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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
31 #include <linux/module.h>
32 #include <linux/types.h>
33 #include <linux/init.h>
34 #include <linux/pci.h>
35 #include <linux/vmalloc.h>
36 #include <linux/pagemap.h>
37 #include <linux/delay.h>
38 #include <linux/netdevice.h>
39 #include <linux/tcp.h>
40 #include <linux/ipv6.h>
41 #include <net/checksum.h>
42 #include <net/ip6_checksum.h>
43 #include <linux/mii.h>
44 #include <linux/ethtool.h>
45 #include <linux/if_vlan.h>
46 #include <linux/cpu.h>
47 #include <linux/smp.h>
48 #include <linux/pm_qos_params.h>
49 #include <linux/pm_runtime.h>
50 #include <linux/aer.h>
54 #define DRV_VERSION "1.0.2-k2"
55 char e1000e_driver_name[] = "e1000e";
56 const char e1000e_driver_version[] = DRV_VERSION;
58 static const struct e1000_info *e1000_info_tbl[] = {
59 [board_82571] = &e1000_82571_info,
60 [board_82572] = &e1000_82572_info,
61 [board_82573] = &e1000_82573_info,
62 [board_82574] = &e1000_82574_info,
63 [board_82583] = &e1000_82583_info,
64 [board_80003es2lan] = &e1000_es2_info,
65 [board_ich8lan] = &e1000_ich8_info,
66 [board_ich9lan] = &e1000_ich9_info,
67 [board_ich10lan] = &e1000_ich10_info,
68 [board_pchlan] = &e1000_pch_info,
72 * e1000_desc_unused - calculate if we have unused descriptors
74 static int e1000_desc_unused(struct e1000_ring *ring)
76 if (ring->next_to_clean > ring->next_to_use)
77 return ring->next_to_clean - ring->next_to_use - 1;
79 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
83 * e1000_receive_skb - helper function to handle Rx indications
84 * @adapter: board private structure
85 * @status: descriptor status field as written by hardware
86 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
87 * @skb: pointer to sk_buff to be indicated to stack
89 static void e1000_receive_skb(struct e1000_adapter *adapter,
90 struct net_device *netdev,
92 u8 status, __le16 vlan)
94 skb->protocol = eth_type_trans(skb, netdev);
96 if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
97 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
98 le16_to_cpu(vlan), skb);
100 napi_gro_receive(&adapter->napi, skb);
104 * e1000_rx_checksum - Receive Checksum Offload for 82543
105 * @adapter: board private structure
106 * @status_err: receive descriptor status and error fields
107 * @csum: receive descriptor csum field
108 * @sk_buff: socket buffer with received data
110 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
111 u32 csum, struct sk_buff *skb)
113 u16 status = (u16)status_err;
114 u8 errors = (u8)(status_err >> 24);
115 skb->ip_summed = CHECKSUM_NONE;
117 /* Ignore Checksum bit is set */
118 if (status & E1000_RXD_STAT_IXSM)
120 /* TCP/UDP checksum error bit is set */
121 if (errors & E1000_RXD_ERR_TCPE) {
122 /* let the stack verify checksum errors */
123 adapter->hw_csum_err++;
127 /* TCP/UDP Checksum has not been calculated */
128 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
131 /* It must be a TCP or UDP packet with a valid checksum */
132 if (status & E1000_RXD_STAT_TCPCS) {
133 /* TCP checksum is good */
134 skb->ip_summed = CHECKSUM_UNNECESSARY;
137 * IP fragment with UDP payload
138 * Hardware complements the payload checksum, so we undo it
139 * and then put the value in host order for further stack use.
141 __sum16 sum = (__force __sum16)htons(csum);
142 skb->csum = csum_unfold(~sum);
143 skb->ip_summed = CHECKSUM_COMPLETE;
145 adapter->hw_csum_good++;
149 * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
150 * @adapter: address of board private structure
152 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
155 struct net_device *netdev = adapter->netdev;
156 struct pci_dev *pdev = adapter->pdev;
157 struct e1000_ring *rx_ring = adapter->rx_ring;
158 struct e1000_rx_desc *rx_desc;
159 struct e1000_buffer *buffer_info;
162 unsigned int bufsz = adapter->rx_buffer_len;
164 i = rx_ring->next_to_use;
165 buffer_info = &rx_ring->buffer_info[i];
167 while (cleaned_count--) {
168 skb = buffer_info->skb;
174 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
176 /* Better luck next round */
177 adapter->alloc_rx_buff_failed++;
181 buffer_info->skb = skb;
183 buffer_info->dma = pci_map_single(pdev, skb->data,
184 adapter->rx_buffer_len,
186 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
187 dev_err(&pdev->dev, "RX DMA map failed\n");
188 adapter->rx_dma_failed++;
192 rx_desc = E1000_RX_DESC(*rx_ring, i);
193 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
196 if (i == rx_ring->count)
198 buffer_info = &rx_ring->buffer_info[i];
201 if (rx_ring->next_to_use != i) {
202 rx_ring->next_to_use = i;
204 i = (rx_ring->count - 1);
207 * Force memory writes to complete before letting h/w
208 * know there are new descriptors to fetch. (Only
209 * applicable for weak-ordered memory model archs,
213 writel(i, adapter->hw.hw_addr + rx_ring->tail);
218 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
219 * @adapter: address of board private structure
221 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
224 struct net_device *netdev = adapter->netdev;
225 struct pci_dev *pdev = adapter->pdev;
226 union e1000_rx_desc_packet_split *rx_desc;
227 struct e1000_ring *rx_ring = adapter->rx_ring;
228 struct e1000_buffer *buffer_info;
229 struct e1000_ps_page *ps_page;
233 i = rx_ring->next_to_use;
234 buffer_info = &rx_ring->buffer_info[i];
236 while (cleaned_count--) {
237 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
239 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
240 ps_page = &buffer_info->ps_pages[j];
241 if (j >= adapter->rx_ps_pages) {
242 /* all unused desc entries get hw null ptr */
243 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
246 if (!ps_page->page) {
247 ps_page->page = alloc_page(GFP_ATOMIC);
248 if (!ps_page->page) {
249 adapter->alloc_rx_buff_failed++;
252 ps_page->dma = pci_map_page(pdev,
256 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
257 dev_err(&adapter->pdev->dev,
258 "RX DMA page map failed\n");
259 adapter->rx_dma_failed++;
264 * Refresh the desc even if buffer_addrs
265 * didn't change because each write-back
268 rx_desc->read.buffer_addr[j+1] =
269 cpu_to_le64(ps_page->dma);
272 skb = netdev_alloc_skb_ip_align(netdev,
273 adapter->rx_ps_bsize0);
276 adapter->alloc_rx_buff_failed++;
280 buffer_info->skb = skb;
281 buffer_info->dma = pci_map_single(pdev, skb->data,
282 adapter->rx_ps_bsize0,
284 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
285 dev_err(&pdev->dev, "RX DMA map failed\n");
286 adapter->rx_dma_failed++;
288 dev_kfree_skb_any(skb);
289 buffer_info->skb = NULL;
293 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
296 if (i == rx_ring->count)
298 buffer_info = &rx_ring->buffer_info[i];
302 if (rx_ring->next_to_use != i) {
303 rx_ring->next_to_use = i;
306 i = (rx_ring->count - 1);
309 * Force memory writes to complete before letting h/w
310 * know there are new descriptors to fetch. (Only
311 * applicable for weak-ordered memory model archs,
316 * Hardware increments by 16 bytes, but packet split
317 * descriptors are 32 bytes...so we increment tail
320 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
325 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
326 * @adapter: address of board private structure
327 * @cleaned_count: number of buffers to allocate this pass
330 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
333 struct net_device *netdev = adapter->netdev;
334 struct pci_dev *pdev = adapter->pdev;
335 struct e1000_rx_desc *rx_desc;
336 struct e1000_ring *rx_ring = adapter->rx_ring;
337 struct e1000_buffer *buffer_info;
340 unsigned int bufsz = 256 - 16 /* for skb_reserve */;
342 i = rx_ring->next_to_use;
343 buffer_info = &rx_ring->buffer_info[i];
345 while (cleaned_count--) {
346 skb = buffer_info->skb;
352 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
353 if (unlikely(!skb)) {
354 /* Better luck next round */
355 adapter->alloc_rx_buff_failed++;
359 buffer_info->skb = skb;
361 /* allocate a new page if necessary */
362 if (!buffer_info->page) {
363 buffer_info->page = alloc_page(GFP_ATOMIC);
364 if (unlikely(!buffer_info->page)) {
365 adapter->alloc_rx_buff_failed++;
370 if (!buffer_info->dma)
371 buffer_info->dma = pci_map_page(pdev,
372 buffer_info->page, 0,
376 rx_desc = E1000_RX_DESC(*rx_ring, i);
377 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
379 if (unlikely(++i == rx_ring->count))
381 buffer_info = &rx_ring->buffer_info[i];
384 if (likely(rx_ring->next_to_use != i)) {
385 rx_ring->next_to_use = i;
386 if (unlikely(i-- == 0))
387 i = (rx_ring->count - 1);
389 /* Force memory writes to complete before letting h/w
390 * know there are new descriptors to fetch. (Only
391 * applicable for weak-ordered memory model archs,
394 writel(i, adapter->hw.hw_addr + rx_ring->tail);
399 * e1000_clean_rx_irq - Send received data up the network stack; legacy
400 * @adapter: board private structure
402 * the return value indicates whether actual cleaning was done, there
403 * is no guarantee that everything was cleaned
405 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
406 int *work_done, int work_to_do)
408 struct net_device *netdev = adapter->netdev;
409 struct pci_dev *pdev = adapter->pdev;
410 struct e1000_hw *hw = &adapter->hw;
411 struct e1000_ring *rx_ring = adapter->rx_ring;
412 struct e1000_rx_desc *rx_desc, *next_rxd;
413 struct e1000_buffer *buffer_info, *next_buffer;
416 int cleaned_count = 0;
418 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
420 i = rx_ring->next_to_clean;
421 rx_desc = E1000_RX_DESC(*rx_ring, i);
422 buffer_info = &rx_ring->buffer_info[i];
424 while (rx_desc->status & E1000_RXD_STAT_DD) {
428 if (*work_done >= work_to_do)
432 status = rx_desc->status;
433 skb = buffer_info->skb;
434 buffer_info->skb = NULL;
436 prefetch(skb->data - NET_IP_ALIGN);
439 if (i == rx_ring->count)
441 next_rxd = E1000_RX_DESC(*rx_ring, i);
444 next_buffer = &rx_ring->buffer_info[i];
448 pci_unmap_single(pdev,
450 adapter->rx_buffer_len,
452 buffer_info->dma = 0;
454 length = le16_to_cpu(rx_desc->length);
457 * !EOP means multiple descriptors were used to store a single
458 * packet, if that's the case we need to toss it. In fact, we
459 * need to toss every packet with the EOP bit clear and the
460 * next frame that _does_ have the EOP bit set, as it is by
461 * definition only a frame fragment
463 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
464 adapter->flags2 |= FLAG2_IS_DISCARDING;
466 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
467 /* All receives must fit into a single buffer */
468 e_dbg("Receive packet consumed multiple buffers\n");
470 buffer_info->skb = skb;
471 if (status & E1000_RXD_STAT_EOP)
472 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
476 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
478 buffer_info->skb = skb;
482 /* adjust length to remove Ethernet CRC */
483 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
486 total_rx_bytes += length;
490 * code added for copybreak, this should improve
491 * performance for small packets with large amounts
492 * of reassembly being done in the stack
494 if (length < copybreak) {
495 struct sk_buff *new_skb =
496 netdev_alloc_skb_ip_align(netdev, length);
498 skb_copy_to_linear_data_offset(new_skb,
504 /* save the skb in buffer_info as good */
505 buffer_info->skb = skb;
508 /* else just continue with the old one */
510 /* end copybreak code */
511 skb_put(skb, length);
513 /* Receive Checksum Offload */
514 e1000_rx_checksum(adapter,
516 ((u32)(rx_desc->errors) << 24),
517 le16_to_cpu(rx_desc->csum), skb);
519 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
524 /* return some buffers to hardware, one at a time is too slow */
525 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
526 adapter->alloc_rx_buf(adapter, cleaned_count);
530 /* use prefetched values */
532 buffer_info = next_buffer;
534 rx_ring->next_to_clean = i;
536 cleaned_count = e1000_desc_unused(rx_ring);
538 adapter->alloc_rx_buf(adapter, cleaned_count);
540 adapter->total_rx_bytes += total_rx_bytes;
541 adapter->total_rx_packets += total_rx_packets;
542 netdev->stats.rx_bytes += total_rx_bytes;
543 netdev->stats.rx_packets += total_rx_packets;
547 static void e1000_put_txbuf(struct e1000_adapter *adapter,
548 struct e1000_buffer *buffer_info)
550 if (buffer_info->dma) {
551 if (buffer_info->mapped_as_page)
552 pci_unmap_page(adapter->pdev, buffer_info->dma,
553 buffer_info->length, PCI_DMA_TODEVICE);
555 pci_unmap_single(adapter->pdev, buffer_info->dma,
