]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/net/e1000e/netdev.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
gianfar iomem misannotations
[net-next-2.6.git] / drivers / net / e1000e / netdev.c
CommitLineData
bc7f75fa
AK		 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"
50char e1000e_driver_name[] = "e1000e";
51const char e1000e_driver_version[] = DRV_VERSION;
52
53static 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 **/
67char *e1000e_get_hw_dev_name(struct e1000_hw *hw)
68{
589c085f 69 return hw->adapter->netdev->name;
bc7f75fa
AK		 70}
71#endif
72
73/**
74 * e1000_desc_unused - calculate if we have unused descriptors
75 **/
76static 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 **/
91static void e1000_receive_skb(struct e1000_adapter *adapter,
92 struct net_device *netdev,
93 struct sk_buff *skb,
a39fe742 94 u8 status, __le16 vlan)
bc7f75fa
AK		 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 **/
115static 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 */
a39fe742
AV		 145 __sum16 sum = (__force __sum16)htons(csum);
146 skb->csum = csum_unfold(~sum);
bc7f75fa
AK		 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 **/
156static 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;
192map_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 **/
229static 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++) {
47f44e40
AK		 248 ps_page = &buffer_info->ps_pages[j];
249 if (j >= adapter->rx_ps_pages) {
250 /* all unused desc entries get hw null ptr */
a39fe742 251 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
47f44e40
AK		 252 continue;
253 }
254 if (!ps_page->page) {
255 ps_page->page = alloc_page(GFP_ATOMIC);
bc7f75fa 256 if (!ps_page->page) {
47f44e40
AK		 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;
bc7f75fa 269 }
bc7f75fa 270 }
47f44e40
AK		 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);
bc7f75fa
AK		 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
315no_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
bc7f75fa
AK		 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 **/
342static 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
bc7f75fa
AK		 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
440next_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;
41988692
AK		 461 adapter->net_stats.rx_packets += total_rx_packets;
462 adapter->net_stats.rx_bytes += total_rx_bytes;
bc7f75fa
AK		 463 return cleaned;
464}
465
bc7f75fa
AK		 466static void e1000_put_txbuf(struct e1000_adapter *adapter,
467 struct e1000_buffer *buffer_info)
468{
469 if (buffer_info->dma) {
470 pci_unmap_page(adapter->pdev, buffer_info->dma,
471 buffer_info->length, PCI_DMA_TODEVICE);
472 buffer_info->dma = 0;
473 }
474 if (buffer_info->skb) {
475 dev_kfree_skb_any(buffer_info->skb);
476 buffer_info->skb = NULL;
477 }
478}
479
480static void e1000_print_tx_hang(struct e1000_adapter *adapter)
481{
482 struct e1000_ring *tx_ring = adapter->tx_ring;
483 unsigned int i = tx_ring->next_to_clean;
484 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
485 struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
486 struct net_device *netdev = adapter->netdev;
487
488 /* detected Tx unit hang */
489 ndev_err(netdev,
490 "Detected Tx Unit Hang:\n"
491 " TDH <%x>\n"
492 " TDT <%x>\n"
493 " next_to_use <%x>\n"
494 " next_to_clean <%x>\n"
495 "buffer_info[next_to_clean]:\n"
496 " time_stamp <%lx>\n"
497 " next_to_watch <%x>\n"
498 " jiffies <%lx>\n"
499 " next_to_watch.status <%x>\n",
500 readl(adapter->hw.hw_addr + tx_ring->head),
501 readl(adapter->hw.hw_addr + tx_ring->tail),
502 tx_ring->next_to_use,
503 tx_ring->next_to_clean,
504 tx_ring->buffer_info[eop].time_stamp,
505 eop,
506 jiffies,
507 eop_desc->upper.fields.status);
508}
509
510/**
511 * e1000_clean_tx_irq - Reclaim resources after transmit completes
512 * @adapter: board private structure
513 *
514 * the return value indicates whether actual cleaning was done, there
515 * is no guarantee that everything was cleaned
516 **/
517static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
518{
519 struct net_device *netdev = adapter->netdev;
520 struct e1000_hw *hw = &adapter->hw;
521 struct e1000_ring *tx_ring = adapter->tx_ring;
522 struct e1000_tx_desc *tx_desc, *eop_desc;
523 struct e1000_buffer *buffer_info;
524 unsigned int i, eop;
525 unsigned int count = 0;
526 bool cleaned = 0;
527 unsigned int total_tx_bytes = 0, total_tx_packets = 0;
528
529 i = tx_ring->next_to_clean;
530 eop = tx_ring->buffer_info[i].next_to_watch;
531 eop_desc = E1000_TX_DESC(*tx_ring, eop);
532
533 while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
534 for (cleaned = 0; !cleaned; ) {
535 tx_desc = E1000_TX_DESC(*tx_ring, i);
536 buffer_info = &tx_ring->buffer_info[i];
537 cleaned = (i == eop);
538
539 if (cleaned) {
540 struct sk_buff *skb = buffer_info->skb;
541 unsigned int segs, bytecount;
542 segs = skb_shinfo(skb)->gso_segs ?: 1;
543 /* multiply data chunks by size of headers */
544 bytecount = ((segs - 1) * skb_headlen(skb)) +
545 skb->len;
546 total_tx_packets += segs;
547 total_tx_bytes += bytecount;
548 }
549
550 e1000_put_txbuf(adapter, buffer_info);
551 tx_desc->upper.data = 0;
552
553 i++;
554 if (i == tx_ring->count)
555 i = 0;
556 }
557
558 eop = tx_ring->buffer_info[i].next_to_watch;
559 eop_desc = E1000_TX_DESC(*tx_ring, eop);
560#define E1000_TX_WEIGHT 64
561 /* weight of a sort for tx, to avoid endless transmit cleanup */
562 if (count++ == E1000_TX_WEIGHT)
563 break;
564 }
565
566 tx_ring->next_to_clean = i;
567
568#define TX_WAKE_THRESHOLD 32
569 if (cleaned && netif_carrier_ok(netdev) &&
570 e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
571 /* Make sure that anybody stopping the queue after this
572 * sees the new next_to_clean.
573 */
574 smp_mb();
575
576 if (netif_queue_stopped(netdev) &&
577 !(test_bit(__E1000_DOWN, &adapter->state))) {
578 netif_wake_queue(netdev);
579 ++adapter->restart_queue;
580 }
581 }
582
583 if (adapter->detect_tx_hung) {
584 /* Detect a transmit hang in hardware, this serializes the
585 * check with the clearing of time_stamp and movement of i */
586 adapter->detect_tx_hung = 0;
587 if (tx_ring->buffer_info[eop].dma &&
588 time_after(jiffies, tx_ring->buffer_info[eop].time_stamp
589 + (adapter->tx_timeout_factor * HZ))
590 && !(er32(STATUS) &
591 E1000_STATUS_TXOFF)) {
592 e1000_print_tx_hang(adapter);
593 netif_stop_queue(netdev);
594 }
595 }
596 adapter->total_tx_bytes += total_tx_bytes;
597 adapter->total_tx_packets += total_tx_packets;
41988692
AK		 598 adapter->net_stats.tx_packets += total_tx_packets;
599 adapter->net_stats.tx_bytes += total_tx_bytes;
bc7f75fa
AK		 600 return cleaned;
601}
602
bc7f75fa
AK		 603/**
604 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
605 * @adapter: board private structure
606 *
607 * the return value indicates whether actual cleaning was done, there
608 * is no guarantee that everything was cleaned
609 **/
610static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
611 int *work_done, int work_to_do)
612{
613 union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
614 struct net_device *netdev = adapter->netdev;
615 struct pci_dev *pdev = adapter->pdev;
616 struct e1000_ring *rx_ring = adapter->rx_ring;
617 struct e1000_buffer *buffer_info, *next_buffer;
618 struct e1000_ps_page *ps_page;
619 struct sk_buff *skb;
620 unsigned int i, j;
621 u32 length, staterr;
622 int cleaned_count = 0;
623 bool cleaned = 0;
624 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
625
626 i = rx_ring->next_to_clean;
627 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
628 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
629 buffer_info = &rx_ring->buffer_info[i];
630
631 while (staterr & E1000_RXD_STAT_DD) {
632 if (*work_done >= work_to_do)
633 break;
634 (*work_done)++;
635 skb = buffer_info->skb;
636
637 /* in the packet split case this is header only */
638 prefetch(skb->data - NET_IP_ALIGN);
639
640 i++;
641 if (i == rx_ring->count)
642 i = 0;
643 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
644 prefetch(next_rxd);
645
646 next_buffer = &rx_ring->buffer_info[i];
647
648 cleaned = 1;
649 cleaned_count++;
650 pci_unmap_single(pdev, buffer_info->dma,
651 adapter->rx_ps_bsize0,
652 PCI_DMA_FROMDEVICE);
653 buffer_info->dma = 0;
654
655 if (!(staterr & E1000_RXD_STAT_EOP)) {
656 ndev_dbg(netdev, "%s: Packet Split buffers didn't pick "
657 "up the full packet\n", netdev->name);
658 dev_kfree_skb_irq(skb);
659 goto next_desc;
660 }
661
662 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
663 dev_kfree_skb_irq(skb);
664 goto next_desc;
665 }
666
667 length = le16_to_cpu(rx_desc->wb.middle.length0);
668
669 if (!length) {
670 ndev_dbg(netdev, "%s: Last part of the packet spanning"
671 " multiple descriptors\n", netdev->name);
672 dev_kfree_skb_irq(skb);
673 goto next_desc;
674 }
675
676 /* Good Receive */
677 skb_put(skb, length);
678
679 {
680 /* this looks ugly, but it seems compiler issues make it
681 more efficient than reusing j */
682 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
683
684 /* page alloc/put takes too long and effects small packet
685 * throughput, so unsplit small packets and save the alloc/put*/
686 if (l1 && (l1 <= copybreak) &&
687 ((length + l1) <= adapter->rx_ps_bsize0)) {
688 u8 *vaddr;
689
47f44e40 690 ps_page = &buffer_info->ps_pages[0];
bc7f75fa
AK		 691
692 /* there is no documentation about how to call
693 * kmap_atomic, so we can't hold the mapping
694 * very long */
695 pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
696 PAGE_SIZE, PCI_DMA_FROMDEVICE);
697 vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
698 memcpy(skb_tail_pointer(skb), vaddr, l1);
699 kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
700 pci_dma_sync_single_for_device(pdev, ps_page->dma,
701 PAGE_SIZE, PCI_DMA_FROMDEVICE);
140a7480 702
bc7f75fa
AK		 703 skb_put(skb, l1);
704 goto copydone;
705 } /* if */
706 }
707
708 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
709 length = le16_to_cpu(rx_desc->wb.upper.length[j]);
710 if (!length)
711 break;
712
47f44e40 713 ps_page = &buffer_info->ps_pages[j];
bc7f75fa
AK		 714 pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
715 PCI_DMA_FROMDEVICE);
716 ps_page->dma = 0;
717 skb_fill_page_desc(skb, j, ps_page->page, 0, length);
718 ps_page->page = NULL;
719 skb->len += length;
720 skb->data_len += length;
721 skb->truesize += length;
722 }
723
bc7f75fa
AK		 724copydone:
725 total_rx_bytes += skb->len;
726 total_rx_packets++;
727
728 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
729 rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
730
731 if (rx_desc->wb.upper.header_status &
732 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
733 adapter->rx_hdr_split++;
734
735 e1000_receive_skb(adapter, netdev, skb,
736 staterr, rx_desc->wb.middle.vlan);
737
738next_desc:
739 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
740 buffer_info->skb = NULL;
741
742 /* return some buffers to hardware, one at a time is too slow */
743 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
744 adapter->alloc_rx_buf(adapter, cleaned_count);
745 cleaned_count = 0;
746 }
747
748 /* use prefetched values */
749 rx_desc = next_rxd;
750 buffer_info = next_buffer;
751
752 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
753 }
754 rx_ring->next_to_clean = i;
755
756 cleaned_count = e1000_desc_unused(rx_ring);
757 if (cleaned_count)
758 adapter->alloc_rx_buf(adapter, cleaned_count);
759
760 adapter->total_rx_packets += total_rx_packets;
761 adapter->total_rx_bytes += total_rx_bytes;
41988692
AK		 762 adapter->net_stats.rx_packets += total_rx_packets;
763 adapter->net_stats.rx_bytes += total_rx_bytes;
bc7f75fa
AK		 764 return cleaned;
765}
766
767/**
768 * e1000_clean_rx_ring - Free Rx Buffers per Queue
769 * @adapter: board private structure
770 **/
771static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
772{
773 struct e1000_ring *rx_ring = adapter->rx_ring;
774 struct e1000_buffer *buffer_info;
775 struct e1000_ps_page *ps_page;
776 struct pci_dev *pdev = adapter->pdev;
bc7f75fa
AK		 777 unsigned int i, j;
778
779 /* Free all the Rx ring sk_buffs */
780 for (i = 0; i < rx_ring->count; i++) {
781 buffer_info = &rx_ring->buffer_info[i];
782 if (buffer_info->dma) {
783 if (adapter->clean_rx == e1000_clean_rx_irq)
784 pci_unmap_single(pdev, buffer_info->dma,
785 adapter->rx_buffer_len,
786 PCI_DMA_FROMDEVICE);
bc7f75fa
AK		 787 else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
788 pci_unmap_single(pdev, buffer_info->dma,
789 adapter->rx_ps_bsize0,
790 PCI_DMA_FROMDEVICE);
791 buffer_info->dma = 0;
792 }
793
bc7f75fa
AK		 794 if (buffer_info->skb) {
795 dev_kfree_skb(buffer_info->skb);
796 buffer_info->skb = NULL;
797 }
798
799 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
47f44e40 800 ps_page = &buffer_info->ps_pages[j];
bc7f75fa
AK		 801 if (!ps_page->page)
802 break;
803 pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
804 PCI_DMA_FROMDEVICE);
805 ps_page->dma = 0;
806 put_page(ps_page->page);
807 ps_page->page = NULL;
808 }
809 }
810
811 /* there also may be some cached data from a chained receive */
812 if (rx_ring->rx_skb_top) {
813 dev_kfree_skb(rx_ring->rx_skb_top);
814 rx_ring->rx_skb_top = NULL;
815 }
816
bc7f75fa
AK		 817 /* Zero out the descriptor ring */
818 memset(rx_ring->desc, 0, rx_ring->size);
819
820 rx_ring->next_to_clean = 0;
821 rx_ring->next_to_use = 0;
822
823 writel(0, adapter->hw.hw_addr + rx_ring->head);
824 writel(0, adapter->hw.hw_addr + rx_ring->tail);
825}
826
827/**
828 * e1000_intr_msi - Interrupt Handler
829 * @irq: interrupt number
830 * @data: pointer to a network interface device structure
831 **/
832static irqreturn_t e1000_intr_msi(int irq, void *data)
833{
834 struct net_device *netdev = data;
835 struct e1000_adapter *adapter = netdev_priv(netdev);
836 struct e1000_hw *hw = &adapter->hw;
837 u32 icr = er32(ICR);
838
839 /* read ICR disables interrupts using IAM, so keep up with our
840 * enable/disable accounting */
841 atomic_inc(&adapter->irq_sem);
842
843 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
844 hw->mac.get_link_status = 1;
845 /* ICH8 workaround-- Call gig speed drop workaround on cable
846 * disconnect (LSC) before accessing any PHY registers */
847 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
848 (!(er32(STATUS) & E1000_STATUS_LU)))
849 e1000e_gig_downshift_workaround_ich8lan(hw);
850
851 /* 80003ES2LAN workaround-- For packet buffer work-around on
852 * link down event; disable receives here in the ISR and reset
853 * adapter in watchdog */
854 if (netif_carrier_ok(netdev) &&
855 adapter->flags & FLAG_RX_NEEDS_RESTART) {
856 /* disable receives */
857 u32 rctl = er32(RCTL);
858 ew32(RCTL, rctl & ~E1000_RCTL_EN);
859 }
860 /* guard against interrupt when we're going down */
861 if (!test_bit(__E1000_DOWN, &adapter->state))
862 mod_timer(&adapter->watchdog_timer, jiffies + 1);
863 }
864
865 if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
866 adapter->total_tx_bytes = 0;
867 adapter->total_tx_packets = 0;
868 adapter->total_rx_bytes = 0;
869 adapter->total_rx_packets = 0;
870 __netif_rx_schedule(netdev, &adapter->napi);
871 } else {
872 atomic_dec(&adapter->irq_sem);
873 }
874
875 return IRQ_HANDLED;
876}
877
878/**
879 * e1000_intr - Interrupt Handler
880 * @irq: interrupt number
881 * @data: pointer to a network interface device structure
882 **/
883static irqreturn_t e1000_intr(int irq, void *data)
884{
885 struct net_device *netdev = data;
886 struct e1000_adapter *adapter = netdev_priv(netdev);
887 struct e1000_hw *hw = &adapter->hw;
888
889 u32 rctl, icr = er32(ICR);
890 if (!icr)
891 return IRQ_NONE; /* Not our interrupt */
892
893 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
894 * not set, then the adapter didn't send an interrupt */
895 if (!(icr & E1000_ICR_INT_ASSERTED))
896 return IRQ_NONE;
897
898 /* Interrupt Auto-Mask...upon reading ICR,
899 * interrupts are masked. No need for the
900 * IMC write, but it does mean we should
901 * account for it ASAP. */
902 atomic_inc(&adapter->irq_sem);
903
904 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
905 hw->mac.get_link_status = 1;
906 /* ICH8 workaround-- Call gig speed drop workaround on cable
907 * disconnect (LSC) before accessing any PHY registers */
908 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
909 (!(er32(STATUS) & E1000_STATUS_LU)))
910 e1000e_gig_downshift_workaround_ich8lan(hw);
911
912 /* 80003ES2LAN workaround--
913 * For packet buffer work-around on link down event;
914 * disable receives here in the ISR and
915 * reset adapter in watchdog
916 */
917 if (netif_carrier_ok(netdev) &&
918 (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
919 /* disable receives */
920 rctl = er32(RCTL);
921 ew32(RCTL, rctl & ~E1000_RCTL_EN);
922 }
923 /* guard against interrupt when we're going down */
924 if (!test_bit(__E1000_DOWN, &adapter->state))
925 mod_timer(&adapter->watchdog_timer, jiffies + 1);
926 }
927
928 if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
929 adapter->total_tx_bytes = 0;
930 adapter->total_tx_packets = 0;
931 adapter->total_rx_bytes = 0;
932 adapter->total_rx_packets = 0;
933 __netif_rx_schedule(netdev, &adapter->napi);
934 } else {
935 atomic_dec(&adapter->irq_sem);
936 }
937
938 return IRQ_HANDLED;
939}
940
941static int e1000_request_irq(struct e1000_adapter *adapter)
942{
943 struct net_device *netdev = adapter->netdev;
a39fe742 944 irq_handler_t handler = e1000_intr;
bc7f75fa
AK		 945 int irq_flags = IRQF_SHARED;
946 int err;
947
9b71c5e0 948 if (!pci_enable_msi(adapter->pdev)) {
bc7f75fa 949 adapter->flags |= FLAG_MSI_ENABLED;
a39fe742 950 handler = e1000_intr_msi;
bc7f75fa
AK		 951 irq_flags = 0;
952 }
953
954 err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
955 netdev);
956 if (err) {
9b71c5e0
AG		 957 ndev_err(netdev,
958 "Unable to allocate %s interrupt (return: %d)\n",
959 adapter->flags & FLAG_MSI_ENABLED ? "MSI":"INTx",
960 err);
bc7f75fa
AK		 961 if (adapter->flags & FLAG_MSI_ENABLED)
962 pci_disable_msi(adapter->pdev);
bc7f75fa
AK		 963 }
964
965 return err;
966}
967
968static void e1000_free_irq(struct e1000_adapter *adapter)
969{
970 struct net_device *netdev = adapter->netdev;
971
972 free_irq(adapter->pdev->irq, netdev);
973 if (adapter->flags & FLAG_MSI_ENABLED) {
974 pci_disable_msi(adapter->pdev);
975 adapter->flags &= ~FLAG_MSI_ENABLED;
976 }
977}
978
979/**
980 * e1000_irq_disable - Mask off interrupt generation on the NIC
981 **/
982static void e1000_irq_disable(struct e1000_adapter *adapter)
983{
984 struct e1000_hw *hw = &adapter->hw;
985
986 atomic_inc(&adapter->irq_sem);
987 ew32(IMC, ~0);
988 e1e_flush();
989 synchronize_irq(adapter->pdev->irq);
990}
991
992/**
993 * e1000_irq_enable - Enable default interrupt generation settings
994 **/
995static void e1000_irq_enable(struct e1000_adapter *adapter)
996{
997 struct e1000_hw *hw = &adapter->hw;
998
999 if (atomic_dec_and_test(&adapter->irq_sem)) {
1000 ew32(IMS, IMS_ENABLE_MASK);
1001 e1e_flush();
1002 }
1003}
1004
1005/**
1006 * e1000_get_hw_control - get control of the h/w from f/w
1007 * @adapter: address of board private structure
1008 *
1009 * e1000_get_hw_control sets {CTRL_EXT|FWSM}:DRV_LOAD bit.
