1 /*******************************************************************************
3 Intel PRO/1000 Linux driver
4 Copyright(c) 1999 - 2009 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 Linux NICS <linux.nics@intel.com>
24 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
27 *******************************************************************************/
29 #include <linux/module.h>
30 #include <linux/types.h>
31 #include <linux/init.h>
32 #include <linux/pci.h>
33 #include <linux/vmalloc.h>
34 #include <linux/pagemap.h>
35 #include <linux/delay.h>
36 #include <linux/netdevice.h>
37 #include <linux/tcp.h>
38 #include <linux/ipv6.h>
39 #include <net/checksum.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/mii.h>
42 #include <linux/ethtool.h>
43 #include <linux/if_vlan.h>
44 #include <linux/cpu.h>
45 #include <linux/smp.h>
46 #include <linux/pm_qos_params.h>
47 #include <linux/aer.h>
51 #define DRV_VERSION "1.0.2-k2"
52 char e1000e_driver_name[] = "e1000e";
53 const char e1000e_driver_version[] = DRV_VERSION;
55 static const struct e1000_info *e1000_info_tbl[] = {
56 [board_82571] = &e1000_82571_info,
57 [board_82572] = &e1000_82572_info,
58 [board_82573] = &e1000_82573_info,
59 [board_82574] = &e1000_82574_info,
60 [board_82583] = &e1000_82583_info,
61 [board_80003es2lan] = &e1000_es2_info,
62 [board_ich8lan] = &e1000_ich8_info,
63 [board_ich9lan] = &e1000_ich9_info,
64 [board_ich10lan] = &e1000_ich10_info,
65 [board_pchlan] = &e1000_pch_info,
69 * e1000_desc_unused - calculate if we have unused descriptors
71 static int e1000_desc_unused(struct e1000_ring *ring)
73 if (ring->next_to_clean > ring->next_to_use)
74 return ring->next_to_clean - ring->next_to_use - 1;
76 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
80 * e1000_receive_skb - helper function to handle Rx indications
81 * @adapter: board private structure
82 * @status: descriptor status field as written by hardware
83 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
84 * @skb: pointer to sk_buff to be indicated to stack
86 static void e1000_receive_skb(struct e1000_adapter *adapter,
87 struct net_device *netdev,
89 u8 status, __le16 vlan)
91 skb->protocol = eth_type_trans(skb, netdev);
93 if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
94 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
95 le16_to_cpu(vlan), skb);
97 napi_gro_receive(&adapter->napi, skb);
101 * e1000_rx_checksum - Receive Checksum Offload for 82543
102 * @adapter: board private structure
103 * @status_err: receive descriptor status and error fields
104 * @csum: receive descriptor csum field
105 * @sk_buff: socket buffer with received data
107 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
108 u32 csum, struct sk_buff *skb)
110 u16 status = (u16)status_err;
111 u8 errors = (u8)(status_err >> 24);
112 skb->ip_summed = CHECKSUM_NONE;
114 /* Ignore Checksum bit is set */
115 if (status & E1000_RXD_STAT_IXSM)
117 /* TCP/UDP checksum error bit is set */
118 if (errors & E1000_RXD_ERR_TCPE) {
119 /* let the stack verify checksum errors */
120 adapter->hw_csum_err++;
124 /* TCP/UDP Checksum has not been calculated */
125 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
128 /* It must be a TCP or UDP packet with a valid checksum */
129 if (status & E1000_RXD_STAT_TCPCS) {
130 /* TCP checksum is good */
131 skb->ip_summed = CHECKSUM_UNNECESSARY;
134 * IP fragment with UDP payload
135 * Hardware complements the payload checksum, so we undo it
136 * and then put the value in host order for further stack use.
138 __sum16 sum = (__force __sum16)htons(csum);
139 skb->csum = csum_unfold(~sum);
140 skb->ip_summed = CHECKSUM_COMPLETE;
142 adapter->hw_csum_good++;
146 * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
147 * @adapter: address of board private structure
149 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
152 struct net_device *netdev = adapter->netdev;
153 struct pci_dev *pdev = adapter->pdev;
154 struct e1000_ring *rx_ring = adapter->rx_ring;
155 struct e1000_rx_desc *rx_desc;
156 struct e1000_buffer *buffer_info;
159 unsigned int bufsz = adapter->rx_buffer_len;
161 i = rx_ring->next_to_use;
162 buffer_info = &rx_ring->buffer_info[i];
164 while (cleaned_count--) {
165 skb = buffer_info->skb;
171 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
173 /* Better luck next round */
174 adapter->alloc_rx_buff_failed++;
178 buffer_info->skb = skb;
180 buffer_info->dma = pci_map_single(pdev, skb->data,
181 adapter->rx_buffer_len,
183 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
184 dev_err(&pdev->dev, "RX DMA map failed\n");
185 adapter->rx_dma_failed++;
189 rx_desc = E1000_RX_DESC(*rx_ring, i);
190 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
193 if (i == rx_ring->count)
195 buffer_info = &rx_ring->buffer_info[i];
198 if (rx_ring->next_to_use != i) {
199 rx_ring->next_to_use = i;
201 i = (rx_ring->count - 1);
204 * Force memory writes to complete before letting h/w
205 * know there are new descriptors to fetch. (Only
206 * applicable for weak-ordered memory model archs,
210 writel(i, adapter->hw.hw_addr + rx_ring->tail);
215 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
216 * @adapter: address of board private structure
218 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
221 struct net_device *netdev = adapter->netdev;
222 struct pci_dev *pdev = adapter->pdev;
223 union e1000_rx_desc_packet_split *rx_desc;
224 struct e1000_ring *rx_ring = adapter->rx_ring;
225 struct e1000_buffer *buffer_info;
226 struct e1000_ps_page *ps_page;
230 i = rx_ring->next_to_use;
231 buffer_info = &rx_ring->buffer_info[i];
233 while (cleaned_count--) {
234 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
236 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
237 ps_page = &buffer_info->ps_pages[j];
238 if (j >= adapter->rx_ps_pages) {
239 /* all unused desc entries get hw null ptr */
240 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
243 if (!ps_page->page) {
244 ps_page->page = alloc_page(GFP_ATOMIC);
245 if (!ps_page->page) {
246 adapter->alloc_rx_buff_failed++;
249 ps_page->dma = pci_map_page(pdev,
253 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
254 dev_err(&adapter->pdev->dev,
255 "RX DMA page map failed\n");
256 adapter->rx_dma_failed++;
261 * Refresh the desc even if buffer_addrs
262 * didn't change because each write-back
265 rx_desc->read.buffer_addr[j+1] =
266 cpu_to_le64(ps_page->dma);
269 skb = netdev_alloc_skb_ip_align(netdev,
270 adapter->rx_ps_bsize0);
273 adapter->alloc_rx_buff_failed++;
277 buffer_info->skb = skb;
278 buffer_info->dma = pci_map_single(pdev, skb->data,
279 adapter->rx_ps_bsize0,
281 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
282 dev_err(&pdev->dev, "RX DMA map failed\n");
283 adapter->rx_dma_failed++;
285 dev_kfree_skb_any(skb);
286 buffer_info->skb = NULL;
290 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
293 if (i == rx_ring->count)
295 buffer_info = &rx_ring->buffer_info[i];
299 if (rx_ring->next_to_use != i) {
300 rx_ring->next_to_use = i;
303 i = (rx_ring->count - 1);
306 * Force memory writes to complete before letting h/w
307 * know there are new descriptors to fetch. (Only
308 * applicable for weak-ordered memory model archs,
313 * Hardware increments by 16 bytes, but packet split
314 * descriptors are 32 bytes...so we increment tail
317 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
322 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
323 * @adapter: address of board private structure
324 * @cleaned_count: number of buffers to allocate this pass
327 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
330 struct net_device *netdev = adapter->netdev;
331 struct pci_dev *pdev = adapter->pdev;
332 struct e1000_rx_desc *rx_desc;
333 struct e1000_ring *rx_ring = adapter->rx_ring;
334 struct e1000_buffer *buffer_info;
337 unsigned int bufsz = 256 - 16 /* for skb_reserve */;
339 i = rx_ring->next_to_use;
340 buffer_info = &rx_ring->buffer_info[i];
342 while (cleaned_count--) {
343 skb = buffer_info->skb;
349 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
350 if (unlikely(!skb)) {
351 /* Better luck next round */
352 adapter->alloc_rx_buff_failed++;
356 buffer_info->skb = skb;
358 /* allocate a new page if necessary */
359 if (!buffer_info->page) {
360 buffer_info->page = alloc_page(GFP_ATOMIC);
361 if (unlikely(!buffer_info->page)) {
362 adapter->alloc_rx_buff_failed++;
367 if (!buffer_info->dma)
368 buffer_info->dma = pci_map_page(pdev,
369 buffer_info->page, 0,
373 rx_desc = E1000_RX_DESC(*rx_ring, i);
374 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
376 if (unlikely(++i == rx_ring->count))
378 buffer_info = &rx_ring->buffer_info[i];
381 if (likely(rx_ring->next_to_use != i)) {
382 rx_ring->next_to_use = i;
383 if (unlikely(i-- == 0))
384 i = (rx_ring->count - 1);
386 /* Force memory writes to complete before letting h/w
387 * know there are new descriptors to fetch. (Only
388 * applicable for weak-ordered memory model archs,
391 writel(i, adapter->hw.hw_addr + rx_ring->tail);
396 * e1000_clean_rx_irq - Send received data up the network stack; legacy
397 * @adapter: board private structure
399 * the return value indicates whether actual cleaning was done, there
400 * is no guarantee that everything was cleaned
402 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
403 int *work_done, int work_to_do)
405 struct net_device *netdev = adapter->netdev;
406 struct pci_dev *pdev = adapter->pdev;
407 struct e1000_hw *hw = &adapter->hw;
408 struct e1000_ring *rx_ring = adapter->rx_ring;
409 struct e1000_rx_desc *rx_desc, *next_rxd;
410 struct e1000_buffer *buffer_info, *next_buffer;
413 int cleaned_count = 0;
415 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
417 i = rx_ring->next_to_clean;
418 rx_desc = E1000_RX_DESC(*rx_ring, i);
419 buffer_info = &rx_ring->buffer_info[i];
421 while (rx_desc->status & E1000_RXD_STAT_DD) {
425 if (*work_done >= work_to_do)
429 status = rx_desc->status;
430 skb = buffer_info->skb;
431 buffer_info->skb = NULL;
433 prefetch(skb->data - NET_IP_ALIGN);
436 if (i == rx_ring->count)
438 next_rxd = E1000_RX_DESC(*rx_ring, i);
441 next_buffer = &rx_ring->buffer_info[i];
445 pci_unmap_single(pdev,
447 adapter->rx_buffer_len,
449 buffer_info->dma = 0;
451 length = le16_to_cpu(rx_desc->length);
453 /* !EOP means multiple descriptors were used to store a single
454 * packet, also make sure the frame isn't just CRC only */
455 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
456 /* All receives must fit into a single buffer */
457 e_dbg("Receive packet consumed multiple buffers\n");
459 buffer_info->skb = skb;
463 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
465 buffer_info->skb = skb;
469 /* adjust length to remove Ethernet CRC */
470 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
473 total_rx_bytes += length;
477 * code added for copybreak, this should improve
478 * performance for small packets with large amounts
479 * of reassembly being done in the stack
481 if (length < copybreak) {
482 struct sk_buff *new_skb =
483 netdev_alloc_skb_ip_align(netdev, length);
485 skb_copy_to_linear_data_offset(new_skb,
491 /* save the skb in buffer_info as good */
492 buffer_info->skb = skb;
495 /* else just continue with the old one */
497 /* end copybreak code */
498 skb_put(skb, length);
500 /* Receive Checksum Offload */
501 e1000_rx_checksum(adapter,
503 ((u32)(rx_desc->errors) << 24),
504 le16_to_cpu(rx_desc->csum), skb);
506 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
511 /* return some buffers to hardware, one at a time is too slow */
512 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
513 adapter->alloc_rx_buf(adapter, cleaned_count);
517 /* use prefetched values */
519 buffer_info = next_buffer;
521 rx_ring->next_to_clean = i;
523 cleaned_count = e1000_desc_unused(rx_ring);
525 adapter->alloc_rx_buf(adapter, cleaned_count);
527 adapter->total_rx_bytes += total_rx_bytes;
528 adapter->total_rx_packets += total_rx_packets;
529 netdev->stats.rx_bytes += total_rx_bytes;
530 netdev->stats.rx_packets += total_rx_packets;
534 static void e1000_put_txbuf(struct e1000_adapter *adapter,
535 struct e1000_buffer *buffer_info)
537 buffer_info->dma = 0;
538 if (buffer_info->skb) {
539 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
541 dev_kfree_skb_any(buffer_info->skb);
542 buffer_info->skb = NULL;
544 buffer_info->time_stamp = 0;
547 static void e1000_print_hw_hang(struct work_struct *work)
549 struct e1000_adapter *adapter = container_of(work,
550 struct e1000_adapter,
552 struct e1000_ring *tx_ring = adapter->tx_ring;
553 unsigned int i = tx_ring->next_to_clean;
554 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
555 struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
556 struct e1000_hw *hw = &adapter->hw;
557 u16 phy_status, phy_1000t_status, phy_ext_status;
560 e1e_rphy(hw, PHY_STATUS, &phy_status);
561 e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
562 e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
564 pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
566 /* detected Hardware unit hang */
567 e_err("Detected Hardware Unit Hang:\n"
570 " next_to_use <%x>\n"
571 " next_to_clean <%x>\n"
572 "buffer_info[next_to_clean]:\n"
573 " time_stamp <%lx>\n"
574 " next_to_watch <%x>\n"
576 " next_to_watch.status <%x>\n"
579 "PHY 1000BASE-T Status <%x>\n"
580 "PHY Extended Status <%x>\n"
582 readl(adapter->hw.hw_addr + tx_ring->head),
583 readl(adapter->hw.hw_addr + tx_ring->tail),
584 tx_ring->next_to_use,
585 tx_ring->next_to_clean,
586 tx_ring->buffer_info[eop].time_stamp,
589 eop_desc->upper.fields.status,
598 * e1000_clean_tx_irq - Reclaim resources after transmit completes
599 * @adapter: board private structure
601 * the return value indicates whether actual cleaning was done, there
602 * is no guarantee that everything was cleaned
604 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
606 struct net_device *netdev = adapter->netdev;
607 struct e1000_hw *hw = &adapter->hw;
608 struct e1000_ring *tx_ring = adapter->tx_ring;
609 struct e1000_tx_desc *tx_desc, *eop_desc;
610 struct e1000_buffer *buffer_info;
612 unsigned int count = 0;
613 unsigned int total_tx_bytes = 0, total_tx_packets = 0;
615 i = tx_ring->next_to_clean;
616 eop = tx_ring->buffer_info[i].next_to_watch;
617 eop_desc = E1000_TX_DESC(*tx_ring, eop);
619 while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
620 (count < tx_ring->count)) {
621 bool cleaned = false;
622 for (; !cleaned; count++) {
623 tx_desc = E1000_TX_DESC(*tx_ring, i);
624 buffer_info = &tx_ring->buffer_info[i];
625 cleaned = (i == eop);
628 struct sk_buff *skb = buffer_info->skb;
629 unsigned int segs, bytecount;
630 segs = skb_shinfo(skb)->gso_segs ?: 1;
631 /* multiply data chunks by size of headers */
632 bytecount = ((segs - 1) * skb_headlen(skb)) +
634 total_tx_packets += segs;
635 total_tx_bytes += bytecount;
638 e1000_put_txbuf(adapter, buffer_info);
639 tx_desc->upper.data = 0;
642 if (i == tx_ring->count)
646 eop = tx_ring->buffer_info[i].next_to_watch;
647 eop_desc = E1000_TX_DESC(*tx_ring, eop);
650 tx_ring->next_to_clean = i;
652 #define TX_WAKE_THRESHOLD 32
653 if (count && netif_carrier_ok(netdev) &&
654 e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
655 /* Make sure that anybody stopping the queue after this
656 * sees the new next_to_clean.
