]> 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
e1000e: fix flow control denial of service possibility
[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
bc7f75fa
AK		 1058/**
1059 * @e1000_alloc_ring - allocate memory for a ring structure
1060 **/
1061static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1062 struct e1000_ring *ring)
1063{
1064 struct pci_dev *pdev = adapter->pdev;
1065
1066 ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1067 GFP_KERNEL);
1068 if (!ring->desc)
1069 return -ENOMEM;
1070
1071 return 0;
1072}
1073
1074/**
1075 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1076 * @adapter: board private structure
1077 *
1078 * Return 0 on success, negative on failure
1079 **/
1080int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1081{
1082 struct e1000_ring *tx_ring = adapter->tx_ring;
1083 int err = -ENOMEM, size;
1084
1085 size = sizeof(struct e1000_buffer) * tx_ring->count;
1086 tx_ring->buffer_info = vmalloc(size);
1087 if (!tx_ring->buffer_info)
1088 goto err;
1089 memset(tx_ring->buffer_info, 0, size);
1090
1091 /* round up to nearest 4K */
1092 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1093 tx_ring->size = ALIGN(tx_ring->size, 4096);
1094
1095 err = e1000_alloc_ring_dma(adapter, tx_ring);
1096 if (err)
1097 goto err;
1098
1099 tx_ring->next_to_use = 0;
1100 tx_ring->next_to_clean = 0;
1101 spin_lock_init(&adapter->tx_queue_lock);
1102
1103 return 0;
1104err:
1105 vfree(tx_ring->buffer_info);
1106 ndev_err(adapter->netdev,
1107 "Unable to allocate memory for the transmit descriptor ring\n");
1108 return err;
1109}
1110
1111/**
1112 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1113 * @adapter: board private structure
1114 *
1115 * Returns 0 on success, negative on failure
1116 **/
1117int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1118{
1119 struct e1000_ring *rx_ring = adapter->rx_ring;
47f44e40
AK		 1120 struct e1000_buffer *buffer_info;
1121 int i, size, desc_len, err = -ENOMEM;
bc7f75fa
AK		 1122
1123 size = sizeof(struct e1000_buffer) * rx_ring->count;
1124 rx_ring->buffer_info = vmalloc(size);
1125 if (!rx_ring->buffer_info)
1126 goto err;
1127 memset(rx_ring->buffer_info, 0, size);
1128
47f44e40
AK		 1129 for (i = 0; i < rx_ring->count; i++) {
1130 buffer_info = &rx_ring->buffer_info[i];
1131 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1132 sizeof(struct e1000_ps_page),
1133 GFP_KERNEL);
1134 if (!buffer_info->ps_pages)
1135 goto err_pages;
1136 }
bc7f75fa
AK		 1137
1138 desc_len = sizeof(union e1000_rx_desc_packet_split);
1139
1140 /* Round up to nearest 4K */
1141 rx_ring->size = rx_ring->count * desc_len;
1142 rx_ring->size = ALIGN(rx_ring->size, 4096);
1143
1144 err = e1000_alloc_ring_dma(adapter, rx_ring);
1145 if (err)
47f44e40 1146 goto err_pages;
bc7f75fa
AK		 1147
1148 rx_ring->next_to_clean = 0;
1149 rx_ring->next_to_use = 0;
1150 rx_ring->rx_skb_top = NULL;
1151
1152 return 0;
47f44e40
AK		 1153
1154err_pages:
1155 for (i = 0; i < rx_ring->count; i++) {
1156 buffer_info = &rx_ring->buffer_info[i];
1157 kfree(buffer_info->ps_pages);
1158 }
bc7f75fa
AK		 1159err:
1160 vfree(rx_ring->buffer_info);
bc7f75fa
AK		 1161 ndev_err(adapter->netdev,
1162 "Unable to allocate memory for the transmit descriptor ring\n");
1163 return err;
1164}
1165
1166/**
1167 * e1000_clean_tx_ring - Free Tx Buffers
1168 * @adapter: board private structure
1169 **/
1170static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1171{
1172 struct e1000_ring *tx_ring = adapter->tx_ring;
1173 struct e1000_buffer *buffer_info;
1174 unsigned long size;
1175 unsigned int i;
1176
1177 for (i = 0; i < tx_ring->count; i++) {
1178 buffer_info = &tx_ring->buffer_info[i];
1179 e1000_put_txbuf(adapter, buffer_info);
1180 }
1181
1182 size = sizeof(struct e1000_buffer) * tx_ring->count;
1183 memset(tx_ring->buffer_info, 0, size);
1184
1185 memset(tx_ring->desc, 0, tx_ring->size);
1186
1187 tx_ring->next_to_use = 0;
1188 tx_ring->next_to_clean = 0;
1189
1190 writel(0, adapter->hw.hw_addr + tx_ring->head);
1191 writel(0, adapter->hw.hw_addr + tx_ring->tail);
1192}
1193
1194/**
1195 * e1000e_free_tx_resources - Free Tx Resources per Queue
1196 * @adapter: board private structure
1197 *
1198 * Free all transmit software resources
1199 **/
1200void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1201{
1202 struct pci_dev *pdev = adapter->pdev;
1203 struct e1000_ring *tx_ring = adapter->tx_ring;
1204
1205 e1000_clean_tx_ring(adapter);
1206
1207 vfree(tx_ring->buffer_info);
1208 tx_ring->buffer_info = NULL;
1209
1210 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1211 tx_ring->dma);
1212 tx_ring->desc = NULL;
1213}
1214
1215/**
1216 * e1000e_free_rx_resources - Free Rx Resources
1217 * @adapter: board private structure
1218 *
1219 * Free all receive software resources
1220 **/
1221
1222void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1223{
1224 struct pci_dev *pdev = adapter->pdev;
1225 struct e1000_ring *rx_ring = adapter->rx_ring;
47f44e40 1226 int i;
bc7f75fa
AK		 1227
1228 e1000_clean_rx_ring(adapter);
1229
47f44e40
AK		 1230 for (i = 0; i < rx_ring->count; i++) {
1231 kfree(rx_ring->buffer_info[i].ps_pages);
1232 }
1233
bc7f75fa
AK		 1234 vfree(rx_ring->buffer_info);
1235 rx_ring->buffer_info = NULL;
1236
bc7f75fa
AK		 1237 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1238 rx_ring->dma);
1239 rx_ring->desc = NULL;
1240}
1241
1242/**
1243 * e1000_update_itr - update the dynamic ITR value based on statistics
1244 * Stores a new ITR value based on packets and byte
1245 * counts during the last interrupt. The advantage of per interrupt
1246 * computation is faster updates and more accurate ITR for the current
1247 * traffic pattern. Constants in this function were computed
1248 * based on theoretical maximum wire speed and thresholds were set based
1249 * on testing data as well as attempting to minimize response time
1250 * while increasing bulk throughput.
1251 * this functionality is controlled by the InterruptThrottleRate module
1252 * parameter (see e1000_param.c)
1253 * @adapter: pointer to adapter
1254 * @itr_setting: current adapter->itr
1255 * @packets: the number of packets during this measurement interval
1256 * @bytes: the number of bytes during this measurement interval
1257 **/
1258static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1259 u16 itr_setting, int packets,
1260 int bytes)
1261{
1262 unsigned int retval = itr_setting;
1263
1264 if (packets == 0)
1265 goto update_itr_done;
1266
1267 switch (itr_setting) {
1268 case lowest_latency:
1269 /* handle TSO and jumbo frames */
1270 if (bytes/packets > 8000)
1271 retval = bulk_latency;
1272 else if ((packets < 5) && (bytes > 512)) {
1273 retval = low_latency;
1274 }
1275 break;
1276 case low_latency: /* 50 usec aka 20000 ints/s */
1277 if (bytes > 10000) {
1278 /* this if handles the TSO accounting */
1279 if (bytes/packets > 8000) {
1280 retval = bulk_latency;
1281 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1282 retval = bulk_latency;
1283 } else if ((packets > 35)) {
1284 retval = lowest_latency;
1285 }
1286 } else if (bytes/packets > 2000) {
1287 retval = bulk_latency;
1288 } else if (packets <= 2 && bytes < 512) {
1289 retval = lowest_latency;
1290 }
1291 break;
1292 case bulk_latency: /* 250 usec aka 4000 ints/s */
1293 if (bytes > 25000) {
1294 if (packets > 35) {
1295 retval = low_latency;
1296 }
1297 } else if (bytes < 6000) {
1298 retval = low_latency;
1299 }
1300 break;
1301 }
1302
1303update_itr_done:
1304 return retval;
1305}
1306
1307static void e1000_set_itr(struct e1000_adapter *adapter)
1308{
1309 struct e1000_hw *hw = &adapter->hw;
1310 u16 current_itr;
1311 u32 new_itr = adapter->itr;
1312
1313 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1314 if (adapter->link_speed != SPEED_1000) {
1315 current_itr = 0;
1316 new_itr = 4000;
1317 goto set_itr_now;
1318 }
1319
1320 adapter->tx_itr = e1000_update_itr(adapter,
1321 adapter->tx_itr,
1322 adapter->total_tx_packets,
1323 adapter->total_tx_bytes);
1324 /* conservative mode (itr 3) eliminates the lowest_latency setting */
1325 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1326 adapter->tx_itr = low_latency;
1327
1328 adapter->rx_itr = e1000_update_itr(adapter,
1329 adapter->rx_itr,
1330 adapter->total_rx_packets,
1331 adapter->total_rx_bytes);
1332 /* conservative mode (itr 3) eliminates the lowest_latency setting */
1333 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1334 adapter->rx_itr = low_latency;
1335
1336 current_itr = max(adapter->rx_itr, adapter->tx_itr);
1337
1338 switch (current_itr) {
1339 /* counts and packets in update_itr are dependent on these numbers */
1340 case lowest_latency:
1341 new_itr = 70000;
1342 break;
1343 case low_latency:
1344 new_itr = 20000; /* aka hwitr = ~200 */
1345 break;
1346 case bulk_latency:
1347 new_itr = 4000;
1348 break;
1349 default:
1350 break;
1351 }
1352
1353set_itr_now:
1354 if (new_itr != adapter->itr) {
1355 /* this attempts to bias the interrupt rate towards Bulk
1356 * by adding intermediate steps when interrupt rate is
1357 * increasing */
1358 new_itr = new_itr > adapter->itr ?
