e1000e: Do not allow requeue of freed skb
[linux-2.6] / drivers / net / e1000e / netdev.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2007 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
47 #include "e1000.h"
48
49 #define DRV_VERSION "0.2.0"
50 char e1000e_driver_name[] = "e1000e";
51 const char e1000e_driver_version[] = DRV_VERSION;
52
53 static const struct e1000_info *e1000_info_tbl[] = {
54         [board_82571]           = &e1000_82571_info,
55         [board_82572]           = &e1000_82572_info,
56         [board_82573]           = &e1000_82573_info,
57         [board_80003es2lan]     = &e1000_es2_info,
58         [board_ich8lan]         = &e1000_ich8_info,
59         [board_ich9lan]         = &e1000_ich9_info,
60 };
61
62 #ifdef DEBUG
63 /**
64  * e1000_get_hw_dev_name - return device name string
65  * used by hardware layer to print debugging information
66  **/
67 char *e1000e_get_hw_dev_name(struct e1000_hw *hw)
68 {
69         struct e1000_adapter *adapter = hw->back;
70         struct net_device *netdev = adapter->netdev;
71         return netdev->name;
72 }
73 #endif
74
75 /**
76  * e1000_desc_unused - calculate if we have unused descriptors
77  **/
78 static int e1000_desc_unused(struct e1000_ring *ring)
79 {
80         if (ring->next_to_clean > ring->next_to_use)
81                 return ring->next_to_clean - ring->next_to_use - 1;
82
83         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
84 }
85
86 /**
87  * e1000_receive_skb - helper function to handle rx indications
88  * @adapter: board private structure
89  * @status: descriptor status field as written by hardware
90  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
91  * @skb: pointer to sk_buff to be indicated to stack
92  **/
93 static void e1000_receive_skb(struct e1000_adapter *adapter,
94                               struct net_device *netdev,
95                               struct sk_buff *skb,
96                               u8 status, u16 vlan)
97 {
98         skb->protocol = eth_type_trans(skb, netdev);
99
100         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
101                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
102                                          le16_to_cpu(vlan) &
103                                          E1000_RXD_SPC_VLAN_MASK);
104         else
105                 netif_receive_skb(skb);
106
107         netdev->last_rx = jiffies;
108 }
109
110 /**
111  * e1000_rx_checksum - Receive Checksum Offload for 82543
112  * @adapter:     board private structure
113  * @status_err:  receive descriptor status and error fields
114  * @csum:       receive descriptor csum field
115  * @sk_buff:     socket buffer with received data
116  **/
117 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
118                               u32 csum, struct sk_buff *skb)
119 {
120         u16 status = (u16)status_err;
121         u8 errors = (u8)(status_err >> 24);
122         skb->ip_summed = CHECKSUM_NONE;
123
124         /* Ignore Checksum bit is set */
125         if (status & E1000_RXD_STAT_IXSM)
126                 return;
127         /* TCP/UDP checksum error bit is set */
128         if (errors & E1000_RXD_ERR_TCPE) {
129                 /* let the stack verify checksum errors */
130                 adapter->hw_csum_err++;
131                 return;
132         }
133
134         /* TCP/UDP Checksum has not been calculated */
135         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
136                 return;
137
138         /* It must be a TCP or UDP packet with a valid checksum */
139         if (status & E1000_RXD_STAT_TCPCS) {
140                 /* TCP checksum is good */
141                 skb->ip_summed = CHECKSUM_UNNECESSARY;
142         } else {
143                 /* IP fragment with UDP payload */
144                 /* Hardware complements the payload checksum, so we undo it
145                  * and then put the value in host order for further stack use.
146                  */
147                 csum = ntohl(csum ^ 0xFFFF);
148                 skb->csum = csum;
149                 skb->ip_summed = CHECKSUM_COMPLETE;
150         }
151         adapter->hw_csum_good++;
152 }
153
154 /**
155  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
156  * @adapter: address of board private structure
157  **/
158 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
159                                    int cleaned_count)
160 {
161         struct net_device *netdev = adapter->netdev;
162         struct pci_dev *pdev = adapter->pdev;
163         struct e1000_ring *rx_ring = adapter->rx_ring;
164         struct e1000_rx_desc *rx_desc;
165         struct e1000_buffer *buffer_info;
166         struct sk_buff *skb;
167         unsigned int i;
168         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
169
170         i = rx_ring->next_to_use;
171         buffer_info = &rx_ring->buffer_info[i];
172
173         while (cleaned_count--) {
174                 skb = buffer_info->skb;
175                 if (skb) {
176                         skb_trim(skb, 0);
177                         goto map_skb;
178                 }
179
180                 skb = netdev_alloc_skb(netdev, bufsz);
181                 if (!skb) {
182                         /* Better luck next round */
183                         adapter->alloc_rx_buff_failed++;
184                         break;
185                 }
186
187                 /* 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(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                 /* Force memory writes to complete before letting h/w
219                  * know there are new descriptors to fetch.  (Only
220                  * applicable for weak-ordered memory model archs,
221                  * such as IA-64). */
222                 wmb();
223                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
224         }
225 }
226
227 /**
228  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
229  * @adapter: address of board private structure
230  **/
231 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
232                                       int cleaned_count)
233 {
234         struct net_device *netdev = adapter->netdev;
235         struct pci_dev *pdev = adapter->pdev;
236         union e1000_rx_desc_packet_split *rx_desc;
237         struct e1000_ring *rx_ring = adapter->rx_ring;
238         struct e1000_buffer *buffer_info;
239         struct e1000_ps_page *ps_page;
240         struct sk_buff *skb;
241         unsigned int i, j;
242
243         i = rx_ring->next_to_use;
244         buffer_info = &rx_ring->buffer_info[i];
245
246         while (cleaned_count--) {
247                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
248
249                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
250                         ps_page = &rx_ring->ps_pages[(i * PS_PAGE_BUFFERS)
251                                                      + j];
252                         if (j < adapter->rx_ps_pages) {
253                                 if (!ps_page->page) {
254                                         ps_page->page = alloc_page(GFP_ATOMIC);
255                                         if (!ps_page->page) {
256                                                 adapter->alloc_rx_buff_failed++;
257                                                 goto no_buffers;
258                                         }
259                                         ps_page->dma = pci_map_page(pdev,
260                                                            ps_page->page,
261                                                            0, PAGE_SIZE,
262                                                            PCI_DMA_FROMDEVICE);
263                                         if (pci_dma_mapping_error(
264                                                         ps_page->dma)) {
265                                                 dev_err(&adapter->pdev->dev,
266                                                   "RX DMA page map failed\n");
267                                                 adapter->rx_dma_failed++;
268                                                 goto no_buffers;
269                                         }
270                                 }
271                                 /*
272                                  * Refresh the desc even if buffer_addrs
273                                  * didn't change because each write-back
274                                  * erases this info.
275                                  */
276                                 rx_desc->read.buffer_addr[j+1] =
277                                      cpu_to_le64(ps_page->dma);
278                         } else {
279                                 rx_desc->read.buffer_addr[j+1] = ~0;
280                         }
281                 }
282
283                 skb = netdev_alloc_skb(netdev,
284                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
285
286                 if (!skb) {
287                         adapter->alloc_rx_buff_failed++;
288                         break;
289                 }
290
291                 /* Make buffer alignment 2 beyond a 16 byte boundary
292                  * this will result in a 16 byte aligned IP header after
293                  * the 14 byte MAC header is removed
294                  */
295                 skb_reserve(skb, NET_IP_ALIGN);
296
297                 buffer_info->skb = skb;
298                 buffer_info->dma = pci_map_single(pdev, skb->data,
299                                                   adapter->rx_ps_bsize0,
300                                                   PCI_DMA_FROMDEVICE);
301                 if (pci_dma_mapping_error(buffer_info->dma)) {
302                         dev_err(&pdev->dev, "RX DMA map failed\n");
303                         adapter->rx_dma_failed++;
304                         /* cleanup skb */
305                         dev_kfree_skb_any(skb);
306                         buffer_info->skb = NULL;
307                         break;
308                 }
309
310                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
311
312                 i++;
313                 if (i == rx_ring->count)
314                         i = 0;
315                 buffer_info = &rx_ring->buffer_info[i];
316         }
317
318 no_buffers:
319         if (rx_ring->next_to_use != i) {
320                 rx_ring->next_to_use = i;
321
322                 if (!(i--))
323                         i = (rx_ring->count - 1);
324
325                 /* Force memory writes to complete before letting h/w
326                  * know there are new descriptors to fetch.  (Only
327                  * applicable for weak-ordered memory model archs,
328                  * such as IA-64). */
329                 wmb();
330                 /* Hardware increments by 16 bytes, but packet split
331                  * descriptors are 32 bytes...so we increment tail
332                  * twice as much.
333                  */
334                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
335         }
336 }
337
338 /**
339  * e1000_alloc_rx_buffers_jumbo - Replace used jumbo receive buffers
340  *
341  * @adapter: address of board private structure
342  * @cleaned_count: number of buffers to allocate this pass
343  **/
344 static void e1000_alloc_rx_buffers_jumbo(struct e1000_adapter *adapter,
345                                          int cleaned_count)
346 {
347         struct net_device *netdev = adapter->netdev;
348         struct pci_dev *pdev = adapter->pdev;
349         struct e1000_ring *rx_ring = adapter->rx_ring;
350         struct e1000_rx_desc *rx_desc;
351         struct e1000_buffer *buffer_info;
352         struct sk_buff *skb;
353         unsigned int i;
354         unsigned int bufsz = 256 -
355                              16 /*for skb_reserve */ -
356                              NET_IP_ALIGN;
357
358         i = rx_ring->next_to_use;
359         buffer_info = &rx_ring->buffer_info[i];
360
361         while (cleaned_count--) {
362                 skb = buffer_info->skb;
363                 if (skb) {
364                         skb_trim(skb, 0);
365                         goto check_page;
366                 }
367
368                 skb = netdev_alloc_skb(netdev, bufsz);
369                 if (!skb) {
370                         /* Better luck next round */
371                         adapter->alloc_rx_buff_failed++;
372                         break;
373                 }
374
375                 /* Make buffer alignment 2 beyond a 16 byte boundary
376                  * this will result in a 16 byte aligned IP header after
377                  * the 14 byte MAC header is removed
378                  */
379                 skb_reserve(skb, NET_IP_ALIGN);
380
381                 buffer_info->skb = skb;
382 check_page:
383                 /* allocate a new page if necessary */
384                 if (!buffer_info->page) {
385                         buffer_info->page = alloc_page(GFP_ATOMIC);
386                         if (!buffer_info->page) {
387                                 adapter->alloc_rx_buff_failed++;
388                                 break;
389                         }
390                 }
391
392                 if (!buffer_info->dma)
393                         buffer_info->dma = pci_map_page(pdev,
394                                                         buffer_info->page, 0,
395                                                         PAGE_SIZE,
396                                                         PCI_DMA_FROMDEVICE);
397                 if (pci_dma_mapping_error(buffer_info->dma)) {
398                         dev_err(&adapter->pdev->dev, "RX DMA page map failed\n");
399                         adapter->rx_dma_failed++;
400                         break;
401                 }
402
403                 rx_desc = E1000_RX_DESC(*rx_ring, i);
404                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
405
406                 i++;
407                 if (i == rx_ring->count)
408                         i = 0;
409                 buffer_info = &rx_ring->buffer_info[i];
410         }
411
412         if (rx_ring->next_to_use != i) {
413                 rx_ring->next_to_use = i;
414                 if (i-- == 0)
415                         i = (rx_ring->count - 1);
416
417                 /* Force memory writes to complete before letting h/w
418                  * know there are new descriptors to fetch.  (Only
419                  * applicable for weak-ordered memory model archs,
420                  * such as IA-64). */
421                 wmb();
422                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
423         }
424 }
425
426 /**
427  * e1000_clean_rx_irq - Send received data up the network stack; legacy
428  * @adapter: board private structure
429  *
430  * the return value indicates whether actual cleaning was done, there
431  * is no guarantee that everything was cleaned
432  **/
433 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
434                                int *work_done, int work_to_do)
435 {
436         struct net_device *netdev = adapter->netdev;
437         struct pci_dev *pdev = adapter->pdev;
438         struct e1000_ring *rx_ring = adapter->rx_ring;
439         struct e1000_rx_desc *rx_desc, *next_rxd;
440         struct e1000_buffer *buffer_info, *next_buffer;
441         u32 length;
442         unsigned int i;
443         int cleaned_count = 0;
444         bool cleaned = 0;
445         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
446
447         i = rx_ring->next_to_clean;
448         rx_desc = E1000_RX_DESC(*rx_ring, i);
449         buffer_info = &rx_ring->buffer_info[i];
450
451         while (rx_desc->status & E1000_RXD_STAT_DD) {
452                 struct sk_buff *skb;
453                 u8 status;
454
455                 if (*work_done >= work_to_do)
456                         break;
457                 (*work_done)++;
458
459                 status = rx_desc->status;
460                 skb = buffer_info->skb;
461                 buffer_info->skb = NULL;
462
463                 prefetch(skb->data - NET_IP_ALIGN);
464
465                 i++;
466                 if (i == rx_ring->count)
467                         i = 0;
468                 next_rxd = E1000_RX_DESC(*rx_ring, i);
469                 prefetch(next_rxd);
470
471                 next_buffer = &rx_ring->buffer_info[i];
472
473                 cleaned = 1;
474                 cleaned_count++;
475                 pci_unmap_single(pdev,
476                                  buffer_info->dma,
477                                  adapter->rx_buffer_len,
478                                  PCI_DMA_FROMDEVICE);
479                 buffer_info->dma = 0;
480
481                 length = le16_to_cpu(rx_desc->length);
482
483                 /* !EOP means multiple descriptors were used to store a single
484                  * packet, also make sure the frame isn't just CRC only */
485                 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
486                         /* All receives must fit into a single buffer */
487                         ndev_dbg(netdev, "%s: Receive packet consumed "
488                                  "multiple buffers\n", netdev->name);
489                         /* recycle */
490                         buffer_info->skb = skb;
491                         goto next_desc;
492                 }
493
494                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
495                         /* recycle */
496                         buffer_info->skb = skb;
497                         goto next_desc;
498                 }
499
500                 /* adjust length to remove Ethernet CRC */
501                 length -= 4;
502
503                 /* probably a little skewed due to removing CRC */
504                 total_rx_bytes += length;
505                 total_rx_packets++;
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                 if (length < copybreak) {
511                         struct sk_buff *new_skb =
512                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
513                         if (new_skb) {
514                                 skb_reserve(new_skb, NET_IP_ALIGN);
515                                 memcpy(new_skb->data - NET_IP_ALIGN,
516                                        skb->data - NET_IP_ALIGN,
517                                        length + NET_IP_ALIGN);
518                                 /* save the skb in buffer_info as good */
519                                 buffer_info->skb = skb;
520                                 skb = new_skb;
521                         }
522                         /* else just continue with the old one */
523                 }
524                 /* end copybreak code */
525                 skb_put(skb, length);
526
527                 /* Receive Checksum Offload */
528                 e1000_rx_checksum(adapter,
529                                   (u32)(status) |
530                                   ((u32)(rx_desc->errors) << 24),
531                                   le16_to_cpu(rx_desc->csum), skb);
532
533                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
534
535 next_desc:
536                 rx_desc->status = 0;
537
538                 /* return some buffers to hardware, one at a time is too slow */
539                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
540                         adapter->alloc_rx_buf(adapter, cleaned_count);
541                         cleaned_count = 0;
542                 }
543
544                 /* use prefetched values */
545                 rx_desc = next_rxd;
546                 buffer_info = next_buffer;
547         }
548         rx_ring->next_to_clean = i;
549
550         cleaned_count = e1000_desc_unused(rx_ring);
551         if (cleaned_count)
552                 adapter->alloc_rx_buf(adapter, cleaned_count);
553
554         adapter->total_rx_packets += total_rx_packets;
555         adapter->total_rx_bytes += total_rx_bytes;
556         return cleaned;
557 }
558
559 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
560                                u16 length)
561 {
562         bi->page = NULL;
563         skb->len += length;
564         skb->data_len += length;
565         skb->truesize += length;
566 }
567
568 static void e1000_put_txbuf(struct e1000_adapter *adapter,
569                              struct e1000_buffer *buffer_info)
570 {
571         if (buffer_info->dma) {
572                 pci_unmap_page(adapter->pdev, buffer_info->dma,
573                                buffer_info->length, PCI_DMA_TODEVICE);
574                 buffer_info->dma = 0;
575         }
576         if (buffer_info->skb) {
577                 dev_kfree_skb_any(buffer_info->skb);
578                 buffer_info->skb = NULL;
579         }
580 }
581
582 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
583 {
584         struct e1000_ring *tx_ring = adapter->tx_ring;
585         unsigned int i = tx_ring->next_to_clean;
586         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
587         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
588         struct net_device *netdev = adapter->netdev;
589
590         /* detected Tx unit hang */
591         ndev_err(netdev,
592                  "Detected Tx Unit Hang:\n"
593                  "  TDH                  <%x>\n"
594                  "  TDT                  <%x>\n"
595                  "  next_to_use          <%x>\n"
596                  "  next_to_clean        <%x>\n"
597                  "buffer_info[next_to_clean]:\n"
598                  "  time_stamp           <%lx>\n"
599                  "  next_to_watch        <%x>\n"
600                  "  jiffies              <%lx>\n"
601                  "  next_to_watch.status <%x>\n",
602                  readl(adapter->hw.hw_addr + tx_ring->head),
603                  readl(adapter->hw.hw_addr + tx_ring->tail),
604                  tx_ring->next_to_use,
605                  tx_ring->next_to_clean,
606                  tx_ring->buffer_info[eop].time_stamp,
607                  eop,
608                  jiffies,
609                  eop_desc->upper.fields.status);
610 }
611
612 /**
613  * e1000_clean_tx_irq - Reclaim resources after transmit completes
614  * @adapter: board private structure
615  *
616  * the return value indicates whether actual cleaning was done, there
617  * is no guarantee that everything was cleaned
618  **/
619 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
620 {
621         struct net_device *netdev = adapter->netdev;
622         struct e1000_hw *hw = &adapter->hw;
623         struct e1000_ring *tx_ring = adapter->tx_ring;
624         struct e1000_tx_desc *tx_desc, *eop_desc;
625         struct e1000_buffer *buffer_info;
626         unsigned int i, eop;
627         unsigned int count = 0;
628         bool cleaned = 0;
629         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
630
631         i = tx_ring->next_to_clean;
632         eop = tx_ring->buffer_info[i].next_to_watch;
633         eop_desc = E1000_TX_DESC(*tx_ring, eop);
634
635         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
636                 for (cleaned = 0; !cleaned; ) {
637                         tx_desc = E1000_TX_DESC(*tx_ring, i);
638                         buffer_info = &tx_ring->buffer_info[i];
639                         cleaned = (i == eop);
640
641                         if (cleaned) {
642                                 struct sk_buff *skb = buffer_info->skb;
643                                 unsigned int segs, bytecount;
644                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
645                                 /* multiply data chunks by size of headers */
646                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
647                                             skb->len;
648                                 total_tx_packets += segs;
649                                 total_tx_bytes += bytecount;
650                         }
651
652                         e1000_put_txbuf(adapter, buffer_info);
653                         tx_desc->upper.data = 0;
654
655                         i++;
656                         if (i == tx_ring->count)
657                                 i = 0;
658                 }
659
660                 eop = tx_ring->buffer_info[i].next_to_watch;
661                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
662 #define E1000_TX_WEIGHT 64
663                 /* weight of a sort for tx, to avoid endless transmit cleanup */
664                 if (count++ == E1000_TX_WEIGHT)
665                         break;
666         }
667
668         tx_ring->next_to_clean = i;
669
670 #define TX_WAKE_THRESHOLD 32
671         if (cleaned && netif_carrier_ok(netdev) &&
672                      e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
673                 /* Make sure that anybody stopping the queue after this
674                  * sees the new next_to_clean.