660 if (netif_queue_stopped(netdev) &&
661 !(test_bit(__E1000_DOWN, &adapter->state))) {
662 netif_wake_queue(netdev);
663 ++adapter->restart_queue;
667 if (adapter->detect_tx_hung) {
669 * Detect a transmit hang in hardware, this serializes the
670 * check with the clearing of time_stamp and movement of i
672 adapter->detect_tx_hung = 0;
673 if (tx_ring->buffer_info[i].time_stamp &&
674 time_after(jiffies, tx_ring->buffer_info[i].time_stamp
675 + (adapter->tx_timeout_factor * HZ))
676 && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
677 schedule_work(&adapter->print_hang_task);
678 netif_stop_queue(netdev);
681 adapter->total_tx_bytes += total_tx_bytes;
682 adapter->total_tx_packets += total_tx_packets;
683 netdev->stats.tx_bytes += total_tx_bytes;
684 netdev->stats.tx_packets += total_tx_packets;
685 return (count < tx_ring->count);
689 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
690 * @adapter: board private structure
692 * the return value indicates whether actual cleaning was done, there
693 * is no guarantee that everything was cleaned
695 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
696 int *work_done, int work_to_do)
698 struct e1000_hw *hw = &adapter->hw;
699 union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
700 struct net_device *netdev = adapter->netdev;
701 struct pci_dev *pdev = adapter->pdev;
702 struct e1000_ring *rx_ring = adapter->rx_ring;
703 struct e1000_buffer *buffer_info, *next_buffer;
704 struct e1000_ps_page *ps_page;
708 int cleaned_count = 0;
710 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
712 i = rx_ring->next_to_clean;
713 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
714 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
715 buffer_info = &rx_ring->buffer_info[i];
717 while (staterr & E1000_RXD_STAT_DD) {
718 if (*work_done >= work_to_do)
721 skb = buffer_info->skb;
723 /* in the packet split case this is header only */
724 prefetch(skb->data - NET_IP_ALIGN);
727 if (i == rx_ring->count)
729 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
732 next_buffer = &rx_ring->buffer_info[i];
736 pci_unmap_single(pdev, buffer_info->dma,
737 adapter->rx_ps_bsize0,
739 buffer_info->dma = 0;
741 if (!(staterr & E1000_RXD_STAT_EOP)) {
742 e_dbg("Packet Split buffers didn't pick up the full "
744 dev_kfree_skb_irq(skb);
748 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
749 dev_kfree_skb_irq(skb);
753 length = le16_to_cpu(rx_desc->wb.middle.length0);
756 e_dbg("Last part of the packet spanning multiple "
758 dev_kfree_skb_irq(skb);
763 skb_put(skb, length);
767 * this looks ugly, but it seems compiler issues make it
768 * more efficient than reusing j
770 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
773 * page alloc/put takes too long and effects small packet
774 * throughput, so unsplit small packets and save the alloc/put
775 * only valid in softirq (napi) context to call kmap_*
777 if (l1 && (l1 <= copybreak) &&
778 ((length + l1) <= adapter->rx_ps_bsize0)) {
781 ps_page = &buffer_info->ps_pages[0];
784 * there is no documentation about how to call
785 * kmap_atomic, so we can't hold the mapping
788 pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
789 PAGE_SIZE, PCI_DMA_FROMDEVICE);
790 vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
791 memcpy(skb_tail_pointer(skb), vaddr, l1);
792 kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
793 pci_dma_sync_single_for_device(pdev, ps_page->dma,
794 PAGE_SIZE, PCI_DMA_FROMDEVICE);
797 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
805 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
806 length = le16_to_cpu(rx_desc->wb.upper.length[j]);
810 ps_page = &buffer_info->ps_pages[j];
811 pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
814 skb_fill_page_desc(skb, j, ps_page->page, 0, length);
815 ps_page->page = NULL;
817 skb->data_len += length;
818 skb->truesize += length;
821 /* strip the ethernet crc, problem is we're using pages now so
822 * this whole operation can get a little cpu intensive
824 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
825 pskb_trim(skb, skb->len - 4);
828 total_rx_bytes += skb->len;
831 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
832 rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
834 if (rx_desc->wb.upper.header_status &
835 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
836 adapter->rx_hdr_split++;
838 e1000_receive_skb(adapter, netdev, skb,
839 staterr, rx_desc->wb.middle.vlan);
842 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
843 buffer_info->skb = NULL;
845 /* return some buffers to hardware, one at a time is too slow */
846 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
847 adapter->alloc_rx_buf(adapter, cleaned_count);
851 /* use prefetched values */
853 buffer_info = next_buffer;
855 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
857 rx_ring->next_to_clean = i;
859 cleaned_count = e1000_desc_unused(rx_ring);
861 adapter->alloc_rx_buf(adapter, cleaned_count);
863 adapter->total_rx_bytes += total_rx_bytes;
864 adapter->total_rx_packets += total_rx_packets;
865 netdev->stats.rx_bytes += total_rx_bytes;
866 netdev->stats.rx_packets += total_rx_packets;
871 * e1000_consume_page - helper function
873 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
878 skb->data_len += length;
879 skb->truesize += length;
883 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
884 * @adapter: board private structure
886 * the return value indicates whether actual cleaning was done, there
887 * is no guarantee that everything was cleaned
890 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
891 int *work_done, int work_to_do)
893 struct net_device *netdev = adapter->netdev;
894 struct pci_dev *pdev = adapter->pdev;
895 struct e1000_ring *rx_ring = adapter->rx_ring;
896 struct e1000_rx_desc *rx_desc, *next_rxd;
897 struct e1000_buffer *buffer_info, *next_buffer;
900 int cleaned_count = 0;
901 bool cleaned = false;
902 unsigned int total_rx_bytes=0, total_rx_packets=0;
904 i = rx_ring->next_to_clean;
905 rx_desc = E1000_RX_DESC(*rx_ring, i);
906 buffer_info = &rx_ring->buffer_info[i];
908 while (rx_desc->status & E1000_RXD_STAT_DD) {
912 if (*work_done >= work_to_do)
916 status = rx_desc->status;
917 skb = buffer_info->skb;
918 buffer_info->skb = NULL;
921 if (i == rx_ring->count)
923 next_rxd = E1000_RX_DESC(*rx_ring, i);
926 next_buffer = &rx_ring->buffer_info[i];
930 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
932 buffer_info->dma = 0;
934 length = le16_to_cpu(rx_desc->length);
936 /* errors is only valid for DD + EOP descriptors */
937 if (unlikely((status & E1000_RXD_STAT_EOP) &&
938 (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
939 /* recycle both page and skb */
940 buffer_info->skb = skb;
941 /* an error means any chain goes out the window
943 if (rx_ring->rx_skb_top)
944 dev_kfree_skb(rx_ring->rx_skb_top);
945 rx_ring->rx_skb_top = NULL;
949 #define rxtop rx_ring->rx_skb_top
950 if (!(status & E1000_RXD_STAT_EOP)) {
951 /* this descriptor is only the beginning (or middle) */
953 /* this is the beginning of a chain */
955 skb_fill_page_desc(rxtop, 0, buffer_info->page,
958 /* this is the middle of a chain */
959 skb_fill_page_desc(rxtop,
960 skb_shinfo(rxtop)->nr_frags,
961 buffer_info->page, 0, length);
962 /* re-use the skb, only consumed the page */
963 buffer_info->skb = skb;
965 e1000_consume_page(buffer_info, rxtop, length);
969 /* end of the chain */
970 skb_fill_page_desc(rxtop,
971 skb_shinfo(rxtop)->nr_frags,
972 buffer_info->page, 0, length);
973 /* re-use the current skb, we only consumed the
975 buffer_info->skb = skb;
978 e1000_consume_page(buffer_info, skb, length);
980 /* no chain, got EOP, this buf is the packet
981 * copybreak to save the put_page/alloc_page */
982 if (length <= copybreak &&
983 skb_tailroom(skb) >= length) {
985 vaddr = kmap_atomic(buffer_info->page,
986 KM_SKB_DATA_SOFTIRQ);
987 memcpy(skb_tail_pointer(skb), vaddr,
990 KM_SKB_DATA_SOFTIRQ);
991 /* re-use the page, so don't erase
992 * buffer_info->page */
993 skb_put(skb, length);
995 skb_fill_page_desc(skb, 0,
996 buffer_info->page, 0,
998 e1000_consume_page(buffer_info, skb,
1004 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1005 e1000_rx_checksum(adapter,
1007 ((u32)(rx_desc->errors) << 24),
1008 le16_to_cpu(rx_desc->csum), skb);
1010 /* probably a little skewed due to removing CRC */
1011 total_rx_bytes += skb->len;
1014 /* eth type trans needs skb->data to point to something */
1015 if (!pskb_may_pull(skb, ETH_HLEN)) {
1016 e_err("pskb_may_pull failed.\n");
1021 e1000_receive_skb(adapter, netdev, skb, status,
1025 rx_desc->status = 0;
1027 /* return some buffers to hardware, one at a time is too slow */
1028 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1029 adapter->alloc_rx_buf(adapter, cleaned_count);
1033 /* use prefetched values */
1035 buffer_info = next_buffer;
1037 rx_ring->next_to_clean = i;
1039 cleaned_count = e1000_desc_unused(rx_ring);
1041 adapter->alloc_rx_buf(adapter, cleaned_count);
1043 adapter->total_rx_bytes += total_rx_bytes;
1044 adapter->total_rx_packets += total_rx_packets;
1045 netdev->stats.rx_bytes += total_rx_bytes;
1046 netdev->stats.rx_packets += total_rx_packets;
1051 * e1000_clean_rx_ring - Free Rx Buffers per Queue
1052 * @adapter: board private structure
1054 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1056 struct e1000_ring *rx_ring = adapter->rx_ring;
1057 struct e1000_buffer *buffer_info;
1058 struct e1000_ps_page *ps_page;
1059 struct pci_dev *pdev = adapter->pdev;
1062 /* Free all the Rx ring sk_buffs */
1063 for (i = 0; i < rx_ring->count; i++) {
1064 buffer_info = &rx_ring->buffer_info[i];
1065 if (buffer_info->dma) {
1066 if (adapter->clean_rx == e1000_clean_rx_irq)
1067 pci_unmap_single(pdev, buffer_info->dma,
1068 adapter->rx_buffer_len,
1069 PCI_DMA_FROMDEVICE);
1070 else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1071 pci_unmap_page(pdev, buffer_info->dma,
1073 PCI_DMA_FROMDEVICE);
1074 else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1075 pci_unmap_single(pdev, buffer_info->dma,
1076 adapter->rx_ps_bsize0,
1077 PCI_DMA_FROMDEVICE);
1078 buffer_info->dma = 0;
1081 if (buffer_info->page) {
1082 put_page(buffer_info->page);
1083 buffer_info->page = NULL;
1086 if (buffer_info->skb) {
1087 dev_kfree_skb(buffer_info->skb);
1088 buffer_info->skb = NULL;
1091 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1092 ps_page = &buffer_info->ps_pages[j];
1095 pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1096 PCI_DMA_FROMDEVICE);
1098 put_page(ps_page->page);
1099 ps_page->page = NULL;
1103 /* there also may be some cached data from a chained receive */
1104 if (rx_ring->rx_skb_top) {
1105 dev_kfree_skb(rx_ring->rx_skb_top);
1106 rx_ring->rx_skb_top = NULL;
1109 /* Zero out the descriptor ring */
1110 memset(rx_ring->desc, 0, rx_ring->size);
1112 rx_ring->next_to_clean = 0;
1113 rx_ring->next_to_use = 0;
1115 writel(0, adapter->hw.hw_addr + rx_ring->head);
1116 writel(0, adapter->hw.hw_addr + rx_ring->tail);
1119 static void e1000e_downshift_workaround(struct work_struct *work)
1121 struct e1000_adapter *adapter = container_of(work,
1122 struct e1000_adapter, downshift_task);
1124 e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1128 * e1000_intr_msi - Interrupt Handler
1129 * @irq: interrupt number
1130 * @data: pointer to a network interface device structure
1132 static irqreturn_t e1000_intr_msi(int irq, void *data)
1134 struct net_device *netdev = data;
1135 struct e1000_adapter *adapter = netdev_priv(netdev);
1136 struct e1000_hw *hw = &adapter->hw;
1137 u32 icr = er32(ICR);
1140 * read ICR disables interrupts using IAM
1143 if (icr & E1000_ICR_LSC) {
1144 hw->mac.get_link_status = 1;
1146 * ICH8 workaround-- Call gig speed drop workaround on cable
1147 * disconnect (LSC) before accessing any PHY registers
1149 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1150 (!(er32(STATUS) & E1000_STATUS_LU)))
1151 schedule_work(&adapter->downshift_task);
1154 * 80003ES2LAN workaround-- For packet buffer work-around on
1155 * link down event; disable receives here in the ISR and reset
1156 * adapter in watchdog
1158 if (netif_carrier_ok(netdev) &&
1159 adapter->flags & FLAG_RX_NEEDS_RESTART) {
1160 /* disable receives */
1161 u32 rctl = er32(RCTL);
1162 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1163 adapter->flags |= FLAG_RX_RESTART_NOW;
1165 /* guard against interrupt when we're going down */
1166 if (!test_bit(__E1000_DOWN, &adapter->state))
1167 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1170 if (napi_schedule_prep(&adapter->napi)) {
1171 adapter->total_tx_bytes = 0;
1172 adapter->total_tx_packets = 0;
1173 adapter->total_rx_bytes = 0;
1174 adapter->total_rx_packets = 0;
1175 __napi_schedule(&adapter->napi);
1182 * e1000_intr - Interrupt Handler
1183 * @irq: interrupt number
1184 * @data: pointer to a network interface device structure
1186 static irqreturn_t e1000_intr(int irq, void *data)
1188 struct net_device *netdev = data;
1189 struct e1000_adapter *adapter = netdev_priv(netdev);
1190 struct e1000_hw *hw = &adapter->hw;
1191 u32 rctl, icr = er32(ICR);
1193 if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1194 return IRQ_NONE; /* Not our interrupt */
1197 * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1198 * not set, then the adapter didn't send an interrupt
1200 if (!(icr & E1000_ICR_INT_ASSERTED))
1204 * Interrupt Auto-Mask...upon reading ICR,
1205 * interrupts are masked. No need for the
1209 if (icr & E1000_ICR_LSC) {
1210 hw->mac.get_link_status = 1;
1212 * ICH8 workaround-- Call gig speed drop workaround on cable
1213 * disconnect (LSC) before accessing any PHY registers
1215 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1216 (!(er32(STATUS) & E1000_STATUS_LU)))
1217 schedule_work(&adapter->downshift_task);
1220 * 80003ES2LAN workaround--
1221 * For packet buffer work-around on link down event;
1222 * disable receives here in the ISR and
1223 * reset adapter in watchdog
1225 if (netif_carrier_ok(netdev) &&
1226 (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1227 /* disable receives */
1229 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1230 adapter->flags |= FLAG_RX_RESTART_NOW;
1232 /* guard against interrupt when we're going down */
1233 if (!test_bit(__E1000_DOWN, &adapter->state))
1234 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1237 if (napi_schedule_prep(&adapter->napi)) {
1238 adapter->total_tx_bytes = 0;
1239 adapter->total_tx_packets = 0;
1240 adapter->total_rx_bytes = 0;
1241 adapter->total_rx_packets = 0;
1242 __napi_schedule(&adapter->napi);
1248 static irqreturn_t e1000_msix_other(int irq, void *data)
1250 struct net_device *netdev = data;
1251 struct e1000_adapter *adapter = netdev_priv(netdev);
1252 struct e1000_hw *hw = &adapter->hw;
1253 u32 icr = er32(ICR);
1255 if (!(icr & E1000_ICR_INT_ASSERTED)) {
1256 if (!test_bit(__E1000_DOWN, &adapter->state))
1257 ew32(IMS, E1000_IMS_OTHER);
1261 if (icr & adapter->eiac_mask)
1262 ew32(ICS, (icr & adapter->eiac_mask));
1264 if (icr & E1000_ICR_OTHER) {
1265 if (!(icr & E1000_ICR_LSC))
1266 goto no_link_interrupt;
1267 hw->mac.get_link_status = 1;
1268 /* guard against interrupt when we're going down */
1269 if (!test_bit(__E1000_DOWN, &adapter->state))
1270 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1274 if (!test_bit(__E1000_DOWN, &adapter->state))
1275 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1281 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1283 struct net_device *netdev = data;
1284 struct e1000_adapter *adapter = netdev_priv(netdev);
1285 struct e1000_hw *hw = &adapter->hw;
1286 struct e1000_ring *tx_ring = adapter->tx_ring;
1289 adapter->total_tx_bytes = 0;
1290 adapter->total_tx_packets = 0;
1292 if (!e1000_clean_tx_irq(adapter))
1293 /* Ring was not completely cleaned, so fire another interrupt */
1294 ew32(ICS, tx_ring->ims_val);
1299 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1301 struct net_device *netdev = data;
1302 struct e1000_adapter *adapter = netdev_priv(netdev);
1304 /* Write the ITR value calculated at the end of the
1305 * previous interrupt.
1307 if (adapter->rx_ring->set_itr) {
1308 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1309 adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1310 adapter->rx_ring->set_itr = 0;
1313 if (napi_schedule_prep(&adapter->napi)) {
1314 adapter->total_rx_bytes = 0;
1315 adapter->total_rx_packets = 0;
1316 __napi_schedule(&adapter->napi);
1322 * e1000_configure_msix - Configure MSI-X hardware
1324 * e1000_configure_msix sets up the hardware to properly
1325 * generate MSI-X interrupts.
1327 static void e1000_configure_msix(struct e1000_adapter *adapter)
1329 struct e1000_hw *hw = &adapter->hw;
1330 struct e1000_ring *rx_ring = adapter->rx_ring;
1331 struct e1000_ring *tx_ring = adapter->tx_ring;
1333 u32 ctrl_ext, ivar = 0;
1335 adapter->eiac_mask = 0;
1337 /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1338 if (hw->mac.type == e1000_82574) {
1339 u32 rfctl = er32(RFCTL);
1340 rfctl |= E1000_RFCTL_ACK_DIS;
1344 #define E1000_IVAR_INT_ALLOC_VALID 0x8
1345 /* Configure Rx vector */
1346 rx_ring->ims_val = E1000_IMS_RXQ0;
1347 adapter->eiac_mask |= rx_ring->ims_val;
1348 if (rx_ring->itr_val)
1349 writel(1000000000 / (rx_ring->itr_val * 256),
1350 hw->hw_addr + rx_ring->itr_register);
1352 writel(1, hw->hw_addr + rx_ring->itr_register);
1353 ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1355 /* Configure Tx vector */
1356 tx_ring->ims_val = E1000_IMS_TXQ0;
1358 if (tx_ring->itr_val)
1359 writel(1000000000 / (tx_ring->itr_val * 256),
1360 hw->hw_addr + tx_ring->itr_register);
1362 writel(1, hw->hw_addr + tx_ring->itr_register);
1363 adapter->eiac_mask |= tx_ring->ims_val;
1364 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1366 /* set vector for Other Causes, e.g. link changes */
1368 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1369 if (rx_ring->itr_val)
1370 writel(1000000000 / (rx_ring->itr_val * 256),
1371 hw->hw_addr + E1000_EITR_82574(vector));
1373 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1375 /* Cause Tx interrupts on every write back */
1380 /* enable MSI-X PBA support */
1381 ctrl_ext = er32(CTRL_EXT);
1382 ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1384 /* Auto-Mask Other interrupts upon ICR read */
1385 #define E1000_EIAC_MASK_82574 0x01F00000
1386 ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1387 ctrl_ext |= E1000_CTRL_EXT_EIAME;
1388 ew32(CTRL_EXT, ctrl_ext);
1392 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1394 if (adapter->msix_entries) {
1395 pci_disable_msix(adapter->pdev);
1396 kfree(adapter->msix_entries);
1397 adapter->msix_entries = NULL;
1398 } else if (adapter->flags & FLAG_MSI_ENABLED) {
1399 pci_disable_msi(adapter->pdev);
1400 adapter->flags &= ~FLAG_MSI_ENABLED;
1407 * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1409 * Attempt to configure interrupts using the best available
1410 * capabilities of the hardware and kernel.
