1 /*******************************************************************************
3 Intel PRO/1000 Linux driver
4 Copyright(c) 1999 - 2007 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>
49 #define DRV_VERSION "0.2.0"
50 char e1000e_driver_name[] = "e1000e";
51 const char e1000e_driver_version[] = DRV_VERSION;
53 static const struct e1000_info *e1000_info_tbl[] = {
54 [board_82571] = &e1000_82571_info,
55 [board_82572] = &e1000_82572_info,
56 [board_82573] = &e1000_82573_info,
57 [board_80003es2lan] = &e1000_es2_info,
58 [board_ich8lan] = &e1000_ich8_info,
59 [board_ich9lan] = &e1000_ich9_info,
64 * e1000_get_hw_dev_name - return device name string
65 * used by hardware layer to print debugging information
67 char *e1000e_get_hw_dev_name(struct e1000_hw *hw)
69 return hw->adapter->netdev->name;
74 * e1000_desc_unused - calculate if we have unused descriptors
76 static int e1000_desc_unused(struct e1000_ring *ring)
78 if (ring->next_to_clean > ring->next_to_use)
79 return ring->next_to_clean - ring->next_to_use - 1;
81 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
85 * e1000_receive_skb - helper function to handle rx indications
86 * @adapter: board private structure
87 * @status: descriptor status field as written by hardware
88 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
89 * @skb: pointer to sk_buff to be indicated to stack
91 static void e1000_receive_skb(struct e1000_adapter *adapter,
92 struct net_device *netdev,
94 u8 status, __le16 vlan)
96 skb->protocol = eth_type_trans(skb, netdev);
98 if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
99 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
101 E1000_RXD_SPC_VLAN_MASK);
103 netif_receive_skb(skb);
105 netdev->last_rx = jiffies;
109 * e1000_rx_checksum - Receive Checksum Offload for 82543
110 * @adapter: board private structure
111 * @status_err: receive descriptor status and error fields
112 * @csum: receive descriptor csum field
113 * @sk_buff: socket buffer with received data
115 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
116 u32 csum, struct sk_buff *skb)
118 u16 status = (u16)status_err;
119 u8 errors = (u8)(status_err >> 24);
120 skb->ip_summed = CHECKSUM_NONE;
122 /* Ignore Checksum bit is set */
123 if (status & E1000_RXD_STAT_IXSM)
125 /* TCP/UDP checksum error bit is set */
126 if (errors & E1000_RXD_ERR_TCPE) {
127 /* let the stack verify checksum errors */
128 adapter->hw_csum_err++;
132 /* TCP/UDP Checksum has not been calculated */
133 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
136 /* It must be a TCP or UDP packet with a valid checksum */
137 if (status & E1000_RXD_STAT_TCPCS) {
138 /* TCP checksum is good */
139 skb->ip_summed = CHECKSUM_UNNECESSARY;
141 /* IP fragment with UDP payload */
142 /* Hardware complements the payload checksum, so we undo it
143 * and then put the value in host order for further stack use.
145 __sum16 sum = (__force __sum16)htons(csum);
146 skb->csum = csum_unfold(~sum);
147 skb->ip_summed = CHECKSUM_COMPLETE;
149 adapter->hw_csum_good++;
153 * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
154 * @adapter: address of board private structure
156 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
159 struct net_device *netdev = adapter->netdev;
160 struct pci_dev *pdev = adapter->pdev;
161 struct e1000_ring *rx_ring = adapter->rx_ring;
162 struct e1000_rx_desc *rx_desc;
163 struct e1000_buffer *buffer_info;
166 unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
168 i = rx_ring->next_to_use;
169 buffer_info = &rx_ring->buffer_info[i];
171 while (cleaned_count--) {
172 skb = buffer_info->skb;
178 skb = netdev_alloc_skb(netdev, bufsz);
180 /* Better luck next round */
181 adapter->alloc_rx_buff_failed++;
185 /* Make buffer alignment 2 beyond a 16 byte boundary
186 * this will result in a 16 byte aligned IP header after
187 * the 14 byte MAC header is removed
189 skb_reserve(skb, NET_IP_ALIGN);
191 buffer_info->skb = skb;
193 buffer_info->dma = pci_map_single(pdev, skb->data,
194 adapter->rx_buffer_len,
196 if (pci_dma_mapping_error(buffer_info->dma)) {
197 dev_err(&pdev->dev, "RX DMA map failed\n");
198 adapter->rx_dma_failed++;
202 rx_desc = E1000_RX_DESC(*rx_ring, i);
203 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
206 if (i == rx_ring->count)
208 buffer_info = &rx_ring->buffer_info[i];
211 if (rx_ring->next_to_use != i) {
212 rx_ring->next_to_use = i;
214 i = (rx_ring->count - 1);
216 /* Force memory writes to complete before letting h/w
217 * know there are new descriptors to fetch. (Only
218 * applicable for weak-ordered memory model archs,
221 writel(i, adapter->hw.hw_addr + rx_ring->tail);
226 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
227 * @adapter: address of board private structure
229 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
232 struct net_device *netdev = adapter->netdev;
233 struct pci_dev *pdev = adapter->pdev;
234 union e1000_rx_desc_packet_split *rx_desc;
235 struct e1000_ring *rx_ring = adapter->rx_ring;
236 struct e1000_buffer *buffer_info;
237 struct e1000_ps_page *ps_page;
241 i = rx_ring->next_to_use;
242 buffer_info = &rx_ring->buffer_info[i];
244 while (cleaned_count--) {
245 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
247 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
248 ps_page = &buffer_info->ps_pages[j];
249 if (j >= adapter->rx_ps_pages) {
250 /* all unused desc entries get hw null ptr */
251 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
254 if (!ps_page->page) {
255 ps_page->page = alloc_page(GFP_ATOMIC);
256 if (!ps_page->page) {
257 adapter->alloc_rx_buff_failed++;
260 ps_page->dma = pci_map_page(pdev,
264 if (pci_dma_mapping_error(ps_page->dma)) {
265 dev_err(&adapter->pdev->dev,
266 "RX DMA page map failed\n");
267 adapter->rx_dma_failed++;
272 * Refresh the desc even if buffer_addrs
273 * didn't change because each write-back
276 rx_desc->read.buffer_addr[j+1] =
277 cpu_to_le64(ps_page->dma);
280 skb = netdev_alloc_skb(netdev,
281 adapter->rx_ps_bsize0 + NET_IP_ALIGN);
284 adapter->alloc_rx_buff_failed++;
288 /* Make buffer alignment 2 beyond a 16 byte boundary
289 * this will result in a 16 byte aligned IP header after
290 * the 14 byte MAC header is removed
292 skb_reserve(skb, NET_IP_ALIGN);
294 buffer_info->skb = skb;
295 buffer_info->dma = pci_map_single(pdev, skb->data,
296 adapter->rx_ps_bsize0,
298 if (pci_dma_mapping_error(buffer_info->dma)) {
299 dev_err(&pdev->dev, "RX DMA map failed\n");
300 adapter->rx_dma_failed++;
302 dev_kfree_skb_any(skb);
303 buffer_info->skb = NULL;
307 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
310 if (i == rx_ring->count)
312 buffer_info = &rx_ring->buffer_info[i];
316 if (rx_ring->next_to_use != i) {
317 rx_ring->next_to_use = i;
320 i = (rx_ring->count - 1);
322 /* Force memory writes to complete before letting h/w
323 * know there are new descriptors to fetch. (Only
324 * applicable for weak-ordered memory model archs,
327 /* Hardware increments by 16 bytes, but packet split
328 * descriptors are 32 bytes...so we increment tail
331 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
336 * e1000_clean_rx_irq - Send received data up the network stack; legacy
337 * @adapter: board private structure
339 * the return value indicates whether actual cleaning was done, there
340 * is no guarantee that everything was cleaned
342 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
343 int *work_done, int work_to_do)
345 struct net_device *netdev = adapter->netdev;
346 struct pci_dev *pdev = adapter->pdev;
347 struct e1000_ring *rx_ring = adapter->rx_ring;
348 struct e1000_rx_desc *rx_desc, *next_rxd;
349 struct e1000_buffer *buffer_info, *next_buffer;
352 int cleaned_count = 0;
354 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
356 i = rx_ring->next_to_clean;
357 rx_desc = E1000_RX_DESC(*rx_ring, i);
358 buffer_info = &rx_ring->buffer_info[i];
360 while (rx_desc->status & E1000_RXD_STAT_DD) {
364 if (*work_done >= work_to_do)
368 status = rx_desc->status;
369 skb = buffer_info->skb;
370 buffer_info->skb = NULL;
372 prefetch(skb->data - NET_IP_ALIGN);
375 if (i == rx_ring->count)
377 next_rxd = E1000_RX_DESC(*rx_ring, i);
380 next_buffer = &rx_ring->buffer_info[i];
384 pci_unmap_single(pdev,
386 adapter->rx_buffer_len,
388 buffer_info->dma = 0;
390 length = le16_to_cpu(rx_desc->length);
392 /* !EOP means multiple descriptors were used to store a single
393 * packet, also make sure the frame isn't just CRC only */
394 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
395 /* All receives must fit into a single buffer */
396 ndev_dbg(netdev, "%s: Receive packet consumed "
397 "multiple buffers\n", netdev->name);
399 buffer_info->skb = skb;
403 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
405 buffer_info->skb = skb;
409 total_rx_bytes += length;
412 /* code added for copybreak, this should improve
413 * performance for small packets with large amounts
414 * of reassembly being done in the stack */
415 if (length < copybreak) {
416 struct sk_buff *new_skb =
417 netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
419 skb_reserve(new_skb, NET_IP_ALIGN);
420 memcpy(new_skb->data - NET_IP_ALIGN,
421 skb->data - NET_IP_ALIGN,
422 length + NET_IP_ALIGN);
423 /* save the skb in buffer_info as good */
424 buffer_info->skb = skb;
427 /* else just continue with the old one */
429 /* end copybreak code */
430 skb_put(skb, length);
432 /* Receive Checksum Offload */
433 e1000_rx_checksum(adapter,
435 ((u32)(rx_desc->errors) << 24),
436 le16_to_cpu(rx_desc->csum), skb);
438 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
443 /* return some buffers to hardware, one at a time is too slow */
444 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
445 adapter->alloc_rx_buf(adapter, cleaned_count);
449 /* use prefetched values */
451 buffer_info = next_buffer;
453 rx_ring->next_to_clean = i;
455 cleaned_count = e1000_desc_unused(rx_ring);
457 adapter->alloc_rx_buf(adapter, cleaned_count);
459 adapter->total_rx_packets += total_rx_packets;
460 adapter->total_rx_bytes += total_rx_bytes;
461 adapter->net_stats.rx_packets += total_rx_packets;
462 adapter->net_stats.rx_bytes += total_rx_bytes;
466 static void e1000_put_txbuf(struct e1000_adapter *adapter,
467 struct e1000_buffer *buffer_info)
469 if (buffer_info->dma) {
470 pci_unmap_page(adapter->pdev, buffer_info->dma,
471 buffer_info->length, PCI_DMA_TODEVICE);
472 buffer_info->dma = 0;
474 if (buffer_info->skb) {
475 dev_kfree_skb_any(buffer_info->skb);
476 buffer_info->skb = NULL;
480 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
482 struct e1000_ring *tx_ring = adapter->tx_ring;
483 unsigned int i = tx_ring->next_to_clean;
484 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
485 struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
486 struct net_device *netdev = adapter->netdev;
488 /* detected Tx unit hang */
490 "Detected Tx Unit Hang:\n"
493 " next_to_use <%x>\n"
494 " next_to_clean <%x>\n"
495 "buffer_info[next_to_clean]:\n"
496 " time_stamp <%lx>\n"
497 " next_to_watch <%x>\n"
499 " next_to_watch.status <%x>\n",
500 readl(adapter->hw.hw_addr + tx_ring->head),
501 readl(adapter->hw.hw_addr + tx_ring->tail),
502 tx_ring->next_to_use,
503 tx_ring->next_to_clean,
504 tx_ring->buffer_info[eop].time_stamp,
507 eop_desc->upper.fields.status);
511 * e1000_clean_tx_irq - Reclaim resources after transmit completes
512 * @adapter: board private structure
514 * the return value indicates whether actual cleaning was done, there
515 * is no guarantee that everything was cleaned
517 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
519 struct net_device *netdev = adapter->netdev;
520 struct e1000_hw *hw = &adapter->hw;
521 struct e1000_ring *tx_ring = adapter->tx_ring;
522 struct e1000_tx_desc *tx_desc, *eop_desc;
523 struct e1000_buffer *buffer_info;
525 unsigned int count = 0;
527 unsigned int total_tx_bytes = 0, total_tx_packets = 0;
529 i = tx_ring->next_to_clean;
530 eop = tx_ring->buffer_info[i].next_to_watch;
531 eop_desc = E1000_TX_DESC(*tx_ring, eop);
533 while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
534 for (cleaned = 0; !cleaned; ) {
535 tx_desc = E1000_TX_DESC(*tx_ring, i);
536 buffer_info = &tx_ring->buffer_info[i];
537 cleaned = (i == eop);
540 struct sk_buff *skb = buffer_info->skb;
541 unsigned int segs, bytecount;
542 segs = skb_shinfo(skb)->gso_segs ?: 1;
543 /* multiply data chunks by size of headers */
544 bytecount = ((segs - 1) * skb_headlen(skb)) +
546 total_tx_packets += segs;
547 total_tx_bytes += bytecount;
550 e1000_put_txbuf(adapter, buffer_info);
551 tx_desc->upper.data = 0;
554 if (i == tx_ring->count)
558 eop = tx_ring->buffer_info[i].next_to_watch;
559 eop_desc = E1000_TX_DESC(*tx_ring, eop);
560 #define E1000_TX_WEIGHT 64
561 /* weight of a sort for tx, to avoid endless transmit cleanup */
562 if (count++ == E1000_TX_WEIGHT)
566 tx_ring->next_to_clean = i;
568 #define TX_WAKE_THRESHOLD 32
569 if (cleaned && netif_carrier_ok(netdev) &&
570 e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
571 /* Make sure that anybody stopping the queue after this
572 * sees the new next_to_clean.
