2265
2266         ew32(TCTL, tctl);
2267
2268         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
2269 }
2270
2271 /**
2272  * e1000_setup_rctl - configure the receive control registers
2273  * @adapter: Board private structure
2274  **/
2275 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2276                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2277 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2278 {
2279         struct e1000_hw *hw = &adapter->hw;
2280         u32 rctl, rfctl;
2281         u32 psrctl = 0;
2282         u32 pages = 0;
2283
2284         /* Program MC offset vector base */
2285         rctl = er32(RCTL);
2286         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2287         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2288                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2289                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2290
2291         /* Do not Store bad packets */
2292         rctl &= ~E1000_RCTL_SBP;
2293
2294         /* Enable Long Packet receive */
2295         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2296                 rctl &= ~E1000_RCTL_LPE;
2297         else
2298                 rctl |= E1000_RCTL_LPE;
2299
2300         /* Some systems expect that the CRC is included in SMBUS traffic. The
2301          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2302          * host memory when this is enabled
2303          */
2304         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2305                 rctl |= E1000_RCTL_SECRC;
2306
2307         /* Setup buffer sizes */
2308         rctl &= ~E1000_RCTL_SZ_4096;
2309         rctl |= E1000_RCTL_BSEX;
2310         switch (adapter->rx_buffer_len) {
2311         case 256:
2312                 rctl |= E1000_RCTL_SZ_256;
2313                 rctl &= ~E1000_RCTL_BSEX;
2314                 break;
2315         case 512:
2316                 rctl |= E1000_RCTL_SZ_512;
2317                 rctl &= ~E1000_RCTL_BSEX;
2318                 break;
2319         case 1024:
2320                 rctl |= E1000_RCTL_SZ_1024;
2321                 rctl &= ~E1000_RCTL_BSEX;
2322                 break;
2323         case 2048:
2324         default:
2325                 rctl |= E1000_RCTL_SZ_2048;
2326                 rctl &= ~E1000_RCTL_BSEX;
2327                 break;
2328         case 4096:
2329                 rctl |= E1000_RCTL_SZ_4096;
2330                 break;
2331         case 8192:
2332                 rctl |= E1000_RCTL_SZ_8192;
2333                 break;
2334         case 16384:
2335                 rctl |= E1000_RCTL_SZ_16384;
2336                 break;
2337         }
2338
2339         /*
2340          * 82571 and greater support packet-split where the protocol
2341          * header is placed in skb->data and the packet data is
2342          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2343          * In the case of a non-split, skb->data is linearly filled,
2344          * followed by the page buffers.  Therefore, skb->data is
2345          * sized to hold the largest protocol header.
2346          *
2347          * allocations using alloc_page take too long for regular MTU
2348          * so only enable packet split for jumbo frames
2349          *
2350          * Using pages when the page size is greater than 16k wastes
2351          * a lot of memory, since we allocate 3 pages at all times
2352          * per packet.
2353          */
2354         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2355         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2356             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2357                 adapter->rx_ps_pages = pages;
2358         else
2359                 adapter->rx_ps_pages = 0;
2360
2361         if (adapter->rx_ps_pages) {
2362                 /* Configure extra packet-split registers */
2363                 rfctl = er32(RFCTL);
2364                 rfctl |= E1000_RFCTL_EXTEN;
2365                 /*
2366                  * disable packet split support for IPv6 extension headers,
2367                  * because some malformed IPv6 headers can hang the Rx
2368                  */
2369                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2370                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2371
2372                 ew32(RFCTL, rfctl);
2373
2374                 /* Enable Packet split descriptors */
2375                 rctl |= E1000_RCTL_DTYP_PS;
2376
2377                 psrctl |= adapter->rx_ps_bsize0 >>
2378                         E1000_PSRCTL_BSIZE0_SHIFT;
2379
2380                 switch (adapter->rx_ps_pages) {
2381                 case 3:
2382                         psrctl |= PAGE_SIZE <<
2383                                 E1000_PSRCTL_BSIZE3_SHIFT;
2384                 case 2:
2385                         psrctl |= PAGE_SIZE <<
2386                                 E1000_PSRCTL_BSIZE2_SHIFT;
2387                 case 1:
2388                         psrctl |= PAGE_SIZE >>
2389                                 E1000_PSRCTL_BSIZE1_SHIFT;
2390                         break;
2391                 }
2392
2393                 ew32(PSRCTL, psrctl);
2394         }
2395
2396         ew32(RCTL, rctl);
2397         /* just started the receive unit, no need to restart */
2398         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2399 }
2400
2401 /**
2402  * e1000_configure_rx - Configure Receive Unit after Reset
2403  * @adapter: board private structure
2404  *
2405  * Configure the Rx unit of the MAC after a reset.
2406  **/
2407 static void e1000_configure_rx(struct e1000_adapter *adapter)
2408 {
2409         struct e1000_hw *hw = &adapter->hw;
2410         struct e1000_ring *rx_ring = adapter->rx_ring;
2411         u64 rdba;
2412         u32 rdlen, rctl, rxcsum, ctrl_ext;
2413
2414         if (adapter->rx_ps_pages) {
2415                 /* this is a 32 byte descriptor */
2416                 rdlen = rx_ring->count *
2417                         sizeof(union e1000_rx_desc_packet_split);
2418                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2419                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2420         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2421                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2422                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2423                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2424         } else {
2425                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2426                 adapter->clean_rx = e1000_clean_rx_irq;
2427                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2428         }
2429
2430         /* disable receives while setting up the descriptors */
2431         rctl = er32(RCTL);
2432         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2433         e1e_flush();
2434         msleep(10);
2435
2436         /* set the Receive Delay Timer Register */
2437         ew32(RDTR, adapter->rx_int_delay);
2438
2439         /* irq moderation */
2440         ew32(RADV, adapter->rx_abs_int_delay);
2441         if (adapter->itr_setting != 0)
2442                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2443
2444         ctrl_ext = er32(CTRL_EXT);
2445         /* Reset delay timers after every interrupt */
2446         ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2447         /* Auto-Mask interrupts upon ICR access */
2448         ctrl_ext |= E1000_CTRL_EXT_IAME;
2449         ew32(IAM, 0xffffffff);
2450         ew32(CTRL_EXT, ctrl_ext);
2451         e1e_flush();
2452
2453         /*
2454          * Setup the HW Rx Head and Tail Descriptor Pointers and
2455          * the Base and Length of the Rx Descriptor Ring
2456          */
2457         rdba = rx_ring->dma;
2458         ew32(RDBAL, (rdba & DMA_32BIT_MASK));
2459         ew32(RDBAH, (rdba >> 32));
2460         ew32(RDLEN, rdlen);
2461         ew32(RDH, 0);
2462         ew32(RDT, 0);
2463         rx_ring->head = E1000_RDH;
2464         rx_ring->tail = E1000_RDT;
2465
2466         /* Enable Receive Checksum Offload for TCP and UDP */
2467         rxcsum = er32(RXCSUM);
2468         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2469                 rxcsum |= E1000_RXCSUM_TUOFL;
2470
2471                 /*
2472                  * IPv4 payload checksum for UDP fragments must be
2473                  * used in conjunction with packet-split.
2474                  */
2475                 if (adapter->rx_ps_pages)
2476                         rxcsum |= E1000_RXCSUM_IPPCSE;
2477         } else {
2478                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2479                 /* no need to clear IPPCSE as it defaults to 0 */
2480         }
2481         ew32(RXCSUM, rxcsum);
2482
2483         /*
2484          * Enable early receives on supported devices, only takes effect when
2485          * packet size is equal or larger than the specified value (in 8 byte
2486          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2487          */
2488         if ((adapter->flags & FLAG_HAS_ERT) &&
2489             (adapter->netdev->mtu > ETH_DATA_LEN)) {
2490                 u32 rxdctl = er32(RXDCTL(0));
2491                 ew32(RXDCTL(0), rxdctl | 0x3);
2492                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2493                 /*
2494                  * With jumbo frames and early-receive enabled, excessive
2495                  * C4->C2 latencies result in dropped transactions.
2496                  */
2497                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2498                                           e1000e_driver_name, 55);
2499         } else {
2500                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2501                                           e1000e_driver_name,
2502                                           PM_QOS_DEFAULT_VALUE);
2503         }
2504
2505         /* Enable Receives */
2506         ew32(RCTL, rctl);
2507 }
2508
2509 /**
2510  *  e1000_update_mc_addr_list - Update Multicast addresses
2511  *  @hw: pointer to the HW structure
2512  *  @mc_addr_list: array of multicast addresses to program
2513  *  @mc_addr_count: number of multicast addresses to program
2514  *  @rar_used_count: the first RAR register free to program
2515  *  @rar_count: total number of supported Receive Address Registers
2516  *
2517  *  Updates the Receive Address Registers and Multicast Table Array.
2518  *  The caller must have a packed mc_addr_list of multicast addresses.
2519  *  The parameter rar_count will usually be hw->mac.rar_entry_count
2520  *  unless there are workarounds that change this.  Currently no func pointer
2521  *  exists and all implementations are handled in the generic version of this
2522  *  function.
2523  **/
2524 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2525                                       u32 mc_addr_count, u32 rar_used_count,
2526                                       u32 rar_count)
2527 {
2528         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2529                                         rar_used_count, rar_count);
2530 }
2531
2532 /**
2533  * e1000_set_multi - Multicast and Promiscuous mode set
2534  * @netdev: network interface device structure
2535  *
2536  * The set_multi entry point is called whenever the multicast address
2537  * list or the network interface flags are updated.  This routine is
2538  * responsible for configuring the hardware for proper multicast,
2539  * promiscuous mode, and all-multi behavior.
2540  **/
2541 static void e1000_set_multi(struct net_device *netdev)
2542 {
2543         struct e1000_adapter *adapter = netdev_priv(netdev);
2544         struct e1000_hw *hw = &adapter->hw;
2545         struct e1000_mac_info *mac = &hw->mac;
2546         struct dev_mc_list *mc_ptr;
2547         u8  *mta_list;
2548         u32 rctl;
2549         int i;
2550
2551         /* Check for Promiscuous and All Multicast modes */
2552
2553         rctl = er32(RCTL);
2554
2555         if (netdev->flags & IFF_PROMISC) {
2556                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2557                 rctl &= ~E1000_RCTL_VFE;
2558         } else {
2559                 if (netdev->flags & IFF_ALLMULTI) {
2560                         rctl |= E1000_RCTL_MPE;
2561                         rctl &= ~E1000_RCTL_UPE;
2562                 } else {
2563                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2564                 }
2565                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2566                         rctl |= E1000_RCTL_VFE;
2567         }
2568
2569         ew32(RCTL, rctl);
2570
2571         if (netdev->mc_count) {
2572                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2573                 if (!mta_list)
2574                         return;
2575
2576                 /* prepare a packed array of only addresses. */
2577                 mc_ptr = netdev->mc_list;
2578
2579                 for (i = 0; i < netdev->mc_count; i++) {
2580                         if (!mc_ptr)
2581                                 break;
2582                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2583                                ETH_ALEN);
2584                         mc_ptr = mc_ptr->next;
2585                 }
2586
2587                 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2588                                           mac->rar_entry_count);
2589                 kfree(mta_list);
2590         } else {
2591                 /*
2592                  * if we're called from probe, we might not have
2593                  * anything to do here, so clear out the list
2594                  */
2595                 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2596         }
2597 }
2598
2599 /**
2600  * e1000_configure - configure the hardware for Rx and Tx
2601  * @adapter: private board structure
2602  **/
2603 static void e1000_configure(struct e1000_adapter *adapter)
2604 {
2605         e1000_set_multi(adapter->netdev);
2606
2607         e1000_restore_vlan(adapter);
2608         e1000_init_manageability(adapter);
2609
2610         e1000_configure_tx(adapter);
2611         e1000_setup_rctl(adapter);
2612         e1000_configure_rx(adapter);
2613         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2614 }
2615
2616 /**
2617  * e1000e_power_up_phy - restore link in case the phy was powered down
2618  * @adapter: address of board private structure
2619  *
2620  * The phy may be powered down to save power and turn off link when the
2621  * driver is unloaded and wake on lan is not enabled (among others)
2622  * *** this routine MUST be followed by a call to e1000e_reset ***
2623  **/
2624 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2625 {
2626         u16 mii_reg = 0;
2627
2628         /* Just clear the power down bit to wake the phy back up */
2629         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2630                 /*
2631                  * According to the manual, the phy will retain its
2632                  * settings across a power-down/up cycle
2633                  */
2634                 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2635                 mii_reg &= ~MII_CR_POWER_DOWN;
2636                 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2637         }
2638
2639         adapter->hw.mac.ops.setup_link(&adapter->hw);
2640 }
2641
2642 /**
2643  * e1000_power_down_phy - Power down the PHY
2644  *
2645  * Power down the PHY so no link is implied when interface is down
2646  * The PHY cannot be powered down is management or WoL is active
2647  */
2648 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2649 {
2650         struct e1000_hw *hw = &adapter->hw;
2651         u16 mii_reg;
2652
2653         /* WoL is enabled */
2654         if (adapter->wol)
2655                 return;
2656
2657         /* non-copper PHY? */
2658         if (adapter->hw.phy.media_type != e1000_media_type_copper)
2659                 return;
2660
2661         /* reset is blocked because of a SoL/IDER session */
2662         if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2663                 return;
2664
2665         /* manageability (AMT) is enabled */
2666         if (er32(MANC) & E1000_MANC_SMBUS_EN)
2667                 return;
2668
2669         /* power down the PHY */
2670         e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2671         mii_reg |= MII_CR_POWER_DOWN;
2672         e1e_wphy(hw, PHY_CONTROL, mii_reg);
2673         mdelay(1);
2674 }
2675
2676 /**
2677  * e1000e_reset - bring the hardware into a known good state
2678  *
2679  * This function boots the hardware and enables some settings that
2680  * require a configuration cycle of the hardware - those cannot be
2681  * set/changed during runtime. After reset the device needs to be
2682  * properly configured for Rx, Tx etc.
