projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
e1000e: fix e1000_netpoll(), remove extraneous e1000_clean_tx_irq() call
[deliverable/linux.git] / drivers / net / e1000e / netdev.c
1 /*******************************************************************************
2
3 Intel PRO/1000 Linux driver
4 Copyright(c) 1999 - 2008 Intel Corporation.
5
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
9
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 more details.
14
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
21
22 Contact Information:
23 Linux NICS <linux.nics@intel.com>
24 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #include <linux/module.h>
30 #include <linux/types.h>
31 #include <linux/init.h>
32 #include <linux/pci.h>
33 #include <linux/vmalloc.h>
34 #include <linux/pagemap.h>
35 #include <linux/delay.h>
36 #include <linux/netdevice.h>
37 #include <linux/tcp.h>
38 #include <linux/ipv6.h>
39 #include <net/checksum.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/mii.h>
42 #include <linux/ethtool.h>
43 #include <linux/if_vlan.h>
44 #include <linux/cpu.h>
45 #include <linux/smp.h>
46 #include <linux/pm_qos_params.h>
47
48 #include "e1000.h"
49
50 #define DRV_VERSION "0.3.3.3-k2"
51 char e1000e_driver_name[] = "e1000e";
52 const char e1000e_driver_version[] = DRV_VERSION;
53
54 static const struct e1000_info *e1000_info_tbl[] = {
55 [board_82571] = &e1000_82571_info,
56 [board_82572] = &e1000_82572_info,
57 [board_82573] = &e1000_82573_info,
58 [board_80003es2lan] = &e1000_es2_info,
59 [board_ich8lan] = &e1000_ich8_info,
60 [board_ich9lan] = &e1000_ich9_info,
61 };
62
63 #ifdef DEBUG
64 /**
65 * e1000_get_hw_dev_name - return device name string
66 * used by hardware layer to print debugging information
67 **/
68 char *e1000e_get_hw_dev_name(struct e1000_hw *hw)
69 {
70 return hw->adapter->netdev->name;
71 }
72 #endif
73
74 /**
75 * e1000_desc_unused - calculate if we have unused descriptors
76 **/
77 static int e1000_desc_unused(struct e1000_ring *ring)
78 {
79 if (ring->next_to_clean > ring->next_to_use)
80 return ring->next_to_clean - ring->next_to_use - 1;
81
82 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
83 }
84
85 /**
86 * e1000_receive_skb - helper function to handle Rx indications
87 * @adapter: board private structure
88 * @status: descriptor status field as written by hardware
89 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
90 * @skb: pointer to sk_buff to be indicated to stack
91 **/
92 static void e1000_receive_skb(struct e1000_adapter *adapter,
93 struct net_device *netdev,
94 struct sk_buff *skb,
95 u8 status, __le16 vlan)
96 {
97 skb->protocol = eth_type_trans(skb, netdev);
98
99 if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
100 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
101 le16_to_cpu(vlan));
102 else
103 netif_receive_skb(skb);
104
105 netdev->last_rx = jiffies;
106 }
107
108 /**
109 * e1000_rx_checksum - Receive Checksum Offload for 82543
110 * @adapter: board private structure
111 * @status_err: receive descriptor status and error fields
112 * @csum: receive descriptor csum field
113 * @sk_buff: socket buffer with received data
114 **/
115 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
116 u32 csum, struct sk_buff *skb)
117 {
118 u16 status = (u16)status_err;
119 u8 errors = (u8)(status_err >> 24);
120 skb->ip_summed = CHECKSUM_NONE;
121
122 /* Ignore Checksum bit is set */
123 if (status & E1000_RXD_STAT_IXSM)
124 return;
125 /* TCP/UDP checksum error bit is set */
126 if (errors & E1000_RXD_ERR_TCPE) {
127 /* let the stack verify checksum errors */
128 adapter->hw_csum_err++;
129 return;
130 }
131
132 /* TCP/UDP Checksum has not been calculated */
133 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
134 return;
135
136 /* It must be a TCP or UDP packet with a valid checksum */
137 if (status & E1000_RXD_STAT_TCPCS) {
138 /* TCP checksum is good */
139 skb->ip_summed = CHECKSUM_UNNECESSARY;
140 } else {
141 /*
142 * IP fragment with UDP payload
143 * Hardware complements the payload checksum, so we undo it
144 * and then put the value in host order for further stack use.
145 */
146 __sum16 sum = (__force __sum16)htons(csum);
147 skb->csum = csum_unfold(~sum);
148 skb->ip_summed = CHECKSUM_COMPLETE;
149 }
150 adapter->hw_csum_good++;
151 }
152
153 /**
154 * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
155 * @adapter: address of board private structure
156 **/
157 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
158 int cleaned_count)
159 {
160 struct net_device *netdev = adapter->netdev;
161 struct pci_dev *pdev = adapter->pdev;
162 struct e1000_ring *rx_ring = adapter->rx_ring;
163 struct e1000_rx_desc *rx_desc;
164 struct e1000_buffer *buffer_info;
165 struct sk_buff *skb;
166 unsigned int i;
167 unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
168
169 i = rx_ring->next_to_use;
170 buffer_info = &rx_ring->buffer_info[i];
171
172 while (cleaned_count--) {
173 skb = buffer_info->skb;
174 if (skb) {
175 skb_trim(skb, 0);
176 goto map_skb;
177 }
178
179 skb = netdev_alloc_skb(netdev, bufsz);
180 if (!skb) {
181 /* Better luck next round */
182 adapter->alloc_rx_buff_failed++;
183 break;
184 }
185
186 /*
187 * Make buffer alignment 2 beyond a 16 byte boundary
188 * this will result in a 16 byte aligned IP header after
189 * the 14 byte MAC header is removed
190 */
191 skb_reserve(skb, NET_IP_ALIGN);
192
193 buffer_info->skb = skb;
194 map_skb:
195 buffer_info->dma = pci_map_single(pdev, skb->data,
196 adapter->rx_buffer_len,
197 PCI_DMA_FROMDEVICE);
198 if (pci_dma_mapping_error(buffer_info->dma)) {
199 dev_err(&pdev->dev, "RX DMA map failed\n");
200 adapter->rx_dma_failed++;
201 break;
202 }
203
204 rx_desc = E1000_RX_DESC(*rx_ring, i);
205 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
206
207 i++;
208 if (i == rx_ring->count)
209 i = 0;
210 buffer_info = &rx_ring->buffer_info[i];
211 }
212
213 if (rx_ring->next_to_use != i) {
214 rx_ring->next_to_use = i;
215 if (i-- == 0)
216 i = (rx_ring->count - 1);
217
218 /*
219 * Force memory writes to complete before letting h/w
220 * know there are new descriptors to fetch. (Only
221 * applicable for weak-ordered memory model archs,
222 * such as IA-64).
223 */
224 wmb();
225 writel(i, adapter->hw.hw_addr + rx_ring->tail);
226 }
227 }
228
229 /**
230 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
231 * @adapter: address of board private structure
232 **/
233 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
234 int cleaned_count)
235 {
236 struct net_device *netdev = adapter->netdev;
237 struct pci_dev *pdev = adapter->pdev;
238 union e1000_rx_desc_packet_split *rx_desc;
239 struct e1000_ring *rx_ring = adapter->rx_ring;
240 struct e1000_buffer *buffer_info;
241 struct e1000_ps_page *ps_page;
242 struct sk_buff *skb;
243 unsigned int i, j;
244
245 i = rx_ring->next_to_use;
246 buffer_info = &rx_ring->buffer_info[i];
247
248 while (cleaned_count--) {
249 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
250
251 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
252 ps_page = &buffer_info->ps_pages[j];
253 if (j >= adapter->rx_ps_pages) {
254 /* all unused desc entries get hw null ptr */
255 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
256 continue;
257 }
258 if (!ps_page->page) {
259 ps_page->page = alloc_page(GFP_ATOMIC);
260 if (!ps_page->page) {
261 adapter->alloc_rx_buff_failed++;
262 goto no_buffers;
263 }
264 ps_page->dma = pci_map_page(pdev,
265 ps_page->page,
266 0, PAGE_SIZE,
267 PCI_DMA_FROMDEVICE);
268 if (pci_dma_mapping_error(ps_page->dma)) {
269 dev_err(&adapter->pdev->dev,
270 "RX DMA page map failed\n");
271 adapter->rx_dma_failed++;
272 goto no_buffers;
273 }
274 }
275 /*
276 * Refresh the desc even if buffer_addrs
277 * didn't change because each write-back
278 * erases this info.
279 */
280 rx_desc->read.buffer_addr[j+1] =
281 cpu_to_le64(ps_page->dma);
282 }
283
284 skb = netdev_alloc_skb(netdev,
285 adapter->rx_ps_bsize0 + NET_IP_ALIGN);
286
287 if (!skb) {
288 adapter->alloc_rx_buff_failed++;
289 break;
290 }
291
292 /*
293 * Make buffer alignment 2 beyond a 16 byte boundary
294 * this will result in a 16 byte aligned IP header after
295 * the 14 byte MAC header is removed
296 */
297 skb_reserve(skb, NET_IP_ALIGN);
298
299 buffer_info->skb = skb;
300 buffer_info->dma = pci_map_single(pdev, skb->data,
301 adapter->rx_ps_bsize0,
302 PCI_DMA_FROMDEVICE);
303 if (pci_dma_mapping_error(buffer_info->dma)) {
304 dev_err(&pdev->dev, "RX DMA map failed\n");
305 adapter->rx_dma_failed++;
306 /* cleanup skb */
307 dev_kfree_skb_any(skb);
308 buffer_info->skb = NULL;
309 break;
310 }
311
312 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
313
314 i++;
315 if (i == rx_ring->count)
316 i = 0;
317 buffer_info = &rx_ring->buffer_info[i];
318 }
319
320 no_buffers:
321 if (rx_ring->next_to_use != i) {
322 rx_ring->next_to_use = i;
323
324 if (!(i--))
325 i = (rx_ring->count - 1);
326
327 /*
328 * Force memory writes to complete before letting h/w
329 * know there are new descriptors to fetch. (Only
330 * applicable for weak-ordered memory model archs,
331 * such as IA-64).
332 */
333 wmb();
334 /*
335 * Hardware increments by 16 bytes, but packet split
336 * descriptors are 32 bytes...so we increment tail
337 * twice as much.
338 */
339 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
340 }
341 }
342
343 /**
344 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
345 * @adapter: address of board private structure
346 * @rx_ring: pointer to receive ring structure
347 * @cleaned_count: number of buffers to allocate this pass
348 **/
349
350 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
351 int cleaned_count)
352 {
353 struct net_device *netdev = adapter->netdev;
354 struct pci_dev *pdev = adapter->pdev;
355 struct e1000_rx_desc *rx_desc;
356 struct e1000_ring *rx_ring = adapter->rx_ring;
357 struct e1000_buffer *buffer_info;
358 struct sk_buff *skb;
359 unsigned int i;
360 unsigned int bufsz = 256 -
361 16 /* for skb_reserve */ -
362 NET_IP_ALIGN;
363
364 i = rx_ring->next_to_use;
365 buffer_info = &rx_ring->buffer_info[i];
366
367 while (cleaned_count--) {
368 skb = buffer_info->skb;
369 if (skb) {
370 skb_trim(skb, 0);
371 goto check_page;
372 }
373
374 skb = netdev_alloc_skb(netdev, bufsz);
375 if (unlikely(!skb)) {
376 /* Better luck next round */
377 adapter->alloc_rx_buff_failed++;
378 break;
379 }
380
381 /* Make buffer alignment 2 beyond a 16 byte boundary
382 * this will result in a 16 byte aligned IP header after
383 * the 14 byte MAC header is removed
384 */
385 skb_reserve(skb, NET_IP_ALIGN);
386
387 buffer_info->skb = skb;
388 check_page:
389 /* allocate a new page if necessary */
390 if (!buffer_info->page) {
391 buffer_info->page = alloc_page(GFP_ATOMIC);
392 if (unlikely(!buffer_info->page)) {
393 adapter->alloc_rx_buff_failed++;
394 break;
395 }
396 }
397
398 if (!buffer_info->dma)
399 buffer_info->dma = pci_map_page(pdev,
400 buffer_info->page, 0,
401 PAGE_SIZE,
402 PCI_DMA_FROMDEVICE);
403
404 rx_desc = E1000_RX_DESC(*rx_ring, i);
405 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
406
407 if (unlikely(++i == rx_ring->count))
408 i = 0;
409 buffer_info = &rx_ring->buffer_info[i];
410 }
411
412 if (likely(rx_ring->next_to_use != i)) {
413 rx_ring->next_to_use = i;
414 if (unlikely(i-- == 0))
415 i = (rx_ring->count - 1);
416
417 /* Force memory writes to complete before letting h/w
418 * know there are new descriptors to fetch. (Only
419 * applicable for weak-ordered memory model archs,
420 * such as IA-64). */
421 wmb();
422 writel(i, adapter->hw.hw_addr + rx_ring->tail);
423 }
424 }
425
426 /**
427 * e1000_clean_rx_irq - Send received data up the network stack; legacy
428 * @adapter: board private structure
429 *
430 * the return value indicates whether actual cleaning was done, there
431 * is no guarantee that everything was cleaned
432 **/
433 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
434 int *work_done, int work_to_do)
435 {
436 struct net_device *netdev = adapter->netdev;
437 struct pci_dev *pdev = adapter->pdev;
438 struct e1000_ring *rx_ring = adapter->rx_ring;
439 struct e1000_rx_desc *rx_desc, *next_rxd;
440 struct e1000_buffer *buffer_info, *next_buffer;
441 u32 length;
442 unsigned int i;
443 int cleaned_count = 0;
444 bool cleaned = 0;
445 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
446
447 i = rx_ring->next_to_clean;
448 rx_desc = E1000_RX_DESC(*rx_ring, i);
449 buffer_info = &rx_ring->buffer_info[i];
450
451 while (rx_desc->status & E1000_RXD_STAT_DD) {
452 struct sk_buff *skb;
453 u8 status;
454
455 if (*work_done >= work_to_do)
456 break;
457 (*work_done)++;
458
459 status = rx_desc->status;
460 skb = buffer_info->skb;
461 buffer_info->skb = NULL;
462
463 prefetch(skb->data - NET_IP_ALIGN);
464
465 i++;
466 if (i == rx_ring->count)
467 i = 0;
468 next_rxd = E1000_RX_DESC(*rx_ring, i);
469 prefetch(next_rxd);
470
471 next_buffer = &rx_ring->buffer_info[i];
472
473 cleaned = 1;
474 cleaned_count++;
475 pci_unmap_single(pdev,
476 buffer_info->dma,
477 adapter->rx_buffer_len,
478 PCI_DMA_FROMDEVICE);
479 buffer_info->dma = 0;
480
481 length = le16_to_cpu(rx_desc->length);
482
483 /* !EOP means multiple descriptors were used to store a single
484 * packet, also make sure the frame isn't just CRC only */
485 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
486 /* All receives must fit into a single buffer */
487 ndev_dbg(netdev, "%s: Receive packet consumed "
488 "multiple buffers\n", netdev->name);
489 /* recycle */
490 buffer_info->skb = skb;
491 goto next_desc;
492 }
493
494 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
495 /* recycle */
496 buffer_info->skb = skb;
497 goto next_desc;
498 }
499
500 total_rx_bytes += length;
501 total_rx_packets++;
502
503 /*
504 * code added for copybreak, this should improve
505 * performance for small packets with large amounts
506 * of reassembly being done in the stack
507 */
508 if (length < copybreak) {
509 struct sk_buff *new_skb =
510 netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
511 if (new_skb) {
512 skb_reserve(new_skb, NET_IP_ALIGN);
513 memcpy(new_skb->data - NET_IP_ALIGN,
514 skb->data - NET_IP_ALIGN,
515 length + NET_IP_ALIGN);
516 /* save the skb in buffer_info as good */
517 buffer_info->skb = skb;
518 skb = new_skb;
519 }
520 /* else just continue with the old one */
521 }
522 /* end copybreak code */
523 skb_put(skb, length);
524
525 /* Receive Checksum Offload */
526 e1000_rx_checksum(adapter,
527 (u32)(status) |
528 ((u32)(rx_desc->errors) << 24),
529 le16_to_cpu(rx_desc->csum), skb);
530
531 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
532
533 next_desc:
534 rx_desc->status = 0;
535
536 /* return some buffers to hardware, one at a time is too slow */
537 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
538 adapter->alloc_rx_buf(adapter, cleaned_count);
539 cleaned_count = 0;
540 }
541
542 /* use prefetched values */
543 rx_desc = next_rxd;
544 buffer_info = next_buffer;
545 }
546 rx_ring->next_to_clean = i;
547
548 cleaned_count = e1000_desc_unused(rx_ring);
549 if (cleaned_count)
550 adapter->alloc_rx_buf(adapter, cleaned_count);
551
552 adapter->total_rx_bytes += total_rx_bytes;
553 adapter->total_rx_packets += total_rx_packets;
554 adapter->net_stats.rx_bytes += total_rx_bytes;
555 adapter->net_stats.rx_packets += total_rx_packets;
556 return cleaned;
557 }
558
559 static void e1000_put_txbuf(struct e1000_adapter *adapter,
560 struct e1000_buffer *buffer_info)
561 {
562 if (buffer_info->dma) {
563 pci_unmap_page(adapter->pdev, buffer_info->dma,
564 buffer_info->length, PCI_DMA_TODEVICE);
565 buffer_info->dma = 0;
566 }
567 if (buffer_info->skb) {
568 dev_kfree_skb_any(buffer_info->skb);
569 buffer_info->skb = NULL;
570 }
571 }
572
573 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
574 {
575 struct e1000_ring *tx_ring = adapter->tx_ring;
576 unsigned int i = tx_ring->next_to_clean;
577 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
578 struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
579 struct net_device *netdev = adapter->netdev;
580
581 /* detected Tx unit hang */
582 ndev_err(netdev,
583 "Detected Tx Unit Hang:\n"
584 " TDH <%x>\n"
585 " TDT <%x>\n"
586 " next_to_use <%x>\n"
587 " next_to_clean <%x>\n"
588 "buffer_info[next_to_clean]:\n"
589 " time_stamp <%lx>\n"
590 " next_to_watch <%x>\n"
591 " jiffies <%lx>\n"
592 " next_to_watch.status <%x>\n",
593 readl(adapter->hw.hw_addr + tx_ring->head),
594 readl(adapter->hw.hw_addr + tx_ring->tail),
595 tx_ring->next_to_use,
596 tx_ring->next_to_clean,
597 tx_ring->buffer_info[eop].time_stamp,
598 eop,
599 jiffies,
600 eop_desc->upper.fields.status);
601 }
602
603 /**
604 * e1000_clean_tx_irq - Reclaim resources after transmit completes
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 **/
610 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
611 {
612 struct net_device *netdev = adapter->netdev;
613 struct e1000_hw *hw = &adapter->hw;
614 struct e1000_ring *tx_ring = adapter->tx_ring;
615 struct e1000_tx_desc *tx_desc, *eop_desc;
616 struct e1000_buffer *buffer_info;
617 unsigned int i, eop;
618 unsigned int count = 0;
619 bool cleaned = 0;
620 unsigned int total_tx_bytes = 0, total_tx_packets = 0;
621
622 i = tx_ring->next_to_clean;
623 eop = tx_ring->buffer_info[i].next_to_watch;
624 eop_desc = E1000_TX_DESC(*tx_ring, eop);
625
626 while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
627 for (cleaned = 0; !cleaned; ) {
628 tx_desc = E1000_TX_DESC(*tx_ring, i);
629 buffer_info = &tx_ring->buffer_info[i];
630 cleaned = (i == eop);
631
632 if (cleaned) {
633 struct sk_buff *skb = buffer_info->skb;
634 unsigned int segs, bytecount;
635 segs = skb_shinfo(skb)->gso_segs ?: 1;
636 /* multiply data chunks by size of headers */
637 bytecount = ((segs - 1) * skb_headlen(skb)) +
638 skb->len;
639 total_tx_packets += segs;
640 total_tx_bytes += bytecount;
641 }
642
643 e1000_put_txbuf(adapter, buffer_info);
644 tx_desc->upper.data = 0;
645
646 i++;
647 if (i == tx_ring->count)
648 i = 0;
649 }
650
651 eop = tx_ring->buffer_info[i].next_to_watch;
652 eop_desc = E1000_TX_DESC(*tx_ring, eop);
653 #define E1000_TX_WEIGHT 64
654 /* weight of a sort for tx, to avoid endless transmit cleanup */
655 if (count++ == E1000_TX_WEIGHT)
656 break;
657 }
658
659 tx_ring->next_to_clean = i;
660
661 #define TX_WAKE_THRESHOLD 32
662 if (cleaned && netif_carrier_ok(netdev) &&
663 e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
664 /* Make sure that anybody stopping the queue after this
665 * sees the new next_to_clean.
