Merge branch 'i2c-next' of git://aeryn.fluff.org.uk/bjdooks/linux
[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-k6"
51 char e1000e_driver_name[] = "e1000e";
52 const char e1000e_driver_version[] = DRV_VERSION;
53
54 static const struct e1000_info *e1000_info_tbl[] = {
55         [board_82571]           = &e1000_82571_info,
56         [board_82572]           = &e1000_82572_info,
57         [board_82573]           = &e1000_82573_info,
58         [board_82574]           = &e1000_82574_info,
59         [board_80003es2lan]     = &e1000_es2_info,
60         [board_ich8lan]         = &e1000_ich8_info,
61         [board_ich9lan]         = &e1000_ich9_info,
62         [board_ich10lan]        = &e1000_ich10_info,
63 };
64
65 #ifdef DEBUG
66 /**
67  * e1000_get_hw_dev_name - return device name string
68  * used by hardware layer to print debugging information
69  **/
70 char *e1000e_get_hw_dev_name(struct e1000_hw *hw)
71 {
72         return hw->adapter->netdev->name;
73 }
74 #endif
75
76 /**
77  * e1000_desc_unused - calculate if we have unused descriptors
78  **/
79 static int e1000_desc_unused(struct e1000_ring *ring)
80 {
81         if (ring->next_to_clean > ring->next_to_use)
82                 return ring->next_to_clean - ring->next_to_use - 1;
83
84         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
85 }
86
87 /**
88  * e1000_receive_skb - helper function to handle Rx indications
89  * @adapter: board private structure
90  * @status: descriptor status field as written by hardware
91  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
92  * @skb: pointer to sk_buff to be indicated to stack
93  **/
94 static void e1000_receive_skb(struct e1000_adapter *adapter,
95                               struct net_device *netdev,
96                               struct sk_buff *skb,
97                               u8 status, __le16 vlan)
98 {
99         skb->protocol = eth_type_trans(skb, netdev);
100
101         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
102                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
103                                          le16_to_cpu(vlan));
104         else
105                 napi_gro_receive(&adapter->napi, skb);
106 }
107
108 /**
109  * e1000_rx_checksum - Receive Checksum Offload for 82543
110  * @adapter:     board private structure
111  * @status_err:  receive descriptor status and error fields
112  * @csum:       receive descriptor csum field
113  * @sk_buff:     socket buffer with received data
114  **/
115 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
116                               u32 csum, struct sk_buff *skb)
117 {
118         u16 status = (u16)status_err;
119         u8 errors = (u8)(status_err >> 24);
120         skb->ip_summed = CHECKSUM_NONE;
121
122         /* Ignore Checksum bit is set */
123         if (status & E1000_RXD_STAT_IXSM)
124                 return;
125         /* TCP/UDP checksum error bit is set */
126         if (errors & E1000_RXD_ERR_TCPE) {
127                 /* let the stack verify checksum errors */
128                 adapter->hw_csum_err++;
129                 return;
130         }
131
132         /* TCP/UDP Checksum has not been calculated */
133         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
134                 return;
135
136         /* It must be a TCP or UDP packet with a valid checksum */
137         if (status & E1000_RXD_STAT_TCPCS) {
138                 /* TCP checksum is good */
139                 skb->ip_summed = CHECKSUM_UNNECESSARY;
140         } else {
141                 /*
142                  * IP fragment with UDP payload
143                  * Hardware complements the payload checksum, so we undo it
144                  * and then put the value in host order for further stack use.
145                  */
146                 __sum16 sum = (__force __sum16)htons(csum);
147                 skb->csum = csum_unfold(~sum);
148                 skb->ip_summed = CHECKSUM_COMPLETE;
149         }
150         adapter->hw_csum_good++;
151 }
152
153 /**
154  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
155  * @adapter: address of board private structure
156  **/
157 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
158                                    int cleaned_count)
159 {
160         struct net_device *netdev = adapter->netdev;
161         struct pci_dev *pdev = adapter->pdev;
162         struct e1000_ring *rx_ring = adapter->rx_ring;
163         struct e1000_rx_desc *rx_desc;
164         struct e1000_buffer *buffer_info;
165         struct sk_buff *skb;
166         unsigned int i;
167         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
168
169         i = rx_ring->next_to_use;
170         buffer_info = &rx_ring->buffer_info[i];
171
172         while (cleaned_count--) {
173                 skb = buffer_info->skb;
174                 if (skb) {
175                         skb_trim(skb, 0);
176                         goto map_skb;
177                 }
178
179                 skb = netdev_alloc_skb(netdev, bufsz);
180                 if (!skb) {
181                         /* Better luck next round */
182                         adapter->alloc_rx_buff_failed++;
183                         break;
184                 }
185
186                 /*
187                  * Make buffer alignment 2 beyond a 16 byte boundary
188                  * this will result in a 16 byte aligned IP header after
189                  * the 14 byte MAC header is removed
190                  */
191                 skb_reserve(skb, NET_IP_ALIGN);
192
193                 buffer_info->skb = skb;
194 map_skb:
195                 buffer_info->dma = pci_map_single(pdev, skb->data,
196                                                   adapter->rx_buffer_len,
197                                                   PCI_DMA_FROMDEVICE);
198                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
199                         dev_err(&pdev->dev, "RX DMA map failed\n");
200                         adapter->rx_dma_failed++;
201                         break;
202                 }
203
204                 rx_desc = E1000_RX_DESC(*rx_ring, i);
205                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
206
207                 i++;
208                 if (i == rx_ring->count)
209                         i = 0;
210                 buffer_info = &rx_ring->buffer_info[i];
211         }
212
213         if (rx_ring->next_to_use != i) {
214                 rx_ring->next_to_use = i;
215                 if (i-- == 0)
216                         i = (rx_ring->count - 1);
217
218                 /*
219                  * Force memory writes to complete before letting h/w
220                  * know there are new descriptors to fetch.  (Only
221                  * applicable for weak-ordered memory model archs,
222                  * such as IA-64).
223                  */
224                 wmb();
225                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
226         }
227 }
228
229 /**
230  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
231  * @adapter: address of board private structure
232  **/
233 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
234                                       int cleaned_count)
235 {
236         struct net_device *netdev = adapter->netdev;
237         struct pci_dev *pdev = adapter->pdev;
238         union e1000_rx_desc_packet_split *rx_desc;
239         struct e1000_ring *rx_ring = adapter->rx_ring;
240         struct e1000_buffer *buffer_info;
241         struct e1000_ps_page *ps_page;
242         struct sk_buff *skb;
243         unsigned int i, j;
244
245         i = rx_ring->next_to_use;
246         buffer_info = &rx_ring->buffer_info[i];
247
248         while (cleaned_count--) {
249                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
250
251                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
252                         ps_page = &buffer_info->ps_pages[j];
253                         if (j >= adapter->rx_ps_pages) {
254                                 /* all unused desc entries get hw null ptr */
255                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
256                                 continue;
257                         }
258                         if (!ps_page->page) {
259                                 ps_page->page = alloc_page(GFP_ATOMIC);
260                                 if (!ps_page->page) {
261                                         adapter->alloc_rx_buff_failed++;
262                                         goto no_buffers;
263                                 }
264                                 ps_page->dma = pci_map_page(pdev,
265                                                    ps_page->page,
266                                                    0, PAGE_SIZE,
267                                                    PCI_DMA_FROMDEVICE);
268                                 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
269                                         dev_err(&adapter->pdev->dev,
270                                           "RX DMA page map failed\n");
271                                         adapter->rx_dma_failed++;
272                                         goto no_buffers;
273                                 }
274                         }
275                         /*
276                          * Refresh the desc even if buffer_addrs
277                          * didn't change because each write-back
278                          * erases this info.
279                          */
280                         rx_desc->read.buffer_addr[j+1] =
281                              cpu_to_le64(ps_page->dma);
282                 }
283
284                 skb = netdev_alloc_skb(netdev,
285                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
286
287                 if (!skb) {
288                         adapter->alloc_rx_buff_failed++;
289                         break;
290                 }
291
292                 /*
293                  * Make buffer alignment 2 beyond a 16 byte boundary
294                  * this will result in a 16 byte aligned IP header after
295                  * the 14 byte MAC header is removed
296                  */
297                 skb_reserve(skb, NET_IP_ALIGN);
298
299                 buffer_info->skb = skb;
300                 buffer_info->dma = pci_map_single(pdev, skb->data,
301                                                   adapter->rx_ps_bsize0,
302                                                   PCI_DMA_FROMDEVICE);
303                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
304                         dev_err(&pdev->dev, "RX DMA map failed\n");
305                         adapter->rx_dma_failed++;
306                         /* cleanup skb */
307                         dev_kfree_skb_any(skb);
308                         buffer_info->skb = NULL;
309                         break;
310                 }
311
312                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
313
314                 i++;
315                 if (i == rx_ring->count)
316                         i = 0;
317                 buffer_info = &rx_ring->buffer_info[i];
318         }
319
320 no_buffers:
321         if (rx_ring->next_to_use != i) {
322                 rx_ring->next_to_use = i;
323
324                 if (!(i--))
325                         i = (rx_ring->count - 1);
326
327                 /*
328                  * Force memory writes to complete before letting h/w
329                  * know there are new descriptors to fetch.  (Only
330                  * applicable for weak-ordered memory model archs,
331                  * such as IA-64).
332                  */
333                 wmb();
334                 /*
335                  * Hardware increments by 16 bytes, but packet split
336                  * descriptors are 32 bytes...so we increment tail
337                  * twice as much.
338                  */
339                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
340         }
341 }
342
343 /**
344  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
345  * @adapter: address of board private structure
346  * @cleaned_count: number of buffers to allocate this pass
347  **/
348
349 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
350                                          int cleaned_count)
351 {
352         struct net_device *netdev = adapter->netdev;
353         struct pci_dev *pdev = adapter->pdev;
354         struct e1000_rx_desc *rx_desc;
355         struct e1000_ring *rx_ring = adapter->rx_ring;
356         struct e1000_buffer *buffer_info;
357         struct sk_buff *skb;
358         unsigned int i;
359         unsigned int bufsz = 256 -
360                              16 /* for skb_reserve */ -
361                              NET_IP_ALIGN;
362
363         i = rx_ring->next_to_use;
364         buffer_info = &rx_ring->buffer_info[i];
365
366         while (cleaned_count--) {
367                 skb = buffer_info->skb;
368                 if (skb) {
369                         skb_trim(skb, 0);
370                         goto check_page;
371                 }
372
373                 skb = netdev_alloc_skb(netdev, bufsz);
374                 if (unlikely(!skb)) {
375                         /* Better luck next round */
376                         adapter->alloc_rx_buff_failed++;
377                         break;
378                 }
379
380                 /* Make buffer alignment 2 beyond a 16 byte boundary
381                  * this will result in a 16 byte aligned IP header after
382                  * the 14 byte MAC header is removed
383                  */
384                 skb_reserve(skb, NET_IP_ALIGN);
385
386                 buffer_info->skb = skb;
387 check_page:
388                 /* allocate a new page if necessary */
389                 if (!buffer_info->page) {
390                         buffer_info->page = alloc_page(GFP_ATOMIC);
391                         if (unlikely(!buffer_info->page)) {
392                                 adapter->alloc_rx_buff_failed++;
393                                 break;
394                         }
395                 }
396
397                 if (!buffer_info->dma)
398                         buffer_info->dma = pci_map_page(pdev,
399                                                         buffer_info->page, 0,
400                                                         PAGE_SIZE,
401                                                         PCI_DMA_FROMDEVICE);
402
403                 rx_desc = E1000_RX_DESC(*rx_ring, i);
404                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
405
406                 if (unlikely(++i == rx_ring->count))
407                         i = 0;
408                 buffer_info = &rx_ring->buffer_info[i];
409         }
410
411         if (likely(rx_ring->next_to_use != i)) {
412                 rx_ring->next_to_use = i;
413                 if (unlikely(i-- == 0))
414                         i = (rx_ring->count - 1);
415
416                 /* Force memory writes to complete before letting h/w
417                  * know there are new descriptors to fetch.  (Only
418                  * applicable for weak-ordered memory model archs,
419                  * such as IA-64). */
420                 wmb();
421                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
422         }
423 }
424
425 /**
426  * e1000_clean_rx_irq - Send received data up the network stack; legacy
427  * @adapter: board private structure
428  *
429  * the return value indicates whether actual cleaning was done, there
430  * is no guarantee that everything was cleaned
431  **/
432 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
433                                int *work_done, int work_to_do)
434 {
435         struct net_device *netdev = adapter->netdev;
436         struct pci_dev *pdev = adapter->pdev;
437         struct e1000_ring *rx_ring = adapter->rx_ring;
438         struct e1000_rx_desc *rx_desc, *next_rxd;
439         struct e1000_buffer *buffer_info, *next_buffer;
440         u32 length;
441         unsigned int i;
442         int cleaned_count = 0;
443         bool cleaned = 0;
444         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
445
446         i = rx_ring->next_to_clean;
447         rx_desc = E1000_RX_DESC(*rx_ring, i);
448         buffer_info = &rx_ring->buffer_info[i];
449
450         while (rx_desc->status & E1000_RXD_STAT_DD) {
451                 struct sk_buff *skb;
452                 u8 status;
453
454                 if (*work_done >= work_to_do)
455                         break;
456                 (*work_done)++;
457
458                 status = rx_desc->status;
459                 skb = buffer_info->skb;
460                 buffer_info->skb = NULL;
461
462                 prefetch(skb->data - NET_IP_ALIGN);
463
464                 i++;
465                 if (i == rx_ring->count)
466                         i = 0;
467                 next_rxd = E1000_RX_DESC(*rx_ring, i);
468                 prefetch(next_rxd);
469
470                 next_buffer = &rx_ring->buffer_info[i];
471
472                 cleaned = 1;
473                 cleaned_count++;
474                 pci_unmap_single(pdev,
475                                  buffer_info->dma,
476                                  adapter->rx_buffer_len,
477                                  PCI_DMA_FROMDEVICE);
478                 buffer_info->dma = 0;
479
480                 length = le16_to_cpu(rx_desc->length);
481
482                 /* !EOP means multiple descriptors were used to store a single
483                  * packet, also make sure the frame isn't just CRC only */
484                 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
485                         /* All receives must fit into a single buffer */
486                         e_dbg("%s: Receive packet consumed multiple buffers\n",
487                               netdev->name);
488                         /* recycle */
489                         buffer_info->skb = skb;
490                         goto next_desc;
491                 }
492
493                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
494                         /* recycle */
495                         buffer_info->skb = skb;
496                         goto next_desc;
497                 }
498
499                 /* adjust length to remove Ethernet CRC */
500                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
501                         length -= 4;
502
503                 total_rx_bytes += length;
504                 total_rx_packets++;
505
506                 /*
507                  * code added for copybreak, this should improve
508                  * performance for small packets with large amounts
509                  * of reassembly being done in the stack
510                  */
511                 if (length < copybreak) {
512                         struct sk_buff *new_skb =
513                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
514                         if (new_skb) {
515                                 skb_reserve(new_skb, NET_IP_ALIGN);
516                                 skb_copy_to_linear_data_offset(new_skb,
517                                                                -NET_IP_ALIGN,
518                                                                (skb->data -
519                                                                 NET_IP_ALIGN),
520                                                                (length +
521                                                                 NET_IP_ALIGN));
522                                 /* save the skb in buffer_info as good */
523                                 buffer_info->skb = skb;
524                                 skb = new_skb;
525                         }
526                         /* else just continue with the old one */
527                 }
528                 /* end copybreak code */
529                 skb_put(skb, length);
530
531                 /* Receive Checksum Offload */
532                 e1000_rx_checksum(adapter,
533                                   (u32)(status) |
534                                   ((u32)(rx_desc->errors) << 24),
535                                   le16_to_cpu(rx_desc->csum), skb);
536
537                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
538
539 next_desc:
540                 rx_desc->status = 0;
541
542                 /* return some buffers to hardware, one at a time is too slow */
543                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
544                         adapter->alloc_rx_buf(adapter, cleaned_count);
545                         cleaned_count = 0;
546                 }
547
548                 /* use prefetched values */
549                 rx_desc = next_rxd;
550                 buffer_info = next_buffer;
551         }
552         rx_ring->next_to_clean = i;
553
554         cleaned_count = e1000_desc_unused(rx_ring);
555         if (cleaned_count)
556                 adapter->alloc_rx_buf(adapter, cleaned_count);
557
558         adapter->total_rx_bytes += total_rx_bytes;
559         adapter->total_rx_packets += total_rx_packets;
560         adapter->net_stats.rx_bytes += total_rx_bytes;
561         adapter->net_stats.rx_packets += total_rx_packets;
562         return cleaned;
563 }
564
565 static void e1000_put_txbuf(struct e1000_adapter *adapter,
566                              struct e1000_buffer *buffer_info)
567 {
568         if (buffer_info->dma) {
569                 pci_unmap_page(adapter->pdev, buffer_info->dma,
570                                buffer_info->length, PCI_DMA_TODEVICE);
571                 buffer_info->dma = 0;
572         }
573         if (buffer_info->skb) {
574                 dev_kfree_skb_any(buffer_info->skb);
575                 buffer_info->skb = NULL;
576         }
577 }
578
579 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
580 {
581         struct e1000_ring *tx_ring = adapter->tx_ring;
582         unsigned int i = tx_ring->next_to_clean;
583         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
584         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
585
586         /* detected Tx unit hang */
587         e_err("Detected Tx Unit Hang:\n"
588               "  TDH                  <%x>\n"
589               "  TDT                  <%x>\n"
590               "  next_to_use          <%x>\n"
591               "  next_to_clean        <%x>\n"
592               "buffer_info[next_to_clean]:\n"
593               "  time_stamp           <%lx>\n"
594               "  next_to_watch        <%x>\n"
595               "  jiffies              <%lx>\n"
596               "  next_to_watch.status <%x>\n",
597               readl(adapter->hw.hw_addr + tx_ring->head),
598               readl(adapter->hw.hw_addr + tx_ring->tail),
599               tx_ring->next_to_use,
600               tx_ring->next_to_clean,
601               tx_ring->buffer_info[eop].time_stamp,
602               eop,
603               jiffies,
604               eop_desc->upper.fields.status);
605 }
606
607 /**
608  * e1000_clean_tx_irq - Reclaim resources after transmit completes
609  * @adapter: board private structure
610  *
611  * the return value indicates whether actual cleaning was done, there
612  * is no guarantee that everything was cleaned
613  **/
614 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
615 {
616         struct net_device *netdev = adapter->netdev;
617         struct e1000_hw *hw = &adapter->hw;
618         struct e1000_ring *tx_ring = adapter->tx_ring;
619         struct e1000_tx_desc *tx_desc, *eop_desc;
620         struct e1000_buffer *buffer_info;
621         unsigned int i, eop;
622         unsigned int count = 0;
623         bool cleaned = 0;
624         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
625
626         i = tx_ring->next_to_clean;
627         eop = tx_ring->buffer_info[i].next_to_watch;
628         eop_desc = E1000_TX_DESC(*tx_ring, eop);
629
630         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
631                 for (cleaned = 0; !cleaned; ) {
632                         tx_desc = E1000_TX_DESC(*tx_ring, i);
633                         buffer_info = &tx_ring->buffer_info[i];
634                         cleaned = (i == eop);
635
636                         if (cleaned) {
637                                 struct sk_buff *skb = buffer_info->skb;
638                                 unsigned int segs, bytecount;
639                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
640                                 /* multiply data chunks by size of headers */
641                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
642                                             skb->len;
643                                 total_tx_packets += segs;
644                                 total_tx_bytes += bytecount;
645                         }
646
647                         e1000_put_txbuf(adapter, buffer_info);
648                         tx_desc->upper.data = 0;
649
650                         i++;
651                         if (i == tx_ring->count)
652                                 i = 0;
653                 }
654
655                 eop = tx_ring->buffer_info[i].next_to_watch;
656                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
657 #define E1000_TX_WEIGHT 64
658                 /* weight of a sort for tx, to avoid endless transmit cleanup */
659                 if (count++ == E1000_TX_WEIGHT)
660                         break;
661         }
662
663         tx_ring->next_to_clean = i;
664
665 #define TX_WAKE_THRESHOLD 32
666         if (cleaned && netif_carrier_ok(netdev) &&
667                      e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
668                 /* Make sure that anybody stopping the queue after this
669                  * sees the new next_to_clean.
