e1000e: correct flow control thresholds
[linux-2.6/mini2440.git] / drivers / net / e1000e / netdev.c
blobb7a46c513783191a86ab27c625dcc929b350a2a8
1 /*******************************************************************************
3 Intel PRO/1000 Linux driver
4 Copyright(c) 1999 - 2008 Intel Corporation.
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.
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.
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.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
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
27 *******************************************************************************/
29 #include <linux/module.h>
30 #include <linux/types.h>
31 #include <linux/init.h>
32 #include <linux/pci.h>
33 #include <linux/vmalloc.h>
34 #include <linux/pagemap.h>
35 #include <linux/delay.h>
36 #include <linux/netdevice.h>
37 #include <linux/tcp.h>
38 #include <linux/ipv6.h>
39 #include <net/checksum.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/mii.h>
42 #include <linux/ethtool.h>
43 #include <linux/if_vlan.h>
44 #include <linux/cpu.h>
45 #include <linux/smp.h>
46 #include <linux/pm_qos_params.h>
47 #include <linux/aer.h>
49 #include "e1000.h"
51 #define DRV_VERSION "0.3.3.4-k4"
52 char e1000e_driver_name[] = "e1000e";
53 const char e1000e_driver_version[] = DRV_VERSION;
55 static const struct e1000_info *e1000_info_tbl[] = {
56 [board_82571] = &e1000_82571_info,
57 [board_82572] = &e1000_82572_info,
58 [board_82573] = &e1000_82573_info,
59 [board_82574] = &e1000_82574_info,
60 [board_82583] = &e1000_82583_info,
61 [board_80003es2lan] = &e1000_es2_info,
62 [board_ich8lan] = &e1000_ich8_info,
63 [board_ich9lan] = &e1000_ich9_info,
64 [board_ich10lan] = &e1000_ich10_info,
67 #ifdef DEBUG
68 /**
69 * e1000_get_hw_dev_name - return device name string
70 * used by hardware layer to print debugging information
71 **/
72 char *e1000e_get_hw_dev_name(struct e1000_hw *hw)
74 return hw->adapter->netdev->name;
76 #endif
78 /**
79 * e1000_desc_unused - calculate if we have unused descriptors
80 **/
81 static int e1000_desc_unused(struct e1000_ring *ring)
83 if (ring->next_to_clean > ring->next_to_use)
84 return ring->next_to_clean - ring->next_to_use - 1;
86 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
89 /**
90 * e1000_receive_skb - helper function to handle Rx indications
91 * @adapter: board private structure
92 * @status: descriptor status field as written by hardware
93 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
94 * @skb: pointer to sk_buff to be indicated to stack
95 **/
96 static void e1000_receive_skb(struct e1000_adapter *adapter,
97 struct net_device *netdev,
98 struct sk_buff *skb,
99 u8 status, __le16 vlan)
101 skb->protocol = eth_type_trans(skb, netdev);
103 if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
104 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
105 le16_to_cpu(vlan), skb);
106 else
107 napi_gro_receive(&adapter->napi, skb);
111 * e1000_rx_checksum - Receive Checksum Offload for 82543
112 * @adapter: board private structure
113 * @status_err: receive descriptor status and error fields
114 * @csum: receive descriptor csum field
115 * @sk_buff: socket buffer with received data
117 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
118 u32 csum, struct sk_buff *skb)
120 u16 status = (u16)status_err;
121 u8 errors = (u8)(status_err >> 24);
122 skb->ip_summed = CHECKSUM_NONE;
124 /* Ignore Checksum bit is set */
125 if (status & E1000_RXD_STAT_IXSM)
126 return;
127 /* TCP/UDP checksum error bit is set */
128 if (errors & E1000_RXD_ERR_TCPE) {
129 /* let the stack verify checksum errors */
130 adapter->hw_csum_err++;
131 return;
134 /* TCP/UDP Checksum has not been calculated */
135 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
136 return;
138 /* It must be a TCP or UDP packet with a valid checksum */
139 if (status & E1000_RXD_STAT_TCPCS) {
140 /* TCP checksum is good */
141 skb->ip_summed = CHECKSUM_UNNECESSARY;
142 } else {
144 * IP fragment with UDP payload
145 * Hardware complements the payload checksum, so we undo it
146 * and then put the value in host order for further stack use.
148 __sum16 sum = (__force __sum16)htons(csum);
149 skb->csum = csum_unfold(~sum);
150 skb->ip_summed = CHECKSUM_COMPLETE;
152 adapter->hw_csum_good++;
156 * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
157 * @adapter: address of board private structure
159 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
160 int cleaned_count)
162 struct net_device *netdev = adapter->netdev;
163 struct pci_dev *pdev = adapter->pdev;
164 struct e1000_ring *rx_ring = adapter->rx_ring;
165 struct e1000_rx_desc *rx_desc;
166 struct e1000_buffer *buffer_info;
167 struct sk_buff *skb;
168 unsigned int i;
169 unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
171 i = rx_ring->next_to_use;
172 buffer_info = &rx_ring->buffer_info[i];
174 while (cleaned_count--) {
175 skb = buffer_info->skb;
176 if (skb) {
177 skb_trim(skb, 0);
178 goto map_skb;
181 skb = netdev_alloc_skb(netdev, bufsz);
182 if (!skb) {
183 /* Better luck next round */
184 adapter->alloc_rx_buff_failed++;
185 break;
189 * Make buffer alignment 2 beyond a 16 byte boundary
190 * this will result in a 16 byte aligned IP header after
191 * the 14 byte MAC header is removed
193 skb_reserve(skb, NET_IP_ALIGN);
195 buffer_info->skb = skb;
196 map_skb:
197 buffer_info->dma = pci_map_single(pdev, skb->data,
198 adapter->rx_buffer_len,
199 PCI_DMA_FROMDEVICE);
200 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
201 dev_err(&pdev->dev, "RX DMA map failed\n");
202 adapter->rx_dma_failed++;
203 break;
206 rx_desc = E1000_RX_DESC(*rx_ring, i);
207 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
209 i++;
210 if (i == rx_ring->count)
211 i = 0;
212 buffer_info = &rx_ring->buffer_info[i];
215 if (rx_ring->next_to_use != i) {
216 rx_ring->next_to_use = i;
217 if (i-- == 0)
218 i = (rx_ring->count - 1);
221 * Force memory writes to complete before letting h/w
222 * know there are new descriptors to fetch. (Only
223 * applicable for weak-ordered memory model archs,
224 * such as IA-64).
226 wmb();
227 writel(i, adapter->hw.hw_addr + rx_ring->tail);
232 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
233 * @adapter: address of board private structure
235 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
236 int cleaned_count)
238 struct net_device *netdev = adapter->netdev;
239 struct pci_dev *pdev = adapter->pdev;
240 union e1000_rx_desc_packet_split *rx_desc;
241 struct e1000_ring *rx_ring = adapter->rx_ring;
242 struct e1000_buffer *buffer_info;
243 struct e1000_ps_page *ps_page;
244 struct sk_buff *skb;
245 unsigned int i, j;
247 i = rx_ring->next_to_use;
248 buffer_info = &rx_ring->buffer_info[i];
250 while (cleaned_count--) {
251 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
253 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
254 ps_page = &buffer_info->ps_pages[j];
255 if (j >= adapter->rx_ps_pages) {
256 /* all unused desc entries get hw null ptr */
257 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
258 continue;
260 if (!ps_page->page) {
261 ps_page->page = alloc_page(GFP_ATOMIC);
262 if (!ps_page->page) {
263 adapter->alloc_rx_buff_failed++;
264 goto no_buffers;
266 ps_page->dma = pci_map_page(pdev,
267 ps_page->page,
268 0, PAGE_SIZE,
269 PCI_DMA_FROMDEVICE);
270 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
271 dev_err(&adapter->pdev->dev,
272 "RX DMA page map failed\n");
273 adapter->rx_dma_failed++;
274 goto no_buffers;
278 * Refresh the desc even if buffer_addrs
279 * didn't change because each write-back
280 * erases this info.
282 rx_desc->read.buffer_addr[j+1] =
283 cpu_to_le64(ps_page->dma);
286 skb = netdev_alloc_skb(netdev,
287 adapter->rx_ps_bsize0 + NET_IP_ALIGN);
289 if (!skb) {
290 adapter->alloc_rx_buff_failed++;
291 break;
295 * Make buffer alignment 2 beyond a 16 byte boundary
296 * this will result in a 16 byte aligned IP header after
297 * the 14 byte MAC header is removed
299 skb_reserve(skb, NET_IP_ALIGN);
301 buffer_info->skb = skb;
302 buffer_info->dma = pci_map_single(pdev, skb->data,
303 adapter->rx_ps_bsize0,
304 PCI_DMA_FROMDEVICE);
305 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
306 dev_err(&pdev->dev, "RX DMA map failed\n");
307 adapter->rx_dma_failed++;
308 /* cleanup skb */
309 dev_kfree_skb_any(skb);
310 buffer_info->skb = NULL;
311 break;
314 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
316 i++;
317 if (i == rx_ring->count)
318 i = 0;
319 buffer_info = &rx_ring->buffer_info[i];
322 no_buffers:
323 if (rx_ring->next_to_use != i) {
324 rx_ring->next_to_use = i;
326 if (!(i--))
327 i = (rx_ring->count - 1);
330 * Force memory writes to complete before letting h/w
331 * know there are new descriptors to fetch. (Only
332 * applicable for weak-ordered memory model archs,
333 * such as IA-64).
335 wmb();
337 * Hardware increments by 16 bytes, but packet split
338 * descriptors are 32 bytes...so we increment tail
339 * twice as much.
341 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
346 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
347 * @adapter: address of board private structure
348 * @cleaned_count: number of buffers to allocate this pass
351 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
352 int cleaned_count)
354 struct net_device *netdev = adapter->netdev;
355 struct pci_dev *pdev = adapter->pdev;
356 struct e1000_rx_desc *rx_desc;
357 struct e1000_ring *rx_ring = adapter->rx_ring;
358 struct e1000_buffer *buffer_info;
359 struct sk_buff *skb;
360 unsigned int i;
361 unsigned int bufsz = 256 -
362 16 /* for skb_reserve */ -
363 NET_IP_ALIGN;
365 i = rx_ring->next_to_use;
366 buffer_info = &rx_ring->buffer_info[i];
368 while (cleaned_count--) {
369 skb = buffer_info->skb;
370 if (skb) {
371 skb_trim(skb, 0);
372 goto check_page;
375 skb = netdev_alloc_skb(netdev, bufsz);
376 if (unlikely(!skb)) {
377 /* Better luck next round */
378 adapter->alloc_rx_buff_failed++;
379 break;
382 /* Make buffer alignment 2 beyond a 16 byte boundary
383 * this will result in a 16 byte aligned IP header after
384 * the 14 byte MAC header is removed
386 skb_reserve(skb, NET_IP_ALIGN);
388 buffer_info->skb = skb;
389 check_page:
390 /* allocate a new page if necessary */
391 if (!buffer_info->page) {
392 buffer_info->page = alloc_page(GFP_ATOMIC);
393 if (unlikely(!buffer_info->page)) {
394 adapter->alloc_rx_buff_failed++;
395 break;
399 if (!buffer_info->dma)
400 buffer_info->dma = pci_map_page(pdev,
401 buffer_info->page, 0,
402 PAGE_SIZE,
403 PCI_DMA_FROMDEVICE);
405 rx_desc = E1000_RX_DESC(*rx_ring, i);
406 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
408 if (unlikely(++i == rx_ring->count))
409 i = 0;
410 buffer_info = &rx_ring->buffer_info[i];
413 if (likely(rx_ring->next_to_use != i)) {
414 rx_ring->next_to_use = i;
415 if (unlikely(i-- == 0))
416 i = (rx_ring->count - 1);
418 /* Force memory writes to complete before letting h/w
419 * know there are new descriptors to fetch. (Only
420 * applicable for weak-ordered memory model archs,
421 * such as IA-64). */
422 wmb();
423 writel(i, adapter->hw.hw_addr + rx_ring->tail);
428 * e1000_clean_rx_irq - Send received data up the network stack; legacy
429 * @adapter: board private structure
431 * the return value indicates whether actual cleaning was done, there
432 * is no guarantee that everything was cleaned
434 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
435 int *work_done, int work_to_do)
437 struct net_device *netdev = adapter->netdev;
438 struct pci_dev *pdev = adapter->pdev;
439 struct e1000_ring *rx_ring = adapter->rx_ring;
440 struct e1000_rx_desc *rx_desc, *next_rxd;
441 struct e1000_buffer *buffer_info, *next_buffer;
442 u32 length;
443 unsigned int i;
444 int cleaned_count = 0;
445 bool cleaned = 0;
446 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
448 i = rx_ring->next_to_clean;
449 rx_desc = E1000_RX_DESC(*rx_ring, i);
450 buffer_info = &rx_ring->buffer_info[i];
452 while (rx_desc->status & E1000_RXD_STAT_DD) {
453 struct sk_buff *skb;
454 u8 status;
456 if (*work_done >= work_to_do)
457 break;
458 (*work_done)++;
460 status = rx_desc->status;
461 skb = buffer_info->skb;
462 buffer_info->skb = NULL;
464 prefetch(skb->data - NET_IP_ALIGN);
466 i++;
467 if (i == rx_ring->count)
468 i = 0;
469 next_rxd = E1000_RX_DESC(*rx_ring, i);
470 prefetch(next_rxd);
472 next_buffer = &rx_ring->buffer_info[i];
474 cleaned = 1;
475 cleaned_count++;
476 pci_unmap_single(pdev,
477 buffer_info->dma,
478 adapter->rx_buffer_len,
479 PCI_DMA_FROMDEVICE);
480 buffer_info->dma = 0;
482 length = le16_to_cpu(rx_desc->length);
484 /* !EOP means multiple descriptors were used to store a single
485 * packet, also make sure the frame isn't just CRC only */
486 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
487 /* All receives must fit into a single buffer */
488 e_dbg("%s: Receive packet consumed multiple buffers\n",
489 netdev->name);
490 /* recycle */
491 buffer_info->skb = skb;
492 goto next_desc;
495 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
496 /* recycle */
497 buffer_info->skb = skb;
498 goto next_desc;
501 /* adjust length to remove Ethernet CRC */
502 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
503 length -= 4;
505 total_rx_bytes += length;
506 total_rx_packets++;
509 * code added for copybreak, this should improve
510 * performance for small packets with large amounts
511 * of reassembly being done in the stack
513 if (length < copybreak) {
514 struct sk_buff *new_skb =
515 netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
516 if (new_skb) {
517 skb_reserve(new_skb, NET_IP_ALIGN);
518 skb_copy_to_linear_data_offset(new_skb,
519 -NET_IP_ALIGN,
520 (skb->data -
521 NET_IP_ALIGN),
522 (length +
523 NET_IP_ALIGN));
524 /* save the skb in buffer_info as good */
525 buffer_info->skb = skb;
526 skb = new_skb;
528 /* else just continue with the old one */
530 /* end copybreak code */
531 skb_put(skb, length);
533 /* Receive Checksum Offload */
534 e1000_rx_checksum(adapter,
535 (u32)(status) |
536 ((u32)(rx_desc->errors) << 24),
537 le16_to_cpu(rx_desc->csum), skb);
539 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
541 next_desc:
542 rx_desc->status = 0;
544 /* return some buffers to hardware, one at a time is too slow */
545 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
546 adapter->alloc_rx_buf(adapter, cleaned_count);
547 cleaned_count = 0;
550 /* use prefetched values */
551 rx_desc = next_rxd;
552 buffer_info = next_buffer;
554 rx_ring->next_to_clean = i;
556 cleaned_count = e1000_desc_unused(rx_ring);
557 if (cleaned_count)
558 adapter->alloc_rx_buf(adapter, cleaned_count);
560 adapter->total_rx_bytes += total_rx_bytes;
561 adapter->total_rx_packets += total_rx_packets;
562 adapter->net_stats.rx_bytes += total_rx_bytes;
563 adapter->net_stats.rx_packets += total_rx_packets;
564 return cleaned;
567 static void e1000_put_txbuf(struct e1000_adapter *adapter,
568 struct e1000_buffer *buffer_info)
570 buffer_info->dma = 0;
571 if (buffer_info->skb) {
572 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
573 DMA_TO_DEVICE);
574 dev_kfree_skb_any(buffer_info->skb);
575 buffer_info->skb = NULL;
577 buffer_info->time_stamp = 0;
580 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
582 struct e1000_ring *tx_ring = adapter->tx_ring;
583 unsigned int i = tx_ring->next_to_clean;
584 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
585 struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
587 /* detected Tx unit hang */
588 e_err("Detected Tx Unit Hang:\n"
589 " TDH <%x>\n"
590 " TDT <%x>\n"
591 " next_to_use <%x>\n"
592 " next_to_clean <%x>\n"
593 "buffer_info[next_to_clean]:\n"
594 " time_stamp <%lx>\n"
595 " next_to_watch <%x>\n"
596 " jiffies <%lx>\n"
597 " next_to_watch.status <%x>\n",
598 readl(adapter->hw.hw_addr + tx_ring->head),
599 readl(adapter->hw.hw_addr + tx_ring->tail),
600 tx_ring->next_to_use,
601 tx_ring->next_to_clean,
602 tx_ring->buffer_info[eop].time_stamp,
603 eop,
604 jiffies,
605 eop_desc->upper.fields.status);
609 * e1000_clean_tx_irq - Reclaim resources after transmit completes
610 * @adapter: board private structure
612 * the return value indicates whether actual cleaning was done, there
613 * is no guarantee that everything was cleaned
615 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
617 struct net_device *netdev = adapter->netdev;
618 struct e1000_hw *hw = &adapter->hw;
619 struct e1000_ring *tx_ring = adapter->tx_ring;
620 struct e1000_tx_desc *tx_desc, *eop_desc;
621 struct e1000_buffer *buffer_info;
622 unsigned int i, eop;
623 unsigned int count = 0;
624 unsigned int total_tx_bytes = 0, total_tx_packets = 0;
626 i = tx_ring->next_to_clean;
627 eop = tx_ring->buffer_info[i].next_to_watch;
628 eop_desc = E1000_TX_DESC(*tx_ring, eop);
630 while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
631 (count < tx_ring->count)) {
632 bool cleaned = false;
633 for (; !cleaned; count++) {
634 tx_desc = E1000_TX_DESC(*tx_ring, i);
635 buffer_info = &tx_ring->buffer_info[i];
636 cleaned = (i == eop);
638 if (cleaned) {
639 struct sk_buff *skb = buffer_info->skb;
640 unsigned int segs, bytecount;
641 segs = skb_shinfo(skb)->gso_segs ?: 1;
642 /* multiply data chunks by size of headers */
643 bytecount = ((segs - 1) * skb_headlen(skb)) +
644 skb->len;
645 total_tx_packets += segs;
646 total_tx_bytes += bytecount;
649 e1000_put_txbuf(adapter, buffer_info);
650 tx_desc->upper.data = 0;
652 i++;
653 if (i == tx_ring->count)
654 i = 0;
657 eop = tx_ring->buffer_info[i].next_to_watch;
658 eop_desc = E1000_TX_DESC(*tx_ring, eop);
661 tx_ring->next_to_clean = i;
663 #define TX_WAKE_THRESHOLD 32
664 if (count && netif_carrier_ok(netdev) &&
665 e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
666 /* Make sure that anybody stopping the queue after this
667 * sees the new next_to_clean.
