]> git.karo-electronics.de Git - mv-sheeva.git/blob - drivers/net/e1000e/netdev.c
e1000e / PCI / PM: Add basic runtime PM support (rev. 4)
[mv-sheeva.git] / drivers / net / e1000e / netdev.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2009 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #include <linux/module.h>
30 #include <linux/types.h>
31 #include <linux/init.h>
32 #include <linux/pci.h>
33 #include <linux/vmalloc.h>
34 #include <linux/pagemap.h>
35 #include <linux/delay.h>
36 #include <linux/netdevice.h>
37 #include <linux/tcp.h>
38 #include <linux/ipv6.h>
39 #include <net/checksum.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/mii.h>
42 #include <linux/ethtool.h>
43 #include <linux/if_vlan.h>
44 #include <linux/cpu.h>
45 #include <linux/smp.h>
46 #include <linux/pm_qos_params.h>
47 #include <linux/pm_runtime.h>
48 #include <linux/aer.h>
49
50 #include "e1000.h"
51
52 #define DRV_VERSION "1.0.2-k2"
53 char e1000e_driver_name[] = "e1000e";
54 const char e1000e_driver_version[] = DRV_VERSION;
55
56 static const struct e1000_info *e1000_info_tbl[] = {
57         [board_82571]           = &e1000_82571_info,
58         [board_82572]           = &e1000_82572_info,
59         [board_82573]           = &e1000_82573_info,
60         [board_82574]           = &e1000_82574_info,
61         [board_82583]           = &e1000_82583_info,
62         [board_80003es2lan]     = &e1000_es2_info,
63         [board_ich8lan]         = &e1000_ich8_info,
64         [board_ich9lan]         = &e1000_ich9_info,
65         [board_ich10lan]        = &e1000_ich10_info,
66         [board_pchlan]          = &e1000_pch_info,
67 };
68
69 /**
70  * e1000_desc_unused - calculate if we have unused descriptors
71  **/
72 static int e1000_desc_unused(struct e1000_ring *ring)
73 {
74         if (ring->next_to_clean > ring->next_to_use)
75                 return ring->next_to_clean - ring->next_to_use - 1;
76
77         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
78 }
79
80 /**
81  * e1000_receive_skb - helper function to handle Rx indications
82  * @adapter: board private structure
83  * @status: descriptor status field as written by hardware
84  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
85  * @skb: pointer to sk_buff to be indicated to stack
86  **/
87 static void e1000_receive_skb(struct e1000_adapter *adapter,
88                               struct net_device *netdev,
89                               struct sk_buff *skb,
90                               u8 status, __le16 vlan)
91 {
92         skb->protocol = eth_type_trans(skb, netdev);
93
94         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
95                 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
96                                  le16_to_cpu(vlan), skb);
97         else
98                 napi_gro_receive(&adapter->napi, skb);
99 }
100
101 /**
102  * e1000_rx_checksum - Receive Checksum Offload for 82543
103  * @adapter:     board private structure
104  * @status_err:  receive descriptor status and error fields
105  * @csum:       receive descriptor csum field
106  * @sk_buff:     socket buffer with received data
107  **/
108 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
109                               u32 csum, struct sk_buff *skb)
110 {
111         u16 status = (u16)status_err;
112         u8 errors = (u8)(status_err >> 24);
113         skb->ip_summed = CHECKSUM_NONE;
114
115         /* Ignore Checksum bit is set */
116         if (status & E1000_RXD_STAT_IXSM)
117                 return;
118         /* TCP/UDP checksum error bit is set */
119         if (errors & E1000_RXD_ERR_TCPE) {
120                 /* let the stack verify checksum errors */
121                 adapter->hw_csum_err++;
122                 return;
123         }
124
125         /* TCP/UDP Checksum has not been calculated */
126         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
127                 return;
128
129         /* It must be a TCP or UDP packet with a valid checksum */
130         if (status & E1000_RXD_STAT_TCPCS) {
131                 /* TCP checksum is good */
132                 skb->ip_summed = CHECKSUM_UNNECESSARY;
133         } else {
134                 /*
135                  * IP fragment with UDP payload
136                  * Hardware complements the payload checksum, so we undo it
137                  * and then put the value in host order for further stack use.
138                  */
139                 __sum16 sum = (__force __sum16)htons(csum);
140                 skb->csum = csum_unfold(~sum);
141                 skb->ip_summed = CHECKSUM_COMPLETE;
142         }
143         adapter->hw_csum_good++;
144 }
145
146 /**
147  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
148  * @adapter: address of board private structure
149  **/
150 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
151                                    int cleaned_count)
152 {
153         struct net_device *netdev = adapter->netdev;
154         struct pci_dev *pdev = adapter->pdev;
155         struct e1000_ring *rx_ring = adapter->rx_ring;
156         struct e1000_rx_desc *rx_desc;
157         struct e1000_buffer *buffer_info;
158         struct sk_buff *skb;
159         unsigned int i;
160         unsigned int bufsz = adapter->rx_buffer_len;
161
162         i = rx_ring->next_to_use;
163         buffer_info = &rx_ring->buffer_info[i];
164
165         while (cleaned_count--) {
166                 skb = buffer_info->skb;
167                 if (skb) {
168                         skb_trim(skb, 0);
169                         goto map_skb;
170                 }
171
172                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
173                 if (!skb) {
174                         /* Better luck next round */
175                         adapter->alloc_rx_buff_failed++;
176                         break;
177                 }
178
179                 buffer_info->skb = skb;
180 map_skb:
181                 buffer_info->dma = pci_map_single(pdev, skb->data,
182                                                   adapter->rx_buffer_len,
183                                                   PCI_DMA_FROMDEVICE);
184                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
185                         dev_err(&pdev->dev, "RX DMA map failed\n");
186                         adapter->rx_dma_failed++;
187                         break;
188                 }
189
190                 rx_desc = E1000_RX_DESC(*rx_ring, i);
191                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
192
193                 i++;
194                 if (i == rx_ring->count)
195                         i = 0;
196                 buffer_info = &rx_ring->buffer_info[i];
197         }
198
199         if (rx_ring->next_to_use != i) {
200                 rx_ring->next_to_use = i;
201                 if (i-- == 0)
202                         i = (rx_ring->count - 1);
203
204                 /*
205                  * Force memory writes to complete before letting h/w
206                  * know there are new descriptors to fetch.  (Only
207                  * applicable for weak-ordered memory model archs,
208                  * such as IA-64).
209                  */
210                 wmb();
211                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
212         }
213 }
214
215 /**
216  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
217  * @adapter: address of board private structure
218  **/
219 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
220                                       int cleaned_count)
221 {
222         struct net_device *netdev = adapter->netdev;
223         struct pci_dev *pdev = adapter->pdev;
224         union e1000_rx_desc_packet_split *rx_desc;
225         struct e1000_ring *rx_ring = adapter->rx_ring;
226         struct e1000_buffer *buffer_info;
227         struct e1000_ps_page *ps_page;
228         struct sk_buff *skb;
229         unsigned int i, j;
230
231         i = rx_ring->next_to_use;
232         buffer_info = &rx_ring->buffer_info[i];
233
234         while (cleaned_count--) {
235                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
236
237                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
238                         ps_page = &buffer_info->ps_pages[j];
239                         if (j >= adapter->rx_ps_pages) {
240                                 /* all unused desc entries get hw null ptr */
241                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
242                                 continue;
243                         }
244                         if (!ps_page->page) {
245                                 ps_page->page = alloc_page(GFP_ATOMIC);
246                                 if (!ps_page->page) {
247                                         adapter->alloc_rx_buff_failed++;
248                                         goto no_buffers;
249                                 }
250                                 ps_page->dma = pci_map_page(pdev,
251                                                    ps_page->page,
252                                                    0, PAGE_SIZE,
253                                                    PCI_DMA_FROMDEVICE);
254                                 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
255                                         dev_err(&adapter->pdev->dev,
256                                           "RX DMA page map failed\n");
257                                         adapter->rx_dma_failed++;
258                                         goto no_buffers;
259                                 }
260                         }
261                         /*
262                          * Refresh the desc even if buffer_addrs
263                          * didn't change because each write-back
264                          * erases this info.
265                          */
266                         rx_desc->read.buffer_addr[j+1] =
267                              cpu_to_le64(ps_page->dma);
268                 }
269
270                 skb = netdev_alloc_skb_ip_align(netdev,
271                                                 adapter->rx_ps_bsize0);
272
273                 if (!skb) {
274                         adapter->alloc_rx_buff_failed++;
275                         break;
276                 }
277
278                 buffer_info->skb = skb;
279                 buffer_info->dma = pci_map_single(pdev, skb->data,
280                                                   adapter->rx_ps_bsize0,
281                                                   PCI_DMA_FROMDEVICE);
282                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
283                         dev_err(&pdev->dev, "RX DMA map failed\n");
284                         adapter->rx_dma_failed++;
285                         /* cleanup skb */
286                         dev_kfree_skb_any(skb);
287                         buffer_info->skb = NULL;
288                         break;
289                 }
290
291                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
292
293                 i++;
294                 if (i == rx_ring->count)
295                         i = 0;
296                 buffer_info = &rx_ring->buffer_info[i];
297         }
298
299 no_buffers:
300         if (rx_ring->next_to_use != i) {
301                 rx_ring->next_to_use = i;
302
303                 if (!(i--))
304                         i = (rx_ring->count - 1);
305
306                 /*
307                  * Force memory writes to complete before letting h/w
308                  * know there are new descriptors to fetch.  (Only
309                  * applicable for weak-ordered memory model archs,
310                  * such as IA-64).
311                  */
312                 wmb();
313                 /*
314                  * Hardware increments by 16 bytes, but packet split
315                  * descriptors are 32 bytes...so we increment tail
316                  * twice as much.
317                  */
318                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
319         }
320 }
321
322 /**
323  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
324  * @adapter: address of board private structure
325  * @cleaned_count: number of buffers to allocate this pass
326  **/
327
328 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
329                                          int cleaned_count)
330 {
331         struct net_device *netdev = adapter->netdev;
332         struct pci_dev *pdev = adapter->pdev;
333         struct e1000_rx_desc *rx_desc;
334         struct e1000_ring *rx_ring = adapter->rx_ring;
335         struct e1000_buffer *buffer_info;
336         struct sk_buff *skb;
337         unsigned int i;
338         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
339
340         i = rx_ring->next_to_use;
341         buffer_info = &rx_ring->buffer_info[i];
342
343         while (cleaned_count--) {
344                 skb = buffer_info->skb;
345                 if (skb) {
346                         skb_trim(skb, 0);
347                         goto check_page;
348                 }
349
350                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
351                 if (unlikely(!skb)) {
352                         /* Better luck next round */
353                         adapter->alloc_rx_buff_failed++;
354                         break;
355                 }
356
357                 buffer_info->skb = skb;
358 check_page:
359                 /* allocate a new page if necessary */
360                 if (!buffer_info->page) {
361                         buffer_info->page = alloc_page(GFP_ATOMIC);
362                         if (unlikely(!buffer_info->page)) {
363                                 adapter->alloc_rx_buff_failed++;
364                                 break;
365                         }
366                 }
367
368                 if (!buffer_info->dma)
369                         buffer_info->dma = pci_map_page(pdev,
370                                                         buffer_info->page, 0,
371                                                         PAGE_SIZE,
372                                                         PCI_DMA_FROMDEVICE);
373
374                 rx_desc = E1000_RX_DESC(*rx_ring, i);
375                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
376
377                 if (unlikely(++i == rx_ring->count))
378                         i = 0;
379                 buffer_info = &rx_ring->buffer_info[i];
380         }
381
382         if (likely(rx_ring->next_to_use != i)) {
383                 rx_ring->next_to_use = i;
384                 if (unlikely(i-- == 0))
385                         i = (rx_ring->count - 1);
386
387                 /* Force memory writes to complete before letting h/w
388                  * know there are new descriptors to fetch.  (Only
389                  * applicable for weak-ordered memory model archs,
390                  * such as IA-64). */
391                 wmb();
392                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
393         }
394 }
395
396 /**
397  * e1000_clean_rx_irq - Send received data up the network stack; legacy
398  * @adapter: board private structure
399  *
400  * the return value indicates whether actual cleaning was done, there
401  * is no guarantee that everything was cleaned
402  **/
403 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
404                                int *work_done, int work_to_do)
405 {
406         struct net_device *netdev = adapter->netdev;
407         struct pci_dev *pdev = adapter->pdev;
408         struct e1000_hw *hw = &adapter->hw;
409         struct e1000_ring *rx_ring = adapter->rx_ring;
410         struct e1000_rx_desc *rx_desc, *next_rxd;
411         struct e1000_buffer *buffer_info, *next_buffer;
412         u32 length;
413         unsigned int i;
414         int cleaned_count = 0;
415         bool cleaned = 0;
416         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
417
418         i = rx_ring->next_to_clean;
419         rx_desc = E1000_RX_DESC(*rx_ring, i);
420         buffer_info = &rx_ring->buffer_info[i];
421
422         while (rx_desc->status & E1000_RXD_STAT_DD) {
423                 struct sk_buff *skb;
424                 u8 status;
425
426                 if (*work_done >= work_to_do)
427                         break;
428                 (*work_done)++;
429
430                 status = rx_desc->status;
431                 skb = buffer_info->skb;
432                 buffer_info->skb = NULL;
433
434                 prefetch(skb->data - NET_IP_ALIGN);
435
436                 i++;
437                 if (i == rx_ring->count)
438                         i = 0;
439                 next_rxd = E1000_RX_DESC(*rx_ring, i);
440                 prefetch(next_rxd);
441
442                 next_buffer = &rx_ring->buffer_info[i];
443
444                 cleaned = 1;
445                 cleaned_count++;
446                 pci_unmap_single(pdev,
447                                  buffer_info->dma,
448                                  adapter->rx_buffer_len,
449                                  PCI_DMA_FROMDEVICE);
450                 buffer_info->dma = 0;
451
452                 length = le16_to_cpu(rx_desc->length);
453
454                 /*
455                  * !EOP means multiple descriptors were used to store a single
456                  * packet, if that's the case we need to toss it.  In fact, we
457                  * need to toss every packet with the EOP bit clear and the
458                  * next frame that _does_ have the EOP bit set, as it is by
459                  * definition only a frame fragment
460                  */
461                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
462                         adapter->flags2 |= FLAG2_IS_DISCARDING;
463
464                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
465                         /* All receives must fit into a single buffer */
466                         e_dbg("Receive packet consumed multiple buffers\n");
467                         /* recycle */
468                         buffer_info->skb = skb;
469                         if (status & E1000_RXD_STAT_EOP)
470                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
471                         goto next_desc;
472                 }
473
474                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
475                         /* recycle */
476                         buffer_info->skb = skb;
477                         goto next_desc;
478                 }
479
480                 /* adjust length to remove Ethernet CRC */
481                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
482                         length -= 4;
483
484                 total_rx_bytes += length;
485                 total_rx_packets++;
486
487                 /*
488                  * code added for copybreak, this should improve
489                  * performance for small packets with large amounts
490                  * of reassembly being done in the stack
491                  */
492                 if (length < copybreak) {
493                         struct sk_buff *new_skb =
494                             netdev_alloc_skb_ip_align(netdev, length);
495                         if (new_skb) {
496                                 skb_copy_to_linear_data_offset(new_skb,
497                                                                -NET_IP_ALIGN,
498                                                                (skb->data -
499                                                                 NET_IP_ALIGN),
500                                                                (length +
501                                                                 NET_IP_ALIGN));
502                                 /* save the skb in buffer_info as good */
503                                 buffer_info->skb = skb;
504                                 skb = new_skb;
505                         }
506                         /* else just continue with the old one */
507                 }
508                 /* end copybreak code */
509                 skb_put(skb, length);
510
511                 /* Receive Checksum Offload */
512                 e1000_rx_checksum(adapter,
513                                   (u32)(status) |
514                                   ((u32)(rx_desc->errors) << 24),
515                                   le16_to_cpu(rx_desc->csum), skb);
516
517                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
518
519 next_desc:
520                 rx_desc->status = 0;
521
522                 /* return some buffers to hardware, one at a time is too slow */
523                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
524                         adapter->alloc_rx_buf(adapter, cleaned_count);
525                         cleaned_count = 0;
526                 }
527
528                 /* use prefetched values */
529                 rx_desc = next_rxd;
530                 buffer_info = next_buffer;
531         }
532         rx_ring->next_to_clean = i;
533
534         cleaned_count = e1000_desc_unused(rx_ring);
535         if (cleaned_count)
536                 adapter->alloc_rx_buf(adapter, cleaned_count);
537
538         adapter->total_rx_bytes += total_rx_bytes;
539         adapter->total_rx_packets += total_rx_packets;
540         netdev->stats.rx_bytes += total_rx_bytes;
541         netdev->stats.rx_packets += total_rx_packets;
542         return cleaned;
543 }
544
545 static void e1000_put_txbuf(struct e1000_adapter *adapter,
546                              struct e1000_buffer *buffer_info)
547 {
548         if (buffer_info->dma) {
549                 if (buffer_info->mapped_as_page)
550                         pci_unmap_page(adapter->pdev, buffer_info->dma,
551                                        buffer_info->length, PCI_DMA_TODEVICE);
552                 else
553                         pci_unmap_single(adapter->pdev, buffer_info->dma,
554                                          buffer_info->length,
555                                          PCI_DMA_TODEVICE);
556                 buffer_info->dma = 0;
557         }
558         if (buffer_info->skb) {
559                 dev_kfree_skb_any(buffer_info->skb);
560                 buffer_info->skb = NULL;
561         }
562         buffer_info->time_stamp = 0;
563 }
564
565 static void e1000_print_hw_hang(struct work_struct *work)
566 {
567         struct e1000_adapter *adapter = container_of(work,
568                                                      struct e1000_adapter,
569                                                      print_hang_task);
570         struct e1000_ring *tx_ring = adapter->tx_ring;
571         unsigned int i = tx_ring->next_to_clean;
572         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
573         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
574         struct e1000_hw *hw = &adapter->hw;
575         u16 phy_status, phy_1000t_status, phy_ext_status;
576         u16 pci_status;
577
578         e1e_rphy(hw, PHY_STATUS, &phy_status);
579         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
580         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
581
582         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
583
584         /* detected Hardware unit hang */
585         e_err("Detected Hardware Unit Hang:\n"
586               "  TDH                  <%x>\n"
587               "  TDT                  <%x>\n"
588               "  next_to_use          <%x>\n"
589               "  next_to_clean        <%x>\n"
590               "buffer_info[next_to_clean]:\n"
591               "  time_stamp           <%lx>\n"
592               "  next_to_watch        <%x>\n"
593               "  jiffies              <%lx>\n"
594               "  next_to_watch.status <%x>\n"
595               "MAC Status             <%x>\n"
596               "PHY Status             <%x>\n"
597               "PHY 1000BASE-T Status  <%x>\n"
598               "PHY Extended Status    <%x>\n"
599               "PCI Status             <%x>\n",
600               readl(adapter->hw.hw_addr + tx_ring->head),
601               readl(adapter->hw.hw_addr + tx_ring->tail),
602               tx_ring->next_to_use,
603               tx_ring->next_to_clean,
604               tx_ring->buffer_info[eop].time_stamp,
605               eop,
606               jiffies,
607               eop_desc->upper.fields.status,
608               er32(STATUS),
609               phy_status,
610               phy_1000t_status,
611               phy_ext_status,
612               pci_status);
613 }
614
615 /**
616  * e1000_clean_tx_irq - Reclaim resources after transmit completes
617  * @adapter: board private structure
618  *
619  * the return value indicates whether actual cleaning was done, there
620  * is no guarantee that everything was cleaned
621  **/
622 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
623 {
624         struct net_device *netdev = adapter->netdev;
625         struct e1000_hw *hw = &adapter->hw;
626         struct e1000_ring *tx_ring = adapter->tx_ring;
627         struct e1000_tx_desc *tx_desc, *eop_desc;
628         struct e1000_buffer *buffer_info;
629         unsigned int i, eop;
630         unsigned int count = 0;
631         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
632
633         i = tx_ring->next_to_clean;
634         eop = tx_ring->buffer_info[i].next_to_watch;
635         eop_desc = E1000_TX_DESC(*tx_ring, eop);
636
637         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
638                (count < tx_ring->count)) {
639                 bool cleaned = false;
640                 for (; !cleaned; count++) {
641                         tx_desc = E1000_TX_DESC(*tx_ring, i);
642                         buffer_info = &tx_ring->buffer_info[i];
643                         cleaned = (i == eop);
644
645                         if (cleaned) {
646                                 struct sk_buff *skb = buffer_info->skb;
647                                 unsigned int segs, bytecount;
648                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
649                                 /* multiply data chunks by size of headers */
650                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
651                                             skb->len;
652                                 total_tx_packets += segs;
653                                 total_tx_bytes += bytecount;
654                         }
655
656                         e1000_put_txbuf(adapter, buffer_info);
657                         tx_desc->upper.data = 0;
658
659                         i++;
660                         if (i == tx_ring->count)
661                                 i = 0;
662                 }
663
664                 eop = tx_ring->buffer_info[i].next_to_watch;
665                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
666         }
667
668         tx_ring->next_to_clean = i;
669
670 #define TX_WAKE_THRESHOLD 32
671         if (count && netif_carrier_ok(netdev) &&
672             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
673                 /* Make sure that anybody stopping the queue after this
674                  * sees the new next_to_clean.