576 if (netif_queue_stopped(netdev) &&
577 !(test_bit(__E1000_DOWN, &adapter->state))) {
578 netif_wake_queue(netdev);
579 ++adapter->restart_queue;
583 if (adapter->detect_tx_hung) {
584 /* Detect a transmit hang in hardware, this serializes the
585 * check with the clearing of time_stamp and movement of i */
586 adapter->detect_tx_hung = 0;
587 if (tx_ring->buffer_info[eop].dma &&
588 time_after(jiffies, tx_ring->buffer_info[eop].time_stamp
589 + (adapter->tx_timeout_factor * HZ))
591 E1000_STATUS_TXOFF)) {
592 e1000_print_tx_hang(adapter);
593 netif_stop_queue(netdev);
596 adapter->total_tx_bytes += total_tx_bytes;
597 adapter->total_tx_packets += total_tx_packets;
598 adapter->net_stats.tx_packets += total_tx_packets;
599 adapter->net_stats.tx_bytes += total_tx_bytes;
604 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
605 * @adapter: board private structure
607 * the return value indicates whether actual cleaning was done, there
608 * is no guarantee that everything was cleaned
610 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
611 int *work_done, int work_to_do)
613 union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
614 struct net_device *netdev = adapter->netdev;
615 struct pci_dev *pdev = adapter->pdev;
616 struct e1000_ring *rx_ring = adapter->rx_ring;
617 struct e1000_buffer *buffer_info, *next_buffer;
618 struct e1000_ps_page *ps_page;
622 int cleaned_count = 0;
624 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
626 i = rx_ring->next_to_clean;
627 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
628 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
629 buffer_info = &rx_ring->buffer_info[i];
631 while (staterr & E1000_RXD_STAT_DD) {
632 if (*work_done >= work_to_do)
635 skb = buffer_info->skb;
637 /* in the packet split case this is header only */
638 prefetch(skb->data - NET_IP_ALIGN);
641 if (i == rx_ring->count)
643 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
646 next_buffer = &rx_ring->buffer_info[i];
650 pci_unmap_single(pdev, buffer_info->dma,
651 adapter->rx_ps_bsize0,
653 buffer_info->dma = 0;
655 if (!(staterr & E1000_RXD_STAT_EOP)) {
656 ndev_dbg(netdev, "%s: Packet Split buffers didn't pick "
657 "up the full packet\n", netdev->name);
658 dev_kfree_skb_irq(skb);
662 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
663 dev_kfree_skb_irq(skb);
667 length = le16_to_cpu(rx_desc->wb.middle.length0);
670 ndev_dbg(netdev, "%s: Last part of the packet spanning"
671 " multiple descriptors\n", netdev->name);
672 dev_kfree_skb_irq(skb);
677 skb_put(skb, length);
680 /* this looks ugly, but it seems compiler issues make it
681 more efficient than reusing j */
682 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
684 /* page alloc/put takes too long and effects small packet
685 * throughput, so unsplit small packets and save the alloc/put*/
686 if (l1 && (l1 <= copybreak) &&
687 ((length + l1) <= adapter->rx_ps_bsize0)) {
690 ps_page = &buffer_info->ps_pages[0];
692 /* there is no documentation about how to call
693 * kmap_atomic, so we can't hold the mapping
695 pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
696 PAGE_SIZE, PCI_DMA_FROMDEVICE);
697 vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
698 memcpy(skb_tail_pointer(skb), vaddr, l1);
699 kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
700 pci_dma_sync_single_for_device(pdev, ps_page->dma,
701 PAGE_SIZE, PCI_DMA_FROMDEVICE);
708 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
709 length = le16_to_cpu(rx_desc->wb.upper.length[j]);
713 ps_page = &buffer_info->ps_pages[j];
714 pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
717 skb_fill_page_desc(skb, j, ps_page->page, 0, length);
718 ps_page->page = NULL;
720 skb->data_len += length;
721 skb->truesize += length;
725 total_rx_bytes += skb->len;
728 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
729 rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
731 if (rx_desc->wb.upper.header_status &
732 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
733 adapter->rx_hdr_split++;
735 e1000_receive_skb(adapter, netdev, skb,
736 staterr, rx_desc->wb.middle.vlan);
739 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
740 buffer_info->skb = NULL;
742 /* return some buffers to hardware, one at a time is too slow */
743 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
744 adapter->alloc_rx_buf(adapter, cleaned_count);
748 /* use prefetched values */
750 buffer_info = next_buffer;
752 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
754 rx_ring->next_to_clean = i;
756 cleaned_count = e1000_desc_unused(rx_ring);
758 adapter->alloc_rx_buf(adapter, cleaned_count);
760 adapter->total_rx_packets += total_rx_packets;
761 adapter->total_rx_bytes += total_rx_bytes;
762 adapter->net_stats.rx_packets += total_rx_packets;
763 adapter->net_stats.rx_bytes += total_rx_bytes;
768 * e1000_clean_rx_ring - Free Rx Buffers per Queue
769 * @adapter: board private structure
771 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
773 struct e1000_ring *rx_ring = adapter->rx_ring;
774 struct e1000_buffer *buffer_info;
775 struct e1000_ps_page *ps_page;
776 struct pci_dev *pdev = adapter->pdev;
779 /* Free all the Rx ring sk_buffs */
780 for (i = 0; i < rx_ring->count; i++) {
781 buffer_info = &rx_ring->buffer_info[i];
782 if (buffer_info->dma) {
783 if (adapter->clean_rx == e1000_clean_rx_irq)
784 pci_unmap_single(pdev, buffer_info->dma,
785 adapter->rx_buffer_len,
787 else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
788 pci_unmap_single(pdev, buffer_info->dma,
789 adapter->rx_ps_bsize0,
791 buffer_info->dma = 0;
794 if (buffer_info->skb) {
795 dev_kfree_skb(buffer_info->skb);
796 buffer_info->skb = NULL;
799 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
800 ps_page = &buffer_info->ps_pages[j];
803 pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
806 put_page(ps_page->page);
807 ps_page->page = NULL;
811 /* there also may be some cached data from a chained receive */
812 if (rx_ring->rx_skb_top) {
813 dev_kfree_skb(rx_ring->rx_skb_top);
814 rx_ring->rx_skb_top = NULL;
817 /* Zero out the descriptor ring */
818 memset(rx_ring->desc, 0, rx_ring->size);
820 rx_ring->next_to_clean = 0;
821 rx_ring->next_to_use = 0;
823 writel(0, adapter->hw.hw_addr + rx_ring->head);
824 writel(0, adapter->hw.hw_addr + rx_ring->tail);
828 * e1000_intr_msi - Interrupt Handler
829 * @irq: interrupt number
830 * @data: pointer to a network interface device structure
832 static irqreturn_t e1000_intr_msi(int irq, void *data)
834 struct net_device *netdev = data;
835 struct e1000_adapter *adapter = netdev_priv(netdev);
836 struct e1000_hw *hw = &adapter->hw;
839 /* read ICR disables interrupts using IAM, so keep up with our
840 * enable/disable accounting */
841 atomic_inc(&adapter->irq_sem);
843 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
844 hw->mac.get_link_status = 1;
845 /* ICH8 workaround-- Call gig speed drop workaround on cable
846 * disconnect (LSC) before accessing any PHY registers */
847 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
848 (!(er32(STATUS) & E1000_STATUS_LU)))
849 e1000e_gig_downshift_workaround_ich8lan(hw);
851 /* 80003ES2LAN workaround-- For packet buffer work-around on
852 * link down event; disable receives here in the ISR and reset
853 * adapter in watchdog */
854 if (netif_carrier_ok(netdev) &&
855 adapter->flags & FLAG_RX_NEEDS_RESTART) {
856 /* disable receives */
857 u32 rctl = er32(RCTL);
858 ew32(RCTL, rctl & ~E1000_RCTL_EN);
860 /* guard against interrupt when we're going down */
861 if (!test_bit(__E1000_DOWN, &adapter->state))
862 mod_timer(&adapter->watchdog_timer, jiffies + 1);
865 if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
866 adapter->total_tx_bytes = 0;
867 adapter->total_tx_packets = 0;
868 adapter->total_rx_bytes = 0;
869 adapter->total_rx_packets = 0;
870 __netif_rx_schedule(netdev, &adapter->napi);
872 atomic_dec(&adapter->irq_sem);
879 * e1000_intr - Interrupt Handler
880 * @irq: interrupt number
881 * @data: pointer to a network interface device structure
883 static irqreturn_t e1000_intr(int irq, void *data)
885 struct net_device *netdev = data;
886 struct e1000_adapter *adapter = netdev_priv(netdev);
887 struct e1000_hw *hw = &adapter->hw;
889 u32 rctl, icr = er32(ICR);
891 return IRQ_NONE; /* Not our interrupt */
893 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
894 * not set, then the adapter didn't send an interrupt */
895 if (!(icr & E1000_ICR_INT_ASSERTED))
898 /* Interrupt Auto-Mask...upon reading ICR,
899 * interrupts are masked. No need for the
900 * IMC write, but it does mean we should
901 * account for it ASAP. */
902 atomic_inc(&adapter->irq_sem);
904 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
905 hw->mac.get_link_status = 1;
906 /* ICH8 workaround-- Call gig speed drop workaround on cable
907 * disconnect (LSC) before accessing any PHY registers */
908 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
909 (!(er32(STATUS) & E1000_STATUS_LU)))
910 e1000e_gig_downshift_workaround_ich8lan(hw);
912 /* 80003ES2LAN workaround--
913 * For packet buffer work-around on link down event;
914 * disable receives here in the ISR and
915 * reset adapter in watchdog
917 if (netif_carrier_ok(netdev) &&
918 (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
919 /* disable receives */
921 ew32(RCTL, rctl & ~E1000_RCTL_EN);
923 /* guard against interrupt when we're going down */
924 if (!test_bit(__E1000_DOWN, &adapter->state))
925 mod_timer(&adapter->watchdog_timer, jiffies + 1);
928 if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
929 adapter->total_tx_bytes = 0;
930 adapter->total_tx_packets = 0;
931 adapter->total_rx_bytes = 0;
932 adapter->total_rx_packets = 0;
933 __netif_rx_schedule(netdev, &adapter->napi);
935 atomic_dec(&adapter->irq_sem);
941 static int e1000_request_irq(struct e1000_adapter *adapter)
943 struct net_device *netdev = adapter->netdev;
944 irq_handler_t handler = e1000_intr;
945 int irq_flags = IRQF_SHARED;
948 err = pci_enable_msi(adapter->pdev);
951 "Unable to allocate MSI interrupt Error: %d\n", err);
953 adapter->flags |= FLAG_MSI_ENABLED;
954 handler = e1000_intr_msi;
958 err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
961 if (adapter->flags & FLAG_MSI_ENABLED)
962 pci_disable_msi(adapter->pdev);
964 "Unable to allocate interrupt Error: %d\n", err);
970 static void e1000_free_irq(struct e1000_adapter *adapter)
972 struct net_device *netdev = adapter->netdev;
974 free_irq(adapter->pdev->irq, netdev);
975 if (adapter->flags & FLAG_MSI_ENABLED) {
976 pci_disable_msi(adapter->pdev);
977 adapter->flags &= ~FLAG_MSI_ENABLED;
982 * e1000_irq_disable - Mask off interrupt generation on the NIC
984 static void e1000_irq_disable(struct e1000_adapter *adapter)
986 struct e1000_hw *hw = &adapter->hw;
988 atomic_inc(&adapter->irq_sem);
991 synchronize_irq(adapter->pdev->irq);
995 * e1000_irq_enable - Enable default interrupt generation settings
997 static void e1000_irq_enable(struct e1000_adapter *adapter)
999 struct e1000_hw *hw = &adapter->hw;
1001 if (atomic_dec_and_test(&adapter->irq_sem)) {
1002 ew32(IMS, IMS_ENABLE_MASK);
1008 * e1000_get_hw_control - get control of the h/w from f/w
1009 * @adapter: address of board private structure
1011 * e1000_get_hw_control sets {CTRL_EXT|FWSM}:DRV_LOAD bit.
1012 * For ASF and Pass Through versions of f/w this means that
1013 * the driver is loaded. For AMT version (only with 82573)
1014 * of the f/w this means that the network i/f is open.
1016 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1018 struct e1000_hw *hw = &adapter->hw;
1022 /* Let firmware know the driver has taken over */
1023 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1025 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1026 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1027 ctrl_ext = er32(CTRL_EXT);
1029 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1034 * e1000_release_hw_control - release control of the h/w to f/w
1035 * @adapter: address of board private structure
1037 * e1000_release_hw_control resets {CTRL_EXT|FWSM}:DRV_LOAD bit.
1038 * For ASF and Pass Through versions of f/w this means that the
1039 * driver is no longer loaded. For AMT version (only with 82573) i
1040 * of the f/w this means that the network i/f is closed.
