Loading...
1// SPDX-License-Identifier: GPL-2.0
2/* Intel PRO/1000 Linux driver
3 * Copyright(c) 1999 - 2015 Intel Corporation.
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 *
14 * The full GNU General Public License is included in this distribution in
15 * the file called "COPYING".
16 *
17 * Contact Information:
18 * Linux NICS <linux.nics@intel.com>
19 * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
20 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
21 */
22
23#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
24
25#include <linux/module.h>
26#include <linux/types.h>
27#include <linux/init.h>
28#include <linux/pci.h>
29#include <linux/vmalloc.h>
30#include <linux/pagemap.h>
31#include <linux/delay.h>
32#include <linux/netdevice.h>
33#include <linux/interrupt.h>
34#include <linux/tcp.h>
35#include <linux/ipv6.h>
36#include <linux/slab.h>
37#include <net/checksum.h>
38#include <net/ip6_checksum.h>
39#include <linux/ethtool.h>
40#include <linux/if_vlan.h>
41#include <linux/cpu.h>
42#include <linux/smp.h>
43#include <linux/pm_qos.h>
44#include <linux/pm_runtime.h>
45#include <linux/aer.h>
46#include <linux/prefetch.h>
47
48#include "e1000.h"
49
50#define DRV_EXTRAVERSION "-k"
51
52#define DRV_VERSION "3.2.6" DRV_EXTRAVERSION
53char e1000e_driver_name[] = "e1000e";
54const char e1000e_driver_version[] = DRV_VERSION;
55
56#define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
57static int debug = -1;
58module_param(debug, int, 0);
59MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
60
61static const struct e1000_info *e1000_info_tbl[] = {
62 [board_82571] = &e1000_82571_info,
63 [board_82572] = &e1000_82572_info,
64 [board_82573] = &e1000_82573_info,
65 [board_82574] = &e1000_82574_info,
66 [board_82583] = &e1000_82583_info,
67 [board_80003es2lan] = &e1000_es2_info,
68 [board_ich8lan] = &e1000_ich8_info,
69 [board_ich9lan] = &e1000_ich9_info,
70 [board_ich10lan] = &e1000_ich10_info,
71 [board_pchlan] = &e1000_pch_info,
72 [board_pch2lan] = &e1000_pch2_info,
73 [board_pch_lpt] = &e1000_pch_lpt_info,
74 [board_pch_spt] = &e1000_pch_spt_info,
75 [board_pch_cnp] = &e1000_pch_cnp_info,
76};
77
78struct e1000_reg_info {
79 u32 ofs;
80 char *name;
81};
82
83static const struct e1000_reg_info e1000_reg_info_tbl[] = {
84 /* General Registers */
85 {E1000_CTRL, "CTRL"},
86 {E1000_STATUS, "STATUS"},
87 {E1000_CTRL_EXT, "CTRL_EXT"},
88
89 /* Interrupt Registers */
90 {E1000_ICR, "ICR"},
91
92 /* Rx Registers */
93 {E1000_RCTL, "RCTL"},
94 {E1000_RDLEN(0), "RDLEN"},
95 {E1000_RDH(0), "RDH"},
96 {E1000_RDT(0), "RDT"},
97 {E1000_RDTR, "RDTR"},
98 {E1000_RXDCTL(0), "RXDCTL"},
99 {E1000_ERT, "ERT"},
100 {E1000_RDBAL(0), "RDBAL"},
101 {E1000_RDBAH(0), "RDBAH"},
102 {E1000_RDFH, "RDFH"},
103 {E1000_RDFT, "RDFT"},
104 {E1000_RDFHS, "RDFHS"},
105 {E1000_RDFTS, "RDFTS"},
106 {E1000_RDFPC, "RDFPC"},
107
108 /* Tx Registers */
109 {E1000_TCTL, "TCTL"},
110 {E1000_TDBAL(0), "TDBAL"},
111 {E1000_TDBAH(0), "TDBAH"},
112 {E1000_TDLEN(0), "TDLEN"},
113 {E1000_TDH(0), "TDH"},
114 {E1000_TDT(0), "TDT"},
115 {E1000_TIDV, "TIDV"},
116 {E1000_TXDCTL(0), "TXDCTL"},
117 {E1000_TADV, "TADV"},
118 {E1000_TARC(0), "TARC"},
119 {E1000_TDFH, "TDFH"},
120 {E1000_TDFT, "TDFT"},
121 {E1000_TDFHS, "TDFHS"},
122 {E1000_TDFTS, "TDFTS"},
123 {E1000_TDFPC, "TDFPC"},
124
125 /* List Terminator */
126 {0, NULL}
127};
128
129/**
130 * __ew32_prepare - prepare to write to MAC CSR register on certain parts
131 * @hw: pointer to the HW structure
132 *
133 * When updating the MAC CSR registers, the Manageability Engine (ME) could
134 * be accessing the registers at the same time. Normally, this is handled in
135 * h/w by an arbiter but on some parts there is a bug that acknowledges Host
136 * accesses later than it should which could result in the register to have
137 * an incorrect value. Workaround this by checking the FWSM register which
138 * has bit 24 set while ME is accessing MAC CSR registers, wait if it is set
139 * and try again a number of times.
140 **/
141s32 __ew32_prepare(struct e1000_hw *hw)
142{
143 s32 i = E1000_ICH_FWSM_PCIM2PCI_COUNT;
144
145 while ((er32(FWSM) & E1000_ICH_FWSM_PCIM2PCI) && --i)
146 udelay(50);
147
148 return i;
149}
150
151void __ew32(struct e1000_hw *hw, unsigned long reg, u32 val)
152{
153 if (hw->adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
154 __ew32_prepare(hw);
155
156 writel(val, hw->hw_addr + reg);
157}
158
159/**
160 * e1000_regdump - register printout routine
161 * @hw: pointer to the HW structure
162 * @reginfo: pointer to the register info table
163 **/
164static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
165{
166 int n = 0;
167 char rname[16];
168 u32 regs[8];
169
170 switch (reginfo->ofs) {
171 case E1000_RXDCTL(0):
172 for (n = 0; n < 2; n++)
173 regs[n] = __er32(hw, E1000_RXDCTL(n));
174 break;
175 case E1000_TXDCTL(0):
176 for (n = 0; n < 2; n++)
177 regs[n] = __er32(hw, E1000_TXDCTL(n));
178 break;
179 case E1000_TARC(0):
180 for (n = 0; n < 2; n++)
181 regs[n] = __er32(hw, E1000_TARC(n));
182 break;
183 default:
184 pr_info("%-15s %08x\n",
185 reginfo->name, __er32(hw, reginfo->ofs));
186 return;
187 }
188
189 snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
190 pr_info("%-15s %08x %08x\n", rname, regs[0], regs[1]);
191}
192
193static void e1000e_dump_ps_pages(struct e1000_adapter *adapter,
194 struct e1000_buffer *bi)
195{
196 int i;
197 struct e1000_ps_page *ps_page;
198
199 for (i = 0; i < adapter->rx_ps_pages; i++) {
200 ps_page = &bi->ps_pages[i];
201
202 if (ps_page->page) {
203 pr_info("packet dump for ps_page %d:\n", i);
204 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
205 16, 1, page_address(ps_page->page),
206 PAGE_SIZE, true);
207 }
208 }
209}
210
211/**
212 * e1000e_dump - Print registers, Tx-ring and Rx-ring
213 * @adapter: board private structure
214 **/
215static void e1000e_dump(struct e1000_adapter *adapter)
216{
217 struct net_device *netdev = adapter->netdev;
218 struct e1000_hw *hw = &adapter->hw;
219 struct e1000_reg_info *reginfo;
220 struct e1000_ring *tx_ring = adapter->tx_ring;
221 struct e1000_tx_desc *tx_desc;
222 struct my_u0 {
223 __le64 a;
224 __le64 b;
225 } *u0;
226 struct e1000_buffer *buffer_info;
227 struct e1000_ring *rx_ring = adapter->rx_ring;
228 union e1000_rx_desc_packet_split *rx_desc_ps;
229 union e1000_rx_desc_extended *rx_desc;
230 struct my_u1 {
231 __le64 a;
232 __le64 b;
233 __le64 c;
234 __le64 d;
235 } *u1;
236 u32 staterr;
237 int i = 0;
238
239 if (!netif_msg_hw(adapter))
240 return;
241
242 /* Print netdevice Info */
243 if (netdev) {
244 dev_info(&adapter->pdev->dev, "Net device Info\n");
245 pr_info("Device Name state trans_start\n");
246 pr_info("%-15s %016lX %016lX\n", netdev->name,
247 netdev->state, dev_trans_start(netdev));
248 }
249
250 /* Print Registers */
251 dev_info(&adapter->pdev->dev, "Register Dump\n");
252 pr_info(" Register Name Value\n");
253 for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
254 reginfo->name; reginfo++) {
255 e1000_regdump(hw, reginfo);
256 }
257
258 /* Print Tx Ring Summary */
259 if (!netdev || !netif_running(netdev))
260 return;
261
262 dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
263 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
264 buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
265 pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
266 0, tx_ring->next_to_use, tx_ring->next_to_clean,
267 (unsigned long long)buffer_info->dma,
268 buffer_info->length,
269 buffer_info->next_to_watch,
270 (unsigned long long)buffer_info->time_stamp);
271
272 /* Print Tx Ring */
273 if (!netif_msg_tx_done(adapter))
274 goto rx_ring_summary;
275
276 dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
277
278 /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
279 *
280 * Legacy Transmit Descriptor
281 * +--------------------------------------------------------------+
282 * 0 | Buffer Address [63:0] (Reserved on Write Back) |
283 * +--------------------------------------------------------------+
284 * 8 | Special | CSS | Status | CMD | CSO | Length |
285 * +--------------------------------------------------------------+
286 * 63 48 47 36 35 32 31 24 23 16 15 0
287 *
288 * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
289 * 63 48 47 40 39 32 31 16 15 8 7 0
290 * +----------------------------------------------------------------+
291 * 0 | TUCSE | TUCS0 | TUCSS | IPCSE | IPCS0 | IPCSS |
292 * +----------------------------------------------------------------+
293 * 8 | MSS | HDRLEN | RSV | STA | TUCMD | DTYP | PAYLEN |
294 * +----------------------------------------------------------------+
295 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
296 *
297 * Extended Data Descriptor (DTYP=0x1)
298 * +----------------------------------------------------------------+
299 * 0 | Buffer Address [63:0] |
300 * +----------------------------------------------------------------+
301 * 8 | VLAN tag | POPTS | Rsvd | Status | Command | DTYP | DTALEN |
302 * +----------------------------------------------------------------+
303 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
304 */
305 pr_info("Tl[desc] [address 63:0 ] [SpeCssSCmCsLen] [bi->dma ] leng ntw timestamp bi->skb <-- Legacy format\n");
306 pr_info("Tc[desc] [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma ] leng ntw timestamp bi->skb <-- Ext Context format\n");
307 pr_info("Td[desc] [address 63:0 ] [VlaPoRSCm1Dlen] [bi->dma ] leng ntw timestamp bi->skb <-- Ext Data format\n");
308 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
309 const char *next_desc;
310 tx_desc = E1000_TX_DESC(*tx_ring, i);
311 buffer_info = &tx_ring->buffer_info[i];
312 u0 = (struct my_u0 *)tx_desc;
313 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
314 next_desc = " NTC/U";
315 else if (i == tx_ring->next_to_use)
316 next_desc = " NTU";
317 else if (i == tx_ring->next_to_clean)
318 next_desc = " NTC";
319 else
320 next_desc = "";
321 pr_info("T%c[0x%03X] %016llX %016llX %016llX %04X %3X %016llX %p%s\n",
322 (!(le64_to_cpu(u0->b) & BIT(29)) ? 'l' :
323 ((le64_to_cpu(u0->b) & BIT(20)) ? 'd' : 'c')),
324 i,
325 (unsigned long long)le64_to_cpu(u0->a),
326 (unsigned long long)le64_to_cpu(u0->b),
327 (unsigned long long)buffer_info->dma,
328 buffer_info->length, buffer_info->next_to_watch,
329 (unsigned long long)buffer_info->time_stamp,
330 buffer_info->skb, next_desc);
331
332 if (netif_msg_pktdata(adapter) && buffer_info->skb)
333 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
334 16, 1, buffer_info->skb->data,
335 buffer_info->skb->len, true);
336 }
337
338 /* Print Rx Ring Summary */
339rx_ring_summary:
340 dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
341 pr_info("Queue [NTU] [NTC]\n");
342 pr_info(" %5d %5X %5X\n",
343 0, rx_ring->next_to_use, rx_ring->next_to_clean);
344
345 /* Print Rx Ring */
346 if (!netif_msg_rx_status(adapter))
347 return;
348
349 dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
350 switch (adapter->rx_ps_pages) {
351 case 1:
352 case 2:
353 case 3:
354 /* [Extended] Packet Split Receive Descriptor Format
355 *
356 * +-----------------------------------------------------+
357 * 0 | Buffer Address 0 [63:0] |
358 * +-----------------------------------------------------+
359 * 8 | Buffer Address 1 [63:0] |
360 * +-----------------------------------------------------+
361 * 16 | Buffer Address 2 [63:0] |
362 * +-----------------------------------------------------+
363 * 24 | Buffer Address 3 [63:0] |
364 * +-----------------------------------------------------+
365 */
366 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");
367 /* [Extended] Receive Descriptor (Write-Back) Format
368 *
369 * 63 48 47 32 31 13 12 8 7 4 3 0
370 * +------------------------------------------------------+
371 * 0 | Packet | IP | Rsvd | MRQ | Rsvd | MRQ RSS |
372 * | Checksum | Ident | | Queue | | Type |
373 * +------------------------------------------------------+
374 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
375 * +------------------------------------------------------+
376 * 63 48 47 32 31 20 19 0
377 */
378 pr_info("RWB[desc] [ck ipid mrqhsh] [vl l0 ee es] [ l3 l2 l1 hs] [reserved ] ---------------- [bi->skb] <-- Ext Rx Write-Back format\n");
379 for (i = 0; i < rx_ring->count; i++) {
380 const char *next_desc;
381 buffer_info = &rx_ring->buffer_info[i];
382 rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
383 u1 = (struct my_u1 *)rx_desc_ps;
384 staterr =
385 le32_to_cpu(rx_desc_ps->wb.middle.status_error);
386
387 if (i == rx_ring->next_to_use)
388 next_desc = " NTU";
389 else if (i == rx_ring->next_to_clean)
390 next_desc = " NTC";
391 else
392 next_desc = "";
393
394 if (staterr & E1000_RXD_STAT_DD) {
395 /* Descriptor Done */
396 pr_info("%s[0x%03X] %016llX %016llX %016llX %016llX ---------------- %p%s\n",
397 "RWB", i,
398 (unsigned long long)le64_to_cpu(u1->a),
399 (unsigned long long)le64_to_cpu(u1->b),
400 (unsigned long long)le64_to_cpu(u1->c),
401 (unsigned long long)le64_to_cpu(u1->d),
402 buffer_info->skb, next_desc);
403 } else {
404 pr_info("%s[0x%03X] %016llX %016llX %016llX %016llX %016llX %p%s\n",
405 "R ", i,
406 (unsigned long long)le64_to_cpu(u1->a),
407 (unsigned long long)le64_to_cpu(u1->b),
408 (unsigned long long)le64_to_cpu(u1->c),
409 (unsigned long long)le64_to_cpu(u1->d),
410 (unsigned long long)buffer_info->dma,
411 buffer_info->skb, next_desc);
412
413 if (netif_msg_pktdata(adapter))
414 e1000e_dump_ps_pages(adapter,
415 buffer_info);
416 }
417 }
418 break;
419 default:
420 case 0:
421 /* Extended Receive Descriptor (Read) Format
422 *
423 * +-----------------------------------------------------+
424 * 0 | Buffer Address [63:0] |
425 * +-----------------------------------------------------+
426 * 8 | Reserved |
427 * +-----------------------------------------------------+
428 */
429 pr_info("R [desc] [buf addr 63:0 ] [reserved 63:0 ] [bi->dma ] [bi->skb] <-- Ext (Read) format\n");
430 /* Extended Receive Descriptor (Write-Back) Format
431 *
432 * 63 48 47 32 31 24 23 4 3 0
433 * +------------------------------------------------------+
434 * | RSS Hash | | | |
435 * 0 +-------------------+ Rsvd | Reserved | MRQ RSS |
436 * | Packet | IP | | | Type |
437 * | Checksum | Ident | | | |
438 * +------------------------------------------------------+
439 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
440 * +------------------------------------------------------+
441 * 63 48 47 32 31 20 19 0
442 */
443 pr_info("RWB[desc] [cs ipid mrq] [vt ln xe xs] [bi->skb] <-- Ext (Write-Back) format\n");
444
445 for (i = 0; i < rx_ring->count; i++) {
446 const char *next_desc;
447
448 buffer_info = &rx_ring->buffer_info[i];
449 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
450 u1 = (struct my_u1 *)rx_desc;
451 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
452
453 if (i == rx_ring->next_to_use)
454 next_desc = " NTU";
455 else if (i == rx_ring->next_to_clean)
456 next_desc = " NTC";
457 else
458 next_desc = "";
459
460 if (staterr & E1000_RXD_STAT_DD) {
461 /* Descriptor Done */
462 pr_info("%s[0x%03X] %016llX %016llX ---------------- %p%s\n",
463 "RWB", i,
464 (unsigned long long)le64_to_cpu(u1->a),
465 (unsigned long long)le64_to_cpu(u1->b),
466 buffer_info->skb, next_desc);
467 } else {
468 pr_info("%s[0x%03X] %016llX %016llX %016llX %p%s\n",
469 "R ", i,
470 (unsigned long long)le64_to_cpu(u1->a),
471 (unsigned long long)le64_to_cpu(u1->b),
472 (unsigned long long)buffer_info->dma,
473 buffer_info->skb, next_desc);
474
475 if (netif_msg_pktdata(adapter) &&
476 buffer_info->skb)
477 print_hex_dump(KERN_INFO, "",
478 DUMP_PREFIX_ADDRESS, 16,
479 1,
480 buffer_info->skb->data,
481 adapter->rx_buffer_len,
482 true);
483 }
484 }
485 }
486}
487
488/**
489 * e1000_desc_unused - calculate if we have unused descriptors
490 **/
491static int e1000_desc_unused(struct e1000_ring *ring)
492{
493 if (ring->next_to_clean > ring->next_to_use)
494 return ring->next_to_clean - ring->next_to_use - 1;
495
496 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
497}
498
499/**
500 * e1000e_systim_to_hwtstamp - convert system time value to hw time stamp
501 * @adapter: board private structure
502 * @hwtstamps: time stamp structure to update
503 * @systim: unsigned 64bit system time value.
504 *
505 * Convert the system time value stored in the RX/TXSTMP registers into a
506 * hwtstamp which can be used by the upper level time stamping functions.
507 *
508 * The 'systim_lock' spinlock is used to protect the consistency of the
509 * system time value. This is needed because reading the 64 bit time
510 * value involves reading two 32 bit registers. The first read latches the
511 * value.
512 **/
513static void e1000e_systim_to_hwtstamp(struct e1000_adapter *adapter,
514 struct skb_shared_hwtstamps *hwtstamps,
515 u64 systim)
516{
517 u64 ns;
518 unsigned long flags;
519
520 spin_lock_irqsave(&adapter->systim_lock, flags);
521 ns = timecounter_cyc2time(&adapter->tc, systim);
522 spin_unlock_irqrestore(&adapter->systim_lock, flags);
523
524 memset(hwtstamps, 0, sizeof(*hwtstamps));
525 hwtstamps->hwtstamp = ns_to_ktime(ns);
526}
527
528/**
529 * e1000e_rx_hwtstamp - utility function which checks for Rx time stamp
530 * @adapter: board private structure
531 * @status: descriptor extended error and status field
532 * @skb: particular skb to include time stamp
533 *
534 * If the time stamp is valid, convert it into the timecounter ns value
535 * and store that result into the shhwtstamps structure which is passed
536 * up the network stack.