675                  */
676                 smp_mb();
677
678                 if (netif_queue_stopped(netdev) &&
679                     !(test_bit(__E1000_DOWN, &adapter->state))) {
680                         netif_wake_queue(netdev);
681                         ++adapter->restart_queue;
682                 }
683         }
684
685         if (adapter->detect_tx_hung) {
686                 /* Detect a transmit hang in hardware, this serializes the
687                  * check with the clearing of time_stamp and movement of i */
688                 adapter->detect_tx_hung = 0;
689                 if (tx_ring->buffer_info[eop].dma &&
690                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp
691                                + (adapter->tx_timeout_factor * HZ))
692                     && !(er32(STATUS) &
693                          E1000_STATUS_TXOFF)) {
694                         e1000_print_tx_hang(adapter);
695                         netif_stop_queue(netdev);
696                 }
697         }
698         adapter->total_tx_bytes += total_tx_bytes;
699         adapter->total_tx_packets += total_tx_packets;
700         return cleaned;
701 }
702
703 /**
704  * e1000_clean_rx_irq_jumbo - Send received data up the network stack; legacy
705  * @adapter: board private structure
706  *
707  * the return value indicates whether actual cleaning was done, there
708  * is no guarantee that everything was cleaned
709  **/
710 static bool e1000_clean_rx_irq_jumbo(struct e1000_adapter *adapter,
711                                      int *work_done, int work_to_do)
712 {
713         struct net_device *netdev = adapter->netdev;
714         struct pci_dev *pdev = adapter->pdev;
715         struct e1000_ring *rx_ring = adapter->rx_ring;
716         struct e1000_rx_desc *rx_desc, *next_rxd;
717         struct e1000_buffer *buffer_info, *next_buffer;
718         u32 length;
719         unsigned int i;
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(*rx_ring, i);
726         buffer_info = &rx_ring->buffer_info[i];
727
728         while (rx_desc->status & E1000_RXD_STAT_DD) {
729                 struct sk_buff *skb;
730                 u8 status;
731
732                 if (*work_done >= work_to_do)
733                         break;
734                 (*work_done)++;
735
736                 status = rx_desc->status;
737                 skb = buffer_info->skb;
738                 buffer_info->skb = NULL;
739
740                 i++;
741                 if (i == rx_ring->count)
742                         i = 0;
743                 next_rxd = E1000_RX_DESC(*rx_ring, i);
744                 prefetch(next_rxd);
745
746                 next_buffer = &rx_ring->buffer_info[i];
747
748                 cleaned = 1;
749                 cleaned_count++;
750                 pci_unmap_page(pdev,
751                                buffer_info->dma,
752                                PAGE_SIZE,
753                                PCI_DMA_FROMDEVICE);
754                 buffer_info->dma = 0;
755
756                 length = le16_to_cpu(rx_desc->length);
757
758                 /* errors is only valid for DD + EOP descriptors */
759                 if ((status & E1000_RXD_STAT_EOP) &&
760                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
761                         /* recycle both page and skb */
762                         buffer_info->skb = skb;
763                         /* an error means any chain goes out the window too */
764                         if (rx_ring->rx_skb_top)
765                                 dev_kfree_skb(rx_ring->rx_skb_top);
766                         rx_ring->rx_skb_top = NULL;
767                         goto next_desc;
768                 }
769
770 #define rxtop rx_ring->rx_skb_top
771                 if (!(status & E1000_RXD_STAT_EOP)) {
772                         /* this descriptor is only the beginning (or middle) */
773                         if (!rxtop) {
774                                 /* this is the beginning of a chain */
775                                 rxtop = skb;
776                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
777                                                    0, length);
778                         } else {
779                                 /* this is the middle of a chain */
780                                 skb_fill_page_desc(rxtop,
781                                                    skb_shinfo(rxtop)->nr_frags,
782                                                    buffer_info->page, 0,
783                                                    length);
784                                 /* re-use the skb, only consumed the page */
785                                 buffer_info->skb = skb;
786                         }
787                         e1000_consume_page(buffer_info, rxtop, length);
788                         goto next_desc;
789                 } else {
790                         if (rxtop) {
791                                 /* end of the chain */
792                                 skb_fill_page_desc(rxtop,
793                                     skb_shinfo(rxtop)->nr_frags,
794                                     buffer_info->page, 0, length);
795                                 /* re-use the current skb, we only consumed the
796                                  * page */
797                                 buffer_info->skb = skb;
798                                 skb = rxtop;
799                                 rxtop = NULL;
800                                 e1000_consume_page(buffer_info, skb, length);
801                         } else {
802                                 /* no chain, got EOP, this buf is the packet
803                                  * copybreak to save the put_page/alloc_page */
804                                 if (length <= copybreak &&
805                                     skb_tailroom(skb) >= length) {
806                                         u8 *vaddr;
807                                         vaddr = kmap_atomic(buffer_info->page,
808                                                            KM_SKB_DATA_SOFTIRQ);
809                                         memcpy(skb_tail_pointer(skb),
810                                                vaddr, length);
811                                         kunmap_atomic(vaddr,
812                                                       KM_SKB_DATA_SOFTIRQ);
813                                         /* re-use the page, so don't erase
814                                          * buffer_info->page */
815                                         skb_put(skb, length);
816                                 } else {
817                                         skb_fill_page_desc(skb, 0,
818                                                            buffer_info->page, 0,
819                                                            length);
820                                         e1000_consume_page(buffer_info, skb,
821                                                            length);
822                                 }
823                         }
824                 }
825
826                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
827                 e1000_rx_checksum(adapter,
828                                   (u32)(status) |
829                                   ((u32)(rx_desc->errors) << 24),
830                                   le16_to_cpu(rx_desc->csum), skb);
831
832                 pskb_trim(skb, skb->len - 4);
833
834                 /* probably a little skewed due to removing CRC */
835                 total_rx_bytes += skb->len;
836                 total_rx_packets++;
837
838                 /* eth type trans needs skb->data to point to something */
839                 if (!pskb_may_pull(skb, ETH_HLEN)) {
840                         ndev_err(netdev, "__pskb_pull_tail failed.\n");
841                         dev_kfree_skb(skb);
842                         goto next_desc;
843                 }
844
845                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
846
847 next_desc:
848                 rx_desc->status = 0;
849
850                 /* return some buffers to hardware, one at a time is too slow */
851                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
852                         adapter->alloc_rx_buf(adapter, cleaned_count);
853                         cleaned_count = 0;
854                 }
855
856                 /* use prefetched values */
857                 rx_desc = next_rxd;
858                 buffer_info = next_buffer;
859         }
860         rx_ring->next_to_clean = i;
861
862         cleaned_count = e1000_desc_unused(rx_ring);
863         if (cleaned_count)
864                 adapter->alloc_rx_buf(adapter, cleaned_count);
865
866         adapter->total_rx_packets += total_rx_packets;
867         adapter->total_rx_bytes += total_rx_bytes;
868         return cleaned;
869 }
870
871 /**
872  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
873  * @adapter: board private structure
874  *
875  * the return value indicates whether actual cleaning was done, there
876  * is no guarantee that everything was cleaned
877  **/
878 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
879                                   int *work_done, int work_to_do)
880 {
881         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
882         struct net_device *netdev = adapter->netdev;
883         struct pci_dev *pdev = adapter->pdev;
884         struct e1000_ring *rx_ring = adapter->rx_ring;
885         struct e1000_buffer *buffer_info, *next_buffer;
886         struct e1000_ps_page *ps_page;
887         struct sk_buff *skb;
888         unsigned int i, j;
889         u32 length, staterr;
890         int cleaned_count = 0;
891         bool cleaned = 0;
892         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
893
894         i = rx_ring->next_to_clean;
895         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
896         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
897         buffer_info = &rx_ring->buffer_info[i];
898
899         while (staterr & E1000_RXD_STAT_DD) {
900                 if (*work_done >= work_to_do)
901                         break;
902                 (*work_done)++;
903                 skb = buffer_info->skb;
904
905                 /* in the packet split case this is header only */
906                 prefetch(skb->data - NET_IP_ALIGN);
907
908                 i++;
909                 if (i == rx_ring->count)
910                         i = 0;
911                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
912                 prefetch(next_rxd);
913
914                 next_buffer = &rx_ring->buffer_info[i];
915
916                 cleaned = 1;
917                 cleaned_count++;
918                 pci_unmap_single(pdev, buffer_info->dma,
919                                  adapter->rx_ps_bsize0,
920                                  PCI_DMA_FROMDEVICE);
921                 buffer_info->dma = 0;
922
923                 if (!(staterr & E1000_RXD_STAT_EOP)) {
924                         ndev_dbg(netdev, "%s: Packet Split buffers didn't pick "
925                                  "up the full packet\n", netdev->name);
926                         dev_kfree_skb_irq(skb);
927                         goto next_desc;
928                 }
929
930                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
931                         dev_kfree_skb_irq(skb);
932                         goto next_desc;
933                 }
934
935                 length = le16_to_cpu(rx_desc->wb.middle.length0);
936
937                 if (!length) {
938                         ndev_dbg(netdev, "%s: Last part of the packet spanning"
939                                  " multiple descriptors\n", netdev->name);
940                         dev_kfree_skb_irq(skb);
941                         goto next_desc;
942                 }
943
944                 /* Good Receive */
945                 skb_put(skb, length);
946
947                 {
948                 /* this looks ugly, but it seems compiler issues make it
949                    more efficient than reusing j */
950                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
951
952                 /* page alloc/put takes too long and effects small packet
953                  * throughput, so unsplit small packets and save the alloc/put*/
954                 if (l1 && (l1 <= copybreak) &&
955                     ((length + l1) <= adapter->rx_ps_bsize0)) {
956                         u8 *vaddr;
957
958                         ps_page = &rx_ring->ps_pages[i * PS_PAGE_BUFFERS];
959
960                         /* there is no documentation about how to call
961                          * kmap_atomic, so we can't hold the mapping
962                          * very long */
963                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
964                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
965                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
966                         memcpy(skb_tail_pointer(skb), vaddr, l1);
967                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
968                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
969                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
970                         /* remove the CRC */
971                         l1 -= 4;
972                         skb_put(skb, l1);
973                         goto copydone;
974                 } /* if */
975                 }
976
977                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
978                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
979                         if (!length)
980                                 break;
981
982                         ps_page = &rx_ring->ps_pages[(i * PS_PAGE_BUFFERS) + j];
983                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
984                                        PCI_DMA_FROMDEVICE);
985                         ps_page->dma = 0;
986                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
987                         ps_page->page = NULL;
988                         skb->len += length;
989                         skb->data_len += length;
990                         skb->truesize += length;
991                 }
992
993                 /* strip the ethernet crc, problem is we're using pages now so
994                  * this whole operation can get a little cpu intensive */
995                 pskb_trim(skb, skb->len - 4);
996
997 copydone:
998                 total_rx_bytes += skb->len;
999                 total_rx_packets++;
1000
1001                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1002                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1003
1004                 if (rx_desc->wb.upper.header_status &
1005                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1006                         adapter->rx_hdr_split++;
1007
1008                 e1000_receive_skb(adapter, netdev, skb,
1009                                   staterr, rx_desc->wb.middle.vlan);
1010
1011 next_desc:
1012                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1013                 buffer_info->skb = NULL;
1014
1015                 /* return some buffers to hardware, one at a time is too slow */
1016                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1017                         adapter->alloc_rx_buf(adapter, cleaned_count);
1018                         cleaned_count = 0;
1019                 }
1020
1021                 /* use prefetched values */
1022                 rx_desc = next_rxd;
1023                 buffer_info = next_buffer;
1024
1025                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1026         }
1027         rx_ring->next_to_clean = i;
1028
1029         cleaned_count = e1000_desc_unused(rx_ring);
1030         if (cleaned_count)
1031                 adapter->alloc_rx_buf(adapter, cleaned_count);
1032
1033         adapter->total_rx_packets += total_rx_packets;
1034         adapter->total_rx_bytes += total_rx_bytes;
1035         return cleaned;
1036 }
1037
1038 /**
1039  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1040  * @adapter: board private structure
1041  **/
1042 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1043 {
1044         struct e1000_ring *rx_ring = adapter->rx_ring;
1045         struct e1000_buffer *buffer_info;
1046         struct e1000_ps_page *ps_page;
1047         struct pci_dev *pdev = adapter->pdev;
1048         unsigned long size;
1049         unsigned int i, j;
1050
1051         /* Free all the Rx ring sk_buffs */
1052         for (i = 0; i < rx_ring->count; i++) {
1053                 buffer_info = &rx_ring->buffer_info[i];
1054                 if (buffer_info->dma) {
1055                         if (adapter->clean_rx == e1000_clean_rx_irq)
1056                                 pci_unmap_single(pdev, buffer_info->dma,
1057                                                  adapter->rx_buffer_len,
1058                                                  PCI_DMA_FROMDEVICE);
1059                         else if (adapter->clean_rx == e1000_clean_rx_irq_jumbo)
1060                                 pci_unmap_page(pdev, buffer_info->dma,
1061                                                PAGE_SIZE, PCI_DMA_FROMDEVICE);
1062                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1063                                 pci_unmap_single(pdev, buffer_info->dma,
1064                                                  adapter->rx_ps_bsize0,
1065                                                  PCI_DMA_FROMDEVICE);
1066                         buffer_info->dma = 0;
1067                 }
1068
1069                 if (buffer_info->page) {
1070                         put_page(buffer_info->page);
1071                         buffer_info->page = NULL;
1072                 }
1073
1074                 if (buffer_info->skb) {
1075                         dev_kfree_skb(buffer_info->skb);
1076                         buffer_info->skb = NULL;
1077                 }
1078
1079                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1080                         ps_page = &rx_ring->ps_pages[(i * PS_PAGE_BUFFERS)
1081                                                      + j];
1082                         if (!ps_page->page)
1083                                 break;
1084                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1085                                        PCI_DMA_FROMDEVICE);
1086                         ps_page->dma = 0;
1087                         put_page(ps_page->page);
1088                         ps_page->page = NULL;
1089                 }
1090         }
1091
1092         /* there also may be some cached data from a chained receive */
1093         if (rx_ring->rx_skb_top) {
1094                 dev_kfree_skb(rx_ring->rx_skb_top);
1095                 rx_ring->rx_skb_top = NULL;
1096         }
1097
1098         size = sizeof(struct e1000_buffer) * rx_ring->count;
1099         memset(rx_ring->buffer_info, 0, size);
1100         size = sizeof(struct e1000_ps_page)
1101                * (rx_ring->count * PS_PAGE_BUFFERS);
1102         memset(rx_ring->ps_pages, 0, size);
1103
1104         /* Zero out the descriptor ring */
1105         memset(rx_ring->desc, 0, rx_ring->size);
1106
1107         rx_ring->next_to_clean = 0;
1108         rx_ring->next_to_use = 0;
1109
1110         writel(0, adapter->hw.hw_addr + rx_ring->head);
1111         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1112 }
1113
1114 /**
1115  * e1000_intr_msi - Interrupt Handler
1116  * @irq: interrupt number
1117  * @data: pointer to a network interface device structure
1118  **/
1119 static irqreturn_t e1000_intr_msi(int irq, void *data)
1120 {
1121         struct net_device *netdev = data;
1122         struct e1000_adapter *adapter = netdev_priv(netdev);
1123         struct e1000_hw *hw = &adapter->hw;
1124         u32 icr = er32(ICR);
1125
1126         /* read ICR disables interrupts using IAM, so keep up with our
1127          * enable/disable accounting */
1128         atomic_inc(&adapter->irq_sem);
1129
1130         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1131                 hw->mac.get_link_status = 1;
1132                 /* ICH8 workaround-- Call gig speed drop workaround on cable
1133                  * disconnect (LSC) before accessing any PHY registers */
1134                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1135                     (!(er32(STATUS) & E1000_STATUS_LU)))
1136                         e1000e_gig_downshift_workaround_ich8lan(hw);
1137
1138                 /* 80003ES2LAN workaround-- For packet buffer work-around on
1139                  * link down event; disable receives here in the ISR and reset
1140                  * adapter in watchdog */
1141                 if (netif_carrier_ok(netdev) &&
1142                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1143                         /* disable receives */
1144                         u32 rctl = er32(RCTL);
1145                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1146                 }
1147                 /* guard against interrupt when we're going down */
1148                 if (!test_bit(__E1000_DOWN, &adapter->state))
1149                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1150         }
1151
1152         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1153                 adapter->total_tx_bytes = 0;
1154                 adapter->total_tx_packets = 0;
1155                 adapter->total_rx_bytes = 0;
1156                 adapter->total_rx_packets = 0;
1157                 __netif_rx_schedule(netdev, &adapter->napi);
1158         } else {
1159                 atomic_dec(&adapter->irq_sem);
1160         }
1161
1162         return IRQ_HANDLED;
1163 }
1164
1165 /**
1166  * e1000_intr - Interrupt Handler
1167  * @irq: interrupt number
1168  * @data: pointer to a network interface device structure
1169  **/
1170 static irqreturn_t e1000_intr(int irq, void *data)
1171 {
1172         struct net_device *netdev = data;
1173         struct e1000_adapter *adapter = netdev_priv(netdev);
1174         struct e1000_hw *hw = &adapter->hw;
1175
1176         u32 rctl, icr = er32(ICR);
1177         if (!icr)
1178                 return IRQ_NONE;  /* Not our interrupt */
1179
1180         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1181          * not set, then the adapter didn't send an interrupt */
1182         if (!(icr & E1000_ICR_INT_ASSERTED))
1183                 return IRQ_NONE;
1184
1185         /* Interrupt Auto-Mask...upon reading ICR,
1186          * interrupts are masked.  No need for the
1187          * IMC write, but it does mean we should
1188          * account for it ASAP. */
1189         atomic_inc(&adapter->irq_sem);
1190
1191         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1192                 hw->mac.get_link_status = 1;
1193                 /* ICH8 workaround-- Call gig speed drop workaround on cable
1194                  * disconnect (LSC) before accessing any PHY registers */
1195                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1196                     (!(er32(STATUS) & E1000_STATUS_LU)))
1197                         e1000e_gig_downshift_workaround_ich8lan(hw);
1198
1199                 /* 80003ES2LAN workaround--
1200                  * For packet buffer work-around on link down event;
1201                  * disable receives here in the ISR and
1202                  * reset adapter in watchdog
1203                  */
1204                 if (netif_carrier_ok(netdev) &&
1205                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1206                         /* disable receives */
1207                         rctl = er32(RCTL);
1208                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1209                 }
1210                 /* guard against interrupt when we're going down */
1211                 if (!test_bit(__E1000_DOWN, &adapter->state))
1212                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1213         }
1214
1215         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1216                 adapter->total_tx_bytes = 0;
1217                 adapter->total_tx_packets = 0;
1218                 adapter->total_rx_bytes = 0;
1219                 adapter->total_rx_packets = 0;
1220                 __netif_rx_schedule(netdev, &adapter->napi);
1221         } else {
1222                 atomic_dec(&adapter->irq_sem);
1223         }
1224
1225         return IRQ_HANDLED;
1226 }
1227
1228 static int e1000_request_irq(struct e1000_adapter *adapter)
1229 {
1230         struct net_device *netdev = adapter->netdev;
1231         void (*handler) = &e1000_intr;
1232         int irq_flags = IRQF_SHARED;
1233         int err;
1234
1235         err = pci_enable_msi(adapter->pdev);
1236         if (err) {
1237                 ndev_warn(netdev,
1238                  "Unable to allocate MSI interrupt Error: %d\n", err);
1239         } else {
1240                 adapter->flags |= FLAG_MSI_ENABLED;
1241                 handler = &e1000_intr_msi;
1242                 irq_flags = 0;
1243         }
1244
1245         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
1246                           netdev);
1247         if (err) {
1248                 if (adapter->flags & FLAG_MSI_ENABLED)
1249                         pci_disable_msi(adapter->pdev);
1250                 ndev_err(netdev,
1251                        "Unable to allocate interrupt Error: %d\n", err);
1252         }
1253
1254         return err;
1255 }
1256
1257 static void e1000_free_irq(struct e1000_adapter *adapter)
1258 {
1259         struct net_device *netdev = adapter->netdev;
1260
1261         free_irq(adapter->pdev->irq, netdev);
1262         if (adapter->flags & FLAG_MSI_ENABLED) {
1263                 pci_disable_msi(adapter->pdev);
1264                 adapter->flags &= ~FLAG_MSI_ENABLED;
1265         }
1266 }
1267
1268 /**
1269  * e1000_irq_disable - Mask off interrupt generation on the NIC
1270  **/
1271 static void e1000_irq_disable(struct e1000_adapter *adapter)
1272 {
1273         struct e1000_hw *hw = &adapter->hw;
1274
1275         atomic_inc(&adapter->irq_sem);
1276         ew32(IMC, ~0);
1277         e1e_flush();
1278         synchronize_irq(adapter->pdev->irq);
1279 }
1280
1281 /**
1282  * e1000_irq_enable - Enable default interrupt generation settings
1283  **/
1284 static void e1000_irq_enable(struct e1000_adapter *adapter)
1285 {
1286         struct e1000_hw *hw = &adapter->hw;
1287
1288         if (atomic_dec_and_test(&adapter->irq_sem)) {
1289                 ew32(IMS, IMS_ENABLE_MASK);
1290                 e1e_flush();
1291         }
1292 }
1293
1294 /**
1295  * e1000_get_hw_control - get control of the h/w from f/w
1296  * @adapter: address of board private structure
1297  *
1298  * e1000_get_hw_control sets {CTRL_EXT|FWSM}:DRV_LOAD bit.
