projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/net...
[deliverable/linux.git] / drivers / net / ethernet / intel / e1000e / netdev.c
1 /*******************************************************************************
2
3 Intel PRO/1000 Linux driver
4 Copyright(c) 1999 - 2011 Intel Corporation.
5
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.
9
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.
14
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.
18
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
21
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
26
27 *******************************************************************************/
28
29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30
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/interrupt.h>
40 #include <linux/tcp.h>
41 #include <linux/ipv6.h>
42 #include <linux/slab.h>
43 #include <net/checksum.h>
44 #include <net/ip6_checksum.h>
45 #include <linux/mii.h>
46 #include <linux/ethtool.h>
47 #include <linux/if_vlan.h>
48 #include <linux/cpu.h>
49 #include <linux/smp.h>
50 #include <linux/pm_qos.h>
51 #include <linux/pm_runtime.h>
52 #include <linux/aer.h>
53 #include <linux/prefetch.h>
54
55 #include "e1000.h"
56
57 #define DRV_EXTRAVERSION "-k"
58
59 #define DRV_VERSION "1.5.1" DRV_EXTRAVERSION
60 char e1000e_driver_name[] = "e1000e";
61 const char e1000e_driver_version[] = DRV_VERSION;
62
63 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state);
64
65 static const struct e1000_info *e1000_info_tbl[] = {
66 [board_82571] = &e1000_82571_info,
67 [board_82572] = &e1000_82572_info,
68 [board_82573] = &e1000_82573_info,
69 [board_82574] = &e1000_82574_info,
70 [board_82583] = &e1000_82583_info,
71 [board_80003es2lan] = &e1000_es2_info,
72 [board_ich8lan] = &e1000_ich8_info,
73 [board_ich9lan] = &e1000_ich9_info,
74 [board_ich10lan] = &e1000_ich10_info,
75 [board_pchlan] = &e1000_pch_info,
76 [board_pch2lan] = &e1000_pch2_info,
77 };
78
79 struct e1000_reg_info {
80 u32 ofs;
81 char *name;
82 };
83
84 #define E1000_RDFH 0x02410 /* Rx Data FIFO Head - RW */
85 #define E1000_RDFT 0x02418 /* Rx Data FIFO Tail - RW */
86 #define E1000_RDFHS 0x02420 /* Rx Data FIFO Head Saved - RW */
87 #define E1000_RDFTS 0x02428 /* Rx Data FIFO Tail Saved - RW */
88 #define E1000_RDFPC 0x02430 /* Rx Data FIFO Packet Count - RW */
89
90 #define E1000_TDFH 0x03410 /* Tx Data FIFO Head - RW */
91 #define E1000_TDFT 0x03418 /* Tx Data FIFO Tail - RW */
92 #define E1000_TDFHS 0x03420 /* Tx Data FIFO Head Saved - RW */
93 #define E1000_TDFTS 0x03428 /* Tx Data FIFO Tail Saved - RW */
94 #define E1000_TDFPC 0x03430 /* Tx Data FIFO Packet Count - RW */
95
96 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
97
98 /* General Registers */
99 {E1000_CTRL, "CTRL"},
100 {E1000_STATUS, "STATUS"},
101 {E1000_CTRL_EXT, "CTRL_EXT"},
102
103 /* Interrupt Registers */
104 {E1000_ICR, "ICR"},
105
106 /* Rx Registers */
107 {E1000_RCTL, "RCTL"},
108 {E1000_RDLEN, "RDLEN"},
109 {E1000_RDH, "RDH"},
110 {E1000_RDT, "RDT"},
111 {E1000_RDTR, "RDTR"},
112 {E1000_RXDCTL(0), "RXDCTL"},
113 {E1000_ERT, "ERT"},
114 {E1000_RDBAL, "RDBAL"},
115 {E1000_RDBAH, "RDBAH"},
116 {E1000_RDFH, "RDFH"},
117 {E1000_RDFT, "RDFT"},
118 {E1000_RDFHS, "RDFHS"},
119 {E1000_RDFTS, "RDFTS"},
120 {E1000_RDFPC, "RDFPC"},
121
122 /* Tx Registers */
123 {E1000_TCTL, "TCTL"},
124 {E1000_TDBAL, "TDBAL"},
125 {E1000_TDBAH, "TDBAH"},
126 {E1000_TDLEN, "TDLEN"},
127 {E1000_TDH, "TDH"},
128 {E1000_TDT, "TDT"},
129 {E1000_TIDV, "TIDV"},
130 {E1000_TXDCTL(0), "TXDCTL"},
131 {E1000_TADV, "TADV"},
132 {E1000_TARC(0), "TARC"},
133 {E1000_TDFH, "TDFH"},
134 {E1000_TDFT, "TDFT"},
135 {E1000_TDFHS, "TDFHS"},
136 {E1000_TDFTS, "TDFTS"},
137 {E1000_TDFPC, "TDFPC"},
138
139 /* List Terminator */
140 {}
141 };
142
143 /*
144 * e1000_regdump - register printout routine
145 */
146 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
147 {
148 int n = 0;
149 char rname[16];
150 u32 regs[8];
151
152 switch (reginfo->ofs) {
153 case E1000_RXDCTL(0):
154 for (n = 0; n < 2; n++)
155 regs[n] = __er32(hw, E1000_RXDCTL(n));
156 break;
157 case E1000_TXDCTL(0):
158 for (n = 0; n < 2; n++)
159 regs[n] = __er32(hw, E1000_TXDCTL(n));
160 break;
161 case E1000_TARC(0):
162 for (n = 0; n < 2; n++)
163 regs[n] = __er32(hw, E1000_TARC(n));
164 break;
165 default:
166 pr_info("%-15s %08x\n",
167 reginfo->name, __er32(hw, reginfo->ofs));
168 return;
169 }
170
171 snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
172 pr_info("%-15s %08x %08x\n", rname, regs[0], regs[1]);
173 }
174
175 /*
176 * e1000e_dump - Print registers, Tx-ring and Rx-ring
177 */
178 static void e1000e_dump(struct e1000_adapter *adapter)
179 {
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 {
186 u64 a;
187 u64 b;
188 } *u0;
189 struct e1000_buffer *buffer_info;
190 struct e1000_ring *rx_ring = adapter->rx_ring;
191 union e1000_rx_desc_packet_split *rx_desc_ps;
192 union e1000_rx_desc_extended *rx_desc;
193 struct my_u1 {
194 u64 a;
195 u64 b;
196 u64 c;
197 u64 d;
198 } *u1;
199 u32 staterr;
200 int i = 0;
201
202 if (!netif_msg_hw(adapter))
203 return;
204
205 /* Print netdevice Info */
206 if (netdev) {
207 dev_info(&adapter->pdev->dev, "Net device Info\n");
208 pr_info("Device Name state trans_start last_rx\n");
209 pr_info("%-15s %016lX %016lX %016lX\n",
210 netdev->name, netdev->state, netdev->trans_start,
211 netdev->last_rx);
212 }
213
214 /* Print Registers */
215 dev_info(&adapter->pdev->dev, "Register Dump\n");
216 pr_info(" Register Name Value\n");
217 for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
218 reginfo->name; reginfo++) {
219 e1000_regdump(hw, reginfo);
220 }
221
222 /* Print Tx Ring Summary */
223 if (!netdev || !netif_running(netdev))
224 goto exit;
225
226 dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
227 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
228 buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
229 pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
230 0, tx_ring->next_to_use, tx_ring->next_to_clean,
231 (unsigned long long)buffer_info->dma,
232 buffer_info->length,
233 buffer_info->next_to_watch,
234 (unsigned long long)buffer_info->time_stamp);
235
236 /* Print Tx Ring */
237 if (!netif_msg_tx_done(adapter))
238 goto rx_ring_summary;
239
240 dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
241
242 /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
243 *
244 * Legacy Transmit Descriptor
245 * +--------------------------------------------------------------+
246 * 0 | Buffer Address [63:0] (Reserved on Write Back) |
247 * +--------------------------------------------------------------+
248 * 8 | Special | CSS | Status | CMD | CSO | Length |
249 * +--------------------------------------------------------------+
250 * 63 48 47 36 35 32 31 24 23 16 15 0
251 *
252 * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
253 * 63 48 47 40 39 32 31 16 15 8 7 0
254 * +----------------------------------------------------------------+
255 * 0 | TUCSE | TUCS0 | TUCSS | IPCSE | IPCS0 | IPCSS |
256 * +----------------------------------------------------------------+
257 * 8 | MSS | HDRLEN | RSV | STA | TUCMD | DTYP | PAYLEN |
258 * +----------------------------------------------------------------+
259 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
260 *
261 * Extended Data Descriptor (DTYP=0x1)
262 * +----------------------------------------------------------------+
263 * 0 | Buffer Address [63:0] |
264 * +----------------------------------------------------------------+
265 * 8 | VLAN tag | POPTS | Rsvd | Status | Command | DTYP | DTALEN |
266 * +----------------------------------------------------------------+
267 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
268 */
269 pr_info("Tl[desc] [address 63:0 ] [SpeCssSCmCsLen] [bi->dma ] leng ntw timestamp bi->skb <-- Legacy format\n");
270 pr_info("Tc[desc] [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma ] leng ntw timestamp bi->skb <-- Ext Context format\n");
271 pr_info("Td[desc] [address 63:0 ] [VlaPoRSCm1Dlen] [bi->dma ] leng ntw timestamp bi->skb <-- Ext Data format\n");
272 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
273 const char *next_desc;
274 tx_desc = E1000_TX_DESC(*tx_ring, i);
275 buffer_info = &tx_ring->buffer_info[i];
276 u0 = (struct my_u0 *)tx_desc;
277 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
278 next_desc = " NTC/U";
279 else if (i == tx_ring->next_to_use)
280 next_desc = " NTU";
281 else if (i == tx_ring->next_to_clean)
282 next_desc = " NTC";
283 else
284 next_desc = "";
285 pr_info("T%c[0x%03X] %016llX %016llX %016llX %04X %3X %016llX %p%s\n",
286 (!(le64_to_cpu(u0->b) & (1 << 29)) ? 'l' :
287 ((le64_to_cpu(u0->b) & (1 << 20)) ? 'd' : 'c')),
288 i,
289 (unsigned long long)le64_to_cpu(u0->a),
290 (unsigned long long)le64_to_cpu(u0->b),
291 (unsigned long long)buffer_info->dma,
292 buffer_info->length, buffer_info->next_to_watch,
293 (unsigned long long)buffer_info->time_stamp,
294 buffer_info->skb, next_desc);
295
296 if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
297 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
298 16, 1, phys_to_virt(buffer_info->dma),
299 buffer_info->length, true);
300 }
301
302 /* Print Rx Ring Summary */
303 rx_ring_summary:
304 dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
305 pr_info("Queue [NTU] [NTC]\n");
306 pr_info(" %5d %5X %5X\n",
307 0, rx_ring->next_to_use, rx_ring->next_to_clean);
308
309 /* Print Rx Ring */
310 if (!netif_msg_rx_status(adapter))
311 goto exit;
312
313 dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
314 switch (adapter->rx_ps_pages) {
315 case 1:
316 case 2:
317 case 3:
318 /* [Extended] Packet Split Receive Descriptor Format
319 *
320 * +-----------------------------------------------------+
321 * 0 | Buffer Address 0 [63:0] |
322 * +-----------------------------------------------------+
323 * 8 | Buffer Address 1 [63:0] |
324 * +-----------------------------------------------------+
325 * 16 | Buffer Address 2 [63:0] |
326 * +-----------------------------------------------------+
327 * 24 | Buffer Address 3 [63:0] |
328 * +-----------------------------------------------------+
329 */
330 pr_info("R [desc] [buffer 0 63:0 ] [buffer 1 63:0 ] [buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma ] [bi->skb] <-- Ext Pkt Split format\n");
331 /* [Extended] Receive Descriptor (Write-Back) Format
332 *
333 * 63 48 47 32 31 13 12 8 7 4 3 0
334 * +------------------------------------------------------+
335 * 0 | Packet | IP | Rsvd | MRQ | Rsvd | MRQ RSS |
336 * | Checksum | Ident | | Queue | | Type |
337 * +------------------------------------------------------+
338 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
339 * +------------------------------------------------------+
340 * 63 48 47 32 31 20 19 0
341 */
342 pr_info("RWB[desc] [ck ipid mrqhsh] [vl l0 ee es] [ l3 l2 l1 hs] [reserved ] ---------------- [bi->skb] <-- Ext Rx Write-Back format\n");
343 for (i = 0; i < rx_ring->count; i++) {
344 const char *next_desc;
345 buffer_info = &rx_ring->buffer_info[i];
346 rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
347 u1 = (struct my_u1 *)rx_desc_ps;
348 staterr =
349 le32_to_cpu(rx_desc_ps->wb.middle.status_error);
350
351 if (i == rx_ring->next_to_use)
352 next_desc = " NTU";
353 else if (i == rx_ring->next_to_clean)
354 next_desc = " NTC";
355 else
356 next_desc = "";
357
358 if (staterr & E1000_RXD_STAT_DD) {
359 /* Descriptor Done */
360 pr_info("%s[0x%03X] %016llX %016llX %016llX %016llX ---------------- %p%s\n",
361 "RWB", i,
362 (unsigned long long)le64_to_cpu(u1->a),
363 (unsigned long long)le64_to_cpu(u1->b),
364 (unsigned long long)le64_to_cpu(u1->c),
365 (unsigned long long)le64_to_cpu(u1->d),
366 buffer_info->skb, next_desc);
367 } else {
368 pr_info("%s[0x%03X] %016llX %016llX %016llX %016llX %016llX %p%s\n",
369 "R ", i,
370 (unsigned long long)le64_to_cpu(u1->a),
371 (unsigned long long)le64_to_cpu(u1->b),
372 (unsigned long long)le64_to_cpu(u1->c),
373 (unsigned long long)le64_to_cpu(u1->d),
374 (unsigned long long)buffer_info->dma,
375 buffer_info->skb, next_desc);
376
377 if (netif_msg_pktdata(adapter))
378 print_hex_dump(KERN_INFO, "",
379 DUMP_PREFIX_ADDRESS, 16, 1,
380 phys_to_virt(buffer_info->dma),
381 adapter->rx_ps_bsize0, true);
382 }
383 }
384 break;
385 default:
386 case 0:
387 /* Extended Receive Descriptor (Read) Format
388 *
389 * +-----------------------------------------------------+
390 * 0 | Buffer Address [63:0] |
391 * +-----------------------------------------------------+
392 * 8 | Reserved |
393 * +-----------------------------------------------------+
394 */
395 pr_info("R [desc] [buf addr 63:0 ] [reserved 63:0 ] [bi->dma ] [bi->skb] <-- Ext (Read) format\n");
396 /* Extended Receive Descriptor (Write-Back) Format
397 *
398 * 63 48 47 32 31 24 23 4 3 0
399 * +------------------------------------------------------+
400 * | RSS Hash | | | |
401 * 0 +-------------------+ Rsvd | Reserved | MRQ RSS |
402 * | Packet | IP | | | Type |
403 * | Checksum | Ident | | | |
404 * +------------------------------------------------------+
405 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
406 * +------------------------------------------------------+
407 * 63 48 47 32 31 20 19 0
408 */
409 pr_info("RWB[desc] [cs ipid mrq] [vt ln xe xs] [bi->skb] <-- Ext (Write-Back) format\n");
410
411 for (i = 0; i < rx_ring->count; i++) {
412 const char *next_desc;
413
414 buffer_info = &rx_ring->buffer_info[i];
415 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
416 u1 = (struct my_u1 *)rx_desc;
417 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
418
419 if (i == rx_ring->next_to_use)
420 next_desc = " NTU";
421 else if (i == rx_ring->next_to_clean)
422 next_desc = " NTC";
423 else
424 next_desc = "";
425
426 if (staterr & E1000_RXD_STAT_DD) {
427 /* Descriptor Done */
428 pr_info("%s[0x%03X] %016llX %016llX ---------------- %p%s\n",
429 "RWB", i,
430 (unsigned long long)le64_to_cpu(u1->a),
431 (unsigned long long)le64_to_cpu(u1->b),
432 buffer_info->skb, next_desc);
433 } else {
434 pr_info("%s[0x%03X] %016llX %016llX %016llX %p%s\n",
435 "R ", i,
436 (unsigned long long)le64_to_cpu(u1->a),
437 (unsigned long long)le64_to_cpu(u1->b),
438 (unsigned long long)buffer_info->dma,
439 buffer_info->skb, next_desc);
440
441 if (netif_msg_pktdata(adapter))
442 print_hex_dump(KERN_INFO, "",
443 DUMP_PREFIX_ADDRESS, 16,
444 1,
445 phys_to_virt
446 (buffer_info->dma),
447 adapter->rx_buffer_len,
448 true);
449 }
450 }
451 }
452
453 exit:
454 return;
455 }
456
457 /**
458 * e1000_desc_unused - calculate if we have unused descriptors
459 **/
460 static int e1000_desc_unused(struct e1000_ring *ring)
461 {
462 if (ring->next_to_clean > ring->next_to_use)
463 return ring->next_to_clean - ring->next_to_use - 1;
464
465 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
466 }
467
468 /**
469 * e1000_receive_skb - helper function to handle Rx indications
470 * @adapter: board private structure
471 * @status: descriptor status field as written by hardware
472 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
473 * @skb: pointer to sk_buff to be indicated to stack
474 **/
475 static void e1000_receive_skb(struct e1000_adapter *adapter,
476 struct net_device *netdev, struct sk_buff *skb,
477 u8 status, __le16 vlan)
478 {
479 u16 tag = le16_to_cpu(vlan);
480 skb->protocol = eth_type_trans(skb, netdev);
481
482 if (status & E1000_RXD_STAT_VP)
483 __vlan_hwaccel_put_tag(skb, tag);
484
485 napi_gro_receive(&adapter->napi, skb);
486 }
487
488 /**
489 * e1000_rx_checksum - Receive Checksum Offload
490 * @adapter: board private structure
491 * @status_err: receive descriptor status and error fields
492 * @csum: receive descriptor csum field
493 * @sk_buff: socket buffer with received data
494 **/
495 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
496 u32 csum, struct sk_buff *skb)
497 {
498 u16 status = (u16)status_err;
499 u8 errors = (u8)(status_err >> 24);
500
501 skb_checksum_none_assert(skb);
502
503 /* Ignore Checksum bit is set */
504 if (status & E1000_RXD_STAT_IXSM)
505 return;
506 /* TCP/UDP checksum error bit is set */
507 if (errors & E1000_RXD_ERR_TCPE) {
508 /* let the stack verify checksum errors */
509 adapter->hw_csum_err++;
510 return;
511 }
512
513 /* TCP/UDP Checksum has not been calculated */
514 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
515 return;
516
517 /* It must be a TCP or UDP packet with a valid checksum */
518 if (status & E1000_RXD_STAT_TCPCS) {
519 /* TCP checksum is good */
520 skb->ip_summed = CHECKSUM_UNNECESSARY;
521 } else {
522 /*
523 * IP fragment with UDP payload
524 * Hardware complements the payload checksum, so we undo it
525 * and then put the value in host order for further stack use.
526 */
527 __sum16 sum = (__force __sum16)htons(csum);
528 skb->csum = csum_unfold(~sum);
529 skb->ip_summed = CHECKSUM_COMPLETE;
530 }
531 adapter->hw_csum_good++;
532 }
533
534 /**
535 * e1000e_update_tail_wa - helper function for e1000e_update_[rt]dt_wa()
536 * @hw: pointer to the HW structure
537 * @tail: address of tail descriptor register
538 * @i: value to write to tail descriptor register
539 *
540 * When updating the tail register, the ME could be accessing Host CSR
541 * registers at the same time. Normally, this is handled in h/w by an
542 * arbiter but on some parts there is a bug that acknowledges Host accesses
543 * later than it should which could result in the descriptor register to
544 * have an incorrect value. Workaround this by checking the FWSM register
545 * which has bit 24 set while ME is accessing Host CSR registers, wait
546 * if it is set and try again a number of times.
547 **/
548 static inline s32 e1000e_update_tail_wa(struct e1000_hw *hw, u8 __iomem * tail,
549 unsigned int i)
550 {
551 unsigned int j = 0;
552
553 while ((j++ < E1000_ICH_FWSM_PCIM2PCI_COUNT) &&
554 (er32(FWSM) & E1000_ICH_FWSM_PCIM2PCI))
555 udelay(50);
556
557 writel(i, tail);
558
559 if ((j == E1000_ICH_FWSM_PCIM2PCI_COUNT) && (i != readl(tail)))
560 return E1000_ERR_SWFW_SYNC;
561
562 return 0;
563 }
564
565 static void e1000e_update_rdt_wa(struct e1000_adapter *adapter, unsigned int i)
566 {
567 u8 __iomem *tail = (adapter->hw.hw_addr + adapter->rx_ring->tail);
568 struct e1000_hw *hw = &adapter->hw;
569
570 if (e1000e_update_tail_wa(hw, tail, i)) {
571 u32 rctl = er32(RCTL);
572 ew32(RCTL, rctl & ~E1000_RCTL_EN);
573 e_err("ME firmware caused invalid RDT - resetting\n");
574 schedule_work(&adapter->reset_task);
575 }
576 }
577
578 static void e1000e_update_tdt_wa(struct e1000_adapter *adapter, unsigned int i)
579 {
580 u8 __iomem *tail = (adapter->hw.hw_addr + adapter->tx_ring->tail);
581 struct e1000_hw *hw = &adapter->hw;
582
583 if (e1000e_update_tail_wa(hw, tail, i)) {
584 u32 tctl = er32(TCTL);
585 ew32(TCTL, tctl & ~E1000_TCTL_EN);
586 e_err("ME firmware caused invalid TDT - resetting\n");
587 schedule_work(&adapter->reset_task);
588 }
589 }
590
591 /**
592 * e1000_alloc_rx_buffers - Replace used receive buffers
593 * @adapter: address of board private structure
594 **/
595 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
596 int cleaned_count, gfp_t gfp)
597 {
598 struct net_device *netdev = adapter->netdev;
599 struct pci_dev *pdev = adapter->pdev;
600 struct e1000_ring *rx_ring = adapter->rx_ring;
601 union e1000_rx_desc_extended *rx_desc;
602 struct e1000_buffer *buffer_info;
603 struct sk_buff *skb;
604 unsigned int i;
605 unsigned int bufsz = adapter->rx_buffer_len;
606
607 i = rx_ring->next_to_use;
608 buffer_info = &rx_ring->buffer_info[i];
609
610 while (cleaned_count--) {
611 skb = buffer_info->skb;
612 if (skb) {
613 skb_trim(skb, 0);
614 goto map_skb;
615 }
616
617 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
618 if (!skb) {
619 /* Better luck next round */
620 adapter->alloc_rx_buff_failed++;
621 break;
622 }
623
624 buffer_info->skb = skb;
625 map_skb:
626 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
627 adapter->rx_buffer_len,
628 DMA_FROM_DEVICE);
629 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
630 dev_err(&pdev->dev, "Rx DMA map failed\n");
631 adapter->rx_dma_failed++;
632 break;
633 }
634
635 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
636 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
637
638 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
639 /*
640 * Force memory writes to complete before letting h/w
641 * know there are new descriptors to fetch. (Only
642 * applicable for weak-ordered memory model archs,
643 * such as IA-64).
644 */
645 wmb();
646 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
647 e1000e_update_rdt_wa(adapter, i);
648 else
649 writel(i, adapter->hw.hw_addr + rx_ring->tail);
650 }
651 i++;
652 if (i == rx_ring->count)
653 i = 0;
654 buffer_info = &rx_ring->buffer_info[i];
655 }
656
657 rx_ring->next_to_use = i;
658 }
659
660 /**
661 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
662 * @adapter: address of board private structure
663 **/
664 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
665 int cleaned_count, gfp_t gfp)
666 {
667 struct net_device *netdev = adapter->netdev;
668 struct pci_dev *pdev = adapter->pdev;
669 union e1000_rx_desc_packet_split *rx_desc;
670 struct e1000_ring *rx_ring = adapter->rx_ring;
671 struct e1000_buffer *buffer_info;
672 struct e1000_ps_page *ps_page;
673 struct sk_buff *skb;
674 unsigned int i, j;
675
676 i = rx_ring->next_to_use;
677 buffer_info = &rx_ring->buffer_info[i];
678
679 while (cleaned_count--) {
680 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
681
682 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
683 ps_page = &buffer_info->ps_pages[j];
684 if (j >= adapter->rx_ps_pages) {
685 /* all unused desc entries get hw null ptr */
686 rx_desc->read.buffer_addr[j + 1] =
687 ~cpu_to_le64(0);
688 continue;
689 }
690 if (!ps_page->page) {
691 ps_page->page = alloc_page(gfp);
692 if (!ps_page->page) {
693 adapter->alloc_rx_buff_failed++;
694 goto no_buffers;
695 }
696 ps_page->dma = dma_map_page(&pdev->dev,
697 ps_page->page,
698 0, PAGE_SIZE,
699 DMA_FROM_DEVICE);
700 if (dma_mapping_error(&pdev->dev,
701 ps_page->dma)) {
702 dev_err(&adapter->pdev->dev,
703 "Rx DMA page map failed\n");
704 adapter->rx_dma_failed++;
705 goto no_buffers;
706 }
707 }
708 /*
709 * Refresh the desc even if buffer_addrs
710 * didn't change because each write-back
711 * erases this info.
