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