Merge branch 'firewire-kernel-streaming' of git://git.alsa-project.org/alsa-kprivate
[firewire-audio.git] / drivers / net / e1000e / netdev.c
blob2e5022849f1828c19a1bc968e7112cdf487d4145
1 /*******************************************************************************
3 Intel PRO/1000 Linux driver
4 Copyright(c) 1999 - 2011 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
13 more details.
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".
22 Contact Information:
23 Linux NICS <linux.nics@intel.com>
24 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
27 *******************************************************************************/
29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
31 #include <linux/module.h>
32 #include <linux/types.h>
33 #include <linux/init.h>
34 #include <linux/pci.h>
35 #include <linux/vmalloc.h>
36 #include <linux/pagemap.h>
37 #include <linux/delay.h>
38 #include <linux/netdevice.h>
39 #include <linux/tcp.h>
40 #include <linux/ipv6.h>
41 #include <linux/slab.h>
42 #include <net/checksum.h>
43 #include <net/ip6_checksum.h>
44 #include <linux/mii.h>
45 #include <linux/ethtool.h>
46 #include <linux/if_vlan.h>
47 #include <linux/cpu.h>
48 #include <linux/smp.h>
49 #include <linux/pm_qos_params.h>
50 #include <linux/pm_runtime.h>
51 #include <linux/aer.h>
53 #include "e1000.h"
55 #define DRV_EXTRAVERSION "-k2"
57 #define DRV_VERSION "1.2.20" DRV_EXTRAVERSION
58 char e1000e_driver_name[] = "e1000e";
59 const char e1000e_driver_version[] = DRV_VERSION;
61 static const struct e1000_info *e1000_info_tbl[] = {
62 [board_82571] = &e1000_82571_info,
63 [board_82572] = &e1000_82572_info,
64 [board_82573] = &e1000_82573_info,
65 [board_82574] = &e1000_82574_info,
66 [board_82583] = &e1000_82583_info,
67 [board_80003es2lan] = &e1000_es2_info,
68 [board_ich8lan] = &e1000_ich8_info,
69 [board_ich9lan] = &e1000_ich9_info,
70 [board_ich10lan] = &e1000_ich10_info,
71 [board_pchlan] = &e1000_pch_info,
72 [board_pch2lan] = &e1000_pch2_info,
75 struct e1000_reg_info {
76 u32 ofs;
77 char *name;
80 #define E1000_RDFH 0x02410 /* Rx Data FIFO Head - RW */
81 #define E1000_RDFT 0x02418 /* Rx Data FIFO Tail - RW */
82 #define E1000_RDFHS 0x02420 /* Rx Data FIFO Head Saved - RW */
83 #define E1000_RDFTS 0x02428 /* Rx Data FIFO Tail Saved - RW */
84 #define E1000_RDFPC 0x02430 /* Rx Data FIFO Packet Count - RW */
86 #define E1000_TDFH 0x03410 /* Tx Data FIFO Head - RW */
87 #define E1000_TDFT 0x03418 /* Tx Data FIFO Tail - RW */
88 #define E1000_TDFHS 0x03420 /* Tx Data FIFO Head Saved - RW */
89 #define E1000_TDFTS 0x03428 /* Tx Data FIFO Tail Saved - RW */
90 #define E1000_TDFPC 0x03430 /* Tx Data FIFO Packet Count - RW */
92 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
94 /* General Registers */
95 {E1000_CTRL, "CTRL"},
96 {E1000_STATUS, "STATUS"},
97 {E1000_CTRL_EXT, "CTRL_EXT"},
99 /* Interrupt Registers */
100 {E1000_ICR, "ICR"},
102 /* Rx Registers */
103 {E1000_RCTL, "RCTL"},
104 {E1000_RDLEN, "RDLEN"},
105 {E1000_RDH, "RDH"},
106 {E1000_RDT, "RDT"},
107 {E1000_RDTR, "RDTR"},
108 {E1000_RXDCTL(0), "RXDCTL"},
109 {E1000_ERT, "ERT"},
110 {E1000_RDBAL, "RDBAL"},
111 {E1000_RDBAH, "RDBAH"},
112 {E1000_RDFH, "RDFH"},
113 {E1000_RDFT, "RDFT"},
114 {E1000_RDFHS, "RDFHS"},
115 {E1000_RDFTS, "RDFTS"},
116 {E1000_RDFPC, "RDFPC"},
118 /* Tx Registers */
119 {E1000_TCTL, "TCTL"},
120 {E1000_TDBAL, "TDBAL"},
121 {E1000_TDBAH, "TDBAH"},
122 {E1000_TDLEN, "TDLEN"},
123 {E1000_TDH, "TDH"},
124 {E1000_TDT, "TDT"},
125 {E1000_TIDV, "TIDV"},
126 {E1000_TXDCTL(0), "TXDCTL"},
127 {E1000_TADV, "TADV"},
128 {E1000_TARC(0), "TARC"},
129 {E1000_TDFH, "TDFH"},
130 {E1000_TDFT, "TDFT"},
131 {E1000_TDFHS, "TDFHS"},
132 {E1000_TDFTS, "TDFTS"},
133 {E1000_TDFPC, "TDFPC"},
135 /* List Terminator */
140 * e1000_regdump - register printout routine
142 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
144 int n = 0;
145 char rname[16];
146 u32 regs[8];
148 switch (reginfo->ofs) {
149 case E1000_RXDCTL(0):
150 for (n = 0; n < 2; n++)
151 regs[n] = __er32(hw, E1000_RXDCTL(n));
152 break;
153 case E1000_TXDCTL(0):
154 for (n = 0; n < 2; n++)
155 regs[n] = __er32(hw, E1000_TXDCTL(n));
156 break;
157 case E1000_TARC(0):
158 for (n = 0; n < 2; n++)
159 regs[n] = __er32(hw, E1000_TARC(n));
160 break;
161 default:
162 printk(KERN_INFO "%-15s %08x\n",
163 reginfo->name, __er32(hw, reginfo->ofs));
164 return;
167 snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
168 printk(KERN_INFO "%-15s ", rname);
169 for (n = 0; n < 2; n++)
170 printk(KERN_CONT "%08x ", regs[n]);
171 printk(KERN_CONT "\n");
175 * e1000e_dump - Print registers, Tx-ring and Rx-ring
177 static void e1000e_dump(struct e1000_adapter *adapter)
179 struct net_device *netdev = adapter->netdev;
180 struct e1000_hw *hw = &adapter->hw;
181 struct e1000_reg_info *reginfo;
182 struct e1000_ring *tx_ring = adapter->tx_ring;
183 struct e1000_tx_desc *tx_desc;
184 struct my_u0 {
185 u64 a;
186 u64 b;
187 } *u0;
188 struct e1000_buffer *buffer_info;
189 struct e1000_ring *rx_ring = adapter->rx_ring;
190 union e1000_rx_desc_packet_split *rx_desc_ps;
191 struct e1000_rx_desc *rx_desc;
192 struct my_u1 {
193 u64 a;
194 u64 b;
195 u64 c;
196 u64 d;
197 } *u1;
198 u32 staterr;
199 int i = 0;
201 if (!netif_msg_hw(adapter))
202 return;
204 /* Print netdevice Info */
205 if (netdev) {
206 dev_info(&adapter->pdev->dev, "Net device Info\n");
207 printk(KERN_INFO "Device Name state "
208 "trans_start last_rx\n");
209 printk(KERN_INFO "%-15s %016lX %016lX %016lX\n",
210 netdev->name, netdev->state, netdev->trans_start,
211 netdev->last_rx);
214 /* Print Registers */
215 dev_info(&adapter->pdev->dev, "Register Dump\n");
216 printk(KERN_INFO " Register Name Value\n");
217 for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
218 reginfo->name; reginfo++) {
219 e1000_regdump(hw, reginfo);
222 /* Print Tx Ring Summary */
223 if (!netdev || !netif_running(netdev))
224 goto exit;
226 dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
227 printk(KERN_INFO "Queue [NTU] [NTC] [bi(ntc)->dma ]"
228 " leng ntw timestamp\n");
229 buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
230 printk(KERN_INFO " %5d %5X %5X %016llX %04X %3X %016llX\n",
231 0, tx_ring->next_to_use, tx_ring->next_to_clean,
232 (unsigned long long)buffer_info->dma,
233 buffer_info->length,
234 buffer_info->next_to_watch,
235 (unsigned long long)buffer_info->time_stamp);
237 /* Print Tx Ring */
238 if (!netif_msg_tx_done(adapter))
239 goto rx_ring_summary;
241 dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
243 /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
245 * Legacy Transmit Descriptor
246 * +--------------------------------------------------------------+
247 * 0 | Buffer Address [63:0] (Reserved on Write Back) |
248 * +--------------------------------------------------------------+
249 * 8 | Special | CSS | Status | CMD | CSO | Length |
250 * +--------------------------------------------------------------+
251 * 63 48 47 36 35 32 31 24 23 16 15 0
253 * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
254 * 63 48 47 40 39 32 31 16 15 8 7 0
255 * +----------------------------------------------------------------+
256 * 0 | TUCSE | TUCS0 | TUCSS | IPCSE | IPCS0 | IPCSS |
257 * +----------------------------------------------------------------+
258 * 8 | MSS | HDRLEN | RSV | STA | TUCMD | DTYP | PAYLEN |
259 * +----------------------------------------------------------------+
260 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
262 * Extended Data Descriptor (DTYP=0x1)
263 * +----------------------------------------------------------------+
264 * 0 | Buffer Address [63:0] |
265 * +----------------------------------------------------------------+
266 * 8 | VLAN tag | POPTS | Rsvd | Status | Command | DTYP | DTALEN |
267 * +----------------------------------------------------------------+
268 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
270 printk(KERN_INFO "Tl[desc] [address 63:0 ] [SpeCssSCmCsLen]"
271 " [bi->dma ] leng ntw timestamp bi->skb "
272 "<-- Legacy format\n");
273 printk(KERN_INFO "Tc[desc] [Ce CoCsIpceCoS] [MssHlRSCm0Plen]"
274 " [bi->dma ] leng ntw timestamp bi->skb "
275 "<-- Ext Context format\n");
276 printk(KERN_INFO "Td[desc] [address 63:0 ] [VlaPoRSCm1Dlen]"
277 " [bi->dma ] leng ntw timestamp bi->skb "
278 "<-- Ext Data format\n");
279 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
280 tx_desc = E1000_TX_DESC(*tx_ring, i);
281 buffer_info = &tx_ring->buffer_info[i];
282 u0 = (struct my_u0 *)tx_desc;
283 printk(KERN_INFO "T%c[0x%03X] %016llX %016llX %016llX "
284 "%04X %3X %016llX %p",
285 (!(le64_to_cpu(u0->b) & (1 << 29)) ? 'l' :
286 ((le64_to_cpu(u0->b) & (1 << 20)) ? 'd' : 'c')), i,
287 (unsigned long long)le64_to_cpu(u0->a),
288 (unsigned long long)le64_to_cpu(u0->b),
289 (unsigned long long)buffer_info->dma,
290 buffer_info->length, buffer_info->next_to_watch,
291 (unsigned long long)buffer_info->time_stamp,
292 buffer_info->skb);
293 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
294 printk(KERN_CONT " NTC/U\n");
295 else if (i == tx_ring->next_to_use)
296 printk(KERN_CONT " NTU\n");
297 else if (i == tx_ring->next_to_clean)
298 printk(KERN_CONT " NTC\n");
299 else
300 printk(KERN_CONT "\n");
302 if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
303 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
304 16, 1, phys_to_virt(buffer_info->dma),
305 buffer_info->length, true);
308 /* Print Rx Ring Summary */
309 rx_ring_summary:
310 dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
311 printk(KERN_INFO "Queue [NTU] [NTC]\n");
312 printk(KERN_INFO " %5d %5X %5X\n", 0,
313 rx_ring->next_to_use, rx_ring->next_to_clean);
315 /* Print Rx Ring */
316 if (!netif_msg_rx_status(adapter))
317 goto exit;
319 dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
320 switch (adapter->rx_ps_pages) {
321 case 1:
322 case 2:
323 case 3:
324 /* [Extended] Packet Split Receive Descriptor Format
326 * +-----------------------------------------------------+
327 * 0 | Buffer Address 0 [63:0] |
328 * +-----------------------------------------------------+
329 * 8 | Buffer Address 1 [63:0] |
330 * +-----------------------------------------------------+
331 * 16 | Buffer Address 2 [63:0] |
332 * +-----------------------------------------------------+
333 * 24 | Buffer Address 3 [63:0] |
334 * +-----------------------------------------------------+
336 printk(KERN_INFO "R [desc] [buffer 0 63:0 ] "
337 "[buffer 1 63:0 ] "
338 "[buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma ] "
339 "[bi->skb] <-- Ext Pkt Split format\n");
340 /* [Extended] Receive Descriptor (Write-Back) Format
342 * 63 48 47 32 31 13 12 8 7 4 3 0
343 * +------------------------------------------------------+
344 * 0 | Packet | IP | Rsvd | MRQ | Rsvd | MRQ RSS |
345 * | Checksum | Ident | | Queue | | Type |
346 * +------------------------------------------------------+
347 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
348 * +------------------------------------------------------+
349 * 63 48 47 32 31 20 19 0
351 printk(KERN_INFO "RWB[desc] [ck ipid mrqhsh] "
352 "[vl l0 ee es] "
353 "[ l3 l2 l1 hs] [reserved ] ---------------- "
354 "[bi->skb] <-- Ext Rx Write-Back format\n");
355 for (i = 0; i < rx_ring->count; i++) {
356 buffer_info = &rx_ring->buffer_info[i];
357 rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
358 u1 = (struct my_u1 *)rx_desc_ps;
359 staterr =
360 le32_to_cpu(rx_desc_ps->wb.middle.status_error);
361 if (staterr & E1000_RXD_STAT_DD) {
362 /* Descriptor Done */
363 printk(KERN_INFO "RWB[0x%03X] %016llX "
364 "%016llX %016llX %016llX "
365 "---------------- %p", i,
366 (unsigned long long)le64_to_cpu(u1->a),
367 (unsigned long long)le64_to_cpu(u1->b),
368 (unsigned long long)le64_to_cpu(u1->c),
369 (unsigned long long)le64_to_cpu(u1->d),
370 buffer_info->skb);
371 } else {
372 printk(KERN_INFO "R [0x%03X] %016llX "
373 "%016llX %016llX %016llX %016llX %p", i,
374 (unsigned long long)le64_to_cpu(u1->a),
375 (unsigned long long)le64_to_cpu(u1->b),
376 (unsigned long long)le64_to_cpu(u1->c),
377 (unsigned long long)le64_to_cpu(u1->d),
378 (unsigned long long)buffer_info->dma,
379 buffer_info->skb);
381 if (netif_msg_pktdata(adapter))
382 print_hex_dump(KERN_INFO, "",
383 DUMP_PREFIX_ADDRESS, 16, 1,
384 phys_to_virt(buffer_info->dma),
385 adapter->rx_ps_bsize0, true);
388 if (i == rx_ring->next_to_use)
389 printk(KERN_CONT " NTU\n");
390 else if (i == rx_ring->next_to_clean)
391 printk(KERN_CONT " NTC\n");
392 else
393 printk(KERN_CONT "\n");
395 break;
396 default:
397 case 0:
398 /* Legacy Receive Descriptor Format
400 * +-----------------------------------------------------+
401 * | Buffer Address [63:0] |
402 * +-----------------------------------------------------+
403 * | VLAN Tag | Errors | Status 0 | Packet csum | Length |
404 * +-----------------------------------------------------+
405 * 63 48 47 40 39 32 31 16 15 0
407 printk(KERN_INFO "Rl[desc] [address 63:0 ] "
408 "[vl er S cks ln] [bi->dma ] [bi->skb] "
409 "<-- Legacy format\n");
410 for (i = 0; rx_ring->desc && (i < rx_ring->count); i++) {
411 rx_desc = E1000_RX_DESC(*rx_ring, i);
412 buffer_info = &rx_ring->buffer_info[i];
413 u0 = (struct my_u0 *)rx_desc;
414 printk(KERN_INFO "Rl[0x%03X] %016llX %016llX "
415 "%016llX %p", i,
416 (unsigned long long)le64_to_cpu(u0->a),
417 (unsigned long long)le64_to_cpu(u0->b),
418 (unsigned long long)buffer_info->dma,
419 buffer_info->skb);
420 if (i == rx_ring->next_to_use)
421 printk(KERN_CONT " NTU\n");
422 else if (i == rx_ring->next_to_clean)
423 printk(KERN_CONT " NTC\n");
424 else
425 printk(KERN_CONT "\n");
427 if (netif_msg_pktdata(adapter))
428 print_hex_dump(KERN_INFO, "",
429 DUMP_PREFIX_ADDRESS,
430 16, 1,
431 phys_to_virt(buffer_info->dma),
432 adapter->rx_buffer_len, true);
436 exit:
437 return;
441 * e1000_desc_unused - calculate if we have unused descriptors
443 static int e1000_desc_unused(struct e1000_ring *ring)
445 if (ring->next_to_clean > ring->next_to_use)
446 return ring->next_to_clean - ring->next_to_use - 1;
448 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
452 * e1000_receive_skb - helper function to handle Rx indications
453 * @adapter: board private structure
454 * @status: descriptor status field as written by hardware
455 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
456 * @skb: pointer to sk_buff to be indicated to stack
458 static void e1000_receive_skb(struct e1000_adapter *adapter,
459 struct net_device *netdev, struct sk_buff *skb,
460 u8 status, __le16 vlan)
462 skb->protocol = eth_type_trans(skb, netdev);
464 if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
465 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
466 le16_to_cpu(vlan), skb);
467 else
468 napi_gro_receive(&adapter->napi, skb);
472 * e1000_rx_checksum - Receive Checksum Offload
473 * @adapter: board private structure
474 * @status_err: receive descriptor status and error fields
475 * @csum: receive descriptor csum field
476 * @sk_buff: socket buffer with received data
478 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
479 u32 csum, struct sk_buff *skb)
481 u16 status = (u16)status_err;
482 u8 errors = (u8)(status_err >> 24);
484 skb_checksum_none_assert(skb);
486 /* Ignore Checksum bit is set */
487 if (status & E1000_RXD_STAT_IXSM)
488 return;
489 /* TCP/UDP checksum error bit is set */
490 if (errors & E1000_RXD_ERR_TCPE) {
491 /* let the stack verify checksum errors */
492 adapter->hw_csum_err++;
493 return;
496 /* TCP/UDP Checksum has not been calculated */
497 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
498 return;
500 /* It must be a TCP or UDP packet with a valid checksum */
501 if (status & E1000_RXD_STAT_TCPCS) {
502 /* TCP checksum is good */
503 skb->ip_summed = CHECKSUM_UNNECESSARY;
504 } else {
506 * IP fragment with UDP payload
507 * Hardware complements the payload checksum, so we undo it
508 * and then put the value in host order for further stack use.
510 __sum16 sum = (__force __sum16)htons(csum);
511 skb->csum = csum_unfold(~sum);
512 skb->ip_summed = CHECKSUM_COMPLETE;
514 adapter->hw_csum_good++;
518 * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
519 * @adapter: address of board private structure
521 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
522 int cleaned_count)
524 struct net_device *netdev = adapter->netdev;
525 struct pci_dev *pdev = adapter->pdev;
526 struct e1000_ring *rx_ring = adapter->rx_ring;
527 struct e1000_rx_desc *rx_desc;
528 struct e1000_buffer *buffer_info;
529 struct sk_buff *skb;
530 unsigned int i;
531 unsigned int bufsz = adapter->rx_buffer_len;
533 i = rx_ring->next_to_use;
534 buffer_info = &rx_ring->buffer_info[i];
536 while (cleaned_count--) {
537 skb = buffer_info->skb;
538 if (skb) {
539 skb_trim(skb, 0);
540 goto map_skb;
543 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
544 if (!skb) {
545 /* Better luck next round */
546 adapter->alloc_rx_buff_failed++;
547 break;
550 buffer_info->skb = skb;
551 map_skb:
552 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
553 adapter->rx_buffer_len,
554 DMA_FROM_DEVICE);
555 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
556 dev_err(&pdev->dev, "Rx DMA map failed\n");
557 adapter->rx_dma_failed++;
558 break;
561 rx_desc = E1000_RX_DESC(*rx_ring, i);
562 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
564 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
566 * Force memory writes to complete before letting h/w
567 * know there are new descriptors to fetch. (Only
568 * applicable for weak-ordered memory model archs,
569 * such as IA-64).
571 wmb();
572 writel(i, adapter->hw.hw_addr + rx_ring->tail);
574 i++;
575 if (i == rx_ring->count)
576 i = 0;
577 buffer_info = &rx_ring->buffer_info[i];
580 rx_ring->next_to_use = i;
584 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
585 * @adapter: address of board private structure
587 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
588 int cleaned_count)
590 struct net_device *netdev = adapter->netdev;
591 struct pci_dev *pdev = adapter->pdev;
592 union e1000_rx_desc_packet_split *rx_desc;
593 struct e1000_ring *rx_ring = adapter->rx_ring;
594 struct e1000_buffer *buffer_info;
595 struct e1000_ps_page *ps_page;
596 struct sk_buff *skb;
597 unsigned int i, j;
599 i = rx_ring->next_to_use;
600 buffer_info = &rx_ring->buffer_info[i];
602 while (cleaned_count--) {
603 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
605 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
606 ps_page = &buffer_info->ps_pages[j];
607 if (j >= adapter->rx_ps_pages) {
608 /* all unused desc entries get hw null ptr */
609 rx_desc->read.buffer_addr[j + 1] =
610 ~cpu_to_le64(0);
611 continue;
613 if (!ps_page->page) {
614 ps_page->page = alloc_page(GFP_ATOMIC);
615 if (!ps_page->page) {
616 adapter->alloc_rx_buff_failed++;
617 goto no_buffers;
619 ps_page->dma = dma_map_page(&pdev->dev,
620 ps_page->page,
621 0, PAGE_SIZE,
622 DMA_FROM_DEVICE);
623 if (dma_mapping_error(&pdev->dev,
624 ps_page->dma)) {
625 dev_err(&adapter->pdev->dev,
626 "Rx DMA page map failed\n");
627 adapter->rx_dma_failed++;
628 goto no_buffers;
632 * Refresh the desc even if buffer_addrs
633 * didn't change because each write-back
634 * erases this info.
636 rx_desc->read.buffer_addr[j + 1] =
637 cpu_to_le64(ps_page->dma);
640 skb = netdev_alloc_skb_ip_align(netdev,
641 adapter->rx_ps_bsize0);
643 if (!skb) {
644 adapter->alloc_rx_buff_failed++;
645 break;
648 buffer_info->skb = skb;
649 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
650 adapter->rx_ps_bsize0,
651 DMA_FROM_DEVICE);
652 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
653 dev_err(&pdev->dev, "Rx DMA map failed\n");
654 adapter->rx_dma_failed++;
655 /* cleanup skb */
656 dev_kfree_skb_any(skb);
657 buffer_info->skb = NULL;
658 break;
661 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
663 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
665 * Force memory writes to complete before letting h/w
666 * know there are new descriptors to fetch. (Only
667 * applicable for weak-ordered memory model archs,
668 * such as IA-64).