666 */
667 smp_mb();
668
669 if (netif_queue_stopped(netdev) &&
670 !(test_bit(__E1000_DOWN, &adapter->state))) {
671 netif_wake_queue(netdev);
672 ++adapter->restart_queue;
673 }
674 }
675
676 if (adapter->detect_tx_hung) {
677 /*
678 * Detect a transmit hang in hardware, this serializes the
679 * check with the clearing of time_stamp and movement of i
680 */
681 adapter->detect_tx_hung = 0;
682 if (tx_ring->buffer_info[eop].dma &&
683 time_after(jiffies, tx_ring->buffer_info[eop].time_stamp
684 + (adapter->tx_timeout_factor * HZ))
685 && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
686 e1000_print_tx_hang(adapter);
687 netif_stop_queue(netdev);
688 }
689 }
690 adapter->total_tx_bytes += total_tx_bytes;
691 adapter->total_tx_packets += total_tx_packets;
692 adapter->net_stats.tx_bytes += total_tx_bytes;
693 adapter->net_stats.tx_packets += total_tx_packets;
694 return cleaned;
695 }
696
697 /**
698 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
699 * @adapter: board private structure
700 *
701 * the return value indicates whether actual cleaning was done, there
702 * is no guarantee that everything was cleaned
703 **/
704 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
705 int *work_done, int work_to_do)
706 {
707 union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
708 struct net_device *netdev = adapter->netdev;
709 struct pci_dev *pdev = adapter->pdev;
710 struct e1000_ring *rx_ring = adapter->rx_ring;
711 struct e1000_buffer *buffer_info, *next_buffer;
712 struct e1000_ps_page *ps_page;
713 struct sk_buff *skb;
714 unsigned int i, j;
715 u32 length, staterr;
716 int cleaned_count = 0;
717 bool cleaned = 0;
718 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
719
720 i = rx_ring->next_to_clean;
721 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
722 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
723 buffer_info = &rx_ring->buffer_info[i];
724
725 while (staterr & E1000_RXD_STAT_DD) {
726 if (*work_done >= work_to_do)
727 break;
728 (*work_done)++;
729 skb = buffer_info->skb;
730
731 /* in the packet split case this is header only */
732 prefetch(skb->data - NET_IP_ALIGN);
733
734 i++;
735 if (i == rx_ring->count)
736 i = 0;
737 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
738 prefetch(next_rxd);
739
740 next_buffer = &rx_ring->buffer_info[i];
741
742 cleaned = 1;
743 cleaned_count++;
744 pci_unmap_single(pdev, buffer_info->dma,
745 adapter->rx_ps_bsize0,
746 PCI_DMA_FROMDEVICE);
747 buffer_info->dma = 0;
748
749 if (!(staterr & E1000_RXD_STAT_EOP)) {
750 ndev_dbg(netdev, "%s: Packet Split buffers didn't pick "
751 "up the full packet\n", netdev->name);
752 dev_kfree_skb_irq(skb);
753 goto next_desc;
754 }
755
756 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
757 dev_kfree_skb_irq(skb);
758 goto next_desc;
759 }
760
761 length = le16_to_cpu(rx_desc->wb.middle.length0);
762
763 if (!length) {
764 ndev_dbg(netdev, "%s: Last part of the packet spanning"
765 " multiple descriptors\n", netdev->name);
766 dev_kfree_skb_irq(skb);
767 goto next_desc;
768 }
769
770 /* Good Receive */
771 skb_put(skb, length);
772
773 {
774 /*
775 * this looks ugly, but it seems compiler issues make it
776 * more efficient than reusing j
777 */
778 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
779
780 /*
781 * page alloc/put takes too long and effects small packet
782 * throughput, so unsplit small packets and save the alloc/put
783 * only valid in softirq (napi) context to call kmap_*
784 */
785 if (l1 && (l1 <= copybreak) &&
786 ((length + l1) <= adapter->rx_ps_bsize0)) {
787 u8 *vaddr;
788
789 ps_page = &buffer_info->ps_pages[0];
790
791 /*
792 * there is no documentation about how to call
793 * kmap_atomic, so we can't hold the mapping
794 * very long
795 */
796 pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
797 PAGE_SIZE, PCI_DMA_FROMDEVICE);
798 vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
799 memcpy(skb_tail_pointer(skb), vaddr, l1);
800 kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
801 pci_dma_sync_single_for_device(pdev, ps_page->dma,
802 PAGE_SIZE, PCI_DMA_FROMDEVICE);
803
804 skb_put(skb, l1);
805 goto copydone;
806 } /* if */
807 }
808
809 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
810 length = le16_to_cpu(rx_desc->wb.upper.length[j]);
811 if (!length)
812 break;
813
814 ps_page = &buffer_info->ps_pages[j];
815 pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
816 PCI_DMA_FROMDEVICE);
817 ps_page->dma = 0;
818 skb_fill_page_desc(skb, j, ps_page->page, 0, length);
819 ps_page->page = NULL;
820 skb->len += length;
821 skb->data_len += length;
822 skb->truesize += length;
823 }
824
825 copydone:
826 total_rx_bytes += skb->len;
827 total_rx_packets++;
828
829 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
830 rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
831
832 if (rx_desc->wb.upper.header_status &
833 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
834 adapter->rx_hdr_split++;
835
836 e1000_receive_skb(adapter, netdev, skb,
837 staterr, rx_desc->wb.middle.vlan);
838
839 next_desc:
840 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
841 buffer_info->skb = NULL;
842
843 /* return some buffers to hardware, one at a time is too slow */
844 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
845 adapter->alloc_rx_buf(adapter, cleaned_count);
846 cleaned_count = 0;
847 }
848
849 /* use prefetched values */
850 rx_desc = next_rxd;
851 buffer_info = next_buffer;
852
853 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
854 }
855 rx_ring->next_to_clean = i;
856
857 cleaned_count = e1000_desc_unused(rx_ring);
858 if (cleaned_count)
859 adapter->alloc_rx_buf(adapter, cleaned_count);
860
861 adapter->total_rx_bytes += total_rx_bytes;
862 adapter->total_rx_packets += total_rx_packets;
863 adapter->net_stats.rx_bytes += total_rx_bytes;
864 adapter->net_stats.rx_packets += total_rx_packets;
865 return cleaned;
866 }
867
868 /**
869 * e1000_consume_page - helper function
870 **/
871 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
872 u16 length)
873 {
874 bi->page = NULL;
875 skb->len += length;
876 skb->data_len += length;
877 skb->truesize += length;
878 }
879
880 /**
881 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
882 * @adapter: board private structure
883 *
884 * the return value indicates whether actual cleaning was done, there
885 * is no guarantee that everything was cleaned
886 **/
887
888 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
889 int *work_done, int work_to_do)
890 {
891 struct net_device *netdev = adapter->netdev;
892 struct pci_dev *pdev = adapter->pdev;
893 struct e1000_ring *rx_ring = adapter->rx_ring;
894 struct e1000_rx_desc *rx_desc, *next_rxd;
895 struct e1000_buffer *buffer_info, *next_buffer;
896 u32 length;
897 unsigned int i;
898 int cleaned_count = 0;
899 bool cleaned = false;
900 unsigned int total_rx_bytes=0, total_rx_packets=0;
901
902 i = rx_ring->next_to_clean;
903 rx_desc = E1000_RX_DESC(*rx_ring, i);
904 buffer_info = &rx_ring->buffer_info[i];
905
906 while (rx_desc->status & E1000_RXD_STAT_DD) {
907 struct sk_buff *skb;
908 u8 status;
909
910 if (*work_done >= work_to_do)
911 break;
912 (*work_done)++;
913
914 status = rx_desc->status;
915 skb = buffer_info->skb;
916 buffer_info->skb = NULL;
917
918 ++i;
919 if (i == rx_ring->count)
920 i = 0;
921 next_rxd = E1000_RX_DESC(*rx_ring, i);
922 prefetch(next_rxd);
923
924 next_buffer = &rx_ring->buffer_info[i];
925
926 cleaned = true;
927 cleaned_count++;
928 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
929 PCI_DMA_FROMDEVICE);
930 buffer_info->dma = 0;
931
932 length = le16_to_cpu(rx_desc->length);
933
934 /* errors is only valid for DD + EOP descriptors */
935 if (unlikely((status & E1000_RXD_STAT_EOP) &&
936 (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
937 /* recycle both page and skb */
938 buffer_info->skb = skb;
939 /* an error means any chain goes out the window
940 * too */
941 if (rx_ring->rx_skb_top)
942 dev_kfree_skb(rx_ring->rx_skb_top);
943 rx_ring->rx_skb_top = NULL;
944 goto next_desc;
945 }
946
947 #define rxtop rx_ring->rx_skb_top
948 if (!(status & E1000_RXD_STAT_EOP)) {
949 /* this descriptor is only the beginning (or middle) */
950 if (!rxtop) {
951 /* this is the beginning of a chain */
952 rxtop = skb;
953 skb_fill_page_desc(rxtop, 0, buffer_info->page,
954 0, length);
955 } else {
956 /* this is the middle of a chain */
957 skb_fill_page_desc(rxtop,
958 skb_shinfo(rxtop)->nr_frags,
959 buffer_info->page, 0, length);
960 /* re-use the skb, only consumed the page */
961 buffer_info->skb = skb;
962 }
963 e1000_consume_page(buffer_info, rxtop, length);
964 goto next_desc;
965 } else {
966 if (rxtop) {
967 /* end of the chain */
968 skb_fill_page_desc(rxtop,
969 skb_shinfo(rxtop)->nr_frags,
970 buffer_info->page, 0, length);
971 /* re-use the current skb, we only consumed the
972 * page */
973 buffer_info->skb = skb;
974 skb = rxtop;
975 rxtop = NULL;
976 e1000_consume_page(buffer_info, skb, length);
977 } else {
978 /* no chain, got EOP, this buf is the packet
979 * copybreak to save the put_page/alloc_page */
980 if (length <= copybreak &&
981 skb_tailroom(skb) >= length) {
982 u8 *vaddr;
983 vaddr = kmap_atomic(buffer_info->page,
984 KM_SKB_DATA_SOFTIRQ);
985 memcpy(skb_tail_pointer(skb), vaddr,
986 length);
987 kunmap_atomic(vaddr,
988 KM_SKB_DATA_SOFTIRQ);
989 /* re-use the page, so don't erase
990 * buffer_info->page */
991 skb_put(skb, length);
992 } else {
993 skb_fill_page_desc(skb, 0,
994 buffer_info->page, 0,
995 length);
996 e1000_consume_page(buffer_info, skb,
997 length);
998 }
999 }
1000 }
1001
1002 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1003 e1000_rx_checksum(adapter,
1004 (u32)(status) |
1005 ((u32)(rx_desc->errors) << 24),
1006 le16_to_cpu(rx_desc->csum), skb);
1007
1008 /* probably a little skewed due to removing CRC */
1009 total_rx_bytes += skb->len;
1010 total_rx_packets++;
1011
1012 /* eth type trans needs skb->data to point to something */
1013 if (!pskb_may_pull(skb, ETH_HLEN)) {
1014 ndev_err(netdev, "pskb_may_pull failed.\n");
1015 dev_kfree_skb(skb);
1016 goto next_desc;
1017 }
1018
1019 e1000_receive_skb(adapter, netdev, skb, status,
1020 rx_desc->special);
1021
1022 next_desc:
1023 rx_desc->status = 0;
1024
1025 /* return some buffers to hardware, one at a time is too slow */
1026 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1027 adapter->alloc_rx_buf(adapter, cleaned_count);
1028 cleaned_count = 0;
1029 }
1030
1031 /* use prefetched values */
1032 rx_desc = next_rxd;
1033 buffer_info = next_buffer;
1034 }
1035 rx_ring->next_to_clean = i;
1036
1037 cleaned_count = e1000_desc_unused(rx_ring);
1038 if (cleaned_count)
1039 adapter->alloc_rx_buf(adapter, cleaned_count);
1040
1041 adapter->total_rx_bytes += total_rx_bytes;
1042 adapter->total_rx_packets += total_rx_packets;
1043 adapter->net_stats.rx_bytes += total_rx_bytes;
1044 adapter->net_stats.rx_packets += total_rx_packets;
1045 return cleaned;
1046 }
1047
1048 /**
1049 * e1000_clean_rx_ring - Free Rx Buffers per Queue
1050 * @adapter: board private structure
1051 **/
1052 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1053 {
1054 struct e1000_ring *rx_ring = adapter->rx_ring;
1055 struct e1000_buffer *buffer_info;
1056 struct e1000_ps_page *ps_page;
1057 struct pci_dev *pdev = adapter->pdev;
1058 unsigned int i, j;
1059
1060 /* Free all the Rx ring sk_buffs */
1061 for (i = 0; i < rx_ring->count; i++) {
1062 buffer_info = &rx_ring->buffer_info[i];
1063 if (buffer_info->dma) {
1064 if (adapter->clean_rx == e1000_clean_rx_irq)
1065 pci_unmap_single(pdev, buffer_info->dma,
1066 adapter->rx_buffer_len,
1067 PCI_DMA_FROMDEVICE);
1068 else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1069 pci_unmap_page(pdev, buffer_info->dma,
1070 PAGE_SIZE,
1071 PCI_DMA_FROMDEVICE);
1072 else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1073 pci_unmap_single(pdev, buffer_info->dma,
1074 adapter->rx_ps_bsize0,
1075 PCI_DMA_FROMDEVICE);
1076 buffer_info->dma = 0;
1077 }
1078
1079 if (buffer_info->page) {
1080 put_page(buffer_info->page);
1081 buffer_info->page = NULL;
1082 }
1083
1084 if (buffer_info->skb) {
1085 dev_kfree_skb(buffer_info->skb);
1086 buffer_info->skb = NULL;
1087 }
1088
1089 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1090 ps_page = &buffer_info->ps_pages[j];
1091 if (!ps_page->page)
1092 break;
1093 pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1094 PCI_DMA_FROMDEVICE);
1095 ps_page->dma = 0;
1096 put_page(ps_page->page);
1097 ps_page->page = NULL;
1098 }
1099 }
1100
1101 /* there also may be some cached data from a chained receive */
1102 if (rx_ring->rx_skb_top) {
1103 dev_kfree_skb(rx_ring->rx_skb_top);
1104 rx_ring->rx_skb_top = NULL;
1105 }
1106
1107 /* Zero out the descriptor ring */
1108 memset(rx_ring->desc, 0, rx_ring->size);
1109
1110 rx_ring->next_to_clean = 0;
1111 rx_ring->next_to_use = 0;
1112
1113 writel(0, adapter->hw.hw_addr + rx_ring->head);
1114 writel(0, adapter->hw.hw_addr + rx_ring->tail);
1115 }
1116
1117 /**
1118 * e1000_intr_msi - Interrupt Handler
1119 * @irq: interrupt number
1120 * @data: pointer to a network interface device structure
1121 **/
1122 static irqreturn_t e1000_intr_msi(int irq, void *data)
1123 {
1124 struct net_device *netdev = data;
1125 struct e1000_adapter *adapter = netdev_priv(netdev);
1126 struct e1000_hw *hw = &adapter->hw;
1127 u32 icr = er32(ICR);
1128
1129 /*
1130 * read ICR disables interrupts using IAM
1131 */
1132
1133 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1134 hw->mac.get_link_status = 1;
1135 /*
1136 * ICH8 workaround-- Call gig speed drop workaround on cable
1137 * disconnect (LSC) before accessing any PHY registers
1138 */
1139 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1140 (!(er32(STATUS) & E1000_STATUS_LU)))
1141 e1000e_gig_downshift_workaround_ich8lan(hw);
1142
1143 /*
1144 * 80003ES2LAN workaround-- For packet buffer work-around on
1145 * link down event; disable receives here in the ISR and reset
1146 * adapter in watchdog
1147 */
1148 if (netif_carrier_ok(netdev) &&
1149 adapter->flags & FLAG_RX_NEEDS_RESTART) {
1150 /* disable receives */
1151 u32 rctl = er32(RCTL);
1152 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1153 adapter->flags |= FLAG_RX_RESTART_NOW;
1154 }
1155 /* guard against interrupt when we're going down */
1156 if (!test_bit(__E1000_DOWN, &adapter->state))
1157 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1158 }
1159
1160 if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1161 adapter->total_tx_bytes = 0;
1162 adapter->total_tx_packets = 0;
1163 adapter->total_rx_bytes = 0;
1164 adapter->total_rx_packets = 0;
1165 __netif_rx_schedule(netdev, &adapter->napi);
1166 }
1167
1168 return IRQ_HANDLED;
1169 }
1170
1171 /**
1172 * e1000_intr - Interrupt Handler
1173 * @irq: interrupt number
1174 * @data: pointer to a network interface device structure
1175 **/
1176 static irqreturn_t e1000_intr(int irq, void *data)
1177 {
1178 struct net_device *netdev = data;
1179 struct e1000_adapter *adapter = netdev_priv(netdev);
1180 struct e1000_hw *hw = &adapter->hw;
1181
1182 u32 rctl, icr = er32(ICR);
1183 if (!icr)
1184 return IRQ_NONE; /* Not our interrupt */
1185
1186 /*
1187 * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1188 * not set, then the adapter didn't send an interrupt
1189 */
1190 if (!(icr & E1000_ICR_INT_ASSERTED))
1191 return IRQ_NONE;
1192
1193 /*
1194 * Interrupt Auto-Mask...upon reading ICR,
1195 * interrupts are masked. No need for the
1196 * IMC write
1197 */
1198
1199 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1200 hw->mac.get_link_status = 1;
1201 /*
1202 * ICH8 workaround-- Call gig speed drop workaround on cable
1203 * disconnect (LSC) before accessing any PHY registers
1204 */
1205 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1206 (!(er32(STATUS) & E1000_STATUS_LU)))
1207 e1000e_gig_downshift_workaround_ich8lan(hw);
1208
1209 /*
1210 * 80003ES2LAN workaround--
1211 * For packet buffer work-around on link down event;
1212 * disable receives here in the ISR and
1213 * reset adapter in watchdog
1214 */
1215 if (netif_carrier_ok(netdev) &&
1216 (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1217 /* disable receives */
1218 rctl = er32(RCTL);
1219 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1220 adapter->flags |= FLAG_RX_RESTART_NOW;
1221 }
1222 /* guard against interrupt when we're going down */
1223 if (!test_bit(__E1000_DOWN, &adapter->state))
1224 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1225 }
1226
1227 if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1228 adapter->total_tx_bytes = 0;
1229 adapter->total_tx_packets = 0;
1230 adapter->total_rx_bytes = 0;
1231 adapter->total_rx_packets = 0;
1232 __netif_rx_schedule(netdev, &adapter->napi);
1233 }
1234
1235 return IRQ_HANDLED;
1236 }
1237
1238 static int e1000_request_irq(struct e1000_adapter *adapter)
1239 {
1240 struct net_device *netdev = adapter->netdev;
1241 irq_handler_t handler = e1000_intr;
1242 int irq_flags = IRQF_SHARED;
1243 int err;
1244
1245 if (!pci_enable_msi(adapter->pdev)) {
1246 adapter->flags |= FLAG_MSI_ENABLED;
1247 handler = e1000_intr_msi;
1248 irq_flags = 0;
1249 }
1250
1251 err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
1252 netdev);
1253 if (err) {
1254 ndev_err(netdev,
1255 "Unable to allocate %s interrupt (return: %d)\n",
1256 adapter->flags & FLAG_MSI_ENABLED ? "MSI":"INTx",
1257 err);
1258 if (adapter->flags & FLAG_MSI_ENABLED)
1259 pci_disable_msi(adapter->pdev);
1260 }
1261
1262 return err;
1263 }
1264
1265 static void e1000_free_irq(struct e1000_adapter *adapter)
1266 {
1267 struct net_device *netdev = adapter->netdev;
1268
1269 free_irq(adapter->pdev->irq, netdev);
1270 if (adapter->flags & FLAG_MSI_ENABLED) {
1271 pci_disable_msi(adapter->pdev);
1272 adapter->flags &= ~FLAG_MSI_ENABLED;
1273 }
1274 }
1275
1276 /**
1277 * e1000_irq_disable - Mask off interrupt generation on the NIC
1278 **/
1279 static void e1000_irq_disable(struct e1000_adapter *adapter)
1280 {
1281 struct e1000_hw *hw = &adapter->hw;
1282
1283 ew32(IMC, ~0);
1284 e1e_flush();
1285 synchronize_irq(adapter->pdev->irq);
1286 }
1287
1288 /**
1289 * e1000_irq_enable - Enable default interrupt generation settings
1290 **/
1291 static void e1000_irq_enable(struct e1000_adapter *adapter)
1292 {
1293 struct e1000_hw *hw = &adapter->hw;
1294
1295 ew32(IMS, IMS_ENABLE_MASK);
1296 e1e_flush();
1297 }
1298
1299 /**
1300 * e1000_get_hw_control - get control of the h/w from f/w
1301 * @adapter: address of board private structure
1302 *
1303 * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1304 * For ASF and Pass Through versions of f/w this means that
1305 * the driver is loaded. For AMT version (only with 82573)
1306 * of the f/w this means that the network i/f is open.