675                  */
676                 smp_mb();
677
678                 if (netif_queue_stopped(netdev) &&
679                     !(test_bit(__E1000_DOWN, &adapter->state))) {
680                         netif_wake_queue(netdev);
681                         ++adapter->restart_queue;
682                 }
683         }
684
685         if (adapter->detect_tx_hung) {
686                 /*
687                  * Detect a transmit hang in hardware, this serializes the
688                  * check with the clearing of time_stamp and movement of i
689                  */
690                 adapter->detect_tx_hung = 0;
691                 if (tx_ring->buffer_info[i].time_stamp &&
692                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
693                                + (adapter->tx_timeout_factor * HZ)) &&
694                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
695                         schedule_work(&adapter->print_hang_task);
696                         netif_stop_queue(netdev);
697                 }
698         }
699         adapter->total_tx_bytes += total_tx_bytes;
700         adapter->total_tx_packets += total_tx_packets;
701         netdev->stats.tx_bytes += total_tx_bytes;
702         netdev->stats.tx_packets += total_tx_packets;
703         return (count < tx_ring->count);
704 }
705
706 /**
707  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
708  * @adapter: board private structure
709  *
710  * the return value indicates whether actual cleaning was done, there
711  * is no guarantee that everything was cleaned
712  **/
713 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
714                                   int *work_done, int work_to_do)
715 {
716         struct e1000_hw *hw = &adapter->hw;
717         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
718         struct net_device *netdev = adapter->netdev;
719         struct pci_dev *pdev = adapter->pdev;
720         struct e1000_ring *rx_ring = adapter->rx_ring;
721         struct e1000_buffer *buffer_info, *next_buffer;
722         struct e1000_ps_page *ps_page;
723         struct sk_buff *skb;
724         unsigned int i, j;
725         u32 length, staterr;
726         int cleaned_count = 0;
727         bool cleaned = 0;
728         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
729
730         i = rx_ring->next_to_clean;
731         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
732         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
733         buffer_info = &rx_ring->buffer_info[i];
734
735         while (staterr & E1000_RXD_STAT_DD) {
736                 if (*work_done >= work_to_do)
737                         break;
738                 (*work_done)++;
739                 skb = buffer_info->skb;
740
741                 /* in the packet split case this is header only */
742                 prefetch(skb->data - NET_IP_ALIGN);
743
744                 i++;
745                 if (i == rx_ring->count)
746                         i = 0;
747                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
748                 prefetch(next_rxd);
749
750                 next_buffer = &rx_ring->buffer_info[i];
751
752                 cleaned = 1;
753                 cleaned_count++;
754                 pci_unmap_single(pdev, buffer_info->dma,
755                                  adapter->rx_ps_bsize0,
756                                  PCI_DMA_FROMDEVICE);
757                 buffer_info->dma = 0;
758
759                 /* see !EOP comment in other rx routine */
760                 if (!(staterr & E1000_RXD_STAT_EOP))
761                         adapter->flags2 |= FLAG2_IS_DISCARDING;
762
763                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
764                         e_dbg("Packet Split buffers didn't pick up the full "
765                               "packet\n");
766                         dev_kfree_skb_irq(skb);
767                         if (staterr & E1000_RXD_STAT_EOP)
768                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
769                         goto next_desc;
770                 }
771
772                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
773                         dev_kfree_skb_irq(skb);
774                         goto next_desc;
775                 }
776
777                 length = le16_to_cpu(rx_desc->wb.middle.length0);
778
779                 if (!length) {
780                         e_dbg("Last part of the packet spanning multiple "
781                               "descriptors\n");
782                         dev_kfree_skb_irq(skb);
783                         goto next_desc;
784                 }
785
786                 /* Good Receive */
787                 skb_put(skb, length);
788
789                 {
790                 /*
791                  * this looks ugly, but it seems compiler issues make it
792                  * more efficient than reusing j
793                  */
794                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
795
796                 /*
797                  * page alloc/put takes too long and effects small packet
798                  * throughput, so unsplit small packets and save the alloc/put
799                  * only valid in softirq (napi) context to call kmap_*
800                  */
801                 if (l1 && (l1 <= copybreak) &&
802                     ((length + l1) <= adapter->rx_ps_bsize0)) {
803                         u8 *vaddr;
804
805                         ps_page = &buffer_info->ps_pages[0];
806
807                         /*
808                          * there is no documentation about how to call
809                          * kmap_atomic, so we can't hold the mapping
810                          * very long
811                          */
812                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
813                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
814                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
815                         memcpy(skb_tail_pointer(skb), vaddr, l1);
816                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
817                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
818                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
819
820                         /* remove the CRC */
821                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
822                                 l1 -= 4;
823
824                         skb_put(skb, l1);
825                         goto copydone;
826                 } /* if */
827                 }
828
829                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
830                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
831                         if (!length)
832                                 break;
833
834                         ps_page = &buffer_info->ps_pages[j];
835                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
836                                        PCI_DMA_FROMDEVICE);
837                         ps_page->dma = 0;
838                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
839                         ps_page->page = NULL;
840                         skb->len += length;
841                         skb->data_len += length;
842                         skb->truesize += length;
843                 }
844
845                 /* strip the ethernet crc, problem is we're using pages now so
846                  * this whole operation can get a little cpu intensive
847                  */
848                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
849                         pskb_trim(skb, skb->len - 4);
850
851 copydone:
852                 total_rx_bytes += skb->len;
853                 total_rx_packets++;
854
855                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
856                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
857
858                 if (rx_desc->wb.upper.header_status &
859                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
860                         adapter->rx_hdr_split++;
861
862                 e1000_receive_skb(adapter, netdev, skb,
863                                   staterr, rx_desc->wb.middle.vlan);
864
865 next_desc:
866                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
867                 buffer_info->skb = NULL;
868
869                 /* return some buffers to hardware, one at a time is too slow */
870                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
871                         adapter->alloc_rx_buf(adapter, cleaned_count);
872                         cleaned_count = 0;
873                 }
874
875                 /* use prefetched values */
876                 rx_desc = next_rxd;
877                 buffer_info = next_buffer;
878
879                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
880         }
881         rx_ring->next_to_clean = i;
882
883         cleaned_count = e1000_desc_unused(rx_ring);
884         if (cleaned_count)
885                 adapter->alloc_rx_buf(adapter, cleaned_count);
886
887         adapter->total_rx_bytes += total_rx_bytes;
888         adapter->total_rx_packets += total_rx_packets;
889         netdev->stats.rx_bytes += total_rx_bytes;
890         netdev->stats.rx_packets += total_rx_packets;
891         return cleaned;
892 }
893
894 /**
895  * e1000_consume_page - helper function
896  **/
897 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
898                                u16 length)
899 {
900         bi->page = NULL;
901         skb->len += length;
902         skb->data_len += length;
903         skb->truesize += length;
904 }
905
906 /**
907  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
908  * @adapter: board private structure
909  *
910  * the return value indicates whether actual cleaning was done, there
911  * is no guarantee that everything was cleaned
912  **/
913
914 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
915                                      int *work_done, int work_to_do)
916 {
917         struct net_device *netdev = adapter->netdev;
918         struct pci_dev *pdev = adapter->pdev;
919         struct e1000_ring *rx_ring = adapter->rx_ring;
920         struct e1000_rx_desc *rx_desc, *next_rxd;
921         struct e1000_buffer *buffer_info, *next_buffer;
922         u32 length;
923         unsigned int i;
924         int cleaned_count = 0;
925         bool cleaned = false;
926         unsigned int total_rx_bytes=0, total_rx_packets=0;
927
928         i = rx_ring->next_to_clean;
929         rx_desc = E1000_RX_DESC(*rx_ring, i);
930         buffer_info = &rx_ring->buffer_info[i];
931
932         while (rx_desc->status & E1000_RXD_STAT_DD) {
933                 struct sk_buff *skb;
934                 u8 status;
935
936                 if (*work_done >= work_to_do)
937                         break;
938                 (*work_done)++;
939
940                 status = rx_desc->status;
941                 skb = buffer_info->skb;
942                 buffer_info->skb = NULL;
943
944                 ++i;
945                 if (i == rx_ring->count)
946                         i = 0;
947                 next_rxd = E1000_RX_DESC(*rx_ring, i);
948                 prefetch(next_rxd);
949
950                 next_buffer = &rx_ring->buffer_info[i];
951
952                 cleaned = true;
953                 cleaned_count++;
954                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
955                                PCI_DMA_FROMDEVICE);
956                 buffer_info->dma = 0;
957
958                 length = le16_to_cpu(rx_desc->length);
959
960                 /* errors is only valid for DD + EOP descriptors */
961                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
962                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
963                                 /* recycle both page and skb */
964                                 buffer_info->skb = skb;
965                                 /* an error means any chain goes out the window
966                                  * too */
967                                 if (rx_ring->rx_skb_top)
968                                         dev_kfree_skb(rx_ring->rx_skb_top);
969                                 rx_ring->rx_skb_top = NULL;
970                                 goto next_desc;
971                 }
972
973 #define rxtop rx_ring->rx_skb_top
974                 if (!(status & E1000_RXD_STAT_EOP)) {
975                         /* this descriptor is only the beginning (or middle) */
976                         if (!rxtop) {
977                                 /* this is the beginning of a chain */
978                                 rxtop = skb;
979                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
980                                                    0, length);
981                         } else {
982                                 /* this is the middle of a chain */
983                                 skb_fill_page_desc(rxtop,
984                                     skb_shinfo(rxtop)->nr_frags,
985                                     buffer_info->page, 0, length);
986                                 /* re-use the skb, only consumed the page */
987                                 buffer_info->skb = skb;
988                         }
989                         e1000_consume_page(buffer_info, rxtop, length);
990                         goto next_desc;
991                 } else {
992                         if (rxtop) {
993                                 /* end of the chain */
994                                 skb_fill_page_desc(rxtop,
995                                     skb_shinfo(rxtop)->nr_frags,
996                                     buffer_info->page, 0, length);
997                                 /* re-use the current skb, we only consumed the
998                                  * page */
999                                 buffer_info->skb = skb;
1000                                 skb = rxtop;
1001                                 rxtop = NULL;
1002                                 e1000_consume_page(buffer_info, skb, length);
1003                         } else {
1004                                 /* no chain, got EOP, this buf is the packet
1005                                  * copybreak to save the put_page/alloc_page */
1006                                 if (length <= copybreak &&
1007                                     skb_tailroom(skb) >= length) {
1008                                         u8 *vaddr;
1009                                         vaddr = kmap_atomic(buffer_info->page,
1010                                                            KM_SKB_DATA_SOFTIRQ);
1011                                         memcpy(skb_tail_pointer(skb), vaddr,
1012                                                length);
1013                                         kunmap_atomic(vaddr,
1014                                                       KM_SKB_DATA_SOFTIRQ);
1015                                         /* re-use the page, so don't erase
1016                                          * buffer_info->page */
1017                                         skb_put(skb, length);
1018                                 } else {
1019                                         skb_fill_page_desc(skb, 0,
1020                                                            buffer_info->page, 0,
1021                                                            length);
1022                                         e1000_consume_page(buffer_info, skb,
1023                                                            length);
1024                                 }
1025                         }
1026                 }
1027
1028                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1029                 e1000_rx_checksum(adapter,
1030                                   (u32)(status) |
1031                                   ((u32)(rx_desc->errors) << 24),
1032                                   le16_to_cpu(rx_desc->csum), skb);
1033
1034                 /* probably a little skewed due to removing CRC */
1035                 total_rx_bytes += skb->len;
1036                 total_rx_packets++;
1037
1038                 /* eth type trans needs skb->data to point to something */
1039                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1040                         e_err("pskb_may_pull failed.\n");
1041                         dev_kfree_skb(skb);
1042                         goto next_desc;
1043                 }
1044
1045                 e1000_receive_skb(adapter, netdev, skb, status,
1046                                   rx_desc->special);
1047
1048 next_desc:
1049                 rx_desc->status = 0;
1050
1051                 /* return some buffers to hardware, one at a time is too slow */
1052                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1053                         adapter->alloc_rx_buf(adapter, cleaned_count);
1054                         cleaned_count = 0;
1055                 }
1056
1057                 /* use prefetched values */
1058                 rx_desc = next_rxd;
1059                 buffer_info = next_buffer;
1060         }
1061         rx_ring->next_to_clean = i;
1062
1063         cleaned_count = e1000_desc_unused(rx_ring);
1064         if (cleaned_count)
1065                 adapter->alloc_rx_buf(adapter, cleaned_count);
1066
1067         adapter->total_rx_bytes += total_rx_bytes;
1068         adapter->total_rx_packets += total_rx_packets;
1069         netdev->stats.rx_bytes += total_rx_bytes;
1070         netdev->stats.rx_packets += total_rx_packets;
1071         return cleaned;
1072 }
1073
1074 /**
1075  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1076  * @adapter: board private structure
1077  **/
1078 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1079 {
1080         struct e1000_ring *rx_ring = adapter->rx_ring;
1081         struct e1000_buffer *buffer_info;
1082         struct e1000_ps_page *ps_page;
1083         struct pci_dev *pdev = adapter->pdev;
1084         unsigned int i, j;
1085
1086         /* Free all the Rx ring sk_buffs */
1087         for (i = 0; i < rx_ring->count; i++) {
1088                 buffer_info = &rx_ring->buffer_info[i];
1089                 if (buffer_info->dma) {
1090                         if (adapter->clean_rx == e1000_clean_rx_irq)
1091                                 pci_unmap_single(pdev, buffer_info->dma,
1092                                                  adapter->rx_buffer_len,
1093                                                  PCI_DMA_FROMDEVICE);
1094                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1095                                 pci_unmap_page(pdev, buffer_info->dma,
1096                                                PAGE_SIZE,
1097                                                PCI_DMA_FROMDEVICE);
1098                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1099                                 pci_unmap_single(pdev, buffer_info->dma,
1100                                                  adapter->rx_ps_bsize0,
1101                                                  PCI_DMA_FROMDEVICE);
1102                         buffer_info->dma = 0;
1103                 }
1104
1105                 if (buffer_info->page) {
1106                         put_page(buffer_info->page);
1107                         buffer_info->page = NULL;
1108                 }
1109
1110                 if (buffer_info->skb) {
1111                         dev_kfree_skb(buffer_info->skb);
1112                         buffer_info->skb = NULL;
1113                 }
1114
1115                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1116                         ps_page = &buffer_info->ps_pages[j];
1117                         if (!ps_page->page)
1118                                 break;
1119                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1120                                        PCI_DMA_FROMDEVICE);
1121                         ps_page->dma = 0;
1122                         put_page(ps_page->page);
1123                         ps_page->page = NULL;
1124                 }
1125         }
1126
1127         /* there also may be some cached data from a chained receive */
1128         if (rx_ring->rx_skb_top) {
1129                 dev_kfree_skb(rx_ring->rx_skb_top);
1130                 rx_ring->rx_skb_top = NULL;
1131         }
1132
1133         /* Zero out the descriptor ring */
1134         memset(rx_ring->desc, 0, rx_ring->size);
1135
1136         rx_ring->next_to_clean = 0;
1137         rx_ring->next_to_use = 0;
1138         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1139
1140         writel(0, adapter->hw.hw_addr + rx_ring->head);
1141         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1142 }
1143
1144 static void e1000e_downshift_workaround(struct work_struct *work)
1145 {
1146         struct e1000_adapter *adapter = container_of(work,
1147                                         struct e1000_adapter, downshift_task);
1148
1149         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1150 }
1151
1152 /**
1153  * e1000_intr_msi - Interrupt Handler
1154  * @irq: interrupt number
1155  * @data: pointer to a network interface device structure
1156  **/
1157 static irqreturn_t e1000_intr_msi(int irq, void *data)
1158 {
1159         struct net_device *netdev = data;
1160         struct e1000_adapter *adapter = netdev_priv(netdev);
1161         struct e1000_hw *hw = &adapter->hw;
1162         u32 icr = er32(ICR);
1163
1164         /*
1165          * read ICR disables interrupts using IAM
1166          */
1167
1168         if (icr & E1000_ICR_LSC) {
1169                 hw->mac.get_link_status = 1;
1170                 /*
1171                  * ICH8 workaround-- Call gig speed drop workaround on cable
1172                  * disconnect (LSC) before accessing any PHY registers
1173                  */
1174                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1175                     (!(er32(STATUS) & E1000_STATUS_LU)))
1176                         schedule_work(&adapter->downshift_task);
1177
1178                 /*
1179                  * 80003ES2LAN workaround-- For packet buffer work-around on
1180                  * link down event; disable receives here in the ISR and reset
1181                  * adapter in watchdog
1182                  */
1183                 if (netif_carrier_ok(netdev) &&
1184                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1185                         /* disable receives */
1186                         u32 rctl = er32(RCTL);
1187                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1188                         adapter->flags |= FLAG_RX_RESTART_NOW;
1189                 }
1190                 /* guard against interrupt when we're going down */
1191                 if (!test_bit(__E1000_DOWN, &adapter->state))
1192                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1193         }
1194
1195         if (napi_schedule_prep(&adapter->napi)) {
1196                 adapter->total_tx_bytes = 0;
1197                 adapter->total_tx_packets = 0;
1198                 adapter->total_rx_bytes = 0;
1199                 adapter->total_rx_packets = 0;
1200                 __napi_schedule(&adapter->napi);
1201         }
1202
1203         return IRQ_HANDLED;
1204 }
1205
1206 /**
1207  * e1000_intr - Interrupt Handler
1208  * @irq: interrupt number
1209  * @data: pointer to a network interface device structure
1210  **/
1211 static irqreturn_t e1000_intr(int irq, void *data)
1212 {
1213         struct net_device *netdev = data;
1214         struct e1000_adapter *adapter = netdev_priv(netdev);
1215         struct e1000_hw *hw = &adapter->hw;
1216         u32 rctl, icr = er32(ICR);
1217
1218         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1219                 return IRQ_NONE;  /* Not our interrupt */
1220
1221         /*
1222          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1223          * not set, then the adapter didn't send an interrupt
1224          */
1225         if (!(icr & E1000_ICR_INT_ASSERTED))
1226                 return IRQ_NONE;
1227
1228         /*
1229          * Interrupt Auto-Mask...upon reading ICR,
1230          * interrupts are masked.  No need for the
1231          * IMC write
1232          */
1233
1234         if (icr & E1000_ICR_LSC) {
1235                 hw->mac.get_link_status = 1;
1236                 /*
1237                  * ICH8 workaround-- Call gig speed drop workaround on cable
1238                  * disconnect (LSC) before accessing any PHY registers
1239                  */
1240                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1241                     (!(er32(STATUS) & E1000_STATUS_LU)))
1242                         schedule_work(&adapter->downshift_task);
1243
1244                 /*
1245                  * 80003ES2LAN workaround--
1246                  * For packet buffer work-around on link down event;
1247                  * disable receives here in the ISR and
1248                  * reset adapter in watchdog
1249                  */
1250                 if (netif_carrier_ok(netdev) &&
1251                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1252                         /* disable receives */
1253                         rctl = er32(RCTL);
1254                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1255                         adapter->flags |= FLAG_RX_RESTART_NOW;
1256                 }
1257                 /* guard against interrupt when we're going down */
1258                 if (!test_bit(__E1000_DOWN, &adapter->state))
1259                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1260         }
1261
1262         if (napi_schedule_prep(&adapter->napi)) {
1263                 adapter->total_tx_bytes = 0;
1264                 adapter->total_tx_packets = 0;
1265                 adapter->total_rx_bytes = 0;
1266                 adapter->total_rx_packets = 0;
1267                 __napi_schedule(&adapter->napi);
1268         }
1269
1270         return IRQ_HANDLED;
1271 }
1272
1273 static irqreturn_t e1000_msix_other(int irq, void *data)
1274 {
1275         struct net_device *netdev = data;
1276         struct e1000_adapter *adapter = netdev_priv(netdev);
1277         struct e1000_hw *hw = &adapter->hw;
1278         u32 icr = er32(ICR);
1279
1280         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1281                 if (!test_bit(__E1000_DOWN, &adapter->state))
1282                         ew32(IMS, E1000_IMS_OTHER);
1283                 return IRQ_NONE;
1284         }
1285
1286         if (icr & adapter->eiac_mask)
1287                 ew32(ICS, (icr & adapter->eiac_mask));
1288
1289         if (icr & E1000_ICR_OTHER) {
1290                 if (!(icr & E1000_ICR_LSC))
1291                         goto no_link_interrupt;
1292                 hw->mac.get_link_status = 1;
1293                 /* guard against interrupt when we're going down */
1294                 if (!test_bit(__E1000_DOWN, &adapter->state))
1295                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1296         }
1297
1298 no_link_interrupt:
1299         if (!test_bit(__E1000_DOWN, &adapter->state))
1300                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1301
1302         return IRQ_HANDLED;
1303 }
1304
1305
1306 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1307 {
1308         struct net_device *netdev = data;
1309         struct e1000_adapter *adapter = netdev_priv(netdev);
1310         struct e1000_hw *hw = &adapter->hw;
1311         struct e1000_ring *tx_ring = adapter->tx_ring;
1312
1313
1314         adapter->total_tx_bytes = 0;
1315         adapter->total_tx_packets = 0;
1316
1317         if (!e1000_clean_tx_irq(adapter))
1318                 /* Ring was not completely cleaned, so fire another interrupt */
1319                 ew32(ICS, tx_ring->ims_val);
1320
1321         return IRQ_HANDLED;
1322 }
1323
1324 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1325 {
1326         struct net_device *netdev = data;
1327         struct e1000_adapter *adapter = netdev_priv(netdev);
1328
1329         /* Write the ITR value calculated at the end of the
1330          * previous interrupt.
1331          */
1332         if (adapter->rx_ring->set_itr) {
1333                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1334                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1335                 adapter->rx_ring->set_itr = 0;
1336         }
1337
1338         if (napi_schedule_prep(&adapter->napi)) {
1339                 adapter->total_rx_bytes = 0;
1340                 adapter->total_rx_packets = 0;
1341                 __napi_schedule(&adapter->napi);
1342         }
1343         return IRQ_HANDLED;
1344 }
1345
1346 /**
1347  * e1000_configure_msix - Configure MSI-X hardware
1348  *
1349  * e1000_configure_msix sets up the hardware to properly
1350  * generate MSI-X interrupts.