558 buffer_info->dma = 0;
560 if (buffer_info->skb) {
561 dev_kfree_skb_any(buffer_info->skb);
562 buffer_info->skb = NULL;
564 buffer_info->time_stamp = 0;
567 static void e1000_print_hw_hang(struct work_struct *work)
569 struct e1000_adapter *adapter = container_of(work,
570 struct e1000_adapter,
572 struct e1000_ring *tx_ring = adapter->tx_ring;
573 unsigned int i = tx_ring->next_to_clean;
574 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
575 struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
576 struct e1000_hw *hw = &adapter->hw;
577 u16 phy_status, phy_1000t_status, phy_ext_status;
580 e1e_rphy(hw, PHY_STATUS, &phy_status);
581 e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
582 e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
584 pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
586 /* detected Hardware unit hang */
587 e_err("Detected Hardware Unit Hang:\n"
590 " next_to_use <%x>\n"
591 " next_to_clean <%x>\n"
592 "buffer_info[next_to_clean]:\n"
593 " time_stamp <%lx>\n"
594 " next_to_watch <%x>\n"
596 " next_to_watch.status <%x>\n"
599 "PHY 1000BASE-T Status <%x>\n"
600 "PHY Extended Status <%x>\n"
602 readl(adapter->hw.hw_addr + tx_ring->head),
603 readl(adapter->hw.hw_addr + tx_ring->tail),
604 tx_ring->next_to_use,
605 tx_ring->next_to_clean,
606 tx_ring->buffer_info[eop].time_stamp,
609 eop_desc->upper.fields.status,
618 * e1000_clean_tx_irq - Reclaim resources after transmit completes
619 * @adapter: board private structure
621 * the return value indicates whether actual cleaning was done, there
622 * is no guarantee that everything was cleaned
624 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
626 struct net_device *netdev = adapter->netdev;
627 struct e1000_hw *hw = &adapter->hw;
628 struct e1000_ring *tx_ring = adapter->tx_ring;
629 struct e1000_tx_desc *tx_desc, *eop_desc;
630 struct e1000_buffer *buffer_info;
632 unsigned int count = 0;
633 unsigned int total_tx_bytes = 0, total_tx_packets = 0;
635 i = tx_ring->next_to_clean;
636 eop = tx_ring->buffer_info[i].next_to_watch;
637 eop_desc = E1000_TX_DESC(*tx_ring, eop);
639 while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
640 (count < tx_ring->count)) {
641 bool cleaned = false;
642 for (; !cleaned; count++) {
643 tx_desc = E1000_TX_DESC(*tx_ring, i);
644 buffer_info = &tx_ring->buffer_info[i];
645 cleaned = (i == eop);
648 struct sk_buff *skb = buffer_info->skb;
649 unsigned int segs, bytecount;
650 segs = skb_shinfo(skb)->gso_segs ?: 1;
651 /* multiply data chunks by size of headers */
652 bytecount = ((segs - 1) * skb_headlen(skb)) +
654 total_tx_packets += segs;
655 total_tx_bytes += bytecount;
658 e1000_put_txbuf(adapter, buffer_info);
659 tx_desc->upper.data = 0;
662 if (i == tx_ring->count)
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);
2296 adapter->tx_queue_len = adapter->netdev->tx_queue_len;
2300 * e1000_setup_rctl - configure the receive control registers
2301 * @adapter: Board private structure
2303 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2304 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2305 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2307 struct e1000_hw *hw = &adapter->hw;
2312 /* Program MC offset vector base */
2314 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2315 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2316 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2317 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2319 /* Do not Store bad packets */
2320 rctl &= ~E1000_RCTL_SBP;
2322 /* Enable Long Packet receive */
2323 if (adapter->netdev->mtu <= ETH_DATA_LEN)
2324 rctl &= ~E1000_RCTL_LPE;
2326 rctl |= E1000_RCTL_LPE;
2328 /* Some systems expect that the CRC is included in SMBUS traffic. The
2329 * hardware strips the CRC before sending to both SMBUS (BMC) and to
2330 * host memory when this is enabled
2332 if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2333 rctl |= E1000_RCTL_SECRC;
2335 /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2336 if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2339 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2341 phy_data |= (1 << 2);
2342 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2344 e1e_rphy(hw, 22, &phy_data);
2346 phy_data |= (1 << 14);
2347 e1e_wphy(hw, 0x10, 0x2823);
2348 e1e_wphy(hw, 0x11, 0x0003);
2349 e1e_wphy(hw, 22, phy_data);
2352 /* Setup buffer sizes */
2353 rctl &= ~E1000_RCTL_SZ_4096;
2354 rctl |= E1000_RCTL_BSEX;
2355 switch (adapter->rx_buffer_len) {
2358 rctl |= E1000_RCTL_SZ_2048;
2359 rctl &= ~E1000_RCTL_BSEX;
2362 rctl |= E1000_RCTL_SZ_4096;
2365 rctl |= E1000_RCTL_SZ_8192;
2368 rctl |= E1000_RCTL_SZ_16384;
2373 * 82571 and greater support packet-split where the protocol
2374 * header is placed in skb->data and the packet data is
2375 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2376 * In the case of a non-split, skb->data is linearly filled,
2377 * followed by the page buffers. Therefore, skb->data is
2378 * sized to hold the largest protocol header.
2380 * allocations using alloc_page take too long for regular MTU
2381 * so only enable packet split for jumbo frames
2383 * Using pages when the page size is greater than 16k wastes
2384 * a lot of memory, since we allocate 3 pages at all times
2387 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2388 if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2389 (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2390 adapter->rx_ps_pages = pages;
2392 adapter->rx_ps_pages = 0;
2394 if (adapter->rx_ps_pages) {
2395 /* Configure extra packet-split registers */
2396 rfctl = er32(RFCTL);
2397 rfctl |= E1000_RFCTL_EXTEN;
2399 * disable packet split support for IPv6 extension headers,
2400 * because some malformed IPv6 headers can hang the Rx
2402 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2403 E1000_RFCTL_NEW_IPV6_EXT_DIS);
2407 /* Enable Packet split descriptors */
2408 rctl |= E1000_RCTL_DTYP_PS;
2410 psrctl |= adapter->rx_ps_bsize0 >>
2411 E1000_PSRCTL_BSIZE0_SHIFT;
2413 switch (adapter->rx_ps_pages) {
2415 psrctl |= PAGE_SIZE <<
2416 E1000_PSRCTL_BSIZE3_SHIFT;
2418 psrctl |= PAGE_SIZE <<
2419 E1000_PSRCTL_BSIZE2_SHIFT;
2421 psrctl |= PAGE_SIZE >>
2422 E1000_PSRCTL_BSIZE1_SHIFT;
2426 ew32(PSRCTL, psrctl);
2430 /* just started the receive unit, no need to restart */
2431 adapter->flags &= ~FLAG_RX_RESTART_NOW;
2435 * e1000_configure_rx - Configure Receive Unit after Reset
2436 * @adapter: board private structure
2438 * Configure the Rx unit of the MAC after a reset.
2440 static void e1000_configure_rx(struct e1000_adapter *adapter)
2442 struct e1000_hw *hw = &adapter->hw;
2443 struct e1000_ring *rx_ring = adapter->rx_ring;
2445 u32 rdlen, rctl, rxcsum, ctrl_ext;
2447 if (adapter->rx_ps_pages) {
2448 /* this is a 32 byte descriptor */
2449 rdlen = rx_ring->count *
2450 sizeof(union e1000_rx_desc_packet_split);
2451 adapter->clean_rx = e1000_clean_rx_irq_ps;
2452 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2453 } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2454 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2455 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2456 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2458 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2459 adapter->clean_rx = e1000_clean_rx_irq;
2460 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2463 /* disable receives while setting up the descriptors */
2465 ew32(RCTL, rctl & ~E1000_RCTL_EN);
2469 /* set the Receive Delay Timer Register */
2470 ew32(RDTR, adapter->rx_int_delay);
2472 /* irq moderation */
2473 ew32(RADV, adapter->rx_abs_int_delay);
2474 if (adapter->itr_setting != 0)
2475 ew32(ITR, 1000000000 / (adapter->itr * 256));
2477 ctrl_ext = er32(CTRL_EXT);
2478 /* Auto-Mask interrupts upon ICR access */
2479 ctrl_ext |= E1000_CTRL_EXT_IAME;
2480 ew32(IAM, 0xffffffff);
2481 ew32(CTRL_EXT, ctrl_ext);
2485 * Setup the HW Rx Head and Tail Descriptor Pointers and
2486 * the Base and Length of the Rx Descriptor Ring
2488 rdba = rx_ring->dma;
2489 ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2490 ew32(RDBAH, (rdba >> 32));
2494 rx_ring->head = E1000_RDH;
2495 rx_ring->tail = E1000_RDT;
2497 /* Enable Receive Checksum Offload for TCP and UDP */
2498 rxcsum = er32(RXCSUM);
2499 if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2500 rxcsum |= E1000_RXCSUM_TUOFL;
2503 * IPv4 payload checksum for UDP fragments must be
2504 * used in conjunction with packet-split.
2506 if (adapter->rx_ps_pages)
2507 rxcsum |= E1000_RXCSUM_IPPCSE;
2509 rxcsum &= ~E1000_RXCSUM_TUOFL;
2510 /* no need to clear IPPCSE as it defaults to 0 */
2512 ew32(RXCSUM, rxcsum);
2515 * Enable early receives on supported devices, only takes effect when
2516 * packet size is equal or larger than the specified value (in 8 byte
2517 * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2519 if (adapter->flags & FLAG_HAS_ERT) {
2520 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2521 u32 rxdctl = er32(RXDCTL(0));
2522 ew32(RXDCTL(0), rxdctl | 0x3);
2523 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2525 * With jumbo frames and early-receive enabled,
2526 * excessive C-state transition latencies result in
2527 * dropped transactions.
2529 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2530 adapter->netdev->name, 55);
2532 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2533 adapter->netdev->name,
2534 PM_QOS_DEFAULT_VALUE);
2538 /* Enable Receives */
2543 * e1000_update_mc_addr_list - Update Multicast addresses
2544 * @hw: pointer to the HW structure
2545 * @mc_addr_list: array of multicast addresses to program
2546 * @mc_addr_count: number of multicast addresses to program
2548 * Updates the Multicast Table Array.
2549 * The caller must have a packed mc_addr_list of multicast addresses.
2551 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2554 hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
2558 * e1000_set_multi - Multicast and Promiscuous mode set
2559 * @netdev: network interface device structure
2561 * The set_multi entry point is called whenever the multicast address
2562 * list or the network interface flags are updated. This routine is
2563 * responsible for configuring the hardware for proper multicast,
2564 * promiscuous mode, and all-multi behavior.
2566 static void e1000_set_multi(struct net_device *netdev)
2568 struct e1000_adapter *adapter = netdev_priv(netdev);
2569 struct e1000_hw *hw = &adapter->hw;
2570 struct dev_mc_list *mc_ptr;
2575 /* Check for Promiscuous and All Multicast modes */
2579 if (netdev->flags & IFF_PROMISC) {
2580 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2581 rctl &= ~E1000_RCTL_VFE;
2583 if (netdev->flags & IFF_ALLMULTI) {
2584 rctl |= E1000_RCTL_MPE;
2585 rctl &= ~E1000_RCTL_UPE;
2587 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2589 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2590 rctl |= E1000_RCTL_VFE;
2595 if (!netdev_mc_empty(netdev)) {
2596 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
2600 /* prepare a packed array of only addresses. */
2602 netdev_for_each_mc_addr(mc_ptr, netdev)
2603 memcpy(mta_list + (i++ * ETH_ALEN),
2604 mc_ptr->dmi_addr, ETH_ALEN);
2606 e1000_update_mc_addr_list(hw, mta_list, i);
2610 * if we're called from probe, we might not have
2611 * anything to do here, so clear out the list
2613 e1000_update_mc_addr_list(hw, NULL, 0);
2618 * e1000_configure - configure the hardware for Rx and Tx
2619 * @adapter: private board structure
2621 static void e1000_configure(struct e1000_adapter *adapter)
2623 e1000_set_multi(adapter->netdev);
2625 e1000_restore_vlan(adapter);
2626 e1000_init_manageability(adapter);
2628 e1000_configure_tx(adapter);
2629 e1000_setup_rctl(adapter);
2630 e1000_configure_rx(adapter);
2631 adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2635 * e1000e_power_up_phy - restore link in case the phy was powered down
2636 * @adapter: address of board private structure
2638 * The phy may be powered down to save power and turn off link when the
2639 * driver is unloaded and wake on lan is not enabled (among others)
2640 * *** this routine MUST be followed by a call to e1000e_reset ***
2642 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2644 if (adapter->hw.phy.ops.power_up)
2645 adapter->hw.phy.ops.power_up(&adapter->hw);
2647 adapter->hw.mac.ops.setup_link(&adapter->hw);
2651 * e1000_power_down_phy - Power down the PHY
2653 * Power down the PHY so no link is implied when interface is down.