1307 **/
1308 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1309 {
1310 struct e1000_hw *hw = &adapter->hw;
1311 u32 ctrl_ext;
1312 u32 swsm;
1313
1314 /* Let firmware know the driver has taken over */
1315 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1316 swsm = er32(SWSM);
1317 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1318 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1319 ctrl_ext = er32(CTRL_EXT);
1320 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1321 }
1322 }
1323
1324 /**
1325 * e1000_release_hw_control - release control of the h/w to f/w
1326 * @adapter: address of board private structure
1327 *
1328 * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1329 * For ASF and Pass Through versions of f/w this means that the
1330 * driver is no longer loaded. For AMT version (only with 82573) i
1331 * of the f/w this means that the network i/f is closed.
1332 *
1333 **/
1334 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1335 {
1336 struct e1000_hw *hw = &adapter->hw;
1337 u32 ctrl_ext;
1338 u32 swsm;
1339
1340 /* Let firmware taken over control of h/w */
1341 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1342 swsm = er32(SWSM);
1343 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1344 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1345 ctrl_ext = er32(CTRL_EXT);
1346 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1347 }
1348 }
1349
1350 /**
1351 * @e1000_alloc_ring - allocate memory for a ring structure
1352 **/
1353 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1354 struct e1000_ring *ring)
1355 {
1356 struct pci_dev *pdev = adapter->pdev;
1357
1358 ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1359 GFP_KERNEL);
1360 if (!ring->desc)
1361 return -ENOMEM;
1362
1363 return 0;
1364 }
1365
1366 /**
1367 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1368 * @adapter: board private structure
1369 *
1370 * Return 0 on success, negative on failure
1371 **/
1372 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1373 {
1374 struct e1000_ring *tx_ring = adapter->tx_ring;
1375 int err = -ENOMEM, size;
1376
1377 size = sizeof(struct e1000_buffer) * tx_ring->count;
1378 tx_ring->buffer_info = vmalloc(size);
1379 if (!tx_ring->buffer_info)
1380 goto err;
1381 memset(tx_ring->buffer_info, 0, size);
1382
1383 /* round up to nearest 4K */
1384 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1385 tx_ring->size = ALIGN(tx_ring->size, 4096);
1386
1387 err = e1000_alloc_ring_dma(adapter, tx_ring);
1388 if (err)
1389 goto err;
1390
1391 tx_ring->next_to_use = 0;
1392 tx_ring->next_to_clean = 0;
1393 spin_lock_init(&adapter->tx_queue_lock);
1394
1395 return 0;
1396 err:
1397 vfree(tx_ring->buffer_info);
1398 ndev_err(adapter->netdev,
1399 "Unable to allocate memory for the transmit descriptor ring\n");
1400 return err;
1401 }
1402
1403 /**
1404 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1405 * @adapter: board private structure
1406 *
1407 * Returns 0 on success, negative on failure
1408 **/
1409 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1410 {
1411 struct e1000_ring *rx_ring = adapter->rx_ring;
1412 struct e1000_buffer *buffer_info;
1413 int i, size, desc_len, err = -ENOMEM;
1414
1415 size = sizeof(struct e1000_buffer) * rx_ring->count;
1416 rx_ring->buffer_info = vmalloc(size);
1417 if (!rx_ring->buffer_info)
1418 goto err;
1419 memset(rx_ring->buffer_info, 0, size);
1420
1421 for (i = 0; i < rx_ring->count; i++) {
1422 buffer_info = &rx_ring->buffer_info[i];
1423 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1424 sizeof(struct e1000_ps_page),
1425 GFP_KERNEL);
1426 if (!buffer_info->ps_pages)
1427 goto err_pages;
1428 }
1429
1430 desc_len = sizeof(union e1000_rx_desc_packet_split);
1431
1432 /* Round up to nearest 4K */
1433 rx_ring->size = rx_ring->count * desc_len;
1434 rx_ring->size = ALIGN(rx_ring->size, 4096);
1435
1436 err = e1000_alloc_ring_dma(adapter, rx_ring);
1437 if (err)
1438 goto err_pages;
1439
1440 rx_ring->next_to_clean = 0;
1441 rx_ring->next_to_use = 0;
1442 rx_ring->rx_skb_top = NULL;
1443
1444 return 0;
1445
1446 err_pages:
1447 for (i = 0; i < rx_ring->count; i++) {
1448 buffer_info = &rx_ring->buffer_info[i];
1449 kfree(buffer_info->ps_pages);
1450 }
1451 err:
1452 vfree(rx_ring->buffer_info);
1453 ndev_err(adapter->netdev,
1454 "Unable to allocate memory for the transmit descriptor ring\n");
1455 return err;
1456 }
1457
1458 /**
1459 * e1000_clean_tx_ring - Free Tx Buffers
1460 * @adapter: board private structure
1461 **/
1462 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1463 {
1464 struct e1000_ring *tx_ring = adapter->tx_ring;
1465 struct e1000_buffer *buffer_info;
1466 unsigned long size;
1467 unsigned int i;
1468
1469 for (i = 0; i < tx_ring->count; i++) {
1470 buffer_info = &tx_ring->buffer_info[i];
1471 e1000_put_txbuf(adapter, buffer_info);
1472 }
1473
1474 size = sizeof(struct e1000_buffer) * tx_ring->count;
1475 memset(tx_ring->buffer_info, 0, size);
1476
1477 memset(tx_ring->desc, 0, tx_ring->size);
1478
1479 tx_ring->next_to_use = 0;
1480 tx_ring->next_to_clean = 0;
1481
1482 writel(0, adapter->hw.hw_addr + tx_ring->head);
1483 writel(0, adapter->hw.hw_addr + tx_ring->tail);
1484 }
1485
1486 /**
1487 * e1000e_free_tx_resources - Free Tx Resources per Queue
1488 * @adapter: board private structure
1489 *
1490 * Free all transmit software resources
1491 **/
1492 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1493 {
1494 struct pci_dev *pdev = adapter->pdev;
1495 struct e1000_ring *tx_ring = adapter->tx_ring;
1496
1497 e1000_clean_tx_ring(adapter);
1498
1499 vfree(tx_ring->buffer_info);
1500 tx_ring->buffer_info = NULL;
1501
1502 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1503 tx_ring->dma);
1504 tx_ring->desc = NULL;
1505 }
1506
1507 /**
1508 * e1000e_free_rx_resources - Free Rx Resources
1509 * @adapter: board private structure
1510 *
1511 * Free all receive software resources
1512 **/
1513
1514 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1515 {
1516 struct pci_dev *pdev = adapter->pdev;
1517 struct e1000_ring *rx_ring = adapter->rx_ring;
1518 int i;
1519
1520 e1000_clean_rx_ring(adapter);
1521
1522 for (i = 0; i < rx_ring->count; i++) {
1523 kfree(rx_ring->buffer_info[i].ps_pages);
1524 }
1525
1526 vfree(rx_ring->buffer_info);
1527 rx_ring->buffer_info = NULL;
1528
1529 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1530 rx_ring->dma);
1531 rx_ring->desc = NULL;
1532 }
1533
1534 /**
1535 * e1000_update_itr - update the dynamic ITR value based on statistics
1536 * @adapter: pointer to adapter
1537 * @itr_setting: current adapter->itr
1538 * @packets: the number of packets during this measurement interval
1539 * @bytes: the number of bytes during this measurement interval
1540 *
1541 * Stores a new ITR value based on packets and byte
1542 * counts during the last interrupt. The advantage of per interrupt
1543 * computation is faster updates and more accurate ITR for the current
1544 * traffic pattern. Constants in this function were computed
1545 * based on theoretical maximum wire speed and thresholds were set based
1546 * on testing data as well as attempting to minimize response time
1547 * while increasing bulk throughput.
1548 * this functionality is controlled by the InterruptThrottleRate module
1549 * parameter (see e1000_param.c)
1550 **/
1551 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1552 u16 itr_setting, int packets,
1553 int bytes)
1554 {
1555 unsigned int retval = itr_setting;
1556
1557 if (packets == 0)
1558 goto update_itr_done;
1559
1560 switch (itr_setting) {
1561 case lowest_latency:
1562 /* handle TSO and jumbo frames */
1563 if (bytes/packets > 8000)
1564 retval = bulk_latency;
1565 else if ((packets < 5) && (bytes > 512)) {
1566 retval = low_latency;
1567 }
1568 break;
1569 case low_latency: /* 50 usec aka 20000 ints/s */
1570 if (bytes > 10000) {
1571 /* this if handles the TSO accounting */
1572 if (bytes/packets > 8000) {
1573 retval = bulk_latency;
1574 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1575 retval = bulk_latency;
1576 } else if ((packets > 35)) {
1577 retval = lowest_latency;
1578 }
1579 } else if (bytes/packets > 2000) {
1580 retval = bulk_latency;
1581 } else if (packets <= 2 && bytes < 512) {
1582 retval = lowest_latency;
1583 }
1584 break;
1585 case bulk_latency: /* 250 usec aka 4000 ints/s */
1586 if (bytes > 25000) {
1587 if (packets > 35) {
1588 retval = low_latency;
1589 }
1590 } else if (bytes < 6000) {
1591 retval = low_latency;
1592 }
1593 break;
1594 }
1595
1596 update_itr_done:
1597 return retval;
1598 }
1599
1600 static void e1000_set_itr(struct e1000_adapter *adapter)
1601 {
1602 struct e1000_hw *hw = &adapter->hw;
1603 u16 current_itr;
1604 u32 new_itr = adapter->itr;
1605
1606 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1607 if (adapter->link_speed != SPEED_1000) {
1608 current_itr = 0;
1609 new_itr = 4000;
1610 goto set_itr_now;
1611 }
1612
1613 adapter->tx_itr = e1000_update_itr(adapter,
1614 adapter->tx_itr,
1615 adapter->total_tx_packets,
1616 adapter->total_tx_bytes);
1617 /* conservative mode (itr 3) eliminates the lowest_latency setting */
1618 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1619 adapter->tx_itr = low_latency;
1620
1621 adapter->rx_itr = e1000_update_itr(adapter,
1622 adapter->rx_itr,
1623 adapter->total_rx_packets,
1624 adapter->total_rx_bytes);
1625 /* conservative mode (itr 3) eliminates the lowest_latency setting */
1626 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1627 adapter->rx_itr = low_latency;
1628
1629 current_itr = max(adapter->rx_itr, adapter->tx_itr);
1630
1631 switch (current_itr) {
1632 /* counts and packets in update_itr are dependent on these numbers */
1633 case lowest_latency:
1634 new_itr = 70000;
1635 break;
1636 case low_latency:
1637 new_itr = 20000; /* aka hwitr = ~200 */
1638 break;
1639 case bulk_latency:
1640 new_itr = 4000;
1641 break;
1642 default:
1643 break;
1644 }
1645
1646 set_itr_now:
1647 if (new_itr != adapter->itr) {
1648 /*
1649 * this attempts to bias the interrupt rate towards Bulk
1650 * by adding intermediate steps when interrupt rate is
1651 * increasing
1652 */
1653 new_itr = new_itr > adapter->itr ?
1654 min(adapter->itr + (new_itr >> 2), new_itr) :
1655 new_itr;
1656 adapter->itr = new_itr;
1657 ew32(ITR, 1000000000 / (new_itr * 256));
1658 }
1659 }
1660
1661 /**
1662 * e1000_clean - NAPI Rx polling callback
1663 * @napi: struct associated with this polling callback
1664 * @budget: amount of packets driver is allowed to process this poll
1665 **/
1666 static int e1000_clean(struct napi_struct *napi, int budget)
1667 {
1668 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1669 struct net_device *poll_dev = adapter->netdev;
1670 int tx_cleaned = 0, work_done = 0;
1671
1672 /* Must NOT use netdev_priv macro here. */
1673 adapter = poll_dev->priv;
1674
1675 /*
1676 * e1000_clean is called per-cpu. This lock protects
1677 * tx_ring from being cleaned by multiple cpus
1678 * simultaneously. A failure obtaining the lock means
1679 * tx_ring is currently being cleaned anyway.
1680 */
1681 if (spin_trylock(&adapter->tx_queue_lock)) {
1682 tx_cleaned = e1000_clean_tx_irq(adapter);
1683 spin_unlock(&adapter->tx_queue_lock);
1684 }
1685
1686 adapter->clean_rx(adapter, &work_done, budget);
1687
1688 if (tx_cleaned)
1689 work_done = budget;
1690
1691 /* If budget not fully consumed, exit the polling mode */
1692 if (work_done < budget) {
1693 if (adapter->itr_setting & 3)
1694 e1000_set_itr(adapter);
1695 netif_rx_complete(poll_dev, napi);
1696 e1000_irq_enable(adapter);
1697 }
1698
1699 return work_done;
1700 }
1701
1702 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1703 {
1704 struct e1000_adapter *adapter = netdev_priv(netdev);
1705 struct e1000_hw *hw = &adapter->hw;
1706 u32 vfta, index;
1707
1708 /* don't update vlan cookie if already programmed */
1709 if ((adapter->hw.mng_cookie.status &
1710 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1711 (vid == adapter->mng_vlan_id))
1712 return;
1713 /* add VID to filter table */
1714 index = (vid >> 5) & 0x7F;
1715 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1716 vfta |= (1 << (vid & 0x1F));
1717 e1000e_write_vfta(hw, index, vfta);
1718 }
1719
1720 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1721 {
1722 struct e1000_adapter *adapter = netdev_priv(netdev);
1723 struct e1000_hw *hw = &adapter->hw;
1724 u32 vfta, index;
1725
1726 if (!test_bit(__E1000_DOWN, &adapter->state))
1727 e1000_irq_disable(adapter);
1728 vlan_group_set_device(adapter->vlgrp, vid, NULL);
1729
1730 if (!test_bit(__E1000_DOWN, &adapter->state))
1731 e1000_irq_enable(adapter);
1732
1733 if ((adapter->hw.mng_cookie.status &
1734 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1735 (vid == adapter->mng_vlan_id)) {
1736 /* release control to f/w */
1737 e1000_release_hw_control(adapter);
1738 return;
1739 }
1740
1741 /* remove VID from filter table */
1742 index = (vid >> 5) & 0x7F;
1743 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1744 vfta &= ~(1 << (vid & 0x1F));
1745 e1000e_write_vfta(hw, index, vfta);
1746 }
1747
1748 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
1749 {
1750 struct net_device *netdev = adapter->netdev;
1751 u16 vid = adapter->hw.mng_cookie.vlan_id;
1752 u16 old_vid = adapter->mng_vlan_id;
1753
1754 if (!adapter->vlgrp)
1755 return;
1756
1757 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
1758 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1759 if (adapter->hw.mng_cookie.status &
1760 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1761 e1000_vlan_rx_add_vid(netdev, vid);
1762 adapter->mng_vlan_id = vid;
1763 }
1764
1765 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
1766 (vid != old_vid) &&
1767 !vlan_group_get_device(adapter->vlgrp, old_vid))
1768 e1000_vlan_rx_kill_vid(netdev, old_vid);
1769 } else {
1770 adapter->mng_vlan_id = vid;
1771 }
1772 }
1773
1774
1775 static void e1000_vlan_rx_register(struct net_device *netdev,
1776 struct vlan_group *grp)
1777 {
1778 struct e1000_adapter *adapter = netdev_priv(netdev);
1779 struct e1000_hw *hw = &adapter->hw;
1780 u32 ctrl, rctl;
1781
1782 if (!test_bit(__E1000_DOWN, &adapter->state))
1783 e1000_irq_disable(adapter);
1784 adapter->vlgrp = grp;
1785
1786 if (grp) {
1787 /* enable VLAN tag insert/strip */
1788 ctrl = er32(CTRL);
1789 ctrl |= E1000_CTRL_VME;
1790 ew32(CTRL, ctrl);
1791
1792 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1793 /* enable VLAN receive filtering */
1794 rctl = er32(RCTL);
1795 rctl &= ~E1000_RCTL_CFIEN;
1796 ew32(RCTL, rctl);
1797 e1000_update_mng_vlan(adapter);
1798 }
1799 } else {
1800 /* disable VLAN tag insert/strip */
1801 ctrl = er32(CTRL);
1802 ctrl &= ~E1000_CTRL_VME;
1803 ew32(CTRL, ctrl);
1804
1805 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1806 if (adapter->mng_vlan_id !=
1807 (u16)E1000_MNG_VLAN_NONE) {
1808 e1000_vlan_rx_kill_vid(netdev,
1809 adapter->mng_vlan_id);
1810 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1811 }
1812 }
1813 }
1814
1815 if (!test_bit(__E1000_DOWN, &adapter->state))
1816 e1000_irq_enable(adapter);
1817 }
1818
1819 static void e1000_restore_vlan(struct e1000_adapter *adapter)
1820 {
1821 u16 vid;
1822
1823 e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
1824
1825 if (!adapter->vlgrp)
1826 return;
1827
1828 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1829 if (!vlan_group_get_device(adapter->vlgrp, vid))
1830 continue;
1831 e1000_vlan_rx_add_vid(adapter->netdev, vid);
1832 }
1833 }
1834
1835 static void e1000_init_manageability(struct e1000_adapter *adapter)
1836 {
1837 struct e1000_hw *hw = &adapter->hw;
1838 u32 manc, manc2h;
1839
1840 if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
1841 return;
1842
1843 manc = er32(MANC);
1844
1845 /*
1846 * enable receiving management packets to the host. this will probably
1847 * generate destination unreachable messages from the host OS, but
1848 * the packets will be handled on SMBUS
1849 */
1850 manc |= E1000_MANC_EN_MNG2HOST;
1851 manc2h = er32(MANC2H);
1852 #define E1000_MNG2HOST_PORT_623 (1 << 5)
1853 #define E1000_MNG2HOST_PORT_664 (1 << 6)
1854 manc2h |= E1000_MNG2HOST_PORT_623;
1855 manc2h |= E1000_MNG2HOST_PORT_664;
1856 ew32(MANC2H, manc2h);
1857 ew32(MANC, manc);
1858 }
1859
1860 /**
1861 * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1862 * @adapter: board private structure
1863 *
1864 * Configure the Tx unit of the MAC after a reset.