537 **/
538static void e1000e_rx_hwtstamp(struct e1000_adapter *adapter, u32 status,
539 struct sk_buff *skb)
540{
541 struct e1000_hw *hw = &adapter->hw;
542 u64 rxstmp;
543
544 if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP) ||
545 !(status & E1000_RXDEXT_STATERR_TST) ||
546 !(er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID))
547 return;
548
549 /* The Rx time stamp registers contain the time stamp. No other
550 * received packet will be time stamped until the Rx time stamp
551 * registers are read. Because only one packet can be time stamped
552 * at a time, the register values must belong to this packet and
553 * therefore none of the other additional attributes need to be
554 * compared.
555 */
556 rxstmp = (u64)er32(RXSTMPL);
557 rxstmp |= (u64)er32(RXSTMPH) << 32;
558 e1000e_systim_to_hwtstamp(adapter, skb_hwtstamps(skb), rxstmp);
559
560 adapter->flags2 &= ~FLAG2_CHECK_RX_HWTSTAMP;
561}
562
563/**
564 * e1000_receive_skb - helper function to handle Rx indications
565 * @adapter: board private structure
566 * @staterr: descriptor extended error and status field as written by hardware
567 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
568 * @skb: pointer to sk_buff to be indicated to stack
569 **/
570static void e1000_receive_skb(struct e1000_adapter *adapter,
571 struct net_device *netdev, struct sk_buff *skb,
572 u32 staterr, __le16 vlan)
573{
574 u16 tag = le16_to_cpu(vlan);
575
576 e1000e_rx_hwtstamp(adapter, staterr, skb);
577
578 skb->protocol = eth_type_trans(skb, netdev);
579
580 if (staterr & E1000_RXD_STAT_VP)
581 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tag);
582
583 napi_gro_receive(&adapter->napi, skb);
584}
585
586/**
587 * e1000_rx_checksum - Receive Checksum Offload
588 * @adapter: board private structure
589 * @status_err: receive descriptor status and error fields
590 * @csum: receive descriptor csum field
591 * @sk_buff: socket buffer with received data
592 **/
593static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
594 struct sk_buff *skb)
595{
596 u16 status = (u16)status_err;
597 u8 errors = (u8)(status_err >> 24);
598
599 skb_checksum_none_assert(skb);
600
601 /* Rx checksum disabled */
602 if (!(adapter->netdev->features & NETIF_F_RXCSUM))
603 return;
604
605 /* Ignore Checksum bit is set */
606 if (status & E1000_RXD_STAT_IXSM)
607 return;
608
609 /* TCP/UDP checksum error bit or IP checksum error bit is set */
610 if (errors & (E1000_RXD_ERR_TCPE | E1000_RXD_ERR_IPE)) {
611 /* let the stack verify checksum errors */
612 adapter->hw_csum_err++;
613 return;
614 }
615
616 /* TCP/UDP Checksum has not been calculated */
617 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
618 return;
619
620 /* It must be a TCP or UDP packet with a valid checksum */
621 skb->ip_summed = CHECKSUM_UNNECESSARY;
622 adapter->hw_csum_good++;
623}
624
625static void e1000e_update_rdt_wa(struct e1000_ring *rx_ring, unsigned int i)
626{
627 struct e1000_adapter *adapter = rx_ring->adapter;
628 struct e1000_hw *hw = &adapter->hw;
629 s32 ret_val = __ew32_prepare(hw);
630
631 writel(i, rx_ring->tail);
632
633 if (unlikely(!ret_val && (i != readl(rx_ring->tail)))) {
634 u32 rctl = er32(RCTL);
635
636 ew32(RCTL, rctl & ~E1000_RCTL_EN);
637 e_err("ME firmware caused invalid RDT - resetting\n");
638 schedule_work(&adapter->reset_task);
639 }
640}
641
642static void e1000e_update_tdt_wa(struct e1000_ring *tx_ring, unsigned int i)
643{
644 struct e1000_adapter *adapter = tx_ring->adapter;
645 struct e1000_hw *hw = &adapter->hw;
646 s32 ret_val = __ew32_prepare(hw);
647
648 writel(i, tx_ring->tail);
649
650 if (unlikely(!ret_val && (i != readl(tx_ring->tail)))) {
651 u32 tctl = er32(TCTL);
652
653 ew32(TCTL, tctl & ~E1000_TCTL_EN);
654 e_err("ME firmware caused invalid TDT - resetting\n");
655 schedule_work(&adapter->reset_task);
656 }
657}
658
659/**
660 * e1000_alloc_rx_buffers - Replace used receive buffers
661 * @rx_ring: Rx descriptor ring
662 **/
663static void e1000_alloc_rx_buffers(struct e1000_ring *rx_ring,
664 int cleaned_count, gfp_t gfp)
665{
666 struct e1000_adapter *adapter = rx_ring->adapter;
667 struct net_device *netdev = adapter->netdev;
668 struct pci_dev *pdev = adapter->pdev;
669 union e1000_rx_desc_extended *rx_desc;
670 struct e1000_buffer *buffer_info;
671 struct sk_buff *skb;
672 unsigned int i;
673 unsigned int bufsz = adapter->rx_buffer_len;
674
675 i = rx_ring->next_to_use;
676 buffer_info = &rx_ring->buffer_info[i];
677
678 while (cleaned_count--) {
679 skb = buffer_info->skb;
680 if (skb) {
681 skb_trim(skb, 0);
682 goto map_skb;
683 }
684
685 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
686 if (!skb) {
687 /* Better luck next round */
688 adapter->alloc_rx_buff_failed++;
689 break;
690 }
691
692 buffer_info->skb = skb;
693map_skb:
694 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
695 adapter->rx_buffer_len,
696 DMA_FROM_DEVICE);
697 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
698 dev_err(&pdev->dev, "Rx DMA map failed\n");
699 adapter->rx_dma_failed++;
700 break;
701 }
702
703 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
704 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
705
706 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
707 /* Force memory writes to complete before letting h/w
708 * know there are new descriptors to fetch. (Only
709 * applicable for weak-ordered memory model archs,
710 * such as IA-64).
711 */
712 wmb();
713 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
714 e1000e_update_rdt_wa(rx_ring, i);
715 else
716 writel(i, rx_ring->tail);
717 }
718 i++;
719 if (i == rx_ring->count)
720 i = 0;
721 buffer_info = &rx_ring->buffer_info[i];
722 }
723
724 rx_ring->next_to_use = i;
725}
726
727/**
728 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
729 * @rx_ring: Rx descriptor ring
730 **/
731static void e1000_alloc_rx_buffers_ps(struct e1000_ring *rx_ring,
732 int cleaned_count, gfp_t gfp)
733{
734 struct e1000_adapter *adapter = rx_ring->adapter;
735 struct net_device *netdev = adapter->netdev;
736 struct pci_dev *pdev = adapter->pdev;
737 union e1000_rx_desc_packet_split *rx_desc;
738 struct e1000_buffer *buffer_info;
739 struct e1000_ps_page *ps_page;
740 struct sk_buff *skb;
741 unsigned int i, j;
742
743 i = rx_ring->next_to_use;
744 buffer_info = &rx_ring->buffer_info[i];
745
746 while (cleaned_count--) {
747 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
748
749 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
750 ps_page = &buffer_info->ps_pages[j];
751 if (j >= adapter->rx_ps_pages) {
752 /* all unused desc entries get hw null ptr */
753 rx_desc->read.buffer_addr[j + 1] =
754 ~cpu_to_le64(0);
755 continue;
756 }
757 if (!ps_page->page) {
758 ps_page->page = alloc_page(gfp);
759 if (!ps_page->page) {
760 adapter->alloc_rx_buff_failed++;
761 goto no_buffers;
762 }
763 ps_page->dma = dma_map_page(&pdev->dev,
764 ps_page->page,
765 0, PAGE_SIZE,
766 DMA_FROM_DEVICE);
767 if (dma_mapping_error(&pdev->dev,
768 ps_page->dma)) {
769 dev_err(&adapter->pdev->dev,
770 "Rx DMA page map failed\n");
771 adapter->rx_dma_failed++;
772 goto no_buffers;
773 }
774 }
775 /* Refresh the desc even if buffer_addrs
776 * didn't change because each write-back
777 * erases this info.
778 */
779 rx_desc->read.buffer_addr[j + 1] =
780 cpu_to_le64(ps_page->dma);
781 }
782
783 skb = __netdev_alloc_skb_ip_align(netdev, adapter->rx_ps_bsize0,
784 gfp);
785
786 if (!skb) {
787 adapter->alloc_rx_buff_failed++;
788 break;
789 }
790
791 buffer_info->skb = skb;
792 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
793 adapter->rx_ps_bsize0,
794 DMA_FROM_DEVICE);
795 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
796 dev_err(&pdev->dev, "Rx DMA map failed\n");
797 adapter->rx_dma_failed++;
798 /* cleanup skb */
799 dev_kfree_skb_any(skb);
800 buffer_info->skb = NULL;
801 break;
802 }
803
804 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
805
806 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
807 /* Force memory writes to complete before letting h/w
808 * know there are new descriptors to fetch. (Only
809 * applicable for weak-ordered memory model archs,
810 * such as IA-64).
811 */
812 wmb();
813 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
814 e1000e_update_rdt_wa(rx_ring, i << 1);
815 else
816 writel(i << 1, rx_ring->tail);
817 }
818
819 i++;
820 if (i == rx_ring->count)
821 i = 0;
822 buffer_info = &rx_ring->buffer_info[i];
823 }
824
825no_buffers:
826 rx_ring->next_to_use = i;
827}
828
829/**
830 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
831 * @rx_ring: Rx descriptor ring
832 * @cleaned_count: number of buffers to allocate this pass
833 **/
834
835static void e1000_alloc_jumbo_rx_buffers(struct e1000_ring *rx_ring,
836 int cleaned_count, gfp_t gfp)
837{
838 struct e1000_adapter *adapter = rx_ring->adapter;
839 struct net_device *netdev = adapter->netdev;
840 struct pci_dev *pdev = adapter->pdev;
841 union e1000_rx_desc_extended *rx_desc;
842 struct e1000_buffer *buffer_info;
843 struct sk_buff *skb;
844 unsigned int i;
845 unsigned int bufsz = 256 - 16; /* for skb_reserve */
846
847 i = rx_ring->next_to_use;
848 buffer_info = &rx_ring->buffer_info[i];
849
850 while (cleaned_count--) {
851 skb = buffer_info->skb;
852 if (skb) {
853 skb_trim(skb, 0);
854 goto check_page;
855 }
856
857 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
858 if (unlikely(!skb)) {
859 /* Better luck next round */
860 adapter->alloc_rx_buff_failed++;
861 break;
862 }
863
864 buffer_info->skb = skb;
865check_page:
866 /* allocate a new page if necessary */
867 if (!buffer_info->page) {
868 buffer_info->page = alloc_page(gfp);
869 if (unlikely(!buffer_info->page)) {
870 adapter->alloc_rx_buff_failed++;
871 break;
872 }
873 }
874
875 if (!buffer_info->dma) {
876 buffer_info->dma = dma_map_page(&pdev->dev,
877 buffer_info->page, 0,
878 PAGE_SIZE,
879 DMA_FROM_DEVICE);
880 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
881 adapter->alloc_rx_buff_failed++;
882 break;
883 }
884 }
885
886 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
887 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
888
889 if (unlikely(++i == rx_ring->count))
890 i = 0;
891 buffer_info = &rx_ring->buffer_info[i];
892 }
893
894 if (likely(rx_ring->next_to_use != i)) {
895 rx_ring->next_to_use = i;
896 if (unlikely(i-- == 0))
897 i = (rx_ring->count - 1);
898
899 /* Force memory writes to complete before letting h/w
900 * know there are new descriptors to fetch. (Only
901 * applicable for weak-ordered memory model archs,
902 * such as IA-64).
903 */
904 wmb();
905 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
906 e1000e_update_rdt_wa(rx_ring, i);
907 else
908 writel(i, rx_ring->tail);
909 }
910}
911
912static inline void e1000_rx_hash(struct net_device *netdev, __le32 rss,
913 struct sk_buff *skb)
914{
915 if (netdev->features & NETIF_F_RXHASH)
916 skb_set_hash(skb, le32_to_cpu(rss), PKT_HASH_TYPE_L3);
917}
918
919/**
920 * e1000_clean_rx_irq - Send received data up the network stack
921 * @rx_ring: Rx descriptor ring
922 *
923 * the return value indicates whether actual cleaning was done, there
924 * is no guarantee that everything was cleaned
925 **/
926static bool e1000_clean_rx_irq(struct e1000_ring *rx_ring, int *work_done,
927 int work_to_do)
928{
929 struct e1000_adapter *adapter = rx_ring->adapter;
930 struct net_device *netdev = adapter->netdev;
931 struct pci_dev *pdev = adapter->pdev;
932 struct e1000_hw *hw = &adapter->hw;
933 union e1000_rx_desc_extended *rx_desc, *next_rxd;
934 struct e1000_buffer *buffer_info, *next_buffer;
935 u32 length, staterr;
936 unsigned int i;
937 int cleaned_count = 0;
938 bool cleaned = false;
939 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
940
941 i = rx_ring->next_to_clean;
942 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
943 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
944 buffer_info = &rx_ring->buffer_info[i];
945
946 while (staterr & E1000_RXD_STAT_DD) {
947 struct sk_buff *skb;
948
949 if (*work_done >= work_to_do)
950 break;
951 (*work_done)++;
952 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
953
954 skb = buffer_info->skb;
955 buffer_info->skb = NULL;
956
957 prefetch(skb->data - NET_IP_ALIGN);
958
959 i++;
960 if (i == rx_ring->count)
961 i = 0;
962 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
963 prefetch(next_rxd);
964
965 next_buffer = &rx_ring->buffer_info[i];
966
967 cleaned = true;
968 cleaned_count++;
969 dma_unmap_single(&pdev->dev, buffer_info->dma,
970 adapter->rx_buffer_len, DMA_FROM_DEVICE);
971 buffer_info->dma = 0;
972
973 length = le16_to_cpu(rx_desc->wb.upper.length);
974
975 /* !EOP means multiple descriptors were used to store a single
976 * packet, if that's the case we need to toss it. In fact, we
977 * need to toss every packet with the EOP bit clear and the
978 * next frame that _does_ have the EOP bit set, as it is by
979 * definition only a frame fragment
980 */
981 if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
982 adapter->flags2 |= FLAG2_IS_DISCARDING;
983
984 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
985 /* All receives must fit into a single buffer */
986 e_dbg("Receive packet consumed multiple buffers\n");
987 /* recycle */
988 buffer_info->skb = skb;
989 if (staterr & E1000_RXD_STAT_EOP)
990 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
991 goto next_desc;
992 }
993
994 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
995 !(netdev->features & NETIF_F_RXALL))) {
996 /* recycle */
997 buffer_info->skb = skb;
998 goto next_desc;
999 }
1000
1001 /* adjust length to remove Ethernet CRC */
1002 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1003 /* If configured to store CRC, don't subtract FCS,
1004 * but keep the FCS bytes out of the total_rx_bytes
1005 * counter
1006 */
1007 if (netdev->features & NETIF_F_RXFCS)
1008 total_rx_bytes -= 4;
1009 else
1010 length -= 4;
1011 }
1012
1013 total_rx_bytes += length;
1014 total_rx_packets++;
1015
1016 /* code added for copybreak, this should improve
1017 * performance for small packets with large amounts
1018 * of reassembly being done in the stack
1019 */
1020 if (length < copybreak) {
1021 struct sk_buff *new_skb =
1022 napi_alloc_skb(&adapter->napi, length);
1023 if (new_skb) {
1024 skb_copy_to_linear_data_offset(new_skb,
1025 -NET_IP_ALIGN,
1026 (skb->data -
1027 NET_IP_ALIGN),
1028 (length +
1029 NET_IP_ALIGN));
1030 /* save the skb in buffer_info as good */
1031 buffer_info->skb = skb;
1032 skb = new_skb;
1033 }
1034 /* else just continue with the old one */
1035 }
1036 /* end copybreak code */
1037 skb_put(skb, length);
1038
1039 /* Receive Checksum Offload */
1040 e1000_rx_checksum(adapter, staterr, skb);
1041
1042 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1043
1044 e1000_receive_skb(adapter, netdev, skb, staterr,
1045 rx_desc->wb.upper.vlan);
1046
1047next_desc:
1048 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1049
1050 /* return some buffers to hardware, one at a time is too slow */
1051 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1052 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1053 GFP_ATOMIC);
1054 cleaned_count = 0;
1055 }
1056
1057 /* use prefetched values */
1058 rx_desc = next_rxd;
1059 buffer_info = next_buffer;
1060
1061 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1062 }
1063 rx_ring->next_to_clean = i;
1064
1065 cleaned_count = e1000_desc_unused(rx_ring);
1066 if (cleaned_count)
1067 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1068
1069 adapter->total_rx_bytes += total_rx_bytes;
1070 adapter->total_rx_packets += total_rx_packets;
1071 return cleaned;
1072}
1073
1074static void e1000_put_txbuf(struct e1000_ring *tx_ring,
1075 struct e1000_buffer *buffer_info,
1076 bool drop)
1077{
1078 struct e1000_adapter *adapter = tx_ring->adapter;
1079
1080 if (buffer_info->dma) {
1081 if (buffer_info->mapped_as_page)
1082 dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1083 buffer_info->length, DMA_TO_DEVICE);
1084 else
1085 dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1086 buffer_info->length, DMA_TO_DEVICE);
1087 buffer_info->dma = 0;
1088 }
1089 if (buffer_info->skb) {
1090 if (drop)
1091 dev_kfree_skb_any(buffer_info->skb);
1092 else
1093 dev_consume_skb_any(buffer_info->skb);
1094 buffer_info->skb = NULL;
1095 }
1096 buffer_info->time_stamp = 0;
1097}
1098
1099static void e1000_print_hw_hang(struct work_struct *work)
1100{
1101 struct e1000_adapter *adapter = container_of(work,
1102 struct e1000_adapter,
1103 print_hang_task);
1104 struct net_device *netdev = adapter->netdev;
1105 struct e1000_ring *tx_ring = adapter->tx_ring;
1106 unsigned int i = tx_ring->next_to_clean;
1107 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1108 struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1109 struct e1000_hw *hw = &adapter->hw;
1110 u16 phy_status, phy_1000t_status, phy_ext_status;
1111 u16 pci_status;
1112
1113 if (test_bit(__E1000_DOWN, &adapter->state))
1114 return;
1115
1116 if (!adapter->tx_hang_recheck && (adapter->flags2 & FLAG2_DMA_BURST)) {
1117 /* May be block on write-back, flush and detect again
1118 * flush pending descriptor writebacks to memory
1119 */
1120 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1121 /* execute the writes immediately */
1122 e1e_flush();
1123 /* Due to rare timing issues, write to TIDV again to ensure
1124 * the write is successful
1125 */
1126 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1127 /* execute the writes immediately */
1128 e1e_flush();
1129 adapter->tx_hang_recheck = true;
1130 return;
1131 }
1132 adapter->tx_hang_recheck = false;
1133
1134 if (er32(TDH(0)) == er32(TDT(0))) {
1135 e_dbg("false hang detected, ignoring\n");
1136 return;
1137 }
1138
1139 /* Real hang detected */
1140 netif_stop_queue(netdev);
1141
1142 e1e_rphy(hw, MII_BMSR, &phy_status);
1143 e1e_rphy(hw, MII_STAT1000, &phy_1000t_status);
1144 e1e_rphy(hw, MII_ESTATUS, &phy_ext_status);
1145
1146 pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1147
1148 /* detected Hardware unit hang */
1149 e_err("Detected Hardware Unit Hang:\n"
1150 " TDH <%x>\n"
1151 " TDT <%x>\n"
1152 " next_to_use <%x>\n"
1153 " next_to_clean <%x>\n"
1154 "buffer_info[next_to_clean]:\n"
1155 " time_stamp <%lx>\n"
1156 " next_to_watch <%x>\n"
1157 " jiffies <%lx>\n"
1158 " next_to_watch.status <%x>\n"
1159 "MAC Status <%x>\n"
1160 "PHY Status <%x>\n"
1161 "PHY 1000BASE-T Status <%x>\n"
1162 "PHY Extended Status <%x>\n"
1163 "PCI Status <%x>\n",
1164 readl(tx_ring->head), readl(tx_ring->tail), tx_ring->next_to_use,
1165 tx_ring->next_to_clean, tx_ring->buffer_info[eop].time_stamp,
1166 eop, jiffies, eop_desc->upper.fields.status, er32(STATUS),
1167 phy_status, phy_1000t_status, phy_ext_status, pci_status);
1168
1169 e1000e_dump(adapter);
1170
1171 /* Suggest workaround for known h/w issue */
1172 if ((hw->mac.type == e1000_pchlan) && (er32(CTRL) & E1000_CTRL_TFCE))
1173 e_err("Try turning off Tx pause (flow control) via ethtool\n");
1174}
1175
1176/**
1177 * e1000e_tx_hwtstamp_work - check for Tx time stamp
1178 * @work: pointer to work struct
1179 *
1180 * This work function polls the TSYNCTXCTL valid bit to determine when a
1181 * timestamp has been taken for the current stored skb. The timestamp must
1182 * be for this skb because only one such packet is allowed in the queue.
