1 /*******************************************************************************
3 Intel PRO/1000 Linux driver
4 Copyright(c) 1999 - 2009 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
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".
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 <linux/slab.h>
40 #include <net/checksum.h>
41 #include <net/ip6_checksum.h>
42 #include <linux/mii.h>
43 #include <linux/ethtool.h>
44 #include <linux/if_vlan.h>
45 #include <linux/cpu.h>
46 #include <linux/smp.h>
47 #include <linux/pm_qos_params.h>
48 #include <linux/aer.h>
52 #define DRV_VERSION "1.0.2-k2"
53 char e1000e_driver_name[] = "e1000e";
54 const char e1000e_driver_version[] = DRV_VERSION;
56 static const struct e1000_info *e1000_info_tbl[] = {
57 [board_82571] = &e1000_82571_info,
58 [board_82572] = &e1000_82572_info,
59 [board_82573] = &e1000_82573_info,
60 [board_82574] = &e1000_82574_info,
61 [board_82583] = &e1000_82583_info,
62 [board_80003es2lan] = &e1000_es2_info,
63 [board_ich8lan] = &e1000_ich8_info,
64 [board_ich9lan] = &e1000_ich9_info,
65 [board_ich10lan] = &e1000_ich10_info,
66 [board_pchlan] = &e1000_pch_info,
70 * e1000_desc_unused - calculate if we have unused descriptors
72 static int e1000_desc_unused(struct e1000_ring *ring)
74 if (ring->next_to_clean > ring->next_to_use)
75 return ring->next_to_clean - ring->next_to_use - 1;
77 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
81 * e1000_receive_skb - helper function to handle Rx indications
82 * @adapter: board private structure
83 * @status: descriptor status field as written by hardware
84 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
85 * @skb: pointer to sk_buff to be indicated to stack
87 static void e1000_receive_skb(struct e1000_adapter *adapter,
88 struct net_device *netdev,
90 u8 status, __le16 vlan)
92 skb->protocol = eth_type_trans(skb, netdev);
94 if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
95 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
96 le16_to_cpu(vlan), skb);
98 napi_gro_receive(&adapter->napi, skb);
102 * e1000_rx_checksum - Receive Checksum Offload for 82543
103 * @adapter: board private structure
104 * @status_err: receive descriptor status and error fields
105 * @csum: receive descriptor csum field
106 * @sk_buff: socket buffer with received data
108 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
109 u32 csum, struct sk_buff *skb)
111 u16 status = (u16)status_err;
112 u8 errors = (u8)(status_err >> 24);
113 skb->ip_summed = CHECKSUM_NONE;
115 /* Ignore Checksum bit is set */
116 if (status & E1000_RXD_STAT_IXSM)
118 /* TCP/UDP checksum error bit is set */
119 if (errors & E1000_RXD_ERR_TCPE) {
120 /* let the stack verify checksum errors */
121 adapter->hw_csum_err++;
125 /* TCP/UDP Checksum has not been calculated */
126 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
129 /* It must be a TCP or UDP packet with a valid checksum */
130 if (status & E1000_RXD_STAT_TCPCS) {
131 /* TCP checksum is good */
132 skb->ip_summed = CHECKSUM_UNNECESSARY;
135 * IP fragment with UDP payload
136 * Hardware complements the payload checksum, so we undo it
137 * and then put the value in host order for further stack use.
139 __sum16 sum = (__force __sum16)htons(csum);
140 skb->csum = csum_unfold(~sum);
141 skb->ip_summed = CHECKSUM_COMPLETE;
143 adapter->hw_csum_good++;
147 * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
148 * @adapter: address of board private structure
150 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
153 struct net_device *netdev = adapter->netdev;
154 struct pci_dev *pdev = adapter->pdev;
155 struct e1000_ring *rx_ring = adapter->rx_ring;
156 struct e1000_rx_desc *rx_desc;
157 struct e1000_buffer *buffer_info;
160 unsigned int bufsz = adapter->rx_buffer_len;
162 i = rx_ring->next_to_use;
163 buffer_info = &rx_ring->buffer_info[i];
165 while (cleaned_count--) {
166 skb = buffer_info->skb;
172 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
174 /* Better luck next round */
175 adapter->alloc_rx_buff_failed++;
179 buffer_info->skb = skb;
181 buffer_info->dma = pci_map_single(pdev, skb->data,
182 adapter->rx_buffer_len,
184 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
185 dev_err(&pdev->dev, "RX DMA map failed\n");
186 adapter->rx_dma_failed++;
190 rx_desc = E1000_RX_DESC(*rx_ring, i);
191 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
194 if (i == rx_ring->count)
196 buffer_info = &rx_ring->buffer_info[i];
199 if (rx_ring->next_to_use != i) {
200 rx_ring->next_to_use = i;
202 i = (rx_ring->count - 1);
205 * Force memory writes to complete before letting h/w
206 * know there are new descriptors to fetch. (Only
207 * applicable for weak-ordered memory model archs,
211 writel(i, adapter->hw.hw_addr + rx_ring->tail);
216 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
217 * @adapter: address of board private structure
219 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
222 struct net_device *netdev = adapter->netdev;
223 struct pci_dev *pdev = adapter->pdev;
224 union e1000_rx_desc_packet_split *rx_desc;
225 struct e1000_ring *rx_ring = adapter->rx_ring;
226 struct e1000_buffer *buffer_info;
227 struct e1000_ps_page *ps_page;
231 i = rx_ring->next_to_use;
232 buffer_info = &rx_ring->buffer_info[i];
234 while (cleaned_count--) {
235 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
237 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
238 ps_page = &buffer_info->ps_pages[j];
239 if (j >= adapter->rx_ps_pages) {
240 /* all unused desc entries get hw null ptr */
241 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
244 if (!ps_page->page) {
245 ps_page->page = alloc_page(GFP_ATOMIC);
246 if (!ps_page->page) {
247 adapter->alloc_rx_buff_failed++;
250 ps_page->dma = pci_map_page(pdev,
254 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
255 dev_err(&adapter->pdev->dev,
256 "RX DMA page map failed\n");
257 adapter->rx_dma_failed++;
262 * Refresh the desc even if buffer_addrs
263 * didn't change because each write-back
266 rx_desc->read.buffer_addr[j+1] =
267 cpu_to_le64(ps_page->dma);
270 skb = netdev_alloc_skb_ip_align(netdev,
271 adapter->rx_ps_bsize0);
274 adapter->alloc_rx_buff_failed++;
278 buffer_info->skb = skb;
279 buffer_info->dma = pci_map_single(pdev, skb->data,
280 adapter->rx_ps_bsize0,
282 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
283 dev_err(&pdev->dev, "RX DMA map failed\n");
284 adapter->rx_dma_failed++;
286 dev_kfree_skb_any(skb);
287 buffer_info->skb = NULL;
291 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
294 if (i == rx_ring->count)
296 buffer_info = &rx_ring->buffer_info[i];
300 if (rx_ring->next_to_use != i) {
301 rx_ring->next_to_use = i;
304 i = (rx_ring->count - 1);
307 * Force memory writes to complete before letting h/w
308 * know there are new descriptors to fetch. (Only
309 * applicable for weak-ordered memory model archs,
314 * Hardware increments by 16 bytes, but packet split
315 * descriptors are 32 bytes...so we increment tail
318 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
323 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
324 * @adapter: address of board private structure
325 * @cleaned_count: number of buffers to allocate this pass
328 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
331 struct net_device *netdev = adapter->netdev;
332 struct pci_dev *pdev = adapter->pdev;
333 struct e1000_rx_desc *rx_desc;
334 struct e1000_ring *rx_ring = adapter->rx_ring;
335 struct e1000_buffer *buffer_info;
338 unsigned int bufsz = 256 - 16 /* for skb_reserve */;
340 i = rx_ring->next_to_use;
341 buffer_info = &rx_ring->buffer_info[i];
343 while (cleaned_count--) {
344 skb = buffer_info->skb;
350 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
351 if (unlikely(!skb)) {
352 /* Better luck next round */
353 adapter->alloc_rx_buff_failed++;
357 buffer_info->skb = skb;
359 /* allocate a new page if necessary */
360 if (!buffer_info->page) {
361 buffer_info->page = alloc_page(GFP_ATOMIC);
362 if (unlikely(!buffer_info->page)) {
363 adapter->alloc_rx_buff_failed++;
368 if (!buffer_info->dma)
369 buffer_info->dma = pci_map_page(pdev,
370 buffer_info->page, 0,
374 rx_desc = E1000_RX_DESC(*rx_ring, i);
375 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
377 if (unlikely(++i == rx_ring->count))
379 buffer_info = &rx_ring->buffer_info[i];
382 if (likely(rx_ring->next_to_use != i)) {
383 rx_ring->next_to_use = i;
384 if (unlikely(i-- == 0))
385 i = (rx_ring->count - 1);
387 /* Force memory writes to complete before letting h/w
388 * know there are new descriptors to fetch. (Only
389 * applicable for weak-ordered memory model archs,
392 writel(i, adapter->hw.hw_addr + rx_ring->tail);
397 * e1000_clean_rx_irq - Send received data up the network stack; legacy
398 * @adapter: board private structure
400 * the return value indicates whether actual cleaning was done, there
401 * is no guarantee that everything was cleaned
403 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
404 int *work_done, int work_to_do)
406 struct net_device *netdev = adapter->netdev;
407 struct pci_dev *pdev = adapter->pdev;
408 struct e1000_hw *hw = &adapter->hw;
409 struct e1000_ring *rx_ring = adapter->rx_ring;
410 struct e1000_rx_desc *rx_desc, *next_rxd;
411 struct e1000_buffer *buffer_info, *next_buffer;
414 int cleaned_count = 0;
416 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
418 i = rx_ring->next_to_clean;
419 rx_desc = E1000_RX_DESC(*rx_ring, i);
420 buffer_info = &rx_ring->buffer_info[i];
422 while (rx_desc->status & E1000_RXD_STAT_DD) {
426 if (*work_done >= work_to_do)
430 status = rx_desc->status;
431 skb = buffer_info->skb;
432 buffer_info->skb = NULL;
434 prefetch(skb->data - NET_IP_ALIGN);
437 if (i == rx_ring->count)
439 next_rxd = E1000_RX_DESC(*rx_ring, i);
442 next_buffer = &rx_ring->buffer_info[i];
446 pci_unmap_single(pdev,
448 adapter->rx_buffer_len,
450 buffer_info->dma = 0;
452 length = le16_to_cpu(rx_desc->length);
455 * !EOP means multiple descriptors were used to store a single
456 * packet, if that's the case we need to toss it. In fact, we
457 * need to toss every packet with the EOP bit clear and the
458 * next frame that _does_ have the EOP bit set, as it is by
459 * definition only a frame fragment
461 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
462 adapter->flags2 |= FLAG2_IS_DISCARDING;
464 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
465 /* All receives must fit into a single buffer */
466 e_dbg("Receive packet consumed multiple buffers\n");
468 buffer_info->skb = skb;
469 if (status & E1000_RXD_STAT_EOP)
470 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
474 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
476 buffer_info->skb = skb;
480 /* adjust length to remove Ethernet CRC */
481 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
484 total_rx_bytes += length;
488 * code added for copybreak, this should improve
489 * performance for small packets with large amounts
490 * of reassembly being done in the stack
492 if (length < copybreak) {
493 struct sk_buff *new_skb =
494 netdev_alloc_skb_ip_align(netdev, length);
496 skb_copy_to_linear_data_offset(new_skb,
502 /* save the skb in buffer_info as good */
503 buffer_info->skb = skb;
506 /* else just continue with the old one */
508 /* end copybreak code */
509 skb_put(skb, length);
511 /* Receive Checksum Offload */
512 e1000_rx_checksum(adapter,
514 ((u32)(rx_desc->errors) << 24),
515 le16_to_cpu(rx_desc->csum), skb);
517 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
522 /* return some buffers to hardware, one at a time is too slow */
523 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
524 adapter->alloc_rx_buf(adapter, cleaned_count);
528 /* use prefetched values */
530 buffer_info = next_buffer;
532 rx_ring->next_to_clean = i;
534 cleaned_count = e1000_desc_unused(rx_ring);
536 adapter->alloc_rx_buf(adapter, cleaned_count);
538 adapter->total_rx_bytes += total_rx_bytes;
539 adapter->total_rx_packets += total_rx_packets;
540 netdev->stats.rx_bytes += total_rx_bytes;
541 netdev->stats.rx_packets += total_rx_packets;
545 static void e1000_put_txbuf(struct e1000_adapter *adapter,
546 struct e1000_buffer *buffer_info)
548 if (buffer_info->dma) {
549 if (buffer_info->mapped_as_page)
550 pci_unmap_page(adapter->pdev, buffer_info->dma,
551 buffer_info->length, PCI_DMA_TODEVICE);
553 pci_unmap_single(adapter->pdev, buffer_info->dma,
556 buffer_info->dma = 0;
558 if (buffer_info->skb) {
559 dev_kfree_skb_any(buffer_info->skb);
560 buffer_info->skb = NULL;
562 buffer_info->time_stamp = 0;
565 static void e1000_print_hw_hang(struct work_struct *work)
567 struct e1000_adapter *adapter = container_of(work,
568 struct e1000_adapter,
570 struct e1000_ring *tx_ring = adapter->tx_ring;
571 unsigned int i = tx_ring->next_to_clean;
572 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
573 struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
574 struct e1000_hw *hw = &adapter->hw;
575 u16 phy_status, phy_1000t_status, phy_ext_status;
578 e1e_rphy(hw, PHY_STATUS, &phy_status);
579 e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
580 e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
582 pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
584 /* detected Hardware unit hang */
585 e_err("Detected Hardware Unit Hang:\n"
588 " next_to_use <%x>\n"
589 " next_to_clean <%x>\n"
590 "buffer_info[next_to_clean]:\n"
591 " time_stamp <%lx>\n"
592 " next_to_watch <%x>\n"
594 " next_to_watch.status <%x>\n"
597 "PHY 1000BASE-T Status <%x>\n"
598 "PHY Extended Status <%x>\n"
600 readl(adapter->hw.hw_addr + tx_ring->head),
601 readl(adapter->hw.hw_addr + tx_ring->tail),
602 tx_ring->next_to_use,
603 tx_ring->next_to_clean,
604 tx_ring->buffer_info[eop].time_stamp,
607 eop_desc->upper.fields.status,
616 * e1000_clean_tx_irq - Reclaim resources after transmit completes
617 * @adapter: board private structure
619 * the return value indicates whether actual cleaning was done, there
620 * is no guarantee that everything was cleaned
622 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
624 struct net_device *netdev = adapter->netdev;
625 struct e1000_hw *hw = &adapter->hw;
626 struct e1000_ring *tx_ring = adapter->tx_ring;
627 struct e1000_tx_desc *tx_desc, *eop_desc;
628 struct e1000_buffer *buffer_info;
630 unsigned int count = 0;
631 unsigned int total_tx_bytes = 0, total_tx_packets = 0;
633 i = tx_ring->next_to_clean;
634 eop = tx_ring->buffer_info[i].next_to_watch;
635 eop_desc = E1000_TX_DESC(*tx_ring, eop);
637 while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
638 (count < tx_ring->count)) {
639 bool cleaned = false;
640 for (; !cleaned; count++) {
641 tx_desc = E1000_TX_DESC(*tx_ring, i);
642 buffer_info = &tx_ring->buffer_info[i];
643 cleaned = (i == eop);
646 struct sk_buff *skb = buffer_info->skb;
647 unsigned int segs, bytecount;
648 segs = skb_shinfo(skb)->gso_segs ?: 1;
649 /* multiply data chunks by size of headers */
650 bytecount = ((segs - 1) * skb_headlen(skb)) +
652 total_tx_packets += segs;
653 total_tx_bytes += bytecount;
656 e1000_put_txbuf(adapter, buffer_info);
657 tx_desc->upper.data = 0;
660 if (i == tx_ring->count)
664 if (i == tx_ring->next_to_use)
666 eop = tx_ring->buffer_info[i].next_to_watch;
667 eop_desc = E1000_TX_DESC(*tx_ring, eop);
670 tx_ring->next_to_clean = i;
672 #define TX_WAKE_THRESHOLD 32
673 if (count && netif_carrier_ok(netdev) &&
674 e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
675 /* Make sure that anybody stopping the queue after this
676 * sees the new next_to_clean.