1865 **/
1866 static void e1000_configure_tx(struct e1000_adapter *adapter)
1867 {
1868 struct e1000_hw *hw = &adapter->hw;
1869 struct e1000_ring *tx_ring = adapter->tx_ring;
1870 u64 tdba;
1871 u32 tdlen, tctl, tipg, tarc;
1872 u32 ipgr1, ipgr2;
1873
1874 /* Setup the HW Tx Head and Tail descriptor pointers */
1875 tdba = tx_ring->dma;
1876 tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
1877 ew32(TDBAL, (tdba & DMA_32BIT_MASK));
1878 ew32(TDBAH, (tdba >> 32));
1879 ew32(TDLEN, tdlen);
1880 ew32(TDH, 0);
1881 ew32(TDT, 0);
1882 tx_ring->head = E1000_TDH;
1883 tx_ring->tail = E1000_TDT;
1884
1885 /* Set the default values for the Tx Inter Packet Gap timer */
1886 tipg = DEFAULT_82543_TIPG_IPGT_COPPER; /* 8 */
1887 ipgr1 = DEFAULT_82543_TIPG_IPGR1; /* 8 */
1888 ipgr2 = DEFAULT_82543_TIPG_IPGR2; /* 6 */
1889
1890 if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
1891 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /* 7 */
1892
1893 tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1894 tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1895 ew32(TIPG, tipg);
1896
1897 /* Set the Tx Interrupt Delay register */
1898 ew32(TIDV, adapter->tx_int_delay);
1899 /* Tx irq moderation */
1900 ew32(TADV, adapter->tx_abs_int_delay);
1901
1902 /* Program the Transmit Control Register */
1903 tctl = er32(TCTL);
1904 tctl &= ~E1000_TCTL_CT;
1905 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1906 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1907
1908 if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
1909 tarc = er32(TARC(0));
1910 /*
1911 * set the speed mode bit, we'll clear it if we're not at
1912 * gigabit link later
1913 */
1914 #define SPEED_MODE_BIT (1 << 21)
1915 tarc |= SPEED_MODE_BIT;
1916 ew32(TARC(0), tarc);
1917 }
1918
1919 /* errata: program both queues to unweighted RR */
1920 if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
1921 tarc = er32(TARC(0));
1922 tarc |= 1;
1923 ew32(TARC(0), tarc);
1924 tarc = er32(TARC(1));
1925 tarc |= 1;
1926 ew32(TARC(1), tarc);
1927 }
1928
1929 e1000e_config_collision_dist(hw);
1930
1931 /* Setup Transmit Descriptor Settings for eop descriptor */
1932 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1933
1934 /* only set IDE if we are delaying interrupts using the timers */
1935 if (adapter->tx_int_delay)
1936 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1937
1938 /* enable Report Status bit */
1939 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1940
1941 ew32(TCTL, tctl);
1942
1943 adapter->tx_queue_len = adapter->netdev->tx_queue_len;
1944 }
1945
1946 /**
1947 * e1000_setup_rctl - configure the receive control registers
1948 * @adapter: Board private structure
1949 **/
1950 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1951 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1952 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1953 {
1954 struct e1000_hw *hw = &adapter->hw;
1955 u32 rctl, rfctl;
1956 u32 psrctl = 0;
1957 u32 pages = 0;
1958
1959 /* Program MC offset vector base */
1960 rctl = er32(RCTL);
1961 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1962 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1963 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1964 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1965
1966 /* Do not Store bad packets */
1967 rctl &= ~E1000_RCTL_SBP;
1968
1969 /* Enable Long Packet receive */
1970 if (adapter->netdev->mtu <= ETH_DATA_LEN)
1971 rctl &= ~E1000_RCTL_LPE;
1972 else
1973 rctl |= E1000_RCTL_LPE;
1974
1975 /* Enable hardware CRC frame stripping */
1976 rctl |= E1000_RCTL_SECRC;
1977
1978 /* Setup buffer sizes */
1979 rctl &= ~E1000_RCTL_SZ_4096;
1980 rctl |= E1000_RCTL_BSEX;
1981 switch (adapter->rx_buffer_len) {
1982 case 256:
1983 rctl |= E1000_RCTL_SZ_256;
1984 rctl &= ~E1000_RCTL_BSEX;
1985 break;
1986 case 512:
1987 rctl |= E1000_RCTL_SZ_512;
1988 rctl &= ~E1000_RCTL_BSEX;
1989 break;
1990 case 1024:
1991 rctl |= E1000_RCTL_SZ_1024;
1992 rctl &= ~E1000_RCTL_BSEX;
1993 break;
1994 case 2048:
1995 default:
1996 rctl |= E1000_RCTL_SZ_2048;
1997 rctl &= ~E1000_RCTL_BSEX;
1998 break;
1999 case 4096:
2000 rctl |= E1000_RCTL_SZ_4096;
2001 break;
2002 case 8192:
2003 rctl |= E1000_RCTL_SZ_8192;
2004 break;
2005 case 16384:
2006 rctl |= E1000_RCTL_SZ_16384;
2007 break;
2008 }
2009
2010 /*
2011 * 82571 and greater support packet-split where the protocol
2012 * header is placed in skb->data and the packet data is
2013 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2014 * In the case of a non-split, skb->data is linearly filled,
2015 * followed by the page buffers. Therefore, skb->data is
2016 * sized to hold the largest protocol header.
2017 *
2018 * allocations using alloc_page take too long for regular MTU
2019 * so only enable packet split for jumbo frames
2020 *
2021 * Using pages when the page size is greater than 16k wastes
2022 * a lot of memory, since we allocate 3 pages at all times
2023 * per packet.
2024 */
2025 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2026 if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2027 (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2028 adapter->rx_ps_pages = pages;
2029 else
2030 adapter->rx_ps_pages = 0;
2031
2032 if (adapter->rx_ps_pages) {
2033 /* Configure extra packet-split registers */
2034 rfctl = er32(RFCTL);
2035 rfctl |= E1000_RFCTL_EXTEN;
2036 /*
2037 * disable packet split support for IPv6 extension headers,
2038 * because some malformed IPv6 headers can hang the Rx
2039 */
2040 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2041 E1000_RFCTL_NEW_IPV6_EXT_DIS);
2042
2043 ew32(RFCTL, rfctl);
2044
2045 /* Enable Packet split descriptors */
2046 rctl |= E1000_RCTL_DTYP_PS;
2047
2048 psrctl |= adapter->rx_ps_bsize0 >>
2049 E1000_PSRCTL_BSIZE0_SHIFT;
2050
2051 switch (adapter->rx_ps_pages) {
2052 case 3:
2053 psrctl |= PAGE_SIZE <<
2054 E1000_PSRCTL_BSIZE3_SHIFT;
2055 case 2:
2056 psrctl |= PAGE_SIZE <<
2057 E1000_PSRCTL_BSIZE2_SHIFT;
2058 case 1:
2059 psrctl |= PAGE_SIZE >>
2060 E1000_PSRCTL_BSIZE1_SHIFT;
2061 break;
2062 }
2063
2064 ew32(PSRCTL, psrctl);
2065 }
2066
2067 ew32(RCTL, rctl);
2068 /* just started the receive unit, no need to restart */
2069 adapter->flags &= ~FLAG_RX_RESTART_NOW;
2070 }
2071
2072 /**
2073 * e1000_configure_rx - Configure Receive Unit after Reset
2074 * @adapter: board private structure
2075 *
2076 * Configure the Rx unit of the MAC after a reset.
2077 **/
2078 static void e1000_configure_rx(struct e1000_adapter *adapter)
2079 {
2080 struct e1000_hw *hw = &adapter->hw;
2081 struct e1000_ring *rx_ring = adapter->rx_ring;
2082 u64 rdba;
2083 u32 rdlen, rctl, rxcsum, ctrl_ext;
2084
2085 if (adapter->rx_ps_pages) {
2086 /* this is a 32 byte descriptor */
2087 rdlen = rx_ring->count *
2088 sizeof(union e1000_rx_desc_packet_split);
2089 adapter->clean_rx = e1000_clean_rx_irq_ps;
2090 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2091 } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2092 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2093 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2094 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2095 } else {
2096 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2097 adapter->clean_rx = e1000_clean_rx_irq;
2098 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2099 }
2100
2101 /* disable receives while setting up the descriptors */
2102 rctl = er32(RCTL);
2103 ew32(RCTL, rctl & ~E1000_RCTL_EN);
2104 e1e_flush();
2105 msleep(10);
2106
2107 /* set the Receive Delay Timer Register */
2108 ew32(RDTR, adapter->rx_int_delay);
2109
2110 /* irq moderation */
2111 ew32(RADV, adapter->rx_abs_int_delay);
2112 if (adapter->itr_setting != 0)
2113 ew32(ITR, 1000000000 / (adapter->itr * 256));
2114
2115 ctrl_ext = er32(CTRL_EXT);
2116 /* Reset delay timers after every interrupt */
2117 ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2118 /* Auto-Mask interrupts upon ICR access */
2119 ctrl_ext |= E1000_CTRL_EXT_IAME;
2120 ew32(IAM, 0xffffffff);
2121 ew32(CTRL_EXT, ctrl_ext);
2122 e1e_flush();
2123
2124 /*
2125 * Setup the HW Rx Head and Tail Descriptor Pointers and
2126 * the Base and Length of the Rx Descriptor Ring
2127 */
2128 rdba = rx_ring->dma;
2129 ew32(RDBAL, (rdba & DMA_32BIT_MASK));
2130 ew32(RDBAH, (rdba >> 32));
2131 ew32(RDLEN, rdlen);
2132 ew32(RDH, 0);
2133 ew32(RDT, 0);
2134 rx_ring->head = E1000_RDH;
2135 rx_ring->tail = E1000_RDT;
2136
2137 /* Enable Receive Checksum Offload for TCP and UDP */
2138 rxcsum = er32(RXCSUM);
2139 if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2140 rxcsum |= E1000_RXCSUM_TUOFL;
2141
2142 /*
2143 * IPv4 payload checksum for UDP fragments must be
2144 * used in conjunction with packet-split.
2145 */
2146 if (adapter->rx_ps_pages)
2147 rxcsum |= E1000_RXCSUM_IPPCSE;
2148 } else {
2149 rxcsum &= ~E1000_RXCSUM_TUOFL;
2150 /* no need to clear IPPCSE as it defaults to 0 */
2151 }
2152 ew32(RXCSUM, rxcsum);
2153
2154 /*
2155 * Enable early receives on supported devices, only takes effect when
2156 * packet size is equal or larger than the specified value (in 8 byte
2157 * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2158 */
2159 if ((adapter->flags & FLAG_HAS_ERT) &&
2160 (adapter->netdev->mtu > ETH_DATA_LEN)) {
2161 u32 rxdctl = er32(RXDCTL(0));
2162 ew32(RXDCTL(0), rxdctl | 0x3);
2163 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2164 /*
2165 * With jumbo frames and early-receive enabled, excessive
2166 * C4->C2 latencies result in dropped transactions.
2167 */
2168 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2169 e1000e_driver_name, 55);
2170 } else {
2171 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2172 e1000e_driver_name,
2173 PM_QOS_DEFAULT_VALUE);
2174 }
2175
2176 /* Enable Receives */
2177 ew32(RCTL, rctl);
2178 }
2179
2180 /**
2181 * e1000_update_mc_addr_list - Update Multicast addresses
2182 * @hw: pointer to the HW structure
2183 * @mc_addr_list: array of multicast addresses to program
2184 * @mc_addr_count: number of multicast addresses to program
2185 * @rar_used_count: the first RAR register free to program
2186 * @rar_count: total number of supported Receive Address Registers
2187 *
2188 * Updates the Receive Address Registers and Multicast Table Array.
2189 * The caller must have a packed mc_addr_list of multicast addresses.
2190 * The parameter rar_count will usually be hw->mac.rar_entry_count
2191 * unless there are workarounds that change this. Currently no func pointer
2192 * exists and all implementations are handled in the generic version of this
2193 * function.
2194 **/
2195 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2196 u32 mc_addr_count, u32 rar_used_count,
2197 u32 rar_count)
2198 {
2199 hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2200 rar_used_count, rar_count);
2201 }
2202
2203 /**
2204 * e1000_set_multi - Multicast and Promiscuous mode set
2205 * @netdev: network interface device structure
2206 *
2207 * The set_multi entry point is called whenever the multicast address
2208 * list or the network interface flags are updated. This routine is
2209 * responsible for configuring the hardware for proper multicast,
2210 * promiscuous mode, and all-multi behavior.
2211 **/
2212 static void e1000_set_multi(struct net_device *netdev)
2213 {
2214 struct e1000_adapter *adapter = netdev_priv(netdev);
2215 struct e1000_hw *hw = &adapter->hw;
2216 struct e1000_mac_info *mac = &hw->mac;
2217 struct dev_mc_list *mc_ptr;
2218 u8 *mta_list;
2219 u32 rctl;
2220 int i;
2221
2222 /* Check for Promiscuous and All Multicast modes */
2223
2224 rctl = er32(RCTL);
2225
2226 if (netdev->flags & IFF_PROMISC) {
2227 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2228 rctl &= ~E1000_RCTL_VFE;
2229 } else {
2230 if (netdev->flags & IFF_ALLMULTI) {
2231 rctl |= E1000_RCTL_MPE;
2232 rctl &= ~E1000_RCTL_UPE;
2233 } else {
2234 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2235 }
2236 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2237 rctl |= E1000_RCTL_VFE;
2238 }
2239
2240 ew32(RCTL, rctl);
2241
2242 if (netdev->mc_count) {
2243 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2244 if (!mta_list)
2245 return;
2246
2247 /* prepare a packed array of only addresses. */
2248 mc_ptr = netdev->mc_list;
2249
2250 for (i = 0; i < netdev->mc_count; i++) {
2251 if (!mc_ptr)
2252 break;
2253 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2254 ETH_ALEN);
2255 mc_ptr = mc_ptr->next;
2256 }
2257
2258 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2259 mac->rar_entry_count);
2260 kfree(mta_list);
2261 } else {
2262 /*
2263 * if we're called from probe, we might not have
2264 * anything to do here, so clear out the list
2265 */
2266 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2267 }
2268 }
2269
2270 /**
2271 * e1000_configure - configure the hardware for Rx and Tx
2272 * @adapter: private board structure
2273 **/
2274 static void e1000_configure(struct e1000_adapter *adapter)
2275 {
2276 e1000_set_multi(adapter->netdev);
2277
2278 e1000_restore_vlan(adapter);
2279 e1000_init_manageability(adapter);
2280
2281 e1000_configure_tx(adapter);
2282 e1000_setup_rctl(adapter);
2283 e1000_configure_rx(adapter);
2284 adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2285 }
2286
2287 /**
2288 * e1000e_power_up_phy - restore link in case the phy was powered down
2289 * @adapter: address of board private structure
2290 *
2291 * The phy may be powered down to save power and turn off link when the
2292 * driver is unloaded and wake on lan is not enabled (among others)
2293 * *** this routine MUST be followed by a call to e1000e_reset ***
2294 **/
2295 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2296 {
2297 u16 mii_reg = 0;
2298
2299 /* Just clear the power down bit to wake the phy back up */
2300 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2301 /*
2302 * According to the manual, the phy will retain its
2303 * settings across a power-down/up cycle
2304 */
2305 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2306 mii_reg &= ~MII_CR_POWER_DOWN;
2307 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2308 }
2309
2310 adapter->hw.mac.ops.setup_link(&adapter->hw);
2311 }
2312
2313 /**
2314 * e1000_power_down_phy - Power down the PHY
2315 *
2316 * Power down the PHY so no link is implied when interface is down
2317 * The PHY cannot be powered down is management or WoL is active
2318 */
2319 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2320 {
2321 struct e1000_hw *hw = &adapter->hw;
2322 u16 mii_reg;
2323
2324 /* WoL is enabled */
2325 if (adapter->wol)
2326 return;
2327
2328 /* non-copper PHY? */
2329 if (adapter->hw.phy.media_type != e1000_media_type_copper)
2330 return;
2331
2332 /* reset is blocked because of a SoL/IDER session */
2333 if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2334 return;
2335
2336 /* manageability (AMT) is enabled */
2337 if (er32(MANC) & E1000_MANC_SMBUS_EN)
2338 return;
2339
2340 /* power down the PHY */
2341 e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2342 mii_reg |= MII_CR_POWER_DOWN;
2343 e1e_wphy(hw, PHY_CONTROL, mii_reg);
2344 mdelay(1);
2345 }
2346
2347 /**
2348 * e1000e_reset - bring the hardware into a known good state
2349 *
2350 * This function boots the hardware and enables some settings that
2351 * require a configuration cycle of the hardware - those cannot be
2352 * set/changed during runtime. After reset the device needs to be
2353 * properly configured for Rx, Tx etc.
2354 */
2355 void e1000e_reset(struct e1000_adapter *adapter)
2356 {
2357 struct e1000_mac_info *mac = &adapter->hw.mac;
2358 struct e1000_fc_info *fc = &adapter->hw.fc;
2359 struct e1000_hw *hw = &adapter->hw;
2360 u32 tx_space, min_tx_space, min_rx_space;
2361 u32 pba = adapter->pba;
2362 u16 hwm;
2363
2364 /* reset Packet Buffer Allocation to default */
2365 ew32(PBA, pba);
2366
2367 if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2368 /*
2369 * To maintain wire speed transmits, the Tx FIFO should be
2370 * large enough to accommodate two full transmit packets,
2371 * rounded up to the next 1KB and expressed in KB. Likewise,
2372 * the Rx FIFO should be large enough to accommodate at least
2373 * one full receive packet and is similarly rounded up and
2374 * expressed in KB.