1359 min(adapter->itr + (new_itr >> 2), new_itr) :
1360 new_itr;
1361 adapter->itr = new_itr;
1362 ew32(ITR, 1000000000 / (new_itr * 256));
1363 }
1364}
1365
1366/**
1367 * e1000_clean - NAPI Rx polling callback
1368 * @adapter: board private structure
1369 **/
1370static int e1000_clean(struct napi_struct *napi, int budget)
1371{
1372 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1373 struct net_device *poll_dev = adapter->netdev;
d2c7ddd6 1374 int tx_cleaned = 0, work_done = 0;
bc7f75fa
AK		 1375
1376 /* Must NOT use netdev_priv macro here. */
1377 adapter = poll_dev->priv;
1378
bc7f75fa
AK		 1379 /* e1000_clean is called per-cpu. This lock protects
1380 * tx_ring from being cleaned by multiple cpus
1381 * simultaneously. A failure obtaining the lock means
1382 * tx_ring is currently being cleaned anyway. */
1383 if (spin_trylock(&adapter->tx_queue_lock)) {
d2c7ddd6 1384 tx_cleaned = e1000_clean_tx_irq(adapter);
bc7f75fa
AK		 1385 spin_unlock(&adapter->tx_queue_lock);
1386 }
1387
1388 adapter->clean_rx(adapter, &work_done, budget);
d2c7ddd6
DM		 1389
1390 if (tx_cleaned)
1391 work_done = budget;
bc7f75fa 1392
53e52c72
DM		 1393 /* If budget not fully consumed, exit the polling mode */
1394 if (work_done < budget) {
bc7f75fa
AK		 1395 if (adapter->itr_setting & 3)
1396 e1000_set_itr(adapter);
1397 netif_rx_complete(poll_dev, napi);
1398 e1000_irq_enable(adapter);
1399 }
1400
1401 return work_done;
1402}
1403
1404static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1405{
1406 struct e1000_adapter *adapter = netdev_priv(netdev);
1407 struct e1000_hw *hw = &adapter->hw;
1408 u32 vfta, index;
1409
1410 /* don't update vlan cookie if already programmed */
1411 if ((adapter->hw.mng_cookie.status &
1412 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1413 (vid == adapter->mng_vlan_id))
1414 return;
1415 /* add VID to filter table */
1416 index = (vid >> 5) & 0x7F;
1417 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1418 vfta |= (1 << (vid & 0x1F));
1419 e1000e_write_vfta(hw, index, vfta);
1420}
1421
1422static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1423{
1424 struct e1000_adapter *adapter = netdev_priv(netdev);
1425 struct e1000_hw *hw = &adapter->hw;
1426 u32 vfta, index;
1427
1428 e1000_irq_disable(adapter);
1429 vlan_group_set_device(adapter->vlgrp, vid, NULL);
1430 e1000_irq_enable(adapter);
1431
1432 if ((adapter->hw.mng_cookie.status &
1433 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1434 (vid == adapter->mng_vlan_id)) {
1435 /* release control to f/w */
1436 e1000_release_hw_control(adapter);
1437 return;
1438 }
1439
1440 /* remove VID from filter table */
1441 index = (vid >> 5) & 0x7F;
1442 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1443 vfta &= ~(1 << (vid & 0x1F));
1444 e1000e_write_vfta(hw, index, vfta);
1445}
1446
1447static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
1448{
1449 struct net_device *netdev = adapter->netdev;
1450 u16 vid = adapter->hw.mng_cookie.vlan_id;
1451 u16 old_vid = adapter->mng_vlan_id;
1452
1453 if (!adapter->vlgrp)
1454 return;
1455
1456 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
1457 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1458 if (adapter->hw.mng_cookie.status &
1459 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1460 e1000_vlan_rx_add_vid(netdev, vid);
1461 adapter->mng_vlan_id = vid;
1462 }
1463
1464 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
1465 (vid != old_vid) &&
1466 !vlan_group_get_device(adapter->vlgrp, old_vid))
1467 e1000_vlan_rx_kill_vid(netdev, old_vid);
1468 } else {
1469 adapter->mng_vlan_id = vid;
1470 }
1471}
1472
1473
1474static void e1000_vlan_rx_register(struct net_device *netdev,
1475 struct vlan_group *grp)
1476{
1477 struct e1000_adapter *adapter = netdev_priv(netdev);
1478 struct e1000_hw *hw = &adapter->hw;
1479 u32 ctrl, rctl;
1480
1481 e1000_irq_disable(adapter);
1482 adapter->vlgrp = grp;
1483
1484 if (grp) {
1485 /* enable VLAN tag insert/strip */
1486 ctrl = er32(CTRL);
1487 ctrl |= E1000_CTRL_VME;
1488 ew32(CTRL, ctrl);
1489
1490 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1491 /* enable VLAN receive filtering */
1492 rctl = er32(RCTL);
1493 rctl |= E1000_RCTL_VFE;
1494 rctl &= ~E1000_RCTL_CFIEN;
1495 ew32(RCTL, rctl);
1496 e1000_update_mng_vlan(adapter);
1497 }
1498 } else {
1499 /* disable VLAN tag insert/strip */
1500 ctrl = er32(CTRL);
1501 ctrl &= ~E1000_CTRL_VME;
1502 ew32(CTRL, ctrl);
1503
1504 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1505 /* disable VLAN filtering */
1506 rctl = er32(RCTL);
1507 rctl &= ~E1000_RCTL_VFE;
1508 ew32(RCTL, rctl);
1509 if (adapter->mng_vlan_id !=
1510 (u16)E1000_MNG_VLAN_NONE) {
1511 e1000_vlan_rx_kill_vid(netdev,
1512 adapter->mng_vlan_id);
1513 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1514 }
1515 }
1516 }
1517
1518 e1000_irq_enable(adapter);
1519}
1520
1521static void e1000_restore_vlan(struct e1000_adapter *adapter)
1522{
1523 u16 vid;
1524
1525 e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
1526
1527 if (!adapter->vlgrp)
1528 return;
1529
1530 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1531 if (!vlan_group_get_device(adapter->vlgrp, vid))
1532 continue;
1533 e1000_vlan_rx_add_vid(adapter->netdev, vid);
1534 }
1535}
1536
1537static void e1000_init_manageability(struct e1000_adapter *adapter)
1538{
1539 struct e1000_hw *hw = &adapter->hw;
1540 u32 manc, manc2h;
1541
1542 if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
1543 return;
1544
1545 manc = er32(MANC);
1546
bc7f75fa
AK		 1547 /* enable receiving management packets to the host. this will probably
1548 * generate destination unreachable messages from the host OS, but
1549 * the packets will be handled on SMBUS */
1550 manc |= E1000_MANC_EN_MNG2HOST;
1551 manc2h = er32(MANC2H);
1552#define E1000_MNG2HOST_PORT_623 (1 << 5)
1553#define E1000_MNG2HOST_PORT_664 (1 << 6)
1554 manc2h |= E1000_MNG2HOST_PORT_623;
1555 manc2h |= E1000_MNG2HOST_PORT_664;
1556 ew32(MANC2H, manc2h);
1557 ew32(MANC, manc);
1558}
1559
1560/**
1561 * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1562 * @adapter: board private structure
1563 *
1564 * Configure the Tx unit of the MAC after a reset.
1565 **/
1566static void e1000_configure_tx(struct e1000_adapter *adapter)
1567{
1568 struct e1000_hw *hw = &adapter->hw;
1569 struct e1000_ring *tx_ring = adapter->tx_ring;
1570 u64 tdba;
1571 u32 tdlen, tctl, tipg, tarc;
1572 u32 ipgr1, ipgr2;
1573
1574 /* Setup the HW Tx Head and Tail descriptor pointers */
1575 tdba = tx_ring->dma;
1576 tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
1577 ew32(TDBAL, (tdba & DMA_32BIT_MASK));
1578 ew32(TDBAH, (tdba >> 32));
1579 ew32(TDLEN, tdlen);
1580 ew32(TDH, 0);
1581 ew32(TDT, 0);
1582 tx_ring->head = E1000_TDH;
1583 tx_ring->tail = E1000_TDT;
1584
1585 /* Set the default values for the Tx Inter Packet Gap timer */
1586 tipg = DEFAULT_82543_TIPG_IPGT_COPPER; /* 8 */
1587 ipgr1 = DEFAULT_82543_TIPG_IPGR1; /* 8 */
1588 ipgr2 = DEFAULT_82543_TIPG_IPGR2; /* 6 */
1589
1590 if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
1591 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /* 7 */
1592
1593 tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1594 tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1595 ew32(TIPG, tipg);
1596
1597 /* Set the Tx Interrupt Delay register */
1598 ew32(TIDV, adapter->tx_int_delay);
1599 /* tx irq moderation */
1600 ew32(TADV, adapter->tx_abs_int_delay);
1601
1602 /* Program the Transmit Control Register */
1603 tctl = er32(TCTL);
1604 tctl &= ~E1000_TCTL_CT;
1605 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1606 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1607
1608 if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
1609 tarc = er32(TARC0);
1610 /* set the speed mode bit, we'll clear it if we're not at
1611 * gigabit link later */
1612#define SPEED_MODE_BIT (1 << 21)
1613 tarc |= SPEED_MODE_BIT;
1614 ew32(TARC0, tarc);
1615 }
1616
1617 /* errata: program both queues to unweighted RR */
1618 if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
1619 tarc = er32(TARC0);
1620 tarc |= 1;
1621 ew32(TARC0, tarc);
1622 tarc = er32(TARC1);
1623 tarc |= 1;
1624 ew32(TARC1, tarc);
1625 }
1626
1627 e1000e_config_collision_dist(hw);
1628
1629 /* Setup Transmit Descriptor Settings for eop descriptor */
1630 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1631
1632 /* only set IDE if we are delaying interrupts using the timers */
1633 if (adapter->tx_int_delay)
1634 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1635
1636 /* enable Report Status bit */
1637 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1638
1639 ew32(TCTL, tctl);
1640
1641 adapter->tx_queue_len = adapter->netdev->tx_queue_len;
1642}
1643
1644/**
1645 * e1000_setup_rctl - configure the receive control registers
1646 * @adapter: Board private structure
1647 **/
1648#define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1649 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1650static void e1000_setup_rctl(struct e1000_adapter *adapter)
1651{
1652 struct e1000_hw *hw = &adapter->hw;
1653 u32 rctl, rfctl;
1654 u32 psrctl = 0;
1655 u32 pages = 0;
1656
1657 /* Program MC offset vector base */
1658 rctl = er32(RCTL);
1659 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1660 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1661 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1662 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1663
1664 /* Do not Store bad packets */
1665 rctl &= ~E1000_RCTL_SBP;
1666
1667 /* Enable Long Packet receive */
1668 if (adapter->netdev->mtu <= ETH_DATA_LEN)
1669 rctl &= ~E1000_RCTL_LPE;
1670 else
1671 rctl |= E1000_RCTL_LPE;
1672
5918bd88
AK		 1673 /* Enable hardware CRC frame stripping */
1674 rctl |= E1000_RCTL_SECRC;
1675
bc7f75fa
AK		 1676 /* Setup buffer sizes */
1677 rctl &= ~E1000_RCTL_SZ_4096;
1678 rctl |= E1000_RCTL_BSEX;
1679 switch (adapter->rx_buffer_len) {
1680 case 256:
1681 rctl |= E1000_RCTL_SZ_256;
1682 rctl &= ~E1000_RCTL_BSEX;
1683 break;
1684 case 512:
1685 rctl |= E1000_RCTL_SZ_512;
1686 rctl &= ~E1000_RCTL_BSEX;
1687 break;
1688 case 1024:
1689 rctl |= E1000_RCTL_SZ_1024;
1690 rctl &= ~E1000_RCTL_BSEX;
1691 break;
1692 case 2048:
1693 default:
1694 rctl |= E1000_RCTL_SZ_2048;
1695 rctl &= ~E1000_RCTL_BSEX;
1696 break;
1697 case 4096:
1698 rctl |= E1000_RCTL_SZ_4096;
1699 break;
1700 case 8192:
1701 rctl |= E1000_RCTL_SZ_8192;
1702 break;
1703 case 16384:
1704 rctl |= E1000_RCTL_SZ_16384;
1705 break;
1706 }
1707
1708 /*
1709 * 82571 and greater support packet-split where the protocol
1710 * header is placed in skb->data and the packet data is
1711 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
1712 * In the case of a non-split, skb->data is linearly filled,
1713 * followed by the page buffers. Therefore, skb->data is
1714 * sized to hold the largest protocol header.
1715 *
1716 * allocations using alloc_page take too long for regular MTU
1717 * so only enable packet split for jumbo frames
1718 *
1719 * Using pages when the page size is greater than 16k wastes
1720 * a lot of memory, since we allocate 3 pages at all times
1721 * per packet.
1722 */
1723 adapter->rx_ps_pages = 0;
1724 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
1725 if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
1726 adapter->rx_ps_pages = pages;
1727
1728 if (adapter->rx_ps_pages) {
1729 /* Configure extra packet-split registers */
1730 rfctl = er32(RFCTL);
1731 rfctl |= E1000_RFCTL_EXTEN;
1732 /* disable packet split support for IPv6 extension headers,
1733 * because some malformed IPv6 headers can hang the RX */
1734 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
1735 E1000_RFCTL_NEW_IPV6_EXT_DIS);
1736
1737 ew32(RFCTL, rfctl);
1738
140a7480
AK		 1739 /* Enable Packet split descriptors */
1740 rctl |= E1000_RCTL_DTYP_PS;
bc7f75fa
AK		 1741
1742 psrctl |= adapter->rx_ps_bsize0 >>
1743 E1000_PSRCTL_BSIZE0_SHIFT;
1744
1745 switch (adapter->rx_ps_pages) {
1746 case 3:
1747 psrctl |= PAGE_SIZE <<
1748 E1000_PSRCTL_BSIZE3_SHIFT;
1749 case 2:
1750 psrctl |= PAGE_SIZE <<
1751 E1000_PSRCTL_BSIZE2_SHIFT;
1752 case 1:
1753 psrctl |= PAGE_SIZE >>
1754 E1000_PSRCTL_BSIZE1_SHIFT;
1755 break;
1756 }
1757
1758 ew32(PSRCTL, psrctl);
1759 }
1760
1761 ew32(RCTL, rctl);
1762}
1763
1764/**
1765 * e1000_configure_rx - Configure Receive Unit after Reset
1766 * @adapter: board private structure
1767 *
1768 * Configure the Rx unit of the MAC after a reset.
1769 **/
1770static void e1000_configure_rx(struct e1000_adapter *adapter)
1771{
1772 struct e1000_hw *hw = &adapter->hw;
1773 struct e1000_ring *rx_ring = adapter->rx_ring;
1774 u64 rdba;
1775 u32 rdlen, rctl, rxcsum, ctrl_ext;
1776
1777 if (adapter->rx_ps_pages) {
1778 /* this is a 32 byte descriptor */
1779 rdlen = rx_ring->count *
1780 sizeof(union e1000_rx_desc_packet_split);
1781 adapter->clean_rx = e1000_clean_rx_irq_ps;
1782 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
bc7f75fa
AK		 1783 } else {
1784 rdlen = rx_ring->count *
1785 sizeof(struct e1000_rx_desc);
1786 adapter->clean_rx = e1000_clean_rx_irq;
1787 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1788 }
1789
1790 /* disable receives while setting up the descriptors */
1791 rctl = er32(RCTL);
1792 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1793 e1e_flush();
1794 msleep(10);
1795
1796 /* set the Receive Delay Timer Register */
1797 ew32(RDTR, adapter->rx_int_delay);
1798
1799 /* irq moderation */
1800 ew32(RADV, adapter->rx_abs_int_delay);
1801 if (adapter->itr_setting != 0)
1802 ew32(ITR,
1803 1000000000 / (adapter->itr * 256));
1804
1805 ctrl_ext = er32(CTRL_EXT);
1806 /* Reset delay timers after every interrupt */
1807 ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
1808 /* Auto-Mask interrupts upon ICR access */
1809 ctrl_ext |= E1000_CTRL_EXT_IAME;
1810 ew32(IAM, 0xffffffff);
1811 ew32(CTRL_EXT, ctrl_ext);
1812 e1e_flush();
1813
1814 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1815 * the Base and Length of the Rx Descriptor Ring */
1816 rdba = rx_ring->dma;
1817 ew32(RDBAL, (rdba & DMA_32BIT_MASK));
1818 ew32(RDBAH, (rdba >> 32));
1819 ew32(RDLEN, rdlen);
1820 ew32(RDH, 0);
1821 ew32(RDT, 0);
1822 rx_ring->head = E1000_RDH;
1823 rx_ring->tail = E1000_RDT;
1824
1825 /* Enable Receive Checksum Offload for TCP and UDP */
1826 rxcsum = er32(RXCSUM);
1827 if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
1828 rxcsum |= E1000_RXCSUM_TUOFL;
1829
1830 /* IPv4 payload checksum for UDP fragments must be
1831 * used in conjunction with packet-split. */
1832 if (adapter->rx_ps_pages)
1833 rxcsum |= E1000_RXCSUM_IPPCSE;
1834 } else {
1835 rxcsum &= ~E1000_RXCSUM_TUOFL;
1836 /* no need to clear IPPCSE as it defaults to 0 */
1837 }
1838 ew32(RXCSUM, rxcsum);
1839
1840 /* Enable early receives on supported devices, only takes effect when
1841 * packet size is equal or larger than the specified value (in 8 byte
1842 * units), e.g. using jumbo frames when setting to E1000_ERT_2048 */
1843 if ((adapter->flags & FLAG_HAS_ERT) &&
1844 (adapter->netdev->mtu > ETH_DATA_LEN))
1845 ew32(ERT, E1000_ERT_2048);
1846
1847 /* Enable Receives */
1848 ew32(RCTL, rctl);
1849}
1850
1851/**
1852 * e1000_mc_addr_list_update - Update Multicast addresses
1853 * @hw: pointer to the HW structure
1854 * @mc_addr_list: array of multicast addresses to program
1855 * @mc_addr_count: number of multicast addresses to program
1856 * @rar_used_count: the first RAR register free to program
1857 * @rar_count: total number of supported Receive Address Registers
1858 *
1859 * Updates the Receive Address Registers and Multicast Table Array.