1010 * For ASF and Pass Through versions of f/w this means that
1011 * the driver is loaded. For AMT version (only with 82573)
1012 * of the f/w this means that the network i/f is open.
1013 **/
1014static void e1000_get_hw_control(struct e1000_adapter *adapter)
1015{
1016 struct e1000_hw *hw = &adapter->hw;
1017 u32 ctrl_ext;
1018 u32 swsm;
1019
1020 /* Let firmware know the driver has taken over */
1021 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1022 swsm = er32(SWSM);
1023 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1024 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1025 ctrl_ext = er32(CTRL_EXT);
1026 ew32(CTRL_EXT,
1027 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1028 }
1029}
1030
1031/**
1032 * e1000_release_hw_control - release control of the h/w to f/w
1033 * @adapter: address of board private structure
1034 *
1035 * e1000_release_hw_control resets {CTRL_EXT|FWSM}:DRV_LOAD bit.
1036 * For ASF and Pass Through versions of f/w this means that the
1037 * driver is no longer loaded. For AMT version (only with 82573) i
1038 * of the f/w this means that the network i/f is closed.
1039 *
1040 **/
1041static void e1000_release_hw_control(struct e1000_adapter *adapter)
1042{
1043 struct e1000_hw *hw = &adapter->hw;
1044 u32 ctrl_ext;
1045 u32 swsm;
1046
1047 /* Let firmware taken over control of h/w */
1048 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1049 swsm = er32(SWSM);
1050 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1051 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1052 ctrl_ext = er32(CTRL_EXT);
1053 ew32(CTRL_EXT,
1054 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1055 }
1056}
1057
1058static void e1000_release_manageability(struct e1000_adapter *adapter)
1059{
1060 if (adapter->flags & FLAG_MNG_PT_ENABLED) {
1061 struct e1000_hw *hw = &adapter->hw;
1062
1063 u32 manc = er32(MANC);
1064
1065 /* re-enable hardware interception of ARP */
1066 manc |= E1000_MANC_ARP_EN;
1067 manc &= ~E1000_MANC_EN_MNG2HOST;
1068
1069 /* don't explicitly have to mess with MANC2H since
1070 * MANC has an enable disable that gates MANC2H */
1071 ew32(MANC, manc);
1072 }
1073}
1074
1075/**
1076 * @e1000_alloc_ring - allocate memory for a ring structure
1077 **/
1078static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1079 struct e1000_ring *ring)
1080{
1081 struct pci_dev *pdev = adapter->pdev;
1082
1083 ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1084 GFP_KERNEL);
1085 if (!ring->desc)
1086 return -ENOMEM;
1087
1088 return 0;
1089}
1090
1091/**
1092 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1093 * @adapter: board private structure
1094 *
1095 * Return 0 on success, negative on failure
1096 **/
1097int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1098{
1099 struct e1000_ring *tx_ring = adapter->tx_ring;
1100 int err = -ENOMEM, size;
1101
1102 size = sizeof(struct e1000_buffer) * tx_ring->count;
1103 tx_ring->buffer_info = vmalloc(size);
1104 if (!tx_ring->buffer_info)
1105 goto err;
1106 memset(tx_ring->buffer_info, 0, size);
1107
1108 /* round up to nearest 4K */
1109 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1110 tx_ring->size = ALIGN(tx_ring->size, 4096);
1111
1112 err = e1000_alloc_ring_dma(adapter, tx_ring);
1113 if (err)
1114 goto err;
1115
1116 tx_ring->next_to_use = 0;
1117 tx_ring->next_to_clean = 0;
1118 spin_lock_init(&adapter->tx_queue_lock);
1119
1120 return 0;
1121err:
1122 vfree(tx_ring->buffer_info);
1123 ndev_err(adapter->netdev,
1124 "Unable to allocate memory for the transmit descriptor ring\n");
1125 return err;
1126}
1127
1128/**
1129 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1130 * @adapter: board private structure
1131 *
1132 * Returns 0 on success, negative on failure
1133 **/
1134int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1135{
1136 struct e1000_ring *rx_ring = adapter->rx_ring;
47f44e40
AK		 1137 struct e1000_buffer *buffer_info;
1138 int i, size, desc_len, err = -ENOMEM;
bc7f75fa
AK		 1139
1140 size = sizeof(struct e1000_buffer) * rx_ring->count;
1141 rx_ring->buffer_info = vmalloc(size);
1142 if (!rx_ring->buffer_info)
1143 goto err;
1144 memset(rx_ring->buffer_info, 0, size);
1145
47f44e40
AK		 1146 for (i = 0; i < rx_ring->count; i++) {
1147 buffer_info = &rx_ring->buffer_info[i];
1148 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1149 sizeof(struct e1000_ps_page),
1150 GFP_KERNEL);
1151 if (!buffer_info->ps_pages)
1152 goto err_pages;
1153 }
bc7f75fa
AK		 1154
1155 desc_len = sizeof(union e1000_rx_desc_packet_split);
1156
1157 /* Round up to nearest 4K */
1158 rx_ring->size = rx_ring->count * desc_len;
1159 rx_ring->size = ALIGN(rx_ring->size, 4096);
1160
1161 err = e1000_alloc_ring_dma(adapter, rx_ring);
1162 if (err)
47f44e40 1163 goto err_pages;
bc7f75fa
AK		 1164
1165 rx_ring->next_to_clean = 0;
1166 rx_ring->next_to_use = 0;
1167 rx_ring->rx_skb_top = NULL;
1168
1169 return 0;
47f44e40
AK		 1170
1171err_pages:
1172 for (i = 0; i < rx_ring->count; i++) {
1173 buffer_info = &rx_ring->buffer_info[i];
1174 kfree(buffer_info->ps_pages);
1175 }
bc7f75fa
AK		 1176err:
1177 vfree(rx_ring->buffer_info);
bc7f75fa
AK		 1178 ndev_err(adapter->netdev,
1179 "Unable to allocate memory for the transmit descriptor ring\n");
1180 return err;
1181}
1182
1183/**
1184 * e1000_clean_tx_ring - Free Tx Buffers
1185 * @adapter: board private structure
1186 **/
1187static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1188{
1189 struct e1000_ring *tx_ring = adapter->tx_ring;
1190 struct e1000_buffer *buffer_info;
1191 unsigned long size;
1192 unsigned int i;
1193
1194 for (i = 0; i < tx_ring->count; i++) {
1195 buffer_info = &tx_ring->buffer_info[i];
1196 e1000_put_txbuf(adapter, buffer_info);
1197 }
1198
1199 size = sizeof(struct e1000_buffer) * tx_ring->count;
1200 memset(tx_ring->buffer_info, 0, size);
1201
1202 memset(tx_ring->desc, 0, tx_ring->size);
1203
1204 tx_ring->next_to_use = 0;
1205 tx_ring->next_to_clean = 0;
1206
1207 writel(0, adapter->hw.hw_addr + tx_ring->head);
1208 writel(0, adapter->hw.hw_addr + tx_ring->tail);
1209}
1210
1211/**
1212 * e1000e_free_tx_resources - Free Tx Resources per Queue
1213 * @adapter: board private structure
1214 *
1215 * Free all transmit software resources
1216 **/
1217void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1218{
1219 struct pci_dev *pdev = adapter->pdev;
1220 struct e1000_ring *tx_ring = adapter->tx_ring;
1221
1222 e1000_clean_tx_ring(adapter);
1223
1224 vfree(tx_ring->buffer_info);
1225 tx_ring->buffer_info = NULL;
1226
1227 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1228 tx_ring->dma);
1229 tx_ring->desc = NULL;
1230}
1231
1232/**
1233 * e1000e_free_rx_resources - Free Rx Resources
1234 * @adapter: board private structure
1235 *
1236 * Free all receive software resources
1237 **/
1238
1239void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1240{
1241 struct pci_dev *pdev = adapter->pdev;
1242 struct e1000_ring *rx_ring = adapter->rx_ring;
47f44e40 1243 int i;
bc7f75fa
AK		 1244
1245 e1000_clean_rx_ring(adapter);
1246
47f44e40
AK		 1247 for (i = 0; i < rx_ring->count; i++) {
1248 kfree(rx_ring->buffer_info[i].ps_pages);
1249 }
1250
bc7f75fa
AK		 1251 vfree(rx_ring->buffer_info);
1252 rx_ring->buffer_info = NULL;
1253
bc7f75fa
AK		 1254 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1255 rx_ring->dma);
1256 rx_ring->desc = NULL;
1257}
1258
1259/**
1260 * e1000_update_itr - update the dynamic ITR value based on statistics
1261 * Stores a new ITR value based on packets and byte
1262 * counts during the last interrupt. The advantage of per interrupt
1263 * computation is faster updates and more accurate ITR for the current
1264 * traffic pattern. Constants in this function were computed
1265 * based on theoretical maximum wire speed and thresholds were set based
1266 * on testing data as well as attempting to minimize response time
1267 * while increasing bulk throughput.
1268 * this functionality is controlled by the InterruptThrottleRate module
1269 * parameter (see e1000_param.c)
1270 * @adapter: pointer to adapter
1271 * @itr_setting: current adapter->itr
1272 * @packets: the number of packets during this measurement interval
1273 * @bytes: the number of bytes during this measurement interval
1274 **/
1275static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1276 u16 itr_setting, int packets,
1277 int bytes)
1278{
1279 unsigned int retval = itr_setting;
1280
1281 if (packets == 0)
1282 goto update_itr_done;
1283
1284 switch (itr_setting) {
1285 case lowest_latency:
1286 /* handle TSO and jumbo frames */
1287 if (bytes/packets > 8000)
1288 retval = bulk_latency;
1289 else if ((packets < 5) && (bytes > 512)) {
1290 retval = low_latency;
1291 }
1292 break;
1293 case low_latency: /* 50 usec aka 20000 ints/s */
1294 if (bytes > 10000) {
1295 /* this if handles the TSO accounting */
1296 if (bytes/packets > 8000) {
1297 retval = bulk_latency;
1298 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1299 retval = bulk_latency;
1300 } else if ((packets > 35)) {
1301 retval = lowest_latency;
1302 }
1303 } else if (bytes/packets > 2000) {
1304 retval = bulk_latency;
1305 } else if (packets <= 2 && bytes < 512) {
1306 retval = lowest_latency;
1307 }
1308 break;
1309 case bulk_latency: /* 250 usec aka 4000 ints/s */
1310 if (bytes > 25000) {
1311 if (packets > 35) {
1312 retval = low_latency;
1313 }
1314 } else if (bytes < 6000) {
1315 retval = low_latency;
1316 }
1317 break;
1318 }
1319
1320update_itr_done:
1321 return retval;
1322}
1323
1324static void e1000_set_itr(struct e1000_adapter *adapter)
1325{
1326 struct e1000_hw *hw = &adapter->hw;
1327 u16 current_itr;
1328 u32 new_itr = adapter->itr;
1329
1330 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1331 if (adapter->link_speed != SPEED_1000) {
1332 current_itr = 0;
1333 new_itr = 4000;
1334 goto set_itr_now;
1335 }
1336
1337 adapter->tx_itr = e1000_update_itr(adapter,
1338 adapter->tx_itr,
1339 adapter->total_tx_packets,
1340 adapter->total_tx_bytes);
1341 /* conservative mode (itr 3) eliminates the lowest_latency setting */
1342 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1343 adapter->tx_itr = low_latency;
1344
1345 adapter->rx_itr = e1000_update_itr(adapter,
1346 adapter->rx_itr,
1347 adapter->total_rx_packets,
1348 adapter->total_rx_bytes);
1349 /* conservative mode (itr 3) eliminates the lowest_latency setting */
1350 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1351 adapter->rx_itr = low_latency;
1352
1353 current_itr = max(adapter->rx_itr, adapter->tx_itr);
1354
1355 switch (current_itr) {
1356 /* counts and packets in update_itr are dependent on these numbers */
1357 case lowest_latency:
1358 new_itr = 70000;
1359 break;
1360 case low_latency:
1361 new_itr = 20000; /* aka hwitr = ~200 */
1362 break;
1363 case bulk_latency:
1364 new_itr = 4000;
1365 break;
1366 default:
1367 break;
1368 }
1369
1370set_itr_now:
1371 if (new_itr != adapter->itr) {
1372 /* this attempts to bias the interrupt rate towards Bulk
1373 * by adding intermediate steps when interrupt rate is
1374 * increasing */
1375 new_itr = new_itr > adapter->itr ?
