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-k6"
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_82574
] = &e1000_82574_info
,
59 [board_80003es2lan
] = &e1000_es2_info
,
60 [board_ich8lan
] = &e1000_ich8_info
,
61 [board_ich9lan
] = &e1000_ich9_info
,
62 [board_ich10lan
] = &e1000_ich10_info
,
67 * e1000_get_hw_dev_name - return device name string
68 * used by hardware layer to print debugging information
70 char *e1000e_get_hw_dev_name(struct e1000_hw
*hw
)
72 return hw
->adapter
->netdev
->name
;
77 * e1000_desc_unused - calculate if we have unused descriptors
79 static int e1000_desc_unused(struct e1000_ring
*ring
)
81 if (ring
->next_to_clean
> ring
->next_to_use
)
82 return ring
->next_to_clean
- ring
->next_to_use
- 1;
84 return ring
->count
+ ring
->next_to_clean
- ring
->next_to_use
- 1;
88 * e1000_receive_skb - helper function to handle Rx indications
89 * @adapter: board private structure
90 * @status: descriptor status field as written by hardware
91 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
92 * @skb: pointer to sk_buff to be indicated to stack
94 static void e1000_receive_skb(struct e1000_adapter
*adapter
,
95 struct net_device
*netdev
,
97 u8 status
, __le16 vlan
)
99 skb
->protocol
= eth_type_trans(skb
, netdev
);
101 if (adapter
->vlgrp
&& (status
& E1000_RXD_STAT_VP
))
102 vlan_hwaccel_receive_skb(skb
, adapter
->vlgrp
,
105 napi_gro_receive(&adapter
->napi
, skb
);
109 * e1000_rx_checksum - Receive Checksum Offload for 82543
110 * @adapter: board private structure
111 * @status_err: receive descriptor status and error fields
112 * @csum: receive descriptor csum field
113 * @sk_buff: socket buffer with received data
115 static void e1000_rx_checksum(struct e1000_adapter
*adapter
, u32 status_err
,
116 u32 csum
, struct sk_buff
*skb
)
118 u16 status
= (u16
)status_err
;
119 u8 errors
= (u8
)(status_err
>> 24);
120 skb
->ip_summed
= CHECKSUM_NONE
;
122 /* Ignore Checksum bit is set */
123 if (status
& E1000_RXD_STAT_IXSM
)
125 /* TCP/UDP checksum error bit is set */
126 if (errors
& E1000_RXD_ERR_TCPE
) {
127 /* let the stack verify checksum errors */
128 adapter
->hw_csum_err
++;
132 /* TCP/UDP Checksum has not been calculated */
133 if (!(status
& (E1000_RXD_STAT_TCPCS
| E1000_RXD_STAT_UDPCS
)))
136 /* It must be a TCP or UDP packet with a valid checksum */
137 if (status
& E1000_RXD_STAT_TCPCS
) {
138 /* TCP checksum is good */
139 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
142 * IP fragment with UDP payload
143 * Hardware complements the payload checksum, so we undo it
144 * and then put the value in host order for further stack use.
146 __sum16 sum
= (__force __sum16
)htons(csum
);
147 skb
->csum
= csum_unfold(~sum
);
148 skb
->ip_summed
= CHECKSUM_COMPLETE
;
150 adapter
->hw_csum_good
++;
154 * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
155 * @adapter: address of board private structure
157 static void e1000_alloc_rx_buffers(struct e1000_adapter
*adapter
,
160 struct net_device
*netdev
= adapter
->netdev
;
161 struct pci_dev
*pdev
= adapter
->pdev
;
162 struct e1000_ring
*rx_ring
= adapter
->rx_ring
;
163 struct e1000_rx_desc
*rx_desc
;
164 struct e1000_buffer
*buffer_info
;
167 unsigned int bufsz
= adapter
->rx_buffer_len
+ NET_IP_ALIGN
;
169 i
= rx_ring
->next_to_use
;
170 buffer_info
= &rx_ring
->buffer_info
[i
];
172 while (cleaned_count
--) {
173 skb
= buffer_info
->skb
;
179 skb
= netdev_alloc_skb(netdev
, bufsz
);
181 /* Better luck next round */
182 adapter
->alloc_rx_buff_failed
++;
187 * Make buffer alignment 2 beyond a 16 byte boundary
188 * this will result in a 16 byte aligned IP header after
189 * the 14 byte MAC header is removed
191 skb_reserve(skb
, NET_IP_ALIGN
);
193 buffer_info
->skb
= skb
;
195 buffer_info
->dma
= pci_map_single(pdev
, skb
->data
,
196 adapter
->rx_buffer_len
,
198 if (pci_dma_mapping_error(pdev
, buffer_info
->dma
)) {
199 dev_err(&pdev
->dev
, "RX DMA map failed\n");
200 adapter
->rx_dma_failed
++;
204 rx_desc
= E1000_RX_DESC(*rx_ring
, i
);
205 rx_desc
->buffer_addr
= cpu_to_le64(buffer_info
->dma
);
208 if (i
== rx_ring
->count
)
210 buffer_info
= &rx_ring
->buffer_info
[i
];
213 if (rx_ring
->next_to_use
!= i
) {
214 rx_ring
->next_to_use
= i
;
216 i
= (rx_ring
->count
- 1);
219 * Force memory writes to complete before letting h/w
220 * know there are new descriptors to fetch. (Only
221 * applicable for weak-ordered memory model archs,
225 writel(i
, adapter
->hw
.hw_addr
+ rx_ring
->tail
);
230 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
231 * @adapter: address of board private structure
233 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter
*adapter
,
236 struct net_device
*netdev
= adapter
->netdev
;
237 struct pci_dev
*pdev
= adapter
->pdev
;
238 union e1000_rx_desc_packet_split
*rx_desc
;
239 struct e1000_ring
*rx_ring
= adapter
->rx_ring
;
240 struct e1000_buffer
*buffer_info
;
241 struct e1000_ps_page
*ps_page
;
245 i
= rx_ring
->next_to_use
;
246 buffer_info
= &rx_ring
->buffer_info
[i
];
248 while (cleaned_count
--) {
249 rx_desc
= E1000_RX_DESC_PS(*rx_ring
, i
);
251 for (j
= 0; j
< PS_PAGE_BUFFERS
; j
++) {
252 ps_page
= &buffer_info
->ps_pages
[j
];
253 if (j
>= adapter
->rx_ps_pages
) {
254 /* all unused desc entries get hw null ptr */
255 rx_desc
->read
.buffer_addr
[j
+1] = ~cpu_to_le64(0);
258 if (!ps_page
->page
) {
259 ps_page
->page
= alloc_page(GFP_ATOMIC
);
260 if (!ps_page
->page
) {
261 adapter
->alloc_rx_buff_failed
++;
264 ps_page
->dma
= pci_map_page(pdev
,
268 if (pci_dma_mapping_error(pdev
, ps_page
->dma
)) {
269 dev_err(&adapter
->pdev
->dev
,
270 "RX DMA page map failed\n");
271 adapter
->rx_dma_failed
++;
276 * Refresh the desc even if buffer_addrs
277 * didn't change because each write-back
280 rx_desc
->read
.buffer_addr
[j
+1] =
281 cpu_to_le64(ps_page
->dma
);
284 skb
= netdev_alloc_skb(netdev
,
285 adapter
->rx_ps_bsize0
+ NET_IP_ALIGN
);
288 adapter
->alloc_rx_buff_failed
++;
293 * Make buffer alignment 2 beyond a 16 byte boundary
294 * this will result in a 16 byte aligned IP header after
295 * the 14 byte MAC header is removed
297 skb_reserve(skb
, NET_IP_ALIGN
);
299 buffer_info
->skb
= skb
;
300 buffer_info
->dma
= pci_map_single(pdev
, skb
->data
,
301 adapter
->rx_ps_bsize0
,
303 if (pci_dma_mapping_error(pdev
, buffer_info
->dma
)) {
304 dev_err(&pdev
->dev
, "RX DMA map failed\n");
305 adapter
->rx_dma_failed
++;
307 dev_kfree_skb_any(skb
);
308 buffer_info
->skb
= NULL
;
312 rx_desc
->read
.buffer_addr
[0] = cpu_to_le64(buffer_info
->dma
);
315 if (i
== rx_ring
->count
)
317 buffer_info
= &rx_ring
->buffer_info
[i
];
321 if (rx_ring
->next_to_use
!= i
) {
322 rx_ring
->next_to_use
= i
;
325 i
= (rx_ring
->count
- 1);
328 * Force memory writes to complete before letting h/w
329 * know there are new descriptors to fetch. (Only
330 * applicable for weak-ordered memory model archs,
335 * Hardware increments by 16 bytes, but packet split
336 * descriptors are 32 bytes...so we increment tail
339 writel(i
<<1, adapter
->hw
.hw_addr
+ rx_ring
->tail
);
344 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
345 * @adapter: address of board private structure
346 * @cleaned_count: number of buffers to allocate this pass
349 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter
*adapter
,
352 struct net_device
*netdev
= adapter
->netdev
;
353 struct pci_dev
*pdev
= adapter
->pdev
;
354 struct e1000_rx_desc
*rx_desc
;
355 struct e1000_ring
*rx_ring
= adapter
->rx_ring
;
356 struct e1000_buffer
*buffer_info
;
359 unsigned int bufsz
= 256 -
360 16 /* for skb_reserve */ -
363 i
= rx_ring
->next_to_use
;
364 buffer_info
= &rx_ring
->buffer_info
[i
];
366 while (cleaned_count
--) {
367 skb
= buffer_info
->skb
;
373 skb
= netdev_alloc_skb(netdev
, bufsz
);
374 if (unlikely(!skb
)) {
375 /* Better luck next round */
376 adapter
->alloc_rx_buff_failed
++;
380 /* Make buffer alignment 2 beyond a 16 byte boundary
381 * this will result in a 16 byte aligned IP header after
382 * the 14 byte MAC header is removed
384 skb_reserve(skb
, NET_IP_ALIGN
);
386 buffer_info
->skb
= skb
;
388 /* allocate a new page if necessary */
389 if (!buffer_info
->page
) {
390 buffer_info
->page
= alloc_page(GFP_ATOMIC
);
391 if (unlikely(!buffer_info
->page
)) {
392 adapter
->alloc_rx_buff_failed
++;
397 if (!buffer_info
->dma
)
398 buffer_info
->dma
= pci_map_page(pdev
,
399 buffer_info
->page
, 0,
403 rx_desc
= E1000_RX_DESC(*rx_ring
, i
);
404 rx_desc
->buffer_addr
= cpu_to_le64(buffer_info
->dma
);
406 if (unlikely(++i
== rx_ring
->count
))
408 buffer_info
= &rx_ring
->buffer_info
[i
];
411 if (likely(rx_ring
->next_to_use
!= i
)) {
412 rx_ring
->next_to_use
= i
;
413 if (unlikely(i
-- == 0))
414 i
= (rx_ring
->count
- 1);
416 /* Force memory writes to complete before letting h/w
417 * know there are new descriptors to fetch. (Only
418 * applicable for weak-ordered memory model archs,
421 writel(i
, adapter
->hw
.hw_addr
+ rx_ring
->tail
);
426 * e1000_clean_rx_irq - Send received data up the network stack; legacy
427 * @adapter: board private structure
429 * the return value indicates whether actual cleaning was done, there
430 * is no guarantee that everything was cleaned
432 static bool e1000_clean_rx_irq(struct e1000_adapter
*adapter
,
433 int *work_done
, int work_to_do
)
435 struct net_device
*netdev
= adapter
->netdev
;
436 struct pci_dev
*pdev
= adapter
->pdev
;
437 struct e1000_ring
*rx_ring
= adapter
->rx_ring
;
438 struct e1000_rx_desc
*rx_desc
, *next_rxd
;
439 struct e1000_buffer
*buffer_info
, *next_buffer
;
442 int cleaned_count
= 0;
444 unsigned int total_rx_bytes
= 0, total_rx_packets
= 0;
446 i
= rx_ring
->next_to_clean
;
447 rx_desc
= E1000_RX_DESC(*rx_ring
, i
);
448 buffer_info
= &rx_ring
->buffer_info
[i
];
450 while (rx_desc
->status
& E1000_RXD_STAT_DD
) {
454 if (*work_done
>= work_to_do
)
458 status
= rx_desc
->status
;
459 skb
= buffer_info
->skb
;
460 buffer_info
->skb
= NULL
;
462 prefetch(skb
->data
- NET_IP_ALIGN
);
465 if (i
== rx_ring
->count
)
467 next_rxd
= E1000_RX_DESC(*rx_ring
, i
);
470 next_buffer
= &rx_ring
->buffer_info
[i
];
474 pci_unmap_single(pdev
,
476 adapter
->rx_buffer_len
,
478 buffer_info
->dma
= 0;
480 length
= le16_to_cpu(rx_desc
->length
);
482 /* !EOP means multiple descriptors were used to store a single
483 * packet, also make sure the frame isn't just CRC only */
484 if (!(status
& E1000_RXD_STAT_EOP
) || (length
<= 4)) {
485 /* All receives must fit into a single buffer */
486 e_dbg("%s: Receive packet consumed multiple buffers\n",
489 buffer_info
->skb
= skb
;
493 if (rx_desc
->errors
& E1000_RXD_ERR_FRAME_ERR_MASK
) {
495 buffer_info
->skb
= skb
;
499 /* adjust length to remove Ethernet CRC */
500 if (!(adapter
->flags2
& FLAG2_CRC_STRIPPING
))
503 total_rx_bytes
+= length
;
507 * code added for copybreak, this should improve
508 * performance for small packets with large amounts
509 * of reassembly being done in the stack
511 if (length
< copybreak
) {
512 struct sk_buff
*new_skb
=
513 netdev_alloc_skb(netdev
, length
+ NET_IP_ALIGN
);
515 skb_reserve(new_skb
, NET_IP_ALIGN
);
516 skb_copy_to_linear_data_offset(new_skb
,
522 /* save the skb in buffer_info as good */
523 buffer_info
->skb
= skb
;
526 /* else just continue with the old one */
528 /* end copybreak code */
529 skb_put(skb
, length
);
531 /* Receive Checksum Offload */
532 e1000_rx_checksum(adapter
,
534 ((u32
)(rx_desc
->errors
) << 24),
535 le16_to_cpu(rx_desc
->csum
), skb
);
537 e1000_receive_skb(adapter
, netdev
, skb
,status
,rx_desc
->special
);
542 /* return some buffers to hardware, one at a time is too slow */
543 if (cleaned_count
>= E1000_RX_BUFFER_WRITE
) {
544 adapter
->alloc_rx_buf(adapter
, cleaned_count
);
548 /* use prefetched values */
550 buffer_info
= next_buffer
;
552 rx_ring
->next_to_clean
= i
;
554 cleaned_count
= e1000_desc_unused(rx_ring
);
556 adapter
->alloc_rx_buf(adapter
, cleaned_count
);
558 adapter
->total_rx_bytes
+= total_rx_bytes
;
559 adapter
->total_rx_packets
+= total_rx_packets
;
560 adapter
->net_stats
.rx_bytes
+= total_rx_bytes
;
561 adapter
->net_stats
.rx_packets
+= total_rx_packets
;
565 static void e1000_put_txbuf(struct e1000_adapter
*adapter
,
566 struct e1000_buffer
*buffer_info
)
568 buffer_info
->dma
= 0;
569 if (buffer_info
->skb
) {
570 skb_dma_unmap(&adapter
->pdev
->dev
, 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;
685 * read barrier to make sure that the ->dma member and time
686 * stamp are updated fully
689 if (tx_ring
->buffer_info
[eop
].dma
&&
690 time_after(jiffies
, tx_ring
->buffer_info
[eop
].time_stamp
691 + (adapter
->tx_timeout_factor
* HZ
))
692 && !(er32(STATUS
) & E1000_STATUS_TXOFF
)) {
693 e1000_print_tx_hang(adapter
);
694 netif_stop_queue(netdev
);
697 adapter
->total_tx_bytes
+= total_tx_bytes
;
698 adapter
->total_tx_packets
+= total_tx_packets
;
699 adapter
->net_stats
.tx_bytes
+= total_tx_bytes
;
700 adapter
->net_stats
.tx_packets
+= total_tx_packets
;
705 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
706 * @adapter: board private structure
708 * the return value indicates whether actual cleaning was done, there
709 * is no guarantee that everything was cleaned
711 static bool e1000_clean_rx_irq_ps(struct e1000_adapter
*adapter
,
712 int *work_done
, int work_to_do
)
714 union e1000_rx_desc_packet_split
*rx_desc
, *next_rxd
;
715 struct net_device
*netdev
= adapter
->netdev
;
716 struct pci_dev
*pdev
= adapter
->pdev
;
717 struct e1000_ring
*rx_ring
= adapter
->rx_ring
;
718 struct e1000_buffer
*buffer_info
, *next_buffer
;
719 struct e1000_ps_page
*ps_page
;
723 int cleaned_count
= 0;
725 unsigned int total_rx_bytes
= 0, total_rx_packets
= 0;
727 i
= rx_ring
->next_to_clean
;
728 rx_desc
= E1000_RX_DESC_PS(*rx_ring
, i
);
729 staterr
= le32_to_cpu(rx_desc
->wb
.middle
.status_error
);
730 buffer_info
= &rx_ring
->buffer_info
[i
];
732 while (staterr
& E1000_RXD_STAT_DD
) {
733 if (*work_done
>= work_to_do
)
736 skb
= buffer_info
->skb
;
738 /* in the packet split case this is header only */
739 prefetch(skb
->data
- NET_IP_ALIGN
);
742 if (i
== rx_ring
->count
)
744 next_rxd
= E1000_RX_DESC_PS(*rx_ring
, i
);
747 next_buffer
= &rx_ring
->buffer_info
[i
];
751 pci_unmap_single(pdev
, buffer_info
->dma
,
752 adapter
->rx_ps_bsize0
,
754 buffer_info
->dma
= 0;
756 if (!(staterr
& E1000_RXD_STAT_EOP
)) {
757 e_dbg("%s: Packet Split buffers didn't pick up the "
758 "full packet\n", netdev
->name
);
759 dev_kfree_skb_irq(skb
);
763 if (staterr
& E1000_RXDEXT_ERR_FRAME_ERR_MASK
) {
764 dev_kfree_skb_irq(skb
);
768 length
= le16_to_cpu(rx_desc
->wb
.middle
.length0
);
771 e_dbg("%s: Last part of the packet spanning multiple "
772 "descriptors\n", netdev
->name
);
773 dev_kfree_skb_irq(skb
);
778 skb_put(skb
, length
);
782 * this looks ugly, but it seems compiler issues make it
783 * more efficient than reusing j
785 int l1
= le16_to_cpu(rx_desc
->wb
.upper
.length
[0]);
788 * page alloc/put takes too long and effects small packet
789 * throughput, so unsplit small packets and save the alloc/put
790 * only valid in softirq (napi) context to call kmap_*
792 if (l1
&& (l1
<= copybreak
) &&
793 ((length
+ l1
) <= adapter
->rx_ps_bsize0
)) {
796 ps_page
= &buffer_info
->ps_pages
[0];
799 * there is no documentation about how to call
800 * kmap_atomic, so we can't hold the mapping
803 pci_dma_sync_single_for_cpu(pdev
, ps_page
->dma
,
804 PAGE_SIZE
, PCI_DMA_FROMDEVICE
);
805 vaddr
= kmap_atomic(ps_page
->page
, KM_SKB_DATA_SOFTIRQ
);
806 memcpy(skb_tail_pointer(skb
), vaddr
, l1
);
807 kunmap_atomic(vaddr
, KM_SKB_DATA_SOFTIRQ
);
808 pci_dma_sync_single_for_device(pdev
, ps_page
->dma
,
809 PAGE_SIZE
, PCI_DMA_FROMDEVICE
);
812 if (!(adapter
->flags2
& FLAG2_CRC_STRIPPING
))
820 for (j
= 0; j
< PS_PAGE_BUFFERS
; j
++) {
821 length
= le16_to_cpu(rx_desc
->wb
.upper
.length
[j
]);
825 ps_page
= &buffer_info
->ps_pages
[j
];
826 pci_unmap_page(pdev
, ps_page
->dma
, PAGE_SIZE
,
829 skb_fill_page_desc(skb
, j
, ps_page
->page
, 0, length
);
830 ps_page
->page
= NULL
;
832 skb
->data_len
+= length
;
833 skb
->truesize
+= length
;
836 /* strip the ethernet crc, problem is we're using pages now so
837 * this whole operation can get a little cpu intensive
839 if (!(adapter
->flags2
& FLAG2_CRC_STRIPPING
))
840 pskb_trim(skb
, skb
->len
- 4);
843 total_rx_bytes
+= skb
->len
;
846 e1000_rx_checksum(adapter
, staterr
, le16_to_cpu(
847 rx_desc
->wb
.lower
.hi_dword
.csum_ip
.csum
), skb
);
849 if (rx_desc
->wb
.upper
.header_status
&
850 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP
))
851 adapter
->rx_hdr_split
++;
853 e1000_receive_skb(adapter
, netdev
, skb
,
854 staterr
, rx_desc
->wb
.middle
.vlan
);
857 rx_desc
->wb
.middle
.status_error
&= cpu_to_le32(~0xFF);
858 buffer_info
->skb
= NULL
;
860 /* return some buffers to hardware, one at a time is too slow */
861 if (cleaned_count
>= E1000_RX_BUFFER_WRITE
) {
862 adapter
->alloc_rx_buf(adapter
, cleaned_count
);
866 /* use prefetched values */
868 buffer_info
= next_buffer
;
870 staterr
= le32_to_cpu(rx_desc
->wb
.middle
.status_error
);
872 rx_ring