1299  * For ASF and Pass Through versions of f/w this means that
1300  * the driver is loaded. For AMT version (only with 82573)
1301  * of the f/w this means that the network i/f is open.
1302  **/
1303 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1304 {
1305         struct e1000_hw *hw = &adapter->hw;
1306         u32 ctrl_ext;
1307         u32 swsm;
1308
1309         /* Let firmware know the driver has taken over */
1310         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1311                 swsm = er32(SWSM);
1312                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1313         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1314                 ctrl_ext = er32(CTRL_EXT);
1315                 ew32(CTRL_EXT,
1316                                 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1317         }
1318 }
1319
1320 /**
1321  * e1000_release_hw_control - release control of the h/w to f/w
1322  * @adapter: address of board private structure
1323  *
1324  * e1000_release_hw_control resets {CTRL_EXT|FWSM}:DRV_LOAD bit.
1325  * For ASF and Pass Through versions of f/w this means that the
1326  * driver is no longer loaded. For AMT version (only with 82573) i
1327  * of the f/w this means that the network i/f is closed.
1328  *
1329  **/
1330 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1331 {
1332         struct e1000_hw *hw = &adapter->hw;
1333         u32 ctrl_ext;
1334         u32 swsm;
1335
1336         /* Let firmware taken over control of h/w */
1337         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1338                 swsm = er32(SWSM);
1339                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1340         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1341                 ctrl_ext = er32(CTRL_EXT);
1342                 ew32(CTRL_EXT,
1343                                 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1344         }
1345 }
1346
1347 static void e1000_release_manageability(struct e1000_adapter *adapter)
1348 {
1349         if (adapter->flags & FLAG_MNG_PT_ENABLED) {
1350                 struct e1000_hw *hw = &adapter->hw;
1351
1352                 u32 manc = er32(MANC);
1353
1354                 /* re-enable hardware interception of ARP */
1355                 manc |= E1000_MANC_ARP_EN;
1356                 manc &= ~E1000_MANC_EN_MNG2HOST;
1357
1358                 /* don't explicitly have to mess with MANC2H since
1359                  * MANC has an enable disable that gates MANC2H */
1360                 ew32(MANC, manc);
1361         }
1362 }
1363
1364 /**
1365  * @e1000_alloc_ring - allocate memory for a ring structure
1366  **/
1367 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1368                                 struct e1000_ring *ring)
1369 {
1370         struct pci_dev *pdev = adapter->pdev;
1371
1372         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1373                                         GFP_KERNEL);
1374         if (!ring->desc)
1375                 return -ENOMEM;
1376
1377         return 0;
1378 }
1379
1380 /**
1381  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1382  * @adapter: board private structure
1383  *
1384  * Return 0 on success, negative on failure
1385  **/
1386 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1387 {
1388         struct e1000_ring *tx_ring = adapter->tx_ring;
1389         int err = -ENOMEM, size;
1390
1391         size = sizeof(struct e1000_buffer) * tx_ring->count;
1392         tx_ring->buffer_info = vmalloc(size);
1393         if (!tx_ring->buffer_info)
1394                 goto err;
1395         memset(tx_ring->buffer_info, 0, size);
1396
1397         /* round up to nearest 4K */
1398         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1399         tx_ring->size = ALIGN(tx_ring->size, 4096);
1400
1401         err = e1000_alloc_ring_dma(adapter, tx_ring);
1402         if (err)
1403                 goto err;
1404
1405         tx_ring->next_to_use = 0;
1406         tx_ring->next_to_clean = 0;
1407         spin_lock_init(&adapter->tx_queue_lock);
1408
1409         return 0;
1410 err:
1411         vfree(tx_ring->buffer_info);
1412         ndev_err(adapter->netdev,
1413         "Unable to allocate memory for the transmit descriptor ring\n");
1414         return err;
1415 }
1416
1417 /**
1418  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1419  * @adapter: board private structure
1420  *
1421  * Returns 0 on success, negative on failure
1422  **/
1423 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1424 {
1425         struct e1000_ring *rx_ring = adapter->rx_ring;
1426         int size, desc_len, err = -ENOMEM;
1427
1428         size = sizeof(struct e1000_buffer) * rx_ring->count;
1429         rx_ring->buffer_info = vmalloc(size);
1430         if (!rx_ring->buffer_info)
1431                 goto err;
1432         memset(rx_ring->buffer_info, 0, size);
1433
1434         rx_ring->ps_pages = kcalloc(rx_ring->count * PS_PAGE_BUFFERS,
1435                                     sizeof(struct e1000_ps_page),
1436                                     GFP_KERNEL);
1437         if (!rx_ring->ps_pages)
1438                 goto err;
1439
1440         desc_len = sizeof(union e1000_rx_desc_packet_split);
1441
1442         /* Round up to nearest 4K */
1443         rx_ring->size = rx_ring->count * desc_len;
1444         rx_ring->size = ALIGN(rx_ring->size, 4096);
1445
1446         err = e1000_alloc_ring_dma(adapter, rx_ring);
1447         if (err)
1448                 goto err;
1449
1450         rx_ring->next_to_clean = 0;
1451         rx_ring->next_to_use = 0;
1452         rx_ring->rx_skb_top = NULL;
1453
1454         return 0;
1455 err:
1456         vfree(rx_ring->buffer_info);
1457         kfree(rx_ring->ps_pages);
1458         ndev_err(adapter->netdev,
1459         "Unable to allocate memory for the transmit descriptor ring\n");
1460         return err;
1461 }
1462
1463 /**
1464  * e1000_clean_tx_ring - Free Tx Buffers
1465  * @adapter: board private structure
1466  **/
1467 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1468 {
1469         struct e1000_ring *tx_ring = adapter->tx_ring;
1470         struct e1000_buffer *buffer_info;
1471         unsigned long size;
1472         unsigned int i;
1473
1474         for (i = 0; i < tx_ring->count; i++) {
1475                 buffer_info = &tx_ring->buffer_info[i];
1476                 e1000_put_txbuf(adapter, buffer_info);
1477         }
1478
1479         size = sizeof(struct e1000_buffer) * tx_ring->count;
1480         memset(tx_ring->buffer_info, 0, size);
1481
1482         memset(tx_ring->desc, 0, tx_ring->size);
1483
1484         tx_ring->next_to_use = 0;
1485         tx_ring->next_to_clean = 0;
1486
1487         writel(0, adapter->hw.hw_addr + tx_ring->head);
1488         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1489 }
1490
1491 /**
1492  * e1000e_free_tx_resources - Free Tx Resources per Queue
1493  * @adapter: board private structure
1494  *
1495  * Free all transmit software resources
1496  **/
1497 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1498 {
1499         struct pci_dev *pdev = adapter->pdev;
1500         struct e1000_ring *tx_ring = adapter->tx_ring;
1501
1502         e1000_clean_tx_ring(adapter);
1503
1504         vfree(tx_ring->buffer_info);
1505         tx_ring->buffer_info = NULL;
1506
1507         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1508                           tx_ring->dma);
1509         tx_ring->desc = NULL;
1510 }
1511
1512 /**
1513  * e1000e_free_rx_resources - Free Rx Resources
1514  * @adapter: board private structure
1515  *
1516  * Free all receive software resources
1517  **/
1518
1519 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1520 {
1521         struct pci_dev *pdev = adapter->pdev;
1522         struct e1000_ring *rx_ring = adapter->rx_ring;
1523
1524         e1000_clean_rx_ring(adapter);
1525
1526         vfree(rx_ring->buffer_info);
1527         rx_ring->buffer_info = NULL;
1528
1529         kfree(rx_ring->ps_pages);
1530         rx_ring->ps_pages = NULL;
1531
1532         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1533                           rx_ring->dma);
1534         rx_ring->desc = NULL;
1535 }
1536
1537 /**
1538  * e1000_update_itr - update the dynamic ITR value based on statistics
1539  *      Stores a new ITR value based on packets and byte
1540  *      counts during the last interrupt.  The advantage of per interrupt
1541  *      computation is faster updates and more accurate ITR for the current
1542  *      traffic pattern.  Constants in this function were computed
1543  *      based on theoretical maximum wire speed and thresholds were set based
1544  *      on testing data as well as attempting to minimize response time
1545  *      while increasing bulk throughput.
1546  *      this functionality is controlled by the InterruptThrottleRate module
1547  *      parameter (see e1000_param.c)
1548  * @adapter: pointer to adapter
1549  * @itr_setting: current adapter->itr
1550  * @packets: the number of packets during this measurement interval
1551  * @bytes: the number of bytes during this measurement interval
1552  **/
1553 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1554                                      u16 itr_setting, int packets,
1555                                      int bytes)
1556 {
1557         unsigned int retval = itr_setting;
1558
1559         if (packets == 0)
1560                 goto update_itr_done;
1561
1562         switch (itr_setting) {
1563         case lowest_latency:
1564                 /* handle TSO and jumbo frames */
1565                 if (bytes/packets > 8000)
1566                         retval = bulk_latency;
1567                 else if ((packets < 5) && (bytes > 512)) {
1568                         retval = low_latency;
1569                 }
1570                 break;
1571         case low_latency:  /* 50 usec aka 20000 ints/s */
1572                 if (bytes > 10000) {
1573                         /* this if handles the TSO accounting */
1574                         if (bytes/packets > 8000) {
1575                                 retval = bulk_latency;
1576                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1577                                 retval = bulk_latency;
1578                         } else if ((packets > 35)) {
1579                                 retval = lowest_latency;
1580                         }
1581                 } else if (bytes/packets > 2000) {
1582                         retval = bulk_latency;
1583                 } else if (packets <= 2 && bytes < 512) {
1584                         retval = lowest_latency;
1585                 }
1586                 break;
1587         case bulk_latency: /* 250 usec aka 4000 ints/s */
1588                 if (bytes > 25000) {
1589                         if (packets > 35) {
1590                                 retval = low_latency;
1591                         }
1592                 } else if (bytes < 6000) {
1593                         retval = low_latency;
1594                 }
1595                 break;
1596         }
1597
1598 update_itr_done:
1599         return retval;
1600 }
1601
1602 static void e1000_set_itr(struct e1000_adapter *adapter)
1603 {
1604         struct e1000_hw *hw = &adapter->hw;
1605         u16 current_itr;
1606         u32 new_itr = adapter->itr;
1607
1608         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1609         if (adapter->link_speed != SPEED_1000) {
1610                 current_itr = 0;
1611                 new_itr = 4000;
1612                 goto set_itr_now;
1613         }
1614
1615         adapter->tx_itr = e1000_update_itr(adapter,
1616                                     adapter->tx_itr,
1617                                     adapter->total_tx_packets,
1618                                     adapter->total_tx_bytes);
1619         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1620         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1621                 adapter->tx_itr = low_latency;
1622
1623         adapter->rx_itr = e1000_update_itr(adapter,
1624                                     adapter->rx_itr,
1625                                     adapter->total_rx_packets,
1626                                     adapter->total_rx_bytes);
1627         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1628         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1629                 adapter->rx_itr = low_latency;
1630
1631         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1632
1633         switch (current_itr) {
1634         /* counts and packets in update_itr are dependent on these numbers */
1635         case lowest_latency:
1636                 new_itr = 70000;
1637                 break;
1638         case low_latency:
1639                 new_itr = 20000; /* aka hwitr = ~200 */
1640                 break;
1641         case bulk_latency:
1642                 new_itr = 4000;
1643                 break;
1644         default:
1645                 break;
1646         }
1647
1648 set_itr_now:
1649         if (new_itr != adapter->itr) {
1650                 /* this attempts to bias the interrupt rate towards Bulk
1651                  * by adding intermediate steps when interrupt rate is
1652                  * increasing */
1653                 new_itr = new_itr > adapter->itr ?