1376 min(adapter->itr + (new_itr >> 2), new_itr) :
1377 new_itr;
1378 adapter->itr = new_itr;
1379 ew32(ITR, 1000000000 / (new_itr * 256));
1380 }
1381}
1382
1383/**
1384 * e1000_clean - NAPI Rx polling callback
1385 * @adapter: board private structure
1386 **/
1387static int e1000_clean(struct napi_struct *napi, int budget)
1388{
1389 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1390 struct net_device *poll_dev = adapter->netdev;
d2c7ddd6 1391 int tx_cleaned = 0, work_done = 0;
bc7f75fa
AK		 1392
1393 /* Must NOT use netdev_priv macro here. */
1394 adapter = poll_dev->priv;
1395
bc7f75fa
AK		 1396 /* e1000_clean is called per-cpu. This lock protects
1397 * tx_ring from being cleaned by multiple cpus
1398 * simultaneously. A failure obtaining the lock means
1399 * tx_ring is currently being cleaned anyway. */
1400 if (spin_trylock(&adapter->tx_queue_lock)) {
d2c7ddd6 1401 tx_cleaned = e1000_clean_tx_irq(adapter);
bc7f75fa
AK		 1402 spin_unlock(&adapter->tx_queue_lock);
1403 }
1404
1405 adapter->clean_rx(adapter, &work_done, budget);
d2c7ddd6
DM		 1406
1407 if (tx_cleaned)
1408 work_done = budget;
bc7f75fa 1409
53e52c72
DM		 1410 /* If budget not fully consumed, exit the polling mode */
1411 if (work_done < budget) {
bc7f75fa
AK		 1412 if (adapter->itr_setting & 3)
1413 e1000_set_itr(adapter);
1414 netif_rx_complete(poll_dev, napi);
1415 e1000_irq_enable(adapter);
1416 }
1417
1418 return work_done;
1419}
1420
1421static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1422{
1423 struct e1000_adapter *adapter = netdev_priv(netdev);
1424 struct e1000_hw *hw = &adapter->hw;
1425 u32 vfta, index;
1426
1427 /* don't update vlan cookie if already programmed */
1428 if ((adapter->hw.mng_cookie.status &
1429 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1430 (vid == adapter->mng_vlan_id))
1431 return;
1432 /* add VID to filter table */
1433 index = (vid >> 5) & 0x7F;
1434 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1435 vfta |= (1 << (vid & 0x1F));
1436 e1000e_write_vfta(hw, index, vfta);
1437}
1438
1439static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1440{
1441 struct e1000_adapter *adapter = netdev_priv(netdev);
1442 struct e1000_hw *hw = &adapter->hw;
1443 u32 vfta, index;
1444
1445 e1000_irq_disable(adapter);
1446 vlan_group_set_device(adapter->vlgrp, vid, NULL);
1447 e1000_irq_enable(adapter);
1448
1449 if ((adapter->hw.mng_cookie.status &
1450 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1451 (vid == adapter->mng_vlan_id)) {
1452 /* release control to f/w */
1453 e1000_release_hw_control(adapter);
1454 return;
1455 }
1456
1457 /* remove VID from filter table */
1458 index = (vid >> 5) & 0x7F;
1459 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1460 vfta &= ~(1 << (vid & 0x1F));
1461 e1000e_write_vfta(hw, index, vfta);
1462}
1463
1464static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
1465{
1466 struct net_device *netdev = adapter->netdev;
1467 u16 vid = adapter->hw.mng_cookie.vlan_id;
1468 u16 old_vid = adapter->mng_vlan_id;
1469
1470 if (!adapter->vlgrp)
1471 return;
1472
1473 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
1474 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1475 if (adapter->hw.mng_cookie.status &
1476 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1477 e1000_vlan_rx_add_vid(netdev, vid);
1478 adapter->mng_vlan_id = vid;
1479 }
1480
1481 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
1482 (vid != old_vid) &&
1483 !vlan_group_get_device(adapter->vlgrp, old_vid))
1484 e1000_vlan_rx_kill_vid(netdev, old_vid);
1485 } else {
1486 adapter->mng_vlan_id = vid;
1487 }
1488}
1489
1490
1491static void e1000_vlan_rx_register(struct net_device *netdev,
1492 struct vlan_group *grp)
1493{
1494 struct e1000_adapter *adapter = netdev_priv(netdev);
1495 struct e1000_hw *hw = &adapter->hw;
1496 u32 ctrl, rctl;
1497
1498 e1000_irq_disable(adapter);
1499 adapter->vlgrp = grp;
1500
1501 if (grp) {
1502 /* enable VLAN tag insert/strip */
1503 ctrl = er32(CTRL);
1504 ctrl |= E1000_CTRL_VME;
1505 ew32(CTRL, ctrl);
1506
1507 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1508 /* enable VLAN receive filtering */
1509 rctl = er32(RCTL);
1510 rctl |= E1000_RCTL_VFE;
1511 rctl &= ~E1000_RCTL_CFIEN;
1512 ew32(RCTL, rctl);
1513 e1000_update_mng_vlan(adapter);
1514 }
1515 } else {
1516 /* disable VLAN tag insert/strip */
1517 ctrl = er32(CTRL);
1518 ctrl &= ~E1000_CTRL_VME;
1519 ew32(CTRL, ctrl);
1520
1521 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1522 /* disable VLAN filtering */
1523 rctl = er32(RCTL);
1524 rctl &= ~E1000_RCTL_VFE;
1525 ew32(RCTL, rctl);
1526 if (adapter->mng_vlan_id !=
1527 (u16)E1000_MNG_VLAN_NONE) {
1528 e1000_vlan_rx_kill_vid(netdev,
1529 adapter->mng_vlan_id);
1530 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1531 }
1532 }
1533 }
1534
1535 e1000_irq_enable(adapter);
1536}
1537
1538static void e1000_restore_vlan(struct e1000_adapter *adapter)
1539{
1540 u16 vid;
1541
1542 e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
1543
1544 if (!adapter->vlgrp)
1545 return;
1546
1547 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1548 if (!vlan_group_get_device(adapter->vlgrp, vid))
1549 continue;
1550 e1000_vlan_rx_add_vid(adapter->netdev, vid);
1551 }
1552}
1553
1554static void e1000_init_manageability(struct e1000_adapter *adapter)
1555{
1556 struct e1000_hw *hw = &adapter->hw;
1557 u32 manc, manc2h;
1558
1559 if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
1560 return;
1561
1562 manc = er32(MANC);
1563
1564 /* disable hardware interception of ARP */
1565 manc &= ~(E1000_MANC_ARP_EN);
1566
1567 /* enable receiving management packets to the host. this will probably
1568 * generate destination unreachable messages from the host OS, but
1569 * the packets will be handled on SMBUS */
1570 manc |= E1000_MANC_EN_MNG2HOST;
1571 manc2h = er32(MANC2H);
1572#define E1000_MNG2HOST_PORT_623 (1 << 5)
1573#define E1000_MNG2HOST_PORT_664 (1 << 6)
1574 manc2h |= E1000_MNG2HOST_PORT_623;
1575 manc2h |= E1000_MNG2HOST_PORT_664;
1576 ew32(MANC2H, manc2h);
1577 ew32(MANC, manc);
1578}
1579
1580/**
1581 * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1582 * @adapter: board private structure
1583 *
1584 * Configure the Tx unit of the MAC after a reset.
1585 **/
1586static void e1000_configure_tx(struct e1000_adapter *adapter)
1587{
1588 struct e1000_hw *hw = &adapter->hw;
1589 struct e1000_ring *tx_ring = adapter->tx_ring;
1590 u64 tdba;
1591 u32 tdlen, tctl, tipg, tarc;
1592 u32 ipgr1, ipgr2;
1593
1594 /* Setup the HW Tx Head and Tail descriptor pointers */
1595 tdba = tx_ring->dma;
1596 tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
1597 ew32(TDBAL, (tdba & DMA_32BIT_MASK));
1598 ew32(TDBAH, (tdba >> 32));
1599 ew32(TDLEN, tdlen);
1600 ew32(TDH, 0);
1601 ew32(TDT, 0);
1602 tx_ring->head = E1000_TDH;
1603 tx_ring->tail = E1000_TDT;
1604
1605 /* Set the default values for the Tx Inter Packet Gap timer */
1606 tipg = DEFAULT_82543_TIPG_IPGT_COPPER; /* 8 */
1607 ipgr1 = DEFAULT_82543_TIPG_IPGR1; /* 8 */
1608 ipgr2 = DEFAULT_82543_TIPG_IPGR2; /* 6 */
1609
1610 if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
1611 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /* 7 */
1612
1613 tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1614 tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1615 ew32(TIPG, tipg);
1616
1617 /* Set the Tx Interrupt Delay register */
1618 ew32(TIDV, adapter->tx_int_delay);
1619 /* tx irq moderation */
1620 ew32(TADV, adapter->tx_abs_int_delay);
1621
1622 /* Program the Transmit Control Register */
1623 tctl = er32(TCTL);
1624 tctl &= ~E1000_TCTL_CT;
1625 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1626 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1627
1628 if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
1629 tarc = er32(TARC0);
1630 /* set the speed mode bit, we'll clear it if we're not at
1631 * gigabit link later */
1632#define SPEED_MODE_BIT (1 << 21)
1633 tarc |= SPEED_MODE_BIT;
1634 ew32(TARC0, tarc);
1635 }
1636
1637 /* errata: program both queues to unweighted RR */
1638 if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
1639 tarc = er32(TARC0);
1640 tarc |= 1;
1641 ew32(TARC0, tarc);
1642 tarc = er32(TARC1);
1643 tarc |= 1;
1644 ew32(TARC1, tarc);
1645 }
1646
1647 e1000e_config_collision_dist(hw);
1648
1649 /* Setup Transmit Descriptor Settings for eop descriptor */
1650 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1651
1652 /* only set IDE if we are delaying interrupts using the timers */
1653 if (adapter->tx_int_delay)
1654 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1655
1656 /* enable Report Status bit */
1657 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1658
1659 ew32(TCTL, tctl);
1660
1661 adapter->tx_queue_len = adapter->netdev->tx_queue_len;
1662}
1663
1664/**
1665 * e1000_setup_rctl - configure the receive control registers
1666 * @adapter: Board private structure
1667 **/
1668#define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1669 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1670static void e1000_setup_rctl(struct e1000_adapter *adapter)
1671{
1672 struct e1000_hw *hw = &adapter->hw;
1673 u32 rctl, rfctl;
1674 u32 psrctl = 0;
1675 u32 pages = 0;
1676
1677 /* Program MC offset vector base */
1678 rctl = er32(RCTL);
1679 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1680 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1681 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1682 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1683
1684 /* Do not Store bad packets */
1685 rctl &= ~E1000_RCTL_SBP;
1686
1687 /* Enable Long Packet receive */
1688 if (adapter->netdev->mtu <= ETH_DATA_LEN)
1689 rctl &= ~E1000_RCTL_LPE;
1690 else
1691 rctl |= E1000_RCTL_LPE;
1692
1693 /* Setup buffer sizes */
1694 rctl &= ~E1000_RCTL_SZ_4096;
1695 rctl |= E1000_RCTL_BSEX;
1696 switch (adapter->rx_buffer_len) {
1697 case 256:
1698 rctl |= E1000_RCTL_SZ_256;
1699 rctl &= ~E1000_RCTL_BSEX;
1700 break;
1701 case 512:
1702 rctl |= E1000_RCTL_SZ_512;
1703 rctl &= ~E1000_RCTL_BSEX;
1704 break;
1705 case 1024:
1706 rctl |= E1000_RCTL_SZ_1024;
1707 rctl &= ~E1000_RCTL_BSEX;
1708 break;
1709 case 2048:
1710 default:
1711 rctl |= E1000_RCTL_SZ_2048;
1712 rctl &= ~E1000_RCTL_BSEX;
1713 break;
1714 case 4096:
1715 rctl |= E1000_RCTL_SZ_4096;
1716 break;
1717 case 8192:
1718 rctl |= E1000_RCTL_SZ_8192;
1719 break;
1720 case 16384:
1721 rctl |= E1000_RCTL_SZ_16384;
1722 break;
1723 }
1724
1725 /*
1726 * 82571 and greater support packet-split where the protocol
1727 * header is placed in skb->data and the packet data is
1728 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
1729 * In the case of a non-split, skb->data is linearly filled,
1730 * followed by the page buffers. Therefore, skb->data is
1731 * sized to hold the largest protocol header.
1732 *
1733 * allocations using alloc_page take too long for regular MTU
1734 * so only enable packet split for jumbo frames
1735 *
1736 * Using pages when the page size is greater than 16k wastes
1737 * a lot of memory, since we allocate 3 pages at all times
1738 * per packet.
1739 */
1740 adapter->rx_ps_pages = 0;
1741 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
1742 if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
1743 adapter->rx_ps_pages = pages;
1744
1745 if (adapter->rx_ps_pages) {
1746 /* Configure extra packet-split registers */
1747 rfctl = er32(RFCTL);
1748 rfctl |= E1000_RFCTL_EXTEN;
1749 /* disable packet split support for IPv6 extension headers,
1750 * because some malformed IPv6 headers can hang the RX */
1751 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
1752 E1000_RFCTL_NEW_IPV6_EXT_DIS);
1753
1754 ew32(RFCTL, rfctl);
1755
140a7480
AK		 1756 /* Enable Packet split descriptors */
1757 rctl |= E1000_RCTL_DTYP_PS;
1758
1759 /* Enable hardware CRC frame stripping */
1760 rctl |= E1000_RCTL_SECRC;
bc7f75fa
AK		 1761
1762 psrctl |= adapter->rx_ps_bsize0 >>
1763 E1000_PSRCTL_BSIZE0_SHIFT;
1764
1765 switch (adapter->rx_ps_pages) {
1766 case 3:
1767 psrctl |= PAGE_SIZE <<
1768 E1000_PSRCTL_BSIZE3_SHIFT;
1769 case 2:
1770 psrctl |= PAGE_SIZE <<
1771 E1000_PSRCTL_BSIZE2_SHIFT;
1772 case 1:
1773 psrctl |= PAGE_SIZE >>
1774 E1000_PSRCTL_BSIZE1_SHIFT;
1775 break;
1776 }
1777
1778 ew32(PSRCTL, psrctl);
1779 }
1780
1781 ew32(RCTL, rctl);
1782}
1783
1784/**
1785 * e1000_configure_rx - Configure Receive Unit after Reset
1786 * @adapter: board private structure
1787 *
1788 * Configure the Rx unit of the MAC after a reset.
1789 **/
1790static void e1000_configure_rx(struct e1000_adapter *adapter)
1791{
1792 struct e1000_hw *hw = &adapter->hw;
1793 struct e1000_ring *rx_ring = adapter->rx_ring;
1794 u64 rdba;
1795 u32 rdlen, rctl, rxcsum, ctrl_ext;
1796
1797 if (adapter->rx_ps_pages) {
1798 /* this is a 32 byte descriptor */
1799 rdlen = rx_ring->count *
1800 sizeof(union e1000_rx_desc_packet_split);
1801 adapter->clean_rx = e1000_clean_rx_irq_ps;
1802 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
bc7f75fa
AK		 1803 } else {
1804 rdlen = rx_ring->count *
1805 sizeof(struct e1000_rx_desc);
1806 adapter->clean_rx = e1000_clean_rx_irq;
1807 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1808 }
1809
1810 /* disable receives while setting up the descriptors */
1811 rctl = er32(RCTL);
1812 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1813 e1e_flush();
1814 msleep(10);
1815
1816 /* set the Receive Delay Timer Register */
1817 ew32(RDTR, adapter->rx_int_delay);
1818
1819 /* irq moderation */
1820 ew32(RADV, adapter->rx_abs_int_delay);
1821 if (adapter->itr_setting != 0)
1822 ew32(ITR,
1823 1000000000 / (adapter->itr * 256));
1824
1825 ctrl_ext = er32(CTRL_EXT);
1826 /* Reset delay timers after every interrupt */
1827 ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
1828 /* Auto-Mask interrupts upon ICR access */
1829 ctrl_ext |= E1000_CTRL_EXT_IAME;
1830 ew32(IAM, 0xffffffff);
1831 ew32(CTRL_EXT, ctrl_ext);
1832 e1e_flush();
1833
1834 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1835 * the Base and Length of the Rx Descriptor Ring */
1836 rdba = rx_ring->dma;
1837 ew32(RDBAL, (rdba & DMA_32BIT_MASK));
1838 ew32(RDBAH, (rdba >> 32));
1839 ew32(RDLEN, rdlen);
1840 ew32(RDH, 0);
1841 ew32(RDT, 0);
1842 rx_ring->head = E1000_RDH;
1843 rx_ring->tail = E1000_RDT;
1844
1845 /* Enable Receive Checksum Offload for TCP and UDP */
1846 rxcsum = er32(RXCSUM);
1847 if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
1848 rxcsum |= E1000_RXCSUM_TUOFL;
1849
1850 /* IPv4 payload checksum for UDP fragments must be
1851 * used in conjunction with packet-split. */
1852 if (adapter->rx_ps_pages)
1853 rxcsum |= E1000_RXCSUM_IPPCSE;
1854 } else {
1855 rxcsum &= ~E1000_RXCSUM_TUOFL;
1856 /* no need to clear IPPCSE as it defaults to 0 */
1857 }
1858 ew32(RXCSUM, rxcsum);
1859
1860 /* Enable early receives on supported devices, only takes effect when
1861 * packet size is equal or larger than the specified value (in 8 byte
1862 * units), e.g. using jumbo frames when setting to E1000_ERT_2048 */
1863 if ((adapter->flags & FLAG_HAS_ERT) &&
1864 (adapter->netdev->mtu > ETH_DATA_LEN))
1865 ew32(ERT, E1000_ERT_2048);
1866
1867 /* Enable Receives */
1868 ew32(RCTL, rctl);
1869}
1870
1871/**
1872 * e1000_mc_addr_list_update - Update Multicast addresses
1873 * @hw: pointer to the HW structure
1874 * @mc_addr_list: array of multicast addresses to program
1875 * @mc_addr_count: number of multicast addresses to program
1876 * @rar_used_count: the first RAR register free to program
1877 * @rar_count: total number of supported Receive Address Registers
1878 *
1879 * Updates the Receive Address Registers and Multicast Table Array.