1183 */
1184static void e1000e_tx_hwtstamp_work(struct work_struct *work)
1185{
1186 struct e1000_adapter *adapter = container_of(work, struct e1000_adapter,
1187 tx_hwtstamp_work);
1188 struct e1000_hw *hw = &adapter->hw;
1189
1190 if (er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID) {
1191 struct sk_buff *skb = adapter->tx_hwtstamp_skb;
1192 struct skb_shared_hwtstamps shhwtstamps;
1193 u64 txstmp;
1194
1195 txstmp = er32(TXSTMPL);
1196 txstmp |= (u64)er32(TXSTMPH) << 32;
1197
1198 e1000e_systim_to_hwtstamp(adapter, &shhwtstamps, txstmp);
1199
1200 /* Clear the global tx_hwtstamp_skb pointer and force writes
1201 * prior to notifying the stack of a Tx timestamp.
1202 */
1203 adapter->tx_hwtstamp_skb = NULL;
1204 wmb(); /* force write prior to skb_tstamp_tx */
1205
1206 skb_tstamp_tx(skb, &shhwtstamps);
1207 dev_consume_skb_any(skb);
1208 } else if (time_after(jiffies, adapter->tx_hwtstamp_start
1209 + adapter->tx_timeout_factor * HZ)) {
1210 dev_kfree_skb_any(adapter->tx_hwtstamp_skb);
1211 adapter->tx_hwtstamp_skb = NULL;
1212 adapter->tx_hwtstamp_timeouts++;
1213 e_warn("clearing Tx timestamp hang\n");
1214 } else {
1215 /* reschedule to check later */
1216 schedule_work(&adapter->tx_hwtstamp_work);
1217 }
1218}
1219
1220/**
1221 * e1000_clean_tx_irq - Reclaim resources after transmit completes
1222 * @tx_ring: Tx descriptor ring
1223 *
1224 * the return value indicates whether actual cleaning was done, there
1225 * is no guarantee that everything was cleaned
1226 **/
1227static bool e1000_clean_tx_irq(struct e1000_ring *tx_ring)
1228{
1229 struct e1000_adapter *adapter = tx_ring->adapter;
1230 struct net_device *netdev = adapter->netdev;
1231 struct e1000_hw *hw = &adapter->hw;
1232 struct e1000_tx_desc *tx_desc, *eop_desc;
1233 struct e1000_buffer *buffer_info;
1234 unsigned int i, eop;
1235 unsigned int count = 0;
1236 unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1237 unsigned int bytes_compl = 0, pkts_compl = 0;
1238
1239 i = tx_ring->next_to_clean;
1240 eop = tx_ring->buffer_info[i].next_to_watch;
1241 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1242
1243 while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1244 (count < tx_ring->count)) {
1245 bool cleaned = false;
1246
1247 dma_rmb(); /* read buffer_info after eop_desc */
1248 for (; !cleaned; count++) {
1249 tx_desc = E1000_TX_DESC(*tx_ring, i);
1250 buffer_info = &tx_ring->buffer_info[i];
1251 cleaned = (i == eop);
1252
1253 if (cleaned) {
1254 total_tx_packets += buffer_info->segs;
1255 total_tx_bytes += buffer_info->bytecount;
1256 if (buffer_info->skb) {
1257 bytes_compl += buffer_info->skb->len;
1258 pkts_compl++;
1259 }
1260 }
1261
1262 e1000_put_txbuf(tx_ring, buffer_info, false);
1263 tx_desc->upper.data = 0;
1264
1265 i++;
1266 if (i == tx_ring->count)
1267 i = 0;
1268 }
1269
1270 if (i == tx_ring->next_to_use)
1271 break;
1272 eop = tx_ring->buffer_info[i].next_to_watch;
1273 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1274 }
1275
1276 tx_ring->next_to_clean = i;
1277
1278 netdev_completed_queue(netdev, pkts_compl, bytes_compl);
1279
1280#define TX_WAKE_THRESHOLD 32
1281 if (count && netif_carrier_ok(netdev) &&
1282 e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1283 /* Make sure that anybody stopping the queue after this
1284 * sees the new next_to_clean.
1285 */
1286 smp_mb();
1287
1288 if (netif_queue_stopped(netdev) &&
1289 !(test_bit(__E1000_DOWN, &adapter->state))) {
1290 netif_wake_queue(netdev);
1291 ++adapter->restart_queue;
1292 }
1293 }
1294
1295 if (adapter->detect_tx_hung) {
1296 /* Detect a transmit hang in hardware, this serializes the
1297 * check with the clearing of time_stamp and movement of i
1298 */
1299 adapter->detect_tx_hung = false;
1300 if (tx_ring->buffer_info[i].time_stamp &&
1301 time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1302 + (adapter->tx_timeout_factor * HZ)) &&
1303 !(er32(STATUS) & E1000_STATUS_TXOFF))
1304 schedule_work(&adapter->print_hang_task);
1305 else
1306 adapter->tx_hang_recheck = false;
1307 }
1308 adapter->total_tx_bytes += total_tx_bytes;
1309 adapter->total_tx_packets += total_tx_packets;
1310 return count < tx_ring->count;
1311}
1312
1313/**
1314 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1315 * @rx_ring: Rx descriptor ring
1316 *
1317 * the return value indicates whether actual cleaning was done, there
1318 * is no guarantee that everything was cleaned
1319 **/
1320static bool e1000_clean_rx_irq_ps(struct e1000_ring *rx_ring, int *work_done,
1321 int work_to_do)
1322{
1323 struct e1000_adapter *adapter = rx_ring->adapter;
1324 struct e1000_hw *hw = &adapter->hw;
1325 union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1326 struct net_device *netdev = adapter->netdev;
1327 struct pci_dev *pdev = adapter->pdev;
1328 struct e1000_buffer *buffer_info, *next_buffer;
1329 struct e1000_ps_page *ps_page;
1330 struct sk_buff *skb;
1331 unsigned int i, j;
1332 u32 length, staterr;
1333 int cleaned_count = 0;
1334 bool cleaned = false;
1335 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1336
1337 i = rx_ring->next_to_clean;
1338 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1339 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1340 buffer_info = &rx_ring->buffer_info[i];
1341
1342 while (staterr & E1000_RXD_STAT_DD) {
1343 if (*work_done >= work_to_do)
1344 break;
1345 (*work_done)++;
1346 skb = buffer_info->skb;
1347 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
1348
1349 /* in the packet split case this is header only */
1350 prefetch(skb->data - NET_IP_ALIGN);
1351
1352 i++;
1353 if (i == rx_ring->count)
1354 i = 0;
1355 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1356 prefetch(next_rxd);
1357
1358 next_buffer = &rx_ring->buffer_info[i];
1359
1360 cleaned = true;
1361 cleaned_count++;
1362 dma_unmap_single(&pdev->dev, buffer_info->dma,
1363 adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1364 buffer_info->dma = 0;
1365
1366 /* see !EOP comment in other Rx routine */
1367 if (!(staterr & E1000_RXD_STAT_EOP))
1368 adapter->flags2 |= FLAG2_IS_DISCARDING;
1369
1370 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1371 e_dbg("Packet Split buffers didn't pick up the full packet\n");
1372 dev_kfree_skb_irq(skb);
1373 if (staterr & E1000_RXD_STAT_EOP)
1374 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1375 goto next_desc;
1376 }
1377
1378 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1379 !(netdev->features & NETIF_F_RXALL))) {
1380 dev_kfree_skb_irq(skb);
1381 goto next_desc;
1382 }
1383
1384 length = le16_to_cpu(rx_desc->wb.middle.length0);
1385
1386 if (!length) {
1387 e_dbg("Last part of the packet spanning multiple descriptors\n");
1388 dev_kfree_skb_irq(skb);
1389 goto next_desc;
1390 }
1391
1392 /* Good Receive */
1393 skb_put(skb, length);
1394
1395 {
1396 /* this looks ugly, but it seems compiler issues make
1397 * it more efficient than reusing j
1398 */
1399 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1400
1401 /* page alloc/put takes too long and effects small
1402 * packet throughput, so unsplit small packets and
1403 * save the alloc/put only valid in softirq (napi)
1404 * context to call kmap_*
1405 */
1406 if (l1 && (l1 <= copybreak) &&
1407 ((length + l1) <= adapter->rx_ps_bsize0)) {
1408 u8 *vaddr;
1409
1410 ps_page = &buffer_info->ps_pages[0];
1411
1412 /* there is no documentation about how to call
1413 * kmap_atomic, so we can't hold the mapping
1414 * very long
1415 */
1416 dma_sync_single_for_cpu(&pdev->dev,
1417 ps_page->dma,
1418 PAGE_SIZE,
1419 DMA_FROM_DEVICE);
1420 vaddr = kmap_atomic(ps_page->page);
1421 memcpy(skb_tail_pointer(skb), vaddr, l1);
1422 kunmap_atomic(vaddr);
1423 dma_sync_single_for_device(&pdev->dev,
1424 ps_page->dma,
1425 PAGE_SIZE,
1426 DMA_FROM_DEVICE);
1427
1428 /* remove the CRC */
1429 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1430 if (!(netdev->features & NETIF_F_RXFCS))
1431 l1 -= 4;
1432 }
1433
1434 skb_put(skb, l1);
1435 goto copydone;
1436 } /* if */
1437 }
1438
1439 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1440 length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1441 if (!length)
1442 break;
1443
1444 ps_page = &buffer_info->ps_pages[j];
1445 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1446 DMA_FROM_DEVICE);
1447 ps_page->dma = 0;
1448 skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1449 ps_page->page = NULL;
1450 skb->len += length;
1451 skb->data_len += length;
1452 skb->truesize += PAGE_SIZE;
1453 }
1454
1455 /* strip the ethernet crc, problem is we're using pages now so
1456 * this whole operation can get a little cpu intensive
1457 */
1458 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1459 if (!(netdev->features & NETIF_F_RXFCS))
1460 pskb_trim(skb, skb->len - 4);
1461 }
1462
1463copydone:
1464 total_rx_bytes += skb->len;
1465 total_rx_packets++;
1466
1467 e1000_rx_checksum(adapter, staterr, skb);
1468
1469 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1470
1471 if (rx_desc->wb.upper.header_status &
1472 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1473 adapter->rx_hdr_split++;
1474
1475 e1000_receive_skb(adapter, netdev, skb, staterr,
1476 rx_desc->wb.middle.vlan);
1477
1478next_desc:
1479 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1480 buffer_info->skb = NULL;
1481
1482 /* return some buffers to hardware, one at a time is too slow */
1483 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1484 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1485 GFP_ATOMIC);
1486 cleaned_count = 0;
1487 }
1488
1489 /* use prefetched values */
1490 rx_desc = next_rxd;
1491 buffer_info = next_buffer;
1492
1493 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1494 }
1495 rx_ring->next_to_clean = i;
1496
1497 cleaned_count = e1000_desc_unused(rx_ring);
1498 if (cleaned_count)
1499 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1500
1501 adapter->total_rx_bytes += total_rx_bytes;
1502 adapter->total_rx_packets += total_rx_packets;
1503 return cleaned;
1504}
1505
1506/**
1507 * e1000_consume_page - helper function
1508 **/
1509static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1510 u16 length)
1511{
1512 bi->page = NULL;
1513 skb->len += length;
1514 skb->data_len += length;
1515 skb->truesize += PAGE_SIZE;
1516}
1517
1518/**
1519 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1520 * @adapter: board private structure
1521 *
1522 * the return value indicates whether actual cleaning was done, there
1523 * is no guarantee that everything was cleaned
1524 **/
1525static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
1526 int work_to_do)
1527{
1528 struct e1000_adapter *adapter = rx_ring->adapter;
1529 struct net_device *netdev = adapter->netdev;
1530 struct pci_dev *pdev = adapter->pdev;
1531 union e1000_rx_desc_extended *rx_desc, *next_rxd;
1532 struct e1000_buffer *buffer_info, *next_buffer;
1533 u32 length, staterr;
1534 unsigned int i;
1535 int cleaned_count = 0;
1536 bool cleaned = false;
1537 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1538 struct skb_shared_info *shinfo;
1539
1540 i = rx_ring->next_to_clean;
1541 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1542 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1543 buffer_info = &rx_ring->buffer_info[i];
1544
1545 while (staterr & E1000_RXD_STAT_DD) {
1546 struct sk_buff *skb;
1547
1548 if (*work_done >= work_to_do)
1549 break;
1550 (*work_done)++;
1551 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
1552
1553 skb = buffer_info->skb;
1554 buffer_info->skb = NULL;
1555
1556 ++i;
1557 if (i == rx_ring->count)
1558 i = 0;
1559 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1560 prefetch(next_rxd);
1561
1562 next_buffer = &rx_ring->buffer_info[i];
1563
1564 cleaned = true;
1565 cleaned_count++;
1566 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1567 DMA_FROM_DEVICE);
1568 buffer_info->dma = 0;
1569
1570 length = le16_to_cpu(rx_desc->wb.upper.length);
1571
1572 /* errors is only valid for DD + EOP descriptors */
1573 if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1574 ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1575 !(netdev->features & NETIF_F_RXALL)))) {
1576 /* recycle both page and skb */
1577 buffer_info->skb = skb;
1578 /* an error means any chain goes out the window too */
1579 if (rx_ring->rx_skb_top)
1580 dev_kfree_skb_irq(rx_ring->rx_skb_top);
1581 rx_ring->rx_skb_top = NULL;
1582 goto next_desc;
1583 }
1584#define rxtop (rx_ring->rx_skb_top)
1585 if (!(staterr & E1000_RXD_STAT_EOP)) {
1586 /* this descriptor is only the beginning (or middle) */
1587 if (!rxtop) {
1588 /* this is the beginning of a chain */
1589 rxtop = skb;
1590 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1591 0, length);
1592 } else {
1593 /* this is the middle of a chain */
1594 shinfo = skb_shinfo(rxtop);
1595 skb_fill_page_desc(rxtop, shinfo->nr_frags,
1596 buffer_info->page, 0,
1597 length);
1598 /* re-use the skb, only consumed the page */
1599 buffer_info->skb = skb;
1600 }
1601 e1000_consume_page(buffer_info, rxtop, length);
1602 goto next_desc;
1603 } else {
1604 if (rxtop) {
1605 /* end of the chain */
1606 shinfo = skb_shinfo(rxtop);
1607 skb_fill_page_desc(rxtop, shinfo->nr_frags,
1608 buffer_info->page, 0,
1609 length);
1610 /* re-use the current skb, we only consumed the
1611 * page
1612 */
1613 buffer_info->skb = skb;
1614 skb = rxtop;
1615 rxtop = NULL;
1616 e1000_consume_page(buffer_info, skb, length);
1617 } else {
1618 /* no chain, got EOP, this buf is the packet
1619 * copybreak to save the put_page/alloc_page
1620 */
1621 if (length <= copybreak &&
1622 skb_tailroom(skb) >= length) {
1623 u8 *vaddr;
1624 vaddr = kmap_atomic(buffer_info->page);
1625 memcpy(skb_tail_pointer(skb), vaddr,
1626 length);
1627 kunmap_atomic(vaddr);
1628 /* re-use the page, so don't erase
1629 * buffer_info->page
1630 */
1631 skb_put(skb, length);
1632 } else {
1633 skb_fill_page_desc(skb, 0,
1634 buffer_info->page, 0,
1635 length);
1636 e1000_consume_page(buffer_info, skb,
1637 length);
1638 }
1639 }
1640 }
1641
1642 /* Receive Checksum Offload */
1643 e1000_rx_checksum(adapter, staterr, skb);
1644
1645 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1646
1647 /* probably a little skewed due to removing CRC */
1648 total_rx_bytes += skb->len;
1649 total_rx_packets++;
1650
1651 /* eth type trans needs skb->data to point to something */
1652 if (!pskb_may_pull(skb, ETH_HLEN)) {
1653 e_err("pskb_may_pull failed.\n");
1654 dev_kfree_skb_irq(skb);
1655 goto next_desc;
1656 }
1657
1658 e1000_receive_skb(adapter, netdev, skb, staterr,
1659 rx_desc->wb.upper.vlan);
1660
1661next_desc:
1662 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1663
1664 /* return some buffers to hardware, one at a time is too slow */
1665 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1666 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1667 GFP_ATOMIC);
1668 cleaned_count = 0;
1669 }
1670
1671 /* use prefetched values */
1672 rx_desc = next_rxd;
1673 buffer_info = next_buffer;
1674
1675 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1676 }
1677 rx_ring->next_to_clean = i;
1678
1679 cleaned_count = e1000_desc_unused(rx_ring);
1680 if (cleaned_count)
1681 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1682
1683 adapter->total_rx_bytes += total_rx_bytes;
1684 adapter->total_rx_packets += total_rx_packets;
1685 return cleaned;
1686}
1687
1688/**
1689 * e1000_clean_rx_ring - Free Rx Buffers per Queue
1690 * @rx_ring: Rx descriptor ring
1691 **/
1692static void e1000_clean_rx_ring(struct e1000_ring *rx_ring)
1693{
1694 struct e1000_adapter *adapter = rx_ring->adapter;
1695 struct e1000_buffer *buffer_info;
1696 struct e1000_ps_page *ps_page;
1697 struct pci_dev *pdev = adapter->pdev;
1698 unsigned int i, j;
1699
1700 /* Free all the Rx ring sk_buffs */
1701 for (i = 0; i < rx_ring->count; i++) {
1702 buffer_info = &rx_ring->buffer_info[i];
1703 if (buffer_info->dma) {
1704 if (adapter->clean_rx == e1000_clean_rx_irq)
1705 dma_unmap_single(&pdev->dev, buffer_info->dma,
1706 adapter->rx_buffer_len,
1707 DMA_FROM_DEVICE);
1708 else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1709 dma_unmap_page(&pdev->dev, buffer_info->dma,
1710 PAGE_SIZE, DMA_FROM_DEVICE);
1711 else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1712 dma_unmap_single(&pdev->dev, buffer_info->dma,
1713 adapter->rx_ps_bsize0,
1714 DMA_FROM_DEVICE);
1715 buffer_info->dma = 0;
1716 }
1717
1718 if (buffer_info->page) {
1719 put_page(buffer_info->page);
1720 buffer_info->page = NULL;
1721 }
1722
1723 if (buffer_info->skb) {
1724 dev_kfree_skb(buffer_info->skb);
1725 buffer_info->skb = NULL;
1726 }
1727
1728 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1729 ps_page = &buffer_info->ps_pages[j];