1654                              min(adapter->itr + (new_itr >> 2), new_itr) :
1655                              new_itr;
1656                 adapter->itr = new_itr;
1657                 ew32(ITR, 1000000000 / (new_itr * 256));
1658         }
1659 }
1660
1661 /**
1662  * e1000_clean - NAPI Rx polling callback
1663  * @adapter: board private structure
1664  **/
1665 static int e1000_clean(struct napi_struct *napi, int budget)
1666 {
1667         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1668         struct net_device *poll_dev = adapter->netdev;
1669         int tx_cleaned = 0, work_done = 0;
1670
1671         /* Must NOT use netdev_priv macro here. */
1672         adapter = poll_dev->priv;
1673
1674         /* Keep link state information with original netdev */
1675         if (!netif_carrier_ok(poll_dev))
1676                 goto quit_polling;
1677
1678         /* e1000_clean is called per-cpu.  This lock protects
1679          * tx_ring from being cleaned by multiple cpus
1680          * simultaneously.  A failure obtaining the lock means
1681          * tx_ring is currently being cleaned anyway. */
1682         if (spin_trylock(&adapter->tx_queue_lock)) {
1683                 tx_cleaned = e1000_clean_tx_irq(adapter);
1684                 spin_unlock(&adapter->tx_queue_lock);
1685         }
1686
1687         adapter->clean_rx(adapter, &work_done, budget);
1688
1689         /* If no Tx and not enough Rx work done, exit the polling mode */
1690         if ((!tx_cleaned && (work_done < budget)) ||
1691            !netif_running(poll_dev)) {
1692 quit_polling:
1693                 if (adapter->itr_setting & 3)
1694                         e1000_set_itr(adapter);
1695                 netif_rx_complete(poll_dev, napi);
1696                 e1000_irq_enable(adapter);
1697         }
1698
1699         return work_done;
1700 }
1701
1702 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1703 {
1704         struct e1000_adapter *adapter = netdev_priv(netdev);
1705         struct e1000_hw *hw = &adapter->hw;
1706         u32 vfta, index;
1707
1708         /* don't update vlan cookie if already programmed */
1709         if ((adapter->hw.mng_cookie.status &
1710              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1711             (vid == adapter->mng_vlan_id))
1712                 return;
1713         /* add VID to filter table */
1714         index = (vid >> 5) & 0x7F;
1715         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1716         vfta |= (1 << (vid & 0x1F));
1717         e1000e_write_vfta(hw, index, vfta);
1718 }
1719
1720 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1721 {
1722         struct e1000_adapter *adapter = netdev_priv(netdev);
1723         struct e1000_hw *hw = &adapter->hw;
1724         u32 vfta, index;
1725
1726         e1000_irq_disable(adapter);
1727         vlan_group_set_device(adapter->vlgrp, vid, NULL);
1728         e1000_irq_enable(adapter);
1729
1730         if ((adapter->hw.mng_cookie.status &
1731              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1732             (vid == adapter->mng_vlan_id)) {
1733                 /* release control to f/w */
1734                 e1000_release_hw_control(adapter);
1735                 return;
1736         }
1737
1738         /* remove VID from filter table */
1739         index = (vid >> 5) & 0x7F;
1740         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1741         vfta &= ~(1 << (vid & 0x1F));
1742         e1000e_write_vfta(hw, index, vfta);
1743 }
1744
1745 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
1746 {
1747         struct net_device *netdev = adapter->netdev;
1748         u16 vid = adapter->hw.mng_cookie.vlan_id;
1749         u16 old_vid = adapter->mng_vlan_id;
1750
1751         if (!adapter->vlgrp)
1752                 return;
1753
1754         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
1755                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1756                 if (adapter->hw.mng_cookie.status &
1757                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1758                         e1000_vlan_rx_add_vid(netdev, vid);
1759                         adapter->mng_vlan_id = vid;
1760                 }
1761
1762                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
1763                                 (vid != old_vid) &&
1764                     !vlan_group_get_device(adapter->vlgrp, old_vid))
1765                         e1000_vlan_rx_kill_vid(netdev, old_vid);
1766         } else {
1767                 adapter->mng_vlan_id = vid;
1768         }
1769 }
1770
1771
1772 static void e1000_vlan_rx_register(struct net_device *netdev,
1773                                    struct vlan_group *grp)
1774 {
1775         struct e1000_adapter *adapter = netdev_priv(netdev);
1776         struct e1000_hw *hw = &adapter->hw;
1777         u32 ctrl, rctl;
1778
1779         e1000_irq_disable(adapter);
1780         adapter->vlgrp = grp;
1781
1782         if (grp) {
1783                 /* enable VLAN tag insert/strip */
1784                 ctrl = er32(CTRL);
1785                 ctrl |= E1000_CTRL_VME;
1786                 ew32(CTRL, ctrl);
1787
1788                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1789                         /* enable VLAN receive filtering */
1790                         rctl = er32(RCTL);
1791                         rctl |= E1000_RCTL_VFE;
1792                         rctl &= ~E1000_RCTL_CFIEN;
1793                         ew32(RCTL, rctl);
1794                         e1000_update_mng_vlan(adapter);
1795                 }
1796         } else {
1797                 /* disable VLAN tag insert/strip */
1798                 ctrl = er32(CTRL);
1799                 ctrl &= ~E1000_CTRL_VME;
1800                 ew32(CTRL, ctrl);
1801
1802                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1803                         /* disable VLAN filtering */
1804                         rctl = er32(RCTL);
1805                         rctl &= ~E1000_RCTL_VFE;
1806                         ew32(RCTL, rctl);
1807                         if (adapter->mng_vlan_id !=
1808                             (u16)E1000_MNG_VLAN_NONE) {
1809                                 e1000_vlan_rx_kill_vid(netdev,
1810                                                        adapter->mng_vlan_id);
1811                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1812                         }
1813                 }
1814         }
1815
1816         e1000_irq_enable(adapter);
1817 }
1818
1819 static void e1000_restore_vlan(struct e1000_adapter *adapter)
1820 {
1821         u16 vid;
1822
1823         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
1824
1825         if (!adapter->vlgrp)
1826                 return;
1827
1828         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1829                 if (!vlan_group_get_device(adapter->vlgrp, vid))
1830                         continue;
1831                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
1832         }
1833 }
1834
1835 static void e1000_init_manageability(struct e1000_adapter *adapter)
1836 {
1837         struct e1000_hw *hw = &adapter->hw;
1838         u32 manc, manc2h;
1839
1840         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
1841                 return;
1842
1843         manc = er32(MANC);
1844
1845         /* disable hardware interception of ARP */
1846         manc &= ~(E1000_MANC_ARP_EN);
1847
1848         /* enable receiving management packets to the host. this will probably
1849          * generate destination unreachable messages from the host OS, but
1850          * the packets will be handled on SMBUS */
1851         manc |= E1000_MANC_EN_MNG2HOST;
1852         manc2h = er32(MANC2H);
1853 #define E1000_MNG2HOST_PORT_623 (1 << 5)
1854 #define E1000_MNG2HOST_PORT_664 (1 << 6)
1855         manc2h |= E1000_MNG2HOST_PORT_623;
1856         manc2h |= E1000_MNG2HOST_PORT_664;
1857         ew32(MANC2H, manc2h);
1858         ew32(MANC, manc);
1859 }
1860
1861 /**
1862  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1863  * @adapter: board private structure
1864  *
1865  * Configure the Tx unit of the MAC after a reset.
1866  **/
1867 static void e1000_configure_tx(struct e1000_adapter *adapter)
1868 {
1869         struct e1000_hw *hw = &adapter->hw;
1870         struct e1000_ring *tx_ring = adapter->tx_ring;
1871         u64 tdba;
1872         u32 tdlen, tctl, tipg, tarc;
1873         u32 ipgr1, ipgr2;
1874
1875         /* Setup the HW Tx Head and Tail descriptor pointers */
1876         tdba = tx_ring->dma;
1877         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
1878         ew32(TDBAL, (tdba & DMA_32BIT_MASK));
1879         ew32(TDBAH, (tdba >> 32));
1880         ew32(TDLEN, tdlen);
1881         ew32(TDH, 0);
1882         ew32(TDT, 0);
1883         tx_ring->head = E1000_TDH;
1884         tx_ring->tail = E1000_TDT;
1885
1886         /* Set the default values for the Tx Inter Packet Gap timer */
1887         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
1888         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
1889         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
1890
1891         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
1892                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
1893
1894         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1895         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1896         ew32(TIPG, tipg);
1897
1898         /* Set the Tx Interrupt Delay register */
1899         ew32(TIDV, adapter->tx_int_delay);
1900         /* tx irq moderation */
1901         ew32(TADV, adapter->tx_abs_int_delay);
1902
1903         /* Program the Transmit Control Register */
1904         tctl = er32(TCTL);
1905         tctl &= ~E1000_TCTL_CT;
1906         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1907                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1908
1909         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
1910                 tarc = er32(TARC0);
1911                 /* set the speed mode bit, we'll clear it if we're not at
1912                  * gigabit link later */
1913 #define SPEED_MODE_BIT (1 << 21)
1914                 tarc |= SPEED_MODE_BIT;
1915                 ew32(TARC0, tarc);
1916         }
1917
1918         /* errata: program both queues to unweighted RR */
1919         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
1920                 tarc = er32(TARC0);
1921                 tarc |= 1;
1922                 ew32(TARC0, tarc);
1923                 tarc = er32(TARC1);
1924                 tarc |= 1;
1925                 ew32(TARC1, tarc);
1926         }
1927
1928         e1000e_config_collision_dist(hw);
1929
1930         /* Setup Transmit Descriptor Settings for eop descriptor */
1931         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1932
1933         /* only set IDE if we are delaying interrupts using the timers */
1934         if (adapter->tx_int_delay)
1935                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1936
1937         /* enable Report Status bit */
1938         adapter->txd_cmd |= E1000_TXD_CMD_RS;
1939
1940         ew32(TCTL, tctl);
1941
1942         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
1943 }
1944
1945 /**
1946  * e1000_setup_rctl - configure the receive control registers
1947  * @adapter: Board private structure
1948  **/
1949 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1950                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1951 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1952 {
1953         struct e1000_hw *hw = &adapter->hw;
1954         u32 rctl, rfctl;
1955         u32 psrctl = 0;
1956         u32 pages = 0;
1957
1958         /* Program MC offset vector base */
1959         rctl = er32(RCTL);
1960         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1961         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1962                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1963                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1964
1965         /* Do not Store bad packets */
1966         rctl &= ~E1000_RCTL_SBP;
1967
1968         /* Enable Long Packet receive */
1969         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1970                 rctl &= ~E1000_RCTL_LPE;
1971         else
1972                 rctl |= E1000_RCTL_LPE;
1973
1974         /* Setup buffer sizes */
1975         rctl &= ~E1000_RCTL_SZ_4096;
1976         rctl |= E1000_RCTL_BSEX;
1977         switch (adapter->rx_buffer_len) {
1978         case 256:
1979                 rctl |= E1000_RCTL_SZ_256;
1980                 rctl &= ~E1000_RCTL_BSEX;
1981                 break;
1982         case 512:
1983                 rctl |= E1000_RCTL_SZ_512;
1984                 rctl &= ~E1000_RCTL_BSEX;
1985                 break;
1986         case 1024:
1987                 rctl |= E1000_RCTL_SZ_1024;
1988                 rctl &= ~E1000_RCTL_BSEX;
1989                 break;
1990         case 2048:
1991         default:
1992                 rctl |= E1000_RCTL_SZ_2048;
1993                 rctl &= ~E1000_RCTL_BSEX;
1994                 break;
1995         case 4096:
1996                 rctl |= E1000_RCTL_SZ_4096;
1997                 break;
1998         case 8192:
1999                 rctl |= E1000_RCTL_SZ_8192;
2000                 break;
2001         case 16384:
2002                 rctl |= E1000_RCTL_SZ_16384;
2003                 break;
2004         }
2005
2006         /*
2007          * 82571 and greater support packet-split where the protocol
2008          * header is placed in skb->data and the packet data is
2009          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2010          * In the case of a non-split, skb->data is linearly filled,
2011          * followed by the page buffers.  Therefore, skb->data is
2012          * sized to hold the largest protocol header.
2013          *
2014          * allocations using alloc_page take too long for regular MTU
2015          * so only enable packet split for jumbo frames
2016          *
2017          * Using pages when the page size is greater than 16k wastes
2018          * a lot of memory, since we allocate 3 pages at all times
2019          * per packet.
2020          */
2021         adapter->rx_ps_pages = 0;
2022         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2023         if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2024                 adapter->rx_ps_pages = pages;
2025
2026         if (adapter->rx_ps_pages) {
2027                 /* Configure extra packet-split registers */
2028                 rfctl = er32(RFCTL);
2029                 rfctl |= E1000_RFCTL_EXTEN;
2030                 /* disable packet split support for IPv6 extension headers,
2031                  * because some malformed IPv6 headers can hang the RX */
2032                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2033                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2034
2035                 ew32(RFCTL, rfctl);
2036
2037                 /* disable the stripping of CRC because it breaks
2038                  * BMC firmware connected over SMBUS */
2039                 rctl |= E1000_RCTL_DTYP_PS /* | E1000_RCTL_SECRC */;
2040
2041                 psrctl |= adapter->rx_ps_bsize0 >>
2042                         E1000_PSRCTL_BSIZE0_SHIFT;
2043
2044                 switch (adapter->rx_ps_pages) {
2045                 case 3:
2046                         psrctl |= PAGE_SIZE <<
2047                                 E1000_PSRCTL_BSIZE3_SHIFT;
2048                 case 2:
2049                         psrctl |= PAGE_SIZE <<
2050                                 E1000_PSRCTL_BSIZE2_SHIFT;
2051                 case 1:
2052                         psrctl |= PAGE_SIZE >>
2053                                 E1000_PSRCTL_BSIZE1_SHIFT;
2054                         break;
2055                 }
2056
2057                 ew32(PSRCTL, psrctl);
2058         }
2059
2060         ew32(RCTL, rctl);
2061 }
2062
2063 /**
2064  * e1000_configure_rx - Configure Receive Unit after Reset
2065  * @adapter: board private structure
2066  *
2067  * Configure the Rx unit of the MAC after a reset.
2068  **/
2069 static void e1000_configure_rx(struct e1000_adapter *adapter)
2070 {
2071         struct e1000_hw *hw = &adapter->hw;
2072         struct e1000_ring *rx_ring = adapter->rx_ring;
2073         u64 rdba;
2074         u32 rdlen, rctl, rxcsum, ctrl_ext;
2075
2076         if (adapter->rx_ps_pages) {
2077                 /* this is a 32 byte descriptor */
2078                 rdlen = rx_ring->count *
2079                         sizeof(union e1000_rx_desc_packet_split);
2080                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2081                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2082         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + VLAN_HLEN + 4) {
2083                 rdlen = rx_ring->count *
2084                         sizeof(struct e1000_rx_desc);
2085                 adapter->clean_rx = e1000_clean_rx_irq_jumbo;
2086                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_jumbo;
2087         } else {
2088                 rdlen = rx_ring->count *
2089                         sizeof(struct e1000_rx_desc);
2090                 adapter->clean_rx = e1000_clean_rx_irq;
2091                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2092         }
2093
2094         /* disable receives while setting up the descriptors */
2095         rctl = er32(RCTL);
2096         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2097         e1e_flush();
2098         msleep(10);
2099
2100         /* set the Receive Delay Timer Register */
2101         ew32(RDTR, adapter->rx_int_delay);
2102
2103         /* irq moderation */
2104         ew32(RADV, adapter->rx_abs_int_delay);
2105         if (adapter->itr_setting != 0)
2106                 ew32(ITR,
2107                         1000000000 / (adapter->itr * 256));
2108
2109         ctrl_ext = er32(CTRL_EXT);
2110         /* Reset delay timers after every interrupt */
2111         ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2112         /* Auto-Mask interrupts upon ICR access */
2113         ctrl_ext |= E1000_CTRL_EXT_IAME;
2114         ew32(IAM, 0xffffffff);
2115         ew32(CTRL_EXT, ctrl_ext);
2116         e1e_flush();
2117
2118         /* Setup the HW Rx Head and Tail Descriptor Pointers and
2119          * the Base and Length of the Rx Descriptor Ring */
2120         rdba = rx_ring->dma;
2121         ew32(RDBAL, (rdba & DMA_32BIT_MASK));
2122         ew32(RDBAH, (rdba >> 32));
2123         ew32(RDLEN, rdlen);
2124         ew32(RDH, 0);
2125         ew32(RDT, 0);
2126         rx_ring->head = E1000_RDH;
2127         rx_ring->tail = E1000_RDT;
2128
2129         /* Enable Receive Checksum Offload for TCP and UDP */
2130         rxcsum = er32(RXCSUM);
2131         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2132                 rxcsum |= E1000_RXCSUM_TUOFL;
2133
2134                 /* IPv4 payload checksum for UDP fragments must be
2135                  * used in conjunction with packet-split. */
2136                 if (adapter->rx_ps_pages)
2137                         rxcsum |= E1000_RXCSUM_IPPCSE;
2138         } else {
2139                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2140                 /* no need to clear IPPCSE as it defaults to 0 */
2141         }
2142         ew32(RXCSUM, rxcsum);
2143
2144         /* Enable early receives on supported devices, only takes effect when
2145          * packet size is equal or larger than the specified value (in 8 byte
2146          * units), e.g. using jumbo frames when setting to E1000_ERT_2048 */
2147         if ((adapter->flags & FLAG_HAS_ERT) &&
2148             (adapter->netdev->mtu > ETH_DATA_LEN))
2149                 ew32(ERT, E1000_ERT_2048);
2150
2151         /* Enable Receives */
2152         ew32(RCTL, rctl);
2153 }
2154
2155 /**
2156  *  e1000_mc_addr_list_update - Update Multicast addresses
2157  *  @hw: pointer to the HW structure
2158  *  @mc_addr_list: array of multicast addresses to program
2159  *  @mc_addr_count: number of multicast addresses to program
2160  *  @rar_used_count: the first RAR register free to program
2161  *  @rar_count: total number of supported Receive Address Registers
2162  *
2163  *  Updates the Receive Address Registers and Multicast Table Array.
2164  *  The caller must have a packed mc_addr_list of multicast addresses.
2165  *  The parameter rar_count will usually be hw->mac.rar_entry_count
2166  *  unless there are workarounds that change this.  Currently no func pointer
2167  *  exists and all implementations are handled in the generic version of this
2168  *  function.
2169  **/
2170 static void e1000_mc_addr_list_update(struct e1000_hw *hw, u8 *mc_addr_list,
2171                                u32 mc_addr_count, u32 rar_used_count,
2172                                u32 rar_count)
2173 {
2174         hw->mac.ops.mc_addr_list_update(hw, mc_addr_list, mc_addr_count,
2175                                         rar_used_count, rar_count);
2176 }
2177
2178 /**
2179  * e1000_set_multi - Multicast and Promiscuous mode set
2180  * @netdev: network interface device structure
2181  *
2182  * The set_multi entry point is called whenever the multicast address
2183  * list or the network interface flags are updated.  This routine is
2184  * responsible for configuring the hardware for proper multicast,
2185  * promiscuous mode, and all-multi behavior.
2186  **/
2187 static void e1000_set_multi(struct net_device *netdev)
2188 {
2189         struct e1000_adapter *adapter = netdev_priv(netdev);
2190         struct e1000_hw *hw = &adapter->hw;
2191         struct e1000_mac_info *mac = &hw->mac;
2192         struct dev_mc_list *mc_ptr;
2193         u8  *mta_list;
2194         u32 rctl;
2195         int i;
2196
2197         /* Check for Promiscuous and All Multicast modes */
2198
2199         rctl = er32(RCTL);
2200
2201         if (netdev->flags & IFF_PROMISC) {
2202                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2203         } else if (netdev->flags & IFF_ALLMULTI) {
2204                 rctl |= E1000_RCTL_MPE;
2205                 rctl &= ~E1000_RCTL_UPE;
2206         } else {
2207                 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2208         }
2209
2210         ew32(RCTL, rctl);
2211
2212         if (netdev->mc_count) {
2213                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2214                 if (!mta_list)
2215                         return;
2216
2217                 /* prepare a packed array of only addresses. */
2218                 mc_ptr = netdev->mc_list;
2219
2220                 for (i = 0; i < netdev->mc_count; i++) {
2221                         if (!mc_ptr)
2222                                 break;
2223                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2224                                ETH_ALEN);
2225                         mc_ptr = mc_ptr->next;
2226                 }
2227
2228                 e1000_mc_addr_list_update(hw, mta_list, i, 1,
2229                                           mac->rar_entry_count);
2230                 kfree(mta_list);
2231         } else {
2232                 /*
2233                  * if we're called from probe, we might not have
2234                  * anything to do here, so clear out the list
2235                  */
2236                 e1000_mc_addr_list_update(hw, NULL, 0, 1,
2237                                           mac->rar_entry_count);
2238         }
2239 }
2240
2241 /**
2242  * e1000_configure - configure the hardware for RX and TX
2243  * @adapter: private board structure
2244  **/
2245 static void e1000_configure(struct e1000_adapter *adapter)
2246 {
2247         e1000_set_multi(adapter->netdev);
2248
2249         e1000_restore_vlan(adapter);
2250         e1000_init_manageability(adapter);
2251
2252         e1000_configure_tx(adapter);
2253         e1000_setup_rctl(adapter);
2254         e1000_configure_rx(adapter);
2255         adapter->alloc_rx_buf(adapter,
2256                               e1000_desc_unused(adapter->rx_ring));
2257 }
2258
2259 /**
2260  * e1000e_power_up_phy - restore link in case the phy was powered down
2261  * @adapter: address of board private structure
2262  *
2263  * The phy may be powered down to save power and turn off link when the
2264  * driver is unloaded and wake on lan is not enabled (among others)
2265  * *** this routine MUST be followed by a call to e1000e_reset ***
2266  **/
2267 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2268 {
2269         u16 mii_reg = 0;
2270
2271         /* Just clear the power down bit to wake the phy back up */
2272         if (adapter->hw.media_type == e1000_media_type_copper) {
2273                 /* according to the manual, the phy will retain its
2274                  * settings across a power-down/up cycle */
2275                 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2276                 mii_reg &= ~MII_CR_POWER_DOWN;
2277                 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2278         }
2279
2280         adapter->hw.mac.ops.setup_link(&adapter->hw);
2281 }
2282
2283 /**
2284  * e1000_power_down_phy - Power down the PHY
2285  *
2286  * Power down the PHY so no link is implied when interface is down
2287  * The PHY cannot be powered down is management or WoL is active
2288  */
2289 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2290 {
2291         struct e1000_hw *hw = &adapter->hw;
2292         u16 mii_reg;
2293
2294         /* WoL is enabled */
2295         if (!adapter->wol)
2296                 return;
2297
2298         /* non-copper PHY? */
2299         if (adapter->hw.media_type != e1000_media_type_copper)
2300                 return;
2301
2302         /* reset is blocked because of a SoL/IDER session */
2303         if (e1000e_check_mng_mode(hw) ||
2304             e1000_check_reset_block(hw))
2305                 return;
2306
2307         /* managebility (AMT) is enabled */
2308         if (er32(MANC) & E1000_MANC_SMBUS_EN)
2309                 return;
2310
2311         /* power down the PHY */
2312         e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2313         mii_reg |= MII_CR_POWER_DOWN;
2314         e1e_wphy(hw, PHY_CONTROL, mii_reg);
2315         mdelay(1);
2316 }
2317
2318 /**
2319  * e1000e_reset - bring the hardware into a known good state
2320  *
2321  * This function boots the hardware and enables some settings that
2322  * require a configuration cycle of the hardware - those cannot be
2323  * set/changed during runtime. After reset the device needs to be
2324  * properly configured for rx, tx etc.
