ath5k: Use new dma_stop function on base.c
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / net / e1000e / netdev.c
blobc4ca1629f532e49b502c78b76e0e99226c728379
1 /*******************************************************************************
3 Intel PRO/1000 Linux driver
4 Copyright(c) 1999 - 2010 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 "1.2.7" 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");
176 * e1000e_dump - Print registers, tx-ring and rx-ring
178 static void e1000e_dump(struct e1000_adapter *adapter)
180 struct net_device *netdev = adapter->netdev;
181 struct e1000_hw *hw = &adapter->hw;
182 struct e1000_reg_info *reginfo;
183 struct e1000_ring *tx_ring = adapter->tx_ring;
184 struct e1000_tx_desc *tx_desc;
185 struct my_u0 { u64 a; u64 b; } *u0;
186 struct e1000_buffer *buffer_info;
187 struct e1000_ring *rx_ring = adapter->rx_ring;
188 union e1000_rx_desc_packet_split *rx_desc_ps;
189 struct e1000_rx_desc *rx_desc;
190 struct my_u1 { u64 a; u64 b; u64 c; u64 d; } *u1;
191 u32 staterr;
192 int i = 0;
194 if (!netif_msg_hw(adapter))
195 return;
197 /* Print netdevice Info */
198 if (netdev) {
199 dev_info(&adapter->pdev->dev, "Net device Info\n");
200 printk(KERN_INFO "Device Name state "
201 "trans_start last_rx\n");
202 printk(KERN_INFO "%-15s %016lX %016lX %016lX\n",
203 netdev->name,
204 netdev->state,
205 netdev->trans_start,
206 netdev->last_rx);
209 /* Print Registers */
210 dev_info(&adapter->pdev->dev, "Register Dump\n");
211 printk(KERN_INFO " Register Name Value\n");
212 for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
213 reginfo->name; reginfo++) {
214 e1000_regdump(hw, reginfo);
217 /* Print TX Ring Summary */
218 if (!netdev || !netif_running(netdev))
219 goto exit;
221 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
222 printk(KERN_INFO "Queue [NTU] [NTC] [bi(ntc)->dma ]"
223 " leng ntw timestamp\n");
224 buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
225 printk(KERN_INFO " %5d %5X %5X %016llX %04X %3X %016llX\n",
226 0, tx_ring->next_to_use, tx_ring->next_to_clean,
227 (unsigned long long)buffer_info->dma,
228 buffer_info->length,
229 buffer_info->next_to_watch,
230 (unsigned long long)buffer_info->time_stamp);
232 /* Print TX Rings */
233 if (!netif_msg_tx_done(adapter))
234 goto rx_ring_summary;
236 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
238 /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
240 * Legacy Transmit Descriptor
241 * +--------------------------------------------------------------+
242 * 0 | Buffer Address [63:0] (Reserved on Write Back) |
243 * +--------------------------------------------------------------+
244 * 8 | Special | CSS | Status | CMD | CSO | Length |
245 * +--------------------------------------------------------------+
246 * 63 48 47 36 35 32 31 24 23 16 15 0
248 * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
249 * 63 48 47 40 39 32 31 16 15 8 7 0
250 * +----------------------------------------------------------------+
251 * 0 | TUCSE | TUCS0 | TUCSS | IPCSE | IPCS0 | IPCSS |
252 * +----------------------------------------------------------------+
253 * 8 | MSS | HDRLEN | RSV | STA | TUCMD | DTYP | PAYLEN |
254 * +----------------------------------------------------------------+
255 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
257 * Extended Data Descriptor (DTYP=0x1)
258 * +----------------------------------------------------------------+
259 * 0 | Buffer Address [63:0] |
260 * +----------------------------------------------------------------+
261 * 8 | VLAN tag | POPTS | Rsvd | Status | Command | DTYP | DTALEN |
262 * +----------------------------------------------------------------+
263 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
265 printk(KERN_INFO "Tl[desc] [address 63:0 ] [SpeCssSCmCsLen]"
266 " [bi->dma ] leng ntw timestamp bi->skb "
267 "<-- Legacy format\n");
268 printk(KERN_INFO "Tc[desc] [Ce CoCsIpceCoS] [MssHlRSCm0Plen]"
269 " [bi->dma ] leng ntw timestamp bi->skb "
270 "<-- Ext Context format\n");
271 printk(KERN_INFO "Td[desc] [address 63:0 ] [VlaPoRSCm1Dlen]"
272 " [bi->dma ] leng ntw timestamp bi->skb "
273 "<-- Ext Data format\n");
274 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
275 tx_desc = E1000_TX_DESC(*tx_ring, i);
276 buffer_info = &tx_ring->buffer_info[i];
277 u0 = (struct my_u0 *)tx_desc;
278 printk(KERN_INFO "T%c[0x%03X] %016llX %016llX %016llX "
279 "%04X %3X %016llX %p",
280 (!(le64_to_cpu(u0->b) & (1<<29)) ? 'l' :
281 ((le64_to_cpu(u0->b) & (1<<20)) ? 'd' : 'c')), i,
282 (unsigned long long)le64_to_cpu(u0->a),
283 (unsigned long long)le64_to_cpu(u0->b),
284 (unsigned long long)buffer_info->dma,
285 buffer_info->length, buffer_info->next_to_watch,
286 (unsigned long long)buffer_info->time_stamp,
287 buffer_info->skb);
288 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
289 printk(KERN_CONT " NTC/U\n");
290 else if (i == tx_ring->next_to_use)
291 printk(KERN_CONT " NTU\n");
292 else if (i == tx_ring->next_to_clean)
293 printk(KERN_CONT " NTC\n");
294 else
295 printk(KERN_CONT "\n");
297 if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
298 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
299 16, 1, phys_to_virt(buffer_info->dma),
300 buffer_info->length, true);
303 /* Print RX Rings Summary */
304 rx_ring_summary:
305 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
306 printk(KERN_INFO "Queue [NTU] [NTC]\n");
307 printk(KERN_INFO " %5d %5X %5X\n", 0,
308 rx_ring->next_to_use, rx_ring->next_to_clean);
310 /* Print RX Rings */
311 if (!netif_msg_rx_status(adapter))
312 goto exit;
314 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
315 switch (adapter->rx_ps_pages) {
316 case 1:
317 case 2:
318 case 3:
319 /* [Extended] Packet Split Receive Descriptor Format
321 * +-----------------------------------------------------+
322 * 0 | Buffer Address 0 [63:0] |
323 * +-----------------------------------------------------+
324 * 8 | Buffer Address 1 [63:0] |
325 * +-----------------------------------------------------+
326 * 16 | Buffer Address 2 [63:0] |
327 * +-----------------------------------------------------+
328 * 24 | Buffer Address 3 [63:0] |
329 * +-----------------------------------------------------+
331 printk(KERN_INFO "R [desc] [buffer 0 63:0 ] "
332 "[buffer 1 63:0 ] "
333 "[buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma ] "
334 "[bi->skb] <-- Ext Pkt Split format\n");
335 /* [Extended] Receive Descriptor (Write-Back) Format
337 * 63 48 47 32 31 13 12 8 7 4 3 0
338 * +------------------------------------------------------+
339 * 0 | Packet | IP | Rsvd | MRQ | Rsvd | MRQ RSS |
340 * | Checksum | Ident | | Queue | | Type |
341 * +------------------------------------------------------+
342 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
343 * +------------------------------------------------------+
344 * 63 48 47 32 31 20 19 0
346 printk(KERN_INFO "RWB[desc] [ck ipid mrqhsh] "
347 "[vl l0 ee es] "
348 "[ l3 l2 l1 hs] [reserved ] ---------------- "
349 "[bi->skb] <-- Ext Rx Write-Back format\n");
350 for (i = 0; i < rx_ring->count; i++) {
351 buffer_info = &rx_ring->buffer_info[i];
352 rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
353 u1 = (struct my_u1 *)rx_desc_ps;
354 staterr =
355 le32_to_cpu(rx_desc_ps->wb.middle.status_error);
356 if (staterr & E1000_RXD_STAT_DD) {
357 /* Descriptor Done */
358 printk(KERN_INFO "RWB[0x%03X] %016llX "
359 "%016llX %016llX %016llX "
360 "---------------- %p", i,
361 (unsigned long long)le64_to_cpu(u1->a),
362 (unsigned long long)le64_to_cpu(u1->b),
363 (unsigned long long)le64_to_cpu(u1->c),
364 (unsigned long long)le64_to_cpu(u1->d),
365 buffer_info->skb);
366 } else {
367 printk(KERN_INFO "R [0x%03X] %016llX "
368 "%016llX %016llX %016llX %016llX %p", i,
369 (unsigned long long)le64_to_cpu(u1->a),
370 (unsigned long long)le64_to_cpu(u1->b),
371 (unsigned long long)le64_to_cpu(u1->c),
372 (unsigned long long)le64_to_cpu(u1->d),
373 (unsigned long long)buffer_info->dma,
374 buffer_info->skb);
376 if (netif_msg_pktdata(adapter))
377 print_hex_dump(KERN_INFO, "",
378 DUMP_PREFIX_ADDRESS, 16, 1,
379 phys_to_virt(buffer_info->dma),
380 adapter->rx_ps_bsize0, true);
383 if (i == rx_ring->next_to_use)
384 printk(KERN_CONT " NTU\n");
385 else if (i == rx_ring->next_to_clean)
386 printk(KERN_CONT " NTC\n");
387 else
388 printk(KERN_CONT "\n");
390 break;
391 default:
392 case 0:
393 /* Legacy Receive Descriptor Format
395 * +-----------------------------------------------------+
396 * | Buffer Address [63:0] |
397 * +-----------------------------------------------------+
398 * | VLAN Tag | Errors | Status 0 | Packet csum | Length |
399 * +-----------------------------------------------------+
400 * 63 48 47 40 39 32 31 16 15 0
402 printk(KERN_INFO "Rl[desc] [address 63:0 ] "
403 "[vl er S cks ln] [bi->dma ] [bi->skb] "
404 "<-- Legacy format\n");
405 for (i = 0; rx_ring->desc && (i < rx_ring->count); i++) {
406 rx_desc = E1000_RX_DESC(*rx_ring, i);
407 buffer_info = &rx_ring->buffer_info[i];
408 u0 = (struct my_u0 *)rx_desc;
409 printk(KERN_INFO "Rl[0x%03X] %016llX %016llX "
410 "%016llX %p", i,
411 (unsigned long long)le64_to_cpu(u0->a),
412 (unsigned long long)le64_to_cpu(u0->b),
413 (unsigned long long)buffer_info->dma,
414 buffer_info->skb);
415 if (i == rx_ring->next_to_use)
416 printk(KERN_CONT " NTU\n");
417 else if (i == rx_ring->next_to_clean)
418 printk(KERN_CONT " NTC\n");
419 else
420 printk(KERN_CONT "\n");
422 if (netif_msg_pktdata(adapter))
423 print_hex_dump(KERN_INFO, "",
424 DUMP_PREFIX_ADDRESS,
425 16, 1, phys_to_virt(buffer_info->dma),
426 adapter->rx_buffer_len, true);
430 exit:
431 return;
435 * e1000_desc_unused - calculate if we have unused descriptors
437 static int e1000_desc_unused(struct e1000_ring *ring)
439 if (ring->next_to_clean > ring->next_to_use)
440 return ring->next_to_clean - ring->next_to_use - 1;
442 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
446 * e1000_receive_skb - helper function to handle Rx indications
447 * @adapter: board private structure
448 * @status: descriptor status field as written by hardware
449 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
450 * @skb: pointer to sk_buff to be indicated to stack
452 static void e1000_receive_skb(struct e1000_adapter *adapter,
453 struct net_device *netdev,
454 struct sk_buff *skb,
455 u8 status, __le16 vlan)
457 skb->protocol = eth_type_trans(skb, netdev);
459 if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
460 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
461 le16_to_cpu(vlan), skb);
462 else
463 napi_gro_receive(&adapter->napi, skb);
467 * e1000_rx_checksum - Receive Checksum Offload for 82543
468 * @adapter: board private structure
469 * @status_err: receive descriptor status and error fields
470 * @csum: receive descriptor csum field
471 * @sk_buff: socket buffer with received data
473 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
474 u32 csum, struct sk_buff *skb)
476 u16 status = (u16)status_err;
477 u8 errors = (u8)(status_err >> 24);
479 skb_checksum_none_assert(skb);
481 /* Ignore Checksum bit is set */
482 if (status & E1000_RXD_STAT_IXSM)
483 return;
484 /* TCP/UDP checksum error bit is set */
485 if (errors & E1000_RXD_ERR_TCPE) {
486 /* let the stack verify checksum errors */
487 adapter->hw_csum_err++;
488 return;
491 /* TCP/UDP Checksum has not been calculated */
492 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
493 return;
495 /* It must be a TCP or UDP packet with a valid checksum */
496 if (status & E1000_RXD_STAT_TCPCS) {
497 /* TCP checksum is good */
498 skb->ip_summed = CHECKSUM_UNNECESSARY;
499 } else {
501 * IP fragment with UDP payload
502 * Hardware complements the payload checksum, so we undo it
503 * and then put the value in host order for further stack use.
505 __sum16 sum = (__force __sum16)htons(csum);
506 skb->csum = csum_unfold(~sum);
507 skb->ip_summed = CHECKSUM_COMPLETE;
509 adapter->hw_csum_good++;
513 * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
514 * @adapter: address of board private structure
516 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
517 int cleaned_count)
519 struct net_device *netdev = adapter->netdev;
520 struct pci_dev *pdev = adapter->pdev;
521 struct e1000_ring *rx_ring = adapter->rx_ring;
522 struct e1000_rx_desc *rx_desc;
523 struct e1000_buffer *buffer_info;
524 struct sk_buff *skb;
525 unsigned int i;
526 unsigned int bufsz = adapter->rx_buffer_len;
528 i = rx_ring->next_to_use;
529 buffer_info = &rx_ring->buffer_info[i];
531 while (cleaned_count--) {
532 skb = buffer_info->skb;
533 if (skb) {
534 skb_trim(skb, 0);
535 goto map_skb;
538 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
539 if (!skb) {
540 /* Better luck next round */
541 adapter->alloc_rx_buff_failed++;
542 break;
545 buffer_info->skb = skb;
546 map_skb:
547 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
548 adapter->rx_buffer_len,
549 DMA_FROM_DEVICE);
550 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
551 dev_err(&pdev->dev, "RX DMA map failed\n");
552 adapter->rx_dma_failed++;
553 break;
556 rx_desc = E1000_RX_DESC(*rx_ring, i);
557 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
559 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
561 * Force memory writes to complete before letting h/w
562 * know there are new descriptors to fetch. (Only
563 * applicable for weak-ordered memory model archs,
564 * such as IA-64).
566 wmb();
567 writel(i, adapter->hw.hw_addr + rx_ring->tail);
569 i++;
570 if (i == rx_ring->count)
571 i = 0;
572 buffer_info = &rx_ring->buffer_info[i];
575 rx_ring->next_to_use = i;
579 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
580 * @adapter: address of board private structure
582 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
583 int cleaned_count)
585 struct net_device *netdev = adapter->netdev;
586 struct pci_dev *pdev = adapter->pdev;
587 union e1000_rx_desc_packet_split *rx_desc;
588 struct e1000_ring *rx_ring = adapter->rx_ring;
589 struct e1000_buffer *buffer_info;
590 struct e1000_ps_page *ps_page;
591 struct sk_buff *skb;
592 unsigned int i, j;
594 i = rx_ring->next_to_use;
595 buffer_info = &rx_ring->buffer_info[i];
597 while (cleaned_count--) {
598 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
600 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
601 ps_page = &buffer_info->ps_pages[j];
602 if (j >= adapter->rx_ps_pages) {
603 /* all unused desc entries get hw null ptr */
604 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
605 continue;
607 if (!ps_page->page) {
608 ps_page->page = alloc_page(GFP_ATOMIC);
609 if (!ps_page->page) {
610 adapter->alloc_rx_buff_failed++;
611 goto no_buffers;
613 ps_page->dma = dma_map_page(&pdev->dev,
614 ps_page->page,
615 0, PAGE_SIZE,
616 DMA_FROM_DEVICE);
617 if (dma_mapping_error(&pdev->dev,
618 ps_page->dma)) {
619 dev_err(&adapter->pdev->dev,
620 "RX DMA page map failed\n");
621 adapter->rx_dma_failed++;
622 goto no_buffers;
626 * Refresh the desc even if buffer_addrs
627 * didn't change because each write-back
628 * erases this info.
630 rx_desc->read.buffer_addr[j+1] =
631 cpu_to_le64(ps_page->dma);
634 skb = netdev_alloc_skb_ip_align(netdev,
635 adapter->rx_ps_bsize0);
637 if (!skb) {
638 adapter->alloc_rx_buff_failed++;
639 break;
642 buffer_info->skb = skb;
643 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
644 adapter->rx_ps_bsize0,
645 DMA_FROM_DEVICE);
646 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
647 dev_err(&pdev->dev, "RX DMA map failed\n");
648 adapter->rx_dma_failed++;
649 /* cleanup skb */
650 dev_kfree_skb_any(skb);
651 buffer_info->skb = NULL;
652 break;
655 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
657 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
659 * Force memory writes to complete before letting h/w
660 * know there are new descriptors to fetch. (Only
661 * applicable for weak-ordered memory model archs,
662 * such as IA-64).
