Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6] / drivers / net / e1000e / netdev.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2008 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #include <linux/module.h>
30 #include <linux/types.h>
31 #include <linux/init.h>
32 #include <linux/pci.h>
33 #include <linux/vmalloc.h>
34 #include <linux/pagemap.h>
35 #include <linux/delay.h>
36 #include <linux/netdevice.h>
37 #include <linux/tcp.h>
38 #include <linux/ipv6.h>
39 #include <net/checksum.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/mii.h>
42 #include <linux/ethtool.h>
43 #include <linux/if_vlan.h>
44 #include <linux/cpu.h>
45 #include <linux/smp.h>
46 #include <linux/pm_qos_params.h>
47
48 #include "e1000.h"
49
50 #define DRV_VERSION "0.3.3.3-k2"
51 char e1000e_driver_name[] = "e1000e";
52 const char e1000e_driver_version[] = DRV_VERSION;
53
54 static const struct e1000_info *e1000_info_tbl[] = {
55         [board_82571]           = &e1000_82571_info,
56         [board_82572]           = &e1000_82572_info,
57         [board_82573]           = &e1000_82573_info,
58         [board_80003es2lan]     = &e1000_es2_info,
59         [board_ich8lan]         = &e1000_ich8_info,
60         [board_ich9lan]         = &e1000_ich9_info,
61 };
62
63 #ifdef DEBUG
64 /**
65  * e1000_get_hw_dev_name - return device name string
66  * used by hardware layer to print debugging information
67  **/
68 char *e1000e_get_hw_dev_name(struct e1000_hw *hw)
69 {
70         return hw->adapter->netdev->name;
71 }
72 #endif
73
74 /**
75  * e1000_desc_unused - calculate if we have unused descriptors
76  **/
77 static int e1000_desc_unused(struct e1000_ring *ring)
78 {
79         if (ring->next_to_clean > ring->next_to_use)
80                 return ring->next_to_clean - ring->next_to_use - 1;
81
82         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
83 }
84
85 /**
86  * e1000_receive_skb - helper function to handle Rx indications
87  * @adapter: board private structure
88  * @status: descriptor status field as written by hardware
89  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
90  * @skb: pointer to sk_buff to be indicated to stack
91  **/
92 static void e1000_receive_skb(struct e1000_adapter *adapter,
93                               struct net_device *netdev,
94                               struct sk_buff *skb,
95                               u8 status, __le16 vlan)
96 {
97         skb->protocol = eth_type_trans(skb, netdev);
98
99         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
100                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
101                                          le16_to_cpu(vlan));
102         else
103                 netif_receive_skb(skb);
104
105         netdev->last_rx = jiffies;
106 }
107
108 /**
109  * e1000_rx_checksum - Receive Checksum Offload for 82543
110  * @adapter:     board private structure
111  * @status_err:  receive descriptor status and error fields
112  * @csum:       receive descriptor csum field
113  * @sk_buff:     socket buffer with received data
114  **/
115 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
116                               u32 csum, struct sk_buff *skb)
117 {
118         u16 status = (u16)status_err;
119         u8 errors = (u8)(status_err >> 24);
120         skb->ip_summed = CHECKSUM_NONE;
121
122         /* Ignore Checksum bit is set */
123         if (status & E1000_RXD_STAT_IXSM)
124                 return;
125         /* TCP/UDP checksum error bit is set */
126         if (errors & E1000_RXD_ERR_TCPE) {
127                 /* let the stack verify checksum errors */
128                 adapter->hw_csum_err++;
129                 return;
130         }
131
132         /* TCP/UDP Checksum has not been calculated */
133         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
134                 return;
135
136         /* It must be a TCP or UDP packet with a valid checksum */
137         if (status & E1000_RXD_STAT_TCPCS) {
138                 /* TCP checksum is good */
139                 skb->ip_summed = CHECKSUM_UNNECESSARY;
140         } else {
141                 /*
142                  * IP fragment with UDP payload
143                  * Hardware complements the payload checksum, so we undo it
144                  * and then put the value in host order for further stack use.
145                  */
146                 __sum16 sum = (__force __sum16)htons(csum);
147                 skb->csum = csum_unfold(~sum);
148                 skb->ip_summed = CHECKSUM_COMPLETE;
149         }
150         adapter->hw_csum_good++;
151 }
152
153 /**
154  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
155  * @adapter: address of board private structure
156  **/
157 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
158                                    int cleaned_count)
159 {
160         struct net_device *netdev = adapter->netdev;
161         struct pci_dev *pdev = adapter->pdev;
162         struct e1000_ring *rx_ring = adapter->rx_ring;
163         struct e1000_rx_desc *rx_desc;
164         struct e1000_buffer *buffer_info;
165         struct sk_buff *skb;
166         unsigned int i;
167         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
168
169         i = rx_ring->next_to_use;
170         buffer_info = &rx_ring->buffer_info[i];
171
172         while (cleaned_count--) {
173                 skb = buffer_info->skb;
174                 if (skb) {
175                         skb_trim(skb, 0);
176                         goto map_skb;
177                 }
178
179                 skb = netdev_alloc_skb(netdev, bufsz);
180                 if (!skb) {
181                         /* Better luck next round */
182                         adapter->alloc_rx_buff_failed++;
183                         break;
184                 }
185
186                 /*
187                  * Make buffer alignment 2 beyond a 16 byte boundary
188                  * this will result in a 16 byte aligned IP header after
189                  * the 14 byte MAC header is removed
190                  */
191                 skb_reserve(skb, NET_IP_ALIGN);
192
193                 buffer_info->skb = skb;
194 map_skb:
195                 buffer_info->dma = pci_map_single(pdev, skb->data,
196                                                   adapter->rx_buffer_len,
197                                                   PCI_DMA_FROMDEVICE);
198                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
199                         dev_err(&pdev->dev, "RX DMA map failed\n");
200                         adapter->rx_dma_failed++;
201                         break;
202                 }
203
204                 rx_desc = E1000_RX_DESC(*rx_ring, i);
205                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
206
207                 i++;
208                 if (i == rx_ring->count)
209                         i = 0;
210                 buffer_info = &rx_ring->buffer_info[i];
211         }
212
213         if (rx_ring->next_to_use != i) {
214                 rx_ring->next_to_use = i;
215                 if (i-- == 0)
216                         i = (rx_ring->count - 1);
217
218                 /*
219                  * Force memory writes to complete before letting h/w
220                  * know there are new descriptors to fetch.  (Only
221                  * applicable for weak-ordered memory model archs,
222                  * such as IA-64).
223                  */
224                 wmb();
225                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
226         }
227 }
228
229 /**
230  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
231  * @adapter: address of board private structure
232  **/
233 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
234                                       int cleaned_count)
235 {
236         struct net_device *netdev = adapter->netdev;
237         struct pci_dev *pdev = adapter->pdev;
238         union e1000_rx_desc_packet_split *rx_desc;
239         struct e1000_ring *rx_ring = adapter->rx_ring;
240         struct e1000_buffer *buffer_info;
241         struct e1000_ps_page *ps_page;
242         struct sk_buff *skb;
243         unsigned int i, j;
244
245         i = rx_ring->next_to_use;
246         buffer_info = &rx_ring->buffer_info[i];
247
248         while (cleaned_count--) {
249                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
250
251                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
252                         ps_page = &buffer_info->ps_pages[j];
253                         if (j >= adapter->rx_ps_pages) {
254                                 /* all unused desc entries get hw null ptr */
255                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
256                                 continue;
257                         }
258                         if (!ps_page->page) {
259                                 ps_page->page = alloc_page(GFP_ATOMIC);
260                                 if (!ps_page->page) {
261                                         adapter->alloc_rx_buff_failed++;
262                                         goto no_buffers;
263                                 }
264                                 ps_page->dma = pci_map_page(pdev,
265                                                    ps_page->page,
266                                                    0, PAGE_SIZE,
267                                                    PCI_DMA_FROMDEVICE);
268                                 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
269                                         dev_err(&adapter->pdev->dev,
270                                           "RX DMA page map failed\n");
271                                         adapter->rx_dma_failed++;
272                                         goto no_buffers;
273                                 }
274                         }
275                         /*
276                          * Refresh the desc even if buffer_addrs
277                          * didn't change because each write-back
278                          * erases this info.
279                          */
280                         rx_desc->read.buffer_addr[j+1] =
281                              cpu_to_le64(ps_page->dma);
282                 }
283
284                 skb = netdev_alloc_skb(netdev,
285                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
286
287                 if (!skb) {
288                         adapter->alloc_rx_buff_failed++;
289                         break;
290                 }
291
292                 /*
293                  * Make buffer alignment 2 beyond a 16 byte boundary
294                  * this will result in a 16 byte aligned IP header after
295                  * the 14 byte MAC header is removed
296                  */
297                 skb_reserve(skb, NET_IP_ALIGN);
298
299                 buffer_info->skb = skb;
300                 buffer_info->dma = pci_map_single(pdev, skb->data,
301                                                   adapter->rx_ps_bsize0,
302                                                   PCI_DMA_FROMDEVICE);
303                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
304                         dev_err(&pdev->dev, "RX DMA map failed\n");
305                         adapter->rx_dma_failed++;
306                         /* cleanup skb */
307                         dev_kfree_skb_any(skb);
308                         buffer_info->skb = NULL;
309                         break;
310                 }
311
312                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
313
314                 i++;
315                 if (i == rx_ring->count)
316                         i = 0;
317                 buffer_info = &rx_ring->buffer_info[i];
318         }
319
320 no_buffers:
321         if (rx_ring->next_to_use != i) {
322                 rx_ring->next_to_use = i;
323
324                 if (!(i--))
325                         i = (rx_ring->count - 1);
326
327                 /*
328                  * Force memory writes to complete before letting h/w
329                  * know there are new descriptors to fetch.  (Only
330                  * applicable for weak-ordered memory model archs,
331                  * such as IA-64).
332                  */
333                 wmb();
334                 /*
335                  * Hardware increments by 16 bytes, but packet split
336                  * descriptors are 32 bytes...so we increment tail
337                  * twice as much.
338                  */
339                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
340         }
341 }
342
343 /**
344  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
345  * @adapter: address of board private structure
346  * @rx_ring: pointer to receive ring structure
347  * @cleaned_count: number of buffers to allocate this pass
348  **/
349
350 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
351                                          int cleaned_count)
352 {
353         struct net_device *netdev = adapter->netdev;
354         struct pci_dev *pdev = adapter->pdev;
355         struct e1000_rx_desc *rx_desc;
356         struct e1000_ring *rx_ring = adapter->rx_ring;
357         struct e1000_buffer *buffer_info;
358         struct sk_buff *skb;
359         unsigned int i;
360         unsigned int bufsz = 256 -
361                              16 /* for skb_reserve */ -
362                              NET_IP_ALIGN;
363
364         i = rx_ring->next_to_use;
365         buffer_info = &rx_ring->buffer_info[i];
366
367         while (cleaned_count--) {
368                 skb = buffer_info->skb;
369                 if (skb) {
370                         skb_trim(skb, 0);
371                         goto check_page;
372                 }
373
374                 skb = netdev_alloc_skb(netdev, bufsz);
375                 if (unlikely(!skb)) {
376                         /* Better luck next round */
377                         adapter->alloc_rx_buff_failed++;
378                         break;
379                 }
380
381                 /* Make buffer alignment 2 beyond a 16 byte boundary
382                  * this will result in a 16 byte aligned IP header after
383                  * the 14 byte MAC header is removed
384                  */
385                 skb_reserve(skb, NET_IP_ALIGN);
386
387                 buffer_info->skb = skb;
388 check_page:
389                 /* allocate a new page if necessary */
390                 if (!buffer_info->page) {
391                         buffer_info->page = alloc_page(GFP_ATOMIC);
392                         if (unlikely(!buffer_info->page)) {
393                                 adapter->alloc_rx_buff_failed++;
394                                 break;
395                         }
396                 }
397
398                 if (!buffer_info->dma)
399                         buffer_info->dma = pci_map_page(pdev,
400                                                         buffer_info->page, 0,
401                                                         PAGE_SIZE,
402                                                         PCI_DMA_FROMDEVICE);
403
404                 rx_desc = E1000_RX_DESC(*rx_ring, i);
405                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
406
407                 if (unlikely(++i == rx_ring->count))
408                         i = 0;
409                 buffer_info = &rx_ring->buffer_info[i];
410         }
411
412         if (likely(rx_ring->next_to_use != i)) {
413                 rx_ring->next_to_use = i;
414                 if (unlikely(i-- == 0))
415                         i = (rx_ring->count - 1);
416
417                 /* Force memory writes to complete before letting h/w
418                  * know there are new descriptors to fetch.  (Only
419                  * applicable for weak-ordered memory model archs,
420                  * such as IA-64). */
421                 wmb();
422                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
423         }
424 }
425
426 /**
427  * e1000_clean_rx_irq - Send received data up the network stack; legacy
428  * @adapter: board private structure
429  *
430  * the return value indicates whether actual cleaning was done, there
431  * is no guarantee that everything was cleaned
432  **/
433 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
434                                int *work_done, int work_to_do)
435 {
436         struct net_device *netdev = adapter->netdev;
437         struct pci_dev *pdev = adapter->pdev;
438         struct e1000_ring *rx_ring = adapter->rx_ring;
439         struct e1000_rx_desc *rx_desc, *next_rxd;
440         struct e1000_buffer *buffer_info, *next_buffer;
441         u32 length;
442         unsigned int i;
443         int cleaned_count = 0;
444         bool cleaned = 0;
445         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
446
447         i = rx_ring->next_to_clean;
448         rx_desc = E1000_RX_DESC(*rx_ring, i);
449         buffer_info = &rx_ring->buffer_info[i];
450
451         while (rx_desc->status & E1000_RXD_STAT_DD) {
452                 struct sk_buff *skb;
453                 u8 status;
454
455                 if (*work_done >= work_to_do)
456                         break;
457                 (*work_done)++;
458
459                 status = rx_desc->status;
460                 skb = buffer_info->skb;
461                 buffer_info->skb = NULL;
462
463                 prefetch(skb->data - NET_IP_ALIGN);
464
465                 i++;
466                 if (i == rx_ring->count)
467                         i = 0;
468                 next_rxd = E1000_RX_DESC(*rx_ring, i);
469                 prefetch(next_rxd);
470
471                 next_buffer = &rx_ring->buffer_info[i];
472
473                 cleaned = 1;
474                 cleaned_count++;
475                 pci_unmap_single(pdev,
476                                  buffer_info->dma,
477                                  adapter->rx_buffer_len,
478                                  PCI_DMA_FROMDEVICE);
479                 buffer_info->dma = 0;
480
481                 length = le16_to_cpu(rx_desc->length);
482
483                 /* !EOP means multiple descriptors were used to store a single
484                  * packet, also make sure the frame isn't just CRC only */
485                 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
486                         /* All receives must fit into a single buffer */
487                         e_dbg("%s: Receive packet consumed multiple buffers\n",
488                               netdev->name);
489                         /* recycle */
490                         buffer_info->skb = skb;
491                         goto next_desc;
492                 }
493
494                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
495                         /* recycle */
496                         buffer_info->skb = skb;
497                         goto next_desc;
498                 }
499
500                 total_rx_bytes += length;
501                 total_rx_packets++;
502
503                 /*
504                  * code added for copybreak, this should improve
505                  * performance for small packets with large amounts
506                  * of reassembly being done in the stack
507                  */
508                 if (length < copybreak) {
509                         struct sk_buff *new_skb =
510                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
511                         if (new_skb) {
512                                 skb_reserve(new_skb, NET_IP_ALIGN);
513                                 skb_copy_to_linear_data_offset(new_skb,
514                                                                -NET_IP_ALIGN,
515                                                                (skb->data -
516                                                                 NET_IP_ALIGN),
517                                                                (length +
518                                                                 NET_IP_ALIGN));
519                                 /* save the skb in buffer_info as good */
520                                 buffer_info->skb = skb;
521                                 skb = new_skb;
522                         }
523                         /* else just continue with the old one */
524                 }
525                 /* end copybreak code */
526                 skb_put(skb, length);
527
528                 /* Receive Checksum Offload */
529                 e1000_rx_checksum(adapter,
530                                   (u32)(status) |
531                                   ((u32)(rx_desc->errors) << 24),
532                                   le16_to_cpu(rx_desc->csum), skb);
533
534                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
535
536 next_desc:
537                 rx_desc->status = 0;
538
539                 /* return some buffers to hardware, one at a time is too slow */
540                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
541                         adapter->alloc_rx_buf(adapter, cleaned_count);
542                         cleaned_count = 0;
543                 }
544
545                 /* use prefetched values */
546                 rx_desc = next_rxd;
547                 buffer_info = next_buffer;
548         }
549         rx_ring->next_to_clean = i;
550
551         cleaned_count = e1000_desc_unused(rx_ring);
552         if (cleaned_count)
553                 adapter->alloc_rx_buf(adapter, cleaned_count);
554
555         adapter->total_rx_bytes += total_rx_bytes;
556         adapter->total_rx_packets += total_rx_packets;
557         adapter->net_stats.rx_bytes += total_rx_bytes;
558         adapter->net_stats.rx_packets += total_rx_packets;
559         return cleaned;
560 }
561
562 static void e1000_put_txbuf(struct e1000_adapter *adapter,
563                              struct e1000_buffer *buffer_info)
564 {
565         if (buffer_info->dma) {
566                 pci_unmap_page(adapter->pdev, buffer_info->dma,
567                                buffer_info->length, PCI_DMA_TODEVICE);
568                 buffer_info->dma = 0;
569         }
570         if (buffer_info->skb) {
571                 dev_kfree_skb_any(buffer_info->skb);
572                 buffer_info->skb = NULL;
573         }
574 }
575
576 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
577 {
578         struct e1000_ring *tx_ring = adapter->tx_ring;
579         unsigned int i = tx_ring->next_to_clean;
580         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
581         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
582
583         /* detected Tx unit hang */
584         e_err("Detected Tx Unit Hang:\n"
585               "  TDH                  <%x>\n"
586               "  TDT                  <%x>\n"
587               "  next_to_use          <%x>\n"
588               "  next_to_clean        <%x>\n"
589               "buffer_info[next_to_clean]:\n"
590               "  time_stamp           <%lx>\n"
591               "  next_to_watch        <%x>\n"
592               "  jiffies              <%lx>\n"
593               "  next_to_watch.status <%x>\n",
594               readl(adapter->hw.hw_addr + tx_ring->head),
595               readl(adapter->hw.hw_addr + tx_ring->tail),
596               tx_ring->next_to_use,
597               tx_ring->next_to_clean,
598               tx_ring->buffer_info[eop].time_stamp,
599               eop,
600               jiffies,
601               eop_desc->upper.fields.status);
602 }
603
604 /**
605  * e1000_clean_tx_irq - Reclaim resources after transmit completes
606  * @adapter: board private structure
607  *
608  * the return value indicates whether actual cleaning was done, there
609  * is no guarantee that everything was cleaned
610  **/
611 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
612 {
613         struct net_device *netdev = adapter->netdev;
614         struct e1000_hw *hw = &adapter->hw;
615         struct e1000_ring *tx_ring = adapter->tx_ring;
616         struct e1000_tx_desc *tx_desc, *eop_desc;
617         struct e1000_buffer *buffer_info;
618         unsigned int i, eop;
619         unsigned int count = 0;
620         bool cleaned = 0;
621         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
622
623         i = tx_ring->next_to_clean;
624         eop = tx_ring->buffer_info[i].next_to_watch;
625         eop_desc = E1000_TX_DESC(*tx_ring, eop);
626
627         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
628                 for (cleaned = 0; !cleaned; ) {
629                         tx_desc = E1000_TX_DESC(*tx_ring, i);
630                         buffer_info = &tx_ring->buffer_info[i];
631                         cleaned = (i == eop);
632
633                         if (cleaned) {
634                                 struct sk_buff *skb = buffer_info->skb;
635                                 unsigned int segs, bytecount;
636                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
637                                 /* multiply data chunks by size of headers */
638                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
639                                             skb->len;
640                                 total_tx_packets += segs;
641                                 total_tx_bytes += bytecount;
642                         }
643
644                         e1000_put_txbuf(adapter, buffer_info);
645                         tx_desc->upper.data = 0;
646
647                         i++;
648                         if (i == tx_ring->count)
649                                 i = 0;
650                 }
651
652                 eop = tx_ring->buffer_info[i].next_to_watch;
653                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
654 #define E1000_TX_WEIGHT 64
655                 /* weight of a sort for tx, to avoid endless transmit cleanup */
656                 if (count++ == E1000_TX_WEIGHT)
657                         break;
658         }
659
660         tx_ring->next_to_clean = i;
661
662 #define TX_WAKE_THRESHOLD 32
663         if (cleaned && netif_carrier_ok(netdev) &&
664                      e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
665                 /* Make sure that anybody stopping the queue after this
666                  * sees the new next_to_clean.