1730 if (!ps_page->page)
1731 break;
1732 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1733 DMA_FROM_DEVICE);
1734 ps_page->dma = 0;
1735 put_page(ps_page->page);
1736 ps_page->page = NULL;
1737 }
1738 }
1739
1740 /* there also may be some cached data from a chained receive */
1741 if (rx_ring->rx_skb_top) {
1742 dev_kfree_skb(rx_ring->rx_skb_top);
1743 rx_ring->rx_skb_top = NULL;
1744 }
1745
1746 /* Zero out the descriptor ring */
1747 memset(rx_ring->desc, 0, rx_ring->size);
1748
1749 rx_ring->next_to_clean = 0;
1750 rx_ring->next_to_use = 0;
1751 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1752}
1753
1754static void e1000e_downshift_workaround(struct work_struct *work)
1755{
1756 struct e1000_adapter *adapter = container_of(work,
1757 struct e1000_adapter,
1758 downshift_task);
1759
1760 if (test_bit(__E1000_DOWN, &adapter->state))
1761 return;
1762
1763 e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1764}
1765
1766/**
1767 * e1000_intr_msi - Interrupt Handler
1768 * @irq: interrupt number
1769 * @data: pointer to a network interface device structure
1770 **/
1771static irqreturn_t e1000_intr_msi(int __always_unused irq, void *data)
1772{
1773 struct net_device *netdev = data;
1774 struct e1000_adapter *adapter = netdev_priv(netdev);
1775 struct e1000_hw *hw = &adapter->hw;
1776 u32 icr = er32(ICR);
1777
1778 /* read ICR disables interrupts using IAM */
1779 if (icr & E1000_ICR_LSC) {
1780 hw->mac.get_link_status = true;
1781 /* ICH8 workaround-- Call gig speed drop workaround on cable
1782 * disconnect (LSC) before accessing any PHY registers
1783 */
1784 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1785 (!(er32(STATUS) & E1000_STATUS_LU)))
1786 schedule_work(&adapter->downshift_task);
1787
1788 /* 80003ES2LAN workaround-- For packet buffer work-around on
1789 * link down event; disable receives here in the ISR and reset
1790 * adapter in watchdog
1791 */
1792 if (netif_carrier_ok(netdev) &&
1793 adapter->flags & FLAG_RX_NEEDS_RESTART) {
1794 /* disable receives */
1795 u32 rctl = er32(RCTL);
1796
1797 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1798 adapter->flags |= FLAG_RESTART_NOW;
1799 }
1800 /* guard against interrupt when we're going down */
1801 if (!test_bit(__E1000_DOWN, &adapter->state))
1802 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1803 }
1804
1805 /* Reset on uncorrectable ECC error */
1806 if ((icr & E1000_ICR_ECCER) && (hw->mac.type >= e1000_pch_lpt)) {
1807 u32 pbeccsts = er32(PBECCSTS);
1808
1809 adapter->corr_errors +=
1810 pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1811 adapter->uncorr_errors +=
1812 (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
1813 E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
1814
1815 /* Do the reset outside of interrupt context */
1816 schedule_work(&adapter->reset_task);
1817
1818 /* return immediately since reset is imminent */
1819 return IRQ_HANDLED;
1820 }
1821
1822 if (napi_schedule_prep(&adapter->napi)) {
1823 adapter->total_tx_bytes = 0;
1824 adapter->total_tx_packets = 0;
1825 adapter->total_rx_bytes = 0;
1826 adapter->total_rx_packets = 0;
1827 __napi_schedule(&adapter->napi);
1828 }
1829
1830 return IRQ_HANDLED;
1831}
1832
1833/**
1834 * e1000_intr - Interrupt Handler
1835 * @irq: interrupt number
1836 * @data: pointer to a network interface device structure
1837 **/
1838static irqreturn_t e1000_intr(int __always_unused irq, void *data)
1839{
1840 struct net_device *netdev = data;
1841 struct e1000_adapter *adapter = netdev_priv(netdev);
1842 struct e1000_hw *hw = &adapter->hw;
1843 u32 rctl, icr = er32(ICR);
1844
1845 if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1846 return IRQ_NONE; /* Not our interrupt */
1847
1848 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1849 * not set, then the adapter didn't send an interrupt
1850 */
1851 if (!(icr & E1000_ICR_INT_ASSERTED))
1852 return IRQ_NONE;
1853
1854 /* Interrupt Auto-Mask...upon reading ICR,
1855 * interrupts are masked. No need for the
1856 * IMC write
1857 */
1858
1859 if (icr & E1000_ICR_LSC) {
1860 hw->mac.get_link_status = true;
1861 /* ICH8 workaround-- Call gig speed drop workaround on cable
1862 * disconnect (LSC) before accessing any PHY registers
1863 */
1864 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1865 (!(er32(STATUS) & E1000_STATUS_LU)))
1866 schedule_work(&adapter->downshift_task);
1867
1868 /* 80003ES2LAN workaround--
1869 * For packet buffer work-around on link down event;
1870 * disable receives here in the ISR and
1871 * reset adapter in watchdog
1872 */
1873 if (netif_carrier_ok(netdev) &&
1874 (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1875 /* disable receives */
1876 rctl = er32(RCTL);
1877 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1878 adapter->flags |= FLAG_RESTART_NOW;
1879 }
1880 /* guard against interrupt when we're going down */
1881 if (!test_bit(__E1000_DOWN, &adapter->state))
1882 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1883 }
1884
1885 /* Reset on uncorrectable ECC error */
1886 if ((icr & E1000_ICR_ECCER) && (hw->mac.type >= e1000_pch_lpt)) {
1887 u32 pbeccsts = er32(PBECCSTS);
1888
1889 adapter->corr_errors +=
1890 pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1891 adapter->uncorr_errors +=
1892 (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
1893 E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
1894
1895 /* Do the reset outside of interrupt context */
1896 schedule_work(&adapter->reset_task);
1897
1898 /* return immediately since reset is imminent */
1899 return IRQ_HANDLED;
1900 }
1901
1902 if (napi_schedule_prep(&adapter->napi)) {
1903 adapter->total_tx_bytes = 0;
1904 adapter->total_tx_packets = 0;
1905 adapter->total_rx_bytes = 0;
1906 adapter->total_rx_packets = 0;
1907 __napi_schedule(&adapter->napi);
1908 }
1909
1910 return IRQ_HANDLED;
1911}
1912
1913static irqreturn_t e1000_msix_other(int __always_unused irq, void *data)
1914{
1915 struct net_device *netdev = data;
1916 struct e1000_adapter *adapter = netdev_priv(netdev);
1917 struct e1000_hw *hw = &adapter->hw;
1918 u32 icr = er32(ICR);
1919
1920 if (icr & adapter->eiac_mask)
1921 ew32(ICS, (icr & adapter->eiac_mask));
1922
1923 if (icr & E1000_ICR_LSC) {
1924 hw->mac.get_link_status = true;
1925 /* guard against interrupt when we're going down */
1926 if (!test_bit(__E1000_DOWN, &adapter->state))
1927 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1928 }
1929
1930 if (!test_bit(__E1000_DOWN, &adapter->state))
1931 ew32(IMS, E1000_IMS_OTHER | IMS_OTHER_MASK);
1932
1933 return IRQ_HANDLED;
1934}
1935
1936static irqreturn_t e1000_intr_msix_tx(int __always_unused irq, void *data)
1937{
1938 struct net_device *netdev = data;
1939 struct e1000_adapter *adapter = netdev_priv(netdev);
1940 struct e1000_hw *hw = &adapter->hw;
1941 struct e1000_ring *tx_ring = adapter->tx_ring;
1942
1943 adapter->total_tx_bytes = 0;
1944 adapter->total_tx_packets = 0;
1945
1946 if (!e1000_clean_tx_irq(tx_ring))
1947 /* Ring was not completely cleaned, so fire another interrupt */
1948 ew32(ICS, tx_ring->ims_val);
1949
1950 if (!test_bit(__E1000_DOWN, &adapter->state))
1951 ew32(IMS, adapter->tx_ring->ims_val);
1952
1953 return IRQ_HANDLED;
1954}
1955
1956static irqreturn_t e1000_intr_msix_rx(int __always_unused irq, void *data)
1957{
1958 struct net_device *netdev = data;
1959 struct e1000_adapter *adapter = netdev_priv(netdev);
1960 struct e1000_ring *rx_ring = adapter->rx_ring;
1961
1962 /* Write the ITR value calculated at the end of the
1963 * previous interrupt.
1964 */
1965 if (rx_ring->set_itr) {
1966 u32 itr = rx_ring->itr_val ?
1967 1000000000 / (rx_ring->itr_val * 256) : 0;
1968
1969 writel(itr, rx_ring->itr_register);
1970 rx_ring->set_itr = 0;
1971 }
1972
1973 if (napi_schedule_prep(&adapter->napi)) {
1974 adapter->total_rx_bytes = 0;
1975 adapter->total_rx_packets = 0;
1976 __napi_schedule(&adapter->napi);
1977 }
1978 return IRQ_HANDLED;
1979}
1980
1981/**
1982 * e1000_configure_msix - Configure MSI-X hardware
1983 *
1984 * e1000_configure_msix sets up the hardware to properly
1985 * generate MSI-X interrupts.
1986 **/
1987static void e1000_configure_msix(struct e1000_adapter *adapter)
1988{
1989 struct e1000_hw *hw = &adapter->hw;
1990 struct e1000_ring *rx_ring = adapter->rx_ring;
1991 struct e1000_ring *tx_ring = adapter->tx_ring;
1992 int vector = 0;
1993 u32 ctrl_ext, ivar = 0;
1994
1995 adapter->eiac_mask = 0;
1996
1997 /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1998 if (hw->mac.type == e1000_82574) {
1999 u32 rfctl = er32(RFCTL);
2000
2001 rfctl |= E1000_RFCTL_ACK_DIS;
2002 ew32(RFCTL, rfctl);
2003 }
2004
2005 /* Configure Rx vector */
2006 rx_ring->ims_val = E1000_IMS_RXQ0;
2007 adapter->eiac_mask |= rx_ring->ims_val;
2008 if (rx_ring->itr_val)
2009 writel(1000000000 / (rx_ring->itr_val * 256),
2010 rx_ring->itr_register);
2011 else
2012 writel(1, rx_ring->itr_register);
2013 ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
2014
2015 /* Configure Tx vector */
2016 tx_ring->ims_val = E1000_IMS_TXQ0;
2017 vector++;
2018 if (tx_ring->itr_val)
2019 writel(1000000000 / (tx_ring->itr_val * 256),
2020 tx_ring->itr_register);
2021 else
2022 writel(1, tx_ring->itr_register);
2023 adapter->eiac_mask |= tx_ring->ims_val;
2024 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
2025
2026 /* set vector for Other Causes, e.g. link changes */
2027 vector++;
2028 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
2029 if (rx_ring->itr_val)
2030 writel(1000000000 / (rx_ring->itr_val * 256),
2031 hw->hw_addr + E1000_EITR_82574(vector));
2032 else
2033 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
2034
2035 /* Cause Tx interrupts on every write back */
2036 ivar |= BIT(31);
2037
2038 ew32(IVAR, ivar);
2039
2040 /* enable MSI-X PBA support */
2041 ctrl_ext = er32(CTRL_EXT) & ~E1000_CTRL_EXT_IAME;
2042 ctrl_ext |= E1000_CTRL_EXT_PBA_CLR | E1000_CTRL_EXT_EIAME;
2043 ew32(CTRL_EXT, ctrl_ext);
2044 e1e_flush();
2045}
2046
2047void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
2048{
2049 if (adapter->msix_entries) {
2050 pci_disable_msix(adapter->pdev);
2051 kfree(adapter->msix_entries);
2052 adapter->msix_entries = NULL;
2053 } else if (adapter->flags & FLAG_MSI_ENABLED) {
2054 pci_disable_msi(adapter->pdev);
2055 adapter->flags &= ~FLAG_MSI_ENABLED;
2056 }
2057}
2058
2059/**
2060 * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
2061 *
2062 * Attempt to configure interrupts using the best available
2063 * capabilities of the hardware and kernel.
2064 **/
2065void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
2066{
2067 int err;
2068 int i;
2069
2070 switch (adapter->int_mode) {
2071 case E1000E_INT_MODE_MSIX:
2072 if (adapter->flags & FLAG_HAS_MSIX) {
2073 adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
2074 adapter->msix_entries = kcalloc(adapter->num_vectors,
2075 sizeof(struct
2076 msix_entry),
2077 GFP_KERNEL);
2078 if (adapter->msix_entries) {
2079 struct e1000_adapter *a = adapter;
2080
2081 for (i = 0; i < adapter->num_vectors; i++)
2082 adapter->msix_entries[i].entry = i;
2083
2084 err = pci_enable_msix_range(a->pdev,
2085 a->msix_entries,
2086 a->num_vectors,
2087 a->num_vectors);
2088 if (err > 0)
2089 return;
2090 }
2091 /* MSI-X failed, so fall through and try MSI */
2092 e_err("Failed to initialize MSI-X interrupts. Falling back to MSI interrupts.\n");
2093 e1000e_reset_interrupt_capability(adapter);
2094 }
2095 adapter->int_mode = E1000E_INT_MODE_MSI;
2096 /* Fall through */
2097 case E1000E_INT_MODE_MSI:
2098 if (!pci_enable_msi(adapter->pdev)) {
2099 adapter->flags |= FLAG_MSI_ENABLED;
2100 } else {
2101 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2102 e_err("Failed to initialize MSI interrupts. Falling back to legacy interrupts.\n");
2103 }
2104 /* Fall through */
2105 case E1000E_INT_MODE_LEGACY:
2106 /* Don't do anything; this is the system default */
2107 break;
2108 }
2109
2110 /* store the number of vectors being used */
2111 adapter->num_vectors = 1;
2112}
2113
2114/**
2115 * e1000_request_msix - Initialize MSI-X interrupts
2116 *
2117 * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
2118 * kernel.
2119 **/
2120static int e1000_request_msix(struct e1000_adapter *adapter)
2121{
2122 struct net_device *netdev = adapter->netdev;
2123 int err = 0, vector = 0;
2124
2125 if (strlen(netdev->name) < (IFNAMSIZ - 5))
2126 snprintf(adapter->rx_ring->name,
2127 sizeof(adapter->rx_ring->name) - 1,
2128 "%s-rx-0", netdev->name);
2129 else
2130 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
2131 err = request_irq(adapter->msix_entries[vector].vector,
2132 e1000_intr_msix_rx, 0, adapter->rx_ring->name,
2133 netdev);
2134 if (err)
2135 return err;
2136 adapter->rx_ring->itr_register = adapter->hw.hw_addr +
2137 E1000_EITR_82574(vector);
2138 adapter->rx_ring->itr_val = adapter->itr;
2139 vector++;
2140
2141 if (strlen(netdev->name) < (IFNAMSIZ - 5))
2142 snprintf(adapter->tx_ring->name,
2143 sizeof(adapter->tx_ring->name) - 1,
2144 "%s-tx-0", netdev->name);
2145 else
2146 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
2147 err = request_irq(adapter->msix_entries[vector].vector,
2148 e1000_intr_msix_tx, 0, adapter->tx_ring->name,
2149 netdev);
2150 if (err)
2151 return err;
2152 adapter->tx_ring->itr_register = adapter->hw.hw_addr +
2153 E1000_EITR_82574(vector);
2154 adapter->tx_ring->itr_val = adapter->itr;
2155 vector++;
2156
2157 err = request_irq(adapter->msix_entries[vector].vector,
2158 e1000_msix_other, 0, netdev->name, netdev);
2159 if (err)
2160 return err;
2161
2162 e1000_configure_msix(adapter);
2163
2164 return 0;
2165}
2166
2167/**
2168 * e1000_request_irq - initialize interrupts
2169 *
2170 * Attempts to configure interrupts using the best available
2171 * capabilities of the hardware and kernel.
2172 **/
2173static int e1000_request_irq(struct e1000_adapter *adapter)
2174{
2175 struct net_device *netdev = adapter->netdev;
2176 int err;
2177
2178 if (adapter->msix_entries) {
2179 err = e1000_request_msix(adapter);
2180 if (!err)
2181 return err;
2182 /* fall back to MSI */
2183 e1000e_reset_interrupt_capability(adapter);
2184 adapter->int_mode = E1000E_INT_MODE_MSI;
2185 e1000e_set_interrupt_capability(adapter);
2186 }
2187 if (adapter->flags & FLAG_MSI_ENABLED) {
2188 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2189 netdev->name, netdev);
2190 if (!err)
2191 return err;
2192
2193 /* fall back to legacy interrupt */
2194 e1000e_reset_interrupt_capability(adapter);
2195 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2196 }
2197
2198 err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2199 netdev->name, netdev);
2200 if (err)
2201 e_err("Unable to allocate interrupt, Error: %d\n", err);
2202
2203 return err;
2204}
2205
2206static void e1000_free_irq(struct e1000_adapter *adapter)
2207{
2208 struct net_device *netdev = adapter->netdev;
2209
2210 if (adapter->msix_entries) {
2211 int vector = 0;
2212
2213 free_irq(adapter->msix_entries[vector].vector, netdev);
2214 vector++;
2215
2216 free_irq(adapter->msix_entries[vector].vector, netdev);
2217 vector++;
2218
2219 /* Other Causes interrupt vector */
2220 free_irq(adapter->msix_entries[vector].vector, netdev);
2221 return;
2222 }
2223
2224 free_irq(adapter->pdev->irq, netdev);
2225}
2226
2227/**
2228 * e1000_irq_disable - Mask off interrupt generation on the NIC
2229 **/
2230static void e1000_irq_disable(struct e1000_adapter *adapter)
2231{
2232 struct e1000_hw *hw = &adapter->hw;
2233
2234 ew32(IMC, ~0);
2235 if (adapter->msix_entries)
2236 ew32(EIAC_82574, 0);
2237 e1e_flush();
2238
2239 if (adapter->msix_entries) {
2240 int i;
2241
2242 for (i = 0; i < adapter->num_vectors; i++)
2243 synchronize_irq(adapter->msix_entries[i].vector);
2244 } else {
2245 synchronize_irq(adapter->pdev->irq);
2246 }
2247}
2248
2249/**
2250 * e1000_irq_enable - Enable default interrupt generation settings
2251 **/
2252static void e1000_irq_enable(struct e1000_adapter *adapter)
2253{
2254 struct e1000_hw *hw = &adapter->hw;
2255
2256 if (adapter->msix_entries) {
2257 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2258 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER |
2259 IMS_OTHER_MASK);
2260 } else if (hw->mac.type >= e1000_pch_lpt) {
2261 ew32(IMS, IMS_ENABLE_MASK | E1000_IMS_ECCER);
2262 } else {
2263 ew32(IMS, IMS_ENABLE_MASK);
2264 }
2265 e1e_flush();
2266}
2267
2268/**
2269 * e1000e_get_hw_control - get control of the h/w from f/w
2270 * @adapter: address of board private structure
2271 *
2272 * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2273 * For ASF and Pass Through versions of f/w this means that
2274 * the driver is loaded. For AMT version (only with 82573)
2275 * of the f/w this means that the network i/f is open.
