1 /*******************************************************************************
3 Intel PRO/1000 Linux driver
4 Copyright(c) 1999 - 2008 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 Linux NICS <linux.nics@intel.com>
24 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
27 *******************************************************************************/
29 #include <linux/module.h>
30 #include <linux/types.h>
31 #include <linux/init.h>
32 #include <linux/pci.h>
33 #include <linux/vmalloc.h>
34 #include <linux/pagemap.h>
35 #include <linux/delay.h>
36 #include <linux/netdevice.h>
37 #include <linux/tcp.h>
38 #include <linux/ipv6.h>
39 #include <net/checksum.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/mii.h>
42 #include <linux/ethtool.h>
43 #include <linux/if_vlan.h>
44 #include <linux/cpu.h>
45 #include <linux/smp.h>
46 #include <linux/pm_qos_params.h>
50 #define DRV_VERSION "0.3.3.3-k2"
51 char e1000e_driver_name[] = "e1000e";
52 const char e1000e_driver_version[] = DRV_VERSION;
54 static const struct e1000_info *e1000_info_tbl[] = {
55 [board_82571] = &e1000_82571_info,
56 [board_82572] = &e1000_82572_info,
57 [board_82573] = &e1000_82573_info,
58 [board_80003es2lan] = &e1000_es2_info,
59 [board_ich8lan] = &e1000_ich8_info,
60 [board_ich9lan] = &e1000_ich9_info,
61 [board_ich10lan] = &e1000_ich10_info,
66 * e1000_get_hw_dev_name - return device name string
67 * used by hardware layer to print debugging information
69 char *e1000e_get_hw_dev_name(struct e1000_hw *hw)
71 return hw->adapter->netdev->name;
76 * e1000_desc_unused - calculate if we have unused descriptors
78 static int e1000_desc_unused(struct e1000_ring *ring)
80 if (ring->next_to_clean > ring->next_to_use)
81 return ring->next_to_clean - ring->next_to_use - 1;
83 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
87 * e1000_receive_skb - helper function to handle Rx indications
88 * @adapter: board private structure
89 * @status: descriptor status field as written by hardware
90 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
91 * @skb: pointer to sk_buff to be indicated to stack
93 static void e1000_receive_skb(struct e1000_adapter *adapter,
94 struct net_device *netdev,
96 u8 status, __le16 vlan)
98 skb->protocol = eth_type_trans(skb, netdev);
100 if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
101 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
104 netif_receive_skb(skb);
106 netdev->last_rx = jiffies;
110 * e1000_rx_checksum - Receive Checksum Offload for 82543
111 * @adapter: board private structure
112 * @status_err: receive descriptor status and error fields
113 * @csum: receive descriptor csum field
114 * @sk_buff: socket buffer with received data
116 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
117 u32 csum, struct sk_buff *skb)
119 u16 status = (u16)status_err;
120 u8 errors = (u8)(status_err >> 24);
121 skb->ip_summed = CHECKSUM_NONE;
123 /* Ignore Checksum bit is set */
124 if (status & E1000_RXD_STAT_IXSM)
126 /* TCP/UDP checksum error bit is set */
127 if (errors & E1000_RXD_ERR_TCPE) {
128 /* let the stack verify checksum errors */
129 adapter->hw_csum_err++;
133 /* TCP/UDP Checksum has not been calculated */
134 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
137 /* It must be a TCP or UDP packet with a valid checksum */
138 if (status & E1000_RXD_STAT_TCPCS) {
139 /* TCP checksum is good */
140 skb->ip_summed = CHECKSUM_UNNECESSARY;
143 * IP fragment with UDP payload
144 * Hardware complements the payload checksum, so we undo it
145 * and then put the value in host order for further stack use.
147 __sum16 sum = (__force __sum16)htons(csum);
148 skb->csum = csum_unfold(~sum);
149 skb->ip_summed = CHECKSUM_COMPLETE;
151 adapter->hw_csum_good++;
155 * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
156 * @adapter: address of board private structure
158 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
161 struct net_device *netdev = adapter->netdev;
162 struct pci_dev *pdev = adapter->pdev;
163 struct e1000_ring *rx_ring = adapter->rx_ring;
164 struct e1000_rx_desc *rx_desc;
165 struct e1000_buffer *buffer_info;
168 unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
170 i = rx_ring->next_to_use;
171 buffer_info = &rx_ring->buffer_info[i];
173 while (cleaned_count--) {
174 skb = buffer_info->skb;
180 skb = netdev_alloc_skb(netdev, bufsz);
182 /* Better luck next round */
183 adapter->alloc_rx_buff_failed++;
188 * Make buffer alignment 2 beyond a 16 byte boundary
189 * this will result in a 16 byte aligned IP header after
190 * the 14 byte MAC header is removed
192 skb_reserve(skb, NET_IP_ALIGN);
194 buffer_info->skb = skb;
196 buffer_info->dma = pci_map_single(pdev, skb->data,
197 adapter->rx_buffer_len,
199 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
200 dev_err(&pdev->dev, "RX DMA map failed\n");
201 adapter->rx_dma_failed++;
205 rx_desc = E1000_RX_DESC(*rx_ring, i);
206 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
209 if (i == rx_ring->count)
211 buffer_info = &rx_ring->buffer_info[i];
214 if (rx_ring->next_to_use != i) {
215 rx_ring->next_to_use = i;
217 i = (rx_ring->count - 1);
220 * Force memory writes to complete before letting h/w
221 * know there are new descriptors to fetch. (Only
222 * applicable for weak-ordered memory model archs,
226 writel(i, adapter->hw.hw_addr + rx_ring->tail);
231 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
232 * @adapter: address of board private structure
234 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
237 struct net_device *netdev = adapter->netdev;
238 struct pci_dev *pdev = adapter->pdev;
239 union e1000_rx_desc_packet_split *rx_desc;
240 struct e1000_ring *rx_ring = adapter->rx_ring;
241 struct e1000_buffer *buffer_info;
242 struct e1000_ps_page *ps_page;
246 i = rx_ring->next_to_use;
247 buffer_info = &rx_ring->buffer_info[i];
249 while (cleaned_count--) {
250 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
252 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
253 ps_page = &buffer_info->ps_pages[j];
254 if (j >= adapter->rx_ps_pages) {
255 /* all unused desc entries get hw null ptr */
256 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
259 if (!ps_page->page) {
260 ps_page->page = alloc_page(GFP_ATOMIC);
261 if (!ps_page->page) {
262 adapter->alloc_rx_buff_failed++;
265 ps_page->dma = pci_map_page(pdev,
269 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
270 dev_err(&adapter->pdev->dev,
271 "RX DMA page map failed\n");
272 adapter->rx_dma_failed++;
277 * Refresh the desc even if buffer_addrs
278 * didn't change because each write-back
281 rx_desc->read.buffer_addr[j+1] =
282 cpu_to_le64(ps_page->dma);
285 skb = netdev_alloc_skb(netdev,
286 adapter->rx_ps_bsize0 + NET_IP_ALIGN);
289 adapter->alloc_rx_buff_failed++;
294 * Make buffer alignment 2 beyond a 16 byte boundary
295 * this will result in a 16 byte aligned IP header after
296 * the 14 byte MAC header is removed
298 skb_reserve(skb, NET_IP_ALIGN);
300 buffer_info->skb = skb;
301 buffer_info->dma = pci_map_single(pdev, skb->data,
302 adapter->rx_ps_bsize0,
304 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
305 dev_err(&pdev->dev, "RX DMA map failed\n");
306 adapter->rx_dma_failed++;
308 dev_kfree_skb_any(skb);
309 buffer_info->skb = NULL;
313 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
316 if (i == rx_ring->count)
318 buffer_info = &rx_ring->buffer_info[i];
322 if (rx_ring->next_to_use != i) {
323 rx_ring->next_to_use = i;
326 i = (rx_ring->count - 1);
329 * Force memory writes to complete before letting h/w
330 * know there are new descriptors to fetch. (Only
331 * applicable for weak-ordered memory model archs,
336 * Hardware increments by 16 bytes, but packet split
337 * descriptors are 32 bytes...so we increment tail
340 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
345 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
346 * @adapter: address of board private structure
347 * @rx_ring: pointer to receive ring structure
348 * @cleaned_count: number of buffers to allocate this pass
351 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
354 struct net_device *netdev = adapter->netdev;
355 struct pci_dev *pdev = adapter->pdev;
356 struct e1000_rx_desc *rx_desc;
357 struct e1000_ring *rx_ring = adapter->rx_ring;
358 struct e1000_buffer *buffer_info;
361 unsigned int bufsz = 256 -
362 16 /* for skb_reserve */ -
365 i = rx_ring->next_to_use;
366 buffer_info = &rx_ring->buffer_info[i];
368 while (cleaned_count--) {
369 skb = buffer_info->skb;
375 skb = netdev_alloc_skb(netdev, bufsz);
376 if (unlikely(!skb)) {
377 /* Better luck next round */
378 adapter->alloc_rx_buff_failed++;
382 /* Make buffer alignment 2 beyond a 16 byte boundary
383 * this will result in a 16 byte aligned IP header after
384 * the 14 byte MAC header is removed
386 skb_reserve(skb, NET_IP_ALIGN);
388 buffer_info->skb = skb;
390 /* allocate a new page if necessary */
391 if (!buffer_info->page) {
392 buffer_info->page = alloc_page(GFP_ATOMIC);
393 if (unlikely(!buffer_info->page)) {
394 adapter->alloc_rx_buff_failed++;
399 if (!buffer_info->dma)
400 buffer_info->dma = pci_map_page(pdev,
401 buffer_info->page, 0,
405 rx_desc = E1000_RX_DESC(*rx_ring, i);
406 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
408 if (unlikely(++i == rx_ring->count))
410 buffer_info = &rx_ring->buffer_info[i];
413 if (likely(rx_ring->next_to_use != i)) {
414 rx_ring->next_to_use = i;
415 if (unlikely(i-- == 0))
416 i = (rx_ring->count - 1);
418 /* Force memory writes to complete before letting h/w
419 * know there are new descriptors to fetch. (Only
420 * applicable for weak-ordered memory model archs,
423 writel(i, adapter->hw.hw_addr + rx_ring->tail);
428 * e1000_clean_rx_irq - Send received data up the network stack; legacy
429 * @adapter: board private structure
431 * the return value indicates whether actual cleaning was done, there
432 * is no guarantee that everything was cleaned
434 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
435 int *work_done, int work_to_do)
437 struct net_device *netdev = adapter->netdev;
438 struct pci_dev *pdev = adapter->pdev;
439 struct e1000_ring *rx_ring = adapter->rx_ring;
440 struct e1000_rx_desc *rx_desc, *next_rxd;
441 struct e1000_buffer *buffer_info, *next_buffer;
444 int cleaned_count = 0;
446 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
448 i = rx_ring->next_to_clean;
449 rx_desc = E1000_RX_DESC(*rx_ring, i);
450 buffer_info = &rx_ring->buffer_info[i];
452 while (rx_desc->status & E1000_RXD_STAT_DD) {
456 if (*work_done >= work_to_do)
460 status = rx_desc->status;
461 skb = buffer_info->skb;
462 buffer_info->skb = NULL;
464 prefetch(skb->data - NET_IP_ALIGN);
467 if (i == rx_ring->count)
469 next_rxd = E1000_RX_DESC(*rx_ring, i);
472 next_buffer = &rx_ring->buffer_info[i];
476 pci_unmap_single(pdev,
478 adapter->rx_buffer_len,
480 buffer_info->dma = 0;
482 length = le16_to_cpu(rx_desc->length);
484 /* !EOP means multiple descriptors were used to store a single
485 * packet, also make sure the frame isn't just CRC only */
486 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
487 /* All receives must fit into a single buffer */
488 e_dbg("%s: Receive packet consumed multiple buffers\n",
491 buffer_info->skb = skb;
495 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
497 buffer_info->skb = skb;
501 total_rx_bytes += length;
505 * code added for copybreak, this should improve
506 * performance for small packets with large amounts
507 * of reassembly being done in the stack
509 if (length < copybreak) {
510 struct sk_buff *new_skb =
511 netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
513 skb_reserve(new_skb, NET_IP_ALIGN);
514 skb_copy_to_linear_data_offset(new_skb,
520 /* save the skb in buffer_info as good */
521 buffer_info->skb = skb;
524 /* else just continue with the old one */
526 /* end copybreak code */
527 skb_put(skb, length);
529 /* Receive Checksum Offload */
530 e1000_rx_checksum(adapter,
532 ((u32)(rx_desc->errors) << 24),
533 le16_to_cpu(rx_desc->csum), skb);
535 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
540 /* return some buffers to hardware, one at a time is too slow */
541 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
542 adapter->alloc_rx_buf(adapter, cleaned_count);
546 /* use prefetched values */
548 buffer_info = next_buffer;
550 rx_ring->next_to_clean = i;
552 cleaned_count = e1000_desc_unused(rx_ring);
554 adapter->alloc_rx_buf(adapter, cleaned_count);
556 adapter->total_rx_bytes += total_rx_bytes;
557 adapter->total_rx_packets += total_rx_packets;
558 adapter->net_stats.rx_bytes += total_rx_bytes;
559 adapter->net_stats.rx_packets += total_rx_packets;
563 static void e1000_put_txbuf(struct e1000_adapter *adapter,
564 struct e1000_buffer *buffer_info)
566 if (buffer_info->dma) {
567 pci_unmap_page(adapter->pdev, buffer_info->dma,
568 buffer_info->length, PCI_DMA_TODEVICE);
569 buffer_info->dma = 0;
571 if (buffer_info->skb) {
572 dev_kfree_skb_any(buffer_info->skb);
573 buffer_info->skb = NULL;
577 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
579 struct e1000_ring *tx_ring = adapter->tx_ring;
580 unsigned int i = tx_ring->next_to_clean;
581 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
582 struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
584 /* detected Tx unit hang */
585 e_err("Detected Tx Unit Hang:\n"
588 " next_to_use <%x>\n"
589 " next_to_clean <%x>\n"
590 "buffer_info[next_to_clean]:\n"
591 " time_stamp <%lx>\n"
592 " next_to_watch <%x>\n"
594 " next_to_watch.status <%x>\n",
595 readl(adapter->hw.hw_addr + tx_ring->head),
596 readl(adapter->hw.hw_addr + tx_ring->tail),
597 tx_ring->next_to_use,
598 tx_ring->next_to_clean,
599 tx_ring->buffer_info[eop].time_stamp,
602 eop_desc->upper.fields.status);
606 * e1000_clean_tx_irq - Reclaim resources after transmit completes
607 * @adapter: board private structure
609 * the return value indicates whether actual cleaning was done, there
610 * is no guarantee that everything was cleaned
612 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
614 struct net_device *netdev = adapter->netdev;
615 struct e1000_hw *hw = &adapter->hw;
616 struct e1000_ring *tx_ring = adapter->tx_ring;
617 struct e1000_tx_desc *tx_desc, *eop_desc;
618 struct e1000_buffer *buffer_info;
620 unsigned int count = 0;
622 unsigned int total_tx_bytes = 0, total_tx_packets = 0;
624 i = tx_ring->next_to_clean;
625 eop = tx_ring->buffer_info[i].next_to_watch;
626 eop_desc = E1000_TX_DESC(*tx_ring, eop);
628 while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
629 for (cleaned = 0; !cleaned; ) {
630 tx_desc = E1000_TX_DESC(*tx_ring, i);
631 buffer_info = &tx_ring->buffer_info[i];
632 cleaned = (i == eop);
635 struct sk_buff *skb = buffer_info->skb;
636 unsigned int segs, bytecount;
637 segs = skb_shinfo(skb)->gso_segs ?: 1;
638 /* multiply data chunks by size of headers */
639 bytecount = ((segs - 1) * skb_headlen(skb)) +
641 total_tx_packets += segs;
642 total_tx_bytes += bytecount;
645 e1000_put_txbuf(adapter, buffer_info);
646 tx_desc->upper.data = 0;
649 if (i == tx_ring->count)
653 eop = tx_ring->buffer_info[i].next_to_watch;
654 eop_desc = E1000_TX_DESC(*tx_ring, eop);
655 #define E1000_TX_WEIGHT 64
656 /* weight of a sort for tx, to avoid endless transmit cleanup */
657 if (count++ == E1000_TX_WEIGHT)
661 tx_ring->next_to_clean = i;
663 #define TX_WAKE_THRESHOLD 32
664 if (cleaned && netif_carrier_ok(netdev) &&
665 e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
666 /* Make sure that anybody stopping the queue after this
667 * sees the new next_to_clean.