2375 */
2376 pba = er32(PBA);
2377 /* upper 16 bits has Tx packet buffer allocation size in KB */
2378 tx_space = pba >> 16;
2379 /* lower 16 bits has Rx packet buffer allocation size in KB */
2380 pba &= 0xffff;
2381 /*
2382 * the Tx fifo also stores 16 bytes of information about the tx
2383 * but don't include ethernet FCS because hardware appends it
2384 */
2385 min_tx_space = (adapter->max_frame_size +
2386 sizeof(struct e1000_tx_desc) -
2387 ETH_FCS_LEN) * 2;
2388 min_tx_space = ALIGN(min_tx_space, 1024);
2389 min_tx_space >>= 10;
2390 /* software strips receive CRC, so leave room for it */
2391 min_rx_space = adapter->max_frame_size;
2392 min_rx_space = ALIGN(min_rx_space, 1024);
2393 min_rx_space >>= 10;
2394
2395 /*
2396 * If current Tx allocation is less than the min Tx FIFO size,
2397 * and the min Tx FIFO size is less than the current Rx FIFO
2398 * allocation, take space away from current Rx allocation
2399 */
2400 if ((tx_space < min_tx_space) &&
2401 ((min_tx_space - tx_space) < pba)) {
2402 pba -= min_tx_space - tx_space;
2403
2404 /*
2405 * if short on Rx space, Rx wins and must trump tx
2406 * adjustment or use Early Receive if available
2407 */
2408 if ((pba < min_rx_space) &&
2409 (!(adapter->flags & FLAG_HAS_ERT)))
2410 /* ERT enabled in e1000_configure_rx */
2411 pba = min_rx_space;
2412 }
2413
2414 ew32(PBA, pba);
2415 }
2416
2417
2418 /*
2419 * flow control settings
2420 *
2421 * The high water mark must be low enough to fit one full frame
2422 * (or the size used for early receive) above it in the Rx FIFO.
2423 * Set it to the lower of:
2424 * - 90% of the Rx FIFO size, and
2425 * - the full Rx FIFO size minus the early receive size (for parts
2426 * with ERT support assuming ERT set to E1000_ERT_2048), or
2427 * - the full Rx FIFO size minus one full frame
2428 */
2429 if (adapter->flags & FLAG_HAS_ERT)
2430 hwm = min(((pba << 10) * 9 / 10),
2431 ((pba << 10) - (E1000_ERT_2048 << 3)));
2432 else
2433 hwm = min(((pba << 10) * 9 / 10),
2434 ((pba << 10) - adapter->max_frame_size));
2435
2436 fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
2437 fc->low_water = fc->high_water - 8;
2438
2439 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2440 fc->pause_time = 0xFFFF;
2441 else
2442 fc->pause_time = E1000_FC_PAUSE_TIME;
2443 fc->send_xon = 1;
2444 fc->type = fc->original_type;
2445
2446 /* Allow time for pending master requests to run */
2447 mac->ops.reset_hw(hw);
2448
2449 /*
2450 * For parts with AMT enabled, let the firmware know
2451 * that the network interface is in control
2452 */
2453 if ((adapter->flags & FLAG_HAS_AMT) && e1000e_check_mng_mode(hw))
2454 e1000_get_hw_control(adapter);
2455
2456 ew32(WUC, 0);
2457
2458 if (mac->ops.init_hw(hw))
2459 ndev_err(adapter->netdev, "Hardware Error\n");
2460
2461 e1000_update_mng_vlan(adapter);
2462
2463 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2464 ew32(VET, ETH_P_8021Q);
2465
2466 e1000e_reset_adaptive(hw);
2467 e1000_get_phy_info(hw);
2468
2469 if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2470 u16 phy_data = 0;
2471 /*
2472 * speed up time to link by disabling smart power down, ignore
2473 * the return value of this function because there is nothing
2474 * different we would do if it failed
2475 */
2476 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2477 phy_data &= ~IGP02E1000_PM_SPD;
2478 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2479 }
2480 }
2481
2482 int e1000e_up(struct e1000_adapter *adapter)
2483 {
2484 struct e1000_hw *hw = &adapter->hw;
2485
2486 /* hardware has been reset, we need to reload some things */
2487 e1000_configure(adapter);
2488
2489 clear_bit(__E1000_DOWN, &adapter->state);
2490
2491 napi_enable(&adapter->napi);
2492 e1000_irq_enable(adapter);
2493
2494 /* fire a link change interrupt to start the watchdog */
2495 ew32(ICS, E1000_ICS_LSC);
2496 return 0;
2497 }
2498
2499 void e1000e_down(struct e1000_adapter *adapter)
2500 {
2501 struct net_device *netdev = adapter->netdev;
2502 struct e1000_hw *hw = &adapter->hw;
2503 u32 tctl, rctl;
2504
2505 /*
2506 * signal that we're down so the interrupt handler does not
2507 * reschedule our watchdog timer
2508 */
2509 set_bit(__E1000_DOWN, &adapter->state);
2510
2511 /* disable receives in the hardware */
2512 rctl = er32(RCTL);
2513 ew32(RCTL, rctl & ~E1000_RCTL_EN);
2514 /* flush and sleep below */
2515
2516 netif_tx_stop_all_queues(netdev);
2517
2518 /* disable transmits in the hardware */
2519 tctl = er32(TCTL);
2520 tctl &= ~E1000_TCTL_EN;
2521 ew32(TCTL, tctl);
2522 /* flush both disables and wait for them to finish */
2523 e1e_flush();
2524 msleep(10);
2525
2526 napi_disable(&adapter->napi);
2527 e1000_irq_disable(adapter);
2528
2529 del_timer_sync(&adapter->watchdog_timer);
2530 del_timer_sync(&adapter->phy_info_timer);
2531
2532 netdev->tx_queue_len = adapter->tx_queue_len;
2533 netif_carrier_off(netdev);
2534 adapter->link_speed = 0;
2535 adapter->link_duplex = 0;
2536
2537 if (!pci_channel_offline(adapter->pdev))
2538 e1000e_reset(adapter);
2539 e1000_clean_tx_ring(adapter);
2540 e1000_clean_rx_ring(adapter);
2541
2542 /*
2543 * TODO: for power management, we could drop the link and
2544 * pci_disable_device here.
2545 */
2546 }
2547
2548 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2549 {
2550 might_sleep();
2551 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2552 msleep(1);
2553 e1000e_down(adapter);
2554 e1000e_up(adapter);
2555 clear_bit(__E1000_RESETTING, &adapter->state);
2556 }
2557
2558 /**
2559 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2560 * @adapter: board private structure to initialize
2561 *
2562 * e1000_sw_init initializes the Adapter private data structure.
2563 * Fields are initialized based on PCI device information and
2564 * OS network device settings (MTU size).
2565 **/
2566 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2567 {
2568 struct net_device *netdev = adapter->netdev;
2569
2570 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2571 adapter->rx_ps_bsize0 = 128;
2572 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2573 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2574
2575 adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2576 if (!adapter->tx_ring)
2577 goto err;
2578
2579 adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2580 if (!adapter->rx_ring)
2581 goto err;
2582
2583 spin_lock_init(&adapter->tx_queue_lock);
2584
2585 /* Explicitly disable IRQ since the NIC can be in any state. */
2586 e1000_irq_disable(adapter);
2587
2588 spin_lock_init(&adapter->stats_lock);
2589
2590 set_bit(__E1000_DOWN, &adapter->state);
2591 return 0;
2592
2593 err:
2594 ndev_err(netdev, "Unable to allocate memory for queues\n");
2595 kfree(adapter->rx_ring);
2596 kfree(adapter->tx_ring);
2597 return -ENOMEM;
2598 }
2599
2600 /**
2601 * e1000_open - Called when a network interface is made active
2602 * @netdev: network interface device structure
2603 *
2604 * Returns 0 on success, negative value on failure
2605 *
2606 * The open entry point is called when a network interface is made
2607 * active by the system (IFF_UP). At this point all resources needed
2608 * for transmit and receive operations are allocated, the interrupt
2609 * handler is registered with the OS, the watchdog timer is started,
2610 * and the stack is notified that the interface is ready.
2611 **/
2612 static int e1000_open(struct net_device *netdev)
2613 {
2614 struct e1000_adapter *adapter = netdev_priv(netdev);
2615 struct e1000_hw *hw = &adapter->hw;
2616 int err;
2617
2618 /* disallow open during test */
2619 if (test_bit(__E1000_TESTING, &adapter->state))
2620 return -EBUSY;
2621
2622 /* allocate transmit descriptors */
2623 err = e1000e_setup_tx_resources(adapter);
2624 if (err)
2625 goto err_setup_tx;
2626
2627 /* allocate receive descriptors */
2628 err = e1000e_setup_rx_resources(adapter);
2629 if (err)
2630 goto err_setup_rx;
2631
2632 e1000e_power_up_phy(adapter);
2633
2634 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2635 if ((adapter->hw.mng_cookie.status &
2636 E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
2637 e1000_update_mng_vlan(adapter);
2638
2639 /*
2640 * If AMT is enabled, let the firmware know that the network
2641 * interface is now open
2642 */
2643 if ((adapter->flags & FLAG_HAS_AMT) &&
2644 e1000e_check_mng_mode(&adapter->hw))
2645 e1000_get_hw_control(adapter);
2646
2647 /*
2648 * before we allocate an interrupt, we must be ready to handle it.
2649 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2650 * as soon as we call pci_request_irq, so we have to setup our
2651 * clean_rx handler before we do so.
2652 */
2653 e1000_configure(adapter);
2654
2655 err = e1000_request_irq(adapter);
2656 if (err)
2657 goto err_req_irq;
2658
2659 /* From here on the code is the same as e1000e_up() */
2660 clear_bit(__E1000_DOWN, &adapter->state);
2661
2662 napi_enable(&adapter->napi);
2663
2664 e1000_irq_enable(adapter);
2665
2666 netif_tx_start_all_queues(netdev);
2667
2668 /* fire a link status change interrupt to start the watchdog */
2669 ew32(ICS, E1000_ICS_LSC);
2670
2671 return 0;
2672
2673 err_req_irq:
2674 e1000_release_hw_control(adapter);
2675 e1000_power_down_phy(adapter);
2676 e1000e_free_rx_resources(adapter);
2677 err_setup_rx:
2678 e1000e_free_tx_resources(adapter);
2679 err_setup_tx:
2680 e1000e_reset(adapter);
2681
2682 return err;
2683 }
2684
2685 /**
2686 * e1000_close - Disables a network interface
2687 * @netdev: network interface device structure
2688 *
2689 * Returns 0, this is not allowed to fail
2690 *
2691 * The close entry point is called when an interface is de-activated
2692 * by the OS. The hardware is still under the drivers control, but
2693 * needs to be disabled. A global MAC reset is issued to stop the
2694 * hardware, and all transmit and receive resources are freed.
2695 **/
2696 static int e1000_close(struct net_device *netdev)
2697 {
2698 struct e1000_adapter *adapter = netdev_priv(netdev);
2699
2700 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
2701 e1000e_down(adapter);
2702 e1000_power_down_phy(adapter);
2703 e1000_free_irq(adapter);
2704
2705 e1000e_free_tx_resources(adapter);
2706 e1000e_free_rx_resources(adapter);
2707
2708 /*
2709 * kill manageability vlan ID if supported, but not if a vlan with
2710 * the same ID is registered on the host OS (let 8021q kill it)
2711 */
2712 if ((adapter->hw.mng_cookie.status &
2713 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2714 !(adapter->vlgrp &&
2715 vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
2716 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2717
2718 /*
2719 * If AMT is enabled, let the firmware know that the network
2720 * interface is now closed
2721 */
2722 if ((adapter->flags & FLAG_HAS_AMT) &&
2723 e1000e_check_mng_mode(&adapter->hw))
2724 e1000_release_hw_control(adapter);
2725
2726 return 0;
2727 }
2728 /**
2729 * e1000_set_mac - Change the Ethernet Address of the NIC
2730 * @netdev: network interface device structure
2731 * @p: pointer to an address structure
2732 *
2733 * Returns 0 on success, negative on failure
2734 **/
2735 static int e1000_set_mac(struct net_device *netdev, void *p)
2736 {
2737 struct e1000_adapter *adapter = netdev_priv(netdev);
2738 struct sockaddr *addr = p;
2739
2740 if (!is_valid_ether_addr(addr->sa_data))
2741 return -EADDRNOTAVAIL;
2742
2743 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2744 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
2745
2746 e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
2747
2748 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
2749 /* activate the work around */
2750 e1000e_set_laa_state_82571(&adapter->hw, 1);
2751
2752 /*
2753 * Hold a copy of the LAA in RAR[14] This is done so that
2754 * between the time RAR[0] gets clobbered and the time it
2755 * gets fixed (in e1000_watchdog), the actual LAA is in one
2756 * of the RARs and no incoming packets directed to this port
2757 * are dropped. Eventually the LAA will be in RAR[0] and
2758 * RAR[14]
2759 */
2760 e1000e_rar_set(&adapter->hw,
2761 adapter->hw.mac.addr,
2762 adapter->hw.mac.rar_entry_count - 1);
2763 }
2764
2765 return 0;
2766 }
2767
2768 /*
2769 * Need to wait a few seconds after link up to get diagnostic information from
2770 * the phy
2771 */
2772 static void e1000_update_phy_info(unsigned long data)
2773 {
2774 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2775 e1000_get_phy_info(&adapter->hw);
2776 }
2777
2778 /**
2779 * e1000e_update_stats - Update the board statistics counters
2780 * @adapter: board private structure
2781 **/
2782 void e1000e_update_stats(struct e1000_adapter *adapter)
2783 {
2784 struct e1000_hw *hw = &adapter->hw;
2785 struct pci_dev *pdev = adapter->pdev;
2786 unsigned long irq_flags;
2787 u16 phy_tmp;
2788
2789 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
2790
2791 /*
2792 * Prevent stats update while adapter is being reset, or if the pci
2793 * connection is down.
2794 */
2795 if (adapter->link_speed == 0)
2796 return;
2797 if (pci_channel_offline(pdev))
2798 return;
2799
2800 spin_lock_irqsave(&adapter->stats_lock, irq_flags);
2801
2802 /*
2803 * these counters are modified from e1000_adjust_tbi_stats,
2804 * called from the interrupt context, so they must only
2805 * be written while holding adapter->stats_lock
2806 */
2807
2808 adapter->stats.crcerrs += er32(CRCERRS);
2809 adapter->stats.gprc += er32(GPRC);
2810 adapter->stats.gorc += er32(GORCL);
2811 er32(GORCH); /* Clear gorc */
2812 adapter->stats.bprc += er32(BPRC);
2813 adapter->stats.mprc += er32(MPRC);
2814 adapter->stats.roc += er32(ROC);
2815
2816 adapter->stats.mpc += er32(MPC);
2817 adapter->stats.scc += er32(SCC);
2818 adapter->stats.ecol += er32(ECOL);
2819 adapter->stats.mcc += er32(MCC);
2820 adapter->stats.latecol += er32(LATECOL);
2821 adapter->stats.dc += er32(DC);
2822 adapter->stats.xonrxc += er32(XONRXC);
2823 adapter->stats.xontxc += er32(XONTXC);
2824 adapter->stats.xoffrxc += er32(XOFFRXC);
2825 adapter->stats.xofftxc += er32(XOFFTXC);
2826 adapter->stats.gptc += er32(GPTC);
2827 adapter->stats.gotc += er32(GOTCL);
2828 er32(GOTCH); /* Clear gotc */
2829 adapter->stats.rnbc += er32(RNBC);
2830 adapter->stats.ruc += er32(RUC);
2831
2832 adapter->stats.mptc += er32(MPTC);
2833 adapter->stats.bptc += er32(BPTC);
2834
2835 /* used for adaptive IFS */
2836
2837 hw->mac.tx_packet_delta = er32(TPT);
2838 adapter->stats.tpt += hw->mac.tx_packet_delta;
2839 hw->mac.collision_delta = er32(COLC);
2840 adapter->stats.colc += hw->mac.collision_delta;
2841
2842 adapter->stats.algnerrc += er32(ALGNERRC);
2843 adapter->stats.rxerrc += er32(RXERRC);
2844 adapter->stats.tncrs += er32(TNCRS);
2845 adapter->stats.cexterr += er32(CEXTERR);
2846 adapter->stats.tsctc += er32(TSCTC);
2847 adapter->stats.tsctfc += er32(TSCTFC);
2848
2849 /* Fill out the OS statistics structure */
2850 adapter->net_stats.multicast = adapter->stats.mprc;
2851 adapter->net_stats.collisions = adapter->stats.colc;
2852
2853 /* Rx Errors */
2854
2855 /*
2856 * RLEC on some newer hardware can be incorrect so build
2857 * our own version based on RUC and ROC
2858 */
2859 adapter->net_stats.rx_errors = adapter->stats.rxerrc +
2860 adapter->stats.crcerrs + adapter->stats.algnerrc +
2861 adapter->stats.ruc + adapter->stats.roc +
2862 adapter->stats.cexterr;
2863 adapter->net_stats.rx_length_errors = adapter->stats.ruc +
2864 adapter->stats.roc;
2865 adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
2866 adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
2867 adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
2868
2869 /* Tx Errors */
2870 adapter->net_stats.tx_errors = adapter->stats.ecol +
2871 adapter->stats.latecol;
2872 adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
2873 adapter->net_stats.tx_window_errors = adapter->stats.latecol;
2874 adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
2875
2876 /* Tx Dropped needs to be maintained elsewhere */
2877
2878 /* Phy Stats */
2879 if (hw->phy.media_type == e1000_media_type_copper) {
2880 if ((adapter->link_speed == SPEED_1000) &&
2881 (!e1e_rphy(hw, PHY_1000T_STATUS, &phy_tmp))) {
2882 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
2883 adapter->phy_stats.idle_errors += phy_tmp;
2884 }
2885 }
2886
2887 /* Management Stats */
2888 adapter->stats.mgptc += er32(MGTPTC);
2889 adapter->stats.mgprc += er32(MGTPRC);
2890 adapter->stats.mgpdc += er32(MGTPDC);
2891
2892 spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
2893 }
2894
2895 /**
2896 * e1000_phy_read_status - Update the PHY register status snapshot
2897 * @adapter: board private structure
2898 **/
2899 static void e1000_phy_read_status(struct e1000_adapter *adapter)
2900 {
2901 struct e1000_hw *hw = &adapter->hw;
2902 struct e1000_phy_regs *phy = &adapter->phy_regs;
2903 int ret_val;
2904 unsigned long irq_flags;
2905
2906
2907 spin_lock_irqsave(&adapter->stats_lock, irq_flags);
2908
2909 if ((er32(STATUS) & E1000_STATUS_LU) &&
2910 (adapter->hw.phy.media_type == e1000_media_type_copper)) {
2911 ret_val = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
2912 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
2913 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
2914 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
2915 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
2916 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
2917 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
2918 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
2919 if (ret_val)
2920 ndev_warn(adapter->netdev,
2921 "Error reading PHY register\n");
2922 } else {
2923 /*
2924 * Do not read PHY registers if link is not up
2925 * Set values to typical power-on defaults
2926 */
2927 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
2928 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
2929 BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
2930 BMSR_ERCAP);
2931 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
2932 ADVERTISE_ALL | ADVERTISE_CSMA);
2933 phy->lpa = 0;
2934 phy->expansion = EXPANSION_ENABLENPAGE;
2935 phy->ctrl1000 = ADVERTISE_1000FULL;
2936 phy->stat1000 = 0;
2937 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
2938 }
2939
2940 spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
2941 }
2942
2943 static void e1000_print_link_info(struct e1000_adapter *adapter)
2944 {
2945 struct e1000_hw *hw = &adapter->hw;
2946 struct net_device *netdev = adapter->netdev;
2947 u32 ctrl = er32(CTRL);
2948
2949 ndev_info(netdev,
2950 "Link is Up %d Mbps %s, Flow Control: %s\n",
2951 adapter->link_speed,
2952 (adapter->link_duplex == FULL_DUPLEX) ?