->next_to_clean
= i
;
874 cleaned_count
= e1000_desc_unused(rx_ring
);
876 adapter
->alloc_rx_buf(adapter
, cleaned_count
);
878 adapter
->total_rx_bytes
+= total_rx_bytes
;
879 adapter
->total_rx_packets
+= total_rx_packets
;
880 adapter
->net_stats
.rx_bytes
+= total_rx_bytes
;
881 adapter
->net_stats
.rx_packets
+= total_rx_packets
;
886 * e1000_consume_page - helper function
888 static void e1000_consume_page(struct e1000_buffer
*bi
, struct sk_buff
*skb
,
893 skb
->data_len
+= length
;
894 skb
->truesize
+= length
;
898 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
899 * @adapter: board private structure
901 * the return value indicates whether actual cleaning was done, there
902 * is no guarantee that everything was cleaned
905 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter
*adapter
,
906 int *work_done
, int work_to_do
)
908 struct net_device
*netdev
= adapter
->netdev
;
909 struct pci_dev
*pdev
= adapter
->pdev
;
910 struct e1000_ring
*rx_ring
= adapter
->rx_ring
;
911 struct e1000_rx_desc
*rx_desc
, *next_rxd
;
912 struct e1000_buffer
*buffer_info
, *next_buffer
;
915 int cleaned_count
= 0;
916 bool cleaned
= false;
917 unsigned int total_rx_bytes
=0, total_rx_packets
=0;
919 i
= rx_ring
->next_to_clean
;
920 rx_desc
= E1000_RX_DESC(*rx_ring
, i
);
921 buffer_info
= &rx_ring
->buffer_info
[i
];
923 while (rx_desc
->status
& E1000_RXD_STAT_DD
) {
927 if (*work_done
>= work_to_do
)
931 status
= rx_desc
->status
;
932 skb
= buffer_info
->skb
;
933 buffer_info
->skb
= NULL
;
936 if (i
== rx_ring
->count
)
938 next_rxd
= E1000_RX_DESC(*rx_ring
, i
);
941 next_buffer
= &rx_ring
->buffer_info
[i
];
945 pci_unmap_page(pdev
, buffer_info
->dma
, PAGE_SIZE
,
947 buffer_info
->dma
= 0;
949 length
= le16_to_cpu(rx_desc
->length
);
951 /* errors is only valid for DD + EOP descriptors */
952 if (unlikely((status
& E1000_RXD_STAT_EOP
) &&
953 (rx_desc
->errors
& E1000_RXD_ERR_FRAME_ERR_MASK
))) {
954 /* recycle both page and skb */
955 buffer_info
->skb
= skb
;
956 /* an error means any chain goes out the window
958 if (rx_ring
->rx_skb_top
)
959 dev_kfree_skb(rx_ring
->rx_skb_top
);
960 rx_ring
->rx_skb_top
= NULL
;
964 #define rxtop rx_ring->rx_skb_top
965 if (!(status
& E1000_RXD_STAT_EOP
)) {
966 /* this descriptor is only the beginning (or middle) */
968 /* this is the beginning of a chain */
970 skb_fill_page_desc(rxtop
, 0, buffer_info
->page
,
973 /* this is the middle of a chain */
974 skb_fill_page_desc(rxtop
,
975 skb_shinfo(rxtop
)->nr_frags
,
976 buffer_info
->page
, 0, length
);
977 /* re-use the skb, only consumed the page */
978 buffer_info
->skb
= skb
;
980 e1000_consume_page(buffer_info
, rxtop
, length
);
984 /* end of the chain */
985 skb_fill_page_desc(rxtop
,
986 skb_shinfo(rxtop
)->nr_frags
,
987 buffer_info
->page
, 0, length
);
988 /* re-use the current skb, we only consumed the
990 buffer_info
->skb
= skb
;
993 e1000_consume_page(buffer_info
, skb
, length
);
995 /* no chain, got EOP, this buf is the packet
996 * copybreak to save the put_page/alloc_page */
997 if (length
<= copybreak
&&
998 skb_tailroom(skb
) >= length
) {
1000 vaddr
= kmap_atomic(buffer_info
->page
,
1001 KM_SKB_DATA_SOFTIRQ
);
1002 memcpy(skb_tail_pointer(skb
), vaddr
,
1004 kunmap_atomic(vaddr
,
1005 KM_SKB_DATA_SOFTIRQ
);
1006 /* re-use the page, so don't erase
1007 * buffer_info->page */
1008 skb_put(skb
, length
);
1010 skb_fill_page_desc(skb
, 0,
1011 buffer_info
->page
, 0,
1013 e1000_consume_page(buffer_info
, skb
,
1019 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1020 e1000_rx_checksum(adapter
,
1022 ((u32
)(rx_desc
->errors
) << 24),
1023 le16_to_cpu(rx_desc
->csum
), skb
);
1025 /* probably a little skewed due to removing CRC */
1026 total_rx_bytes
+= skb
->len
;
1029 /* eth type trans needs skb->data to point to something */
1030 if (!pskb_may_pull(skb
, ETH_HLEN
)) {
1031 e_err("pskb_may_pull failed.\n");
1036 e1000_receive_skb(adapter
, netdev
, skb
, status
,
1040 rx_desc
->status
= 0;
1042 /* return some buffers to hardware, one at a time is too slow */
1043 if (unlikely(cleaned_count
>= E1000_RX_BUFFER_WRITE
)) {
1044 adapter
->alloc_rx_buf(adapter
, cleaned_count
);
1048 /* use prefetched values */
1050 buffer_info
= next_buffer
;
1052 rx_ring
->next_to_clean
= i
;
1054 cleaned_count
= e1000_desc_unused(rx_ring
);
1056 adapter
->alloc_rx_buf(adapter
, cleaned_count
);
1058 adapter
->total_rx_bytes
+= total_rx_bytes
;
1059 adapter
->total_rx_packets
+= total_rx_packets
;
1060 adapter
->net_stats
.rx_bytes
+= total_rx_bytes
;
1061 adapter
->net_stats
.rx_packets
+= total_rx_packets
;
1066 * e1000_clean_rx_ring - Free Rx Buffers per Queue
1067 * @adapter: board private structure
1069 static void e1000_clean_rx_ring(struct e1000_adapter
*adapter
)
1071 struct e1000_ring
*rx_ring
= adapter
->rx_ring
;
1072 struct e1000_buffer
*buffer_info
;
1073 struct e1000_ps_page
*ps_page
;
1074 struct pci_dev
*pdev
= adapter
->pdev
;
1077 /* Free all the Rx ring sk_buffs */
1078 for (i
= 0; i
< rx_ring
->count
; i
++) {
1079 buffer_info
= &rx_ring
->buffer_info
[i
];
1080 if (buffer_info
->dma
) {
1081 if (adapter
->clean_rx
== e1000_clean_rx_irq
)
1082 pci_unmap_single(pdev
, buffer_info
->dma
,
1083 adapter
->rx_buffer_len
,
1084 PCI_DMA_FROMDEVICE
);
1085 else if (adapter
->clean_rx
== e1000_clean_jumbo_rx_irq
)
1086 pci_unmap_page(pdev
, buffer_info
->dma
,
1088 PCI_DMA_FROMDEVICE
);
1089 else if (adapter
->clean_rx
== e1000_clean_rx_irq_ps
)
1090 pci_unmap_single(pdev
, buffer_info
->dma
,
1091 adapter
->rx_ps_bsize0
,
1092 PCI_DMA_FROMDEVICE
);
1093 buffer_info
->dma
= 0;
1096 if (buffer_info
->page
) {
1097 put_page(buffer_info
->page
);
1098 buffer_info
->page
= NULL
;
1101 if (buffer_info
->skb
) {
1102 dev_kfree_skb(buffer_info
->skb
);
1103 buffer_info
->skb
= NULL
;
1106 for (j
= 0; j
< PS_PAGE_BUFFERS
; j
++) {
1107 ps_page
= &buffer_info
->ps_pages
[j
];
1110 pci_unmap_page(pdev
, ps_page
->dma
, PAGE_SIZE
,
1111 PCI_DMA_FROMDEVICE
);
1113 put_page(ps_page
->page
);
1114 ps_page
->page
= NULL
;
1118 /* there also may be some cached data from a chained receive */
1119 if (rx_ring
->rx_skb_top
) {
1120 dev_kfree_skb(rx_ring
->rx_skb_top
);
1121 rx_ring
->rx_skb_top
= NULL
;
1124 /* Zero out the descriptor ring */
1125 memset(rx_ring
->desc
, 0, rx_ring
->size
);
1127 rx_ring
->next_to_clean
= 0;
1128 rx_ring
->next_to_use
= 0;
1130 writel(0, adapter
->hw
.hw_addr
+ rx_ring
->head
);
1131 writel(0, adapter
->hw
.hw_addr
+ rx_ring
->tail
);
1134 static void e1000e_downshift_workaround(struct work_struct
*work
)
1136 struct e1000_adapter
*adapter
= container_of(work
,
1137 struct e1000_adapter
, downshift_task
);
1139 e1000e_gig_downshift_workaround_ich8lan(&adapter
->hw
);
1143 * e1000_intr_msi - Interrupt Handler
1144 * @irq: interrupt number
1145 * @data: pointer to a network interface device structure
1147 static irqreturn_t
e1000_intr_msi(int irq
, void *data
)
1149 struct net_device
*netdev
= data
;
1150 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
1151 struct e1000_hw
*hw
= &adapter
->hw
;
1152 u32 icr
= er32(ICR
);
1155 * read ICR disables interrupts using IAM
1158 if (icr
& (E1000_ICR_RXSEQ
| E1000_ICR_LSC
)) {
1159 hw
->mac
.get_link_status
= 1;
1161 * ICH8 workaround-- Call gig speed drop workaround on cable
1162 * disconnect (LSC) before accessing any PHY registers
1164 if ((adapter
->flags
& FLAG_LSC_GIG_SPEED_DROP
) &&
1165 (!(er32(STATUS
) & E1000_STATUS_LU
)))
1166 schedule_work(&adapter
->downshift_task
);
1169 * 80003ES2LAN workaround-- For packet buffer work-around on
1170 * link down event; disable receives here in the ISR and reset
1171 * adapter in watchdog
1173 if (netif_carrier_ok(netdev
) &&
1174 adapter
->flags
& FLAG_RX_NEEDS_RESTART
) {
1175 /* disable receives */
1176 u32 rctl
= er32(RCTL
);
1177 ew32(RCTL
, rctl
& ~E1000_RCTL_EN
);
1178 adapter
->flags
|= FLAG_RX_RESTART_NOW
;
1180 /* guard against interrupt when we're going down */
1181 if (!test_bit(__E1000_DOWN
, &adapter
->state
))
1182 mod_timer(&adapter
->watchdog_timer
, jiffies
+ 1);
1185 if (netif_rx_schedule_prep(&adapter
->napi
)) {
1186 adapter
->total_tx_bytes
= 0;
1187 adapter
->total_tx_packets
= 0;
1188 adapter
->total_rx_bytes
= 0;
1189 adapter
->total_rx_packets
= 0;
1190 __netif_rx_schedule(&adapter
->napi
);
1197 * e1000_intr - Interrupt Handler
1198 * @irq: interrupt number
1199 * @data: pointer to a network interface device structure
1201 static irqreturn_t
e1000_intr(int irq
, void *data
)
1203 struct net_device
*netdev
= data
;
1204 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
1205 struct e1000_hw
*hw
= &adapter
->hw
;
1206 u32 rctl
, icr
= er32(ICR
);
1209 return IRQ_NONE
; /* Not our interrupt */
1212 * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1213 * not set, then the adapter didn't send an interrupt
1215 if (!(icr
& E1000_ICR_INT_ASSERTED
))
1219 * Interrupt Auto-Mask...upon reading ICR,
1220 * interrupts are masked. No need for the
1224 if (icr
& (E1000_ICR_RXSEQ
| E1000_ICR_LSC
)) {
1225 hw
->mac
.get_link_status
= 1;
1227 * ICH8 workaround-- Call gig speed drop workaround on cable
1228 * disconnect (LSC) before accessing any PHY registers
1230 if ((adapter
->flags
& FLAG_LSC_GIG_SPEED_DROP
) &&
1231 (!(er32(STATUS
) & E1000_STATUS_LU
)))
1232 schedule_work(&adapter
->downshift_task
);
1235 * 80003ES2LAN workaround--
1236 * For packet buffer work-around on link down event;
1237 * disable receives here in the ISR and
1238 * reset adapter in watchdog
1240 if (netif_carrier_ok(netdev
) &&
1241 (adapter
->flags
& FLAG_RX_NEEDS_RESTART
)) {
1242 /* disable receives */
1244 ew32(RCTL
, rctl
& ~E1000_RCTL_EN
);
1245 adapter
->flags
|= FLAG_RX_RESTART_NOW
;
1247 /* guard against interrupt when we're going down */
1248 if (!test_bit(__E1000_DOWN
, &adapter
->state
))
1249 mod_timer(&adapter
->watchdog_timer
, jiffies
+ 1);
1252 if (netif_rx_schedule_prep(&adapter
->napi
)) {
1253 adapter
->total_tx_bytes
= 0;
1254 adapter
->total_tx_packets
= 0;
1255 adapter
->total_rx_bytes
= 0;
1256 adapter
->total_rx_packets
= 0;
1257 __netif_rx_schedule(&adapter
->napi
);
1263 static irqreturn_t
e1000_msix_other(int irq
, void *data
)
1265 struct net_device
*netdev
= data
;
1266 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
1267 struct e1000_hw
*hw
= &adapter
->hw
;
1268 u32 icr
= er32(ICR
);
1270 if (!(icr
& E1000_ICR_INT_ASSERTED
)) {
1271 ew32(IMS
, E1000_IMS_OTHER
);
1275 if (icr
& adapter
->eiac_mask
)
1276 ew32(ICS
, (icr
& adapter
->eiac_mask
));
1278 if (icr
& E1000_ICR_OTHER
) {
1279 if (!(icr
& E1000_ICR_LSC
))
1280 goto no_link_interrupt
;
1281 hw
->mac
.get_link_status
= 1;
1282 /* guard against interrupt when we're going down */
1283 if (!test_bit(__E1000_DOWN
, &adapter
->state
))
1284 mod_timer(&adapter
->watchdog_timer
, jiffies
+ 1);
1288 ew32(IMS
, E1000_IMS_LSC
| E1000_IMS_OTHER
);
1294 static irqreturn_t
e1000_intr_msix_tx(int irq
, void *data
)
1296 struct net_device
*netdev
= data
;
1297 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
1298 struct e1000_hw
*hw
= &adapter
->hw
;
1299 struct e1000_ring
*tx_ring
= adapter
->tx_ring
;
1302 adapter
->total_tx_bytes
= 0;
1303 adapter
->total_tx_packets
= 0;
1305 if (!e1000_clean_tx_irq(adapter
))
1306 /* Ring was not completely cleaned, so fire another interrupt */
1307 ew32(ICS
, tx_ring
->ims_val
);
1312 static irqreturn_t
e1000_intr_msix_rx(int irq
, void *data
)
1314 struct net_device
*netdev
= data
;
1315 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
1317 /* Write the ITR value calculated at the end of the
1318 * previous interrupt.
1320 if (adapter
->rx_ring
->set_itr
) {
1321 writel(1000000000 / (adapter
->rx_ring
->itr_val
* 256),
1322 adapter
->hw
.hw_addr
+ adapter
->rx_ring
->itr_register
);
1323 adapter
->rx_ring
->set_itr
= 0;
1326 if (netif_rx_schedule_prep(&adapter
->napi
)) {
1327 adapter
->total_rx_bytes
= 0;
1328 adapter
->total_rx_packets
= 0;
1329 __netif_rx_schedule(&adapter
->napi
);
1335 * e1000_configure_msix - Configure MSI-X hardware
1337 * e1000_configure_msix sets up the hardware to properly
1338 * generate MSI-X interrupts.
1340 static void e1000_configure_msix(struct e1000_adapter
*adapter
)
1342 struct e1000_hw
*hw
= &adapter
->hw
;
1343 struct e1000_ring
*rx_ring
= adapter
->rx_ring
;
1344 struct e1000_ring
*tx_ring
= adapter
->tx_ring
;
1346 u32 ctrl_ext
, ivar
= 0;
1348 adapter
->eiac_mask
= 0;
1350 /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1351 if (hw
->mac
.type
== e1000_82574
) {
1352 u32 rfctl
= er32(RFCTL
);
1353 rfctl
|= E1000_RFCTL_ACK_DIS
;
1357 #define E1000_IVAR_INT_ALLOC_VALID 0x8
1358 /* Configure Rx vector */
1359 rx_ring
->ims_val
= E1000_IMS_RXQ0
;
1360 adapter
->eiac_mask
|= rx_ring
->ims_val
;
1361 if (rx_ring
->itr_val
)
1362 writel(1000000000 / (rx_ring
->itr_val
* 256),
1363 hw
->hw_addr
+ rx_ring
->itr_register
);
1365 writel(1, hw
->hw_addr
+ rx_ring
->itr_register
);
1366 ivar
= E1000_IVAR_INT_ALLOC_VALID
| vector
;
1368 /* Configure Tx vector */
1369 tx_ring
->ims_val
= E1000_IMS_TXQ0
;
1371 if (tx_ring
->itr_val
)
1372 writel(1000000000 / (tx_ring
->itr_val
* 256),
1373 hw
->hw_addr
+ tx_ring
->itr_register
);
1375 writel(1, hw
->hw_addr
+ tx_ring
->itr_register
);
1376 adapter
->eiac_mask
|= tx_ring
->ims_val
;
1377 ivar
|= ((E1000_IVAR_INT_ALLOC_VALID
| vector
) << 8);
1379 /* set vector for Other Causes, e.g. link changes */
1381 ivar
|= ((E1000_IVAR_INT_ALLOC_VALID
| vector
) << 16);
1382 if (rx_ring
->itr_val
)
1383 writel(1000000000 / (rx_ring
->itr_val
* 256),
1384 hw
->hw_addr
+ E1000_EITR_82574(vector
));
1386 writel(1, hw
->hw_addr
+ E1000_EITR_82574(vector
));
1388 /* Cause Tx interrupts on every write back */
1393 /* enable MSI-X PBA support */
1394 ctrl_ext
= er32(CTRL_EXT
);
1395 ctrl_ext
|= E1000_CTRL_EXT_PBA_CLR
;
1397 /* Auto-Mask Other interrupts upon ICR read */
1398 #define E1000_EIAC_MASK_82574 0x01F00000
1399 ew32(IAM
, ~E1000_EIAC_MASK_82574
| E1000_IMS_OTHER
);
1400 ctrl_ext
|= E1000_CTRL_EXT_EIAME
;
1401 ew32(CTRL_EXT
, ctrl_ext
);
1405 void e1000e_reset_interrupt_capability(struct e1000_adapter
*adapter
)
1407 if (adapter
->msix_entries
) {
1408 pci_disable_msix(adapter
->pdev
);
1409 kfree(adapter
->msix_entries
);
1410 adapter
->msix_entries
= NULL
;
1411 } else if (adapter
->flags
& FLAG_MSI_ENABLED
) {
1412 pci_disable_msi(adapter
->pdev
);
1413 adapter
->flags
&= ~FLAG_MSI_ENABLED
;
1420 * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1422 * Attempt to configure interrupts using the best available
1423 * capabilities of the hardware and kernel.
1425 void e1000e_set_interrupt_capability(struct e1000_adapter
*adapter
)
1431 switch (adapter
->int_mode
) {
1432 case E1000E_INT_MODE_MSIX
:
1433 if (adapter
->flags
& FLAG_HAS_MSIX
) {
1434 numvecs
= 3; /* RxQ0, TxQ0 and other */
1435 adapter
->msix_entries
= kcalloc(numvecs
,
1436 sizeof(struct msix_entry
),
1438 if (adapter
->msix_entries
) {
1439 for (i
= 0; i
< numvecs
; i
++)
1440 adapter
->msix_entries
[i
].entry
= i
;
1442 err
= pci_enable_msix(adapter
->pdev
,
1443 adapter
->msix_entries
,
1448 /* MSI-X failed, so fall through and try MSI */
1449 e_err("Failed to initialize MSI-X interrupts. "
1450 "Falling back to MSI interrupts.\n");
1451 e1000e_reset_interrupt_capability(adapter
);
1453 adapter
->int_mode
= E1000E_INT_MODE_MSI
;
1455 case E1000E_INT_MODE_MSI
:
1456 if (!pci_enable_msi(adapter
->pdev
)) {
1457 adapter
->flags
|= FLAG_MSI_ENABLED
;
1459 adapter
->int_mode
= E1000E_INT_MODE_LEGACY
;
1460 e_err("Failed to initialize MSI interrupts. Falling "
1461 "back to legacy interrupts.\n");
1464 case E1000E_INT_MODE_LEGACY
:
1465 /* Don't do anything; this is the system default */
1473 * e1000_request_msix - Initialize MSI-X interrupts
1475 * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1478 static int e1000_request_msix(struct e1000_adapter
*adapter
)
1480 struct net_device
*netdev
= adapter
->netdev
;
1481 int err
= 0, vector
= 0;
1483 if (strlen(netdev
->name
) < (IFNAMSIZ
- 5))
1484 sprintf(adapter
->rx_ring
->name
, "%s-rx-0", netdev
->name
);
1486 memcpy(adapter
->rx_ring
->name
, netdev
->name
, IFNAMSIZ
);
1487 err
= request_irq(adapter
->msix_entries
[vector
].vector
,
1488 &e1000_intr_msix_rx
, 0, adapter
->rx_ring
->name
,
1492 adapter
->rx_ring
->itr_register
= E1000_EITR_82574(vector
);
1493 adapter
->rx_ring
->itr_val
= adapter
->itr
;
1496 if (strlen(netdev
->name
) < (IFNAMSIZ
- 5))
1497 sprintf(adapter
->tx_ring
->name
, "%s-tx-0", netdev
->name
);
1499 memcpy(adapter
->tx_ring
->name
, netdev
->name
, IFNAMSIZ
);
1500 err
= request_irq(adapter
->msix_entries
[vector
].vector
,
1501 &e1000_intr_msix_tx
, 0, adapter
->tx_ring
->name
,
1505 adapter
->tx_ring
->itr_register
= E1000_EITR_82574(vector
);
1506 adapter
->tx_ring
->itr_val
= adapter
->itr
;
1509 err
= request_irq(adapter
->msix_entries
[vector
].vector
,
1510 &e1000_msix_other
, 0, netdev
->name
, netdev
);
1514 e1000_configure_msix(adapter
);
1521 * e1000_request_irq - initialize interrupts
1523 * Attempts to configure interrupts using the best available
1524 * capabilities of the hardware and kernel.