680 if (netif_queue_stopped(netdev) &&
681 !(test_bit(__E1000_DOWN, &adapter->state))) {
682 netif_wake_queue(netdev);
683 ++adapter->restart_queue;
687 if (adapter->detect_tx_hung) {
689 * Detect a transmit hang in hardware, this serializes the
690 * check with the clearing of time_stamp and movement of i
692 adapter->detect_tx_hung = 0;
693 if (tx_ring->buffer_info[i].time_stamp &&
694 time_after(jiffies, tx_ring->buffer_info[i].time_stamp
695 + (adapter->tx_timeout_factor * HZ)) &&
696 !(er32(STATUS) & E1000_STATUS_TXOFF)) {
697 schedule_work(&adapter->print_hang_task);
698 netif_stop_queue(netdev);
701 adapter->total_tx_bytes += total_tx_bytes;
702 adapter->total_tx_packets += total_tx_packets;
703 netdev->stats.tx_bytes += total_tx_bytes;
704 netdev->stats.tx_packets += total_tx_packets;
705 return (count < tx_ring->count);
709 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
710 * @adapter: board private structure
712 * the return value indicates whether actual cleaning was done, there
713 * is no guarantee that everything was cleaned
715 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
716 int *work_done, int work_to_do)
718 struct e1000_hw *hw = &adapter->hw;
719 union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
720 struct net_device *netdev = adapter->netdev;
721 struct pci_dev *pdev = adapter->pdev;
722 struct e1000_ring *rx_ring = adapter->rx_ring;
723 struct e1000_buffer *buffer_info, *next_buffer;
724 struct e1000_ps_page *ps_page;
728 int cleaned_count = 0;
730 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
732 i = rx_ring->next_to_clean;
733 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
734 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
735 buffer_info = &rx_ring->buffer_info[i];
737 while (staterr & E1000_RXD_STAT_DD) {
738 if (*work_done >= work_to_do)
741 skb = buffer_info->skb;
743 /* in the packet split case this is header only */
744 prefetch(skb->data - NET_IP_ALIGN);
747 if (i == rx_ring->count)
749 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
752 next_buffer = &rx_ring->buffer_info[i];
756 pci_unmap_single(pdev, buffer_info->dma,
757 adapter->rx_ps_bsize0,
759 buffer_info->dma = 0;
761 /* see !EOP comment in other rx routine */
762 if (!(staterr & E1000_RXD_STAT_EOP))
763 adapter->flags2 |= FLAG2_IS_DISCARDING;
765 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
766 e_dbg("Packet Split buffers didn't pick up the full "
768 dev_kfree_skb_irq(skb);
769 if (staterr & E1000_RXD_STAT_EOP)
770 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
774 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
775 dev_kfree_skb_irq(skb);
779 length = le16_to_cpu(rx_desc->wb.middle.length0);
782 e_dbg("Last part of the packet spanning multiple "
784 dev_kfree_skb_irq(skb);
789 skb_put(skb, length);
793 * this looks ugly, but it seems compiler issues make it
794 * more efficient than reusing j
796 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
799 * page alloc/put takes too long and effects small packet
800 * throughput, so unsplit small packets and save the alloc/put
801 * only valid in softirq (napi) context to call kmap_*
803 if (l1 && (l1 <= copybreak) &&
804 ((length + l1) <= adapter->rx_ps_bsize0)) {
807 ps_page = &buffer_info->ps_pages[0];
810 * there is no documentation about how to call
811 * kmap_atomic, so we can't hold the mapping
814 pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
815 PAGE_SIZE, PCI_DMA_FROMDEVICE);
816 vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
817 memcpy(skb_tail_pointer(skb), vaddr, l1);
818 kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
819 pci_dma_sync_single_for_device(pdev, ps_page->dma,
820 PAGE_SIZE, PCI_DMA_FROMDEVICE);
823 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
831 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
832 length = le16_to_cpu(rx_desc->wb.upper.length[j]);
836 ps_page = &buffer_info->ps_pages[j];
837 pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
840 skb_fill_page_desc(skb, j, ps_page->page, 0, length);
841 ps_page->page = NULL;
843 skb->data_len += length;
844 skb->truesize += length;
847 /* strip the ethernet crc, problem is we're using pages now so
848 * this whole operation can get a little cpu intensive
850 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
851 pskb_trim(skb, skb->len - 4);
854 total_rx_bytes += skb->len;
857 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
858 rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
860 if (rx_desc->wb.upper.header_status &
861 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
862 adapter->rx_hdr_split++;
864 e1000_receive_skb(adapter, netdev, skb,
865 staterr, rx_desc->wb.middle.vlan);
868 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
869 buffer_info->skb = NULL;
871 /* return some buffers to hardware, one at a time is too slow */
872 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
873 adapter->alloc_rx_buf(adapter, cleaned_count);
877 /* use prefetched values */
879 buffer_info = next_buffer;
881 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
883 rx_ring->next_to_clean = i;
885 cleaned_count = e1000_desc_unused(rx_ring);
887 adapter->alloc_rx_buf(adapter, cleaned_count);
889 adapter->total_rx_bytes += total_rx_bytes;
890 adapter->total_rx_packets += total_rx_packets;
891 netdev->stats.rx_bytes += total_rx_bytes;
892 netdev->stats.rx_packets += total_rx_packets;
897 * e1000_consume_page - helper function
899 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
904 skb->data_len += length;
905 skb->truesize += length;
909 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
910 * @adapter: board private structure
912 * the return value indicates whether actual cleaning was done, there
913 * is no guarantee that everything was cleaned
916 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
917 int *work_done, int work_to_do)
919 struct net_device *netdev = adapter->netdev;
920 struct pci_dev *pdev = adapter->pdev;
921 struct e1000_ring *rx_ring = adapter->rx_ring;
922 struct e1000_rx_desc *rx_desc, *next_rxd;
923 struct e1000_buffer *buffer_info, *next_buffer;
926 int cleaned_count = 0;
927 bool cleaned = false;
928 unsigned int total_rx_bytes=0, total_rx_packets=0;
930 i = rx_ring->next_to_clean;
931 rx_desc = E1000_RX_DESC(*rx_ring, i);
932 buffer_info = &rx_ring->buffer_info[i];
934 while (rx_desc->status & E1000_RXD_STAT_DD) {
938 if (*work_done >= work_to_do)
942 status = rx_desc->status;
943 skb = buffer_info->skb;
944 buffer_info->skb = NULL;
947 if (i == rx_ring->count)
949 next_rxd = E1000_RX_DESC(*rx_ring, i);
952 next_buffer = &rx_ring->buffer_info[i];
956 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
958 buffer_info->dma = 0;
960 length = le16_to_cpu(rx_desc->length);
962 /* errors is only valid for DD + EOP descriptors */
963 if (unlikely((status & E1000_RXD_STAT_EOP) &&
964 (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
965 /* recycle both page and skb */
966 buffer_info->skb = skb;
967 /* an error means any chain goes out the window
969 if (rx_ring->rx_skb_top)
970 dev_kfree_skb(rx_ring->rx_skb_top);
971 rx_ring->rx_skb_top = NULL;
975 #define rxtop rx_ring->rx_skb_top
976 if (!(status & E1000_RXD_STAT_EOP)) {
977 /* this descriptor is only the beginning (or middle) */
979 /* this is the beginning of a chain */
981 skb_fill_page_desc(rxtop, 0, buffer_info->page,
984 /* this is the middle of a chain */
985 skb_fill_page_desc(rxtop,
986 skb_shinfo(rxtop)->nr_frags,
987 buffer_info->page, 0, length);
988 /* re-use the skb, only consumed the page */
989 buffer_info->skb = skb;
991 e1000_consume_page(buffer_info, rxtop, length);
995 /* end of the chain */
996 skb_fill_page_desc(rxtop,
997 skb_shinfo(rxtop)->nr_frags,
998 buffer_info->page, 0, length);
999 /* re-use the current skb, we only consumed the
1001 buffer_info->skb = skb;
1004 e1000_consume_page(buffer_info, skb, length);
1006 /* no chain, got EOP, this buf is the packet
1007 * copybreak to save the put_page/alloc_page */
1008 if (length <= copybreak &&
1009 skb_tailroom(skb) >= length) {
1011 vaddr = kmap_atomic(buffer_info->page,
1012 KM_SKB_DATA_SOFTIRQ);
1013 memcpy(skb_tail_pointer(skb), vaddr,
1015 kunmap_atomic(vaddr,
1016 KM_SKB_DATA_SOFTIRQ);
1017 /* re-use the page, so don't erase
1018 * buffer_info->page */
1019 skb_put(skb, length);
1021 skb_fill_page_desc(skb, 0,
1022 buffer_info->page, 0,
1024 e1000_consume_page(buffer_info, skb,
1030 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1031 e1000_rx_checksum(adapter,
1033 ((u32)(rx_desc->errors) << 24),
1034 le16_to_cpu(rx_desc->csum), skb);
1036 /* probably a little skewed due to removing CRC */
1037 total_rx_bytes += skb->len;
1040 /* eth type trans needs skb->data to point to something */
1041 if (!pskb_may_pull(skb, ETH_HLEN)) {
1042 e_err("pskb_may_pull failed.\n");
1047 e1000_receive_skb(adapter, netdev, skb, status,
1051 rx_desc->status = 0;
1053 /* return some buffers to hardware, one at a time is too slow */
1054 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1055 adapter->alloc_rx_buf(adapter, cleaned_count);
1059 /* use prefetched values */
1061 buffer_info = next_buffer;
1063 rx_ring->next_to_clean = i;
1065 cleaned_count = e1000_desc_unused(rx_ring);
1067 adapter->alloc_rx_buf(adapter, cleaned_count);
1069 adapter->total_rx_bytes += total_rx_bytes;
1070 adapter->total_rx_packets += total_rx_packets;
1071 netdev->stats.rx_bytes += total_rx_bytes;
1072 netdev->stats.rx_packets += total_rx_packets;
1077 * e1000_clean_rx_ring - Free Rx Buffers per Queue
1078 * @adapter: board private structure
1080 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1082 struct e1000_ring *rx_ring = adapter->rx_ring;
1083 struct e1000_buffer *buffer_info;
1084 struct e1000_ps_page *ps_page;
1085 struct pci_dev *pdev = adapter->pdev;
1088 /* Free all the Rx ring sk_buffs */
1089 for (i = 0; i < rx_ring->count; i++) {
1090 buffer_info = &rx_ring->buffer_info[i];
1091 if (buffer_info->dma) {
1092 if (adapter->clean_rx == e1000_clean_rx_irq)
1093 pci_unmap_single(pdev, buffer_info->dma,
1094 adapter->rx_buffer_len,
1095 PCI_DMA_FROMDEVICE);
1096 else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1097 pci_unmap_page(pdev, buffer_info->dma,
1099 PCI_DMA_FROMDEVICE);
1100 else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1101 pci_unmap_single(pdev, buffer_info->dma,
1102 adapter->rx_ps_bsize0,
1103 PCI_DMA_FROMDEVICE);
1104 buffer_info->dma = 0;
1107 if (buffer_info->page) {
1108 put_page(buffer_info->page);
1109 buffer_info->page = NULL;
1112 if (buffer_info->skb) {
1113 dev_kfree_skb(buffer_info->skb);
1114 buffer_info->skb = NULL;
1117 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1118 ps_page = &buffer_info->ps_pages[j];
1121 pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1122 PCI_DMA_FROMDEVICE);
1124 put_page(ps_page->page);
1125 ps_page->page = NULL;
1129 /* there also may be some cached data from a chained receive */
1130 if (rx_ring->rx_skb_top) {
1131 dev_kfree_skb(rx_ring->rx_skb_top);
1132 rx_ring->rx_skb_top = NULL;
1135 /* Zero out the descriptor ring */
1136 memset(rx_ring->desc, 0, rx_ring->size);
1138 rx_ring->next_to_clean = 0;
1139 rx_ring->next_to_use = 0;
1140 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1142 writel(0, adapter->hw.hw_addr + rx_ring->head);
1143 writel(0, adapter->hw.hw_addr + rx_ring->tail);
1146 static void e1000e_downshift_workaround(struct work_struct *work)
1148 struct e1000_adapter *adapter = container_of(work,
1149 struct e1000_adapter, downshift_task);
1151 e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1155 * e1000_intr_msi - Interrupt Handler
1156 * @irq: interrupt number
1157 * @data: pointer to a network interface device structure
1159 static irqreturn_t e1000_intr_msi(int irq, void *data)
1161 struct net_device *netdev = data;
1162 struct e1000_adapter *adapter = netdev_priv(netdev);
1163 struct e1000_hw *hw = &adapter->hw;
1164 u32 icr = er32(ICR);
1167 * read ICR disables interrupts using IAM
1170 if (icr & E1000_ICR_LSC) {
1171 hw->mac.get_link_status = 1;
1173 * ICH8 workaround-- Call gig speed drop workaround on cable
1174 * disconnect (LSC) before accessing any PHY registers
1176 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1177 (!(er32(STATUS) & E1000_STATUS_LU)))
1178 schedule_work(&adapter->downshift_task);
1181 * 80003ES2LAN workaround-- For packet buffer work-around on
1182 * link down event; disable receives here in the ISR and reset
1183 * adapter in watchdog
1185 if (netif_carrier_ok(netdev) &&
1186 adapter->flags & FLAG_RX_NEEDS_RESTART) {
1187 /* disable receives */
1188 u32 rctl = er32(RCTL);
1189 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1190 adapter->flags |= FLAG_RX_RESTART_NOW;
1192 /* guard against interrupt when we're going down */
1193 if (!test_bit(__E1000_DOWN, &adapter->state))
1194 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1197 if (napi_schedule_prep(&adapter->napi)) {
1198 adapter->total_tx_bytes = 0;
1199 adapter->total_tx_packets = 0;
1200 adapter->total_rx_bytes = 0;
1201 adapter->total_rx_packets = 0;
1202 __napi_schedule(&adapter->napi);
1209 * e1000_intr - Interrupt Handler
1210 * @irq: interrupt number
1211 * @data: pointer to a network interface device structure
1213 static irqreturn_t e1000_intr(int irq, void *data)
1215 struct net_device *netdev = data;
1216 struct e1000_adapter *adapter = netdev_priv(netdev);
1217 struct e1000_hw *hw = &adapter->hw;
1218 u32 rctl, icr = er32(ICR);
1220 if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1221 return IRQ_NONE; /* Not our interrupt */
1224 * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1225 * not set, then the adapter didn't send an interrupt
1227 if (!(icr & E1000_ICR_INT_ASSERTED))
1231 * Interrupt Auto-Mask...upon reading ICR,
1232 * interrupts are masked. No need for the
1236 if (icr & E1000_ICR_LSC) {
1237 hw->mac.get_link_status = 1;
1239 * ICH8 workaround-- Call gig speed drop workaround on cable
1240 * disconnect (LSC) before accessing any PHY registers
1242 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1243 (!(er32(STATUS) & E1000_STATUS_LU)))
1244 schedule_work(&adapter->downshift_task);
1247 * 80003ES2LAN workaround--
1248 * For packet buffer work-around on link down event;
1249 * disable receives here in the ISR and
1250 * reset adapter in watchdog
1252 if (netif_carrier_ok(netdev) &&
1253 (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1254 /* disable receives */
1256 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1257 adapter->flags |= FLAG_RX_RESTART_NOW;
1259 /* guard against interrupt when we're going down */
1260 if (!test_bit(__E1000_DOWN, &adapter->state))
1261 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1264 if (napi_schedule_prep(&adapter->napi)) {
1265 adapter->total_tx_bytes = 0;
1266 adapter->total_tx_packets = 0;
1267 adapter->total_rx_bytes = 0;
1268 adapter->total_rx_packets = 0;
1269 __napi_schedule(&adapter->napi);
1275 static irqreturn_t e1000_msix_other(int irq, void *data)
1277 struct net_device *netdev = data;
1278 struct e1000_adapter *adapter = netdev_priv(netdev);
1279 struct e1000_hw *hw = &adapter->hw;
1280 u32 icr = er32(ICR);
1282 if (!(icr & E1000_ICR_INT_ASSERTED)) {
1283 if (!test_bit(__E1000_DOWN, &adapter->state))
1284 ew32(IMS, E1000_IMS_OTHER);
1288 if (icr & adapter->eiac_mask)
1289 ew32(ICS, (icr & adapter->eiac_mask));
1291 if (icr & E1000_ICR_OTHER) {
1292 if (!(icr & E1000_ICR_LSC))
1293 goto no_link_interrupt;
1294 hw->mac.get_link_status = 1;
1295 /* guard against interrupt when we're going down */
1296 if (!test_bit(__E1000_DOWN, &adapter->state))
1297 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1301 if (!test_bit(__E1000_DOWN, &adapter->state))
1302 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1308 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1310 struct net_device *netdev = data;
1311 struct e1000_adapter *adapter = netdev_priv(netdev);
1312 struct e1000_hw *hw = &adapter->hw;
1313 struct e1000_ring *tx_ring = adapter->tx_ring;
1316 adapter->total_tx_bytes = 0;
1317 adapter->total_tx_packets = 0;
1319 if (!e1000_clean_tx_irq(adapter))
1320 /* Ring was not completely cleaned, so fire another interrupt */
1321 ew32(ICS, tx_ring->ims_val);
1326 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1328 struct net_device *netdev = data;
1329 struct e1000_adapter *adapter = netdev_priv(netdev);
1331 /* Write the ITR value calculated at the end of the
1332 * previous interrupt.
1334 if (adapter->rx_ring->set_itr) {
1335 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1336 adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1337 adapter->rx_ring->set_itr = 0;
1340 if (napi_schedule_prep(&adapter->napi)) {
1341 adapter->total_rx_bytes = 0;
1342 adapter->total_rx_packets = 0;
1343 __napi_schedule(&adapter->napi);
1349 * e1000_configure_msix - Configure MSI-X hardware
1351 * e1000_configure_msix sets up the hardware to properly
1352 * generate MSI-X interrupts.
