Merge git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
[linux-2.6] / drivers / net / e1000e / netdev.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2007 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #include <linux/module.h>
30 #include <linux/types.h>
31 #include <linux/init.h>
32 #include <linux/pci.h>
33 #include <linux/vmalloc.h>
34 #include <linux/pagemap.h>
35 #include <linux/delay.h>
36 #include <linux/netdevice.h>
37 #include <linux/tcp.h>
38 #include <linux/ipv6.h>
39 #include <net/checksum.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/mii.h>
42 #include <linux/ethtool.h>
43 #include <linux/if_vlan.h>
44 #include <linux/cpu.h>
45 #include <linux/smp.h>
46
47 #include "e1000.h"
48
49 #define DRV_VERSION "0.2.0"
50 char e1000e_driver_name[] = "e1000e";
51 const char e1000e_driver_version[] = DRV_VERSION;
52
53 static const struct e1000_info *e1000_info_tbl[] = {
54         [board_82571]           = &e1000_82571_info,
55         [board_82572]           = &e1000_82572_info,
56         [board_82573]           = &e1000_82573_info,
57         [board_80003es2lan]     = &e1000_es2_info,
58         [board_ich8lan]         = &e1000_ich8_info,
59         [board_ich9lan]         = &e1000_ich9_info,
60 };
61
62 #ifdef DEBUG
63 /**
64  * e1000_get_hw_dev_name - return device name string
65  * used by hardware layer to print debugging information
66  **/
67 char *e1000e_get_hw_dev_name(struct e1000_hw *hw)
68 {
69         return hw->adapter->netdev->name;
70 }
71 #endif
72
73 /**
74  * e1000_desc_unused - calculate if we have unused descriptors
75  **/
76 static int e1000_desc_unused(struct e1000_ring *ring)
77 {
78         if (ring->next_to_clean > ring->next_to_use)
79                 return ring->next_to_clean - ring->next_to_use - 1;
80
81         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
82 }
83
84 /**
85  * e1000_receive_skb - helper function to handle rx indications
86  * @adapter: board private structure
87  * @status: descriptor status field as written by hardware
88  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
89  * @skb: pointer to sk_buff to be indicated to stack
90  **/
91 static void e1000_receive_skb(struct e1000_adapter *adapter,
92                               struct net_device *netdev,
93                               struct sk_buff *skb,
94                               u8 status, u16 vlan)
95 {
96         skb->protocol = eth_type_trans(skb, netdev);
97
98         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
99                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
100                                          le16_to_cpu(vlan) &
101                                          E1000_RXD_SPC_VLAN_MASK);
102         else
103                 netif_receive_skb(skb);
104
105         netdev->last_rx = jiffies;
106 }
107
108 /**
109  * e1000_rx_checksum - Receive Checksum Offload for 82543
110  * @adapter:     board private structure
111  * @status_err:  receive descriptor status and error fields
112  * @csum:       receive descriptor csum field
113  * @sk_buff:     socket buffer with received data
114  **/
115 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
116                               u32 csum, struct sk_buff *skb)
117 {
118         u16 status = (u16)status_err;
119         u8 errors = (u8)(status_err >> 24);
120         skb->ip_summed = CHECKSUM_NONE;
121
122         /* Ignore Checksum bit is set */
123         if (status & E1000_RXD_STAT_IXSM)
124                 return;
125         /* TCP/UDP checksum error bit is set */
126         if (errors & E1000_RXD_ERR_TCPE) {
127                 /* let the stack verify checksum errors */
128                 adapter->hw_csum_err++;
129                 return;
130         }
131
132         /* TCP/UDP Checksum has not been calculated */
133         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
134                 return;
135
136         /* It must be a TCP or UDP packet with a valid checksum */
137         if (status & E1000_RXD_STAT_TCPCS) {
138                 /* TCP checksum is good */
139                 skb->ip_summed = CHECKSUM_UNNECESSARY;
140         } else {
141                 /* IP fragment with UDP payload */
142                 /* Hardware complements the payload checksum, so we undo it
143                  * and then put the value in host order for further stack use.
144                  */
145                 csum = ntohl(csum ^ 0xFFFF);
146                 skb->csum = csum;
147                 skb->ip_summed = CHECKSUM_COMPLETE;
148         }
149         adapter->hw_csum_good++;
150 }
151
152 /**
153  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
154  * @adapter: address of board private structure
155  **/
156 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
157                                    int cleaned_count)
158 {
159         struct net_device *netdev = adapter->netdev;
160         struct pci_dev *pdev = adapter->pdev;
161         struct e1000_ring *rx_ring = adapter->rx_ring;
162         struct e1000_rx_desc *rx_desc;
163         struct e1000_buffer *buffer_info;
164         struct sk_buff *skb;
165         unsigned int i;
166         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
167
168         i = rx_ring->next_to_use;
169         buffer_info = &rx_ring->buffer_info[i];
170
171         while (cleaned_count--) {
172                 skb = buffer_info->skb;
173                 if (skb) {
174                         skb_trim(skb, 0);
175                         goto map_skb;
176                 }
177
178                 skb = netdev_alloc_skb(netdev, bufsz);
179                 if (!skb) {
180                         /* Better luck next round */
181                         adapter->alloc_rx_buff_failed++;
182                         break;
183                 }
184
185                 /* Make buffer alignment 2 beyond a 16 byte boundary
186                  * this will result in a 16 byte aligned IP header after
187                  * the 14 byte MAC header is removed
188                  */
189                 skb_reserve(skb, NET_IP_ALIGN);
190
191                 buffer_info->skb = skb;
192 map_skb:
193                 buffer_info->dma = pci_map_single(pdev, skb->data,
194                                                   adapter->rx_buffer_len,
195                                                   PCI_DMA_FROMDEVICE);
196                 if (pci_dma_mapping_error(buffer_info->dma)) {
197                         dev_err(&pdev->dev, "RX DMA map failed\n");
198                         adapter->rx_dma_failed++;
199                         break;
200                 }
201
202                 rx_desc = E1000_RX_DESC(*rx_ring, i);
203                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
204
205                 i++;
206                 if (i == rx_ring->count)
207                         i = 0;
208                 buffer_info = &rx_ring->buffer_info[i];
209         }
210
211         if (rx_ring->next_to_use != i) {
212                 rx_ring->next_to_use = i;
213                 if (i-- == 0)
214                         i = (rx_ring->count - 1);
215
216                 /* Force memory writes to complete before letting h/w
217                  * know there are new descriptors to fetch.  (Only
218                  * applicable for weak-ordered memory model archs,
219                  * such as IA-64). */
220                 wmb();
221                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
222         }
223 }
224
225 /**
226  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
227  * @adapter: address of board private structure
228  **/
229 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
230                                       int cleaned_count)
231 {
232         struct net_device *netdev = adapter->netdev;
233         struct pci_dev *pdev = adapter->pdev;
234         union e1000_rx_desc_packet_split *rx_desc;
235         struct e1000_ring *rx_ring = adapter->rx_ring;
236         struct e1000_buffer *buffer_info;
237         struct e1000_ps_page *ps_page;
238         struct sk_buff *skb;
239         unsigned int i, j;
240
241         i = rx_ring->next_to_use;
242         buffer_info = &rx_ring->buffer_info[i];
243
244         while (cleaned_count--) {
245                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
246
247                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
248                         ps_page = &buffer_info->ps_pages[j];
249                         if (j >= adapter->rx_ps_pages) {
250                                 /* all unused desc entries get hw null ptr */
251                                 rx_desc->read.buffer_addr[j+1] = ~0;
252                                 continue;
253                         }
254                         if (!ps_page->page) {
255                                 ps_page->page = alloc_page(GFP_ATOMIC);
256                                 if (!ps_page->page) {
257                                         adapter->alloc_rx_buff_failed++;
258                                         goto no_buffers;
259                                 }
260                                 ps_page->dma = pci_map_page(pdev,
261                                                    ps_page->page,
262                                                    0, PAGE_SIZE,
263                                                    PCI_DMA_FROMDEVICE);
264                                 if (pci_dma_mapping_error(ps_page->dma)) {
265                                         dev_err(&adapter->pdev->dev,
266                                           "RX DMA page map failed\n");
267                                         adapter->rx_dma_failed++;
268                                         goto no_buffers;
269                                 }
270                         }
271                         /*
272                          * Refresh the desc even if buffer_addrs
273                          * didn't change because each write-back
274                          * erases this info.
275                          */
276                         rx_desc->read.buffer_addr[j+1] =
277                              cpu_to_le64(ps_page->dma);
278                 }
279
280                 skb = netdev_alloc_skb(netdev,
281                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
282
283                 if (!skb) {
284                         adapter->alloc_rx_buff_failed++;
285                         break;
286                 }
287
288                 /* Make buffer alignment 2 beyond a 16 byte boundary
289                  * this will result in a 16 byte aligned IP header after
290                  * the 14 byte MAC header is removed
291                  */
292                 skb_reserve(skb, NET_IP_ALIGN);
293
294                 buffer_info->skb = skb;
295                 buffer_info->dma = pci_map_single(pdev, skb->data,
296                                                   adapter->rx_ps_bsize0,
297                                                   PCI_DMA_FROMDEVICE);
298                 if (pci_dma_mapping_error(buffer_info->dma)) {
299                         dev_err(&pdev->dev, "RX DMA map failed\n");
300                         adapter->rx_dma_failed++;
301                         /* cleanup skb */
302                         dev_kfree_skb_any(skb);
303                         buffer_info->skb = NULL;
304                         break;
305                 }
306
307                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
308
309                 i++;
310                 if (i == rx_ring->count)
311                         i = 0;
312                 buffer_info = &rx_ring->buffer_info[i];
313         }
314
315 no_buffers:
316         if (rx_ring->next_to_use != i) {
317                 rx_ring->next_to_use = i;
318
319                 if (!(i--))
320                         i = (rx_ring->count - 1);
321
322                 /* Force memory writes to complete before letting h/w
323                  * know there are new descriptors to fetch.  (Only
324                  * applicable for weak-ordered memory model archs,
325                  * such as IA-64). */
326                 wmb();
327                 /* Hardware increments by 16 bytes, but packet split
328                  * descriptors are 32 bytes...so we increment tail
329                  * twice as much.
330                  */
331                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
332         }
333 }
334
335 /**
336  * e1000_clean_rx_irq - Send received data up the network stack; legacy
337  * @adapter: board private structure
338  *
339  * the return value indicates whether actual cleaning was done, there
340  * is no guarantee that everything was cleaned
341  **/
342 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
343                                int *work_done, int work_to_do)
344 {
345         struct net_device *netdev = adapter->netdev;
346         struct pci_dev *pdev = adapter->pdev;
347         struct e1000_ring *rx_ring = adapter->rx_ring;
348         struct e1000_rx_desc *rx_desc, *next_rxd;
349         struct e1000_buffer *buffer_info, *next_buffer;
350         u32 length;
351         unsigned int i;
352         int cleaned_count = 0;
353         bool cleaned = 0;
354         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
355
356         i = rx_ring->next_to_clean;
357         rx_desc = E1000_RX_DESC(*rx_ring, i);
358         buffer_info = &rx_ring->buffer_info[i];
359
360         while (rx_desc->status & E1000_RXD_STAT_DD) {
361                 struct sk_buff *skb;
362                 u8 status;
363
364                 if (*work_done >= work_to_do)
365                         break;
366                 (*work_done)++;
367
368                 status = rx_desc->status;
369                 skb = buffer_info->skb;
370                 buffer_info->skb = NULL;
371
372                 prefetch(skb->data - NET_IP_ALIGN);
373
374                 i++;
375                 if (i == rx_ring->count)
376                         i = 0;
377                 next_rxd = E1000_RX_DESC(*rx_ring, i);
378                 prefetch(next_rxd);
379
380                 next_buffer = &rx_ring->buffer_info[i];
381
382                 cleaned = 1;
383                 cleaned_count++;
384                 pci_unmap_single(pdev,
385                                  buffer_info->dma,
386                                  adapter->rx_buffer_len,
387                                  PCI_DMA_FROMDEVICE);
388                 buffer_info->dma = 0;
389
390                 length = le16_to_cpu(rx_desc->length);
391
392                 /* !EOP means multiple descriptors were used to store a single
393                  * packet, also make sure the frame isn't just CRC only */
394                 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
395                         /* All receives must fit into a single buffer */
396                         ndev_dbg(netdev, "%s: Receive packet consumed "
397                                  "multiple buffers\n", netdev->name);
398                         /* recycle */
399                         buffer_info->skb = skb;
400                         goto next_desc;
401                 }
402
403                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
404                         /* recycle */
405                         buffer_info->skb = skb;
406                         goto next_desc;
407                 }
408
409                 total_rx_bytes += length;
410                 total_rx_packets++;
411
412                 /* code added for copybreak, this should improve
413                  * performance for small packets with large amounts
414                  * of reassembly being done in the stack */
415                 if (length < copybreak) {
416                         struct sk_buff *new_skb =
417                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
418                         if (new_skb) {
419                                 skb_reserve(new_skb, NET_IP_ALIGN);
420                                 memcpy(new_skb->data - NET_IP_ALIGN,
421                                        skb->data - NET_IP_ALIGN,
422                                        length + NET_IP_ALIGN);
423                                 /* save the skb in buffer_info as good */
424                                 buffer_info->skb = skb;
425                                 skb = new_skb;
426                         }
427                         /* else just continue with the old one */
428                 }
429                 /* end copybreak code */
430                 skb_put(skb, length);
431
432                 /* Receive Checksum Offload */
433                 e1000_rx_checksum(adapter,
434                                   (u32)(status) |
435                                   ((u32)(rx_desc->errors) << 24),
436                                   le16_to_cpu(rx_desc->csum), skb);
437
438                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
439
440 next_desc:
441                 rx_desc->status = 0;
442
443                 /* return some buffers to hardware, one at a time is too slow */
444                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
445                         adapter->alloc_rx_buf(adapter, cleaned_count);
446                         cleaned_count = 0;
447                 }
448
449                 /* use prefetched values */
450                 rx_desc = next_rxd;
451                 buffer_info = next_buffer;
452         }
453         rx_ring->next_to_clean = i;
454
455         cleaned_count = e1000_desc_unused(rx_ring);
456         if (cleaned_count)
457                 adapter->alloc_rx_buf(adapter, cleaned_count);
458
459         adapter->total_rx_packets += total_rx_packets;
460         adapter->total_rx_bytes += total_rx_bytes;
461         return cleaned;
462 }
463
464 static void e1000_put_txbuf(struct e1000_adapter *adapter,
465                              struct e1000_buffer *buffer_info)
466 {
467         if (buffer_info->dma) {
468                 pci_unmap_page(adapter->pdev, buffer_info->dma,
469                                buffer_info->length, PCI_DMA_TODEVICE);
470                 buffer_info->dma = 0;
471         }
472         if (buffer_info->skb) {
473                 dev_kfree_skb_any(buffer_info->skb);
474                 buffer_info->skb = NULL;
475         }
476 }
477
478 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
479 {
480         struct e1000_ring *tx_ring = adapter->tx_ring;
481         unsigned int i = tx_ring->next_to_clean;
482         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
483         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
484         struct net_device *netdev = adapter->netdev;
485
486         /* detected Tx unit hang */
487         ndev_err(netdev,
488                  "Detected Tx Unit Hang:\n"
489                  "  TDH                  <%x>\n"
490                  "  TDT                  <%x>\n"
491                  "  next_to_use          <%x>\n"
492                  "  next_to_clean        <%x>\n"
493                  "buffer_info[next_to_clean]:\n"
494                  "  time_stamp           <%lx>\n"
495                  "  next_to_watch        <%x>\n"
496                  "  jiffies              <%lx>\n"
497                  "  next_to_watch.status <%x>\n",
498                  readl(adapter->hw.hw_addr + tx_ring->head),
499                  readl(adapter->hw.hw_addr + tx_ring->tail),
500                  tx_ring->next_to_use,
501                  tx_ring->next_to_clean,
502                  tx_ring->buffer_info[eop].time_stamp,
503                  eop,
504                  jiffies,
505                  eop_desc->upper.fields.status);
506 }
507
508 /**
509  * e1000_clean_tx_irq - Reclaim resources after transmit completes
510  * @adapter: board private structure
511  *
512  * the return value indicates whether actual cleaning was done, there
513  * is no guarantee that everything was cleaned
514  **/
515 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
516 {
517         struct net_device *netdev = adapter->netdev;
518         struct e1000_hw *hw = &adapter->hw;
519         struct e1000_ring *tx_ring = adapter->tx_ring;
520         struct e1000_tx_desc *tx_desc, *eop_desc;
521         struct e1000_buffer *buffer_info;
522         unsigned int i, eop;
523         unsigned int count = 0;
524         bool cleaned = 0;
525         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
526
527         i = tx_ring->next_to_clean;
528         eop = tx_ring->buffer_info[i].next_to_watch;
529         eop_desc = E1000_TX_DESC(*tx_ring, eop);
530
531         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
532                 for (cleaned = 0; !cleaned; ) {
533                         tx_desc = E1000_TX_DESC(*tx_ring, i);
534                         buffer_info = &tx_ring->buffer_info[i];
535                         cleaned = (i == eop);
536
537                         if (cleaned) {
538                                 struct sk_buff *skb = buffer_info->skb;
539                                 unsigned int segs, bytecount;
540                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
541                                 /* multiply data chunks by size of headers */
542                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
543                                             skb->len;
544                                 total_tx_packets += segs;
545                                 total_tx_bytes += bytecount;
546                         }
547
548                         e1000_put_txbuf(adapter, buffer_info);
549                         tx_desc->upper.data = 0;
550
551                         i++;
552                         if (i == tx_ring->count)
553                                 i = 0;
554                 }
555
556                 eop = tx_ring->buffer_info[i].next_to_watch;
557                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
558 #define E1000_TX_WEIGHT 64
559                 /* weight of a sort for tx, to avoid endless transmit cleanup */
560                 if (count++ == E1000_TX_WEIGHT)
561                         break;
562         }
563
564         tx_ring->next_to_clean = i;
565
566 #define TX_WAKE_THRESHOLD 32
567         if (cleaned && netif_carrier_ok(netdev) &&
568                      e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
569                 /* Make sure that anybody stopping the queue after this
570                  * sees the new next_to_clean.