1412 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1418 switch (adapter->int_mode) {
1419 case E1000E_INT_MODE_MSIX:
1420 if (adapter->flags & FLAG_HAS_MSIX) {
1421 numvecs = 3; /* RxQ0, TxQ0 and other */
1422 adapter->msix_entries = kcalloc(numvecs,
1423 sizeof(struct msix_entry),
1425 if (adapter->msix_entries) {
1426 for (i = 0; i < numvecs; i++)
1427 adapter->msix_entries[i].entry = i;
1429 err = pci_enable_msix(adapter->pdev,
1430 adapter->msix_entries,
1435 /* MSI-X failed, so fall through and try MSI */
1436 e_err("Failed to initialize MSI-X interrupts. "
1437 "Falling back to MSI interrupts.\n");
1438 e1000e_reset_interrupt_capability(adapter);
1440 adapter->int_mode = E1000E_INT_MODE_MSI;
1442 case E1000E_INT_MODE_MSI:
1443 if (!pci_enable_msi(adapter->pdev)) {
1444 adapter->flags |= FLAG_MSI_ENABLED;
1446 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1447 e_err("Failed to initialize MSI interrupts. Falling "
1448 "back to legacy interrupts.\n");
1451 case E1000E_INT_MODE_LEGACY:
1452 /* Don't do anything; this is the system default */
1460 * e1000_request_msix - Initialize MSI-X interrupts
1462 * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1465 static int e1000_request_msix(struct e1000_adapter *adapter)
1467 struct net_device *netdev = adapter->netdev;
1468 int err = 0, vector = 0;
1470 if (strlen(netdev->name) < (IFNAMSIZ - 5))
1471 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1473 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1474 err = request_irq(adapter->msix_entries[vector].vector,
1475 e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1479 adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1480 adapter->rx_ring->itr_val = adapter->itr;
1483 if (strlen(netdev->name) < (IFNAMSIZ - 5))
1484 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1486 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1487 err = request_irq(adapter->msix_entries[vector].vector,
1488 e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1492 adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1493 adapter->tx_ring->itr_val = adapter->itr;
1496 err = request_irq(adapter->msix_entries[vector].vector,
1497 e1000_msix_other, 0, netdev->name, netdev);
1501 e1000_configure_msix(adapter);
1508 * e1000_request_irq - initialize interrupts
1510 * Attempts to configure interrupts using the best available
1511 * capabilities of the hardware and kernel.
1513 static int e1000_request_irq(struct e1000_adapter *adapter)
1515 struct net_device *netdev = adapter->netdev;
1518 if (adapter->msix_entries) {
1519 err = e1000_request_msix(adapter);
1522 /* fall back to MSI */
1523 e1000e_reset_interrupt_capability(adapter);
1524 adapter->int_mode = E1000E_INT_MODE_MSI;
1525 e1000e_set_interrupt_capability(adapter);
1527 if (adapter->flags & FLAG_MSI_ENABLED) {
1528 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1529 netdev->name, netdev);
1533 /* fall back to legacy interrupt */
1534 e1000e_reset_interrupt_capability(adapter);
1535 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1538 err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1539 netdev->name, netdev);
1541 e_err("Unable to allocate interrupt, Error: %d\n", err);
1546 static void e1000_free_irq(struct e1000_adapter *adapter)
1548 struct net_device *netdev = adapter->netdev;
1550 if (adapter->msix_entries) {
1553 free_irq(adapter->msix_entries[vector].vector, netdev);
1556 free_irq(adapter->msix_entries[vector].vector, netdev);
1559 /* Other Causes interrupt vector */
1560 free_irq(adapter->msix_entries[vector].vector, netdev);
1564 free_irq(adapter->pdev->irq, netdev);
1568 * e1000_irq_disable - Mask off interrupt generation on the NIC
1570 static void e1000_irq_disable(struct e1000_adapter *adapter)
1572 struct e1000_hw *hw = &adapter->hw;
1575 if (adapter->msix_entries)
1576 ew32(EIAC_82574, 0);
1578 synchronize_irq(adapter->pdev->irq);
1582 * e1000_irq_enable - Enable default interrupt generation settings
1584 static void e1000_irq_enable(struct e1000_adapter *adapter)
1586 struct e1000_hw *hw = &adapter->hw;
1588 if (adapter->msix_entries) {
1589 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1590 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1592 ew32(IMS, IMS_ENABLE_MASK);
1598 * e1000_get_hw_control - get control of the h/w from f/w
1599 * @adapter: address of board private structure
1601 * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1602 * For ASF and Pass Through versions of f/w this means that
1603 * the driver is loaded. For AMT version (only with 82573)
1604 * of the f/w this means that the network i/f is open.
1606 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1608 struct e1000_hw *hw = &adapter->hw;
1612 /* Let firmware know the driver has taken over */
1613 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1615 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1616 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1617 ctrl_ext = er32(CTRL_EXT);
1618 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1623 * e1000_release_hw_control - release control of the h/w to f/w
1624 * @adapter: address of board private structure
1626 * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1627 * For ASF and Pass Through versions of f/w this means that the
1628 * driver is no longer loaded. For AMT version (only with 82573) i
1629 * of the f/w this means that the network i/f is closed.
1632 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1634 struct e1000_hw *hw = &adapter->hw;
1638 /* Let firmware taken over control of h/w */
1639 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1641 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1642 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1643 ctrl_ext = er32(CTRL_EXT);
1644 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1649 * @e1000_alloc_ring - allocate memory for a ring structure
1651 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1652 struct e1000_ring *ring)
1654 struct pci_dev *pdev = adapter->pdev;
1656 ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1665 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1666 * @adapter: board private structure
1668 * Return 0 on success, negative on failure
1670 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1672 struct e1000_ring *tx_ring = adapter->tx_ring;
1673 int err = -ENOMEM, size;
1675 size = sizeof(struct e1000_buffer) * tx_ring->count;
1676 tx_ring->buffer_info = vmalloc(size);
1677 if (!tx_ring->buffer_info)
1679 memset(tx_ring->buffer_info, 0, size);
1681 /* round up to nearest 4K */
1682 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1683 tx_ring->size = ALIGN(tx_ring->size, 4096);
1685 err = e1000_alloc_ring_dma(adapter, tx_ring);
1689 tx_ring->next_to_use = 0;
1690 tx_ring->next_to_clean = 0;
1694 vfree(tx_ring->buffer_info);
1695 e_err("Unable to allocate memory for the transmit descriptor ring\n");
1700 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1701 * @adapter: board private structure
1703 * Returns 0 on success, negative on failure
1705 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1707 struct e1000_ring *rx_ring = adapter->rx_ring;
1708 struct e1000_buffer *buffer_info;
1709 int i, size, desc_len, err = -ENOMEM;
1711 size = sizeof(struct e1000_buffer) * rx_ring->count;
1712 rx_ring->buffer_info = vmalloc(size);
1713 if (!rx_ring->buffer_info)
1715 memset(rx_ring->buffer_info, 0, size);
1717 for (i = 0; i < rx_ring->count; i++) {
1718 buffer_info = &rx_ring->buffer_info[i];
1719 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1720 sizeof(struct e1000_ps_page),
1722 if (!buffer_info->ps_pages)
1726 desc_len = sizeof(union e1000_rx_desc_packet_split);
1728 /* Round up to nearest 4K */
1729 rx_ring->size = rx_ring->count * desc_len;
1730 rx_ring->size = ALIGN(rx_ring->size, 4096);
1732 err = e1000_alloc_ring_dma(adapter, rx_ring);
1736 rx_ring->next_to_clean = 0;
1737 rx_ring->next_to_use = 0;
1738 rx_ring->rx_skb_top = NULL;
1743 for (i = 0; i < rx_ring->count; i++) {
1744 buffer_info = &rx_ring->buffer_info[i];
1745 kfree(buffer_info->ps_pages);
1748 vfree(rx_ring->buffer_info);
1749 e_err("Unable to allocate memory for the transmit descriptor ring\n");
1754 * e1000_clean_tx_ring - Free Tx Buffers
1755 * @adapter: board private structure
1757 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1759 struct e1000_ring *tx_ring = adapter->tx_ring;
1760 struct e1000_buffer *buffer_info;
1764 for (i = 0; i < tx_ring->count; i++) {
1765 buffer_info = &tx_ring->buffer_info[i];
1766 e1000_put_txbuf(adapter, buffer_info);
1769 size = sizeof(struct e1000_buffer) * tx_ring->count;
1770 memset(tx_ring->buffer_info, 0, size);
1772 memset(tx_ring->desc, 0, tx_ring->size);
1774 tx_ring->next_to_use = 0;
1775 tx_ring->next_to_clean = 0;
1777 writel(0, adapter->hw.hw_addr + tx_ring->head);
1778 writel(0, adapter->hw.hw_addr + tx_ring->tail);
1782 * e1000e_free_tx_resources - Free Tx Resources per Queue
1783 * @adapter: board private structure
1785 * Free all transmit software resources
1787 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1789 struct pci_dev *pdev = adapter->pdev;
1790 struct e1000_ring *tx_ring = adapter->tx_ring;
1792 e1000_clean_tx_ring(adapter);
1794 vfree(tx_ring->buffer_info);
1795 tx_ring->buffer_info = NULL;
1797 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1799 tx_ring->desc = NULL;
1803 * e1000e_free_rx_resources - Free Rx Resources
1804 * @adapter: board private structure
1806 * Free all receive software resources
1809 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1811 struct pci_dev *pdev = adapter->pdev;
1812 struct e1000_ring *rx_ring = adapter->rx_ring;
1815 e1000_clean_rx_ring(adapter);
1817 for (i = 0; i < rx_ring->count; i++) {
1818 kfree(rx_ring->buffer_info[i].ps_pages);
1821 vfree(rx_ring->buffer_info);
1822 rx_ring->buffer_info = NULL;
1824 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1826 rx_ring->desc = NULL;
1830 * e1000_update_itr - update the dynamic ITR value based on statistics
1831 * @adapter: pointer to adapter
1832 * @itr_setting: current adapter->itr
1833 * @packets: the number of packets during this measurement interval
1834 * @bytes: the number of bytes during this measurement interval
1836 * Stores a new ITR value based on packets and byte
1837 * counts during the last interrupt. The advantage of per interrupt
1838 * computation is faster updates and more accurate ITR for the current
1839 * traffic pattern. Constants in this function were computed
1840 * based on theoretical maximum wire speed and thresholds were set based
1841 * on testing data as well as attempting to minimize response time
1842 * while increasing bulk throughput. This functionality is controlled
1843 * by the InterruptThrottleRate module parameter.
1845 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1846 u16 itr_setting, int packets,
1849 unsigned int retval = itr_setting;
1852 goto update_itr_done;
1854 switch (itr_setting) {
1855 case lowest_latency:
1856 /* handle TSO and jumbo frames */
1857 if (bytes/packets > 8000)
1858 retval = bulk_latency;
1859 else if ((packets < 5) && (bytes > 512)) {
1860 retval = low_latency;
1863 case low_latency: /* 50 usec aka 20000 ints/s */
1864 if (bytes > 10000) {
1865 /* this if handles the TSO accounting */
1866 if (bytes/packets > 8000) {
1867 retval = bulk_latency;
1868 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1869 retval = bulk_latency;
1870 } else if ((packets > 35)) {
1871 retval = lowest_latency;
1873 } else if (bytes/packets > 2000) {
1874 retval = bulk_latency;
1875 } else if (packets <= 2 && bytes < 512) {
1876 retval = lowest_latency;
1879 case bulk_latency: /* 250 usec aka 4000 ints/s */
1880 if (bytes > 25000) {
1882 retval = low_latency;
1884 } else if (bytes < 6000) {
1885 retval = low_latency;
1894 static void e1000_set_itr(struct e1000_adapter *adapter)
1896 struct e1000_hw *hw = &adapter->hw;
1898 u32 new_itr = adapter->itr;
1900 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1901 if (adapter->link_speed != SPEED_1000) {
1907 adapter->tx_itr = e1000_update_itr(adapter,
1909 adapter->total_tx_packets,
1910 adapter->total_tx_bytes);
1911 /* conservative mode (itr 3) eliminates the lowest_latency setting */
1912 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1913 adapter->tx_itr = low_latency;
1915 adapter->rx_itr = e1000_update_itr(adapter,
1917 adapter->total_rx_packets,
1918 adapter->total_rx_bytes);
1919 /* conservative mode (itr 3) eliminates the lowest_latency setting */
1920 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1921 adapter->rx_itr = low_latency;
1923 current_itr = max(adapter->rx_itr, adapter->tx_itr);
1925 switch (current_itr) {
1926 /* counts and packets in update_itr are dependent on these numbers */
1927 case lowest_latency:
1931 new_itr = 20000; /* aka hwitr = ~200 */
1941 if (new_itr != adapter->itr) {
1943 * this attempts to bias the interrupt rate towards Bulk
1944 * by adding intermediate steps when interrupt rate is
1947 new_itr = new_itr > adapter->itr ?
1948 min(adapter->itr + (new_itr >> 2), new_itr) :
1950 adapter->itr = new_itr;
1951 adapter->rx_ring->itr_val = new_itr;
1952 if (adapter->msix_entries)
1953 adapter->rx_ring->set_itr = 1;
1955 ew32(ITR, 1000000000 / (new_itr * 256));
1960 * e1000_alloc_queues - Allocate memory for all rings
1961 * @adapter: board private structure to initialize
1963 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1965 adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1966 if (!adapter->tx_ring)
1969 adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1970 if (!adapter->rx_ring)
1975 e_err("Unable to allocate memory for queues\n");
1976 kfree(adapter->rx_ring);
1977 kfree(adapter->tx_ring);
1982 * e1000_clean - NAPI Rx polling callback
1983 * @napi: struct associated with this polling callback
1984 * @budget: amount of packets driver is allowed to process this poll
1986 static int e1000_clean(struct napi_struct *napi, int budget)
1988 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1989 struct e1000_hw *hw = &adapter->hw;
1990 struct net_device *poll_dev = adapter->netdev;
1991 int tx_cleaned = 1, work_done = 0;
1993 adapter = netdev_priv(poll_dev);
1995 if (adapter->msix_entries &&
1996 !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
1999 tx_cleaned = e1000_clean_tx_irq(adapter);
2002 adapter->clean_rx(adapter, &work_done, budget);
2007 /* If budget not fully consumed, exit the polling mode */
2008 if (work_done < budget) {
2009 if (adapter->itr_setting & 3)
2010 e1000_set_itr(adapter);
2011 napi_complete(napi);
2012 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2013 if (adapter->msix_entries)
2014 ew32(IMS, adapter->rx_ring->ims_val);
2016 e1000_irq_enable(adapter);
2023 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2025 struct e1000_adapter *adapter = netdev_priv(netdev);
2026 struct e1000_hw *hw = &adapter->hw;
2029 /* don't update vlan cookie if already programmed */
2030 if ((adapter->hw.mng_cookie.status &
2031 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2032 (vid == adapter->mng_vlan_id))
2034 /* add VID to filter table */
2035 index = (vid >> 5) & 0x7F;
2036 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2037 vfta |= (1 << (vid & 0x1F));
2038 e1000e_write_vfta(hw, index, vfta);
2041 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2043 struct e1000_adapter *adapter = netdev_priv(netdev);
2044 struct e1000_hw *hw = &adapter->hw;
2047 if (!test_bit(__E1000_DOWN, &adapter->state))
2048 e1000_irq_disable(adapter);
2049 vlan_group_set_device(adapter->vlgrp, vid, NULL);
2051 if (!test_bit(__E1000_DOWN, &adapter->state))
2052 e1000_irq_enable(adapter);
2054 if ((adapter->hw.mng_cookie.status &
2055 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2056 (vid == adapter->mng_vlan_id)) {
2057 /* release control to f/w */
2058 e1000_release_hw_control(adapter);
2062 /* remove VID from filter table */
2063 index = (vid >> 5) & 0x7F;
2064 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2065 vfta &= ~(1 << (vid & 0x1F));
2066 e1000e_write_vfta(hw, index, vfta);
2069 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2071 struct net_device *netdev = adapter->netdev;
2072 u16 vid = adapter->hw.mng_cookie.vlan_id;
2073 u16 old_vid = adapter->mng_vlan_id;
2075 if (!adapter->vlgrp)
2078 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2079 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2080 if (adapter->hw.mng_cookie.status &
2081 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2082 e1000_vlan_rx_add_vid(netdev, vid);
2083 adapter->mng_vlan_id = vid;
2086 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2088 !vlan_group_get_device(adapter->vlgrp, old_vid))
2089 e1000_vlan_rx_kill_vid(netdev, old_vid);
2091 adapter->mng_vlan_id = vid;
2096 static void e1000_vlan_rx_register(struct net_device *netdev,
2097 struct vlan_group *grp)
2099 struct e1000_adapter *adapter = netdev_priv(netdev);
2100 struct e1000_hw *hw = &adapter->hw;
2103 if (!test_bit(__E1000_DOWN, &adapter->state))
2104 e1000_irq_disable(adapter);
2105 adapter->vlgrp = grp;
2108 /* enable VLAN tag insert/strip */
2110 ctrl |= E1000_CTRL_VME;
2113 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2114 /* enable VLAN receive filtering */
2116 rctl &= ~E1000_RCTL_CFIEN;
2118 e1000_update_mng_vlan(adapter);
2121 /* disable VLAN tag insert/strip */
2123 ctrl &= ~E1000_CTRL_VME;
2126 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2127 if (adapter->mng_vlan_id !=
2128 (u16)E1000_MNG_VLAN_NONE) {
2129 e1000_vlan_rx_kill_vid(netdev,
2130 adapter->mng_vlan_id);
2131 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2136 if (!test_bit(__E1000_DOWN, &adapter->state))
2137 e1000_irq_enable(adapter);
2140 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2144 e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2146 if (!adapter->vlgrp)
2149 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2150 if (!vlan_group_get_device(adapter->vlgrp, vid))
2152 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2156 static void e1000_init_manageability(struct e1000_adapter *adapter)
2158 struct e1000_hw *hw = &adapter->hw;
2161 if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2167 * enable receiving management packets to the host. this will probably
2168 * generate destination unreachable messages from the host OS, but
2169 * the packets will be handled on SMBUS
2171 manc |= E1000_MANC_EN_MNG2HOST;
2172 manc2h = er32(MANC2H);
2173 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2174 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2175 manc2h |= E1000_MNG2HOST_PORT_623;
2176 manc2h |= E1000_MNG2HOST_PORT_664;
2177 ew32(MANC2H, manc2h);
2182 * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2183 * @adapter: board private structure
2185 * Configure the Tx unit of the MAC after a reset.