664 wmb();
665 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
668 i++;
669 if (i == rx_ring->count)
670 i = 0;
671 buffer_info = &rx_ring->buffer_info[i];
674 no_buffers:
675 rx_ring->next_to_use = i;
679 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
680 * @adapter: address of board private structure
681 * @cleaned_count: number of buffers to allocate this pass
684 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
685 int cleaned_count)
687 struct net_device *netdev = adapter->netdev;
688 struct pci_dev *pdev = adapter->pdev;
689 struct e1000_rx_desc *rx_desc;
690 struct e1000_ring *rx_ring = adapter->rx_ring;
691 struct e1000_buffer *buffer_info;
692 struct sk_buff *skb;
693 unsigned int i;
694 unsigned int bufsz = 256 - 16 /* for skb_reserve */;
696 i = rx_ring->next_to_use;
697 buffer_info = &rx_ring->buffer_info[i];
699 while (cleaned_count--) {
700 skb = buffer_info->skb;
701 if (skb) {
702 skb_trim(skb, 0);
703 goto check_page;
706 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
707 if (unlikely(!skb)) {
708 /* Better luck next round */
709 adapter->alloc_rx_buff_failed++;
710 break;
713 buffer_info->skb = skb;
714 check_page:
715 /* allocate a new page if necessary */
716 if (!buffer_info->page) {
717 buffer_info->page = alloc_page(GFP_ATOMIC);
718 if (unlikely(!buffer_info->page)) {
719 adapter->alloc_rx_buff_failed++;
720 break;
724 if (!buffer_info->dma)
725 buffer_info->dma = dma_map_page(&pdev->dev,
726 buffer_info->page, 0,
727 PAGE_SIZE,
728 DMA_FROM_DEVICE);
730 rx_desc = E1000_RX_DESC(*rx_ring, i);
731 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
733 if (unlikely(++i == rx_ring->count))
734 i = 0;
735 buffer_info = &rx_ring->buffer_info[i];
738 if (likely(rx_ring->next_to_use != i)) {
739 rx_ring->next_to_use = i;
740 if (unlikely(i-- == 0))
741 i = (rx_ring->count - 1);
743 /* Force memory writes to complete before letting h/w
744 * know there are new descriptors to fetch. (Only
745 * applicable for weak-ordered memory model archs,
746 * such as IA-64). */
747 wmb();
748 writel(i, adapter->hw.hw_addr + rx_ring->tail);
753 * e1000_clean_rx_irq - Send received data up the network stack; legacy
754 * @adapter: board private structure
756 * the return value indicates whether actual cleaning was done, there
757 * is no guarantee that everything was cleaned
759 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
760 int *work_done, int work_to_do)
762 struct net_device *netdev = adapter->netdev;
763 struct pci_dev *pdev = adapter->pdev;
764 struct e1000_hw *hw = &adapter->hw;
765 struct e1000_ring *rx_ring = adapter->rx_ring;
766 struct e1000_rx_desc *rx_desc, *next_rxd;
767 struct e1000_buffer *buffer_info, *next_buffer;
768 u32 length;
769 unsigned int i;
770 int cleaned_count = 0;
771 bool cleaned = 0;
772 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
774 i = rx_ring->next_to_clean;
775 rx_desc = E1000_RX_DESC(*rx_ring, i);
776 buffer_info = &rx_ring->buffer_info[i];
778 while (rx_desc->status & E1000_RXD_STAT_DD) {
779 struct sk_buff *skb;
780 u8 status;
782 if (*work_done >= work_to_do)
783 break;
784 (*work_done)++;
785 rmb(); /* read descriptor and rx_buffer_info after status DD */
787 status = rx_desc->status;
788 skb = buffer_info->skb;
789 buffer_info->skb = NULL;
791 prefetch(skb->data - NET_IP_ALIGN);
793 i++;
794 if (i == rx_ring->count)
795 i = 0;
796 next_rxd = E1000_RX_DESC(*rx_ring, i);
797 prefetch(next_rxd);
799 next_buffer = &rx_ring->buffer_info[i];
801 cleaned = 1;
802 cleaned_count++;
803 dma_unmap_single(&pdev->dev,
804 buffer_info->dma,
805 adapter->rx_buffer_len,
806 DMA_FROM_DEVICE);
807 buffer_info->dma = 0;
809 length = le16_to_cpu(rx_desc->length);
812 * !EOP means multiple descriptors were used to store a single
813 * packet, if that's the case we need to toss it. In fact, we
814 * need to toss every packet with the EOP bit clear and the
815 * next frame that _does_ have the EOP bit set, as it is by
816 * definition only a frame fragment
818 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
819 adapter->flags2 |= FLAG2_IS_DISCARDING;
821 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
822 /* All receives must fit into a single buffer */
823 e_dbg("Receive packet consumed multiple buffers\n");
824 /* recycle */
825 buffer_info->skb = skb;
826 if (status & E1000_RXD_STAT_EOP)
827 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
828 goto next_desc;
831 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
832 /* recycle */
833 buffer_info->skb = skb;
834 goto next_desc;
837 /* adjust length to remove Ethernet CRC */
838 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
839 length -= 4;
841 total_rx_bytes += length;
842 total_rx_packets++;
845 * code added for copybreak, this should improve
846 * performance for small packets with large amounts
847 * of reassembly being done in the stack
849 if (length < copybreak) {
850 struct sk_buff *new_skb =
851 netdev_alloc_skb_ip_align(netdev, length);
852 if (new_skb) {
853 skb_copy_to_linear_data_offset(new_skb,
854 -NET_IP_ALIGN,
855 (skb->data -
856 NET_IP_ALIGN),
857 (length +
858 NET_IP_ALIGN));
859 /* save the skb in buffer_info as good */
860 buffer_info->skb = skb;
861 skb = new_skb;
863 /* else just continue with the old one */
865 /* end copybreak code */
866 skb_put(skb, length);
868 /* Receive Checksum Offload */
869 e1000_rx_checksum(adapter,
870 (u32)(status) |
871 ((u32)(rx_desc->errors) << 24),
872 le16_to_cpu(rx_desc->csum), skb);
874 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
876 next_desc:
877 rx_desc->status = 0;
879 /* return some buffers to hardware, one at a time is too slow */
880 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
881 adapter->alloc_rx_buf(adapter, cleaned_count);
882 cleaned_count = 0;
885 /* use prefetched values */
886 rx_desc = next_rxd;
887 buffer_info = next_buffer;
889 rx_ring->next_to_clean = i;
891 cleaned_count = e1000_desc_unused(rx_ring);
892 if (cleaned_count)
893 adapter->alloc_rx_buf(adapter, cleaned_count);
895 adapter->total_rx_bytes += total_rx_bytes;
896 adapter->total_rx_packets += total_rx_packets;
897 netdev->stats.rx_bytes += total_rx_bytes;
898 netdev->stats.rx_packets += total_rx_packets;
899 return cleaned;
902 static void e1000_put_txbuf(struct e1000_adapter *adapter,
903 struct e1000_buffer *buffer_info)
905 if (buffer_info->dma) {
906 if (buffer_info->mapped_as_page)
907 dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
908 buffer_info->length, DMA_TO_DEVICE);
909 else
910 dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
911 buffer_info->length, DMA_TO_DEVICE);
912 buffer_info->dma = 0;
914 if (buffer_info->skb) {
915 dev_kfree_skb_any(buffer_info->skb);
916 buffer_info->skb = NULL;
918 buffer_info->time_stamp = 0;
921 static void e1000_print_hw_hang(struct work_struct *work)
923 struct e1000_adapter *adapter = container_of(work,
924 struct e1000_adapter,
925 print_hang_task);
926 struct e1000_ring *tx_ring = adapter->tx_ring;
927 unsigned int i = tx_ring->next_to_clean;
928 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
929 struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
930 struct e1000_hw *hw = &adapter->hw;
931 u16 phy_status, phy_1000t_status, phy_ext_status;
932 u16 pci_status;
934 e1e_rphy(hw, PHY_STATUS, &phy_status);
935 e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
936 e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
938 pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
940 /* detected Hardware unit hang */
941 e_err("Detected Hardware Unit Hang:\n"
942 " TDH <%x>\n"
943 " TDT <%x>\n"
944 " next_to_use <%x>\n"
945 " next_to_clean <%x>\n"
946 "buffer_info[next_to_clean]:\n"
947 " time_stamp <%lx>\n"
948 " next_to_watch <%x>\n"
949 " jiffies <%lx>\n"
950 " next_to_watch.status <%x>\n"
951 "MAC Status <%x>\n"
952 "PHY Status <%x>\n"
953 "PHY 1000BASE-T Status <%x>\n"
954 "PHY Extended Status <%x>\n"
955 "PCI Status <%x>\n",
956 readl(adapter->hw.hw_addr + tx_ring->head),
957 readl(adapter->hw.hw_addr + tx_ring->tail),
958 tx_ring->next_to_use,
959 tx_ring->next_to_clean,
960 tx_ring->buffer_info[eop].time_stamp,
961 eop,
962 jiffies,
963 eop_desc->upper.fields.status,
964 er32(STATUS),
965 phy_status,
966 phy_1000t_status,
967 phy_ext_status,
968 pci_status);
972 * e1000_clean_tx_irq - Reclaim resources after transmit completes
973 * @adapter: board private structure
975 * the return value indicates whether actual cleaning was done, there
976 * is no guarantee that everything was cleaned
978 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
980 struct net_device *netdev = adapter->netdev;
981 struct e1000_hw *hw = &adapter->hw;
982 struct e1000_ring *tx_ring = adapter->tx_ring;
983 struct e1000_tx_desc *tx_desc, *eop_desc;
984 struct e1000_buffer *buffer_info;
985 unsigned int i, eop;
986 unsigned int count = 0;
987 unsigned int total_tx_bytes = 0, total_tx_packets = 0;
989 i = tx_ring->next_to_clean;
990 eop = tx_ring->buffer_info[i].next_to_watch;
991 eop_desc = E1000_TX_DESC(*tx_ring, eop);
993 while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
994 (count < tx_ring->count)) {
995 bool cleaned = false;
996 rmb(); /* read buffer_info after eop_desc */
997 for (; !cleaned; count++) {
998 tx_desc = E1000_TX_DESC(*tx_ring, i);
999 buffer_info = &tx_ring->buffer_info[i];
1000 cleaned = (i == eop);
1002 if (cleaned) {
1003 total_tx_packets += buffer_info->segs;
1004 total_tx_bytes += buffer_info->bytecount;
1007 e1000_put_txbuf(adapter, buffer_info);
1008 tx_desc->upper.data = 0;
1010 i++;
1011 if (i == tx_ring->count)
1012 i = 0;
1015 if (i == tx_ring->next_to_use)
1016 break;
1017 eop = tx_ring->buffer_info[i].next_to_watch;
1018 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1021 tx_ring->next_to_clean = i;
1023 #define TX_WAKE_THRESHOLD 32
1024 if (count && netif_carrier_ok(netdev) &&
1025 e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1026 /* Make sure that anybody stopping the queue after this
1027 * sees the new next_to_clean.
1029 smp_mb();
1031 if (netif_queue_stopped(netdev) &&
1032 !(test_bit(__E1000_DOWN, &adapter->state))) {
1033 netif_wake_queue(netdev);
1034 ++adapter->restart_queue;
1038 if (adapter->detect_tx_hung) {
1040 * Detect a transmit hang in hardware, this serializes the
1041 * check with the clearing of time_stamp and movement of i
1043 adapter->detect_tx_hung = 0;
1044 if (tx_ring->buffer_info[i].time_stamp &&
1045 time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1046 + (adapter->tx_timeout_factor * HZ)) &&
1047 !(er32(STATUS) & E1000_STATUS_TXOFF)) {
1048 schedule_work(&adapter->print_hang_task);
1049 netif_stop_queue(netdev);
1052 adapter->total_tx_bytes += total_tx_bytes;
1053 adapter->total_tx_packets += total_tx_packets;
1054 netdev->stats.tx_bytes += total_tx_bytes;
1055 netdev->stats.tx_packets += total_tx_packets;
1056 return count < tx_ring->count;
1060 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1061 * @adapter: board private structure
1063 * the return value indicates whether actual cleaning was done, there
1064 * is no guarantee that everything was cleaned
1066 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1067 int *work_done, int work_to_do)
1069 struct e1000_hw *hw = &adapter->hw;
1070 union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1071 struct net_device *netdev = adapter->netdev;
1072 struct pci_dev *pdev = adapter->pdev;
1073 struct e1000_ring *rx_ring = adapter->rx_ring;
1074 struct e1000_buffer *buffer_info, *next_buffer;
1075 struct e1000_ps_page *ps_page;
1076 struct sk_buff *skb;
1077 unsigned int i, j;
1078 u32 length, staterr;
1079 int cleaned_count = 0;
1080 bool cleaned = 0;
1081 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1083 i = rx_ring->next_to_clean;
1084 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1085 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1086 buffer_info = &rx_ring->buffer_info[i];
1088 while (staterr & E1000_RXD_STAT_DD) {
1089 if (*work_done >= work_to_do)
1090 break;
1091 (*work_done)++;
1092 skb = buffer_info->skb;
1093 rmb(); /* read descriptor and rx_buffer_info after status DD */
1095 /* in the packet split case this is header only */
1096 prefetch(skb->data - NET_IP_ALIGN);
1098 i++;
1099 if (i == rx_ring->count)
1100 i = 0;
1101 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1102 prefetch(next_rxd);
1104 next_buffer = &rx_ring->buffer_info[i];
1106 cleaned = 1;
1107 cleaned_count++;
1108 dma_unmap_single(&pdev->dev, buffer_info->dma,
1109 adapter->rx_ps_bsize0,
1110 DMA_FROM_DEVICE);
1111 buffer_info->dma = 0;
1113 /* see !EOP comment in other rx routine */
1114 if (!(staterr & E1000_RXD_STAT_EOP))
1115 adapter->flags2 |= FLAG2_IS_DISCARDING;
1117 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1118 e_dbg("Packet Split buffers didn't pick up the full "
1119 "packet\n");
1120 dev_kfree_skb_irq(skb);
1121 if (staterr & E1000_RXD_STAT_EOP)
1122 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1123 goto next_desc;
1126 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1127 dev_kfree_skb_irq(skb);
1128 goto next_desc;
1131 length = le16_to_cpu(rx_desc->wb.middle.length0);
1133 if (!length) {
1134 e_dbg("Last part of the packet spanning multiple "
1135 "descriptors\n");
1136 dev_kfree_skb_irq(skb);
1137 goto next_desc;
1140 /* Good Receive */
1141 skb_put(skb, length);
1145 * this looks ugly, but it seems compiler issues make it
1146 * more efficient than reusing j
1148 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1151 * page alloc/put takes too long and effects small packet
1152 * throughput, so unsplit small packets and save the alloc/put
1153 * only valid in softirq (napi) context to call kmap_*
1155 if (l1 && (l1 <= copybreak) &&
1156 ((length + l1) <= adapter->rx_ps_bsize0)) {
1157 u8 *vaddr;
1159 ps_page = &buffer_info->ps_pages[0];
1162 * there is no documentation about how to call
1163 * kmap_atomic, so we can't hold the mapping
1164 * very long
1166 dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1167 PAGE_SIZE, DMA_FROM_DEVICE);
1168 vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1169 memcpy(skb_tail_pointer(skb), vaddr, l1);
1170 kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1171 dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1172 PAGE_SIZE, DMA_FROM_DEVICE);
1174 /* remove the CRC */
1175 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1176 l1 -= 4;
1178 skb_put(skb, l1);
1179 goto copydone;
1180 } /* if */
1183 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1184 length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1185 if (!length)
1186 break;
1188 ps_page = &buffer_info->ps_pages[j];
1189 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1190 DMA_FROM_DEVICE);
1191 ps_page->dma = 0;
1192 skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1193 ps_page->page = NULL;
1194 skb->len += length;
1195 skb->data_len += length;
1196 skb->truesize += length;
1199 /* strip the ethernet crc, problem is we're using pages now so
1200 * this whole operation can get a little cpu intensive
1202 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1203 pskb_trim(skb, skb->len - 4);
1205 copydone:
1206 total_rx_bytes += skb->len;
1207 total_rx_packets++;
1209 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1210 rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1212 if (rx_desc->wb.upper.header_status &
1213 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1214 adapter->rx_hdr_split++;
1216 e1000_receive_skb(adapter, netdev, skb,
1217 staterr, rx_desc->wb.middle.vlan);
1219 next_desc:
1220 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1221 buffer_info->skb = NULL;
1223 /* return some buffers to hardware, one at a time is too slow */
1224 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1225 adapter->alloc_rx_buf(adapter, cleaned_count);
1226 cleaned_count = 0;
1229 /* use prefetched values */
1230 rx_desc = next_rxd;
1231 buffer_info = next_buffer;
1233 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1235 rx_ring->next_to_clean = i;
1237 cleaned_count = e1000_desc_unused(rx_ring);
1238 if (cleaned_count)
1239 adapter->alloc_rx_buf(adapter, cleaned_count);
1241 adapter->total_rx_bytes += total_rx_bytes;
1242 adapter->total_rx_packets += total_rx_packets;
1243 netdev->stats.rx_bytes += total_rx_bytes;
1244 netdev->stats.rx_packets += total_rx_packets;
1245 return cleaned;
1249 * e1000_consume_page - helper function
1251 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1252 u16 length)
1254 bi->page = NULL;
1255 skb->len += length;
1256 skb->data_len += length;
1257 skb->truesize += length;
1261 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1262 * @adapter: board private structure
1264 * the return value indicates whether actual cleaning was done, there
1265 * is no guarantee that everything was cleaned
1268 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1269 int *work_done, int work_to_do)
1271 struct net_device *netdev = adapter->netdev;
1272 struct pci_dev *pdev = adapter->pdev;
1273 struct e1000_ring *rx_ring = adapter->rx_ring;
1274 struct e1000_rx_desc *rx_desc, *next_rxd;
1275 struct e1000_buffer *buffer_info, *next_buffer;
1276 u32 length;
1277 unsigned int i;
1278 int cleaned_count = 0;
1279 bool cleaned = false;
1280 unsigned int total_rx_bytes=0, total_rx_packets=0;
1282 i = rx_ring->next_to_clean;
1283 rx_desc = E1000_RX_DESC(*rx_ring, i);
1284 buffer_info = &rx_ring->buffer_info[i];
1286 while (rx_desc->status & E1000_RXD_STAT_DD) {
1287 struct sk_buff *skb;
1288 u8 status;
1290 if (*work_done >= work_to_do)
1291 break;
1292 (*work_done)++;
1293 rmb(); /* read descriptor and rx_buffer_info after status DD */
1295 status = rx_desc->status;
1296 skb = buffer_info->skb;
1297 buffer_info->skb = NULL;
1299 ++i;
1300 if (i == rx_ring->count)
1301 i = 0;
1302 next_rxd = E1000_RX_DESC(*rx_ring, i);
1303 prefetch(next_rxd);
1305 next_buffer = &rx_ring->buffer_info[i];
1307 cleaned = true;
1308 cleaned_count++;
1309 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1310 DMA_FROM_DEVICE);
1311 buffer_info->dma = 0;
1313 length = le16_to_cpu(rx_desc->length);
1315 /* errors is only valid for DD + EOP descriptors */
1316 if (unlikely((status & E1000_RXD_STAT_EOP) &&
1317 (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
1318 /* recycle both page and skb */
1319 buffer_info->skb = skb;
1320 /* an error means any chain goes out the window
1321 * too */
1322 if (rx_ring->rx_skb_top)
1323 dev_kfree_skb(rx_ring->rx_skb_top);
1324 rx_ring->rx_skb_top = NULL;
1325 goto next_desc;
1328 #define rxtop rx_ring->rx_skb_top
1329 if (!(status & E1000_RXD_STAT_EOP)) {
1330 /* this descriptor is only the beginning (or middle) */
1331 if (!rxtop) {
1332 /* this is the beginning of a chain */
1333 rxtop = skb;
1334 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1335 0, length);
1336 } else {
1337 /* this is the middle of a chain */
1338 skb_fill_page_desc(rxtop,
1339 skb_shinfo(rxtop)->nr_frags,
1340 buffer_info->page, 0, length);
1341 /* re-use the skb, only consumed the page */
1342 buffer_info->skb = skb;
1344 e1000_consume_page(buffer_info, rxtop, length);
1345 goto next_desc;
1346 } else {
1347 if (rxtop) {
1348 /* end of the chain */
1349 skb_fill_page_desc(rxtop,
1350 skb_shinfo(rxtop)->nr_frags,
1351 buffer_info->page, 0, length);
1352 /* re-use the current skb, we only consumed the
1353 * page */
1354 buffer_info->skb = skb;
1355 skb = rxtop;
1356 rxtop = NULL;
1357 e1000_consume_page(buffer_info, skb, length);
1358 } else {
1359 /* no chain, got EOP, this buf is the packet
1360 * copybreak to save the put_page/alloc_page */
1361 if (length <= copybreak &&
1362 skb_tailroom(skb) >= length) {
1363 u8 *vaddr;
1364 vaddr = kmap_atomic(buffer_info->page,
1365 KM_SKB_DATA_SOFTIRQ);
1366 memcpy(skb_tail_pointer(skb), vaddr,
1367 length);
1368 kunmap_atomic(vaddr,
1369 KM_SKB_DATA_SOFTIRQ);
1370 /* re-use the page, so don't erase
1371 * buffer_info->page */
1372 skb_put(skb, length);
1373 } else {
1374 skb_fill_page_desc(skb, 0,
1375 buffer_info->page, 0,
1376 length);
1377 e1000_consume_page(buffer_info, skb,
1378 length);
1383 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1384 e1000_rx_checksum(adapter,
1385 (u32)(status) |
1386 ((u32)(rx_desc->errors) << 24),
1387 le16_to_cpu(rx_desc->csum), skb);
1389 /* probably a little skewed due to removing CRC */
1390 total_rx_bytes += skb->len;
1391 total_rx_packets++;
1393 /* eth type trans needs skb->data to point to something */
1394 if (!pskb_may_pull(skb, ETH_HLEN)) {
1395 e_err("pskb_may_pull failed.\n");
1396 dev_kfree_skb(skb);
1397 goto next_desc;
1400 e1000_receive_skb(adapter, netdev, skb, status,
1401 rx_desc->special);
1403 next_desc:
1404 rx_desc->status = 0;
1406 /* return some buffers to hardware, one at a time is too slow */
1407 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1408 adapter->alloc_rx_buf(adapter, cleaned_count);
1409 cleaned_count = 0;
1412 /* use prefetched values */
1413 rx_desc = next_rxd;
1414 buffer_info = next_buffer;
1416 rx_ring->next_to_clean = i;
1418 cleaned_count = e1000_desc_unused(rx_ring);
1419 if (cleaned_count)
1420 adapter->alloc_rx_buf(adapter, cleaned_count);
1422 adapter->total_rx_bytes += total_rx_bytes;
1423 adapter->total_rx_packets += total_rx_packets;
1424 netdev->stats.rx_bytes += total_rx_bytes;
1425 netdev->stats.rx_packets += total_rx_packets;
1426 return cleaned;
1430 * e1000_clean_rx_ring - Free Rx Buffers per Queue
1431 * @adapter: board private structure
1433 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1435 struct e1000_ring *rx_ring = adapter->rx_ring;
1436 struct e1000_buffer *buffer_info;
1437 struct e1000_ps_page *ps_page;
1438 struct pci_dev *pdev = adapter->pdev;
1439 unsigned int i, j;
1441 /* Free all the Rx ring sk_buffs */
1442 for (i = 0; i < rx_ring->count; i++) {
1443 buffer_info = &rx_ring->buffer_info[i];
1444 if (buffer_info->dma) {
1445 if (adapter->clean_rx == e1000_clean_rx_irq)
1446 dma_unmap_single(&pdev->dev, buffer_info->dma,
1447 adapter->rx_buffer_len,
1448 DMA_FROM_DEVICE);
1449 else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1450 dma_unmap_page(&pdev->dev, buffer_info->dma,
1451 PAGE_SIZE,
1452 DMA_FROM_DEVICE);
1453 else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1454 dma_unmap_single(&pdev->dev, buffer_info->dma,
1455 adapter->rx_ps_bsize0,
1456 DMA_FROM_DEVICE);
1457 buffer_info->dma = 0;
1460 if (buffer_info->page) {
1461 put_page(buffer_info->page);
1462 buffer_info->page = NULL;
1465 if (buffer_info->skb) {
1466 dev_kfree_skb(buffer_info->skb);
1467 buffer_info->skb = NULL;
1470 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1471 ps_page = &buffer_info->ps_pages[j];
1472 if (!ps_page->page)
1473 break;
1474 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1475 DMA_FROM_DEVICE);
1476 ps_page->dma = 0;
1477 put_page(ps_page->page);
1478 ps_page->page = NULL;
1482 /* there also may be some cached data from a chained receive */
1483 if (rx_ring->rx_skb_top) {
1484 dev_kfree_skb(rx_ring->rx_skb_top);
1485 rx_ring->rx_skb_top = NULL;
1488 /* Zero out the descriptor ring */
1489 memset(rx_ring->desc, 0, rx_ring->size);
1491 rx_ring->next_to_clean = 0;
1492 rx_ring->next_to_use = 0;
1493 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1495 writel(0, adapter->hw.hw_addr + rx_ring->head);
1496 writel(0, adapter->hw.hw_addr + rx_ring->tail);
1499 static void e1000e_downshift_workaround(struct work_struct *work)
1501 struct e1000_adapter *adapter = container_of(work,
1502 struct e1000_adapter, downshift_task);
1504 e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1508 * e1000_intr_msi - Interrupt Handler
1509 * @irq: interrupt number
1510 * @data: pointer to a network interface device structure
1512 static irqreturn_t e1000_intr_msi(int irq, void *data)
1514 struct net_device *netdev = data;
1515 struct e1000_adapter *adapter = netdev_priv(netdev);
1516 struct e1000_hw *hw = &adapter->hw;
1517 u32 icr = er32(ICR);
1520 * read ICR disables interrupts using IAM
1523 if (icr & E1000_ICR_LSC) {
1524 hw->mac.get_link_status = 1;
1526 * ICH8 workaround-- Call gig speed drop workaround on cable
1527 * disconnect (LSC) before accessing any PHY registers
1529 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1530 (!(er32(STATUS) & E1000_STATUS_LU)))
1531 schedule_work(&adapter->downshift_task);
1534 * 80003ES2LAN workaround-- For packet buffer work-around on
1535 * link down event; disable receives here in the ISR and reset
1536 * adapter in watchdog
1538 if (netif_carrier_ok(netdev) &&
1539 adapter->flags & FLAG_RX_NEEDS_RESTART) {
1540 /* disable receives */
1541 u32 rctl = er32(RCTL);
1542 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1543 adapter->flags |= FLAG_RX_RESTART_NOW;
1545 /* guard against interrupt when we're going down */
1546 if (!test_bit(__E1000_DOWN, &adapter->state))
1547 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1550 if (napi_schedule_prep(&adapter->napi)) {
1551 adapter->total_tx_bytes = 0;
1552 adapter->total_tx_packets = 0;
1553 adapter->total_rx_bytes = 0;
1554 adapter->total_rx_packets = 0;
1555 __napi_schedule(&adapter->napi);
1558 return IRQ_HANDLED;
1562 * e1000_intr - Interrupt Handler
1563 * @irq: interrupt number
1564 * @data: pointer to a network interface device structure
1566 static irqreturn_t e1000_intr(int irq, void *data)
1568 struct net_device *netdev = data;
1569 struct e1000_adapter *adapter = netdev_priv(netdev);
1570 struct e1000_hw *hw = &adapter->hw;
1571 u32 rctl, icr = er32(ICR);
1573 if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1574 return IRQ_NONE; /* Not our interrupt */
1577 * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1578 * not set, then the adapter didn't send an interrupt
1580 if (!(icr & E1000_ICR_INT_ASSERTED))
1581 return IRQ_NONE;
1584 * Interrupt Auto-Mask...upon reading ICR,
1585 * interrupts are masked. No need for the
1586 * IMC write
1589 if (icr & E1000_ICR_LSC) {
1590 hw->mac.get_link_status = 1;
1592 * ICH8 workaround-- Call gig speed drop workaround on cable
1593 * disconnect (LSC) before accessing any PHY registers
1595 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1596 (!(er32(STATUS) & E1000_STATUS_LU)))
1597 schedule_work(&adapter->downshift_task);
1600 * 80003ES2LAN workaround--
1601 * For packet buffer work-around on link down event;
1602 * disable receives here in the ISR and
1603 * reset adapter in watchdog
1605 if (netif_carrier_ok(netdev) &&
1606 (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1607 /* disable receives */
1608 rctl = er32(RCTL);
1609 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1610 adapter->flags |= FLAG_RX_RESTART_NOW;
1612 /* guard against interrupt when we're going down */
1613 if (!test_bit(__E1000_DOWN, &adapter->state))
1614 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1617 if (napi_schedule_prep(&adapter->napi)) {
1618 adapter->total_tx_bytes = 0;
1619 adapter->total_tx_packets = 0;
1620 adapter->total_rx_bytes = 0;
1621 adapter->total_rx_packets = 0;
1622 __napi_schedule(&adapter->napi);
1625 return IRQ_HANDLED;
1628 static irqreturn_t e1000_msix_other(int irq, void *data)
1630 struct net_device *netdev = data;
1631 struct e1000_adapter *adapter = netdev_priv(netdev);
1632 struct e1000_hw *hw = &adapter->hw;
1633 u32 icr = er32(ICR);
1635 if (!(icr & E1000_ICR_INT_ASSERTED)) {
1636 if (!test_bit(__E1000_DOWN, &adapter->state))
1637 ew32(IMS, E1000_IMS_OTHER);
1638 return IRQ_NONE;
1641 if (icr & adapter->eiac_mask)
1642 ew32(ICS, (icr & adapter->eiac_mask));
1644 if (icr & E1000_ICR_OTHER) {
1645 if (!(icr & E1000_ICR_LSC))
1646 goto no_link_interrupt;
1647 hw->mac.get_link_status = 1;
1648 /* guard against interrupt when we're going down */
1649 if (!test_bit(__E1000_DOWN, &adapter->state))
1650 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1653 no_link_interrupt:
1654 if (!test_bit(__E1000_DOWN, &adapter->state))
1655 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1657 return IRQ_HANDLED;
1661 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1663 struct net_device *netdev = data;
1664 struct e1000_adapter *adapter = netdev_priv(netdev);
1665 struct e1000_hw *hw = &adapter->hw;
1666 struct e1000_ring *tx_ring = adapter->tx_ring;
1669 adapter->total_tx_bytes = 0;
1670 adapter->total_tx_packets = 0;
1672 if (!e1000_clean_tx_irq(adapter))
1673 /* Ring was not completely cleaned, so fire another interrupt */
1674 ew32(ICS, tx_ring->ims_val);
1676 return IRQ_HANDLED;
1679 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1681 struct net_device *netdev = data;
1682 struct e1000_adapter *adapter = netdev_priv(netdev);
1684 /* Write the ITR value calculated at the end of the
1685 * previous interrupt.