2654 * The PHY cannot be powered down if management or WoL is active.
2656 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2658 /* WoL is enabled */
2662 if (adapter->hw.phy.ops.power_down)
2663 adapter->hw.phy.ops.power_down(&adapter->hw);
2667 * e1000e_reset - bring the hardware into a known good state
2669 * This function boots the hardware and enables some settings that
2670 * require a configuration cycle of the hardware - those cannot be
2671 * set/changed during runtime. After reset the device needs to be
2672 * properly configured for Rx, Tx etc.
2674 void e1000e_reset(struct e1000_adapter *adapter)
2676 struct e1000_mac_info *mac = &adapter->hw.mac;
2677 struct e1000_fc_info *fc = &adapter->hw.fc;
2678 struct e1000_hw *hw = &adapter->hw;
2679 u32 tx_space, min_tx_space, min_rx_space;
2680 u32 pba = adapter->pba;
2683 /* reset Packet Buffer Allocation to default */
2686 if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2688 * To maintain wire speed transmits, the Tx FIFO should be
2689 * large enough to accommodate two full transmit packets,
2690 * rounded up to the next 1KB and expressed in KB. Likewise,
2691 * the Rx FIFO should be large enough to accommodate at least
2692 * one full receive packet and is similarly rounded up and
2696 /* upper 16 bits has Tx packet buffer allocation size in KB */
2697 tx_space = pba >> 16;
2698 /* lower 16 bits has Rx packet buffer allocation size in KB */
2701 * the Tx fifo also stores 16 bytes of information about the tx
2702 * but don't include ethernet FCS because hardware appends it
2704 min_tx_space = (adapter->max_frame_size +
2705 sizeof(struct e1000_tx_desc) -
2707 min_tx_space = ALIGN(min_tx_space, 1024);
2708 min_tx_space >>= 10;
2709 /* software strips receive CRC, so leave room for it */
2710 min_rx_space = adapter->max_frame_size;
2711 min_rx_space = ALIGN(min_rx_space, 1024);
2712 min_rx_space >>= 10;
2715 * If current Tx allocation is less than the min Tx FIFO size,
2716 * and the min Tx FIFO size is less than the current Rx FIFO
2717 * allocation, take space away from current Rx allocation
2719 if ((tx_space < min_tx_space) &&
2720 ((min_tx_space - tx_space) < pba)) {
2721 pba -= min_tx_space - tx_space;
2724 * if short on Rx space, Rx wins and must trump tx
2725 * adjustment or use Early Receive if available
2727 if ((pba < min_rx_space) &&
2728 (!(adapter->flags & FLAG_HAS_ERT)))
2729 /* ERT enabled in e1000_configure_rx */
2738 * flow control settings
2740 * The high water mark must be low enough to fit one full frame
2741 * (or the size used for early receive) above it in the Rx FIFO.
2742 * Set it to the lower of:
2743 * - 90% of the Rx FIFO size, and
2744 * - the full Rx FIFO size minus the early receive size (for parts
2745 * with ERT support assuming ERT set to E1000_ERT_2048), or
2746 * - the full Rx FIFO size minus one full frame
2748 if (hw->mac.type == e1000_pchlan) {
2750 * Workaround PCH LOM adapter hangs with certain network
2751 * loads. If hangs persist, try disabling Tx flow control.
2753 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2754 fc->high_water = 0x3500;
2755 fc->low_water = 0x1500;
2757 fc->high_water = 0x5000;
2758 fc->low_water = 0x3000;
2761 if ((adapter->flags & FLAG_HAS_ERT) &&
2762 (adapter->netdev->mtu > ETH_DATA_LEN))
2763 hwm = min(((pba << 10) * 9 / 10),
2764 ((pba << 10) - (E1000_ERT_2048 << 3)));
2766 hwm = min(((pba << 10) * 9 / 10),
2767 ((pba << 10) - adapter->max_frame_size));
2769 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
2770 fc->low_water = fc->high_water - 8;
2773 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2774 fc->pause_time = 0xFFFF;
2776 fc->pause_time = E1000_FC_PAUSE_TIME;
2778 fc->current_mode = fc->requested_mode;
2780 /* Allow time for pending master requests to run */
2781 mac->ops.reset_hw(hw);
2784 * For parts with AMT enabled, let the firmware know
2785 * that the network interface is in control
2787 if (adapter->flags & FLAG_HAS_AMT)
2788 e1000_get_hw_control(adapter);
2791 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP)
2792 e1e_wphy(&adapter->hw, BM_WUC, 0);
2794 if (mac->ops.init_hw(hw))
2795 e_err("Hardware Error\n");
2797 /* additional part of the flow-control workaround above */
2798 if (hw->mac.type == e1000_pchlan)
2799 ew32(FCRTV_PCH, 0x1000);
2801 e1000_update_mng_vlan(adapter);
2803 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2804 ew32(VET, ETH_P_8021Q);
2806 e1000e_reset_adaptive(hw);
2807 e1000_get_phy_info(hw);
2809 if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
2810 !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2813 * speed up time to link by disabling smart power down, ignore
2814 * the return value of this function because there is nothing
2815 * different we would do if it failed
2817 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2818 phy_data &= ~IGP02E1000_PM_SPD;
2819 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2823 int e1000e_up(struct e1000_adapter *adapter)
2825 struct e1000_hw *hw = &adapter->hw;
2827 /* DMA latency requirement to workaround early-receive/jumbo issue */
2828 if (adapter->flags & FLAG_HAS_ERT)
2829 pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY,
2830 adapter->netdev->name,
2831 PM_QOS_DEFAULT_VALUE);
2833 /* hardware has been reset, we need to reload some things */
2834 e1000_configure(adapter);
2836 clear_bit(__E1000_DOWN, &adapter->state);
2838 napi_enable(&adapter->napi);
2839 if (adapter->msix_entries)
2840 e1000_configure_msix(adapter);
2841 e1000_irq_enable(adapter);
2843 netif_wake_queue(adapter->netdev);
2845 /* fire a link change interrupt to start the watchdog */
2846 ew32(ICS, E1000_ICS_LSC);
2850 void e1000e_down(struct e1000_adapter *adapter)
2852 struct net_device *netdev = adapter->netdev;
2853 struct e1000_hw *hw = &adapter->hw;
2857 * signal that we're down so the interrupt handler does not
2858 * reschedule our watchdog timer
2860 set_bit(__E1000_DOWN, &adapter->state);
2862 /* disable receives in the hardware */
2864 ew32(RCTL, rctl & ~E1000_RCTL_EN);
2865 /* flush and sleep below */
2867 netif_stop_queue(netdev);
2869 /* disable transmits in the hardware */
2871 tctl &= ~E1000_TCTL_EN;
2873 /* flush both disables and wait for them to finish */
2877 napi_disable(&adapter->napi);
2878 e1000_irq_disable(adapter);
2880 del_timer_sync(&adapter->watchdog_timer);
2881 del_timer_sync(&adapter->phy_info_timer);
2883 netdev->tx_queue_len = adapter->tx_queue_len;
2884 netif_carrier_off(netdev);
2885 adapter->link_speed = 0;
2886 adapter->link_duplex = 0;
2888 if (!pci_channel_offline(adapter->pdev))
2889 e1000e_reset(adapter);
2890 e1000_clean_tx_ring(adapter);
2891 e1000_clean_rx_ring(adapter);
2893 if (adapter->flags & FLAG_HAS_ERT)
2894 pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY,
2895 adapter->netdev->name);
2898 * TODO: for power management, we could drop the link and
2899 * pci_disable_device here.
2903 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2906 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2908 e1000e_down(adapter);
2910 clear_bit(__E1000_RESETTING, &adapter->state);
2914 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2915 * @adapter: board private structure to initialize
2917 * e1000_sw_init initializes the Adapter private data structure.
2918 * Fields are initialized based on PCI device information and
2919 * OS network device settings (MTU size).
2921 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2923 struct net_device *netdev = adapter->netdev;
2925 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2926 adapter->rx_ps_bsize0 = 128;
2927 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2928 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2930 e1000e_set_interrupt_capability(adapter);
2932 if (e1000_alloc_queues(adapter))
2935 /* Explicitly disable IRQ since the NIC can be in any state. */
2936 e1000_irq_disable(adapter);
2938 set_bit(__E1000_DOWN, &adapter->state);
2943 * e1000_intr_msi_test - Interrupt Handler
2944 * @irq: interrupt number
2945 * @data: pointer to a network interface device structure
2947 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2949 struct net_device *netdev = data;
2950 struct e1000_adapter *adapter = netdev_priv(netdev);
2951 struct e1000_hw *hw = &adapter->hw;
2952 u32 icr = er32(ICR);
2954 e_dbg("icr is %08X\n", icr);
2955 if (icr & E1000_ICR_RXSEQ) {
2956 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2964 * e1000_test_msi_interrupt - Returns 0 for successful test
2965 * @adapter: board private struct
2967 * code flow taken from tg3.c
2969 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2971 struct net_device *netdev = adapter->netdev;
2972 struct e1000_hw *hw = &adapter->hw;
2975 /* poll_enable hasn't been called yet, so don't need disable */
2976 /* clear any pending events */
2979 /* free the real vector and request a test handler */
2980 e1000_free_irq(adapter);
2981 e1000e_reset_interrupt_capability(adapter);
2983 /* Assume that the test fails, if it succeeds then the test
2984 * MSI irq handler will unset this flag */
2985 adapter->flags |= FLAG_MSI_TEST_FAILED;
2987 err = pci_enable_msi(adapter->pdev);
2989 goto msi_test_failed;
2991 err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
2992 netdev->name, netdev);
2994 pci_disable_msi(adapter->pdev);
2995 goto msi_test_failed;
3000 e1000_irq_enable(adapter);
3002 /* fire an unusual interrupt on the test handler */
3003 ew32(ICS, E1000_ICS_RXSEQ);
3007 e1000_irq_disable(adapter);
3011 if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3012 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3014 e_info("MSI interrupt test failed!\n");
3017 free_irq(adapter->pdev->irq, netdev);
3018 pci_disable_msi(adapter->pdev);
3021 goto msi_test_failed;
3023 /* okay so the test worked, restore settings */
3024 e_dbg("MSI interrupt test succeeded!\n");
3026 e1000e_set_interrupt_capability(adapter);
3027 e1000_request_irq(adapter);
3032 * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3033 * @adapter: board private struct
3035 * code flow taken from tg3.c, called with e1000 interrupts disabled.
3037 static int e1000_test_msi(struct e1000_adapter *adapter)
3042 if (!(adapter->flags & FLAG_MSI_ENABLED))
3045 /* disable SERR in case the MSI write causes a master abort */
3046 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3047 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3048 pci_cmd & ~PCI_COMMAND_SERR);
3050 err = e1000_test_msi_interrupt(adapter);
3052 /* restore previous setting of command word */
3053 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3059 /* EIO means MSI test failed */
3063 /* back to INTx mode */
3064 e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3066 e1000_free_irq(adapter);
3068 err = e1000_request_irq(adapter);
3074 * e1000_open - Called when a network interface is made active
3075 * @netdev: network interface device structure
3077 * Returns 0 on success, negative value on failure
3079 * The open entry point is called when a network interface is made
3080 * active by the system (IFF_UP). At this point all resources needed
3081 * for transmit and receive operations are allocated, the interrupt
3082 * handler is registered with the OS, the watchdog timer is started,
3083 * and the stack is notified that the interface is ready.
3085 static int e1000_open(struct net_device *netdev)
3087 struct e1000_adapter *adapter = netdev_priv(netdev);
3088 struct e1000_hw *hw = &adapter->hw;
3089 struct pci_dev *pdev = adapter->pdev;
3092 /* disallow open during test */
3093 if (test_bit(__E1000_TESTING, &adapter->state))
3096 pm_runtime_get_sync(&pdev->dev);
3098 netif_carrier_off(netdev);
3100 /* allocate transmit descriptors */
3101 err = e1000e_setup_tx_resources(adapter);
3105 /* allocate receive descriptors */
3106 err = e1000e_setup_rx_resources(adapter);
3110 e1000e_power_up_phy(adapter);
3112 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3113 if ((adapter->hw.mng_cookie.status &
3114 E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3115 e1000_update_mng_vlan(adapter);
3118 * If AMT is enabled, let the firmware know that the network
3119 * interface is now open
3121 if (adapter->flags & FLAG_HAS_AMT)
3122 e1000_get_hw_control(adapter);
3125 * before we allocate an interrupt, we must be ready to handle it.
3126 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3127 * as soon as we call pci_request_irq, so we have to setup our
3128 * clean_rx handler before we do so.