669 smp_mb();
671 if (netif_queue_stopped(netdev) &&
672 !(test_bit(__E1000_DOWN, &adapter->state))) {
673 netif_wake_queue(netdev);
674 ++adapter->restart_queue;
678 if (adapter->detect_tx_hung) {
679 /* Detect a transmit hang in hardware, this serializes the
680 * check with the clearing of time_stamp and movement of i */
681 adapter->detect_tx_hung = 0;
682 if (tx_ring->buffer_info[i].time_stamp &&
683 time_after(jiffies, tx_ring->buffer_info[i].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);
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 (count < tx_ring->count);
698 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
699 * @adapter: board private structure
701 * the return value indicates whether actual cleaning was done, there
702 * is no guarantee that everything was cleaned
704 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
705 int *work_done, int work_to_do)
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;
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];
725 while (staterr & E1000_RXD_STAT_DD) {
726 if (*work_done >= work_to_do)
727 break;
728 (*work_done)++;
729 skb = buffer_info->skb;
731 /* in the packet split case this is header only */
732 prefetch(skb->data - NET_IP_ALIGN);
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);
740 next_buffer = &rx_ring->buffer_info[i];
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;
749 if (!(staterr & E1000_RXD_STAT_EOP)) {
750 e_dbg("%s: Packet Split buffers didn't pick up the "
751 "full packet\n", netdev->name);
752 dev_kfree_skb_irq(skb);
753 goto next_desc;
756 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
757 dev_kfree_skb_irq(skb);
758 goto next_desc;
761 length = le16_to_cpu(rx_desc->wb.middle.length0);
763 if (!length) {
764 e_dbg("%s: Last part of the packet spanning multiple "
765 "descriptors\n", netdev->name);
766 dev_kfree_skb_irq(skb);
767 goto next_desc;
770 /* Good Receive */
771 skb_put(skb, length);
775 * this looks ugly, but it seems compiler issues make it
776 * more efficient than reusing j
778 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
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_*
785 if (l1 && (l1 <= copybreak) &&
786 ((length + l1) <= adapter->rx_ps_bsize0)) {
787 u8 *vaddr;
789 ps_page = &buffer_info->ps_pages[0];
792 * there is no documentation about how to call
793 * kmap_atomic, so we can't hold the mapping
794 * very long
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);
804 /* remove the CRC */
805 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
806 l1 -= 4;
808 skb_put(skb, l1);
809 goto copydone;
810 } /* if */
813 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
814 length = le16_to_cpu(rx_desc->wb.upper.length[j]);
815 if (!length)
816 break;
818 ps_page = &buffer_info->ps_pages[j];
819 pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
820 PCI_DMA_FROMDEVICE);
821 ps_page->dma = 0;
822 skb_fill_page_desc(skb, j, ps_page->page, 0, length);
823 ps_page->page = NULL;
824 skb->len += length;
825 skb->data_len += length;
826 skb->truesize += length;
829 /* strip the ethernet crc, problem is we're using pages now so
830 * this whole operation can get a little cpu intensive
832 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
833 pskb_trim(skb, skb->len - 4);
835 copydone:
836 total_rx_bytes += skb->len;
837 total_rx_packets++;
839 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
840 rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
842 if (rx_desc->wb.upper.header_status &
843 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
844 adapter->rx_hdr_split++;
846 e1000_receive_skb(adapter, netdev, skb,
847 staterr, rx_desc->wb.middle.vlan);
849 next_desc:
850 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
851 buffer_info->skb = NULL;
853 /* return some buffers to hardware, one at a time is too slow */
854 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
855 adapter->alloc_rx_buf(adapter, cleaned_count);
856 cleaned_count = 0;
859 /* use prefetched values */
860 rx_desc = next_rxd;
861 buffer_info = next_buffer;
863 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
865 rx_ring->next_to_clean = i;
867 cleaned_count = e1000_desc_unused(rx_ring);
868 if (cleaned_count)
869 adapter->alloc_rx_buf(adapter, cleaned_count);
871 adapter->total_rx_bytes += total_rx_bytes;
872 adapter->total_rx_packets += total_rx_packets;
873 adapter->net_stats.rx_bytes += total_rx_bytes;
874 adapter->net_stats.rx_packets += total_rx_packets;
875 return cleaned;
879 * e1000_consume_page - helper function
881 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
882 u16 length)
884 bi->page = NULL;
885 skb->len += length;
886 skb->data_len += length;
887 skb->truesize += length;
891 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
892 * @adapter: board private structure
894 * the return value indicates whether actual cleaning was done, there
895 * is no guarantee that everything was cleaned
898 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
899 int *work_done, int work_to_do)
901 struct net_device *netdev = adapter->netdev;
902 struct pci_dev *pdev = adapter->pdev;
903 struct e1000_ring *rx_ring = adapter->rx_ring;
904 struct e1000_rx_desc *rx_desc, *next_rxd;
905 struct e1000_buffer *buffer_info, *next_buffer;
906 u32 length;
907 unsigned int i;
908 int cleaned_count = 0;
909 bool cleaned = false;
910 unsigned int total_rx_bytes=0, total_rx_packets=0;
912 i = rx_ring->next_to_clean;
913 rx_desc = E1000_RX_DESC(*rx_ring, i);
914 buffer_info = &rx_ring->buffer_info[i];
916 while (rx_desc->status & E1000_RXD_STAT_DD) {
917 struct sk_buff *skb;
918 u8 status;
920 if (*work_done >= work_to_do)
921 break;
922 (*work_done)++;
924 status = rx_desc->status;
925 skb = buffer_info->skb;
926 buffer_info->skb = NULL;
928 ++i;
929 if (i == rx_ring->count)
930 i = 0;
931 next_rxd = E1000_RX_DESC(*rx_ring, i);
932 prefetch(next_rxd);
934 next_buffer = &rx_ring->buffer_info[i];
936 cleaned = true;
937 cleaned_count++;
938 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
939 PCI_DMA_FROMDEVICE);
940 buffer_info->dma = 0;
942 length = le16_to_cpu(rx_desc->length);
944 /* errors is only valid for DD + EOP descriptors */
945 if (unlikely((status & E1000_RXD_STAT_EOP) &&
946 (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
947 /* recycle both page and skb */
948 buffer_info->skb = skb;
949 /* an error means any chain goes out the window
950 * too */
951 if (rx_ring->rx_skb_top)
952 dev_kfree_skb(rx_ring->rx_skb_top);
953 rx_ring->rx_skb_top = NULL;
954 goto next_desc;
957 #define rxtop rx_ring->rx_skb_top
958 if (!(status & E1000_RXD_STAT_EOP)) {
959 /* this descriptor is only the beginning (or middle) */
960 if (!rxtop) {
961 /* this is the beginning of a chain */
962 rxtop = skb;
963 skb_fill_page_desc(rxtop, 0, buffer_info->page,
964 0, length);
965 } else {
966 /* this is the middle of a chain */
967 skb_fill_page_desc(rxtop,
968 skb_shinfo(rxtop)->nr_frags,
969 buffer_info->page, 0, length);
970 /* re-use the skb, only consumed the page */
971 buffer_info->skb = skb;
973 e1000_consume_page(buffer_info, rxtop, length);
974 goto next_desc;
975 } else {
976 if (rxtop) {
977 /* end of the chain */
978 skb_fill_page_desc(rxtop,
979 skb_shinfo(rxtop)->nr_frags,
980 buffer_info->page, 0, length);
981 /* re-use the current skb, we only consumed the
982 * page */
983 buffer_info->skb = skb;
984 skb = rxtop;
985 rxtop = NULL;
986 e1000_consume_page(buffer_info, skb, length);
987 } else {
988 /* no chain, got EOP, this buf is the packet
989 * copybreak to save the put_page/alloc_page */
990 if (length <= copybreak &&
991 skb_tailroom(skb) >= length) {
992 u8 *vaddr;
993 vaddr = kmap_atomic(buffer_info->page,
994 KM_SKB_DATA_SOFTIRQ);
995 memcpy(skb_tail_pointer(skb), vaddr,
996 length);
997 kunmap_atomic(vaddr,
998 KM_SKB_DATA_SOFTIRQ);
999 /* re-use the page, so don't erase
1000 * buffer_info->page */
1001 skb_put(skb, length);
1002 } else {
1003 skb_fill_page_desc(skb, 0,
1004 buffer_info->page, 0,
1005 length);
1006 e1000_consume_page(buffer_info, skb,
1007 length);
1012 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1013 e1000_rx_checksum(adapter,
1014 (u32)(status) |
1015 ((u32)(rx_desc->errors) << 24),
1016 le16_to_cpu(rx_desc->csum), skb);
1018 /* probably a little skewed due to removing CRC */
1019 total_rx_bytes += skb->len;
1020 total_rx_packets++;
1022 /* eth type trans needs skb->data to point to something */
1023 if (!pskb_may_pull(skb, ETH_HLEN)) {
1024 e_err("pskb_may_pull failed.\n");
1025 dev_kfree_skb(skb);
1026 goto next_desc;
1029 e1000_receive_skb(adapter, netdev, skb, status,
1030 rx_desc->special);
1032 next_desc:
1033 rx_desc->status = 0;
1035 /* return some buffers to hardware, one at a time is too slow */
1036 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1037 adapter->alloc_rx_buf(adapter, cleaned_count);
1038 cleaned_count = 0;
1041 /* use prefetched values */
1042 rx_desc = next_rxd;
1043 buffer_info = next_buffer;
1045 rx_ring->next_to_clean = i;
1047 cleaned_count = e1000_desc_unused(rx_ring);
1048 if (cleaned_count)
1049 adapter->alloc_rx_buf(adapter, cleaned_count);
1051 adapter->total_rx_bytes += total_rx_bytes;
1052 adapter->total_rx_packets += total_rx_packets;
1053 adapter->net_stats.rx_bytes += total_rx_bytes;
1054 adapter->net_stats.rx_packets += total_rx_packets;
1055 return cleaned;
1059 * e1000_clean_rx_ring - Free Rx Buffers per Queue
1060 * @adapter: board private structure
1062 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1064 struct e1000_ring *rx_ring = adapter->rx_ring;
1065 struct e1000_buffer *buffer_info;
1066 struct e1000_ps_page *ps_page;
1067 struct pci_dev *pdev = adapter->pdev;
1068 unsigned int i, j;
1070 /* Free all the Rx ring sk_buffs */
1071 for (i = 0; i < rx_ring->count; i++) {
1072 buffer_info = &rx_ring->buffer_info[i];
1073 if (buffer_info->dma) {
1074 if (adapter->clean_rx == e1000_clean_rx_irq)
1075 pci_unmap_single(pdev, buffer_info->dma,
1076 adapter->rx_buffer_len,
1077 PCI_DMA_FROMDEVICE);
1078 else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1079 pci_unmap_page(pdev, buffer_info->dma,
1080 PAGE_SIZE,
1081 PCI_DMA_FROMDEVICE);
1082 else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1083 pci_unmap_single(pdev, buffer_info->dma,
1084 adapter->rx_ps_bsize0,
1085 PCI_DMA_FROMDEVICE);
1086 buffer_info->dma = 0;
1089 if (buffer_info->page) {
1090 put_page(buffer_info->page);
1091 buffer_info->page = NULL;
1094 if (buffer_info->skb) {
1095 dev_kfree_skb(buffer_info->skb);
1096 buffer_info->skb = NULL;
1099 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1100 ps_page = &buffer_info->ps_pages[j];
1101 if (!ps_page->page)
1102 break;
1103 pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1104 PCI_DMA_FROMDEVICE);
1105 ps_page->dma = 0;
1106 put_page(ps_page->page);
1107 ps_page->page = NULL;
1111 /* there also may be some cached data from a chained receive */
1112 if (rx_ring->rx_skb_top) {
1113 dev_kfree_skb(rx_ring->rx_skb_top);
1114 rx_ring->rx_skb_top = NULL;
1117 /* Zero out the descriptor ring */
1118 memset(rx_ring->desc, 0, rx_ring->size);
1120 rx_ring->next_to_clean = 0;
1121 rx_ring->next_to_use = 0;
1123 writel(0, adapter->hw.hw_addr + rx_ring->head);
1124 writel(0, adapter->hw.hw_addr + rx_ring->tail);
1127 static void e1000e_downshift_workaround(struct work_struct *work)
1129 struct e1000_adapter *adapter = container_of(work,
1130 struct e1000_adapter, downshift_task);
1132 e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1136 * e1000_intr_msi - Interrupt Handler
1137 * @irq: interrupt number
1138 * @data: pointer to a network interface device structure
1140 static irqreturn_t e1000_intr_msi(int irq, void *data)
1142 struct net_device *netdev = data;
1143 struct e1000_adapter *adapter = netdev_priv(netdev);
1144 struct e1000_hw *hw = &adapter->hw;
1145 u32 icr = er32(ICR);
1148 * read ICR disables interrupts using IAM
1151 if (icr & E1000_ICR_LSC) {
1152 hw->mac.get_link_status = 1;
1154 * ICH8 workaround-- Call gig speed drop workaround on cable
1155 * disconnect (LSC) before accessing any PHY registers
1157 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1158 (!(er32(STATUS) & E1000_STATUS_LU)))
1159 schedule_work(&adapter->downshift_task);
1162 * 80003ES2LAN workaround-- For packet buffer work-around on
1163 * link down event; disable receives here in the ISR and reset
1164 * adapter in watchdog
1166 if (netif_carrier_ok(netdev) &&
1167 adapter->flags & FLAG_RX_NEEDS_RESTART) {
1168 /* disable receives */
1169 u32 rctl = er32(RCTL);
1170 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1171 adapter->flags |= FLAG_RX_RESTART_NOW;
1173 /* guard against interrupt when we're going down */
1174 if (!test_bit(__E1000_DOWN, &adapter->state))
1175 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1178 if (napi_schedule_prep(&adapter->napi)) {
1179 adapter->total_tx_bytes = 0;
1180 adapter->total_tx_packets = 0;
1181 adapter->total_rx_bytes = 0;
1182 adapter->total_rx_packets = 0;
1183 __napi_schedule(&adapter->napi);
1186 return IRQ_HANDLED;
1190 * e1000_intr - Interrupt Handler
1191 * @irq: interrupt number
1192 * @data: pointer to a network interface device structure
1194 static irqreturn_t e1000_intr(int irq, void *data)
1196 struct net_device *netdev = data;
1197 struct e1000_adapter *adapter = netdev_priv(netdev);
1198 struct e1000_hw *hw = &adapter->hw;
1199 u32 rctl, icr = er32(ICR);
1201 if (!icr)
1202 return IRQ_NONE; /* Not our interrupt */
1205 * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1206 * not set, then the adapter didn't send an interrupt
1208 if (!(icr & E1000_ICR_INT_ASSERTED))
1209 return IRQ_NONE;
1212 * Interrupt Auto-Mask...upon reading ICR,
1213 * interrupts are masked. No need for the
1214 * IMC write
1217 if (icr & E1000_ICR_LSC) {
1218 hw->mac.get_link_status = 1;
1220 * ICH8 workaround-- Call gig speed drop workaround on cable
1221 * disconnect (LSC) before accessing any PHY registers
1223 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1224 (!(er32(STATUS) & E1000_STATUS_LU)))
1225 schedule_work(&adapter->downshift_task);
1228 * 80003ES2LAN workaround--
1229 * For packet buffer work-around on link down event;
1230 * disable receives here in the ISR and
1231 * reset adapter in watchdog
1233 if (netif_carrier_ok(netdev) &&
1234 (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1235 /* disable receives */
1236 rctl = er32(RCTL);
1237 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1238 adapter->flags |= FLAG_RX_RESTART_NOW;
1240 /* guard against interrupt when we're going down */
1241 if (!test_bit(__E1000_DOWN, &adapter->state))
1242 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1245 if (napi_schedule_prep(&adapter->napi)) {
1246 adapter->total_tx_bytes = 0;
1247 adapter->total_tx_packets = 0;
1248 adapter->total_rx_bytes = 0;
1249 adapter->total_rx_packets = 0;
1250 __napi_schedule(&adapter->napi);
1253 return IRQ_HANDLED;
1256 static irqreturn_t e1000_msix_other(int irq, void *data)
1258 struct net_device *netdev = data;
1259 struct e1000_adapter *adapter = netdev_priv(netdev);
1260 struct e1000_hw *hw = &adapter->hw;
1261 u32 icr = er32(ICR);
1263 if (!(icr & E1000_ICR_INT_ASSERTED)) {
1264 if (!test_bit(__E1000_DOWN, &adapter->state))
1265 ew32(IMS, E1000_IMS_OTHER);
1266 return IRQ_NONE;
1269 if (icr & adapter->eiac_mask)
1270 ew32(ICS, (icr & adapter->eiac_mask));
1272 if (icr & E1000_ICR_OTHER) {
1273 if (!(icr & E1000_ICR_LSC))
1274 goto no_link_interrupt;
1275 hw->mac.get_link_status = 1;
1276 /* guard against interrupt when we're going down */
1277 if (!test_bit(__E1000_DOWN, &adapter->state))
1278 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1281 no_link_interrupt:
1282 if (!test_bit(__E1000_DOWN, &adapter->state))
1283 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1285 return IRQ_HANDLED;
1289 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1291 struct net_device *netdev = data;
1292 struct e1000_adapter *adapter = netdev_priv(netdev);
1293 struct e1000_hw *hw = &adapter->hw;
1294 struct e1000_ring *tx_ring = adapter->tx_ring;
1297 adapter->total_tx_bytes = 0;
1298 adapter->total_tx_packets = 0;
1300 if (!e1000_clean_tx_irq(adapter))
1301 /* Ring was not completely cleaned, so fire another interrupt */
1302 ew32(ICS, tx_ring->ims_val);
1304 return IRQ_HANDLED;
1307 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1309 struct net_device *netdev = data;
1310 struct e1000_adapter *adapter = netdev_priv(netdev);
1312 /* Write the ITR value calculated at the end of the
1313 * previous interrupt.
1315 if (adapter->rx_ring->set_itr) {
1316 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1317 adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1318 adapter->rx_ring->set_itr = 0;
1321 if (napi_schedule_prep(&adapter->napi)) {
1322 adapter->total_rx_bytes = 0;
1323 adapter->total_rx_packets = 0;
1324 __napi_schedule(&adapter->napi);
1326 return IRQ_HANDLED;
1330 * e1000_configure_msix - Configure MSI-X hardware
1332 * e1000_configure_msix sets up the hardware to properly
1333 * generate MSI-X interrupts.