1526 static int e1000_request_irq(struct e1000_adapter
*adapter
)
1528 struct net_device
*netdev
= adapter
->netdev
;
1531 if (adapter
->msix_entries
) {
1532 err
= e1000_request_msix(adapter
);
1535 /* fall back to MSI */
1536 e1000e_reset_interrupt_capability(adapter
);
1537 adapter
->int_mode
= E1000E_INT_MODE_MSI
;
1538 e1000e_set_interrupt_capability(adapter
);
1540 if (adapter
->flags
& FLAG_MSI_ENABLED
) {
1541 err
= request_irq(adapter
->pdev
->irq
, &e1000_intr_msi
, 0,
1542 netdev
->name
, netdev
);
1546 /* fall back to legacy interrupt */
1547 e1000e_reset_interrupt_capability(adapter
);
1548 adapter
->int_mode
= E1000E_INT_MODE_LEGACY
;
1551 err
= request_irq(adapter
->pdev
->irq
, &e1000_intr
, IRQF_SHARED
,
1552 netdev
->name
, netdev
);
1554 e_err("Unable to allocate interrupt, Error: %d\n", err
);
1559 static void e1000_free_irq(struct e1000_adapter
*adapter
)
1561 struct net_device
*netdev
= adapter
->netdev
;
1563 if (adapter
->msix_entries
) {
1566 free_irq(adapter
->msix_entries
[vector
].vector
, netdev
);
1569 free_irq(adapter
->msix_entries
[vector
].vector
, netdev
);
1572 /* Other Causes interrupt vector */
1573 free_irq(adapter
->msix_entries
[vector
].vector
, netdev
);
1577 free_irq(adapter
->pdev
->irq
, netdev
);
1581 * e1000_irq_disable - Mask off interrupt generation on the NIC
1583 static void e1000_irq_disable(struct e1000_adapter
*adapter
)
1585 struct e1000_hw
*hw
= &adapter
->hw
;
1588 if (adapter
->msix_entries
)
1589 ew32(EIAC_82574
, 0);
1591 synchronize_irq(adapter
->pdev
->irq
);
1595 * e1000_irq_enable - Enable default interrupt generation settings
1597 static void e1000_irq_enable(struct e1000_adapter
*adapter
)
1599 struct e1000_hw
*hw
= &adapter
->hw
;
1601 if (adapter
->msix_entries
) {
1602 ew32(EIAC_82574
, adapter
->eiac_mask
& E1000_EIAC_MASK_82574
);
1603 ew32(IMS
, adapter
->eiac_mask
| E1000_IMS_OTHER
| E1000_IMS_LSC
);
1605 ew32(IMS
, IMS_ENABLE_MASK
);
1611 * e1000_get_hw_control - get control of the h/w from f/w
1612 * @adapter: address of board private structure
1614 * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1615 * For ASF and Pass Through versions of f/w this means that
1616 * the driver is loaded. For AMT version (only with 82573)
1617 * of the f/w this means that the network i/f is open.
1619 static void e1000_get_hw_control(struct e1000_adapter
*adapter
)
1621 struct e1000_hw
*hw
= &adapter
->hw
;
1625 /* Let firmware know the driver has taken over */
1626 if (adapter
->flags
& FLAG_HAS_SWSM_ON_LOAD
) {
1628 ew32(SWSM
, swsm
| E1000_SWSM_DRV_LOAD
);
1629 } else if (adapter
->flags
& FLAG_HAS_CTRLEXT_ON_LOAD
) {
1630 ctrl_ext
= er32(CTRL_EXT
);
1631 ew32(CTRL_EXT
, ctrl_ext
| E1000_CTRL_EXT_DRV_LOAD
);
1636 * e1000_release_hw_control - release control of the h/w to f/w
1637 * @adapter: address of board private structure
1639 * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1640 * For ASF and Pass Through versions of f/w this means that the
1641 * driver is no longer loaded. For AMT version (only with 82573) i
1642 * of the f/w this means that the network i/f is closed.
1645 static void e1000_release_hw_control(struct e1000_adapter
*adapter
)
1647 struct e1000_hw
*hw
= &adapter
->hw
;
1651 /* Let firmware taken over control of h/w */
1652 if (adapter
->flags
& FLAG_HAS_SWSM_ON_LOAD
) {
1654 ew32(SWSM
, swsm
& ~E1000_SWSM_DRV_LOAD
);
1655 } else if (adapter
->flags
& FLAG_HAS_CTRLEXT_ON_LOAD
) {
1656 ctrl_ext
= er32(CTRL_EXT
);
1657 ew32(CTRL_EXT
, ctrl_ext
& ~E1000_CTRL_EXT_DRV_LOAD
);
1662 * @e1000_alloc_ring - allocate memory for a ring structure
1664 static int e1000_alloc_ring_dma(struct e1000_adapter
*adapter
,
1665 struct e1000_ring
*ring
)
1667 struct pci_dev
*pdev
= adapter
->pdev
;
1669 ring
->desc
= dma_alloc_coherent(&pdev
->dev
, ring
->size
, &ring
->dma
,
1678 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1679 * @adapter: board private structure
1681 * Return 0 on success, negative on failure
1683 int e1000e_setup_tx_resources(struct e1000_adapter
*adapter
)
1685 struct e1000_ring
*tx_ring
= adapter
->tx_ring
;
1686 int err
= -ENOMEM
, size
;
1688 size
= sizeof(struct e1000_buffer
) * tx_ring
->count
;
1689 tx_ring
->buffer_info
= vmalloc(size
);
1690 if (!tx_ring
->buffer_info
)
1692 memset(tx_ring
->buffer_info
, 0, size
);
1694 /* round up to nearest 4K */
1695 tx_ring
->size
= tx_ring
->count
* sizeof(struct e1000_tx_desc
);
1696 tx_ring
->size
= ALIGN(tx_ring
->size
, 4096);
1698 err
= e1000_alloc_ring_dma(adapter
, tx_ring
);
1702 tx_ring
->next_to_use
= 0;
1703 tx_ring
->next_to_clean
= 0;
1704 spin_lock_init(&adapter
->tx_queue_lock
);
1708 vfree(tx_ring
->buffer_info
);
1709 e_err("Unable to allocate memory for the transmit descriptor ring\n");
1714 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1715 * @adapter: board private structure
1717 * Returns 0 on success, negative on failure
1719 int e1000e_setup_rx_resources(struct e1000_adapter
*adapter
)
1721 struct e1000_ring
*rx_ring
= adapter
->rx_ring
;
1722 struct e1000_buffer
*buffer_info
;
1723 int i
, size
, desc_len
, err
= -ENOMEM
;
1725 size
= sizeof(struct e1000_buffer
) * rx_ring
->count
;
1726 rx_ring
->buffer_info
= vmalloc(size
);
1727 if (!rx_ring
->buffer_info
)
1729 memset(rx_ring
->buffer_info
, 0, size
);
1731 for (i
= 0; i
< rx_ring
->count
; i
++) {
1732 buffer_info
= &rx_ring
->buffer_info
[i
];
1733 buffer_info
->ps_pages
= kcalloc(PS_PAGE_BUFFERS
,
1734 sizeof(struct e1000_ps_page
),
1736 if (!buffer_info
->ps_pages
)
1740 desc_len
= sizeof(union e1000_rx_desc_packet_split
);
1742 /* Round up to nearest 4K */
1743 rx_ring
->size
= rx_ring
->count
* desc_len
;
1744 rx_ring
->size
= ALIGN(rx_ring
->size
, 4096);
1746 err
= e1000_alloc_ring_dma(adapter
, rx_ring
);
1750 rx_ring
->next_to_clean
= 0;
1751 rx_ring
->next_to_use
= 0;
1752 rx_ring
->rx_skb_top
= NULL
;
1757 for (i
= 0; i
< rx_ring
->count
; i
++) {
1758 buffer_info
= &rx_ring
->buffer_info
[i
];
1759 kfree(buffer_info
->ps_pages
);
1762 vfree(rx_ring
->buffer_info
);
1763 e_err("Unable to allocate memory for the transmit descriptor ring\n");
1768 * e1000_clean_tx_ring - Free Tx Buffers
1769 * @adapter: board private structure
1771 static void e1000_clean_tx_ring(struct e1000_adapter
*adapter
)
1773 struct e1000_ring
*tx_ring
= adapter
->tx_ring
;
1774 struct e1000_buffer
*buffer_info
;
1778 for (i
= 0; i
< tx_ring
->count
; i
++) {
1779 buffer_info
= &tx_ring
->buffer_info
[i
];
1780 e1000_put_txbuf(adapter
, buffer_info
);
1783 size
= sizeof(struct e1000_buffer
) * tx_ring
->count
;
1784 memset(tx_ring
->buffer_info
, 0, size
);
1786 memset(tx_ring
->desc
, 0, tx_ring
->size
);
1788 tx_ring
->next_to_use
= 0;
1789 tx_ring
->next_to_clean
= 0;
1791 writel(0, adapter
->hw
.hw_addr
+ tx_ring
->head
);
1792 writel(0, adapter
->hw
.hw_addr
+ tx_ring
->tail
);
1796 * e1000e_free_tx_resources - Free Tx Resources per Queue
1797 * @adapter: board private structure
1799 * Free all transmit software resources
1801 void e1000e_free_tx_resources(struct e1000_adapter
*adapter
)
1803 struct pci_dev
*pdev
= adapter
->pdev
;
1804 struct e1000_ring
*tx_ring
= adapter
->tx_ring
;
1806 e1000_clean_tx_ring(adapter
);
1808 vfree(tx_ring
->buffer_info
);
1809 tx_ring
->buffer_info
= NULL
;
1811 dma_free_coherent(&pdev
->dev
, tx_ring
->size
, tx_ring
->desc
,
1813 tx_ring
->desc
= NULL
;
1817 * e1000e_free_rx_resources - Free Rx Resources
1818 * @adapter: board private structure
1820 * Free all receive software resources
1823 void e1000e_free_rx_resources(struct e1000_adapter
*adapter
)
1825 struct pci_dev
*pdev
= adapter
->pdev
;
1826 struct e1000_ring
*rx_ring
= adapter
->rx_ring
;
1829 e1000_clean_rx_ring(adapter
);
1831 for (i
= 0; i
< rx_ring
->count
; i
++) {
1832 kfree(rx_ring
->buffer_info
[i
].ps_pages
);
1835 vfree(rx_ring
->buffer_info
);
1836 rx_ring
->buffer_info
= NULL
;
1838 dma_free_coherent(&pdev
->dev
, rx_ring
->size
, rx_ring
->desc
,
1840 rx_ring
->desc
= NULL
;
1844 * e1000_update_itr - update the dynamic ITR value based on statistics
1845 * @adapter: pointer to adapter
1846 * @itr_setting: current adapter->itr
1847 * @packets: the number of packets during this measurement interval
1848 * @bytes: the number of bytes during this measurement interval
1850 * Stores a new ITR value based on packets and byte
1851 * counts during the last interrupt. The advantage of per interrupt
1852 * computation is faster updates and more accurate ITR for the current
1853 * traffic pattern. Constants in this function were computed
1854 * based on theoretical maximum wire speed and thresholds were set based
1855 * on testing data as well as attempting to minimize response time
1856 * while increasing bulk throughput. This functionality is controlled
1857 * by the InterruptThrottleRate module parameter.
1859 static unsigned int e1000_update_itr(struct e1000_adapter
*adapter
,
1860 u16 itr_setting
, int packets
,
1863 unsigned int retval
= itr_setting
;
1866 goto update_itr_done
;
1868 switch (itr_setting
) {
1869 case lowest_latency
:
1870 /* handle TSO and jumbo frames */
1871 if (bytes
/packets
> 8000)
1872 retval
= bulk_latency
;
1873 else if ((packets
< 5) && (bytes
> 512)) {
1874 retval
= low_latency
;
1877 case low_latency
: /* 50 usec aka 20000 ints/s */
1878 if (bytes
> 10000) {
1879 /* this if handles the TSO accounting */
1880 if (bytes
/packets
> 8000) {
1881 retval
= bulk_latency
;
1882 } else if ((packets
< 10) || ((bytes
/packets
) > 1200)) {
1883 retval
= bulk_latency
;
1884 } else if ((packets
> 35)) {
1885 retval
= lowest_latency
;
1887 } else if (bytes
/packets
> 2000) {
1888 retval
= bulk_latency
;
1889 } else if (packets
<= 2 && bytes
< 512) {
1890 retval
= lowest_latency
;
1893 case bulk_latency
: /* 250 usec aka 4000 ints/s */
1894 if (bytes
> 25000) {
1896 retval
= low_latency
;
1898 } else if (bytes
< 6000) {
1899 retval
= low_latency
;
1908 static void e1000_set_itr(struct e1000_adapter
*adapter
)
1910 struct e1000_hw
*hw
= &adapter
->hw
;
1912 u32 new_itr
= adapter
->itr
;
1914 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1915 if (adapter
->link_speed
!= SPEED_1000
) {
1921 adapter
->tx_itr
= e1000_update_itr(adapter
,
1923 adapter
->total_tx_packets
,
1924 adapter
->total_tx_bytes
);
1925 /* conservative mode (itr 3) eliminates the lowest_latency setting */
1926 if (adapter
->itr_setting
== 3 && adapter
->tx_itr
== lowest_latency
)
1927 adapter
->tx_itr
= low_latency
;
1929 adapter
->rx_itr
= e1000_update_itr(adapter
,
1931 adapter
->total_rx_packets
,
1932 adapter
->total_rx_bytes
);
1933 /* conservative mode (itr 3) eliminates the lowest_latency setting */
1934 if (adapter
->itr_setting
== 3 && adapter
->rx_itr
== lowest_latency
)
1935 adapter
->rx_itr
= low_latency
;
1937 current_itr
= max(adapter
->rx_itr
, adapter
->tx_itr
);
1939 switch (current_itr
) {
1940 /* counts and packets in update_itr are dependent on these numbers */
1941 case lowest_latency
:
1945 new_itr
= 20000; /* aka hwitr = ~200 */
1955 if (new_itr
!= adapter
->itr
) {
1957 * this attempts to bias the interrupt rate towards Bulk
1958 * by adding intermediate steps when interrupt rate is
1961 new_itr
= new_itr
> adapter
->itr
?
1962 min(adapter
->itr
+ (new_itr
>> 2), new_itr
) :
1964 adapter
->itr
= new_itr
;
1965 adapter
->rx_ring
->itr_val
= new_itr
;
1966 if (adapter
->msix_entries
)
1967 adapter
->rx_ring
->set_itr
= 1;
1969 ew32(ITR
, 1000000000 / (new_itr
* 256));
1974 * e1000_alloc_queues - Allocate memory for all rings
1975 * @adapter: board private structure to initialize
1977 static int __devinit
e1000_alloc_queues(struct e1000_adapter
*adapter
)
1979 adapter
->tx_ring
= kzalloc(sizeof(struct e1000_ring
), GFP_KERNEL
);
1980 if (!adapter
->tx_ring
)
1983 adapter
->rx_ring
= kzalloc(sizeof(struct e1000_ring
), GFP_KERNEL
);
1984 if (!adapter
->rx_ring
)
1989 e_err("Unable to allocate memory for queues\n");
1990 kfree(adapter
->rx_ring
);
1991 kfree(adapter
->tx_ring
);
1996 * e1000_clean - NAPI Rx polling callback
1997 * @napi: struct associated with this polling callback
1998 * @budget: amount of packets driver is allowed to process this poll
2000 static int e1000_clean(struct napi_struct
*napi
, int budget
)
2002 struct e1000_adapter
*adapter
= container_of(napi
, struct e1000_adapter
, napi
);
2003 struct e1000_hw
*hw
= &adapter
->hw
;
2004 struct net_device
*poll_dev
= adapter
->netdev
;
2005 int tx_cleaned
= 0, work_done
= 0;
2007 adapter
= netdev_priv(poll_dev
);
2009 if (adapter
->msix_entries
&&
2010 !(adapter
->rx_ring
->ims_val
& adapter
->tx_ring
->ims_val
))
2014 * e1000_clean is called per-cpu. This lock protects
2015 * tx_ring from being cleaned by multiple cpus
2016 * simultaneously. A failure obtaining the lock means
2017 * tx_ring is currently being cleaned anyway.
2019 if (spin_trylock(&adapter
->tx_queue_lock
)) {
2020 tx_cleaned
= e1000_clean_tx_irq(adapter
);
2021 spin_unlock(&adapter
->tx_queue_lock
);
2025 adapter
->clean_rx(adapter
, &work_done
, budget
);
2030 /* If budget not fully consumed, exit the polling mode */
2031 if (work_done
< budget
) {
2032 if (adapter
->itr_setting
& 3)
2033 e1000_set_itr(adapter
);
2034 netif_rx_complete(napi
);
2035 if (adapter
->msix_entries
)
2036 ew32(IMS
, adapter
->rx_ring
->ims_val
);
2038 e1000_irq_enable(adapter
);
2044 static void e1000_vlan_rx_add_vid(struct net_device
*netdev
, u16 vid
)
2046 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
2047 struct e1000_hw
*hw
= &adapter
->hw
;
2050 /* don't update vlan cookie if already programmed */
2051 if ((adapter
->hw
.mng_cookie
.status
&
2052 E1000_MNG_DHCP_COOKIE_STATUS_VLAN
) &&
2053 (vid
== adapter
->mng_vlan_id
))
2055 /* add VID to filter table */
2056 index
= (vid
>> 5) & 0x7F;
2057 vfta
= E1000_READ_REG_ARRAY(hw
, E1000_VFTA
, index
);
2058 vfta
|= (1 << (vid
& 0x1F));
2059 e1000e_write_vfta(hw
, index
, vfta
);
2062 static void e1000_vlan_rx_kill_vid(struct net_device
*netdev
, u16 vid
)
2064 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
2065 struct e1000_hw
*hw
= &adapter
->hw
;
2068 if (!test_bit(__E1000_DOWN
, &adapter
->state
))
2069 e1000_irq_disable(adapter
);
2070 vlan_group_set_device(adapter
->vlgrp
, vid
, NULL
);
2072 if (!test_bit(__E1000_DOWN
, &adapter
->state
))
2073 e1000_irq_enable(adapter
);
2075 if ((adapter
->hw
.mng_cookie
.status
&
2076 E1000_MNG_DHCP_COOKIE_STATUS_VLAN
) &&
2077 (vid
== adapter
->mng_vlan_id
)) {
2078 /* release control to f/w */
2079 e1000_release_hw_control(adapter
);
2083 /* remove VID from filter table */
2084 index
= (vid
>> 5) & 0x7F;
2085 vfta
= E1000_READ_REG_ARRAY(hw
, E1000_VFTA
, index
);
2086 vfta
&= ~(1 << (vid
& 0x1F));
2087 e1000e_write_vfta(hw
, index
, vfta
);
2090 static void e1000_update_mng_vlan(struct e1000_adapter
*adapter
)
2092 struct net_device
*netdev
= adapter
->netdev
;
2093 u16 vid
= adapter
->hw
.mng_cookie
.vlan_id
;
2094 u16 old_vid
= adapter
->mng_vlan_id
;
2096 if (!adapter
->vlgrp
)
2099 if (!vlan_group_get_device(adapter
->vlgrp
, vid
)) {
2100 adapter
->mng_vlan_id
= E1000_MNG_VLAN_NONE
;
2101 if (adapter
->hw
.mng_cookie
.status
&
2102 E1000_MNG_DHCP_COOKIE_STATUS_VLAN
) {
2103 e1000_vlan_rx_add_vid(netdev
, vid
);
2104 adapter
->mng_vlan_id
= vid
;
2107 if ((old_vid
!= (u16
)E1000_MNG_VLAN_NONE
) &&
2109 !vlan_group_get_device(adapter
->vlgrp
, old_vid
))
2110 e1000_vlan_rx_kill_vid(netdev
, old_vid
);
2112 adapter
->mng_vlan_id
= vid
;
2117 static void e1000_vlan_rx_register(struct net_device
*netdev
,
2118 struct vlan_group
*grp
)
2120 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
2121 struct e1000_hw
*hw
= &adapter
->hw
;
2124 if (!test_bit(__E1000_DOWN
, &adapter
->state
))
2125 e1000_irq_disable(adapter
);
2126 adapter
->vlgrp
= grp
;
2129 /* enable VLAN tag insert/strip */
2131 ctrl
|= E1000_CTRL_VME
;
2134 if (adapter
->flags
& FLAG_HAS_HW_VLAN_FILTER
) {
2135 /* enable VLAN receive filtering */
2137 rctl
&= ~E1000_RCTL_CFIEN
;
2139 e1000_update_mng_vlan(adapter
);
2142 /* disable VLAN tag insert/strip */
2144 ctrl
&= ~E1000_CTRL_VME
;
2147 if (adapter
->flags
& FLAG_HAS_HW_VLAN_FILTER
) {
2148 if (adapter
->mng_vlan_id
!=
2149 (u16
)E1000_MNG_VLAN_NONE
) {
2150 e1000_vlan_rx_kill_vid(netdev
,
2151 adapter
->mng_vlan_id
);
2152 adapter
->mng_vlan_id
= E1000_MNG_VLAN_NONE
;
2157 if (!test_bit(__E1000_DOWN
, &adapter
->state
))
2158 e1000_irq_enable(adapter
);
2161 static void e1000_restore_vlan(struct e1000_adapter
*adapter
)
2165 e1000_vlan_rx_register(adapter
->netdev
, adapter
->vlgrp
);
2167 if (!adapter
->vlgrp
)
2170 for (vid
= 0; vid
< VLAN_GROUP_ARRAY_LEN
; vid
++) {
2171 if (!vlan_group_get_device(adapter
->vlgrp
, vid
))
2173 e1000_vlan_rx_add_vid(adapter
->netdev
, vid
);
2177 static void e1000_init_manageability(struct e1000_adapter
*adapter
)
2179 struct e1000_hw
*hw
= &adapter
->hw
;
2182 if (!(adapter
->flags
& FLAG_MNG_PT_ENABLED
))
2188 * enable receiving management packets to the host. this will probably
2189 * generate destination unreachable messages from the host OS, but
2190 * the packets will be handled on SMBUS
2192 manc
|= E1000_MANC_EN_MNG2HOST
;
2193 manc2h
= er32(MANC2H
);
2194 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2195 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2196 manc2h
|= E1000_MNG2HOST_PORT_623
;
2197 manc2h
|= E1000_MNG2HOST_PORT_664
;
2198 ew32(MANC2H
, manc2h
);
2203 * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2204 * @adapter: board private structure
2206 * Configure the Tx unit of the MAC after a reset.