667                  */
668                 smp_mb();
669
670                 if (netif_queue_stopped(netdev) &&
671                     !(test_bit(__E1000_DOWN, &adapter->state))) {
672                         netif_wake_queue(netdev);
673                         ++adapter->restart_queue;
674                 }
675         }
676
677         if (adapter->detect_tx_hung) {
678                 /*
679                  * Detect a transmit hang in hardware, this serializes the
680                  * check with the clearing of time_stamp and movement of i
681                  */
682                 adapter->detect_tx_hung = 0;
683                 if (tx_ring->buffer_info[eop].dma &&
684                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp
685                                + (adapter->tx_timeout_factor * HZ))
686                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
687                         e1000_print_tx_hang(adapter);
688                         netif_stop_queue(netdev);
689                 }
690         }
691         adapter->total_tx_bytes += total_tx_bytes;
692         adapter->total_tx_packets += total_tx_packets;
693         adapter->net_stats.tx_bytes += total_tx_bytes;
694         adapter->net_stats.tx_packets += total_tx_packets;
695         return cleaned;
696 }
697
698 /**
699  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
700  * @adapter: board private structure
701  *
702  * the return value indicates whether actual cleaning was done, there
703  * is no guarantee that everything was cleaned
704  **/
705 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
706                                   int *work_done, int work_to_do)
707 {
708         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
709         struct net_device *netdev = adapter->netdev;
710         struct pci_dev *pdev = adapter->pdev;
711         struct e1000_ring *rx_ring = adapter->rx_ring;
712         struct e1000_buffer *buffer_info, *next_buffer;
713         struct e1000_ps_page *ps_page;
714         struct sk_buff *skb;
715         unsigned int i, j;
716         u32 length, staterr;
717         int cleaned_count = 0;
718         bool cleaned = 0;
719         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
720
721         i = rx_ring->next_to_clean;
722         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
723         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
724         buffer_info = &rx_ring->buffer_info[i];
725
726         while (staterr & E1000_RXD_STAT_DD) {
727                 if (*work_done >= work_to_do)
728                         break;
729                 (*work_done)++;
730                 skb = buffer_info->skb;
731
732                 /* in the packet split case this is header only */
733                 prefetch(skb->data - NET_IP_ALIGN);
734
735                 i++;
736                 if (i == rx_ring->count)
737                         i = 0;
738                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
739                 prefetch(next_rxd);
740
741                 next_buffer = &rx_ring->buffer_info[i];
742
743                 cleaned = 1;
744                 cleaned_count++;
745                 pci_unmap_single(pdev, buffer_info->dma,
746                                  adapter->rx_ps_bsize0,
747                                  PCI_DMA_FROMDEVICE);
748                 buffer_info->dma = 0;
749
750                 if (!(staterr & E1000_RXD_STAT_EOP)) {
751                         e_dbg("%s: Packet Split buffers didn't pick up the "
752                               "full packet\n", netdev->name);
753                         dev_kfree_skb_irq(skb);
754                         goto next_desc;
755                 }
756
757                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
758                         dev_kfree_skb_irq(skb);
759                         goto next_desc;
760                 }
761
762                 length = le16_to_cpu(rx_desc->wb.middle.length0);
763
764                 if (!length) {
765                         e_dbg("%s: Last part of the packet spanning multiple "
766                               "descriptors\n", netdev->name);
767                         dev_kfree_skb_irq(skb);
768                         goto next_desc;
769                 }
770
771                 /* Good Receive */
772                 skb_put(skb, length);
773
774                 {
775                 /*
776                  * this looks ugly, but it seems compiler issues make it
777                  * more efficient than reusing j
778                  */
779                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
780
781                 /*
782                  * page alloc/put takes too long and effects small packet
783                  * throughput, so unsplit small packets and save the alloc/put
784                  * only valid in softirq (napi) context to call kmap_*
785                  */
786                 if (l1 && (l1 <= copybreak) &&
787                     ((length + l1) <= adapter->rx_ps_bsize0)) {
788                         u8 *vaddr;
789
790                         ps_page = &buffer_info->ps_pages[0];
791
792                         /*
793                          * there is no documentation about how to call
794                          * kmap_atomic, so we can't hold the mapping
795                          * very long
796                          */
797                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
798                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
799                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
800                         memcpy(skb_tail_pointer(skb), vaddr, l1);
801                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
802                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
803                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
804
805                         skb_put(skb, l1);
806                         goto copydone;
807                 } /* if */
808                 }
809
810                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
811                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
812                         if (!length)
813                                 break;
814
815                         ps_page = &buffer_info->ps_pages[j];
816                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
817                                        PCI_DMA_FROMDEVICE);
818                         ps_page->dma = 0;
819                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
820                         ps_page->page = NULL;
821                         skb->len += length;
822                         skb->data_len += length;
823                         skb->truesize += length;
824                 }
825
826 copydone:
827                 total_rx_bytes += skb->len;
828                 total_rx_packets++;
829
830                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
831                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
832
833                 if (rx_desc->wb.upper.header_status &
834                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
835                         adapter->rx_hdr_split++;
836
837                 e1000_receive_skb(adapter, netdev, skb,
838                                   staterr, rx_desc->wb.middle.vlan);
839
840 next_desc:
841                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
842                 buffer_info->skb = NULL;
843
844                 /* return some buffers to hardware, one at a time is too slow */
845                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
846                         adapter->alloc_rx_buf(adapter, cleaned_count);
847                         cleaned_count = 0;
848                 }
849
850                 /* use prefetched values */
851                 rx_desc = next_rxd;
852                 buffer_info = next_buffer;
853
854                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
855         }
856         rx_ring->next_to_clean = i;
857
858         cleaned_count = e1000_desc_unused(rx_ring);
859         if (cleaned_count)
860                 adapter->alloc_rx_buf(adapter, cleaned_count);
861
862         adapter->total_rx_bytes += total_rx_bytes;
863         adapter->total_rx_packets += total_rx_packets;
864         adapter->net_stats.rx_bytes += total_rx_bytes;
865         adapter->net_stats.rx_packets += total_rx_packets;
866         return cleaned;
867 }
868
869 /**
870  * e1000_consume_page - helper function
871  **/
872 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
873                                u16 length)
874 {
875         bi->page = NULL;
876         skb->len += length;
877         skb->data_len += length;
878         skb->truesize += length;
879 }
880
881 /**
882  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
883  * @adapter: board private structure
884  *
885  * the return value indicates whether actual cleaning was done, there
886  * is no guarantee that everything was cleaned
887  **/
888
889 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
890                                      int *work_done, int work_to_do)
891 {
892         struct net_device *netdev = adapter->netdev;
893         struct pci_dev *pdev = adapter->pdev;
894         struct e1000_ring *rx_ring = adapter->rx_ring;
895         struct e1000_rx_desc *rx_desc, *next_rxd;
896         struct e1000_buffer *buffer_info, *next_buffer;
897         u32 length;
898         unsigned int i;
899         int cleaned_count = 0;
900         bool cleaned = false;
901         unsigned int total_rx_bytes=0, total_rx_packets=0;
902
903         i = rx_ring->next_to_clean;
904         rx_desc = E1000_RX_DESC(*rx_ring, i);
905         buffer_info = &rx_ring->buffer_info[i];
906
907         while (rx_desc->status & E1000_RXD_STAT_DD) {
908                 struct sk_buff *skb;
909                 u8 status;
910
911                 if (*work_done >= work_to_do)
912                         break;
913                 (*work_done)++;
914
915                 status = rx_desc->status;
916                 skb = buffer_info->skb;
917                 buffer_info->skb = NULL;
918
919                 ++i;
920                 if (i == rx_ring->count)
921                         i = 0;
922                 next_rxd = E1000_RX_DESC(*rx_ring, i);
923                 prefetch(next_rxd);
924
925                 next_buffer = &rx_ring->buffer_info[i];
926
927                 cleaned = true;
928                 cleaned_count++;
929                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
930                                PCI_DMA_FROMDEVICE);
931                 buffer_info->dma = 0;
932
933                 length = le16_to_cpu(rx_desc->length);
934
935                 /* errors is only valid for DD + EOP descriptors */
936                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
937                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
938                                 /* recycle both page and skb */
939                                 buffer_info->skb = skb;
940                                 /* an error means any chain goes out the window
941                                  * too */
942                                 if (rx_ring->rx_skb_top)
943                                         dev_kfree_skb(rx_ring->rx_skb_top);
944                                 rx_ring->rx_skb_top = NULL;
945                                 goto next_desc;
946                 }
947
948 #define rxtop rx_ring->rx_skb_top
949                 if (!(status & E1000_RXD_STAT_EOP)) {
950                         /* this descriptor is only the beginning (or middle) */
951                         if (!rxtop) {
952                                 /* this is the beginning of a chain */
953                                 rxtop = skb;
954                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
955                                                    0, length);
956                         } else {
957                                 /* this is the middle of a chain */
958                                 skb_fill_page_desc(rxtop,
959                                     skb_shinfo(rxtop)->nr_frags,
960                                     buffer_info->page, 0, length);
961                                 /* re-use the skb, only consumed the page */
962                                 buffer_info->skb = skb;
963                         }
964                         e1000_consume_page(buffer_info, rxtop, length);
965                         goto next_desc;
966                 } else {
967                         if (rxtop) {
968                                 /* end of the chain */
969                                 skb_fill_page_desc(rxtop,
970                                     skb_shinfo(rxtop)->nr_frags,
971                                     buffer_info->page, 0, length);
972                                 /* re-use the current skb, we only consumed the
973                                  * page */
974                                 buffer_info->skb = skb;
975                                 skb = rxtop;
976                                 rxtop = NULL;
977                                 e1000_consume_page(buffer_info, skb, length);
978                         } else {
979                                 /* no chain, got EOP, this buf is the packet
980                                  * copybreak to save the put_page/alloc_page */
981                                 if (length <= copybreak &&
982                                     skb_tailroom(skb) >= length) {
983                                         u8 *vaddr;
984                                         vaddr = kmap_atomic(buffer_info->page,
985                                                            KM_SKB_DATA_SOFTIRQ);
986                                         memcpy(skb_tail_pointer(skb), vaddr,
987                                                length);
988                                         kunmap_atomic(vaddr,
989                                                       KM_SKB_DATA_SOFTIRQ);
990                                         /* re-use the page, so don't erase
991                                          * buffer_info->page */
992                                         skb_put(skb, length);
993                                 } else {
994                                         skb_fill_page_desc(skb, 0,
995                                                            buffer_info->page, 0,
996                                                            length);
997                                         e1000_consume_page(buffer_info, skb,
998                                                            length);
999                                 }
1000                         }
1001                 }
1002
1003                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1004                 e1000_rx_checksum(adapter,
1005                                   (u32)(status) |
1006                                   ((u32)(rx_desc->errors) << 24),
1007                                   le16_to_cpu(rx_desc->csum), skb);
1008
1009                 /* probably a little skewed due to removing CRC */
1010                 total_rx_bytes += skb->len;
1011                 total_rx_packets++;
1012
1013                 /* eth type trans needs skb->data to point to something */
1014                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1015                         e_err("pskb_may_pull failed.\n");
1016                         dev_kfree_skb(skb);
1017                         goto next_desc;
1018                 }
1019
1020                 e1000_receive_skb(adapter, netdev, skb, status,
1021                                   rx_desc->special);
1022
1023 next_desc:
1024                 rx_desc->status = 0;
1025
1026                 /* return some buffers to hardware, one at a time is too slow */
1027                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1028                         adapter->alloc_rx_buf(adapter, cleaned_count);
1029                         cleaned_count = 0;
1030                 }
1031
1032                 /* use prefetched values */
1033                 rx_desc = next_rxd;
1034                 buffer_info = next_buffer;
1035         }
1036         rx_ring->next_to_clean = i;
1037
1038         cleaned_count = e1000_desc_unused(rx_ring);
1039         if (cleaned_count)
1040                 adapter->alloc_rx_buf(adapter, cleaned_count);
1041
1042         adapter->total_rx_bytes += total_rx_bytes;
1043         adapter->total_rx_packets += total_rx_packets;
1044         adapter->net_stats.rx_bytes += total_rx_bytes;
1045         adapter->net_stats.rx_packets += total_rx_packets;
1046         return cleaned;
1047 }
1048
1049 /**
1050  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1051  * @adapter: board private structure
1052  **/
1053 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1054 {
1055         struct e1000_ring *rx_ring = adapter->rx_ring;
1056         struct e1000_buffer *buffer_info;
1057         struct e1000_ps_page *ps_page;
1058         struct pci_dev *pdev = adapter->pdev;
1059         unsigned int i, j;
1060
1061         /* Free all the Rx ring sk_buffs */
1062         for (i = 0; i < rx_ring->count; i++) {
1063                 buffer_info = &rx_ring->buffer_info[i];
1064                 if (buffer_info->dma) {
1065                         if (adapter->clean_rx == e1000_clean_rx_irq)
1066                                 pci_unmap_single(pdev, buffer_info->dma,
1067                                                  adapter->rx_buffer_len,
1068                                                  PCI_DMA_FROMDEVICE);
1069                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1070                                 pci_unmap_page(pdev, buffer_info->dma,
1071                                                PAGE_SIZE,
1072                                                PCI_DMA_FROMDEVICE);
1073                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1074                                 pci_unmap_single(pdev, buffer_info->dma,
1075                                                  adapter->rx_ps_bsize0,
1076                                                  PCI_DMA_FROMDEVICE);
1077                         buffer_info->dma = 0;
1078                 }
1079
1080                 if (buffer_info->page) {
1081                         put_page(buffer_info->page);
1082                         buffer_info->page = NULL;
1083                 }
1084
1085                 if (buffer_info->skb) {
1086                         dev_kfree_skb(buffer_info->skb);
1087                         buffer_info->skb = NULL;
1088                 }
1089
1090                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1091                         ps_page = &buffer_info->ps_pages[j];
1092                         if (!ps_page->page)
1093                                 break;
1094                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1095                                        PCI_DMA_FROMDEVICE);
1096                         ps_page->dma = 0;
1097                         put_page(ps_page->page);
1098                         ps_page->page = NULL;
1099                 }
1100         }
1101
1102         /* there also may be some cached data from a chained receive */
1103         if (rx_ring->rx_skb_top) {
1104                 dev_kfree_skb(rx_ring->rx_skb_top);
1105                 rx_ring->rx_skb_top = NULL;
1106         }
1107
1108         /* Zero out the descriptor ring */
1109         memset(rx_ring->desc, 0, rx_ring->size);
1110
1111         rx_ring->next_to_clean = 0;
1112         rx_ring->next_to_use = 0;
1113
1114         writel(0, adapter->hw.hw_addr + rx_ring->head);
1115         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1116 }
1117
1118 /**
1119  * e1000_intr_msi - Interrupt Handler
1120  * @irq: interrupt number
1121  * @data: pointer to a network interface device structure
1122  **/
1123 static irqreturn_t e1000_intr_msi(int irq, void *data)
1124 {
1125         struct net_device *netdev = data;
1126         struct e1000_adapter *adapter = netdev_priv(netdev);
1127         struct e1000_hw *hw = &adapter->hw;
1128         u32 icr = er32(ICR);
1129
1130         /*
1131          * read ICR disables interrupts using IAM
1132          */
1133
1134         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1135                 hw->mac.get_link_status = 1;
1136                 /*
1137                  * ICH8 workaround-- Call gig speed drop workaround on cable
1138                  * disconnect (LSC) before accessing any PHY registers
1139                  */
1140                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1141                     (!(er32(STATUS) & E1000_STATUS_LU)))
1142                         e1000e_gig_downshift_workaround_ich8lan(hw);
1143
1144                 /*
1145                  * 80003ES2LAN workaround-- For packet buffer work-around on
1146                  * link down event; disable receives here in the ISR and reset
1147                  * adapter in watchdog
1148                  */
1149                 if (netif_carrier_ok(netdev) &&
1150                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1151                         /* disable receives */
1152                         u32 rctl = er32(RCTL);
1153                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1154                         adapter->flags |= FLAG_RX_RESTART_NOW;
1155                 }
1156                 /* guard against interrupt when we're going down */
1157                 if (!test_bit(__E1000_DOWN, &adapter->state))
1158                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1159         }
1160
1161         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1162                 adapter->total_tx_bytes = 0;
1163                 adapter->total_tx_packets = 0;
1164                 adapter->total_rx_bytes = 0;
1165                 adapter->total_rx_packets = 0;
1166                 __netif_rx_schedule(netdev, &adapter->napi);
1167         }
1168
1169         return IRQ_HANDLED;
1170 }
1171
1172 /**
1173  * e1000_intr - Interrupt Handler
1174  * @irq: interrupt number
1175  * @data: pointer to a network interface device structure
1176  **/
1177 static irqreturn_t e1000_intr(int irq, void *data)
1178 {
1179         struct net_device *netdev = data;
1180         struct e1000_adapter *adapter = netdev_priv(netdev);
1181         struct e1000_hw *hw = &adapter->hw;
1182
1183         u32 rctl, icr = er32(ICR);
1184         if (!icr)
1185                 return IRQ_NONE;  /* Not our interrupt */
1186
1187         /*
1188          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1189          * not set, then the adapter didn't send an interrupt
1190          */
1191         if (!(icr & E1000_ICR_INT_ASSERTED))
1192                 return IRQ_NONE;
1193
1194         /*
1195          * Interrupt Auto-Mask...upon reading ICR,
1196          * interrupts are masked.  No need for the
1197          * IMC write
1198          */
1199
1200         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1201                 hw->mac.get_link_status = 1;
1202                 /*
1203                  * ICH8 workaround-- Call gig speed drop workaround on cable
1204                  * disconnect (LSC) before accessing any PHY registers
1205                  */
1206                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1207                     (!(er32(STATUS) & E1000_STATUS_LU)))
1208                         e1000e_gig_downshift_workaround_ich8lan(hw);
1209
1210                 /*
1211                  * 80003ES2LAN workaround--
1212                  * For packet buffer work-around on link down event;
1213                  * disable receives here in the ISR and
1214                  * reset adapter in watchdog
1215                  */
1216                 if (netif_carrier_ok(netdev) &&
1217                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1218                         /* disable receives */
1219                         rctl = er32(RCTL);
1220                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1221                         adapter->flags |= FLAG_RX_RESTART_NOW;
1222                 }
1223                 /* guard against interrupt when we're going down */
1224                 if (!test_bit(__E1000_DOWN, &adapter->state))
1225                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1226         }
1227
1228         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1229                 adapter->total_tx_bytes = 0;
1230                 adapter->total_tx_packets = 0;
1231                 adapter->total_rx_bytes = 0;
1232                 adapter->total_rx_packets = 0;
1233                 __netif_rx_schedule(netdev, &adapter->napi);
1234         }
1235
1236         return IRQ_HANDLED;
1237 }
1238
1239 /**
1240  * e1000_request_irq - initialize interrupts
1241  *
1242  * Attempts to configure interrupts using the best available
1243  * capabilities of the hardware and kernel.