2276 **/
2277void e1000e_get_hw_control(struct e1000_adapter *adapter)
2278{
2279 struct e1000_hw *hw = &adapter->hw;
2280 u32 ctrl_ext;
2281 u32 swsm;
2282
2283 /* Let firmware know the driver has taken over */
2284 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2285 swsm = er32(SWSM);
2286 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2287 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2288 ctrl_ext = er32(CTRL_EXT);
2289 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2290 }
2291}
2292
2293/**
2294 * e1000e_release_hw_control - release control of the h/w to f/w
2295 * @adapter: address of board private structure
2296 *
2297 * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2298 * For ASF and Pass Through versions of f/w this means that the
2299 * driver is no longer loaded. For AMT version (only with 82573) i
2300 * of the f/w this means that the network i/f is closed.
2301 *
2302 **/
2303void e1000e_release_hw_control(struct e1000_adapter *adapter)
2304{
2305 struct e1000_hw *hw = &adapter->hw;
2306 u32 ctrl_ext;
2307 u32 swsm;
2308
2309 /* Let firmware taken over control of h/w */
2310 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2311 swsm = er32(SWSM);
2312 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2313 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2314 ctrl_ext = er32(CTRL_EXT);
2315 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2316 }
2317}
2318
2319/**
2320 * e1000_alloc_ring_dma - allocate memory for a ring structure
2321 **/
2322static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2323 struct e1000_ring *ring)
2324{
2325 struct pci_dev *pdev = adapter->pdev;
2326
2327 ring->desc = dma_zalloc_coherent(&pdev->dev, ring->size, &ring->dma,
2328 GFP_KERNEL);
2329 if (!ring->desc)
2330 return -ENOMEM;
2331
2332 return 0;
2333}
2334
2335/**
2336 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2337 * @tx_ring: Tx descriptor ring
2338 *
2339 * Return 0 on success, negative on failure
2340 **/
2341int e1000e_setup_tx_resources(struct e1000_ring *tx_ring)
2342{
2343 struct e1000_adapter *adapter = tx_ring->adapter;
2344 int err = -ENOMEM, size;
2345
2346 size = sizeof(struct e1000_buffer) * tx_ring->count;
2347 tx_ring->buffer_info = vzalloc(size);
2348 if (!tx_ring->buffer_info)
2349 goto err;
2350
2351 /* round up to nearest 4K */
2352 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2353 tx_ring->size = ALIGN(tx_ring->size, 4096);
2354
2355 err = e1000_alloc_ring_dma(adapter, tx_ring);
2356 if (err)
2357 goto err;
2358
2359 tx_ring->next_to_use = 0;
2360 tx_ring->next_to_clean = 0;
2361
2362 return 0;
2363err:
2364 vfree(tx_ring->buffer_info);
2365 e_err("Unable to allocate memory for the transmit descriptor ring\n");
2366 return err;
2367}
2368
2369/**
2370 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2371 * @rx_ring: Rx descriptor ring
2372 *
2373 * Returns 0 on success, negative on failure
2374 **/
2375int e1000e_setup_rx_resources(struct e1000_ring *rx_ring)
2376{
2377 struct e1000_adapter *adapter = rx_ring->adapter;
2378 struct e1000_buffer *buffer_info;
2379 int i, size, desc_len, err = -ENOMEM;
2380
2381 size = sizeof(struct e1000_buffer) * rx_ring->count;
2382 rx_ring->buffer_info = vzalloc(size);
2383 if (!rx_ring->buffer_info)
2384 goto err;
2385
2386 for (i = 0; i < rx_ring->count; i++) {
2387 buffer_info = &rx_ring->buffer_info[i];
2388 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2389 sizeof(struct e1000_ps_page),
2390 GFP_KERNEL);
2391 if (!buffer_info->ps_pages)
2392 goto err_pages;
2393 }
2394
2395 desc_len = sizeof(union e1000_rx_desc_packet_split);
2396
2397 /* Round up to nearest 4K */
2398 rx_ring->size = rx_ring->count * desc_len;
2399 rx_ring->size = ALIGN(rx_ring->size, 4096);
2400
2401 err = e1000_alloc_ring_dma(adapter, rx_ring);
2402 if (err)
2403 goto err_pages;
2404
2405 rx_ring->next_to_clean = 0;
2406 rx_ring->next_to_use = 0;
2407 rx_ring->rx_skb_top = NULL;
2408
2409 return 0;
2410
2411err_pages:
2412 for (i = 0; i < rx_ring->count; i++) {
2413 buffer_info = &rx_ring->buffer_info[i];
2414 kfree(buffer_info->ps_pages);
2415 }
2416err:
2417 vfree(rx_ring->buffer_info);
2418 e_err("Unable to allocate memory for the receive descriptor ring\n");
2419 return err;
2420}
2421
2422/**
2423 * e1000_clean_tx_ring - Free Tx Buffers
2424 * @tx_ring: Tx descriptor ring
2425 **/
2426static void e1000_clean_tx_ring(struct e1000_ring *tx_ring)
2427{
2428 struct e1000_adapter *adapter = tx_ring->adapter;
2429 struct e1000_buffer *buffer_info;
2430 unsigned long size;
2431 unsigned int i;
2432
2433 for (i = 0; i < tx_ring->count; i++) {
2434 buffer_info = &tx_ring->buffer_info[i];
2435 e1000_put_txbuf(tx_ring, buffer_info, false);
2436 }
2437
2438 netdev_reset_queue(adapter->netdev);
2439 size = sizeof(struct e1000_buffer) * tx_ring->count;
2440 memset(tx_ring->buffer_info, 0, size);
2441
2442 memset(tx_ring->desc, 0, tx_ring->size);
2443
2444 tx_ring->next_to_use = 0;
2445 tx_ring->next_to_clean = 0;
2446}
2447
2448/**
2449 * e1000e_free_tx_resources - Free Tx Resources per Queue
2450 * @tx_ring: Tx descriptor ring
2451 *
2452 * Free all transmit software resources
2453 **/
2454void e1000e_free_tx_resources(struct e1000_ring *tx_ring)
2455{
2456 struct e1000_adapter *adapter = tx_ring->adapter;
2457 struct pci_dev *pdev = adapter->pdev;
2458
2459 e1000_clean_tx_ring(tx_ring);
2460
2461 vfree(tx_ring->buffer_info);
2462 tx_ring->buffer_info = NULL;
2463
2464 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2465 tx_ring->dma);
2466 tx_ring->desc = NULL;
2467}
2468
2469/**
2470 * e1000e_free_rx_resources - Free Rx Resources
2471 * @rx_ring: Rx descriptor ring
2472 *
2473 * Free all receive software resources
2474 **/
2475void e1000e_free_rx_resources(struct e1000_ring *rx_ring)
2476{
2477 struct e1000_adapter *adapter = rx_ring->adapter;
2478 struct pci_dev *pdev = adapter->pdev;
2479 int i;
2480
2481 e1000_clean_rx_ring(rx_ring);
2482
2483 for (i = 0; i < rx_ring->count; i++)
2484 kfree(rx_ring->buffer_info[i].ps_pages);
2485
2486 vfree(rx_ring->buffer_info);
2487 rx_ring->buffer_info = NULL;
2488
2489 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2490 rx_ring->dma);
2491 rx_ring->desc = NULL;
2492}
2493
2494/**
2495 * e1000_update_itr - update the dynamic ITR value based on statistics
2496 * @adapter: pointer to adapter
2497 * @itr_setting: current adapter->itr
2498 * @packets: the number of packets during this measurement interval
2499 * @bytes: the number of bytes during this measurement interval
2500 *
2501 * Stores a new ITR value based on packets and byte
2502 * counts during the last interrupt. The advantage of per interrupt
2503 * computation is faster updates and more accurate ITR for the current
2504 * traffic pattern. Constants in this function were computed
2505 * based on theoretical maximum wire speed and thresholds were set based
2506 * on testing data as well as attempting to minimize response time
2507 * while increasing bulk throughput. This functionality is controlled
2508 * by the InterruptThrottleRate module parameter.
2509 **/
2510static unsigned int e1000_update_itr(u16 itr_setting, int packets, int bytes)
2511{
2512 unsigned int retval = itr_setting;
2513
2514 if (packets == 0)
2515 return itr_setting;
2516
2517 switch (itr_setting) {
2518 case lowest_latency:
2519 /* handle TSO and jumbo frames */
2520 if (bytes / packets > 8000)
2521 retval = bulk_latency;
2522 else if ((packets < 5) && (bytes > 512))
2523 retval = low_latency;
2524 break;
2525 case low_latency: /* 50 usec aka 20000 ints/s */
2526 if (bytes > 10000) {
2527 /* this if handles the TSO accounting */
2528 if (bytes / packets > 8000)
2529 retval = bulk_latency;
2530 else if ((packets < 10) || ((bytes / packets) > 1200))
2531 retval = bulk_latency;
2532 else if ((packets > 35))
2533 retval = lowest_latency;
2534 } else if (bytes / packets > 2000) {
2535 retval = bulk_latency;
2536 } else if (packets <= 2 && bytes < 512) {
2537 retval = lowest_latency;
2538 }
2539 break;
2540 case bulk_latency: /* 250 usec aka 4000 ints/s */
2541 if (bytes > 25000) {
2542 if (packets > 35)
2543 retval = low_latency;
2544 } else if (bytes < 6000) {
2545 retval = low_latency;
2546 }
2547 break;
2548 }
2549
2550 return retval;
2551}
2552
2553static void e1000_set_itr(struct e1000_adapter *adapter)
2554{
2555 u16 current_itr;
2556 u32 new_itr = adapter->itr;
2557
2558 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2559 if (adapter->link_speed != SPEED_1000) {
2560 current_itr = 0;
2561 new_itr = 4000;
2562 goto set_itr_now;
2563 }
2564
2565 if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2566 new_itr = 0;
2567 goto set_itr_now;
2568 }
2569
2570 adapter->tx_itr = e1000_update_itr(adapter->tx_itr,
2571 adapter->total_tx_packets,
2572 adapter->total_tx_bytes);
2573 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2574 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2575 adapter->tx_itr = low_latency;
2576
2577 adapter->rx_itr = e1000_update_itr(adapter->rx_itr,
2578 adapter->total_rx_packets,
2579 adapter->total_rx_bytes);
2580 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2581 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2582 adapter->rx_itr = low_latency;
2583
2584 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2585
2586 /* counts and packets in update_itr are dependent on these numbers */
2587 switch (current_itr) {
2588 case lowest_latency:
2589 new_itr = 70000;
2590 break;
2591 case low_latency:
2592 new_itr = 20000; /* aka hwitr = ~200 */
2593 break;
2594 case bulk_latency:
2595 new_itr = 4000;
2596 break;
2597 default:
2598 break;
2599 }
2600
2601set_itr_now:
2602 if (new_itr != adapter->itr) {
2603 /* this attempts to bias the interrupt rate towards Bulk
2604 * by adding intermediate steps when interrupt rate is
2605 * increasing
2606 */
2607 new_itr = new_itr > adapter->itr ?
2608 min(adapter->itr + (new_itr >> 2), new_itr) : new_itr;
2609 adapter->itr = new_itr;
2610 adapter->rx_ring->itr_val = new_itr;
2611 if (adapter->msix_entries)
2612 adapter->rx_ring->set_itr = 1;
2613 else
2614 e1000e_write_itr(adapter, new_itr);
2615 }
2616}
2617
2618/**
2619 * e1000e_write_itr - write the ITR value to the appropriate registers
2620 * @adapter: address of board private structure
2621 * @itr: new ITR value to program
2622 *
2623 * e1000e_write_itr determines if the adapter is in MSI-X mode
2624 * and, if so, writes the EITR registers with the ITR value.
2625 * Otherwise, it writes the ITR value into the ITR register.
2626 **/
2627void e1000e_write_itr(struct e1000_adapter *adapter, u32 itr)
2628{
2629 struct e1000_hw *hw = &adapter->hw;
2630 u32 new_itr = itr ? 1000000000 / (itr * 256) : 0;
2631
2632 if (adapter->msix_entries) {
2633 int vector;
2634
2635 for (vector = 0; vector < adapter->num_vectors; vector++)
2636 writel(new_itr, hw->hw_addr + E1000_EITR_82574(vector));
2637 } else {
2638 ew32(ITR, new_itr);
2639 }
2640}
2641
2642/**
2643 * e1000_alloc_queues - Allocate memory for all rings
2644 * @adapter: board private structure to initialize
2645 **/
2646static int e1000_alloc_queues(struct e1000_adapter *adapter)
2647{
2648 int size = sizeof(struct e1000_ring);
2649
2650 adapter->tx_ring = kzalloc(size, GFP_KERNEL);
2651 if (!adapter->tx_ring)
2652 goto err;
2653 adapter->tx_ring->count = adapter->tx_ring_count;
2654 adapter->tx_ring->adapter = adapter;
2655
2656 adapter->rx_ring = kzalloc(size, GFP_KERNEL);
2657 if (!adapter->rx_ring)
2658 goto err;
2659 adapter->rx_ring->count = adapter->rx_ring_count;
2660 adapter->rx_ring->adapter = adapter;
2661
2662 return 0;
2663err:
2664 e_err("Unable to allocate memory for queues\n");
2665 kfree(adapter->rx_ring);
2666 kfree(adapter->tx_ring);
2667 return -ENOMEM;
2668}
2669
2670/**
2671 * e1000e_poll - NAPI Rx polling callback
2672 * @napi: struct associated with this polling callback
2673 * @weight: number of packets driver is allowed to process this poll
2674 **/
2675static int e1000e_poll(struct napi_struct *napi, int weight)
2676{
2677 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter,
2678 napi);
2679 struct e1000_hw *hw = &adapter->hw;
2680 struct net_device *poll_dev = adapter->netdev;
2681 int tx_cleaned = 1, work_done = 0;
2682
2683 adapter = netdev_priv(poll_dev);
2684
2685 if (!adapter->msix_entries ||
2686 (adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2687 tx_cleaned = e1000_clean_tx_irq(adapter->tx_ring);
2688
2689 adapter->clean_rx(adapter->rx_ring, &work_done, weight);
2690
2691 if (!tx_cleaned)
2692 work_done = weight;
2693
2694 /* If weight not fully consumed, exit the polling mode */
2695 if (work_done < weight) {
2696 if (adapter->itr_setting & 3)
2697 e1000_set_itr(adapter);
2698 napi_complete_done(napi, work_done);
2699 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2700 if (adapter->msix_entries)
2701 ew32(IMS, adapter->rx_ring->ims_val);
2702 else
2703 e1000_irq_enable(adapter);
2704 }
2705 }
2706
2707 return work_done;
2708}
2709
2710static int e1000_vlan_rx_add_vid(struct net_device *netdev,
2711 __always_unused __be16 proto, u16 vid)
2712{
2713 struct e1000_adapter *adapter = netdev_priv(netdev);
2714 struct e1000_hw *hw = &adapter->hw;
2715 u32 vfta, index;
2716
2717 /* don't update vlan cookie if already programmed */
2718 if ((adapter->hw.mng_cookie.status &
2719 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2720 (vid == adapter->mng_vlan_id))
2721 return 0;
2722
2723 /* add VID to filter table */
2724 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2725 index = (vid >> 5) & 0x7F;
2726 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2727 vfta |= BIT((vid & 0x1F));
2728 hw->mac.ops.write_vfta(hw, index, vfta);
2729 }
2730
2731 set_bit(vid, adapter->active_vlans);
2732
2733 return 0;
2734}
2735
2736static int e1000_vlan_rx_kill_vid(struct net_device *netdev,
2737 __always_unused __be16 proto, u16 vid)
2738{
2739 struct e1000_adapter *adapter = netdev_priv(netdev);
2740 struct e1000_hw *hw = &adapter->hw;
2741 u32 vfta, index;
2742
2743 if ((adapter->hw.mng_cookie.status &
2744 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2745 (vid == adapter->mng_vlan_id)) {
2746 /* release control to f/w */
2747 e1000e_release_hw_control(adapter);
2748 return 0;
2749 }
2750
2751 /* remove VID from filter table */
2752 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2753 index = (vid >> 5) & 0x7F;
2754 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2755 vfta &= ~BIT((vid & 0x1F));
2756 hw->mac.ops.write_vfta(hw, index, vfta);
2757 }
2758
2759 clear_bit(vid, adapter->active_vlans);
2760
2761 return 0;
2762}
2763
2764/**
2765 * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2766 * @adapter: board private structure to initialize
2767 **/
2768static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2769{
2770 struct net_device *netdev = adapter->netdev;
2771 struct e1000_hw *hw = &adapter->hw;
2772 u32 rctl;
2773
2774 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2775 /* disable VLAN receive filtering */
2776 rctl = er32(RCTL);
2777 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2778 ew32(RCTL, rctl);
2779
2780 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2781 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
2782 adapter->mng_vlan_id);
2783 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2784 }
2785 }
2786}
2787
2788/**
2789 * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2790 * @adapter: board private structure to initialize
2791 **/
2792static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2793{
2794 struct e1000_hw *hw = &adapter->hw;
2795 u32 rctl;
2796
2797 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2798 /* enable VLAN receive filtering */
2799 rctl = er32(RCTL);
2800 rctl |= E1000_RCTL_VFE;
2801 rctl &= ~E1000_RCTL_CFIEN;
2802 ew32(RCTL, rctl);
2803 }
2804}
2805
2806/**
2807 * e1000e_vlan_strip_disable - helper to disable HW VLAN stripping
2808 * @adapter: board private structure to initialize
2809 **/
2810static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2811{
2812 struct e1000_hw *hw = &adapter->hw;
2813 u32 ctrl;
2814
2815 /* disable VLAN tag insert/strip */
2816 ctrl = er32(CTRL);
2817 ctrl &= ~E1000_CTRL_VME;
2818 ew32(CTRL, ctrl);
2819}
2820
2821/**
2822 * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2823 * @adapter: board private structure to initialize
2824 **/
2825static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2826{
2827 struct e1000_hw *hw = &adapter->hw;
2828 u32 ctrl;
2829
2830 /* enable VLAN tag insert/strip */
2831 ctrl = er32(CTRL);
2832 ctrl |= E1000_CTRL_VME;
2833 ew32(CTRL, ctrl);
2834}
2835
2836static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2837{
2838 struct net_device *netdev = adapter->netdev;
2839 u16 vid = adapter->hw.mng_cookie.vlan_id;
2840 u16 old_vid = adapter->mng_vlan_id;
2841
2842 if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2843 e1000_vlan_rx_add_vid(netdev, htons(ETH_P_8021Q), vid);
2844 adapter->mng_vlan_id = vid;
2845 }
2846
2847 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2848 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q), old_vid);
2849}
2850
2851static void e1000_restore_vlan(struct e1000_adapter *adapter)
2852{
2853 u16 vid;
2854
2855 e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
2856
2857 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2858 e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
2859}
2860
2861static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2862{
2863 struct e1000_hw *hw = &adapter->hw;
2864 u32 manc, manc2h, mdef, i, j;
2865
2866 if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2867 return;
2868
2869 manc = er32(MANC);
2870
2871 /* enable receiving management packets to the host. this will probably
2872 * generate destination unreachable messages from the host OS, but
2873 * the packets will be handled on SMBUS
2874 */
2875 manc |= E1000_MANC_EN_MNG2HOST;
2876 manc2h = er32(MANC2H);
2877
2878 switch (hw->mac.type) {
2879 default:
2880 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2881 break;
2882 case e1000_82574:
2883 case e1000_82583:
2884 /* Check if IPMI pass-through decision filter already exists;
2885 * if so, enable it.