712 */
713 rx_desc->read.buffer_addr[j + 1] =
714 cpu_to_le64(ps_page->dma);
715 }
716
717 skb = __netdev_alloc_skb_ip_align(netdev,
718 adapter->rx_ps_bsize0,
719 gfp);
720
721 if (!skb) {
722 adapter->alloc_rx_buff_failed++;
723 break;
724 }
725
726 buffer_info->skb = skb;
727 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
728 adapter->rx_ps_bsize0,
729 DMA_FROM_DEVICE);
730 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
731 dev_err(&pdev->dev, "Rx DMA map failed\n");
732 adapter->rx_dma_failed++;
733 /* cleanup skb */
734 dev_kfree_skb_any(skb);
735 buffer_info->skb = NULL;
736 break;
737 }
738
739 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
740
741 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
742 /*
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 */
748 wmb();
749 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
750 e1000e_update_rdt_wa(adapter, i << 1);
751 else
752 writel(i << 1,
753 adapter->hw.hw_addr + rx_ring->tail);
754 }
755
756 i++;
757 if (i == rx_ring->count)
758 i = 0;
759 buffer_info = &rx_ring->buffer_info[i];
760 }
761
762 no_buffers:
763 rx_ring->next_to_use = i;
764 }
765
766 /**
767 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
768 * @adapter: address of board private structure
769 * @cleaned_count: number of buffers to allocate this pass
770 **/
771
772 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
773 int cleaned_count, gfp_t gfp)
774 {
775 struct net_device *netdev = adapter->netdev;
776 struct pci_dev *pdev = adapter->pdev;
777 union e1000_rx_desc_extended *rx_desc;
778 struct e1000_ring *rx_ring = adapter->rx_ring;
779 struct e1000_buffer *buffer_info;
780 struct sk_buff *skb;
781 unsigned int i;
782 unsigned int bufsz = 256 - 16 /* for skb_reserve */;
783
784 i = rx_ring->next_to_use;
785 buffer_info = &rx_ring->buffer_info[i];
786
787 while (cleaned_count--) {
788 skb = buffer_info->skb;
789 if (skb) {
790 skb_trim(skb, 0);
791 goto check_page;
792 }
793
794 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
795 if (unlikely(!skb)) {
796 /* Better luck next round */
797 adapter->alloc_rx_buff_failed++;
798 break;
799 }
800
801 buffer_info->skb = skb;
802 check_page:
803 /* allocate a new page if necessary */
804 if (!buffer_info->page) {
805 buffer_info->page = alloc_page(gfp);
806 if (unlikely(!buffer_info->page)) {
807 adapter->alloc_rx_buff_failed++;
808 break;
809 }
810 }
811
812 if (!buffer_info->dma)
813 buffer_info->dma = dma_map_page(&pdev->dev,
814 buffer_info->page, 0,
815 PAGE_SIZE,
816 DMA_FROM_DEVICE);
817
818 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
819 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
820
821 if (unlikely(++i == rx_ring->count))
822 i = 0;
823 buffer_info = &rx_ring->buffer_info[i];
824 }
825
826 if (likely(rx_ring->next_to_use != i)) {
827 rx_ring->next_to_use = i;
828 if (unlikely(i-- == 0))
829 i = (rx_ring->count - 1);
830
831 /* Force memory writes to complete before letting h/w
832 * know there are new descriptors to fetch. (Only
833 * applicable for weak-ordered memory model archs,
834 * such as IA-64). */
835 wmb();
836 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
837 e1000e_update_rdt_wa(adapter, i);
838 else
839 writel(i, adapter->hw.hw_addr + rx_ring->tail);
840 }
841 }
842
843 /**
844 * e1000_clean_rx_irq - Send received data up the network stack; legacy
845 * @adapter: board private structure
846 *
847 * the return value indicates whether actual cleaning was done, there
848 * is no guarantee that everything was cleaned
849 **/
850 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
851 int *work_done, int work_to_do)
852 {
853 struct net_device *netdev = adapter->netdev;
854 struct pci_dev *pdev = adapter->pdev;
855 struct e1000_hw *hw = &adapter->hw;
856 struct e1000_ring *rx_ring = adapter->rx_ring;
857 union e1000_rx_desc_extended *rx_desc, *next_rxd;
858 struct e1000_buffer *buffer_info, *next_buffer;
859 u32 length, staterr;
860 unsigned int i;
861 int cleaned_count = 0;
862 bool cleaned = 0;
863 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
864
865 i = rx_ring->next_to_clean;
866 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
867 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
868 buffer_info = &rx_ring->buffer_info[i];
869
870 while (staterr & E1000_RXD_STAT_DD) {
871 struct sk_buff *skb;
872
873 if (*work_done >= work_to_do)
874 break;
875 (*work_done)++;
876 rmb(); /* read descriptor and rx_buffer_info after status DD */
877
878 skb = buffer_info->skb;
879 buffer_info->skb = NULL;
880
881 prefetch(skb->data - NET_IP_ALIGN);
882
883 i++;
884 if (i == rx_ring->count)
885 i = 0;
886 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
887 prefetch(next_rxd);
888
889 next_buffer = &rx_ring->buffer_info[i];
890
891 cleaned = 1;
892 cleaned_count++;
893 dma_unmap_single(&pdev->dev,
894 buffer_info->dma,
895 adapter->rx_buffer_len,
896 DMA_FROM_DEVICE);
897 buffer_info->dma = 0;
898
899 length = le16_to_cpu(rx_desc->wb.upper.length);
900
901 /*
902 * !EOP means multiple descriptors were used to store a single
903 * packet, if that's the case we need to toss it. In fact, we
904 * need to toss every packet with the EOP bit clear and the
905 * next frame that _does_ have the EOP bit set, as it is by
906 * definition only a frame fragment
907 */
908 if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
909 adapter->flags2 |= FLAG2_IS_DISCARDING;
910
911 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
912 /* All receives must fit into a single buffer */
913 e_dbg("Receive packet consumed multiple buffers\n");
914 /* recycle */
915 buffer_info->skb = skb;
916 if (staterr & E1000_RXD_STAT_EOP)
917 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
918 goto next_desc;
919 }
920
921 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
922 /* recycle */
923 buffer_info->skb = skb;
924 goto next_desc;
925 }
926
927 /* adjust length to remove Ethernet CRC */
928 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
929 length -= 4;
930
931 total_rx_bytes += length;
932 total_rx_packets++;
933
934 /*
935 * code added for copybreak, this should improve
936 * performance for small packets with large amounts
937 * of reassembly being done in the stack
938 */
939 if (length < copybreak) {
940 struct sk_buff *new_skb =
941 netdev_alloc_skb_ip_align(netdev, length);
942 if (new_skb) {
943 skb_copy_to_linear_data_offset(new_skb,
944 -NET_IP_ALIGN,
945 (skb->data -
946 NET_IP_ALIGN),
947 (length +
948 NET_IP_ALIGN));
949 /* save the skb in buffer_info as good */
950 buffer_info->skb = skb;
951 skb = new_skb;
952 }
953 /* else just continue with the old one */
954 }
955 /* end copybreak code */
956 skb_put(skb, length);
957
958 /* Receive Checksum Offload */
959 e1000_rx_checksum(adapter, staterr,
960 le16_to_cpu(rx_desc->wb.lower.hi_dword.
961 csum_ip.csum), skb);
962
963 e1000_receive_skb(adapter, netdev, skb, staterr,
964 rx_desc->wb.upper.vlan);
965
966 next_desc:
967 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
968
969 /* return some buffers to hardware, one at a time is too slow */
970 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
971 adapter->alloc_rx_buf(adapter, cleaned_count,
972 GFP_ATOMIC);
973 cleaned_count = 0;
974 }
975
976 /* use prefetched values */
977 rx_desc = next_rxd;
978 buffer_info = next_buffer;
979
980 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
981 }
982 rx_ring->next_to_clean = i;
983
984 cleaned_count = e1000_desc_unused(rx_ring);
985 if (cleaned_count)
986 adapter->alloc_rx_buf(adapter, cleaned_count, GFP_ATOMIC);
987
988 adapter->total_rx_bytes += total_rx_bytes;
989 adapter->total_rx_packets += total_rx_packets;
990 return cleaned;
991 }
992
993 static void e1000_put_txbuf(struct e1000_adapter *adapter,
994 struct e1000_buffer *buffer_info)
995 {
996 if (buffer_info->dma) {
997 if (buffer_info->mapped_as_page)
998 dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
999 buffer_info->length, DMA_TO_DEVICE);
1000 else
1001 dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1002 buffer_info->length, DMA_TO_DEVICE);
1003 buffer_info->dma = 0;
1004 }
1005 if (buffer_info->skb) {
1006 dev_kfree_skb_any(buffer_info->skb);
1007 buffer_info->skb = NULL;
1008 }
1009 buffer_info->time_stamp = 0;
1010 }
1011
1012 static void e1000_print_hw_hang(struct work_struct *work)
1013 {
1014 struct e1000_adapter *adapter = container_of(work,
1015 struct e1000_adapter,
1016 print_hang_task);
1017 struct e1000_ring *tx_ring = adapter->tx_ring;
1018 unsigned int i = tx_ring->next_to_clean;
1019 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1020 struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1021 struct e1000_hw *hw = &adapter->hw;
1022 u16 phy_status, phy_1000t_status, phy_ext_status;
1023 u16 pci_status;
1024
1025 if (test_bit(__E1000_DOWN, &adapter->state))
1026 return;
1027
1028 e1e_rphy(hw, PHY_STATUS, &phy_status);
1029 e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
1030 e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
1031
1032 pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1033
1034 /* detected Hardware unit hang */
1035 e_err("Detected Hardware Unit Hang:\n"
1036 " TDH <%x>\n"
1037 " TDT <%x>\n"
1038 " next_to_use <%x>\n"
1039 " next_to_clean <%x>\n"
1040 "buffer_info[next_to_clean]:\n"
1041 " time_stamp <%lx>\n"
1042 " next_to_watch <%x>\n"
1043 " jiffies <%lx>\n"
1044 " next_to_watch.status <%x>\n"
1045 "MAC Status <%x>\n"
1046 "PHY Status <%x>\n"
1047 "PHY 1000BASE-T Status <%x>\n"
1048 "PHY Extended Status <%x>\n"
1049 "PCI Status <%x>\n",
1050 readl(adapter->hw.hw_addr + tx_ring->head),
1051 readl(adapter->hw.hw_addr + tx_ring->tail),
1052 tx_ring->next_to_use,
1053 tx_ring->next_to_clean,
1054 tx_ring->buffer_info[eop].time_stamp,
1055 eop,
1056 jiffies,
1057 eop_desc->upper.fields.status,
1058 er32(STATUS),
1059 phy_status,
1060 phy_1000t_status,
1061 phy_ext_status,
1062 pci_status);
1063 }
1064
1065 /**
1066 * e1000_clean_tx_irq - Reclaim resources after transmit completes
1067 * @adapter: board private structure
1068 *
1069 * the return value indicates whether actual cleaning was done, there
1070 * is no guarantee that everything was cleaned
1071 **/
1072 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
1073 {
1074 struct net_device *netdev = adapter->netdev;
1075 struct e1000_hw *hw = &adapter->hw;
1076 struct e1000_ring *tx_ring = adapter->tx_ring;
1077 struct e1000_tx_desc *tx_desc, *eop_desc;
1078 struct e1000_buffer *buffer_info;
1079 unsigned int i, eop;
1080 unsigned int count = 0;
1081 unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1082
1083 i = tx_ring->next_to_clean;
1084 eop = tx_ring->buffer_info[i].next_to_watch;
1085 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1086
1087 while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1088 (count < tx_ring->count)) {
1089 bool cleaned = false;
1090 rmb(); /* read buffer_info after eop_desc */
1091 for (; !cleaned; count++) {
1092 tx_desc = E1000_TX_DESC(*tx_ring, i);
1093 buffer_info = &tx_ring->buffer_info[i];
1094 cleaned = (i == eop);
1095
1096 if (cleaned) {
1097 total_tx_packets += buffer_info->segs;
1098 total_tx_bytes += buffer_info->bytecount;
1099 }
1100
1101 e1000_put_txbuf(adapter, buffer_info);
1102 tx_desc->upper.data = 0;
1103
1104 i++;
1105 if (i == tx_ring->count)
1106 i = 0;
1107 }
1108
1109 if (i == tx_ring->next_to_use)
1110 break;
1111 eop = tx_ring->buffer_info[i].next_to_watch;
1112 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1113 }
1114
1115 tx_ring->next_to_clean = i;
1116
1117 #define TX_WAKE_THRESHOLD 32
1118 if (count && netif_carrier_ok(netdev) &&
1119 e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1120 /* Make sure that anybody stopping the queue after this
1121 * sees the new next_to_clean.
1122 */
1123 smp_mb();
1124
1125 if (netif_queue_stopped(netdev) &&
1126 !(test_bit(__E1000_DOWN, &adapter->state))) {
1127 netif_wake_queue(netdev);
1128 ++adapter->restart_queue;
1129 }
1130 }
1131
1132 if (adapter->detect_tx_hung) {
1133 /*
1134 * Detect a transmit hang in hardware, this serializes the
1135 * check with the clearing of time_stamp and movement of i
1136 */
1137 adapter->detect_tx_hung = 0;
1138 if (tx_ring->buffer_info[i].time_stamp &&
1139 time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1140 + (adapter->tx_timeout_factor * HZ)) &&
1141 !(er32(STATUS) & E1000_STATUS_TXOFF)) {
1142 schedule_work(&adapter->print_hang_task);
1143 netif_stop_queue(netdev);
1144 }
1145 }
1146 adapter->total_tx_bytes += total_tx_bytes;
1147 adapter->total_tx_packets += total_tx_packets;
1148 return count < tx_ring->count;
1149 }
1150
1151 /**
1152 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1153 * @adapter: board private structure
1154 *
1155 * the return value indicates whether actual cleaning was done, there
1156 * is no guarantee that everything was cleaned
1157 **/
1158 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1159 int *work_done, int work_to_do)
1160 {
1161 struct e1000_hw *hw = &adapter->hw;
1162 union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1163 struct net_device *netdev = adapter->netdev;
1164 struct pci_dev *pdev = adapter->pdev;
1165 struct e1000_ring *rx_ring = adapter->rx_ring;
1166 struct e1000_buffer *buffer_info, *next_buffer;
1167 struct e1000_ps_page *ps_page;
1168 struct sk_buff *skb;
1169 unsigned int i, j;
1170 u32 length, staterr;
1171 int cleaned_count = 0;
1172 bool cleaned = 0;
1173 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1174
1175 i = rx_ring->next_to_clean;
1176 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1177 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1178 buffer_info = &rx_ring->buffer_info[i];
1179
1180 while (staterr & E1000_RXD_STAT_DD) {
1181 if (*work_done >= work_to_do)
1182 break;
1183 (*work_done)++;
1184 skb = buffer_info->skb;
1185 rmb(); /* read descriptor and rx_buffer_info after status DD */
1186
1187 /* in the packet split case this is header only */
1188 prefetch(skb->data - NET_IP_ALIGN);
1189
1190 i++;
1191 if (i == rx_ring->count)
1192 i = 0;
1193 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1194 prefetch(next_rxd);
1195
1196 next_buffer = &rx_ring->buffer_info[i];
1197
1198 cleaned = 1;
1199 cleaned_count++;
1200 dma_unmap_single(&pdev->dev, buffer_info->dma,
1201 adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1202 buffer_info->dma = 0;
1203
1204 /* see !EOP comment in other Rx routine */
1205 if (!(staterr & E1000_RXD_STAT_EOP))
1206 adapter->flags2 |= FLAG2_IS_DISCARDING;
1207
1208 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1209 e_dbg("Packet Split buffers didn't pick up the full packet\n");
1210 dev_kfree_skb_irq(skb);
1211 if (staterr & E1000_RXD_STAT_EOP)
1212 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1213 goto next_desc;
1214 }
1215
1216 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1217 dev_kfree_skb_irq(skb);
1218 goto next_desc;
1219 }
1220
1221 length = le16_to_cpu(rx_desc->wb.middle.length0);
1222
1223 if (!length) {
1224 e_dbg("Last part of the packet spanning multiple descriptors\n");
1225 dev_kfree_skb_irq(skb);
1226 goto next_desc;
1227 }
1228
1229 /* Good Receive */
1230 skb_put(skb, length);
1231
1232 {
1233 /*
1234 * this looks ugly, but it seems compiler issues make it
1235 * more efficient than reusing j
1236 */
1237 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1238
1239 /*
1240 * page alloc/put takes too long and effects small packet
1241 * throughput, so unsplit small packets and save the alloc/put
1242 * only valid in softirq (napi) context to call kmap_*
1243 */
1244 if (l1 && (l1 <= copybreak) &&
1245 ((length + l1) <= adapter->rx_ps_bsize0)) {
1246 u8 *vaddr;
1247
1248 ps_page = &buffer_info->ps_pages[0];
1249
1250 /*
1251 * there is no documentation about how to call
1252 * kmap_atomic, so we can't hold the mapping
1253 * very long
1254 */
1255 dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1256 PAGE_SIZE, DMA_FROM_DEVICE);
1257 vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1258 memcpy(skb_tail_pointer(skb), vaddr, l1);
1259 kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1260 dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1261 PAGE_SIZE, DMA_FROM_DEVICE);
1262
1263 /* remove the CRC */
1264 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1265 l1 -= 4;
1266
1267 skb_put(skb, l1);
1268 goto copydone;
1269 } /* if */
1270 }
1271
1272 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1273 length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1274 if (!length)
1275 break;
1276
1277 ps_page = &buffer_info->ps_pages[j];
1278 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1279 DMA_FROM_DEVICE);
1280 ps_page->dma = 0;
1281 skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1282 ps_page->page = NULL;
1283 skb->len += length;
1284 skb->data_len += length;
1285 skb->truesize += PAGE_SIZE;
1286 }
1287
1288 /* strip the ethernet crc, problem is we're using pages now so
1289 * this whole operation can get a little cpu intensive
1290 */
1291 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1292 pskb_trim(skb, skb->len - 4);
1293
1294 copydone:
1295 total_rx_bytes += skb->len;
1296 total_rx_packets++;
1297
1298 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1299 rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1300
1301 if (rx_desc->wb.upper.header_status &
1302 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1303 adapter->rx_hdr_split++;
1304
1305 e1000_receive_skb(adapter, netdev, skb,
1306 staterr, rx_desc->wb.middle.vlan);
1307
1308 next_desc:
1309 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1310 buffer_info->skb = NULL;
1311
1312 /* return some buffers to hardware, one at a time is too slow */
1313 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1314 adapter->alloc_rx_buf(adapter, cleaned_count,
1315 GFP_ATOMIC);
1316 cleaned_count = 0;
1317 }
1318
1319 /* use prefetched values */
1320 rx_desc = next_rxd;
1321 buffer_info = next_buffer;
1322
1323 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1324 }
1325 rx_ring->next_to_clean = i;
1326
1327 cleaned_count = e1000_desc_unused(rx_ring);
1328 if (cleaned_count)
1329 adapter->alloc_rx_buf(adapter, cleaned_count, GFP_ATOMIC);
1330
1331 adapter->total_rx_bytes += total_rx_bytes;
1332 adapter->total_rx_packets += total_rx_packets;
1333 return cleaned;
1334 }
1335
1336 /**
1337 * e1000_consume_page - helper function
1338 **/
1339 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1340 u16 length)
1341 {
1342 bi->page = NULL;
1343 skb->len += length;
1344 skb->data_len += length;
1345 skb->truesize += PAGE_SIZE;
1346 }
1347
1348 /**
1349 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1350 * @adapter: board private structure
1351 *
1352 * the return value indicates whether actual cleaning was done, there
1353 * is no guarantee that everything was cleaned
1354 **/
1355
1356 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1357 int *work_done, int work_to_do)
1358 {
1359 struct net_device *netdev = adapter->netdev;
1360 struct pci_dev *pdev = adapter->pdev;
1361 struct e1000_ring *rx_ring = adapter->rx_ring;
1362 union e1000_rx_desc_extended *rx_desc, *next_rxd;
1363 struct e1000_buffer *buffer_info, *next_buffer;
1364 u32 length, staterr;
1365 unsigned int i;
1366 int cleaned_count = 0;
1367 bool cleaned = false;
1368 unsigned int total_rx_bytes=0, total_rx_packets=0;
1369
1370 i = rx_ring->next_to_clean;
1371 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1372 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1373 buffer_info = &rx_ring->buffer_info[i];
1374
1375 while (staterr & E1000_RXD_STAT_DD) {
1376 struct sk_buff *skb;
1377
1378 if (*work_done >= work_to_do)
1379 break;
1380 (*work_done)++;
1381 rmb(); /* read descriptor and rx_buffer_info after status DD */
1382
1383 skb = buffer_info->skb;
1384 buffer_info->skb = NULL;
1385
1386 ++i;
1387 if (i == rx_ring->count)
1388 i = 0;
1389 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1390 prefetch(next_rxd);
1391
1392 next_buffer = &rx_ring->buffer_info[i];
1393
1394 cleaned = true;
1395 cleaned_count++;
1396 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1397 DMA_FROM_DEVICE);
1398 buffer_info->dma = 0;
1399
1400 length = le16_to_cpu(rx_desc->wb.upper.length);
1401
1402 /* errors is only valid for DD + EOP descriptors */
1403 if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1404 (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK))) {
1405 /* recycle both page and skb */
1406 buffer_info->skb = skb;
1407 /* an error means any chain goes out the window too */
1408 if (rx_ring->rx_skb_top)
1409 dev_kfree_skb_irq(rx_ring->rx_skb_top);
1410 rx_ring->rx_skb_top = NULL;
1411 goto next_desc;
1412 }
1413
1414 #define rxtop (rx_ring->rx_skb_top)
1415 if (!(staterr & E1000_RXD_STAT_EOP)) {
1416 /* this descriptor is only the beginning (or middle) */
1417 if (!rxtop) {
1418 /* this is the beginning of a chain */
1419 rxtop = skb;
1420 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1421 0, length);
1422 } else {
1423 /* this is the middle of a chain */
1424 skb_fill_page_desc(rxtop,
1425 skb_shinfo(rxtop)->nr_frags,
1426 buffer_info->page, 0, length);
1427 /* re-use the skb, only consumed the page */
1428 buffer_info->skb = skb;
1429 }
1430 e1000_consume_page(buffer_info, rxtop, length);
1431 goto next_desc;
1432 } else {
1433 if (rxtop) {
1434 /* end of the chain */
1435 skb_fill_page_desc(rxtop,
1436 skb_shinfo(rxtop)->nr_frags,
1437 buffer_info->page, 0, length);
1438 /* re-use the current skb, we only consumed the
1439 * page */
1440 buffer_info->skb = skb;
1441 skb = rxtop;
1442 rxtop = NULL;
1443 e1000_consume_page(buffer_info, skb, length);
1444 } else {
1445 /* no chain, got EOP, this buf is the packet
1446 * copybreak to save the put_page/alloc_page */
1447 if (length <= copybreak &&
1448 skb_tailroom(skb) >= length) {
1449 u8 *vaddr;
1450 vaddr = kmap_atomic(buffer_info->page,
1451 KM_SKB_DATA_SOFTIRQ);
1452 memcpy(skb_tail_pointer(skb), vaddr,
1453 length);
1454 kunmap_atomic(vaddr,
1455 KM_SKB_DATA_SOFTIRQ);
1456 /* re-use the page, so don't erase
1457 * buffer_info->page */
1458 skb_put(skb, length);
1459 } else {
1460 skb_fill_page_desc(skb, 0,
1461 buffer_info->page, 0,
1462 length);
1463 e1000_consume_page(buffer_info, skb,
1464 length);
1465 }
1466 }
1467 }
1468
1469 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1470 e1000_rx_checksum(adapter, staterr,
1471 le16_to_cpu(rx_desc->wb.lower.hi_dword.