3130 e1000_configure(adapter);
3132 err = e1000_request_irq(adapter);
3137 * Work around PCIe errata with MSI interrupts causing some chipsets to
3138 * ignore e1000e MSI messages, which means we need to test our MSI
3141 if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3142 err = e1000_test_msi(adapter);
3144 e_err("Interrupt allocation failed\n");
3149 /* From here on the code is the same as e1000e_up() */
3150 clear_bit(__E1000_DOWN, &adapter->state);
3152 napi_enable(&adapter->napi);
3154 e1000_irq_enable(adapter);
3156 netif_start_queue(netdev);
3158 adapter->idle_check = true;
3159 pm_runtime_put(&pdev->dev);
3161 /* fire a link status change interrupt to start the watchdog */
3162 ew32(ICS, E1000_ICS_LSC);
3167 e1000_release_hw_control(adapter);
3168 e1000_power_down_phy(adapter);
3169 e1000e_free_rx_resources(adapter);
3171 e1000e_free_tx_resources(adapter);
3173 e1000e_reset(adapter);
3174 pm_runtime_put_sync(&pdev->dev);
3180 * e1000_close - Disables a network interface
3181 * @netdev: network interface device structure
3183 * Returns 0, this is not allowed to fail
3185 * The close entry point is called when an interface is de-activated
3186 * by the OS. The hardware is still under the drivers control, but
3187 * needs to be disabled. A global MAC reset is issued to stop the
3188 * hardware, and all transmit and receive resources are freed.
3190 static int e1000_close(struct net_device *netdev)
3192 struct e1000_adapter *adapter = netdev_priv(netdev);
3193 struct pci_dev *pdev = adapter->pdev;
3195 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3197 pm_runtime_get_sync(&pdev->dev);
3199 if (!test_bit(__E1000_DOWN, &adapter->state)) {
3200 e1000e_down(adapter);
3201 e1000_free_irq(adapter);
3203 e1000_power_down_phy(adapter);
3205 e1000e_free_tx_resources(adapter);
3206 e1000e_free_rx_resources(adapter);
3209 * kill manageability vlan ID if supported, but not if a vlan with
3210 * the same ID is registered on the host OS (let 8021q kill it)
3212 if ((adapter->hw.mng_cookie.status &
3213 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3215 vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3216 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3219 * If AMT is enabled, let the firmware know that the network
3220 * interface is now closed
3222 if (adapter->flags & FLAG_HAS_AMT)
3223 e1000_release_hw_control(adapter);
3225 pm_runtime_put_sync(&pdev->dev);
3230 * e1000_set_mac - Change the Ethernet Address of the NIC
3231 * @netdev: network interface device structure
3232 * @p: pointer to an address structure
3234 * Returns 0 on success, negative on failure
3236 static int e1000_set_mac(struct net_device *netdev, void *p)
3238 struct e1000_adapter *adapter = netdev_priv(netdev);
3239 struct sockaddr *addr = p;
3241 if (!is_valid_ether_addr(addr->sa_data))
3242 return -EADDRNOTAVAIL;
3244 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3245 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3247 e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3249 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3250 /* activate the work around */
3251 e1000e_set_laa_state_82571(&adapter->hw, 1);
3254 * Hold a copy of the LAA in RAR[14] This is done so that
3255 * between the time RAR[0] gets clobbered and the time it
3256 * gets fixed (in e1000_watchdog), the actual LAA is in one
3257 * of the RARs and no incoming packets directed to this port
3258 * are dropped. Eventually the LAA will be in RAR[0] and
3261 e1000e_rar_set(&adapter->hw,
3262 adapter->hw.mac.addr,
3263 adapter->hw.mac.rar_entry_count - 1);
3270 * e1000e_update_phy_task - work thread to update phy
3271 * @work: pointer to our work struct
3273 * this worker thread exists because we must acquire a
3274 * semaphore to read the phy, which we could msleep while
3275 * waiting for it, and we can't msleep in a timer.
3277 static void e1000e_update_phy_task(struct work_struct *work)
3279 struct e1000_adapter *adapter = container_of(work,
3280 struct e1000_adapter, update_phy_task);
3281 e1000_get_phy_info(&adapter->hw);
3285 * Need to wait a few seconds after link up to get diagnostic information from
3288 static void e1000_update_phy_info(unsigned long data)
3290 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3291 schedule_work(&adapter->update_phy_task);
3295 * e1000e_update_stats - Update the board statistics counters
3296 * @adapter: board private structure
3298 void e1000e_update_stats(struct e1000_adapter *adapter)
3300 struct net_device *netdev = adapter->netdev;
3301 struct e1000_hw *hw = &adapter->hw;
3302 struct pci_dev *pdev = adapter->pdev;
3306 * Prevent stats update while adapter is being reset, or if the pci
3307 * connection is down.
3309 if (adapter->link_speed == 0)
3311 if (pci_channel_offline(pdev))
3314 adapter->stats.crcerrs += er32(CRCERRS);
3315 adapter->stats.gprc += er32(GPRC);
3316 adapter->stats.gorc += er32(GORCL);
3317 er32(GORCH); /* Clear gorc */
3318 adapter->stats.bprc += er32(BPRC);
3319 adapter->stats.mprc += er32(MPRC);
3320 adapter->stats.roc += er32(ROC);
3322 adapter->stats.mpc += er32(MPC);
3323 if ((hw->phy.type == e1000_phy_82578) ||
3324 (hw->phy.type == e1000_phy_82577)) {
3325 e1e_rphy(hw, HV_SCC_UPPER, &phy_data);
3326 if (!e1e_rphy(hw, HV_SCC_LOWER, &phy_data))
3327 adapter->stats.scc += phy_data;
3329 e1e_rphy(hw, HV_ECOL_UPPER, &phy_data);
3330 if (!e1e_rphy(hw, HV_ECOL_LOWER, &phy_data))
3331 adapter->stats.ecol += phy_data;
3333 e1e_rphy(hw, HV_MCC_UPPER, &phy_data);
3334 if (!e1e_rphy(hw, HV_MCC_LOWER, &phy_data))
3335 adapter->stats.mcc += phy_data;
3337 e1e_rphy(hw, HV_LATECOL_UPPER, &phy_data);
3338 if (!e1e_rphy(hw, HV_LATECOL_LOWER, &phy_data))
3339 adapter->stats.latecol += phy_data;
3341 e1e_rphy(hw, HV_DC_UPPER, &phy_data);
3342 if (!e1e_rphy(hw, HV_DC_LOWER, &phy_data))
3343 adapter->stats.dc += phy_data;
3345 adapter->stats.scc += er32(SCC);
3346 adapter->stats.ecol += er32(ECOL);
3347 adapter->stats.mcc += er32(MCC);
3348 adapter->stats.latecol += er32(LATECOL);
3349 adapter->stats.dc += er32(DC);
3351 adapter->stats.xonrxc += er32(XONRXC);
3352 adapter->stats.xontxc += er32(XONTXC);
3353 adapter->stats.xoffrxc += er32(XOFFRXC);
3354 adapter->stats.xofftxc += er32(XOFFTXC);
3355 adapter->stats.gptc += er32(GPTC);
3356 adapter->stats.gotc += er32(GOTCL);
3357 er32(GOTCH); /* Clear gotc */
3358 adapter->stats.rnbc += er32(RNBC);
3359 adapter->stats.ruc += er32(RUC);
3361 adapter->stats.mptc += er32(MPTC);
3362 adapter->stats.bptc += er32(BPTC);
3364 /* used for adaptive IFS */
3366 hw->mac.tx_packet_delta = er32(TPT);
3367 adapter->stats.tpt += hw->mac.tx_packet_delta;
3368 if ((hw->phy.type == e1000_phy_82578) ||
3369 (hw->phy.type == e1000_phy_82577)) {
3370 e1e_rphy(hw, HV_COLC_UPPER, &phy_data);
3371 if (!e1e_rphy(hw, HV_COLC_LOWER, &phy_data))
3372 hw->mac.collision_delta = phy_data;
3374 hw->mac.collision_delta = er32(COLC);
3376 adapter->stats.colc += hw->mac.collision_delta;
3378 adapter->stats.algnerrc += er32(ALGNERRC);
3379 adapter->stats.rxerrc += er32(RXERRC);
3380 if ((hw->phy.type == e1000_phy_82578) ||
3381 (hw->phy.type == e1000_phy_82577)) {
3382 e1e_rphy(hw, HV_TNCRS_UPPER, &phy_data);
3383 if (!e1e_rphy(hw, HV_TNCRS_LOWER, &phy_data))
3384 adapter->stats.tncrs += phy_data;
3386 if ((hw->mac.type != e1000_82574) &&
3387 (hw->mac.type != e1000_82583))
3388 adapter->stats.tncrs += er32(TNCRS);
3390 adapter->stats.cexterr += er32(CEXTERR);
3391 adapter->stats.tsctc += er32(TSCTC);
3392 adapter->stats.tsctfc += er32(TSCTFC);
3394 /* Fill out the OS statistics structure */
3395 netdev->stats.multicast = adapter->stats.mprc;
3396 netdev->stats.collisions = adapter->stats.colc;
3401 * RLEC on some newer hardware can be incorrect so build
3402 * our own version based on RUC and ROC
3404 netdev->stats.rx_errors = adapter->stats.rxerrc +
3405 adapter->stats.crcerrs + adapter->stats.algnerrc +
3406 adapter->stats.ruc + adapter->stats.roc +
3407 adapter->stats.cexterr;
3408 netdev->stats.rx_length_errors = adapter->stats.ruc +
3410 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3411 netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3412 netdev->stats.rx_missed_errors = adapter->stats.mpc;
3415 netdev->stats.tx_errors = adapter->stats.ecol +
3416 adapter->stats.latecol;
3417 netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3418 netdev->stats.tx_window_errors = adapter->stats.latecol;
3419 netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3421 /* Tx Dropped needs to be maintained elsewhere */
3423 /* Management Stats */
3424 adapter->stats.mgptc += er32(MGTPTC);
3425 adapter->stats.mgprc += er32(MGTPRC);
3426 adapter->stats.mgpdc += er32(MGTPDC);
3430 * e1000_phy_read_status - Update the PHY register status snapshot
3431 * @adapter: board private structure
3433 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3435 struct e1000_hw *hw = &adapter->hw;
3436 struct e1000_phy_regs *phy = &adapter->phy_regs;
3439 if ((er32(STATUS) & E1000_STATUS_LU) &&
3440 (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3441 ret_val = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3442 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3443 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3444 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3445 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3446 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3447 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3448 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3450 e_warn("Error reading PHY register\n");
3453 * Do not read PHY registers if link is not up
3454 * Set values to typical power-on defaults
3456 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3457 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3458 BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3460 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3461 ADVERTISE_ALL | ADVERTISE_CSMA);
3463 phy->expansion = EXPANSION_ENABLENPAGE;
3464 phy->ctrl1000 = ADVERTISE_1000FULL;
3466 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3470 static void e1000_print_link_info(struct e1000_adapter *adapter)
3472 struct e1000_hw *hw = &adapter->hw;
3473 u32 ctrl = er32(CTRL);
3475 /* Link status message must follow this format for user tools */
3476 printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3477 "Flow Control: %s\n",
3478 adapter->netdev->name,
3479 adapter->link_speed,
3480 (adapter->link_duplex == FULL_DUPLEX) ?