1351  **/
1352 static void e1000_configure_msix(struct e1000_adapter *adapter)
1353 {
1354         struct e1000_hw *hw = &adapter->hw;
1355         struct e1000_ring *rx_ring = adapter->rx_ring;
1356         struct e1000_ring *tx_ring = adapter->tx_ring;
1357         int vector = 0;
1358         u32 ctrl_ext, ivar = 0;
1359
1360         adapter->eiac_mask = 0;
1361
1362         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1363         if (hw->mac.type == e1000_82574) {
1364                 u32 rfctl = er32(RFCTL);
1365                 rfctl |= E1000_RFCTL_ACK_DIS;
1366                 ew32(RFCTL, rfctl);
1367         }
1368
1369 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1370         /* Configure Rx vector */
1371         rx_ring->ims_val = E1000_IMS_RXQ0;
1372         adapter->eiac_mask |= rx_ring->ims_val;
1373         if (rx_ring->itr_val)
1374                 writel(1000000000 / (rx_ring->itr_val * 256),
1375                        hw->hw_addr + rx_ring->itr_register);
1376         else
1377                 writel(1, hw->hw_addr + rx_ring->itr_register);
1378         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1379
1380         /* Configure Tx vector */
1381         tx_ring->ims_val = E1000_IMS_TXQ0;
1382         vector++;
1383         if (tx_ring->itr_val)
1384                 writel(1000000000 / (tx_ring->itr_val * 256),
1385                        hw->hw_addr + tx_ring->itr_register);
1386         else
1387                 writel(1, hw->hw_addr + tx_ring->itr_register);
1388         adapter->eiac_mask |= tx_ring->ims_val;
1389         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1390
1391         /* set vector for Other Causes, e.g. link changes */
1392         vector++;
1393         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1394         if (rx_ring->itr_val)
1395                 writel(1000000000 / (rx_ring->itr_val * 256),
1396                        hw->hw_addr + E1000_EITR_82574(vector));
1397         else
1398                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1399
1400         /* Cause Tx interrupts on every write back */
1401         ivar |= (1 << 31);
1402
1403         ew32(IVAR, ivar);
1404
1405         /* enable MSI-X PBA support */
1406         ctrl_ext = er32(CTRL_EXT);
1407         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1408
1409         /* Auto-Mask Other interrupts upon ICR read */
1410 #define E1000_EIAC_MASK_82574   0x01F00000
1411         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1412         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1413         ew32(CTRL_EXT, ctrl_ext);
1414         e1e_flush();
1415 }
1416
1417 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1418 {
1419         if (adapter->msix_entries) {
1420                 pci_disable_msix(adapter->pdev);
1421                 kfree(adapter->msix_entries);
1422                 adapter->msix_entries = NULL;
1423         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1424                 pci_disable_msi(adapter->pdev);
1425                 adapter->flags &= ~FLAG_MSI_ENABLED;
1426         }
1427
1428         return;
1429 }
1430
1431 /**
1432  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1433  *
1434  * Attempt to configure interrupts using the best available
1435  * capabilities of the hardware and kernel.
1436  **/
1437 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1438 {
1439         int err;
1440         int numvecs, i;
1441
1442
1443         switch (adapter->int_mode) {
1444         case E1000E_INT_MODE_MSIX:
1445                 if (adapter->flags & FLAG_HAS_MSIX) {
1446                         numvecs = 3; /* RxQ0, TxQ0 and other */
1447                         adapter->msix_entries = kcalloc(numvecs,
1448                                                       sizeof(struct msix_entry),
1449                                                       GFP_KERNEL);
1450                         if (adapter->msix_entries) {
1451                                 for (i = 0; i < numvecs; i++)
1452                                         adapter->msix_entries[i].entry = i;
1453
1454                                 err = pci_enable_msix(adapter->pdev,
1455                                                       adapter->msix_entries,
1456                                                       numvecs);
1457                                 if (err == 0)
1458                                         return;
1459                         }
1460                         /* MSI-X failed, so fall through and try MSI */
1461                         e_err("Failed to initialize MSI-X interrupts.  "
1462                               "Falling back to MSI interrupts.\n");
1463                         e1000e_reset_interrupt_capability(adapter);
1464                 }
1465                 adapter->int_mode = E1000E_INT_MODE_MSI;
1466                 /* Fall through */
1467         case E1000E_INT_MODE_MSI:
1468                 if (!pci_enable_msi(adapter->pdev)) {
1469                         adapter->flags |= FLAG_MSI_ENABLED;
1470                 } else {
1471                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1472                         e_err("Failed to initialize MSI interrupts.  Falling "
1473                               "back to legacy interrupts.\n");
1474                 }
1475                 /* Fall through */
1476         case E1000E_INT_MODE_LEGACY:
1477                 /* Don't do anything; this is the system default */
1478                 break;
1479         }
1480
1481         return;
1482 }
1483
1484 /**
1485  * e1000_request_msix - Initialize MSI-X interrupts
1486  *
1487  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1488  * kernel.
1489  **/
1490 static int e1000_request_msix(struct e1000_adapter *adapter)
1491 {
1492         struct net_device *netdev = adapter->netdev;
1493         int err = 0, vector = 0;
1494
1495         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1496                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1497         else
1498                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1499         err = request_irq(adapter->msix_entries[vector].vector,
1500                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1501                           netdev);
1502         if (err)
1503                 goto out;
1504         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1505         adapter->rx_ring->itr_val = adapter->itr;
1506         vector++;
1507
1508         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1509                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1510         else
1511                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1512         err = request_irq(adapter->msix_entries[vector].vector,
1513                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1514                           netdev);
1515         if (err)
1516                 goto out;
1517         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1518         adapter->tx_ring->itr_val = adapter->itr;
1519         vector++;
1520
1521         err = request_irq(adapter->msix_entries[vector].vector,
1522                           e1000_msix_other, 0, netdev->name, netdev);
1523         if (err)
1524                 goto out;
1525
1526         e1000_configure_msix(adapter);
1527         return 0;
1528 out:
1529         return err;
1530 }
1531
1532 /**
1533  * e1000_request_irq - initialize interrupts
1534  *
1535  * Attempts to configure interrupts using the best available
1536  * capabilities of the hardware and kernel.
1537  **/
1538 static int e1000_request_irq(struct e1000_adapter *adapter)
1539 {
1540         struct net_device *netdev = adapter->netdev;
1541         int err;
1542
1543         if (adapter->msix_entries) {
1544                 err = e1000_request_msix(adapter);
1545                 if (!err)
1546                         return err;
1547                 /* fall back to MSI */
1548                 e1000e_reset_interrupt_capability(adapter);
1549                 adapter->int_mode = E1000E_INT_MODE_MSI;
1550                 e1000e_set_interrupt_capability(adapter);
1551         }
1552         if (adapter->flags & FLAG_MSI_ENABLED) {
1553                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1554                                   netdev->name, netdev);
1555                 if (!err)
1556                         return err;
1557
1558                 /* fall back to legacy interrupt */
1559                 e1000e_reset_interrupt_capability(adapter);
1560                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1561         }
1562
1563         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1564                           netdev->name, netdev);
1565         if (err)
1566                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1567
1568         return err;
1569 }
1570
1571 static void e1000_free_irq(struct e1000_adapter *adapter)
1572 {
1573         struct net_device *netdev = adapter->netdev;
1574
1575         if (adapter->msix_entries) {
1576                 int vector = 0;
1577
1578                 free_irq(adapter->msix_entries[vector].vector, netdev);
1579                 vector++;
1580
1581                 free_irq(adapter->msix_entries[vector].vector, netdev);
1582                 vector++;
1583
1584                 /* Other Causes interrupt vector */
1585                 free_irq(adapter->msix_entries[vector].vector, netdev);
1586                 return;
1587         }
1588
1589         free_irq(adapter->pdev->irq, netdev);
1590 }
1591
1592 /**
1593  * e1000_irq_disable - Mask off interrupt generation on the NIC
1594  **/
1595 static void e1000_irq_disable(struct e1000_adapter *adapter)
1596 {
1597         struct e1000_hw *hw = &adapter->hw;
1598
1599         ew32(IMC, ~0);
1600         if (adapter->msix_entries)
1601                 ew32(EIAC_82574, 0);
1602         e1e_flush();
1603         synchronize_irq(adapter->pdev->irq);
1604 }
1605
1606 /**
1607  * e1000_irq_enable - Enable default interrupt generation settings
1608  **/
1609 static void e1000_irq_enable(struct e1000_adapter *adapter)
1610 {
1611         struct e1000_hw *hw = &adapter->hw;
1612
1613         if (adapter->msix_entries) {
1614                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1615                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1616         } else {
1617                 ew32(IMS, IMS_ENABLE_MASK);
1618         }
1619         e1e_flush();
1620 }
1621
1622 /**
1623  * e1000_get_hw_control - get control of the h/w from f/w
1624  * @adapter: address of board private structure
1625  *
1626  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1627  * For ASF and Pass Through versions of f/w this means that
1628  * the driver is loaded. For AMT version (only with 82573)
1629  * of the f/w this means that the network i/f is open.
1630  **/
1631 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1632 {
1633         struct e1000_hw *hw = &adapter->hw;
1634         u32 ctrl_ext;
1635         u32 swsm;
1636
1637         /* Let firmware know the driver has taken over */
1638         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1639                 swsm = er32(SWSM);
1640                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1641         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1642                 ctrl_ext = er32(CTRL_EXT);
1643                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1644         }
1645 }
1646
1647 /**
1648  * e1000_release_hw_control - release control of the h/w to f/w
1649  * @adapter: address of board private structure
1650  *
1651  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1652  * For ASF and Pass Through versions of f/w this means that the
1653  * driver is no longer loaded. For AMT version (only with 82573) i
1654  * of the f/w this means that the network i/f is closed.
1655  *
1656  **/
1657 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1658 {
1659         struct e1000_hw *hw = &adapter->hw;
1660         u32 ctrl_ext;
1661         u32 swsm;
1662
1663         /* Let firmware taken over control of h/w */
1664         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1665                 swsm = er32(SWSM);
1666                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1667         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1668                 ctrl_ext = er32(CTRL_EXT);
1669                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1670         }
1671 }
1672
1673 /**
1674  * @e1000_alloc_ring - allocate memory for a ring structure
1675  **/
1676 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1677                                 struct e1000_ring *ring)
1678 {
1679         struct pci_dev *pdev = adapter->pdev;
1680
1681         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1682                                         GFP_KERNEL);
1683         if (!ring->desc)
1684                 return -ENOMEM;
1685
1686         return 0;
1687 }
1688
1689 /**
1690  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1691  * @adapter: board private structure
1692  *
1693  * Return 0 on success, negative on failure
1694  **/
1695 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1696 {
1697         struct e1000_ring *tx_ring = adapter->tx_ring;
1698         int err = -ENOMEM, size;
1699
1700         size = sizeof(struct e1000_buffer) * tx_ring->count;
1701         tx_ring->buffer_info = vmalloc(size);
1702         if (!tx_ring->buffer_info)
1703                 goto err;
1704         memset(tx_ring->buffer_info, 0, size);
1705
1706         /* round up to nearest 4K */
1707         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1708         tx_ring->size = ALIGN(tx_ring->size, 4096);
1709
1710         err = e1000_alloc_ring_dma(adapter, tx_ring);
1711         if (err)
1712                 goto err;
1713
1714         tx_ring->next_to_use = 0;
1715         tx_ring->next_to_clean = 0;
1716
1717         return 0;
1718 err:
1719         vfree(tx_ring->buffer_info);
1720         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1721         return err;
1722 }
1723
1724 /**
1725  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1726  * @adapter: board private structure
1727  *
1728  * Returns 0 on success, negative on failure
1729  **/
1730 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1731 {
1732         struct e1000_ring *rx_ring = adapter->rx_ring;
1733         struct e1000_buffer *buffer_info;
1734         int i, size, desc_len, err = -ENOMEM;
1735
1736         size = sizeof(struct e1000_buffer) * rx_ring->count;
1737         rx_ring->buffer_info = vmalloc(size);
1738         if (!rx_ring->buffer_info)
1739                 goto err;
1740         memset(rx_ring->buffer_info, 0, size);
1741
1742         for (i = 0; i < rx_ring->count; i++) {
1743                 buffer_info = &rx_ring->buffer_info[i];
1744                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1745                                                 sizeof(struct e1000_ps_page),
1746                                                 GFP_KERNEL);
1747                 if (!buffer_info->ps_pages)
1748                         goto err_pages;
1749         }
1750
1751         desc_len = sizeof(union e1000_rx_desc_packet_split);
1752
1753         /* Round up to nearest 4K */
1754         rx_ring->size = rx_ring->count * desc_len;
1755         rx_ring->size = ALIGN(rx_ring->size, 4096);
1756
1757         err = e1000_alloc_ring_dma(adapter, rx_ring);
1758         if (err)
1759                 goto err_pages;
1760
1761         rx_ring->next_to_clean = 0;
1762         rx_ring->next_to_use = 0;
1763         rx_ring->rx_skb_top = NULL;
1764
1765         return 0;
1766
1767 err_pages:
1768         for (i = 0; i < rx_ring->count; i++) {
1769                 buffer_info = &rx_ring->buffer_info[i];
1770                 kfree(buffer_info->ps_pages);
1771         }
1772 err:
1773         vfree(rx_ring->buffer_info);
1774         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1775         return err;
1776 }
1777
1778 /**
1779  * e1000_clean_tx_ring - Free Tx Buffers
1780  * @adapter: board private structure
1781  **/
1782 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1783 {
1784         struct e1000_ring *tx_ring = adapter->tx_ring;
1785         struct e1000_buffer *buffer_info;
1786         unsigned long size;
1787         unsigned int i;
1788
1789         for (i = 0; i < tx_ring->count; i++) {
1790                 buffer_info = &tx_ring->buffer_info[i];
1791                 e1000_put_txbuf(adapter, buffer_info);
1792         }
1793
1794         size = sizeof(struct e1000_buffer) * tx_ring->count;
1795         memset(tx_ring->buffer_info, 0, size);
1796
1797         memset(tx_ring->desc, 0, tx_ring->size);
1798
1799         tx_ring->next_to_use = 0;
1800         tx_ring->next_to_clean = 0;
1801
1802         writel(0, adapter->hw.hw_addr + tx_ring->head);
1803         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1804 }
1805
1806 /**
1807  * e1000e_free_tx_resources - Free Tx Resources per Queue
1808  * @adapter: board private structure
1809  *
1810  * Free all transmit software resources
1811  **/
1812 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1813 {
1814         struct pci_dev *pdev = adapter->pdev;
1815         struct e1000_ring *tx_ring = adapter->tx_ring;
1816
1817         e1000_clean_tx_ring(adapter);
1818
1819         vfree(tx_ring->buffer_info);
1820         tx_ring->buffer_info = NULL;
1821
1822         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1823                           tx_ring->dma);
1824         tx_ring->desc = NULL;
1825 }
1826
1827 /**
1828  * e1000e_free_rx_resources - Free Rx Resources
1829  * @adapter: board private structure
1830  *
1831  * Free all receive software resources
1832  **/
1833
1834 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1835 {
1836         struct pci_dev *pdev = adapter->pdev;
1837         struct e1000_ring *rx_ring = adapter->rx_ring;
1838         int i;
1839
1840         e1000_clean_rx_ring(adapter);
1841
1842         for (i = 0; i < rx_ring->count; i++) {
1843                 kfree(rx_ring->buffer_info[i].ps_pages);
1844         }
1845
1846         vfree(rx_ring->buffer_info);
1847         rx_ring->buffer_info = NULL;
1848
1849         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1850                           rx_ring->dma);
1851         rx_ring->desc = NULL;
1852 }
1853
1854 /**
1855  * e1000_update_itr - update the dynamic ITR value based on statistics
1856  * @adapter: pointer to adapter
1857  * @itr_setting: current adapter->itr
1858  * @packets: the number of packets during this measurement interval
1859  * @bytes: the number of bytes during this measurement interval
1860  *
1861  *      Stores a new ITR value based on packets and byte
1862  *      counts during the last interrupt.  The advantage of per interrupt
1863  *      computation is faster updates and more accurate ITR for the current
1864  *      traffic pattern.  Constants in this function were computed
1865  *      based on theoretical maximum wire speed and thresholds were set based
1866  *      on testing data as well as attempting to minimize response time
1867  *      while increasing bulk throughput.  This functionality is controlled
1868  *      by the InterruptThrottleRate module parameter.
1869  **/
1870 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1871                                      u16 itr_setting, int packets,
1872                                      int bytes)
1873 {
1874         unsigned int retval = itr_setting;
1875
1876         if (packets == 0)
1877                 goto update_itr_done;
1878
1879         switch (itr_setting) {
1880         case lowest_latency:
1881                 /* handle TSO and jumbo frames */
1882                 if (bytes/packets > 8000)
1883                         retval = bulk_latency;
1884                 else if ((packets < 5) && (bytes > 512)) {
1885                         retval = low_latency;
1886                 }
1887                 break;
1888         case low_latency:  /* 50 usec aka 20000 ints/s */
1889                 if (bytes > 10000) {
1890                         /* this if handles the TSO accounting */
1891                         if (bytes/packets > 8000) {
1892                                 retval = bulk_latency;
1893                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1894                                 retval = bulk_latency;
1895                         } else if ((packets > 35)) {
1896                                 retval = lowest_latency;
1897                         }
1898                 } else if (bytes/packets > 2000) {
1899                         retval = bulk_latency;
1900                 } else if (packets <= 2 && bytes < 512) {
1901                         retval = lowest_latency;
1902                 }
1903                 break;
1904         case bulk_latency: /* 250 usec aka 4000 ints/s */
1905                 if (bytes > 25000) {
1906                         if (packets > 35) {
1907                                 retval = low_latency;
1908                         }
1909                 } else if (bytes < 6000) {
1910                         retval = low_latency;
1911                 }
1912                 break;
1913         }
1914
1915 update_itr_done:
1916         return retval;
1917 }
1918
1919 static void e1000_set_itr(struct e1000_adapter *adapter)
1920 {
1921         struct e1000_hw *hw = &adapter->hw;
1922         u16 current_itr;
1923         u32 new_itr = adapter->itr;
1924
1925         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1926         if (adapter->link_speed != SPEED_1000) {
1927                 current_itr = 0;
1928                 new_itr = 4000;
1929                 goto set_itr_now;
1930         }
1931
1932         adapter->tx_itr = e1000_update_itr(adapter,
1933                                     adapter->tx_itr,
1934                                     adapter->total_tx_packets,
1935                                     adapter->total_tx_bytes);
1936         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1937         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1938                 adapter->tx_itr = low_latency;
1939
1940         adapter->rx_itr = e1000_update_itr(adapter,
1941                                     adapter->rx_itr,
1942                                     adapter->total_rx_packets,
1943                                     adapter->total_rx_bytes);
1944         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1945         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1946                 adapter->rx_itr = low_latency;
1947
1948         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1949
1950         switch (current_itr) {
1951         /* counts and packets in update_itr are dependent on these numbers */
1952         case lowest_latency:
1953                 new_itr = 70000;
1954                 break;
1955         case low_latency:
1956                 new_itr = 20000; /* aka hwitr = ~200 */
1957                 break;
1958         case bulk_latency:
1959                 new_itr = 4000;
1960                 break;
1961         default:
1962                 break;
1963         }
1964
1965 set_itr_now:
1966         if (new_itr != adapter->itr) {
1967                 /*
1968                  * this attempts to bias the interrupt rate towards Bulk
1969                  * by adding intermediate steps when interrupt rate is
1970                  * increasing
1971                  */
1972                 new_itr = new_itr > adapter->itr ?
1973                              min(adapter->itr + (new_itr >> 2), new_itr) :
1974                              new_itr;
1975                 adapter->itr = new_itr;
1976                 adapter->rx_ring->itr_val = new_itr;
1977                 if (adapter->msix_entries)
1978                         adapter->rx_ring->set_itr = 1;
1979                 else
1980                         ew32(ITR, 1000000000 / (new_itr * 256));
1981         }
1982 }
1983
1984 /**
1985  * e1000_alloc_queues - Allocate memory for all rings
1986  * @adapter: board private structure to initialize
1987  **/
1988 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1989 {
1990         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1991         if (!adapter->tx_ring)
1992                 goto err;
1993
1994         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1995         if (!adapter->rx_ring)
1996                 goto err;
1997
1998         return 0;
1999 err:
2000         e_err("Unable to allocate memory for queues\n");
2001         kfree(adapter->rx_ring);
2002         kfree(adapter->tx_ring);
2003         return -ENOMEM;
2004 }
2005
2006 /**
2007  * e1000_clean - NAPI Rx polling callback
2008  * @napi: struct associated with this polling callback
2009  * @budget: amount of packets driver is allowed to process this poll
2010  **/
2011 static int e1000_clean(struct napi_struct *napi, int budget)
2012 {
2013         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2014         struct e1000_hw *hw = &adapter->hw;
2015         struct net_device *poll_dev = adapter->netdev;
2016         int tx_cleaned = 1, work_done = 0;
2017
2018         adapter = netdev_priv(poll_dev);
2019
2020         if (adapter->msix_entries &&
2021             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2022                 goto clean_rx;
2023
2024         tx_cleaned = e1000_clean_tx_irq(adapter);
2025
2026 clean_rx:
2027         adapter->clean_rx(adapter, &work_done, budget);
2028
2029         if (!tx_cleaned)
2030                 work_done = budget;
2031
2032         /* If budget not fully consumed, exit the polling mode */
2033         if (work_done < budget) {
2034                 if (adapter->itr_setting & 3)
2035                         e1000_set_itr(adapter);
2036                 napi_complete(napi);
2037                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2038                         if (adapter->msix_entries)
2039                                 ew32(IMS, adapter->rx_ring->ims_val);
2040                         else
2041                                 e1000_irq_enable(adapter);
2042                 }
2043         }
2044
2045         return work_done;
2046 }
2047
2048 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2049 {
2050         struct e1000_adapter *adapter = netdev_priv(netdev);
2051         struct e1000_hw *hw = &adapter->hw;
2052         u32 vfta, index;
2053
2054         /* don't update vlan cookie if already programmed */
2055         if ((adapter->hw.mng_cookie.status &
2056              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2057             (vid == adapter->mng_vlan_id))
2058                 return;
2059
2060         /* add VID to filter table */
2061         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2062                 index = (vid >> 5) & 0x7F;
2063                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2064                 vfta |= (1 << (vid & 0x1F));
2065                 hw->mac.ops.write_vfta(hw, index, vfta);
2066         }
2067 }
2068
2069 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2070 {
2071         struct e1000_adapter *adapter = netdev_priv(netdev);
2072         struct e1000_hw *hw = &adapter->hw;
2073         u32 vfta, index;
2074
2075         if (!test_bit(__E1000_DOWN, &adapter->state))
2076                 e1000_irq_disable(adapter);
2077         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2078
2079         if (!test_bit(__E1000_DOWN, &adapter->state))
2080                 e1000_irq_enable(adapter);
2081
2082         if ((adapter->hw.mng_cookie.status &
2083              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2084             (vid == adapter->mng_vlan_id)) {
2085                 /* release control to f/w */
2086                 e1000_release_hw_control(adapter);
2087                 return;
2088         }
2089
2090         /* remove VID from filter table */
2091         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2092                 index = (vid >> 5) & 0x7F;
2093                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2094                 vfta &= ~(1 << (vid & 0x1F));
2095                 hw->mac.ops.write_vfta(hw, index, vfta);
2096         }
2097 }
2098
2099 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2100 {
2101         struct net_device *netdev = adapter->netdev;
2102         u16 vid = adapter->hw.mng_cookie.vlan_id;
2103         u16 old_vid = adapter->mng_vlan_id;
2104
2105         if (!adapter->vlgrp)
2106                 return;
2107
2108         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2109                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2110                 if (adapter->hw.mng_cookie.status &
2111                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2112                         e1000_vlan_rx_add_vid(netdev, vid);
2113                         adapter->mng_vlan_id = vid;
2114                 }
2115
2116                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2117                                 (vid != old_vid) &&
2118                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2119                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2120         } else {
2121                 adapter->mng_vlan_id = vid;
2122         }
2123 }
2124
2125
2126 static void e1000_vlan_rx_register(struct net_device *netdev,
2127                                    struct vlan_group *grp)
2128 {
2129         struct e1000_adapter *adapter = netdev_priv(netdev);
2130         struct e1000_hw *hw = &adapter->hw;
2131         u32 ctrl, rctl;
2132
2133         if (!test_bit(__E1000_DOWN, &adapter->state))
2134                 e1000_irq_disable(adapter);
2135         adapter->vlgrp = grp;
2136
2137         if (grp) {
2138                 /* enable VLAN tag insert/strip */
2139                 ctrl = er32(CTRL);
2140                 ctrl |= E1000_CTRL_VME;
2141                 ew32(CTRL, ctrl);
2142
2143                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2144                         /* enable VLAN receive filtering */
2145                         rctl = er32(RCTL);
2146                         rctl &= ~E1000_RCTL_CFIEN;
2147                         ew32(RCTL, rctl);
2148                         e1000_update_mng_vlan(adapter);
2149                 }
2150         } else {
2151                 /* disable VLAN tag insert/strip */
2152                 ctrl = er32(CTRL);
2153                 ctrl &= ~E1000_CTRL_VME;
2154                 ew32(CTRL, ctrl);
2155
2156                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2157                         if (adapter->mng_vlan_id !=
2158                             (u16)E1000_MNG_VLAN_NONE) {
2159                                 e1000_vlan_rx_kill_vid(netdev,
2160                                                        adapter->mng_vlan_id);
2161                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2162                         }
2163                 }
2164         }
2165
2166         if (!test_bit(__E1000_DOWN, &adapter->state))
2167                 e1000_irq_enable(adapter);
2168 }
2169
2170 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2171 {
2172         u16 vid;
2173
2174         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2175
2176         if (!adapter->vlgrp)
2177                 return;
2178
2179         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2180                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2181                         continue;
2182                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2183         }
2184 }
2185
2186 static void e1000_init_manageability(struct e1000_adapter *adapter)
2187 {
2188         struct e1000_hw *hw = &adapter->hw;
2189         u32 manc, manc2h;
2190
2191         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2192                 return;
2193
2194         manc = er32(MANC);
2195
2196         /*
2197          * enable receiving management packets to the host. this will probably
2198          * generate destination unreachable messages from the host OS, but
2199          * the packets will be handled on SMBUS
2200          */
2201         manc |= E1000_MANC_EN_MNG2HOST;
2202         manc2h = er32(MANC2H);
2203 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2204 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2205         manc2h |= E1000_MNG2HOST_PORT_623;
2206         manc2h |= E1000_MNG2HOST_PORT_664;
2207         ew32(MANC2H, manc2h);
2208         ew32(MANC, manc);
2209 }
2210
2211 /**
2212  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2213  * @adapter: board private structure
2214  *
2215  * Configure the Tx unit of the MAC after a reset.