2187 static void e1000_configure_tx(struct e1000_adapter *adapter)
2189 struct e1000_hw *hw = &adapter->hw;
2190 struct e1000_ring *tx_ring = adapter->tx_ring;
2192 u32 tdlen, tctl, tipg, tarc;
2195 /* Setup the HW Tx Head and Tail descriptor pointers */
2196 tdba = tx_ring->dma;
2197 tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2198 ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2199 ew32(TDBAH, (tdba >> 32));
2203 tx_ring->head = E1000_TDH;
2204 tx_ring->tail = E1000_TDT;
2206 /* Set the default values for the Tx Inter Packet Gap timer */
2207 tipg = DEFAULT_82543_TIPG_IPGT_COPPER; /* 8 */
2208 ipgr1 = DEFAULT_82543_TIPG_IPGR1; /* 8 */
2209 ipgr2 = DEFAULT_82543_TIPG_IPGR2; /* 6 */
2211 if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2212 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /* 7 */
2214 tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2215 tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2218 /* Set the Tx Interrupt Delay register */
2219 ew32(TIDV, adapter->tx_int_delay);
2220 /* Tx irq moderation */
2221 ew32(TADV, adapter->tx_abs_int_delay);
2223 /* Program the Transmit Control Register */
2225 tctl &= ~E1000_TCTL_CT;
2226 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2227 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2229 if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2230 tarc = er32(TARC(0));
2232 * set the speed mode bit, we'll clear it if we're not at
2233 * gigabit link later
2235 #define SPEED_MODE_BIT (1 << 21)
2236 tarc |= SPEED_MODE_BIT;
2237 ew32(TARC(0), tarc);
2240 /* errata: program both queues to unweighted RR */
2241 if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2242 tarc = er32(TARC(0));
2244 ew32(TARC(0), tarc);
2245 tarc = er32(TARC(1));
2247 ew32(TARC(1), tarc);
2250 /* Setup Transmit Descriptor Settings for eop descriptor */
2251 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2253 /* only set IDE if we are delaying interrupts using the timers */
2254 if (adapter->tx_int_delay)
2255 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2257 /* enable Report Status bit */
2258 adapter->txd_cmd |= E1000_TXD_CMD_RS;
2262 e1000e_config_collision_dist(hw);
2264 adapter->tx_queue_len = adapter->netdev->tx_queue_len;
2268 * e1000_setup_rctl - configure the receive control registers
2269 * @adapter: Board private structure
2271 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2272 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2273 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2275 struct e1000_hw *hw = &adapter->hw;
2280 /* Program MC offset vector base */
2282 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2283 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2284 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2285 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2287 /* Do not Store bad packets */
2288 rctl &= ~E1000_RCTL_SBP;
2290 /* Enable Long Packet receive */
2291 if (adapter->netdev->mtu <= ETH_DATA_LEN)
2292 rctl &= ~E1000_RCTL_LPE;
2294 rctl |= E1000_RCTL_LPE;
2296 /* Some systems expect that the CRC is included in SMBUS traffic. The
2297 * hardware strips the CRC before sending to both SMBUS (BMC) and to
2298 * host memory when this is enabled
2300 if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2301 rctl |= E1000_RCTL_SECRC;
2303 /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2304 if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2307 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2309 phy_data |= (1 << 2);
2310 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2312 e1e_rphy(hw, 22, &phy_data);
2314 phy_data |= (1 << 14);
2315 e1e_wphy(hw, 0x10, 0x2823);
2316 e1e_wphy(hw, 0x11, 0x0003);
2317 e1e_wphy(hw, 22, phy_data);
2320 /* Setup buffer sizes */
2321 rctl &= ~E1000_RCTL_SZ_4096;
2322 rctl |= E1000_RCTL_BSEX;
2323 switch (adapter->rx_buffer_len) {
2325 rctl |= E1000_RCTL_SZ_256;
2326 rctl &= ~E1000_RCTL_BSEX;
2329 rctl |= E1000_RCTL_SZ_512;
2330 rctl &= ~E1000_RCTL_BSEX;
2333 rctl |= E1000_RCTL_SZ_1024;
2334 rctl &= ~E1000_RCTL_BSEX;
2338 rctl |= E1000_RCTL_SZ_2048;
2339 rctl &= ~E1000_RCTL_BSEX;
2342 rctl |= E1000_RCTL_SZ_4096;
2345 rctl |= E1000_RCTL_SZ_8192;
2348 rctl |= E1000_RCTL_SZ_16384;
2353 * 82571 and greater support packet-split where the protocol
2354 * header is placed in skb->data and the packet data is
2355 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2356 * In the case of a non-split, skb->data is linearly filled,
2357 * followed by the page buffers. Therefore, skb->data is
2358 * sized to hold the largest protocol header.
2360 * allocations using alloc_page take too long for regular MTU
2361 * so only enable packet split for jumbo frames
2363 * Using pages when the page size is greater than 16k wastes
2364 * a lot of memory, since we allocate 3 pages at all times
2367 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2368 if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2369 (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2370 adapter->rx_ps_pages = pages;
2372 adapter->rx_ps_pages = 0;
2374 if (adapter->rx_ps_pages) {
2375 /* Configure extra packet-split registers */
2376 rfctl = er32(RFCTL);
2377 rfctl |= E1000_RFCTL_EXTEN;
2379 * disable packet split support for IPv6 extension headers,
2380 * because some malformed IPv6 headers can hang the Rx
2382 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2383 E1000_RFCTL_NEW_IPV6_EXT_DIS);
2387 /* Enable Packet split descriptors */
2388 rctl |= E1000_RCTL_DTYP_PS;
2390 psrctl |= adapter->rx_ps_bsize0 >>
2391 E1000_PSRCTL_BSIZE0_SHIFT;
2393 switch (adapter->rx_ps_pages) {
2395 psrctl |= PAGE_SIZE <<
2396 E1000_PSRCTL_BSIZE3_SHIFT;
2398 psrctl |= PAGE_SIZE <<
2399 E1000_PSRCTL_BSIZE2_SHIFT;
2401 psrctl |= PAGE_SIZE >>
2402 E1000_PSRCTL_BSIZE1_SHIFT;
2406 ew32(PSRCTL, psrctl);
2410 /* just started the receive unit, no need to restart */
2411 adapter->flags &= ~FLAG_RX_RESTART_NOW;
2415 * e1000_configure_rx - Configure Receive Unit after Reset
2416 * @adapter: board private structure
2418 * Configure the Rx unit of the MAC after a reset.
2420 static void e1000_configure_rx(struct e1000_adapter *adapter)
2422 struct e1000_hw *hw = &adapter->hw;
2423 struct e1000_ring *rx_ring = adapter->rx_ring;
2425 u32 rdlen, rctl, rxcsum, ctrl_ext;
2427 if (adapter->rx_ps_pages) {
2428 /* this is a 32 byte descriptor */
2429 rdlen = rx_ring->count *
2430 sizeof(union e1000_rx_desc_packet_split);
2431 adapter->clean_rx = e1000_clean_rx_irq_ps;
2432 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2433 } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2434 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2435 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2436 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2438 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2439 adapter->clean_rx = e1000_clean_rx_irq;
2440 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2443 /* disable receives while setting up the descriptors */
2445 ew32(RCTL, rctl & ~E1000_RCTL_EN);
2449 /* set the Receive Delay Timer Register */
2450 ew32(RDTR, adapter->rx_int_delay);
2452 /* irq moderation */
2453 ew32(RADV, adapter->rx_abs_int_delay);
2454 if (adapter->itr_setting != 0)
2455 ew32(ITR, 1000000000 / (adapter->itr * 256));
2457 ctrl_ext = er32(CTRL_EXT);
2458 /* Auto-Mask interrupts upon ICR access */
2459 ctrl_ext |= E1000_CTRL_EXT_IAME;
2460 ew32(IAM, 0xffffffff);
2461 ew32(CTRL_EXT, ctrl_ext);
2465 * Setup the HW Rx Head and Tail Descriptor Pointers and
2466 * the Base and Length of the Rx Descriptor Ring
2468 rdba = rx_ring->dma;
2469 ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2470 ew32(RDBAH, (rdba >> 32));
2474 rx_ring->head = E1000_RDH;
2475 rx_ring->tail = E1000_RDT;
2477 /* Enable Receive Checksum Offload for TCP and UDP */
2478 rxcsum = er32(RXCSUM);
2479 if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2480 rxcsum |= E1000_RXCSUM_TUOFL;
2483 * IPv4 payload checksum for UDP fragments must be
2484 * used in conjunction with packet-split.
2486 if (adapter->rx_ps_pages)
2487 rxcsum |= E1000_RXCSUM_IPPCSE;
2489 rxcsum &= ~E1000_RXCSUM_TUOFL;
2490 /* no need to clear IPPCSE as it defaults to 0 */
2492 ew32(RXCSUM, rxcsum);
2495 * Enable early receives on supported devices, only takes effect when
2496 * packet size is equal or larger than the specified value (in 8 byte
2497 * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2499 if (adapter->flags & FLAG_HAS_ERT) {
2500 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2501 u32 rxdctl = er32(RXDCTL(0));
2502 ew32(RXDCTL(0), rxdctl | 0x3);
2503 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2505 * With jumbo frames and early-receive enabled,
2506 * excessive C-state transition latencies result in
2507 * dropped transactions.
2509 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2510 adapter->netdev->name, 55);
2512 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2513 adapter->netdev->name,
2514 PM_QOS_DEFAULT_VALUE);
2518 /* Enable Receives */
2523 * e1000_update_mc_addr_list - Update Multicast addresses
2524 * @hw: pointer to the HW structure
2525 * @mc_addr_list: array of multicast addresses to program
2526 * @mc_addr_count: number of multicast addresses to program
2527 * @rar_used_count: the first RAR register free to program
2528 * @rar_count: total number of supported Receive Address Registers
2530 * Updates the Receive Address Registers and Multicast Table Array.
2531 * The caller must have a packed mc_addr_list of multicast addresses.
2532 * The parameter rar_count will usually be hw->mac.rar_entry_count
2533 * unless there are workarounds that change this. Currently no func pointer
2534 * exists and all implementations are handled in the generic version of this
2537 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2538 u32 mc_addr_count, u32 rar_used_count,
2541 hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2542 rar_used_count, rar_count);
2546 * e1000_set_multi - Multicast and Promiscuous mode set
2547 * @netdev: network interface device structure
2549 * The set_multi entry point is called whenever the multicast address
2550 * list or the network interface flags are updated. This routine is
2551 * responsible for configuring the hardware for proper multicast,
2552 * promiscuous mode, and all-multi behavior.
2554 static void e1000_set_multi(struct net_device *netdev)
2556 struct e1000_adapter *adapter = netdev_priv(netdev);
2557 struct e1000_hw *hw = &adapter->hw;
2558 struct e1000_mac_info *mac = &hw->mac;
2559 struct dev_mc_list *mc_ptr;
2564 /* Check for Promiscuous and All Multicast modes */
2568 if (netdev->flags & IFF_PROMISC) {
2569 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2570 rctl &= ~E1000_RCTL_VFE;
2572 if (netdev->flags & IFF_ALLMULTI) {
2573 rctl |= E1000_RCTL_MPE;
2574 rctl &= ~E1000_RCTL_UPE;
2576 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2578 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2579 rctl |= E1000_RCTL_VFE;
2584 if (netdev->mc_count) {
2585 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2589 /* prepare a packed array of only addresses. */
2590 mc_ptr = netdev->mc_list;
2592 for (i = 0; i < netdev->mc_count; i++) {
2595 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2597 mc_ptr = mc_ptr->next;
2600 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2601 mac->rar_entry_count);
2605 * if we're called from probe, we might not have
2606 * anything to do here, so clear out the list
2608 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2613 * e1000_configure - configure the hardware for Rx and Tx
2614 * @adapter: private board structure
2616 static void e1000_configure(struct e1000_adapter *adapter)
2618 e1000_set_multi(adapter->netdev);
2620 e1000_restore_vlan(adapter);
2621 e1000_init_manageability(adapter);
2623 e1000_configure_tx(adapter);
2624 e1000_setup_rctl(adapter);
2625 e1000_configure_rx(adapter);
2626 adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2630 * e1000e_power_up_phy - restore link in case the phy was powered down
2631 * @adapter: address of board private structure
2633 * The phy may be powered down to save power and turn off link when the
2634 * driver is unloaded and wake on lan is not enabled (among others)
2635 * *** this routine MUST be followed by a call to e1000e_reset ***
2637 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2641 /* Just clear the power down bit to wake the phy back up */
2642 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2644 * According to the manual, the phy will retain its
2645 * settings across a power-down/up cycle
2647 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2648 mii_reg &= ~MII_CR_POWER_DOWN;
2649 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2652 adapter->hw.mac.ops.setup_link(&adapter->hw);
2656 * e1000_power_down_phy - Power down the PHY
2658 * Power down the PHY so no link is implied when interface is down
2659 * The PHY cannot be powered down is management or WoL is active
2661 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2663 struct e1000_hw *hw = &adapter->hw;
2666 /* WoL is enabled */
2670 /* non-copper PHY? */
2671 if (adapter->hw.phy.media_type != e1000_media_type_copper)
2674 /* reset is blocked because of a SoL/IDER session */
2675 if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2678 /* manageability (AMT) is enabled */
2679 if (er32(MANC) & E1000_MANC_SMBUS_EN)
2682 /* power down the PHY */
2683 e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2684 mii_reg |= MII_CR_POWER_DOWN;
2685 e1e_wphy(hw, PHY_CONTROL, mii_reg);
2690 * e1000e_reset - bring the hardware into a known good state
2692 * This function boots the hardware and enables some settings that
2693 * require a configuration cycle of the hardware - those cannot be
2694 * set/changed during runtime. After reset the device needs to be
2695 * properly configured for Rx, Tx etc.
2697 void e1000e_reset(struct e1000_adapter *adapter)
2699 struct e1000_mac_info *mac = &adapter->hw.mac;
2700 struct e1000_fc_info *fc = &adapter->hw.fc;
2701 struct e1000_hw *hw = &adapter->hw;
2702 u32 tx_space, min_tx_space, min_rx_space;
2703 u32 pba = adapter->pba;
2706 /* reset Packet Buffer Allocation to default */
2709 if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2711 * To maintain wire speed transmits, the Tx FIFO should be
2712 * large enough to accommodate two full transmit packets,
2713 * rounded up to the next 1KB and expressed in KB. Likewise,
2714 * the Rx FIFO should be large enough to accommodate at least
2715 * one full receive packet and is similarly rounded up and
2719 /* upper 16 bits has Tx packet buffer allocation size in KB */
2720 tx_space = pba >> 16;
2721 /* lower 16 bits has Rx packet buffer allocation size in KB */
2724 * the Tx fifo also stores 16 bytes of information about the tx
2725 * but don't include ethernet FCS because hardware appends it
2727 min_tx_space = (adapter->max_frame_size +
2728 sizeof(struct e1000_tx_desc) -
2730 min_tx_space = ALIGN(min_tx_space, 1024);
2731 min_tx_space >>= 10;
2732 /* software strips receive CRC, so leave room for it */
2733 min_rx_space = adapter->max_frame_size;
2734 min_rx_space = ALIGN(min_rx_space, 1024);
2735 min_rx_space >>= 10;
2738 * If current Tx allocation is less than the min Tx FIFO size,
2739 * and the min Tx FIFO size is less than the current Rx FIFO
2740 * allocation, take space away from current Rx allocation
2742 if ((tx_space < min_tx_space) &&
2743 ((min_tx_space - tx_space) < pba)) {
2744 pba -= min_tx_space - tx_space;
2747 * if short on Rx space, Rx wins and must trump tx
2748 * adjustment or use Early Receive if available
2750 if ((pba < min_rx_space) &&
2751 (!(adapter->flags & FLAG_HAS_ERT)))
2752 /* ERT enabled in e1000_configure_rx */
2761 * flow control settings
2763 * The high water mark must be low enough to fit two full frame
2764 * (or the size used for early receive) above it in the Rx FIFO.