1687 if (adapter->rx_ring->set_itr) {
1688 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1689 adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1690 adapter->rx_ring->set_itr = 0;
1693 if (napi_schedule_prep(&adapter->napi)) {
1694 adapter->total_rx_bytes = 0;
1695 adapter->total_rx_packets = 0;
1696 __napi_schedule(&adapter->napi);
1698 return IRQ_HANDLED;
1702 * e1000_configure_msix - Configure MSI-X hardware
1704 * e1000_configure_msix sets up the hardware to properly
1705 * generate MSI-X interrupts.
1707 static void e1000_configure_msix(struct e1000_adapter *adapter)
1709 struct e1000_hw *hw = &adapter->hw;
1710 struct e1000_ring *rx_ring = adapter->rx_ring;
1711 struct e1000_ring *tx_ring = adapter->tx_ring;
1712 int vector = 0;
1713 u32 ctrl_ext, ivar = 0;
1715 adapter->eiac_mask = 0;
1717 /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1718 if (hw->mac.type == e1000_82574) {
1719 u32 rfctl = er32(RFCTL);
1720 rfctl |= E1000_RFCTL_ACK_DIS;
1721 ew32(RFCTL, rfctl);
1724 #define E1000_IVAR_INT_ALLOC_VALID 0x8
1725 /* Configure Rx vector */
1726 rx_ring->ims_val = E1000_IMS_RXQ0;
1727 adapter->eiac_mask |= rx_ring->ims_val;
1728 if (rx_ring->itr_val)
1729 writel(1000000000 / (rx_ring->itr_val * 256),
1730 hw->hw_addr + rx_ring->itr_register);
1731 else
1732 writel(1, hw->hw_addr + rx_ring->itr_register);
1733 ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1735 /* Configure Tx vector */
1736 tx_ring->ims_val = E1000_IMS_TXQ0;
1737 vector++;
1738 if (tx_ring->itr_val)
1739 writel(1000000000 / (tx_ring->itr_val * 256),
1740 hw->hw_addr + tx_ring->itr_register);
1741 else
1742 writel(1, hw->hw_addr + tx_ring->itr_register);
1743 adapter->eiac_mask |= tx_ring->ims_val;
1744 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1746 /* set vector for Other Causes, e.g. link changes */
1747 vector++;
1748 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1749 if (rx_ring->itr_val)
1750 writel(1000000000 / (rx_ring->itr_val * 256),
1751 hw->hw_addr + E1000_EITR_82574(vector));
1752 else
1753 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1755 /* Cause Tx interrupts on every write back */
1756 ivar |= (1 << 31);
1758 ew32(IVAR, ivar);
1760 /* enable MSI-X PBA support */
1761 ctrl_ext = er32(CTRL_EXT);
1762 ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1764 /* Auto-Mask Other interrupts upon ICR read */
1765 #define E1000_EIAC_MASK_82574 0x01F00000
1766 ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1767 ctrl_ext |= E1000_CTRL_EXT_EIAME;
1768 ew32(CTRL_EXT, ctrl_ext);
1769 e1e_flush();
1772 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1774 if (adapter->msix_entries) {
1775 pci_disable_msix(adapter->pdev);
1776 kfree(adapter->msix_entries);
1777 adapter->msix_entries = NULL;
1778 } else if (adapter->flags & FLAG_MSI_ENABLED) {
1779 pci_disable_msi(adapter->pdev);
1780 adapter->flags &= ~FLAG_MSI_ENABLED;
1785 * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1787 * Attempt to configure interrupts using the best available
1788 * capabilities of the hardware and kernel.
1790 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1792 int err;
1793 int i;
1795 switch (adapter->int_mode) {
1796 case E1000E_INT_MODE_MSIX:
1797 if (adapter->flags & FLAG_HAS_MSIX) {
1798 adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1799 adapter->msix_entries = kcalloc(adapter->num_vectors,
1800 sizeof(struct msix_entry),
1801 GFP_KERNEL);
1802 if (adapter->msix_entries) {
1803 for (i = 0; i < adapter->num_vectors; i++)
1804 adapter->msix_entries[i].entry = i;
1806 err = pci_enable_msix(adapter->pdev,
1807 adapter->msix_entries,
1808 adapter->num_vectors);
1809 if (err == 0) {
1810 return;
1813 /* MSI-X failed, so fall through and try MSI */
1814 e_err("Failed to initialize MSI-X interrupts. "
1815 "Falling back to MSI interrupts.\n");
1816 e1000e_reset_interrupt_capability(adapter);
1818 adapter->int_mode = E1000E_INT_MODE_MSI;
1819 /* Fall through */
1820 case E1000E_INT_MODE_MSI:
1821 if (!pci_enable_msi(adapter->pdev)) {
1822 adapter->flags |= FLAG_MSI_ENABLED;
1823 } else {
1824 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1825 e_err("Failed to initialize MSI interrupts. Falling "
1826 "back to legacy interrupts.\n");
1828 /* Fall through */
1829 case E1000E_INT_MODE_LEGACY:
1830 /* Don't do anything; this is the system default */
1831 break;
1834 /* store the number of vectors being used */
1835 adapter->num_vectors = 1;
1839 * e1000_request_msix - Initialize MSI-X interrupts
1841 * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1842 * kernel.
1844 static int e1000_request_msix(struct e1000_adapter *adapter)
1846 struct net_device *netdev = adapter->netdev;
1847 int err = 0, vector = 0;
1849 if (strlen(netdev->name) < (IFNAMSIZ - 5))
1850 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1851 else
1852 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1853 err = request_irq(adapter->msix_entries[vector].vector,
1854 e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1855 netdev);
1856 if (err)
1857 goto out;
1858 adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1859 adapter->rx_ring->itr_val = adapter->itr;
1860 vector++;
1862 if (strlen(netdev->name) < (IFNAMSIZ - 5))
1863 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1864 else
1865 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1866 err = request_irq(adapter->msix_entries[vector].vector,
1867 e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1868 netdev);
1869 if (err)
1870 goto out;
1871 adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1872 adapter->tx_ring->itr_val = adapter->itr;
1873 vector++;
1875 err = request_irq(adapter->msix_entries[vector].vector,
1876 e1000_msix_other, 0, netdev->name, netdev);
1877 if (err)
1878 goto out;
1880 e1000_configure_msix(adapter);
1881 return 0;
1882 out:
1883 return err;
1887 * e1000_request_irq - initialize interrupts
1889 * Attempts to configure interrupts using the best available
1890 * capabilities of the hardware and kernel.
1892 static int e1000_request_irq(struct e1000_adapter *adapter)
1894 struct net_device *netdev = adapter->netdev;
1895 int err;
1897 if (adapter->msix_entries) {
1898 err = e1000_request_msix(adapter);
1899 if (!err)
1900 return err;
1901 /* fall back to MSI */
1902 e1000e_reset_interrupt_capability(adapter);
1903 adapter->int_mode = E1000E_INT_MODE_MSI;
1904 e1000e_set_interrupt_capability(adapter);
1906 if (adapter->flags & FLAG_MSI_ENABLED) {
1907 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1908 netdev->name, netdev);
1909 if (!err)
1910 return err;
1912 /* fall back to legacy interrupt */
1913 e1000e_reset_interrupt_capability(adapter);
1914 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1917 err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1918 netdev->name, netdev);
1919 if (err)
1920 e_err("Unable to allocate interrupt, Error: %d\n", err);
1922 return err;
1925 static void e1000_free_irq(struct e1000_adapter *adapter)
1927 struct net_device *netdev = adapter->netdev;
1929 if (adapter->msix_entries) {
1930 int vector = 0;
1932 free_irq(adapter->msix_entries[vector].vector, netdev);
1933 vector++;
1935 free_irq(adapter->msix_entries[vector].vector, netdev);
1936 vector++;
1938 /* Other Causes interrupt vector */
1939 free_irq(adapter->msix_entries[vector].vector, netdev);
1940 return;
1943 free_irq(adapter->pdev->irq, netdev);
1947 * e1000_irq_disable - Mask off interrupt generation on the NIC
1949 static void e1000_irq_disable(struct e1000_adapter *adapter)
1951 struct e1000_hw *hw = &adapter->hw;
1953 ew32(IMC, ~0);
1954 if (adapter->msix_entries)
1955 ew32(EIAC_82574, 0);
1956 e1e_flush();
1958 if (adapter->msix_entries) {
1959 int i;
1960 for (i = 0; i < adapter->num_vectors; i++)
1961 synchronize_irq(adapter->msix_entries[i].vector);
1962 } else {
1963 synchronize_irq(adapter->pdev->irq);
1968 * e1000_irq_enable - Enable default interrupt generation settings
1970 static void e1000_irq_enable(struct e1000_adapter *adapter)
1972 struct e1000_hw *hw = &adapter->hw;
1974 if (adapter->msix_entries) {
1975 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1976 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1977 } else {
1978 ew32(IMS, IMS_ENABLE_MASK);
1980 e1e_flush();
1984 * e1000_get_hw_control - get control of the h/w from f/w
1985 * @adapter: address of board private structure
1987 * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1988 * For ASF and Pass Through versions of f/w this means that
1989 * the driver is loaded. For AMT version (only with 82573)
1990 * of the f/w this means that the network i/f is open.
1992 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1994 struct e1000_hw *hw = &adapter->hw;
1995 u32 ctrl_ext;
1996 u32 swsm;
1998 /* Let firmware know the driver has taken over */
1999 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2000 swsm = er32(SWSM);
2001 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2002 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2003 ctrl_ext = er32(CTRL_EXT);
2004 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2009 * e1000_release_hw_control - release control of the h/w to f/w
2010 * @adapter: address of board private structure
2012 * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2013 * For ASF and Pass Through versions of f/w this means that the
2014 * driver is no longer loaded. For AMT version (only with 82573) i
2015 * of the f/w this means that the network i/f is closed.
2018 static void e1000_release_hw_control(struct e1000_adapter *adapter)
2020 struct e1000_hw *hw = &adapter->hw;
2021 u32 ctrl_ext;
2022 u32 swsm;
2024 /* Let firmware taken over control of h/w */
2025 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2026 swsm = er32(SWSM);
2027 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2028 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2029 ctrl_ext = er32(CTRL_EXT);
2030 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2035 * @e1000_alloc_ring - allocate memory for a ring structure
2037 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2038 struct e1000_ring *ring)
2040 struct pci_dev *pdev = adapter->pdev;
2042 ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2043 GFP_KERNEL);
2044 if (!ring->desc)
2045 return -ENOMEM;
2047 return 0;
2051 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2052 * @adapter: board private structure
2054 * Return 0 on success, negative on failure
2056 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2058 struct e1000_ring *tx_ring = adapter->tx_ring;
2059 int err = -ENOMEM, size;
2061 size = sizeof(struct e1000_buffer) * tx_ring->count;
2062 tx_ring->buffer_info = vmalloc(size);
2063 if (!tx_ring->buffer_info)
2064 goto err;
2065 memset(tx_ring->buffer_info, 0, size);
2067 /* round up to nearest 4K */
2068 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2069 tx_ring->size = ALIGN(tx_ring->size, 4096);
2071 err = e1000_alloc_ring_dma(adapter, tx_ring);
2072 if (err)
2073 goto err;
2075 tx_ring->next_to_use = 0;
2076 tx_ring->next_to_clean = 0;
2078 return 0;
2079 err:
2080 vfree(tx_ring->buffer_info);
2081 e_err("Unable to allocate memory for the transmit descriptor ring\n");
2082 return err;
2086 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2087 * @adapter: board private structure
2089 * Returns 0 on success, negative on failure
2091 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2093 struct e1000_ring *rx_ring = adapter->rx_ring;
2094 struct e1000_buffer *buffer_info;
2095 int i, size, desc_len, err = -ENOMEM;
2097 size = sizeof(struct e1000_buffer) * rx_ring->count;
2098 rx_ring->buffer_info = vmalloc(size);
2099 if (!rx_ring->buffer_info)
2100 goto err;
2101 memset(rx_ring->buffer_info, 0, size);
2103 for (i = 0; i < rx_ring->count; i++) {
2104 buffer_info = &rx_ring->buffer_info[i];
2105 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2106 sizeof(struct e1000_ps_page),
2107 GFP_KERNEL);
2108 if (!buffer_info->ps_pages)
2109 goto err_pages;
2112 desc_len = sizeof(union e1000_rx_desc_packet_split);
2114 /* Round up to nearest 4K */
2115 rx_ring->size = rx_ring->count * desc_len;
2116 rx_ring->size = ALIGN(rx_ring->size, 4096);
2118 err = e1000_alloc_ring_dma(adapter, rx_ring);
2119 if (err)
2120 goto err_pages;
2122 rx_ring->next_to_clean = 0;
2123 rx_ring->next_to_use = 0;
2124 rx_ring->rx_skb_top = NULL;
2126 return 0;
2128 err_pages:
2129 for (i = 0; i < rx_ring->count; i++) {
2130 buffer_info = &rx_ring->buffer_info[i];
2131 kfree(buffer_info->ps_pages);
2133 err:
2134 vfree(rx_ring->buffer_info);
2135 e_err("Unable to allocate memory for the transmit descriptor ring\n");
2136 return err;
2140 * e1000_clean_tx_ring - Free Tx Buffers
2141 * @adapter: board private structure
2143 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2145 struct e1000_ring *tx_ring = adapter->tx_ring;
2146 struct e1000_buffer *buffer_info;
2147 unsigned long size;
2148 unsigned int i;
2150 for (i = 0; i < tx_ring->count; i++) {
2151 buffer_info = &tx_ring->buffer_info[i];
2152 e1000_put_txbuf(adapter, buffer_info);
2155 size = sizeof(struct e1000_buffer) * tx_ring->count;
2156 memset(tx_ring->buffer_info, 0, size);
2158 memset(tx_ring->desc, 0, tx_ring->size);
2160 tx_ring->next_to_use = 0;
2161 tx_ring->next_to_clean = 0;
2163 writel(0, adapter->hw.hw_addr + tx_ring->head);
2164 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2168 * e1000e_free_tx_resources - Free Tx Resources per Queue
2169 * @adapter: board private structure
2171 * Free all transmit software resources
2173 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2175 struct pci_dev *pdev = adapter->pdev;
2176 struct e1000_ring *tx_ring = adapter->tx_ring;
2178 e1000_clean_tx_ring(adapter);
2180 vfree(tx_ring->buffer_info);
2181 tx_ring->buffer_info = NULL;
2183 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2184 tx_ring->dma);
2185 tx_ring->desc = NULL;
2189 * e1000e_free_rx_resources - Free Rx Resources
2190 * @adapter: board private structure
2192 * Free all receive software resources
2195 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2197 struct pci_dev *pdev = adapter->pdev;
2198 struct e1000_ring *rx_ring = adapter->rx_ring;
2199 int i;
2201 e1000_clean_rx_ring(adapter);
2203 for (i = 0; i < rx_ring->count; i++) {
2204 kfree(rx_ring->buffer_info[i].ps_pages);
2207 vfree(rx_ring->buffer_info);
2208 rx_ring->buffer_info = NULL;
2210 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2211 rx_ring->dma);
2212 rx_ring->desc = NULL;
2216 * e1000_update_itr - update the dynamic ITR value based on statistics
2217 * @adapter: pointer to adapter
2218 * @itr_setting: current adapter->itr
2219 * @packets: the number of packets during this measurement interval
2220 * @bytes: the number of bytes during this measurement interval
2222 * Stores a new ITR value based on packets and byte
2223 * counts during the last interrupt. The advantage of per interrupt
2224 * computation is faster updates and more accurate ITR for the current
2225 * traffic pattern. Constants in this function were computed
2226 * based on theoretical maximum wire speed and thresholds were set based
2227 * on testing data as well as attempting to minimize response time
2228 * while increasing bulk throughput. This functionality is controlled
2229 * by the InterruptThrottleRate module parameter.
2231 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2232 u16 itr_setting, int packets,
2233 int bytes)
2235 unsigned int retval = itr_setting;
2237 if (packets == 0)
2238 goto update_itr_done;
2240 switch (itr_setting) {
2241 case lowest_latency:
2242 /* handle TSO and jumbo frames */
2243 if (bytes/packets > 8000)
2244 retval = bulk_latency;
2245 else if ((packets < 5) && (bytes > 512)) {
2246 retval = low_latency;
2248 break;
2249 case low_latency: /* 50 usec aka 20000 ints/s */
2250 if (bytes > 10000) {
2251 /* this if handles the TSO accounting */
2252 if (bytes/packets > 8000) {
2253 retval = bulk_latency;
2254 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2255 retval = bulk_latency;
2256 } else if ((packets > 35)) {
2257 retval = lowest_latency;
2259 } else if (bytes/packets > 2000) {
2260 retval = bulk_latency;
2261 } else if (packets <= 2 && bytes < 512) {
2262 retval = lowest_latency;
2264 break;
2265 case bulk_latency: /* 250 usec aka 4000 ints/s */
2266 if (bytes > 25000) {
2267 if (packets > 35) {
2268 retval = low_latency;
2270 } else if (bytes < 6000) {
2271 retval = low_latency;
2273 break;
2276 update_itr_done:
2277 return retval;
2280 static void e1000_set_itr(struct e1000_adapter *adapter)
2282 struct e1000_hw *hw = &adapter->hw;
2283 u16 current_itr;
2284 u32 new_itr = adapter->itr;
2286 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2287 if (adapter->link_speed != SPEED_1000) {
2288 current_itr = 0;
2289 new_itr = 4000;
2290 goto set_itr_now;
2293 if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2294 new_itr = 0;
2295 goto set_itr_now;
2298 adapter->tx_itr = e1000_update_itr(adapter,
2299 adapter->tx_itr,
2300 adapter->total_tx_packets,
2301 adapter->total_tx_bytes);
2302 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2303 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2304 adapter->tx_itr = low_latency;
2306 adapter->rx_itr = e1000_update_itr(adapter,
2307 adapter->rx_itr,
2308 adapter->total_rx_packets,
2309 adapter->total_rx_bytes);
2310 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2311 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2312 adapter->rx_itr = low_latency;
2314 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2316 switch (current_itr) {
2317 /* counts and packets in update_itr are dependent on these numbers */
2318 case lowest_latency:
2319 new_itr = 70000;
2320 break;
2321 case low_latency:
2322 new_itr = 20000; /* aka hwitr = ~200 */
2323 break;
2324 case bulk_latency:
2325 new_itr = 4000;
2326 break;
2327 default:
2328 break;
2331 set_itr_now:
2332 if (new_itr != adapter->itr) {
2334 * this attempts to bias the interrupt rate towards Bulk
2335 * by adding intermediate steps when interrupt rate is
2336 * increasing
2338 new_itr = new_itr > adapter->itr ?