1860 * The caller must have a packed mc_addr_list of multicast addresses.
1861 * The parameter rar_count will usually be hw->mac.rar_entry_count
1862 * unless there are workarounds that change this. Currently no func pointer
1863 * exists and all implementations are handled in the generic version of this
1864 * function.
1865 **/
1866static void e1000_mc_addr_list_update(struct e1000_hw *hw, u8 *mc_addr_list,
1867 u32 mc_addr_count, u32 rar_used_count,
1868 u32 rar_count)
1869{
1870 hw->mac.ops.mc_addr_list_update(hw, mc_addr_list, mc_addr_count,
1871 rar_used_count, rar_count);
1872}
1873
1874/**
1875 * e1000_set_multi - Multicast and Promiscuous mode set
1876 * @netdev: network interface device structure
1877 *
1878 * The set_multi entry point is called whenever the multicast address
1879 * list or the network interface flags are updated. This routine is
1880 * responsible for configuring the hardware for proper multicast,
1881 * promiscuous mode, and all-multi behavior.
1882 **/
1883static void e1000_set_multi(struct net_device *netdev)
1884{
1885 struct e1000_adapter *adapter = netdev_priv(netdev);
1886 struct e1000_hw *hw = &adapter->hw;
1887 struct e1000_mac_info *mac = &hw->mac;
1888 struct dev_mc_list *mc_ptr;
1889 u8 *mta_list;
1890 u32 rctl;
1891 int i;
1892
1893 /* Check for Promiscuous and All Multicast modes */
1894
1895 rctl = er32(RCTL);
1896
1897 if (netdev->flags & IFF_PROMISC) {
1898 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
1899 } else if (netdev->flags & IFF_ALLMULTI) {
1900 rctl |= E1000_RCTL_MPE;
1901 rctl &= ~E1000_RCTL_UPE;
1902 } else {
1903 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
1904 }
1905
1906 ew32(RCTL, rctl);
1907
1908 if (netdev->mc_count) {
1909 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
1910 if (!mta_list)
1911 return;
1912
1913 /* prepare a packed array of only addresses. */
1914 mc_ptr = netdev->mc_list;
1915
1916 for (i = 0; i < netdev->mc_count; i++) {
1917 if (!mc_ptr)
1918 break;
1919 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
1920 ETH_ALEN);
1921 mc_ptr = mc_ptr->next;
1922 }
1923
1924 e1000_mc_addr_list_update(hw, mta_list, i, 1,
1925 mac->rar_entry_count);
1926 kfree(mta_list);
1927 } else {
1928 /*
1929 * if we're called from probe, we might not have
1930 * anything to do here, so clear out the list
1931 */
1932 e1000_mc_addr_list_update(hw, NULL, 0, 1,
1933 mac->rar_entry_count);
1934 }
1935}
1936
1937/**
1938 * e1000_configure - configure the hardware for RX and TX
1939 * @adapter: private board structure
1940 **/
1941static void e1000_configure(struct e1000_adapter *adapter)
1942{
1943 e1000_set_multi(adapter->netdev);
1944
1945 e1000_restore_vlan(adapter);
1946 e1000_init_manageability(adapter);
1947
1948 e1000_configure_tx(adapter);
1949 e1000_setup_rctl(adapter);
1950 e1000_configure_rx(adapter);
1951 adapter->alloc_rx_buf(adapter,
1952 e1000_desc_unused(adapter->rx_ring));
1953}
1954
1955/**
1956 * e1000e_power_up_phy - restore link in case the phy was powered down
1957 * @adapter: address of board private structure
1958 *
1959 * The phy may be powered down to save power and turn off link when the
1960 * driver is unloaded and wake on lan is not enabled (among others)
1961 * *** this routine MUST be followed by a call to e1000e_reset ***
1962 **/
1963void e1000e_power_up_phy(struct e1000_adapter *adapter)
1964{
1965 u16 mii_reg = 0;
1966
1967 /* Just clear the power down bit to wake the phy back up */
1968 if (adapter->hw.media_type == e1000_media_type_copper) {
1969 /* according to the manual, the phy will retain its
1970 * settings across a power-down/up cycle */
1971 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
1972 mii_reg &= ~MII_CR_POWER_DOWN;
1973 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
1974 }
1975
1976 adapter->hw.mac.ops.setup_link(&adapter->hw);
1977}
1978
1979/**
1980 * e1000_power_down_phy - Power down the PHY
1981 *
1982 * Power down the PHY so no link is implied when interface is down
1983 * The PHY cannot be powered down is management or WoL is active
1984 */
1985static void e1000_power_down_phy(struct e1000_adapter *adapter)
1986{
1987 struct e1000_hw *hw = &adapter->hw;
1988 u16 mii_reg;
1989
1990 /* WoL is enabled */
23b66e2b 1991 if (adapter->wol)
bc7f75fa
AK		 1992 return;
1993
1994 /* non-copper PHY? */
1995 if (adapter->hw.media_type != e1000_media_type_copper)
1996 return;
1997
1998 /* reset is blocked because of a SoL/IDER session */
1999 if (e1000e_check_mng_mode(hw) ||
2000 e1000_check_reset_block(hw))
2001 return;
2002
2003 /* managebility (AMT) is enabled */
2004 if (er32(MANC) & E1000_MANC_SMBUS_EN)
2005 return;
2006
2007 /* power down the PHY */
2008 e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2009 mii_reg |= MII_CR_POWER_DOWN;
2010 e1e_wphy(hw, PHY_CONTROL, mii_reg);
2011 mdelay(1);
2012}
2013
2014/**
2015 * e1000e_reset - bring the hardware into a known good state
2016 *
2017 * This function boots the hardware and enables some settings that
2018 * require a configuration cycle of the hardware - those cannot be
2019 * set/changed during runtime. After reset the device needs to be
2020 * properly configured for rx, tx etc.
2021 */
2022void e1000e_reset(struct e1000_adapter *adapter)
2023{
2024 struct e1000_mac_info *mac = &adapter->hw.mac;
2025 struct e1000_hw *hw = &adapter->hw;
2026 u32 tx_space, min_tx_space, min_rx_space;
df762464 2027 u32 pba;
bc7f75fa
AK		 2028 u16 hwm;
2029
df762464
AK		 2030 ew32(PBA, adapter->pba);
2031
bc7f75fa
AK		 2032 if (mac->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN ) {
2033 /* To maintain wire speed transmits, the Tx FIFO should be
2034 * large enough to accommodate two full transmit packets,
2035 * rounded up to the next 1KB and expressed in KB. Likewise,
2036 * the Rx FIFO should be large enough to accommodate at least
2037 * one full receive packet and is similarly rounded up and
2038 * expressed in KB. */
df762464 2039 pba = er32(PBA);
bc7f75fa 2040 /* upper 16 bits has Tx packet buffer allocation size in KB */
df762464 2041 tx_space = pba >> 16;
bc7f75fa 2042 /* lower 16 bits has Rx packet buffer allocation size in KB */
df762464 2043 pba &= 0xffff;
bc7f75fa
AK		 2044 /* the tx fifo also stores 16 bytes of information about the tx
2045 * but don't include ethernet FCS because hardware appends it */
2046 min_tx_space = (mac->max_frame_size +
2047 sizeof(struct e1000_tx_desc) -
2048 ETH_FCS_LEN) * 2;
2049 min_tx_space = ALIGN(min_tx_space, 1024);
2050 min_tx_space >>= 10;
2051 /* software strips receive CRC, so leave room for it */
2052 min_rx_space = mac->max_frame_size;
2053 min_rx_space = ALIGN(min_rx_space, 1024);
2054 min_rx_space >>= 10;
2055
2056 /* If current Tx allocation is less than the min Tx FIFO size,
2057 * and the min Tx FIFO size is less than the current Rx FIFO
2058 * allocation, take space away from current Rx allocation */
df762464
AK		 2059 if ((tx_space < min_tx_space) &&
2060 ((min_tx_space - tx_space) < pba)) {
2061 pba -= min_tx_space - tx_space;
bc7f75fa
AK		 2062
2063 /* if short on rx space, rx wins and must trump tx
2064 * adjustment or use Early Receive if available */
df762464 2065 if ((pba < min_rx_space) &&
bc7f75fa
AK		 2066 (!(adapter->flags & FLAG_HAS_ERT)))
2067 /* ERT enabled in e1000_configure_rx */
df762464 2068 pba = min_rx_space;
bc7f75fa 2069 }
df762464
AK		 2070
2071 ew32(PBA, pba);
bc7f75fa
AK		 2072 }
2073
bc7f75fa
AK		 2074
2075 /* flow control settings */
2076 /* The high water mark must be low enough to fit one full frame
2077 * (or the size used for early receive) above it in the Rx FIFO.
2078 * Set it to the lower of:
2079 * - 90% of the Rx FIFO size, and
2080 * - the full Rx FIFO size minus the early receive size (for parts
2081 * with ERT support assuming ERT set to E1000_ERT_2048), or
2082 * - the full Rx FIFO size minus one full frame */
2083 if (adapter->flags & FLAG_HAS_ERT)
2084 hwm = min(((adapter->pba << 10) * 9 / 10),
2085 ((adapter->pba << 10) - (E1000_ERT_2048 << 3)));
2086 else
2087 hwm = min(((adapter->pba << 10) * 9 / 10),
2088 ((adapter->pba << 10) - mac->max_frame_size));
2089
2090 mac->fc_high_water = hwm & 0xFFF8; /* 8-byte granularity */
2091 mac->fc_low_water = mac->fc_high_water - 8;
2092
2093 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2094 mac->fc_pause_time = 0xFFFF;
2095 else
2096 mac->fc_pause_time = E1000_FC_PAUSE_TIME;
2097 mac->fc = mac->original_fc;
2098
2099 /* Allow time for pending master requests to run */
2100 mac->ops.reset_hw(hw);
2101 ew32(WUC, 0);
2102
2103 if (mac->ops.init_hw(hw))
2104 ndev_err(adapter->netdev, "Hardware Error\n");
2105
2106 e1000_update_mng_vlan(adapter);
2107
2108 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2109 ew32(VET, ETH_P_8021Q);
2110
2111 e1000e_reset_adaptive(hw);
2112 e1000_get_phy_info(hw);
2113
2114 if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2115 u16 phy_data = 0;
2116 /* speed up time to link by disabling smart power down, ignore
2117 * the return value of this function because there is nothing
2118 * different we would do if it failed */
2119 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2120 phy_data &= ~IGP02E1000_PM_SPD;
2121 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2122 }
bc7f75fa
AK		 2123}
2124
2125int e1000e_up(struct e1000_adapter *adapter)
2126{
2127 struct e1000_hw *hw = &adapter->hw;
2128
2129 /* hardware has been reset, we need to reload some things */
2130 e1000_configure(adapter);
2131
2132 clear_bit(__E1000_DOWN, &adapter->state);
2133
2134 napi_enable(&adapter->napi);
2135 e1000_irq_enable(adapter);
2136
2137 /* fire a link change interrupt to start the watchdog */
2138 ew32(ICS, E1000_ICS_LSC);
2139 return 0;
2140}
2141
2142void e1000e_down(struct e1000_adapter *adapter)
2143{
2144 struct net_device *netdev = adapter->netdev;
2145 struct e1000_hw *hw = &adapter->hw;
2146 u32 tctl, rctl;
2147
2148 /* signal that we're down so the interrupt handler does not
2149 * reschedule our watchdog timer */
2150 set_bit(__E1000_DOWN, &adapter->state);
2151
2152 /* disable receives in the hardware */
2153 rctl = er32(RCTL);
2154 ew32(RCTL, rctl & ~E1000_RCTL_EN);
2155 /* flush and sleep below */
2156
2157 netif_stop_queue(netdev);
2158
2159 /* disable transmits in the hardware */
2160 tctl = er32(TCTL);
2161 tctl &= ~E1000_TCTL_EN;
2162 ew32(TCTL, tctl);
2163 /* flush both disables and wait for them to finish */
2164 e1e_flush();
2165 msleep(10);
2166
2167 napi_disable(&adapter->napi);
49d85c50 2168 atomic_set(&adapter->irq_sem, 0);
bc7f75fa
AK		 2169 e1000_irq_disable(adapter);
2170
2171 del_timer_sync(&adapter->watchdog_timer);
2172 del_timer_sync(&adapter->phy_info_timer);
2173
2174 netdev->tx_queue_len = adapter->tx_queue_len;
2175 netif_carrier_off(netdev);
2176 adapter->link_speed = 0;
2177 adapter->link_duplex = 0;
2178
2179 e1000e_reset(adapter);
2180 e1000_clean_tx_ring(adapter);
2181 e1000_clean_rx_ring(adapter);
2182
2183 /*
2184 * TODO: for power management, we could drop the link and
2185 * pci_disable_device here.