1354 static void e1000_configure_msix(struct e1000_adapter *adapter)
1356 struct e1000_hw *hw = &adapter->hw;
1357 struct e1000_ring *rx_ring = adapter->rx_ring;
1358 struct e1000_ring *tx_ring = adapter->tx_ring;
1360 u32 ctrl_ext, ivar = 0;
1362 adapter->eiac_mask = 0;
1364 /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1365 if (hw->mac.type == e1000_82574) {
1366 u32 rfctl = er32(RFCTL);
1367 rfctl |= E1000_RFCTL_ACK_DIS;
1371 #define E1000_IVAR_INT_ALLOC_VALID 0x8
1372 /* Configure Rx vector */
1373 rx_ring->ims_val = E1000_IMS_RXQ0;
1374 adapter->eiac_mask |= rx_ring->ims_val;
1375 if (rx_ring->itr_val)
1376 writel(1000000000 / (rx_ring->itr_val * 256),
1377 hw->hw_addr + rx_ring->itr_register);
1379 writel(1, hw->hw_addr + rx_ring->itr_register);
1380 ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1382 /* Configure Tx vector */
1383 tx_ring->ims_val = E1000_IMS_TXQ0;
1385 if (tx_ring->itr_val)
1386 writel(1000000000 / (tx_ring->itr_val * 256),
1387 hw->hw_addr + tx_ring->itr_register);
1389 writel(1, hw->hw_addr + tx_ring->itr_register);
1390 adapter->eiac_mask |= tx_ring->ims_val;
1391 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1393 /* set vector for Other Causes, e.g. link changes */
1395 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1396 if (rx_ring->itr_val)
1397 writel(1000000000 / (rx_ring->itr_val * 256),
1398 hw->hw_addr + E1000_EITR_82574(vector));
1400 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1402 /* Cause Tx interrupts on every write back */
1407 /* enable MSI-X PBA support */
1408 ctrl_ext = er32(CTRL_EXT);
1409 ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1411 /* Auto-Mask Other interrupts upon ICR read */
1412 #define E1000_EIAC_MASK_82574 0x01F00000
1413 ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1414 ctrl_ext |= E1000_CTRL_EXT_EIAME;
1415 ew32(CTRL_EXT, ctrl_ext);
1419 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1421 if (adapter->msix_entries) {
1422 pci_disable_msix(adapter->pdev);
1423 kfree(adapter->msix_entries);
1424 adapter->msix_entries = NULL;
1425 } else if (adapter->flags & FLAG_MSI_ENABLED) {
1426 pci_disable_msi(adapter->pdev);
1427 adapter->flags &= ~FLAG_MSI_ENABLED;
1434 * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1436 * Attempt to configure interrupts using the best available
1437 * capabilities of the hardware and kernel.
1439 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1445 switch (adapter->int_mode) {
1446 case E1000E_INT_MODE_MSIX:
1447 if (adapter->flags & FLAG_HAS_MSIX) {
1448 numvecs = 3; /* RxQ0, TxQ0 and other */
1449 adapter->msix_entries = kcalloc(numvecs,
1450 sizeof(struct msix_entry),
1452 if (adapter->msix_entries) {
1453 for (i = 0; i < numvecs; i++)
1454 adapter->msix_entries[i].entry = i;
1456 err = pci_enable_msix(adapter->pdev,
1457 adapter->msix_entries,
1462 /* MSI-X failed, so fall through and try MSI */
1463 e_err("Failed to initialize MSI-X interrupts. "
1464 "Falling back to MSI interrupts.\n");
1465 e1000e_reset_interrupt_capability(adapter);
1467 adapter->int_mode = E1000E_INT_MODE_MSI;
1469 case E1000E_INT_MODE_MSI:
1470 if (!pci_enable_msi(adapter->pdev)) {
1471 adapter->flags |= FLAG_MSI_ENABLED;
1473 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1474 e_err("Failed to initialize MSI interrupts. Falling "
1475 "back to legacy interrupts.\n");
1478 case E1000E_INT_MODE_LEGACY:
1479 /* Don't do anything; this is the system default */
1487 * e1000_request_msix - Initialize MSI-X interrupts
1489 * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1492 static int e1000_request_msix(struct e1000_adapter *adapter)
1494 struct net_device *netdev = adapter->netdev;
1495 int err = 0, vector = 0;
1497 if (strlen(netdev->name) < (IFNAMSIZ - 5))
1498 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1500 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1501 err = request_irq(adapter->msix_entries[vector].vector,
1502 e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1506 adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1507 adapter->rx_ring->itr_val = adapter->itr;
1510 if (strlen(netdev->name) < (IFNAMSIZ - 5))
1511 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1513 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1514 err = request_irq(adapter->msix_entries[vector].vector,
1515 e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1519 adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1520 adapter->tx_ring->itr_val = adapter->itr;
1523 err = request_irq(adapter->msix_entries[vector].vector,
1524 e1000_msix_other, 0, netdev->name, netdev);
1528 e1000_configure_msix(adapter);
1535 * e1000_request_irq - initialize interrupts
1537 * Attempts to configure interrupts using the best available
1538 * capabilities of the hardware and kernel.
1540 static int e1000_request_irq(struct e1000_adapter *adapter)
1542 struct net_device *netdev = adapter->netdev;
1545 if (adapter->msix_entries) {
1546 err = e1000_request_msix(adapter);
1549 /* fall back to MSI */
1550 e1000e_reset_interrupt_capability(adapter);
1551 adapter->int_mode = E1000E_INT_MODE_MSI;
1552 e1000e_set_interrupt_capability(adapter);
1554 if (adapter->flags & FLAG_MSI_ENABLED) {
1555 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1556 netdev->name, netdev);
1560 /* fall back to legacy interrupt */
1561 e1000e_reset_interrupt_capability(adapter);
1562 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1565 err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1566 netdev->name, netdev);
1568 e_err("Unable to allocate interrupt, Error: %d\n", err);
1573 static void e1000_free_irq(struct e1000_adapter *adapter)
1575 struct net_device *netdev = adapter->netdev;
1577 if (adapter->msix_entries) {
1580 free_irq(adapter->msix_entries[vector].vector, netdev);
1583 free_irq(adapter->msix_entries[vector].vector, netdev);
1586 /* Other Causes interrupt vector */
1587 free_irq(adapter->msix_entries[vector].vector, netdev);
1591 free_irq(adapter->pdev->irq, netdev);
1595 * e1000_irq_disable - Mask off interrupt generation on the NIC
1597 static void e1000_irq_disable(struct e1000_adapter *adapter)
1599 struct e1000_hw *hw = &adapter->hw;
1602 if (adapter->msix_entries)
1603 ew32(EIAC_82574, 0);
1605 synchronize_irq(adapter->pdev->irq);
1609 * e1000_irq_enable - Enable default interrupt generation settings
1611 static void e1000_irq_enable(struct e1000_adapter *adapter)
1613 struct e1000_hw *hw = &adapter->hw;
1615 if (adapter->msix_entries) {
1616 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1617 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1619 ew32(IMS, IMS_ENABLE_MASK);
1625 * e1000_get_hw_control - get control of the h/w from f/w
1626 * @adapter: address of board private structure
1628 * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1629 * For ASF and Pass Through versions of f/w this means that
1630 * the driver is loaded. For AMT version (only with 82573)
1631 * of the f/w this means that the network i/f is open.
1633 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1635 struct e1000_hw *hw = &adapter->hw;
1639 /* Let firmware know the driver has taken over */
1640 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1642 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1643 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1644 ctrl_ext = er32(CTRL_EXT);
1645 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1650 * e1000_release_hw_control - release control of the h/w to f/w
1651 * @adapter: address of board private structure
1653 * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1654 * For ASF and Pass Through versions of f/w this means that the
1655 * driver is no longer loaded. For AMT version (only with 82573) i
1656 * of the f/w this means that the network i/f is closed.
1659 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1661 struct e1000_hw *hw = &adapter->hw;
1665 /* Let firmware taken over control of h/w */
1666 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1668 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1669 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1670 ctrl_ext = er32(CTRL_EXT);
1671 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1676 * @e1000_alloc_ring - allocate memory for a ring structure
1678 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1679 struct e1000_ring *ring)
1681 struct pci_dev *pdev = adapter->pdev;
1683 ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1692 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1693 * @adapter: board private structure
1695 * Return 0 on success, negative on failure
1697 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1699 struct e1000_ring *tx_ring = adapter->tx_ring;
1700 int err = -ENOMEM, size;
1702 size = sizeof(struct e1000_buffer) * tx_ring->count;
1703 tx_ring->buffer_info = vmalloc(size);
1704 if (!tx_ring->buffer_info)
1706 memset(tx_ring->buffer_info, 0, size);
1708 /* round up to nearest 4K */
1709 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1710 tx_ring->size = ALIGN(tx_ring->size, 4096);
1712 err = e1000_alloc_ring_dma(adapter, tx_ring);
1716 tx_ring->next_to_use = 0;
1717 tx_ring->next_to_clean = 0;
1721 vfree(tx_ring->buffer_info);
1722 e_err("Unable to allocate memory for the transmit descriptor ring\n");
1727 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1728 * @adapter: board private structure
1730 * Returns 0 on success, negative on failure
1732 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1734 struct e1000_ring *rx_ring = adapter->rx_ring;
1735 struct e1000_buffer *buffer_info;
1736 int i, size, desc_len, err = -ENOMEM;
1738 size = sizeof(struct e1000_buffer) * rx_ring->count;
1739 rx_ring->buffer_info = vmalloc(size);
1740 if (!rx_ring->buffer_info)
1742 memset(rx_ring->buffer_info, 0, size);
1744 for (i = 0; i < rx_ring->count; i++) {
1745 buffer_info = &rx_ring->buffer_info[i];
1746 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1747 sizeof(struct e1000_ps_page),
1749 if (!buffer_info->ps_pages)
1753 desc_len = sizeof(union e1000_rx_desc_packet_split);
1755 /* Round up to nearest 4K */
1756 rx_ring->size = rx_ring->count * desc_len;
1757 rx_ring->size = ALIGN(rx_ring->size, 4096);
1759 err = e1000_alloc_ring_dma(adapter, rx_ring);
1763 rx_ring->next_to_clean = 0;
1764 rx_ring->next_to_use = 0;
1765 rx_ring->rx_skb_top = NULL;
1770 for (i = 0; i < rx_ring->count; i++) {
1771 buffer_info = &rx_ring->buffer_info[i];
1772 kfree(buffer_info->ps_pages);
1775 vfree(rx_ring->buffer_info);
1776 e_err("Unable to allocate memory for the transmit descriptor ring\n");
1781 * e1000_clean_tx_ring - Free Tx Buffers
1782 * @adapter: board private structure
1784 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1786 struct e1000_ring *tx_ring = adapter->tx_ring;
1787 struct e1000_buffer *buffer_info;
1791 for (i = 0; i < tx_ring->count; i++) {
1792 buffer_info = &tx_ring->buffer_info[i];
1793 e1000_put_txbuf(adapter, buffer_info);
1796 size = sizeof(struct e1000_buffer) * tx_ring->count;
1797 memset(tx_ring->buffer_info, 0, size);
1799 memset(tx_ring->desc, 0, tx_ring->size);
1801 tx_ring->next_to_use = 0;
1802 tx_ring->next_to_clean = 0;
1804 writel(0, adapter->hw.hw_addr + tx_ring->head);
1805 writel(0, adapter->hw.hw_addr + tx_ring->tail);
1809 * e1000e_free_tx_resources - Free Tx Resources per Queue
1810 * @adapter: board private structure
1812 * Free all transmit software resources
1814 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1816 struct pci_dev *pdev = adapter->pdev;
1817 struct e1000_ring *tx_ring = adapter->tx_ring;
1819 e1000_clean_tx_ring(adapter);
1821 vfree(tx_ring->buffer_info);
1822 tx_ring->buffer_info = NULL;
1824 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1826 tx_ring->desc = NULL;
1830 * e1000e_free_rx_resources - Free Rx Resources
1831 * @adapter: board private structure
1833 * Free all receive software resources
1836 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1838 struct pci_dev *pdev = adapter->pdev;
1839 struct e1000_ring *rx_ring = adapter->rx_ring;
1842 e1000_clean_rx_ring(adapter);
1844 for (i = 0; i < rx_ring->count; i++) {
1845 kfree(rx_ring->buffer_info[i].ps_pages);
1848 vfree(rx_ring->buffer_info);
1849 rx_ring->buffer_info = NULL;
1851 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1853 rx_ring->desc = NULL;
1857 * e1000_update_itr - update the dynamic ITR value based on statistics
1858 * @adapter: pointer to adapter
1859 * @itr_setting: current adapter->itr
1860 * @packets: the number of packets during this measurement interval
1861 * @bytes: the number of bytes during this measurement interval
1863 * Stores a new ITR value based on packets and byte
1864 * counts during the last interrupt. The advantage of per interrupt
1865 * computation is faster updates and more accurate ITR for the current
1866 * traffic pattern. Constants in this function were computed
1867 * based on theoretical maximum wire speed and thresholds were set based
1868 * on testing data as well as attempting to minimize response time
1869 * while increasing bulk throughput. This functionality is controlled
1870 * by the InterruptThrottleRate module parameter.
1872 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1873 u16 itr_setting, int packets,
1876 unsigned int retval = itr_setting;
1879 goto update_itr_done;
1881 switch (itr_setting) {
1882 case lowest_latency:
1883 /* handle TSO and jumbo frames */
1884 if (bytes/packets > 8000)
1885 retval = bulk_latency;
1886 else if ((packets < 5) && (bytes > 512)) {
1887 retval = low_latency;
1890 case low_latency: /* 50 usec aka 20000 ints/s */
1891 if (bytes > 10000) {
1892 /* this if handles the TSO accounting */
1893 if (bytes/packets > 8000) {
1894 retval = bulk_latency;
1895 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1896 retval = bulk_latency;
1897 } else if ((packets > 35)) {
1898 retval = lowest_latency;
1900 } else if (bytes/packets > 2000) {
1901 retval = bulk_latency;
1902 } else if (packets <= 2 && bytes < 512) {
1903 retval = lowest_latency;
1906 case bulk_latency: /* 250 usec aka 4000 ints/s */
1907 if (bytes > 25000) {
1909 retval = low_latency;
1911 } else if (bytes < 6000) {
1912 retval = low_latency;
1921 static void e1000_set_itr(struct e1000_adapter *adapter)
1923 struct e1000_hw *hw = &adapter->hw;
1925 u32 new_itr = adapter->itr;
1927 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1928 if (adapter->link_speed != SPEED_1000) {
1934 adapter->tx_itr = e1000_update_itr(adapter,
1936 adapter->total_tx_packets,
1937 adapter->total_tx_bytes);
1938 /* conservative mode (itr 3) eliminates the lowest_latency setting */
1939 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1940 adapter->tx_itr = low_latency;
1942 adapter->rx_itr = e1000_update_itr(adapter,
1944 adapter->total_rx_packets,
1945 adapter->total_rx_bytes);
1946 /* conservative mode (itr 3) eliminates the lowest_latency setting */
1947 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1948 adapter->rx_itr = low_latency;
1950 current_itr = max(adapter->rx_itr, adapter->tx_itr);
1952 switch (current_itr) {
1953 /* counts and packets in update_itr are dependent on these numbers */
1954 case lowest_latency:
1958 new_itr = 20000; /* aka hwitr = ~200 */
1968 if (new_itr != adapter->itr) {
1970 * this attempts to bias the interrupt rate towards Bulk
1971 * by adding intermediate steps when interrupt rate is
1974 new_itr = new_itr > adapter->itr ?
1975 min(adapter->itr + (new_itr >> 2), new_itr) :
1977 adapter->itr = new_itr;
1978 adapter->rx_ring->itr_val = new_itr;
1979 if (adapter->msix_entries)
1980 adapter->rx_ring->set_itr = 1;
1982 ew32(ITR, 1000000000 / (new_itr * 256));
1987 * e1000_alloc_queues - Allocate memory for all rings
1988 * @adapter: board private structure to initialize
1990 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1992 adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1993 if (!adapter->tx_ring)
1996 adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1997 if (!adapter->rx_ring)
2002 e_err("Unable to allocate memory for queues\n");
2003 kfree(adapter->rx_ring);
2004 kfree(adapter->tx_ring);
2009 * e1000_clean - NAPI Rx polling callback
2010 * @napi: struct associated with this polling callback
2011 * @budget: amount of packets driver is allowed to process this poll
2013 static int e1000_clean(struct napi_struct *napi, int budget)
2015 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2016 struct e1000_hw *hw = &adapter->hw;
2017 struct net_device *poll_dev = adapter->netdev;
2018 int tx_cleaned = 1, work_done = 0;
2020 adapter = netdev_priv(poll_dev);
2022 if (adapter->msix_entries &&
2023 !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2026 tx_cleaned = e1000_clean_tx_irq(adapter);
2029 adapter->clean_rx(adapter, &work_done, budget);
2034 /* If budget not fully consumed, exit the polling mode */
2035 if (work_done < budget) {
2036 if (adapter->itr_setting & 3)
2037 e1000_set_itr(adapter);
2038 napi_complete(napi);
2039 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2040 if (adapter->msix_entries)
2041 ew32(IMS, adapter->rx_ring->ims_val);
2043 e1000_irq_enable(adapter);
2050 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2052 struct e1000_adapter *adapter = netdev_priv(netdev);
2053 struct e1000_hw *hw = &adapter->hw;
2056 /* don't update vlan cookie if already programmed */
2057 if ((adapter->hw.mng_cookie.status &
2058 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2059 (vid == adapter->mng_vlan_id))
2062 /* add VID to filter table */
2063 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2064 index = (vid >> 5) & 0x7F;
2065 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2066 vfta |= (1 << (vid & 0x1F));
2067 hw->mac.ops.write_vfta(hw, index, vfta);
2071 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2073 struct e1000_adapter *adapter = netdev_priv(netdev);
2074 struct e1000_hw *hw = &adapter->hw;
2077 if (!test_bit(__E1000_DOWN, &adapter->state))
2078 e1000_irq_disable(adapter);
2079 vlan_group_set_device(adapter->vlgrp, vid, NULL);
2081 if (!test_bit(__E1000_DOWN, &adapter->state))
2082 e1000_irq_enable(adapter);
2084 if ((adapter->hw.mng_cookie.status &
2085 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2086 (vid == adapter->mng_vlan_id)) {
2087 /* release control to f/w */
2088 e1000_release_hw_control(adapter);
2092 /* remove VID from filter table */
2093 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2094 index = (vid >> 5) & 0x7F;
2095 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2096 vfta &= ~(1 << (vid & 0x1F));
2097 hw->mac.ops.write_vfta(hw, index, vfta);
2101 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2103 struct net_device *netdev = adapter->netdev;
2104 u16 vid = adapter->hw.mng_cookie.vlan_id;
2105 u16 old_vid = adapter->mng_vlan_id;
2107 if (!adapter->vlgrp)
2110 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2111 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2112 if (adapter->hw.mng_cookie.status &
2113 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2114 e1000_vlan_rx_add_vid(netdev, vid);
2115 adapter->mng_vlan_id = vid;
2118 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2120 !vlan_group_get_device(adapter->vlgrp, old_vid))
2121 e1000_vlan_rx_kill_vid(netdev, old_vid);
2123 adapter->mng_vlan_id = vid;
2128 static void e1000_vlan_rx_register(struct net_device *netdev,
2129 struct vlan_group *grp)
2131 struct e1000_adapter *adapter = netdev_priv(netdev);
2132 struct e1000_hw *hw = &adapter->hw;
2135 if (!test_bit(__E1000_DOWN, &adapter->state))
2136 e1000_irq_disable(adapter);
2137 adapter->vlgrp = grp;
2140 /* enable VLAN tag insert/strip */
2142 ctrl |= E1000_CTRL_VME;
2145 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2146 /* enable VLAN receive filtering */
2148 rctl &= ~E1000_RCTL_CFIEN;
2150 e1000_update_mng_vlan(adapter);
2153 /* disable VLAN tag insert/strip */
2155 ctrl &= ~E1000_CTRL_VME;
2158 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2159 if (adapter->mng_vlan_id !=
2160 (u16)E1000_MNG_VLAN_NONE) {
2161 e1000_vlan_rx_kill_vid(netdev,
2162 adapter->mng_vlan_id);
2163 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2168 if (!test_bit(__E1000_DOWN, &adapter->state))
2169 e1000_irq_enable(adapter);
2172 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2176 e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2178 if (!adapter->vlgrp)
2181 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2182 if (!vlan_group_get_device(adapter->vlgrp, vid))
2184 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2188 static void e1000_init_manageability(struct e1000_adapter *adapter)
2190 struct e1000_hw *hw = &adapter->hw;
2193 if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2199 * enable receiving management packets to the host. this will probably
2200 * generate destination unreachable messages from the host OS, but
2201 * the packets will be handled on SMBUS
2203 manc |= E1000_MANC_EN_MNG2HOST;
2204 manc2h = er32(MANC2H);
2205 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2206 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2207 manc2h |= E1000_MNG2HOST_PORT_623;
2208 manc2h |= E1000_MNG2HOST_PORT_664;
2209 ew32(MANC2H, manc2h);
2214 * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2215 * @adapter: board private structure
2217 * Configure the Tx unit of the MAC after a reset.