1472 csum_ip.csum), skb);
1473
1474 /* probably a little skewed due to removing CRC */
1475 total_rx_bytes += skb->len;
1476 total_rx_packets++;
1477
1478 /* eth type trans needs skb->data to point to something */
1479 if (!pskb_may_pull(skb, ETH_HLEN)) {
1480 e_err("pskb_may_pull failed.\n");
1481 dev_kfree_skb_irq(skb);
1482 goto next_desc;
1483 }
1484
1485 e1000_receive_skb(adapter, netdev, skb, staterr,
1486 rx_desc->wb.upper.vlan);
1487
1488 next_desc:
1489 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1490
1491 /* return some buffers to hardware, one at a time is too slow */
1492 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1493 adapter->alloc_rx_buf(adapter, cleaned_count,
1494 GFP_ATOMIC);
1495 cleaned_count = 0;
1496 }
1497
1498 /* use prefetched values */
1499 rx_desc = next_rxd;
1500 buffer_info = next_buffer;
1501
1502 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1503 }
1504 rx_ring->next_to_clean = i;
1505
1506 cleaned_count = e1000_desc_unused(rx_ring);
1507 if (cleaned_count)
1508 adapter->alloc_rx_buf(adapter, cleaned_count, GFP_ATOMIC);
1509
1510 adapter->total_rx_bytes += total_rx_bytes;
1511 adapter->total_rx_packets += total_rx_packets;
1512 return cleaned;
1513 }
1514
1515 /**
1516 * e1000_clean_rx_ring - Free Rx Buffers per Queue
1517 * @adapter: board private structure
1518 **/
1519 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1520 {
1521 struct e1000_ring *rx_ring = adapter->rx_ring;
1522 struct e1000_buffer *buffer_info;
1523 struct e1000_ps_page *ps_page;
1524 struct pci_dev *pdev = adapter->pdev;
1525 unsigned int i, j;
1526
1527 /* Free all the Rx ring sk_buffs */
1528 for (i = 0; i < rx_ring->count; i++) {
1529 buffer_info = &rx_ring->buffer_info[i];
1530 if (buffer_info->dma) {
1531 if (adapter->clean_rx == e1000_clean_rx_irq)
1532 dma_unmap_single(&pdev->dev, buffer_info->dma,
1533 adapter->rx_buffer_len,
1534 DMA_FROM_DEVICE);
1535 else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1536 dma_unmap_page(&pdev->dev, buffer_info->dma,
1537 PAGE_SIZE,
1538 DMA_FROM_DEVICE);
1539 else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1540 dma_unmap_single(&pdev->dev, buffer_info->dma,
1541 adapter->rx_ps_bsize0,
1542 DMA_FROM_DEVICE);
1543 buffer_info->dma = 0;
1544 }
1545
1546 if (buffer_info->page) {
1547 put_page(buffer_info->page);
1548 buffer_info->page = NULL;
1549 }
1550
1551 if (buffer_info->skb) {
1552 dev_kfree_skb(buffer_info->skb);
1553 buffer_info->skb = NULL;
1554 }
1555
1556 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1557 ps_page = &buffer_info->ps_pages[j];
1558 if (!ps_page->page)
1559 break;
1560 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1561 DMA_FROM_DEVICE);
1562 ps_page->dma = 0;
1563 put_page(ps_page->page);
1564 ps_page->page = NULL;
1565 }
1566 }
1567
1568 /* there also may be some cached data from a chained receive */
1569 if (rx_ring->rx_skb_top) {
1570 dev_kfree_skb(rx_ring->rx_skb_top);
1571 rx_ring->rx_skb_top = NULL;
1572 }
1573
1574 /* Zero out the descriptor ring */
1575 memset(rx_ring->desc, 0, rx_ring->size);
1576
1577 rx_ring->next_to_clean = 0;
1578 rx_ring->next_to_use = 0;
1579 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1580
1581 writel(0, adapter->hw.hw_addr + rx_ring->head);
1582 writel(0, adapter->hw.hw_addr + rx_ring->tail);
1583 }
1584
1585 static void e1000e_downshift_workaround(struct work_struct *work)
1586 {
1587 struct e1000_adapter *adapter = container_of(work,
1588 struct e1000_adapter, downshift_task);
1589
1590 if (test_bit(__E1000_DOWN, &adapter->state))
1591 return;
1592
1593 e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1594 }
1595
1596 /**
1597 * e1000_intr_msi - Interrupt Handler
1598 * @irq: interrupt number
1599 * @data: pointer to a network interface device structure
1600 **/
1601 static irqreturn_t e1000_intr_msi(int irq, void *data)
1602 {
1603 struct net_device *netdev = data;
1604 struct e1000_adapter *adapter = netdev_priv(netdev);
1605 struct e1000_hw *hw = &adapter->hw;
1606 u32 icr = er32(ICR);
1607
1608 /*
1609 * read ICR disables interrupts using IAM
1610 */
1611
1612 if (icr & E1000_ICR_LSC) {
1613 hw->mac.get_link_status = 1;
1614 /*
1615 * ICH8 workaround-- Call gig speed drop workaround on cable
1616 * disconnect (LSC) before accessing any PHY registers
1617 */
1618 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1619 (!(er32(STATUS) & E1000_STATUS_LU)))
1620 schedule_work(&adapter->downshift_task);
1621
1622 /*
1623 * 80003ES2LAN workaround-- For packet buffer work-around on
1624 * link down event; disable receives here in the ISR and reset
1625 * adapter in watchdog
1626 */
1627 if (netif_carrier_ok(netdev) &&
1628 adapter->flags & FLAG_RX_NEEDS_RESTART) {
1629 /* disable receives */
1630 u32 rctl = er32(RCTL);
1631 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1632 adapter->flags |= FLAG_RX_RESTART_NOW;
1633 }
1634 /* guard against interrupt when we're going down */
1635 if (!test_bit(__E1000_DOWN, &adapter->state))
1636 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1637 }
1638
1639 if (napi_schedule_prep(&adapter->napi)) {
1640 adapter->total_tx_bytes = 0;
1641 adapter->total_tx_packets = 0;
1642 adapter->total_rx_bytes = 0;
1643 adapter->total_rx_packets = 0;
1644 __napi_schedule(&adapter->napi);
1645 }
1646
1647 return IRQ_HANDLED;
1648 }
1649
1650 /**
1651 * e1000_intr - Interrupt Handler
1652 * @irq: interrupt number
1653 * @data: pointer to a network interface device structure
1654 **/
1655 static irqreturn_t e1000_intr(int irq, void *data)
1656 {
1657 struct net_device *netdev = data;
1658 struct e1000_adapter *adapter = netdev_priv(netdev);
1659 struct e1000_hw *hw = &adapter->hw;
1660 u32 rctl, icr = er32(ICR);
1661
1662 if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1663 return IRQ_NONE; /* Not our interrupt */
1664
1665 /*
1666 * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1667 * not set, then the adapter didn't send an interrupt
1668 */
1669 if (!(icr & E1000_ICR_INT_ASSERTED))
1670 return IRQ_NONE;
1671
1672 /*
1673 * Interrupt Auto-Mask...upon reading ICR,
1674 * interrupts are masked. No need for the
1675 * IMC write
1676 */
1677
1678 if (icr & E1000_ICR_LSC) {
1679 hw->mac.get_link_status = 1;
1680 /*
1681 * ICH8 workaround-- Call gig speed drop workaround on cable
1682 * disconnect (LSC) before accessing any PHY registers
1683 */
1684 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1685 (!(er32(STATUS) & E1000_STATUS_LU)))
1686 schedule_work(&adapter->downshift_task);
1687
1688 /*
1689 * 80003ES2LAN workaround--
1690 * For packet buffer work-around on link down event;
1691 * disable receives here in the ISR and
1692 * reset adapter in watchdog
1693 */
1694 if (netif_carrier_ok(netdev) &&
1695 (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1696 /* disable receives */
1697 rctl = er32(RCTL);
1698 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1699 adapter->flags |= FLAG_RX_RESTART_NOW;
1700 }
1701 /* guard against interrupt when we're going down */
1702 if (!test_bit(__E1000_DOWN, &adapter->state))
1703 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1704 }
1705
1706 if (napi_schedule_prep(&adapter->napi)) {
1707 adapter->total_tx_bytes = 0;
1708 adapter->total_tx_packets = 0;
1709 adapter->total_rx_bytes = 0;
1710 adapter->total_rx_packets = 0;
1711 __napi_schedule(&adapter->napi);
1712 }
1713
1714 return IRQ_HANDLED;
1715 }
1716
1717 static irqreturn_t e1000_msix_other(int irq, void *data)
1718 {
1719 struct net_device *netdev = data;
1720 struct e1000_adapter *adapter = netdev_priv(netdev);
1721 struct e1000_hw *hw = &adapter->hw;
1722 u32 icr = er32(ICR);
1723
1724 if (!(icr & E1000_ICR_INT_ASSERTED)) {
1725 if (!test_bit(__E1000_DOWN, &adapter->state))
1726 ew32(IMS, E1000_IMS_OTHER);
1727 return IRQ_NONE;
1728 }
1729
1730 if (icr & adapter->eiac_mask)
1731 ew32(ICS, (icr & adapter->eiac_mask));
1732
1733 if (icr & E1000_ICR_OTHER) {
1734 if (!(icr & E1000_ICR_LSC))
1735 goto no_link_interrupt;
1736 hw->mac.get_link_status = 1;
1737 /* guard against interrupt when we're going down */
1738 if (!test_bit(__E1000_DOWN, &adapter->state))
1739 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1740 }
1741
1742 no_link_interrupt:
1743 if (!test_bit(__E1000_DOWN, &adapter->state))
1744 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1745
1746 return IRQ_HANDLED;
1747 }
1748
1749
1750 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1751 {
1752 struct net_device *netdev = data;
1753 struct e1000_adapter *adapter = netdev_priv(netdev);
1754 struct e1000_hw *hw = &adapter->hw;
1755 struct e1000_ring *tx_ring = adapter->tx_ring;
1756
1757
1758 adapter->total_tx_bytes = 0;
1759 adapter->total_tx_packets = 0;
1760
1761 if (!e1000_clean_tx_irq(adapter))
1762 /* Ring was not completely cleaned, so fire another interrupt */
1763 ew32(ICS, tx_ring->ims_val);
1764
1765 return IRQ_HANDLED;
1766 }
1767
1768 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1769 {
1770 struct net_device *netdev = data;
1771 struct e1000_adapter *adapter = netdev_priv(netdev);
1772
1773 /* Write the ITR value calculated at the end of the
1774 * previous interrupt.
1775 */
1776 if (adapter->rx_ring->set_itr) {
1777 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1778 adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1779 adapter->rx_ring->set_itr = 0;
1780 }
1781
1782 if (napi_schedule_prep(&adapter->napi)) {
1783 adapter->total_rx_bytes = 0;
1784 adapter->total_rx_packets = 0;
1785 __napi_schedule(&adapter->napi);
1786 }
1787 return IRQ_HANDLED;
1788 }
1789
1790 /**
1791 * e1000_configure_msix - Configure MSI-X hardware
1792 *
1793 * e1000_configure_msix sets up the hardware to properly
1794 * generate MSI-X interrupts.
1795 **/
1796 static void e1000_configure_msix(struct e1000_adapter *adapter)
1797 {
1798 struct e1000_hw *hw = &adapter->hw;
1799 struct e1000_ring *rx_ring = adapter->rx_ring;
1800 struct e1000_ring *tx_ring = adapter->tx_ring;
1801 int vector = 0;
1802 u32 ctrl_ext, ivar = 0;
1803
1804 adapter->eiac_mask = 0;
1805
1806 /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1807 if (hw->mac.type == e1000_82574) {
1808 u32 rfctl = er32(RFCTL);
1809 rfctl |= E1000_RFCTL_ACK_DIS;
1810 ew32(RFCTL, rfctl);
1811 }
1812
1813 #define E1000_IVAR_INT_ALLOC_VALID 0x8
1814 /* Configure Rx vector */
1815 rx_ring->ims_val = E1000_IMS_RXQ0;
1816 adapter->eiac_mask |= rx_ring->ims_val;
1817 if (rx_ring->itr_val)
1818 writel(1000000000 / (rx_ring->itr_val * 256),
1819 hw->hw_addr + rx_ring->itr_register);
1820 else
1821 writel(1, hw->hw_addr + rx_ring->itr_register);
1822 ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1823
1824 /* Configure Tx vector */
1825 tx_ring->ims_val = E1000_IMS_TXQ0;
1826 vector++;
1827 if (tx_ring->itr_val)
1828 writel(1000000000 / (tx_ring->itr_val * 256),
1829 hw->hw_addr + tx_ring->itr_register);
1830 else
1831 writel(1, hw->hw_addr + tx_ring->itr_register);
1832 adapter->eiac_mask |= tx_ring->ims_val;
1833 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1834
1835 /* set vector for Other Causes, e.g. link changes */
1836 vector++;
1837 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1838 if (rx_ring->itr_val)
1839 writel(1000000000 / (rx_ring->itr_val * 256),
1840 hw->hw_addr + E1000_EITR_82574(vector));
1841 else
1842 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1843
1844 /* Cause Tx interrupts on every write back */
1845 ivar |= (1 << 31);
1846
1847 ew32(IVAR, ivar);
1848
1849 /* enable MSI-X PBA support */
1850 ctrl_ext = er32(CTRL_EXT);
1851 ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1852
1853 /* Auto-Mask Other interrupts upon ICR read */
1854 #define E1000_EIAC_MASK_82574 0x01F00000
1855 ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1856 ctrl_ext |= E1000_CTRL_EXT_EIAME;
1857 ew32(CTRL_EXT, ctrl_ext);
1858 e1e_flush();
1859 }
1860
1861 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1862 {
1863 if (adapter->msix_entries) {
1864 pci_disable_msix(adapter->pdev);
1865 kfree(adapter->msix_entries);
1866 adapter->msix_entries = NULL;
1867 } else if (adapter->flags & FLAG_MSI_ENABLED) {
1868 pci_disable_msi(adapter->pdev);
1869 adapter->flags &= ~FLAG_MSI_ENABLED;
1870 }
1871 }
1872
1873 /**
1874 * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1875 *
1876 * Attempt to configure interrupts using the best available
1877 * capabilities of the hardware and kernel.
1878 **/
1879 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1880 {
1881 int err;
1882 int i;
1883
1884 switch (adapter->int_mode) {
1885 case E1000E_INT_MODE_MSIX:
1886 if (adapter->flags & FLAG_HAS_MSIX) {
1887 adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1888 adapter->msix_entries = kcalloc(adapter->num_vectors,
1889 sizeof(struct msix_entry),
1890 GFP_KERNEL);
1891 if (adapter->msix_entries) {
1892 for (i = 0; i < adapter->num_vectors; i++)
1893 adapter->msix_entries[i].entry = i;
1894
1895 err = pci_enable_msix(adapter->pdev,
1896 adapter->msix_entries,
1897 adapter->num_vectors);
1898 if (err == 0)
1899 return;
1900 }
1901 /* MSI-X failed, so fall through and try MSI */
1902 e_err("Failed to initialize MSI-X interrupts. Falling back to MSI interrupts.\n");
1903 e1000e_reset_interrupt_capability(adapter);
1904 }
1905 adapter->int_mode = E1000E_INT_MODE_MSI;
1906 /* Fall through */
1907 case E1000E_INT_MODE_MSI:
1908 if (!pci_enable_msi(adapter->pdev)) {
1909 adapter->flags |= FLAG_MSI_ENABLED;
1910 } else {
1911 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1912 e_err("Failed to initialize MSI interrupts. Falling back to legacy interrupts.\n");
1913 }
1914 /* Fall through */
1915 case E1000E_INT_MODE_LEGACY:
1916 /* Don't do anything; this is the system default */
1917 break;
1918 }
1919
1920 /* store the number of vectors being used */
1921 adapter->num_vectors = 1;
1922 }
1923
1924 /**
1925 * e1000_request_msix - Initialize MSI-X interrupts
1926 *
1927 * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1928 * kernel.
1929 **/
1930 static int e1000_request_msix(struct e1000_adapter *adapter)
1931 {
1932 struct net_device *netdev = adapter->netdev;
1933 int err = 0, vector = 0;
1934
1935 if (strlen(netdev->name) < (IFNAMSIZ - 5))
1936 snprintf(adapter->rx_ring->name,
1937 sizeof(adapter->rx_ring->name) - 1,
1938 "%s-rx-0", netdev->name);
1939 else
1940 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1941 err = request_irq(adapter->msix_entries[vector].vector,
1942 e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1943 netdev);
1944 if (err)
1945 goto out;
1946 adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1947 adapter->rx_ring->itr_val = adapter->itr;
1948 vector++;
1949
1950 if (strlen(netdev->name) < (IFNAMSIZ - 5))
1951 snprintf(adapter->tx_ring->name,
1952 sizeof(adapter->tx_ring->name) - 1,
1953 "%s-tx-0", netdev->name);
1954 else
1955 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1956 err = request_irq(adapter->msix_entries[vector].vector,
1957 e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1958 netdev);
1959 if (err)
1960 goto out;
1961 adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1962 adapter->tx_ring->itr_val = adapter->itr;
1963 vector++;
1964
1965 err = request_irq(adapter->msix_entries[vector].vector,
1966 e1000_msix_other, 0, netdev->name, netdev);
1967 if (err)
1968 goto out;
1969
1970 e1000_configure_msix(adapter);
1971 return 0;
1972 out:
1973 return err;
1974 }
1975
1976 /**
1977 * e1000_request_irq - initialize interrupts
1978 *
1979 * Attempts to configure interrupts using the best available
1980 * capabilities of the hardware and kernel.
1981 **/
1982 static int e1000_request_irq(struct e1000_adapter *adapter)
1983 {
1984 struct net_device *netdev = adapter->netdev;
1985 int err;
1986
1987 if (adapter->msix_entries) {
1988 err = e1000_request_msix(adapter);
1989 if (!err)
1990 return err;
1991 /* fall back to MSI */
1992 e1000e_reset_interrupt_capability(adapter);
1993 adapter->int_mode = E1000E_INT_MODE_MSI;
1994 e1000e_set_interrupt_capability(adapter);
1995 }
1996 if (adapter->flags & FLAG_MSI_ENABLED) {
1997 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1998 netdev->name, netdev);
1999 if (!err)
2000 return err;
2001
2002 /* fall back to legacy interrupt */
2003 e1000e_reset_interrupt_capability(adapter);
2004 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2005 }
2006
2007 err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2008 netdev->name, netdev);
2009 if (err)
2010 e_err("Unable to allocate interrupt, Error: %d\n", err);
2011
2012 return err;
2013 }
2014
2015 static void e1000_free_irq(struct e1000_adapter *adapter)
2016 {
2017 struct net_device *netdev = adapter->netdev;
2018
2019 if (adapter->msix_entries) {
2020 int vector = 0;
2021
2022 free_irq(adapter->msix_entries[vector].vector, netdev);
2023 vector++;
2024
2025 free_irq(adapter->msix_entries[vector].vector, netdev);
2026 vector++;
2027
2028 /* Other Causes interrupt vector */
2029 free_irq(adapter->msix_entries[vector].vector, netdev);
2030 return;
2031 }
2032
2033 free_irq(adapter->pdev->irq, netdev);
2034 }
2035
2036 /**
2037 * e1000_irq_disable - Mask off interrupt generation on the NIC
2038 **/
2039 static void e1000_irq_disable(struct e1000_adapter *adapter)
2040 {
2041 struct e1000_hw *hw = &adapter->hw;
2042
2043 ew32(IMC, ~0);
2044 if (adapter->msix_entries)
2045 ew32(EIAC_82574, 0);
2046 e1e_flush();
2047
2048 if (adapter->msix_entries) {
2049 int i;
2050 for (i = 0; i < adapter->num_vectors; i++)
2051 synchronize_irq(adapter->msix_entries[i].vector);
2052 } else {
2053 synchronize_irq(adapter->pdev->irq);
2054 }
2055 }
2056
2057 /**
2058 * e1000_irq_enable - Enable default interrupt generation settings
2059 **/
2060 static void e1000_irq_enable(struct e1000_adapter *adapter)
2061 {
2062 struct e1000_hw *hw = &adapter->hw;
2063
2064 if (adapter->msix_entries) {
2065 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2066 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
2067 } else {
2068 ew32(IMS, IMS_ENABLE_MASK);
2069 }
2070 e1e_flush();
2071 }
2072
2073 /**
2074 * e1000e_get_hw_control - get control of the h/w from f/w
2075 * @adapter: address of board private structure
2076 *
2077 * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2078 * For ASF and Pass Through versions of f/w this means that
2079 * the driver is loaded. For AMT version (only with 82573)
2080 * of the f/w this means that the network i/f is open.
2081 **/
2082 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2083 {
2084 struct e1000_hw *hw = &adapter->hw;
2085 u32 ctrl_ext;
2086 u32 swsm;
2087
2088 /* Let firmware know the driver has taken over */
2089 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2090 swsm = er32(SWSM);
2091 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2092 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2093 ctrl_ext = er32(CTRL_EXT);
2094 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2095 }
2096 }
2097
2098 /**
2099 * e1000e_release_hw_control - release control of the h/w to f/w
2100 * @adapter: address of board private structure
2101 *
2102 * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2103 * For ASF and Pass Through versions of f/w this means that the
2104 * driver is no longer loaded. For AMT version (only with 82573) i
2105 * of the f/w this means that the network i/f is closed.
2106 *
2107 **/
2108 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2109 {
2110 struct e1000_hw *hw = &adapter->hw;
2111 u32 ctrl_ext;
2112 u32 swsm;
2113
2114 /* Let firmware taken over control of h/w */
2115 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2116 swsm = er32(SWSM);
2117 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2118 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2119 ctrl_ext = er32(CTRL_EXT);
2120 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2121 }
2122 }
2123
2124 /**
2125 * @e1000_alloc_ring - allocate memory for a ring structure
2126 **/
2127 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2128 struct e1000_ring *ring)
2129 {
2130 struct pci_dev *pdev = adapter->pdev;
2131
2132 ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2133 GFP_KERNEL);
2134 if (!ring->desc)
2135 return -ENOMEM;
2136
2137 return 0;
2138 }
2139
2140 /**
2141 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2142 * @adapter: board private structure
2143 *
2144 * Return 0 on success, negative on failure
2145 **/
2146 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2147 {
2148 struct e1000_ring *tx_ring = adapter->tx_ring;
2149 int err = -ENOMEM, size;
2150
2151 size = sizeof(struct e1000_buffer) * tx_ring->count;
2152 tx_ring->buffer_info = vzalloc(size);
2153 if (!tx_ring->buffer_info)
2154 goto err;
2155
2156 /* round up to nearest 4K */
2157 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2158 tx_ring->size = ALIGN(tx_ring->size, 4096);
2159
2160 err = e1000_alloc_ring_dma(adapter, tx_ring);
2161 if (err)
2162 goto err;
2163
2164 tx_ring->next_to_use = 0;
2165 tx_ring->next_to_clean = 0;
2166
2167 return 0;
2168 err:
2169 vfree(tx_ring->buffer_info);
2170 e_err("Unable to allocate memory for the transmit descriptor ring\n");
2171 return err;
2172 }
2173
2174 /**
2175 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2176 * @adapter: board private structure
2177 *
2178 * Returns 0 on success, negative on failure
2179 **/
2180 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2181 {
2182 struct e1000_ring *rx_ring = adapter->rx_ring;
2183 struct e1000_buffer *buffer_info;
2184 int i, size, desc_len, err = -ENOMEM;
2185
2186 size = sizeof(struct e1000_buffer) * rx_ring->count;
2187 rx_ring->buffer_info = vzalloc(size);
2188 if (!rx_ring->buffer_info)
2189 goto err;
2190
2191 for (i = 0; i < rx_ring->count; i++) {
2192 buffer_info = &rx_ring->buffer_info[i];
2193 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2194 sizeof(struct e1000_ps_page),
2195 GFP_KERNEL);
2196 if (!buffer_info->ps_pages)
2197 goto err_pages;
2198 }
2199
2200 desc_len = sizeof(union e1000_rx_desc_packet_split);
2201
2202 /* Round up to nearest 4K */
2203 rx_ring->size = rx_ring->count * desc_len;
2204 rx_ring->size = ALIGN(rx_ring->size, 4096);
2205
2206 err = e1000_alloc_ring_dma(adapter, rx_ring);
2207 if (err)
2208 goto err_pages;
2209
2210 rx_ring->next_to_clean = 0;
2211 rx_ring->next_to_use = 0;
2212 rx_ring->rx_skb_top = NULL;
2213
2214 return 0;
2215
2216 err_pages:
2217 for (i = 0; i < rx_ring->count; i++) {
2218 buffer_info = &rx_ring->buffer_info[i];
2219 kfree(buffer_info->ps_pages);
2220 }
2221 err:
2222 vfree(rx_ring->buffer_info);
2223 e_err("Unable to allocate memory for the receive descriptor ring\n");
2224 return err;
2225 }
2226
2227 /**
2228 * e1000_clean_tx_ring - Free Tx Buffers
2229 * @adapter: board private structure
2230 **/
2231 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2232 {
2233 struct e1000_ring *tx_ring = adapter->tx_ring;
2234 struct e1000_buffer *buffer_info;
2235 unsigned long size;
2236 unsigned int i;
2237
2238 for (i = 0; i < tx_ring->count; i++) {
2239 buffer_info = &tx_ring->buffer_info[i];
2240 e1000_put_txbuf(adapter, buffer_info);
2241 }
2242
2243 size = sizeof(struct e1000_buffer) * tx_ring->count;
2244 memset(tx_ring->buffer_info, 0, size);
2245
2246 memset(tx_ring->desc, 0, tx_ring->size);
2247
2248 tx_ring->next_to_use = 0;
2249 tx_ring->next_to_clean = 0;
2250
2251 writel(0, adapter->hw.hw_addr + tx_ring->head);
2252 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2253 }
2254
2255 /**
2256 * e1000e_free_tx_resources - Free Tx Resources per Queue
2257 * @adapter: board private structure
2258 *
2259 * Free all transmit software resources
2260 **/
2261 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2262 {
2263 struct pci_dev *pdev = adapter->pdev;
2264 struct e1000_ring *tx_ring = adapter->tx_ring;
2265
2266 e1000_clean_tx_ring(adapter);
2267
2268 vfree(tx_ring->buffer_info);
2269 tx_ring->buffer_info = NULL;
2270
2271 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2272 tx_ring->dma);
2273 tx_ring->desc = NULL;
2274 }
2275
2276 /**
2277 * e1000e_free_rx_resources - Free Rx Resources
2278 * @adapter: board private structure
2279 *
2280 * Free all receive software resources
2281 **/
2282
2283 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2284 {
2285 struct pci_dev *pdev = adapter->pdev;
2286 struct e1000_ring *rx_ring = adapter->rx_ring;
2287 int i;
2288
2289 e1000_clean_rx_ring(adapter);
2290
2291 for (i = 0; i < rx_ring->count; i++)
2292 kfree(rx_ring->buffer_info[i].ps_pages);
2293
2294 vfree(rx_ring->buffer_info);
2295 rx_ring->buffer_info = NULL;
2296
2297 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2298 rx_ring->dma);
2299 rx_ring->desc = NULL;
2300 }
2301
2302 /**
2303 * e1000_update_itr - update the dynamic ITR value based on statistics
2304 * @adapter: pointer to adapter
2305 * @itr_setting: current adapter->itr
2306 * @packets: the number of packets during this measurement interval
2307 * @bytes: the number of bytes during this measurement interval
2308 *
2309 * Stores a new ITR value based on packets and byte
2310 * counts during the last interrupt. The advantage of per interrupt
2311 * computation is faster updates and more accurate ITR for the current
2312 * traffic pattern. Constants in this function were computed
2313 * based on theoretical maximum wire speed and thresholds were set based
2314 * on testing data as well as attempting to minimize response time
2315 * while increasing bulk throughput. This functionality is controlled
2316 * by the InterruptThrottleRate module parameter.
2317 **/
2318 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2319 u16 itr_setting, int packets,
2320 int bytes)
2321 {
2322 unsigned int retval = itr_setting;
2323
2324 if (packets == 0)
2325 goto update_itr_done;
2326
2327 switch (itr_setting) {
2328 case lowest_latency:
2329 /* handle TSO and jumbo frames */
2330 if (bytes/packets > 8000)
2331 retval = bulk_latency;
2332 else if ((packets < 5) && (bytes > 512))
2333 retval = low_latency;
2334 break;
2335 case low_latency: /* 50 usec aka 20000 ints/s */
2336 if (bytes > 10000) {
2337 /* this if handles the TSO accounting */
2338 if (bytes/packets > 8000)
2339 retval = bulk_latency;
2340 else if ((packets < 10) || ((bytes/packets) > 1200))
2341 retval = bulk_latency;
2342 else if ((packets > 35))
2343 retval = lowest_latency;
2344 } else if (bytes/packets > 2000) {
2345 retval = bulk_latency;
2346 } else if (packets <= 2 && bytes < 512) {
2347 retval = lowest_latency;
2348 }
2349 break;
2350 case bulk_latency: /* 250 usec aka 4000 ints/s */
2351 if (bytes > 25000) {
2352 if (packets > 35)
2353 retval = low_latency;
2354 } else if (bytes < 6000) {
2355 retval = low_latency;
2356 }
2357 break;
2358 }
2359
2360 update_itr_done:
2361 return retval;
2362 }
2363
2364 static void e1000_set_itr(struct e1000_adapter *adapter)
2365 {
2366 struct e1000_hw *hw = &adapter->hw;
2367 u16 current_itr;
2368 u32 new_itr = adapter->itr;
2369
2370 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2371 if (adapter->link_speed != SPEED_1000) {
2372 current_itr = 0;
2373 new_itr = 4000;
2374 goto set_itr_now;
2375 }
2376
2377 if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2378 new_itr = 0;
2379 goto set_itr_now;
2380 }
2381
2382 adapter->tx_itr = e1000_update_itr(adapter,
2383 adapter->tx_itr,
2384 adapter->total_tx_packets,
2385 adapter->total_tx_bytes);
2386 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2387 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2388 adapter->tx_itr = low_latency;
2389
2390 adapter->rx_itr = e1000_update_itr(adapter,
2391 adapter->rx_itr,
2392 adapter->total_rx_packets,
2393 adapter->total_rx_bytes);
2394 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2395 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2396 adapter->rx_itr = low_latency;
2397
2398 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2399
2400 switch (current_itr) {
2401 /* counts and packets in update_itr are dependent on these numbers */
2402 case lowest_latency:
2403 new_itr = 70000;
2404 break;
2405 case low_latency:
2406 new_itr = 20000; /* aka hwitr = ~200 */
2407 break;
2408 case bulk_latency:
2409 new_itr = 4000;
2410 break;
2411 default:
2412 break;
2413 }
2414
2415 set_itr_now:
2416 if (new_itr != adapter->itr) {
2417 /*
2418 * this attempts to bias the interrupt rate towards Bulk
2419 * by adding intermediate steps when interrupt rate is
2420 * increasing
2421 */
2422 new_itr = new_itr > adapter->itr ?