2186 */
2187}
2188
2189void e1000e_reinit_locked(struct e1000_adapter *adapter)
2190{
2191 might_sleep();
2192 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2193 msleep(1);
2194 e1000e_down(adapter);
2195 e1000e_up(adapter);
2196 clear_bit(__E1000_RESETTING, &adapter->state);
2197}
2198
2199/**
2200 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2201 * @adapter: board private structure to initialize
2202 *
2203 * e1000_sw_init initializes the Adapter private data structure.
2204 * Fields are initialized based on PCI device information and
2205 * OS network device settings (MTU size).
2206 **/
2207static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2208{
2209 struct e1000_hw *hw = &adapter->hw;
2210 struct net_device *netdev = adapter->netdev;
2211
2212 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2213 adapter->rx_ps_bsize0 = 128;
2214 hw->mac.max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2215 hw->mac.min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2216
2217 adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2218 if (!adapter->tx_ring)
2219 goto err;
2220
2221 adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2222 if (!adapter->rx_ring)
2223 goto err;
2224
2225 spin_lock_init(&adapter->tx_queue_lock);
2226
2227 /* Explicitly disable IRQ since the NIC can be in any state. */
2228 atomic_set(&adapter->irq_sem, 0);
2229 e1000_irq_disable(adapter);
2230
2231 spin_lock_init(&adapter->stats_lock);
2232
2233 set_bit(__E1000_DOWN, &adapter->state);
2234 return 0;
2235
2236err:
2237 ndev_err(netdev, "Unable to allocate memory for queues\n");
2238 kfree(adapter->rx_ring);
2239 kfree(adapter->tx_ring);
2240 return -ENOMEM;
2241}
2242
2243/**
2244 * e1000_open - Called when a network interface is made active
2245 * @netdev: network interface device structure
2246 *
2247 * Returns 0 on success, negative value on failure
2248 *
2249 * The open entry point is called when a network interface is made
2250 * active by the system (IFF_UP). At this point all resources needed
2251 * for transmit and receive operations are allocated, the interrupt
2252 * handler is registered with the OS, the watchdog timer is started,
2253 * and the stack is notified that the interface is ready.
2254 **/
2255static int e1000_open(struct net_device *netdev)
2256{
2257 struct e1000_adapter *adapter = netdev_priv(netdev);
2258 struct e1000_hw *hw = &adapter->hw;
2259 int err;
2260
2261 /* disallow open during test */
2262 if (test_bit(__E1000_TESTING, &adapter->state))
2263 return -EBUSY;
2264
2265 /* allocate transmit descriptors */
2266 err = e1000e_setup_tx_resources(adapter);
2267 if (err)
2268 goto err_setup_tx;
2269
2270 /* allocate receive descriptors */
2271 err = e1000e_setup_rx_resources(adapter);
2272 if (err)
2273 goto err_setup_rx;
2274
2275 e1000e_power_up_phy(adapter);
2276
2277 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2278 if ((adapter->hw.mng_cookie.status &
2279 E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
2280 e1000_update_mng_vlan(adapter);
2281
2282 /* If AMT is enabled, let the firmware know that the network
2283 * interface is now open */
2284 if ((adapter->flags & FLAG_HAS_AMT) &&
2285 e1000e_check_mng_mode(&adapter->hw))
2286 e1000_get_hw_control(adapter);
2287
2288 /* before we allocate an interrupt, we must be ready to handle it.
2289 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2290 * as soon as we call pci_request_irq, so we have to setup our
2291 * clean_rx handler before we do so. */
2292 e1000_configure(adapter);
2293
2294 err = e1000_request_irq(adapter);
2295 if (err)
2296 goto err_req_irq;
2297
2298 /* From here on the code is the same as e1000e_up() */
2299 clear_bit(__E1000_DOWN, &adapter->state);
2300
2301 napi_enable(&adapter->napi);
2302
2303 e1000_irq_enable(adapter);
2304
2305 /* fire a link status change interrupt to start the watchdog */
2306 ew32(ICS, E1000_ICS_LSC);
2307
2308 return 0;
2309
2310err_req_irq:
2311 e1000_release_hw_control(adapter);
2312 e1000_power_down_phy(adapter);
2313 e1000e_free_rx_resources(adapter);
2314err_setup_rx:
2315 e1000e_free_tx_resources(adapter);
2316err_setup_tx:
2317 e1000e_reset(adapter);
2318
2319 return err;
2320}
2321
2322/**
2323 * e1000_close - Disables a network interface
2324 * @netdev: network interface device structure
2325 *
2326 * Returns 0, this is not allowed to fail
2327 *
2328 * The close entry point is called when an interface is de-activated
2329 * by the OS. The hardware is still under the drivers control, but
2330 * needs to be disabled. A global MAC reset is issued to stop the
2331 * hardware, and all transmit and receive resources are freed.
2332 **/
2333static int e1000_close(struct net_device *netdev)
2334{
2335 struct e1000_adapter *adapter = netdev_priv(netdev);
2336
2337 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
2338 e1000e_down(adapter);
2339 e1000_power_down_phy(adapter);
2340 e1000_free_irq(adapter);
2341
2342 e1000e_free_tx_resources(adapter);
2343 e1000e_free_rx_resources(adapter);
2344
2345 /* kill manageability vlan ID if supported, but not if a vlan with
2346 * the same ID is registered on the host OS (let 8021q kill it) */
2347 if ((adapter->hw.mng_cookie.status &
2348 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2349 !(adapter->vlgrp &&
2350 vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
2351 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2352
2353 /* If AMT is enabled, let the firmware know that the network
2354 * interface is now closed */
2355 if ((adapter->flags & FLAG_HAS_AMT) &&
2356 e1000e_check_mng_mode(&adapter->hw))
2357 e1000_release_hw_control(adapter);
2358
2359 return 0;
2360}
2361/**
2362 * e1000_set_mac - Change the Ethernet Address of the NIC
2363 * @netdev: network interface device structure
2364 * @p: pointer to an address structure
2365 *
2366 * Returns 0 on success, negative on failure
2367 **/
2368static int e1000_set_mac(struct net_device *netdev, void *p)
2369{
2370 struct e1000_adapter *adapter = netdev_priv(netdev);
2371 struct sockaddr *addr = p;
2372
2373 if (!is_valid_ether_addr(addr->sa_data))
2374 return -EADDRNOTAVAIL;
2375
2376 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2377 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
2378
2379 e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
2380
2381 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
2382 /* activate the work around */
2383 e1000e_set_laa_state_82571(&adapter->hw, 1);
2384
2385 /* Hold a copy of the LAA in RAR[14] This is done so that
2386 * between the time RAR[0] gets clobbered and the time it
2387 * gets fixed (in e1000_watchdog), the actual LAA is in one
2388 * of the RARs and no incoming packets directed to this port
2389 * are dropped. Eventually the LAA will be in RAR[0] and
2390 * RAR[14] */
2391 e1000e_rar_set(&adapter->hw,
2392 adapter->hw.mac.addr,
2393 adapter->hw.mac.rar_entry_count - 1);
2394 }
2395
2396 return 0;
2397}
2398
2399/* Need to wait a few seconds after link up to get diagnostic information from
2400 * the phy */
2401static void e1000_update_phy_info(unsigned long data)
2402{
2403 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2404 e1000_get_phy_info(&adapter->hw);
2405}
2406
2407/**
2408 * e1000e_update_stats - Update the board statistics counters
2409 * @adapter: board private structure
2410 **/
2411void e1000e_update_stats(struct e1000_adapter *adapter)
2412{
2413 struct e1000_hw *hw = &adapter->hw;
2414 struct pci_dev *pdev = adapter->pdev;
2415 unsigned long irq_flags;
2416 u16 phy_tmp;
2417
2418#define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
2419
2420 /*
2421 * Prevent stats update while adapter is being reset, or if the pci
2422 * connection is down.
2423 */
2424 if (adapter->link_speed == 0)
2425 return;
2426 if (pci_channel_offline(pdev))
2427 return;
2428
2429 spin_lock_irqsave(&adapter->stats_lock, irq_flags);
2430
2431 /* these counters are modified from e1000_adjust_tbi_stats,
2432 * called from the interrupt context, so they must only
2433 * be written while holding adapter->stats_lock
2434 */
2435
2436 adapter->stats.crcerrs += er32(CRCERRS);
2437 adapter->stats.gprc += er32(GPRC);
2438 adapter->stats.gorcl += er32(GORCL);
2439 adapter->stats.gorch += er32(GORCH);
2440 adapter->stats.bprc += er32(BPRC);
2441 adapter->stats.mprc += er32(MPRC);
2442 adapter->stats.roc += er32(ROC);
2443
2444 if (adapter->flags & FLAG_HAS_STATS_PTC_PRC) {
2445 adapter->stats.prc64 += er32(PRC64);
2446 adapter->stats.prc127 += er32(PRC127);
2447 adapter->stats.prc255 += er32(PRC255);
2448 adapter->stats.prc511 += er32(PRC511);
2449 adapter->stats.prc1023 += er32(PRC1023);
2450 adapter->stats.prc1522 += er32(PRC1522);
2451 adapter->stats.symerrs += er32(SYMERRS);
2452 adapter->stats.sec += er32(SEC);
2453 }
2454
2455 adapter->stats.mpc += er32(MPC);
2456 adapter->stats.scc += er32(SCC);
2457 adapter->stats.ecol += er32(ECOL);
2458 adapter->stats.mcc += er32(MCC);
2459 adapter->stats.latecol += er32(LATECOL);
2460 adapter->stats.dc += er32(DC);
2461 adapter->stats.rlec += er32(RLEC);
2462 adapter->stats.xonrxc += er32(XONRXC);
2463 adapter->stats.xontxc += er32(XONTXC);
2464 adapter->stats.xoffrxc += er32(XOFFRXC);
2465 adapter->stats.xofftxc += er32(XOFFTXC);
2466 adapter->stats.fcruc += er32(FCRUC);
2467 adapter->stats.gptc += er32(GPTC);
2468 adapter->stats.gotcl += er32(GOTCL);
2469 adapter->stats.gotch += er32(GOTCH);
2470 adapter->stats.rnbc += er32(RNBC);
2471 adapter->stats.ruc += er32(RUC);
2472 adapter->stats.rfc += er32(RFC);
2473 adapter->stats.rjc += er32(RJC);
2474 adapter->stats.torl += er32(TORL);
2475 adapter->stats.torh += er32(TORH);
2476 adapter->stats.totl += er32(TOTL);
2477 adapter->stats.toth += er32(TOTH);
2478 adapter->stats.tpr += er32(TPR);
2479
2480 if (adapter->flags & FLAG_HAS_STATS_PTC_PRC) {
2481 adapter->stats.ptc64 += er32(PTC64);
2482 adapter->stats.ptc127 += er32(PTC127);
2483 adapter->stats.ptc255 += er32(PTC255);
2484 adapter->stats.ptc511 += er32(PTC511);
2485 adapter->stats.ptc1023 += er32(PTC1023);
2486 adapter->stats.ptc1522 += er32(PTC1522);
2487 }
2488
2489 adapter->stats.mptc += er32(MPTC);
2490 adapter->stats.bptc += er32(BPTC);
2491
2492 /* used for adaptive IFS */
2493
2494 hw->mac.tx_packet_delta = er32(TPT);
2495 adapter->stats.tpt += hw->mac.tx_packet_delta;
2496 hw->mac.collision_delta = er32(COLC);
2497 adapter->stats.colc += hw->mac.collision_delta;
2498
2499 adapter->stats.algnerrc += er32(ALGNERRC);
2500 adapter->stats.rxerrc += er32(RXERRC);
2501 adapter->stats.tncrs += er32(TNCRS);
2502 adapter->stats.cexterr += er32(CEXTERR);
2503 adapter->stats.tsctc += er32(TSCTC);
2504 adapter->stats.tsctfc += er32(TSCTFC);
2505
2506 adapter->stats.iac += er32(IAC);
2507
2508 if (adapter->flags & FLAG_HAS_STATS_ICR_ICT) {
2509 adapter->stats.icrxoc += er32(ICRXOC);
2510 adapter->stats.icrxptc += er32(ICRXPTC);
2511 adapter->stats.icrxatc += er32(ICRXATC);
2512 adapter->stats.ictxptc += er32(ICTXPTC);
2513 adapter->stats.ictxatc += er32(ICTXATC);
2514 adapter->stats.ictxqec += er32(ICTXQEC);
2515 adapter->stats.ictxqmtc += er32(ICTXQMTC);
2516 adapter->stats.icrxdmtc += er32(ICRXDMTC);
2517 }
2518
2519 /* Fill out the OS statistics structure */
bc7f75fa
AK		 2520 adapter->net_stats.multicast = adapter->stats.mprc;
2521 adapter->net_stats.collisions = adapter->stats.colc;
2522
2523 /* Rx Errors */
2524
2525 /* RLEC on some newer hardware can be incorrect so build
2526 * our own version based on RUC and ROC */
2527 adapter->net_stats.rx_errors = adapter->stats.rxerrc +
2528 adapter->stats.crcerrs + adapter->stats.algnerrc +
2529 adapter->stats.ruc + adapter->stats.roc +
2530 adapter->stats.cexterr;
2531 adapter->net_stats.rx_length_errors = adapter->stats.ruc +
2532 adapter->stats.roc;
2533 adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
2534 adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
2535 adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
2536
2537 /* Tx Errors */
2538 adapter->net_stats.tx_errors = adapter->stats.ecol +
2539 adapter->stats.latecol;
2540 adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
2541 adapter->net_stats.tx_window_errors = adapter->stats.latecol;
2542 adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
2543
2544 /* Tx Dropped needs to be maintained elsewhere */
2545
2546 /* Phy Stats */
2547 if (hw->media_type == e1000_media_type_copper) {
2548 if ((adapter->link_speed == SPEED_1000) &&
2549 (!e1e_rphy(hw, PHY_1000T_STATUS, &phy_tmp))) {
2550 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
2551 adapter->phy_stats.idle_errors += phy_tmp;
2552 }
2553 }
2554
2555 /* Management Stats */
2556 adapter->stats.mgptc += er32(MGTPTC);
2557 adapter->stats.mgprc += er32(MGTPRC);
2558 adapter->stats.mgpdc += er32(MGTPDC);
2559
2560 spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
2561}
2562
2563static void e1000_print_link_info(struct e1000_adapter *adapter)
2564{
2565 struct net_device *netdev = adapter->netdev;
2566 struct e1000_hw *hw = &adapter->hw;
2567 u32 ctrl = er32(CTRL);
2568
2569 ndev_info(netdev,
2570 "Link is Up %d Mbps %s, Flow Control: %s\n",
2571 adapter->link_speed,
2572 (adapter->link_duplex == FULL_DUPLEX) ?