1244  **/
1245 static int e1000_request_irq(struct e1000_adapter *adapter)
1246 {
1247         struct net_device *netdev = adapter->netdev;
1248         int irq_flags = IRQF_SHARED;
1249         int err;
1250
1251         if (!(adapter->flags & FLAG_MSI_TEST_FAILED)) {
1252                 err = pci_enable_msi(adapter->pdev);
1253                 if (!err) {
1254                         adapter->flags |= FLAG_MSI_ENABLED;
1255                         irq_flags = 0;
1256                 }
1257         }
1258
1259         err = request_irq(adapter->pdev->irq,
1260                           ((adapter->flags & FLAG_MSI_ENABLED) ?
1261                                 &e1000_intr_msi : &e1000_intr),
1262                           irq_flags, netdev->name, netdev);
1263         if (err) {
1264                 if (adapter->flags & FLAG_MSI_ENABLED) {
1265                         pci_disable_msi(adapter->pdev);
1266                         adapter->flags &= ~FLAG_MSI_ENABLED;
1267                 }
1268                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1269         }
1270
1271         return err;
1272 }
1273
1274 static void e1000_free_irq(struct e1000_adapter *adapter)
1275 {
1276         struct net_device *netdev = adapter->netdev;
1277
1278         free_irq(adapter->pdev->irq, netdev);
1279         if (adapter->flags & FLAG_MSI_ENABLED) {
1280                 pci_disable_msi(adapter->pdev);
1281                 adapter->flags &= ~FLAG_MSI_ENABLED;
1282         }
1283 }
1284
1285 /**
1286  * e1000_irq_disable - Mask off interrupt generation on the NIC
1287  **/
1288 static void e1000_irq_disable(struct e1000_adapter *adapter)
1289 {
1290         struct e1000_hw *hw = &adapter->hw;
1291
1292         ew32(IMC, ~0);
1293         e1e_flush();
1294         synchronize_irq(adapter->pdev->irq);
1295 }
1296
1297 /**
1298  * e1000_irq_enable - Enable default interrupt generation settings
1299  **/
1300 static void e1000_irq_enable(struct e1000_adapter *adapter)
1301 {
1302         struct e1000_hw *hw = &adapter->hw;
1303
1304         ew32(IMS, IMS_ENABLE_MASK);
1305         e1e_flush();
1306 }
1307
1308 /**
1309  * e1000_get_hw_control - get control of the h/w from f/w
1310  * @adapter: address of board private structure
1311  *
1312  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1313  * For ASF and Pass Through versions of f/w this means that
1314  * the driver is loaded. For AMT version (only with 82573)
1315  * of the f/w this means that the network i/f is open.
1316  **/
1317 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1318 {
1319         struct e1000_hw *hw = &adapter->hw;
1320         u32 ctrl_ext;
1321         u32 swsm;
1322
1323         /* Let firmware know the driver has taken over */
1324         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1325                 swsm = er32(SWSM);
1326                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1327         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1328                 ctrl_ext = er32(CTRL_EXT);
1329                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1330         }
1331 }
1332
1333 /**
1334  * e1000_release_hw_control - release control of the h/w to f/w
1335  * @adapter: address of board private structure
1336  *
1337  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1338  * For ASF and Pass Through versions of f/w this means that the
1339  * driver is no longer loaded. For AMT version (only with 82573) i
1340  * of the f/w this means that the network i/f is closed.
1341  *
1342  **/
1343 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1344 {
1345         struct e1000_hw *hw = &adapter->hw;
1346         u32 ctrl_ext;
1347         u32 swsm;
1348
1349         /* Let firmware taken over control of h/w */
1350         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1351                 swsm = er32(SWSM);
1352                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1353         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1354                 ctrl_ext = er32(CTRL_EXT);
1355                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1356         }
1357 }
1358
1359 /**
1360  * @e1000_alloc_ring - allocate memory for a ring structure
1361  **/
1362 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1363                                 struct e1000_ring *ring)
1364 {
1365         struct pci_dev *pdev = adapter->pdev;
1366
1367         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1368                                         GFP_KERNEL);
1369         if (!ring->desc)
1370                 return -ENOMEM;
1371
1372         return 0;
1373 }
1374
1375 /**
1376  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1377  * @adapter: board private structure
1378  *
1379  * Return 0 on success, negative on failure
1380  **/
1381 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1382 {
1383         struct e1000_ring *tx_ring = adapter->tx_ring;
1384         int err = -ENOMEM, size;
1385
1386         size = sizeof(struct e1000_buffer) * tx_ring->count;
1387         tx_ring->buffer_info = vmalloc(size);
1388         if (!tx_ring->buffer_info)
1389                 goto err;
1390         memset(tx_ring->buffer_info, 0, size);
1391
1392         /* round up to nearest 4K */
1393         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1394         tx_ring->size = ALIGN(tx_ring->size, 4096);
1395
1396         err = e1000_alloc_ring_dma(adapter, tx_ring);
1397         if (err)
1398                 goto err;
1399
1400         tx_ring->next_to_use = 0;
1401         tx_ring->next_to_clean = 0;
1402         spin_lock_init(&adapter->tx_queue_lock);
1403
1404         return 0;
1405 err:
1406         vfree(tx_ring->buffer_info);
1407         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1408         return err;
1409 }
1410
1411 /**
1412  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1413  * @adapter: board private structure
1414  *
1415  * Returns 0 on success, negative on failure
1416  **/
1417 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1418 {
1419         struct e1000_ring *rx_ring = adapter->rx_ring;
1420         struct e1000_buffer *buffer_info;
1421         int i, size, desc_len, err = -ENOMEM;
1422
1423         size = sizeof(struct e1000_buffer) * rx_ring->count;
1424         rx_ring->buffer_info = vmalloc(size);
1425         if (!rx_ring->buffer_info)
1426                 goto err;
1427         memset(rx_ring->buffer_info, 0, size);
1428
1429         for (i = 0; i < rx_ring->count; i++) {
1430                 buffer_info = &rx_ring->buffer_info[i];
1431                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1432                                                 sizeof(struct e1000_ps_page),
1433                                                 GFP_KERNEL);
1434                 if (!buffer_info->ps_pages)
1435                         goto err_pages;
1436         }
1437
1438         desc_len = sizeof(union e1000_rx_desc_packet_split);
1439
1440         /* Round up to nearest 4K */
1441         rx_ring->size = rx_ring->count * desc_len;
1442         rx_ring->size = ALIGN(rx_ring->size, 4096);
1443
1444         err = e1000_alloc_ring_dma(adapter, rx_ring);
1445         if (err)
1446                 goto err_pages;
1447
1448         rx_ring->next_to_clean = 0;
1449         rx_ring->next_to_use = 0;
1450         rx_ring->rx_skb_top = NULL;
1451
1452         return 0;
1453
1454 err_pages:
1455         for (i = 0; i < rx_ring->count; i++) {
1456                 buffer_info = &rx_ring->buffer_info[i];
1457                 kfree(buffer_info->ps_pages);
1458         }
1459 err:
1460         vfree(rx_ring->buffer_info);
1461         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1462         return err;
1463 }
1464
1465 /**
1466  * e1000_clean_tx_ring - Free Tx Buffers
1467  * @adapter: board private structure
1468  **/
1469 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1470 {
1471         struct e1000_ring *tx_ring = adapter->tx_ring;
1472         struct e1000_buffer *buffer_info;
1473         unsigned long size;
1474         unsigned int i;
1475
1476         for (i = 0; i < tx_ring->count; i++) {
1477                 buffer_info = &tx_ring->buffer_info[i];
1478                 e1000_put_txbuf(adapter, buffer_info);
1479         }
1480
1481         size = sizeof(struct e1000_buffer) * tx_ring->count;
1482         memset(tx_ring->buffer_info, 0, size);
1483
1484         memset(tx_ring->desc, 0, tx_ring->size);
1485
1486         tx_ring->next_to_use = 0;
1487         tx_ring->next_to_clean = 0;
1488
1489         writel(0, adapter->hw.hw_addr + tx_ring->head);
1490         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1491 }
1492
1493 /**
1494  * e1000e_free_tx_resources - Free Tx Resources per Queue
1495  * @adapter: board private structure
1496  *
1497  * Free all transmit software resources
1498  **/
1499 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1500 {
1501         struct pci_dev *pdev = adapter->pdev;
1502         struct e1000_ring *tx_ring = adapter->tx_ring;
1503
1504         e1000_clean_tx_ring(adapter);
1505
1506         vfree(tx_ring->buffer_info);
1507         tx_ring->buffer_info = NULL;
1508
1509         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1510                           tx_ring->dma);
1511         tx_ring->desc = NULL;
1512 }
1513
1514 /**
1515  * e1000e_free_rx_resources - Free Rx Resources
1516  * @adapter: board private structure
1517  *
1518  * Free all receive software resources
1519  **/
1520
1521 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1522 {
1523         struct pci_dev *pdev = adapter->pdev;
1524         struct e1000_ring *rx_ring = adapter->rx_ring;
1525         int i;
1526
1527         e1000_clean_rx_ring(adapter);
1528
1529         for (i = 0; i < rx_ring->count; i++) {
1530                 kfree(rx_ring->buffer_info[i].ps_pages);
1531         }
1532
1533         vfree(rx_ring->buffer_info);
1534         rx_ring->buffer_info = NULL;
1535
1536         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1537                           rx_ring->dma);
1538         rx_ring->desc = NULL;
1539 }
1540
1541 /**
1542  * e1000_update_itr - update the dynamic ITR value based on statistics
1543  * @adapter: pointer to adapter
1544  * @itr_setting: current adapter->itr
1545  * @packets: the number of packets during this measurement interval
1546  * @bytes: the number of bytes during this measurement interval
1547  *
1548  *      Stores a new ITR value based on packets and byte
1549  *      counts during the last interrupt.  The advantage of per interrupt
1550  *      computation is faster updates and more accurate ITR for the current
1551  *      traffic pattern.  Constants in this function were computed
1552  *      based on theoretical maximum wire speed and thresholds were set based
1553  *      on testing data as well as attempting to minimize response time
1554  *      while increasing bulk throughput.
1555  *      this functionality is controlled by the InterruptThrottleRate module
1556  *      parameter (see e1000_param.c)
1557  **/
1558 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1559                                      u16 itr_setting, int packets,
1560                                      int bytes)
1561 {
1562         unsigned int retval = itr_setting;
1563
1564         if (packets == 0)
1565                 goto update_itr_done;
1566
1567         switch (itr_setting) {
1568         case lowest_latency:
1569                 /* handle TSO and jumbo frames */
1570                 if (bytes/packets > 8000)
1571                         retval = bulk_latency;
1572                 else if ((packets < 5) && (bytes > 512)) {
1573                         retval = low_latency;
1574                 }
1575                 break;
1576         case low_latency:  /* 50 usec aka 20000 ints/s */
1577                 if (bytes > 10000) {
1578                         /* this if handles the TSO accounting */
1579                         if (bytes/packets > 8000) {
1580                                 retval = bulk_latency;
1581                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1582                                 retval = bulk_latency;
1583                         } else if ((packets > 35)) {
1584                                 retval = lowest_latency;
1585                         }
1586                 } else if (bytes/packets > 2000) {
1587                         retval = bulk_latency;
1588                 } else if (packets <= 2 && bytes < 512) {
1589                         retval = lowest_latency;
1590                 }
1591                 break;
1592         case bulk_latency: /* 250 usec aka 4000 ints/s */
1593                 if (bytes > 25000) {
1594                         if (packets > 35) {
1595                                 retval = low_latency;
1596                         }
1597                 } else if (bytes < 6000) {
1598                         retval = low_latency;
1599                 }
1600                 break;
1601         }
1602
1603 update_itr_done:
1604         return retval;
1605 }
1606
1607 static void e1000_set_itr(struct e1000_adapter *adapter)
1608 {
1609         struct e1000_hw *hw = &adapter->hw;
1610         u16 current_itr;
1611         u32 new_itr = adapter->itr;
1612
1613         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1614         if (adapter->link_speed != SPEED_1000) {
1615                 current_itr = 0;
1616                 new_itr = 4000;
1617                 goto set_itr_now;
1618         }
1619
1620         adapter->tx_itr = e1000_update_itr(adapter,
1621                                     adapter->tx_itr,
1622                                     adapter->total_tx_packets,
1623                                     adapter->total_tx_bytes);
1624         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1625         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1626                 adapter->tx_itr = low_latency;
1627
1628         adapter->rx_itr = e1000_update_itr(adapter,
1629                                     adapter->rx_itr,
1630                                     adapter->total_rx_packets,
1631                                     adapter->total_rx_bytes);
1632         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1633         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1634                 adapter->rx_itr = low_latency;
1635
1636         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1637
1638         switch (current_itr) {
1639         /* counts and packets in update_itr are dependent on these numbers */
1640         case lowest_latency:
1641                 new_itr = 70000;
1642                 break;
1643         case low_latency:
1644                 new_itr = 20000; /* aka hwitr = ~200 */
1645                 break;
1646         case bulk_latency:
1647                 new_itr = 4000;
1648                 break;
1649         default:
1650                 break;
1651         }
1652
1653 set_itr_now:
1654         if (new_itr != adapter->itr) {
1655                 /*
1656                  * this attempts to bias the interrupt rate towards Bulk
1657                  * by adding intermediate steps when interrupt rate is
1658                  * increasing
1659                  */
1660                 new_itr = new_itr > adapter->itr ?
1661                              min(adapter->itr + (new_itr >> 2), new_itr) :
1662                              new_itr;
1663                 adapter->itr = new_itr;
1664                 ew32(ITR, 1000000000 / (new_itr * 256));
1665         }
1666 }
1667
1668 /**
1669  * e1000_clean - NAPI Rx polling callback
1670  * @napi: struct associated with this polling callback
1671  * @budget: amount of packets driver is allowed to process this poll
1672  **/
1673 static int e1000_clean(struct napi_struct *napi, int budget)
1674 {
1675         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1676         struct net_device *poll_dev = adapter->netdev;
1677         int tx_cleaned = 0, work_done = 0;
1678
1679         /* Must NOT use netdev_priv macro here. */
1680         adapter = poll_dev->priv;
1681
1682         /*
1683          * e1000_clean is called per-cpu.  This lock protects
1684          * tx_ring from being cleaned by multiple cpus
1685          * simultaneously.  A failure obtaining the lock means
1686          * tx_ring is currently being cleaned anyway.
1687          */
1688         if (spin_trylock(&adapter->tx_queue_lock)) {
1689                 tx_cleaned = e1000_clean_tx_irq(adapter);
1690                 spin_unlock(&adapter->tx_queue_lock);
1691         }
1692
1693         adapter->clean_rx(adapter, &work_done, budget);
1694
1695         if (tx_cleaned)
1696                 work_done = budget;
1697
1698         /* If budget not fully consumed, exit the polling mode */
1699         if (work_done < budget) {
1700                 if (adapter->itr_setting & 3)
1701                         e1000_set_itr(adapter);
1702                 netif_rx_complete(poll_dev, napi);
1703                 e1000_irq_enable(adapter);
1704         }
1705
1706         return work_done;
1707 }
1708
1709 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1710 {
1711         struct e1000_adapter *adapter = netdev_priv(netdev);
1712         struct e1000_hw *hw = &adapter->hw;
1713         u32 vfta, index;
1714
1715         /* don't update vlan cookie if already programmed */
1716         if ((adapter->hw.mng_cookie.status &
1717              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1718             (vid == adapter->mng_vlan_id))
1719                 return;
1720         /* add VID to filter table */
1721         index = (vid >> 5) & 0x7F;
1722         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1723         vfta |= (1 << (vid & 0x1F));
1724         e1000e_write_vfta(hw, index, vfta);
1725 }
1726
1727 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1728 {
1729         struct e1000_adapter *adapter = netdev_priv(netdev);
1730         struct e1000_hw *hw = &adapter->hw;
1731         u32 vfta, index;
1732
1733         if (!test_bit(__E1000_DOWN, &adapter->state))
1734                 e1000_irq_disable(adapter);
1735         vlan_group_set_device(adapter->vlgrp, vid, NULL);
1736
1737         if (!test_bit(__E1000_DOWN, &adapter->state))
1738                 e1000_irq_enable(adapter);
1739
1740         if ((adapter->hw.mng_cookie.status &
1741              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1742             (vid == adapter->mng_vlan_id)) {
1743                 /* release control to f/w */
1744                 e1000_release_hw_control(adapter);
1745                 return;
1746         }
1747
1748         /* remove VID from filter table */
1749         index = (vid >> 5) & 0x7F;
1750         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1751         vfta &= ~(1 << (vid & 0x1F));
1752         e1000e_write_vfta(hw, index, vfta);
1753 }
1754
1755 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
1756 {
1757         struct net_device *netdev = adapter->netdev;
1758         u16 vid = adapter->hw.mng_cookie.vlan_id;
1759         u16 old_vid = adapter->mng_vlan_id;
1760
1761         if (!adapter->vlgrp)
1762                 return;
1763
1764         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
1765                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1766                 if (adapter->hw.mng_cookie.status &
1767                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1768                         e1000_vlan_rx_add_vid(netdev, vid);
1769                         adapter->mng_vlan_id = vid;
1770                 }
1771
1772                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
1773                                 (vid != old_vid) &&
1774                     !vlan_group_get_device(adapter->vlgrp, old_vid))
1775                         e1000_vlan_rx_kill_vid(netdev, old_vid);
1776         } else {
1777                 adapter->mng_vlan_id = vid;
1778         }
1779 }
1780
1781
1782 static void e1000_vlan_rx_register(struct net_device *netdev,
1783                                    struct vlan_group *grp)
1784 {
1785         struct e1000_adapter *adapter = netdev_priv(netdev);
1786         struct e1000_hw *hw = &adapter->hw;
1787         u32 ctrl, rctl;
1788
1789         if (!test_bit(__E1000_DOWN, &adapter->state))
1790                 e1000_irq_disable(adapter);
1791         adapter->vlgrp = grp;
1792
1793         if (grp) {
1794                 /* enable VLAN tag insert/strip */
1795                 ctrl = er32(CTRL);
1796                 ctrl |= E1000_CTRL_VME;
1797                 ew32(CTRL, ctrl);
1798
1799                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1800                         /* enable VLAN receive filtering */
1801                         rctl = er32(RCTL);
1802                         rctl &= ~E1000_RCTL_CFIEN;
1803                         ew32(RCTL, rctl);
1804                         e1000_update_mng_vlan(adapter);
1805                 }
1806         } else {
1807                 /* disable VLAN tag insert/strip */
1808                 ctrl = er32(CTRL);
1809                 ctrl &= ~E1000_CTRL_VME;
1810                 ew32(CTRL, ctrl);
1811
1812                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1813                         if (adapter->mng_vlan_id !=
1814                             (u16)E1000_MNG_VLAN_NONE) {
1815                                 e1000_vlan_rx_kill_vid(netdev,
1816                                                        adapter->mng_vlan_id);
1817                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1818                         }
1819                 }
1820         }
1821
1822         if (!test_bit(__E1000_DOWN, &adapter->state))
1823                 e1000_irq_enable(adapter);
1824 }
1825
1826 static void e1000_restore_vlan(struct e1000_adapter *adapter)
1827 {
1828         u16 vid;
1829
1830         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
1831
1832         if (!adapter->vlgrp)
1833                 return;
1834
1835         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1836                 if (!vlan_group_get_device(adapter->vlgrp, vid))
1837                         continue;
1838                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
1839         }
1840 }
1841
1842 static void e1000_init_manageability(struct e1000_adapter *adapter)
1843 {
1844         struct e1000_hw *hw = &adapter->hw;
1845         u32 manc, manc2h;
1846
1847         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
1848                 return;
1849
1850         manc = er32(MANC);
1851
1852         /*
1853          * enable receiving management packets to the host. this will probably
1854          * generate destination unreachable messages from the host OS, but
1855          * the packets will be handled on SMBUS
1856          */
1857         manc |= E1000_MANC_EN_MNG2HOST;
1858         manc2h = er32(MANC2H);
1859 #define E1000_MNG2HOST_PORT_623 (1 << 5)
1860 #define E1000_MNG2HOST_PORT_664 (1 << 6)
1861         manc2h |= E1000_MNG2HOST_PORT_623;
1862         manc2h |= E1000_MNG2HOST_PORT_664;
1863         ew32(MANC2H, manc2h);
1864         ew32(MANC, manc);
1865 }
1866
1867 /**
1868  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1869  * @adapter: board private structure
1870  *
1871  * Configure the Tx unit of the MAC after a reset.