1335 static void e1000_configure_msix(struct e1000_adapter *adapter)
1337 struct e1000_hw *hw = &adapter->hw;
1338 struct e1000_ring *rx_ring = adapter->rx_ring;
1339 struct e1000_ring *tx_ring = adapter->tx_ring;
1340 int vector = 0;
1341 u32 ctrl_ext, ivar = 0;
1343 adapter->eiac_mask = 0;
1345 /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1346 if (hw->mac.type == e1000_82574) {
1347 u32 rfctl = er32(RFCTL);
1348 rfctl |= E1000_RFCTL_ACK_DIS;
1349 ew32(RFCTL, rfctl);
1352 #define E1000_IVAR_INT_ALLOC_VALID 0x8
1353 /* Configure Rx vector */
1354 rx_ring->ims_val = E1000_IMS_RXQ0;
1355 adapter->eiac_mask |= rx_ring->ims_val;
1356 if (rx_ring->itr_val)
1357 writel(1000000000 / (rx_ring->itr_val * 256),
1358 hw->hw_addr + rx_ring->itr_register);
1359 else
1360 writel(1, hw->hw_addr + rx_ring->itr_register);
1361 ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1363 /* Configure Tx vector */
1364 tx_ring->ims_val = E1000_IMS_TXQ0;
1365 vector++;
1366 if (tx_ring->itr_val)
1367 writel(1000000000 / (tx_ring->itr_val * 256),
1368 hw->hw_addr + tx_ring->itr_register);
1369 else
1370 writel(1, hw->hw_addr + tx_ring->itr_register);
1371 adapter->eiac_mask |= tx_ring->ims_val;
1372 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1374 /* set vector for Other Causes, e.g. link changes */
1375 vector++;
1376 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1377 if (rx_ring->itr_val)
1378 writel(1000000000 / (rx_ring->itr_val * 256),
1379 hw->hw_addr + E1000_EITR_82574(vector));
1380 else
1381 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1383 /* Cause Tx interrupts on every write back */
1384 ivar |= (1 << 31);
1386 ew32(IVAR, ivar);
1388 /* enable MSI-X PBA support */
1389 ctrl_ext = er32(CTRL_EXT);
1390 ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1392 /* Auto-Mask Other interrupts upon ICR read */
1393 #define E1000_EIAC_MASK_82574 0x01F00000
1394 ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1395 ctrl_ext |= E1000_CTRL_EXT_EIAME;
1396 ew32(CTRL_EXT, ctrl_ext);
1397 e1e_flush();
1400 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1402 if (adapter->msix_entries) {
1403 pci_disable_msix(adapter->pdev);
1404 kfree(adapter->msix_entries);
1405 adapter->msix_entries = NULL;
1406 } else if (adapter->flags & FLAG_MSI_ENABLED) {
1407 pci_disable_msi(adapter->pdev);
1408 adapter->flags &= ~FLAG_MSI_ENABLED;
1411 return;
1415 * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1417 * Attempt to configure interrupts using the best available
1418 * capabilities of the hardware and kernel.
1420 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1422 int err;
1423 int numvecs, i;
1426 switch (adapter->int_mode) {
1427 case E1000E_INT_MODE_MSIX:
1428 if (adapter->flags & FLAG_HAS_MSIX) {
1429 numvecs = 3; /* RxQ0, TxQ0 and other */
1430 adapter->msix_entries = kcalloc(numvecs,
1431 sizeof(struct msix_entry),
1432 GFP_KERNEL);
1433 if (adapter->msix_entries) {
1434 for (i = 0; i < numvecs; i++)
1435 adapter->msix_entries[i].entry = i;
1437 err = pci_enable_msix(adapter->pdev,
1438 adapter->msix_entries,
1439 numvecs);
1440 if (err == 0)
1441 return;
1443 /* MSI-X failed, so fall through and try MSI */
1444 e_err("Failed to initialize MSI-X interrupts. "
1445 "Falling back to MSI interrupts.\n");
1446 e1000e_reset_interrupt_capability(adapter);
1448 adapter->int_mode = E1000E_INT_MODE_MSI;
1449 /* Fall through */
1450 case E1000E_INT_MODE_MSI:
1451 if (!pci_enable_msi(adapter->pdev)) {
1452 adapter->flags |= FLAG_MSI_ENABLED;
1453 } else {
1454 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1455 e_err("Failed to initialize MSI interrupts. Falling "
1456 "back to legacy interrupts.\n");
1458 /* Fall through */
1459 case E1000E_INT_MODE_LEGACY:
1460 /* Don't do anything; this is the system default */
1461 break;
1464 return;
1468 * e1000_request_msix - Initialize MSI-X interrupts
1470 * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1471 * kernel.
1473 static int e1000_request_msix(struct e1000_adapter *adapter)
1475 struct net_device *netdev = adapter->netdev;
1476 int err = 0, vector = 0;
1478 if (strlen(netdev->name) < (IFNAMSIZ - 5))
1479 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1480 else
1481 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1482 err = request_irq(adapter->msix_entries[vector].vector,
1483 &e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1484 netdev);
1485 if (err)
1486 goto out;
1487 adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1488 adapter->rx_ring->itr_val = adapter->itr;
1489 vector++;
1491 if (strlen(netdev->name) < (IFNAMSIZ - 5))
1492 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1493 else
1494 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1495 err = request_irq(adapter->msix_entries[vector].vector,
1496 &e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1497 netdev);
1498 if (err)
1499 goto out;
1500 adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1501 adapter->tx_ring->itr_val = adapter->itr;
1502 vector++;
1504 err = request_irq(adapter->msix_entries[vector].vector,
1505 &e1000_msix_other, 0, netdev->name, netdev);
1506 if (err)
1507 goto out;
1509 e1000_configure_msix(adapter);
1510 return 0;
1511 out:
1512 return err;
1516 * e1000_request_irq - initialize interrupts
1518 * Attempts to configure interrupts using the best available
1519 * capabilities of the hardware and kernel.
1521 static int e1000_request_irq(struct e1000_adapter *adapter)
1523 struct net_device *netdev = adapter->netdev;
1524 int err;
1526 if (adapter->msix_entries) {
1527 err = e1000_request_msix(adapter);
1528 if (!err)
1529 return err;
1530 /* fall back to MSI */
1531 e1000e_reset_interrupt_capability(adapter);
1532 adapter->int_mode = E1000E_INT_MODE_MSI;
1533 e1000e_set_interrupt_capability(adapter);
1535 if (adapter->flags & FLAG_MSI_ENABLED) {
1536 err = request_irq(adapter->pdev->irq, &e1000_intr_msi, 0,
1537 netdev->name, netdev);
1538 if (!err)
1539 return err;
1541 /* fall back to legacy interrupt */
1542 e1000e_reset_interrupt_capability(adapter);
1543 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1546 err = request_irq(adapter->pdev->irq, &e1000_intr, IRQF_SHARED,
1547 netdev->name, netdev);
1548 if (err)
1549 e_err("Unable to allocate interrupt, Error: %d\n", err);
1551 return err;
1554 static void e1000_free_irq(struct e1000_adapter *adapter)
1556 struct net_device *netdev = adapter->netdev;
1558 if (adapter->msix_entries) {
1559 int vector = 0;
1561 free_irq(adapter->msix_entries[vector].vector, netdev);
1562 vector++;
1564 free_irq(adapter->msix_entries[vector].vector, netdev);
1565 vector++;
1567 /* Other Causes interrupt vector */
1568 free_irq(adapter->msix_entries[vector].vector, netdev);
1569 return;
1572 free_irq(adapter->pdev->irq, netdev);
1576 * e1000_irq_disable - Mask off interrupt generation on the NIC
1578 static void e1000_irq_disable(struct e1000_adapter *adapter)
1580 struct e1000_hw *hw = &adapter->hw;
1582 ew32(IMC, ~0);
1583 if (adapter->msix_entries)
1584 ew32(EIAC_82574, 0);
1585 e1e_flush();
1586 synchronize_irq(adapter->pdev->irq);
1590 * e1000_irq_enable - Enable default interrupt generation settings
1592 static void e1000_irq_enable(struct e1000_adapter *adapter)
1594 struct e1000_hw *hw = &adapter->hw;
1596 if (adapter->msix_entries) {
1597 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1598 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1599 } else {
1600 ew32(IMS, IMS_ENABLE_MASK);
1602 e1e_flush();
1606 * e1000_get_hw_control - get control of the h/w from f/w
1607 * @adapter: address of board private structure
1609 * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1610 * For ASF and Pass Through versions of f/w this means that
1611 * the driver is loaded. For AMT version (only with 82573)
1612 * of the f/w this means that the network i/f is open.
1614 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1616 struct e1000_hw *hw = &adapter->hw;
1617 u32 ctrl_ext;
1618 u32 swsm;
1620 /* Let firmware know the driver has taken over */
1621 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1622 swsm = er32(SWSM);
1623 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1624 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1625 ctrl_ext = er32(CTRL_EXT);
1626 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1631 * e1000_release_hw_control - release control of the h/w to f/w
1632 * @adapter: address of board private structure
1634 * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1635 * For ASF and Pass Through versions of f/w this means that the
1636 * driver is no longer loaded. For AMT version (only with 82573) i
1637 * of the f/w this means that the network i/f is closed.
1640 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1642 struct e1000_hw *hw = &adapter->hw;
1643 u32 ctrl_ext;
1644 u32 swsm;
1646 /* Let firmware taken over control of h/w */
1647 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1648 swsm = er32(SWSM);
1649 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1650 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1651 ctrl_ext = er32(CTRL_EXT);
1652 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1657 * @e1000_alloc_ring - allocate memory for a ring structure
1659 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1660 struct e1000_ring *ring)
1662 struct pci_dev *pdev = adapter->pdev;
1664 ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1665 GFP_KERNEL);
1666 if (!ring->desc)
1667 return -ENOMEM;
1669 return 0;
1673 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1674 * @adapter: board private structure
1676 * Return 0 on success, negative on failure
1678 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1680 struct e1000_ring *tx_ring = adapter->tx_ring;
1681 int err = -ENOMEM, size;
1683 size = sizeof(struct e1000_buffer) * tx_ring->count;
1684 tx_ring->buffer_info = vmalloc(size);
1685 if (!tx_ring->buffer_info)
1686 goto err;
1687 memset(tx_ring->buffer_info, 0, size);
1689 /* round up to nearest 4K */
1690 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1691 tx_ring->size = ALIGN(tx_ring->size, 4096);
1693 err = e1000_alloc_ring_dma(adapter, tx_ring);
1694 if (err)
1695 goto err;
1697 tx_ring->next_to_use = 0;
1698 tx_ring->next_to_clean = 0;
1700 return 0;
1701 err:
1702 vfree(tx_ring->buffer_info);
1703 e_err("Unable to allocate memory for the transmit descriptor ring\n");
1704 return err;
1708 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1709 * @adapter: board private structure
1711 * Returns 0 on success, negative on failure
1713 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1715 struct e1000_ring *rx_ring = adapter->rx_ring;
1716 struct e1000_buffer *buffer_info;
1717 int i, size, desc_len, err = -ENOMEM;
1719 size = sizeof(struct e1000_buffer) * rx_ring->count;
1720 rx_ring->buffer_info = vmalloc(size);
1721 if (!rx_ring->buffer_info)
1722 goto err;
1723 memset(rx_ring->buffer_info, 0, size);
1725 for (i = 0; i < rx_ring->count; i++) {
1726 buffer_info = &rx_ring->buffer_info[i];
1727 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1728 sizeof(struct e1000_ps_page),
1729 GFP_KERNEL);
1730 if (!buffer_info->ps_pages)
1731 goto err_pages;
1734 desc_len = sizeof(union e1000_rx_desc_packet_split);
1736 /* Round up to nearest 4K */
1737 rx_ring->size = rx_ring->count * desc_len;
1738 rx_ring->size = ALIGN(rx_ring->size, 4096);
1740 err = e1000_alloc_ring_dma(adapter, rx_ring);
1741 if (err)
1742 goto err_pages;
1744 rx_ring->next_to_clean = 0;
1745 rx_ring->next_to_use = 0;
1746 rx_ring->rx_skb_top = NULL;
1748 return 0;
1750 err_pages:
1751 for (i = 0; i < rx_ring->count; i++) {
1752 buffer_info = &rx_ring->buffer_info[i];
1753 kfree(buffer_info->ps_pages);
1755 err:
1756 vfree(rx_ring->buffer_info);
1757 e_err("Unable to allocate memory for the transmit descriptor ring\n");
1758 return err;
1762 * e1000_clean_tx_ring - Free Tx Buffers
1763 * @adapter: board private structure
1765 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1767 struct e1000_ring *tx_ring = adapter->tx_ring;
1768 struct e1000_buffer *buffer_info;
1769 unsigned long size;
1770 unsigned int i;
1772 for (i = 0; i < tx_ring->count; i++) {
1773 buffer_info = &tx_ring->buffer_info[i];
1774 e1000_put_txbuf(adapter, buffer_info);
1777 size = sizeof(struct e1000_buffer) * tx_ring->count;
1778 memset(tx_ring->buffer_info, 0, size);
1780 memset(tx_ring->desc, 0, tx_ring->size);
1782 tx_ring->next_to_use = 0;
1783 tx_ring->next_to_clean = 0;
1785 writel(0, adapter->hw.hw_addr + tx_ring->head);
1786 writel(0, adapter->hw.hw_addr + tx_ring->tail);
1790 * e1000e_free_tx_resources - Free Tx Resources per Queue
1791 * @adapter: board private structure
1793 * Free all transmit software resources
1795 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1797 struct pci_dev *pdev = adapter->pdev;
1798 struct e1000_ring *tx_ring = adapter->tx_ring;
1800 e1000_clean_tx_ring(adapter);
1802 vfree(tx_ring->buffer_info);
1803 tx_ring->buffer_info = NULL;
1805 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1806 tx_ring->dma);
1807 tx_ring->desc = NULL;
1811 * e1000e_free_rx_resources - Free Rx Resources
1812 * @adapter: board private structure
1814 * Free all receive software resources
1817 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1819 struct pci_dev *pdev = adapter->pdev;
1820 struct e1000_ring *rx_ring = adapter->rx_ring;
1821 int i;
1823 e1000_clean_rx_ring(adapter);
1825 for (i = 0; i < rx_ring->count; i++) {
1826 kfree(rx_ring->buffer_info[i].ps_pages);
1829 vfree(rx_ring->buffer_info);
1830 rx_ring->buffer_info = NULL;
1832 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1833 rx_ring->dma);
1834 rx_ring->desc = NULL;
1838 * e1000_update_itr - update the dynamic ITR value based on statistics
1839 * @adapter: pointer to adapter
1840 * @itr_setting: current adapter->itr
1841 * @packets: the number of packets during this measurement interval
1842 * @bytes: the number of bytes during this measurement interval
1844 * Stores a new ITR value based on packets and byte
1845 * counts during the last interrupt. The advantage of per interrupt
1846 * computation is faster updates and more accurate ITR for the current
1847 * traffic pattern. Constants in this function were computed
1848 * based on theoretical maximum wire speed and thresholds were set based
1849 * on testing data as well as attempting to minimize response time
1850 * while increasing bulk throughput. This functionality is controlled
1851 * by the InterruptThrottleRate module parameter.
1853 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1854 u16 itr_setting, int packets,
1855 int bytes)
1857 unsigned int retval = itr_setting;
1859 if (packets == 0)
1860 goto update_itr_done;
1862 switch (itr_setting) {
1863 case lowest_latency:
1864 /* handle TSO and jumbo frames */
1865 if (bytes/packets > 8000)
1866 retval = bulk_latency;
1867 else if ((packets < 5) && (bytes > 512)) {
1868 retval = low_latency;
1870 break;
1871 case low_latency: /* 50 usec aka 20000 ints/s */
1872 if (bytes > 10000) {
1873 /* this if handles the TSO accounting */
1874 if (bytes/packets > 8000) {
1875 retval = bulk_latency;
1876 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1877 retval = bulk_latency;
1878 } else if ((packets > 35)) {
1879 retval = lowest_latency;
1881 } else if (bytes/packets > 2000) {
1882 retval = bulk_latency;
1883 } else if (packets <= 2 && bytes < 512) {
1884 retval = lowest_latency;
1886 break;
1887 case bulk_latency: /* 250 usec aka 4000 ints/s */
1888 if (bytes > 25000) {
1889 if (packets > 35) {
1890 retval = low_latency;
1892 } else if (bytes < 6000) {
1893 retval = low_latency;
1895 break;
1898 update_itr_done:
1899 return retval;
1902 static void e1000_set_itr(struct e1000_adapter *adapter)
1904 struct e1000_hw *hw = &adapter->hw;
1905 u16 current_itr;
1906 u32 new_itr = adapter->itr;
1908 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1909 if (adapter->link_speed != SPEED_1000) {
1910 current_itr = 0;
1911 new_itr = 4000;
1912 goto set_itr_now;
1915 adapter->tx_itr = e1000_update_itr(adapter,
1916 adapter->tx_itr,
1917 adapter->total_tx_packets,
1918 adapter->total_tx_bytes);
1919 /* conservative mode (itr 3) eliminates the lowest_latency setting */
1920 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1921 adapter->tx_itr = low_latency;
1923 adapter->rx_itr = e1000_update_itr(adapter,
1924 adapter->rx_itr,
1925 adapter->total_rx_packets,
1926 adapter->total_rx_bytes);
1927 /* conservative mode (itr 3) eliminates the lowest_latency setting */
1928 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1929 adapter->rx_itr = low_latency;
1931 current_itr = max(adapter->rx_itr, adapter->tx_itr);
1933 switch (current_itr) {
1934 /* counts and packets in update_itr are dependent on these numbers */
1935 case lowest_latency:
1936 new_itr = 70000;
1937 break;
1938 case low_latency:
1939 new_itr = 20000; /* aka hwitr = ~200 */
1940 break;
1941 case bulk_latency:
1942 new_itr = 4000;
1943 break;
1944 default:
1945 break;
1948 set_itr_now:
1949 if (new_itr != adapter->itr) {
1951 * this attempts to bias the interrupt rate towards Bulk
1952 * by adding intermediate steps when interrupt rate is
1953 * increasing
1955 new_itr = new_itr > adapter->itr ?