2423 min(adapter->itr + (new_itr >> 2), new_itr) :
2424 new_itr;
2425 adapter->itr = new_itr;
2426 adapter->rx_ring->itr_val = new_itr;
2427 if (adapter->msix_entries)
2428 adapter->rx_ring->set_itr = 1;
2429 else
2430 if (new_itr)
2431 ew32(ITR, 1000000000 / (new_itr * 256));
2432 else
2433 ew32(ITR, 0);
2434 }
2435 }
2436
2437 /**
2438 * e1000_alloc_queues - Allocate memory for all rings
2439 * @adapter: board private structure to initialize
2440 **/
2441 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2442 {
2443 adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2444 if (!adapter->tx_ring)
2445 goto err;
2446
2447 adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2448 if (!adapter->rx_ring)
2449 goto err;
2450
2451 return 0;
2452 err:
2453 e_err("Unable to allocate memory for queues\n");
2454 kfree(adapter->rx_ring);
2455 kfree(adapter->tx_ring);
2456 return -ENOMEM;
2457 }
2458
2459 /**
2460 * e1000_clean - NAPI Rx polling callback
2461 * @napi: struct associated with this polling callback
2462 * @budget: amount of packets driver is allowed to process this poll
2463 **/
2464 static int e1000_clean(struct napi_struct *napi, int budget)
2465 {
2466 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2467 struct e1000_hw *hw = &adapter->hw;
2468 struct net_device *poll_dev = adapter->netdev;
2469 int tx_cleaned = 1, work_done = 0;
2470
2471 adapter = netdev_priv(poll_dev);
2472
2473 if (adapter->msix_entries &&
2474 !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2475 goto clean_rx;
2476
2477 tx_cleaned = e1000_clean_tx_irq(adapter);
2478
2479 clean_rx:
2480 adapter->clean_rx(adapter, &work_done, budget);
2481
2482 if (!tx_cleaned)
2483 work_done = budget;
2484
2485 /* If budget not fully consumed, exit the polling mode */
2486 if (work_done < budget) {
2487 if (adapter->itr_setting & 3)
2488 e1000_set_itr(adapter);
2489 napi_complete(napi);
2490 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2491 if (adapter->msix_entries)
2492 ew32(IMS, adapter->rx_ring->ims_val);
2493 else
2494 e1000_irq_enable(adapter);
2495 }
2496 }
2497
2498 return work_done;
2499 }
2500
2501 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2502 {
2503 struct e1000_adapter *adapter = netdev_priv(netdev);
2504 struct e1000_hw *hw = &adapter->hw;
2505 u32 vfta, index;
2506
2507 /* don't update vlan cookie if already programmed */
2508 if ((adapter->hw.mng_cookie.status &
2509 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2510 (vid == adapter->mng_vlan_id))
2511 return;
2512
2513 /* add VID to filter table */
2514 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2515 index = (vid >> 5) & 0x7F;
2516 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2517 vfta |= (1 << (vid & 0x1F));
2518 hw->mac.ops.write_vfta(hw, index, vfta);
2519 }
2520
2521 set_bit(vid, adapter->active_vlans);
2522 }
2523
2524 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2525 {
2526 struct e1000_adapter *adapter = netdev_priv(netdev);
2527 struct e1000_hw *hw = &adapter->hw;
2528 u32 vfta, index;
2529
2530 if ((adapter->hw.mng_cookie.status &
2531 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2532 (vid == adapter->mng_vlan_id)) {
2533 /* release control to f/w */
2534 e1000e_release_hw_control(adapter);
2535 return;
2536 }
2537
2538 /* remove VID from filter table */
2539 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2540 index = (vid >> 5) & 0x7F;
2541 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2542 vfta &= ~(1 << (vid & 0x1F));
2543 hw->mac.ops.write_vfta(hw, index, vfta);
2544 }
2545
2546 clear_bit(vid, adapter->active_vlans);
2547 }
2548
2549 /**
2550 * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2551 * @adapter: board private structure to initialize
2552 **/
2553 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2554 {
2555 struct net_device *netdev = adapter->netdev;
2556 struct e1000_hw *hw = &adapter->hw;
2557 u32 rctl;
2558
2559 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2560 /* disable VLAN receive filtering */
2561 rctl = er32(RCTL);
2562 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2563 ew32(RCTL, rctl);
2564
2565 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2566 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2567 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2568 }
2569 }
2570 }
2571
2572 /**
2573 * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2574 * @adapter: board private structure to initialize
2575 **/
2576 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2577 {
2578 struct e1000_hw *hw = &adapter->hw;
2579 u32 rctl;
2580
2581 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2582 /* enable VLAN receive filtering */
2583 rctl = er32(RCTL);
2584 rctl |= E1000_RCTL_VFE;
2585 rctl &= ~E1000_RCTL_CFIEN;
2586 ew32(RCTL, rctl);
2587 }
2588 }
2589
2590 /**
2591 * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2592 * @adapter: board private structure to initialize
2593 **/
2594 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2595 {
2596 struct e1000_hw *hw = &adapter->hw;
2597 u32 ctrl;
2598
2599 /* disable VLAN tag insert/strip */
2600 ctrl = er32(CTRL);
2601 ctrl &= ~E1000_CTRL_VME;
2602 ew32(CTRL, ctrl);
2603 }
2604
2605 /**
2606 * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2607 * @adapter: board private structure to initialize
2608 **/
2609 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2610 {
2611 struct e1000_hw *hw = &adapter->hw;
2612 u32 ctrl;
2613
2614 /* enable VLAN tag insert/strip */
2615 ctrl = er32(CTRL);
2616 ctrl |= E1000_CTRL_VME;
2617 ew32(CTRL, ctrl);
2618 }
2619
2620 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2621 {
2622 struct net_device *netdev = adapter->netdev;
2623 u16 vid = adapter->hw.mng_cookie.vlan_id;
2624 u16 old_vid = adapter->mng_vlan_id;
2625
2626 if (adapter->hw.mng_cookie.status &
2627 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2628 e1000_vlan_rx_add_vid(netdev, vid);
2629 adapter->mng_vlan_id = vid;
2630 }
2631
2632 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2633 e1000_vlan_rx_kill_vid(netdev, old_vid);
2634 }
2635
2636 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2637 {
2638 u16 vid;
2639
2640 e1000_vlan_rx_add_vid(adapter->netdev, 0);
2641
2642 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2643 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2644 }
2645
2646 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2647 {
2648 struct e1000_hw *hw = &adapter->hw;
2649 u32 manc, manc2h, mdef, i, j;
2650
2651 if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2652 return;
2653
2654 manc = er32(MANC);
2655
2656 /*
2657 * enable receiving management packets to the host. this will probably
2658 * generate destination unreachable messages from the host OS, but
2659 * the packets will be handled on SMBUS
2660 */
2661 manc |= E1000_MANC_EN_MNG2HOST;
2662 manc2h = er32(MANC2H);
2663
2664 switch (hw->mac.type) {
2665 default:
2666 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2667 break;
2668 case e1000_82574:
2669 case e1000_82583:
2670 /*
2671 * Check if IPMI pass-through decision filter already exists;
2672 * if so, enable it.
2673 */
2674 for (i = 0, j = 0; i < 8; i++) {
2675 mdef = er32(MDEF(i));
2676
2677 /* Ignore filters with anything other than IPMI ports */
2678 if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2679 continue;
2680
2681 /* Enable this decision filter in MANC2H */
2682 if (mdef)
2683 manc2h |= (1 << i);
2684
2685 j |= mdef;
2686 }
2687
2688 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2689 break;
2690
2691 /* Create new decision filter in an empty filter */
2692 for (i = 0, j = 0; i < 8; i++)
2693 if (er32(MDEF(i)) == 0) {
2694 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2695 E1000_MDEF_PORT_664));
2696 manc2h |= (1 << 1);
2697 j++;
2698 break;
2699 }
2700
2701 if (!j)
2702 e_warn("Unable to create IPMI pass-through filter\n");
2703 break;
2704 }
2705
2706 ew32(MANC2H, manc2h);
2707 ew32(MANC, manc);
2708 }
2709
2710 /**
2711 * e1000_configure_tx - Configure Transmit Unit after Reset
2712 * @adapter: board private structure
2713 *
2714 * Configure the Tx unit of the MAC after a reset.
2715 **/
2716 static void e1000_configure_tx(struct e1000_adapter *adapter)
2717 {
2718 struct e1000_hw *hw = &adapter->hw;
2719 struct e1000_ring *tx_ring = adapter->tx_ring;
2720 u64 tdba;
2721 u32 tdlen, tctl, tipg, tarc;
2722 u32 ipgr1, ipgr2;
2723
2724 /* Setup the HW Tx Head and Tail descriptor pointers */
2725 tdba = tx_ring->dma;
2726 tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2727 ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2728 ew32(TDBAH, (tdba >> 32));
2729 ew32(TDLEN, tdlen);
2730 ew32(TDH, 0);
2731 ew32(TDT, 0);
2732 tx_ring->head = E1000_TDH;
2733 tx_ring->tail = E1000_TDT;
2734
2735 /* Set the default values for the Tx Inter Packet Gap timer */
2736 tipg = DEFAULT_82543_TIPG_IPGT_COPPER; /* 8 */
2737 ipgr1 = DEFAULT_82543_TIPG_IPGR1; /* 8 */
2738 ipgr2 = DEFAULT_82543_TIPG_IPGR2; /* 6 */
2739
2740 if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2741 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /* 7 */
2742
2743 tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2744 tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2745 ew32(TIPG, tipg);
2746
2747 /* Set the Tx Interrupt Delay register */
2748 ew32(TIDV, adapter->tx_int_delay);
2749 /* Tx irq moderation */
2750 ew32(TADV, adapter->tx_abs_int_delay);
2751
2752 if (adapter->flags2 & FLAG2_DMA_BURST) {
2753 u32 txdctl = er32(TXDCTL(0));
2754 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2755 E1000_TXDCTL_WTHRESH);
2756 /*
2757 * set up some performance related parameters to encourage the
2758 * hardware to use the bus more efficiently in bursts, depends
2759 * on the tx_int_delay to be enabled,
2760 * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2761 * hthresh = 1 ==> prefetch when one or more available
2762 * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2763 * BEWARE: this seems to work but should be considered first if
2764 * there are Tx hangs or other Tx related bugs
2765 */
2766 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2767 ew32(TXDCTL(0), txdctl);
2768 /* erratum work around: set txdctl the same for both queues */
2769 ew32(TXDCTL(1), txdctl);
2770 }
2771
2772 /* Program the Transmit Control Register */
2773 tctl = er32(TCTL);
2774 tctl &= ~E1000_TCTL_CT;
2775 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2776 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2777
2778 if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2779 tarc = er32(TARC(0));
2780 /*
2781 * set the speed mode bit, we'll clear it if we're not at
2782 * gigabit link later
2783 */
2784 #define SPEED_MODE_BIT (1 << 21)
2785 tarc |= SPEED_MODE_BIT;
2786 ew32(TARC(0), tarc);
2787 }
2788
2789 /* errata: program both queues to unweighted RR */
2790 if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2791 tarc = er32(TARC(0));
2792 tarc |= 1;
2793 ew32(TARC(0), tarc);
2794 tarc = er32(TARC(1));
2795 tarc |= 1;
2796 ew32(TARC(1), tarc);
2797 }
2798
2799 /* Setup Transmit Descriptor Settings for eop descriptor */
2800 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2801
2802 /* only set IDE if we are delaying interrupts using the timers */
2803 if (adapter->tx_int_delay)
2804 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2805
2806 /* enable Report Status bit */
2807 adapter->txd_cmd |= E1000_TXD_CMD_RS;
2808
2809 ew32(TCTL, tctl);
2810
2811 e1000e_config_collision_dist(hw);
2812 }
2813
2814 /**
2815 * e1000_setup_rctl - configure the receive control registers
2816 * @adapter: Board private structure
2817 **/
2818 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2819 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2820 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2821 {
2822 struct e1000_hw *hw = &adapter->hw;
2823 u32 rctl, rfctl;
2824 u32 pages = 0;
2825
2826 /* Workaround Si errata on 82579 - configure jumbo frame flow */
2827 if (hw->mac.type == e1000_pch2lan) {
2828 s32 ret_val;
2829
2830 if (adapter->netdev->mtu > ETH_DATA_LEN)
2831 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2832 else
2833 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2834
2835 if (ret_val)
2836 e_dbg("failed to enable jumbo frame workaround mode\n");
2837 }
2838
2839 /* Program MC offset vector base */
2840 rctl = er32(RCTL);
2841 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2842 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2843 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2844 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2845
2846 /* Do not Store bad packets */
2847 rctl &= ~E1000_RCTL_SBP;
2848
2849 /* Enable Long Packet receive */
2850 if (adapter->netdev->mtu <= ETH_DATA_LEN)
2851 rctl &= ~E1000_RCTL_LPE;
2852 else
2853 rctl |= E1000_RCTL_LPE;
2854
2855 /* Some systems expect that the CRC is included in SMBUS traffic. The
2856 * hardware strips the CRC before sending to both SMBUS (BMC) and to
2857 * host memory when this is enabled
2858 */
2859 if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2860 rctl |= E1000_RCTL_SECRC;
2861
2862 /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2863 if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2864 u16 phy_data;
2865
2866 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2867 phy_data &= 0xfff8;
2868 phy_data |= (1 << 2);
2869 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2870
2871 e1e_rphy(hw, 22, &phy_data);
2872 phy_data &= 0x0fff;
2873 phy_data |= (1 << 14);
2874 e1e_wphy(hw, 0x10, 0x2823);
2875 e1e_wphy(hw, 0x11, 0x0003);
2876 e1e_wphy(hw, 22, phy_data);
2877 }
2878
2879 /* Setup buffer sizes */
2880 rctl &= ~E1000_RCTL_SZ_4096;
2881 rctl |= E1000_RCTL_BSEX;
2882 switch (adapter->rx_buffer_len) {
2883 case 2048:
2884 default:
2885 rctl |= E1000_RCTL_SZ_2048;
2886 rctl &= ~E1000_RCTL_BSEX;
2887 break;
2888 case 4096:
2889 rctl |= E1000_RCTL_SZ_4096;
2890 break;
2891 case 8192:
2892 rctl |= E1000_RCTL_SZ_8192;
2893 break;
2894 case 16384:
2895 rctl |= E1000_RCTL_SZ_16384;
2896 break;
2897 }
2898
2899 /* Enable Extended Status in all Receive Descriptors */
2900 rfctl = er32(RFCTL);
2901 rfctl |= E1000_RFCTL_EXTEN;
2902
2903 /*
2904 * 82571 and greater support packet-split where the protocol
2905 * header is placed in skb->data and the packet data is
2906 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2907 * In the case of a non-split, skb->data is linearly filled,
2908 * followed by the page buffers. Therefore, skb->data is
2909 * sized to hold the largest protocol header.
2910 *
2911 * allocations using alloc_page take too long for regular MTU
2912 * so only enable packet split for jumbo frames
2913 *
2914 * Using pages when the page size is greater than 16k wastes
2915 * a lot of memory, since we allocate 3 pages at all times
2916 * per packet.
2917 */
2918 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2919 if (!(adapter->flags & FLAG_HAS_ERT) && (pages <= 3) &&
2920 (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2921 adapter->rx_ps_pages = pages;
2922 else
2923 adapter->rx_ps_pages = 0;
2924
2925 if (adapter->rx_ps_pages) {
2926 u32 psrctl = 0;
2927
2928 /*
2929 * disable packet split support for IPv6 extension headers,
2930 * because some malformed IPv6 headers can hang the Rx
2931 */
2932 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2933 E1000_RFCTL_NEW_IPV6_EXT_DIS);
2934
2935 /* Enable Packet split descriptors */
2936 rctl |= E1000_RCTL_DTYP_PS;
2937
2938 psrctl |= adapter->rx_ps_bsize0 >>
2939 E1000_PSRCTL_BSIZE0_SHIFT;
2940
2941 switch (adapter->rx_ps_pages) {
2942 case 3:
2943 psrctl |= PAGE_SIZE <<
2944 E1000_PSRCTL_BSIZE3_SHIFT;
2945 case 2:
2946 psrctl |= PAGE_SIZE <<
2947 E1000_PSRCTL_BSIZE2_SHIFT;
2948 case 1:
2949 psrctl |= PAGE_SIZE >>
2950 E1000_PSRCTL_BSIZE1_SHIFT;
2951 break;
2952 }
2953
2954 ew32(PSRCTL, psrctl);
2955 }
2956
2957 ew32(RFCTL, rfctl);
2958 ew32(RCTL, rctl);
2959 /* just started the receive unit, no need to restart */
2960 adapter->flags &= ~FLAG_RX_RESTART_NOW;
2961 }
2962
2963 /**
2964 * e1000_configure_rx - Configure Receive Unit after Reset
2965 * @adapter: board private structure
2966 *
2967 * Configure the Rx unit of the MAC after a reset.
2968 **/
2969 static void e1000_configure_rx(struct e1000_adapter *adapter)
2970 {
2971 struct e1000_hw *hw = &adapter->hw;
2972 struct e1000_ring *rx_ring = adapter->rx_ring;
2973 u64 rdba;
2974 u32 rdlen, rctl, rxcsum, ctrl_ext;
2975
2976 if (adapter->rx_ps_pages) {
2977 /* this is a 32 byte descriptor */
2978 rdlen = rx_ring->count *
2979 sizeof(union e1000_rx_desc_packet_split);
2980 adapter->clean_rx = e1000_clean_rx_irq_ps;
2981 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2982 } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2983 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
2984 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2985 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2986 } else {
2987 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
2988 adapter->clean_rx = e1000_clean_rx_irq;
2989 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2990 }
2991
2992 /* disable receives while setting up the descriptors */
2993 rctl = er32(RCTL);
2994 if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
2995 ew32(RCTL, rctl & ~E1000_RCTL_EN);
2996 e1e_flush();
2997 usleep_range(10000, 20000);
2998
2999 if (adapter->flags2 & FLAG2_DMA_BURST) {
3000 /*
3001 * set the writeback threshold (only takes effect if the RDTR
3002 * is set). set GRAN=1 and write back up to 0x4 worth, and
3003 * enable prefetching of 0x20 Rx descriptors
3004 * granularity = 01
3005 * wthresh = 04,
3006 * hthresh = 04,
3007 * pthresh = 0x20
3008 */
3009 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3010 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3011
3012 /*
3013 * override the delay timers for enabling bursting, only if
3014 * the value was not set by the user via module options
3015 */
3016 if (adapter->rx_int_delay == DEFAULT_RDTR)
3017 adapter->rx_int_delay = BURST_RDTR;
3018 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
3019 adapter->rx_abs_int_delay = BURST_RADV;
3020 }
3021
3022 /* set the Receive Delay Timer Register */
3023 ew32(RDTR, adapter->rx_int_delay);
3024
3025 /* irq moderation */
3026 ew32(RADV, adapter->rx_abs_int_delay);
3027 if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3028 ew32(ITR, 1000000000 / (adapter->itr * 256));
3029
3030 ctrl_ext = er32(CTRL_EXT);
3031 /* Auto-Mask interrupts upon ICR access */
3032 ctrl_ext |= E1000_CTRL_EXT_IAME;
3033 ew32(IAM, 0xffffffff);
3034 ew32(CTRL_EXT, ctrl_ext);
3035 e1e_flush();
3036
3037 /*
3038 * Setup the HW Rx Head and Tail Descriptor Pointers and
3039 * the Base and Length of the Rx Descriptor Ring
3040 */
3041 rdba = rx_ring->dma;
3042 ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
3043 ew32(RDBAH, (rdba >> 32));
3044 ew32(RDLEN, rdlen);
3045 ew32(RDH, 0);
3046 ew32(RDT, 0);
3047 rx_ring->head = E1000_RDH;
3048 rx_ring->tail = E1000_RDT;
3049
3050 /* Enable Receive Checksum Offload for TCP and UDP */
3051 rxcsum = er32(RXCSUM);
3052 if (adapter->netdev->features & NETIF_F_RXCSUM) {
3053 rxcsum |= E1000_RXCSUM_TUOFL;
3054
3055 /*
3056 * IPv4 payload checksum for UDP fragments must be
3057 * used in conjunction with packet-split.
3058 */
3059 if (adapter->rx_ps_pages)
3060 rxcsum |= E1000_RXCSUM_IPPCSE;
3061 } else {
3062 rxcsum &= ~E1000_RXCSUM_TUOFL;
3063 /* no need to clear IPPCSE as it defaults to 0 */
3064 }
3065 ew32(RXCSUM, rxcsum);
3066
3067 /*
3068 * Enable early receives on supported devices, only takes effect when
3069 * packet size is equal or larger than the specified value (in 8 byte
3070 * units), e.g. using jumbo frames when setting to E1000_ERT_2048
3071 */
3072 if ((adapter->flags & FLAG_HAS_ERT) ||
3073 (adapter->hw.mac.type == e1000_pch2lan)) {
3074 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3075 u32 rxdctl = er32(RXDCTL(0));
3076 ew32(RXDCTL(0), rxdctl | 0x3);
3077 if (adapter->flags & FLAG_HAS_ERT)
3078 ew32(ERT, E1000_ERT_2048 | (1 << 13));
3079 /*
3080 * With jumbo frames and early-receive enabled,
3081 * excessive C-state transition latencies result in
3082 * dropped transactions.
3083 */
3084 pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
3085 } else {
3086 pm_qos_update_request(&adapter->netdev->pm_qos_req,
3087 PM_QOS_DEFAULT_VALUE);
3088 }
3089 }
3090
3091 /* Enable Receives */
3092 ew32(RCTL, rctl);
3093 }
3094
3095 /**
3096 * e1000e_write_mc_addr_list - write multicast addresses to MTA
3097 * @netdev: network interface device structure
3098 *
3099 * Writes multicast address list to the MTA hash table.
3100 * Returns: -ENOMEM on failure
3101 * 0 on no addresses written
3102 * X on writing X addresses to MTA
3103 */
3104 static int e1000e_write_mc_addr_list(struct net_device *netdev)
3105 {
3106 struct e1000_adapter *adapter = netdev_priv(netdev);
3107 struct e1000_hw *hw = &adapter->hw;
3108 struct netdev_hw_addr *ha;
3109 u8 *mta_list;
3110 int i;
3111
3112 if (netdev_mc_empty(netdev)) {
3113 /* nothing to program, so clear mc list */
3114 hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3115 return 0;
3116 }
3117
3118 mta_list = kzalloc(netdev_mc_count(netdev) * ETH_ALEN, GFP_ATOMIC);
3119 if (!mta_list)
3120 return -ENOMEM;
3121
3122 /* update_mc_addr_list expects a packed array of only addresses. */
3123 i = 0;
3124 netdev_for_each_mc_addr(ha, netdev)
3125 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3126
3127 hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3128 kfree(mta_list);
3129
3130 return netdev_mc_count(netdev);
3131 }
3132
3133 /**
3134 * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3135 * @netdev: network interface device structure
3136 *
3137 * Writes unicast address list to the RAR table.
3138 * Returns: -ENOMEM on failure/insufficient address space
3139 * 0 on no addresses written
3140 * X on writing X addresses to the RAR table
3141 **/
3142 static int e1000e_write_uc_addr_list(struct net_device *netdev)
3143 {
3144 struct e1000_adapter *adapter = netdev_priv(netdev);
3145 struct e1000_hw *hw = &adapter->hw;
3146 unsigned int rar_entries = hw->mac.rar_entry_count;
3147 int count = 0;
3148
3149 /* save a rar entry for our hardware address */
3150 rar_entries--;
3151
3152 /* save a rar entry for the LAA workaround */
3153 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3154 rar_entries--;
3155
3156 /* return ENOMEM indicating insufficient memory for addresses */
3157 if (netdev_uc_count(netdev) > rar_entries)
3158 return -ENOMEM;
3159
3160 if (!netdev_uc_empty(netdev) && rar_entries) {
3161 struct netdev_hw_addr *ha;
3162
3163 /*
3164 * write the addresses in reverse order to avoid write
3165 * combining
3166 */
3167 netdev_for_each_uc_addr(ha, netdev) {
3168 if (!rar_entries)
3169 break;
3170 e1000e_rar_set(hw, ha->addr, rar_entries--);
3171 count++;
3172 }
3173 }
3174
3175 /* zero out the remaining RAR entries not used above */
3176 for (; rar_entries > 0; rar_entries--) {
3177 ew32(RAH(rar_entries), 0);
3178 ew32(RAL(rar_entries), 0);
3179 }
3180 e1e_flush();
3181
3182 return count;
3183 }
3184
3185 /**
3186 * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3187 * @netdev: network interface device structure
3188 *
3189 * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3190 * address list or the network interface flags are updated. This routine is
3191 * responsible for configuring the hardware for proper unicast, multicast,
3192 * promiscuous mode, and all-multi behavior.