571                  */
572                 smp_mb();
573
574                 if (netif_queue_stopped(netdev) &&
575                     !(test_bit(__E1000_DOWN, &adapter->state))) {
576                         netif_wake_queue(netdev);
577                         ++adapter->restart_queue;
578                 }
579         }
580
581         if (adapter->detect_tx_hung) {
582                 /* Detect a transmit hang in hardware, this serializes the
583                  * check with the clearing of time_stamp and movement of i */
584                 adapter->detect_tx_hung = 0;
585                 if (tx_ring->buffer_info[eop].dma &&
586                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp
587                                + (adapter->tx_timeout_factor * HZ))
588                     && !(er32(STATUS) &
589                          E1000_STATUS_TXOFF)) {
590                         e1000_print_tx_hang(adapter);
591                         netif_stop_queue(netdev);
592                 }
593         }
594         adapter->total_tx_bytes += total_tx_bytes;
595         adapter->total_tx_packets += total_tx_packets;
596         return cleaned;
597 }
598
599 /**
600  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
601  * @adapter: board private structure
602  *
603  * the return value indicates whether actual cleaning was done, there
604  * is no guarantee that everything was cleaned
605  **/
606 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
607                                   int *work_done, int work_to_do)
608 {
609         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
610         struct net_device *netdev = adapter->netdev;
611         struct pci_dev *pdev = adapter->pdev;
612         struct e1000_ring *rx_ring = adapter->rx_ring;
613         struct e1000_buffer *buffer_info, *next_buffer;
614         struct e1000_ps_page *ps_page;
615         struct sk_buff *skb;
616         unsigned int i, j;
617         u32 length, staterr;
618         int cleaned_count = 0;
619         bool cleaned = 0;
620         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
621
622         i = rx_ring->next_to_clean;
623         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
624         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
625         buffer_info = &rx_ring->buffer_info[i];
626
627         while (staterr & E1000_RXD_STAT_DD) {
628                 if (*work_done >= work_to_do)
629                         break;
630                 (*work_done)++;
631                 skb = buffer_info->skb;
632
633                 /* in the packet split case this is header only */
634                 prefetch(skb->data - NET_IP_ALIGN);
635
636                 i++;
637                 if (i == rx_ring->count)
638                         i = 0;
639                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
640                 prefetch(next_rxd);
641
642                 next_buffer = &rx_ring->buffer_info[i];
643
644                 cleaned = 1;
645                 cleaned_count++;
646                 pci_unmap_single(pdev, buffer_info->dma,
647                                  adapter->rx_ps_bsize0,
648                                  PCI_DMA_FROMDEVICE);
649                 buffer_info->dma = 0;
650
651                 if (!(staterr & E1000_RXD_STAT_EOP)) {
652                         ndev_dbg(netdev, "%s: Packet Split buffers didn't pick "
653                                  "up the full packet\n", netdev->name);
654                         dev_kfree_skb_irq(skb);
655                         goto next_desc;
656                 }
657
658                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
659                         dev_kfree_skb_irq(skb);
660                         goto next_desc;
661                 }
662
663                 length = le16_to_cpu(rx_desc->wb.middle.length0);
664
665                 if (!length) {
666                         ndev_dbg(netdev, "%s: Last part of the packet spanning"
667                                  " multiple descriptors\n", netdev->name);
668                         dev_kfree_skb_irq(skb);
669                         goto next_desc;
670                 }
671
672                 /* Good Receive */
673                 skb_put(skb, length);
674
675                 {
676                 /* this looks ugly, but it seems compiler issues make it
677                    more efficient than reusing j */
678                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
679
680                 /* page alloc/put takes too long and effects small packet
681                  * throughput, so unsplit small packets and save the alloc/put*/
682                 if (l1 && (l1 <= copybreak) &&
683                     ((length + l1) <= adapter->rx_ps_bsize0)) {
684                         u8 *vaddr;
685
686                         ps_page = &buffer_info->ps_pages[0];
687
688                         /* there is no documentation about how to call
689                          * kmap_atomic, so we can't hold the mapping
690                          * very long */
691                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
692                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
693                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
694                         memcpy(skb_tail_pointer(skb), vaddr, l1);
695                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
696                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
697                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
698
699                         skb_put(skb, l1);
700                         goto copydone;
701                 } /* if */
702                 }
703
704                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
705                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
706                         if (!length)
707                                 break;
708
709                         ps_page = &buffer_info->ps_pages[j];
710                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
711                                        PCI_DMA_FROMDEVICE);
712                         ps_page->dma = 0;
713                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
714                         ps_page->page = NULL;
715                         skb->len += length;
716                         skb->data_len += length;
717                         skb->truesize += length;
718                 }
719
720 copydone:
721                 total_rx_bytes += skb->len;
722                 total_rx_packets++;
723
724                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
725                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
726
727                 if (rx_desc->wb.upper.header_status &
728                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
729                         adapter->rx_hdr_split++;
730
731                 e1000_receive_skb(adapter, netdev, skb,
732                                   staterr, rx_desc->wb.middle.vlan);
733
734 next_desc:
735                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
736                 buffer_info->skb = NULL;
737
738                 /* return some buffers to hardware, one at a time is too slow */
739                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
740                         adapter->alloc_rx_buf(adapter, cleaned_count);
741                         cleaned_count = 0;
742                 }
743
744                 /* use prefetched values */
745                 rx_desc = next_rxd;
746                 buffer_info = next_buffer;
747
748                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
749         }
750         rx_ring->next_to_clean = i;
751
752         cleaned_count = e1000_desc_unused(rx_ring);
753         if (cleaned_count)
754                 adapter->alloc_rx_buf(adapter, cleaned_count);
755
756         adapter->total_rx_packets += total_rx_packets;
757         adapter->total_rx_bytes += total_rx_bytes;
758         return cleaned;
759 }
760
761 /**
762  * e1000_clean_rx_ring - Free Rx Buffers per Queue
763  * @adapter: board private structure
764  **/
765 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
766 {
767         struct e1000_ring *rx_ring = adapter->rx_ring;
768         struct e1000_buffer *buffer_info;
769         struct e1000_ps_page *ps_page;
770         struct pci_dev *pdev = adapter->pdev;
771         unsigned int i, j;
772
773         /* Free all the Rx ring sk_buffs */
774         for (i = 0; i < rx_ring->count; i++) {
775                 buffer_info = &rx_ring->buffer_info[i];
776                 if (buffer_info->dma) {
777                         if (adapter->clean_rx == e1000_clean_rx_irq)
778                                 pci_unmap_single(pdev, buffer_info->dma,
779                                                  adapter->rx_buffer_len,
780                                                  PCI_DMA_FROMDEVICE);
781                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
782                                 pci_unmap_single(pdev, buffer_info->dma,
783                                                  adapter->rx_ps_bsize0,
784                                                  PCI_DMA_FROMDEVICE);
785                         buffer_info->dma = 0;
786                 }
787
788                 if (buffer_info->skb) {
789                         dev_kfree_skb(buffer_info->skb);
790                         buffer_info->skb = NULL;
791                 }
792
793                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
794                         ps_page = &buffer_info->ps_pages[j];
795                         if (!ps_page->page)
796                                 break;
797                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
798                                        PCI_DMA_FROMDEVICE);
799                         ps_page->dma = 0;
800                         put_page(ps_page->page);
801                         ps_page->page = NULL;
802                 }
803         }
804
805         /* there also may be some cached data from a chained receive */
806         if (rx_ring->rx_skb_top) {
807                 dev_kfree_skb(rx_ring->rx_skb_top);
808                 rx_ring->rx_skb_top = NULL;
809         }
810
811         /* Zero out the descriptor ring */
812         memset(rx_ring->desc, 0, rx_ring->size);
813
814         rx_ring->next_to_clean = 0;
815         rx_ring->next_to_use = 0;
816
817         writel(0, adapter->hw.hw_addr + rx_ring->head);
818         writel(0, adapter->hw.hw_addr + rx_ring->tail);
819 }
820
821 /**
822  * e1000_intr_msi - Interrupt Handler
823  * @irq: interrupt number
824  * @data: pointer to a network interface device structure
825  **/
826 static irqreturn_t e1000_intr_msi(int irq, void *data)
827 {
828         struct net_device *netdev = data;
829         struct e1000_adapter *adapter = netdev_priv(netdev);
830         struct e1000_hw *hw = &adapter->hw;
831         u32 icr = er32(ICR);
832
833         /* read ICR disables interrupts using IAM, so keep up with our
834          * enable/disable accounting */
835         atomic_inc(&adapter->irq_sem);
836
837         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
838                 hw->mac.get_link_status = 1;
839                 /* ICH8 workaround-- Call gig speed drop workaround on cable
840                  * disconnect (LSC) before accessing any PHY registers */
841                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
842                     (!(er32(STATUS) & E1000_STATUS_LU)))
843                         e1000e_gig_downshift_workaround_ich8lan(hw);
844
845                 /* 80003ES2LAN workaround-- For packet buffer work-around on
846                  * link down event; disable receives here in the ISR and reset
847                  * adapter in watchdog */
848                 if (netif_carrier_ok(netdev) &&
849                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
850                         /* disable receives */
851                         u32 rctl = er32(RCTL);
852                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
853                 }
854                 /* guard against interrupt when we're going down */
855                 if (!test_bit(__E1000_DOWN, &adapter->state))
856                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
857         }
858
859         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
860                 adapter->total_tx_bytes = 0;
861                 adapter->total_tx_packets = 0;
862                 adapter->total_rx_bytes = 0;
863                 adapter->total_rx_packets = 0;
864                 __netif_rx_schedule(netdev, &adapter->napi);
865         } else {
866                 atomic_dec(&adapter->irq_sem);
867         }
868
869         return IRQ_HANDLED;
870 }
871
872 /**
873  * e1000_intr - Interrupt Handler
874  * @irq: interrupt number
875  * @data: pointer to a network interface device structure
876  **/
877 static irqreturn_t e1000_intr(int irq, void *data)
878 {
879         struct net_device *netdev = data;
880         struct e1000_adapter *adapter = netdev_priv(netdev);
881         struct e1000_hw *hw = &adapter->hw;
882
883         u32 rctl, icr = er32(ICR);
884         if (!icr)
885                 return IRQ_NONE;  /* Not our interrupt */
886
887         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
888          * not set, then the adapter didn't send an interrupt */
889         if (!(icr & E1000_ICR_INT_ASSERTED))
890                 return IRQ_NONE;
891
892         /* Interrupt Auto-Mask...upon reading ICR,
893          * interrupts are masked.  No need for the
894          * IMC write, but it does mean we should
895          * account for it ASAP. */
896         atomic_inc(&adapter->irq_sem);
897
898         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
899                 hw->mac.get_link_status = 1;
900                 /* ICH8 workaround-- Call gig speed drop workaround on cable
901                  * disconnect (LSC) before accessing any PHY registers */
902                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
903                     (!(er32(STATUS) & E1000_STATUS_LU)))
904                         e1000e_gig_downshift_workaround_ich8lan(hw);
905
906                 /* 80003ES2LAN workaround--
907                  * For packet buffer work-around on link down event;
908                  * disable receives here in the ISR and
909                  * reset adapter in watchdog
910                  */
911                 if (netif_carrier_ok(netdev) &&
912                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
913                         /* disable receives */
914                         rctl = er32(RCTL);
915                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
916                 }
917                 /* guard against interrupt when we're going down */
918                 if (!test_bit(__E1000_DOWN, &adapter->state))
919                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
920         }
921
922         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
923                 adapter->total_tx_bytes = 0;
924                 adapter->total_tx_packets = 0;
925                 adapter->total_rx_bytes = 0;
926                 adapter->total_rx_packets = 0;
927                 __netif_rx_schedule(netdev, &adapter->napi);
928         } else {
929                 atomic_dec(&adapter->irq_sem);
930         }
931
932         return IRQ_HANDLED;
933 }
934
935 static int e1000_request_irq(struct e1000_adapter *adapter)
936 {
937         struct net_device *netdev = adapter->netdev;
938         void (*handler) = &e1000_intr;
939         int irq_flags = IRQF_SHARED;
940         int err;
941
942         err = pci_enable_msi(adapter->pdev);
943         if (err) {
944                 ndev_warn(netdev,
945                  "Unable to allocate MSI interrupt Error: %d\n", err);
946         } else {
947                 adapter->flags |= FLAG_MSI_ENABLED;
948                 handler = &e1000_intr_msi;
949                 irq_flags = 0;
950         }
951
952         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
953                           netdev);
954         if (err) {
955                 if (adapter->flags & FLAG_MSI_ENABLED)
956                         pci_disable_msi(adapter->pdev);
957                 ndev_err(netdev,
958                        "Unable to allocate interrupt Error: %d\n", err);
959         }
960
961         return err;
962 }
963
964 static void e1000_free_irq(struct e1000_adapter *adapter)
965 {
966         struct net_device *netdev = adapter->netdev;
967
968         free_irq(adapter->pdev->irq, netdev);
969         if (adapter->flags & FLAG_MSI_ENABLED) {
970                 pci_disable_msi(adapter->pdev);
971                 adapter->flags &= ~FLAG_MSI_ENABLED;
972         }
973 }
974
975 /**
976  * e1000_irq_disable - Mask off interrupt generation on the NIC
977  **/
978 static void e1000_irq_disable(struct e1000_adapter *adapter)
979 {
980         struct e1000_hw *hw = &adapter->hw;
981
982         atomic_inc(&adapter->irq_sem);
983         ew32(IMC, ~0);
984         e1e_flush();
985         synchronize_irq(adapter->pdev->irq);
986 }
987
988 /**
989  * e1000_irq_enable - Enable default interrupt generation settings
990  **/
991 static void e1000_irq_enable(struct e1000_adapter *adapter)
992 {
993         struct e1000_hw *hw = &adapter->hw;
994
995         if (atomic_dec_and_test(&adapter->irq_sem)) {
996                 ew32(IMS, IMS_ENABLE_MASK);
997                 e1e_flush();
998         }
999 }
1000
1001 /**
1002  * e1000_get_hw_control - get control of the h/w from f/w
1003  * @adapter: address of board private structure
1004  *
1005  * e1000_get_hw_control sets {CTRL_EXT|FWSM}:DRV_LOAD bit.
1006  * For ASF and Pass Through versions of f/w this means that
1007  * the driver is loaded. For AMT version (only with 82573)
1008  * of the f/w this means that the network i/f is open.
1009  **/
1010 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1011 {
1012         struct e1000_hw *hw = &adapter->hw;
1013         u32 ctrl_ext;
1014         u32 swsm;
1015
1016         /* Let firmware know the driver has taken over */
1017         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1018                 swsm = er32(SWSM);
1019                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1020         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1021                 ctrl_ext = er32(CTRL_EXT);
1022                 ew32(CTRL_EXT,
1023                                 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1024         }
1025 }
1026
1027 /**
1028  * e1000_release_hw_control - release control of the h/w to f/w
1029  * @adapter: address of board private structure
1030  *
1031  * e1000_release_hw_control resets {CTRL_EXT|FWSM}:DRV_LOAD bit.
1032  * For ASF and Pass Through versions of f/w this means that the
1033  * driver is no longer loaded. For AMT version (only with 82573) i
1034  * of the f/w this means that the network i/f is closed.