671 if (netif_queue_stopped(netdev) &&
672 !(test_bit(__E1000_DOWN, &adapter->state))) {
673 netif_wake_queue(netdev);
674 ++adapter->restart_queue;
678 if (adapter->detect_tx_hung) {
680 * Detect a transmit hang in hardware, this serializes the
681 * check with the clearing of time_stamp and movement of i
683 adapter->detect_tx_hung = 0;
684 if (tx_ring->buffer_info[eop].dma &&
685 time_after(jiffies, tx_ring->buffer_info[eop].time_stamp
686 + (adapter->tx_timeout_factor * HZ))
687 && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
688 e1000_print_tx_hang(adapter);
689 netif_stop_queue(netdev);
692 adapter->total_tx_bytes += total_tx_bytes;
693 adapter->total_tx_packets += total_tx_packets;
694 adapter->net_stats.tx_bytes += total_tx_bytes;
695 adapter->net_stats.tx_packets += total_tx_packets;
700 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
701 * @adapter: board private structure
703 * the return value indicates whether actual cleaning was done, there
704 * is no guarantee that everything was cleaned
706 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
707 int *work_done, int work_to_do)
709 union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
710 struct net_device *netdev = adapter->netdev;
711 struct pci_dev *pdev = adapter->pdev;
712 struct e1000_ring *rx_ring = adapter->rx_ring;
713 struct e1000_buffer *buffer_info, *next_buffer;
714 struct e1000_ps_page *ps_page;
718 int cleaned_count = 0;
720 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
722 i = rx_ring->next_to_clean;
723 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
724 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
725 buffer_info = &rx_ring->buffer_info[i];
727 while (staterr & E1000_RXD_STAT_DD) {
728 if (*work_done >= work_to_do)
731 skb = buffer_info->skb;
733 /* in the packet split case this is header only */
734 prefetch(skb->data - NET_IP_ALIGN);
737 if (i == rx_ring->count)
739 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
742 next_buffer = &rx_ring->buffer_info[i];
746 pci_unmap_single(pdev, buffer_info->dma,
747 adapter->rx_ps_bsize0,
749 buffer_info->dma = 0;
751 if (!(staterr & E1000_RXD_STAT_EOP)) {
752 e_dbg("%s: Packet Split buffers didn't pick up the "
753 "full packet\n", netdev->name);
754 dev_kfree_skb_irq(skb);
758 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
759 dev_kfree_skb_irq(skb);
763 length = le16_to_cpu(rx_desc->wb.middle.length0);
766 e_dbg("%s: Last part of the packet spanning multiple "
767 "descriptors\n", netdev->name);
768 dev_kfree_skb_irq(skb);
773 skb_put(skb, length);
777 * this looks ugly, but it seems compiler issues make it
778 * more efficient than reusing j
780 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
783 * page alloc/put takes too long and effects small packet
784 * throughput, so unsplit small packets and save the alloc/put
785 * only valid in softirq (napi) context to call kmap_*
787 if (l1 && (l1 <= copybreak) &&
788 ((length + l1) <= adapter->rx_ps_bsize0)) {
791 ps_page = &buffer_info->ps_pages[0];
794 * there is no documentation about how to call
795 * kmap_atomic, so we can't hold the mapping
798 pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
799 PAGE_SIZE, PCI_DMA_FROMDEVICE);
800 vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
801 memcpy(skb_tail_pointer(skb), vaddr, l1);
802 kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
803 pci_dma_sync_single_for_device(pdev, ps_page->dma,
804 PAGE_SIZE, PCI_DMA_FROMDEVICE);
811 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
812 length = le16_to_cpu(rx_desc->wb.upper.length[j]);
816 ps_page = &buffer_info->ps_pages[j];
817 pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
820 skb_fill_page_desc(skb, j, ps_page->page, 0, length);
821 ps_page->page = NULL;
823 skb->data_len += length;
824 skb->truesize += length;
828 total_rx_bytes += skb->len;
831 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
832 rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
834 if (rx_desc->wb.upper.header_status &
835 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
836 adapter->rx_hdr_split++;
838 e1000_receive_skb(adapter, netdev, skb,
839 staterr, rx_desc->wb.middle.vlan);
842 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
843 buffer_info->skb = NULL;
845 /* return some buffers to hardware, one at a time is too slow */
846 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
847 adapter->alloc_rx_buf(adapter, cleaned_count);
851 /* use prefetched values */
853 buffer_info = next_buffer;
855 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
857 rx_ring->next_to_clean = i;
859 cleaned_count = e1000_desc_unused(rx_ring);
861 adapter->alloc_rx_buf(adapter, cleaned_count);
863 adapter->total_rx_bytes += total_rx_bytes;
864 adapter->total_rx_packets += total_rx_packets;
865 adapter->net_stats.rx_bytes += total_rx_bytes;
866 adapter->net_stats.rx_packets += total_rx_packets;
871 * e1000_consume_page - helper function
873 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
878 skb->data_len += length;
879 skb->truesize += length;
883 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
884 * @adapter: board private structure
886 * the return value indicates whether actual cleaning was done, there
887 * is no guarantee that everything was cleaned
890 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
891 int *work_done, int work_to_do)
893 struct net_device *netdev = adapter->netdev;
894 struct pci_dev *pdev = adapter->pdev;
895 struct e1000_ring *rx_ring = adapter->rx_ring;
896 struct e1000_rx_desc *rx_desc, *next_rxd;
897 struct e1000_buffer *buffer_info, *next_buffer;
900 int cleaned_count = 0;
901 bool cleaned = false;
902 unsigned int total_rx_bytes=0, total_rx_packets=0;
904 i = rx_ring->next_to_clean;
905 rx_desc = E1000_RX_DESC(*rx_ring, i);
906 buffer_info = &rx_ring->buffer_info[i];
908 while (rx_desc->status & E1000_RXD_STAT_DD) {
912 if (*work_done >= work_to_do)
916 status = rx_desc->status;
917 skb = buffer_info->skb;
918 buffer_info->skb = NULL;
921 if (i == rx_ring->count)
923 next_rxd = E1000_RX_DESC(*rx_ring, i);
926 next_buffer = &rx_ring->buffer_info[i];
930 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
932 buffer_info->dma = 0;
934 length = le16_to_cpu(rx_desc->length);
936 /* errors is only valid for DD + EOP descriptors */
937 if (unlikely((status & E1000_RXD_STAT_EOP) &&
938 (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
939 /* recycle both page and skb */
940 buffer_info->skb = skb;
941 /* an error means any chain goes out the window
943 if (rx_ring->rx_skb_top)
944 dev_kfree_skb(rx_ring->rx_skb_top);
945 rx_ring->rx_skb_top = NULL;
949 #define rxtop rx_ring->rx_skb_top
950 if (!(status & E1000_RXD_STAT_EOP)) {
951 /* this descriptor is only the beginning (or middle) */
953 /* this is the beginning of a chain */
955 skb_fill_page_desc(rxtop, 0, buffer_info->page,
958 /* this is the middle of a chain */
959 skb_fill_page_desc(rxtop,
960 skb_shinfo(rxtop)->nr_frags,
961 buffer_info->page, 0, length);
962 /* re-use the skb, only consumed the page */
963 buffer_info->skb = skb;
965 e1000_consume_page(buffer_info, rxtop, length);
969 /* end of the chain */
970 skb_fill_page_desc(rxtop,
971 skb_shinfo(rxtop)->nr_frags,
972 buffer_info->page, 0, length);
973 /* re-use the current skb, we only consumed the
975 buffer_info->skb = skb;
978 e1000_consume_page(buffer_info, skb, length);
980 /* no chain, got EOP, this buf is the packet
981 * copybreak to save the put_page/alloc_page */
982 if (length <= copybreak &&
983 skb_tailroom(skb) >= length) {
985 vaddr = kmap_atomic(buffer_info->page,
986 KM_SKB_DATA_SOFTIRQ);
987 memcpy(skb_tail_pointer(skb), vaddr,
990 KM_SKB_DATA_SOFTIRQ);
991 /* re-use the page, so don't erase
992 * buffer_info->page */
993 skb_put(skb, length);
995 skb_fill_page_desc(skb, 0,
996 buffer_info->page, 0,
998 e1000_consume_page(buffer_info, skb,
1004 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1005 e1000_rx_checksum(adapter,
1007 ((u32)(rx_desc->errors) << 24),
1008 le16_to_cpu(rx_desc->csum), skb);
1010 /* probably a little skewed due to removing CRC */
1011 total_rx_bytes += skb->len;
1014 /* eth type trans needs skb->data to point to something */
1015 if (!pskb_may_pull(skb, ETH_HLEN)) {
1016 e_err("pskb_may_pull failed.\n");
1021 e1000_receive_skb(adapter, netdev, skb, status,
1025 rx_desc->status = 0;
1027 /* return some buffers to hardware, one at a time is too slow */
1028 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1029 adapter->alloc_rx_buf(adapter, cleaned_count);
1033 /* use prefetched values */
1035 buffer_info = next_buffer;
1037 rx_ring->next_to_clean = i;
1039 cleaned_count = e1000_desc_unused(rx_ring);
1041 adapter->alloc_rx_buf(adapter, cleaned_count);
1043 adapter->total_rx_bytes += total_rx_bytes;
1044 adapter->total_rx_packets += total_rx_packets;
1045 adapter->net_stats.rx_bytes += total_rx_bytes;
1046 adapter->net_stats.rx_packets += total_rx_packets;
1051 * e1000_clean_rx_ring - Free Rx Buffers per Queue
1052 * @adapter: board private structure
1054 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1056 struct e1000_ring *rx_ring = adapter->rx_ring;
1057 struct e1000_buffer *buffer_info;
1058 struct e1000_ps_page *ps_page;
1059 struct pci_dev *pdev = adapter->pdev;
1062 /* Free all the Rx ring sk_buffs */
1063 for (i = 0; i < rx_ring->count; i++) {
1064 buffer_info = &rx_ring->buffer_info[i];
1065 if (buffer_info->dma) {
1066 if (adapter->clean_rx == e1000_clean_rx_irq)
1067 pci_unmap_single(pdev, buffer_info->dma,
1068 adapter->rx_buffer_len,
1069 PCI_DMA_FROMDEVICE);
1070 else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1071 pci_unmap_page(pdev, buffer_info->dma,
1073 PCI_DMA_FROMDEVICE);
1074 else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1075 pci_unmap_single(pdev, buffer_info->dma,
1076 adapter->rx_ps_bsize0,
1077 PCI_DMA_FROMDEVICE);
1078 buffer_info->dma = 0;
1081 if (buffer_info->page) {
1082 put_page(buffer_info->page);
1083 buffer_info->page = NULL;
1086 if (buffer_info->skb) {
1087 dev_kfree_skb(buffer_info->skb);
1088 buffer_info->skb = NULL;
1091 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1092 ps_page = &buffer_info->ps_pages[j];
1095 pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1096 PCI_DMA_FROMDEVICE);
1098 put_page(ps_page->page);
1099 ps_page->page = NULL;
1103 /* there also may be some cached data from a chained receive */
1104 if (rx_ring->rx_skb_top) {
1105 dev_kfree_skb(rx_ring->rx_skb_top);
1106 rx_ring->rx_skb_top = NULL;
1109 /* Zero out the descriptor ring */
1110 memset(rx_ring->desc, 0, rx_ring->size);
1112 rx_ring->next_to_clean = 0;
1113 rx_ring->next_to_use = 0;
1115 writel(0, adapter->hw.hw_addr + rx_ring->head);
1116 writel(0, adapter->hw.hw_addr + rx_ring->tail);
1120 * e1000_intr_msi - Interrupt Handler
1121 * @irq: interrupt number
1122 * @data: pointer to a network interface device structure
1124 static irqreturn_t e1000_intr_msi(int irq, void *data)
1126 struct net_device *netdev = data;
1127 struct e1000_adapter *adapter = netdev_priv(netdev);
1128 struct e1000_hw *hw = &adapter->hw;
1129 u32 icr = er32(ICR);
1132 * read ICR disables interrupts using IAM
1135 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1136 hw->mac.get_link_status = 1;
1138 * ICH8 workaround-- Call gig speed drop workaround on cable
1139 * disconnect (LSC) before accessing any PHY registers
1141 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1142 (!(er32(STATUS) & E1000_STATUS_LU)))
1143 e1000e_gig_downshift_workaround_ich8lan(hw);
1146 * 80003ES2LAN workaround-- For packet buffer work-around on
1147 * link down event; disable receives here in the ISR and reset
1148 * adapter in watchdog
1150 if (netif_carrier_ok(netdev) &&
1151 adapter->flags & FLAG_RX_NEEDS_RESTART) {
1152 /* disable receives */
1153 u32 rctl = er32(RCTL);
1154 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1155 adapter->flags |= FLAG_RX_RESTART_NOW;
1157 /* guard against interrupt when we're going down */
1158 if (!test_bit(__E1000_DOWN, &adapter->state))
1159 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1162 if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1163 adapter->total_tx_bytes = 0;
1164 adapter->total_tx_packets = 0;
1165 adapter->total_rx_bytes = 0;
1166 adapter->total_rx_packets = 0;
1167 __netif_rx_schedule(netdev, &adapter->napi);
1174 * e1000_intr - Interrupt Handler
1175 * @irq: interrupt number
1176 * @data: pointer to a network interface device structure
1178 static irqreturn_t e1000_intr(int irq, void *data)
1180 struct net_device *netdev = data;
1181 struct e1000_adapter *adapter = netdev_priv(netdev);
1182 struct e1000_hw *hw = &adapter->hw;
1184 u32 rctl, icr = er32(ICR);
1186 return IRQ_NONE; /* Not our interrupt */
1189 * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1190 * not set, then the adapter didn't send an interrupt
1192 if (!(icr & E1000_ICR_INT_ASSERTED))
1196 * Interrupt Auto-Mask...upon reading ICR,
1197 * interrupts are masked. No need for the
1201 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1202 hw->mac.get_link_status = 1;
1204 * ICH8 workaround-- Call gig speed drop workaround on cable
1205 * disconnect (LSC) before accessing any PHY registers
1207 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1208 (!(er32(STATUS) & E1000_STATUS_LU)))
1209 e1000e_gig_downshift_workaround_ich8lan(hw);
1212 * 80003ES2LAN workaround--
1213 * For packet buffer work-around on link down event;
1214 * disable receives here in the ISR and
1215 * reset adapter in watchdog
1217 if (netif_carrier_ok(netdev) &&
1218 (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1219 /* disable receives */
1221 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1222 adapter->flags |= FLAG_RX_RESTART_NOW;
1224 /* guard against interrupt when we're going down */
1225 if (!test_bit(__E1000_DOWN, &adapter->state))
1226 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1229 if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1230 adapter->total_tx_bytes = 0;
1231 adapter->total_tx_packets = 0;
1232 adapter->total_rx_bytes = 0;
1233 adapter->total_rx_packets = 0;
1234 __netif_rx_schedule(netdev, &adapter->napi);
1241 * e1000_request_irq - initialize interrupts
1243 * Attempts to configure interrupts using the best available
1244 * capabilities of the hardware and kernel.