670 wmb();
671 writel(i << 1, adapter->hw.hw_addr + rx_ring->tail);
674 i++;
675 if (i == rx_ring->count)
676 i = 0;
677 buffer_info = &rx_ring->buffer_info[i];
680 no_buffers:
681 rx_ring->next_to_use = i;
685 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
686 * @adapter: address of board private structure
687 * @cleaned_count: number of buffers to allocate this pass
690 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
691 int cleaned_count)
693 struct net_device *netdev = adapter->netdev;
694 struct pci_dev *pdev = adapter->pdev;
695 struct e1000_rx_desc *rx_desc;
696 struct e1000_ring *rx_ring = adapter->rx_ring;
697 struct e1000_buffer *buffer_info;
698 struct sk_buff *skb;
699 unsigned int i;
700 unsigned int bufsz = 256 - 16 /* for skb_reserve */;
702 i = rx_ring->next_to_use;
703 buffer_info = &rx_ring->buffer_info[i];
705 while (cleaned_count--) {
706 skb = buffer_info->skb;
707 if (skb) {
708 skb_trim(skb, 0);
709 goto check_page;
712 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
713 if (unlikely(!skb)) {
714 /* Better luck next round */
715 adapter->alloc_rx_buff_failed++;
716 break;
719 buffer_info->skb = skb;
720 check_page:
721 /* allocate a new page if necessary */
722 if (!buffer_info->page) {
723 buffer_info->page = alloc_page(GFP_ATOMIC);
724 if (unlikely(!buffer_info->page)) {
725 adapter->alloc_rx_buff_failed++;
726 break;
730 if (!buffer_info->dma)
731 buffer_info->dma = dma_map_page(&pdev->dev,
732 buffer_info->page, 0,
733 PAGE_SIZE,
734 DMA_FROM_DEVICE);
736 rx_desc = E1000_RX_DESC(*rx_ring, i);
737 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
739 if (unlikely(++i == rx_ring->count))
740 i = 0;
741 buffer_info = &rx_ring->buffer_info[i];
744 if (likely(rx_ring->next_to_use != i)) {
745 rx_ring->next_to_use = i;
746 if (unlikely(i-- == 0))
747 i = (rx_ring->count - 1);
749 /* Force memory writes to complete before letting h/w
750 * know there are new descriptors to fetch. (Only
751 * applicable for weak-ordered memory model archs,
752 * such as IA-64). */
753 wmb();
754 writel(i, adapter->hw.hw_addr + rx_ring->tail);
759 * e1000_clean_rx_irq - Send received data up the network stack; legacy
760 * @adapter: board private structure
762 * the return value indicates whether actual cleaning was done, there
763 * is no guarantee that everything was cleaned
765 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
766 int *work_done, int work_to_do)
768 struct net_device *netdev = adapter->netdev;
769 struct pci_dev *pdev = adapter->pdev;
770 struct e1000_hw *hw = &adapter->hw;
771 struct e1000_ring *rx_ring = adapter->rx_ring;
772 struct e1000_rx_desc *rx_desc, *next_rxd;
773 struct e1000_buffer *buffer_info, *next_buffer;
774 u32 length;
775 unsigned int i;
776 int cleaned_count = 0;
777 bool cleaned = 0;
778 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
780 i = rx_ring->next_to_clean;
781 rx_desc = E1000_RX_DESC(*rx_ring, i);
782 buffer_info = &rx_ring->buffer_info[i];
784 while (rx_desc->status & E1000_RXD_STAT_DD) {
785 struct sk_buff *skb;
786 u8 status;
788 if (*work_done >= work_to_do)
789 break;
790 (*work_done)++;
791 rmb(); /* read descriptor and rx_buffer_info after status DD */
793 status = rx_desc->status;
794 skb = buffer_info->skb;
795 buffer_info->skb = NULL;
797 prefetch(skb->data - NET_IP_ALIGN);
799 i++;
800 if (i == rx_ring->count)
801 i = 0;
802 next_rxd = E1000_RX_DESC(*rx_ring, i);
803 prefetch(next_rxd);
805 next_buffer = &rx_ring->buffer_info[i];
807 cleaned = 1;
808 cleaned_count++;
809 dma_unmap_single(&pdev->dev,
810 buffer_info->dma,
811 adapter->rx_buffer_len,
812 DMA_FROM_DEVICE);
813 buffer_info->dma = 0;
815 length = le16_to_cpu(rx_desc->length);
818 * !EOP means multiple descriptors were used to store a single
819 * packet, if that's the case we need to toss it. In fact, we
820 * need to toss every packet with the EOP bit clear and the
821 * next frame that _does_ have the EOP bit set, as it is by
822 * definition only a frame fragment
824 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
825 adapter->flags2 |= FLAG2_IS_DISCARDING;
827 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
828 /* All receives must fit into a single buffer */
829 e_dbg("Receive packet consumed multiple buffers\n");
830 /* recycle */
831 buffer_info->skb = skb;
832 if (status & E1000_RXD_STAT_EOP)
833 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
834 goto next_desc;
837 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
838 /* recycle */
839 buffer_info->skb = skb;
840 goto next_desc;
843 /* adjust length to remove Ethernet CRC */
844 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
845 length -= 4;
847 total_rx_bytes += length;
848 total_rx_packets++;
851 * code added for copybreak, this should improve
852 * performance for small packets with large amounts
853 * of reassembly being done in the stack
855 if (length < copybreak) {
856 struct sk_buff *new_skb =
857 netdev_alloc_skb_ip_align(netdev, length);
858 if (new_skb) {
859 skb_copy_to_linear_data_offset(new_skb,
860 -NET_IP_ALIGN,
861 (skb->data -
862 NET_IP_ALIGN),
863 (length +
864 NET_IP_ALIGN));
865 /* save the skb in buffer_info as good */
866 buffer_info->skb = skb;
867 skb = new_skb;
869 /* else just continue with the old one */
871 /* end copybreak code */
872 skb_put(skb, length);
874 /* Receive Checksum Offload */
875 e1000_rx_checksum(adapter,
876 (u32)(status) |
877 ((u32)(rx_desc->errors) << 24),
878 le16_to_cpu(rx_desc->csum), skb);
880 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
882 next_desc:
883 rx_desc->status = 0;
885 /* return some buffers to hardware, one at a time is too slow */
886 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
887 adapter->alloc_rx_buf(adapter, cleaned_count);
888 cleaned_count = 0;
891 /* use prefetched values */
892 rx_desc = next_rxd;
893 buffer_info = next_buffer;
895 rx_ring->next_to_clean = i;
897 cleaned_count = e1000_desc_unused(rx_ring);
898 if (cleaned_count)
899 adapter->alloc_rx_buf(adapter, cleaned_count);
901 adapter->total_rx_bytes += total_rx_bytes;
902 adapter->total_rx_packets += total_rx_packets;
903 netdev->stats.rx_bytes += total_rx_bytes;
904 netdev->stats.rx_packets += total_rx_packets;
905 return cleaned;
908 static void e1000_put_txbuf(struct e1000_adapter *adapter,
909 struct e1000_buffer *buffer_info)
911 if (buffer_info->dma) {
912 if (buffer_info->mapped_as_page)
913 dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
914 buffer_info->length, DMA_TO_DEVICE);
915 else
916 dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
917 buffer_info->length, DMA_TO_DEVICE);
918 buffer_info->dma = 0;
920 if (buffer_info->skb) {
921 dev_kfree_skb_any(buffer_info->skb);
922 buffer_info->skb = NULL;
924 buffer_info->time_stamp = 0;
927 static void e1000_print_hw_hang(struct work_struct *work)
929 struct e1000_adapter *adapter = container_of(work,
930 struct e1000_adapter,
931 print_hang_task);
932 struct e1000_ring *tx_ring = adapter->tx_ring;
933 unsigned int i = tx_ring->next_to_clean;
934 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
935 struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
936 struct e1000_hw *hw = &adapter->hw;
937 u16 phy_status, phy_1000t_status, phy_ext_status;
938 u16 pci_status;
940 if (test_bit(__E1000_DOWN, &adapter->state))
941 return;
943 e1e_rphy(hw, PHY_STATUS, &phy_status);
944 e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
945 e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
947 pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
949 /* detected Hardware unit hang */
950 e_err("Detected Hardware Unit Hang:\n"
951 " TDH <%x>\n"
952 " TDT <%x>\n"
953 " next_to_use <%x>\n"
954 " next_to_clean <%x>\n"
955 "buffer_info[next_to_clean]:\n"
956 " time_stamp <%lx>\n"
957 " next_to_watch <%x>\n"
958 " jiffies <%lx>\n"
959 " next_to_watch.status <%x>\n"
960 "MAC Status <%x>\n"
961 "PHY Status <%x>\n"
962 "PHY 1000BASE-T Status <%x>\n"
963 "PHY Extended Status <%x>\n"
964 "PCI Status <%x>\n",
965 readl(adapter->hw.hw_addr + tx_ring->head),
966 readl(adapter->hw.hw_addr + tx_ring->tail),
967 tx_ring->next_to_use,
968 tx_ring->next_to_clean,
969 tx_ring->buffer_info[eop].time_stamp,
970 eop,
971 jiffies,
972 eop_desc->upper.fields.status,
973 er32(STATUS),
974 phy_status,
975 phy_1000t_status,
976 phy_ext_status,
977 pci_status);
981 * e1000_clean_tx_irq - Reclaim resources after transmit completes
982 * @adapter: board private structure
984 * the return value indicates whether actual cleaning was done, there
985 * is no guarantee that everything was cleaned
987 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
989 struct net_device *netdev = adapter->netdev;
990 struct e1000_hw *hw = &adapter->hw;
991 struct e1000_ring *tx_ring = adapter->tx_ring;
992 struct e1000_tx_desc *tx_desc, *eop_desc;
993 struct e1000_buffer *buffer_info;
994 unsigned int i, eop;
995 unsigned int count = 0;
996 unsigned int total_tx_bytes = 0, total_tx_packets = 0;
998 i = tx_ring->next_to_clean;
999 eop = tx_ring->buffer_info[i].next_to_watch;
1000 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1002 while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1003 (count < tx_ring->count)) {
1004 bool cleaned = false;
1005 rmb(); /* read buffer_info after eop_desc */
1006 for (; !cleaned; count++) {
1007 tx_desc = E1000_TX_DESC(*tx_ring, i);
1008 buffer_info = &tx_ring->buffer_info[i];
1009 cleaned = (i == eop);
1011 if (cleaned) {
1012 total_tx_packets += buffer_info->segs;
1013 total_tx_bytes += buffer_info->bytecount;
1016 e1000_put_txbuf(adapter, buffer_info);
1017 tx_desc->upper.data = 0;
1019 i++;
1020 if (i == tx_ring->count)
1021 i = 0;
1024 if (i == tx_ring->next_to_use)
1025 break;
1026 eop = tx_ring->buffer_info[i].next_to_watch;
1027 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1030 tx_ring->next_to_clean = i;
1032 #define TX_WAKE_THRESHOLD 32
1033 if (count && netif_carrier_ok(netdev) &&
1034 e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1035 /* Make sure that anybody stopping the queue after this
1036 * sees the new next_to_clean.
1038 smp_mb();
1040 if (netif_queue_stopped(netdev) &&
1041 !(test_bit(__E1000_DOWN, &adapter->state))) {
1042 netif_wake_queue(netdev);
1043 ++adapter->restart_queue;
1047 if (adapter->detect_tx_hung) {
1049 * Detect a transmit hang in hardware, this serializes the
1050 * check with the clearing of time_stamp and movement of i
1052 adapter->detect_tx_hung = 0;
1053 if (tx_ring->buffer_info[i].time_stamp &&
1054 time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1055 + (adapter->tx_timeout_factor * HZ)) &&
1056 !(er32(STATUS) & E1000_STATUS_TXOFF)) {
1057 schedule_work(&adapter->print_hang_task);
1058 netif_stop_queue(netdev);
1061 adapter->total_tx_bytes += total_tx_bytes;
1062 adapter->total_tx_packets += total_tx_packets;
1063 netdev->stats.tx_bytes += total_tx_bytes;
1064 netdev->stats.tx_packets += total_tx_packets;
1065 return count < tx_ring->count;
1069 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1070 * @adapter: board private structure
1072 * the return value indicates whether actual cleaning was done, there
1073 * is no guarantee that everything was cleaned
1075 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1076 int *work_done, int work_to_do)
1078 struct e1000_hw *hw = &adapter->hw;
1079 union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1080 struct net_device *netdev = adapter->netdev;
1081 struct pci_dev *pdev = adapter->pdev;
1082 struct e1000_ring *rx_ring = adapter->rx_ring;
1083 struct e1000_buffer *buffer_info, *next_buffer;
1084 struct e1000_ps_page *ps_page;
1085 struct sk_buff *skb;
1086 unsigned int i, j;
1087 u32 length, staterr;
1088 int cleaned_count = 0;
1089 bool cleaned = 0;
1090 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1092 i = rx_ring->next_to_clean;
1093 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1094 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1095 buffer_info = &rx_ring->buffer_info[i];
1097 while (staterr & E1000_RXD_STAT_DD) {
1098 if (*work_done >= work_to_do)
1099 break;
1100 (*work_done)++;
1101 skb = buffer_info->skb;
1102 rmb(); /* read descriptor and rx_buffer_info after status DD */
1104 /* in the packet split case this is header only */
1105 prefetch(skb->data - NET_IP_ALIGN);
1107 i++;
1108 if (i == rx_ring->count)
1109 i = 0;
1110 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1111 prefetch(next_rxd);
1113 next_buffer = &rx_ring->buffer_info[i];
1115 cleaned = 1;
1116 cleaned_count++;
1117 dma_unmap_single(&pdev->dev, buffer_info->dma,
1118 adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1119 buffer_info->dma = 0;
1121 /* see !EOP comment in other Rx routine */
1122 if (!(staterr & E1000_RXD_STAT_EOP))
1123 adapter->flags2 |= FLAG2_IS_DISCARDING;
1125 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1126 e_dbg("Packet Split buffers didn't pick up the full "
1127 "packet\n");
1128 dev_kfree_skb_irq(skb);
1129 if (staterr & E1000_RXD_STAT_EOP)
1130 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1131 goto next_desc;
1134 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1135 dev_kfree_skb_irq(skb);
1136 goto next_desc;
1139 length = le16_to_cpu(rx_desc->wb.middle.length0);
1141 if (!length) {
1142 e_dbg("Last part of the packet spanning multiple "
1143 "descriptors\n");
1144 dev_kfree_skb_irq(skb);
1145 goto next_desc;
1148 /* Good Receive */
1149 skb_put(skb, length);
1153 * this looks ugly, but it seems compiler issues make it
1154 * more efficient than reusing j
1156 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1159 * page alloc/put takes too long and effects small packet
1160 * throughput, so unsplit small packets and save the alloc/put
1161 * only valid in softirq (napi) context to call kmap_*
1163 if (l1 && (l1 <= copybreak) &&
1164 ((length + l1) <= adapter->rx_ps_bsize0)) {
1165 u8 *vaddr;
1167 ps_page = &buffer_info->ps_pages[0];
1170 * there is no documentation about how to call
1171 * kmap_atomic, so we can't hold the mapping
1172 * very long
1174 dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1175 PAGE_SIZE, DMA_FROM_DEVICE);
1176 vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1177 memcpy(skb_tail_pointer(skb), vaddr, l1);
1178 kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1179 dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1180 PAGE_SIZE, DMA_FROM_DEVICE);
1182 /* remove the CRC */
1183 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1184 l1 -= 4;
1186 skb_put(skb, l1);
1187 goto copydone;
1188 } /* if */
1191 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1192 length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1193 if (!length)
1194 break;
1196 ps_page = &buffer_info->ps_pages[j];
1197 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1198 DMA_FROM_DEVICE);
1199 ps_page->dma = 0;
1200 skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1201 ps_page->page = NULL;
1202 skb->len += length;
1203 skb->data_len += length;
1204 skb->truesize += length;
1207 /* strip the ethernet crc, problem is we're using pages now so
1208 * this whole operation can get a little cpu intensive
1210 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1211 pskb_trim(skb, skb->len - 4);
1213 copydone:
1214 total_rx_bytes += skb->len;
1215 total_rx_packets++;
1217 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1218 rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1220 if (rx_desc->wb.upper.header_status &
1221 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1222 adapter->rx_hdr_split++;
1224 e1000_receive_skb(adapter, netdev, skb,
1225 staterr, rx_desc->wb.middle.vlan);
1227 next_desc:
1228 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1229 buffer_info->skb = NULL;
1231 /* return some buffers to hardware, one at a time is too slow */
1232 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1233 adapter->alloc_rx_buf(adapter, cleaned_count);
1234 cleaned_count = 0;
1237 /* use prefetched values */
1238 rx_desc = next_rxd;
1239 buffer_info = next_buffer;
1241 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1243 rx_ring->next_to_clean = i;
1245 cleaned_count = e1000_desc_unused(rx_ring);
1246 if (cleaned_count)
1247 adapter->alloc_rx_buf(adapter, cleaned_count);
1249 adapter->total_rx_bytes += total_rx_bytes;
1250 adapter->total_rx_packets += total_rx_packets;
1251 netdev->stats.rx_bytes += total_rx_bytes;
1252 netdev->stats.rx_packets += total_rx_packets;
1253 return cleaned;
1257 * e1000_consume_page - helper function
1259 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1260 u16 length)
1262 bi->page = NULL;
1263 skb->len += length;
1264 skb->data_len += length;
1265 skb->truesize += length;
1269 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1270 * @adapter: board private structure
1272 * the return value indicates whether actual cleaning was done, there
1273 * is no guarantee that everything was cleaned
1276 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1277 int *work_done, int work_to_do)
1279 struct net_device *netdev = adapter->netdev;
1280 struct pci_dev *pdev = adapter->pdev;
1281 struct e1000_ring *rx_ring = adapter->rx_ring;
1282 struct e1000_rx_desc *rx_desc, *next_rxd;
1283 struct e1000_buffer *buffer_info, *next_buffer;
1284 u32 length;
1285 unsigned int i;
1286 int cleaned_count = 0;
1287 bool cleaned = false;
1288 unsigned int total_rx_bytes=0, total_rx_packets=0;
1290 i = rx_ring->next_to_clean;
1291 rx_desc = E1000_RX_DESC(*rx_ring, i);
1292 buffer_info = &rx_ring->buffer_info[i];
1294 while (rx_desc->status & E1000_RXD_STAT_DD) {
1295 struct sk_buff *skb;
1296 u8 status;
1298 if (*work_done >= work_to_do)
1299 break;
1300 (*work_done)++;
1301 rmb(); /* read descriptor and rx_buffer_info after status DD */
1303 status = rx_desc->status;
1304 skb = buffer_info->skb;
1305 buffer_info->skb = NULL;
1307 ++i;
1308 if (i == rx_ring->count)
1309 i = 0;
1310 next_rxd = E1000_RX_DESC(*rx_ring, i);
1311 prefetch(next_rxd);
1313 next_buffer = &rx_ring->buffer_info[i];
1315 cleaned = true;
1316 cleaned_count++;
1317 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1318 DMA_FROM_DEVICE);
1319 buffer_info->dma = 0;
1321 length = le16_to_cpu(rx_desc->length);
1323 /* errors is only valid for DD + EOP descriptors */
1324 if (unlikely((status & E1000_RXD_STAT_EOP) &&
1325 (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
1326 /* recycle both page and skb */
1327 buffer_info->skb = skb;
1328 /* an error means any chain goes out the window
1329 * too */
1330 if (rx_ring->rx_skb_top)
1331 dev_kfree_skb(rx_ring->rx_skb_top);
1332 rx_ring->rx_skb_top = NULL;
1333 goto next_desc;
1336 #define rxtop (rx_ring->rx_skb_top)
1337 if (!(status & E1000_RXD_STAT_EOP)) {
1338 /* this descriptor is only the beginning (or middle) */
1339 if (!rxtop) {
1340 /* this is the beginning of a chain */
1341 rxtop = skb;
1342 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1343 0, length);
1344 } else {
1345 /* this is the middle of a chain */
1346 skb_fill_page_desc(rxtop,
1347 skb_shinfo(rxtop)->nr_frags,
1348 buffer_info->page, 0, length);
1349 /* re-use the skb, only consumed the page */
1350 buffer_info->skb = skb;
1352 e1000_consume_page(buffer_info, rxtop, length);
1353 goto next_desc;
1354 } else {
1355 if (rxtop) {
1356 /* end of the chain */
1357 skb_fill_page_desc(rxtop,
1358 skb_shinfo(rxtop)->nr_frags,
1359 buffer_info->page, 0, length);
1360 /* re-use the current skb, we only consumed the
1361 * page */
1362 buffer_info->skb = skb;
1363 skb = rxtop;
1364 rxtop = NULL;
1365 e1000_consume_page(buffer_info, skb, length);
1366 } else {
1367 /* no chain, got EOP, this buf is the packet
1368 * copybreak to save the put_page/alloc_page */
1369 if (length <= copybreak &&
1370 skb_tailroom(skb) >= length) {
1371 u8 *vaddr;
1372 vaddr = kmap_atomic(buffer_info->page,
1373 KM_SKB_DATA_SOFTIRQ);
1374 memcpy(skb_tail_pointer(skb), vaddr,
1375 length);
1376 kunmap_atomic(vaddr,
1377 KM_SKB_DATA_SOFTIRQ);
1378 /* re-use the page, so don't erase
1379 * buffer_info->page */
1380 skb_put(skb, length);
1381 } else {
1382 skb_fill_page_desc(skb, 0,
1383 buffer_info->page, 0,
1384 length);
1385 e1000_consume_page(buffer_info, skb,
1386 length);
1391 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1392 e1000_rx_checksum(adapter,
1393 (u32)(status) |
1394 ((u32)(rx_desc->errors) << 24),
1395 le16_to_cpu(rx_desc->csum), skb);
1397 /* probably a little skewed due to removing CRC */
1398 total_rx_bytes += skb->len;
1399 total_rx_packets++;
1401 /* eth type trans needs skb->data to point to something */
1402 if (!pskb_may_pull(skb, ETH_HLEN)) {
1403 e_err("pskb_may_pull failed.\n");
1404 dev_kfree_skb(skb);
1405 goto next_desc;
1408 e1000_receive_skb(adapter, netdev, skb, status,
1409 rx_desc->special);
1411 next_desc:
1412 rx_desc->status = 0;
1414 /* return some buffers to hardware, one at a time is too slow */
1415 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1416 adapter->alloc_rx_buf(adapter, cleaned_count);
1417 cleaned_count = 0;
1420 /* use prefetched values */
1421 rx_desc = next_rxd;
1422 buffer_info = next_buffer;
1424 rx_ring->next_to_clean = i;
1426 cleaned_count = e1000_desc_unused(rx_ring);
1427 if (cleaned_count)
1428 adapter->alloc_rx_buf(adapter, cleaned_count);
1430 adapter->total_rx_bytes += total_rx_bytes;
1431 adapter->total_rx_packets += total_rx_packets;
1432 netdev->stats.rx_bytes += total_rx_bytes;
1433 netdev->stats.rx_packets += total_rx_packets;
1434 return cleaned;
1438 * e1000_clean_rx_ring - Free Rx Buffers per Queue
1439 * @adapter: board private structure
1441 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1443 struct e1000_ring *rx_ring = adapter->rx_ring;
1444 struct e1000_buffer *buffer_info;
1445 struct e1000_ps_page *ps_page;
1446 struct pci_dev *pdev = adapter->pdev;
1447 unsigned int i, j;
1449 /* Free all the Rx ring sk_buffs */
1450 for (i = 0; i < rx_ring->count; i++) {
1451 buffer_info = &rx_ring->buffer_info[i];
1452 if (buffer_info->dma) {
1453 if (adapter->clean_rx == e1000_clean_rx_irq)
1454 dma_unmap_single(&pdev->dev, buffer_info->dma,
1455 adapter->rx_buffer_len,
1456 DMA_FROM_DEVICE);
1457 else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1458 dma_unmap_page(&pdev->dev, buffer_info->dma,
1459 PAGE_SIZE,
1460 DMA_FROM_DEVICE);
1461 else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1462 dma_unmap_single(&pdev->dev, buffer_info->dma,
1463 adapter->rx_ps_bsize0,
1464 DMA_FROM_DEVICE);
1465 buffer_info->dma = 0;
1468 if (buffer_info->page) {
1469 put_page(buffer_info->page);
1470 buffer_info->page = NULL;
1473 if (buffer_info->skb) {
1474 dev_kfree_skb(buffer_info->skb);
1475 buffer_info->skb = NULL;
1478 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1479 ps_page = &buffer_info->ps_pages[j];
1480 if (!ps_page->page)
1481 break;
1482 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1483 DMA_FROM_DEVICE);
1484 ps_page->dma = 0;
1485 put_page(ps_page->page);
1486 ps_page->page = NULL;
1490 /* there also may be some cached data from a chained receive */
1491 if (rx_ring->rx_skb_top) {
1492 dev_kfree_skb(rx_ring->rx_skb_top);
1493 rx_ring->rx_skb_top = NULL;
1496 /* Zero out the descriptor ring */
1497 memset(rx_ring->desc, 0, rx_ring->size);
1499 rx_ring->next_to_clean = 0;
1500 rx_ring->next_to_use = 0;
1501 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1503 writel(0, adapter->hw.hw_addr + rx_ring->head);
1504 writel(0, adapter->hw.hw_addr + rx_ring->tail);
1507 static void e1000e_downshift_workaround(struct work_struct *work)
1509 struct e1000_adapter *adapter = container_of(work,
1510 struct e1000_adapter, downshift_task);
1512 if (test_bit(__E1000_DOWN, &adapter->state))
1513 return;
1515 e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1519 * e1000_intr_msi - Interrupt Handler
1520 * @irq: interrupt number
1521 * @data: pointer to a network interface device structure
1523 static irqreturn_t e1000_intr_msi(int irq, void *data)
1525 struct net_device *netdev = data;
1526 struct e1000_adapter *adapter = netdev_priv(netdev);
1527 struct e1000_hw *hw = &adapter->hw;
1528 u32 icr = er32(ICR);
1531 * read ICR disables interrupts using IAM
1534 if (icr & E1000_ICR_LSC) {
1535 hw->mac.get_link_status = 1;
1537 * ICH8 workaround-- Call gig speed drop workaround on cable
1538 * disconnect (LSC) before accessing any PHY registers
1540 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1541 (!(er32(STATUS) & E1000_STATUS_LU)))
1542 schedule_work(&adapter->downshift_task);
1545 * 80003ES2LAN workaround-- For packet buffer work-around on
1546 * link down event; disable receives here in the ISR and reset
1547 * adapter in watchdog
1549 if (netif_carrier_ok(netdev) &&
1550 adapter->flags & FLAG_RX_NEEDS_RESTART) {
1551 /* disable receives */
1552 u32 rctl = er32(RCTL);
1553 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1554 adapter->flags |= FLAG_RX_RESTART_NOW;
1556 /* guard against interrupt when we're going down */
1557 if (!test_bit(__E1000_DOWN, &adapter->state))
1558 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1561 if (napi_schedule_prep(&adapter->napi)) {
1562 adapter->total_tx_bytes = 0;
1563 adapter->total_tx_packets = 0;
1564 adapter->total_rx_bytes = 0;
1565 adapter->total_rx_packets = 0;
1566 __napi_schedule(&adapter->napi);
1569 return IRQ_HANDLED;
1573 * e1000_intr - Interrupt Handler
1574 * @irq: interrupt number
1575 * @data: pointer to a network interface device structure
1577 static irqreturn_t e1000_intr(int irq, void *data)
1579 struct net_device *netdev = data;
1580 struct e1000_adapter *adapter = netdev_priv(netdev);
1581 struct e1000_hw *hw = &adapter->hw;
1582 u32 rctl, icr = er32(ICR);
1584 if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1585 return IRQ_NONE; /* Not our interrupt */
1588 * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1589 * not set, then the adapter didn't send an interrupt
1591 if (!(icr & E1000_ICR_INT_ASSERTED))
1592 return IRQ_NONE;
1595 * Interrupt Auto-Mask...upon reading ICR,
1596 * interrupts are masked. No need for the
1597 * IMC write
1600 if (icr & E1000_ICR_LSC) {
1601 hw->mac.get_link_status = 1;
1603 * ICH8 workaround-- Call gig speed drop workaround on cable
1604 * disconnect (LSC) before accessing any PHY registers
1606 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1607 (!(er32(STATUS) & E1000_STATUS_LU)))
1608 schedule_work(&adapter->downshift_task);
1611 * 80003ES2LAN workaround--
1612 * For packet buffer work-around on link down event;
1613 * disable receives here in the ISR and
1614 * reset adapter in watchdog
1616 if (netif_carrier_ok(netdev) &&
1617 (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1618 /* disable receives */
1619 rctl = er32(RCTL);
1620 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1621 adapter->flags |= FLAG_RX_RESTART_NOW;
1623 /* guard against interrupt when we're going down */
1624 if (!test_bit(__E1000_DOWN, &adapter->state))
1625 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1628 if (napi_schedule_prep(&adapter->napi)) {
1629 adapter->total_tx_bytes = 0;
1630 adapter->total_tx_packets = 0;
1631 adapter->total_rx_bytes = 0;
1632 adapter->total_rx_packets = 0;
1633 __napi_schedule(&adapter->napi);
1636 return IRQ_HANDLED;
1639 static irqreturn_t e1000_msix_other(int irq, void *data)
1641 struct net_device *netdev = data;
1642 struct e1000_adapter *adapter = netdev_priv(netdev);
1643 struct e1000_hw *hw = &adapter->hw;
1644 u32 icr = er32(ICR);
1646 if (!(icr & E1000_ICR_INT_ASSERTED)) {
1647 if (!test_bit(__E1000_DOWN, &adapter->state))
1648 ew32(IMS, E1000_IMS_OTHER);
1649 return IRQ_NONE;
1652 if (icr & adapter->eiac_mask)
1653 ew32(ICS, (icr & adapter->eiac_mask));
1655 if (icr & E1000_ICR_OTHER) {
1656 if (!(icr & E1000_ICR_LSC))
1657 goto no_link_interrupt;
1658 hw->mac.get_link_status = 1;
1659 /* guard against interrupt when we're going down */
1660 if (!test_bit(__E1000_DOWN, &adapter->state))
1661 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1664 no_link_interrupt:
1665 if (!test_bit(__E1000_DOWN, &adapter->state))
1666 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1668 return IRQ_HANDLED;
1672 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1674 struct net_device *netdev = data;
1675 struct e1000_adapter *adapter = netdev_priv(netdev);
1676 struct e1000_hw *hw = &adapter->hw;
1677 struct e1000_ring *tx_ring = adapter->tx_ring;
1680 adapter->total_tx_bytes = 0;
1681 adapter->total_tx_packets = 0;
1683 if (!e1000_clean_tx_irq(adapter))
1684 /* Ring was not completely cleaned, so fire another interrupt */
1685 ew32(ICS, tx_ring->ims_val);
1687 return IRQ_HANDLED;
1690 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1692 struct net_device *netdev = data;
1693 struct e1000_adapter *adapter = netdev_priv(netdev);
1695 /* Write the ITR value calculated at the end of the
1696 * previous interrupt.