2573 "Full Duplex" : "Half Duplex",
2574 ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
2575 "RX/TX" :
2576 ((ctrl & E1000_CTRL_RFCE) ? "RX" :
2577 ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
2578}
2579
2580/**
2581 * e1000_watchdog - Timer Call-back
2582 * @data: pointer to adapter cast into an unsigned long
2583 **/
2584static void e1000_watchdog(unsigned long data)
2585{
2586 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2587
2588 /* Do the rest outside of interrupt context */
2589 schedule_work(&adapter->watchdog_task);
2590
2591 /* TODO: make this use queue_delayed_work() */
2592}
2593
2594static void e1000_watchdog_task(struct work_struct *work)
2595{
2596 struct e1000_adapter *adapter = container_of(work,
2597 struct e1000_adapter, watchdog_task);
2598
2599 struct net_device *netdev = adapter->netdev;
2600 struct e1000_mac_info *mac = &adapter->hw.mac;
2601 struct e1000_ring *tx_ring = adapter->tx_ring;
2602 struct e1000_hw *hw = &adapter->hw;
2603 u32 link, tctl;
2604 s32 ret_val;
2605 int tx_pending = 0;
2606
2607 if ((netif_carrier_ok(netdev)) &&
2608 (er32(STATUS) & E1000_STATUS_LU))
2609 goto link_up;
2610
2611 ret_val = mac->ops.check_for_link(hw);
2612 if ((ret_val == E1000_ERR_PHY) &&
2613 (adapter->hw.phy.type == e1000_phy_igp_3) &&
2614 (er32(CTRL) &
2615 E1000_PHY_CTRL_GBE_DISABLE)) {
2616 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
2617 ndev_info(netdev,
2618 "Gigabit has been disabled, downgrading speed\n");
2619 }
2620
2621 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
2622 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
2623 e1000_update_mng_vlan(adapter);
2624
2625 if ((adapter->hw.media_type == e1000_media_type_internal_serdes) &&
2626 !(er32(TXCW) & E1000_TXCW_ANE))
2627 link = adapter->hw.mac.serdes_has_link;
2628 else
2629 link = er32(STATUS) & E1000_STATUS_LU;
2630
2631 if (link) {
2632 if (!netif_carrier_ok(netdev)) {
2633 bool txb2b = 1;
2634 mac->ops.get_link_up_info(&adapter->hw,
2635 &adapter->link_speed,
2636 &adapter->link_duplex);
2637 e1000_print_link_info(adapter);
2638 /* tweak tx_queue_len according to speed/duplex
2639 * and adjust the timeout factor */
2640 netdev->tx_queue_len = adapter->tx_queue_len;
2641 adapter->tx_timeout_factor = 1;
2642 switch (adapter->link_speed) {
2643 case SPEED_10:
2644 txb2b = 0;
2645 netdev->tx_queue_len = 10;
2646 adapter->tx_timeout_factor = 14;
2647 break;
2648 case SPEED_100:
2649 txb2b = 0;
2650 netdev->tx_queue_len = 100;
2651 /* maybe add some timeout factor ? */
2652 break;
2653 }
2654
2655 /* workaround: re-program speed mode bit after
2656 * link-up event */
2657 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
2658 !txb2b) {
2659 u32 tarc0;
2660 tarc0 = er32(TARC0);
2661 tarc0 &= ~SPEED_MODE_BIT;
2662 ew32(TARC0, tarc0);
2663 }
2664
2665 /* disable TSO for pcie and 10/100 speeds, to avoid
2666 * some hardware issues */
2667 if (!(adapter->flags & FLAG_TSO_FORCE)) {
2668 switch (adapter->link_speed) {
2669 case SPEED_10:
2670 case SPEED_100:
2671 ndev_info(netdev,
2672 "10/100 speed: disabling TSO\n");
2673 netdev->features &= ~NETIF_F_TSO;
2674 netdev->features &= ~NETIF_F_TSO6;
2675 break;
2676 case SPEED_1000:
2677 netdev->features |= NETIF_F_TSO;
2678 netdev->features |= NETIF_F_TSO6;
2679 break;
2680 default:
2681 /* oops */
2682 break;
2683 }
2684 }
2685
2686 /* enable transmits in the hardware, need to do this
2687 * after setting TARC0 */
2688 tctl = er32(TCTL);
2689 tctl |= E1000_TCTL_EN;
2690 ew32(TCTL, tctl);
2691
2692 netif_carrier_on(netdev);
2693 netif_wake_queue(netdev);
2694
2695 if (!test_bit(__E1000_DOWN, &adapter->state))
2696 mod_timer(&adapter->phy_info_timer,
2697 round_jiffies(jiffies + 2 * HZ));
2698 } else {
2699 /* make sure the receive unit is started */
2700 if (adapter->flags & FLAG_RX_NEEDS_RESTART) {
2701 u32 rctl = er32(RCTL);
2702 ew32(RCTL, rctl |
2703 E1000_RCTL_EN);
2704 }
2705 }
2706 } else {
2707 if (netif_carrier_ok(netdev)) {
2708 adapter->link_speed = 0;
2709 adapter->link_duplex = 0;
2710 ndev_info(netdev, "Link is Down\n");
2711 netif_carrier_off(netdev);
2712 netif_stop_queue(netdev);
2713 if (!test_bit(__E1000_DOWN, &adapter->state))
2714 mod_timer(&adapter->phy_info_timer,
2715 round_jiffies(jiffies + 2 * HZ));
2716
2717 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
2718 schedule_work(&adapter->reset_task);
2719 }
2720 }
2721
2722link_up:
2723 e1000e_update_stats(adapter);
2724
2725 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2726 adapter->tpt_old = adapter->stats.tpt;
2727 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
2728 adapter->colc_old = adapter->stats.colc;
2729
2730 adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2731 adapter->gorcl_old = adapter->stats.gorcl;
2732 adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2733 adapter->gotcl_old = adapter->stats.gotcl;
2734
2735 e1000e_update_adaptive(&adapter->hw);
2736
2737 if (!netif_carrier_ok(netdev)) {
2738 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
2739 tx_ring->count);
2740 if (tx_pending) {
2741 /* We've lost link, so the controller stops DMA,
2742 * but we've got queued Tx work that's never going
2743 * to get done, so reset controller to flush Tx.
2744 * (Do the reset outside of interrupt context). */
2745 adapter->tx_timeout_count++;
2746 schedule_work(&adapter->reset_task);
2747 }
2748 }
2749
2750 /* Cause software interrupt to ensure rx ring is cleaned */
2751 ew32(ICS, E1000_ICS_RXDMT0);
2752
2753 /* Force detection of hung controller every watchdog period */
2754 adapter->detect_tx_hung = 1;
2755
2756 /* With 82571 controllers, LAA may be overwritten due to controller
2757 * reset from the other port. Set the appropriate LAA in RAR[0] */
2758 if (e1000e_get_laa_state_82571(hw))
2759 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
2760
2761 /* Reset the timer */
2762 if (!test_bit(__E1000_DOWN, &adapter->state))
2763 mod_timer(&adapter->watchdog_timer,
2764 round_jiffies(jiffies + 2 * HZ));
2765}
2766
2767#define E1000_TX_FLAGS_CSUM 0x00000001
2768#define E1000_TX_FLAGS_VLAN 0x00000002
2769#define E1000_TX_FLAGS_TSO 0x00000004
2770#define E1000_TX_FLAGS_IPV4 0x00000008
2771#define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
2772#define E1000_TX_FLAGS_VLAN_SHIFT 16
2773
2774static int e1000_tso(struct e1000_adapter *adapter,
2775 struct sk_buff *skb)
2776{
2777 struct e1000_ring *tx_ring = adapter->tx_ring;
2778 struct e1000_context_desc *context_desc;
2779 struct e1000_buffer *buffer_info;
2780 unsigned int i;
2781 u32 cmd_length = 0;
2782 u16 ipcse = 0, tucse, mss;
2783 u8 ipcss, ipcso, tucss, tucso, hdr_len;
2784 int err;
2785
2786 if (skb_is_gso(skb)) {
2787 if (skb_header_cloned(skb)) {
2788 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2789 if (err)
2790 return err;
2791 }
2792
2793 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2794 mss = skb_shinfo(skb)->gso_size;
2795 if (skb->protocol == htons(ETH_P_IP)) {
2796 struct iphdr *iph = ip_hdr(skb);
2797 iph->tot_len = 0;
2798 iph->check = 0;
2799 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2800 iph->daddr, 0,
2801 IPPROTO_TCP,
2802 0);
2803 cmd_length = E1000_TXD_CMD_IP;
2804 ipcse = skb_transport_offset(skb) - 1;
2805 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
2806 ipv6_hdr(skb)->payload_len = 0;
2807 tcp_hdr(skb)->check =
2808 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2809 &ipv6_hdr(skb)->daddr,
2810 0, IPPROTO_TCP, 0);
2811 ipcse = 0;
2812 }
2813 ipcss = skb_network_offset(skb);
2814 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2815 tucss = skb_transport_offset(skb);
2816 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2817 tucse = 0;
2818
2819 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2820 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2821
2822 i = tx_ring->next_to_use;
2823 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2824 buffer_info = &tx_ring->buffer_info[i];
2825
2826 context_desc->lower_setup.ip_fields.ipcss = ipcss;
2827 context_desc->lower_setup.ip_fields.ipcso = ipcso;
2828 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
2829 context_desc->upper_setup.tcp_fields.tucss = tucss;
2830 context_desc->upper_setup.tcp_fields.tucso = tucso;
2831 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2832 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
2833 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2834 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2835
2836 buffer_info->time_stamp = jiffies;
2837 buffer_info->next_to_watch = i;
2838
2839 i++;
2840 if (i == tx_ring->count)
2841 i = 0;
2842 tx_ring->next_to_use = i;
2843
2844 return 1;
2845 }
2846
2847 return 0;
2848}
2849
2850static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
2851{
2852 struct e1000_ring *tx_ring = adapter->tx_ring;
2853 struct e1000_context_desc *context_desc;
2854 struct e1000_buffer *buffer_info;
2855 unsigned int i;
2856 u8 css;
2857
2858 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2859 css = skb_transport_offset(skb);
2860
2861 i = tx_ring->next_to_use;
2862 buffer_info = &tx_ring->buffer_info[i];
2863 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2864
2865 context_desc->lower_setup.ip_config = 0;
2866 context_desc->upper_setup.tcp_fields.tucss = css;
2867 context_desc->upper_setup.tcp_fields.tucso =
2868 css + skb->csum_offset;
2869 context_desc->upper_setup.tcp_fields.tucse = 0;
2870 context_desc->tcp_seg_setup.data = 0;
2871 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
2872
2873 buffer_info->time_stamp = jiffies;
2874 buffer_info->next_to_watch = i;
2875
2876 i++;
2877 if (i == tx_ring->count)
2878 i = 0;
2879 tx_ring->next_to_use = i;
2880
2881 return 1;
2882 }
2883
2884 return 0;
2885}
2886
2887#define E1000_MAX_PER_TXD 8192
2888#define E1000_MAX_TXD_PWR 12
2889
2890static int e1000_tx_map(struct e1000_adapter *adapter,
2891 struct sk_buff *skb, unsigned int first,
2892 unsigned int max_per_txd, unsigned int nr_frags,
2893 unsigned int mss)
2894{
2895 struct e1000_ring *tx_ring = adapter->tx_ring;
2896 struct e1000_buffer *buffer_info;
2897 unsigned int len = skb->len - skb->data_len;
2898 unsigned int offset = 0, size, count = 0, i;
2899 unsigned int f;
2900
2901 i = tx_ring->next_to_use;
2902
2903 while (len) {
2904 buffer_info = &tx_ring->buffer_info[i];
2905 size = min(len, max_per_txd);
2906
2907 /* Workaround for premature desc write-backs
2908 * in TSO mode. Append 4-byte sentinel desc */
2909 if (mss && !nr_frags && size == len && size > 8)
2910 size -= 4;
2911
2912 buffer_info->length = size;
2913 /* set time_stamp *before* dma to help avoid a possible race */
2914 buffer_info->time_stamp = jiffies;
2915 buffer_info->dma =
2916 pci_map_single(adapter->pdev,
2917 skb->data + offset,
2918 size,
2919 PCI_DMA_TODEVICE);
2920 if (pci_dma_mapping_error(buffer_info->dma)) {
2921 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
2922 adapter->tx_dma_failed++;
2923 return -1;
2924 }
2925 buffer_info->next_to_watch = i;
2926
2927 len -= size;
2928 offset += size;
2929 count++;
2930 i++;
2931 if (i == tx_ring->count)
2932 i = 0;
2933 }
2934
2935 for (f = 0; f < nr_frags; f++) {
2936 struct skb_frag_struct *frag;
2937
2938 frag = &skb_shinfo(skb)->frags[f];
2939 len = frag->size;
2940 offset = frag->page_offset;
2941
2942 while (len) {
2943 buffer_info = &tx_ring->buffer_info[i];
2944 size = min(len, max_per_txd);
2945 /* Workaround for premature desc write-backs