2208 static void e1000_configure_tx(struct e1000_adapter
*adapter
)
2210 struct e1000_hw
*hw
= &adapter
->hw
;
2211 struct e1000_ring
*tx_ring
= adapter
->tx_ring
;
2213 u32 tdlen
, tctl
, tipg
, tarc
;
2216 /* Setup the HW Tx Head and Tail descriptor pointers */
2217 tdba
= tx_ring
->dma
;
2218 tdlen
= tx_ring
->count
* sizeof(struct e1000_tx_desc
);
2219 ew32(TDBAL
, (tdba
& DMA_32BIT_MASK
));
2220 ew32(TDBAH
, (tdba
>> 32));
2224 tx_ring
->head
= E1000_TDH
;
2225 tx_ring
->tail
= E1000_TDT
;
2227 /* Set the default values for the Tx Inter Packet Gap timer */
2228 tipg
= DEFAULT_82543_TIPG_IPGT_COPPER
; /* 8 */
2229 ipgr1
= DEFAULT_82543_TIPG_IPGR1
; /* 8 */
2230 ipgr2
= DEFAULT_82543_TIPG_IPGR2
; /* 6 */
2232 if (adapter
->flags
& FLAG_TIPG_MEDIUM_FOR_80003ESLAN
)
2233 ipgr2
= DEFAULT_80003ES2LAN_TIPG_IPGR2
; /* 7 */
2235 tipg
|= ipgr1
<< E1000_TIPG_IPGR1_SHIFT
;
2236 tipg
|= ipgr2
<< E1000_TIPG_IPGR2_SHIFT
;
2239 /* Set the Tx Interrupt Delay register */
2240 ew32(TIDV
, adapter
->tx_int_delay
);
2241 /* Tx irq moderation */
2242 ew32(TADV
, adapter
->tx_abs_int_delay
);
2244 /* Program the Transmit Control Register */
2246 tctl
&= ~E1000_TCTL_CT
;
2247 tctl
|= E1000_TCTL_PSP
| E1000_TCTL_RTLC
|
2248 (E1000_COLLISION_THRESHOLD
<< E1000_CT_SHIFT
);
2250 if (adapter
->flags
& FLAG_TARC_SPEED_MODE_BIT
) {
2251 tarc
= er32(TARC(0));
2253 * set the speed mode bit, we'll clear it if we're not at
2254 * gigabit link later
2256 #define SPEED_MODE_BIT (1 << 21)
2257 tarc
|= SPEED_MODE_BIT
;
2258 ew32(TARC(0), tarc
);
2261 /* errata: program both queues to unweighted RR */
2262 if (adapter
->flags
& FLAG_TARC_SET_BIT_ZERO
) {
2263 tarc
= er32(TARC(0));
2265 ew32(TARC(0), tarc
);
2266 tarc
= er32(TARC(1));
2268 ew32(TARC(1), tarc
);
2271 e1000e_config_collision_dist(hw
);
2273 /* Setup Transmit Descriptor Settings for eop descriptor */
2274 adapter
->txd_cmd
= E1000_TXD_CMD_EOP
| E1000_TXD_CMD_IFCS
;
2276 /* only set IDE if we are delaying interrupts using the timers */
2277 if (adapter
->tx_int_delay
)
2278 adapter
->txd_cmd
|= E1000_TXD_CMD_IDE
;
2280 /* enable Report Status bit */
2281 adapter
->txd_cmd
|= E1000_TXD_CMD_RS
;
2285 adapter
->tx_queue_len
= adapter
->netdev
->tx_queue_len
;
2289 * e1000_setup_rctl - configure the receive control registers
2290 * @adapter: Board private structure
2292 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2293 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2294 static void e1000_setup_rctl(struct e1000_adapter
*adapter
)
2296 struct e1000_hw
*hw
= &adapter
->hw
;
2301 /* Program MC offset vector base */
2303 rctl
&= ~(3 << E1000_RCTL_MO_SHIFT
);
2304 rctl
|= E1000_RCTL_EN
| E1000_RCTL_BAM
|
2305 E1000_RCTL_LBM_NO
| E1000_RCTL_RDMTS_HALF
|
2306 (adapter
->hw
.mac
.mc_filter_type
<< E1000_RCTL_MO_SHIFT
);
2308 /* Do not Store bad packets */
2309 rctl
&= ~E1000_RCTL_SBP
;
2311 /* Enable Long Packet receive */
2312 if (adapter
->netdev
->mtu
<= ETH_DATA_LEN
)
2313 rctl
&= ~E1000_RCTL_LPE
;
2315 rctl
|= E1000_RCTL_LPE
;
2317 /* Some systems expect that the CRC is included in SMBUS traffic. The
2318 * hardware strips the CRC before sending to both SMBUS (BMC) and to
2319 * host memory when this is enabled
2321 if (adapter
->flags2
& FLAG2_CRC_STRIPPING
)
2322 rctl
|= E1000_RCTL_SECRC
;
2324 /* Setup buffer sizes */
2325 rctl
&= ~E1000_RCTL_SZ_4096
;
2326 rctl
|= E1000_RCTL_BSEX
;
2327 switch (adapter
->rx_buffer_len
) {
2329 rctl
|= E1000_RCTL_SZ_256
;
2330 rctl
&= ~E1000_RCTL_BSEX
;
2333 rctl
|= E1000_RCTL_SZ_512
;
2334 rctl
&= ~E1000_RCTL_BSEX
;
2337 rctl
|= E1000_RCTL_SZ_1024
;
2338 rctl
&= ~E1000_RCTL_BSEX
;
2342 rctl
|= E1000_RCTL_SZ_2048
;
2343 rctl
&= ~E1000_RCTL_BSEX
;
2346 rctl
|= E1000_RCTL_SZ_4096
;
2349 rctl
|= E1000_RCTL_SZ_8192
;
2352 rctl
|= E1000_RCTL_SZ_16384
;
2357 * 82571 and greater support packet-split where the protocol
2358 * header is placed in skb->data and the packet data is
2359 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2360 * In the case of a non-split, skb->data is linearly filled,
2361 * followed by the page buffers. Therefore, skb->data is
2362 * sized to hold the largest protocol header.
2364 * allocations using alloc_page take too long for regular MTU
2365 * so only enable packet split for jumbo frames
2367 * Using pages when the page size is greater than 16k wastes
2368 * a lot of memory, since we allocate 3 pages at all times
2371 pages
= PAGE_USE_COUNT(adapter
->netdev
->mtu
);
2372 if (!(adapter
->flags
& FLAG_IS_ICH
) && (pages
<= 3) &&
2373 (PAGE_SIZE
<= 16384) && (rctl
& E1000_RCTL_LPE
))
2374 adapter
->rx_ps_pages
= pages
;
2376 adapter
->rx_ps_pages
= 0;
2378 if (adapter
->rx_ps_pages
) {
2379 /* Configure extra packet-split registers */
2380 rfctl
= er32(RFCTL
);
2381 rfctl
|= E1000_RFCTL_EXTEN
;
2383 * disable packet split support for IPv6 extension headers,
2384 * because some malformed IPv6 headers can hang the Rx
2386 rfctl
|= (E1000_RFCTL_IPV6_EX_DIS
|
2387 E1000_RFCTL_NEW_IPV6_EXT_DIS
);
2391 /* Enable Packet split descriptors */
2392 rctl
|= E1000_RCTL_DTYP_PS
;
2394 psrctl
|= adapter
->rx_ps_bsize0
>>
2395 E1000_PSRCTL_BSIZE0_SHIFT
;
2397 switch (adapter
->rx_ps_pages
) {
2399 psrctl
|= PAGE_SIZE
<<
2400 E1000_PSRCTL_BSIZE3_SHIFT
;
2402 psrctl
|= PAGE_SIZE
<<
2403 E1000_PSRCTL_BSIZE2_SHIFT
;
2405 psrctl
|= PAGE_SIZE
>>
2406 E1000_PSRCTL_BSIZE1_SHIFT
;
2410 ew32(PSRCTL
, psrctl
);
2414 /* just started the receive unit, no need to restart */
2415 adapter
->flags
&= ~FLAG_RX_RESTART_NOW
;
2419 * e1000_configure_rx - Configure Receive Unit after Reset
2420 * @adapter: board private structure
2422 * Configure the Rx unit of the MAC after a reset.
2424 static void e1000_configure_rx(struct e1000_adapter
*adapter
)
2426 struct e1000_hw
*hw
= &adapter
->hw
;
2427 struct e1000_ring
*rx_ring
= adapter
->rx_ring
;
2429 u32 rdlen
, rctl
, rxcsum
, ctrl_ext
;
2431 if (adapter
->rx_ps_pages
) {
2432 /* this is a 32 byte descriptor */
2433 rdlen
= rx_ring
->count
*
2434 sizeof(union e1000_rx_desc_packet_split
);
2435 adapter
->clean_rx
= e1000_clean_rx_irq_ps
;
2436 adapter
->alloc_rx_buf
= e1000_alloc_rx_buffers_ps
;
2437 } else if (adapter
->netdev
->mtu
> ETH_FRAME_LEN
+ ETH_FCS_LEN
) {
2438 rdlen
= rx_ring
->count
* sizeof(struct e1000_rx_desc
);
2439 adapter
->clean_rx
= e1000_clean_jumbo_rx_irq
;
2440 adapter
->alloc_rx_buf
= e1000_alloc_jumbo_rx_buffers
;
2442 rdlen
= rx_ring
->count
* sizeof(struct e1000_rx_desc
);
2443 adapter
->clean_rx
= e1000_clean_rx_irq
;
2444 adapter
->alloc_rx_buf
= e1000_alloc_rx_buffers
;
2447 /* disable receives while setting up the descriptors */
2449 ew32(RCTL
, rctl
& ~E1000_RCTL_EN
);
2453 /* set the Receive Delay Timer Register */
2454 ew32(RDTR
, adapter
->rx_int_delay
);
2456 /* irq moderation */
2457 ew32(RADV
, adapter
->rx_abs_int_delay
);
2458 if (adapter
->itr_setting
!= 0)
2459 ew32(ITR
, 1000000000 / (adapter
->itr
* 256));
2461 ctrl_ext
= er32(CTRL_EXT
);
2462 /* Reset delay timers after every interrupt */
2463 ctrl_ext
|= E1000_CTRL_EXT_INT_TIMER_CLR
;
2464 /* Auto-Mask interrupts upon ICR access */
2465 ctrl_ext
|= E1000_CTRL_EXT_IAME
;
2466 ew32(IAM
, 0xffffffff);
2467 ew32(CTRL_EXT
, ctrl_ext
);
2471 * Setup the HW Rx Head and Tail Descriptor Pointers and
2472 * the Base and Length of the Rx Descriptor Ring
2474 rdba
= rx_ring
->dma
;
2475 ew32(RDBAL
, (rdba
& DMA_32BIT_MASK
));
2476 ew32(RDBAH
, (rdba
>> 32));
2480 rx_ring
->head
= E1000_RDH
;
2481 rx_ring
->tail
= E1000_RDT
;
2483 /* Enable Receive Checksum Offload for TCP and UDP */
2484 rxcsum
= er32(RXCSUM
);
2485 if (adapter
->flags
& FLAG_RX_CSUM_ENABLED
) {
2486 rxcsum
|= E1000_RXCSUM_TUOFL
;
2489 * IPv4 payload checksum for UDP fragments must be
2490 * used in conjunction with packet-split.
2492 if (adapter
->rx_ps_pages
)
2493 rxcsum
|= E1000_RXCSUM_IPPCSE
;
2495 rxcsum
&= ~E1000_RXCSUM_TUOFL
;
2496 /* no need to clear IPPCSE as it defaults to 0 */
2498 ew32(RXCSUM
, rxcsum
);
2501 * Enable early receives on supported devices, only takes effect when
2502 * packet size is equal or larger than the specified value (in 8 byte
2503 * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2505 if ((adapter
->flags
& FLAG_HAS_ERT
) &&
2506 (adapter
->netdev
->mtu
> ETH_DATA_LEN
)) {
2507 u32 rxdctl
= er32(RXDCTL(0));
2508 ew32(RXDCTL(0), rxdctl
| 0x3);
2509 ew32(ERT
, E1000_ERT_2048
| (1 << 13));
2511 * With jumbo frames and early-receive enabled, excessive
2512 * C4->C2 latencies result in dropped transactions.
2514 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY
,
2515 e1000e_driver_name
, 55);
2517 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY
,
2519 PM_QOS_DEFAULT_VALUE
);
2522 /* Enable Receives */
2527 * e1000_update_mc_addr_list - Update Multicast addresses
2528 * @hw: pointer to the HW structure
2529 * @mc_addr_list: array of multicast addresses to program
2530 * @mc_addr_count: number of multicast addresses to program
2531 * @rar_used_count: the first RAR register free to program
2532 * @rar_count: total number of supported Receive Address Registers
2534 * Updates the Receive Address Registers and Multicast Table Array.
2535 * The caller must have a packed mc_addr_list of multicast addresses.
2536 * The parameter rar_count will usually be hw->mac.rar_entry_count
2537 * unless there are workarounds that change this. Currently no func pointer
2538 * exists and all implementations are handled in the generic version of this
2541 static void e1000_update_mc_addr_list(struct e1000_hw
*hw
, u8
*mc_addr_list
,
2542 u32 mc_addr_count
, u32 rar_used_count
,
2545 hw
->mac
.ops
.update_mc_addr_list(hw
, mc_addr_list
, mc_addr_count
,
2546 rar_used_count
, rar_count
);
2550 * e1000_set_multi - Multicast and Promiscuous mode set
2551 * @netdev: network interface device structure
2553 * The set_multi entry point is called whenever the multicast address
2554 * list or the network interface flags are updated. This routine is
2555 * responsible for configuring the hardware for proper multicast,
2556 * promiscuous mode, and all-multi behavior.
2558 static void e1000_set_multi(struct net_device
*netdev
)
2560 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
2561 struct e1000_hw
*hw
= &adapter
->hw
;
2562 struct e1000_mac_info
*mac
= &hw
->mac
;
2563 struct dev_mc_list
*mc_ptr
;
2568 /* Check for Promiscuous and All Multicast modes */
2572 if (netdev
->flags
& IFF_PROMISC
) {
2573 rctl
|= (E1000_RCTL_UPE
| E1000_RCTL_MPE
);
2574 rctl
&= ~E1000_RCTL_VFE
;
2576 if (netdev
->flags
& IFF_ALLMULTI
) {
2577 rctl
|= E1000_RCTL_MPE
;
2578 rctl
&= ~E1000_RCTL_UPE
;
2580 rctl
&= ~(E1000_RCTL_UPE
| E1000_RCTL_MPE
);
2582 if (adapter
->flags
& FLAG_HAS_HW_VLAN_FILTER
)
2583 rctl
|= E1000_RCTL_VFE
;
2588 if (netdev
->mc_count
) {
2589 mta_list
= kmalloc(netdev
->mc_count
* 6, GFP_ATOMIC
);
2593 /* prepare a packed array of only addresses. */
2594 mc_ptr
= netdev
->mc_list
;
2596 for (i
= 0; i
< netdev
->mc_count
; i
++) {
2599 memcpy(mta_list
+ (i
*ETH_ALEN
), mc_ptr
->dmi_addr
,
2601 mc_ptr
= mc_ptr
->next
;
2604 e1000_update_mc_addr_list(hw
, mta_list
, i
, 1,
2605 mac
->rar_entry_count
);
2609 * if we're called from probe, we might not have
2610 * anything to do here, so clear out the list
2612 e1000_update_mc_addr_list(hw
, NULL
, 0, 1, mac
->rar_entry_count
);
2617 * e1000_configure - configure the hardware for Rx and Tx
2618 * @adapter: private board structure
2620 static void e1000_configure(struct e1000_adapter
*adapter
)
2622 e1000_set_multi(adapter
->netdev
);
2624 e1000_restore_vlan(adapter
);
2625 e1000_init_manageability(adapter
);
2627 e1000_configure_tx(adapter
);
2628 e1000_setup_rctl(adapter
);
2629 e1000_configure_rx(adapter
);
2630 adapter
->alloc_rx_buf(adapter
, e1000_desc_unused(adapter
->rx_ring
));
2634 * e1000e_power_up_phy - restore link in case the phy was powered down
2635 * @adapter: address of board private structure
2637 * The phy may be powered down to save power and turn off link when the
2638 * driver is unloaded and wake on lan is not enabled (among others)
2639 * *** this routine MUST be followed by a call to e1000e_reset ***
2641 void e1000e_power_up_phy(struct e1000_adapter
*adapter
)
2645 /* Just clear the power down bit to wake the phy back up */
2646 if (adapter
->hw
.phy
.media_type
== e1000_media_type_copper
) {
2648 * According to the manual, the phy will retain its
2649 * settings across a power-down/up cycle
2651 e1e_rphy(&adapter
->hw
, PHY_CONTROL
, &mii_reg
);
2652 mii_reg
&= ~MII_CR_POWER_DOWN
;
2653 e1e_wphy(&adapter
->hw
, PHY_CONTROL
, mii_reg
);
2656 adapter
->hw
.mac
.ops
.setup_link(&adapter
->hw
);
2660 * e1000_power_down_phy - Power down the PHY
2662 * Power down the PHY so no link is implied when interface is down
2663 * The PHY cannot be powered down is management or WoL is active
2665 static void e1000_power_down_phy(struct e1000_adapter
*adapter
)
2667 struct e1000_hw
*hw
= &adapter
->hw
;
2670 /* WoL is enabled */
2674 /* non-copper PHY? */
2675 if (adapter
->hw
.phy
.media_type
!= e1000_media_type_copper
)
2678 /* reset is blocked because of a SoL/IDER session */
2679 if (e1000e_check_mng_mode(hw
) || e1000_check_reset_block(hw
))
2682 /* manageability (AMT) is enabled */
2683 if (er32(MANC
) & E1000_MANC_SMBUS_EN
)
2686 /* power down the PHY */
2687 e1e_rphy(hw
, PHY_CONTROL
, &mii_reg
);
2688 mii_reg
|= MII_CR_POWER_DOWN
;
2689 e1e_wphy(hw
, PHY_CONTROL
, mii_reg
);
2694 * e1000e_reset - bring the hardware into a known good state
2696 * This function boots the hardware and enables some settings that
2697 * require a configuration cycle of the hardware - those cannot be
2698 * set/changed during runtime. After reset the device needs to be
2699 * properly configured for Rx, Tx etc.
2701 void e1000e_reset(struct e1000_adapter
*adapter
)
2703 struct e1000_mac_info
*mac
= &adapter
->hw
.mac
;
2704 struct e1000_fc_info
*fc
= &adapter
->hw
.fc
;
2705 struct e1000_hw
*hw
= &adapter
->hw
;
2706 u32 tx_space
, min_tx_space
, min_rx_space
;
2707 u32 pba
= adapter
->pba
;
2710 /* reset Packet Buffer Allocation to default */
2713 if (adapter
->max_frame_size
> ETH_FRAME_LEN
+ ETH_FCS_LEN
) {
2715 * To maintain wire speed transmits, the Tx FIFO should be
2716 * large enough to accommodate two full transmit packets,
2717 * rounded up to the next 1KB and expressed in KB. Likewise,
2718 * the Rx FIFO should be large enough to accommodate at least
2719 * one full receive packet and is similarly rounded up and
2723 /* upper 16 bits has Tx packet buffer allocation size in KB */
2724 tx_space
= pba
>> 16;
2725 /* lower 16 bits has Rx packet buffer allocation size in KB */
2728 * the Tx fifo also stores 16 bytes of information about the tx
2729 * but don't include ethernet FCS because hardware appends it
2731 min_tx_space
= (adapter
->max_frame_size
+
2732 sizeof(struct e1000_tx_desc
) -
2734 min_tx_space
= ALIGN(min_tx_space
, 1024);
2735 min_tx_space
>>= 10;
2736 /* software strips receive CRC, so leave room for it */
2737 min_rx_space
= adapter
->max_frame_size
;
2738 min_rx_space
= ALIGN(min_rx_space
, 1024);
2739 min_rx_space
>>= 10;
2742 * If current Tx allocation is less than the min Tx FIFO size,
2743 * and the min Tx FIFO size is less than the current Rx FIFO
2744 * allocation, take space away from current Rx allocation
2746 if ((tx_space
< min_tx_space
) &&
2747 ((min_tx_space
- tx_space
) < pba
)) {
2748 pba
-= min_tx_space
- tx_space
;
2751 * if short on Rx space, Rx wins and must trump tx
2752 * adjustment or use Early Receive if available
2754 if ((pba
< min_rx_space
) &&
2755 (!(adapter
->flags
& FLAG_HAS_ERT
)))
2756 /* ERT enabled in e1000_configure_rx */
2765 * flow control settings
2767 * The high water mark must be low enough to fit one full frame
2768 * (or the size used for early receive) above it in the Rx FIFO.