1035  *
1036  **/
1037 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1038 {
1039         struct e1000_hw *hw = &adapter->hw;
1040         u32 ctrl_ext;
1041         u32 swsm;
1042
1043         /* Let firmware taken over control of h/w */
1044         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1045                 swsm = er32(SWSM);
1046                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1047         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1048                 ctrl_ext = er32(CTRL_EXT);
1049                 ew32(CTRL_EXT,
1050                                 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1051         }
1052 }
1053
1054 static void e1000_release_manageability(struct e1000_adapter *adapter)
1055 {
1056         if (adapter->flags & FLAG_MNG_PT_ENABLED) {
1057                 struct e1000_hw *hw = &adapter->hw;
1058
1059                 u32 manc = er32(MANC);
1060
1061                 /* re-enable hardware interception of ARP */
1062                 manc |= E1000_MANC_ARP_EN;
1063                 manc &= ~E1000_MANC_EN_MNG2HOST;
1064
1065                 /* don't explicitly have to mess with MANC2H since
1066                  * MANC has an enable disable that gates MANC2H */
1067                 ew32(MANC, manc);
1068         }
1069 }
1070
1071 /**
1072  * @e1000_alloc_ring - allocate memory for a ring structure
1073  **/
1074 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1075                                 struct e1000_ring *ring)
1076 {
1077         struct pci_dev *pdev = adapter->pdev;
1078
1079         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1080                                         GFP_KERNEL);
1081         if (!ring->desc)
1082                 return -ENOMEM;
1083
1084         return 0;
1085 }
1086
1087 /**
1088  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1089  * @adapter: board private structure
1090  *
1091  * Return 0 on success, negative on failure
1092  **/
1093 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1094 {
1095         struct e1000_ring *tx_ring = adapter->tx_ring;
1096         int err = -ENOMEM, size;
1097
1098         size = sizeof(struct e1000_buffer) * tx_ring->count;
1099         tx_ring->buffer_info = vmalloc(size);
1100         if (!tx_ring->buffer_info)
1101                 goto err;
1102         memset(tx_ring->buffer_info, 0, size);
1103
1104         /* round up to nearest 4K */
1105         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1106         tx_ring->size = ALIGN(tx_ring->size, 4096);
1107
1108         err = e1000_alloc_ring_dma(adapter, tx_ring);
1109         if (err)
1110                 goto err;
1111
1112         tx_ring->next_to_use = 0;
1113         tx_ring->next_to_clean = 0;
1114         spin_lock_init(&adapter->tx_queue_lock);
1115
1116         return 0;
1117 err:
1118         vfree(tx_ring->buffer_info);
1119         ndev_err(adapter->netdev,
1120         "Unable to allocate memory for the transmit descriptor ring\n");
1121         return err;
1122 }
1123
1124 /**
1125  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1126  * @adapter: board private structure
1127  *
1128  * Returns 0 on success, negative on failure
1129  **/
1130 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1131 {
1132         struct e1000_ring *rx_ring = adapter->rx_ring;
1133         struct e1000_buffer *buffer_info;
1134         int i, size, desc_len, err = -ENOMEM;
1135
1136         size = sizeof(struct e1000_buffer) * rx_ring->count;
1137         rx_ring->buffer_info = vmalloc(size);
1138         if (!rx_ring->buffer_info)
1139                 goto err;
1140         memset(rx_ring->buffer_info, 0, size);
1141
1142         for (i = 0; i < rx_ring->count; i++) {
1143                 buffer_info = &rx_ring->buffer_info[i];
1144                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1145                                                 sizeof(struct e1000_ps_page),
1146                                                 GFP_KERNEL);
1147                 if (!buffer_info->ps_pages)
1148                         goto err_pages;
1149         }
1150
1151         desc_len = sizeof(union e1000_rx_desc_packet_split);
1152
1153         /* Round up to nearest 4K */
1154         rx_ring->size = rx_ring->count * desc_len;
1155         rx_ring->size = ALIGN(rx_ring->size, 4096);
1156
1157         err = e1000_alloc_ring_dma(adapter, rx_ring);
1158         if (err)
1159                 goto err_pages;
1160
1161         rx_ring->next_to_clean = 0;
1162         rx_ring->next_to_use = 0;
1163         rx_ring->rx_skb_top = NULL;
1164
1165         return 0;
1166
1167 err_pages:
1168         for (i = 0; i < rx_ring->count; i++) {
1169                 buffer_info = &rx_ring->buffer_info[i];
1170                 kfree(buffer_info->ps_pages);
1171         }
1172 err:
1173         vfree(rx_ring->buffer_info);
1174         ndev_err(adapter->netdev,
1175         "Unable to allocate memory for the transmit descriptor ring\n");
1176         return err;
1177 }
1178
1179 /**
1180  * e1000_clean_tx_ring - Free Tx Buffers
1181  * @adapter: board private structure
1182  **/
1183 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1184 {
1185         struct e1000_ring *tx_ring = adapter->tx_ring;
1186         struct e1000_buffer *buffer_info;
1187         unsigned long size;
1188         unsigned int i;
1189
1190         for (i = 0; i < tx_ring->count; i++) {
1191                 buffer_info = &tx_ring->buffer_info[i];
1192                 e1000_put_txbuf(adapter, buffer_info);
1193         }
1194
1195         size = sizeof(struct e1000_buffer) * tx_ring->count;
1196         memset(tx_ring->buffer_info, 0, size);
1197
1198         memset(tx_ring->desc, 0, tx_ring->size);
1199
1200         tx_ring->next_to_use = 0;
1201         tx_ring->next_to_clean = 0;
1202
1203         writel(0, adapter->hw.hw_addr + tx_ring->head);
1204         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1205 }
1206
1207 /**
1208  * e1000e_free_tx_resources - Free Tx Resources per Queue
1209  * @adapter: board private structure
1210  *
1211  * Free all transmit software resources
1212  **/
1213 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1214 {
1215         struct pci_dev *pdev = adapter->pdev;
1216         struct e1000_ring *tx_ring = adapter->tx_ring;
1217
1218         e1000_clean_tx_ring(adapter);
1219
1220         vfree(tx_ring->buffer_info);
1221         tx_ring->buffer_info = NULL;
1222
1223         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1224                           tx_ring->dma);
1225         tx_ring->desc = NULL;
1226 }
1227
1228 /**
1229  * e1000e_free_rx_resources - Free Rx Resources
1230  * @adapter: board private structure
1231  *
1232  * Free all receive software resources
1233  **/
1234
1235 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1236 {
1237         struct pci_dev *pdev = adapter->pdev;
1238         struct e1000_ring *rx_ring = adapter->rx_ring;
1239         int i;
1240
1241         e1000_clean_rx_ring(adapter);
1242
1243         for (i = 0; i < rx_ring->count; i++) {
1244                 kfree(rx_ring->buffer_info[i].ps_pages);
1245         }
1246
1247         vfree(rx_ring->buffer_info);
1248         rx_ring->buffer_info = NULL;
1249
1250         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1251                           rx_ring->dma);
1252         rx_ring->desc = NULL;
1253 }
1254
1255 /**
1256  * e1000_update_itr - update the dynamic ITR value based on statistics
1257  *      Stores a new ITR value based on packets and byte
1258  *      counts during the last interrupt.  The advantage of per interrupt
1259  *      computation is faster updates and more accurate ITR for the current
1260  *      traffic pattern.  Constants in this function were computed
1261  *      based on theoretical maximum wire speed and thresholds were set based
1262  *      on testing data as well as attempting to minimize response time
1263  *      while increasing bulk throughput.
1264  *      this functionality is controlled by the InterruptThrottleRate module
1265  *      parameter (see e1000_param.c)
1266  * @adapter: pointer to adapter
1267  * @itr_setting: current adapter->itr
1268  * @packets: the number of packets during this measurement interval
1269  * @bytes: the number of bytes during this measurement interval
1270  **/
1271 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1272                                      u16 itr_setting, int packets,
1273                                      int bytes)
1274 {
1275         unsigned int retval = itr_setting;
1276
1277         if (packets == 0)
1278                 goto update_itr_done;
1279
1280         switch (itr_setting) {
1281         case lowest_latency:
1282                 /* handle TSO and jumbo frames */
1283                 if (bytes/packets > 8000)
1284                         retval = bulk_latency;
1285                 else if ((packets < 5) && (bytes > 512)) {
1286                         retval = low_latency;
1287                 }
1288                 break;
1289         case low_latency:  /* 50 usec aka 20000 ints/s */
1290                 if (bytes > 10000) {
1291                         /* this if handles the TSO accounting */
1292                         if (bytes/packets > 8000) {
1293                                 retval = bulk_latency;
1294                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1295                                 retval = bulk_latency;
1296                         } else if ((packets > 35)) {
1297                                 retval = lowest_latency;
1298                         }
1299                 } else if (bytes/packets > 2000) {
1300                         retval = bulk_latency;
1301                 } else if (packets <= 2 && bytes < 512) {
1302                         retval = lowest_latency;
1303                 }
1304                 break;
1305         case bulk_latency: /* 250 usec aka 4000 ints/s */
1306                 if (bytes > 25000) {
1307                         if (packets > 35) {
1308                                 retval = low_latency;
1309                         }
1310                 } else if (bytes < 6000) {
1311                         retval = low_latency;
1312                 }
1313                 break;
1314         }
1315
1316 update_itr_done:
1317         return retval;
1318 }
1319
1320 static void e1000_set_itr(struct e1000_adapter *adapter)
1321 {
1322         struct e1000_hw *hw = &adapter->hw;
1323         u16 current_itr;
1324         u32 new_itr = adapter->itr;
1325
1326         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1327         if (adapter->link_speed != SPEED_1000) {
1328                 current_itr = 0;
1329                 new_itr = 4000;
1330                 goto set_itr_now;
1331         }
1332
1333         adapter->tx_itr = e1000_update_itr(adapter,
1334                                     adapter->tx_itr,
1335                                     adapter->total_tx_packets,
1336                                     adapter->total_tx_bytes);
1337         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1338         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1339                 adapter->tx_itr = low_latency;
1340
1341         adapter->rx_itr = e1000_update_itr(adapter,
1342                                     adapter->rx_itr,
1343                                     adapter->total_rx_packets,
1344                                     adapter->total_rx_bytes);
1345         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1346         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1347                 adapter->rx_itr = low_latency;
1348
1349         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1350
1351         switch (current_itr) {
1352         /* counts and packets in update_itr are dependent on these numbers */
1353         case lowest_latency:
1354                 new_itr = 70000;
1355                 break;
1356         case low_latency:
1357                 new_itr = 20000; /* aka hwitr = ~200 */
1358                 break;
1359         case bulk_latency:
1360                 new_itr = 4000;
1361                 break;
1362         default:
1363                 break;
1364         }
1365
1366 set_itr_now:
1367         if (new_itr != adapter->itr) {
1368                 /* this attempts to bias the interrupt rate towards Bulk
1369                  * by adding intermediate steps when interrupt rate is
1370                  * increasing */
1371                 new_itr = new_itr > adapter->itr ?
1372                              min(adapter->itr + (new_itr >> 2), new_itr) :
1373                              new_itr;
1374                 adapter->itr = new_itr;
1375                 ew32(ITR, 1000000000 / (new_itr * 256));
1376         }
1377 }
1378
1379 /**
1380  * e1000_clean - NAPI Rx polling callback
1381  * @adapter: board private structure
1382  **/
1383 static int e1000_clean(struct napi_struct *napi, int budget)
1384 {
1385         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1386         struct net_device *poll_dev = adapter->netdev;
1387         int work_done = 0;
1388
1389         /* Must NOT use netdev_priv macro here. */
1390         adapter = poll_dev->priv;
1391
1392         /* e1000_clean is called per-cpu.  This lock protects
1393          * tx_ring from being cleaned by multiple cpus
1394          * simultaneously.  A failure obtaining the lock means
1395          * tx_ring is currently being cleaned anyway. */
1396         if (spin_trylock(&adapter->tx_queue_lock)) {
1397                 e1000_clean_tx_irq(adapter);
1398                 spin_unlock(&adapter->tx_queue_lock);
1399         }
1400
1401         adapter->clean_rx(adapter, &work_done, budget);
1402
1403         /* If budget not fully consumed, exit the polling mode */
1404         if (work_done < budget) {
1405                 if (adapter->itr_setting & 3)
1406                         e1000_set_itr(adapter);
1407                 netif_rx_complete(poll_dev, napi);
1408                 e1000_irq_enable(adapter);
1409         }
1410
1411         return work_done;
1412 }
1413
1414 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1415 {
1416         struct e1000_adapter *adapter = netdev_priv(netdev);
1417         struct e1000_hw *hw = &adapter->hw;
1418         u32 vfta, index;
1419
1420         /* don't update vlan cookie if already programmed */
1421         if ((adapter->hw.mng_cookie.status &
1422              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1423             (vid == adapter->mng_vlan_id))
1424                 return;
1425         /* add VID to filter table */
1426         index = (vid >> 5) & 0x7F;
1427         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1428         vfta |= (1 << (vid & 0x1F));
1429         e1000e_write_vfta(hw, index, vfta);
1430 }
1431
1432 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1433 {
1434         struct e1000_adapter *adapter = netdev_priv(netdev);
1435         struct e1000_hw *hw = &adapter->hw;
1436         u32 vfta, index;
1437
1438         e1000_irq_disable(adapter);
1439         vlan_group_set_device(adapter->vlgrp, vid, NULL);
1440         e1000_irq_enable(adapter);
1441
1442         if ((adapter->hw.mng_cookie.status &
1443              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1444             (vid == adapter->mng_vlan_id)) {
1445                 /* release control to f/w */
1446                 e1000_release_hw_control(adapter);
1447                 return;
1448         }
1449
1450         /* remove VID from filter table */
1451         index = (vid >> 5) & 0x7F;
1452         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1453         vfta &= ~(1 << (vid & 0x1F));
1454         e1000e_write_vfta(hw, index, vfta);
1455 }
1456
1457 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
1458 {
1459         struct net_device *netdev = adapter->netdev;
1460         u16 vid = adapter->hw.mng_cookie.vlan_id;
1461         u16 old_vid = adapter->mng_vlan_id;
1462
1463         if (!adapter->vlgrp)
1464                 return;
1465
1466         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
1467                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1468                 if (adapter->hw.mng_cookie.status &
1469                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1470                         e1000_vlan_rx_add_vid(netdev, vid);
1471                         adapter->mng_vlan_id = vid;
1472                 }
1473
1474                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
1475                                 (vid != old_vid) &&
1476                     !vlan_group_get_device(adapter->vlgrp, old_vid))
1477                         e1000_vlan_rx_kill_vid(netdev, old_vid);
1478         } else {
1479                 adapter->mng_vlan_id = vid;
1480         }
1481 }
1482
1483
1484 static void e1000_vlan_rx_register(struct net_device *netdev,
1485                                    struct vlan_group *grp)
1486 {
1487         struct e1000_adapter *adapter = netdev_priv(netdev);
1488         struct e1000_hw *hw = &adapter->hw;
1489         u32 ctrl, rctl;
1490
1491         e1000_irq_disable(adapter);
1492         adapter->vlgrp = grp;
1493
1494         if (grp) {
1495                 /* enable VLAN tag insert/strip */
1496                 ctrl = er32(CTRL);
1497                 ctrl |= E1000_CTRL_VME;
1498                 ew32(CTRL, ctrl);
1499
1500                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1501                         /* enable VLAN receive filtering */
1502                         rctl = er32(RCTL);
1503                         rctl |= E1000_RCTL_VFE;
1504                         rctl &= ~E1000_RCTL_CFIEN;
1505                         ew32(RCTL, rctl);
1506                         e1000_update_mng_vlan(adapter);
1507                 }
1508         } else {
1509                 /* disable VLAN tag insert/strip */
1510                 ctrl = er32(CTRL);
1511                 ctrl &= ~E1000_CTRL_VME;
1512                 ew32(CTRL, ctrl);
1513
1514                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1515                         /* disable VLAN filtering */
1516                         rctl = er32(RCTL);
1517                         rctl &= ~E1000_RCTL_VFE;
1518                         ew32(RCTL, rctl);
1519                         if (adapter->mng_vlan_id !=
1520                             (u16)E1000_MNG_VLAN_NONE) {
1521                                 e1000_vlan_rx_kill_vid(netdev,
1522                                                        adapter->mng_vlan_id);
1523                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1524                         }
1525                 }
1526         }
1527
1528         e1000_irq_enable(adapter);
1529 }
1530
1531 static void e1000_restore_vlan(struct e1000_adapter *adapter)
1532 {
1533         u16 vid;
1534
1535         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
1536
1537         if (!adapter->vlgrp)
1538                 return;
1539
1540         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1541                 if (!vlan_group_get_device(adapter->vlgrp, vid))
1542                         continue;
1543                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
1544         }
1545 }
1546
1547 static void e1000_init_manageability(struct e1000_adapter *adapter)
1548 {
1549         struct e1000_hw *hw = &adapter->hw;
1550         u32 manc, manc2h;
1551
1552         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
1553                 return;
1554
1555         manc = er32(MANC);
1556
1557         /* disable hardware interception of ARP */
1558         manc &= ~(E1000_MANC_ARP_EN);
1559
1560         /* enable receiving management packets to the host. this will probably
1561          * generate destination unreachable messages from the host OS, but
1562          * the packets will be handled on SMBUS */
1563         manc |= E1000_MANC_EN_MNG2HOST;
1564         manc2h = er32(MANC2H);
1565 #define E1000_MNG2HOST_PORT_623 (1 << 5)
1566 #define E1000_MNG2HOST_PORT_664 (1 << 6)
1567         manc2h |= E1000_MNG2HOST_PORT_623;
1568         manc2h |= E1000_MNG2HOST_PORT_664;
1569         ew32(MANC2H, manc2h);
1570         ew32(MANC, manc);
1571 }
1572
1573 /**
1574  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1575  * @adapter: board private structure
1576  *
1577  * Configure the Tx unit of the MAC after a reset.