1880 * The caller must have a packed mc_addr_list of multicast addresses.
1881 * The parameter rar_count will usually be hw->mac.rar_entry_count
1882 * unless there are workarounds that change this. Currently no func pointer
1883 * exists and all implementations are handled in the generic version of this
1884 * function.
1885 **/
1886static void e1000_mc_addr_list_update(struct e1000_hw *hw, u8 *mc_addr_list,
1887 u32 mc_addr_count, u32 rar_used_count,
1888 u32 rar_count)
1889{
1890 hw->mac.ops.mc_addr_list_update(hw, mc_addr_list, mc_addr_count,
1891 rar_used_count, rar_count);
1892}
1893
1894/**
1895 * e1000_set_multi - Multicast and Promiscuous mode set
1896 * @netdev: network interface device structure
1897 *
1898 * The set_multi entry point is called whenever the multicast address
1899 * list or the network interface flags are updated. This routine is
1900 * responsible for configuring the hardware for proper multicast,
1901 * promiscuous mode, and all-multi behavior.
1902 **/
1903static void e1000_set_multi(struct net_device *netdev)
1904{
1905 struct e1000_adapter *adapter = netdev_priv(netdev);
1906 struct e1000_hw *hw = &adapter->hw;
1907 struct e1000_mac_info *mac = &hw->mac;
1908 struct dev_mc_list *mc_ptr;
1909 u8 *mta_list;
1910 u32 rctl;
1911 int i;
1912
1913 /* Check for Promiscuous and All Multicast modes */
1914
1915 rctl = er32(RCTL);
1916
1917 if (netdev->flags & IFF_PROMISC) {
1918 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
1919 } else if (netdev->flags & IFF_ALLMULTI) {
1920 rctl |= E1000_RCTL_MPE;
1921 rctl &= ~E1000_RCTL_UPE;
1922 } else {
1923 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
1924 }
1925
1926 ew32(RCTL, rctl);
1927
1928 if (netdev->mc_count) {
1929 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
1930 if (!mta_list)
1931 return;
1932
1933 /* prepare a packed array of only addresses. */
1934 mc_ptr = netdev->mc_list;
1935
1936 for (i = 0; i < netdev->mc_count; i++) {
1937 if (!mc_ptr)
1938 break;
1939 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
1940 ETH_ALEN);
1941 mc_ptr = mc_ptr->next;
1942 }
1943
1944 e1000_mc_addr_list_update(hw, mta_list, i, 1,
1945 mac->rar_entry_count);
1946 kfree(mta_list);
1947 } else {
1948 /*
1949 * if we're called from probe, we might not have
1950 * anything to do here, so clear out the list
1951 */
1952 e1000_mc_addr_list_update(hw, NULL, 0, 1,
1953 mac->rar_entry_count);
1954 }
1955}
1956
1957/**
1958 * e1000_configure - configure the hardware for RX and TX
1959 * @adapter: private board structure
1960 **/
1961static void e1000_configure(struct e1000_adapter *adapter)
1962{
1963 e1000_set_multi(adapter->netdev);
1964
1965 e1000_restore_vlan(adapter);
1966 e1000_init_manageability(adapter);
1967
1968 e1000_configure_tx(adapter);
1969 e1000_setup_rctl(adapter);
1970 e1000_configure_rx(adapter);
1971 adapter->alloc_rx_buf(adapter,
1972 e1000_desc_unused(adapter->rx_ring));
1973}
1974
1975/**
1976 * e1000e_power_up_phy - restore link in case the phy was powered down
1977 * @adapter: address of board private structure
1978 *
1979 * The phy may be powered down to save power and turn off link when the
1980 * driver is unloaded and wake on lan is not enabled (among others)
1981 * *** this routine MUST be followed by a call to e1000e_reset ***
1982 **/
1983void e1000e_power_up_phy(struct e1000_adapter *adapter)
1984{
1985 u16 mii_reg = 0;
1986
1987 /* Just clear the power down bit to wake the phy back up */
1988 if (adapter->hw.media_type == e1000_media_type_copper) {
1989 /* according to the manual, the phy will retain its
1990 * settings across a power-down/up cycle */
1991 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
1992 mii_reg &= ~MII_CR_POWER_DOWN;
1993 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
1994 }
1995
1996 adapter->hw.mac.ops.setup_link(&adapter->hw);
1997}
1998
1999/**
2000 * e1000_power_down_phy - Power down the PHY
2001 *
2002 * Power down the PHY so no link is implied when interface is down
2003 * The PHY cannot be powered down is management or WoL is active
2004 */
2005static void e1000_power_down_phy(struct e1000_adapter *adapter)
2006{
2007 struct e1000_hw *hw = &adapter->hw;
2008 u16 mii_reg;
2009
2010 /* WoL is enabled */
23b66e2b 2011 if (adapter->wol)
bc7f75fa
AK		 2012 return;
2013
2014 /* non-copper PHY? */
2015 if (adapter->hw.media_type != e1000_media_type_copper)
2016 return;
2017
2018 /* reset is blocked because of a SoL/IDER session */
2019 if (e1000e_check_mng_mode(hw) ||
2020 e1000_check_reset_block(hw))
2021 return;
2022
2023 /* managebility (AMT) is enabled */
2024 if (er32(MANC) & E1000_MANC_SMBUS_EN)
2025 return;
2026
2027 /* power down the PHY */
2028 e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2029 mii_reg |= MII_CR_POWER_DOWN;
2030 e1e_wphy(hw, PHY_CONTROL, mii_reg);
2031 mdelay(1);
2032}
2033
2034/**
2035 * e1000e_reset - bring the hardware into a known good state
2036 *
2037 * This function boots the hardware and enables some settings that
2038 * require a configuration cycle of the hardware - those cannot be
2039 * set/changed during runtime. After reset the device needs to be
2040 * properly configured for rx, tx etc.
2041 */
2042void e1000e_reset(struct e1000_adapter *adapter)
2043{
2044 struct e1000_mac_info *mac = &adapter->hw.mac;
2045 struct e1000_hw *hw = &adapter->hw;
2046 u32 tx_space, min_tx_space, min_rx_space;
df762464 2047 u32 pba;
bc7f75fa
AK		 2048 u16 hwm;
2049
df762464
AK		 2050 ew32(PBA, adapter->pba);
2051
bc7f75fa
AK		 2052 if (mac->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN ) {
2053 /* To maintain wire speed transmits, the Tx FIFO should be
2054 * large enough to accommodate two full transmit packets,
2055 * rounded up to the next 1KB and expressed in KB. Likewise,
2056 * the Rx FIFO should be large enough to accommodate at least
2057 * one full receive packet and is similarly rounded up and
2058 * expressed in KB. */
df762464 2059 pba = er32(PBA);
bc7f75fa 2060 /* upper 16 bits has Tx packet buffer allocation size in KB */
df762464 2061 tx_space = pba >> 16;
bc7f75fa 2062 /* lower 16 bits has Rx packet buffer allocation size in KB */
df762464 2063 pba &= 0xffff;
bc7f75fa
AK		 2064 /* the tx fifo also stores 16 bytes of information about the tx
2065 * but don't include ethernet FCS because hardware appends it */
2066 min_tx_space = (mac->max_frame_size +
2067 sizeof(struct e1000_tx_desc) -
2068 ETH_FCS_LEN) * 2;
2069 min_tx_space = ALIGN(min_tx_space, 1024);
2070 min_tx_space >>= 10;
2071 /* software strips receive CRC, so leave room for it */
2072 min_rx_space = mac->max_frame_size;
2073 min_rx_space = ALIGN(min_rx_space, 1024);
2074 min_rx_space >>= 10;
2075
2076 /* If current Tx allocation is less than the min Tx FIFO size,
2077 * and the min Tx FIFO size is less than the current Rx FIFO
2078 * allocation, take space away from current Rx allocation */
df762464
AK		 2079 if ((tx_space < min_tx_space) &&
2080 ((min_tx_space - tx_space) < pba)) {
2081 pba -= min_tx_space - tx_space;
bc7f75fa
AK		 2082
2083 /* if short on rx space, rx wins and must trump tx
2084 * adjustment or use Early Receive if available */
df762464 2085 if ((pba < min_rx_space) &&
bc7f75fa
AK		 2086 (!(adapter->flags & FLAG_HAS_ERT)))
2087 /* ERT enabled in e1000_configure_rx */
df762464 2088 pba = min_rx_space;
bc7f75fa 2089 }
df762464
AK		 2090
2091 ew32(PBA, pba);
bc7f75fa
AK		 2092 }
2093
bc7f75fa
AK		 2094
2095 /* flow control settings */
2096 /* The high water mark must be low enough to fit one full frame
2097 * (or the size used for early receive) above it in the Rx FIFO.
2098 * Set it to the lower of:
2099 * - 90% of the Rx FIFO size, and
2100 * - the full Rx FIFO size minus the early receive size (for parts
2101 * with ERT support assuming ERT set to E1000_ERT_2048), or
2102 * - the full Rx FIFO size minus one full frame */
2103 if (adapter->flags & FLAG_HAS_ERT)
2104 hwm = min(((adapter->pba << 10) * 9 / 10),
2105 ((adapter->pba << 10) - (E1000_ERT_2048 << 3)));
2106 else
2107 hwm = min(((adapter->pba << 10) * 9 / 10),
2108 ((adapter->pba << 10) - mac->max_frame_size));
2109
2110 mac->fc_high_water = hwm & 0xFFF8; /* 8-byte granularity */
2111 mac->fc_low_water = mac->fc_high_water - 8;
2112
2113 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2114 mac->fc_pause_time = 0xFFFF;
2115 else
2116 mac->fc_pause_time = E1000_FC_PAUSE_TIME;
2117 mac->fc = mac->original_fc;
2118
2119 /* Allow time for pending master requests to run */
2120 mac->ops.reset_hw(hw);
2121 ew32(WUC, 0);
2122
2123 if (mac->ops.init_hw(hw))
2124 ndev_err(adapter->netdev, "Hardware Error\n");
2125
2126 e1000_update_mng_vlan(adapter);
2127
2128 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2129 ew32(VET, ETH_P_8021Q);
2130
2131 e1000e_reset_adaptive(hw);
2132 e1000_get_phy_info(hw);
2133
2134 if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2135 u16 phy_data = 0;
2136 /* speed up time to link by disabling smart power down, ignore
2137 * the return value of this function because there is nothing
2138 * different we would do if it failed */
2139 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2140 phy_data &= ~IGP02E1000_PM_SPD;
2141 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2142 }
2143
2144 e1000_release_manageability(adapter);
2145}
2146
2147int e1000e_up(struct e1000_adapter *adapter)
2148{
2149 struct e1000_hw *hw = &adapter->hw;
2150
2151 /* hardware has been reset, we need to reload some things */
2152 e1000_configure(adapter);
2153
2154 clear_bit(__E1000_DOWN, &adapter->state);
2155
2156 napi_enable(&adapter->napi);
2157 e1000_irq_enable(adapter);
2158
2159 /* fire a link change interrupt to start the watchdog */
2160 ew32(ICS, E1000_ICS_LSC);
2161 return 0;
2162}
2163
2164void e1000e_down(struct e1000_adapter *adapter)
2165{
2166 struct net_device *netdev = adapter->netdev;
2167 struct e1000_hw *hw = &adapter->hw;
2168 u32 tctl, rctl;
2169
2170 /* signal that we're down so the interrupt handler does not
2171 * reschedule our watchdog timer */
2172 set_bit(__E1000_DOWN, &adapter->state);
2173
2174 /* disable receives in the hardware */
2175 rctl = er32(RCTL);
2176 ew32(RCTL, rctl & ~E1000_RCTL_EN);
2177 /* flush and sleep below */
2178
2179 netif_stop_queue(netdev);
2180
2181 /* disable transmits in the hardware */
2182 tctl = er32(TCTL);
2183 tctl &= ~E1000_TCTL_EN;
2184 ew32(TCTL, tctl);
2185 /* flush both disables and wait for them to finish */
2186 e1e_flush();
2187 msleep(10);
2188
2189 napi_disable(&adapter->napi);
49d85c50 2190 atomic_set(&adapter->irq_sem, 0);
bc7f75fa
AK		 2191 e1000_irq_disable(adapter);
2192
2193 del_timer_sync(&adapter->watchdog_timer);
2194 del_timer_sync(&adapter->phy_info_timer);
2195
2196 netdev->tx_queue_len = adapter->tx_queue_len;
2197 netif_carrier_off(netdev);
2198 adapter->link_speed = 0;
2199 adapter->link_duplex = 0;
2200
2201 e1000e_reset(adapter);
2202 e1000_clean_tx_ring(adapter);
2203 e1000_clean_rx_ring(adapter);
2204
2205 /*
2206 * TODO: for power management, we could drop the link and
2207 * pci_disable_device here.
2208 */
2209}
2210
2211void e1000e_reinit_locked(struct e1000_adapter *adapter)
2212{
2213 might_sleep();
2214 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2215 msleep(1);
2216 e1000e_down(adapter);
2217 e1000e_up(adapter);
2218 clear_bit(__E1000_RESETTING, &adapter->state);
2219}
2220
2221/**
2222 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2223 * @adapter: board private structure to initialize
2224 *
2225 * e1000_sw_init initializes the Adapter private data structure.
2226 * Fields are initialized based on PCI device information and
2227 * OS network device settings (MTU size).
2228 **/
2229static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2230{
2231 struct e1000_hw *hw = &adapter->hw;
2232 struct net_device *netdev = adapter->netdev;
2233
2234 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2235 adapter->rx_ps_bsize0 = 128;
2236 hw->mac.max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2237 hw->mac.min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2238
2239 adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2240 if (!adapter->tx_ring)
2241 goto err;
2242
2243 adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2244 if (!adapter->rx_ring)
2245 goto err;
2246
2247 spin_lock_init(&adapter->tx_queue_lock);
2248
2249 /* Explicitly disable IRQ since the NIC can be in any state. */
2250 atomic_set(&adapter->irq_sem, 0);
2251 e1000_irq_disable(adapter);
2252
2253 spin_lock_init(&adapter->stats_lock);
2254
2255 set_bit(__E1000_DOWN, &adapter->state);
2256 return 0;
2257
2258err:
2259 ndev_err(netdev, "Unable to allocate memory for queues\n");
2260 kfree(adapter->rx_ring);
2261 kfree(adapter->tx_ring);
2262 return -ENOMEM;
2263}
2264
2265/**
2266 * e1000_open - Called when a network interface is made active
2267 * @netdev: network interface device structure
2268 *
2269 * Returns 0 on success, negative value on failure
2270 *
2271 * The open entry point is called when a network interface is made
2272 * active by the system (IFF_UP). At this point all resources needed
2273 * for transmit and receive operations are allocated, the interrupt
2274 * handler is registered with the OS, the watchdog timer is started,
2275 * and the stack is notified that the interface is ready.