1246 static int e1000_request_irq(struct e1000_adapter *adapter)
1248 struct net_device *netdev = adapter->netdev;
1249 int irq_flags = IRQF_SHARED;
1252 if (!(adapter->flags & FLAG_MSI_TEST_FAILED)) {
1253 err = pci_enable_msi(adapter->pdev);
1255 adapter->flags |= FLAG_MSI_ENABLED;
1260 err = request_irq(adapter->pdev->irq,
1261 ((adapter->flags & FLAG_MSI_ENABLED) ?
1262 &e1000_intr_msi : &e1000_intr),
1263 irq_flags, netdev->name, netdev);
1265 if (adapter->flags & FLAG_MSI_ENABLED) {
1266 pci_disable_msi(adapter->pdev);
1267 adapter->flags &= ~FLAG_MSI_ENABLED;
1269 e_err("Unable to allocate interrupt, Error: %d\n", err);
1275 static void e1000_free_irq(struct e1000_adapter *adapter)
1277 struct net_device *netdev = adapter->netdev;
1279 free_irq(adapter->pdev->irq, netdev);
1280 if (adapter->flags & FLAG_MSI_ENABLED) {
1281 pci_disable_msi(adapter->pdev);
1282 adapter->flags &= ~FLAG_MSI_ENABLED;
1287 * e1000_irq_disable - Mask off interrupt generation on the NIC
1289 static void e1000_irq_disable(struct e1000_adapter *adapter)
1291 struct e1000_hw *hw = &adapter->hw;
1295 synchronize_irq(adapter->pdev->irq);
1299 * e1000_irq_enable - Enable default interrupt generation settings
1301 static void e1000_irq_enable(struct e1000_adapter *adapter)
1303 struct e1000_hw *hw = &adapter->hw;
1305 ew32(IMS, IMS_ENABLE_MASK);
1310 * e1000_get_hw_control - get control of the h/w from f/w
1311 * @adapter: address of board private structure
1313 * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1314 * For ASF and Pass Through versions of f/w this means that
1315 * the driver is loaded. For AMT version (only with 82573)
1316 * of the f/w this means that the network i/f is open.
1318 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1320 struct e1000_hw *hw = &adapter->hw;
1324 /* Let firmware know the driver has taken over */
1325 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1327 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1328 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1329 ctrl_ext = er32(CTRL_EXT);
1330 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1335 * e1000_release_hw_control - release control of the h/w to f/w
1336 * @adapter: address of board private structure
1338 * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1339 * For ASF and Pass Through versions of f/w this means that the
1340 * driver is no longer loaded. For AMT version (only with 82573) i
1341 * of the f/w this means that the network i/f is closed.
1344 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1346 struct e1000_hw *hw = &adapter->hw;
1350 /* Let firmware taken over control of h/w */
1351 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1353 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1354 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1355 ctrl_ext = er32(CTRL_EXT);
1356 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1361 * @e1000_alloc_ring - allocate memory for a ring structure
1363 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1364 struct e1000_ring *ring)
1366 struct pci_dev *pdev = adapter->pdev;
1368 ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1377 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1378 * @adapter: board private structure
1380 * Return 0 on success, negative on failure
1382 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1384 struct e1000_ring *tx_ring = adapter->tx_ring;
1385 int err = -ENOMEM, size;
1387 size = sizeof(struct e1000_buffer) * tx_ring->count;
1388 tx_ring->buffer_info = vmalloc(size);
1389 if (!tx_ring->buffer_info)
1391 memset(tx_ring->buffer_info, 0, size);
1393 /* round up to nearest 4K */
1394 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1395 tx_ring->size = ALIGN(tx_ring->size, 4096);
1397 err = e1000_alloc_ring_dma(adapter, tx_ring);
1401 tx_ring->next_to_use = 0;
1402 tx_ring->next_to_clean = 0;
1403 spin_lock_init(&adapter->tx_queue_lock);
1407 vfree(tx_ring->buffer_info);
1408 e_err("Unable to allocate memory for the transmit descriptor ring\n");
1413 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1414 * @adapter: board private structure
1416 * Returns 0 on success, negative on failure
1418 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1420 struct e1000_ring *rx_ring = adapter->rx_ring;
1421 struct e1000_buffer *buffer_info;
1422 int i, size, desc_len, err = -ENOMEM;
1424 size = sizeof(struct e1000_buffer) * rx_ring->count;
1425 rx_ring->buffer_info = vmalloc(size);
1426 if (!rx_ring->buffer_info)
1428 memset(rx_ring->buffer_info, 0, size);
1430 for (i = 0; i < rx_ring->count; i++) {
1431 buffer_info = &rx_ring->buffer_info[i];
1432 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1433 sizeof(struct e1000_ps_page),
1435 if (!buffer_info->ps_pages)
1439 desc_len = sizeof(union e1000_rx_desc_packet_split);
1441 /* Round up to nearest 4K */
1442 rx_ring->size = rx_ring->count * desc_len;
1443 rx_ring->size = ALIGN(rx_ring->size, 4096);
1445 err = e1000_alloc_ring_dma(adapter, rx_ring);
1449 rx_ring->next_to_clean = 0;
1450 rx_ring->next_to_use = 0;
1451 rx_ring->rx_skb_top = NULL;
1456 for (i = 0; i < rx_ring->count; i++) {
1457 buffer_info = &rx_ring->buffer_info[i];
1458 kfree(buffer_info->ps_pages);
1461 vfree(rx_ring->buffer_info);
1462 e_err("Unable to allocate memory for the transmit descriptor ring\n");
1467 * e1000_clean_tx_ring - Free Tx Buffers
1468 * @adapter: board private structure
1470 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1472 struct e1000_ring *tx_ring = adapter->tx_ring;
1473 struct e1000_buffer *buffer_info;
1477 for (i = 0; i < tx_ring->count; i++) {
1478 buffer_info = &tx_ring->buffer_info[i];
1479 e1000_put_txbuf(adapter, buffer_info);
1482 size = sizeof(struct e1000_buffer) * tx_ring->count;
1483 memset(tx_ring->buffer_info, 0, size);
1485 memset(tx_ring->desc, 0, tx_ring->size);
1487 tx_ring->next_to_use = 0;
1488 tx_ring->next_to_clean = 0;
1490 writel(0, adapter->hw.hw_addr + tx_ring->head);
1491 writel(0, adapter->hw.hw_addr + tx_ring->tail);
1495 * e1000e_free_tx_resources - Free Tx Resources per Queue
1496 * @adapter: board private structure
1498 * Free all transmit software resources
1500 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1502 struct pci_dev *pdev = adapter->pdev;
1503 struct e1000_ring *tx_ring = adapter->tx_ring;
1505 e1000_clean_tx_ring(adapter);
1507 vfree(tx_ring->buffer_info);
1508 tx_ring->buffer_info = NULL;
1510 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1512 tx_ring->desc = NULL;
1516 * e1000e_free_rx_resources - Free Rx Resources
1517 * @adapter: board private structure
1519 * Free all receive software resources
1522 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1524 struct pci_dev *pdev = adapter->pdev;
1525 struct e1000_ring *rx_ring = adapter->rx_ring;
1528 e1000_clean_rx_ring(adapter);
1530 for (i = 0; i < rx_ring->count; i++) {
1531 kfree(rx_ring->buffer_info[i].ps_pages);
1534 vfree(rx_ring->buffer_info);
1535 rx_ring->buffer_info = NULL;
1537 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1539 rx_ring->desc = NULL;
1543 * e1000_update_itr - update the dynamic ITR value based on statistics
1544 * @adapter: pointer to adapter
1545 * @itr_setting: current adapter->itr
1546 * @packets: the number of packets during this measurement interval
1547 * @bytes: the number of bytes during this measurement interval
1549 * Stores a new ITR value based on packets and byte
1550 * counts during the last interrupt. The advantage of per interrupt
1551 * computation is faster updates and more accurate ITR for the current
1552 * traffic pattern. Constants in this function were computed
1553 * based on theoretical maximum wire speed and thresholds were set based
1554 * on testing data as well as attempting to minimize response time
1555 * while increasing bulk throughput.
1556 * this functionality is controlled by the InterruptThrottleRate module
1557 * parameter (see e1000_param.c)
1559 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1560 u16 itr_setting, int packets,
1563 unsigned int retval = itr_setting;
1566 goto update_itr_done;
1568 switch (itr_setting) {
1569 case lowest_latency:
1570 /* handle TSO and jumbo frames */
1571 if (bytes/packets > 8000)
1572 retval = bulk_latency;
1573 else if ((packets < 5) && (bytes > 512)) {
1574 retval = low_latency;
1577 case low_latency: /* 50 usec aka 20000 ints/s */
1578 if (bytes > 10000) {
1579 /* this if handles the TSO accounting */
1580 if (bytes/packets > 8000) {
1581 retval = bulk_latency;
1582 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1583 retval = bulk_latency;
1584 } else if ((packets > 35)) {
1585 retval = lowest_latency;
1587 } else if (bytes/packets > 2000) {
1588 retval = bulk_latency;
1589 } else if (packets <= 2 && bytes < 512) {
1590 retval = lowest_latency;
1593 case bulk_latency: /* 250 usec aka 4000 ints/s */
1594 if (bytes > 25000) {
1596 retval = low_latency;
1598 } else if (bytes < 6000) {
1599 retval = low_latency;
1608 static void e1000_set_itr(struct e1000_adapter *adapter)
1610 struct e1000_hw *hw = &adapter->hw;
1612 u32 new_itr = adapter->itr;
1614 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1615 if (adapter->link_speed != SPEED_1000) {
1621 adapter->tx_itr = e1000_update_itr(adapter,
1623 adapter->total_tx_packets,
1624 adapter->total_tx_bytes);
1625 /* conservative mode (itr 3) eliminates the lowest_latency setting */
1626 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1627 adapter->tx_itr = low_latency;
1629 adapter->rx_itr = e1000_update_itr(adapter,
1631 adapter->total_rx_packets,
1632 adapter->total_rx_bytes);
1633 /* conservative mode (itr 3) eliminates the lowest_latency setting */
1634 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1635 adapter->rx_itr = low_latency;
1637 current_itr = max(adapter->rx_itr, adapter->tx_itr);
1639 switch (current_itr) {
1640 /* counts and packets in update_itr are dependent on these numbers */
1641 case lowest_latency:
1645 new_itr = 20000; /* aka hwitr = ~200 */
1655 if (new_itr != adapter->itr) {
1657 * this attempts to bias the interrupt rate towards Bulk
1658 * by adding intermediate steps when interrupt rate is
1661 new_itr = new_itr > adapter->itr ?
1662 min(adapter->itr + (new_itr >> 2), new_itr) :
1664 adapter->itr = new_itr;
1665 ew32(ITR, 1000000000 / (new_itr * 256));
1670 * e1000_clean - NAPI Rx polling callback
1671 * @napi: struct associated with this polling callback
1672 * @budget: amount of packets driver is allowed to process this poll
1674 static int e1000_clean(struct napi_struct *napi, int budget)
1676 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1677 struct net_device *poll_dev = adapter->netdev;
1678 int tx_cleaned = 0, work_done = 0;
1680 /* Must NOT use netdev_priv macro here. */
1681 adapter = poll_dev->priv;
1684 * e1000_clean is called per-cpu. This lock protects
1685 * tx_ring from being cleaned by multiple cpus
1686 * simultaneously. A failure obtaining the lock means
1687 * tx_ring is currently being cleaned anyway.
1689 if (spin_trylock(&adapter->tx_queue_lock)) {
1690 tx_cleaned = e1000_clean_tx_irq(adapter);
1691 spin_unlock(&adapter->tx_queue_lock);
1694 adapter->clean_rx(adapter, &work_done, budget);
1699 /* If budget not fully consumed, exit the polling mode */
1700 if (work_done < budget) {
1701 if (adapter->itr_setting & 3)
1702 e1000_set_itr(adapter);
1703 netif_rx_complete(poll_dev, napi);
1704 e1000_irq_enable(adapter);
1710 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1712 struct e1000_adapter *adapter = netdev_priv(netdev);
1713 struct e1000_hw *hw = &adapter->hw;
1716 /* don't update vlan cookie if already programmed */
1717 if ((adapter->hw.mng_cookie.status &
1718 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1719 (vid == adapter->mng_vlan_id))
1721 /* add VID to filter table */
1722 index = (vid >> 5) & 0x7F;
1723 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1724 vfta |= (1 << (vid & 0x1F));
1725 e1000e_write_vfta(hw, index, vfta);
1728 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1730 struct e1000_adapter *adapter = netdev_priv(netdev);
1731 struct e1000_hw *hw = &adapter->hw;
1734 if (!test_bit(__E1000_DOWN, &adapter->state))
1735 e1000_irq_disable(adapter);
1736 vlan_group_set_device(adapter->vlgrp, vid, NULL);
1738 if (!test_bit(__E1000_DOWN, &adapter->state))
1739 e1000_irq_enable(adapter);
1741 if ((adapter->hw.mng_cookie.status &
1742 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1743 (vid == adapter->mng_vlan_id)) {
1744 /* release control to f/w */
1745 e1000_release_hw_control(adapter);
1749 /* remove VID from filter table */
1750 index = (vid >> 5) & 0x7F;
1751 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1752 vfta &= ~(1 << (vid & 0x1F));
1753 e1000e_write_vfta(hw, index, vfta);
1756 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
1758 struct net_device *netdev = adapter->netdev;
1759 u16 vid = adapter->hw.mng_cookie.vlan_id;
1760 u16 old_vid = adapter->mng_vlan_id;
1762 if (!adapter->vlgrp)
1765 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
1766 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1767 if (adapter->hw.mng_cookie.status &
1768 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1769 e1000_vlan_rx_add_vid(netdev, vid);
1770 adapter->mng_vlan_id = vid;
1773 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
1775 !vlan_group_get_device(adapter->vlgrp, old_vid))
1776 e1000_vlan_rx_kill_vid(netdev, old_vid);
1778 adapter->mng_vlan_id = vid;
1783 static void e1000_vlan_rx_register(struct net_device *netdev,
1784 struct vlan_group *grp)
1786 struct e1000_adapter *adapter = netdev_priv(netdev);
1787 struct e1000_hw *hw = &adapter->hw;
1790 if (!test_bit(__E1000_DOWN, &adapter->state))
1791 e1000_irq_disable(adapter);
1792 adapter->vlgrp = grp;
1795 /* enable VLAN tag insert/strip */
1797 ctrl |= E1000_CTRL_VME;
1800 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1801 /* enable VLAN receive filtering */
1803 rctl &= ~E1000_RCTL_CFIEN;
1805 e1000_update_mng_vlan(adapter);
1808 /* disable VLAN tag insert/strip */
1810 ctrl &= ~E1000_CTRL_VME;
1813 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1814 if (adapter->mng_vlan_id !=
1815 (u16)E1000_MNG_VLAN_NONE) {
1816 e1000_vlan_rx_kill_vid(netdev,
1817 adapter->mng_vlan_id);
1818 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1823 if (!test_bit(__E1000_DOWN, &adapter->state))
1824 e1000_irq_enable(adapter);
1827 static void e1000_restore_vlan(struct e1000_adapter *adapter)
1831 e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
1833 if (!adapter->vlgrp)
1836 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1837 if (!vlan_group_get_device(adapter->vlgrp, vid))
1839 e1000_vlan_rx_add_vid(adapter->netdev, vid);
1843 static void e1000_init_manageability(struct e1000_adapter *adapter)
1845 struct e1000_hw *hw = &adapter->hw;
1848 if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
1854 * enable receiving management packets to the host. this will probably
1855 * generate destination unreachable messages from the host OS, but
1856 * the packets will be handled on SMBUS
1858 manc |= E1000_MANC_EN_MNG2HOST;
1859 manc2h = er32(MANC2H);
1860 #define E1000_MNG2HOST_PORT_623 (1 << 5)
1861 #define E1000_MNG2HOST_PORT_664 (1 << 6)
1862 manc2h |= E1000_MNG2HOST_PORT_623;
1863 manc2h |= E1000_MNG2HOST_PORT_664;
1864 ew32(MANC2H, manc2h);
1869 * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1870 * @adapter: board private structure
1872 * Configure the Tx unit of the MAC after a reset.