1578  **/
1579 static void e1000_configure_tx(struct e1000_adapter *adapter)
1580 {
1581         struct e1000_hw *hw = &adapter->hw;
1582         struct e1000_ring *tx_ring = adapter->tx_ring;
1583         u64 tdba;
1584         u32 tdlen, tctl, tipg, tarc;
1585         u32 ipgr1, ipgr2;
1586
1587         /* Setup the HW Tx Head and Tail descriptor pointers */
1588         tdba = tx_ring->dma;
1589         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
1590         ew32(TDBAL, (tdba & DMA_32BIT_MASK));
1591         ew32(TDBAH, (tdba >> 32));
1592         ew32(TDLEN, tdlen);
1593         ew32(TDH, 0);
1594         ew32(TDT, 0);
1595         tx_ring->head = E1000_TDH;
1596         tx_ring->tail = E1000_TDT;
1597
1598         /* Set the default values for the Tx Inter Packet Gap timer */
1599         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
1600         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
1601         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
1602
1603         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
1604                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
1605
1606         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1607         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1608         ew32(TIPG, tipg);
1609
1610         /* Set the Tx Interrupt Delay register */
1611         ew32(TIDV, adapter->tx_int_delay);
1612         /* tx irq moderation */
1613         ew32(TADV, adapter->tx_abs_int_delay);
1614
1615         /* Program the Transmit Control Register */
1616         tctl = er32(TCTL);
1617         tctl &= ~E1000_TCTL_CT;
1618         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1619                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1620
1621         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
1622                 tarc = er32(TARC0);
1623                 /* set the speed mode bit, we'll clear it if we're not at
1624                  * gigabit link later */
1625 #define SPEED_MODE_BIT (1 << 21)
1626                 tarc |= SPEED_MODE_BIT;
1627                 ew32(TARC0, tarc);
1628         }
1629
1630         /* errata: program both queues to unweighted RR */
1631         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
1632                 tarc = er32(TARC0);
1633                 tarc |= 1;
1634                 ew32(TARC0, tarc);
1635                 tarc = er32(TARC1);
1636                 tarc |= 1;
1637                 ew32(TARC1, tarc);
1638         }
1639
1640         e1000e_config_collision_dist(hw);
1641
1642         /* Setup Transmit Descriptor Settings for eop descriptor */
1643         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1644
1645         /* only set IDE if we are delaying interrupts using the timers */
1646         if (adapter->tx_int_delay)
1647                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1648
1649         /* enable Report Status bit */
1650         adapter->txd_cmd |= E1000_TXD_CMD_RS;
1651
1652         ew32(TCTL, tctl);
1653
1654         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
1655 }
1656
1657 /**
1658  * e1000_setup_rctl - configure the receive control registers
1659  * @adapter: Board private structure
1660  **/
1661 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1662                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1663 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1664 {
1665         struct e1000_hw *hw = &adapter->hw;
1666         u32 rctl, rfctl;
1667         u32 psrctl = 0;
1668         u32 pages = 0;
1669
1670         /* Program MC offset vector base */
1671         rctl = er32(RCTL);
1672         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1673         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1674                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1675                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1676
1677         /* Do not Store bad packets */
1678         rctl &= ~E1000_RCTL_SBP;
1679
1680         /* Enable Long Packet receive */
1681         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1682                 rctl &= ~E1000_RCTL_LPE;
1683         else
1684                 rctl |= E1000_RCTL_LPE;
1685
1686         /* Setup buffer sizes */
1687         rctl &= ~E1000_RCTL_SZ_4096;
1688         rctl |= E1000_RCTL_BSEX;
1689         switch (adapter->rx_buffer_len) {
1690         case 256:
1691                 rctl |= E1000_RCTL_SZ_256;
1692                 rctl &= ~E1000_RCTL_BSEX;
1693                 break;
1694         case 512:
1695                 rctl |= E1000_RCTL_SZ_512;
1696                 rctl &= ~E1000_RCTL_BSEX;
1697                 break;
1698         case 1024:
1699                 rctl |= E1000_RCTL_SZ_1024;
1700                 rctl &= ~E1000_RCTL_BSEX;
1701                 break;
1702         case 2048:
1703         default:
1704                 rctl |= E1000_RCTL_SZ_2048;
1705                 rctl &= ~E1000_RCTL_BSEX;
1706                 break;
1707         case 4096:
1708                 rctl |= E1000_RCTL_SZ_4096;
1709                 break;
1710         case 8192:
1711                 rctl |= E1000_RCTL_SZ_8192;
1712                 break;
1713         case 16384:
1714                 rctl |= E1000_RCTL_SZ_16384;
1715                 break;
1716         }
1717
1718         /*
1719          * 82571 and greater support packet-split where the protocol
1720          * header is placed in skb->data and the packet data is
1721          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
1722          * In the case of a non-split, skb->data is linearly filled,
1723          * followed by the page buffers.  Therefore, skb->data is
1724          * sized to hold the largest protocol header.
1725          *
1726          * allocations using alloc_page take too long for regular MTU
1727          * so only enable packet split for jumbo frames
1728          *
1729          * Using pages when the page size is greater than 16k wastes
1730          * a lot of memory, since we allocate 3 pages at all times
1731          * per packet.
1732          */
1733         adapter->rx_ps_pages = 0;
1734         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
1735         if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
1736                 adapter->rx_ps_pages = pages;
1737
1738         if (adapter->rx_ps_pages) {
1739                 /* Configure extra packet-split registers */
1740                 rfctl = er32(RFCTL);
1741                 rfctl |= E1000_RFCTL_EXTEN;
1742                 /* disable packet split support for IPv6 extension headers,
1743                  * because some malformed IPv6 headers can hang the RX */
1744                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
1745                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
1746
1747                 ew32(RFCTL, rfctl);
1748
1749                 /* Enable Packet split descriptors */
1750                 rctl |= E1000_RCTL_DTYP_PS;
1751                 
1752                 /* Enable hardware CRC frame stripping */
1753                 rctl |= E1000_RCTL_SECRC;
1754
1755                 psrctl |= adapter->rx_ps_bsize0 >>
1756                         E1000_PSRCTL_BSIZE0_SHIFT;
1757
1758                 switch (adapter->rx_ps_pages) {
1759                 case 3:
1760                         psrctl |= PAGE_SIZE <<
1761                                 E1000_PSRCTL_BSIZE3_SHIFT;
1762                 case 2:
1763                         psrctl |= PAGE_SIZE <<
1764                                 E1000_PSRCTL_BSIZE2_SHIFT;
1765                 case 1:
1766                         psrctl |= PAGE_SIZE >>
1767                                 E1000_PSRCTL_BSIZE1_SHIFT;
1768                         break;
1769                 }
1770
1771                 ew32(PSRCTL, psrctl);
1772         }
1773
1774         ew32(RCTL, rctl);
1775 }
1776
1777 /**
1778  * e1000_configure_rx - Configure Receive Unit after Reset
1779  * @adapter: board private structure
1780  *
1781  * Configure the Rx unit of the MAC after a reset.
1782  **/
1783 static void e1000_configure_rx(struct e1000_adapter *adapter)
1784 {
1785         struct e1000_hw *hw = &adapter->hw;
1786         struct e1000_ring *rx_ring = adapter->rx_ring;
1787         u64 rdba;
1788         u32 rdlen, rctl, rxcsum, ctrl_ext;
1789
1790         if (adapter->rx_ps_pages) {
1791                 /* this is a 32 byte descriptor */
1792                 rdlen = rx_ring->count *
1793                         sizeof(union e1000_rx_desc_packet_split);
1794                 adapter->clean_rx = e1000_clean_rx_irq_ps;
1795                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
1796         } else {
1797                 rdlen = rx_ring->count *
1798                         sizeof(struct e1000_rx_desc);
1799                 adapter->clean_rx = e1000_clean_rx_irq;
1800                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1801         }
1802
1803         /* disable receives while setting up the descriptors */
1804         rctl = er32(RCTL);
1805         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1806         e1e_flush();
1807         msleep(10);
1808
1809         /* set the Receive Delay Timer Register */
1810         ew32(RDTR, adapter->rx_int_delay);
1811
1812         /* irq moderation */
1813         ew32(RADV, adapter->rx_abs_int_delay);
1814         if (adapter->itr_setting != 0)
1815                 ew32(ITR,
1816                         1000000000 / (adapter->itr * 256));
1817
1818         ctrl_ext = er32(CTRL_EXT);
1819         /* Reset delay timers after every interrupt */
1820         ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
1821         /* Auto-Mask interrupts upon ICR access */
1822         ctrl_ext |= E1000_CTRL_EXT_IAME;
1823         ew32(IAM, 0xffffffff);
1824         ew32(CTRL_EXT, ctrl_ext);
1825         e1e_flush();
1826
1827         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1828          * the Base and Length of the Rx Descriptor Ring */
1829         rdba = rx_ring->dma;
1830         ew32(RDBAL, (rdba & DMA_32BIT_MASK));
1831         ew32(RDBAH, (rdba >> 32));
1832         ew32(RDLEN, rdlen);
1833         ew32(RDH, 0);
1834         ew32(RDT, 0);
1835         rx_ring->head = E1000_RDH;
1836         rx_ring->tail = E1000_RDT;
1837
1838         /* Enable Receive Checksum Offload for TCP and UDP */
1839         rxcsum = er32(RXCSUM);
1840         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
1841                 rxcsum |= E1000_RXCSUM_TUOFL;
1842
1843                 /* IPv4 payload checksum for UDP fragments must be
1844                  * used in conjunction with packet-split. */
1845                 if (adapter->rx_ps_pages)
1846                         rxcsum |= E1000_RXCSUM_IPPCSE;
1847         } else {
1848                 rxcsum &= ~E1000_RXCSUM_TUOFL;
1849                 /* no need to clear IPPCSE as it defaults to 0 */
1850         }
1851         ew32(RXCSUM, rxcsum);
1852
1853         /* Enable early receives on supported devices, only takes effect when
1854          * packet size is equal or larger than the specified value (in 8 byte
1855          * units), e.g. using jumbo frames when setting to E1000_ERT_2048 */
1856         if ((adapter->flags & FLAG_HAS_ERT) &&
1857             (adapter->netdev->mtu > ETH_DATA_LEN))
1858                 ew32(ERT, E1000_ERT_2048);
1859
1860         /* Enable Receives */
1861         ew32(RCTL, rctl);
1862 }
1863
1864 /**
1865  *  e1000_mc_addr_list_update - Update Multicast addresses
1866  *  @hw: pointer to the HW structure
1867  *  @mc_addr_list: array of multicast addresses to program
1868  *  @mc_addr_count: number of multicast addresses to program
1869  *  @rar_used_count: the first RAR register free to program
1870  *  @rar_count: total number of supported Receive Address Registers
1871  *
1872  *  Updates the Receive Address Registers and Multicast Table Array.
1873  *  The caller must have a packed mc_addr_list of multicast addresses.
1874  *  The parameter rar_count will usually be hw->mac.rar_entry_count
1875  *  unless there are workarounds that change this.  Currently no func pointer
1876  *  exists and all implementations are handled in the generic version of this
1877  *  function.
1878  **/
1879 static void e1000_mc_addr_list_update(struct e1000_hw *hw, u8 *mc_addr_list,
1880                                u32 mc_addr_count, u32 rar_used_count,
1881                                u32 rar_count)
1882 {
1883         hw->mac.ops.mc_addr_list_update(hw, mc_addr_list, mc_addr_count,
1884                                         rar_used_count, rar_count);
1885 }
1886
1887 /**
1888  * e1000_set_multi - Multicast and Promiscuous mode set
1889  * @netdev: network interface device structure
1890  *
1891  * The set_multi entry point is called whenever the multicast address
1892  * list or the network interface flags are updated.  This routine is
1893  * responsible for configuring the hardware for proper multicast,
1894  * promiscuous mode, and all-multi behavior.
1895  **/
1896 static void e1000_set_multi(struct net_device *netdev)
1897 {
1898         struct e1000_adapter *adapter = netdev_priv(netdev);
1899         struct e1000_hw *hw = &adapter->hw;
1900         struct e1000_mac_info *mac = &hw->mac;
1901         struct dev_mc_list *mc_ptr;
1902         u8  *mta_list;
1903         u32 rctl;
1904         int i;
1905
1906         /* Check for Promiscuous and All Multicast modes */
1907
1908         rctl = er32(RCTL);
1909
1910         if (netdev->flags & IFF_PROMISC) {
1911                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
1912         } else if (netdev->flags & IFF_ALLMULTI) {
1913                 rctl |= E1000_RCTL_MPE;
1914                 rctl &= ~E1000_RCTL_UPE;
1915         } else {
1916                 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
1917         }
1918
1919         ew32(RCTL, rctl);
1920
1921         if (netdev->mc_count) {
1922                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
1923                 if (!mta_list)
1924                         return;
1925
1926                 /* prepare a packed array of only addresses. */
1927                 mc_ptr = netdev->mc_list;
1928
1929                 for (i = 0; i < netdev->mc_count; i++) {
1930                         if (!mc_ptr)
1931                                 break;
1932                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
1933                                ETH_ALEN);
1934                         mc_ptr = mc_ptr->next;
1935                 }
1936
1937                 e1000_mc_addr_list_update(hw, mta_list, i, 1,
1938                                           mac->rar_entry_count);
1939                 kfree(mta_list);
1940         } else {
1941                 /*
1942                  * if we're called from probe, we might not have
1943                  * anything to do here, so clear out the list
1944                  */
1945                 e1000_mc_addr_list_update(hw, NULL, 0, 1,
1946                                           mac->rar_entry_count);
1947         }
1948 }
1949
1950 /**
1951  * e1000_configure - configure the hardware for RX and TX
1952  * @adapter: private board structure
1953  **/
1954 static void e1000_configure(struct e1000_adapter *adapter)
1955 {
1956         e1000_set_multi(adapter->netdev);
1957
1958         e1000_restore_vlan(adapter);
1959         e1000_init_manageability(adapter);
1960
1961         e1000_configure_tx(adapter);
1962         e1000_setup_rctl(adapter);
1963         e1000_configure_rx(adapter);
1964         adapter->alloc_rx_buf(adapter,
1965                               e1000_desc_unused(adapter->rx_ring));
1966 }
1967
1968 /**
1969  * e1000e_power_up_phy - restore link in case the phy was powered down
1970  * @adapter: address of board private structure
1971  *
1972  * The phy may be powered down to save power and turn off link when the
1973  * driver is unloaded and wake on lan is not enabled (among others)
1974  * *** this routine MUST be followed by a call to e1000e_reset ***
1975  **/
1976 void e1000e_power_up_phy(struct e1000_adapter *adapter)
1977 {
1978         u16 mii_reg = 0;
1979
1980         /* Just clear the power down bit to wake the phy back up */
1981         if (adapter->hw.media_type == e1000_media_type_copper) {
1982                 /* according to the manual, the phy will retain its
1983                  * settings across a power-down/up cycle */
1984                 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
1985                 mii_reg &= ~MII_CR_POWER_DOWN;
1986                 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
1987         }
1988
1989         adapter->hw.mac.ops.setup_link(&adapter->hw);
1990 }
1991
1992 /**
1993  * e1000_power_down_phy - Power down the PHY
1994  *
1995  * Power down the PHY so no link is implied when interface is down
1996  * The PHY cannot be powered down is management or WoL is active
1997  */
1998 static void e1000_power_down_phy(struct e1000_adapter *adapter)
1999 {
2000         struct e1000_hw *hw = &adapter->hw;
2001         u16 mii_reg;
2002
2003         /* WoL is enabled */
2004         if (!adapter->wol)
2005                 return;
2006
2007         /* non-copper PHY? */
2008         if (adapter->hw.media_type != e1000_media_type_copper)
2009                 return;
2010
2011         /* reset is blocked because of a SoL/IDER session */
2012         if (e1000e_check_mng_mode(hw) ||
2013             e1000_check_reset_block(hw))
2014                 return;
2015
2016         /* managebility (AMT) is enabled */
2017         if (er32(MANC) & E1000_MANC_SMBUS_EN)
2018                 return;
2019
2020         /* power down the PHY */
2021         e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2022         mii_reg |= MII_CR_POWER_DOWN;
2023         e1e_wphy(hw, PHY_CONTROL, mii_reg);
2024         mdelay(1);
2025 }
2026
2027 /**
2028  * e1000e_reset - bring the hardware into a known good state
2029  *
2030  * This function boots the hardware and enables some settings that
2031  * require a configuration cycle of the hardware - those cannot be
2032  * set/changed during runtime. After reset the device needs to be
2033  * properly configured for rx, tx etc.
2034  */
2035 void e1000e_reset(struct e1000_adapter *adapter)
2036 {
2037         struct e1000_mac_info *mac = &adapter->hw.mac;
2038         struct e1000_hw *hw = &adapter->hw;
2039         u32 tx_space, min_tx_space, min_rx_space;
2040         u32 pba;
2041         u16 hwm;
2042
2043         ew32(PBA, adapter->pba);
2044
2045         if (mac->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN ) {
2046                 /* To maintain wire speed transmits, the Tx FIFO should be
2047                  * large enough to accommodate two full transmit packets,
2048                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2049                  * the Rx FIFO should be large enough to accommodate at least
2050                  * one full receive packet and is similarly rounded up and
2051                  * expressed in KB. */
2052                 pba = er32(PBA);
2053                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2054                 tx_space = pba >> 16;
2055                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2056                 pba &= 0xffff;
2057                 /* the tx fifo also stores 16 bytes of information about the tx
2058                  * but don't include ethernet FCS because hardware appends it */
2059                 min_tx_space = (mac->max_frame_size +
2060                                 sizeof(struct e1000_tx_desc) -
2061                                 ETH_FCS_LEN) * 2;
2062                 min_tx_space = ALIGN(min_tx_space, 1024);
2063                 min_tx_space >>= 10;
2064                 /* software strips receive CRC, so leave room for it */
2065                 min_rx_space = mac->max_frame_size;
2066                 min_rx_space = ALIGN(min_rx_space, 1024);
2067                 min_rx_space >>= 10;
2068
2069                 /* If current Tx allocation is less than the min Tx FIFO size,
2070                  * and the min Tx FIFO size is less than the current Rx FIFO
2071                  * allocation, take space away from current Rx allocation */
2072                 if ((tx_space < min_tx_space) &&
2073                     ((min_tx_space - tx_space) < pba)) {
2074                         pba -= min_tx_space - tx_space;
2075
2076                         /* if short on rx space, rx wins and must trump tx
2077                          * adjustment or use Early Receive if available */
2078                         if ((pba < min_rx_space) &&
2079                             (!(adapter->flags & FLAG_HAS_ERT)))
2080                                 /* ERT enabled in e1000_configure_rx */
2081                                 pba = min_rx_space;
2082                 }
2083
2084                 ew32(PBA, pba);
2085         }
2086
2087
2088         /* flow control settings */
2089         /* The high water mark must be low enough to fit one full frame
2090          * (or the size used for early receive) above it in the Rx FIFO.