2216  **/
2217 static void e1000_configure_tx(struct e1000_adapter *adapter)
2218 {
2219         struct e1000_hw *hw = &adapter->hw;
2220         struct e1000_ring *tx_ring = adapter->tx_ring;
2221         u64 tdba;
2222         u32 tdlen, tctl, tipg, tarc;
2223         u32 ipgr1, ipgr2;
2224
2225         /* Setup the HW Tx Head and Tail descriptor pointers */
2226         tdba = tx_ring->dma;
2227         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2228         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2229         ew32(TDBAH, (tdba >> 32));
2230         ew32(TDLEN, tdlen);
2231         ew32(TDH, 0);
2232         ew32(TDT, 0);
2233         tx_ring->head = E1000_TDH;
2234         tx_ring->tail = E1000_TDT;
2235
2236         /* Set the default values for the Tx Inter Packet Gap timer */
2237         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2238         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2239         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2240
2241         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2242                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2243
2244         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2245         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2246         ew32(TIPG, tipg);
2247
2248         /* Set the Tx Interrupt Delay register */
2249         ew32(TIDV, adapter->tx_int_delay);
2250         /* Tx irq moderation */
2251         ew32(TADV, adapter->tx_abs_int_delay);
2252
2253         /* Program the Transmit Control Register */
2254         tctl = er32(TCTL);
2255         tctl &= ~E1000_TCTL_CT;
2256         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2257                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2258
2259         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2260                 tarc = er32(TARC(0));
2261                 /*
2262                  * set the speed mode bit, we'll clear it if we're not at
2263                  * gigabit link later
2264                  */
2265 #define SPEED_MODE_BIT (1 << 21)
2266                 tarc |= SPEED_MODE_BIT;
2267                 ew32(TARC(0), tarc);
2268         }
2269
2270         /* errata: program both queues to unweighted RR */
2271         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2272                 tarc = er32(TARC(0));
2273                 tarc |= 1;
2274                 ew32(TARC(0), tarc);
2275                 tarc = er32(TARC(1));
2276                 tarc |= 1;
2277                 ew32(TARC(1), tarc);
2278         }
2279
2280         /* Setup Transmit Descriptor Settings for eop descriptor */
2281         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2282
2283         /* only set IDE if we are delaying interrupts using the timers */
2284         if (adapter->tx_int_delay)
2285                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2286
2287         /* enable Report Status bit */
2288         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2289
2290         ew32(TCTL, tctl);
2291
2292         e1000e_config_collision_dist(hw);
2293
2294         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
2295 }
2296
2297 /**
2298  * e1000_setup_rctl - configure the receive control registers
2299  * @adapter: Board private structure
2300  **/
2301 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2302                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2303 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2304 {
2305         struct e1000_hw *hw = &adapter->hw;
2306         u32 rctl, rfctl;
2307         u32 psrctl = 0;
2308         u32 pages = 0;
2309
2310         /* Program MC offset vector base */
2311         rctl = er32(RCTL);
2312         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2313         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2314                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2315                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2316
2317         /* Do not Store bad packets */
2318         rctl &= ~E1000_RCTL_SBP;
2319
2320         /* Enable Long Packet receive */
2321         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2322                 rctl &= ~E1000_RCTL_LPE;
2323         else
2324                 rctl |= E1000_RCTL_LPE;
2325
2326         /* Some systems expect that the CRC is included in SMBUS traffic. The
2327          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2328          * host memory when this is enabled
2329          */
2330         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2331                 rctl |= E1000_RCTL_SECRC;
2332
2333         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2334         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2335                 u16 phy_data;
2336
2337                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2338                 phy_data &= 0xfff8;
2339                 phy_data |= (1 << 2);
2340                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2341
2342                 e1e_rphy(hw, 22, &phy_data);
2343                 phy_data &= 0x0fff;
2344                 phy_data |= (1 << 14);
2345                 e1e_wphy(hw, 0x10, 0x2823);
2346                 e1e_wphy(hw, 0x11, 0x0003);
2347                 e1e_wphy(hw, 22, phy_data);
2348         }
2349
2350         /* Setup buffer sizes */
2351         rctl &= ~E1000_RCTL_SZ_4096;
2352         rctl |= E1000_RCTL_BSEX;
2353         switch (adapter->rx_buffer_len) {
2354         case 2048:
2355         default:
2356                 rctl |= E1000_RCTL_SZ_2048;
2357                 rctl &= ~E1000_RCTL_BSEX;
2358                 break;
2359         case 4096:
2360                 rctl |= E1000_RCTL_SZ_4096;
2361                 break;
2362         case 8192:
2363                 rctl |= E1000_RCTL_SZ_8192;
2364                 break;
2365         case 16384:
2366                 rctl |= E1000_RCTL_SZ_16384;
2367                 break;
2368         }
2369
2370         /*
2371          * 82571 and greater support packet-split where the protocol
2372          * header is placed in skb->data and the packet data is
2373          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2374          * In the case of a non-split, skb->data is linearly filled,
2375          * followed by the page buffers.  Therefore, skb->data is
2376          * sized to hold the largest protocol header.
2377          *
2378          * allocations using alloc_page take too long for regular MTU
2379          * so only enable packet split for jumbo frames
2380          *
2381          * Using pages when the page size is greater than 16k wastes
2382          * a lot of memory, since we allocate 3 pages at all times
2383          * per packet.
2384          */
2385         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2386         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2387             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2388                 adapter->rx_ps_pages = pages;
2389         else
2390                 adapter->rx_ps_pages = 0;
2391
2392         if (adapter->rx_ps_pages) {
2393                 /* Configure extra packet-split registers */
2394                 rfctl = er32(RFCTL);
2395                 rfctl |= E1000_RFCTL_EXTEN;
2396                 /*
2397                  * disable packet split support for IPv6 extension headers,
2398                  * because some malformed IPv6 headers can hang the Rx
2399                  */
2400                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2401                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2402
2403                 ew32(RFCTL, rfctl);
2404
2405                 /* Enable Packet split descriptors */
2406                 rctl |= E1000_RCTL_DTYP_PS;
2407
2408                 psrctl |= adapter->rx_ps_bsize0 >>
2409                         E1000_PSRCTL_BSIZE0_SHIFT;
2410
2411                 switch (adapter->rx_ps_pages) {
2412                 case 3:
2413                         psrctl |= PAGE_SIZE <<
2414                                 E1000_PSRCTL_BSIZE3_SHIFT;
2415                 case 2:
2416                         psrctl |= PAGE_SIZE <<
2417                                 E1000_PSRCTL_BSIZE2_SHIFT;
2418                 case 1:
2419                         psrctl |= PAGE_SIZE >>
2420                                 E1000_PSRCTL_BSIZE1_SHIFT;
2421                         break;
2422                 }
2423
2424                 ew32(PSRCTL, psrctl);
2425         }
2426
2427         ew32(RCTL, rctl);
2428         /* just started the receive unit, no need to restart */
2429         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2430 }
2431
2432 /**
2433  * e1000_configure_rx - Configure Receive Unit after Reset
2434  * @adapter: board private structure
2435  *
2436  * Configure the Rx unit of the MAC after a reset.
2437  **/
2438 static void e1000_configure_rx(struct e1000_adapter *adapter)
2439 {
2440         struct e1000_hw *hw = &adapter->hw;
2441         struct e1000_ring *rx_ring = adapter->rx_ring;
2442         u64 rdba;
2443         u32 rdlen, rctl, rxcsum, ctrl_ext;
2444
2445         if (adapter->rx_ps_pages) {
2446                 /* this is a 32 byte descriptor */
2447                 rdlen = rx_ring->count *
2448                         sizeof(union e1000_rx_desc_packet_split);
2449                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2450                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2451         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2452                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2453                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2454                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2455         } else {
2456                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2457                 adapter->clean_rx = e1000_clean_rx_irq;
2458                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2459         }
2460
2461         /* disable receives while setting up the descriptors */
2462         rctl = er32(RCTL);
2463         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2464         e1e_flush();
2465         msleep(10);
2466
2467         /* set the Receive Delay Timer Register */
2468         ew32(RDTR, adapter->rx_int_delay);
2469
2470         /* irq moderation */
2471         ew32(RADV, adapter->rx_abs_int_delay);
2472         if (adapter->itr_setting != 0)
2473                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2474
2475         ctrl_ext = er32(CTRL_EXT);
2476         /* Auto-Mask interrupts upon ICR access */
2477         ctrl_ext |= E1000_CTRL_EXT_IAME;
2478         ew32(IAM, 0xffffffff);
2479         ew32(CTRL_EXT, ctrl_ext);
2480         e1e_flush();
2481
2482         /*
2483          * Setup the HW Rx Head and Tail Descriptor Pointers and
2484          * the Base and Length of the Rx Descriptor Ring
2485          */
2486         rdba = rx_ring->dma;
2487         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2488         ew32(RDBAH, (rdba >> 32));
2489         ew32(RDLEN, rdlen);
2490         ew32(RDH, 0);
2491         ew32(RDT, 0);
2492         rx_ring->head = E1000_RDH;
2493         rx_ring->tail = E1000_RDT;
2494
2495         /* Enable Receive Checksum Offload for TCP and UDP */
2496         rxcsum = er32(RXCSUM);
2497         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2498                 rxcsum |= E1000_RXCSUM_TUOFL;
2499
2500                 /*
2501                  * IPv4 payload checksum for UDP fragments must be
2502                  * used in conjunction with packet-split.
2503                  */
2504                 if (adapter->rx_ps_pages)
2505                         rxcsum |= E1000_RXCSUM_IPPCSE;
2506         } else {
2507                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2508                 /* no need to clear IPPCSE as it defaults to 0 */
2509         }
2510         ew32(RXCSUM, rxcsum);
2511
2512         /*
2513          * Enable early receives on supported devices, only takes effect when
2514          * packet size is equal or larger than the specified value (in 8 byte
2515          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2516          */
2517         if (adapter->flags & FLAG_HAS_ERT) {
2518                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2519                         u32 rxdctl = er32(RXDCTL(0));
2520                         ew32(RXDCTL(0), rxdctl | 0x3);
2521                         ew32(ERT, E1000_ERT_2048 | (1 << 13));
2522                         /*
2523                          * With jumbo frames and early-receive enabled,
2524                          * excessive C-state transition latencies result in
2525                          * dropped transactions.
2526                          */
2527                         pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2528                                                   adapter->netdev->name, 55);
2529                 } else {
2530                         pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2531                                                   adapter->netdev->name,
2532                                                   PM_QOS_DEFAULT_VALUE);
2533                 }
2534         }
2535
2536         /* Enable Receives */
2537         ew32(RCTL, rctl);
2538 }
2539
2540 /**
2541  *  e1000_update_mc_addr_list - Update Multicast addresses
2542  *  @hw: pointer to the HW structure
2543  *  @mc_addr_list: array of multicast addresses to program
2544  *  @mc_addr_count: number of multicast addresses to program
2545  *
2546  *  Updates the Multicast Table Array.
2547  *  The caller must have a packed mc_addr_list of multicast addresses.
2548  **/
2549 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2550                                       u32 mc_addr_count)
2551 {
2552         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
2553 }
2554
2555 /**
2556  * e1000_set_multi - Multicast and Promiscuous mode set
2557  * @netdev: network interface device structure
2558  *
2559  * The set_multi entry point is called whenever the multicast address
2560  * list or the network interface flags are updated.  This routine is
2561  * responsible for configuring the hardware for proper multicast,
2562  * promiscuous mode, and all-multi behavior.
2563  **/
2564 static void e1000_set_multi(struct net_device *netdev)
2565 {
2566         struct e1000_adapter *adapter = netdev_priv(netdev);
2567         struct e1000_hw *hw = &adapter->hw;
2568         struct dev_mc_list *mc_ptr;
2569         u8  *mta_list;
2570         u32 rctl;
2571         int i;
2572
2573         /* Check for Promiscuous and All Multicast modes */
2574
2575         rctl = er32(RCTL);
2576
2577         if (netdev->flags & IFF_PROMISC) {
2578                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2579                 rctl &= ~E1000_RCTL_VFE;
2580         } else {
2581                 if (netdev->flags & IFF_ALLMULTI) {
2582                         rctl |= E1000_RCTL_MPE;
2583                         rctl &= ~E1000_RCTL_UPE;
2584                 } else {
2585                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2586                 }
2587                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2588                         rctl |= E1000_RCTL_VFE;
2589         }
2590
2591         ew32(RCTL, rctl);
2592
2593         if (!netdev_mc_empty(netdev)) {
2594                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
2595                 if (!mta_list)
2596                         return;
2597
2598                 /* prepare a packed array of only addresses. */
2599                 i = 0;
2600                 netdev_for_each_mc_addr(mc_ptr, netdev)
2601                         memcpy(mta_list + (i++ * ETH_ALEN),
2602                                mc_ptr->dmi_addr, ETH_ALEN);
2603
2604                 e1000_update_mc_addr_list(hw, mta_list, i);
2605                 kfree(mta_list);
2606         } else {
2607                 /*
2608                  * if we're called from probe, we might not have
2609                  * anything to do here, so clear out the list
2610                  */
2611                 e1000_update_mc_addr_list(hw, NULL, 0);
2612         }
2613 }
2614
2615 /**
2616  * e1000_configure - configure the hardware for Rx and Tx
2617  * @adapter: private board structure
2618  **/
2619 static void e1000_configure(struct e1000_adapter *adapter)
2620 {
2621         e1000_set_multi(adapter->netdev);
2622
2623         e1000_restore_vlan(adapter);
2624         e1000_init_manageability(adapter);
2625
2626         e1000_configure_tx(adapter);
2627         e1000_setup_rctl(adapter);
2628         e1000_configure_rx(adapter);
2629         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2630 }
2631
2632 /**
2633  * e1000e_power_up_phy - restore link in case the phy was powered down
2634  * @adapter: address of board private structure
2635  *
2636  * The phy may be powered down to save power and turn off link when the
2637  * driver is unloaded and wake on lan is not enabled (among others)
2638  * *** this routine MUST be followed by a call to e1000e_reset ***
2639  **/
2640 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2641 {
2642         if (adapter->hw.phy.ops.power_up)
2643                 adapter->hw.phy.ops.power_up(&adapter->hw);
2644
2645         adapter->hw.mac.ops.setup_link(&adapter->hw);
2646 }
2647
2648 /**
2649  * e1000_power_down_phy - Power down the PHY
2650  *
2651  * Power down the PHY so no link is implied when interface is down.
2652  * The PHY cannot be powered down if management or WoL is active.
2653  */
2654 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2655 {
2656         /* WoL is enabled */
2657         if (adapter->wol)
2658                 return;
2659
2660         if (adapter->hw.phy.ops.power_down)
2661                 adapter->hw.phy.ops.power_down(&adapter->hw);
2662 }
2663
2664 /**
2665  * e1000e_reset - bring the hardware into a known good state
2666  *
2667  * This function boots the hardware and enables some settings that
2668  * require a configuration cycle of the hardware - those cannot be
2669  * set/changed during runtime. After reset the device needs to be
2670  * properly configured for Rx, Tx etc.
2671  */
2672 void e1000e_reset(struct e1000_adapter *adapter)
2673 {
2674         struct e1000_mac_info *mac = &adapter->hw.mac;
2675         struct e1000_fc_info *fc = &adapter->hw.fc;
2676         struct e1000_hw *hw = &adapter->hw;
2677         u32 tx_space, min_tx_space, min_rx_space;
2678         u32 pba = adapter->pba;
2679         u16 hwm;
2680
2681         /* reset Packet Buffer Allocation to default */
2682         ew32(PBA, pba);
2683
2684         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2685                 /*
2686                  * To maintain wire speed transmits, the Tx FIFO should be
2687                  * large enough to accommodate two full transmit packets,
2688                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2689                  * the Rx FIFO should be large enough to accommodate at least
2690                  * one full receive packet and is similarly rounded up and
2691                  * expressed in KB.
2692                  */
2693                 pba = er32(PBA);
2694                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2695                 tx_space = pba >> 16;
2696                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2697                 pba &= 0xffff;
2698                 /*
2699                  * the Tx fifo also stores 16 bytes of information about the tx
2700                  * but don't include ethernet FCS because hardware appends it
2701                  */
2702                 min_tx_space = (adapter->max_frame_size +
2703                                 sizeof(struct e1000_tx_desc) -
2704                                 ETH_FCS_LEN) * 2;
2705                 min_tx_space = ALIGN(min_tx_space, 1024);
2706                 min_tx_space >>= 10;
2707                 /* software strips receive CRC, so leave room for it */
2708                 min_rx_space = adapter->max_frame_size;
2709                 min_rx_space = ALIGN(min_rx_space, 1024);
2710                 min_rx_space >>= 10;
2711
2712                 /*
2713                  * If current Tx allocation is less than the min Tx FIFO size,
2714                  * and the min Tx FIFO size is less than the current Rx FIFO
2715                  * allocation, take space away from current Rx allocation
2716                  */
2717                 if ((tx_space < min_tx_space) &&
2718                     ((min_tx_space - tx_space) < pba)) {
2719                         pba -= min_tx_space - tx_space;
2720
2721                         /*
2722                          * if short on Rx space, Rx wins and must trump tx
2723                          * adjustment or use Early Receive if available
2724                          */
2725                         if ((pba < min_rx_space) &&
2726                             (!(adapter->flags & FLAG_HAS_ERT)))
2727                                 /* ERT enabled in e1000_configure_rx */
2728                                 pba = min_rx_space;
2729                 }
2730
2731                 ew32(PBA, pba);
2732         }
2733
2734
2735         /*
2736          * flow control settings
2737          *
2738          * The high water mark must be low enough to fit one full frame
2739          * (or the size used for early receive) above it in the Rx FIFO.
2740          * Set it to the lower of:
2741          * - 90% of the Rx FIFO size, and
2742          * - the full Rx FIFO size minus the early receive size (for parts
2743          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2744          * - the full Rx FIFO size minus one full frame
2745          */
2746         if (hw->mac.type == e1000_pchlan) {
2747                 /*
2748                  * Workaround PCH LOM adapter hangs with certain network
2749                  * loads.  If hangs persist, try disabling Tx flow control.