1698 if (adapter->rx_ring->set_itr) {
1699 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1700 adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1701 adapter->rx_ring->set_itr = 0;
1704 if (napi_schedule_prep(&adapter->napi)) {
1705 adapter->total_rx_bytes = 0;
1706 adapter->total_rx_packets = 0;
1707 __napi_schedule(&adapter->napi);
1709 return IRQ_HANDLED;
1713 * e1000_configure_msix - Configure MSI-X hardware
1715 * e1000_configure_msix sets up the hardware to properly
1716 * generate MSI-X interrupts.
1718 static void e1000_configure_msix(struct e1000_adapter *adapter)
1720 struct e1000_hw *hw = &adapter->hw;
1721 struct e1000_ring *rx_ring = adapter->rx_ring;
1722 struct e1000_ring *tx_ring = adapter->tx_ring;
1723 int vector = 0;
1724 u32 ctrl_ext, ivar = 0;
1726 adapter->eiac_mask = 0;
1728 /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1729 if (hw->mac.type == e1000_82574) {
1730 u32 rfctl = er32(RFCTL);
1731 rfctl |= E1000_RFCTL_ACK_DIS;
1732 ew32(RFCTL, rfctl);
1735 #define E1000_IVAR_INT_ALLOC_VALID 0x8
1736 /* Configure Rx vector */
1737 rx_ring->ims_val = E1000_IMS_RXQ0;
1738 adapter->eiac_mask |= rx_ring->ims_val;
1739 if (rx_ring->itr_val)
1740 writel(1000000000 / (rx_ring->itr_val * 256),
1741 hw->hw_addr + rx_ring->itr_register);
1742 else
1743 writel(1, hw->hw_addr + rx_ring->itr_register);
1744 ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1746 /* Configure Tx vector */
1747 tx_ring->ims_val = E1000_IMS_TXQ0;
1748 vector++;
1749 if (tx_ring->itr_val)
1750 writel(1000000000 / (tx_ring->itr_val * 256),
1751 hw->hw_addr + tx_ring->itr_register);
1752 else
1753 writel(1, hw->hw_addr + tx_ring->itr_register);
1754 adapter->eiac_mask |= tx_ring->ims_val;
1755 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1757 /* set vector for Other Causes, e.g. link changes */
1758 vector++;
1759 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1760 if (rx_ring->itr_val)
1761 writel(1000000000 / (rx_ring->itr_val * 256),
1762 hw->hw_addr + E1000_EITR_82574(vector));
1763 else
1764 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1766 /* Cause Tx interrupts on every write back */
1767 ivar |= (1 << 31);
1769 ew32(IVAR, ivar);
1771 /* enable MSI-X PBA support */
1772 ctrl_ext = er32(CTRL_EXT);
1773 ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1775 /* Auto-Mask Other interrupts upon ICR read */
1776 #define E1000_EIAC_MASK_82574 0x01F00000
1777 ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1778 ctrl_ext |= E1000_CTRL_EXT_EIAME;
1779 ew32(CTRL_EXT, ctrl_ext);
1780 e1e_flush();
1783 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1785 if (adapter->msix_entries) {
1786 pci_disable_msix(adapter->pdev);
1787 kfree(adapter->msix_entries);
1788 adapter->msix_entries = NULL;
1789 } else if (adapter->flags & FLAG_MSI_ENABLED) {
1790 pci_disable_msi(adapter->pdev);
1791 adapter->flags &= ~FLAG_MSI_ENABLED;
1796 * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1798 * Attempt to configure interrupts using the best available
1799 * capabilities of the hardware and kernel.
1801 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1803 int err;
1804 int i;
1806 switch (adapter->int_mode) {
1807 case E1000E_INT_MODE_MSIX:
1808 if (adapter->flags & FLAG_HAS_MSIX) {
1809 adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1810 adapter->msix_entries = kcalloc(adapter->num_vectors,
1811 sizeof(struct msix_entry),
1812 GFP_KERNEL);
1813 if (adapter->msix_entries) {
1814 for (i = 0; i < adapter->num_vectors; i++)
1815 adapter->msix_entries[i].entry = i;
1817 err = pci_enable_msix(adapter->pdev,
1818 adapter->msix_entries,
1819 adapter->num_vectors);
1820 if (err == 0)
1821 return;
1823 /* MSI-X failed, so fall through and try MSI */
1824 e_err("Failed to initialize MSI-X interrupts. "
1825 "Falling back to MSI interrupts.\n");
1826 e1000e_reset_interrupt_capability(adapter);
1828 adapter->int_mode = E1000E_INT_MODE_MSI;
1829 /* Fall through */
1830 case E1000E_INT_MODE_MSI:
1831 if (!pci_enable_msi(adapter->pdev)) {
1832 adapter->flags |= FLAG_MSI_ENABLED;
1833 } else {
1834 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1835 e_err("Failed to initialize MSI interrupts. Falling "
1836 "back to legacy interrupts.\n");
1838 /* Fall through */
1839 case E1000E_INT_MODE_LEGACY:
1840 /* Don't do anything; this is the system default */
1841 break;
1844 /* store the number of vectors being used */
1845 adapter->num_vectors = 1;
1849 * e1000_request_msix - Initialize MSI-X interrupts
1851 * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1852 * kernel.
1854 static int e1000_request_msix(struct e1000_adapter *adapter)
1856 struct net_device *netdev = adapter->netdev;
1857 int err = 0, vector = 0;
1859 if (strlen(netdev->name) < (IFNAMSIZ - 5))
1860 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1861 else
1862 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1863 err = request_irq(adapter->msix_entries[vector].vector,
1864 e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1865 netdev);
1866 if (err)
1867 goto out;
1868 adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1869 adapter->rx_ring->itr_val = adapter->itr;
1870 vector++;
1872 if (strlen(netdev->name) < (IFNAMSIZ - 5))
1873 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1874 else
1875 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1876 err = request_irq(adapter->msix_entries[vector].vector,
1877 e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1878 netdev);
1879 if (err)
1880 goto out;
1881 adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1882 adapter->tx_ring->itr_val = adapter->itr;
1883 vector++;
1885 err = request_irq(adapter->msix_entries[vector].vector,
1886 e1000_msix_other, 0, netdev->name, netdev);
1887 if (err)
1888 goto out;
1890 e1000_configure_msix(adapter);
1891 return 0;
1892 out:
1893 return err;
1897 * e1000_request_irq - initialize interrupts
1899 * Attempts to configure interrupts using the best available
1900 * capabilities of the hardware and kernel.
1902 static int e1000_request_irq(struct e1000_adapter *adapter)
1904 struct net_device *netdev = adapter->netdev;
1905 int err;
1907 if (adapter->msix_entries) {
1908 err = e1000_request_msix(adapter);
1909 if (!err)
1910 return err;
1911 /* fall back to MSI */
1912 e1000e_reset_interrupt_capability(adapter);
1913 adapter->int_mode = E1000E_INT_MODE_MSI;
1914 e1000e_set_interrupt_capability(adapter);
1916 if (adapter->flags & FLAG_MSI_ENABLED) {
1917 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1918 netdev->name, netdev);
1919 if (!err)
1920 return err;
1922 /* fall back to legacy interrupt */
1923 e1000e_reset_interrupt_capability(adapter);
1924 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1927 err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1928 netdev->name, netdev);
1929 if (err)
1930 e_err("Unable to allocate interrupt, Error: %d\n", err);
1932 return err;
1935 static void e1000_free_irq(struct e1000_adapter *adapter)
1937 struct net_device *netdev = adapter->netdev;
1939 if (adapter->msix_entries) {
1940 int vector = 0;
1942 free_irq(adapter->msix_entries[vector].vector, netdev);
1943 vector++;
1945 free_irq(adapter->msix_entries[vector].vector, netdev);
1946 vector++;
1948 /* Other Causes interrupt vector */
1949 free_irq(adapter->msix_entries[vector].vector, netdev);
1950 return;
1953 free_irq(adapter->pdev->irq, netdev);
1957 * e1000_irq_disable - Mask off interrupt generation on the NIC
1959 static void e1000_irq_disable(struct e1000_adapter *adapter)
1961 struct e1000_hw *hw = &adapter->hw;
1963 ew32(IMC, ~0);
1964 if (adapter->msix_entries)
1965 ew32(EIAC_82574, 0);
1966 e1e_flush();
1968 if (adapter->msix_entries) {
1969 int i;
1970 for (i = 0; i < adapter->num_vectors; i++)
1971 synchronize_irq(adapter->msix_entries[i].vector);
1972 } else {
1973 synchronize_irq(adapter->pdev->irq);
1978 * e1000_irq_enable - Enable default interrupt generation settings
1980 static void e1000_irq_enable(struct e1000_adapter *adapter)
1982 struct e1000_hw *hw = &adapter->hw;
1984 if (adapter->msix_entries) {
1985 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1986 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1987 } else {
1988 ew32(IMS, IMS_ENABLE_MASK);
1990 e1e_flush();
1994 * e1000e_get_hw_control - get control of the h/w from f/w
1995 * @adapter: address of board private structure
1997 * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1998 * For ASF and Pass Through versions of f/w this means that
1999 * the driver is loaded. For AMT version (only with 82573)
2000 * of the f/w this means that the network i/f is open.
2002 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2004 struct e1000_hw *hw = &adapter->hw;
2005 u32 ctrl_ext;
2006 u32 swsm;
2008 /* Let firmware know the driver has taken over */
2009 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2010 swsm = er32(SWSM);
2011 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2012 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2013 ctrl_ext = er32(CTRL_EXT);
2014 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2019 * e1000e_release_hw_control - release control of the h/w to f/w
2020 * @adapter: address of board private structure
2022 * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2023 * For ASF and Pass Through versions of f/w this means that the
2024 * driver is no longer loaded. For AMT version (only with 82573) i
2025 * of the f/w this means that the network i/f is closed.
2028 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2030 struct e1000_hw *hw = &adapter->hw;
2031 u32 ctrl_ext;
2032 u32 swsm;
2034 /* Let firmware taken over control of h/w */
2035 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2036 swsm = er32(SWSM);
2037 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2038 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2039 ctrl_ext = er32(CTRL_EXT);
2040 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2045 * @e1000_alloc_ring - allocate memory for a ring structure
2047 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2048 struct e1000_ring *ring)
2050 struct pci_dev *pdev = adapter->pdev;
2052 ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2053 GFP_KERNEL);
2054 if (!ring->desc)
2055 return -ENOMEM;
2057 return 0;
2061 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2062 * @adapter: board private structure
2064 * Return 0 on success, negative on failure
2066 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2068 struct e1000_ring *tx_ring = adapter->tx_ring;
2069 int err = -ENOMEM, size;
2071 size = sizeof(struct e1000_buffer) * tx_ring->count;
2072 tx_ring->buffer_info = vzalloc(size);
2073 if (!tx_ring->buffer_info)
2074 goto err;
2076 /* round up to nearest 4K */
2077 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2078 tx_ring->size = ALIGN(tx_ring->size, 4096);
2080 err = e1000_alloc_ring_dma(adapter, tx_ring);
2081 if (err)
2082 goto err;
2084 tx_ring->next_to_use = 0;
2085 tx_ring->next_to_clean = 0;
2087 return 0;
2088 err:
2089 vfree(tx_ring->buffer_info);
2090 e_err("Unable to allocate memory for the transmit descriptor ring\n");
2091 return err;
2095 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2096 * @adapter: board private structure
2098 * Returns 0 on success, negative on failure
2100 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2102 struct e1000_ring *rx_ring = adapter->rx_ring;
2103 struct e1000_buffer *buffer_info;
2104 int i, size, desc_len, err = -ENOMEM;
2106 size = sizeof(struct e1000_buffer) * rx_ring->count;
2107 rx_ring->buffer_info = vzalloc(size);
2108 if (!rx_ring->buffer_info)
2109 goto err;
2111 for (i = 0; i < rx_ring->count; i++) {
2112 buffer_info = &rx_ring->buffer_info[i];
2113 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2114 sizeof(struct e1000_ps_page),
2115 GFP_KERNEL);
2116 if (!buffer_info->ps_pages)
2117 goto err_pages;
2120 desc_len = sizeof(union e1000_rx_desc_packet_split);
2122 /* Round up to nearest 4K */
2123 rx_ring->size = rx_ring->count * desc_len;
2124 rx_ring->size = ALIGN(rx_ring->size, 4096);
2126 err = e1000_alloc_ring_dma(adapter, rx_ring);
2127 if (err)
2128 goto err_pages;
2130 rx_ring->next_to_clean = 0;
2131 rx_ring->next_to_use = 0;
2132 rx_ring->rx_skb_top = NULL;
2134 return 0;
2136 err_pages:
2137 for (i = 0; i < rx_ring->count; i++) {
2138 buffer_info = &rx_ring->buffer_info[i];
2139 kfree(buffer_info->ps_pages);
2141 err:
2142 vfree(rx_ring->buffer_info);
2143 e_err("Unable to allocate memory for the receive descriptor ring\n");
2144 return err;
2148 * e1000_clean_tx_ring - Free Tx Buffers
2149 * @adapter: board private structure
2151 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2153 struct e1000_ring *tx_ring = adapter->tx_ring;
2154 struct e1000_buffer *buffer_info;
2155 unsigned long size;
2156 unsigned int i;
2158 for (i = 0; i < tx_ring->count; i++) {
2159 buffer_info = &tx_ring->buffer_info[i];
2160 e1000_put_txbuf(adapter, buffer_info);
2163 size = sizeof(struct e1000_buffer) * tx_ring->count;
2164 memset(tx_ring->buffer_info, 0, size);
2166 memset(tx_ring->desc, 0, tx_ring->size);
2168 tx_ring->next_to_use = 0;
2169 tx_ring->next_to_clean = 0;
2171 writel(0, adapter->hw.hw_addr + tx_ring->head);
2172 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2176 * e1000e_free_tx_resources - Free Tx Resources per Queue
2177 * @adapter: board private structure
2179 * Free all transmit software resources
2181 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2183 struct pci_dev *pdev = adapter->pdev;
2184 struct e1000_ring *tx_ring = adapter->tx_ring;
2186 e1000_clean_tx_ring(adapter);
2188 vfree(tx_ring->buffer_info);
2189 tx_ring->buffer_info = NULL;
2191 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2192 tx_ring->dma);
2193 tx_ring->desc = NULL;
2197 * e1000e_free_rx_resources - Free Rx Resources
2198 * @adapter: board private structure
2200 * Free all receive software resources
2203 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2205 struct pci_dev *pdev = adapter->pdev;
2206 struct e1000_ring *rx_ring = adapter->rx_ring;
2207 int i;
2209 e1000_clean_rx_ring(adapter);
2211 for (i = 0; i < rx_ring->count; i++)
2212 kfree(rx_ring->buffer_info[i].ps_pages);
2214 vfree(rx_ring->buffer_info);
2215 rx_ring->buffer_info = NULL;
2217 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2218 rx_ring->dma);
2219 rx_ring->desc = NULL;
2223 * e1000_update_itr - update the dynamic ITR value based on statistics
2224 * @adapter: pointer to adapter
2225 * @itr_setting: current adapter->itr
2226 * @packets: the number of packets during this measurement interval
2227 * @bytes: the number of bytes during this measurement interval
2229 * Stores a new ITR value based on packets and byte
2230 * counts during the last interrupt. The advantage of per interrupt
2231 * computation is faster updates and more accurate ITR for the current
2232 * traffic pattern. Constants in this function were computed
2233 * based on theoretical maximum wire speed and thresholds were set based
2234 * on testing data as well as attempting to minimize response time
2235 * while increasing bulk throughput. This functionality is controlled
2236 * by the InterruptThrottleRate module parameter.