2886 */
2887 for (i = 0, j = 0; i < 8; i++) {
2888 mdef = er32(MDEF(i));
2889
2890 /* Ignore filters with anything other than IPMI ports */
2891 if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2892 continue;
2893
2894 /* Enable this decision filter in MANC2H */
2895 if (mdef)
2896 manc2h |= BIT(i);
2897
2898 j |= mdef;
2899 }
2900
2901 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2902 break;
2903
2904 /* Create new decision filter in an empty filter */
2905 for (i = 0, j = 0; i < 8; i++)
2906 if (er32(MDEF(i)) == 0) {
2907 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2908 E1000_MDEF_PORT_664));
2909 manc2h |= BIT(1);
2910 j++;
2911 break;
2912 }
2913
2914 if (!j)
2915 e_warn("Unable to create IPMI pass-through filter\n");
2916 break;
2917 }
2918
2919 ew32(MANC2H, manc2h);
2920 ew32(MANC, manc);
2921}
2922
2923/**
2924 * e1000_configure_tx - Configure Transmit Unit after Reset
2925 * @adapter: board private structure
2926 *
2927 * Configure the Tx unit of the MAC after a reset.
2928 **/
2929static void e1000_configure_tx(struct e1000_adapter *adapter)
2930{
2931 struct e1000_hw *hw = &adapter->hw;
2932 struct e1000_ring *tx_ring = adapter->tx_ring;
2933 u64 tdba;
2934 u32 tdlen, tctl, tarc;
2935
2936 /* Setup the HW Tx Head and Tail descriptor pointers */
2937 tdba = tx_ring->dma;
2938 tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2939 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
2940 ew32(TDBAH(0), (tdba >> 32));
2941 ew32(TDLEN(0), tdlen);
2942 ew32(TDH(0), 0);
2943 ew32(TDT(0), 0);
2944 tx_ring->head = adapter->hw.hw_addr + E1000_TDH(0);
2945 tx_ring->tail = adapter->hw.hw_addr + E1000_TDT(0);
2946
2947 writel(0, tx_ring->head);
2948 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
2949 e1000e_update_tdt_wa(tx_ring, 0);
2950 else
2951 writel(0, tx_ring->tail);
2952
2953 /* Set the Tx Interrupt Delay register */
2954 ew32(TIDV, adapter->tx_int_delay);
2955 /* Tx irq moderation */
2956 ew32(TADV, adapter->tx_abs_int_delay);
2957
2958 if (adapter->flags2 & FLAG2_DMA_BURST) {
2959 u32 txdctl = er32(TXDCTL(0));
2960
2961 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2962 E1000_TXDCTL_WTHRESH);
2963 /* set up some performance related parameters to encourage the
2964 * hardware to use the bus more efficiently in bursts, depends
2965 * on the tx_int_delay to be enabled,
2966 * wthresh = 1 ==> burst write is disabled to avoid Tx stalls
2967 * hthresh = 1 ==> prefetch when one or more available
2968 * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2969 * BEWARE: this seems to work but should be considered first if
2970 * there are Tx hangs or other Tx related bugs
2971 */
2972 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2973 ew32(TXDCTL(0), txdctl);
2974 }
2975 /* erratum work around: set txdctl the same for both queues */
2976 ew32(TXDCTL(1), er32(TXDCTL(0)));
2977
2978 /* Program the Transmit Control Register */
2979 tctl = er32(TCTL);
2980 tctl &= ~E1000_TCTL_CT;
2981 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2982 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2983
2984 if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2985 tarc = er32(TARC(0));
2986 /* set the speed mode bit, we'll clear it if we're not at
2987 * gigabit link later
2988 */
2989#define SPEED_MODE_BIT BIT(21)
2990 tarc |= SPEED_MODE_BIT;
2991 ew32(TARC(0), tarc);
2992 }
2993
2994 /* errata: program both queues to unweighted RR */
2995 if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2996 tarc = er32(TARC(0));
2997 tarc |= 1;
2998 ew32(TARC(0), tarc);
2999 tarc = er32(TARC(1));
3000 tarc |= 1;
3001 ew32(TARC(1), tarc);
3002 }
3003
3004 /* Setup Transmit Descriptor Settings for eop descriptor */
3005 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
3006
3007 /* only set IDE if we are delaying interrupts using the timers */
3008 if (adapter->tx_int_delay)
3009 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
3010
3011 /* enable Report Status bit */
3012 adapter->txd_cmd |= E1000_TXD_CMD_RS;
3013
3014 ew32(TCTL, tctl);
3015
3016 hw->mac.ops.config_collision_dist(hw);
3017
3018 /* SPT and KBL Si errata workaround to avoid data corruption */
3019 if (hw->mac.type == e1000_pch_spt) {
3020 u32 reg_val;
3021
3022 reg_val = er32(IOSFPC);
3023 reg_val |= E1000_RCTL_RDMTS_HEX;
3024 ew32(IOSFPC, reg_val);
3025
3026 reg_val = er32(TARC(0));
3027 /* SPT and KBL Si errata workaround to avoid Tx hang.
3028 * Dropping the number of outstanding requests from
3029 * 3 to 2 in order to avoid a buffer overrun.
3030 */
3031 reg_val &= ~E1000_TARC0_CB_MULTIQ_3_REQ;
3032 reg_val |= E1000_TARC0_CB_MULTIQ_2_REQ;
3033 ew32(TARC(0), reg_val);
3034 }
3035}
3036
3037/**
3038 * e1000_setup_rctl - configure the receive control registers
3039 * @adapter: Board private structure
3040 **/
3041#define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
3042 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
3043static void e1000_setup_rctl(struct e1000_adapter *adapter)
3044{
3045 struct e1000_hw *hw = &adapter->hw;
3046 u32 rctl, rfctl;
3047 u32 pages = 0;
3048
3049 /* Workaround Si errata on PCHx - configure jumbo frame flow.
3050 * If jumbo frames not set, program related MAC/PHY registers
3051 * to h/w defaults
3052 */
3053 if (hw->mac.type >= e1000_pch2lan) {
3054 s32 ret_val;
3055
3056 if (adapter->netdev->mtu > ETH_DATA_LEN)
3057 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
3058 else
3059 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
3060
3061 if (ret_val)
3062 e_dbg("failed to enable|disable jumbo frame workaround mode\n");
3063 }
3064
3065 /* Program MC offset vector base */
3066 rctl = er32(RCTL);
3067 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3068 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
3069 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
3070 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3071
3072 /* Do not Store bad packets */
3073 rctl &= ~E1000_RCTL_SBP;
3074
3075 /* Enable Long Packet receive */
3076 if (adapter->netdev->mtu <= ETH_DATA_LEN)
3077 rctl &= ~E1000_RCTL_LPE;
3078 else
3079 rctl |= E1000_RCTL_LPE;
3080
3081 /* Some systems expect that the CRC is included in SMBUS traffic. The
3082 * hardware strips the CRC before sending to both SMBUS (BMC) and to
3083 * host memory when this is enabled
3084 */
3085 if (adapter->flags2 & FLAG2_CRC_STRIPPING)
3086 rctl |= E1000_RCTL_SECRC;
3087
3088 /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
3089 if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
3090 u16 phy_data;
3091
3092 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
3093 phy_data &= 0xfff8;
3094 phy_data |= BIT(2);
3095 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
3096
3097 e1e_rphy(hw, 22, &phy_data);
3098 phy_data &= 0x0fff;
3099 phy_data |= BIT(14);
3100 e1e_wphy(hw, 0x10, 0x2823);
3101 e1e_wphy(hw, 0x11, 0x0003);
3102 e1e_wphy(hw, 22, phy_data);
3103 }
3104
3105 /* Setup buffer sizes */
3106 rctl &= ~E1000_RCTL_SZ_4096;
3107 rctl |= E1000_RCTL_BSEX;
3108 switch (adapter->rx_buffer_len) {
3109 case 2048:
3110 default:
3111 rctl |= E1000_RCTL_SZ_2048;
3112 rctl &= ~E1000_RCTL_BSEX;
3113 break;
3114 case 4096:
3115 rctl |= E1000_RCTL_SZ_4096;
3116 break;
3117 case 8192:
3118 rctl |= E1000_RCTL_SZ_8192;
3119 break;
3120 case 16384:
3121 rctl |= E1000_RCTL_SZ_16384;
3122 break;
3123 }
3124
3125 /* Enable Extended Status in all Receive Descriptors */
3126 rfctl = er32(RFCTL);
3127 rfctl |= E1000_RFCTL_EXTEN;
3128 ew32(RFCTL, rfctl);
3129
3130 /* 82571 and greater support packet-split where the protocol
3131 * header is placed in skb->data and the packet data is
3132 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
3133 * In the case of a non-split, skb->data is linearly filled,
3134 * followed by the page buffers. Therefore, skb->data is
3135 * sized to hold the largest protocol header.
3136 *
3137 * allocations using alloc_page take too long for regular MTU
3138 * so only enable packet split for jumbo frames
3139 *
3140 * Using pages when the page size is greater than 16k wastes
3141 * a lot of memory, since we allocate 3 pages at all times
3142 * per packet.
3143 */
3144 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
3145 if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
3146 adapter->rx_ps_pages = pages;
3147 else
3148 adapter->rx_ps_pages = 0;
3149
3150 if (adapter->rx_ps_pages) {
3151 u32 psrctl = 0;
3152
3153 /* Enable Packet split descriptors */
3154 rctl |= E1000_RCTL_DTYP_PS;
3155
3156 psrctl |= adapter->rx_ps_bsize0 >> E1000_PSRCTL_BSIZE0_SHIFT;
3157
3158 switch (adapter->rx_ps_pages) {
3159 case 3:
3160 psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE3_SHIFT;
3161 /* fall-through */
3162 case 2:
3163 psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE2_SHIFT;
3164 /* fall-through */
3165 case 1:
3166 psrctl |= PAGE_SIZE >> E1000_PSRCTL_BSIZE1_SHIFT;
3167 break;
3168 }
3169
3170 ew32(PSRCTL, psrctl);
3171 }
3172
3173 /* This is useful for sniffing bad packets. */
3174 if (adapter->netdev->features & NETIF_F_RXALL) {
3175 /* UPE and MPE will be handled by normal PROMISC logic
3176 * in e1000e_set_rx_mode
3177 */
3178 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3179 E1000_RCTL_BAM | /* RX All Bcast Pkts */
3180 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3181
3182 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3183 E1000_RCTL_DPF | /* Allow filtered pause */
3184 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3185 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3186 * and that breaks VLANs.
3187 */
3188 }
3189
3190 ew32(RCTL, rctl);
3191 /* just started the receive unit, no need to restart */
3192 adapter->flags &= ~FLAG_RESTART_NOW;
3193}
3194
3195/**
3196 * e1000_configure_rx - Configure Receive Unit after Reset
3197 * @adapter: board private structure
3198 *
3199 * Configure the Rx unit of the MAC after a reset.
3200 **/
3201static void e1000_configure_rx(struct e1000_adapter *adapter)
3202{
3203 struct e1000_hw *hw = &adapter->hw;
3204 struct e1000_ring *rx_ring = adapter->rx_ring;
3205 u64 rdba;
3206 u32 rdlen, rctl, rxcsum, ctrl_ext;
3207
3208 if (adapter->rx_ps_pages) {
3209 /* this is a 32 byte descriptor */
3210 rdlen = rx_ring->count *
3211 sizeof(union e1000_rx_desc_packet_split);
3212 adapter->clean_rx = e1000_clean_rx_irq_ps;
3213 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3214 } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3215 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3216 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3217 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3218 } else {
3219 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3220 adapter->clean_rx = e1000_clean_rx_irq;
3221 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3222 }
3223
3224 /* disable receives while setting up the descriptors */
3225 rctl = er32(RCTL);
3226 if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3227 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3228 e1e_flush();
3229 usleep_range(10000, 20000);
3230
3231 if (adapter->flags2 & FLAG2_DMA_BURST) {
3232 /* set the writeback threshold (only takes effect if the RDTR
3233 * is set). set GRAN=1 and write back up to 0x4 worth, and
3234 * enable prefetching of 0x20 Rx descriptors
3235 * granularity = 01
3236 * wthresh = 04,
3237 * hthresh = 04,
3238 * pthresh = 0x20
3239 */
3240 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3241 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3242 }
3243
3244 /* set the Receive Delay Timer Register */
3245 ew32(RDTR, adapter->rx_int_delay);
3246
3247 /* irq moderation */
3248 ew32(RADV, adapter->rx_abs_int_delay);
3249 if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3250 e1000e_write_itr(adapter, adapter->itr);
3251
3252 ctrl_ext = er32(CTRL_EXT);
3253 /* Auto-Mask interrupts upon ICR access */
3254 ctrl_ext |= E1000_CTRL_EXT_IAME;
3255 ew32(IAM, 0xffffffff);
3256 ew32(CTRL_EXT, ctrl_ext);
3257 e1e_flush();
3258
3259 /* Setup the HW Rx Head and Tail Descriptor Pointers and
3260 * the Base and Length of the Rx Descriptor Ring
3261 */
3262 rdba = rx_ring->dma;
3263 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
3264 ew32(RDBAH(0), (rdba >> 32));
3265 ew32(RDLEN(0), rdlen);
3266 ew32(RDH(0), 0);
3267 ew32(RDT(0), 0);
3268 rx_ring->head = adapter->hw.hw_addr + E1000_RDH(0);
3269 rx_ring->tail = adapter->hw.hw_addr + E1000_RDT(0);
3270
3271 writel(0, rx_ring->head);
3272 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
3273 e1000e_update_rdt_wa(rx_ring, 0);
3274 else
3275 writel(0, rx_ring->tail);
3276
3277 /* Enable Receive Checksum Offload for TCP and UDP */
3278 rxcsum = er32(RXCSUM);
3279 if (adapter->netdev->features & NETIF_F_RXCSUM)
3280 rxcsum |= E1000_RXCSUM_TUOFL;
3281 else
3282 rxcsum &= ~E1000_RXCSUM_TUOFL;
3283 ew32(RXCSUM, rxcsum);
3284
3285 /* With jumbo frames, excessive C-state transition latencies result
3286 * in dropped transactions.
3287 */
3288 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3289 u32 lat =
3290 ((er32(PBA) & E1000_PBA_RXA_MASK) * 1024 -
3291 adapter->max_frame_size) * 8 / 1000;
3292
3293 if (adapter->flags & FLAG_IS_ICH) {
3294 u32 rxdctl = er32(RXDCTL(0));
3295
3296 ew32(RXDCTL(0), rxdctl | 0x3 | BIT(8));
3297 }
3298
3299 dev_info(&adapter->pdev->dev,
3300 "Some CPU C-states have been disabled in order to enable jumbo frames\n");
3301 pm_qos_update_request(&adapter->pm_qos_req, lat);
3302 } else {
3303 pm_qos_update_request(&adapter->pm_qos_req,
3304 PM_QOS_DEFAULT_VALUE);
3305 }
3306
3307 /* Enable Receives */
3308 ew32(RCTL, rctl);
3309}
3310
3311/**
3312 * e1000e_write_mc_addr_list - write multicast addresses to MTA
3313 * @netdev: network interface device structure
3314 *
3315 * Writes multicast address list to the MTA hash table.
3316 * Returns: -ENOMEM on failure
3317 * 0 on no addresses written
3318 * X on writing X addresses to MTA
3319 */
3320static int e1000e_write_mc_addr_list(struct net_device *netdev)
3321{
3322 struct e1000_adapter *adapter = netdev_priv(netdev);
3323 struct e1000_hw *hw = &adapter->hw;
3324 struct netdev_hw_addr *ha;
3325 u8 *mta_list;
3326 int i;
3327
3328 if (netdev_mc_empty(netdev)) {
3329 /* nothing to program, so clear mc list */
3330 hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3331 return 0;
3332 }
3333
3334 mta_list = kzalloc(netdev_mc_count(netdev) * ETH_ALEN, GFP_ATOMIC);
3335 if (!mta_list)
3336 return -ENOMEM;
3337
3338 /* update_mc_addr_list expects a packed array of only addresses. */
3339 i = 0;
3340 netdev_for_each_mc_addr(ha, netdev)
3341 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3342
3343 hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3344 kfree(mta_list);
3345
3346 return netdev_mc_count(netdev);
3347}
3348
3349/**
3350 * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3351 * @netdev: network interface device structure
3352 *
3353 * Writes unicast address list to the RAR table.
3354 * Returns: -ENOMEM on failure/insufficient address space
3355 * 0 on no addresses written
3356 * X on writing X addresses to the RAR table
3357 **/
3358static int e1000e_write_uc_addr_list(struct net_device *netdev)
3359{
3360 struct e1000_adapter *adapter = netdev_priv(netdev);
3361 struct e1000_hw *hw = &adapter->hw;
3362 unsigned int rar_entries;
3363 int count = 0;
3364
3365 rar_entries = hw->mac.ops.rar_get_count(hw);
3366
3367 /* save a rar entry for our hardware address */
3368 rar_entries--;
3369
3370 /* save a rar entry for the LAA workaround */
3371 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3372 rar_entries--;
3373
3374 /* return ENOMEM indicating insufficient memory for addresses */
3375 if (netdev_uc_count(netdev) > rar_entries)
3376 return -ENOMEM;
3377
3378 if (!netdev_uc_empty(netdev) && rar_entries) {
3379 struct netdev_hw_addr *ha;
3380
3381 /* write the addresses in reverse order to avoid write
3382 * combining
3383 */
3384 netdev_for_each_uc_addr(ha, netdev) {
3385 int ret_val;
3386
3387 if (!rar_entries)
3388 break;
3389 ret_val = hw->mac.ops.rar_set(hw, ha->addr, rar_entries--);
3390 if (ret_val < 0)
3391 return -ENOMEM;
3392 count++;
3393 }
3394 }
3395
3396 /* zero out the remaining RAR entries not used above */
3397 for (; rar_entries > 0; rar_entries--) {
3398 ew32(RAH(rar_entries), 0);
3399 ew32(RAL(rar_entries), 0);
3400 }
3401 e1e_flush();
3402
3403 return count;
3404}
3405
3406/**
3407 * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3408 * @netdev: network interface device structure
3409 *
3410 * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3411 * address list or the network interface flags are updated. This routine is
3412 * responsible for configuring the hardware for proper unicast, multicast,
3413 * promiscuous mode, and all-multi behavior.
3414 **/
3415static void e1000e_set_rx_mode(struct net_device *netdev)
3416{
3417 struct e1000_adapter *adapter = netdev_priv(netdev);
3418 struct e1000_hw *hw = &adapter->hw;
3419 u32 rctl;
3420
3421 if (pm_runtime_suspended(netdev->dev.parent))
3422 return;
3423
3424 /* Check for Promiscuous and All Multicast modes */
3425 rctl = er32(RCTL);
3426
3427 /* clear the affected bits */
3428 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3429
3430 if (netdev->flags & IFF_PROMISC) {
3431 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3432 /* Do not hardware filter VLANs in promisc mode */
3433 e1000e_vlan_filter_disable(adapter);
3434 } else {
3435 int count;
3436
3437 if (netdev->flags & IFF_ALLMULTI) {
3438 rctl |= E1000_RCTL_MPE;
3439 } else {
3440 /* Write addresses to the MTA, if the attempt fails
3441 * then we should just turn on promiscuous mode so
3442 * that we can at least receive multicast traffic
3443 */
3444 count = e1000e_write_mc_addr_list(netdev);
3445 if (count < 0)
3446 rctl |= E1000_RCTL_MPE;
3447 }
3448 e1000e_vlan_filter_enable(adapter);
3449 /* Write addresses to available RAR registers, if there is not
3450 * sufficient space to store all the addresses then enable
3451 * unicast promiscuous mode
3452 */
3453 count = e1000e_write_uc_addr_list(netdev);
3454 if (count < 0)
3455 rctl |= E1000_RCTL_UPE;
3456 }
3457
3458 ew32(RCTL, rctl);
3459
3460 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
3461 e1000e_vlan_strip_enable(adapter);
3462 else
3463 e1000e_vlan_strip_disable(adapter);
3464}
3465
3466static void e1000e_setup_rss_hash(struct e1000_adapter *adapter)
3467{
3468 struct e1000_hw *hw = &adapter->hw;
3469 u32 mrqc, rxcsum;
3470 u32 rss_key[10];
3471 int i;
3472
3473 netdev_rss_key_fill(rss_key, sizeof(rss_key));
3474 for (i = 0; i < 10; i++)
3475 ew32(RSSRK(i), rss_key[i]);
3476
3477 /* Direct all traffic to queue 0 */
3478 for (i = 0; i < 32; i++)
3479 ew32(RETA(i), 0);
3480
3481 /* Disable raw packet checksumming so that RSS hash is placed in
3482 * descriptor on writeback.