2946 * in TSO mode. Append 4-byte sentinel desc */
2947 if (mss && f == (nr_frags-1) && size == len && size > 8)
2948 size -= 4;
2949
2950 buffer_info->length = size;
2951 buffer_info->time_stamp = jiffies;
2952 buffer_info->dma =
2953 pci_map_page(adapter->pdev,
2954 frag->page,
2955 offset,
2956 size,
2957 PCI_DMA_TODEVICE);
2958 if (pci_dma_mapping_error(buffer_info->dma)) {
2959 dev_err(&adapter->pdev->dev,
2960 "TX DMA page map failed\n");
2961 adapter->tx_dma_failed++;
2962 return -1;
2963 }
2964
2965 buffer_info->next_to_watch = i;
2966
2967 len -= size;
2968 offset += size;
2969 count++;
2970
2971 i++;
2972 if (i == tx_ring->count)
2973 i = 0;
2974 }
2975 }
2976
2977 if (i == 0)
2978 i = tx_ring->count - 1;
2979 else
2980 i--;
2981
2982 tx_ring->buffer_info[i].skb = skb;
2983 tx_ring->buffer_info[first].next_to_watch = i;
2984
2985 return count;
2986}
2987
2988static void e1000_tx_queue(struct e1000_adapter *adapter,
2989 int tx_flags, int count)
2990{
2991 struct e1000_ring *tx_ring = adapter->tx_ring;
2992 struct e1000_tx_desc *tx_desc = NULL;
2993 struct e1000_buffer *buffer_info;
2994 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
2995 unsigned int i;
2996
2997 if (tx_flags & E1000_TX_FLAGS_TSO) {
2998 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
2999 E1000_TXD_CMD_TSE;
3000 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3001
3002 if (tx_flags & E1000_TX_FLAGS_IPV4)
3003 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3004 }
3005
3006 if (tx_flags & E1000_TX_FLAGS_CSUM) {
3007 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3008 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3009 }
3010
3011 if (tx_flags & E1000_TX_FLAGS_VLAN) {
3012 txd_lower |= E1000_TXD_CMD_VLE;
3013 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3014 }
3015
3016 i = tx_ring->next_to_use;
3017
3018 while (count--) {
3019 buffer_info = &tx_ring->buffer_info[i];
3020 tx_desc = E1000_TX_DESC(*tx_ring, i);
3021 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3022 tx_desc->lower.data =
3023 cpu_to_le32(txd_lower | buffer_info->length);
3024 tx_desc->upper.data = cpu_to_le32(txd_upper);
3025
3026 i++;
3027 if (i == tx_ring->count)
3028 i = 0;
3029 }
3030
3031 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3032
3033 /* Force memory writes to complete before letting h/w
3034 * know there are new descriptors to fetch. (Only
3035 * applicable for weak-ordered memory model archs,
3036 * such as IA-64). */
3037 wmb();
3038
3039 tx_ring->next_to_use = i;
3040 writel(i, adapter->hw.hw_addr + tx_ring->tail);
3041 /* we need this if more than one processor can write to our tail
3042 * at a time, it synchronizes IO on IA64/Altix systems */
3043 mmiowb();
3044}
3045
3046#define MINIMUM_DHCP_PACKET_SIZE 282
3047static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3048 struct sk_buff *skb)
3049{
3050 struct e1000_hw *hw = &adapter->hw;
3051 u16 length, offset;
3052
3053 if (vlan_tx_tag_present(skb)) {
3054 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3055 && (adapter->hw.mng_cookie.status &
3056 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3057 return 0;
3058 }
3059
3060 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3061 return 0;
3062
3063 if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3064 return 0;
3065
3066 {
3067 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3068 struct udphdr *udp;
3069
3070 if (ip->protocol != IPPROTO_UDP)
3071 return 0;
3072
3073 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3074 if (ntohs(udp->dest) != 67)
3075 return 0;
3076
3077 offset = (u8 *)udp + 8 - skb->data;
3078 length = skb->len - offset;
3079 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
3080 }
3081
3082 return 0;
3083}
3084
3085static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3086{
3087 struct e1000_adapter *adapter = netdev_priv(netdev);
3088
3089 netif_stop_queue(netdev);
3090 /* Herbert's original patch had:
3091 * smp_mb__after_netif_stop_queue();
3092 * but since that doesn't exist yet, just open code it. */
3093 smp_mb();
3094
3095 /* We need to check again in a case another CPU has just
3096 * made room available. */
3097 if (e1000_desc_unused(adapter->tx_ring) < size)
3098 return -EBUSY;
3099
3100 /* A reprieve! */
3101 netif_start_queue(netdev);
3102 ++adapter->restart_queue;
3103 return 0;
3104}
3105
3106static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
3107{
3108 struct e1000_adapter *adapter = netdev_priv(netdev);
3109
3110 if (e1000_desc_unused(adapter->tx_ring) >= size)
3111 return 0;
3112 return __e1000_maybe_stop_tx(netdev, size);
3113}
3114
3115#define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3116static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3117{
3118 struct e1000_adapter *adapter = netdev_priv(netdev);
3119 struct e1000_ring *tx_ring = adapter->tx_ring;
3120 unsigned int first;
3121 unsigned int max_per_txd = E1000_MAX_PER_TXD;
3122 unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3123 unsigned int tx_flags = 0;
4e6c709c 3124 unsigned int len = skb->len - skb->data_len;
bc7f75fa 3125 unsigned long irq_flags;
4e6c709c
AK		 3126 unsigned int nr_frags;
3127 unsigned int mss;
bc7f75fa
AK		 3128 int count = 0;
3129 int tso;
3130 unsigned int f;
bc7f75fa
AK		 3131
3132 if (test_bit(__E1000_DOWN, &adapter->state)) {
3133 dev_kfree_skb_any(skb);
3134 return NETDEV_TX_OK;
3135 }
3136
3137 if (skb->len <= 0) {
3138 dev_kfree_skb_any(skb);
3139 return NETDEV_TX_OK;
3140 }
3141
3142 mss = skb_shinfo(skb)->gso_size;
3143 /* The controller does a simple calculation to
3144 * make sure there is enough room in the FIFO before
3145 * initiating the DMA for each buffer. The calc is:
3146 * 4 = ceil(buffer len/mss). To make sure we don't
3147 * overrun the FIFO, adjust the max buffer len if mss
3148 * drops. */
3149 if (mss) {
3150 u8 hdr_len;
3151 max_per_txd = min(mss << 2, max_per_txd);
3152 max_txd_pwr = fls(max_per_txd) - 1;
3153
3154 /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
3155 * points to just header, pull a few bytes of payload from
3156 * frags into skb->data */
3157 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4e6c709c 3158 if (skb->data_len && (hdr_len == len)) {
bc7f75fa
AK		 3159 unsigned int pull_size;
3160
3161 pull_size = min((unsigned int)4, skb->data_len);
3162 if (!__pskb_pull_tail(skb, pull_size)) {
3163 ndev_err(netdev,
3164 "__pskb_pull_tail failed.\n");
3165 dev_kfree_skb_any(skb);
3166 return NETDEV_TX_OK;
3167 }
3168 len = skb->len - skb->data_len;
3169 }
3170 }
3171
3172 /* reserve a descriptor for the offload context */
3173 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3174 count++;
3175 count++;
3176
3177 count += TXD_USE_COUNT(len, max_txd_pwr);
3178
3179 nr_frags = skb_shinfo(skb)->nr_frags;
3180 for (f = 0; f < nr_frags; f++)
3181 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3182 max_txd_pwr);
3183
3184 if (adapter->hw.mac.tx_pkt_filtering)
3185 e1000_transfer_dhcp_info(adapter, skb);
3186
3187 if (!spin_trylock_irqsave(&adapter->tx_queue_lock, irq_flags))
3188 /* Collision - tell upper layer to requeue */
3189 return NETDEV_TX_LOCKED;
3190
3191 /* need: count + 2 desc gap to keep tail from touching
3192 * head, otherwise try next time */
3193 if (e1000_maybe_stop_tx(netdev, count + 2)) {
3194 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3195 return NETDEV_TX_BUSY;
3196 }
3197
3198 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3199 tx_flags |= E1000_TX_FLAGS_VLAN;
3200 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3201 }
3202
3203 first = tx_ring->next_to_use;
3204
3205 tso = e1000_tso(adapter, skb);
3206 if (tso < 0) {
3207 dev_kfree_skb_any(skb);
3208 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3209 return NETDEV_TX_OK;
3210 }
3211
3212 if (tso)
3213 tx_flags |= E1000_TX_FLAGS_TSO;
3214 else if (e1000_tx_csum(adapter, skb))
3215 tx_flags |= E1000_TX_FLAGS_CSUM;
3216
3217 /* Old method was to assume IPv4 packet by default if TSO was enabled.
3218 * 82571 hardware supports TSO capabilities for IPv6 as well...
3219 * no longer assume, we must. */
3220 if (skb->protocol == htons(ETH_P_IP))
3221 tx_flags |= E1000_TX_FLAGS_IPV4;
3222
3223 count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
3224 if (count < 0) {
3225 /* handle pci_map_single() error in e1000_tx_map */
3226 dev_kfree_skb_any(skb);
3227 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
7b5dfe1a 3228 return NETDEV_TX_OK;
bc7f75fa
AK		 3229 }
3230
3231 e1000_tx_queue(adapter, tx_flags, count);
3232
3233 netdev->trans_start = jiffies;
3234
3235 /* Make sure there is space in the ring for the next send. */
3236 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
3237
3238 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3239 return NETDEV_TX_OK;
3240}
3241
3242/**
3243 * e1000_tx_timeout - Respond to a Tx Hang
3244 * @netdev: network interface device structure
3245 **/
3246static void e1000_tx_timeout(struct net_device *netdev)
3247{
3248 struct e1000_adapter *adapter = netdev_priv(netdev);
3249
3250 /* Do the reset outside of interrupt context */
3251 adapter->tx_timeout_count++;
3252 schedule_work(&adapter->reset_task);
3253}
3254
3255static void e1000_reset_task(struct work_struct *work)
3256{
3257 struct e1000_adapter *adapter;
3258 adapter = container_of(work, struct e1000_adapter, reset_task);
3259
3260 e1000e_reinit_locked(adapter);
3261}
3262
3263/**
3264 * e1000_get_stats - Get System Network Statistics
3265 * @netdev: network interface device structure
3266 *
3267 * Returns the address of the device statistics structure.
3268 * The statistics are actually updated from the timer callback.
3269 **/
3270static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3271{
3272 struct e1000_adapter *adapter = netdev_priv(netdev);
3273
3274 /* only return the current stats */
3275 return &adapter->net_stats;
3276}
3277
3278/**
3279 * e1000_change_mtu - Change the Maximum Transfer Unit
3280 * @netdev: network interface device structure
3281 * @new_mtu: new value for maximum frame size
3282 *
3283 * Returns 0 on success, negative on failure
3284 **/
3285static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3286{
3287 struct e1000_adapter *adapter = netdev_priv(netdev);
3288 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3289
3290 if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
3291 (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3292 ndev_err(netdev, "Invalid MTU setting\n");
3293 return -EINVAL;
3294 }
3295
3296 /* Jumbo frame size limits */
3297 if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
3298 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
3299 ndev_err(netdev, "Jumbo Frames not supported.\n");
3300 return -EINVAL;
3301 }
3302 if (adapter->hw.phy.type == e1000_phy_ife) {
3303 ndev_err(netdev, "Jumbo Frames not supported.\n");
3304 return -EINVAL;
3305 }
3306 }
3307
3308#define MAX_STD_JUMBO_FRAME_SIZE 9234
3309 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3310 ndev_err(netdev, "MTU > 9216 not supported.\n");
3311 return -EINVAL;
3312 }
3313
3314 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3315 msleep(1);
3316 /* e1000e_down has a dependency on max_frame_size */
3317 adapter->hw.mac.max_frame_size = max_frame;
3318 if (netif_running(netdev))
3319 e1000e_down(adapter);
3320
3321 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3322 * means we reserve 2 more, this pushes us to allocate from the next
3323 * larger slab size.