1872  **/
1873 static void e1000_configure_tx(struct e1000_adapter *adapter)
1874 {
1875         struct e1000_hw *hw = &adapter->hw;
1876         struct e1000_ring *tx_ring = adapter->tx_ring;
1877         u64 tdba;
1878         u32 tdlen, tctl, tipg, tarc;
1879         u32 ipgr1, ipgr2;
1880
1881         /* Setup the HW Tx Head and Tail descriptor pointers */
1882         tdba = tx_ring->dma;
1883         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
1884         ew32(TDBAL, (tdba & DMA_32BIT_MASK));
1885         ew32(TDBAH, (tdba >> 32));
1886         ew32(TDLEN, tdlen);
1887         ew32(TDH, 0);
1888         ew32(TDT, 0);
1889         tx_ring->head = E1000_TDH;
1890         tx_ring->tail = E1000_TDT;
1891
1892         /* Set the default values for the Tx Inter Packet Gap timer */
1893         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
1894         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
1895         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
1896
1897         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
1898                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
1899
1900         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1901         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1902         ew32(TIPG, tipg);
1903
1904         /* Set the Tx Interrupt Delay register */
1905         ew32(TIDV, adapter->tx_int_delay);
1906         /* Tx irq moderation */
1907         ew32(TADV, adapter->tx_abs_int_delay);
1908
1909         /* Program the Transmit Control Register */
1910         tctl = er32(TCTL);
1911         tctl &= ~E1000_TCTL_CT;
1912         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1913                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1914
1915         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
1916                 tarc = er32(TARC(0));
1917                 /*
1918                  * set the speed mode bit, we'll clear it if we're not at
1919                  * gigabit link later
1920                  */
1921 #define SPEED_MODE_BIT (1 << 21)
1922                 tarc |= SPEED_MODE_BIT;
1923                 ew32(TARC(0), tarc);
1924         }
1925
1926         /* errata: program both queues to unweighted RR */
1927         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
1928                 tarc = er32(TARC(0));
1929                 tarc |= 1;
1930                 ew32(TARC(0), tarc);
1931                 tarc = er32(TARC(1));
1932                 tarc |= 1;
1933                 ew32(TARC(1), tarc);
1934         }
1935
1936         e1000e_config_collision_dist(hw);
1937
1938         /* Setup Transmit Descriptor Settings for eop descriptor */
1939         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1940
1941         /* only set IDE if we are delaying interrupts using the timers */
1942         if (adapter->tx_int_delay)
1943                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1944
1945         /* enable Report Status bit */
1946         adapter->txd_cmd |= E1000_TXD_CMD_RS;
1947
1948         ew32(TCTL, tctl);
1949
1950         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
1951 }
1952
1953 /**
1954  * e1000_setup_rctl - configure the receive control registers
1955  * @adapter: Board private structure
1956  **/
1957 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1958                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1959 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1960 {
1961         struct e1000_hw *hw = &adapter->hw;
1962         u32 rctl, rfctl;
1963         u32 psrctl = 0;
1964         u32 pages = 0;
1965
1966         /* Program MC offset vector base */
1967         rctl = er32(RCTL);
1968         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1969         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1970                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1971                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1972
1973         /* Do not Store bad packets */
1974         rctl &= ~E1000_RCTL_SBP;
1975
1976         /* Enable Long Packet receive */
1977         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1978                 rctl &= ~E1000_RCTL_LPE;
1979         else
1980                 rctl |= E1000_RCTL_LPE;
1981
1982         /* Enable hardware CRC frame stripping */
1983         rctl |= E1000_RCTL_SECRC;
1984
1985         /* Setup buffer sizes */
1986         rctl &= ~E1000_RCTL_SZ_4096;
1987         rctl |= E1000_RCTL_BSEX;
1988         switch (adapter->rx_buffer_len) {
1989         case 256:
1990                 rctl |= E1000_RCTL_SZ_256;
1991                 rctl &= ~E1000_RCTL_BSEX;
1992                 break;
1993         case 512:
1994                 rctl |= E1000_RCTL_SZ_512;
1995                 rctl &= ~E1000_RCTL_BSEX;
1996                 break;
1997         case 1024:
1998                 rctl |= E1000_RCTL_SZ_1024;
1999                 rctl &= ~E1000_RCTL_BSEX;
2000                 break;
2001         case 2048:
2002         default:
2003                 rctl |= E1000_RCTL_SZ_2048;
2004                 rctl &= ~E1000_RCTL_BSEX;
2005                 break;
2006         case 4096:
2007                 rctl |= E1000_RCTL_SZ_4096;
2008                 break;
2009         case 8192:
2010                 rctl |= E1000_RCTL_SZ_8192;
2011                 break;
2012         case 16384:
2013                 rctl |= E1000_RCTL_SZ_16384;
2014                 break;
2015         }
2016
2017         /*
2018          * 82571 and greater support packet-split where the protocol
2019          * header is placed in skb->data and the packet data is
2020          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2021          * In the case of a non-split, skb->data is linearly filled,
2022          * followed by the page buffers.  Therefore, skb->data is
2023          * sized to hold the largest protocol header.
2024          *
2025          * allocations using alloc_page take too long for regular MTU
2026          * so only enable packet split for jumbo frames
2027          *
2028          * Using pages when the page size is greater than 16k wastes
2029          * a lot of memory, since we allocate 3 pages at all times
2030          * per packet.
2031          */
2032         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2033         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2034             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2035                 adapter->rx_ps_pages = pages;
2036         else
2037                 adapter->rx_ps_pages = 0;
2038
2039         if (adapter->rx_ps_pages) {
2040                 /* Configure extra packet-split registers */
2041                 rfctl = er32(RFCTL);
2042                 rfctl |= E1000_RFCTL_EXTEN;
2043                 /*
2044                  * disable packet split support for IPv6 extension headers,
2045                  * because some malformed IPv6 headers can hang the Rx
2046                  */
2047                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2048                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2049
2050                 ew32(RFCTL, rfctl);
2051
2052                 /* Enable Packet split descriptors */
2053                 rctl |= E1000_RCTL_DTYP_PS;
2054
2055                 psrctl |= adapter->rx_ps_bsize0 >>
2056                         E1000_PSRCTL_BSIZE0_SHIFT;
2057
2058                 switch (adapter->rx_ps_pages) {
2059                 case 3:
2060                         psrctl |= PAGE_SIZE <<
2061                                 E1000_PSRCTL_BSIZE3_SHIFT;
2062                 case 2:
2063                         psrctl |= PAGE_SIZE <<
2064                                 E1000_PSRCTL_BSIZE2_SHIFT;
2065                 case 1:
2066                         psrctl |= PAGE_SIZE >>
2067                                 E1000_PSRCTL_BSIZE1_SHIFT;
2068                         break;
2069                 }
2070
2071                 ew32(PSRCTL, psrctl);
2072         }
2073
2074         ew32(RCTL, rctl);
2075         /* just started the receive unit, no need to restart */
2076         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2077 }
2078
2079 /**
2080  * e1000_configure_rx - Configure Receive Unit after Reset
2081  * @adapter: board private structure
2082  *
2083  * Configure the Rx unit of the MAC after a reset.
2084  **/
2085 static void e1000_configure_rx(struct e1000_adapter *adapter)
2086 {
2087         struct e1000_hw *hw = &adapter->hw;
2088         struct e1000_ring *rx_ring = adapter->rx_ring;
2089         u64 rdba;
2090         u32 rdlen, rctl, rxcsum, ctrl_ext;
2091
2092         if (adapter->rx_ps_pages) {
2093                 /* this is a 32 byte descriptor */
2094                 rdlen = rx_ring->count *
2095                         sizeof(union e1000_rx_desc_packet_split);
2096                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2097                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2098         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2099                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2100                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2101                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2102         } else {
2103                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2104                 adapter->clean_rx = e1000_clean_rx_irq;
2105                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2106         }
2107
2108         /* disable receives while setting up the descriptors */
2109         rctl = er32(RCTL);
2110         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2111         e1e_flush();
2112         msleep(10);
2113
2114         /* set the Receive Delay Timer Register */
2115         ew32(RDTR, adapter->rx_int_delay);
2116
2117         /* irq moderation */
2118         ew32(RADV, adapter->rx_abs_int_delay);
2119         if (adapter->itr_setting != 0)
2120                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2121
2122         ctrl_ext = er32(CTRL_EXT);
2123         /* Reset delay timers after every interrupt */
2124         ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2125         /* Auto-Mask interrupts upon ICR access */
2126         ctrl_ext |= E1000_CTRL_EXT_IAME;
2127         ew32(IAM, 0xffffffff);
2128         ew32(CTRL_EXT, ctrl_ext);
2129         e1e_flush();
2130
2131         /*
2132          * Setup the HW Rx Head and Tail Descriptor Pointers and
2133          * the Base and Length of the Rx Descriptor Ring
2134          */
2135         rdba = rx_ring->dma;
2136         ew32(RDBAL, (rdba & DMA_32BIT_MASK));
2137         ew32(RDBAH, (rdba >> 32));
2138         ew32(RDLEN, rdlen);
2139         ew32(RDH, 0);
2140         ew32(RDT, 0);
2141         rx_ring->head = E1000_RDH;
2142         rx_ring->tail = E1000_RDT;
2143
2144         /* Enable Receive Checksum Offload for TCP and UDP */
2145         rxcsum = er32(RXCSUM);
2146         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2147                 rxcsum |= E1000_RXCSUM_TUOFL;
2148
2149                 /*
2150                  * IPv4 payload checksum for UDP fragments must be
2151                  * used in conjunction with packet-split.
2152                  */
2153                 if (adapter->rx_ps_pages)
2154                         rxcsum |= E1000_RXCSUM_IPPCSE;
2155         } else {
2156                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2157                 /* no need to clear IPPCSE as it defaults to 0 */
2158         }
2159         ew32(RXCSUM, rxcsum);
2160
2161         /*
2162          * Enable early receives on supported devices, only takes effect when
2163          * packet size is equal or larger than the specified value (in 8 byte
2164          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2165          */
2166         if ((adapter->flags & FLAG_HAS_ERT) &&
2167             (adapter->netdev->mtu > ETH_DATA_LEN)) {
2168                 u32 rxdctl = er32(RXDCTL(0));
2169                 ew32(RXDCTL(0), rxdctl | 0x3);
2170                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2171                 /*
2172                  * With jumbo frames and early-receive enabled, excessive
2173                  * C4->C2 latencies result in dropped transactions.
2174                  */
2175                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2176                                           e1000e_driver_name, 55);
2177         } else {
2178                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2179                                           e1000e_driver_name,
2180                                           PM_QOS_DEFAULT_VALUE);
2181         }
2182
2183         /* Enable Receives */
2184         ew32(RCTL, rctl);
2185 }
2186
2187 /**
2188  *  e1000_update_mc_addr_list - Update Multicast addresses
2189  *  @hw: pointer to the HW structure
2190  *  @mc_addr_list: array of multicast addresses to program
2191  *  @mc_addr_count: number of multicast addresses to program
2192  *  @rar_used_count: the first RAR register free to program
2193  *  @rar_count: total number of supported Receive Address Registers
2194  *
2195  *  Updates the Receive Address Registers and Multicast Table Array.
2196  *  The caller must have a packed mc_addr_list of multicast addresses.
2197  *  The parameter rar_count will usually be hw->mac.rar_entry_count
2198  *  unless there are workarounds that change this.  Currently no func pointer
2199  *  exists and all implementations are handled in the generic version of this
2200  *  function.
2201  **/
2202 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2203                                       u32 mc_addr_count, u32 rar_used_count,
2204                                       u32 rar_count)
2205 {
2206         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2207                                         rar_used_count, rar_count);
2208 }
2209
2210 /**
2211  * e1000_set_multi - Multicast and Promiscuous mode set
2212  * @netdev: network interface device structure
2213  *
2214  * The set_multi entry point is called whenever the multicast address
2215  * list or the network interface flags are updated.  This routine is
2216  * responsible for configuring the hardware for proper multicast,
2217  * promiscuous mode, and all-multi behavior.
2218  **/
2219 static void e1000_set_multi(struct net_device *netdev)
2220 {
2221         struct e1000_adapter *adapter = netdev_priv(netdev);
2222         struct e1000_hw *hw = &adapter->hw;
2223         struct e1000_mac_info *mac = &hw->mac;
2224         struct dev_mc_list *mc_ptr;
2225         u8  *mta_list;
2226         u32 rctl;
2227         int i;
2228
2229         /* Check for Promiscuous and All Multicast modes */
2230
2231         rctl = er32(RCTL);
2232
2233         if (netdev->flags & IFF_PROMISC) {
2234                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2235                 rctl &= ~E1000_RCTL_VFE;
2236         } else {
2237                 if (netdev->flags & IFF_ALLMULTI) {
2238                         rctl |= E1000_RCTL_MPE;
2239                         rctl &= ~E1000_RCTL_UPE;
2240                 } else {
2241                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2242                 }
2243                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2244                         rctl |= E1000_RCTL_VFE;
2245         }
2246
2247         ew32(RCTL, rctl);
2248
2249         if (netdev->mc_count) {
2250                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2251                 if (!mta_list)
2252                         return;
2253
2254                 /* prepare a packed array of only addresses. */
2255                 mc_ptr = netdev->mc_list;
2256
2257                 for (i = 0; i < netdev->mc_count; i++) {
2258                         if (!mc_ptr)
2259                                 break;
2260                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2261                                ETH_ALEN);
2262                         mc_ptr = mc_ptr->next;
2263                 }
2264
2265                 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2266                                           mac->rar_entry_count);
2267                 kfree(mta_list);
2268         } else {
2269                 /*
2270                  * if we're called from probe, we might not have
2271                  * anything to do here, so clear out the list
2272                  */
2273                 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2274         }
2275 }
2276
2277 /**
2278  * e1000_configure - configure the hardware for Rx and Tx
2279  * @adapter: private board structure
2280  **/
2281 static void e1000_configure(struct e1000_adapter *adapter)
2282 {
2283         e1000_set_multi(adapter->netdev);
2284
2285         e1000_restore_vlan(adapter);
2286         e1000_init_manageability(adapter);
2287
2288         e1000_configure_tx(adapter);
2289         e1000_setup_rctl(adapter);
2290         e1000_configure_rx(adapter);
2291         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2292 }
2293
2294 /**
2295  * e1000e_power_up_phy - restore link in case the phy was powered down
2296  * @adapter: address of board private structure
2297  *
2298  * The phy may be powered down to save power and turn off link when the
2299  * driver is unloaded and wake on lan is not enabled (among others)
2300  * *** this routine MUST be followed by a call to e1000e_reset ***
2301  **/
2302 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2303 {
2304         u16 mii_reg = 0;
2305
2306         /* Just clear the power down bit to wake the phy back up */
2307         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2308                 /*
2309                  * According to the manual, the phy will retain its
2310                  * settings across a power-down/up cycle
2311                  */
2312                 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2313                 mii_reg &= ~MII_CR_POWER_DOWN;
2314                 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2315         }
2316
2317         adapter->hw.mac.ops.setup_link(&adapter->hw);
2318 }
2319
2320 /**
2321  * e1000_power_down_phy - Power down the PHY
2322  *
2323  * Power down the PHY so no link is implied when interface is down
2324  * The PHY cannot be powered down is management or WoL is active
2325  */
2326 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2327 {
2328         struct e1000_hw *hw = &adapter->hw;
2329         u16 mii_reg;
2330
2331         /* WoL is enabled */
2332         if (adapter->wol)
2333                 return;
2334
2335         /* non-copper PHY? */
2336         if (adapter->hw.phy.media_type != e1000_media_type_copper)
2337                 return;
2338
2339         /* reset is blocked because of a SoL/IDER session */
2340         if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2341                 return;
2342
2343         /* manageability (AMT) is enabled */
2344         if (er32(MANC) & E1000_MANC_SMBUS_EN)
2345                 return;
2346
2347         /* power down the PHY */
2348         e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2349         mii_reg |= MII_CR_POWER_DOWN;
2350         e1e_wphy(hw, PHY_CONTROL, mii_reg);
2351         mdelay(1);
2352 }
2353
2354 /**
2355  * e1000e_reset - bring the hardware into a known good state
2356  *
2357  * This function boots the hardware and enables some settings that
2358  * require a configuration cycle of the hardware - those cannot be
2359  * set/changed during runtime. After reset the device needs to be
2360  * properly configured for Rx, Tx etc.
2361  */
2362 void e1000e_reset(struct e1000_adapter *adapter)
2363 {
2364         struct e1000_mac_info *mac = &adapter->hw.mac;
2365         struct e1000_fc_info *fc = &adapter->hw.fc;
2366         struct e1000_hw *hw = &adapter->hw;
2367         u32 tx_space, min_tx_space, min_rx_space;
2368         u32 pba = adapter->pba;
2369         u16 hwm;
2370
2371         /* reset Packet Buffer Allocation to default */
2372         ew32(PBA, pba);
2373
2374         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2375                 /*
2376                  * To maintain wire speed transmits, the Tx FIFO should be
2377                  * large enough to accommodate two full transmit packets,
2378                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2379                  * the Rx FIFO should be large enough to accommodate at least
2380                  * one full receive packet and is similarly rounded up and
2381                  * expressed in KB.
2382                  */
2383                 pba = er32(PBA);
2384                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2385                 tx_space = pba >> 16;
2386                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2387                 pba &= 0xffff;
2388                 /*
2389                  * the Tx fifo also stores 16 bytes of information about the tx
2390                  * but don't include ethernet FCS because hardware appends it
2391                  */
2392                 min_tx_space = (adapter->max_frame_size +
2393                                 sizeof(struct e1000_tx_desc) -
2394                                 ETH_FCS_LEN) * 2;
2395                 min_tx_space = ALIGN(min_tx_space, 1024);
2396                 min_tx_space >>= 10;
2397                 /* software strips receive CRC, so leave room for it */
2398                 min_rx_space = adapter->max_frame_size;
2399                 min_rx_space = ALIGN(min_rx_space, 1024);
2400                 min_rx_space >>= 10;
2401
2402                 /*
2403                  * If current Tx allocation is less than the min Tx FIFO size,
2404                  * and the min Tx FIFO size is less than the current Rx FIFO
2405                  * allocation, take space away from current Rx allocation
2406                  */
2407                 if ((tx_space < min_tx_space) &&
2408                     ((min_tx_space - tx_space) < pba)) {
2409                         pba -= min_tx_space - tx_space;
2410
2411                         /*
2412                          * if short on Rx space, Rx wins and must trump tx
2413                          * adjustment or use Early Receive if available
2414                          */
2415                         if ((pba < min_rx_space) &&
2416                             (!(adapter->flags & FLAG_HAS_ERT)))
2417                                 /* ERT enabled in e1000_configure_rx */
2418                                 pba = min_rx_space;
2419                 }
2420
2421                 ew32(PBA, pba);
2422         }
2423
2424
2425         /*
2426          * flow control settings
2427          *
2428          * The high water mark must be low enough to fit one full frame
2429          * (or the size used for early receive) above it in the Rx FIFO.