3481 "Full Duplex" : "Half Duplex",
3482 ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3484 ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3485 ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3488 bool e1000e_has_link(struct e1000_adapter *adapter)
3490 struct e1000_hw *hw = &adapter->hw;
3491 bool link_active = 0;
3495 * get_link_status is set on LSC (link status) interrupt or
3496 * Rx sequence error interrupt. get_link_status will stay
3497 * false until the check_for_link establishes link
3498 * for copper adapters ONLY
3500 switch (hw->phy.media_type) {
3501 case e1000_media_type_copper:
3502 if (hw->mac.get_link_status) {
3503 ret_val = hw->mac.ops.check_for_link(hw);
3504 link_active = !hw->mac.get_link_status;
3509 case e1000_media_type_fiber:
3510 ret_val = hw->mac.ops.check_for_link(hw);
3511 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3513 case e1000_media_type_internal_serdes:
3514 ret_val = hw->mac.ops.check_for_link(hw);
3515 link_active = adapter->hw.mac.serdes_has_link;
3518 case e1000_media_type_unknown:
3522 if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3523 (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3524 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3525 e_info("Gigabit has been disabled, downgrading speed\n");
3531 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3533 /* make sure the receive unit is started */
3534 if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3535 (adapter->flags & FLAG_RX_RESTART_NOW)) {
3536 struct e1000_hw *hw = &adapter->hw;
3537 u32 rctl = er32(RCTL);
3538 ew32(RCTL, rctl | E1000_RCTL_EN);
3539 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3544 * e1000_watchdog - Timer Call-back
3545 * @data: pointer to adapter cast into an unsigned long
3547 static void e1000_watchdog(unsigned long data)
3549 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3551 /* Do the rest outside of interrupt context */
3552 schedule_work(&adapter->watchdog_task);
3554 /* TODO: make this use queue_delayed_work() */
3557 static void e1000_watchdog_task(struct work_struct *work)
3559 struct e1000_adapter *adapter = container_of(work,
3560 struct e1000_adapter, watchdog_task);
3561 struct net_device *netdev = adapter->netdev;
3562 struct e1000_mac_info *mac = &adapter->hw.mac;
3563 struct e1000_phy_info *phy = &adapter->hw.phy;
3564 struct e1000_ring *tx_ring = adapter->tx_ring;
3565 struct e1000_hw *hw = &adapter->hw;
3569 link = e1000e_has_link(adapter);
3570 if ((netif_carrier_ok(netdev)) && link) {
3571 /* Cancel scheduled suspend requests. */
3572 pm_runtime_resume(netdev->dev.parent);
3574 e1000e_enable_receives(adapter);
3578 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3579 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3580 e1000_update_mng_vlan(adapter);
3583 if (!netif_carrier_ok(netdev)) {
3586 /* Cancel scheduled suspend requests. */
3587 pm_runtime_resume(netdev->dev.parent);
3589 /* update snapshot of PHY registers on LSC */
3590 e1000_phy_read_status(adapter);
3591 mac->ops.get_link_up_info(&adapter->hw,
3592 &adapter->link_speed,
3593 &adapter->link_duplex);
3594 e1000_print_link_info(adapter);
3596 * On supported PHYs, check for duplex mismatch only
3597 * if link has autonegotiated at 10/100 half
3599 if ((hw->phy.type == e1000_phy_igp_3 ||
3600 hw->phy.type == e1000_phy_bm) &&
3601 (hw->mac.autoneg == true) &&
3602 (adapter->link_speed == SPEED_10 ||
3603 adapter->link_speed == SPEED_100) &&
3604 (adapter->link_duplex == HALF_DUPLEX)) {
3607 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3609 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3610 e_info("Autonegotiated half duplex but"
3611 " link partner cannot autoneg. "
3612 " Try forcing full duplex if "
3613 "link gets many collisions.\n");
3617 * tweak tx_queue_len according to speed/duplex
3618 * and adjust the timeout factor
3620 netdev->tx_queue_len = adapter->tx_queue_len;
3621 adapter->tx_timeout_factor = 1;
3622 switch (adapter->link_speed) {
3625 netdev->tx_queue_len = 10;
3626 adapter->tx_timeout_factor = 16;
3630 netdev->tx_queue_len = 100;
3631 adapter->tx_timeout_factor = 10;
3636 * workaround: re-program speed mode bit after
3639 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3642 tarc0 = er32(TARC(0));
3643 tarc0 &= ~SPEED_MODE_BIT;
3644 ew32(TARC(0), tarc0);
3648 * disable TSO for pcie and 10/100 speeds, to avoid
3649 * some hardware issues
3651 if (!(adapter->flags & FLAG_TSO_FORCE)) {
3652 switch (adapter->link_speed) {
3655 e_info("10/100 speed: disabling TSO\n");
3656 netdev->features &= ~NETIF_F_TSO;
3657 netdev->features &= ~NETIF_F_TSO6;
3660 netdev->features |= NETIF_F_TSO;
3661 netdev->features |= NETIF_F_TSO6;
3670 * enable transmits in the hardware, need to do this
3671 * after setting TARC(0)
3674 tctl |= E1000_TCTL_EN;
3678 * Perform any post-link-up configuration before
3679 * reporting link up.
3681 if (phy->ops.cfg_on_link_up)
3682 phy->ops.cfg_on_link_up(hw);
3684 netif_carrier_on(netdev);
3686 if (!test_bit(__E1000_DOWN, &adapter->state))
3687 mod_timer(&adapter->phy_info_timer,
3688 round_jiffies(jiffies + 2 * HZ));
3691 if (netif_carrier_ok(netdev)) {
3692 adapter->link_speed = 0;
3693 adapter->link_duplex = 0;
3694 /* Link status message must follow this format */
3695 printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
3696 adapter->netdev->name);
3697 netif_carrier_off(netdev);
3698 if (!test_bit(__E1000_DOWN, &adapter->state))
3699 mod_timer(&adapter->phy_info_timer,
3700 round_jiffies(jiffies + 2 * HZ));
3702 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3703 schedule_work(&adapter->reset_task);
3705 pm_schedule_suspend(netdev->dev.parent,
3711 e1000e_update_stats(adapter);
3713 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3714 adapter->tpt_old = adapter->stats.tpt;
3715 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3716 adapter->colc_old = adapter->stats.colc;
3718 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3719 adapter->gorc_old = adapter->stats.gorc;
3720 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3721 adapter->gotc_old = adapter->stats.gotc;
3723 e1000e_update_adaptive(&adapter->hw);
3725 if (!netif_carrier_ok(netdev)) {
3726 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3730 * We've lost link, so the controller stops DMA,
3731 * but we've got queued Tx work that's never going
3732 * to get done, so reset controller to flush Tx.
3733 * (Do the reset outside of interrupt context).
3735 adapter->tx_timeout_count++;
3736 schedule_work(&adapter->reset_task);
3737 /* return immediately since reset is imminent */
3742 /* Cause software interrupt to ensure Rx ring is cleaned */
3743 if (adapter->msix_entries)
3744 ew32(ICS, adapter->rx_ring->ims_val);
3746 ew32(ICS, E1000_ICS_RXDMT0);
3748 /* Force detection of hung controller every watchdog period */
3749 adapter->detect_tx_hung = 1;
3752 * With 82571 controllers, LAA may be overwritten due to controller
3753 * reset from the other port. Set the appropriate LAA in RAR[0]
3755 if (e1000e_get_laa_state_82571(hw))
3756 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3758 /* Reset the timer */
3759 if (!test_bit(__E1000_DOWN, &adapter->state))
3760 mod_timer(&adapter->watchdog_timer,
3761 round_jiffies(jiffies + 2 * HZ));
3764 #define E1000_TX_FLAGS_CSUM 0x00000001
3765 #define E1000_TX_FLAGS_VLAN 0x00000002
3766 #define E1000_TX_FLAGS_TSO 0x00000004
3767 #define E1000_TX_FLAGS_IPV4 0x00000008
3768 #define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
3769 #define E1000_TX_FLAGS_VLAN_SHIFT 16
3771 static int e1000_tso(struct e1000_adapter *adapter,
3772 struct sk_buff *skb)
3774 struct e1000_ring *tx_ring = adapter->tx_ring;
3775 struct e1000_context_desc *context_desc;
3776 struct e1000_buffer *buffer_info;
3779 u16 ipcse = 0, tucse, mss;
3780 u8 ipcss, ipcso, tucss, tucso, hdr_len;
3783 if (!skb_is_gso(skb))
3786 if (skb_header_cloned(skb)) {
3787 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3792 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3793 mss = skb_shinfo(skb)->gso_size;
3794 if (skb->protocol == htons(ETH_P_IP)) {
3795 struct iphdr *iph = ip_hdr(skb);
3798 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
3800 cmd_length = E1000_TXD_CMD_IP;
3801 ipcse = skb_transport_offset(skb) - 1;
3802 } else if (skb_is_gso_v6(skb)) {
3803 ipv6_hdr(skb)->payload_len = 0;
3804 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3805 &ipv6_hdr(skb)->daddr,
3809 ipcss = skb_network_offset(skb);
3810 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3811 tucss = skb_transport_offset(skb);
3812 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3815 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3816 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3818 i = tx_ring->next_to_use;
3819 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3820 buffer_info = &tx_ring->buffer_info[i];
3822 context_desc->lower_setup.ip_fields.ipcss = ipcss;
3823 context_desc->lower_setup.ip_fields.ipcso = ipcso;
3824 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
3825 context_desc->upper_setup.tcp_fields.tucss = tucss;
3826 context_desc->upper_setup.tcp_fields.tucso = tucso;
3827 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3828 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
3829 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3830 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3832 buffer_info->time_stamp = jiffies;
3833 buffer_info->next_to_watch = i;
3836 if (i == tx_ring->count)
3838 tx_ring->next_to_use = i;
3843 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3845 struct e1000_ring *tx_ring = adapter->tx_ring;
3846 struct e1000_context_desc *context_desc;
3847 struct e1000_buffer *buffer_info;
3850 u32 cmd_len = E1000_TXD_CMD_DEXT;
3853 if (skb->ip_summed != CHECKSUM_PARTIAL)
3856 if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
3857 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
3859 protocol = skb->protocol;
3862 case cpu_to_be16(ETH_P_IP):
3863 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3864 cmd_len |= E1000_TXD_CMD_TCP;
3866 case cpu_to_be16(ETH_P_IPV6):
3867 /* XXX not handling all IPV6 headers */
3868 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3869 cmd_len |= E1000_TXD_CMD_TCP;
3872 if (unlikely(net_ratelimit()))
3873 e_warn("checksum_partial proto=%x!\n",
3874 be16_to_cpu(protocol));
3878 css = skb_transport_offset(skb);
3880 i = tx_ring->next_to_use;
3881 buffer_info = &tx_ring->buffer_info[i];
3882 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3884 context_desc->lower_setup.ip_config = 0;
3885 context_desc->upper_setup.tcp_fields.tucss = css;
3886 context_desc->upper_setup.tcp_fields.tucso =
3887 css + skb->csum_offset;
3888 context_desc->upper_setup.tcp_fields.tucse = 0;
3889 context_desc->tcp_seg_setup.data = 0;
3890 context_desc->cmd_and_length = cpu_to_le32(cmd_len);
3892 buffer_info->time_stamp = jiffies;
3893 buffer_info->next_to_watch = i;
3896 if (i == tx_ring->count)
3898 tx_ring->next_to_use = i;
3903 #define E1000_MAX_PER_TXD 8192
3904 #define E1000_MAX_TXD_PWR 12
3906 static int e1000_tx_map(struct e1000_adapter *adapter,
3907 struct sk_buff *skb, unsigned int first,
3908 unsigned int max_per_txd, unsigned int nr_frags,
3911 struct e1000_ring *tx_ring = adapter->tx_ring;
3912 struct pci_dev *pdev = adapter->pdev;
3913 struct e1000_buffer *buffer_info;
3914 unsigned int len = skb_headlen(skb);
3915 unsigned int offset = 0, size, count = 0, i;
3918 i = tx_ring->next_to_use;
3921 buffer_info = &tx_ring->buffer_info[i];
3922 size = min(len, max_per_txd);
3924 buffer_info->length = size;
3925 buffer_info->time_stamp = jiffies;
3926 buffer_info->next_to_watch = i;
3927 buffer_info->dma = pci_map_single(pdev, skb->data + offset,
3928 size, PCI_DMA_TODEVICE);
3929 buffer_info->mapped_as_page = false;
3930 if (pci_dma_mapping_error(pdev, buffer_info->dma))
3939 if (i == tx_ring->count)
3944 for (f = 0; f < nr_frags; f++) {
3945 struct skb_frag_struct *frag;
3947 frag = &skb_shinfo(skb)->frags[f];
3949 offset = frag->page_offset;
3953 if (i == tx_ring->count)
3956 buffer_info = &tx_ring->buffer_info[i];
3957 size = min(len, max_per_txd);
3959 buffer_info->length = size;
3960 buffer_info->time_stamp = jiffies;
3961 buffer_info->next_to_watch = i;
3962 buffer_info->dma = pci_map_page(pdev, frag->page,
3965 buffer_info->mapped_as_page = true;
3966 if (pci_dma_mapping_error(pdev, buffer_info->dma))
3975 tx_ring->buffer_info[i].skb = skb;
3976 tx_ring->buffer_info[first].next_to_watch = i;
3981 dev_err(&pdev->dev, "TX DMA map failed\n");
3982 buffer_info->dma = 0;
3988 i += tx_ring->count;
3990 buffer_info = &tx_ring->buffer_info[i];
3991 e1000_put_txbuf(adapter, buffer_info);;
3997 static void e1000_tx_queue(struct e1000_adapter *adapter,
3998 int tx_flags, int count)
4000 struct e1000_ring *tx_ring = adapter->tx_ring;
4001 struct e1000_tx_desc *tx_desc = NULL;
4002 struct e1000_buffer *buffer_info;
4003 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4006 if (tx_flags & E1000_TX_FLAGS_TSO) {
4007 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4009 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4011 if (tx_flags & E1000_TX_FLAGS_IPV4)
4012 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4015 if (tx_flags & E1000_TX_FLAGS_CSUM) {
4016 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4017 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4020 if (tx_flags & E1000_TX_FLAGS_VLAN) {
4021 txd_lower |= E1000_TXD_CMD_VLE;
4022 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4025 i = tx_ring->next_to_use;
4028 buffer_info = &tx_ring->buffer_info[i];
4029 tx_desc = E1000_TX_DESC(*tx_ring, i);
4030 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4031 tx_desc->lower.data =
4032 cpu_to_le32(txd_lower | buffer_info->length);
4033 tx_desc->upper.data = cpu_to_le32(txd_upper);
4036 if (i == tx_ring->count)
4040 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4043 * Force memory writes to complete before letting h/w
4044 * know there are new descriptors to fetch. (Only
4045 * applicable for weak-ordered memory model archs,
4050 tx_ring->next_to_use = i;
4051 writel(i, adapter->hw.hw_addr + tx_ring->tail);
4053 * we need this if more than one processor can write to our tail
4054 * at a time, it synchronizes IO on IA64/Altix systems
4059 #define MINIMUM_DHCP_PACKET_SIZE 282
4060 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4061 struct sk_buff *skb)
4063 struct e1000_hw *hw = &adapter->hw;
4066 if (vlan_tx_tag_present(skb)) {
4067 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4068 (adapter->hw.mng_cookie.status &
4069 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4073 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4076 if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4080 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4083 if (ip->protocol != IPPROTO_UDP)
4086 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4087 if (ntohs(udp->dest) != 67)
4090 offset = (u8 *)udp + 8 - skb->data;
4091 length = skb->len - offset;
4092 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4098 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4100 struct e1000_adapter *adapter = netdev_priv(netdev);
4102 netif_stop_queue(netdev);
4104 * Herbert's original patch had:
4105 * smp_mb__after_netif_stop_queue();
4106 * but since that doesn't exist yet, just open code it.