1043 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1045 struct e1000_hw *hw = &adapter->hw;
1049 /* Let firmware taken over control of h/w */
1050 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1052 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1053 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1054 ctrl_ext = er32(CTRL_EXT);
1056 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1060 static void e1000_release_manageability(struct e1000_adapter *adapter)
1062 if (adapter->flags & FLAG_MNG_PT_ENABLED) {
1063 struct e1000_hw *hw = &adapter->hw;
1065 u32 manc = er32(MANC);
1067 /* re-enable hardware interception of ARP */
1068 manc |= E1000_MANC_ARP_EN;
1069 manc &= ~E1000_MANC_EN_MNG2HOST;
1071 /* don't explicitly have to mess with MANC2H since
1072 * MANC has an enable disable that gates MANC2H */
1078 * @e1000_alloc_ring - allocate memory for a ring structure
1080 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1081 struct e1000_ring *ring)
1083 struct pci_dev *pdev = adapter->pdev;
1085 ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1094 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1095 * @adapter: board private structure
1097 * Return 0 on success, negative on failure
1099 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1101 struct e1000_ring *tx_ring = adapter->tx_ring;
1102 int err = -ENOMEM, size;
1104 size = sizeof(struct e1000_buffer) * tx_ring->count;
1105 tx_ring->buffer_info = vmalloc(size);
1106 if (!tx_ring->buffer_info)
1108 memset(tx_ring->buffer_info, 0, size);
1110 /* round up to nearest 4K */
1111 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1112 tx_ring->size = ALIGN(tx_ring->size, 4096);
1114 err = e1000_alloc_ring_dma(adapter, tx_ring);
1118 tx_ring->next_to_use = 0;
1119 tx_ring->next_to_clean = 0;
1120 spin_lock_init(&adapter->tx_queue_lock);
1124 vfree(tx_ring->buffer_info);
1125 ndev_err(adapter->netdev,
1126 "Unable to allocate memory for the transmit descriptor ring\n");
1131 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1132 * @adapter: board private structure
1134 * Returns 0 on success, negative on failure
1136 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1138 struct e1000_ring *rx_ring = adapter->rx_ring;
1139 struct e1000_buffer *buffer_info;
1140 int i, size, desc_len, err = -ENOMEM;
1142 size = sizeof(struct e1000_buffer) * rx_ring->count;
1143 rx_ring->buffer_info = vmalloc(size);
1144 if (!rx_ring->buffer_info)
1146 memset(rx_ring->buffer_info, 0, size);
1148 for (i = 0; i < rx_ring->count; i++) {
1149 buffer_info = &rx_ring->buffer_info[i];
1150 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1151 sizeof(struct e1000_ps_page),
1153 if (!buffer_info->ps_pages)
1157 desc_len = sizeof(union e1000_rx_desc_packet_split);
1159 /* Round up to nearest 4K */
1160 rx_ring->size = rx_ring->count * desc_len;
1161 rx_ring->size = ALIGN(rx_ring->size, 4096);
1163 err = e1000_alloc_ring_dma(adapter, rx_ring);
1167 rx_ring->next_to_clean = 0;
1168 rx_ring->next_to_use = 0;
1169 rx_ring->rx_skb_top = NULL;
1174 for (i = 0; i < rx_ring->count; i++) {
1175 buffer_info = &rx_ring->buffer_info[i];
1176 kfree(buffer_info->ps_pages);
1179 vfree(rx_ring->buffer_info);
1180 ndev_err(adapter->netdev,
1181 "Unable to allocate memory for the transmit descriptor ring\n");
1186 * e1000_clean_tx_ring - Free Tx Buffers
1187 * @adapter: board private structure
1189 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1191 struct e1000_ring *tx_ring = adapter->tx_ring;
1192 struct e1000_buffer *buffer_info;
1196 for (i = 0; i < tx_ring->count; i++) {
1197 buffer_info = &tx_ring->buffer_info[i];
1198 e1000_put_txbuf(adapter, buffer_info);
1201 size = sizeof(struct e1000_buffer) * tx_ring->count;
1202 memset(tx_ring->buffer_info, 0, size);
1204 memset(tx_ring->desc, 0, tx_ring->size);
1206 tx_ring->next_to_use = 0;
1207 tx_ring->next_to_clean = 0;
1209 writel(0, adapter->hw.hw_addr + tx_ring->head);
1210 writel(0, adapter->hw.hw_addr + tx_ring->tail);
1214 * e1000e_free_tx_resources - Free Tx Resources per Queue
1215 * @adapter: board private structure
1217 * Free all transmit software resources
1219 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1221 struct pci_dev *pdev = adapter->pdev;
1222 struct e1000_ring *tx_ring = adapter->tx_ring;
1224 e1000_clean_tx_ring(adapter);
1226 vfree(tx_ring->buffer_info);
1227 tx_ring->buffer_info = NULL;
1229 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1231 tx_ring->desc = NULL;
1235 * e1000e_free_rx_resources - Free Rx Resources
1236 * @adapter: board private structure
1238 * Free all receive software resources
1241 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1243 struct pci_dev *pdev = adapter->pdev;
1244 struct e1000_ring *rx_ring = adapter->rx_ring;
1247 e1000_clean_rx_ring(adapter);
1249 for (i = 0; i < rx_ring->count; i++) {
1250 kfree(rx_ring->buffer_info[i].ps_pages);
1253 vfree(rx_ring->buffer_info);
1254 rx_ring->buffer_info = NULL;
1256 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1258 rx_ring->desc = NULL;
1262 * e1000_update_itr - update the dynamic ITR value based on statistics
1263 * Stores a new ITR value based on packets and byte
1264 * counts during the last interrupt. The advantage of per interrupt
1265 * computation is faster updates and more accurate ITR for the current
1266 * traffic pattern. Constants in this function were computed
1267 * based on theoretical maximum wire speed and thresholds were set based
1268 * on testing data as well as attempting to minimize response time
1269 * while increasing bulk throughput.
1270 * this functionality is controlled by the InterruptThrottleRate module
1271 * parameter (see e1000_param.c)
1272 * @adapter: pointer to adapter
1273 * @itr_setting: current adapter->itr
1274 * @packets: the number of packets during this measurement interval
1275 * @bytes: the number of bytes during this measurement interval
1277 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1278 u16 itr_setting, int packets,
1281 unsigned int retval = itr_setting;
1284 goto update_itr_done;
1286 switch (itr_setting) {
1287 case lowest_latency:
1288 /* handle TSO and jumbo frames */
1289 if (bytes/packets > 8000)
1290 retval = bulk_latency;
1291 else if ((packets < 5) && (bytes > 512)) {
1292 retval = low_latency;
1295 case low_latency: /* 50 usec aka 20000 ints/s */
1296 if (bytes > 10000) {
1297 /* this if handles the TSO accounting */
1298 if (bytes/packets > 8000) {
1299 retval = bulk_latency;
1300 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1301 retval = bulk_latency;
1302 } else if ((packets > 35)) {
1303 retval = lowest_latency;
1305 } else if (bytes/packets > 2000) {
1306 retval = bulk_latency;
1307 } else if (packets <= 2 && bytes < 512) {
1308 retval = lowest_latency;
1311 case bulk_latency: /* 250 usec aka 4000 ints/s */
1312 if (bytes > 25000) {
1314 retval = low_latency;
1316 } else if (bytes < 6000) {
1317 retval = low_latency;
1326 static void e1000_set_itr(struct e1000_adapter *adapter)
1328 struct e1000_hw *hw = &adapter->hw;
1330 u32 new_itr = adapter->itr;
1332 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1333 if (adapter->link_speed != SPEED_1000) {
1339 adapter->tx_itr = e1000_update_itr(adapter,
1341 adapter->total_tx_packets,
1342 adapter->total_tx_bytes);
1343 /* conservative mode (itr 3) eliminates the lowest_latency setting */
1344 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1345 adapter->tx_itr = low_latency;
1347 adapter->rx_itr = e1000_update_itr(adapter,
1349 adapter->total_rx_packets,
1350 adapter->total_rx_bytes);
1351 /* conservative mode (itr 3) eliminates the lowest_latency setting */
1352 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1353 adapter->rx_itr = low_latency;
1355 current_itr = max(adapter->rx_itr, adapter->tx_itr);
1357 switch (current_itr) {
1358 /* counts and packets in update_itr are dependent on these numbers */
1359 case lowest_latency:
1363 new_itr = 20000; /* aka hwitr = ~200 */
1373 if (new_itr != adapter->itr) {
1374 /* this attempts to bias the interrupt rate towards Bulk
1375 * by adding intermediate steps when interrupt rate is
1377 new_itr = new_itr > adapter->itr ?
1378 min(adapter->itr + (new_itr >> 2), new_itr) :
1380 adapter->itr = new_itr;
1381 ew32(ITR, 1000000000 / (new_itr * 256));
1386 * e1000_clean - NAPI Rx polling callback
1387 * @adapter: board private structure
1389 static int e1000_clean(struct napi_struct *napi, int budget)
1391 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1392 struct net_device *poll_dev = adapter->netdev;
1393 int tx_cleaned = 0, work_done = 0;
1395 /* Must NOT use netdev_priv macro here. */
1396 adapter = poll_dev->priv;
1398 /* e1000_clean is called per-cpu. This lock protects
1399 * tx_ring from being cleaned by multiple cpus
1400 * simultaneously. A failure obtaining the lock means
1401 * tx_ring is currently being cleaned anyway. */
1402 if (spin_trylock(&adapter->tx_queue_lock)) {
1403 tx_cleaned = e1000_clean_tx_irq(adapter);
1404 spin_unlock(&adapter->tx_queue_lock);
1407 adapter->clean_rx(adapter, &work_done, budget);
1412 /* If budget not fully consumed, exit the polling mode */
1413 if (work_done < budget) {
1414 if (adapter->itr_setting & 3)
1415 e1000_set_itr(adapter);
1416 netif_rx_complete(poll_dev, napi);
1417 e1000_irq_enable(adapter);
1423 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1425 struct e1000_adapter *adapter = netdev_priv(netdev);
1426 struct e1000_hw *hw = &adapter->hw;
1429 /* don't update vlan cookie if already programmed */
1430 if ((adapter->hw.mng_cookie.status &
1431 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1432 (vid == adapter->mng_vlan_id))
1434 /* add VID to filter table */
1435 index = (vid >> 5) & 0x7F;
1436 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1437 vfta |= (1 << (vid & 0x1F));
1438 e1000e_write_vfta(hw, index, vfta);
1441 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1443 struct e1000_adapter *adapter = netdev_priv(netdev);
1444 struct e1000_hw *hw = &adapter->hw;
1447 e1000_irq_disable(adapter);
1448 vlan_group_set_device(adapter->vlgrp, vid, NULL);
1449 e1000_irq_enable(adapter);
1451 if ((adapter->hw.mng_cookie.status &
1452 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1453 (vid == adapter->mng_vlan_id)) {
1454 /* release control to f/w */
1455 e1000_release_hw_control(adapter);
1459 /* remove VID from filter table */
1460 index = (vid >> 5) & 0x7F;
1461 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1462 vfta &= ~(1 << (vid & 0x1F));
1463 e1000e_write_vfta(hw, index, vfta);
1466 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
1468 struct net_device *netdev = adapter->netdev;
1469 u16 vid = adapter->hw.mng_cookie.vlan_id;
1470 u16 old_vid = adapter->mng_vlan_id;
1472 if (!adapter->vlgrp)
1475 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
1476 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1477 if (adapter->hw.mng_cookie.status &
1478 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1479 e1000_vlan_rx_add_vid(netdev, vid);
1480 adapter->mng_vlan_id = vid;
1483 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
1485 !vlan_group_get_device(adapter->vlgrp, old_vid))
1486 e1000_vlan_rx_kill_vid(netdev, old_vid);
1488 adapter->mng_vlan_id = vid;
1493 static void e1000_vlan_rx_register(struct net_device *netdev,
1494 struct vlan_group *grp)
1496 struct e1000_adapter *adapter = netdev_priv(netdev);
1497 struct e1000_hw *hw = &adapter->hw;
1500 e1000_irq_disable(adapter);
1501 adapter->vlgrp = grp;
1504 /* enable VLAN tag insert/strip */
1506 ctrl |= E1000_CTRL_VME;
1509 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1510 /* enable VLAN receive filtering */
1512 rctl |= E1000_RCTL_VFE;
1513 rctl &= ~E1000_RCTL_CFIEN;
1515 e1000_update_mng_vlan(adapter);
1518 /* disable VLAN tag insert/strip */
1520 ctrl &= ~E1000_CTRL_VME;
1523 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1524 /* disable VLAN filtering */
1526 rctl &= ~E1000_RCTL_VFE;
1528 if (adapter->mng_vlan_id !=
1529 (u16)E1000_MNG_VLAN_NONE) {
1530 e1000_vlan_rx_kill_vid(netdev,
1531 adapter->mng_vlan_id);
1532 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1537 e1000_irq_enable(adapter);
1540 static void e1000_restore_vlan(struct e1000_adapter *adapter)
1544 e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
1546 if (!adapter->vlgrp)
1549 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1550 if (!vlan_group_get_device(adapter->vlgrp, vid))
1552 e1000_vlan_rx_add_vid(adapter->netdev, vid);
1556 static void e1000_init_manageability(struct e1000_adapter *adapter)
1558 struct e1000_hw *hw = &adapter->hw;
1561 if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
1566 /* disable hardware interception of ARP */
1567 manc &= ~(E1000_MANC_ARP_EN);
1569 /* enable receiving management packets to the host. this will probably
1570 * generate destination unreachable messages from the host OS, but
1571 * the packets will be handled on SMBUS */
1572 manc |= E1000_MANC_EN_MNG2HOST;
1573 manc2h = er32(MANC2H);
1574 #define E1000_MNG2HOST_PORT_623 (1 << 5)
1575 #define E1000_MNG2HOST_PORT_664 (1 << 6)
1576 manc2h |= E1000_MNG2HOST_PORT_623;
1577 manc2h |= E1000_MNG2HOST_PORT_664;
1578 ew32(MANC2H, manc2h);
1583 * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1584 * @adapter: board private structure
1586 * Configure the Tx unit of the MAC after a reset.
1588 static void e1000_configure_tx(struct e1000_adapter *adapter)
1590 struct e1000_hw *hw = &adapter->hw;
1591 struct e1000_ring *tx_ring = adapter->tx_ring;
1593 u32 tdlen, tctl, tipg, tarc;
1596 /* Setup the HW Tx Head and Tail descriptor pointers */
1597 tdba = tx_ring->dma;
1598 tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
1599 ew32(TDBAL, (tdba & DMA_32BIT_MASK));
1600 ew32(TDBAH, (tdba >> 32));
1604 tx_ring->head = E1000_TDH;
1605 tx_ring->tail = E1000_TDT;
1607 /* Set the default values for the Tx Inter Packet Gap timer */
1608 tipg = DEFAULT_82543_TIPG_IPGT_COPPER; /* 8 */
1609 ipgr1 = DEFAULT_82543_TIPG_IPGR1; /* 8 */
1610 ipgr2 = DEFAULT_82543_TIPG_IPGR2; /* 6 */
1612 if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
1613 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /* 7 */
1615 tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1616 tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1619 /* Set the Tx Interrupt Delay register */
1620 ew32(TIDV, adapter->tx_int_delay);
1621 /* tx irq moderation */
1622 ew32(TADV, adapter->tx_abs_int_delay);
1624 /* Program the Transmit Control Register */
1626 tctl &= ~E1000_TCTL_CT;
1627 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1628 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1630 if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
1632 /* set the speed mode bit, we'll clear it if we're not at
1633 * gigabit link later */
1634 #define SPEED_MODE_BIT (1 << 21)
1635 tarc |= SPEED_MODE_BIT;
1639 /* errata: program both queues to unweighted RR */
1640 if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
1649 e1000e_config_collision_dist(hw);
1651 /* Setup Transmit Descriptor Settings for eop descriptor */
1652 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1654 /* only set IDE if we are delaying interrupts using the timers */
1655 if (adapter->tx_int_delay)
1656 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1658 /* enable Report Status bit */
1659 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1663 adapter->tx_queue_len = adapter->netdev->tx_queue_len;
1667 * e1000_setup_rctl - configure the receive control registers
1668 * @adapter: Board private structure
1670 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1671 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1672 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1674 struct e1000_hw *hw = &adapter->hw;
1679 /* Program MC offset vector base */
1681 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1682 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1683 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1684 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1686 /* Do not Store bad packets */
1687 rctl &= ~E1000_RCTL_SBP;
1689 /* Enable Long Packet receive */
1690 if (adapter->netdev->mtu <= ETH_DATA_LEN)
1691 rctl &= ~E1000_RCTL_LPE;
1693 rctl |= E1000_RCTL_LPE;
1695 /* Setup buffer sizes */
1696 rctl &= ~E1000_RCTL_SZ_4096;
1697 rctl |= E1000_RCTL_BSEX;
1698 switch (adapter->rx_buffer_len) {
1700 rctl |= E1000_RCTL_SZ_256;
1701 rctl &= ~E1000_RCTL_BSEX;
1704 rctl |= E1000_RCTL_SZ_512;
1705 rctl &= ~E1000_RCTL_BSEX;
1708 rctl |= E1000_RCTL_SZ_1024;
1709 rctl &= ~E1000_RCTL_BSEX;
1713 rctl |= E1000_RCTL_SZ_2048;
1714 rctl &= ~E1000_RCTL_BSEX;
1717 rctl |= E1000_RCTL_SZ_4096;
1720 rctl |= E1000_RCTL_SZ_8192;
1723 rctl |= E1000_RCTL_SZ_16384;
1728 * 82571 and greater support packet-split where the protocol
1729 * header is placed in skb->data and the packet data is
1730 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
1731 * In the case of a non-split, skb->data is linearly filled,
1732 * followed by the page buffers. Therefore, skb->data is
1733 * sized to hold the largest protocol header.