2276 **/
2277static int e1000_open(struct net_device *netdev)
2278{
2279 struct e1000_adapter *adapter = netdev_priv(netdev);
2280 struct e1000_hw *hw = &adapter->hw;
2281 int err;
2282
2283 /* disallow open during test */
2284 if (test_bit(__E1000_TESTING, &adapter->state))
2285 return -EBUSY;
2286
2287 /* allocate transmit descriptors */
2288 err = e1000e_setup_tx_resources(adapter);
2289 if (err)
2290 goto err_setup_tx;
2291
2292 /* allocate receive descriptors */
2293 err = e1000e_setup_rx_resources(adapter);
2294 if (err)
2295 goto err_setup_rx;
2296
2297 e1000e_power_up_phy(adapter);
2298
2299 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2300 if ((adapter->hw.mng_cookie.status &
2301 E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
2302 e1000_update_mng_vlan(adapter);
2303
2304 /* If AMT is enabled, let the firmware know that the network
2305 * interface is now open */
2306 if ((adapter->flags & FLAG_HAS_AMT) &&
2307 e1000e_check_mng_mode(&adapter->hw))
2308 e1000_get_hw_control(adapter);
2309
2310 /* before we allocate an interrupt, we must be ready to handle it.
2311 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2312 * as soon as we call pci_request_irq, so we have to setup our
2313 * clean_rx handler before we do so. */
2314 e1000_configure(adapter);
2315
2316 err = e1000_request_irq(adapter);
2317 if (err)
2318 goto err_req_irq;
2319
2320 /* From here on the code is the same as e1000e_up() */
2321 clear_bit(__E1000_DOWN, &adapter->state);
2322
2323 napi_enable(&adapter->napi);
2324
2325 e1000_irq_enable(adapter);
2326
2327 /* fire a link status change interrupt to start the watchdog */
2328 ew32(ICS, E1000_ICS_LSC);
2329
2330 return 0;
2331
2332err_req_irq:
2333 e1000_release_hw_control(adapter);
2334 e1000_power_down_phy(adapter);
2335 e1000e_free_rx_resources(adapter);
2336err_setup_rx:
2337 e1000e_free_tx_resources(adapter);
2338err_setup_tx:
2339 e1000e_reset(adapter);
2340
2341 return err;
2342}
2343
2344/**
2345 * e1000_close - Disables a network interface
2346 * @netdev: network interface device structure
2347 *
2348 * Returns 0, this is not allowed to fail
2349 *
2350 * The close entry point is called when an interface is de-activated
2351 * by the OS. The hardware is still under the drivers control, but
2352 * needs to be disabled. A global MAC reset is issued to stop the
2353 * hardware, and all transmit and receive resources are freed.
2354 **/
2355static int e1000_close(struct net_device *netdev)
2356{
2357 struct e1000_adapter *adapter = netdev_priv(netdev);
2358
2359 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
2360 e1000e_down(adapter);
2361 e1000_power_down_phy(adapter);
2362 e1000_free_irq(adapter);
2363
2364 e1000e_free_tx_resources(adapter);
2365 e1000e_free_rx_resources(adapter);
2366
2367 /* kill manageability vlan ID if supported, but not if a vlan with
2368 * the same ID is registered on the host OS (let 8021q kill it) */
2369 if ((adapter->hw.mng_cookie.status &
2370 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2371 !(adapter->vlgrp &&
2372 vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
2373 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2374
2375 /* If AMT is enabled, let the firmware know that the network
2376 * interface is now closed */
2377 if ((adapter->flags & FLAG_HAS_AMT) &&
2378 e1000e_check_mng_mode(&adapter->hw))
2379 e1000_release_hw_control(adapter);
2380
2381 return 0;
2382}
2383/**
2384 * e1000_set_mac - Change the Ethernet Address of the NIC
2385 * @netdev: network interface device structure
2386 * @p: pointer to an address structure
2387 *
2388 * Returns 0 on success, negative on failure
2389 **/
2390static int e1000_set_mac(struct net_device *netdev, void *p)
2391{
2392 struct e1000_adapter *adapter = netdev_priv(netdev);
2393 struct sockaddr *addr = p;
2394
2395 if (!is_valid_ether_addr(addr->sa_data))
2396 return -EADDRNOTAVAIL;
2397
2398 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2399 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
2400
2401 e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
2402
2403 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
2404 /* activate the work around */
2405 e1000e_set_laa_state_82571(&adapter->hw, 1);
2406
2407 /* Hold a copy of the LAA in RAR[14] This is done so that
2408 * between the time RAR[0] gets clobbered and the time it
2409 * gets fixed (in e1000_watchdog), the actual LAA is in one
2410 * of the RARs and no incoming packets directed to this port
2411 * are dropped. Eventually the LAA will be in RAR[0] and
2412 * RAR[14] */
2413 e1000e_rar_set(&adapter->hw,
2414 adapter->hw.mac.addr,
2415 adapter->hw.mac.rar_entry_count - 1);
2416 }
2417
2418 return 0;
2419}
2420
2421/* Need to wait a few seconds after link up to get diagnostic information from
2422 * the phy */
2423static void e1000_update_phy_info(unsigned long data)
2424{
2425 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2426 e1000_get_phy_info(&adapter->hw);
2427}
2428
2429/**
2430 * e1000e_update_stats - Update the board statistics counters
2431 * @adapter: board private structure
2432 **/
2433void e1000e_update_stats(struct e1000_adapter *adapter)
2434{
2435 struct e1000_hw *hw = &adapter->hw;
2436 struct pci_dev *pdev = adapter->pdev;
2437 unsigned long irq_flags;
2438 u16 phy_tmp;
2439
2440#define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
2441
2442 /*
2443 * Prevent stats update while adapter is being reset, or if the pci
2444 * connection is down.
2445 */
2446 if (adapter->link_speed == 0)
2447 return;
2448 if (pci_channel_offline(pdev))
2449 return;
2450
2451 spin_lock_irqsave(&adapter->stats_lock, irq_flags);
2452
2453 /* these counters are modified from e1000_adjust_tbi_stats,
2454 * called from the interrupt context, so they must only
2455 * be written while holding adapter->stats_lock
2456 */
2457
2458 adapter->stats.crcerrs += er32(CRCERRS);
2459 adapter->stats.gprc += er32(GPRC);
2460 adapter->stats.gorcl += er32(GORCL);
2461 adapter->stats.gorch += er32(GORCH);
2462 adapter->stats.bprc += er32(BPRC);
2463 adapter->stats.mprc += er32(MPRC);
2464 adapter->stats.roc += er32(ROC);
2465
2466 if (adapter->flags & FLAG_HAS_STATS_PTC_PRC) {
2467 adapter->stats.prc64 += er32(PRC64);
2468 adapter->stats.prc127 += er32(PRC127);
2469 adapter->stats.prc255 += er32(PRC255);
2470 adapter->stats.prc511 += er32(PRC511);
2471 adapter->stats.prc1023 += er32(PRC1023);
2472 adapter->stats.prc1522 += er32(PRC1522);
2473 adapter->stats.symerrs += er32(SYMERRS);
2474 adapter->stats.sec += er32(SEC);
2475 }
2476
2477 adapter->stats.mpc += er32(MPC);
2478 adapter->stats.scc += er32(SCC);
2479 adapter->stats.ecol += er32(ECOL);
2480 adapter->stats.mcc += er32(MCC);
2481 adapter->stats.latecol += er32(LATECOL);
2482 adapter->stats.dc += er32(DC);
2483 adapter->stats.rlec += er32(RLEC);
2484 adapter->stats.xonrxc += er32(XONRXC);
2485 adapter->stats.xontxc += er32(XONTXC);
2486 adapter->stats.xoffrxc += er32(XOFFRXC);
2487 adapter->stats.xofftxc += er32(XOFFTXC);
2488 adapter->stats.fcruc += er32(FCRUC);
2489 adapter->stats.gptc += er32(GPTC);
2490 adapter->stats.gotcl += er32(GOTCL);
2491 adapter->stats.gotch += er32(GOTCH);
2492 adapter->stats.rnbc += er32(RNBC);
2493 adapter->stats.ruc += er32(RUC);
2494 adapter->stats.rfc += er32(RFC);
2495 adapter->stats.rjc += er32(RJC);
2496 adapter->stats.torl += er32(TORL);
2497 adapter->stats.torh += er32(TORH);
2498 adapter->stats.totl += er32(TOTL);
2499 adapter->stats.toth += er32(TOTH);
2500 adapter->stats.tpr += er32(TPR);
2501
2502 if (adapter->flags & FLAG_HAS_STATS_PTC_PRC) {
2503 adapter->stats.ptc64 += er32(PTC64);
2504 adapter->stats.ptc127 += er32(PTC127);
2505 adapter->stats.ptc255 += er32(PTC255);
2506 adapter->stats.ptc511 += er32(PTC511);
2507 adapter->stats.ptc1023 += er32(PTC1023);
2508 adapter->stats.ptc1522 += er32(PTC1522);
2509 }
2510
2511 adapter->stats.mptc += er32(MPTC);
2512 adapter->stats.bptc += er32(BPTC);
2513
2514 /* used for adaptive IFS */
2515
2516 hw->mac.tx_packet_delta = er32(TPT);
2517 adapter->stats.tpt += hw->mac.tx_packet_delta;
2518 hw->mac.collision_delta = er32(COLC);
2519 adapter->stats.colc += hw->mac.collision_delta;
2520
2521 adapter->stats.algnerrc += er32(ALGNERRC);
2522 adapter->stats.rxerrc += er32(RXERRC);
2523 adapter->stats.tncrs += er32(TNCRS);
2524 adapter->stats.cexterr += er32(CEXTERR);
2525 adapter->stats.tsctc += er32(TSCTC);
2526 adapter->stats.tsctfc += er32(TSCTFC);
2527
2528 adapter->stats.iac += er32(IAC);
2529
2530 if (adapter->flags & FLAG_HAS_STATS_ICR_ICT) {
2531 adapter->stats.icrxoc += er32(ICRXOC);
2532 adapter->stats.icrxptc += er32(ICRXPTC);
2533 adapter->stats.icrxatc += er32(ICRXATC);
2534 adapter->stats.ictxptc += er32(ICTXPTC);
2535 adapter->stats.ictxatc += er32(ICTXATC);
2536 adapter->stats.ictxqec += er32(ICTXQEC);
2537 adapter->stats.ictxqmtc += er32(ICTXQMTC);
2538 adapter->stats.icrxdmtc += er32(ICRXDMTC);
2539 }
2540
2541 /* Fill out the OS statistics structure */
bc7f75fa
AK		 2542 adapter->net_stats.multicast = adapter->stats.mprc;
2543 adapter->net_stats.collisions = adapter->stats.colc;
2544
2545 /* Rx Errors */
2546
2547 /* RLEC on some newer hardware can be incorrect so build
2548 * our own version based on RUC and ROC */
2549 adapter->net_stats.rx_errors = adapter->stats.rxerrc +
2550 adapter->stats.crcerrs + adapter->stats.algnerrc +
2551 adapter->stats.ruc + adapter->stats.roc +
2552 adapter->stats.cexterr;
2553 adapter->net_stats.rx_length_errors = adapter->stats.ruc +
2554 adapter->stats.roc;
2555 adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
2556 adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
2557 adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
2558
2559 /* Tx Errors */
2560 adapter->net_stats.tx_errors = adapter->stats.ecol +
2561 adapter->stats.latecol;
2562 adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
2563 adapter->net_stats.tx_window_errors = adapter->stats.latecol;
2564 adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
2565
2566 /* Tx Dropped needs to be maintained elsewhere */
2567
2568 /* Phy Stats */
2569 if (hw->media_type == e1000_media_type_copper) {
2570 if ((adapter->link_speed == SPEED_1000) &&
2571 (!e1e_rphy(hw, PHY_1000T_STATUS, &phy_tmp))) {
2572 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
2573 adapter->phy_stats.idle_errors += phy_tmp;
2574 }
2575 }
2576
2577 /* Management Stats */
2578 adapter->stats.mgptc += er32(MGTPTC);
2579 adapter->stats.mgprc += er32(MGTPRC);
2580 adapter->stats.mgpdc += er32(MGTPDC);
2581
2582 spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
2583}
2584
2585static void e1000_print_link_info(struct e1000_adapter *adapter)
2586{
2587 struct net_device *netdev = adapter->netdev;
2588 struct e1000_hw *hw = &adapter->hw;
2589 u32 ctrl = er32(CTRL);
2590
2591 ndev_info(netdev,
2592 "Link is Up %d Mbps %s, Flow Control: %s\n",
2593 adapter->link_speed,
2594 (adapter->link_duplex == FULL_DUPLEX) ?
2595 "Full Duplex" : "Half Duplex",
2596 ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
2597 "RX/TX" :
2598 ((ctrl & E1000_CTRL_RFCE) ? "RX" :
2599 ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
2600}
2601
2602/**
2603 * e1000_watchdog - Timer Call-back
2604 * @data: pointer to adapter cast into an unsigned long
2605 **/
2606static void e1000_watchdog(unsigned long data)
2607{
2608 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2609
2610 /* Do the rest outside of interrupt context */
2611 schedule_work(&adapter->watchdog_task);
2612
2613 /* TODO: make this use queue_delayed_work() */
2614}
2615
2616static void e1000_watchdog_task(struct work_struct *work)
2617{
2618 struct e1000_adapter *adapter = container_of(work,
2619 struct e1000_adapter, watchdog_task);
2620
2621 struct net_device *netdev = adapter->netdev;
2622 struct e1000_mac_info *mac = &adapter->hw.mac;
2623 struct e1000_ring *tx_ring = adapter->tx_ring;
2624 struct e1000_hw *hw = &adapter->hw;
2625 u32 link, tctl;
2626 s32 ret_val;
2627 int tx_pending = 0;
2628
2629 if ((netif_carrier_ok(netdev)) &&
2630 (er32(STATUS) & E1000_STATUS_LU))
2631 goto link_up;
2632
2633 ret_val = mac->ops.check_for_link(hw);
2634 if ((ret_val == E1000_ERR_PHY) &&
2635 (adapter->hw.phy.type == e1000_phy_igp_3) &&
2636 (er32(CTRL) &
2637 E1000_PHY_CTRL_GBE_DISABLE)) {
2638 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
2639 ndev_info(netdev,
2640 "Gigabit has been disabled, downgrading speed\n");
2641 }
2642
2643 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
2644 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
2645 e1000_update_mng_vlan(adapter);
2646
2647 if ((adapter->hw.media_type == e1000_media_type_internal_serdes) &&
2648 !(er32(TXCW) & E1000_TXCW_ANE))
2649 link = adapter->hw.mac.serdes_has_link;
2650 else
2651 link = er32(STATUS) & E1000_STATUS_LU;
2652
2653 if (link) {
2654 if (!netif_carrier_ok(netdev)) {
2655 bool txb2b = 1;
2656 mac->ops.get_link_up_info(&adapter->hw,
2657 &adapter->link_speed,
2658 &adapter->link_duplex);
2659 e1000_print_link_info(adapter);
2660 /* tweak tx_queue_len according to speed/duplex
2661 * and adjust the timeout factor */
2662 netdev->tx_queue_len = adapter->tx_queue_len;
2663 adapter->tx_timeout_factor = 1;
2664 switch (adapter->link_speed) {
2665 case SPEED_10:
2666 txb2b = 0;
2667 netdev->tx_queue_len = 10;
2668 adapter->tx_timeout_factor = 14;
2669 break;
2670 case SPEED_100:
2671 txb2b = 0;
2672 netdev->tx_queue_len = 100;
2673 /* maybe add some timeout factor ? */
2674 break;
2675 }
2676
2677 /* workaround: re-program speed mode bit after
2678 * link-up event */
2679 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
2680 !txb2b) {
2681 u32 tarc0;
2682 tarc0 = er32(TARC0);
2683 tarc0 &= ~SPEED_MODE_BIT;
2684 ew32(TARC0, tarc0);
2685 }
2686
2687 /* disable TSO for pcie and 10/100 speeds, to avoid
2688 * some hardware issues */
2689 if (!(adapter->flags & FLAG_TSO_FORCE)) {
2690 switch (adapter->link_speed) {
2691 case SPEED_10:
2692 case SPEED_100:
2693 ndev_info(netdev,
2694 "10/100 speed: disabling TSO\n");
2695 netdev->features &= ~NETIF_F_TSO;
2696 netdev->features &= ~NETIF_F_TSO6;
2697 break;
2698 case SPEED_1000:
2699 netdev->features |= NETIF_F_TSO;
2700 netdev->features |= NETIF_F_TSO6;
2701 break;
2702 default:
2703 /* oops */
2704 break;
2705 }
2706 }
2707
2708 /* enable transmits in the hardware, need to do this
2709 * after setting TARC0 */
2710 tctl = er32(TCTL);
2711 tctl |= E1000_TCTL_EN;
2712 ew32(TCTL, tctl);
2713
2714 netif_carrier_on(netdev);
2715 netif_wake_queue(netdev);
2716
2717 if (!test_bit(__E1000_DOWN, &adapter->state))
2718 mod_timer(&adapter->phy_info_timer,
2719 round_jiffies(jiffies + 2 * HZ));
2720 } else {
2721 /* make sure the receive unit is started */
2722 if (adapter->flags & FLAG_RX_NEEDS_RESTART) {
2723 u32 rctl = er32(RCTL);
2724 ew32(RCTL, rctl |
2725 E1000_RCTL_EN);
2726 }
2727 }
2728 } else {
2729 if (netif_carrier_ok(netdev)) {
2730 adapter->link_speed = 0;
2731 adapter->link_duplex = 0;
2732 ndev_info(netdev, "Link is Down\n");
2733 netif_carrier_off(netdev);
2734 netif_stop_queue(netdev);
2735 if (!test_bit(__E1000_DOWN, &adapter->state))
2736 mod_timer(&adapter->phy_info_timer,
2737 round_jiffies(jiffies + 2 * HZ));
2738
2739 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
2740 schedule_work(&adapter->reset_task);
2741 }
2742 }
2743
2744link_up:
2745 e1000e_update_stats(adapter);
2746
2747 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2748 adapter->tpt_old = adapter->stats.tpt;
2749 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
2750 adapter->colc_old = adapter->stats.colc;
2751
2752 adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2753 adapter->gorcl_old = adapter->stats.gorcl;
2754 adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2755 adapter->gotcl_old = adapter->stats.gotcl;
2756
2757 e1000e_update_adaptive(&adapter->hw);
2758
2759 if (!netif_carrier_ok(netdev)) {
2760 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
2761 tx_ring->count);
2762 if (tx_pending) {
2763 /* We've lost link, so the controller stops DMA,
2764 * but we've got queued Tx work that's never going
2765 * to get done, so reset controller to flush Tx.