2683  */
2684 void e1000e_reset(struct e1000_adapter *adapter)
2685 {
2686         struct e1000_mac_info *mac = &adapter->hw.mac;
2687         struct e1000_fc_info *fc = &adapter->hw.fc;
2688         struct e1000_hw *hw = &adapter->hw;
2689         u32 tx_space, min_tx_space, min_rx_space;
2690         u32 pba = adapter->pba;
2691         u16 hwm;
2692
2693         /* reset Packet Buffer Allocation to default */
2694         ew32(PBA, pba);
2695
2696         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2697                 /*
2698                  * To maintain wire speed transmits, the Tx FIFO should be
2699                  * large enough to accommodate two full transmit packets,
2700                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2701                  * the Rx FIFO should be large enough to accommodate at least
2702                  * one full receive packet and is similarly rounded up and
2703                  * expressed in KB.
2704                  */
2705                 pba = er32(PBA);
2706                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2707                 tx_space = pba >> 16;
2708                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2709                 pba &= 0xffff;
2710                 /*
2711                  * the Tx fifo also stores 16 bytes of information about the tx
2712                  * but don't include ethernet FCS because hardware appends it
2713                  */
2714                 min_tx_space = (adapter->max_frame_size +
2715                                 sizeof(struct e1000_tx_desc) -
2716                                 ETH_FCS_LEN) * 2;
2717                 min_tx_space = ALIGN(min_tx_space, 1024);
2718                 min_tx_space >>= 10;
2719                 /* software strips receive CRC, so leave room for it */
2720                 min_rx_space = adapter->max_frame_size;
2721                 min_rx_space = ALIGN(min_rx_space, 1024);
2722                 min_rx_space >>= 10;
2723
2724                 /*
2725                  * If current Tx allocation is less than the min Tx FIFO size,
2726                  * and the min Tx FIFO size is less than the current Rx FIFO
2727                  * allocation, take space away from current Rx allocation
2728                  */
2729                 if ((tx_space < min_tx_space) &&
2730                     ((min_tx_space - tx_space) < pba)) {
2731                         pba -= min_tx_space - tx_space;
2732
2733                         /*
2734                          * if short on Rx space, Rx wins and must trump tx
2735                          * adjustment or use Early Receive if available
2736                          */
2737                         if ((pba < min_rx_space) &&
2738                             (!(adapter->flags & FLAG_HAS_ERT)))
2739                                 /* ERT enabled in e1000_configure_rx */
2740                                 pba = min_rx_space;
2741                 }
2742
2743                 ew32(PBA, pba);
2744         }
2745
2746
2747         /*
2748          * flow control settings
2749          *
2750          * The high water mark must be low enough to fit one full frame
2751          * (or the size used for early receive) above it in the Rx FIFO.
2752          * Set it to the lower of:
2753          * - 90% of the Rx FIFO size, and
2754          * - the full Rx FIFO size minus the early receive size (for parts
2755          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2756          * - the full Rx FIFO size minus one full frame
2757          */
2758         if (adapter->flags & FLAG_HAS_ERT)
2759                 hwm = min(((pba << 10) * 9 / 10),
2760                           ((pba << 10) - (E1000_ERT_2048 << 3)));
2761         else
2762                 hwm = min(((pba << 10) * 9 / 10),
2763                           ((pba << 10) - adapter->max_frame_size));
2764
2765         fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
2766         fc->low_water = fc->high_water - 8;
2767
2768         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2769                 fc->pause_time = 0xFFFF;
2770         else
2771                 fc->pause_time = E1000_FC_PAUSE_TIME;
2772         fc->send_xon = 1;
2773         fc->current_mode = fc->requested_mode;
2774
2775         /* Allow time for pending master requests to run */
2776         mac->ops.reset_hw(hw);
2777
2778         /*
2779          * For parts with AMT enabled, let the firmware know
2780          * that the network interface is in control
2781          */
2782         if (adapter->flags & FLAG_HAS_AMT)
2783                 e1000_get_hw_control(adapter);
2784
2785         ew32(WUC, 0);
2786
2787         if (mac->ops.init_hw(hw))
2788                 e_err("Hardware Error\n");
2789
2790         e1000_update_mng_vlan(adapter);
2791
2792         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2793         ew32(VET, ETH_P_8021Q);
2794
2795         e1000e_reset_adaptive(hw);
2796         e1000_get_phy_info(hw);
2797
2798         if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2799                 u16 phy_data = 0;
2800                 /*
2801                  * speed up time to link by disabling smart power down, ignore
2802                  * the return value of this function because there is nothing
2803                  * different we would do if it failed
2804                  */
2805                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2806                 phy_data &= ~IGP02E1000_PM_SPD;
2807                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2808         }
2809 }
2810
2811 int e1000e_up(struct e1000_adapter *adapter)
2812 {
2813         struct e1000_hw *hw = &adapter->hw;
2814
2815         /* hardware has been reset, we need to reload some things */
2816         e1000_configure(adapter);
2817
2818         clear_bit(__E1000_DOWN, &adapter->state);
2819
2820         napi_enable(&adapter->napi);
2821         if (adapter->msix_entries)
2822                 e1000_configure_msix(adapter);
2823         e1000_irq_enable(adapter);
2824
2825         /* fire a link change interrupt to start the watchdog */
2826         ew32(ICS, E1000_ICS_LSC);
2827         return 0;
2828 }
2829
2830 void e1000e_down(struct e1000_adapter *adapter)
2831 {
2832         struct net_device *netdev = adapter->netdev;
2833         struct e1000_hw *hw = &adapter->hw;
2834         u32 tctl, rctl;
2835
2836         /*
2837          * signal that we're down so the interrupt handler does not
2838          * reschedule our watchdog timer
2839          */
2840         set_bit(__E1000_DOWN, &adapter->state);
2841
2842         /* disable receives in the hardware */
2843         rctl = er32(RCTL);
2844         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2845         /* flush and sleep below */
2846
2847         netif_tx_stop_all_queues(netdev);
2848
2849         /* disable transmits in the hardware */
2850         tctl = er32(TCTL);
2851         tctl &= ~E1000_TCTL_EN;
2852         ew32(TCTL, tctl);
2853         /* flush both disables and wait for them to finish */
2854         e1e_flush();
2855         msleep(10);
2856
2857         napi_disable(&adapter->napi);
2858         e1000_irq_disable(adapter);
2859
2860         del_timer_sync(&adapter->watchdog_timer);
2861         del_timer_sync(&adapter->phy_info_timer);
2862
2863         netdev->tx_queue_len = adapter->tx_queue_len;
2864         netif_carrier_off(netdev);
2865         adapter->link_speed = 0;
2866         adapter->link_duplex = 0;
2867
2868         if (!pci_channel_offline(adapter->pdev))
2869                 e1000e_reset(adapter);
2870         e1000_clean_tx_ring(adapter);
2871         e1000_clean_rx_ring(adapter);
2872
2873         /*
2874          * TODO: for power management, we could drop the link and
2875          * pci_disable_device here.
2876          */
2877 }
2878
2879 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2880 {
2881         might_sleep();
2882         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2883                 msleep(1);
2884         e1000e_down(adapter);
2885         e1000e_up(adapter);
2886         clear_bit(__E1000_RESETTING, &adapter->state);
2887 }
2888
2889 /**
2890  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2891  * @adapter: board private structure to initialize
2892  *
2893  * e1000_sw_init initializes the Adapter private data structure.
2894  * Fields are initialized based on PCI device information and
2895  * OS network device settings (MTU size).
2896  **/
2897 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2898 {
2899         struct net_device *netdev = adapter->netdev;
2900
2901         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2902         adapter->rx_ps_bsize0 = 128;
2903         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2904         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2905
2906         e1000e_set_interrupt_capability(adapter);
2907
2908         if (e1000_alloc_queues(adapter))
2909                 return -ENOMEM;
2910
2911         /* Explicitly disable IRQ since the NIC can be in any state. */
2912         e1000_irq_disable(adapter);
2913
2914         set_bit(__E1000_DOWN, &adapter->state);
2915         return 0;
2916 }
2917
2918 /**
2919  * e1000_intr_msi_test - Interrupt Handler
2920  * @irq: interrupt number
2921  * @data: pointer to a network interface device structure
2922  **/
2923 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2924 {
2925         struct net_device *netdev = data;
2926         struct e1000_adapter *adapter = netdev_priv(netdev);
2927         struct e1000_hw *hw = &adapter->hw;
2928         u32 icr = er32(ICR);
2929
2930         e_dbg("%s: icr is %08X\n", netdev->name, icr);
2931         if (icr & E1000_ICR_RXSEQ) {
2932                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2933                 wmb();
2934         }
2935
2936         return IRQ_HANDLED;
2937 }
2938
2939 /**
2940  * e1000_test_msi_interrupt - Returns 0 for successful test
2941  * @adapter: board private struct
2942  *
2943  * code flow taken from tg3.c
2944  **/
2945 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2946 {
2947         struct net_device *netdev = adapter->netdev;
2948         struct e1000_hw *hw = &adapter->hw;
2949         int err;
2950
2951         /* poll_enable hasn't been called yet, so don't need disable */
2952         /* clear any pending events */
2953         er32(ICR);
2954
2955         /* free the real vector and request a test handler */
2956         e1000_free_irq(adapter);
2957         e1000e_reset_interrupt_capability(adapter);
2958
2959         /* Assume that the test fails, if it succeeds then the test
2960          * MSI irq handler will unset this flag */
2961         adapter->flags |= FLAG_MSI_TEST_FAILED;
2962
2963         err = pci_enable_msi(adapter->pdev);
2964         if (err)
2965                 goto msi_test_failed;
2966
2967         err = request_irq(adapter->pdev->irq, &e1000_intr_msi_test, 0,
2968                           netdev->name, netdev);
2969         if (err) {
2970                 pci_disable_msi(adapter->pdev);
2971                 goto msi_test_failed;
2972         }
2973
2974         wmb();
2975
2976         e1000_irq_enable(adapter);
2977
2978         /* fire an unusual interrupt on the test handler */
2979         ew32(ICS, E1000_ICS_RXSEQ);
2980         e1e_flush();
2981         msleep(50);
2982
2983         e1000_irq_disable(adapter);
2984
2985         rmb();
2986
2987         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
2988                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2989                 err = -EIO;
2990                 e_info("MSI interrupt test failed!\n");
2991         }
2992
2993         free_irq(adapter->pdev->irq, netdev);
2994         pci_disable_msi(adapter->pdev);
2995
2996         if (err == -EIO)
2997                 goto msi_test_failed;
2998
2999         /* okay so the test worked, restore settings */
3000         e_dbg("%s: MSI interrupt test succeeded!\n", netdev->name);
3001 msi_test_failed:
3002         e1000e_set_interrupt_capability(adapter);
3003         e1000_request_irq(adapter);
3004         return err;
3005 }
3006
3007 /**
3008  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3009  * @adapter: board private struct
3010  *
3011  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3012  **/
3013 static int e1000_test_msi(struct e1000_adapter *adapter)
3014 {
3015         int err;
3016         u16 pci_cmd;
3017
3018         if (!(adapter->flags & FLAG_MSI_ENABLED))
3019                 return 0;
3020
3021         /* disable SERR in case the MSI write causes a master abort */
3022         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3023         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3024                               pci_cmd & ~PCI_COMMAND_SERR);
3025
3026         err = e1000_test_msi_interrupt(adapter);
3027
3028         /* restore previous setting of command word */
3029         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3030
3031         /* success ! */
3032         if (!err)
3033                 return 0;
3034
3035         /* EIO means MSI test failed */
3036         if (err != -EIO)
3037                 return err;
3038
3039         /* back to INTx mode */
3040         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3041
3042         e1000_free_irq(adapter);
3043
3044         err = e1000_request_irq(adapter);
3045
3046         return err;
3047 }
3048
3049 /**
3050  * e1000_open - Called when a network interface is made active
3051  * @netdev: network interface device structure
3052  *
3053  * Returns 0 on success, negative value on failure
3054  *
3055  * The open entry point is called when a network interface is made
3056  * active by the system (IFF_UP).  At this point all resources needed
3057  * for transmit and receive operations are allocated, the interrupt
3058  * handler is registered with the OS, the watchdog timer is started,
3059  * and the stack is notified that the interface is ready.