1735 * allocations using alloc_page take too long for regular MTU
1736 * so only enable packet split for jumbo frames
1738 * Using pages when the page size is greater than 16k wastes
1739 * a lot of memory, since we allocate 3 pages at all times
1742 adapter->rx_ps_pages = 0;
1743 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
1744 if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
1745 adapter->rx_ps_pages = pages;
1747 if (adapter->rx_ps_pages) {
1748 /* Configure extra packet-split registers */
1749 rfctl = er32(RFCTL);
1750 rfctl |= E1000_RFCTL_EXTEN;
1751 /* disable packet split support for IPv6 extension headers,
1752 * because some malformed IPv6 headers can hang the RX */
1753 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
1754 E1000_RFCTL_NEW_IPV6_EXT_DIS);
1758 /* Enable Packet split descriptors */
1759 rctl |= E1000_RCTL_DTYP_PS;
1761 /* Enable hardware CRC frame stripping */
1762 rctl |= E1000_RCTL_SECRC;
1764 psrctl |= adapter->rx_ps_bsize0 >>
1765 E1000_PSRCTL_BSIZE0_SHIFT;
1767 switch (adapter->rx_ps_pages) {
1769 psrctl |= PAGE_SIZE <<
1770 E1000_PSRCTL_BSIZE3_SHIFT;
1772 psrctl |= PAGE_SIZE <<
1773 E1000_PSRCTL_BSIZE2_SHIFT;
1775 psrctl |= PAGE_SIZE >>
1776 E1000_PSRCTL_BSIZE1_SHIFT;
1780 ew32(PSRCTL, psrctl);
1787 * e1000_configure_rx - Configure Receive Unit after Reset
1788 * @adapter: board private structure
1790 * Configure the Rx unit of the MAC after a reset.
1792 static void e1000_configure_rx(struct e1000_adapter *adapter)
1794 struct e1000_hw *hw = &adapter->hw;
1795 struct e1000_ring *rx_ring = adapter->rx_ring;
1797 u32 rdlen, rctl, rxcsum, ctrl_ext;
1799 if (adapter->rx_ps_pages) {
1800 /* this is a 32 byte descriptor */
1801 rdlen = rx_ring->count *
1802 sizeof(union e1000_rx_desc_packet_split);
1803 adapter->clean_rx = e1000_clean_rx_irq_ps;
1804 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
1806 rdlen = rx_ring->count *
1807 sizeof(struct e1000_rx_desc);
1808 adapter->clean_rx = e1000_clean_rx_irq;
1809 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1812 /* disable receives while setting up the descriptors */
1814 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1818 /* set the Receive Delay Timer Register */
1819 ew32(RDTR, adapter->rx_int_delay);
1821 /* irq moderation */
1822 ew32(RADV, adapter->rx_abs_int_delay);
1823 if (adapter->itr_setting != 0)
1825 1000000000 / (adapter->itr * 256));
1827 ctrl_ext = er32(CTRL_EXT);
1828 /* Reset delay timers after every interrupt */
1829 ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
1830 /* Auto-Mask interrupts upon ICR access */
1831 ctrl_ext |= E1000_CTRL_EXT_IAME;
1832 ew32(IAM, 0xffffffff);
1833 ew32(CTRL_EXT, ctrl_ext);
1836 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1837 * the Base and Length of the Rx Descriptor Ring */
1838 rdba = rx_ring->dma;
1839 ew32(RDBAL, (rdba & DMA_32BIT_MASK));
1840 ew32(RDBAH, (rdba >> 32));
1844 rx_ring->head = E1000_RDH;
1845 rx_ring->tail = E1000_RDT;
1847 /* Enable Receive Checksum Offload for TCP and UDP */
1848 rxcsum = er32(RXCSUM);
1849 if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
1850 rxcsum |= E1000_RXCSUM_TUOFL;
1852 /* IPv4 payload checksum for UDP fragments must be
1853 * used in conjunction with packet-split. */
1854 if (adapter->rx_ps_pages)
1855 rxcsum |= E1000_RXCSUM_IPPCSE;
1857 rxcsum &= ~E1000_RXCSUM_TUOFL;
1858 /* no need to clear IPPCSE as it defaults to 0 */
1860 ew32(RXCSUM, rxcsum);
1862 /* Enable early receives on supported devices, only takes effect when
1863 * packet size is equal or larger than the specified value (in 8 byte
1864 * units), e.g. using jumbo frames when setting to E1000_ERT_2048 */
1865 if ((adapter->flags & FLAG_HAS_ERT) &&
1866 (adapter->netdev->mtu > ETH_DATA_LEN))
1867 ew32(ERT, E1000_ERT_2048);
1869 /* Enable Receives */
1874 * e1000_mc_addr_list_update - Update Multicast addresses
1875 * @hw: pointer to the HW structure
1876 * @mc_addr_list: array of multicast addresses to program
1877 * @mc_addr_count: number of multicast addresses to program
1878 * @rar_used_count: the first RAR register free to program
1879 * @rar_count: total number of supported Receive Address Registers
1881 * Updates the Receive Address Registers and Multicast Table Array.
1882 * The caller must have a packed mc_addr_list of multicast addresses.
1883 * The parameter rar_count will usually be hw->mac.rar_entry_count
1884 * unless there are workarounds that change this. Currently no func pointer
1885 * exists and all implementations are handled in the generic version of this
1888 static void e1000_mc_addr_list_update(struct e1000_hw *hw, u8 *mc_addr_list,
1889 u32 mc_addr_count, u32 rar_used_count,
1892 hw->mac.ops.mc_addr_list_update(hw, mc_addr_list, mc_addr_count,
1893 rar_used_count, rar_count);
1897 * e1000_set_multi - Multicast and Promiscuous mode set
1898 * @netdev: network interface device structure
1900 * The set_multi entry point is called whenever the multicast address
1901 * list or the network interface flags are updated. This routine is
1902 * responsible for configuring the hardware for proper multicast,
1903 * promiscuous mode, and all-multi behavior.
1905 static void e1000_set_multi(struct net_device *netdev)
1907 struct e1000_adapter *adapter = netdev_priv(netdev);
1908 struct e1000_hw *hw = &adapter->hw;
1909 struct e1000_mac_info *mac = &hw->mac;
1910 struct dev_mc_list *mc_ptr;
1915 /* Check for Promiscuous and All Multicast modes */
1919 if (netdev->flags & IFF_PROMISC) {
1920 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
1921 } else if (netdev->flags & IFF_ALLMULTI) {
1922 rctl |= E1000_RCTL_MPE;
1923 rctl &= ~E1000_RCTL_UPE;
1925 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
1930 if (netdev->mc_count) {
1931 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
1935 /* prepare a packed array of only addresses. */
1936 mc_ptr = netdev->mc_list;
1938 for (i = 0; i < netdev->mc_count; i++) {
1941 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
1943 mc_ptr = mc_ptr->next;
1946 e1000_mc_addr_list_update(hw, mta_list, i, 1,
1947 mac->rar_entry_count);
1951 * if we're called from probe, we might not have
1952 * anything to do here, so clear out the list
1954 e1000_mc_addr_list_update(hw, NULL, 0, 1,
1955 mac->rar_entry_count);
1960 * e1000_configure - configure the hardware for RX and TX
1961 * @adapter: private board structure
1963 static void e1000_configure(struct e1000_adapter *adapter)
1965 e1000_set_multi(adapter->netdev);
1967 e1000_restore_vlan(adapter);
1968 e1000_init_manageability(adapter);
1970 e1000_configure_tx(adapter);
1971 e1000_setup_rctl(adapter);
1972 e1000_configure_rx(adapter);
1973 adapter->alloc_rx_buf(adapter,
1974 e1000_desc_unused(adapter->rx_ring));
1978 * e1000e_power_up_phy - restore link in case the phy was powered down
1979 * @adapter: address of board private structure
1981 * The phy may be powered down to save power and turn off link when the
1982 * driver is unloaded and wake on lan is not enabled (among others)
1983 * *** this routine MUST be followed by a call to e1000e_reset ***
1985 void e1000e_power_up_phy(struct e1000_adapter *adapter)
1989 /* Just clear the power down bit to wake the phy back up */
1990 if (adapter->hw.media_type == e1000_media_type_copper) {
1991 /* according to the manual, the phy will retain its
1992 * settings across a power-down/up cycle */
1993 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
1994 mii_reg &= ~MII_CR_POWER_DOWN;
1995 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
1998 adapter->hw.mac.ops.setup_link(&adapter->hw);
2002 * e1000_power_down_phy - Power down the PHY
2004 * Power down the PHY so no link is implied when interface is down
2005 * The PHY cannot be powered down is management or WoL is active
2007 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2009 struct e1000_hw *hw = &adapter->hw;
2012 /* WoL is enabled */
2016 /* non-copper PHY? */
2017 if (adapter->hw.media_type != e1000_media_type_copper)
2020 /* reset is blocked because of a SoL/IDER session */
2021 if (e1000e_check_mng_mode(hw) ||
2022 e1000_check_reset_block(hw))
2025 /* managebility (AMT) is enabled */
2026 if (er32(MANC) & E1000_MANC_SMBUS_EN)
2029 /* power down the PHY */
2030 e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2031 mii_reg |= MII_CR_POWER_DOWN;
2032 e1e_wphy(hw, PHY_CONTROL, mii_reg);
2037 * e1000e_reset - bring the hardware into a known good state
2039 * This function boots the hardware and enables some settings that
2040 * require a configuration cycle of the hardware - those cannot be
2041 * set/changed during runtime. After reset the device needs to be
2042 * properly configured for rx, tx etc.
2044 void e1000e_reset(struct e1000_adapter *adapter)
2046 struct e1000_mac_info *mac = &adapter->hw.mac;
2047 struct e1000_hw *hw = &adapter->hw;
2048 u32 tx_space, min_tx_space, min_rx_space;
2052 ew32(PBA, adapter->pba);
2054 if (mac->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN ) {
2055 /* To maintain wire speed transmits, the Tx FIFO should be
2056 * large enough to accommodate two full transmit packets,
2057 * rounded up to the next 1KB and expressed in KB. Likewise,
2058 * the Rx FIFO should be large enough to accommodate at least
2059 * one full receive packet and is similarly rounded up and
2060 * expressed in KB. */
2062 /* upper 16 bits has Tx packet buffer allocation size in KB */
2063 tx_space = pba >> 16;
2064 /* lower 16 bits has Rx packet buffer allocation size in KB */
2066 /* the tx fifo also stores 16 bytes of information about the tx
2067 * but don't include ethernet FCS because hardware appends it */
2068 min_tx_space = (mac->max_frame_size +
2069 sizeof(struct e1000_tx_desc) -
2071 min_tx_space = ALIGN(min_tx_space, 1024);
2072 min_tx_space >>= 10;
2073 /* software strips receive CRC, so leave room for it */
2074 min_rx_space = mac->max_frame_size;
2075 min_rx_space = ALIGN(min_rx_space, 1024);
2076 min_rx_space >>= 10;
2078 /* If current Tx allocation is less than the min Tx FIFO size,
2079 * and the min Tx FIFO size is less than the current Rx FIFO
2080 * allocation, take space away from current Rx allocation */
2081 if ((tx_space < min_tx_space) &&
2082 ((min_tx_space - tx_space) < pba)) {
2083 pba -= min_tx_space - tx_space;
2085 /* if short on rx space, rx wins and must trump tx
2086 * adjustment or use Early Receive if available */
2087 if ((pba < min_rx_space) &&
2088 (!(adapter->flags & FLAG_HAS_ERT)))
2089 /* ERT enabled in e1000_configure_rx */
2097 /* flow control settings */
2098 /* The high water mark must be low enough to fit one full frame
2099 * (or the size used for early receive) above it in the Rx FIFO.