2953 "Full Duplex" : "Half Duplex",
2954 ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
2955 "RX/TX" :
2956 ((ctrl & E1000_CTRL_RFCE) ? "RX" :
2957 ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
2958 }
2959
2960 static bool e1000_has_link(struct e1000_adapter *adapter)
2961 {
2962 struct e1000_hw *hw = &adapter->hw;
2963 bool link_active = 0;
2964 s32 ret_val = 0;
2965
2966 /*
2967 * get_link_status is set on LSC (link status) interrupt or
2968 * Rx sequence error interrupt. get_link_status will stay
2969 * false until the check_for_link establishes link
2970 * for copper adapters ONLY
2971 */
2972 switch (hw->phy.media_type) {
2973 case e1000_media_type_copper:
2974 if (hw->mac.get_link_status) {
2975 ret_val = hw->mac.ops.check_for_link(hw);
2976 link_active = !hw->mac.get_link_status;
2977 } else {
2978 link_active = 1;
2979 }
2980 break;
2981 case e1000_media_type_fiber:
2982 ret_val = hw->mac.ops.check_for_link(hw);
2983 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2984 break;
2985 case e1000_media_type_internal_serdes:
2986 ret_val = hw->mac.ops.check_for_link(hw);
2987 link_active = adapter->hw.mac.serdes_has_link;
2988 break;
2989 default:
2990 case e1000_media_type_unknown:
2991 break;
2992 }
2993
2994 if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
2995 (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
2996 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
2997 ndev_info(adapter->netdev,
2998 "Gigabit has been disabled, downgrading speed\n");
2999 }
3000
3001 return link_active;
3002 }
3003
3004 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3005 {
3006 /* make sure the receive unit is started */
3007 if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3008 (adapter->flags & FLAG_RX_RESTART_NOW)) {
3009 struct e1000_hw *hw = &adapter->hw;
3010 u32 rctl = er32(RCTL);
3011 ew32(RCTL, rctl | E1000_RCTL_EN);
3012 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3013 }
3014 }
3015
3016 /**
3017 * e1000_watchdog - Timer Call-back
3018 * @data: pointer to adapter cast into an unsigned long
3019 **/
3020 static void e1000_watchdog(unsigned long data)
3021 {
3022 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3023
3024 /* Do the rest outside of interrupt context */
3025 schedule_work(&adapter->watchdog_task);
3026
3027 /* TODO: make this use queue_delayed_work() */
3028 }
3029
3030 static void e1000_watchdog_task(struct work_struct *work)
3031 {
3032 struct e1000_adapter *adapter = container_of(work,
3033 struct e1000_adapter, watchdog_task);
3034 struct net_device *netdev = adapter->netdev;
3035 struct e1000_mac_info *mac = &adapter->hw.mac;
3036 struct e1000_ring *tx_ring = adapter->tx_ring;
3037 struct e1000_hw *hw = &adapter->hw;
3038 u32 link, tctl;
3039 int tx_pending = 0;
3040
3041 link = e1000_has_link(adapter);
3042 if ((netif_carrier_ok(netdev)) && link) {
3043 e1000e_enable_receives(adapter);
3044 goto link_up;
3045 }
3046
3047 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3048 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3049 e1000_update_mng_vlan(adapter);
3050
3051 if (link) {
3052 if (!netif_carrier_ok(netdev)) {
3053 bool txb2b = 1;
3054 /* update snapshot of PHY registers on LSC */
3055 e1000_phy_read_status(adapter);
3056 mac->ops.get_link_up_info(&adapter->hw,
3057 &adapter->link_speed,
3058 &adapter->link_duplex);
3059 e1000_print_link_info(adapter);
3060 /*
3061 * tweak tx_queue_len according to speed/duplex
3062 * and adjust the timeout factor
3063 */
3064 netdev->tx_queue_len = adapter->tx_queue_len;
3065 adapter->tx_timeout_factor = 1;
3066 switch (adapter->link_speed) {
3067 case SPEED_10:
3068 txb2b = 0;
3069 netdev->tx_queue_len = 10;
3070 adapter->tx_timeout_factor = 14;
3071 break;
3072 case SPEED_100:
3073 txb2b = 0;
3074 netdev->tx_queue_len = 100;
3075 /* maybe add some timeout factor ? */
3076 break;
3077 }
3078
3079 /*
3080 * workaround: re-program speed mode bit after
3081 * link-up event
3082 */
3083 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3084 !txb2b) {
3085 u32 tarc0;
3086 tarc0 = er32(TARC(0));
3087 tarc0 &= ~SPEED_MODE_BIT;
3088 ew32(TARC(0), tarc0);
3089 }
3090
3091 /*
3092 * disable TSO for pcie and 10/100 speeds, to avoid
3093 * some hardware issues
3094 */
3095 if (!(adapter->flags & FLAG_TSO_FORCE)) {
3096 switch (adapter->link_speed) {
3097 case SPEED_10:
3098 case SPEED_100:
3099 ndev_info(netdev,
3100 "10/100 speed: disabling TSO\n");
3101 netdev->features &= ~NETIF_F_TSO;
3102 netdev->features &= ~NETIF_F_TSO6;
3103 break;
3104 case SPEED_1000:
3105 netdev->features |= NETIF_F_TSO;
3106 netdev->features |= NETIF_F_TSO6;
3107 break;
3108 default:
3109 /* oops */
3110 break;
3111 }
3112 }
3113
3114 /*
3115 * enable transmits in the hardware, need to do this
3116 * after setting TARC(0)
3117 */
3118 tctl = er32(TCTL);
3119 tctl |= E1000_TCTL_EN;
3120 ew32(TCTL, tctl);
3121
3122 netif_carrier_on(netdev);
3123 netif_tx_wake_all_queues(netdev);
3124
3125 if (!test_bit(__E1000_DOWN, &adapter->state))
3126 mod_timer(&adapter->phy_info_timer,
3127 round_jiffies(jiffies + 2 * HZ));
3128 }
3129 } else {
3130 if (netif_carrier_ok(netdev)) {
3131 adapter->link_speed = 0;
3132 adapter->link_duplex = 0;
3133 ndev_info(netdev, "Link is Down\n");
3134 netif_carrier_off(netdev);
3135 netif_tx_stop_all_queues(netdev);
3136 if (!test_bit(__E1000_DOWN, &adapter->state))
3137 mod_timer(&adapter->phy_info_timer,
3138 round_jiffies(jiffies + 2 * HZ));
3139
3140 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3141 schedule_work(&adapter->reset_task);
3142 }
3143 }
3144
3145 link_up:
3146 e1000e_update_stats(adapter);
3147
3148 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3149 adapter->tpt_old = adapter->stats.tpt;
3150 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3151 adapter->colc_old = adapter->stats.colc;
3152
3153 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3154 adapter->gorc_old = adapter->stats.gorc;
3155 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3156 adapter->gotc_old = adapter->stats.gotc;
3157
3158 e1000e_update_adaptive(&adapter->hw);
3159
3160 if (!netif_carrier_ok(netdev)) {
3161 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3162 tx_ring->count);
3163 if (tx_pending) {
3164 /*
3165 * We've lost link, so the controller stops DMA,
3166 * but we've got queued Tx work that's never going
3167 * to get done, so reset controller to flush Tx.
3168 * (Do the reset outside of interrupt context).
3169 */
3170 adapter->tx_timeout_count++;
3171 schedule_work(&adapter->reset_task);
3172 }
3173 }
3174
3175 /* Cause software interrupt to ensure Rx ring is cleaned */
3176 ew32(ICS, E1000_ICS_RXDMT0);
3177
3178 /* Force detection of hung controller every watchdog period */
3179 adapter->detect_tx_hung = 1;
3180
3181 /*
3182 * With 82571 controllers, LAA may be overwritten due to controller
3183 * reset from the other port. Set the appropriate LAA in RAR[0]
3184 */
3185 if (e1000e_get_laa_state_82571(hw))
3186 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3187
3188 /* Reset the timer */
3189 if (!test_bit(__E1000_DOWN, &adapter->state))
3190 mod_timer(&adapter->watchdog_timer,
3191 round_jiffies(jiffies + 2 * HZ));
3192 }
3193
3194 #define E1000_TX_FLAGS_CSUM 0x00000001
3195 #define E1000_TX_FLAGS_VLAN 0x00000002
3196 #define E1000_TX_FLAGS_TSO 0x00000004
3197 #define E1000_TX_FLAGS_IPV4 0x00000008
3198 #define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
3199 #define E1000_TX_FLAGS_VLAN_SHIFT 16
3200
3201 static int e1000_tso(struct e1000_adapter *adapter,
3202 struct sk_buff *skb)
3203 {
3204 struct e1000_ring *tx_ring = adapter->tx_ring;
3205 struct e1000_context_desc *context_desc;
3206 struct e1000_buffer *buffer_info;
3207 unsigned int i;
3208 u32 cmd_length = 0;
3209 u16 ipcse = 0, tucse, mss;
3210 u8 ipcss, ipcso, tucss, tucso, hdr_len;
3211 int err;
3212
3213 if (skb_is_gso(skb)) {
3214 if (skb_header_cloned(skb)) {
3215 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3216 if (err)
3217 return err;
3218 }
3219
3220 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3221 mss = skb_shinfo(skb)->gso_size;
3222 if (skb->protocol == htons(ETH_P_IP)) {
3223 struct iphdr *iph = ip_hdr(skb);
3224 iph->tot_len = 0;
3225 iph->check = 0;
3226 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3227 iph->daddr, 0,
3228 IPPROTO_TCP,
3229 0);
3230 cmd_length = E1000_TXD_CMD_IP;
3231 ipcse = skb_transport_offset(skb) - 1;
3232 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3233 ipv6_hdr(skb)->payload_len = 0;
3234 tcp_hdr(skb)->check =
3235 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3236 &ipv6_hdr(skb)->daddr,
3237 0, IPPROTO_TCP, 0);
3238 ipcse = 0;
3239 }
3240 ipcss = skb_network_offset(skb);
3241 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3242 tucss = skb_transport_offset(skb);
3243 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3244 tucse = 0;
3245
3246 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3247 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3248
3249 i = tx_ring->next_to_use;
3250 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3251 buffer_info = &tx_ring->buffer_info[i];
3252
3253 context_desc->lower_setup.ip_fields.ipcss = ipcss;
3254 context_desc->lower_setup.ip_fields.ipcso = ipcso;
3255 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
3256 context_desc->upper_setup.tcp_fields.tucss = tucss;
3257 context_desc->upper_setup.tcp_fields.tucso = tucso;
3258 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3259 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
3260 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3261 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3262
3263 buffer_info->time_stamp = jiffies;
3264 buffer_info->next_to_watch = i;
3265
3266 i++;
3267 if (i == tx_ring->count)
3268 i = 0;
3269 tx_ring->next_to_use = i;
3270
3271 return 1;
3272 }
3273
3274 return 0;
3275 }
3276
3277 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3278 {
3279 struct e1000_ring *tx_ring = adapter->tx_ring;
3280 struct e1000_context_desc *context_desc;
3281 struct e1000_buffer *buffer_info;
3282 unsigned int i;
3283 u8 css;
3284
3285 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3286 css = skb_transport_offset(skb);
3287
3288 i = tx_ring->next_to_use;
3289 buffer_info = &tx_ring->buffer_info[i];
3290 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3291
3292 context_desc->lower_setup.ip_config = 0;
3293 context_desc->upper_setup.tcp_fields.tucss = css;
3294 context_desc->upper_setup.tcp_fields.tucso =
3295 css + skb->csum_offset;
3296 context_desc->upper_setup.tcp_fields.tucse = 0;
3297 context_desc->tcp_seg_setup.data = 0;
3298 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
3299
3300 buffer_info->time_stamp = jiffies;
3301 buffer_info->next_to_watch = i;
3302
3303 i++;
3304 if (i == tx_ring->count)
3305 i = 0;
3306 tx_ring->next_to_use = i;
3307
3308 return 1;
3309 }
3310
3311 return 0;
3312 }
3313
3314 #define E1000_MAX_PER_TXD 8192
3315 #define E1000_MAX_TXD_PWR 12
3316
3317 static int e1000_tx_map(struct e1000_adapter *adapter,
3318 struct sk_buff *skb, unsigned int first,
3319 unsigned int max_per_txd, unsigned int nr_frags,
3320 unsigned int mss)
3321 {
3322 struct e1000_ring *tx_ring = adapter->tx_ring;
3323 struct e1000_buffer *buffer_info;
3324 unsigned int len = skb->len - skb->data_len;
3325 unsigned int offset = 0, size, count = 0, i;
3326 unsigned int f;
3327
3328 i = tx_ring->next_to_use;
3329
3330 while (len) {
3331 buffer_info = &tx_ring->buffer_info[i];
3332 size = min(len, max_per_txd);
3333
3334 /* Workaround for premature desc write-backs
3335 * in TSO mode. Append 4-byte sentinel desc */
3336 if (mss && !nr_frags && size == len && size > 8)
3337 size -= 4;
3338
3339 buffer_info->length = size;
3340 /* set time_stamp *before* dma to help avoid a possible race */
3341 buffer_info->time_stamp = jiffies;
3342 buffer_info->dma =
3343 pci_map_single(adapter->pdev,
3344 skb->data + offset,
3345 size,
3346 PCI_DMA_TODEVICE);
3347 if (pci_dma_mapping_error(buffer_info->dma)) {
3348 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3349 adapter->tx_dma_failed++;
3350 return -1;
3351 }
3352 buffer_info->next_to_watch = i;
3353
3354 len -= size;
3355 offset += size;
3356 count++;
3357 i++;
3358 if (i == tx_ring->count)
3359 i = 0;
3360 }
3361
3362 for (f = 0; f < nr_frags; f++) {
3363 struct skb_frag_struct *frag;
3364
3365 frag = &skb_shinfo(skb)->frags[f];
3366 len = frag->size;
3367 offset = frag->page_offset;
3368
3369 while (len) {
3370 buffer_info = &tx_ring->buffer_info[i];
3371 size = min(len, max_per_txd);
3372 /* Workaround for premature desc write-backs
3373 * in TSO mode. Append 4-byte sentinel desc */
3374 if (mss && f == (nr_frags-1) && size == len && size > 8)
3375 size -= 4;
3376
3377 buffer_info->length = size;
3378 buffer_info->time_stamp = jiffies;
3379 buffer_info->dma =
3380 pci_map_page(adapter->pdev,
3381 frag->page,
3382 offset,
3383 size,
3384 PCI_DMA_TODEVICE);
3385 if (pci_dma_mapping_error(buffer_info->dma)) {
3386 dev_err(&adapter->pdev->dev,
3387 "TX DMA page map failed\n");
3388 adapter->tx_dma_failed++;
3389 return -1;
3390 }
3391
3392 buffer_info->next_to_watch = i;
3393
3394 len -= size;
3395 offset += size;
3396 count++;
3397
3398 i++;
3399 if (i == tx_ring->count)
3400 i = 0;
3401 }
3402 }
3403
3404 if (i == 0)
3405 i = tx_ring->count - 1;
3406 else
3407 i--;
3408
3409 tx_ring->buffer_info[i].skb = skb;
3410 tx_ring->buffer_info[first].next_to_watch = i;
3411
3412 return count;
3413 }
3414
3415 static void e1000_tx_queue(struct e1000_adapter *adapter,
3416 int tx_flags, int count)
3417 {
3418 struct e1000_ring *tx_ring = adapter->tx_ring;
3419 struct e1000_tx_desc *tx_desc = NULL;
3420 struct e1000_buffer *buffer_info;
3421 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3422 unsigned int i;
3423
3424 if (tx_flags & E1000_TX_FLAGS_TSO) {
3425 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3426 E1000_TXD_CMD_TSE;
3427 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3428
3429 if (tx_flags & E1000_TX_FLAGS_IPV4)
3430 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3431 }
3432
3433 if (tx_flags & E1000_TX_FLAGS_CSUM) {
3434 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3435 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3436 }
3437
3438 if (tx_flags & E1000_TX_FLAGS_VLAN) {
3439 txd_lower |= E1000_TXD_CMD_VLE;
3440 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3441 }
3442
3443 i = tx_ring->next_to_use;
3444
3445 while (count--) {
3446 buffer_info = &tx_ring->buffer_info[i];
3447 tx_desc = E1000_TX_DESC(*tx_ring, i);
3448 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3449 tx_desc->lower.data =
3450 cpu_to_le32(txd_lower | buffer_info->length);
3451 tx_desc->upper.data = cpu_to_le32(txd_upper);
3452
3453 i++;
3454 if (i == tx_ring->count)
3455 i = 0;
3456 }
3457
3458 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3459
3460 /*
3461 * Force memory writes to complete before letting h/w
3462 * know there are new descriptors to fetch. (Only
3463 * applicable for weak-ordered memory model archs,
3464 * such as IA-64).
3465 */
3466 wmb();
3467
3468 tx_ring->next_to_use = i;
3469 writel(i, adapter->hw.hw_addr + tx_ring->tail);
3470 /*
3471 * we need this if more than one processor can write to our tail
3472 * at a time, it synchronizes IO on IA64/Altix systems
3473 */
3474 mmiowb();
3475 }
3476
3477 #define MINIMUM_DHCP_PACKET_SIZE 282
3478 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3479 struct sk_buff *skb)
3480 {
3481 struct e1000_hw *hw = &adapter->hw;
3482 u16 length, offset;
3483
3484 if (vlan_tx_tag_present(skb)) {
3485 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3486 && (adapter->hw.mng_cookie.status &
3487 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3488 return 0;
3489 }
3490
3491 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3492 return 0;
3493
3494 if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3495 return 0;
3496
3497 {
3498 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3499 struct udphdr *udp;
3500
3501 if (ip->protocol != IPPROTO_UDP)
3502 return 0;
3503
3504 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3505 if (ntohs(udp->dest) != 67)
3506 return 0;
3507
3508 offset = (u8 *)udp + 8 - skb->data;
3509 length = skb->len - offset;
3510 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
3511 }
3512
3513 return 0;
3514 }
3515
3516 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3517 {
3518 struct e1000_adapter *adapter = netdev_priv(netdev);
3519
3520 netif_stop_queue(netdev);
3521 /*
3522 * Herbert's original patch had:
3523 * smp_mb__after_netif_stop_queue();
3524 * but since that doesn't exist yet, just open code it.
3525 */
3526 smp_mb();
3527
3528 /*
3529 * We need to check again in a case another CPU has just
3530 * made room available.