3060  **/
3061 static int e1000_open(struct net_device *netdev)
3062 {
3063         struct e1000_adapter *adapter = netdev_priv(netdev);
3064         struct e1000_hw *hw = &adapter->hw;
3065         int err;
3066
3067         /* disallow open during test */
3068         if (test_bit(__E1000_TESTING, &adapter->state))
3069                 return -EBUSY;
3070
3071         /* allocate transmit descriptors */
3072         err = e1000e_setup_tx_resources(adapter);
3073         if (err)
3074                 goto err_setup_tx;
3075
3076         /* allocate receive descriptors */
3077         err = e1000e_setup_rx_resources(adapter);
3078         if (err)
3079                 goto err_setup_rx;
3080
3081         e1000e_power_up_phy(adapter);
3082
3083         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3084         if ((adapter->hw.mng_cookie.status &
3085              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3086                 e1000_update_mng_vlan(adapter);
3087
3088         /*
3089          * If AMT is enabled, let the firmware know that the network
3090          * interface is now open
3091          */
3092         if (adapter->flags & FLAG_HAS_AMT)
3093                 e1000_get_hw_control(adapter);
3094
3095         /*
3096          * before we allocate an interrupt, we must be ready to handle it.
3097          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3098          * as soon as we call pci_request_irq, so we have to setup our
3099          * clean_rx handler before we do so.
3100          */
3101         e1000_configure(adapter);
3102
3103         err = e1000_request_irq(adapter);
3104         if (err)
3105                 goto err_req_irq;
3106
3107         /*
3108          * Work around PCIe errata with MSI interrupts causing some chipsets to
3109          * ignore e1000e MSI messages, which means we need to test our MSI
3110          * interrupt now
3111          */
3112         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3113                 err = e1000_test_msi(adapter);
3114                 if (err) {
3115                         e_err("Interrupt allocation failed\n");
3116                         goto err_req_irq;
3117                 }
3118         }
3119
3120         /* From here on the code is the same as e1000e_up() */
3121         clear_bit(__E1000_DOWN, &adapter->state);
3122
3123         napi_enable(&adapter->napi);
3124
3125         e1000_irq_enable(adapter);
3126
3127         netif_tx_start_all_queues(netdev);
3128
3129         /* fire a link status change interrupt to start the watchdog */
3130         ew32(ICS, E1000_ICS_LSC);
3131
3132         return 0;
3133
3134 err_req_irq:
3135         e1000_release_hw_control(adapter);
3136         e1000_power_down_phy(adapter);
3137         e1000e_free_rx_resources(adapter);
3138 err_setup_rx:
3139         e1000e_free_tx_resources(adapter);
3140 err_setup_tx:
3141         e1000e_reset(adapter);
3142
3143         return err;
3144 }
3145
3146 /**
3147  * e1000_close - Disables a network interface
3148  * @netdev: network interface device structure
3149  *
3150  * Returns 0, this is not allowed to fail
3151  *
3152  * The close entry point is called when an interface is de-activated
3153  * by the OS.  The hardware is still under the drivers control, but
3154  * needs to be disabled.  A global MAC reset is issued to stop the
3155  * hardware, and all transmit and receive resources are freed.
3156  **/
3157 static int e1000_close(struct net_device *netdev)
3158 {
3159         struct e1000_adapter *adapter = netdev_priv(netdev);
3160
3161         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3162         e1000e_down(adapter);
3163         e1000_power_down_phy(adapter);
3164         e1000_free_irq(adapter);
3165
3166         e1000e_free_tx_resources(adapter);
3167         e1000e_free_rx_resources(adapter);
3168
3169         /*
3170          * kill manageability vlan ID if supported, but not if a vlan with
3171          * the same ID is registered on the host OS (let 8021q kill it)
3172          */
3173         if ((adapter->hw.mng_cookie.status &
3174                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3175              !(adapter->vlgrp &&
3176                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3177                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3178
3179         /*
3180          * If AMT is enabled, let the firmware know that the network
3181          * interface is now closed
3182          */
3183         if (adapter->flags & FLAG_HAS_AMT)
3184                 e1000_release_hw_control(adapter);
3185
3186         return 0;
3187 }
3188 /**
3189  * e1000_set_mac - Change the Ethernet Address of the NIC
3190  * @netdev: network interface device structure
3191  * @p: pointer to an address structure
3192  *
3193  * Returns 0 on success, negative on failure
3194  **/
3195 static int e1000_set_mac(struct net_device *netdev, void *p)
3196 {
3197         struct e1000_adapter *adapter = netdev_priv(netdev);
3198         struct sockaddr *addr = p;
3199
3200         if (!is_valid_ether_addr(addr->sa_data))
3201                 return -EADDRNOTAVAIL;
3202
3203         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3204         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3205
3206         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3207
3208         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3209                 /* activate the work around */
3210                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3211
3212                 /*
3213                  * Hold a copy of the LAA in RAR[14] This is done so that
3214                  * between the time RAR[0] gets clobbered  and the time it
3215                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3216                  * of the RARs and no incoming packets directed to this port
3217                  * are dropped. Eventually the LAA will be in RAR[0] and
3218                  * RAR[14]
3219                  */
3220                 e1000e_rar_set(&adapter->hw,
3221                               adapter->hw.mac.addr,
3222                               adapter->hw.mac.rar_entry_count - 1);
3223         }
3224
3225         return 0;
3226 }
3227
3228 /**
3229  * e1000e_update_phy_task - work thread to update phy
3230  * @work: pointer to our work struct
3231  *
3232  * this worker thread exists because we must acquire a
3233  * semaphore to read the phy, which we could msleep while
3234  * waiting for it, and we can't msleep in a timer.
3235  **/
3236 static void e1000e_update_phy_task(struct work_struct *work)
3237 {
3238         struct e1000_adapter *adapter = container_of(work,
3239                                         struct e1000_adapter, update_phy_task);
3240         e1000_get_phy_info(&adapter->hw);
3241 }
3242
3243 /*
3244  * Need to wait a few seconds after link up to get diagnostic information from
3245  * the phy
3246  */
3247 static void e1000_update_phy_info(unsigned long data)
3248 {
3249         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3250         schedule_work(&adapter->update_phy_task);
3251 }
3252
3253 /**
3254  * e1000e_update_stats - Update the board statistics counters
3255  * @adapter: board private structure
3256  **/
3257 void e1000e_update_stats(struct e1000_adapter *adapter)
3258 {
3259         struct e1000_hw *hw = &adapter->hw;
3260         struct pci_dev *pdev = adapter->pdev;
3261
3262         /*
3263          * Prevent stats update while adapter is being reset, or if the pci
3264          * connection is down.
3265          */
3266         if (adapter->link_speed == 0)
3267                 return;
3268         if (pci_channel_offline(pdev))
3269                 return;
3270
3271         adapter->stats.crcerrs += er32(CRCERRS);
3272         adapter->stats.gprc += er32(GPRC);
3273         adapter->stats.gorc += er32(GORCL);
3274         er32(GORCH); /* Clear gorc */
3275         adapter->stats.bprc += er32(BPRC);
3276         adapter->stats.mprc += er32(MPRC);
3277         adapter->stats.roc += er32(ROC);
3278
3279         adapter->stats.mpc += er32(MPC);
3280         adapter->stats.scc += er32(SCC);
3281         adapter->stats.ecol += er32(ECOL);
3282         adapter->stats.mcc += er32(MCC);
3283         adapter->stats.latecol += er32(LATECOL);
3284         adapter->stats.dc += er32(DC);
3285         adapter->stats.xonrxc += er32(XONRXC);
3286         adapter->stats.xontxc += er32(XONTXC);
3287         adapter->stats.xoffrxc += er32(XOFFRXC);
3288         adapter->stats.xofftxc += er32(XOFFTXC);
3289         adapter->stats.gptc += er32(GPTC);
3290         adapter->stats.gotc += er32(GOTCL);
3291         er32(GOTCH); /* Clear gotc */
3292         adapter->stats.rnbc += er32(RNBC);
3293         adapter->stats.ruc += er32(RUC);
3294
3295         adapter->stats.mptc += er32(MPTC);
3296         adapter->stats.bptc += er32(BPTC);
3297
3298         /* used for adaptive IFS */
3299
3300         hw->mac.tx_packet_delta = er32(TPT);
3301         adapter->stats.tpt += hw->mac.tx_packet_delta;
3302         hw->mac.collision_delta = er32(COLC);
3303         adapter->stats.colc += hw->mac.collision_delta;
3304
3305         adapter->stats.algnerrc += er32(ALGNERRC);
3306         adapter->stats.rxerrc += er32(RXERRC);
3307         if ((hw->mac.type != e1000_82574) && (hw->mac.type != e1000_82583))
3308                 adapter->stats.tncrs += er32(TNCRS);
3309         adapter->stats.cexterr += er32(CEXTERR);
3310         adapter->stats.tsctc += er32(TSCTC);
3311         adapter->stats.tsctfc += er32(TSCTFC);
3312
3313         /* Fill out the OS statistics structure */
3314         adapter->net_stats.multicast = adapter->stats.mprc;
3315         adapter->net_stats.collisions = adapter->stats.colc;
3316
3317         /* Rx Errors */
3318
3319         /*
3320          * RLEC on some newer hardware can be incorrect so build
3321          * our own version based on RUC and ROC
3322          */
3323         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3324                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3325                 adapter->stats.ruc + adapter->stats.roc +
3326                 adapter->stats.cexterr;
3327         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3328                                               adapter->stats.roc;
3329         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3330         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3331         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3332
3333         /* Tx Errors */
3334         adapter->net_stats.tx_errors = adapter->stats.ecol +
3335                                        adapter->stats.latecol;
3336         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3337         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3338         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3339
3340         /* Tx Dropped needs to be maintained elsewhere */
3341
3342         /* Management Stats */
3343         adapter->stats.mgptc += er32(MGTPTC);
3344         adapter->stats.mgprc += er32(MGTPRC);
3345         adapter->stats.mgpdc += er32(MGTPDC);
3346 }
3347
3348 /**
3349  * e1000_phy_read_status - Update the PHY register status snapshot
3350  * @adapter: board private structure
3351  **/
3352 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3353 {
3354         struct e1000_hw *hw = &adapter->hw;
3355         struct e1000_phy_regs *phy = &adapter->phy_regs;
3356         int ret_val;
3357
3358         if ((er32(STATUS) & E1000_STATUS_LU) &&
3359             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3360                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3361                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3362                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3363                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3364                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3365                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3366                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3367                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3368                 if (ret_val)
3369                         e_warn("Error reading PHY register\n");
3370         } else {
3371                 /*
3372                  * Do not read PHY registers if link is not up
3373                  * Set values to typical power-on defaults
3374                  */
3375                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3376                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3377                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3378                              BMSR_ERCAP);
3379                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3380                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3381                 phy->lpa = 0;
3382                 phy->expansion = EXPANSION_ENABLENPAGE;
3383                 phy->ctrl1000 = ADVERTISE_1000FULL;
3384                 phy->stat1000 = 0;
3385                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3386         }
3387 }
3388
3389 static void e1000_print_link_info(struct e1000_adapter *adapter)
3390 {
3391         struct e1000_hw *hw = &adapter->hw;
3392         u32 ctrl = er32(CTRL);
3393
3394         /* Link status message must follow this format for user tools */
3395         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3396                "Flow Control: %s\n",
3397                adapter->netdev->name,
3398                adapter->link_speed,
3399                (adapter->link_duplex == FULL_DUPLEX) ?