2766 * (Do the reset outside of interrupt context). */
2767 adapter->tx_timeout_count++;
2768 schedule_work(&adapter->reset_task);
2769 }
2770 }
2771
2772 /* Cause software interrupt to ensure rx ring is cleaned */
2773 ew32(ICS, E1000_ICS_RXDMT0);
2774
2775 /* Force detection of hung controller every watchdog period */
2776 adapter->detect_tx_hung = 1;
2777
2778 /* With 82571 controllers, LAA may be overwritten due to controller
2779 * reset from the other port. Set the appropriate LAA in RAR[0] */
2780 if (e1000e_get_laa_state_82571(hw))
2781 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
2782
2783 /* Reset the timer */
2784 if (!test_bit(__E1000_DOWN, &adapter->state))
2785 mod_timer(&adapter->watchdog_timer,
2786 round_jiffies(jiffies + 2 * HZ));
2787}
2788
2789#define E1000_TX_FLAGS_CSUM 0x00000001
2790#define E1000_TX_FLAGS_VLAN 0x00000002
2791#define E1000_TX_FLAGS_TSO 0x00000004
2792#define E1000_TX_FLAGS_IPV4 0x00000008
2793#define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
2794#define E1000_TX_FLAGS_VLAN_SHIFT 16
2795
2796static int e1000_tso(struct e1000_adapter *adapter,
2797 struct sk_buff *skb)
2798{
2799 struct e1000_ring *tx_ring = adapter->tx_ring;
2800 struct e1000_context_desc *context_desc;
2801 struct e1000_buffer *buffer_info;
2802 unsigned int i;
2803 u32 cmd_length = 0;
2804 u16 ipcse = 0, tucse, mss;
2805 u8 ipcss, ipcso, tucss, tucso, hdr_len;
2806 int err;
2807
2808 if (skb_is_gso(skb)) {
2809 if (skb_header_cloned(skb)) {
2810 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2811 if (err)
2812 return err;
2813 }
2814
2815 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2816 mss = skb_shinfo(skb)->gso_size;
2817 if (skb->protocol == htons(ETH_P_IP)) {
2818 struct iphdr *iph = ip_hdr(skb);
2819 iph->tot_len = 0;
2820 iph->check = 0;
2821 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2822 iph->daddr, 0,
2823 IPPROTO_TCP,
2824 0);
2825 cmd_length = E1000_TXD_CMD_IP;
2826 ipcse = skb_transport_offset(skb) - 1;
2827 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
2828 ipv6_hdr(skb)->payload_len = 0;
2829 tcp_hdr(skb)->check =
2830 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2831 &ipv6_hdr(skb)->daddr,
2832 0, IPPROTO_TCP, 0);
2833 ipcse = 0;
2834 }
2835 ipcss = skb_network_offset(skb);
2836 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2837 tucss = skb_transport_offset(skb);
2838 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2839 tucse = 0;
2840
2841 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2842 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2843
2844 i = tx_ring->next_to_use;
2845 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2846 buffer_info = &tx_ring->buffer_info[i];
2847
2848 context_desc->lower_setup.ip_fields.ipcss = ipcss;
2849 context_desc->lower_setup.ip_fields.ipcso = ipcso;
2850 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
2851 context_desc->upper_setup.tcp_fields.tucss = tucss;
2852 context_desc->upper_setup.tcp_fields.tucso = tucso;
2853 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2854 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
2855 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2856 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2857
2858 buffer_info->time_stamp = jiffies;
2859 buffer_info->next_to_watch = i;
2860
2861 i++;
2862 if (i == tx_ring->count)
2863 i = 0;
2864 tx_ring->next_to_use = i;
2865
2866 return 1;
2867 }
2868
2869 return 0;
2870}
2871
2872static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
2873{
2874 struct e1000_ring *tx_ring = adapter->tx_ring;
2875 struct e1000_context_desc *context_desc;
2876 struct e1000_buffer *buffer_info;
2877 unsigned int i;
2878 u8 css;
2879
2880 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2881 css = skb_transport_offset(skb);
2882
2883 i = tx_ring->next_to_use;
2884 buffer_info = &tx_ring->buffer_info[i];
2885 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2886
2887 context_desc->lower_setup.ip_config = 0;
2888 context_desc->upper_setup.tcp_fields.tucss = css;
2889 context_desc->upper_setup.tcp_fields.tucso =
2890 css + skb->csum_offset;
2891 context_desc->upper_setup.tcp_fields.tucse = 0;
2892 context_desc->tcp_seg_setup.data = 0;
2893 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
2894
2895 buffer_info->time_stamp = jiffies;
2896 buffer_info->next_to_watch = i;
2897
2898 i++;
2899 if (i == tx_ring->count)
2900 i = 0;
2901 tx_ring->next_to_use = i;
2902
2903 return 1;
2904 }
2905
2906 return 0;
2907}
2908
2909#define E1000_MAX_PER_TXD 8192
2910#define E1000_MAX_TXD_PWR 12
2911
2912static int e1000_tx_map(struct e1000_adapter *adapter,
2913 struct sk_buff *skb, unsigned int first,
2914 unsigned int max_per_txd, unsigned int nr_frags,
2915 unsigned int mss)
2916{
2917 struct e1000_ring *tx_ring = adapter->tx_ring;
2918 struct e1000_buffer *buffer_info;
2919 unsigned int len = skb->len - skb->data_len;
2920 unsigned int offset = 0, size, count = 0, i;
2921 unsigned int f;
2922
2923 i = tx_ring->next_to_use;
2924
2925 while (len) {
2926 buffer_info = &tx_ring->buffer_info[i];
2927 size = min(len, max_per_txd);
2928
2929 /* Workaround for premature desc write-backs
2930 * in TSO mode. Append 4-byte sentinel desc */
2931 if (mss && !nr_frags && size == len && size > 8)
2932 size -= 4;
2933
2934 buffer_info->length = size;
2935 /* set time_stamp *before* dma to help avoid a possible race */
2936 buffer_info->time_stamp = jiffies;
2937 buffer_info->dma =
2938 pci_map_single(adapter->pdev,
2939 skb->data + offset,
2940 size,
2941 PCI_DMA_TODEVICE);
2942 if (pci_dma_mapping_error(buffer_info->dma)) {
2943 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
2944 adapter->tx_dma_failed++;
2945 return -1;
2946 }
2947 buffer_info->next_to_watch = i;
2948
2949 len -= size;
2950 offset += size;
2951 count++;
2952 i++;
2953 if (i == tx_ring->count)
2954 i = 0;
2955 }
2956
2957 for (f = 0; f < nr_frags; f++) {
2958 struct skb_frag_struct *frag;
2959
2960 frag = &skb_shinfo(skb)->frags[f];
2961 len = frag->size;
2962 offset = frag->page_offset;
2963
2964 while (len) {
2965 buffer_info = &tx_ring->buffer_info[i];
2966 size = min(len, max_per_txd);
2967 /* Workaround for premature desc write-backs
2968 * in TSO mode. Append 4-byte sentinel desc */
2969 if (mss && f == (nr_frags-1) && size == len && size > 8)
2970 size -= 4;
2971
2972 buffer_info->length = size;
2973 buffer_info->time_stamp = jiffies;
2974 buffer_info->dma =
2975 pci_map_page(adapter->pdev,
2976 frag->page,
2977 offset,
2978 size,
2979 PCI_DMA_TODEVICE);
2980 if (pci_dma_mapping_error(buffer_info->dma)) {
2981 dev_err(&adapter->pdev->dev,
2982 "TX DMA page map failed\n");
2983 adapter->tx_dma_failed++;
2984 return -1;
2985 }
2986
2987 buffer_info->next_to_watch = i;
2988
2989 len -= size;
2990 offset += size;
2991 count++;
2992
2993 i++;
2994 if (i == tx_ring->count)
2995 i = 0;
2996 }
2997 }
2998
2999 if (i == 0)
3000 i = tx_ring->count - 1;
3001 else
3002 i--;
3003
3004 tx_ring->buffer_info[i].skb = skb;
3005 tx_ring->buffer_info[first].next_to_watch = i;
3006
3007 return count;
3008}
3009
3010static void e1000_tx_queue(struct e1000_adapter *adapter,
3011 int tx_flags, int count)
3012{
3013 struct e1000_ring *tx_ring = adapter->tx_ring;
3014 struct e1000_tx_desc *tx_desc = NULL;
3015 struct e1000_buffer *buffer_info;
3016 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3017 unsigned int i;
3018
3019 if (tx_flags & E1000_TX_FLAGS_TSO) {
3020 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3021 E1000_TXD_CMD_TSE;
3022 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3023
3024 if (tx_flags & E1000_TX_FLAGS_IPV4)
3025 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3026 }
3027
3028 if (tx_flags & E1000_TX_FLAGS_CSUM) {
3029 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3030 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3031 }
3032
3033 if (tx_flags & E1000_TX_FLAGS_VLAN) {
3034 txd_lower |= E1000_TXD_CMD_VLE;
3035 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3036 }
3037
3038 i = tx_ring->next_to_use;
3039
3040 while (count--) {
3041 buffer_info = &tx_ring->buffer_info[i];
3042 tx_desc = E1000_TX_DESC(*tx_ring, i);
3043 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3044 tx_desc->lower.data =
3045 cpu_to_le32(txd_lower | buffer_info->length);
3046 tx_desc->upper.data = cpu_to_le32(txd_upper);
3047
3048 i++;
3049 if (i == tx_ring->count)
3050 i = 0;
3051 }
3052
3053 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3054
3055 /* Force memory writes to complete before letting h/w
3056 * know there are new descriptors to fetch. (Only
3057 * applicable for weak-ordered memory model archs,
3058 * such as IA-64). */
3059 wmb();
3060
3061 tx_ring->next_to_use = i;
3062 writel(i, adapter->hw.hw_addr + tx_ring->tail);
3063 /* we need this if more than one processor can write to our tail
3064 * at a time, it synchronizes IO on IA64/Altix systems */
3065 mmiowb();
3066}
3067
3068#define MINIMUM_DHCP_PACKET_SIZE 282
3069static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3070 struct sk_buff *skb)
3071{
3072 struct e1000_hw *hw = &adapter->hw;
3073 u16 length, offset;
3074
3075 if (vlan_tx_tag_present(skb)) {
3076 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3077 && (adapter->hw.mng_cookie.status &
3078 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3079 return 0;
3080 }
3081
3082 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3083 return 0;
3084
3085 if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3086 return 0;
3087
3088 {
3089 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3090 struct udphdr *udp;
3091
3092 if (ip->protocol != IPPROTO_UDP)
3093 return 0;
3094
3095 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3096 if (ntohs(udp->dest) != 67)
3097 return 0;
3098
3099 offset = (u8 *)udp + 8 - skb->data;
3100 length = skb->len - offset;
3101 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
3102 }
3103
3104 return 0;
3105}
3106
3107static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3108{
3109 struct e1000_adapter *adapter = netdev_priv(netdev);
3110
3111 netif_stop_queue(netdev);
3112 /* Herbert's original patch had:
3113 * smp_mb__after_netif_stop_queue();
3114 * but since that doesn't exist yet, just open code it. */
3115 smp_mb();
3116
3117 /* We need to check again in a case another CPU has just
3118 * made room available. */
3119 if (e1000_desc_unused(adapter->tx_ring) < size)
3120 return -EBUSY;
3121
3122 /* A reprieve! */
3123 netif_start_queue(netdev);
3124 ++adapter->restart_queue;
3125 return 0;
3126}
3127
3128static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
3129{
3130 struct e1000_adapter *adapter = netdev_priv(netdev);
3131
3132 if (e1000_desc_unused(adapter->tx_ring) >= size)
3133 return 0;
3134 return __e1000_maybe_stop_tx(netdev, size);
3135}
3136
3137#define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3138static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3139{
3140 struct e1000_adapter *adapter = netdev_priv(netdev);
3141 struct e1000_ring *tx_ring = adapter->tx_ring;
3142 unsigned int first;
3143 unsigned int max_per_txd = E1000_MAX_PER_TXD;
3144 unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3145 unsigned int tx_flags = 0;
4e6c709c 3146 unsigned int len = skb->len - skb->data_len;
bc7f75fa 3147 unsigned long irq_flags;
4e6c709c
AK		 3148 unsigned int nr_frags;
3149 unsigned int mss;
bc7f75fa
AK		 3150 int count = 0;
3151 int tso;
3152 unsigned int f;
bc7f75fa
AK		 3153
3154 if (test_bit(__E1000_DOWN, &adapter->state)) {
3155 dev_kfree_skb_any(skb);
3156 return NETDEV_TX_OK;
3157 }
3158
3159 if (skb->len <= 0) {
3160 dev_kfree_skb_any(skb);
3161 return NETDEV_TX_OK;
3162 }
3163
3164 mss = skb_shinfo(skb)->gso_size;
3165 /* The controller does a simple calculation to
3166 * make sure there is enough room in the FIFO before
3167 * initiating the DMA for each buffer. The calc is:
3168 * 4 = ceil(buffer len/mss). To make sure we don't
3169 * overrun the FIFO, adjust the max buffer len if mss
3170 * drops. */
3171 if (mss) {
3172 u8 hdr_len;
3173 max_per_txd = min(mss << 2, max_per_txd);
3174 max_txd_pwr = fls(max_per_txd) - 1;
3175
3176 /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
3177 * points to just header, pull a few bytes of payload from
3178 * frags into skb->data */
3179 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4e6c709c 3180 if (skb->data_len && (hdr_len == len)) {
bc7f75fa
AK		 3181 unsigned int pull_size;
3182
3183 pull_size = min((unsigned int)4, skb->data_len);
3184 if (!__pskb_pull_tail(skb, pull_size)) {
3185 ndev_err(netdev,
3186 "__pskb_pull_tail failed.\n");
3187 dev_kfree_skb_any(skb);
3188 return NETDEV_TX_OK;
3189 }
3190 len = skb->len - skb->data_len;
3191 }
3192 }
3193
3194 /* reserve a descriptor for the offload context */
3195 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3196 count++;
3197 count++;
3198
3199 count += TXD_USE_COUNT(len, max_txd_pwr);
3200
3201 nr_frags = skb_shinfo(skb)->nr_frags;
3202 for (f = 0; f < nr_frags; f++)
3203 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3204 max_txd_pwr);
3205
3206 if (adapter->hw.mac.tx_pkt_filtering)
3207 e1000_transfer_dhcp_info(adapter, skb);
3208
3209 if (!spin_trylock_irqsave(&adapter->tx_queue_lock, irq_flags))
3210 /* Collision - tell upper layer to requeue */
3211 return NETDEV_TX_LOCKED;
3212
3213 /* need: count + 2 desc gap to keep tail from touching
3214 * head, otherwise try next time */
3215 if (e1000_maybe_stop_tx(netdev, count + 2)) {
3216 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3217 return NETDEV_TX_BUSY;
3218 }
3219
3220 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3221 tx_flags |= E1000_TX_FLAGS_VLAN;
3222 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3223 }
3224
3225 first = tx_ring->next_to_use;
3226
3227 tso = e1000_tso(adapter, skb);
3228 if (tso < 0) {
3229 dev_kfree_skb_any(skb);
3230 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3231 return NETDEV_TX_OK;
3232 }
3233
3234 if (tso)
3235 tx_flags |= E1000_TX_FLAGS_TSO;
3236 else if (e1000_tx_csum(adapter, skb))
3237 tx_flags |= E1000_TX_FLAGS_CSUM;
3238
3239 /* Old method was to assume IPv4 packet by default if TSO was enabled.