2339 min(adapter->itr + (new_itr >> 2), new_itr) :
2340 new_itr;
2341 adapter->itr = new_itr;
2342 adapter->rx_ring->itr_val = new_itr;
2343 if (adapter->msix_entries)
2344 adapter->rx_ring->set_itr = 1;
2345 else
2346 if (new_itr)
2347 ew32(ITR, 1000000000 / (new_itr * 256));
2348 else
2349 ew32(ITR, 0);
2354 * e1000_alloc_queues - Allocate memory for all rings
2355 * @adapter: board private structure to initialize
2357 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2359 adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2360 if (!adapter->tx_ring)
2361 goto err;
2363 adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2364 if (!adapter->rx_ring)
2365 goto err;
2367 return 0;
2368 err:
2369 e_err("Unable to allocate memory for queues\n");
2370 kfree(adapter->rx_ring);
2371 kfree(adapter->tx_ring);
2372 return -ENOMEM;
2376 * e1000_clean - NAPI Rx polling callback
2377 * @napi: struct associated with this polling callback
2378 * @budget: amount of packets driver is allowed to process this poll
2380 static int e1000_clean(struct napi_struct *napi, int budget)
2382 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2383 struct e1000_hw *hw = &adapter->hw;
2384 struct net_device *poll_dev = adapter->netdev;
2385 int tx_cleaned = 1, work_done = 0;
2387 adapter = netdev_priv(poll_dev);
2389 if (adapter->msix_entries &&
2390 !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2391 goto clean_rx;
2393 tx_cleaned = e1000_clean_tx_irq(adapter);
2395 clean_rx:
2396 adapter->clean_rx(adapter, &work_done, budget);
2398 if (!tx_cleaned)
2399 work_done = budget;
2401 /* If budget not fully consumed, exit the polling mode */
2402 if (work_done < budget) {
2403 if (adapter->itr_setting & 3)
2404 e1000_set_itr(adapter);
2405 napi_complete(napi);
2406 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2407 if (adapter->msix_entries)
2408 ew32(IMS, adapter->rx_ring->ims_val);
2409 else
2410 e1000_irq_enable(adapter);
2414 return work_done;
2417 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2419 struct e1000_adapter *adapter = netdev_priv(netdev);
2420 struct e1000_hw *hw = &adapter->hw;
2421 u32 vfta, index;
2423 /* don't update vlan cookie if already programmed */
2424 if ((adapter->hw.mng_cookie.status &
2425 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2426 (vid == adapter->mng_vlan_id))
2427 return;
2429 /* add VID to filter table */
2430 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2431 index = (vid >> 5) & 0x7F;
2432 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2433 vfta |= (1 << (vid & 0x1F));
2434 hw->mac.ops.write_vfta(hw, index, vfta);
2438 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2440 struct e1000_adapter *adapter = netdev_priv(netdev);
2441 struct e1000_hw *hw = &adapter->hw;
2442 u32 vfta, index;
2444 if (!test_bit(__E1000_DOWN, &adapter->state))
2445 e1000_irq_disable(adapter);
2446 vlan_group_set_device(adapter->vlgrp, vid, NULL);
2448 if (!test_bit(__E1000_DOWN, &adapter->state))
2449 e1000_irq_enable(adapter);
2451 if ((adapter->hw.mng_cookie.status &
2452 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2453 (vid == adapter->mng_vlan_id)) {
2454 /* release control to f/w */
2455 e1000_release_hw_control(adapter);
2456 return;
2459 /* remove VID from filter table */
2460 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2461 index = (vid >> 5) & 0x7F;
2462 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2463 vfta &= ~(1 << (vid & 0x1F));
2464 hw->mac.ops.write_vfta(hw, index, vfta);
2468 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2470 struct net_device *netdev = adapter->netdev;
2471 u16 vid = adapter->hw.mng_cookie.vlan_id;
2472 u16 old_vid = adapter->mng_vlan_id;
2474 if (!adapter->vlgrp)
2475 return;
2477 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2478 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2479 if (adapter->hw.mng_cookie.status &
2480 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2481 e1000_vlan_rx_add_vid(netdev, vid);
2482 adapter->mng_vlan_id = vid;
2485 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2486 (vid != old_vid) &&
2487 !vlan_group_get_device(adapter->vlgrp, old_vid))
2488 e1000_vlan_rx_kill_vid(netdev, old_vid);
2489 } else {
2490 adapter->mng_vlan_id = vid;
2495 static void e1000_vlan_rx_register(struct net_device *netdev,
2496 struct vlan_group *grp)
2498 struct e1000_adapter *adapter = netdev_priv(netdev);
2499 struct e1000_hw *hw = &adapter->hw;
2500 u32 ctrl, rctl;
2502 if (!test_bit(__E1000_DOWN, &adapter->state))
2503 e1000_irq_disable(adapter);
2504 adapter->vlgrp = grp;
2506 if (grp) {
2507 /* enable VLAN tag insert/strip */
2508 ctrl = er32(CTRL);
2509 ctrl |= E1000_CTRL_VME;
2510 ew32(CTRL, ctrl);
2512 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2513 /* enable VLAN receive filtering */
2514 rctl = er32(RCTL);
2515 rctl &= ~E1000_RCTL_CFIEN;
2516 ew32(RCTL, rctl);
2517 e1000_update_mng_vlan(adapter);
2519 } else {
2520 /* disable VLAN tag insert/strip */
2521 ctrl = er32(CTRL);
2522 ctrl &= ~E1000_CTRL_VME;
2523 ew32(CTRL, ctrl);
2525 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2526 if (adapter->mng_vlan_id !=
2527 (u16)E1000_MNG_VLAN_NONE) {
2528 e1000_vlan_rx_kill_vid(netdev,
2529 adapter->mng_vlan_id);
2530 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2535 if (!test_bit(__E1000_DOWN, &adapter->state))
2536 e1000_irq_enable(adapter);
2539 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2541 u16 vid;
2543 e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2545 if (!adapter->vlgrp)
2546 return;
2548 for (vid = 0; vid < VLAN_N_VID; vid++) {
2549 if (!vlan_group_get_device(adapter->vlgrp, vid))
2550 continue;
2551 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2555 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2557 struct e1000_hw *hw = &adapter->hw;
2558 u32 manc, manc2h, mdef, i, j;
2560 if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2561 return;
2563 manc = er32(MANC);
2566 * enable receiving management packets to the host. this will probably
2567 * generate destination unreachable messages from the host OS, but
2568 * the packets will be handled on SMBUS
2570 manc |= E1000_MANC_EN_MNG2HOST;
2571 manc2h = er32(MANC2H);
2573 switch (hw->mac.type) {
2574 default:
2575 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2576 break;
2577 case e1000_82574:
2578 case e1000_82583:
2580 * Check if IPMI pass-through decision filter already exists;
2581 * if so, enable it.
2583 for (i = 0, j = 0; i < 8; i++) {
2584 mdef = er32(MDEF(i));
2586 /* Ignore filters with anything other than IPMI ports */
2587 if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2588 continue;
2590 /* Enable this decision filter in MANC2H */
2591 if (mdef)
2592 manc2h |= (1 << i);
2594 j |= mdef;
2597 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2598 break;
2600 /* Create new decision filter in an empty filter */
2601 for (i = 0, j = 0; i < 8; i++)
2602 if (er32(MDEF(i)) == 0) {
2603 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2604 E1000_MDEF_PORT_664));
2605 manc2h |= (1 << 1);
2606 j++;
2607 break;
2610 if (!j)
2611 e_warn("Unable to create IPMI pass-through filter\n");
2612 break;
2615 ew32(MANC2H, manc2h);
2616 ew32(MANC, manc);
2620 * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2621 * @adapter: board private structure
2623 * Configure the Tx unit of the MAC after a reset.
2625 static void e1000_configure_tx(struct e1000_adapter *adapter)
2627 struct e1000_hw *hw = &adapter->hw;
2628 struct e1000_ring *tx_ring = adapter->tx_ring;
2629 u64 tdba;
2630 u32 tdlen, tctl, tipg, tarc;
2631 u32 ipgr1, ipgr2;
2633 /* Setup the HW Tx Head and Tail descriptor pointers */
2634 tdba = tx_ring->dma;
2635 tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2636 ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2637 ew32(TDBAH, (tdba >> 32));
2638 ew32(TDLEN, tdlen);
2639 ew32(TDH, 0);
2640 ew32(TDT, 0);
2641 tx_ring->head = E1000_TDH;
2642 tx_ring->tail = E1000_TDT;
2644 /* Set the default values for the Tx Inter Packet Gap timer */
2645 tipg = DEFAULT_82543_TIPG_IPGT_COPPER; /* 8 */
2646 ipgr1 = DEFAULT_82543_TIPG_IPGR1; /* 8 */
2647 ipgr2 = DEFAULT_82543_TIPG_IPGR2; /* 6 */
2649 if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2650 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /* 7 */
2652 tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2653 tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2654 ew32(TIPG, tipg);
2656 /* Set the Tx Interrupt Delay register */
2657 ew32(TIDV, adapter->tx_int_delay);
2658 /* Tx irq moderation */
2659 ew32(TADV, adapter->tx_abs_int_delay);
2661 if (adapter->flags2 & FLAG2_DMA_BURST) {
2662 u32 txdctl = er32(TXDCTL(0));
2663 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2664 E1000_TXDCTL_WTHRESH);
2666 * set up some performance related parameters to encourage the
2667 * hardware to use the bus more efficiently in bursts, depends
2668 * on the tx_int_delay to be enabled,
2669 * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2670 * hthresh = 1 ==> prefetch when one or more available
2671 * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2672 * BEWARE: this seems to work but should be considered first if
2673 * there are tx hangs or other tx related bugs
2675 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2676 ew32(TXDCTL(0), txdctl);
2677 /* erratum work around: set txdctl the same for both queues */
2678 ew32(TXDCTL(1), txdctl);
2681 /* Program the Transmit Control Register */
2682 tctl = er32(TCTL);
2683 tctl &= ~E1000_TCTL_CT;
2684 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2685 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2687 if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2688 tarc = er32(TARC(0));
2690 * set the speed mode bit, we'll clear it if we're not at
2691 * gigabit link later
2693 #define SPEED_MODE_BIT (1 << 21)
2694 tarc |= SPEED_MODE_BIT;
2695 ew32(TARC(0), tarc);
2698 /* errata: program both queues to unweighted RR */
2699 if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2700 tarc = er32(TARC(0));
2701 tarc |= 1;
2702 ew32(TARC(0), tarc);
2703 tarc = er32(TARC(1));
2704 tarc |= 1;
2705 ew32(TARC(1), tarc);
2708 /* Setup Transmit Descriptor Settings for eop descriptor */
2709 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2711 /* only set IDE if we are delaying interrupts using the timers */
2712 if (adapter->tx_int_delay)
2713 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2715 /* enable Report Status bit */
2716 adapter->txd_cmd |= E1000_TXD_CMD_RS;
2718 ew32(TCTL, tctl);
2720 e1000e_config_collision_dist(hw);
2724 * e1000_setup_rctl - configure the receive control registers
2725 * @adapter: Board private structure
2727 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2728 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2729 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2731 struct e1000_hw *hw = &adapter->hw;
2732 u32 rctl, rfctl;
2733 u32 psrctl = 0;
2734 u32 pages = 0;
2736 /* Workaround Si errata on 82579 - configure jumbo frame flow */
2737 if (hw->mac.type == e1000_pch2lan) {
2738 s32 ret_val;
2740 if (adapter->netdev->mtu > ETH_DATA_LEN)
2741 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2742 else
2743 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2746 /* Program MC offset vector base */
2747 rctl = er32(RCTL);
2748 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2749 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2750 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2751 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2753 /* Do not Store bad packets */
2754 rctl &= ~E1000_RCTL_SBP;
2756 /* Enable Long Packet receive */
2757 if (adapter->netdev->mtu <= ETH_DATA_LEN)
2758 rctl &= ~E1000_RCTL_LPE;
2759 else
2760 rctl |= E1000_RCTL_LPE;
2762 /* Some systems expect that the CRC is included in SMBUS traffic. The
2763 * hardware strips the CRC before sending to both SMBUS (BMC) and to
2764 * host memory when this is enabled
2766 if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2767 rctl |= E1000_RCTL_SECRC;
2769 /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2770 if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2771 u16 phy_data;
2773 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2774 phy_data &= 0xfff8;
2775 phy_data |= (1 << 2);
2776 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2778 e1e_rphy(hw, 22, &phy_data);
2779 phy_data &= 0x0fff;
2780 phy_data |= (1 << 14);
2781 e1e_wphy(hw, 0x10, 0x2823);
2782 e1e_wphy(hw, 0x11, 0x0003);
2783 e1e_wphy(hw, 22, phy_data);
2786 /* Setup buffer sizes */
2787 rctl &= ~E1000_RCTL_SZ_4096;
2788 rctl |= E1000_RCTL_BSEX;
2789 switch (adapter->rx_buffer_len) {
2790 case 2048:
2791 default:
2792 rctl |= E1000_RCTL_SZ_2048;
2793 rctl &= ~E1000_RCTL_BSEX;
2794 break;
2795 case 4096:
2796 rctl |= E1000_RCTL_SZ_4096;
2797 break;
2798 case 8192:
2799 rctl |= E1000_RCTL_SZ_8192;
2800 break;
2801 case 16384:
2802 rctl |= E1000_RCTL_SZ_16384;
2803 break;
2807 * 82571 and greater support packet-split where the protocol
2808 * header is placed in skb->data and the packet data is
2809 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2810 * In the case of a non-split, skb->data is linearly filled,
2811 * followed by the page buffers. Therefore, skb->data is
2812 * sized to hold the largest protocol header.
2814 * allocations using alloc_page take too long for regular MTU
2815 * so only enable packet split for jumbo frames
2817 * Using pages when the page size is greater than 16k wastes
2818 * a lot of memory, since we allocate 3 pages at all times
2819 * per packet.
2821 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2822 if (!(adapter->flags & FLAG_HAS_ERT) && (pages <= 3) &&
2823 (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2824 adapter->rx_ps_pages = pages;
2825 else
2826 adapter->rx_ps_pages = 0;
2828 if (adapter->rx_ps_pages) {
2829 /* Configure extra packet-split registers */
2830 rfctl = er32(RFCTL);
2831 rfctl |= E1000_RFCTL_EXTEN;
2833 * disable packet split support for IPv6 extension headers,
2834 * because some malformed IPv6 headers can hang the Rx
2836 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2837 E1000_RFCTL_NEW_IPV6_EXT_DIS);
2839 ew32(RFCTL, rfctl);
2841 /* Enable Packet split descriptors */
2842 rctl |= E1000_RCTL_DTYP_PS;
2844 psrctl |= adapter->rx_ps_bsize0 >>
2845 E1000_PSRCTL_BSIZE0_SHIFT;
2847 switch (adapter->rx_ps_pages) {
2848 case 3:
2849 psrctl |= PAGE_SIZE <<
2850 E1000_PSRCTL_BSIZE3_SHIFT;
2851 case 2:
2852 psrctl |= PAGE_SIZE <<
2853 E1000_PSRCTL_BSIZE2_SHIFT;
2854 case 1:
2855 psrctl |= PAGE_SIZE >>
2856 E1000_PSRCTL_BSIZE1_SHIFT;
2857 break;
2860 ew32(PSRCTL, psrctl);
2863 ew32(RCTL, rctl);
2864 /* just started the receive unit, no need to restart */
2865 adapter->flags &= ~FLAG_RX_RESTART_NOW;
2869 * e1000_configure_rx - Configure Receive Unit after Reset
2870 * @adapter: board private structure
2872 * Configure the Rx unit of the MAC after a reset.
2874 static void e1000_configure_rx(struct e1000_adapter *adapter)
2876 struct e1000_hw *hw = &adapter->hw;
2877 struct e1000_ring *rx_ring = adapter->rx_ring;
2878 u64 rdba;
2879 u32 rdlen, rctl, rxcsum, ctrl_ext;
2881 if (adapter->rx_ps_pages) {
2882 /* this is a 32 byte descriptor */
2883 rdlen = rx_ring->count *
2884 sizeof(union e1000_rx_desc_packet_split);
2885 adapter->clean_rx = e1000_clean_rx_irq_ps;
2886 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2887 } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2888 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2889 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2890 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2891 } else {
2892 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2893 adapter->clean_rx = e1000_clean_rx_irq;
2894 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2897 /* disable receives while setting up the descriptors */
2898 rctl = er32(RCTL);
2899 ew32(RCTL, rctl & ~E1000_RCTL_EN);
2900 e1e_flush();
2901 msleep(10);
2903 if (adapter->flags2 & FLAG2_DMA_BURST) {
2905 * set the writeback threshold (only takes effect if the RDTR
2906 * is set). set GRAN=1 and write back up to 0x4 worth, and
2907 * enable prefetching of 0x20 rx descriptors
2908 * granularity = 01
2909 * wthresh = 04,
2910 * hthresh = 04,
2911 * pthresh = 0x20
2913 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
2914 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
2917 * override the delay timers for enabling bursting, only if
2918 * the value was not set by the user via module options
2920 if (adapter->rx_int_delay == DEFAULT_RDTR)
2921 adapter->rx_int_delay = BURST_RDTR;
2922 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
2923 adapter->rx_abs_int_delay = BURST_RADV;
2926 /* set the Receive Delay Timer Register */
2927 ew32(RDTR, adapter->rx_int_delay);
2929 /* irq moderation */
2930 ew32(RADV, adapter->rx_abs_int_delay);
2931 if ((adapter->itr_setting != 0) && (adapter->itr != 0))
2932 ew32(ITR, 1000000000 / (adapter->itr * 256));
2934 ctrl_ext = er32(CTRL_EXT);
2935 /* Auto-Mask interrupts upon ICR access */
2936 ctrl_ext |= E1000_CTRL_EXT_IAME;
2937 ew32(IAM, 0xffffffff);
2938 ew32(CTRL_EXT, ctrl_ext);
2939 e1e_flush();
2942 * Setup the HW Rx Head and Tail Descriptor Pointers and
2943 * the Base and Length of the Rx Descriptor Ring
2945 rdba = rx_ring->dma;
2946 ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2947 ew32(RDBAH, (rdba >> 32));
2948 ew32(RDLEN, rdlen);
2949 ew32(RDH, 0);
2950 ew32(RDT, 0);
2951 rx_ring->head = E1000_RDH;
2952 rx_ring->tail = E1000_RDT;
2954 /* Enable Receive Checksum Offload for TCP and UDP */
2955 rxcsum = er32(RXCSUM);
2956 if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2957 rxcsum |= E1000_RXCSUM_TUOFL;
2960 * IPv4 payload checksum for UDP fragments must be
2961 * used in conjunction with packet-split.
2963 if (adapter->rx_ps_pages)
2964 rxcsum |= E1000_RXCSUM_IPPCSE;
2965 } else {
2966 rxcsum &= ~E1000_RXCSUM_TUOFL;
2967 /* no need to clear IPPCSE as it defaults to 0 */
2969 ew32(RXCSUM, rxcsum);
2972 * Enable early receives on supported devices, only takes effect when
2973 * packet size is equal or larger than the specified value (in 8 byte
2974 * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2976 if ((adapter->flags & FLAG_HAS_ERT) ||
2977 (adapter->hw.mac.type == e1000_pch2lan)) {
2978 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2979 u32 rxdctl = er32(RXDCTL(0));
2980 ew32(RXDCTL(0), rxdctl | 0x3);
2981 if (adapter->flags & FLAG_HAS_ERT)
2982 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2984 * With jumbo frames and early-receive enabled,
2985 * excessive C-state transition latencies result in
2986 * dropped transactions.
2988 pm_qos_update_request(
2989 &adapter->netdev->pm_qos_req, 55);
2990 } else {
2991 pm_qos_update_request(
2992 &adapter->netdev->pm_qos_req,
2993 PM_QOS_DEFAULT_VALUE);
2997 /* Enable Receives */
2998 ew32(RCTL, rctl);
3002 * e1000_update_mc_addr_list - Update Multicast addresses
3003 * @hw: pointer to the HW structure
3004 * @mc_addr_list: array of multicast addresses to program
3005 * @mc_addr_count: number of multicast addresses to program
3007 * Updates the Multicast Table Array.
3008 * The caller must have a packed mc_addr_list of multicast addresses.
3010 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
3011 u32 mc_addr_count)
3013 hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
3017 * e1000_set_multi - Multicast and Promiscuous mode set
3018 * @netdev: network interface device structure
3020 * The set_multi entry point is called whenever the multicast address
3021 * list or the network interface flags are updated. This routine is
3022 * responsible for configuring the hardware for proper multicast,
3023 * promiscuous mode, and all-multi behavior.
3025 static void e1000_set_multi(struct net_device *netdev)
3027 struct e1000_adapter *adapter = netdev_priv(netdev);
3028 struct e1000_hw *hw = &adapter->hw;
3029 struct netdev_hw_addr *ha;
3030 u8 *mta_list;
3031 u32 rctl;
3032 int i;
3034 /* Check for Promiscuous and All Multicast modes */
3036 rctl = er32(RCTL);
3038 if (netdev->flags & IFF_PROMISC) {
3039 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3040 rctl &= ~E1000_RCTL_VFE;
3041 } else {
3042 if (netdev->flags & IFF_ALLMULTI) {
3043 rctl |= E1000_RCTL_MPE;
3044 rctl &= ~E1000_RCTL_UPE;
3045 } else {
3046 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3048 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
3049 rctl |= E1000_RCTL_VFE;
3052 ew32(RCTL, rctl);
3054 if (!netdev_mc_empty(netdev)) {
3055 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3056 if (!mta_list)
3057 return;
3059 /* prepare a packed array of only addresses. */
3060 i = 0;
3061 netdev_for_each_mc_addr(ha, netdev)
3062 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3064 e1000_update_mc_addr_list(hw, mta_list, i);
3065 kfree(mta_list);
3066 } else {
3068 * if we're called from probe, we might not have
3069 * anything to do here, so clear out the list
3071 e1000_update_mc_addr_list(hw, NULL, 0);
3076 * e1000_configure - configure the hardware for Rx and Tx
3077 * @adapter: private board structure
3079 static void e1000_configure(struct e1000_adapter *adapter)
3081 e1000_set_multi(adapter->netdev);
3083 e1000_restore_vlan(adapter);
3084 e1000_init_manageability_pt(adapter);
3086 e1000_configure_tx(adapter);
3087 e1000_setup_rctl(adapter);
3088 e1000_configure_rx(adapter);
3089 adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
3093 * e1000e_power_up_phy - restore link in case the phy was powered down
3094 * @adapter: address of board private structure
3096 * The phy may be powered down to save power and turn off link when the
3097 * driver is unloaded and wake on lan is not enabled (among others)
3098 * *** this routine MUST be followed by a call to e1000e_reset ***
3100 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3102 if (adapter->hw.phy.ops.power_up)
3103 adapter->hw.phy.ops.power_up(&adapter->hw);
3105 adapter->hw.mac.ops.setup_link(&adapter->hw);
3109 * e1000_power_down_phy - Power down the PHY
3111 * Power down the PHY so no link is implied when interface is down.
3112 * The PHY cannot be powered down if management or WoL is active.
3114 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3116 /* WoL is enabled */
3117 if (adapter->wol)
3118 return;
3120 if (adapter->hw.phy.ops.power_down)
3121 adapter->hw.phy.ops.power_down(&adapter->hw);
3125 * e1000e_reset - bring the hardware into a known good state
3127 * This function boots the hardware and enables some settings that
3128 * require a configuration cycle of the hardware - those cannot be
3129 * set/changed during runtime. After reset the device needs to be
3130 * properly configured for Rx, Tx etc.