2765 * Set it to the lower of:
2766 * - 90% of the Rx FIFO size, and
2767 * - the full Rx FIFO size minus the early receive size (for parts
2768 * with ERT support assuming ERT set to E1000_ERT_2048), or
2769 * - the full Rx FIFO size minus two full frames
2771 if ((adapter->flags & FLAG_HAS_ERT) &&
2772 (adapter->netdev->mtu > ETH_DATA_LEN))
2773 hwm = min(((pba << 10) * 9 / 10),
2774 ((pba << 10) - (E1000_ERT_2048 << 3)));
2776 hwm = min(((pba << 10) * 9 / 10),
2777 ((pba << 10) - (2 * adapter->max_frame_size)));
2779 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
2780 fc->low_water = (fc->high_water - (2 * adapter->max_frame_size));
2781 fc->low_water &= E1000_FCRTL_RTL; /* 8-byte granularity */
2783 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2784 fc->pause_time = 0xFFFF;
2786 fc->pause_time = E1000_FC_PAUSE_TIME;
2788 fc->current_mode = fc->requested_mode;
2790 /* Allow time for pending master requests to run */
2791 mac->ops.reset_hw(hw);
2794 * For parts with AMT enabled, let the firmware know
2795 * that the network interface is in control
2797 if (adapter->flags & FLAG_HAS_AMT)
2798 e1000_get_hw_control(adapter);
2801 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP)
2802 e1e_wphy(&adapter->hw, BM_WUC, 0);
2804 if (mac->ops.init_hw(hw))
2805 e_err("Hardware Error\n");
2807 e1000_update_mng_vlan(adapter);
2809 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2810 ew32(VET, ETH_P_8021Q);
2812 e1000e_reset_adaptive(hw);
2813 e1000_get_phy_info(hw);
2815 if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
2816 !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2819 * speed up time to link by disabling smart power down, ignore
2820 * the return value of this function because there is nothing
2821 * different we would do if it failed
2823 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2824 phy_data &= ~IGP02E1000_PM_SPD;
2825 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2829 int e1000e_up(struct e1000_adapter *adapter)
2831 struct e1000_hw *hw = &adapter->hw;
2833 /* DMA latency requirement to workaround early-receive/jumbo issue */
2834 if (adapter->flags & FLAG_HAS_ERT)
2835 pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY,
2836 adapter->netdev->name,
2837 PM_QOS_DEFAULT_VALUE);
2839 /* hardware has been reset, we need to reload some things */
2840 e1000_configure(adapter);
2842 clear_bit(__E1000_DOWN, &adapter->state);
2844 napi_enable(&adapter->napi);
2845 if (adapter->msix_entries)
2846 e1000_configure_msix(adapter);
2847 e1000_irq_enable(adapter);
2849 netif_wake_queue(adapter->netdev);
2851 /* fire a link change interrupt to start the watchdog */
2852 ew32(ICS, E1000_ICS_LSC);
2856 void e1000e_down(struct e1000_adapter *adapter)
2858 struct net_device *netdev = adapter->netdev;
2859 struct e1000_hw *hw = &adapter->hw;
2863 * signal that we're down so the interrupt handler does not
2864 * reschedule our watchdog timer
2866 set_bit(__E1000_DOWN, &adapter->state);
2868 /* disable receives in the hardware */
2870 ew32(RCTL, rctl & ~E1000_RCTL_EN);
2871 /* flush and sleep below */
2873 netif_stop_queue(netdev);
2875 /* disable transmits in the hardware */
2877 tctl &= ~E1000_TCTL_EN;
2879 /* flush both disables and wait for them to finish */
2883 napi_disable(&adapter->napi);
2884 e1000_irq_disable(adapter);
2886 del_timer_sync(&adapter->watchdog_timer);
2887 del_timer_sync(&adapter->phy_info_timer);
2889 netdev->tx_queue_len = adapter->tx_queue_len;
2890 netif_carrier_off(netdev);
2891 adapter->link_speed = 0;
2892 adapter->link_duplex = 0;
2894 if (!pci_channel_offline(adapter->pdev))
2895 e1000e_reset(adapter);
2896 e1000_clean_tx_ring(adapter);
2897 e1000_clean_rx_ring(adapter);
2899 if (adapter->flags & FLAG_HAS_ERT)
2900 pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY,
2901 adapter->netdev->name);
2904 * TODO: for power management, we could drop the link and
2905 * pci_disable_device here.
2909 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2912 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2914 e1000e_down(adapter);
2916 clear_bit(__E1000_RESETTING, &adapter->state);
2920 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2921 * @adapter: board private structure to initialize
2923 * e1000_sw_init initializes the Adapter private data structure.
2924 * Fields are initialized based on PCI device information and
2925 * OS network device settings (MTU size).
2927 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2929 struct net_device *netdev = adapter->netdev;
2931 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2932 adapter->rx_ps_bsize0 = 128;
2933 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2934 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2936 e1000e_set_interrupt_capability(adapter);
2938 if (e1000_alloc_queues(adapter))
2941 /* Explicitly disable IRQ since the NIC can be in any state. */
2942 e1000_irq_disable(adapter);
2944 set_bit(__E1000_DOWN, &adapter->state);
2949 * e1000_intr_msi_test - Interrupt Handler
2950 * @irq: interrupt number
2951 * @data: pointer to a network interface device structure
2953 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2955 struct net_device *netdev = data;
2956 struct e1000_adapter *adapter = netdev_priv(netdev);
2957 struct e1000_hw *hw = &adapter->hw;
2958 u32 icr = er32(ICR);
2960 e_dbg("icr is %08X\n", icr);
2961 if (icr & E1000_ICR_RXSEQ) {
2962 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2970 * e1000_test_msi_interrupt - Returns 0 for successful test
2971 * @adapter: board private struct
2973 * code flow taken from tg3.c
2975 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2977 struct net_device *netdev = adapter->netdev;
2978 struct e1000_hw *hw = &adapter->hw;
2981 /* poll_enable hasn't been called yet, so don't need disable */
2982 /* clear any pending events */
2985 /* free the real vector and request a test handler */
2986 e1000_free_irq(adapter);
2987 e1000e_reset_interrupt_capability(adapter);
2989 /* Assume that the test fails, if it succeeds then the test
2990 * MSI irq handler will unset this flag */
2991 adapter->flags |= FLAG_MSI_TEST_FAILED;
2993 err = pci_enable_msi(adapter->pdev);
2995 goto msi_test_failed;
2997 err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
2998 netdev->name, netdev);
3000 pci_disable_msi(adapter->pdev);
3001 goto msi_test_failed;
3006 e1000_irq_enable(adapter);
3008 /* fire an unusual interrupt on the test handler */
3009 ew32(ICS, E1000_ICS_RXSEQ);
3013 e1000_irq_disable(adapter);
3017 if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3018 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3020 e_info("MSI interrupt test failed!\n");
3023 free_irq(adapter->pdev->irq, netdev);
3024 pci_disable_msi(adapter->pdev);
3027 goto msi_test_failed;
3029 /* okay so the test worked, restore settings */
3030 e_dbg("MSI interrupt test succeeded!\n");
3032 e1000e_set_interrupt_capability(adapter);
3033 e1000_request_irq(adapter);
3038 * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3039 * @adapter: board private struct
3041 * code flow taken from tg3.c, called with e1000 interrupts disabled.
3043 static int e1000_test_msi(struct e1000_adapter *adapter)
3048 if (!(adapter->flags & FLAG_MSI_ENABLED))
3051 /* disable SERR in case the MSI write causes a master abort */
3052 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3053 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3054 pci_cmd & ~PCI_COMMAND_SERR);
3056 err = e1000_test_msi_interrupt(adapter);
3058 /* restore previous setting of command word */
3059 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3065 /* EIO means MSI test failed */
3069 /* back to INTx mode */
3070 e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3072 e1000_free_irq(adapter);
3074 err = e1000_request_irq(adapter);
3080 * e1000_open - Called when a network interface is made active
3081 * @netdev: network interface device structure
3083 * Returns 0 on success, negative value on failure
3085 * The open entry point is called when a network interface is made
3086 * active by the system (IFF_UP). At this point all resources needed
3087 * for transmit and receive operations are allocated, the interrupt
3088 * handler is registered with the OS, the watchdog timer is started,
3089 * and the stack is notified that the interface is ready.
3091 static int e1000_open(struct net_device *netdev)
3093 struct e1000_adapter *adapter = netdev_priv(netdev);
3094 struct e1000_hw *hw = &adapter->hw;
3097 /* disallow open during test */
3098 if (test_bit(__E1000_TESTING, &adapter->state))
3101 netif_carrier_off(netdev);
3103 /* allocate transmit descriptors */
3104 err = e1000e_setup_tx_resources(adapter);
3108 /* allocate receive descriptors */
3109 err = e1000e_setup_rx_resources(adapter);
3113 e1000e_power_up_phy(adapter);
3115 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3116 if ((adapter->hw.mng_cookie.status &
3117 E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3118 e1000_update_mng_vlan(adapter);
3121 * If AMT is enabled, let the firmware know that the network
3122 * interface is now open
3124 if (adapter->flags & FLAG_HAS_AMT)
3125 e1000_get_hw_control(adapter);
3128 * before we allocate an interrupt, we must be ready to handle it.
3129 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3130 * as soon as we call pci_request_irq, so we have to setup our
3131 * clean_rx handler before we do so.
3133 e1000_configure(adapter);
3135 err = e1000_request_irq(adapter);
3140 * Work around PCIe errata with MSI interrupts causing some chipsets to
3141 * ignore e1000e MSI messages, which means we need to test our MSI
3144 if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3145 err = e1000_test_msi(adapter);
3147 e_err("Interrupt allocation failed\n");
3152 /* From here on the code is the same as e1000e_up() */
3153 clear_bit(__E1000_DOWN, &adapter->state);
3155 napi_enable(&adapter->napi);
3157 e1000_irq_enable(adapter);
3159 netif_start_queue(netdev);
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);
3179 * e1000_close - Disables a network interface
3180 * @netdev: network interface device structure
3182 * Returns 0, this is not allowed to fail
3184 * The close entry point is called when an interface is de-activated
3185 * by the OS. The hardware is still under the drivers control, but
3186 * needs to be disabled. A global MAC reset is issued to stop the
3187 * hardware, and all transmit and receive resources are freed.
3189 static int e1000_close(struct net_device *netdev)
3191 struct e1000_adapter *adapter = netdev_priv(netdev);
3193 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3194 e1000e_down(adapter);
3195 e1000_power_down_phy(adapter);
3196 e1000_free_irq(adapter);
3198 e1000e_free_tx_resources(adapter);
3199 e1000e_free_rx_resources(adapter);
3202 * kill manageability vlan ID if supported, but not if a vlan with
3203 * the same ID is registered on the host OS (let 8021q kill it)
3205 if ((adapter->hw.mng_cookie.status &
3206 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3208 vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3209 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3212 * If AMT is enabled, let the firmware know that the network
3213 * interface is now closed
3215 if (adapter->flags & FLAG_HAS_AMT)
3216 e1000_release_hw_control(adapter);
3221 * e1000_set_mac - Change the Ethernet Address of the NIC
3222 * @netdev: network interface device structure
3223 * @p: pointer to an address structure
3225 * Returns 0 on success, negative on failure
3227 static int e1000_set_mac(struct net_device *netdev, void *p)
3229 struct e1000_adapter *adapter = netdev_priv(netdev);
3230 struct sockaddr *addr = p;
3232 if (!is_valid_ether_addr(addr->sa_data))
3233 return -EADDRNOTAVAIL;
3235 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3236 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3238 e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3240 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3241 /* activate the work around */
3242 e1000e_set_laa_state_82571(&adapter->hw, 1);
3245 * Hold a copy of the LAA in RAR[14] This is done so that
3246 * between the time RAR[0] gets clobbered and the time it
3247 * gets fixed (in e1000_watchdog), the actual LAA is in one
3248 * of the RARs and no incoming packets directed to this port
3249 * are dropped. Eventually the LAA will be in RAR[0] and
3252 e1000e_rar_set(&adapter->hw,
3253 adapter->hw.mac.addr,
3254 adapter->hw.mac.rar_entry_count - 1);
3261 * e1000e_update_phy_task - work thread to update phy
3262 * @work: pointer to our work struct
3264 * this worker thread exists because we must acquire a
3265 * semaphore to read the phy, which we could msleep while
3266 * waiting for it, and we can't msleep in a timer.
3268 static void e1000e_update_phy_task(struct work_struct *work)
3270 struct e1000_adapter *adapter = container_of(work,
3271 struct e1000_adapter, update_phy_task);
3272 e1000_get_phy_info(&adapter->hw);
3276 * Need to wait a few seconds after link up to get diagnostic information from
3279 static void e1000_update_phy_info(unsigned long data)
3281 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3282 schedule_work(&adapter->update_phy_task);
3286 * e1000e_update_stats - Update the board statistics counters
3287 * @adapter: board private structure
3289 void e1000e_update_stats(struct e1000_adapter *adapter)
3291 struct net_device *netdev = adapter->netdev;
3292 struct e1000_hw *hw = &adapter->hw;
3293 struct pci_dev *pdev = adapter->pdev;
3297 * Prevent stats update while adapter is being reset, or if the pci
3298 * connection is down.
3300 if (adapter->link_speed == 0)
3302 if (pci_channel_offline(pdev))
3305 adapter->stats.crcerrs += er32(CRCERRS);
3306 adapter->stats.gprc += er32(GPRC);
3307 adapter->stats.gorc += er32(GORCL);
3308 er32(GORCH); /* Clear gorc */
3309 adapter->stats.bprc += er32(BPRC);
3310 adapter->stats.mprc += er32(MPRC);
3311 adapter->stats.roc += er32(ROC);
3313 adapter->stats.mpc += er32(MPC);
3314 if ((hw->phy.type == e1000_phy_82578) ||
3315 (hw->phy.type == e1000_phy_82577)) {
3316 e1e_rphy(hw, HV_SCC_UPPER, &phy_data);
3317 e1e_rphy(hw, HV_SCC_LOWER, &phy_data);
3318 adapter->stats.scc += phy_data;
3320 e1e_rphy(hw, HV_ECOL_UPPER, &phy_data);
3321 e1e_rphy(hw, HV_ECOL_LOWER, &phy_data);
3322 adapter->stats.ecol += phy_data;
3324 e1e_rphy(hw, HV_MCC_UPPER, &phy_data);
3325 e1e_rphy(hw, HV_MCC_LOWER, &phy_data);
3326 adapter->stats.mcc += phy_data;
3328 e1e_rphy(hw, HV_LATECOL_UPPER, &phy_data);
3329 e1e_rphy(hw, HV_LATECOL_LOWER, &phy_data);
3330 adapter->stats.latecol += phy_data;
3332 e1e_rphy(hw, HV_DC_UPPER, &phy_data);
3333 e1e_rphy(hw, HV_DC_LOWER, &phy_data);
3334 adapter->stats.dc += phy_data;
3336 adapter->stats.scc += er32(SCC);
3337 adapter->stats.ecol += er32(ECOL);
3338 adapter->stats.mcc += er32(MCC);
3339 adapter->stats.latecol += er32(LATECOL);
3340 adapter->stats.dc += er32(DC);
3342 adapter->stats.xonrxc += er32(XONRXC);
3343 adapter->stats.xontxc += er32(XONTXC);
3344 adapter->stats.xoffrxc += er32(XOFFRXC);
3345 adapter->stats.xofftxc += er32(XOFFTXC);
3346 adapter->stats.gptc += er32(GPTC);
3347 adapter->stats.gotc += er32(GOTCL);
3348 er32(GOTCH); /* Clear gotc */
3349 adapter->stats.rnbc += er32(RNBC);
3350 adapter->stats.ruc += er32(RUC);
3352 adapter->stats.mptc += er32(MPTC);
3353 adapter->stats.bptc += er32(BPTC);
3355 /* used for adaptive IFS */
3357 hw->mac.tx_packet_delta = er32(TPT);
3358 adapter->stats.tpt += hw->mac.tx_packet_delta;
3359 if ((hw->phy.type == e1000_phy_82578) ||
3360 (hw->phy.type == e1000_phy_82577)) {
3361 e1e_rphy(hw, HV_COLC_UPPER, &phy_data);
3362 e1e_rphy(hw, HV_COLC_LOWER, &phy_data);
3363 hw->mac.collision_delta = phy_data;
3365 hw->mac.collision_delta = er32(COLC);
3367 adapter->stats.colc += hw->mac.collision_delta;
3369 adapter->stats.algnerrc += er32(ALGNERRC);
3370 adapter->stats.rxerrc += er32(RXERRC);
3371 if ((hw->phy.type == e1000_phy_82578) ||
3372 (hw->phy.type == e1000_phy_82577)) {
3373 e1e_rphy(hw, HV_TNCRS_UPPER, &phy_data);
3374 e1e_rphy(hw, HV_TNCRS_LOWER, &phy_data);
3375 adapter->stats.tncrs += phy_data;
3377 if ((hw->mac.type != e1000_82574) &&
3378 (hw->mac.type != e1000_82583))
3379 adapter->stats.tncrs += er32(TNCRS);
3381 adapter->stats.cexterr += er32(CEXTERR);
3382 adapter->stats.tsctc += er32(TSCTC);
3383 adapter->stats.tsctfc += er32(TSCTFC);
3385 /* Fill out the OS statistics structure */
3386 netdev->stats.multicast = adapter->stats.mprc;
3387 netdev->stats.collisions = adapter->stats.colc;
3392 * RLEC on some newer hardware can be incorrect so build
3393 * our own version based on RUC and ROC
3395 netdev->stats.rx_errors = adapter->stats.rxerrc +
3396 adapter->stats.crcerrs + adapter->stats.algnerrc +
3397 adapter->stats.ruc + adapter->stats.roc +
3398 adapter->stats.cexterr;
3399 netdev->stats.rx_length_errors = adapter->stats.ruc +
3401 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3402 netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3403 netdev->stats.rx_missed_errors = adapter->stats.mpc;
3406 netdev->stats.tx_errors = adapter->stats.ecol +
3407 adapter->stats.latecol;
3408 netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3409 netdev->stats.tx_window_errors = adapter->stats.latecol;
3410 netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3412 /* Tx Dropped needs to be maintained elsewhere */
3414 /* Management Stats */
3415 adapter->stats.mgptc += er32(MGTPTC);
3416 adapter->stats.mgprc += er32(MGTPRC);
3417 adapter->stats.mgpdc += er32(MGTPDC);
3421 * e1000_phy_read_status - Update the PHY register status snapshot
3422 * @adapter: board private structure
3424 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3426 struct e1000_hw *hw = &adapter->hw;
3427 struct e1000_phy_regs *phy = &adapter->phy_regs;
3430 if ((er32(STATUS) & E1000_STATUS_LU) &&
3431 (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3432 ret_val = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3433 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3434 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3435 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3436 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3437 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3438 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3439 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3441 e_warn("Error reading PHY register\n");
3444 * Do not read PHY registers if link is not up
3445 * Set values to typical power-on defaults
3447 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3448 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3449 BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3451 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3452 ADVERTISE_ALL | ADVERTISE_CSMA);
3454 phy->expansion = EXPANSION_ENABLENPAGE;
3455 phy->ctrl1000 = ADVERTISE_1000FULL;
3457 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3461 static void e1000_print_link_info(struct e1000_adapter *adapter)
3463 struct e1000_hw *hw = &adapter->hw;
3464 u32 ctrl = er32(CTRL);
3466 /* Link status message must follow this format for user tools */
3467 printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3468 "Flow Control: %s\n",
3469 adapter->netdev->name,
3470 adapter->link_speed,
3471 (adapter->link_duplex == FULL_DUPLEX) ?