2219 static void e1000_configure_tx(struct e1000_adapter *adapter)
2221 struct e1000_hw *hw = &adapter->hw;
2222 struct e1000_ring *tx_ring = adapter->tx_ring;
2224 u32 tdlen, tctl, tipg, tarc;
2227 /* Setup the HW Tx Head and Tail descriptor pointers */
2228 tdba = tx_ring->dma;
2229 tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2230 ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2231 ew32(TDBAH, (tdba >> 32));
2235 tx_ring->head = E1000_TDH;
2236 tx_ring->tail = E1000_TDT;
2238 /* Set the default values for the Tx Inter Packet Gap timer */
2239 tipg = DEFAULT_82543_TIPG_IPGT_COPPER; /* 8 */
2240 ipgr1 = DEFAULT_82543_TIPG_IPGR1; /* 8 */
2241 ipgr2 = DEFAULT_82543_TIPG_IPGR2; /* 6 */
2243 if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2244 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /* 7 */
2246 tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2247 tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2250 /* Set the Tx Interrupt Delay register */
2251 ew32(TIDV, adapter->tx_int_delay);
2252 /* Tx irq moderation */
2253 ew32(TADV, adapter->tx_abs_int_delay);
2255 /* Program the Transmit Control Register */
2257 tctl &= ~E1000_TCTL_CT;
2258 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2259 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2261 if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2262 tarc = er32(TARC(0));
2264 * set the speed mode bit, we'll clear it if we're not at
2265 * gigabit link later
2267 #define SPEED_MODE_BIT (1 << 21)
2268 tarc |= SPEED_MODE_BIT;
2269 ew32(TARC(0), tarc);
2272 /* errata: program both queues to unweighted RR */
2273 if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2274 tarc = er32(TARC(0));
2276 ew32(TARC(0), tarc);
2277 tarc = er32(TARC(1));
2279 ew32(TARC(1), tarc);
2282 /* Setup Transmit Descriptor Settings for eop descriptor */
2283 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2285 /* only set IDE if we are delaying interrupts using the timers */
2286 if (adapter->tx_int_delay)
2287 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2289 /* enable Report Status bit */
2290 adapter->txd_cmd |= E1000_TXD_CMD_RS;
2294 e1000e_config_collision_dist(hw);
2298 * e1000_setup_rctl - configure the receive control registers
2299 * @adapter: Board private structure
2301 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2302 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2303 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2305 struct e1000_hw *hw = &adapter->hw;
2310 /* Program MC offset vector base */
2312 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2313 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2314 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2315 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2317 /* Do not Store bad packets */
2318 rctl &= ~E1000_RCTL_SBP;
2320 /* Enable Long Packet receive */
2321 if (adapter->netdev->mtu <= ETH_DATA_LEN)
2322 rctl &= ~E1000_RCTL_LPE;
2324 rctl |= E1000_RCTL_LPE;
2326 /* Some systems expect that the CRC is included in SMBUS traffic. The
2327 * hardware strips the CRC before sending to both SMBUS (BMC) and to
2328 * host memory when this is enabled
2330 if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2331 rctl |= E1000_RCTL_SECRC;
2333 /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2334 if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2337 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2339 phy_data |= (1 << 2);
2340 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2342 e1e_rphy(hw, 22, &phy_data);
2344 phy_data |= (1 << 14);
2345 e1e_wphy(hw, 0x10, 0x2823);
2346 e1e_wphy(hw, 0x11, 0x0003);
2347 e1e_wphy(hw, 22, phy_data);
2350 /* Setup buffer sizes */
2351 rctl &= ~E1000_RCTL_SZ_4096;
2352 rctl |= E1000_RCTL_BSEX;
2353 switch (adapter->rx_buffer_len) {
2356 rctl |= E1000_RCTL_SZ_2048;
2357 rctl &= ~E1000_RCTL_BSEX;
2360 rctl |= E1000_RCTL_SZ_4096;
2363 rctl |= E1000_RCTL_SZ_8192;
2366 rctl |= E1000_RCTL_SZ_16384;
2371 * 82571 and greater support packet-split where the protocol
2372 * header is placed in skb->data and the packet data is
2373 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2374 * In the case of a non-split, skb->data is linearly filled,
2375 * followed by the page buffers. Therefore, skb->data is
2376 * sized to hold the largest protocol header.
2378 * allocations using alloc_page take too long for regular MTU
2379 * so only enable packet split for jumbo frames
2381 * Using pages when the page size is greater than 16k wastes
2382 * a lot of memory, since we allocate 3 pages at all times
2385 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2386 if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2387 (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2388 adapter->rx_ps_pages = pages;
2390 adapter->rx_ps_pages = 0;
2392 if (adapter->rx_ps_pages) {
2393 /* Configure extra packet-split registers */
2394 rfctl = er32(RFCTL);
2395 rfctl |= E1000_RFCTL_EXTEN;
2397 * disable packet split support for IPv6 extension headers,
2398 * because some malformed IPv6 headers can hang the Rx
2400 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2401 E1000_RFCTL_NEW_IPV6_EXT_DIS);
2405 /* Enable Packet split descriptors */
2406 rctl |= E1000_RCTL_DTYP_PS;
2408 psrctl |= adapter->rx_ps_bsize0 >>
2409 E1000_PSRCTL_BSIZE0_SHIFT;
2411 switch (adapter->rx_ps_pages) {
2413 psrctl |= PAGE_SIZE <<
2414 E1000_PSRCTL_BSIZE3_SHIFT;
2416 psrctl |= PAGE_SIZE <<
2417 E1000_PSRCTL_BSIZE2_SHIFT;
2419 psrctl |= PAGE_SIZE >>
2420 E1000_PSRCTL_BSIZE1_SHIFT;
2424 ew32(PSRCTL, psrctl);
2428 /* just started the receive unit, no need to restart */
2429 adapter->flags &= ~FLAG_RX_RESTART_NOW;
2433 * e1000_configure_rx - Configure Receive Unit after Reset
2434 * @adapter: board private structure
2436 * Configure the Rx unit of the MAC after a reset.
2438 static void e1000_configure_rx(struct e1000_adapter *adapter)
2440 struct e1000_hw *hw = &adapter->hw;
2441 struct e1000_ring *rx_ring = adapter->rx_ring;
2443 u32 rdlen, rctl, rxcsum, ctrl_ext;
2445 if (adapter->rx_ps_pages) {
2446 /* this is a 32 byte descriptor */
2447 rdlen = rx_ring->count *
2448 sizeof(union e1000_rx_desc_packet_split);
2449 adapter->clean_rx = e1000_clean_rx_irq_ps;
2450 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2451 } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2452 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2453 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2454 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2456 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2457 adapter->clean_rx = e1000_clean_rx_irq;
2458 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2461 /* disable receives while setting up the descriptors */
2463 ew32(RCTL, rctl & ~E1000_RCTL_EN);
2467 /* set the Receive Delay Timer Register */
2468 ew32(RDTR, adapter->rx_int_delay);
2470 /* irq moderation */
2471 ew32(RADV, adapter->rx_abs_int_delay);
2472 if (adapter->itr_setting != 0)
2473 ew32(ITR, 1000000000 / (adapter->itr * 256));
2475 ctrl_ext = er32(CTRL_EXT);
2476 /* Auto-Mask interrupts upon ICR access */
2477 ctrl_ext |= E1000_CTRL_EXT_IAME;
2478 ew32(IAM, 0xffffffff);
2479 ew32(CTRL_EXT, ctrl_ext);
2483 * Setup the HW Rx Head and Tail Descriptor Pointers and
2484 * the Base and Length of the Rx Descriptor Ring
2486 rdba = rx_ring->dma;
2487 ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2488 ew32(RDBAH, (rdba >> 32));
2492 rx_ring->head = E1000_RDH;
2493 rx_ring->tail = E1000_RDT;
2495 /* Enable Receive Checksum Offload for TCP and UDP */
2496 rxcsum = er32(RXCSUM);
2497 if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2498 rxcsum |= E1000_RXCSUM_TUOFL;
2501 * IPv4 payload checksum for UDP fragments must be
2502 * used in conjunction with packet-split.
2504 if (adapter->rx_ps_pages)
2505 rxcsum |= E1000_RXCSUM_IPPCSE;
2507 rxcsum &= ~E1000_RXCSUM_TUOFL;
2508 /* no need to clear IPPCSE as it defaults to 0 */
2510 ew32(RXCSUM, rxcsum);
2513 * Enable early receives on supported devices, only takes effect when
2514 * packet size is equal or larger than the specified value (in 8 byte
2515 * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2517 if (adapter->flags & FLAG_HAS_ERT) {
2518 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2519 u32 rxdctl = er32(RXDCTL(0));
2520 ew32(RXDCTL(0), rxdctl | 0x3);
2521 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2523 * With jumbo frames and early-receive enabled,
2524 * excessive C-state transition latencies result in
2525 * dropped transactions.
2527 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2528 adapter->netdev->name, 55);
2530 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2531 adapter->netdev->name,
2532 PM_QOS_DEFAULT_VALUE);
2536 /* Enable Receives */
2541 * e1000_update_mc_addr_list - Update Multicast addresses
2542 * @hw: pointer to the HW structure
2543 * @mc_addr_list: array of multicast addresses to program
2544 * @mc_addr_count: number of multicast addresses to program
2546 * Updates the Multicast Table Array.
2547 * The caller must have a packed mc_addr_list of multicast addresses.
2549 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2552 hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
2556 * e1000_set_multi - Multicast and Promiscuous mode set
2557 * @netdev: network interface device structure
2559 * The set_multi entry point is called whenever the multicast address
2560 * list or the network interface flags are updated. This routine is
2561 * responsible for configuring the hardware for proper multicast,
2562 * promiscuous mode, and all-multi behavior.
2564 static void e1000_set_multi(struct net_device *netdev)
2566 struct e1000_adapter *adapter = netdev_priv(netdev);
2567 struct e1000_hw *hw = &adapter->hw;
2568 struct dev_mc_list *mc_ptr;
2573 /* Check for Promiscuous and All Multicast modes */
2577 if (netdev->flags & IFF_PROMISC) {
2578 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2579 rctl &= ~E1000_RCTL_VFE;
2581 if (netdev->flags & IFF_ALLMULTI) {
2582 rctl |= E1000_RCTL_MPE;
2583 rctl &= ~E1000_RCTL_UPE;
2585 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2587 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2588 rctl |= E1000_RCTL_VFE;
2593 if (!netdev_mc_empty(netdev)) {
2594 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
2598 /* prepare a packed array of only addresses. */
2600 netdev_for_each_mc_addr(mc_ptr, netdev)
2601 memcpy(mta_list + (i++ * ETH_ALEN),
2602 mc_ptr->dmi_addr, ETH_ALEN);
2604 e1000_update_mc_addr_list(hw, mta_list, i);
2608 * if we're called from probe, we might not have
2609 * anything to do here, so clear out the list
2611 e1000_update_mc_addr_list(hw, NULL, 0);
2616 * e1000_configure - configure the hardware for Rx and Tx
2617 * @adapter: private board structure
2619 static void e1000_configure(struct e1000_adapter *adapter)
2621 e1000_set_multi(adapter->netdev);
2623 e1000_restore_vlan(adapter);
2624 e1000_init_manageability(adapter);
2626 e1000_configure_tx(adapter);
2627 e1000_setup_rctl(adapter);
2628 e1000_configure_rx(adapter);
2629 adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2633 * e1000e_power_up_phy - restore link in case the phy was powered down
2634 * @adapter: address of board private structure
2636 * The phy may be powered down to save power and turn off link when the
2637 * driver is unloaded and wake on lan is not enabled (among others)
2638 * *** this routine MUST be followed by a call to e1000e_reset ***
2640 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2642 if (adapter->hw.phy.ops.power_up)
2643 adapter->hw.phy.ops.power_up(&adapter->hw);
2645 adapter->hw.mac.ops.setup_link(&adapter->hw);
2649 * e1000_power_down_phy - Power down the PHY
2651 * Power down the PHY so no link is implied when interface is down.
2652 * The PHY cannot be powered down if management or WoL is active.
2654 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2656 /* WoL is enabled */
2660 if (adapter->hw.phy.ops.power_down)
2661 adapter->hw.phy.ops.power_down(&adapter->hw);
2665 * e1000e_reset - bring the hardware into a known good state
2667 * This function boots the hardware and enables some settings that
2668 * require a configuration cycle of the hardware - those cannot be
2669 * set/changed during runtime. After reset the device needs to be
2670 * properly configured for Rx, Tx etc.
2672 void e1000e_reset(struct e1000_adapter *adapter)
2674 struct e1000_mac_info *mac = &adapter->hw.mac;
2675 struct e1000_fc_info *fc = &adapter->hw.fc;
2676 struct e1000_hw *hw = &adapter->hw;
2677 u32 tx_space, min_tx_space, min_rx_space;
2678 u32 pba = adapter->pba;
2681 /* reset Packet Buffer Allocation to default */
2684 if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2686 * To maintain wire speed transmits, the Tx FIFO should be
2687 * large enough to accommodate two full transmit packets,
2688 * rounded up to the next 1KB and expressed in KB. Likewise,
2689 * the Rx FIFO should be large enough to accommodate at least
2690 * one full receive packet and is similarly rounded up and
2694 /* upper 16 bits has Tx packet buffer allocation size in KB */
2695 tx_space = pba >> 16;
2696 /* lower 16 bits has Rx packet buffer allocation size in KB */
2699 * the Tx fifo also stores 16 bytes of information about the tx
2700 * but don't include ethernet FCS because hardware appends it
2702 min_tx_space = (adapter->max_frame_size +
2703 sizeof(struct e1000_tx_desc) -
2705 min_tx_space = ALIGN(min_tx_space, 1024);
2706 min_tx_space >>= 10;
2707 /* software strips receive CRC, so leave room for it */
2708 min_rx_space = adapter->max_frame_size;
2709 min_rx_space = ALIGN(min_rx_space, 1024);
2710 min_rx_space >>= 10;
2713 * If current Tx allocation is less than the min Tx FIFO size,
2714 * and the min Tx FIFO size is less than the current Rx FIFO
2715 * allocation, take space away from current Rx allocation
2717 if ((tx_space < min_tx_space) &&
2718 ((min_tx_space - tx_space) < pba)) {
2719 pba -= min_tx_space - tx_space;
2722 * if short on Rx space, Rx wins and must trump tx
2723 * adjustment or use Early Receive if available
2725 if ((pba < min_rx_space) &&
2726 (!(adapter->flags & FLAG_HAS_ERT)))
2727 /* ERT enabled in e1000_configure_rx */
2736 * flow control settings
2738 * The high water mark must be low enough to fit one full frame
2739 * (or the size used for early receive) above it in the Rx FIFO.