3193 **/
3194 static void e1000e_set_rx_mode(struct net_device *netdev)
3195 {
3196 struct e1000_adapter *adapter = netdev_priv(netdev);
3197 struct e1000_hw *hw = &adapter->hw;
3198 u32 rctl;
3199
3200 /* Check for Promiscuous and All Multicast modes */
3201 rctl = er32(RCTL);
3202
3203 /* clear the affected bits */
3204 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3205
3206 if (netdev->flags & IFF_PROMISC) {
3207 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3208 /* Do not hardware filter VLANs in promisc mode */
3209 e1000e_vlan_filter_disable(adapter);
3210 } else {
3211 int count;
3212 if (netdev->flags & IFF_ALLMULTI) {
3213 rctl |= E1000_RCTL_MPE;
3214 } else {
3215 /*
3216 * Write addresses to the MTA, if the attempt fails
3217 * then we should just turn on promiscuous mode so
3218 * that we can at least receive multicast traffic
3219 */
3220 count = e1000e_write_mc_addr_list(netdev);
3221 if (count < 0)
3222 rctl |= E1000_RCTL_MPE;
3223 }
3224 e1000e_vlan_filter_enable(adapter);
3225 /*
3226 * Write addresses to available RAR registers, if there is not
3227 * sufficient space to store all the addresses then enable
3228 * unicast promiscuous mode
3229 */
3230 count = e1000e_write_uc_addr_list(netdev);
3231 if (count < 0)
3232 rctl |= E1000_RCTL_UPE;
3233 }
3234
3235 ew32(RCTL, rctl);
3236
3237 if (netdev->features & NETIF_F_HW_VLAN_RX)
3238 e1000e_vlan_strip_enable(adapter);
3239 else
3240 e1000e_vlan_strip_disable(adapter);
3241 }
3242
3243 /**
3244 * e1000_configure - configure the hardware for Rx and Tx
3245 * @adapter: private board structure
3246 **/
3247 static void e1000_configure(struct e1000_adapter *adapter)
3248 {
3249 e1000e_set_rx_mode(adapter->netdev);
3250
3251 e1000_restore_vlan(adapter);
3252 e1000_init_manageability_pt(adapter);
3253
3254 e1000_configure_tx(adapter);
3255 e1000_setup_rctl(adapter);
3256 e1000_configure_rx(adapter);
3257 adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring),
3258 GFP_KERNEL);
3259 }
3260
3261 /**
3262 * e1000e_power_up_phy - restore link in case the phy was powered down
3263 * @adapter: address of board private structure
3264 *
3265 * The phy may be powered down to save power and turn off link when the
3266 * driver is unloaded and wake on lan is not enabled (among others)
3267 * *** this routine MUST be followed by a call to e1000e_reset ***
3268 **/
3269 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3270 {
3271 if (adapter->hw.phy.ops.power_up)
3272 adapter->hw.phy.ops.power_up(&adapter->hw);
3273
3274 adapter->hw.mac.ops.setup_link(&adapter->hw);
3275 }
3276
3277 /**
3278 * e1000_power_down_phy - Power down the PHY
3279 *
3280 * Power down the PHY so no link is implied when interface is down.
3281 * The PHY cannot be powered down if management or WoL is active.
3282 */
3283 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3284 {
3285 /* WoL is enabled */
3286 if (adapter->wol)
3287 return;
3288
3289 if (adapter->hw.phy.ops.power_down)
3290 adapter->hw.phy.ops.power_down(&adapter->hw);
3291 }
3292
3293 /**
3294 * e1000e_reset - bring the hardware into a known good state
3295 *
3296 * This function boots the hardware and enables some settings that
3297 * require a configuration cycle of the hardware - those cannot be
3298 * set/changed during runtime. After reset the device needs to be
3299 * properly configured for Rx, Tx etc.
3300 */
3301 void e1000e_reset(struct e1000_adapter *adapter)
3302 {
3303 struct e1000_mac_info *mac = &adapter->hw.mac;
3304 struct e1000_fc_info *fc = &adapter->hw.fc;
3305 struct e1000_hw *hw = &adapter->hw;
3306 u32 tx_space, min_tx_space, min_rx_space;
3307 u32 pba = adapter->pba;
3308 u16 hwm;
3309
3310 /* reset Packet Buffer Allocation to default */
3311 ew32(PBA, pba);
3312
3313 if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3314 /*
3315 * To maintain wire speed transmits, the Tx FIFO should be
3316 * large enough to accommodate two full transmit packets,
3317 * rounded up to the next 1KB and expressed in KB. Likewise,
3318 * the Rx FIFO should be large enough to accommodate at least
3319 * one full receive packet and is similarly rounded up and
3320 * expressed in KB.
3321 */
3322 pba = er32(PBA);
3323 /* upper 16 bits has Tx packet buffer allocation size in KB */
3324 tx_space = pba >> 16;
3325 /* lower 16 bits has Rx packet buffer allocation size in KB */
3326 pba &= 0xffff;
3327 /*
3328 * the Tx fifo also stores 16 bytes of information about the Tx
3329 * but don't include ethernet FCS because hardware appends it
3330 */
3331 min_tx_space = (adapter->max_frame_size +
3332 sizeof(struct e1000_tx_desc) -
3333 ETH_FCS_LEN) * 2;
3334 min_tx_space = ALIGN(min_tx_space, 1024);
3335 min_tx_space >>= 10;
3336 /* software strips receive CRC, so leave room for it */
3337 min_rx_space = adapter->max_frame_size;
3338 min_rx_space = ALIGN(min_rx_space, 1024);
3339 min_rx_space >>= 10;
3340
3341 /*
3342 * If current Tx allocation is less than the min Tx FIFO size,
3343 * and the min Tx FIFO size is less than the current Rx FIFO
3344 * allocation, take space away from current Rx allocation
3345 */
3346 if ((tx_space < min_tx_space) &&
3347 ((min_tx_space - tx_space) < pba)) {
3348 pba -= min_tx_space - tx_space;
3349
3350 /*
3351 * if short on Rx space, Rx wins and must trump Tx
3352 * adjustment or use Early Receive if available
3353 */
3354 if ((pba < min_rx_space) &&
3355 (!(adapter->flags & FLAG_HAS_ERT)))
3356 /* ERT enabled in e1000_configure_rx */
3357 pba = min_rx_space;
3358 }
3359
3360 ew32(PBA, pba);
3361 }
3362
3363 /*
3364 * flow control settings
3365 *
3366 * The high water mark must be low enough to fit one full frame
3367 * (or the size used for early receive) above it in the Rx FIFO.
3368 * Set it to the lower of:
3369 * - 90% of the Rx FIFO size, and
3370 * - the full Rx FIFO size minus the early receive size (for parts
3371 * with ERT support assuming ERT set to E1000_ERT_2048), or
3372 * - the full Rx FIFO size minus one full frame
3373 */
3374 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3375 fc->pause_time = 0xFFFF;
3376 else
3377 fc->pause_time = E1000_FC_PAUSE_TIME;
3378 fc->send_xon = 1;
3379 fc->current_mode = fc->requested_mode;
3380
3381 switch (hw->mac.type) {
3382 default:
3383 if ((adapter->flags & FLAG_HAS_ERT) &&
3384 (adapter->netdev->mtu > ETH_DATA_LEN))
3385 hwm = min(((pba << 10) * 9 / 10),
3386 ((pba << 10) - (E1000_ERT_2048 << 3)));
3387 else
3388 hwm = min(((pba << 10) * 9 / 10),
3389 ((pba << 10) - adapter->max_frame_size));
3390
3391 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3392 fc->low_water = fc->high_water - 8;
3393 break;
3394 case e1000_pchlan:
3395 /*
3396 * Workaround PCH LOM adapter hangs with certain network
3397 * loads. If hangs persist, try disabling Tx flow control.
3398 */
3399 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3400 fc->high_water = 0x3500;
3401 fc->low_water = 0x1500;
3402 } else {
3403 fc->high_water = 0x5000;
3404 fc->low_water = 0x3000;
3405 }
3406 fc->refresh_time = 0x1000;
3407 break;
3408 case e1000_pch2lan:
3409 fc->high_water = 0x05C20;
3410 fc->low_water = 0x05048;
3411 fc->pause_time = 0x0650;
3412 fc->refresh_time = 0x0400;
3413 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3414 pba = 14;
3415 ew32(PBA, pba);
3416 }
3417 break;
3418 }
3419
3420 /*
3421 * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3422 * fit in receive buffer and early-receive not supported.
3423 */
3424 if (adapter->itr_setting & 0x3) {
3425 if (((adapter->max_frame_size * 2) > (pba << 10)) &&
3426 !(adapter->flags & FLAG_HAS_ERT)) {
3427 if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3428 dev_info(&adapter->pdev->dev,
3429 "Interrupt Throttle Rate turned off\n");
3430 adapter->flags2 |= FLAG2_DISABLE_AIM;
3431 ew32(ITR, 0);
3432 }
3433 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3434 dev_info(&adapter->pdev->dev,
3435 "Interrupt Throttle Rate turned on\n");
3436 adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3437 adapter->itr = 20000;
3438 ew32(ITR, 1000000000 / (adapter->itr * 256));
3439 }
3440 }
3441
3442 /* Allow time for pending master requests to run */
3443 mac->ops.reset_hw(hw);
3444
3445 /*
3446 * For parts with AMT enabled, let the firmware know
3447 * that the network interface is in control
3448 */
3449 if (adapter->flags & FLAG_HAS_AMT)
3450 e1000e_get_hw_control(adapter);
3451
3452 ew32(WUC, 0);
3453
3454 if (mac->ops.init_hw(hw))
3455 e_err("Hardware Error\n");
3456
3457 e1000_update_mng_vlan(adapter);
3458
3459 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3460 ew32(VET, ETH_P_8021Q);
3461
3462 e1000e_reset_adaptive(hw);
3463
3464 if (!netif_running(adapter->netdev) &&
3465 !test_bit(__E1000_TESTING, &adapter->state)) {
3466 e1000_power_down_phy(adapter);
3467 return;
3468 }
3469
3470 e1000_get_phy_info(hw);
3471
3472 if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3473 !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3474 u16 phy_data = 0;
3475 /*
3476 * speed up time to link by disabling smart power down, ignore
3477 * the return value of this function because there is nothing
3478 * different we would do if it failed
3479 */
3480 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3481 phy_data &= ~IGP02E1000_PM_SPD;
3482 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3483 }
3484 }
3485
3486 int e1000e_up(struct e1000_adapter *adapter)
3487 {
3488 struct e1000_hw *hw = &adapter->hw;
3489
3490 /* hardware has been reset, we need to reload some things */
3491 e1000_configure(adapter);
3492
3493 clear_bit(__E1000_DOWN, &adapter->state);
3494
3495 napi_enable(&adapter->napi);
3496 if (adapter->msix_entries)
3497 e1000_configure_msix(adapter);
3498 e1000_irq_enable(adapter);
3499
3500 netif_start_queue(adapter->netdev);
3501
3502 /* fire a link change interrupt to start the watchdog */
3503 if (adapter->msix_entries)
3504 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3505 else
3506 ew32(ICS, E1000_ICS_LSC);
3507
3508 return 0;
3509 }
3510
3511 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3512 {
3513 struct e1000_hw *hw = &adapter->hw;
3514
3515 if (!(adapter->flags2 & FLAG2_DMA_BURST))
3516 return;
3517
3518 /* flush pending descriptor writebacks to memory */
3519 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3520 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3521
3522 /* execute the writes immediately */
3523 e1e_flush();
3524 }
3525
3526 static void e1000e_update_stats(struct e1000_adapter *adapter);
3527
3528 void e1000e_down(struct e1000_adapter *adapter)
3529 {
3530 struct net_device *netdev = adapter->netdev;
3531 struct e1000_hw *hw = &adapter->hw;
3532 u32 tctl, rctl;
3533
3534 /*
3535 * signal that we're down so the interrupt handler does not
3536 * reschedule our watchdog timer
3537 */
3538 set_bit(__E1000_DOWN, &adapter->state);
3539
3540 /* disable receives in the hardware */
3541 rctl = er32(RCTL);
3542 if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3543 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3544 /* flush and sleep below */
3545
3546 netif_stop_queue(netdev);
3547
3548 /* disable transmits in the hardware */
3549 tctl = er32(TCTL);
3550 tctl &= ~E1000_TCTL_EN;
3551 ew32(TCTL, tctl);
3552
3553 /* flush both disables and wait for them to finish */
3554 e1e_flush();
3555 usleep_range(10000, 20000);
3556
3557 napi_disable(&adapter->napi);
3558 e1000_irq_disable(adapter);
3559
3560 del_timer_sync(&adapter->watchdog_timer);
3561 del_timer_sync(&adapter->phy_info_timer);
3562
3563 netif_carrier_off(netdev);
3564
3565 spin_lock(&adapter->stats64_lock);
3566 e1000e_update_stats(adapter);
3567 spin_unlock(&adapter->stats64_lock);
3568
3569 e1000e_flush_descriptors(adapter);
3570 e1000_clean_tx_ring(adapter);
3571 e1000_clean_rx_ring(adapter);
3572
3573 adapter->link_speed = 0;
3574 adapter->link_duplex = 0;
3575
3576 if (!pci_channel_offline(adapter->pdev))
3577 e1000e_reset(adapter);
3578
3579 /*
3580 * TODO: for power management, we could drop the link and
3581 * pci_disable_device here.
3582 */
3583 }
3584
3585 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3586 {
3587 might_sleep();
3588 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3589 usleep_range(1000, 2000);
3590 e1000e_down(adapter);
3591 e1000e_up(adapter);
3592 clear_bit(__E1000_RESETTING, &adapter->state);
3593 }
3594
3595 /**
3596 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3597 * @adapter: board private structure to initialize
3598 *
3599 * e1000_sw_init initializes the Adapter private data structure.
3600 * Fields are initialized based on PCI device information and
3601 * OS network device settings (MTU size).
3602 **/
3603 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3604 {
3605 struct net_device *netdev = adapter->netdev;
3606
3607 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3608 adapter->rx_ps_bsize0 = 128;
3609 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3610 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3611
3612 spin_lock_init(&adapter->stats64_lock);
3613
3614 e1000e_set_interrupt_capability(adapter);
3615
3616 if (e1000_alloc_queues(adapter))
3617 return -ENOMEM;
3618
3619 /* Explicitly disable IRQ since the NIC can be in any state. */
3620 e1000_irq_disable(adapter);
3621
3622 set_bit(__E1000_DOWN, &adapter->state);
3623 return 0;
3624 }
3625
3626 /**
3627 * e1000_intr_msi_test - Interrupt Handler
3628 * @irq: interrupt number
3629 * @data: pointer to a network interface device structure
3630 **/
3631 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3632 {
3633 struct net_device *netdev = data;
3634 struct e1000_adapter *adapter = netdev_priv(netdev);
3635 struct e1000_hw *hw = &adapter->hw;
3636 u32 icr = er32(ICR);
3637
3638 e_dbg("icr is %08X\n", icr);
3639 if (icr & E1000_ICR_RXSEQ) {
3640 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3641 wmb();
3642 }
3643
3644 return IRQ_HANDLED;
3645 }
3646
3647 /**
3648 * e1000_test_msi_interrupt - Returns 0 for successful test
3649 * @adapter: board private struct
3650 *
3651 * code flow taken from tg3.c
3652 **/
3653 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3654 {
3655 struct net_device *netdev = adapter->netdev;
3656 struct e1000_hw *hw = &adapter->hw;
3657 int err;
3658
3659 /* poll_enable hasn't been called yet, so don't need disable */
3660 /* clear any pending events */
3661 er32(ICR);
3662
3663 /* free the real vector and request a test handler */
3664 e1000_free_irq(adapter);
3665 e1000e_reset_interrupt_capability(adapter);
3666
3667 /* Assume that the test fails, if it succeeds then the test
3668 * MSI irq handler will unset this flag */
3669 adapter->flags |= FLAG_MSI_TEST_FAILED;
3670
3671 err = pci_enable_msi(adapter->pdev);
3672 if (err)
3673 goto msi_test_failed;
3674
3675 err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3676 netdev->name, netdev);
3677 if (err) {
3678 pci_disable_msi(adapter->pdev);
3679 goto msi_test_failed;
3680 }
3681
3682 wmb();
3683
3684 e1000_irq_enable(adapter);
3685
3686 /* fire an unusual interrupt on the test handler */
3687 ew32(ICS, E1000_ICS_RXSEQ);
3688 e1e_flush();
3689 msleep(50);
3690
3691 e1000_irq_disable(adapter);
3692
3693 rmb();
3694
3695 if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3696 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3697 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3698 } else
3699 e_dbg("MSI interrupt test succeeded!\n");
3700
3701 free_irq(adapter->pdev->irq, netdev);
3702 pci_disable_msi(adapter->pdev);
3703
3704 msi_test_failed:
3705 e1000e_set_interrupt_capability(adapter);
3706 return e1000_request_irq(adapter);
3707 }
3708
3709 /**
3710 * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3711 * @adapter: board private struct
3712 *
3713 * code flow taken from tg3.c, called with e1000 interrupts disabled.
3714 **/
3715 static int e1000_test_msi(struct e1000_adapter *adapter)
3716 {
3717 int err;
3718 u16 pci_cmd;
3719
3720 if (!(adapter->flags & FLAG_MSI_ENABLED))
3721 return 0;
3722
3723 /* disable SERR in case the MSI write causes a master abort */
3724 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3725 if (pci_cmd & PCI_COMMAND_SERR)
3726 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3727 pci_cmd & ~PCI_COMMAND_SERR);
3728
3729 err = e1000_test_msi_interrupt(adapter);
3730
3731 /* re-enable SERR */
3732 if (pci_cmd & PCI_COMMAND_SERR) {
3733 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3734 pci_cmd |= PCI_COMMAND_SERR;
3735 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3736 }
3737
3738 return err;
3739 }
3740
3741 /**
3742 * e1000_open - Called when a network interface is made active
3743 * @netdev: network interface device structure
3744 *
3745 * Returns 0 on success, negative value on failure
3746 *
3747 * The open entry point is called when a network interface is made
3748 * active by the system (IFF_UP). At this point all resources needed
3749 * for transmit and receive operations are allocated, the interrupt
3750 * handler is registered with the OS, the watchdog timer is started,
3751 * and the stack is notified that the interface is ready.
3752 **/
3753 static int e1000_open(struct net_device *netdev)
3754 {
3755 struct e1000_adapter *adapter = netdev_priv(netdev);
3756 struct e1000_hw *hw = &adapter->hw;
3757 struct pci_dev *pdev = adapter->pdev;
3758 int err;
3759
3760 /* disallow open during test */
3761 if (test_bit(__E1000_TESTING, &adapter->state))
3762 return -EBUSY;
3763
3764 pm_runtime_get_sync(&pdev->dev);
3765
3766 netif_carrier_off(netdev);
3767
3768 /* allocate transmit descriptors */
3769 err = e1000e_setup_tx_resources(adapter);
3770 if (err)
3771 goto err_setup_tx;
3772
3773 /* allocate receive descriptors */
3774 err = e1000e_setup_rx_resources(adapter);
3775 if (err)
3776 goto err_setup_rx;
3777
3778 /*
3779 * If AMT is enabled, let the firmware know that the network
3780 * interface is now open and reset the part to a known state.
3781 */
3782 if (adapter->flags & FLAG_HAS_AMT) {
3783 e1000e_get_hw_control(adapter);
3784 e1000e_reset(adapter);
3785 }
3786
3787 e1000e_power_up_phy(adapter);
3788
3789 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3790 if ((adapter->hw.mng_cookie.status &
3791 E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3792 e1000_update_mng_vlan(adapter);
3793
3794 /* DMA latency requirement to workaround early-receive/jumbo issue */
3795 if ((adapter->flags & FLAG_HAS_ERT) ||
3796 (adapter->hw.mac.type == e1000_pch2lan))
3797 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3798 PM_QOS_CPU_DMA_LATENCY,
3799 PM_QOS_DEFAULT_VALUE);
3800
3801 /*
3802 * before we allocate an interrupt, we must be ready to handle it.
3803 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3804 * as soon as we call pci_request_irq, so we have to setup our
3805 * clean_rx handler before we do so.
3806 */
3807 e1000_configure(adapter);
3808
3809 err = e1000_request_irq(adapter);
3810 if (err)
3811 goto err_req_irq;
3812
3813 /*
3814 * Work around PCIe errata with MSI interrupts causing some chipsets to
3815 * ignore e1000e MSI messages, which means we need to test our MSI
3816 * interrupt now
3817 */
3818 if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3819 err = e1000_test_msi(adapter);
3820 if (err) {
3821 e_err("Interrupt allocation failed\n");
3822 goto err_req_irq;
3823 }
3824 }
3825
3826 /* From here on the code is the same as e1000e_up() */
3827 clear_bit(__E1000_DOWN, &adapter->state);
3828
3829 napi_enable(&adapter->napi);
3830
3831 e1000_irq_enable(adapter);
3832
3833 netif_start_queue(netdev);
3834
3835 adapter->idle_check = true;
3836 pm_runtime_put(&pdev->dev);
3837
3838 /* fire a link status change interrupt to start the watchdog */
3839 if (adapter->msix_entries)
3840 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3841 else
3842 ew32(ICS, E1000_ICS_LSC);
3843
3844 return 0;
3845
3846 err_req_irq:
3847 e1000e_release_hw_control(adapter);
3848 e1000_power_down_phy(adapter);
3849 e1000e_free_rx_resources(adapter);
3850 err_setup_rx:
3851 e1000e_free_tx_resources(adapter);
3852 err_setup_tx:
3853 e1000e_reset(adapter);
3854 pm_runtime_put_sync(&pdev->dev);
3855
3856 return err;
3857 }
3858
3859 /**
3860 * e1000_close - Disables a network interface
3861 * @netdev: network interface device structure
3862 *
3863 * Returns 0, this is not allowed to fail
3864 *
3865 * The close entry point is called when an interface is de-activated
3866 * by the OS. The hardware is still under the drivers control, but
3867 * needs to be disabled. A global MAC reset is issued to stop the
3868 * hardware, and all transmit and receive resources are freed.
3869 **/
3870 static int e1000_close(struct net_device *netdev)
3871 {
3872 struct e1000_adapter *adapter = netdev_priv(netdev);
3873 struct pci_dev *pdev = adapter->pdev;
3874
3875 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3876
3877 pm_runtime_get_sync(&pdev->dev);
3878
3879 if (!test_bit(__E1000_DOWN, &adapter->state)) {
3880 e1000e_down(adapter);
3881 e1000_free_irq(adapter);
3882 }
3883 e1000_power_down_phy(adapter);
3884
3885 e1000e_free_tx_resources(adapter);
3886 e1000e_free_rx_resources(adapter);
3887
3888 /*
3889 * kill manageability vlan ID if supported, but not if a vlan with
3890 * the same ID is registered on the host OS (let 8021q kill it)
3891 */
3892 if (adapter->hw.mng_cookie.status &
3893 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
3894 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3895
3896 /*
3897 * If AMT is enabled, let the firmware know that the network
3898 * interface is now closed
3899 */
3900 if ((adapter->flags & FLAG_HAS_AMT) &&
3901 !test_bit(__E1000_TESTING, &adapter->state))
3902 e1000e_release_hw_control(adapter);
3903
3904 if ((adapter->flags & FLAG_HAS_ERT) ||
3905 (adapter->hw.mac.type == e1000_pch2lan))
3906 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3907
3908 pm_runtime_put_sync(&pdev->dev);
3909
3910 return 0;
3911 }
3912 /**
3913 * e1000_set_mac - Change the Ethernet Address of the NIC
3914 * @netdev: network interface device structure
3915 * @p: pointer to an address structure
3916 *
3917 * Returns 0 on success, negative on failure
3918 **/
3919 static int e1000_set_mac(struct net_device *netdev, void *p)
3920 {
3921 struct e1000_adapter *adapter = netdev_priv(netdev);
3922 struct sockaddr *addr = p;
3923
3924 if (!is_valid_ether_addr(addr->sa_data))
3925 return -EADDRNOTAVAIL;
3926
3927 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3928 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3929
3930 e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3931
3932 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3933 /* activate the work around */
3934 e1000e_set_laa_state_82571(&adapter->hw, 1);
3935
3936 /*
3937 * Hold a copy of the LAA in RAR[14] This is done so that
3938 * between the time RAR[0] gets clobbered and the time it
3939 * gets fixed (in e1000_watchdog), the actual LAA is in one
3940 * of the RARs and no incoming packets directed to this port
3941 * are dropped. Eventually the LAA will be in RAR[0] and
3942 * RAR[14]
3943 */
3944 e1000e_rar_set(&adapter->hw,
3945 adapter->hw.mac.addr,
3946 adapter->hw.mac.rar_entry_count - 1);
3947 }
3948
3949 return 0;
3950 }
3951
3952 /**
3953 * e1000e_update_phy_task - work thread to update phy
3954 * @work: pointer to our work struct
3955 *
3956 * this worker thread exists because we must acquire a
3957 * semaphore to read the phy, which we could msleep while
3958 * waiting for it, and we can't msleep in a timer.
3959 **/
3960 static void e1000e_update_phy_task(struct work_struct *work)
3961 {
3962 struct e1000_adapter *adapter = container_of(work,
3963 struct e1000_adapter, update_phy_task);
3964
3965 if (test_bit(__E1000_DOWN, &adapter->state))
3966 return;
3967
3968 e1000_get_phy_info(&adapter->hw);
3969 }
3970
3971 /*
3972 * Need to wait a few seconds after link up to get diagnostic information from
3973 * the phy
3974 */
3975 static void e1000_update_phy_info(unsigned long data)
3976 {
3977 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3978
3979 if (test_bit(__E1000_DOWN, &adapter->state))
3980 return;
3981
3982 schedule_work(&adapter->update_phy_task);
3983 }
3984
3985 /**
3986 * e1000e_update_phy_stats - Update the PHY statistics counters
3987 * @adapter: board private structure
3988 *
3989 * Read/clear the upper 16-bit PHY registers and read/accumulate lower
3990 **/
3991 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
3992 {
3993 struct e1000_hw *hw = &adapter->hw;
3994 s32 ret_val;
3995 u16 phy_data;
3996
3997 ret_val = hw->phy.ops.acquire(hw);
3998 if (ret_val)
3999 return;
4000
4001 /*
4002 * A page set is expensive so check if already on desired page.
4003 * If not, set to the page with the PHY status registers.
4004 */
4005 hw->phy.addr = 1;
4006 ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4007 &phy_data);
4008 if (ret_val)
4009 goto release;
4010 if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4011 ret_val = hw->phy.ops.set_page(hw,
4012 HV_STATS_PAGE << IGP_PAGE_SHIFT);
4013 if (ret_val)
4014 goto release;
4015 }
4016
4017 /* Single Collision Count */
4018 hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4019 ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4020 if (!ret_val)
4021 adapter->stats.scc += phy_data;
4022
4023 /* Excessive Collision Count */
4024 hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4025 ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4026 if (!ret_val)
4027 adapter->stats.ecol += phy_data;
4028
4029 /* Multiple Collision Count */
4030 hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4031 ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4032 if (!ret_val)
4033 adapter->stats.mcc += phy_data;
4034
4035 /* Late Collision Count */
4036 hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4037 ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4038 if (!ret_val)
4039 adapter->stats.latecol += phy_data;
4040
4041 /* Collision Count - also used for adaptive IFS */
4042 hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4043 ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4044 if (!ret_val)
4045 hw->mac.collision_delta = phy_data;
4046
4047 /* Defer Count */
4048 hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4049 ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4050 if (!ret_val)
4051 adapter->stats.dc += phy_data;
4052
4053 /* Transmit with no CRS */
4054 hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4055 ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4056 if (!ret_val)
4057 adapter->stats.tncrs += phy_data;
4058
4059 release:
4060 hw->phy.ops.release(hw);
4061 }
4062
4063 /**
4064 * e1000e_update_stats - Update the board statistics counters
4065 * @adapter: board private structure
4066 **/
4067 static void e1000e_update_stats(struct e1000_adapter *adapter)
4068 {
4069 struct net_device *netdev = adapter->netdev;
4070 struct e1000_hw *hw = &adapter->hw;
4071 struct pci_dev *pdev = adapter->pdev;
4072
4073 /*
4074 * Prevent stats update while adapter is being reset, or if the pci
4075 * connection is down.