1956 min(adapter->itr + (new_itr >> 2), new_itr) :
1957 new_itr;
1958 adapter->itr = new_itr;
1959 adapter->rx_ring->itr_val = new_itr;
1960 if (adapter->msix_entries)
1961 adapter->rx_ring->set_itr = 1;
1962 else
1963 ew32(ITR, 1000000000 / (new_itr * 256));
1968 * e1000_alloc_queues - Allocate memory for all rings
1969 * @adapter: board private structure to initialize
1971 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1973 adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1974 if (!adapter->tx_ring)
1975 goto err;
1977 adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1978 if (!adapter->rx_ring)
1979 goto err;
1981 return 0;
1982 err:
1983 e_err("Unable to allocate memory for queues\n");
1984 kfree(adapter->rx_ring);
1985 kfree(adapter->tx_ring);
1986 return -ENOMEM;
1990 * e1000_clean - NAPI Rx polling callback
1991 * @napi: struct associated with this polling callback
1992 * @budget: amount of packets driver is allowed to process this poll
1994 static int e1000_clean(struct napi_struct *napi, int budget)
1996 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1997 struct e1000_hw *hw = &adapter->hw;
1998 struct net_device *poll_dev = adapter->netdev;
1999 int tx_cleaned = 0, work_done = 0;
2001 adapter = netdev_priv(poll_dev);
2003 if (adapter->msix_entries &&
2004 !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2005 goto clean_rx;
2007 tx_cleaned = e1000_clean_tx_irq(adapter);
2009 clean_rx:
2010 adapter->clean_rx(adapter, &work_done, budget);
2012 if (!tx_cleaned)
2013 work_done = budget;
2015 /* If budget not fully consumed, exit the polling mode */
2016 if (work_done < budget) {
2017 if (adapter->itr_setting & 3)
2018 e1000_set_itr(adapter);
2019 napi_complete(napi);
2020 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2021 if (adapter->msix_entries)
2022 ew32(IMS, adapter->rx_ring->ims_val);
2023 else
2024 e1000_irq_enable(adapter);
2028 return work_done;
2031 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2033 struct e1000_adapter *adapter = netdev_priv(netdev);
2034 struct e1000_hw *hw = &adapter->hw;
2035 u32 vfta, index;
2037 /* don't update vlan cookie if already programmed */
2038 if ((adapter->hw.mng_cookie.status &
2039 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2040 (vid == adapter->mng_vlan_id))
2041 return;
2042 /* add VID to filter table */
2043 index = (vid >> 5) & 0x7F;
2044 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2045 vfta |= (1 << (vid & 0x1F));
2046 e1000e_write_vfta(hw, index, vfta);
2049 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2051 struct e1000_adapter *adapter = netdev_priv(netdev);
2052 struct e1000_hw *hw = &adapter->hw;
2053 u32 vfta, index;
2055 if (!test_bit(__E1000_DOWN, &adapter->state))
2056 e1000_irq_disable(adapter);
2057 vlan_group_set_device(adapter->vlgrp, vid, NULL);
2059 if (!test_bit(__E1000_DOWN, &adapter->state))
2060 e1000_irq_enable(adapter);
2062 if ((adapter->hw.mng_cookie.status &
2063 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2064 (vid == adapter->mng_vlan_id)) {
2065 /* release control to f/w */
2066 e1000_release_hw_control(adapter);
2067 return;
2070 /* remove VID from filter table */
2071 index = (vid >> 5) & 0x7F;
2072 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2073 vfta &= ~(1 << (vid & 0x1F));
2074 e1000e_write_vfta(hw, index, vfta);
2077 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2079 struct net_device *netdev = adapter->netdev;
2080 u16 vid = adapter->hw.mng_cookie.vlan_id;
2081 u16 old_vid = adapter->mng_vlan_id;
2083 if (!adapter->vlgrp)
2084 return;
2086 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2087 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2088 if (adapter->hw.mng_cookie.status &
2089 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2090 e1000_vlan_rx_add_vid(netdev, vid);
2091 adapter->mng_vlan_id = vid;
2094 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2095 (vid != old_vid) &&
2096 !vlan_group_get_device(adapter->vlgrp, old_vid))
2097 e1000_vlan_rx_kill_vid(netdev, old_vid);
2098 } else {
2099 adapter->mng_vlan_id = vid;
2104 static void e1000_vlan_rx_register(struct net_device *netdev,
2105 struct vlan_group *grp)
2107 struct e1000_adapter *adapter = netdev_priv(netdev);
2108 struct e1000_hw *hw = &adapter->hw;
2109 u32 ctrl, rctl;
2111 if (!test_bit(__E1000_DOWN, &adapter->state))
2112 e1000_irq_disable(adapter);
2113 adapter->vlgrp = grp;
2115 if (grp) {
2116 /* enable VLAN tag insert/strip */
2117 ctrl = er32(CTRL);
2118 ctrl |= E1000_CTRL_VME;
2119 ew32(CTRL, ctrl);
2121 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2122 /* enable VLAN receive filtering */
2123 rctl = er32(RCTL);
2124 rctl &= ~E1000_RCTL_CFIEN;
2125 ew32(RCTL, rctl);
2126 e1000_update_mng_vlan(adapter);
2128 } else {
2129 /* disable VLAN tag insert/strip */
2130 ctrl = er32(CTRL);
2131 ctrl &= ~E1000_CTRL_VME;
2132 ew32(CTRL, ctrl);
2134 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2135 if (adapter->mng_vlan_id !=
2136 (u16)E1000_MNG_VLAN_NONE) {
2137 e1000_vlan_rx_kill_vid(netdev,
2138 adapter->mng_vlan_id);
2139 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2144 if (!test_bit(__E1000_DOWN, &adapter->state))
2145 e1000_irq_enable(adapter);
2148 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2150 u16 vid;
2152 e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2154 if (!adapter->vlgrp)
2155 return;
2157 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2158 if (!vlan_group_get_device(adapter->vlgrp, vid))
2159 continue;
2160 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2164 static void e1000_init_manageability(struct e1000_adapter *adapter)
2166 struct e1000_hw *hw = &adapter->hw;
2167 u32 manc, manc2h;
2169 if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2170 return;
2172 manc = er32(MANC);
2175 * enable receiving management packets to the host. this will probably
2176 * generate destination unreachable messages from the host OS, but
2177 * the packets will be handled on SMBUS
2179 manc |= E1000_MANC_EN_MNG2HOST;
2180 manc2h = er32(MANC2H);
2181 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2182 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2183 manc2h |= E1000_MNG2HOST_PORT_623;
2184 manc2h |= E1000_MNG2HOST_PORT_664;
2185 ew32(MANC2H, manc2h);
2186 ew32(MANC, manc);
2190 * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2191 * @adapter: board private structure
2193 * Configure the Tx unit of the MAC after a reset.
2195 static void e1000_configure_tx(struct e1000_adapter *adapter)
2197 struct e1000_hw *hw = &adapter->hw;
2198 struct e1000_ring *tx_ring = adapter->tx_ring;
2199 u64 tdba;
2200 u32 tdlen, tctl, tipg, tarc;
2201 u32 ipgr1, ipgr2;
2203 /* Setup the HW Tx Head and Tail descriptor pointers */
2204 tdba = tx_ring->dma;
2205 tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2206 ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2207 ew32(TDBAH, (tdba >> 32));
2208 ew32(TDLEN, tdlen);
2209 ew32(TDH, 0);
2210 ew32(TDT, 0);
2211 tx_ring->head = E1000_TDH;
2212 tx_ring->tail = E1000_TDT;
2214 /* Set the default values for the Tx Inter Packet Gap timer */
2215 tipg = DEFAULT_82543_TIPG_IPGT_COPPER; /* 8 */
2216 ipgr1 = DEFAULT_82543_TIPG_IPGR1; /* 8 */
2217 ipgr2 = DEFAULT_82543_TIPG_IPGR2; /* 6 */
2219 if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2220 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /* 7 */
2222 tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2223 tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2224 ew32(TIPG, tipg);
2226 /* Set the Tx Interrupt Delay register */
2227 ew32(TIDV, adapter->tx_int_delay);
2228 /* Tx irq moderation */
2229 ew32(TADV, adapter->tx_abs_int_delay);
2231 /* Program the Transmit Control Register */
2232 tctl = er32(TCTL);
2233 tctl &= ~E1000_TCTL_CT;
2234 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2235 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2237 if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2238 tarc = er32(TARC(0));
2240 * set the speed mode bit, we'll clear it if we're not at
2241 * gigabit link later
2243 #define SPEED_MODE_BIT (1 << 21)
2244 tarc |= SPEED_MODE_BIT;
2245 ew32(TARC(0), tarc);
2248 /* errata: program both queues to unweighted RR */
2249 if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2250 tarc = er32(TARC(0));
2251 tarc |= 1;
2252 ew32(TARC(0), tarc);
2253 tarc = er32(TARC(1));
2254 tarc |= 1;
2255 ew32(TARC(1), tarc);
2258 e1000e_config_collision_dist(hw);
2260 /* Setup Transmit Descriptor Settings for eop descriptor */
2261 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2263 /* only set IDE if we are delaying interrupts using the timers */
2264 if (adapter->tx_int_delay)
2265 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2267 /* enable Report Status bit */
2268 adapter->txd_cmd |= E1000_TXD_CMD_RS;
2270 ew32(TCTL, tctl);
2272 adapter->tx_queue_len = adapter->netdev->tx_queue_len;
2276 * e1000_setup_rctl - configure the receive control registers
2277 * @adapter: Board private structure
2279 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2280 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2281 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2283 struct e1000_hw *hw = &adapter->hw;
2284 u32 rctl, rfctl;
2285 u32 psrctl = 0;
2286 u32 pages = 0;
2288 /* Program MC offset vector base */
2289 rctl = er32(RCTL);
2290 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2291 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2292 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2293 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2295 /* Do not Store bad packets */
2296 rctl &= ~E1000_RCTL_SBP;
2298 /* Enable Long Packet receive */
2299 if (adapter->netdev->mtu <= ETH_DATA_LEN)
2300 rctl &= ~E1000_RCTL_LPE;
2301 else
2302 rctl |= E1000_RCTL_LPE;
2304 /* Some systems expect that the CRC is included in SMBUS traffic. The
2305 * hardware strips the CRC before sending to both SMBUS (BMC) and to
2306 * host memory when this is enabled
2308 if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2309 rctl |= E1000_RCTL_SECRC;
2311 /* Setup buffer sizes */
2312 rctl &= ~E1000_RCTL_SZ_4096;
2313 rctl |= E1000_RCTL_BSEX;
2314 switch (adapter->rx_buffer_len) {
2315 case 256:
2316 rctl |= E1000_RCTL_SZ_256;
2317 rctl &= ~E1000_RCTL_BSEX;
2318 break;
2319 case 512:
2320 rctl |= E1000_RCTL_SZ_512;
2321 rctl &= ~E1000_RCTL_BSEX;
2322 break;
2323 case 1024:
2324 rctl |= E1000_RCTL_SZ_1024;
2325 rctl &= ~E1000_RCTL_BSEX;
2326 break;
2327 case 2048:
2328 default:
2329 rctl |= E1000_RCTL_SZ_2048;
2330 rctl &= ~E1000_RCTL_BSEX;
2331 break;
2332 case 4096:
2333 rctl |= E1000_RCTL_SZ_4096;
2334 break;
2335 case 8192:
2336 rctl |= E1000_RCTL_SZ_8192;
2337 break;
2338 case 16384:
2339 rctl |= E1000_RCTL_SZ_16384;
2340 break;
2344 * 82571 and greater support packet-split where the protocol
2345 * header is placed in skb->data and the packet data is
2346 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2347 * In the case of a non-split, skb->data is linearly filled,
2348 * followed by the page buffers. Therefore, skb->data is
2349 * sized to hold the largest protocol header.
2351 * allocations using alloc_page take too long for regular MTU
2352 * so only enable packet split for jumbo frames
2354 * Using pages when the page size is greater than 16k wastes
2355 * a lot of memory, since we allocate 3 pages at all times
2356 * per packet.
2358 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2359 if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2360 (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2361 adapter->rx_ps_pages = pages;
2362 else
2363 adapter->rx_ps_pages = 0;
2365 if (adapter->rx_ps_pages) {
2366 /* Configure extra packet-split registers */
2367 rfctl = er32(RFCTL);
2368 rfctl |= E1000_RFCTL_EXTEN;
2370 * disable packet split support for IPv6 extension headers,
2371 * because some malformed IPv6 headers can hang the Rx
2373 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2374 E1000_RFCTL_NEW_IPV6_EXT_DIS);
2376 ew32(RFCTL, rfctl);
2378 /* Enable Packet split descriptors */
2379 rctl |= E1000_RCTL_DTYP_PS;
2381 psrctl |= adapter->rx_ps_bsize0 >>
2382 E1000_PSRCTL_BSIZE0_SHIFT;
2384 switch (adapter->rx_ps_pages) {
2385 case 3:
2386 psrctl |= PAGE_SIZE <<
2387 E1000_PSRCTL_BSIZE3_SHIFT;
2388 case 2:
2389 psrctl |= PAGE_SIZE <<
2390 E1000_PSRCTL_BSIZE2_SHIFT;
2391 case 1:
2392 psrctl |= PAGE_SIZE >>
2393 E1000_PSRCTL_BSIZE1_SHIFT;
2394 break;
2397 ew32(PSRCTL, psrctl);
2400 ew32(RCTL, rctl);
2401 /* just started the receive unit, no need to restart */
2402 adapter->flags &= ~FLAG_RX_RESTART_NOW;
2406 * e1000_configure_rx - Configure Receive Unit after Reset
2407 * @adapter: board private structure
2409 * Configure the Rx unit of the MAC after a reset.
2411 static void e1000_configure_rx(struct e1000_adapter *adapter)
2413 struct e1000_hw *hw = &adapter->hw;
2414 struct e1000_ring *rx_ring = adapter->rx_ring;
2415 u64 rdba;
2416 u32 rdlen, rctl, rxcsum, ctrl_ext;
2418 if (adapter->rx_ps_pages) {
2419 /* this is a 32 byte descriptor */
2420 rdlen = rx_ring->count *
2421 sizeof(union e1000_rx_desc_packet_split);
2422 adapter->clean_rx = e1000_clean_rx_irq_ps;
2423 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2424 } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2425 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2426 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2427 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2428 } else {
2429 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2430 adapter->clean_rx = e1000_clean_rx_irq;
2431 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2434 /* disable receives while setting up the descriptors */
2435 rctl = er32(RCTL);
2436 ew32(RCTL, rctl & ~E1000_RCTL_EN);
2437 e1e_flush();
2438 msleep(10);
2440 /* set the Receive Delay Timer Register */
2441 ew32(RDTR, adapter->rx_int_delay);
2443 /* irq moderation */
2444 ew32(RADV, adapter->rx_abs_int_delay);
2445 if (adapter->itr_setting != 0)
2446 ew32(ITR, 1000000000 / (adapter->itr * 256));
2448 ctrl_ext = er32(CTRL_EXT);
2449 /* Reset delay timers after every interrupt */
2450 ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2451 /* Auto-Mask interrupts upon ICR access */
2452 ctrl_ext |= E1000_CTRL_EXT_IAME;
2453 ew32(IAM, 0xffffffff);
2454 ew32(CTRL_EXT, ctrl_ext);
2455 e1e_flush();
2458 * Setup the HW Rx Head and Tail Descriptor Pointers and
2459 * the Base and Length of the Rx Descriptor Ring
2461 rdba = rx_ring->dma;
2462 ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2463 ew32(RDBAH, (rdba >> 32));
2464 ew32(RDLEN, rdlen);
2465 ew32(RDH, 0);
2466 ew32(RDT, 0);
2467 rx_ring->head = E1000_RDH;
2468 rx_ring->tail = E1000_RDT;
2470 /* Enable Receive Checksum Offload for TCP and UDP */
2471 rxcsum = er32(RXCSUM);
2472 if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2473 rxcsum |= E1000_RXCSUM_TUOFL;
2476 * IPv4 payload checksum for UDP fragments must be
2477 * used in conjunction with packet-split.
2479 if (adapter->rx_ps_pages)
2480 rxcsum |= E1000_RXCSUM_IPPCSE;
2481 } else {
2482 rxcsum &= ~E1000_RXCSUM_TUOFL;
2483 /* no need to clear IPPCSE as it defaults to 0 */
2485 ew32(RXCSUM, rxcsum);
2488 * Enable early receives on supported devices, only takes effect when
2489 * packet size is equal or larger than the specified value (in 8 byte
2490 * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2492 if ((adapter->flags & FLAG_HAS_ERT) &&
2493 (adapter->netdev->mtu > ETH_DATA_LEN)) {
2494 u32 rxdctl = er32(RXDCTL(0));
2495 ew32(RXDCTL(0), rxdctl | 0x3);
2496 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2498 * With jumbo frames and early-receive enabled, excessive
2499 * C4->C2 latencies result in dropped transactions.
2501 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2502 e1000e_driver_name, 55);
2503 } else {
2504 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2505 e1000e_driver_name,
2506 PM_QOS_DEFAULT_VALUE);
2509 /* Enable Receives */
2510 ew32(RCTL, rctl);
2514 * e1000_update_mc_addr_list - Update Multicast addresses
2515 * @hw: pointer to the HW structure
2516 * @mc_addr_list: array of multicast addresses to program
2517 * @mc_addr_count: number of multicast addresses to program
2518 * @rar_used_count: the first RAR register free to program
2519 * @rar_count: total number of supported Receive Address Registers
2521 * Updates the Receive Address Registers and Multicast Table Array.
2522 * The caller must have a packed mc_addr_list of multicast addresses.
2523 * The parameter rar_count will usually be hw->mac.rar_entry_count
2524 * unless there are workarounds that change this. Currently no func pointer
2525 * exists and all implementations are handled in the generic version of this
2526 * function.
2528 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2529 u32 mc_addr_count, u32 rar_used_count,
2530 u32 rar_count)
2532 hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2533 rar_used_count, rar_count);
2537 * e1000_set_multi - Multicast and Promiscuous mode set
2538 * @netdev: network interface device structure
2540 * The set_multi entry point is called whenever the multicast address
2541 * list or the network interface flags are updated. This routine is
2542 * responsible for configuring the hardware for proper multicast,
2543 * promiscuous mode, and all-multi behavior.
2545 static void e1000_set_multi(struct net_device *netdev)
2547 struct e1000_adapter *adapter = netdev_priv(netdev);
2548 struct e1000_hw *hw = &adapter->hw;
2549 struct e1000_mac_info *mac = &hw->mac;
2550 struct dev_mc_list *mc_ptr;
2551 u8 *mta_list;
2552 u32 rctl;
2553 int i;
2555 /* Check for Promiscuous and All Multicast modes */
2557 rctl = er32(RCTL);
2559 if (netdev->flags & IFF_PROMISC) {
2560 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2561 rctl &= ~E1000_RCTL_VFE;
2562 } else {
2563 if (netdev->flags & IFF_ALLMULTI) {
2564 rctl |= E1000_RCTL_MPE;
2565 rctl &= ~E1000_RCTL_UPE;
2566 } else {
2567 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2569 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2570 rctl |= E1000_RCTL_VFE;
2573 ew32(RCTL, rctl);
2575 if (netdev->mc_count) {
2576 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2577 if (!mta_list)
2578 return;
2580 /* prepare a packed array of only addresses. */
2581 mc_ptr = netdev->mc_list;
2583 for (i = 0; i < netdev->mc_count; i++) {
2584 if (!mc_ptr)
2585 break;
2586 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2587 ETH_ALEN);
2588 mc_ptr = mc_ptr->next;
2591 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2592 mac->rar_entry_count);
2593 kfree(mta_list);
2594 } else {
2596 * if we're called from probe, we might not have
2597 * anything to do here, so clear out the list
2599 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2604 * e1000_configure - configure the hardware for Rx and Tx
2605 * @adapter: private board structure
2607 static void e1000_configure(struct e1000_adapter *adapter)
2609 e1000_set_multi(adapter->netdev);
2611 e1000_restore_vlan(adapter);
2612 e1000_init_manageability(adapter);
2614 e1000_configure_tx(adapter);
2615 e1000_setup_rctl(adapter);
2616 e1000_configure_rx(adapter);
2617 adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2621 * e1000e_power_up_phy - restore link in case the phy was powered down
2622 * @adapter: address of board private structure
2624 * The phy may be powered down to save power and turn off link when the
2625 * driver is unloaded and wake on lan is not enabled (among others)
2626 * *** this routine MUST be followed by a call to e1000e_reset ***
2628 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2630 u16 mii_reg = 0;
2632 /* Just clear the power down bit to wake the phy back up */
2633 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2635 * According to the manual, the phy will retain its
2636 * settings across a power-down/up cycle
2638 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2639 mii_reg &= ~MII_CR_POWER_DOWN;
2640 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2643 adapter->hw.mac.ops.setup_link(&adapter->hw);
2647 * e1000_power_down_phy - Power down the PHY
2649 * Power down the PHY so no link is implied when interface is down
2650 * The PHY cannot be powered down is management or WoL is active
2652 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2654 struct e1000_hw *hw = &adapter->hw;
2655 u16 mii_reg;
2657 /* WoL is enabled */
2658 if (adapter->wol)
2659 return;
2661 /* non-copper PHY? */
2662 if (adapter->hw.phy.media_type != e1000_media_type_copper)
2663 return;
2665 /* reset is blocked because of a SoL/IDER session */
2666 if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2667 return;
2669 /* manageability (AMT) is enabled */
2670 if (er32(MANC) & E1000_MANC_SMBUS_EN)
2671 return;
2673 /* power down the PHY */
2674 e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2675 mii_reg |= MII_CR_POWER_DOWN;
2676 e1e_wphy(hw, PHY_CONTROL, mii_reg);
2677 mdelay(1);
2681 * e1000e_reset - bring the hardware into a known good state
2683 * This function boots the hardware and enables some settings that
2684 * require a configuration cycle of the hardware - those cannot be
2685 * set/changed during runtime. After reset the device needs to be
2686 * properly configured for Rx, Tx etc.