3132 void e1000e_reset(struct e1000_adapter *adapter)
3134 struct e1000_mac_info *mac = &adapter->hw.mac;
3135 struct e1000_fc_info *fc = &adapter->hw.fc;
3136 struct e1000_hw *hw = &adapter->hw;
3137 u32 tx_space, min_tx_space, min_rx_space;
3138 u32 pba = adapter->pba;
3139 u16 hwm;
3141 /* reset Packet Buffer Allocation to default */
3142 ew32(PBA, pba);
3144 if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3146 * To maintain wire speed transmits, the Tx FIFO should be
3147 * large enough to accommodate two full transmit packets,
3148 * rounded up to the next 1KB and expressed in KB. Likewise,
3149 * the Rx FIFO should be large enough to accommodate at least
3150 * one full receive packet and is similarly rounded up and
3151 * expressed in KB.
3153 pba = er32(PBA);
3154 /* upper 16 bits has Tx packet buffer allocation size in KB */
3155 tx_space = pba >> 16;
3156 /* lower 16 bits has Rx packet buffer allocation size in KB */
3157 pba &= 0xffff;
3159 * the Tx fifo also stores 16 bytes of information about the tx
3160 * but don't include ethernet FCS because hardware appends it
3162 min_tx_space = (adapter->max_frame_size +
3163 sizeof(struct e1000_tx_desc) -
3164 ETH_FCS_LEN) * 2;
3165 min_tx_space = ALIGN(min_tx_space, 1024);
3166 min_tx_space >>= 10;
3167 /* software strips receive CRC, so leave room for it */
3168 min_rx_space = adapter->max_frame_size;
3169 min_rx_space = ALIGN(min_rx_space, 1024);
3170 min_rx_space >>= 10;
3173 * If current Tx allocation is less than the min Tx FIFO size,
3174 * and the min Tx FIFO size is less than the current Rx FIFO
3175 * allocation, take space away from current Rx allocation
3177 if ((tx_space < min_tx_space) &&
3178 ((min_tx_space - tx_space) < pba)) {
3179 pba -= min_tx_space - tx_space;
3182 * if short on Rx space, Rx wins and must trump tx
3183 * adjustment or use Early Receive if available
3185 if ((pba < min_rx_space) &&
3186 (!(adapter->flags & FLAG_HAS_ERT)))
3187 /* ERT enabled in e1000_configure_rx */
3188 pba = min_rx_space;
3191 ew32(PBA, pba);
3196 * flow control settings
3198 * The high water mark must be low enough to fit one full frame
3199 * (or the size used for early receive) above it in the Rx FIFO.
3200 * Set it to the lower of:
3201 * - 90% of the Rx FIFO size, and
3202 * - the full Rx FIFO size minus the early receive size (for parts
3203 * with ERT support assuming ERT set to E1000_ERT_2048), or
3204 * - the full Rx FIFO size minus one full frame
3206 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3207 fc->pause_time = 0xFFFF;
3208 else
3209 fc->pause_time = E1000_FC_PAUSE_TIME;
3210 fc->send_xon = 1;
3211 fc->current_mode = fc->requested_mode;
3213 switch (hw->mac.type) {
3214 default:
3215 if ((adapter->flags & FLAG_HAS_ERT) &&
3216 (adapter->netdev->mtu > ETH_DATA_LEN))
3217 hwm = min(((pba << 10) * 9 / 10),
3218 ((pba << 10) - (E1000_ERT_2048 << 3)));
3219 else
3220 hwm = min(((pba << 10) * 9 / 10),
3221 ((pba << 10) - adapter->max_frame_size));
3223 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3224 fc->low_water = fc->high_water - 8;
3225 break;
3226 case e1000_pchlan:
3228 * Workaround PCH LOM adapter hangs with certain network
3229 * loads. If hangs persist, try disabling Tx flow control.
3231 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3232 fc->high_water = 0x3500;
3233 fc->low_water = 0x1500;
3234 } else {
3235 fc->high_water = 0x5000;
3236 fc->low_water = 0x3000;
3238 fc->refresh_time = 0x1000;
3239 break;
3240 case e1000_pch2lan:
3241 fc->high_water = 0x05C20;
3242 fc->low_water = 0x05048;
3243 fc->pause_time = 0x0650;
3244 fc->refresh_time = 0x0400;
3245 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3246 pba = 14;
3247 ew32(PBA, pba);
3249 break;
3253 * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3254 * fit in receive buffer and early-receive not supported.
3256 if (adapter->itr_setting & 0x3) {
3257 if (((adapter->max_frame_size * 2) > (pba << 10)) &&
3258 !(adapter->flags & FLAG_HAS_ERT)) {
3259 if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3260 dev_info(&adapter->pdev->dev,
3261 "Interrupt Throttle Rate turned off\n");
3262 adapter->flags2 |= FLAG2_DISABLE_AIM;
3263 ew32(ITR, 0);
3265 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3266 dev_info(&adapter->pdev->dev,
3267 "Interrupt Throttle Rate turned on\n");
3268 adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3269 adapter->itr = 20000;
3270 ew32(ITR, 1000000000 / (adapter->itr * 256));
3274 /* Allow time for pending master requests to run */
3275 mac->ops.reset_hw(hw);
3278 * For parts with AMT enabled, let the firmware know
3279 * that the network interface is in control
3281 if (adapter->flags & FLAG_HAS_AMT)
3282 e1000_get_hw_control(adapter);
3284 ew32(WUC, 0);
3286 if (mac->ops.init_hw(hw))
3287 e_err("Hardware Error\n");
3289 e1000_update_mng_vlan(adapter);
3291 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3292 ew32(VET, ETH_P_8021Q);
3294 e1000e_reset_adaptive(hw);
3295 e1000_get_phy_info(hw);
3297 if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3298 !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3299 u16 phy_data = 0;
3301 * speed up time to link by disabling smart power down, ignore
3302 * the return value of this function because there is nothing
3303 * different we would do if it failed
3305 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3306 phy_data &= ~IGP02E1000_PM_SPD;
3307 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3311 int e1000e_up(struct e1000_adapter *adapter)
3313 struct e1000_hw *hw = &adapter->hw;
3315 /* hardware has been reset, we need to reload some things */
3316 e1000_configure(adapter);
3318 clear_bit(__E1000_DOWN, &adapter->state);
3320 napi_enable(&adapter->napi);
3321 if (adapter->msix_entries)
3322 e1000_configure_msix(adapter);
3323 e1000_irq_enable(adapter);
3325 netif_wake_queue(adapter->netdev);
3327 /* fire a link change interrupt to start the watchdog */
3328 if (adapter->msix_entries)
3329 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3330 else
3331 ew32(ICS, E1000_ICS_LSC);
3333 return 0;
3336 void e1000e_down(struct e1000_adapter *adapter)
3338 struct net_device *netdev = adapter->netdev;
3339 struct e1000_hw *hw = &adapter->hw;
3340 u32 tctl, rctl;
3343 * signal that we're down so the interrupt handler does not
3344 * reschedule our watchdog timer
3346 set_bit(__E1000_DOWN, &adapter->state);
3348 /* disable receives in the hardware */
3349 rctl = er32(RCTL);
3350 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3351 /* flush and sleep below */
3353 netif_stop_queue(netdev);
3355 /* disable transmits in the hardware */
3356 tctl = er32(TCTL);
3357 tctl &= ~E1000_TCTL_EN;
3358 ew32(TCTL, tctl);
3359 /* flush both disables and wait for them to finish */
3360 e1e_flush();
3361 msleep(10);
3363 napi_disable(&adapter->napi);
3364 e1000_irq_disable(adapter);
3366 del_timer_sync(&adapter->watchdog_timer);
3367 del_timer_sync(&adapter->phy_info_timer);
3369 netif_carrier_off(netdev);
3370 adapter->link_speed = 0;
3371 adapter->link_duplex = 0;
3373 if (!pci_channel_offline(adapter->pdev))
3374 e1000e_reset(adapter);
3375 e1000_clean_tx_ring(adapter);
3376 e1000_clean_rx_ring(adapter);
3379 * TODO: for power management, we could drop the link and
3380 * pci_disable_device here.
3384 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3386 might_sleep();
3387 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3388 msleep(1);
3389 e1000e_down(adapter);
3390 e1000e_up(adapter);
3391 clear_bit(__E1000_RESETTING, &adapter->state);
3395 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3396 * @adapter: board private structure to initialize
3398 * e1000_sw_init initializes the Adapter private data structure.
3399 * Fields are initialized based on PCI device information and
3400 * OS network device settings (MTU size).
3402 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3404 struct net_device *netdev = adapter->netdev;
3406 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3407 adapter->rx_ps_bsize0 = 128;
3408 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3409 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3411 e1000e_set_interrupt_capability(adapter);
3413 if (e1000_alloc_queues(adapter))
3414 return -ENOMEM;
3416 /* Explicitly disable IRQ since the NIC can be in any state. */
3417 e1000_irq_disable(adapter);
3419 set_bit(__E1000_DOWN, &adapter->state);
3420 return 0;
3424 * e1000_intr_msi_test - Interrupt Handler
3425 * @irq: interrupt number
3426 * @data: pointer to a network interface device structure
3428 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3430 struct net_device *netdev = data;
3431 struct e1000_adapter *adapter = netdev_priv(netdev);
3432 struct e1000_hw *hw = &adapter->hw;
3433 u32 icr = er32(ICR);
3435 e_dbg("icr is %08X\n", icr);
3436 if (icr & E1000_ICR_RXSEQ) {
3437 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3438 wmb();
3441 return IRQ_HANDLED;
3445 * e1000_test_msi_interrupt - Returns 0 for successful test
3446 * @adapter: board private struct
3448 * code flow taken from tg3.c
3450 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3452 struct net_device *netdev = adapter->netdev;
3453 struct e1000_hw *hw = &adapter->hw;
3454 int err;
3456 /* poll_enable hasn't been called yet, so don't need disable */
3457 /* clear any pending events */
3458 er32(ICR);
3460 /* free the real vector and request a test handler */
3461 e1000_free_irq(adapter);
3462 e1000e_reset_interrupt_capability(adapter);
3464 /* Assume that the test fails, if it succeeds then the test
3465 * MSI irq handler will unset this flag */
3466 adapter->flags |= FLAG_MSI_TEST_FAILED;
3468 err = pci_enable_msi(adapter->pdev);
3469 if (err)
3470 goto msi_test_failed;
3472 err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3473 netdev->name, netdev);
3474 if (err) {
3475 pci_disable_msi(adapter->pdev);
3476 goto msi_test_failed;
3479 wmb();
3481 e1000_irq_enable(adapter);
3483 /* fire an unusual interrupt on the test handler */
3484 ew32(ICS, E1000_ICS_RXSEQ);
3485 e1e_flush();
3486 msleep(50);
3488 e1000_irq_disable(adapter);
3490 rmb();
3492 if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3493 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3494 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3495 } else
3496 e_dbg("MSI interrupt test succeeded!\n");
3498 free_irq(adapter->pdev->irq, netdev);
3499 pci_disable_msi(adapter->pdev);
3501 msi_test_failed:
3502 e1000e_set_interrupt_capability(adapter);
3503 return e1000_request_irq(adapter);
3507 * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3508 * @adapter: board private struct
3510 * code flow taken from tg3.c, called with e1000 interrupts disabled.
3512 static int e1000_test_msi(struct e1000_adapter *adapter)
3514 int err;
3515 u16 pci_cmd;
3517 if (!(adapter->flags & FLAG_MSI_ENABLED))
3518 return 0;
3520 /* disable SERR in case the MSI write causes a master abort */
3521 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3522 if (pci_cmd & PCI_COMMAND_SERR)
3523 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3524 pci_cmd & ~PCI_COMMAND_SERR);
3526 err = e1000_test_msi_interrupt(adapter);
3528 /* re-enable SERR */
3529 if (pci_cmd & PCI_COMMAND_SERR) {
3530 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3531 pci_cmd |= PCI_COMMAND_SERR;
3532 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3535 return err;
3539 * e1000_open - Called when a network interface is made active
3540 * @netdev: network interface device structure
3542 * Returns 0 on success, negative value on failure
3544 * The open entry point is called when a network interface is made
3545 * active by the system (IFF_UP). At this point all resources needed
3546 * for transmit and receive operations are allocated, the interrupt
3547 * handler is registered with the OS, the watchdog timer is started,
3548 * and the stack is notified that the interface is ready.
3550 static int e1000_open(struct net_device *netdev)
3552 struct e1000_adapter *adapter = netdev_priv(netdev);
3553 struct e1000_hw *hw = &adapter->hw;
3554 struct pci_dev *pdev = adapter->pdev;
3555 int err;
3557 /* disallow open during test */
3558 if (test_bit(__E1000_TESTING, &adapter->state))
3559 return -EBUSY;
3561 pm_runtime_get_sync(&pdev->dev);
3563 netif_carrier_off(netdev);
3565 /* allocate transmit descriptors */
3566 err = e1000e_setup_tx_resources(adapter);
3567 if (err)
3568 goto err_setup_tx;
3570 /* allocate receive descriptors */
3571 err = e1000e_setup_rx_resources(adapter);
3572 if (err)
3573 goto err_setup_rx;
3576 * If AMT is enabled, let the firmware know that the network
3577 * interface is now open and reset the part to a known state.
3579 if (adapter->flags & FLAG_HAS_AMT) {
3580 e1000_get_hw_control(adapter);
3581 e1000e_reset(adapter);
3584 e1000e_power_up_phy(adapter);
3586 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3587 if ((adapter->hw.mng_cookie.status &
3588 E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3589 e1000_update_mng_vlan(adapter);
3591 /* DMA latency requirement to workaround early-receive/jumbo issue */
3592 if ((adapter->flags & FLAG_HAS_ERT) ||
3593 (adapter->hw.mac.type == e1000_pch2lan))
3594 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3595 PM_QOS_CPU_DMA_LATENCY,
3596 PM_QOS_DEFAULT_VALUE);
3599 * before we allocate an interrupt, we must be ready to handle it.
3600 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3601 * as soon as we call pci_request_irq, so we have to setup our
3602 * clean_rx handler before we do so.
3604 e1000_configure(adapter);
3606 err = e1000_request_irq(adapter);
3607 if (err)
3608 goto err_req_irq;
3611 * Work around PCIe errata with MSI interrupts causing some chipsets to
3612 * ignore e1000e MSI messages, which means we need to test our MSI
3613 * interrupt now
3615 if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3616 err = e1000_test_msi(adapter);
3617 if (err) {
3618 e_err("Interrupt allocation failed\n");
3619 goto err_req_irq;
3623 /* From here on the code is the same as e1000e_up() */
3624 clear_bit(__E1000_DOWN, &adapter->state);
3626 napi_enable(&adapter->napi);
3628 e1000_irq_enable(adapter);
3630 netif_start_queue(netdev);
3632 adapter->idle_check = true;
3633 pm_runtime_put(&pdev->dev);
3635 /* fire a link status change interrupt to start the watchdog */
3636 if (adapter->msix_entries)
3637 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3638 else
3639 ew32(ICS, E1000_ICS_LSC);
3641 return 0;
3643 err_req_irq:
3644 e1000_release_hw_control(adapter);
3645 e1000_power_down_phy(adapter);
3646 e1000e_free_rx_resources(adapter);
3647 err_setup_rx:
3648 e1000e_free_tx_resources(adapter);
3649 err_setup_tx:
3650 e1000e_reset(adapter);
3651 pm_runtime_put_sync(&pdev->dev);
3653 return err;
3657 * e1000_close - Disables a network interface
3658 * @netdev: network interface device structure
3660 * Returns 0, this is not allowed to fail
3662 * The close entry point is called when an interface is de-activated
3663 * by the OS. The hardware is still under the drivers control, but
3664 * needs to be disabled. A global MAC reset is issued to stop the
3665 * hardware, and all transmit and receive resources are freed.
3667 static int e1000_close(struct net_device *netdev)
3669 struct e1000_adapter *adapter = netdev_priv(netdev);
3670 struct pci_dev *pdev = adapter->pdev;
3672 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3674 pm_runtime_get_sync(&pdev->dev);
3676 if (!test_bit(__E1000_DOWN, &adapter->state)) {
3677 e1000e_down(adapter);
3678 e1000_free_irq(adapter);
3680 e1000_power_down_phy(adapter);
3682 e1000e_free_tx_resources(adapter);
3683 e1000e_free_rx_resources(adapter);
3686 * kill manageability vlan ID if supported, but not if a vlan with
3687 * the same ID is registered on the host OS (let 8021q kill it)
3689 if ((adapter->hw.mng_cookie.status &
3690 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3691 !(adapter->vlgrp &&
3692 vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3693 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3696 * If AMT is enabled, let the firmware know that the network
3697 * interface is now closed
3699 if (adapter->flags & FLAG_HAS_AMT)
3700 e1000_release_hw_control(adapter);
3702 if ((adapter->flags & FLAG_HAS_ERT) ||
3703 (adapter->hw.mac.type == e1000_pch2lan))
3704 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3706 pm_runtime_put_sync(&pdev->dev);
3708 return 0;
3711 * e1000_set_mac - Change the Ethernet Address of the NIC
3712 * @netdev: network interface device structure
3713 * @p: pointer to an address structure
3715 * Returns 0 on success, negative on failure
3717 static int e1000_set_mac(struct net_device *netdev, void *p)
3719 struct e1000_adapter *adapter = netdev_priv(netdev);
3720 struct sockaddr *addr = p;
3722 if (!is_valid_ether_addr(addr->sa_data))
3723 return -EADDRNOTAVAIL;
3725 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3726 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3728 e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3730 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3731 /* activate the work around */
3732 e1000e_set_laa_state_82571(&adapter->hw, 1);
3735 * Hold a copy of the LAA in RAR[14] This is done so that
3736 * between the time RAR[0] gets clobbered and the time it
3737 * gets fixed (in e1000_watchdog), the actual LAA is in one
3738 * of the RARs and no incoming packets directed to this port
3739 * are dropped. Eventually the LAA will be in RAR[0] and
3740 * RAR[14]
3742 e1000e_rar_set(&adapter->hw,
3743 adapter->hw.mac.addr,
3744 adapter->hw.mac.rar_entry_count - 1);
3747 return 0;
3751 * e1000e_update_phy_task - work thread to update phy
3752 * @work: pointer to our work struct
3754 * this worker thread exists because we must acquire a
3755 * semaphore to read the phy, which we could msleep while
3756 * waiting for it, and we can't msleep in a timer.
3758 static void e1000e_update_phy_task(struct work_struct *work)
3760 struct e1000_adapter *adapter = container_of(work,
3761 struct e1000_adapter, update_phy_task);
3762 e1000_get_phy_info(&adapter->hw);
3766 * Need to wait a few seconds after link up to get diagnostic information from
3767 * the phy
3769 static void e1000_update_phy_info(unsigned long data)
3771 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3772 schedule_work(&adapter->update_phy_task);
3776 * e1000e_update_phy_stats - Update the PHY statistics counters
3777 * @adapter: board private structure
3779 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
3781 struct e1000_hw *hw = &adapter->hw;
3782 s32 ret_val;
3783 u16 phy_data;
3785 ret_val = hw->phy.ops.acquire(hw);
3786 if (ret_val)
3787 return;
3789 hw->phy.addr = 1;
3791 #define HV_PHY_STATS_PAGE 778
3793 * A page set is expensive so check if already on desired page.
3794 * If not, set to the page with the PHY status registers.
3796 ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
3797 &phy_data);
3798 if (ret_val)
3799 goto release;
3800 if (phy_data != (HV_PHY_STATS_PAGE << IGP_PAGE_SHIFT)) {
3801 ret_val = e1000e_write_phy_reg_mdic(hw,
3802 IGP01E1000_PHY_PAGE_SELECT,
3803 (HV_PHY_STATS_PAGE <<
3804 IGP_PAGE_SHIFT));
3805 if (ret_val)
3806 goto release;
3809 /* Read/clear the upper 16-bit registers and read/accumulate lower */
3811 /* Single Collision Count */
3812 e1000e_read_phy_reg_mdic(hw, HV_SCC_UPPER & MAX_PHY_REG_ADDRESS,
3813 &phy_data);
3814 ret_val = e1000e_read_phy_reg_mdic(hw,
3815 HV_SCC_LOWER & MAX_PHY_REG_ADDRESS,
3816 &phy_data);
3817 if (!ret_val)
3818 adapter->stats.scc += phy_data;
3820 /* Excessive Collision Count */
3821 e1000e_read_phy_reg_mdic(hw, HV_ECOL_UPPER & MAX_PHY_REG_ADDRESS,
3822 &phy_data);
3823 ret_val = e1000e_read_phy_reg_mdic(hw,
3824 HV_ECOL_LOWER & MAX_PHY_REG_ADDRESS,
3825 &phy_data);
3826 if (!ret_val)
3827 adapter->stats.ecol += phy_data;
3829 /* Multiple Collision Count */
3830 e1000e_read_phy_reg_mdic(hw, HV_MCC_UPPER & MAX_PHY_REG_ADDRESS,
3831 &phy_data);
3832 ret_val = e1000e_read_phy_reg_mdic(hw,
3833 HV_MCC_LOWER & MAX_PHY_REG_ADDRESS,
3834 &phy_data);
3835 if (!ret_val)
3836 adapter->stats.mcc += phy_data;
3838 /* Late Collision Count */
3839 e1000e_read_phy_reg_mdic(hw, HV_LATECOL_UPPER & MAX_PHY_REG_ADDRESS,
3840 &phy_data);
3841 ret_val = e1000e_read_phy_reg_mdic(hw,
3842 HV_LATECOL_LOWER &
3843 MAX_PHY_REG_ADDRESS,
3844 &phy_data);
3845 if (!ret_val)
3846 adapter->stats.latecol += phy_data;
3848 /* Collision Count - also used for adaptive IFS */
3849 e1000e_read_phy_reg_mdic(hw, HV_COLC_UPPER & MAX_PHY_REG_ADDRESS,
3850 &phy_data);
3851 ret_val = e1000e_read_phy_reg_mdic(hw,
3852 HV_COLC_LOWER & MAX_PHY_REG_ADDRESS,
3853 &phy_data);
3854 if (!ret_val)
3855 hw->mac.collision_delta = phy_data;
3857 /* Defer Count */
3858 e1000e_read_phy_reg_mdic(hw, HV_DC_UPPER & MAX_PHY_REG_ADDRESS,
3859 &phy_data);
3860 ret_val = e1000e_read_phy_reg_mdic(hw,
3861 HV_DC_LOWER & MAX_PHY_REG_ADDRESS,
3862 &phy_data);
3863 if (!ret_val)
3864 adapter->stats.dc += phy_data;
3866 /* Transmit with no CRS */
3867 e1000e_read_phy_reg_mdic(hw, HV_TNCRS_UPPER & MAX_PHY_REG_ADDRESS,
3868 &phy_data);
3869 ret_val = e1000e_read_phy_reg_mdic(hw,
3870 HV_TNCRS_LOWER & MAX_PHY_REG_ADDRESS,
3871 &phy_data);
3872 if (!ret_val)
3873 adapter->stats.tncrs += phy_data;
3875 release:
3876 hw->phy.ops.release(hw);
3880 * e1000e_update_stats - Update the board statistics counters
3881 * @adapter: board private structure
3883 void e1000e_update_stats(struct e1000_adapter *adapter)
3885 struct net_device *netdev = adapter->netdev;
3886 struct e1000_hw *hw = &adapter->hw;
3887 struct pci_dev *pdev = adapter->pdev;
3890 * Prevent stats update while adapter is being reset, or if the pci
3891 * connection is down.