670                  */
671                 smp_mb();
672
673                 if (netif_queue_stopped(netdev) &&
674                     !(test_bit(__E1000_DOWN, &adapter->state))) {
675                         netif_wake_queue(netdev);
676                         ++adapter->restart_queue;
677                 }
678         }
679
680         if (adapter->detect_tx_hung) {
681                 /*
682                  * Detect a transmit hang in hardware, this serializes the
683                  * check with the clearing of time_stamp and movement of i
684                  */
685                 adapter->detect_tx_hung = 0;
686                 if (tx_ring->buffer_info[eop].dma &&
687                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp
688                                + (adapter->tx_timeout_factor * HZ))
689                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
690                         e1000_print_tx_hang(adapter);
691                         netif_stop_queue(netdev);
692                 }
693         }
694         adapter->total_tx_bytes += total_tx_bytes;
695         adapter->total_tx_packets += total_tx_packets;
696         adapter->net_stats.tx_bytes += total_tx_bytes;
697         adapter->net_stats.tx_packets += total_tx_packets;
698         return cleaned;
699 }
700
701 /**
702  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
703  * @adapter: board private structure
704  *
705  * the return value indicates whether actual cleaning was done, there
706  * is no guarantee that everything was cleaned
707  **/
708 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
709                                   int *work_done, int work_to_do)
710 {
711         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
712         struct net_device *netdev = adapter->netdev;
713         struct pci_dev *pdev = adapter->pdev;
714         struct e1000_ring *rx_ring = adapter->rx_ring;
715         struct e1000_buffer *buffer_info, *next_buffer;
716         struct e1000_ps_page *ps_page;
717         struct sk_buff *skb;
718         unsigned int i, j;
719         u32 length, staterr;
720         int cleaned_count = 0;
721         bool cleaned = 0;
722         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
723
724         i = rx_ring->next_to_clean;
725         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
726         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
727         buffer_info = &rx_ring->buffer_info[i];
728
729         while (staterr & E1000_RXD_STAT_DD) {
730                 if (*work_done >= work_to_do)
731                         break;
732                 (*work_done)++;
733                 skb = buffer_info->skb;
734
735                 /* in the packet split case this is header only */
736                 prefetch(skb->data - NET_IP_ALIGN);
737
738                 i++;
739                 if (i == rx_ring->count)
740                         i = 0;
741                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
742                 prefetch(next_rxd);
743
744                 next_buffer = &rx_ring->buffer_info[i];
745
746                 cleaned = 1;
747                 cleaned_count++;
748                 pci_unmap_single(pdev, buffer_info->dma,
749                                  adapter->rx_ps_bsize0,
750                                  PCI_DMA_FROMDEVICE);
751                 buffer_info->dma = 0;
752
753                 if (!(staterr & E1000_RXD_STAT_EOP)) {
754                         e_dbg("%s: Packet Split buffers didn't pick up the "
755                               "full packet\n", netdev->name);
756                         dev_kfree_skb_irq(skb);
757                         goto next_desc;
758                 }
759
760                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
761                         dev_kfree_skb_irq(skb);
762                         goto next_desc;
763                 }
764
765                 length = le16_to_cpu(rx_desc->wb.middle.length0);
766
767                 if (!length) {
768                         e_dbg("%s: Last part of the packet spanning multiple "
769                               "descriptors\n", netdev->name);
770                         dev_kfree_skb_irq(skb);
771                         goto next_desc;
772                 }
773
774                 /* Good Receive */
775                 skb_put(skb, length);
776
777                 {
778                 /*
779                  * this looks ugly, but it seems compiler issues make it
780                  * more efficient than reusing j
781                  */
782                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
783
784                 /*
785                  * page alloc/put takes too long and effects small packet
786                  * throughput, so unsplit small packets and save the alloc/put
787                  * only valid in softirq (napi) context to call kmap_*
788                  */
789                 if (l1 && (l1 <= copybreak) &&
790                     ((length + l1) <= adapter->rx_ps_bsize0)) {
791                         u8 *vaddr;
792
793                         ps_page = &buffer_info->ps_pages[0];
794
795                         /*
796                          * there is no documentation about how to call
797                          * kmap_atomic, so we can't hold the mapping
798                          * very long
799                          */
800                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
801                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
802                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
803                         memcpy(skb_tail_pointer(skb), vaddr, l1);
804                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
805                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
806                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
807
808                         /* remove the CRC */
809                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
810                                 l1 -= 4;
811
812                         skb_put(skb, l1);
813                         goto copydone;
814                 } /* if */
815                 }
816
817                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
818                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
819                         if (!length)
820                                 break;
821
822                         ps_page = &buffer_info->ps_pages[j];
823                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
824                                        PCI_DMA_FROMDEVICE);
825                         ps_page->dma = 0;
826                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
827                         ps_page->page = NULL;
828                         skb->len += length;
829                         skb->data_len += length;
830                         skb->truesize += length;
831                 }
832
833                 /* strip the ethernet crc, problem is we're using pages now so
834                  * this whole operation can get a little cpu intensive
835                  */
836                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
837                         pskb_trim(skb, skb->len - 4);
838
839 copydone:
840                 total_rx_bytes += skb->len;
841                 total_rx_packets++;
842
843                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
844                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
845
846                 if (rx_desc->wb.upper.header_status &
847                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
848                         adapter->rx_hdr_split++;
849
850                 e1000_receive_skb(adapter, netdev, skb,
851                                   staterr, rx_desc->wb.middle.vlan);
852
853 next_desc:
854                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
855                 buffer_info->skb = NULL;
856
857                 /* return some buffers to hardware, one at a time is too slow */
858                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
859                         adapter->alloc_rx_buf(adapter, cleaned_count);
860                         cleaned_count = 0;
861                 }
862
863                 /* use prefetched values */
864                 rx_desc = next_rxd;
865                 buffer_info = next_buffer;
866
867                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
868         }
869         rx_ring->next_to_clean = i;
870
871         cleaned_count = e1000_desc_unused(rx_ring);
872         if (cleaned_count)
873                 adapter->alloc_rx_buf(adapter, cleaned_count);
874
875         adapter->total_rx_bytes += total_rx_bytes;
876         adapter->total_rx_packets += total_rx_packets;
877         adapter->net_stats.rx_bytes += total_rx_bytes;
878         adapter->net_stats.rx_packets += total_rx_packets;
879         return cleaned;
880 }
881
882 /**
883  * e1000_consume_page - helper function
884  **/
885 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
886                                u16 length)
887 {
888         bi->page = NULL;
889         skb->len += length;
890         skb->data_len += length;
891         skb->truesize += length;
892 }
893
894 /**
895  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
896  * @adapter: board private structure
897  *
898  * the return value indicates whether actual cleaning was done, there
899  * is no guarantee that everything was cleaned
900  **/
901
902 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
903                                      int *work_done, int work_to_do)
904 {
905         struct net_device *netdev = adapter->netdev;
906         struct pci_dev *pdev = adapter->pdev;
907         struct e1000_ring *rx_ring = adapter->rx_ring;
908         struct e1000_rx_desc *rx_desc, *next_rxd;
909         struct e1000_buffer *buffer_info, *next_buffer;
910         u32 length;
911         unsigned int i;
912         int cleaned_count = 0;
913         bool cleaned = false;
914         unsigned int total_rx_bytes=0, total_rx_packets=0;
915
916         i = rx_ring->next_to_clean;
917         rx_desc = E1000_RX_DESC(*rx_ring, i);
918         buffer_info = &rx_ring->buffer_info[i];
919
920         while (rx_desc->status & E1000_RXD_STAT_DD) {
921                 struct sk_buff *skb;
922                 u8 status;
923
924                 if (*work_done >= work_to_do)
925                         break;
926                 (*work_done)++;
927
928                 status = rx_desc->status;
929                 skb = buffer_info->skb;
930                 buffer_info->skb = NULL;
931
932                 ++i;
933                 if (i == rx_ring->count)
934                         i = 0;
935                 next_rxd = E1000_RX_DESC(*rx_ring, i);
936                 prefetch(next_rxd);
937
938                 next_buffer = &rx_ring->buffer_info[i];
939
940                 cleaned = true;
941                 cleaned_count++;
942                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
943                                PCI_DMA_FROMDEVICE);
944                 buffer_info->dma = 0;
945
946                 length = le16_to_cpu(rx_desc->length);
947
948                 /* errors is only valid for DD + EOP descriptors */
949                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
950                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
951                                 /* recycle both page and skb */
952                                 buffer_info->skb = skb;
953                                 /* an error means any chain goes out the window
954                                  * too */
955                                 if (rx_ring->rx_skb_top)
956                                         dev_kfree_skb(rx_ring->rx_skb_top);
957                                 rx_ring->rx_skb_top = NULL;
958                                 goto next_desc;
959                 }
960
961 #define rxtop rx_ring->rx_skb_top
962                 if (!(status & E1000_RXD_STAT_EOP)) {
963                         /* this descriptor is only the beginning (or middle) */
964                         if (!rxtop) {
965                                 /* this is the beginning of a chain */
966                                 rxtop = skb;
967                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
968                                                    0, length);
969                         } else {
970                                 /* this is the middle of a chain */
971                                 skb_fill_page_desc(rxtop,
972                                     skb_shinfo(rxtop)->nr_frags,
973                                     buffer_info->page, 0, length);
974                                 /* re-use the skb, only consumed the page */
975                                 buffer_info->skb = skb;
976                         }
977                         e1000_consume_page(buffer_info, rxtop, length);
978                         goto next_desc;
979                 } else {
980                         if (rxtop) {
981                                 /* end of the chain */
982                                 skb_fill_page_desc(rxtop,
983                                     skb_shinfo(rxtop)->nr_frags,
984                                     buffer_info->page, 0, length);
985                                 /* re-use the current skb, we only consumed the
986                                  * page */
987                                 buffer_info->skb = skb;
988                                 skb = rxtop;
989                                 rxtop = NULL;
990                                 e1000_consume_page(buffer_info, skb, length);
991                         } else {
992                                 /* no chain, got EOP, this buf is the packet
993                                  * copybreak to save the put_page/alloc_page */
994                                 if (length <= copybreak &&
995                                     skb_tailroom(skb) >= length) {
996                                         u8 *vaddr;
997                                         vaddr = kmap_atomic(buffer_info->page,
998                                                            KM_SKB_DATA_SOFTIRQ);
999                                         memcpy(skb_tail_pointer(skb), vaddr,
1000                                                length);
1001                                         kunmap_atomic(vaddr,
1002                                                       KM_SKB_DATA_SOFTIRQ);
1003                                         /* re-use the page, so don't erase
1004                                          * buffer_info->page */
1005                                         skb_put(skb, length);
1006                                 } else {
1007                                         skb_fill_page_desc(skb, 0,
1008                                                            buffer_info->page, 0,
1009                                                            length);
1010                                         e1000_consume_page(buffer_info, skb,
1011                                                            length);
1012                                 }
1013                         }
1014                 }
1015
1016                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1017                 e1000_rx_checksum(adapter,
1018                                   (u32)(status) |
1019                                   ((u32)(rx_desc->errors) << 24),
1020                                   le16_to_cpu(rx_desc->csum), skb);
1021
1022                 /* probably a little skewed due to removing CRC */
1023                 total_rx_bytes += skb->len;
1024                 total_rx_packets++;
1025
1026                 /* eth type trans needs skb->data to point to something */
1027                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1028                         e_err("pskb_may_pull failed.\n");
1029                         dev_kfree_skb(skb);
1030                         goto next_desc;
1031                 }
1032
1033                 e1000_receive_skb(adapter, netdev, skb, status,
1034                                   rx_desc->special);
1035
1036 next_desc:
1037                 rx_desc->status = 0;
1038
1039                 /* return some buffers to hardware, one at a time is too slow */
1040                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1041                         adapter->alloc_rx_buf(adapter, cleaned_count);
1042                         cleaned_count = 0;
1043                 }
1044
1045                 /* use prefetched values */
1046                 rx_desc = next_rxd;
1047                 buffer_info = next_buffer;
1048         }
1049         rx_ring->next_to_clean = i;
1050
1051         cleaned_count = e1000_desc_unused(rx_ring);
1052         if (cleaned_count)
1053                 adapter->alloc_rx_buf(adapter, cleaned_count);
1054
1055         adapter->total_rx_bytes += total_rx_bytes;
1056         adapter->total_rx_packets += total_rx_packets;
1057         adapter->net_stats.rx_bytes += total_rx_bytes;
1058         adapter->net_stats.rx_packets += total_rx_packets;
1059         return cleaned;
1060 }
1061
1062 /**
1063  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1064  * @adapter: board private structure
1065  **/
1066 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1067 {
1068         struct e1000_ring *rx_ring = adapter->rx_ring;
1069         struct e1000_buffer *buffer_info;
1070         struct e1000_ps_page *ps_page;
1071         struct pci_dev *pdev = adapter->pdev;
1072         unsigned int i, j;
1073
1074         /* Free all the Rx ring sk_buffs */
1075         for (i = 0; i < rx_ring->count; i++) {
1076                 buffer_info = &rx_ring->buffer_info[i];
1077                 if (buffer_info->dma) {
1078                         if (adapter->clean_rx == e1000_clean_rx_irq)
1079                                 pci_unmap_single(pdev, buffer_info->dma,
1080                                                  adapter->rx_buffer_len,
1081                                                  PCI_DMA_FROMDEVICE);
1082                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1083                                 pci_unmap_page(pdev, buffer_info->dma,
1084                                                PAGE_SIZE,
1085                                                PCI_DMA_FROMDEVICE);
1086                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1087                                 pci_unmap_single(pdev, buffer_info->dma,
1088                                                  adapter->rx_ps_bsize0,
1089                                                  PCI_DMA_FROMDEVICE);
1090                         buffer_info->dma = 0;
1091                 }
1092
1093                 if (buffer_info->page) {
1094                         put_page(buffer_info->page);
1095                         buffer_info->page = NULL;
1096                 }
1097
1098                 if (buffer_info->skb) {
1099                         dev_kfree_skb(buffer_info->skb);
1100                         buffer_info->skb = NULL;
1101                 }
1102
1103                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1104                         ps_page = &buffer_info->ps_pages[j];
1105                         if (!ps_page->page)
1106                                 break;
1107                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1108                                        PCI_DMA_FROMDEVICE);
1109                         ps_page->dma = 0;
1110                         put_page(ps_page->page);
1111                         ps_page->page = NULL;
1112                 }
1113         }
1114
1115         /* there also may be some cached data from a chained receive */
1116         if (rx_ring->rx_skb_top) {
1117                 dev_kfree_skb(rx_ring->rx_skb_top);
1118                 rx_ring->rx_skb_top = NULL;
1119         }
1120
1121         /* Zero out the descriptor ring */
1122         memset(rx_ring->desc, 0, rx_ring->size);
1123
1124         rx_ring->next_to_clean = 0;
1125         rx_ring->next_to_use = 0;
1126
1127         writel(0, adapter->hw.hw_addr + rx_ring->head);
1128         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1129 }
1130
1131 static void e1000e_downshift_workaround(struct work_struct *work)
1132 {
1133         struct e1000_adapter *adapter = container_of(work,
1134                                         struct e1000_adapter, downshift_task);
1135
1136         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1137 }
1138
1139 /**
1140  * e1000_intr_msi - Interrupt Handler
1141  * @irq: interrupt number
1142  * @data: pointer to a network interface device structure
1143  **/
1144 static irqreturn_t e1000_intr_msi(int irq, void *data)
1145 {
1146         struct net_device *netdev = data;
1147         struct e1000_adapter *adapter = netdev_priv(netdev);
1148         struct e1000_hw *hw = &adapter->hw;
1149         u32 icr = er32(ICR);
1150
1151         /*
1152          * read ICR disables interrupts using IAM
1153          */
1154
1155         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1156                 hw->mac.get_link_status = 1;
1157                 /*
1158                  * ICH8 workaround-- Call gig speed drop workaround on cable
1159                  * disconnect (LSC) before accessing any PHY registers
1160                  */
1161                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1162                     (!(er32(STATUS) & E1000_STATUS_LU)))
1163                         schedule_work(&adapter->downshift_task);
1164
1165                 /*
1166                  * 80003ES2LAN workaround-- For packet buffer work-around on
1167                  * link down event; disable receives here in the ISR and reset
1168                  * adapter in watchdog
1169                  */
1170                 if (netif_carrier_ok(netdev) &&
1171                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1172                         /* disable receives */
1173                         u32 rctl = er32(RCTL);
1174                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1175                         adapter->flags |= FLAG_RX_RESTART_NOW;
1176                 }
1177                 /* guard against interrupt when we're going down */
1178                 if (!test_bit(__E1000_DOWN, &adapter->state))
1179                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1180         }
1181
1182         if (netif_rx_schedule_prep(&adapter->napi)) {
1183                 adapter->total_tx_bytes = 0;
1184                 adapter->total_tx_packets = 0;
1185                 adapter->total_rx_bytes = 0;
1186                 adapter->total_rx_packets = 0;
1187                 __netif_rx_schedule(&adapter->napi);
1188         }
1189
1190         return IRQ_HANDLED;
1191 }
1192
1193 /**
1194  * e1000_intr - Interrupt Handler
1195  * @irq: interrupt number
1196  * @data: pointer to a network interface device structure
1197  **/
1198 static irqreturn_t e1000_intr(int irq, void *data)
1199 {
1200         struct net_device *netdev = data;
1201         struct e1000_adapter *adapter = netdev_priv(netdev);
1202         struct e1000_hw *hw = &adapter->hw;
1203         u32 rctl, icr = er32(ICR);
1204
1205         if (!icr)
1206                 return IRQ_NONE;  /* Not our interrupt */
1207
1208         /*
1209          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1210          * not set, then the adapter didn't send an interrupt
1211          */
1212         if (!(icr & E1000_ICR_INT_ASSERTED))
1213                 return IRQ_NONE;
1214
1215         /*
1216          * Interrupt Auto-Mask...upon reading ICR,
1217          * interrupts are masked.  No need for the
1218          * IMC write
1219          */
1220
1221         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1222                 hw->mac.get_link_status = 1;
1223                 /*
1224                  * ICH8 workaround-- Call gig speed drop workaround on cable
1225                  * disconnect (LSC) before accessing any PHY registers
1226                  */
1227                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1228                     (!(er32(STATUS) & E1000_STATUS_LU)))
1229                         schedule_work(&adapter->downshift_task);
1230
1231                 /*
1232                  * 80003ES2LAN workaround--
1233                  * For packet buffer work-around on link down event;
1234                  * disable receives here in the ISR and
1235                  * reset adapter in watchdog
1236                  */
1237                 if (netif_carrier_ok(netdev) &&
1238                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1239                         /* disable receives */
1240                         rctl = er32(RCTL);
1241                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1242                         adapter->flags |= FLAG_RX_RESTART_NOW;
1243                 }
1244                 /* guard against interrupt when we're going down */
1245                 if (!test_bit(__E1000_DOWN, &adapter->state))
1246                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1247         }
1248
1249         if (netif_rx_schedule_prep(&adapter->napi)) {
1250                 adapter->total_tx_bytes = 0;
1251                 adapter->total_tx_packets = 0;
1252                 adapter->total_rx_bytes = 0;
1253                 adapter->total_rx_packets = 0;
1254                 __netif_rx_schedule(&adapter->napi);
1255         }
1256
1257         return IRQ_HANDLED;
1258 }
1259
1260 static irqreturn_t e1000_msix_other(int irq, void *data)
1261 {
1262         struct net_device *netdev = data;
1263         struct e1000_adapter *adapter = netdev_priv(netdev);
1264         struct e1000_hw *hw = &adapter->hw;
1265         u32 icr = er32(ICR);
1266
1267         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1268                 ew32(IMS, E1000_IMS_OTHER);
1269                 return IRQ_NONE;
1270         }
1271
1272         if (icr & adapter->eiac_mask)
1273                 ew32(ICS, (icr & adapter->eiac_mask));
1274
1275         if (icr & E1000_ICR_OTHER) {
1276                 if (!(icr & E1000_ICR_LSC))
1277                         goto no_link_interrupt;
1278                 hw->mac.get_link_status = 1;
1279                 /* guard against interrupt when we're going down */
1280                 if (!test_bit(__E1000_DOWN, &adapter->state))
1281                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1282         }
1283
1284 no_link_interrupt:
1285         ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1286
1287         return IRQ_HANDLED;
1288 }
1289
1290
1291 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1292 {
1293         struct net_device *netdev = data;
1294         struct e1000_adapter *adapter = netdev_priv(netdev);
1295         struct e1000_hw *hw = &adapter->hw;
1296         struct e1000_ring *tx_ring = adapter->tx_ring;
1297
1298
1299         adapter->total_tx_bytes = 0;
1300         adapter->total_tx_packets = 0;
1301
1302         if (!e1000_clean_tx_irq(adapter))
1303                 /* Ring was not completely cleaned, so fire another interrupt */
1304                 ew32(ICS, tx_ring->ims_val);
1305
1306         return IRQ_HANDLED;
1307 }
1308
1309 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1310 {
1311         struct net_device *netdev = data;
1312         struct e1000_adapter *adapter = netdev_priv(netdev);
1313
1314         /* Write the ITR value calculated at the end of the
1315          * previous interrupt.
1316          */
1317         if (adapter->rx_ring->set_itr) {
1318                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1319                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1320                 adapter->rx_ring->set_itr = 0;
1321         }
1322
1323         if (netif_rx_schedule_prep(&adapter->napi)) {
1324                 adapter->total_rx_bytes = 0;
1325                 adapter->total_rx_packets = 0;
1326                 __netif_rx_schedule(&adapter->napi);
1327         }
1328         return IRQ_HANDLED;
1329 }
1330
1331 /**
1332  * e1000_configure_msix - Configure MSI-X hardware
1333  *
1334  * e1000_configure_msix sets up the hardware to properly
1335  * generate MSI-X interrupts.