2100 * Set it to the lower of:
2101 * - 90% of the Rx FIFO size, and
2102 * - the full Rx FIFO size minus the early receive size (for parts
2103 * with ERT support assuming ERT set to E1000_ERT_2048), or
2104 * - the full Rx FIFO size minus one full frame */
2105 if (adapter->flags & FLAG_HAS_ERT)
2106 hwm = min(((adapter->pba << 10) * 9 / 10),
2107 ((adapter->pba << 10) - (E1000_ERT_2048 << 3)));
2109 hwm = min(((adapter->pba << 10) * 9 / 10),
2110 ((adapter->pba << 10) - mac->max_frame_size));
2112 mac->fc_high_water = hwm & 0xFFF8; /* 8-byte granularity */
2113 mac->fc_low_water = mac->fc_high_water - 8;
2115 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2116 mac->fc_pause_time = 0xFFFF;
2118 mac->fc_pause_time = E1000_FC_PAUSE_TIME;
2119 mac->fc = mac->original_fc;
2121 /* Allow time for pending master requests to run */
2122 mac->ops.reset_hw(hw);
2125 if (mac->ops.init_hw(hw))
2126 ndev_err(adapter->netdev, "Hardware Error\n");
2128 e1000_update_mng_vlan(adapter);
2130 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2131 ew32(VET, ETH_P_8021Q);
2133 e1000e_reset_adaptive(hw);
2134 e1000_get_phy_info(hw);
2136 if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2138 /* speed up time to link by disabling smart power down, ignore
2139 * the return value of this function because there is nothing
2140 * different we would do if it failed */
2141 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2142 phy_data &= ~IGP02E1000_PM_SPD;
2143 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2146 e1000_release_manageability(adapter);
2149 int e1000e_up(struct e1000_adapter *adapter)
2151 struct e1000_hw *hw = &adapter->hw;
2153 /* hardware has been reset, we need to reload some things */
2154 e1000_configure(adapter);
2156 clear_bit(__E1000_DOWN, &adapter->state);
2158 napi_enable(&adapter->napi);
2159 e1000_irq_enable(adapter);
2161 /* fire a link change interrupt to start the watchdog */
2162 ew32(ICS, E1000_ICS_LSC);
2166 void e1000e_down(struct e1000_adapter *adapter)
2168 struct net_device *netdev = adapter->netdev;
2169 struct e1000_hw *hw = &adapter->hw;
2172 /* signal that we're down so the interrupt handler does not
2173 * reschedule our watchdog timer */
2174 set_bit(__E1000_DOWN, &adapter->state);
2176 /* disable receives in the hardware */
2178 ew32(RCTL, rctl & ~E1000_RCTL_EN);
2179 /* flush and sleep below */
2181 netif_stop_queue(netdev);
2183 /* disable transmits in the hardware */
2185 tctl &= ~E1000_TCTL_EN;
2187 /* flush both disables and wait for them to finish */
2191 napi_disable(&adapter->napi);
2192 atomic_set(&adapter->irq_sem, 0);
2193 e1000_irq_disable(adapter);
2195 del_timer_sync(&adapter->watchdog_timer);
2196 del_timer_sync(&adapter->phy_info_timer);
2198 netdev->tx_queue_len = adapter->tx_queue_len;
2199 netif_carrier_off(netdev);
2200 adapter->link_speed = 0;
2201 adapter->link_duplex = 0;
2203 e1000e_reset(adapter);
2204 e1000_clean_tx_ring(adapter);
2205 e1000_clean_rx_ring(adapter);
2208 * TODO: for power management, we could drop the link and
2209 * pci_disable_device here.
2213 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2216 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2218 e1000e_down(adapter);
2220 clear_bit(__E1000_RESETTING, &adapter->state);
2224 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2225 * @adapter: board private structure to initialize
2227 * e1000_sw_init initializes the Adapter private data structure.
2228 * Fields are initialized based on PCI device information and
2229 * OS network device settings (MTU size).
2231 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2233 struct e1000_hw *hw = &adapter->hw;
2234 struct net_device *netdev = adapter->netdev;
2236 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2237 adapter->rx_ps_bsize0 = 128;
2238 hw->mac.max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2239 hw->mac.min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2241 adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2242 if (!adapter->tx_ring)
2245 adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2246 if (!adapter->rx_ring)
2249 spin_lock_init(&adapter->tx_queue_lock);
2251 /* Explicitly disable IRQ since the NIC can be in any state. */
2252 atomic_set(&adapter->irq_sem, 0);
2253 e1000_irq_disable(adapter);
2255 spin_lock_init(&adapter->stats_lock);
2257 set_bit(__E1000_DOWN, &adapter->state);
2261 ndev_err(netdev, "Unable to allocate memory for queues\n");
2262 kfree(adapter->rx_ring);
2263 kfree(adapter->tx_ring);
2268 * e1000_open - Called when a network interface is made active
2269 * @netdev: network interface device structure
2271 * Returns 0 on success, negative value on failure
2273 * The open entry point is called when a network interface is made
2274 * active by the system (IFF_UP). At this point all resources needed
2275 * for transmit and receive operations are allocated, the interrupt
2276 * handler is registered with the OS, the watchdog timer is started,
2277 * and the stack is notified that the interface is ready.
2279 static int e1000_open(struct net_device *netdev)
2281 struct e1000_adapter *adapter = netdev_priv(netdev);
2282 struct e1000_hw *hw = &adapter->hw;
2285 /* disallow open during test */
2286 if (test_bit(__E1000_TESTING, &adapter->state))
2289 /* allocate transmit descriptors */
2290 err = e1000e_setup_tx_resources(adapter);
2294 /* allocate receive descriptors */
2295 err = e1000e_setup_rx_resources(adapter);
2299 e1000e_power_up_phy(adapter);
2301 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2302 if ((adapter->hw.mng_cookie.status &
2303 E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
2304 e1000_update_mng_vlan(adapter);
2306 /* If AMT is enabled, let the firmware know that the network
2307 * interface is now open */
2308 if ((adapter->flags & FLAG_HAS_AMT) &&
2309 e1000e_check_mng_mode(&adapter->hw))
2310 e1000_get_hw_control(adapter);
2312 /* before we allocate an interrupt, we must be ready to handle it.
2313 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2314 * as soon as we call pci_request_irq, so we have to setup our
2315 * clean_rx handler before we do so. */
2316 e1000_configure(adapter);
2318 err = e1000_request_irq(adapter);
2322 /* From here on the code is the same as e1000e_up() */
2323 clear_bit(__E1000_DOWN, &adapter->state);
2325 napi_enable(&adapter->napi);
2327 e1000_irq_enable(adapter);
2329 /* fire a link status change interrupt to start the watchdog */
2330 ew32(ICS, E1000_ICS_LSC);
2335 e1000_release_hw_control(adapter);
2336 e1000_power_down_phy(adapter);
2337 e1000e_free_rx_resources(adapter);
2339 e1000e_free_tx_resources(adapter);
2341 e1000e_reset(adapter);
2347 * e1000_close - Disables a network interface
2348 * @netdev: network interface device structure
2350 * Returns 0, this is not allowed to fail
2352 * The close entry point is called when an interface is de-activated
2353 * by the OS. The hardware is still under the drivers control, but
2354 * needs to be disabled. A global MAC reset is issued to stop the
2355 * hardware, and all transmit and receive resources are freed.
2357 static int e1000_close(struct net_device *netdev)
2359 struct e1000_adapter *adapter = netdev_priv(netdev);
2361 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
2362 e1000e_down(adapter);
2363 e1000_power_down_phy(adapter);
2364 e1000_free_irq(adapter);
2366 e1000e_free_tx_resources(adapter);
2367 e1000e_free_rx_resources(adapter);
2369 /* kill manageability vlan ID if supported, but not if a vlan with
2370 * the same ID is registered on the host OS (let 8021q kill it) */
2371 if ((adapter->hw.mng_cookie.status &
2372 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2374 vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
2375 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2377 /* If AMT is enabled, let the firmware know that the network
2378 * interface is now closed */
2379 if ((adapter->flags & FLAG_HAS_AMT) &&
2380 e1000e_check_mng_mode(&adapter->hw))
2381 e1000_release_hw_control(adapter);
2386 * e1000_set_mac - Change the Ethernet Address of the NIC
2387 * @netdev: network interface device structure
2388 * @p: pointer to an address structure
2390 * Returns 0 on success, negative on failure
2392 static int e1000_set_mac(struct net_device *netdev, void *p)
2394 struct e1000_adapter *adapter = netdev_priv(netdev);
2395 struct sockaddr *addr = p;
2397 if (!is_valid_ether_addr(addr->sa_data))
2398 return -EADDRNOTAVAIL;
2400 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2401 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
2403 e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
2405 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
2406 /* activate the work around */
2407 e1000e_set_laa_state_82571(&adapter->hw, 1);
2409 /* Hold a copy of the LAA in RAR[14] This is done so that
2410 * between the time RAR[0] gets clobbered and the time it
2411 * gets fixed (in e1000_watchdog), the actual LAA is in one
2412 * of the RARs and no incoming packets directed to this port
2413 * are dropped. Eventually the LAA will be in RAR[0] and
2415 e1000e_rar_set(&adapter->hw,
2416 adapter->hw.mac.addr,
2417 adapter->hw.mac.rar_entry_count - 1);
2423 /* Need to wait a few seconds after link up to get diagnostic information from
2425 static void e1000_update_phy_info(unsigned long data)
2427 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2428 e1000_get_phy_info(&adapter->hw);
2432 * e1000e_update_stats - Update the board statistics counters
2433 * @adapter: board private structure
2435 void e1000e_update_stats(struct e1000_adapter *adapter)
2437 struct e1000_hw *hw = &adapter->hw;
2438 struct pci_dev *pdev = adapter->pdev;
2439 unsigned long irq_flags;
2442 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
2445 * Prevent stats update while adapter is being reset, or if the pci
2446 * connection is down.
2448 if (adapter->link_speed == 0)
2450 if (pci_channel_offline(pdev))
2453 spin_lock_irqsave(&adapter->stats_lock, irq_flags);
2455 /* these counters are modified from e1000_adjust_tbi_stats,
2456 * called from the interrupt context, so they must only
2457 * be written while holding adapter->stats_lock
2460 adapter->stats.crcerrs += er32(CRCERRS);
2461 adapter->stats.gprc += er32(GPRC);
2462 adapter->stats.gorcl += er32(GORCL);
2463 adapter->stats.gorch += er32(GORCH);
2464 adapter->stats.bprc += er32(BPRC);
2465 adapter->stats.mprc += er32(MPRC);
2466 adapter->stats.roc += er32(ROC);
2468 if (adapter->flags & FLAG_HAS_STATS_PTC_PRC) {
2469 adapter->stats.prc64 += er32(PRC64);
2470 adapter->stats.prc127 += er32(PRC127);
2471 adapter->stats.prc255 += er32(PRC255);
2472 adapter->stats.prc511 += er32(PRC511);
2473 adapter->stats.prc1023 += er32(PRC1023);
2474 adapter->stats.prc1522 += er32(PRC1522);
2475 adapter->stats.symerrs += er32(SYMERRS);
2476 adapter->stats.sec += er32(SEC);
2479 adapter->stats.mpc += er32(MPC);
2480 adapter->stats.scc += er32(SCC);
2481 adapter->stats.ecol += er32(ECOL);
2482 adapter->stats.mcc += er32(MCC);
2483 adapter->stats.latecol += er32(LATECOL);
2484 adapter->stats.dc += er32(DC);
2485 adapter->stats.rlec += er32(RLEC);
2486 adapter->stats.xonrxc += er32(XONRXC);
2487 adapter->stats.xontxc += er32(XONTXC);
2488 adapter->stats.xoffrxc += er32(XOFFRXC);
2489 adapter->stats.xofftxc += er32(XOFFTXC);
2490 adapter->stats.fcruc += er32(FCRUC);
2491 adapter->stats.gptc += er32(GPTC);
2492 adapter->stats.gotcl += er32(GOTCL);
2493 adapter->stats.gotch += er32(GOTCH);
2494 adapter->stats.rnbc += er32(RNBC);
2495 adapter->stats.ruc += er32(RUC);
2496 adapter->stats.rfc += er32(RFC);
2497 adapter->stats.rjc += er32(RJC);
2498 adapter->stats.torl += er32(TORL);
2499 adapter->stats.torh += er32(TORH);
2500 adapter->stats.totl += er32(TOTL);
2501 adapter->stats.toth += er32(TOTH);
2502 adapter->stats.tpr += er32(TPR);
2504 if (adapter->flags & FLAG_HAS_STATS_PTC_PRC) {
2505 adapter->stats.ptc64 += er32(PTC64);
2506 adapter->stats.ptc127 += er32(PTC127);
2507 adapter->stats.ptc255 += er32(PTC255);
2508 adapter->stats.ptc511 += er32(PTC511);
2509 adapter->stats.ptc1023 += er32(PTC1023);
2510 adapter->stats.ptc1522 += er32(PTC1522);
2513 adapter->stats.mptc += er32(MPTC);
2514 adapter->stats.bptc += er32(BPTC);
2516 /* used for adaptive IFS */
2518 hw->mac.tx_packet_delta = er32(TPT);
2519 adapter->stats.tpt += hw->mac.tx_packet_delta;
2520 hw->mac.collision_delta = er32(COLC);
2521 adapter->stats.colc += hw->mac.collision_delta;
2523 adapter->stats.algnerrc += er32(ALGNERRC);
2524 adapter->stats.rxerrc += er32(RXERRC);
2525 adapter->stats.tncrs += er32(TNCRS);
2526 adapter->stats.cexterr += er32(CEXTERR);
2527 adapter->stats.tsctc += er32(TSCTC);
2528 adapter->stats.tsctfc += er32(TSCTFC);
2530 adapter->stats.iac += er32(IAC);
2532 if (adapter->flags & FLAG_HAS_STATS_ICR_ICT) {
2533 adapter->stats.icrxoc += er32(ICRXOC);
2534 adapter->stats.icrxptc += er32(ICRXPTC);
2535 adapter->stats.icrxatc += er32(ICRXATC);
2536 adapter->stats.ictxptc += er32(ICTXPTC);
2537 adapter->stats.ictxatc += er32(ICTXATC);
2538 adapter->stats.ictxqec += er32(ICTXQEC);
2539 adapter->stats.ictxqmtc += er32(ICTXQMTC);
2540 adapter->stats.icrxdmtc += er32(ICRXDMTC);
2543 /* Fill out the OS statistics structure */
2544 adapter->net_stats.multicast = adapter->stats.mprc;
2545 adapter->net_stats.collisions = adapter->stats.colc;
2549 /* RLEC on some newer hardware can be incorrect so build
2550 * our own version based on RUC and ROC */
2551 adapter->net_stats.rx_errors = adapter->stats.rxerrc +
2552 adapter->stats.crcerrs + adapter->stats.algnerrc +
2553 adapter->stats.ruc + adapter->stats.roc +
2554 adapter->stats.cexterr;
2555 adapter->net_stats.rx_length_errors = adapter->stats.ruc +
2557 adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
2558 adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
2559 adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
2562 adapter->net_stats.tx_errors = adapter->stats.ecol +
2563 adapter->stats.latecol;
2564 adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
2565 adapter->net_stats.tx_window_errors = adapter->stats.latecol;
2566 adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
2568 /* Tx Dropped needs to be maintained elsewhere */
2571 if (hw->media_type == e1000_media_type_copper) {
2572 if ((adapter->link_speed == SPEED_1000) &&
2573 (!e1e_rphy(hw, PHY_1000T_STATUS, &phy_tmp))) {
2574 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
2575 adapter->phy_stats.idle_errors += phy_tmp;
2579 /* Management Stats */
2580 adapter->stats.mgptc += er32(MGTPTC);
2581 adapter->stats.mgprc += er32(MGTPRC);
2582 adapter->stats.mgpdc += er32(MGTPDC);
2584 spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
2587 static void e1000_print_link_info(struct e1000_adapter *adapter)
2589 struct net_device *netdev = adapter->netdev;
2590 struct e1000_hw *hw = &adapter->hw;
2591 u32 ctrl = er32(CTRL);
2594 "Link is Up %d Mbps %s, Flow Control: %s\n",
2595 adapter->link_speed,
2596 (adapter->link_duplex == FULL_DUPLEX) ?