2740 * Set it to the lower of:
2741 * - 90% of the Rx FIFO size, and
2742 * - the full Rx FIFO size minus the early receive size (for parts
2743 * with ERT support assuming ERT set to E1000_ERT_2048), or
2744 * - the full Rx FIFO size minus one full frame
2746 if (hw->mac.type == e1000_pchlan) {
2748 * Workaround PCH LOM adapter hangs with certain network
2749 * loads. If hangs persist, try disabling Tx flow control.
2751 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2752 fc->high_water = 0x3500;
2753 fc->low_water = 0x1500;
2755 fc->high_water = 0x5000;
2756 fc->low_water = 0x3000;
2759 if ((adapter->flags & FLAG_HAS_ERT) &&
2760 (adapter->netdev->mtu > ETH_DATA_LEN))
2761 hwm = min(((pba << 10) * 9 / 10),
2762 ((pba << 10) - (E1000_ERT_2048 << 3)));
2764 hwm = min(((pba << 10) * 9 / 10),
2765 ((pba << 10) - adapter->max_frame_size));
2767 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
2768 fc->low_water = fc->high_water - 8;
2771 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2772 fc->pause_time = 0xFFFF;
2774 fc->pause_time = E1000_FC_PAUSE_TIME;
2776 fc->current_mode = fc->requested_mode;
2778 /* Allow time for pending master requests to run */
2779 mac->ops.reset_hw(hw);
2782 * For parts with AMT enabled, let the firmware know
2783 * that the network interface is in control
2785 if (adapter->flags & FLAG_HAS_AMT)
2786 e1000_get_hw_control(adapter);
2789 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP)
2790 e1e_wphy(&adapter->hw, BM_WUC, 0);
2792 if (mac->ops.init_hw(hw))
2793 e_err("Hardware Error\n");
2795 /* additional part of the flow-control workaround above */
2796 if (hw->mac.type == e1000_pchlan)
2797 ew32(FCRTV_PCH, 0x1000);
2799 e1000_update_mng_vlan(adapter);
2801 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2802 ew32(VET, ETH_P_8021Q);
2804 e1000e_reset_adaptive(hw);
2805 e1000_get_phy_info(hw);
2807 if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
2808 !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2811 * speed up time to link by disabling smart power down, ignore
2812 * the return value of this function because there is nothing
2813 * different we would do if it failed
2815 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2816 phy_data &= ~IGP02E1000_PM_SPD;
2817 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2821 int e1000e_up(struct e1000_adapter *adapter)
2823 struct e1000_hw *hw = &adapter->hw;
2825 /* DMA latency requirement to workaround early-receive/jumbo issue */
2826 if (adapter->flags & FLAG_HAS_ERT)
2827 pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY,
2828 adapter->netdev->name,
2829 PM_QOS_DEFAULT_VALUE);
2831 /* hardware has been reset, we need to reload some things */
2832 e1000_configure(adapter);
2834 clear_bit(__E1000_DOWN, &adapter->state);
2836 napi_enable(&adapter->napi);
2837 if (adapter->msix_entries)
2838 e1000_configure_msix(adapter);
2839 e1000_irq_enable(adapter);
2841 netif_wake_queue(adapter->netdev);
2843 /* fire a link change interrupt to start the watchdog */
2844 ew32(ICS, E1000_ICS_LSC);
2848 void e1000e_down(struct e1000_adapter *adapter)
2850 struct net_device *netdev = adapter->netdev;
2851 struct e1000_hw *hw = &adapter->hw;
2855 * signal that we're down so the interrupt handler does not
2856 * reschedule our watchdog timer
2858 set_bit(__E1000_DOWN, &adapter->state);
2860 /* disable receives in the hardware */
2862 ew32(RCTL, rctl & ~E1000_RCTL_EN);
2863 /* flush and sleep below */
2865 netif_stop_queue(netdev);
2867 /* disable transmits in the hardware */
2869 tctl &= ~E1000_TCTL_EN;
2871 /* flush both disables and wait for them to finish */
2875 napi_disable(&adapter->napi);
2876 e1000_irq_disable(adapter);
2878 del_timer_sync(&adapter->watchdog_timer);
2879 del_timer_sync(&adapter->phy_info_timer);
2881 netif_carrier_off(netdev);
2882 adapter->link_speed = 0;
2883 adapter->link_duplex = 0;
2885 if (!pci_channel_offline(adapter->pdev))
2886 e1000e_reset(adapter);
2887 e1000_clean_tx_ring(adapter);
2888 e1000_clean_rx_ring(adapter);
2890 if (adapter->flags & FLAG_HAS_ERT)
2891 pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY,
2892 adapter->netdev->name);
2895 * TODO: for power management, we could drop the link and
2896 * pci_disable_device here.
2900 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2903 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2905 e1000e_down(adapter);
2907 clear_bit(__E1000_RESETTING, &adapter->state);
2911 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2912 * @adapter: board private structure to initialize
2914 * e1000_sw_init initializes the Adapter private data structure.
2915 * Fields are initialized based on PCI device information and
2916 * OS network device settings (MTU size).
2918 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2920 struct net_device *netdev = adapter->netdev;
2922 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2923 adapter->rx_ps_bsize0 = 128;
2924 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2925 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2927 e1000e_set_interrupt_capability(adapter);
2929 if (e1000_alloc_queues(adapter))
2932 /* Explicitly disable IRQ since the NIC can be in any state. */
2933 e1000_irq_disable(adapter);
2935 set_bit(__E1000_DOWN, &adapter->state);
2940 * e1000_intr_msi_test - Interrupt Handler
2941 * @irq: interrupt number
2942 * @data: pointer to a network interface device structure
2944 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2946 struct net_device *netdev = data;
2947 struct e1000_adapter *adapter = netdev_priv(netdev);
2948 struct e1000_hw *hw = &adapter->hw;
2949 u32 icr = er32(ICR);
2951 e_dbg("icr is %08X\n", icr);
2952 if (icr & E1000_ICR_RXSEQ) {
2953 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2961 * e1000_test_msi_interrupt - Returns 0 for successful test
2962 * @adapter: board private struct
2964 * code flow taken from tg3.c
2966 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2968 struct net_device *netdev = adapter->netdev;
2969 struct e1000_hw *hw = &adapter->hw;
2972 /* poll_enable hasn't been called yet, so don't need disable */
2973 /* clear any pending events */
2976 /* free the real vector and request a test handler */
2977 e1000_free_irq(adapter);
2978 e1000e_reset_interrupt_capability(adapter);
2980 /* Assume that the test fails, if it succeeds then the test
2981 * MSI irq handler will unset this flag */
2982 adapter->flags |= FLAG_MSI_TEST_FAILED;
2984 err = pci_enable_msi(adapter->pdev);
2986 goto msi_test_failed;
2988 err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
2989 netdev->name, netdev);
2991 pci_disable_msi(adapter->pdev);
2992 goto msi_test_failed;
2997 e1000_irq_enable(adapter);
2999 /* fire an unusual interrupt on the test handler */
3000 ew32(ICS, E1000_ICS_RXSEQ);
3004 e1000_irq_disable(adapter);
3008 if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3009 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3011 e_info("MSI interrupt test failed!\n");
3014 free_irq(adapter->pdev->irq, netdev);
3015 pci_disable_msi(adapter->pdev);
3018 goto msi_test_failed;
3020 /* okay so the test worked, restore settings */
3021 e_dbg("MSI interrupt test succeeded!\n");
3023 e1000e_set_interrupt_capability(adapter);
3024 e1000_request_irq(adapter);
3029 * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3030 * @adapter: board private struct
3032 * code flow taken from tg3.c, called with e1000 interrupts disabled.
3034 static int e1000_test_msi(struct e1000_adapter *adapter)
3039 if (!(adapter->flags & FLAG_MSI_ENABLED))
3042 /* disable SERR in case the MSI write causes a master abort */
3043 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3044 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3045 pci_cmd & ~PCI_COMMAND_SERR);
3047 err = e1000_test_msi_interrupt(adapter);
3049 /* restore previous setting of command word */
3050 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3056 /* EIO means MSI test failed */
3060 /* back to INTx mode */
3061 e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3063 e1000_free_irq(adapter);
3065 err = e1000_request_irq(adapter);
3071 * e1000_open - Called when a network interface is made active
3072 * @netdev: network interface device structure
3074 * Returns 0 on success, negative value on failure
3076 * The open entry point is called when a network interface is made
3077 * active by the system (IFF_UP). At this point all resources needed
3078 * for transmit and receive operations are allocated, the interrupt
3079 * handler is registered with the OS, the watchdog timer is started,
3080 * and the stack is notified that the interface is ready.
3082 static int e1000_open(struct net_device *netdev)
3084 struct e1000_adapter *adapter = netdev_priv(netdev);
3085 struct e1000_hw *hw = &adapter->hw;
3088 /* disallow open during test */
3089 if (test_bit(__E1000_TESTING, &adapter->state))
3092 netif_carrier_off(netdev);
3094 /* allocate transmit descriptors */
3095 err = e1000e_setup_tx_resources(adapter);
3099 /* allocate receive descriptors */
3100 err = e1000e_setup_rx_resources(adapter);
3104 e1000e_power_up_phy(adapter);
3106 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3107 if ((adapter->hw.mng_cookie.status &
3108 E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3109 e1000_update_mng_vlan(adapter);
3112 * If AMT is enabled, let the firmware know that the network
3113 * interface is now open
3115 if (adapter->flags & FLAG_HAS_AMT)
3116 e1000_get_hw_control(adapter);
3119 * before we allocate an interrupt, we must be ready to handle it.
3120 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3121 * as soon as we call pci_request_irq, so we have to setup our
3122 * clean_rx handler before we do so.
3124 e1000_configure(adapter);
3126 err = e1000_request_irq(adapter);
3131 * Work around PCIe errata with MSI interrupts causing some chipsets to
3132 * ignore e1000e MSI messages, which means we need to test our MSI
3135 if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3136 err = e1000_test_msi(adapter);
3138 e_err("Interrupt allocation failed\n");
3143 /* From here on the code is the same as e1000e_up() */
3144 clear_bit(__E1000_DOWN, &adapter->state);
3146 napi_enable(&adapter->napi);
3148 e1000_irq_enable(adapter);
3150 netif_start_queue(netdev);
3152 /* fire a link status change interrupt to start the watchdog */
3153 ew32(ICS, E1000_ICS_LSC);
3158 e1000_release_hw_control(adapter);
3159 e1000_power_down_phy(adapter);
3160 e1000e_free_rx_resources(adapter);
3162 e1000e_free_tx_resources(adapter);
3164 e1000e_reset(adapter);
3170 * e1000_close - Disables a network interface
3171 * @netdev: network interface device structure
3173 * Returns 0, this is not allowed to fail
3175 * The close entry point is called when an interface is de-activated
3176 * by the OS. The hardware is still under the drivers control, but
3177 * needs to be disabled. A global MAC reset is issued to stop the
3178 * hardware, and all transmit and receive resources are freed.
3180 static int e1000_close(struct net_device *netdev)
3182 struct e1000_adapter *adapter = netdev_priv(netdev);
3184 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3185 e1000e_down(adapter);
3186 e1000_power_down_phy(adapter);
3187 e1000_free_irq(adapter);
3189 e1000e_free_tx_resources(adapter);
3190 e1000e_free_rx_resources(adapter);
3193 * kill manageability vlan ID if supported, but not if a vlan with
3194 * the same ID is registered on the host OS (let 8021q kill it)
3196 if ((adapter->hw.mng_cookie.status &
3197 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3199 vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3200 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3203 * If AMT is enabled, let the firmware know that the network
3204 * interface is now closed
3206 if (adapter->flags & FLAG_HAS_AMT)
3207 e1000_release_hw_control(adapter);
3212 * e1000_set_mac - Change the Ethernet Address of the NIC
3213 * @netdev: network interface device structure
3214 * @p: pointer to an address structure
3216 * Returns 0 on success, negative on failure
3218 static int e1000_set_mac(struct net_device *netdev, void *p)
3220 struct e1000_adapter *adapter = netdev_priv(netdev);
3221 struct sockaddr *addr = p;
3223 if (!is_valid_ether_addr(addr->sa_data))
3224 return -EADDRNOTAVAIL;
3226 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3227 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3229 e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3231 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3232 /* activate the work around */
3233 e1000e_set_laa_state_82571(&adapter->hw, 1);
3236 * Hold a copy of the LAA in RAR[14] This is done so that
3237 * between the time RAR[0] gets clobbered and the time it
3238 * gets fixed (in e1000_watchdog), the actual LAA is in one
3239 * of the RARs and no incoming packets directed to this port
3240 * are dropped. Eventually the LAA will be in RAR[0] and
3243 e1000e_rar_set(&adapter->hw,
3244 adapter->hw.mac.addr,
3245 adapter->hw.mac.rar_entry_count - 1);
3252 * e1000e_update_phy_task - work thread to update phy
3253 * @work: pointer to our work struct
3255 * this worker thread exists because we must acquire a
3256 * semaphore to read the phy, which we could msleep while
3257 * waiting for it, and we can't msleep in a timer.
3259 static void e1000e_update_phy_task(struct work_struct *work)
3261 struct e1000_adapter *adapter = container_of(work,
3262 struct e1000_adapter, update_phy_task);
3263 e1000_get_phy_info(&adapter->hw);
3267 * Need to wait a few seconds after link up to get diagnostic information from
3270 static void e1000_update_phy_info(unsigned long data)
3272 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3273 schedule_work(&adapter->update_phy_task);
3277 * e1000e_update_stats - Update the board statistics counters
3278 * @adapter: board private structure
3280 void e1000e_update_stats(struct e1000_adapter *adapter)
3282 struct net_device *netdev = adapter->netdev;
3283 struct e1000_hw *hw = &adapter->hw;
3284 struct pci_dev *pdev = adapter->pdev;
3288 * Prevent stats update while adapter is being reset, or if the pci
3289 * connection is down.
3291 if (adapter->link_speed == 0)
3293 if (pci_channel_offline(pdev))
3296 adapter->stats.crcerrs += er32(CRCERRS);
3297 adapter->stats.gprc += er32(GPRC);
3298 adapter->stats.gorc += er32(GORCL);
3299 er32(GORCH); /* Clear gorc */
3300 adapter->stats.bprc += er32(BPRC);
3301 adapter->stats.mprc += er32(MPRC);
3302 adapter->stats.roc += er32(ROC);
3304 adapter->stats.mpc += er32(MPC);
3305 if ((hw->phy.type == e1000_phy_82578) ||
3306 (hw->phy.type == e1000_phy_82577)) {
3307 e1e_rphy(hw, HV_SCC_UPPER, &phy_data);
3308 if (!e1e_rphy(hw, HV_SCC_LOWER, &phy_data))
3309 adapter->stats.scc += phy_data;
3311 e1e_rphy(hw, HV_ECOL_UPPER, &phy_data);
3312 if (!e1e_rphy(hw, HV_ECOL_LOWER, &phy_data))
3313 adapter->stats.ecol += phy_data;
3315 e1e_rphy(hw, HV_MCC_UPPER, &phy_data);
3316 if (!e1e_rphy(hw, HV_MCC_LOWER, &phy_data))
3317 adapter->stats.mcc += phy_data;
3319 e1e_rphy(hw, HV_LATECOL_UPPER, &phy_data);
3320 if (!e1e_rphy(hw, HV_LATECOL_LOWER, &phy_data))
3321 adapter->stats.latecol += phy_data;
3323 e1e_rphy(hw, HV_DC_UPPER, &phy_data);
3324 if (!e1e_rphy(hw, HV_DC_LOWER, &phy_data))
3325 adapter->stats.dc += phy_data;
3327 adapter->stats.scc += er32(SCC);
3328 adapter->stats.ecol += er32(ECOL);
3329 adapter->stats.mcc += er32(MCC);
3330 adapter->stats.latecol += er32(LATECOL);
3331 adapter->stats.dc += er32(DC);
3333 adapter->stats.xonrxc += er32(XONRXC);
3334 adapter->stats.xontxc += er32(XONTXC);
3335 adapter->stats.xoffrxc += er32(XOFFRXC);
3336 adapter->stats.xofftxc += er32(XOFFTXC);
3337 adapter->stats.gptc += er32(GPTC);
3338 adapter->stats.gotc += er32(GOTCL);
3339 er32(GOTCH); /* Clear gotc */
3340 adapter->stats.rnbc += er32(RNBC);
3341 adapter->stats.ruc += er32(RUC);
3343 adapter->stats.mptc += er32(MPTC);
3344 adapter->stats.bptc += er32(BPTC);
3346 /* used for adaptive IFS */
3348 hw->mac.tx_packet_delta = er32(TPT);
3349 adapter->stats.tpt += hw->mac.tx_packet_delta;
3350 if ((hw->phy.type == e1000_phy_82578) ||
3351 (hw->phy.type == e1000_phy_82577)) {
3352 e1e_rphy(hw, HV_COLC_UPPER, &phy_data);
3353 if (!e1e_rphy(hw, HV_COLC_LOWER, &phy_data))
3354 hw->mac.collision_delta = phy_data;
3356 hw->mac.collision_delta = er32(COLC);
3358 adapter->stats.colc += hw->mac.collision_delta;
3360 adapter->stats.algnerrc += er32(ALGNERRC);
3361 adapter->stats.rxerrc += er32(RXERRC);
3362 if ((hw->phy.type == e1000_phy_82578) ||
3363 (hw->phy.type == e1000_phy_82577)) {
3364 e1e_rphy(hw, HV_TNCRS_UPPER, &phy_data);
3365 if (!e1e_rphy(hw, HV_TNCRS_LOWER, &phy_data))
3366 adapter->stats.tncrs += phy_data;
3368 if ((hw->mac.type != e1000_82574) &&
3369 (hw->mac.type != e1000_82583))
3370 adapter->stats.tncrs += er32(TNCRS);
3372 adapter->stats.cexterr += er32(CEXTERR);
3373 adapter->stats.tsctc += er32(TSCTC);
3374 adapter->stats.tsctfc += er32(TSCTFC);
3376 /* Fill out the OS statistics structure */
3377 netdev->stats.multicast = adapter->stats.mprc;
3378 netdev->stats.collisions = adapter->stats.colc;
3383 * RLEC on some newer hardware can be incorrect so build
3384 * our own version based on RUC and ROC
3386 netdev->stats.rx_errors = adapter->stats.rxerrc +
3387 adapter->stats.crcerrs + adapter->stats.algnerrc +
3388 adapter->stats.ruc + adapter->stats.roc +
3389 adapter->stats.cexterr;
3390 netdev->stats.rx_length_errors = adapter->stats.ruc +
3392 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3393 netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3394 netdev->stats.rx_missed_errors = adapter->stats.mpc;
3397 netdev->stats.tx_errors = adapter->stats.ecol +
3398 adapter->stats.latecol;
3399 netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3400 netdev->stats.tx_window_errors = adapter->stats.latecol;
3401 netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3403 /* Tx Dropped needs to be maintained elsewhere */
3405 /* Management Stats */
3406 adapter->stats.mgptc += er32(MGTPTC);
3407 adapter->stats.mgprc += er32(MGTPRC);
3408 adapter->stats.mgpdc += er32(MGTPDC);
3412 * e1000_phy_read_status - Update the PHY register status snapshot
3413 * @adapter: board private structure
3415 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3417 struct e1000_hw *hw = &adapter->hw;
3418 struct e1000_phy_regs *phy = &adapter->phy_regs;
3421 if ((er32(STATUS) & E1000_STATUS_LU) &&
3422 (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3423 ret_val = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3424 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3425 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3426 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3427 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3428 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3429 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3430 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3432 e_warn("Error reading PHY register\n");
3435 * Do not read PHY registers if link is not up
3436 * Set values to typical power-on defaults
3438 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3439 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3440 BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3442 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3443 ADVERTISE_ALL | ADVERTISE_CSMA);
3445 phy->expansion = EXPANSION_ENABLENPAGE;
3446 phy->ctrl1000 = ADVERTISE_1000FULL;
3448 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3452 static void e1000_print_link_info(struct e1000_adapter *adapter)
3454 struct e1000_hw *hw = &adapter->hw;
3455 u32 ctrl = er32(CTRL);
3457 /* Link status message must follow this format for user tools */
3458 printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3459 "Flow Control: %s\n",
3460 adapter->netdev->name,
3461 adapter->link_speed,
3462 (adapter->link_duplex == FULL_DUPLEX) ?