1336  **/
1337 static void e1000_configure_msix(struct e1000_adapter *adapter)
1338 {
1339         struct e1000_hw *hw = &adapter->hw;
1340         struct e1000_ring *rx_ring = adapter->rx_ring;
1341         struct e1000_ring *tx_ring = adapter->tx_ring;
1342         int vector = 0;
1343         u32 ctrl_ext, ivar = 0;
1344
1345         adapter->eiac_mask = 0;
1346
1347         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1348         if (hw->mac.type == e1000_82574) {
1349                 u32 rfctl = er32(RFCTL);
1350                 rfctl |= E1000_RFCTL_ACK_DIS;
1351                 ew32(RFCTL, rfctl);
1352         }
1353
1354 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1355         /* Configure Rx vector */
1356         rx_ring->ims_val = E1000_IMS_RXQ0;
1357         adapter->eiac_mask |= rx_ring->ims_val;
1358         if (rx_ring->itr_val)
1359                 writel(1000000000 / (rx_ring->itr_val * 256),
1360                        hw->hw_addr + rx_ring->itr_register);
1361         else
1362                 writel(1, hw->hw_addr + rx_ring->itr_register);
1363         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1364
1365         /* Configure Tx vector */
1366         tx_ring->ims_val = E1000_IMS_TXQ0;
1367         vector++;
1368         if (tx_ring->itr_val)
1369                 writel(1000000000 / (tx_ring->itr_val * 256),
1370                        hw->hw_addr + tx_ring->itr_register);
1371         else
1372                 writel(1, hw->hw_addr + tx_ring->itr_register);
1373         adapter->eiac_mask |= tx_ring->ims_val;
1374         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1375
1376         /* set vector for Other Causes, e.g. link changes */
1377         vector++;
1378         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1379         if (rx_ring->itr_val)
1380                 writel(1000000000 / (rx_ring->itr_val * 256),
1381                        hw->hw_addr + E1000_EITR_82574(vector));
1382         else
1383                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1384
1385         /* Cause Tx interrupts on every write back */
1386         ivar |= (1 << 31);
1387
1388         ew32(IVAR, ivar);
1389
1390         /* enable MSI-X PBA support */
1391         ctrl_ext = er32(CTRL_EXT);
1392         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1393
1394         /* Auto-Mask Other interrupts upon ICR read */
1395 #define E1000_EIAC_MASK_82574   0x01F00000
1396         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1397         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1398         ew32(CTRL_EXT, ctrl_ext);
1399         e1e_flush();
1400 }
1401
1402 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1403 {
1404         if (adapter->msix_entries) {
1405                 pci_disable_msix(adapter->pdev);
1406                 kfree(adapter->msix_entries);
1407                 adapter->msix_entries = NULL;
1408         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1409                 pci_disable_msi(adapter->pdev);
1410                 adapter->flags &= ~FLAG_MSI_ENABLED;
1411         }
1412
1413         return;
1414 }
1415
1416 /**
1417  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1418  *
1419  * Attempt to configure interrupts using the best available
1420  * capabilities of the hardware and kernel.
1421  **/
1422 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1423 {
1424         int err;
1425         int numvecs, i;
1426
1427
1428         switch (adapter->int_mode) {
1429         case E1000E_INT_MODE_MSIX:
1430                 if (adapter->flags & FLAG_HAS_MSIX) {
1431                         numvecs = 3; /* RxQ0, TxQ0 and other */
1432                         adapter->msix_entries = kcalloc(numvecs,
1433                                                       sizeof(struct msix_entry),
1434                                                       GFP_KERNEL);
1435                         if (adapter->msix_entries) {
1436                                 for (i = 0; i < numvecs; i++)
1437                                         adapter->msix_entries[i].entry = i;
1438
1439                                 err = pci_enable_msix(adapter->pdev,
1440                                                       adapter->msix_entries,
1441                                                       numvecs);
1442                                 if (err == 0)
1443                                         return;
1444                         }
1445                         /* MSI-X failed, so fall through and try MSI */
1446                         e_err("Failed to initialize MSI-X interrupts.  "
1447                               "Falling back to MSI interrupts.\n");
1448                         e1000e_reset_interrupt_capability(adapter);
1449                 }
1450                 adapter->int_mode = E1000E_INT_MODE_MSI;
1451                 /* Fall through */
1452         case E1000E_INT_MODE_MSI:
1453                 if (!pci_enable_msi(adapter->pdev)) {
1454                         adapter->flags |= FLAG_MSI_ENABLED;
1455                 } else {
1456                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1457                         e_err("Failed to initialize MSI interrupts.  Falling "
1458                               "back to legacy interrupts.\n");
1459                 }
1460                 /* Fall through */
1461         case E1000E_INT_MODE_LEGACY:
1462                 /* Don't do anything; this is the system default */
1463                 break;
1464         }
1465
1466         return;
1467 }
1468
1469 /**
1470  * e1000_request_msix - Initialize MSI-X interrupts
1471  *
1472  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1473  * kernel.
1474  **/
1475 static int e1000_request_msix(struct e1000_adapter *adapter)
1476 {
1477         struct net_device *netdev = adapter->netdev;
1478         int err = 0, vector = 0;
1479
1480         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1481                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1482         else
1483                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1484         err = request_irq(adapter->msix_entries[vector].vector,
1485                           &e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1486                           netdev);
1487         if (err)
1488                 goto out;
1489         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1490         adapter->rx_ring->itr_val = adapter->itr;
1491         vector++;
1492
1493         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1494                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1495         else
1496                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1497         err = request_irq(adapter->msix_entries[vector].vector,
1498                           &e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1499                           netdev);
1500         if (err)
1501                 goto out;
1502         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1503         adapter->tx_ring->itr_val = adapter->itr;
1504         vector++;
1505
1506         err = request_irq(adapter->msix_entries[vector].vector,
1507                           &e1000_msix_other, 0, netdev->name, netdev);
1508         if (err)
1509                 goto out;
1510
1511         e1000_configure_msix(adapter);
1512         return 0;
1513 out:
1514         return err;
1515 }
1516
1517 /**
1518  * e1000_request_irq - initialize interrupts
1519  *
1520  * Attempts to configure interrupts using the best available
1521  * capabilities of the hardware and kernel.
1522  **/
1523 static int e1000_request_irq(struct e1000_adapter *adapter)
1524 {
1525         struct net_device *netdev = adapter->netdev;
1526         int err;
1527
1528         if (adapter->msix_entries) {
1529                 err = e1000_request_msix(adapter);
1530                 if (!err)
1531                         return err;
1532                 /* fall back to MSI */
1533                 e1000e_reset_interrupt_capability(adapter);
1534                 adapter->int_mode = E1000E_INT_MODE_MSI;
1535                 e1000e_set_interrupt_capability(adapter);
1536         }
1537         if (adapter->flags & FLAG_MSI_ENABLED) {
1538                 err = request_irq(adapter->pdev->irq, &e1000_intr_msi, 0,
1539                                   netdev->name, netdev);
1540                 if (!err)
1541                         return err;
1542
1543                 /* fall back to legacy interrupt */
1544                 e1000e_reset_interrupt_capability(adapter);
1545                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1546         }
1547
1548         err = request_irq(adapter->pdev->irq, &e1000_intr, IRQF_SHARED,
1549                           netdev->name, netdev);
1550         if (err)
1551                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1552
1553         return err;
1554 }
1555
1556 static void e1000_free_irq(struct e1000_adapter *adapter)
1557 {
1558         struct net_device *netdev = adapter->netdev;
1559
1560         if (adapter->msix_entries) {
1561                 int vector = 0;
1562
1563                 free_irq(adapter->msix_entries[vector].vector, netdev);
1564                 vector++;
1565
1566                 free_irq(adapter->msix_entries[vector].vector, netdev);
1567                 vector++;
1568
1569                 /* Other Causes interrupt vector */
1570                 free_irq(adapter->msix_entries[vector].vector, netdev);
1571                 return;
1572         }
1573
1574         free_irq(adapter->pdev->irq, netdev);
1575 }
1576
1577 /**
1578  * e1000_irq_disable - Mask off interrupt generation on the NIC
1579  **/
1580 static void e1000_irq_disable(struct e1000_adapter *adapter)
1581 {
1582         struct e1000_hw *hw = &adapter->hw;
1583
1584         ew32(IMC, ~0);
1585         if (adapter->msix_entries)
1586                 ew32(EIAC_82574, 0);
1587         e1e_flush();
1588         synchronize_irq(adapter->pdev->irq);
1589 }
1590
1591 /**
1592  * e1000_irq_enable - Enable default interrupt generation settings
1593  **/
1594 static void e1000_irq_enable(struct e1000_adapter *adapter)
1595 {
1596         struct e1000_hw *hw = &adapter->hw;
1597
1598         if (adapter->msix_entries) {
1599                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1600                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1601         } else {
1602                 ew32(IMS, IMS_ENABLE_MASK);
1603         }
1604         e1e_flush();
1605 }
1606
1607 /**
1608  * e1000_get_hw_control - get control of the h/w from f/w
1609  * @adapter: address of board private structure
1610  *
1611  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1612  * For ASF and Pass Through versions of f/w this means that
1613  * the driver is loaded. For AMT version (only with 82573)
1614  * of the f/w this means that the network i/f is open.
1615  **/
1616 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1617 {
1618         struct e1000_hw *hw = &adapter->hw;
1619         u32 ctrl_ext;
1620         u32 swsm;
1621
1622         /* Let firmware know the driver has taken over */
1623         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1624                 swsm = er32(SWSM);
1625                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1626         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1627                 ctrl_ext = er32(CTRL_EXT);
1628                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1629         }
1630 }
1631
1632 /**
1633  * e1000_release_hw_control - release control of the h/w to f/w
1634  * @adapter: address of board private structure
1635  *
1636  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1637  * For ASF and Pass Through versions of f/w this means that the
1638  * driver is no longer loaded. For AMT version (only with 82573) i
1639  * of the f/w this means that the network i/f is closed.
1640  *
1641  **/
1642 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1643 {
1644         struct e1000_hw *hw = &adapter->hw;
1645         u32 ctrl_ext;
1646         u32 swsm;
1647
1648         /* Let firmware taken over control of h/w */
1649         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1650                 swsm = er32(SWSM);
1651                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1652         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1653                 ctrl_ext = er32(CTRL_EXT);
1654                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1655         }
1656 }
1657
1658 /**
1659  * @e1000_alloc_ring - allocate memory for a ring structure
1660  **/
1661 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1662                                 struct e1000_ring *ring)
1663 {
1664         struct pci_dev *pdev = adapter->pdev;
1665
1666         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1667                                         GFP_KERNEL);
1668         if (!ring->desc)
1669                 return -ENOMEM;
1670
1671         return 0;
1672 }
1673
1674 /**
1675  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1676  * @adapter: board private structure
1677  *
1678  * Return 0 on success, negative on failure
1679  **/
1680 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1681 {
1682         struct e1000_ring *tx_ring = adapter->tx_ring;
1683         int err = -ENOMEM, size;
1684
1685         size = sizeof(struct e1000_buffer) * tx_ring->count;
1686         tx_ring->buffer_info = vmalloc(size);
1687         if (!tx_ring->buffer_info)
1688                 goto err;
1689         memset(tx_ring->buffer_info, 0, size);
1690
1691         /* round up to nearest 4K */
1692         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1693         tx_ring->size = ALIGN(tx_ring->size, 4096);
1694
1695         err = e1000_alloc_ring_dma(adapter, tx_ring);
1696         if (err)
1697                 goto err;
1698
1699         tx_ring->next_to_use = 0;
1700         tx_ring->next_to_clean = 0;
1701         spin_lock_init(&adapter->tx_queue_lock);
1702
1703         return 0;
1704 err:
1705         vfree(tx_ring->buffer_info);
1706         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1707         return err;
1708 }
1709
1710 /**
1711  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1712  * @adapter: board private structure
1713  *
1714  * Returns 0 on success, negative on failure
1715  **/
1716 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1717 {
1718         struct e1000_ring *rx_ring = adapter->rx_ring;
1719         struct e1000_buffer *buffer_info;
1720         int i, size, desc_len, err = -ENOMEM;
1721
1722         size = sizeof(struct e1000_buffer) * rx_ring->count;
1723         rx_ring->buffer_info = vmalloc(size);
1724         if (!rx_ring->buffer_info)
1725                 goto err;
1726         memset(rx_ring->buffer_info, 0, size);
1727
1728         for (i = 0; i < rx_ring->count; i++) {
1729                 buffer_info = &rx_ring->buffer_info[i];
1730                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1731                                                 sizeof(struct e1000_ps_page),
1732                                                 GFP_KERNEL);
1733                 if (!buffer_info->ps_pages)
1734                         goto err_pages;
1735         }
1736
1737         desc_len = sizeof(union e1000_rx_desc_packet_split);
1738
1739         /* Round up to nearest 4K */
1740         rx_ring->size = rx_ring->count * desc_len;
1741         rx_ring->size = ALIGN(rx_ring->size, 4096);
1742
1743         err = e1000_alloc_ring_dma(adapter, rx_ring);
1744         if (err)
1745                 goto err_pages;
1746
1747         rx_ring->next_to_clean = 0;
1748         rx_ring->next_to_use = 0;
1749         rx_ring->rx_skb_top = NULL;
1750
1751         return 0;
1752
1753 err_pages:
1754         for (i = 0; i < rx_ring->count; i++) {
1755                 buffer_info = &rx_ring->buffer_info[i];
1756                 kfree(buffer_info->ps_pages);
1757         }
1758 err:
1759         vfree(rx_ring->buffer_info);
1760         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1761         return err;
1762 }
1763
1764 /**
1765  * e1000_clean_tx_ring - Free Tx Buffers
1766  * @adapter: board private structure
1767  **/
1768 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1769 {
1770         struct e1000_ring *tx_ring = adapter->tx_ring;
1771         struct e1000_buffer *buffer_info;
1772         unsigned long size;
1773         unsigned int i;
1774
1775         for (i = 0; i < tx_ring->count; i++) {
1776                 buffer_info = &tx_ring->buffer_info[i];
1777                 e1000_put_txbuf(adapter, buffer_info);
1778         }
1779
1780         size = sizeof(struct e1000_buffer) * tx_ring->count;
1781         memset(tx_ring->buffer_info, 0, size);
1782
1783         memset(tx_ring->desc, 0, tx_ring->size);
1784
1785         tx_ring->next_to_use = 0;
1786         tx_ring->next_to_clean = 0;
1787
1788         writel(0, adapter->hw.hw_addr + tx_ring->head);
1789         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1790 }
1791
1792 /**
1793  * e1000e_free_tx_resources - Free Tx Resources per Queue
1794  * @adapter: board private structure
1795  *
1796  * Free all transmit software resources
1797  **/
1798 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1799 {
1800         struct pci_dev *pdev = adapter->pdev;
1801         struct e1000_ring *tx_ring = adapter->tx_ring;
1802
1803         e1000_clean_tx_ring(adapter);
1804
1805         vfree(tx_ring->buffer_info);
1806         tx_ring->buffer_info = NULL;
1807
1808         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1809                           tx_ring->dma);
1810         tx_ring->desc = NULL;
1811 }
1812
1813 /**
1814  * e1000e_free_rx_resources - Free Rx Resources
1815  * @adapter: board private structure
1816  *
1817  * Free all receive software resources
1818  **/
1819
1820 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1821 {
1822         struct pci_dev *pdev = adapter->pdev;
1823         struct e1000_ring *rx_ring = adapter->rx_ring;
1824         int i;
1825
1826         e1000_clean_rx_ring(adapter);
1827
1828         for (i = 0; i < rx_ring->count; i++) {
1829                 kfree(rx_ring->buffer_info[i].ps_pages);
1830         }
1831
1832         vfree(rx_ring->buffer_info);
1833         rx_ring->buffer_info = NULL;
1834
1835         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1836                           rx_ring->dma);
1837         rx_ring->desc = NULL;
1838 }
1839
1840 /**
1841  * e1000_update_itr - update the dynamic ITR value based on statistics
1842  * @adapter: pointer to adapter
1843  * @itr_setting: current adapter->itr
1844  * @packets: the number of packets during this measurement interval
1845  * @bytes: the number of bytes during this measurement interval
1846  *
1847  *      Stores a new ITR value based on packets and byte
1848  *      counts during the last interrupt.  The advantage of per interrupt
1849  *      computation is faster updates and more accurate ITR for the current
1850  *      traffic pattern.  Constants in this function were computed
1851  *      based on theoretical maximum wire speed and thresholds were set based
1852  *      on testing data as well as attempting to minimize response time
1853  *      while increasing bulk throughput.  This functionality is controlled
1854  *      by the InterruptThrottleRate module parameter.
1855  **/
1856 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1857                                      u16 itr_setting, int packets,
1858                                      int bytes)
1859 {
1860         unsigned int retval = itr_setting;
1861
1862         if (packets == 0)
1863                 goto update_itr_done;
1864
1865         switch (itr_setting) {
1866         case lowest_latency:
1867                 /* handle TSO and jumbo frames */
1868                 if (bytes/packets > 8000)
1869                         retval = bulk_latency;
1870                 else if ((packets < 5) && (bytes > 512)) {
1871                         retval = low_latency;
1872                 }
1873                 break;
1874         case low_latency:  /* 50 usec aka 20000 ints/s */
1875                 if (bytes > 10000) {
1876                         /* this if handles the TSO accounting */
1877                         if (bytes/packets > 8000) {
1878                                 retval = bulk_latency;
1879                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1880                                 retval = bulk_latency;
1881                         } else if ((packets > 35)) {
1882                                 retval = lowest_latency;
1883                         }
1884                 } else if (bytes/packets > 2000) {
1885                         retval = bulk_latency;
1886                 } else if (packets <= 2 && bytes < 512) {
1887                         retval = lowest_latency;
1888                 }
1889                 break;
1890         case bulk_latency: /* 250 usec aka 4000 ints/s */
1891                 if (bytes > 25000) {
1892                         if (packets > 35) {
1893                                 retval = low_latency;
1894                         }
1895                 } else if (bytes < 6000) {
1896                         retval = low_latency;
1897                 }
1898                 break;
1899         }
1900
1901 update_itr_done:
1902         return retval;
1903 }
1904
1905 static void e1000_set_itr(struct e1000_adapter *adapter)
1906 {
1907         struct e1000_hw *hw = &adapter->hw;
1908         u16 current_itr;
1909         u32 new_itr = adapter->itr;
1910
1911         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1912         if (adapter->link_speed != SPEED_1000) {
1913                 current_itr = 0;
1914                 new_itr = 4000;
1915                 goto set_itr_now;
1916         }
1917
1918         adapter->tx_itr = e1000_update_itr(adapter,
1919                                     adapter->tx_itr,
1920                                     adapter->total_tx_packets,
1921                                     adapter->total_tx_bytes);
1922         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1923         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1924                 adapter->tx_itr = low_latency;
1925
1926         adapter->rx_itr = e1000_update_itr(adapter,
1927                                     adapter->rx_itr,
1928                                     adapter->total_rx_packets,
1929                                     adapter->total_rx_bytes);
1930         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1931         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1932                 adapter->rx_itr = low_latency;
1933
1934         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1935
1936         switch (current_itr) {
1937         /* counts and packets in update_itr are dependent on these numbers */
1938         case lowest_latency:
1939                 new_itr = 70000;
1940                 break;
1941         case low_latency:
1942                 new_itr = 20000; /* aka hwitr = ~200 */
1943                 break;
1944         case bulk_latency:
1945                 new_itr = 4000;
1946                 break;
1947         default:
1948                 break;
1949         }
1950
1951 set_itr_now:
1952         if (new_itr != adapter->itr) {
1953                 /*
1954                  * this attempts to bias the interrupt rate towards Bulk
1955                  * by adding intermediate steps when interrupt rate is
1956                  * increasing
1957                  */
1958                 new_itr = new_itr > adapter->itr ?
1959                              min(adapter->itr + (new_itr >> 2), new_itr) :
1960                              new_itr;
1961                 adapter->itr = new_itr;
1962                 adapter->rx_ring->itr_val = new_itr;
1963                 if (adapter->msix_entries)
1964                         adapter->rx_ring->set_itr = 1;
1965                 else
1966                         ew32(ITR, 1000000000 / (new_itr * 256));
1967         }
1968 }
1969
1970 /**
1971  * e1000_alloc_queues - Allocate memory for all rings
1972  * @adapter: board private structure to initialize
1973  **/
1974 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1975 {
1976         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1977         if (!adapter->tx_ring)
1978                 goto err;
1979
1980         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1981         if (!adapter->rx_ring)
1982                 goto err;
1983
1984         return 0;
1985 err:
1986         e_err("Unable to allocate memory for queues\n");
1987         kfree(adapter->rx_ring);
1988         kfree(adapter->tx_ring);
1989         return -ENOMEM;
1990 }
1991
1992 /**
1993  * e1000_clean - NAPI Rx polling callback
1994  * @napi: struct associated with this polling callback
1995  * @budget: amount of packets driver is allowed to process this poll
1996  **/
1997 static int e1000_clean(struct napi_struct *napi, int budget)
1998 {
1999         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2000         struct e1000_hw *hw = &adapter->hw;
2001         struct net_device *poll_dev = adapter->netdev;
2002         int tx_cleaned = 0, work_done = 0;
2003
2004         adapter = netdev_priv(poll_dev);
2005
2006         if (adapter->msix_entries &&
2007             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2008                 goto clean_rx;
2009
2010         /*
2011          * e1000_clean is called per-cpu.  This lock protects
2012          * tx_ring from being cleaned by multiple cpus
2013          * simultaneously.  A failure obtaining the lock means
2014          * tx_ring is currently being cleaned anyway.