2325  */
2326 void e1000e_reset(struct e1000_adapter *adapter)
2327 {
2328         struct e1000_mac_info *mac = &adapter->hw.mac;
2329         struct e1000_hw *hw = &adapter->hw;
2330         u32 tx_space, min_tx_space, min_rx_space;
2331         u16 hwm;
2332
2333         if (mac->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN ) {
2334                 /* To maintain wire speed transmits, the Tx FIFO should be
2335                  * large enough to accommodate two full transmit packets,
2336                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2337                  * the Rx FIFO should be large enough to accommodate at least
2338                  * one full receive packet and is similarly rounded up and
2339                  * expressed in KB. */
2340                 adapter->pba = er32(PBA);
2341                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2342                 tx_space = adapter->pba >> 16;
2343                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2344                 adapter->pba &= 0xffff;
2345                 /* the tx fifo also stores 16 bytes of information about the tx
2346                  * but don't include ethernet FCS because hardware appends it */
2347                 min_tx_space = (mac->max_frame_size +
2348                                 sizeof(struct e1000_tx_desc) -
2349                                 ETH_FCS_LEN) * 2;
2350                 min_tx_space = ALIGN(min_tx_space, 1024);
2351                 min_tx_space >>= 10;
2352                 /* software strips receive CRC, so leave room for it */
2353                 min_rx_space = mac->max_frame_size;
2354                 min_rx_space = ALIGN(min_rx_space, 1024);
2355                 min_rx_space >>= 10;
2356
2357                 /* If current Tx allocation is less than the min Tx FIFO size,
2358                  * and the min Tx FIFO size is less than the current Rx FIFO
2359                  * allocation, take space away from current Rx allocation */
2360                 if (tx_space < min_tx_space &&
2361                     ((min_tx_space - tx_space) < adapter->pba)) {
2362                         adapter->pba -= - (min_tx_space - tx_space);
2363
2364                         /* if short on rx space, rx wins and must trump tx
2365                          * adjustment or use Early Receive if available */
2366                         if ((adapter->pba < min_rx_space) &&
2367                             (!(adapter->flags & FLAG_HAS_ERT)))
2368                                 /* ERT enabled in e1000_configure_rx */
2369                                 adapter->pba = min_rx_space;
2370                 }
2371         }
2372
2373         ew32(PBA, adapter->pba);
2374
2375         /* flow control settings */
2376         /* The high water mark must be low enough to fit one full frame
2377          * (or the size used for early receive) above it in the Rx FIFO.
2378          * Set it to the lower of:
2379          * - 90% of the Rx FIFO size, and
2380          * - the full Rx FIFO size minus the early receive size (for parts
2381          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2382          * - the full Rx FIFO size minus one full frame */
2383         if (adapter->flags & FLAG_HAS_ERT)
2384                 hwm = min(((adapter->pba << 10) * 9 / 10),
2385                           ((adapter->pba << 10) - (E1000_ERT_2048 << 3)));
2386         else
2387                 hwm = min(((adapter->pba << 10) * 9 / 10),
2388                           ((adapter->pba << 10) - mac->max_frame_size));
2389
2390         mac->fc_high_water = hwm & 0xFFF8; /* 8-byte granularity */
2391         mac->fc_low_water = mac->fc_high_water - 8;
2392
2393         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2394                 mac->fc_pause_time = 0xFFFF;
2395         else
2396                 mac->fc_pause_time = E1000_FC_PAUSE_TIME;
2397         mac->fc = mac->original_fc;
2398
2399         /* Allow time for pending master requests to run */
2400         mac->ops.reset_hw(hw);
2401         ew32(WUC, 0);
2402
2403         if (mac->ops.init_hw(hw))
2404                 ndev_err(adapter->netdev, "Hardware Error\n");
2405
2406         e1000_update_mng_vlan(adapter);
2407
2408         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2409         ew32(VET, ETH_P_8021Q);
2410
2411         e1000e_reset_adaptive(hw);
2412         e1000_get_phy_info(hw);
2413
2414         if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2415                 u16 phy_data = 0;
2416                 /* speed up time to link by disabling smart power down, ignore
2417                  * the return value of this function because there is nothing
2418                  * different we would do if it failed */
2419                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2420                 phy_data &= ~IGP02E1000_PM_SPD;
2421                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2422         }
2423
2424         e1000_release_manageability(adapter);
2425 }
2426
2427 int e1000e_up(struct e1000_adapter *adapter)
2428 {
2429         struct e1000_hw *hw = &adapter->hw;
2430
2431         /* hardware has been reset, we need to reload some things */
2432         e1000_configure(adapter);
2433
2434         clear_bit(__E1000_DOWN, &adapter->state);
2435
2436         napi_enable(&adapter->napi);
2437         e1000_irq_enable(adapter);
2438
2439         /* fire a link change interrupt to start the watchdog */
2440         ew32(ICS, E1000_ICS_LSC);
2441         return 0;
2442 }
2443
2444 void e1000e_down(struct e1000_adapter *adapter)
2445 {
2446         struct net_device *netdev = adapter->netdev;
2447         struct e1000_hw *hw = &adapter->hw;
2448         u32 tctl, rctl;
2449
2450         /* signal that we're down so the interrupt handler does not
2451          * reschedule our watchdog timer */
2452         set_bit(__E1000_DOWN, &adapter->state);
2453
2454         /* disable receives in the hardware */
2455         rctl = er32(RCTL);
2456         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2457         /* flush and sleep below */
2458
2459         netif_stop_queue(netdev);
2460
2461         /* disable transmits in the hardware */
2462         tctl = er32(TCTL);
2463         tctl &= ~E1000_TCTL_EN;
2464         ew32(TCTL, tctl);
2465         /* flush both disables and wait for them to finish */
2466         e1e_flush();
2467         msleep(10);
2468
2469         napi_disable(&adapter->napi);
2470         e1000_irq_disable(adapter);
2471
2472         del_timer_sync(&adapter->watchdog_timer);
2473         del_timer_sync(&adapter->phy_info_timer);
2474
2475         netdev->tx_queue_len = adapter->tx_queue_len;
2476         netif_carrier_off(netdev);
2477         adapter->link_speed = 0;
2478         adapter->link_duplex = 0;
2479
2480         e1000e_reset(adapter);
2481         e1000_clean_tx_ring(adapter);
2482         e1000_clean_rx_ring(adapter);
2483
2484         /*
2485          * TODO: for power management, we could drop the link and
2486          * pci_disable_device here.
2487          */
2488 }
2489
2490 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2491 {
2492         might_sleep();
2493         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2494                 msleep(1);
2495         e1000e_down(adapter);
2496         e1000e_up(adapter);
2497         clear_bit(__E1000_RESETTING, &adapter->state);
2498 }
2499
2500 /**
2501  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2502  * @adapter: board private structure to initialize
2503  *
2504  * e1000_sw_init initializes the Adapter private data structure.
2505  * Fields are initialized based on PCI device information and
2506  * OS network device settings (MTU size).
2507  **/
2508 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2509 {
2510         struct e1000_hw *hw = &adapter->hw;
2511         struct net_device *netdev = adapter->netdev;
2512
2513         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2514         adapter->rx_ps_bsize0 = 128;
2515         hw->mac.max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2516         hw->mac.min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2517
2518         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2519         if (!adapter->tx_ring)
2520                 goto err;
2521
2522         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2523         if (!adapter->rx_ring)
2524                 goto err;
2525
2526         spin_lock_init(&adapter->tx_queue_lock);
2527
2528         /* Explicitly disable IRQ since the NIC can be in any state. */
2529         atomic_set(&adapter->irq_sem, 0);
2530         e1000_irq_disable(adapter);
2531
2532         spin_lock_init(&adapter->stats_lock);
2533
2534         set_bit(__E1000_DOWN, &adapter->state);
2535         return 0;
2536
2537 err:
2538         ndev_err(netdev, "Unable to allocate memory for queues\n");
2539         kfree(adapter->rx_ring);
2540         kfree(adapter->tx_ring);
2541         return -ENOMEM;
2542 }
2543
2544 /**
2545  * e1000_open - Called when a network interface is made active
2546  * @netdev: network interface device structure
2547  *
2548  * Returns 0 on success, negative value on failure
2549  *
2550  * The open entry point is called when a network interface is made
2551  * active by the system (IFF_UP).  At this point all resources needed
2552  * for transmit and receive operations are allocated, the interrupt
2553  * handler is registered with the OS, the watchdog timer is started,
2554  * and the stack is notified that the interface is ready.
2555  **/
2556 static int e1000_open(struct net_device *netdev)
2557 {
2558         struct e1000_adapter *adapter = netdev_priv(netdev);
2559         struct e1000_hw *hw = &adapter->hw;
2560         int err;
2561
2562         /* disallow open during test */
2563         if (test_bit(__E1000_TESTING, &adapter->state))
2564                 return -EBUSY;
2565
2566         /* allocate transmit descriptors */
2567         err = e1000e_setup_tx_resources(adapter);
2568         if (err)
2569                 goto err_setup_tx;
2570
2571         /* allocate receive descriptors */
2572         err = e1000e_setup_rx_resources(adapter);
2573         if (err)
2574                 goto err_setup_rx;
2575
2576         e1000e_power_up_phy(adapter);
2577
2578         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2579         if ((adapter->hw.mng_cookie.status &
2580              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
2581                 e1000_update_mng_vlan(adapter);
2582
2583         /* If AMT is enabled, let the firmware know that the network
2584          * interface is now open */
2585         if ((adapter->flags & FLAG_HAS_AMT) &&
2586             e1000e_check_mng_mode(&adapter->hw))
2587                 e1000_get_hw_control(adapter);
2588
2589         /* before we allocate an interrupt, we must be ready to handle it.
2590          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2591          * as soon as we call pci_request_irq, so we have to setup our
2592          * clean_rx handler before we do so.  */
2593         e1000_configure(adapter);
2594
2595         err = e1000_request_irq(adapter);
2596         if (err)
2597                 goto err_req_irq;
2598
2599         /* From here on the code is the same as e1000e_up() */
2600         clear_bit(__E1000_DOWN, &adapter->state);
2601
2602         napi_enable(&adapter->napi);
2603
2604         e1000_irq_enable(adapter);
2605
2606         /* fire a link status change interrupt to start the watchdog */
2607         ew32(ICS, E1000_ICS_LSC);
2608
2609         return 0;
2610
2611 err_req_irq:
2612         e1000_release_hw_control(adapter);
2613         e1000_power_down_phy(adapter);
2614         e1000e_free_rx_resources(adapter);
2615 err_setup_rx:
2616         e1000e_free_tx_resources(adapter);
2617 err_setup_tx:
2618         e1000e_reset(adapter);
2619
2620         return err;
2621 }
2622
2623 /**
2624  * e1000_close - Disables a network interface
2625  * @netdev: network interface device structure
2626  *
2627  * Returns 0, this is not allowed to fail
2628  *
2629  * The close entry point is called when an interface is de-activated
2630  * by the OS.  The hardware is still under the drivers control, but
2631  * needs to be disabled.  A global MAC reset is issued to stop the
2632  * hardware, and all transmit and receive resources are freed.
2633  **/
2634 static int e1000_close(struct net_device *netdev)
2635 {
2636         struct e1000_adapter *adapter = netdev_priv(netdev);
2637
2638         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
2639         e1000e_down(adapter);
2640         e1000_power_down_phy(adapter);
2641         e1000_free_irq(adapter);
2642
2643         e1000e_free_tx_resources(adapter);
2644         e1000e_free_rx_resources(adapter);
2645
2646         /* kill manageability vlan ID if supported, but not if a vlan with
2647          * the same ID is registered on the host OS (let 8021q kill it) */
2648         if ((adapter->hw.mng_cookie.status &
2649                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2650              !(adapter->vlgrp &&
2651                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
2652                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2653
2654         /* If AMT is enabled, let the firmware know that the network
2655          * interface is now closed */
2656         if ((adapter->flags & FLAG_HAS_AMT) &&
2657             e1000e_check_mng_mode(&adapter->hw))
2658                 e1000_release_hw_control(adapter);
2659
2660         return 0;
2661 }
2662 /**
2663  * e1000_set_mac - Change the Ethernet Address of the NIC
2664  * @netdev: network interface device structure
2665  * @p: pointer to an address structure
2666  *
2667  * Returns 0 on success, negative on failure
2668  **/
2669 static int e1000_set_mac(struct net_device *netdev, void *p)
2670 {
2671         struct e1000_adapter *adapter = netdev_priv(netdev);
2672         struct sockaddr *addr = p;
2673
2674         if (!is_valid_ether_addr(addr->sa_data))
2675                 return -EADDRNOTAVAIL;
2676
2677         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2678         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
2679
2680         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
2681
2682         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
2683                 /* activate the work around */
2684                 e1000e_set_laa_state_82571(&adapter->hw, 1);
2685
2686                 /* Hold a copy of the LAA in RAR[14] This is done so that
2687                  * between the time RAR[0] gets clobbered  and the time it
2688                  * gets fixed (in e1000_watchdog), the actual LAA is in one
2689                  * of the RARs and no incoming packets directed to this port
2690                  * are dropped. Eventually the LAA will be in RAR[0] and
2691                  * RAR[14] */
2692                 e1000e_rar_set(&adapter->hw,
2693                               adapter->hw.mac.addr,
2694                               adapter->hw.mac.rar_entry_count - 1);
2695         }
2696
2697         return 0;
2698 }
2699
2700 /* Need to wait a few seconds after link up to get diagnostic information from
2701  * the phy */
2702 static void e1000_update_phy_info(unsigned long data)
2703 {
2704         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2705         e1000_get_phy_info(&adapter->hw);
2706 }
2707
2708 /**
2709  * e1000e_update_stats - Update the board statistics counters
2710  * @adapter: board private structure
2711  **/
2712 void e1000e_update_stats(struct e1000_adapter *adapter)
2713 {
2714         struct e1000_hw *hw = &adapter->hw;
2715         struct pci_dev *pdev = adapter->pdev;
2716         unsigned long irq_flags;
2717         u16 phy_tmp;
2718
2719 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
2720
2721         /*
2722          * Prevent stats update while adapter is being reset, or if the pci
2723          * connection is down.
2724          */
2725         if (adapter->link_speed == 0)
2726                 return;
2727         if (pci_channel_offline(pdev))
2728                 return;
2729
2730         spin_lock_irqsave(&adapter->stats_lock, irq_flags);
2731
2732         /* these counters are modified from e1000_adjust_tbi_stats,
2733          * called from the interrupt context, so they must only
2734          * be written while holding adapter->stats_lock
2735          */
2736
2737         adapter->stats.crcerrs += er32(CRCERRS);
2738         adapter->stats.gprc += er32(GPRC);
2739         adapter->stats.gorcl += er32(GORCL);
2740         adapter->stats.gorch += er32(GORCH);
2741         adapter->stats.bprc += er32(BPRC);
2742         adapter->stats.mprc += er32(MPRC);
2743         adapter->stats.roc += er32(ROC);
2744
2745         if (adapter->flags & FLAG_HAS_STATS_PTC_PRC) {
2746                 adapter->stats.prc64 += er32(PRC64);
2747                 adapter->stats.prc127 += er32(PRC127);
2748                 adapter->stats.prc255 += er32(PRC255);
2749                 adapter->stats.prc511 += er32(PRC511);
2750                 adapter->stats.prc1023 += er32(PRC1023);
2751                 adapter->stats.prc1522 += er32(PRC1522);
2752                 adapter->stats.symerrs += er32(SYMERRS);
2753                 adapter->stats.sec += er32(SEC);
2754         }
2755
2756         adapter->stats.mpc += er32(MPC);
2757         adapter->stats.scc += er32(SCC);
2758         adapter->stats.ecol += er32(ECOL);
2759         adapter->stats.mcc += er32(MCC);
2760         adapter->stats.latecol += er32(LATECOL);
2761         adapter->stats.dc += er32(DC);
2762         adapter->stats.rlec += er32(RLEC);
2763         adapter->stats.xonrxc += er32(XONRXC);
2764         adapter->stats.xontxc += er32(XONTXC);
2765         adapter->stats.xoffrxc += er32(XOFFRXC);
2766         adapter->stats.xofftxc += er32(XOFFTXC);
2767         adapter->stats.fcruc += er32(FCRUC);
2768         adapter->stats.gptc += er32(GPTC);
2769         adapter->stats.gotcl += er32(GOTCL);
2770         adapter->stats.gotch += er32(GOTCH);
2771         adapter->stats.rnbc += er32(RNBC);
2772         adapter->stats.ruc += er32(RUC);
2773         adapter->stats.rfc += er32(RFC);
2774         adapter->stats.rjc += er32(RJC);
2775         adapter->stats.torl += er32(TORL);
2776         adapter->stats.torh += er32(TORH);
2777         adapter->stats.totl += er32(TOTL);
2778         adapter->stats.toth += er32(TOTH);
2779         adapter->stats.tpr += er32(TPR);
2780
2781         if (adapter->flags & FLAG_HAS_STATS_PTC_PRC) {
2782                 adapter->stats.ptc64 += er32(PTC64);
2783                 adapter->stats.ptc127 += er32(PTC127);
2784                 adapter->stats.ptc255 += er32(PTC255);
2785                 adapter->stats.ptc511 += er32(PTC511);
2786                 adapter->stats.ptc1023 += er32(PTC1023);
2787                 adapter->stats.ptc1522 += er32(PTC1522);
2788         }
2789
2790         adapter->stats.mptc += er32(MPTC);
2791         adapter->stats.bptc += er32(BPTC);
2792
2793         /* used for adaptive IFS */
2794
2795         hw->mac.tx_packet_delta = er32(TPT);
2796         adapter->stats.tpt += hw->mac.tx_packet_delta;
2797         hw->mac.collision_delta = er32(COLC);
2798         adapter->stats.colc += hw->mac.collision_delta;
2799
2800         adapter->stats.algnerrc += er32(ALGNERRC);
2801         adapter->stats.rxerrc += er32(RXERRC);
2802         adapter->stats.tncrs += er32(TNCRS);
2803         adapter->stats.cexterr += er32(CEXTERR);
2804         adapter->stats.tsctc += er32(TSCTC);
2805         adapter->stats.tsctfc += er32(TSCTFC);
2806
2807         adapter->stats.iac += er32(IAC);
2808
2809         if (adapter->flags & FLAG_HAS_STATS_ICR_ICT) {
2810                 adapter->stats.icrxoc += er32(ICRXOC);
2811                 adapter->stats.icrxptc += er32(ICRXPTC);
2812                 adapter->stats.icrxatc += er32(ICRXATC);
2813                 adapter->stats.ictxptc += er32(ICTXPTC);
2814                 adapter->stats.ictxatc += er32(ICTXATC);
2815                 adapter->stats.ictxqec += er32(ICTXQEC);
2816                 adapter->stats.ictxqmtc += er32(ICTXQMTC);
2817                 adapter->stats.icrxdmtc += er32(ICRXDMTC);
2818         }
2819
2820         /* Fill out the OS statistics structure */
2821         adapter->net_stats.rx_packets = adapter->stats.gprc;
2822         adapter->net_stats.tx_packets = adapter->stats.gptc;
2823         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
2824         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
2825         adapter->net_stats.multicast = adapter->stats.mprc;
2826         adapter->net_stats.collisions = adapter->stats.colc;
2827
2828         /* Rx Errors */
2829
2830         /* RLEC on some newer hardware can be incorrect so build
2831         * our own version based on RUC and ROC */
2832         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
2833                 adapter->stats.crcerrs + adapter->stats.algnerrc +
2834                 adapter->stats.ruc + adapter->stats.roc +
2835                 adapter->stats.cexterr;
2836         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
2837                                               adapter->stats.roc;
2838         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
2839         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
2840         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
2841
2842         /* Tx Errors */
2843         adapter->net_stats.tx_errors = adapter->stats.ecol +
2844                                        adapter->stats.latecol;
2845         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
2846         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
2847         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
2848
2849         /* Tx Dropped needs to be maintained elsewhere */
2850
2851         /* Phy Stats */
2852         if (hw->media_type == e1000_media_type_copper) {
2853                 if ((adapter->link_speed == SPEED_1000) &&
2854                    (!e1e_rphy(hw, PHY_1000T_STATUS, &phy_tmp))) {
2855                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
2856                         adapter->phy_stats.idle_errors += phy_tmp;
2857                 }
2858         }
2859
2860         /* Management Stats */
2861         adapter->stats.mgptc += er32(MGTPTC);
2862         adapter->stats.mgprc += er32(MGTPRC);
2863         adapter->stats.mgpdc += er32(MGTPDC);
2864
2865         spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
2866 }
2867
2868 static void e1000_print_link_info(struct e1000_adapter *adapter)
2869 {
2870         struct net_device *netdev = adapter->netdev;
2871         struct e1000_hw *hw = &adapter->hw;
2872         u32 ctrl = er32(CTRL);
2873
2874         ndev_info(netdev,
2875                 "Link is Up %d Mbps %s, Flow Control: %s\n",
2876                 adapter->link_speed,
2877                 (adapter->link_duplex == FULL_DUPLEX) ?