4076 */
4077 if (adapter->link_speed == 0)
4078 return;
4079 if (pci_channel_offline(pdev))
4080 return;
4081
4082 adapter->stats.crcerrs += er32(CRCERRS);
4083 adapter->stats.gprc += er32(GPRC);
4084 adapter->stats.gorc += er32(GORCL);
4085 er32(GORCH); /* Clear gorc */
4086 adapter->stats.bprc += er32(BPRC);
4087 adapter->stats.mprc += er32(MPRC);
4088 adapter->stats.roc += er32(ROC);
4089
4090 adapter->stats.mpc += er32(MPC);
4091
4092 /* Half-duplex statistics */
4093 if (adapter->link_duplex == HALF_DUPLEX) {
4094 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4095 e1000e_update_phy_stats(adapter);
4096 } else {
4097 adapter->stats.scc += er32(SCC);
4098 adapter->stats.ecol += er32(ECOL);
4099 adapter->stats.mcc += er32(MCC);
4100 adapter->stats.latecol += er32(LATECOL);
4101 adapter->stats.dc += er32(DC);
4102
4103 hw->mac.collision_delta = er32(COLC);
4104
4105 if ((hw->mac.type != e1000_82574) &&
4106 (hw->mac.type != e1000_82583))
4107 adapter->stats.tncrs += er32(TNCRS);
4108 }
4109 adapter->stats.colc += hw->mac.collision_delta;
4110 }
4111
4112 adapter->stats.xonrxc += er32(XONRXC);
4113 adapter->stats.xontxc += er32(XONTXC);
4114 adapter->stats.xoffrxc += er32(XOFFRXC);
4115 adapter->stats.xofftxc += er32(XOFFTXC);
4116 adapter->stats.gptc += er32(GPTC);
4117 adapter->stats.gotc += er32(GOTCL);
4118 er32(GOTCH); /* Clear gotc */
4119 adapter->stats.rnbc += er32(RNBC);
4120 adapter->stats.ruc += er32(RUC);
4121
4122 adapter->stats.mptc += er32(MPTC);
4123 adapter->stats.bptc += er32(BPTC);
4124
4125 /* used for adaptive IFS */
4126
4127 hw->mac.tx_packet_delta = er32(TPT);
4128 adapter->stats.tpt += hw->mac.tx_packet_delta;
4129
4130 adapter->stats.algnerrc += er32(ALGNERRC);
4131 adapter->stats.rxerrc += er32(RXERRC);
4132 adapter->stats.cexterr += er32(CEXTERR);
4133 adapter->stats.tsctc += er32(TSCTC);
4134 adapter->stats.tsctfc += er32(TSCTFC);
4135
4136 /* Fill out the OS statistics structure */
4137 netdev->stats.multicast = adapter->stats.mprc;
4138 netdev->stats.collisions = adapter->stats.colc;
4139
4140 /* Rx Errors */
4141
4142 /*
4143 * RLEC on some newer hardware can be incorrect so build
4144 * our own version based on RUC and ROC
4145 */
4146 netdev->stats.rx_errors = adapter->stats.rxerrc +
4147 adapter->stats.crcerrs + adapter->stats.algnerrc +
4148 adapter->stats.ruc + adapter->stats.roc +
4149 adapter->stats.cexterr;
4150 netdev->stats.rx_length_errors = adapter->stats.ruc +
4151 adapter->stats.roc;
4152 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4153 netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4154 netdev->stats.rx_missed_errors = adapter->stats.mpc;
4155
4156 /* Tx Errors */
4157 netdev->stats.tx_errors = adapter->stats.ecol +
4158 adapter->stats.latecol;
4159 netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4160 netdev->stats.tx_window_errors = adapter->stats.latecol;
4161 netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4162
4163 /* Tx Dropped needs to be maintained elsewhere */
4164
4165 /* Management Stats */
4166 adapter->stats.mgptc += er32(MGTPTC);
4167 adapter->stats.mgprc += er32(MGTPRC);
4168 adapter->stats.mgpdc += er32(MGTPDC);
4169 }
4170
4171 /**
4172 * e1000_phy_read_status - Update the PHY register status snapshot
4173 * @adapter: board private structure
4174 **/
4175 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4176 {
4177 struct e1000_hw *hw = &adapter->hw;
4178 struct e1000_phy_regs *phy = &adapter->phy_regs;
4179
4180 if ((er32(STATUS) & E1000_STATUS_LU) &&
4181 (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4182 int ret_val;
4183
4184 ret_val = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4185 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4186 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4187 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4188 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4189 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4190 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4191 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4192 if (ret_val)
4193 e_warn("Error reading PHY register\n");
4194 } else {
4195 /*
4196 * Do not read PHY registers if link is not up
4197 * Set values to typical power-on defaults
4198 */
4199 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4200 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4201 BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4202 BMSR_ERCAP);
4203 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4204 ADVERTISE_ALL | ADVERTISE_CSMA);
4205 phy->lpa = 0;
4206 phy->expansion = EXPANSION_ENABLENPAGE;
4207 phy->ctrl1000 = ADVERTISE_1000FULL;
4208 phy->stat1000 = 0;
4209 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4210 }
4211 }
4212
4213 static void e1000_print_link_info(struct e1000_adapter *adapter)
4214 {
4215 struct e1000_hw *hw = &adapter->hw;
4216 u32 ctrl = er32(CTRL);
4217
4218 /* Link status message must follow this format for user tools */
4219 printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4220 adapter->netdev->name,
4221 adapter->link_speed,
4222 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
4223 (ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
4224 (ctrl & E1000_CTRL_RFCE) ? "Rx" :
4225 (ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
4226 }
4227
4228 static bool e1000e_has_link(struct e1000_adapter *adapter)
4229 {
4230 struct e1000_hw *hw = &adapter->hw;
4231 bool link_active = 0;
4232 s32 ret_val = 0;
4233
4234 /*
4235 * get_link_status is set on LSC (link status) interrupt or
4236 * Rx sequence error interrupt. get_link_status will stay
4237 * false until the check_for_link establishes link
4238 * for copper adapters ONLY
4239 */
4240 switch (hw->phy.media_type) {
4241 case e1000_media_type_copper:
4242 if (hw->mac.get_link_status) {
4243 ret_val = hw->mac.ops.check_for_link(hw);
4244 link_active = !hw->mac.get_link_status;
4245 } else {
4246 link_active = 1;
4247 }
4248 break;
4249 case e1000_media_type_fiber:
4250 ret_val = hw->mac.ops.check_for_link(hw);
4251 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4252 break;
4253 case e1000_media_type_internal_serdes:
4254 ret_val = hw->mac.ops.check_for_link(hw);
4255 link_active = adapter->hw.mac.serdes_has_link;
4256 break;
4257 default:
4258 case e1000_media_type_unknown:
4259 break;
4260 }
4261
4262 if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4263 (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4264 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4265 e_info("Gigabit has been disabled, downgrading speed\n");
4266 }
4267
4268 return link_active;
4269 }
4270
4271 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4272 {
4273 /* make sure the receive unit is started */
4274 if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4275 (adapter->flags & FLAG_RX_RESTART_NOW)) {
4276 struct e1000_hw *hw = &adapter->hw;
4277 u32 rctl = er32(RCTL);
4278 ew32(RCTL, rctl | E1000_RCTL_EN);
4279 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4280 }
4281 }
4282
4283 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4284 {
4285 struct e1000_hw *hw = &adapter->hw;
4286
4287 /*
4288 * With 82574 controllers, PHY needs to be checked periodically
4289 * for hung state and reset, if two calls return true
4290 */
4291 if (e1000_check_phy_82574(hw))
4292 adapter->phy_hang_count++;
4293 else
4294 adapter->phy_hang_count = 0;
4295
4296 if (adapter->phy_hang_count > 1) {
4297 adapter->phy_hang_count = 0;
4298 schedule_work(&adapter->reset_task);
4299 }
4300 }
4301
4302 /**
4303 * e1000_watchdog - Timer Call-back
4304 * @data: pointer to adapter cast into an unsigned long
4305 **/
4306 static void e1000_watchdog(unsigned long data)
4307 {
4308 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4309
4310 /* Do the rest outside of interrupt context */
4311 schedule_work(&adapter->watchdog_task);
4312
4313 /* TODO: make this use queue_delayed_work() */
4314 }
4315
4316 static void e1000_watchdog_task(struct work_struct *work)
4317 {
4318 struct e1000_adapter *adapter = container_of(work,
4319 struct e1000_adapter, watchdog_task);
4320 struct net_device *netdev = adapter->netdev;
4321 struct e1000_mac_info *mac = &adapter->hw.mac;
4322 struct e1000_phy_info *phy = &adapter->hw.phy;
4323 struct e1000_ring *tx_ring = adapter->tx_ring;
4324 struct e1000_hw *hw = &adapter->hw;
4325 u32 link, tctl;
4326
4327 if (test_bit(__E1000_DOWN, &adapter->state))
4328 return;
4329
4330 link = e1000e_has_link(adapter);
4331 if ((netif_carrier_ok(netdev)) && link) {
4332 /* Cancel scheduled suspend requests. */
4333 pm_runtime_resume(netdev->dev.parent);
4334
4335 e1000e_enable_receives(adapter);
4336 goto link_up;
4337 }
4338
4339 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4340 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4341 e1000_update_mng_vlan(adapter);
4342
4343 if (link) {
4344 if (!netif_carrier_ok(netdev)) {
4345 bool txb2b = 1;
4346
4347 /* Cancel scheduled suspend requests. */
4348 pm_runtime_resume(netdev->dev.parent);
4349
4350 /* update snapshot of PHY registers on LSC */
4351 e1000_phy_read_status(adapter);
4352 mac->ops.get_link_up_info(&adapter->hw,
4353 &adapter->link_speed,
4354 &adapter->link_duplex);
4355 e1000_print_link_info(adapter);
4356 /*
4357 * On supported PHYs, check for duplex mismatch only
4358 * if link has autonegotiated at 10/100 half
4359 */
4360 if ((hw->phy.type == e1000_phy_igp_3 ||
4361 hw->phy.type == e1000_phy_bm) &&
4362 (hw->mac.autoneg == true) &&
4363 (adapter->link_speed == SPEED_10 ||
4364 adapter->link_speed == SPEED_100) &&
4365 (adapter->link_duplex == HALF_DUPLEX)) {
4366 u16 autoneg_exp;
4367
4368 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4369
4370 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4371 e_info("Autonegotiated half duplex but link partner cannot autoneg. Try forcing full duplex if link gets many collisions.\n");
4372 }
4373
4374 /* adjust timeout factor according to speed/duplex */
4375 adapter->tx_timeout_factor = 1;
4376 switch (adapter->link_speed) {
4377 case SPEED_10:
4378 txb2b = 0;
4379 adapter->tx_timeout_factor = 16;
4380 break;
4381 case SPEED_100:
4382 txb2b = 0;
4383 adapter->tx_timeout_factor = 10;
4384 break;
4385 }
4386
4387 /*
4388 * workaround: re-program speed mode bit after
4389 * link-up event
4390 */
4391 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4392 !txb2b) {
4393 u32 tarc0;
4394 tarc0 = er32(TARC(0));
4395 tarc0 &= ~SPEED_MODE_BIT;
4396 ew32(TARC(0), tarc0);
4397 }
4398
4399 /*
4400 * disable TSO for pcie and 10/100 speeds, to avoid
4401 * some hardware issues
4402 */
4403 if (!(adapter->flags & FLAG_TSO_FORCE)) {
4404 switch (adapter->link_speed) {
4405 case SPEED_10:
4406 case SPEED_100:
4407 e_info("10/100 speed: disabling TSO\n");
4408 netdev->features &= ~NETIF_F_TSO;
4409 netdev->features &= ~NETIF_F_TSO6;
4410 break;
4411 case SPEED_1000:
4412 netdev->features |= NETIF_F_TSO;
4413 netdev->features |= NETIF_F_TSO6;
4414 break;
4415 default:
4416 /* oops */
4417 break;
4418 }
4419 }
4420
4421 /*
4422 * enable transmits in the hardware, need to do this
4423 * after setting TARC(0)
4424 */
4425 tctl = er32(TCTL);
4426 tctl |= E1000_TCTL_EN;
4427 ew32(TCTL, tctl);
4428
4429 /*
4430 * Perform any post-link-up configuration before
4431 * reporting link up.
4432 */
4433 if (phy->ops.cfg_on_link_up)
4434 phy->ops.cfg_on_link_up(hw);
4435
4436 netif_carrier_on(netdev);
4437
4438 if (!test_bit(__E1000_DOWN, &adapter->state))
4439 mod_timer(&adapter->phy_info_timer,
4440 round_jiffies(jiffies + 2 * HZ));
4441 }
4442 } else {
4443 if (netif_carrier_ok(netdev)) {
4444 adapter->link_speed = 0;
4445 adapter->link_duplex = 0;
4446 /* Link status message must follow this format */
4447 printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4448 adapter->netdev->name);
4449 netif_carrier_off(netdev);
4450 if (!test_bit(__E1000_DOWN, &adapter->state))
4451 mod_timer(&adapter->phy_info_timer,
4452 round_jiffies(jiffies + 2 * HZ));
4453
4454 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4455 schedule_work(&adapter->reset_task);
4456 else
4457 pm_schedule_suspend(netdev->dev.parent,
4458 LINK_TIMEOUT);
4459 }
4460 }
4461
4462 link_up:
4463 spin_lock(&adapter->stats64_lock);
4464 e1000e_update_stats(adapter);
4465
4466 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4467 adapter->tpt_old = adapter->stats.tpt;
4468 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4469 adapter->colc_old = adapter->stats.colc;
4470
4471 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4472 adapter->gorc_old = adapter->stats.gorc;
4473 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4474 adapter->gotc_old = adapter->stats.gotc;
4475 spin_unlock(&adapter->stats64_lock);
4476
4477 e1000e_update_adaptive(&adapter->hw);
4478
4479 if (!netif_carrier_ok(netdev) &&
4480 (e1000_desc_unused(tx_ring) + 1 < tx_ring->count)) {
4481 /*
4482 * We've lost link, so the controller stops DMA,
4483 * but we've got queued Tx work that's never going
4484 * to get done, so reset controller to flush Tx.
4485 * (Do the reset outside of interrupt context).
4486 */
4487 schedule_work(&adapter->reset_task);
4488 /* return immediately since reset is imminent */
4489 return;
4490 }
4491
4492 /* Simple mode for Interrupt Throttle Rate (ITR) */
4493 if (adapter->itr_setting == 4) {
4494 /*
4495 * Symmetric Tx/Rx gets a reduced ITR=2000;
4496 * Total asymmetrical Tx or Rx gets ITR=8000;
4497 * everyone else is between 2000-8000.
4498 */
4499 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4500 u32 dif = (adapter->gotc > adapter->gorc ?
4501 adapter->gotc - adapter->gorc :
4502 adapter->gorc - adapter->gotc) / 10000;
4503 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4504
4505 ew32(ITR, 1000000000 / (itr * 256));
4506 }
4507
4508 /* Cause software interrupt to ensure Rx ring is cleaned */
4509 if (adapter->msix_entries)
4510 ew32(ICS, adapter->rx_ring->ims_val);
4511 else
4512 ew32(ICS, E1000_ICS_RXDMT0);
4513
4514 /* flush pending descriptors to memory before detecting Tx hang */
4515 e1000e_flush_descriptors(adapter);
4516
4517 /* Force detection of hung controller every watchdog period */
4518 adapter->detect_tx_hung = 1;
4519
4520 /*
4521 * With 82571 controllers, LAA may be overwritten due to controller
4522 * reset from the other port. Set the appropriate LAA in RAR[0]
4523 */
4524 if (e1000e_get_laa_state_82571(hw))
4525 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4526
4527 if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4528 e1000e_check_82574_phy_workaround(adapter);
4529
4530 /* Reset the timer */
4531 if (!test_bit(__E1000_DOWN, &adapter->state))
4532 mod_timer(&adapter->watchdog_timer,
4533 round_jiffies(jiffies + 2 * HZ));
4534 }
4535
4536 #define E1000_TX_FLAGS_CSUM 0x00000001
4537 #define E1000_TX_FLAGS_VLAN 0x00000002
4538 #define E1000_TX_FLAGS_TSO 0x00000004
4539 #define E1000_TX_FLAGS_IPV4 0x00000008
4540 #define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
4541 #define E1000_TX_FLAGS_VLAN_SHIFT 16
4542
4543 static int e1000_tso(struct e1000_adapter *adapter,
4544 struct sk_buff *skb)
4545 {
4546 struct e1000_ring *tx_ring = adapter->tx_ring;
4547 struct e1000_context_desc *context_desc;
4548 struct e1000_buffer *buffer_info;
4549 unsigned int i;
4550 u32 cmd_length = 0;
4551 u16 ipcse = 0, tucse, mss;
4552 u8 ipcss, ipcso, tucss, tucso, hdr_len;
4553
4554 if (!skb_is_gso(skb))
4555 return 0;
4556
4557 if (skb_header_cloned(skb)) {
4558 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4559
4560 if (err)
4561 return err;
4562 }
4563
4564 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4565 mss = skb_shinfo(skb)->gso_size;
4566 if (skb->protocol == htons(ETH_P_IP)) {
4567 struct iphdr *iph = ip_hdr(skb);
4568 iph->tot_len = 0;
4569 iph->check = 0;
4570 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4571 0, IPPROTO_TCP, 0);
4572 cmd_length = E1000_TXD_CMD_IP;
4573 ipcse = skb_transport_offset(skb) - 1;
4574 } else if (skb_is_gso_v6(skb)) {
4575 ipv6_hdr(skb)->payload_len = 0;
4576 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4577 &ipv6_hdr(skb)->daddr,
4578 0, IPPROTO_TCP, 0);
4579 ipcse = 0;
4580 }
4581 ipcss = skb_network_offset(skb);
4582 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4583 tucss = skb_transport_offset(skb);
4584 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4585 tucse = 0;
4586
4587 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4588 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4589
4590 i = tx_ring->next_to_use;
4591 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4592 buffer_info = &tx_ring->buffer_info[i];
4593
4594 context_desc->lower_setup.ip_fields.ipcss = ipcss;
4595 context_desc->lower_setup.ip_fields.ipcso = ipcso;
4596 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
4597 context_desc->upper_setup.tcp_fields.tucss = tucss;
4598 context_desc->upper_setup.tcp_fields.tucso = tucso;
4599 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4600 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
4601 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4602 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4603
4604 buffer_info->time_stamp = jiffies;
4605 buffer_info->next_to_watch = i;
4606
4607 i++;
4608 if (i == tx_ring->count)
4609 i = 0;
4610 tx_ring->next_to_use = i;
4611
4612 return 1;
4613 }
4614
4615 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4616 {
4617 struct e1000_ring *tx_ring = adapter->tx_ring;
4618 struct e1000_context_desc *context_desc;
4619 struct e1000_buffer *buffer_info;
4620 unsigned int i;
4621 u8 css;
4622 u32 cmd_len = E1000_TXD_CMD_DEXT;
4623 __be16 protocol;
4624
4625 if (skb->ip_summed != CHECKSUM_PARTIAL)
4626 return 0;
4627
4628 if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4629 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4630 else
4631 protocol = skb->protocol;
4632
4633 switch (protocol) {
4634 case cpu_to_be16(ETH_P_IP):
4635 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4636 cmd_len |= E1000_TXD_CMD_TCP;
4637 break;
4638 case cpu_to_be16(ETH_P_IPV6):
4639 /* XXX not handling all IPV6 headers */
4640 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4641 cmd_len |= E1000_TXD_CMD_TCP;
4642 break;
4643 default:
4644 if (unlikely(net_ratelimit()))
4645 e_warn("checksum_partial proto=%x!\n",
4646 be16_to_cpu(protocol));
4647 break;
4648 }
4649
4650 css = skb_checksum_start_offset(skb);
4651
4652 i = tx_ring->next_to_use;
4653 buffer_info = &tx_ring->buffer_info[i];
4654 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4655
4656 context_desc->lower_setup.ip_config = 0;
4657 context_desc->upper_setup.tcp_fields.tucss = css;
4658 context_desc->upper_setup.tcp_fields.tucso =
4659 css + skb->csum_offset;
4660 context_desc->upper_setup.tcp_fields.tucse = 0;
4661 context_desc->tcp_seg_setup.data = 0;
4662 context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4663
4664 buffer_info->time_stamp = jiffies;
4665 buffer_info->next_to_watch = i;
4666
4667 i++;
4668 if (i == tx_ring->count)
4669 i = 0;
4670 tx_ring->next_to_use = i;
4671
4672 return 1;
4673 }
4674
4675 #define E1000_MAX_PER_TXD 8192
4676 #define E1000_MAX_TXD_PWR 12
4677
4678 static int e1000_tx_map(struct e1000_adapter *adapter,
4679 struct sk_buff *skb, unsigned int first,
4680 unsigned int max_per_txd, unsigned int nr_frags,
4681 unsigned int mss)
4682 {
4683 struct e1000_ring *tx_ring = adapter->tx_ring;
4684 struct pci_dev *pdev = adapter->pdev;
4685 struct e1000_buffer *buffer_info;
4686 unsigned int len = skb_headlen(skb);
4687 unsigned int offset = 0, size, count = 0, i;
4688 unsigned int f, bytecount, segs;
4689
4690 i = tx_ring->next_to_use;
4691
4692 while (len) {
4693 buffer_info = &tx_ring->buffer_info[i];
4694 size = min(len, max_per_txd);
4695
4696 buffer_info->length = size;
4697 buffer_info->time_stamp = jiffies;
4698 buffer_info->next_to_watch = i;
4699 buffer_info->dma = dma_map_single(&pdev->dev,
4700 skb->data + offset,
4701 size, DMA_TO_DEVICE);
4702 buffer_info->mapped_as_page = false;
4703 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4704 goto dma_error;
4705
4706 len -= size;
4707 offset += size;
4708 count++;
4709
4710 if (len) {
4711 i++;
4712 if (i == tx_ring->count)
4713 i = 0;
4714 }
4715 }
4716
4717 for (f = 0; f < nr_frags; f++) {
4718 const struct skb_frag_struct *frag;
4719
4720 frag = &skb_shinfo(skb)->frags[f];
4721 len = skb_frag_size(frag);
4722 offset = 0;
4723
4724 while (len) {
4725 i++;
4726 if (i == tx_ring->count)
4727 i = 0;
4728
4729 buffer_info = &tx_ring->buffer_info[i];
4730 size = min(len, max_per_txd);
4731
4732 buffer_info->length = size;
4733 buffer_info->time_stamp = jiffies;
4734 buffer_info->next_to_watch = i;
4735 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
4736 offset, size, DMA_TO_DEVICE);
4737 buffer_info->mapped_as_page = true;
4738 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4739 goto dma_error;
4740
4741 len -= size;
4742 offset += size;
4743 count++;
4744 }
4745 }
4746
4747 segs = skb_shinfo(skb)->gso_segs ? : 1;
4748 /* multiply data chunks by size of headers */
4749 bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4750
4751 tx_ring->buffer_info[i].skb = skb;
4752 tx_ring->buffer_info[i].segs = segs;
4753 tx_ring->buffer_info[i].bytecount = bytecount;
4754 tx_ring->buffer_info[first].next_to_watch = i;
4755
4756 return count;
4757
4758 dma_error:
4759 dev_err(&pdev->dev, "Tx DMA map failed\n");
4760 buffer_info->dma = 0;
4761 if (count)
4762 count--;
4763
4764 while (count--) {
4765 if (i == 0)
4766 i += tx_ring->count;
4767 i--;
4768 buffer_info = &tx_ring->buffer_info[i];
4769 e1000_put_txbuf(adapter, buffer_info);
4770 }
4771
4772 return 0;
4773 }
4774
4775 static void e1000_tx_queue(struct e1000_adapter *adapter,
4776 int tx_flags, int count)
4777 {
4778 struct e1000_ring *tx_ring = adapter->tx_ring;
4779 struct e1000_tx_desc *tx_desc = NULL;
4780 struct e1000_buffer *buffer_info;
4781 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4782 unsigned int i;
4783
4784 if (tx_flags & E1000_TX_FLAGS_TSO) {
4785 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4786 E1000_TXD_CMD_TSE;
4787 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4788
4789 if (tx_flags & E1000_TX_FLAGS_IPV4)
4790 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4791 }
4792
4793 if (tx_flags & E1000_TX_FLAGS_CSUM) {
4794 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4795 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4796 }
4797
4798 if (tx_flags & E1000_TX_FLAGS_VLAN) {
4799 txd_lower |= E1000_TXD_CMD_VLE;
4800 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4801 }
4802
4803 i = tx_ring->next_to_use;
4804
4805 do {
4806 buffer_info = &tx_ring->buffer_info[i];
4807 tx_desc = E1000_TX_DESC(*tx_ring, i);
4808 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4809 tx_desc->lower.data =
4810 cpu_to_le32(txd_lower | buffer_info->length);
4811 tx_desc->upper.data = cpu_to_le32(txd_upper);
4812
4813 i++;
4814 if (i == tx_ring->count)
4815 i = 0;
4816 } while (--count > 0);
4817
4818 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4819
4820 /*
4821 * Force memory writes to complete before letting h/w
4822 * know there are new descriptors to fetch. (Only
4823 * applicable for weak-ordered memory model archs,
4824 * such as IA-64).