2015          */
2016         if (spin_trylock(&adapter->tx_queue_lock)) {
2017                 tx_cleaned = e1000_clean_tx_irq(adapter);
2018                 spin_unlock(&adapter->tx_queue_lock);
2019         }
2020
2021 clean_rx:
2022         adapter->clean_rx(adapter, &work_done, budget);
2023
2024         if (tx_cleaned)
2025                 work_done = budget;
2026
2027         /* If budget not fully consumed, exit the polling mode */
2028         if (work_done < budget) {
2029                 if (adapter->itr_setting & 3)
2030                         e1000_set_itr(adapter);
2031                 netif_rx_complete(napi);
2032                 if (adapter->msix_entries)
2033                         ew32(IMS, adapter->rx_ring->ims_val);
2034                 else
2035                         e1000_irq_enable(adapter);
2036         }
2037
2038         return work_done;
2039 }
2040
2041 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2042 {
2043         struct e1000_adapter *adapter = netdev_priv(netdev);
2044         struct e1000_hw *hw = &adapter->hw;
2045         u32 vfta, index;
2046
2047         /* don't update vlan cookie if already programmed */
2048         if ((adapter->hw.mng_cookie.status &
2049              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2050             (vid == adapter->mng_vlan_id))
2051                 return;
2052         /* add VID to filter table */
2053         index = (vid >> 5) & 0x7F;
2054         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2055         vfta |= (1 << (vid & 0x1F));
2056         e1000e_write_vfta(hw, index, vfta);
2057 }
2058
2059 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2060 {
2061         struct e1000_adapter *adapter = netdev_priv(netdev);
2062         struct e1000_hw *hw = &adapter->hw;
2063         u32 vfta, index;
2064
2065         if (!test_bit(__E1000_DOWN, &adapter->state))
2066                 e1000_irq_disable(adapter);
2067         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2068
2069         if (!test_bit(__E1000_DOWN, &adapter->state))
2070                 e1000_irq_enable(adapter);
2071
2072         if ((adapter->hw.mng_cookie.status &
2073              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2074             (vid == adapter->mng_vlan_id)) {
2075                 /* release control to f/w */
2076                 e1000_release_hw_control(adapter);
2077                 return;
2078         }
2079
2080         /* remove VID from filter table */
2081         index = (vid >> 5) & 0x7F;
2082         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2083         vfta &= ~(1 << (vid & 0x1F));
2084         e1000e_write_vfta(hw, index, vfta);
2085 }
2086
2087 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2088 {
2089         struct net_device *netdev = adapter->netdev;
2090         u16 vid = adapter->hw.mng_cookie.vlan_id;
2091         u16 old_vid = adapter->mng_vlan_id;
2092
2093         if (!adapter->vlgrp)
2094                 return;
2095
2096         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2097                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2098                 if (adapter->hw.mng_cookie.status &
2099                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2100                         e1000_vlan_rx_add_vid(netdev, vid);
2101                         adapter->mng_vlan_id = vid;
2102                 }
2103
2104                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2105                                 (vid != old_vid) &&
2106                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2107                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2108         } else {
2109                 adapter->mng_vlan_id = vid;
2110         }
2111 }
2112
2113
2114 static void e1000_vlan_rx_register(struct net_device *netdev,
2115                                    struct vlan_group *grp)
2116 {
2117         struct e1000_adapter *adapter = netdev_priv(netdev);
2118         struct e1000_hw *hw = &adapter->hw;
2119         u32 ctrl, rctl;
2120
2121         if (!test_bit(__E1000_DOWN, &adapter->state))
2122                 e1000_irq_disable(adapter);
2123         adapter->vlgrp = grp;
2124
2125         if (grp) {
2126                 /* enable VLAN tag insert/strip */
2127                 ctrl = er32(CTRL);
2128                 ctrl |= E1000_CTRL_VME;
2129                 ew32(CTRL, ctrl);
2130
2131                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2132                         /* enable VLAN receive filtering */
2133                         rctl = er32(RCTL);
2134                         rctl &= ~E1000_RCTL_CFIEN;
2135                         ew32(RCTL, rctl);
2136                         e1000_update_mng_vlan(adapter);
2137                 }
2138         } else {
2139                 /* disable VLAN tag insert/strip */
2140                 ctrl = er32(CTRL);
2141                 ctrl &= ~E1000_CTRL_VME;
2142                 ew32(CTRL, ctrl);
2143
2144                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2145                         if (adapter->mng_vlan_id !=
2146                             (u16)E1000_MNG_VLAN_NONE) {
2147                                 e1000_vlan_rx_kill_vid(netdev,
2148                                                        adapter->mng_vlan_id);
2149                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2150                         }
2151                 }
2152         }
2153
2154         if (!test_bit(__E1000_DOWN, &adapter->state))
2155                 e1000_irq_enable(adapter);
2156 }
2157
2158 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2159 {
2160         u16 vid;
2161
2162         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2163
2164         if (!adapter->vlgrp)
2165                 return;
2166
2167         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2168                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2169                         continue;
2170                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2171         }
2172 }
2173
2174 static void e1000_init_manageability(struct e1000_adapter *adapter)
2175 {
2176         struct e1000_hw *hw = &adapter->hw;
2177         u32 manc, manc2h;
2178
2179         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2180                 return;
2181
2182         manc = er32(MANC);
2183
2184         /*
2185          * enable receiving management packets to the host. this will probably
2186          * generate destination unreachable messages from the host OS, but
2187          * the packets will be handled on SMBUS
2188          */
2189         manc |= E1000_MANC_EN_MNG2HOST;
2190         manc2h = er32(MANC2H);
2191 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2192 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2193         manc2h |= E1000_MNG2HOST_PORT_623;
2194         manc2h |= E1000_MNG2HOST_PORT_664;
2195         ew32(MANC2H, manc2h);
2196         ew32(MANC, manc);
2197 }
2198
2199 /**
2200  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2201  * @adapter: board private structure
2202  *
2203  * Configure the Tx unit of the MAC after a reset.
2204  **/
2205 static void e1000_configure_tx(struct e1000_adapter *adapter)
2206 {
2207         struct e1000_hw *hw = &adapter->hw;
2208         struct e1000_ring *tx_ring = adapter->tx_ring;
2209         u64 tdba;
2210         u32 tdlen, tctl, tipg, tarc;
2211         u32 ipgr1, ipgr2;
2212
2213         /* Setup the HW Tx Head and Tail descriptor pointers */
2214         tdba = tx_ring->dma;
2215         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2216         ew32(TDBAL, (tdba & DMA_32BIT_MASK));
2217         ew32(TDBAH, (tdba >> 32));
2218         ew32(TDLEN, tdlen);
2219         ew32(TDH, 0);
2220         ew32(TDT, 0);
2221         tx_ring->head = E1000_TDH;
2222         tx_ring->tail = E1000_TDT;
2223
2224         /* Set the default values for the Tx Inter Packet Gap timer */
2225         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2226         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2227         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2228
2229         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2230                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2231
2232         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2233         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2234         ew32(TIPG, tipg);
2235
2236         /* Set the Tx Interrupt Delay register */
2237         ew32(TIDV, adapter->tx_int_delay);
2238         /* Tx irq moderation */
2239         ew32(TADV, adapter->tx_abs_int_delay);
2240
2241         /* Program the Transmit Control Register */
2242         tctl = er32(TCTL);
2243         tctl &= ~E1000_TCTL_CT;
2244         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2245                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2246
2247         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2248                 tarc = er32(TARC(0));
2249                 /*
2250                  * set the speed mode bit, we'll clear it if we're not at
2251                  * gigabit link later
2252                  */
2253 #define SPEED_MODE_BIT (1 << 21)
2254                 tarc |= SPEED_MODE_BIT;
2255                 ew32(TARC(0), tarc);
2256         }
2257
2258         /* errata: program both queues to unweighted RR */
2259         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2260                 tarc = er32(TARC(0));
2261                 tarc |= 1;
2262                 ew32(TARC(0), tarc);
2263                 tarc = er32(TARC(1));
2264                 tarc |= 1;
2265                 ew32(TARC(1), tarc);
2266         }
2267
2268         e1000e_config_collision_dist(hw);
2269
2270         /* Setup Transmit Descriptor Settings for eop descriptor */
2271         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2272
2273         /* only set IDE if we are delaying interrupts using the timers */
2274         if (adapter->tx_int_delay)
2275                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2276
2277         /* enable Report Status bit */
2278         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2279
2280         ew32(TCTL, tctl);
2281
2282         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
2283 }
2284
2285 /**
2286  * e1000_setup_rctl - configure the receive control registers
2287  * @adapter: Board private structure
2288  **/
2289 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2290                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2291 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2292 {
2293         struct e1000_hw *hw = &adapter->hw;
2294         u32 rctl, rfctl;
2295         u32 psrctl = 0;
2296         u32 pages = 0;
2297
2298         /* Program MC offset vector base */
2299         rctl = er32(RCTL);
2300         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2301         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2302                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2303                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2304
2305         /* Do not Store bad packets */
2306         rctl &= ~E1000_RCTL_SBP;
2307
2308         /* Enable Long Packet receive */
2309         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2310                 rctl &= ~E1000_RCTL_LPE;
2311         else
2312                 rctl |= E1000_RCTL_LPE;
2313
2314         /* Some systems expect that the CRC is included in SMBUS traffic. The
2315          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2316          * host memory when this is enabled
2317          */
2318         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2319                 rctl |= E1000_RCTL_SECRC;
2320
2321         /* Setup buffer sizes */
2322         rctl &= ~E1000_RCTL_SZ_4096;
2323         rctl |= E1000_RCTL_BSEX;
2324         switch (adapter->rx_buffer_len) {
2325         case 256:
2326                 rctl |= E1000_RCTL_SZ_256;
2327                 rctl &= ~E1000_RCTL_BSEX;
2328                 break;
2329         case 512:
2330                 rctl |= E1000_RCTL_SZ_512;
2331                 rctl &= ~E1000_RCTL_BSEX;
2332                 break;
2333         case 1024:
2334                 rctl |= E1000_RCTL_SZ_1024;
2335                 rctl &= ~E1000_RCTL_BSEX;
2336                 break;
2337         case 2048:
2338         default:
2339                 rctl |= E1000_RCTL_SZ_2048;
2340                 rctl &= ~E1000_RCTL_BSEX;
2341                 break;
2342         case 4096:
2343                 rctl |= E1000_RCTL_SZ_4096;
2344                 break;
2345         case 8192:
2346                 rctl |= E1000_RCTL_SZ_8192;
2347                 break;
2348         case 16384:
2349                 rctl |= E1000_RCTL_SZ_16384;
2350                 break;
2351         }
2352
2353         /*
2354          * 82571 and greater support packet-split where the protocol
2355          * header is placed in skb->data and the packet data is
2356          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2357          * In the case of a non-split, skb->data is linearly filled,
2358          * followed by the page buffers.  Therefore, skb->data is
2359          * sized to hold the largest protocol header.
2360          *
2361          * allocations using alloc_page take too long for regular MTU
2362          * so only enable packet split for jumbo frames
2363          *
2364          * Using pages when the page size is greater than 16k wastes
2365          * a lot of memory, since we allocate 3 pages at all times
2366          * per packet.
2367          */
2368         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2369         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2370             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2371                 adapter->rx_ps_pages = pages;
2372         else
2373                 adapter->rx_ps_pages = 0;
2374
2375         if (adapter->rx_ps_pages) {
2376                 /* Configure extra packet-split registers */
2377                 rfctl = er32(RFCTL);
2378                 rfctl |= E1000_RFCTL_EXTEN;
2379                 /*
2380                  * disable packet split support for IPv6 extension headers,
2381                  * because some malformed IPv6 headers can hang the Rx
2382                  */
2383                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2384                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2385
2386                 ew32(RFCTL, rfctl);
2387
2388                 /* Enable Packet split descriptors */
2389                 rctl |= E1000_RCTL_DTYP_PS;
2390
2391                 psrctl |= adapter->rx_ps_bsize0 >>
2392                         E1000_PSRCTL_BSIZE0_SHIFT;
2393
2394                 switch (adapter->rx_ps_pages) {
2395                 case 3:
2396                         psrctl |= PAGE_SIZE <<
2397                                 E1000_PSRCTL_BSIZE3_SHIFT;
2398                 case 2:
2399                         psrctl |= PAGE_SIZE <<
2400                                 E1000_PSRCTL_BSIZE2_SHIFT;
2401                 case 1:
2402                         psrctl |= PAGE_SIZE >>
2403                                 E1000_PSRCTL_BSIZE1_SHIFT;
2404                         break;
2405                 }
2406
2407                 ew32(PSRCTL, psrctl);
2408         }
2409
2410         ew32(RCTL, rctl);
2411         /* just started the receive unit, no need to restart */
2412         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2413 }
2414
2415 /**
2416  * e1000_configure_rx - Configure Receive Unit after Reset
2417  * @adapter: board private structure
2418  *
2419  * Configure the Rx unit of the MAC after a reset.
2420  **/
2421 static void e1000_configure_rx(struct e1000_adapter *adapter)
2422 {
2423         struct e1000_hw *hw = &adapter->hw;
2424         struct e1000_ring *rx_ring = adapter->rx_ring;
2425         u64 rdba;
2426         u32 rdlen, rctl, rxcsum, ctrl_ext;
2427
2428         if (adapter->rx_ps_pages) {
2429                 /* this is a 32 byte descriptor */
2430                 rdlen = rx_ring->count *
2431                         sizeof(union e1000_rx_desc_packet_split);
2432                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2433                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2434         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2435                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2436                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2437                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2438         } else {
2439                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2440                 adapter->clean_rx = e1000_clean_rx_irq;
2441                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2442         }
2443
2444         /* disable receives while setting up the descriptors */
2445         rctl = er32(RCTL);
2446         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2447         e1e_flush();
2448         msleep(10);
2449
2450         /* set the Receive Delay Timer Register */
2451         ew32(RDTR, adapter->rx_int_delay);
2452
2453         /* irq moderation */
2454         ew32(RADV, adapter->rx_abs_int_delay);
2455         if (adapter->itr_setting != 0)
2456                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2457
2458         ctrl_ext = er32(CTRL_EXT);
2459         /* Reset delay timers after every interrupt */
2460         ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2461         /* Auto-Mask interrupts upon ICR access */
2462         ctrl_ext |= E1000_CTRL_EXT_IAME;
2463         ew32(IAM, 0xffffffff);
2464         ew32(CTRL_EXT, ctrl_ext);
2465         e1e_flush();
2466
2467         /*
2468          * Setup the HW Rx Head and Tail Descriptor Pointers and
2469          * the Base and Length of the Rx Descriptor Ring
2470          */
2471         rdba = rx_ring->dma;
2472         ew32(RDBAL, (rdba & DMA_32BIT_MASK));
2473         ew32(RDBAH, (rdba >> 32));
2474         ew32(RDLEN, rdlen);
2475         ew32(RDH, 0);
2476         ew32(RDT, 0);
2477         rx_ring->head = E1000_RDH;
2478         rx_ring->tail = E1000_RDT;
2479
2480         /* Enable Receive Checksum Offload for TCP and UDP */
2481         rxcsum = er32(RXCSUM);
2482         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2483                 rxcsum |= E1000_RXCSUM_TUOFL;
2484
2485                 /*
2486                  * IPv4 payload checksum for UDP fragments must be
2487                  * used in conjunction with packet-split.
2488                  */
2489                 if (adapter->rx_ps_pages)
2490                         rxcsum |= E1000_RXCSUM_IPPCSE;
2491         } else {
2492                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2493                 /* no need to clear IPPCSE as it defaults to 0 */
2494         }
2495         ew32(RXCSUM, rxcsum);
2496
2497         /*
2498          * Enable early receives on supported devices, only takes effect when
2499          * packet size is equal or larger than the specified value (in 8 byte
2500          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2501          */
2502         if ((adapter->flags & FLAG_HAS_ERT) &&
2503             (adapter->netdev->mtu > ETH_DATA_LEN)) {
2504                 u32 rxdctl = er32(RXDCTL(0));
2505                 ew32(RXDCTL(0), rxdctl | 0x3);
2506                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2507                 /*
2508                  * With jumbo frames and early-receive enabled, excessive
2509                  * C4->C2 latencies result in dropped transactions.
2510                  */
2511                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2512                                           e1000e_driver_name, 55);
2513         } else {
2514                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2515                                           e1000e_driver_name,
2516                                           PM_QOS_DEFAULT_VALUE);
2517         }
2518
2519         /* Enable Receives */
2520         ew32(RCTL, rctl);
2521 }
2522
2523 /**
2524  *  e1000_update_mc_addr_list - Update Multicast addresses
2525  *  @hw: pointer to the HW structure
2526  *  @mc_addr_list: array of multicast addresses to program
2527  *  @mc_addr_count: number of multicast addresses to program
2528  *  @rar_used_count: the first RAR register free to program
2529  *  @rar_count: total number of supported Receive Address Registers
2530  *
2531  *  Updates the Receive Address Registers and Multicast Table Array.
2532  *  The caller must have a packed mc_addr_list of multicast addresses.
2533  *  The parameter rar_count will usually be hw->mac.rar_entry_count
2534  *  unless there are workarounds that change this.  Currently no func pointer
2535  *  exists and all implementations are handled in the generic version of this
2536  *  function.
2537  **/
2538 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2539                                       u32 mc_addr_count, u32 rar_used_count,
2540                                       u32 rar_count)
2541 {
2542         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2543                                         rar_used_count, rar_count);
2544 }
2545
2546 /**
2547  * e1000_set_multi - Multicast and Promiscuous mode set
2548  * @netdev: network interface device structure
2549  *
2550  * The set_multi entry point is called whenever the multicast address
2551  * list or the network interface flags are updated.  This routine is
2552  * responsible for configuring the hardware for proper multicast,
2553  * promiscuous mode, and all-multi behavior.
2554  **/
2555 static void e1000_set_multi(struct net_device *netdev)
2556 {
2557         struct e1000_adapter *adapter = netdev_priv(netdev);
2558         struct e1000_hw *hw = &adapter->hw;
2559         struct e1000_mac_info *mac = &hw->mac;
2560         struct dev_mc_list *mc_ptr;
2561         u8  *mta_list;
2562         u32 rctl;
2563         int i;
2564
2565         /* Check for Promiscuous and All Multicast modes */
2566
2567         rctl = er32(RCTL);
2568
2569         if (netdev->flags & IFF_PROMISC) {
2570                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2571                 rctl &= ~E1000_RCTL_VFE;
2572         } else {
2573                 if (netdev->flags & IFF_ALLMULTI) {
2574                         rctl |= E1000_RCTL_MPE;
2575                         rctl &= ~E1000_RCTL_UPE;
2576                 } else {
2577                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2578                 }
2579                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2580                         rctl |= E1000_RCTL_VFE;
2581         }
2582
2583         ew32(RCTL, rctl);
2584
2585         if (netdev->mc_count) {
2586                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2587                 if (!mta_list)
2588                         return;
2589
2590                 /* prepare a packed array of only addresses. */
2591                 mc_ptr = netdev->mc_list;
2592
2593                 for (i = 0; i < netdev->mc_count; i++) {
2594                         if (!mc_ptr)
2595                                 break;
2596                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2597                                ETH_ALEN);
2598                         mc_ptr = mc_ptr->next;
2599                 }
2600
2601                 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2602                                           mac->rar_entry_count);
2603                 kfree(mta_list);
2604         } else {
2605                 /*
2606                  * if we're called from probe, we might not have
2607                  * anything to do here, so clear out the list
2608                  */
2609                 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2610         }
2611 }
2612
2613 /**
2614  * e1000_configure - configure the hardware for Rx and Tx
2615  * @adapter: private board structure
2616  **/
2617 static void e1000_configure(struct e1000_adapter *adapter)
2618 {
2619         e1000_set_multi(adapter->netdev);
2620
2621         e1000_restore_vlan(adapter);
2622         e1000_init_manageability(adapter);
2623
2624         e1000_configure_tx(adapter);
2625         e1000_setup_rctl(adapter);
2626         e1000_configure_rx(adapter);
2627         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2628 }
2629
2630 /**
2631  * e1000e_power_up_phy - restore link in case the phy was powered down
2632  * @adapter: address of board private structure
2633  *
2634  * The phy may be powered down to save power and turn off link when the
2635  * driver is unloaded and wake on lan is not enabled (among others)
2636  * *** this routine MUST be followed by a call to e1000e_reset ***
2637  **/
2638 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2639 {
2640         u16 mii_reg = 0;
2641
2642         /* Just clear the power down bit to wake the phy back up */
2643         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2644                 /*
2645                  * According to the manual, the phy will retain its
2646                  * settings across a power-down/up cycle
2647                  */
2648                 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2649                 mii_reg &= ~MII_CR_POWER_DOWN;
2650                 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2651         }
2652
2653         adapter->hw.mac.ops.setup_link(&adapter->hw);
2654 }
2655
2656 /**
2657  * e1000_power_down_phy - Power down the PHY
2658  *
2659  * Power down the PHY so no link is implied when interface is down
2660  * The PHY cannot be powered down is management or WoL is active
2661  */
2662 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2663 {
2664         struct e1000_hw *hw = &adapter->hw;
2665         u16 mii_reg;
2666
2667         /* WoL is enabled */
2668         if (adapter->wol)
2669                 return;
2670
2671         /* non-copper PHY? */
2672         if (adapter->hw.phy.media_type != e1000_media_type_copper)
2673                 return;
2674
2675         /* reset is blocked because of a SoL/IDER session */
2676         if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2677                 return;
2678
2679         /* manageability (AMT) is enabled */
2680         if (er32(MANC) & E1000_MANC_SMBUS_EN)
2681                 return;
2682
2683         /* power down the PHY */
2684         e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2685         mii_reg |= MII_CR_POWER_DOWN;
2686         e1e_wphy(hw, PHY_CONTROL, mii_reg);
2687         mdelay(1);
2688 }
2689
2690 /**
2691  * e1000e_reset - bring the hardware into a known good state
2692  *
2693  * This function boots the hardware and enables some settings that
2694  * require a configuration cycle of the hardware - those cannot be
2695  * set/changed during runtime. After reset the device needs to be
2696  * properly configured for Rx, Tx etc.