2238 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2239 u16 itr_setting, int packets,
2240 int bytes)
2242 unsigned int retval = itr_setting;
2244 if (packets == 0)
2245 goto update_itr_done;
2247 switch (itr_setting) {
2248 case lowest_latency:
2249 /* handle TSO and jumbo frames */
2250 if (bytes/packets > 8000)
2251 retval = bulk_latency;
2252 else if ((packets < 5) && (bytes > 512))
2253 retval = low_latency;
2254 break;
2255 case low_latency: /* 50 usec aka 20000 ints/s */
2256 if (bytes > 10000) {
2257 /* this if handles the TSO accounting */
2258 if (bytes/packets > 8000)
2259 retval = bulk_latency;
2260 else if ((packets < 10) || ((bytes/packets) > 1200))
2261 retval = bulk_latency;
2262 else if ((packets > 35))
2263 retval = lowest_latency;
2264 } else if (bytes/packets > 2000) {
2265 retval = bulk_latency;
2266 } else if (packets <= 2 && bytes < 512) {
2267 retval = lowest_latency;
2269 break;
2270 case bulk_latency: /* 250 usec aka 4000 ints/s */
2271 if (bytes > 25000) {
2272 if (packets > 35)
2273 retval = low_latency;
2274 } else if (bytes < 6000) {
2275 retval = low_latency;
2277 break;
2280 update_itr_done:
2281 return retval;
2284 static void e1000_set_itr(struct e1000_adapter *adapter)
2286 struct e1000_hw *hw = &adapter->hw;
2287 u16 current_itr;
2288 u32 new_itr = adapter->itr;
2290 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2291 if (adapter->link_speed != SPEED_1000) {
2292 current_itr = 0;
2293 new_itr = 4000;
2294 goto set_itr_now;
2297 if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2298 new_itr = 0;
2299 goto set_itr_now;
2302 adapter->tx_itr = e1000_update_itr(adapter,
2303 adapter->tx_itr,
2304 adapter->total_tx_packets,
2305 adapter->total_tx_bytes);
2306 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2307 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2308 adapter->tx_itr = low_latency;
2310 adapter->rx_itr = e1000_update_itr(adapter,
2311 adapter->rx_itr,
2312 adapter->total_rx_packets,
2313 adapter->total_rx_bytes);
2314 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2315 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2316 adapter->rx_itr = low_latency;
2318 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2320 switch (current_itr) {
2321 /* counts and packets in update_itr are dependent on these numbers */
2322 case lowest_latency:
2323 new_itr = 70000;
2324 break;
2325 case low_latency:
2326 new_itr = 20000; /* aka hwitr = ~200 */
2327 break;
2328 case bulk_latency:
2329 new_itr = 4000;
2330 break;
2331 default:
2332 break;
2335 set_itr_now:
2336 if (new_itr != adapter->itr) {
2338 * this attempts to bias the interrupt rate towards Bulk
2339 * by adding intermediate steps when interrupt rate is
2340 * increasing
2342 new_itr = new_itr > adapter->itr ?
2343 min(adapter->itr + (new_itr >> 2), new_itr) :
2344 new_itr;
2345 adapter->itr = new_itr;
2346 adapter->rx_ring->itr_val = new_itr;
2347 if (adapter->msix_entries)
2348 adapter->rx_ring->set_itr = 1;
2349 else
2350 if (new_itr)
2351 ew32(ITR, 1000000000 / (new_itr * 256));
2352 else
2353 ew32(ITR, 0);
2358 * e1000_alloc_queues - Allocate memory for all rings
2359 * @adapter: board private structure to initialize
2361 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2363 adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2364 if (!adapter->tx_ring)
2365 goto err;
2367 adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2368 if (!adapter->rx_ring)
2369 goto err;
2371 return 0;
2372 err:
2373 e_err("Unable to allocate memory for queues\n");
2374 kfree(adapter->rx_ring);
2375 kfree(adapter->tx_ring);
2376 return -ENOMEM;
2380 * e1000_clean - NAPI Rx polling callback
2381 * @napi: struct associated with this polling callback
2382 * @budget: amount of packets driver is allowed to process this poll
2384 static int e1000_clean(struct napi_struct *napi, int budget)
2386 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2387 struct e1000_hw *hw = &adapter->hw;
2388 struct net_device *poll_dev = adapter->netdev;
2389 int tx_cleaned = 1, work_done = 0;
2391 adapter = netdev_priv(poll_dev);
2393 if (adapter->msix_entries &&
2394 !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2395 goto clean_rx;
2397 tx_cleaned = e1000_clean_tx_irq(adapter);
2399 clean_rx:
2400 adapter->clean_rx(adapter, &work_done, budget);
2402 if (!tx_cleaned)
2403 work_done = budget;
2405 /* If budget not fully consumed, exit the polling mode */
2406 if (work_done < budget) {
2407 if (adapter->itr_setting & 3)
2408 e1000_set_itr(adapter);
2409 napi_complete(napi);
2410 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2411 if (adapter->msix_entries)
2412 ew32(IMS, adapter->rx_ring->ims_val);
2413 else
2414 e1000_irq_enable(adapter);
2418 return work_done;
2421 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2423 struct e1000_adapter *adapter = netdev_priv(netdev);
2424 struct e1000_hw *hw = &adapter->hw;
2425 u32 vfta, index;
2427 /* don't update vlan cookie if already programmed */
2428 if ((adapter->hw.mng_cookie.status &
2429 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2430 (vid == adapter->mng_vlan_id))
2431 return;
2433 /* add VID to filter table */
2434 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2435 index = (vid >> 5) & 0x7F;
2436 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2437 vfta |= (1 << (vid & 0x1F));
2438 hw->mac.ops.write_vfta(hw, index, vfta);
2442 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2444 struct e1000_adapter *adapter = netdev_priv(netdev);
2445 struct e1000_hw *hw = &adapter->hw;
2446 u32 vfta, index;
2448 if (!test_bit(__E1000_DOWN, &adapter->state))
2449 e1000_irq_disable(adapter);
2450 vlan_group_set_device(adapter->vlgrp, vid, NULL);
2452 if (!test_bit(__E1000_DOWN, &adapter->state))
2453 e1000_irq_enable(adapter);
2455 if ((adapter->hw.mng_cookie.status &
2456 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2457 (vid == adapter->mng_vlan_id)) {
2458 /* release control to f/w */
2459 e1000e_release_hw_control(adapter);
2460 return;
2463 /* remove VID from filter table */
2464 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2465 index = (vid >> 5) & 0x7F;
2466 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2467 vfta &= ~(1 << (vid & 0x1F));
2468 hw->mac.ops.write_vfta(hw, index, vfta);
2472 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2474 struct net_device *netdev = adapter->netdev;
2475 u16 vid = adapter->hw.mng_cookie.vlan_id;
2476 u16 old_vid = adapter->mng_vlan_id;
2478 if (!adapter->vlgrp)
2479 return;
2481 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2482 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2483 if (adapter->hw.mng_cookie.status &
2484 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2485 e1000_vlan_rx_add_vid(netdev, vid);
2486 adapter->mng_vlan_id = vid;
2489 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2490 (vid != old_vid) &&
2491 !vlan_group_get_device(adapter->vlgrp, old_vid))
2492 e1000_vlan_rx_kill_vid(netdev, old_vid);
2493 } else {
2494 adapter->mng_vlan_id = vid;
2499 static void e1000_vlan_rx_register(struct net_device *netdev,
2500 struct vlan_group *grp)
2502 struct e1000_adapter *adapter = netdev_priv(netdev);
2503 struct e1000_hw *hw = &adapter->hw;
2504 u32 ctrl, rctl;
2506 if (!test_bit(__E1000_DOWN, &adapter->state))
2507 e1000_irq_disable(adapter);
2508 adapter->vlgrp = grp;
2510 if (grp) {
2511 /* enable VLAN tag insert/strip */
2512 ctrl = er32(CTRL);
2513 ctrl |= E1000_CTRL_VME;
2514 ew32(CTRL, ctrl);
2516 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2517 /* enable VLAN receive filtering */
2518 rctl = er32(RCTL);
2519 rctl &= ~E1000_RCTL_CFIEN;
2520 ew32(RCTL, rctl);
2521 e1000_update_mng_vlan(adapter);
2523 } else {
2524 /* disable VLAN tag insert/strip */
2525 ctrl = er32(CTRL);
2526 ctrl &= ~E1000_CTRL_VME;
2527 ew32(CTRL, ctrl);
2529 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2530 if (adapter->mng_vlan_id !=
2531 (u16)E1000_MNG_VLAN_NONE) {
2532 e1000_vlan_rx_kill_vid(netdev,
2533 adapter->mng_vlan_id);
2534 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2539 if (!test_bit(__E1000_DOWN, &adapter->state))
2540 e1000_irq_enable(adapter);
2543 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2545 u16 vid;
2547 e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2549 if (!adapter->vlgrp)
2550 return;
2552 for (vid = 0; vid < VLAN_N_VID; vid++) {
2553 if (!vlan_group_get_device(adapter->vlgrp, vid))
2554 continue;
2555 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2559 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2561 struct e1000_hw *hw = &adapter->hw;
2562 u32 manc, manc2h, mdef, i, j;
2564 if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2565 return;
2567 manc = er32(MANC);
2570 * enable receiving management packets to the host. this will probably
2571 * generate destination unreachable messages from the host OS, but
2572 * the packets will be handled on SMBUS
2574 manc |= E1000_MANC_EN_MNG2HOST;
2575 manc2h = er32(MANC2H);
2577 switch (hw->mac.type) {
2578 default:
2579 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2580 break;
2581 case e1000_82574:
2582 case e1000_82583:
2584 * Check if IPMI pass-through decision filter already exists;
2585 * if so, enable it.
2587 for (i = 0, j = 0; i < 8; i++) {
2588 mdef = er32(MDEF(i));
2590 /* Ignore filters with anything other than IPMI ports */
2591 if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2592 continue;
2594 /* Enable this decision filter in MANC2H */
2595 if (mdef)
2596 manc2h |= (1 << i);
2598 j |= mdef;
2601 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2602 break;
2604 /* Create new decision filter in an empty filter */
2605 for (i = 0, j = 0; i < 8; i++)
2606 if (er32(MDEF(i)) == 0) {
2607 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2608 E1000_MDEF_PORT_664));
2609 manc2h |= (1 << 1);
2610 j++;
2611 break;
2614 if (!j)
2615 e_warn("Unable to create IPMI pass-through filter\n");
2616 break;
2619 ew32(MANC2H, manc2h);
2620 ew32(MANC, manc);
2624 * e1000_configure_tx - Configure Transmit Unit after Reset
2625 * @adapter: board private structure
2627 * Configure the Tx unit of the MAC after a reset.
2629 static void e1000_configure_tx(struct e1000_adapter *adapter)
2631 struct e1000_hw *hw = &adapter->hw;
2632 struct e1000_ring *tx_ring = adapter->tx_ring;
2633 u64 tdba;
2634 u32 tdlen, tctl, tipg, tarc;
2635 u32 ipgr1, ipgr2;
2637 /* Setup the HW Tx Head and Tail descriptor pointers */
2638 tdba = tx_ring->dma;
2639 tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2640 ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2641 ew32(TDBAH, (tdba >> 32));
2642 ew32(TDLEN, tdlen);
2643 ew32(TDH, 0);
2644 ew32(TDT, 0);
2645 tx_ring->head = E1000_TDH;
2646 tx_ring->tail = E1000_TDT;
2648 /* Set the default values for the Tx Inter Packet Gap timer */
2649 tipg = DEFAULT_82543_TIPG_IPGT_COPPER; /* 8 */
2650 ipgr1 = DEFAULT_82543_TIPG_IPGR1; /* 8 */
2651 ipgr2 = DEFAULT_82543_TIPG_IPGR2; /* 6 */
2653 if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2654 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /* 7 */
2656 tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2657 tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2658 ew32(TIPG, tipg);
2660 /* Set the Tx Interrupt Delay register */
2661 ew32(TIDV, adapter->tx_int_delay);
2662 /* Tx irq moderation */
2663 ew32(TADV, adapter->tx_abs_int_delay);
2665 if (adapter->flags2 & FLAG2_DMA_BURST) {
2666 u32 txdctl = er32(TXDCTL(0));
2667 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2668 E1000_TXDCTL_WTHRESH);
2670 * set up some performance related parameters to encourage the
2671 * hardware to use the bus more efficiently in bursts, depends
2672 * on the tx_int_delay to be enabled,
2673 * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2674 * hthresh = 1 ==> prefetch when one or more available
2675 * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2676 * BEWARE: this seems to work but should be considered first if
2677 * there are Tx hangs or other Tx related bugs
2679 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2680 ew32(TXDCTL(0), txdctl);
2681 /* erratum work around: set txdctl the same for both queues */
2682 ew32(TXDCTL(1), txdctl);
2685 /* Program the Transmit Control Register */
2686 tctl = er32(TCTL);
2687 tctl &= ~E1000_TCTL_CT;
2688 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2689 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2691 if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2692 tarc = er32(TARC(0));
2694 * set the speed mode bit, we'll clear it if we're not at
2695 * gigabit link later
2697 #define SPEED_MODE_BIT (1 << 21)
2698 tarc |= SPEED_MODE_BIT;
2699 ew32(TARC(0), tarc);
2702 /* errata: program both queues to unweighted RR */
2703 if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2704 tarc = er32(TARC(0));
2705 tarc |= 1;
2706 ew32(TARC(0), tarc);
2707 tarc = er32(TARC(1));
2708 tarc |= 1;
2709 ew32(TARC(1), tarc);
2712 /* Setup Transmit Descriptor Settings for eop descriptor */
2713 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2715 /* only set IDE if we are delaying interrupts using the timers */
2716 if (adapter->tx_int_delay)
2717 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2719 /* enable Report Status bit */
2720 adapter->txd_cmd |= E1000_TXD_CMD_RS;
2722 ew32(TCTL, tctl);
2724 e1000e_config_collision_dist(hw);
2728 * e1000_setup_rctl - configure the receive control registers
2729 * @adapter: Board private structure
2731 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2732 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2733 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2735 struct e1000_hw *hw = &adapter->hw;
2736 u32 rctl, rfctl;
2737 u32 psrctl = 0;
2738 u32 pages = 0;
2740 /* Workaround Si errata on 82579 - configure jumbo frame flow */
2741 if (hw->mac.type == e1000_pch2lan) {
2742 s32 ret_val;
2744 if (adapter->netdev->mtu > ETH_DATA_LEN)
2745 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2746 else
2747 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2749 if (ret_val)
2750 e_dbg("failed to enable jumbo frame workaround mode\n");
2753 /* Program MC offset vector base */
2754 rctl = er32(RCTL);
2755 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2756 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2757 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2758 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2760 /* Do not Store bad packets */
2761 rctl &= ~E1000_RCTL_SBP;
2763 /* Enable Long Packet receive */
2764 if (adapter->netdev->mtu <= ETH_DATA_LEN)
2765 rctl &= ~E1000_RCTL_LPE;
2766 else
2767 rctl |= E1000_RCTL_LPE;
2769 /* Some systems expect that the CRC is included in SMBUS traffic. The
2770 * hardware strips the CRC before sending to both SMBUS (BMC) and to
2771 * host memory when this is enabled
2773 if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2774 rctl |= E1000_RCTL_SECRC;
2776 /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2777 if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2778 u16 phy_data;
2780 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2781 phy_data &= 0xfff8;
2782 phy_data |= (1 << 2);
2783 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2785 e1e_rphy(hw, 22, &phy_data);
2786 phy_data &= 0x0fff;
2787 phy_data |= (1 << 14);
2788 e1e_wphy(hw, 0x10, 0x2823);
2789 e1e_wphy(hw, 0x11, 0x0003);
2790 e1e_wphy(hw, 22, phy_data);
2793 /* Setup buffer sizes */
2794 rctl &= ~E1000_RCTL_SZ_4096;
2795 rctl |= E1000_RCTL_BSEX;
2796 switch (adapter->rx_buffer_len) {
2797 case 2048:
2798 default:
2799 rctl |= E1000_RCTL_SZ_2048;
2800 rctl &= ~E1000_RCTL_BSEX;
2801 break;
2802 case 4096:
2803 rctl |= E1000_RCTL_SZ_4096;
2804 break;
2805 case 8192:
2806 rctl |= E1000_RCTL_SZ_8192;
2807 break;
2808 case 16384:
2809 rctl |= E1000_RCTL_SZ_16384;
2810 break;
2814 * 82571 and greater support packet-split where the protocol
2815 * header is placed in skb->data and the packet data is
2816 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2817 * In the case of a non-split, skb->data is linearly filled,
2818 * followed by the page buffers. Therefore, skb->data is
2819 * sized to hold the largest protocol header.
2821 * allocations using alloc_page take too long for regular MTU
2822 * so only enable packet split for jumbo frames
2824 * Using pages when the page size is greater than 16k wastes
2825 * a lot of memory, since we allocate 3 pages at all times
2826 * per packet.
2828 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2829 if (!(adapter->flags & FLAG_HAS_ERT) && (pages <= 3) &&
2830 (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2831 adapter->rx_ps_pages = pages;
2832 else
2833 adapter->rx_ps_pages = 0;
2835 if (adapter->rx_ps_pages) {
2836 /* Configure extra packet-split registers */
2837 rfctl = er32(RFCTL);
2838 rfctl |= E1000_RFCTL_EXTEN;
2840 * disable packet split support for IPv6 extension headers,
2841 * because some malformed IPv6 headers can hang the Rx
2843 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2844 E1000_RFCTL_NEW_IPV6_EXT_DIS);
2846 ew32(RFCTL, rfctl);
2848 /* Enable Packet split descriptors */
2849 rctl |= E1000_RCTL_DTYP_PS;
2851 psrctl |= adapter->rx_ps_bsize0 >>
2852 E1000_PSRCTL_BSIZE0_SHIFT;
2854 switch (adapter->rx_ps_pages) {
2855 case 3:
2856 psrctl |= PAGE_SIZE <<
2857 E1000_PSRCTL_BSIZE3_SHIFT;
2858 case 2:
2859 psrctl |= PAGE_SIZE <<
2860 E1000_PSRCTL_BSIZE2_SHIFT;
2861 case 1:
2862 psrctl |= PAGE_SIZE >>
2863 E1000_PSRCTL_BSIZE1_SHIFT;
2864 break;
2867 ew32(PSRCTL, psrctl);
2870 ew32(RCTL, rctl);
2871 /* just started the receive unit, no need to restart */
2872 adapter->flags &= ~FLAG_RX_RESTART_NOW;
2876 * e1000_configure_rx - Configure Receive Unit after Reset
2877 * @adapter: board private structure
2879 * Configure the Rx unit of the MAC after a reset.
2881 static void e1000_configure_rx(struct e1000_adapter *adapter)
2883 struct e1000_hw *hw = &adapter->hw;
2884 struct e1000_ring *rx_ring = adapter->rx_ring;
2885 u64 rdba;
2886 u32 rdlen, rctl, rxcsum, ctrl_ext;
2888 if (adapter->rx_ps_pages) {
2889 /* this is a 32 byte descriptor */
2890 rdlen = rx_ring->count *
2891 sizeof(union e1000_rx_desc_packet_split);
2892 adapter->clean_rx = e1000_clean_rx_irq_ps;
2893 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2894 } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2895 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2896 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2897 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2898 } else {
2899 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2900 adapter->clean_rx = e1000_clean_rx_irq;
2901 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2904 /* disable receives while setting up the descriptors */
2905 rctl = er32(RCTL);
2906 ew32(RCTL, rctl & ~E1000_RCTL_EN);
2907 e1e_flush();
2908 msleep(10);
2910 if (adapter->flags2 & FLAG2_DMA_BURST) {
2912 * set the writeback threshold (only takes effect if the RDTR
2913 * is set). set GRAN=1 and write back up to 0x4 worth, and
2914 * enable prefetching of 0x20 Rx descriptors
2915 * granularity = 01
2916 * wthresh = 04,
2917 * hthresh = 04,
2918 * pthresh = 0x20
2920 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
2921 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
2924 * override the delay timers for enabling bursting, only if
2925 * the value was not set by the user via module options
2927 if (adapter->rx_int_delay == DEFAULT_RDTR)
2928 adapter->rx_int_delay = BURST_RDTR;
2929 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
2930 adapter->rx_abs_int_delay = BURST_RADV;
2933 /* set the Receive Delay Timer Register */
2934 ew32(RDTR, adapter->rx_int_delay);
2936 /* irq moderation */
2937 ew32(RADV, adapter->rx_abs_int_delay);
2938 if ((adapter->itr_setting != 0) && (adapter->itr != 0))
2939 ew32(ITR, 1000000000 / (adapter->itr * 256));
2941 ctrl_ext = er32(CTRL_EXT);
2942 /* Auto-Mask interrupts upon ICR access */
2943 ctrl_ext |= E1000_CTRL_EXT_IAME;
2944 ew32(IAM, 0xffffffff);
2945 ew32(CTRL_EXT, ctrl_ext);
2946 e1e_flush();
2949 * Setup the HW Rx Head and Tail Descriptor Pointers and
2950 * the Base and Length of the Rx Descriptor Ring
2952 rdba = rx_ring->dma;
2953 ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2954 ew32(RDBAH, (rdba >> 32));
2955 ew32(RDLEN, rdlen);
2956 ew32(RDH, 0);
2957 ew32(RDT, 0);
2958 rx_ring->head = E1000_RDH;
2959 rx_ring->tail = E1000_RDT;
2961 /* Enable Receive Checksum Offload for TCP and UDP */
2962 rxcsum = er32(RXCSUM);
2963 if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2964 rxcsum |= E1000_RXCSUM_TUOFL;
2967 * IPv4 payload checksum for UDP fragments must be
2968 * used in conjunction with packet-split.
2970 if (adapter->rx_ps_pages)
2971 rxcsum |= E1000_RXCSUM_IPPCSE;
2972 } else {
2973 rxcsum &= ~E1000_RXCSUM_TUOFL;
2974 /* no need to clear IPPCSE as it defaults to 0 */
2976 ew32(RXCSUM, rxcsum);
2979 * Enable early receives on supported devices, only takes effect when
2980 * packet size is equal or larger than the specified value (in 8 byte
2981 * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2983 if ((adapter->flags & FLAG_HAS_ERT) ||
2984 (adapter->hw.mac.type == e1000_pch2lan)) {
2985 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2986 u32 rxdctl = er32(RXDCTL(0));
2987 ew32(RXDCTL(0), rxdctl | 0x3);
2988 if (adapter->flags & FLAG_HAS_ERT)
2989 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2991 * With jumbo frames and early-receive enabled,
2992 * excessive C-state transition latencies result in
2993 * dropped transactions.
2995 pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
2996 } else {
2997 pm_qos_update_request(&adapter->netdev->pm_qos_req,
2998 PM_QOS_DEFAULT_VALUE);
3002 /* Enable Receives */
3003 ew32(RCTL, rctl);
3007 * e1000_update_mc_addr_list - Update Multicast addresses
3008 * @hw: pointer to the HW structure
3009 * @mc_addr_list: array of multicast addresses to program
3010 * @mc_addr_count: number of multicast addresses to program
3012 * Updates the Multicast Table Array.
3013 * The caller must have a packed mc_addr_list of multicast addresses.
3015 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
3016 u32 mc_addr_count)
3018 hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
3022 * e1000_set_multi - Multicast and Promiscuous mode set
3023 * @netdev: network interface device structure
3025 * The set_multi entry point is called whenever the multicast address
3026 * list or the network interface flags are updated. This routine is
3027 * responsible for configuring the hardware for proper multicast,
3028 * promiscuous mode, and all-multi behavior.
3030 static void e1000_set_multi(struct net_device *netdev)
3032 struct e1000_adapter *adapter = netdev_priv(netdev);
3033 struct e1000_hw *hw = &adapter->hw;
3034 struct netdev_hw_addr *ha;
3035 u8 *mta_list;
3036 u32 rctl;
3037 int i;
3039 /* Check for Promiscuous and All Multicast modes */
3041 rctl = er32(RCTL);
3043 if (netdev->flags & IFF_PROMISC) {
3044 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3045 rctl &= ~E1000_RCTL_VFE;
3046 } else {
3047 if (netdev->flags & IFF_ALLMULTI) {
3048 rctl |= E1000_RCTL_MPE;
3049 rctl &= ~E1000_RCTL_UPE;
3050 } else {
3051 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3053 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
3054 rctl |= E1000_RCTL_VFE;
3057 ew32(RCTL, rctl);
3059 if (!netdev_mc_empty(netdev)) {
3060 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3061 if (!mta_list)
3062 return;
3064 /* prepare a packed array of only addresses. */
3065 i = 0;
3066 netdev_for_each_mc_addr(ha, netdev)
3067 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3069 e1000_update_mc_addr_list(hw, mta_list, i);
3070 kfree(mta_list);
3071 } else {
3073 * if we're called from probe, we might not have
3074 * anything to do here, so clear out the list
3076 e1000_update_mc_addr_list(hw, NULL, 0);
3081 * e1000_configure - configure the hardware for Rx and Tx
3082 * @adapter: private board structure
3084 static void e1000_configure(struct e1000_adapter *adapter)
3086 e1000_set_multi(adapter->netdev);
3088 e1000_restore_vlan(adapter);
3089 e1000_init_manageability_pt(adapter);
3091 e1000_configure_tx(adapter);
3092 e1000_setup_rctl(adapter);
3093 e1000_configure_rx(adapter);
3094 adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
3098 * e1000e_power_up_phy - restore link in case the phy was powered down
3099 * @adapter: address of board private structure
3101 * The phy may be powered down to save power and turn off link when the
3102 * driver is unloaded and wake on lan is not enabled (among others)
3103 * *** this routine MUST be followed by a call to e1000e_reset ***
3105 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3107 if (adapter->hw.phy.ops.power_up)
3108 adapter->hw.phy.ops.power_up(&adapter->hw);
3110 adapter->hw.mac.ops.setup_link(&adapter->hw);
3114 * e1000_power_down_phy - Power down the PHY
3116 * Power down the PHY so no link is implied when interface is down.