2597 "Full Duplex" : "Half Duplex",
2598 ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
2600 ((ctrl & E1000_CTRL_RFCE) ? "RX" :
2601 ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
2605 * e1000_watchdog - Timer Call-back
2606 * @data: pointer to adapter cast into an unsigned long
2608 static void e1000_watchdog(unsigned long data)
2610 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2612 /* Do the rest outside of interrupt context */
2613 schedule_work(&adapter->watchdog_task);
2615 /* TODO: make this use queue_delayed_work() */
2618 static void e1000_watchdog_task(struct work_struct *work)
2620 struct e1000_adapter *adapter = container_of(work,
2621 struct e1000_adapter, watchdog_task);
2623 struct net_device *netdev = adapter->netdev;
2624 struct e1000_mac_info *mac = &adapter->hw.mac;
2625 struct e1000_ring *tx_ring = adapter->tx_ring;
2626 struct e1000_hw *hw = &adapter->hw;
2631 if ((netif_carrier_ok(netdev)) &&
2632 (er32(STATUS) & E1000_STATUS_LU))
2635 ret_val = mac->ops.check_for_link(hw);
2636 if ((ret_val == E1000_ERR_PHY) &&
2637 (adapter->hw.phy.type == e1000_phy_igp_3) &&
2639 E1000_PHY_CTRL_GBE_DISABLE)) {
2640 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
2642 "Gigabit has been disabled, downgrading speed\n");
2645 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
2646 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
2647 e1000_update_mng_vlan(adapter);
2649 if ((adapter->hw.media_type == e1000_media_type_internal_serdes) &&
2650 !(er32(TXCW) & E1000_TXCW_ANE))
2651 link = adapter->hw.mac.serdes_has_link;
2653 link = er32(STATUS) & E1000_STATUS_LU;
2656 if (!netif_carrier_ok(netdev)) {
2658 mac->ops.get_link_up_info(&adapter->hw,
2659 &adapter->link_speed,
2660 &adapter->link_duplex);
2661 e1000_print_link_info(adapter);
2662 /* tweak tx_queue_len according to speed/duplex
2663 * and adjust the timeout factor */
2664 netdev->tx_queue_len = adapter->tx_queue_len;
2665 adapter->tx_timeout_factor = 1;
2666 switch (adapter->link_speed) {
2669 netdev->tx_queue_len = 10;
2670 adapter->tx_timeout_factor = 14;
2674 netdev->tx_queue_len = 100;
2675 /* maybe add some timeout factor ? */
2679 /* workaround: re-program speed mode bit after
2681 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
2684 tarc0 = er32(TARC0);
2685 tarc0 &= ~SPEED_MODE_BIT;
2689 /* disable TSO for pcie and 10/100 speeds, to avoid
2690 * some hardware issues */
2691 if (!(adapter->flags & FLAG_TSO_FORCE)) {
2692 switch (adapter->link_speed) {
2696 "10/100 speed: disabling TSO\n");
2697 netdev->features &= ~NETIF_F_TSO;
2698 netdev->features &= ~NETIF_F_TSO6;
2701 netdev->features |= NETIF_F_TSO;
2702 netdev->features |= NETIF_F_TSO6;
2710 /* enable transmits in the hardware, need to do this
2711 * after setting TARC0 */
2713 tctl |= E1000_TCTL_EN;
2716 netif_carrier_on(netdev);
2717 netif_wake_queue(netdev);
2719 if (!test_bit(__E1000_DOWN, &adapter->state))
2720 mod_timer(&adapter->phy_info_timer,
2721 round_jiffies(jiffies + 2 * HZ));
2723 /* make sure the receive unit is started */
2724 if (adapter->flags & FLAG_RX_NEEDS_RESTART) {
2725 u32 rctl = er32(RCTL);
2731 if (netif_carrier_ok(netdev)) {
2732 adapter->link_speed = 0;
2733 adapter->link_duplex = 0;
2734 ndev_info(netdev, "Link is Down\n");
2735 netif_carrier_off(netdev);
2736 netif_stop_queue(netdev);
2737 if (!test_bit(__E1000_DOWN, &adapter->state))
2738 mod_timer(&adapter->phy_info_timer,
2739 round_jiffies(jiffies + 2 * HZ));
2741 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
2742 schedule_work(&adapter->reset_task);
2747 e1000e_update_stats(adapter);
2749 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2750 adapter->tpt_old = adapter->stats.tpt;
2751 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
2752 adapter->colc_old = adapter->stats.colc;
2754 adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2755 adapter->gorcl_old = adapter->stats.gorcl;
2756 adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2757 adapter->gotcl_old = adapter->stats.gotcl;
2759 e1000e_update_adaptive(&adapter->hw);
2761 if (!netif_carrier_ok(netdev)) {
2762 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
2765 /* We've lost link, so the controller stops DMA,
2766 * but we've got queued Tx work that's never going
2767 * to get done, so reset controller to flush Tx.
2768 * (Do the reset outside of interrupt context). */
2769 adapter->tx_timeout_count++;
2770 schedule_work(&adapter->reset_task);
2774 /* Cause software interrupt to ensure rx ring is cleaned */
2775 ew32(ICS, E1000_ICS_RXDMT0);
2777 /* Force detection of hung controller every watchdog period */
2778 adapter->detect_tx_hung = 1;
2780 /* With 82571 controllers, LAA may be overwritten due to controller
2781 * reset from the other port. Set the appropriate LAA in RAR[0] */
2782 if (e1000e_get_laa_state_82571(hw))
2783 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
2785 /* Reset the timer */
2786 if (!test_bit(__E1000_DOWN, &adapter->state))
2787 mod_timer(&adapter->watchdog_timer,
2788 round_jiffies(jiffies + 2 * HZ));
2791 #define E1000_TX_FLAGS_CSUM 0x00000001
2792 #define E1000_TX_FLAGS_VLAN 0x00000002
2793 #define E1000_TX_FLAGS_TSO 0x00000004
2794 #define E1000_TX_FLAGS_IPV4 0x00000008
2795 #define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
2796 #define E1000_TX_FLAGS_VLAN_SHIFT 16
2798 static int e1000_tso(struct e1000_adapter *adapter,
2799 struct sk_buff *skb)
2801 struct e1000_ring *tx_ring = adapter->tx_ring;
2802 struct e1000_context_desc *context_desc;
2803 struct e1000_buffer *buffer_info;
2806 u16 ipcse = 0, tucse, mss;
2807 u8 ipcss, ipcso, tucss, tucso, hdr_len;
2810 if (skb_is_gso(skb)) {
2811 if (skb_header_cloned(skb)) {
2812 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2817 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2818 mss = skb_shinfo(skb)->gso_size;
2819 if (skb->protocol == htons(ETH_P_IP)) {
2820 struct iphdr *iph = ip_hdr(skb);
2823 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2827 cmd_length = E1000_TXD_CMD_IP;
2828 ipcse = skb_transport_offset(skb) - 1;
2829 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
2830 ipv6_hdr(skb)->payload_len = 0;
2831 tcp_hdr(skb)->check =
2832 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2833 &ipv6_hdr(skb)->daddr,
2837 ipcss = skb_network_offset(skb);
2838 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2839 tucss = skb_transport_offset(skb);
2840 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2843 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2844 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2846 i = tx_ring->next_to_use;
2847 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2848 buffer_info = &tx_ring->buffer_info[i];
2850 context_desc->lower_setup.ip_fields.ipcss = ipcss;
2851 context_desc->lower_setup.ip_fields.ipcso = ipcso;
2852 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
2853 context_desc->upper_setup.tcp_fields.tucss = tucss;
2854 context_desc->upper_setup.tcp_fields.tucso = tucso;
2855 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2856 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
2857 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2858 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2860 buffer_info->time_stamp = jiffies;
2861 buffer_info->next_to_watch = i;
2864 if (i == tx_ring->count)
2866 tx_ring->next_to_use = i;
2874 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
2876 struct e1000_ring *tx_ring = adapter->tx_ring;
2877 struct e1000_context_desc *context_desc;
2878 struct e1000_buffer *buffer_info;
2882 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2883 css = skb_transport_offset(skb);
2885 i = tx_ring->next_to_use;
2886 buffer_info = &tx_ring->buffer_info[i];
2887 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2889 context_desc->lower_setup.ip_config = 0;
2890 context_desc->upper_setup.tcp_fields.tucss = css;
2891 context_desc->upper_setup.tcp_fields.tucso =
2892 css + skb->csum_offset;
2893 context_desc->upper_setup.tcp_fields.tucse = 0;
2894 context_desc->tcp_seg_setup.data = 0;
2895 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
2897 buffer_info->time_stamp = jiffies;
2898 buffer_info->next_to_watch = i;
2901 if (i == tx_ring->count)
2903 tx_ring->next_to_use = i;
2911 #define E1000_MAX_PER_TXD 8192
2912 #define E1000_MAX_TXD_PWR 12
2914 static int e1000_tx_map(struct e1000_adapter *adapter,
2915 struct sk_buff *skb, unsigned int first,
2916 unsigned int max_per_txd, unsigned int nr_frags,
2919 struct e1000_ring *tx_ring = adapter->tx_ring;
2920 struct e1000_buffer *buffer_info;
2921 unsigned int len = skb->len - skb->data_len;
2922 unsigned int offset = 0, size, count = 0, i;
2925 i = tx_ring->next_to_use;
2928 buffer_info = &tx_ring->buffer_info[i];
2929 size = min(len, max_per_txd);
2931 /* Workaround for premature desc write-backs
2932 * in TSO mode. Append 4-byte sentinel desc */
2933 if (mss && !nr_frags && size == len && size > 8)
2936 buffer_info->length = size;
2937 /* set time_stamp *before* dma to help avoid a possible race */
2938 buffer_info->time_stamp = jiffies;
2940 pci_map_single(adapter->pdev,
2944 if (pci_dma_mapping_error(buffer_info->dma)) {
2945 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
2946 adapter->tx_dma_failed++;
2949 buffer_info->next_to_watch = i;
2955 if (i == tx_ring->count)
2959 for (f = 0; f < nr_frags; f++) {
2960 struct skb_frag_struct *frag;
2962 frag = &skb_shinfo(skb)->frags[f];
2964 offset = frag->page_offset;
2967 buffer_info = &tx_ring->buffer_info[i];
2968 size = min(len, max_per_txd);
2969 /* Workaround for premature desc write-backs
2970 * in TSO mode. Append 4-byte sentinel desc */
2971 if (mss && f == (nr_frags-1) && size == len && size > 8)
2974 buffer_info->length = size;
2975 buffer_info->time_stamp = jiffies;
2977 pci_map_page(adapter->pdev,
2982 if (pci_dma_mapping_error(buffer_info->dma)) {
2983 dev_err(&adapter->pdev->dev,
2984 "TX DMA page map failed\n");
2985 adapter->tx_dma_failed++;
2989 buffer_info->next_to_watch = i;
2996 if (i == tx_ring->count)
3002 i = tx_ring->count - 1;
3006 tx_ring->buffer_info[i].skb = skb;
3007 tx_ring->buffer_info[first].next_to_watch = i;
3012 static void e1000_tx_queue(struct e1000_adapter *adapter,
3013 int tx_flags, int count)
3015 struct e1000_ring *tx_ring = adapter->tx_ring;
3016 struct e1000_tx_desc *tx_desc = NULL;
3017 struct e1000_buffer *buffer_info;
3018 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3021 if (tx_flags & E1000_TX_FLAGS_TSO) {
3022 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3024 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3026 if (tx_flags & E1000_TX_FLAGS_IPV4)
3027 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3030 if (tx_flags & E1000_TX_FLAGS_CSUM) {
3031 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3032 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3035 if (tx_flags & E1000_TX_FLAGS_VLAN) {
3036 txd_lower |= E1000_TXD_CMD_VLE;
3037 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3040 i = tx_ring->next_to_use;
3043 buffer_info = &tx_ring->buffer_info[i];
3044 tx_desc = E1000_TX_DESC(*tx_ring, i);
3045 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3046 tx_desc->lower.data =
3047 cpu_to_le32(txd_lower | buffer_info->length);
3048 tx_desc->upper.data = cpu_to_le32(txd_upper);
3051 if (i == tx_ring->count)
3055 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3057 /* Force memory writes to complete before letting h/w
3058 * know there are new descriptors to fetch. (Only
3059 * applicable for weak-ordered memory model archs,
3060 * such as IA-64). */
3063 tx_ring->next_to_use = i;
3064 writel(i, adapter->hw.hw_addr + tx_ring->tail);
3065 /* we need this if more than one processor can write to our tail
3066 * at a time, it synchronizes IO on IA64/Altix systems */
3070 #define MINIMUM_DHCP_PACKET_SIZE 282
3071 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3072 struct sk_buff *skb)
3074 struct e1000_hw *hw = &adapter->hw;
3077 if (vlan_tx_tag_present(skb)) {
3078 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3079 && (adapter->hw.mng_cookie.status &
3080 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3084 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3087 if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3091 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3094 if (ip->protocol != IPPROTO_UDP)
3097 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3098 if (ntohs(udp->dest) != 67)
3101 offset = (u8 *)udp + 8 - skb->data;
3102 length = skb->len - offset;
3103 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
3109 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3111 struct e1000_adapter *adapter = netdev_priv(netdev);
3113 netif_stop_queue(netdev);
3114 /* Herbert's original patch had:
3115 * smp_mb__after_netif_stop_queue();
3116 * but since that doesn't exist yet, just open code it. */
3119 /* We need to check again in a case another CPU has just
3120 * made room available. */
3121 if (e1000_desc_unused(adapter->tx_ring) < size)
3125 netif_start_queue(netdev);
3126 ++adapter->restart_queue;
3130 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
3132 struct e1000_adapter *adapter = netdev_priv(netdev);
3134 if (e1000_desc_unused(adapter->tx_ring) >= size)
3136 return __e1000_maybe_stop_tx(netdev, size);
3139 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3140 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3142 struct e1000_adapter *adapter = netdev_priv(netdev);
3143 struct e1000_ring *tx_ring = adapter->tx_ring;
3145 unsigned int max_per_txd = E1000_MAX_PER_TXD;
3146 unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3147 unsigned int tx_flags = 0;
3148 unsigned int len = skb->len - skb->data_len;
3149 unsigned long irq_flags;
3150 unsigned int nr_frags;
3156 if (test_bit(__E1000_DOWN, &adapter->state)) {
3157 dev_kfree_skb_any(skb);
3158 return NETDEV_TX_OK;
3161 if (skb->len <= 0) {
3162 dev_kfree_skb_any(skb);
3163 return NETDEV_TX_OK;
3166 mss = skb_shinfo(skb)->gso_size;
3167 /* The controller does a simple calculation to
3168 * make sure there is enough room in the FIFO before
3169 * initiating the DMA for each buffer. The calc is:
3170 * 4 = ceil(buffer len/mss). To make sure we don't
3171 * overrun the FIFO, adjust the max buffer len if mss
3175 max_per_txd = min(mss << 2, max_per_txd);
3176 max_txd_pwr = fls(max_per_txd) - 1;
3178 /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
3179 * points to just header, pull a few bytes of payload from
3180 * frags into skb->data */
3181 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3182 if (skb->data_len && (hdr_len == len)) {
3183 unsigned int pull_size;
3185 pull_size = min((unsigned int)4, skb->data_len);
3186 if (!__pskb_pull_tail(skb, pull_size)) {
3188 "__pskb_pull_tail failed.\n");
3189 dev_kfree_skb_any(skb);
3190 return NETDEV_TX_OK;
3192 len = skb->len - skb->data_len;
3196 /* reserve a descriptor for the offload context */
3197 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3201 count += TXD_USE_COUNT(len, max_txd_pwr);
3203 nr_frags = skb_shinfo(skb)->nr_frags;
3204 for (f = 0; f < nr_frags; f++)
3205 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3208 if (adapter->hw.mac.tx_pkt_filtering)
3209 e1000_transfer_dhcp_info(adapter, skb);
3211 if (!spin_trylock_irqsave(&adapter->tx_queue_lock, irq_flags))
3212 /* Collision - tell upper layer to requeue */
3213 return NETDEV_TX_LOCKED;
3215 /* need: count + 2 desc gap to keep tail from touching
3216 * head, otherwise try next time */
3217 if (e1000_maybe_stop_tx(netdev, count + 2)) {
3218 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3219 return NETDEV_TX_BUSY;
3222 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3223 tx_flags |= E1000_TX_FLAGS_VLAN;
3224 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3227 first = tx_ring->next_to_use;
3229 tso = e1000_tso(adapter, skb);
3231 dev_kfree_skb_any(skb);
3232 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3233 return NETDEV_TX_OK;
3237 tx_flags |= E1000_TX_FLAGS_TSO;
3238 else if (e1000_tx_csum(adapter, skb))
3239 tx_flags |= E1000_TX_FLAGS_CSUM;
3241 /* Old method was to assume IPv4 packet by default if TSO was enabled.