3531 */
3532 if (e1000_desc_unused(adapter->tx_ring) < size)
3533 return -EBUSY;
3534
3535 /* A reprieve! */
3536 netif_start_queue(netdev);
3537 ++adapter->restart_queue;
3538 return 0;
3539 }
3540
3541 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
3542 {
3543 struct e1000_adapter *adapter = netdev_priv(netdev);
3544
3545 if (e1000_desc_unused(adapter->tx_ring) >= size)
3546 return 0;
3547 return __e1000_maybe_stop_tx(netdev, size);
3548 }
3549
3550 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3551 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3552 {
3553 struct e1000_adapter *adapter = netdev_priv(netdev);
3554 struct e1000_ring *tx_ring = adapter->tx_ring;
3555 unsigned int first;
3556 unsigned int max_per_txd = E1000_MAX_PER_TXD;
3557 unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3558 unsigned int tx_flags = 0;
3559 unsigned int len = skb->len - skb->data_len;
3560 unsigned long irq_flags;
3561 unsigned int nr_frags;
3562 unsigned int mss;
3563 int count = 0;
3564 int tso;
3565 unsigned int f;
3566
3567 if (test_bit(__E1000_DOWN, &adapter->state)) {
3568 dev_kfree_skb_any(skb);
3569 return NETDEV_TX_OK;
3570 }
3571
3572 if (skb->len <= 0) {
3573 dev_kfree_skb_any(skb);
3574 return NETDEV_TX_OK;
3575 }
3576
3577 mss = skb_shinfo(skb)->gso_size;
3578 /*
3579 * The controller does a simple calculation to
3580 * make sure there is enough room in the FIFO before
3581 * initiating the DMA for each buffer. The calc is:
3582 * 4 = ceil(buffer len/mss). To make sure we don't
3583 * overrun the FIFO, adjust the max buffer len if mss
3584 * drops.
3585 */
3586 if (mss) {
3587 u8 hdr_len;
3588 max_per_txd = min(mss << 2, max_per_txd);
3589 max_txd_pwr = fls(max_per_txd) - 1;
3590
3591 /*
3592 * TSO Workaround for 82571/2/3 Controllers -- if skb->data
3593 * points to just header, pull a few bytes of payload from
3594 * frags into skb->data
3595 */
3596 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3597 /*
3598 * we do this workaround for ES2LAN, but it is un-necessary,
3599 * avoiding it could save a lot of cycles
3600 */
3601 if (skb->data_len && (hdr_len == len)) {
3602 unsigned int pull_size;
3603
3604 pull_size = min((unsigned int)4, skb->data_len);
3605 if (!__pskb_pull_tail(skb, pull_size)) {
3606 ndev_err(netdev,
3607 "__pskb_pull_tail failed.\n");
3608 dev_kfree_skb_any(skb);
3609 return NETDEV_TX_OK;
3610 }
3611 len = skb->len - skb->data_len;
3612 }
3613 }
3614
3615 /* reserve a descriptor for the offload context */
3616 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3617 count++;
3618 count++;
3619
3620 count += TXD_USE_COUNT(len, max_txd_pwr);
3621
3622 nr_frags = skb_shinfo(skb)->nr_frags;
3623 for (f = 0; f < nr_frags; f++)
3624 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3625 max_txd_pwr);
3626
3627 if (adapter->hw.mac.tx_pkt_filtering)
3628 e1000_transfer_dhcp_info(adapter, skb);
3629
3630 if (!spin_trylock_irqsave(&adapter->tx_queue_lock, irq_flags))
3631 /* Collision - tell upper layer to requeue */
3632 return NETDEV_TX_LOCKED;
3633
3634 /*
3635 * need: count + 2 desc gap to keep tail from touching
3636 * head, otherwise try next time
3637 */
3638 if (e1000_maybe_stop_tx(netdev, count + 2)) {
3639 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3640 return NETDEV_TX_BUSY;
3641 }
3642
3643 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3644 tx_flags |= E1000_TX_FLAGS_VLAN;
3645 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3646 }
3647
3648 first = tx_ring->next_to_use;
3649
3650 tso = e1000_tso(adapter, skb);
3651 if (tso < 0) {
3652 dev_kfree_skb_any(skb);
3653 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3654 return NETDEV_TX_OK;
3655 }
3656
3657 if (tso)
3658 tx_flags |= E1000_TX_FLAGS_TSO;
3659 else if (e1000_tx_csum(adapter, skb))
3660 tx_flags |= E1000_TX_FLAGS_CSUM;
3661
3662 /*
3663 * Old method was to assume IPv4 packet by default if TSO was enabled.
3664 * 82571 hardware supports TSO capabilities for IPv6 as well...
3665 * no longer assume, we must.
3666 */
3667 if (skb->protocol == htons(ETH_P_IP))
3668 tx_flags |= E1000_TX_FLAGS_IPV4;
3669
3670 count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
3671 if (count < 0) {
3672 /* handle pci_map_single() error in e1000_tx_map */
3673 dev_kfree_skb_any(skb);
3674 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3675 return NETDEV_TX_OK;
3676 }
3677
3678 e1000_tx_queue(adapter, tx_flags, count);
3679
3680 netdev->trans_start = jiffies;
3681
3682 /* Make sure there is space in the ring for the next send. */
3683 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
3684
3685 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3686 return NETDEV_TX_OK;
3687 }
3688
3689 /**
3690 * e1000_tx_timeout - Respond to a Tx Hang
3691 * @netdev: network interface device structure
3692 **/
3693 static void e1000_tx_timeout(struct net_device *netdev)
3694 {
3695 struct e1000_adapter *adapter = netdev_priv(netdev);
3696
3697 /* Do the reset outside of interrupt context */
3698 adapter->tx_timeout_count++;
3699 schedule_work(&adapter->reset_task);
3700 }
3701
3702 static void e1000_reset_task(struct work_struct *work)
3703 {
3704 struct e1000_adapter *adapter;
3705 adapter = container_of(work, struct e1000_adapter, reset_task);
3706
3707 e1000e_reinit_locked(adapter);
3708 }
3709
3710 /**
3711 * e1000_get_stats - Get System Network Statistics
3712 * @netdev: network interface device structure
3713 *
3714 * Returns the address of the device statistics structure.
3715 * The statistics are actually updated from the timer callback.
3716 **/
3717 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3718 {
3719 struct e1000_adapter *adapter = netdev_priv(netdev);
3720
3721 /* only return the current stats */
3722 return &adapter->net_stats;
3723 }
3724
3725 /**
3726 * e1000_change_mtu - Change the Maximum Transfer Unit
3727 * @netdev: network interface device structure
3728 * @new_mtu: new value for maximum frame size
3729 *
3730 * Returns 0 on success, negative on failure
3731 **/
3732 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3733 {
3734 struct e1000_adapter *adapter = netdev_priv(netdev);
3735 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3736
3737 if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
3738 (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3739 ndev_err(netdev, "Invalid MTU setting\n");
3740 return -EINVAL;
3741 }
3742
3743 /* Jumbo frame size limits */
3744 if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
3745 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
3746 ndev_err(netdev, "Jumbo Frames not supported.\n");
3747 return -EINVAL;
3748 }
3749 if (adapter->hw.phy.type == e1000_phy_ife) {
3750 ndev_err(netdev, "Jumbo Frames not supported.\n");
3751 return -EINVAL;
3752 }
3753 }
3754
3755 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3756 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3757 ndev_err(netdev, "MTU > 9216 not supported.\n");
3758 return -EINVAL;
3759 }
3760
3761 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3762 msleep(1);
3763 /* e1000e_down has a dependency on max_frame_size */
3764 adapter->max_frame_size = max_frame;
3765 if (netif_running(netdev))
3766 e1000e_down(adapter);
3767
3768 /*
3769 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3770 * means we reserve 2 more, this pushes us to allocate from the next
3771 * larger slab size.
3772 * i.e. RXBUFFER_2048 --> size-4096 slab
3773 * However with the new *_jumbo_rx* routines, jumbo receives will use
3774 * fragmented skbs
3775 */
3776
3777 if (max_frame <= 256)
3778 adapter->rx_buffer_len = 256;
3779 else if (max_frame <= 512)
3780 adapter->rx_buffer_len = 512;
3781 else if (max_frame <= 1024)
3782 adapter->rx_buffer_len = 1024;
3783 else if (max_frame <= 2048)
3784 adapter->rx_buffer_len = 2048;
3785 else
3786 adapter->rx_buffer_len = 4096;
3787
3788 /* adjust allocation if LPE protects us, and we aren't using SBP */
3789 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3790 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
3791 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
3792 + ETH_FCS_LEN;
3793
3794 ndev_info(netdev, "changing MTU from %d to %d\n",
3795 netdev->mtu, new_mtu);
3796 netdev->mtu = new_mtu;
3797
3798 if (netif_running(netdev))
3799 e1000e_up(adapter);
3800 else
3801 e1000e_reset(adapter);
3802
3803 clear_bit(__E1000_RESETTING, &adapter->state);
3804
3805 return 0;
3806 }
3807
3808 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
3809 int cmd)
3810 {
3811 struct e1000_adapter *adapter = netdev_priv(netdev);
3812 struct mii_ioctl_data *data = if_mii(ifr);
3813
3814 if (adapter->hw.phy.media_type != e1000_media_type_copper)
3815 return -EOPNOTSUPP;
3816
3817 switch (cmd) {
3818 case SIOCGMIIPHY:
3819 data->phy_id = adapter->hw.phy.addr;
3820 break;
3821 case SIOCGMIIREG:
3822 if (!capable(CAP_NET_ADMIN))
3823 return -EPERM;
3824 switch (data->reg_num & 0x1F) {
3825 case MII_BMCR:
3826 data->val_out = adapter->phy_regs.bmcr;
3827 break;
3828 case MII_BMSR:
3829 data->val_out = adapter->phy_regs.bmsr;
3830 break;
3831 case MII_PHYSID1:
3832 data->val_out = (adapter->hw.phy.id >> 16);
3833 break;
3834 case MII_PHYSID2:
3835 data->val_out = (adapter->hw.phy.id & 0xFFFF);
3836 break;
3837 case MII_ADVERTISE:
3838 data->val_out = adapter->phy_regs.advertise;
3839 break;
3840 case MII_LPA:
3841 data->val_out = adapter->phy_regs.lpa;
3842 break;
3843 case MII_EXPANSION:
3844 data->val_out = adapter->phy_regs.expansion;
3845 break;
3846 case MII_CTRL1000:
3847 data->val_out = adapter->phy_regs.ctrl1000;
3848 break;
3849 case MII_STAT1000:
3850 data->val_out = adapter->phy_regs.stat1000;
3851 break;
3852 case MII_ESTATUS:
3853 data->val_out = adapter->phy_regs.estatus;
3854 break;
3855 default:
3856 return -EIO;
3857 }
3858 break;
3859 case SIOCSMIIREG:
3860 default:
3861 return -EOPNOTSUPP;
3862 }
3863 return 0;
3864 }
3865
3866 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
3867 {
3868 switch (cmd) {
3869 case SIOCGMIIPHY:
3870 case SIOCGMIIREG:
3871 case SIOCSMIIREG:
3872 return e1000_mii_ioctl(netdev, ifr, cmd);
3873 default:
3874 return -EOPNOTSUPP;
3875 }
3876 }
3877
3878 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
3879 {
3880 struct net_device *netdev = pci_get_drvdata(pdev);
3881 struct e1000_adapter *adapter = netdev_priv(netdev);
3882 struct e1000_hw *hw = &adapter->hw;
3883 u32 ctrl, ctrl_ext, rctl, status;
3884 u32 wufc = adapter->wol;
3885 int retval = 0;
3886
3887 netif_device_detach(netdev);
3888
3889 if (netif_running(netdev)) {
3890 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3891 e1000e_down(adapter);
3892 e1000_free_irq(adapter);
3893 }
3894
3895 retval = pci_save_state(pdev);
3896 if (retval)
3897 return retval;
3898
3899 status = er32(STATUS);
3900 if (status & E1000_STATUS_LU)
3901 wufc &= ~E1000_WUFC_LNKC;
3902
3903 if (wufc) {
3904 e1000_setup_rctl(adapter);
3905 e1000_set_multi(netdev);
3906
3907 /* turn on all-multi mode if wake on multicast is enabled */
3908 if (wufc & E1000_WUFC_MC) {
3909 rctl = er32(RCTL);
3910 rctl |= E1000_RCTL_MPE;
3911 ew32(RCTL, rctl);
3912 }
3913
3914 ctrl = er32(CTRL);
3915 /* advertise wake from D3Cold */
3916 #define E1000_CTRL_ADVD3WUC 0x00100000
3917 /* phy power management enable */
3918 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
3919 ctrl |= E1000_CTRL_ADVD3WUC |
3920 E1000_CTRL_EN_PHY_PWR_MGMT;
3921 ew32(CTRL, ctrl);
3922
3923 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
3924 adapter->hw.phy.media_type ==
3925 e1000_media_type_internal_serdes) {
3926 /* keep the laser running in D3 */
3927 ctrl_ext = er32(CTRL_EXT);
3928 ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
3929 ew32(CTRL_EXT, ctrl_ext);
3930 }
3931
3932 if (adapter->flags & FLAG_IS_ICH)
3933 e1000e_disable_gig_wol_ich8lan(&adapter->hw);
3934
3935 /* Allow time for pending master requests to run */
3936 e1000e_disable_pcie_master(&adapter->hw);
3937
3938 ew32(WUC, E1000_WUC_PME_EN);
3939 ew32(WUFC, wufc);
3940 pci_enable_wake(pdev, PCI_D3hot, 1);
3941 pci_enable_wake(pdev, PCI_D3cold, 1);
3942 } else {
3943 ew32(WUC, 0);
3944 ew32(WUFC, 0);
3945 pci_enable_wake(pdev, PCI_D3hot, 0);
3946 pci_enable_wake(pdev, PCI_D3cold, 0);
3947 }
3948
3949 /* make sure adapter isn't asleep if manageability is enabled */
3950 if (adapter->flags & FLAG_MNG_PT_ENABLED) {
3951 pci_enable_wake(pdev, PCI_D3hot, 1);
3952 pci_enable_wake(pdev, PCI_D3cold, 1);
3953 }
3954
3955 if (adapter->hw.phy.type == e1000_phy_igp_3)
3956 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
3957
3958 /*
3959 * Release control of h/w to f/w. If f/w is AMT enabled, this
3960 * would have already happened in close and is redundant.
3961 */
3962 e1000_release_hw_control(adapter);
3963
3964 pci_disable_device(pdev);
3965
3966 pci_set_power_state(pdev, pci_choose_state(pdev, state));
3967
3968 return 0;
3969 }
3970
3971 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
3972 {
3973 int pos;
3974 u16 val;
3975
3976 /*
3977 * 82573 workaround - disable L1 ASPM on mobile chipsets
3978 *
3979 * L1 ASPM on various mobile (ich7) chipsets do not behave properly
3980 * resulting in lost data or garbage information on the pci-e link
3981 * level. This could result in (false) bad EEPROM checksum errors,
3982 * long ping times (up to 2s) or even a system freeze/hang.
3983 *
3984 * Unfortunately this feature saves about 1W power consumption when
3985 * active.
3986 */
3987 pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
3988 pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
3989 if (val & 0x2) {
3990 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
3991 val &= ~0x2;
3992 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
3993 }
3994 }
3995
3996 #ifdef CONFIG_PM
3997 static int e1000_resume(struct pci_dev *pdev)
3998 {
3999 struct net_device *netdev = pci_get_drvdata(pdev);
4000 struct e1000_adapter *adapter = netdev_priv(netdev);
4001 struct e1000_hw *hw = &adapter->hw;
4002 u32 err;
4003
4004 pci_set_power_state(pdev, PCI_D0);
4005 pci_restore_state(pdev);
4006 e1000e_disable_l1aspm(pdev);
4007
4008 if (adapter->need_ioport)
4009 err = pci_enable_device(pdev);
4010 else
4011 err = pci_enable_device_mem(pdev);
4012 if (err) {
4013 dev_err(&pdev->dev,
4014 "Cannot enable PCI device from suspend\n");
4015 return err;
4016 }
4017
4018 pci_set_master(pdev);
4019
4020 pci_enable_wake(pdev, PCI_D3hot, 0);
4021 pci_enable_wake(pdev, PCI_D3cold, 0);
4022
4023 if (netif_running(netdev)) {
4024 err = e1000_request_irq(adapter);
4025 if (err)
4026 return err;
4027 }
4028
4029 e1000e_power_up_phy(adapter);
4030 e1000e_reset(adapter);
4031 ew32(WUS, ~0);
4032
4033 e1000_init_manageability(adapter);
4034
4035 if (netif_running(netdev))
4036 e1000e_up(adapter);
4037
4038 netif_device_attach(netdev);
4039
4040 /*
4041 * If the controller has AMT, do not set DRV_LOAD until the interface
4042 * is up. For all other cases, let the f/w know that the h/w is now
4043 * under the control of the driver.
4044 */
4045 if (!(adapter->flags & FLAG_HAS_AMT) || !e1000e_check_mng_mode(&adapter->hw))
4046 e1000_get_hw_control(adapter);
4047
4048 return 0;
4049 }
4050 #endif
4051
4052 static void e1000_shutdown(struct pci_dev *pdev)
4053 {
4054 e1000_suspend(pdev, PMSG_SUSPEND);
4055 }
4056
4057 #ifdef CONFIG_NET_POLL_CONTROLLER
4058 /*
4059 * Polling 'interrupt' - used by things like netconsole to send skbs
4060 * without having to re-enable interrupts. It's not called while
4061 * the interrupt routine is executing.
4062 */
4063 static void e1000_netpoll(struct net_device *netdev)
4064 {
4065 struct e1000_adapter *adapter = netdev_priv(netdev);
4066
4067 disable_irq(adapter->pdev->irq);
4068 e1000_intr(adapter->pdev->irq, netdev);
4069
4070 enable_irq(adapter->pdev->irq);
4071 }
4072 #endif
4073
4074 /**
4075 * e1000_io_error_detected - called when PCI error is detected
4076 * @pdev: Pointer to PCI device
4077 * @state: The current pci connection state
4078 *
4079 * This function is called after a PCI bus error affecting
4080 * this device has been detected.
4081 */
4082 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4083 pci_channel_state_t state)
4084 {
4085 struct net_device *netdev = pci_get_drvdata(pdev);
4086 struct e1000_adapter *adapter = netdev_priv(netdev);
4087
4088 netif_device_detach(netdev);
4089
4090 if (netif_running(netdev))
4091 e1000e_down(adapter);
4092 pci_disable_device(pdev);
4093
4094 /* Request a slot slot reset. */
4095 return PCI_ERS_RESULT_NEED_RESET;
4096 }
4097
4098 /**
4099 * e1000_io_slot_reset - called after the pci bus has been reset.
4100 * @pdev: Pointer to PCI device
4101 *
4102 * Restart the card from scratch, as if from a cold-boot. Implementation
4103 * resembles the first-half of the e1000_resume routine.
4104 */
4105 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4106 {
4107 struct net_device *netdev = pci_get_drvdata(pdev);
4108 struct e1000_adapter *adapter = netdev_priv(netdev);
4109 struct e1000_hw *hw = &adapter->hw;
4110 int err;
4111
4112 e1000e_disable_l1aspm(pdev);
4113 if (adapter->need_ioport)
4114 err = pci_enable_device(pdev);
4115 else
4116 err = pci_enable_device_mem(pdev);
4117 if (err) {
4118 dev_err(&pdev->dev,
4119 "Cannot re-enable PCI device after reset.\n");
4120 return PCI_ERS_RESULT_DISCONNECT;
4121 }
4122 pci_set_master(pdev);
4123 pci_restore_state(pdev);
4124
4125 pci_enable_wake(pdev, PCI_D3hot, 0);
4126 pci_enable_wake(pdev, PCI_D3cold, 0);
4127
4128 e1000e_reset(adapter);
4129 ew32(WUS, ~0);
4130
4131 return PCI_ERS_RESULT_RECOVERED;
4132 }
4133
4134 /**
4135 * e1000_io_resume - called when traffic can start flowing again.
4136 * @pdev: Pointer to PCI device
4137 *
4138 * This callback is called when the error recovery driver tells us that
4139 * its OK to resume normal operation. Implementation resembles the
4140 * second-half of the e1000_resume routine.