4111 * We need to check again in a case another CPU has just
4112 * made room available.
4114 if (e1000_desc_unused(adapter->tx_ring) < size)
4118 netif_start_queue(netdev);
4119 ++adapter->restart_queue;
4123 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4125 struct e1000_adapter *adapter = netdev_priv(netdev);
4127 if (e1000_desc_unused(adapter->tx_ring) >= size)
4129 return __e1000_maybe_stop_tx(netdev, size);
4132 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4133 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4134 struct net_device *netdev)
4136 struct e1000_adapter *adapter = netdev_priv(netdev);
4137 struct e1000_ring *tx_ring = adapter->tx_ring;
4139 unsigned int max_per_txd = E1000_MAX_PER_TXD;
4140 unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4141 unsigned int tx_flags = 0;
4142 unsigned int len = skb->len - skb->data_len;
4143 unsigned int nr_frags;
4149 if (test_bit(__E1000_DOWN, &adapter->state)) {
4150 dev_kfree_skb_any(skb);
4151 return NETDEV_TX_OK;
4154 if (skb->len <= 0) {
4155 dev_kfree_skb_any(skb);
4156 return NETDEV_TX_OK;
4159 mss = skb_shinfo(skb)->gso_size;
4161 * The controller does a simple calculation to
4162 * make sure there is enough room in the FIFO before
4163 * initiating the DMA for each buffer. The calc is:
4164 * 4 = ceil(buffer len/mss). To make sure we don't
4165 * overrun the FIFO, adjust the max buffer len if mss
4170 max_per_txd = min(mss << 2, max_per_txd);
4171 max_txd_pwr = fls(max_per_txd) - 1;
4174 * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4175 * points to just header, pull a few bytes of payload from
4176 * frags into skb->data
4178 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4180 * we do this workaround for ES2LAN, but it is un-necessary,
4181 * avoiding it could save a lot of cycles
4183 if (skb->data_len && (hdr_len == len)) {
4184 unsigned int pull_size;
4186 pull_size = min((unsigned int)4, skb->data_len);
4187 if (!__pskb_pull_tail(skb, pull_size)) {
4188 e_err("__pskb_pull_tail failed.\n");
4189 dev_kfree_skb_any(skb);
4190 return NETDEV_TX_OK;
4192 len = skb->len - skb->data_len;
4196 /* reserve a descriptor for the offload context */
4197 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4201 count += TXD_USE_COUNT(len, max_txd_pwr);
4203 nr_frags = skb_shinfo(skb)->nr_frags;
4204 for (f = 0; f < nr_frags; f++)
4205 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4208 if (adapter->hw.mac.tx_pkt_filtering)
4209 e1000_transfer_dhcp_info(adapter, skb);
4212 * need: count + 2 desc gap to keep tail from touching
4213 * head, otherwise try next time
4215 if (e1000_maybe_stop_tx(netdev, count + 2))
4216 return NETDEV_TX_BUSY;
4218 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4219 tx_flags |= E1000_TX_FLAGS_VLAN;
4220 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4223 first = tx_ring->next_to_use;
4225 tso = e1000_tso(adapter, skb);
4227 dev_kfree_skb_any(skb);
4228 return NETDEV_TX_OK;
4232 tx_flags |= E1000_TX_FLAGS_TSO;
4233 else if (e1000_tx_csum(adapter, skb))
4234 tx_flags |= E1000_TX_FLAGS_CSUM;
4237 * Old method was to assume IPv4 packet by default if TSO was enabled.
4238 * 82571 hardware supports TSO capabilities for IPv6 as well...
4239 * no longer assume, we must.
4241 if (skb->protocol == htons(ETH_P_IP))
4242 tx_flags |= E1000_TX_FLAGS_IPV4;
4244 /* if count is 0 then mapping error has occured */
4245 count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4247 e1000_tx_queue(adapter, tx_flags, count);
4248 /* Make sure there is space in the ring for the next send. */
4249 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4252 dev_kfree_skb_any(skb);
4253 tx_ring->buffer_info[first].time_stamp = 0;
4254 tx_ring->next_to_use = first;
4257 return NETDEV_TX_OK;
4261 * e1000_tx_timeout - Respond to a Tx Hang
4262 * @netdev: network interface device structure
4264 static void e1000_tx_timeout(struct net_device *netdev)
4266 struct e1000_adapter *adapter = netdev_priv(netdev);
4268 /* Do the reset outside of interrupt context */
4269 adapter->tx_timeout_count++;
4270 schedule_work(&adapter->reset_task);
4273 static void e1000_reset_task(struct work_struct *work)
4275 struct e1000_adapter *adapter;
4276 adapter = container_of(work, struct e1000_adapter, reset_task);
4278 e1000e_reinit_locked(adapter);
4282 * e1000_get_stats - Get System Network Statistics
4283 * @netdev: network interface device structure
4285 * Returns the address of the device statistics structure.
4286 * The statistics are actually updated from the timer callback.
4288 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4290 /* only return the current stats */
4291 return &netdev->stats;
4295 * e1000_change_mtu - Change the Maximum Transfer Unit
4296 * @netdev: network interface device structure
4297 * @new_mtu: new value for maximum frame size
4299 * Returns 0 on success, negative on failure
4301 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4303 struct e1000_adapter *adapter = netdev_priv(netdev);
4304 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4306 /* Jumbo frame support */
4307 if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4308 !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4309 e_err("Jumbo Frames not supported.\n");
4313 /* Supported frame sizes */
4314 if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4315 (max_frame > adapter->max_hw_frame_size)) {
4316 e_err("Unsupported MTU setting\n");
4320 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4322 /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
4323 adapter->max_frame_size = max_frame;
4324 e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4325 netdev->mtu = new_mtu;
4326 if (netif_running(netdev))
4327 e1000e_down(adapter);
4330 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4331 * means we reserve 2 more, this pushes us to allocate from the next
4333 * i.e. RXBUFFER_2048 --> size-4096 slab
4334 * However with the new *_jumbo_rx* routines, jumbo receives will use
4338 if (max_frame <= 2048)
4339 adapter->rx_buffer_len = 2048;
4341 adapter->rx_buffer_len = 4096;
4343 /* adjust allocation if LPE protects us, and we aren't using SBP */
4344 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4345 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4346 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4349 if (netif_running(netdev))
4352 e1000e_reset(adapter);
4354 clear_bit(__E1000_RESETTING, &adapter->state);
4359 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4362 struct e1000_adapter *adapter = netdev_priv(netdev);
4363 struct mii_ioctl_data *data = if_mii(ifr);
4365 if (adapter->hw.phy.media_type != e1000_media_type_copper)
4370 data->phy_id = adapter->hw.phy.addr;
4373 e1000_phy_read_status(adapter);
4375 switch (data->reg_num & 0x1F) {
4377 data->val_out = adapter->phy_regs.bmcr;
4380 data->val_out = adapter->phy_regs.bmsr;
4383 data->val_out = (adapter->hw.phy.id >> 16);
4386 data->val_out = (adapter->hw.phy.id & 0xFFFF);
4389 data->val_out = adapter->phy_regs.advertise;
4392 data->val_out = adapter->phy_regs.lpa;
4395 data->val_out = adapter->phy_regs.expansion;
4398 data->val_out = adapter->phy_regs.ctrl1000;
4401 data->val_out = adapter->phy_regs.stat1000;
4404 data->val_out = adapter->phy_regs.estatus;
4417 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4423 return e1000_mii_ioctl(netdev, ifr, cmd);
4429 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
4431 struct e1000_hw *hw = &adapter->hw;
4436 /* copy MAC RARs to PHY RARs */
4437 for (i = 0; i < adapter->hw.mac.rar_entry_count; i++) {
4438 mac_reg = er32(RAL(i));
4439 e1e_wphy(hw, BM_RAR_L(i), (u16)(mac_reg & 0xFFFF));
4440 e1e_wphy(hw, BM_RAR_M(i), (u16)((mac_reg >> 16) & 0xFFFF));
4441 mac_reg = er32(RAH(i));
4442 e1e_wphy(hw, BM_RAR_H(i), (u16)(mac_reg & 0xFFFF));
4443 e1e_wphy(hw, BM_RAR_CTRL(i), (u16)((mac_reg >> 16) & 0xFFFF));
4446 /* copy MAC MTA to PHY MTA */
4447 for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
4448 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
4449 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
4450 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
4453 /* configure PHY Rx Control register */
4454 e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
4455 mac_reg = er32(RCTL);
4456 if (mac_reg & E1000_RCTL_UPE)
4457 phy_reg |= BM_RCTL_UPE;
4458 if (mac_reg & E1000_RCTL_MPE)
4459 phy_reg |= BM_RCTL_MPE;
4460 phy_reg &= ~(BM_RCTL_MO_MASK);
4461 if (mac_reg & E1000_RCTL_MO_3)
4462 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
4463 << BM_RCTL_MO_SHIFT);
4464 if (mac_reg & E1000_RCTL_BAM)
4465 phy_reg |= BM_RCTL_BAM;
4466 if (mac_reg & E1000_RCTL_PMCF)
4467 phy_reg |= BM_RCTL_PMCF;
4468 mac_reg = er32(CTRL);
4469 if (mac_reg & E1000_CTRL_RFCE)
4470 phy_reg |= BM_RCTL_RFCE;
4471 e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
4473 /* enable PHY wakeup in MAC register */
4475 ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
4477 /* configure and enable PHY wakeup in PHY registers */
4478 e1e_wphy(&adapter->hw, BM_WUFC, wufc);
4479 e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
4481 /* activate PHY wakeup */
4482 retval = hw->phy.ops.acquire(hw);
4484 e_err("Could not acquire PHY\n");
4487 e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4488 (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
4489 retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
4491 e_err("Could not read PHY page 769\n");
4494 phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
4495 retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
4497 e_err("Could not set PHY Host Wakeup bit\n");
4499 hw->phy.ops.release(hw);
4504 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
4507 struct net_device *netdev = pci_get_drvdata(pdev);
4508 struct e1000_adapter *adapter = netdev_priv(netdev);
4509 struct e1000_hw *hw = &adapter->hw;
4510 u32 ctrl, ctrl_ext, rctl, status;
4511 /* Runtime suspend should only enable wakeup for link changes */
4512 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
4515 netif_device_detach(netdev);
4517 if (netif_running(netdev)) {
4518 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4519 e1000e_down(adapter);
4520 e1000_free_irq(adapter);
4522 e1000e_reset_interrupt_capability(adapter);
4524 retval = pci_save_state(pdev);
4528 status = er32(STATUS);
4529 if (status & E1000_STATUS_LU)
4530 wufc &= ~E1000_WUFC_LNKC;
4533 e1000_setup_rctl(adapter);
4534 e1000_set_multi(netdev);
4536 /* turn on all-multi mode if wake on multicast is enabled */
4537 if (wufc & E1000_WUFC_MC) {
4539 rctl |= E1000_RCTL_MPE;
4544 /* advertise wake from D3Cold */
4545 #define E1000_CTRL_ADVD3WUC 0x00100000
4546 /* phy power management enable */
4547 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4548 ctrl |= E1000_CTRL_ADVD3WUC;
4549 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
4550 ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
4553 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4554 adapter->hw.phy.media_type ==
4555 e1000_media_type_internal_serdes) {
4556 /* keep the laser running in D3 */
4557 ctrl_ext = er32(CTRL_EXT);
4558 ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
4559 ew32(CTRL_EXT, ctrl_ext);
4562 if (adapter->flags & FLAG_IS_ICH)
4563 e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4565 /* Allow time for pending master requests to run */
4566 e1000e_disable_pcie_master(&adapter->hw);
4568 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4569 /* enable wakeup by the PHY */
4570 retval = e1000_init_phy_wakeup(adapter, wufc);
4574 /* enable wakeup by the MAC */
4576 ew32(WUC, E1000_WUC_PME_EN);
4583 *enable_wake = !!wufc;
4585 /* make sure adapter isn't asleep if manageability is enabled */
4586 if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
4587 (hw->mac.ops.check_mng_mode(hw)))
4588 *enable_wake = true;
4590 if (adapter->hw.phy.type == e1000_phy_igp_3)
4591 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4594 * Release control of h/w to f/w. If f/w is AMT enabled, this
4595 * would have already happened in close and is redundant.