3117 * The PHY cannot be powered down if management or WoL is active.
3119 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3121 /* WoL is enabled */
3122 if (adapter->wol)
3123 return;
3125 if (adapter->hw.phy.ops.power_down)
3126 adapter->hw.phy.ops.power_down(&adapter->hw);
3130 * e1000e_reset - bring the hardware into a known good state
3132 * This function boots the hardware and enables some settings that
3133 * require a configuration cycle of the hardware - those cannot be
3134 * set/changed during runtime. After reset the device needs to be
3135 * properly configured for Rx, Tx etc.
3137 void e1000e_reset(struct e1000_adapter *adapter)
3139 struct e1000_mac_info *mac = &adapter->hw.mac;
3140 struct e1000_fc_info *fc = &adapter->hw.fc;
3141 struct e1000_hw *hw = &adapter->hw;
3142 u32 tx_space, min_tx_space, min_rx_space;
3143 u32 pba = adapter->pba;
3144 u16 hwm;
3146 /* reset Packet Buffer Allocation to default */
3147 ew32(PBA, pba);
3149 if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3151 * To maintain wire speed transmits, the Tx FIFO should be
3152 * large enough to accommodate two full transmit packets,
3153 * rounded up to the next 1KB and expressed in KB. Likewise,
3154 * the Rx FIFO should be large enough to accommodate at least
3155 * one full receive packet and is similarly rounded up and
3156 * expressed in KB.
3158 pba = er32(PBA);
3159 /* upper 16 bits has Tx packet buffer allocation size in KB */
3160 tx_space = pba >> 16;
3161 /* lower 16 bits has Rx packet buffer allocation size in KB */
3162 pba &= 0xffff;
3164 * the Tx fifo also stores 16 bytes of information about the Tx
3165 * but don't include ethernet FCS because hardware appends it
3167 min_tx_space = (adapter->max_frame_size +
3168 sizeof(struct e1000_tx_desc) -
3169 ETH_FCS_LEN) * 2;
3170 min_tx_space = ALIGN(min_tx_space, 1024);
3171 min_tx_space >>= 10;
3172 /* software strips receive CRC, so leave room for it */
3173 min_rx_space = adapter->max_frame_size;
3174 min_rx_space = ALIGN(min_rx_space, 1024);
3175 min_rx_space >>= 10;
3178 * If current Tx allocation is less than the min Tx FIFO size,
3179 * and the min Tx FIFO size is less than the current Rx FIFO
3180 * allocation, take space away from current Rx allocation
3182 if ((tx_space < min_tx_space) &&
3183 ((min_tx_space - tx_space) < pba)) {
3184 pba -= min_tx_space - tx_space;
3187 * if short on Rx space, Rx wins and must trump Tx
3188 * adjustment or use Early Receive if available
3190 if ((pba < min_rx_space) &&
3191 (!(adapter->flags & FLAG_HAS_ERT)))
3192 /* ERT enabled in e1000_configure_rx */
3193 pba = min_rx_space;
3196 ew32(PBA, pba);
3200 * flow control settings
3202 * The high water mark must be low enough to fit one full frame
3203 * (or the size used for early receive) above it in the Rx FIFO.
3204 * Set it to the lower of:
3205 * - 90% of the Rx FIFO size, and
3206 * - the full Rx FIFO size minus the early receive size (for parts
3207 * with ERT support assuming ERT set to E1000_ERT_2048), or
3208 * - the full Rx FIFO size minus one full frame
3210 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3211 fc->pause_time = 0xFFFF;
3212 else
3213 fc->pause_time = E1000_FC_PAUSE_TIME;
3214 fc->send_xon = 1;
3215 fc->current_mode = fc->requested_mode;
3217 switch (hw->mac.type) {
3218 default:
3219 if ((adapter->flags & FLAG_HAS_ERT) &&
3220 (adapter->netdev->mtu > ETH_DATA_LEN))
3221 hwm = min(((pba << 10) * 9 / 10),
3222 ((pba << 10) - (E1000_ERT_2048 << 3)));
3223 else
3224 hwm = min(((pba << 10) * 9 / 10),
3225 ((pba << 10) - adapter->max_frame_size));
3227 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3228 fc->low_water = fc->high_water - 8;
3229 break;
3230 case e1000_pchlan:
3232 * Workaround PCH LOM adapter hangs with certain network
3233 * loads. If hangs persist, try disabling Tx flow control.
3235 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3236 fc->high_water = 0x3500;
3237 fc->low_water = 0x1500;
3238 } else {
3239 fc->high_water = 0x5000;
3240 fc->low_water = 0x3000;
3242 fc->refresh_time = 0x1000;
3243 break;
3244 case e1000_pch2lan:
3245 fc->high_water = 0x05C20;
3246 fc->low_water = 0x05048;
3247 fc->pause_time = 0x0650;
3248 fc->refresh_time = 0x0400;
3249 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3250 pba = 14;
3251 ew32(PBA, pba);
3253 break;
3257 * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3258 * fit in receive buffer and early-receive not supported.
3260 if (adapter->itr_setting & 0x3) {
3261 if (((adapter->max_frame_size * 2) > (pba << 10)) &&
3262 !(adapter->flags & FLAG_HAS_ERT)) {
3263 if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3264 dev_info(&adapter->pdev->dev,
3265 "Interrupt Throttle Rate turned off\n");
3266 adapter->flags2 |= FLAG2_DISABLE_AIM;
3267 ew32(ITR, 0);
3269 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3270 dev_info(&adapter->pdev->dev,
3271 "Interrupt Throttle Rate turned on\n");
3272 adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3273 adapter->itr = 20000;
3274 ew32(ITR, 1000000000 / (adapter->itr * 256));
3278 /* Allow time for pending master requests to run */
3279 mac->ops.reset_hw(hw);
3282 * For parts with AMT enabled, let the firmware know
3283 * that the network interface is in control
3285 if (adapter->flags & FLAG_HAS_AMT)
3286 e1000e_get_hw_control(adapter);
3288 ew32(WUC, 0);
3290 if (mac->ops.init_hw(hw))
3291 e_err("Hardware Error\n");
3293 e1000_update_mng_vlan(adapter);
3295 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3296 ew32(VET, ETH_P_8021Q);
3298 e1000e_reset_adaptive(hw);
3300 if (!netif_running(adapter->netdev) &&
3301 !test_bit(__E1000_TESTING, &adapter->state)) {
3302 e1000_power_down_phy(adapter);
3303 return;
3306 e1000_get_phy_info(hw);
3308 if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3309 !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3310 u16 phy_data = 0;
3312 * speed up time to link by disabling smart power down, ignore
3313 * the return value of this function because there is nothing
3314 * different we would do if it failed
3316 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3317 phy_data &= ~IGP02E1000_PM_SPD;
3318 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3322 int e1000e_up(struct e1000_adapter *adapter)
3324 struct e1000_hw *hw = &adapter->hw;
3326 /* hardware has been reset, we need to reload some things */
3327 e1000_configure(adapter);
3329 clear_bit(__E1000_DOWN, &adapter->state);
3331 napi_enable(&adapter->napi);
3332 if (adapter->msix_entries)
3333 e1000_configure_msix(adapter);
3334 e1000_irq_enable(adapter);
3336 netif_wake_queue(adapter->netdev);
3338 /* fire a link change interrupt to start the watchdog */
3339 if (adapter->msix_entries)
3340 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3341 else
3342 ew32(ICS, E1000_ICS_LSC);
3344 return 0;
3347 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3349 struct e1000_hw *hw = &adapter->hw;
3351 if (!(adapter->flags2 & FLAG2_DMA_BURST))
3352 return;
3354 /* flush pending descriptor writebacks to memory */
3355 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3356 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3358 /* execute the writes immediately */
3359 e1e_flush();
3362 void e1000e_down(struct e1000_adapter *adapter)
3364 struct net_device *netdev = adapter->netdev;
3365 struct e1000_hw *hw = &adapter->hw;
3366 u32 tctl, rctl;
3369 * signal that we're down so the interrupt handler does not
3370 * reschedule our watchdog timer
3372 set_bit(__E1000_DOWN, &adapter->state);
3374 /* disable receives in the hardware */
3375 rctl = er32(RCTL);
3376 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3377 /* flush and sleep below */
3379 netif_stop_queue(netdev);
3381 /* disable transmits in the hardware */
3382 tctl = er32(TCTL);
3383 tctl &= ~E1000_TCTL_EN;
3384 ew32(TCTL, tctl);
3385 /* flush both disables and wait for them to finish */
3386 e1e_flush();
3387 msleep(10);
3389 napi_disable(&adapter->napi);
3390 e1000_irq_disable(adapter);
3392 del_timer_sync(&adapter->watchdog_timer);
3393 del_timer_sync(&adapter->phy_info_timer);
3395 netif_carrier_off(netdev);
3396 adapter->link_speed = 0;
3397 adapter->link_duplex = 0;
3399 if (!pci_channel_offline(adapter->pdev))
3400 e1000e_reset(adapter);
3402 e1000e_flush_descriptors(adapter);
3404 e1000_clean_tx_ring(adapter);
3405 e1000_clean_rx_ring(adapter);
3408 * TODO: for power management, we could drop the link and
3409 * pci_disable_device here.
3413 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3415 might_sleep();
3416 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3417 msleep(1);
3418 e1000e_down(adapter);
3419 e1000e_up(adapter);
3420 clear_bit(__E1000_RESETTING, &adapter->state);
3424 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3425 * @adapter: board private structure to initialize
3427 * e1000_sw_init initializes the Adapter private data structure.
3428 * Fields are initialized based on PCI device information and
3429 * OS network device settings (MTU size).
3431 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3433 struct net_device *netdev = adapter->netdev;
3435 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3436 adapter->rx_ps_bsize0 = 128;
3437 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3438 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3440 e1000e_set_interrupt_capability(adapter);
3442 if (e1000_alloc_queues(adapter))
3443 return -ENOMEM;
3445 /* Explicitly disable IRQ since the NIC can be in any state. */
3446 e1000_irq_disable(adapter);
3448 set_bit(__E1000_DOWN, &adapter->state);
3449 return 0;
3453 * e1000_intr_msi_test - Interrupt Handler
3454 * @irq: interrupt number
3455 * @data: pointer to a network interface device structure
3457 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3459 struct net_device *netdev = data;
3460 struct e1000_adapter *adapter = netdev_priv(netdev);
3461 struct e1000_hw *hw = &adapter->hw;
3462 u32 icr = er32(ICR);
3464 e_dbg("icr is %08X\n", icr);
3465 if (icr & E1000_ICR_RXSEQ) {
3466 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3467 wmb();
3470 return IRQ_HANDLED;
3474 * e1000_test_msi_interrupt - Returns 0 for successful test
3475 * @adapter: board private struct
3477 * code flow taken from tg3.c
3479 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3481 struct net_device *netdev = adapter->netdev;
3482 struct e1000_hw *hw = &adapter->hw;
3483 int err;
3485 /* poll_enable hasn't been called yet, so don't need disable */
3486 /* clear any pending events */
3487 er32(ICR);
3489 /* free the real vector and request a test handler */
3490 e1000_free_irq(adapter);
3491 e1000e_reset_interrupt_capability(adapter);
3493 /* Assume that the test fails, if it succeeds then the test
3494 * MSI irq handler will unset this flag */
3495 adapter->flags |= FLAG_MSI_TEST_FAILED;
3497 err = pci_enable_msi(adapter->pdev);
3498 if (err)
3499 goto msi_test_failed;
3501 err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3502 netdev->name, netdev);
3503 if (err) {
3504 pci_disable_msi(adapter->pdev);
3505 goto msi_test_failed;
3508 wmb();
3510 e1000_irq_enable(adapter);
3512 /* fire an unusual interrupt on the test handler */
3513 ew32(ICS, E1000_ICS_RXSEQ);
3514 e1e_flush();
3515 msleep(50);
3517 e1000_irq_disable(adapter);
3519 rmb();
3521 if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3522 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3523 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3524 } else
3525 e_dbg("MSI interrupt test succeeded!\n");
3527 free_irq(adapter->pdev->irq, netdev);
3528 pci_disable_msi(adapter->pdev);
3530 msi_test_failed:
3531 e1000e_set_interrupt_capability(adapter);
3532 return e1000_request_irq(adapter);
3536 * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3537 * @adapter: board private struct
3539 * code flow taken from tg3.c, called with e1000 interrupts disabled.
3541 static int e1000_test_msi(struct e1000_adapter *adapter)
3543 int err;
3544 u16 pci_cmd;
3546 if (!(adapter->flags & FLAG_MSI_ENABLED))
3547 return 0;
3549 /* disable SERR in case the MSI write causes a master abort */
3550 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3551 if (pci_cmd & PCI_COMMAND_SERR)
3552 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3553 pci_cmd & ~PCI_COMMAND_SERR);
3555 err = e1000_test_msi_interrupt(adapter);
3557 /* re-enable SERR */
3558 if (pci_cmd & PCI_COMMAND_SERR) {
3559 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3560 pci_cmd |= PCI_COMMAND_SERR;
3561 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3564 return err;
3568 * e1000_open - Called when a network interface is made active
3569 * @netdev: network interface device structure
3571 * Returns 0 on success, negative value on failure
3573 * The open entry point is called when a network interface is made
3574 * active by the system (IFF_UP). At this point all resources needed
3575 * for transmit and receive operations are allocated, the interrupt
3576 * handler is registered with the OS, the watchdog timer is started,
3577 * and the stack is notified that the interface is ready.
3579 static int e1000_open(struct net_device *netdev)
3581 struct e1000_adapter *adapter = netdev_priv(netdev);
3582 struct e1000_hw *hw = &adapter->hw;
3583 struct pci_dev *pdev = adapter->pdev;
3584 int err;
3586 /* disallow open during test */
3587 if (test_bit(__E1000_TESTING, &adapter->state))
3588 return -EBUSY;
3590 pm_runtime_get_sync(&pdev->dev);
3592 netif_carrier_off(netdev);
3594 /* allocate transmit descriptors */
3595 err = e1000e_setup_tx_resources(adapter);
3596 if (err)
3597 goto err_setup_tx;
3599 /* allocate receive descriptors */
3600 err = e1000e_setup_rx_resources(adapter);
3601 if (err)
3602 goto err_setup_rx;
3605 * If AMT is enabled, let the firmware know that the network
3606 * interface is now open and reset the part to a known state.
3608 if (adapter->flags & FLAG_HAS_AMT) {
3609 e1000e_get_hw_control(adapter);
3610 e1000e_reset(adapter);
3613 e1000e_power_up_phy(adapter);
3615 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3616 if ((adapter->hw.mng_cookie.status &
3617 E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3618 e1000_update_mng_vlan(adapter);
3620 /* DMA latency requirement to workaround early-receive/jumbo issue */
3621 if ((adapter->flags & FLAG_HAS_ERT) ||
3622 (adapter->hw.mac.type == e1000_pch2lan))
3623 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3624 PM_QOS_CPU_DMA_LATENCY,
3625 PM_QOS_DEFAULT_VALUE);
3628 * before we allocate an interrupt, we must be ready to handle it.
3629 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3630 * as soon as we call pci_request_irq, so we have to setup our
3631 * clean_rx handler before we do so.
3633 e1000_configure(adapter);
3635 err = e1000_request_irq(adapter);
3636 if (err)
3637 goto err_req_irq;
3640 * Work around PCIe errata with MSI interrupts causing some chipsets to
3641 * ignore e1000e MSI messages, which means we need to test our MSI
3642 * interrupt now
3644 if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3645 err = e1000_test_msi(adapter);
3646 if (err) {
3647 e_err("Interrupt allocation failed\n");
3648 goto err_req_irq;
3652 /* From here on the code is the same as e1000e_up() */
3653 clear_bit(__E1000_DOWN, &adapter->state);
3655 napi_enable(&adapter->napi);
3657 e1000_irq_enable(adapter);
3659 netif_start_queue(netdev);
3661 adapter->idle_check = true;
3662 pm_runtime_put(&pdev->dev);
3664 /* fire a link status change interrupt to start the watchdog */
3665 if (adapter->msix_entries)
3666 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3667 else
3668 ew32(ICS, E1000_ICS_LSC);
3670 return 0;
3672 err_req_irq:
3673 e1000e_release_hw_control(adapter);
3674 e1000_power_down_phy(adapter);
3675 e1000e_free_rx_resources(adapter);
3676 err_setup_rx:
3677 e1000e_free_tx_resources(adapter);
3678 err_setup_tx:
3679 e1000e_reset(adapter);
3680 pm_runtime_put_sync(&pdev->dev);
3682 return err;
3686 * e1000_close - Disables a network interface
3687 * @netdev: network interface device structure
3689 * Returns 0, this is not allowed to fail
3691 * The close entry point is called when an interface is de-activated
3692 * by the OS. The hardware is still under the drivers control, but
3693 * needs to be disabled. A global MAC reset is issued to stop the
3694 * hardware, and all transmit and receive resources are freed.
3696 static int e1000_close(struct net_device *netdev)
3698 struct e1000_adapter *adapter = netdev_priv(netdev);
3699 struct pci_dev *pdev = adapter->pdev;
3701 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3703 pm_runtime_get_sync(&pdev->dev);
3705 if (!test_bit(__E1000_DOWN, &adapter->state)) {
3706 e1000e_down(adapter);
3707 e1000_free_irq(adapter);
3709 e1000_power_down_phy(adapter);
3711 e1000e_free_tx_resources(adapter);
3712 e1000e_free_rx_resources(adapter);
3715 * kill manageability vlan ID if supported, but not if a vlan with
3716 * the same ID is registered on the host OS (let 8021q kill it)
3718 if ((adapter->hw.mng_cookie.status &
3719 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3720 !(adapter->vlgrp &&
3721 vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3722 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3725 * If AMT is enabled, let the firmware know that the network
3726 * interface is now closed
3728 if ((adapter->flags & FLAG_HAS_AMT) &&
3729 !test_bit(__E1000_TESTING, &adapter->state))
3730 e1000e_release_hw_control(adapter);
3732 if ((adapter->flags & FLAG_HAS_ERT) ||
3733 (adapter->hw.mac.type == e1000_pch2lan))
3734 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3736 pm_runtime_put_sync(&pdev->dev);
3738 return 0;
3741 * e1000_set_mac - Change the Ethernet Address of the NIC
3742 * @netdev: network interface device structure
3743 * @p: pointer to an address structure
3745 * Returns 0 on success, negative on failure
3747 static int e1000_set_mac(struct net_device *netdev, void *p)
3749 struct e1000_adapter *adapter = netdev_priv(netdev);
3750 struct sockaddr *addr = p;
3752 if (!is_valid_ether_addr(addr->sa_data))
3753 return -EADDRNOTAVAIL;
3755 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3756 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3758 e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3760 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3761 /* activate the work around */
3762 e1000e_set_laa_state_82571(&adapter->hw, 1);
3765 * Hold a copy of the LAA in RAR[14] This is done so that
3766 * between the time RAR[0] gets clobbered and the time it
3767 * gets fixed (in e1000_watchdog), the actual LAA is in one
3768 * of the RARs and no incoming packets directed to this port
3769 * are dropped. Eventually the LAA will be in RAR[0] and
3770 * RAR[14]
3772 e1000e_rar_set(&adapter->hw,
3773 adapter->hw.mac.addr,
3774 adapter->hw.mac.rar_entry_count - 1);
3777 return 0;
3781 * e1000e_update_phy_task - work thread to update phy
3782 * @work: pointer to our work struct
3784 * this worker thread exists because we must acquire a
3785 * semaphore to read the phy, which we could msleep while
3786 * waiting for it, and we can't msleep in a timer.
3788 static void e1000e_update_phy_task(struct work_struct *work)
3790 struct e1000_adapter *adapter = container_of(work,
3791 struct e1000_adapter, update_phy_task);
3793 if (test_bit(__E1000_DOWN, &adapter->state))
3794 return;
3796 e1000_get_phy_info(&adapter->hw);
3800 * Need to wait a few seconds after link up to get diagnostic information from
3801 * the phy
3803 static void e1000_update_phy_info(unsigned long data)
3805 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3807 if (test_bit(__E1000_DOWN, &adapter->state))
3808 return;
3810 schedule_work(&adapter->update_phy_task);
3814 * e1000e_update_phy_stats - Update the PHY statistics counters
3815 * @adapter: board private structure
3817 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
3819 struct e1000_hw *hw = &adapter->hw;
3820 s32 ret_val;
3821 u16 phy_data;
3823 ret_val = hw->phy.ops.acquire(hw);
3824 if (ret_val)
3825 return;
3827 hw->phy.addr = 1;
3829 #define HV_PHY_STATS_PAGE 778
3831 * A page set is expensive so check if already on desired page.
3832 * If not, set to the page with the PHY status registers.
3834 ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
3835 &phy_data);
3836 if (ret_val)
3837 goto release;
3838 if (phy_data != (HV_PHY_STATS_PAGE << IGP_PAGE_SHIFT)) {
3839 ret_val = e1000e_write_phy_reg_mdic(hw,
3840 IGP01E1000_PHY_PAGE_SELECT,
3841 (HV_PHY_STATS_PAGE <<
3842 IGP_PAGE_SHIFT));
3843 if (ret_val)
3844 goto release;
3847 /* Read/clear the upper 16-bit registers and read/accumulate lower */
3849 /* Single Collision Count */
3850 e1000e_read_phy_reg_mdic(hw, HV_SCC_UPPER & MAX_PHY_REG_ADDRESS,
3851 &phy_data);
3852 ret_val = e1000e_read_phy_reg_mdic(hw,
3853 HV_SCC_LOWER & MAX_PHY_REG_ADDRESS,
3854 &phy_data);
3855 if (!ret_val)
3856 adapter->stats.scc += phy_data;
3858 /* Excessive Collision Count */
3859 e1000e_read_phy_reg_mdic(hw, HV_ECOL_UPPER & MAX_PHY_REG_ADDRESS,
3860 &phy_data);
3861 ret_val = e1000e_read_phy_reg_mdic(hw,
3862 HV_ECOL_LOWER & MAX_PHY_REG_ADDRESS,
3863 &phy_data);
3864 if (!ret_val)
3865 adapter->stats.ecol += phy_data;
3867 /* Multiple Collision Count */
3868 e1000e_read_phy_reg_mdic(hw, HV_MCC_UPPER & MAX_PHY_REG_ADDRESS,
3869 &phy_data);
3870 ret_val = e1000e_read_phy_reg_mdic(hw,
3871 HV_MCC_LOWER & MAX_PHY_REG_ADDRESS,
3872 &phy_data);
3873 if (!ret_val)
3874 adapter->stats.mcc += phy_data;
3876 /* Late Collision Count */
3877 e1000e_read_phy_reg_mdic(hw, HV_LATECOL_UPPER & MAX_PHY_REG_ADDRESS,
3878 &phy_data);
3879 ret_val = e1000e_read_phy_reg_mdic(hw,
3880 HV_LATECOL_LOWER &
3881 MAX_PHY_REG_ADDRESS,
3882 &phy_data);
3883 if (!ret_val)
3884 adapter->stats.latecol += phy_data;
3886 /* Collision Count - also used for adaptive IFS */
3887 e1000e_read_phy_reg_mdic(hw, HV_COLC_UPPER & MAX_PHY_REG_ADDRESS,
3888 &phy_data);
3889 ret_val = e1000e_read_phy_reg_mdic(hw,
3890 HV_COLC_LOWER & MAX_PHY_REG_ADDRESS,
3891 &phy_data);
3892 if (!ret_val)
3893 hw->mac.collision_delta = phy_data;
3895 /* Defer Count */
3896 e1000e_read_phy_reg_mdic(hw, HV_DC_UPPER & MAX_PHY_REG_ADDRESS,
3897 &phy_data);
3898 ret_val = e1000e_read_phy_reg_mdic(hw,
3899 HV_DC_LOWER & MAX_PHY_REG_ADDRESS,
3900 &phy_data);
3901 if (!ret_val)
3902 adapter->stats.dc += phy_data;
3904 /* Transmit with no CRS */
3905 e1000e_read_phy_reg_mdic(hw, HV_TNCRS_UPPER & MAX_PHY_REG_ADDRESS,
3906 &phy_data);
3907 ret_val = e1000e_read_phy_reg_mdic(hw,
3908 HV_TNCRS_LOWER & MAX_PHY_REG_ADDRESS,
3909 &phy_data);
3910 if (!ret_val)
3911 adapter->stats.tncrs += phy_data;
3913 release:
3914 hw->phy.ops.release(hw);
3918 * e1000e_update_stats - Update the board statistics counters
3919 * @adapter: board private structure
3921 void e1000e_update_stats(struct e1000_adapter *adapter)
3923 struct net_device *netdev = adapter->netdev;
3924 struct e1000_hw *hw = &adapter->hw;
3925 struct pci_dev *pdev = adapter->pdev;
3928 * Prevent stats update while adapter is being reset, or if the pci
3929 * connection is down.