2091          * Set it to the lower of:
2092          * - 90% of the Rx FIFO size, and
2093          * - the full Rx FIFO size minus the early receive size (for parts
2094          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2095          * - the full Rx FIFO size minus one full frame */
2096         if (adapter->flags & FLAG_HAS_ERT)
2097                 hwm = min(((adapter->pba << 10) * 9 / 10),
2098                           ((adapter->pba << 10) - (E1000_ERT_2048 << 3)));
2099         else
2100                 hwm = min(((adapter->pba << 10) * 9 / 10),
2101                           ((adapter->pba << 10) - mac->max_frame_size));
2102
2103         mac->fc_high_water = hwm & 0xFFF8; /* 8-byte granularity */
2104         mac->fc_low_water = mac->fc_high_water - 8;
2105
2106         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2107                 mac->fc_pause_time = 0xFFFF;
2108         else
2109                 mac->fc_pause_time = E1000_FC_PAUSE_TIME;
2110         mac->fc = mac->original_fc;
2111
2112         /* Allow time for pending master requests to run */
2113         mac->ops.reset_hw(hw);
2114         ew32(WUC, 0);
2115
2116         if (mac->ops.init_hw(hw))
2117                 ndev_err(adapter->netdev, "Hardware Error\n");
2118
2119         e1000_update_mng_vlan(adapter);
2120
2121         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2122         ew32(VET, ETH_P_8021Q);
2123
2124         e1000e_reset_adaptive(hw);
2125         e1000_get_phy_info(hw);
2126
2127         if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2128                 u16 phy_data = 0;
2129                 /* speed up time to link by disabling smart power down, ignore
2130                  * the return value of this function because there is nothing
2131                  * different we would do if it failed */
2132                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2133                 phy_data &= ~IGP02E1000_PM_SPD;
2134                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2135         }
2136
2137         e1000_release_manageability(adapter);
2138 }
2139
2140 int e1000e_up(struct e1000_adapter *adapter)
2141 {
2142         struct e1000_hw *hw = &adapter->hw;
2143
2144         /* hardware has been reset, we need to reload some things */
2145         e1000_configure(adapter);
2146
2147         clear_bit(__E1000_DOWN, &adapter->state);
2148
2149         napi_enable(&adapter->napi);
2150         e1000_irq_enable(adapter);
2151
2152         /* fire a link change interrupt to start the watchdog */
2153         ew32(ICS, E1000_ICS_LSC);
2154         return 0;
2155 }
2156
2157 void e1000e_down(struct e1000_adapter *adapter)
2158 {
2159         struct net_device *netdev = adapter->netdev;
2160         struct e1000_hw *hw = &adapter->hw;
2161         u32 tctl, rctl;
2162
2163         /* signal that we're down so the interrupt handler does not
2164          * reschedule our watchdog timer */
2165         set_bit(__E1000_DOWN, &adapter->state);
2166
2167         /* disable receives in the hardware */
2168         rctl = er32(RCTL);
2169         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2170         /* flush and sleep below */
2171
2172         netif_stop_queue(netdev);
2173
2174         /* disable transmits in the hardware */
2175         tctl = er32(TCTL);
2176         tctl &= ~E1000_TCTL_EN;
2177         ew32(TCTL, tctl);
2178         /* flush both disables and wait for them to finish */
2179         e1e_flush();
2180         msleep(10);
2181
2182         napi_disable(&adapter->napi);
2183         e1000_irq_disable(adapter);
2184
2185         del_timer_sync(&adapter->watchdog_timer);
2186         del_timer_sync(&adapter->phy_info_timer);
2187
2188         netdev->tx_queue_len = adapter->tx_queue_len;
2189         netif_carrier_off(netdev);
2190         adapter->link_speed = 0;
2191         adapter->link_duplex = 0;
2192
2193         e1000e_reset(adapter);
2194         e1000_clean_tx_ring(adapter);
2195         e1000_clean_rx_ring(adapter);
2196
2197         /*
2198          * TODO: for power management, we could drop the link and
2199          * pci_disable_device here.
2200          */
2201 }
2202
2203 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2204 {
2205         might_sleep();
2206         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2207                 msleep(1);
2208         e1000e_down(adapter);
2209         e1000e_up(adapter);
2210         clear_bit(__E1000_RESETTING, &adapter->state);
2211 }
2212
2213 /**
2214  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2215  * @adapter: board private structure to initialize
2216  *
2217  * e1000_sw_init initializes the Adapter private data structure.
2218  * Fields are initialized based on PCI device information and
2219  * OS network device settings (MTU size).
2220  **/
2221 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2222 {
2223         struct e1000_hw *hw = &adapter->hw;
2224         struct net_device *netdev = adapter->netdev;
2225
2226         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2227         adapter->rx_ps_bsize0 = 128;
2228         hw->mac.max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2229         hw->mac.min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2230
2231         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2232         if (!adapter->tx_ring)
2233                 goto err;
2234
2235         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2236         if (!adapter->rx_ring)
2237                 goto err;
2238
2239         spin_lock_init(&adapter->tx_queue_lock);
2240
2241         /* Explicitly disable IRQ since the NIC can be in any state. */
2242         atomic_set(&adapter->irq_sem, 0);
2243         e1000_irq_disable(adapter);
2244
2245         spin_lock_init(&adapter->stats_lock);
2246
2247         set_bit(__E1000_DOWN, &adapter->state);
2248         return 0;
2249
2250 err:
2251         ndev_err(netdev, "Unable to allocate memory for queues\n");
2252         kfree(adapter->rx_ring);
2253         kfree(adapter->tx_ring);
2254         return -ENOMEM;
2255 }
2256
2257 /**
2258  * e1000_open - Called when a network interface is made active
2259  * @netdev: network interface device structure
2260  *
2261  * Returns 0 on success, negative value on failure
2262  *
2263  * The open entry point is called when a network interface is made
2264  * active by the system (IFF_UP).  At this point all resources needed
2265  * for transmit and receive operations are allocated, the interrupt
2266  * handler is registered with the OS, the watchdog timer is started,
2267  * and the stack is notified that the interface is ready.
2268  **/
2269 static int e1000_open(struct net_device *netdev)
2270 {
2271         struct e1000_adapter *adapter = netdev_priv(netdev);
2272         struct e1000_hw *hw = &adapter->hw;
2273         int err;
2274
2275         /* disallow open during test */
2276         if (test_bit(__E1000_TESTING, &adapter->state))
2277                 return -EBUSY;
2278
2279         /* allocate transmit descriptors */
2280         err = e1000e_setup_tx_resources(adapter);
2281         if (err)
2282                 goto err_setup_tx;
2283
2284         /* allocate receive descriptors */
2285         err = e1000e_setup_rx_resources(adapter);
2286         if (err)
2287                 goto err_setup_rx;
2288
2289         e1000e_power_up_phy(adapter);
2290
2291         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2292         if ((adapter->hw.mng_cookie.status &
2293              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
2294                 e1000_update_mng_vlan(adapter);
2295
2296         /* If AMT is enabled, let the firmware know that the network
2297          * interface is now open */
2298         if ((adapter->flags & FLAG_HAS_AMT) &&
2299             e1000e_check_mng_mode(&adapter->hw))
2300                 e1000_get_hw_control(adapter);
2301
2302         /* before we allocate an interrupt, we must be ready to handle it.
2303          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2304          * as soon as we call pci_request_irq, so we have to setup our
2305          * clean_rx handler before we do so.  */
2306         e1000_configure(adapter);
2307
2308         err = e1000_request_irq(adapter);
2309         if (err)
2310                 goto err_req_irq;
2311
2312         /* From here on the code is the same as e1000e_up() */
2313         clear_bit(__E1000_DOWN, &adapter->state);
2314
2315         napi_enable(&adapter->napi);
2316
2317         e1000_irq_enable(adapter);
2318
2319         /* fire a link status change interrupt to start the watchdog */
2320         ew32(ICS, E1000_ICS_LSC);
2321
2322         return 0;
2323
2324 err_req_irq:
2325         e1000_release_hw_control(adapter);
2326         e1000_power_down_phy(adapter);
2327         e1000e_free_rx_resources(adapter);
2328 err_setup_rx:
2329         e1000e_free_tx_resources(adapter);
2330 err_setup_tx:
2331         e1000e_reset(adapter);
2332
2333         return err;
2334 }
2335
2336 /**
2337  * e1000_close - Disables a network interface
2338  * @netdev: network interface device structure
2339  *
2340  * Returns 0, this is not allowed to fail
2341  *
2342  * The close entry point is called when an interface is de-activated
2343  * by the OS.  The hardware is still under the drivers control, but
2344  * needs to be disabled.  A global MAC reset is issued to stop the
2345  * hardware, and all transmit and receive resources are freed.
2346  **/
2347 static int e1000_close(struct net_device *netdev)
2348 {
2349         struct e1000_adapter *adapter = netdev_priv(netdev);
2350
2351         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
2352         e1000e_down(adapter);
2353         e1000_power_down_phy(adapter);
2354         e1000_free_irq(adapter);
2355
2356         e1000e_free_tx_resources(adapter);
2357         e1000e_free_rx_resources(adapter);
2358
2359         /* kill manageability vlan ID if supported, but not if a vlan with
2360          * the same ID is registered on the host OS (let 8021q kill it) */
2361         if ((adapter->hw.mng_cookie.status &
2362                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2363              !(adapter->vlgrp &&
2364                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
2365                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2366
2367         /* If AMT is enabled, let the firmware know that the network
2368          * interface is now closed */
2369         if ((adapter->flags & FLAG_HAS_AMT) &&
2370             e1000e_check_mng_mode(&adapter->hw))
2371                 e1000_release_hw_control(adapter);
2372
2373         return 0;
2374 }
2375 /**
2376  * e1000_set_mac - Change the Ethernet Address of the NIC
2377  * @netdev: network interface device structure
2378  * @p: pointer to an address structure
2379  *
2380  * Returns 0 on success, negative on failure
2381  **/
2382 static int e1000_set_mac(struct net_device *netdev, void *p)
2383 {
2384         struct e1000_adapter *adapter = netdev_priv(netdev);
2385         struct sockaddr *addr = p;
2386
2387         if (!is_valid_ether_addr(addr->sa_data))
2388                 return -EADDRNOTAVAIL;
2389
2390         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2391         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
2392
2393         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
2394
2395         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
2396                 /* activate the work around */
2397                 e1000e_set_laa_state_82571(&adapter->hw, 1);
2398
2399                 /* Hold a copy of the LAA in RAR[14] This is done so that
2400                  * between the time RAR[0] gets clobbered  and the time it
2401                  * gets fixed (in e1000_watchdog), the actual LAA is in one
2402                  * of the RARs and no incoming packets directed to this port
2403                  * are dropped. Eventually the LAA will be in RAR[0] and
2404                  * RAR[14] */
2405                 e1000e_rar_set(&adapter->hw,
2406                               adapter->hw.mac.addr,
2407                               adapter->hw.mac.rar_entry_count - 1);
2408         }
2409
2410         return 0;
2411 }
2412
2413 /* Need to wait a few seconds after link up to get diagnostic information from
2414  * the phy */
2415 static void e1000_update_phy_info(unsigned long data)
2416 {
2417         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2418         e1000_get_phy_info(&adapter->hw);
2419 }
2420
2421 /**
2422  * e1000e_update_stats - Update the board statistics counters
2423  * @adapter: board private structure
2424  **/
2425 void e1000e_update_stats(struct e1000_adapter *adapter)
2426 {
2427         struct e1000_hw *hw = &adapter->hw;
2428         struct pci_dev *pdev = adapter->pdev;
2429         unsigned long irq_flags;
2430         u16 phy_tmp;
2431
2432 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
2433
2434         /*
2435          * Prevent stats update while adapter is being reset, or if the pci
2436          * connection is down.
2437          */
2438         if (adapter->link_speed == 0)
2439                 return;
2440         if (pci_channel_offline(pdev))
2441                 return;
2442
2443         spin_lock_irqsave(&adapter->stats_lock, irq_flags);
2444
2445         /* these counters are modified from e1000_adjust_tbi_stats,
2446          * called from the interrupt context, so they must only
2447          * be written while holding adapter->stats_lock
2448          */
2449
2450         adapter->stats.crcerrs += er32(CRCERRS);
2451         adapter->stats.gprc += er32(GPRC);
2452         adapter->stats.gorcl += er32(GORCL);
2453         adapter->stats.gorch += er32(GORCH);
2454         adapter->stats.bprc += er32(BPRC);
2455         adapter->stats.mprc += er32(MPRC);
2456         adapter->stats.roc += er32(ROC);
2457
2458         if (adapter->flags & FLAG_HAS_STATS_PTC_PRC) {
2459                 adapter->stats.prc64 += er32(PRC64);
2460                 adapter->stats.prc127 += er32(PRC127);
2461                 adapter->stats.prc255 += er32(PRC255);
2462                 adapter->stats.prc511 += er32(PRC511);
2463                 adapter->stats.prc1023 += er32(PRC1023);
2464                 adapter->stats.prc1522 += er32(PRC1522);
2465                 adapter->stats.symerrs += er32(SYMERRS);
2466                 adapter->stats.sec += er32(SEC);
2467         }
2468
2469         adapter->stats.mpc += er32(MPC);
2470         adapter->stats.scc += er32(SCC);
2471         adapter->stats.ecol += er32(ECOL);
2472         adapter->stats.mcc += er32(MCC);
2473         adapter->stats.latecol += er32(LATECOL);
2474         adapter->stats.dc += er32(DC);
2475         adapter->stats.rlec += er32(RLEC);
2476         adapter->stats.xonrxc += er32(XONRXC);
2477         adapter->stats.xontxc += er32(XONTXC);
2478         adapter->stats.xoffrxc += er32(XOFFRXC);
2479         adapter->stats.xofftxc += er32(XOFFTXC);
2480         adapter->stats.fcruc += er32(FCRUC);
2481         adapter->stats.gptc += er32(GPTC);
2482         adapter->stats.gotcl += er32(GOTCL);
2483         adapter->stats.gotch += er32(GOTCH);
2484         adapter->stats.rnbc += er32(RNBC);
2485         adapter->stats.ruc += er32(RUC);
2486         adapter->stats.rfc += er32(RFC);
2487         adapter->stats.rjc += er32(RJC);
2488         adapter->stats.torl += er32(TORL);
2489         adapter->stats.torh += er32(TORH);
2490         adapter->stats.totl += er32(TOTL);
2491         adapter->stats.toth += er32(TOTH);
2492         adapter->stats.tpr += er32(TPR);
2493
2494         if (adapter->flags & FLAG_HAS_STATS_PTC_PRC) {
2495                 adapter->stats.ptc64 += er32(PTC64);
2496                 adapter->stats.ptc127 += er32(PTC127);
2497                 adapter->stats.ptc255 += er32(PTC255);
2498                 adapter->stats.ptc511 += er32(PTC511);
2499                 adapter->stats.ptc1023 += er32(PTC1023);
2500                 adapter->stats.ptc1522 += er32(PTC1522);
2501         }
2502
2503         adapter->stats.mptc += er32(MPTC);
2504         adapter->stats.bptc += er32(BPTC);
2505
2506         /* used for adaptive IFS */
2507
2508         hw->mac.tx_packet_delta = er32(TPT);
2509         adapter->stats.tpt += hw->mac.tx_packet_delta;
2510         hw->mac.collision_delta = er32(COLC);
2511         adapter->stats.colc += hw->mac.collision_delta;
2512
2513         adapter->stats.algnerrc += er32(ALGNERRC);
2514         adapter->stats.rxerrc += er32(RXERRC);
2515         adapter->stats.tncrs += er32(TNCRS);
2516         adapter->stats.cexterr += er32(CEXTERR);
2517         adapter->stats.tsctc += er32(TSCTC);
2518         adapter->stats.tsctfc += er32(TSCTFC);
2519
2520         adapter->stats.iac += er32(IAC);
2521
2522         if (adapter->flags & FLAG_HAS_STATS_ICR_ICT) {
2523                 adapter->stats.icrxoc += er32(ICRXOC);
2524                 adapter->stats.icrxptc += er32(ICRXPTC);
2525                 adapter->stats.icrxatc += er32(ICRXATC);
2526                 adapter->stats.ictxptc += er32(ICTXPTC);
2527                 adapter->stats.ictxatc += er32(ICTXATC);
2528                 adapter->stats.ictxqec += er32(ICTXQEC);
2529                 adapter->stats.ictxqmtc += er32(ICTXQMTC);
2530                 adapter->stats.icrxdmtc += er32(ICRXDMTC);
2531         }
2532
2533         /* Fill out the OS statistics structure */
2534         adapter->net_stats.rx_packets = adapter->stats.gprc;
2535         adapter->net_stats.tx_packets = adapter->stats.gptc;
2536         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
2537         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
2538         adapter->net_stats.multicast = adapter->stats.mprc;
2539         adapter->net_stats.collisions = adapter->stats.colc;
2540
2541         /* Rx Errors */
2542
2543         /* RLEC on some newer hardware can be incorrect so build
2544         * our own version based on RUC and ROC */
2545         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
2546                 adapter->stats.crcerrs + adapter->stats.algnerrc +
2547                 adapter->stats.ruc + adapter->stats.roc +
2548                 adapter->stats.cexterr;
2549         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
2550                                               adapter->stats.roc;
2551         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
2552         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
2553         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
2554
2555         /* Tx Errors */
2556         adapter->net_stats.tx_errors = adapter->stats.ecol +
2557                                        adapter->stats.latecol;
2558         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
2559         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
2560         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
2561
2562         /* Tx Dropped needs to be maintained elsewhere */
2563
2564         /* Phy Stats */
2565         if (hw->media_type == e1000_media_type_copper) {
2566                 if ((adapter->link_speed == SPEED_1000) &&
2567                    (!e1e_rphy(hw, PHY_1000T_STATUS, &phy_tmp))) {
2568                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
2569                         adapter->phy_stats.idle_errors += phy_tmp;
2570                 }
2571         }
2572
2573         /* Management Stats */
2574         adapter->stats.mgptc += er32(MGTPTC);
2575         adapter->stats.mgprc += er32(MGTPRC);
2576         adapter->stats.mgpdc += er32(MGTPDC);
2577
2578         spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
2579 }
2580
2581 static void e1000_print_link_info(struct e1000_adapter *adapter)
2582 {
2583         struct net_device *netdev = adapter->netdev;
2584         struct e1000_hw *hw = &adapter->hw;
2585         u32 ctrl = er32(CTRL);
2586
2587         ndev_info(netdev,
2588                 "Link is Up %d Mbps %s, Flow Control: %s\n",
2589                 adapter->link_speed,
2590                 (adapter->link_duplex == FULL_DUPLEX) ?