4597 e1000_release_hw_control(adapter);
4599 pci_disable_device(pdev);
4604 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
4606 if (sleep && wake) {
4607 pci_prepare_to_sleep(pdev);
4611 pci_wake_from_d3(pdev, wake);
4612 pci_set_power_state(pdev, PCI_D3hot);
4615 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
4618 struct net_device *netdev = pci_get_drvdata(pdev);
4619 struct e1000_adapter *adapter = netdev_priv(netdev);
4622 * The pci-e switch on some quad port adapters will report a
4623 * correctable error when the MAC transitions from D0 to D3. To
4624 * prevent this we need to mask off the correctable errors on the
4625 * downstream port of the pci-e switch.
4627 if (adapter->flags & FLAG_IS_QUAD_PORT) {
4628 struct pci_dev *us_dev = pdev->bus->self;
4629 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
4632 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
4633 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
4634 (devctl & ~PCI_EXP_DEVCTL_CERE));
4636 e1000_power_off(pdev, sleep, wake);
4638 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
4640 e1000_power_off(pdev, sleep, wake);
4644 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4650 * 82573 workaround - disable L1 ASPM on mobile chipsets
4652 * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4653 * resulting in lost data or garbage information on the pci-e link
4654 * level. This could result in (false) bad EEPROM checksum errors,
4655 * long ping times (up to 2s) or even a system freeze/hang.
4657 * Unfortunately this feature saves about 1W power consumption when
4660 pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4661 pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4663 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4665 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4669 #ifdef CONFIG_PM_OPS
4670 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
4672 return !!adapter->tx_ring->buffer_info;
4675 static int __e1000_resume(struct pci_dev *pdev)
4677 struct net_device *netdev = pci_get_drvdata(pdev);
4678 struct e1000_adapter *adapter = netdev_priv(netdev);
4679 struct e1000_hw *hw = &adapter->hw;
4682 e1000e_disable_l1aspm(pdev);
4684 e1000e_set_interrupt_capability(adapter);
4685 if (netif_running(netdev)) {
4686 err = e1000_request_irq(adapter);
4691 e1000e_power_up_phy(adapter);
4693 /* report the system wakeup cause from S3/S4 */
4694 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4697 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
4699 e_info("PHY Wakeup cause - %s\n",
4700 phy_data & E1000_WUS_EX ? "Unicast Packet" :
4701 phy_data & E1000_WUS_MC ? "Multicast Packet" :
4702 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
4703 phy_data & E1000_WUS_MAG ? "Magic Packet" :
4704 phy_data & E1000_WUS_LNKC ? "Link Status "
4705 " Change" : "other");
4707 e1e_wphy(&adapter->hw, BM_WUS, ~0);
4709 u32 wus = er32(WUS);
4711 e_info("MAC Wakeup cause - %s\n",
4712 wus & E1000_WUS_EX ? "Unicast Packet" :
4713 wus & E1000_WUS_MC ? "Multicast Packet" :
4714 wus & E1000_WUS_BC ? "Broadcast Packet" :
4715 wus & E1000_WUS_MAG ? "Magic Packet" :
4716 wus & E1000_WUS_LNKC ? "Link Status Change" :
4722 e1000e_reset(adapter);
4724 e1000_init_manageability(adapter);
4726 if (netif_running(netdev))
4729 netif_device_attach(netdev);
4732 * If the controller has AMT, do not set DRV_LOAD until the interface
4733 * is up. For all other cases, let the f/w know that the h/w is now
4734 * under the control of the driver.
4736 if (!(adapter->flags & FLAG_HAS_AMT))
4737 e1000_get_hw_control(adapter);
4742 #ifdef CONFIG_PM_SLEEP
4743 static int e1000_suspend(struct device *dev)
4745 struct pci_dev *pdev = to_pci_dev(dev);
4749 retval = __e1000_shutdown(pdev, &wake, false);
4751 e1000_complete_shutdown(pdev, true, wake);
4756 static int e1000_resume(struct device *dev)
4758 struct pci_dev *pdev = to_pci_dev(dev);
4759 struct net_device *netdev = pci_get_drvdata(pdev);
4760 struct e1000_adapter *adapter = netdev_priv(netdev);
4762 if (e1000e_pm_ready(adapter))
4763 adapter->idle_check = true;
4765 return __e1000_resume(pdev);
4767 #endif /* CONFIG_PM_SLEEP */
4769 #ifdef CONFIG_PM_RUNTIME
4770 static int e1000_runtime_suspend(struct device *dev)
4772 struct pci_dev *pdev = to_pci_dev(dev);
4773 struct net_device *netdev = pci_get_drvdata(pdev);
4774 struct e1000_adapter *adapter = netdev_priv(netdev);
4776 if (e1000e_pm_ready(adapter)) {
4779 __e1000_shutdown(pdev, &wake, true);
4785 static int e1000_idle(struct device *dev)
4787 struct pci_dev *pdev = to_pci_dev(dev);
4788 struct net_device *netdev = pci_get_drvdata(pdev);
4789 struct e1000_adapter *adapter = netdev_priv(netdev);
4791 if (!e1000e_pm_ready(adapter))
4794 if (adapter->idle_check) {
4795 adapter->idle_check = false;
4796 if (!e1000e_has_link(adapter))
4797 pm_schedule_suspend(dev, MSEC_PER_SEC);
4803 static int e1000_runtime_resume(struct device *dev)
4805 struct pci_dev *pdev = to_pci_dev(dev);
4806 struct net_device *netdev = pci_get_drvdata(pdev);
4807 struct e1000_adapter *adapter = netdev_priv(netdev);
4809 if (!e1000e_pm_ready(adapter))
4812 adapter->idle_check = !dev->power.runtime_auto;
4813 return __e1000_resume(pdev);
4815 #endif /* CONFIG_PM_RUNTIME */
4816 #endif /* CONFIG_PM_OPS */
4818 static void e1000_shutdown(struct pci_dev *pdev)
4822 __e1000_shutdown(pdev, &wake, false);
4824 if (system_state == SYSTEM_POWER_OFF)
4825 e1000_complete_shutdown(pdev, false, wake);
4828 #ifdef CONFIG_NET_POLL_CONTROLLER
4830 * Polling 'interrupt' - used by things like netconsole to send skbs
4831 * without having to re-enable interrupts. It's not called while
4832 * the interrupt routine is executing.
4834 static void e1000_netpoll(struct net_device *netdev)
4836 struct e1000_adapter *adapter = netdev_priv(netdev);
4838 disable_irq(adapter->pdev->irq);
4839 e1000_intr(adapter->pdev->irq, netdev);
4841 enable_irq(adapter->pdev->irq);
4846 * e1000_io_error_detected - called when PCI error is detected
4847 * @pdev: Pointer to PCI device
4848 * @state: The current pci connection state
4850 * This function is called after a PCI bus error affecting
4851 * this device has been detected.
4853 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4854 pci_channel_state_t state)
4856 struct net_device *netdev = pci_get_drvdata(pdev);
4857 struct e1000_adapter *adapter = netdev_priv(netdev);
4859 netif_device_detach(netdev);
4861 if (state == pci_channel_io_perm_failure)
4862 return PCI_ERS_RESULT_DISCONNECT;
4864 if (netif_running(netdev))
4865 e1000e_down(adapter);
4866 pci_disable_device(pdev);
4868 /* Request a slot slot reset. */
4869 return PCI_ERS_RESULT_NEED_RESET;
4873 * e1000_io_slot_reset - called after the pci bus has been reset.
4874 * @pdev: Pointer to PCI device
4876 * Restart the card from scratch, as if from a cold-boot. Implementation
4877 * resembles the first-half of the e1000_resume routine.
4879 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4881 struct net_device *netdev = pci_get_drvdata(pdev);
4882 struct e1000_adapter *adapter = netdev_priv(netdev);
4883 struct e1000_hw *hw = &adapter->hw;
4885 pci_ers_result_t result;
4887 e1000e_disable_l1aspm(pdev);
4888 err = pci_enable_device_mem(pdev);
4891 "Cannot re-enable PCI device after reset.\n");
4892 result = PCI_ERS_RESULT_DISCONNECT;
4894 pci_set_master(pdev);
4895 pdev->state_saved = true;
4896 pci_restore_state(pdev);
4898 pci_enable_wake(pdev, PCI_D3hot, 0);
4899 pci_enable_wake(pdev, PCI_D3cold, 0);
4901 e1000e_reset(adapter);
4903 result = PCI_ERS_RESULT_RECOVERED;
4906 pci_cleanup_aer_uncorrect_error_status(pdev);
4912 * e1000_io_resume - called when traffic can start flowing again.
4913 * @pdev: Pointer to PCI device
4915 * This callback is called when the error recovery driver tells us that
4916 * its OK to resume normal operation. Implementation resembles the
4917 * second-half of the e1000_resume routine.
4919 static void e1000_io_resume(struct pci_dev *pdev)
4921 struct net_device *netdev = pci_get_drvdata(pdev);
4922 struct e1000_adapter *adapter = netdev_priv(netdev);
4924 e1000_init_manageability(adapter);
4926 if (netif_running(netdev)) {
4927 if (e1000e_up(adapter)) {
4929 "can't bring device back up after reset\n");
4934 netif_device_attach(netdev);
4937 * If the controller has AMT, do not set DRV_LOAD until the interface
4938 * is up. For all other cases, let the f/w know that the h/w is now
4939 * under the control of the driver.
4941 if (!(adapter->flags & FLAG_HAS_AMT))
4942 e1000_get_hw_control(adapter);
4946 static void e1000_print_device_info(struct e1000_adapter *adapter)
4948 struct e1000_hw *hw = &adapter->hw;
4949 struct net_device *netdev = adapter->netdev;
4952 /* print bus type/speed/width info */
4953 e_info("(PCI Express:2.5GB/s:%s) %pM\n",
4955 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4959 e_info("Intel(R) PRO/%s Network Connection\n",
4960 (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4961 e1000e_read_pba_num(hw, &pba_num);
4962 e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4963 hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4966 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4968 struct e1000_hw *hw = &adapter->hw;
4972 if (hw->mac.type != e1000_82573)
4975 ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4976 if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
4977 /* Deep Smart Power Down (DSPD) */
4978 dev_warn(&adapter->pdev->dev,
4979 "Warning: detected DSPD enabled in EEPROM\n");
4982 ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4983 if (!ret_val && (le16_to_cpu(buf) & (3 << 2))) {
4985 dev_warn(&adapter->pdev->dev,
4986 "Warning: detected ASPM enabled in EEPROM\n");
4990 static const struct net_device_ops e1000e_netdev_ops = {
4991 .ndo_open = e1000_open,
4992 .ndo_stop = e1000_close,
4993 .ndo_start_xmit = e1000_xmit_frame,
4994 .ndo_get_stats = e1000_get_stats,
4995 .ndo_set_multicast_list = e1000_set_multi,
4996 .ndo_set_mac_address = e1000_set_mac,
4997 .ndo_change_mtu = e1000_change_mtu,
4998 .ndo_do_ioctl = e1000_ioctl,
4999 .ndo_tx_timeout = e1000_tx_timeout,
5000 .ndo_validate_addr = eth_validate_addr,
5002 .ndo_vlan_rx_register = e1000_vlan_rx_register,
5003 .ndo_vlan_rx_add_vid = e1000_vlan_rx_add_vid,
5004 .ndo_vlan_rx_kill_vid = e1000_vlan_rx_kill_vid,
5005 #ifdef CONFIG_NET_POLL_CONTROLLER
5006 .ndo_poll_controller = e1000_netpoll,
5011 * e1000_probe - Device Initialization Routine
5012 * @pdev: PCI device information struct
5013 * @ent: entry in e1000_pci_tbl
5015 * Returns 0 on success, negative on failure
5017 * e1000_probe initializes an adapter identified by a pci_dev structure.
5018 * The OS initialization, configuring of the adapter private structure,
5019 * and a hardware reset occur.