3400                                 "Full Duplex" : "Half Duplex",
3401                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3402                                 "RX/TX" :
3403                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3404                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3405 }
3406
3407 bool e1000_has_link(struct e1000_adapter *adapter)
3408 {
3409         struct e1000_hw *hw = &adapter->hw;
3410         bool link_active = 0;
3411         s32 ret_val = 0;
3412
3413         /*
3414          * get_link_status is set on LSC (link status) interrupt or
3415          * Rx sequence error interrupt.  get_link_status will stay
3416          * false until the check_for_link establishes link
3417          * for copper adapters ONLY
3418          */
3419         switch (hw->phy.media_type) {
3420         case e1000_media_type_copper:
3421                 if (hw->mac.get_link_status) {
3422                         ret_val = hw->mac.ops.check_for_link(hw);
3423                         link_active = !hw->mac.get_link_status;
3424                 } else {
3425                         link_active = 1;
3426                 }
3427                 break;
3428         case e1000_media_type_fiber:
3429                 ret_val = hw->mac.ops.check_for_link(hw);
3430                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3431                 break;
3432         case e1000_media_type_internal_serdes:
3433                 ret_val = hw->mac.ops.check_for_link(hw);
3434                 link_active = adapter->hw.mac.serdes_has_link;
3435                 break;
3436         default:
3437         case e1000_media_type_unknown:
3438                 break;
3439         }
3440
3441         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3442             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3443                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3444                 e_info("Gigabit has been disabled, downgrading speed\n");
3445         }
3446
3447         return link_active;
3448 }
3449
3450 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3451 {
3452         /* make sure the receive unit is started */
3453         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3454             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3455                 struct e1000_hw *hw = &adapter->hw;
3456                 u32 rctl = er32(RCTL);
3457                 ew32(RCTL, rctl | E1000_RCTL_EN);
3458                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3459         }
3460 }
3461
3462 /**
3463  * e1000_watchdog - Timer Call-back
3464  * @data: pointer to adapter cast into an unsigned long
3465  **/
3466 static void e1000_watchdog(unsigned long data)
3467 {
3468         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3469
3470         /* Do the rest outside of interrupt context */
3471         schedule_work(&adapter->watchdog_task);
3472
3473         /* TODO: make this use queue_delayed_work() */
3474 }
3475
3476 static void e1000_watchdog_task(struct work_struct *work)
3477 {
3478         struct e1000_adapter *adapter = container_of(work,
3479                                         struct e1000_adapter, watchdog_task);
3480         struct net_device *netdev = adapter->netdev;
3481         struct e1000_mac_info *mac = &adapter->hw.mac;
3482         struct e1000_phy_info *phy = &adapter->hw.phy;
3483         struct e1000_ring *tx_ring = adapter->tx_ring;
3484         struct e1000_hw *hw = &adapter->hw;
3485         u32 link, tctl;
3486         int tx_pending = 0;
3487
3488         link = e1000_has_link(adapter);
3489         if ((netif_carrier_ok(netdev)) && link) {
3490                 e1000e_enable_receives(adapter);
3491                 goto link_up;
3492         }
3493
3494         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3495             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3496                 e1000_update_mng_vlan(adapter);
3497
3498         if (link) {
3499                 if (!netif_carrier_ok(netdev)) {
3500                         bool txb2b = 1;
3501                         /* update snapshot of PHY registers on LSC */
3502                         e1000_phy_read_status(adapter);
3503                         mac->ops.get_link_up_info(&adapter->hw,
3504                                                    &adapter->link_speed,
3505                                                    &adapter->link_duplex);
3506                         e1000_print_link_info(adapter);
3507                         /*
3508                          * On supported PHYs, check for duplex mismatch only
3509                          * if link has autonegotiated at 10/100 half
3510                          */
3511                         if ((hw->phy.type == e1000_phy_igp_3 ||
3512                              hw->phy.type == e1000_phy_bm) &&
3513                             (hw->mac.autoneg == true) &&
3514                             (adapter->link_speed == SPEED_10 ||
3515                              adapter->link_speed == SPEED_100) &&
3516                             (adapter->link_duplex == HALF_DUPLEX)) {
3517                                 u16 autoneg_exp;
3518
3519                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3520
3521                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3522                                         e_info("Autonegotiated half duplex but"
3523                                                " link partner cannot autoneg. "
3524                                                " Try forcing full duplex if "
3525                                                "link gets many collisions.\n");
3526                         }
3527
3528                         /*
3529                          * tweak tx_queue_len according to speed/duplex
3530                          * and adjust the timeout factor
3531                          */
3532                         netdev->tx_queue_len = adapter->tx_queue_len;
3533                         adapter->tx_timeout_factor = 1;
3534                         switch (adapter->link_speed) {
3535                         case SPEED_10:
3536                                 txb2b = 0;
3537                                 netdev->tx_queue_len = 10;
3538                                 adapter->tx_timeout_factor = 16;
3539                                 break;
3540                         case SPEED_100:
3541                                 txb2b = 0;
3542                                 netdev->tx_queue_len = 100;
3543                                 /* maybe add some timeout factor ? */
3544                                 break;
3545                         }
3546
3547                         /*
3548                          * workaround: re-program speed mode bit after
3549                          * link-up event
3550                          */
3551                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3552                             !txb2b) {
3553                                 u32 tarc0;
3554                                 tarc0 = er32(TARC(0));
3555                                 tarc0 &= ~SPEED_MODE_BIT;
3556                                 ew32(TARC(0), tarc0);
3557                         }
3558
3559                         /*
3560                          * disable TSO for pcie and 10/100 speeds, to avoid
3561                          * some hardware issues
3562                          */
3563                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3564                                 switch (adapter->link_speed) {
3565                                 case SPEED_10:
3566                                 case SPEED_100:
3567                                         e_info("10/100 speed: disabling TSO\n");
3568                                         netdev->features &= ~NETIF_F_TSO;
3569                                         netdev->features &= ~NETIF_F_TSO6;
3570                                         break;
3571                                 case SPEED_1000:
3572                                         netdev->features |= NETIF_F_TSO;
3573                                         netdev->features |= NETIF_F_TSO6;
3574                                         break;
3575                                 default:
3576                                         /* oops */
3577                                         break;
3578                                 }
3579                         }
3580
3581                         /*
3582                          * enable transmits in the hardware, need to do this
3583                          * after setting TARC(0)
3584                          */
3585                         tctl = er32(TCTL);
3586                         tctl |= E1000_TCTL_EN;
3587                         ew32(TCTL, tctl);
3588
3589                         /*
3590                          * Perform any post-link-up configuration before
3591                          * reporting link up.
3592                          */
3593                         if (phy->ops.cfg_on_link_up)
3594                                 phy->ops.cfg_on_link_up(hw);
3595
3596                         netif_carrier_on(netdev);
3597                         netif_tx_wake_all_queues(netdev);
3598
3599                         if (!test_bit(__E1000_DOWN, &adapter->state))
3600                                 mod_timer(&adapter->phy_info_timer,
3601                                           round_jiffies(jiffies + 2 * HZ));
3602                 }
3603         } else {
3604                 if (netif_carrier_ok(netdev)) {
3605                         adapter->link_speed = 0;
3606                         adapter->link_duplex = 0;
3607                         /* Link status message must follow this format */
3608                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
3609                                adapter->netdev->name);
3610                         netif_carrier_off(netdev);
3611                         netif_tx_stop_all_queues(netdev);
3612                         if (!test_bit(__E1000_DOWN, &adapter->state))
3613                                 mod_timer(&adapter->phy_info_timer,
3614                                           round_jiffies(jiffies + 2 * HZ));
3615
3616                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3617                                 schedule_work(&adapter->reset_task);
3618                 }
3619         }
3620
3621 link_up:
3622         e1000e_update_stats(adapter);
3623
3624         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3625         adapter->tpt_old = adapter->stats.tpt;
3626         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3627         adapter->colc_old = adapter->stats.colc;
3628
3629         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3630         adapter->gorc_old = adapter->stats.gorc;
3631         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3632         adapter->gotc_old = adapter->stats.gotc;
3633
3634         e1000e_update_adaptive(&adapter->hw);
3635
3636         if (!netif_carrier_ok(netdev)) {
3637                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3638                                tx_ring->count);
3639                 if (tx_pending) {
3640                         /*
3641                          * We've lost link, so the controller stops DMA,
3642                          * but we've got queued Tx work that's never going
3643                          * to get done, so reset controller to flush Tx.
3644                          * (Do the reset outside of interrupt context).