f920c186 3324 * i.e. RXBUFFER_2048 --> size-4096 slab */
bc7f75fa
AK		 3325
3326 if (max_frame <= 256)
3327 adapter->rx_buffer_len = 256;
3328 else if (max_frame <= 512)
3329 adapter->rx_buffer_len = 512;
3330 else if (max_frame <= 1024)
3331 adapter->rx_buffer_len = 1024;
3332 else if (max_frame <= 2048)
3333 adapter->rx_buffer_len = 2048;
3334 else
3335 adapter->rx_buffer_len = 4096;
3336
3337 /* adjust allocation if LPE protects us, and we aren't using SBP */
3338 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3339 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
3340 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
3341 + ETH_FCS_LEN ;
3342
3343 ndev_info(netdev, "changing MTU from %d to %d\n",
3344 netdev->mtu, new_mtu);
3345 netdev->mtu = new_mtu;
3346
3347 if (netif_running(netdev))
3348 e1000e_up(adapter);
3349 else
3350 e1000e_reset(adapter);
3351
3352 clear_bit(__E1000_RESETTING, &adapter->state);
3353
3354 return 0;
3355}
3356
3357static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
3358 int cmd)
3359{
3360 struct e1000_adapter *adapter = netdev_priv(netdev);
3361 struct mii_ioctl_data *data = if_mii(ifr);
3362 unsigned long irq_flags;
3363
3364 if (adapter->hw.media_type != e1000_media_type_copper)
3365 return -EOPNOTSUPP;
3366
3367 switch (cmd) {
3368 case SIOCGMIIPHY:
3369 data->phy_id = adapter->hw.phy.addr;
3370 break;
3371 case SIOCGMIIREG:
3372 if (!capable(CAP_NET_ADMIN))
3373 return -EPERM;
3374 spin_lock_irqsave(&adapter->stats_lock, irq_flags);
3375 if (e1e_rphy(&adapter->hw, data->reg_num & 0x1F,
3376 &data->val_out)) {
3377 spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
3378 return -EIO;
3379 }
3380 spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
3381 break;
3382 case SIOCSMIIREG:
3383 default:
3384 return -EOPNOTSUPP;
3385 }
3386 return 0;
3387}
3388
3389static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
3390{
3391 switch (cmd) {
3392 case SIOCGMIIPHY:
3393 case SIOCGMIIREG:
3394 case SIOCSMIIREG:
3395 return e1000_mii_ioctl(netdev, ifr, cmd);
3396 default:
3397 return -EOPNOTSUPP;
3398 }
3399}
3400
3401static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
3402{
3403 struct net_device *netdev = pci_get_drvdata(pdev);
3404 struct e1000_adapter *adapter = netdev_priv(netdev);
3405 struct e1000_hw *hw = &adapter->hw;
3406 u32 ctrl, ctrl_ext, rctl, status;
3407 u32 wufc = adapter->wol;
3408 int retval = 0;
3409
3410 netif_device_detach(netdev);
3411
3412 if (netif_running(netdev)) {
3413 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3414 e1000e_down(adapter);
3415 e1000_free_irq(adapter);
3416 }
3417
3418 retval = pci_save_state(pdev);
3419 if (retval)
3420 return retval;
3421
3422 status = er32(STATUS);
3423 if (status & E1000_STATUS_LU)
3424 wufc &= ~E1000_WUFC_LNKC;
3425
3426 if (wufc) {
3427 e1000_setup_rctl(adapter);
3428 e1000_set_multi(netdev);
3429
3430 /* turn on all-multi mode if wake on multicast is enabled */
3431 if (wufc & E1000_WUFC_MC) {
3432 rctl = er32(RCTL);
3433 rctl |= E1000_RCTL_MPE;
3434 ew32(RCTL, rctl);
3435 }
3436
3437 ctrl = er32(CTRL);
3438 /* advertise wake from D3Cold */
3439 #define E1000_CTRL_ADVD3WUC 0x00100000
3440 /* phy power management enable */
3441 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
3442 ctrl |= E1000_CTRL_ADVD3WUC |
3443 E1000_CTRL_EN_PHY_PWR_MGMT;
3444 ew32(CTRL, ctrl);
3445
3446 if (adapter->hw.media_type == e1000_media_type_fiber ||
3447 adapter->hw.media_type == e1000_media_type_internal_serdes) {
3448 /* keep the laser running in D3 */
3449 ctrl_ext = er32(CTRL_EXT);
3450 ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
3451 ew32(CTRL_EXT, ctrl_ext);
3452 }
3453
3454 /* Allow time for pending master requests to run */
3455 e1000e_disable_pcie_master(&adapter->hw);
3456
3457 ew32(WUC, E1000_WUC_PME_EN);
3458 ew32(WUFC, wufc);
3459 pci_enable_wake(pdev, PCI_D3hot, 1);
3460 pci_enable_wake(pdev, PCI_D3cold, 1);
3461 } else {
3462 ew32(WUC, 0);
3463 ew32(WUFC, 0);
3464 pci_enable_wake(pdev, PCI_D3hot, 0);
3465 pci_enable_wake(pdev, PCI_D3cold, 0);
3466 }
3467
bc7f75fa
AK		 3468 /* make sure adapter isn't asleep if manageability is enabled */
3469 if (adapter->flags & FLAG_MNG_PT_ENABLED) {
3470 pci_enable_wake(pdev, PCI_D3hot, 1);
3471 pci_enable_wake(pdev, PCI_D3cold, 1);
3472 }
3473
3474 if (adapter->hw.phy.type == e1000_phy_igp_3)
3475 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
3476
3477 /* Release control of h/w to f/w. If f/w is AMT enabled, this
3478 * would have already happened in close and is redundant. */
3479 e1000_release_hw_control(adapter);
3480
3481 pci_disable_device(pdev);
3482
3483 pci_set_power_state(pdev, pci_choose_state(pdev, state));
3484
3485 return 0;
3486}
3487
1eae4eb2
AK		 3488static void e1000e_disable_l1aspm(struct pci_dev *pdev)
3489{
3490 int pos;
3491 u32 cap;
3492 u16 val;
3493
3494 /*
3495 * 82573 workaround - disable L1 ASPM on mobile chipsets
3496 *
3497 * L1 ASPM on various mobile (ich7) chipsets do not behave properly
3498 * resulting in lost data or garbage information on the pci-e link
3499 * level. This could result in (false) bad EEPROM checksum errors,
3500 * long ping times (up to 2s) or even a system freeze/hang.
3501 *
3502 * Unfortunately this feature saves about 1W power consumption when
3503 * active.
3504 */
3505 pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
3506 pci_read_config_dword(pdev, pos + PCI_EXP_LNKCAP, &cap);
3507 pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
3508 if (val & 0x2) {
3509 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
3510 val &= ~0x2;
3511 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
3512 }
3513}
3514
bc7f75fa
AK		 3515#ifdef CONFIG_PM
3516static int e1000_resume(struct pci_dev *pdev)
3517{
3518 struct net_device *netdev = pci_get_drvdata(pdev);
3519 struct e1000_adapter *adapter = netdev_priv(netdev);
3520 struct e1000_hw *hw = &adapter->hw;
3521 u32 err;
3522
3523 pci_set_power_state(pdev, PCI_D0);
3524 pci_restore_state(pdev);
1eae4eb2 3525 e1000e_disable_l1aspm(pdev);
bc7f75fa
AK		 3526 err = pci_enable_device(pdev);
3527 if (err) {
3528 dev_err(&pdev->dev,
3529 "Cannot enable PCI device from suspend\n");
3530 return err;
3531 }
3532
3533 pci_set_master(pdev);
3534
3535 pci_enable_wake(pdev, PCI_D3hot, 0);
3536 pci_enable_wake(pdev, PCI_D3cold, 0);
3537
3538 if (netif_running(netdev)) {
3539 err = e1000_request_irq(adapter);
3540 if (err)
3541 return err;
3542 }
3543
3544 e1000e_power_up_phy(adapter);
3545 e1000e_reset(adapter);
3546 ew32(WUS, ~0);
3547
3548 e1000_init_manageability(adapter);
3549
3550 if (netif_running(netdev))
3551 e1000e_up(adapter);
3552
3553 netif_device_attach(netdev);
3554
3555 /* If the controller has AMT, do not set DRV_LOAD until the interface
3556 * is up. For all other cases, let the f/w know that the h/w is now
3557 * under the control of the driver. */
3558 if (!(adapter->flags & FLAG_HAS_AMT) || !e1000e_check_mng_mode(&adapter->hw))
3559 e1000_get_hw_control(adapter);
3560
3561 return 0;
3562}
3563#endif
3564
3565static void e1000_shutdown(struct pci_dev *pdev)
3566{
3567 e1000_suspend(pdev, PMSG_SUSPEND);
3568}
3569
3570#ifdef CONFIG_NET_POLL_CONTROLLER
3571/*
3572 * Polling 'interrupt' - used by things like netconsole to send skbs
3573 * without having to re-enable interrupts. It's not called while
3574 * the interrupt routine is executing.
3575 */
3576static void e1000_netpoll(struct net_device *netdev)
3577{
3578 struct e1000_adapter *adapter = netdev_priv(netdev);
3579
3580 disable_irq(adapter->pdev->irq);
3581 e1000_intr(adapter->pdev->irq, netdev);
3582
3583 e1000_clean_tx_irq(adapter);
3584
3585 enable_irq(adapter->pdev->irq);
3586}
3587#endif
3588
3589/**
3590 * e1000_io_error_detected - called when PCI error is detected
3591 * @pdev: Pointer to PCI device
3592 * @state: The current pci connection state
3593 *
3594 * This function is called after a PCI bus error affecting
3595 * this device has been detected.
3596 */
3597static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
3598 pci_channel_state_t state)
3599{
3600 struct net_device *netdev = pci_get_drvdata(pdev);
3601 struct e1000_adapter *adapter = netdev_priv(netdev);
3602
3603 netif_device_detach(netdev);
3604
3605 if (netif_running(netdev))
3606 e1000e_down(adapter);
3607 pci_disable_device(pdev);
3608
3609 /* Request a slot slot reset. */
3610 return PCI_ERS_RESULT_NEED_RESET;
3611}
3612
3613/**
3614 * e1000_io_slot_reset - called after the pci bus has been reset.
3615 * @pdev: Pointer to PCI device
3616 *
3617 * Restart the card from scratch, as if from a cold-boot. Implementation
3618 * resembles the first-half of the e1000_resume routine.
3619 */
3620static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
3621{
3622 struct net_device *netdev = pci_get_drvdata(pdev);
3623 struct e1000_adapter *adapter = netdev_priv(netdev);
3624 struct e1000_hw *hw = &adapter->hw;
3625
1eae4eb2 3626 e1000e_disable_l1aspm(pdev);
bc7f75fa
AK		 3627 if (pci_enable_device(pdev)) {
3628 dev_err(&pdev->dev,
3629 "Cannot re-enable PCI device after reset.\n");
3630 return PCI_ERS_RESULT_DISCONNECT;
3631 }
3632 pci_set_master(pdev);
3633
3634 pci_enable_wake(pdev, PCI_D3hot, 0);
3635 pci_enable_wake(pdev, PCI_D3cold, 0);
3636
3637 e1000e_reset(adapter);
3638 ew32(WUS, ~0);
3639
3640 return PCI_ERS_RESULT_RECOVERED;
3641}
3642
3643/**
3644 * e1000_io_resume - called when traffic can start flowing again.
3645 * @pdev: Pointer to PCI device
3646 *
3647 * This callback is called when the error recovery driver tells us that
3648 * its OK to resume normal operation. Implementation resembles the
3649 * second-half of the e1000_resume routine.
3650 */
3651static void e1000_io_resume(struct pci_dev *pdev)
3652{
3653 struct net_device *netdev = pci_get_drvdata(pdev);
3654 struct e1000_adapter *adapter = netdev_priv(netdev);
3655
3656 e1000_init_manageability(adapter);
3657
3658 if (netif_running(netdev)) {
3659 if (e1000e_up(adapter)) {
3660 dev_err(&pdev->dev,
3661 "can't bring device back up after reset\n");
3662 return;
3663 }
3664 }
3665
3666 netif_device_attach(netdev);
3667
3668 /* If the controller has AMT, do not set DRV_LOAD until the interface
3669 * is up. For all other cases, let the f/w know that the h/w is now
3670 * under the control of the driver. */
3671 if (!(adapter->flags & FLAG_HAS_AMT) ||
3672 !e1000e_check_mng_mode(&adapter->hw))
3673 e1000_get_hw_control(adapter);
3674
3675}
3676
3677static void e1000_print_device_info(struct e1000_adapter *adapter)
3678{
3679 struct e1000_hw *hw = &adapter->hw;
3680 struct net_device *netdev = adapter->netdev;
3681 u32 part_num;
3682
3683 /* print bus type/speed/width info */
3684 ndev_info(netdev, "(PCI Express:2.5GB/s:%s) "
3685 "%02x:%02x:%02x:%02x:%02x:%02x\n",
3686 /* bus width */
3687 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
3688 "Width x1"),
3689 /* MAC address */
3690 netdev->dev_addr[0], netdev->dev_addr[1],
3691 netdev->dev_addr[2], netdev->dev_addr[3],
3692 netdev->dev_addr[4], netdev->dev_addr[5]);
3693 ndev_info(netdev, "Intel(R) PRO/%s Network Connection\n",
3694 (hw->phy.type == e1000_phy_ife)
3695 ? "10/100" : "1000");
3696 e1000e_read_part_num(hw, &part_num);
3697 ndev_info(netdev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
3698 hw->mac.type, hw->phy.type,
3699 (part_num >> 8), (part_num & 0xff));
3700}
3701
3702/**
3703 * e1000_probe - Device Initialization Routine
3704 * @pdev: PCI device information struct
3705 * @ent: entry in e1000_pci_tbl
3706 *
3707 * Returns 0 on success, negative on failure
3708 *
3709 * e1000_probe initializes an adapter identified by a pci_dev structure.
3710 * The OS initialization, configuring of the adapter private structure,
3711 * and a hardware reset occur.