3483 */
3484 rxcsum = er32(RXCSUM);
3485 rxcsum |= E1000_RXCSUM_PCSD;
3486
3487 ew32(RXCSUM, rxcsum);
3488
3489 mrqc = (E1000_MRQC_RSS_FIELD_IPV4 |
3490 E1000_MRQC_RSS_FIELD_IPV4_TCP |
3491 E1000_MRQC_RSS_FIELD_IPV6 |
3492 E1000_MRQC_RSS_FIELD_IPV6_TCP |
3493 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
3494
3495 ew32(MRQC, mrqc);
3496}
3497
3498/**
3499 * e1000e_get_base_timinca - get default SYSTIM time increment attributes
3500 * @adapter: board private structure
3501 * @timinca: pointer to returned time increment attributes
3502 *
3503 * Get attributes for incrementing the System Time Register SYSTIML/H at
3504 * the default base frequency, and set the cyclecounter shift value.
3505 **/
3506s32 e1000e_get_base_timinca(struct e1000_adapter *adapter, u32 *timinca)
3507{
3508 struct e1000_hw *hw = &adapter->hw;
3509 u32 incvalue, incperiod, shift;
3510
3511 /* Make sure clock is enabled on I217/I218/I219 before checking
3512 * the frequency
3513 */
3514 if ((hw->mac.type >= e1000_pch_lpt) &&
3515 !(er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) &&
3516 !(er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_ENABLED)) {
3517 u32 fextnvm7 = er32(FEXTNVM7);
3518
3519 if (!(fextnvm7 & BIT(0))) {
3520 ew32(FEXTNVM7, fextnvm7 | BIT(0));
3521 e1e_flush();
3522 }
3523 }
3524
3525 switch (hw->mac.type) {
3526 case e1000_pch2lan:
3527 /* Stable 96MHz frequency */
3528 incperiod = INCPERIOD_96MHZ;
3529 incvalue = INCVALUE_96MHZ;
3530 shift = INCVALUE_SHIFT_96MHZ;
3531 adapter->cc.shift = shift + INCPERIOD_SHIFT_96MHZ;
3532 break;
3533 case e1000_pch_lpt:
3534 if (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI) {
3535 /* Stable 96MHz frequency */
3536 incperiod = INCPERIOD_96MHZ;
3537 incvalue = INCVALUE_96MHZ;
3538 shift = INCVALUE_SHIFT_96MHZ;
3539 adapter->cc.shift = shift + INCPERIOD_SHIFT_96MHZ;
3540 } else {
3541 /* Stable 25MHz frequency */
3542 incperiod = INCPERIOD_25MHZ;
3543 incvalue = INCVALUE_25MHZ;
3544 shift = INCVALUE_SHIFT_25MHZ;
3545 adapter->cc.shift = shift;
3546 }
3547 break;
3548 case e1000_pch_spt:
3549 if (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI) {
3550 /* Stable 24MHz frequency */
3551 incperiod = INCPERIOD_24MHZ;
3552 incvalue = INCVALUE_24MHZ;
3553 shift = INCVALUE_SHIFT_24MHZ;
3554 adapter->cc.shift = shift;
3555 break;
3556 }
3557 return -EINVAL;
3558 case e1000_pch_cnp:
3559 if (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI) {
3560 /* Stable 24MHz frequency */
3561 incperiod = INCPERIOD_24MHZ;
3562 incvalue = INCVALUE_24MHZ;
3563 shift = INCVALUE_SHIFT_24MHZ;
3564 adapter->cc.shift = shift;
3565 } else {
3566 /* Stable 38400KHz frequency */
3567 incperiod = INCPERIOD_38400KHZ;
3568 incvalue = INCVALUE_38400KHZ;
3569 shift = INCVALUE_SHIFT_38400KHZ;
3570 adapter->cc.shift = shift;
3571 }
3572 break;
3573 case e1000_82574:
3574 case e1000_82583:
3575 /* Stable 25MHz frequency */
3576 incperiod = INCPERIOD_25MHZ;
3577 incvalue = INCVALUE_25MHZ;
3578 shift = INCVALUE_SHIFT_25MHZ;
3579 adapter->cc.shift = shift;
3580 break;
3581 default:
3582 return -EINVAL;
3583 }
3584
3585 *timinca = ((incperiod << E1000_TIMINCA_INCPERIOD_SHIFT) |
3586 ((incvalue << shift) & E1000_TIMINCA_INCVALUE_MASK));
3587
3588 return 0;
3589}
3590
3591/**
3592 * e1000e_config_hwtstamp - configure the hwtstamp registers and enable/disable
3593 * @adapter: board private structure
3594 *
3595 * Outgoing time stamping can be enabled and disabled. Play nice and
3596 * disable it when requested, although it shouldn't cause any overhead
3597 * when no packet needs it. At most one packet in the queue may be
3598 * marked for time stamping, otherwise it would be impossible to tell
3599 * for sure to which packet the hardware time stamp belongs.
3600 *
3601 * Incoming time stamping has to be configured via the hardware filters.
3602 * Not all combinations are supported, in particular event type has to be
3603 * specified. Matching the kind of event packet is not supported, with the
3604 * exception of "all V2 events regardless of level 2 or 4".
3605 **/
3606static int e1000e_config_hwtstamp(struct e1000_adapter *adapter,
3607 struct hwtstamp_config *config)
3608{
3609 struct e1000_hw *hw = &adapter->hw;
3610 u32 tsync_tx_ctl = E1000_TSYNCTXCTL_ENABLED;
3611 u32 tsync_rx_ctl = E1000_TSYNCRXCTL_ENABLED;
3612 u32 rxmtrl = 0;
3613 u16 rxudp = 0;
3614 bool is_l4 = false;
3615 bool is_l2 = false;
3616 u32 regval;
3617
3618 if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP))
3619 return -EINVAL;
3620
3621 /* flags reserved for future extensions - must be zero */
3622 if (config->flags)
3623 return -EINVAL;
3624
3625 switch (config->tx_type) {
3626 case HWTSTAMP_TX_OFF:
3627 tsync_tx_ctl = 0;
3628 break;
3629 case HWTSTAMP_TX_ON:
3630 break;
3631 default:
3632 return -ERANGE;
3633 }
3634
3635 switch (config->rx_filter) {
3636 case HWTSTAMP_FILTER_NONE:
3637 tsync_rx_ctl = 0;
3638 break;
3639 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3640 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3641 rxmtrl = E1000_RXMTRL_PTP_V1_SYNC_MESSAGE;
3642 is_l4 = true;
3643 break;
3644 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3645 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3646 rxmtrl = E1000_RXMTRL_PTP_V1_DELAY_REQ_MESSAGE;
3647 is_l4 = true;
3648 break;
3649 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
3650 /* Also time stamps V2 L2 Path Delay Request/Response */
3651 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3652 rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3653 is_l2 = true;
3654 break;
3655 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
3656 /* Also time stamps V2 L2 Path Delay Request/Response. */
3657 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3658 rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3659 is_l2 = true;
3660 break;
3661 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3662 /* Hardware cannot filter just V2 L4 Sync messages;
3663 * fall-through to V2 (both L2 and L4) Sync.
3664 */
3665 case HWTSTAMP_FILTER_PTP_V2_SYNC:
3666 /* Also time stamps V2 Path Delay Request/Response. */
3667 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3668 rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3669 is_l2 = true;
3670 is_l4 = true;
3671 break;
3672 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3673 /* Hardware cannot filter just V2 L4 Delay Request messages;
3674 * fall-through to V2 (both L2 and L4) Delay Request.
3675 */
3676 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
3677 /* Also time stamps V2 Path Delay Request/Response. */
3678 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3679 rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3680 is_l2 = true;
3681 is_l4 = true;
3682 break;
3683 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3684 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
3685 /* Hardware cannot filter just V2 L4 or L2 Event messages;
3686 * fall-through to all V2 (both L2 and L4) Events.
3687 */
3688 case HWTSTAMP_FILTER_PTP_V2_EVENT:
3689 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_EVENT_V2;
3690 config->rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
3691 is_l2 = true;
3692 is_l4 = true;
3693 break;
3694 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3695 /* For V1, the hardware can only filter Sync messages or
3696 * Delay Request messages but not both so fall-through to
3697 * time stamp all packets.
3698 */
3699 case HWTSTAMP_FILTER_NTP_ALL:
3700 case HWTSTAMP_FILTER_ALL:
3701 is_l2 = true;
3702 is_l4 = true;
3703 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_ALL;
3704 config->rx_filter = HWTSTAMP_FILTER_ALL;
3705 break;
3706 default:
3707 return -ERANGE;
3708 }
3709
3710 adapter->hwtstamp_config = *config;
3711
3712 /* enable/disable Tx h/w time stamping */
3713 regval = er32(TSYNCTXCTL);
3714 regval &= ~E1000_TSYNCTXCTL_ENABLED;
3715 regval |= tsync_tx_ctl;
3716 ew32(TSYNCTXCTL, regval);
3717 if ((er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) !=
3718 (regval & E1000_TSYNCTXCTL_ENABLED)) {
3719 e_err("Timesync Tx Control register not set as expected\n");
3720 return -EAGAIN;
3721 }
3722
3723 /* enable/disable Rx h/w time stamping */
3724 regval = er32(TSYNCRXCTL);
3725 regval &= ~(E1000_TSYNCRXCTL_ENABLED | E1000_TSYNCRXCTL_TYPE_MASK);
3726 regval |= tsync_rx_ctl;
3727 ew32(TSYNCRXCTL, regval);
3728 if ((er32(TSYNCRXCTL) & (E1000_TSYNCRXCTL_ENABLED |
3729 E1000_TSYNCRXCTL_TYPE_MASK)) !=
3730 (regval & (E1000_TSYNCRXCTL_ENABLED |
3731 E1000_TSYNCRXCTL_TYPE_MASK))) {
3732 e_err("Timesync Rx Control register not set as expected\n");
3733 return -EAGAIN;
3734 }
3735
3736 /* L2: define ethertype filter for time stamped packets */
3737 if (is_l2)
3738 rxmtrl |= ETH_P_1588;
3739
3740 /* define which PTP packets get time stamped */
3741 ew32(RXMTRL, rxmtrl);
3742
3743 /* Filter by destination port */
3744 if (is_l4) {
3745 rxudp = PTP_EV_PORT;
3746 cpu_to_be16s(&rxudp);
3747 }
3748 ew32(RXUDP, rxudp);
3749
3750 e1e_flush();
3751
3752 /* Clear TSYNCRXCTL_VALID & TSYNCTXCTL_VALID bit */
3753 er32(RXSTMPH);
3754 er32(TXSTMPH);
3755
3756 return 0;
3757}
3758
3759/**
3760 * e1000_configure - configure the hardware for Rx and Tx
3761 * @adapter: private board structure
3762 **/
3763static void e1000_configure(struct e1000_adapter *adapter)
3764{
3765 struct e1000_ring *rx_ring = adapter->rx_ring;
3766
3767 e1000e_set_rx_mode(adapter->netdev);
3768
3769 e1000_restore_vlan(adapter);
3770 e1000_init_manageability_pt(adapter);
3771
3772 e1000_configure_tx(adapter);
3773
3774 if (adapter->netdev->features & NETIF_F_RXHASH)
3775 e1000e_setup_rss_hash(adapter);
3776 e1000_setup_rctl(adapter);
3777 e1000_configure_rx(adapter);
3778 adapter->alloc_rx_buf(rx_ring, e1000_desc_unused(rx_ring), GFP_KERNEL);
3779}
3780
3781/**
3782 * e1000e_power_up_phy - restore link in case the phy was powered down
3783 * @adapter: address of board private structure
3784 *
3785 * The phy may be powered down to save power and turn off link when the
3786 * driver is unloaded and wake on lan is not enabled (among others)
3787 * *** this routine MUST be followed by a call to e1000e_reset ***
3788 **/
3789void e1000e_power_up_phy(struct e1000_adapter *adapter)
3790{
3791 if (adapter->hw.phy.ops.power_up)
3792 adapter->hw.phy.ops.power_up(&adapter->hw);
3793
3794 adapter->hw.mac.ops.setup_link(&adapter->hw);
3795}
3796
3797/**
3798 * e1000_power_down_phy - Power down the PHY
3799 *
3800 * Power down the PHY so no link is implied when interface is down.
3801 * The PHY cannot be powered down if management or WoL is active.
3802 */
3803static void e1000_power_down_phy(struct e1000_adapter *adapter)
3804{
3805 if (adapter->hw.phy.ops.power_down)
3806 adapter->hw.phy.ops.power_down(&adapter->hw);
3807}
3808
3809/**
3810 * e1000_flush_tx_ring - remove all descriptors from the tx_ring
3811 *
3812 * We want to clear all pending descriptors from the TX ring.
3813 * zeroing happens when the HW reads the regs. We assign the ring itself as
3814 * the data of the next descriptor. We don't care about the data we are about
3815 * to reset the HW.
3816 */
3817static void e1000_flush_tx_ring(struct e1000_adapter *adapter)
3818{
3819 struct e1000_hw *hw = &adapter->hw;
3820 struct e1000_ring *tx_ring = adapter->tx_ring;
3821 struct e1000_tx_desc *tx_desc = NULL;
3822 u32 tdt, tctl, txd_lower = E1000_TXD_CMD_IFCS;
3823 u16 size = 512;
3824
3825 tctl = er32(TCTL);
3826 ew32(TCTL, tctl | E1000_TCTL_EN);
3827 tdt = er32(TDT(0));
3828 BUG_ON(tdt != tx_ring->next_to_use);
3829 tx_desc = E1000_TX_DESC(*tx_ring, tx_ring->next_to_use);
3830 tx_desc->buffer_addr = tx_ring->dma;
3831
3832 tx_desc->lower.data = cpu_to_le32(txd_lower | size);
3833 tx_desc->upper.data = 0;
3834 /* flush descriptors to memory before notifying the HW */
3835 wmb();
3836 tx_ring->next_to_use++;
3837 if (tx_ring->next_to_use == tx_ring->count)
3838 tx_ring->next_to_use = 0;
3839 ew32(TDT(0), tx_ring->next_to_use);
3840 mmiowb();
3841 usleep_range(200, 250);
3842}
3843
3844/**
3845 * e1000_flush_rx_ring - remove all descriptors from the rx_ring
3846 *
3847 * Mark all descriptors in the RX ring as consumed and disable the rx ring
3848 */
3849static void e1000_flush_rx_ring(struct e1000_adapter *adapter)
3850{
3851 u32 rctl, rxdctl;
3852 struct e1000_hw *hw = &adapter->hw;
3853
3854 rctl = er32(RCTL);
3855 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3856 e1e_flush();
3857 usleep_range(100, 150);
3858
3859 rxdctl = er32(RXDCTL(0));
3860 /* zero the lower 14 bits (prefetch and host thresholds) */
3861 rxdctl &= 0xffffc000;
3862
3863 /* update thresholds: prefetch threshold to 31, host threshold to 1
3864 * and make sure the granularity is "descriptors" and not "cache lines"
3865 */
3866 rxdctl |= (0x1F | BIT(8) | E1000_RXDCTL_THRESH_UNIT_DESC);
3867
3868 ew32(RXDCTL(0), rxdctl);
3869 /* momentarily enable the RX ring for the changes to take effect */
3870 ew32(RCTL, rctl | E1000_RCTL_EN);
3871 e1e_flush();
3872 usleep_range(100, 150);
3873 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3874}
3875
3876/**
3877 * e1000_flush_desc_rings - remove all descriptors from the descriptor rings
3878 *
3879 * In i219, the descriptor rings must be emptied before resetting the HW
3880 * or before changing the device state to D3 during runtime (runtime PM).
3881 *
3882 * Failure to do this will cause the HW to enter a unit hang state which can
3883 * only be released by PCI reset on the device
3884 *
3885 */
3886
3887static void e1000_flush_desc_rings(struct e1000_adapter *adapter)
3888{
3889 u16 hang_state;
3890 u32 fext_nvm11, tdlen;
3891 struct e1000_hw *hw = &adapter->hw;
3892
3893 /* First, disable MULR fix in FEXTNVM11 */
3894 fext_nvm11 = er32(FEXTNVM11);
3895 fext_nvm11 |= E1000_FEXTNVM11_DISABLE_MULR_FIX;
3896 ew32(FEXTNVM11, fext_nvm11);
3897 /* do nothing if we're not in faulty state, or if the queue is empty */
3898 tdlen = er32(TDLEN(0));
3899 pci_read_config_word(adapter->pdev, PCICFG_DESC_RING_STATUS,
3900 &hang_state);
3901 if (!(hang_state & FLUSH_DESC_REQUIRED) || !tdlen)
3902 return;
3903 e1000_flush_tx_ring(adapter);
3904 /* recheck, maybe the fault is caused by the rx ring */
3905 pci_read_config_word(adapter->pdev, PCICFG_DESC_RING_STATUS,
3906 &hang_state);
3907 if (hang_state & FLUSH_DESC_REQUIRED)
3908 e1000_flush_rx_ring(adapter);
3909}
3910
3911/**
3912 * e1000e_systim_reset - reset the timesync registers after a hardware reset
3913 * @adapter: board private structure
3914 *
3915 * When the MAC is reset, all hardware bits for timesync will be reset to the
3916 * default values. This function will restore the settings last in place.
3917 * Since the clock SYSTIME registers are reset, we will simply restore the
3918 * cyclecounter to the kernel real clock time.
3919 **/
3920static void e1000e_systim_reset(struct e1000_adapter *adapter)
3921{
3922 struct ptp_clock_info *info = &adapter->ptp_clock_info;
3923 struct e1000_hw *hw = &adapter->hw;
3924 unsigned long flags;
3925 u32 timinca;
3926 s32 ret_val;
3927
3928 if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP))
3929 return;
3930
3931 if (info->adjfreq) {
3932 /* restore the previous ptp frequency delta */
3933 ret_val = info->adjfreq(info, adapter->ptp_delta);
3934 } else {
3935 /* set the default base frequency if no adjustment possible */
3936 ret_val = e1000e_get_base_timinca(adapter, &timinca);
3937 if (!ret_val)
3938 ew32(TIMINCA, timinca);
3939 }
3940
3941 if (ret_val) {
3942 dev_warn(&adapter->pdev->dev,
3943 "Failed to restore TIMINCA clock rate delta: %d\n",
3944 ret_val);
3945 return;
3946 }
3947
3948 /* reset the systim ns time counter */
3949 spin_lock_irqsave(&adapter->systim_lock, flags);
3950 timecounter_init(&adapter->tc, &adapter->cc,
3951 ktime_to_ns(ktime_get_real()));
3952 spin_unlock_irqrestore(&adapter->systim_lock, flags);
3953
3954 /* restore the previous hwtstamp configuration settings */
3955 e1000e_config_hwtstamp(adapter, &adapter->hwtstamp_config);
3956}
3957
3958/**
3959 * e1000e_reset - bring the hardware into a known good state
3960 *
3961 * This function boots the hardware and enables some settings that
3962 * require a configuration cycle of the hardware - those cannot be
3963 * set/changed during runtime. After reset the device needs to be
3964 * properly configured for Rx, Tx etc.