2750                  */
2751                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2752                         fc->high_water = 0x3500;
2753                         fc->low_water  = 0x1500;
2754                 } else {
2755                         fc->high_water = 0x5000;
2756                         fc->low_water  = 0x3000;
2757                 }
2758         } else {
2759                 if ((adapter->flags & FLAG_HAS_ERT) &&
2760                     (adapter->netdev->mtu > ETH_DATA_LEN))
2761                         hwm = min(((pba << 10) * 9 / 10),
2762                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
2763                 else
2764                         hwm = min(((pba << 10) * 9 / 10),
2765                                   ((pba << 10) - adapter->max_frame_size));
2766
2767                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
2768                 fc->low_water = fc->high_water - 8;
2769         }
2770
2771         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2772                 fc->pause_time = 0xFFFF;
2773         else
2774                 fc->pause_time = E1000_FC_PAUSE_TIME;
2775         fc->send_xon = 1;
2776         fc->current_mode = fc->requested_mode;
2777
2778         /* Allow time for pending master requests to run */
2779         mac->ops.reset_hw(hw);
2780
2781         /*
2782          * For parts with AMT enabled, let the firmware know
2783          * that the network interface is in control
2784          */
2785         if (adapter->flags & FLAG_HAS_AMT)
2786                 e1000_get_hw_control(adapter);
2787
2788         ew32(WUC, 0);
2789         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP)
2790                 e1e_wphy(&adapter->hw, BM_WUC, 0);
2791
2792         if (mac->ops.init_hw(hw))
2793                 e_err("Hardware Error\n");
2794
2795         /* additional part of the flow-control workaround above */
2796         if (hw->mac.type == e1000_pchlan)
2797                 ew32(FCRTV_PCH, 0x1000);
2798
2799         e1000_update_mng_vlan(adapter);
2800
2801         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2802         ew32(VET, ETH_P_8021Q);
2803
2804         e1000e_reset_adaptive(hw);
2805         e1000_get_phy_info(hw);
2806
2807         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
2808             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2809                 u16 phy_data = 0;
2810                 /*
2811                  * speed up time to link by disabling smart power down, ignore
2812                  * the return value of this function because there is nothing
2813                  * different we would do if it failed
2814                  */
2815                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2816                 phy_data &= ~IGP02E1000_PM_SPD;
2817                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2818         }
2819 }
2820
2821 int e1000e_up(struct e1000_adapter *adapter)
2822 {
2823         struct e1000_hw *hw = &adapter->hw;
2824
2825         /* DMA latency requirement to workaround early-receive/jumbo issue */
2826         if (adapter->flags & FLAG_HAS_ERT)
2827                 pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY,
2828                                        adapter->netdev->name,
2829                                        PM_QOS_DEFAULT_VALUE);
2830
2831         /* hardware has been reset, we need to reload some things */
2832         e1000_configure(adapter);
2833
2834         clear_bit(__E1000_DOWN, &adapter->state);
2835
2836         napi_enable(&adapter->napi);
2837         if (adapter->msix_entries)
2838                 e1000_configure_msix(adapter);
2839         e1000_irq_enable(adapter);
2840
2841         netif_wake_queue(adapter->netdev);
2842
2843         /* fire a link change interrupt to start the watchdog */
2844         ew32(ICS, E1000_ICS_LSC);
2845         return 0;
2846 }
2847
2848 void e1000e_down(struct e1000_adapter *adapter)
2849 {
2850         struct net_device *netdev = adapter->netdev;
2851         struct e1000_hw *hw = &adapter->hw;
2852         u32 tctl, rctl;
2853
2854         /*
2855          * signal that we're down so the interrupt handler does not
2856          * reschedule our watchdog timer
2857          */
2858         set_bit(__E1000_DOWN, &adapter->state);
2859
2860         /* disable receives in the hardware */
2861         rctl = er32(RCTL);
2862         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2863         /* flush and sleep below */
2864
2865         netif_stop_queue(netdev);
2866
2867         /* disable transmits in the hardware */
2868         tctl = er32(TCTL);
2869         tctl &= ~E1000_TCTL_EN;
2870         ew32(TCTL, tctl);
2871         /* flush both disables and wait for them to finish */
2872         e1e_flush();
2873         msleep(10);
2874
2875         napi_disable(&adapter->napi);
2876         e1000_irq_disable(adapter);
2877
2878         del_timer_sync(&adapter->watchdog_timer);
2879         del_timer_sync(&adapter->phy_info_timer);
2880
2881         netdev->tx_queue_len = adapter->tx_queue_len;
2882         netif_carrier_off(netdev);
2883         adapter->link_speed = 0;
2884         adapter->link_duplex = 0;
2885
2886         if (!pci_channel_offline(adapter->pdev))
2887                 e1000e_reset(adapter);
2888         e1000_clean_tx_ring(adapter);
2889         e1000_clean_rx_ring(adapter);
2890
2891         if (adapter->flags & FLAG_HAS_ERT)
2892                 pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY,
2893                                           adapter->netdev->name);
2894
2895         /*
2896          * TODO: for power management, we could drop the link and
2897          * pci_disable_device here.
2898          */
2899 }
2900
2901 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2902 {
2903         might_sleep();
2904         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2905                 msleep(1);
2906         e1000e_down(adapter);
2907         e1000e_up(adapter);
2908         clear_bit(__E1000_RESETTING, &adapter->state);
2909 }
2910
2911 /**
2912  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2913  * @adapter: board private structure to initialize
2914  *
2915  * e1000_sw_init initializes the Adapter private data structure.
2916  * Fields are initialized based on PCI device information and
2917  * OS network device settings (MTU size).
2918  **/
2919 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2920 {
2921         struct net_device *netdev = adapter->netdev;
2922
2923         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2924         adapter->rx_ps_bsize0 = 128;
2925         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2926         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2927
2928         e1000e_set_interrupt_capability(adapter);
2929
2930         if (e1000_alloc_queues(adapter))
2931                 return -ENOMEM;
2932
2933         /* Explicitly disable IRQ since the NIC can be in any state. */
2934         e1000_irq_disable(adapter);
2935
2936         set_bit(__E1000_DOWN, &adapter->state);
2937         return 0;
2938 }
2939
2940 /**
2941  * e1000_intr_msi_test - Interrupt Handler
2942  * @irq: interrupt number
2943  * @data: pointer to a network interface device structure
2944  **/
2945 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2946 {
2947         struct net_device *netdev = data;
2948         struct e1000_adapter *adapter = netdev_priv(netdev);
2949         struct e1000_hw *hw = &adapter->hw;
2950         u32 icr = er32(ICR);
2951
2952         e_dbg("icr is %08X\n", icr);
2953         if (icr & E1000_ICR_RXSEQ) {
2954                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2955                 wmb();
2956         }
2957
2958         return IRQ_HANDLED;
2959 }
2960
2961 /**
2962  * e1000_test_msi_interrupt - Returns 0 for successful test
2963  * @adapter: board private struct
2964  *
2965  * code flow taken from tg3.c
2966  **/
2967 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2968 {
2969         struct net_device *netdev = adapter->netdev;
2970         struct e1000_hw *hw = &adapter->hw;
2971         int err;
2972
2973         /* poll_enable hasn't been called yet, so don't need disable */
2974         /* clear any pending events */
2975         er32(ICR);
2976
2977         /* free the real vector and request a test handler */
2978         e1000_free_irq(adapter);
2979         e1000e_reset_interrupt_capability(adapter);
2980
2981         /* Assume that the test fails, if it succeeds then the test
2982          * MSI irq handler will unset this flag */
2983         adapter->flags |= FLAG_MSI_TEST_FAILED;
2984
2985         err = pci_enable_msi(adapter->pdev);
2986         if (err)
2987                 goto msi_test_failed;
2988
2989         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
2990                           netdev->name, netdev);
2991         if (err) {
2992                 pci_disable_msi(adapter->pdev);
2993                 goto msi_test_failed;
2994         }
2995
2996         wmb();
2997
2998         e1000_irq_enable(adapter);
2999
3000         /* fire an unusual interrupt on the test handler */
3001         ew32(ICS, E1000_ICS_RXSEQ);
3002         e1e_flush();
3003         msleep(50);
3004
3005         e1000_irq_disable(adapter);
3006
3007         rmb();
3008
3009         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3010                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3011                 err = -EIO;
3012                 e_info("MSI interrupt test failed!\n");
3013         }
3014
3015         free_irq(adapter->pdev->irq, netdev);
3016         pci_disable_msi(adapter->pdev);
3017
3018         if (err == -EIO)
3019                 goto msi_test_failed;
3020
3021         /* okay so the test worked, restore settings */
3022         e_dbg("MSI interrupt test succeeded!\n");
3023 msi_test_failed:
3024         e1000e_set_interrupt_capability(adapter);
3025         e1000_request_irq(adapter);
3026         return err;
3027 }
3028
3029 /**
3030  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3031  * @adapter: board private struct
3032  *
3033  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3034  **/
3035 static int e1000_test_msi(struct e1000_adapter *adapter)
3036 {
3037         int err;
3038         u16 pci_cmd;
3039
3040         if (!(adapter->flags & FLAG_MSI_ENABLED))
3041                 return 0;
3042
3043         /* disable SERR in case the MSI write causes a master abort */
3044         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3045         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3046                               pci_cmd & ~PCI_COMMAND_SERR);
3047
3048         err = e1000_test_msi_interrupt(adapter);
3049
3050         /* restore previous setting of command word */
3051         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3052
3053         /* success ! */
3054         if (!err)
3055                 return 0;
3056
3057         /* EIO means MSI test failed */
3058         if (err != -EIO)
3059                 return err;
3060
3061         /* back to INTx mode */
3062         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3063
3064         e1000_free_irq(adapter);
3065
3066         err = e1000_request_irq(adapter);
3067
3068         return err;
3069 }
3070
3071 /**
3072  * e1000_open - Called when a network interface is made active
3073  * @netdev: network interface device structure
3074  *
3075  * Returns 0 on success, negative value on failure
3076  *
3077  * The open entry point is called when a network interface is made
3078  * active by the system (IFF_UP).  At this point all resources needed
3079  * for transmit and receive operations are allocated, the interrupt
3080  * handler is registered with the OS, the watchdog timer is started,
3081  * and the stack is notified that the interface is ready.
3082  **/
3083 static int e1000_open(struct net_device *netdev)
3084 {
3085         struct e1000_adapter *adapter = netdev_priv(netdev);
3086         struct e1000_hw *hw = &adapter->hw;
3087         struct pci_dev *pdev = adapter->pdev;
3088         int err;
3089
3090         /* disallow open during test */
3091         if (test_bit(__E1000_TESTING, &adapter->state))
3092                 return -EBUSY;
3093
3094         pm_runtime_get_sync(&pdev->dev);
3095
3096         netif_carrier_off(netdev);
3097
3098         /* allocate transmit descriptors */
3099         err = e1000e_setup_tx_resources(adapter);
3100         if (err)
3101                 goto err_setup_tx;
3102
3103         /* allocate receive descriptors */
3104         err = e1000e_setup_rx_resources(adapter);
3105         if (err)
3106                 goto err_setup_rx;
3107
3108         e1000e_power_up_phy(adapter);
3109
3110         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3111         if ((adapter->hw.mng_cookie.status &
3112              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3113                 e1000_update_mng_vlan(adapter);
3114
3115         /*
3116          * If AMT is enabled, let the firmware know that the network
3117          * interface is now open
3118          */
3119         if (adapter->flags & FLAG_HAS_AMT)
3120                 e1000_get_hw_control(adapter);
3121
3122         /*
3123          * before we allocate an interrupt, we must be ready to handle it.
3124          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3125          * as soon as we call pci_request_irq, so we have to setup our
3126          * clean_rx handler before we do so.
3127          */
3128         e1000_configure(adapter);
3129
3130         err = e1000_request_irq(adapter);
3131         if (err)
3132                 goto err_req_irq;
3133
3134         /*
3135          * Work around PCIe errata with MSI interrupts causing some chipsets to
3136          * ignore e1000e MSI messages, which means we need to test our MSI
3137          * interrupt now
3138          */
3139         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3140                 err = e1000_test_msi(adapter);
3141                 if (err) {
3142                         e_err("Interrupt allocation failed\n");
3143                         goto err_req_irq;
3144                 }
3145         }
3146
3147         /* From here on the code is the same as e1000e_up() */
3148         clear_bit(__E1000_DOWN, &adapter->state);
3149
3150         napi_enable(&adapter->napi);
3151
3152         e1000_irq_enable(adapter);
3153
3154         netif_start_queue(netdev);
3155
3156         adapter->idle_check = true;
3157         pm_runtime_put(&pdev->dev);
3158
3159         /* fire a link status change interrupt to start the watchdog */
3160         ew32(ICS, E1000_ICS_LSC);
3161
3162         return 0;
3163
3164 err_req_irq:
3165         e1000_release_hw_control(adapter);
3166         e1000_power_down_phy(adapter);
3167         e1000e_free_rx_resources(adapter);
3168 err_setup_rx:
3169         e1000e_free_tx_resources(adapter);
3170 err_setup_tx:
3171         e1000e_reset(adapter);
3172         pm_runtime_put_sync(&pdev->dev);
3173
3174         return err;
3175 }
3176
3177 /**
3178  * e1000_close - Disables a network interface
3179  * @netdev: network interface device structure
3180  *
3181  * Returns 0, this is not allowed to fail
3182  *
3183  * The close entry point is called when an interface is de-activated
3184  * by the OS.  The hardware is still under the drivers control, but
3185  * needs to be disabled.  A global MAC reset is issued to stop the
3186  * hardware, and all transmit and receive resources are freed.
3187  **/
3188 static int e1000_close(struct net_device *netdev)
3189 {
3190         struct e1000_adapter *adapter = netdev_priv(netdev);
3191         struct pci_dev *pdev = adapter->pdev;
3192
3193         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3194
3195         pm_runtime_get_sync(&pdev->dev);
3196
3197         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3198                 e1000e_down(adapter);
3199                 e1000_free_irq(adapter);
3200         }
3201         e1000_power_down_phy(adapter);
3202
3203         e1000e_free_tx_resources(adapter);
3204         e1000e_free_rx_resources(adapter);
3205
3206         /*
3207          * kill manageability vlan ID if supported, but not if a vlan with
3208          * the same ID is registered on the host OS (let 8021q kill it)
3209          */
3210         if ((adapter->hw.mng_cookie.status &
3211                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3212              !(adapter->vlgrp &&
3213                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3214                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3215
3216         /*
3217          * If AMT is enabled, let the firmware know that the network
3218          * interface is now closed
3219          */
3220         if (adapter->flags & FLAG_HAS_AMT)
3221                 e1000_release_hw_control(adapter);
3222
3223         pm_runtime_put_sync(&pdev->dev);
3224
3225         return 0;
3226 }
3227 /**
3228  * e1000_set_mac - Change the Ethernet Address of the NIC
3229  * @netdev: network interface device structure
3230  * @p: pointer to an address structure
3231  *
3232  * Returns 0 on success, negative on failure
3233  **/
3234 static int e1000_set_mac(struct net_device *netdev, void *p)
3235 {
3236         struct e1000_adapter *adapter = netdev_priv(netdev);
3237         struct sockaddr *addr = p;
3238
3239         if (!is_valid_ether_addr(addr->sa_data))
3240                 return -EADDRNOTAVAIL;
3241
3242         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3243         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3244
3245         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3246
3247         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3248                 /* activate the work around */
3249                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3250
3251                 /*
3252                  * Hold a copy of the LAA in RAR[14] This is done so that
3253                  * between the time RAR[0] gets clobbered  and the time it
3254                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3255                  * of the RARs and no incoming packets directed to this port
3256                  * are dropped. Eventually the LAA will be in RAR[0] and
3257                  * RAR[14]
3258                  */
3259                 e1000e_rar_set(&adapter->hw,
3260                               adapter->hw.mac.addr,
3261                               adapter->hw.mac.rar_entry_count - 1);
3262         }
3263
3264         return 0;
3265 }
3266
3267 /**
3268  * e1000e_update_phy_task - work thread to update phy
3269  * @work: pointer to our work struct
3270  *
3271  * this worker thread exists because we must acquire a
3272  * semaphore to read the phy, which we could msleep while
3273  * waiting for it, and we can't msleep in a timer.
3274  **/
3275 static void e1000e_update_phy_task(struct work_struct *work)
3276 {
3277         struct e1000_adapter *adapter = container_of(work,
3278                                         struct e1000_adapter, update_phy_task);
3279         e1000_get_phy_info(&adapter->hw);
3280 }
3281
3282 /*
3283  * Need to wait a few seconds after link up to get diagnostic information from
3284  * the phy
3285  */
3286 static void e1000_update_phy_info(unsigned long data)
3287 {
3288         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3289         schedule_work(&adapter->update_phy_task);
3290 }
3291
3292 /**
3293  * e1000e_update_stats - Update the board statistics counters
3294  * @adapter: board private structure
3295  **/
3296 void e1000e_update_stats(struct e1000_adapter *adapter)
3297 {
3298         struct net_device *netdev = adapter->netdev;
3299         struct e1000_hw *hw = &adapter->hw;
3300         struct pci_dev *pdev = adapter->pdev;
3301         u16 phy_data;
3302
3303         /*
3304          * Prevent stats update while adapter is being reset, or if the pci
3305          * connection is down.
3306          */
3307         if (adapter->link_speed == 0)
3308                 return;
3309         if (pci_channel_offline(pdev))
3310                 return;
3311
3312         adapter->stats.crcerrs += er32(CRCERRS);
3313         adapter->stats.gprc += er32(GPRC);
3314         adapter->stats.gorc += er32(GORCL);
3315         er32(GORCH); /* Clear gorc */
3316         adapter->stats.bprc += er32(BPRC);
3317         adapter->stats.mprc += er32(MPRC);
3318         adapter->stats.roc += er32(ROC);
3319
3320         adapter->stats.mpc += er32(MPC);
3321         if ((hw->phy.type == e1000_phy_82578) ||
3322             (hw->phy.type == e1000_phy_82577)) {
3323                 e1e_rphy(hw, HV_SCC_UPPER, &phy_data);
3324                 if (!e1e_rphy(hw, HV_SCC_LOWER, &phy_data))
3325                         adapter->stats.scc += phy_data;
3326
3327                 e1e_rphy(hw, HV_ECOL_UPPER, &phy_data);
3328                 if (!e1e_rphy(hw, HV_ECOL_LOWER, &phy_data))
3329                         adapter->stats.ecol += phy_data;
3330
3331                 e1e_rphy(hw, HV_MCC_UPPER, &phy_data);
3332                 if (!e1e_rphy(hw, HV_MCC_LOWER, &phy_data))
3333                         adapter->stats.mcc += phy_data;
3334
3335                 e1e_rphy(hw, HV_LATECOL_UPPER, &phy_data);
3336                 if (!e1e_rphy(hw, HV_LATECOL_LOWER, &phy_data))
3337                         adapter->stats.latecol += phy_data;
3338
3339                 e1e_rphy(hw, HV_DC_UPPER, &phy_data);
3340                 if (!e1e_rphy(hw, HV_DC_LOWER, &phy_data))
3341                         adapter->stats.dc += phy_data;
3342         } else {
3343                 adapter->stats.scc += er32(SCC);
3344                 adapter->stats.ecol += er32(ECOL);
3345                 adapter->stats.mcc += er32(MCC);
3346                 adapter->stats.latecol += er32(LATECOL);
3347                 adapter->stats.dc += er32(DC);
3348         }
3349         adapter->stats.xonrxc += er32(XONRXC);
3350         adapter->stats.xontxc += er32(XONTXC);
3351         adapter->stats.xoffrxc += er32(XOFFRXC);
3352         adapter->stats.xofftxc += er32(XOFFTXC);
3353         adapter->stats.gptc += er32(GPTC);
3354         adapter->stats.gotc += er32(GOTCL);
3355         er32(GOTCH); /* Clear gotc */
3356         adapter->stats.rnbc += er32(RNBC);
3357         adapter->stats.ruc += er32(RUC);
3358
3359         adapter->stats.mptc += er32(MPTC);
3360         adapter->stats.bptc += er32(BPTC);
3361
3362         /* used for adaptive IFS */
3363
3364         hw->mac.tx_packet_delta = er32(TPT);
3365         adapter->stats.tpt += hw->mac.tx_packet_delta;
3366         if ((hw->phy.type == e1000_phy_82578) ||
3367             (hw->phy.type == e1000_phy_82577)) {
3368                 e1e_rphy(hw, HV_COLC_UPPER, &phy_data);
3369                 if (!e1e_rphy(hw, HV_COLC_LOWER, &phy_data))
3370                         hw->mac.collision_delta = phy_data;
3371         } else {
3372                 hw->mac.collision_delta = er32(COLC);
3373         }
3374         adapter->stats.colc += hw->mac.collision_delta;
3375
3376         adapter->stats.algnerrc += er32(ALGNERRC);
3377         adapter->stats.rxerrc += er32(RXERRC);
3378         if ((hw->phy.type == e1000_phy_82578) ||
3379             (hw->phy.type == e1000_phy_82577)) {
3380                 e1e_rphy(hw, HV_TNCRS_UPPER, &phy_data);
3381                 if (!e1e_rphy(hw, HV_TNCRS_LOWER, &phy_data))
3382                         adapter->stats.tncrs += phy_data;
3383         } else {
3384                 if ((hw->mac.type != e1000_82574) &&
3385                     (hw->mac.type != e1000_82583))
3386                         adapter->stats.tncrs += er32(TNCRS);
3387         }
3388         adapter->stats.cexterr += er32(CEXTERR);
3389         adapter->stats.tsctc += er32(TSCTC);
3390         adapter->stats.tsctfc += er32(TSCTFC);
3391
3392         /* Fill out the OS statistics structure */
3393         netdev->stats.multicast = adapter->stats.mprc;
3394         netdev->stats.collisions = adapter->stats.colc;
3395
3396         /* Rx Errors */
3397
3398         /*
3399          * RLEC on some newer hardware can be incorrect so build
3400          * our own version based on RUC and ROC
3401          */
3402         netdev->stats.rx_errors = adapter->stats.rxerrc +
3403                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3404                 adapter->stats.ruc + adapter->stats.roc +
3405                 adapter->stats.cexterr;
3406         netdev->stats.rx_length_errors = adapter->stats.ruc +
3407                                               adapter->stats.roc;
3408         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3409         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3410         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3411
3412         /* Tx Errors */
3413         netdev->stats.tx_errors = adapter->stats.ecol +
3414                                        adapter->stats.latecol;
3415         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3416         netdev->stats.tx_window_errors = adapter->stats.latecol;
3417         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3418
3419         /* Tx Dropped needs to be maintained elsewhere */
3420
3421         /* Management Stats */
3422         adapter->stats.mgptc += er32(MGTPTC);
3423         adapter->stats.mgprc += er32(MGTPRC);
3424         adapter->stats.mgpdc += er32(MGTPDC);
3425 }
3426
3427 /**
3428  * e1000_phy_read_status - Update the PHY register status snapshot
3429  * @adapter: board private structure
3430  **/
3431 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3432 {
3433         struct e1000_hw *hw = &adapter->hw;
3434         struct e1000_phy_regs *phy = &adapter->phy_regs;
3435         int ret_val;
3436
3437         if ((er32(STATUS) & E1000_STATUS_LU) &&
3438             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3439                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3440                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3441                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3442                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3443                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3444                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3445                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3446                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3447                 if (ret_val)
3448                         e_warn("Error reading PHY register\n");
3449         } else {
3450                 /*
3451                  * Do not read PHY registers if link is not up
3452                  * Set values to typical power-on defaults
3453                  */
3454                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3455                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3456                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3457                              BMSR_ERCAP);
3458                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3459                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3460                 phy->lpa = 0;
3461                 phy->expansion = EXPANSION_ENABLENPAGE;
3462                 phy->ctrl1000 = ADVERTISE_1000FULL;
3463                 phy->stat1000 = 0;
3464                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3465         }
3466 }
3467
3468 static void e1000_print_link_info(struct e1000_adapter *adapter)
3469 {
3470         struct e1000_hw *hw = &adapter->hw;
3471         u32 ctrl = er32(CTRL);
3472
3473         /* Link status message must follow this format for user tools */
3474         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3475                "Flow Control: %s\n",
3476                adapter->netdev->name,
3477                adapter->link_speed,
3478                (adapter->link_duplex == FULL_DUPLEX) ?