3463 "Full Duplex" : "Half Duplex",
3464 ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3466 ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3467 ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3470 bool e1000e_has_link(struct e1000_adapter *adapter)
3472 struct e1000_hw *hw = &adapter->hw;
3473 bool link_active = 0;
3477 * get_link_status is set on LSC (link status) interrupt or
3478 * Rx sequence error interrupt. get_link_status will stay
3479 * false until the check_for_link establishes link
3480 * for copper adapters ONLY
3482 switch (hw->phy.media_type) {
3483 case e1000_media_type_copper:
3484 if (hw->mac.get_link_status) {
3485 ret_val = hw->mac.ops.check_for_link(hw);
3486 link_active = !hw->mac.get_link_status;
3491 case e1000_media_type_fiber:
3492 ret_val = hw->mac.ops.check_for_link(hw);
3493 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3495 case e1000_media_type_internal_serdes:
3496 ret_val = hw->mac.ops.check_for_link(hw);
3497 link_active = adapter->hw.mac.serdes_has_link;
3500 case e1000_media_type_unknown:
3504 if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3505 (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3506 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3507 e_info("Gigabit has been disabled, downgrading speed\n");
3513 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3515 /* make sure the receive unit is started */
3516 if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3517 (adapter->flags & FLAG_RX_RESTART_NOW)) {
3518 struct e1000_hw *hw = &adapter->hw;
3519 u32 rctl = er32(RCTL);
3520 ew32(RCTL, rctl | E1000_RCTL_EN);
3521 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3526 * e1000_watchdog - Timer Call-back
3527 * @data: pointer to adapter cast into an unsigned long
3529 static void e1000_watchdog(unsigned long data)
3531 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3533 /* Do the rest outside of interrupt context */
3534 schedule_work(&adapter->watchdog_task);
3536 /* TODO: make this use queue_delayed_work() */
3539 static void e1000_watchdog_task(struct work_struct *work)
3541 struct e1000_adapter *adapter = container_of(work,
3542 struct e1000_adapter, watchdog_task);
3543 struct net_device *netdev = adapter->netdev;
3544 struct e1000_mac_info *mac = &adapter->hw.mac;
3545 struct e1000_phy_info *phy = &adapter->hw.phy;
3546 struct e1000_ring *tx_ring = adapter->tx_ring;
3547 struct e1000_hw *hw = &adapter->hw;
3551 link = e1000e_has_link(adapter);
3552 if ((netif_carrier_ok(netdev)) && link) {
3553 e1000e_enable_receives(adapter);
3557 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3558 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3559 e1000_update_mng_vlan(adapter);
3562 if (!netif_carrier_ok(netdev)) {
3564 /* update snapshot of PHY registers on LSC */
3565 e1000_phy_read_status(adapter);
3566 mac->ops.get_link_up_info(&adapter->hw,
3567 &adapter->link_speed,
3568 &adapter->link_duplex);
3569 e1000_print_link_info(adapter);
3571 * On supported PHYs, check for duplex mismatch only
3572 * if link has autonegotiated at 10/100 half
3574 if ((hw->phy.type == e1000_phy_igp_3 ||
3575 hw->phy.type == e1000_phy_bm) &&
3576 (hw->mac.autoneg == true) &&
3577 (adapter->link_speed == SPEED_10 ||
3578 adapter->link_speed == SPEED_100) &&
3579 (adapter->link_duplex == HALF_DUPLEX)) {
3582 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3584 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3585 e_info("Autonegotiated half duplex but"
3586 " link partner cannot autoneg. "
3587 " Try forcing full duplex if "
3588 "link gets many collisions.\n");
3591 /* adjust timeout factor according to speed/duplex */
3592 adapter->tx_timeout_factor = 1;
3593 switch (adapter->link_speed) {
3596 adapter->tx_timeout_factor = 16;
3600 adapter->tx_timeout_factor = 10;
3605 * workaround: re-program speed mode bit after
3608 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3611 tarc0 = er32(TARC(0));
3612 tarc0 &= ~SPEED_MODE_BIT;
3613 ew32(TARC(0), tarc0);
3617 * disable TSO for pcie and 10/100 speeds, to avoid
3618 * some hardware issues
3620 if (!(adapter->flags & FLAG_TSO_FORCE)) {
3621 switch (adapter->link_speed) {
3624 e_info("10/100 speed: disabling TSO\n");
3625 netdev->features &= ~NETIF_F_TSO;
3626 netdev->features &= ~NETIF_F_TSO6;
3629 netdev->features |= NETIF_F_TSO;
3630 netdev->features |= NETIF_F_TSO6;
3639 * enable transmits in the hardware, need to do this
3640 * after setting TARC(0)
3643 tctl |= E1000_TCTL_EN;
3647 * Perform any post-link-up configuration before
3648 * reporting link up.
3650 if (phy->ops.cfg_on_link_up)
3651 phy->ops.cfg_on_link_up(hw);
3653 netif_carrier_on(netdev);
3655 if (!test_bit(__E1000_DOWN, &adapter->state))
3656 mod_timer(&adapter->phy_info_timer,
3657 round_jiffies(jiffies + 2 * HZ));
3660 if (netif_carrier_ok(netdev)) {
3661 adapter->link_speed = 0;
3662 adapter->link_duplex = 0;
3663 /* Link status message must follow this format */
3664 printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
3665 adapter->netdev->name);
3666 netif_carrier_off(netdev);
3667 if (!test_bit(__E1000_DOWN, &adapter->state))
3668 mod_timer(&adapter->phy_info_timer,
3669 round_jiffies(jiffies + 2 * HZ));
3671 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3672 schedule_work(&adapter->reset_task);
3677 e1000e_update_stats(adapter);
3679 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3680 adapter->tpt_old = adapter->stats.tpt;
3681 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3682 adapter->colc_old = adapter->stats.colc;
3684 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3685 adapter->gorc_old = adapter->stats.gorc;
3686 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3687 adapter->gotc_old = adapter->stats.gotc;
3689 e1000e_update_adaptive(&adapter->hw);
3691 if (!netif_carrier_ok(netdev)) {
3692 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3696 * We've lost link, so the controller stops DMA,
3697 * but we've got queued Tx work that's never going
3698 * to get done, so reset controller to flush Tx.
3699 * (Do the reset outside of interrupt context).
3701 adapter->tx_timeout_count++;
3702 schedule_work(&adapter->reset_task);
3703 /* return immediately since reset is imminent */
3708 /* Cause software interrupt to ensure Rx ring is cleaned */
3709 if (adapter->msix_entries)
3710 ew32(ICS, adapter->rx_ring->ims_val);
3712 ew32(ICS, E1000_ICS_RXDMT0);
3714 /* Force detection of hung controller every watchdog period */
3715 adapter->detect_tx_hung = 1;
3718 * With 82571 controllers, LAA may be overwritten due to controller
3719 * reset from the other port. Set the appropriate LAA in RAR[0]
3721 if (e1000e_get_laa_state_82571(hw))
3722 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3724 /* Reset the timer */
3725 if (!test_bit(__E1000_DOWN, &adapter->state))
3726 mod_timer(&adapter->watchdog_timer,
3727 round_jiffies(jiffies + 2 * HZ));
3730 #define E1000_TX_FLAGS_CSUM 0x00000001
3731 #define E1000_TX_FLAGS_VLAN 0x00000002
3732 #define E1000_TX_FLAGS_TSO 0x00000004
3733 #define E1000_TX_FLAGS_IPV4 0x00000008
3734 #define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
3735 #define E1000_TX_FLAGS_VLAN_SHIFT 16
3737 static int e1000_tso(struct e1000_adapter *adapter,
3738 struct sk_buff *skb)
3740 struct e1000_ring *tx_ring = adapter->tx_ring;
3741 struct e1000_context_desc *context_desc;
3742 struct e1000_buffer *buffer_info;
3745 u16 ipcse = 0, tucse, mss;
3746 u8 ipcss, ipcso, tucss, tucso, hdr_len;
3749 if (!skb_is_gso(skb))
3752 if (skb_header_cloned(skb)) {
3753 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3758 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3759 mss = skb_shinfo(skb)->gso_size;
3760 if (skb->protocol == htons(ETH_P_IP)) {
3761 struct iphdr *iph = ip_hdr(skb);
3764 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
3766 cmd_length = E1000_TXD_CMD_IP;
3767 ipcse = skb_transport_offset(skb) - 1;
3768 } else if (skb_is_gso_v6(skb)) {
3769 ipv6_hdr(skb)->payload_len = 0;
3770 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3771 &ipv6_hdr(skb)->daddr,
3775 ipcss = skb_network_offset(skb);
3776 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3777 tucss = skb_transport_offset(skb);
3778 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3781 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3782 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3784 i = tx_ring->next_to_use;
3785 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3786 buffer_info = &tx_ring->buffer_info[i];
3788 context_desc->lower_setup.ip_fields.ipcss = ipcss;
3789 context_desc->lower_setup.ip_fields.ipcso = ipcso;
3790 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
3791 context_desc->upper_setup.tcp_fields.tucss = tucss;
3792 context_desc->upper_setup.tcp_fields.tucso = tucso;
3793 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3794 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
3795 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3796 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3798 buffer_info->time_stamp = jiffies;
3799 buffer_info->next_to_watch = i;
3802 if (i == tx_ring->count)
3804 tx_ring->next_to_use = i;
3809 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3811 struct e1000_ring *tx_ring = adapter->tx_ring;
3812 struct e1000_context_desc *context_desc;
3813 struct e1000_buffer *buffer_info;
3816 u32 cmd_len = E1000_TXD_CMD_DEXT;
3819 if (skb->ip_summed != CHECKSUM_PARTIAL)
3822 if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
3823 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
3825 protocol = skb->protocol;
3828 case cpu_to_be16(ETH_P_IP):
3829 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3830 cmd_len |= E1000_TXD_CMD_TCP;
3832 case cpu_to_be16(ETH_P_IPV6):
3833 /* XXX not handling all IPV6 headers */
3834 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3835 cmd_len |= E1000_TXD_CMD_TCP;
3838 if (unlikely(net_ratelimit()))
3839 e_warn("checksum_partial proto=%x!\n",
3840 be16_to_cpu(protocol));
3844 css = skb_transport_offset(skb);
3846 i = tx_ring->next_to_use;
3847 buffer_info = &tx_ring->buffer_info[i];
3848 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3850 context_desc->lower_setup.ip_config = 0;
3851 context_desc->upper_setup.tcp_fields.tucss = css;
3852 context_desc->upper_setup.tcp_fields.tucso =
3853 css + skb->csum_offset;
3854 context_desc->upper_setup.tcp_fields.tucse = 0;
3855 context_desc->tcp_seg_setup.data = 0;
3856 context_desc->cmd_and_length = cpu_to_le32(cmd_len);
3858 buffer_info->time_stamp = jiffies;
3859 buffer_info->next_to_watch = i;
3862 if (i == tx_ring->count)
3864 tx_ring->next_to_use = i;
3869 #define E1000_MAX_PER_TXD 8192
3870 #define E1000_MAX_TXD_PWR 12
3872 static int e1000_tx_map(struct e1000_adapter *adapter,
3873 struct sk_buff *skb, unsigned int first,
3874 unsigned int max_per_txd, unsigned int nr_frags,
3877 struct e1000_ring *tx_ring = adapter->tx_ring;
3878 struct pci_dev *pdev = adapter->pdev;
3879 struct e1000_buffer *buffer_info;
3880 unsigned int len = skb_headlen(skb);
3881 unsigned int offset = 0, size, count = 0, i;
3884 i = tx_ring->next_to_use;
3887 buffer_info = &tx_ring->buffer_info[i];
3888 size = min(len, max_per_txd);
3890 buffer_info->length = size;
3891 buffer_info->time_stamp = jiffies;
3892 buffer_info->next_to_watch = i;
3893 buffer_info->dma = pci_map_single(pdev, skb->data + offset,
3894 size, PCI_DMA_TODEVICE);
3895 buffer_info->mapped_as_page = false;
3896 if (pci_dma_mapping_error(pdev, buffer_info->dma))
3905 if (i == tx_ring->count)
3910 for (f = 0; f < nr_frags; f++) {
3911 struct skb_frag_struct *frag;
3913 frag = &skb_shinfo(skb)->frags[f];
3915 offset = frag->page_offset;
3919 if (i == tx_ring->count)
3922 buffer_info = &tx_ring->buffer_info[i];
3923 size = min(len, max_per_txd);
3925 buffer_info->length = size;
3926 buffer_info->time_stamp = jiffies;
3927 buffer_info->next_to_watch = i;
3928 buffer_info->dma = pci_map_page(pdev, frag->page,
3931 buffer_info->mapped_as_page = true;
3932 if (pci_dma_mapping_error(pdev, buffer_info->dma))
3941 tx_ring->buffer_info[i].skb = skb;
3942 tx_ring->buffer_info[first].next_to_watch = i;
3947 dev_err(&pdev->dev, "TX DMA map failed\n");
3948 buffer_info->dma = 0;
3954 i += tx_ring->count;
3956 buffer_info = &tx_ring->buffer_info[i];
3957 e1000_put_txbuf(adapter, buffer_info);;
3963 static void e1000_tx_queue(struct e1000_adapter *adapter,
3964 int tx_flags, int count)
3966 struct e1000_ring *tx_ring = adapter->tx_ring;
3967 struct e1000_tx_desc *tx_desc = NULL;
3968 struct e1000_buffer *buffer_info;
3969 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3972 if (tx_flags & E1000_TX_FLAGS_TSO) {
3973 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3975 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3977 if (tx_flags & E1000_TX_FLAGS_IPV4)
3978 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3981 if (tx_flags & E1000_TX_FLAGS_CSUM) {
3982 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3983 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3986 if (tx_flags & E1000_TX_FLAGS_VLAN) {
3987 txd_lower |= E1000_TXD_CMD_VLE;
3988 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3991 i = tx_ring->next_to_use;
3994 buffer_info = &tx_ring->buffer_info[i];
3995 tx_desc = E1000_TX_DESC(*tx_ring, i);
3996 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3997 tx_desc->lower.data =
3998 cpu_to_le32(txd_lower | buffer_info->length);
3999 tx_desc->upper.data = cpu_to_le32(txd_upper);
4002 if (i == tx_ring->count)
4006 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4009 * Force memory writes to complete before letting h/w
4010 * know there are new descriptors to fetch. (Only
4011 * applicable for weak-ordered memory model archs,
4016 tx_ring->next_to_use = i;
4017 writel(i, adapter->hw.hw_addr + tx_ring->tail);
4019 * we need this if more than one processor can write to our tail
4020 * at a time, it synchronizes IO on IA64/Altix systems
4025 #define MINIMUM_DHCP_PACKET_SIZE 282
4026 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4027 struct sk_buff *skb)
4029 struct e1000_hw *hw = &adapter->hw;
4032 if (vlan_tx_tag_present(skb)) {
4033 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4034 (adapter->hw.mng_cookie.status &
4035 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4039 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4042 if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4046 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4049 if (ip->protocol != IPPROTO_UDP)
4052 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4053 if (ntohs(udp->dest) != 67)
4056 offset = (u8 *)udp + 8 - skb->data;
4057 length = skb->len - offset;
4058 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4064 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4066 struct e1000_adapter *adapter = netdev_priv(netdev);
4068 netif_stop_queue(netdev);
4070 * Herbert's original patch had:
4071 * smp_mb__after_netif_stop_queue();
4072 * but since that doesn't exist yet, just open code it.