3893 if (adapter->link_speed == 0)
3894 return;
3895 if (pci_channel_offline(pdev))
3896 return;
3898 adapter->stats.crcerrs += er32(CRCERRS);
3899 adapter->stats.gprc += er32(GPRC);
3900 adapter->stats.gorc += er32(GORCL);
3901 er32(GORCH); /* Clear gorc */
3902 adapter->stats.bprc += er32(BPRC);
3903 adapter->stats.mprc += er32(MPRC);
3904 adapter->stats.roc += er32(ROC);
3906 adapter->stats.mpc += er32(MPC);
3908 /* Half-duplex statistics */
3909 if (adapter->link_duplex == HALF_DUPLEX) {
3910 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
3911 e1000e_update_phy_stats(adapter);
3912 } else {
3913 adapter->stats.scc += er32(SCC);
3914 adapter->stats.ecol += er32(ECOL);
3915 adapter->stats.mcc += er32(MCC);
3916 adapter->stats.latecol += er32(LATECOL);
3917 adapter->stats.dc += er32(DC);
3919 hw->mac.collision_delta = er32(COLC);
3921 if ((hw->mac.type != e1000_82574) &&
3922 (hw->mac.type != e1000_82583))
3923 adapter->stats.tncrs += er32(TNCRS);
3925 adapter->stats.colc += hw->mac.collision_delta;
3928 adapter->stats.xonrxc += er32(XONRXC);
3929 adapter->stats.xontxc += er32(XONTXC);
3930 adapter->stats.xoffrxc += er32(XOFFRXC);
3931 adapter->stats.xofftxc += er32(XOFFTXC);
3932 adapter->stats.gptc += er32(GPTC);
3933 adapter->stats.gotc += er32(GOTCL);
3934 er32(GOTCH); /* Clear gotc */
3935 adapter->stats.rnbc += er32(RNBC);
3936 adapter->stats.ruc += er32(RUC);
3938 adapter->stats.mptc += er32(MPTC);
3939 adapter->stats.bptc += er32(BPTC);
3941 /* used for adaptive IFS */
3943 hw->mac.tx_packet_delta = er32(TPT);
3944 adapter->stats.tpt += hw->mac.tx_packet_delta;
3946 adapter->stats.algnerrc += er32(ALGNERRC);
3947 adapter->stats.rxerrc += er32(RXERRC);
3948 adapter->stats.cexterr += er32(CEXTERR);
3949 adapter->stats.tsctc += er32(TSCTC);
3950 adapter->stats.tsctfc += er32(TSCTFC);
3952 /* Fill out the OS statistics structure */
3953 netdev->stats.multicast = adapter->stats.mprc;
3954 netdev->stats.collisions = adapter->stats.colc;
3956 /* Rx Errors */
3959 * RLEC on some newer hardware can be incorrect so build
3960 * our own version based on RUC and ROC
3962 netdev->stats.rx_errors = adapter->stats.rxerrc +
3963 adapter->stats.crcerrs + adapter->stats.algnerrc +
3964 adapter->stats.ruc + adapter->stats.roc +
3965 adapter->stats.cexterr;
3966 netdev->stats.rx_length_errors = adapter->stats.ruc +
3967 adapter->stats.roc;
3968 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3969 netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3970 netdev->stats.rx_missed_errors = adapter->stats.mpc;
3972 /* Tx Errors */
3973 netdev->stats.tx_errors = adapter->stats.ecol +
3974 adapter->stats.latecol;
3975 netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3976 netdev->stats.tx_window_errors = adapter->stats.latecol;
3977 netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3979 /* Tx Dropped needs to be maintained elsewhere */
3981 /* Management Stats */
3982 adapter->stats.mgptc += er32(MGTPTC);
3983 adapter->stats.mgprc += er32(MGTPRC);
3984 adapter->stats.mgpdc += er32(MGTPDC);
3988 * e1000_phy_read_status - Update the PHY register status snapshot
3989 * @adapter: board private structure
3991 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3993 struct e1000_hw *hw = &adapter->hw;
3994 struct e1000_phy_regs *phy = &adapter->phy_regs;
3995 int ret_val;
3997 if ((er32(STATUS) & E1000_STATUS_LU) &&
3998 (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3999 ret_val = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4000 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4001 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4002 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4003 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4004 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4005 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4006 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4007 if (ret_val)
4008 e_warn("Error reading PHY register\n");
4009 } else {
4011 * Do not read PHY registers if link is not up
4012 * Set values to typical power-on defaults
4014 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4015 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4016 BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4017 BMSR_ERCAP);
4018 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4019 ADVERTISE_ALL | ADVERTISE_CSMA);
4020 phy->lpa = 0;
4021 phy->expansion = EXPANSION_ENABLENPAGE;
4022 phy->ctrl1000 = ADVERTISE_1000FULL;
4023 phy->stat1000 = 0;
4024 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4028 static void e1000_print_link_info(struct e1000_adapter *adapter)
4030 struct e1000_hw *hw = &adapter->hw;
4031 u32 ctrl = er32(CTRL);
4033 /* Link status message must follow this format for user tools */
4034 printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
4035 "Flow Control: %s\n",
4036 adapter->netdev->name,
4037 adapter->link_speed,
4038 (adapter->link_duplex == FULL_DUPLEX) ?
4039 "Full Duplex" : "Half Duplex",
4040 ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
4041 "RX/TX" :
4042 ((ctrl & E1000_CTRL_RFCE) ? "RX" :
4043 ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
4046 static bool e1000e_has_link(struct e1000_adapter *adapter)
4048 struct e1000_hw *hw = &adapter->hw;
4049 bool link_active = 0;
4050 s32 ret_val = 0;
4053 * get_link_status is set on LSC (link status) interrupt or
4054 * Rx sequence error interrupt. get_link_status will stay
4055 * false until the check_for_link establishes link
4056 * for copper adapters ONLY
4058 switch (hw->phy.media_type) {
4059 case e1000_media_type_copper:
4060 if (hw->mac.get_link_status) {
4061 ret_val = hw->mac.ops.check_for_link(hw);
4062 link_active = !hw->mac.get_link_status;
4063 } else {
4064 link_active = 1;
4066 break;
4067 case e1000_media_type_fiber:
4068 ret_val = hw->mac.ops.check_for_link(hw);
4069 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4070 break;
4071 case e1000_media_type_internal_serdes:
4072 ret_val = hw->mac.ops.check_for_link(hw);
4073 link_active = adapter->hw.mac.serdes_has_link;
4074 break;
4075 default:
4076 case e1000_media_type_unknown:
4077 break;
4080 if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4081 (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4082 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4083 e_info("Gigabit has been disabled, downgrading speed\n");
4086 return link_active;
4089 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4091 /* make sure the receive unit is started */
4092 if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4093 (adapter->flags & FLAG_RX_RESTART_NOW)) {
4094 struct e1000_hw *hw = &adapter->hw;
4095 u32 rctl = er32(RCTL);
4096 ew32(RCTL, rctl | E1000_RCTL_EN);
4097 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4101 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4103 struct e1000_hw *hw = &adapter->hw;
4106 * With 82574 controllers, PHY needs to be checked periodically
4107 * for hung state and reset, if two calls return true
4109 if (e1000_check_phy_82574(hw))
4110 adapter->phy_hang_count++;
4111 else
4112 adapter->phy_hang_count = 0;
4114 if (adapter->phy_hang_count > 1) {
4115 adapter->phy_hang_count = 0;
4116 schedule_work(&adapter->reset_task);
4121 * e1000_watchdog - Timer Call-back
4122 * @data: pointer to adapter cast into an unsigned long
4124 static void e1000_watchdog(unsigned long data)
4126 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4128 /* Do the rest outside of interrupt context */
4129 schedule_work(&adapter->watchdog_task);
4131 /* TODO: make this use queue_delayed_work() */
4134 static void e1000_watchdog_task(struct work_struct *work)
4136 struct e1000_adapter *adapter = container_of(work,
4137 struct e1000_adapter, watchdog_task);
4138 struct net_device *netdev = adapter->netdev;
4139 struct e1000_mac_info *mac = &adapter->hw.mac;
4140 struct e1000_phy_info *phy = &adapter->hw.phy;
4141 struct e1000_ring *tx_ring = adapter->tx_ring;
4142 struct e1000_hw *hw = &adapter->hw;
4143 u32 link, tctl;
4144 int tx_pending = 0;
4146 link = e1000e_has_link(adapter);
4147 if ((netif_carrier_ok(netdev)) && link) {
4148 /* Cancel scheduled suspend requests. */
4149 pm_runtime_resume(netdev->dev.parent);
4151 e1000e_enable_receives(adapter);
4152 goto link_up;
4155 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4156 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4157 e1000_update_mng_vlan(adapter);
4159 if (link) {
4160 if (!netif_carrier_ok(netdev)) {
4161 bool txb2b = 1;
4163 /* Cancel scheduled suspend requests. */
4164 pm_runtime_resume(netdev->dev.parent);
4166 /* update snapshot of PHY registers on LSC */
4167 e1000_phy_read_status(adapter);
4168 mac->ops.get_link_up_info(&adapter->hw,
4169 &adapter->link_speed,
4170 &adapter->link_duplex);
4171 e1000_print_link_info(adapter);
4173 * On supported PHYs, check for duplex mismatch only
4174 * if link has autonegotiated at 10/100 half
4176 if ((hw->phy.type == e1000_phy_igp_3 ||
4177 hw->phy.type == e1000_phy_bm) &&
4178 (hw->mac.autoneg == true) &&
4179 (adapter->link_speed == SPEED_10 ||
4180 adapter->link_speed == SPEED_100) &&
4181 (adapter->link_duplex == HALF_DUPLEX)) {
4182 u16 autoneg_exp;
4184 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4186 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4187 e_info("Autonegotiated half duplex but"
4188 " link partner cannot autoneg. "
4189 " Try forcing full duplex if "
4190 "link gets many collisions.\n");
4193 /* adjust timeout factor according to speed/duplex */
4194 adapter->tx_timeout_factor = 1;
4195 switch (adapter->link_speed) {
4196 case SPEED_10:
4197 txb2b = 0;
4198 adapter->tx_timeout_factor = 16;
4199 break;
4200 case SPEED_100:
4201 txb2b = 0;
4202 adapter->tx_timeout_factor = 10;
4203 break;
4207 * workaround: re-program speed mode bit after
4208 * link-up event
4210 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4211 !txb2b) {
4212 u32 tarc0;
4213 tarc0 = er32(TARC(0));
4214 tarc0 &= ~SPEED_MODE_BIT;
4215 ew32(TARC(0), tarc0);
4219 * disable TSO for pcie and 10/100 speeds, to avoid
4220 * some hardware issues
4222 if (!(adapter->flags & FLAG_TSO_FORCE)) {
4223 switch (adapter->link_speed) {
4224 case SPEED_10:
4225 case SPEED_100:
4226 e_info("10/100 speed: disabling TSO\n");
4227 netdev->features &= ~NETIF_F_TSO;
4228 netdev->features &= ~NETIF_F_TSO6;
4229 break;
4230 case SPEED_1000:
4231 netdev->features |= NETIF_F_TSO;
4232 netdev->features |= NETIF_F_TSO6;
4233 break;
4234 default:
4235 /* oops */
4236 break;
4241 * enable transmits in the hardware, need to do this
4242 * after setting TARC(0)
4244 tctl = er32(TCTL);
4245 tctl |= E1000_TCTL_EN;
4246 ew32(TCTL, tctl);
4249 * Perform any post-link-up configuration before
4250 * reporting link up.
4252 if (phy->ops.cfg_on_link_up)
4253 phy->ops.cfg_on_link_up(hw);
4255 netif_carrier_on(netdev);
4257 if (!test_bit(__E1000_DOWN, &adapter->state))
4258 mod_timer(&adapter->phy_info_timer,
4259 round_jiffies(jiffies + 2 * HZ));
4261 } else {
4262 if (netif_carrier_ok(netdev)) {
4263 adapter->link_speed = 0;
4264 adapter->link_duplex = 0;
4265 /* Link status message must follow this format */
4266 printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4267 adapter->netdev->name);
4268 netif_carrier_off(netdev);
4269 if (!test_bit(__E1000_DOWN, &adapter->state))
4270 mod_timer(&adapter->phy_info_timer,
4271 round_jiffies(jiffies + 2 * HZ));
4273 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4274 schedule_work(&adapter->reset_task);
4275 else
4276 pm_schedule_suspend(netdev->dev.parent,
4277 LINK_TIMEOUT);
4281 link_up:
4282 e1000e_update_stats(adapter);
4284 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4285 adapter->tpt_old = adapter->stats.tpt;
4286 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4287 adapter->colc_old = adapter->stats.colc;
4289 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4290 adapter->gorc_old = adapter->stats.gorc;
4291 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4292 adapter->gotc_old = adapter->stats.gotc;
4294 e1000e_update_adaptive(&adapter->hw);
4296 if (!netif_carrier_ok(netdev)) {
4297 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
4298 tx_ring->count);
4299 if (tx_pending) {
4301 * We've lost link, so the controller stops DMA,
4302 * but we've got queued Tx work that's never going
4303 * to get done, so reset controller to flush Tx.
4304 * (Do the reset outside of interrupt context).
4306 adapter->tx_timeout_count++;
4307 schedule_work(&adapter->reset_task);
4308 /* return immediately since reset is imminent */
4309 return;
4313 /* Simple mode for Interrupt Throttle Rate (ITR) */
4314 if (adapter->itr_setting == 4) {
4316 * Symmetric Tx/Rx gets a reduced ITR=2000;
4317 * Total asymmetrical Tx or Rx gets ITR=8000;
4318 * everyone else is between 2000-8000.
4320 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4321 u32 dif = (adapter->gotc > adapter->gorc ?
4322 adapter->gotc - adapter->gorc :
4323 adapter->gorc - adapter->gotc) / 10000;
4324 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4326 ew32(ITR, 1000000000 / (itr * 256));
4329 /* Cause software interrupt to ensure Rx ring is cleaned */
4330 if (adapter->msix_entries)
4331 ew32(ICS, adapter->rx_ring->ims_val);
4332 else
4333 ew32(ICS, E1000_ICS_RXDMT0);
4335 /* Force detection of hung controller every watchdog period */
4336 adapter->detect_tx_hung = 1;
4338 /* flush partial descriptors to memory before detecting tx hang */
4339 if (adapter->flags2 & FLAG2_DMA_BURST) {
4340 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4341 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4343 * no need to flush the writes because the timeout code does
4344 * an er32 first thing
4349 * With 82571 controllers, LAA may be overwritten due to controller
4350 * reset from the other port. Set the appropriate LAA in RAR[0]
4352 if (e1000e_get_laa_state_82571(hw))
4353 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4355 if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4356 e1000e_check_82574_phy_workaround(adapter);
4358 /* Reset the timer */
4359 if (!test_bit(__E1000_DOWN, &adapter->state))
4360 mod_timer(&adapter->watchdog_timer,
4361 round_jiffies(jiffies + 2 * HZ));
4364 #define E1000_TX_FLAGS_CSUM 0x00000001
4365 #define E1000_TX_FLAGS_VLAN 0x00000002
4366 #define E1000_TX_FLAGS_TSO 0x00000004
4367 #define E1000_TX_FLAGS_IPV4 0x00000008
4368 #define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
4369 #define E1000_TX_FLAGS_VLAN_SHIFT 16
4371 static int e1000_tso(struct e1000_adapter *adapter,
4372 struct sk_buff *skb)
4374 struct e1000_ring *tx_ring = adapter->tx_ring;
4375 struct e1000_context_desc *context_desc;
4376 struct e1000_buffer *buffer_info;
4377 unsigned int i;
4378 u32 cmd_length = 0;
4379 u16 ipcse = 0, tucse, mss;
4380 u8 ipcss, ipcso, tucss, tucso, hdr_len;
4381 int err;
4383 if (!skb_is_gso(skb))
4384 return 0;
4386 if (skb_header_cloned(skb)) {
4387 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4388 if (err)
4389 return err;
4392 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4393 mss = skb_shinfo(skb)->gso_size;
4394 if (skb->protocol == htons(ETH_P_IP)) {
4395 struct iphdr *iph = ip_hdr(skb);
4396 iph->tot_len = 0;
4397 iph->check = 0;
4398 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4399 0, IPPROTO_TCP, 0);
4400 cmd_length = E1000_TXD_CMD_IP;
4401 ipcse = skb_transport_offset(skb) - 1;
4402 } else if (skb_is_gso_v6(skb)) {
4403 ipv6_hdr(skb)->payload_len = 0;
4404 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4405 &ipv6_hdr(skb)->daddr,
4406 0, IPPROTO_TCP, 0);
4407 ipcse = 0;
4409 ipcss = skb_network_offset(skb);
4410 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4411 tucss = skb_transport_offset(skb);
4412 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4413 tucse = 0;
4415 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4416 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4418 i = tx_ring->next_to_use;
4419 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4420 buffer_info = &tx_ring->buffer_info[i];
4422 context_desc->lower_setup.ip_fields.ipcss = ipcss;
4423 context_desc->lower_setup.ip_fields.ipcso = ipcso;
4424 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
4425 context_desc->upper_setup.tcp_fields.tucss = tucss;
4426 context_desc->upper_setup.tcp_fields.tucso = tucso;
4427 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4428 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
4429 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4430 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4432 buffer_info->time_stamp = jiffies;
4433 buffer_info->next_to_watch = i;
4435 i++;
4436 if (i == tx_ring->count)
4437 i = 0;
4438 tx_ring->next_to_use = i;
4440 return 1;
4443 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4445 struct e1000_ring *tx_ring = adapter->tx_ring;
4446 struct e1000_context_desc *context_desc;
4447 struct e1000_buffer *buffer_info;
4448 unsigned int i;
4449 u8 css;
4450 u32 cmd_len = E1000_TXD_CMD_DEXT;
4451 __be16 protocol;
4453 if (skb->ip_summed != CHECKSUM_PARTIAL)
4454 return 0;
4456 if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4457 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4458 else
4459 protocol = skb->protocol;
4461 switch (protocol) {
4462 case cpu_to_be16(ETH_P_IP):
4463 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4464 cmd_len |= E1000_TXD_CMD_TCP;
4465 break;
4466 case cpu_to_be16(ETH_P_IPV6):
4467 /* XXX not handling all IPV6 headers */
4468 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4469 cmd_len |= E1000_TXD_CMD_TCP;
4470 break;
4471 default:
4472 if (unlikely(net_ratelimit()))
4473 e_warn("checksum_partial proto=%x!\n",
4474 be16_to_cpu(protocol));
4475 break;
4478 css = skb_transport_offset(skb);
4480 i = tx_ring->next_to_use;
4481 buffer_info = &tx_ring->buffer_info[i];
4482 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4484 context_desc->lower_setup.ip_config = 0;
4485 context_desc->upper_setup.tcp_fields.tucss = css;
4486 context_desc->upper_setup.tcp_fields.tucso =
4487 css + skb->csum_offset;
4488 context_desc->upper_setup.tcp_fields.tucse = 0;
4489 context_desc->tcp_seg_setup.data = 0;
4490 context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4492 buffer_info->time_stamp = jiffies;
4493 buffer_info->next_to_watch = i;
4495 i++;
4496 if (i == tx_ring->count)
4497 i = 0;
4498 tx_ring->next_to_use = i;
4500 return 1;
4503 #define E1000_MAX_PER_TXD 8192
4504 #define E1000_MAX_TXD_PWR 12
4506 static int e1000_tx_map(struct e1000_adapter *adapter,
4507 struct sk_buff *skb, unsigned int first,
4508 unsigned int max_per_txd, unsigned int nr_frags,
4509 unsigned int mss)
4511 struct e1000_ring *tx_ring = adapter->tx_ring;
4512 struct pci_dev *pdev = adapter->pdev;
4513 struct e1000_buffer *buffer_info;
4514 unsigned int len = skb_headlen(skb);
4515 unsigned int offset = 0, size, count = 0, i;
4516 unsigned int f, bytecount, segs;
4518 i = tx_ring->next_to_use;
4520 while (len) {
4521 buffer_info = &tx_ring->buffer_info[i];
4522 size = min(len, max_per_txd);
4524 buffer_info->length = size;
4525 buffer_info->time_stamp = jiffies;
4526 buffer_info->next_to_watch = i;
4527 buffer_info->dma = dma_map_single(&pdev->dev,
4528 skb->data + offset,
4529 size, DMA_TO_DEVICE);
4530 buffer_info->mapped_as_page = false;
4531 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4532 goto dma_error;
4534 len -= size;
4535 offset += size;
4536 count++;
4538 if (len) {
4539 i++;
4540 if (i == tx_ring->count)
4541 i = 0;
4545 for (f = 0; f < nr_frags; f++) {
4546 struct skb_frag_struct *frag;
4548 frag = &skb_shinfo(skb)->frags[f];
4549 len = frag->size;
4550 offset = frag->page_offset;
4552 while (len) {
4553 i++;
4554 if (i == tx_ring->count)
4555 i = 0;
4557 buffer_info = &tx_ring->buffer_info[i];
4558 size = min(len, max_per_txd);
4560 buffer_info->length = size;
4561 buffer_info->time_stamp = jiffies;
4562 buffer_info->next_to_watch = i;
4563 buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
4564 offset, size,
4565 DMA_TO_DEVICE);
4566 buffer_info->mapped_as_page = true;
4567 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4568 goto dma_error;
4570 len -= size;
4571 offset += size;
4572 count++;
4576 segs = skb_shinfo(skb)->gso_segs ?: 1;
4577 /* multiply data chunks by size of headers */
4578 bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4580 tx_ring->buffer_info[i].skb = skb;
4581 tx_ring->buffer_info[i].segs = segs;
4582 tx_ring->buffer_info[i].bytecount = bytecount;
4583 tx_ring->buffer_info[first].next_to_watch = i;
4585 return count;
4587 dma_error:
4588 dev_err(&pdev->dev, "TX DMA map failed\n");
4589 buffer_info->dma = 0;
4590 if (count)
4591 count--;
4593 while (count--) {
4594 if (i==0)
4595 i += tx_ring->count;
4596 i--;
4597 buffer_info = &tx_ring->buffer_info[i];
4598 e1000_put_txbuf(adapter, buffer_info);;
4601 return 0;
4604 static void e1000_tx_queue(struct e1000_adapter *adapter,
4605 int tx_flags, int count)
4607 struct e1000_ring *tx_ring = adapter->tx_ring;
4608 struct e1000_tx_desc *tx_desc = NULL;
4609 struct e1000_buffer *buffer_info;
4610 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4611 unsigned int i;
4613 if (tx_flags & E1000_TX_FLAGS_TSO) {
4614 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4615 E1000_TXD_CMD_TSE;
4616 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4618 if (tx_flags & E1000_TX_FLAGS_IPV4)
4619 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4622 if (tx_flags & E1000_TX_FLAGS_CSUM) {
4623 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4624 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4627 if (tx_flags & E1000_TX_FLAGS_VLAN) {
4628 txd_lower |= E1000_TXD_CMD_VLE;
4629 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4632 i = tx_ring->next_to_use;
4634 while (count--) {
4635 buffer_info = &tx_ring->buffer_info[i];
4636 tx_desc = E1000_TX_DESC(*tx_ring, i);
4637 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4638 tx_desc->lower.data =
4639 cpu_to_le32(txd_lower | buffer_info->length);
4640 tx_desc->upper.data = cpu_to_le32(txd_upper);
4642 i++;
4643 if (i == tx_ring->count)
4644 i = 0;
4647 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4650 * Force memory writes to complete before letting h/w
4651 * know there are new descriptors to fetch. (Only
4652 * applicable for weak-ordered memory model archs,
4653 * such as IA-64).