3931 if (adapter->link_speed == 0)
3932 return;
3933 if (pci_channel_offline(pdev))
3934 return;
3936 adapter->stats.crcerrs += er32(CRCERRS);
3937 adapter->stats.gprc += er32(GPRC);
3938 adapter->stats.gorc += er32(GORCL);
3939 er32(GORCH); /* Clear gorc */
3940 adapter->stats.bprc += er32(BPRC);
3941 adapter->stats.mprc += er32(MPRC);
3942 adapter->stats.roc += er32(ROC);
3944 adapter->stats.mpc += er32(MPC);
3946 /* Half-duplex statistics */
3947 if (adapter->link_duplex == HALF_DUPLEX) {
3948 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
3949 e1000e_update_phy_stats(adapter);
3950 } else {
3951 adapter->stats.scc += er32(SCC);
3952 adapter->stats.ecol += er32(ECOL);
3953 adapter->stats.mcc += er32(MCC);
3954 adapter->stats.latecol += er32(LATECOL);
3955 adapter->stats.dc += er32(DC);
3957 hw->mac.collision_delta = er32(COLC);
3959 if ((hw->mac.type != e1000_82574) &&
3960 (hw->mac.type != e1000_82583))
3961 adapter->stats.tncrs += er32(TNCRS);
3963 adapter->stats.colc += hw->mac.collision_delta;
3966 adapter->stats.xonrxc += er32(XONRXC);
3967 adapter->stats.xontxc += er32(XONTXC);
3968 adapter->stats.xoffrxc += er32(XOFFRXC);
3969 adapter->stats.xofftxc += er32(XOFFTXC);
3970 adapter->stats.gptc += er32(GPTC);
3971 adapter->stats.gotc += er32(GOTCL);
3972 er32(GOTCH); /* Clear gotc */
3973 adapter->stats.rnbc += er32(RNBC);
3974 adapter->stats.ruc += er32(RUC);
3976 adapter->stats.mptc += er32(MPTC);
3977 adapter->stats.bptc += er32(BPTC);
3979 /* used for adaptive IFS */
3981 hw->mac.tx_packet_delta = er32(TPT);
3982 adapter->stats.tpt += hw->mac.tx_packet_delta;
3984 adapter->stats.algnerrc += er32(ALGNERRC);
3985 adapter->stats.rxerrc += er32(RXERRC);
3986 adapter->stats.cexterr += er32(CEXTERR);
3987 adapter->stats.tsctc += er32(TSCTC);
3988 adapter->stats.tsctfc += er32(TSCTFC);
3990 /* Fill out the OS statistics structure */
3991 netdev->stats.multicast = adapter->stats.mprc;
3992 netdev->stats.collisions = adapter->stats.colc;
3994 /* Rx Errors */
3997 * RLEC on some newer hardware can be incorrect so build
3998 * our own version based on RUC and ROC
4000 netdev->stats.rx_errors = adapter->stats.rxerrc +
4001 adapter->stats.crcerrs + adapter->stats.algnerrc +
4002 adapter->stats.ruc + adapter->stats.roc +
4003 adapter->stats.cexterr;
4004 netdev->stats.rx_length_errors = adapter->stats.ruc +
4005 adapter->stats.roc;
4006 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4007 netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4008 netdev->stats.rx_missed_errors = adapter->stats.mpc;
4010 /* Tx Errors */
4011 netdev->stats.tx_errors = adapter->stats.ecol +
4012 adapter->stats.latecol;
4013 netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4014 netdev->stats.tx_window_errors = adapter->stats.latecol;
4015 netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4017 /* Tx Dropped needs to be maintained elsewhere */
4019 /* Management Stats */
4020 adapter->stats.mgptc += er32(MGTPTC);
4021 adapter->stats.mgprc += er32(MGTPRC);
4022 adapter->stats.mgpdc += er32(MGTPDC);
4026 * e1000_phy_read_status - Update the PHY register status snapshot
4027 * @adapter: board private structure
4029 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4031 struct e1000_hw *hw = &adapter->hw;
4032 struct e1000_phy_regs *phy = &adapter->phy_regs;
4033 int ret_val;
4035 if ((er32(STATUS) & E1000_STATUS_LU) &&
4036 (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4037 ret_val = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4038 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4039 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4040 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4041 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4042 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4043 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4044 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4045 if (ret_val)
4046 e_warn("Error reading PHY register\n");
4047 } else {
4049 * Do not read PHY registers if link is not up
4050 * Set values to typical power-on defaults
4052 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4053 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4054 BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4055 BMSR_ERCAP);
4056 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4057 ADVERTISE_ALL | ADVERTISE_CSMA);
4058 phy->lpa = 0;
4059 phy->expansion = EXPANSION_ENABLENPAGE;
4060 phy->ctrl1000 = ADVERTISE_1000FULL;
4061 phy->stat1000 = 0;
4062 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4066 static void e1000_print_link_info(struct e1000_adapter *adapter)
4068 struct e1000_hw *hw = &adapter->hw;
4069 u32 ctrl = er32(CTRL);
4071 /* Link status message must follow this format for user tools */
4072 printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
4073 "Flow Control: %s\n",
4074 adapter->netdev->name,
4075 adapter->link_speed,
4076 (adapter->link_duplex == FULL_DUPLEX) ?
4077 "Full Duplex" : "Half Duplex",
4078 ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
4079 "Rx/Tx" :
4080 ((ctrl & E1000_CTRL_RFCE) ? "Rx" :
4081 ((ctrl & E1000_CTRL_TFCE) ? "Tx" : "None")));
4084 static bool e1000e_has_link(struct e1000_adapter *adapter)
4086 struct e1000_hw *hw = &adapter->hw;
4087 bool link_active = 0;
4088 s32 ret_val = 0;
4091 * get_link_status is set on LSC (link status) interrupt or
4092 * Rx sequence error interrupt. get_link_status will stay
4093 * false until the check_for_link establishes link
4094 * for copper adapters ONLY
4096 switch (hw->phy.media_type) {
4097 case e1000_media_type_copper:
4098 if (hw->mac.get_link_status) {
4099 ret_val = hw->mac.ops.check_for_link(hw);
4100 link_active = !hw->mac.get_link_status;
4101 } else {
4102 link_active = 1;
4104 break;
4105 case e1000_media_type_fiber:
4106 ret_val = hw->mac.ops.check_for_link(hw);
4107 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4108 break;
4109 case e1000_media_type_internal_serdes:
4110 ret_val = hw->mac.ops.check_for_link(hw);
4111 link_active = adapter->hw.mac.serdes_has_link;
4112 break;
4113 default:
4114 case e1000_media_type_unknown:
4115 break;
4118 if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4119 (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4120 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4121 e_info("Gigabit has been disabled, downgrading speed\n");
4124 return link_active;
4127 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4129 /* make sure the receive unit is started */
4130 if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4131 (adapter->flags & FLAG_RX_RESTART_NOW)) {
4132 struct e1000_hw *hw = &adapter->hw;
4133 u32 rctl = er32(RCTL);
4134 ew32(RCTL, rctl | E1000_RCTL_EN);
4135 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4139 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4141 struct e1000_hw *hw = &adapter->hw;
4144 * With 82574 controllers, PHY needs to be checked periodically
4145 * for hung state and reset, if two calls return true
4147 if (e1000_check_phy_82574(hw))
4148 adapter->phy_hang_count++;
4149 else
4150 adapter->phy_hang_count = 0;
4152 if (adapter->phy_hang_count > 1) {
4153 adapter->phy_hang_count = 0;
4154 schedule_work(&adapter->reset_task);
4159 * e1000_watchdog - Timer Call-back
4160 * @data: pointer to adapter cast into an unsigned long
4162 static void e1000_watchdog(unsigned long data)
4164 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4166 /* Do the rest outside of interrupt context */
4167 schedule_work(&adapter->watchdog_task);
4169 /* TODO: make this use queue_delayed_work() */
4172 static void e1000_watchdog_task(struct work_struct *work)
4174 struct e1000_adapter *adapter = container_of(work,
4175 struct e1000_adapter, watchdog_task);
4176 struct net_device *netdev = adapter->netdev;
4177 struct e1000_mac_info *mac = &adapter->hw.mac;
4178 struct e1000_phy_info *phy = &adapter->hw.phy;
4179 struct e1000_ring *tx_ring = adapter->tx_ring;
4180 struct e1000_hw *hw = &adapter->hw;
4181 u32 link, tctl;
4182 int tx_pending = 0;
4184 if (test_bit(__E1000_DOWN, &adapter->state))
4185 return;
4187 link = e1000e_has_link(adapter);
4188 if ((netif_carrier_ok(netdev)) && link) {
4189 /* Cancel scheduled suspend requests. */
4190 pm_runtime_resume(netdev->dev.parent);
4192 e1000e_enable_receives(adapter);
4193 goto link_up;
4196 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4197 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4198 e1000_update_mng_vlan(adapter);
4200 if (link) {
4201 if (!netif_carrier_ok(netdev)) {
4202 bool txb2b = 1;
4204 /* Cancel scheduled suspend requests. */
4205 pm_runtime_resume(netdev->dev.parent);
4207 /* update snapshot of PHY registers on LSC */
4208 e1000_phy_read_status(adapter);
4209 mac->ops.get_link_up_info(&adapter->hw,
4210 &adapter->link_speed,
4211 &adapter->link_duplex);
4212 e1000_print_link_info(adapter);
4214 * On supported PHYs, check for duplex mismatch only
4215 * if link has autonegotiated at 10/100 half
4217 if ((hw->phy.type == e1000_phy_igp_3 ||
4218 hw->phy.type == e1000_phy_bm) &&
4219 (hw->mac.autoneg == true) &&
4220 (adapter->link_speed == SPEED_10 ||
4221 adapter->link_speed == SPEED_100) &&
4222 (adapter->link_duplex == HALF_DUPLEX)) {
4223 u16 autoneg_exp;
4225 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4227 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4228 e_info("Autonegotiated half duplex but"
4229 " link partner cannot autoneg. "
4230 " Try forcing full duplex if "
4231 "link gets many collisions.\n");
4234 /* adjust timeout factor according to speed/duplex */
4235 adapter->tx_timeout_factor = 1;
4236 switch (adapter->link_speed) {
4237 case SPEED_10:
4238 txb2b = 0;
4239 adapter->tx_timeout_factor = 16;
4240 break;
4241 case SPEED_100:
4242 txb2b = 0;
4243 adapter->tx_timeout_factor = 10;
4244 break;
4248 * workaround: re-program speed mode bit after
4249 * link-up event
4251 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4252 !txb2b) {
4253 u32 tarc0;
4254 tarc0 = er32(TARC(0));
4255 tarc0 &= ~SPEED_MODE_BIT;
4256 ew32(TARC(0), tarc0);
4260 * disable TSO for pcie and 10/100 speeds, to avoid
4261 * some hardware issues
4263 if (!(adapter->flags & FLAG_TSO_FORCE)) {
4264 switch (adapter->link_speed) {
4265 case SPEED_10:
4266 case SPEED_100:
4267 e_info("10/100 speed: disabling TSO\n");
4268 netdev->features &= ~NETIF_F_TSO;
4269 netdev->features &= ~NETIF_F_TSO6;
4270 break;
4271 case SPEED_1000:
4272 netdev->features |= NETIF_F_TSO;
4273 netdev->features |= NETIF_F_TSO6;
4274 break;
4275 default:
4276 /* oops */
4277 break;
4282 * enable transmits in the hardware, need to do this
4283 * after setting TARC(0)
4285 tctl = er32(TCTL);
4286 tctl |= E1000_TCTL_EN;
4287 ew32(TCTL, tctl);
4290 * Perform any post-link-up configuration before
4291 * reporting link up.
4293 if (phy->ops.cfg_on_link_up)
4294 phy->ops.cfg_on_link_up(hw);
4296 netif_carrier_on(netdev);
4298 if (!test_bit(__E1000_DOWN, &adapter->state))
4299 mod_timer(&adapter->phy_info_timer,
4300 round_jiffies(jiffies + 2 * HZ));
4302 } else {
4303 if (netif_carrier_ok(netdev)) {
4304 adapter->link_speed = 0;
4305 adapter->link_duplex = 0;
4306 /* Link status message must follow this format */
4307 printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4308 adapter->netdev->name);
4309 netif_carrier_off(netdev);
4310 if (!test_bit(__E1000_DOWN, &adapter->state))
4311 mod_timer(&adapter->phy_info_timer,
4312 round_jiffies(jiffies + 2 * HZ));
4314 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4315 schedule_work(&adapter->reset_task);
4316 else
4317 pm_schedule_suspend(netdev->dev.parent,
4318 LINK_TIMEOUT);
4322 link_up:
4323 e1000e_update_stats(adapter);
4325 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4326 adapter->tpt_old = adapter->stats.tpt;
4327 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4328 adapter->colc_old = adapter->stats.colc;
4330 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4331 adapter->gorc_old = adapter->stats.gorc;
4332 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4333 adapter->gotc_old = adapter->stats.gotc;
4335 e1000e_update_adaptive(&adapter->hw);
4337 if (!netif_carrier_ok(netdev)) {
4338 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
4339 tx_ring->count);
4340 if (tx_pending) {
4342 * We've lost link, so the controller stops DMA,
4343 * but we've got queued Tx work that's never going
4344 * to get done, so reset controller to flush Tx.
4345 * (Do the reset outside of interrupt context).
4347 schedule_work(&adapter->reset_task);
4348 /* return immediately since reset is imminent */
4349 return;
4353 /* Simple mode for Interrupt Throttle Rate (ITR) */
4354 if (adapter->itr_setting == 4) {
4356 * Symmetric Tx/Rx gets a reduced ITR=2000;
4357 * Total asymmetrical Tx or Rx gets ITR=8000;
4358 * everyone else is between 2000-8000.
4360 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4361 u32 dif = (adapter->gotc > adapter->gorc ?
4362 adapter->gotc - adapter->gorc :
4363 adapter->gorc - adapter->gotc) / 10000;
4364 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4366 ew32(ITR, 1000000000 / (itr * 256));
4369 /* Cause software interrupt to ensure Rx ring is cleaned */
4370 if (adapter->msix_entries)
4371 ew32(ICS, adapter->rx_ring->ims_val);
4372 else
4373 ew32(ICS, E1000_ICS_RXDMT0);
4375 /* flush pending descriptors to memory before detecting Tx hang */
4376 e1000e_flush_descriptors(adapter);
4378 /* Force detection of hung controller every watchdog period */
4379 adapter->detect_tx_hung = 1;
4382 * With 82571 controllers, LAA may be overwritten due to controller
4383 * reset from the other port. Set the appropriate LAA in RAR[0]
4385 if (e1000e_get_laa_state_82571(hw))
4386 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4388 if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4389 e1000e_check_82574_phy_workaround(adapter);
4391 /* Reset the timer */
4392 if (!test_bit(__E1000_DOWN, &adapter->state))
4393 mod_timer(&adapter->watchdog_timer,
4394 round_jiffies(jiffies + 2 * HZ));
4397 #define E1000_TX_FLAGS_CSUM 0x00000001
4398 #define E1000_TX_FLAGS_VLAN 0x00000002
4399 #define E1000_TX_FLAGS_TSO 0x00000004
4400 #define E1000_TX_FLAGS_IPV4 0x00000008
4401 #define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
4402 #define E1000_TX_FLAGS_VLAN_SHIFT 16
4404 static int e1000_tso(struct e1000_adapter *adapter,
4405 struct sk_buff *skb)
4407 struct e1000_ring *tx_ring = adapter->tx_ring;
4408 struct e1000_context_desc *context_desc;
4409 struct e1000_buffer *buffer_info;
4410 unsigned int i;
4411 u32 cmd_length = 0;
4412 u16 ipcse = 0, tucse, mss;
4413 u8 ipcss, ipcso, tucss, tucso, hdr_len;
4414 int err;
4416 if (!skb_is_gso(skb))
4417 return 0;
4419 if (skb_header_cloned(skb)) {
4420 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4421 if (err)
4422 return err;
4425 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4426 mss = skb_shinfo(skb)->gso_size;
4427 if (skb->protocol == htons(ETH_P_IP)) {
4428 struct iphdr *iph = ip_hdr(skb);
4429 iph->tot_len = 0;
4430 iph->check = 0;
4431 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4432 0, IPPROTO_TCP, 0);
4433 cmd_length = E1000_TXD_CMD_IP;
4434 ipcse = skb_transport_offset(skb) - 1;
4435 } else if (skb_is_gso_v6(skb)) {
4436 ipv6_hdr(skb)->payload_len = 0;
4437 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4438 &ipv6_hdr(skb)->daddr,
4439 0, IPPROTO_TCP, 0);
4440 ipcse = 0;
4442 ipcss = skb_network_offset(skb);
4443 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4444 tucss = skb_transport_offset(skb);
4445 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4446 tucse = 0;
4448 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4449 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4451 i = tx_ring->next_to_use;
4452 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4453 buffer_info = &tx_ring->buffer_info[i];
4455 context_desc->lower_setup.ip_fields.ipcss = ipcss;
4456 context_desc->lower_setup.ip_fields.ipcso = ipcso;
4457 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
4458 context_desc->upper_setup.tcp_fields.tucss = tucss;
4459 context_desc->upper_setup.tcp_fields.tucso = tucso;
4460 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4461 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
4462 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4463 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4465 buffer_info->time_stamp = jiffies;
4466 buffer_info->next_to_watch = i;
4468 i++;
4469 if (i == tx_ring->count)
4470 i = 0;
4471 tx_ring->next_to_use = i;
4473 return 1;
4476 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4478 struct e1000_ring *tx_ring = adapter->tx_ring;
4479 struct e1000_context_desc *context_desc;
4480 struct e1000_buffer *buffer_info;
4481 unsigned int i;
4482 u8 css;
4483 u32 cmd_len = E1000_TXD_CMD_DEXT;
4484 __be16 protocol;
4486 if (skb->ip_summed != CHECKSUM_PARTIAL)
4487 return 0;
4489 if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4490 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4491 else
4492 protocol = skb->protocol;
4494 switch (protocol) {
4495 case cpu_to_be16(ETH_P_IP):
4496 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4497 cmd_len |= E1000_TXD_CMD_TCP;
4498 break;
4499 case cpu_to_be16(ETH_P_IPV6):
4500 /* XXX not handling all IPV6 headers */
4501 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4502 cmd_len |= E1000_TXD_CMD_TCP;
4503 break;
4504 default:
4505 if (unlikely(net_ratelimit()))
4506 e_warn("checksum_partial proto=%x!\n",
4507 be16_to_cpu(protocol));
4508 break;
4511 css = skb_checksum_start_offset(skb);
4513 i = tx_ring->next_to_use;
4514 buffer_info = &tx_ring->buffer_info[i];
4515 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4517 context_desc->lower_setup.ip_config = 0;
4518 context_desc->upper_setup.tcp_fields.tucss = css;
4519 context_desc->upper_setup.tcp_fields.tucso =
4520 css + skb->csum_offset;
4521 context_desc->upper_setup.tcp_fields.tucse = 0;
4522 context_desc->tcp_seg_setup.data = 0;
4523 context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4525 buffer_info->time_stamp = jiffies;
4526 buffer_info->next_to_watch = i;
4528 i++;
4529 if (i == tx_ring->count)
4530 i = 0;
4531 tx_ring->next_to_use = i;
4533 return 1;
4536 #define E1000_MAX_PER_TXD 8192
4537 #define E1000_MAX_TXD_PWR 12
4539 static int e1000_tx_map(struct e1000_adapter *adapter,
4540 struct sk_buff *skb, unsigned int first,
4541 unsigned int max_per_txd, unsigned int nr_frags,
4542 unsigned int mss)
4544 struct e1000_ring *tx_ring = adapter->tx_ring;
4545 struct pci_dev *pdev = adapter->pdev;
4546 struct e1000_buffer *buffer_info;
4547 unsigned int len = skb_headlen(skb);
4548 unsigned int offset = 0, size, count = 0, i;
4549 unsigned int f, bytecount, segs;
4551 i = tx_ring->next_to_use;
4553 while (len) {
4554 buffer_info = &tx_ring->buffer_info[i];
4555 size = min(len, max_per_txd);
4557 buffer_info->length = size;
4558 buffer_info->time_stamp = jiffies;
4559 buffer_info->next_to_watch = i;
4560 buffer_info->dma = dma_map_single(&pdev->dev,
4561 skb->data + offset,
4562 size, DMA_TO_DEVICE);
4563 buffer_info->mapped_as_page = false;
4564 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4565 goto dma_error;
4567 len -= size;
4568 offset += size;
4569 count++;
4571 if (len) {
4572 i++;
4573 if (i == tx_ring->count)
4574 i = 0;
4578 for (f = 0; f < nr_frags; f++) {
4579 struct skb_frag_struct *frag;
4581 frag = &skb_shinfo(skb)->frags[f];
4582 len = frag->size;
4583 offset = frag->page_offset;
4585 while (len) {
4586 i++;
4587 if (i == tx_ring->count)
4588 i = 0;
4590 buffer_info = &tx_ring->buffer_info[i];
4591 size = min(len, max_per_txd);
4593 buffer_info->length = size;
4594 buffer_info->time_stamp = jiffies;
4595 buffer_info->next_to_watch = i;
4596 buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
4597 offset, size,
4598 DMA_TO_DEVICE);
4599 buffer_info->mapped_as_page = true;
4600 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4601 goto dma_error;
4603 len -= size;
4604 offset += size;
4605 count++;
4609 segs = skb_shinfo(skb)->gso_segs ? : 1;
4610 /* multiply data chunks by size of headers */
4611 bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4613 tx_ring->buffer_info[i].skb = skb;
4614 tx_ring->buffer_info[i].segs = segs;
4615 tx_ring->buffer_info[i].bytecount = bytecount;
4616 tx_ring->buffer_info[first].next_to_watch = i;
4618 return count;
4620 dma_error:
4621 dev_err(&pdev->dev, "Tx DMA map failed\n");
4622 buffer_info->dma = 0;
4623 if (count)
4624 count--;
4626 while (count--) {
4627 if (i == 0)
4628 i += tx_ring->count;
4629 i--;
4630 buffer_info = &tx_ring->buffer_info[i];
4631 e1000_put_txbuf(adapter, buffer_info);
4634 return 0;
4637 static void e1000_tx_queue(struct e1000_adapter *adapter,
4638 int tx_flags, int count)
4640 struct e1000_ring *tx_ring = adapter->tx_ring;
4641 struct e1000_tx_desc *tx_desc = NULL;
4642 struct e1000_buffer *buffer_info;
4643 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4644 unsigned int i;
4646 if (tx_flags & E1000_TX_FLAGS_TSO) {
4647 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4648 E1000_TXD_CMD_TSE;
4649 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4651 if (tx_flags & E1000_TX_FLAGS_IPV4)
4652 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4655 if (tx_flags & E1000_TX_FLAGS_CSUM) {
4656 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4657 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4660 if (tx_flags & E1000_TX_FLAGS_VLAN) {
4661 txd_lower |= E1000_TXD_CMD_VLE;
4662 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4665 i = tx_ring->next_to_use;
4667 do {
4668 buffer_info = &tx_ring->buffer_info[i];
4669 tx_desc = E1000_TX_DESC(*tx_ring, i);
4670 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4671 tx_desc->lower.data =
4672 cpu_to_le32(txd_lower | buffer_info->length);
4673 tx_desc->upper.data = cpu_to_le32(txd_upper);
4675 i++;
4676 if (i == tx_ring->count)
4677 i = 0;
4678 } while (--count > 0);
4680 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4683 * Force memory writes to complete before letting h/w
4684 * know there are new descriptors to fetch. (Only
4685 * applicable for weak-ordered memory model archs,
4686 * such as IA-64).