4077 * We need to check again in a case another CPU has just
4078 * made room available.
4080 if (e1000_desc_unused(adapter->tx_ring) < size)
4084 netif_start_queue(netdev);
4085 ++adapter->restart_queue;
4089 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4091 struct e1000_adapter *adapter = netdev_priv(netdev);
4093 if (e1000_desc_unused(adapter->tx_ring) >= size)
4095 return __e1000_maybe_stop_tx(netdev, size);
4098 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4099 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4100 struct net_device *netdev)
4102 struct e1000_adapter *adapter = netdev_priv(netdev);
4103 struct e1000_ring *tx_ring = adapter->tx_ring;
4105 unsigned int max_per_txd = E1000_MAX_PER_TXD;
4106 unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4107 unsigned int tx_flags = 0;
4108 unsigned int len = skb->len - skb->data_len;
4109 unsigned int nr_frags;
4115 if (test_bit(__E1000_DOWN, &adapter->state)) {
4116 dev_kfree_skb_any(skb);
4117 return NETDEV_TX_OK;
4120 if (skb->len <= 0) {
4121 dev_kfree_skb_any(skb);
4122 return NETDEV_TX_OK;
4125 mss = skb_shinfo(skb)->gso_size;
4127 * The controller does a simple calculation to
4128 * make sure there is enough room in the FIFO before
4129 * initiating the DMA for each buffer. The calc is:
4130 * 4 = ceil(buffer len/mss). To make sure we don't
4131 * overrun the FIFO, adjust the max buffer len if mss
4136 max_per_txd = min(mss << 2, max_per_txd);
4137 max_txd_pwr = fls(max_per_txd) - 1;
4140 * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4141 * points to just header, pull a few bytes of payload from
4142 * frags into skb->data
4144 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4146 * we do this workaround for ES2LAN, but it is un-necessary,
4147 * avoiding it could save a lot of cycles
4149 if (skb->data_len && (hdr_len == len)) {
4150 unsigned int pull_size;
4152 pull_size = min((unsigned int)4, skb->data_len);
4153 if (!__pskb_pull_tail(skb, pull_size)) {
4154 e_err("__pskb_pull_tail failed.\n");
4155 dev_kfree_skb_any(skb);
4156 return NETDEV_TX_OK;
4158 len = skb->len - skb->data_len;
4162 /* reserve a descriptor for the offload context */
4163 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4167 count += TXD_USE_COUNT(len, max_txd_pwr);
4169 nr_frags = skb_shinfo(skb)->nr_frags;
4170 for (f = 0; f < nr_frags; f++)
4171 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4174 if (adapter->hw.mac.tx_pkt_filtering)
4175 e1000_transfer_dhcp_info(adapter, skb);
4178 * need: count + 2 desc gap to keep tail from touching
4179 * head, otherwise try next time
4181 if (e1000_maybe_stop_tx(netdev, count + 2))
4182 return NETDEV_TX_BUSY;
4184 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4185 tx_flags |= E1000_TX_FLAGS_VLAN;
4186 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4189 first = tx_ring->next_to_use;
4191 tso = e1000_tso(adapter, skb);
4193 dev_kfree_skb_any(skb);
4194 return NETDEV_TX_OK;
4198 tx_flags |= E1000_TX_FLAGS_TSO;
4199 else if (e1000_tx_csum(adapter, skb))
4200 tx_flags |= E1000_TX_FLAGS_CSUM;
4203 * Old method was to assume IPv4 packet by default if TSO was enabled.
4204 * 82571 hardware supports TSO capabilities for IPv6 as well...
4205 * no longer assume, we must.
4207 if (skb->protocol == htons(ETH_P_IP))
4208 tx_flags |= E1000_TX_FLAGS_IPV4;
4210 /* if count is 0 then mapping error has occured */
4211 count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4213 e1000_tx_queue(adapter, tx_flags, count);
4214 /* Make sure there is space in the ring for the next send. */
4215 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4218 dev_kfree_skb_any(skb);
4219 tx_ring->buffer_info[first].time_stamp = 0;
4220 tx_ring->next_to_use = first;
4223 return NETDEV_TX_OK;
4227 * e1000_tx_timeout - Respond to a Tx Hang
4228 * @netdev: network interface device structure
4230 static void e1000_tx_timeout(struct net_device *netdev)
4232 struct e1000_adapter *adapter = netdev_priv(netdev);
4234 /* Do the reset outside of interrupt context */
4235 adapter->tx_timeout_count++;
4236 schedule_work(&adapter->reset_task);
4239 static void e1000_reset_task(struct work_struct *work)
4241 struct e1000_adapter *adapter;
4242 adapter = container_of(work, struct e1000_adapter, reset_task);
4244 e1000e_reinit_locked(adapter);
4248 * e1000_get_stats - Get System Network Statistics
4249 * @netdev: network interface device structure
4251 * Returns the address of the device statistics structure.
4252 * The statistics are actually updated from the timer callback.
4254 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4256 /* only return the current stats */
4257 return &netdev->stats;
4261 * e1000_change_mtu - Change the Maximum Transfer Unit
4262 * @netdev: network interface device structure
4263 * @new_mtu: new value for maximum frame size
4265 * Returns 0 on success, negative on failure
4267 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4269 struct e1000_adapter *adapter = netdev_priv(netdev);
4270 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4272 /* Jumbo frame support */
4273 if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4274 !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4275 e_err("Jumbo Frames not supported.\n");
4279 /* Supported frame sizes */
4280 if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4281 (max_frame > adapter->max_hw_frame_size)) {
4282 e_err("Unsupported MTU setting\n");
4286 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4288 /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
4289 adapter->max_frame_size = max_frame;
4290 e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4291 netdev->mtu = new_mtu;
4292 if (netif_running(netdev))
4293 e1000e_down(adapter);
4296 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4297 * means we reserve 2 more, this pushes us to allocate from the next
4299 * i.e. RXBUFFER_2048 --> size-4096 slab
4300 * However with the new *_jumbo_rx* routines, jumbo receives will use
4304 if (max_frame <= 2048)
4305 adapter->rx_buffer_len = 2048;
4307 adapter->rx_buffer_len = 4096;
4309 /* adjust allocation if LPE protects us, and we aren't using SBP */
4310 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4311 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4312 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4315 if (netif_running(netdev))
4318 e1000e_reset(adapter);
4320 clear_bit(__E1000_RESETTING, &adapter->state);
4325 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4328 struct e1000_adapter *adapter = netdev_priv(netdev);
4329 struct mii_ioctl_data *data = if_mii(ifr);
4331 if (adapter->hw.phy.media_type != e1000_media_type_copper)
4336 data->phy_id = adapter->hw.phy.addr;
4339 e1000_phy_read_status(adapter);
4341 switch (data->reg_num & 0x1F) {
4343 data->val_out = adapter->phy_regs.bmcr;
4346 data->val_out = adapter->phy_regs.bmsr;
4349 data->val_out = (adapter->hw.phy.id >> 16);
4352 data->val_out = (adapter->hw.phy.id & 0xFFFF);
4355 data->val_out = adapter->phy_regs.advertise;
4358 data->val_out = adapter->phy_regs.lpa;
4361 data->val_out = adapter->phy_regs.expansion;
4364 data->val_out = adapter->phy_regs.ctrl1000;
4367 data->val_out = adapter->phy_regs.stat1000;
4370 data->val_out = adapter->phy_regs.estatus;
4383 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4389 return e1000_mii_ioctl(netdev, ifr, cmd);
4395 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
4397 struct e1000_hw *hw = &adapter->hw;
4402 /* copy MAC RARs to PHY RARs */
4403 for (i = 0; i < adapter->hw.mac.rar_entry_count; i++) {
4404 mac_reg = er32(RAL(i));
4405 e1e_wphy(hw, BM_RAR_L(i), (u16)(mac_reg & 0xFFFF));
4406 e1e_wphy(hw, BM_RAR_M(i), (u16)((mac_reg >> 16) & 0xFFFF));
4407 mac_reg = er32(RAH(i));
4408 e1e_wphy(hw, BM_RAR_H(i), (u16)(mac_reg & 0xFFFF));
4409 e1e_wphy(hw, BM_RAR_CTRL(i), (u16)((mac_reg >> 16) & 0xFFFF));
4412 /* copy MAC MTA to PHY MTA */
4413 for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
4414 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
4415 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
4416 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
4419 /* configure PHY Rx Control register */
4420 e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
4421 mac_reg = er32(RCTL);
4422 if (mac_reg & E1000_RCTL_UPE)
4423 phy_reg |= BM_RCTL_UPE;
4424 if (mac_reg & E1000_RCTL_MPE)
4425 phy_reg |= BM_RCTL_MPE;
4426 phy_reg &= ~(BM_RCTL_MO_MASK);
4427 if (mac_reg & E1000_RCTL_MO_3)
4428 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
4429 << BM_RCTL_MO_SHIFT);
4430 if (mac_reg & E1000_RCTL_BAM)
4431 phy_reg |= BM_RCTL_BAM;
4432 if (mac_reg & E1000_RCTL_PMCF)
4433 phy_reg |= BM_RCTL_PMCF;
4434 mac_reg = er32(CTRL);
4435 if (mac_reg & E1000_CTRL_RFCE)
4436 phy_reg |= BM_RCTL_RFCE;
4437 e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
4439 /* enable PHY wakeup in MAC register */
4441 ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
4443 /* configure and enable PHY wakeup in PHY registers */
4444 e1e_wphy(&adapter->hw, BM_WUFC, wufc);
4445 e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
4447 /* activate PHY wakeup */
4448 retval = hw->phy.ops.acquire(hw);
4450 e_err("Could not acquire PHY\n");
4453 e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4454 (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
4455 retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
4457 e_err("Could not read PHY page 769\n");
4460 phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
4461 retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
4463 e_err("Could not set PHY Host Wakeup bit\n");
4465 hw->phy.ops.release(hw);
4470 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4472 struct net_device *netdev = pci_get_drvdata(pdev);
4473 struct e1000_adapter *adapter = netdev_priv(netdev);
4474 struct e1000_hw *hw = &adapter->hw;
4475 u32 ctrl, ctrl_ext, rctl, status;
4476 u32 wufc = adapter->wol;
4479 netif_device_detach(netdev);
4481 if (netif_running(netdev)) {
4482 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4483 e1000e_down(adapter);
4484 e1000_free_irq(adapter);
4486 e1000e_reset_interrupt_capability(adapter);
4488 retval = pci_save_state(pdev);
4492 status = er32(STATUS);
4493 if (status & E1000_STATUS_LU)
4494 wufc &= ~E1000_WUFC_LNKC;
4497 e1000_setup_rctl(adapter);
4498 e1000_set_multi(netdev);
4500 /* turn on all-multi mode if wake on multicast is enabled */
4501 if (wufc & E1000_WUFC_MC) {
4503 rctl |= E1000_RCTL_MPE;
4508 /* advertise wake from D3Cold */
4509 #define E1000_CTRL_ADVD3WUC 0x00100000
4510 /* phy power management enable */
4511 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4512 ctrl |= E1000_CTRL_ADVD3WUC;
4513 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
4514 ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
4517 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4518 adapter->hw.phy.media_type ==
4519 e1000_media_type_internal_serdes) {
4520 /* keep the laser running in D3 */
4521 ctrl_ext = er32(CTRL_EXT);
4522 ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
4523 ew32(CTRL_EXT, ctrl_ext);
4526 if (adapter->flags & FLAG_IS_ICH)
4527 e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4529 /* Allow time for pending master requests to run */
4530 e1000e_disable_pcie_master(&adapter->hw);
4532 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4533 /* enable wakeup by the PHY */
4534 retval = e1000_init_phy_wakeup(adapter, wufc);
4538 /* enable wakeup by the MAC */
4540 ew32(WUC, E1000_WUC_PME_EN);
4547 *enable_wake = !!wufc;
4549 /* make sure adapter isn't asleep if manageability is enabled */
4550 if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
4551 (hw->mac.ops.check_mng_mode(hw)))
4552 *enable_wake = true;
4554 if (adapter->hw.phy.type == e1000_phy_igp_3)
4555 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4558 * Release control of h/w to f/w. If f/w is AMT enabled, this
4559 * would have already happened in close and is redundant.
4561 e1000_release_hw_control(adapter);
4563 pci_disable_device(pdev);
4568 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
4570 if (sleep && wake) {
4571 pci_prepare_to_sleep(pdev);
4575 pci_wake_from_d3(pdev, wake);
4576 pci_set_power_state(pdev, PCI_D3hot);
4579 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
4582 struct net_device *netdev = pci_get_drvdata(pdev);
4583 struct e1000_adapter *adapter = netdev_priv(netdev);
4586 * The pci-e switch on some quad port adapters will report a
4587 * correctable error when the MAC transitions from D0 to D3. To
4588 * prevent this we need to mask off the correctable errors on the
4589 * downstream port of the pci-e switch.
4591 if (adapter->flags & FLAG_IS_QUAD_PORT) {
4592 struct pci_dev *us_dev = pdev->bus->self;
4593 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
4596 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
4597 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
4598 (devctl & ~PCI_EXP_DEVCTL_CERE));
4600 e1000_power_off(pdev, sleep, wake);
4602 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
4604 e1000_power_off(pdev, sleep, wake);
4608 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4614 * 82573 workaround - disable L1 ASPM on mobile chipsets
4616 * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4617 * resulting in lost data or garbage information on the pci-e link
4618 * level. This could result in (false) bad EEPROM checksum errors,
4619 * long ping times (up to 2s) or even a system freeze/hang.
4621 * Unfortunately this feature saves about 1W power consumption when
4624 pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4625 pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4627 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4629 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4634 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4639 retval = __e1000_shutdown(pdev, &wake);
4641 e1000_complete_shutdown(pdev, true, wake);
4646 static int e1000_resume(struct pci_dev *pdev)
4648 struct net_device *netdev = pci_get_drvdata(pdev);
4649 struct e1000_adapter *adapter = netdev_priv(netdev);
4650 struct e1000_hw *hw = &adapter->hw;
4653 pci_set_power_state(pdev, PCI_D0);
4654 pci_restore_state(pdev);
4655 pci_save_state(pdev);
4656 e1000e_disable_l1aspm(pdev);
4658 err = pci_enable_device_mem(pdev);
4661 "Cannot enable PCI device from suspend\n");
4665 pci_set_master(pdev);
4667 pci_enable_wake(pdev, PCI_D3hot, 0);
4668 pci_enable_wake(pdev, PCI_D3cold, 0);
4670 e1000e_set_interrupt_capability(adapter);
4671 if (netif_running(netdev)) {
4672 err = e1000_request_irq(adapter);
4677 e1000e_power_up_phy(adapter);
4679 /* report the system wakeup cause from S3/S4 */
4680 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4683 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
4685 e_info("PHY Wakeup cause - %s\n",
4686 phy_data & E1000_WUS_EX ? "Unicast Packet" :
4687 phy_data & E1000_WUS_MC ? "Multicast Packet" :
4688 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
4689 phy_data & E1000_WUS_MAG ? "Magic Packet" :
4690 phy_data & E1000_WUS_LNKC ? "Link Status "
4691 " Change" : "other");
4693 e1e_wphy(&adapter->hw, BM_WUS, ~0);
4695 u32 wus = er32(WUS);
4697 e_info("MAC Wakeup cause - %s\n",
4698 wus & E1000_WUS_EX ? "Unicast Packet" :
4699 wus & E1000_WUS_MC ? "Multicast Packet" :
4700 wus & E1000_WUS_BC ? "Broadcast Packet" :
4701 wus & E1000_WUS_MAG ? "Magic Packet" :
4702 wus & E1000_WUS_LNKC ? "Link Status Change" :
4708 e1000e_reset(adapter);
4710 e1000_init_manageability(adapter);
4712 if (netif_running(netdev))
4715 netif_device_attach(netdev);
4718 * If the controller has AMT, do not set DRV_LOAD until the interface
4719 * is up. For all other cases, let the f/w know that the h/w is now
4720 * under the control of the driver.
4722 if (!(adapter->flags & FLAG_HAS_AMT))
4723 e1000_get_hw_control(adapter);
4729 static void e1000_shutdown(struct pci_dev *pdev)
4733 __e1000_shutdown(pdev, &wake);
4735 if (system_state == SYSTEM_POWER_OFF)
4736 e1000_complete_shutdown(pdev, false, wake);
4739 #ifdef CONFIG_NET_POLL_CONTROLLER
4741 * Polling 'interrupt' - used by things like netconsole to send skbs
4742 * without having to re-enable interrupts. It's not called while
4743 * the interrupt routine is executing.
4745 static void e1000_netpoll(struct net_device *netdev)
4747 struct e1000_adapter *adapter = netdev_priv(netdev);
4749 disable_irq(adapter->pdev->irq);
4750 e1000_intr(adapter->pdev->irq, netdev);
4752 enable_irq(adapter->pdev->irq);
4757 * e1000_io_error_detected - called when PCI error is detected
4758 * @pdev: Pointer to PCI device
4759 * @state: The current pci connection state
4761 * This function is called after a PCI bus error affecting
4762 * this device has been detected.