1874 static void e1000_configure_tx(struct e1000_adapter *adapter)
1876 struct e1000_hw *hw = &adapter->hw;
1877 struct e1000_ring *tx_ring = adapter->tx_ring;
1879 u32 tdlen, tctl, tipg, tarc;
1882 /* Setup the HW Tx Head and Tail descriptor pointers */
1883 tdba = tx_ring->dma;
1884 tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
1885 ew32(TDBAL, (tdba & DMA_32BIT_MASK));
1886 ew32(TDBAH, (tdba >> 32));
1890 tx_ring->head = E1000_TDH;
1891 tx_ring->tail = E1000_TDT;
1893 /* Set the default values for the Tx Inter Packet Gap timer */
1894 tipg = DEFAULT_82543_TIPG_IPGT_COPPER; /* 8 */
1895 ipgr1 = DEFAULT_82543_TIPG_IPGR1; /* 8 */
1896 ipgr2 = DEFAULT_82543_TIPG_IPGR2; /* 6 */
1898 if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
1899 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /* 7 */
1901 tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1902 tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1905 /* Set the Tx Interrupt Delay register */
1906 ew32(TIDV, adapter->tx_int_delay);
1907 /* Tx irq moderation */
1908 ew32(TADV, adapter->tx_abs_int_delay);
1910 /* Program the Transmit Control Register */
1912 tctl &= ~E1000_TCTL_CT;
1913 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1914 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1916 if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
1917 tarc = er32(TARC(0));
1919 * set the speed mode bit, we'll clear it if we're not at
1920 * gigabit link later
1922 #define SPEED_MODE_BIT (1 << 21)
1923 tarc |= SPEED_MODE_BIT;
1924 ew32(TARC(0), tarc);
1927 /* errata: program both queues to unweighted RR */
1928 if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
1929 tarc = er32(TARC(0));
1931 ew32(TARC(0), tarc);
1932 tarc = er32(TARC(1));
1934 ew32(TARC(1), tarc);
1937 e1000e_config_collision_dist(hw);
1939 /* Setup Transmit Descriptor Settings for eop descriptor */
1940 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1942 /* only set IDE if we are delaying interrupts using the timers */
1943 if (adapter->tx_int_delay)
1944 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1946 /* enable Report Status bit */
1947 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1951 adapter->tx_queue_len = adapter->netdev->tx_queue_len;
1955 * e1000_setup_rctl - configure the receive control registers
1956 * @adapter: Board private structure
1958 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1959 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1960 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1962 struct e1000_hw *hw = &adapter->hw;
1967 /* Program MC offset vector base */
1969 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1970 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1971 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1972 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1974 /* Do not Store bad packets */
1975 rctl &= ~E1000_RCTL_SBP;
1977 /* Enable Long Packet receive */
1978 if (adapter->netdev->mtu <= ETH_DATA_LEN)
1979 rctl &= ~E1000_RCTL_LPE;
1981 rctl |= E1000_RCTL_LPE;
1983 /* Enable hardware CRC frame stripping */
1984 rctl |= E1000_RCTL_SECRC;
1986 /* Setup buffer sizes */
1987 rctl &= ~E1000_RCTL_SZ_4096;
1988 rctl |= E1000_RCTL_BSEX;
1989 switch (adapter->rx_buffer_len) {
1991 rctl |= E1000_RCTL_SZ_256;
1992 rctl &= ~E1000_RCTL_BSEX;
1995 rctl |= E1000_RCTL_SZ_512;
1996 rctl &= ~E1000_RCTL_BSEX;
1999 rctl |= E1000_RCTL_SZ_1024;
2000 rctl &= ~E1000_RCTL_BSEX;
2004 rctl |= E1000_RCTL_SZ_2048;
2005 rctl &= ~E1000_RCTL_BSEX;
2008 rctl |= E1000_RCTL_SZ_4096;
2011 rctl |= E1000_RCTL_SZ_8192;
2014 rctl |= E1000_RCTL_SZ_16384;
2019 * 82571 and greater support packet-split where the protocol
2020 * header is placed in skb->data and the packet data is
2021 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2022 * In the case of a non-split, skb->data is linearly filled,
2023 * followed by the page buffers. Therefore, skb->data is
2024 * sized to hold the largest protocol header.
2026 * allocations using alloc_page take too long for regular MTU
2027 * so only enable packet split for jumbo frames
2029 * Using pages when the page size is greater than 16k wastes
2030 * a lot of memory, since we allocate 3 pages at all times
2033 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2034 if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2035 (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2036 adapter->rx_ps_pages = pages;
2038 adapter->rx_ps_pages = 0;
2040 if (adapter->rx_ps_pages) {
2041 /* Configure extra packet-split registers */
2042 rfctl = er32(RFCTL);
2043 rfctl |= E1000_RFCTL_EXTEN;
2045 * disable packet split support for IPv6 extension headers,
2046 * because some malformed IPv6 headers can hang the Rx
2048 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2049 E1000_RFCTL_NEW_IPV6_EXT_DIS);
2053 /* Enable Packet split descriptors */
2054 rctl |= E1000_RCTL_DTYP_PS;
2056 psrctl |= adapter->rx_ps_bsize0 >>
2057 E1000_PSRCTL_BSIZE0_SHIFT;
2059 switch (adapter->rx_ps_pages) {
2061 psrctl |= PAGE_SIZE <<
2062 E1000_PSRCTL_BSIZE3_SHIFT;
2064 psrctl |= PAGE_SIZE <<
2065 E1000_PSRCTL_BSIZE2_SHIFT;
2067 psrctl |= PAGE_SIZE >>
2068 E1000_PSRCTL_BSIZE1_SHIFT;
2072 ew32(PSRCTL, psrctl);
2076 /* just started the receive unit, no need to restart */
2077 adapter->flags &= ~FLAG_RX_RESTART_NOW;
2081 * e1000_configure_rx - Configure Receive Unit after Reset
2082 * @adapter: board private structure
2084 * Configure the Rx unit of the MAC after a reset.
2086 static void e1000_configure_rx(struct e1000_adapter *adapter)
2088 struct e1000_hw *hw = &adapter->hw;
2089 struct e1000_ring *rx_ring = adapter->rx_ring;
2091 u32 rdlen, rctl, rxcsum, ctrl_ext;
2093 if (adapter->rx_ps_pages) {
2094 /* this is a 32 byte descriptor */
2095 rdlen = rx_ring->count *
2096 sizeof(union e1000_rx_desc_packet_split);
2097 adapter->clean_rx = e1000_clean_rx_irq_ps;
2098 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2099 } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2100 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2101 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2102 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2104 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2105 adapter->clean_rx = e1000_clean_rx_irq;
2106 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2109 /* disable receives while setting up the descriptors */
2111 ew32(RCTL, rctl & ~E1000_RCTL_EN);
2115 /* set the Receive Delay Timer Register */
2116 ew32(RDTR, adapter->rx_int_delay);
2118 /* irq moderation */
2119 ew32(RADV, adapter->rx_abs_int_delay);
2120 if (adapter->itr_setting != 0)
2121 ew32(ITR, 1000000000 / (adapter->itr * 256));
2123 ctrl_ext = er32(CTRL_EXT);
2124 /* Reset delay timers after every interrupt */
2125 ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2126 /* Auto-Mask interrupts upon ICR access */
2127 ctrl_ext |= E1000_CTRL_EXT_IAME;
2128 ew32(IAM, 0xffffffff);
2129 ew32(CTRL_EXT, ctrl_ext);
2133 * Setup the HW Rx Head and Tail Descriptor Pointers and
2134 * the Base and Length of the Rx Descriptor Ring
2136 rdba = rx_ring->dma;
2137 ew32(RDBAL, (rdba & DMA_32BIT_MASK));
2138 ew32(RDBAH, (rdba >> 32));
2142 rx_ring->head = E1000_RDH;
2143 rx_ring->tail = E1000_RDT;
2145 /* Enable Receive Checksum Offload for TCP and UDP */
2146 rxcsum = er32(RXCSUM);
2147 if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2148 rxcsum |= E1000_RXCSUM_TUOFL;
2151 * IPv4 payload checksum for UDP fragments must be
2152 * used in conjunction with packet-split.
2154 if (adapter->rx_ps_pages)
2155 rxcsum |= E1000_RXCSUM_IPPCSE;
2157 rxcsum &= ~E1000_RXCSUM_TUOFL;
2158 /* no need to clear IPPCSE as it defaults to 0 */
2160 ew32(RXCSUM, rxcsum);
2163 * Enable early receives on supported devices, only takes effect when
2164 * packet size is equal or larger than the specified value (in 8 byte
2165 * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2167 if ((adapter->flags & FLAG_HAS_ERT) &&
2168 (adapter->netdev->mtu > ETH_DATA_LEN)) {
2169 u32 rxdctl = er32(RXDCTL(0));
2170 ew32(RXDCTL(0), rxdctl | 0x3);
2171 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2173 * With jumbo frames and early-receive enabled, excessive
2174 * C4->C2 latencies result in dropped transactions.
2176 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2177 e1000e_driver_name, 55);
2179 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2181 PM_QOS_DEFAULT_VALUE);
2184 /* Enable Receives */
2189 * e1000_update_mc_addr_list - Update Multicast addresses
2190 * @hw: pointer to the HW structure
2191 * @mc_addr_list: array of multicast addresses to program
2192 * @mc_addr_count: number of multicast addresses to program
2193 * @rar_used_count: the first RAR register free to program
2194 * @rar_count: total number of supported Receive Address Registers
2196 * Updates the Receive Address Registers and Multicast Table Array.
2197 * The caller must have a packed mc_addr_list of multicast addresses.
2198 * The parameter rar_count will usually be hw->mac.rar_entry_count
2199 * unless there are workarounds that change this. Currently no func pointer
2200 * exists and all implementations are handled in the generic version of this
2203 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2204 u32 mc_addr_count, u32 rar_used_count,
2207 hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2208 rar_used_count, rar_count);
2212 * e1000_set_multi - Multicast and Promiscuous mode set
2213 * @netdev: network interface device structure
2215 * The set_multi entry point is called whenever the multicast address
2216 * list or the network interface flags are updated. This routine is
2217 * responsible for configuring the hardware for proper multicast,
2218 * promiscuous mode, and all-multi behavior.
2220 static void e1000_set_multi(struct net_device *netdev)
2222 struct e1000_adapter *adapter = netdev_priv(netdev);
2223 struct e1000_hw *hw = &adapter->hw;
2224 struct e1000_mac_info *mac = &hw->mac;
2225 struct dev_mc_list *mc_ptr;
2230 /* Check for Promiscuous and All Multicast modes */
2234 if (netdev->flags & IFF_PROMISC) {
2235 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2236 rctl &= ~E1000_RCTL_VFE;
2238 if (netdev->flags & IFF_ALLMULTI) {
2239 rctl |= E1000_RCTL_MPE;
2240 rctl &= ~E1000_RCTL_UPE;
2242 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2244 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2245 rctl |= E1000_RCTL_VFE;
2250 if (netdev->mc_count) {
2251 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2255 /* prepare a packed array of only addresses. */
2256 mc_ptr = netdev->mc_list;
2258 for (i = 0; i < netdev->mc_count; i++) {
2261 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2263 mc_ptr = mc_ptr->next;
2266 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2267 mac->rar_entry_count);
2271 * if we're called from probe, we might not have
2272 * anything to do here, so clear out the list
2274 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2279 * e1000_configure - configure the hardware for Rx and Tx
2280 * @adapter: private board structure
2282 static void e1000_configure(struct e1000_adapter *adapter)
2284 e1000_set_multi(adapter->netdev);
2286 e1000_restore_vlan(adapter);
2287 e1000_init_manageability(adapter);
2289 e1000_configure_tx(adapter);
2290 e1000_setup_rctl(adapter);
2291 e1000_configure_rx(adapter);
2292 adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2296 * e1000e_power_up_phy - restore link in case the phy was powered down
2297 * @adapter: address of board private structure
2299 * The phy may be powered down to save power and turn off link when the
2300 * driver is unloaded and wake on lan is not enabled (among others)
2301 * *** this routine MUST be followed by a call to e1000e_reset ***
2303 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2307 /* Just clear the power down bit to wake the phy back up */
2308 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2310 * According to the manual, the phy will retain its
2311 * settings across a power-down/up cycle
2313 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2314 mii_reg &= ~MII_CR_POWER_DOWN;
2315 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2318 adapter->hw.mac.ops.setup_link(&adapter->hw);
2322 * e1000_power_down_phy - Power down the PHY
2324 * Power down the PHY so no link is implied when interface is down
2325 * The PHY cannot be powered down is management or WoL is active
2327 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2329 struct e1000_hw *hw = &adapter->hw;
2332 /* WoL is enabled */
2336 /* non-copper PHY? */
2337 if (adapter->hw.phy.media_type != e1000_media_type_copper)
2340 /* reset is blocked because of a SoL/IDER session */
2341 if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2344 /* manageability (AMT) is enabled */
2345 if (er32(MANC) & E1000_MANC_SMBUS_EN)
2348 /* power down the PHY */
2349 e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2350 mii_reg |= MII_CR_POWER_DOWN;
2351 e1e_wphy(hw, PHY_CONTROL, mii_reg);
2356 * e1000e_reset - bring the hardware into a known good state
2358 * This function boots the hardware and enables some settings that
2359 * require a configuration cycle of the hardware - those cannot be
2360 * set/changed during runtime. After reset the device needs to be
2361 * properly configured for Rx, Tx etc.
2363 void e1000e_reset(struct e1000_adapter *adapter)
2365 struct e1000_mac_info *mac = &adapter->hw.mac;
2366 struct e1000_fc_info *fc = &adapter->hw.fc;
2367 struct e1000_hw *hw = &adapter->hw;
2368 u32 tx_space, min_tx_space, min_rx_space;
2369 u32 pba = adapter->pba;
2372 /* reset Packet Buffer Allocation to default */
2375 if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2377 * To maintain wire speed transmits, the Tx FIFO should be
2378 * large enough to accommodate two full transmit packets,
2379 * rounded up to the next 1KB and expressed in KB. Likewise,
2380 * the Rx FIFO should be large enough to accommodate at least
2381 * one full receive packet and is similarly rounded up and
2385 /* upper 16 bits has Tx packet buffer allocation size in KB */
2386 tx_space = pba >> 16;
2387 /* lower 16 bits has Rx packet buffer allocation size in KB */
2390 * the Tx fifo also stores 16 bytes of information about the tx
2391 * but don't include ethernet FCS because hardware appends it
2393 min_tx_space = (adapter->max_frame_size +
2394 sizeof(struct e1000_tx_desc) -
2396 min_tx_space = ALIGN(min_tx_space, 1024);
2397 min_tx_space >>= 10;
2398 /* software strips receive CRC, so leave room for it */
2399 min_rx_space = adapter->max_frame_size;
2400 min_rx_space = ALIGN(min_rx_space, 1024);
2401 min_rx_space >>= 10;
2404 * If current Tx allocation is less than the min Tx FIFO size,
2405 * and the min Tx FIFO size is less than the current Rx FIFO
2406 * allocation, take space away from current Rx allocation
2408 if ((tx_space < min_tx_space) &&
2409 ((min_tx_space - tx_space) < pba)) {
2410 pba -= min_tx_space - tx_space;
2413 * if short on Rx space, Rx wins and must trump tx
2414 * adjustment or use Early Receive if available
2416 if ((pba < min_rx_space) &&
2417 (!(adapter->flags & FLAG_HAS_ERT)))
2418 /* ERT enabled in e1000_configure_rx */
2427 * flow control settings
2429 * The high water mark must be low enough to fit one full frame
2430 * (or the size used for early receive) above it in the Rx FIFO.