3472 "Full Duplex" : "Half Duplex",
3473 ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3475 ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3476 ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3479 bool e1000_has_link(struct e1000_adapter *adapter)
3481 struct e1000_hw *hw = &adapter->hw;
3482 bool link_active = 0;
3486 * get_link_status is set on LSC (link status) interrupt or
3487 * Rx sequence error interrupt. get_link_status will stay
3488 * false until the check_for_link establishes link
3489 * for copper adapters ONLY
3491 switch (hw->phy.media_type) {
3492 case e1000_media_type_copper:
3493 if (hw->mac.get_link_status) {
3494 ret_val = hw->mac.ops.check_for_link(hw);
3495 link_active = !hw->mac.get_link_status;
3500 case e1000_media_type_fiber:
3501 ret_val = hw->mac.ops.check_for_link(hw);
3502 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3504 case e1000_media_type_internal_serdes:
3505 ret_val = hw->mac.ops.check_for_link(hw);
3506 link_active = adapter->hw.mac.serdes_has_link;
3509 case e1000_media_type_unknown:
3513 if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3514 (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3515 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3516 e_info("Gigabit has been disabled, downgrading speed\n");
3522 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3524 /* make sure the receive unit is started */
3525 if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3526 (adapter->flags & FLAG_RX_RESTART_NOW)) {
3527 struct e1000_hw *hw = &adapter->hw;
3528 u32 rctl = er32(RCTL);
3529 ew32(RCTL, rctl | E1000_RCTL_EN);
3530 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3535 * e1000_watchdog - Timer Call-back
3536 * @data: pointer to adapter cast into an unsigned long
3538 static void e1000_watchdog(unsigned long data)
3540 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3542 /* Do the rest outside of interrupt context */
3543 schedule_work(&adapter->watchdog_task);
3545 /* TODO: make this use queue_delayed_work() */
3548 static void e1000_watchdog_task(struct work_struct *work)
3550 struct e1000_adapter *adapter = container_of(work,
3551 struct e1000_adapter, watchdog_task);
3552 struct net_device *netdev = adapter->netdev;
3553 struct e1000_mac_info *mac = &adapter->hw.mac;
3554 struct e1000_phy_info *phy = &adapter->hw.phy;
3555 struct e1000_ring *tx_ring = adapter->tx_ring;
3556 struct e1000_hw *hw = &adapter->hw;
3560 link = e1000_has_link(adapter);
3561 if ((netif_carrier_ok(netdev)) && link) {
3562 e1000e_enable_receives(adapter);
3566 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3567 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3568 e1000_update_mng_vlan(adapter);
3571 if (!netif_carrier_ok(netdev)) {
3573 /* update snapshot of PHY registers on LSC */
3574 e1000_phy_read_status(adapter);
3575 mac->ops.get_link_up_info(&adapter->hw,
3576 &adapter->link_speed,
3577 &adapter->link_duplex);
3578 e1000_print_link_info(adapter);
3580 * On supported PHYs, check for duplex mismatch only
3581 * if link has autonegotiated at 10/100 half
3583 if ((hw->phy.type == e1000_phy_igp_3 ||
3584 hw->phy.type == e1000_phy_bm) &&
3585 (hw->mac.autoneg == true) &&
3586 (adapter->link_speed == SPEED_10 ||
3587 adapter->link_speed == SPEED_100) &&
3588 (adapter->link_duplex == HALF_DUPLEX)) {
3591 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3593 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3594 e_info("Autonegotiated half duplex but"
3595 " link partner cannot autoneg. "
3596 " Try forcing full duplex if "
3597 "link gets many collisions.\n");
3601 * tweak tx_queue_len according to speed/duplex
3602 * and adjust the timeout factor
3604 netdev->tx_queue_len = adapter->tx_queue_len;
3605 adapter->tx_timeout_factor = 1;
3606 switch (adapter->link_speed) {
3609 netdev->tx_queue_len = 10;
3610 adapter->tx_timeout_factor = 16;
3614 netdev->tx_queue_len = 100;
3615 /* maybe add some timeout factor ? */
3620 * workaround: re-program speed mode bit after
3623 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3626 tarc0 = er32(TARC(0));
3627 tarc0 &= ~SPEED_MODE_BIT;
3628 ew32(TARC(0), tarc0);
3632 * disable TSO for pcie and 10/100 speeds, to avoid
3633 * some hardware issues
3635 if (!(adapter->flags & FLAG_TSO_FORCE)) {
3636 switch (adapter->link_speed) {
3639 e_info("10/100 speed: disabling TSO\n");
3640 netdev->features &= ~NETIF_F_TSO;
3641 netdev->features &= ~NETIF_F_TSO6;
3644 netdev->features |= NETIF_F_TSO;
3645 netdev->features |= NETIF_F_TSO6;
3654 * enable transmits in the hardware, need to do this
3655 * after setting TARC(0)
3658 tctl |= E1000_TCTL_EN;
3662 * Perform any post-link-up configuration before
3663 * reporting link up.
3665 if (phy->ops.cfg_on_link_up)
3666 phy->ops.cfg_on_link_up(hw);
3668 netif_carrier_on(netdev);
3670 if (!test_bit(__E1000_DOWN, &adapter->state))
3671 mod_timer(&adapter->phy_info_timer,
3672 round_jiffies(jiffies + 2 * HZ));
3675 if (netif_carrier_ok(netdev)) {
3676 adapter->link_speed = 0;
3677 adapter->link_duplex = 0;
3678 /* Link status message must follow this format */
3679 printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
3680 adapter->netdev->name);
3681 netif_carrier_off(netdev);
3682 if (!test_bit(__E1000_DOWN, &adapter->state))
3683 mod_timer(&adapter->phy_info_timer,
3684 round_jiffies(jiffies + 2 * HZ));
3686 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3687 schedule_work(&adapter->reset_task);
3692 e1000e_update_stats(adapter);
3694 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3695 adapter->tpt_old = adapter->stats.tpt;
3696 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3697 adapter->colc_old = adapter->stats.colc;
3699 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3700 adapter->gorc_old = adapter->stats.gorc;
3701 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3702 adapter->gotc_old = adapter->stats.gotc;
3704 e1000e_update_adaptive(&adapter->hw);
3706 if (!netif_carrier_ok(netdev)) {
3707 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3711 * We've lost link, so the controller stops DMA,
3712 * but we've got queued Tx work that's never going
3713 * to get done, so reset controller to flush Tx.
3714 * (Do the reset outside of interrupt context).
3716 adapter->tx_timeout_count++;
3717 schedule_work(&adapter->reset_task);
3718 /* return immediately since reset is imminent */
3723 /* Cause software interrupt to ensure Rx ring is cleaned */
3724 if (adapter->msix_entries)
3725 ew32(ICS, adapter->rx_ring->ims_val);
3727 ew32(ICS, E1000_ICS_RXDMT0);
3729 /* Force detection of hung controller every watchdog period */
3730 adapter->detect_tx_hung = 1;
3733 * With 82571 controllers, LAA may be overwritten due to controller
3734 * reset from the other port. Set the appropriate LAA in RAR[0]
3736 if (e1000e_get_laa_state_82571(hw))
3737 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3739 /* Reset the timer */
3740 if (!test_bit(__E1000_DOWN, &adapter->state))
3741 mod_timer(&adapter->watchdog_timer,
3742 round_jiffies(jiffies + 2 * HZ));
3745 #define E1000_TX_FLAGS_CSUM 0x00000001
3746 #define E1000_TX_FLAGS_VLAN 0x00000002
3747 #define E1000_TX_FLAGS_TSO 0x00000004
3748 #define E1000_TX_FLAGS_IPV4 0x00000008
3749 #define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
3750 #define E1000_TX_FLAGS_VLAN_SHIFT 16
3752 static int e1000_tso(struct e1000_adapter *adapter,
3753 struct sk_buff *skb)
3755 struct e1000_ring *tx_ring = adapter->tx_ring;
3756 struct e1000_context_desc *context_desc;
3757 struct e1000_buffer *buffer_info;
3760 u16 ipcse = 0, tucse, mss;
3761 u8 ipcss, ipcso, tucss, tucso, hdr_len;
3764 if (!skb_is_gso(skb))
3767 if (skb_header_cloned(skb)) {
3768 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3773 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3774 mss = skb_shinfo(skb)->gso_size;
3775 if (skb->protocol == htons(ETH_P_IP)) {
3776 struct iphdr *iph = ip_hdr(skb);
3779 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
3781 cmd_length = E1000_TXD_CMD_IP;
3782 ipcse = skb_transport_offset(skb) - 1;
3783 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3784 ipv6_hdr(skb)->payload_len = 0;
3785 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3786 &ipv6_hdr(skb)->daddr,
3790 ipcss = skb_network_offset(skb);
3791 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3792 tucss = skb_transport_offset(skb);
3793 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3796 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3797 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3799 i = tx_ring->next_to_use;
3800 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3801 buffer_info = &tx_ring->buffer_info[i];
3803 context_desc->lower_setup.ip_fields.ipcss = ipcss;
3804 context_desc->lower_setup.ip_fields.ipcso = ipcso;
3805 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
3806 context_desc->upper_setup.tcp_fields.tucss = tucss;
3807 context_desc->upper_setup.tcp_fields.tucso = tucso;
3808 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3809 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
3810 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3811 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3813 buffer_info->time_stamp = jiffies;
3814 buffer_info->next_to_watch = i;
3817 if (i == tx_ring->count)
3819 tx_ring->next_to_use = i;
3824 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3826 struct e1000_ring *tx_ring = adapter->tx_ring;
3827 struct e1000_context_desc *context_desc;
3828 struct e1000_buffer *buffer_info;
3831 u32 cmd_len = E1000_TXD_CMD_DEXT;
3834 if (skb->ip_summed != CHECKSUM_PARTIAL)
3837 if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
3838 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
3840 protocol = skb->protocol;
3843 case cpu_to_be16(ETH_P_IP):
3844 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3845 cmd_len |= E1000_TXD_CMD_TCP;
3847 case cpu_to_be16(ETH_P_IPV6):
3848 /* XXX not handling all IPV6 headers */
3849 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3850 cmd_len |= E1000_TXD_CMD_TCP;
3853 if (unlikely(net_ratelimit()))
3854 e_warn("checksum_partial proto=%x!\n",
3855 be16_to_cpu(protocol));
3859 css = skb_transport_offset(skb);
3861 i = tx_ring->next_to_use;
3862 buffer_info = &tx_ring->buffer_info[i];
3863 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3865 context_desc->lower_setup.ip_config = 0;
3866 context_desc->upper_setup.tcp_fields.tucss = css;
3867 context_desc->upper_setup.tcp_fields.tucso =
3868 css + skb->csum_offset;
3869 context_desc->upper_setup.tcp_fields.tucse = 0;
3870 context_desc->tcp_seg_setup.data = 0;
3871 context_desc->cmd_and_length = cpu_to_le32(cmd_len);
3873 buffer_info->time_stamp = jiffies;
3874 buffer_info->next_to_watch = i;
3877 if (i == tx_ring->count)
3879 tx_ring->next_to_use = i;
3884 #define E1000_MAX_PER_TXD 8192
3885 #define E1000_MAX_TXD_PWR 12
3887 static int e1000_tx_map(struct e1000_adapter *adapter,
3888 struct sk_buff *skb, unsigned int first,
3889 unsigned int max_per_txd, unsigned int nr_frags,
3892 struct e1000_ring *tx_ring = adapter->tx_ring;
3893 struct e1000_buffer *buffer_info;
3894 unsigned int len = skb_headlen(skb);
3895 unsigned int offset, size, count = 0, i;
3899 i = tx_ring->next_to_use;
3901 if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
3902 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3903 adapter->tx_dma_failed++;
3907 map = skb_shinfo(skb)->dma_maps;
3911 buffer_info = &tx_ring->buffer_info[i];
3912 size = min(len, max_per_txd);
3914 buffer_info->length = size;
3915 buffer_info->time_stamp = jiffies;
3916 buffer_info->next_to_watch = i;
3917 buffer_info->dma = skb_shinfo(skb)->dma_head + offset;
3925 if (i == tx_ring->count)
3930 for (f = 0; f < nr_frags; f++) {
3931 struct skb_frag_struct *frag;
3933 frag = &skb_shinfo(skb)->frags[f];
3939 if (i == tx_ring->count)
3942 buffer_info = &tx_ring->buffer_info[i];
3943 size = min(len, max_per_txd);
3945 buffer_info->length = size;
3946 buffer_info->time_stamp = jiffies;
3947 buffer_info->next_to_watch = i;
3948 buffer_info->dma = map[f] + offset;
3956 tx_ring->buffer_info[i].skb = skb;
3957 tx_ring->buffer_info[first].next_to_watch = i;
3962 static void e1000_tx_queue(struct e1000_adapter *adapter,
3963 int tx_flags, int count)
3965 struct e1000_ring *tx_ring = adapter->tx_ring;
3966 struct e1000_tx_desc *tx_desc = NULL;
3967 struct e1000_buffer *buffer_info;
3968 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3971 if (tx_flags & E1000_TX_FLAGS_TSO) {
3972 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3974 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3976 if (tx_flags & E1000_TX_FLAGS_IPV4)
3977 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3980 if (tx_flags & E1000_TX_FLAGS_CSUM) {
3981 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3982 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3985 if (tx_flags & E1000_TX_FLAGS_VLAN) {
3986 txd_lower |= E1000_TXD_CMD_VLE;
3987 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3990 i = tx_ring->next_to_use;
3993 buffer_info = &tx_ring->buffer_info[i];
3994 tx_desc = E1000_TX_DESC(*tx_ring, i);
3995 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3996 tx_desc->lower.data =
3997 cpu_to_le32(txd_lower | buffer_info->length);
3998 tx_desc->upper.data = cpu_to_le32(txd_upper);
4001 if (i == tx_ring->count)
4005 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4008 * Force memory writes to complete before letting h/w
4009 * know there are new descriptors to fetch. (Only
4010 * applicable for weak-ordered memory model archs,
4015 tx_ring->next_to_use = i;
4016 writel(i, adapter->hw.hw_addr + tx_ring->tail);
4018 * we need this if more than one processor can write to our tail
4019 * at a time, it synchronizes IO on IA64/Altix systems
4024 #define MINIMUM_DHCP_PACKET_SIZE 282
4025 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4026 struct sk_buff *skb)
4028 struct e1000_hw *hw = &adapter->hw;
4031 if (vlan_tx_tag_present(skb)) {
4032 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
4033 && (adapter->hw.mng_cookie.status &
4034 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4038 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4041 if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4045 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4048 if (ip->protocol != IPPROTO_UDP)
4051 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4052 if (ntohs(udp->dest) != 67)
4055 offset = (u8 *)udp + 8 - skb->data;
4056 length = skb->len - offset;
4057 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4063 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4065 struct e1000_adapter *adapter = netdev_priv(netdev);
4067 netif_stop_queue(netdev);
4069 * Herbert's original patch had:
4070 * smp_mb__after_netif_stop_queue();
4071 * but since that doesn't exist yet, just open code it.