4825 */
4826 wmb();
4827
4828 tx_ring->next_to_use = i;
4829
4830 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
4831 e1000e_update_tdt_wa(adapter, i);
4832 else
4833 writel(i, adapter->hw.hw_addr + tx_ring->tail);
4834
4835 /*
4836 * we need this if more than one processor can write to our tail
4837 * at a time, it synchronizes IO on IA64/Altix systems
4838 */
4839 mmiowb();
4840 }
4841
4842 #define MINIMUM_DHCP_PACKET_SIZE 282
4843 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4844 struct sk_buff *skb)
4845 {
4846 struct e1000_hw *hw = &adapter->hw;
4847 u16 length, offset;
4848
4849 if (vlan_tx_tag_present(skb)) {
4850 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4851 (adapter->hw.mng_cookie.status &
4852 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4853 return 0;
4854 }
4855
4856 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4857 return 0;
4858
4859 if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4860 return 0;
4861
4862 {
4863 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4864 struct udphdr *udp;
4865
4866 if (ip->protocol != IPPROTO_UDP)
4867 return 0;
4868
4869 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4870 if (ntohs(udp->dest) != 67)
4871 return 0;
4872
4873 offset = (u8 *)udp + 8 - skb->data;
4874 length = skb->len - offset;
4875 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4876 }
4877
4878 return 0;
4879 }
4880
4881 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4882 {
4883 struct e1000_adapter *adapter = netdev_priv(netdev);
4884
4885 netif_stop_queue(netdev);
4886 /*
4887 * Herbert's original patch had:
4888 * smp_mb__after_netif_stop_queue();
4889 * but since that doesn't exist yet, just open code it.
4890 */
4891 smp_mb();
4892
4893 /*
4894 * We need to check again in a case another CPU has just
4895 * made room available.
4896 */
4897 if (e1000_desc_unused(adapter->tx_ring) < size)
4898 return -EBUSY;
4899
4900 /* A reprieve! */
4901 netif_start_queue(netdev);
4902 ++adapter->restart_queue;
4903 return 0;
4904 }
4905
4906 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4907 {
4908 struct e1000_adapter *adapter = netdev_priv(netdev);
4909
4910 if (e1000_desc_unused(adapter->tx_ring) >= size)
4911 return 0;
4912 return __e1000_maybe_stop_tx(netdev, size);
4913 }
4914
4915 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4916 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4917 struct net_device *netdev)
4918 {
4919 struct e1000_adapter *adapter = netdev_priv(netdev);
4920 struct e1000_ring *tx_ring = adapter->tx_ring;
4921 unsigned int first;
4922 unsigned int max_per_txd = E1000_MAX_PER_TXD;
4923 unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4924 unsigned int tx_flags = 0;
4925 unsigned int len = skb_headlen(skb);
4926 unsigned int nr_frags;
4927 unsigned int mss;
4928 int count = 0;
4929 int tso;
4930 unsigned int f;
4931
4932 if (test_bit(__E1000_DOWN, &adapter->state)) {
4933 dev_kfree_skb_any(skb);
4934 return NETDEV_TX_OK;
4935 }
4936
4937 if (skb->len <= 0) {
4938 dev_kfree_skb_any(skb);
4939 return NETDEV_TX_OK;
4940 }
4941
4942 mss = skb_shinfo(skb)->gso_size;
4943 /*
4944 * The controller does a simple calculation to
4945 * make sure there is enough room in the FIFO before
4946 * initiating the DMA for each buffer. The calc is:
4947 * 4 = ceil(buffer len/mss). To make sure we don't
4948 * overrun the FIFO, adjust the max buffer len if mss
4949 * drops.
4950 */
4951 if (mss) {
4952 u8 hdr_len;
4953 max_per_txd = min(mss << 2, max_per_txd);
4954 max_txd_pwr = fls(max_per_txd) - 1;
4955
4956 /*
4957 * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4958 * points to just header, pull a few bytes of payload from
4959 * frags into skb->data
4960 */
4961 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4962 /*
4963 * we do this workaround for ES2LAN, but it is un-necessary,
4964 * avoiding it could save a lot of cycles
4965 */
4966 if (skb->data_len && (hdr_len == len)) {
4967 unsigned int pull_size;
4968
4969 pull_size = min((unsigned int)4, skb->data_len);
4970 if (!__pskb_pull_tail(skb, pull_size)) {
4971 e_err("__pskb_pull_tail failed.\n");
4972 dev_kfree_skb_any(skb);
4973 return NETDEV_TX_OK;
4974 }
4975 len = skb_headlen(skb);
4976 }
4977 }
4978
4979 /* reserve a descriptor for the offload context */
4980 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4981 count++;
4982 count++;
4983
4984 count += TXD_USE_COUNT(len, max_txd_pwr);
4985
4986 nr_frags = skb_shinfo(skb)->nr_frags;
4987 for (f = 0; f < nr_frags; f++)
4988 count += TXD_USE_COUNT(skb_frag_size(&skb_shinfo(skb)->frags[f]),
4989 max_txd_pwr);
4990
4991 if (adapter->hw.mac.tx_pkt_filtering)
4992 e1000_transfer_dhcp_info(adapter, skb);
4993
4994 /*
4995 * need: count + 2 desc gap to keep tail from touching
4996 * head, otherwise try next time
4997 */
4998 if (e1000_maybe_stop_tx(netdev, count + 2))
4999 return NETDEV_TX_BUSY;
5000
5001 if (vlan_tx_tag_present(skb)) {
5002 tx_flags |= E1000_TX_FLAGS_VLAN;
5003 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
5004 }
5005
5006 first = tx_ring->next_to_use;
5007
5008 tso = e1000_tso(adapter, skb);
5009 if (tso < 0) {
5010 dev_kfree_skb_any(skb);
5011 return NETDEV_TX_OK;
5012 }
5013
5014 if (tso)
5015 tx_flags |= E1000_TX_FLAGS_TSO;
5016 else if (e1000_tx_csum(adapter, skb))
5017 tx_flags |= E1000_TX_FLAGS_CSUM;
5018
5019 /*
5020 * Old method was to assume IPv4 packet by default if TSO was enabled.
5021 * 82571 hardware supports TSO capabilities for IPv6 as well...
5022 * no longer assume, we must.
5023 */
5024 if (skb->protocol == htons(ETH_P_IP))
5025 tx_flags |= E1000_TX_FLAGS_IPV4;
5026
5027 /* if count is 0 then mapping error has occurred */
5028 count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
5029 if (count) {
5030 e1000_tx_queue(adapter, tx_flags, count);
5031 /* Make sure there is space in the ring for the next send. */
5032 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
5033
5034 } else {
5035 dev_kfree_skb_any(skb);
5036 tx_ring->buffer_info[first].time_stamp = 0;
5037 tx_ring->next_to_use = first;
5038 }
5039
5040 return NETDEV_TX_OK;
5041 }
5042
5043 /**
5044 * e1000_tx_timeout - Respond to a Tx Hang
5045 * @netdev: network interface device structure
5046 **/
5047 static void e1000_tx_timeout(struct net_device *netdev)
5048 {
5049 struct e1000_adapter *adapter = netdev_priv(netdev);
5050
5051 /* Do the reset outside of interrupt context */
5052 adapter->tx_timeout_count++;
5053 schedule_work(&adapter->reset_task);
5054 }
5055
5056 static void e1000_reset_task(struct work_struct *work)
5057 {
5058 struct e1000_adapter *adapter;
5059 adapter = container_of(work, struct e1000_adapter, reset_task);
5060
5061 /* don't run the task if already down */
5062 if (test_bit(__E1000_DOWN, &adapter->state))
5063 return;
5064
5065 if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5066 (adapter->flags & FLAG_RX_RESTART_NOW))) {
5067 e1000e_dump(adapter);
5068 e_err("Reset adapter\n");
5069 }
5070 e1000e_reinit_locked(adapter);
5071 }
5072
5073 /**
5074 * e1000_get_stats64 - Get System Network Statistics
5075 * @netdev: network interface device structure
5076 * @stats: rtnl_link_stats64 pointer
5077 *
5078 * Returns the address of the device statistics structure.
5079 **/
5080 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
5081 struct rtnl_link_stats64 *stats)
5082 {
5083 struct e1000_adapter *adapter = netdev_priv(netdev);
5084
5085 memset(stats, 0, sizeof(struct rtnl_link_stats64));
5086 spin_lock(&adapter->stats64_lock);
5087 e1000e_update_stats(adapter);
5088 /* Fill out the OS statistics structure */
5089 stats->rx_bytes = adapter->stats.gorc;
5090 stats->rx_packets = adapter->stats.gprc;
5091 stats->tx_bytes = adapter->stats.gotc;
5092 stats->tx_packets = adapter->stats.gptc;
5093 stats->multicast = adapter->stats.mprc;
5094 stats->collisions = adapter->stats.colc;
5095
5096 /* Rx Errors */
5097
5098 /*
5099 * RLEC on some newer hardware can be incorrect so build
5100 * our own version based on RUC and ROC
5101 */
5102 stats->rx_errors = adapter->stats.rxerrc +
5103 adapter->stats.crcerrs + adapter->stats.algnerrc +
5104 adapter->stats.ruc + adapter->stats.roc +
5105 adapter->stats.cexterr;
5106 stats->rx_length_errors = adapter->stats.ruc +
5107 adapter->stats.roc;
5108 stats->rx_crc_errors = adapter->stats.crcerrs;
5109 stats->rx_frame_errors = adapter->stats.algnerrc;
5110 stats->rx_missed_errors = adapter->stats.mpc;
5111
5112 /* Tx Errors */
5113 stats->tx_errors = adapter->stats.ecol +
5114 adapter->stats.latecol;
5115 stats->tx_aborted_errors = adapter->stats.ecol;
5116 stats->tx_window_errors = adapter->stats.latecol;
5117 stats->tx_carrier_errors = adapter->stats.tncrs;
5118
5119 /* Tx Dropped needs to be maintained elsewhere */
5120
5121 spin_unlock(&adapter->stats64_lock);
5122 return stats;
5123 }
5124
5125 /**
5126 * e1000_change_mtu - Change the Maximum Transfer Unit
5127 * @netdev: network interface device structure
5128 * @new_mtu: new value for maximum frame size
5129 *
5130 * Returns 0 on success, negative on failure
5131 **/
5132 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5133 {
5134 struct e1000_adapter *adapter = netdev_priv(netdev);
5135 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5136
5137 /* Jumbo frame support */
5138 if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
5139 !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5140 e_err("Jumbo Frames not supported.\n");
5141 return -EINVAL;
5142 }
5143
5144 /* Supported frame sizes */
5145 if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
5146 (max_frame > adapter->max_hw_frame_size)) {
5147 e_err("Unsupported MTU setting\n");
5148 return -EINVAL;
5149 }
5150
5151 /* Jumbo frame workaround on 82579 requires CRC be stripped */
5152 if ((adapter->hw.mac.type == e1000_pch2lan) &&
5153 !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5154 (new_mtu > ETH_DATA_LEN)) {
5155 e_err("Jumbo Frames not supported on 82579 when CRC stripping is disabled.\n");
5156 return -EINVAL;
5157 }
5158
5159 /* 82573 Errata 17 */
5160 if (((adapter->hw.mac.type == e1000_82573) ||
5161 (adapter->hw.mac.type == e1000_82574)) &&
5162 (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
5163 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
5164 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
5165 }
5166
5167 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5168 usleep_range(1000, 2000);
5169 /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5170 adapter->max_frame_size = max_frame;
5171 e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5172 netdev->mtu = new_mtu;
5173 if (netif_running(netdev))
5174 e1000e_down(adapter);
5175
5176 /*
5177 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5178 * means we reserve 2 more, this pushes us to allocate from the next
5179 * larger slab size.
5180 * i.e. RXBUFFER_2048 --> size-4096 slab
5181 * However with the new *_jumbo_rx* routines, jumbo receives will use
5182 * fragmented skbs
5183 */
5184
5185 if (max_frame <= 2048)
5186 adapter->rx_buffer_len = 2048;
5187 else
5188 adapter->rx_buffer_len = 4096;
5189
5190 /* adjust allocation if LPE protects us, and we aren't using SBP */
5191 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5192 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5193 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5194 + ETH_FCS_LEN;
5195
5196 if (netif_running(netdev))
5197 e1000e_up(adapter);
5198 else
5199 e1000e_reset(adapter);
5200
5201 clear_bit(__E1000_RESETTING, &adapter->state);
5202
5203 return 0;
5204 }
5205
5206 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5207 int cmd)
5208 {
5209 struct e1000_adapter *adapter = netdev_priv(netdev);
5210 struct mii_ioctl_data *data = if_mii(ifr);
5211
5212 if (adapter->hw.phy.media_type != e1000_media_type_copper)
5213 return -EOPNOTSUPP;
5214
5215 switch (cmd) {
5216 case SIOCGMIIPHY:
5217 data->phy_id = adapter->hw.phy.addr;
5218 break;
5219 case SIOCGMIIREG:
5220 e1000_phy_read_status(adapter);
5221
5222 switch (data->reg_num & 0x1F) {
5223 case MII_BMCR:
5224 data->val_out = adapter->phy_regs.bmcr;
5225 break;
5226 case MII_BMSR:
5227 data->val_out = adapter->phy_regs.bmsr;
5228 break;
5229 case MII_PHYSID1:
5230 data->val_out = (adapter->hw.phy.id >> 16);
5231 break;
5232 case MII_PHYSID2:
5233 data->val_out = (adapter->hw.phy.id & 0xFFFF);
5234 break;
5235 case MII_ADVERTISE:
5236 data->val_out = adapter->phy_regs.advertise;
5237 break;
5238 case MII_LPA:
5239 data->val_out = adapter->phy_regs.lpa;
5240 break;
5241 case MII_EXPANSION:
5242 data->val_out = adapter->phy_regs.expansion;
5243 break;
5244 case MII_CTRL1000:
5245 data->val_out = adapter->phy_regs.ctrl1000;
5246 break;
5247 case MII_STAT1000:
5248 data->val_out = adapter->phy_regs.stat1000;
5249 break;
5250 case MII_ESTATUS:
5251 data->val_out = adapter->phy_regs.estatus;
5252 break;
5253 default:
5254 return -EIO;
5255 }
5256 break;
5257 case SIOCSMIIREG:
5258 default:
5259 return -EOPNOTSUPP;
5260 }
5261 return 0;
5262 }
5263
5264 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5265 {
5266 switch (cmd) {
5267 case SIOCGMIIPHY:
5268 case SIOCGMIIREG:
5269 case SIOCSMIIREG:
5270 return e1000_mii_ioctl(netdev, ifr, cmd);
5271 default:
5272 return -EOPNOTSUPP;
5273 }
5274 }
5275
5276 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5277 {
5278 struct e1000_hw *hw = &adapter->hw;
5279 u32 i, mac_reg;
5280 u16 phy_reg, wuc_enable;
5281 int retval = 0;
5282
5283 /* copy MAC RARs to PHY RARs */
5284 e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5285
5286 retval = hw->phy.ops.acquire(hw);
5287 if (retval) {
5288 e_err("Could not acquire PHY\n");
5289 return retval;
5290 }
5291
5292 /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
5293 retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5294 if (retval)
5295 goto out;
5296
5297 /* copy MAC MTA to PHY MTA - only needed for pchlan */
5298 for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5299 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5300 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
5301 (u16)(mac_reg & 0xFFFF));
5302 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
5303 (u16)((mac_reg >> 16) & 0xFFFF));
5304 }
5305
5306 /* configure PHY Rx Control register */
5307 hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
5308 mac_reg = er32(RCTL);
5309 if (mac_reg & E1000_RCTL_UPE)
5310 phy_reg |= BM_RCTL_UPE;
5311 if (mac_reg & E1000_RCTL_MPE)
5312 phy_reg |= BM_RCTL_MPE;
5313 phy_reg &= ~(BM_RCTL_MO_MASK);
5314 if (mac_reg & E1000_RCTL_MO_3)
5315 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5316 << BM_RCTL_MO_SHIFT);
5317 if (mac_reg & E1000_RCTL_BAM)
5318 phy_reg |= BM_RCTL_BAM;
5319 if (mac_reg & E1000_RCTL_PMCF)
5320 phy_reg |= BM_RCTL_PMCF;
5321 mac_reg = er32(CTRL);
5322 if (mac_reg & E1000_CTRL_RFCE)
5323 phy_reg |= BM_RCTL_RFCE;
5324 hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
5325
5326 /* enable PHY wakeup in MAC register */
5327 ew32(WUFC, wufc);
5328 ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5329
5330 /* configure and enable PHY wakeup in PHY registers */
5331 hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
5332 hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5333
5334 /* activate PHY wakeup */
5335 wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5336 retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5337 if (retval)
5338 e_err("Could not set PHY Host Wakeup bit\n");
5339 out:
5340 hw->phy.ops.release(hw);
5341
5342 return retval;
5343 }
5344
5345 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5346 bool runtime)
5347 {
5348 struct net_device *netdev = pci_get_drvdata(pdev);
5349 struct e1000_adapter *adapter = netdev_priv(netdev);
5350 struct e1000_hw *hw = &adapter->hw;
5351 u32 ctrl, ctrl_ext, rctl, status;
5352 /* Runtime suspend should only enable wakeup for link changes */
5353 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5354 int retval = 0;
5355
5356 netif_device_detach(netdev);
5357
5358 if (netif_running(netdev)) {
5359 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5360 e1000e_down(adapter);
5361 e1000_free_irq(adapter);
5362 }
5363 e1000e_reset_interrupt_capability(adapter);
5364
5365 retval = pci_save_state(pdev);
5366 if (retval)
5367 return retval;
5368
5369 status = er32(STATUS);
5370 if (status & E1000_STATUS_LU)
5371 wufc &= ~E1000_WUFC_LNKC;
5372
5373 if (wufc) {
5374 e1000_setup_rctl(adapter);
5375 e1000e_set_rx_mode(netdev);
5376
5377 /* turn on all-multi mode if wake on multicast is enabled */
5378 if (wufc & E1000_WUFC_MC) {
5379 rctl = er32(RCTL);
5380 rctl |= E1000_RCTL_MPE;
5381 ew32(RCTL, rctl);
5382 }
5383
5384 ctrl = er32(CTRL);
5385 /* advertise wake from D3Cold */
5386 #define E1000_CTRL_ADVD3WUC 0x00100000
5387 /* phy power management enable */
5388 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5389 ctrl |= E1000_CTRL_ADVD3WUC;
5390 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5391 ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5392 ew32(CTRL, ctrl);
5393
5394 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5395 adapter->hw.phy.media_type ==
5396 e1000_media_type_internal_serdes) {
5397 /* keep the laser running in D3 */
5398 ctrl_ext = er32(CTRL_EXT);
5399 ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5400 ew32(CTRL_EXT, ctrl_ext);
5401 }
5402
5403 if (adapter->flags & FLAG_IS_ICH)
5404 e1000_suspend_workarounds_ich8lan(&adapter->hw);
5405
5406 /* Allow time for pending master requests to run */
5407 e1000e_disable_pcie_master(&adapter->hw);
5408
5409 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5410 /* enable wakeup by the PHY */
5411 retval = e1000_init_phy_wakeup(adapter, wufc);
5412 if (retval)
5413 return retval;
5414 } else {
5415 /* enable wakeup by the MAC */
5416 ew32(WUFC, wufc);
5417 ew32(WUC, E1000_WUC_PME_EN);
5418 }
5419 } else {
5420 ew32(WUC, 0);
5421 ew32(WUFC, 0);
5422 }
5423
5424 *enable_wake = !!wufc;
5425
5426 /* make sure adapter isn't asleep if manageability is enabled */
5427 if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5428 (hw->mac.ops.check_mng_mode(hw)))
5429 *enable_wake = true;
5430
5431 if (adapter->hw.phy.type == e1000_phy_igp_3)
5432 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5433
5434 /*
5435 * Release control of h/w to f/w. If f/w is AMT enabled, this
5436 * would have already happened in close and is redundant.
5437 */
5438 e1000e_release_hw_control(adapter);
5439
5440 pci_disable_device(pdev);
5441
5442 return 0;
5443 }
5444
5445 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5446 {
5447 if (sleep && wake) {
5448 pci_prepare_to_sleep(pdev);
5449 return;
5450 }
5451
5452 pci_wake_from_d3(pdev, wake);
5453 pci_set_power_state(pdev, PCI_D3hot);
5454 }
5455
5456 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5457 bool wake)
5458 {
5459 struct net_device *netdev = pci_get_drvdata(pdev);
5460 struct e1000_adapter *adapter = netdev_priv(netdev);
5461
5462 /*
5463 * The pci-e switch on some quad port adapters will report a
5464 * correctable error when the MAC transitions from D0 to D3. To
5465 * prevent this we need to mask off the correctable errors on the
5466 * downstream port of the pci-e switch.
5467 */
5468 if (adapter->flags & FLAG_IS_QUAD_PORT) {
5469 struct pci_dev *us_dev = pdev->bus->self;
5470 int pos = pci_pcie_cap(us_dev);
5471 u16 devctl;
5472
5473 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5474 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5475 (devctl & ~PCI_EXP_DEVCTL_CERE));
5476
5477 e1000_power_off(pdev, sleep, wake);
5478
5479 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5480 } else {
5481 e1000_power_off(pdev, sleep, wake);
5482 }
5483 }
5484
5485 #ifdef CONFIG_PCIEASPM
5486 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5487 {
5488 pci_disable_link_state_locked(pdev, state);
5489 }
5490 #else
5491 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5492 {
5493 int pos;
5494 u16 reg16;
5495
5496 /*
5497 * Both device and parent should have the same ASPM setting.
5498 * Disable ASPM in downstream component first and then upstream.
5499 */
5500 pos = pci_pcie_cap(pdev);
5501 pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5502 reg16 &= ~state;
5503 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5504
5505 if (!pdev->bus->self)
5506 return;
5507
5508 pos = pci_pcie_cap(pdev->bus->self);
5509 pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5510 reg16 &= ~state;
5511 pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5512 }
5513 #endif
5514 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5515 {
5516 dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5517 (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5518 (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5519
5520 __e1000e_disable_aspm(pdev, state);
5521 }
5522
5523 #ifdef CONFIG_PM
5524 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5525 {
5526 return !!adapter->tx_ring->buffer_info;
5527 }
5528
5529 static int __e1000_resume(struct pci_dev *pdev)
5530 {
5531 struct net_device *netdev = pci_get_drvdata(pdev);
5532 struct e1000_adapter *adapter = netdev_priv(netdev);
5533 struct e1000_hw *hw = &adapter->hw;
5534 u16 aspm_disable_flag = 0;
5535 u32 err;
5536
5537 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5538 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5539 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5540 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5541 if (aspm_disable_flag)
5542 e1000e_disable_aspm(pdev, aspm_disable_flag);
5543
5544 pci_set_power_state(pdev, PCI_D0);
5545 pci_restore_state(pdev);
5546 pci_save_state(pdev);
5547
5548 e1000e_set_interrupt_capability(adapter);
5549 if (netif_running(netdev)) {
5550 err = e1000_request_irq(adapter);
5551 if (err)
5552 return err;
5553 }
5554
5555 if (hw->mac.type == e1000_pch2lan)
5556 e1000_resume_workarounds_pchlan(&adapter->hw);
5557
5558 e1000e_power_up_phy(adapter);
5559
5560 /* report the system wakeup cause from S3/S4 */
5561 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5562 u16 phy_data;
5563
5564 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5565 if (phy_data) {
5566 e_info("PHY Wakeup cause - %s\n",
5567 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5568 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5569 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5570 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5571 phy_data & E1000_WUS_LNKC ?
5572 "Link Status Change" : "other");
5573 }
5574 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5575 } else {
5576 u32 wus = er32(WUS);
5577 if (wus) {
5578 e_info("MAC Wakeup cause - %s\n",
5579 wus & E1000_WUS_EX ? "Unicast Packet" :
5580 wus & E1000_WUS_MC ? "Multicast Packet" :
5581 wus & E1000_WUS_BC ? "Broadcast Packet" :
5582 wus & E1000_WUS_MAG ? "Magic Packet" :
5583 wus & E1000_WUS_LNKC ? "Link Status Change" :
5584 "other");
5585 }
5586 ew32(WUS, ~0);
5587 }
5588
5589 e1000e_reset(adapter);
5590
5591 e1000_init_manageability_pt(adapter);
5592
5593 if (netif_running(netdev))
5594 e1000e_up(adapter);
5595
5596 netif_device_attach(netdev);
5597
5598 /*
5599 * If the controller has AMT, do not set DRV_LOAD until the interface
5600 * is up. For all other cases, let the f/w know that the h/w is now
5601 * under the control of the driver.
5602 */
5603 if (!(adapter->flags & FLAG_HAS_AMT))
5604 e1000e_get_hw_control(adapter);
5605
5606 return 0;
5607 }
5608
5609 #ifdef CONFIG_PM_SLEEP
5610 static int e1000_suspend(struct device *dev)
5611 {
5612 struct pci_dev *pdev = to_pci_dev(dev);
5613 int retval;
5614 bool wake;
5615
5616 retval = __e1000_shutdown(pdev, &wake, false);
5617 if (!retval)
5618 e1000_complete_shutdown(pdev, true, wake);
5619
5620 return retval;
5621 }
5622
5623 static int e1000_resume(struct device *dev)
5624 {
5625 struct pci_dev *pdev = to_pci_dev(dev);
5626 struct net_device *netdev = pci_get_drvdata(pdev);
5627 struct e1000_adapter *adapter = netdev_priv(netdev);
5628
5629 if (e1000e_pm_ready(adapter))
5630 adapter->idle_check = true;
5631
5632 return __e1000_resume(pdev);
5633 }
5634 #endif /* CONFIG_PM_SLEEP */
5635
5636 #ifdef CONFIG_PM_RUNTIME
5637 static int e1000_runtime_suspend(struct device *dev)
5638 {
5639 struct pci_dev *pdev = to_pci_dev(dev);
5640 struct net_device *netdev = pci_get_drvdata(pdev);
5641 struct e1000_adapter *adapter = netdev_priv(netdev);
5642
5643 if (e1000e_pm_ready(adapter)) {
5644 bool wake;
5645
5646 __e1000_shutdown(pdev, &wake, true);
5647 }
5648
5649 return 0;
5650 }
5651
5652 static int e1000_idle(struct device *dev)
5653 {
5654 struct pci_dev *pdev = to_pci_dev(dev);
5655 struct net_device *netdev = pci_get_drvdata(pdev);
5656 struct e1000_adapter *adapter = netdev_priv(netdev);
5657
5658 if (!e1000e_pm_ready(adapter))
5659 return 0;
5660
5661 if (adapter->idle_check) {
5662 adapter->idle_check = false;
5663 if (!e1000e_has_link(adapter))
5664 pm_schedule_suspend(dev, MSEC_PER_SEC);
5665 }
5666
5667 return -EBUSY;
5668 }
5669
5670 static int e1000_runtime_resume(struct device *dev)
5671 {
5672 struct pci_dev *pdev = to_pci_dev(dev);
5673 struct net_device *netdev = pci_get_drvdata(pdev);
5674 struct e1000_adapter *adapter = netdev_priv(netdev);
5675
5676 if (!e1000e_pm_ready(adapter))
5677 return 0;
5678
5679 adapter->idle_check = !dev->power.runtime_auto;
5680 return __e1000_resume(pdev);
5681 }
5682 #endif /* CONFIG_PM_RUNTIME */
5683 #endif /* CONFIG_PM */
5684
5685 static void e1000_shutdown(struct pci_dev *pdev)
5686 {
5687 bool wake = false;
5688
5689 __e1000_shutdown(pdev, &wake, false);
5690
5691 if (system_state == SYSTEM_POWER_OFF)
5692 e1000_complete_shutdown(pdev, false, wake);
5693 }
5694
5695 #ifdef CONFIG_NET_POLL_CONTROLLER
5696
5697 static irqreturn_t e1000_intr_msix(int irq, void *data)
5698 {
5699 struct net_device *netdev = data;
5700 struct e1000_adapter *adapter = netdev_priv(netdev);
5701
5702 if (adapter->msix_entries) {
5703 int vector, msix_irq;
5704
5705 vector = 0;
5706 msix_irq = adapter->msix_entries[vector].vector;
5707 disable_irq(msix_irq);
5708 e1000_intr_msix_rx(msix_irq, netdev);
5709 enable_irq(msix_irq);
5710
5711 vector++;
5712 msix_irq = adapter->msix_entries[vector].vector;
5713 disable_irq(msix_irq);
5714 e1000_intr_msix_tx(msix_irq, netdev);
5715 enable_irq(msix_irq);
5716
5717 vector++;
5718 msix_irq = adapter->msix_entries[vector].vector;
5719 disable_irq(msix_irq);
5720 e1000_msix_other(msix_irq, netdev);
5721 enable_irq(msix_irq);
5722 }
5723
5724 return IRQ_HANDLED;
5725 }
5726
5727 /*
5728 * Polling 'interrupt' - used by things like netconsole to send skbs
5729 * without having to re-enable interrupts. It's not called while
5730 * the interrupt routine is executing.