2769 * Set it to the lower of:
2770 * - 90% of the Rx FIFO size, and
2771 * - the full Rx FIFO size minus the early receive size (for parts
2772 * with ERT support assuming ERT set to E1000_ERT_2048), or
2773 * - the full Rx FIFO size minus one full frame
2775 if (adapter
->flags
& FLAG_HAS_ERT
)
2776 hwm
= min(((pba
<< 10) * 9 / 10),
2777 ((pba
<< 10) - (E1000_ERT_2048
<< 3)));
2779 hwm
= min(((pba
<< 10) * 9 / 10),
2780 ((pba
<< 10) - adapter
->max_frame_size
));
2782 fc
->high_water
= hwm
& 0xFFF8; /* 8-byte granularity */
2783 fc
->low_water
= fc
->high_water
- 8;
2785 if (adapter
->flags
& FLAG_DISABLE_FC_PAUSE_TIME
)
2786 fc
->pause_time
= 0xFFFF;
2788 fc
->pause_time
= E1000_FC_PAUSE_TIME
;
2790 fc
->current_mode
= fc
->requested_mode
;
2792 /* Allow time for pending master requests to run */
2793 mac
->ops
.reset_hw(hw
);
2796 * For parts with AMT enabled, let the firmware know
2797 * that the network interface is in control
2799 if (adapter
->flags
& FLAG_HAS_AMT
)
2800 e1000_get_hw_control(adapter
);
2804 if (mac
->ops
.init_hw(hw
))
2805 e_err("Hardware Error\n");
2807 e1000_update_mng_vlan(adapter
);
2809 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2810 ew32(VET
, ETH_P_8021Q
);
2812 e1000e_reset_adaptive(hw
);
2813 e1000_get_phy_info(hw
);
2815 if (!(adapter
->flags
& FLAG_SMART_POWER_DOWN
)) {
2818 * speed up time to link by disabling smart power down, ignore
2819 * the return value of this function because there is nothing
2820 * different we would do if it failed
2822 e1e_rphy(hw
, IGP02E1000_PHY_POWER_MGMT
, &phy_data
);
2823 phy_data
&= ~IGP02E1000_PM_SPD
;
2824 e1e_wphy(hw
, IGP02E1000_PHY_POWER_MGMT
, phy_data
);
2828 int e1000e_up(struct e1000_adapter
*adapter
)
2830 struct e1000_hw
*hw
= &adapter
->hw
;
2832 /* hardware has been reset, we need to reload some things */
2833 e1000_configure(adapter
);
2835 clear_bit(__E1000_DOWN
, &adapter
->state
);
2837 napi_enable(&adapter
->napi
);
2838 if (adapter
->msix_entries
)
2839 e1000_configure_msix(adapter
);
2840 e1000_irq_enable(adapter
);
2842 /* fire a link change interrupt to start the watchdog */
2843 ew32(ICS
, E1000_ICS_LSC
);
2847 void e1000e_down(struct e1000_adapter
*adapter
)
2849 struct net_device
*netdev
= adapter
->netdev
;
2850 struct e1000_hw
*hw
= &adapter
->hw
;
2854 * signal that we're down so the interrupt handler does not
2855 * reschedule our watchdog timer
2857 set_bit(__E1000_DOWN
, &adapter
->state
);
2859 /* disable receives in the hardware */
2861 ew32(RCTL
, rctl
& ~E1000_RCTL_EN
);
2862 /* flush and sleep below */
2864 netif_tx_stop_all_queues(netdev
);
2866 /* disable transmits in the hardware */
2868 tctl
&= ~E1000_TCTL_EN
;
2870 /* flush both disables and wait for them to finish */
2874 napi_disable(&adapter
->napi
);
2875 e1000_irq_disable(adapter
);
2877 del_timer_sync(&adapter
->watchdog_timer
);
2878 del_timer_sync(&adapter
->phy_info_timer
);
2880 netdev
->tx_queue_len
= adapter
->tx_queue_len
;
2881 netif_carrier_off(netdev
);
2882 adapter
->link_speed
= 0;
2883 adapter
->link_duplex
= 0;
2885 if (!pci_channel_offline(adapter
->pdev
))
2886 e1000e_reset(adapter
);
2887 e1000_clean_tx_ring(adapter
);
2888 e1000_clean_rx_ring(adapter
);
2891 * TODO: for power management, we could drop the link and
2892 * pci_disable_device here.
2896 void e1000e_reinit_locked(struct e1000_adapter
*adapter
)
2899 while (test_and_set_bit(__E1000_RESETTING
, &adapter
->state
))
2901 e1000e_down(adapter
);
2903 clear_bit(__E1000_RESETTING
, &adapter
->state
);
2907 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2908 * @adapter: board private structure to initialize
2910 * e1000_sw_init initializes the Adapter private data structure.
2911 * Fields are initialized based on PCI device information and
2912 * OS network device settings (MTU size).
2914 static int __devinit
e1000_sw_init(struct e1000_adapter
*adapter
)
2916 struct net_device
*netdev
= adapter
->netdev
;
2918 adapter
->rx_buffer_len
= ETH_FRAME_LEN
+ VLAN_HLEN
+ ETH_FCS_LEN
;
2919 adapter
->rx_ps_bsize0
= 128;
2920 adapter
->max_frame_size
= netdev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
;
2921 adapter
->min_frame_size
= ETH_ZLEN
+ ETH_FCS_LEN
;
2923 e1000e_set_interrupt_capability(adapter
);
2925 if (e1000_alloc_queues(adapter
))
2928 spin_lock_init(&adapter
->tx_queue_lock
);
2930 /* Explicitly disable IRQ since the NIC can be in any state. */
2931 e1000_irq_disable(adapter
);
2933 set_bit(__E1000_DOWN
, &adapter
->state
);
2938 * e1000_intr_msi_test - Interrupt Handler
2939 * @irq: interrupt number
2940 * @data: pointer to a network interface device structure
2942 static irqreturn_t
e1000_intr_msi_test(int irq
, void *data
)
2944 struct net_device
*netdev
= data
;
2945 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
2946 struct e1000_hw
*hw
= &adapter
->hw
;
2947 u32 icr
= er32(ICR
);
2949 e_dbg("%s: icr is %08X\n", netdev
->name
, icr
);
2950 if (icr
& E1000_ICR_RXSEQ
) {
2951 adapter
->flags
&= ~FLAG_MSI_TEST_FAILED
;
2959 * e1000_test_msi_interrupt - Returns 0 for successful test
2960 * @adapter: board private struct
2962 * code flow taken from tg3.c
2964 static int e1000_test_msi_interrupt(struct e1000_adapter
*adapter
)
2966 struct net_device
*netdev
= adapter
->netdev
;
2967 struct e1000_hw
*hw
= &adapter
->hw
;
2970 /* poll_enable hasn't been called yet, so don't need disable */
2971 /* clear any pending events */
2974 /* free the real vector and request a test handler */
2975 e1000_free_irq(adapter
);
2976 e1000e_reset_interrupt_capability(adapter
);
2978 /* Assume that the test fails, if it succeeds then the test
2979 * MSI irq handler will unset this flag */
2980 adapter
->flags
|= FLAG_MSI_TEST_FAILED
;
2982 err
= pci_enable_msi(adapter
->pdev
);
2984 goto msi_test_failed
;
2986 err
= request_irq(adapter
->pdev
->irq
, &e1000_intr_msi_test
, 0,
2987 netdev
->name
, netdev
);
2989 pci_disable_msi(adapter
->pdev
);
2990 goto msi_test_failed
;
2995 e1000_irq_enable(adapter
);
2997 /* fire an unusual interrupt on the test handler */
2998 ew32(ICS
, E1000_ICS_RXSEQ
);
3002 e1000_irq_disable(adapter
);
3006 if (adapter
->flags
& FLAG_MSI_TEST_FAILED
) {
3007 adapter
->int_mode
= E1000E_INT_MODE_LEGACY
;
3009 e_info("MSI interrupt test failed!\n");
3012 free_irq(adapter
->pdev
->irq
, netdev
);
3013 pci_disable_msi(adapter
->pdev
);
3016 goto msi_test_failed
;
3018 /* okay so the test worked, restore settings */
3019 e_dbg("%s: MSI interrupt test succeeded!\n", netdev
->name
);
3021 e1000e_set_interrupt_capability(adapter
);
3022 e1000_request_irq(adapter
);
3027 * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3028 * @adapter: board private struct
3030 * code flow taken from tg3.c, called with e1000 interrupts disabled.
3032 static int e1000_test_msi(struct e1000_adapter
*adapter
)
3037 if (!(adapter
->flags
& FLAG_MSI_ENABLED
))
3040 /* disable SERR in case the MSI write causes a master abort */
3041 pci_read_config_word(adapter
->pdev
, PCI_COMMAND
, &pci_cmd
);
3042 pci_write_config_word(adapter
->pdev
, PCI_COMMAND
,
3043 pci_cmd
& ~PCI_COMMAND_SERR
);
3045 err
= e1000_test_msi_interrupt(adapter
);
3047 /* restore previous setting of command word */
3048 pci_write_config_word(adapter
->pdev
, PCI_COMMAND
, pci_cmd
);
3054 /* EIO means MSI test failed */
3058 /* back to INTx mode */
3059 e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3061 e1000_free_irq(adapter
);
3063 err
= e1000_request_irq(adapter
);
3069 * e1000_open - Called when a network interface is made active
3070 * @netdev: network interface device structure
3072 * Returns 0 on success, negative value on failure
3074 * The open entry point is called when a network interface is made
3075 * active by the system (IFF_UP). At this point all resources needed
3076 * for transmit and receive operations are allocated, the interrupt
3077 * handler is registered with the OS, the watchdog timer is started,
3078 * and the stack is notified that the interface is ready.
3080 static int e1000_open(struct net_device
*netdev
)
3082 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
3083 struct e1000_hw
*hw
= &adapter
->hw
;
3086 /* disallow open during test */
3087 if (test_bit(__E1000_TESTING
, &adapter
->state
))
3090 /* allocate transmit descriptors */
3091 err
= e1000e_setup_tx_resources(adapter
);
3095 /* allocate receive descriptors */
3096 err
= e1000e_setup_rx_resources(adapter
);
3100 e1000e_power_up_phy(adapter
);
3102 adapter
->mng_vlan_id
= E1000_MNG_VLAN_NONE
;
3103 if ((adapter
->hw
.mng_cookie
.status
&
3104 E1000_MNG_DHCP_COOKIE_STATUS_VLAN
))
3105 e1000_update_mng_vlan(adapter
);
3108 * If AMT is enabled, let the firmware know that the network
3109 * interface is now open
3111 if (adapter
->flags
& FLAG_HAS_AMT
)
3112 e1000_get_hw_control(adapter
);
3115 * before we allocate an interrupt, we must be ready to handle it.
3116 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3117 * as soon as we call pci_request_irq, so we have to setup our
3118 * clean_rx handler before we do so.
3120 e1000_configure(adapter
);
3122 err
= e1000_request_irq(adapter
);
3127 * Work around PCIe errata with MSI interrupts causing some chipsets to
3128 * ignore e1000e MSI messages, which means we need to test our MSI
3131 if (adapter
->int_mode
!= E1000E_INT_MODE_LEGACY
) {
3132 err
= e1000_test_msi(adapter
);
3134 e_err("Interrupt allocation failed\n");
3139 /* From here on the code is the same as e1000e_up() */
3140 clear_bit(__E1000_DOWN
, &adapter
->state
);
3142 napi_enable(&adapter
->napi
);
3144 e1000_irq_enable(adapter
);
3146 netif_tx_start_all_queues(netdev
);
3148 /* fire a link status change interrupt to start the watchdog */
3149 ew32(ICS
, E1000_ICS_LSC
);
3154 e1000_release_hw_control(adapter
);
3155 e1000_power_down_phy(adapter
);
3156 e1000e_free_rx_resources(adapter
);
3158 e1000e_free_tx_resources(adapter
);
3160 e1000e_reset(adapter
);
3166 * e1000_close - Disables a network interface
3167 * @netdev: network interface device structure
3169 * Returns 0, this is not allowed to fail
3171 * The close entry point is called when an interface is de-activated
3172 * by the OS. The hardware is still under the drivers control, but
3173 * needs to be disabled. A global MAC reset is issued to stop the
3174 * hardware, and all transmit and receive resources are freed.
3176 static int e1000_close(struct net_device
*netdev
)
3178 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
3180 WARN_ON(test_bit(__E1000_RESETTING
, &adapter
->state
));
3181 e1000e_down(adapter
);
3182 e1000_power_down_phy(adapter
);
3183 e1000_free_irq(adapter
);
3185 e1000e_free_tx_resources(adapter
);
3186 e1000e_free_rx_resources(adapter
);
3189 * kill manageability vlan ID if supported, but not if a vlan with
3190 * the same ID is registered on the host OS (let 8021q kill it)
3192 if ((adapter
->hw
.mng_cookie
.status
&
3193 E1000_MNG_DHCP_COOKIE_STATUS_VLAN
) &&
3195 vlan_group_get_device(adapter
->vlgrp
, adapter
->mng_vlan_id
)))
3196 e1000_vlan_rx_kill_vid(netdev
, adapter
->mng_vlan_id
);
3199 * If AMT is enabled, let the firmware know that the network
3200 * interface is now closed
3202 if (adapter
->flags
& FLAG_HAS_AMT
)
3203 e1000_release_hw_control(adapter
);
3208 * e1000_set_mac - Change the Ethernet Address of the NIC
3209 * @netdev: network interface device structure
3210 * @p: pointer to an address structure
3212 * Returns 0 on success, negative on failure
3214 static int e1000_set_mac(struct net_device
*netdev
, void *p
)
3216 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
3217 struct sockaddr
*addr
= p
;
3219 if (!is_valid_ether_addr(addr
->sa_data
))
3220 return -EADDRNOTAVAIL
;
3222 memcpy(netdev
->dev_addr
, addr
->sa_data
, netdev
->addr_len
);
3223 memcpy(adapter
->hw
.mac
.addr
, addr
->sa_data
, netdev
->addr_len
);
3225 e1000e_rar_set(&adapter
->hw
, adapter
->hw
.mac
.addr
, 0);
3227 if (adapter
->flags
& FLAG_RESET_OVERWRITES_LAA
) {
3228 /* activate the work around */
3229 e1000e_set_laa_state_82571(&adapter
->hw
, 1);
3232 * Hold a copy of the LAA in RAR[14] This is done so that
3233 * between the time RAR[0] gets clobbered and the time it
3234 * gets fixed (in e1000_watchdog), the actual LAA is in one
3235 * of the RARs and no incoming packets directed to this port
3236 * are dropped. Eventually the LAA will be in RAR[0] and
3239 e1000e_rar_set(&adapter
->hw
,
3240 adapter
->hw
.mac
.addr
,
3241 adapter
->hw
.mac
.rar_entry_count
- 1);
3248 * e1000e_update_phy_task - work thread to update phy
3249 * @work: pointer to our work struct
3251 * this worker thread exists because we must acquire a
3252 * semaphore to read the phy, which we could msleep while
3253 * waiting for it, and we can't msleep in a timer.
3255 static void e1000e_update_phy_task(struct work_struct
*work
)
3257 struct e1000_adapter
*adapter
= container_of(work
,
3258 struct e1000_adapter
, update_phy_task
);
3259 e1000_get_phy_info(&adapter
->hw
);
3263 * Need to wait a few seconds after link up to get diagnostic information from
3266 static void e1000_update_phy_info(unsigned long data
)
3268 struct e1000_adapter
*adapter
= (struct e1000_adapter
*) data
;
3269 schedule_work(&adapter
->update_phy_task
);
3273 * e1000e_update_stats - Update the board statistics counters
3274 * @adapter: board private structure
3276 void e1000e_update_stats(struct e1000_adapter
*adapter
)
3278 struct e1000_hw
*hw
= &adapter
->hw
;
3279 struct pci_dev
*pdev
= adapter
->pdev
;
3282 * Prevent stats update while adapter is being reset, or if the pci
3283 * connection is down.
3285 if (adapter
->link_speed
== 0)
3287 if (pci_channel_offline(pdev
))
3290 adapter
->stats
.crcerrs
+= er32(CRCERRS
);
3291 adapter
->stats
.gprc
+= er32(GPRC
);
3292 adapter
->stats
.gorc
+= er32(GORCL
);
3293 er32(GORCH
); /* Clear gorc */
3294 adapter
->stats
.bprc
+= er32(BPRC
);
3295 adapter
->stats
.mprc
+= er32(MPRC
);
3296 adapter
->stats
.roc
+= er32(ROC
);
3298 adapter
->stats
.mpc
+= er32(MPC
);
3299 adapter
->stats
.scc
+= er32(SCC
);
3300 adapter
->stats
.ecol
+= er32(ECOL
);
3301 adapter
->stats
.mcc
+= er32(MCC
);
3302 adapter
->stats
.latecol
+= er32(LATECOL
);
3303 adapter
->stats
.dc
+= er32(DC
);
3304 adapter
->stats
.xonrxc
+= er32(XONRXC
);
3305 adapter
->stats
.xontxc
+= er32(XONTXC
);
3306 adapter
->stats
.xoffrxc
+= er32(XOFFRXC
);
3307 adapter
->stats
.xofftxc
+= er32(XOFFTXC
);
3308 adapter
->stats
.gptc
+= er32(GPTC
);
3309 adapter
->stats
.gotc
+= er32(GOTCL
);
3310 er32(GOTCH
); /* Clear gotc */
3311 adapter
->stats
.rnbc
+= er32(RNBC
);
3312 adapter
->stats
.ruc
+= er32(RUC
);
3314 adapter
->stats
.mptc
+= er32(MPTC
);
3315 adapter
->stats
.bptc
+= er32(BPTC
);
3317 /* used for adaptive IFS */
3319 hw
->mac
.tx_packet_delta
= er32(TPT
);
3320 adapter
->stats
.tpt
+= hw
->mac
.tx_packet_delta
;
3321 hw
->mac
.collision_delta
= er32(COLC
);
3322 adapter
->stats
.colc
+= hw
->mac
.collision_delta
;
3324 adapter
->stats
.algnerrc
+= er32(ALGNERRC
);
3325 adapter
->stats
.rxerrc
+= er32(RXERRC
);
3326 if (hw
->mac
.type
!= e1000_82574
)
3327 adapter
->stats
.tncrs
+= er32(TNCRS
);
3328 adapter
->stats
.cexterr
+= er32(CEXTERR
);
3329 adapter
->stats
.tsctc
+= er32(TSCTC
);
3330 adapter
->stats
.tsctfc
+= er32(TSCTFC
);
3332 /* Fill out the OS statistics structure */
3333 adapter
->net_stats
.multicast
= adapter
->stats
.mprc
;
3334 adapter
->net_stats
.collisions
= adapter
->stats
.colc
;
3339 * RLEC on some newer hardware can be incorrect so build
3340 * our own version based on RUC and ROC
3342 adapter
->net_stats
.rx_errors
= adapter
->stats
.rxerrc
+
3343 adapter
->stats
.crcerrs
+ adapter
->stats
.algnerrc
+
3344 adapter
->stats
.ruc
+ adapter
->stats
.roc
+
3345 adapter
->stats
.cexterr
;
3346 adapter
->net_stats
.rx_length_errors
= adapter
->stats
.ruc
+
3348 adapter
->net_stats
.rx_crc_errors
= adapter
->stats
.crcerrs
;
3349 adapter
->net_stats
.rx_frame_errors
= adapter
->stats
.algnerrc
;
3350 adapter
->net_stats
.rx_missed_errors
= adapter
->stats
.mpc
;
3353 adapter
->net_stats
.tx_errors
= adapter
->stats
.ecol
+
3354 adapter
->stats
.latecol
;
3355 adapter
->net_stats
.tx_aborted_errors
= adapter
->stats
.ecol
;
3356 adapter
->net_stats
.tx_window_errors
= adapter
->stats
.latecol
;
3357 adapter
->net_stats
.tx_carrier_errors
= adapter
->stats
.tncrs
;
3359 /* Tx Dropped needs to be maintained elsewhere */
3361 /* Management Stats */
3362 adapter
->stats
.mgptc
+= er32(MGTPTC
);
3363 adapter
->stats
.mgprc
+= er32(MGTPRC
);
3364 adapter
->stats
.mgpdc
+= er32(MGTPDC
);
3368 * e1000_phy_read_status - Update the PHY register status snapshot
3369 * @adapter: board private structure
3371 static void e1000_phy_read_status(struct e1000_adapter
*adapter
)
3373 struct e1000_hw
*hw
= &adapter
->hw
;
3374 struct e1000_phy_regs
*phy
= &adapter
->phy_regs
;
3377 if ((er32(STATUS
) & E1000_STATUS_LU
) &&
3378 (adapter
->hw
.phy
.media_type
== e1000_media_type_copper
)) {
3379 ret_val
= e1e_rphy(hw
, PHY_CONTROL
, &phy
->bmcr
);
3380 ret_val
|= e1e_rphy(hw
, PHY_STATUS
, &phy
->bmsr
);
3381 ret_val
|= e1e_rphy(hw
, PHY_AUTONEG_ADV
, &phy
->advertise
);
3382 ret_val
|= e1e_rphy(hw
, PHY_LP_ABILITY
, &phy
->lpa
);
3383 ret_val
|= e1e_rphy(hw
, PHY_AUTONEG_EXP
, &phy
->expansion
);
3384 ret_val
|= e1e_rphy(hw
, PHY_1000T_CTRL
, &phy
->ctrl1000
);
3385 ret_val
|= e1e_rphy(hw
, PHY_1000T_STATUS
, &phy
->stat1000
);
3386 ret_val
|= e1e_rphy(hw
, PHY_EXT_STATUS
, &phy
->estatus
);
3388 e_warn("Error reading PHY register\n");
3391 * Do not read PHY registers if link is not up
3392 * Set values to typical power-on defaults
3394 phy
->bmcr
= (BMCR_SPEED1000
| BMCR_ANENABLE
| BMCR_FULLDPLX
);
3395 phy
->bmsr
= (BMSR_100FULL
| BMSR_100HALF
| BMSR_10FULL
|
3396 BMSR_10HALF
| BMSR_ESTATEN
| BMSR_ANEGCAPABLE
|
3398 phy
->advertise
= (ADVERTISE_PAUSE_ASYM
| ADVERTISE_PAUSE_CAP
|
3399 ADVERTISE_ALL
| ADVERTISE_CSMA
);
3401 phy
->expansion
= EXPANSION_ENABLENPAGE
;
3402 phy
->ctrl1000
= ADVERTISE_1000FULL
;
3404 phy
->estatus
= (ESTATUS_1000_TFULL
| ESTATUS_1000_THALF
);
3408 static void e1000_print_link_info(struct e1000_adapter
*adapter
)
3410 struct e1000_hw
*hw
= &adapter
->hw
;
3411 u32 ctrl
= er32(CTRL
);
3413 /* Link status message must follow this format for user tools */
3414 printk(KERN_INFO
"e1000e: %s NIC Link is Up %d Mbps %s, "
3415 "Flow Control: %s\n",
3416 adapter
->netdev
->name
,
3417 adapter
->link_speed
,
3418 (adapter
->link_duplex
== FULL_DUPLEX
) ?