4655 wmb();
4657 tx_ring->next_to_use = i;
4658 writel(i, adapter->hw.hw_addr + tx_ring->tail);
4660 * we need this if more than one processor can write to our tail
4661 * at a time, it synchronizes IO on IA64/Altix systems
4663 mmiowb();
4666 #define MINIMUM_DHCP_PACKET_SIZE 282
4667 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4668 struct sk_buff *skb)
4670 struct e1000_hw *hw = &adapter->hw;
4671 u16 length, offset;
4673 if (vlan_tx_tag_present(skb)) {
4674 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4675 (adapter->hw.mng_cookie.status &
4676 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4677 return 0;
4680 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4681 return 0;
4683 if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4684 return 0;
4687 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4688 struct udphdr *udp;
4690 if (ip->protocol != IPPROTO_UDP)
4691 return 0;
4693 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4694 if (ntohs(udp->dest) != 67)
4695 return 0;
4697 offset = (u8 *)udp + 8 - skb->data;
4698 length = skb->len - offset;
4699 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4702 return 0;
4705 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4707 struct e1000_adapter *adapter = netdev_priv(netdev);
4709 netif_stop_queue(netdev);
4711 * Herbert's original patch had:
4712 * smp_mb__after_netif_stop_queue();
4713 * but since that doesn't exist yet, just open code it.
4715 smp_mb();
4718 * We need to check again in a case another CPU has just
4719 * made room available.
4721 if (e1000_desc_unused(adapter->tx_ring) < size)
4722 return -EBUSY;
4724 /* A reprieve! */
4725 netif_start_queue(netdev);
4726 ++adapter->restart_queue;
4727 return 0;
4730 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4732 struct e1000_adapter *adapter = netdev_priv(netdev);
4734 if (e1000_desc_unused(adapter->tx_ring) >= size)
4735 return 0;
4736 return __e1000_maybe_stop_tx(netdev, size);
4739 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4740 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4741 struct net_device *netdev)
4743 struct e1000_adapter *adapter = netdev_priv(netdev);
4744 struct e1000_ring *tx_ring = adapter->tx_ring;
4745 unsigned int first;
4746 unsigned int max_per_txd = E1000_MAX_PER_TXD;
4747 unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4748 unsigned int tx_flags = 0;
4749 unsigned int len = skb_headlen(skb);
4750 unsigned int nr_frags;
4751 unsigned int mss;
4752 int count = 0;
4753 int tso;
4754 unsigned int f;
4756 if (test_bit(__E1000_DOWN, &adapter->state)) {
4757 dev_kfree_skb_any(skb);
4758 return NETDEV_TX_OK;
4761 if (skb->len <= 0) {
4762 dev_kfree_skb_any(skb);
4763 return NETDEV_TX_OK;
4766 mss = skb_shinfo(skb)->gso_size;
4768 * The controller does a simple calculation to
4769 * make sure there is enough room in the FIFO before
4770 * initiating the DMA for each buffer. The calc is:
4771 * 4 = ceil(buffer len/mss). To make sure we don't
4772 * overrun the FIFO, adjust the max buffer len if mss
4773 * drops.
4775 if (mss) {
4776 u8 hdr_len;
4777 max_per_txd = min(mss << 2, max_per_txd);
4778 max_txd_pwr = fls(max_per_txd) - 1;
4781 * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4782 * points to just header, pull a few bytes of payload from
4783 * frags into skb->data
4785 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4787 * we do this workaround for ES2LAN, but it is un-necessary,
4788 * avoiding it could save a lot of cycles
4790 if (skb->data_len && (hdr_len == len)) {
4791 unsigned int pull_size;
4793 pull_size = min((unsigned int)4, skb->data_len);
4794 if (!__pskb_pull_tail(skb, pull_size)) {
4795 e_err("__pskb_pull_tail failed.\n");
4796 dev_kfree_skb_any(skb);
4797 return NETDEV_TX_OK;
4799 len = skb_headlen(skb);
4803 /* reserve a descriptor for the offload context */
4804 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4805 count++;
4806 count++;
4808 count += TXD_USE_COUNT(len, max_txd_pwr);
4810 nr_frags = skb_shinfo(skb)->nr_frags;
4811 for (f = 0; f < nr_frags; f++)
4812 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4813 max_txd_pwr);
4815 if (adapter->hw.mac.tx_pkt_filtering)
4816 e1000_transfer_dhcp_info(adapter, skb);
4819 * need: count + 2 desc gap to keep tail from touching
4820 * head, otherwise try next time
4822 if (e1000_maybe_stop_tx(netdev, count + 2))
4823 return NETDEV_TX_BUSY;
4825 if (vlan_tx_tag_present(skb)) {
4826 tx_flags |= E1000_TX_FLAGS_VLAN;
4827 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4830 first = tx_ring->next_to_use;
4832 tso = e1000_tso(adapter, skb);
4833 if (tso < 0) {
4834 dev_kfree_skb_any(skb);
4835 return NETDEV_TX_OK;
4838 if (tso)
4839 tx_flags |= E1000_TX_FLAGS_TSO;
4840 else if (e1000_tx_csum(adapter, skb))
4841 tx_flags |= E1000_TX_FLAGS_CSUM;
4844 * Old method was to assume IPv4 packet by default if TSO was enabled.
4845 * 82571 hardware supports TSO capabilities for IPv6 as well...
4846 * no longer assume, we must.
4848 if (skb->protocol == htons(ETH_P_IP))
4849 tx_flags |= E1000_TX_FLAGS_IPV4;
4851 /* if count is 0 then mapping error has occured */
4852 count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4853 if (count) {
4854 e1000_tx_queue(adapter, tx_flags, count);
4855 /* Make sure there is space in the ring for the next send. */
4856 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4858 } else {
4859 dev_kfree_skb_any(skb);
4860 tx_ring->buffer_info[first].time_stamp = 0;
4861 tx_ring->next_to_use = first;
4864 return NETDEV_TX_OK;
4868 * e1000_tx_timeout - Respond to a Tx Hang
4869 * @netdev: network interface device structure
4871 static void e1000_tx_timeout(struct net_device *netdev)
4873 struct e1000_adapter *adapter = netdev_priv(netdev);
4875 /* Do the reset outside of interrupt context */
4876 adapter->tx_timeout_count++;
4877 schedule_work(&adapter->reset_task);
4880 static void e1000_reset_task(struct work_struct *work)
4882 struct e1000_adapter *adapter;
4883 adapter = container_of(work, struct e1000_adapter, reset_task);
4885 if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4886 (adapter->flags & FLAG_RX_RESTART_NOW))) {
4887 e1000e_dump(adapter);
4888 e_err("Reset adapter\n");
4890 e1000e_reinit_locked(adapter);
4894 * e1000_get_stats - Get System Network Statistics
4895 * @netdev: network interface device structure
4897 * Returns the address of the device statistics structure.
4898 * The statistics are actually updated from the timer callback.
4900 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4902 /* only return the current stats */
4903 return &netdev->stats;
4907 * e1000_change_mtu - Change the Maximum Transfer Unit
4908 * @netdev: network interface device structure
4909 * @new_mtu: new value for maximum frame size
4911 * Returns 0 on success, negative on failure
4913 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4915 struct e1000_adapter *adapter = netdev_priv(netdev);
4916 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4918 /* Jumbo frame support */
4919 if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4920 !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4921 e_err("Jumbo Frames not supported.\n");
4922 return -EINVAL;
4925 /* Supported frame sizes */
4926 if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4927 (max_frame > adapter->max_hw_frame_size)) {
4928 e_err("Unsupported MTU setting\n");
4929 return -EINVAL;
4932 /* Jumbo frame workaround on 82579 requires CRC be stripped */
4933 if ((adapter->hw.mac.type == e1000_pch2lan) &&
4934 !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
4935 (new_mtu > ETH_DATA_LEN)) {
4936 e_err("Jumbo Frames not supported on 82579 when CRC "
4937 "stripping is disabled.\n");
4938 return -EINVAL;
4941 /* 82573 Errata 17 */
4942 if (((adapter->hw.mac.type == e1000_82573) ||
4943 (adapter->hw.mac.type == e1000_82574)) &&
4944 (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
4945 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
4946 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
4949 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4950 msleep(1);
4951 /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
4952 adapter->max_frame_size = max_frame;
4953 e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4954 netdev->mtu = new_mtu;
4955 if (netif_running(netdev))
4956 e1000e_down(adapter);
4959 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4960 * means we reserve 2 more, this pushes us to allocate from the next
4961 * larger slab size.
4962 * i.e. RXBUFFER_2048 --> size-4096 slab
4963 * However with the new *_jumbo_rx* routines, jumbo receives will use
4964 * fragmented skbs
4967 if (max_frame <= 2048)
4968 adapter->rx_buffer_len = 2048;
4969 else
4970 adapter->rx_buffer_len = 4096;
4972 /* adjust allocation if LPE protects us, and we aren't using SBP */
4973 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4974 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4975 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4976 + ETH_FCS_LEN;
4978 if (netif_running(netdev))
4979 e1000e_up(adapter);
4980 else
4981 e1000e_reset(adapter);
4983 clear_bit(__E1000_RESETTING, &adapter->state);
4985 return 0;
4988 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4989 int cmd)
4991 struct e1000_adapter *adapter = netdev_priv(netdev);
4992 struct mii_ioctl_data *data = if_mii(ifr);
4994 if (adapter->hw.phy.media_type != e1000_media_type_copper)
4995 return -EOPNOTSUPP;
4997 switch (cmd) {
4998 case SIOCGMIIPHY:
4999 data->phy_id = adapter->hw.phy.addr;
5000 break;
5001 case SIOCGMIIREG:
5002 e1000_phy_read_status(adapter);
5004 switch (data->reg_num & 0x1F) {
5005 case MII_BMCR:
5006 data->val_out = adapter->phy_regs.bmcr;
5007 break;
5008 case MII_BMSR:
5009 data->val_out = adapter->phy_regs.bmsr;
5010 break;
5011 case MII_PHYSID1:
5012 data->val_out = (adapter->hw.phy.id >> 16);
5013 break;
5014 case MII_PHYSID2:
5015 data->val_out = (adapter->hw.phy.id & 0xFFFF);
5016 break;
5017 case MII_ADVERTISE:
5018 data->val_out = adapter->phy_regs.advertise;
5019 break;
5020 case MII_LPA:
5021 data->val_out = adapter->phy_regs.lpa;
5022 break;
5023 case MII_EXPANSION:
5024 data->val_out = adapter->phy_regs.expansion;
5025 break;
5026 case MII_CTRL1000:
5027 data->val_out = adapter->phy_regs.ctrl1000;
5028 break;
5029 case MII_STAT1000:
5030 data->val_out = adapter->phy_regs.stat1000;
5031 break;
5032 case MII_ESTATUS:
5033 data->val_out = adapter->phy_regs.estatus;
5034 break;
5035 default:
5036 return -EIO;
5038 break;
5039 case SIOCSMIIREG:
5040 default:
5041 return -EOPNOTSUPP;
5043 return 0;
5046 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5048 switch (cmd) {
5049 case SIOCGMIIPHY:
5050 case SIOCGMIIREG:
5051 case SIOCSMIIREG:
5052 return e1000_mii_ioctl(netdev, ifr, cmd);
5053 default:
5054 return -EOPNOTSUPP;
5058 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5060 struct e1000_hw *hw = &adapter->hw;
5061 u32 i, mac_reg;
5062 u16 phy_reg;
5063 int retval = 0;
5065 /* copy MAC RARs to PHY RARs */
5066 e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5068 /* copy MAC MTA to PHY MTA */
5069 for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5070 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5071 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
5072 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
5075 /* configure PHY Rx Control register */
5076 e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
5077 mac_reg = er32(RCTL);
5078 if (mac_reg & E1000_RCTL_UPE)
5079 phy_reg |= BM_RCTL_UPE;
5080 if (mac_reg & E1000_RCTL_MPE)
5081 phy_reg |= BM_RCTL_MPE;
5082 phy_reg &= ~(BM_RCTL_MO_MASK);
5083 if (mac_reg & E1000_RCTL_MO_3)
5084 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5085 << BM_RCTL_MO_SHIFT);
5086 if (mac_reg & E1000_RCTL_BAM)
5087 phy_reg |= BM_RCTL_BAM;
5088 if (mac_reg & E1000_RCTL_PMCF)
5089 phy_reg |= BM_RCTL_PMCF;
5090 mac_reg = er32(CTRL);
5091 if (mac_reg & E1000_CTRL_RFCE)
5092 phy_reg |= BM_RCTL_RFCE;
5093 e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
5095 /* enable PHY wakeup in MAC register */
5096 ew32(WUFC, wufc);
5097 ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5099 /* configure and enable PHY wakeup in PHY registers */
5100 e1e_wphy(&adapter->hw, BM_WUFC, wufc);
5101 e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5103 /* activate PHY wakeup */
5104 retval = hw->phy.ops.acquire(hw);
5105 if (retval) {
5106 e_err("Could not acquire PHY\n");
5107 return retval;
5109 e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
5110 (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
5111 retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
5112 if (retval) {
5113 e_err("Could not read PHY page 769\n");
5114 goto out;
5116 phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5117 retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
5118 if (retval)
5119 e_err("Could not set PHY Host Wakeup bit\n");
5120 out:
5121 hw->phy.ops.release(hw);
5123 return retval;
5126 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5127 bool runtime)
5129 struct net_device *netdev = pci_get_drvdata(pdev);
5130 struct e1000_adapter *adapter = netdev_priv(netdev);
5131 struct e1000_hw *hw = &adapter->hw;
5132 u32 ctrl, ctrl_ext, rctl, status;
5133 /* Runtime suspend should only enable wakeup for link changes */
5134 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5135 int retval = 0;
5137 netif_device_detach(netdev);
5139 if (netif_running(netdev)) {
5140 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5141 e1000e_down(adapter);
5142 e1000_free_irq(adapter);
5144 e1000e_reset_interrupt_capability(adapter);
5146 retval = pci_save_state(pdev);
5147 if (retval)
5148 return retval;
5150 status = er32(STATUS);
5151 if (status & E1000_STATUS_LU)
5152 wufc &= ~E1000_WUFC_LNKC;
5154 if (wufc) {
5155 e1000_setup_rctl(adapter);
5156 e1000_set_multi(netdev);
5158 /* turn on all-multi mode if wake on multicast is enabled */
5159 if (wufc & E1000_WUFC_MC) {
5160 rctl = er32(RCTL);
5161 rctl |= E1000_RCTL_MPE;
5162 ew32(RCTL, rctl);
5165 ctrl = er32(CTRL);
5166 /* advertise wake from D3Cold */
5167 #define E1000_CTRL_ADVD3WUC 0x00100000
5168 /* phy power management enable */
5169 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5170 ctrl |= E1000_CTRL_ADVD3WUC;
5171 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5172 ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5173 ew32(CTRL, ctrl);
5175 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5176 adapter->hw.phy.media_type ==
5177 e1000_media_type_internal_serdes) {
5178 /* keep the laser running in D3 */
5179 ctrl_ext = er32(CTRL_EXT);
5180 ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5181 ew32(CTRL_EXT, ctrl_ext);
5184 if (adapter->flags & FLAG_IS_ICH)
5185 e1000e_disable_gig_wol_ich8lan(&adapter->hw);
5187 /* Allow time for pending master requests to run */
5188 e1000e_disable_pcie_master(&adapter->hw);
5190 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5191 /* enable wakeup by the PHY */
5192 retval = e1000_init_phy_wakeup(adapter, wufc);
5193 if (retval)
5194 return retval;
5195 } else {
5196 /* enable wakeup by the MAC */
5197 ew32(WUFC, wufc);
5198 ew32(WUC, E1000_WUC_PME_EN);
5200 } else {
5201 ew32(WUC, 0);
5202 ew32(WUFC, 0);
5205 *enable_wake = !!wufc;
5207 /* make sure adapter isn't asleep if manageability is enabled */
5208 if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5209 (hw->mac.ops.check_mng_mode(hw)))
5210 *enable_wake = true;
5212 if (adapter->hw.phy.type == e1000_phy_igp_3)
5213 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5216 * Release control of h/w to f/w. If f/w is AMT enabled, this
5217 * would have already happened in close and is redundant.
5219 e1000_release_hw_control(adapter);
5221 pci_disable_device(pdev);
5223 return 0;
5226 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5228 if (sleep && wake) {
5229 pci_prepare_to_sleep(pdev);
5230 return;
5233 pci_wake_from_d3(pdev, wake);
5234 pci_set_power_state(pdev, PCI_D3hot);
5237 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5238 bool wake)
5240 struct net_device *netdev = pci_get_drvdata(pdev);
5241 struct e1000_adapter *adapter = netdev_priv(netdev);
5244 * The pci-e switch on some quad port adapters will report a
5245 * correctable error when the MAC transitions from D0 to D3. To
5246 * prevent this we need to mask off the correctable errors on the
5247 * downstream port of the pci-e switch.
5249 if (adapter->flags & FLAG_IS_QUAD_PORT) {
5250 struct pci_dev *us_dev = pdev->bus->self;
5251 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
5252 u16 devctl;
5254 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5255 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5256 (devctl & ~PCI_EXP_DEVCTL_CERE));
5258 e1000_power_off(pdev, sleep, wake);
5260 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5261 } else {
5262 e1000_power_off(pdev, sleep, wake);
5266 #ifdef CONFIG_PCIEASPM
5267 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5269 pci_disable_link_state(pdev, state);
5271 #else
5272 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5274 int pos;
5275 u16 reg16;
5278 * Both device and parent should have the same ASPM setting.
5279 * Disable ASPM in downstream component first and then upstream.
5281 pos = pci_pcie_cap(pdev);
5282 pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5283 reg16 &= ~state;
5284 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5286 if (!pdev->bus->self)
5287 return;
5289 pos = pci_pcie_cap(pdev->bus->self);
5290 pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5291 reg16 &= ~state;
5292 pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5294 #endif
5295 void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5297 dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5298 (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5299 (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5301 __e1000e_disable_aspm(pdev, state);
5304 #ifdef CONFIG_PM_OPS
5305 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5307 return !!adapter->tx_ring->buffer_info;
5310 static int __e1000_resume(struct pci_dev *pdev)
5312 struct net_device *netdev = pci_get_drvdata(pdev);
5313 struct e1000_adapter *adapter = netdev_priv(netdev);
5314 struct e1000_hw *hw = &adapter->hw;
5315 u32 err;
5317 pci_set_power_state(pdev, PCI_D0);
5318 pci_restore_state(pdev);
5319 pci_save_state(pdev);
5320 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5321 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5323 e1000e_set_interrupt_capability(adapter);
5324 if (netif_running(netdev)) {
5325 err = e1000_request_irq(adapter);
5326 if (err)
5327 return err;
5330 e1000e_power_up_phy(adapter);
5332 /* report the system wakeup cause from S3/S4 */
5333 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5334 u16 phy_data;
5336 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5337 if (phy_data) {
5338 e_info("PHY Wakeup cause - %s\n",
5339 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5340 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5341 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5342 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5343 phy_data & E1000_WUS_LNKC ? "Link Status "
5344 " Change" : "other");
5346 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5347 } else {
5348 u32 wus = er32(WUS);
5349 if (wus) {
5350 e_info("MAC Wakeup cause - %s\n",
5351 wus & E1000_WUS_EX ? "Unicast Packet" :
5352 wus & E1000_WUS_MC ? "Multicast Packet" :
5353 wus & E1000_WUS_BC ? "Broadcast Packet" :
5354 wus & E1000_WUS_MAG ? "Magic Packet" :
5355 wus & E1000_WUS_LNKC ? "Link Status Change" :
5356 "other");
5358 ew32(WUS, ~0);
5361 e1000e_reset(adapter);
5363 e1000_init_manageability_pt(adapter);
5365 if (netif_running(netdev))
5366 e1000e_up(adapter);
5368 netif_device_attach(netdev);
5371 * If the controller has AMT, do not set DRV_LOAD until the interface
5372 * is up. For all other cases, let the f/w know that the h/w is now
5373 * under the control of the driver.