5021 static int __devinit e1000_probe(struct pci_dev *pdev,
5022 const struct pci_device_id *ent)
5024 struct net_device *netdev;
5025 struct e1000_adapter *adapter;
5026 struct e1000_hw *hw;
5027 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5028 resource_size_t mmio_start, mmio_len;
5029 resource_size_t flash_start, flash_len;
5031 static int cards_found;
5032 int i, err, pci_using_dac;
5033 u16 eeprom_data = 0;
5034 u16 eeprom_apme_mask = E1000_EEPROM_APME;
5036 e1000e_disable_l1aspm(pdev);
5038 err = pci_enable_device_mem(pdev);
5043 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
5045 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
5049 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
5051 err = pci_set_consistent_dma_mask(pdev,
5054 dev_err(&pdev->dev, "No usable DMA "
5055 "configuration, aborting\n");
5061 err = pci_request_selected_regions_exclusive(pdev,
5062 pci_select_bars(pdev, IORESOURCE_MEM),
5063 e1000e_driver_name);
5067 /* AER (Advanced Error Reporting) hooks */
5068 pci_enable_pcie_error_reporting(pdev);
5070 pci_set_master(pdev);
5071 /* PCI config space info */
5072 err = pci_save_state(pdev);
5074 goto err_alloc_etherdev;
5077 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5079 goto err_alloc_etherdev;
5081 SET_NETDEV_DEV(netdev, &pdev->dev);
5083 pci_set_drvdata(pdev, netdev);
5084 adapter = netdev_priv(netdev);
5086 adapter->netdev = netdev;
5087 adapter->pdev = pdev;
5089 adapter->pba = ei->pba;
5090 adapter->flags = ei->flags;
5091 adapter->flags2 = ei->flags2;
5092 adapter->hw.adapter = adapter;
5093 adapter->hw.mac.type = ei->mac;
5094 adapter->max_hw_frame_size = ei->max_hw_frame_size;
5095 adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5097 mmio_start = pci_resource_start(pdev, 0);
5098 mmio_len = pci_resource_len(pdev, 0);
5101 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5102 if (!adapter->hw.hw_addr)
5105 if ((adapter->flags & FLAG_HAS_FLASH) &&
5106 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5107 flash_start = pci_resource_start(pdev, 1);
5108 flash_len = pci_resource_len(pdev, 1);
5109 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5110 if (!adapter->hw.flash_address)
5114 /* construct the net_device struct */
5115 netdev->netdev_ops = &e1000e_netdev_ops;
5116 e1000e_set_ethtool_ops(netdev);
5117 netdev->watchdog_timeo = 5 * HZ;
5118 netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5119 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5121 netdev->mem_start = mmio_start;
5122 netdev->mem_end = mmio_start + mmio_len;
5124 adapter->bd_number = cards_found++;
5126 e1000e_check_options(adapter);
5128 /* setup adapter struct */
5129 err = e1000_sw_init(adapter);
5135 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5136 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5137 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5139 err = ei->get_variants(adapter);
5143 if ((adapter->flags & FLAG_IS_ICH) &&
5144 (adapter->flags & FLAG_READ_ONLY_NVM))
5145 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5147 hw->mac.ops.get_bus_info(&adapter->hw);
5149 adapter->hw.phy.autoneg_wait_to_complete = 0;
5151 /* Copper options */
5152 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5153 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5154 adapter->hw.phy.disable_polarity_correction = 0;
5155 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5158 if (e1000_check_reset_block(&adapter->hw))
5159 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5161 netdev->features = NETIF_F_SG |
5163 NETIF_F_HW_VLAN_TX |
5166 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5167 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5169 netdev->features |= NETIF_F_TSO;
5170 netdev->features |= NETIF_F_TSO6;
5172 netdev->vlan_features |= NETIF_F_TSO;
5173 netdev->vlan_features |= NETIF_F_TSO6;
5174 netdev->vlan_features |= NETIF_F_HW_CSUM;
5175 netdev->vlan_features |= NETIF_F_SG;
5178 netdev->features |= NETIF_F_HIGHDMA;
5180 if (e1000e_enable_mng_pass_thru(&adapter->hw))
5181 adapter->flags |= FLAG_MNG_PT_ENABLED;
5184 * before reading the NVM, reset the controller to
5185 * put the device in a known good starting state
5187 adapter->hw.mac.ops.reset_hw(&adapter->hw);
5190 * systems with ASPM and others may see the checksum fail on the first
5191 * attempt. Let's give it a few tries
5194 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5197 e_err("The NVM Checksum Is Not Valid\n");
5203 e1000_eeprom_checks(adapter);
5205 /* copy the MAC address */
5206 if (e1000e_read_mac_addr(&adapter->hw))
5207 e_err("NVM Read Error while reading MAC address\n");
5209 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5210 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5212 if (!is_valid_ether_addr(netdev->perm_addr)) {
5213 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5218 init_timer(&adapter->watchdog_timer);
5219 adapter->watchdog_timer.function = &e1000_watchdog;
5220 adapter->watchdog_timer.data = (unsigned long) adapter;
5222 init_timer(&adapter->phy_info_timer);
5223 adapter->phy_info_timer.function = &e1000_update_phy_info;
5224 adapter->phy_info_timer.data = (unsigned long) adapter;
5226 INIT_WORK(&adapter->reset_task, e1000_reset_task);
5227 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5228 INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5229 INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5230 INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5232 /* Initialize link parameters. User can change them with ethtool */
5233 adapter->hw.mac.autoneg = 1;
5234 adapter->fc_autoneg = 1;
5235 adapter->hw.fc.requested_mode = e1000_fc_default;
5236 adapter->hw.fc.current_mode = e1000_fc_default;
5237 adapter->hw.phy.autoneg_advertised = 0x2f;
5239 /* ring size defaults */
5240 adapter->rx_ring->count = 256;
5241 adapter->tx_ring->count = 256;
5244 * Initial Wake on LAN setting - If APM wake is enabled in
5245 * the EEPROM, enable the ACPI Magic Packet filter
5247 if (adapter->flags & FLAG_APME_IN_WUC) {
5248 /* APME bit in EEPROM is mapped to WUC.APME */
5249 eeprom_data = er32(WUC);
5250 eeprom_apme_mask = E1000_WUC_APME;
5251 if (eeprom_data & E1000_WUC_PHY_WAKE)
5252 adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5253 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5254 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5255 (adapter->hw.bus.func == 1))
5256 e1000_read_nvm(&adapter->hw,
5257 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5259 e1000_read_nvm(&adapter->hw,
5260 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5263 /* fetch WoL from EEPROM */
5264 if (eeprom_data & eeprom_apme_mask)
5265 adapter->eeprom_wol |= E1000_WUFC_MAG;
5268 * now that we have the eeprom settings, apply the special cases
5269 * where the eeprom may be wrong or the board simply won't support
5270 * wake on lan on a particular port
5272 if (!(adapter->flags & FLAG_HAS_WOL))
5273 adapter->eeprom_wol = 0;
5275 /* initialize the wol settings based on the eeprom settings */
5276 adapter->wol = adapter->eeprom_wol;
5277 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5279 /* save off EEPROM version number */
5280 e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5282 /* reset the hardware with the new settings */
5283 e1000e_reset(adapter);
5286 * If the controller has AMT, do not set DRV_LOAD until the interface
5287 * is up. For all other cases, let the f/w know that the h/w is now
5288 * under the control of the driver.
5290 if (!(adapter->flags & FLAG_HAS_AMT))
5291 e1000_get_hw_control(adapter);
5293 strcpy(netdev->name, "eth%d");
5294 err = register_netdev(netdev);
5298 /* carrier off reporting is important to ethtool even BEFORE open */
5299 netif_carrier_off(netdev);
5301 e1000_print_device_info(adapter);
5303 if (pci_dev_run_wake(pdev)) {
5304 pm_runtime_set_active(&pdev->dev);
5305 pm_runtime_enable(&pdev->dev);
5307 pm_schedule_suspend(&pdev->dev, MSEC_PER_SEC);
5312 if (!(adapter->flags & FLAG_HAS_AMT))
5313 e1000_release_hw_control(adapter);
5315 if (!e1000_check_reset_block(&adapter->hw))
5316 e1000_phy_hw_reset(&adapter->hw);
5319 kfree(adapter->tx_ring);
5320 kfree(adapter->rx_ring);
5322 if (adapter->hw.flash_address)
5323 iounmap(adapter->hw.flash_address);
5324 e1000e_reset_interrupt_capability(adapter);
5326 iounmap(adapter->hw.hw_addr);
5328 free_netdev(netdev);
5330 pci_release_selected_regions(pdev,
5331 pci_select_bars(pdev, IORESOURCE_MEM));
5334 pci_disable_device(pdev);
5339 * e1000_remove - Device Removal Routine
5340 * @pdev: PCI device information struct
5342 * e1000_remove is called by the PCI subsystem to alert the driver
5343 * that it should release a PCI device. The could be caused by a
5344 * Hot-Plug event, or because the driver is going to be removed from
5347 static void __devexit e1000_remove(struct pci_dev *pdev)
5349 struct net_device *netdev = pci_get_drvdata(pdev);
5350 struct e1000_adapter *adapter = netdev_priv(netdev);
5351 bool down = test_bit(__E1000_DOWN, &adapter->state);
5353 pm_runtime_get_sync(&pdev->dev);
5356 * flush_scheduled work may reschedule our watchdog task, so
5357 * explicitly disable watchdog tasks from being rescheduled
5360 set_bit(__E1000_DOWN, &adapter->state);
5361 del_timer_sync(&adapter->watchdog_timer);
5362 del_timer_sync(&adapter->phy_info_timer);
5364 cancel_work_sync(&adapter->reset_task);
5365 cancel_work_sync(&adapter->watchdog_task);
5366 cancel_work_sync(&adapter->downshift_task);
5367 cancel_work_sync(&adapter->update_phy_task);
5368 cancel_work_sync(&adapter->print_hang_task);
5369 flush_scheduled_work();
5371 if (!(netdev->flags & IFF_UP))
5372 e1000_power_down_phy(adapter);
5374 /* Don't lie to e1000_close() down the road. */
5376 clear_bit(__E1000_DOWN, &adapter->state);
5377 unregister_netdev(netdev);
5379 if (pci_dev_run_wake(pdev)) {
5380 pm_runtime_disable(&pdev->dev);
5381 pm_runtime_set_suspended(&pdev->dev);
5383 pm_runtime_put_noidle(&pdev->dev);
5386 * Release control of h/w to f/w. If f/w is AMT enabled, this
5387 * would have already happened in close and is redundant.
5389 e1000_release_hw_control(adapter);
5391 e1000e_reset_interrupt_capability(adapter);
5392 kfree(adapter->tx_ring);
5393 kfree(adapter->rx_ring);
5395 iounmap(adapter->hw.hw_addr);
5396 if (adapter->hw.flash_address)
5397 iounmap(adapter->hw.flash_address);
5398 pci_release_selected_regions(pdev,
5399 pci_select_bars(pdev, IORESOURCE_MEM));
5401 free_netdev(netdev);
5404 pci_disable_pcie_error_reporting(pdev);
5406 pci_disable_device(pdev);
5409 /* PCI Error Recovery (ERS) */
5410 static struct pci_error_handlers e1000_err_handler = {
5411 .error_detected = e1000_io_error_detected,
5412 .slot_reset = e1000_io_slot_reset,
5413 .resume = e1000_io_resume,
5416 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
5417 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5418 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5419 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5420 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5421 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5422 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5423 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5424 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5425 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5427 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5428 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5429 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5430 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5432 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5433 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5434 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5436 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5437 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
5438 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
5440 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5441 board_80003es2lan },
5442 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5443 board_80003es2lan },
5444 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5445 board_80003es2lan },
5446 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5447 board_80003es2lan },
5449 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5450 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5451 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5452 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5453 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5454 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5455 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5456 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
5458 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5459 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5460 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5461 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5462 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5463 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5464 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5465 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5466 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5468 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5469 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5470 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5472 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5473 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5475 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
5476 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
5477 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
5478 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
5480 { } /* terminate list */
5482 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5484 #ifdef CONFIG_PM_OPS
5485 static const struct dev_pm_ops e1000_pm_ops = {
5486 SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
5487 SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
5488 e1000_runtime_resume, e1000_idle)
5492 /* PCI Device API Driver */
5493 static struct pci_driver e1000_driver = {
5494 .name = e1000e_driver_name,
5495 .id_table = e1000_pci_tbl,
5496 .probe = e1000_probe,
5497 .remove = __devexit_p(e1000_remove),
5498 #ifdef CONFIG_PM_OPS
5499 .driver.pm = &e1000_pm_ops,
5501 .shutdown = e1000_shutdown,
5502 .err_handler = &e1000_err_handler
5506 * e1000_init_module - Driver Registration Routine
5508 * e1000_init_module is the first routine called when the driver is
5509 * loaded. All it does is register with the PCI subsystem.
5511 static int __init e1000_init_module(void)
5514 pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
5515 e1000e_driver_version);
5516 pr_info("Copyright (c) 1999 - 2009 Intel Corporation.\n");
5517 ret = pci_register_driver(&e1000_driver);
5521 module_init(e1000_init_module);
5524 * e1000_exit_module - Driver Exit Cleanup Routine
5526 * e1000_exit_module is called just before the driver is removed
5529 static void __exit e1000_exit_module(void)
5531 pci_unregister_driver(&e1000_driver);
5533 module_exit(e1000_exit_module);
5536 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5537 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5538 MODULE_LICENSE("GPL");
5539 MODULE_VERSION(DRV_VERSION);