2430          * Set it to the lower of:
2431          * - 90% of the Rx FIFO size, and
2432          * - the full Rx FIFO size minus the early receive size (for parts
2433          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2434          * - the full Rx FIFO size minus one full frame
2435          */
2436         if (adapter->flags & FLAG_HAS_ERT)
2437                 hwm = min(((pba << 10) * 9 / 10),
2438                           ((pba << 10) - (E1000_ERT_2048 << 3)));
2439         else
2440                 hwm = min(((pba << 10) * 9 / 10),
2441                           ((pba << 10) - adapter->max_frame_size));
2442
2443         fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
2444         fc->low_water = fc->high_water - 8;
2445
2446         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2447                 fc->pause_time = 0xFFFF;
2448         else
2449                 fc->pause_time = E1000_FC_PAUSE_TIME;
2450         fc->send_xon = 1;
2451         fc->type = fc->original_type;
2452
2453         /* Allow time for pending master requests to run */
2454         mac->ops.reset_hw(hw);
2455
2456         /*
2457          * For parts with AMT enabled, let the firmware know
2458          * that the network interface is in control
2459          */
2460         if (adapter->flags & FLAG_HAS_AMT)
2461                 e1000_get_hw_control(adapter);
2462
2463         ew32(WUC, 0);
2464
2465         if (mac->ops.init_hw(hw))
2466                 e_err("Hardware Error\n");
2467
2468         e1000_update_mng_vlan(adapter);
2469
2470         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2471         ew32(VET, ETH_P_8021Q);
2472
2473         e1000e_reset_adaptive(hw);
2474         e1000_get_phy_info(hw);
2475
2476         if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2477                 u16 phy_data = 0;
2478                 /*
2479                  * speed up time to link by disabling smart power down, ignore
2480                  * the return value of this function because there is nothing
2481                  * different we would do if it failed
2482                  */
2483                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2484                 phy_data &= ~IGP02E1000_PM_SPD;
2485                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2486         }
2487 }
2488
2489 int e1000e_up(struct e1000_adapter *adapter)
2490 {
2491         struct e1000_hw *hw = &adapter->hw;
2492
2493         /* hardware has been reset, we need to reload some things */
2494         e1000_configure(adapter);
2495
2496         clear_bit(__E1000_DOWN, &adapter->state);
2497
2498         napi_enable(&adapter->napi);
2499         e1000_irq_enable(adapter);
2500
2501         /* fire a link change interrupt to start the watchdog */
2502         ew32(ICS, E1000_ICS_LSC);
2503         return 0;
2504 }
2505
2506 void e1000e_down(struct e1000_adapter *adapter)
2507 {
2508         struct net_device *netdev = adapter->netdev;
2509         struct e1000_hw *hw = &adapter->hw;
2510         u32 tctl, rctl;
2511
2512         /*
2513          * signal that we're down so the interrupt handler does not
2514          * reschedule our watchdog timer
2515          */
2516         set_bit(__E1000_DOWN, &adapter->state);
2517
2518         /* disable receives in the hardware */
2519         rctl = er32(RCTL);
2520         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2521         /* flush and sleep below */
2522
2523         netif_tx_stop_all_queues(netdev);
2524
2525         /* disable transmits in the hardware */
2526         tctl = er32(TCTL);
2527         tctl &= ~E1000_TCTL_EN;
2528         ew32(TCTL, tctl);
2529         /* flush both disables and wait for them to finish */
2530         e1e_flush();
2531         msleep(10);
2532
2533         napi_disable(&adapter->napi);
2534         e1000_irq_disable(adapter);
2535
2536         del_timer_sync(&adapter->watchdog_timer);
2537         del_timer_sync(&adapter->phy_info_timer);
2538
2539         netdev->tx_queue_len = adapter->tx_queue_len;
2540         netif_carrier_off(netdev);
2541         adapter->link_speed = 0;
2542         adapter->link_duplex = 0;
2543
2544         if (!pci_channel_offline(adapter->pdev))
2545                 e1000e_reset(adapter);
2546         e1000_clean_tx_ring(adapter);
2547         e1000_clean_rx_ring(adapter);
2548
2549         /*
2550          * TODO: for power management, we could drop the link and
2551          * pci_disable_device here.
2552          */
2553 }
2554
2555 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2556 {
2557         might_sleep();
2558         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2559                 msleep(1);
2560         e1000e_down(adapter);
2561         e1000e_up(adapter);
2562         clear_bit(__E1000_RESETTING, &adapter->state);
2563 }
2564
2565 /**
2566  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2567  * @adapter: board private structure to initialize
2568  *
2569  * e1000_sw_init initializes the Adapter private data structure.
2570  * Fields are initialized based on PCI device information and
2571  * OS network device settings (MTU size).
2572  **/
2573 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2574 {
2575         struct net_device *netdev = adapter->netdev;
2576
2577         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2578         adapter->rx_ps_bsize0 = 128;
2579         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2580         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2581
2582         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2583         if (!adapter->tx_ring)
2584                 goto err;
2585
2586         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2587         if (!adapter->rx_ring)
2588                 goto err;
2589
2590         spin_lock_init(&adapter->tx_queue_lock);
2591
2592         /* Explicitly disable IRQ since the NIC can be in any state. */
2593         e1000_irq_disable(adapter);
2594
2595         spin_lock_init(&adapter->stats_lock);
2596
2597         set_bit(__E1000_DOWN, &adapter->state);
2598         return 0;
2599
2600 err:
2601         e_err("Unable to allocate memory for queues\n");
2602         kfree(adapter->rx_ring);
2603         kfree(adapter->tx_ring);
2604         return -ENOMEM;
2605 }
2606
2607 /**
2608  * e1000_intr_msi_test - Interrupt Handler
2609  * @irq: interrupt number
2610  * @data: pointer to a network interface device structure
2611  **/
2612 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2613 {
2614         struct net_device *netdev = data;
2615         struct e1000_adapter *adapter = netdev_priv(netdev);
2616         struct e1000_hw *hw = &adapter->hw;
2617         u32 icr = er32(ICR);
2618
2619         e_dbg("%s: icr is %08X\n", netdev->name, icr);
2620         if (icr & E1000_ICR_RXSEQ) {
2621                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2622                 wmb();
2623         }
2624
2625         return IRQ_HANDLED;
2626 }
2627
2628 /**
2629  * e1000_test_msi_interrupt - Returns 0 for successful test
2630  * @adapter: board private struct
2631  *
2632  * code flow taken from tg3.c
2633  **/
2634 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2635 {
2636         struct net_device *netdev = adapter->netdev;
2637         struct e1000_hw *hw = &adapter->hw;
2638         int err;
2639
2640         /* poll_enable hasn't been called yet, so don't need disable */
2641         /* clear any pending events */
2642         er32(ICR);
2643
2644         /* free the real vector and request a test handler */
2645         e1000_free_irq(adapter);
2646
2647         /* Assume that the test fails, if it succeeds then the test
2648          * MSI irq handler will unset this flag */
2649         adapter->flags |= FLAG_MSI_TEST_FAILED;
2650
2651         err = pci_enable_msi(adapter->pdev);
2652         if (err)
2653                 goto msi_test_failed;
2654
2655         err = request_irq(adapter->pdev->irq, &e1000_intr_msi_test, 0,
2656                           netdev->name, netdev);
2657         if (err) {
2658                 pci_disable_msi(adapter->pdev);
2659                 goto msi_test_failed;
2660         }
2661
2662         wmb();
2663
2664         e1000_irq_enable(adapter);
2665
2666         /* fire an unusual interrupt on the test handler */
2667         ew32(ICS, E1000_ICS_RXSEQ);
2668         e1e_flush();
2669         msleep(50);
2670
2671         e1000_irq_disable(adapter);
2672
2673         rmb();
2674
2675         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
2676                 err = -EIO;
2677                 e_info("MSI interrupt test failed!\n");
2678         }
2679
2680         free_irq(adapter->pdev->irq, netdev);
2681         pci_disable_msi(adapter->pdev);
2682
2683         if (err == -EIO)
2684                 goto msi_test_failed;
2685
2686         /* okay so the test worked, restore settings */
2687         e_dbg("%s: MSI interrupt test succeeded!\n", netdev->name);
2688 msi_test_failed:
2689         /* restore the original vector, even if it failed */
2690         e1000_request_irq(adapter);
2691         return err;
2692 }
2693
2694 /**
2695  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
2696  * @adapter: board private struct
2697  *
2698  * code flow taken from tg3.c, called with e1000 interrupts disabled.
2699  **/
2700 static int e1000_test_msi(struct e1000_adapter *adapter)
2701 {
2702         int err;
2703         u16 pci_cmd;
2704
2705         if (!(adapter->flags & FLAG_MSI_ENABLED))
2706                 return 0;
2707
2708         /* disable SERR in case the MSI write causes a master abort */
2709         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
2710         pci_write_config_word(adapter->pdev, PCI_COMMAND,
2711                               pci_cmd & ~PCI_COMMAND_SERR);
2712
2713         err = e1000_test_msi_interrupt(adapter);
2714
2715         /* restore previous setting of command word */
2716         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
2717
2718         /* success ! */
2719         if (!err)
2720                 return 0;
2721
2722         /* EIO means MSI test failed */
2723         if (err != -EIO)
2724                 return err;
2725
2726         /* back to INTx mode */
2727         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
2728
2729         e1000_free_irq(adapter);
2730
2731         err = e1000_request_irq(adapter);
2732
2733         return err;
2734 }
2735
2736 /**
2737  * e1000_open - Called when a network interface is made active
2738  * @netdev: network interface device structure
2739  *
2740  * Returns 0 on success, negative value on failure
2741  *
2742  * The open entry point is called when a network interface is made
2743  * active by the system (IFF_UP).  At this point all resources needed
2744  * for transmit and receive operations are allocated, the interrupt
2745  * handler is registered with the OS, the watchdog timer is started,
2746  * and the stack is notified that the interface is ready.
2747  **/
2748 static int e1000_open(struct net_device *netdev)
2749 {
2750         struct e1000_adapter *adapter = netdev_priv(netdev);
2751         struct e1000_hw *hw = &adapter->hw;
2752         int err;
2753
2754         /* disallow open during test */
2755         if (test_bit(__E1000_TESTING, &adapter->state))
2756                 return -EBUSY;
2757
2758         /* allocate transmit descriptors */
2759         err = e1000e_setup_tx_resources(adapter);
2760         if (err)
2761                 goto err_setup_tx;
2762
2763         /* allocate receive descriptors */
2764         err = e1000e_setup_rx_resources(adapter);
2765         if (err)
2766                 goto err_setup_rx;
2767
2768         e1000e_power_up_phy(adapter);
2769
2770         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2771         if ((adapter->hw.mng_cookie.status &
2772              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
2773                 e1000_update_mng_vlan(adapter);
2774
2775         /*
2776          * If AMT is enabled, let the firmware know that the network
2777          * interface is now open
2778          */
2779         if (adapter->flags & FLAG_HAS_AMT)
2780                 e1000_get_hw_control(adapter);
2781
2782         /*
2783          * before we allocate an interrupt, we must be ready to handle it.
2784          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2785          * as soon as we call pci_request_irq, so we have to setup our
2786          * clean_rx handler before we do so.
2787          */
2788         e1000_configure(adapter);
2789
2790         err = e1000_request_irq(adapter);
2791         if (err)
2792                 goto err_req_irq;
2793
2794         /*
2795          * Work around PCIe errata with MSI interrupts causing some chipsets to
2796          * ignore e1000e MSI messages, which means we need to test our MSI
2797          * interrupt now
2798          */
2799         {
2800                 err = e1000_test_msi(adapter);
2801                 if (err) {
2802                         e_err("Interrupt allocation failed\n");
2803                         goto err_req_irq;
2804                 }
2805         }
2806
2807         /* From here on the code is the same as e1000e_up() */
2808         clear_bit(__E1000_DOWN, &adapter->state);
2809
2810         napi_enable(&adapter->napi);
2811
2812         e1000_irq_enable(adapter);
2813
2814         netif_tx_start_all_queues(netdev);
2815
2816         /* fire a link status change interrupt to start the watchdog */
2817         ew32(ICS, E1000_ICS_LSC);
2818
2819         return 0;
2820
2821 err_req_irq:
2822         e1000_release_hw_control(adapter);
2823         e1000_power_down_phy(adapter);
2824         e1000e_free_rx_resources(adapter);
2825 err_setup_rx:
2826         e1000e_free_tx_resources(adapter);
2827 err_setup_tx:
2828         e1000e_reset(adapter);
2829
2830         return err;
2831 }
2832
2833 /**
2834  * e1000_close - Disables a network interface
2835  * @netdev: network interface device structure
2836  *
2837  * Returns 0, this is not allowed to fail
2838  *
2839  * The close entry point is called when an interface is de-activated
2840  * by the OS.  The hardware is still under the drivers control, but
2841  * needs to be disabled.  A global MAC reset is issued to stop the
2842  * hardware, and all transmit and receive resources are freed.
2843  **/
2844 static int e1000_close(struct net_device *netdev)
2845 {
2846         struct e1000_adapter *adapter = netdev_priv(netdev);
2847
2848         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
2849         e1000e_down(adapter);
2850         e1000_power_down_phy(adapter);
2851         e1000_free_irq(adapter);
2852
2853         e1000e_free_tx_resources(adapter);
2854         e1000e_free_rx_resources(adapter);
2855
2856         /*
2857          * kill manageability vlan ID if supported, but not if a vlan with
2858          * the same ID is registered on the host OS (let 8021q kill it)
2859          */
2860         if ((adapter->hw.mng_cookie.status &
2861                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2862              !(adapter->vlgrp &&
2863                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
2864                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2865
2866         /*
2867          * If AMT is enabled, let the firmware know that the network
2868          * interface is now closed
2869          */
2870         if (adapter->flags & FLAG_HAS_AMT)
2871                 e1000_release_hw_control(adapter);
2872
2873         return 0;
2874 }
2875 /**
2876  * e1000_set_mac - Change the Ethernet Address of the NIC
2877  * @netdev: network interface device structure
2878  * @p: pointer to an address structure
2879  *
2880  * Returns 0 on success, negative on failure
2881  **/
2882 static int e1000_set_mac(struct net_device *netdev, void *p)
2883 {
2884         struct e1000_adapter *adapter = netdev_priv(netdev);
2885         struct sockaddr *addr = p;
2886
2887         if (!is_valid_ether_addr(addr->sa_data))
2888                 return -EADDRNOTAVAIL;
2889
2890         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2891         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
2892
2893         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
2894
2895         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
2896                 /* activate the work around */
2897                 e1000e_set_laa_state_82571(&adapter->hw, 1);
2898
2899                 /*
2900                  * Hold a copy of the LAA in RAR[14] This is done so that
2901                  * between the time RAR[0] gets clobbered  and the time it
2902                  * gets fixed (in e1000_watchdog), the actual LAA is in one
2903                  * of the RARs and no incoming packets directed to this port
2904                  * are dropped. Eventually the LAA will be in RAR[0] and
2905                  * RAR[14]
2906                  */
2907                 e1000e_rar_set(&adapter->hw,
2908                               adapter->hw.mac.addr,
2909                               adapter->hw.mac.rar_entry_count - 1);
2910         }
2911
2912         return 0;
2913 }
2914
2915 /*
2916  * Need to wait a few seconds after link up to get diagnostic information from
2917  * the phy
2918  */
2919 static void e1000_update_phy_info(unsigned long data)
2920 {
2921         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2922         e1000_get_phy_info(&adapter->hw);
2923 }
2924
2925 /**
2926  * e1000e_update_stats - Update the board statistics counters
2927  * @adapter: board private structure
2928  **/
2929 void e1000e_update_stats(struct e1000_adapter *adapter)
2930 {
2931         struct e1000_hw *hw = &adapter->hw;
2932         struct pci_dev *pdev = adapter->pdev;
2933         unsigned long irq_flags;
2934         u16 phy_tmp;
2935
2936 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
2937
2938         /*
2939          * Prevent stats update while adapter is being reset, or if the pci
2940          * connection is down.
2941          */
2942         if (adapter->link_speed == 0)
2943                 return;
2944         if (pci_channel_offline(pdev))
2945                 return;
2946
2947         spin_lock_irqsave(&adapter->stats_lock, irq_flags);
2948
2949         /*
2950          * these counters are modified from e1000_adjust_tbi_stats,
2951          * called from the interrupt context, so they must only
2952          * be written while holding adapter->stats_lock
2953          */
2954
2955         adapter->stats.crcerrs += er32(CRCERRS);
2956         adapter->stats.gprc += er32(GPRC);
2957         adapter->stats.gorc += er32(GORCL);
2958         er32(GORCH); /* Clear gorc */
2959         adapter->stats.bprc += er32(BPRC);
2960         adapter->stats.mprc += er32(MPRC);
2961         adapter->stats.roc += er32(ROC);
2962
2963         adapter->stats.mpc += er32(MPC);
2964         adapter->stats.scc += er32(SCC);
2965         adapter->stats.ecol += er32(ECOL);
2966         adapter->stats.mcc += er32(MCC);
2967         adapter->stats.latecol += er32(LATECOL);
2968         adapter->stats.dc += er32(DC);
2969         adapter->stats.xonrxc += er32(XONRXC);
2970         adapter->stats.xontxc += er32(XONTXC);
2971         adapter->stats.xoffrxc += er32(XOFFRXC);
2972         adapter->stats.xofftxc += er32(XOFFTXC);
2973         adapter->stats.gptc += er32(GPTC);
2974         adapter->stats.gotc += er32(GOTCL);
2975         er32(GOTCH); /* Clear gotc */
2976         adapter->stats.rnbc += er32(RNBC);
2977         adapter->stats.ruc += er32(RUC);
2978
2979         adapter->stats.mptc += er32(MPTC);
2980         adapter->stats.bptc += er32(BPTC);
2981
2982         /* used for adaptive IFS */
2983
2984         hw->mac.tx_packet_delta = er32(TPT);
2985         adapter->stats.tpt += hw->mac.tx_packet_delta;
2986         hw->mac.collision_delta = er32(COLC);
2987         adapter->stats.colc += hw->mac.collision_delta;
2988
2989         adapter->stats.algnerrc += er32(ALGNERRC);
2990         adapter->stats.rxerrc += er32(RXERRC);
2991         adapter->stats.tncrs += er32(TNCRS);
2992         adapter->stats.cexterr += er32(CEXTERR);
2993         adapter->stats.tsctc += er32(TSCTC);
2994         adapter->stats.tsctfc += er32(TSCTFC);
2995
2996         /* Fill out the OS statistics structure */
2997         adapter->net_stats.multicast = adapter->stats.mprc;
2998         adapter->net_stats.collisions = adapter->stats.colc;
2999
3000         /* Rx Errors */
3001
3002         /*
3003          * RLEC on some newer hardware can be incorrect so build
3004          * our own version based on RUC and ROC
3005          */
3006         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3007                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3008                 adapter->stats.ruc + adapter->stats.roc +
3009                 adapter->stats.cexterr;
3010         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3011                                               adapter->stats.roc;
3012         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3013         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3014         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3015
3016         /* Tx Errors */
3017         adapter->net_stats.tx_errors = adapter->stats.ecol +
3018                                        adapter->stats.latecol;
3019         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3020         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3021         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3022
3023         /* Tx Dropped needs to be maintained elsewhere */
3024
3025         /* Phy Stats */
3026         if (hw->phy.media_type == e1000_media_type_copper) {
3027                 if ((adapter->link_speed == SPEED_1000) &&
3028                    (!e1e_rphy(hw, PHY_1000T_STATUS, &phy_tmp))) {
3029                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3030                         adapter->phy_stats.idle_errors += phy_tmp;
3031                 }
3032         }
3033
3034         /* Management Stats */
3035         adapter->stats.mgptc += er32(MGTPTC);
3036         adapter->stats.mgprc += er32(MGTPRC);
3037         adapter->stats.mgpdc += er32(MGTPDC);
3038
3039         spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
3040 }
3041
3042 /**
3043  * e1000_phy_read_status - Update the PHY register status snapshot
3044  * @adapter: board private structure
3045  **/
3046 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3047 {
3048         struct e1000_hw *hw = &adapter->hw;
3049         struct e1000_phy_regs *phy = &adapter->phy_regs;
3050         int ret_val;
3051         unsigned long irq_flags;
3052
3053
3054         spin_lock_irqsave(&adapter->stats_lock, irq_flags);
3055
3056         if ((er32(STATUS) & E1000_STATUS_LU) &&
3057             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3058                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3059                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3060                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3061                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3062                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3063                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3064                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3065                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3066                 if (ret_val)
3067                         e_warn("Error reading PHY register\n");
3068         } else {
3069                 /*
3070                  * Do not read PHY registers if link is not up
3071                  * Set values to typical power-on defaults
3072                  */
3073                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3074                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3075                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3076                              BMSR_ERCAP);
3077                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3078                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3079                 phy->lpa = 0;
3080                 phy->expansion = EXPANSION_ENABLENPAGE;
3081                 phy->ctrl1000 = ADVERTISE_1000FULL;
3082                 phy->stat1000 = 0;
3083                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3084         }
3085
3086         spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
3087 }
3088
3089 static void e1000_print_link_info(struct e1000_adapter *adapter)
3090 {
3091         struct e1000_hw *hw = &adapter->hw;
3092         u32 ctrl = er32(CTRL);
3093
3094         e_info("Link is Up %d Mbps %s, Flow Control: %s\n",
3095                adapter->link_speed,
3096                (adapter->link_duplex == FULL_DUPLEX) ?