3419 "Full Duplex" : "Half Duplex",
3420 ((ctrl
& E1000_CTRL_TFCE
) && (ctrl
& E1000_CTRL_RFCE
)) ?
3422 ((ctrl
& E1000_CTRL_RFCE
) ? "RX" :
3423 ((ctrl
& E1000_CTRL_TFCE
) ? "TX" : "None" )));
3426 bool e1000_has_link(struct e1000_adapter
*adapter
)
3428 struct e1000_hw
*hw
= &adapter
->hw
;
3429 bool link_active
= 0;
3433 * get_link_status is set on LSC (link status) interrupt or
3434 * Rx sequence error interrupt. get_link_status will stay
3435 * false until the check_for_link establishes link
3436 * for copper adapters ONLY
3438 switch (hw
->phy
.media_type
) {
3439 case e1000_media_type_copper
:
3440 if (hw
->mac
.get_link_status
) {
3441 ret_val
= hw
->mac
.ops
.check_for_link(hw
);
3442 link_active
= !hw
->mac
.get_link_status
;
3447 case e1000_media_type_fiber
:
3448 ret_val
= hw
->mac
.ops
.check_for_link(hw
);
3449 link_active
= !!(er32(STATUS
) & E1000_STATUS_LU
);
3451 case e1000_media_type_internal_serdes
:
3452 ret_val
= hw
->mac
.ops
.check_for_link(hw
);
3453 link_active
= adapter
->hw
.mac
.serdes_has_link
;
3456 case e1000_media_type_unknown
:
3460 if ((ret_val
== E1000_ERR_PHY
) && (hw
->phy
.type
== e1000_phy_igp_3
) &&
3461 (er32(CTRL
) & E1000_PHY_CTRL_GBE_DISABLE
)) {
3462 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3463 e_info("Gigabit has been disabled, downgrading speed\n");
3469 static void e1000e_enable_receives(struct e1000_adapter
*adapter
)
3471 /* make sure the receive unit is started */
3472 if ((adapter
->flags
& FLAG_RX_NEEDS_RESTART
) &&
3473 (adapter
->flags
& FLAG_RX_RESTART_NOW
)) {
3474 struct e1000_hw
*hw
= &adapter
->hw
;
3475 u32 rctl
= er32(RCTL
);
3476 ew32(RCTL
, rctl
| E1000_RCTL_EN
);
3477 adapter
->flags
&= ~FLAG_RX_RESTART_NOW
;
3482 * e1000_watchdog - Timer Call-back
3483 * @data: pointer to adapter cast into an unsigned long
3485 static void e1000_watchdog(unsigned long data
)
3487 struct e1000_adapter
*adapter
= (struct e1000_adapter
*) data
;
3489 /* Do the rest outside of interrupt context */
3490 schedule_work(&adapter
->watchdog_task
);
3492 /* TODO: make this use queue_delayed_work() */
3495 static void e1000_watchdog_task(struct work_struct
*work
)
3497 struct e1000_adapter
*adapter
= container_of(work
,
3498 struct e1000_adapter
, watchdog_task
);
3499 struct net_device
*netdev
= adapter
->netdev
;
3500 struct e1000_mac_info
*mac
= &adapter
->hw
.mac
;
3501 struct e1000_phy_info
*phy
= &adapter
->hw
.phy
;
3502 struct e1000_ring
*tx_ring
= adapter
->tx_ring
;
3503 struct e1000_hw
*hw
= &adapter
->hw
;
3507 link
= e1000_has_link(adapter
);
3508 if ((netif_carrier_ok(netdev
)) && link
) {
3509 e1000e_enable_receives(adapter
);
3513 if ((e1000e_enable_tx_pkt_filtering(hw
)) &&
3514 (adapter
->mng_vlan_id
!= adapter
->hw
.mng_cookie
.vlan_id
))
3515 e1000_update_mng_vlan(adapter
);
3518 if (!netif_carrier_ok(netdev
)) {
3520 /* update snapshot of PHY registers on LSC */
3521 e1000_phy_read_status(adapter
);
3522 mac
->ops
.get_link_up_info(&adapter
->hw
,
3523 &adapter
->link_speed
,
3524 &adapter
->link_duplex
);
3525 e1000_print_link_info(adapter
);
3527 * On supported PHYs, check for duplex mismatch only
3528 * if link has autonegotiated at 10/100 half
3530 if ((hw
->phy
.type
== e1000_phy_igp_3
||
3531 hw
->phy
.type
== e1000_phy_bm
) &&
3532 (hw
->mac
.autoneg
== true) &&
3533 (adapter
->link_speed
== SPEED_10
||
3534 adapter
->link_speed
== SPEED_100
) &&
3535 (adapter
->link_duplex
== HALF_DUPLEX
)) {
3538 e1e_rphy(hw
, PHY_AUTONEG_EXP
, &autoneg_exp
);
3540 if (!(autoneg_exp
& NWAY_ER_LP_NWAY_CAPS
))
3541 e_info("Autonegotiated half duplex but"
3542 " link partner cannot autoneg. "
3543 " Try forcing full duplex if "
3544 "link gets many collisions.\n");
3548 * tweak tx_queue_len according to speed/duplex
3549 * and adjust the timeout factor
3551 netdev
->tx_queue_len
= adapter
->tx_queue_len
;
3552 adapter
->tx_timeout_factor
= 1;
3553 switch (adapter
->link_speed
) {
3556 netdev
->tx_queue_len
= 10;
3557 adapter
->tx_timeout_factor
= 16;
3561 netdev
->tx_queue_len
= 100;
3562 /* maybe add some timeout factor ? */
3567 * workaround: re-program speed mode bit after
3570 if ((adapter
->flags
& FLAG_TARC_SPEED_MODE_BIT
) &&
3573 tarc0
= er32(TARC(0));
3574 tarc0
&= ~SPEED_MODE_BIT
;
3575 ew32(TARC(0), tarc0
);
3579 * disable TSO for pcie and 10/100 speeds, to avoid
3580 * some hardware issues
3582 if (!(adapter
->flags
& FLAG_TSO_FORCE
)) {
3583 switch (adapter
->link_speed
) {
3586 e_info("10/100 speed: disabling TSO\n");
3587 netdev
->features
&= ~NETIF_F_TSO
;
3588 netdev
->features
&= ~NETIF_F_TSO6
;
3591 netdev
->features
|= NETIF_F_TSO
;
3592 netdev
->features
|= NETIF_F_TSO6
;
3601 * enable transmits in the hardware, need to do this
3602 * after setting TARC(0)
3605 tctl
|= E1000_TCTL_EN
;
3609 * Perform any post-link-up configuration before
3610 * reporting link up.
3612 if (phy
->ops
.cfg_on_link_up
)
3613 phy
->ops
.cfg_on_link_up(hw
);
3615 netif_carrier_on(netdev
);
3616 netif_tx_wake_all_queues(netdev
);
3618 if (!test_bit(__E1000_DOWN
, &adapter
->state
))
3619 mod_timer(&adapter
->phy_info_timer
,
3620 round_jiffies(jiffies
+ 2 * HZ
));
3623 if (netif_carrier_ok(netdev
)) {
3624 adapter
->link_speed
= 0;
3625 adapter
->link_duplex
= 0;
3626 /* Link status message must follow this format */
3627 printk(KERN_INFO
"e1000e: %s NIC Link is Down\n",
3628 adapter
->netdev
->name
);
3629 netif_carrier_off(netdev
);
3630 netif_tx_stop_all_queues(netdev
);
3631 if (!test_bit(__E1000_DOWN
, &adapter
->state
))
3632 mod_timer(&adapter
->phy_info_timer
,
3633 round_jiffies(jiffies
+ 2 * HZ
));
3635 if (adapter
->flags
& FLAG_RX_NEEDS_RESTART
)
3636 schedule_work(&adapter
->reset_task
);
3641 e1000e_update_stats(adapter
);
3643 mac
->tx_packet_delta
= adapter
->stats
.tpt
- adapter
->tpt_old
;
3644 adapter
->tpt_old
= adapter
->stats
.tpt
;
3645 mac
->collision_delta
= adapter
->stats
.colc
- adapter
->colc_old
;
3646 adapter
->colc_old
= adapter
->stats
.colc
;
3648 adapter
->gorc
= adapter
->stats
.gorc
- adapter
->gorc_old
;
3649 adapter
->gorc_old
= adapter
->stats
.gorc
;
3650 adapter
->gotc
= adapter
->stats
.gotc
- adapter
->gotc_old
;
3651 adapter
->gotc_old
= adapter
->stats
.gotc
;
3653 e1000e_update_adaptive(&adapter
->hw
);
3655 if (!netif_carrier_ok(netdev
)) {
3656 tx_pending
= (e1000_desc_unused(tx_ring
) + 1 <
3660 * We've lost link, so the controller stops DMA,
3661 * but we've got queued Tx work that's never going
3662 * to get done, so reset controller to flush Tx.
3663 * (Do the reset outside of interrupt context).
3665 adapter
->tx_timeout_count
++;
3666 schedule_work(&adapter
->reset_task
);
3670 /* Cause software interrupt to ensure Rx ring is cleaned */
3671 if (adapter
->msix_entries
)
3672 ew32(ICS
, adapter
->rx_ring
->ims_val
);
3674 ew32(ICS
, E1000_ICS_RXDMT0
);
3676 /* Force detection of hung controller every watchdog period */
3677 adapter
->detect_tx_hung
= 1;
3680 * With 82571 controllers, LAA may be overwritten due to controller
3681 * reset from the other port. Set the appropriate LAA in RAR[0]
3683 if (e1000e_get_laa_state_82571(hw
))
3684 e1000e_rar_set(hw
, adapter
->hw
.mac
.addr
, 0);
3686 /* Reset the timer */
3687 if (!test_bit(__E1000_DOWN
, &adapter
->state
))
3688 mod_timer(&adapter
->watchdog_timer
,
3689 round_jiffies(jiffies
+ 2 * HZ
));
3692 #define E1000_TX_FLAGS_CSUM 0x00000001
3693 #define E1000_TX_FLAGS_VLAN 0x00000002
3694 #define E1000_TX_FLAGS_TSO 0x00000004
3695 #define E1000_TX_FLAGS_IPV4 0x00000008
3696 #define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
3697 #define E1000_TX_FLAGS_VLAN_SHIFT 16
3699 static int e1000_tso(struct e1000_adapter
*adapter
,
3700 struct sk_buff
*skb
)
3702 struct e1000_ring
*tx_ring
= adapter
->tx_ring
;
3703 struct e1000_context_desc
*context_desc
;
3704 struct e1000_buffer
*buffer_info
;
3707 u16 ipcse
= 0, tucse
, mss
;
3708 u8 ipcss
, ipcso
, tucss
, tucso
, hdr_len
;
3711 if (skb_is_gso(skb
)) {
3712 if (skb_header_cloned(skb
)) {
3713 err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
3718 hdr_len
= skb_transport_offset(skb
) + tcp_hdrlen(skb
);
3719 mss
= skb_shinfo(skb
)->gso_size
;
3720 if (skb
->protocol
== htons(ETH_P_IP
)) {
3721 struct iphdr
*iph
= ip_hdr(skb
);
3724 tcp_hdr(skb
)->check
= ~csum_tcpudp_magic(iph
->saddr
,
3728 cmd_length
= E1000_TXD_CMD_IP
;
3729 ipcse
= skb_transport_offset(skb
) - 1;
3730 } else if (skb_shinfo(skb
)->gso_type
== SKB_GSO_TCPV6
) {
3731 ipv6_hdr(skb
)->payload_len
= 0;
3732 tcp_hdr(skb
)->check
=
3733 ~csum_ipv6_magic(&ipv6_hdr(skb
)->saddr
,
3734 &ipv6_hdr(skb
)->daddr
,
3738 ipcss
= skb_network_offset(skb
);
3739 ipcso
= (void *)&(ip_hdr(skb
)->check
) - (void *)skb
->data
;
3740 tucss
= skb_transport_offset(skb
);
3741 tucso
= (void *)&(tcp_hdr(skb
)->check
) - (void *)skb
->data
;
3744 cmd_length
|= (E1000_TXD_CMD_DEXT
| E1000_TXD_CMD_TSE
|
3745 E1000_TXD_CMD_TCP
| (skb
->len
- (hdr_len
)));
3747 i
= tx_ring
->next_to_use
;
3748 context_desc
= E1000_CONTEXT_DESC(*tx_ring
, i
);
3749 buffer_info
= &tx_ring
->buffer_info
[i
];
3751 context_desc
->lower_setup
.ip_fields
.ipcss
= ipcss
;
3752 context_desc
->lower_setup
.ip_fields
.ipcso
= ipcso
;
3753 context_desc
->lower_setup
.ip_fields
.ipcse
= cpu_to_le16(ipcse
);
3754 context_desc
->upper_setup
.tcp_fields
.tucss
= tucss
;
3755 context_desc
->upper_setup
.tcp_fields
.tucso
= tucso
;
3756 context_desc
->upper_setup
.tcp_fields
.tucse
= cpu_to_le16(tucse
);
3757 context_desc
->tcp_seg_setup
.fields
.mss
= cpu_to_le16(mss
);
3758 context_desc
->tcp_seg_setup
.fields
.hdr_len
= hdr_len
;
3759 context_desc
->cmd_and_length
= cpu_to_le32(cmd_length
);
3761 buffer_info
->time_stamp
= jiffies
;
3762 buffer_info
->next_to_watch
= i
;
3765 if (i
== tx_ring
->count
)
3767 tx_ring
->next_to_use
= i
;
3775 static bool e1000_tx_csum(struct e1000_adapter
*adapter
, struct sk_buff
*skb
)
3777 struct e1000_ring
*tx_ring
= adapter
->tx_ring
;
3778 struct e1000_context_desc
*context_desc
;
3779 struct e1000_buffer
*buffer_info
;
3782 u32 cmd_len
= E1000_TXD_CMD_DEXT
;
3784 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
3787 switch (skb
->protocol
) {
3788 case __constant_htons(ETH_P_IP
):
3789 if (ip_hdr(skb
)->protocol
== IPPROTO_TCP
)
3790 cmd_len
|= E1000_TXD_CMD_TCP
;
3792 case __constant_htons(ETH_P_IPV6
):
3793 /* XXX not handling all IPV6 headers */
3794 if (ipv6_hdr(skb
)->nexthdr
== IPPROTO_TCP
)
3795 cmd_len
|= E1000_TXD_CMD_TCP
;
3798 if (unlikely(net_ratelimit()))
3799 e_warn("checksum_partial proto=%x!\n", skb
->protocol
);
3803 css
= skb_transport_offset(skb
);
3805 i
= tx_ring
->next_to_use
;
3806 buffer_info
= &tx_ring
->buffer_info
[i
];
3807 context_desc
= E1000_CONTEXT_DESC(*tx_ring
, i
);
3809 context_desc
->lower_setup
.ip_config
= 0;
3810 context_desc
->upper_setup
.tcp_fields
.tucss
= css
;
3811 context_desc
->upper_setup
.tcp_fields
.tucso
=
3812 css
+ skb
->csum_offset
;
3813 context_desc
->upper_setup
.tcp_fields
.tucse
= 0;
3814 context_desc
->tcp_seg_setup
.data
= 0;
3815 context_desc
->cmd_and_length
= cpu_to_le32(cmd_len
);
3817 buffer_info
->time_stamp
= jiffies
;
3818 buffer_info
->next_to_watch
= i
;
3821 if (i
== tx_ring
->count
)
3823 tx_ring
->next_to_use
= i
;
3828 #define E1000_MAX_PER_TXD 8192
3829 #define E1000_MAX_TXD_PWR 12
3831 static int e1000_tx_map(struct e1000_adapter
*adapter
,
3832 struct sk_buff
*skb
, unsigned int first
,
3833 unsigned int max_per_txd
, unsigned int nr_frags
,
3836 struct e1000_ring
*tx_ring
= adapter
->tx_ring
;
3837 unsigned int len
= skb_headlen(skb
);
3838 unsigned int offset
, size
, count
= 0, i
;
3842 i
= tx_ring
->next_to_use
;
3844 if (skb_dma_map(&adapter
->pdev
->dev
, skb
, DMA_TO_DEVICE
)) {
3845 dev_err(&adapter
->pdev
->dev
, "TX DMA map failed\n");
3846 adapter
->tx_dma_failed
++;
3851 map
= skb_shinfo(skb
)->dma_maps
[0];
3855 struct e1000_buffer
*buffer_info
= &tx_ring
->buffer_info
[i
];
3856 size
= min(len
, max_per_txd
);
3858 /* Workaround for premature desc write-backs
3859 * in TSO mode. Append 4-byte sentinel desc */
3860 if (mss
&& !nr_frags
&& size
== len
&& size
> 8)
3863 buffer_info
->length
= size
;
3864 /* set time_stamp *before* dma to help avoid a possible race */
3865 buffer_info
->time_stamp
= jiffies
;
3866 buffer_info
->dma
= map
+ offset
;
3867 buffer_info
->next_to_watch
= i
;
3873 if (i
== tx_ring
->count
)
3877 for (f
= 0; f
< nr_frags
; f
++) {
3878 struct skb_frag_struct
*frag
;
3880 frag
= &skb_shinfo(skb
)->frags
[f
];
3882 map
= skb_shinfo(skb
)->dma_maps
[f
+ 1];
3886 struct e1000_buffer
*buffer_info
;
3887 buffer_info
= &tx_ring
->buffer_info
[i
];
3888 size
= min(len
, max_per_txd
);
3889 /* Workaround for premature desc write-backs
3890 * in TSO mode. Append 4-byte sentinel desc */
3891 if (mss
&& f
== (nr_frags
-1) && size
== len
&& size
> 8)
3894 buffer_info
->length
= size
;
3895 buffer_info
->time_stamp
= jiffies
;
3896 buffer_info
->dma
= map
+ offset
;
3897 buffer_info
->next_to_watch
= i
;
3904 if (i
== tx_ring
->count
)
3910 i
= tx_ring
->count
- 1;
3914 tx_ring
->buffer_info
[i
].skb
= skb
;
3915 tx_ring
->buffer_info
[first
].next_to_watch
= i
;
3921 static void e1000_tx_queue(struct e1000_adapter
*adapter
,
3922 int tx_flags
, int count
)
3924 struct e1000_ring
*tx_ring
= adapter
->tx_ring
;
3925 struct e1000_tx_desc
*tx_desc
= NULL
;
3926 struct e1000_buffer
*buffer_info
;
3927 u32 txd_upper
= 0, txd_lower
= E1000_TXD_CMD_IFCS
;
3930 if (tx_flags
& E1000_TX_FLAGS_TSO
) {
3931 txd_lower
|= E1000_TXD_CMD_DEXT
| E1000_TXD_DTYP_D
|
3933 txd_upper
|= E1000_TXD_POPTS_TXSM
<< 8;
3935 if (tx_flags
& E1000_TX_FLAGS_IPV4
)
3936 txd_upper
|= E1000_TXD_POPTS_IXSM
<< 8;
3939 if (tx_flags
& E1000_TX_FLAGS_CSUM
) {
3940 txd_lower
|= E1000_TXD_CMD_DEXT
| E1000_TXD_DTYP_D
;
3941 txd_upper
|= E1000_TXD_POPTS_TXSM
<< 8;
3944 if (tx_flags
& E1000_TX_FLAGS_VLAN
) {
3945 txd_lower
|= E1000_TXD_CMD_VLE
;
3946 txd_upper
|= (tx_flags
& E1000_TX_FLAGS_VLAN_MASK
);
3949 i
= tx_ring
->next_to_use
;
3952 buffer_info
= &tx_ring
->buffer_info
[i
];
3953 tx_desc
= E1000_TX_DESC(*tx_ring
, i
);
3954 tx_desc
->buffer_addr
= cpu_to_le64(buffer_info
->dma
);
3955 tx_desc
->lower
.data
=
3956 cpu_to_le32(txd_lower
| buffer_info
->length
);
3957 tx_desc
->upper
.data
= cpu_to_le32(txd_upper
);
3960 if (i
== tx_ring
->count
)
3964 tx_desc
->lower
.data
|= cpu_to_le32(adapter
->txd_cmd
);
3967 * Force memory writes to complete before letting h/w
3968 * know there are new descriptors to fetch. (Only
3969 * applicable for weak-ordered memory model archs,
3974 tx_ring
->next_to_use
= i
;
3975 writel(i
, adapter
->hw
.hw_addr
+ tx_ring
->tail
);
3977 * we need this if more than one processor can write to our tail
3978 * at a time, it synchronizes IO on IA64/Altix systems
3983 #define MINIMUM_DHCP_PACKET_SIZE 282
3984 static int e1000_transfer_dhcp_info(struct e1000_adapter
*adapter
,
3985 struct sk_buff
*skb
)
3987 struct e1000_hw
*hw
= &adapter
->hw
;
3990 if (vlan_tx_tag_present(skb
)) {
3991 if (!((vlan_tx_tag_get(skb
) == adapter
->hw
.mng_cookie
.vlan_id
)
3992 && (adapter
->hw
.mng_cookie
.status
&
3993 E1000_MNG_DHCP_COOKIE_STATUS_VLAN
)))
3997 if (skb
->len
<= MINIMUM_DHCP_PACKET_SIZE
)
4000 if (((struct ethhdr
*) skb
->data
)->h_proto
!= htons(ETH_P_IP
))
4004 const struct iphdr
*ip
= (struct iphdr
*)((u8
*)skb
->data
+14);
4007 if (ip
->protocol
!= IPPROTO_UDP
)
4010 udp
= (struct udphdr
*)((u8
*)ip
+ (ip
->ihl
<< 2));
4011 if (ntohs(udp
->dest
) != 67)
4014 offset
= (u8
*)udp
+ 8 - skb
->data
;
4015 length
= skb
->len
- offset
;
4016 return e1000e_mng_write_dhcp_info(hw
, (u8
*)udp
+ 8, length
);
4022 static int __e1000_maybe_stop_tx(struct net_device
*netdev
, int size
)
4024 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
4026 netif_stop_queue(netdev
);
4028 * Herbert's original patch had:
4029 * smp_mb__after_netif_stop_queue();
4030 * but since that doesn't exist yet, just open code it.