2431 * Set it to the lower of:
2432 * - 90% of the Rx FIFO size, and
2433 * - the full Rx FIFO size minus the early receive size (for parts
2434 * with ERT support assuming ERT set to E1000_ERT_2048), or
2435 * - the full Rx FIFO size minus one full frame
2437 if (adapter->flags & FLAG_HAS_ERT)
2438 hwm = min(((pba << 10) * 9 / 10),
2439 ((pba << 10) - (E1000_ERT_2048 << 3)));
2441 hwm = min(((pba << 10) * 9 / 10),
2442 ((pba << 10) - adapter->max_frame_size));
2444 fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
2445 fc->low_water = fc->high_water - 8;
2447 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2448 fc->pause_time = 0xFFFF;
2450 fc->pause_time = E1000_FC_PAUSE_TIME;
2452 fc->type = fc->original_type;
2454 /* Allow time for pending master requests to run */
2455 mac->ops.reset_hw(hw);
2458 * For parts with AMT enabled, let the firmware know
2459 * that the network interface is in control
2461 if (adapter->flags & FLAG_HAS_AMT)
2462 e1000_get_hw_control(adapter);
2466 if (mac->ops.init_hw(hw))
2467 e_err("Hardware Error\n");
2469 e1000_update_mng_vlan(adapter);
2471 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2472 ew32(VET, ETH_P_8021Q);
2474 e1000e_reset_adaptive(hw);
2475 e1000_get_phy_info(hw);
2477 if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2480 * speed up time to link by disabling smart power down, ignore
2481 * the return value of this function because there is nothing
2482 * different we would do if it failed
2484 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2485 phy_data &= ~IGP02E1000_PM_SPD;
2486 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2490 int e1000e_up(struct e1000_adapter *adapter)
2492 struct e1000_hw *hw = &adapter->hw;
2494 /* hardware has been reset, we need to reload some things */
2495 e1000_configure(adapter);
2497 clear_bit(__E1000_DOWN, &adapter->state);
2499 napi_enable(&adapter->napi);
2500 e1000_irq_enable(adapter);
2502 /* fire a link change interrupt to start the watchdog */
2503 ew32(ICS, E1000_ICS_LSC);
2507 void e1000e_down(struct e1000_adapter *adapter)
2509 struct net_device *netdev = adapter->netdev;
2510 struct e1000_hw *hw = &adapter->hw;
2514 * signal that we're down so the interrupt handler does not
2515 * reschedule our watchdog timer
2517 set_bit(__E1000_DOWN, &adapter->state);
2519 /* disable receives in the hardware */
2521 ew32(RCTL, rctl & ~E1000_RCTL_EN);
2522 /* flush and sleep below */
2524 netif_tx_stop_all_queues(netdev);
2526 /* disable transmits in the hardware */
2528 tctl &= ~E1000_TCTL_EN;
2530 /* flush both disables and wait for them to finish */
2534 napi_disable(&adapter->napi);
2535 e1000_irq_disable(adapter);
2537 del_timer_sync(&adapter->watchdog_timer);
2538 del_timer_sync(&adapter->phy_info_timer);
2540 netdev->tx_queue_len = adapter->tx_queue_len;
2541 netif_carrier_off(netdev);
2542 adapter->link_speed = 0;
2543 adapter->link_duplex = 0;
2545 if (!pci_channel_offline(adapter->pdev))
2546 e1000e_reset(adapter);
2547 e1000_clean_tx_ring(adapter);
2548 e1000_clean_rx_ring(adapter);
2551 * TODO: for power management, we could drop the link and
2552 * pci_disable_device here.
2556 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2559 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2561 e1000e_down(adapter);
2563 clear_bit(__E1000_RESETTING, &adapter->state);
2567 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2568 * @adapter: board private structure to initialize
2570 * e1000_sw_init initializes the Adapter private data structure.
2571 * Fields are initialized based on PCI device information and
2572 * OS network device settings (MTU size).
2574 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2576 struct net_device *netdev = adapter->netdev;
2578 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2579 adapter->rx_ps_bsize0 = 128;
2580 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2581 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2583 adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2584 if (!adapter->tx_ring)
2587 adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2588 if (!adapter->rx_ring)
2591 spin_lock_init(&adapter->tx_queue_lock);
2593 /* Explicitly disable IRQ since the NIC can be in any state. */
2594 e1000_irq_disable(adapter);
2596 spin_lock_init(&adapter->stats_lock);
2598 set_bit(__E1000_DOWN, &adapter->state);
2602 e_err("Unable to allocate memory for queues\n");
2603 kfree(adapter->rx_ring);
2604 kfree(adapter->tx_ring);
2609 * e1000_intr_msi_test - Interrupt Handler
2610 * @irq: interrupt number
2611 * @data: pointer to a network interface device structure
2613 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2615 struct net_device *netdev = data;
2616 struct e1000_adapter *adapter = netdev_priv(netdev);
2617 struct e1000_hw *hw = &adapter->hw;
2618 u32 icr = er32(ICR);
2620 e_dbg("%s: icr is %08X\n", netdev->name, icr);
2621 if (icr & E1000_ICR_RXSEQ) {
2622 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2630 * e1000_test_msi_interrupt - Returns 0 for successful test
2631 * @adapter: board private struct
2633 * code flow taken from tg3.c
2635 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2637 struct net_device *netdev = adapter->netdev;
2638 struct e1000_hw *hw = &adapter->hw;
2641 /* poll_enable hasn't been called yet, so don't need disable */
2642 /* clear any pending events */
2645 /* free the real vector and request a test handler */
2646 e1000_free_irq(adapter);
2648 /* Assume that the test fails, if it succeeds then the test
2649 * MSI irq handler will unset this flag */
2650 adapter->flags |= FLAG_MSI_TEST_FAILED;
2652 err = pci_enable_msi(adapter->pdev);
2654 goto msi_test_failed;
2656 err = request_irq(adapter->pdev->irq, &e1000_intr_msi_test, 0,
2657 netdev->name, netdev);
2659 pci_disable_msi(adapter->pdev);
2660 goto msi_test_failed;
2665 e1000_irq_enable(adapter);
2667 /* fire an unusual interrupt on the test handler */
2668 ew32(ICS, E1000_ICS_RXSEQ);
2672 e1000_irq_disable(adapter);
2676 if (adapter->flags & FLAG_MSI_TEST_FAILED) {
2678 e_info("MSI interrupt test failed!\n");
2681 free_irq(adapter->pdev->irq, netdev);
2682 pci_disable_msi(adapter->pdev);
2685 goto msi_test_failed;
2687 /* okay so the test worked, restore settings */
2688 e_dbg("%s: MSI interrupt test succeeded!\n", netdev->name);
2690 /* restore the original vector, even if it failed */
2691 e1000_request_irq(adapter);
2696 * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
2697 * @adapter: board private struct
2699 * code flow taken from tg3.c, called with e1000 interrupts disabled.
2701 static int e1000_test_msi(struct e1000_adapter *adapter)
2706 if (!(adapter->flags & FLAG_MSI_ENABLED))
2709 /* disable SERR in case the MSI write causes a master abort */
2710 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
2711 pci_write_config_word(adapter->pdev, PCI_COMMAND,
2712 pci_cmd & ~PCI_COMMAND_SERR);
2714 err = e1000_test_msi_interrupt(adapter);
2716 /* restore previous setting of command word */
2717 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
2723 /* EIO means MSI test failed */
2727 /* back to INTx mode */
2728 e_warn("MSI interrupt test failed, using legacy interrupt.\n");
2730 e1000_free_irq(adapter);
2732 err = e1000_request_irq(adapter);
2738 * e1000_open - Called when a network interface is made active
2739 * @netdev: network interface device structure
2741 * Returns 0 on success, negative value on failure
2743 * The open entry point is called when a network interface is made
2744 * active by the system (IFF_UP). At this point all resources needed
2745 * for transmit and receive operations are allocated, the interrupt
2746 * handler is registered with the OS, the watchdog timer is started,
2747 * and the stack is notified that the interface is ready.
2749 static int e1000_open(struct net_device *netdev)
2751 struct e1000_adapter *adapter = netdev_priv(netdev);
2752 struct e1000_hw *hw = &adapter->hw;
2755 /* disallow open during test */
2756 if (test_bit(__E1000_TESTING, &adapter->state))
2759 /* allocate transmit descriptors */
2760 err = e1000e_setup_tx_resources(adapter);
2764 /* allocate receive descriptors */
2765 err = e1000e_setup_rx_resources(adapter);
2769 e1000e_power_up_phy(adapter);
2771 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2772 if ((adapter->hw.mng_cookie.status &
2773 E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
2774 e1000_update_mng_vlan(adapter);
2777 * If AMT is enabled, let the firmware know that the network
2778 * interface is now open
2780 if (adapter->flags & FLAG_HAS_AMT)
2781 e1000_get_hw_control(adapter);
2784 * before we allocate an interrupt, we must be ready to handle it.
2785 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2786 * as soon as we call pci_request_irq, so we have to setup our
2787 * clean_rx handler before we do so.
2789 e1000_configure(adapter);
2791 err = e1000_request_irq(adapter);
2796 * Work around PCIe errata with MSI interrupts causing some chipsets to
2797 * ignore e1000e MSI messages, which means we need to test our MSI
2801 err = e1000_test_msi(adapter);
2803 e_err("Interrupt allocation failed\n");
2808 /* From here on the code is the same as e1000e_up() */
2809 clear_bit(__E1000_DOWN, &adapter->state);
2811 napi_enable(&adapter->napi);
2813 e1000_irq_enable(adapter);
2815 netif_tx_start_all_queues(netdev);
2817 /* fire a link status change interrupt to start the watchdog */
2818 ew32(ICS, E1000_ICS_LSC);
2823 e1000_release_hw_control(adapter);
2824 e1000_power_down_phy(adapter);
2825 e1000e_free_rx_resources(adapter);
2827 e1000e_free_tx_resources(adapter);
2829 e1000e_reset(adapter);
2835 * e1000_close - Disables a network interface
2836 * @netdev: network interface device structure
2838 * Returns 0, this is not allowed to fail
2840 * The close entry point is called when an interface is de-activated
2841 * by the OS. The hardware is still under the drivers control, but
2842 * needs to be disabled. A global MAC reset is issued to stop the
2843 * hardware, and all transmit and receive resources are freed.
2845 static int e1000_close(struct net_device *netdev)
2847 struct e1000_adapter *adapter = netdev_priv(netdev);
2849 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
2850 e1000e_down(adapter);
2851 e1000_power_down_phy(adapter);
2852 e1000_free_irq(adapter);
2854 e1000e_free_tx_resources(adapter);
2855 e1000e_free_rx_resources(adapter);
2858 * kill manageability vlan ID if supported, but not if a vlan with
2859 * the same ID is registered on the host OS (let 8021q kill it)
2861 if ((adapter->hw.mng_cookie.status &
2862 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2864 vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
2865 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2868 * If AMT is enabled, let the firmware know that the network
2869 * interface is now closed
2871 if (adapter->flags & FLAG_HAS_AMT)
2872 e1000_release_hw_control(adapter);
2877 * e1000_set_mac - Change the Ethernet Address of the NIC
2878 * @netdev: network interface device structure
2879 * @p: pointer to an address structure
2881 * Returns 0 on success, negative on failure
2883 static int e1000_set_mac(struct net_device *netdev, void *p)
2885 struct e1000_adapter *adapter = netdev_priv(netdev);
2886 struct sockaddr *addr = p;
2888 if (!is_valid_ether_addr(addr->sa_data))
2889 return -EADDRNOTAVAIL;
2891 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2892 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
2894 e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
2896 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
2897 /* activate the work around */
2898 e1000e_set_laa_state_82571(&adapter->hw, 1);
2901 * Hold a copy of the LAA in RAR[14] This is done so that
2902 * between the time RAR[0] gets clobbered and the time it
2903 * gets fixed (in e1000_watchdog), the actual LAA is in one
2904 * of the RARs and no incoming packets directed to this port
2905 * are dropped. Eventually the LAA will be in RAR[0] and
2908 e1000e_rar_set(&adapter->hw,
2909 adapter->hw.mac.addr,
2910 adapter->hw.mac.rar_entry_count - 1);
2917 * Need to wait a few seconds after link up to get diagnostic information from
2920 static void e1000_update_phy_info(unsigned long data)
2922 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2923 e1000_get_phy_info(&adapter->hw);
2927 * e1000e_update_stats - Update the board statistics counters
2928 * @adapter: board private structure
2930 void e1000e_update_stats(struct e1000_adapter *adapter)
2932 struct e1000_hw *hw = &adapter->hw;
2933 struct pci_dev *pdev = adapter->pdev;
2934 unsigned long irq_flags;
2937 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
2940 * Prevent stats update while adapter is being reset, or if the pci
2941 * connection is down.
2943 if (adapter->link_speed == 0)
2945 if (pci_channel_offline(pdev))
2948 spin_lock_irqsave(&adapter->stats_lock, irq_flags);
2951 * these counters are modified from e1000_adjust_tbi_stats,
2952 * called from the interrupt context, so they must only
2953 * be written while holding adapter->stats_lock
2956 adapter->stats.crcerrs += er32(CRCERRS);
2957 adapter->stats.gprc += er32(GPRC);
2958 adapter->stats.gorc += er32(GORCL);
2959 er32(GORCH); /* Clear gorc */
2960 adapter->stats.bprc += er32(BPRC);
2961 adapter->stats.mprc += er32(MPRC);
2962 adapter->stats.roc += er32(ROC);
2964 adapter->stats.mpc += er32(MPC);
2965 adapter->stats.scc += er32(SCC);
2966 adapter->stats.ecol += er32(ECOL);
2967 adapter->stats.mcc += er32(MCC);
2968 adapter->stats.latecol += er32(LATECOL);
2969 adapter->stats.dc += er32(DC);
2970 adapter->stats.xonrxc += er32(XONRXC);
2971 adapter->stats.xontxc += er32(XONTXC);
2972 adapter->stats.xoffrxc += er32(XOFFRXC);
2973 adapter->stats.xofftxc += er32(XOFFTXC);
2974 adapter->stats.gptc += er32(GPTC);
2975 adapter->stats.gotc += er32(GOTCL);
2976 er32(GOTCH); /* Clear gotc */
2977 adapter->stats.rnbc += er32(RNBC);
2978 adapter->stats.ruc += er32(RUC);
2980 adapter->stats.mptc += er32(MPTC);
2981 adapter->stats.bptc += er32(BPTC);
2983 /* used for adaptive IFS */
2985 hw->mac.tx_packet_delta = er32(TPT);
2986 adapter->stats.tpt += hw->mac.tx_packet_delta;
2987 hw->mac.collision_delta = er32(COLC);
2988 adapter->stats.colc += hw->mac.collision_delta;
2990 adapter->stats.algnerrc += er32(ALGNERRC);
2991 adapter->stats.rxerrc += er32(RXERRC);
2992 adapter->stats.tncrs += er32(TNCRS);
2993 adapter->stats.cexterr += er32(CEXTERR);
2994 adapter->stats.tsctc += er32(TSCTC);
2995 adapter->stats.tsctfc += er32(TSCTFC);
2997 /* Fill out the OS statistics structure */
2998 adapter->net_stats.multicast = adapter->stats.mprc;
2999 adapter->net_stats.collisions = adapter->stats.colc;
3004 * RLEC on some newer hardware can be incorrect so build
3005 * our own version based on RUC and ROC
3007 adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3008 adapter->stats.crcerrs + adapter->stats.algnerrc +
3009 adapter->stats.ruc + adapter->stats.roc +
3010 adapter->stats.cexterr;
3011 adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3013 adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3014 adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3015 adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3018 adapter->net_stats.tx_errors = adapter->stats.ecol +
3019 adapter->stats.latecol;
3020 adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3021 adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3022 adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3024 /* Tx Dropped needs to be maintained elsewhere */
3027 if (hw->phy.media_type == e1000_media_type_copper) {
3028 if ((adapter->link_speed == SPEED_1000) &&
3029 (!e1e_rphy(hw, PHY_1000T_STATUS, &phy_tmp))) {
3030 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3031 adapter->phy_stats.idle_errors += phy_tmp;
3035 /* Management Stats */
3036 adapter->stats.mgptc += er32(MGTPTC);
3037 adapter->stats.mgprc += er32(MGTPRC);
3038 adapter->stats.mgpdc += er32(MGTPDC);
3040 spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
3044 * e1000_phy_read_status - Update the PHY register status snapshot
3045 * @adapter: board private structure
3047 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3049 struct e1000_hw *hw = &adapter->hw;
3050 struct e1000_phy_regs *phy = &adapter->phy_regs;
3052 unsigned long irq_flags;
3055 spin_lock_irqsave(&adapter->stats_lock, irq_flags);
3057 if ((er32(STATUS) & E1000_STATUS_LU) &&
3058 (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3059 ret_val = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3060 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3061 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3062 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3063 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3064 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3065 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3066 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3068 e_warn("Error reading PHY register\n");
3071 * Do not read PHY registers if link is not up
3072 * Set values to typical power-on defaults
3074 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3075 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3076 BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3078 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3079 ADVERTISE_ALL | ADVERTISE_CSMA);
3081 phy->expansion = EXPANSION_ENABLENPAGE;
3082 phy->ctrl1000 = ADVERTISE_1000FULL;
3084 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3087 spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
3090 static void e1000_print_link_info(struct e1000_adapter *adapter)
3092 struct e1000_hw *hw = &adapter->hw;
3093 u32 ctrl = er32(CTRL);
3095 e_info("Link is Up %d Mbps %s, Flow Control: %s\n",
3096 adapter->link_speed,
3097 (adapter->link_duplex == FULL_DUPLEX) ?