3645                          */
3646                         adapter->tx_timeout_count++;
3647                         schedule_work(&adapter->reset_task);
3648                 }
3649         }
3650
3651         /* Cause software interrupt to ensure Rx ring is cleaned */
3652         if (adapter->msix_entries)
3653                 ew32(ICS, adapter->rx_ring->ims_val);
3654         else
3655                 ew32(ICS, E1000_ICS_RXDMT0);
3656
3657         /* Force detection of hung controller every watchdog period */
3658         adapter->detect_tx_hung = 1;
3659
3660         /*
3661          * With 82571 controllers, LAA may be overwritten due to controller
3662          * reset from the other port. Set the appropriate LAA in RAR[0]
3663          */
3664         if (e1000e_get_laa_state_82571(hw))
3665                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3666
3667         /* Reset the timer */
3668         if (!test_bit(__E1000_DOWN, &adapter->state))
3669                 mod_timer(&adapter->watchdog_timer,
3670                           round_jiffies(jiffies + 2 * HZ));
3671 }
3672
3673 #define E1000_TX_FLAGS_CSUM             0x00000001
3674 #define E1000_TX_FLAGS_VLAN             0x00000002
3675 #define E1000_TX_FLAGS_TSO              0x00000004
3676 #define E1000_TX_FLAGS_IPV4             0x00000008
3677 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3678 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3679
3680 static int e1000_tso(struct e1000_adapter *adapter,
3681                      struct sk_buff *skb)
3682 {
3683         struct e1000_ring *tx_ring = adapter->tx_ring;
3684         struct e1000_context_desc *context_desc;
3685         struct e1000_buffer *buffer_info;
3686         unsigned int i;
3687         u32 cmd_length = 0;
3688         u16 ipcse = 0, tucse, mss;
3689         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3690         int err;
3691
3692         if (skb_is_gso(skb)) {
3693                 if (skb_header_cloned(skb)) {
3694                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3695                         if (err)
3696                                 return err;
3697                 }
3698
3699                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3700                 mss = skb_shinfo(skb)->gso_size;
3701                 if (skb->protocol == htons(ETH_P_IP)) {
3702                         struct iphdr *iph = ip_hdr(skb);
3703                         iph->tot_len = 0;
3704                         iph->check = 0;
3705                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3706                                                                  iph->daddr, 0,
3707                                                                  IPPROTO_TCP,
3708                                                                  0);
3709                         cmd_length = E1000_TXD_CMD_IP;
3710                         ipcse = skb_transport_offset(skb) - 1;
3711                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3712                         ipv6_hdr(skb)->payload_len = 0;
3713                         tcp_hdr(skb)->check =
3714                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3715                                                  &ipv6_hdr(skb)->daddr,
3716                                                  0, IPPROTO_TCP, 0);
3717                         ipcse = 0;
3718                 }
3719                 ipcss = skb_network_offset(skb);
3720                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3721                 tucss = skb_transport_offset(skb);
3722                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3723                 tucse = 0;
3724
3725                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3726                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3727
3728                 i = tx_ring->next_to_use;
3729                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3730                 buffer_info = &tx_ring->buffer_info[i];
3731
3732                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3733                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3734                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3735                 context_desc->upper_setup.tcp_fields.tucss = tucss;
3736                 context_desc->upper_setup.tcp_fields.tucso = tucso;
3737                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3738                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3739                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3740                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3741
3742                 buffer_info->time_stamp = jiffies;
3743                 buffer_info->next_to_watch = i;
3744
3745                 i++;
3746                 if (i == tx_ring->count)
3747                         i = 0;
3748                 tx_ring->next_to_use = i;
3749
3750                 return 1;
3751         }
3752
3753         return 0;
3754 }
3755
3756 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3757 {
3758         struct e1000_ring *tx_ring = adapter->tx_ring;
3759         struct e1000_context_desc *context_desc;
3760         struct e1000_buffer *buffer_info;
3761         unsigned int i;
3762         u8 css;
3763         u32 cmd_len = E1000_TXD_CMD_DEXT;
3764         __be16 protocol;
3765
3766         if (skb->ip_summed != CHECKSUM_PARTIAL)
3767                 return 0;
3768
3769         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
3770                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
3771         else
3772                 protocol = skb->protocol;
3773
3774         switch (protocol) {
3775         case cpu_to_be16(ETH_P_IP):
3776                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3777                         cmd_len |= E1000_TXD_CMD_TCP;
3778                 break;
3779         case cpu_to_be16(ETH_P_IPV6):
3780                 /* XXX not handling all IPV6 headers */
3781                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3782                         cmd_len |= E1000_TXD_CMD_TCP;
3783                 break;
3784         default:
3785                 if (unlikely(net_ratelimit()))
3786                         e_warn("checksum_partial proto=%x!\n",
3787                                be16_to_cpu(protocol));
3788                 break;
3789         }
3790
3791         css = skb_transport_offset(skb);
3792
3793         i = tx_ring->next_to_use;
3794         buffer_info = &tx_ring->buffer_info[i];
3795         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3796
3797         context_desc->lower_setup.ip_config = 0;
3798         context_desc->upper_setup.tcp_fields.tucss = css;
3799         context_desc->upper_setup.tcp_fields.tucso =
3800                                 css + skb->csum_offset;
3801         context_desc->upper_setup.tcp_fields.tucse = 0;
3802         context_desc->tcp_seg_setup.data = 0;
3803         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
3804
3805         buffer_info->time_stamp = jiffies;
3806         buffer_info->next_to_watch = i;
3807
3808         i++;
3809         if (i == tx_ring->count)
3810                 i = 0;
3811         tx_ring->next_to_use = i;
3812
3813         return 1;
3814 }
3815
3816 #define E1000_MAX_PER_TXD       8192
3817 #define E1000_MAX_TXD_PWR       12
3818
3819 static int e1000_tx_map(struct e1000_adapter *adapter,
3820                         struct sk_buff *skb, unsigned int first,
3821                         unsigned int max_per_txd, unsigned int nr_frags,
3822                         unsigned int mss)
3823 {
3824         struct e1000_ring *tx_ring = adapter->tx_ring;
3825         struct e1000_buffer *buffer_info;
3826         unsigned int len = skb_headlen(skb);
3827         unsigned int offset, size, count = 0, i;
3828         unsigned int f;
3829         dma_addr_t *map;
3830
3831         i = tx_ring->next_to_use;
3832
3833         if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
3834                 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3835                 adapter->tx_dma_failed++;
3836                 return 0;
3837         }
3838
3839         map = skb_shinfo(skb)->dma_maps;
3840         offset = 0;
3841
3842         while (len) {
3843                 buffer_info = &tx_ring->buffer_info[i];
3844                 size = min(len, max_per_txd);
3845
3846                 buffer_info->length = size;
3847                 buffer_info->time_stamp = jiffies;
3848                 buffer_info->next_to_watch = i;
3849                 buffer_info->dma = map[0] + offset;
3850                 count++;
3851
3852                 len -= size;
3853                 offset += size;
3854
3855                 if (len) {
3856                         i++;
3857                         if (i == tx_ring->count)
3858                                 i = 0;
3859                 }
3860         }
3861
3862         for (f = 0; f < nr_frags; f++) {
3863                 struct skb_frag_struct *frag;
3864
3865                 frag = &skb_shinfo(skb)->frags[f];
3866                 len = frag->size;
3867                 offset = 0;
3868
3869                 while (len) {
3870                         i++;
3871                         if (i == tx_ring->count)
3872                                 i = 0;
3873
3874                         buffer_info = &tx_ring->buffer_info[i];
3875                         size = min(len, max_per_txd);
3876
3877                         buffer_info->length = size;
3878                         buffer_info->time_stamp = jiffies;
3879                         buffer_info->next_to_watch = i;
3880                         buffer_info->dma = map[f + 1] + offset;
3881
3882                         len -= size;
3883                         offset += size;
3884                         count++;
3885                 }
3886         }
3887
3888         tx_ring->buffer_info[i].skb = skb;
3889         tx_ring->buffer_info[first].next_to_watch = i;
3890
3891         return count;
3892 }
3893
3894 static void e1000_tx_queue(struct e1000_adapter *adapter,
3895                            int tx_flags, int count)
3896 {
3897         struct e1000_ring *tx_ring = adapter->tx_ring;
3898         struct e1000_tx_desc *tx_desc = NULL;
3899         struct e1000_buffer *buffer_info;
3900         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3901         unsigned int i;
3902
3903         if (tx_flags & E1000_TX_FLAGS_TSO) {
3904                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3905                              E1000_TXD_CMD_TSE;
3906                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3907
3908                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3909                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3910         }
3911
3912         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3913                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3914                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3915         }
3916
3917         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3918                 txd_lower |= E1000_TXD_CMD_VLE;
3919                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3920         }
3921
3922         i = tx_ring->next_to_use;
3923
3924         while (count--) {
3925                 buffer_info = &tx_ring->buffer_info[i];
3926                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3927                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3928                 tx_desc->lower.data =
3929                         cpu_to_le32(txd_lower | buffer_info->length);
3930                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3931
3932                 i++;
3933                 if (i == tx_ring->count)
3934                         i = 0;
3935         }
3936
3937         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3938
3939         /*
3940          * Force memory writes to complete before letting h/w
3941          * know there are new descriptors to fetch.  (Only
3942          * applicable for weak-ordered memory model archs,
3943          * such as IA-64).
3944          */
3945         wmb();
3946
3947         tx_ring->next_to_use = i;
3948         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3949         /*
3950          * we need this if more than one processor can write to our tail
3951          * at a time, it synchronizes IO on IA64/Altix systems
3952          */
3953         mmiowb();
3954 }
3955
3956 #define MINIMUM_DHCP_PACKET_SIZE 282
3957 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3958                                     struct sk_buff *skb)
3959 {
3960         struct e1000_hw *hw =  &adapter->hw;
3961         u16 length, offset;
3962
3963         if (vlan_tx_tag_present(skb)) {
3964                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3965                     && (adapter->hw.mng_cookie.status &
3966                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3967                         return 0;
3968         }
3969
3970         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3971                 return 0;
3972
3973         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3974                 return 0;
3975
3976         {
3977                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3978                 struct udphdr *udp;
3979
3980                 if (ip->protocol != IPPROTO_UDP)
3981                         return 0;
3982
3983                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3984                 if (ntohs(udp->dest) != 67)
3985                         return 0;
3986
3987                 offset = (u8 *)udp + 8 - skb->data;
3988                 length = skb->len - offset;
3989                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
3990         }
3991
3992         return 0;
3993 }
3994
3995 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3996 {
3997         struct e1000_adapter *adapter = netdev_priv(netdev);
3998
3999         netif_stop_queue(netdev);
4000         /*
4001          * Herbert's original patch had:
4002          *  smp_mb__after_netif_stop_queue();
4003          * but since that doesn't exist yet, just open code it.
4004          */
4005         smp_mb();
4006
4007         /*
4008          * We need to check again in a case another CPU has just
4009          * made room available.
4010          */
4011         if (e1000_desc_unused(adapter->tx_ring) < size)
4012                 return -EBUSY;
4013
4014         /* A reprieve! */
4015         netif_start_queue(netdev);
4016         ++adapter->restart_queue;
4017         return 0;
4018 }
4019
4020 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4021 {
4022         struct e1000_adapter *adapter = netdev_priv(netdev);
4023
4024         if (e1000_desc_unused(adapter->tx_ring) >= size)
4025                 return 0;
4026         return __e1000_maybe_stop_tx(netdev, size);
4027 }
4028
4029 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4030 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4031 {
4032         struct e1000_adapter *adapter = netdev_priv(netdev);
4033         struct e1000_ring *tx_ring = adapter->tx_ring;
4034         unsigned int first;
4035         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4036         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4037         unsigned int tx_flags = 0;
4038         unsigned int len = skb->len - skb->data_len;
4039         unsigned int nr_frags;
4040         unsigned int mss;
4041         int count = 0;
4042         int tso;
4043         unsigned int f;
4044
4045         if (test_bit(__E1000_DOWN, &adapter->state)) {
4046                 dev_kfree_skb_any(skb);
4047                 return NETDEV_TX_OK;
4048         }
4049
4050         if (skb->len <= 0) {
4051                 dev_kfree_skb_any(skb);
4052                 return NETDEV_TX_OK;
4053         }
4054
4055         mss = skb_shinfo(skb)->gso_size;
4056         /*
4057          * The controller does a simple calculation to
4058          * make sure there is enough room in the FIFO before
4059          * initiating the DMA for each buffer.  The calc is:
4060          * 4 = ceil(buffer len/mss).  To make sure we don't
4061          * overrun the FIFO, adjust the max buffer len if mss
4062          * drops.
4063          */
4064         if (mss) {
4065                 u8 hdr_len;
4066                 max_per_txd = min(mss << 2, max_per_txd);
4067                 max_txd_pwr = fls(max_per_txd) - 1;
4068
4069                 /*
4070                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4071                  * points to just header, pull a few bytes of payload from
4072                  * frags into skb->data
4073                  */
4074                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4075                 /*
4076                  * we do this workaround for ES2LAN, but it is un-necessary,
4077                  * avoiding it could save a lot of cycles
4078                  */
4079                 if (skb->data_len && (hdr_len == len)) {
4080                         unsigned int pull_size;
4081
4082                         pull_size = min((unsigned int)4, skb->data_len);
4083                         if (!__pskb_pull_tail(skb, pull_size)) {
4084                                 e_err("__pskb_pull_tail failed.\n");
4085                                 dev_kfree_skb_any(skb);
4086                                 return NETDEV_TX_OK;
4087                         }
4088                         len = skb->len - skb->data_len;
4089                 }
4090         }
4091
4092         /* reserve a descriptor for the offload context */
4093         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4094                 count++;
4095         count++;
4096
4097         count += TXD_USE_COUNT(len, max_txd_pwr);
4098
4099         nr_frags = skb_shinfo(skb)->nr_frags;
4100         for (f = 0; f < nr_frags; f++)
4101                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4102                                        max_txd_pwr);
4103
4104         if (adapter->hw.mac.tx_pkt_filtering)
4105                 e1000_transfer_dhcp_info(adapter, skb);
4106
4107         /*
4108          * need: count + 2 desc gap to keep tail from touching
4109          * head, otherwise try next time
4110          */
4111         if (e1000_maybe_stop_tx(netdev, count + 2))
4112                 return NETDEV_TX_BUSY;
4113
4114         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4115                 tx_flags |= E1000_TX_FLAGS_VLAN;
4116                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4117         }
4118
4119         first = tx_ring->next_to_use;
4120
4121         tso = e1000_tso(adapter, skb);
4122         if (tso < 0) {
4123                 dev_kfree_skb_any(skb);
4124                 return NETDEV_TX_OK;
4125         }
4126
4127         if (tso)
4128                 tx_flags |= E1000_TX_FLAGS_TSO;
4129         else if (e1000_tx_csum(adapter, skb))
4130                 tx_flags |= E1000_TX_FLAGS_CSUM;
4131
4132         /*
4133          * Old method was to assume IPv4 packet by default if TSO was enabled.