2688 void e1000e_reset(struct e1000_adapter *adapter)
2690 struct e1000_mac_info *mac = &adapter->hw.mac;
2691 struct e1000_fc_info *fc = &adapter->hw.fc;
2692 struct e1000_hw *hw = &adapter->hw;
2693 u32 tx_space, min_tx_space, min_rx_space;
2694 u32 pba = adapter->pba;
2695 u16 hwm;
2697 /* reset Packet Buffer Allocation to default */
2698 ew32(PBA, pba);
2700 if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2702 * To maintain wire speed transmits, the Tx FIFO should be
2703 * large enough to accommodate two full transmit packets,
2704 * rounded up to the next 1KB and expressed in KB. Likewise,
2705 * the Rx FIFO should be large enough to accommodate at least
2706 * one full receive packet and is similarly rounded up and
2707 * expressed in KB.
2709 pba = er32(PBA);
2710 /* upper 16 bits has Tx packet buffer allocation size in KB */
2711 tx_space = pba >> 16;
2712 /* lower 16 bits has Rx packet buffer allocation size in KB */
2713 pba &= 0xffff;
2715 * the Tx fifo also stores 16 bytes of information about the tx
2716 * but don't include ethernet FCS because hardware appends it
2718 min_tx_space = (adapter->max_frame_size +
2719 sizeof(struct e1000_tx_desc) -
2720 ETH_FCS_LEN) * 2;
2721 min_tx_space = ALIGN(min_tx_space, 1024);
2722 min_tx_space >>= 10;
2723 /* software strips receive CRC, so leave room for it */
2724 min_rx_space = adapter->max_frame_size;
2725 min_rx_space = ALIGN(min_rx_space, 1024);
2726 min_rx_space >>= 10;
2729 * If current Tx allocation is less than the min Tx FIFO size,
2730 * and the min Tx FIFO size is less than the current Rx FIFO
2731 * allocation, take space away from current Rx allocation
2733 if ((tx_space < min_tx_space) &&
2734 ((min_tx_space - tx_space) < pba)) {
2735 pba -= min_tx_space - tx_space;
2738 * if short on Rx space, Rx wins and must trump tx
2739 * adjustment or use Early Receive if available
2741 if ((pba < min_rx_space) &&
2742 (!(adapter->flags & FLAG_HAS_ERT)))
2743 /* ERT enabled in e1000_configure_rx */
2744 pba = min_rx_space;
2747 ew32(PBA, pba);
2752 * flow control settings
2754 * The high water mark must be low enough to fit two full frame
2755 * (or the size used for early receive) above it in the Rx FIFO.
2756 * Set it to the lower of:
2757 * - 90% of the Rx FIFO size, and
2758 * - the full Rx FIFO size minus the early receive size (for parts
2759 * with ERT support assuming ERT set to E1000_ERT_2048), or
2760 * - the full Rx FIFO size minus two full frames
2762 if ((adapter->flags & FLAG_HAS_ERT) &&
2763 (adapter->netdev->mtu > ETH_DATA_LEN))
2764 hwm = min(((pba << 10) * 9 / 10),
2765 ((pba << 10) - (E1000_ERT_2048 << 3)));
2766 else
2767 hwm = min(((pba << 10) * 9 / 10),
2768 ((pba << 10) - (2 * adapter->max_frame_size)));
2770 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
2771 fc->low_water = (fc->high_water - (2 * adapter->max_frame_size));
2772 fc->low_water &= E1000_FCRTL_RTL; /* 8-byte granularity */
2774 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2775 fc->pause_time = 0xFFFF;
2776 else
2777 fc->pause_time = E1000_FC_PAUSE_TIME;
2778 fc->send_xon = 1;
2779 fc->current_mode = fc->requested_mode;
2781 /* Allow time for pending master requests to run */
2782 mac->ops.reset_hw(hw);
2785 * For parts with AMT enabled, let the firmware know
2786 * that the network interface is in control
2788 if (adapter->flags & FLAG_HAS_AMT)
2789 e1000_get_hw_control(adapter);
2791 ew32(WUC, 0);
2793 if (mac->ops.init_hw(hw))
2794 e_err("Hardware Error\n");
2796 e1000_update_mng_vlan(adapter);
2798 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2799 ew32(VET, ETH_P_8021Q);
2801 e1000e_reset_adaptive(hw);
2802 e1000_get_phy_info(hw);
2804 if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
2805 !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2806 u16 phy_data = 0;
2808 * speed up time to link by disabling smart power down, ignore
2809 * the return value of this function because there is nothing
2810 * different we would do if it failed
2812 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2813 phy_data &= ~IGP02E1000_PM_SPD;
2814 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2818 int e1000e_up(struct e1000_adapter *adapter)
2820 struct e1000_hw *hw = &adapter->hw;
2822 /* hardware has been reset, we need to reload some things */
2823 e1000_configure(adapter);
2825 clear_bit(__E1000_DOWN, &adapter->state);
2827 napi_enable(&adapter->napi);
2828 if (adapter->msix_entries)
2829 e1000_configure_msix(adapter);
2830 e1000_irq_enable(adapter);
2832 netif_wake_queue(adapter->netdev);
2834 /* fire a link change interrupt to start the watchdog */
2835 ew32(ICS, E1000_ICS_LSC);
2836 return 0;
2839 void e1000e_down(struct e1000_adapter *adapter)
2841 struct net_device *netdev = adapter->netdev;
2842 struct e1000_hw *hw = &adapter->hw;
2843 u32 tctl, rctl;
2846 * signal that we're down so the interrupt handler does not
2847 * reschedule our watchdog timer
2849 set_bit(__E1000_DOWN, &adapter->state);
2851 /* disable receives in the hardware */
2852 rctl = er32(RCTL);
2853 ew32(RCTL, rctl & ~E1000_RCTL_EN);
2854 /* flush and sleep below */
2856 netif_stop_queue(netdev);
2858 /* disable transmits in the hardware */
2859 tctl = er32(TCTL);
2860 tctl &= ~E1000_TCTL_EN;
2861 ew32(TCTL, tctl);
2862 /* flush both disables and wait for them to finish */
2863 e1e_flush();
2864 msleep(10);
2866 napi_disable(&adapter->napi);
2867 e1000_irq_disable(adapter);
2869 del_timer_sync(&adapter->watchdog_timer);
2870 del_timer_sync(&adapter->phy_info_timer);
2872 netdev->tx_queue_len = adapter->tx_queue_len;
2873 netif_carrier_off(netdev);
2874 adapter->link_speed = 0;
2875 adapter->link_duplex = 0;
2877 if (!pci_channel_offline(adapter->pdev))
2878 e1000e_reset(adapter);
2879 e1000_clean_tx_ring(adapter);
2880 e1000_clean_rx_ring(adapter);
2883 * TODO: for power management, we could drop the link and
2884 * pci_disable_device here.
2888 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2890 might_sleep();
2891 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2892 msleep(1);
2893 e1000e_down(adapter);
2894 e1000e_up(adapter);
2895 clear_bit(__E1000_RESETTING, &adapter->state);
2899 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2900 * @adapter: board private structure to initialize
2902 * e1000_sw_init initializes the Adapter private data structure.
2903 * Fields are initialized based on PCI device information and
2904 * OS network device settings (MTU size).
2906 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2908 struct net_device *netdev = adapter->netdev;
2910 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2911 adapter->rx_ps_bsize0 = 128;
2912 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2913 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2915 e1000e_set_interrupt_capability(adapter);
2917 if (e1000_alloc_queues(adapter))
2918 return -ENOMEM;
2920 /* Explicitly disable IRQ since the NIC can be in any state. */
2921 e1000_irq_disable(adapter);
2923 set_bit(__E1000_DOWN, &adapter->state);
2924 return 0;
2928 * e1000_intr_msi_test - Interrupt Handler
2929 * @irq: interrupt number
2930 * @data: pointer to a network interface device structure
2932 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2934 struct net_device *netdev = data;
2935 struct e1000_adapter *adapter = netdev_priv(netdev);
2936 struct e1000_hw *hw = &adapter->hw;
2937 u32 icr = er32(ICR);
2939 e_dbg("%s: icr is %08X\n", netdev->name, icr);
2940 if (icr & E1000_ICR_RXSEQ) {
2941 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2942 wmb();
2945 return IRQ_HANDLED;
2949 * e1000_test_msi_interrupt - Returns 0 for successful test
2950 * @adapter: board private struct
2952 * code flow taken from tg3.c
2954 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2956 struct net_device *netdev = adapter->netdev;
2957 struct e1000_hw *hw = &adapter->hw;
2958 int err;
2960 /* poll_enable hasn't been called yet, so don't need disable */
2961 /* clear any pending events */
2962 er32(ICR);
2964 /* free the real vector and request a test handler */
2965 e1000_free_irq(adapter);
2966 e1000e_reset_interrupt_capability(adapter);
2968 /* Assume that the test fails, if it succeeds then the test
2969 * MSI irq handler will unset this flag */
2970 adapter->flags |= FLAG_MSI_TEST_FAILED;
2972 err = pci_enable_msi(adapter->pdev);
2973 if (err)
2974 goto msi_test_failed;
2976 err = request_irq(adapter->pdev->irq, &e1000_intr_msi_test, 0,
2977 netdev->name, netdev);
2978 if (err) {
2979 pci_disable_msi(adapter->pdev);
2980 goto msi_test_failed;
2983 wmb();
2985 e1000_irq_enable(adapter);
2987 /* fire an unusual interrupt on the test handler */
2988 ew32(ICS, E1000_ICS_RXSEQ);
2989 e1e_flush();
2990 msleep(50);
2992 e1000_irq_disable(adapter);
2994 rmb();
2996 if (adapter->flags & FLAG_MSI_TEST_FAILED) {
2997 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2998 err = -EIO;
2999 e_info("MSI interrupt test failed!\n");
3002 free_irq(adapter->pdev->irq, netdev);
3003 pci_disable_msi(adapter->pdev);
3005 if (err == -EIO)
3006 goto msi_test_failed;
3008 /* okay so the test worked, restore settings */
3009 e_dbg("%s: MSI interrupt test succeeded!\n", netdev->name);
3010 msi_test_failed:
3011 e1000e_set_interrupt_capability(adapter);
3012 e1000_request_irq(adapter);
3013 return err;
3017 * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3018 * @adapter: board private struct
3020 * code flow taken from tg3.c, called with e1000 interrupts disabled.
3022 static int e1000_test_msi(struct e1000_adapter *adapter)
3024 int err;
3025 u16 pci_cmd;
3027 if (!(adapter->flags & FLAG_MSI_ENABLED))
3028 return 0;
3030 /* disable SERR in case the MSI write causes a master abort */
3031 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3032 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3033 pci_cmd & ~PCI_COMMAND_SERR);
3035 err = e1000_test_msi_interrupt(adapter);
3037 /* restore previous setting of command word */
3038 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3040 /* success ! */
3041 if (!err)
3042 return 0;
3044 /* EIO means MSI test failed */
3045 if (err != -EIO)
3046 return err;
3048 /* back to INTx mode */
3049 e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3051 e1000_free_irq(adapter);
3053 err = e1000_request_irq(adapter);
3055 return err;
3059 * e1000_open - Called when a network interface is made active
3060 * @netdev: network interface device structure
3062 * Returns 0 on success, negative value on failure
3064 * The open entry point is called when a network interface is made
3065 * active by the system (IFF_UP). At this point all resources needed
3066 * for transmit and receive operations are allocated, the interrupt
3067 * handler is registered with the OS, the watchdog timer is started,
3068 * and the stack is notified that the interface is ready.
3070 static int e1000_open(struct net_device *netdev)
3072 struct e1000_adapter *adapter = netdev_priv(netdev);
3073 struct e1000_hw *hw = &adapter->hw;
3074 int err;
3076 /* disallow open during test */
3077 if (test_bit(__E1000_TESTING, &adapter->state))
3078 return -EBUSY;
3080 netif_carrier_off(netdev);
3082 /* allocate transmit descriptors */
3083 err = e1000e_setup_tx_resources(adapter);
3084 if (err)
3085 goto err_setup_tx;
3087 /* allocate receive descriptors */
3088 err = e1000e_setup_rx_resources(adapter);
3089 if (err)
3090 goto err_setup_rx;
3092 e1000e_power_up_phy(adapter);
3094 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3095 if ((adapter->hw.mng_cookie.status &
3096 E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3097 e1000_update_mng_vlan(adapter);
3100 * If AMT is enabled, let the firmware know that the network
3101 * interface is now open
3103 if (adapter->flags & FLAG_HAS_AMT)
3104 e1000_get_hw_control(adapter);
3107 * before we allocate an interrupt, we must be ready to handle it.
3108 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3109 * as soon as we call pci_request_irq, so we have to setup our
3110 * clean_rx handler before we do so.
3112 e1000_configure(adapter);
3114 err = e1000_request_irq(adapter);
3115 if (err)
3116 goto err_req_irq;
3119 * Work around PCIe errata with MSI interrupts causing some chipsets to
3120 * ignore e1000e MSI messages, which means we need to test our MSI
3121 * interrupt now
3123 if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3124 err = e1000_test_msi(adapter);
3125 if (err) {
3126 e_err("Interrupt allocation failed\n");
3127 goto err_req_irq;
3131 /* From here on the code is the same as e1000e_up() */
3132 clear_bit(__E1000_DOWN, &adapter->state);
3134 napi_enable(&adapter->napi);
3136 e1000_irq_enable(adapter);
3138 netif_start_queue(netdev);
3140 /* fire a link status change interrupt to start the watchdog */
3141 ew32(ICS, E1000_ICS_LSC);
3143 return 0;
3145 err_req_irq:
3146 e1000_release_hw_control(adapter);
3147 e1000_power_down_phy(adapter);
3148 e1000e_free_rx_resources(adapter);
3149 err_setup_rx:
3150 e1000e_free_tx_resources(adapter);
3151 err_setup_tx:
3152 e1000e_reset(adapter);
3154 return err;
3158 * e1000_close - Disables a network interface
3159 * @netdev: network interface device structure
3161 * Returns 0, this is not allowed to fail
3163 * The close entry point is called when an interface is de-activated
3164 * by the OS. The hardware is still under the drivers control, but
3165 * needs to be disabled. A global MAC reset is issued to stop the
3166 * hardware, and all transmit and receive resources are freed.
3168 static int e1000_close(struct net_device *netdev)
3170 struct e1000_adapter *adapter = netdev_priv(netdev);
3172 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3173 e1000e_down(adapter);
3174 e1000_power_down_phy(adapter);
3175 e1000_free_irq(adapter);
3177 e1000e_free_tx_resources(adapter);
3178 e1000e_free_rx_resources(adapter);
3181 * kill manageability vlan ID if supported, but not if a vlan with
3182 * the same ID is registered on the host OS (let 8021q kill it)
3184 if ((adapter->hw.mng_cookie.status &
3185 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3186 !(adapter->vlgrp &&
3187 vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3188 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3191 * If AMT is enabled, let the firmware know that the network
3192 * interface is now closed
3194 if (adapter->flags & FLAG_HAS_AMT)
3195 e1000_release_hw_control(adapter);
3197 return 0;
3200 * e1000_set_mac - Change the Ethernet Address of the NIC
3201 * @netdev: network interface device structure
3202 * @p: pointer to an address structure
3204 * Returns 0 on success, negative on failure
3206 static int e1000_set_mac(struct net_device *netdev, void *p)
3208 struct e1000_adapter *adapter = netdev_priv(netdev);
3209 struct sockaddr *addr = p;
3211 if (!is_valid_ether_addr(addr->sa_data))
3212 return -EADDRNOTAVAIL;
3214 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3215 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3217 e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3219 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3220 /* activate the work around */
3221 e1000e_set_laa_state_82571(&adapter->hw, 1);
3224 * Hold a copy of the LAA in RAR[14] This is done so that
3225 * between the time RAR[0] gets clobbered and the time it
3226 * gets fixed (in e1000_watchdog), the actual LAA is in one
3227 * of the RARs and no incoming packets directed to this port
3228 * are dropped. Eventually the LAA will be in RAR[0] and
3229 * RAR[14]
3231 e1000e_rar_set(&adapter->hw,
3232 adapter->hw.mac.addr,
3233 adapter->hw.mac.rar_entry_count - 1);
3236 return 0;
3240 * e1000e_update_phy_task - work thread to update phy
3241 * @work: pointer to our work struct
3243 * this worker thread exists because we must acquire a
3244 * semaphore to read the phy, which we could msleep while
3245 * waiting for it, and we can't msleep in a timer.
3247 static void e1000e_update_phy_task(struct work_struct *work)
3249 struct e1000_adapter *adapter = container_of(work,
3250 struct e1000_adapter, update_phy_task);
3251 e1000_get_phy_info(&adapter->hw);
3255 * Need to wait a few seconds after link up to get diagnostic information from
3256 * the phy
3258 static void e1000_update_phy_info(unsigned long data)
3260 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3261 schedule_work(&adapter->update_phy_task);
3265 * e1000e_update_stats - Update the board statistics counters
3266 * @adapter: board private structure
3268 void e1000e_update_stats(struct e1000_adapter *adapter)
3270 struct e1000_hw *hw = &adapter->hw;
3271 struct pci_dev *pdev = adapter->pdev;
3274 * Prevent stats update while adapter is being reset, or if the pci
3275 * connection is down.