3098 "Full Duplex" : "Half Duplex",
3099 ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3101 ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3102 ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3105 static bool e1000_has_link(struct e1000_adapter *adapter)
3107 struct e1000_hw *hw = &adapter->hw;
3108 bool link_active = 0;
3112 * get_link_status is set on LSC (link status) interrupt or
3113 * Rx sequence error interrupt. get_link_status will stay
3114 * false until the check_for_link establishes link
3115 * for copper adapters ONLY
3117 switch (hw->phy.media_type) {
3118 case e1000_media_type_copper:
3119 if (hw->mac.get_link_status) {
3120 ret_val = hw->mac.ops.check_for_link(hw);
3121 link_active = !hw->mac.get_link_status;
3126 case e1000_media_type_fiber:
3127 ret_val = hw->mac.ops.check_for_link(hw);
3128 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3130 case e1000_media_type_internal_serdes:
3131 ret_val = hw->mac.ops.check_for_link(hw);
3132 link_active = adapter->hw.mac.serdes_has_link;
3135 case e1000_media_type_unknown:
3139 if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3140 (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3141 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3142 e_info("Gigabit has been disabled, downgrading speed\n");
3148 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3150 /* make sure the receive unit is started */
3151 if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3152 (adapter->flags & FLAG_RX_RESTART_NOW)) {
3153 struct e1000_hw *hw = &adapter->hw;
3154 u32 rctl = er32(RCTL);
3155 ew32(RCTL, rctl | E1000_RCTL_EN);
3156 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3161 * e1000_watchdog - Timer Call-back
3162 * @data: pointer to adapter cast into an unsigned long
3164 static void e1000_watchdog(unsigned long data)
3166 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3168 /* Do the rest outside of interrupt context */
3169 schedule_work(&adapter->watchdog_task);
3171 /* TODO: make this use queue_delayed_work() */
3174 static void e1000_watchdog_task(struct work_struct *work)
3176 struct e1000_adapter *adapter = container_of(work,
3177 struct e1000_adapter, watchdog_task);
3178 struct net_device *netdev = adapter->netdev;
3179 struct e1000_mac_info *mac = &adapter->hw.mac;
3180 struct e1000_ring *tx_ring = adapter->tx_ring;
3181 struct e1000_hw *hw = &adapter->hw;
3185 link = e1000_has_link(adapter);
3186 if ((netif_carrier_ok(netdev)) && link) {
3187 e1000e_enable_receives(adapter);
3191 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3192 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3193 e1000_update_mng_vlan(adapter);
3196 if (!netif_carrier_ok(netdev)) {
3198 /* update snapshot of PHY registers on LSC */
3199 e1000_phy_read_status(adapter);
3200 mac->ops.get_link_up_info(&adapter->hw,
3201 &adapter->link_speed,
3202 &adapter->link_duplex);
3203 e1000_print_link_info(adapter);
3205 * On supported PHYs, check for duplex mismatch only
3206 * if link has autonegotiated at 10/100 half
3208 if ((hw->phy.type == e1000_phy_igp_3 ||
3209 hw->phy.type == e1000_phy_bm) &&
3210 (hw->mac.autoneg == true) &&
3211 (adapter->link_speed == SPEED_10 ||
3212 adapter->link_speed == SPEED_100) &&
3213 (adapter->link_duplex == HALF_DUPLEX)) {
3216 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3218 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3219 e_info("Autonegotiated half duplex but"
3220 " link partner cannot autoneg. "
3221 " Try forcing full duplex if "
3222 "link gets many collisions.\n");
3226 * tweak tx_queue_len according to speed/duplex
3227 * and adjust the timeout factor
3229 netdev->tx_queue_len = adapter->tx_queue_len;
3230 adapter->tx_timeout_factor = 1;
3231 switch (adapter->link_speed) {
3234 netdev->tx_queue_len = 10;
3235 adapter->tx_timeout_factor = 16;
3239 netdev->tx_queue_len = 100;
3240 /* maybe add some timeout factor ? */
3245 * workaround: re-program speed mode bit after
3248 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3251 tarc0 = er32(TARC(0));
3252 tarc0 &= ~SPEED_MODE_BIT;
3253 ew32(TARC(0), tarc0);
3257 * disable TSO for pcie and 10/100 speeds, to avoid
3258 * some hardware issues
3260 if (!(adapter->flags & FLAG_TSO_FORCE)) {
3261 switch (adapter->link_speed) {
3264 e_info("10/100 speed: disabling TSO\n");
3265 netdev->features &= ~NETIF_F_TSO;
3266 netdev->features &= ~NETIF_F_TSO6;
3269 netdev->features |= NETIF_F_TSO;
3270 netdev->features |= NETIF_F_TSO6;
3279 * enable transmits in the hardware, need to do this
3280 * after setting TARC(0)
3283 tctl |= E1000_TCTL_EN;
3286 netif_carrier_on(netdev);
3287 netif_tx_wake_all_queues(netdev);
3289 if (!test_bit(__E1000_DOWN, &adapter->state))
3290 mod_timer(&adapter->phy_info_timer,
3291 round_jiffies(jiffies + 2 * HZ));
3294 if (netif_carrier_ok(netdev)) {
3295 adapter->link_speed = 0;
3296 adapter->link_duplex = 0;
3297 e_info("Link is Down\n");
3298 netif_carrier_off(netdev);
3299 netif_tx_stop_all_queues(netdev);
3300 if (!test_bit(__E1000_DOWN, &adapter->state))
3301 mod_timer(&adapter->phy_info_timer,
3302 round_jiffies(jiffies + 2 * HZ));
3304 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3305 schedule_work(&adapter->reset_task);
3310 e1000e_update_stats(adapter);
3312 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3313 adapter->tpt_old = adapter->stats.tpt;
3314 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3315 adapter->colc_old = adapter->stats.colc;
3317 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3318 adapter->gorc_old = adapter->stats.gorc;
3319 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3320 adapter->gotc_old = adapter->stats.gotc;
3322 e1000e_update_adaptive(&adapter->hw);
3324 if (!netif_carrier_ok(netdev)) {
3325 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3329 * We've lost link, so the controller stops DMA,
3330 * but we've got queued Tx work that's never going
3331 * to get done, so reset controller to flush Tx.
3332 * (Do the reset outside of interrupt context).
3334 adapter->tx_timeout_count++;
3335 schedule_work(&adapter->reset_task);
3339 /* Cause software interrupt to ensure Rx ring is cleaned */
3340 ew32(ICS, E1000_ICS_RXDMT0);
3342 /* Force detection of hung controller every watchdog period */
3343 adapter->detect_tx_hung = 1;
3346 * With 82571 controllers, LAA may be overwritten due to controller
3347 * reset from the other port. Set the appropriate LAA in RAR[0]
3349 if (e1000e_get_laa_state_82571(hw))
3350 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3352 /* Reset the timer */
3353 if (!test_bit(__E1000_DOWN, &adapter->state))
3354 mod_timer(&adapter->watchdog_timer,
3355 round_jiffies(jiffies + 2 * HZ));
3358 #define E1000_TX_FLAGS_CSUM 0x00000001
3359 #define E1000_TX_FLAGS_VLAN 0x00000002
3360 #define E1000_TX_FLAGS_TSO 0x00000004
3361 #define E1000_TX_FLAGS_IPV4 0x00000008
3362 #define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
3363 #define E1000_TX_FLAGS_VLAN_SHIFT 16
3365 static int e1000_tso(struct e1000_adapter *adapter,
3366 struct sk_buff *skb)
3368 struct e1000_ring *tx_ring = adapter->tx_ring;
3369 struct e1000_context_desc *context_desc;
3370 struct e1000_buffer *buffer_info;
3373 u16 ipcse = 0, tucse, mss;
3374 u8 ipcss, ipcso, tucss, tucso, hdr_len;
3377 if (skb_is_gso(skb)) {
3378 if (skb_header_cloned(skb)) {
3379 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3384 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3385 mss = skb_shinfo(skb)->gso_size;
3386 if (skb->protocol == htons(ETH_P_IP)) {
3387 struct iphdr *iph = ip_hdr(skb);
3390 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3394 cmd_length = E1000_TXD_CMD_IP;
3395 ipcse = skb_transport_offset(skb) - 1;
3396 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3397 ipv6_hdr(skb)->payload_len = 0;
3398 tcp_hdr(skb)->check =
3399 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3400 &ipv6_hdr(skb)->daddr,
3404 ipcss = skb_network_offset(skb);
3405 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3406 tucss = skb_transport_offset(skb);
3407 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3410 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3411 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3413 i = tx_ring->next_to_use;
3414 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3415 buffer_info = &tx_ring->buffer_info[i];
3417 context_desc->lower_setup.ip_fields.ipcss = ipcss;
3418 context_desc->lower_setup.ip_fields.ipcso = ipcso;
3419 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
3420 context_desc->upper_setup.tcp_fields.tucss = tucss;
3421 context_desc->upper_setup.tcp_fields.tucso = tucso;
3422 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3423 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
3424 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3425 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3427 buffer_info->time_stamp = jiffies;
3428 buffer_info->next_to_watch = i;
3431 if (i == tx_ring->count)
3433 tx_ring->next_to_use = i;
3441 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3443 struct e1000_ring *tx_ring = adapter->tx_ring;
3444 struct e1000_context_desc *context_desc;
3445 struct e1000_buffer *buffer_info;
3449 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3450 css = skb_transport_offset(skb);
3452 i = tx_ring->next_to_use;
3453 buffer_info = &tx_ring->buffer_info[i];
3454 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3456 context_desc->lower_setup.ip_config = 0;
3457 context_desc->upper_setup.tcp_fields.tucss = css;
3458 context_desc->upper_setup.tcp_fields.tucso =
3459 css + skb->csum_offset;
3460 context_desc->upper_setup.tcp_fields.tucse = 0;
3461 context_desc->tcp_seg_setup.data = 0;
3462 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
3464 buffer_info->time_stamp = jiffies;
3465 buffer_info->next_to_watch = i;
3468 if (i == tx_ring->count)
3470 tx_ring->next_to_use = i;
3478 #define E1000_MAX_PER_TXD 8192
3479 #define E1000_MAX_TXD_PWR 12
3481 static int e1000_tx_map(struct e1000_adapter *adapter,
3482 struct sk_buff *skb, unsigned int first,
3483 unsigned int max_per_txd, unsigned int nr_frags,
3486 struct e1000_ring *tx_ring = adapter->tx_ring;
3487 struct e1000_buffer *buffer_info;
3488 unsigned int len = skb->len - skb->data_len;
3489 unsigned int offset = 0, size, count = 0, i;
3492 i = tx_ring->next_to_use;
3495 buffer_info = &tx_ring->buffer_info[i];
3496 size = min(len, max_per_txd);
3498 /* Workaround for premature desc write-backs
3499 * in TSO mode. Append 4-byte sentinel desc */
3500 if (mss && !nr_frags && size == len && size > 8)
3503 buffer_info->length = size;
3504 /* set time_stamp *before* dma to help avoid a possible race */
3505 buffer_info->time_stamp = jiffies;
3507 pci_map_single(adapter->pdev,
3511 if (pci_dma_mapping_error(adapter->pdev, buffer_info->dma)) {
3512 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3513 adapter->tx_dma_failed++;
3516 buffer_info->next_to_watch = i;
3522 if (i == tx_ring->count)
3526 for (f = 0; f < nr_frags; f++) {
3527 struct skb_frag_struct *frag;
3529 frag = &skb_shinfo(skb)->frags[f];
3531 offset = frag->page_offset;
3534 buffer_info = &tx_ring->buffer_info[i];
3535 size = min(len, max_per_txd);
3536 /* Workaround for premature desc write-backs
3537 * in TSO mode. Append 4-byte sentinel desc */
3538 if (mss && f == (nr_frags-1) && size == len && size > 8)
3541 buffer_info->length = size;
3542 buffer_info->time_stamp = jiffies;
3544 pci_map_page(adapter->pdev,
3549 if (pci_dma_mapping_error(adapter->pdev,
3550 buffer_info->dma)) {
3551 dev_err(&adapter->pdev->dev,
3552 "TX DMA page map failed\n");
3553 adapter->tx_dma_failed++;
3557 buffer_info->next_to_watch = i;
3564 if (i == tx_ring->count)
3570 i = tx_ring->count - 1;
3574 tx_ring->buffer_info[i].skb = skb;
3575 tx_ring->buffer_info[first].next_to_watch = i;
3580 static void e1000_tx_queue(struct e1000_adapter *adapter,
3581 int tx_flags, int count)
3583 struct e1000_ring *tx_ring = adapter->tx_ring;
3584 struct e1000_tx_desc *tx_desc = NULL;
3585 struct e1000_buffer *buffer_info;
3586 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3589 if (tx_flags & E1000_TX_FLAGS_TSO) {
3590 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3592 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3594 if (tx_flags & E1000_TX_FLAGS_IPV4)
3595 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3598 if (tx_flags & E1000_TX_FLAGS_CSUM) {
3599 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3600 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3603 if (tx_flags & E1000_TX_FLAGS_VLAN) {
3604 txd_lower |= E1000_TXD_CMD_VLE;
3605 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3608 i = tx_ring->next_to_use;
3611 buffer_info = &tx_ring->buffer_info[i];
3612 tx_desc = E1000_TX_DESC(*tx_ring, i);
3613 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3614 tx_desc->lower.data =
3615 cpu_to_le32(txd_lower | buffer_info->length);
3616 tx_desc->upper.data = cpu_to_le32(txd_upper);
3619 if (i == tx_ring->count)
3623 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3626 * Force memory writes to complete before letting h/w
3627 * know there are new descriptors to fetch. (Only
3628 * applicable for weak-ordered memory model archs,
3633 tx_ring->next_to_use = i;
3634 writel(i, adapter->hw.hw_addr + tx_ring->tail);
3636 * we need this if more than one processor can write to our tail
3637 * at a time, it synchronizes IO on IA64/Altix systems
3642 #define MINIMUM_DHCP_PACKET_SIZE 282
3643 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3644 struct sk_buff *skb)
3646 struct e1000_hw *hw = &adapter->hw;
3649 if (vlan_tx_tag_present(skb)) {
3650 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3651 && (adapter->hw.mng_cookie.status &
3652 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3656 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3659 if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3663 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3666 if (ip->protocol != IPPROTO_UDP)
3669 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3670 if (ntohs(udp->dest) != 67)
3673 offset = (u8 *)udp + 8 - skb->data;
3674 length = skb->len - offset;
3675 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
3681 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3683 struct e1000_adapter *adapter = netdev_priv(netdev);
3685 netif_stop_queue(netdev);
3687 * Herbert's original patch had:
3688 * smp_mb__after_netif_stop_queue();
3689 * but since that doesn't exist yet, just open code it.