3479                                 "Full Duplex" : "Half Duplex",
3480                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3481                                 "RX/TX" :
3482                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3483                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3484 }
3485
3486 bool e1000e_has_link(struct e1000_adapter *adapter)
3487 {
3488         struct e1000_hw *hw = &adapter->hw;
3489         bool link_active = 0;
3490         s32 ret_val = 0;
3491
3492         /*
3493          * get_link_status is set on LSC (link status) interrupt or
3494          * Rx sequence error interrupt.  get_link_status will stay
3495          * false until the check_for_link establishes link
3496          * for copper adapters ONLY
3497          */
3498         switch (hw->phy.media_type) {
3499         case e1000_media_type_copper:
3500                 if (hw->mac.get_link_status) {
3501                         ret_val = hw->mac.ops.check_for_link(hw);
3502                         link_active = !hw->mac.get_link_status;
3503                 } else {
3504                         link_active = 1;
3505                 }
3506                 break;
3507         case e1000_media_type_fiber:
3508                 ret_val = hw->mac.ops.check_for_link(hw);
3509                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3510                 break;
3511         case e1000_media_type_internal_serdes:
3512                 ret_val = hw->mac.ops.check_for_link(hw);
3513                 link_active = adapter->hw.mac.serdes_has_link;
3514                 break;
3515         default:
3516         case e1000_media_type_unknown:
3517                 break;
3518         }
3519
3520         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3521             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3522                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3523                 e_info("Gigabit has been disabled, downgrading speed\n");
3524         }
3525
3526         return link_active;
3527 }
3528
3529 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3530 {
3531         /* make sure the receive unit is started */
3532         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3533             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3534                 struct e1000_hw *hw = &adapter->hw;
3535                 u32 rctl = er32(RCTL);
3536                 ew32(RCTL, rctl | E1000_RCTL_EN);
3537                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3538         }
3539 }
3540
3541 /**
3542  * e1000_watchdog - Timer Call-back
3543  * @data: pointer to adapter cast into an unsigned long
3544  **/
3545 static void e1000_watchdog(unsigned long data)
3546 {
3547         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3548
3549         /* Do the rest outside of interrupt context */
3550         schedule_work(&adapter->watchdog_task);
3551
3552         /* TODO: make this use queue_delayed_work() */
3553 }
3554
3555 static void e1000_watchdog_task(struct work_struct *work)
3556 {
3557         struct e1000_adapter *adapter = container_of(work,
3558                                         struct e1000_adapter, watchdog_task);
3559         struct net_device *netdev = adapter->netdev;
3560         struct e1000_mac_info *mac = &adapter->hw.mac;
3561         struct e1000_phy_info *phy = &adapter->hw.phy;
3562         struct e1000_ring *tx_ring = adapter->tx_ring;
3563         struct e1000_hw *hw = &adapter->hw;
3564         u32 link, tctl;
3565         int tx_pending = 0;
3566
3567         link = e1000e_has_link(adapter);
3568         if ((netif_carrier_ok(netdev)) && link) {
3569                 /* Cancel scheduled suspend requests. */
3570                 pm_runtime_resume(netdev->dev.parent);
3571
3572                 e1000e_enable_receives(adapter);
3573                 goto link_up;
3574         }
3575
3576         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3577             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3578                 e1000_update_mng_vlan(adapter);
3579
3580         if (link) {
3581                 if (!netif_carrier_ok(netdev)) {
3582                         bool txb2b = 1;
3583
3584                         /* Cancel scheduled suspend requests. */
3585                         pm_runtime_resume(netdev->dev.parent);
3586
3587                         /* update snapshot of PHY registers on LSC */
3588                         e1000_phy_read_status(adapter);
3589                         mac->ops.get_link_up_info(&adapter->hw,
3590                                                    &adapter->link_speed,
3591                                                    &adapter->link_duplex);
3592                         e1000_print_link_info(adapter);
3593                         /*
3594                          * On supported PHYs, check for duplex mismatch only
3595                          * if link has autonegotiated at 10/100 half
3596                          */
3597                         if ((hw->phy.type == e1000_phy_igp_3 ||
3598                              hw->phy.type == e1000_phy_bm) &&
3599                             (hw->mac.autoneg == true) &&
3600                             (adapter->link_speed == SPEED_10 ||
3601                              adapter->link_speed == SPEED_100) &&
3602                             (adapter->link_duplex == HALF_DUPLEX)) {
3603                                 u16 autoneg_exp;
3604
3605                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3606
3607                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3608                                         e_info("Autonegotiated half duplex but"
3609                                                " link partner cannot autoneg. "
3610                                                " Try forcing full duplex if "
3611                                                "link gets many collisions.\n");
3612                         }
3613
3614                         /*
3615                          * tweak tx_queue_len according to speed/duplex
3616                          * and adjust the timeout factor
3617                          */
3618                         netdev->tx_queue_len = adapter->tx_queue_len;
3619                         adapter->tx_timeout_factor = 1;
3620                         switch (adapter->link_speed) {
3621                         case SPEED_10:
3622                                 txb2b = 0;
3623                                 netdev->tx_queue_len = 10;
3624                                 adapter->tx_timeout_factor = 16;
3625                                 break;
3626                         case SPEED_100:
3627                                 txb2b = 0;
3628                                 netdev->tx_queue_len = 100;
3629                                 adapter->tx_timeout_factor = 10;
3630                                 break;
3631                         }
3632
3633                         /*
3634                          * workaround: re-program speed mode bit after
3635                          * link-up event
3636                          */
3637                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3638                             !txb2b) {
3639                                 u32 tarc0;
3640                                 tarc0 = er32(TARC(0));
3641                                 tarc0 &= ~SPEED_MODE_BIT;
3642                                 ew32(TARC(0), tarc0);
3643                         }
3644
3645                         /*
3646                          * disable TSO for pcie and 10/100 speeds, to avoid
3647                          * some hardware issues
3648                          */
3649                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3650                                 switch (adapter->link_speed) {
3651                                 case SPEED_10:
3652                                 case SPEED_100:
3653                                         e_info("10/100 speed: disabling TSO\n");
3654                                         netdev->features &= ~NETIF_F_TSO;
3655                                         netdev->features &= ~NETIF_F_TSO6;
3656                                         break;
3657                                 case SPEED_1000:
3658                                         netdev->features |= NETIF_F_TSO;
3659                                         netdev->features |= NETIF_F_TSO6;
3660                                         break;
3661                                 default:
3662                                         /* oops */
3663                                         break;
3664                                 }
3665                         }
3666
3667                         /*
3668                          * enable transmits in the hardware, need to do this
3669                          * after setting TARC(0)
3670                          */
3671                         tctl = er32(TCTL);
3672                         tctl |= E1000_TCTL_EN;
3673                         ew32(TCTL, tctl);
3674
3675                         /*
3676                          * Perform any post-link-up configuration before
3677                          * reporting link up.
3678                          */
3679                         if (phy->ops.cfg_on_link_up)
3680                                 phy->ops.cfg_on_link_up(hw);
3681
3682                         netif_carrier_on(netdev);
3683
3684                         if (!test_bit(__E1000_DOWN, &adapter->state))
3685                                 mod_timer(&adapter->phy_info_timer,
3686                                           round_jiffies(jiffies + 2 * HZ));
3687                 }
3688         } else {
3689                 if (netif_carrier_ok(netdev)) {
3690                         adapter->link_speed = 0;
3691                         adapter->link_duplex = 0;
3692                         /* Link status message must follow this format */
3693                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
3694                                adapter->netdev->name);
3695                         netif_carrier_off(netdev);
3696                         if (!test_bit(__E1000_DOWN, &adapter->state))
3697                                 mod_timer(&adapter->phy_info_timer,
3698                                           round_jiffies(jiffies + 2 * HZ));
3699
3700                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3701                                 schedule_work(&adapter->reset_task);
3702                         else
3703                                 pm_schedule_suspend(netdev->dev.parent,
3704                                                         LINK_TIMEOUT);
3705                 }
3706         }
3707
3708 link_up:
3709         e1000e_update_stats(adapter);
3710
3711         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3712         adapter->tpt_old = adapter->stats.tpt;
3713         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3714         adapter->colc_old = adapter->stats.colc;
3715
3716         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3717         adapter->gorc_old = adapter->stats.gorc;
3718         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3719         adapter->gotc_old = adapter->stats.gotc;
3720
3721         e1000e_update_adaptive(&adapter->hw);
3722
3723         if (!netif_carrier_ok(netdev)) {
3724                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3725                                tx_ring->count);
3726                 if (tx_pending) {
3727                         /*
3728                          * We've lost link, so the controller stops DMA,
3729                          * but we've got queued Tx work that's never going
3730                          * to get done, so reset controller to flush Tx.
3731                          * (Do the reset outside of interrupt context).
3732                          */
3733                         adapter->tx_timeout_count++;
3734                         schedule_work(&adapter->reset_task);
3735                         /* return immediately since reset is imminent */
3736                         return;
3737                 }
3738         }
3739
3740         /* Cause software interrupt to ensure Rx ring is cleaned */
3741         if (adapter->msix_entries)
3742                 ew32(ICS, adapter->rx_ring->ims_val);
3743         else
3744                 ew32(ICS, E1000_ICS_RXDMT0);
3745
3746         /* Force detection of hung controller every watchdog period */
3747         adapter->detect_tx_hung = 1;
3748
3749         /*
3750          * With 82571 controllers, LAA may be overwritten due to controller
3751          * reset from the other port. Set the appropriate LAA in RAR[0]
3752          */
3753         if (e1000e_get_laa_state_82571(hw))
3754                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3755
3756         /* Reset the timer */
3757         if (!test_bit(__E1000_DOWN, &adapter->state))
3758                 mod_timer(&adapter->watchdog_timer,
3759                           round_jiffies(jiffies + 2 * HZ));
3760 }
3761
3762 #define E1000_TX_FLAGS_CSUM             0x00000001
3763 #define E1000_TX_FLAGS_VLAN             0x00000002
3764 #define E1000_TX_FLAGS_TSO              0x00000004
3765 #define E1000_TX_FLAGS_IPV4             0x00000008
3766 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3767 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3768
3769 static int e1000_tso(struct e1000_adapter *adapter,
3770                      struct sk_buff *skb)
3771 {
3772         struct e1000_ring *tx_ring = adapter->tx_ring;
3773         struct e1000_context_desc *context_desc;
3774         struct e1000_buffer *buffer_info;
3775         unsigned int i;
3776         u32 cmd_length = 0;
3777         u16 ipcse = 0, tucse, mss;
3778         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3779         int err;
3780
3781         if (!skb_is_gso(skb))
3782                 return 0;
3783
3784         if (skb_header_cloned(skb)) {
3785                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3786                 if (err)
3787                         return err;
3788         }
3789
3790         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3791         mss = skb_shinfo(skb)->gso_size;
3792         if (skb->protocol == htons(ETH_P_IP)) {
3793                 struct iphdr *iph = ip_hdr(skb);
3794                 iph->tot_len = 0;
3795                 iph->check = 0;
3796                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
3797                                                          0, IPPROTO_TCP, 0);
3798                 cmd_length = E1000_TXD_CMD_IP;
3799                 ipcse = skb_transport_offset(skb) - 1;
3800         } else if (skb_is_gso_v6(skb)) {
3801                 ipv6_hdr(skb)->payload_len = 0;
3802                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3803                                                        &ipv6_hdr(skb)->daddr,
3804                                                        0, IPPROTO_TCP, 0);
3805                 ipcse = 0;
3806         }
3807         ipcss = skb_network_offset(skb);
3808         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3809         tucss = skb_transport_offset(skb);
3810         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3811         tucse = 0;
3812
3813         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3814                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3815
3816         i = tx_ring->next_to_use;
3817         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3818         buffer_info = &tx_ring->buffer_info[i];
3819
3820         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3821         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3822         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3823         context_desc->upper_setup.tcp_fields.tucss = tucss;
3824         context_desc->upper_setup.tcp_fields.tucso = tucso;
3825         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3826         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3827         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3828         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3829
3830         buffer_info->time_stamp = jiffies;
3831         buffer_info->next_to_watch = i;
3832
3833         i++;
3834         if (i == tx_ring->count)
3835                 i = 0;
3836         tx_ring->next_to_use = i;
3837
3838         return 1;
3839 }
3840
3841 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3842 {
3843         struct e1000_ring *tx_ring = adapter->tx_ring;
3844         struct e1000_context_desc *context_desc;
3845         struct e1000_buffer *buffer_info;
3846         unsigned int i;
3847         u8 css;
3848         u32 cmd_len = E1000_TXD_CMD_DEXT;
3849         __be16 protocol;
3850
3851         if (skb->ip_summed != CHECKSUM_PARTIAL)
3852                 return 0;
3853
3854         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
3855                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
3856         else
3857                 protocol = skb->protocol;
3858
3859         switch (protocol) {
3860         case cpu_to_be16(ETH_P_IP):
3861                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3862                         cmd_len |= E1000_TXD_CMD_TCP;
3863                 break;
3864         case cpu_to_be16(ETH_P_IPV6):
3865                 /* XXX not handling all IPV6 headers */
3866                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3867                         cmd_len |= E1000_TXD_CMD_TCP;
3868                 break;
3869         default:
3870                 if (unlikely(net_ratelimit()))
3871                         e_warn("checksum_partial proto=%x!\n",
3872                                be16_to_cpu(protocol));
3873                 break;
3874         }
3875
3876         css = skb_transport_offset(skb);
3877
3878         i = tx_ring->next_to_use;
3879         buffer_info = &tx_ring->buffer_info[i];
3880         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3881
3882         context_desc->lower_setup.ip_config = 0;
3883         context_desc->upper_setup.tcp_fields.tucss = css;
3884         context_desc->upper_setup.tcp_fields.tucso =
3885                                 css + skb->csum_offset;
3886         context_desc->upper_setup.tcp_fields.tucse = 0;
3887         context_desc->tcp_seg_setup.data = 0;
3888         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
3889
3890         buffer_info->time_stamp = jiffies;
3891         buffer_info->next_to_watch = i;
3892
3893         i++;
3894         if (i == tx_ring->count)
3895                 i = 0;
3896         tx_ring->next_to_use = i;
3897
3898         return 1;
3899 }
3900
3901 #define E1000_MAX_PER_TXD       8192
3902 #define E1000_MAX_TXD_PWR       12
3903
3904 static int e1000_tx_map(struct e1000_adapter *adapter,
3905                         struct sk_buff *skb, unsigned int first,
3906                         unsigned int max_per_txd, unsigned int nr_frags,
3907                         unsigned int mss)
3908 {
3909         struct e1000_ring *tx_ring = adapter->tx_ring;
3910         struct pci_dev *pdev = adapter->pdev;
3911         struct e1000_buffer *buffer_info;
3912         unsigned int len = skb_headlen(skb);
3913         unsigned int offset = 0, size, count = 0, i;
3914         unsigned int f;
3915
3916         i = tx_ring->next_to_use;
3917
3918         while (len) {
3919                 buffer_info = &tx_ring->buffer_info[i];
3920                 size = min(len, max_per_txd);
3921
3922                 buffer_info->length = size;
3923                 buffer_info->time_stamp = jiffies;
3924                 buffer_info->next_to_watch = i;
3925                 buffer_info->dma = pci_map_single(pdev, skb->data + offset,
3926                                                   size, PCI_DMA_TODEVICE);
3927                 buffer_info->mapped_as_page = false;
3928                 if (pci_dma_mapping_error(pdev, buffer_info->dma))
3929                         goto dma_error;
3930
3931                 len -= size;
3932                 offset += size;
3933                 count++;
3934
3935                 if (len) {
3936                         i++;
3937                         if (i == tx_ring->count)
3938                                 i = 0;
3939                 }
3940         }
3941
3942         for (f = 0; f < nr_frags; f++) {
3943                 struct skb_frag_struct *frag;
3944
3945                 frag = &skb_shinfo(skb)->frags[f];
3946                 len = frag->size;
3947                 offset = frag->page_offset;
3948
3949                 while (len) {
3950                         i++;
3951                         if (i == tx_ring->count)
3952                                 i = 0;
3953
3954                         buffer_info = &tx_ring->buffer_info[i];
3955                         size = min(len, max_per_txd);
3956
3957                         buffer_info->length = size;
3958                         buffer_info->time_stamp = jiffies;
3959                         buffer_info->next_to_watch = i;
3960                         buffer_info->dma = pci_map_page(pdev, frag->page,
3961                                                         offset, size,
3962                                                         PCI_DMA_TODEVICE);
3963                         buffer_info->mapped_as_page = true;
3964                         if (pci_dma_mapping_error(pdev, buffer_info->dma))
3965                                 goto dma_error;
3966
3967                         len -= size;
3968                         offset += size;
3969                         count++;
3970                 }
3971         }
3972
3973         tx_ring->buffer_info[i].skb = skb;
3974         tx_ring->buffer_info[first].next_to_watch = i;
3975
3976         return count;
3977
3978 dma_error:
3979         dev_err(&pdev->dev, "TX DMA map failed\n");
3980         buffer_info->dma = 0;
3981         if (count)
3982                 count--;
3983
3984         while (count--) {
3985                 if (i==0)
3986                         i += tx_ring->count;
3987                 i--;
3988                 buffer_info = &tx_ring->buffer_info[i];
3989                 e1000_put_txbuf(adapter, buffer_info);;
3990         }
3991
3992         return 0;
3993 }
3994
3995 static void e1000_tx_queue(struct e1000_adapter *adapter,
3996                            int tx_flags, int count)
3997 {
3998         struct e1000_ring *tx_ring = adapter->tx_ring;
3999         struct e1000_tx_desc *tx_desc = NULL;
4000         struct e1000_buffer *buffer_info;
4001         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4002         unsigned int i;
4003
4004         if (tx_flags & E1000_TX_FLAGS_TSO) {
4005                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4006                              E1000_TXD_CMD_TSE;
4007                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4008
4009                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4010                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4011         }
4012
4013         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4014                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4015                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4016         }
4017
4018         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4019                 txd_lower |= E1000_TXD_CMD_VLE;
4020                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4021         }
4022
4023         i = tx_ring->next_to_use;
4024
4025         while (count--) {
4026                 buffer_info = &tx_ring->buffer_info[i];
4027                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4028                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4029                 tx_desc->lower.data =
4030                         cpu_to_le32(txd_lower | buffer_info->length);
4031                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4032
4033                 i++;
4034                 if (i == tx_ring->count)
4035                         i = 0;
4036         }
4037
4038         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4039
4040         /*
4041          * Force memory writes to complete before letting h/w
4042          * know there are new descriptors to fetch.  (Only
4043          * applicable for weak-ordered memory model archs,
4044          * such as IA-64).
4045          */
4046         wmb();
4047
4048         tx_ring->next_to_use = i;
4049         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4050         /*
4051          * we need this if more than one processor can write to our tail
4052          * at a time, it synchronizes IO on IA64/Altix systems
4053          */
4054         mmiowb();
4055 }
4056
4057 #define MINIMUM_DHCP_PACKET_SIZE 282
4058 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4059                                     struct sk_buff *skb)
4060 {
4061         struct e1000_hw *hw =  &adapter->hw;
4062         u16 length, offset;
4063
4064         if (vlan_tx_tag_present(skb)) {
4065                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4066                     (adapter->hw.mng_cookie.status &
4067                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4068                         return 0;
4069         }
4070
4071         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4072                 return 0;
4073
4074         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4075                 return 0;
4076
4077         {
4078                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4079                 struct udphdr *udp;
4080
4081                 if (ip->protocol != IPPROTO_UDP)
4082                         return 0;
4083
4084                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4085                 if (ntohs(udp->dest) != 67)
4086                         return 0;
4087
4088                 offset = (u8 *)udp + 8 - skb->data;
4089                 length = skb->len - offset;
4090                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4091         }
4092
4093         return 0;
4094 }
4095
4096 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4097 {
4098         struct e1000_adapter *adapter = netdev_priv(netdev);
4099
4100         netif_stop_queue(netdev);
4101         /*
4102          * Herbert's original patch had:
4103          *  smp_mb__after_netif_stop_queue();
4104          * but since that doesn't exist yet, just open code it.
4105          */
4106         smp_mb();
4107
4108         /*
4109          * We need to check again in a case another CPU has just
4110          * made room available.
4111          */
4112         if (e1000_desc_unused(adapter->tx_ring) < size)
4113                 return -EBUSY;
4114
4115         /* A reprieve! */
4116         netif_start_queue(netdev);
4117         ++adapter->restart_queue;
4118         return 0;
4119 }
4120
4121 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4122 {
4123         struct e1000_adapter *adapter = netdev_priv(netdev);
4124
4125         if (e1000_desc_unused(adapter->tx_ring) >= size)
4126                 return 0;
4127         return __e1000_maybe_stop_tx(netdev, size);
4128 }
4129
4130 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4131 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4132                                     struct net_device *netdev)
4133 {
4134         struct e1000_adapter *adapter = netdev_priv(netdev);
4135         struct e1000_ring *tx_ring = adapter->tx_ring;
4136         unsigned int first;
4137         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4138         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4139         unsigned int tx_flags = 0;
4140         unsigned int len = skb->len - skb->data_len;
4141         unsigned int nr_frags;
4142         unsigned int mss;
4143         int count = 0;
4144         int tso;
4145         unsigned int f;
4146
4147         if (test_bit(__E1000_DOWN, &adapter->state)) {
4148                 dev_kfree_skb_any(skb);
4149                 return NETDEV_TX_OK;
4150         }
4151
4152         if (skb->len <= 0) {
4153                 dev_kfree_skb_any(skb);
4154                 return NETDEV_TX_OK;
4155         }
4156
4157         mss = skb_shinfo(skb)->gso_size;
4158         /*
4159          * The controller does a simple calculation to
4160          * make sure there is enough room in the FIFO before
4161          * initiating the DMA for each buffer.  The calc is:
4162          * 4 = ceil(buffer len/mss).  To make sure we don't
4163          * overrun the FIFO, adjust the max buffer len if mss
4164          * drops.