3097                                 "Full Duplex" : "Half Duplex",
3098                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3099                                 "RX/TX" :
3100                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3101                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3102 }
3103
3104 static bool e1000_has_link(struct e1000_adapter *adapter)
3105 {
3106         struct e1000_hw *hw = &adapter->hw;
3107         bool link_active = 0;
3108         s32 ret_val = 0;
3109
3110         /*
3111          * get_link_status is set on LSC (link status) interrupt or
3112          * Rx sequence error interrupt.  get_link_status will stay
3113          * false until the check_for_link establishes link
3114          * for copper adapters ONLY
3115          */
3116         switch (hw->phy.media_type) {
3117         case e1000_media_type_copper:
3118                 if (hw->mac.get_link_status) {
3119                         ret_val = hw->mac.ops.check_for_link(hw);
3120                         link_active = !hw->mac.get_link_status;
3121                 } else {
3122                         link_active = 1;
3123                 }
3124                 break;
3125         case e1000_media_type_fiber:
3126                 ret_val = hw->mac.ops.check_for_link(hw);
3127                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3128                 break;
3129         case e1000_media_type_internal_serdes:
3130                 ret_val = hw->mac.ops.check_for_link(hw);
3131                 link_active = adapter->hw.mac.serdes_has_link;
3132                 break;
3133         default:
3134         case e1000_media_type_unknown:
3135                 break;
3136         }
3137
3138         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3139             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3140                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3141                 e_info("Gigabit has been disabled, downgrading speed\n");
3142         }
3143
3144         return link_active;
3145 }
3146
3147 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3148 {
3149         /* make sure the receive unit is started */
3150         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3151             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3152                 struct e1000_hw *hw = &adapter->hw;
3153                 u32 rctl = er32(RCTL);
3154                 ew32(RCTL, rctl | E1000_RCTL_EN);
3155                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3156         }
3157 }
3158
3159 /**
3160  * e1000_watchdog - Timer Call-back
3161  * @data: pointer to adapter cast into an unsigned long
3162  **/
3163 static void e1000_watchdog(unsigned long data)
3164 {
3165         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3166
3167         /* Do the rest outside of interrupt context */
3168         schedule_work(&adapter->watchdog_task);
3169
3170         /* TODO: make this use queue_delayed_work() */
3171 }
3172
3173 static void e1000_watchdog_task(struct work_struct *work)
3174 {
3175         struct e1000_adapter *adapter = container_of(work,
3176                                         struct e1000_adapter, watchdog_task);
3177         struct net_device *netdev = adapter->netdev;
3178         struct e1000_mac_info *mac = &adapter->hw.mac;
3179         struct e1000_ring *tx_ring = adapter->tx_ring;
3180         struct e1000_hw *hw = &adapter->hw;
3181         u32 link, tctl;
3182         int tx_pending = 0;
3183
3184         link = e1000_has_link(adapter);
3185         if ((netif_carrier_ok(netdev)) && link) {
3186                 e1000e_enable_receives(adapter);
3187                 goto link_up;
3188         }
3189
3190         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3191             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3192                 e1000_update_mng_vlan(adapter);
3193
3194         if (link) {
3195                 if (!netif_carrier_ok(netdev)) {
3196                         bool txb2b = 1;
3197                         /* update snapshot of PHY registers on LSC */
3198                         e1000_phy_read_status(adapter);
3199                         mac->ops.get_link_up_info(&adapter->hw,
3200                                                    &adapter->link_speed,
3201                                                    &adapter->link_duplex);
3202                         e1000_print_link_info(adapter);
3203                         /*
3204                          * tweak tx_queue_len according to speed/duplex
3205                          * and adjust the timeout factor
3206                          */
3207                         netdev->tx_queue_len = adapter->tx_queue_len;
3208                         adapter->tx_timeout_factor = 1;
3209                         switch (adapter->link_speed) {
3210                         case SPEED_10:
3211                                 txb2b = 0;
3212                                 netdev->tx_queue_len = 10;
3213                                 adapter->tx_timeout_factor = 16;
3214                                 break;
3215                         case SPEED_100:
3216                                 txb2b = 0;
3217                                 netdev->tx_queue_len = 100;
3218                                 /* maybe add some timeout factor ? */
3219                                 break;
3220                         }
3221
3222                         /*
3223                          * workaround: re-program speed mode bit after
3224                          * link-up event
3225                          */
3226                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3227                             !txb2b) {
3228                                 u32 tarc0;
3229                                 tarc0 = er32(TARC(0));
3230                                 tarc0 &= ~SPEED_MODE_BIT;
3231                                 ew32(TARC(0), tarc0);
3232                         }
3233
3234                         /*
3235                          * disable TSO for pcie and 10/100 speeds, to avoid
3236                          * some hardware issues
3237                          */
3238                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3239                                 switch (adapter->link_speed) {
3240                                 case SPEED_10:
3241                                 case SPEED_100:
3242                                         e_info("10/100 speed: disabling TSO\n");
3243                                         netdev->features &= ~NETIF_F_TSO;
3244                                         netdev->features &= ~NETIF_F_TSO6;
3245                                         break;
3246                                 case SPEED_1000:
3247                                         netdev->features |= NETIF_F_TSO;
3248                                         netdev->features |= NETIF_F_TSO6;
3249                                         break;
3250                                 default:
3251                                         /* oops */
3252                                         break;
3253                                 }
3254                         }
3255
3256                         /*
3257                          * enable transmits in the hardware, need to do this
3258                          * after setting TARC(0)
3259                          */
3260                         tctl = er32(TCTL);
3261                         tctl |= E1000_TCTL_EN;
3262                         ew32(TCTL, tctl);
3263
3264                         netif_carrier_on(netdev);
3265                         netif_tx_wake_all_queues(netdev);
3266
3267                         if (!test_bit(__E1000_DOWN, &adapter->state))
3268                                 mod_timer(&adapter->phy_info_timer,
3269                                           round_jiffies(jiffies + 2 * HZ));
3270                 }
3271         } else {
3272                 if (netif_carrier_ok(netdev)) {
3273                         adapter->link_speed = 0;
3274                         adapter->link_duplex = 0;
3275                         e_info("Link is Down\n");
3276                         netif_carrier_off(netdev);
3277                         netif_tx_stop_all_queues(netdev);
3278                         if (!test_bit(__E1000_DOWN, &adapter->state))
3279                                 mod_timer(&adapter->phy_info_timer,
3280                                           round_jiffies(jiffies + 2 * HZ));
3281
3282                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3283                                 schedule_work(&adapter->reset_task);
3284                 }
3285         }
3286
3287 link_up:
3288         e1000e_update_stats(adapter);
3289
3290         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3291         adapter->tpt_old = adapter->stats.tpt;
3292         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3293         adapter->colc_old = adapter->stats.colc;
3294
3295         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3296         adapter->gorc_old = adapter->stats.gorc;
3297         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3298         adapter->gotc_old = adapter->stats.gotc;
3299
3300         e1000e_update_adaptive(&adapter->hw);
3301
3302         if (!netif_carrier_ok(netdev)) {
3303                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3304                                tx_ring->count);
3305                 if (tx_pending) {
3306                         /*
3307                          * We've lost link, so the controller stops DMA,
3308                          * but we've got queued Tx work that's never going
3309                          * to get done, so reset controller to flush Tx.
3310                          * (Do the reset outside of interrupt context).
3311                          */
3312                         adapter->tx_timeout_count++;
3313                         schedule_work(&adapter->reset_task);
3314                 }
3315         }
3316
3317         /* Cause software interrupt to ensure Rx ring is cleaned */
3318         ew32(ICS, E1000_ICS_RXDMT0);
3319
3320         /* Force detection of hung controller every watchdog period */
3321         adapter->detect_tx_hung = 1;
3322
3323         /*
3324          * With 82571 controllers, LAA may be overwritten due to controller
3325          * reset from the other port. Set the appropriate LAA in RAR[0]
3326          */
3327         if (e1000e_get_laa_state_82571(hw))
3328                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3329
3330         /* Reset the timer */
3331         if (!test_bit(__E1000_DOWN, &adapter->state))
3332                 mod_timer(&adapter->watchdog_timer,
3333                           round_jiffies(jiffies + 2 * HZ));
3334 }
3335
3336 #define E1000_TX_FLAGS_CSUM             0x00000001
3337 #define E1000_TX_FLAGS_VLAN             0x00000002
3338 #define E1000_TX_FLAGS_TSO              0x00000004
3339 #define E1000_TX_FLAGS_IPV4             0x00000008
3340 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3341 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3342
3343 static int e1000_tso(struct e1000_adapter *adapter,
3344                      struct sk_buff *skb)
3345 {
3346         struct e1000_ring *tx_ring = adapter->tx_ring;
3347         struct e1000_context_desc *context_desc;
3348         struct e1000_buffer *buffer_info;
3349         unsigned int i;
3350         u32 cmd_length = 0;
3351         u16 ipcse = 0, tucse, mss;
3352         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3353         int err;
3354
3355         if (skb_is_gso(skb)) {
3356                 if (skb_header_cloned(skb)) {
3357                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3358                         if (err)
3359                                 return err;
3360                 }
3361
3362                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3363                 mss = skb_shinfo(skb)->gso_size;
3364                 if (skb->protocol == htons(ETH_P_IP)) {
3365                         struct iphdr *iph = ip_hdr(skb);
3366                         iph->tot_len = 0;
3367                         iph->check = 0;
3368                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3369                                                                  iph->daddr, 0,
3370                                                                  IPPROTO_TCP,
3371                                                                  0);
3372                         cmd_length = E1000_TXD_CMD_IP;
3373                         ipcse = skb_transport_offset(skb) - 1;
3374                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3375                         ipv6_hdr(skb)->payload_len = 0;
3376                         tcp_hdr(skb)->check =
3377                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3378                                                  &ipv6_hdr(skb)->daddr,
3379                                                  0, IPPROTO_TCP, 0);
3380                         ipcse = 0;
3381                 }
3382                 ipcss = skb_network_offset(skb);
3383                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3384                 tucss = skb_transport_offset(skb);
3385                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3386                 tucse = 0;
3387
3388                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3389                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3390
3391                 i = tx_ring->next_to_use;
3392                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3393                 buffer_info = &tx_ring->buffer_info[i];
3394
3395                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3396                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3397                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3398                 context_desc->upper_setup.tcp_fields.tucss = tucss;
3399                 context_desc->upper_setup.tcp_fields.tucso = tucso;
3400                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3401                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3402                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3403                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3404
3405                 buffer_info->time_stamp = jiffies;
3406                 buffer_info->next_to_watch = i;
3407
3408                 i++;
3409                 if (i == tx_ring->count)
3410                         i = 0;
3411                 tx_ring->next_to_use = i;
3412
3413                 return 1;
3414         }
3415
3416         return 0;
3417 }
3418
3419 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3420 {
3421         struct e1000_ring *tx_ring = adapter->tx_ring;
3422         struct e1000_context_desc *context_desc;
3423         struct e1000_buffer *buffer_info;
3424         unsigned int i;
3425         u8 css;
3426
3427         if (skb->ip_summed == CHECKSUM_PARTIAL) {
3428                 css = skb_transport_offset(skb);
3429
3430                 i = tx_ring->next_to_use;
3431                 buffer_info = &tx_ring->buffer_info[i];
3432                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3433
3434                 context_desc->lower_setup.ip_config = 0;
3435                 context_desc->upper_setup.tcp_fields.tucss = css;
3436                 context_desc->upper_setup.tcp_fields.tucso =
3437                                         css + skb->csum_offset;
3438                 context_desc->upper_setup.tcp_fields.tucse = 0;
3439                 context_desc->tcp_seg_setup.data = 0;
3440                 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
3441
3442                 buffer_info->time_stamp = jiffies;
3443                 buffer_info->next_to_watch = i;
3444
3445                 i++;
3446                 if (i == tx_ring->count)
3447                         i = 0;
3448                 tx_ring->next_to_use = i;
3449
3450                 return 1;
3451         }
3452
3453         return 0;
3454 }
3455
3456 #define E1000_MAX_PER_TXD       8192
3457 #define E1000_MAX_TXD_PWR       12
3458
3459 static int e1000_tx_map(struct e1000_adapter *adapter,
3460                         struct sk_buff *skb, unsigned int first,
3461                         unsigned int max_per_txd, unsigned int nr_frags,
3462                         unsigned int mss)
3463 {
3464         struct e1000_ring *tx_ring = adapter->tx_ring;
3465         struct e1000_buffer *buffer_info;
3466         unsigned int len = skb->len - skb->data_len;
3467         unsigned int offset = 0, size, count = 0, i;
3468         unsigned int f;
3469
3470         i = tx_ring->next_to_use;
3471
3472         while (len) {
3473                 buffer_info = &tx_ring->buffer_info[i];
3474                 size = min(len, max_per_txd);
3475
3476                 /* Workaround for premature desc write-backs
3477                  * in TSO mode.  Append 4-byte sentinel desc */
3478                 if (mss && !nr_frags && size == len && size > 8)
3479                         size -= 4;
3480
3481                 buffer_info->length = size;
3482                 /* set time_stamp *before* dma to help avoid a possible race */
3483                 buffer_info->time_stamp = jiffies;
3484                 buffer_info->dma =
3485                         pci_map_single(adapter->pdev,
3486                                 skb->data + offset,
3487                                 size,
3488                                 PCI_DMA_TODEVICE);
3489                 if (pci_dma_mapping_error(adapter->pdev, buffer_info->dma)) {
3490                         dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3491                         adapter->tx_dma_failed++;
3492                         return -1;
3493                 }
3494                 buffer_info->next_to_watch = i;
3495
3496                 len -= size;
3497                 offset += size;
3498                 count++;
3499                 i++;
3500                 if (i == tx_ring->count)
3501                         i = 0;
3502         }
3503
3504         for (f = 0; f < nr_frags; f++) {
3505                 struct skb_frag_struct *frag;
3506
3507                 frag = &skb_shinfo(skb)->frags[f];
3508                 len = frag->size;
3509                 offset = frag->page_offset;
3510
3511                 while (len) {
3512                         buffer_info = &tx_ring->buffer_info[i];
3513                         size = min(len, max_per_txd);
3514                         /* Workaround for premature desc write-backs
3515                          * in TSO mode.  Append 4-byte sentinel desc */
3516                         if (mss && f == (nr_frags-1) && size == len && size > 8)
3517                                 size -= 4;
3518
3519                         buffer_info->length = size;
3520                         buffer_info->time_stamp = jiffies;
3521                         buffer_info->dma =
3522                                 pci_map_page(adapter->pdev,
3523                                         frag->page,
3524                                         offset,
3525                                         size,
3526                                         PCI_DMA_TODEVICE);
3527                         if (pci_dma_mapping_error(adapter->pdev,
3528                                                   buffer_info->dma)) {
3529                                 dev_err(&adapter->pdev->dev,
3530                                         "TX DMA page map failed\n");
3531                                 adapter->tx_dma_failed++;
3532                                 return -1;
3533                         }
3534
3535                         buffer_info->next_to_watch = i;
3536
3537                         len -= size;
3538                         offset += size;
3539                         count++;
3540
3541                         i++;
3542                         if (i == tx_ring->count)
3543                                 i = 0;
3544                 }
3545         }
3546
3547         if (i == 0)
3548                 i = tx_ring->count - 1;
3549         else
3550                 i--;
3551
3552         tx_ring->buffer_info[i].skb = skb;
3553         tx_ring->buffer_info[first].next_to_watch = i;
3554
3555         return count;
3556 }
3557
3558 static void e1000_tx_queue(struct e1000_adapter *adapter,
3559                            int tx_flags, int count)
3560 {
3561         struct e1000_ring *tx_ring = adapter->tx_ring;
3562         struct e1000_tx_desc *tx_desc = NULL;
3563         struct e1000_buffer *buffer_info;
3564         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3565         unsigned int i;
3566
3567         if (tx_flags & E1000_TX_FLAGS_TSO) {
3568                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3569                              E1000_TXD_CMD_TSE;
3570                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3571
3572                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3573                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3574         }
3575
3576         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3577                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3578                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3579         }
3580
3581         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3582                 txd_lower |= E1000_TXD_CMD_VLE;
3583                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3584         }
3585
3586         i = tx_ring->next_to_use;
3587
3588         while (count--) {
3589                 buffer_info = &tx_ring->buffer_info[i];
3590                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3591                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3592                 tx_desc->lower.data =
3593                         cpu_to_le32(txd_lower | buffer_info->length);
3594                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3595
3596                 i++;
3597                 if (i == tx_ring->count)
3598                         i = 0;
3599         }
3600
3601         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3602
3603         /*
3604          * Force memory writes to complete before letting h/w
3605          * know there are new descriptors to fetch.  (Only
3606          * applicable for weak-ordered memory model archs,
3607          * such as IA-64).