4134          * 82571 hardware supports TSO capabilities for IPv6 as well...
4135          * no longer assume, we must.
4136          */
4137         if (skb->protocol == htons(ETH_P_IP))
4138                 tx_flags |= E1000_TX_FLAGS_IPV4;
4139
4140         /* if count is 0 then mapping error has occured */
4141         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4142         if (count) {
4143                 e1000_tx_queue(adapter, tx_flags, count);
4144                 netdev->trans_start = jiffies;
4145                 /* Make sure there is space in the ring for the next send. */
4146                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4147
4148         } else {
4149                 dev_kfree_skb_any(skb);
4150                 tx_ring->buffer_info[first].time_stamp = 0;
4151                 tx_ring->next_to_use = first;
4152         }
4153
4154         return NETDEV_TX_OK;
4155 }
4156
4157 /**
4158  * e1000_tx_timeout - Respond to a Tx Hang
4159  * @netdev: network interface device structure
4160  **/
4161 static void e1000_tx_timeout(struct net_device *netdev)
4162 {
4163         struct e1000_adapter *adapter = netdev_priv(netdev);
4164
4165         /* Do the reset outside of interrupt context */
4166         adapter->tx_timeout_count++;
4167         schedule_work(&adapter->reset_task);
4168 }
4169
4170 static void e1000_reset_task(struct work_struct *work)
4171 {
4172         struct e1000_adapter *adapter;
4173         adapter = container_of(work, struct e1000_adapter, reset_task);
4174
4175         e1000e_reinit_locked(adapter);
4176 }
4177
4178 /**
4179  * e1000_get_stats - Get System Network Statistics
4180  * @netdev: network interface device structure
4181  *
4182  * Returns the address of the device statistics structure.
4183  * The statistics are actually updated from the timer callback.
4184  **/
4185 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4186 {
4187         struct e1000_adapter *adapter = netdev_priv(netdev);
4188
4189         /* only return the current stats */
4190         return &adapter->net_stats;
4191 }
4192
4193 /**
4194  * e1000_change_mtu - Change the Maximum Transfer Unit
4195  * @netdev: network interface device structure
4196  * @new_mtu: new value for maximum frame size
4197  *
4198  * Returns 0 on success, negative on failure
4199  **/
4200 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4201 {
4202         struct e1000_adapter *adapter = netdev_priv(netdev);
4203         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4204
4205         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4206             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
4207                 e_err("Invalid MTU setting\n");
4208                 return -EINVAL;
4209         }
4210
4211         /* Jumbo frame size limits */
4212         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
4213                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4214                         e_err("Jumbo Frames not supported.\n");
4215                         return -EINVAL;
4216                 }
4217                 if (adapter->hw.phy.type == e1000_phy_ife) {
4218                         e_err("Jumbo Frames not supported.\n");
4219                         return -EINVAL;
4220                 }
4221         }
4222
4223 #define MAX_STD_JUMBO_FRAME_SIZE 9234
4224         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
4225                 e_err("MTU > 9216 not supported.\n");
4226                 return -EINVAL;
4227         }
4228
4229         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4230                 msleep(1);
4231         /* e1000e_down has a dependency on max_frame_size */
4232         adapter->max_frame_size = max_frame;
4233         if (netif_running(netdev))
4234                 e1000e_down(adapter);
4235
4236         /*
4237          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4238          * means we reserve 2 more, this pushes us to allocate from the next
4239          * larger slab size.
4240          * i.e. RXBUFFER_2048 --> size-4096 slab
4241          * However with the new *_jumbo_rx* routines, jumbo receives will use
4242          * fragmented skbs
4243          */
4244
4245         if (max_frame <= 256)
4246                 adapter->rx_buffer_len = 256;
4247         else if (max_frame <= 512)
4248                 adapter->rx_buffer_len = 512;
4249         else if (max_frame <= 1024)
4250                 adapter->rx_buffer_len = 1024;
4251         else if (max_frame <= 2048)
4252                 adapter->rx_buffer_len = 2048;
4253         else
4254                 adapter->rx_buffer_len = 4096;
4255
4256         /* adjust allocation if LPE protects us, and we aren't using SBP */
4257         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4258              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4259                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4260                                          + ETH_FCS_LEN;
4261
4262         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4263         netdev->mtu = new_mtu;
4264
4265         if (netif_running(netdev))
4266                 e1000e_up(adapter);
4267         else
4268                 e1000e_reset(adapter);
4269
4270         clear_bit(__E1000_RESETTING, &adapter->state);
4271
4272         return 0;
4273 }
4274
4275 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4276                            int cmd)
4277 {
4278         struct e1000_adapter *adapter = netdev_priv(netdev);
4279         struct mii_ioctl_data *data = if_mii(ifr);
4280
4281         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4282                 return -EOPNOTSUPP;
4283
4284         switch (cmd) {
4285         case SIOCGMIIPHY:
4286                 data->phy_id = adapter->hw.phy.addr;
4287                 break;
4288         case SIOCGMIIREG:
4289                 if (!capable(CAP_NET_ADMIN))
4290                         return -EPERM;
4291                 switch (data->reg_num & 0x1F) {
4292                 case MII_BMCR:
4293                         data->val_out = adapter->phy_regs.bmcr;
4294                         break;
4295                 case MII_BMSR:
4296                         data->val_out = adapter->phy_regs.bmsr;
4297                         break;
4298                 case MII_PHYSID1:
4299                         data->val_out = (adapter->hw.phy.id >> 16);
4300                         break;
4301                 case MII_PHYSID2:
4302                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4303                         break;
4304                 case MII_ADVERTISE:
4305                         data->val_out = adapter->phy_regs.advertise;
4306                         break;
4307                 case MII_LPA:
4308                         data->val_out = adapter->phy_regs.lpa;
4309                         break;
4310                 case MII_EXPANSION:
4311                         data->val_out = adapter->phy_regs.expansion;
4312                         break;
4313                 case MII_CTRL1000:
4314                         data->val_out = adapter->phy_regs.ctrl1000;
4315                         break;
4316                 case MII_STAT1000:
4317                         data->val_out = adapter->phy_regs.stat1000;
4318                         break;
4319                 case MII_ESTATUS:
4320                         data->val_out = adapter->phy_regs.estatus;
4321                         break;
4322                 default:
4323                         return -EIO;
4324                 }
4325                 break;
4326         case SIOCSMIIREG:
4327         default:
4328                 return -EOPNOTSUPP;
4329         }
4330         return 0;
4331 }
4332
4333 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4334 {
4335         switch (cmd) {
4336         case SIOCGMIIPHY:
4337         case SIOCGMIIREG:
4338         case SIOCSMIIREG:
4339                 return e1000_mii_ioctl(netdev, ifr, cmd);
4340         default:
4341                 return -EOPNOTSUPP;
4342         }
4343 }
4344
4345 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4346 {
4347         struct net_device *netdev = pci_get_drvdata(pdev);
4348         struct e1000_adapter *adapter = netdev_priv(netdev);
4349         struct e1000_hw *hw = &adapter->hw;
4350         u32 ctrl, ctrl_ext, rctl, status;
4351         u32 wufc = adapter->wol;
4352         int retval = 0;
4353
4354         netif_device_detach(netdev);
4355
4356         if (netif_running(netdev)) {
4357                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4358                 e1000e_down(adapter);
4359                 e1000_free_irq(adapter);
4360         }
4361         e1000e_reset_interrupt_capability(adapter);
4362
4363         retval = pci_save_state(pdev);
4364         if (retval)
4365                 return retval;
4366
4367         status = er32(STATUS);
4368         if (status & E1000_STATUS_LU)
4369                 wufc &= ~E1000_WUFC_LNKC;
4370
4371         if (wufc) {
4372                 e1000_setup_rctl(adapter);
4373                 e1000_set_multi(netdev);
4374
4375                 /* turn on all-multi mode if wake on multicast is enabled */
4376                 if (wufc & E1000_WUFC_MC) {
4377                         rctl = er32(RCTL);
4378                         rctl |= E1000_RCTL_MPE;
4379                         ew32(RCTL, rctl);
4380                 }
4381
4382                 ctrl = er32(CTRL);
4383                 /* advertise wake from D3Cold */
4384                 #define E1000_CTRL_ADVD3WUC 0x00100000
4385                 /* phy power management enable */
4386                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4387                 ctrl |= E1000_CTRL_ADVD3WUC |
4388                         E1000_CTRL_EN_PHY_PWR_MGMT;
4389                 ew32(CTRL, ctrl);
4390
4391                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4392                     adapter->hw.phy.media_type ==
4393                     e1000_media_type_internal_serdes) {
4394                         /* keep the laser running in D3 */
4395                         ctrl_ext = er32(CTRL_EXT);
4396                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4397                         ew32(CTRL_EXT, ctrl_ext);
4398                 }
4399
4400                 if (adapter->flags & FLAG_IS_ICH)
4401                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4402
4403                 /* Allow time for pending master requests to run */
4404                 e1000e_disable_pcie_master(&adapter->hw);
4405
4406                 ew32(WUC, E1000_WUC_PME_EN);
4407                 ew32(WUFC, wufc);
4408                 pci_enable_wake(pdev, PCI_D3hot, 1);
4409                 pci_enable_wake(pdev, PCI_D3cold, 1);
4410         } else {
4411                 ew32(WUC, 0);
4412                 ew32(WUFC, 0);
4413                 pci_enable_wake(pdev, PCI_D3hot, 0);
4414                 pci_enable_wake(pdev, PCI_D3cold, 0);
4415         }
4416
4417         /* make sure adapter isn't asleep if manageability is enabled */
4418         if (adapter->flags & FLAG_MNG_PT_ENABLED) {
4419                 pci_enable_wake(pdev, PCI_D3hot, 1);
4420                 pci_enable_wake(pdev, PCI_D3cold, 1);
4421         }
4422
4423         if (adapter->hw.phy.type == e1000_phy_igp_3)
4424                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4425
4426         /*
4427          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4428          * would have already happened in close and is redundant.
4429          */
4430         e1000_release_hw_control(adapter);
4431
4432         pci_disable_device(pdev);
4433
4434         /*
4435          * The pci-e switch on some quad port adapters will report a
4436          * correctable error when the MAC transitions from D0 to D3.  To
4437          * prevent this we need to mask off the correctable errors on the
4438          * downstream port of the pci-e switch.
4439          */
4440         if (adapter->flags & FLAG_IS_QUAD_PORT) {
4441                 struct pci_dev *us_dev = pdev->bus->self;
4442                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
4443                 u16 devctl;
4444
4445                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
4446                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
4447                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
4448
4449                 pci_set_power_state(pdev, pci_choose_state(pdev, state));
4450
4451                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
4452         } else {
4453                 pci_set_power_state(pdev, pci_choose_state(pdev, state));
4454         }
4455
4456         return 0;
4457 }
4458
4459 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4460 {
4461         int pos;
4462         u16 val;
4463
4464         /*
4465          * 82573 workaround - disable L1 ASPM on mobile chipsets
4466          *
4467          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4468          * resulting in lost data or garbage information on the pci-e link
4469          * level. This could result in (false) bad EEPROM checksum errors,
4470          * long ping times (up to 2s) or even a system freeze/hang.
4471          *
4472          * Unfortunately this feature saves about 1W power consumption when
4473          * active.