4035 * We need to check again in a case another CPU has just
4036 * made room available.
4038 if (e1000_desc_unused(adapter
->tx_ring
) < size
)
4042 netif_start_queue(netdev
);
4043 ++adapter
->restart_queue
;
4047 static int e1000_maybe_stop_tx(struct net_device
*netdev
, int size
)
4049 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
4051 if (e1000_desc_unused(adapter
->tx_ring
) >= size
)
4053 return __e1000_maybe_stop_tx(netdev
, size
);
4056 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4057 static int e1000_xmit_frame(struct sk_buff
*skb
, struct net_device
*netdev
)
4059 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
4060 struct e1000_ring
*tx_ring
= adapter
->tx_ring
;
4062 unsigned int max_per_txd
= E1000_MAX_PER_TXD
;
4063 unsigned int max_txd_pwr
= E1000_MAX_TXD_PWR
;
4064 unsigned int tx_flags
= 0;
4065 unsigned int len
= skb
->len
- skb
->data_len
;
4066 unsigned long irq_flags
;
4067 unsigned int nr_frags
;
4073 if (test_bit(__E1000_DOWN
, &adapter
->state
)) {
4074 dev_kfree_skb_any(skb
);
4075 return NETDEV_TX_OK
;
4078 if (skb
->len
<= 0) {
4079 dev_kfree_skb_any(skb
);
4080 return NETDEV_TX_OK
;
4083 mss
= skb_shinfo(skb
)->gso_size
;
4085 * The controller does a simple calculation to
4086 * make sure there is enough room in the FIFO before
4087 * initiating the DMA for each buffer. The calc is:
4088 * 4 = ceil(buffer len/mss). To make sure we don't
4089 * overrun the FIFO, adjust the max buffer len if mss
4094 max_per_txd
= min(mss
<< 2, max_per_txd
);
4095 max_txd_pwr
= fls(max_per_txd
) - 1;
4098 * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4099 * points to just header, pull a few bytes of payload from
4100 * frags into skb->data
4102 hdr_len
= skb_transport_offset(skb
) + tcp_hdrlen(skb
);
4104 * we do this workaround for ES2LAN, but it is un-necessary,
4105 * avoiding it could save a lot of cycles
4107 if (skb
->data_len
&& (hdr_len
== len
)) {
4108 unsigned int pull_size
;
4110 pull_size
= min((unsigned int)4, skb
->data_len
);
4111 if (!__pskb_pull_tail(skb
, pull_size
)) {
4112 e_err("__pskb_pull_tail failed.\n");
4113 dev_kfree_skb_any(skb
);
4114 return NETDEV_TX_OK
;
4116 len
= skb
->len
- skb
->data_len
;
4120 /* reserve a descriptor for the offload context */
4121 if ((mss
) || (skb
->ip_summed
== CHECKSUM_PARTIAL
))
4125 count
+= TXD_USE_COUNT(len
, max_txd_pwr
);
4127 nr_frags
= skb_shinfo(skb
)->nr_frags
;
4128 for (f
= 0; f
< nr_frags
; f
++)
4129 count
+= TXD_USE_COUNT(skb_shinfo(skb
)->frags
[f
].size
,
4132 if (adapter
->hw
.mac
.tx_pkt_filtering
)
4133 e1000_transfer_dhcp_info(adapter
, skb
);
4135 spin_lock_irqsave(&adapter
->tx_queue_lock
, irq_flags
);
4138 * need: count + 2 desc gap to keep tail from touching
4139 * head, otherwise try next time
4141 if (e1000_maybe_stop_tx(netdev
, count
+ 2)) {
4142 spin_unlock_irqrestore(&adapter
->tx_queue_lock
, irq_flags
);
4143 return NETDEV_TX_BUSY
;
4146 if (adapter
->vlgrp
&& vlan_tx_tag_present(skb
)) {
4147 tx_flags
|= E1000_TX_FLAGS_VLAN
;
4148 tx_flags
|= (vlan_tx_tag_get(skb
) << E1000_TX_FLAGS_VLAN_SHIFT
);
4151 first
= tx_ring
->next_to_use
;
4153 tso
= e1000_tso(adapter
, skb
);
4155 dev_kfree_skb_any(skb
);
4156 spin_unlock_irqrestore(&adapter
->tx_queue_lock
, irq_flags
);
4157 return NETDEV_TX_OK
;
4161 tx_flags
|= E1000_TX_FLAGS_TSO
;
4162 else if (e1000_tx_csum(adapter
, skb
))
4163 tx_flags
|= E1000_TX_FLAGS_CSUM
;
4166 * Old method was to assume IPv4 packet by default if TSO was enabled.
4167 * 82571 hardware supports TSO capabilities for IPv6 as well...
4168 * no longer assume, we must.
4170 if (skb
->protocol
== htons(ETH_P_IP
))
4171 tx_flags
|= E1000_TX_FLAGS_IPV4
;
4173 count
= e1000_tx_map(adapter
, skb
, first
, max_per_txd
, nr_frags
, mss
);
4175 /* handle pci_map_single() error in e1000_tx_map */
4176 dev_kfree_skb_any(skb
);
4177 spin_unlock_irqrestore(&adapter
->tx_queue_lock
, irq_flags
);
4178 return NETDEV_TX_OK
;
4181 e1000_tx_queue(adapter
, tx_flags
, count
);
4183 netdev
->trans_start
= jiffies
;
4185 /* Make sure there is space in the ring for the next send. */
4186 e1000_maybe_stop_tx(netdev
, MAX_SKB_FRAGS
+ 2);
4188 spin_unlock_irqrestore(&adapter
->tx_queue_lock
, irq_flags
);
4189 return NETDEV_TX_OK
;
4193 * e1000_tx_timeout - Respond to a Tx Hang
4194 * @netdev: network interface device structure
4196 static void e1000_tx_timeout(struct net_device
*netdev
)
4198 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
4200 /* Do the reset outside of interrupt context */
4201 adapter
->tx_timeout_count
++;
4202 schedule_work(&adapter
->reset_task
);
4205 static void e1000_reset_task(struct work_struct
*work
)
4207 struct e1000_adapter
*adapter
;
4208 adapter
= container_of(work
, struct e1000_adapter
, reset_task
);
4210 e1000e_reinit_locked(adapter
);
4214 * e1000_get_stats - Get System Network Statistics
4215 * @netdev: network interface device structure
4217 * Returns the address of the device statistics structure.
4218 * The statistics are actually updated from the timer callback.
4220 static struct net_device_stats
*e1000_get_stats(struct net_device
*netdev
)
4222 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
4224 /* only return the current stats */
4225 return &adapter
->net_stats
;
4229 * e1000_change_mtu - Change the Maximum Transfer Unit
4230 * @netdev: network interface device structure
4231 * @new_mtu: new value for maximum frame size
4233 * Returns 0 on success, negative on failure
4235 static int e1000_change_mtu(struct net_device
*netdev
, int new_mtu
)
4237 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
4238 int max_frame
= new_mtu
+ ETH_HLEN
+ ETH_FCS_LEN
;
4240 if ((new_mtu
< ETH_ZLEN
+ ETH_FCS_LEN
+ VLAN_HLEN
) ||
4241 (max_frame
> MAX_JUMBO_FRAME_SIZE
)) {
4242 e_err("Invalid MTU setting\n");
4246 /* Jumbo frame size limits */
4247 if (max_frame
> ETH_FRAME_LEN
+ ETH_FCS_LEN
) {
4248 if (!(adapter
->flags
& FLAG_HAS_JUMBO_FRAMES
)) {
4249 e_err("Jumbo Frames not supported.\n");
4252 if (adapter
->hw
.phy
.type
== e1000_phy_ife
) {
4253 e_err("Jumbo Frames not supported.\n");
4258 #define MAX_STD_JUMBO_FRAME_SIZE 9234
4259 if (max_frame
> MAX_STD_JUMBO_FRAME_SIZE
) {
4260 e_err("MTU > 9216 not supported.\n");
4264 while (test_and_set_bit(__E1000_RESETTING
, &adapter
->state
))
4266 /* e1000e_down has a dependency on max_frame_size */
4267 adapter
->max_frame_size
= max_frame
;
4268 if (netif_running(netdev
))
4269 e1000e_down(adapter
);
4272 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4273 * means we reserve 2 more, this pushes us to allocate from the next
4275 * i.e. RXBUFFER_2048 --> size-4096 slab
4276 * However with the new *_jumbo_rx* routines, jumbo receives will use
4280 if (max_frame
<= 256)
4281 adapter
->rx_buffer_len
= 256;
4282 else if (max_frame
<= 512)
4283 adapter
->rx_buffer_len
= 512;
4284 else if (max_frame
<= 1024)
4285 adapter
->rx_buffer_len
= 1024;
4286 else if (max_frame
<= 2048)
4287 adapter
->rx_buffer_len
= 2048;
4289 adapter
->rx_buffer_len
= 4096;
4291 /* adjust allocation if LPE protects us, and we aren't using SBP */
4292 if ((max_frame
== ETH_FRAME_LEN
+ ETH_FCS_LEN
) ||
4293 (max_frame
== ETH_FRAME_LEN
+ VLAN_HLEN
+ ETH_FCS_LEN
))
4294 adapter
->rx_buffer_len
= ETH_FRAME_LEN
+ VLAN_HLEN
4297 e_info("changing MTU from %d to %d\n", netdev
->mtu
, new_mtu
);
4298 netdev
->mtu
= new_mtu
;
4300 if (netif_running(netdev
))
4303 e1000e_reset(adapter
);
4305 clear_bit(__E1000_RESETTING
, &adapter
->state
);
4310 static int e1000_mii_ioctl(struct net_device
*netdev
, struct ifreq
*ifr
,
4313 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
4314 struct mii_ioctl_data
*data
= if_mii(ifr
);
4316 if (adapter
->hw
.phy
.media_type
!= e1000_media_type_copper
)
4321 data
->phy_id
= adapter
->hw
.phy
.addr
;
4324 if (!capable(CAP_NET_ADMIN
))
4326 switch (data
->reg_num
& 0x1F) {
4328 data
->val_out
= adapter
->phy_regs
.bmcr
;
4331 data
->val_out
= adapter
->phy_regs
.bmsr
;
4334 data
->val_out
= (adapter
->hw
.phy
.id
>> 16);
4337 data
->val_out
= (adapter
->hw
.phy
.id
& 0xFFFF);
4340 data
->val_out
= adapter
->phy_regs
.advertise
;
4343 data
->val_out
= adapter
->phy_regs
.lpa
;
4346 data
->val_out
= adapter
->phy_regs
.expansion
;
4349 data
->val_out
= adapter
->phy_regs
.ctrl1000
;
4352 data
->val_out
= adapter
->phy_regs
.stat1000
;
4355 data
->val_out
= adapter
->phy_regs
.estatus
;
4368 static int e1000_ioctl(struct net_device
*netdev
, struct ifreq
*ifr
, int cmd
)
4374 return e1000_mii_ioctl(netdev
, ifr
, cmd
);
4380 static int e1000_suspend(struct pci_dev
*pdev
, pm_message_t state
)
4382 struct net_device
*netdev
= pci_get_drvdata(pdev
);
4383 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
4384 struct e1000_hw
*hw
= &adapter
->hw
;
4385 u32 ctrl
, ctrl_ext
, rctl
, status
;
4386 u32 wufc
= adapter
->wol
;
4389 netif_device_detach(netdev
);
4391 if (netif_running(netdev
)) {
4392 WARN_ON(test_bit(__E1000_RESETTING
, &adapter
->state
));
4393 e1000e_down(adapter
);
4394 e1000_free_irq(adapter
);
4396 e1000e_reset_interrupt_capability(adapter
);
4398 retval
= pci_save_state(pdev
);
4402 status
= er32(STATUS
);
4403 if (status
& E1000_STATUS_LU
)
4404 wufc
&= ~E1000_WUFC_LNKC
;
4407 e1000_setup_rctl(adapter
);
4408 e1000_set_multi(netdev
);
4410 /* turn on all-multi mode if wake on multicast is enabled */
4411 if (wufc
& E1000_WUFC_MC
) {
4413 rctl
|= E1000_RCTL_MPE
;
4418 /* advertise wake from D3Cold */
4419 #define E1000_CTRL_ADVD3WUC 0x00100000
4420 /* phy power management enable */
4421 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4422 ctrl
|= E1000_CTRL_ADVD3WUC
|
4423 E1000_CTRL_EN_PHY_PWR_MGMT
;
4426 if (adapter
->hw
.phy
.media_type
== e1000_media_type_fiber
||
4427 adapter
->hw
.phy
.media_type
==
4428 e1000_media_type_internal_serdes
) {
4429 /* keep the laser running in D3 */
4430 ctrl_ext
= er32(CTRL_EXT
);
4431 ctrl_ext
|= E1000_CTRL_EXT_SDP7_DATA
;
4432 ew32(CTRL_EXT
, ctrl_ext
);
4435 if (adapter
->flags
& FLAG_IS_ICH
)
4436 e1000e_disable_gig_wol_ich8lan(&adapter
->hw
);
4438 /* Allow time for pending master requests to run */
4439 e1000e_disable_pcie_master(&adapter
->hw
);
4441 ew32(WUC
, E1000_WUC_PME_EN
);
4443 pci_enable_wake(pdev
, PCI_D3hot
, 1);
4444 pci_enable_wake(pdev
, PCI_D3cold
, 1);
4448 pci_enable_wake(pdev
, PCI_D3hot
, 0);
4449 pci_enable_wake(pdev
, PCI_D3cold
, 0);
4452 /* make sure adapter isn't asleep if manageability is enabled */
4453 if (adapter
->flags
& FLAG_MNG_PT_ENABLED
) {
4454 pci_enable_wake(pdev
, PCI_D3hot
, 1);
4455 pci_enable_wake(pdev
, PCI_D3cold
, 1);
4458 if (adapter
->hw
.phy
.type
== e1000_phy_igp_3
)
4459 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter
->hw
);
4462 * Release control of h/w to f/w. If f/w is AMT enabled, this
4463 * would have already happened in close and is redundant.
4465 e1000_release_hw_control(adapter
);
4467 pci_disable_device(pdev
);
4470 * The pci-e switch on some quad port adapters will report a
4471 * correctable error when the MAC transitions from D0 to D3. To
4472 * prevent this we need to mask off the correctable errors on the
4473 * downstream port of the pci-e switch.
4475 if (adapter
->flags
& FLAG_IS_QUAD_PORT
) {
4476 struct pci_dev
*us_dev
= pdev
->bus
->self
;
4477 int pos
= pci_find_capability(us_dev
, PCI_CAP_ID_EXP
);
4480 pci_read_config_word(us_dev
, pos
+ PCI_EXP_DEVCTL
, &devctl
);
4481 pci_write_config_word(us_dev
, pos
+ PCI_EXP_DEVCTL
,
4482 (devctl
& ~PCI_EXP_DEVCTL_CERE
));
4484 pci_set_power_state(pdev
, pci_choose_state(pdev
, state
));
4486 pci_write_config_word(us_dev
, pos
+ PCI_EXP_DEVCTL
, devctl
);
4488 pci_set_power_state(pdev
, pci_choose_state(pdev
, state
));
4494 static void e1000e_disable_l1aspm(struct pci_dev
*pdev
)
4500 * 82573 workaround - disable L1 ASPM on mobile chipsets
4502 * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4503 * resulting in lost data or garbage information on the pci-e link
4504 * level. This could result in (false) bad EEPROM checksum errors,
4505 * long ping times (up to 2s) or even a system freeze/hang.
4507 * Unfortunately this feature saves about 1W power consumption when
4510 pos
= pci_find_capability(pdev
, PCI_CAP_ID_EXP
);
4511 pci_read_config_word(pdev
, pos
+ PCI_EXP_LNKCTL
, &val
);
4513 dev_warn(&pdev
->dev
, "Disabling L1 ASPM\n");
4515 pci_write_config_word(pdev
, pos
+ PCI_EXP_LNKCTL
, val
);
4520 static int e1000_resume(struct pci_dev
*pdev
)
4522 struct net_device
*netdev
= pci_get_drvdata(pdev
);
4523 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
4524 struct e1000_hw
*hw
= &adapter
->hw
;
4527 pci_set_power_state(pdev
, PCI_D0
);
4528 pci_restore_state(pdev
);
4529 e1000e_disable_l1aspm(pdev
);
4531 err
= pci_enable_device_mem(pdev
);
4534 "Cannot enable PCI device from suspend\n");
4538 pci_set_master(pdev
);
4540 pci_enable_wake(pdev
, PCI_D3hot
, 0);
4541 pci_enable_wake(pdev
, PCI_D3cold
, 0);
4543 e1000e_set_interrupt_capability(adapter
);
4544 if (netif_running(netdev
)) {
4545 err
= e1000_request_irq(adapter
);
4550 e1000e_power_up_phy(adapter
);
4551 e1000e_reset(adapter
);
4554 e1000_init_manageability(adapter
);
4556 if (netif_running(netdev
))
4559 netif_device_attach(netdev
);
4562 * If the controller has AMT, do not set DRV_LOAD until the interface
4563 * is up. For all other cases, let the f/w know that the h/w is now
4564 * under the control of the driver.
4566 if (!(adapter
->flags
& FLAG_HAS_AMT
))
4567 e1000_get_hw_control(adapter
);
4573 static void e1000_shutdown(struct pci_dev
*pdev
)
4575 e1000_suspend(pdev
, PMSG_SUSPEND
);
4578 #ifdef CONFIG_NET_POLL_CONTROLLER
4580 * Polling 'interrupt' - used by things like netconsole to send skbs
4581 * without having to re-enable interrupts. It's not called while
4582 * the interrupt routine is executing.
4584 static void e1000_netpoll(struct net_device
*netdev
)
4586 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
4588 disable_irq(adapter
->pdev
->irq
);
4589 e1000_intr(adapter
->pdev
->irq
, netdev
);
4591 enable_irq(adapter
->pdev
->irq
);
4596 * e1000_io_error_detected - called when PCI error is detected
4597 * @pdev: Pointer to PCI device
4598 * @state: The current pci connection state
4600 * This function is called after a PCI bus error affecting
4601 * this device has been detected.
4603 static pci_ers_result_t
e1000_io_error_detected(struct pci_dev
*pdev
,
4604 pci_channel_state_t state
)
4606 struct net_device
*netdev
= pci_get_drvdata(pdev
);
4607 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
4609 netif_device_detach(netdev
);
4611 if (netif_running(netdev
))
4612 e1000e_down(adapter
);
4613 pci_disable_device(pdev
);
4615 /* Request a slot slot reset. */
4616 return PCI_ERS_RESULT_NEED_RESET
;
4620 * e1000_io_slot_reset - called after the pci bus has been reset.
4621 * @pdev: Pointer to PCI device
4623 * Restart the card from scratch, as if from a cold-boot. Implementation
4624 * resembles the first-half of the e1000_resume routine.
4626 static pci_ers_result_t
e1000_io_slot_reset(struct pci_dev
*pdev
)
4628 struct net_device
*netdev
= pci_get_drvdata(pdev
);
4629 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
4630 struct e1000_hw
*hw
= &adapter
->hw
;
4633 e1000e_disable_l1aspm(pdev
);
4634 err
= pci_enable_device_mem(pdev
);
4637 "Cannot re-enable PCI device after reset.\n");
4638 return PCI_ERS_RESULT_DISCONNECT
;
4640 pci_set_master(pdev
);
4641 pci_restore_state(pdev
);
4643 pci_enable_wake(pdev
, PCI_D3hot
, 0);
4644 pci_enable_wake(pdev
, PCI_D3cold
, 0);
4646 e1000e_reset(adapter
);
4649 return PCI_ERS_RESULT_RECOVERED
;
4653 * e1000_io_resume - called when traffic can start flowing again.
4654 * @pdev: Pointer to PCI device
4656 * This callback is called when the error recovery driver tells us that
4657 * its OK to resume normal operation. Implementation resembles the
4658 * second-half of the e1000_resume routine.