3277 if (adapter->link_speed == 0)
3278 return;
3279 if (pci_channel_offline(pdev))
3280 return;
3282 adapter->stats.crcerrs += er32(CRCERRS);
3283 adapter->stats.gprc += er32(GPRC);
3284 adapter->stats.gorc += er32(GORCL);
3285 er32(GORCH); /* Clear gorc */
3286 adapter->stats.bprc += er32(BPRC);
3287 adapter->stats.mprc += er32(MPRC);
3288 adapter->stats.roc += er32(ROC);
3290 adapter->stats.mpc += er32(MPC);
3291 adapter->stats.scc += er32(SCC);
3292 adapter->stats.ecol += er32(ECOL);
3293 adapter->stats.mcc += er32(MCC);
3294 adapter->stats.latecol += er32(LATECOL);
3295 adapter->stats.dc += er32(DC);
3296 adapter->stats.xonrxc += er32(XONRXC);
3297 adapter->stats.xontxc += er32(XONTXC);
3298 adapter->stats.xoffrxc += er32(XOFFRXC);
3299 adapter->stats.xofftxc += er32(XOFFTXC);
3300 adapter->stats.gptc += er32(GPTC);
3301 adapter->stats.gotc += er32(GOTCL);
3302 er32(GOTCH); /* Clear gotc */
3303 adapter->stats.rnbc += er32(RNBC);
3304 adapter->stats.ruc += er32(RUC);
3306 adapter->stats.mptc += er32(MPTC);
3307 adapter->stats.bptc += er32(BPTC);
3309 /* used for adaptive IFS */
3311 hw->mac.tx_packet_delta = er32(TPT);
3312 adapter->stats.tpt += hw->mac.tx_packet_delta;
3313 hw->mac.collision_delta = er32(COLC);
3314 adapter->stats.colc += hw->mac.collision_delta;
3316 adapter->stats.algnerrc += er32(ALGNERRC);
3317 adapter->stats.rxerrc += er32(RXERRC);
3318 if ((hw->mac.type != e1000_82574) && (hw->mac.type != e1000_82583))
3319 adapter->stats.tncrs += er32(TNCRS);
3320 adapter->stats.cexterr += er32(CEXTERR);
3321 adapter->stats.tsctc += er32(TSCTC);
3322 adapter->stats.tsctfc += er32(TSCTFC);
3324 /* Fill out the OS statistics structure */
3325 adapter->net_stats.multicast = adapter->stats.mprc;
3326 adapter->net_stats.collisions = adapter->stats.colc;
3328 /* Rx Errors */
3331 * RLEC on some newer hardware can be incorrect so build
3332 * our own version based on RUC and ROC
3334 adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3335 adapter->stats.crcerrs + adapter->stats.algnerrc +
3336 adapter->stats.ruc + adapter->stats.roc +
3337 adapter->stats.cexterr;
3338 adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3339 adapter->stats.roc;
3340 adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3341 adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3342 adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3344 /* Tx Errors */
3345 adapter->net_stats.tx_errors = adapter->stats.ecol +
3346 adapter->stats.latecol;
3347 adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3348 adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3349 adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3351 /* Tx Dropped needs to be maintained elsewhere */
3353 /* Management Stats */
3354 adapter->stats.mgptc += er32(MGTPTC);
3355 adapter->stats.mgprc += er32(MGTPRC);
3356 adapter->stats.mgpdc += er32(MGTPDC);
3360 * e1000_phy_read_status - Update the PHY register status snapshot
3361 * @adapter: board private structure
3363 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3365 struct e1000_hw *hw = &adapter->hw;
3366 struct e1000_phy_regs *phy = &adapter->phy_regs;
3367 int ret_val;
3369 if ((er32(STATUS) & E1000_STATUS_LU) &&
3370 (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3371 ret_val = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3372 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3373 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3374 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3375 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3376 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3377 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3378 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3379 if (ret_val)
3380 e_warn("Error reading PHY register\n");
3381 } else {
3383 * Do not read PHY registers if link is not up
3384 * Set values to typical power-on defaults
3386 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3387 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3388 BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3389 BMSR_ERCAP);
3390 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3391 ADVERTISE_ALL | ADVERTISE_CSMA);
3392 phy->lpa = 0;
3393 phy->expansion = EXPANSION_ENABLENPAGE;
3394 phy->ctrl1000 = ADVERTISE_1000FULL;
3395 phy->stat1000 = 0;
3396 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3400 static void e1000_print_link_info(struct e1000_adapter *adapter)
3402 struct e1000_hw *hw = &adapter->hw;
3403 u32 ctrl = er32(CTRL);
3405 /* Link status message must follow this format for user tools */
3406 printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3407 "Flow Control: %s\n",
3408 adapter->netdev->name,
3409 adapter->link_speed,
3410 (adapter->link_duplex == FULL_DUPLEX) ?
3411 "Full Duplex" : "Half Duplex",
3412 ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3413 "RX/TX" :
3414 ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3415 ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3418 bool e1000_has_link(struct e1000_adapter *adapter)
3420 struct e1000_hw *hw = &adapter->hw;
3421 bool link_active = 0;
3422 s32 ret_val = 0;
3425 * get_link_status is set on LSC (link status) interrupt or
3426 * Rx sequence error interrupt. get_link_status will stay
3427 * false until the check_for_link establishes link
3428 * for copper adapters ONLY
3430 switch (hw->phy.media_type) {
3431 case e1000_media_type_copper:
3432 if (hw->mac.get_link_status) {
3433 ret_val = hw->mac.ops.check_for_link(hw);
3434 link_active = !hw->mac.get_link_status;
3435 } else {
3436 link_active = 1;
3438 break;
3439 case e1000_media_type_fiber:
3440 ret_val = hw->mac.ops.check_for_link(hw);
3441 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3442 break;
3443 case e1000_media_type_internal_serdes:
3444 ret_val = hw->mac.ops.check_for_link(hw);
3445 link_active = adapter->hw.mac.serdes_has_link;
3446 break;
3447 default:
3448 case e1000_media_type_unknown:
3449 break;
3452 if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3453 (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3454 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3455 e_info("Gigabit has been disabled, downgrading speed\n");
3458 return link_active;
3461 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3463 /* make sure the receive unit is started */
3464 if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3465 (adapter->flags & FLAG_RX_RESTART_NOW)) {
3466 struct e1000_hw *hw = &adapter->hw;
3467 u32 rctl = er32(RCTL);
3468 ew32(RCTL, rctl | E1000_RCTL_EN);
3469 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3474 * e1000_watchdog - Timer Call-back
3475 * @data: pointer to adapter cast into an unsigned long
3477 static void e1000_watchdog(unsigned long data)
3479 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3481 /* Do the rest outside of interrupt context */
3482 schedule_work(&adapter->watchdog_task);
3484 /* TODO: make this use queue_delayed_work() */
3487 static void e1000_watchdog_task(struct work_struct *work)
3489 struct e1000_adapter *adapter = container_of(work,
3490 struct e1000_adapter, watchdog_task);
3491 struct net_device *netdev = adapter->netdev;
3492 struct e1000_mac_info *mac = &adapter->hw.mac;
3493 struct e1000_phy_info *phy = &adapter->hw.phy;
3494 struct e1000_ring *tx_ring = adapter->tx_ring;
3495 struct e1000_hw *hw = &adapter->hw;
3496 u32 link, tctl;
3497 int tx_pending = 0;
3499 link = e1000_has_link(adapter);
3500 if ((netif_carrier_ok(netdev)) && link) {
3501 e1000e_enable_receives(adapter);
3502 goto link_up;
3505 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3506 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3507 e1000_update_mng_vlan(adapter);
3509 if (link) {
3510 if (!netif_carrier_ok(netdev)) {
3511 bool txb2b = 1;
3512 /* update snapshot of PHY registers on LSC */
3513 e1000_phy_read_status(adapter);
3514 mac->ops.get_link_up_info(&adapter->hw,
3515 &adapter->link_speed,
3516 &adapter->link_duplex);
3517 e1000_print_link_info(adapter);
3519 * On supported PHYs, check for duplex mismatch only
3520 * if link has autonegotiated at 10/100 half
3522 if ((hw->phy.type == e1000_phy_igp_3 ||
3523 hw->phy.type == e1000_phy_bm) &&
3524 (hw->mac.autoneg == true) &&
3525 (adapter->link_speed == SPEED_10 ||
3526 adapter->link_speed == SPEED_100) &&
3527 (adapter->link_duplex == HALF_DUPLEX)) {
3528 u16 autoneg_exp;
3530 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3532 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3533 e_info("Autonegotiated half duplex but"
3534 " link partner cannot autoneg. "
3535 " Try forcing full duplex if "
3536 "link gets many collisions.\n");
3540 * tweak tx_queue_len according to speed/duplex
3541 * and adjust the timeout factor
3543 netdev->tx_queue_len = adapter->tx_queue_len;
3544 adapter->tx_timeout_factor = 1;
3545 switch (adapter->link_speed) {
3546 case SPEED_10:
3547 txb2b = 0;
3548 netdev->tx_queue_len = 10;
3549 adapter->tx_timeout_factor = 16;
3550 break;
3551 case SPEED_100:
3552 txb2b = 0;
3553 netdev->tx_queue_len = 100;
3554 /* maybe add some timeout factor ? */
3555 break;
3559 * workaround: re-program speed mode bit after
3560 * link-up event
3562 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3563 !txb2b) {
3564 u32 tarc0;
3565 tarc0 = er32(TARC(0));
3566 tarc0 &= ~SPEED_MODE_BIT;
3567 ew32(TARC(0), tarc0);
3571 * disable TSO for pcie and 10/100 speeds, to avoid
3572 * some hardware issues
3574 if (!(adapter->flags & FLAG_TSO_FORCE)) {
3575 switch (adapter->link_speed) {
3576 case SPEED_10:
3577 case SPEED_100:
3578 e_info("10/100 speed: disabling TSO\n");
3579 netdev->features &= ~NETIF_F_TSO;
3580 netdev->features &= ~NETIF_F_TSO6;
3581 break;
3582 case SPEED_1000:
3583 netdev->features |= NETIF_F_TSO;
3584 netdev->features |= NETIF_F_TSO6;
3585 break;
3586 default:
3587 /* oops */
3588 break;
3593 * enable transmits in the hardware, need to do this
3594 * after setting TARC(0)
3596 tctl = er32(TCTL);
3597 tctl |= E1000_TCTL_EN;
3598 ew32(TCTL, tctl);
3601 * Perform any post-link-up configuration before
3602 * reporting link up.
3604 if (phy->ops.cfg_on_link_up)
3605 phy->ops.cfg_on_link_up(hw);
3607 netif_carrier_on(netdev);
3609 if (!test_bit(__E1000_DOWN, &adapter->state))
3610 mod_timer(&adapter->phy_info_timer,
3611 round_jiffies(jiffies + 2 * HZ));
3613 } else {
3614 if (netif_carrier_ok(netdev)) {
3615 adapter->link_speed = 0;
3616 adapter->link_duplex = 0;
3617 /* Link status message must follow this format */
3618 printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
3619 adapter->netdev->name);
3620 netif_carrier_off(netdev);
3621 if (!test_bit(__E1000_DOWN, &adapter->state))
3622 mod_timer(&adapter->phy_info_timer,
3623 round_jiffies(jiffies + 2 * HZ));
3625 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3626 schedule_work(&adapter->reset_task);
3630 link_up:
3631 e1000e_update_stats(adapter);
3633 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3634 adapter->tpt_old = adapter->stats.tpt;
3635 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3636 adapter->colc_old = adapter->stats.colc;
3638 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3639 adapter->gorc_old = adapter->stats.gorc;
3640 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3641 adapter->gotc_old = adapter->stats.gotc;
3643 e1000e_update_adaptive(&adapter->hw);
3645 if (!netif_carrier_ok(netdev)) {
3646 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3647 tx_ring->count);
3648 if (tx_pending) {
3650 * We've lost link, so the controller stops DMA,
3651 * but we've got queued Tx work that's never going
3652 * to get done, so reset controller to flush Tx.
3653 * (Do the reset outside of interrupt context).
3655 adapter->tx_timeout_count++;
3656 schedule_work(&adapter->reset_task);
3657 /* return immediately since reset is imminent */
3658 return;
3662 /* Cause software interrupt to ensure Rx ring is cleaned */
3663 if (adapter->msix_entries)
3664 ew32(ICS, adapter->rx_ring->ims_val);
3665 else
3666 ew32(ICS, E1000_ICS_RXDMT0);
3668 /* Force detection of hung controller every watchdog period */
3669 adapter->detect_tx_hung = 1;
3672 * With 82571 controllers, LAA may be overwritten due to controller
3673 * reset from the other port. Set the appropriate LAA in RAR[0]
3675 if (e1000e_get_laa_state_82571(hw))
3676 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3678 /* Reset the timer */
3679 if (!test_bit(__E1000_DOWN, &adapter->state))
3680 mod_timer(&adapter->watchdog_timer,
3681 round_jiffies(jiffies + 2 * HZ));
3684 #define E1000_TX_FLAGS_CSUM 0x00000001
3685 #define E1000_TX_FLAGS_VLAN 0x00000002
3686 #define E1000_TX_FLAGS_TSO 0x00000004
3687 #define E1000_TX_FLAGS_IPV4 0x00000008
3688 #define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
3689 #define E1000_TX_FLAGS_VLAN_SHIFT 16
3691 static int e1000_tso(struct e1000_adapter *adapter,
3692 struct sk_buff *skb)
3694 struct e1000_ring *tx_ring = adapter->tx_ring;
3695 struct e1000_context_desc *context_desc;
3696 struct e1000_buffer *buffer_info;
3697 unsigned int i;
3698 u32 cmd_length = 0;
3699 u16 ipcse = 0, tucse, mss;
3700 u8 ipcss, ipcso, tucss, tucso, hdr_len;
3701 int err;
3703 if (skb_is_gso(skb)) {
3704 if (skb_header_cloned(skb)) {
3705 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3706 if (err)
3707 return err;
3710 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3711 mss = skb_shinfo(skb)->gso_size;
3712 if (skb->protocol == htons(ETH_P_IP)) {
3713 struct iphdr *iph = ip_hdr(skb);
3714 iph->tot_len = 0;
3715 iph->check = 0;
3716 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3717 iph->daddr, 0,
3718 IPPROTO_TCP,
3720 cmd_length = E1000_TXD_CMD_IP;
3721 ipcse = skb_transport_offset(skb) - 1;
3722 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3723 ipv6_hdr(skb)->payload_len = 0;
3724 tcp_hdr(skb)->check =
3725 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3726 &ipv6_hdr(skb)->daddr,
3727 0, IPPROTO_TCP, 0);
3728 ipcse = 0;
3730 ipcss = skb_network_offset(skb);
3731 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3732 tucss = skb_transport_offset(skb);
3733 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3734 tucse = 0;
3736 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3737 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3739 i = tx_ring->next_to_use;
3740 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3741 buffer_info = &tx_ring->buffer_info[i];
3743 context_desc->lower_setup.ip_fields.ipcss = ipcss;
3744 context_desc->lower_setup.ip_fields.ipcso = ipcso;
3745 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
3746 context_desc->upper_setup.tcp_fields.tucss = tucss;
3747 context_desc->upper_setup.tcp_fields.tucso = tucso;
3748 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3749 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
3750 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3751 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3753 buffer_info->time_stamp = jiffies;
3754 buffer_info->next_to_watch = i;
3756 i++;
3757 if (i == tx_ring->count)
3758 i = 0;
3759 tx_ring->next_to_use = i;
3761 return 1;
3764 return 0;
3767 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3769 struct e1000_ring *tx_ring = adapter->tx_ring;
3770 struct e1000_context_desc *context_desc;
3771 struct e1000_buffer *buffer_info;
3772 unsigned int i;
3773 u8 css;
3774 u32 cmd_len = E1000_TXD_CMD_DEXT;
3775 __be16 protocol;
3777 if (skb->ip_summed != CHECKSUM_PARTIAL)
3778 return 0;
3780 if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
3781 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
3782 else
3783 protocol = skb->protocol;
3785 switch (protocol) {
3786 case cpu_to_be16(ETH_P_IP):
3787 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3788 cmd_len |= E1000_TXD_CMD_TCP;
3789 break;
3790 case cpu_to_be16(ETH_P_IPV6):
3791 /* XXX not handling all IPV6 headers */
3792 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3793 cmd_len |= E1000_TXD_CMD_TCP;
3794 break;
3795 default:
3796 if (unlikely(net_ratelimit()))
3797 e_warn("checksum_partial proto=%x!\n",
3798 be16_to_cpu(protocol));
3799 break;
3802 css = skb_transport_offset(skb);
3804 i = tx_ring->next_to_use;
3805 buffer_info = &tx_ring->buffer_info[i];
3806 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3808 context_desc->lower_setup.ip_config = 0;
3809 context_desc->upper_setup.tcp_fields.tucss = css;
3810 context_desc->upper_setup.tcp_fields.tucso =
3811 css + skb->csum_offset;
3812 context_desc->upper_setup.tcp_fields.tucse = 0;
3813 context_desc->tcp_seg_setup.data = 0;
3814 context_desc->cmd_and_length = cpu_to_le32(cmd_len);
3816 buffer_info->time_stamp = jiffies;
3817 buffer_info->next_to_watch = i;
3819 i++;
3820 if (i == tx_ring->count)
3821 i = 0;
3822 tx_ring->next_to_use = i;
3824 return 1;
3827 #define E1000_MAX_PER_TXD 8192
3828 #define E1000_MAX_TXD_PWR 12
3830 static int e1000_tx_map(struct e1000_adapter *adapter,
3831 struct sk_buff *skb, unsigned int first,
3832 unsigned int max_per_txd, unsigned int nr_frags,
3833 unsigned int mss)
3835 struct e1000_ring *tx_ring = adapter->tx_ring;
3836 struct e1000_buffer *buffer_info;
3837 unsigned int len = skb_headlen(skb);
3838 unsigned int offset, size, count = 0, i;
3839 unsigned int f;
3840 dma_addr_t *map;
3842 i = tx_ring->next_to_use;
3844 if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
3845 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3846 adapter->tx_dma_failed++;
3847 return 0;
3850 map = skb_shinfo(skb)->dma_maps;
3851 offset = 0;
3853 while (len) {
3854 buffer_info = &tx_ring->buffer_info[i];
3855 size = min(len, max_per_txd);
3857 buffer_info->length = size;
3858 buffer_info->time_stamp = jiffies;
3859 buffer_info->next_to_watch = i;
3860 buffer_info->dma = map[0] + offset;
3861 count++;
3863 len -= size;
3864 offset += size;
3866 if (len) {
3867 i++;
3868 if (i == tx_ring->count)
3869 i = 0;
3873 for (f = 0; f < nr_frags; f++) {
3874 struct skb_frag_struct *frag;
3876 frag = &skb_shinfo(skb)->frags[f];
3877 len = frag->size;
3878 offset = 0;
3880 while (len) {
3881 i++;
3882 if (i == tx_ring->count)
3883 i = 0;
3885 buffer_info = &tx_ring->buffer_info[i];
3886 size = min(len, max_per_txd);
3888 buffer_info->length = size;
3889 buffer_info->time_stamp = jiffies;
3890 buffer_info->next_to_watch = i;
3891 buffer_info->dma = map[f + 1] + offset;
3893 len -= size;
3894 offset += size;
3895 count++;
3899 tx_ring->buffer_info[i].skb = skb;
3900 tx_ring->buffer_info[first].next_to_watch = i;
3902 return count;
3905 static void e1000_tx_queue(struct e1000_adapter *adapter,
3906 int tx_flags, int count)
3908 struct e1000_ring *tx_ring = adapter->tx_ring;
3909 struct e1000_tx_desc *tx_desc = NULL;
3910 struct e1000_buffer *buffer_info;
3911 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3912 unsigned int i;
3914 if (tx_flags & E1000_TX_FLAGS_TSO) {
3915 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3916 E1000_TXD_CMD_TSE;
3917 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3919 if (tx_flags & E1000_TX_FLAGS_IPV4)
3920 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3923 if (tx_flags & E1000_TX_FLAGS_CSUM) {
3924 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3925 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3928 if (tx_flags & E1000_TX_FLAGS_VLAN) {
3929 txd_lower |= E1000_TXD_CMD_VLE;
3930 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3933 i = tx_ring->next_to_use;
3935 while (count--) {
3936 buffer_info = &tx_ring->buffer_info[i];
3937 tx_desc = E1000_TX_DESC(*tx_ring, i);
3938 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3939 tx_desc->lower.data =
3940 cpu_to_le32(txd_lower | buffer_info->length);
3941 tx_desc->upper.data = cpu_to_le32(txd_upper);
3943 i++;
3944 if (i == tx_ring->count)
3945 i = 0;
3948 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3951 * Force memory writes to complete before letting h/w
3952 * know there are new descriptors to fetch. (Only
3953 * applicable for weak-ordered memory model archs,
3954 * such as IA-64).