2697  */
2698 void e1000e_reset(struct e1000_adapter *adapter)
2699 {
2700         struct e1000_mac_info *mac = &adapter->hw.mac;
2701         struct e1000_fc_info *fc = &adapter->hw.fc;
2702         struct e1000_hw *hw = &adapter->hw;
2703         u32 tx_space, min_tx_space, min_rx_space;
2704         u32 pba = adapter->pba;
2705         u16 hwm;
2706
2707         /* reset Packet Buffer Allocation to default */
2708         ew32(PBA, pba);
2709
2710         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2711                 /*
2712                  * To maintain wire speed transmits, the Tx FIFO should be
2713                  * large enough to accommodate two full transmit packets,
2714                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2715                  * the Rx FIFO should be large enough to accommodate at least
2716                  * one full receive packet and is similarly rounded up and
2717                  * expressed in KB.
2718                  */
2719                 pba = er32(PBA);
2720                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2721                 tx_space = pba >> 16;
2722                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2723                 pba &= 0xffff;
2724                 /*
2725                  * the Tx fifo also stores 16 bytes of information about the tx
2726                  * but don't include ethernet FCS because hardware appends it
2727                  */
2728                 min_tx_space = (adapter->max_frame_size +
2729                                 sizeof(struct e1000_tx_desc) -
2730                                 ETH_FCS_LEN) * 2;
2731                 min_tx_space = ALIGN(min_tx_space, 1024);
2732                 min_tx_space >>= 10;
2733                 /* software strips receive CRC, so leave room for it */
2734                 min_rx_space = adapter->max_frame_size;
2735                 min_rx_space = ALIGN(min_rx_space, 1024);
2736                 min_rx_space >>= 10;
2737
2738                 /*
2739                  * If current Tx allocation is less than the min Tx FIFO size,
2740                  * and the min Tx FIFO size is less than the current Rx FIFO
2741                  * allocation, take space away from current Rx allocation
2742                  */
2743                 if ((tx_space < min_tx_space) &&
2744                     ((min_tx_space - tx_space) < pba)) {
2745                         pba -= min_tx_space - tx_space;
2746
2747                         /*
2748                          * if short on Rx space, Rx wins and must trump tx
2749                          * adjustment or use Early Receive if available
2750                          */
2751                         if ((pba < min_rx_space) &&
2752                             (!(adapter->flags & FLAG_HAS_ERT)))
2753                                 /* ERT enabled in e1000_configure_rx */
2754                                 pba = min_rx_space;
2755                 }
2756
2757                 ew32(PBA, pba);
2758         }
2759
2760
2761         /*
2762          * flow control settings
2763          *
2764          * The high water mark must be low enough to fit one full frame
2765          * (or the size used for early receive) above it in the Rx FIFO.
2766          * Set it to the lower of:
2767          * - 90% of the Rx FIFO size, and
2768          * - the full Rx FIFO size minus the early receive size (for parts
2769          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2770          * - the full Rx FIFO size minus one full frame
2771          */
2772         if (adapter->flags & FLAG_HAS_ERT)
2773                 hwm = min(((pba << 10) * 9 / 10),
2774                           ((pba << 10) - (E1000_ERT_2048 << 3)));
2775         else
2776                 hwm = min(((pba << 10) * 9 / 10),
2777                           ((pba << 10) - adapter->max_frame_size));
2778
2779         fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
2780         fc->low_water = fc->high_water - 8;
2781
2782         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2783                 fc->pause_time = 0xFFFF;
2784         else
2785                 fc->pause_time = E1000_FC_PAUSE_TIME;
2786         fc->send_xon = 1;
2787         fc->current_mode = fc->requested_mode;
2788
2789         /* Allow time for pending master requests to run */
2790         mac->ops.reset_hw(hw);
2791
2792         /*
2793          * For parts with AMT enabled, let the firmware know
2794          * that the network interface is in control
2795          */
2796         if (adapter->flags & FLAG_HAS_AMT)
2797                 e1000_get_hw_control(adapter);
2798
2799         ew32(WUC, 0);
2800
2801         if (mac->ops.init_hw(hw))
2802                 e_err("Hardware Error\n");
2803
2804         e1000_update_mng_vlan(adapter);
2805
2806         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2807         ew32(VET, ETH_P_8021Q);
2808
2809         e1000e_reset_adaptive(hw);
2810         e1000_get_phy_info(hw);
2811
2812         if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2813                 u16 phy_data = 0;
2814                 /*
2815                  * speed up time to link by disabling smart power down, ignore
2816                  * the return value of this function because there is nothing
2817                  * different we would do if it failed
2818                  */
2819                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2820                 phy_data &= ~IGP02E1000_PM_SPD;
2821                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2822         }
2823 }
2824
2825 int e1000e_up(struct e1000_adapter *adapter)
2826 {
2827         struct e1000_hw *hw = &adapter->hw;
2828
2829         /* hardware has been reset, we need to reload some things */
2830         e1000_configure(adapter);
2831
2832         clear_bit(__E1000_DOWN, &adapter->state);
2833
2834         napi_enable(&adapter->napi);
2835         if (adapter->msix_entries)
2836                 e1000_configure_msix(adapter);
2837         e1000_irq_enable(adapter);
2838
2839         /* fire a link change interrupt to start the watchdog */
2840         ew32(ICS, E1000_ICS_LSC);
2841         return 0;
2842 }
2843
2844 void e1000e_down(struct e1000_adapter *adapter)
2845 {
2846         struct net_device *netdev = adapter->netdev;
2847         struct e1000_hw *hw = &adapter->hw;
2848         u32 tctl, rctl;
2849
2850         /*
2851          * signal that we're down so the interrupt handler does not
2852          * reschedule our watchdog timer
2853          */
2854         set_bit(__E1000_DOWN, &adapter->state);
2855
2856         /* disable receives in the hardware */
2857         rctl = er32(RCTL);
2858         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2859         /* flush and sleep below */
2860
2861         netif_tx_stop_all_queues(netdev);
2862
2863         /* disable transmits in the hardware */
2864         tctl = er32(TCTL);
2865         tctl &= ~E1000_TCTL_EN;
2866         ew32(TCTL, tctl);
2867         /* flush both disables and wait for them to finish */
2868         e1e_flush();
2869         msleep(10);
2870
2871         napi_disable(&adapter->napi);
2872         e1000_irq_disable(adapter);
2873
2874         del_timer_sync(&adapter->watchdog_timer);
2875         del_timer_sync(&adapter->phy_info_timer);
2876
2877         netdev->tx_queue_len = adapter->tx_queue_len;
2878         netif_carrier_off(netdev);
2879         adapter->link_speed = 0;
2880         adapter->link_duplex = 0;
2881
2882         if (!pci_channel_offline(adapter->pdev))
2883                 e1000e_reset(adapter);
2884         e1000_clean_tx_ring(adapter);
2885         e1000_clean_rx_ring(adapter);
2886
2887         /*
2888          * TODO: for power management, we could drop the link and
2889          * pci_disable_device here.
2890          */
2891 }
2892
2893 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2894 {
2895         might_sleep();
2896         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2897                 msleep(1);
2898         e1000e_down(adapter);
2899         e1000e_up(adapter);
2900         clear_bit(__E1000_RESETTING, &adapter->state);
2901 }
2902
2903 /**
2904  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2905  * @adapter: board private structure to initialize
2906  *
2907  * e1000_sw_init initializes the Adapter private data structure.
2908  * Fields are initialized based on PCI device information and
2909  * OS network device settings (MTU size).
2910  **/
2911 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2912 {
2913         struct net_device *netdev = adapter->netdev;
2914
2915         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2916         adapter->rx_ps_bsize0 = 128;
2917         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2918         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2919
2920         e1000e_set_interrupt_capability(adapter);
2921
2922         if (e1000_alloc_queues(adapter))
2923                 return -ENOMEM;
2924
2925         spin_lock_init(&adapter->tx_queue_lock);
2926
2927         /* Explicitly disable IRQ since the NIC can be in any state. */
2928         e1000_irq_disable(adapter);
2929
2930         set_bit(__E1000_DOWN, &adapter->state);
2931         return 0;
2932 }
2933
2934 /**
2935  * e1000_intr_msi_test - Interrupt Handler
2936  * @irq: interrupt number
2937  * @data: pointer to a network interface device structure
2938  **/
2939 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2940 {
2941         struct net_device *netdev = data;
2942         struct e1000_adapter *adapter = netdev_priv(netdev);
2943         struct e1000_hw *hw = &adapter->hw;
2944         u32 icr = er32(ICR);
2945
2946         e_dbg("%s: icr is %08X\n", netdev->name, icr);
2947         if (icr & E1000_ICR_RXSEQ) {
2948                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2949                 wmb();
2950         }
2951
2952         return IRQ_HANDLED;
2953 }
2954
2955 /**
2956  * e1000_test_msi_interrupt - Returns 0 for successful test
2957  * @adapter: board private struct
2958  *
2959  * code flow taken from tg3.c
2960  **/
2961 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2962 {
2963         struct net_device *netdev = adapter->netdev;
2964         struct e1000_hw *hw = &adapter->hw;
2965         int err;
2966
2967         /* poll_enable hasn't been called yet, so don't need disable */
2968         /* clear any pending events */
2969         er32(ICR);
2970
2971         /* free the real vector and request a test handler */
2972         e1000_free_irq(adapter);
2973         e1000e_reset_interrupt_capability(adapter);
2974
2975         /* Assume that the test fails, if it succeeds then the test
2976          * MSI irq handler will unset this flag */
2977         adapter->flags |= FLAG_MSI_TEST_FAILED;
2978
2979         err = pci_enable_msi(adapter->pdev);
2980         if (err)
2981                 goto msi_test_failed;
2982
2983         err = request_irq(adapter->pdev->irq, &e1000_intr_msi_test, 0,
2984                           netdev->name, netdev);
2985         if (err) {
2986                 pci_disable_msi(adapter->pdev);
2987                 goto msi_test_failed;
2988         }
2989
2990         wmb();
2991
2992         e1000_irq_enable(adapter);
2993
2994         /* fire an unusual interrupt on the test handler */
2995         ew32(ICS, E1000_ICS_RXSEQ);
2996         e1e_flush();
2997         msleep(50);
2998
2999         e1000_irq_disable(adapter);
3000
3001         rmb();
3002
3003         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3004                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3005                 err = -EIO;
3006                 e_info("MSI interrupt test failed!\n");
3007         }
3008
3009         free_irq(adapter->pdev->irq, netdev);
3010         pci_disable_msi(adapter->pdev);
3011
3012         if (err == -EIO)
3013                 goto msi_test_failed;
3014
3015         /* okay so the test worked, restore settings */
3016         e_dbg("%s: MSI interrupt test succeeded!\n", netdev->name);
3017 msi_test_failed:
3018         e1000e_set_interrupt_capability(adapter);
3019         e1000_request_irq(adapter);
3020         return err;
3021 }
3022
3023 /**
3024  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3025  * @adapter: board private struct
3026  *
3027  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3028  **/
3029 static int e1000_test_msi(struct e1000_adapter *adapter)
3030 {
3031         int err;
3032         u16 pci_cmd;
3033
3034         if (!(adapter->flags & FLAG_MSI_ENABLED))
3035                 return 0;
3036
3037         /* disable SERR in case the MSI write causes a master abort */
3038         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3039         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3040                               pci_cmd & ~PCI_COMMAND_SERR);
3041
3042         err = e1000_test_msi_interrupt(adapter);
3043
3044         /* restore previous setting of command word */
3045         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3046
3047         /* success ! */
3048         if (!err)
3049                 return 0;
3050
3051         /* EIO means MSI test failed */
3052         if (err != -EIO)
3053                 return err;
3054
3055         /* back to INTx mode */
3056         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3057
3058         e1000_free_irq(adapter);
3059
3060         err = e1000_request_irq(adapter);
3061
3062         return err;
3063 }
3064
3065 /**
3066  * e1000_open - Called when a network interface is made active
3067  * @netdev: network interface device structure
3068  *
3069  * Returns 0 on success, negative value on failure
3070  *
3071  * The open entry point is called when a network interface is made
3072  * active by the system (IFF_UP).  At this point all resources needed
3073  * for transmit and receive operations are allocated, the interrupt
3074  * handler is registered with the OS, the watchdog timer is started,
3075  * and the stack is notified that the interface is ready.
3076  **/
3077 static int e1000_open(struct net_device *netdev)
3078 {
3079         struct e1000_adapter *adapter = netdev_priv(netdev);
3080         struct e1000_hw *hw = &adapter->hw;
3081         int err;
3082
3083         /* disallow open during test */
3084         if (test_bit(__E1000_TESTING, &adapter->state))
3085                 return -EBUSY;
3086
3087         /* allocate transmit descriptors */
3088         err = e1000e_setup_tx_resources(adapter);
3089         if (err)
3090                 goto err_setup_tx;
3091
3092         /* allocate receive descriptors */
3093         err = e1000e_setup_rx_resources(adapter);
3094         if (err)
3095                 goto err_setup_rx;
3096
3097         e1000e_power_up_phy(adapter);
3098
3099         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3100         if ((adapter->hw.mng_cookie.status &
3101              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3102                 e1000_update_mng_vlan(adapter);
3103
3104         /*
3105          * If AMT is enabled, let the firmware know that the network
3106          * interface is now open
3107          */
3108         if (adapter->flags & FLAG_HAS_AMT)
3109                 e1000_get_hw_control(adapter);
3110
3111         /*
3112          * before we allocate an interrupt, we must be ready to handle it.
3113          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3114          * as soon as we call pci_request_irq, so we have to setup our
3115          * clean_rx handler before we do so.
3116          */
3117         e1000_configure(adapter);
3118
3119         err = e1000_request_irq(adapter);
3120         if (err)
3121                 goto err_req_irq;
3122
3123         /*
3124          * Work around PCIe errata with MSI interrupts causing some chipsets to
3125          * ignore e1000e MSI messages, which means we need to test our MSI
3126          * interrupt now
3127          */
3128         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3129                 err = e1000_test_msi(adapter);
3130                 if (err) {
3131                         e_err("Interrupt allocation failed\n");
3132                         goto err_req_irq;
3133                 }
3134         }
3135
3136         /* From here on the code is the same as e1000e_up() */
3137         clear_bit(__E1000_DOWN, &adapter->state);
3138
3139         napi_enable(&adapter->napi);
3140
3141         e1000_irq_enable(adapter);
3142
3143         netif_tx_start_all_queues(netdev);
3144
3145         /* fire a link status change interrupt to start the watchdog */
3146         ew32(ICS, E1000_ICS_LSC);
3147
3148         return 0;
3149
3150 err_req_irq:
3151         e1000_release_hw_control(adapter);
3152         e1000_power_down_phy(adapter);
3153         e1000e_free_rx_resources(adapter);
3154 err_setup_rx:
3155         e1000e_free_tx_resources(adapter);
3156 err_setup_tx:
3157         e1000e_reset(adapter);
3158
3159         return err;
3160 }
3161
3162 /**
3163  * e1000_close - Disables a network interface
3164  * @netdev: network interface device structure
3165  *
3166  * Returns 0, this is not allowed to fail
3167  *
3168  * The close entry point is called when an interface is de-activated
3169  * by the OS.  The hardware is still under the drivers control, but
3170  * needs to be disabled.  A global MAC reset is issued to stop the
3171  * hardware, and all transmit and receive resources are freed.
3172  **/
3173 static int e1000_close(struct net_device *netdev)
3174 {
3175         struct e1000_adapter *adapter = netdev_priv(netdev);
3176
3177         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3178         e1000e_down(adapter);
3179         e1000_power_down_phy(adapter);
3180         e1000_free_irq(adapter);
3181
3182         e1000e_free_tx_resources(adapter);
3183         e1000e_free_rx_resources(adapter);
3184
3185         /*
3186          * kill manageability vlan ID if supported, but not if a vlan with
3187          * the same ID is registered on the host OS (let 8021q kill it)
3188          */
3189         if ((adapter->hw.mng_cookie.status &
3190                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3191              !(adapter->vlgrp &&
3192                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3193                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3194
3195         /*
3196          * If AMT is enabled, let the firmware know that the network
3197          * interface is now closed
3198          */
3199         if (adapter->flags & FLAG_HAS_AMT)
3200                 e1000_release_hw_control(adapter);
3201
3202         return 0;
3203 }
3204 /**
3205  * e1000_set_mac - Change the Ethernet Address of the NIC
3206  * @netdev: network interface device structure
3207  * @p: pointer to an address structure
3208  *
3209  * Returns 0 on success, negative on failure
3210  **/
3211 static int e1000_set_mac(struct net_device *netdev, void *p)
3212 {
3213         struct e1000_adapter *adapter = netdev_priv(netdev);
3214         struct sockaddr *addr = p;
3215
3216         if (!is_valid_ether_addr(addr->sa_data))
3217                 return -EADDRNOTAVAIL;
3218
3219         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3220         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3221
3222         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3223
3224         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3225                 /* activate the work around */
3226                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3227
3228                 /*
3229                  * Hold a copy of the LAA in RAR[14] This is done so that
3230                  * between the time RAR[0] gets clobbered  and the time it
3231                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3232                  * of the RARs and no incoming packets directed to this port
3233                  * are dropped. Eventually the LAA will be in RAR[0] and
3234                  * RAR[14]
3235                  */
3236                 e1000e_rar_set(&adapter->hw,
3237                               adapter->hw.mac.addr,
3238                               adapter->hw.mac.rar_entry_count - 1);
3239         }
3240
3241         return 0;
3242 }
3243
3244 /**
3245  * e1000e_update_phy_task - work thread to update phy
3246  * @work: pointer to our work struct
3247  *
3248  * this worker thread exists because we must acquire a
3249  * semaphore to read the phy, which we could msleep while
3250  * waiting for it, and we can't msleep in a timer.
3251  **/
3252 static void e1000e_update_phy_task(struct work_struct *work)
3253 {
3254         struct e1000_adapter *adapter = container_of(work,
3255                                         struct e1000_adapter, update_phy_task);
3256         e1000_get_phy_info(&adapter->hw);
3257 }
3258
3259 /*
3260  * Need to wait a few seconds after link up to get diagnostic information from
3261  * the phy
3262  */
3263 static void e1000_update_phy_info(unsigned long data)
3264 {
3265         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3266         schedule_work(&adapter->update_phy_task);
3267 }
3268
3269 /**
3270  * e1000e_update_stats - Update the board statistics counters
3271  * @adapter: board private structure
3272  **/
3273 void e1000e_update_stats(struct e1000_adapter *adapter)
3274 {
3275         struct e1000_hw *hw = &adapter->hw;
3276         struct pci_dev *pdev = adapter->pdev;
3277
3278         /*
3279          * Prevent stats update while adapter is being reset, or if the pci
3280          * connection is down.