2591                                 "Full Duplex" : "Half Duplex",
2592                 ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
2593                                 "RX/TX" :
2594                 ((ctrl & E1000_CTRL_RFCE) ? "RX" :
2595                 ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
2596 }
2597
2598 /**
2599  * e1000_watchdog - Timer Call-back
2600  * @data: pointer to adapter cast into an unsigned long
2601  **/
2602 static void e1000_watchdog(unsigned long data)
2603 {
2604         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2605
2606         /* Do the rest outside of interrupt context */
2607         schedule_work(&adapter->watchdog_task);
2608
2609         /* TODO: make this use queue_delayed_work() */
2610 }
2611
2612 static void e1000_watchdog_task(struct work_struct *work)
2613 {
2614         struct e1000_adapter *adapter = container_of(work,
2615                                         struct e1000_adapter, watchdog_task);
2616
2617         struct net_device *netdev = adapter->netdev;
2618         struct e1000_mac_info *mac = &adapter->hw.mac;
2619         struct e1000_ring *tx_ring = adapter->tx_ring;
2620         struct e1000_hw *hw = &adapter->hw;
2621         u32 link, tctl;
2622         s32 ret_val;
2623         int tx_pending = 0;
2624
2625         if ((netif_carrier_ok(netdev)) &&
2626             (er32(STATUS) & E1000_STATUS_LU))
2627                 goto link_up;
2628
2629         ret_val = mac->ops.check_for_link(hw);
2630         if ((ret_val == E1000_ERR_PHY) &&
2631             (adapter->hw.phy.type == e1000_phy_igp_3) &&
2632             (er32(CTRL) &
2633              E1000_PHY_CTRL_GBE_DISABLE)) {
2634                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
2635                 ndev_info(netdev,
2636                         "Gigabit has been disabled, downgrading speed\n");
2637         }
2638
2639         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
2640             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
2641                 e1000_update_mng_vlan(adapter);
2642
2643         if ((adapter->hw.media_type == e1000_media_type_internal_serdes) &&
2644            !(er32(TXCW) & E1000_TXCW_ANE))
2645                 link = adapter->hw.mac.serdes_has_link;
2646         else
2647                 link = er32(STATUS) & E1000_STATUS_LU;
2648
2649         if (link) {
2650                 if (!netif_carrier_ok(netdev)) {
2651                         bool txb2b = 1;
2652                         mac->ops.get_link_up_info(&adapter->hw,
2653                                                    &adapter->link_speed,
2654                                                    &adapter->link_duplex);
2655                         e1000_print_link_info(adapter);
2656                         /* tweak tx_queue_len according to speed/duplex
2657                          * and adjust the timeout factor */
2658                         netdev->tx_queue_len = adapter->tx_queue_len;
2659                         adapter->tx_timeout_factor = 1;
2660                         switch (adapter->link_speed) {
2661                         case SPEED_10:
2662                                 txb2b = 0;
2663                                 netdev->tx_queue_len = 10;
2664                                 adapter->tx_timeout_factor = 14;
2665                                 break;
2666                         case SPEED_100:
2667                                 txb2b = 0;
2668                                 netdev->tx_queue_len = 100;
2669                                 /* maybe add some timeout factor ? */
2670                                 break;
2671                         }
2672
2673                         /* workaround: re-program speed mode bit after
2674                          * link-up event */
2675                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
2676                             !txb2b) {
2677                                 u32 tarc0;
2678                                 tarc0 = er32(TARC0);
2679                                 tarc0 &= ~SPEED_MODE_BIT;
2680                                 ew32(TARC0, tarc0);
2681                         }
2682
2683                         /* disable TSO for pcie and 10/100 speeds, to avoid
2684                          * some hardware issues */
2685                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
2686                                 switch (adapter->link_speed) {
2687                                 case SPEED_10:
2688                                 case SPEED_100:
2689                                         ndev_info(netdev,
2690                                         "10/100 speed: disabling TSO\n");
2691                                         netdev->features &= ~NETIF_F_TSO;
2692                                         netdev->features &= ~NETIF_F_TSO6;
2693                                         break;
2694                                 case SPEED_1000:
2695                                         netdev->features |= NETIF_F_TSO;
2696                                         netdev->features |= NETIF_F_TSO6;
2697                                         break;
2698                                 default:
2699                                         /* oops */
2700                                         break;
2701                                 }
2702                         }
2703
2704                         /* enable transmits in the hardware, need to do this
2705                          * after setting TARC0 */
2706                         tctl = er32(TCTL);
2707                         tctl |= E1000_TCTL_EN;
2708                         ew32(TCTL, tctl);
2709
2710                         netif_carrier_on(netdev);
2711                         netif_wake_queue(netdev);
2712
2713                         if (!test_bit(__E1000_DOWN, &adapter->state))
2714                                 mod_timer(&adapter->phy_info_timer,
2715                                           round_jiffies(jiffies + 2 * HZ));
2716                 } else {
2717                         /* make sure the receive unit is started */
2718                         if (adapter->flags & FLAG_RX_NEEDS_RESTART) {
2719                                 u32 rctl = er32(RCTL);
2720                                 ew32(RCTL, rctl |
2721                                                 E1000_RCTL_EN);
2722                         }
2723                 }
2724         } else {
2725                 if (netif_carrier_ok(netdev)) {
2726                         adapter->link_speed = 0;
2727                         adapter->link_duplex = 0;
2728                         ndev_info(netdev, "Link is Down\n");
2729                         netif_carrier_off(netdev);
2730                         netif_stop_queue(netdev);
2731                         if (!test_bit(__E1000_DOWN, &adapter->state))
2732                                 mod_timer(&adapter->phy_info_timer,
2733                                           round_jiffies(jiffies + 2 * HZ));
2734
2735                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
2736                                 schedule_work(&adapter->reset_task);
2737                 }
2738         }
2739
2740 link_up:
2741         e1000e_update_stats(adapter);
2742
2743         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2744         adapter->tpt_old = adapter->stats.tpt;
2745         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
2746         adapter->colc_old = adapter->stats.colc;
2747
2748         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2749         adapter->gorcl_old = adapter->stats.gorcl;
2750         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2751         adapter->gotcl_old = adapter->stats.gotcl;
2752
2753         e1000e_update_adaptive(&adapter->hw);
2754
2755         if (!netif_carrier_ok(netdev)) {
2756                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
2757                                tx_ring->count);
2758                 if (tx_pending) {
2759                         /* We've lost link, so the controller stops DMA,
2760                          * but we've got queued Tx work that's never going
2761                          * to get done, so reset controller to flush Tx.
2762                          * (Do the reset outside of interrupt context). */
2763                         adapter->tx_timeout_count++;
2764                         schedule_work(&adapter->reset_task);
2765                 }
2766         }
2767
2768         /* Cause software interrupt to ensure rx ring is cleaned */
2769         ew32(ICS, E1000_ICS_RXDMT0);
2770
2771         /* Force detection of hung controller every watchdog period */
2772         adapter->detect_tx_hung = 1;
2773
2774         /* With 82571 controllers, LAA may be overwritten due to controller
2775          * reset from the other port. Set the appropriate LAA in RAR[0] */
2776         if (e1000e_get_laa_state_82571(hw))
2777                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
2778
2779         /* Reset the timer */
2780         if (!test_bit(__E1000_DOWN, &adapter->state))
2781                 mod_timer(&adapter->watchdog_timer,
2782                           round_jiffies(jiffies + 2 * HZ));
2783 }
2784
2785 #define E1000_TX_FLAGS_CSUM             0x00000001
2786 #define E1000_TX_FLAGS_VLAN             0x00000002
2787 #define E1000_TX_FLAGS_TSO              0x00000004
2788 #define E1000_TX_FLAGS_IPV4             0x00000008
2789 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2790 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2791
2792 static int e1000_tso(struct e1000_adapter *adapter,
2793                      struct sk_buff *skb)
2794 {
2795         struct e1000_ring *tx_ring = adapter->tx_ring;
2796         struct e1000_context_desc *context_desc;
2797         struct e1000_buffer *buffer_info;
2798         unsigned int i;
2799         u32 cmd_length = 0;
2800         u16 ipcse = 0, tucse, mss;
2801         u8 ipcss, ipcso, tucss, tucso, hdr_len;
2802         int err;
2803
2804         if (skb_is_gso(skb)) {
2805                 if (skb_header_cloned(skb)) {
2806                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2807                         if (err)
2808                                 return err;
2809                 }
2810
2811                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2812                 mss = skb_shinfo(skb)->gso_size;
2813                 if (skb->protocol == htons(ETH_P_IP)) {
2814                         struct iphdr *iph = ip_hdr(skb);
2815                         iph->tot_len = 0;
2816                         iph->check = 0;
2817                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2818                                                                  iph->daddr, 0,
2819                                                                  IPPROTO_TCP,
2820                                                                  0);
2821                         cmd_length = E1000_TXD_CMD_IP;
2822                         ipcse = skb_transport_offset(skb) - 1;
2823                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
2824                         ipv6_hdr(skb)->payload_len = 0;
2825                         tcp_hdr(skb)->check =
2826                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2827                                                  &ipv6_hdr(skb)->daddr,
2828                                                  0, IPPROTO_TCP, 0);
2829                         ipcse = 0;
2830                 }
2831                 ipcss = skb_network_offset(skb);
2832                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2833                 tucss = skb_transport_offset(skb);
2834                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2835                 tucse = 0;
2836
2837                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2838                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2839
2840                 i = tx_ring->next_to_use;
2841                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2842                 buffer_info = &tx_ring->buffer_info[i];
2843
2844                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2845                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2846                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2847                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2848                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2849                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2850                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2851                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2852                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2853
2854                 buffer_info->time_stamp = jiffies;
2855                 buffer_info->next_to_watch = i;
2856
2857                 i++;
2858                 if (i == tx_ring->count)
2859                         i = 0;
2860                 tx_ring->next_to_use = i;
2861
2862                 return 1;
2863         }
2864
2865         return 0;
2866 }
2867
2868 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
2869 {
2870         struct e1000_ring *tx_ring = adapter->tx_ring;
2871         struct e1000_context_desc *context_desc;
2872         struct e1000_buffer *buffer_info;
2873         unsigned int i;
2874         u8 css;
2875
2876         if (skb->ip_summed == CHECKSUM_PARTIAL) {
2877                 css = skb_transport_offset(skb);
2878
2879                 i = tx_ring->next_to_use;
2880                 buffer_info = &tx_ring->buffer_info[i];
2881                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2882
2883                 context_desc->lower_setup.ip_config = 0;
2884                 context_desc->upper_setup.tcp_fields.tucss = css;
2885                 context_desc->upper_setup.tcp_fields.tucso =
2886                                         css + skb->csum_offset;
2887                 context_desc->upper_setup.tcp_fields.tucse = 0;
2888                 context_desc->tcp_seg_setup.data = 0;
2889                 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
2890
2891                 buffer_info->time_stamp = jiffies;
2892                 buffer_info->next_to_watch = i;
2893
2894                 i++;
2895                 if (i == tx_ring->count)
2896                         i = 0;
2897                 tx_ring->next_to_use = i;
2898
2899                 return 1;
2900         }
2901
2902         return 0;
2903 }
2904
2905 #define E1000_MAX_PER_TXD       8192
2906 #define E1000_MAX_TXD_PWR       12
2907
2908 static int e1000_tx_map(struct e1000_adapter *adapter,
2909                         struct sk_buff *skb, unsigned int first,
2910                         unsigned int max_per_txd, unsigned int nr_frags,
2911                         unsigned int mss)
2912 {
2913         struct e1000_ring *tx_ring = adapter->tx_ring;
2914         struct e1000_buffer *buffer_info;
2915         unsigned int len = skb->len - skb->data_len;
2916         unsigned int offset = 0, size, count = 0, i;
2917         unsigned int f;
2918
2919         i = tx_ring->next_to_use;
2920
2921         while (len) {
2922                 buffer_info = &tx_ring->buffer_info[i];
2923                 size = min(len, max_per_txd);
2924
2925                 /* Workaround for premature desc write-backs
2926                  * in TSO mode.  Append 4-byte sentinel desc */
2927                 if (mss && !nr_frags && size == len && size > 8)
2928                         size -= 4;
2929
2930                 buffer_info->length = size;
2931                 /* set time_stamp *before* dma to help avoid a possible race */
2932                 buffer_info->time_stamp = jiffies;
2933                 buffer_info->dma =
2934                         pci_map_single(adapter->pdev,
2935                                 skb->data + offset,
2936                                 size,
2937                                 PCI_DMA_TODEVICE);
2938                 if (pci_dma_mapping_error(buffer_info->dma)) {
2939                         dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
2940                         adapter->tx_dma_failed++;
2941                         return -1;
2942                 }
2943                 buffer_info->next_to_watch = i;
2944
2945                 len -= size;
2946                 offset += size;
2947                 count++;
2948                 i++;
2949                 if (i == tx_ring->count)
2950                         i = 0;
2951         }
2952
2953         for (f = 0; f < nr_frags; f++) {
2954                 struct skb_frag_struct *frag;
2955
2956                 frag = &skb_shinfo(skb)->frags[f];
2957                 len = frag->size;
2958                 offset = frag->page_offset;
2959
2960                 while (len) {
2961                         buffer_info = &tx_ring->buffer_info[i];
2962                         size = min(len, max_per_txd);
2963                         /* Workaround for premature desc write-backs
2964                          * in TSO mode.  Append 4-byte sentinel desc */
2965                         if (mss && f == (nr_frags-1) && size == len && size > 8)
2966                                 size -= 4;
2967
2968                         buffer_info->length = size;
2969                         buffer_info->time_stamp = jiffies;
2970                         buffer_info->dma =
2971                                 pci_map_page(adapter->pdev,
2972                                         frag->page,
2973                                         offset,
2974                                         size,
2975                                         PCI_DMA_TODEVICE);
2976                         if (pci_dma_mapping_error(buffer_info->dma)) {
2977                                 dev_err(&adapter->pdev->dev,
2978                                         "TX DMA page map failed\n");
2979                                 adapter->tx_dma_failed++;
2980                                 return -1;
2981                         }
2982
2983                         buffer_info->next_to_watch = i;
2984
2985                         len -= size;
2986                         offset += size;
2987                         count++;
2988
2989                         i++;
2990                         if (i == tx_ring->count)
2991                                 i = 0;
2992                 }
2993         }
2994
2995         if (i == 0)
2996                 i = tx_ring->count - 1;
2997         else
2998                 i--;
2999
3000         tx_ring->buffer_info[i].skb = skb;
3001         tx_ring->buffer_info[first].next_to_watch = i;
3002
3003         return count;
3004 }
3005
3006 static void e1000_tx_queue(struct e1000_adapter *adapter,
3007                            int tx_flags, int count)
3008 {
3009         struct e1000_ring *tx_ring = adapter->tx_ring;
3010         struct e1000_tx_desc *tx_desc = NULL;
3011         struct e1000_buffer *buffer_info;
3012         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3013         unsigned int i;
3014
3015         if (tx_flags & E1000_TX_FLAGS_TSO) {
3016                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3017                              E1000_TXD_CMD_TSE;
3018                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3019
3020                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3021                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3022         }
3023
3024         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3025                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3026                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3027         }
3028
3029         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3030                 txd_lower |= E1000_TXD_CMD_VLE;
3031                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3032         }
3033
3034         i = tx_ring->next_to_use;
3035
3036         while (count--) {
3037                 buffer_info = &tx_ring->buffer_info[i];
3038                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3039                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3040                 tx_desc->lower.data =
3041                         cpu_to_le32(txd_lower | buffer_info->length);
3042                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3043
3044                 i++;
3045                 if (i == tx_ring->count)
3046                         i = 0;
3047         }
3048
3049         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3050
3051         /* Force memory writes to complete before letting h/w
3052          * know there are new descriptors to fetch.  (Only
3053          * applicable for weak-ordered memory model archs,
3054          * such as IA-64). */
3055         wmb();
3056
3057         tx_ring->next_to_use = i;
3058         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3059         /* we need this if more than one processor can write to our tail
3060          * at a time, it synchronizes IO on IA64/Altix systems */
3061         mmiowb();
3062 }
3063
3064 #define MINIMUM_DHCP_PACKET_SIZE 282
3065 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3066                                     struct sk_buff *skb)
3067 {
3068         struct e1000_hw *hw =  &adapter->hw;
3069         u16 length, offset;
3070
3071         if (vlan_tx_tag_present(skb)) {
3072                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3073                     && (adapter->hw.mng_cookie.status &
3074                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3075                         return 0;
3076         }
3077
3078         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3079                 return 0;
3080
3081         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3082                 return 0;
3083
3084         {
3085                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3086                 struct udphdr *udp;
3087
3088                 if (ip->protocol != IPPROTO_UDP)
3089                         return 0;
3090
3091                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3092                 if (ntohs(udp->dest) != 67)
3093                         return 0;
3094
3095                 offset = (u8 *)udp + 8 - skb->data;
3096                 length = skb->len - offset;
3097                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
3098         }
3099
3100         return 0;
3101 }
3102
3103 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3104 {
3105         struct e1000_adapter *adapter = netdev_priv(netdev);
3106
3107         netif_stop_queue(netdev);
3108         /* Herbert's original patch had:
3109          *  smp_mb__after_netif_stop_queue();
3110          * but since that doesn't exist yet, just open code it. */
3111         smp_mb();
3112
3113         /* We need to check again in a case another CPU has just
3114          * made room available. */
3115         if (e1000_desc_unused(adapter->tx_ring) < size)
3116                 return -EBUSY;
3117
3118         /* A reprieve! */
3119         netif_start_queue(netdev);
3120         ++adapter->restart_queue;
3121         return 0;
3122 }
3123
3124 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
3125 {
3126         struct e1000_adapter *adapter = netdev_priv(netdev);
3127
3128         if (e1000_desc_unused(adapter->tx_ring) >= size)
3129                 return 0;
3130         return __e1000_maybe_stop_tx(netdev, size);
3131 }
3132
3133 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3134 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3135 {
3136         struct e1000_adapter *adapter = netdev_priv(netdev);
3137         struct e1000_ring *tx_ring = adapter->tx_ring;
3138         unsigned int first;
3139         unsigned int max_per_txd = E1000_MAX_PER_TXD;
3140         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3141         unsigned int tx_flags = 0;
3142         unsigned int len = skb->len - skb->data_len;
3143         unsigned long irq_flags;
3144         unsigned int nr_frags;
3145         unsigned int mss;
3146         int count = 0;
3147         int tso;
3148         unsigned int f;
3149
3150         if (test_bit(__E1000_DOWN, &adapter->state)) {
3151                 dev_kfree_skb_any(skb);
3152                 return NETDEV_TX_OK;
3153         }
3154
3155         if (skb->len <= 0) {
3156                 dev_kfree_skb_any(skb);
3157                 return NETDEV_TX_OK;
3158         }
3159
3160         mss = skb_shinfo(skb)->gso_size;
3161         /* The controller does a simple calculation to
3162          * make sure there is enough room in the FIFO before
3163          * initiating the DMA for each buffer.  The calc is:
3164          * 4 = ceil(buffer len/mss).  To make sure we don't
3165          * overrun the FIFO, adjust the max buffer len if mss
3166          * drops. */
3167         if (mss) {
3168                 u8 hdr_len;
3169                 max_per_txd = min(mss << 2, max_per_txd);
3170                 max_txd_pwr = fls(max_per_txd) - 1;
3171
3172                 /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
3173                 * points to just header, pull a few bytes of payload from
3174                 * frags into skb->data */
3175                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3176                 if (skb->data_len && (hdr_len == len)) {
3177                         unsigned int pull_size;
3178
3179                         pull_size = min((unsigned int)4, skb->data_len);
3180                         if (!__pskb_pull_tail(skb, pull_size)) {
3181                                 ndev_err(netdev,
3182                                          "__pskb_pull_tail failed.\n");
3183                                 dev_kfree_skb_any(skb);
3184                                 return NETDEV_TX_OK;
3185                         }
3186                         len = skb->len - skb->data_len;
3187                 }
3188         }
3189
3190         /* reserve a descriptor for the offload context */
3191         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3192                 count++;
3193         count++;
3194
3195         count += TXD_USE_COUNT(len, max_txd_pwr);
3196
3197         nr_frags = skb_shinfo(skb)->nr_frags;
3198         for (f = 0; f < nr_frags; f++)
3199                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3200                                        max_txd_pwr);
3201
3202         if (adapter->hw.mac.tx_pkt_filtering)
3203                 e1000_transfer_dhcp_info(adapter, skb);
3204
3205         if (!spin_trylock_irqsave(&adapter->tx_queue_lock, irq_flags))
3206                 /* Collision - tell upper layer to requeue */
3207                 return NETDEV_TX_LOCKED;
3208
3209         /* need: count + 2 desc gap to keep tail from touching
3210          * head, otherwise try next time */
3211         if (e1000_maybe_stop_tx(netdev, count + 2)) {
3212                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3213                 return NETDEV_TX_BUSY;
3214         }
3215
3216         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3217                 tx_flags |= E1000_TX_FLAGS_VLAN;
3218                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3219         }
3220
3221         first = tx_ring->next_to_use;
3222
3223         tso = e1000_tso(adapter, skb);
3224         if (tso < 0) {
3225                 dev_kfree_skb_any(skb);
3226                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3227                 return NETDEV_TX_OK;
3228         }
3229
3230         if (tso)
3231                 tx_flags |= E1000_TX_FLAGS_TSO;
3232         else if (e1000_tx_csum(adapter, skb))
3233                 tx_flags |= E1000_TX_FLAGS_CSUM;
3234
3235         /* Old method was to assume IPv4 packet by default if TSO was enabled.