5375 if (!(adapter->flags & FLAG_HAS_AMT))
5376 e1000_get_hw_control(adapter);
5378 return 0;
5381 #ifdef CONFIG_PM_SLEEP
5382 static int e1000_suspend(struct device *dev)
5384 struct pci_dev *pdev = to_pci_dev(dev);
5385 int retval;
5386 bool wake;
5388 retval = __e1000_shutdown(pdev, &wake, false);
5389 if (!retval)
5390 e1000_complete_shutdown(pdev, true, wake);
5392 return retval;
5395 static int e1000_resume(struct device *dev)
5397 struct pci_dev *pdev = to_pci_dev(dev);
5398 struct net_device *netdev = pci_get_drvdata(pdev);
5399 struct e1000_adapter *adapter = netdev_priv(netdev);
5401 if (e1000e_pm_ready(adapter))
5402 adapter->idle_check = true;
5404 return __e1000_resume(pdev);
5406 #endif /* CONFIG_PM_SLEEP */
5408 #ifdef CONFIG_PM_RUNTIME
5409 static int e1000_runtime_suspend(struct device *dev)
5411 struct pci_dev *pdev = to_pci_dev(dev);
5412 struct net_device *netdev = pci_get_drvdata(pdev);
5413 struct e1000_adapter *adapter = netdev_priv(netdev);
5415 if (e1000e_pm_ready(adapter)) {
5416 bool wake;
5418 __e1000_shutdown(pdev, &wake, true);
5421 return 0;
5424 static int e1000_idle(struct device *dev)
5426 struct pci_dev *pdev = to_pci_dev(dev);
5427 struct net_device *netdev = pci_get_drvdata(pdev);
5428 struct e1000_adapter *adapter = netdev_priv(netdev);
5430 if (!e1000e_pm_ready(adapter))
5431 return 0;
5433 if (adapter->idle_check) {
5434 adapter->idle_check = false;
5435 if (!e1000e_has_link(adapter))
5436 pm_schedule_suspend(dev, MSEC_PER_SEC);
5439 return -EBUSY;
5442 static int e1000_runtime_resume(struct device *dev)
5444 struct pci_dev *pdev = to_pci_dev(dev);
5445 struct net_device *netdev = pci_get_drvdata(pdev);
5446 struct e1000_adapter *adapter = netdev_priv(netdev);
5448 if (!e1000e_pm_ready(adapter))
5449 return 0;
5451 adapter->idle_check = !dev->power.runtime_auto;
5452 return __e1000_resume(pdev);
5454 #endif /* CONFIG_PM_RUNTIME */
5455 #endif /* CONFIG_PM_OPS */
5457 static void e1000_shutdown(struct pci_dev *pdev)
5459 bool wake = false;
5461 __e1000_shutdown(pdev, &wake, false);
5463 if (system_state == SYSTEM_POWER_OFF)
5464 e1000_complete_shutdown(pdev, false, wake);
5467 #ifdef CONFIG_NET_POLL_CONTROLLER
5469 * Polling 'interrupt' - used by things like netconsole to send skbs
5470 * without having to re-enable interrupts. It's not called while
5471 * the interrupt routine is executing.
5473 static void e1000_netpoll(struct net_device *netdev)
5475 struct e1000_adapter *adapter = netdev_priv(netdev);
5477 disable_irq(adapter->pdev->irq);
5478 e1000_intr(adapter->pdev->irq, netdev);
5480 enable_irq(adapter->pdev->irq);
5482 #endif
5485 * e1000_io_error_detected - called when PCI error is detected
5486 * @pdev: Pointer to PCI device
5487 * @state: The current pci connection state
5489 * This function is called after a PCI bus error affecting
5490 * this device has been detected.
5492 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5493 pci_channel_state_t state)
5495 struct net_device *netdev = pci_get_drvdata(pdev);
5496 struct e1000_adapter *adapter = netdev_priv(netdev);
5498 netif_device_detach(netdev);
5500 if (state == pci_channel_io_perm_failure)
5501 return PCI_ERS_RESULT_DISCONNECT;
5503 if (netif_running(netdev))
5504 e1000e_down(adapter);
5505 pci_disable_device(pdev);
5507 /* Request a slot slot reset. */
5508 return PCI_ERS_RESULT_NEED_RESET;
5512 * e1000_io_slot_reset - called after the pci bus has been reset.
5513 * @pdev: Pointer to PCI device
5515 * Restart the card from scratch, as if from a cold-boot. Implementation
5516 * resembles the first-half of the e1000_resume routine.
5518 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5520 struct net_device *netdev = pci_get_drvdata(pdev);
5521 struct e1000_adapter *adapter = netdev_priv(netdev);
5522 struct e1000_hw *hw = &adapter->hw;
5523 int err;
5524 pci_ers_result_t result;
5526 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5527 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5528 err = pci_enable_device_mem(pdev);
5529 if (err) {
5530 dev_err(&pdev->dev,
5531 "Cannot re-enable PCI device after reset.\n");
5532 result = PCI_ERS_RESULT_DISCONNECT;
5533 } else {
5534 pci_set_master(pdev);
5535 pdev->state_saved = true;
5536 pci_restore_state(pdev);
5538 pci_enable_wake(pdev, PCI_D3hot, 0);
5539 pci_enable_wake(pdev, PCI_D3cold, 0);
5541 e1000e_reset(adapter);
5542 ew32(WUS, ~0);
5543 result = PCI_ERS_RESULT_RECOVERED;
5546 pci_cleanup_aer_uncorrect_error_status(pdev);
5548 return result;
5552 * e1000_io_resume - called when traffic can start flowing again.
5553 * @pdev: Pointer to PCI device
5555 * This callback is called when the error recovery driver tells us that
5556 * its OK to resume normal operation. Implementation resembles the
5557 * second-half of the e1000_resume routine.
5559 static void e1000_io_resume(struct pci_dev *pdev)
5561 struct net_device *netdev = pci_get_drvdata(pdev);
5562 struct e1000_adapter *adapter = netdev_priv(netdev);
5564 e1000_init_manageability_pt(adapter);
5566 if (netif_running(netdev)) {
5567 if (e1000e_up(adapter)) {
5568 dev_err(&pdev->dev,
5569 "can't bring device back up after reset\n");
5570 return;
5574 netif_device_attach(netdev);
5577 * If the controller has AMT, do not set DRV_LOAD until the interface
5578 * is up. For all other cases, let the f/w know that the h/w is now
5579 * under the control of the driver.
5581 if (!(adapter->flags & FLAG_HAS_AMT))
5582 e1000_get_hw_control(adapter);
5586 static void e1000_print_device_info(struct e1000_adapter *adapter)
5588 struct e1000_hw *hw = &adapter->hw;
5589 struct net_device *netdev = adapter->netdev;
5590 u32 pba_num;
5592 /* print bus type/speed/width info */
5593 e_info("(PCI Express:2.5GB/s:%s) %pM\n",
5594 /* bus width */
5595 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5596 "Width x1"),
5597 /* MAC address */
5598 netdev->dev_addr);
5599 e_info("Intel(R) PRO/%s Network Connection\n",
5600 (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5601 e1000e_read_pba_num(hw, &pba_num);
5602 e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
5603 hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
5606 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5608 struct e1000_hw *hw = &adapter->hw;
5609 int ret_val;
5610 u16 buf = 0;
5612 if (hw->mac.type != e1000_82573)
5613 return;
5615 ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5616 if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5617 /* Deep Smart Power Down (DSPD) */
5618 dev_warn(&adapter->pdev->dev,
5619 "Warning: detected DSPD enabled in EEPROM\n");
5623 static const struct net_device_ops e1000e_netdev_ops = {
5624 .ndo_open = e1000_open,
5625 .ndo_stop = e1000_close,
5626 .ndo_start_xmit = e1000_xmit_frame,
5627 .ndo_get_stats = e1000_get_stats,
5628 .ndo_set_multicast_list = e1000_set_multi,
5629 .ndo_set_mac_address = e1000_set_mac,
5630 .ndo_change_mtu = e1000_change_mtu,
5631 .ndo_do_ioctl = e1000_ioctl,
5632 .ndo_tx_timeout = e1000_tx_timeout,
5633 .ndo_validate_addr = eth_validate_addr,
5635 .ndo_vlan_rx_register = e1000_vlan_rx_register,
5636 .ndo_vlan_rx_add_vid = e1000_vlan_rx_add_vid,
5637 .ndo_vlan_rx_kill_vid = e1000_vlan_rx_kill_vid,
5638 #ifdef CONFIG_NET_POLL_CONTROLLER
5639 .ndo_poll_controller = e1000_netpoll,
5640 #endif
5644 * e1000_probe - Device Initialization Routine
5645 * @pdev: PCI device information struct
5646 * @ent: entry in e1000_pci_tbl
5648 * Returns 0 on success, negative on failure
5650 * e1000_probe initializes an adapter identified by a pci_dev structure.
5651 * The OS initialization, configuring of the adapter private structure,
5652 * and a hardware reset occur.
5654 static int __devinit e1000_probe(struct pci_dev *pdev,
5655 const struct pci_device_id *ent)
5657 struct net_device *netdev;
5658 struct e1000_adapter *adapter;
5659 struct e1000_hw *hw;
5660 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5661 resource_size_t mmio_start, mmio_len;
5662 resource_size_t flash_start, flash_len;
5664 static int cards_found;
5665 int i, err, pci_using_dac;
5666 u16 eeprom_data = 0;
5667 u16 eeprom_apme_mask = E1000_EEPROM_APME;
5669 if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
5670 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5672 err = pci_enable_device_mem(pdev);
5673 if (err)
5674 return err;
5676 pci_using_dac = 0;
5677 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
5678 if (!err) {
5679 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
5680 if (!err)
5681 pci_using_dac = 1;
5682 } else {
5683 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
5684 if (err) {
5685 err = dma_set_coherent_mask(&pdev->dev,
5686 DMA_BIT_MASK(32));
5687 if (err) {
5688 dev_err(&pdev->dev, "No usable DMA "
5689 "configuration, aborting\n");
5690 goto err_dma;
5695 err = pci_request_selected_regions_exclusive(pdev,
5696 pci_select_bars(pdev, IORESOURCE_MEM),
5697 e1000e_driver_name);
5698 if (err)
5699 goto err_pci_reg;
5701 /* AER (Advanced Error Reporting) hooks */
5702 pci_enable_pcie_error_reporting(pdev);
5704 pci_set_master(pdev);
5705 /* PCI config space info */
5706 err = pci_save_state(pdev);
5707 if (err)
5708 goto err_alloc_etherdev;
5710 err = -ENOMEM;
5711 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5712 if (!netdev)
5713 goto err_alloc_etherdev;
5715 SET_NETDEV_DEV(netdev, &pdev->dev);
5717 netdev->irq = pdev->irq;
5719 pci_set_drvdata(pdev, netdev);
5720 adapter = netdev_priv(netdev);
5721 hw = &adapter->hw;
5722 adapter->netdev = netdev;
5723 adapter->pdev = pdev;
5724 adapter->ei = ei;
5725 adapter->pba = ei->pba;
5726 adapter->flags = ei->flags;
5727 adapter->flags2 = ei->flags2;
5728 adapter->hw.adapter = adapter;
5729 adapter->hw.mac.type = ei->mac;
5730 adapter->max_hw_frame_size = ei->max_hw_frame_size;
5731 adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5733 mmio_start = pci_resource_start(pdev, 0);
5734 mmio_len = pci_resource_len(pdev, 0);
5736 err = -EIO;
5737 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5738 if (!adapter->hw.hw_addr)
5739 goto err_ioremap;
5741 if ((adapter->flags & FLAG_HAS_FLASH) &&
5742 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5743 flash_start = pci_resource_start(pdev, 1);
5744 flash_len = pci_resource_len(pdev, 1);
5745 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5746 if (!adapter->hw.flash_address)
5747 goto err_flashmap;
5750 /* construct the net_device struct */
5751 netdev->netdev_ops = &e1000e_netdev_ops;
5752 e1000e_set_ethtool_ops(netdev);
5753 netdev->watchdog_timeo = 5 * HZ;
5754 netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5755 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5757 netdev->mem_start = mmio_start;
5758 netdev->mem_end = mmio_start + mmio_len;
5760 adapter->bd_number = cards_found++;
5762 e1000e_check_options(adapter);
5764 /* setup adapter struct */
5765 err = e1000_sw_init(adapter);
5766 if (err)
5767 goto err_sw_init;
5769 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5770 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5771 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5773 err = ei->get_variants(adapter);
5774 if (err)
5775 goto err_hw_init;
5777 if ((adapter->flags & FLAG_IS_ICH) &&
5778 (adapter->flags & FLAG_READ_ONLY_NVM))
5779 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5781 hw->mac.ops.get_bus_info(&adapter->hw);
5783 adapter->hw.phy.autoneg_wait_to_complete = 0;
5785 /* Copper options */
5786 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5787 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5788 adapter->hw.phy.disable_polarity_correction = 0;
5789 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5792 if (e1000_check_reset_block(&adapter->hw))
5793 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5795 netdev->features = NETIF_F_SG |
5796 NETIF_F_HW_CSUM |
5797 NETIF_F_HW_VLAN_TX |
5798 NETIF_F_HW_VLAN_RX;
5800 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5801 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5803 netdev->features |= NETIF_F_TSO;
5804 netdev->features |= NETIF_F_TSO6;
5806 netdev->vlan_features |= NETIF_F_TSO;
5807 netdev->vlan_features |= NETIF_F_TSO6;
5808 netdev->vlan_features |= NETIF_F_HW_CSUM;
5809 netdev->vlan_features |= NETIF_F_SG;
5811 if (pci_using_dac) {
5812 netdev->features |= NETIF_F_HIGHDMA;
5813 netdev->vlan_features |= NETIF_F_HIGHDMA;
5816 if (e1000e_enable_mng_pass_thru(&adapter->hw))
5817 adapter->flags |= FLAG_MNG_PT_ENABLED;
5820 * before reading the NVM, reset the controller to
5821 * put the device in a known good starting state
5823 adapter->hw.mac.ops.reset_hw(&adapter->hw);
5826 * systems with ASPM and others may see the checksum fail on the first
5827 * attempt. Let's give it a few tries
5829 for (i = 0;; i++) {
5830 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5831 break;
5832 if (i == 2) {
5833 e_err("The NVM Checksum Is Not Valid\n");
5834 err = -EIO;
5835 goto err_eeprom;
5839 e1000_eeprom_checks(adapter);
5841 /* copy the MAC address */
5842 if (e1000e_read_mac_addr(&adapter->hw))
5843 e_err("NVM Read Error while reading MAC address\n");
5845 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5846 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5848 if (!is_valid_ether_addr(netdev->perm_addr)) {
5849 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5850 err = -EIO;
5851 goto err_eeprom;
5854 init_timer(&adapter->watchdog_timer);
5855 adapter->watchdog_timer.function = e1000_watchdog;
5856 adapter->watchdog_timer.data = (unsigned long) adapter;
5858 init_timer(&adapter->phy_info_timer);
5859 adapter->phy_info_timer.function = e1000_update_phy_info;
5860 adapter->phy_info_timer.data = (unsigned long) adapter;
5862 INIT_WORK(&adapter->reset_task, e1000_reset_task);
5863 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5864 INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5865 INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5866 INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5868 /* Initialize link parameters. User can change them with ethtool */
5869 adapter->hw.mac.autoneg = 1;
5870 adapter->fc_autoneg = 1;
5871 adapter->hw.fc.requested_mode = e1000_fc_default;
5872 adapter->hw.fc.current_mode = e1000_fc_default;
5873 adapter->hw.phy.autoneg_advertised = 0x2f;
5875 /* ring size defaults */
5876 adapter->rx_ring->count = 256;
5877 adapter->tx_ring->count = 256;
5880 * Initial Wake on LAN setting - If APM wake is enabled in
5881 * the EEPROM, enable the ACPI Magic Packet filter
5883 if (adapter->flags & FLAG_APME_IN_WUC) {
5884 /* APME bit in EEPROM is mapped to WUC.APME */
5885 eeprom_data = er32(WUC);
5886 eeprom_apme_mask = E1000_WUC_APME;
5887 if (eeprom_data & E1000_WUC_PHY_WAKE)
5888 adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5889 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5890 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5891 (adapter->hw.bus.func == 1))
5892 e1000_read_nvm(&adapter->hw,
5893 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5894 else
5895 e1000_read_nvm(&adapter->hw,
5896 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5899 /* fetch WoL from EEPROM */
5900 if (eeprom_data & eeprom_apme_mask)
5901 adapter->eeprom_wol |= E1000_WUFC_MAG;
5904 * now that we have the eeprom settings, apply the special cases
5905 * where the eeprom may be wrong or the board simply won't support
5906 * wake on lan on a particular port
5908 if (!(adapter->flags & FLAG_HAS_WOL))
5909 adapter->eeprom_wol = 0;
5911 /* initialize the wol settings based on the eeprom settings */
5912 adapter->wol = adapter->eeprom_wol;
5913 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5915 /* save off EEPROM version number */
5916 e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5918 /* reset the hardware with the new settings */
5919 e1000e_reset(adapter);
5922 * If the controller has AMT, do not set DRV_LOAD until the interface
5923 * is up. For all other cases, let the f/w know that the h/w is now
5924 * under the control of the driver.
5926 if (!(adapter->flags & FLAG_HAS_AMT))
5927 e1000_get_hw_control(adapter);
5929 strcpy(netdev->name, "eth%d");
5930 err = register_netdev(netdev);
5931 if (err)
5932 goto err_register;
5934 /* carrier off reporting is important to ethtool even BEFORE open */
5935 netif_carrier_off(netdev);
5937 e1000_print_device_info(adapter);
5939 if (pci_dev_run_wake(pdev))
5940 pm_runtime_put_noidle(&pdev->dev);
5942 return 0;
5944 err_register:
5945 if (!(adapter->flags & FLAG_HAS_AMT))
5946 e1000_release_hw_control(adapter);
5947 err_eeprom:
5948 if (!e1000_check_reset_block(&adapter->hw))
5949 e1000_phy_hw_reset(&adapter->hw);
5950 err_hw_init:
5952 kfree(adapter->tx_ring);
5953 kfree(adapter->rx_ring);
5954 err_sw_init:
5955 if (adapter->hw.flash_address)
5956 iounmap(adapter->hw.flash_address);
5957 e1000e_reset_interrupt_capability(adapter);
5958 err_flashmap:
5959 iounmap(adapter->hw.hw_addr);
5960 err_ioremap:
5961 free_netdev(netdev);
5962 err_alloc_etherdev:
5963 pci_release_selected_regions(pdev,
5964 pci_select_bars(pdev, IORESOURCE_MEM));
5965 err_pci_reg:
5966 err_dma:
5967 pci_disable_device(pdev);
5968 return err;
5972 * e1000_remove - Device Removal Routine
5973 * @pdev: PCI device information struct
5975 * e1000_remove is called by the PCI subsystem to alert the driver
5976 * that it should release a PCI device. The could be caused by a
5977 * Hot-Plug event, or because the driver is going to be removed from
5978 * memory.
5980 static void __devexit e1000_remove(struct pci_dev *pdev)
5982 struct net_device *netdev = pci_get_drvdata(pdev);
5983 struct e1000_adapter *adapter = netdev_priv(netdev);
5984 bool down = test_bit(__E1000_DOWN, &adapter->state);
5987 * flush_scheduled work may reschedule our watchdog task, so
5988 * explicitly disable watchdog tasks from being rescheduled
5990 if (!down)
5991 set_bit(__E1000_DOWN, &adapter->state);
5992 del_timer_sync(&adapter->watchdog_timer);
5993 del_timer_sync(&adapter->phy_info_timer);
5995 cancel_work_sync(&adapter->reset_task);
5996 cancel_work_sync(&adapter->watchdog_task);
5997 cancel_work_sync(&adapter->downshift_task);
5998 cancel_work_sync(&adapter->update_phy_task);
5999 cancel_work_sync(&adapter->print_hang_task);
6000 flush_scheduled_work();
6002 if (!(netdev->flags & IFF_UP))
6003 e1000_power_down_phy(adapter);
6005 /* Don't lie to e1000_close() down the road. */
6006 if (!down)
6007 clear_bit(__E1000_DOWN, &adapter->state);
6008 unregister_netdev(netdev);
6010 if (pci_dev_run_wake(pdev))
6011 pm_runtime_get_noresume(&pdev->dev);
6014 * Release control of h/w to f/w. If f/w is AMT enabled, this
6015 * would have already happened in close and is redundant.
6017 e1000_release_hw_control(adapter);
6019 e1000e_reset_interrupt_capability(adapter);
6020 kfree(adapter->tx_ring);
6021 kfree(adapter->rx_ring);
6023 iounmap(adapter->hw.hw_addr);
6024 if (adapter->hw.flash_address)
6025 iounmap(adapter->hw.flash_address);
6026 pci_release_selected_regions(pdev,
6027 pci_select_bars(pdev, IORESOURCE_MEM));
6029 free_netdev(netdev);
6031 /* AER disable */
6032 pci_disable_pcie_error_reporting(pdev);
6034 pci_disable_device(pdev);
6037 /* PCI Error Recovery (ERS) */
6038 static struct pci_error_handlers e1000_err_handler = {
6039 .error_detected = e1000_io_error_detected,
6040 .slot_reset = e1000_io_slot_reset,
6041 .resume = e1000_io_resume,
6044 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6045 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6046 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6047 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6048 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6049 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6050 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6051 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6052 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6053 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6055 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6056 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6057 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6058 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6060 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6061 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6062 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6064 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6065 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6066 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6068 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6069 board_80003es2lan },
6070 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6071 board_80003es2lan },
6072 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6073 board_80003es2lan },
6074 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6075 board_80003es2lan },
6077 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6078 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6079 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6080 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6081 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6082 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6083 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6084 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6086 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6087 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6088 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6089 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6090 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6091 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6092 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6093 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6094 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6096 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6097 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6098 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6100 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6101 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6102 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6104 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6105 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6106 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6107 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6109 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6110 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6112 { } /* terminate list */
6114 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6116 #ifdef CONFIG_PM_OPS
6117 static const struct dev_pm_ops e1000_pm_ops = {
6118 SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6119 SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6120 e1000_runtime_resume, e1000_idle)
6122 #endif
6124 /* PCI Device API Driver */
6125 static struct pci_driver e1000_driver = {
6126 .name = e1000e_driver_name,
6127 .id_table = e1000_pci_tbl,
6128 .probe = e1000_probe,
6129 .remove = __devexit_p(e1000_remove),
6130 #ifdef CONFIG_PM_OPS
6131 .driver.pm = &e1000_pm_ops,
6132 #endif
6133 .shutdown = e1000_shutdown,
6134 .err_handler = &e1000_err_handler
6138 * e1000_init_module - Driver Registration Routine
6140 * e1000_init_module is the first routine called when the driver is
6141 * loaded. All it does is register with the PCI subsystem.
6143 static int __init e1000_init_module(void)
6145 int ret;
6146 pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6147 e1000e_driver_version);
6148 pr_info("Copyright (c) 1999 - 2010 Intel Corporation.\n");
6149 ret = pci_register_driver(&e1000_driver);
6151 return ret;
6153 module_init(e1000_init_module);
6156 * e1000_exit_module - Driver Exit Cleanup Routine
6158 * e1000_exit_module is called just before the driver is removed
6159 * from memory.
6161 static void __exit e1000_exit_module(void)
6163 pci_unregister_driver(&e1000_driver);
6165 module_exit(e1000_exit_module);
6168 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6169 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6170 MODULE_LICENSE("GPL");
6171 MODULE_VERSION(DRV_VERSION);
6173 /* e1000_main.c */