4474          */
4475         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4476         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4477         if (val & 0x2) {
4478                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4479                 val &= ~0x2;
4480                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4481         }
4482 }
4483
4484 #ifdef CONFIG_PM
4485 static int e1000_resume(struct pci_dev *pdev)
4486 {
4487         struct net_device *netdev = pci_get_drvdata(pdev);
4488         struct e1000_adapter *adapter = netdev_priv(netdev);
4489         struct e1000_hw *hw = &adapter->hw;
4490         u32 err;
4491
4492         pci_set_power_state(pdev, PCI_D0);
4493         pci_restore_state(pdev);
4494         e1000e_disable_l1aspm(pdev);
4495
4496         err = pci_enable_device_mem(pdev);
4497         if (err) {
4498                 dev_err(&pdev->dev,
4499                         "Cannot enable PCI device from suspend\n");
4500                 return err;
4501         }
4502
4503         /* AER (Advanced Error Reporting) hooks */
4504         err = pci_enable_pcie_error_reporting(pdev);
4505         if (err) {
4506                 dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
4507                                     "0x%x\n", err);
4508                 /* non-fatal, continue */
4509         }
4510
4511         pci_set_master(pdev);
4512
4513         pci_enable_wake(pdev, PCI_D3hot, 0);
4514         pci_enable_wake(pdev, PCI_D3cold, 0);
4515
4516         e1000e_set_interrupt_capability(adapter);
4517         if (netif_running(netdev)) {
4518                 err = e1000_request_irq(adapter);
4519                 if (err)
4520                         return err;
4521         }
4522
4523         e1000e_power_up_phy(adapter);
4524         e1000e_reset(adapter);
4525         ew32(WUS, ~0);
4526
4527         e1000_init_manageability(adapter);
4528
4529         if (netif_running(netdev))
4530                 e1000e_up(adapter);
4531
4532         netif_device_attach(netdev);
4533
4534         /*
4535          * If the controller has AMT, do not set DRV_LOAD until the interface
4536          * is up.  For all other cases, let the f/w know that the h/w is now
4537          * under the control of the driver.
4538          */
4539         if (!(adapter->flags & FLAG_HAS_AMT))
4540                 e1000_get_hw_control(adapter);
4541
4542         return 0;
4543 }
4544 #endif
4545
4546 static void e1000_shutdown(struct pci_dev *pdev)
4547 {
4548         e1000_suspend(pdev, PMSG_SUSPEND);
4549 }
4550
4551 #ifdef CONFIG_NET_POLL_CONTROLLER
4552 /*
4553  * Polling 'interrupt' - used by things like netconsole to send skbs
4554  * without having to re-enable interrupts. It's not called while
4555  * the interrupt routine is executing.
4556  */
4557 static void e1000_netpoll(struct net_device *netdev)
4558 {
4559         struct e1000_adapter *adapter = netdev_priv(netdev);
4560
4561         disable_irq(adapter->pdev->irq);
4562         e1000_intr(adapter->pdev->irq, netdev);
4563
4564         enable_irq(adapter->pdev->irq);
4565 }
4566 #endif
4567
4568 /**
4569  * e1000_io_error_detected - called when PCI error is detected
4570  * @pdev: Pointer to PCI device
4571  * @state: The current pci connection state
4572  *
4573  * This function is called after a PCI bus error affecting
4574  * this device has been detected.
4575  */
4576 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4577                                                 pci_channel_state_t state)
4578 {
4579         struct net_device *netdev = pci_get_drvdata(pdev);
4580         struct e1000_adapter *adapter = netdev_priv(netdev);
4581
4582         netif_device_detach(netdev);
4583
4584         if (netif_running(netdev))
4585                 e1000e_down(adapter);
4586         pci_disable_device(pdev);
4587
4588         /* Request a slot slot reset. */
4589         return PCI_ERS_RESULT_NEED_RESET;
4590 }
4591
4592 /**
4593  * e1000_io_slot_reset - called after the pci bus has been reset.
4594  * @pdev: Pointer to PCI device
4595  *
4596  * Restart the card from scratch, as if from a cold-boot. Implementation
4597  * resembles the first-half of the e1000_resume routine.
4598  */
4599 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4600 {
4601         struct net_device *netdev = pci_get_drvdata(pdev);
4602         struct e1000_adapter *adapter = netdev_priv(netdev);
4603         struct e1000_hw *hw = &adapter->hw;
4604         int err;
4605         pci_ers_result_t result;
4606
4607         e1000e_disable_l1aspm(pdev);
4608         err = pci_enable_device_mem(pdev);
4609         if (err) {
4610                 dev_err(&pdev->dev,
4611                         "Cannot re-enable PCI device after reset.\n");
4612                 result = PCI_ERS_RESULT_DISCONNECT;
4613         } else {
4614                 pci_set_master(pdev);
4615                 pci_restore_state(pdev);
4616
4617                 pci_enable_wake(pdev, PCI_D3hot, 0);
4618                 pci_enable_wake(pdev, PCI_D3cold, 0);
4619
4620                 e1000e_reset(adapter);
4621                 ew32(WUS, ~0);
4622                 result = PCI_ERS_RESULT_RECOVERED;
4623         }
4624
4625         pci_cleanup_aer_uncorrect_error_status(pdev);
4626
4627         return result;
4628 }
4629
4630 /**
4631  * e1000_io_resume - called when traffic can start flowing again.
4632  * @pdev: Pointer to PCI device
4633  *
4634  * This callback is called when the error recovery driver tells us that
4635  * its OK to resume normal operation. Implementation resembles the
4636  * second-half of the e1000_resume routine.
4637  */
4638 static void e1000_io_resume(struct pci_dev *pdev)
4639 {
4640         struct net_device *netdev = pci_get_drvdata(pdev);
4641         struct e1000_adapter *adapter = netdev_priv(netdev);
4642
4643         e1000_init_manageability(adapter);
4644
4645         if (netif_running(netdev)) {
4646                 if (e1000e_up(adapter)) {
4647                         dev_err(&pdev->dev,
4648                                 "can't bring device back up after reset\n");
4649                         return;
4650                 }
4651         }
4652
4653         netif_device_attach(netdev);
4654
4655         /*
4656          * If the controller has AMT, do not set DRV_LOAD until the interface
4657          * is up.  For all other cases, let the f/w know that the h/w is now
4658          * under the control of the driver.
4659          */
4660         if (!(adapter->flags & FLAG_HAS_AMT))
4661                 e1000_get_hw_control(adapter);
4662
4663 }
4664
4665 static void e1000_print_device_info(struct e1000_adapter *adapter)
4666 {
4667         struct e1000_hw *hw = &adapter->hw;
4668         struct net_device *netdev = adapter->netdev;
4669         u32 pba_num;
4670
4671         /* print bus type/speed/width info */
4672         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
4673                /* bus width */
4674                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4675                 "Width x1"),
4676                /* MAC address */
4677                netdev->dev_addr);
4678         e_info("Intel(R) PRO/%s Network Connection\n",
4679                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4680         e1000e_read_pba_num(hw, &pba_num);
4681         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4682                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4683 }
4684
4685 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4686 {
4687         struct e1000_hw *hw = &adapter->hw;
4688         int ret_val;
4689         u16 buf = 0;
4690
4691         if (hw->mac.type != e1000_82573)
4692                 return;
4693
4694         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4695         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
4696                 /* Deep Smart Power Down (DSPD) */
4697                 dev_warn(&adapter->pdev->dev,
4698                          "Warning: detected DSPD enabled in EEPROM\n");
4699         }
4700
4701         ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4702         if (!ret_val && (le16_to_cpu(buf) & (3 << 2))) {
4703                 /* ASPM enable */
4704                 dev_warn(&adapter->pdev->dev,
4705                          "Warning: detected ASPM enabled in EEPROM\n");
4706         }
4707 }
4708
4709 static const struct net_device_ops e1000e_netdev_ops = {
4710         .ndo_open               = e1000_open,
4711         .ndo_stop               = e1000_close,
4712         .ndo_start_xmit         = e1000_xmit_frame,
4713         .ndo_get_stats          = e1000_get_stats,
4714         .ndo_set_multicast_list = e1000_set_multi,
4715         .ndo_set_mac_address    = e1000_set_mac,
4716         .ndo_change_mtu         = e1000_change_mtu,
4717         .ndo_do_ioctl           = e1000_ioctl,
4718         .ndo_tx_timeout         = e1000_tx_timeout,
4719         .ndo_validate_addr      = eth_validate_addr,
4720
4721         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
4722         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
4723         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
4724 #ifdef CONFIG_NET_POLL_CONTROLLER
4725         .ndo_poll_controller    = e1000_netpoll,
4726 #endif
4727 };
4728
4729 /**
4730  * e1000_probe - Device Initialization Routine
4731  * @pdev: PCI device information struct
4732  * @ent: entry in e1000_pci_tbl
4733  *
4734  * Returns 0 on success, negative on failure
4735  *
4736  * e1000_probe initializes an adapter identified by a pci_dev structure.
4737  * The OS initialization, configuring of the adapter private structure,
4738  * and a hardware reset occur.