3956 wmb();
3958 tx_ring->next_to_use = i;
3959 writel(i, adapter->hw.hw_addr + tx_ring->tail);
3961 * we need this if more than one processor can write to our tail
3962 * at a time, it synchronizes IO on IA64/Altix systems
3964 mmiowb();
3967 #define MINIMUM_DHCP_PACKET_SIZE 282
3968 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3969 struct sk_buff *skb)
3971 struct e1000_hw *hw = &adapter->hw;
3972 u16 length, offset;
3974 if (vlan_tx_tag_present(skb)) {
3975 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3976 && (adapter->hw.mng_cookie.status &
3977 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3978 return 0;
3981 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3982 return 0;
3984 if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3985 return 0;
3988 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3989 struct udphdr *udp;
3991 if (ip->protocol != IPPROTO_UDP)
3992 return 0;
3994 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3995 if (ntohs(udp->dest) != 67)
3996 return 0;
3998 offset = (u8 *)udp + 8 - skb->data;
3999 length = skb->len - offset;
4000 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4003 return 0;
4006 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4008 struct e1000_adapter *adapter = netdev_priv(netdev);
4010 netif_stop_queue(netdev);
4012 * Herbert's original patch had:
4013 * smp_mb__after_netif_stop_queue();
4014 * but since that doesn't exist yet, just open code it.
4016 smp_mb();
4019 * We need to check again in a case another CPU has just
4020 * made room available.
4022 if (e1000_desc_unused(adapter->tx_ring) < size)
4023 return -EBUSY;
4025 /* A reprieve! */
4026 netif_start_queue(netdev);
4027 ++adapter->restart_queue;
4028 return 0;
4031 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4033 struct e1000_adapter *adapter = netdev_priv(netdev);
4035 if (e1000_desc_unused(adapter->tx_ring) >= size)
4036 return 0;
4037 return __e1000_maybe_stop_tx(netdev, size);
4040 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4041 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4043 struct e1000_adapter *adapter = netdev_priv(netdev);
4044 struct e1000_ring *tx_ring = adapter->tx_ring;
4045 unsigned int first;
4046 unsigned int max_per_txd = E1000_MAX_PER_TXD;
4047 unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4048 unsigned int tx_flags = 0;
4049 unsigned int len = skb->len - skb->data_len;
4050 unsigned int nr_frags;
4051 unsigned int mss;
4052 int count = 0;
4053 int tso;
4054 unsigned int f;
4056 if (test_bit(__E1000_DOWN, &adapter->state)) {
4057 dev_kfree_skb_any(skb);
4058 return NETDEV_TX_OK;
4061 if (skb->len <= 0) {
4062 dev_kfree_skb_any(skb);
4063 return NETDEV_TX_OK;
4066 mss = skb_shinfo(skb)->gso_size;
4068 * The controller does a simple calculation to
4069 * make sure there is enough room in the FIFO before
4070 * initiating the DMA for each buffer. The calc is:
4071 * 4 = ceil(buffer len/mss). To make sure we don't
4072 * overrun the FIFO, adjust the max buffer len if mss
4073 * drops.
4075 if (mss) {
4076 u8 hdr_len;
4077 max_per_txd = min(mss << 2, max_per_txd);
4078 max_txd_pwr = fls(max_per_txd) - 1;
4081 * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4082 * points to just header, pull a few bytes of payload from
4083 * frags into skb->data
4085 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4087 * we do this workaround for ES2LAN, but it is un-necessary,
4088 * avoiding it could save a lot of cycles
4090 if (skb->data_len && (hdr_len == len)) {
4091 unsigned int pull_size;
4093 pull_size = min((unsigned int)4, skb->data_len);
4094 if (!__pskb_pull_tail(skb, pull_size)) {
4095 e_err("__pskb_pull_tail failed.\n");
4096 dev_kfree_skb_any(skb);
4097 return NETDEV_TX_OK;
4099 len = skb->len - skb->data_len;
4103 /* reserve a descriptor for the offload context */
4104 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4105 count++;
4106 count++;
4108 count += TXD_USE_COUNT(len, max_txd_pwr);
4110 nr_frags = skb_shinfo(skb)->nr_frags;
4111 for (f = 0; f < nr_frags; f++)
4112 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4113 max_txd_pwr);
4115 if (adapter->hw.mac.tx_pkt_filtering)
4116 e1000_transfer_dhcp_info(adapter, skb);
4119 * need: count + 2 desc gap to keep tail from touching
4120 * head, otherwise try next time
4122 if (e1000_maybe_stop_tx(netdev, count + 2))
4123 return NETDEV_TX_BUSY;
4125 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4126 tx_flags |= E1000_TX_FLAGS_VLAN;
4127 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4130 first = tx_ring->next_to_use;
4132 tso = e1000_tso(adapter, skb);
4133 if (tso < 0) {
4134 dev_kfree_skb_any(skb);
4135 return NETDEV_TX_OK;
4138 if (tso)
4139 tx_flags |= E1000_TX_FLAGS_TSO;
4140 else if (e1000_tx_csum(adapter, skb))
4141 tx_flags |= E1000_TX_FLAGS_CSUM;
4144 * Old method was to assume IPv4 packet by default if TSO was enabled.
4145 * 82571 hardware supports TSO capabilities for IPv6 as well...
4146 * no longer assume, we must.
4148 if (skb->protocol == htons(ETH_P_IP))
4149 tx_flags |= E1000_TX_FLAGS_IPV4;
4151 /* if count is 0 then mapping error has occured */
4152 count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4153 if (count) {
4154 e1000_tx_queue(adapter, tx_flags, count);
4155 /* Make sure there is space in the ring for the next send. */
4156 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4158 } else {
4159 dev_kfree_skb_any(skb);
4160 tx_ring->buffer_info[first].time_stamp = 0;
4161 tx_ring->next_to_use = first;
4164 return NETDEV_TX_OK;
4168 * e1000_tx_timeout - Respond to a Tx Hang
4169 * @netdev: network interface device structure
4171 static void e1000_tx_timeout(struct net_device *netdev)
4173 struct e1000_adapter *adapter = netdev_priv(netdev);
4175 /* Do the reset outside of interrupt context */
4176 adapter->tx_timeout_count++;
4177 schedule_work(&adapter->reset_task);
4180 static void e1000_reset_task(struct work_struct *work)
4182 struct e1000_adapter *adapter;
4183 adapter = container_of(work, struct e1000_adapter, reset_task);
4185 e1000e_reinit_locked(adapter);
4189 * e1000_get_stats - Get System Network Statistics
4190 * @netdev: network interface device structure
4192 * Returns the address of the device statistics structure.
4193 * The statistics are actually updated from the timer callback.
4195 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4197 struct e1000_adapter *adapter = netdev_priv(netdev);
4199 /* only return the current stats */
4200 return &adapter->net_stats;
4204 * e1000_change_mtu - Change the Maximum Transfer Unit
4205 * @netdev: network interface device structure
4206 * @new_mtu: new value for maximum frame size
4208 * Returns 0 on success, negative on failure
4210 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4212 struct e1000_adapter *adapter = netdev_priv(netdev);
4213 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4215 if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4216 (max_frame > MAX_JUMBO_FRAME_SIZE)) {
4217 e_err("Invalid MTU setting\n");
4218 return -EINVAL;
4221 /* Jumbo frame size limits */
4222 if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
4223 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4224 e_err("Jumbo Frames not supported.\n");
4225 return -EINVAL;
4227 if (adapter->hw.phy.type == e1000_phy_ife) {
4228 e_err("Jumbo Frames not supported.\n");
4229 return -EINVAL;
4233 #define MAX_STD_JUMBO_FRAME_SIZE 9234
4234 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
4235 e_err("MTU > 9216 not supported.\n");
4236 return -EINVAL;
4239 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4240 msleep(1);
4241 /* e1000e_down has a dependency on max_frame_size */
4242 adapter->max_frame_size = max_frame;
4243 if (netif_running(netdev))
4244 e1000e_down(adapter);
4247 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4248 * means we reserve 2 more, this pushes us to allocate from the next
4249 * larger slab size.
4250 * i.e. RXBUFFER_2048 --> size-4096 slab
4251 * However with the new *_jumbo_rx* routines, jumbo receives will use
4252 * fragmented skbs
4255 if (max_frame <= 256)
4256 adapter->rx_buffer_len = 256;
4257 else if (max_frame <= 512)
4258 adapter->rx_buffer_len = 512;
4259 else if (max_frame <= 1024)
4260 adapter->rx_buffer_len = 1024;
4261 else if (max_frame <= 2048)
4262 adapter->rx_buffer_len = 2048;
4263 else
4264 adapter->rx_buffer_len = 4096;
4266 /* adjust allocation if LPE protects us, and we aren't using SBP */
4267 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4268 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4269 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4270 + ETH_FCS_LEN;
4272 e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4273 netdev->mtu = new_mtu;
4275 if (netif_running(netdev))
4276 e1000e_up(adapter);
4277 else
4278 e1000e_reset(adapter);
4280 clear_bit(__E1000_RESETTING, &adapter->state);
4282 return 0;
4285 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4286 int cmd)
4288 struct e1000_adapter *adapter = netdev_priv(netdev);
4289 struct mii_ioctl_data *data = if_mii(ifr);
4291 if (adapter->hw.phy.media_type != e1000_media_type_copper)
4292 return -EOPNOTSUPP;
4294 switch (cmd) {
4295 case SIOCGMIIPHY:
4296 data->phy_id = adapter->hw.phy.addr;
4297 break;
4298 case SIOCGMIIREG:
4299 if (!capable(CAP_NET_ADMIN))
4300 return -EPERM;
4301 switch (data->reg_num & 0x1F) {
4302 case MII_BMCR:
4303 data->val_out = adapter->phy_regs.bmcr;
4304 break;
4305 case MII_BMSR:
4306 data->val_out = adapter->phy_regs.bmsr;
4307 break;
4308 case MII_PHYSID1:
4309 data->val_out = (adapter->hw.phy.id >> 16);
4310 break;
4311 case MII_PHYSID2:
4312 data->val_out = (adapter->hw.phy.id & 0xFFFF);
4313 break;
4314 case MII_ADVERTISE:
4315 data->val_out = adapter->phy_regs.advertise;
4316 break;
4317 case MII_LPA:
4318 data->val_out = adapter->phy_regs.lpa;
4319 break;
4320 case MII_EXPANSION:
4321 data->val_out = adapter->phy_regs.expansion;
4322 break;
4323 case MII_CTRL1000:
4324 data->val_out = adapter->phy_regs.ctrl1000;
4325 break;
4326 case MII_STAT1000:
4327 data->val_out = adapter->phy_regs.stat1000;
4328 break;
4329 case MII_ESTATUS:
4330 data->val_out = adapter->phy_regs.estatus;
4331 break;
4332 default:
4333 return -EIO;
4335 break;
4336 case SIOCSMIIREG:
4337 default:
4338 return -EOPNOTSUPP;
4340 return 0;
4343 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4345 switch (cmd) {
4346 case SIOCGMIIPHY:
4347 case SIOCGMIIREG:
4348 case SIOCSMIIREG:
4349 return e1000_mii_ioctl(netdev, ifr, cmd);
4350 default:
4351 return -EOPNOTSUPP;
4355 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4357 struct net_device *netdev = pci_get_drvdata(pdev);
4358 struct e1000_adapter *adapter = netdev_priv(netdev);
4359 struct e1000_hw *hw = &adapter->hw;
4360 u32 ctrl, ctrl_ext, rctl, status;
4361 u32 wufc = adapter->wol;
4362 int retval = 0;
4364 netif_device_detach(netdev);
4366 if (netif_running(netdev)) {
4367 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4368 e1000e_down(adapter);
4369 e1000_free_irq(adapter);
4371 e1000e_reset_interrupt_capability(adapter);
4373 retval = pci_save_state(pdev);
4374 if (retval)
4375 return retval;
4377 status = er32(STATUS);
4378 if (status & E1000_STATUS_LU)
4379 wufc &= ~E1000_WUFC_LNKC;
4381 if (wufc) {
4382 e1000_setup_rctl(adapter);
4383 e1000_set_multi(netdev);
4385 /* turn on all-multi mode if wake on multicast is enabled */
4386 if (wufc & E1000_WUFC_MC) {
4387 rctl = er32(RCTL);
4388 rctl |= E1000_RCTL_MPE;
4389 ew32(RCTL, rctl);
4392 ctrl = er32(CTRL);
4393 /* advertise wake from D3Cold */
4394 #define E1000_CTRL_ADVD3WUC 0x00100000
4395 /* phy power management enable */
4396 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4397 ctrl |= E1000_CTRL_ADVD3WUC |
4398 E1000_CTRL_EN_PHY_PWR_MGMT;
4399 ew32(CTRL, ctrl);
4401 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4402 adapter->hw.phy.media_type ==
4403 e1000_media_type_internal_serdes) {
4404 /* keep the laser running in D3 */
4405 ctrl_ext = er32(CTRL_EXT);
4406 ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4407 ew32(CTRL_EXT, ctrl_ext);
4410 if (adapter->flags & FLAG_IS_ICH)
4411 e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4413 /* Allow time for pending master requests to run */
4414 e1000e_disable_pcie_master(&adapter->hw);
4416 ew32(WUC, E1000_WUC_PME_EN);
4417 ew32(WUFC, wufc);
4418 } else {
4419 ew32(WUC, 0);
4420 ew32(WUFC, 0);
4423 *enable_wake = !!wufc;
4425 /* make sure adapter isn't asleep if manageability is enabled */
4426 if (adapter->flags & FLAG_MNG_PT_ENABLED)
4427 *enable_wake = true;
4429 if (adapter->hw.phy.type == e1000_phy_igp_3)
4430 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4433 * Release control of h/w to f/w. If f/w is AMT enabled, this
4434 * would have already happened in close and is redundant.
4436 e1000_release_hw_control(adapter);
4438 pci_disable_device(pdev);
4440 return 0;
4443 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
4445 if (sleep && wake) {
4446 pci_prepare_to_sleep(pdev);
4447 return;
4450 pci_wake_from_d3(pdev, wake);
4451 pci_set_power_state(pdev, PCI_D3hot);
4454 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
4455 bool wake)
4457 struct net_device *netdev = pci_get_drvdata(pdev);
4458 struct e1000_adapter *adapter = netdev_priv(netdev);
4461 * The pci-e switch on some quad port adapters will report a
4462 * correctable error when the MAC transitions from D0 to D3. To
4463 * prevent this we need to mask off the correctable errors on the
4464 * downstream port of the pci-e switch.
4466 if (adapter->flags & FLAG_IS_QUAD_PORT) {
4467 struct pci_dev *us_dev = pdev->bus->self;
4468 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
4469 u16 devctl;
4471 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
4472 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
4473 (devctl & ~PCI_EXP_DEVCTL_CERE));
4475 e1000_power_off(pdev, sleep, wake);
4477 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
4478 } else {
4479 e1000_power_off(pdev, sleep, wake);
4483 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4485 int pos;
4486 u16 val;
4489 * 82573 workaround - disable L1 ASPM on mobile chipsets
4491 * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4492 * resulting in lost data or garbage information on the pci-e link
4493 * level. This could result in (false) bad EEPROM checksum errors,
4494 * long ping times (up to 2s) or even a system freeze/hang.
4496 * Unfortunately this feature saves about 1W power consumption when
4497 * active.
4499 pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4500 pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4501 if (val & 0x2) {
4502 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4503 val &= ~0x2;
4504 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4508 #ifdef CONFIG_PM
4509 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4511 int retval;
4512 bool wake;
4514 retval = __e1000_shutdown(pdev, &wake);
4515 if (!retval)
4516 e1000_complete_shutdown(pdev, true, wake);
4518 return retval;
4521 static int e1000_resume(struct pci_dev *pdev)
4523 struct net_device *netdev = pci_get_drvdata(pdev);
4524 struct e1000_adapter *adapter = netdev_priv(netdev);
4525 struct e1000_hw *hw = &adapter->hw;
4526 u32 err;
4528 pci_set_power_state(pdev, PCI_D0);
4529 pci_restore_state(pdev);
4530 e1000e_disable_l1aspm(pdev);
4532 err = pci_enable_device_mem(pdev);
4533 if (err) {
4534 dev_err(&pdev->dev,
4535 "Cannot enable PCI device from suspend\n");
4536 return err;
4539 /* AER (Advanced Error Reporting) hooks */
4540 err = pci_enable_pcie_error_reporting(pdev);
4541 if (err) {
4542 dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
4543 "0x%x\n", err);
4544 /* non-fatal, continue */
4547 pci_set_master(pdev);
4549 pci_enable_wake(pdev, PCI_D3hot, 0);
4550 pci_enable_wake(pdev, PCI_D3cold, 0);
4552 e1000e_set_interrupt_capability(adapter);
4553 if (netif_running(netdev)) {
4554 err = e1000_request_irq(adapter);
4555 if (err)
4556 return err;
4559 e1000e_power_up_phy(adapter);
4560 e1000e_reset(adapter);
4561 ew32(WUS, ~0);
4563 e1000_init_manageability(adapter);
4565 if (netif_running(netdev))
4566 e1000e_up(adapter);
4568 netif_device_attach(netdev);
4571 * If the controller has AMT, do not set DRV_LOAD until the interface
4572 * is up. For all other cases, let the f/w know that the h/w is now
4573 * under the control of the driver.
4575 if (!(adapter->flags & FLAG_HAS_AMT))
4576 e1000_get_hw_control(adapter);
4578 return 0;
4580 #endif
4582 static void e1000_shutdown(struct pci_dev *pdev)
4584 bool wake = false;
4586 __e1000_shutdown(pdev, &wake);
4588 if (system_state == SYSTEM_POWER_OFF)
4589 e1000_complete_shutdown(pdev, false, wake);
4592 #ifdef CONFIG_NET_POLL_CONTROLLER
4594 * Polling 'interrupt' - used by things like netconsole to send skbs
4595 * without having to re-enable interrupts. It's not called while
4596 * the interrupt routine is executing.
4598 static void e1000_netpoll(struct net_device *netdev)
4600 struct e1000_adapter *adapter = netdev_priv(netdev);
4602 disable_irq(adapter->pdev->irq);
4603 e1000_intr(adapter->pdev->irq, netdev);
4605 enable_irq(adapter->pdev->irq);
4607 #endif
4610 * e1000_io_error_detected - called when PCI error is detected
4611 * @pdev: Pointer to PCI device
4612 * @state: The current pci connection state
4614 * This function is called after a PCI bus error affecting
4615 * this device has been detected.
4617 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4618 pci_channel_state_t state)
4620 struct net_device *netdev = pci_get_drvdata(pdev);
4621 struct e1000_adapter *adapter = netdev_priv(netdev);
4623 netif_device_detach(netdev);
4625 if (netif_running(netdev))
4626 e1000e_down(adapter);
4627 pci_disable_device(pdev);
4629 /* Request a slot slot reset. */
4630 return PCI_ERS_RESULT_NEED_RESET;
4634 * e1000_io_slot_reset - called after the pci bus has been reset.