3242 * 82571 hardware supports TSO capabilities for IPv6 as well...
3243 * no longer assume, we must. */
3244 if (skb->protocol == htons(ETH_P_IP))
3245 tx_flags |= E1000_TX_FLAGS_IPV4;
3247 count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
3249 /* handle pci_map_single() error in e1000_tx_map */
3250 dev_kfree_skb_any(skb);
3251 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3252 return NETDEV_TX_OK;
3255 e1000_tx_queue(adapter, tx_flags, count);
3257 netdev->trans_start = jiffies;
3259 /* Make sure there is space in the ring for the next send. */
3260 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
3262 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3263 return NETDEV_TX_OK;
3267 * e1000_tx_timeout - Respond to a Tx Hang
3268 * @netdev: network interface device structure
3270 static void e1000_tx_timeout(struct net_device *netdev)
3272 struct e1000_adapter *adapter = netdev_priv(netdev);
3274 /* Do the reset outside of interrupt context */
3275 adapter->tx_timeout_count++;
3276 schedule_work(&adapter->reset_task);
3279 static void e1000_reset_task(struct work_struct *work)
3281 struct e1000_adapter *adapter;
3282 adapter = container_of(work, struct e1000_adapter, reset_task);
3284 e1000e_reinit_locked(adapter);
3288 * e1000_get_stats - Get System Network Statistics
3289 * @netdev: network interface device structure
3291 * Returns the address of the device statistics structure.
3292 * The statistics are actually updated from the timer callback.
3294 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3296 struct e1000_adapter *adapter = netdev_priv(netdev);
3298 /* only return the current stats */
3299 return &adapter->net_stats;
3303 * e1000_change_mtu - Change the Maximum Transfer Unit
3304 * @netdev: network interface device structure
3305 * @new_mtu: new value for maximum frame size
3307 * Returns 0 on success, negative on failure
3309 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3311 struct e1000_adapter *adapter = netdev_priv(netdev);
3312 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3314 if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
3315 (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3316 ndev_err(netdev, "Invalid MTU setting\n");
3320 /* Jumbo frame size limits */
3321 if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
3322 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
3323 ndev_err(netdev, "Jumbo Frames not supported.\n");
3326 if (adapter->hw.phy.type == e1000_phy_ife) {
3327 ndev_err(netdev, "Jumbo Frames not supported.\n");
3332 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3333 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3334 ndev_err(netdev, "MTU > 9216 not supported.\n");
3338 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3340 /* e1000e_down has a dependency on max_frame_size */
3341 adapter->hw.mac.max_frame_size = max_frame;
3342 if (netif_running(netdev))
3343 e1000e_down(adapter);
3345 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3346 * means we reserve 2 more, this pushes us to allocate from the next
3348 * i.e. RXBUFFER_2048 --> size-4096 slab */
3350 if (max_frame <= 256)
3351 adapter->rx_buffer_len = 256;
3352 else if (max_frame <= 512)
3353 adapter->rx_buffer_len = 512;
3354 else if (max_frame <= 1024)
3355 adapter->rx_buffer_len = 1024;
3356 else if (max_frame <= 2048)
3357 adapter->rx_buffer_len = 2048;
3359 adapter->rx_buffer_len = 4096;
3361 /* adjust allocation if LPE protects us, and we aren't using SBP */
3362 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3363 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
3364 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
3367 ndev_info(netdev, "changing MTU from %d to %d\n",
3368 netdev->mtu, new_mtu);
3369 netdev->mtu = new_mtu;
3371 if (netif_running(netdev))
3374 e1000e_reset(adapter);
3376 clear_bit(__E1000_RESETTING, &adapter->state);
3381 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
3384 struct e1000_adapter *adapter = netdev_priv(netdev);
3385 struct mii_ioctl_data *data = if_mii(ifr);
3386 unsigned long irq_flags;
3388 if (adapter->hw.media_type != e1000_media_type_copper)
3393 data->phy_id = adapter->hw.phy.addr;
3396 if (!capable(CAP_NET_ADMIN))
3398 spin_lock_irqsave(&adapter->stats_lock, irq_flags);
3399 if (e1e_rphy(&adapter->hw, data->reg_num & 0x1F,
3401 spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
3404 spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
3413 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
3419 return e1000_mii_ioctl(netdev, ifr, cmd);
3425 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
3427 struct net_device *netdev = pci_get_drvdata(pdev);
3428 struct e1000_adapter *adapter = netdev_priv(netdev);
3429 struct e1000_hw *hw = &adapter->hw;
3430 u32 ctrl, ctrl_ext, rctl, status;
3431 u32 wufc = adapter->wol;
3434 netif_device_detach(netdev);
3436 if (netif_running(netdev)) {
3437 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3438 e1000e_down(adapter);
3439 e1000_free_irq(adapter);
3442 retval = pci_save_state(pdev);
3446 status = er32(STATUS);
3447 if (status & E1000_STATUS_LU)
3448 wufc &= ~E1000_WUFC_LNKC;
3451 e1000_setup_rctl(adapter);
3452 e1000_set_multi(netdev);
3454 /* turn on all-multi mode if wake on multicast is enabled */
3455 if (wufc & E1000_WUFC_MC) {
3457 rctl |= E1000_RCTL_MPE;
3462 /* advertise wake from D3Cold */
3463 #define E1000_CTRL_ADVD3WUC 0x00100000
3464 /* phy power management enable */
3465 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
3466 ctrl |= E1000_CTRL_ADVD3WUC |
3467 E1000_CTRL_EN_PHY_PWR_MGMT;
3470 if (adapter->hw.media_type == e1000_media_type_fiber ||
3471 adapter->hw.media_type == e1000_media_type_internal_serdes) {
3472 /* keep the laser running in D3 */
3473 ctrl_ext = er32(CTRL_EXT);
3474 ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
3475 ew32(CTRL_EXT, ctrl_ext);
3478 /* Allow time for pending master requests to run */
3479 e1000e_disable_pcie_master(&adapter->hw);
3481 ew32(WUC, E1000_WUC_PME_EN);
3483 pci_enable_wake(pdev, PCI_D3hot, 1);
3484 pci_enable_wake(pdev, PCI_D3cold, 1);
3488 pci_enable_wake(pdev, PCI_D3hot, 0);
3489 pci_enable_wake(pdev, PCI_D3cold, 0);
3492 e1000_release_manageability(adapter);
3494 /* make sure adapter isn't asleep if manageability is enabled */
3495 if (adapter->flags & FLAG_MNG_PT_ENABLED) {
3496 pci_enable_wake(pdev, PCI_D3hot, 1);
3497 pci_enable_wake(pdev, PCI_D3cold, 1);
3500 if (adapter->hw.phy.type == e1000_phy_igp_3)
3501 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
3503 /* Release control of h/w to f/w. If f/w is AMT enabled, this
3504 * would have already happened in close and is redundant. */
3505 e1000_release_hw_control(adapter);
3507 pci_disable_device(pdev);
3509 pci_set_power_state(pdev, pci_choose_state(pdev, state));
3514 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
3521 * 82573 workaround - disable L1 ASPM on mobile chipsets
3523 * L1 ASPM on various mobile (ich7) chipsets do not behave properly
3524 * resulting in lost data or garbage information on the pci-e link
3525 * level. This could result in (false) bad EEPROM checksum errors,
3526 * long ping times (up to 2s) or even a system freeze/hang.
3528 * Unfortunately this feature saves about 1W power consumption when
3531 pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
3532 pci_read_config_dword(pdev, pos + PCI_EXP_LNKCAP, &cap);
3533 pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
3535 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
3537 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
3542 static int e1000_resume(struct pci_dev *pdev)
3544 struct net_device *netdev = pci_get_drvdata(pdev);
3545 struct e1000_adapter *adapter = netdev_priv(netdev);
3546 struct e1000_hw *hw = &adapter->hw;
3549 pci_set_power_state(pdev, PCI_D0);
3550 pci_restore_state(pdev);
3551 e1000e_disable_l1aspm(pdev);
3552 err = pci_enable_device(pdev);
3555 "Cannot enable PCI device from suspend\n");
3559 pci_set_master(pdev);
3561 pci_enable_wake(pdev, PCI_D3hot, 0);
3562 pci_enable_wake(pdev, PCI_D3cold, 0);
3564 if (netif_running(netdev)) {
3565 err = e1000_request_irq(adapter);
3570 e1000e_power_up_phy(adapter);
3571 e1000e_reset(adapter);
3574 e1000_init_manageability(adapter);
3576 if (netif_running(netdev))
3579 netif_device_attach(netdev);
3581 /* If the controller has AMT, do not set DRV_LOAD until the interface
3582 * is up. For all other cases, let the f/w know that the h/w is now
3583 * under the control of the driver. */
3584 if (!(adapter->flags & FLAG_HAS_AMT) || !e1000e_check_mng_mode(&adapter->hw))
3585 e1000_get_hw_control(adapter);
3591 static void e1000_shutdown(struct pci_dev *pdev)
3593 e1000_suspend(pdev, PMSG_SUSPEND);
3596 #ifdef CONFIG_NET_POLL_CONTROLLER
3598 * Polling 'interrupt' - used by things like netconsole to send skbs
3599 * without having to re-enable interrupts. It's not called while
3600 * the interrupt routine is executing.
3602 static void e1000_netpoll(struct net_device *netdev)
3604 struct e1000_adapter *adapter = netdev_priv(netdev);
3606 disable_irq(adapter->pdev->irq);
3607 e1000_intr(adapter->pdev->irq, netdev);
3609 e1000_clean_tx_irq(adapter);
3611 enable_irq(adapter->pdev->irq);
3616 * e1000_io_error_detected - called when PCI error is detected
3617 * @pdev: Pointer to PCI device
3618 * @state: The current pci connection state
3620 * This function is called after a PCI bus error affecting
3621 * this device has been detected.
3623 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
3624 pci_channel_state_t state)
3626 struct net_device *netdev = pci_get_drvdata(pdev);
3627 struct e1000_adapter *adapter = netdev_priv(netdev);
3629 netif_device_detach(netdev);
3631 if (netif_running(netdev))
3632 e1000e_down(adapter);
3633 pci_disable_device(pdev);
3635 /* Request a slot slot reset. */
3636 return PCI_ERS_RESULT_NEED_RESET;
3640 * e1000_io_slot_reset - called after the pci bus has been reset.
3641 * @pdev: Pointer to PCI device
3643 * Restart the card from scratch, as if from a cold-boot. Implementation
3644 * resembles the first-half of the e1000_resume routine.
3646 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
3648 struct net_device *netdev = pci_get_drvdata(pdev);
3649 struct e1000_adapter *adapter = netdev_priv(netdev);
3650 struct e1000_hw *hw = &adapter->hw;
3652 e1000e_disable_l1aspm(pdev);
3653 if (pci_enable_device(pdev)) {
3655 "Cannot re-enable PCI device after reset.\n");
3656 return PCI_ERS_RESULT_DISCONNECT;
3658 pci_set_master(pdev);
3660 pci_enable_wake(pdev, PCI_D3hot, 0);
3661 pci_enable_wake(pdev, PCI_D3cold, 0);
3663 e1000e_reset(adapter);
3666 return PCI_ERS_RESULT_RECOVERED;
3670 * e1000_io_resume - called when traffic can start flowing again.
3671 * @pdev: Pointer to PCI device
3673 * This callback is called when the error recovery driver tells us that
3674 * its OK to resume normal operation. Implementation resembles the
3675 * second-half of the e1000_resume routine.