3281          */
3282         if (adapter->link_speed == 0)
3283                 return;
3284         if (pci_channel_offline(pdev))
3285                 return;
3286
3287         adapter->stats.crcerrs += er32(CRCERRS);
3288         adapter->stats.gprc += er32(GPRC);
3289         adapter->stats.gorc += er32(GORCL);
3290         er32(GORCH); /* Clear gorc */
3291         adapter->stats.bprc += er32(BPRC);
3292         adapter->stats.mprc += er32(MPRC);
3293         adapter->stats.roc += er32(ROC);
3294
3295         adapter->stats.mpc += er32(MPC);
3296         adapter->stats.scc += er32(SCC);
3297         adapter->stats.ecol += er32(ECOL);
3298         adapter->stats.mcc += er32(MCC);
3299         adapter->stats.latecol += er32(LATECOL);
3300         adapter->stats.dc += er32(DC);
3301         adapter->stats.xonrxc += er32(XONRXC);
3302         adapter->stats.xontxc += er32(XONTXC);
3303         adapter->stats.xoffrxc += er32(XOFFRXC);
3304         adapter->stats.xofftxc += er32(XOFFTXC);
3305         adapter->stats.gptc += er32(GPTC);
3306         adapter->stats.gotc += er32(GOTCL);
3307         er32(GOTCH); /* Clear gotc */
3308         adapter->stats.rnbc += er32(RNBC);
3309         adapter->stats.ruc += er32(RUC);
3310
3311         adapter->stats.mptc += er32(MPTC);
3312         adapter->stats.bptc += er32(BPTC);
3313
3314         /* used for adaptive IFS */
3315
3316         hw->mac.tx_packet_delta = er32(TPT);
3317         adapter->stats.tpt += hw->mac.tx_packet_delta;
3318         hw->mac.collision_delta = er32(COLC);
3319         adapter->stats.colc += hw->mac.collision_delta;
3320
3321         adapter->stats.algnerrc += er32(ALGNERRC);
3322         adapter->stats.rxerrc += er32(RXERRC);
3323         if (hw->mac.type != e1000_82574)
3324                 adapter->stats.tncrs += er32(TNCRS);
3325         adapter->stats.cexterr += er32(CEXTERR);
3326         adapter->stats.tsctc += er32(TSCTC);
3327         adapter->stats.tsctfc += er32(TSCTFC);
3328
3329         /* Fill out the OS statistics structure */
3330         adapter->net_stats.multicast = adapter->stats.mprc;
3331         adapter->net_stats.collisions = adapter->stats.colc;
3332
3333         /* Rx Errors */
3334
3335         /*
3336          * RLEC on some newer hardware can be incorrect so build
3337          * our own version based on RUC and ROC
3338          */
3339         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3340                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3341                 adapter->stats.ruc + adapter->stats.roc +
3342                 adapter->stats.cexterr;
3343         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3344                                               adapter->stats.roc;
3345         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3346         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3347         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3348
3349         /* Tx Errors */
3350         adapter->net_stats.tx_errors = adapter->stats.ecol +
3351                                        adapter->stats.latecol;
3352         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3353         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3354         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3355
3356         /* Tx Dropped needs to be maintained elsewhere */
3357
3358         /* Management Stats */
3359         adapter->stats.mgptc += er32(MGTPTC);
3360         adapter->stats.mgprc += er32(MGTPRC);
3361         adapter->stats.mgpdc += er32(MGTPDC);
3362 }
3363
3364 /**
3365  * e1000_phy_read_status - Update the PHY register status snapshot
3366  * @adapter: board private structure
3367  **/
3368 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3369 {
3370         struct e1000_hw *hw = &adapter->hw;
3371         struct e1000_phy_regs *phy = &adapter->phy_regs;
3372         int ret_val;
3373
3374         if ((er32(STATUS) & E1000_STATUS_LU) &&
3375             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3376                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3377                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3378                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3379                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3380                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3381                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3382                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3383                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3384                 if (ret_val)
3385                         e_warn("Error reading PHY register\n");
3386         } else {
3387                 /*
3388                  * Do not read PHY registers if link is not up
3389                  * Set values to typical power-on defaults
3390                  */
3391                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3392                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3393                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3394                              BMSR_ERCAP);
3395                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3396                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3397                 phy->lpa = 0;
3398                 phy->expansion = EXPANSION_ENABLENPAGE;
3399                 phy->ctrl1000 = ADVERTISE_1000FULL;
3400                 phy->stat1000 = 0;
3401                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3402         }
3403 }
3404
3405 static void e1000_print_link_info(struct e1000_adapter *adapter)
3406 {
3407         struct e1000_hw *hw = &adapter->hw;
3408         u32 ctrl = er32(CTRL);
3409
3410         /* Link status message must follow this format for user tools */
3411         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3412                "Flow Control: %s\n",
3413                adapter->netdev->name,
3414                adapter->link_speed,
3415                (adapter->link_duplex == FULL_DUPLEX) ?
3416                                 "Full Duplex" : "Half Duplex",
3417                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3418                                 "RX/TX" :
3419                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3420                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3421 }
3422
3423 bool e1000_has_link(struct e1000_adapter *adapter)
3424 {
3425         struct e1000_hw *hw = &adapter->hw;
3426         bool link_active = 0;
3427         s32 ret_val = 0;
3428
3429         /*
3430          * get_link_status is set on LSC (link status) interrupt or
3431          * Rx sequence error interrupt.  get_link_status will stay
3432          * false until the check_for_link establishes link
3433          * for copper adapters ONLY
3434          */
3435         switch (hw->phy.media_type) {
3436         case e1000_media_type_copper:
3437                 if (hw->mac.get_link_status) {
3438                         ret_val = hw->mac.ops.check_for_link(hw);
3439                         link_active = !hw->mac.get_link_status;
3440                 } else {
3441                         link_active = 1;
3442                 }
3443                 break;
3444         case e1000_media_type_fiber:
3445                 ret_val = hw->mac.ops.check_for_link(hw);
3446                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3447                 break;
3448         case e1000_media_type_internal_serdes:
3449                 ret_val = hw->mac.ops.check_for_link(hw);
3450                 link_active = adapter->hw.mac.serdes_has_link;
3451                 break;
3452         default:
3453         case e1000_media_type_unknown:
3454                 break;
3455         }
3456
3457         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3458             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3459                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3460                 e_info("Gigabit has been disabled, downgrading speed\n");
3461         }
3462
3463         return link_active;
3464 }
3465
3466 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3467 {
3468         /* make sure the receive unit is started */
3469         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3470             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3471                 struct e1000_hw *hw = &adapter->hw;
3472                 u32 rctl = er32(RCTL);
3473                 ew32(RCTL, rctl | E1000_RCTL_EN);
3474                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3475         }
3476 }
3477
3478 /**
3479  * e1000_watchdog - Timer Call-back
3480  * @data: pointer to adapter cast into an unsigned long
3481  **/
3482 static void e1000_watchdog(unsigned long data)
3483 {
3484         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3485
3486         /* Do the rest outside of interrupt context */
3487         schedule_work(&adapter->watchdog_task);
3488
3489         /* TODO: make this use queue_delayed_work() */
3490 }
3491
3492 static void e1000_watchdog_task(struct work_struct *work)
3493 {
3494         struct e1000_adapter *adapter = container_of(work,
3495                                         struct e1000_adapter, watchdog_task);
3496         struct net_device *netdev = adapter->netdev;
3497         struct e1000_mac_info *mac = &adapter->hw.mac;
3498         struct e1000_phy_info *phy = &adapter->hw.phy;
3499         struct e1000_ring *tx_ring = adapter->tx_ring;
3500         struct e1000_hw *hw = &adapter->hw;
3501         u32 link, tctl;
3502         int tx_pending = 0;
3503
3504         link = e1000_has_link(adapter);
3505         if ((netif_carrier_ok(netdev)) && link) {
3506                 e1000e_enable_receives(adapter);
3507                 goto link_up;
3508         }
3509
3510         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3511             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3512                 e1000_update_mng_vlan(adapter);
3513
3514         if (link) {
3515                 if (!netif_carrier_ok(netdev)) {
3516                         bool txb2b = 1;
3517                         /* update snapshot of PHY registers on LSC */
3518                         e1000_phy_read_status(adapter);
3519                         mac->ops.get_link_up_info(&adapter->hw,
3520                                                    &adapter->link_speed,
3521                                                    &adapter->link_duplex);
3522                         e1000_print_link_info(adapter);
3523                         /*
3524                          * On supported PHYs, check for duplex mismatch only
3525                          * if link has autonegotiated at 10/100 half
3526                          */
3527                         if ((hw->phy.type == e1000_phy_igp_3 ||
3528                              hw->phy.type == e1000_phy_bm) &&
3529                             (hw->mac.autoneg == true) &&
3530                             (adapter->link_speed == SPEED_10 ||
3531                              adapter->link_speed == SPEED_100) &&
3532                             (adapter->link_duplex == HALF_DUPLEX)) {
3533                                 u16 autoneg_exp;
3534
3535                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3536
3537                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3538                                         e_info("Autonegotiated half duplex but"
3539                                                " link partner cannot autoneg. "
3540                                                " Try forcing full duplex if "
3541                                                "link gets many collisions.\n");
3542                         }
3543
3544                         /*
3545                          * tweak tx_queue_len according to speed/duplex
3546                          * and adjust the timeout factor
3547                          */
3548                         netdev->tx_queue_len = adapter->tx_queue_len;
3549                         adapter->tx_timeout_factor = 1;
3550                         switch (adapter->link_speed) {
3551                         case SPEED_10:
3552                                 txb2b = 0;
3553                                 netdev->tx_queue_len = 10;
3554                                 adapter->tx_timeout_factor = 16;
3555                                 break;
3556                         case SPEED_100:
3557                                 txb2b = 0;
3558                                 netdev->tx_queue_len = 100;
3559                                 /* maybe add some timeout factor ? */
3560                                 break;
3561                         }
3562
3563                         /*
3564                          * workaround: re-program speed mode bit after
3565                          * link-up event
3566                          */
3567                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3568                             !txb2b) {
3569                                 u32 tarc0;
3570                                 tarc0 = er32(TARC(0));
3571                                 tarc0 &= ~SPEED_MODE_BIT;
3572                                 ew32(TARC(0), tarc0);
3573                         }
3574
3575                         /*
3576                          * disable TSO for pcie and 10/100 speeds, to avoid
3577                          * some hardware issues
3578                          */
3579                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3580                                 switch (adapter->link_speed) {
3581                                 case SPEED_10:
3582                                 case SPEED_100:
3583                                         e_info("10/100 speed: disabling TSO\n");
3584                                         netdev->features &= ~NETIF_F_TSO;
3585                                         netdev->features &= ~NETIF_F_TSO6;
3586                                         break;
3587                                 case SPEED_1000:
3588                                         netdev->features |= NETIF_F_TSO;
3589                                         netdev->features |= NETIF_F_TSO6;
3590                                         break;
3591                                 default:
3592                                         /* oops */
3593                                         break;
3594                                 }
3595                         }
3596
3597                         /*
3598                          * enable transmits in the hardware, need to do this
3599                          * after setting TARC(0)
3600                          */
3601                         tctl = er32(TCTL);
3602                         tctl |= E1000_TCTL_EN;
3603                         ew32(TCTL, tctl);
3604
3605                         /*
3606                          * Perform any post-link-up configuration before
3607                          * reporting link up.
3608                          */
3609                         if (phy->ops.cfg_on_link_up)
3610                                 phy->ops.cfg_on_link_up(hw);
3611
3612                         netif_carrier_on(netdev);
3613                         netif_tx_wake_all_queues(netdev);
3614
3615                         if (!test_bit(__E1000_DOWN, &adapter->state))
3616                                 mod_timer(&adapter->phy_info_timer,
3617                                           round_jiffies(jiffies + 2 * HZ));
3618                 }
3619         } else {
3620                 if (netif_carrier_ok(netdev)) {
3621                         adapter->link_speed = 0;
3622                         adapter->link_duplex = 0;
3623                         /* Link status message must follow this format */
3624                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
3625                                adapter->netdev->name);
3626                         netif_carrier_off(netdev);
3627                         netif_tx_stop_all_queues(netdev);
3628                         if (!test_bit(__E1000_DOWN, &adapter->state))
3629                                 mod_timer(&adapter->phy_info_timer,
3630                                           round_jiffies(jiffies + 2 * HZ));
3631
3632                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3633                                 schedule_work(&adapter->reset_task);
3634                 }
3635         }
3636
3637 link_up:
3638         e1000e_update_stats(adapter);
3639
3640         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3641         adapter->tpt_old = adapter->stats.tpt;
3642         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3643         adapter->colc_old = adapter->stats.colc;
3644
3645         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3646         adapter->gorc_old = adapter->stats.gorc;
3647         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3648         adapter->gotc_old = adapter->stats.gotc;
3649
3650         e1000e_update_adaptive(&adapter->hw);
3651
3652         if (!netif_carrier_ok(netdev)) {
3653                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3654                                tx_ring->count);
3655                 if (tx_pending) {
3656                         /*
3657                          * We've lost link, so the controller stops DMA,
3658                          * but we've got queued Tx work that's never going
3659                          * to get done, so reset controller to flush Tx.
3660                          * (Do the reset outside of interrupt context).
3661                          */
3662                         adapter->tx_timeout_count++;
3663                         schedule_work(&adapter->reset_task);
3664                 }
3665         }
3666
3667         /* Cause software interrupt to ensure Rx ring is cleaned */
3668         if (adapter->msix_entries)
3669                 ew32(ICS, adapter->rx_ring->ims_val);
3670         else
3671                 ew32(ICS, E1000_ICS_RXDMT0);
3672
3673         /* Force detection of hung controller every watchdog period */
3674         adapter->detect_tx_hung = 1;
3675
3676         /*
3677          * With 82571 controllers, LAA may be overwritten due to controller
3678          * reset from the other port. Set the appropriate LAA in RAR[0]
3679          */
3680         if (e1000e_get_laa_state_82571(hw))
3681                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3682
3683         /* Reset the timer */
3684         if (!test_bit(__E1000_DOWN, &adapter->state))
3685                 mod_timer(&adapter->watchdog_timer,
3686                           round_jiffies(jiffies + 2 * HZ));
3687 }
3688
3689 #define E1000_TX_FLAGS_CSUM             0x00000001
3690 #define E1000_TX_FLAGS_VLAN             0x00000002
3691 #define E1000_TX_FLAGS_TSO              0x00000004
3692 #define E1000_TX_FLAGS_IPV4             0x00000008
3693 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3694 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3695
3696 static int e1000_tso(struct e1000_adapter *adapter,
3697                      struct sk_buff *skb)
3698 {
3699         struct e1000_ring *tx_ring = adapter->tx_ring;
3700         struct e1000_context_desc *context_desc;
3701         struct e1000_buffer *buffer_info;
3702         unsigned int i;
3703         u32 cmd_length = 0;
3704         u16 ipcse = 0, tucse, mss;
3705         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3706         int err;
3707
3708         if (skb_is_gso(skb)) {
3709                 if (skb_header_cloned(skb)) {
3710                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3711                         if (err)
3712                                 return err;
3713                 }
3714
3715                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3716                 mss = skb_shinfo(skb)->gso_size;
3717                 if (skb->protocol == htons(ETH_P_IP)) {
3718                         struct iphdr *iph = ip_hdr(skb);
3719                         iph->tot_len = 0;
3720                         iph->check = 0;
3721                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3722                                                                  iph->daddr, 0,
3723                                                                  IPPROTO_TCP,
3724                                                                  0);
3725                         cmd_length = E1000_TXD_CMD_IP;
3726                         ipcse = skb_transport_offset(skb) - 1;
3727                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3728                         ipv6_hdr(skb)->payload_len = 0;
3729                         tcp_hdr(skb)->check =
3730                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3731                                                  &ipv6_hdr(skb)->daddr,
3732                                                  0, IPPROTO_TCP, 0);
3733                         ipcse = 0;
3734                 }
3735                 ipcss = skb_network_offset(skb);
3736                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3737                 tucss = skb_transport_offset(skb);
3738                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3739                 tucse = 0;
3740
3741                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3742                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3743
3744                 i = tx_ring->next_to_use;
3745                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3746                 buffer_info = &tx_ring->buffer_info[i];
3747
3748                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3749                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3750                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3751                 context_desc->upper_setup.tcp_fields.tucss = tucss;
3752                 context_desc->upper_setup.tcp_fields.tucso = tucso;
3753                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3754                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3755                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3756                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3757
3758                 buffer_info->time_stamp = jiffies;
3759                 buffer_info->next_to_watch = i;
3760
3761                 i++;
3762                 if (i == tx_ring->count)
3763                         i = 0;
3764                 tx_ring->next_to_use = i;
3765
3766                 return 1;
3767         }
3768
3769         return 0;
3770 }
3771
3772 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3773 {
3774         struct e1000_ring *tx_ring = adapter->tx_ring;
3775         struct e1000_context_desc *context_desc;
3776         struct e1000_buffer *buffer_info;
3777         unsigned int i;
3778         u8 css;
3779         u32 cmd_len = E1000_TXD_CMD_DEXT;
3780
3781         if (skb->ip_summed != CHECKSUM_PARTIAL)
3782                 return 0;
3783
3784         switch (skb->protocol) {
3785         case __constant_htons(ETH_P_IP):
3786                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3787                         cmd_len |= E1000_TXD_CMD_TCP;
3788                 break;
3789         case __constant_htons(ETH_P_IPV6):
3790                 /* XXX not handling all IPV6 headers */
3791                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3792                         cmd_len |= E1000_TXD_CMD_TCP;
3793                 break;
3794         default:
3795                 if (unlikely(net_ratelimit()))
3796                         e_warn("checksum_partial proto=%x!\n", skb->protocol);
3797                 break;
3798         }
3799
3800         css = skb_transport_offset(skb);
3801
3802         i = tx_ring->next_to_use;
3803         buffer_info = &tx_ring->buffer_info[i];
3804         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3805
3806         context_desc->lower_setup.ip_config = 0;
3807         context_desc->upper_setup.tcp_fields.tucss = css;
3808         context_desc->upper_setup.tcp_fields.tucso =
3809                                 css + skb->csum_offset;
3810         context_desc->upper_setup.tcp_fields.tucse = 0;
3811         context_desc->tcp_seg_setup.data = 0;
3812         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
3813
3814         buffer_info->time_stamp = jiffies;
3815         buffer_info->next_to_watch = i;
3816
3817         i++;
3818         if (i == tx_ring->count)
3819                 i = 0;
3820         tx_ring->next_to_use = i;
3821
3822         return 1;
3823 }
3824
3825 #define E1000_MAX_PER_TXD       8192
3826 #define E1000_MAX_TXD_PWR       12
3827
3828 static int e1000_tx_map(struct e1000_adapter *adapter,
3829                         struct sk_buff *skb, unsigned int first,
3830                         unsigned int max_per_txd, unsigned int nr_frags,
3831                         unsigned int mss)
3832 {
3833         struct e1000_ring *tx_ring = adapter->tx_ring;
3834         struct e1000_buffer *buffer_info;
3835         unsigned int len = skb->len - skb->data_len;
3836         unsigned int offset = 0, size, count = 0, i;
3837         unsigned int f;
3838
3839         i = tx_ring->next_to_use;
3840
3841         while (len) {
3842                 buffer_info = &tx_ring->buffer_info[i];
3843                 size = min(len, max_per_txd);
3844
3845                 /* Workaround for premature desc write-backs
3846                  * in TSO mode.  Append 4-byte sentinel desc */
3847                 if (mss && !nr_frags && size == len && size > 8)
3848                         size -= 4;
3849
3850                 buffer_info->length = size;
3851                 /* set time_stamp *before* dma to help avoid a possible race */
3852                 buffer_info->time_stamp = jiffies;
3853                 buffer_info->dma =
3854                         pci_map_single(adapter->pdev,
3855                                 skb->data + offset,
3856                                 size,
3857                                 PCI_DMA_TODEVICE);
3858                 if (pci_dma_mapping_error(adapter->pdev, buffer_info->dma)) {
3859                         dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3860                         adapter->tx_dma_failed++;
3861                         return -1;
3862                 }
3863                 buffer_info->next_to_watch = i;
3864
3865                 len -= size;
3866                 offset += size;
3867                 count++;
3868                 i++;
3869                 if (i == tx_ring->count)
3870                         i = 0;
3871         }
3872
3873         for (f = 0; f < nr_frags; f++) {
3874                 struct skb_frag_struct *frag;
3875
3876                 frag = &skb_shinfo(skb)->frags[f];
3877                 len = frag->size;
3878                 offset = frag->page_offset;
3879
3880                 while (len) {
3881                         buffer_info = &tx_ring->buffer_info[i];
3882                         size = min(len, max_per_txd);
3883                         /* Workaround for premature desc write-backs
3884                          * in TSO mode.  Append 4-byte sentinel desc */
3885                         if (mss && f == (nr_frags-1) && size == len && size > 8)
3886                                 size -= 4;
3887
3888                         buffer_info->length = size;
3889                         buffer_info->time_stamp = jiffies;
3890                         buffer_info->dma =
3891                                 pci_map_page(adapter->pdev,
3892                                         frag->page,
3893                                         offset,
3894                                         size,
3895                                         PCI_DMA_TODEVICE);
3896                         if (pci_dma_mapping_error(adapter->pdev,
3897                                                   buffer_info->dma)) {
3898                                 dev_err(&adapter->pdev->dev,
3899                                         "TX DMA page map failed\n");
3900                                 adapter->tx_dma_failed++;
3901                                 return -1;
3902                         }
3903
3904                         buffer_info->next_to_watch = i;
3905
3906                         len -= size;
3907                         offset += size;
3908                         count++;
3909
3910                         i++;
3911                         if (i == tx_ring->count)
3912                                 i = 0;
3913                 }
3914         }
3915
3916         if (i == 0)
3917                 i = tx_ring->count - 1;
3918         else
3919                 i--;
3920
3921         tx_ring->buffer_info[i].skb = skb;
3922         tx_ring->buffer_info[first].next_to_watch = i;
3923
3924         return count;
3925 }
3926
3927 static void e1000_tx_queue(struct e1000_adapter *adapter,
3928                            int tx_flags, int count)
3929 {
3930         struct e1000_ring *tx_ring = adapter->tx_ring;
3931         struct e1000_tx_desc *tx_desc = NULL;
3932         struct e1000_buffer *buffer_info;
3933         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3934         unsigned int i;
3935
3936         if (tx_flags & E1000_TX_FLAGS_TSO) {
3937                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3938                              E1000_TXD_CMD_TSE;
3939                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3940
3941                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3942                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3943         }
3944
3945         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3946                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3947                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3948         }
3949
3950         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3951                 txd_lower |= E1000_TXD_CMD_VLE;
3952                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3953         }
3954
3955         i = tx_ring->next_to_use;
3956
3957         while (count--) {
3958                 buffer_info = &tx_ring->buffer_info[i];
3959                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3960                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3961                 tx_desc->lower.data =
3962                         cpu_to_le32(txd_lower | buffer_info->length);
3963                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3964
3965                 i++;
3966                 if (i == tx_ring->count)
3967                         i = 0;
3968         }
3969
3970         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3971
3972         /*
3973          * Force memory writes to complete before letting h/w
3974          * know there are new descriptors to fetch.  (Only
3975          * applicable for weak-ordered memory model archs,
3976          * such as IA-64).