4739  **/
4740 static int __devinit e1000_probe(struct pci_dev *pdev,
4741                                  const struct pci_device_id *ent)
4742 {
4743         struct net_device *netdev;
4744         struct e1000_adapter *adapter;
4745         struct e1000_hw *hw;
4746         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4747         resource_size_t mmio_start, mmio_len;
4748         resource_size_t flash_start, flash_len;
4749
4750         static int cards_found;
4751         int i, err, pci_using_dac;
4752         u16 eeprom_data = 0;
4753         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4754
4755         e1000e_disable_l1aspm(pdev);
4756
4757         err = pci_enable_device_mem(pdev);
4758         if (err)
4759                 return err;
4760
4761         pci_using_dac = 0;
4762         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
4763         if (!err) {
4764                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
4765                 if (!err)
4766                         pci_using_dac = 1;
4767         } else {
4768                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
4769                 if (err) {
4770                         err = pci_set_consistent_dma_mask(pdev,
4771                                                           DMA_32BIT_MASK);
4772                         if (err) {
4773                                 dev_err(&pdev->dev, "No usable DMA "
4774                                         "configuration, aborting\n");
4775                                 goto err_dma;
4776                         }
4777                 }
4778         }
4779
4780         err = pci_request_selected_regions_exclusive(pdev,
4781                                           pci_select_bars(pdev, IORESOURCE_MEM),
4782                                           e1000e_driver_name);
4783         if (err)
4784                 goto err_pci_reg;
4785
4786         pci_set_master(pdev);
4787         /* PCI config space info */
4788         err = pci_save_state(pdev);
4789         if (err)
4790                 goto err_alloc_etherdev;
4791
4792         err = -ENOMEM;
4793         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4794         if (!netdev)
4795                 goto err_alloc_etherdev;
4796
4797         SET_NETDEV_DEV(netdev, &pdev->dev);
4798
4799         pci_set_drvdata(pdev, netdev);
4800         adapter = netdev_priv(netdev);
4801         hw = &adapter->hw;
4802         adapter->netdev = netdev;
4803         adapter->pdev = pdev;
4804         adapter->ei = ei;
4805         adapter->pba = ei->pba;
4806         adapter->flags = ei->flags;
4807         adapter->flags2 = ei->flags2;
4808         adapter->hw.adapter = adapter;
4809         adapter->hw.mac.type = ei->mac;
4810         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4811
4812         mmio_start = pci_resource_start(pdev, 0);
4813         mmio_len = pci_resource_len(pdev, 0);
4814
4815         err = -EIO;
4816         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4817         if (!adapter->hw.hw_addr)
4818                 goto err_ioremap;
4819
4820         if ((adapter->flags & FLAG_HAS_FLASH) &&
4821             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
4822                 flash_start = pci_resource_start(pdev, 1);
4823                 flash_len = pci_resource_len(pdev, 1);
4824                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
4825                 if (!adapter->hw.flash_address)
4826                         goto err_flashmap;
4827         }
4828
4829         /* construct the net_device struct */
4830         netdev->netdev_ops              = &e1000e_netdev_ops;
4831         e1000e_set_ethtool_ops(netdev);
4832         netdev->watchdog_timeo          = 5 * HZ;
4833         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
4834         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
4835
4836         netdev->mem_start = mmio_start;
4837         netdev->mem_end = mmio_start + mmio_len;
4838
4839         adapter->bd_number = cards_found++;
4840
4841         e1000e_check_options(adapter);
4842
4843         /* setup adapter struct */
4844         err = e1000_sw_init(adapter);
4845         if (err)
4846                 goto err_sw_init;
4847
4848         err = -EIO;
4849
4850         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4851         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
4852         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4853
4854         err = ei->get_variants(adapter);
4855         if (err)
4856                 goto err_hw_init;
4857
4858         if ((adapter->flags & FLAG_IS_ICH) &&
4859             (adapter->flags & FLAG_READ_ONLY_NVM))
4860                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
4861
4862         hw->mac.ops.get_bus_info(&adapter->hw);
4863
4864         adapter->hw.phy.autoneg_wait_to_complete = 0;
4865
4866         /* Copper options */
4867         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
4868                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
4869                 adapter->hw.phy.disable_polarity_correction = 0;
4870                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
4871         }
4872
4873         if (e1000_check_reset_block(&adapter->hw))
4874                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
4875
4876         netdev->features = NETIF_F_SG |
4877                            NETIF_F_HW_CSUM |
4878                            NETIF_F_HW_VLAN_TX |
4879                            NETIF_F_HW_VLAN_RX;
4880
4881         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
4882                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
4883
4884         netdev->features |= NETIF_F_TSO;
4885         netdev->features |= NETIF_F_TSO6;
4886
4887         netdev->vlan_features |= NETIF_F_TSO;
4888         netdev->vlan_features |= NETIF_F_TSO6;
4889         netdev->vlan_features |= NETIF_F_HW_CSUM;
4890         netdev->vlan_features |= NETIF_F_SG;
4891
4892         if (pci_using_dac)
4893                 netdev->features |= NETIF_F_HIGHDMA;
4894
4895         if (e1000e_enable_mng_pass_thru(&adapter->hw))
4896                 adapter->flags |= FLAG_MNG_PT_ENABLED;
4897
4898         /*
4899          * before reading the NVM, reset the controller to
4900          * put the device in a known good starting state
4901          */
4902         adapter->hw.mac.ops.reset_hw(&adapter->hw);
4903
4904         /*
4905          * systems with ASPM and others may see the checksum fail on the first
4906          * attempt. Let's give it a few tries
4907          */
4908         for (i = 0;; i++) {
4909                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
4910                         break;
4911                 if (i == 2) {
4912                         e_err("The NVM Checksum Is Not Valid\n");
4913                         err = -EIO;
4914                         goto err_eeprom;
4915                 }
4916         }
4917
4918         e1000_eeprom_checks(adapter);
4919
4920         /* copy the MAC address out of the NVM */
4921         if (e1000e_read_mac_addr(&adapter->hw))
4922                 e_err("NVM Read Error while reading MAC address\n");
4923
4924         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
4925         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
4926
4927         if (!is_valid_ether_addr(netdev->perm_addr)) {
4928                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
4929                 err = -EIO;
4930                 goto err_eeprom;
4931         }
4932
4933         init_timer(&adapter->watchdog_timer);
4934         adapter->watchdog_timer.function = &e1000_watchdog;
4935         adapter->watchdog_timer.data = (unsigned long) adapter;
4936
4937         init_timer(&adapter->phy_info_timer);
4938         adapter->phy_info_timer.function = &e1000_update_phy_info;
4939         adapter->phy_info_timer.data = (unsigned long) adapter;
4940
4941         INIT_WORK(&adapter->reset_task, e1000_reset_task);
4942         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
4943         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
4944         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
4945
4946         /* Initialize link parameters. User can change them with ethtool */
4947         adapter->hw.mac.autoneg = 1;
4948         adapter->fc_autoneg = 1;
4949         adapter->hw.fc.requested_mode = e1000_fc_default;
4950         adapter->hw.fc.current_mode = e1000_fc_default;
4951         adapter->hw.phy.autoneg_advertised = 0x2f;
4952
4953         /* ring size defaults */
4954         adapter->rx_ring->count = 256;
4955         adapter->tx_ring->count = 256;
4956
4957         /*
4958          * Initial Wake on LAN setting - If APM wake is enabled in
4959          * the EEPROM, enable the ACPI Magic Packet filter
4960          */
4961         if (adapter->flags & FLAG_APME_IN_WUC) {
4962                 /* APME bit in EEPROM is mapped to WUC.APME */
4963                 eeprom_data = er32(WUC);
4964                 eeprom_apme_mask = E1000_WUC_APME;
4965         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
4966                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
4967                     (adapter->hw.bus.func == 1))
4968                         e1000_read_nvm(&adapter->hw,
4969                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
4970                 else
4971                         e1000_read_nvm(&adapter->hw,
4972                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
4973         }
4974
4975         /* fetch WoL from EEPROM */
4976         if (eeprom_data & eeprom_apme_mask)
4977                 adapter->eeprom_wol |= E1000_WUFC_MAG;
4978
4979         /*
4980          * now that we have the eeprom settings, apply the special cases
4981          * where the eeprom may be wrong or the board simply won't support
4982          * wake on lan on a particular port
4983          */
4984         if (!(adapter->flags & FLAG_HAS_WOL))
4985                 adapter->eeprom_wol = 0;
4986
4987         /* initialize the wol settings based on the eeprom settings */
4988         adapter->wol = adapter->eeprom_wol;
4989         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
4990
4991         /* save off EEPROM version number */
4992         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
4993
4994         /* reset the hardware with the new settings */
4995         e1000e_reset(adapter);
4996
4997         /*
4998          * If the controller has AMT, do not set DRV_LOAD until the interface
4999          * is up.  For all other cases, let the f/w know that the h/w is now
5000          * under the control of the driver.
5001          */
5002         if (!(adapter->flags & FLAG_HAS_AMT))
5003                 e1000_get_hw_control(adapter);
5004
5005         /* tell the stack to leave us alone until e1000_open() is called */
5006         netif_carrier_off(netdev);
5007         netif_tx_stop_all_queues(netdev);
5008
5009         strcpy(netdev->name, "eth%d");
5010         err = register_netdev(netdev);
5011         if (err)
5012                 goto err_register;
5013
5014         e1000_print_device_info(adapter);
5015
5016         return 0;
5017
5018 err_register:
5019         if (!(adapter->flags & FLAG_HAS_AMT))
5020                 e1000_release_hw_control(adapter);
5021 err_eeprom:
5022         if (!e1000_check_reset_block(&adapter->hw))
5023                 e1000_phy_hw_reset(&adapter->hw);
5024 err_hw_init:
5025
5026         kfree(adapter->tx_ring);
5027         kfree(adapter->rx_ring);
5028 err_sw_init:
5029         if (adapter->hw.flash_address)
5030                 iounmap(adapter->hw.flash_address);
5031         e1000e_reset_interrupt_capability(adapter);
5032 err_flashmap:
5033         iounmap(adapter->hw.hw_addr);
5034 err_ioremap:
5035         free_netdev(netdev);
5036 err_alloc_etherdev:
5037         pci_release_selected_regions(pdev,
5038                                      pci_select_bars(pdev, IORESOURCE_MEM));
5039 err_pci_reg:
5040 err_dma:
5041         pci_disable_device(pdev);
5042         return err;
5043 }
5044
5045 /**
5046  * e1000_remove - Device Removal Routine
5047  * @pdev: PCI device information struct
5048  *
5049  * e1000_remove is called by the PCI subsystem to alert the driver
5050  * that it should release a PCI device.  The could be caused by a
5051  * Hot-Plug event, or because the driver is going to be removed from
5052  * memory.
5053  **/
5054 static void __devexit e1000_remove(struct pci_dev *pdev)
5055 {
5056         struct net_device *netdev = pci_get_drvdata(pdev);
5057         struct e1000_adapter *adapter = netdev_priv(netdev);
5058         int err;
5059
5060         /*
5061          * flush_scheduled work may reschedule our watchdog task, so
5062          * explicitly disable watchdog tasks from being rescheduled
5063          */
5064         set_bit(__E1000_DOWN, &adapter->state);
5065         del_timer_sync(&adapter->watchdog_timer);
5066         del_timer_sync(&adapter->phy_info_timer);
5067
5068         flush_scheduled_work();
5069
5070         /*
5071          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5072          * would have already happened in close and is redundant.
5073          */
5074         e1000_release_hw_control(adapter);
5075
5076         unregister_netdev(netdev);
5077
5078         if (!e1000_check_reset_block(&adapter->hw))
5079                 e1000_phy_hw_reset(&adapter->hw);
5080
5081         e1000e_reset_interrupt_capability(adapter);
5082         kfree(adapter->tx_ring);
5083         kfree(adapter->rx_ring);
5084
5085         iounmap(adapter->hw.hw_addr);
5086         if (adapter->hw.flash_address)
5087                 iounmap(adapter->hw.flash_address);
5088         pci_release_selected_regions(pdev,
5089                                      pci_select_bars(pdev, IORESOURCE_MEM));
5090
5091         free_netdev(netdev);
5092
5093         /* AER disable */
5094         err = pci_disable_pcie_error_reporting(pdev);
5095         if (err)
5096                 dev_err(&pdev->dev,
5097                         "pci_disable_pcie_error_reporting failed 0x%x\n", err);
5098
5099         pci_disable_device(pdev);
5100 }
5101
5102 /* PCI Error Recovery (ERS) */
5103 static struct pci_error_handlers e1000_err_handler = {
5104         .error_detected = e1000_io_error_detected,
5105         .slot_reset = e1000_io_slot_reset,
5106         .resume = e1000_io_resume,
5107 };
5108
5109 static struct pci_device_id e1000_pci_tbl[] = {
5110         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5111         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5112         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5113         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5114         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5115         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5116         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5117         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5118         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5119
5120         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5121         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5122         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5123         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5124
5125         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5126         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5127         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5128
5129         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5130         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
5131         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
5132
5133         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5134           board_80003es2lan },
5135         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5136           board_80003es2lan },
5137         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5138           board_80003es2lan },
5139         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5140           board_80003es2lan },
5141
5142         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5143         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5144         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5145         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5146         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5147         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5148         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5149
5150         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5151         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5152         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5153         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5154         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5155         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5156         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5157         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5158         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5159
5160         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5161         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5162         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5163
5164         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5165         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5166
5167         { }     /* terminate list */
5168 };
5169 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5170
5171 /* PCI Device API Driver */
5172 static struct pci_driver e1000_driver = {
5173         .name     = e1000e_driver_name,
5174         .id_table = e1000_pci_tbl,
5175         .probe    = e1000_probe,
5176         .remove   = __devexit_p(e1000_remove),
5177 #ifdef CONFIG_PM
5178         /* Power Management Hooks */
5179         .suspend  = e1000_suspend,
5180         .resume   = e1000_resume,
5181 #endif
5182         .shutdown = e1000_shutdown,
5183         .err_handler = &e1000_err_handler
5184 };
5185
5186 /**
5187  * e1000_init_module - Driver Registration Routine
5188  *
5189  * e1000_init_module is the first routine called when the driver is
5190  * loaded. All it does is register with the PCI subsystem.
5191  **/
5192 static int __init e1000_init_module(void)
5193 {
5194         int ret;
5195         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
5196                e1000e_driver_name, e1000e_driver_version);
5197         printk(KERN_INFO "%s: Copyright (c) 1999-2008 Intel Corporation.\n",
5198                e1000e_driver_name);
5199         ret = pci_register_driver(&e1000_driver);
5200         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name,
5201                                PM_QOS_DEFAULT_VALUE);
5202                                 
5203         return ret;
5204 }
5205 module_init(e1000_init_module);
5206
5207 /**
5208  * e1000_exit_module - Driver Exit Cleanup Routine
5209  *
5210  * e1000_exit_module is called just before the driver is removed
5211  * from memory.
5212  **/
5213 static void __exit e1000_exit_module(void)
5214 {
5215         pci_unregister_driver(&e1000_driver);
5216         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name);
5217 }
5218 module_exit(e1000_exit_module);
5219
5220
5221 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5222 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5223 MODULE_LICENSE("GPL");
5224 MODULE_VERSION(DRV_VERSION);
5225
5226 /* e1000_main.c */