3240 * 82571 hardware supports TSO capabilities for IPv6 as well...
3241 * no longer assume, we must. */
3242 if (skb->protocol == htons(ETH_P_IP))
3243 tx_flags |= E1000_TX_FLAGS_IPV4;
3244
3245 count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
3246 if (count < 0) {
3247 /* handle pci_map_single() error in e1000_tx_map */
3248 dev_kfree_skb_any(skb);
3249 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
7b5dfe1a 3250 return NETDEV_TX_OK;
bc7f75fa
AK		 3251 }
3252
3253 e1000_tx_queue(adapter, tx_flags, count);
3254
3255 netdev->trans_start = jiffies;
3256
3257 /* Make sure there is space in the ring for the next send. */
3258 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
3259
3260 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3261 return NETDEV_TX_OK;
3262}
3263
3264/**
3265 * e1000_tx_timeout - Respond to a Tx Hang
3266 * @netdev: network interface device structure
3267 **/
3268static void e1000_tx_timeout(struct net_device *netdev)
3269{
3270 struct e1000_adapter *adapter = netdev_priv(netdev);
3271
3272 /* Do the reset outside of interrupt context */
3273 adapter->tx_timeout_count++;
3274 schedule_work(&adapter->reset_task);
3275}
3276
3277static void e1000_reset_task(struct work_struct *work)
3278{
3279 struct e1000_adapter *adapter;
3280 adapter = container_of(work, struct e1000_adapter, reset_task);
3281
3282 e1000e_reinit_locked(adapter);
3283}
3284
3285/**
3286 * e1000_get_stats - Get System Network Statistics
3287 * @netdev: network interface device structure
3288 *
3289 * Returns the address of the device statistics structure.
3290 * The statistics are actually updated from the timer callback.
3291 **/
3292static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3293{
3294 struct e1000_adapter *adapter = netdev_priv(netdev);
3295
3296 /* only return the current stats */
3297 return &adapter->net_stats;
3298}
3299
3300/**
3301 * e1000_change_mtu - Change the Maximum Transfer Unit
3302 * @netdev: network interface device structure
3303 * @new_mtu: new value for maximum frame size
3304 *
3305 * Returns 0 on success, negative on failure
3306 **/
3307static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3308{
3309 struct e1000_adapter *adapter = netdev_priv(netdev);
3310 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3311
3312 if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
3313 (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3314 ndev_err(netdev, "Invalid MTU setting\n");
3315 return -EINVAL;
3316 }
3317
3318 /* Jumbo frame size limits */
3319 if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
3320 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
3321 ndev_err(netdev, "Jumbo Frames not supported.\n");
3322 return -EINVAL;
3323 }
3324 if (adapter->hw.phy.type == e1000_phy_ife) {
3325 ndev_err(netdev, "Jumbo Frames not supported.\n");
3326 return -EINVAL;
3327 }
3328 }
3329
3330#define MAX_STD_JUMBO_FRAME_SIZE 9234
3331 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3332 ndev_err(netdev, "MTU > 9216 not supported.\n");
3333 return -EINVAL;
3334 }
3335
3336 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3337 msleep(1);
3338 /* e1000e_down has a dependency on max_frame_size */
3339 adapter->hw.mac.max_frame_size = max_frame;
3340 if (netif_running(netdev))
3341 e1000e_down(adapter);
3342
3343 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3344 * means we reserve 2 more, this pushes us to allocate from the next
3345 * larger slab size.
f920c186 3346 * i.e. RXBUFFER_2048 --> size-4096 slab */
bc7f75fa
AK		 3347
3348 if (max_frame <= 256)
3349 adapter->rx_buffer_len = 256;
3350 else if (max_frame <= 512)
3351 adapter->rx_buffer_len = 512;
3352 else if (max_frame <= 1024)
3353 adapter->rx_buffer_len = 1024;
3354 else if (max_frame <= 2048)
3355 adapter->rx_buffer_len = 2048;
3356 else
3357 adapter->rx_buffer_len = 4096;
3358
3359 /* adjust allocation if LPE protects us, and we aren't using SBP */
3360 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3361 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
3362 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
3363 + ETH_FCS_LEN ;
3364
3365 ndev_info(netdev, "changing MTU from %d to %d\n",
3366 netdev->mtu, new_mtu);
3367 netdev->mtu = new_mtu;
3368
3369 if (netif_running(netdev))
3370 e1000e_up(adapter);
3371 else
3372 e1000e_reset(adapter);
3373
3374 clear_bit(__E1000_RESETTING, &adapter->state);
3375
3376 return 0;
3377}
3378
3379static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
3380 int cmd)
3381{
3382 struct e1000_adapter *adapter = netdev_priv(netdev);
3383 struct mii_ioctl_data *data = if_mii(ifr);
3384 unsigned long irq_flags;
3385
3386 if (adapter->hw.media_type != e1000_media_type_copper)
3387 return -EOPNOTSUPP;
3388
3389 switch (cmd) {
3390 case SIOCGMIIPHY:
3391 data->phy_id = adapter->hw.phy.addr;
3392 break;
3393 case SIOCGMIIREG:
3394 if (!capable(CAP_NET_ADMIN))
3395 return -EPERM;
3396 spin_lock_irqsave(&adapter->stats_lock, irq_flags);
3397 if (e1e_rphy(&adapter->hw, data->reg_num & 0x1F,
3398 &data->val_out)) {
3399 spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
3400 return -EIO;
3401 }
3402 spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
3403 break;
3404 case SIOCSMIIREG:
3405 default:
3406 return -EOPNOTSUPP;
3407 }
3408 return 0;
3409}
3410
3411static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
3412{
3413 switch (cmd) {
3414 case SIOCGMIIPHY:
3415 case SIOCGMIIREG:
3416 case SIOCSMIIREG:
3417 return e1000_mii_ioctl(netdev, ifr, cmd);
3418 default:
3419 return -EOPNOTSUPP;
3420 }
3421}
3422
3423static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
3424{
3425 struct net_device *netdev = pci_get_drvdata(pdev);
3426 struct e1000_adapter *adapter = netdev_priv(netdev);
3427 struct e1000_hw *hw = &adapter->hw;
3428 u32 ctrl, ctrl_ext, rctl, status;
3429 u32 wufc = adapter->wol;
3430 int retval = 0;
3431
3432 netif_device_detach(netdev);
3433
3434 if (netif_running(netdev)) {
3435 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3436 e1000e_down(adapter);
3437 e1000_free_irq(adapter);
3438 }
3439
3440 retval = pci_save_state(pdev);
3441 if (retval)
3442 return retval;
3443
3444 status = er32(STATUS);
3445 if (status & E1000_STATUS_LU)
3446 wufc &= ~E1000_WUFC_LNKC;
3447
3448 if (wufc) {
3449 e1000_setup_rctl(adapter);
3450 e1000_set_multi(netdev);
3451
3452 /* turn on all-multi mode if wake on multicast is enabled */
3453 if (wufc & E1000_WUFC_MC) {
3454 rctl = er32(RCTL);
3455 rctl |= E1000_RCTL_MPE;
3456 ew32(RCTL, rctl);
3457 }
3458
3459 ctrl = er32(CTRL);
3460 /* advertise wake from D3Cold */
3461 #define E1000_CTRL_ADVD3WUC 0x00100000
3462 /* phy power management enable */
3463 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
3464 ctrl |= E1000_CTRL_ADVD3WUC |
3465 E1000_CTRL_EN_PHY_PWR_MGMT;
3466 ew32(CTRL, ctrl);
3467
3468 if (adapter->hw.media_type == e1000_media_type_fiber ||
3469 adapter->hw.media_type == e1000_media_type_internal_serdes) {
3470 /* keep the laser running in D3 */
3471 ctrl_ext = er32(CTRL_EXT);
3472 ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
3473 ew32(CTRL_EXT, ctrl_ext);
3474 }
3475
3476 /* Allow time for pending master requests to run */
3477 e1000e_disable_pcie_master(&adapter->hw);
3478
3479 ew32(WUC, E1000_WUC_PME_EN);
3480 ew32(WUFC, wufc);
3481 pci_enable_wake(pdev, PCI_D3hot, 1);
3482 pci_enable_wake(pdev, PCI_D3cold, 1);
3483 } else {
3484 ew32(WUC, 0);
3485 ew32(WUFC, 0);
3486 pci_enable_wake(pdev, PCI_D3hot, 0);
3487 pci_enable_wake(pdev, PCI_D3cold, 0);
3488 }
3489
3490 e1000_release_manageability(adapter);
3491
3492 /* make sure adapter isn't asleep if manageability is enabled */
3493 if (adapter->flags & FLAG_MNG_PT_ENABLED) {
3494 pci_enable_wake(pdev, PCI_D3hot, 1);
3495 pci_enable_wake(pdev, PCI_D3cold, 1);
3496 }
3497
3498 if (adapter->hw.phy.type == e1000_phy_igp_3)
3499 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
3500
3501 /* Release control of h/w to f/w. If f/w is AMT enabled, this
3502 * would have already happened in close and is redundant. */
3503 e1000_release_hw_control(adapter);
3504
3505 pci_disable_device(pdev);
3506
3507 pci_set_power_state(pdev, pci_choose_state(pdev, state));
3508
3509 return 0;
3510}
3511
1eae4eb2
AK		 3512static void e1000e_disable_l1aspm(struct pci_dev *pdev)
3513{
3514 int pos;
3515 u32 cap;
3516 u16 val;
3517
3518 /*
3519 * 82573 workaround - disable L1 ASPM on mobile chipsets
3520 *
3521 * L1 ASPM on various mobile (ich7) chipsets do not behave properly
3522 * resulting in lost data or garbage information on the pci-e link
3523 * level. This could result in (false) bad EEPROM checksum errors,
3524 * long ping times (up to 2s) or even a system freeze/hang.
3525 *
3526 * Unfortunately this feature saves about 1W power consumption when
3527 * active.
3528 */
3529 pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
3530 pci_read_config_dword(pdev, pos + PCI_EXP_LNKCAP, &cap);
3531 pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
3532 if (val & 0x2) {
3533 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
3534 val &= ~0x2;
3535 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
3536 }
3537}
3538
bc7f75fa
AK		 3539#ifdef CONFIG_PM
3540static int e1000_resume(struct pci_dev *pdev)
3541{
3542 struct net_device *netdev = pci_get_drvdata(pdev);
3543 struct e1000_adapter *adapter = netdev_priv(netdev);
3544 struct e1000_hw *hw = &adapter->hw;
3545 u32 err;
3546
3547 pci_set_power_state(pdev, PCI_D0);
3548 pci_restore_state(pdev);
1eae4eb2 3549 e1000e_disable_l1aspm(pdev);
bc7f75fa
AK		 3550 err = pci_enable_device(pdev);
3551 if (err) {
3552 dev_err(&pdev->dev,
3553 "Cannot enable PCI device from suspend\n");
3554 return err;
3555 }
3556
3557 pci_set_master(pdev);
3558
3559 pci_enable_wake(pdev, PCI_D3hot, 0);
3560 pci_enable_wake(pdev, PCI_D3cold, 0);
3561
3562 if (netif_running(netdev)) {
3563 err = e1000_request_irq(adapter);
3564 if (err)
3565 return err;
3566 }
3567
3568 e1000e_power_up_phy(adapter);
3569 e1000e_reset(adapter);
3570 ew32(WUS, ~0);
3571
3572 e1000_init_manageability(adapter);
3573
3574 if (netif_running(netdev))
3575 e1000e_up(adapter);
3576
3577 netif_device_attach(netdev);
3578
3579 /* If the controller has AMT, do not set DRV_LOAD until the interface
3580 * is up. For all other cases, let the f/w know that the h/w is now
3581 * under the control of the driver. */
3582 if (!(adapter->flags & FLAG_HAS_AMT) || !e1000e_check_mng_mode(&adapter->hw))
3583 e1000_get_hw_control(adapter);
3584
3585 return 0;
3586}
3587#endif
3588
3589static void e1000_shutdown(struct pci_dev *pdev)
3590{
3591 e1000_suspend(pdev, PMSG_SUSPEND);
3592}
3593
3594#ifdef CONFIG_NET_POLL_CONTROLLER
3595/*
3596 * Polling 'interrupt' - used by things like netconsole to send skbs
3597 * without having to re-enable interrupts. It's not called while
3598 * the interrupt routine is executing.
3599 */
3600static void e1000_netpoll(struct net_device *netdev)
3601{
3602 struct e1000_adapter *adapter = netdev_priv(netdev);
3603
3604 disable_irq(adapter->pdev->irq);
3605 e1000_intr(adapter->pdev->irq, netdev);
3606
3607 e1000_clean_tx_irq(adapter);
3608
3609 enable_irq(adapter->pdev->irq);
3610}
3611#endif
3612
3613/**
3614 * e1000_io_error_detected - called when PCI error is detected
3615 * @pdev: Pointer to PCI device
3616 * @state: The current pci connection state
3617 *
3618 * This function is called after a PCI bus error affecting
3619 * this device has been detected.
3620 */
3621static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
3622 pci_channel_state_t state)
3623{
3624 struct net_device *netdev = pci_get_drvdata(pdev);
3625 struct e1000_adapter *adapter = netdev_priv(netdev);
3626
3627 netif_device_detach(netdev);
3628
3629 if (netif_running(netdev))
3630 e1000e_down(adapter);
3631 pci_disable_device(pdev);
3632
3633 /* Request a slot slot reset. */
3634 return PCI_ERS_RESULT_NEED_RESET;
3635}
3636
3637/**
3638 * e1000_io_slot_reset - called after the pci bus has been reset.
3639 * @pdev: Pointer to PCI device
3640 *
3641 * Restart the card from scratch, as if from a cold-boot. Implementation
3642 * resembles the first-half of the e1000_resume routine.
3643 */
3644static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
3645{
3646 struct net_device *netdev = pci_get_drvdata(pdev);
3647 struct e1000_adapter *adapter = netdev_priv(netdev);
3648 struct e1000_hw *hw = &adapter->hw;
3649
1eae4eb2 3650 e1000e_disable_l1aspm(pdev);
bc7f75fa
AK		 3651 if (pci_enable_device(pdev)) {
3652 dev_err(&pdev->dev,
3653 "Cannot re-enable PCI device after reset.\n");
3654 return PCI_ERS_RESULT_DISCONNECT;
3655 }
3656 pci_set_master(pdev);
3657
3658 pci_enable_wake(pdev, PCI_D3hot, 0);
3659 pci_enable_wake(pdev, PCI_D3cold, 0);
3660
3661 e1000e_reset(adapter);
3662 ew32(WUS, ~0);
3663
3664 return PCI_ERS_RESULT_RECOVERED;
3665}
3666
3667/**
3668 * e1000_io_resume - called when traffic can start flowing again.
3669 * @pdev: Pointer to PCI device
3670 *
3671 * This callback is called when the error recovery driver tells us that
3672 * its OK to resume normal operation. Implementation resembles the
3673 * second-half of the e1000_resume routine.
3674 */
3675static void e1000_io_resume(struct pci_dev *pdev)
3676{
3677 struct net_device *netdev = pci_get_drvdata(pdev);
3678 struct e1000_adapter *adapter = netdev_priv(netdev);
3679
3680 e1000_init_manageability(adapter);
3681
3682 if (netif_running(netdev)) {
3683 if (e1000e_up(adapter)) {
3684 dev_err(&pdev->dev,
3685 "can't bring device back up after reset\n");
3686 return;
3687 }
3688 }
3689
3690 netif_device_attach(netdev);
3691
3692 /* If the controller has AMT, do not set DRV_LOAD until the interface
3693 * is up. For all other cases, let the f/w know that the h/w is now
3694 * under the control of the driver. */
3695 if (!(adapter->flags & FLAG_HAS_AMT) ||
3696 !e1000e_check_mng_mode(&adapter->hw))
3697 e1000_get_hw_control(adapter);
3698
3699}
3700
3701static void e1000_print_device_info(struct e1000_adapter *adapter)
3702{
3703 struct e1000_hw *hw = &adapter->hw;
3704 struct net_device *netdev = adapter->netdev;
3705 u32 part_num;
3706
3707 /* print bus type/speed/width info */
3708 ndev_info(netdev, "(PCI Express:2.5GB/s:%s) "
3709 "%02x:%02x:%02x:%02x:%02x:%02x\n",
3710 /* bus width */
3711 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
3712 "Width x1"),
3713 /* MAC address */
3714 netdev->dev_addr[0], netdev->dev_addr[1],
3715 netdev->dev_addr[2], netdev->dev_addr[3],
3716 netdev->dev_addr[4], netdev->dev_addr[5]);
3717 ndev_info(netdev, "Intel(R) PRO/%s Network Connection\n",
3718 (hw->phy.type == e1000_phy_ife)
3719 ? "10/100" : "1000");
3720 e1000e_read_part_num(hw, &part_num);
3721 ndev_info(netdev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
3722 hw->mac.type, hw->phy.type,
3723 (part_num >> 8), (part_num & 0xff));
3724}
3725
3726/**
3727 * e1000_probe - Device Initialization Routine
3728 * @pdev: PCI device information struct
3729 * @ent: entry in e1000_pci_tbl
3730 *
3731 * Returns 0 on success, negative on failure
3732 *
3733 * e1000_probe initializes an adapter identified by a pci_dev structure.