4165          */
4166         if (mss) {
4167                 u8 hdr_len;
4168                 max_per_txd = min(mss << 2, max_per_txd);
4169                 max_txd_pwr = fls(max_per_txd) - 1;
4170
4171                 /*
4172                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4173                  * points to just header, pull a few bytes of payload from
4174                  * frags into skb->data
4175                  */
4176                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4177                 /*
4178                  * we do this workaround for ES2LAN, but it is un-necessary,
4179                  * avoiding it could save a lot of cycles
4180                  */
4181                 if (skb->data_len && (hdr_len == len)) {
4182                         unsigned int pull_size;
4183
4184                         pull_size = min((unsigned int)4, skb->data_len);
4185                         if (!__pskb_pull_tail(skb, pull_size)) {
4186                                 e_err("__pskb_pull_tail failed.\n");
4187                                 dev_kfree_skb_any(skb);
4188                                 return NETDEV_TX_OK;
4189                         }
4190                         len = skb->len - skb->data_len;
4191                 }
4192         }
4193
4194         /* reserve a descriptor for the offload context */
4195         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4196                 count++;
4197         count++;
4198
4199         count += TXD_USE_COUNT(len, max_txd_pwr);
4200
4201         nr_frags = skb_shinfo(skb)->nr_frags;
4202         for (f = 0; f < nr_frags; f++)
4203                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4204                                        max_txd_pwr);
4205
4206         if (adapter->hw.mac.tx_pkt_filtering)
4207                 e1000_transfer_dhcp_info(adapter, skb);
4208
4209         /*
4210          * need: count + 2 desc gap to keep tail from touching
4211          * head, otherwise try next time
4212          */
4213         if (e1000_maybe_stop_tx(netdev, count + 2))
4214                 return NETDEV_TX_BUSY;
4215
4216         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4217                 tx_flags |= E1000_TX_FLAGS_VLAN;
4218                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4219         }
4220
4221         first = tx_ring->next_to_use;
4222
4223         tso = e1000_tso(adapter, skb);
4224         if (tso < 0) {
4225                 dev_kfree_skb_any(skb);
4226                 return NETDEV_TX_OK;
4227         }
4228
4229         if (tso)
4230                 tx_flags |= E1000_TX_FLAGS_TSO;
4231         else if (e1000_tx_csum(adapter, skb))
4232                 tx_flags |= E1000_TX_FLAGS_CSUM;
4233
4234         /*
4235          * Old method was to assume IPv4 packet by default if TSO was enabled.
4236          * 82571 hardware supports TSO capabilities for IPv6 as well...
4237          * no longer assume, we must.
4238          */
4239         if (skb->protocol == htons(ETH_P_IP))
4240                 tx_flags |= E1000_TX_FLAGS_IPV4;
4241
4242         /* if count is 0 then mapping error has occured */
4243         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4244         if (count) {
4245                 e1000_tx_queue(adapter, tx_flags, count);
4246                 /* Make sure there is space in the ring for the next send. */
4247                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4248
4249         } else {
4250                 dev_kfree_skb_any(skb);
4251                 tx_ring->buffer_info[first].time_stamp = 0;
4252                 tx_ring->next_to_use = first;
4253         }
4254
4255         return NETDEV_TX_OK;
4256 }
4257
4258 /**
4259  * e1000_tx_timeout - Respond to a Tx Hang
4260  * @netdev: network interface device structure
4261  **/
4262 static void e1000_tx_timeout(struct net_device *netdev)
4263 {
4264         struct e1000_adapter *adapter = netdev_priv(netdev);
4265
4266         /* Do the reset outside of interrupt context */
4267         adapter->tx_timeout_count++;
4268         schedule_work(&adapter->reset_task);
4269 }
4270
4271 static void e1000_reset_task(struct work_struct *work)
4272 {
4273         struct e1000_adapter *adapter;
4274         adapter = container_of(work, struct e1000_adapter, reset_task);
4275
4276         e1000e_reinit_locked(adapter);
4277 }
4278
4279 /**
4280  * e1000_get_stats - Get System Network Statistics
4281  * @netdev: network interface device structure
4282  *
4283  * Returns the address of the device statistics structure.
4284  * The statistics are actually updated from the timer callback.
4285  **/
4286 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4287 {
4288         /* only return the current stats */
4289         return &netdev->stats;
4290 }
4291
4292 /**
4293  * e1000_change_mtu - Change the Maximum Transfer Unit
4294  * @netdev: network interface device structure
4295  * @new_mtu: new value for maximum frame size
4296  *
4297  * Returns 0 on success, negative on failure
4298  **/
4299 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4300 {
4301         struct e1000_adapter *adapter = netdev_priv(netdev);
4302         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4303
4304         /* Jumbo frame support */
4305         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4306             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4307                 e_err("Jumbo Frames not supported.\n");
4308                 return -EINVAL;
4309         }
4310
4311         /* Supported frame sizes */
4312         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4313             (max_frame > adapter->max_hw_frame_size)) {
4314                 e_err("Unsupported MTU setting\n");
4315                 return -EINVAL;
4316         }
4317
4318         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4319                 msleep(1);
4320         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
4321         adapter->max_frame_size = max_frame;
4322         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4323         netdev->mtu = new_mtu;
4324         if (netif_running(netdev))
4325                 e1000e_down(adapter);
4326
4327         /*
4328          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4329          * means we reserve 2 more, this pushes us to allocate from the next
4330          * larger slab size.
4331          * i.e. RXBUFFER_2048 --> size-4096 slab
4332          * However with the new *_jumbo_rx* routines, jumbo receives will use
4333          * fragmented skbs
4334          */
4335
4336         if (max_frame <= 2048)
4337                 adapter->rx_buffer_len = 2048;
4338         else
4339                 adapter->rx_buffer_len = 4096;
4340
4341         /* adjust allocation if LPE protects us, and we aren't using SBP */
4342         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4343              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4344                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4345                                          + ETH_FCS_LEN;
4346
4347         if (netif_running(netdev))
4348                 e1000e_up(adapter);
4349         else
4350                 e1000e_reset(adapter);
4351
4352         clear_bit(__E1000_RESETTING, &adapter->state);
4353
4354         return 0;
4355 }
4356
4357 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4358                            int cmd)
4359 {
4360         struct e1000_adapter *adapter = netdev_priv(netdev);
4361         struct mii_ioctl_data *data = if_mii(ifr);
4362
4363         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4364                 return -EOPNOTSUPP;
4365
4366         switch (cmd) {
4367         case SIOCGMIIPHY:
4368                 data->phy_id = adapter->hw.phy.addr;
4369                 break;
4370         case SIOCGMIIREG:
4371                 e1000_phy_read_status(adapter);
4372
4373                 switch (data->reg_num & 0x1F) {
4374                 case MII_BMCR:
4375                         data->val_out = adapter->phy_regs.bmcr;
4376                         break;
4377                 case MII_BMSR:
4378                         data->val_out = adapter->phy_regs.bmsr;
4379                         break;
4380                 case MII_PHYSID1:
4381                         data->val_out = (adapter->hw.phy.id >> 16);
4382                         break;
4383                 case MII_PHYSID2:
4384                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4385                         break;
4386                 case MII_ADVERTISE:
4387                         data->val_out = adapter->phy_regs.advertise;
4388                         break;
4389                 case MII_LPA:
4390                         data->val_out = adapter->phy_regs.lpa;
4391                         break;
4392                 case MII_EXPANSION:
4393                         data->val_out = adapter->phy_regs.expansion;
4394                         break;
4395                 case MII_CTRL1000:
4396                         data->val_out = adapter->phy_regs.ctrl1000;
4397                         break;
4398                 case MII_STAT1000:
4399                         data->val_out = adapter->phy_regs.stat1000;
4400                         break;
4401                 case MII_ESTATUS:
4402                         data->val_out = adapter->phy_regs.estatus;
4403                         break;
4404                 default:
4405                         return -EIO;
4406                 }
4407                 break;
4408         case SIOCSMIIREG:
4409         default:
4410                 return -EOPNOTSUPP;
4411         }
4412         return 0;
4413 }
4414
4415 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4416 {
4417         switch (cmd) {
4418         case SIOCGMIIPHY:
4419         case SIOCGMIIREG:
4420         case SIOCSMIIREG:
4421                 return e1000_mii_ioctl(netdev, ifr, cmd);
4422         default:
4423                 return -EOPNOTSUPP;
4424         }
4425 }
4426
4427 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
4428 {
4429         struct e1000_hw *hw = &adapter->hw;
4430         u32 i, mac_reg;
4431         u16 phy_reg;
4432         int retval = 0;
4433
4434         /* copy MAC RARs to PHY RARs */
4435         for (i = 0; i < adapter->hw.mac.rar_entry_count; i++) {
4436                 mac_reg = er32(RAL(i));
4437                 e1e_wphy(hw, BM_RAR_L(i), (u16)(mac_reg & 0xFFFF));
4438                 e1e_wphy(hw, BM_RAR_M(i), (u16)((mac_reg >> 16) & 0xFFFF));
4439                 mac_reg = er32(RAH(i));
4440                 e1e_wphy(hw, BM_RAR_H(i), (u16)(mac_reg & 0xFFFF));
4441                 e1e_wphy(hw, BM_RAR_CTRL(i), (u16)((mac_reg >> 16) & 0xFFFF));
4442         }
4443
4444         /* copy MAC MTA to PHY MTA */
4445         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
4446                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
4447                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
4448                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
4449         }
4450
4451         /* configure PHY Rx Control register */
4452         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
4453         mac_reg = er32(RCTL);
4454         if (mac_reg & E1000_RCTL_UPE)
4455                 phy_reg |= BM_RCTL_UPE;
4456         if (mac_reg & E1000_RCTL_MPE)
4457                 phy_reg |= BM_RCTL_MPE;
4458         phy_reg &= ~(BM_RCTL_MO_MASK);
4459         if (mac_reg & E1000_RCTL_MO_3)
4460                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
4461                                 << BM_RCTL_MO_SHIFT);
4462         if (mac_reg & E1000_RCTL_BAM)
4463                 phy_reg |= BM_RCTL_BAM;
4464         if (mac_reg & E1000_RCTL_PMCF)
4465                 phy_reg |= BM_RCTL_PMCF;
4466         mac_reg = er32(CTRL);
4467         if (mac_reg & E1000_CTRL_RFCE)
4468                 phy_reg |= BM_RCTL_RFCE;
4469         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
4470
4471         /* enable PHY wakeup in MAC register */
4472         ew32(WUFC, wufc);
4473         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
4474
4475         /* configure and enable PHY wakeup in PHY registers */
4476         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
4477         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
4478
4479         /* activate PHY wakeup */
4480         retval = hw->phy.ops.acquire(hw);
4481         if (retval) {
4482                 e_err("Could not acquire PHY\n");
4483                 return retval;
4484         }
4485         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4486                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
4487         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
4488         if (retval) {
4489                 e_err("Could not read PHY page 769\n");
4490                 goto out;
4491         }
4492         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
4493         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
4494         if (retval)
4495                 e_err("Could not set PHY Host Wakeup bit\n");
4496 out:
4497         hw->phy.ops.release(hw);
4498
4499         return retval;
4500 }
4501
4502 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
4503                             bool runtime)
4504 {
4505         struct net_device *netdev = pci_get_drvdata(pdev);
4506         struct e1000_adapter *adapter = netdev_priv(netdev);
4507         struct e1000_hw *hw = &adapter->hw;
4508         u32 ctrl, ctrl_ext, rctl, status;
4509         /* Runtime suspend should only enable wakeup for link changes */
4510         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
4511         int retval = 0;
4512
4513         netif_device_detach(netdev);
4514
4515         if (netif_running(netdev)) {
4516                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4517                 e1000e_down(adapter);
4518                 e1000_free_irq(adapter);
4519         }
4520         e1000e_reset_interrupt_capability(adapter);
4521
4522         retval = pci_save_state(pdev);
4523         if (retval)
4524                 return retval;
4525
4526         status = er32(STATUS);
4527         if (status & E1000_STATUS_LU)
4528                 wufc &= ~E1000_WUFC_LNKC;
4529
4530         if (wufc) {
4531                 e1000_setup_rctl(adapter);
4532                 e1000_set_multi(netdev);
4533
4534                 /* turn on all-multi mode if wake on multicast is enabled */
4535                 if (wufc & E1000_WUFC_MC) {
4536                         rctl = er32(RCTL);
4537                         rctl |= E1000_RCTL_MPE;
4538                         ew32(RCTL, rctl);
4539                 }
4540
4541                 ctrl = er32(CTRL);
4542                 /* advertise wake from D3Cold */
4543                 #define E1000_CTRL_ADVD3WUC 0x00100000
4544                 /* phy power management enable */
4545                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4546                 ctrl |= E1000_CTRL_ADVD3WUC;
4547                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
4548                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
4549                 ew32(CTRL, ctrl);
4550
4551                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4552                     adapter->hw.phy.media_type ==
4553                     e1000_media_type_internal_serdes) {
4554                         /* keep the laser running in D3 */
4555                         ctrl_ext = er32(CTRL_EXT);
4556                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
4557                         ew32(CTRL_EXT, ctrl_ext);
4558                 }
4559
4560                 if (adapter->flags & FLAG_IS_ICH)
4561                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4562
4563                 /* Allow time for pending master requests to run */
4564                 e1000e_disable_pcie_master(&adapter->hw);
4565
4566                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4567                         /* enable wakeup by the PHY */
4568                         retval = e1000_init_phy_wakeup(adapter, wufc);
4569                         if (retval)
4570                                 return retval;
4571                 } else {
4572                         /* enable wakeup by the MAC */
4573                         ew32(WUFC, wufc);
4574                         ew32(WUC, E1000_WUC_PME_EN);
4575                 }
4576         } else {
4577                 ew32(WUC, 0);
4578                 ew32(WUFC, 0);
4579         }
4580
4581         *enable_wake = !!wufc;
4582
4583         /* make sure adapter isn't asleep if manageability is enabled */
4584         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
4585             (hw->mac.ops.check_mng_mode(hw)))
4586                 *enable_wake = true;
4587
4588         if (adapter->hw.phy.type == e1000_phy_igp_3)
4589                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4590
4591         /*
4592          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4593          * would have already happened in close and is redundant.
4594          */
4595         e1000_release_hw_control(adapter);
4596
4597         pci_disable_device(pdev);
4598
4599         return 0;
4600 }
4601
4602 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
4603 {
4604         if (sleep && wake) {
4605                 pci_prepare_to_sleep(pdev);
4606                 return;
4607         }
4608
4609         pci_wake_from_d3(pdev, wake);
4610         pci_set_power_state(pdev, PCI_D3hot);
4611 }
4612
4613 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
4614                                     bool wake)
4615 {
4616         struct net_device *netdev = pci_get_drvdata(pdev);
4617         struct e1000_adapter *adapter = netdev_priv(netdev);
4618
4619         /*
4620          * The pci-e switch on some quad port adapters will report a
4621          * correctable error when the MAC transitions from D0 to D3.  To
4622          * prevent this we need to mask off the correctable errors on the
4623          * downstream port of the pci-e switch.
4624          */
4625         if (adapter->flags & FLAG_IS_QUAD_PORT) {
4626                 struct pci_dev *us_dev = pdev->bus->self;
4627                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
4628                 u16 devctl;
4629
4630                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
4631                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
4632                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
4633
4634                 e1000_power_off(pdev, sleep, wake);
4635
4636                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
4637         } else {
4638                 e1000_power_off(pdev, sleep, wake);
4639         }
4640 }
4641
4642 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4643 {
4644         int pos;
4645         u16 val;
4646
4647         /*
4648          * 82573 workaround - disable L1 ASPM on mobile chipsets
4649          *
4650          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4651          * resulting in lost data or garbage information on the pci-e link
4652          * level. This could result in (false) bad EEPROM checksum errors,
4653          * long ping times (up to 2s) or even a system freeze/hang.
4654          *
4655          * Unfortunately this feature saves about 1W power consumption when
4656          * active.
4657          */
4658         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4659         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4660         if (val & 0x2) {
4661                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4662                 val &= ~0x2;
4663                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4664         }
4665 }
4666
4667 #ifdef CONFIG_PM
4668 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
4669 {
4670         return !!adapter->tx_ring->buffer_info;
4671 }
4672
4673 static int e1000_idle(struct device *dev)
4674 {
4675         struct pci_dev *pdev = to_pci_dev(dev);
4676         struct net_device *netdev = pci_get_drvdata(pdev);
4677         struct e1000_adapter *adapter = netdev_priv(netdev);
4678
4679         if (!e1000e_pm_ready(adapter))
4680                 return 0;
4681
4682         if (adapter->idle_check) {
4683                 adapter->idle_check = false;
4684                 if (!e1000e_has_link(adapter))
4685                         pm_schedule_suspend(dev, MSEC_PER_SEC);
4686         }
4687
4688         return -EBUSY;
4689 }
4690
4691 static int e1000_suspend(struct device *dev)
4692 {
4693         struct pci_dev *pdev = to_pci_dev(dev);
4694         int retval;
4695         bool wake;
4696
4697         retval = __e1000_shutdown(pdev, &wake, false);
4698         if (!retval)
4699                 e1000_complete_shutdown(pdev, true, wake);
4700
4701         return retval;
4702 }
4703
4704 static int e1000_runtime_suspend(struct device *dev)
4705 {
4706         struct pci_dev *pdev = to_pci_dev(dev);
4707         struct net_device *netdev = pci_get_drvdata(pdev);
4708         struct e1000_adapter *adapter = netdev_priv(netdev);
4709
4710         if (e1000e_pm_ready(adapter)) {
4711                 bool wake;
4712
4713                 __e1000_shutdown(pdev, &wake, true);
4714         }
4715
4716         return 0;
4717 }
4718
4719 static int __e1000_resume(struct pci_dev *pdev)
4720 {
4721         struct net_device *netdev = pci_get_drvdata(pdev);
4722         struct e1000_adapter *adapter = netdev_priv(netdev);
4723         struct e1000_hw *hw = &adapter->hw;
4724         u32 err;
4725
4726         e1000e_disable_l1aspm(pdev);
4727
4728         e1000e_set_interrupt_capability(adapter);
4729         if (netif_running(netdev)) {
4730                 err = e1000_request_irq(adapter);
4731                 if (err)
4732                         return err;
4733         }
4734
4735         e1000e_power_up_phy(adapter);
4736
4737         /* report the system wakeup cause from S3/S4 */
4738         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4739                 u16 phy_data;
4740
4741                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
4742                 if (phy_data) {
4743                         e_info("PHY Wakeup cause - %s\n",
4744                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
4745                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
4746                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
4747                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
4748                                 phy_data & E1000_WUS_LNKC ? "Link Status "
4749                                 " Change" : "other");
4750                 }
4751                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
4752         } else {
4753                 u32 wus = er32(WUS);
4754                 if (wus) {
4755                         e_info("MAC Wakeup cause - %s\n",
4756                                 wus & E1000_WUS_EX ? "Unicast Packet" :
4757                                 wus & E1000_WUS_MC ? "Multicast Packet" :
4758                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
4759                                 wus & E1000_WUS_MAG ? "Magic Packet" :
4760                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
4761                                 "other");
4762                 }
4763                 ew32(WUS, ~0);
4764         }
4765
4766         e1000e_reset(adapter);
4767
4768         e1000_init_manageability(adapter);
4769
4770         if (netif_running(netdev))
4771                 e1000e_up(adapter);
4772
4773         netif_device_attach(netdev);
4774
4775         /*
4776          * If the controller has AMT, do not set DRV_LOAD until the interface
4777          * is up.  For all other cases, let the f/w know that the h/w is now
4778          * under the control of the driver.
4779          */
4780         if (!(adapter->flags & FLAG_HAS_AMT))
4781                 e1000_get_hw_control(adapter);
4782
4783         return 0;
4784 }
4785
4786 static int e1000_resume(struct device *dev)
4787 {
4788         struct pci_dev *pdev = to_pci_dev(dev);
4789         struct net_device *netdev = pci_get_drvdata(pdev);
4790         struct e1000_adapter *adapter = netdev_priv(netdev);
4791
4792         if (e1000e_pm_ready(adapter))
4793                 adapter->idle_check = true;
4794
4795         return __e1000_resume(pdev);
4796 }
4797
4798 static int e1000_runtime_resume(struct device *dev)
4799 {
4800         struct pci_dev *pdev = to_pci_dev(dev);
4801         struct net_device *netdev = pci_get_drvdata(pdev);
4802         struct e1000_adapter *adapter = netdev_priv(netdev);
4803
4804         if (!e1000e_pm_ready(adapter))
4805                 return 0;
4806
4807         adapter->idle_check = !dev->power.runtime_auto;
4808         return __e1000_resume(pdev);
4809 }
4810 #endif
4811
4812 static void e1000_shutdown(struct pci_dev *pdev)
4813 {
4814         bool wake = false;
4815
4816         __e1000_shutdown(pdev, &wake, false);
4817
4818         if (system_state == SYSTEM_POWER_OFF)
4819                 e1000_complete_shutdown(pdev, false, wake);
4820 }
4821
4822 #ifdef CONFIG_NET_POLL_CONTROLLER
4823 /*
4824  * Polling 'interrupt' - used by things like netconsole to send skbs
4825  * without having to re-enable interrupts. It's not called while
4826  * the interrupt routine is executing.
4827  */
4828 static void e1000_netpoll(struct net_device *netdev)
4829 {
4830         struct e1000_adapter *adapter = netdev_priv(netdev);
4831
4832         disable_irq(adapter->pdev->irq);
4833         e1000_intr(adapter->pdev->irq, netdev);
4834
4835         enable_irq(adapter->pdev->irq);
4836 }
4837 #endif
4838
4839 /**
4840  * e1000_io_error_detected - called when PCI error is detected
4841  * @pdev: Pointer to PCI device
4842  * @state: The current pci connection state
4843  *
4844  * This function is called after a PCI bus error affecting
4845  * this device has been detected.