3694 * We need to check again in a case another CPU has just
3695 * made room available.
3697 if (e1000_desc_unused(adapter->tx_ring) < size)
3701 netif_start_queue(netdev);
3702 ++adapter->restart_queue;
3706 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
3708 struct e1000_adapter *adapter = netdev_priv(netdev);
3710 if (e1000_desc_unused(adapter->tx_ring) >= size)
3712 return __e1000_maybe_stop_tx(netdev, size);
3715 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3716 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3718 struct e1000_adapter *adapter = netdev_priv(netdev);
3719 struct e1000_ring *tx_ring = adapter->tx_ring;
3721 unsigned int max_per_txd = E1000_MAX_PER_TXD;
3722 unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3723 unsigned int tx_flags = 0;
3724 unsigned int len = skb->len - skb->data_len;
3725 unsigned long irq_flags;
3726 unsigned int nr_frags;
3732 if (test_bit(__E1000_DOWN, &adapter->state)) {
3733 dev_kfree_skb_any(skb);
3734 return NETDEV_TX_OK;
3737 if (skb->len <= 0) {
3738 dev_kfree_skb_any(skb);
3739 return NETDEV_TX_OK;
3742 mss = skb_shinfo(skb)->gso_size;
3744 * The controller does a simple calculation to
3745 * make sure there is enough room in the FIFO before
3746 * initiating the DMA for each buffer. The calc is:
3747 * 4 = ceil(buffer len/mss). To make sure we don't
3748 * overrun the FIFO, adjust the max buffer len if mss
3753 max_per_txd = min(mss << 2, max_per_txd);
3754 max_txd_pwr = fls(max_per_txd) - 1;
3757 * TSO Workaround for 82571/2/3 Controllers -- if skb->data
3758 * points to just header, pull a few bytes of payload from
3759 * frags into skb->data
3761 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3763 * we do this workaround for ES2LAN, but it is un-necessary,
3764 * avoiding it could save a lot of cycles
3766 if (skb->data_len && (hdr_len == len)) {
3767 unsigned int pull_size;
3769 pull_size = min((unsigned int)4, skb->data_len);
3770 if (!__pskb_pull_tail(skb, pull_size)) {
3771 e_err("__pskb_pull_tail failed.\n");
3772 dev_kfree_skb_any(skb);
3773 return NETDEV_TX_OK;
3775 len = skb->len - skb->data_len;
3779 /* reserve a descriptor for the offload context */
3780 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3784 count += TXD_USE_COUNT(len, max_txd_pwr);
3786 nr_frags = skb_shinfo(skb)->nr_frags;
3787 for (f = 0; f < nr_frags; f++)
3788 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3791 if (adapter->hw.mac.tx_pkt_filtering)
3792 e1000_transfer_dhcp_info(adapter, skb);
3794 if (!spin_trylock_irqsave(&adapter->tx_queue_lock, irq_flags))
3795 /* Collision - tell upper layer to requeue */
3796 return NETDEV_TX_LOCKED;
3799 * need: count + 2 desc gap to keep tail from touching
3800 * head, otherwise try next time
3802 if (e1000_maybe_stop_tx(netdev, count + 2)) {
3803 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3804 return NETDEV_TX_BUSY;
3807 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3808 tx_flags |= E1000_TX_FLAGS_VLAN;
3809 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3812 first = tx_ring->next_to_use;
3814 tso = e1000_tso(adapter, skb);
3816 dev_kfree_skb_any(skb);
3817 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3818 return NETDEV_TX_OK;
3822 tx_flags |= E1000_TX_FLAGS_TSO;
3823 else if (e1000_tx_csum(adapter, skb))
3824 tx_flags |= E1000_TX_FLAGS_CSUM;
3827 * Old method was to assume IPv4 packet by default if TSO was enabled.
3828 * 82571 hardware supports TSO capabilities for IPv6 as well...
3829 * no longer assume, we must.
3831 if (skb->protocol == htons(ETH_P_IP))
3832 tx_flags |= E1000_TX_FLAGS_IPV4;
3834 count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
3836 /* handle pci_map_single() error in e1000_tx_map */
3837 dev_kfree_skb_any(skb);
3838 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3839 return NETDEV_TX_OK;
3842 e1000_tx_queue(adapter, tx_flags, count);
3844 netdev->trans_start = jiffies;
3846 /* Make sure there is space in the ring for the next send. */
3847 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
3849 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3850 return NETDEV_TX_OK;
3854 * e1000_tx_timeout - Respond to a Tx Hang
3855 * @netdev: network interface device structure
3857 static void e1000_tx_timeout(struct net_device *netdev)
3859 struct e1000_adapter *adapter = netdev_priv(netdev);
3861 /* Do the reset outside of interrupt context */
3862 adapter->tx_timeout_count++;
3863 schedule_work(&adapter->reset_task);
3866 static void e1000_reset_task(struct work_struct *work)
3868 struct e1000_adapter *adapter;
3869 adapter = container_of(work, struct e1000_adapter, reset_task);
3871 e1000e_reinit_locked(adapter);
3875 * e1000_get_stats - Get System Network Statistics
3876 * @netdev: network interface device structure
3878 * Returns the address of the device statistics structure.
3879 * The statistics are actually updated from the timer callback.
3881 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3883 struct e1000_adapter *adapter = netdev_priv(netdev);
3885 /* only return the current stats */
3886 return &adapter->net_stats;
3890 * e1000_change_mtu - Change the Maximum Transfer Unit
3891 * @netdev: network interface device structure
3892 * @new_mtu: new value for maximum frame size
3894 * Returns 0 on success, negative on failure
3896 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3898 struct e1000_adapter *adapter = netdev_priv(netdev);
3899 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3901 if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
3902 (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3903 e_err("Invalid MTU setting\n");
3907 /* Jumbo frame size limits */
3908 if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
3909 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
3910 e_err("Jumbo Frames not supported.\n");
3913 if (adapter->hw.phy.type == e1000_phy_ife) {
3914 e_err("Jumbo Frames not supported.\n");
3919 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3920 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3921 e_err("MTU > 9216 not supported.\n");
3925 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3927 /* e1000e_down has a dependency on max_frame_size */
3928 adapter->max_frame_size = max_frame;
3929 if (netif_running(netdev))
3930 e1000e_down(adapter);
3933 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3934 * means we reserve 2 more, this pushes us to allocate from the next
3936 * i.e. RXBUFFER_2048 --> size-4096 slab
3937 * However with the new *_jumbo_rx* routines, jumbo receives will use
3941 if (max_frame <= 256)
3942 adapter->rx_buffer_len = 256;
3943 else if (max_frame <= 512)
3944 adapter->rx_buffer_len = 512;
3945 else if (max_frame <= 1024)
3946 adapter->rx_buffer_len = 1024;
3947 else if (max_frame <= 2048)
3948 adapter->rx_buffer_len = 2048;
3950 adapter->rx_buffer_len = 4096;
3952 /* adjust allocation if LPE protects us, and we aren't using SBP */
3953 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3954 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
3955 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
3958 e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
3959 netdev->mtu = new_mtu;
3961 if (netif_running(netdev))
3964 e1000e_reset(adapter);
3966 clear_bit(__E1000_RESETTING, &adapter->state);
3971 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
3974 struct e1000_adapter *adapter = netdev_priv(netdev);
3975 struct mii_ioctl_data *data = if_mii(ifr);
3977 if (adapter->hw.phy.media_type != e1000_media_type_copper)
3982 data->phy_id = adapter->hw.phy.addr;
3985 if (!capable(CAP_NET_ADMIN))
3987 switch (data->reg_num & 0x1F) {
3989 data->val_out = adapter->phy_regs.bmcr;
3992 data->val_out = adapter->phy_regs.bmsr;
3995 data->val_out = (adapter->hw.phy.id >> 16);
3998 data->val_out = (adapter->hw.phy.id & 0xFFFF);
4001 data->val_out = adapter->phy_regs.advertise;
4004 data->val_out = adapter->phy_regs.lpa;
4007 data->val_out = adapter->phy_regs.expansion;
4010 data->val_out = adapter->phy_regs.ctrl1000;
4013 data->val_out = adapter->phy_regs.stat1000;
4016 data->val_out = adapter->phy_regs.estatus;
4029 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4035 return e1000_mii_ioctl(netdev, ifr, cmd);
4041 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4043 struct net_device *netdev = pci_get_drvdata(pdev);
4044 struct e1000_adapter *adapter = netdev_priv(netdev);
4045 struct e1000_hw *hw = &adapter->hw;
4046 u32 ctrl, ctrl_ext, rctl, status;
4047 u32 wufc = adapter->wol;
4050 netif_device_detach(netdev);
4052 if (netif_running(netdev)) {
4053 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4054 e1000e_down(adapter);
4055 e1000_free_irq(adapter);
4058 retval = pci_save_state(pdev);
4062 status = er32(STATUS);
4063 if (status & E1000_STATUS_LU)
4064 wufc &= ~E1000_WUFC_LNKC;
4067 e1000_setup_rctl(adapter);
4068 e1000_set_multi(netdev);
4070 /* turn on all-multi mode if wake on multicast is enabled */
4071 if (wufc & E1000_WUFC_MC) {
4073 rctl |= E1000_RCTL_MPE;
4078 /* advertise wake from D3Cold */
4079 #define E1000_CTRL_ADVD3WUC 0x00100000
4080 /* phy power management enable */
4081 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4082 ctrl |= E1000_CTRL_ADVD3WUC |
4083 E1000_CTRL_EN_PHY_PWR_MGMT;
4086 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4087 adapter->hw.phy.media_type ==
4088 e1000_media_type_internal_serdes) {
4089 /* keep the laser running in D3 */
4090 ctrl_ext = er32(CTRL_EXT);
4091 ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4092 ew32(CTRL_EXT, ctrl_ext);
4095 if (adapter->flags & FLAG_IS_ICH)
4096 e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4098 /* Allow time for pending master requests to run */
4099 e1000e_disable_pcie_master(&adapter->hw);
4101 ew32(WUC, E1000_WUC_PME_EN);
4103 pci_enable_wake(pdev, PCI_D3hot, 1);
4104 pci_enable_wake(pdev, PCI_D3cold, 1);
4108 pci_enable_wake(pdev, PCI_D3hot, 0);
4109 pci_enable_wake(pdev, PCI_D3cold, 0);
4112 /* make sure adapter isn't asleep if manageability is enabled */
4113 if (adapter->flags & FLAG_MNG_PT_ENABLED) {
4114 pci_enable_wake(pdev, PCI_D3hot, 1);
4115 pci_enable_wake(pdev, PCI_D3cold, 1);
4118 if (adapter->hw.phy.type == e1000_phy_igp_3)
4119 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4122 * Release control of h/w to f/w. If f/w is AMT enabled, this
4123 * would have already happened in close and is redundant.
4125 e1000_release_hw_control(adapter);
4127 pci_disable_device(pdev);
4129 pci_set_power_state(pdev, pci_choose_state(pdev, state));
4134 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4140 * 82573 workaround - disable L1 ASPM on mobile chipsets
4142 * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4143 * resulting in lost data or garbage information on the pci-e link
4144 * level. This could result in (false) bad EEPROM checksum errors,
4145 * long ping times (up to 2s) or even a system freeze/hang.
4147 * Unfortunately this feature saves about 1W power consumption when
4150 pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4151 pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4153 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4155 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4160 static int e1000_resume(struct pci_dev *pdev)
4162 struct net_device *netdev = pci_get_drvdata(pdev);
4163 struct e1000_adapter *adapter = netdev_priv(netdev);
4164 struct e1000_hw *hw = &adapter->hw;
4167 pci_set_power_state(pdev, PCI_D0);
4168 pci_restore_state(pdev);
4169 e1000e_disable_l1aspm(pdev);
4171 err = pci_enable_device_mem(pdev);
4174 "Cannot enable PCI device from suspend\n");
4178 pci_set_master(pdev);
4180 pci_enable_wake(pdev, PCI_D3hot, 0);
4181 pci_enable_wake(pdev, PCI_D3cold, 0);
4183 if (netif_running(netdev)) {
4184 err = e1000_request_irq(adapter);
4189 e1000e_power_up_phy(adapter);
4190 e1000e_reset(adapter);
4193 e1000_init_manageability(adapter);
4195 if (netif_running(netdev))
4198 netif_device_attach(netdev);
4201 * If the controller has AMT, do not set DRV_LOAD until the interface
4202 * is up. For all other cases, let the f/w know that the h/w is now
4203 * under the control of the driver.
4205 if (!(adapter->flags & FLAG_HAS_AMT))
4206 e1000_get_hw_control(adapter);
4212 static void e1000_shutdown(struct pci_dev *pdev)
4214 e1000_suspend(pdev, PMSG_SUSPEND);
4217 #ifdef CONFIG_NET_POLL_CONTROLLER
4219 * Polling 'interrupt' - used by things like netconsole to send skbs
4220 * without having to re-enable interrupts. It's not called while
4221 * the interrupt routine is executing.
4223 static void e1000_netpoll(struct net_device *netdev)
4225 struct e1000_adapter *adapter = netdev_priv(netdev);
4227 disable_irq(adapter->pdev->irq);
4228 e1000_intr(adapter->pdev->irq, netdev);
4230 enable_irq(adapter->pdev->irq);
4235 * e1000_io_error_detected - called when PCI error is detected
4236 * @pdev: Pointer to PCI device
4237 * @state: The current pci connection state
4239 * This function is called after a PCI bus error affecting
4240 * this device has been detected.
4242 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4243 pci_channel_state_t state)
4245 struct net_device *netdev = pci_get_drvdata(pdev);
4246 struct e1000_adapter *adapter = netdev_priv(netdev);
4248 netif_device_detach(netdev);
4250 if (netif_running(netdev))
4251 e1000e_down(adapter);
4252 pci_disable_device(pdev);
4254 /* Request a slot slot reset. */
4255 return PCI_ERS_RESULT_NEED_RESET;
4259 * e1000_io_slot_reset - called after the pci bus has been reset.
4260 * @pdev: Pointer to PCI device
4262 * Restart the card from scratch, as if from a cold-boot. Implementation
4263 * resembles the first-half of the e1000_resume routine.
4265 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4267 struct net_device *netdev = pci_get_drvdata(pdev);
4268 struct e1000_adapter *adapter = netdev_priv(netdev);
4269 struct e1000_hw *hw = &adapter->hw;
4272 e1000e_disable_l1aspm(pdev);
4273 err = pci_enable_device_mem(pdev);
4276 "Cannot re-enable PCI device after reset.\n");
4277 return PCI_ERS_RESULT_DISCONNECT;
4279 pci_set_master(pdev);
4280 pci_restore_state(pdev);
4282 pci_enable_wake(pdev, PCI_D3hot, 0);
4283 pci_enable_wake(pdev, PCI_D3cold, 0);
4285 e1000e_reset(adapter);
4288 return PCI_ERS_RESULT_RECOVERED;
4292 * e1000_io_resume - called when traffic can start flowing again.