3608          */
3609         wmb();
3610
3611         tx_ring->next_to_use = i;
3612         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3613         /*
3614          * we need this if more than one processor can write to our tail
3615          * at a time, it synchronizes IO on IA64/Altix systems
3616          */
3617         mmiowb();
3618 }
3619
3620 #define MINIMUM_DHCP_PACKET_SIZE 282
3621 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3622                                     struct sk_buff *skb)
3623 {
3624         struct e1000_hw *hw =  &adapter->hw;
3625         u16 length, offset;
3626
3627         if (vlan_tx_tag_present(skb)) {
3628                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3629                     && (adapter->hw.mng_cookie.status &
3630                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3631                         return 0;
3632         }
3633
3634         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3635                 return 0;
3636
3637         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3638                 return 0;
3639
3640         {
3641                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3642                 struct udphdr *udp;
3643
3644                 if (ip->protocol != IPPROTO_UDP)
3645                         return 0;
3646
3647                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3648                 if (ntohs(udp->dest) != 67)
3649                         return 0;
3650
3651                 offset = (u8 *)udp + 8 - skb->data;
3652                 length = skb->len - offset;
3653                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
3654         }
3655
3656         return 0;
3657 }
3658
3659 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3660 {
3661         struct e1000_adapter *adapter = netdev_priv(netdev);
3662
3663         netif_stop_queue(netdev);
3664         /*
3665          * Herbert's original patch had:
3666          *  smp_mb__after_netif_stop_queue();
3667          * but since that doesn't exist yet, just open code it.
3668          */
3669         smp_mb();
3670
3671         /*
3672          * We need to check again in a case another CPU has just
3673          * made room available.
3674          */
3675         if (e1000_desc_unused(adapter->tx_ring) < size)
3676                 return -EBUSY;
3677
3678         /* A reprieve! */
3679         netif_start_queue(netdev);
3680         ++adapter->restart_queue;
3681         return 0;
3682 }
3683
3684 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
3685 {
3686         struct e1000_adapter *adapter = netdev_priv(netdev);
3687
3688         if (e1000_desc_unused(adapter->tx_ring) >= size)
3689                 return 0;
3690         return __e1000_maybe_stop_tx(netdev, size);
3691 }
3692
3693 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3694 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3695 {
3696         struct e1000_adapter *adapter = netdev_priv(netdev);
3697         struct e1000_ring *tx_ring = adapter->tx_ring;
3698         unsigned int first;
3699         unsigned int max_per_txd = E1000_MAX_PER_TXD;
3700         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3701         unsigned int tx_flags = 0;
3702         unsigned int len = skb->len - skb->data_len;
3703         unsigned long irq_flags;
3704         unsigned int nr_frags;
3705         unsigned int mss;
3706         int count = 0;
3707         int tso;
3708         unsigned int f;
3709
3710         if (test_bit(__E1000_DOWN, &adapter->state)) {
3711                 dev_kfree_skb_any(skb);
3712                 return NETDEV_TX_OK;
3713         }
3714
3715         if (skb->len <= 0) {
3716                 dev_kfree_skb_any(skb);
3717                 return NETDEV_TX_OK;
3718         }
3719
3720         mss = skb_shinfo(skb)->gso_size;
3721         /*
3722          * The controller does a simple calculation to
3723          * make sure there is enough room in the FIFO before
3724          * initiating the DMA for each buffer.  The calc is:
3725          * 4 = ceil(buffer len/mss).  To make sure we don't
3726          * overrun the FIFO, adjust the max buffer len if mss
3727          * drops.
3728          */
3729         if (mss) {
3730                 u8 hdr_len;
3731                 max_per_txd = min(mss << 2, max_per_txd);
3732                 max_txd_pwr = fls(max_per_txd) - 1;
3733
3734                 /*
3735                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
3736                  * points to just header, pull a few bytes of payload from
3737                  * frags into skb->data
3738                  */
3739                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3740                 /*
3741                  * we do this workaround for ES2LAN, but it is un-necessary,
3742                  * avoiding it could save a lot of cycles
3743                  */
3744                 if (skb->data_len && (hdr_len == len)) {
3745                         unsigned int pull_size;
3746
3747                         pull_size = min((unsigned int)4, skb->data_len);
3748                         if (!__pskb_pull_tail(skb, pull_size)) {
3749                                 e_err("__pskb_pull_tail failed.\n");
3750                                 dev_kfree_skb_any(skb);
3751                                 return NETDEV_TX_OK;
3752                         }
3753                         len = skb->len - skb->data_len;
3754                 }
3755         }
3756
3757         /* reserve a descriptor for the offload context */
3758         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3759                 count++;
3760         count++;
3761
3762         count += TXD_USE_COUNT(len, max_txd_pwr);
3763
3764         nr_frags = skb_shinfo(skb)->nr_frags;
3765         for (f = 0; f < nr_frags; f++)
3766                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3767                                        max_txd_pwr);
3768
3769         if (adapter->hw.mac.tx_pkt_filtering)
3770                 e1000_transfer_dhcp_info(adapter, skb);
3771
3772         if (!spin_trylock_irqsave(&adapter->tx_queue_lock, irq_flags))
3773                 /* Collision - tell upper layer to requeue */
3774                 return NETDEV_TX_LOCKED;
3775
3776         /*
3777          * need: count + 2 desc gap to keep tail from touching
3778          * head, otherwise try next time
3779          */
3780         if (e1000_maybe_stop_tx(netdev, count + 2)) {
3781                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3782                 return NETDEV_TX_BUSY;
3783         }
3784
3785         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3786                 tx_flags |= E1000_TX_FLAGS_VLAN;
3787                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3788         }
3789
3790         first = tx_ring->next_to_use;
3791
3792         tso = e1000_tso(adapter, skb);
3793         if (tso < 0) {
3794                 dev_kfree_skb_any(skb);
3795                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3796                 return NETDEV_TX_OK;
3797         }
3798
3799         if (tso)
3800                 tx_flags |= E1000_TX_FLAGS_TSO;
3801         else if (e1000_tx_csum(adapter, skb))
3802                 tx_flags |= E1000_TX_FLAGS_CSUM;
3803
3804         /*
3805          * Old method was to assume IPv4 packet by default if TSO was enabled.
3806          * 82571 hardware supports TSO capabilities for IPv6 as well...
3807          * no longer assume, we must.
3808          */
3809         if (skb->protocol == htons(ETH_P_IP))
3810                 tx_flags |= E1000_TX_FLAGS_IPV4;
3811
3812         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
3813         if (count < 0) {
3814                 /* handle pci_map_single() error in e1000_tx_map */
3815                 dev_kfree_skb_any(skb);
3816                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3817                 return NETDEV_TX_OK;
3818         }
3819
3820         e1000_tx_queue(adapter, tx_flags, count);
3821
3822         netdev->trans_start = jiffies;
3823
3824         /* Make sure there is space in the ring for the next send. */
3825         e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
3826
3827         spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3828         return NETDEV_TX_OK;
3829 }
3830
3831 /**
3832  * e1000_tx_timeout - Respond to a Tx Hang
3833  * @netdev: network interface device structure
3834  **/
3835 static void e1000_tx_timeout(struct net_device *netdev)
3836 {
3837         struct e1000_adapter *adapter = netdev_priv(netdev);
3838
3839         /* Do the reset outside of interrupt context */
3840         adapter->tx_timeout_count++;
3841         schedule_work(&adapter->reset_task);
3842 }
3843
3844 static void e1000_reset_task(struct work_struct *work)
3845 {
3846         struct e1000_adapter *adapter;
3847         adapter = container_of(work, struct e1000_adapter, reset_task);
3848
3849         e1000e_reinit_locked(adapter);
3850 }
3851
3852 /**
3853  * e1000_get_stats - Get System Network Statistics
3854  * @netdev: network interface device structure
3855  *
3856  * Returns the address of the device statistics structure.
3857  * The statistics are actually updated from the timer callback.
3858  **/
3859 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3860 {
3861         struct e1000_adapter *adapter = netdev_priv(netdev);
3862
3863         /* only return the current stats */
3864         return &adapter->net_stats;
3865 }
3866
3867 /**
3868  * e1000_change_mtu - Change the Maximum Transfer Unit
3869  * @netdev: network interface device structure
3870  * @new_mtu: new value for maximum frame size
3871  *
3872  * Returns 0 on success, negative on failure
3873  **/
3874 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3875 {
3876         struct e1000_adapter *adapter = netdev_priv(netdev);
3877         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3878
3879         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
3880             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3881                 e_err("Invalid MTU setting\n");
3882                 return -EINVAL;
3883         }
3884
3885         /* Jumbo frame size limits */
3886         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
3887                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
3888                         e_err("Jumbo Frames not supported.\n");
3889                         return -EINVAL;
3890                 }
3891                 if (adapter->hw.phy.type == e1000_phy_ife) {
3892                         e_err("Jumbo Frames not supported.\n");
3893                         return -EINVAL;
3894                 }
3895         }
3896
3897 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3898         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3899                 e_err("MTU > 9216 not supported.\n");
3900                 return -EINVAL;
3901         }
3902
3903         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3904                 msleep(1);
3905         /* e1000e_down has a dependency on max_frame_size */
3906         adapter->max_frame_size = max_frame;
3907         if (netif_running(netdev))
3908                 e1000e_down(adapter);
3909
3910         /*
3911          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3912          * means we reserve 2 more, this pushes us to allocate from the next
3913          * larger slab size.
3914          * i.e. RXBUFFER_2048 --> size-4096 slab
3915          * However with the new *_jumbo_rx* routines, jumbo receives will use
3916          * fragmented skbs
3917          */
3918
3919         if (max_frame <= 256)
3920                 adapter->rx_buffer_len = 256;
3921         else if (max_frame <= 512)
3922                 adapter->rx_buffer_len = 512;
3923         else if (max_frame <= 1024)
3924                 adapter->rx_buffer_len = 1024;
3925         else if (max_frame <= 2048)
3926                 adapter->rx_buffer_len = 2048;
3927         else
3928                 adapter->rx_buffer_len = 4096;
3929
3930         /* adjust allocation if LPE protects us, and we aren't using SBP */
3931         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3932              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
3933                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
3934                                          + ETH_FCS_LEN;
3935
3936         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
3937         netdev->mtu = new_mtu;
3938
3939         if (netif_running(netdev))
3940                 e1000e_up(adapter);
3941         else
3942                 e1000e_reset(adapter);
3943
3944         clear_bit(__E1000_RESETTING, &adapter->state);
3945
3946         return 0;
3947 }
3948
3949 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
3950                            int cmd)
3951 {
3952         struct e1000_adapter *adapter = netdev_priv(netdev);
3953         struct mii_ioctl_data *data = if_mii(ifr);
3954
3955         if (adapter->hw.phy.media_type != e1000_media_type_copper)
3956                 return -EOPNOTSUPP;
3957
3958         switch (cmd) {
3959         case SIOCGMIIPHY:
3960                 data->phy_id = adapter->hw.phy.addr;
3961                 break;
3962         case SIOCGMIIREG:
3963                 if (!capable(CAP_NET_ADMIN))
3964                         return -EPERM;
3965                 switch (data->reg_num & 0x1F) {
3966                 case MII_BMCR:
3967                         data->val_out = adapter->phy_regs.bmcr;
3968                         break;
3969                 case MII_BMSR:
3970                         data->val_out = adapter->phy_regs.bmsr;
3971                         break;
3972                 case MII_PHYSID1:
3973                         data->val_out = (adapter->hw.phy.id >> 16);
3974                         break;
3975                 case MII_PHYSID2:
3976                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
3977                         break;
3978                 case MII_ADVERTISE:
3979                         data->val_out = adapter->phy_regs.advertise;
3980                         break;
3981                 case MII_LPA:
3982                         data->val_out = adapter->phy_regs.lpa;
3983                         break;
3984                 case MII_EXPANSION:
3985                         data->val_out = adapter->phy_regs.expansion;
3986                         break;
3987                 case MII_CTRL1000:
3988                         data->val_out = adapter->phy_regs.ctrl1000;
3989                         break;
3990                 case MII_STAT1000:
3991                         data->val_out = adapter->phy_regs.stat1000;
3992                         break;
3993                 case MII_ESTATUS:
3994                         data->val_out = adapter->phy_regs.estatus;
3995                         break;
3996                 default:
3997                         return -EIO;
3998                 }
3999                 break;
4000         case SIOCSMIIREG:
4001         default:
4002                 return -EOPNOTSUPP;
4003         }
4004         return 0;
4005 }
4006
4007 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4008 {
4009         switch (cmd) {
4010         case SIOCGMIIPHY:
4011         case SIOCGMIIREG:
4012         case SIOCSMIIREG:
4013                 return e1000_mii_ioctl(netdev, ifr, cmd);
4014         default:
4015                 return -EOPNOTSUPP;
4016         }
4017 }
4018
4019 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4020 {
4021         struct net_device *netdev = pci_get_drvdata(pdev);
4022         struct e1000_adapter *adapter = netdev_priv(netdev);
4023         struct e1000_hw *hw = &adapter->hw;
4024         u32 ctrl, ctrl_ext, rctl, status;
4025         u32 wufc = adapter->wol;
4026         int retval = 0;
4027
4028         netif_device_detach(netdev);
4029
4030         if (netif_running(netdev)) {
4031                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4032                 e1000e_down(adapter);
4033                 e1000_free_irq(adapter);
4034         }
4035
4036         retval = pci_save_state(pdev);
4037         if (retval)
4038                 return retval;
4039
4040         status = er32(STATUS);
4041         if (status & E1000_STATUS_LU)
4042                 wufc &= ~E1000_WUFC_LNKC;
4043
4044         if (wufc) {
4045                 e1000_setup_rctl(adapter);
4046                 e1000_set_multi(netdev);
4047
4048                 /* turn on all-multi mode if wake on multicast is enabled */
4049                 if (wufc & E1000_WUFC_MC) {
4050                         rctl = er32(RCTL);
4051                         rctl |= E1000_RCTL_MPE;
4052                         ew32(RCTL, rctl);
4053                 }
4054
4055                 ctrl = er32(CTRL);
4056                 /* advertise wake from D3Cold */
4057                 #define E1000_CTRL_ADVD3WUC 0x00100000
4058                 /* phy power management enable */
4059                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4060                 ctrl |= E1000_CTRL_ADVD3WUC |
4061                         E1000_CTRL_EN_PHY_PWR_MGMT;
4062                 ew32(CTRL, ctrl);
4063
4064                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4065                     adapter->hw.phy.media_type ==
4066                     e1000_media_type_internal_serdes) {
4067                         /* keep the laser running in D3 */
4068                         ctrl_ext = er32(CTRL_EXT);
4069                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4070                         ew32(CTRL_EXT, ctrl_ext);
4071                 }
4072
4073                 if (adapter->flags & FLAG_IS_ICH)
4074                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4075
4076                 /* Allow time for pending master requests to run */
4077                 e1000e_disable_pcie_master(&adapter->hw);
4078
4079                 ew32(WUC, E1000_WUC_PME_EN);
4080                 ew32(WUFC, wufc);
4081                 pci_enable_wake(pdev, PCI_D3hot, 1);
4082                 pci_enable_wake(pdev, PCI_D3cold, 1);
4083         } else {
4084                 ew32(WUC, 0);
4085                 ew32(WUFC, 0);
4086                 pci_enable_wake(pdev, PCI_D3hot, 0);
4087                 pci_enable_wake(pdev, PCI_D3cold, 0);
4088         }
4089
4090         /* make sure adapter isn't asleep if manageability is enabled */
4091         if (adapter->flags & FLAG_MNG_PT_ENABLED) {
4092                 pci_enable_wake(pdev, PCI_D3hot, 1);
4093                 pci_enable_wake(pdev, PCI_D3cold, 1);
4094         }
4095
4096         if (adapter->hw.phy.type == e1000_phy_igp_3)
4097                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4098
4099         /*
4100          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4101          * would have already happened in close and is redundant.
4102          */
4103         e1000_release_hw_control(adapter);
4104
4105         pci_disable_device(pdev);
4106
4107         pci_set_power_state(pdev, pci_choose_state(pdev, state));
4108
4109         return 0;
4110 }
4111
4112 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4113 {
4114         int pos;
4115         u16 val;
4116
4117         /*
4118          * 82573 workaround - disable L1 ASPM on mobile chipsets
4119          *
4120          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4121          * resulting in lost data or garbage information on the pci-e link
4122          * level. This could result in (false) bad EEPROM checksum errors,
4123          * long ping times (up to 2s) or even a system freeze/hang.
4124          *
4125          * Unfortunately this feature saves about 1W power consumption when
4126          * active.
4127          */
4128         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4129         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4130         if (val & 0x2) {
4131                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4132                 val &= ~0x2;
4133                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4134         }
4135 }
4136
4137 #ifdef CONFIG_PM
4138 static int e1000_resume(struct pci_dev *pdev)
4139 {
4140         struct net_device *netdev = pci_get_drvdata(pdev);
4141         struct e1000_adapter *adapter = netdev_priv(netdev);
4142         struct e1000_hw *hw = &adapter->hw;
4143         u32 err;
4144
4145         pci_set_power_state(pdev, PCI_D0);
4146         pci_restore_state(pdev);
4147         e1000e_disable_l1aspm(pdev);
4148
4149         err = pci_enable_device_mem(pdev);
4150         if (err) {
4151                 dev_err(&pdev->dev,
4152                         "Cannot enable PCI device from suspend\n");
4153                 return err;
4154         }
4155
4156         pci_set_master(pdev);
4157
4158         pci_enable_wake(pdev, PCI_D3hot, 0);
4159         pci_enable_wake(pdev, PCI_D3cold, 0);
4160
4161         if (netif_running(netdev)) {
4162                 err = e1000_request_irq(adapter);
4163                 if (err)
4164                         return err;
4165         }
4166
4167         e1000e_power_up_phy(adapter);
4168         e1000e_reset(adapter);
4169         ew32(WUS, ~0);
4170
4171         e1000_init_manageability(adapter);
4172
4173         if (netif_running(netdev))
4174                 e1000e_up(adapter);
4175
4176         netif_device_attach(netdev);
4177
4178         /*
4179          * If the controller has AMT, do not set DRV_LOAD until the interface
4180          * is up.  For all other cases, let the f/w know that the h/w is now
4181          * under the control of the driver.
4182          */
4183         if (!(adapter->flags & FLAG_HAS_AMT))
4184                 e1000_get_hw_control(adapter);
4185
4186         return 0;
4187 }
4188 #endif
4189
4190 static void e1000_shutdown(struct pci_dev *pdev)
4191 {
4192         e1000_suspend(pdev, PMSG_SUSPEND);
4193 }
4194
4195 #ifdef CONFIG_NET_POLL_CONTROLLER
4196 /*
4197  * Polling 'interrupt' - used by things like netconsole to send skbs
4198  * without having to re-enable interrupts. It's not called while
4199  * the interrupt routine is executing.
4200  */
4201 static void e1000_netpoll(struct net_device *netdev)
4202 {
4203         struct e1000_adapter *adapter = netdev_priv(netdev);
4204
4205         disable_irq(adapter->pdev->irq);
4206         e1000_intr(adapter->pdev->irq, netdev);
4207
4208         enable_irq(adapter->pdev->irq);
4209 }
4210 #endif
4211
4212 /**
4213  * e1000_io_error_detected - called when PCI error is detected
4214  * @pdev: Pointer to PCI device
4215  * @state: The current pci connection state
4216  *
4217  * This function is called after a PCI bus error affecting
4218  * this device has been detected.
4219  */
4220 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4221                                                 pci_channel_state_t state)
4222 {
4223         struct net_device *netdev = pci_get_drvdata(pdev);
4224         struct e1000_adapter *adapter = netdev_priv(netdev);
4225
4226         netif_device_detach(netdev);
4227
4228         if (netif_running(netdev))
4229                 e1000e_down(adapter);
4230         pci_disable_device(pdev);
4231
4232         /* Request a slot slot reset. */
4233         return PCI_ERS_RESULT_NEED_RESET;
4234 }
4235
4236 /**
4237  * e1000_io_slot_reset - called after the pci bus has been reset.
4238  * @pdev: Pointer to PCI device
4239  *
4240  * Restart the card from scratch, as if from a cold-boot. Implementation
4241  * resembles the first-half of the e1000_resume routine.