5731 */
5732 static void e1000_netpoll(struct net_device *netdev)
5733 {
5734 struct e1000_adapter *adapter = netdev_priv(netdev);
5735
5736 switch (adapter->int_mode) {
5737 case E1000E_INT_MODE_MSIX:
5738 e1000_intr_msix(adapter->pdev->irq, netdev);
5739 break;
5740 case E1000E_INT_MODE_MSI:
5741 disable_irq(adapter->pdev->irq);
5742 e1000_intr_msi(adapter->pdev->irq, netdev);
5743 enable_irq(adapter->pdev->irq);
5744 break;
5745 default: /* E1000E_INT_MODE_LEGACY */
5746 disable_irq(adapter->pdev->irq);
5747 e1000_intr(adapter->pdev->irq, netdev);
5748 enable_irq(adapter->pdev->irq);
5749 break;
5750 }
5751 }
5752 #endif
5753
5754 /**
5755 * e1000_io_error_detected - called when PCI error is detected
5756 * @pdev: Pointer to PCI device
5757 * @state: The current pci connection state
5758 *
5759 * This function is called after a PCI bus error affecting
5760 * this device has been detected.
5761 */
5762 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5763 pci_channel_state_t state)
5764 {
5765 struct net_device *netdev = pci_get_drvdata(pdev);
5766 struct e1000_adapter *adapter = netdev_priv(netdev);
5767
5768 netif_device_detach(netdev);
5769
5770 if (state == pci_channel_io_perm_failure)
5771 return PCI_ERS_RESULT_DISCONNECT;
5772
5773 if (netif_running(netdev))
5774 e1000e_down(adapter);
5775 pci_disable_device(pdev);
5776
5777 /* Request a slot slot reset. */
5778 return PCI_ERS_RESULT_NEED_RESET;
5779 }
5780
5781 /**
5782 * e1000_io_slot_reset - called after the pci bus has been reset.
5783 * @pdev: Pointer to PCI device
5784 *
5785 * Restart the card from scratch, as if from a cold-boot. Implementation
5786 * resembles the first-half of the e1000_resume routine.
5787 */
5788 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5789 {
5790 struct net_device *netdev = pci_get_drvdata(pdev);
5791 struct e1000_adapter *adapter = netdev_priv(netdev);
5792 struct e1000_hw *hw = &adapter->hw;
5793 u16 aspm_disable_flag = 0;
5794 int err;
5795 pci_ers_result_t result;
5796
5797 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5798 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5799 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5800 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5801 if (aspm_disable_flag)
5802 e1000e_disable_aspm(pdev, aspm_disable_flag);
5803
5804 err = pci_enable_device_mem(pdev);
5805 if (err) {
5806 dev_err(&pdev->dev,
5807 "Cannot re-enable PCI device after reset.\n");
5808 result = PCI_ERS_RESULT_DISCONNECT;
5809 } else {
5810 pci_set_master(pdev);
5811 pdev->state_saved = true;
5812 pci_restore_state(pdev);
5813
5814 pci_enable_wake(pdev, PCI_D3hot, 0);
5815 pci_enable_wake(pdev, PCI_D3cold, 0);
5816
5817 e1000e_reset(adapter);
5818 ew32(WUS, ~0);
5819 result = PCI_ERS_RESULT_RECOVERED;
5820 }
5821
5822 pci_cleanup_aer_uncorrect_error_status(pdev);
5823
5824 return result;
5825 }
5826
5827 /**
5828 * e1000_io_resume - called when traffic can start flowing again.
5829 * @pdev: Pointer to PCI device
5830 *
5831 * This callback is called when the error recovery driver tells us that
5832 * its OK to resume normal operation. Implementation resembles the
5833 * second-half of the e1000_resume routine.
5834 */
5835 static void e1000_io_resume(struct pci_dev *pdev)
5836 {
5837 struct net_device *netdev = pci_get_drvdata(pdev);
5838 struct e1000_adapter *adapter = netdev_priv(netdev);
5839
5840 e1000_init_manageability_pt(adapter);
5841
5842 if (netif_running(netdev)) {
5843 if (e1000e_up(adapter)) {
5844 dev_err(&pdev->dev,
5845 "can't bring device back up after reset\n");
5846 return;
5847 }
5848 }
5849
5850 netif_device_attach(netdev);
5851
5852 /*
5853 * If the controller has AMT, do not set DRV_LOAD until the interface
5854 * is up. For all other cases, let the f/w know that the h/w is now
5855 * under the control of the driver.
5856 */
5857 if (!(adapter->flags & FLAG_HAS_AMT))
5858 e1000e_get_hw_control(adapter);
5859
5860 }
5861
5862 static void e1000_print_device_info(struct e1000_adapter *adapter)
5863 {
5864 struct e1000_hw *hw = &adapter->hw;
5865 struct net_device *netdev = adapter->netdev;
5866 u32 ret_val;
5867 u8 pba_str[E1000_PBANUM_LENGTH];
5868
5869 /* print bus type/speed/width info */
5870 e_info("(PCI Express:2.5GT/s:%s) %pM\n",
5871 /* bus width */
5872 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5873 "Width x1"),
5874 /* MAC address */
5875 netdev->dev_addr);
5876 e_info("Intel(R) PRO/%s Network Connection\n",
5877 (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5878 ret_val = e1000_read_pba_string_generic(hw, pba_str,
5879 E1000_PBANUM_LENGTH);
5880 if (ret_val)
5881 strncpy((char *)pba_str, "Unknown", sizeof(pba_str) - 1);
5882 e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5883 hw->mac.type, hw->phy.type, pba_str);
5884 }
5885
5886 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5887 {
5888 struct e1000_hw *hw = &adapter->hw;
5889 int ret_val;
5890 u16 buf = 0;
5891
5892 if (hw->mac.type != e1000_82573)
5893 return;
5894
5895 ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5896 if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5897 /* Deep Smart Power Down (DSPD) */
5898 dev_warn(&adapter->pdev->dev,
5899 "Warning: detected DSPD enabled in EEPROM\n");
5900 }
5901 }
5902
5903 static int e1000_set_features(struct net_device *netdev,
5904 netdev_features_t features)
5905 {
5906 struct e1000_adapter *adapter = netdev_priv(netdev);
5907 netdev_features_t changed = features ^ netdev->features;
5908
5909 if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
5910 adapter->flags |= FLAG_TSO_FORCE;
5911
5912 if (!(changed & (NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX |
5913 NETIF_F_RXCSUM)))
5914 return 0;
5915
5916 if (netif_running(netdev))
5917 e1000e_reinit_locked(adapter);
5918 else
5919 e1000e_reset(adapter);
5920
5921 return 0;
5922 }
5923
5924 static const struct net_device_ops e1000e_netdev_ops = {
5925 .ndo_open = e1000_open,
5926 .ndo_stop = e1000_close,
5927 .ndo_start_xmit = e1000_xmit_frame,
5928 .ndo_get_stats64 = e1000e_get_stats64,
5929 .ndo_set_rx_mode = e1000e_set_rx_mode,
5930 .ndo_set_mac_address = e1000_set_mac,
5931 .ndo_change_mtu = e1000_change_mtu,
5932 .ndo_do_ioctl = e1000_ioctl,
5933 .ndo_tx_timeout = e1000_tx_timeout,
5934 .ndo_validate_addr = eth_validate_addr,
5935
5936 .ndo_vlan_rx_add_vid = e1000_vlan_rx_add_vid,
5937 .ndo_vlan_rx_kill_vid = e1000_vlan_rx_kill_vid,
5938 #ifdef CONFIG_NET_POLL_CONTROLLER
5939 .ndo_poll_controller = e1000_netpoll,
5940 #endif
5941 .ndo_set_features = e1000_set_features,
5942 };
5943
5944 /**
5945 * e1000_probe - Device Initialization Routine
5946 * @pdev: PCI device information struct
5947 * @ent: entry in e1000_pci_tbl
5948 *
5949 * Returns 0 on success, negative on failure
5950 *
5951 * e1000_probe initializes an adapter identified by a pci_dev structure.
5952 * The OS initialization, configuring of the adapter private structure,
5953 * and a hardware reset occur.
5954 **/
5955 static int __devinit e1000_probe(struct pci_dev *pdev,
5956 const struct pci_device_id *ent)
5957 {
5958 struct net_device *netdev;
5959 struct e1000_adapter *adapter;
5960 struct e1000_hw *hw;
5961 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5962 resource_size_t mmio_start, mmio_len;
5963 resource_size_t flash_start, flash_len;
5964
5965 static int cards_found;
5966 u16 aspm_disable_flag = 0;
5967 int i, err, pci_using_dac;
5968 u16 eeprom_data = 0;
5969 u16 eeprom_apme_mask = E1000_EEPROM_APME;
5970
5971 if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
5972 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5973 if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
5974 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5975 if (aspm_disable_flag)
5976 e1000e_disable_aspm(pdev, aspm_disable_flag);
5977
5978 err = pci_enable_device_mem(pdev);
5979 if (err)
5980 return err;
5981
5982 pci_using_dac = 0;
5983 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
5984 if (!err) {
5985 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
5986 if (!err)
5987 pci_using_dac = 1;
5988 } else {
5989 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
5990 if (err) {
5991 err = dma_set_coherent_mask(&pdev->dev,
5992 DMA_BIT_MASK(32));
5993 if (err) {
5994 dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
5995 goto err_dma;
5996 }
5997 }
5998 }
5999
6000 err = pci_request_selected_regions_exclusive(pdev,
6001 pci_select_bars(pdev, IORESOURCE_MEM),
6002 e1000e_driver_name);
6003 if (err)
6004 goto err_pci_reg;
6005
6006 /* AER (Advanced Error Reporting) hooks */
6007 pci_enable_pcie_error_reporting(pdev);
6008
6009 pci_set_master(pdev);
6010 /* PCI config space info */
6011 err = pci_save_state(pdev);
6012 if (err)
6013 goto err_alloc_etherdev;
6014
6015 err = -ENOMEM;
6016 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
6017 if (!netdev)
6018 goto err_alloc_etherdev;
6019
6020 SET_NETDEV_DEV(netdev, &pdev->dev);
6021
6022 netdev->irq = pdev->irq;
6023
6024 pci_set_drvdata(pdev, netdev);
6025 adapter = netdev_priv(netdev);
6026 hw = &adapter->hw;
6027 adapter->netdev = netdev;
6028 adapter->pdev = pdev;
6029 adapter->ei = ei;
6030 adapter->pba = ei->pba;
6031 adapter->flags = ei->flags;
6032 adapter->flags2 = ei->flags2;
6033 adapter->hw.adapter = adapter;
6034 adapter->hw.mac.type = ei->mac;
6035 adapter->max_hw_frame_size = ei->max_hw_frame_size;
6036 adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
6037
6038 mmio_start = pci_resource_start(pdev, 0);
6039 mmio_len = pci_resource_len(pdev, 0);
6040
6041 err = -EIO;
6042 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
6043 if (!adapter->hw.hw_addr)
6044 goto err_ioremap;
6045
6046 if ((adapter->flags & FLAG_HAS_FLASH) &&
6047 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
6048 flash_start = pci_resource_start(pdev, 1);
6049 flash_len = pci_resource_len(pdev, 1);
6050 adapter->hw.flash_address = ioremap(flash_start, flash_len);
6051 if (!adapter->hw.flash_address)
6052 goto err_flashmap;
6053 }
6054
6055 /* construct the net_device struct */
6056 netdev->netdev_ops = &e1000e_netdev_ops;
6057 e1000e_set_ethtool_ops(netdev);
6058 netdev->watchdog_timeo = 5 * HZ;
6059 netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
6060 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
6061
6062 netdev->mem_start = mmio_start;
6063 netdev->mem_end = mmio_start + mmio_len;
6064
6065 adapter->bd_number = cards_found++;
6066
6067 e1000e_check_options(adapter);
6068
6069 /* setup adapter struct */
6070 err = e1000_sw_init(adapter);
6071 if (err)
6072 goto err_sw_init;
6073
6074 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
6075 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
6076 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
6077
6078 err = ei->get_variants(adapter);
6079 if (err)
6080 goto err_hw_init;
6081
6082 if ((adapter->flags & FLAG_IS_ICH) &&
6083 (adapter->flags & FLAG_READ_ONLY_NVM))
6084 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
6085
6086 hw->mac.ops.get_bus_info(&adapter->hw);
6087
6088 adapter->hw.phy.autoneg_wait_to_complete = 0;
6089
6090 /* Copper options */
6091 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
6092 adapter->hw.phy.mdix = AUTO_ALL_MODES;
6093 adapter->hw.phy.disable_polarity_correction = 0;
6094 adapter->hw.phy.ms_type = e1000_ms_hw_default;
6095 }
6096
6097 if (e1000_check_reset_block(&adapter->hw))
6098 e_info("PHY reset is blocked due to SOL/IDER session.\n");
6099
6100 /* Set initial default active device features */
6101 netdev->features = (NETIF_F_SG |
6102 NETIF_F_HW_VLAN_RX |
6103 NETIF_F_HW_VLAN_TX |
6104 NETIF_F_TSO |
6105 NETIF_F_TSO6 |
6106 NETIF_F_RXCSUM |
6107 NETIF_F_HW_CSUM);
6108
6109 /* Set user-changeable features (subset of all device features) */
6110 netdev->hw_features = netdev->features;
6111
6112 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
6113 netdev->features |= NETIF_F_HW_VLAN_FILTER;
6114
6115 netdev->vlan_features |= (NETIF_F_SG |
6116 NETIF_F_TSO |
6117 NETIF_F_TSO6 |
6118 NETIF_F_HW_CSUM);
6119
6120 netdev->priv_flags |= IFF_UNICAST_FLT;
6121
6122 if (pci_using_dac) {
6123 netdev->features |= NETIF_F_HIGHDMA;
6124 netdev->vlan_features |= NETIF_F_HIGHDMA;
6125 }
6126
6127 if (e1000e_enable_mng_pass_thru(&adapter->hw))
6128 adapter->flags |= FLAG_MNG_PT_ENABLED;
6129
6130 /*
6131 * before reading the NVM, reset the controller to
6132 * put the device in a known good starting state
6133 */
6134 adapter->hw.mac.ops.reset_hw(&adapter->hw);
6135
6136 /*
6137 * systems with ASPM and others may see the checksum fail on the first
6138 * attempt. Let's give it a few tries
6139 */
6140 for (i = 0;; i++) {
6141 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
6142 break;
6143 if (i == 2) {
6144 e_err("The NVM Checksum Is Not Valid\n");
6145 err = -EIO;
6146 goto err_eeprom;
6147 }
6148 }
6149
6150 e1000_eeprom_checks(adapter);
6151
6152 /* copy the MAC address */
6153 if (e1000e_read_mac_addr(&adapter->hw))
6154 e_err("NVM Read Error while reading MAC address\n");
6155
6156 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
6157 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
6158
6159 if (!is_valid_ether_addr(netdev->perm_addr)) {
6160 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
6161 err = -EIO;
6162 goto err_eeprom;
6163 }
6164
6165 init_timer(&adapter->watchdog_timer);
6166 adapter->watchdog_timer.function = e1000_watchdog;
6167 adapter->watchdog_timer.data = (unsigned long) adapter;
6168
6169 init_timer(&adapter->phy_info_timer);
6170 adapter->phy_info_timer.function = e1000_update_phy_info;
6171 adapter->phy_info_timer.data = (unsigned long) adapter;
6172
6173 INIT_WORK(&adapter->reset_task, e1000_reset_task);
6174 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
6175 INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
6176 INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
6177 INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
6178
6179 /* Initialize link parameters. User can change them with ethtool */
6180 adapter->hw.mac.autoneg = 1;
6181 adapter->fc_autoneg = 1;
6182 adapter->hw.fc.requested_mode = e1000_fc_default;
6183 adapter->hw.fc.current_mode = e1000_fc_default;
6184 adapter->hw.phy.autoneg_advertised = 0x2f;
6185
6186 /* ring size defaults */
6187 adapter->rx_ring->count = 256;
6188 adapter->tx_ring->count = 256;
6189
6190 /*
6191 * Initial Wake on LAN setting - If APM wake is enabled in
6192 * the EEPROM, enable the ACPI Magic Packet filter
6193 */
6194 if (adapter->flags & FLAG_APME_IN_WUC) {
6195 /* APME bit in EEPROM is mapped to WUC.APME */
6196 eeprom_data = er32(WUC);
6197 eeprom_apme_mask = E1000_WUC_APME;
6198 if ((hw->mac.type > e1000_ich10lan) &&
6199 (eeprom_data & E1000_WUC_PHY_WAKE))
6200 adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
6201 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
6202 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
6203 (adapter->hw.bus.func == 1))
6204 e1000_read_nvm(&adapter->hw,
6205 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
6206 else
6207 e1000_read_nvm(&adapter->hw,
6208 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
6209 }
6210
6211 /* fetch WoL from EEPROM */
6212 if (eeprom_data & eeprom_apme_mask)
6213 adapter->eeprom_wol |= E1000_WUFC_MAG;
6214
6215 /*
6216 * now that we have the eeprom settings, apply the special cases
6217 * where the eeprom may be wrong or the board simply won't support
6218 * wake on lan on a particular port
6219 */
6220 if (!(adapter->flags & FLAG_HAS_WOL))
6221 adapter->eeprom_wol = 0;
6222
6223 /* initialize the wol settings based on the eeprom settings */
6224 adapter->wol = adapter->eeprom_wol;
6225 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6226
6227 /* save off EEPROM version number */
6228 e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6229
6230 /* reset the hardware with the new settings */
6231 e1000e_reset(adapter);
6232
6233 /*
6234 * If the controller has AMT, do not set DRV_LOAD until the interface
6235 * is up. For all other cases, let the f/w know that the h/w is now
6236 * under the control of the driver.
6237 */
6238 if (!(adapter->flags & FLAG_HAS_AMT))
6239 e1000e_get_hw_control(adapter);
6240
6241 strncpy(netdev->name, "eth%d", sizeof(netdev->name) - 1);
6242 err = register_netdev(netdev);
6243 if (err)
6244 goto err_register;
6245
6246 /* carrier off reporting is important to ethtool even BEFORE open */
6247 netif_carrier_off(netdev);
6248
6249 e1000_print_device_info(adapter);
6250
6251 if (pci_dev_run_wake(pdev))
6252 pm_runtime_put_noidle(&pdev->dev);
6253
6254 return 0;
6255
6256 err_register:
6257 if (!(adapter->flags & FLAG_HAS_AMT))
6258 e1000e_release_hw_control(adapter);
6259 err_eeprom:
6260 if (!e1000_check_reset_block(&adapter->hw))
6261 e1000_phy_hw_reset(&adapter->hw);
6262 err_hw_init:
6263 kfree(adapter->tx_ring);
6264 kfree(adapter->rx_ring);
6265 err_sw_init:
6266 if (adapter->hw.flash_address)
6267 iounmap(adapter->hw.flash_address);
6268 e1000e_reset_interrupt_capability(adapter);
6269 err_flashmap:
6270 iounmap(adapter->hw.hw_addr);
6271 err_ioremap:
6272 free_netdev(netdev);
6273 err_alloc_etherdev:
6274 pci_release_selected_regions(pdev,
6275 pci_select_bars(pdev, IORESOURCE_MEM));
6276 err_pci_reg:
6277 err_dma:
6278 pci_disable_device(pdev);
6279 return err;
6280 }
6281
6282 /**
6283 * e1000_remove - Device Removal Routine
6284 * @pdev: PCI device information struct
6285 *
6286 * e1000_remove is called by the PCI subsystem to alert the driver
6287 * that it should release a PCI device. The could be caused by a
6288 * Hot-Plug event, or because the driver is going to be removed from
6289 * memory.
6290 **/
6291 static void __devexit e1000_remove(struct pci_dev *pdev)
6292 {
6293 struct net_device *netdev = pci_get_drvdata(pdev);
6294 struct e1000_adapter *adapter = netdev_priv(netdev);
6295 bool down = test_bit(__E1000_DOWN, &adapter->state);
6296
6297 /*
6298 * The timers may be rescheduled, so explicitly disable them
6299 * from being rescheduled.
6300 */
6301 if (!down)
6302 set_bit(__E1000_DOWN, &adapter->state);
6303 del_timer_sync(&adapter->watchdog_timer);
6304 del_timer_sync(&adapter->phy_info_timer);
6305
6306 cancel_work_sync(&adapter->reset_task);
6307 cancel_work_sync(&adapter->watchdog_task);
6308 cancel_work_sync(&adapter->downshift_task);
6309 cancel_work_sync(&adapter->update_phy_task);
6310 cancel_work_sync(&adapter->print_hang_task);
6311
6312 if (!(netdev->flags & IFF_UP))
6313 e1000_power_down_phy(adapter);
6314
6315 /* Don't lie to e1000_close() down the road. */
6316 if (!down)
6317 clear_bit(__E1000_DOWN, &adapter->state);
6318 unregister_netdev(netdev);
6319
6320 if (pci_dev_run_wake(pdev))
6321 pm_runtime_get_noresume(&pdev->dev);
6322
6323 /*
6324 * Release control of h/w to f/w. If f/w is AMT enabled, this
6325 * would have already happened in close and is redundant.
6326 */
6327 e1000e_release_hw_control(adapter);
6328
6329 e1000e_reset_interrupt_capability(adapter);
6330 kfree(adapter->tx_ring);
6331 kfree(adapter->rx_ring);
6332
6333 iounmap(adapter->hw.hw_addr);
6334 if (adapter->hw.flash_address)
6335 iounmap(adapter->hw.flash_address);
6336 pci_release_selected_regions(pdev,
6337 pci_select_bars(pdev, IORESOURCE_MEM));
6338
6339 free_netdev(netdev);
6340
6341 /* AER disable */
6342 pci_disable_pcie_error_reporting(pdev);
6343
6344 pci_disable_device(pdev);
6345 }
6346
6347 /* PCI Error Recovery (ERS) */
6348 static struct pci_error_handlers e1000_err_handler = {
6349 .error_detected = e1000_io_error_detected,
6350 .slot_reset = e1000_io_slot_reset,
6351 .resume = e1000_io_resume,
6352 };
6353
6354 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6355 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6356 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6357 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6358 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6359 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6360 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6361 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6362 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6363 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6364
6365 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6366 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6367 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6368 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6369
6370 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6371 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6372 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6373
6374 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6375 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6376 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6377
6378 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6379 board_80003es2lan },
6380 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6381 board_80003es2lan },
6382 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6383 board_80003es2lan },
6384 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6385 board_80003es2lan },
6386
6387 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6388 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6389 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6390 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6391 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6392 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6393 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6394 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6395
6396 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6397 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6398 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6399 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6400 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6401 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6402 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6403 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6404 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6405
6406 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6407 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6408 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6409
6410 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6411 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6412 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6413
6414 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6415 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6416 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6417 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6418
6419 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6420 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6421
6422 { } /* terminate list */
6423 };
6424 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6425
6426 #ifdef CONFIG_PM
6427 static const struct dev_pm_ops e1000_pm_ops = {
6428 SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6429 SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6430 e1000_runtime_resume, e1000_idle)
6431 };
6432 #endif
6433
6434 /* PCI Device API Driver */
6435 static struct pci_driver e1000_driver = {
6436 .name = e1000e_driver_name,
6437 .id_table = e1000_pci_tbl,
6438 .probe = e1000_probe,
6439 .remove = __devexit_p(e1000_remove),
6440 #ifdef CONFIG_PM
6441 .driver.pm = &e1000_pm_ops,
6442 #endif
6443 .shutdown = e1000_shutdown,
6444 .err_handler = &e1000_err_handler
6445 };
6446
6447 /**
6448 * e1000_init_module - Driver Registration Routine
6449 *
6450 * e1000_init_module is the first routine called when the driver is
6451 * loaded. All it does is register with the PCI subsystem.
6452 **/
6453 static int __init e1000_init_module(void)
6454 {
6455 int ret;
6456 pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6457 e1000e_driver_version);
6458 pr_info("Copyright(c) 1999 - 2011 Intel Corporation.\n");
6459 ret = pci_register_driver(&e1000_driver);
6460
6461 return ret;
6462 }
6463 module_init(e1000_init_module);
6464
6465 /**
6466 * e1000_exit_module - Driver Exit Cleanup Routine
6467 *
6468 * e1000_exit_module is called just before the driver is removed
6469 * from memory.
6470 **/
6471 static void __exit e1000_exit_module(void)
6472 {
6473 pci_unregister_driver(&e1000_driver);
6474 }
6475 module_exit(e1000_exit_module);
6476
6477
6478 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6479 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6480 MODULE_LICENSE("GPL");
6481 MODULE_VERSION(DRV_VERSION);
6482
6483 /* e1000_main.c */
Linux kernel - Deliverables
This page took 0.172364 seconds and 5 git commands to generate.