2878                                 "Full Duplex" : "Half Duplex",
2879                 ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
2880                                 "RX/TX" :
2881                 ((ctrl & E1000_CTRL_RFCE) ? "RX" :
2882                 ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
2883 }
2884
2885 /**
2886  * e1000_watchdog - Timer Call-back
2887  * @data: pointer to adapter cast into an unsigned long
2888  **/
2889 static void e1000_watchdog(unsigned long data)
2890 {
2891         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2892
2893         /* Do the rest outside of interrupt context */
2894         schedule_work(&adapter->watchdog_task);
2895
2896         /* TODO: make this use queue_delayed_work() */
2897 }
2898
2899 static void e1000_watchdog_task(struct work_struct *work)
2900 {
2901         struct e1000_adapter *adapter = container_of(work,
2902                                         struct e1000_adapter, watchdog_task);
2903
2904         struct net_device *netdev = adapter->netdev;
2905         struct e1000_mac_info *mac = &adapter->hw.mac;
2906         struct e1000_ring *tx_ring = adapter->tx_ring;
2907         struct e1000_hw *hw = &adapter->hw;
2908         u32 link, tctl;
2909         s32 ret_val;
2910         int tx_pending = 0;
2911
2912         if ((netif_carrier_ok(netdev)) &&
2913             (er32(STATUS) & E1000_STATUS_LU))
2914                 goto link_up;
2915
2916         ret_val = mac->ops.check_for_link(hw);
2917         if ((ret_val == E1000_ERR_PHY) &&
2918             (adapter->hw.phy.type == e1000_phy_igp_3) &&
2919             (er32(CTRL) &
2920              E1000_PHY_CTRL_GBE_DISABLE)) {
2921                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
2922                 ndev_info(netdev,
2923                         "Gigabit has been disabled, downgrading speed\n");
2924         }
2925
2926         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
2927             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
2928                 e1000_update_mng_vlan(adapter);
2929
2930         if ((adapter->hw.media_type == e1000_media_type_internal_serdes) &&
2931            !(er32(TXCW) & E1000_TXCW_ANE))
2932                 link = adapter->hw.mac.serdes_has_link;
2933         else
2934                 link = er32(STATUS) & E1000_STATUS_LU;
2935
2936         if (link) {
2937                 if (!netif_carrier_ok(netdev)) {
2938                         bool txb2b = 1;
2939                         mac->ops.get_link_up_info(&adapter->hw,
2940                                                    &adapter->link_speed,
2941                                                    &adapter->link_duplex);
2942                         e1000_print_link_info(adapter);
2943                         /* tweak tx_queue_len according to speed/duplex
2944                          * and adjust the timeout factor */
2945                         netdev->tx_queue_len = adapter->tx_queue_len;
2946                         adapter->tx_timeout_factor = 1;
2947                         switch (adapter->link_speed) {
2948                         case SPEED_10:
2949                                 txb2b = 0;
2950                                 netdev->tx_queue_len = 10;
2951                                 adapter->tx_timeout_factor = 14;
2952                                 break;
2953                         case SPEED_100:
2954                                 txb2b = 0;
2955                                 netdev->tx_queue_len = 100;
2956                                 /* maybe add some timeout factor ? */
2957                                 break;
2958                         }
2959
2960                         /* workaround: re-program speed mode bit after
2961                          * link-up event */
2962                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
2963                             !txb2b) {
2964                                 u32 tarc0;
2965                                 tarc0 = er32(TARC0);
2966                                 tarc0 &= ~SPEED_MODE_BIT;
2967                                 ew32(TARC0, tarc0);
2968                         }
2969
2970                         /* disable TSO for pcie and 10/100 speeds, to avoid
2971                          * some hardware issues */
2972                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
2973                                 switch (adapter->link_speed) {
2974                                 case SPEED_10:
2975                                 case SPEED_100:
2976                                         ndev_info(netdev,
2977                                         "10/100 speed: disabling TSO\n");
2978                                         netdev->features &= ~NETIF_F_TSO;
2979                                         netdev->features &= ~NETIF_F_TSO6;
2980                                         break;
2981                                 case SPEED_1000:
2982                                         netdev->features |= NETIF_F_TSO;
2983                                         netdev->features |= NETIF_F_TSO6;
2984                                         break;
2985                                 default:
2986                                         /* oops */
2987                                         break;
2988                                 }
2989                         }
2990
2991                         /* enable transmits in the hardware, need to do this
2992                          * after setting TARC0 */
2993                         tctl = er32(TCTL);
2994                         tctl |= E1000_TCTL_EN;
2995                         ew32(TCTL, tctl);
2996
2997                         netif_carrier_on(netdev);
2998                         netif_wake_queue(netdev);
2999
3000                         if (!test_bit(__E1000_DOWN, &adapter->state))
3001                                 mod_timer(&adapter->phy_info_timer,
3002                                           round_jiffies(jiffies + 2 * HZ));
3003                 } else {
3004                         /* make sure the receive unit is started */
3005                         if (adapter->flags & FLAG_RX_NEEDS_RESTART) {
3006                                 u32 rctl = er32(RCTL);
3007                                 ew32(RCTL, rctl |
3008                                                 E1000_RCTL_EN);
3009                         }
3010                 }
3011         } else {
3012                 if (netif_carrier_ok(netdev)) {
3013                         adapter->link_speed = 0;
3014                         adapter->link_duplex = 0;
3015                         ndev_info(netdev, "Link is Down\n");
3016                         netif_carrier_off(netdev);
3017                         netif_stop_queue(netdev);
3018                         if (!test_bit(__E1000_DOWN, &adapter->state))
3019                                 mod_timer(&adapter->phy_info_timer,
3020                                           round_jiffies(jiffies + 2 * HZ));
3021
3022                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3023                                 schedule_work(&adapter->reset_task);
3024                 }
3025         }
3026
3027 link_up:
3028         e1000e_update_stats(adapter);
3029
3030         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3031         adapter->tpt_old = adapter->stats.tpt;
3032         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3033         adapter->colc_old = adapter->stats.colc;
3034
3035         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
3036         adapter->gorcl_old = adapter->stats.gorcl;
3037         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
3038         adapter->gotcl_old = adapter->stats.gotcl;
3039
3040         e1000e_update_adaptive(&adapter->hw);
3041
3042         if (!netif_carrier_ok(netdev)) {
3043                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3044                                tx_ring->count);
3045                 if (tx_pending) {
3046                         /* We've lost link, so the controller stops DMA,
3047                          * but we've got queued Tx work that's never going
3048                          * to get done, so reset controller to flush Tx.
3049                          * (Do the reset outside of interrupt context). */
3050                         adapter->tx_timeout_count++;
3051                         schedule_work(&adapter->reset_task);
3052                 }
3053         }
3054
3055         /* Cause software interrupt to ensure rx ring is cleaned */
3056         ew32(ICS, E1000_ICS_RXDMT0);
3057
3058         /* Force detection of hung controller every watchdog period */
3059         adapter->detect_tx_hung = 1;
3060
3061         /* With 82571 controllers, LAA may be overwritten due to controller
3062          * reset from the other port. Set the appropriate LAA in RAR[0] */
3063         if (e1000e_get_laa_state_82571(hw))
3064                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3065
3066         /* Reset the timer */
3067         if (!test_bit(__E1000_DOWN, &adapter->state))
3068                 mod_timer(&adapter->watchdog_timer,
3069                           round_jiffies(jiffies + 2 * HZ));
3070 }
3071
3072 #define E1000_TX_FLAGS_CSUM             0x00000001
3073 #define E1000_TX_FLAGS_VLAN             0x00000002
3074 #define E1000_TX_FLAGS_TSO              0x00000004
3075 #define E1000_TX_FLAGS_IPV4             0x00000008
3076 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3077 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3078
3079 static int e1000_tso(struct e1000_adapter *adapter,
3080                      struct sk_buff *skb)
3081 {
3082         struct e1000_ring *tx_ring = adapter->tx_ring;
3083         struct e1000_context_desc *context_desc;
3084         struct e1000_buffer *buffer_info;
3085         unsigned int i;
3086         u32 cmd_length = 0;
3087         u16 ipcse = 0, tucse, mss;
3088         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3089         int err;
3090
3091         if (skb_is_gso(skb)) {
3092                 if (skb_header_cloned(skb)) {
3093                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3094                         if (err)
3095                                 return err;
3096                 }
3097
3098                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3099                 mss = skb_shinfo(skb)->gso_size;
3100                 if (skb->protocol == htons(ETH_P_IP)) {
3101                         struct iphdr *iph = ip_hdr(skb);
3102                         iph->tot_len = 0;
3103                         iph->check = 0;
3104                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3105                                                                  iph->daddr, 0,
3106                                                                  IPPROTO_TCP,
3107                                                                  0);
3108                         cmd_length = E1000_TXD_CMD_IP;
3109                         ipcse = skb_transport_offset(skb) - 1;
3110                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3111                         ipv6_hdr(skb)->payload_len = 0;
3112                         tcp_hdr(skb)->check =
3113                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3114                                                  &ipv6_hdr(skb)->daddr,
3115                                                  0, IPPROTO_TCP, 0);
3116                         ipcse = 0;
3117                 }
3118                 ipcss = skb_network_offset(skb);
3119                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3120                 tucss = skb_transport_offset(skb);
3121                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3122                 tucse = 0;
3123
3124                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3125                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3126
3127                 i = tx_ring->next_to_use;
3128                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3129                 buffer_info = &tx_ring->buffer_info[i];
3130
3131                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3132                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3133                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3134                 context_desc->upper_setup.tcp_fields.tucss = tucss;
3135                 context_desc->upper_setup.tcp_fields.tucso = tucso;
3136                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3137                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3138                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3139                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3140
3141                 buffer_info->time_stamp = jiffies;
3142                 buffer_info->next_to_watch = i;
3143
3144                 i++;
3145                 if (i == tx_ring->count)
3146                         i = 0;
3147                 tx_ring->next_to_use = i;
3148
3149                 return 1;
3150         }
3151
3152         return 0;
3153 }
3154
3155 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3156 {
3157         struct e1000_ring *tx_ring = adapter->tx_ring;
3158         struct e1000_context_desc *context_desc;
3159         struct e1000_buffer *buffer_info;
3160         unsigned int i;
3161         u8 css;
3162
3163         if (skb->ip_summed == CHECKSUM_PARTIAL) {
3164                 css = skb_transport_offset(skb);
3165
3166                 i = tx_ring->next_to_use;
3167                 buffer_info = &tx_ring->buffer_info[i];
3168                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3169
3170                 context_desc->lower_setup.ip_config = 0;
3171                 context_desc->upper_setup.tcp_fields.tucss = css;
3172                 context_desc->upper_setup.tcp_fields.tucso =
3173                                         css + skb->csum_offset;
3174                 context_desc->upper_setup.tcp_fields.tucse = 0;
3175                 context_desc->tcp_seg_setup.data = 0;
3176                 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
3177
3178                 buffer_info->time_stamp = jiffies;
3179                 buffer_info->next_to_watch = i;
3180
3181                 i++;
3182                 if (i == tx_ring->count)
3183                         i = 0;
3184                 tx_ring->next_to_use = i;
3185
3186                 return 1;
3187         }
3188
3189         return 0;
3190 }
3191
3192 #define E1000_MAX_PER_TXD       8192
3193 #define E1000_MAX_TXD_PWR       12
3194
3195 static int e1000_tx_map(struct e1000_adapter *adapter,
3196                         struct sk_buff *skb, unsigned int first,
3197                         unsigned int max_per_txd, unsigned int nr_frags,
3198                         unsigned int mss)
3199 {
3200         struct e1000_ring *tx_ring = adapter->tx_ring;
3201         struct e1000_buffer *buffer_info;
3202         unsigned int len = skb->len - skb->data_len;
3203         unsigned int offset = 0, size, count = 0, i;
3204         unsigned int f;
3205
3206         i = tx_ring->next_to_use;
3207
3208         while (len) {
3209                 buffer_info = &tx_ring->buffer_info[i];
3210                 size = min(len, max_per_txd);
3211
3212                 /* Workaround for premature desc write-backs
3213                  * in TSO mode.  Append 4-byte sentinel desc */
3214                 if (mss && !nr_frags && size == len && size > 8)
3215                         size -= 4;
3216
3217                 buffer_info->length = size;
3218                 /* set time_stamp *before* dma to help avoid a possible race */
3219                 buffer_info->time_stamp = jiffies;
3220                 buffer_info->dma =
3221                         pci_map_single(adapter->pdev,
3222                                 skb->data + offset,
3223                                 size,
3224                                 PCI_DMA_TODEVICE);
3225                 if (pci_dma_mapping_error(buffer_info->dma)) {
3226                         dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3227                         adapter->tx_dma_failed++;
3228                         return -1;
3229                 }
3230                 buffer_info->next_to_watch = i;
3231
3232                 len -= size;
3233                 offset += size;
3234                 count++;
3235                 i++;
3236                 if (i == tx_ring->count)
3237                         i = 0;
3238         }
3239
3240         for (f = 0; f < nr_frags; f++) {
3241                 struct skb_frag_struct *frag;
3242
3243                 frag = &skb_shinfo(skb)->frags[f];
3244                 len = frag->size;
3245                 offset = frag->page_offset;
3246
3247                 while (len) {
3248                         buffer_info = &tx_ring->buffer_info[i];
3249                         size = min(len, max_per_txd);
3250                         /* Workaround for premature desc write-backs
3251                          * in TSO mode.  Append 4-byte sentinel desc */
3252                         if (mss && f == (nr_frags-1) && size == len && size > 8)
3253                                 size -= 4;
3254
3255                         buffer_info->length = size;
3256                         buffer_info->time_stamp = jiffies;
3257                         buffer_info->dma =
3258                                 pci_map_page(adapter->pdev,
3259                                         frag->page,
3260                                         offset,
3261                                         size,
3262                                         PCI_DMA_TODEVICE);
3263                         if (pci_dma_mapping_error(buffer_info->dma)) {
3264                                 dev_err(&adapter->pdev->dev,
3265                                         "TX DMA page map failed\n");
3266                                 adapter->tx_dma_failed++;
3267                                 return -1;
3268                         }
3269
3270                         buffer_info->next_to_watch = i;
3271
3272                         len -= size;
3273                         offset += size;
3274                         count++;
3275
3276                         i++;
3277                         if (i == tx_ring->count)
3278                                 i = 0;
3279                 }
3280         }
3281
3282         if (i == 0)
3283                 i = tx_ring->count - 1;
3284         else
3285                 i--;
3286
3287         tx_ring->buffer_info[i].skb = skb;
3288         tx_ring->buffer_info[first].next_to_watch = i;
3289
3290         return count;
3291 }
3292
3293 static void e1000_tx_queue(struct e1000_adapter *adapter,
3294                            int tx_flags, int count)
3295 {
3296         struct e1000_ring *tx_ring = adapter->tx_ring;
3297         struct e1000_tx_desc *tx_desc = NULL;
3298         struct e1000_buffer *buffer_info;
3299         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3300         unsigned int i;
3301
3302         if (tx_flags & E1000_TX_FLAGS_TSO) {
3303                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3304                              E1000_TXD_CMD_TSE;
3305                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3306
3307                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3308                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3309         }
3310
3311         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3312                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3313                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3314         }
3315
3316         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3317                 txd_lower |= E1000_TXD_CMD_VLE;
3318                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3319         }
3320
3321         i = tx_ring->next_to_use;
3322
3323         while (count--) {
3324                 buffer_info = &tx_ring->buffer_info[i];
3325                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3326                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3327                 tx_desc->lower.data =
3328                         cpu_to_le32(txd_lower | buffer_info->length);
3329                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3330
3331                 i++;
3332                 if (i == tx_ring->count)
3333                         i = 0;
3334         }
3335
3336         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3337
3338         /* Force memory writes to complete before letting h/w
3339          * know there are new descriptors to fetch.  (Only
3340          * applicable for weak-ordered memory model archs,
3341          * such as IA-64). */
3342         wmb();