3236          * 82571 hardware supports TSO capabilities for IPv6 as well...
3237          * no longer assume, we must. */
3238         if (skb->protocol == htons(ETH_P_IP))
3239                 tx_flags |= E1000_TX_FLAGS_IPV4;
3240
3241         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
3242         if (count < 0) {
3243                 /* handle pci_map_single() error in e1000_tx_map */
3244                 dev_kfree_skb_any(skb);
3245                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3246                 return NETDEV_TX_OK;
3247         }
3248
3249         e1000_tx_queue(adapter, tx_flags, count);
3250
3251         netdev->trans_start = jiffies;
3252
3253         /* Make sure there is space in the ring for the next send. */
3254         e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
3255
3256         spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3257         return NETDEV_TX_OK;
3258 }
3259
3260 /**
3261  * e1000_tx_timeout - Respond to a Tx Hang
3262  * @netdev: network interface device structure
3263  **/
3264 static void e1000_tx_timeout(struct net_device *netdev)
3265 {
3266         struct e1000_adapter *adapter = netdev_priv(netdev);
3267
3268         /* Do the reset outside of interrupt context */
3269         adapter->tx_timeout_count++;
3270         schedule_work(&adapter->reset_task);
3271 }
3272
3273 static void e1000_reset_task(struct work_struct *work)
3274 {
3275         struct e1000_adapter *adapter;
3276         adapter = container_of(work, struct e1000_adapter, reset_task);
3277
3278         e1000e_reinit_locked(adapter);
3279 }
3280
3281 /**
3282  * e1000_get_stats - Get System Network Statistics
3283  * @netdev: network interface device structure
3284  *
3285  * Returns the address of the device statistics structure.
3286  * The statistics are actually updated from the timer callback.
3287  **/
3288 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3289 {
3290         struct e1000_adapter *adapter = netdev_priv(netdev);
3291
3292         /* only return the current stats */
3293         return &adapter->net_stats;
3294 }
3295
3296 /**
3297  * e1000_change_mtu - Change the Maximum Transfer Unit
3298  * @netdev: network interface device structure
3299  * @new_mtu: new value for maximum frame size
3300  *
3301  * Returns 0 on success, negative on failure
3302  **/
3303 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3304 {
3305         struct e1000_adapter *adapter = netdev_priv(netdev);
3306         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3307
3308         if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
3309             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3310                 ndev_err(netdev, "Invalid MTU setting\n");
3311                 return -EINVAL;
3312         }
3313
3314         /* Jumbo frame size limits */
3315         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
3316                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
3317                         ndev_err(netdev, "Jumbo Frames not supported.\n");
3318                         return -EINVAL;
3319                 }
3320                 if (adapter->hw.phy.type == e1000_phy_ife) {
3321                         ndev_err(netdev, "Jumbo Frames not supported.\n");
3322                         return -EINVAL;
3323                 }
3324         }
3325
3326 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3327         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3328                 ndev_err(netdev, "MTU > 9216 not supported.\n");
3329                 return -EINVAL;
3330         }
3331
3332         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3333                 msleep(1);
3334         /* e1000e_down has a dependency on max_frame_size */
3335         adapter->hw.mac.max_frame_size = max_frame;
3336         if (netif_running(netdev))
3337                 e1000e_down(adapter);
3338
3339         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3340          * means we reserve 2 more, this pushes us to allocate from the next
3341          * larger slab size.
3342          * i.e. RXBUFFER_2048 --> size-4096 slab */
3343
3344         if (max_frame <= 256)
3345                 adapter->rx_buffer_len = 256;
3346         else if (max_frame <= 512)
3347                 adapter->rx_buffer_len = 512;
3348         else if (max_frame <= 1024)
3349                 adapter->rx_buffer_len = 1024;
3350         else if (max_frame <= 2048)
3351                 adapter->rx_buffer_len = 2048;
3352         else
3353                 adapter->rx_buffer_len = 4096;
3354
3355         /* adjust allocation if LPE protects us, and we aren't using SBP */
3356         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3357              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
3358                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
3359                                          + ETH_FCS_LEN ;
3360
3361         ndev_info(netdev, "changing MTU from %d to %d\n",
3362                 netdev->mtu, new_mtu);
3363         netdev->mtu = new_mtu;
3364
3365         if (netif_running(netdev))
3366                 e1000e_up(adapter);
3367         else
3368                 e1000e_reset(adapter);
3369
3370         clear_bit(__E1000_RESETTING, &adapter->state);
3371
3372         return 0;
3373 }
3374
3375 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
3376                            int cmd)
3377 {
3378         struct e1000_adapter *adapter = netdev_priv(netdev);
3379         struct mii_ioctl_data *data = if_mii(ifr);
3380         unsigned long irq_flags;
3381
3382         if (adapter->hw.media_type != e1000_media_type_copper)
3383                 return -EOPNOTSUPP;
3384
3385         switch (cmd) {
3386         case SIOCGMIIPHY:
3387                 data->phy_id = adapter->hw.phy.addr;
3388                 break;
3389         case SIOCGMIIREG:
3390                 if (!capable(CAP_NET_ADMIN))
3391                         return -EPERM;
3392                 spin_lock_irqsave(&adapter->stats_lock, irq_flags);
3393                 if (e1e_rphy(&adapter->hw, data->reg_num & 0x1F,
3394                                    &data->val_out)) {
3395                         spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
3396                         return -EIO;
3397                 }
3398                 spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
3399                 break;
3400         case SIOCSMIIREG:
3401         default:
3402                 return -EOPNOTSUPP;
3403         }
3404         return 0;
3405 }
3406
3407 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
3408 {
3409         switch (cmd) {
3410         case SIOCGMIIPHY:
3411         case SIOCGMIIREG:
3412         case SIOCSMIIREG:
3413                 return e1000_mii_ioctl(netdev, ifr, cmd);
3414         default:
3415                 return -EOPNOTSUPP;
3416         }
3417 }
3418
3419 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
3420 {
3421         struct net_device *netdev = pci_get_drvdata(pdev);
3422         struct e1000_adapter *adapter = netdev_priv(netdev);
3423         struct e1000_hw *hw = &adapter->hw;
3424         u32 ctrl, ctrl_ext, rctl, status;
3425         u32 wufc = adapter->wol;
3426         int retval = 0;
3427
3428         netif_device_detach(netdev);
3429
3430         if (netif_running(netdev)) {
3431                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3432                 e1000e_down(adapter);
3433                 e1000_free_irq(adapter);
3434         }
3435
3436         retval = pci_save_state(pdev);
3437         if (retval)
3438                 return retval;
3439
3440         status = er32(STATUS);
3441         if (status & E1000_STATUS_LU)
3442                 wufc &= ~E1000_WUFC_LNKC;
3443
3444         if (wufc) {
3445                 e1000_setup_rctl(adapter);
3446                 e1000_set_multi(netdev);
3447
3448                 /* turn on all-multi mode if wake on multicast is enabled */
3449                 if (wufc & E1000_WUFC_MC) {
3450                         rctl = er32(RCTL);
3451                         rctl |= E1000_RCTL_MPE;
3452                         ew32(RCTL, rctl);
3453                 }
3454
3455                 ctrl = er32(CTRL);
3456                 /* advertise wake from D3Cold */
3457                 #define E1000_CTRL_ADVD3WUC 0x00100000
3458                 /* phy power management enable */
3459                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
3460                 ctrl |= E1000_CTRL_ADVD3WUC |
3461                         E1000_CTRL_EN_PHY_PWR_MGMT;
3462                 ew32(CTRL, ctrl);
3463
3464                 if (adapter->hw.media_type == e1000_media_type_fiber ||
3465                    adapter->hw.media_type == e1000_media_type_internal_serdes) {
3466                         /* keep the laser running in D3 */
3467                         ctrl_ext = er32(CTRL_EXT);
3468                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
3469                         ew32(CTRL_EXT, ctrl_ext);
3470                 }
3471
3472                 /* Allow time for pending master requests to run */
3473                 e1000e_disable_pcie_master(&adapter->hw);
3474
3475                 ew32(WUC, E1000_WUC_PME_EN);
3476                 ew32(WUFC, wufc);
3477                 pci_enable_wake(pdev, PCI_D3hot, 1);
3478                 pci_enable_wake(pdev, PCI_D3cold, 1);
3479         } else {
3480                 ew32(WUC, 0);
3481                 ew32(WUFC, 0);
3482                 pci_enable_wake(pdev, PCI_D3hot, 0);
3483                 pci_enable_wake(pdev, PCI_D3cold, 0);
3484         }
3485
3486         e1000_release_manageability(adapter);
3487
3488         /* make sure adapter isn't asleep if manageability is enabled */
3489         if (adapter->flags & FLAG_MNG_PT_ENABLED) {
3490                 pci_enable_wake(pdev, PCI_D3hot, 1);
3491                 pci_enable_wake(pdev, PCI_D3cold, 1);
3492         }
3493
3494         if (adapter->hw.phy.type == e1000_phy_igp_3)
3495                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
3496
3497         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
3498          * would have already happened in close and is redundant. */
3499         e1000_release_hw_control(adapter);
3500
3501         pci_disable_device(pdev);
3502
3503         pci_set_power_state(pdev, pci_choose_state(pdev, state));
3504
3505         return 0;
3506 }
3507
3508 #ifdef CONFIG_PM
3509 static int e1000_resume(struct pci_dev *pdev)
3510 {
3511         struct net_device *netdev = pci_get_drvdata(pdev);
3512         struct e1000_adapter *adapter = netdev_priv(netdev);
3513         struct e1000_hw *hw = &adapter->hw;
3514         u32 err;
3515
3516         pci_set_power_state(pdev, PCI_D0);
3517         pci_restore_state(pdev);
3518         err = pci_enable_device(pdev);
3519         if (err) {
3520                 dev_err(&pdev->dev,
3521                         "Cannot enable PCI device from suspend\n");
3522                 return err;
3523         }
3524
3525         pci_set_master(pdev);
3526
3527         pci_enable_wake(pdev, PCI_D3hot, 0);
3528         pci_enable_wake(pdev, PCI_D3cold, 0);
3529
3530         if (netif_running(netdev)) {
3531                 err = e1000_request_irq(adapter);
3532                 if (err)
3533                         return err;
3534         }
3535
3536         e1000e_power_up_phy(adapter);
3537         e1000e_reset(adapter);
3538         ew32(WUS, ~0);
3539
3540         e1000_init_manageability(adapter);
3541
3542         if (netif_running(netdev))
3543                 e1000e_up(adapter);
3544
3545         netif_device_attach(netdev);
3546
3547         /* If the controller has AMT, do not set DRV_LOAD until the interface
3548          * is up.  For all other cases, let the f/w know that the h/w is now
3549          * under the control of the driver. */
3550         if (!(adapter->flags & FLAG_HAS_AMT) || !e1000e_check_mng_mode(&adapter->hw))
3551                 e1000_get_hw_control(adapter);
3552
3553         return 0;
3554 }
3555 #endif
3556
3557 static void e1000_shutdown(struct pci_dev *pdev)
3558 {
3559         e1000_suspend(pdev, PMSG_SUSPEND);
3560 }
3561
3562 #ifdef CONFIG_NET_POLL_CONTROLLER
3563 /*
3564  * Polling 'interrupt' - used by things like netconsole to send skbs
3565  * without having to re-enable interrupts. It's not called while
3566  * the interrupt routine is executing.
3567  */
3568 static void e1000_netpoll(struct net_device *netdev)
3569 {
3570         struct e1000_adapter *adapter = netdev_priv(netdev);
3571
3572         disable_irq(adapter->pdev->irq);
3573         e1000_intr(adapter->pdev->irq, netdev);
3574
3575         e1000_clean_tx_irq(adapter);
3576
3577         enable_irq(adapter->pdev->irq);
3578 }
3579 #endif
3580
3581 /**
3582  * e1000_io_error_detected - called when PCI error is detected
3583  * @pdev: Pointer to PCI device
3584  * @state: The current pci connection state
3585  *
3586  * This function is called after a PCI bus error affecting
3587  * this device has been detected.
3588  */
3589 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
3590                                                 pci_channel_state_t state)
3591 {
3592         struct net_device *netdev = pci_get_drvdata(pdev);
3593         struct e1000_adapter *adapter = netdev_priv(netdev);
3594
3595         netif_device_detach(netdev);
3596
3597         if (netif_running(netdev))
3598                 e1000e_down(adapter);
3599         pci_disable_device(pdev);
3600
3601         /* Request a slot slot reset. */
3602         return PCI_ERS_RESULT_NEED_RESET;
3603 }
3604
3605 /**
3606  * e1000_io_slot_reset - called after the pci bus has been reset.
3607  * @pdev: Pointer to PCI device
3608  *
3609  * Restart the card from scratch, as if from a cold-boot. Implementation
3610  * resembles the first-half of the e1000_resume routine.
3611  */
3612 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
3613 {
3614         struct net_device *netdev = pci_get_drvdata(pdev);
3615         struct e1000_adapter *adapter = netdev_priv(netdev);
3616         struct e1000_hw *hw = &adapter->hw;
3617
3618         if (pci_enable_device(pdev)) {
3619                 dev_err(&pdev->dev,
3620                         "Cannot re-enable PCI device after reset.\n");
3621                 return PCI_ERS_RESULT_DISCONNECT;
3622         }
3623         pci_set_master(pdev);
3624
3625         pci_enable_wake(pdev, PCI_D3hot, 0);
3626         pci_enable_wake(pdev, PCI_D3cold, 0);
3627
3628         e1000e_reset(adapter);
3629         ew32(WUS, ~0);
3630
3631         return PCI_ERS_RESULT_RECOVERED;
3632 }
3633
3634 /**
3635  * e1000_io_resume - called when traffic can start flowing again.
3636  * @pdev: Pointer to PCI device
3637  *
3638  * This callback is called when the error recovery driver tells us that
3639  * its OK to resume normal operation. Implementation resembles the
3640  * second-half of the e1000_resume routine.