4764 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4765 pci_channel_state_t state)
4767 struct net_device *netdev = pci_get_drvdata(pdev);
4768 struct e1000_adapter *adapter = netdev_priv(netdev);
4770 netif_device_detach(netdev);
4772 if (state == pci_channel_io_perm_failure)
4773 return PCI_ERS_RESULT_DISCONNECT;
4775 if (netif_running(netdev))
4776 e1000e_down(adapter);
4777 pci_disable_device(pdev);
4779 /* Request a slot slot reset. */
4780 return PCI_ERS_RESULT_NEED_RESET;
4784 * e1000_io_slot_reset - called after the pci bus has been reset.
4785 * @pdev: Pointer to PCI device
4787 * Restart the card from scratch, as if from a cold-boot. Implementation
4788 * resembles the first-half of the e1000_resume routine.
4790 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4792 struct net_device *netdev = pci_get_drvdata(pdev);
4793 struct e1000_adapter *adapter = netdev_priv(netdev);
4794 struct e1000_hw *hw = &adapter->hw;
4796 pci_ers_result_t result;
4798 e1000e_disable_l1aspm(pdev);
4799 err = pci_enable_device_mem(pdev);
4802 "Cannot re-enable PCI device after reset.\n");
4803 result = PCI_ERS_RESULT_DISCONNECT;
4805 pci_set_master(pdev);
4806 pci_restore_state(pdev);
4807 pci_save_state(pdev);
4809 pci_enable_wake(pdev, PCI_D3hot, 0);
4810 pci_enable_wake(pdev, PCI_D3cold, 0);
4812 e1000e_reset(adapter);
4814 result = PCI_ERS_RESULT_RECOVERED;
4817 pci_cleanup_aer_uncorrect_error_status(pdev);
4823 * e1000_io_resume - called when traffic can start flowing again.
4824 * @pdev: Pointer to PCI device
4826 * This callback is called when the error recovery driver tells us that
4827 * its OK to resume normal operation. Implementation resembles the
4828 * second-half of the e1000_resume routine.
4830 static void e1000_io_resume(struct pci_dev *pdev)
4832 struct net_device *netdev = pci_get_drvdata(pdev);
4833 struct e1000_adapter *adapter = netdev_priv(netdev);
4835 e1000_init_manageability(adapter);
4837 if (netif_running(netdev)) {
4838 if (e1000e_up(adapter)) {
4840 "can't bring device back up after reset\n");
4845 netif_device_attach(netdev);
4848 * If the controller has AMT, do not set DRV_LOAD until the interface
4849 * is up. For all other cases, let the f/w know that the h/w is now
4850 * under the control of the driver.
4852 if (!(adapter->flags & FLAG_HAS_AMT))
4853 e1000_get_hw_control(adapter);
4857 static void e1000_print_device_info(struct e1000_adapter *adapter)
4859 struct e1000_hw *hw = &adapter->hw;
4860 struct net_device *netdev = adapter->netdev;
4863 /* print bus type/speed/width info */
4864 e_info("(PCI Express:2.5GB/s:%s) %pM\n",
4866 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4870 e_info("Intel(R) PRO/%s Network Connection\n",
4871 (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4872 e1000e_read_pba_num(hw, &pba_num);
4873 e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4874 hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4877 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4879 struct e1000_hw *hw = &adapter->hw;
4883 if (hw->mac.type != e1000_82573)
4886 ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4887 if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
4888 /* Deep Smart Power Down (DSPD) */
4889 dev_warn(&adapter->pdev->dev,
4890 "Warning: detected DSPD enabled in EEPROM\n");
4893 ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4894 if (!ret_val && (le16_to_cpu(buf) & (3 << 2))) {
4896 dev_warn(&adapter->pdev->dev,
4897 "Warning: detected ASPM enabled in EEPROM\n");
4901 static const struct net_device_ops e1000e_netdev_ops = {
4902 .ndo_open = e1000_open,
4903 .ndo_stop = e1000_close,
4904 .ndo_start_xmit = e1000_xmit_frame,
4905 .ndo_get_stats = e1000_get_stats,
4906 .ndo_set_multicast_list = e1000_set_multi,
4907 .ndo_set_mac_address = e1000_set_mac,
4908 .ndo_change_mtu = e1000_change_mtu,
4909 .ndo_do_ioctl = e1000_ioctl,
4910 .ndo_tx_timeout = e1000_tx_timeout,
4911 .ndo_validate_addr = eth_validate_addr,
4913 .ndo_vlan_rx_register = e1000_vlan_rx_register,
4914 .ndo_vlan_rx_add_vid = e1000_vlan_rx_add_vid,
4915 .ndo_vlan_rx_kill_vid = e1000_vlan_rx_kill_vid,
4916 #ifdef CONFIG_NET_POLL_CONTROLLER
4917 .ndo_poll_controller = e1000_netpoll,
4922 * e1000_probe - Device Initialization Routine
4923 * @pdev: PCI device information struct
4924 * @ent: entry in e1000_pci_tbl
4926 * Returns 0 on success, negative on failure
4928 * e1000_probe initializes an adapter identified by a pci_dev structure.
4929 * The OS initialization, configuring of the adapter private structure,
4930 * and a hardware reset occur.
4932 static int __devinit e1000_probe(struct pci_dev *pdev,
4933 const struct pci_device_id *ent)
4935 struct net_device *netdev;
4936 struct e1000_adapter *adapter;
4937 struct e1000_hw *hw;
4938 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4939 resource_size_t mmio_start, mmio_len;
4940 resource_size_t flash_start, flash_len;
4942 static int cards_found;
4943 int i, err, pci_using_dac;
4944 u16 eeprom_data = 0;
4945 u16 eeprom_apme_mask = E1000_EEPROM_APME;
4947 e1000e_disable_l1aspm(pdev);
4949 err = pci_enable_device_mem(pdev);
4954 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
4956 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
4960 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
4962 err = pci_set_consistent_dma_mask(pdev,
4965 dev_err(&pdev->dev, "No usable DMA "
4966 "configuration, aborting\n");
4972 err = pci_request_selected_regions_exclusive(pdev,
4973 pci_select_bars(pdev, IORESOURCE_MEM),
4974 e1000e_driver_name);
4978 /* AER (Advanced Error Reporting) hooks */
4979 pci_enable_pcie_error_reporting(pdev);
4981 pci_set_master(pdev);
4982 /* PCI config space info */
4983 err = pci_save_state(pdev);
4985 goto err_alloc_etherdev;
4988 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4990 goto err_alloc_etherdev;
4992 SET_NETDEV_DEV(netdev, &pdev->dev);
4994 pci_set_drvdata(pdev, netdev);
4995 adapter = netdev_priv(netdev);
4997 adapter->netdev = netdev;
4998 adapter->pdev = pdev;
5000 adapter->pba = ei->pba;
5001 adapter->flags = ei->flags;
5002 adapter->flags2 = ei->flags2;
5003 adapter->hw.adapter = adapter;
5004 adapter->hw.mac.type = ei->mac;
5005 adapter->max_hw_frame_size = ei->max_hw_frame_size;
5006 adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5008 mmio_start = pci_resource_start(pdev, 0);
5009 mmio_len = pci_resource_len(pdev, 0);
5012 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5013 if (!adapter->hw.hw_addr)
5016 if ((adapter->flags & FLAG_HAS_FLASH) &&
5017 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5018 flash_start = pci_resource_start(pdev, 1);
5019 flash_len = pci_resource_len(pdev, 1);
5020 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5021 if (!adapter->hw.flash_address)
5025 /* construct the net_device struct */
5026 netdev->netdev_ops = &e1000e_netdev_ops;
5027 e1000e_set_ethtool_ops(netdev);
5028 netdev->watchdog_timeo = 5 * HZ;
5029 netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5030 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5032 netdev->mem_start = mmio_start;
5033 netdev->mem_end = mmio_start + mmio_len;
5035 adapter->bd_number = cards_found++;
5037 e1000e_check_options(adapter);
5039 /* setup adapter struct */
5040 err = e1000_sw_init(adapter);
5046 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5047 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5048 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5050 err = ei->get_variants(adapter);
5054 if ((adapter->flags & FLAG_IS_ICH) &&
5055 (adapter->flags & FLAG_READ_ONLY_NVM))
5056 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5058 hw->mac.ops.get_bus_info(&adapter->hw);
5060 adapter->hw.phy.autoneg_wait_to_complete = 0;
5062 /* Copper options */
5063 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5064 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5065 adapter->hw.phy.disable_polarity_correction = 0;
5066 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5069 if (e1000_check_reset_block(&adapter->hw))
5070 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5072 netdev->features = NETIF_F_SG |
5074 NETIF_F_HW_VLAN_TX |
5077 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5078 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5080 netdev->features |= NETIF_F_TSO;
5081 netdev->features |= NETIF_F_TSO6;
5083 netdev->vlan_features |= NETIF_F_TSO;
5084 netdev->vlan_features |= NETIF_F_TSO6;
5085 netdev->vlan_features |= NETIF_F_HW_CSUM;
5086 netdev->vlan_features |= NETIF_F_SG;
5089 netdev->features |= NETIF_F_HIGHDMA;
5091 if (e1000e_enable_mng_pass_thru(&adapter->hw))
5092 adapter->flags |= FLAG_MNG_PT_ENABLED;
5095 * before reading the NVM, reset the controller to
5096 * put the device in a known good starting state
5098 adapter->hw.mac.ops.reset_hw(&adapter->hw);
5101 * systems with ASPM and others may see the checksum fail on the first
5102 * attempt. Let's give it a few tries
5105 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5108 e_err("The NVM Checksum Is Not Valid\n");
5114 e1000_eeprom_checks(adapter);
5116 /* copy the MAC address */
5117 if (e1000e_read_mac_addr(&adapter->hw))
5118 e_err("NVM Read Error while reading MAC address\n");
5120 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5121 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5123 if (!is_valid_ether_addr(netdev->perm_addr)) {
5124 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5129 init_timer(&adapter->watchdog_timer);
5130 adapter->watchdog_timer.function = &e1000_watchdog;
5131 adapter->watchdog_timer.data = (unsigned long) adapter;
5133 init_timer(&adapter->phy_info_timer);
5134 adapter->phy_info_timer.function = &e1000_update_phy_info;
5135 adapter->phy_info_timer.data = (unsigned long) adapter;
5137 INIT_WORK(&adapter->reset_task, e1000_reset_task);
5138 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5139 INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5140 INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5141 INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5143 /* Initialize link parameters. User can change them with ethtool */
5144 adapter->hw.mac.autoneg = 1;
5145 adapter->fc_autoneg = 1;
5146 adapter->hw.fc.requested_mode = e1000_fc_default;
5147 adapter->hw.fc.current_mode = e1000_fc_default;
5148 adapter->hw.phy.autoneg_advertised = 0x2f;
5150 /* ring size defaults */
5151 adapter->rx_ring->count = 256;
5152 adapter->tx_ring->count = 256;
5155 * Initial Wake on LAN setting - If APM wake is enabled in
5156 * the EEPROM, enable the ACPI Magic Packet filter
5158 if (adapter->flags & FLAG_APME_IN_WUC) {
5159 /* APME bit in EEPROM is mapped to WUC.APME */
5160 eeprom_data = er32(WUC);
5161 eeprom_apme_mask = E1000_WUC_APME;
5162 if (eeprom_data & E1000_WUC_PHY_WAKE)
5163 adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5164 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5165 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5166 (adapter->hw.bus.func == 1))
5167 e1000_read_nvm(&adapter->hw,
5168 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5170 e1000_read_nvm(&adapter->hw,
5171 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5174 /* fetch WoL from EEPROM */
5175 if (eeprom_data & eeprom_apme_mask)
5176 adapter->eeprom_wol |= E1000_WUFC_MAG;
5179 * now that we have the eeprom settings, apply the special cases
5180 * where the eeprom may be wrong or the board simply won't support
5181 * wake on lan on a particular port
5183 if (!(adapter->flags & FLAG_HAS_WOL))
5184 adapter->eeprom_wol = 0;
5186 /* initialize the wol settings based on the eeprom settings */
5187 adapter->wol = adapter->eeprom_wol;
5188 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5190 /* save off EEPROM version number */
5191 e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5193 /* reset the hardware with the new settings */
5194 e1000e_reset(adapter);
5197 * If the controller has AMT, do not set DRV_LOAD until the interface
5198 * is up. For all other cases, let the f/w know that the h/w is now
5199 * under the control of the driver.
5201 if (!(adapter->flags & FLAG_HAS_AMT))
5202 e1000_get_hw_control(adapter);
5204 strcpy(netdev->name, "eth%d");
5205 err = register_netdev(netdev);
5209 /* carrier off reporting is important to ethtool even BEFORE open */
5210 netif_carrier_off(netdev);
5212 e1000_print_device_info(adapter);
5217 if (!(adapter->flags & FLAG_HAS_AMT))
5218 e1000_release_hw_control(adapter);
5220 if (!e1000_check_reset_block(&adapter->hw))
5221 e1000_phy_hw_reset(&adapter->hw);
5224 kfree(adapter->tx_ring);
5225 kfree(adapter->rx_ring);
5227 if (adapter->hw.flash_address)
5228 iounmap(adapter->hw.flash_address);
5229 e1000e_reset_interrupt_capability(adapter);
5231 iounmap(adapter->hw.hw_addr);
5233 free_netdev(netdev);
5235 pci_release_selected_regions(pdev,
5236 pci_select_bars(pdev, IORESOURCE_MEM));
5239 pci_disable_device(pdev);
5244 * e1000_remove - Device Removal Routine
5245 * @pdev: PCI device information struct
5247 * e1000_remove is called by the PCI subsystem to alert the driver
5248 * that it should release a PCI device. The could be caused by a
5249 * Hot-Plug event, or because the driver is going to be removed from
5252 static void __devexit e1000_remove(struct pci_dev *pdev)
5254 struct net_device *netdev = pci_get_drvdata(pdev);
5255 struct e1000_adapter *adapter = netdev_priv(netdev);
5258 * flush_scheduled work may reschedule our watchdog task, so
5259 * explicitly disable watchdog tasks from being rescheduled
5261 set_bit(__E1000_DOWN, &adapter->state);
5262 del_timer_sync(&adapter->watchdog_timer);
5263 del_timer_sync(&adapter->phy_info_timer);
5265 cancel_work_sync(&adapter->reset_task);
5266 cancel_work_sync(&adapter->watchdog_task);
5267 cancel_work_sync(&adapter->downshift_task);
5268 cancel_work_sync(&adapter->update_phy_task);
5269 cancel_work_sync(&adapter->print_hang_task);
5270 flush_scheduled_work();
5272 if (!(netdev->flags & IFF_UP))
5273 e1000_power_down_phy(adapter);
5275 unregister_netdev(netdev);
5278 * Release control of h/w to f/w. If f/w is AMT enabled, this
5279 * would have already happened in close and is redundant.
5281 e1000_release_hw_control(adapter);
5283 e1000e_reset_interrupt_capability(adapter);
5284 kfree(adapter->tx_ring);
5285 kfree(adapter->rx_ring);
5287 iounmap(adapter->hw.hw_addr);
5288 if (adapter->hw.flash_address)
5289 iounmap(adapter->hw.flash_address);
5290 pci_release_selected_regions(pdev,
5291 pci_select_bars(pdev, IORESOURCE_MEM));
5293 free_netdev(netdev);
5296 pci_disable_pcie_error_reporting(pdev);
5298 pci_disable_device(pdev);
5301 /* PCI Error Recovery (ERS) */
5302 static struct pci_error_handlers e1000_err_handler = {
5303 .error_detected = e1000_io_error_detected,
5304 .slot_reset = e1000_io_slot_reset,
5305 .resume = e1000_io_resume,
5308 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
5309 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5310 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5311 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5312 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5313 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5314 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5315 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5316 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5317 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5319 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5320 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5321 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5322 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5324 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5325 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5326 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5328 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5329 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
5330 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
5332 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5333 board_80003es2lan },
5334 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5335 board_80003es2lan },
5336 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5337 board_80003es2lan },
5338 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5339 board_80003es2lan },
5341 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5342 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5343 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5344 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5345 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5346 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5347 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5348 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
5350 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5351 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5352 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5353 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5354 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5355 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5356 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5357 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5358 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5360 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5361 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5362 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5364 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5365 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5367 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
5368 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
5369 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
5370 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
5372 { } /* terminate list */
5374 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5376 /* PCI Device API Driver */
5377 static struct pci_driver e1000_driver = {
5378 .name = e1000e_driver_name,
5379 .id_table = e1000_pci_tbl,
5380 .probe = e1000_probe,
5381 .remove = __devexit_p(e1000_remove),
5383 /* Power Management Hooks */
5384 .suspend = e1000_suspend,
5385 .resume = e1000_resume,
5387 .shutdown = e1000_shutdown,
5388 .err_handler = &e1000_err_handler
5392 * e1000_init_module - Driver Registration Routine
5394 * e1000_init_module is the first routine called when the driver is
5395 * loaded. All it does is register with the PCI subsystem.
5397 static int __init e1000_init_module(void)
5400 printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
5401 e1000e_driver_name, e1000e_driver_version);
5402 printk(KERN_INFO "%s: Copyright (c) 1999 - 2009 Intel Corporation.\n",
5403 e1000e_driver_name);
5404 ret = pci_register_driver(&e1000_driver);
5408 module_init(e1000_init_module);
5411 * e1000_exit_module - Driver Exit Cleanup Routine
5413 * e1000_exit_module is called just before the driver is removed
5416 static void __exit e1000_exit_module(void)
5418 pci_unregister_driver(&e1000_driver);
5420 module_exit(e1000_exit_module);
5423 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5424 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5425 MODULE_LICENSE("GPL");
5426 MODULE_VERSION(DRV_VERSION);