4076 * We need to check again in a case another CPU has just
4077 * made room available.
4079 if (e1000_desc_unused(adapter->tx_ring) < size)
4083 netif_start_queue(netdev);
4084 ++adapter->restart_queue;
4088 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4090 struct e1000_adapter *adapter = netdev_priv(netdev);
4092 if (e1000_desc_unused(adapter->tx_ring) >= size)
4094 return __e1000_maybe_stop_tx(netdev, size);
4097 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4098 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4099 struct net_device *netdev)
4101 struct e1000_adapter *adapter = netdev_priv(netdev);
4102 struct e1000_ring *tx_ring = adapter->tx_ring;
4104 unsigned int max_per_txd = E1000_MAX_PER_TXD;
4105 unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4106 unsigned int tx_flags = 0;
4107 unsigned int len = skb->len - skb->data_len;
4108 unsigned int nr_frags;
4114 if (test_bit(__E1000_DOWN, &adapter->state)) {
4115 dev_kfree_skb_any(skb);
4116 return NETDEV_TX_OK;
4119 if (skb->len <= 0) {
4120 dev_kfree_skb_any(skb);
4121 return NETDEV_TX_OK;
4124 mss = skb_shinfo(skb)->gso_size;
4126 * The controller does a simple calculation to
4127 * make sure there is enough room in the FIFO before
4128 * initiating the DMA for each buffer. The calc is:
4129 * 4 = ceil(buffer len/mss). To make sure we don't
4130 * overrun the FIFO, adjust the max buffer len if mss
4135 max_per_txd = min(mss << 2, max_per_txd);
4136 max_txd_pwr = fls(max_per_txd) - 1;
4139 * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4140 * points to just header, pull a few bytes of payload from
4141 * frags into skb->data
4143 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4145 * we do this workaround for ES2LAN, but it is un-necessary,
4146 * avoiding it could save a lot of cycles
4148 if (skb->data_len && (hdr_len == len)) {
4149 unsigned int pull_size;
4151 pull_size = min((unsigned int)4, skb->data_len);
4152 if (!__pskb_pull_tail(skb, pull_size)) {
4153 e_err("__pskb_pull_tail failed.\n");
4154 dev_kfree_skb_any(skb);
4155 return NETDEV_TX_OK;
4157 len = skb->len - skb->data_len;
4161 /* reserve a descriptor for the offload context */
4162 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4166 count += TXD_USE_COUNT(len, max_txd_pwr);
4168 nr_frags = skb_shinfo(skb)->nr_frags;
4169 for (f = 0; f < nr_frags; f++)
4170 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4173 if (adapter->hw.mac.tx_pkt_filtering)
4174 e1000_transfer_dhcp_info(adapter, skb);
4177 * need: count + 2 desc gap to keep tail from touching
4178 * head, otherwise try next time
4180 if (e1000_maybe_stop_tx(netdev, count + 2))
4181 return NETDEV_TX_BUSY;
4183 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4184 tx_flags |= E1000_TX_FLAGS_VLAN;
4185 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4188 first = tx_ring->next_to_use;
4190 tso = e1000_tso(adapter, skb);
4192 dev_kfree_skb_any(skb);
4193 return NETDEV_TX_OK;
4197 tx_flags |= E1000_TX_FLAGS_TSO;
4198 else if (e1000_tx_csum(adapter, skb))
4199 tx_flags |= E1000_TX_FLAGS_CSUM;
4202 * Old method was to assume IPv4 packet by default if TSO was enabled.
4203 * 82571 hardware supports TSO capabilities for IPv6 as well...
4204 * no longer assume, we must.
4206 if (skb->protocol == htons(ETH_P_IP))
4207 tx_flags |= E1000_TX_FLAGS_IPV4;
4209 /* if count is 0 then mapping error has occured */
4210 count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4212 e1000_tx_queue(adapter, tx_flags, count);
4213 /* Make sure there is space in the ring for the next send. */
4214 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4217 dev_kfree_skb_any(skb);
4218 tx_ring->buffer_info[first].time_stamp = 0;
4219 tx_ring->next_to_use = first;
4222 return NETDEV_TX_OK;
4226 * e1000_tx_timeout - Respond to a Tx Hang
4227 * @netdev: network interface device structure
4229 static void e1000_tx_timeout(struct net_device *netdev)
4231 struct e1000_adapter *adapter = netdev_priv(netdev);
4233 /* Do the reset outside of interrupt context */
4234 adapter->tx_timeout_count++;
4235 schedule_work(&adapter->reset_task);
4238 static void e1000_reset_task(struct work_struct *work)
4240 struct e1000_adapter *adapter;
4241 adapter = container_of(work, struct e1000_adapter, reset_task);
4243 e1000e_reinit_locked(adapter);
4247 * e1000_get_stats - Get System Network Statistics
4248 * @netdev: network interface device structure
4250 * Returns the address of the device statistics structure.
4251 * The statistics are actually updated from the timer callback.
4253 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4255 /* only return the current stats */
4256 return &netdev->stats;
4260 * e1000_change_mtu - Change the Maximum Transfer Unit
4261 * @netdev: network interface device structure
4262 * @new_mtu: new value for maximum frame size
4264 * Returns 0 on success, negative on failure
4266 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4268 struct e1000_adapter *adapter = netdev_priv(netdev);
4269 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4271 /* Jumbo frame support */
4272 if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4273 !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4274 e_err("Jumbo Frames not supported.\n");
4278 /* Supported frame sizes */
4279 if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4280 (max_frame > adapter->max_hw_frame_size)) {
4281 e_err("Unsupported MTU setting\n");
4285 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4287 /* e1000e_down has a dependency on max_frame_size */
4288 adapter->max_frame_size = max_frame;
4289 if (netif_running(netdev))
4290 e1000e_down(adapter);
4293 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4294 * means we reserve 2 more, this pushes us to allocate from the next
4296 * i.e. RXBUFFER_2048 --> size-4096 slab
4297 * However with the new *_jumbo_rx* routines, jumbo receives will use
4301 if (max_frame <= 256)
4302 adapter->rx_buffer_len = 256;
4303 else if (max_frame <= 512)
4304 adapter->rx_buffer_len = 512;
4305 else if (max_frame <= 1024)
4306 adapter->rx_buffer_len = 1024;
4307 else if (max_frame <= 2048)
4308 adapter->rx_buffer_len = 2048;
4310 adapter->rx_buffer_len = 4096;
4312 /* adjust allocation if LPE protects us, and we aren't using SBP */
4313 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4314 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4315 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4318 e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4319 netdev->mtu = new_mtu;
4321 if (netif_running(netdev))
4324 e1000e_reset(adapter);
4326 clear_bit(__E1000_RESETTING, &adapter->state);
4331 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4334 struct e1000_adapter *adapter = netdev_priv(netdev);
4335 struct mii_ioctl_data *data = if_mii(ifr);
4337 if (adapter->hw.phy.media_type != e1000_media_type_copper)
4342 data->phy_id = adapter->hw.phy.addr;
4345 e1000_phy_read_status(adapter);
4347 switch (data->reg_num & 0x1F) {
4349 data->val_out = adapter->phy_regs.bmcr;
4352 data->val_out = adapter->phy_regs.bmsr;
4355 data->val_out = (adapter->hw.phy.id >> 16);
4358 data->val_out = (adapter->hw.phy.id & 0xFFFF);
4361 data->val_out = adapter->phy_regs.advertise;
4364 data->val_out = adapter->phy_regs.lpa;
4367 data->val_out = adapter->phy_regs.expansion;
4370 data->val_out = adapter->phy_regs.ctrl1000;
4373 data->val_out = adapter->phy_regs.stat1000;
4376 data->val_out = adapter->phy_regs.estatus;
4389 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4395 return e1000_mii_ioctl(netdev, ifr, cmd);
4401 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
4403 struct e1000_hw *hw = &adapter->hw;
4408 /* copy MAC RARs to PHY RARs */
4409 for (i = 0; i < adapter->hw.mac.rar_entry_count; i++) {
4410 mac_reg = er32(RAL(i));
4411 e1e_wphy(hw, BM_RAR_L(i), (u16)(mac_reg & 0xFFFF));
4412 e1e_wphy(hw, BM_RAR_M(i), (u16)((mac_reg >> 16) & 0xFFFF));
4413 mac_reg = er32(RAH(i));
4414 e1e_wphy(hw, BM_RAR_H(i), (u16)(mac_reg & 0xFFFF));
4415 e1e_wphy(hw, BM_RAR_CTRL(i), (u16)((mac_reg >> 16) & 0xFFFF));
4418 /* copy MAC MTA to PHY MTA */
4419 for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
4420 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
4421 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
4422 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
4425 /* configure PHY Rx Control register */
4426 e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
4427 mac_reg = er32(RCTL);
4428 if (mac_reg & E1000_RCTL_UPE)
4429 phy_reg |= BM_RCTL_UPE;
4430 if (mac_reg & E1000_RCTL_MPE)
4431 phy_reg |= BM_RCTL_MPE;
4432 phy_reg &= ~(BM_RCTL_MO_MASK);
4433 if (mac_reg & E1000_RCTL_MO_3)
4434 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
4435 << BM_RCTL_MO_SHIFT);
4436 if (mac_reg & E1000_RCTL_BAM)
4437 phy_reg |= BM_RCTL_BAM;
4438 if (mac_reg & E1000_RCTL_PMCF)
4439 phy_reg |= BM_RCTL_PMCF;
4440 mac_reg = er32(CTRL);
4441 if (mac_reg & E1000_CTRL_RFCE)
4442 phy_reg |= BM_RCTL_RFCE;
4443 e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
4445 /* enable PHY wakeup in MAC register */
4447 ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
4449 /* configure and enable PHY wakeup in PHY registers */
4450 e1e_wphy(&adapter->hw, BM_WUFC, wufc);
4451 e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
4453 /* activate PHY wakeup */
4454 retval = hw->phy.ops.acquire(hw);
4456 e_err("Could not acquire PHY\n");
4459 e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4460 (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
4461 retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
4463 e_err("Could not read PHY page 769\n");
4466 phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
4467 retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
4469 e_err("Could not set PHY Host Wakeup bit\n");
4471 hw->phy.ops.release(hw);
4476 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4478 struct net_device *netdev = pci_get_drvdata(pdev);
4479 struct e1000_adapter *adapter = netdev_priv(netdev);
4480 struct e1000_hw *hw = &adapter->hw;
4481 u32 ctrl, ctrl_ext, rctl, status;
4482 u32 wufc = adapter->wol;
4485 netif_device_detach(netdev);
4487 if (netif_running(netdev)) {
4488 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4489 e1000e_down(adapter);
4490 e1000_free_irq(adapter);
4492 e1000e_reset_interrupt_capability(adapter);
4494 retval = pci_save_state(pdev);
4498 status = er32(STATUS);
4499 if (status & E1000_STATUS_LU)
4500 wufc &= ~E1000_WUFC_LNKC;
4503 e1000_setup_rctl(adapter);
4504 e1000_set_multi(netdev);
4506 /* turn on all-multi mode if wake on multicast is enabled */
4507 if (wufc & E1000_WUFC_MC) {
4509 rctl |= E1000_RCTL_MPE;
4514 /* advertise wake from D3Cold */
4515 #define E1000_CTRL_ADVD3WUC 0x00100000
4516 /* phy power management enable */
4517 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4518 ctrl |= E1000_CTRL_ADVD3WUC;
4519 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
4520 ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
4523 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4524 adapter->hw.phy.media_type ==
4525 e1000_media_type_internal_serdes) {
4526 /* keep the laser running in D3 */
4527 ctrl_ext = er32(CTRL_EXT);
4528 ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4529 ew32(CTRL_EXT, ctrl_ext);
4532 if (adapter->flags & FLAG_IS_ICH)
4533 e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4535 /* Allow time for pending master requests to run */
4536 e1000e_disable_pcie_master(&adapter->hw);
4538 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4539 /* enable wakeup by the PHY */
4540 retval = e1000_init_phy_wakeup(adapter, wufc);
4544 /* enable wakeup by the MAC */
4546 ew32(WUC, E1000_WUC_PME_EN);
4553 *enable_wake = !!wufc;
4555 /* make sure adapter isn't asleep if manageability is enabled */
4556 if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
4557 (hw->mac.ops.check_mng_mode(hw)))
4558 *enable_wake = true;
4560 if (adapter->hw.phy.type == e1000_phy_igp_3)
4561 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4564 * Release control of h/w to f/w. If f/w is AMT enabled, this
4565 * would have already happened in close and is redundant.
4567 e1000_release_hw_control(adapter);
4569 pci_disable_device(pdev);
4574 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
4576 if (sleep && wake) {
4577 pci_prepare_to_sleep(pdev);
4581 pci_wake_from_d3(pdev, wake);
4582 pci_set_power_state(pdev, PCI_D3hot);
4585 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
4588 struct net_device *netdev = pci_get_drvdata(pdev);
4589 struct e1000_adapter *adapter = netdev_priv(netdev);
4592 * The pci-e switch on some quad port adapters will report a
4593 * correctable error when the MAC transitions from D0 to D3. To
4594 * prevent this we need to mask off the correctable errors on the
4595 * downstream port of the pci-e switch.
4597 if (adapter->flags & FLAG_IS_QUAD_PORT) {
4598 struct pci_dev *us_dev = pdev->bus->self;
4599 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
4602 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
4603 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
4604 (devctl & ~PCI_EXP_DEVCTL_CERE));
4606 e1000_power_off(pdev, sleep, wake);
4608 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
4610 e1000_power_off(pdev, sleep, wake);
4614 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4620 * 82573 workaround - disable L1 ASPM on mobile chipsets
4622 * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4623 * resulting in lost data or garbage information on the pci-e link
4624 * level. This could result in (false) bad EEPROM checksum errors,
4625 * long ping times (up to 2s) or even a system freeze/hang.
4627 * Unfortunately this feature saves about 1W power consumption when
4630 pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4631 pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4633 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4635 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4640 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4645 retval = __e1000_shutdown(pdev, &wake);
4647 e1000_complete_shutdown(pdev, true, wake);
4652 static int e1000_resume(struct pci_dev *pdev)
4654 struct net_device *netdev = pci_get_drvdata(pdev);
4655 struct e1000_adapter *adapter = netdev_priv(netdev);
4656 struct e1000_hw *hw = &adapter->hw;
4659 pci_set_power_state(pdev, PCI_D0);
4660 pci_restore_state(pdev);
4661 e1000e_disable_l1aspm(pdev);
4663 err = pci_enable_device_mem(pdev);
4666 "Cannot enable PCI device from suspend\n");
4670 pci_set_master(pdev);
4672 pci_enable_wake(pdev, PCI_D3hot, 0);
4673 pci_enable_wake(pdev, PCI_D3cold, 0);
4675 e1000e_set_interrupt_capability(adapter);
4676 if (netif_running(netdev)) {
4677 err = e1000_request_irq(adapter);
4682 e1000e_power_up_phy(adapter);
4684 /* report the system wakeup cause from S3/S4 */
4685 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4688 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
4690 e_info("PHY Wakeup cause - %s\n",
4691 phy_data & E1000_WUS_EX ? "Unicast Packet" :
4692 phy_data & E1000_WUS_MC ? "Multicast Packet" :
4693 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
4694 phy_data & E1000_WUS_MAG ? "Magic Packet" :
4695 phy_data & E1000_WUS_LNKC ? "Link Status "
4696 " Change" : "other");
4698 e1e_wphy(&adapter->hw, BM_WUS, ~0);
4700 u32 wus = er32(WUS);
4702 e_info("MAC Wakeup cause - %s\n",
4703 wus & E1000_WUS_EX ? "Unicast Packet" :
4704 wus & E1000_WUS_MC ? "Multicast Packet" :
4705 wus & E1000_WUS_BC ? "Broadcast Packet" :
4706 wus & E1000_WUS_MAG ? "Magic Packet" :
4707 wus & E1000_WUS_LNKC ? "Link Status Change" :
4713 e1000e_reset(adapter);
4715 e1000_init_manageability(adapter);
4717 if (netif_running(netdev))
4720 netif_device_attach(netdev);
4723 * If the controller has AMT, do not set DRV_LOAD until the interface
4724 * is up. For all other cases, let the f/w know that the h/w is now
4725 * under the control of the driver.
4727 if (!(adapter->flags & FLAG_HAS_AMT))
4728 e1000_get_hw_control(adapter);
4734 static void e1000_shutdown(struct pci_dev *pdev)
4738 __e1000_shutdown(pdev, &wake);
4740 if (system_state == SYSTEM_POWER_OFF)
4741 e1000_complete_shutdown(pdev, false, wake);
4744 #ifdef CONFIG_NET_POLL_CONTROLLER
4746 * Polling 'interrupt' - used by things like netconsole to send skbs
4747 * without having to re-enable interrupts. It's not called while
4748 * the interrupt routine is executing.
4750 static void e1000_netpoll(struct net_device *netdev)
4752 struct e1000_adapter *adapter = netdev_priv(netdev);
4754 disable_irq(adapter->pdev->irq);
4755 e1000_intr(adapter->pdev->irq, netdev);
4757 enable_irq(adapter->pdev->irq);
4762 * e1000_io_error_detected - called when PCI error is detected
4763 * @pdev: Pointer to PCI device
4764 * @state: The current pci connection state
4766 * This function is called after a PCI bus error affecting
4767 * this device has been detected.