3677 static void e1000_io_resume(struct pci_dev *pdev)
3679 struct net_device *netdev = pci_get_drvdata(pdev);
3680 struct e1000_adapter *adapter = netdev_priv(netdev);
3682 e1000_init_manageability(adapter);
3684 if (netif_running(netdev)) {
3685 if (e1000e_up(adapter)) {
3687 "can't bring device back up after reset\n");
3692 netif_device_attach(netdev);
3694 /* If the controller has AMT, do not set DRV_LOAD until the interface
3695 * is up. For all other cases, let the f/w know that the h/w is now
3696 * under the control of the driver. */
3697 if (!(adapter->flags & FLAG_HAS_AMT) ||
3698 !e1000e_check_mng_mode(&adapter->hw))
3699 e1000_get_hw_control(adapter);
3703 static void e1000_print_device_info(struct e1000_adapter *adapter)
3705 struct e1000_hw *hw = &adapter->hw;
3706 struct net_device *netdev = adapter->netdev;
3709 /* print bus type/speed/width info */
3710 ndev_info(netdev, "(PCI Express:2.5GB/s:%s) "
3711 "%02x:%02x:%02x:%02x:%02x:%02x\n",
3713 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
3716 netdev->dev_addr[0], netdev->dev_addr[1],
3717 netdev->dev_addr[2], netdev->dev_addr[3],
3718 netdev->dev_addr[4], netdev->dev_addr[5]);
3719 ndev_info(netdev, "Intel(R) PRO/%s Network Connection\n",
3720 (hw->phy.type == e1000_phy_ife)
3721 ? "10/100" : "1000");
3722 e1000e_read_part_num(hw, &part_num);
3723 ndev_info(netdev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
3724 hw->mac.type, hw->phy.type,
3725 (part_num >> 8), (part_num & 0xff));
3729 * e1000_probe - Device Initialization Routine
3730 * @pdev: PCI device information struct
3731 * @ent: entry in e1000_pci_tbl
3733 * Returns 0 on success, negative on failure
3735 * e1000_probe initializes an adapter identified by a pci_dev structure.
3736 * The OS initialization, configuring of the adapter private structure,
3737 * and a hardware reset occur.
3739 static int __devinit e1000_probe(struct pci_dev *pdev,
3740 const struct pci_device_id *ent)
3742 struct net_device *netdev;
3743 struct e1000_adapter *adapter;
3744 struct e1000_hw *hw;
3745 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
3746 unsigned long mmio_start, mmio_len;
3747 unsigned long flash_start, flash_len;
3749 static int cards_found;
3750 int i, err, pci_using_dac;
3751 u16 eeprom_data = 0;
3752 u16 eeprom_apme_mask = E1000_EEPROM_APME;
3754 e1000e_disable_l1aspm(pdev);
3755 err = pci_enable_device(pdev);
3760 err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
3762 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
3766 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
3768 err = pci_set_consistent_dma_mask(pdev,
3771 dev_err(&pdev->dev, "No usable DMA "
3772 "configuration, aborting\n");
3778 err = pci_request_regions(pdev, e1000e_driver_name);
3782 pci_set_master(pdev);
3785 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
3787 goto err_alloc_etherdev;
3789 SET_NETDEV_DEV(netdev, &pdev->dev);
3791 pci_set_drvdata(pdev, netdev);
3792 adapter = netdev_priv(netdev);
3794 adapter->netdev = netdev;
3795 adapter->pdev = pdev;
3797 adapter->pba = ei->pba;
3798 adapter->flags = ei->flags;
3799 adapter->hw.adapter = adapter;
3800 adapter->hw.mac.type = ei->mac;
3801 adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
3803 mmio_start = pci_resource_start(pdev, 0);
3804 mmio_len = pci_resource_len(pdev, 0);
3807 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
3808 if (!adapter->hw.hw_addr)
3811 if ((adapter->flags & FLAG_HAS_FLASH) &&
3812 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
3813 flash_start = pci_resource_start(pdev, 1);
3814 flash_len = pci_resource_len(pdev, 1);
3815 adapter->hw.flash_address = ioremap(flash_start, flash_len);
3816 if (!adapter->hw.flash_address)
3820 /* construct the net_device struct */
3821 netdev->open = &e1000_open;
3822 netdev->stop = &e1000_close;
3823 netdev->hard_start_xmit = &e1000_xmit_frame;
3824 netdev->get_stats = &e1000_get_stats;
3825 netdev->set_multicast_list = &e1000_set_multi;
3826 netdev->set_mac_address = &e1000_set_mac;
3827 netdev->change_mtu = &e1000_change_mtu;
3828 netdev->do_ioctl = &e1000_ioctl;
3829 e1000e_set_ethtool_ops(netdev);
3830 netdev->tx_timeout = &e1000_tx_timeout;
3831 netdev->watchdog_timeo = 5 * HZ;
3832 netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
3833 netdev->vlan_rx_register = e1000_vlan_rx_register;
3834 netdev->vlan_rx_add_vid = e1000_vlan_rx_add_vid;
3835 netdev->vlan_rx_kill_vid = e1000_vlan_rx_kill_vid;
3836 #ifdef CONFIG_NET_POLL_CONTROLLER
3837 netdev->poll_controller = e1000_netpoll;
3839 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
3841 netdev->mem_start = mmio_start;
3842 netdev->mem_end = mmio_start + mmio_len;
3844 adapter->bd_number = cards_found++;
3846 /* setup adapter struct */
3847 err = e1000_sw_init(adapter);
3853 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
3854 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
3855 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
3857 err = ei->get_invariants(adapter);
3861 hw->mac.ops.get_bus_info(&adapter->hw);
3863 adapter->hw.phy.wait_for_link = 0;
3865 /* Copper options */
3866 if (adapter->hw.media_type == e1000_media_type_copper) {
3867 adapter->hw.phy.mdix = AUTO_ALL_MODES;
3868 adapter->hw.phy.disable_polarity_correction = 0;
3869 adapter->hw.phy.ms_type = e1000_ms_hw_default;
3872 if (e1000_check_reset_block(&adapter->hw))
3874 "PHY reset is blocked due to SOL/IDER session.\n");
3876 netdev->features = NETIF_F_SG |
3878 NETIF_F_HW_VLAN_TX |
3881 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
3882 netdev->features |= NETIF_F_HW_VLAN_FILTER;
3884 netdev->features |= NETIF_F_TSO;
3885 netdev->features |= NETIF_F_TSO6;
3888 netdev->features |= NETIF_F_HIGHDMA;
3890 /* We should not be using LLTX anymore, but we are still TX faster with
3892 netdev->features |= NETIF_F_LLTX;
3894 if (e1000e_enable_mng_pass_thru(&adapter->hw))
3895 adapter->flags |= FLAG_MNG_PT_ENABLED;
3897 /* before reading the NVM, reset the controller to
3898 * put the device in a known good starting state */
3899 adapter->hw.mac.ops.reset_hw(&adapter->hw);
3902 * systems with ASPM and others may see the checksum fail on the first
3903 * attempt. Let's give it a few tries
3906 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
3909 ndev_err(netdev, "The NVM Checksum Is Not Valid\n");
3915 /* copy the MAC address out of the NVM */
3916 if (e1000e_read_mac_addr(&adapter->hw))
3917 ndev_err(netdev, "NVM Read Error while reading MAC address\n");
3919 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
3920 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
3922 if (!is_valid_ether_addr(netdev->perm_addr)) {
3923 ndev_err(netdev, "Invalid MAC Address: "
3924 "%02x:%02x:%02x:%02x:%02x:%02x\n",
3925 netdev->perm_addr[0], netdev->perm_addr[1],
3926 netdev->perm_addr[2], netdev->perm_addr[3],
3927 netdev->perm_addr[4], netdev->perm_addr[5]);
3932 init_timer(&adapter->watchdog_timer);
3933 adapter->watchdog_timer.function = &e1000_watchdog;
3934 adapter->watchdog_timer.data = (unsigned long) adapter;
3936 init_timer(&adapter->phy_info_timer);
3937 adapter->phy_info_timer.function = &e1000_update_phy_info;
3938 adapter->phy_info_timer.data = (unsigned long) adapter;
3940 INIT_WORK(&adapter->reset_task, e1000_reset_task);
3941 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
3943 e1000e_check_options(adapter);
3945 /* Initialize link parameters. User can change them with ethtool */
3946 adapter->hw.mac.autoneg = 1;
3947 adapter->fc_autoneg = 1;
3948 adapter->hw.mac.original_fc = e1000_fc_default;
3949 adapter->hw.mac.fc = e1000_fc_default;
3950 adapter->hw.phy.autoneg_advertised = 0x2f;
3952 /* ring size defaults */
3953 adapter->rx_ring->count = 256;
3954 adapter->tx_ring->count = 256;
3957 * Initial Wake on LAN setting - If APM wake is enabled in
3958 * the EEPROM, enable the ACPI Magic Packet filter
3960 if (adapter->flags & FLAG_APME_IN_WUC) {
3961 /* APME bit in EEPROM is mapped to WUC.APME */
3962 eeprom_data = er32(WUC);
3963 eeprom_apme_mask = E1000_WUC_APME;
3964 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
3965 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
3966 (adapter->hw.bus.func == 1))
3967 e1000_read_nvm(&adapter->hw,
3968 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
3970 e1000_read_nvm(&adapter->hw,
3971 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
3974 /* fetch WoL from EEPROM */
3975 if (eeprom_data & eeprom_apme_mask)
3976 adapter->eeprom_wol |= E1000_WUFC_MAG;
3979 * now that we have the eeprom settings, apply the special cases
3980 * where the eeprom may be wrong or the board simply won't support
3981 * wake on lan on a particular port
3983 if (!(adapter->flags & FLAG_HAS_WOL))
3984 adapter->eeprom_wol = 0;
3986 /* initialize the wol settings based on the eeprom settings */
3987 adapter->wol = adapter->eeprom_wol;
3989 /* reset the hardware with the new settings */
3990 e1000e_reset(adapter);
3992 /* If the controller has AMT, do not set DRV_LOAD until the interface
3993 * is up. For all other cases, let the f/w know that the h/w is now
3994 * under the control of the driver. */
3995 if (!(adapter->flags & FLAG_HAS_AMT) ||
3996 !e1000e_check_mng_mode(&adapter->hw))
3997 e1000_get_hw_control(adapter);
3999 /* tell the stack to leave us alone until e1000_open() is called */
4000 netif_carrier_off(netdev);
4001 netif_stop_queue(netdev);
4003 strcpy(netdev->name, "eth%d");
4004 err = register_netdev(netdev);
4008 e1000_print_device_info(adapter);
4014 e1000_release_hw_control(adapter);
4016 if (!e1000_check_reset_block(&adapter->hw))
4017 e1000_phy_hw_reset(&adapter->hw);
4019 if (adapter->hw.flash_address)
4020 iounmap(adapter->hw.flash_address);
4023 kfree(adapter->tx_ring);
4024 kfree(adapter->rx_ring);
4026 iounmap(adapter->hw.hw_addr);
4028 free_netdev(netdev);
4030 pci_release_regions(pdev);
4033 pci_disable_device(pdev);
4038 * e1000_remove - Device Removal Routine
4039 * @pdev: PCI device information struct
4041 * e1000_remove is called by the PCI subsystem to alert the driver
4042 * that it should release a PCI device. The could be caused by a
4043 * Hot-Plug event, or because the driver is going to be removed from
4046 static void __devexit e1000_remove(struct pci_dev *pdev)
4048 struct net_device *netdev = pci_get_drvdata(pdev);
4049 struct e1000_adapter *adapter = netdev_priv(netdev);
4051 /* flush_scheduled work may reschedule our watchdog task, so
4052 * explicitly disable watchdog tasks from being rescheduled */
4053 set_bit(__E1000_DOWN, &adapter->state);
4054 del_timer_sync(&adapter->watchdog_timer);
4055 del_timer_sync(&adapter->phy_info_timer);
4057 flush_scheduled_work();
4059 e1000_release_manageability(adapter);
4061 /* Release control of h/w to f/w. If f/w is AMT enabled, this
4062 * would have already happened in close and is redundant. */
4063 e1000_release_hw_control(adapter);
4065 unregister_netdev(netdev);
4067 if (!e1000_check_reset_block(&adapter->hw))
4068 e1000_phy_hw_reset(&adapter->hw);
4070 kfree(adapter->tx_ring);
4071 kfree(adapter->rx_ring);
4073 iounmap(adapter->hw.hw_addr);
4074 if (adapter->hw.flash_address)
4075 iounmap(adapter->hw.flash_address);
4076 pci_release_regions(pdev);
4078 free_netdev(netdev);
4080 pci_disable_device(pdev);
4083 /* PCI Error Recovery (ERS) */
4084 static struct pci_error_handlers e1000_err_handler = {
4085 .error_detected = e1000_io_error_detected,
4086 .slot_reset = e1000_io_slot_reset,
4087 .resume = e1000_io_resume,
4090 static struct pci_device_id e1000_pci_tbl[] = {
4091 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
4092 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
4093 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
4094 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
4095 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
4096 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
4097 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
4098 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
4099 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
4100 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
4101 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
4102 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
4103 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
4104 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
4105 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
4106 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
4107 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
4108 board_80003es2lan },
4109 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
4110 board_80003es2lan },
4111 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
4112 board_80003es2lan },
4113 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
4114 board_80003es2lan },
4115 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
4116 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
4117 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
4118 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
4119 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
4120 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
4121 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
4122 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
4123 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
4124 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
4125 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
4126 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
4128 { } /* terminate list */
4130 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
4132 /* PCI Device API Driver */
4133 static struct pci_driver e1000_driver = {
4134 .name = e1000e_driver_name,
4135 .id_table = e1000_pci_tbl,
4136 .probe = e1000_probe,
4137 .remove = __devexit_p(e1000_remove),
4139 /* Power Managment Hooks */
4140 .suspend = e1000_suspend,
4141 .resume = e1000_resume,
4143 .shutdown = e1000_shutdown,
4144 .err_handler = &e1000_err_handler
4148 * e1000_init_module - Driver Registration Routine
4150 * e1000_init_module is the first routine called when the driver is
4151 * loaded. All it does is register with the PCI subsystem.
4153 static int __init e1000_init_module(void)
4156 printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
4157 e1000e_driver_name, e1000e_driver_version);
4158 printk(KERN_INFO "%s: Copyright (c) 1999-2007 Intel Corporation.\n",
4159 e1000e_driver_name);
4160 ret = pci_register_driver(&e1000_driver);
4164 module_init(e1000_init_module);
4167 * e1000_exit_module - Driver Exit Cleanup Routine
4169 * e1000_exit_module is called just before the driver is removed
4172 static void __exit e1000_exit_module(void)
4174 pci_unregister_driver(&e1000_driver);
4176 module_exit(e1000_exit_module);
4179 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
4180 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
4181 MODULE_LICENSE("GPL");
4182 MODULE_VERSION(DRV_VERSION);