4293 * @pdev: Pointer to PCI device
4295 * This callback is called when the error recovery driver tells us that
4296 * its OK to resume normal operation. Implementation resembles the
4297 * second-half of the e1000_resume routine.
4299 static void e1000_io_resume(struct pci_dev *pdev)
4301 struct net_device *netdev = pci_get_drvdata(pdev);
4302 struct e1000_adapter *adapter = netdev_priv(netdev);
4304 e1000_init_manageability(adapter);
4306 if (netif_running(netdev)) {
4307 if (e1000e_up(adapter)) {
4309 "can't bring device back up after reset\n");
4314 netif_device_attach(netdev);
4317 * If the controller has AMT, do not set DRV_LOAD until the interface
4318 * is up. For all other cases, let the f/w know that the h/w is now
4319 * under the control of the driver.
4321 if (!(adapter->flags & FLAG_HAS_AMT))
4322 e1000_get_hw_control(adapter);
4326 static void e1000_print_device_info(struct e1000_adapter *adapter)
4328 struct e1000_hw *hw = &adapter->hw;
4329 struct net_device *netdev = adapter->netdev;
4332 /* print bus type/speed/width info */
4333 e_info("(PCI Express:2.5GB/s:%s) %02x:%02x:%02x:%02x:%02x:%02x\n",
4335 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4338 netdev->dev_addr[0], netdev->dev_addr[1],
4339 netdev->dev_addr[2], netdev->dev_addr[3],
4340 netdev->dev_addr[4], netdev->dev_addr[5]);
4341 e_info("Intel(R) PRO/%s Network Connection\n",
4342 (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4343 e1000e_read_pba_num(hw, &pba_num);
4344 e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4345 hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4348 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4350 struct e1000_hw *hw = &adapter->hw;
4354 if (hw->mac.type != e1000_82573)
4357 ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4358 if (!(le16_to_cpu(buf) & (1 << 0))) {
4359 /* Deep Smart Power Down (DSPD) */
4360 e_warn("Warning: detected DSPD enabled in EEPROM\n");
4363 ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4364 if (le16_to_cpu(buf) & (3 << 2)) {
4366 e_warn("Warning: detected ASPM enabled in EEPROM\n");
4371 * e1000_probe - Device Initialization Routine
4372 * @pdev: PCI device information struct
4373 * @ent: entry in e1000_pci_tbl
4375 * Returns 0 on success, negative on failure
4377 * e1000_probe initializes an adapter identified by a pci_dev structure.
4378 * The OS initialization, configuring of the adapter private structure,
4379 * and a hardware reset occur.
4381 static int __devinit e1000_probe(struct pci_dev *pdev,
4382 const struct pci_device_id *ent)
4384 struct net_device *netdev;
4385 struct e1000_adapter *adapter;
4386 struct e1000_hw *hw;
4387 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4388 resource_size_t mmio_start, mmio_len;
4389 resource_size_t flash_start, flash_len;
4391 static int cards_found;
4392 int i, err, pci_using_dac;
4393 u16 eeprom_data = 0;
4394 u16 eeprom_apme_mask = E1000_EEPROM_APME;
4396 e1000e_disable_l1aspm(pdev);
4398 err = pci_enable_device_mem(pdev);
4403 err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
4405 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
4409 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
4411 err = pci_set_consistent_dma_mask(pdev,
4414 dev_err(&pdev->dev, "No usable DMA "
4415 "configuration, aborting\n");
4421 err = pci_request_selected_regions(pdev,
4422 pci_select_bars(pdev, IORESOURCE_MEM),
4423 e1000e_driver_name);
4427 pci_set_master(pdev);
4428 pci_save_state(pdev);
4431 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4433 goto err_alloc_etherdev;
4435 SET_NETDEV_DEV(netdev, &pdev->dev);
4437 pci_set_drvdata(pdev, netdev);
4438 adapter = netdev_priv(netdev);
4440 adapter->netdev = netdev;
4441 adapter->pdev = pdev;
4443 adapter->pba = ei->pba;
4444 adapter->flags = ei->flags;
4445 adapter->hw.adapter = adapter;
4446 adapter->hw.mac.type = ei->mac;
4447 adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4449 mmio_start = pci_resource_start(pdev, 0);
4450 mmio_len = pci_resource_len(pdev, 0);
4453 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4454 if (!adapter->hw.hw_addr)
4457 if ((adapter->flags & FLAG_HAS_FLASH) &&
4458 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
4459 flash_start = pci_resource_start(pdev, 1);
4460 flash_len = pci_resource_len(pdev, 1);
4461 adapter->hw.flash_address = ioremap(flash_start, flash_len);
4462 if (!adapter->hw.flash_address)
4466 /* construct the net_device struct */
4467 netdev->open = &e1000_open;
4468 netdev->stop = &e1000_close;
4469 netdev->hard_start_xmit = &e1000_xmit_frame;
4470 netdev->get_stats = &e1000_get_stats;
4471 netdev->set_multicast_list = &e1000_set_multi;
4472 netdev->set_mac_address = &e1000_set_mac;
4473 netdev->change_mtu = &e1000_change_mtu;
4474 netdev->do_ioctl = &e1000_ioctl;
4475 e1000e_set_ethtool_ops(netdev);
4476 netdev->tx_timeout = &e1000_tx_timeout;
4477 netdev->watchdog_timeo = 5 * HZ;
4478 netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
4479 netdev->vlan_rx_register = e1000_vlan_rx_register;
4480 netdev->vlan_rx_add_vid = e1000_vlan_rx_add_vid;
4481 netdev->vlan_rx_kill_vid = e1000_vlan_rx_kill_vid;
4482 #ifdef CONFIG_NET_POLL_CONTROLLER
4483 netdev->poll_controller = e1000_netpoll;
4485 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
4487 netdev->mem_start = mmio_start;
4488 netdev->mem_end = mmio_start + mmio_len;
4490 adapter->bd_number = cards_found++;
4492 /* setup adapter struct */
4493 err = e1000_sw_init(adapter);
4499 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4500 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
4501 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4503 err = ei->get_variants(adapter);
4507 hw->mac.ops.get_bus_info(&adapter->hw);
4509 adapter->hw.phy.autoneg_wait_to_complete = 0;
4511 /* Copper options */
4512 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
4513 adapter->hw.phy.mdix = AUTO_ALL_MODES;
4514 adapter->hw.phy.disable_polarity_correction = 0;
4515 adapter->hw.phy.ms_type = e1000_ms_hw_default;
4518 if (e1000_check_reset_block(&adapter->hw))
4519 e_info("PHY reset is blocked due to SOL/IDER session.\n");
4521 netdev->features = NETIF_F_SG |
4523 NETIF_F_HW_VLAN_TX |
4526 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
4527 netdev->features |= NETIF_F_HW_VLAN_FILTER;
4529 netdev->features |= NETIF_F_TSO;
4530 netdev->features |= NETIF_F_TSO6;
4532 netdev->vlan_features |= NETIF_F_TSO;
4533 netdev->vlan_features |= NETIF_F_TSO6;
4534 netdev->vlan_features |= NETIF_F_HW_CSUM;
4535 netdev->vlan_features |= NETIF_F_SG;
4538 netdev->features |= NETIF_F_HIGHDMA;
4541 * We should not be using LLTX anymore, but we are still Tx faster with
4544 netdev->features |= NETIF_F_LLTX;
4546 if (e1000e_enable_mng_pass_thru(&adapter->hw))
4547 adapter->flags |= FLAG_MNG_PT_ENABLED;
4550 * before reading the NVM, reset the controller to
4551 * put the device in a known good starting state
4553 adapter->hw.mac.ops.reset_hw(&adapter->hw);
4556 * systems with ASPM and others may see the checksum fail on the first
4557 * attempt. Let's give it a few tries
4560 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
4563 e_err("The NVM Checksum Is Not Valid\n");
4569 e1000_eeprom_checks(adapter);
4571 /* copy the MAC address out of the NVM */
4572 if (e1000e_read_mac_addr(&adapter->hw))
4573 e_err("NVM Read Error while reading MAC address\n");
4575 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
4576 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
4578 if (!is_valid_ether_addr(netdev->perm_addr)) {
4579 e_err("Invalid MAC Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
4580 netdev->perm_addr[0], netdev->perm_addr[1],
4581 netdev->perm_addr[2], netdev->perm_addr[3],
4582 netdev->perm_addr[4], netdev->perm_addr[5]);
4587 init_timer(&adapter->watchdog_timer);
4588 adapter->watchdog_timer.function = &e1000_watchdog;
4589 adapter->watchdog_timer.data = (unsigned long) adapter;
4591 init_timer(&adapter->phy_info_timer);
4592 adapter->phy_info_timer.function = &e1000_update_phy_info;
4593 adapter->phy_info_timer.data = (unsigned long) adapter;
4595 INIT_WORK(&adapter->reset_task, e1000_reset_task);
4596 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
4598 e1000e_check_options(adapter);
4600 /* Initialize link parameters. User can change them with ethtool */
4601 adapter->hw.mac.autoneg = 1;
4602 adapter->fc_autoneg = 1;
4603 adapter->hw.fc.original_type = e1000_fc_default;
4604 adapter->hw.fc.type = e1000_fc_default;
4605 adapter->hw.phy.autoneg_advertised = 0x2f;
4607 /* ring size defaults */
4608 adapter->rx_ring->count = 256;
4609 adapter->tx_ring->count = 256;
4612 * Initial Wake on LAN setting - If APM wake is enabled in
4613 * the EEPROM, enable the ACPI Magic Packet filter
4615 if (adapter->flags & FLAG_APME_IN_WUC) {
4616 /* APME bit in EEPROM is mapped to WUC.APME */
4617 eeprom_data = er32(WUC);
4618 eeprom_apme_mask = E1000_WUC_APME;
4619 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
4620 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
4621 (adapter->hw.bus.func == 1))
4622 e1000_read_nvm(&adapter->hw,
4623 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
4625 e1000_read_nvm(&adapter->hw,
4626 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
4629 /* fetch WoL from EEPROM */
4630 if (eeprom_data & eeprom_apme_mask)
4631 adapter->eeprom_wol |= E1000_WUFC_MAG;
4634 * now that we have the eeprom settings, apply the special cases
4635 * where the eeprom may be wrong or the board simply won't support
4636 * wake on lan on a particular port
4638 if (!(adapter->flags & FLAG_HAS_WOL))
4639 adapter->eeprom_wol = 0;
4641 /* initialize the wol settings based on the eeprom settings */
4642 adapter->wol = adapter->eeprom_wol;
4644 /* reset the hardware with the new settings */
4645 e1000e_reset(adapter);
4648 * If the controller has AMT, do not set DRV_LOAD until the interface
4649 * is up. For all other cases, let the f/w know that the h/w is now
4650 * under the control of the driver.
4652 if (!(adapter->flags & FLAG_HAS_AMT))
4653 e1000_get_hw_control(adapter);
4655 /* tell the stack to leave us alone until e1000_open() is called */
4656 netif_carrier_off(netdev);
4657 netif_tx_stop_all_queues(netdev);
4659 strcpy(netdev->name, "eth%d");
4660 err = register_netdev(netdev);
4664 e1000_print_device_info(adapter);
4669 if (!(adapter->flags & FLAG_HAS_AMT))
4670 e1000_release_hw_control(adapter);
4672 if (!e1000_check_reset_block(&adapter->hw))
4673 e1000_phy_hw_reset(&adapter->hw);
4676 kfree(adapter->tx_ring);
4677 kfree(adapter->rx_ring);
4679 if (adapter->hw.flash_address)
4680 iounmap(adapter->hw.flash_address);
4682 iounmap(adapter->hw.hw_addr);
4684 free_netdev(netdev);
4686 pci_release_selected_regions(pdev,
4687 pci_select_bars(pdev, IORESOURCE_MEM));
4690 pci_disable_device(pdev);
4695 * e1000_remove - Device Removal Routine
4696 * @pdev: PCI device information struct
4698 * e1000_remove is called by the PCI subsystem to alert the driver
4699 * that it should release a PCI device. The could be caused by a
4700 * Hot-Plug event, or because the driver is going to be removed from
4703 static void __devexit e1000_remove(struct pci_dev *pdev)
4705 struct net_device *netdev = pci_get_drvdata(pdev);
4706 struct e1000_adapter *adapter = netdev_priv(netdev);
4709 * flush_scheduled work may reschedule our watchdog task, so
4710 * explicitly disable watchdog tasks from being rescheduled
4712 set_bit(__E1000_DOWN, &adapter->state);
4713 del_timer_sync(&adapter->watchdog_timer);
4714 del_timer_sync(&adapter->phy_info_timer);
4716 flush_scheduled_work();
4719 * Release control of h/w to f/w. If f/w is AMT enabled, this
4720 * would have already happened in close and is redundant.
4722 e1000_release_hw_control(adapter);
4724 unregister_netdev(netdev);
4726 if (!e1000_check_reset_block(&adapter->hw))
4727 e1000_phy_hw_reset(&adapter->hw);
4729 kfree(adapter->tx_ring);
4730 kfree(adapter->rx_ring);
4732 iounmap(adapter->hw.hw_addr);
4733 if (adapter->hw.flash_address)
4734 iounmap(adapter->hw.flash_address);
4735 pci_release_selected_regions(pdev,
4736 pci_select_bars(pdev, IORESOURCE_MEM));
4738 free_netdev(netdev);
4740 pci_disable_device(pdev);
4743 /* PCI Error Recovery (ERS) */
4744 static struct pci_error_handlers e1000_err_handler = {
4745 .error_detected = e1000_io_error_detected,
4746 .slot_reset = e1000_io_slot_reset,
4747 .resume = e1000_io_resume,
4750 static struct pci_device_id e1000_pci_tbl[] = {
4751 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
4752 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
4753 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
4754 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
4755 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
4756 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
4757 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
4758 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
4759 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
4761 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
4762 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
4763 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
4764 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
4766 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
4767 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
4768 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
4770 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
4771 board_80003es2lan },
4772 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
4773 board_80003es2lan },
4774 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
4775 board_80003es2lan },
4776 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
4777 board_80003es2lan },
4779 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
4780 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
4781 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
4782 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
4783 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
4784 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
4785 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
4787 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
4788 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
4789 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
4790 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
4791 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
4792 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
4793 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
4794 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
4795 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
4797 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
4798 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
4799 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
4801 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
4802 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
4804 { } /* terminate list */
4806 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
4808 /* PCI Device API Driver */
4809 static struct pci_driver e1000_driver = {
4810 .name = e1000e_driver_name,
4811 .id_table = e1000_pci_tbl,
4812 .probe = e1000_probe,
4813 .remove = __devexit_p(e1000_remove),
4815 /* Power Management Hooks */
4816 .suspend = e1000_suspend,
4817 .resume = e1000_resume,
4819 .shutdown = e1000_shutdown,
4820 .err_handler = &e1000_err_handler
4824 * e1000_init_module - Driver Registration Routine
4826 * e1000_init_module is the first routine called when the driver is
4827 * loaded. All it does is register with the PCI subsystem.
4829 static int __init e1000_init_module(void)
4832 printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
4833 e1000e_driver_name, e1000e_driver_version);
4834 printk(KERN_INFO "%s: Copyright (c) 1999-2008 Intel Corporation.\n",
4835 e1000e_driver_name);
4836 ret = pci_register_driver(&e1000_driver);
4837 pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name,
4838 PM_QOS_DEFAULT_VALUE);
4842 module_init(e1000_init_module);
4845 * e1000_exit_module - Driver Exit Cleanup Routine
4847 * e1000_exit_module is called just before the driver is removed
4850 static void __exit e1000_exit_module(void)
4852 pci_unregister_driver(&e1000_driver);
4853 pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name);
4855 module_exit(e1000_exit_module);
4858 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
4859 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
4860 MODULE_LICENSE("GPL");
4861 MODULE_VERSION(DRV_VERSION);