3977          */
3978         wmb();
3979
3980         tx_ring->next_to_use = i;
3981         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3982         /*
3983          * we need this if more than one processor can write to our tail
3984          * at a time, it synchronizes IO on IA64/Altix systems
3985          */
3986         mmiowb();
3987 }
3988
3989 #define MINIMUM_DHCP_PACKET_SIZE 282
3990 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3991                                     struct sk_buff *skb)
3992 {
3993         struct e1000_hw *hw =  &adapter->hw;
3994         u16 length, offset;
3995
3996         if (vlan_tx_tag_present(skb)) {
3997                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3998                     && (adapter->hw.mng_cookie.status &
3999                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4000                         return 0;
4001         }
4002
4003         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4004                 return 0;
4005
4006         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4007                 return 0;
4008
4009         {
4010                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4011                 struct udphdr *udp;
4012
4013                 if (ip->protocol != IPPROTO_UDP)
4014                         return 0;
4015
4016                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4017                 if (ntohs(udp->dest) != 67)
4018                         return 0;
4019
4020                 offset = (u8 *)udp + 8 - skb->data;
4021                 length = skb->len - offset;
4022                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4023         }
4024
4025         return 0;
4026 }
4027
4028 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4029 {
4030         struct e1000_adapter *adapter = netdev_priv(netdev);
4031
4032         netif_stop_queue(netdev);
4033         /*
4034          * Herbert's original patch had:
4035          *  smp_mb__after_netif_stop_queue();
4036          * but since that doesn't exist yet, just open code it.
4037          */
4038         smp_mb();
4039
4040         /*
4041          * We need to check again in a case another CPU has just
4042          * made room available.
4043          */
4044         if (e1000_desc_unused(adapter->tx_ring) < size)
4045                 return -EBUSY;
4046
4047         /* A reprieve! */
4048         netif_start_queue(netdev);
4049         ++adapter->restart_queue;
4050         return 0;
4051 }
4052
4053 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4054 {
4055         struct e1000_adapter *adapter = netdev_priv(netdev);
4056
4057         if (e1000_desc_unused(adapter->tx_ring) >= size)
4058                 return 0;
4059         return __e1000_maybe_stop_tx(netdev, size);
4060 }
4061
4062 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4063 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4064 {
4065         struct e1000_adapter *adapter = netdev_priv(netdev);
4066         struct e1000_ring *tx_ring = adapter->tx_ring;
4067         unsigned int first;
4068         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4069         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4070         unsigned int tx_flags = 0;
4071         unsigned int len = skb->len - skb->data_len;
4072         unsigned long irq_flags;
4073         unsigned int nr_frags;
4074         unsigned int mss;
4075         int count = 0;
4076         int tso;
4077         unsigned int f;
4078
4079         if (test_bit(__E1000_DOWN, &adapter->state)) {
4080                 dev_kfree_skb_any(skb);
4081                 return NETDEV_TX_OK;
4082         }
4083
4084         if (skb->len <= 0) {
4085                 dev_kfree_skb_any(skb);
4086                 return NETDEV_TX_OK;
4087         }
4088
4089         mss = skb_shinfo(skb)->gso_size;
4090         /*
4091          * The controller does a simple calculation to
4092          * make sure there is enough room in the FIFO before
4093          * initiating the DMA for each buffer.  The calc is:
4094          * 4 = ceil(buffer len/mss).  To make sure we don't
4095          * overrun the FIFO, adjust the max buffer len if mss
4096          * drops.
4097          */
4098         if (mss) {
4099                 u8 hdr_len;
4100                 max_per_txd = min(mss << 2, max_per_txd);
4101                 max_txd_pwr = fls(max_per_txd) - 1;
4102
4103                 /*
4104                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4105                  * points to just header, pull a few bytes of payload from
4106                  * frags into skb->data
4107                  */
4108                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4109                 /*
4110                  * we do this workaround for ES2LAN, but it is un-necessary,
4111                  * avoiding it could save a lot of cycles
4112                  */
4113                 if (skb->data_len && (hdr_len == len)) {
4114                         unsigned int pull_size;
4115
4116                         pull_size = min((unsigned int)4, skb->data_len);
4117                         if (!__pskb_pull_tail(skb, pull_size)) {
4118                                 e_err("__pskb_pull_tail failed.\n");
4119                                 dev_kfree_skb_any(skb);
4120                                 return NETDEV_TX_OK;
4121                         }
4122                         len = skb->len - skb->data_len;
4123                 }
4124         }
4125
4126         /* reserve a descriptor for the offload context */
4127         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4128                 count++;
4129         count++;
4130
4131         count += TXD_USE_COUNT(len, max_txd_pwr);
4132
4133         nr_frags = skb_shinfo(skb)->nr_frags;
4134         for (f = 0; f < nr_frags; f++)
4135                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4136                                        max_txd_pwr);
4137
4138         if (adapter->hw.mac.tx_pkt_filtering)
4139                 e1000_transfer_dhcp_info(adapter, skb);
4140
4141         if (!spin_trylock_irqsave(&adapter->tx_queue_lock, irq_flags))
4142                 /* Collision - tell upper layer to requeue */
4143                 return NETDEV_TX_LOCKED;
4144
4145         /*
4146          * need: count + 2 desc gap to keep tail from touching
4147          * head, otherwise try next time
4148          */
4149         if (e1000_maybe_stop_tx(netdev, count + 2)) {
4150                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
4151                 return NETDEV_TX_BUSY;
4152         }
4153
4154         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4155                 tx_flags |= E1000_TX_FLAGS_VLAN;
4156                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4157         }
4158
4159         first = tx_ring->next_to_use;
4160
4161         tso = e1000_tso(adapter, skb);
4162         if (tso < 0) {
4163                 dev_kfree_skb_any(skb);
4164                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
4165                 return NETDEV_TX_OK;
4166         }
4167
4168         if (tso)
4169                 tx_flags |= E1000_TX_FLAGS_TSO;
4170         else if (e1000_tx_csum(adapter, skb))
4171                 tx_flags |= E1000_TX_FLAGS_CSUM;
4172
4173         /*
4174          * Old method was to assume IPv4 packet by default if TSO was enabled.
4175          * 82571 hardware supports TSO capabilities for IPv6 as well...
4176          * no longer assume, we must.
4177          */
4178         if (skb->protocol == htons(ETH_P_IP))
4179                 tx_flags |= E1000_TX_FLAGS_IPV4;
4180
4181         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4182         if (count < 0) {
4183                 /* handle pci_map_single() error in e1000_tx_map */
4184                 dev_kfree_skb_any(skb);
4185                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
4186                 return NETDEV_TX_OK;
4187         }
4188
4189         e1000_tx_queue(adapter, tx_flags, count);
4190
4191         netdev->trans_start = jiffies;
4192
4193         /* Make sure there is space in the ring for the next send. */
4194         e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4195
4196         spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
4197         return NETDEV_TX_OK;
4198 }
4199
4200 /**
4201  * e1000_tx_timeout - Respond to a Tx Hang
4202  * @netdev: network interface device structure
4203  **/
4204 static void e1000_tx_timeout(struct net_device *netdev)
4205 {
4206         struct e1000_adapter *adapter = netdev_priv(netdev);
4207
4208         /* Do the reset outside of interrupt context */
4209         adapter->tx_timeout_count++;
4210         schedule_work(&adapter->reset_task);
4211 }
4212
4213 static void e1000_reset_task(struct work_struct *work)
4214 {
4215         struct e1000_adapter *adapter;
4216         adapter = container_of(work, struct e1000_adapter, reset_task);
4217
4218         e1000e_reinit_locked(adapter);
4219 }
4220
4221 /**
4222  * e1000_get_stats - Get System Network Statistics
4223  * @netdev: network interface device structure
4224  *
4225  * Returns the address of the device statistics structure.
4226  * The statistics are actually updated from the timer callback.
4227  **/
4228 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4229 {
4230         struct e1000_adapter *adapter = netdev_priv(netdev);
4231
4232         /* only return the current stats */
4233         return &adapter->net_stats;
4234 }
4235
4236 /**
4237  * e1000_change_mtu - Change the Maximum Transfer Unit
4238  * @netdev: network interface device structure
4239  * @new_mtu: new value for maximum frame size
4240  *
4241  * Returns 0 on success, negative on failure
4242  **/
4243 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4244 {
4245         struct e1000_adapter *adapter = netdev_priv(netdev);
4246         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4247
4248         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4249             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
4250                 e_err("Invalid MTU setting\n");
4251                 return -EINVAL;
4252         }
4253
4254         /* Jumbo frame size limits */
4255         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
4256                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4257                         e_err("Jumbo Frames not supported.\n");
4258                         return -EINVAL;
4259                 }
4260                 if (adapter->hw.phy.type == e1000_phy_ife) {
4261                         e_err("Jumbo Frames not supported.\n");
4262                         return -EINVAL;
4263                 }
4264         }
4265
4266 #define MAX_STD_JUMBO_FRAME_SIZE 9234
4267         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
4268                 e_err("MTU > 9216 not supported.\n");
4269                 return -EINVAL;
4270         }
4271
4272         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4273                 msleep(1);
4274         /* e1000e_down has a dependency on max_frame_size */
4275         adapter->max_frame_size = max_frame;
4276         if (netif_running(netdev))
4277                 e1000e_down(adapter);
4278
4279         /*
4280          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4281          * means we reserve 2 more, this pushes us to allocate from the next
4282          * larger slab size.
4283          * i.e. RXBUFFER_2048 --> size-4096 slab
4284          * However with the new *_jumbo_rx* routines, jumbo receives will use
4285          * fragmented skbs
4286          */
4287
4288         if (max_frame <= 256)
4289                 adapter->rx_buffer_len = 256;
4290         else if (max_frame <= 512)
4291                 adapter->rx_buffer_len = 512;
4292         else if (max_frame <= 1024)
4293                 adapter->rx_buffer_len = 1024;
4294         else if (max_frame <= 2048)
4295                 adapter->rx_buffer_len = 2048;
4296         else
4297                 adapter->rx_buffer_len = 4096;
4298
4299         /* adjust allocation if LPE protects us, and we aren't using SBP */
4300         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4301              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4302                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4303                                          + ETH_FCS_LEN;
4304
4305         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4306         netdev->mtu = new_mtu;
4307
4308         if (netif_running(netdev))
4309                 e1000e_up(adapter);
4310         else
4311                 e1000e_reset(adapter);
4312
4313         clear_bit(__E1000_RESETTING, &adapter->state);
4314
4315         return 0;
4316 }
4317
4318 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4319                            int cmd)
4320 {
4321         struct e1000_adapter *adapter = netdev_priv(netdev);
4322         struct mii_ioctl_data *data = if_mii(ifr);
4323
4324         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4325                 return -EOPNOTSUPP;
4326
4327         switch (cmd) {
4328         case SIOCGMIIPHY:
4329                 data->phy_id = adapter->hw.phy.addr;
4330                 break;
4331         case SIOCGMIIREG:
4332                 if (!capable(CAP_NET_ADMIN))
4333                         return -EPERM;
4334                 switch (data->reg_num & 0x1F) {
4335                 case MII_BMCR:
4336                         data->val_out = adapter->phy_regs.bmcr;
4337                         break;
4338                 case MII_BMSR:
4339                         data->val_out = adapter->phy_regs.bmsr;
4340                         break;
4341                 case MII_PHYSID1:
4342                         data->val_out = (adapter->hw.phy.id >> 16);
4343                         break;
4344                 case MII_PHYSID2:
4345                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4346                         break;
4347                 case MII_ADVERTISE:
4348                         data->val_out = adapter->phy_regs.advertise;
4349                         break;
4350                 case MII_LPA:
4351                         data->val_out = adapter->phy_regs.lpa;
4352                         break;
4353                 case MII_EXPANSION:
4354                         data->val_out = adapter->phy_regs.expansion;
4355                         break;
4356                 case MII_CTRL1000:
4357                         data->val_out = adapter->phy_regs.ctrl1000;
4358                         break;
4359                 case MII_STAT1000:
4360                         data->val_out = adapter->phy_regs.stat1000;
4361                         break;
4362                 case MII_ESTATUS:
4363                         data->val_out = adapter->phy_regs.estatus;
4364                         break;
4365                 default:
4366                         return -EIO;
4367                 }
4368                 break;
4369         case SIOCSMIIREG:
4370         default:
4371                 return -EOPNOTSUPP;
4372         }
4373         return 0;
4374 }
4375
4376 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4377 {
4378         switch (cmd) {
4379         case SIOCGMIIPHY:
4380         case SIOCGMIIREG:
4381         case SIOCSMIIREG:
4382                 return e1000_mii_ioctl(netdev, ifr, cmd);
4383         default:
4384                 return -EOPNOTSUPP;
4385         }
4386 }
4387
4388 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4389 {
4390         struct net_device *netdev = pci_get_drvdata(pdev);
4391         struct e1000_adapter *adapter = netdev_priv(netdev);
4392         struct e1000_hw *hw = &adapter->hw;
4393         u32 ctrl, ctrl_ext, rctl, status;
4394         u32 wufc = adapter->wol;
4395         int retval = 0;
4396
4397         netif_device_detach(netdev);
4398
4399         if (netif_running(netdev)) {
4400                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4401                 e1000e_down(adapter);
4402                 e1000_free_irq(adapter);
4403         }
4404         e1000e_reset_interrupt_capability(adapter);
4405
4406         retval = pci_save_state(pdev);
4407         if (retval)
4408                 return retval;
4409
4410         status = er32(STATUS);
4411         if (status & E1000_STATUS_LU)
4412                 wufc &= ~E1000_WUFC_LNKC;
4413
4414         if (wufc) {
4415                 e1000_setup_rctl(adapter);
4416                 e1000_set_multi(netdev);
4417
4418                 /* turn on all-multi mode if wake on multicast is enabled */
4419                 if (wufc & E1000_WUFC_MC) {
4420                         rctl = er32(RCTL);
4421                         rctl |= E1000_RCTL_MPE;
4422                         ew32(RCTL, rctl);
4423                 }
4424
4425                 ctrl = er32(CTRL);
4426                 /* advertise wake from D3Cold */
4427                 #define E1000_CTRL_ADVD3WUC 0x00100000
4428                 /* phy power management enable */
4429                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4430                 ctrl |= E1000_CTRL_ADVD3WUC |
4431                         E1000_CTRL_EN_PHY_PWR_MGMT;
4432                 ew32(CTRL, ctrl);
4433
4434                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4435                     adapter->hw.phy.media_type ==
4436                     e1000_media_type_internal_serdes) {
4437                         /* keep the laser running in D3 */
4438                         ctrl_ext = er32(CTRL_EXT);
4439                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4440                         ew32(CTRL_EXT, ctrl_ext);
4441                 }
4442
4443                 if (adapter->flags & FLAG_IS_ICH)
4444                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4445
4446                 /* Allow time for pending master requests to run */
4447                 e1000e_disable_pcie_master(&adapter->hw);
4448
4449                 ew32(WUC, E1000_WUC_PME_EN);
4450                 ew32(WUFC, wufc);
4451                 pci_enable_wake(pdev, PCI_D3hot, 1);
4452                 pci_enable_wake(pdev, PCI_D3cold, 1);
4453         } else {
4454                 ew32(WUC, 0);
4455                 ew32(WUFC, 0);
4456                 pci_enable_wake(pdev, PCI_D3hot, 0);
4457                 pci_enable_wake(pdev, PCI_D3cold, 0);
4458         }
4459
4460         /* make sure adapter isn't asleep if manageability is enabled */
4461         if (adapter->flags & FLAG_MNG_PT_ENABLED) {
4462                 pci_enable_wake(pdev, PCI_D3hot, 1);
4463                 pci_enable_wake(pdev, PCI_D3cold, 1);
4464         }
4465
4466         if (adapter->hw.phy.type == e1000_phy_igp_3)
4467                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4468
4469         /*
4470          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4471          * would have already happened in close and is redundant.
4472          */
4473         e1000_release_hw_control(adapter);
4474
4475         pci_disable_device(pdev);
4476
4477         /*
4478          * The pci-e switch on some quad port adapters will report a
4479          * correctable error when the MAC transitions from D0 to D3.  To
4480          * prevent this we need to mask off the correctable errors on the
4481          * downstream port of the pci-e switch.
4482          */
4483         if (adapter->flags & FLAG_IS_QUAD_PORT) {
4484                 struct pci_dev *us_dev = pdev->bus->self;
4485                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
4486                 u16 devctl;
4487
4488                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
4489                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
4490                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
4491
4492                 pci_set_power_state(pdev, pci_choose_state(pdev, state));
4493
4494                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
4495         } else {
4496                 pci_set_power_state(pdev, pci_choose_state(pdev, state));
4497         }
4498
4499         return 0;
4500 }
4501
4502 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4503 {
4504         int pos;
4505         u16 val;
4506
4507         /*
4508          * 82573 workaround - disable L1 ASPM on mobile chipsets
4509          *
4510          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4511          * resulting in lost data or garbage information on the pci-e link
4512          * level. This could result in (false) bad EEPROM checksum errors,
4513          * long ping times (up to 2s) or even a system freeze/hang.
4514          *
4515          * Unfortunately this feature saves about 1W power consumption when
4516          * active.
4517          */
4518         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4519         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4520         if (val & 0x2) {
4521                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4522                 val &= ~0x2;
4523                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4524         }
4525 }
4526
4527 #ifdef CONFIG_PM
4528 static int e1000_resume(struct pci_dev *pdev)
4529 {
4530         struct net_device *netdev = pci_get_drvdata(pdev);
4531         struct e1000_adapter *adapter = netdev_priv(netdev);
4532         struct e1000_hw *hw = &adapter->hw;
4533         u32 err;
4534
4535         pci_set_power_state(pdev, PCI_D0);
4536         pci_restore_state(pdev);
4537         e1000e_disable_l1aspm(pdev);
4538
4539         err = pci_enable_device_mem(pdev);
4540         if (err) {
4541                 dev_err(&pdev->dev,
4542                         "Cannot enable PCI device from suspend\n");
4543                 return err;
4544         }
4545
4546         pci_set_master(pdev);
4547
4548         pci_enable_wake(pdev, PCI_D3hot, 0);
4549         pci_enable_wake(pdev, PCI_D3cold, 0);
4550
4551         e1000e_set_interrupt_capability(adapter);
4552         if (netif_running(netdev)) {
4553                 err = e1000_request_irq(adapter);
4554                 if (err)
4555                         return err;
4556         }
4557
4558         e1000e_power_up_phy(adapter);
4559         e1000e_reset(adapter);
4560         ew32(WUS, ~0);
4561
4562         e1000_init_manageability(adapter);
4563
4564         if (netif_running(netdev))
4565                 e1000e_up(adapter);
4566
4567         netif_device_attach(netdev);
4568
4569         /*
4570          * If the controller has AMT, do not set DRV_LOAD until the interface
4571          * is up.  For all other cases, let the f/w know that the h/w is now
4572          * under the control of the driver.
4573          */
4574         if (!(adapter->flags & FLAG_HAS_AMT))
4575                 e1000_get_hw_control(adapter);
4576
4577         return 0;
4578 }
4579 #endif
4580
4581 static void e1000_shutdown(struct pci_dev *pdev)
4582 {
4583         e1000_suspend(pdev, PMSG_SUSPEND);
4584 }
4585
4586 #ifdef CONFIG_NET_POLL_CONTROLLER
4587 /*
4588  * Polling 'interrupt' - used by things like netconsole to send skbs
4589  * without having to re-enable interrupts. It's not called while
4590  * the interrupt routine is executing.