4846  */
4847 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4848                                                 pci_channel_state_t state)
4849 {
4850         struct net_device *netdev = pci_get_drvdata(pdev);
4851         struct e1000_adapter *adapter = netdev_priv(netdev);
4852
4853         netif_device_detach(netdev);
4854
4855         if (state == pci_channel_io_perm_failure)
4856                 return PCI_ERS_RESULT_DISCONNECT;
4857
4858         if (netif_running(netdev))
4859                 e1000e_down(adapter);
4860         pci_disable_device(pdev);
4861
4862         /* Request a slot slot reset. */
4863         return PCI_ERS_RESULT_NEED_RESET;
4864 }
4865
4866 /**
4867  * e1000_io_slot_reset - called after the pci bus has been reset.
4868  * @pdev: Pointer to PCI device
4869  *
4870  * Restart the card from scratch, as if from a cold-boot. Implementation
4871  * resembles the first-half of the e1000_resume routine.
4872  */
4873 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4874 {
4875         struct net_device *netdev = pci_get_drvdata(pdev);
4876         struct e1000_adapter *adapter = netdev_priv(netdev);
4877         struct e1000_hw *hw = &adapter->hw;
4878         int err;
4879         pci_ers_result_t result;
4880
4881         e1000e_disable_l1aspm(pdev);
4882         err = pci_enable_device_mem(pdev);
4883         if (err) {
4884                 dev_err(&pdev->dev,
4885                         "Cannot re-enable PCI device after reset.\n");
4886                 result = PCI_ERS_RESULT_DISCONNECT;
4887         } else {
4888                 pci_set_master(pdev);
4889                 pdev->state_saved = true;
4890                 pci_restore_state(pdev);
4891
4892                 pci_enable_wake(pdev, PCI_D3hot, 0);
4893                 pci_enable_wake(pdev, PCI_D3cold, 0);
4894
4895                 e1000e_reset(adapter);
4896                 ew32(WUS, ~0);
4897                 result = PCI_ERS_RESULT_RECOVERED;
4898         }
4899
4900         pci_cleanup_aer_uncorrect_error_status(pdev);
4901
4902         return result;
4903 }
4904
4905 /**
4906  * e1000_io_resume - called when traffic can start flowing again.
4907  * @pdev: Pointer to PCI device
4908  *
4909  * This callback is called when the error recovery driver tells us that
4910  * its OK to resume normal operation. Implementation resembles the
4911  * second-half of the e1000_resume routine.
4912  */
4913 static void e1000_io_resume(struct pci_dev *pdev)
4914 {
4915         struct net_device *netdev = pci_get_drvdata(pdev);
4916         struct e1000_adapter *adapter = netdev_priv(netdev);
4917
4918         e1000_init_manageability(adapter);
4919
4920         if (netif_running(netdev)) {
4921                 if (e1000e_up(adapter)) {
4922                         dev_err(&pdev->dev,
4923                                 "can't bring device back up after reset\n");
4924                         return;
4925                 }
4926         }
4927
4928         netif_device_attach(netdev);
4929
4930         /*
4931          * If the controller has AMT, do not set DRV_LOAD until the interface
4932          * is up.  For all other cases, let the f/w know that the h/w is now
4933          * under the control of the driver.
4934          */
4935         if (!(adapter->flags & FLAG_HAS_AMT))
4936                 e1000_get_hw_control(adapter);
4937
4938 }
4939
4940 static void e1000_print_device_info(struct e1000_adapter *adapter)
4941 {
4942         struct e1000_hw *hw = &adapter->hw;
4943         struct net_device *netdev = adapter->netdev;
4944         u32 pba_num;
4945
4946         /* print bus type/speed/width info */
4947         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
4948                /* bus width */
4949                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4950                 "Width x1"),
4951                /* MAC address */
4952                netdev->dev_addr);
4953         e_info("Intel(R) PRO/%s Network Connection\n",
4954                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4955         e1000e_read_pba_num(hw, &pba_num);
4956         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4957                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4958 }
4959
4960 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4961 {
4962         struct e1000_hw *hw = &adapter->hw;
4963         int ret_val;
4964         u16 buf = 0;
4965
4966         if (hw->mac.type != e1000_82573)
4967                 return;
4968
4969         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4970         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
4971                 /* Deep Smart Power Down (DSPD) */
4972                 dev_warn(&adapter->pdev->dev,
4973                          "Warning: detected DSPD enabled in EEPROM\n");
4974         }
4975
4976         ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4977         if (!ret_val && (le16_to_cpu(buf) & (3 << 2))) {
4978                 /* ASPM enable */
4979                 dev_warn(&adapter->pdev->dev,
4980                          "Warning: detected ASPM enabled in EEPROM\n");
4981         }
4982 }
4983
4984 static const struct net_device_ops e1000e_netdev_ops = {
4985         .ndo_open               = e1000_open,
4986         .ndo_stop               = e1000_close,
4987         .ndo_start_xmit         = e1000_xmit_frame,
4988         .ndo_get_stats          = e1000_get_stats,
4989         .ndo_set_multicast_list = e1000_set_multi,
4990         .ndo_set_mac_address    = e1000_set_mac,
4991         .ndo_change_mtu         = e1000_change_mtu,
4992         .ndo_do_ioctl           = e1000_ioctl,
4993         .ndo_tx_timeout         = e1000_tx_timeout,
4994         .ndo_validate_addr      = eth_validate_addr,
4995
4996         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
4997         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
4998         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
4999 #ifdef CONFIG_NET_POLL_CONTROLLER
5000         .ndo_poll_controller    = e1000_netpoll,
5001 #endif
5002 };
5003
5004 /**
5005  * e1000_probe - Device Initialization Routine
5006  * @pdev: PCI device information struct
5007  * @ent: entry in e1000_pci_tbl
5008  *
5009  * Returns 0 on success, negative on failure
5010  *
5011  * e1000_probe initializes an adapter identified by a pci_dev structure.
5012  * The OS initialization, configuring of the adapter private structure,
5013  * and a hardware reset occur.
5014  **/
5015 static int __devinit e1000_probe(struct pci_dev *pdev,
5016                                  const struct pci_device_id *ent)
5017 {
5018         struct net_device *netdev;
5019         struct e1000_adapter *adapter;
5020         struct e1000_hw *hw;
5021         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5022         resource_size_t mmio_start, mmio_len;
5023         resource_size_t flash_start, flash_len;
5024
5025         static int cards_found;
5026         int i, err, pci_using_dac;
5027         u16 eeprom_data = 0;
5028         u16 eeprom_apme_mask = E1000_EEPROM_APME;
5029
5030         e1000e_disable_l1aspm(pdev);
5031
5032         err = pci_enable_device_mem(pdev);
5033         if (err)
5034                 return err;
5035
5036         pci_using_dac = 0;
5037         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
5038         if (!err) {
5039                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
5040                 if (!err)
5041                         pci_using_dac = 1;
5042         } else {
5043                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
5044                 if (err) {
5045                         err = pci_set_consistent_dma_mask(pdev,
5046                                                           DMA_BIT_MASK(32));
5047                         if (err) {
5048                                 dev_err(&pdev->dev, "No usable DMA "
5049                                         "configuration, aborting\n");
5050                                 goto err_dma;
5051                         }
5052                 }
5053         }
5054
5055         err = pci_request_selected_regions_exclusive(pdev,
5056                                           pci_select_bars(pdev, IORESOURCE_MEM),
5057                                           e1000e_driver_name);
5058         if (err)
5059                 goto err_pci_reg;
5060
5061         /* AER (Advanced Error Reporting) hooks */
5062         pci_enable_pcie_error_reporting(pdev);
5063
5064         pci_set_master(pdev);
5065         /* PCI config space info */
5066         err = pci_save_state(pdev);
5067         if (err)
5068                 goto err_alloc_etherdev;
5069
5070         err = -ENOMEM;
5071         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5072         if (!netdev)
5073                 goto err_alloc_etherdev;
5074
5075         SET_NETDEV_DEV(netdev, &pdev->dev);
5076
5077         pci_set_drvdata(pdev, netdev);
5078         adapter = netdev_priv(netdev);
5079         hw = &adapter->hw;
5080         adapter->netdev = netdev;
5081         adapter->pdev = pdev;
5082         adapter->ei = ei;
5083         adapter->pba = ei->pba;
5084         adapter->flags = ei->flags;
5085         adapter->flags2 = ei->flags2;
5086         adapter->hw.adapter = adapter;
5087         adapter->hw.mac.type = ei->mac;
5088         adapter->max_hw_frame_size = ei->max_hw_frame_size;
5089         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5090
5091         mmio_start = pci_resource_start(pdev, 0);
5092         mmio_len = pci_resource_len(pdev, 0);
5093
5094         err = -EIO;
5095         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5096         if (!adapter->hw.hw_addr)
5097                 goto err_ioremap;
5098
5099         if ((adapter->flags & FLAG_HAS_FLASH) &&
5100             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5101                 flash_start = pci_resource_start(pdev, 1);
5102                 flash_len = pci_resource_len(pdev, 1);
5103                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5104                 if (!adapter->hw.flash_address)
5105                         goto err_flashmap;
5106         }
5107
5108         /* construct the net_device struct */
5109         netdev->netdev_ops              = &e1000e_netdev_ops;
5110         e1000e_set_ethtool_ops(netdev);
5111         netdev->watchdog_timeo          = 5 * HZ;
5112         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5113         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5114
5115         netdev->mem_start = mmio_start;
5116         netdev->mem_end = mmio_start + mmio_len;
5117
5118         adapter->bd_number = cards_found++;
5119
5120         e1000e_check_options(adapter);
5121
5122         /* setup adapter struct */
5123         err = e1000_sw_init(adapter);
5124         if (err)
5125                 goto err_sw_init;
5126
5127         err = -EIO;
5128
5129         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5130         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5131         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5132
5133         err = ei->get_variants(adapter);
5134         if (err)
5135                 goto err_hw_init;
5136
5137         if ((adapter->flags & FLAG_IS_ICH) &&
5138             (adapter->flags & FLAG_READ_ONLY_NVM))
5139                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5140
5141         hw->mac.ops.get_bus_info(&adapter->hw);
5142
5143         adapter->hw.phy.autoneg_wait_to_complete = 0;
5144
5145         /* Copper options */
5146         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5147                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5148                 adapter->hw.phy.disable_polarity_correction = 0;
5149                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5150         }
5151
5152         if (e1000_check_reset_block(&adapter->hw))
5153                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5154
5155         netdev->features = NETIF_F_SG |
5156                            NETIF_F_HW_CSUM |
5157                            NETIF_F_HW_VLAN_TX |
5158                            NETIF_F_HW_VLAN_RX;
5159
5160         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5161                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5162
5163         netdev->features |= NETIF_F_TSO;
5164         netdev->features |= NETIF_F_TSO6;
5165
5166         netdev->vlan_features |= NETIF_F_TSO;
5167         netdev->vlan_features |= NETIF_F_TSO6;
5168         netdev->vlan_features |= NETIF_F_HW_CSUM;
5169         netdev->vlan_features |= NETIF_F_SG;
5170
5171         if (pci_using_dac)
5172                 netdev->features |= NETIF_F_HIGHDMA;
5173
5174         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5175                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5176
5177         /*
5178          * before reading the NVM, reset the controller to
5179          * put the device in a known good starting state
5180          */
5181         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5182
5183         /*
5184          * systems with ASPM and others may see the checksum fail on the first
5185          * attempt. Let's give it a few tries
5186          */
5187         for (i = 0;; i++) {
5188                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5189                         break;
5190                 if (i == 2) {
5191                         e_err("The NVM Checksum Is Not Valid\n");
5192                         err = -EIO;
5193                         goto err_eeprom;
5194                 }
5195         }
5196
5197         e1000_eeprom_checks(adapter);
5198
5199         /* copy the MAC address */
5200         if (e1000e_read_mac_addr(&adapter->hw))
5201                 e_err("NVM Read Error while reading MAC address\n");
5202
5203         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5204         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5205
5206         if (!is_valid_ether_addr(netdev->perm_addr)) {
5207                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5208                 err = -EIO;
5209                 goto err_eeprom;
5210         }
5211
5212         init_timer(&adapter->watchdog_timer);
5213         adapter->watchdog_timer.function = &e1000_watchdog;
5214         adapter->watchdog_timer.data = (unsigned long) adapter;
5215
5216         init_timer(&adapter->phy_info_timer);
5217         adapter->phy_info_timer.function = &e1000_update_phy_info;
5218         adapter->phy_info_timer.data = (unsigned long) adapter;
5219
5220         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5221         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5222         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5223         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5224         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5225
5226         /* Initialize link parameters. User can change them with ethtool */
5227         adapter->hw.mac.autoneg = 1;
5228         adapter->fc_autoneg = 1;
5229         adapter->hw.fc.requested_mode = e1000_fc_default;
5230         adapter->hw.fc.current_mode = e1000_fc_default;
5231         adapter->hw.phy.autoneg_advertised = 0x2f;
5232
5233         /* ring size defaults */
5234         adapter->rx_ring->count = 256;
5235         adapter->tx_ring->count = 256;
5236
5237         /*
5238          * Initial Wake on LAN setting - If APM wake is enabled in
5239          * the EEPROM, enable the ACPI Magic Packet filter
5240          */
5241         if (adapter->flags & FLAG_APME_IN_WUC) {
5242                 /* APME bit in EEPROM is mapped to WUC.APME */
5243                 eeprom_data = er32(WUC);
5244                 eeprom_apme_mask = E1000_WUC_APME;
5245                 if (eeprom_data & E1000_WUC_PHY_WAKE)
5246                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5247         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5248                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5249                     (adapter->hw.bus.func == 1))
5250                         e1000_read_nvm(&adapter->hw,
5251                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5252                 else
5253                         e1000_read_nvm(&adapter->hw,
5254                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5255         }
5256
5257         /* fetch WoL from EEPROM */
5258         if (eeprom_data & eeprom_apme_mask)
5259                 adapter->eeprom_wol |= E1000_WUFC_MAG;
5260
5261         /*
5262          * now that we have the eeprom settings, apply the special cases
5263          * where the eeprom may be wrong or the board simply won't support
5264          * wake on lan on a particular port
5265          */
5266         if (!(adapter->flags & FLAG_HAS_WOL))
5267                 adapter->eeprom_wol = 0;
5268
5269         /* initialize the wol settings based on the eeprom settings */
5270         adapter->wol = adapter->eeprom_wol;
5271         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5272
5273         /* save off EEPROM version number */
5274         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5275
5276         /* reset the hardware with the new settings */
5277         e1000e_reset(adapter);
5278
5279         /*
5280          * If the controller has AMT, do not set DRV_LOAD until the interface
5281          * is up.  For all other cases, let the f/w know that the h/w is now
5282          * under the control of the driver.
5283          */
5284         if (!(adapter->flags & FLAG_HAS_AMT))
5285                 e1000_get_hw_control(adapter);
5286
5287         strcpy(netdev->name, "eth%d");
5288         err = register_netdev(netdev);
5289         if (err)
5290                 goto err_register;
5291
5292         /* carrier off reporting is important to ethtool even BEFORE open */
5293         netif_carrier_off(netdev);
5294
5295         e1000_print_device_info(adapter);
5296
5297         if (pci_dev_run_wake(pdev)) {
5298                 pm_runtime_set_active(&pdev->dev);
5299                 pm_runtime_enable(&pdev->dev);
5300         }
5301         pm_schedule_suspend(&pdev->dev, MSEC_PER_SEC);
5302
5303         return 0;
5304
5305 err_register:
5306         if (!(adapter->flags & FLAG_HAS_AMT))
5307                 e1000_release_hw_control(adapter);
5308 err_eeprom:
5309         if (!e1000_check_reset_block(&adapter->hw))
5310                 e1000_phy_hw_reset(&adapter->hw);
5311 err_hw_init:
5312
5313         kfree(adapter->tx_ring);
5314         kfree(adapter->rx_ring);
5315 err_sw_init:
5316         if (adapter->hw.flash_address)
5317                 iounmap(adapter->hw.flash_address);
5318         e1000e_reset_interrupt_capability(adapter);
5319 err_flashmap:
5320         iounmap(adapter->hw.hw_addr);
5321 err_ioremap:
5322         free_netdev(netdev);
5323 err_alloc_etherdev:
5324         pci_release_selected_regions(pdev,
5325                                      pci_select_bars(pdev, IORESOURCE_MEM));
5326 err_pci_reg:
5327 err_dma:
5328         pci_disable_device(pdev);
5329         return err;
5330 }
5331
5332 /**
5333  * e1000_remove - Device Removal Routine
5334  * @pdev: PCI device information struct
5335  *
5336  * e1000_remove is called by the PCI subsystem to alert the driver
5337  * that it should release a PCI device.  The could be caused by a
5338  * Hot-Plug event, or because the driver is going to be removed from
5339  * memory.
5340  **/
5341 static void __devexit e1000_remove(struct pci_dev *pdev)
5342 {
5343         struct net_device *netdev = pci_get_drvdata(pdev);
5344         struct e1000_adapter *adapter = netdev_priv(netdev);
5345         bool down = test_bit(__E1000_DOWN, &adapter->state);
5346
5347         pm_runtime_get_sync(&pdev->dev);
5348
5349         /*
5350          * flush_scheduled work may reschedule our watchdog task, so
5351          * explicitly disable watchdog tasks from being rescheduled
5352          */
5353         if (!down)
5354                 set_bit(__E1000_DOWN, &adapter->state);
5355         del_timer_sync(&adapter->watchdog_timer);
5356         del_timer_sync(&adapter->phy_info_timer);
5357
5358         cancel_work_sync(&adapter->reset_task);
5359         cancel_work_sync(&adapter->watchdog_task);
5360         cancel_work_sync(&adapter->downshift_task);
5361         cancel_work_sync(&adapter->update_phy_task);
5362         cancel_work_sync(&adapter->print_hang_task);
5363         flush_scheduled_work();
5364
5365         if (!(netdev->flags & IFF_UP))
5366                 e1000_power_down_phy(adapter);
5367
5368         /* Don't lie to e1000_close() down the road. */
5369         if (!down)
5370                 clear_bit(__E1000_DOWN, &adapter->state);
5371         unregister_netdev(netdev);
5372
5373         if (pci_dev_run_wake(pdev)) {
5374                 pm_runtime_disable(&pdev->dev);
5375                 pm_runtime_set_suspended(&pdev->dev);
5376         }
5377         pm_runtime_put_noidle(&pdev->dev);
5378
5379         /*
5380          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5381          * would have already happened in close and is redundant.
5382          */
5383         e1000_release_hw_control(adapter);
5384
5385         e1000e_reset_interrupt_capability(adapter);
5386         kfree(adapter->tx_ring);
5387         kfree(adapter->rx_ring);
5388
5389         iounmap(adapter->hw.hw_addr);
5390         if (adapter->hw.flash_address)
5391                 iounmap(adapter->hw.flash_address);
5392         pci_release_selected_regions(pdev,
5393                                      pci_select_bars(pdev, IORESOURCE_MEM));
5394
5395         free_netdev(netdev);
5396
5397         /* AER disable */
5398         pci_disable_pcie_error_reporting(pdev);
5399
5400         pci_disable_device(pdev);
5401 }
5402
5403 /* PCI Error Recovery (ERS) */
5404 static struct pci_error_handlers e1000_err_handler = {
5405         .error_detected = e1000_io_error_detected,
5406         .slot_reset = e1000_io_slot_reset,
5407         .resume = e1000_io_resume,
5408 };
5409
5410 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
5411         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5412         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5413         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5414         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5415         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5416         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5417         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5418         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5419         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5420
5421         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5422         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5423         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5424         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5425
5426         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5427         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5428         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5429
5430         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5431         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
5432         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
5433
5434         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5435           board_80003es2lan },
5436         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5437           board_80003es2lan },
5438         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5439           board_80003es2lan },
5440         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5441           board_80003es2lan },
5442
5443         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5444         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5445         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5446         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5447         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5448         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5449         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5450         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
5451
5452         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5453         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5454         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5455         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5456         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5457         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5458         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5459         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5460         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5461
5462         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5463         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5464         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5465
5466         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5467         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5468
5469         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
5470         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
5471         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
5472         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
5473
5474         { }     /* terminate list */
5475 };
5476 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5477
5478 static const struct dev_pm_ops e1000_pm_ops = {
5479         .suspend  = e1000_suspend,
5480         .resume   = e1000_resume,
5481         .freeze = e1000_suspend,
5482         .thaw = e1000_resume,
5483         .poweroff = e1000_suspend,
5484         .restore = e1000_resume,
5485         .runtime_suspend = e1000_runtime_suspend,
5486         .runtime_resume = e1000_runtime_resume,
5487         .runtime_idle = e1000_idle,
5488 };
5489
5490 /* PCI Device API Driver */
5491 static struct pci_driver e1000_driver = {
5492         .name     = e1000e_driver_name,
5493         .id_table = e1000_pci_tbl,
5494         .probe    = e1000_probe,
5495         .remove   = __devexit_p(e1000_remove),
5496 #ifdef CONFIG_PM
5497         .driver.pm = &e1000_pm_ops,
5498 #endif
5499         .shutdown = e1000_shutdown,
5500         .err_handler = &e1000_err_handler
5501 };
5502
5503 /**
5504  * e1000_init_module - Driver Registration Routine
5505  *
5506  * e1000_init_module is the first routine called when the driver is
5507  * loaded. All it does is register with the PCI subsystem.
5508  **/
5509 static int __init e1000_init_module(void)
5510 {
5511         int ret;
5512         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
5513                e1000e_driver_name, e1000e_driver_version);
5514         printk(KERN_INFO "%s: Copyright (c) 1999 - 2009 Intel Corporation.\n",
5515                e1000e_driver_name);
5516         ret = pci_register_driver(&e1000_driver);
5517
5518         return ret;
5519 }
5520 module_init(e1000_init_module);
5521
5522 /**
5523  * e1000_exit_module - Driver Exit Cleanup Routine
5524  *
5525  * e1000_exit_module is called just before the driver is removed
5526  * from memory.
5527  **/
5528 static void __exit e1000_exit_module(void)
5529 {
5530         pci_unregister_driver(&e1000_driver);
5531 }
5532 module_exit(e1000_exit_module);
5533
5534
5535 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5536 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5537 MODULE_LICENSE("GPL");
5538 MODULE_VERSION(DRV_VERSION);
5539
5540 /* e1000_main.c */