3343
3344         tx_ring->next_to_use = i;
3345         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3346         /* we need this if more than one processor can write to our tail
3347          * at a time, it synchronizes IO on IA64/Altix systems */
3348         mmiowb();
3349 }
3350
3351 #define MINIMUM_DHCP_PACKET_SIZE 282
3352 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3353                                     struct sk_buff *skb)
3354 {
3355         struct e1000_hw *hw =  &adapter->hw;
3356         u16 length, offset;
3357
3358         if (vlan_tx_tag_present(skb)) {
3359                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3360                     && (adapter->hw.mng_cookie.status &
3361                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3362                         return 0;
3363         }
3364
3365         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3366                 return 0;
3367
3368         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3369                 return 0;
3370
3371         {
3372                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3373                 struct udphdr *udp;
3374
3375                 if (ip->protocol != IPPROTO_UDP)
3376                         return 0;
3377
3378                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3379                 if (ntohs(udp->dest) != 67)
3380                         return 0;
3381
3382                 offset = (u8 *)udp + 8 - skb->data;
3383                 length = skb->len - offset;
3384                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
3385         }
3386
3387         return 0;
3388 }
3389
3390 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3391 {
3392         struct e1000_adapter *adapter = netdev_priv(netdev);
3393
3394         netif_stop_queue(netdev);
3395         /* Herbert's original patch had:
3396          *  smp_mb__after_netif_stop_queue();
3397          * but since that doesn't exist yet, just open code it. */
3398         smp_mb();
3399
3400         /* We need to check again in a case another CPU has just
3401          * made room available. */
3402         if (e1000_desc_unused(adapter->tx_ring) < size)
3403                 return -EBUSY;
3404
3405         /* A reprieve! */
3406         netif_start_queue(netdev);
3407         ++adapter->restart_queue;
3408         return 0;
3409 }
3410
3411 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
3412 {
3413         struct e1000_adapter *adapter = netdev_priv(netdev);
3414
3415         if (e1000_desc_unused(adapter->tx_ring) >= size)
3416                 return 0;
3417         return __e1000_maybe_stop_tx(netdev, size);
3418 }
3419
3420 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3421 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3422 {
3423         struct e1000_adapter *adapter = netdev_priv(netdev);
3424         struct e1000_ring *tx_ring = adapter->tx_ring;
3425         unsigned int first;
3426         unsigned int max_per_txd = E1000_MAX_PER_TXD;
3427         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3428         unsigned int tx_flags = 0;
3429         unsigned int len = skb->len;
3430         unsigned long irq_flags;
3431         unsigned int nr_frags = 0;
3432         unsigned int mss = 0;
3433         int count = 0;
3434         int tso;
3435         unsigned int f;
3436         len -= skb->data_len;
3437
3438         if (test_bit(__E1000_DOWN, &adapter->state)) {
3439                 dev_kfree_skb_any(skb);
3440                 return NETDEV_TX_OK;
3441         }
3442
3443         if (skb->len <= 0) {
3444                 dev_kfree_skb_any(skb);
3445                 return NETDEV_TX_OK;
3446         }
3447
3448         mss = skb_shinfo(skb)->gso_size;
3449         /* The controller does a simple calculation to
3450          * make sure there is enough room in the FIFO before
3451          * initiating the DMA for each buffer.  The calc is:
3452          * 4 = ceil(buffer len/mss).  To make sure we don't
3453          * overrun the FIFO, adjust the max buffer len if mss
3454          * drops. */
3455         if (mss) {
3456                 u8 hdr_len;
3457                 max_per_txd = min(mss << 2, max_per_txd);
3458                 max_txd_pwr = fls(max_per_txd) - 1;
3459
3460                 /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
3461                 * points to just header, pull a few bytes of payload from
3462                 * frags into skb->data */
3463                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3464                 if (skb->data_len && (hdr_len == (skb->len - skb->data_len))) {
3465                         unsigned int pull_size;
3466
3467                         pull_size = min((unsigned int)4, skb->data_len);
3468                         if (!__pskb_pull_tail(skb, pull_size)) {
3469                                 ndev_err(netdev,
3470                                          "__pskb_pull_tail failed.\n");
3471                                 dev_kfree_skb_any(skb);
3472                                 return NETDEV_TX_OK;
3473                         }
3474                         len = skb->len - skb->data_len;
3475                 }
3476         }
3477
3478         /* reserve a descriptor for the offload context */
3479         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3480                 count++;
3481         count++;
3482
3483         count += TXD_USE_COUNT(len, max_txd_pwr);
3484
3485         nr_frags = skb_shinfo(skb)->nr_frags;
3486         for (f = 0; f < nr_frags; f++)
3487                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3488                                        max_txd_pwr);
3489
3490         if (adapter->hw.mac.tx_pkt_filtering)
3491                 e1000_transfer_dhcp_info(adapter, skb);
3492
3493         if (!spin_trylock_irqsave(&adapter->tx_queue_lock, irq_flags))
3494                 /* Collision - tell upper layer to requeue */
3495                 return NETDEV_TX_LOCKED;
3496
3497         /* need: count + 2 desc gap to keep tail from touching
3498          * head, otherwise try next time */
3499         if (e1000_maybe_stop_tx(netdev, count + 2)) {
3500                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3501                 return NETDEV_TX_BUSY;
3502         }
3503
3504         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3505                 tx_flags |= E1000_TX_FLAGS_VLAN;
3506                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3507         }
3508
3509         first = tx_ring->next_to_use;
3510
3511         tso = e1000_tso(adapter, skb);
3512         if (tso < 0) {
3513                 dev_kfree_skb_any(skb);
3514                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3515                 return NETDEV_TX_OK;
3516         }
3517
3518         if (tso)
3519                 tx_flags |= E1000_TX_FLAGS_TSO;
3520         else if (e1000_tx_csum(adapter, skb))
3521                 tx_flags |= E1000_TX_FLAGS_CSUM;
3522
3523         /* Old method was to assume IPv4 packet by default if TSO was enabled.
3524          * 82571 hardware supports TSO capabilities for IPv6 as well...
3525          * no longer assume, we must. */
3526         if (skb->protocol == htons(ETH_P_IP))
3527                 tx_flags |= E1000_TX_FLAGS_IPV4;
3528
3529         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
3530         if (count < 0) {
3531                 /* handle pci_map_single() error in e1000_tx_map */
3532                 dev_kfree_skb_any(skb);
3533                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3534                 return NETDEV_TX_OK;
3535         }
3536
3537         e1000_tx_queue(adapter, tx_flags, count);
3538
3539         netdev->trans_start = jiffies;
3540
3541         /* Make sure there is space in the ring for the next send. */
3542         e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
3543
3544         spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3545         return NETDEV_TX_OK;
3546 }
3547
3548 /**
3549  * e1000_tx_timeout - Respond to a Tx Hang
3550  * @netdev: network interface device structure
3551  **/
3552 static void e1000_tx_timeout(struct net_device *netdev)
3553 {
3554         struct e1000_adapter *adapter = netdev_priv(netdev);
3555
3556         /* Do the reset outside of interrupt context */
3557         adapter->tx_timeout_count++;
3558         schedule_work(&adapter->reset_task);
3559 }
3560
3561 static void e1000_reset_task(struct work_struct *work)
3562 {
3563         struct e1000_adapter *adapter;
3564         adapter = container_of(work, struct e1000_adapter, reset_task);
3565
3566         e1000e_reinit_locked(adapter);
3567 }
3568
3569 /**
3570  * e1000_get_stats - Get System Network Statistics
3571  * @netdev: network interface device structure
3572  *
3573  * Returns the address of the device statistics structure.
3574  * The statistics are actually updated from the timer callback.
3575  **/
3576 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3577 {
3578         struct e1000_adapter *adapter = netdev_priv(netdev);
3579
3580         /* only return the current stats */
3581         return &adapter->net_stats;
3582 }
3583
3584 /**
3585  * e1000_change_mtu - Change the Maximum Transfer Unit
3586  * @netdev: network interface device structure
3587  * @new_mtu: new value for maximum frame size
3588  *
3589  * Returns 0 on success, negative on failure
3590  **/
3591 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3592 {
3593         struct e1000_adapter *adapter = netdev_priv(netdev);
3594         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3595
3596         if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
3597             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3598                 ndev_err(netdev, "Invalid MTU setting\n");
3599                 return -EINVAL;
3600         }
3601
3602         /* Jumbo frame size limits */
3603         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
3604                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
3605                         ndev_err(netdev, "Jumbo Frames not supported.\n");
3606                         return -EINVAL;
3607                 }
3608                 if (adapter->hw.phy.type == e1000_phy_ife) {
3609                         ndev_err(netdev, "Jumbo Frames not supported.\n");
3610                         return -EINVAL;
3611                 }
3612         }
3613
3614 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3615         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3616                 ndev_err(netdev, "MTU > 9216 not supported.\n");
3617                 return -EINVAL;
3618         }
3619
3620         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3621                 msleep(1);
3622         /* e1000e_down has a dependency on max_frame_size */
3623         adapter->hw.mac.max_frame_size = max_frame;
3624         if (netif_running(netdev))
3625                 e1000e_down(adapter);
3626
3627         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3628          * means we reserve 2 more, this pushes us to allocate from the next
3629          * larger slab size.
3630          * i.e. RXBUFFER_2048 --> size-4096 slab
3631          *  however with the new *_jumbo* routines, jumbo receives will use
3632          *  fragmented skbs */
3633
3634         if (max_frame <= 256)
3635                 adapter->rx_buffer_len = 256;
3636         else if (max_frame <= 512)
3637                 adapter->rx_buffer_len = 512;
3638         else if (max_frame <= 1024)
3639                 adapter->rx_buffer_len = 1024;
3640         else if (max_frame <= 2048)
3641                 adapter->rx_buffer_len = 2048;
3642         else
3643                 adapter->rx_buffer_len = 4096;
3644
3645         /* adjust allocation if LPE protects us, and we aren't using SBP */
3646         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3647              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
3648                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
3649                                          + ETH_FCS_LEN ;
3650
3651         ndev_info(netdev, "changing MTU from %d to %d\n",
3652                 netdev->mtu, new_mtu);
3653         netdev->mtu = new_mtu;
3654
3655         if (netif_running(netdev))
3656                 e1000e_up(adapter);
3657         else
3658                 e1000e_reset(adapter);
3659
3660         clear_bit(__E1000_RESETTING, &adapter->state);
3661
3662         return 0;
3663 }
3664
3665 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
3666                            int cmd)
3667 {
3668         struct e1000_adapter *adapter = netdev_priv(netdev);
3669         struct mii_ioctl_data *data = if_mii(ifr);
3670         unsigned long irq_flags;
3671
3672         if (adapter->hw.media_type != e1000_media_type_copper)
3673                 return -EOPNOTSUPP;
3674
3675         switch (cmd) {
3676         case SIOCGMIIPHY:
3677                 data->phy_id = adapter->hw.phy.addr;
3678                 break;
3679         case SIOCGMIIREG:
3680                 if (!capable(CAP_NET_ADMIN))
3681                         return -EPERM;
3682                 spin_lock_irqsave(&adapter->stats_lock, irq_flags);
3683                 if (e1e_rphy(&adapter->hw, data->reg_num & 0x1F,
3684                                    &data->val_out)) {
3685                         spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
3686                         return -EIO;
3687                 }
3688                 spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
3689                 break;
3690         case SIOCSMIIREG:
3691         default:
3692                 return -EOPNOTSUPP;
3693         }
3694         return 0;
3695 }
3696
3697 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
3698 {
3699         switch (cmd) {
3700         case SIOCGMIIPHY:
3701         case SIOCGMIIREG:
3702         case SIOCSMIIREG:
3703                 return e1000_mii_ioctl(netdev, ifr, cmd);
3704         default:
3705                 return -EOPNOTSUPP;
3706         }
3707 }
3708
3709 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
3710 {
3711         struct net_device *netdev = pci_get_drvdata(pdev);
3712         struct e1000_adapter *adapter = netdev_priv(netdev);
3713         struct e1000_hw *hw = &adapter->hw;
3714         u32 ctrl, ctrl_ext, rctl, status;
3715         u32 wufc = adapter->wol;
3716         int retval = 0;
3717
3718         netif_device_detach(netdev);
3719
3720         if (netif_running(netdev)) {
3721                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3722                 e1000e_down(adapter);
3723                 e1000_free_irq(adapter);
3724         }
3725
3726         retval = pci_save_state(pdev);
3727         if (retval)
3728                 return retval;
3729
3730         status = er32(STATUS);
3731         if (status & E1000_STATUS_LU)
3732                 wufc &= ~E1000_WUFC_LNKC;
3733
3734         if (wufc) {
3735                 e1000_setup_rctl(adapter);
3736                 e1000_set_multi(netdev);
3737
3738                 /* turn on all-multi mode if wake on multicast is enabled */
3739                 if (wufc & E1000_WUFC_MC) {
3740                         rctl = er32(RCTL);
3741                         rctl |= E1000_RCTL_MPE;
3742                         ew32(RCTL, rctl);
3743                 }
3744
3745                 ctrl = er32(CTRL);
3746                 /* advertise wake from D3Cold */
3747                 #define E1000_CTRL_ADVD3WUC 0x00100000
3748                 /* phy power management enable */
3749                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
3750                 ctrl |= E1000_CTRL_ADVD3WUC |
3751                         E1000_CTRL_EN_PHY_PWR_MGMT;
3752                 ew32(CTRL, ctrl);
3753
3754                 if (adapter->hw.media_type == e1000_media_type_fiber ||
3755                    adapter->hw.media_type == e1000_media_type_internal_serdes) {
3756                         /* keep the laser running in D3 */
3757                         ctrl_ext = er32(CTRL_EXT);
3758                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
3759                         ew32(CTRL_EXT, ctrl_ext);
3760                 }
3761
3762                 /* Allow time for pending master requests to run */
3763                 e1000e_disable_pcie_master(&adapter->hw);
3764
3765                 ew32(WUC, E1000_WUC_PME_EN);
3766                 ew32(WUFC, wufc);
3767                 pci_enable_wake(pdev, PCI_D3hot, 1);
3768                 pci_enable_wake(pdev, PCI_D3cold, 1);
3769         } else {
3770                 ew32(WUC, 0);
3771                 ew32(WUFC, 0);
3772                 pci_enable_wake(pdev, PCI_D3hot, 0);
3773                 pci_enable_wake(pdev, PCI_D3cold, 0);
3774         }
3775
3776         e1000_release_manageability(adapter);
3777
3778         /* make sure adapter isn't asleep if manageability is enabled */
3779         if (adapter->flags & FLAG_MNG_PT_ENABLED) {
3780                 pci_enable_wake(pdev, PCI_D3hot, 1);
3781                 pci_enable_wake(pdev, PCI_D3cold, 1);
3782         }
3783
3784         if (adapter->hw.phy.type == e1000_phy_igp_3)
3785                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
3786
3787         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
3788          * would have already happened in close and is redundant. */
3789         e1000_release_hw_control(adapter);
3790
3791         pci_disable_device(pdev);
3792
3793         pci_set_power_state(pdev, pci_choose_state(pdev, state));
3794
3795         return 0;
3796 }
3797
3798 #ifdef CONFIG_PM
3799 static int e1000_resume(struct pci_dev *pdev)
3800 {
3801         struct net_device *netdev = pci_get_drvdata(pdev);
3802         struct e1000_adapter *adapter = netdev_priv(netdev);
3803         struct e1000_hw *hw = &adapter->hw;
3804         u32 err;
3805
3806         pci_set_power_state(pdev, PCI_D0);
3807         pci_restore_state(pdev);
3808         err = pci_enable_device(pdev);
3809         if (err) {
3810                 dev_err(&pdev->dev,
3811                         "Cannot enable PCI device from suspend\n");
3812                 return err;
3813         }
3814
3815         pci_set_master(pdev);
3816
3817         pci_enable_wake(pdev, PCI_D3hot, 0);
3818         pci_enable_wake(pdev, PCI_D3cold, 0);
3819
3820         if (netif_running(netdev)) {
3821                 err = e1000_request_irq(adapter);
3822                 if (err)
3823                         return err;
3824         }
3825
3826         e1000e_power_up_phy(adapter);
3827         e1000e_reset(adapter);
3828         ew32(WUS, ~0);
3829
3830         e1000_init_manageability(adapter);
3831
3832         if (netif_running(netdev))
3833                 e1000e_up(adapter);
3834
3835         netif_device_attach(netdev);
3836
3837         /* If the controller has AMT, do not set DRV_LOAD until the interface
3838          * is up.  For all other cases, let the f/w know that the h/w is now
3839          * under the control of the driver. */
3840         if (!(adapter->flags & FLAG_HAS_AMT) || !e1000e_check_mng_mode(&adapter->hw))
3841                 e1000_get_hw_control(adapter);
3842
3843         return 0;
3844 }
3845 #endif
3846
3847 static void e1000_shutdown(struct pci_dev *pdev)
3848 {
3849         e1000_suspend(pdev, PMSG_SUSPEND);
3850 }
3851
3852 #ifdef CONFIG_NET_POLL_CONTROLLER
3853 /*
3854  * Polling 'interrupt' - used by things like netconsole to send skbs
3855  * without having to re-enable interrupts. It's not called while
3856  * the interrupt routine is executing.
3857  */
3858 static void e1000_netpoll(struct net_device *netdev)
3859 {
3860         struct e1000_adapter *adapter = netdev_priv(netdev);
3861
3862         disable_irq(adapter->pdev->irq);
3863         e1000_intr(adapter->pdev->irq, netdev);
3864
3865         e1000_clean_tx_irq(adapter);
3866
3867         enable_irq(adapter->pdev->irq);
3868 }
3869 #endif
3870
3871 /**
3872  * e1000_io_error_detected - called when PCI error is detected
3873  * @pdev: Pointer to PCI device
3874  * @state: The current pci connection state
3875  *
3876  * This function is called after a PCI bus error affecting
3877  * this device has been detected.
3878  */
3879 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
3880                                                 pci_channel_state_t state)
3881 {
3882         struct net_device *netdev = pci_get_drvdata(pdev);
3883         struct e1000_adapter *adapter = netdev_priv(netdev);
3884
3885         netif_device_detach(netdev);
3886
3887         if (netif_running(netdev))
3888                 e1000e_down(adapter);
3889         pci_disable_device(pdev);
3890
3891         /* Request a slot slot reset. */
3892         return PCI_ERS_RESULT_NEED_RESET;
3893 }
3894
3895 /**
3896  * e1000_io_slot_reset - called after the pci bus has been reset.