3641  */
3642 static void e1000_io_resume(struct pci_dev *pdev)
3643 {
3644         struct net_device *netdev = pci_get_drvdata(pdev);
3645         struct e1000_adapter *adapter = netdev_priv(netdev);
3646
3647         e1000_init_manageability(adapter);
3648
3649         if (netif_running(netdev)) {
3650                 if (e1000e_up(adapter)) {
3651                         dev_err(&pdev->dev,
3652                                 "can't bring device back up after reset\n");
3653                         return;
3654                 }
3655         }
3656
3657         netif_device_attach(netdev);
3658
3659         /* If the controller has AMT, do not set DRV_LOAD until the interface
3660          * is up.  For all other cases, let the f/w know that the h/w is now
3661          * under the control of the driver. */
3662         if (!(adapter->flags & FLAG_HAS_AMT) ||
3663             !e1000e_check_mng_mode(&adapter->hw))
3664                 e1000_get_hw_control(adapter);
3665
3666 }
3667
3668 static void e1000_print_device_info(struct e1000_adapter *adapter)
3669 {
3670         struct e1000_hw *hw = &adapter->hw;
3671         struct net_device *netdev = adapter->netdev;
3672         u32 part_num;
3673
3674         /* print bus type/speed/width info */
3675         ndev_info(netdev, "(PCI Express:2.5GB/s:%s) "
3676                   "%02x:%02x:%02x:%02x:%02x:%02x\n",
3677                   /* bus width */
3678                  ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
3679                   "Width x1"),
3680                   /* MAC address */
3681                   netdev->dev_addr[0], netdev->dev_addr[1],
3682                   netdev->dev_addr[2], netdev->dev_addr[3],
3683                   netdev->dev_addr[4], netdev->dev_addr[5]);
3684         ndev_info(netdev, "Intel(R) PRO/%s Network Connection\n",
3685                   (hw->phy.type == e1000_phy_ife)
3686                    ? "10/100" : "1000");
3687         e1000e_read_part_num(hw, &part_num);
3688         ndev_info(netdev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
3689                   hw->mac.type, hw->phy.type,
3690                   (part_num >> 8), (part_num & 0xff));
3691 }
3692
3693 /**
3694  * e1000_probe - Device Initialization Routine
3695  * @pdev: PCI device information struct
3696  * @ent: entry in e1000_pci_tbl
3697  *
3698  * Returns 0 on success, negative on failure
3699  *
3700  * e1000_probe initializes an adapter identified by a pci_dev structure.
3701  * The OS initialization, configuring of the adapter private structure,
3702  * and a hardware reset occur.
3703  **/
3704 static int __devinit e1000_probe(struct pci_dev *pdev,
3705                                  const struct pci_device_id *ent)
3706 {
3707         struct net_device *netdev;
3708         struct e1000_adapter *adapter;
3709         struct e1000_hw *hw;
3710         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
3711         unsigned long mmio_start, mmio_len;
3712         unsigned long flash_start, flash_len;
3713
3714         static int cards_found;
3715         int i, err, pci_using_dac;
3716         u16 eeprom_data = 0;
3717         u16 eeprom_apme_mask = E1000_EEPROM_APME;
3718
3719         err = pci_enable_device(pdev);
3720         if (err)
3721                 return err;
3722
3723         pci_using_dac = 0;
3724         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
3725         if (!err) {
3726                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
3727                 if (!err)
3728                         pci_using_dac = 1;
3729         } else {
3730                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
3731                 if (err) {
3732                         err = pci_set_consistent_dma_mask(pdev,
3733                                                           DMA_32BIT_MASK);
3734                         if (err) {
3735                                 dev_err(&pdev->dev, "No usable DMA "
3736                                         "configuration, aborting\n");
3737                                 goto err_dma;
3738                         }
3739                 }
3740         }
3741
3742         err = pci_request_regions(pdev, e1000e_driver_name);
3743         if (err)
3744                 goto err_pci_reg;
3745
3746         pci_set_master(pdev);
3747
3748         err = -ENOMEM;
3749         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
3750         if (!netdev)
3751                 goto err_alloc_etherdev;
3752
3753         SET_NETDEV_DEV(netdev, &pdev->dev);
3754
3755         pci_set_drvdata(pdev, netdev);
3756         adapter = netdev_priv(netdev);
3757         hw = &adapter->hw;
3758         adapter->netdev = netdev;
3759         adapter->pdev = pdev;
3760         adapter->ei = ei;
3761         adapter->pba = ei->pba;
3762         adapter->flags = ei->flags;
3763         adapter->hw.adapter = adapter;
3764         adapter->hw.mac.type = ei->mac;
3765         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
3766
3767         mmio_start = pci_resource_start(pdev, 0);
3768         mmio_len = pci_resource_len(pdev, 0);
3769
3770         err = -EIO;
3771         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
3772         if (!adapter->hw.hw_addr)
3773                 goto err_ioremap;
3774
3775         if ((adapter->flags & FLAG_HAS_FLASH) &&
3776             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
3777                 flash_start = pci_resource_start(pdev, 1);
3778                 flash_len = pci_resource_len(pdev, 1);
3779                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
3780                 if (!adapter->hw.flash_address)
3781                         goto err_flashmap;
3782         }
3783
3784         /* construct the net_device struct */
3785         netdev->open                    = &e1000_open;
3786         netdev->stop                    = &e1000_close;
3787         netdev->hard_start_xmit         = &e1000_xmit_frame;
3788         netdev->get_stats               = &e1000_get_stats;
3789         netdev->set_multicast_list      = &e1000_set_multi;
3790         netdev->set_mac_address         = &e1000_set_mac;
3791         netdev->change_mtu              = &e1000_change_mtu;
3792         netdev->do_ioctl                = &e1000_ioctl;
3793         e1000e_set_ethtool_ops(netdev);
3794         netdev->tx_timeout              = &e1000_tx_timeout;
3795         netdev->watchdog_timeo          = 5 * HZ;
3796         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
3797         netdev->vlan_rx_register        = e1000_vlan_rx_register;
3798         netdev->vlan_rx_add_vid         = e1000_vlan_rx_add_vid;
3799         netdev->vlan_rx_kill_vid        = e1000_vlan_rx_kill_vid;
3800 #ifdef CONFIG_NET_POLL_CONTROLLER
3801         netdev->poll_controller         = e1000_netpoll;
3802 #endif
3803         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
3804
3805         netdev->mem_start = mmio_start;
3806         netdev->mem_end = mmio_start + mmio_len;
3807
3808         adapter->bd_number = cards_found++;
3809
3810         /* setup adapter struct */
3811         err = e1000_sw_init(adapter);
3812         if (err)
3813                 goto err_sw_init;
3814
3815         err = -EIO;
3816
3817         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
3818         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
3819         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
3820
3821         err = ei->get_invariants(adapter);
3822         if (err)
3823                 goto err_hw_init;
3824
3825         hw->mac.ops.get_bus_info(&adapter->hw);
3826
3827         adapter->hw.phy.wait_for_link = 0;
3828
3829         /* Copper options */
3830         if (adapter->hw.media_type == e1000_media_type_copper) {
3831                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
3832                 adapter->hw.phy.disable_polarity_correction = 0;
3833                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
3834         }
3835
3836         if (e1000_check_reset_block(&adapter->hw))
3837                 ndev_info(netdev,
3838                           "PHY reset is blocked due to SOL/IDER session.\n");
3839
3840         netdev->features = NETIF_F_SG |
3841                            NETIF_F_HW_CSUM |
3842                            NETIF_F_HW_VLAN_TX |
3843                            NETIF_F_HW_VLAN_RX;
3844
3845         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
3846                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
3847
3848         netdev->features |= NETIF_F_TSO;
3849         netdev->features |= NETIF_F_TSO6;
3850
3851         if (pci_using_dac)
3852                 netdev->features |= NETIF_F_HIGHDMA;
3853
3854         /* We should not be using LLTX anymore, but we are still TX faster with
3855          * it. */
3856         netdev->features |= NETIF_F_LLTX;
3857
3858         if (e1000e_enable_mng_pass_thru(&adapter->hw))
3859                 adapter->flags |= FLAG_MNG_PT_ENABLED;
3860
3861         /* before reading the NVM, reset the controller to
3862          * put the device in a known good starting state */
3863         adapter->hw.mac.ops.reset_hw(&adapter->hw);
3864
3865         /*
3866          * systems with ASPM and others may see the checksum fail on the first
3867          * attempt. Let's give it a few tries
3868          */
3869         for (i = 0;; i++) {
3870                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
3871                         break;
3872                 if (i == 2) {
3873                         ndev_err(netdev, "The NVM Checksum Is Not Valid\n");
3874                         err = -EIO;
3875                         goto err_eeprom;
3876                 }
3877         }
3878
3879         /* copy the MAC address out of the NVM */
3880         if (e1000e_read_mac_addr(&adapter->hw))
3881                 ndev_err(netdev, "NVM Read Error while reading MAC address\n");
3882
3883         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
3884         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
3885
3886         if (!is_valid_ether_addr(netdev->perm_addr)) {
3887                 ndev_err(netdev, "Invalid MAC Address: "
3888                          "%02x:%02x:%02x:%02x:%02x:%02x\n",
3889                          netdev->perm_addr[0], netdev->perm_addr[1],
3890                          netdev->perm_addr[2], netdev->perm_addr[3],
3891                          netdev->perm_addr[4], netdev->perm_addr[5]);
3892                 err = -EIO;
3893                 goto err_eeprom;
3894         }
3895
3896         init_timer(&adapter->watchdog_timer);
3897         adapter->watchdog_timer.function = &e1000_watchdog;
3898         adapter->watchdog_timer.data = (unsigned long) adapter;
3899
3900         init_timer(&adapter->phy_info_timer);
3901         adapter->phy_info_timer.function = &e1000_update_phy_info;
3902         adapter->phy_info_timer.data = (unsigned long) adapter;
3903
3904         INIT_WORK(&adapter->reset_task, e1000_reset_task);
3905         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
3906
3907         e1000e_check_options(adapter);
3908
3909         /* Initialize link parameters. User can change them with ethtool */
3910         adapter->hw.mac.autoneg = 1;
3911         adapter->fc_autoneg = 1;
3912         adapter->hw.mac.original_fc = e1000_fc_default;
3913         adapter->hw.mac.fc = e1000_fc_default;
3914         adapter->hw.phy.autoneg_advertised = 0x2f;
3915
3916         /* ring size defaults */
3917         adapter->rx_ring->count = 256;
3918         adapter->tx_ring->count = 256;
3919
3920         /*
3921          * Initial Wake on LAN setting - If APM wake is enabled in
3922          * the EEPROM, enable the ACPI Magic Packet filter
3923          */
3924         if (adapter->flags & FLAG_APME_IN_WUC) {
3925                 /* APME bit in EEPROM is mapped to WUC.APME */
3926                 eeprom_data = er32(WUC);
3927                 eeprom_apme_mask = E1000_WUC_APME;
3928         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
3929                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
3930                     (adapter->hw.bus.func == 1))
3931                         e1000_read_nvm(&adapter->hw,
3932                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
3933                 else
3934                         e1000_read_nvm(&adapter->hw,
3935                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
3936         }
3937
3938         /* fetch WoL from EEPROM */
3939         if (eeprom_data & eeprom_apme_mask)
3940                 adapter->eeprom_wol |= E1000_WUFC_MAG;
3941
3942         /*
3943          * now that we have the eeprom settings, apply the special cases
3944          * where the eeprom may be wrong or the board simply won't support
3945          * wake on lan on a particular port
3946          */
3947         if (!(adapter->flags & FLAG_HAS_WOL))
3948                 adapter->eeprom_wol = 0;
3949
3950         /* initialize the wol settings based on the eeprom settings */
3951         adapter->wol = adapter->eeprom_wol;
3952
3953         /* reset the hardware with the new settings */
3954         e1000e_reset(adapter);
3955
3956         /* If the controller has AMT, do not set DRV_LOAD until the interface
3957          * is up.  For all other cases, let the f/w know that the h/w is now
3958          * under the control of the driver. */
3959         if (!(adapter->flags & FLAG_HAS_AMT) ||
3960             !e1000e_check_mng_mode(&adapter->hw))
3961                 e1000_get_hw_control(adapter);
3962
3963         /* tell the stack to leave us alone until e1000_open() is called */
3964         netif_carrier_off(netdev);
3965         netif_stop_queue(netdev);
3966
3967         strcpy(netdev->name, "eth%d");
3968         err = register_netdev(netdev);
3969         if (err)
3970                 goto err_register;
3971
3972         e1000_print_device_info(adapter);
3973
3974         return 0;
3975
3976 err_register:
3977 err_hw_init:
3978         e1000_release_hw_control(adapter);
3979 err_eeprom:
3980         if (!e1000_check_reset_block(&adapter->hw))
3981                 e1000_phy_hw_reset(&adapter->hw);
3982
3983         if (adapter->hw.flash_address)
3984                 iounmap(adapter->hw.flash_address);
3985
3986 err_flashmap:
3987         kfree(adapter->tx_ring);
3988         kfree(adapter->rx_ring);
3989 err_sw_init:
3990         iounmap(adapter->hw.hw_addr);
3991 err_ioremap:
3992         free_netdev(netdev);
3993 err_alloc_etherdev:
3994         pci_release_regions(pdev);
3995 err_pci_reg:
3996 err_dma:
3997         pci_disable_device(pdev);
3998         return err;
3999 }
4000
4001 /**
4002  * e1000_remove - Device Removal Routine
4003  * @pdev: PCI device information struct
4004  *
4005  * e1000_remove is called by the PCI subsystem to alert the driver
4006  * that it should release a PCI device.  The could be caused by a
4007  * Hot-Plug event, or because the driver is going to be removed from
4008  * memory.
4009  **/
4010 static void __devexit e1000_remove(struct pci_dev *pdev)
4011 {
4012         struct net_device *netdev = pci_get_drvdata(pdev);
4013         struct e1000_adapter *adapter = netdev_priv(netdev);
4014
4015         /* flush_scheduled work may reschedule our watchdog task, so
4016          * explicitly disable watchdog tasks from being rescheduled  */
4017         set_bit(__E1000_DOWN, &adapter->state);
4018         del_timer_sync(&adapter->watchdog_timer);
4019         del_timer_sync(&adapter->phy_info_timer);
4020
4021         flush_scheduled_work();
4022
4023         e1000_release_manageability(adapter);
4024
4025         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
4026          * would have already happened in close and is redundant. */
4027         e1000_release_hw_control(adapter);
4028
4029         unregister_netdev(netdev);
4030
4031         if (!e1000_check_reset_block(&adapter->hw))
4032                 e1000_phy_hw_reset(&adapter->hw);
4033
4034         kfree(adapter->tx_ring);
4035         kfree(adapter->rx_ring);
4036
4037         iounmap(adapter->hw.hw_addr);
4038         if (adapter->hw.flash_address)
4039                 iounmap(adapter->hw.flash_address);
4040         pci_release_regions(pdev);
4041
4042         free_netdev(netdev);
4043
4044         pci_disable_device(pdev);
4045 }
4046
4047 /* PCI Error Recovery (ERS) */
4048 static struct pci_error_handlers e1000_err_handler = {
4049         .error_detected = e1000_io_error_detected,
4050         .slot_reset = e1000_io_slot_reset,
4051         .resume = e1000_io_resume,
4052 };
4053
4054 static struct pci_device_id e1000_pci_tbl[] = {
4055         /*
4056          * Support for 82571/2/3, es2lan and ich8 will be phased in
4057          * stepwise.
4058
4059         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
4060         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
4061         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
4062         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
4063         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
4064         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
4065         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
4066         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
4067         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
4068         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
4069         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
4070         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
4071         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
4072         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
4073           board_80003es2lan },
4074         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
4075           board_80003es2lan },
4076         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
4077           board_80003es2lan },
4078         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
4079           board_80003es2lan },
4080         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
4081         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
4082         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
4083         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
4084         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
4085         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
4086         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
4087         */
4088
4089         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
4090         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
4091         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
4092         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
4093         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
4094
4095         { }     /* terminate list */
4096 };
4097 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
4098
4099 /* PCI Device API Driver */
4100 static struct pci_driver e1000_driver = {
4101         .name     = e1000e_driver_name,
4102         .id_table = e1000_pci_tbl,
4103         .probe    = e1000_probe,
4104         .remove   = __devexit_p(e1000_remove),
4105 #ifdef CONFIG_PM
4106         /* Power Managment Hooks */
4107         .suspend  = e1000_suspend,
4108         .resume   = e1000_resume,
4109 #endif
4110         .shutdown = e1000_shutdown,
4111         .err_handler = &e1000_err_handler
4112 };
4113
4114 /**
4115  * e1000_init_module - Driver Registration Routine
4116  *
4117  * e1000_init_module is the first routine called when the driver is
4118  * loaded. All it does is register with the PCI subsystem.
4119  **/
4120 static int __init e1000_init_module(void)
4121 {
4122         int ret;
4123         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
4124                e1000e_driver_name, e1000e_driver_version);
4125         printk(KERN_INFO "%s: Copyright (c) 1999-2007 Intel Corporation.\n",
4126                e1000e_driver_name);
4127         ret = pci_register_driver(&e1000_driver);
4128
4129         return ret;
4130 }
4131 module_init(e1000_init_module);
4132
4133 /**
4134  * e1000_exit_module - Driver Exit Cleanup Routine
4135  *
4136  * e1000_exit_module is called just before the driver is removed
4137  * from memory.
4138  **/
4139 static void __exit e1000_exit_module(void)
4140 {
4141         pci_unregister_driver(&e1000_driver);
4142 }
4143 module_exit(e1000_exit_module);
4144
4145
4146 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
4147 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
4148 MODULE_LICENSE("GPL");
4149 MODULE_VERSION(DRV_VERSION);
4150
4151 /* e1000_main.c */