4688 wmb();
4690 tx_ring->next_to_use = i;
4691 writel(i, adapter->hw.hw_addr + tx_ring->tail);
4693 * we need this if more than one processor can write to our tail
4694 * at a time, it synchronizes IO on IA64/Altix systems
4696 mmiowb();
4699 #define MINIMUM_DHCP_PACKET_SIZE 282
4700 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4701 struct sk_buff *skb)
4703 struct e1000_hw *hw = &adapter->hw;
4704 u16 length, offset;
4706 if (vlan_tx_tag_present(skb)) {
4707 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4708 (adapter->hw.mng_cookie.status &
4709 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4710 return 0;
4713 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4714 return 0;
4716 if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4717 return 0;
4720 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4721 struct udphdr *udp;
4723 if (ip->protocol != IPPROTO_UDP)
4724 return 0;
4726 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4727 if (ntohs(udp->dest) != 67)
4728 return 0;
4730 offset = (u8 *)udp + 8 - skb->data;
4731 length = skb->len - offset;
4732 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4735 return 0;
4738 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4740 struct e1000_adapter *adapter = netdev_priv(netdev);
4742 netif_stop_queue(netdev);
4744 * Herbert's original patch had:
4745 * smp_mb__after_netif_stop_queue();
4746 * but since that doesn't exist yet, just open code it.
4748 smp_mb();
4751 * We need to check again in a case another CPU has just
4752 * made room available.
4754 if (e1000_desc_unused(adapter->tx_ring) < size)
4755 return -EBUSY;
4757 /* A reprieve! */
4758 netif_start_queue(netdev);
4759 ++adapter->restart_queue;
4760 return 0;
4763 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4765 struct e1000_adapter *adapter = netdev_priv(netdev);
4767 if (e1000_desc_unused(adapter->tx_ring) >= size)
4768 return 0;
4769 return __e1000_maybe_stop_tx(netdev, size);
4772 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4773 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4774 struct net_device *netdev)
4776 struct e1000_adapter *adapter = netdev_priv(netdev);
4777 struct e1000_ring *tx_ring = adapter->tx_ring;
4778 unsigned int first;
4779 unsigned int max_per_txd = E1000_MAX_PER_TXD;
4780 unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4781 unsigned int tx_flags = 0;
4782 unsigned int len = skb_headlen(skb);
4783 unsigned int nr_frags;
4784 unsigned int mss;
4785 int count = 0;
4786 int tso;
4787 unsigned int f;
4789 if (test_bit(__E1000_DOWN, &adapter->state)) {
4790 dev_kfree_skb_any(skb);
4791 return NETDEV_TX_OK;
4794 if (skb->len <= 0) {
4795 dev_kfree_skb_any(skb);
4796 return NETDEV_TX_OK;
4799 mss = skb_shinfo(skb)->gso_size;
4801 * The controller does a simple calculation to
4802 * make sure there is enough room in the FIFO before
4803 * initiating the DMA for each buffer. The calc is:
4804 * 4 = ceil(buffer len/mss). To make sure we don't
4805 * overrun the FIFO, adjust the max buffer len if mss
4806 * drops.
4808 if (mss) {
4809 u8 hdr_len;
4810 max_per_txd = min(mss << 2, max_per_txd);
4811 max_txd_pwr = fls(max_per_txd) - 1;
4814 * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4815 * points to just header, pull a few bytes of payload from
4816 * frags into skb->data
4818 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4820 * we do this workaround for ES2LAN, but it is un-necessary,
4821 * avoiding it could save a lot of cycles
4823 if (skb->data_len && (hdr_len == len)) {
4824 unsigned int pull_size;
4826 pull_size = min((unsigned int)4, skb->data_len);
4827 if (!__pskb_pull_tail(skb, pull_size)) {
4828 e_err("__pskb_pull_tail failed.\n");
4829 dev_kfree_skb_any(skb);
4830 return NETDEV_TX_OK;
4832 len = skb_headlen(skb);
4836 /* reserve a descriptor for the offload context */
4837 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4838 count++;
4839 count++;
4841 count += TXD_USE_COUNT(len, max_txd_pwr);
4843 nr_frags = skb_shinfo(skb)->nr_frags;
4844 for (f = 0; f < nr_frags; f++)
4845 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4846 max_txd_pwr);
4848 if (adapter->hw.mac.tx_pkt_filtering)
4849 e1000_transfer_dhcp_info(adapter, skb);
4852 * need: count + 2 desc gap to keep tail from touching
4853 * head, otherwise try next time
4855 if (e1000_maybe_stop_tx(netdev, count + 2))
4856 return NETDEV_TX_BUSY;
4858 if (vlan_tx_tag_present(skb)) {
4859 tx_flags |= E1000_TX_FLAGS_VLAN;
4860 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4863 first = tx_ring->next_to_use;
4865 tso = e1000_tso(adapter, skb);
4866 if (tso < 0) {
4867 dev_kfree_skb_any(skb);
4868 return NETDEV_TX_OK;
4871 if (tso)
4872 tx_flags |= E1000_TX_FLAGS_TSO;
4873 else if (e1000_tx_csum(adapter, skb))
4874 tx_flags |= E1000_TX_FLAGS_CSUM;
4877 * Old method was to assume IPv4 packet by default if TSO was enabled.
4878 * 82571 hardware supports TSO capabilities for IPv6 as well...
4879 * no longer assume, we must.
4881 if (skb->protocol == htons(ETH_P_IP))
4882 tx_flags |= E1000_TX_FLAGS_IPV4;
4884 /* if count is 0 then mapping error has occured */
4885 count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4886 if (count) {
4887 e1000_tx_queue(adapter, tx_flags, count);
4888 /* Make sure there is space in the ring for the next send. */
4889 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4891 } else {
4892 dev_kfree_skb_any(skb);
4893 tx_ring->buffer_info[first].time_stamp = 0;
4894 tx_ring->next_to_use = first;
4897 return NETDEV_TX_OK;
4901 * e1000_tx_timeout - Respond to a Tx Hang
4902 * @netdev: network interface device structure
4904 static void e1000_tx_timeout(struct net_device *netdev)
4906 struct e1000_adapter *adapter = netdev_priv(netdev);
4908 /* Do the reset outside of interrupt context */
4909 adapter->tx_timeout_count++;
4910 schedule_work(&adapter->reset_task);
4913 static void e1000_reset_task(struct work_struct *work)
4915 struct e1000_adapter *adapter;
4916 adapter = container_of(work, struct e1000_adapter, reset_task);
4918 /* don't run the task if already down */
4919 if (test_bit(__E1000_DOWN, &adapter->state))
4920 return;
4922 if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4923 (adapter->flags & FLAG_RX_RESTART_NOW))) {
4924 e1000e_dump(adapter);
4925 e_err("Reset adapter\n");
4927 e1000e_reinit_locked(adapter);
4931 * e1000_get_stats - Get System Network Statistics
4932 * @netdev: network interface device structure
4934 * Returns the address of the device statistics structure.
4935 * The statistics are actually updated from the timer callback.
4937 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4939 /* only return the current stats */
4940 return &netdev->stats;
4944 * e1000_change_mtu - Change the Maximum Transfer Unit
4945 * @netdev: network interface device structure
4946 * @new_mtu: new value for maximum frame size
4948 * Returns 0 on success, negative on failure
4950 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4952 struct e1000_adapter *adapter = netdev_priv(netdev);
4953 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4955 /* Jumbo frame support */
4956 if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4957 !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4958 e_err("Jumbo Frames not supported.\n");
4959 return -EINVAL;
4962 /* Supported frame sizes */
4963 if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4964 (max_frame > adapter->max_hw_frame_size)) {
4965 e_err("Unsupported MTU setting\n");
4966 return -EINVAL;
4969 /* Jumbo frame workaround on 82579 requires CRC be stripped */
4970 if ((adapter->hw.mac.type == e1000_pch2lan) &&
4971 !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
4972 (new_mtu > ETH_DATA_LEN)) {
4973 e_err("Jumbo Frames not supported on 82579 when CRC "
4974 "stripping is disabled.\n");
4975 return -EINVAL;
4978 /* 82573 Errata 17 */
4979 if (((adapter->hw.mac.type == e1000_82573) ||
4980 (adapter->hw.mac.type == e1000_82574)) &&
4981 (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
4982 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
4983 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
4986 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4987 msleep(1);
4988 /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
4989 adapter->max_frame_size = max_frame;
4990 e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4991 netdev->mtu = new_mtu;
4992 if (netif_running(netdev))
4993 e1000e_down(adapter);
4996 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4997 * means we reserve 2 more, this pushes us to allocate from the next
4998 * larger slab size.
4999 * i.e. RXBUFFER_2048 --> size-4096 slab
5000 * However with the new *_jumbo_rx* routines, jumbo receives will use
5001 * fragmented skbs
5004 if (max_frame <= 2048)
5005 adapter->rx_buffer_len = 2048;
5006 else
5007 adapter->rx_buffer_len = 4096;
5009 /* adjust allocation if LPE protects us, and we aren't using SBP */
5010 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5011 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5012 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5013 + ETH_FCS_LEN;
5015 if (netif_running(netdev))
5016 e1000e_up(adapter);
5017 else
5018 e1000e_reset(adapter);
5020 clear_bit(__E1000_RESETTING, &adapter->state);
5022 return 0;
5025 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5026 int cmd)
5028 struct e1000_adapter *adapter = netdev_priv(netdev);
5029 struct mii_ioctl_data *data = if_mii(ifr);
5031 if (adapter->hw.phy.media_type != e1000_media_type_copper)
5032 return -EOPNOTSUPP;
5034 switch (cmd) {
5035 case SIOCGMIIPHY:
5036 data->phy_id = adapter->hw.phy.addr;
5037 break;
5038 case SIOCGMIIREG:
5039 e1000_phy_read_status(adapter);
5041 switch (data->reg_num & 0x1F) {
5042 case MII_BMCR:
5043 data->val_out = adapter->phy_regs.bmcr;
5044 break;
5045 case MII_BMSR:
5046 data->val_out = adapter->phy_regs.bmsr;
5047 break;
5048 case MII_PHYSID1:
5049 data->val_out = (adapter->hw.phy.id >> 16);
5050 break;
5051 case MII_PHYSID2:
5052 data->val_out = (adapter->hw.phy.id & 0xFFFF);
5053 break;
5054 case MII_ADVERTISE:
5055 data->val_out = adapter->phy_regs.advertise;
5056 break;
5057 case MII_LPA:
5058 data->val_out = adapter->phy_regs.lpa;
5059 break;
5060 case MII_EXPANSION:
5061 data->val_out = adapter->phy_regs.expansion;
5062 break;
5063 case MII_CTRL1000:
5064 data->val_out = adapter->phy_regs.ctrl1000;
5065 break;
5066 case MII_STAT1000:
5067 data->val_out = adapter->phy_regs.stat1000;
5068 break;
5069 case MII_ESTATUS:
5070 data->val_out = adapter->phy_regs.estatus;
5071 break;
5072 default:
5073 return -EIO;
5075 break;
5076 case SIOCSMIIREG:
5077 default:
5078 return -EOPNOTSUPP;
5080 return 0;
5083 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5085 switch (cmd) {
5086 case SIOCGMIIPHY:
5087 case SIOCGMIIREG:
5088 case SIOCSMIIREG:
5089 return e1000_mii_ioctl(netdev, ifr, cmd);
5090 default:
5091 return -EOPNOTSUPP;
5095 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5097 struct e1000_hw *hw = &adapter->hw;
5098 u32 i, mac_reg;
5099 u16 phy_reg;
5100 int retval = 0;
5102 /* copy MAC RARs to PHY RARs */
5103 e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5105 /* copy MAC MTA to PHY MTA */
5106 for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5107 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5108 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
5109 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
5112 /* configure PHY Rx Control register */
5113 e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
5114 mac_reg = er32(RCTL);
5115 if (mac_reg & E1000_RCTL_UPE)
5116 phy_reg |= BM_RCTL_UPE;
5117 if (mac_reg & E1000_RCTL_MPE)
5118 phy_reg |= BM_RCTL_MPE;
5119 phy_reg &= ~(BM_RCTL_MO_MASK);
5120 if (mac_reg & E1000_RCTL_MO_3)
5121 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5122 << BM_RCTL_MO_SHIFT);
5123 if (mac_reg & E1000_RCTL_BAM)
5124 phy_reg |= BM_RCTL_BAM;
5125 if (mac_reg & E1000_RCTL_PMCF)
5126 phy_reg |= BM_RCTL_PMCF;
5127 mac_reg = er32(CTRL);
5128 if (mac_reg & E1000_CTRL_RFCE)
5129 phy_reg |= BM_RCTL_RFCE;
5130 e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
5132 /* enable PHY wakeup in MAC register */
5133 ew32(WUFC, wufc);
5134 ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5136 /* configure and enable PHY wakeup in PHY registers */
5137 e1e_wphy(&adapter->hw, BM_WUFC, wufc);
5138 e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5140 /* activate PHY wakeup */
5141 retval = hw->phy.ops.acquire(hw);
5142 if (retval) {
5143 e_err("Could not acquire PHY\n");
5144 return retval;
5146 e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
5147 (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
5148 retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
5149 if (retval) {
5150 e_err("Could not read PHY page 769\n");
5151 goto out;
5153 phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5154 retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
5155 if (retval)
5156 e_err("Could not set PHY Host Wakeup bit\n");
5157 out:
5158 hw->phy.ops.release(hw);
5160 return retval;
5163 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5164 bool runtime)
5166 struct net_device *netdev = pci_get_drvdata(pdev);
5167 struct e1000_adapter *adapter = netdev_priv(netdev);
5168 struct e1000_hw *hw = &adapter->hw;
5169 u32 ctrl, ctrl_ext, rctl, status;
5170 /* Runtime suspend should only enable wakeup for link changes */
5171 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5172 int retval = 0;
5174 netif_device_detach(netdev);
5176 if (netif_running(netdev)) {
5177 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5178 e1000e_down(adapter);
5179 e1000_free_irq(adapter);
5181 e1000e_reset_interrupt_capability(adapter);
5183 retval = pci_save_state(pdev);
5184 if (retval)
5185 return retval;
5187 status = er32(STATUS);
5188 if (status & E1000_STATUS_LU)
5189 wufc &= ~E1000_WUFC_LNKC;
5191 if (wufc) {
5192 e1000_setup_rctl(adapter);
5193 e1000_set_multi(netdev);
5195 /* turn on all-multi mode if wake on multicast is enabled */
5196 if (wufc & E1000_WUFC_MC) {
5197 rctl = er32(RCTL);
5198 rctl |= E1000_RCTL_MPE;
5199 ew32(RCTL, rctl);
5202 ctrl = er32(CTRL);
5203 /* advertise wake from D3Cold */
5204 #define E1000_CTRL_ADVD3WUC 0x00100000
5205 /* phy power management enable */
5206 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5207 ctrl |= E1000_CTRL_ADVD3WUC;
5208 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5209 ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5210 ew32(CTRL, ctrl);
5212 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5213 adapter->hw.phy.media_type ==
5214 e1000_media_type_internal_serdes) {
5215 /* keep the laser running in D3 */
5216 ctrl_ext = er32(CTRL_EXT);
5217 ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5218 ew32(CTRL_EXT, ctrl_ext);
5221 if (adapter->flags & FLAG_IS_ICH)
5222 e1000e_disable_gig_wol_ich8lan(&adapter->hw);
5224 /* Allow time for pending master requests to run */
5225 e1000e_disable_pcie_master(&adapter->hw);
5227 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5228 /* enable wakeup by the PHY */
5229 retval = e1000_init_phy_wakeup(adapter, wufc);
5230 if (retval)
5231 return retval;
5232 } else {
5233 /* enable wakeup by the MAC */
5234 ew32(WUFC, wufc);
5235 ew32(WUC, E1000_WUC_PME_EN);
5237 } else {
5238 ew32(WUC, 0);
5239 ew32(WUFC, 0);
5242 *enable_wake = !!wufc;
5244 /* make sure adapter isn't asleep if manageability is enabled */
5245 if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5246 (hw->mac.ops.check_mng_mode(hw)))
5247 *enable_wake = true;
5249 if (adapter->hw.phy.type == e1000_phy_igp_3)
5250 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5253 * Release control of h/w to f/w. If f/w is AMT enabled, this
5254 * would have already happened in close and is redundant.
5256 e1000e_release_hw_control(adapter);
5258 pci_disable_device(pdev);
5260 return 0;
5263 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5265 if (sleep && wake) {
5266 pci_prepare_to_sleep(pdev);
5267 return;
5270 pci_wake_from_d3(pdev, wake);
5271 pci_set_power_state(pdev, PCI_D3hot);
5274 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5275 bool wake)
5277 struct net_device *netdev = pci_get_drvdata(pdev);
5278 struct e1000_adapter *adapter = netdev_priv(netdev);
5281 * The pci-e switch on some quad port adapters will report a
5282 * correctable error when the MAC transitions from D0 to D3. To
5283 * prevent this we need to mask off the correctable errors on the
5284 * downstream port of the pci-e switch.
5286 if (adapter->flags & FLAG_IS_QUAD_PORT) {
5287 struct pci_dev *us_dev = pdev->bus->self;
5288 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
5289 u16 devctl;
5291 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5292 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5293 (devctl & ~PCI_EXP_DEVCTL_CERE));
5295 e1000_power_off(pdev, sleep, wake);
5297 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5298 } else {
5299 e1000_power_off(pdev, sleep, wake);
5303 #ifdef CONFIG_PCIEASPM
5304 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5306 pci_disable_link_state(pdev, state);
5308 #else
5309 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5311 int pos;
5312 u16 reg16;
5315 * Both device and parent should have the same ASPM setting.
5316 * Disable ASPM in downstream component first and then upstream.
5318 pos = pci_pcie_cap(pdev);
5319 pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5320 reg16 &= ~state;
5321 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5323 if (!pdev->bus->self)
5324 return;
5326 pos = pci_pcie_cap(pdev->bus->self);
5327 pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5328 reg16 &= ~state;
5329 pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5331 #endif
5332 void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5334 dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5335 (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5336 (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5338 __e1000e_disable_aspm(pdev, state);
5341 #ifdef CONFIG_PM_OPS
5342 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5344 return !!adapter->tx_ring->buffer_info;
5347 static int __e1000_resume(struct pci_dev *pdev)
5349 struct net_device *netdev = pci_get_drvdata(pdev);
5350 struct e1000_adapter *adapter = netdev_priv(netdev);
5351 struct e1000_hw *hw = &adapter->hw;
5352 u32 err;
5354 pci_set_power_state(pdev, PCI_D0);
5355 pci_restore_state(pdev);
5356 pci_save_state(pdev);
5357 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5358 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5360 e1000e_set_interrupt_capability(adapter);
5361 if (netif_running(netdev)) {
5362 err = e1000_request_irq(adapter);
5363 if (err)
5364 return err;
5367 e1000e_power_up_phy(adapter);
5369 /* report the system wakeup cause from S3/S4 */
5370 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5371 u16 phy_data;
5373 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5374 if (phy_data) {
5375 e_info("PHY Wakeup cause - %s\n",
5376 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5377 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5378 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5379 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5380 phy_data & E1000_WUS_LNKC ? "Link Status "
5381 " Change" : "other");
5383 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5384 } else {
5385 u32 wus = er32(WUS);
5386 if (wus) {
5387 e_info("MAC Wakeup cause - %s\n",
5388 wus & E1000_WUS_EX ? "Unicast Packet" :
5389 wus & E1000_WUS_MC ? "Multicast Packet" :
5390 wus & E1000_WUS_BC ? "Broadcast Packet" :
5391 wus & E1000_WUS_MAG ? "Magic Packet" :
5392 wus & E1000_WUS_LNKC ? "Link Status Change" :
5393 "other");
5395 ew32(WUS, ~0);
5398 e1000e_reset(adapter);
5400 e1000_init_manageability_pt(adapter);
5402 if (netif_running(netdev))
5403 e1000e_up(adapter);
5405 netif_device_attach(netdev);
5408 * If the controller has AMT, do not set DRV_LOAD until the interface
5409 * is up. For all other cases, let the f/w know that the h/w is now
5410 * under the control of the driver.