3734 * The OS initialization, configuring of the adapter private structure,
3735 * and a hardware reset occur.
3736 **/
3737static int __devinit e1000_probe(struct pci_dev *pdev,
3738 const struct pci_device_id *ent)
3739{
3740 struct net_device *netdev;
3741 struct e1000_adapter *adapter;
3742 struct e1000_hw *hw;
3743 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
3744 unsigned long mmio_start, mmio_len;
3745 unsigned long flash_start, flash_len;
3746
3747 static int cards_found;
3748 int i, err, pci_using_dac;
3749 u16 eeprom_data = 0;
3750 u16 eeprom_apme_mask = E1000_EEPROM_APME;
3751
1eae4eb2 3752 e1000e_disable_l1aspm(pdev);
bc7f75fa
AK		 3753 err = pci_enable_device(pdev);
3754 if (err)
3755 return err;
3756
3757 pci_using_dac = 0;
3758 err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
3759 if (!err) {
3760 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
3761 if (!err)
3762 pci_using_dac = 1;
3763 } else {
3764 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
3765 if (err) {
3766 err = pci_set_consistent_dma_mask(pdev,
3767 DMA_32BIT_MASK);
3768 if (err) {
3769 dev_err(&pdev->dev, "No usable DMA "
3770 "configuration, aborting\n");
3771 goto err_dma;
3772 }
3773 }
3774 }
3775
3776 err = pci_request_regions(pdev, e1000e_driver_name);
3777 if (err)
3778 goto err_pci_reg;
3779
3780 pci_set_master(pdev);
3781
3782 err = -ENOMEM;
3783 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
3784 if (!netdev)
3785 goto err_alloc_etherdev;
3786
bc7f75fa
AK		 3787 SET_NETDEV_DEV(netdev, &pdev->dev);
3788
3789 pci_set_drvdata(pdev, netdev);
3790 adapter = netdev_priv(netdev);
3791 hw = &adapter->hw;
3792 adapter->netdev = netdev;
3793 adapter->pdev = pdev;
3794 adapter->ei = ei;
3795 adapter->pba = ei->pba;
3796 adapter->flags = ei->flags;
3797 adapter->hw.adapter = adapter;
3798 adapter->hw.mac.type = ei->mac;
3799 adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
3800
3801 mmio_start = pci_resource_start(pdev, 0);
3802 mmio_len = pci_resource_len(pdev, 0);
3803
3804 err = -EIO;
3805 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
3806 if (!adapter->hw.hw_addr)
3807 goto err_ioremap;
3808
3809 if ((adapter->flags & FLAG_HAS_FLASH) &&
3810 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
3811 flash_start = pci_resource_start(pdev, 1);
3812 flash_len = pci_resource_len(pdev, 1);
3813 adapter->hw.flash_address = ioremap(flash_start, flash_len);
3814 if (!adapter->hw.flash_address)
3815 goto err_flashmap;
3816 }
3817
3818 /* construct the net_device struct */
3819 netdev->open = &e1000_open;
3820 netdev->stop = &e1000_close;
3821 netdev->hard_start_xmit = &e1000_xmit_frame;
3822 netdev->get_stats = &e1000_get_stats;
3823 netdev->set_multicast_list = &e1000_set_multi;
3824 netdev->set_mac_address = &e1000_set_mac;
3825 netdev->change_mtu = &e1000_change_mtu;
3826 netdev->do_ioctl = &e1000_ioctl;
3827 e1000e_set_ethtool_ops(netdev);
3828 netdev->tx_timeout = &e1000_tx_timeout;
3829 netdev->watchdog_timeo = 5 * HZ;
3830 netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
3831 netdev->vlan_rx_register = e1000_vlan_rx_register;
3832 netdev->vlan_rx_add_vid = e1000_vlan_rx_add_vid;
3833 netdev->vlan_rx_kill_vid = e1000_vlan_rx_kill_vid;
3834#ifdef CONFIG_NET_POLL_CONTROLLER
3835 netdev->poll_controller = e1000_netpoll;
3836#endif
3837 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
3838
3839 netdev->mem_start = mmio_start;
3840 netdev->mem_end = mmio_start + mmio_len;
3841
3842 adapter->bd_number = cards_found++;
3843
3844 /* setup adapter struct */
3845 err = e1000_sw_init(adapter);
3846 if (err)
3847 goto err_sw_init;
3848
3849 err = -EIO;
3850
3851 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
3852 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
3853 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
3854
3855 err = ei->get_invariants(adapter);
3856 if (err)
3857 goto err_hw_init;
3858
3859 hw->mac.ops.get_bus_info(&adapter->hw);
3860
3861 adapter->hw.phy.wait_for_link = 0;
3862
3863 /* Copper options */
3864 if (adapter->hw.media_type == e1000_media_type_copper) {
3865 adapter->hw.phy.mdix = AUTO_ALL_MODES;
3866 adapter->hw.phy.disable_polarity_correction = 0;
3867 adapter->hw.phy.ms_type = e1000_ms_hw_default;
3868 }
3869
3870 if (e1000_check_reset_block(&adapter->hw))
3871 ndev_info(netdev,
3872 "PHY reset is blocked due to SOL/IDER session.\n");
3873
3874 netdev->features = NETIF_F_SG |
3875 NETIF_F_HW_CSUM |
3876 NETIF_F_HW_VLAN_TX |
3877 NETIF_F_HW_VLAN_RX;
3878
3879 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
3880 netdev->features |= NETIF_F_HW_VLAN_FILTER;
3881
3882 netdev->features |= NETIF_F_TSO;
3883 netdev->features |= NETIF_F_TSO6;
3884
3885 if (pci_using_dac)
3886 netdev->features |= NETIF_F_HIGHDMA;
3887
3888 /* We should not be using LLTX anymore, but we are still TX faster with
3889 * it. */
3890 netdev->features |= NETIF_F_LLTX;
3891
3892 if (e1000e_enable_mng_pass_thru(&adapter->hw))
3893 adapter->flags |= FLAG_MNG_PT_ENABLED;
3894
3895 /* before reading the NVM, reset the controller to
3896 * put the device in a known good starting state */
3897 adapter->hw.mac.ops.reset_hw(&adapter->hw);
3898
3899 /*
3900 * systems with ASPM and others may see the checksum fail on the first
3901 * attempt. Let's give it a few tries
3902 */
3903 for (i = 0;; i++) {
3904 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
3905 break;
3906 if (i == 2) {
3907 ndev_err(netdev, "The NVM Checksum Is Not Valid\n");
3908 err = -EIO;
3909 goto err_eeprom;
3910 }
3911 }
3912
3913 /* copy the MAC address out of the NVM */
3914 if (e1000e_read_mac_addr(&adapter->hw))
3915 ndev_err(netdev, "NVM Read Error while reading MAC address\n");
3916
3917 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
3918 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
3919
3920 if (!is_valid_ether_addr(netdev->perm_addr)) {
3921 ndev_err(netdev, "Invalid MAC Address: "
3922 "%02x:%02x:%02x:%02x:%02x:%02x\n",
3923 netdev->perm_addr[0], netdev->perm_addr[1],
3924 netdev->perm_addr[2], netdev->perm_addr[3],
3925 netdev->perm_addr[4], netdev->perm_addr[5]);
3926 err = -EIO;
3927 goto err_eeprom;
3928 }
3929
3930 init_timer(&adapter->watchdog_timer);
3931 adapter->watchdog_timer.function = &e1000_watchdog;
3932 adapter->watchdog_timer.data = (unsigned long) adapter;
3933
3934 init_timer(&adapter->phy_info_timer);
3935 adapter->phy_info_timer.function = &e1000_update_phy_info;
3936 adapter->phy_info_timer.data = (unsigned long) adapter;
3937
3938 INIT_WORK(&adapter->reset_task, e1000_reset_task);
3939 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
3940
3941 e1000e_check_options(adapter);
3942
3943 /* Initialize link parameters. User can change them with ethtool */
3944 adapter->hw.mac.autoneg = 1;
309af40b 3945 adapter->fc_autoneg = 1;
bc7f75fa
AK		 3946 adapter->hw.mac.original_fc = e1000_fc_default;
3947 adapter->hw.mac.fc = e1000_fc_default;
3948 adapter->hw.phy.autoneg_advertised = 0x2f;
3949
3950 /* ring size defaults */
3951 adapter->rx_ring->count = 256;
3952 adapter->tx_ring->count = 256;
3953
3954 /*
3955 * Initial Wake on LAN setting - If APM wake is enabled in
3956 * the EEPROM, enable the ACPI Magic Packet filter
3957 */
3958 if (adapter->flags & FLAG_APME_IN_WUC) {
3959 /* APME bit in EEPROM is mapped to WUC.APME */
3960 eeprom_data = er32(WUC);
3961 eeprom_apme_mask = E1000_WUC_APME;
3962 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
3963 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
3964 (adapter->hw.bus.func == 1))
3965 e1000_read_nvm(&adapter->hw,
3966 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
3967 else
3968 e1000_read_nvm(&adapter->hw,
3969 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
3970 }
3971
3972 /* fetch WoL from EEPROM */
3973 if (eeprom_data & eeprom_apme_mask)
3974 adapter->eeprom_wol |= E1000_WUFC_MAG;
3975
3976 /*
3977 * now that we have the eeprom settings, apply the special cases
3978 * where the eeprom may be wrong or the board simply won't support
3979 * wake on lan on a particular port
3980 */
3981 if (!(adapter->flags & FLAG_HAS_WOL))
3982 adapter->eeprom_wol = 0;
3983
3984 /* initialize the wol settings based on the eeprom settings */
3985 adapter->wol = adapter->eeprom_wol;
3986
3987 /* reset the hardware with the new settings */
3988 e1000e_reset(adapter);
3989
3990 /* If the controller has AMT, do not set DRV_LOAD until the interface
3991 * is up. For all other cases, let the f/w know that the h/w is now
3992 * under the control of the driver. */
3993 if (!(adapter->flags & FLAG_HAS_AMT) ||
3994 !e1000e_check_mng_mode(&adapter->hw))
3995 e1000_get_hw_control(adapter);
3996
3997 /* tell the stack to leave us alone until e1000_open() is called */
3998 netif_carrier_off(netdev);
3999 netif_stop_queue(netdev);
4000
4001 strcpy(netdev->name, "eth%d");
4002 err = register_netdev(netdev);
4003 if (err)
4004 goto err_register;
4005
4006 e1000_print_device_info(adapter);
4007
4008 return 0;
4009
4010err_register:
4011err_hw_init:
4012 e1000_release_hw_control(adapter);
4013err_eeprom:
4014 if (!e1000_check_reset_block(&adapter->hw))
4015 e1000_phy_hw_reset(&adapter->hw);
4016
4017 if (adapter->hw.flash_address)
4018 iounmap(adapter->hw.flash_address);
4019
4020err_flashmap:
4021 kfree(adapter->tx_ring);
4022 kfree(adapter->rx_ring);
4023err_sw_init:
4024 iounmap(adapter->hw.hw_addr);
4025err_ioremap:
4026 free_netdev(netdev);
4027err_alloc_etherdev:
4028 pci_release_regions(pdev);
4029err_pci_reg:
4030err_dma:
4031 pci_disable_device(pdev);
4032 return err;
4033}
4034
4035/**
4036 * e1000_remove - Device Removal Routine
4037 * @pdev: PCI device information struct
4038 *
4039 * e1000_remove is called by the PCI subsystem to alert the driver
4040 * that it should release a PCI device. The could be caused by a
4041 * Hot-Plug event, or because the driver is going to be removed from
4042 * memory.
4043 **/
4044static void __devexit e1000_remove(struct pci_dev *pdev)
4045{
4046 struct net_device *netdev = pci_get_drvdata(pdev);
4047 struct e1000_adapter *adapter = netdev_priv(netdev);
4048
4049 /* flush_scheduled work may reschedule our watchdog task, so
4050 * explicitly disable watchdog tasks from being rescheduled */
4051 set_bit(__E1000_DOWN, &adapter->state);
4052 del_timer_sync(&adapter->watchdog_timer);
4053 del_timer_sync(&adapter->phy_info_timer);
4054
4055 flush_scheduled_work();
4056
4057 e1000_release_manageability(adapter);
4058
4059 /* Release control of h/w to f/w. If f/w is AMT enabled, this
4060 * would have already happened in close and is redundant. */
4061 e1000_release_hw_control(adapter);
4062
4063 unregister_netdev(netdev);
4064
4065 if (!e1000_check_reset_block(&adapter->hw))
4066 e1000_phy_hw_reset(&adapter->hw);
4067
4068 kfree(adapter->tx_ring);
4069 kfree(adapter->rx_ring);
4070
4071 iounmap(adapter->hw.hw_addr);
4072 if (adapter->hw.flash_address)
4073 iounmap(adapter->hw.flash_address);
4074 pci_release_regions(pdev);
4075
4076 free_netdev(netdev);
4077
4078 pci_disable_device(pdev);
4079}
4080
4081/* PCI Error Recovery (ERS) */
4082static struct pci_error_handlers e1000_err_handler = {
4083 .error_detected = e1000_io_error_detected,
4084 .slot_reset = e1000_io_slot_reset,
4085 .resume = e1000_io_resume,
4086};
4087
4088static struct pci_device_id e1000_pci_tbl[] = {
bc7f75fa
AK		 4089 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
4090 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
4091 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
4092 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
4093 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
4094 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
040babf9
AK		 4095 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
4096 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
4097 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
bc7f75fa
AK		 4098 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
4099 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
4100 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
4101 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
4102 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
4103 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
4104 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
4105 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
4106 board_80003es2lan },
4107 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
4108 board_80003es2lan },
4109 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
4110 board_80003es2lan },
4111 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
4112 board_80003es2lan },
4113 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
4114 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
4115 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
4116 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
4117 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
4118 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
4119 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
bc7f75fa
AK		 4120 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
4121 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
4122 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
4123 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
4124 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
4125
4126 { } /* terminate list */
4127};
4128MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
4129
4130/* PCI Device API Driver */
4131static struct pci_driver e1000_driver = {
4132 .name = e1000e_driver_name,
4133 .id_table = e1000_pci_tbl,
4134 .probe = e1000_probe,
4135 .remove = __devexit_p(e1000_remove),
4136#ifdef CONFIG_PM
4137 /* Power Managment Hooks */
4138 .suspend = e1000_suspend,
4139 .resume = e1000_resume,
4140#endif
4141 .shutdown = e1000_shutdown,
4142 .err_handler = &e1000_err_handler
4143};
4144
4145/**
4146 * e1000_init_module - Driver Registration Routine
4147 *
4148 * e1000_init_module is the first routine called when the driver is
4149 * loaded. All it does is register with the PCI subsystem.
4150 **/
4151static int __init e1000_init_module(void)
4152{
4153 int ret;
4154 printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
4155 e1000e_driver_name, e1000e_driver_version);
4156 printk(KERN_INFO "%s: Copyright (c) 1999-2007 Intel Corporation.\n",
4157 e1000e_driver_name);
4158 ret = pci_register_driver(&e1000_driver);
4159
4160 return ret;
4161}
4162module_init(e1000_init_module);
4163
4164/**
4165 * e1000_exit_module - Driver Exit Cleanup Routine
4166 *
4167 * e1000_exit_module is called just before the driver is removed
4168 * from memory.
4169 **/
4170static void __exit e1000_exit_module(void)
4171{
4172 pci_unregister_driver(&e1000_driver);
4173}
4174module_exit(e1000_exit_module);
4175
4176
4177MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
4178MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
4179MODULE_LICENSE("GPL");
4180MODULE_VERSION(DRV_VERSION);
4181
4182/* e1000_main.c */
Clone of davem's net-next-2.6 tree