4591  */
4592 static void e1000_netpoll(struct net_device *netdev)
4593 {
4594         struct e1000_adapter *adapter = netdev_priv(netdev);
4595
4596         disable_irq(adapter->pdev->irq);
4597         e1000_intr(adapter->pdev->irq, netdev);
4598
4599         enable_irq(adapter->pdev->irq);
4600 }
4601 #endif
4602
4603 /**
4604  * e1000_io_error_detected - called when PCI error is detected
4605  * @pdev: Pointer to PCI device
4606  * @state: The current pci connection state
4607  *
4608  * This function is called after a PCI bus error affecting
4609  * this device has been detected.
4610  */
4611 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4612                                                 pci_channel_state_t state)
4613 {
4614         struct net_device *netdev = pci_get_drvdata(pdev);
4615         struct e1000_adapter *adapter = netdev_priv(netdev);
4616
4617         netif_device_detach(netdev);
4618
4619         if (netif_running(netdev))
4620                 e1000e_down(adapter);
4621         pci_disable_device(pdev);
4622
4623         /* Request a slot slot reset. */
4624         return PCI_ERS_RESULT_NEED_RESET;
4625 }
4626
4627 /**
4628  * e1000_io_slot_reset - called after the pci bus has been reset.
4629  * @pdev: Pointer to PCI device
4630  *
4631  * Restart the card from scratch, as if from a cold-boot. Implementation
4632  * resembles the first-half of the e1000_resume routine.
4633  */
4634 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4635 {
4636         struct net_device *netdev = pci_get_drvdata(pdev);
4637         struct e1000_adapter *adapter = netdev_priv(netdev);
4638         struct e1000_hw *hw = &adapter->hw;
4639         int err;
4640
4641         e1000e_disable_l1aspm(pdev);
4642         err = pci_enable_device_mem(pdev);
4643         if (err) {
4644                 dev_err(&pdev->dev,
4645                         "Cannot re-enable PCI device after reset.\n");
4646                 return PCI_ERS_RESULT_DISCONNECT;
4647         }
4648         pci_set_master(pdev);
4649         pci_restore_state(pdev);
4650
4651         pci_enable_wake(pdev, PCI_D3hot, 0);
4652         pci_enable_wake(pdev, PCI_D3cold, 0);
4653
4654         e1000e_reset(adapter);
4655         ew32(WUS, ~0);
4656
4657         return PCI_ERS_RESULT_RECOVERED;
4658 }
4659
4660 /**
4661  * e1000_io_resume - called when traffic can start flowing again.
4662  * @pdev: Pointer to PCI device
4663  *
4664  * This callback is called when the error recovery driver tells us that
4665  * its OK to resume normal operation. Implementation resembles the
4666  * second-half of the e1000_resume routine.
4667  */
4668 static void e1000_io_resume(struct pci_dev *pdev)
4669 {
4670         struct net_device *netdev = pci_get_drvdata(pdev);
4671         struct e1000_adapter *adapter = netdev_priv(netdev);
4672
4673         e1000_init_manageability(adapter);
4674
4675         if (netif_running(netdev)) {
4676                 if (e1000e_up(adapter)) {
4677                         dev_err(&pdev->dev,
4678                                 "can't bring device back up after reset\n");
4679                         return;
4680                 }
4681         }
4682
4683         netif_device_attach(netdev);
4684
4685         /*
4686          * If the controller has AMT, do not set DRV_LOAD until the interface
4687          * is up.  For all other cases, let the f/w know that the h/w is now
4688          * under the control of the driver.
4689          */
4690         if (!(adapter->flags & FLAG_HAS_AMT))
4691                 e1000_get_hw_control(adapter);
4692
4693 }
4694
4695 static void e1000_print_device_info(struct e1000_adapter *adapter)
4696 {
4697         struct e1000_hw *hw = &adapter->hw;
4698         struct net_device *netdev = adapter->netdev;
4699         u32 pba_num;
4700
4701         /* print bus type/speed/width info */
4702         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
4703                /* bus width */
4704                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4705                 "Width x1"),
4706                /* MAC address */
4707                netdev->dev_addr);
4708         e_info("Intel(R) PRO/%s Network Connection\n",
4709                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4710         e1000e_read_pba_num(hw, &pba_num);
4711         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4712                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4713 }
4714
4715 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4716 {
4717         struct e1000_hw *hw = &adapter->hw;
4718         int ret_val;
4719         u16 buf = 0;
4720
4721         if (hw->mac.type != e1000_82573)
4722                 return;
4723
4724         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4725         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
4726                 /* Deep Smart Power Down (DSPD) */
4727                 dev_warn(&adapter->pdev->dev,
4728                          "Warning: detected DSPD enabled in EEPROM\n");
4729         }
4730
4731         ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4732         if (!ret_val && (le16_to_cpu(buf) & (3 << 2))) {
4733                 /* ASPM enable */
4734                 dev_warn(&adapter->pdev->dev,
4735                          "Warning: detected ASPM enabled in EEPROM\n");
4736         }
4737 }
4738
4739 static const struct net_device_ops e1000e_netdev_ops = {
4740         .ndo_open               = e1000_open,
4741         .ndo_stop               = e1000_close,
4742         .ndo_start_xmit         = e1000_xmit_frame,
4743         .ndo_get_stats          = e1000_get_stats,
4744         .ndo_set_multicast_list = e1000_set_multi,
4745         .ndo_set_mac_address    = e1000_set_mac,
4746         .ndo_change_mtu         = e1000_change_mtu,
4747         .ndo_do_ioctl           = e1000_ioctl,
4748         .ndo_tx_timeout         = e1000_tx_timeout,
4749         .ndo_validate_addr      = eth_validate_addr,
4750
4751         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
4752         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
4753         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
4754 #ifdef CONFIG_NET_POLL_CONTROLLER
4755         .ndo_poll_controller    = e1000_netpoll,
4756 #endif
4757 };
4758
4759 /**
4760  * e1000_probe - Device Initialization Routine
4761  * @pdev: PCI device information struct
4762  * @ent: entry in e1000_pci_tbl
4763  *
4764  * Returns 0 on success, negative on failure
4765  *
4766  * e1000_probe initializes an adapter identified by a pci_dev structure.
4767  * The OS initialization, configuring of the adapter private structure,
4768  * and a hardware reset occur.
4769  **/
4770 static int __devinit e1000_probe(struct pci_dev *pdev,
4771                                  const struct pci_device_id *ent)
4772 {
4773         struct net_device *netdev;
4774         struct e1000_adapter *adapter;
4775         struct e1000_hw *hw;
4776         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4777         resource_size_t mmio_start, mmio_len;
4778         resource_size_t flash_start, flash_len;
4779
4780         static int cards_found;
4781         int i, err, pci_using_dac;
4782         u16 eeprom_data = 0;
4783         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4784
4785         e1000e_disable_l1aspm(pdev);
4786
4787         err = pci_enable_device_mem(pdev);
4788         if (err)
4789                 return err;
4790
4791         pci_using_dac = 0;
4792         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
4793         if (!err) {
4794                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
4795                 if (!err)
4796                         pci_using_dac = 1;
4797         } else {
4798                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
4799                 if (err) {
4800                         err = pci_set_consistent_dma_mask(pdev,
4801                                                           DMA_32BIT_MASK);
4802                         if (err) {
4803                                 dev_err(&pdev->dev, "No usable DMA "
4804                                         "configuration, aborting\n");
4805                                 goto err_dma;
4806                         }
4807                 }
4808         }
4809
4810         err = pci_request_selected_regions(pdev,
4811                                           pci_select_bars(pdev, IORESOURCE_MEM),
4812                                           e1000e_driver_name);
4813         if (err)
4814                 goto err_pci_reg;
4815
4816         pci_set_master(pdev);
4817         /* PCI config space info */
4818         err = pci_save_state(pdev);
4819         if (err)
4820                 goto err_alloc_etherdev;
4821
4822         err = -ENOMEM;
4823         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4824         if (!netdev)
4825                 goto err_alloc_etherdev;
4826
4827         SET_NETDEV_DEV(netdev, &pdev->dev);
4828
4829         pci_set_drvdata(pdev, netdev);
4830         adapter = netdev_priv(netdev);
4831         hw = &adapter->hw;
4832         adapter->netdev = netdev;
4833         adapter->pdev = pdev;
4834         adapter->ei = ei;
4835         adapter->pba = ei->pba;
4836         adapter->flags = ei->flags;
4837         adapter->flags2 = ei->flags2;
4838         adapter->hw.adapter = adapter;
4839         adapter->hw.mac.type = ei->mac;
4840         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4841
4842         mmio_start = pci_resource_start(pdev, 0);
4843         mmio_len = pci_resource_len(pdev, 0);
4844
4845         err = -EIO;
4846         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4847         if (!adapter->hw.hw_addr)
4848                 goto err_ioremap;
4849
4850         if ((adapter->flags & FLAG_HAS_FLASH) &&
4851             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
4852                 flash_start = pci_resource_start(pdev, 1);
4853                 flash_len = pci_resource_len(pdev, 1);
4854                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
4855                 if (!adapter->hw.flash_address)
4856                         goto err_flashmap;
4857         }
4858
4859         /* construct the net_device struct */
4860         netdev->netdev_ops              = &e1000e_netdev_ops;
4861         e1000e_set_ethtool_ops(netdev);
4862         netdev->watchdog_timeo          = 5 * HZ;
4863         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
4864         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
4865
4866         netdev->mem_start = mmio_start;
4867         netdev->mem_end = mmio_start + mmio_len;
4868
4869         adapter->bd_number = cards_found++;
4870
4871         e1000e_check_options(adapter);
4872
4873         /* setup adapter struct */
4874         err = e1000_sw_init(adapter);
4875         if (err)
4876                 goto err_sw_init;
4877
4878         err = -EIO;
4879
4880         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4881         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
4882         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4883
4884         err = ei->get_variants(adapter);
4885         if (err)
4886                 goto err_hw_init;
4887
4888         if ((adapter->flags & FLAG_IS_ICH) &&
4889             (adapter->flags & FLAG_READ_ONLY_NVM))
4890                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
4891
4892         hw->mac.ops.get_bus_info(&adapter->hw);
4893
4894         adapter->hw.phy.autoneg_wait_to_complete = 0;
4895
4896         /* Copper options */
4897         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
4898                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
4899                 adapter->hw.phy.disable_polarity_correction = 0;
4900                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
4901         }
4902
4903         if (e1000_check_reset_block(&adapter->hw))
4904                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
4905
4906         netdev->features = NETIF_F_SG |
4907                            NETIF_F_HW_CSUM |
4908                            NETIF_F_HW_VLAN_TX |
4909                            NETIF_F_HW_VLAN_RX;
4910
4911         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
4912                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
4913
4914         netdev->features |= NETIF_F_TSO;
4915         netdev->features |= NETIF_F_TSO6;
4916
4917         netdev->vlan_features |= NETIF_F_TSO;
4918         netdev->vlan_features |= NETIF_F_TSO6;
4919         netdev->vlan_features |= NETIF_F_HW_CSUM;
4920         netdev->vlan_features |= NETIF_F_SG;
4921
4922         if (pci_using_dac)
4923                 netdev->features |= NETIF_F_HIGHDMA;
4924
4925         /*
4926          * We should not be using LLTX anymore, but we are still Tx faster with
4927          * it.
4928          */
4929         netdev->features |= NETIF_F_LLTX;
4930
4931         if (e1000e_enable_mng_pass_thru(&adapter->hw))
4932                 adapter->flags |= FLAG_MNG_PT_ENABLED;
4933
4934         /*
4935          * before reading the NVM, reset the controller to
4936          * put the device in a known good starting state
4937          */
4938         adapter->hw.mac.ops.reset_hw(&adapter->hw);
4939
4940         /*
4941          * systems with ASPM and others may see the checksum fail on the first
4942          * attempt. Let's give it a few tries
4943          */
4944         for (i = 0;; i++) {
4945                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
4946                         break;
4947                 if (i == 2) {
4948                         e_err("The NVM Checksum Is Not Valid\n");
4949                         err = -EIO;
4950                         goto err_eeprom;
4951                 }
4952         }
4953
4954         e1000_eeprom_checks(adapter);
4955
4956         /* copy the MAC address out of the NVM */
4957         if (e1000e_read_mac_addr(&adapter->hw))
4958                 e_err("NVM Read Error while reading MAC address\n");
4959
4960         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
4961         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
4962
4963         if (!is_valid_ether_addr(netdev->perm_addr)) {
4964                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
4965                 err = -EIO;
4966                 goto err_eeprom;
4967         }
4968
4969         init_timer(&adapter->watchdog_timer);
4970         adapter->watchdog_timer.function = &e1000_watchdog;
4971         adapter->watchdog_timer.data = (unsigned long) adapter;
4972
4973         init_timer(&adapter->phy_info_timer);
4974         adapter->phy_info_timer.function = &e1000_update_phy_info;
4975         adapter->phy_info_timer.data = (unsigned long) adapter;
4976
4977         INIT_WORK(&adapter->reset_task, e1000_reset_task);
4978         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
4979         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
4980         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
4981
4982         /* Initialize link parameters. User can change them with ethtool */
4983         adapter->hw.mac.autoneg = 1;
4984         adapter->fc_autoneg = 1;
4985         adapter->hw.fc.requested_mode = e1000_fc_default;
4986         adapter->hw.fc.current_mode = e1000_fc_default;
4987         adapter->hw.phy.autoneg_advertised = 0x2f;
4988
4989         /* ring size defaults */
4990         adapter->rx_ring->count = 256;
4991         adapter->tx_ring->count = 256;
4992
4993         /*
4994          * Initial Wake on LAN setting - If APM wake is enabled in
4995          * the EEPROM, enable the ACPI Magic Packet filter
4996          */
4997         if (adapter->flags & FLAG_APME_IN_WUC) {
4998                 /* APME bit in EEPROM is mapped to WUC.APME */
4999                 eeprom_data = er32(WUC);
5000                 eeprom_apme_mask = E1000_WUC_APME;
5001         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5002                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5003                     (adapter->hw.bus.func == 1))
5004                         e1000_read_nvm(&adapter->hw,
5005                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5006                 else
5007                         e1000_read_nvm(&adapter->hw,
5008                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5009         }
5010
5011         /* fetch WoL from EEPROM */
5012         if (eeprom_data & eeprom_apme_mask)
5013                 adapter->eeprom_wol |= E1000_WUFC_MAG;
5014
5015         /*
5016          * now that we have the eeprom settings, apply the special cases
5017          * where the eeprom may be wrong or the board simply won't support
5018          * wake on lan on a particular port
5019          */
5020         if (!(adapter->flags & FLAG_HAS_WOL))
5021                 adapter->eeprom_wol = 0;
5022
5023         /* initialize the wol settings based on the eeprom settings */
5024         adapter->wol = adapter->eeprom_wol;
5025         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5026
5027         /* save off EEPROM version number */
5028         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5029
5030         /* reset the hardware with the new settings */
5031         e1000e_reset(adapter);
5032
5033         /*
5034          * If the controller has AMT, do not set DRV_LOAD until the interface
5035          * is up.  For all other cases, let the f/w know that the h/w is now
5036          * under the control of the driver.
5037          */
5038         if (!(adapter->flags & FLAG_HAS_AMT))
5039                 e1000_get_hw_control(adapter);
5040
5041         /* tell the stack to leave us alone until e1000_open() is called */
5042         netif_carrier_off(netdev);
5043         netif_tx_stop_all_queues(netdev);
5044
5045         strcpy(netdev->name, "eth%d");
5046         err = register_netdev(netdev);
5047         if (err)
5048                 goto err_register;
5049
5050         e1000_print_device_info(adapter);
5051
5052         return 0;
5053
5054 err_register:
5055         if (!(adapter->flags & FLAG_HAS_AMT))
5056                 e1000_release_hw_control(adapter);
5057 err_eeprom:
5058         if (!e1000_check_reset_block(&adapter->hw))
5059                 e1000_phy_hw_reset(&adapter->hw);
5060 err_hw_init:
5061
5062         kfree(adapter->tx_ring);
5063         kfree(adapter->rx_ring);
5064 err_sw_init:
5065         if (adapter->hw.flash_address)
5066                 iounmap(adapter->hw.flash_address);
5067         e1000e_reset_interrupt_capability(adapter);
5068 err_flashmap:
5069         iounmap(adapter->hw.hw_addr);
5070 err_ioremap:
5071         free_netdev(netdev);
5072 err_alloc_etherdev:
5073         pci_release_selected_regions(pdev,
5074                                      pci_select_bars(pdev, IORESOURCE_MEM));
5075 err_pci_reg:
5076 err_dma:
5077         pci_disable_device(pdev);
5078         return err;
5079 }
5080
5081 /**
5082  * e1000_remove - Device Removal Routine
5083  * @pdev: PCI device information struct
5084  *
5085  * e1000_remove is called by the PCI subsystem to alert the driver
5086  * that it should release a PCI device.  The could be caused by a
5087  * Hot-Plug event, or because the driver is going to be removed from
5088  * memory.
5089  **/
5090 static void __devexit e1000_remove(struct pci_dev *pdev)
5091 {
5092         struct net_device *netdev = pci_get_drvdata(pdev);
5093         struct e1000_adapter *adapter = netdev_priv(netdev);
5094
5095         /*
5096          * flush_scheduled work may reschedule our watchdog task, so
5097          * explicitly disable watchdog tasks from being rescheduled
5098          */
5099         set_bit(__E1000_DOWN, &adapter->state);
5100         del_timer_sync(&adapter->watchdog_timer);
5101         del_timer_sync(&adapter->phy_info_timer);
5102
5103         flush_scheduled_work();
5104
5105         /*
5106          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5107          * would have already happened in close and is redundant.
5108          */
5109         e1000_release_hw_control(adapter);
5110
5111         unregister_netdev(netdev);
5112
5113         if (!e1000_check_reset_block(&adapter->hw))
5114                 e1000_phy_hw_reset(&adapter->hw);
5115
5116         e1000e_reset_interrupt_capability(adapter);
5117         kfree(adapter->tx_ring);
5118         kfree(adapter->rx_ring);
5119
5120         iounmap(adapter->hw.hw_addr);
5121         if (adapter->hw.flash_address)
5122                 iounmap(adapter->hw.flash_address);
5123         pci_release_selected_regions(pdev,
5124                                      pci_select_bars(pdev, IORESOURCE_MEM));
5125
5126         free_netdev(netdev);
5127
5128         pci_disable_device(pdev);
5129 }
5130
5131 /* PCI Error Recovery (ERS) */
5132 static struct pci_error_handlers e1000_err_handler = {
5133         .error_detected = e1000_io_error_detected,
5134         .slot_reset = e1000_io_slot_reset,
5135         .resume = e1000_io_resume,
5136 };
5137
5138 static struct pci_device_id e1000_pci_tbl[] = {
5139         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5140         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5141         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5142         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5143         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5144         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5145         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5146         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5147         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5148
5149         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5150         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5151         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5152         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5153
5154         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5155         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5156         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5157
5158         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5159
5160         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5161           board_80003es2lan },
5162         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5163           board_80003es2lan },
5164         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5165           board_80003es2lan },
5166         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5167           board_80003es2lan },
5168
5169         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5170         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5171         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5172         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5173         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5174         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5175         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5176
5177         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5178         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5179         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5180         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5181         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5182         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5183         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5184         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5185         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5186
5187         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5188         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5189         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5190
5191         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5192         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5193
5194         { }     /* terminate list */
5195 };
5196 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5197
5198 /* PCI Device API Driver */
5199 static struct pci_driver e1000_driver = {
5200         .name     = e1000e_driver_name,
5201         .id_table = e1000_pci_tbl,
5202         .probe    = e1000_probe,
5203         .remove   = __devexit_p(e1000_remove),
5204 #ifdef CONFIG_PM
5205         /* Power Management Hooks */
5206         .suspend  = e1000_suspend,
5207         .resume   = e1000_resume,
5208 #endif
5209         .shutdown = e1000_shutdown,
5210         .err_handler = &e1000_err_handler
5211 };
5212
5213 /**
5214  * e1000_init_module - Driver Registration Routine
5215  *
5216  * e1000_init_module is the first routine called when the driver is
5217  * loaded. All it does is register with the PCI subsystem.
5218  **/
5219 static int __init e1000_init_module(void)
5220 {
5221         int ret;
5222         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
5223                e1000e_driver_name, e1000e_driver_version);
5224         printk(KERN_INFO "%s: Copyright (c) 1999-2008 Intel Corporation.\n",
5225                e1000e_driver_name);
5226         ret = pci_register_driver(&e1000_driver);
5227         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name,
5228                                PM_QOS_DEFAULT_VALUE);
5229                                 
5230         return ret;
5231 }
5232 module_init(e1000_init_module);
5233
5234 /**
5235  * e1000_exit_module - Driver Exit Cleanup Routine
5236  *
5237  * e1000_exit_module is called just before the driver is removed
5238  * from memory.
5239  **/
5240 static void __exit e1000_exit_module(void)
5241 {
5242         pci_unregister_driver(&e1000_driver);
5243         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name);
5244 }
5245 module_exit(e1000_exit_module);
5246
5247
5248 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5249 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5250 MODULE_LICENSE("GPL");
5251 MODULE_VERSION(DRV_VERSION);
5252
5253 /* e1000_main.c */