4769 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4770 pci_channel_state_t state)
4772 struct net_device *netdev = pci_get_drvdata(pdev);
4773 struct e1000_adapter *adapter = netdev_priv(netdev);
4775 netif_device_detach(netdev);
4777 if (state == pci_channel_io_perm_failure)
4778 return PCI_ERS_RESULT_DISCONNECT;
4780 if (netif_running(netdev))
4781 e1000e_down(adapter);
4782 pci_disable_device(pdev);
4784 /* Request a slot slot reset. */
4785 return PCI_ERS_RESULT_NEED_RESET;
4789 * e1000_io_slot_reset - called after the pci bus has been reset.
4790 * @pdev: Pointer to PCI device
4792 * Restart the card from scratch, as if from a cold-boot. Implementation
4793 * resembles the first-half of the e1000_resume routine.
4795 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4797 struct net_device *netdev = pci_get_drvdata(pdev);
4798 struct e1000_adapter *adapter = netdev_priv(netdev);
4799 struct e1000_hw *hw = &adapter->hw;
4801 pci_ers_result_t result;
4803 e1000e_disable_l1aspm(pdev);
4804 err = pci_enable_device_mem(pdev);
4807 "Cannot re-enable PCI device after reset.\n");
4808 result = PCI_ERS_RESULT_DISCONNECT;
4810 pci_set_master(pdev);
4811 pci_restore_state(pdev);
4813 pci_enable_wake(pdev, PCI_D3hot, 0);
4814 pci_enable_wake(pdev, PCI_D3cold, 0);
4816 e1000e_reset(adapter);
4818 result = PCI_ERS_RESULT_RECOVERED;
4821 pci_cleanup_aer_uncorrect_error_status(pdev);
4827 * e1000_io_resume - called when traffic can start flowing again.
4828 * @pdev: Pointer to PCI device
4830 * This callback is called when the error recovery driver tells us that
4831 * its OK to resume normal operation. Implementation resembles the
4832 * second-half of the e1000_resume routine.
4834 static void e1000_io_resume(struct pci_dev *pdev)
4836 struct net_device *netdev = pci_get_drvdata(pdev);
4837 struct e1000_adapter *adapter = netdev_priv(netdev);
4839 e1000_init_manageability(adapter);
4841 if (netif_running(netdev)) {
4842 if (e1000e_up(adapter)) {
4844 "can't bring device back up after reset\n");
4849 netif_device_attach(netdev);
4852 * If the controller has AMT, do not set DRV_LOAD until the interface
4853 * is up. For all other cases, let the f/w know that the h/w is now
4854 * under the control of the driver.
4856 if (!(adapter->flags & FLAG_HAS_AMT))
4857 e1000_get_hw_control(adapter);
4861 static void e1000_print_device_info(struct e1000_adapter *adapter)
4863 struct e1000_hw *hw = &adapter->hw;
4864 struct net_device *netdev = adapter->netdev;
4867 /* print bus type/speed/width info */
4868 e_info("(PCI Express:2.5GB/s:%s) %pM\n",
4870 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4874 e_info("Intel(R) PRO/%s Network Connection\n",
4875 (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4876 e1000e_read_pba_num(hw, &pba_num);
4877 e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4878 hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4881 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4883 struct e1000_hw *hw = &adapter->hw;
4887 if (hw->mac.type != e1000_82573)
4890 ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4891 if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
4892 /* Deep Smart Power Down (DSPD) */
4893 dev_warn(&adapter->pdev->dev,
4894 "Warning: detected DSPD enabled in EEPROM\n");
4897 ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4898 if (!ret_val && (le16_to_cpu(buf) & (3 << 2))) {
4900 dev_warn(&adapter->pdev->dev,
4901 "Warning: detected ASPM enabled in EEPROM\n");
4905 static const struct net_device_ops e1000e_netdev_ops = {
4906 .ndo_open = e1000_open,
4907 .ndo_stop = e1000_close,
4908 .ndo_start_xmit = e1000_xmit_frame,
4909 .ndo_get_stats = e1000_get_stats,
4910 .ndo_set_multicast_list = e1000_set_multi,
4911 .ndo_set_mac_address = e1000_set_mac,
4912 .ndo_change_mtu = e1000_change_mtu,
4913 .ndo_do_ioctl = e1000_ioctl,
4914 .ndo_tx_timeout = e1000_tx_timeout,
4915 .ndo_validate_addr = eth_validate_addr,
4917 .ndo_vlan_rx_register = e1000_vlan_rx_register,
4918 .ndo_vlan_rx_add_vid = e1000_vlan_rx_add_vid,
4919 .ndo_vlan_rx_kill_vid = e1000_vlan_rx_kill_vid,
4920 #ifdef CONFIG_NET_POLL_CONTROLLER
4921 .ndo_poll_controller = e1000_netpoll,
4926 * e1000_probe - Device Initialization Routine
4927 * @pdev: PCI device information struct
4928 * @ent: entry in e1000_pci_tbl
4930 * Returns 0 on success, negative on failure
4932 * e1000_probe initializes an adapter identified by a pci_dev structure.
4933 * The OS initialization, configuring of the adapter private structure,
4934 * and a hardware reset occur.
4936 static int __devinit e1000_probe(struct pci_dev *pdev,
4937 const struct pci_device_id *ent)
4939 struct net_device *netdev;
4940 struct e1000_adapter *adapter;
4941 struct e1000_hw *hw;
4942 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4943 resource_size_t mmio_start, mmio_len;
4944 resource_size_t flash_start, flash_len;
4946 static int cards_found;
4947 int i, err, pci_using_dac;
4948 u16 eeprom_data = 0;
4949 u16 eeprom_apme_mask = E1000_EEPROM_APME;
4951 e1000e_disable_l1aspm(pdev);
4953 err = pci_enable_device_mem(pdev);
4958 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
4960 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
4964 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
4966 err = pci_set_consistent_dma_mask(pdev,
4969 dev_err(&pdev->dev, "No usable DMA "
4970 "configuration, aborting\n");
4976 err = pci_request_selected_regions_exclusive(pdev,
4977 pci_select_bars(pdev, IORESOURCE_MEM),
4978 e1000e_driver_name);
4982 /* AER (Advanced Error Reporting) hooks */
4983 pci_enable_pcie_error_reporting(pdev);
4985 pci_set_master(pdev);
4986 /* PCI config space info */
4987 err = pci_save_state(pdev);
4989 goto err_alloc_etherdev;
4992 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4994 goto err_alloc_etherdev;
4996 SET_NETDEV_DEV(netdev, &pdev->dev);
4998 pci_set_drvdata(pdev, netdev);
4999 adapter = netdev_priv(netdev);
5001 adapter->netdev = netdev;
5002 adapter->pdev = pdev;
5004 adapter->pba = ei->pba;
5005 adapter->flags = ei->flags;
5006 adapter->flags2 = ei->flags2;
5007 adapter->hw.adapter = adapter;
5008 adapter->hw.mac.type = ei->mac;
5009 adapter->max_hw_frame_size = ei->max_hw_frame_size;
5010 adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5012 mmio_start = pci_resource_start(pdev, 0);
5013 mmio_len = pci_resource_len(pdev, 0);
5016 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5017 if (!adapter->hw.hw_addr)
5020 if ((adapter->flags & FLAG_HAS_FLASH) &&
5021 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5022 flash_start = pci_resource_start(pdev, 1);
5023 flash_len = pci_resource_len(pdev, 1);
5024 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5025 if (!adapter->hw.flash_address)
5029 /* construct the net_device struct */
5030 netdev->netdev_ops = &e1000e_netdev_ops;
5031 e1000e_set_ethtool_ops(netdev);
5032 netdev->watchdog_timeo = 5 * HZ;
5033 netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5034 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5036 netdev->mem_start = mmio_start;
5037 netdev->mem_end = mmio_start + mmio_len;
5039 adapter->bd_number = cards_found++;
5041 e1000e_check_options(adapter);
5043 /* setup adapter struct */
5044 err = e1000_sw_init(adapter);
5050 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5051 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5052 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5054 err = ei->get_variants(adapter);
5058 if ((adapter->flags & FLAG_IS_ICH) &&
5059 (adapter->flags & FLAG_READ_ONLY_NVM))
5060 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5062 hw->mac.ops.get_bus_info(&adapter->hw);
5064 adapter->hw.phy.autoneg_wait_to_complete = 0;
5066 /* Copper options */
5067 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5068 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5069 adapter->hw.phy.disable_polarity_correction = 0;
5070 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5073 if (e1000_check_reset_block(&adapter->hw))
5074 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5076 netdev->features = NETIF_F_SG |
5078 NETIF_F_HW_VLAN_TX |
5081 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5082 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5084 netdev->features |= NETIF_F_TSO;
5085 netdev->features |= NETIF_F_TSO6;
5087 netdev->vlan_features |= NETIF_F_TSO;
5088 netdev->vlan_features |= NETIF_F_TSO6;
5089 netdev->vlan_features |= NETIF_F_HW_CSUM;
5090 netdev->vlan_features |= NETIF_F_SG;
5093 netdev->features |= NETIF_F_HIGHDMA;
5095 if (e1000e_enable_mng_pass_thru(&adapter->hw))
5096 adapter->flags |= FLAG_MNG_PT_ENABLED;
5099 * before reading the NVM, reset the controller to
5100 * put the device in a known good starting state
5102 adapter->hw.mac.ops.reset_hw(&adapter->hw);
5105 * systems with ASPM and others may see the checksum fail on the first
5106 * attempt. Let's give it a few tries
5109 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5112 e_err("The NVM Checksum Is Not Valid\n");
5118 e1000_eeprom_checks(adapter);
5120 /* copy the MAC address out of the NVM */
5121 if (e1000e_read_mac_addr(&adapter->hw))
5122 e_err("NVM Read Error while reading MAC address\n");
5124 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5125 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5127 if (!is_valid_ether_addr(netdev->perm_addr)) {
5128 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5133 init_timer(&adapter->watchdog_timer);
5134 adapter->watchdog_timer.function = &e1000_watchdog;
5135 adapter->watchdog_timer.data = (unsigned long) adapter;
5137 init_timer(&adapter->phy_info_timer);
5138 adapter->phy_info_timer.function = &e1000_update_phy_info;
5139 adapter->phy_info_timer.data = (unsigned long) adapter;
5141 INIT_WORK(&adapter->reset_task, e1000_reset_task);
5142 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5143 INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5144 INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5145 INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5147 /* Initialize link parameters. User can change them with ethtool */
5148 adapter->hw.mac.autoneg = 1;
5149 adapter->fc_autoneg = 1;
5150 adapter->hw.fc.requested_mode = e1000_fc_default;
5151 adapter->hw.fc.current_mode = e1000_fc_default;
5152 adapter->hw.phy.autoneg_advertised = 0x2f;
5154 /* ring size defaults */
5155 adapter->rx_ring->count = 256;
5156 adapter->tx_ring->count = 256;
5159 * Initial Wake on LAN setting - If APM wake is enabled in
5160 * the EEPROM, enable the ACPI Magic Packet filter
5162 if (adapter->flags & FLAG_APME_IN_WUC) {
5163 /* APME bit in EEPROM is mapped to WUC.APME */
5164 eeprom_data = er32(WUC);
5165 eeprom_apme_mask = E1000_WUC_APME;
5166 if (eeprom_data & E1000_WUC_PHY_WAKE)
5167 adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5168 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5169 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5170 (adapter->hw.bus.func == 1))
5171 e1000_read_nvm(&adapter->hw,
5172 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5174 e1000_read_nvm(&adapter->hw,
5175 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5178 /* fetch WoL from EEPROM */
5179 if (eeprom_data & eeprom_apme_mask)
5180 adapter->eeprom_wol |= E1000_WUFC_MAG;
5183 * now that we have the eeprom settings, apply the special cases
5184 * where the eeprom may be wrong or the board simply won't support
5185 * wake on lan on a particular port
5187 if (!(adapter->flags & FLAG_HAS_WOL))
5188 adapter->eeprom_wol = 0;
5190 /* initialize the wol settings based on the eeprom settings */
5191 adapter->wol = adapter->eeprom_wol;
5192 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5194 /* save off EEPROM version number */
5195 e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5197 /* reset the hardware with the new settings */
5198 e1000e_reset(adapter);
5201 * If the controller has AMT, do not set DRV_LOAD until the interface
5202 * is up. For all other cases, let the f/w know that the h/w is now
5203 * under the control of the driver.
5205 if (!(adapter->flags & FLAG_HAS_AMT))
5206 e1000_get_hw_control(adapter);
5208 strcpy(netdev->name, "eth%d");
5209 err = register_netdev(netdev);
5213 /* carrier off reporting is important to ethtool even BEFORE open */
5214 netif_carrier_off(netdev);
5216 e1000_print_device_info(adapter);
5221 if (!(adapter->flags & FLAG_HAS_AMT))
5222 e1000_release_hw_control(adapter);
5224 if (!e1000_check_reset_block(&adapter->hw))
5225 e1000_phy_hw_reset(&adapter->hw);
5228 kfree(adapter->tx_ring);
5229 kfree(adapter->rx_ring);
5231 if (adapter->hw.flash_address)
5232 iounmap(adapter->hw.flash_address);
5233 e1000e_reset_interrupt_capability(adapter);
5235 iounmap(adapter->hw.hw_addr);
5237 free_netdev(netdev);
5239 pci_release_selected_regions(pdev,
5240 pci_select_bars(pdev, IORESOURCE_MEM));
5243 pci_disable_device(pdev);
5248 * e1000_remove - Device Removal Routine
5249 * @pdev: PCI device information struct
5251 * e1000_remove is called by the PCI subsystem to alert the driver
5252 * that it should release a PCI device. The could be caused by a
5253 * Hot-Plug event, or because the driver is going to be removed from
5256 static void __devexit e1000_remove(struct pci_dev *pdev)
5258 struct net_device *netdev = pci_get_drvdata(pdev);
5259 struct e1000_adapter *adapter = netdev_priv(netdev);
5262 * flush_scheduled work may reschedule our watchdog task, so
5263 * explicitly disable watchdog tasks from being rescheduled
5265 set_bit(__E1000_DOWN, &adapter->state);
5266 del_timer_sync(&adapter->watchdog_timer);
5267 del_timer_sync(&adapter->phy_info_timer);
5269 cancel_work_sync(&adapter->reset_task);
5270 cancel_work_sync(&adapter->watchdog_task);
5271 cancel_work_sync(&adapter->downshift_task);
5272 cancel_work_sync(&adapter->update_phy_task);
5273 cancel_work_sync(&adapter->print_hang_task);
5274 flush_scheduled_work();
5277 * Release control of h/w to f/w. If f/w is AMT enabled, this
5278 * would have already happened in close and is redundant.
5280 e1000_release_hw_control(adapter);
5282 unregister_netdev(netdev);
5284 if (!e1000_check_reset_block(&adapter->hw))
5285 e1000_phy_hw_reset(&adapter->hw);
5287 e1000e_reset_interrupt_capability(adapter);
5288 kfree(adapter->tx_ring);
5289 kfree(adapter->rx_ring);
5291 iounmap(adapter->hw.hw_addr);
5292 if (adapter->hw.flash_address)
5293 iounmap(adapter->hw.flash_address);
5294 pci_release_selected_regions(pdev,
5295 pci_select_bars(pdev, IORESOURCE_MEM));
5297 free_netdev(netdev);
5300 pci_disable_pcie_error_reporting(pdev);
5302 pci_disable_device(pdev);
5305 /* PCI Error Recovery (ERS) */
5306 static struct pci_error_handlers e1000_err_handler = {
5307 .error_detected = e1000_io_error_detected,
5308 .slot_reset = e1000_io_slot_reset,
5309 .resume = e1000_io_resume,
5312 static struct pci_device_id e1000_pci_tbl[] = {
5313 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5314 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5315 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5316 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5317 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5318 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5319 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5320 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5321 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5323 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5324 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5325 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5326 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5328 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5329 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5330 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5332 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5333 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
5334 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
5336 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5337 board_80003es2lan },
5338 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5339 board_80003es2lan },
5340 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5341 board_80003es2lan },
5342 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5343 board_80003es2lan },
5345 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5346 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5347 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5348 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5349 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5350 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5351 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5353 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5354 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5355 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5356 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5357 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5358 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5359 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5360 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5361 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5363 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5364 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5365 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5367 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5368 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5370 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
5371 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
5372 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
5373 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
5375 { } /* terminate list */
5377 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5379 /* PCI Device API Driver */
5380 static struct pci_driver e1000_driver = {
5381 .name = e1000e_driver_name,
5382 .id_table = e1000_pci_tbl,
5383 .probe = e1000_probe,
5384 .remove = __devexit_p(e1000_remove),
5386 /* Power Management Hooks */
5387 .suspend = e1000_suspend,
5388 .resume = e1000_resume,
5390 .shutdown = e1000_shutdown,
5391 .err_handler = &e1000_err_handler
5395 * e1000_init_module - Driver Registration Routine
5397 * e1000_init_module is the first routine called when the driver is
5398 * loaded. All it does is register with the PCI subsystem.
5400 static int __init e1000_init_module(void)
5403 printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
5404 e1000e_driver_name, e1000e_driver_version);
5405 printk(KERN_INFO "%s: Copyright (c) 1999 - 2009 Intel Corporation.\n",
5406 e1000e_driver_name);
5407 ret = pci_register_driver(&e1000_driver);
5411 module_init(e1000_init_module);
5414 * e1000_exit_module - Driver Exit Cleanup Routine
5416 * e1000_exit_module is called just before the driver is removed
5419 static void __exit e1000_exit_module(void)
5421 pci_unregister_driver(&e1000_driver);
5423 module_exit(e1000_exit_module);
5426 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5427 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5428 MODULE_LICENSE("GPL");
5429 MODULE_VERSION(DRV_VERSION);