3712 **/
3713static int __devinit e1000_probe(struct pci_dev *pdev,
3714 const struct pci_device_id *ent)
3715{
3716 struct net_device *netdev;
3717 struct e1000_adapter *adapter;
3718 struct e1000_hw *hw;
3719 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
3720 unsigned long mmio_start, mmio_len;
3721 unsigned long flash_start, flash_len;
3722
3723 static int cards_found;
3724 int i, err, pci_using_dac;
3725 u16 eeprom_data = 0;
3726 u16 eeprom_apme_mask = E1000_EEPROM_APME;
3727
1eae4eb2 3728 e1000e_disable_l1aspm(pdev);
bc7f75fa
AK		 3729 err = pci_enable_device(pdev);
3730 if (err)
3731 return err;
3732
3733 pci_using_dac = 0;
3734 err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
3735 if (!err) {
3736 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
3737 if (!err)
3738 pci_using_dac = 1;
3739 } else {
3740 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
3741 if (err) {
3742 err = pci_set_consistent_dma_mask(pdev,
3743 DMA_32BIT_MASK);
3744 if (err) {
3745 dev_err(&pdev->dev, "No usable DMA "
3746 "configuration, aborting\n");
3747 goto err_dma;
3748 }
3749 }
3750 }
3751
3752 err = pci_request_regions(pdev, e1000e_driver_name);
3753 if (err)
3754 goto err_pci_reg;
3755
3756 pci_set_master(pdev);
3757
3758 err = -ENOMEM;
3759 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
3760 if (!netdev)
3761 goto err_alloc_etherdev;
3762
bc7f75fa
AK		 3763 SET_NETDEV_DEV(netdev, &pdev->dev);
3764
3765 pci_set_drvdata(pdev, netdev);
3766 adapter = netdev_priv(netdev);
3767 hw = &adapter->hw;
3768 adapter->netdev = netdev;
3769 adapter->pdev = pdev;
3770 adapter->ei = ei;
3771 adapter->pba = ei->pba;
3772 adapter->flags = ei->flags;
3773 adapter->hw.adapter = adapter;
3774 adapter->hw.mac.type = ei->mac;
3775 adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
3776
3777 mmio_start = pci_resource_start(pdev, 0);
3778 mmio_len = pci_resource_len(pdev, 0);
3779
3780 err = -EIO;
3781 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
3782 if (!adapter->hw.hw_addr)
3783 goto err_ioremap;
3784
3785 if ((adapter->flags & FLAG_HAS_FLASH) &&
3786 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
3787 flash_start = pci_resource_start(pdev, 1);
3788 flash_len = pci_resource_len(pdev, 1);
3789 adapter->hw.flash_address = ioremap(flash_start, flash_len);
3790 if (!adapter->hw.flash_address)
3791 goto err_flashmap;
3792 }
3793
3794 /* construct the net_device struct */
3795 netdev->open = &e1000_open;
3796 netdev->stop = &e1000_close;
3797 netdev->hard_start_xmit = &e1000_xmit_frame;
3798 netdev->get_stats = &e1000_get_stats;
3799 netdev->set_multicast_list = &e1000_set_multi;
3800 netdev->set_mac_address = &e1000_set_mac;
3801 netdev->change_mtu = &e1000_change_mtu;
3802 netdev->do_ioctl = &e1000_ioctl;
3803 e1000e_set_ethtool_ops(netdev);
3804 netdev->tx_timeout = &e1000_tx_timeout;
3805 netdev->watchdog_timeo = 5 * HZ;
3806 netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
3807 netdev->vlan_rx_register = e1000_vlan_rx_register;
3808 netdev->vlan_rx_add_vid = e1000_vlan_rx_add_vid;
3809 netdev->vlan_rx_kill_vid = e1000_vlan_rx_kill_vid;
3810#ifdef CONFIG_NET_POLL_CONTROLLER
3811 netdev->poll_controller = e1000_netpoll;
3812#endif
3813 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
3814
3815 netdev->mem_start = mmio_start;
3816 netdev->mem_end = mmio_start + mmio_len;
3817
3818 adapter->bd_number = cards_found++;
3819
3820 /* setup adapter struct */
3821 err = e1000_sw_init(adapter);
3822 if (err)
3823 goto err_sw_init;
3824
3825 err = -EIO;
3826
3827 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
3828 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
3829 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
3830
3831 err = ei->get_invariants(adapter);
3832 if (err)
3833 goto err_hw_init;
3834
3835 hw->mac.ops.get_bus_info(&adapter->hw);
3836
3837 adapter->hw.phy.wait_for_link = 0;
3838
3839 /* Copper options */
3840 if (adapter->hw.media_type == e1000_media_type_copper) {
3841 adapter->hw.phy.mdix = AUTO_ALL_MODES;
3842 adapter->hw.phy.disable_polarity_correction = 0;
3843 adapter->hw.phy.ms_type = e1000_ms_hw_default;
3844 }
3845
3846 if (e1000_check_reset_block(&adapter->hw))
3847 ndev_info(netdev,
3848 "PHY reset is blocked due to SOL/IDER session.\n");
3849
3850 netdev->features = NETIF_F_SG |
3851 NETIF_F_HW_CSUM |
3852 NETIF_F_HW_VLAN_TX |
3853 NETIF_F_HW_VLAN_RX;
3854
3855 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
3856 netdev->features |= NETIF_F_HW_VLAN_FILTER;
3857
3858 netdev->features |= NETIF_F_TSO;
3859 netdev->features |= NETIF_F_TSO6;
3860
3861 if (pci_using_dac)
3862 netdev->features |= NETIF_F_HIGHDMA;
3863
3864 /* We should not be using LLTX anymore, but we are still TX faster with
3865 * it. */
3866 netdev->features |= NETIF_F_LLTX;
3867
3868 if (e1000e_enable_mng_pass_thru(&adapter->hw))
3869 adapter->flags |= FLAG_MNG_PT_ENABLED;
3870
3871 /* before reading the NVM, reset the controller to
3872 * put the device in a known good starting state */
3873 adapter->hw.mac.ops.reset_hw(&adapter->hw);
3874
3875 /*
3876 * systems with ASPM and others may see the checksum fail on the first
3877 * attempt. Let's give it a few tries
3878 */
3879 for (i = 0;; i++) {
3880 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
3881 break;
3882 if (i == 2) {
3883 ndev_err(netdev, "The NVM Checksum Is Not Valid\n");
3884 err = -EIO;
3885 goto err_eeprom;
3886 }
3887 }
3888
3889 /* copy the MAC address out of the NVM */
3890 if (e1000e_read_mac_addr(&adapter->hw))
3891 ndev_err(netdev, "NVM Read Error while reading MAC address\n");
3892
3893 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
3894 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
3895
3896 if (!is_valid_ether_addr(netdev->perm_addr)) {
3897 ndev_err(netdev, "Invalid MAC Address: "
3898 "%02x:%02x:%02x:%02x:%02x:%02x\n",
3899 netdev->perm_addr[0], netdev->perm_addr[1],
3900 netdev->perm_addr[2], netdev->perm_addr[3],
3901 netdev->perm_addr[4], netdev->perm_addr[5]);
3902 err = -EIO;
3903 goto err_eeprom;
3904 }
3905
3906 init_timer(&adapter->watchdog_timer);
3907 adapter->watchdog_timer.function = &e1000_watchdog;
3908 adapter->watchdog_timer.data = (unsigned long) adapter;
3909
3910 init_timer(&adapter->phy_info_timer);
3911 adapter->phy_info_timer.function = &e1000_update_phy_info;
3912 adapter->phy_info_timer.data = (unsigned long) adapter;
3913
3914 INIT_WORK(&adapter->reset_task, e1000_reset_task);
3915 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
3916
3917 e1000e_check_options(adapter);
3918
3919 /* Initialize link parameters. User can change them with ethtool */
3920 adapter->hw.mac.autoneg = 1;
309af40b 3921 adapter->fc_autoneg = 1;
bc7f75fa
AK		 3922 adapter->hw.mac.original_fc = e1000_fc_default;
3923 adapter->hw.mac.fc = e1000_fc_default;
3924 adapter->hw.phy.autoneg_advertised = 0x2f;
3925
3926 /* ring size defaults */
3927 adapter->rx_ring->count = 256;
3928 adapter->tx_ring->count = 256;
3929
3930 /*
3931 * Initial Wake on LAN setting - If APM wake is enabled in
3932 * the EEPROM, enable the ACPI Magic Packet filter
3933 */
3934 if (adapter->flags & FLAG_APME_IN_WUC) {
3935 /* APME bit in EEPROM is mapped to WUC.APME */
3936 eeprom_data = er32(WUC);
3937 eeprom_apme_mask = E1000_WUC_APME;
3938 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
3939 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
3940 (adapter->hw.bus.func == 1))
3941 e1000_read_nvm(&adapter->hw,
3942 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
3943 else
3944 e1000_read_nvm(&adapter->hw,
3945 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
3946 }
3947
3948 /* fetch WoL from EEPROM */
3949 if (eeprom_data & eeprom_apme_mask)
3950 adapter->eeprom_wol |= E1000_WUFC_MAG;
3951
3952 /*
3953 * now that we have the eeprom settings, apply the special cases
3954 * where the eeprom may be wrong or the board simply won't support
3955 * wake on lan on a particular port
3956 */
3957 if (!(adapter->flags & FLAG_HAS_WOL))
3958 adapter->eeprom_wol = 0;
3959
3960 /* initialize the wol settings based on the eeprom settings */
3961 adapter->wol = adapter->eeprom_wol;
3962
3963 /* reset the hardware with the new settings */
3964 e1000e_reset(adapter);
3965
3966 /* If the controller has AMT, do not set DRV_LOAD until the interface
3967 * is up. For all other cases, let the f/w know that the h/w is now
3968 * under the control of the driver. */
3969 if (!(adapter->flags & FLAG_HAS_AMT) ||
3970 !e1000e_check_mng_mode(&adapter->hw))
3971 e1000_get_hw_control(adapter);
3972
3973 /* tell the stack to leave us alone until e1000_open() is called */
3974 netif_carrier_off(netdev);
3975 netif_stop_queue(netdev);
3976
3977 strcpy(netdev->name, "eth%d");
3978 err = register_netdev(netdev);
3979 if (err)
3980 goto err_register;
3981
3982 e1000_print_device_info(adapter);
3983
3984 return 0;
3985
3986err_register:
3987err_hw_init:
3988 e1000_release_hw_control(adapter);
3989err_eeprom:
3990 if (!e1000_check_reset_block(&adapter->hw))
3991 e1000_phy_hw_reset(&adapter->hw);
3992
3993 if (adapter->hw.flash_address)
3994 iounmap(adapter->hw.flash_address);
3995
3996err_flashmap:
3997 kfree(adapter->tx_ring);
3998 kfree(adapter->rx_ring);
3999err_sw_init:
4000 iounmap(adapter->hw.hw_addr);
4001err_ioremap:
4002 free_netdev(netdev);
4003err_alloc_etherdev:
4004 pci_release_regions(pdev);
4005err_pci_reg:
4006err_dma:
4007 pci_disable_device(pdev);
4008 return err;
4009}
4010
4011/**
4012 * e1000_remove - Device Removal Routine
4013 * @pdev: PCI device information struct
4014 *
4015 * e1000_remove is called by the PCI subsystem to alert the driver
4016 * that it should release a PCI device. The could be caused by a
4017 * Hot-Plug event, or because the driver is going to be removed from
4018 * memory.
4019 **/
4020static void __devexit e1000_remove(struct pci_dev *pdev)
4021{
4022 struct net_device *netdev = pci_get_drvdata(pdev);
4023 struct e1000_adapter *adapter = netdev_priv(netdev);
4024
4025 /* flush_scheduled work may reschedule our watchdog task, so
4026 * explicitly disable watchdog tasks from being rescheduled */
4027 set_bit(__E1000_DOWN, &adapter->state);
4028 del_timer_sync(&adapter->watchdog_timer);
4029 del_timer_sync(&adapter->phy_info_timer);
4030
4031 flush_scheduled_work();
4032
bc7f75fa
AK		 4033 /* Release control of h/w to f/w. If f/w is AMT enabled, this
4034 * would have already happened in close and is redundant. */
4035 e1000_release_hw_control(adapter);
4036
4037 unregister_netdev(netdev);
4038
4039 if (!e1000_check_reset_block(&adapter->hw))
4040 e1000_phy_hw_reset(&adapter->hw);
4041
4042 kfree(adapter->tx_ring);
4043 kfree(adapter->rx_ring);
4044
4045 iounmap(adapter->hw.hw_addr);
4046 if (adapter->hw.flash_address)
4047 iounmap(adapter->hw.flash_address);
4048 pci_release_regions(pdev);
4049
4050 free_netdev(netdev);
4051
4052 pci_disable_device(pdev);
4053}
4054
4055/* PCI Error Recovery (ERS) */
4056static struct pci_error_handlers e1000_err_handler = {
4057 .error_detected = e1000_io_error_detected,
4058 .slot_reset = e1000_io_slot_reset,
4059 .resume = e1000_io_resume,
4060};
4061
4062static struct pci_device_id e1000_pci_tbl[] = {
bc7f75fa
AK		 4063 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
4064 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
4065 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
4066 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
4067 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
4068 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
040babf9
AK		 4069 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
4070 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
4071 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
bc7f75fa
AK		 4072 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
4073 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
4074 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
4075 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
4076 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
4077 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
4078 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
4079 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
4080 board_80003es2lan },
4081 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
4082 board_80003es2lan },
4083 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
4084 board_80003es2lan },
4085 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
4086 board_80003es2lan },
4087 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
4088 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
4089 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
4090 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
4091 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
4092 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
4093 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
bc7f75fa
AK		 4094 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
4095 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
4096 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
4097 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
4098 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
4099
4100 { } /* terminate list */
4101};
4102MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
4103
4104/* PCI Device API Driver */
4105static struct pci_driver e1000_driver = {
4106 .name = e1000e_driver_name,
4107 .id_table = e1000_pci_tbl,
4108 .probe = e1000_probe,
4109 .remove = __devexit_p(e1000_remove),
4110#ifdef CONFIG_PM
4111 /* Power Managment Hooks */
4112 .suspend = e1000_suspend,
4113 .resume = e1000_resume,
4114#endif
4115 .shutdown = e1000_shutdown,
4116 .err_handler = &e1000_err_handler
4117};
4118
4119/**
4120 * e1000_init_module - Driver Registration Routine
4121 *
4122 * e1000_init_module is the first routine called when the driver is
4123 * loaded. All it does is register with the PCI subsystem.
4124 **/
4125static int __init e1000_init_module(void)
4126{
4127 int ret;
4128 printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
4129 e1000e_driver_name, e1000e_driver_version);
4130 printk(KERN_INFO "%s: Copyright (c) 1999-2007 Intel Corporation.\n",
4131 e1000e_driver_name);
4132 ret = pci_register_driver(&e1000_driver);
4133
4134 return ret;
4135}
4136module_init(e1000_init_module);
4137
4138/**
4139 * e1000_exit_module - Driver Exit Cleanup Routine
4140 *
4141 * e1000_exit_module is called just before the driver is removed
4142 * from memory.
4143 **/
4144static void __exit e1000_exit_module(void)
4145{
4146 pci_unregister_driver(&e1000_driver);
4147}
4148module_exit(e1000_exit_module);
4149
4150
4151MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
4152MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
4153MODULE_LICENSE("GPL");
4154MODULE_VERSION(DRV_VERSION);
4155
4156/* e1000_main.c */
Clone of davem's net-next-2.6 tree