4635 * @pdev: Pointer to PCI device
4637 * Restart the card from scratch, as if from a cold-boot. Implementation
4638 * resembles the first-half of the e1000_resume routine.
4640 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4642 struct net_device *netdev = pci_get_drvdata(pdev);
4643 struct e1000_adapter *adapter = netdev_priv(netdev);
4644 struct e1000_hw *hw = &adapter->hw;
4645 int err;
4646 pci_ers_result_t result;
4648 e1000e_disable_l1aspm(pdev);
4649 err = pci_enable_device_mem(pdev);
4650 if (err) {
4651 dev_err(&pdev->dev,
4652 "Cannot re-enable PCI device after reset.\n");
4653 result = PCI_ERS_RESULT_DISCONNECT;
4654 } else {
4655 pci_set_master(pdev);
4656 pci_restore_state(pdev);
4658 pci_enable_wake(pdev, PCI_D3hot, 0);
4659 pci_enable_wake(pdev, PCI_D3cold, 0);
4661 e1000e_reset(adapter);
4662 ew32(WUS, ~0);
4663 result = PCI_ERS_RESULT_RECOVERED;
4666 pci_cleanup_aer_uncorrect_error_status(pdev);
4668 return result;
4672 * e1000_io_resume - called when traffic can start flowing again.
4673 * @pdev: Pointer to PCI device
4675 * This callback is called when the error recovery driver tells us that
4676 * its OK to resume normal operation. Implementation resembles the
4677 * second-half of the e1000_resume routine.
4679 static void e1000_io_resume(struct pci_dev *pdev)
4681 struct net_device *netdev = pci_get_drvdata(pdev);
4682 struct e1000_adapter *adapter = netdev_priv(netdev);
4684 e1000_init_manageability(adapter);
4686 if (netif_running(netdev)) {
4687 if (e1000e_up(adapter)) {
4688 dev_err(&pdev->dev,
4689 "can't bring device back up after reset\n");
4690 return;
4694 netif_device_attach(netdev);
4697 * If the controller has AMT, do not set DRV_LOAD until the interface
4698 * is up. For all other cases, let the f/w know that the h/w is now
4699 * under the control of the driver.
4701 if (!(adapter->flags & FLAG_HAS_AMT))
4702 e1000_get_hw_control(adapter);
4706 static void e1000_print_device_info(struct e1000_adapter *adapter)
4708 struct e1000_hw *hw = &adapter->hw;
4709 struct net_device *netdev = adapter->netdev;
4710 u32 pba_num;
4712 /* print bus type/speed/width info */
4713 e_info("(PCI Express:2.5GB/s:%s) %pM\n",
4714 /* bus width */
4715 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4716 "Width x1"),
4717 /* MAC address */
4718 netdev->dev_addr);
4719 e_info("Intel(R) PRO/%s Network Connection\n",
4720 (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4721 e1000e_read_pba_num(hw, &pba_num);
4722 e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4723 hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4726 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4728 struct e1000_hw *hw = &adapter->hw;
4729 int ret_val;
4730 u16 buf = 0;
4732 if (hw->mac.type != e1000_82573)
4733 return;
4735 ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4736 if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
4737 /* Deep Smart Power Down (DSPD) */
4738 dev_warn(&adapter->pdev->dev,
4739 "Warning: detected DSPD enabled in EEPROM\n");
4742 ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4743 if (!ret_val && (le16_to_cpu(buf) & (3 << 2))) {
4744 /* ASPM enable */
4745 dev_warn(&adapter->pdev->dev,
4746 "Warning: detected ASPM enabled in EEPROM\n");
4750 static const struct net_device_ops e1000e_netdev_ops = {
4751 .ndo_open = e1000_open,
4752 .ndo_stop = e1000_close,
4753 .ndo_start_xmit = e1000_xmit_frame,
4754 .ndo_get_stats = e1000_get_stats,
4755 .ndo_set_multicast_list = e1000_set_multi,
4756 .ndo_set_mac_address = e1000_set_mac,
4757 .ndo_change_mtu = e1000_change_mtu,
4758 .ndo_do_ioctl = e1000_ioctl,
4759 .ndo_tx_timeout = e1000_tx_timeout,
4760 .ndo_validate_addr = eth_validate_addr,
4762 .ndo_vlan_rx_register = e1000_vlan_rx_register,
4763 .ndo_vlan_rx_add_vid = e1000_vlan_rx_add_vid,
4764 .ndo_vlan_rx_kill_vid = e1000_vlan_rx_kill_vid,
4765 #ifdef CONFIG_NET_POLL_CONTROLLER
4766 .ndo_poll_controller = e1000_netpoll,
4767 #endif
4771 * e1000_probe - Device Initialization Routine
4772 * @pdev: PCI device information struct
4773 * @ent: entry in e1000_pci_tbl
4775 * Returns 0 on success, negative on failure
4777 * e1000_probe initializes an adapter identified by a pci_dev structure.
4778 * The OS initialization, configuring of the adapter private structure,
4779 * and a hardware reset occur.
4781 static int __devinit e1000_probe(struct pci_dev *pdev,
4782 const struct pci_device_id *ent)
4784 struct net_device *netdev;
4785 struct e1000_adapter *adapter;
4786 struct e1000_hw *hw;
4787 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4788 resource_size_t mmio_start, mmio_len;
4789 resource_size_t flash_start, flash_len;
4791 static int cards_found;
4792 int i, err, pci_using_dac;
4793 u16 eeprom_data = 0;
4794 u16 eeprom_apme_mask = E1000_EEPROM_APME;
4796 e1000e_disable_l1aspm(pdev);
4798 err = pci_enable_device_mem(pdev);
4799 if (err)
4800 return err;
4802 pci_using_dac = 0;
4803 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
4804 if (!err) {
4805 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
4806 if (!err)
4807 pci_using_dac = 1;
4808 } else {
4809 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
4810 if (err) {
4811 err = pci_set_consistent_dma_mask(pdev,
4812 DMA_BIT_MASK(32));
4813 if (err) {
4814 dev_err(&pdev->dev, "No usable DMA "
4815 "configuration, aborting\n");
4816 goto err_dma;
4821 err = pci_request_selected_regions_exclusive(pdev,
4822 pci_select_bars(pdev, IORESOURCE_MEM),
4823 e1000e_driver_name);
4824 if (err)
4825 goto err_pci_reg;
4827 pci_set_master(pdev);
4828 /* PCI config space info */
4829 err = pci_save_state(pdev);
4830 if (err)
4831 goto err_alloc_etherdev;
4833 err = -ENOMEM;
4834 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4835 if (!netdev)
4836 goto err_alloc_etherdev;
4838 SET_NETDEV_DEV(netdev, &pdev->dev);
4840 pci_set_drvdata(pdev, netdev);
4841 adapter = netdev_priv(netdev);
4842 hw = &adapter->hw;
4843 adapter->netdev = netdev;
4844 adapter->pdev = pdev;
4845 adapter->ei = ei;
4846 adapter->pba = ei->pba;
4847 adapter->flags = ei->flags;
4848 adapter->flags2 = ei->flags2;
4849 adapter->hw.adapter = adapter;
4850 adapter->hw.mac.type = ei->mac;
4851 adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4853 mmio_start = pci_resource_start(pdev, 0);
4854 mmio_len = pci_resource_len(pdev, 0);
4856 err = -EIO;
4857 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4858 if (!adapter->hw.hw_addr)
4859 goto err_ioremap;
4861 if ((adapter->flags & FLAG_HAS_FLASH) &&
4862 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
4863 flash_start = pci_resource_start(pdev, 1);
4864 flash_len = pci_resource_len(pdev, 1);
4865 adapter->hw.flash_address = ioremap(flash_start, flash_len);
4866 if (!adapter->hw.flash_address)
4867 goto err_flashmap;
4870 /* construct the net_device struct */
4871 netdev->netdev_ops = &e1000e_netdev_ops;
4872 e1000e_set_ethtool_ops(netdev);
4873 netdev->watchdog_timeo = 5 * HZ;
4874 netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
4875 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
4877 netdev->mem_start = mmio_start;
4878 netdev->mem_end = mmio_start + mmio_len;
4880 adapter->bd_number = cards_found++;
4882 e1000e_check_options(adapter);
4884 /* setup adapter struct */
4885 err = e1000_sw_init(adapter);
4886 if (err)
4887 goto err_sw_init;
4889 err = -EIO;
4891 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4892 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
4893 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4895 err = ei->get_variants(adapter);
4896 if (err)
4897 goto err_hw_init;
4899 if ((adapter->flags & FLAG_IS_ICH) &&
4900 (adapter->flags & FLAG_READ_ONLY_NVM))
4901 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
4903 hw->mac.ops.get_bus_info(&adapter->hw);
4905 adapter->hw.phy.autoneg_wait_to_complete = 0;
4907 /* Copper options */
4908 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
4909 adapter->hw.phy.mdix = AUTO_ALL_MODES;
4910 adapter->hw.phy.disable_polarity_correction = 0;
4911 adapter->hw.phy.ms_type = e1000_ms_hw_default;
4914 if (e1000_check_reset_block(&adapter->hw))
4915 e_info("PHY reset is blocked due to SOL/IDER session.\n");
4917 netdev->features = NETIF_F_SG |
4918 NETIF_F_HW_CSUM |
4919 NETIF_F_HW_VLAN_TX |
4920 NETIF_F_HW_VLAN_RX;
4922 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
4923 netdev->features |= NETIF_F_HW_VLAN_FILTER;
4925 netdev->features |= NETIF_F_TSO;
4926 netdev->features |= NETIF_F_TSO6;
4928 netdev->vlan_features |= NETIF_F_TSO;
4929 netdev->vlan_features |= NETIF_F_TSO6;
4930 netdev->vlan_features |= NETIF_F_HW_CSUM;
4931 netdev->vlan_features |= NETIF_F_SG;
4933 if (pci_using_dac)
4934 netdev->features |= NETIF_F_HIGHDMA;
4936 if (e1000e_enable_mng_pass_thru(&adapter->hw))
4937 adapter->flags |= FLAG_MNG_PT_ENABLED;
4940 * before reading the NVM, reset the controller to
4941 * put the device in a known good starting state
4943 adapter->hw.mac.ops.reset_hw(&adapter->hw);
4946 * systems with ASPM and others may see the checksum fail on the first
4947 * attempt. Let's give it a few tries
4949 for (i = 0;; i++) {
4950 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
4951 break;
4952 if (i == 2) {
4953 e_err("The NVM Checksum Is Not Valid\n");
4954 err = -EIO;
4955 goto err_eeprom;
4959 e1000_eeprom_checks(adapter);
4961 /* copy the MAC address out of the NVM */
4962 if (e1000e_read_mac_addr(&adapter->hw))
4963 e_err("NVM Read Error while reading MAC address\n");
4965 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
4966 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
4968 if (!is_valid_ether_addr(netdev->perm_addr)) {
4969 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
4970 err = -EIO;
4971 goto err_eeprom;
4974 init_timer(&adapter->watchdog_timer);
4975 adapter->watchdog_timer.function = &e1000_watchdog;
4976 adapter->watchdog_timer.data = (unsigned long) adapter;
4978 init_timer(&adapter->phy_info_timer);
4979 adapter->phy_info_timer.function = &e1000_update_phy_info;
4980 adapter->phy_info_timer.data = (unsigned long) adapter;
4982 INIT_WORK(&adapter->reset_task, e1000_reset_task);
4983 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
4984 INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
4985 INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
4987 /* Initialize link parameters. User can change them with ethtool */
4988 adapter->hw.mac.autoneg = 1;
4989 adapter->fc_autoneg = 1;
4990 adapter->hw.fc.requested_mode = e1000_fc_default;
4991 adapter->hw.fc.current_mode = e1000_fc_default;
4992 adapter->hw.phy.autoneg_advertised = 0x2f;
4994 /* ring size defaults */
4995 adapter->rx_ring->count = 256;
4996 adapter->tx_ring->count = 256;
4999 * Initial Wake on LAN setting - If APM wake is enabled in
5000 * the EEPROM, enable the ACPI Magic Packet filter
5002 if (adapter->flags & FLAG_APME_IN_WUC) {
5003 /* APME bit in EEPROM is mapped to WUC.APME */
5004 eeprom_data = er32(WUC);
5005 eeprom_apme_mask = E1000_WUC_APME;
5006 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5007 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5008 (adapter->hw.bus.func == 1))
5009 e1000_read_nvm(&adapter->hw,
5010 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5011 else
5012 e1000_read_nvm(&adapter->hw,
5013 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5016 /* fetch WoL from EEPROM */
5017 if (eeprom_data & eeprom_apme_mask)
5018 adapter->eeprom_wol |= E1000_WUFC_MAG;
5021 * now that we have the eeprom settings, apply the special cases
5022 * where the eeprom may be wrong or the board simply won't support
5023 * wake on lan on a particular port
5025 if (!(adapter->flags & FLAG_HAS_WOL))
5026 adapter->eeprom_wol = 0;
5028 /* initialize the wol settings based on the eeprom settings */
5029 adapter->wol = adapter->eeprom_wol;
5030 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5032 /* save off EEPROM version number */
5033 e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5035 /* reset the hardware with the new settings */
5036 e1000e_reset(adapter);
5039 * If the controller has AMT, do not set DRV_LOAD until the interface
5040 * is up. For all other cases, let the f/w know that the h/w is now
5041 * under the control of the driver.
5043 if (!(adapter->flags & FLAG_HAS_AMT))
5044 e1000_get_hw_control(adapter);
5046 strcpy(netdev->name, "eth%d");
5047 err = register_netdev(netdev);
5048 if (err)
5049 goto err_register;
5051 /* carrier off reporting is important to ethtool even BEFORE open */
5052 netif_carrier_off(netdev);
5054 e1000_print_device_info(adapter);
5056 return 0;
5058 err_register:
5059 if (!(adapter->flags & FLAG_HAS_AMT))
5060 e1000_release_hw_control(adapter);
5061 err_eeprom:
5062 if (!e1000_check_reset_block(&adapter->hw))
5063 e1000_phy_hw_reset(&adapter->hw);
5064 err_hw_init:
5066 kfree(adapter->tx_ring);
5067 kfree(adapter->rx_ring);
5068 err_sw_init:
5069 if (adapter->hw.flash_address)
5070 iounmap(adapter->hw.flash_address);
5071 e1000e_reset_interrupt_capability(adapter);
5072 err_flashmap:
5073 iounmap(adapter->hw.hw_addr);
5074 err_ioremap:
5075 free_netdev(netdev);
5076 err_alloc_etherdev:
5077 pci_release_selected_regions(pdev,
5078 pci_select_bars(pdev, IORESOURCE_MEM));
5079 err_pci_reg:
5080 err_dma:
5081 pci_disable_device(pdev);
5082 return err;
5086 * e1000_remove - Device Removal Routine
5087 * @pdev: PCI device information struct
5089 * e1000_remove is called by the PCI subsystem to alert the driver
5090 * that it should release a PCI device. The could be caused by a
5091 * Hot-Plug event, or because the driver is going to be removed from
5092 * memory.
5094 static void __devexit e1000_remove(struct pci_dev *pdev)
5096 struct net_device *netdev = pci_get_drvdata(pdev);
5097 struct e1000_adapter *adapter = netdev_priv(netdev);
5098 int err;
5101 * flush_scheduled work may reschedule our watchdog task, so
5102 * explicitly disable watchdog tasks from being rescheduled
5104 set_bit(__E1000_DOWN, &adapter->state);
5105 del_timer_sync(&adapter->watchdog_timer);
5106 del_timer_sync(&adapter->phy_info_timer);
5108 flush_scheduled_work();
5111 * Release control of h/w to f/w. If f/w is AMT enabled, this
5112 * would have already happened in close and is redundant.
5114 e1000_release_hw_control(adapter);
5116 unregister_netdev(netdev);
5118 if (!e1000_check_reset_block(&adapter->hw))
5119 e1000_phy_hw_reset(&adapter->hw);
5121 e1000e_reset_interrupt_capability(adapter);
5122 kfree(adapter->tx_ring);
5123 kfree(adapter->rx_ring);
5125 iounmap(adapter->hw.hw_addr);
5126 if (adapter->hw.flash_address)
5127 iounmap(adapter->hw.flash_address);
5128 pci_release_selected_regions(pdev,
5129 pci_select_bars(pdev, IORESOURCE_MEM));
5131 free_netdev(netdev);
5133 /* AER disable */
5134 err = pci_disable_pcie_error_reporting(pdev);
5135 if (err)
5136 dev_err(&pdev->dev,
5137 "pci_disable_pcie_error_reporting failed 0x%x\n", err);
5139 pci_disable_device(pdev);
5142 /* PCI Error Recovery (ERS) */
5143 static struct pci_error_handlers e1000_err_handler = {
5144 .error_detected = e1000_io_error_detected,
5145 .slot_reset = e1000_io_slot_reset,
5146 .resume = e1000_io_resume,
5149 static struct pci_device_id e1000_pci_tbl[] = {
5150 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5151 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5152 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5153 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5154 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5155 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5156 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5157 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5158 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5160 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5161 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5162 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5163 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5165 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5166 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5167 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5169 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5170 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
5171 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
5173 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5174 board_80003es2lan },
5175 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5176 board_80003es2lan },
5177 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5178 board_80003es2lan },
5179 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5180 board_80003es2lan },
5182 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5183 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5184 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5185 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5186 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5187 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5188 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5190 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5191 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5192 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5193 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5194 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5195 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5196 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5197 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5198 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5200 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5201 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5202 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5204 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5205 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5207 { } /* terminate list */
5209 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5211 /* PCI Device API Driver */
5212 static struct pci_driver e1000_driver = {
5213 .name = e1000e_driver_name,
5214 .id_table = e1000_pci_tbl,
5215 .probe = e1000_probe,
5216 .remove = __devexit_p(e1000_remove),
5217 #ifdef CONFIG_PM
5218 /* Power Management Hooks */
5219 .suspend = e1000_suspend,
5220 .resume = e1000_resume,
5221 #endif
5222 .shutdown = e1000_shutdown,
5223 .err_handler = &e1000_err_handler
5227 * e1000_init_module - Driver Registration Routine
5229 * e1000_init_module is the first routine called when the driver is
5230 * loaded. All it does is register with the PCI subsystem.
5232 static int __init e1000_init_module(void)
5234 int ret;
5235 printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
5236 e1000e_driver_name, e1000e_driver_version);
5237 printk(KERN_INFO "%s: Copyright (c) 1999-2008 Intel Corporation.\n",
5238 e1000e_driver_name);
5239 ret = pci_register_driver(&e1000_driver);
5240 pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name,
5241 PM_QOS_DEFAULT_VALUE);
5243 return ret;
5245 module_init(e1000_init_module);
5248 * e1000_exit_module - Driver Exit Cleanup Routine
5250 * e1000_exit_module is called just before the driver is removed
5251 * from memory.
5253 static void __exit e1000_exit_module(void)
5255 pci_unregister_driver(&e1000_driver);
5256 pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name);
5258 module_exit(e1000_exit_module);
5261 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5262 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5263 MODULE_LICENSE("GPL");
5264 MODULE_VERSION(DRV_VERSION);
5266 /* e1000_main.c */