4141 */
4142 static void e1000_io_resume(struct pci_dev *pdev)
4143 {
4144 struct net_device *netdev = pci_get_drvdata(pdev);
4145 struct e1000_adapter *adapter = netdev_priv(netdev);
4146
4147 e1000_init_manageability(adapter);
4148
4149 if (netif_running(netdev)) {
4150 if (e1000e_up(adapter)) {
4151 dev_err(&pdev->dev,
4152 "can't bring device back up after reset\n");
4153 return;
4154 }
4155 }
4156
4157 netif_device_attach(netdev);
4158
4159 /*
4160 * If the controller has AMT, do not set DRV_LOAD until the interface
4161 * is up. For all other cases, let the f/w know that the h/w is now
4162 * under the control of the driver.
4163 */
4164 if (!(adapter->flags & FLAG_HAS_AMT) ||
4165 !e1000e_check_mng_mode(&adapter->hw))
4166 e1000_get_hw_control(adapter);
4167
4168 }
4169
4170 static void e1000_print_device_info(struct e1000_adapter *adapter)
4171 {
4172 struct e1000_hw *hw = &adapter->hw;
4173 struct net_device *netdev = adapter->netdev;
4174 u32 pba_num;
4175
4176 /* print bus type/speed/width info */
4177 ndev_info(netdev, "(PCI Express:2.5GB/s:%s) "
4178 "%02x:%02x:%02x:%02x:%02x:%02x\n",
4179 /* bus width */
4180 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4181 "Width x1"),
4182 /* MAC address */
4183 netdev->dev_addr[0], netdev->dev_addr[1],
4184 netdev->dev_addr[2], netdev->dev_addr[3],
4185 netdev->dev_addr[4], netdev->dev_addr[5]);
4186 ndev_info(netdev, "Intel(R) PRO/%s Network Connection\n",
4187 (hw->phy.type == e1000_phy_ife)
4188 ? "10/100" : "1000");
4189 e1000e_read_pba_num(hw, &pba_num);
4190 ndev_info(netdev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4191 hw->mac.type, hw->phy.type,
4192 (pba_num >> 8), (pba_num & 0xff));
4193 }
4194
4195 /**
4196 * e1000e_is_need_ioport - determine if an adapter needs ioport resources or not
4197 * @pdev: PCI device information struct
4198 *
4199 * Returns true if an adapters needs ioport resources
4200 **/
4201 static int e1000e_is_need_ioport(struct pci_dev *pdev)
4202 {
4203 switch (pdev->device) {
4204 /* Currently there are no adapters that need ioport resources */
4205 default:
4206 return false;
4207 }
4208 }
4209
4210 /**
4211 * e1000_probe - Device Initialization Routine
4212 * @pdev: PCI device information struct
4213 * @ent: entry in e1000_pci_tbl
4214 *
4215 * Returns 0 on success, negative on failure
4216 *
4217 * e1000_probe initializes an adapter identified by a pci_dev structure.
4218 * The OS initialization, configuring of the adapter private structure,
4219 * and a hardware reset occur.
4220 **/
4221 static int __devinit e1000_probe(struct pci_dev *pdev,
4222 const struct pci_device_id *ent)
4223 {
4224 struct net_device *netdev;
4225 struct e1000_adapter *adapter;
4226 struct e1000_hw *hw;
4227 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4228 resource_size_t mmio_start, mmio_len;
4229 resource_size_t flash_start, flash_len;
4230
4231 static int cards_found;
4232 int i, err, pci_using_dac;
4233 u16 eeprom_data = 0;
4234 u16 eeprom_apme_mask = E1000_EEPROM_APME;
4235 int bars, need_ioport;
4236
4237 e1000e_disable_l1aspm(pdev);
4238
4239 /* do not allocate ioport bars when not needed */
4240 need_ioport = e1000e_is_need_ioport(pdev);
4241 if (need_ioport) {
4242 bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
4243 err = pci_enable_device(pdev);
4244 } else {
4245 bars = pci_select_bars(pdev, IORESOURCE_MEM);
4246 err = pci_enable_device_mem(pdev);
4247 }
4248 if (err)
4249 return err;
4250
4251 pci_using_dac = 0;
4252 err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
4253 if (!err) {
4254 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
4255 if (!err)
4256 pci_using_dac = 1;
4257 } else {
4258 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
4259 if (err) {
4260 err = pci_set_consistent_dma_mask(pdev,
4261 DMA_32BIT_MASK);
4262 if (err) {
4263 dev_err(&pdev->dev, "No usable DMA "
4264 "configuration, aborting\n");
4265 goto err_dma;
4266 }
4267 }
4268 }
4269
4270 err = pci_request_selected_regions(pdev, bars, e1000e_driver_name);
4271 if (err)
4272 goto err_pci_reg;
4273
4274 pci_set_master(pdev);
4275 pci_save_state(pdev);
4276
4277 err = -ENOMEM;
4278 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4279 if (!netdev)
4280 goto err_alloc_etherdev;
4281
4282 SET_NETDEV_DEV(netdev, &pdev->dev);
4283
4284 pci_set_drvdata(pdev, netdev);
4285 adapter = netdev_priv(netdev);
4286 hw = &adapter->hw;
4287 adapter->netdev = netdev;
4288 adapter->pdev = pdev;
4289 adapter->ei = ei;
4290 adapter->pba = ei->pba;
4291 adapter->flags = ei->flags;
4292 adapter->hw.adapter = adapter;
4293 adapter->hw.mac.type = ei->mac;
4294 adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4295 adapter->bars = bars;
4296 adapter->need_ioport = need_ioport;
4297
4298 mmio_start = pci_resource_start(pdev, 0);
4299 mmio_len = pci_resource_len(pdev, 0);
4300
4301 err = -EIO;
4302 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4303 if (!adapter->hw.hw_addr)
4304 goto err_ioremap;
4305
4306 if ((adapter->flags & FLAG_HAS_FLASH) &&
4307 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
4308 flash_start = pci_resource_start(pdev, 1);
4309 flash_len = pci_resource_len(pdev, 1);
4310 adapter->hw.flash_address = ioremap(flash_start, flash_len);
4311 if (!adapter->hw.flash_address)
4312 goto err_flashmap;
4313 }
4314
4315 /* construct the net_device struct */
4316 netdev->open = &e1000_open;
4317 netdev->stop = &e1000_close;
4318 netdev->hard_start_xmit = &e1000_xmit_frame;
4319 netdev->get_stats = &e1000_get_stats;
4320 netdev->set_multicast_list = &e1000_set_multi;
4321 netdev->set_mac_address = &e1000_set_mac;
4322 netdev->change_mtu = &e1000_change_mtu;
4323 netdev->do_ioctl = &e1000_ioctl;
4324 e1000e_set_ethtool_ops(netdev);
4325 netdev->tx_timeout = &e1000_tx_timeout;
4326 netdev->watchdog_timeo = 5 * HZ;
4327 netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
4328 netdev->vlan_rx_register = e1000_vlan_rx_register;
4329 netdev->vlan_rx_add_vid = e1000_vlan_rx_add_vid;
4330 netdev->vlan_rx_kill_vid = e1000_vlan_rx_kill_vid;
4331 #ifdef CONFIG_NET_POLL_CONTROLLER
4332 netdev->poll_controller = e1000_netpoll;
4333 #endif
4334 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
4335
4336 netdev->mem_start = mmio_start;
4337 netdev->mem_end = mmio_start + mmio_len;
4338
4339 adapter->bd_number = cards_found++;
4340
4341 /* setup adapter struct */
4342 err = e1000_sw_init(adapter);
4343 if (err)
4344 goto err_sw_init;
4345
4346 err = -EIO;
4347
4348 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4349 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
4350 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4351
4352 err = ei->get_variants(adapter);
4353 if (err)
4354 goto err_hw_init;
4355
4356 hw->mac.ops.get_bus_info(&adapter->hw);
4357
4358 adapter->hw.phy.autoneg_wait_to_complete = 0;
4359
4360 /* Copper options */
4361 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
4362 adapter->hw.phy.mdix = AUTO_ALL_MODES;
4363 adapter->hw.phy.disable_polarity_correction = 0;
4364 adapter->hw.phy.ms_type = e1000_ms_hw_default;
4365 }
4366
4367 if (e1000_check_reset_block(&adapter->hw))
4368 ndev_info(netdev,
4369 "PHY reset is blocked due to SOL/IDER session.\n");
4370
4371 netdev->features = NETIF_F_SG |
4372 NETIF_F_HW_CSUM |
4373 NETIF_F_HW_VLAN_TX |
4374 NETIF_F_HW_VLAN_RX;
4375
4376 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
4377 netdev->features |= NETIF_F_HW_VLAN_FILTER;
4378
4379 netdev->features |= NETIF_F_TSO;
4380 netdev->features |= NETIF_F_TSO6;
4381
4382 netdev->vlan_features |= NETIF_F_TSO;
4383 netdev->vlan_features |= NETIF_F_TSO6;
4384 netdev->vlan_features |= NETIF_F_HW_CSUM;
4385 netdev->vlan_features |= NETIF_F_SG;
4386
4387 if (pci_using_dac)
4388 netdev->features |= NETIF_F_HIGHDMA;
4389
4390 /*
4391 * We should not be using LLTX anymore, but we are still Tx faster with
4392 * it.
4393 */
4394 netdev->features |= NETIF_F_LLTX;
4395
4396 if (e1000e_enable_mng_pass_thru(&adapter->hw))
4397 adapter->flags |= FLAG_MNG_PT_ENABLED;
4398
4399 /*
4400 * before reading the NVM, reset the controller to
4401 * put the device in a known good starting state
4402 */
4403 adapter->hw.mac.ops.reset_hw(&adapter->hw);
4404
4405 /*
4406 * systems with ASPM and others may see the checksum fail on the first
4407 * attempt. Let's give it a few tries
4408 */
4409 for (i = 0;; i++) {
4410 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
4411 break;
4412 if (i == 2) {
4413 ndev_err(netdev, "The NVM Checksum Is Not Valid\n");
4414 err = -EIO;
4415 goto err_eeprom;
4416 }
4417 }
4418
4419 /* copy the MAC address out of the NVM */
4420 if (e1000e_read_mac_addr(&adapter->hw))
4421 ndev_err(netdev, "NVM Read Error while reading MAC address\n");
4422
4423 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
4424 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
4425
4426 if (!is_valid_ether_addr(netdev->perm_addr)) {
4427 ndev_err(netdev, "Invalid MAC Address: "
4428 "%02x:%02x:%02x:%02x:%02x:%02x\n",
4429 netdev->perm_addr[0], netdev->perm_addr[1],
4430 netdev->perm_addr[2], netdev->perm_addr[3],
4431 netdev->perm_addr[4], netdev->perm_addr[5]);
4432 err = -EIO;
4433 goto err_eeprom;
4434 }
4435
4436 init_timer(&adapter->watchdog_timer);
4437 adapter->watchdog_timer.function = &e1000_watchdog;
4438 adapter->watchdog_timer.data = (unsigned long) adapter;
4439
4440 init_timer(&adapter->phy_info_timer);
4441 adapter->phy_info_timer.function = &e1000_update_phy_info;
4442 adapter->phy_info_timer.data = (unsigned long) adapter;
4443
4444 INIT_WORK(&adapter->reset_task, e1000_reset_task);
4445 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
4446
4447 e1000e_check_options(adapter);
4448
4449 /* Initialize link parameters. User can change them with ethtool */
4450 adapter->hw.mac.autoneg = 1;
4451 adapter->fc_autoneg = 1;
4452 adapter->hw.fc.original_type = e1000_fc_default;
4453 adapter->hw.fc.type = e1000_fc_default;
4454 adapter->hw.phy.autoneg_advertised = 0x2f;
4455
4456 /* ring size defaults */
4457 adapter->rx_ring->count = 256;
4458 adapter->tx_ring->count = 256;
4459
4460 /*
4461 * Initial Wake on LAN setting - If APM wake is enabled in
4462 * the EEPROM, enable the ACPI Magic Packet filter
4463 */
4464 if (adapter->flags & FLAG_APME_IN_WUC) {
4465 /* APME bit in EEPROM is mapped to WUC.APME */
4466 eeprom_data = er32(WUC);
4467 eeprom_apme_mask = E1000_WUC_APME;
4468 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
4469 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
4470 (adapter->hw.bus.func == 1))
4471 e1000_read_nvm(&adapter->hw,
4472 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
4473 else
4474 e1000_read_nvm(&adapter->hw,
4475 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
4476 }
4477
4478 /* fetch WoL from EEPROM */
4479 if (eeprom_data & eeprom_apme_mask)
4480 adapter->eeprom_wol |= E1000_WUFC_MAG;
4481
4482 /*
4483 * now that we have the eeprom settings, apply the special cases
4484 * where the eeprom may be wrong or the board simply won't support
4485 * wake on lan on a particular port
4486 */
4487 if (!(adapter->flags & FLAG_HAS_WOL))
4488 adapter->eeprom_wol = 0;
4489
4490 /* initialize the wol settings based on the eeprom settings */
4491 adapter->wol = adapter->eeprom_wol;
4492
4493 /* reset the hardware with the new settings */
4494 e1000e_reset(adapter);
4495
4496 /*
4497 * If the controller has AMT, do not set DRV_LOAD until the interface
4498 * is up. For all other cases, let the f/w know that the h/w is now
4499 * under the control of the driver.
4500 */
4501 if (!(adapter->flags & FLAG_HAS_AMT) ||
4502 !e1000e_check_mng_mode(&adapter->hw))
4503 e1000_get_hw_control(adapter);
4504
4505 /* tell the stack to leave us alone until e1000_open() is called */
4506 netif_carrier_off(netdev);
4507 netif_tx_stop_all_queues(netdev);
4508
4509 strcpy(netdev->name, "eth%d");
4510 err = register_netdev(netdev);
4511 if (err)
4512 goto err_register;
4513
4514 e1000_print_device_info(adapter);
4515
4516 return 0;
4517
4518 err_register:
4519 err_hw_init:
4520 e1000_release_hw_control(adapter);
4521 err_eeprom:
4522 if (!e1000_check_reset_block(&adapter->hw))
4523 e1000_phy_hw_reset(&adapter->hw);
4524
4525 if (adapter->hw.flash_address)
4526 iounmap(adapter->hw.flash_address);
4527
4528 err_flashmap:
4529 kfree(adapter->tx_ring);
4530 kfree(adapter->rx_ring);
4531 err_sw_init:
4532 iounmap(adapter->hw.hw_addr);
4533 err_ioremap:
4534 free_netdev(netdev);
4535 err_alloc_etherdev:
4536 pci_release_selected_regions(pdev, bars);
4537 err_pci_reg:
4538 err_dma:
4539 pci_disable_device(pdev);
4540 return err;
4541 }
4542
4543 /**
4544 * e1000_remove - Device Removal Routine
4545 * @pdev: PCI device information struct
4546 *
4547 * e1000_remove is called by the PCI subsystem to alert the driver
4548 * that it should release a PCI device. The could be caused by a
4549 * Hot-Plug event, or because the driver is going to be removed from
4550 * memory.
4551 **/
4552 static void __devexit e1000_remove(struct pci_dev *pdev)
4553 {
4554 struct net_device *netdev = pci_get_drvdata(pdev);
4555 struct e1000_adapter *adapter = netdev_priv(netdev);
4556
4557 /*
4558 * flush_scheduled work may reschedule our watchdog task, so
4559 * explicitly disable watchdog tasks from being rescheduled
4560 */
4561 set_bit(__E1000_DOWN, &adapter->state);
4562 del_timer_sync(&adapter->watchdog_timer);
4563 del_timer_sync(&adapter->phy_info_timer);
4564
4565 flush_scheduled_work();
4566
4567 /*
4568 * Release control of h/w to f/w. If f/w is AMT enabled, this
4569 * would have already happened in close and is redundant.
4570 */
4571 e1000_release_hw_control(adapter);
4572
4573 unregister_netdev(netdev);
4574
4575 if (!e1000_check_reset_block(&adapter->hw))
4576 e1000_phy_hw_reset(&adapter->hw);
4577
4578 kfree(adapter->tx_ring);
4579 kfree(adapter->rx_ring);
4580
4581 iounmap(adapter->hw.hw_addr);
4582 if (adapter->hw.flash_address)
4583 iounmap(adapter->hw.flash_address);
4584 pci_release_selected_regions(pdev, adapter->bars);
4585
4586 free_netdev(netdev);
4587
4588 pci_disable_device(pdev);
4589 }
4590
4591 /* PCI Error Recovery (ERS) */
4592 static struct pci_error_handlers e1000_err_handler = {
4593 .error_detected = e1000_io_error_detected,
4594 .slot_reset = e1000_io_slot_reset,
4595 .resume = e1000_io_resume,
4596 };
4597
4598 static struct pci_device_id e1000_pci_tbl[] = {
4599 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
4600 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
4601 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
4602 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
4603 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
4604 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
4605 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
4606 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
4607 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
4608
4609 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
4610 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
4611 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
4612 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
4613
4614 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
4615 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
4616 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
4617
4618 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
4619 board_80003es2lan },
4620 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
4621 board_80003es2lan },
4622 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
4623 board_80003es2lan },
4624 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
4625 board_80003es2lan },
4626
4627 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
4628 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
4629 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
4630 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
4631 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
4632 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
4633 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
4634
4635 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
4636 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
4637 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
4638 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
4639 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
4640 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
4641 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
4642 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
4643
4644 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
4645 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
4646 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
4647
4648 { } /* terminate list */
4649 };
4650 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
4651
4652 /* PCI Device API Driver */
4653 static struct pci_driver e1000_driver = {
4654 .name = e1000e_driver_name,
4655 .id_table = e1000_pci_tbl,
4656 .probe = e1000_probe,
4657 .remove = __devexit_p(e1000_remove),
4658 #ifdef CONFIG_PM
4659 /* Power Management Hooks */
4660 .suspend = e1000_suspend,
4661 .resume = e1000_resume,
4662 #endif
4663 .shutdown = e1000_shutdown,
4664 .err_handler = &e1000_err_handler
4665 };
4666
4667 /**
4668 * e1000_init_module - Driver Registration Routine
4669 *
4670 * e1000_init_module is the first routine called when the driver is
4671 * loaded. All it does is register with the PCI subsystem.
4672 **/
4673 static int __init e1000_init_module(void)
4674 {
4675 int ret;
4676 printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
4677 e1000e_driver_name, e1000e_driver_version);
4678 printk(KERN_INFO "%s: Copyright (c) 1999-2008 Intel Corporation.\n",
4679 e1000e_driver_name);
4680 ret = pci_register_driver(&e1000_driver);
4681 pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name,
4682 PM_QOS_DEFAULT_VALUE);
4683
4684 return ret;
4685 }
4686 module_init(e1000_init_module);
4687
4688 /**
4689 * e1000_exit_module - Driver Exit Cleanup Routine
4690 *
4691 * e1000_exit_module is called just before the driver is removed
4692 * from memory.
4693 **/
4694 static void __exit e1000_exit_module(void)
4695 {
4696 pci_unregister_driver(&e1000_driver);
4697 pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name);
4698 }
4699 module_exit(e1000_exit_module);
4700
4701
4702 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
4703 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
4704 MODULE_LICENSE("GPL");
4705 MODULE_VERSION(DRV_VERSION);
4706
4707 /* e1000_main.c */
Linux kernel - Deliverables
This page took 0.271003 seconds and 5 git commands to generate.