4660 static void e1000_io_resume(struct pci_dev
*pdev
)
4662 struct net_device
*netdev
= pci_get_drvdata(pdev
);
4663 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
4665 e1000_init_manageability(adapter
);
4667 if (netif_running(netdev
)) {
4668 if (e1000e_up(adapter
)) {
4670 "can't bring device back up after reset\n");
4675 netif_device_attach(netdev
);
4678 * If the controller has AMT, do not set DRV_LOAD until the interface
4679 * is up. For all other cases, let the f/w know that the h/w is now
4680 * under the control of the driver.
4682 if (!(adapter
->flags
& FLAG_HAS_AMT
))
4683 e1000_get_hw_control(adapter
);
4687 static void e1000_print_device_info(struct e1000_adapter
*adapter
)
4689 struct e1000_hw
*hw
= &adapter
->hw
;
4690 struct net_device
*netdev
= adapter
->netdev
;
4693 /* print bus type/speed/width info */
4694 e_info("(PCI Express:2.5GB/s:%s) %pM\n",
4696 ((hw
->bus
.width
== e1000_bus_width_pcie_x4
) ? "Width x4" :
4700 e_info("Intel(R) PRO/%s Network Connection\n",
4701 (hw
->phy
.type
== e1000_phy_ife
) ? "10/100" : "1000");
4702 e1000e_read_pba_num(hw
, &pba_num
);
4703 e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4704 hw
->mac
.type
, hw
->phy
.type
, (pba_num
>> 8), (pba_num
& 0xff));
4707 static void e1000_eeprom_checks(struct e1000_adapter
*adapter
)
4709 struct e1000_hw
*hw
= &adapter
->hw
;
4713 if (hw
->mac
.type
!= e1000_82573
)
4716 ret_val
= e1000_read_nvm(hw
, NVM_INIT_CONTROL2_REG
, 1, &buf
);
4717 if (!ret_val
&& (!(le16_to_cpu(buf
) & (1 << 0)))) {
4718 /* Deep Smart Power Down (DSPD) */
4719 dev_warn(&adapter
->pdev
->dev
,
4720 "Warning: detected DSPD enabled in EEPROM\n");
4723 ret_val
= e1000_read_nvm(hw
, NVM_INIT_3GIO_3
, 1, &buf
);
4724 if (!ret_val
&& (le16_to_cpu(buf
) & (3 << 2))) {
4726 dev_warn(&adapter
->pdev
->dev
,
4727 "Warning: detected ASPM enabled in EEPROM\n");
4731 static const struct net_device_ops e1000e_netdev_ops
= {
4732 .ndo_open
= e1000_open
,
4733 .ndo_stop
= e1000_close
,
4734 .ndo_start_xmit
= e1000_xmit_frame
,
4735 .ndo_get_stats
= e1000_get_stats
,
4736 .ndo_set_multicast_list
= e1000_set_multi
,
4737 .ndo_set_mac_address
= e1000_set_mac
,
4738 .ndo_change_mtu
= e1000_change_mtu
,
4739 .ndo_do_ioctl
= e1000_ioctl
,
4740 .ndo_tx_timeout
= e1000_tx_timeout
,
4741 .ndo_validate_addr
= eth_validate_addr
,
4743 .ndo_vlan_rx_register
= e1000_vlan_rx_register
,
4744 .ndo_vlan_rx_add_vid
= e1000_vlan_rx_add_vid
,
4745 .ndo_vlan_rx_kill_vid
= e1000_vlan_rx_kill_vid
,
4746 #ifdef CONFIG_NET_POLL_CONTROLLER
4747 .ndo_poll_controller
= e1000_netpoll
,
4752 * e1000_probe - Device Initialization Routine
4753 * @pdev: PCI device information struct
4754 * @ent: entry in e1000_pci_tbl
4756 * Returns 0 on success, negative on failure
4758 * e1000_probe initializes an adapter identified by a pci_dev structure.
4759 * The OS initialization, configuring of the adapter private structure,
4760 * and a hardware reset occur.
4762 static int __devinit
e1000_probe(struct pci_dev
*pdev
,
4763 const struct pci_device_id
*ent
)
4765 struct net_device
*netdev
;
4766 struct e1000_adapter
*adapter
;
4767 struct e1000_hw
*hw
;
4768 const struct e1000_info
*ei
= e1000_info_tbl
[ent
->driver_data
];
4769 resource_size_t mmio_start
, mmio_len
;
4770 resource_size_t flash_start
, flash_len
;
4772 static int cards_found
;
4773 int i
, err
, pci_using_dac
;
4774 u16 eeprom_data
= 0;
4775 u16 eeprom_apme_mask
= E1000_EEPROM_APME
;
4777 e1000e_disable_l1aspm(pdev
);
4779 err
= pci_enable_device_mem(pdev
);
4784 err
= pci_set_dma_mask(pdev
, DMA_64BIT_MASK
);
4786 err
= pci_set_consistent_dma_mask(pdev
, DMA_64BIT_MASK
);
4790 err
= pci_set_dma_mask(pdev
, DMA_32BIT_MASK
);
4792 err
= pci_set_consistent_dma_mask(pdev
,
4795 dev_err(&pdev
->dev
, "No usable DMA "
4796 "configuration, aborting\n");
4802 err
= pci_request_selected_regions_exclusive(pdev
,
4803 pci_select_bars(pdev
, IORESOURCE_MEM
),
4804 e1000e_driver_name
);
4808 pci_set_master(pdev
);
4809 /* PCI config space info */
4810 err
= pci_save_state(pdev
);
4812 goto err_alloc_etherdev
;
4815 netdev
= alloc_etherdev(sizeof(struct e1000_adapter
));
4817 goto err_alloc_etherdev
;
4819 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
4821 pci_set_drvdata(pdev
, netdev
);
4822 adapter
= netdev_priv(netdev
);
4824 adapter
->netdev
= netdev
;
4825 adapter
->pdev
= pdev
;
4827 adapter
->pba
= ei
->pba
;
4828 adapter
->flags
= ei
->flags
;
4829 adapter
->flags2
= ei
->flags2
;
4830 adapter
->hw
.adapter
= adapter
;
4831 adapter
->hw
.mac
.type
= ei
->mac
;
4832 adapter
->msg_enable
= (1 << NETIF_MSG_DRV
| NETIF_MSG_PROBE
) - 1;
4834 mmio_start
= pci_resource_start(pdev
, 0);
4835 mmio_len
= pci_resource_len(pdev
, 0);
4838 adapter
->hw
.hw_addr
= ioremap(mmio_start
, mmio_len
);
4839 if (!adapter
->hw
.hw_addr
)
4842 if ((adapter
->flags
& FLAG_HAS_FLASH
) &&
4843 (pci_resource_flags(pdev
, 1) & IORESOURCE_MEM
)) {
4844 flash_start
= pci_resource_start(pdev
, 1);
4845 flash_len
= pci_resource_len(pdev
, 1);
4846 adapter
->hw
.flash_address
= ioremap(flash_start
, flash_len
);
4847 if (!adapter
->hw
.flash_address
)
4851 /* construct the net_device struct */
4852 netdev
->netdev_ops
= &e1000e_netdev_ops
;
4853 e1000e_set_ethtool_ops(netdev
);
4854 netdev
->watchdog_timeo
= 5 * HZ
;
4855 netif_napi_add(netdev
, &adapter
->napi
, e1000_clean
, 64);
4856 strncpy(netdev
->name
, pci_name(pdev
), sizeof(netdev
->name
) - 1);
4858 netdev
->mem_start
= mmio_start
;
4859 netdev
->mem_end
= mmio_start
+ mmio_len
;
4861 adapter
->bd_number
= cards_found
++;
4863 e1000e_check_options(adapter
);
4865 /* setup adapter struct */
4866 err
= e1000_sw_init(adapter
);
4872 memcpy(&hw
->mac
.ops
, ei
->mac_ops
, sizeof(hw
->mac
.ops
));
4873 memcpy(&hw
->nvm
.ops
, ei
->nvm_ops
, sizeof(hw
->nvm
.ops
));
4874 memcpy(&hw
->phy
.ops
, ei
->phy_ops
, sizeof(hw
->phy
.ops
));
4876 err
= ei
->get_variants(adapter
);
4880 if ((adapter
->flags
& FLAG_IS_ICH
) &&
4881 (adapter
->flags
& FLAG_READ_ONLY_NVM
))
4882 e1000e_write_protect_nvm_ich8lan(&adapter
->hw
);
4884 hw
->mac
.ops
.get_bus_info(&adapter
->hw
);
4886 adapter
->hw
.phy
.autoneg_wait_to_complete
= 0;
4888 /* Copper options */
4889 if (adapter
->hw
.phy
.media_type
== e1000_media_type_copper
) {
4890 adapter
->hw
.phy
.mdix
= AUTO_ALL_MODES
;
4891 adapter
->hw
.phy
.disable_polarity_correction
= 0;
4892 adapter
->hw
.phy
.ms_type
= e1000_ms_hw_default
;
4895 if (e1000_check_reset_block(&adapter
->hw
))
4896 e_info("PHY reset is blocked due to SOL/IDER session.\n");
4898 netdev
->features
= NETIF_F_SG
|
4900 NETIF_F_HW_VLAN_TX
|
4903 if (adapter
->flags
& FLAG_HAS_HW_VLAN_FILTER
)
4904 netdev
->features
|= NETIF_F_HW_VLAN_FILTER
;
4906 netdev
->features
|= NETIF_F_TSO
;
4907 netdev
->features
|= NETIF_F_TSO6
;
4909 netdev
->vlan_features
|= NETIF_F_TSO
;
4910 netdev
->vlan_features
|= NETIF_F_TSO6
;
4911 netdev
->vlan_features
|= NETIF_F_HW_CSUM
;
4912 netdev
->vlan_features
|= NETIF_F_SG
;
4915 netdev
->features
|= NETIF_F_HIGHDMA
;
4918 * We should not be using LLTX anymore, but we are still Tx faster with
4921 netdev
->features
|= NETIF_F_LLTX
;
4923 if (e1000e_enable_mng_pass_thru(&adapter
->hw
))
4924 adapter
->flags
|= FLAG_MNG_PT_ENABLED
;
4927 * before reading the NVM, reset the controller to
4928 * put the device in a known good starting state
4930 adapter
->hw
.mac
.ops
.reset_hw(&adapter
->hw
);
4933 * systems with ASPM and others may see the checksum fail on the first
4934 * attempt. Let's give it a few tries
4937 if (e1000_validate_nvm_checksum(&adapter
->hw
) >= 0)
4940 e_err("The NVM Checksum Is Not Valid\n");
4946 e1000_eeprom_checks(adapter
);
4948 /* copy the MAC address out of the NVM */
4949 if (e1000e_read_mac_addr(&adapter
->hw
))
4950 e_err("NVM Read Error while reading MAC address\n");
4952 memcpy(netdev
->dev_addr
, adapter
->hw
.mac
.addr
, netdev
->addr_len
);
4953 memcpy(netdev
->perm_addr
, adapter
->hw
.mac
.addr
, netdev
->addr_len
);
4955 if (!is_valid_ether_addr(netdev
->perm_addr
)) {
4956 e_err("Invalid MAC Address: %pM\n", netdev
->perm_addr
);
4961 init_timer(&adapter
->watchdog_timer
);
4962 adapter
->watchdog_timer
.function
= &e1000_watchdog
;
4963 adapter
->watchdog_timer
.data
= (unsigned long) adapter
;
4965 init_timer(&adapter
->phy_info_timer
);
4966 adapter
->phy_info_timer
.function
= &e1000_update_phy_info
;
4967 adapter
->phy_info_timer
.data
= (unsigned long) adapter
;
4969 INIT_WORK(&adapter
->reset_task
, e1000_reset_task
);
4970 INIT_WORK(&adapter
->watchdog_task
, e1000_watchdog_task
);
4971 INIT_WORK(&adapter
->downshift_task
, e1000e_downshift_workaround
);
4972 INIT_WORK(&adapter
->update_phy_task
, e1000e_update_phy_task
);
4974 /* Initialize link parameters. User can change them with ethtool */
4975 adapter
->hw
.mac
.autoneg
= 1;
4976 adapter
->fc_autoneg
= 1;
4977 adapter
->hw
.fc
.requested_mode
= e1000_fc_default
;
4978 adapter
->hw
.fc
.current_mode
= e1000_fc_default
;
4979 adapter
->hw
.phy
.autoneg_advertised
= 0x2f;
4981 /* ring size defaults */
4982 adapter
->rx_ring
->count
= 256;
4983 adapter
->tx_ring
->count
= 256;
4986 * Initial Wake on LAN setting - If APM wake is enabled in
4987 * the EEPROM, enable the ACPI Magic Packet filter
4989 if (adapter
->flags
& FLAG_APME_IN_WUC
) {
4990 /* APME bit in EEPROM is mapped to WUC.APME */
4991 eeprom_data
= er32(WUC
);
4992 eeprom_apme_mask
= E1000_WUC_APME
;
4993 } else if (adapter
->flags
& FLAG_APME_IN_CTRL3
) {
4994 if (adapter
->flags
& FLAG_APME_CHECK_PORT_B
&&
4995 (adapter
->hw
.bus
.func
== 1))
4996 e1000_read_nvm(&adapter
->hw
,
4997 NVM_INIT_CONTROL3_PORT_B
, 1, &eeprom_data
);
4999 e1000_read_nvm(&adapter
->hw
,
5000 NVM_INIT_CONTROL3_PORT_A
, 1, &eeprom_data
);
5003 /* fetch WoL from EEPROM */
5004 if (eeprom_data
& eeprom_apme_mask
)
5005 adapter
->eeprom_wol
|= E1000_WUFC_MAG
;
5008 * now that we have the eeprom settings, apply the special cases
5009 * where the eeprom may be wrong or the board simply won't support
5010 * wake on lan on a particular port
5012 if (!(adapter
->flags
& FLAG_HAS_WOL
))
5013 adapter
->eeprom_wol
= 0;
5015 /* initialize the wol settings based on the eeprom settings */
5016 adapter
->wol
= adapter
->eeprom_wol
;
5017 device_set_wakeup_enable(&adapter
->pdev
->dev
, adapter
->wol
);
5019 /* save off EEPROM version number */
5020 e1000_read_nvm(&adapter
->hw
, 5, 1, &adapter
->eeprom_vers
);
5022 /* reset the hardware with the new settings */
5023 e1000e_reset(adapter
);
5026 * If the controller has AMT, do not set DRV_LOAD until the interface
5027 * is up. For all other cases, let the f/w know that the h/w is now
5028 * under the control of the driver.
5030 if (!(adapter
->flags
& FLAG_HAS_AMT
))
5031 e1000_get_hw_control(adapter
);
5033 /* tell the stack to leave us alone until e1000_open() is called */
5034 netif_carrier_off(netdev
);
5035 netif_tx_stop_all_queues(netdev
);
5037 strcpy(netdev
->name
, "eth%d");
5038 err
= register_netdev(netdev
);
5042 e1000_print_device_info(adapter
);
5047 if (!(adapter
->flags
& FLAG_HAS_AMT
))
5048 e1000_release_hw_control(adapter
);
5050 if (!e1000_check_reset_block(&adapter
->hw
))
5051 e1000_phy_hw_reset(&adapter
->hw
);
5054 kfree(adapter
->tx_ring
);
5055 kfree(adapter
->rx_ring
);
5057 if (adapter
->hw
.flash_address
)
5058 iounmap(adapter
->hw
.flash_address
);
5059 e1000e_reset_interrupt_capability(adapter
);
5061 iounmap(adapter
->hw
.hw_addr
);
5063 free_netdev(netdev
);
5065 pci_release_selected_regions(pdev
,
5066 pci_select_bars(pdev
, IORESOURCE_MEM
));
5069 pci_disable_device(pdev
);
5074 * e1000_remove - Device Removal Routine
5075 * @pdev: PCI device information struct
5077 * e1000_remove is called by the PCI subsystem to alert the driver
5078 * that it should release a PCI device. The could be caused by a
5079 * Hot-Plug event, or because the driver is going to be removed from
5082 static void __devexit
e1000_remove(struct pci_dev
*pdev
)
5084 struct net_device
*netdev
= pci_get_drvdata(pdev
);
5085 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
5088 * flush_scheduled work may reschedule our watchdog task, so
5089 * explicitly disable watchdog tasks from being rescheduled
5091 set_bit(__E1000_DOWN
, &adapter
->state
);
5092 del_timer_sync(&adapter
->watchdog_timer
);
5093 del_timer_sync(&adapter
->phy_info_timer
);
5095 flush_scheduled_work();
5098 * Release control of h/w to f/w. If f/w is AMT enabled, this
5099 * would have already happened in close and is redundant.
5101 e1000_release_hw_control(adapter
);
5103 unregister_netdev(netdev
);
5105 if (!e1000_check_reset_block(&adapter
->hw
))
5106 e1000_phy_hw_reset(&adapter
->hw
);
5108 e1000e_reset_interrupt_capability(adapter
);
5109 kfree(adapter
->tx_ring
);
5110 kfree(adapter
->rx_ring
);
5112 iounmap(adapter
->hw
.hw_addr
);
5113 if (adapter
->hw
.flash_address
)
5114 iounmap(adapter
->hw
.flash_address
);
5115 pci_release_selected_regions(pdev
,
5116 pci_select_bars(pdev
, IORESOURCE_MEM
));
5118 free_netdev(netdev
);
5120 pci_disable_device(pdev
);
5123 /* PCI Error Recovery (ERS) */
5124 static struct pci_error_handlers e1000_err_handler
= {
5125 .error_detected
= e1000_io_error_detected
,
5126 .slot_reset
= e1000_io_slot_reset
,
5127 .resume
= e1000_io_resume
,
5130 static struct pci_device_id e1000_pci_tbl
[] = {
5131 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82571EB_COPPER
), board_82571
},
5132 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82571EB_FIBER
), board_82571
},
5133 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82571EB_QUAD_COPPER
), board_82571
},
5134 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82571EB_QUAD_COPPER_LP
), board_82571
},
5135 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82571EB_QUAD_FIBER
), board_82571
},
5136 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82571EB_SERDES
), board_82571
},
5137 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82571EB_SERDES_DUAL
), board_82571
},
5138 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82571EB_SERDES_QUAD
), board_82571
},
5139 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82571PT_QUAD_COPPER
), board_82571
},
5141 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82572EI
), board_82572
},
5142 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82572EI_COPPER
), board_82572
},
5143 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82572EI_FIBER
), board_82572
},
5144 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82572EI_SERDES
), board_82572
},
5146 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82573E
), board_82573
},
5147 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82573E_IAMT
), board_82573
},
5148 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82573L
), board_82573
},
5150 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82574L
), board_82574
},
5152 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_80003ES2LAN_COPPER_DPT
),
5153 board_80003es2lan
},
5154 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_80003ES2LAN_COPPER_SPT
),
5155 board_80003es2lan
},
5156 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_80003ES2LAN_SERDES_DPT
),
5157 board_80003es2lan
},
5158 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_80003ES2LAN_SERDES_SPT
),
5159 board_80003es2lan
},
5161 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH8_IFE
), board_ich8lan
},
5162 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH8_IFE_G
), board_ich8lan
},
5163 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH8_IFE_GT
), board_ich8lan
},
5164 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH8_IGP_AMT
), board_ich8lan
},
5165 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH8_IGP_C
), board_ich8lan
},
5166 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH8_IGP_M
), board_ich8lan
},
5167 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH8_IGP_M_AMT
), board_ich8lan
},
5169 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH9_IFE
), board_ich9lan
},
5170 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH9_IFE_G
), board_ich9lan
},
5171 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH9_IFE_GT
), board_ich9lan
},
5172 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH9_IGP_AMT
), board_ich9lan
},
5173 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH9_IGP_C
), board_ich9lan
},
5174 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH9_BM
), board_ich9lan
},
5175 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH9_IGP_M
), board_ich9lan
},
5176 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH9_IGP_M_AMT
), board_ich9lan
},
5177 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH9_IGP_M_V
), board_ich9lan
},
5179 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH10_R_BM_LM
), board_ich9lan
},
5180 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH10_R_BM_LF
), board_ich9lan
},
5181 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH10_R_BM_V
), board_ich9lan
},
5183 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH10_D_BM_LM
), board_ich10lan
},
5184 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH10_D_BM_LF
), board_ich10lan
},
5186 { } /* terminate list */
5188 MODULE_DEVICE_TABLE(pci
, e1000_pci_tbl
);
5190 /* PCI Device API Driver */
5191 static struct pci_driver e1000_driver
= {
5192 .name
= e1000e_driver_name
,
5193 .id_table
= e1000_pci_tbl
,
5194 .probe
= e1000_probe
,
5195 .remove
= __devexit_p(e1000_remove
),
5197 /* Power Management Hooks */
5198 .suspend
= e1000_suspend
,
5199 .resume
= e1000_resume
,
5201 .shutdown
= e1000_shutdown
,
5202 .err_handler
= &e1000_err_handler
5206 * e1000_init_module - Driver Registration Routine
5208 * e1000_init_module is the first routine called when the driver is
5209 * loaded. All it does is register with the PCI subsystem.
5211 static int __init
e1000_init_module(void)
5214 printk(KERN_INFO
"%s: Intel(R) PRO/1000 Network Driver - %s\n",
5215 e1000e_driver_name
, e1000e_driver_version
);
5216 printk(KERN_INFO
"%s: Copyright (c) 1999-2008 Intel Corporation.\n",
5217 e1000e_driver_name
);
5218 ret
= pci_register_driver(&e1000_driver
);
5219 pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY
, e1000e_driver_name
,
5220 PM_QOS_DEFAULT_VALUE
);
5224 module_init(e1000_init_module
);
5227 * e1000_exit_module - Driver Exit Cleanup Routine
5229 * e1000_exit_module is called just before the driver is removed
5232 static void __exit
e1000_exit_module(void)
5234 pci_unregister_driver(&e1000_driver
);
5235 pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY
, e1000e_driver_name
);
5237 module_exit(e1000_exit_module
);
5240 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5241 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5242 MODULE_LICENSE("GPL");
5243 MODULE_VERSION(DRV_VERSION
);