3897  * @pdev: Pointer to PCI device
3898  *
3899  * Restart the card from scratch, as if from a cold-boot. Implementation
3900  * resembles the first-half of the e1000_resume routine.
3901  */
3902 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
3903 {
3904         struct net_device *netdev = pci_get_drvdata(pdev);
3905         struct e1000_adapter *adapter = netdev_priv(netdev);
3906         struct e1000_hw *hw = &adapter->hw;
3907
3908         if (pci_enable_device(pdev)) {
3909                 dev_err(&pdev->dev,
3910                         "Cannot re-enable PCI device after reset.\n");
3911                 return PCI_ERS_RESULT_DISCONNECT;
3912         }
3913         pci_set_master(pdev);
3914
3915         pci_enable_wake(pdev, PCI_D3hot, 0);
3916         pci_enable_wake(pdev, PCI_D3cold, 0);
3917
3918         e1000e_reset(adapter);
3919         ew32(WUS, ~0);
3920
3921         return PCI_ERS_RESULT_RECOVERED;
3922 }
3923
3924 /**
3925  * e1000_io_resume - called when traffic can start flowing again.
3926  * @pdev: Pointer to PCI device
3927  *
3928  * This callback is called when the error recovery driver tells us that
3929  * its OK to resume normal operation. Implementation resembles the
3930  * second-half of the e1000_resume routine.
3931  */
3932 static void e1000_io_resume(struct pci_dev *pdev)
3933 {
3934         struct net_device *netdev = pci_get_drvdata(pdev);
3935         struct e1000_adapter *adapter = netdev_priv(netdev);
3936
3937         e1000_init_manageability(adapter);
3938
3939         if (netif_running(netdev)) {
3940                 if (e1000e_up(adapter)) {
3941                         dev_err(&pdev->dev,
3942                                 "can't bring device back up after reset\n");
3943                         return;
3944                 }
3945         }
3946
3947         netif_device_attach(netdev);
3948
3949         /* If the controller has AMT, do not set DRV_LOAD until the interface
3950          * is up.  For all other cases, let the f/w know that the h/w is now
3951          * under the control of the driver. */
3952         if (!(adapter->flags & FLAG_HAS_AMT) ||
3953             !e1000e_check_mng_mode(&adapter->hw))
3954                 e1000_get_hw_control(adapter);
3955
3956 }
3957
3958 static void e1000_print_device_info(struct e1000_adapter *adapter)
3959 {
3960         struct e1000_hw *hw = &adapter->hw;
3961         struct net_device *netdev = adapter->netdev;
3962         u32 part_num;
3963
3964         /* print bus type/speed/width info */
3965         ndev_info(netdev, "(PCI Express:2.5GB/s:%s) "
3966                   "%02x:%02x:%02x:%02x:%02x:%02x\n",
3967                   /* bus width */
3968                  ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
3969                   "Width x1"),
3970                   /* MAC address */
3971                   netdev->dev_addr[0], netdev->dev_addr[1],
3972                   netdev->dev_addr[2], netdev->dev_addr[3],
3973                   netdev->dev_addr[4], netdev->dev_addr[5]);
3974         ndev_info(netdev, "Intel(R) PRO/%s Network Connection\n",
3975                   (hw->phy.type == e1000_phy_ife)
3976                    ? "10/100" : "1000");
3977         e1000e_read_part_num(hw, &part_num);
3978         ndev_info(netdev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
3979                   hw->mac.type, hw->phy.type,
3980                   (part_num >> 8), (part_num & 0xff));
3981 }
3982
3983 /**
3984  * e1000_probe - Device Initialization Routine
3985  * @pdev: PCI device information struct
3986  * @ent: entry in e1000_pci_tbl
3987  *
3988  * Returns 0 on success, negative on failure
3989  *
3990  * e1000_probe initializes an adapter identified by a pci_dev structure.
3991  * The OS initialization, configuring of the adapter private structure,
3992  * and a hardware reset occur.
3993  **/
3994 static int __devinit e1000_probe(struct pci_dev *pdev,
3995                                  const struct pci_device_id *ent)
3996 {
3997         struct net_device *netdev;
3998         struct e1000_adapter *adapter;
3999         struct e1000_hw *hw;
4000         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4001         unsigned long mmio_start, mmio_len;
4002         unsigned long flash_start, flash_len;
4003
4004         static int cards_found;
4005         int i, err, pci_using_dac;
4006         u16 eeprom_data = 0;
4007         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4008
4009         err = pci_enable_device(pdev);
4010         if (err)
4011                 return err;
4012
4013         pci_using_dac = 0;
4014         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
4015         if (!err) {
4016                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
4017                 if (!err)
4018                         pci_using_dac = 1;
4019         } else {
4020                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
4021                 if (err) {
4022                         err = pci_set_consistent_dma_mask(pdev,
4023                                                           DMA_32BIT_MASK);
4024                         if (err) {
4025                                 dev_err(&pdev->dev, "No usable DMA "
4026                                         "configuration, aborting\n");
4027                                 goto err_dma;
4028                         }
4029                 }
4030         }
4031
4032         err = pci_request_regions(pdev, e1000e_driver_name);
4033         if (err)
4034                 goto err_pci_reg;
4035
4036         pci_set_master(pdev);
4037
4038         err = -ENOMEM;
4039         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4040         if (!netdev)
4041                 goto err_alloc_etherdev;
4042
4043         SET_NETDEV_DEV(netdev, &pdev->dev);
4044
4045         pci_set_drvdata(pdev, netdev);
4046         adapter = netdev_priv(netdev);
4047         hw = &adapter->hw;
4048         adapter->netdev = netdev;
4049         adapter->pdev = pdev;
4050         adapter->ei = ei;
4051         adapter->pba = ei->pba;
4052         adapter->flags = ei->flags;
4053         adapter->hw.adapter = adapter;
4054         adapter->hw.mac.type = ei->mac;
4055         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4056
4057         mmio_start = pci_resource_start(pdev, 0);
4058         mmio_len = pci_resource_len(pdev, 0);
4059
4060         err = -EIO;
4061         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4062         if (!adapter->hw.hw_addr)
4063                 goto err_ioremap;
4064
4065         if ((adapter->flags & FLAG_HAS_FLASH) &&
4066             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
4067                 flash_start = pci_resource_start(pdev, 1);
4068                 flash_len = pci_resource_len(pdev, 1);
4069                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
4070                 if (!adapter->hw.flash_address)
4071                         goto err_flashmap;
4072         }
4073
4074         /* construct the net_device struct */
4075         netdev->open                    = &e1000_open;
4076         netdev->stop                    = &e1000_close;
4077         netdev->hard_start_xmit         = &e1000_xmit_frame;
4078         netdev->get_stats               = &e1000_get_stats;
4079         netdev->set_multicast_list      = &e1000_set_multi;
4080         netdev->set_mac_address         = &e1000_set_mac;
4081         netdev->change_mtu              = &e1000_change_mtu;
4082         netdev->do_ioctl                = &e1000_ioctl;
4083         e1000e_set_ethtool_ops(netdev);
4084         netdev->tx_timeout              = &e1000_tx_timeout;
4085         netdev->watchdog_timeo          = 5 * HZ;
4086         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
4087         netdev->vlan_rx_register        = e1000_vlan_rx_register;
4088         netdev->vlan_rx_add_vid         = e1000_vlan_rx_add_vid;
4089         netdev->vlan_rx_kill_vid        = e1000_vlan_rx_kill_vid;
4090 #ifdef CONFIG_NET_POLL_CONTROLLER
4091         netdev->poll_controller         = e1000_netpoll;
4092 #endif
4093         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
4094
4095         netdev->mem_start = mmio_start;
4096         netdev->mem_end = mmio_start + mmio_len;
4097
4098         adapter->bd_number = cards_found++;
4099
4100         /* setup adapter struct */
4101         err = e1000_sw_init(adapter);
4102         if (err)
4103                 goto err_sw_init;
4104
4105         err = -EIO;
4106
4107         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4108         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
4109         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4110
4111         err = ei->get_invariants(adapter);
4112         if (err)
4113                 goto err_hw_init;
4114
4115         hw->mac.ops.get_bus_info(&adapter->hw);
4116
4117         adapter->hw.phy.wait_for_link = 0;
4118
4119         /* Copper options */
4120         if (adapter->hw.media_type == e1000_media_type_copper) {
4121                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
4122                 adapter->hw.phy.disable_polarity_correction = 0;
4123                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
4124         }
4125
4126         if (e1000_check_reset_block(&adapter->hw))
4127                 ndev_info(netdev,
4128                           "PHY reset is blocked due to SOL/IDER session.\n");
4129
4130         netdev->features = NETIF_F_SG |
4131                            NETIF_F_HW_CSUM |
4132                            NETIF_F_HW_VLAN_TX |
4133                            NETIF_F_HW_VLAN_RX;
4134
4135         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
4136                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
4137
4138         netdev->features |= NETIF_F_TSO;
4139         netdev->features |= NETIF_F_TSO6;
4140
4141         if (pci_using_dac)
4142                 netdev->features |= NETIF_F_HIGHDMA;
4143
4144         /* We should not be using LLTX anymore, but we are still TX faster with
4145          * it. */
4146         netdev->features |= NETIF_F_LLTX;
4147
4148         if (e1000e_enable_mng_pass_thru(&adapter->hw))
4149                 adapter->flags |= FLAG_MNG_PT_ENABLED;
4150
4151         /* before reading the NVM, reset the controller to
4152          * put the device in a known good starting state */
4153         adapter->hw.mac.ops.reset_hw(&adapter->hw);
4154
4155         /*
4156          * systems with ASPM and others may see the checksum fail on the first
4157          * attempt. Let's give it a few tries
4158          */
4159         for (i = 0;; i++) {
4160                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
4161                         break;
4162                 if (i == 2) {
4163                         ndev_err(netdev, "The NVM Checksum Is Not Valid\n");
4164                         err = -EIO;
4165                         goto err_eeprom;
4166                 }
4167         }
4168
4169         /* copy the MAC address out of the NVM */
4170         if (e1000e_read_mac_addr(&adapter->hw))
4171                 ndev_err(netdev, "NVM Read Error while reading MAC address\n");
4172
4173         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
4174         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
4175
4176         if (!is_valid_ether_addr(netdev->perm_addr)) {
4177                 ndev_err(netdev, "Invalid MAC Address: "
4178                          "%02x:%02x:%02x:%02x:%02x:%02x\n",
4179                          netdev->perm_addr[0], netdev->perm_addr[1],
4180                          netdev->perm_addr[2], netdev->perm_addr[3],
4181                          netdev->perm_addr[4], netdev->perm_addr[5]);
4182                 err = -EIO;
4183                 goto err_eeprom;
4184         }
4185
4186         init_timer(&adapter->watchdog_timer);
4187         adapter->watchdog_timer.function = &e1000_watchdog;
4188         adapter->watchdog_timer.data = (unsigned long) adapter;
4189
4190         init_timer(&adapter->phy_info_timer);
4191         adapter->phy_info_timer.function = &e1000_update_phy_info;
4192         adapter->phy_info_timer.data = (unsigned long) adapter;
4193
4194         INIT_WORK(&adapter->reset_task, e1000_reset_task);
4195         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
4196
4197         e1000e_check_options(adapter);
4198
4199         /* Initialize link parameters. User can change them with ethtool */
4200         adapter->hw.mac.autoneg = 1;
4201         adapter->hw.mac.original_fc = e1000_fc_default;
4202         adapter->hw.mac.fc = e1000_fc_default;
4203         adapter->hw.phy.autoneg_advertised = 0x2f;
4204
4205         /* ring size defaults */
4206         adapter->rx_ring->count = 256;
4207         adapter->tx_ring->count = 256;
4208
4209         /*
4210          * Initial Wake on LAN setting - If APM wake is enabled in
4211          * the EEPROM, enable the ACPI Magic Packet filter
4212          */
4213         if (adapter->flags & FLAG_APME_IN_WUC) {
4214                 /* APME bit in EEPROM is mapped to WUC.APME */
4215                 eeprom_data = er32(WUC);
4216                 eeprom_apme_mask = E1000_WUC_APME;
4217         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
4218                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
4219                     (adapter->hw.bus.func == 1))
4220                         e1000_read_nvm(&adapter->hw,
4221                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
4222                 else
4223                         e1000_read_nvm(&adapter->hw,
4224                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
4225         }
4226
4227         /* fetch WoL from EEPROM */
4228         if (eeprom_data & eeprom_apme_mask)
4229                 adapter->eeprom_wol |= E1000_WUFC_MAG;
4230
4231         /*
4232          * now that we have the eeprom settings, apply the special cases
4233          * where the eeprom may be wrong or the board simply won't support
4234          * wake on lan on a particular port
4235          */
4236         if (!(adapter->flags & FLAG_HAS_WOL))
4237                 adapter->eeprom_wol = 0;
4238
4239         /* initialize the wol settings based on the eeprom settings */
4240         adapter->wol = adapter->eeprom_wol;
4241
4242         /* reset the hardware with the new settings */
4243         e1000e_reset(adapter);
4244
4245         /* If the controller has AMT, do not set DRV_LOAD until the interface
4246          * is up.  For all other cases, let the f/w know that the h/w is now
4247          * under the control of the driver. */
4248         if (!(adapter->flags & FLAG_HAS_AMT) ||
4249             !e1000e_check_mng_mode(&adapter->hw))
4250                 e1000_get_hw_control(adapter);
4251
4252         /* tell the stack to leave us alone until e1000_open() is called */
4253         netif_carrier_off(netdev);
4254         netif_stop_queue(netdev);
4255
4256         strcpy(netdev->name, "eth%d");
4257         err = register_netdev(netdev);
4258         if (err)
4259                 goto err_register;
4260
4261         e1000_print_device_info(adapter);
4262
4263         return 0;
4264
4265 err_register:
4266 err_hw_init:
4267         e1000_release_hw_control(adapter);
4268 err_eeprom:
4269         if (!e1000_check_reset_block(&adapter->hw))
4270                 e1000_phy_hw_reset(&adapter->hw);
4271
4272         if (adapter->hw.flash_address)
4273                 iounmap(adapter->hw.flash_address);
4274
4275 err_flashmap:
4276         kfree(adapter->tx_ring);
4277         kfree(adapter->rx_ring);
4278 err_sw_init:
4279         iounmap(adapter->hw.hw_addr);
4280 err_ioremap:
4281         free_netdev(netdev);
4282 err_alloc_etherdev:
4283         pci_release_regions(pdev);
4284 err_pci_reg:
4285 err_dma:
4286         pci_disable_device(pdev);
4287         return err;
4288 }
4289
4290 /**
4291  * e1000_remove - Device Removal Routine
4292  * @pdev: PCI device information struct
4293  *
4294  * e1000_remove is called by the PCI subsystem to alert the driver
4295  * that it should release a PCI device.  The could be caused by a
4296  * Hot-Plug event, or because the driver is going to be removed from
4297  * memory.
4298  **/
4299 static void __devexit e1000_remove(struct pci_dev *pdev)
4300 {
4301         struct net_device *netdev = pci_get_drvdata(pdev);
4302         struct e1000_adapter *adapter = netdev_priv(netdev);
4303
4304         /* flush_scheduled work may reschedule our watchdog task, so
4305          * explicitly disable watchdog tasks from being rescheduled  */
4306         set_bit(__E1000_DOWN, &adapter->state);
4307         del_timer_sync(&adapter->watchdog_timer);
4308         del_timer_sync(&adapter->phy_info_timer);
4309
4310         flush_scheduled_work();
4311
4312         e1000_release_manageability(adapter);
4313
4314         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
4315          * would have already happened in close and is redundant. */
4316         e1000_release_hw_control(adapter);
4317
4318         unregister_netdev(netdev);
4319
4320         if (!e1000_check_reset_block(&adapter->hw))
4321                 e1000_phy_hw_reset(&adapter->hw);
4322
4323         kfree(adapter->tx_ring);
4324         kfree(adapter->rx_ring);
4325
4326         iounmap(adapter->hw.hw_addr);
4327         if (adapter->hw.flash_address)
4328                 iounmap(adapter->hw.flash_address);
4329         pci_release_regions(pdev);
4330
4331         free_netdev(netdev);
4332
4333         pci_disable_device(pdev);
4334 }
4335
4336 /* PCI Error Recovery (ERS) */
4337 static struct pci_error_handlers e1000_err_handler = {
4338         .error_detected = e1000_io_error_detected,
4339         .slot_reset = e1000_io_slot_reset,
4340         .resume = e1000_io_resume,
4341 };
4342
4343 static struct pci_device_id e1000_pci_tbl[] = {
4344         /*
4345          * Support for 82571/2/3, es2lan and ich8 will be phased in
4346          * stepwise.
4347
4348         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
4349         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
4350         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
4351         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
4352         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
4353         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
4354         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
4355         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
4356         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
4357         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
4358         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
4359         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
4360         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
4361         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
4362           board_80003es2lan },
4363         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
4364           board_80003es2lan },
4365         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
4366           board_80003es2lan },
4367         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
4368           board_80003es2lan },
4369         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
4370         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
4371         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
4372         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
4373         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
4374         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
4375         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
4376         */
4377
4378         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
4379         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
4380         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
4381         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
4382         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
4383
4384         { }     /* terminate list */
4385 };
4386 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
4387
4388 /* PCI Device API Driver */
4389 static struct pci_driver e1000_driver = {
4390         .name     = e1000e_driver_name,
4391         .id_table = e1000_pci_tbl,
4392         .probe    = e1000_probe,
4393         .remove   = __devexit_p(e1000_remove),
4394 #ifdef CONFIG_PM
4395         /* Power Managment Hooks */
4396         .suspend  = e1000_suspend,
4397         .resume   = e1000_resume,
4398 #endif
4399         .shutdown = e1000_shutdown,
4400         .err_handler = &e1000_err_handler
4401 };
4402
4403 /**
4404  * e1000_init_module - Driver Registration Routine
4405  *
4406  * e1000_init_module is the first routine called when the driver is
4407  * loaded. All it does is register with the PCI subsystem.
4408  **/
4409 static int __init e1000_init_module(void)
4410 {
4411         int ret;
4412         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
4413                e1000e_driver_name, e1000e_driver_version);
4414         printk(KERN_INFO "%s: Copyright (c) 1999-2007 Intel Corporation.\n",
4415                e1000e_driver_name);
4416         ret = pci_register_driver(&e1000_driver);
4417
4418         return ret;
4419 }
4420 module_init(e1000_init_module);
4421
4422 /**
4423  * e1000_exit_module - Driver Exit Cleanup Routine
4424  *
4425  * e1000_exit_module is called just before the driver is removed
4426  * from memory.
4427  **/
4428 static void __exit e1000_exit_module(void)
4429 {
4430         pci_unregister_driver(&e1000_driver);
4431 }
4432 module_exit(e1000_exit_module);
4433
4434
4435 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
4436 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
4437 MODULE_LICENSE("GPL");
4438 MODULE_VERSION(DRV_VERSION);
4439
4440 /* e1000_main.c */