5412 if (!(adapter->flags & FLAG_HAS_AMT))
5413 e1000e_get_hw_control(adapter);
5415 return 0;
5418 #ifdef CONFIG_PM_SLEEP
5419 static int e1000_suspend(struct device *dev)
5421 struct pci_dev *pdev = to_pci_dev(dev);
5422 int retval;
5423 bool wake;
5425 retval = __e1000_shutdown(pdev, &wake, false);
5426 if (!retval)
5427 e1000_complete_shutdown(pdev, true, wake);
5429 return retval;
5432 static int e1000_resume(struct device *dev)
5434 struct pci_dev *pdev = to_pci_dev(dev);
5435 struct net_device *netdev = pci_get_drvdata(pdev);
5436 struct e1000_adapter *adapter = netdev_priv(netdev);
5438 if (e1000e_pm_ready(adapter))
5439 adapter->idle_check = true;
5441 return __e1000_resume(pdev);
5443 #endif /* CONFIG_PM_SLEEP */
5445 #ifdef CONFIG_PM_RUNTIME
5446 static int e1000_runtime_suspend(struct device *dev)
5448 struct pci_dev *pdev = to_pci_dev(dev);
5449 struct net_device *netdev = pci_get_drvdata(pdev);
5450 struct e1000_adapter *adapter = netdev_priv(netdev);
5452 if (e1000e_pm_ready(adapter)) {
5453 bool wake;
5455 __e1000_shutdown(pdev, &wake, true);
5458 return 0;
5461 static int e1000_idle(struct device *dev)
5463 struct pci_dev *pdev = to_pci_dev(dev);
5464 struct net_device *netdev = pci_get_drvdata(pdev);
5465 struct e1000_adapter *adapter = netdev_priv(netdev);
5467 if (!e1000e_pm_ready(adapter))
5468 return 0;
5470 if (adapter->idle_check) {
5471 adapter->idle_check = false;
5472 if (!e1000e_has_link(adapter))
5473 pm_schedule_suspend(dev, MSEC_PER_SEC);
5476 return -EBUSY;
5479 static int e1000_runtime_resume(struct device *dev)
5481 struct pci_dev *pdev = to_pci_dev(dev);
5482 struct net_device *netdev = pci_get_drvdata(pdev);
5483 struct e1000_adapter *adapter = netdev_priv(netdev);
5485 if (!e1000e_pm_ready(adapter))
5486 return 0;
5488 adapter->idle_check = !dev->power.runtime_auto;
5489 return __e1000_resume(pdev);
5491 #endif /* CONFIG_PM_RUNTIME */
5492 #endif /* CONFIG_PM_OPS */
5494 static void e1000_shutdown(struct pci_dev *pdev)
5496 bool wake = false;
5498 __e1000_shutdown(pdev, &wake, false);
5500 if (system_state == SYSTEM_POWER_OFF)
5501 e1000_complete_shutdown(pdev, false, wake);
5504 #ifdef CONFIG_NET_POLL_CONTROLLER
5506 static irqreturn_t e1000_intr_msix(int irq, void *data)
5508 struct net_device *netdev = data;
5509 struct e1000_adapter *adapter = netdev_priv(netdev);
5510 int vector, msix_irq;
5512 if (adapter->msix_entries) {
5513 vector = 0;
5514 msix_irq = adapter->msix_entries[vector].vector;
5515 disable_irq(msix_irq);
5516 e1000_intr_msix_rx(msix_irq, netdev);
5517 enable_irq(msix_irq);
5519 vector++;
5520 msix_irq = adapter->msix_entries[vector].vector;
5521 disable_irq(msix_irq);
5522 e1000_intr_msix_tx(msix_irq, netdev);
5523 enable_irq(msix_irq);
5525 vector++;
5526 msix_irq = adapter->msix_entries[vector].vector;
5527 disable_irq(msix_irq);
5528 e1000_msix_other(msix_irq, netdev);
5529 enable_irq(msix_irq);
5532 return IRQ_HANDLED;
5536 * Polling 'interrupt' - used by things like netconsole to send skbs
5537 * without having to re-enable interrupts. It's not called while
5538 * the interrupt routine is executing.
5540 static void e1000_netpoll(struct net_device *netdev)
5542 struct e1000_adapter *adapter = netdev_priv(netdev);
5544 switch (adapter->int_mode) {
5545 case E1000E_INT_MODE_MSIX:
5546 e1000_intr_msix(adapter->pdev->irq, netdev);
5547 break;
5548 case E1000E_INT_MODE_MSI:
5549 disable_irq(adapter->pdev->irq);
5550 e1000_intr_msi(adapter->pdev->irq, netdev);
5551 enable_irq(adapter->pdev->irq);
5552 break;
5553 default: /* E1000E_INT_MODE_LEGACY */
5554 disable_irq(adapter->pdev->irq);
5555 e1000_intr(adapter->pdev->irq, netdev);
5556 enable_irq(adapter->pdev->irq);
5557 break;
5560 #endif
5563 * e1000_io_error_detected - called when PCI error is detected
5564 * @pdev: Pointer to PCI device
5565 * @state: The current pci connection state
5567 * This function is called after a PCI bus error affecting
5568 * this device has been detected.
5570 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5571 pci_channel_state_t state)
5573 struct net_device *netdev = pci_get_drvdata(pdev);
5574 struct e1000_adapter *adapter = netdev_priv(netdev);
5576 netif_device_detach(netdev);
5578 if (state == pci_channel_io_perm_failure)
5579 return PCI_ERS_RESULT_DISCONNECT;
5581 if (netif_running(netdev))
5582 e1000e_down(adapter);
5583 pci_disable_device(pdev);
5585 /* Request a slot slot reset. */
5586 return PCI_ERS_RESULT_NEED_RESET;
5590 * e1000_io_slot_reset - called after the pci bus has been reset.
5591 * @pdev: Pointer to PCI device
5593 * Restart the card from scratch, as if from a cold-boot. Implementation
5594 * resembles the first-half of the e1000_resume routine.
5596 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5598 struct net_device *netdev = pci_get_drvdata(pdev);
5599 struct e1000_adapter *adapter = netdev_priv(netdev);
5600 struct e1000_hw *hw = &adapter->hw;
5601 int err;
5602 pci_ers_result_t result;
5604 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5605 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5606 err = pci_enable_device_mem(pdev);
5607 if (err) {
5608 dev_err(&pdev->dev,
5609 "Cannot re-enable PCI device after reset.\n");
5610 result = PCI_ERS_RESULT_DISCONNECT;
5611 } else {
5612 pci_set_master(pdev);
5613 pdev->state_saved = true;
5614 pci_restore_state(pdev);
5616 pci_enable_wake(pdev, PCI_D3hot, 0);
5617 pci_enable_wake(pdev, PCI_D3cold, 0);
5619 e1000e_reset(adapter);
5620 ew32(WUS, ~0);
5621 result = PCI_ERS_RESULT_RECOVERED;
5624 pci_cleanup_aer_uncorrect_error_status(pdev);
5626 return result;
5630 * e1000_io_resume - called when traffic can start flowing again.
5631 * @pdev: Pointer to PCI device
5633 * This callback is called when the error recovery driver tells us that
5634 * its OK to resume normal operation. Implementation resembles the
5635 * second-half of the e1000_resume routine.
5637 static void e1000_io_resume(struct pci_dev *pdev)
5639 struct net_device *netdev = pci_get_drvdata(pdev);
5640 struct e1000_adapter *adapter = netdev_priv(netdev);
5642 e1000_init_manageability_pt(adapter);
5644 if (netif_running(netdev)) {
5645 if (e1000e_up(adapter)) {
5646 dev_err(&pdev->dev,
5647 "can't bring device back up after reset\n");
5648 return;
5652 netif_device_attach(netdev);
5655 * If the controller has AMT, do not set DRV_LOAD until the interface
5656 * is up. For all other cases, let the f/w know that the h/w is now
5657 * under the control of the driver.
5659 if (!(adapter->flags & FLAG_HAS_AMT))
5660 e1000e_get_hw_control(adapter);
5664 static void e1000_print_device_info(struct e1000_adapter *adapter)
5666 struct e1000_hw *hw = &adapter->hw;
5667 struct net_device *netdev = adapter->netdev;
5668 u32 ret_val;
5669 u8 pba_str[E1000_PBANUM_LENGTH];
5671 /* print bus type/speed/width info */
5672 e_info("(PCI Express:2.5GB/s:%s) %pM\n",
5673 /* bus width */
5674 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5675 "Width x1"),
5676 /* MAC address */
5677 netdev->dev_addr);
5678 e_info("Intel(R) PRO/%s Network Connection\n",
5679 (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5680 ret_val = e1000_read_pba_string_generic(hw, pba_str,
5681 E1000_PBANUM_LENGTH);
5682 if (ret_val)
5683 strncpy((char *)pba_str, "Unknown", sizeof(pba_str) - 1);
5684 e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5685 hw->mac.type, hw->phy.type, pba_str);
5688 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5690 struct e1000_hw *hw = &adapter->hw;
5691 int ret_val;
5692 u16 buf = 0;
5694 if (hw->mac.type != e1000_82573)
5695 return;
5697 ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5698 if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5699 /* Deep Smart Power Down (DSPD) */
5700 dev_warn(&adapter->pdev->dev,
5701 "Warning: detected DSPD enabled in EEPROM\n");
5705 static const struct net_device_ops e1000e_netdev_ops = {
5706 .ndo_open = e1000_open,
5707 .ndo_stop = e1000_close,
5708 .ndo_start_xmit = e1000_xmit_frame,
5709 .ndo_get_stats = e1000_get_stats,
5710 .ndo_set_multicast_list = e1000_set_multi,
5711 .ndo_set_mac_address = e1000_set_mac,
5712 .ndo_change_mtu = e1000_change_mtu,
5713 .ndo_do_ioctl = e1000_ioctl,
5714 .ndo_tx_timeout = e1000_tx_timeout,
5715 .ndo_validate_addr = eth_validate_addr,
5717 .ndo_vlan_rx_register = e1000_vlan_rx_register,
5718 .ndo_vlan_rx_add_vid = e1000_vlan_rx_add_vid,
5719 .ndo_vlan_rx_kill_vid = e1000_vlan_rx_kill_vid,
5720 #ifdef CONFIG_NET_POLL_CONTROLLER
5721 .ndo_poll_controller = e1000_netpoll,
5722 #endif
5726 * e1000_probe - Device Initialization Routine
5727 * @pdev: PCI device information struct
5728 * @ent: entry in e1000_pci_tbl
5730 * Returns 0 on success, negative on failure
5732 * e1000_probe initializes an adapter identified by a pci_dev structure.
5733 * The OS initialization, configuring of the adapter private structure,
5734 * and a hardware reset occur.
5736 static int __devinit e1000_probe(struct pci_dev *pdev,
5737 const struct pci_device_id *ent)
5739 struct net_device *netdev;
5740 struct e1000_adapter *adapter;
5741 struct e1000_hw *hw;
5742 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5743 resource_size_t mmio_start, mmio_len;
5744 resource_size_t flash_start, flash_len;
5746 static int cards_found;
5747 int i, err, pci_using_dac;
5748 u16 eeprom_data = 0;
5749 u16 eeprom_apme_mask = E1000_EEPROM_APME;
5751 if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
5752 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5754 err = pci_enable_device_mem(pdev);
5755 if (err)
5756 return err;
5758 pci_using_dac = 0;
5759 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
5760 if (!err) {
5761 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
5762 if (!err)
5763 pci_using_dac = 1;
5764 } else {
5765 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
5766 if (err) {
5767 err = dma_set_coherent_mask(&pdev->dev,
5768 DMA_BIT_MASK(32));
5769 if (err) {
5770 dev_err(&pdev->dev, "No usable DMA "
5771 "configuration, aborting\n");
5772 goto err_dma;
5777 err = pci_request_selected_regions_exclusive(pdev,
5778 pci_select_bars(pdev, IORESOURCE_MEM),
5779 e1000e_driver_name);
5780 if (err)
5781 goto err_pci_reg;
5783 /* AER (Advanced Error Reporting) hooks */
5784 pci_enable_pcie_error_reporting(pdev);
5786 pci_set_master(pdev);
5787 /* PCI config space info */
5788 err = pci_save_state(pdev);
5789 if (err)
5790 goto err_alloc_etherdev;
5792 err = -ENOMEM;
5793 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5794 if (!netdev)
5795 goto err_alloc_etherdev;
5797 SET_NETDEV_DEV(netdev, &pdev->dev);
5799 netdev->irq = pdev->irq;
5801 pci_set_drvdata(pdev, netdev);
5802 adapter = netdev_priv(netdev);
5803 hw = &adapter->hw;
5804 adapter->netdev = netdev;
5805 adapter->pdev = pdev;
5806 adapter->ei = ei;
5807 adapter->pba = ei->pba;
5808 adapter->flags = ei->flags;
5809 adapter->flags2 = ei->flags2;
5810 adapter->hw.adapter = adapter;
5811 adapter->hw.mac.type = ei->mac;
5812 adapter->max_hw_frame_size = ei->max_hw_frame_size;
5813 adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5815 mmio_start = pci_resource_start(pdev, 0);
5816 mmio_len = pci_resource_len(pdev, 0);
5818 err = -EIO;
5819 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5820 if (!adapter->hw.hw_addr)
5821 goto err_ioremap;
5823 if ((adapter->flags & FLAG_HAS_FLASH) &&
5824 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5825 flash_start = pci_resource_start(pdev, 1);
5826 flash_len = pci_resource_len(pdev, 1);
5827 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5828 if (!adapter->hw.flash_address)
5829 goto err_flashmap;
5832 /* construct the net_device struct */
5833 netdev->netdev_ops = &e1000e_netdev_ops;
5834 e1000e_set_ethtool_ops(netdev);
5835 netdev->watchdog_timeo = 5 * HZ;
5836 netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5837 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5839 netdev->mem_start = mmio_start;
5840 netdev->mem_end = mmio_start + mmio_len;
5842 adapter->bd_number = cards_found++;
5844 e1000e_check_options(adapter);
5846 /* setup adapter struct */
5847 err = e1000_sw_init(adapter);
5848 if (err)
5849 goto err_sw_init;
5851 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5852 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5853 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5855 err = ei->get_variants(adapter);
5856 if (err)
5857 goto err_hw_init;
5859 if ((adapter->flags & FLAG_IS_ICH) &&
5860 (adapter->flags & FLAG_READ_ONLY_NVM))
5861 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5863 hw->mac.ops.get_bus_info(&adapter->hw);
5865 adapter->hw.phy.autoneg_wait_to_complete = 0;
5867 /* Copper options */
5868 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5869 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5870 adapter->hw.phy.disable_polarity_correction = 0;
5871 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5874 if (e1000_check_reset_block(&adapter->hw))
5875 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5877 netdev->features = NETIF_F_SG |
5878 NETIF_F_HW_CSUM |
5879 NETIF_F_HW_VLAN_TX |
5880 NETIF_F_HW_VLAN_RX;
5882 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5883 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5885 netdev->features |= NETIF_F_TSO;
5886 netdev->features |= NETIF_F_TSO6;
5888 netdev->vlan_features |= NETIF_F_TSO;
5889 netdev->vlan_features |= NETIF_F_TSO6;
5890 netdev->vlan_features |= NETIF_F_HW_CSUM;
5891 netdev->vlan_features |= NETIF_F_SG;
5893 if (pci_using_dac) {
5894 netdev->features |= NETIF_F_HIGHDMA;
5895 netdev->vlan_features |= NETIF_F_HIGHDMA;
5898 if (e1000e_enable_mng_pass_thru(&adapter->hw))
5899 adapter->flags |= FLAG_MNG_PT_ENABLED;
5902 * before reading the NVM, reset the controller to
5903 * put the device in a known good starting state
5905 adapter->hw.mac.ops.reset_hw(&adapter->hw);
5908 * systems with ASPM and others may see the checksum fail on the first
5909 * attempt. Let's give it a few tries
5911 for (i = 0;; i++) {
5912 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5913 break;
5914 if (i == 2) {
5915 e_err("The NVM Checksum Is Not Valid\n");
5916 err = -EIO;
5917 goto err_eeprom;
5921 e1000_eeprom_checks(adapter);
5923 /* copy the MAC address */
5924 if (e1000e_read_mac_addr(&adapter->hw))
5925 e_err("NVM Read Error while reading MAC address\n");
5927 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5928 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5930 if (!is_valid_ether_addr(netdev->perm_addr)) {
5931 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5932 err = -EIO;
5933 goto err_eeprom;
5936 init_timer(&adapter->watchdog_timer);
5937 adapter->watchdog_timer.function = e1000_watchdog;
5938 adapter->watchdog_timer.data = (unsigned long) adapter;
5940 init_timer(&adapter->phy_info_timer);
5941 adapter->phy_info_timer.function = e1000_update_phy_info;
5942 adapter->phy_info_timer.data = (unsigned long) adapter;
5944 INIT_WORK(&adapter->reset_task, e1000_reset_task);
5945 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5946 INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5947 INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5948 INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5949 INIT_WORK(&adapter->led_blink_task, e1000e_led_blink_task);
5951 /* Initialize link parameters. User can change them with ethtool */
5952 adapter->hw.mac.autoneg = 1;
5953 adapter->fc_autoneg = 1;
5954 adapter->hw.fc.requested_mode = e1000_fc_default;
5955 adapter->hw.fc.current_mode = e1000_fc_default;
5956 adapter->hw.phy.autoneg_advertised = 0x2f;
5958 /* ring size defaults */
5959 adapter->rx_ring->count = 256;
5960 adapter->tx_ring->count = 256;
5963 * Initial Wake on LAN setting - If APM wake is enabled in
5964 * the EEPROM, enable the ACPI Magic Packet filter
5966 if (adapter->flags & FLAG_APME_IN_WUC) {
5967 /* APME bit in EEPROM is mapped to WUC.APME */
5968 eeprom_data = er32(WUC);
5969 eeprom_apme_mask = E1000_WUC_APME;
5970 if ((hw->mac.type > e1000_ich10lan) &&
5971 (eeprom_data & E1000_WUC_PHY_WAKE))
5972 adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5973 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5974 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5975 (adapter->hw.bus.func == 1))
5976 e1000_read_nvm(&adapter->hw,
5977 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5978 else
5979 e1000_read_nvm(&adapter->hw,
5980 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5983 /* fetch WoL from EEPROM */
5984 if (eeprom_data & eeprom_apme_mask)
5985 adapter->eeprom_wol |= E1000_WUFC_MAG;
5988 * now that we have the eeprom settings, apply the special cases
5989 * where the eeprom may be wrong or the board simply won't support
5990 * wake on lan on a particular port
5992 if (!(adapter->flags & FLAG_HAS_WOL))
5993 adapter->eeprom_wol = 0;
5995 /* initialize the wol settings based on the eeprom settings */
5996 adapter->wol = adapter->eeprom_wol;
5997 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5999 /* save off EEPROM version number */
6000 e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6002 /* reset the hardware with the new settings */
6003 e1000e_reset(adapter);
6006 * If the controller has AMT, do not set DRV_LOAD until the interface
6007 * is up. For all other cases, let the f/w know that the h/w is now
6008 * under the control of the driver.
6010 if (!(adapter->flags & FLAG_HAS_AMT))
6011 e1000e_get_hw_control(adapter);
6013 strncpy(netdev->name, "eth%d", sizeof(netdev->name) - 1);
6014 err = register_netdev(netdev);
6015 if (err)
6016 goto err_register;
6018 /* carrier off reporting is important to ethtool even BEFORE open */
6019 netif_carrier_off(netdev);
6021 e1000_print_device_info(adapter);
6023 if (pci_dev_run_wake(pdev))
6024 pm_runtime_put_noidle(&pdev->dev);
6026 return 0;
6028 err_register:
6029 if (!(adapter->flags & FLAG_HAS_AMT))
6030 e1000e_release_hw_control(adapter);
6031 err_eeprom:
6032 if (!e1000_check_reset_block(&adapter->hw))
6033 e1000_phy_hw_reset(&adapter->hw);
6034 err_hw_init:
6035 kfree(adapter->tx_ring);
6036 kfree(adapter->rx_ring);
6037 err_sw_init:
6038 if (adapter->hw.flash_address)
6039 iounmap(adapter->hw.flash_address);
6040 e1000e_reset_interrupt_capability(adapter);
6041 err_flashmap:
6042 iounmap(adapter->hw.hw_addr);
6043 err_ioremap:
6044 free_netdev(netdev);
6045 err_alloc_etherdev:
6046 pci_release_selected_regions(pdev,
6047 pci_select_bars(pdev, IORESOURCE_MEM));
6048 err_pci_reg:
6049 err_dma:
6050 pci_disable_device(pdev);
6051 return err;
6055 * e1000_remove - Device Removal Routine
6056 * @pdev: PCI device information struct
6058 * e1000_remove is called by the PCI subsystem to alert the driver
6059 * that it should release a PCI device. The could be caused by a
6060 * Hot-Plug event, or because the driver is going to be removed from
6061 * memory.
6063 static void __devexit e1000_remove(struct pci_dev *pdev)
6065 struct net_device *netdev = pci_get_drvdata(pdev);
6066 struct e1000_adapter *adapter = netdev_priv(netdev);
6067 bool down = test_bit(__E1000_DOWN, &adapter->state);
6070 * The timers may be rescheduled, so explicitly disable them
6071 * from being rescheduled.
6073 if (!down)
6074 set_bit(__E1000_DOWN, &adapter->state);
6075 del_timer_sync(&adapter->watchdog_timer);
6076 del_timer_sync(&adapter->phy_info_timer);
6078 cancel_work_sync(&adapter->reset_task);
6079 cancel_work_sync(&adapter->watchdog_task);
6080 cancel_work_sync(&adapter->downshift_task);
6081 cancel_work_sync(&adapter->update_phy_task);
6082 cancel_work_sync(&adapter->led_blink_task);
6083 cancel_work_sync(&adapter->print_hang_task);
6085 if (!(netdev->flags & IFF_UP))
6086 e1000_power_down_phy(adapter);
6088 /* Don't lie to e1000_close() down the road. */
6089 if (!down)
6090 clear_bit(__E1000_DOWN, &adapter->state);
6091 unregister_netdev(netdev);
6093 if (pci_dev_run_wake(pdev))
6094 pm_runtime_get_noresume(&pdev->dev);
6097 * Release control of h/w to f/w. If f/w is AMT enabled, this
6098 * would have already happened in close and is redundant.
6100 e1000e_release_hw_control(adapter);
6102 e1000e_reset_interrupt_capability(adapter);
6103 kfree(adapter->tx_ring);
6104 kfree(adapter->rx_ring);
6106 iounmap(adapter->hw.hw_addr);
6107 if (adapter->hw.flash_address)
6108 iounmap(adapter->hw.flash_address);
6109 pci_release_selected_regions(pdev,
6110 pci_select_bars(pdev, IORESOURCE_MEM));
6112 free_netdev(netdev);
6114 /* AER disable */
6115 pci_disable_pcie_error_reporting(pdev);
6117 pci_disable_device(pdev);
6120 /* PCI Error Recovery (ERS) */
6121 static struct pci_error_handlers e1000_err_handler = {
6122 .error_detected = e1000_io_error_detected,
6123 .slot_reset = e1000_io_slot_reset,
6124 .resume = e1000_io_resume,
6127 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6128 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6129 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6130 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6131 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6132 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6133 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6134 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6135 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6136 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6138 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6139 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6140 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6141 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6143 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6144 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6145 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6147 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6148 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6149 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6151 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6152 board_80003es2lan },
6153 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6154 board_80003es2lan },
6155 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6156 board_80003es2lan },
6157 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6158 board_80003es2lan },
6160 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6161 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6162 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6163 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6164 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6165 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6166 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6167 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6169 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6170 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6171 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6172 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6173 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6174 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6175 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6176 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6177 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6179 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6180 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6181 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6183 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6184 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6185 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6187 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6188 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6189 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6190 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6192 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6193 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6195 { } /* terminate list */
6197 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6199 #ifdef CONFIG_PM_OPS
6200 static const struct dev_pm_ops e1000_pm_ops = {
6201 SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6202 SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6203 e1000_runtime_resume, e1000_idle)
6205 #endif
6207 /* PCI Device API Driver */
6208 static struct pci_driver e1000_driver = {
6209 .name = e1000e_driver_name,
6210 .id_table = e1000_pci_tbl,
6211 .probe = e1000_probe,
6212 .remove = __devexit_p(e1000_remove),
6213 #ifdef CONFIG_PM_OPS
6214 .driver.pm = &e1000_pm_ops,
6215 #endif
6216 .shutdown = e1000_shutdown,
6217 .err_handler = &e1000_err_handler
6221 * e1000_init_module - Driver Registration Routine
6223 * e1000_init_module is the first routine called when the driver is
6224 * loaded. All it does is register with the PCI subsystem.
6226 static int __init e1000_init_module(void)
6228 int ret;
6229 pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6230 e1000e_driver_version);
6231 pr_info("Copyright(c) 1999 - 2011 Intel Corporation.\n");
6232 ret = pci_register_driver(&e1000_driver);
6234 return ret;
6236 module_init(e1000_init_module);
6239 * e1000_exit_module - Driver Exit Cleanup Routine
6241 * e1000_exit_module is called just before the driver is removed
6242 * from memory.
6244 static void __exit e1000_exit_module(void)
6246 pci_unregister_driver(&e1000_driver);
6248 module_exit(e1000_exit_module);
6251 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6252 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6253 MODULE_LICENSE("GPL");
6254 MODULE_VERSION(DRV_VERSION);
6256 /* e1000_main.c */