4242  */
4243 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4244 {
4245         struct net_device *netdev = pci_get_drvdata(pdev);
4246         struct e1000_adapter *adapter = netdev_priv(netdev);
4247         struct e1000_hw *hw = &adapter->hw;
4248         int err;
4249
4250         e1000e_disable_l1aspm(pdev);
4251         err = pci_enable_device_mem(pdev);
4252         if (err) {
4253                 dev_err(&pdev->dev,
4254                         "Cannot re-enable PCI device after reset.\n");
4255                 return PCI_ERS_RESULT_DISCONNECT;
4256         }
4257         pci_set_master(pdev);
4258         pci_restore_state(pdev);
4259
4260         pci_enable_wake(pdev, PCI_D3hot, 0);
4261         pci_enable_wake(pdev, PCI_D3cold, 0);
4262
4263         e1000e_reset(adapter);
4264         ew32(WUS, ~0);
4265
4266         return PCI_ERS_RESULT_RECOVERED;
4267 }
4268
4269 /**
4270  * e1000_io_resume - called when traffic can start flowing again.
4271  * @pdev: Pointer to PCI device
4272  *
4273  * This callback is called when the error recovery driver tells us that
4274  * its OK to resume normal operation. Implementation resembles the
4275  * second-half of the e1000_resume routine.
4276  */
4277 static void e1000_io_resume(struct pci_dev *pdev)
4278 {
4279         struct net_device *netdev = pci_get_drvdata(pdev);
4280         struct e1000_adapter *adapter = netdev_priv(netdev);
4281
4282         e1000_init_manageability(adapter);
4283
4284         if (netif_running(netdev)) {
4285                 if (e1000e_up(adapter)) {
4286                         dev_err(&pdev->dev,
4287                                 "can't bring device back up after reset\n");
4288                         return;
4289                 }
4290         }
4291
4292         netif_device_attach(netdev);
4293
4294         /*
4295          * If the controller has AMT, do not set DRV_LOAD until the interface
4296          * is up.  For all other cases, let the f/w know that the h/w is now
4297          * under the control of the driver.
4298          */
4299         if (!(adapter->flags & FLAG_HAS_AMT))
4300                 e1000_get_hw_control(adapter);
4301
4302 }
4303
4304 static void e1000_print_device_info(struct e1000_adapter *adapter)
4305 {
4306         struct e1000_hw *hw = &adapter->hw;
4307         struct net_device *netdev = adapter->netdev;
4308         u32 pba_num;
4309
4310         /* print bus type/speed/width info */
4311         e_info("(PCI Express:2.5GB/s:%s) %02x:%02x:%02x:%02x:%02x:%02x\n",
4312                /* bus width */
4313                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4314                 "Width x1"),
4315                /* MAC address */
4316                netdev->dev_addr[0], netdev->dev_addr[1],
4317                netdev->dev_addr[2], netdev->dev_addr[3],
4318                netdev->dev_addr[4], netdev->dev_addr[5]);
4319         e_info("Intel(R) PRO/%s Network Connection\n",
4320                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4321         e1000e_read_pba_num(hw, &pba_num);
4322         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4323                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4324 }
4325
4326 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4327 {
4328         struct e1000_hw *hw = &adapter->hw;
4329         int ret_val;
4330         u16 buf = 0;
4331
4332         if (hw->mac.type != e1000_82573)
4333                 return;
4334
4335         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4336         if (!(le16_to_cpu(buf) & (1 << 0))) {
4337                 /* Deep Smart Power Down (DSPD) */
4338                 e_warn("Warning: detected DSPD enabled in EEPROM\n");
4339         }
4340
4341         ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4342         if (le16_to_cpu(buf) & (3 << 2)) {
4343                 /* ASPM enable */
4344                 e_warn("Warning: detected ASPM enabled in EEPROM\n");
4345         }
4346 }
4347
4348 /**
4349  * e1000_probe - Device Initialization Routine
4350  * @pdev: PCI device information struct
4351  * @ent: entry in e1000_pci_tbl
4352  *
4353  * Returns 0 on success, negative on failure
4354  *
4355  * e1000_probe initializes an adapter identified by a pci_dev structure.
4356  * The OS initialization, configuring of the adapter private structure,
4357  * and a hardware reset occur.
4358  **/
4359 static int __devinit e1000_probe(struct pci_dev *pdev,
4360                                  const struct pci_device_id *ent)
4361 {
4362         struct net_device *netdev;
4363         struct e1000_adapter *adapter;
4364         struct e1000_hw *hw;
4365         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4366         resource_size_t mmio_start, mmio_len;
4367         resource_size_t flash_start, flash_len;
4368
4369         static int cards_found;
4370         int i, err, pci_using_dac;
4371         u16 eeprom_data = 0;
4372         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4373
4374         e1000e_disable_l1aspm(pdev);
4375
4376         err = pci_enable_device_mem(pdev);
4377         if (err)
4378                 return err;
4379
4380         pci_using_dac = 0;
4381         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
4382         if (!err) {
4383                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
4384                 if (!err)
4385                         pci_using_dac = 1;
4386         } else {
4387                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
4388                 if (err) {
4389                         err = pci_set_consistent_dma_mask(pdev,
4390                                                           DMA_32BIT_MASK);
4391                         if (err) {
4392                                 dev_err(&pdev->dev, "No usable DMA "
4393                                         "configuration, aborting\n");
4394                                 goto err_dma;
4395                         }
4396                 }
4397         }
4398
4399         err = pci_request_selected_regions(pdev,
4400                                           pci_select_bars(pdev, IORESOURCE_MEM),
4401                                           e1000e_driver_name);
4402         if (err)
4403                 goto err_pci_reg;
4404
4405         pci_set_master(pdev);
4406         pci_save_state(pdev);
4407
4408         err = -ENOMEM;
4409         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4410         if (!netdev)
4411                 goto err_alloc_etherdev;
4412
4413         SET_NETDEV_DEV(netdev, &pdev->dev);
4414
4415         pci_set_drvdata(pdev, netdev);
4416         adapter = netdev_priv(netdev);
4417         hw = &adapter->hw;
4418         adapter->netdev = netdev;
4419         adapter->pdev = pdev;
4420         adapter->ei = ei;
4421         adapter->pba = ei->pba;
4422         adapter->flags = ei->flags;
4423         adapter->hw.adapter = adapter;
4424         adapter->hw.mac.type = ei->mac;
4425         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4426
4427         mmio_start = pci_resource_start(pdev, 0);
4428         mmio_len = pci_resource_len(pdev, 0);
4429
4430         err = -EIO;
4431         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4432         if (!adapter->hw.hw_addr)
4433                 goto err_ioremap;
4434
4435         if ((adapter->flags & FLAG_HAS_FLASH) &&
4436             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
4437                 flash_start = pci_resource_start(pdev, 1);
4438                 flash_len = pci_resource_len(pdev, 1);
4439                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
4440                 if (!adapter->hw.flash_address)
4441                         goto err_flashmap;
4442         }
4443
4444         /* construct the net_device struct */
4445         netdev->open                    = &e1000_open;
4446         netdev->stop                    = &e1000_close;
4447         netdev->hard_start_xmit         = &e1000_xmit_frame;
4448         netdev->get_stats               = &e1000_get_stats;
4449         netdev->set_multicast_list      = &e1000_set_multi;
4450         netdev->set_mac_address         = &e1000_set_mac;
4451         netdev->change_mtu              = &e1000_change_mtu;
4452         netdev->do_ioctl                = &e1000_ioctl;
4453         e1000e_set_ethtool_ops(netdev);
4454         netdev->tx_timeout              = &e1000_tx_timeout;
4455         netdev->watchdog_timeo          = 5 * HZ;
4456         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
4457         netdev->vlan_rx_register        = e1000_vlan_rx_register;
4458         netdev->vlan_rx_add_vid         = e1000_vlan_rx_add_vid;
4459         netdev->vlan_rx_kill_vid        = e1000_vlan_rx_kill_vid;
4460 #ifdef CONFIG_NET_POLL_CONTROLLER
4461         netdev->poll_controller         = e1000_netpoll;
4462 #endif
4463         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
4464
4465         netdev->mem_start = mmio_start;
4466         netdev->mem_end = mmio_start + mmio_len;
4467
4468         adapter->bd_number = cards_found++;
4469
4470         /* setup adapter struct */
4471         err = e1000_sw_init(adapter);
4472         if (err)
4473                 goto err_sw_init;
4474
4475         err = -EIO;
4476
4477         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4478         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
4479         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4480
4481         err = ei->get_variants(adapter);
4482         if (err)
4483                 goto err_hw_init;
4484
4485         hw->mac.ops.get_bus_info(&adapter->hw);
4486
4487         adapter->hw.phy.autoneg_wait_to_complete = 0;
4488
4489         /* Copper options */
4490         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
4491                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
4492                 adapter->hw.phy.disable_polarity_correction = 0;
4493                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
4494         }
4495
4496         if (e1000_check_reset_block(&adapter->hw))
4497                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
4498
4499         netdev->features = NETIF_F_SG |
4500                            NETIF_F_HW_CSUM |
4501                            NETIF_F_HW_VLAN_TX |
4502                            NETIF_F_HW_VLAN_RX;
4503
4504         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
4505                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
4506
4507         netdev->features |= NETIF_F_TSO;
4508         netdev->features |= NETIF_F_TSO6;
4509
4510         netdev->vlan_features |= NETIF_F_TSO;
4511         netdev->vlan_features |= NETIF_F_TSO6;
4512         netdev->vlan_features |= NETIF_F_HW_CSUM;
4513         netdev->vlan_features |= NETIF_F_SG;
4514
4515         if (pci_using_dac)
4516                 netdev->features |= NETIF_F_HIGHDMA;
4517
4518         /*
4519          * We should not be using LLTX anymore, but we are still Tx faster with
4520          * it.
4521          */
4522         netdev->features |= NETIF_F_LLTX;
4523
4524         if (e1000e_enable_mng_pass_thru(&adapter->hw))
4525                 adapter->flags |= FLAG_MNG_PT_ENABLED;
4526
4527         /*
4528          * before reading the NVM, reset the controller to
4529          * put the device in a known good starting state
4530          */
4531         adapter->hw.mac.ops.reset_hw(&adapter->hw);
4532
4533         /*
4534          * systems with ASPM and others may see the checksum fail on the first
4535          * attempt. Let's give it a few tries
4536          */
4537         for (i = 0;; i++) {
4538                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
4539                         break;
4540                 if (i == 2) {
4541                         e_err("The NVM Checksum Is Not Valid\n");
4542                         err = -EIO;
4543                         goto err_eeprom;
4544                 }
4545         }
4546
4547         e1000_eeprom_checks(adapter);
4548
4549         /* copy the MAC address out of the NVM */
4550         if (e1000e_read_mac_addr(&adapter->hw))
4551                 e_err("NVM Read Error while reading MAC address\n");
4552
4553         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
4554         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
4555
4556         if (!is_valid_ether_addr(netdev->perm_addr)) {
4557                 e_err("Invalid MAC Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
4558                       netdev->perm_addr[0], netdev->perm_addr[1],
4559                       netdev->perm_addr[2], netdev->perm_addr[3],
4560                       netdev->perm_addr[4], netdev->perm_addr[5]);
4561                 err = -EIO;
4562                 goto err_eeprom;
4563         }
4564
4565         init_timer(&adapter->watchdog_timer);
4566         adapter->watchdog_timer.function = &e1000_watchdog;
4567         adapter->watchdog_timer.data = (unsigned long) adapter;
4568
4569         init_timer(&adapter->phy_info_timer);
4570         adapter->phy_info_timer.function = &e1000_update_phy_info;
4571         adapter->phy_info_timer.data = (unsigned long) adapter;
4572
4573         INIT_WORK(&adapter->reset_task, e1000_reset_task);
4574         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
4575
4576         e1000e_check_options(adapter);
4577
4578         /* Initialize link parameters. User can change them with ethtool */
4579         adapter->hw.mac.autoneg = 1;
4580         adapter->fc_autoneg = 1;
4581         adapter->hw.fc.original_type = e1000_fc_default;
4582         adapter->hw.fc.type = e1000_fc_default;
4583         adapter->hw.phy.autoneg_advertised = 0x2f;
4584
4585         /* ring size defaults */
4586         adapter->rx_ring->count = 256;
4587         adapter->tx_ring->count = 256;
4588
4589         /*
4590          * Initial Wake on LAN setting - If APM wake is enabled in
4591          * the EEPROM, enable the ACPI Magic Packet filter
4592          */
4593         if (adapter->flags & FLAG_APME_IN_WUC) {
4594                 /* APME bit in EEPROM is mapped to WUC.APME */
4595                 eeprom_data = er32(WUC);
4596                 eeprom_apme_mask = E1000_WUC_APME;
4597         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
4598                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
4599                     (adapter->hw.bus.func == 1))
4600                         e1000_read_nvm(&adapter->hw,
4601                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
4602                 else
4603                         e1000_read_nvm(&adapter->hw,
4604                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
4605         }
4606
4607         /* fetch WoL from EEPROM */
4608         if (eeprom_data & eeprom_apme_mask)
4609                 adapter->eeprom_wol |= E1000_WUFC_MAG;
4610
4611         /*
4612          * now that we have the eeprom settings, apply the special cases
4613          * where the eeprom may be wrong or the board simply won't support
4614          * wake on lan on a particular port
4615          */
4616         if (!(adapter->flags & FLAG_HAS_WOL))
4617                 adapter->eeprom_wol = 0;
4618
4619         /* initialize the wol settings based on the eeprom settings */
4620         adapter->wol = adapter->eeprom_wol;
4621
4622         /* reset the hardware with the new settings */
4623         e1000e_reset(adapter);
4624
4625         /*
4626          * If the controller has AMT, do not set DRV_LOAD until the interface
4627          * is up.  For all other cases, let the f/w know that the h/w is now
4628          * under the control of the driver.
4629          */
4630         if (!(adapter->flags & FLAG_HAS_AMT))
4631                 e1000_get_hw_control(adapter);
4632
4633         /* tell the stack to leave us alone until e1000_open() is called */
4634         netif_carrier_off(netdev);
4635         netif_tx_stop_all_queues(netdev);
4636
4637         strcpy(netdev->name, "eth%d");
4638         err = register_netdev(netdev);
4639         if (err)
4640                 goto err_register;
4641
4642         e1000_print_device_info(adapter);
4643
4644         return 0;
4645
4646 err_register:
4647         if (!(adapter->flags & FLAG_HAS_AMT))
4648                 e1000_release_hw_control(adapter);
4649 err_eeprom:
4650         if (!e1000_check_reset_block(&adapter->hw))
4651                 e1000_phy_hw_reset(&adapter->hw);
4652 err_hw_init:
4653
4654         kfree(adapter->tx_ring);
4655         kfree(adapter->rx_ring);
4656 err_sw_init:
4657         if (adapter->hw.flash_address)
4658                 iounmap(adapter->hw.flash_address);
4659 err_flashmap:
4660         iounmap(adapter->hw.hw_addr);
4661 err_ioremap:
4662         free_netdev(netdev);
4663 err_alloc_etherdev:
4664         pci_release_selected_regions(pdev,
4665                                      pci_select_bars(pdev, IORESOURCE_MEM));
4666 err_pci_reg:
4667 err_dma:
4668         pci_disable_device(pdev);
4669         return err;
4670 }
4671
4672 /**
4673  * e1000_remove - Device Removal Routine
4674  * @pdev: PCI device information struct
4675  *
4676  * e1000_remove is called by the PCI subsystem to alert the driver
4677  * that it should release a PCI device.  The could be caused by a
4678  * Hot-Plug event, or because the driver is going to be removed from
4679  * memory.
4680  **/
4681 static void __devexit e1000_remove(struct pci_dev *pdev)
4682 {
4683         struct net_device *netdev = pci_get_drvdata(pdev);
4684         struct e1000_adapter *adapter = netdev_priv(netdev);
4685
4686         /*
4687          * flush_scheduled work may reschedule our watchdog task, so
4688          * explicitly disable watchdog tasks from being rescheduled
4689          */
4690         set_bit(__E1000_DOWN, &adapter->state);
4691         del_timer_sync(&adapter->watchdog_timer);
4692         del_timer_sync(&adapter->phy_info_timer);
4693
4694         flush_scheduled_work();
4695
4696         /*
4697          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4698          * would have already happened in close and is redundant.
4699          */
4700         e1000_release_hw_control(adapter);
4701
4702         unregister_netdev(netdev);
4703
4704         if (!e1000_check_reset_block(&adapter->hw))
4705                 e1000_phy_hw_reset(&adapter->hw);
4706
4707         kfree(adapter->tx_ring);
4708         kfree(adapter->rx_ring);
4709
4710         iounmap(adapter->hw.hw_addr);
4711         if (adapter->hw.flash_address)
4712                 iounmap(adapter->hw.flash_address);
4713         pci_release_selected_regions(pdev,
4714                                      pci_select_bars(pdev, IORESOURCE_MEM));
4715
4716         free_netdev(netdev);
4717
4718         pci_disable_device(pdev);
4719 }
4720
4721 /* PCI Error Recovery (ERS) */
4722 static struct pci_error_handlers e1000_err_handler = {
4723         .error_detected = e1000_io_error_detected,
4724         .slot_reset = e1000_io_slot_reset,
4725         .resume = e1000_io_resume,
4726 };
4727
4728 static struct pci_device_id e1000_pci_tbl[] = {
4729         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
4730         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
4731         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
4732         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
4733         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
4734         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
4735         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
4736         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
4737         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
4738
4739         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
4740         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
4741         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
4742         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
4743
4744         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
4745         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
4746         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
4747
4748         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
4749           board_80003es2lan },
4750         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
4751           board_80003es2lan },
4752         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
4753           board_80003es2lan },
4754         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
4755           board_80003es2lan },
4756
4757         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
4758         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
4759         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
4760         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
4761         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
4762         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
4763         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
4764
4765         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
4766         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
4767         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
4768         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
4769         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
4770         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
4771         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
4772         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
4773
4774         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
4775         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
4776         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
4777
4778         { }     /* terminate list */
4779 };
4780 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
4781
4782 /* PCI Device API Driver */
4783 static struct pci_driver e1000_driver = {
4784         .name     = e1000e_driver_name,
4785         .id_table = e1000_pci_tbl,
4786         .probe    = e1000_probe,
4787         .remove   = __devexit_p(e1000_remove),
4788 #ifdef CONFIG_PM
4789         /* Power Management Hooks */
4790         .suspend  = e1000_suspend,
4791         .resume   = e1000_resume,
4792 #endif
4793         .shutdown = e1000_shutdown,
4794         .err_handler = &e1000_err_handler
4795 };
4796
4797 /**
4798  * e1000_init_module - Driver Registration Routine
4799  *
4800  * e1000_init_module is the first routine called when the driver is
4801  * loaded. All it does is register with the PCI subsystem.
4802  **/
4803 static int __init e1000_init_module(void)
4804 {
4805         int ret;
4806         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
4807                e1000e_driver_name, e1000e_driver_version);
4808         printk(KERN_INFO "%s: Copyright (c) 1999-2008 Intel Corporation.\n",
4809                e1000e_driver_name);
4810         ret = pci_register_driver(&e1000_driver);
4811         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name,
4812                                PM_QOS_DEFAULT_VALUE);
4813                                 
4814         return ret;
4815 }
4816 module_init(e1000_init_module);
4817
4818 /**
4819  * e1000_exit_module - Driver Exit Cleanup Routine
4820  *
4821  * e1000_exit_module is called just before the driver is removed
4822  * from memory.
4823  **/
4824 static void __exit e1000_exit_module(void)
4825 {
4826         pci_unregister_driver(&e1000_driver);
4827         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name);
4828 }
4829 module_exit(e1000_exit_module);
4830
4831
4832 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
4833 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
4834 MODULE_LICENSE("GPL");
4835 MODULE_VERSION(DRV_VERSION);
4836
4837 /* e1000_main.c */