3965 */
3966void e1000e_reset(struct e1000_adapter *adapter)
3967{
3968 struct e1000_mac_info *mac = &adapter->hw.mac;
3969 struct e1000_fc_info *fc = &adapter->hw.fc;
3970 struct e1000_hw *hw = &adapter->hw;
3971 u32 tx_space, min_tx_space, min_rx_space;
3972 u32 pba = adapter->pba;
3973 u16 hwm;
3974
3975 /* reset Packet Buffer Allocation to default */
3976 ew32(PBA, pba);
3977
3978 if (adapter->max_frame_size > (VLAN_ETH_FRAME_LEN + ETH_FCS_LEN)) {
3979 /* To maintain wire speed transmits, the Tx FIFO should be
3980 * large enough to accommodate two full transmit packets,
3981 * rounded up to the next 1KB and expressed in KB. Likewise,
3982 * the Rx FIFO should be large enough to accommodate at least
3983 * one full receive packet and is similarly rounded up and
3984 * expressed in KB.
3985 */
3986 pba = er32(PBA);
3987 /* upper 16 bits has Tx packet buffer allocation size in KB */
3988 tx_space = pba >> 16;
3989 /* lower 16 bits has Rx packet buffer allocation size in KB */
3990 pba &= 0xffff;
3991 /* the Tx fifo also stores 16 bytes of information about the Tx
3992 * but don't include ethernet FCS because hardware appends it
3993 */
3994 min_tx_space = (adapter->max_frame_size +
3995 sizeof(struct e1000_tx_desc) - ETH_FCS_LEN) * 2;
3996 min_tx_space = ALIGN(min_tx_space, 1024);
3997 min_tx_space >>= 10;
3998 /* software strips receive CRC, so leave room for it */
3999 min_rx_space = adapter->max_frame_size;
4000 min_rx_space = ALIGN(min_rx_space, 1024);
4001 min_rx_space >>= 10;
4002
4003 /* If current Tx allocation is less than the min Tx FIFO size,
4004 * and the min Tx FIFO size is less than the current Rx FIFO
4005 * allocation, take space away from current Rx allocation
4006 */
4007 if ((tx_space < min_tx_space) &&
4008 ((min_tx_space - tx_space) < pba)) {
4009 pba -= min_tx_space - tx_space;
4010
4011 /* if short on Rx space, Rx wins and must trump Tx
4012 * adjustment
4013 */
4014 if (pba < min_rx_space)
4015 pba = min_rx_space;
4016 }
4017
4018 ew32(PBA, pba);
4019 }
4020
4021 /* flow control settings
4022 *
4023 * The high water mark must be low enough to fit one full frame
4024 * (or the size used for early receive) above it in the Rx FIFO.
4025 * Set it to the lower of:
4026 * - 90% of the Rx FIFO size, and
4027 * - the full Rx FIFO size minus one full frame
4028 */
4029 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
4030 fc->pause_time = 0xFFFF;
4031 else
4032 fc->pause_time = E1000_FC_PAUSE_TIME;
4033 fc->send_xon = true;
4034 fc->current_mode = fc->requested_mode;
4035
4036 switch (hw->mac.type) {
4037 case e1000_ich9lan:
4038 case e1000_ich10lan:
4039 if (adapter->netdev->mtu > ETH_DATA_LEN) {
4040 pba = 14;
4041 ew32(PBA, pba);
4042 fc->high_water = 0x2800;
4043 fc->low_water = fc->high_water - 8;
4044 break;
4045 }
4046 /* fall-through */
4047 default:
4048 hwm = min(((pba << 10) * 9 / 10),
4049 ((pba << 10) - adapter->max_frame_size));
4050
4051 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
4052 fc->low_water = fc->high_water - 8;
4053 break;
4054 case e1000_pchlan:
4055 /* Workaround PCH LOM adapter hangs with certain network
4056 * loads. If hangs persist, try disabling Tx flow control.
4057 */
4058 if (adapter->netdev->mtu > ETH_DATA_LEN) {
4059 fc->high_water = 0x3500;
4060 fc->low_water = 0x1500;
4061 } else {
4062 fc->high_water = 0x5000;
4063 fc->low_water = 0x3000;
4064 }
4065 fc->refresh_time = 0x1000;
4066 break;
4067 case e1000_pch2lan:
4068 case e1000_pch_lpt:
4069 case e1000_pch_spt:
4070 case e1000_pch_cnp:
4071 fc->refresh_time = 0x0400;
4072
4073 if (adapter->netdev->mtu <= ETH_DATA_LEN) {
4074 fc->high_water = 0x05C20;
4075 fc->low_water = 0x05048;
4076 fc->pause_time = 0x0650;
4077 break;
4078 }
4079
4080 pba = 14;
4081 ew32(PBA, pba);
4082 fc->high_water = ((pba << 10) * 9 / 10) & E1000_FCRTH_RTH;
4083 fc->low_water = ((pba << 10) * 8 / 10) & E1000_FCRTL_RTL;
4084 break;
4085 }
4086
4087 /* Alignment of Tx data is on an arbitrary byte boundary with the
4088 * maximum size per Tx descriptor limited only to the transmit
4089 * allocation of the packet buffer minus 96 bytes with an upper
4090 * limit of 24KB due to receive synchronization limitations.
4091 */
4092 adapter->tx_fifo_limit = min_t(u32, ((er32(PBA) >> 16) << 10) - 96,
4093 24 << 10);
4094
4095 /* Disable Adaptive Interrupt Moderation if 2 full packets cannot
4096 * fit in receive buffer.
4097 */
4098 if (adapter->itr_setting & 0x3) {
4099 if ((adapter->max_frame_size * 2) > (pba << 10)) {
4100 if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
4101 dev_info(&adapter->pdev->dev,
4102 "Interrupt Throttle Rate off\n");
4103 adapter->flags2 |= FLAG2_DISABLE_AIM;
4104 e1000e_write_itr(adapter, 0);
4105 }
4106 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
4107 dev_info(&adapter->pdev->dev,
4108 "Interrupt Throttle Rate on\n");
4109 adapter->flags2 &= ~FLAG2_DISABLE_AIM;
4110 adapter->itr = 20000;
4111 e1000e_write_itr(adapter, adapter->itr);
4112 }
4113 }
4114
4115 if (hw->mac.type >= e1000_pch_spt)
4116 e1000_flush_desc_rings(adapter);
4117 /* Allow time for pending master requests to run */
4118 mac->ops.reset_hw(hw);
4119
4120 /* For parts with AMT enabled, let the firmware know
4121 * that the network interface is in control
4122 */
4123 if (adapter->flags & FLAG_HAS_AMT)
4124 e1000e_get_hw_control(adapter);
4125
4126 ew32(WUC, 0);
4127
4128 if (mac->ops.init_hw(hw))
4129 e_err("Hardware Error\n");
4130
4131 e1000_update_mng_vlan(adapter);
4132
4133 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
4134 ew32(VET, ETH_P_8021Q);
4135
4136 e1000e_reset_adaptive(hw);
4137
4138 /* restore systim and hwtstamp settings */
4139 e1000e_systim_reset(adapter);
4140
4141 /* Set EEE advertisement as appropriate */
4142 if (adapter->flags2 & FLAG2_HAS_EEE) {
4143 s32 ret_val;
4144 u16 adv_addr;
4145
4146 switch (hw->phy.type) {
4147 case e1000_phy_82579:
4148 adv_addr = I82579_EEE_ADVERTISEMENT;
4149 break;
4150 case e1000_phy_i217:
4151 adv_addr = I217_EEE_ADVERTISEMENT;
4152 break;
4153 default:
4154 dev_err(&adapter->pdev->dev,
4155 "Invalid PHY type setting EEE advertisement\n");
4156 return;
4157 }
4158
4159 ret_val = hw->phy.ops.acquire(hw);
4160 if (ret_val) {
4161 dev_err(&adapter->pdev->dev,
4162 "EEE advertisement - unable to acquire PHY\n");
4163 return;
4164 }
4165
4166 e1000_write_emi_reg_locked(hw, adv_addr,
4167 hw->dev_spec.ich8lan.eee_disable ?
4168 0 : adapter->eee_advert);
4169
4170 hw->phy.ops.release(hw);
4171 }
4172
4173 if (!netif_running(adapter->netdev) &&
4174 !test_bit(__E1000_TESTING, &adapter->state))
4175 e1000_power_down_phy(adapter);
4176
4177 e1000_get_phy_info(hw);
4178
4179 if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
4180 !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
4181 u16 phy_data = 0;
4182 /* speed up time to link by disabling smart power down, ignore
4183 * the return value of this function because there is nothing
4184 * different we would do if it failed
4185 */
4186 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
4187 phy_data &= ~IGP02E1000_PM_SPD;
4188 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
4189 }
4190 if (hw->mac.type >= e1000_pch_spt && adapter->int_mode == 0) {
4191 u32 reg;
4192
4193 /* Fextnvm7 @ 0xe4[2] = 1 */
4194 reg = er32(FEXTNVM7);
4195 reg |= E1000_FEXTNVM7_SIDE_CLK_UNGATE;
4196 ew32(FEXTNVM7, reg);
4197 /* Fextnvm9 @ 0x5bb4[13:12] = 11 */
4198 reg = er32(FEXTNVM9);
4199 reg |= E1000_FEXTNVM9_IOSFSB_CLKGATE_DIS |
4200 E1000_FEXTNVM9_IOSFSB_CLKREQ_DIS;
4201 ew32(FEXTNVM9, reg);
4202 }
4203
4204}
4205
4206/**
4207 * e1000e_trigger_lsc - trigger an LSC interrupt
4208 * @adapter:
4209 *
4210 * Fire a link status change interrupt to start the watchdog.
4211 **/
4212static void e1000e_trigger_lsc(struct e1000_adapter *adapter)
4213{
4214 struct e1000_hw *hw = &adapter->hw;
4215
4216 if (adapter->msix_entries)
4217 ew32(ICS, E1000_ICS_LSC | E1000_ICS_OTHER);
4218 else
4219 ew32(ICS, E1000_ICS_LSC);
4220}
4221
4222void e1000e_up(struct e1000_adapter *adapter)
4223{
4224 /* hardware has been reset, we need to reload some things */
4225 e1000_configure(adapter);
4226
4227 clear_bit(__E1000_DOWN, &adapter->state);
4228
4229 if (adapter->msix_entries)
4230 e1000_configure_msix(adapter);
4231 e1000_irq_enable(adapter);
4232
4233 netif_start_queue(adapter->netdev);
4234
4235 e1000e_trigger_lsc(adapter);
4236}
4237
4238static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
4239{
4240 struct e1000_hw *hw = &adapter->hw;
4241
4242 if (!(adapter->flags2 & FLAG2_DMA_BURST))
4243 return;
4244
4245 /* flush pending descriptor writebacks to memory */
4246 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4247 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4248
4249 /* execute the writes immediately */
4250 e1e_flush();
4251
4252 /* due to rare timing issues, write to TIDV/RDTR again to ensure the
4253 * write is successful
4254 */
4255 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4256 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4257
4258 /* execute the writes immediately */
4259 e1e_flush();
4260}
4261
4262static void e1000e_update_stats(struct e1000_adapter *adapter);
4263
4264/**
4265 * e1000e_down - quiesce the device and optionally reset the hardware
4266 * @adapter: board private structure
4267 * @reset: boolean flag to reset the hardware or not
4268 */
4269void e1000e_down(struct e1000_adapter *adapter, bool reset)
4270{
4271 struct net_device *netdev = adapter->netdev;
4272 struct e1000_hw *hw = &adapter->hw;
4273 u32 tctl, rctl;
4274
4275 /* signal that we're down so the interrupt handler does not
4276 * reschedule our watchdog timer
4277 */
4278 set_bit(__E1000_DOWN, &adapter->state);
4279
4280 netif_carrier_off(netdev);
4281
4282 /* disable receives in the hardware */
4283 rctl = er32(RCTL);
4284 if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
4285 ew32(RCTL, rctl & ~E1000_RCTL_EN);
4286 /* flush and sleep below */
4287
4288 netif_stop_queue(netdev);
4289
4290 /* disable transmits in the hardware */
4291 tctl = er32(TCTL);
4292 tctl &= ~E1000_TCTL_EN;
4293 ew32(TCTL, tctl);
4294
4295 /* flush both disables and wait for them to finish */
4296 e1e_flush();
4297 usleep_range(10000, 20000);
4298
4299 e1000_irq_disable(adapter);
4300
4301 napi_synchronize(&adapter->napi);
4302
4303 del_timer_sync(&adapter->watchdog_timer);
4304 del_timer_sync(&adapter->phy_info_timer);
4305
4306 spin_lock(&adapter->stats64_lock);
4307 e1000e_update_stats(adapter);
4308 spin_unlock(&adapter->stats64_lock);
4309
4310 e1000e_flush_descriptors(adapter);
4311
4312 adapter->link_speed = 0;
4313 adapter->link_duplex = 0;
4314
4315 /* Disable Si errata workaround on PCHx for jumbo frame flow */
4316 if ((hw->mac.type >= e1000_pch2lan) &&
4317 (adapter->netdev->mtu > ETH_DATA_LEN) &&
4318 e1000_lv_jumbo_workaround_ich8lan(hw, false))
4319 e_dbg("failed to disable jumbo frame workaround mode\n");
4320
4321 if (!pci_channel_offline(adapter->pdev)) {
4322 if (reset)
4323 e1000e_reset(adapter);
4324 else if (hw->mac.type >= e1000_pch_spt)
4325 e1000_flush_desc_rings(adapter);
4326 }
4327 e1000_clean_tx_ring(adapter->tx_ring);
4328 e1000_clean_rx_ring(adapter->rx_ring);
4329}
4330
4331void e1000e_reinit_locked(struct e1000_adapter *adapter)
4332{
4333 might_sleep();
4334 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4335 usleep_range(1000, 2000);
4336 e1000e_down(adapter, true);
4337 e1000e_up(adapter);
4338 clear_bit(__E1000_RESETTING, &adapter->state);
4339}
4340
4341/**
4342 * e1000e_sanitize_systim - sanitize raw cycle counter reads
4343 * @hw: pointer to the HW structure
4344 * @systim: time value read, sanitized and returned
4345 *
4346 * Errata for 82574/82583 possible bad bits read from SYSTIMH/L:
4347 * check to see that the time is incrementing at a reasonable
4348 * rate and is a multiple of incvalue.
4349 **/
4350static u64 e1000e_sanitize_systim(struct e1000_hw *hw, u64 systim)
4351{
4352 u64 time_delta, rem, temp;
4353 u64 systim_next;
4354 u32 incvalue;
4355 int i;
4356
4357 incvalue = er32(TIMINCA) & E1000_TIMINCA_INCVALUE_MASK;
4358 for (i = 0; i < E1000_MAX_82574_SYSTIM_REREADS; i++) {
4359 /* latch SYSTIMH on read of SYSTIML */
4360 systim_next = (u64)er32(SYSTIML);
4361 systim_next |= (u64)er32(SYSTIMH) << 32;
4362
4363 time_delta = systim_next - systim;
4364 temp = time_delta;
4365 /* VMWare users have seen incvalue of zero, don't div / 0 */
4366 rem = incvalue ? do_div(temp, incvalue) : (time_delta != 0);
4367
4368 systim = systim_next;
4369
4370 if ((time_delta < E1000_82574_SYSTIM_EPSILON) && (rem == 0))
4371 break;
4372 }
4373
4374 return systim;
4375}
4376
4377/**
4378 * e1000e_cyclecounter_read - read raw cycle counter (used by time counter)
4379 * @cc: cyclecounter structure
4380 **/
4381static u64 e1000e_cyclecounter_read(const struct cyclecounter *cc)
4382{
4383 struct e1000_adapter *adapter = container_of(cc, struct e1000_adapter,
4384 cc);
4385 struct e1000_hw *hw = &adapter->hw;
4386 u32 systimel, systimeh;
4387 u64 systim;
4388 /* SYSTIMH latching upon SYSTIML read does not work well.
4389 * This means that if SYSTIML overflows after we read it but before
4390 * we read SYSTIMH, the value of SYSTIMH has been incremented and we
4391 * will experience a huge non linear increment in the systime value
4392 * to fix that we test for overflow and if true, we re-read systime.
4393 */
4394 systimel = er32(SYSTIML);
4395 systimeh = er32(SYSTIMH);
4396 /* Is systimel is so large that overflow is possible? */
4397 if (systimel >= (u32)0xffffffff - E1000_TIMINCA_INCVALUE_MASK) {
4398 u32 systimel_2 = er32(SYSTIML);
4399 if (systimel > systimel_2) {
4400 /* There was an overflow, read again SYSTIMH, and use
4401 * systimel_2
4402 */
4403 systimeh = er32(SYSTIMH);
4404 systimel = systimel_2;
4405 }
4406 }
4407 systim = (u64)systimel;
4408 systim |= (u64)systimeh << 32;
4409
4410 if (adapter->flags2 & FLAG2_CHECK_SYSTIM_OVERFLOW)
4411 systim = e1000e_sanitize_systim(hw, systim);
4412
4413 return systim;
4414}
4415
4416/**
4417 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
4418 * @adapter: board private structure to initialize
4419 *
4420 * e1000_sw_init initializes the Adapter private data structure.
4421 * Fields are initialized based on PCI device information and
4422 * OS network device settings (MTU size).
4423 **/
4424static int e1000_sw_init(struct e1000_adapter *adapter)
4425{
4426 struct net_device *netdev = adapter->netdev;
4427
4428 adapter->rx_buffer_len = VLAN_ETH_FRAME_LEN + ETH_FCS_LEN;
4429 adapter->rx_ps_bsize0 = 128;
4430 adapter->max_frame_size = netdev->mtu + VLAN_ETH_HLEN + ETH_FCS_LEN;
4431 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
4432 adapter->tx_ring_count = E1000_DEFAULT_TXD;
4433 adapter->rx_ring_count = E1000_DEFAULT_RXD;
4434
4435 spin_lock_init(&adapter->stats64_lock);
4436
4437 e1000e_set_interrupt_capability(adapter);
4438
4439 if (e1000_alloc_queues(adapter))
4440 return -ENOMEM;
4441
4442 /* Setup hardware time stamping cyclecounter */
4443 if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
4444 adapter->cc.read = e1000e_cyclecounter_read;
4445 adapter->cc.mask = CYCLECOUNTER_MASK(64);
4446 adapter->cc.mult = 1;
4447 /* cc.shift set in e1000e_get_base_tininca() */
4448
4449 spin_lock_init(&adapter->systim_lock);
4450 INIT_WORK(&adapter->tx_hwtstamp_work, e1000e_tx_hwtstamp_work);
4451 }
4452
4453 /* Explicitly disable IRQ since the NIC can be in any state. */
4454 e1000_irq_disable(adapter);
4455
4456 set_bit(__E1000_DOWN, &adapter->state);
4457 return 0;
4458}
4459
4460/**
4461 * e1000_intr_msi_test - Interrupt Handler
4462 * @irq: interrupt number
4463 * @data: pointer to a network interface device structure
4464 **/
4465static irqreturn_t e1000_intr_msi_test(int __always_unused irq, void *data)
4466{
4467 struct net_device *netdev = data;
4468 struct e1000_adapter *adapter = netdev_priv(netdev);
4469 struct e1000_hw *hw = &adapter->hw;
4470 u32 icr = er32(ICR);
4471
4472 e_dbg("icr is %08X\n", icr);
4473 if (icr & E1000_ICR_RXSEQ) {
4474 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
4475 /* Force memory writes to complete before acknowledging the
4476 * interrupt is handled.
4477 */
4478 wmb();
4479 }
4480
4481 return IRQ_HANDLED;
4482}
4483
4484/**
4485 * e1000_test_msi_interrupt - Returns 0 for successful test
4486 * @adapter: board private struct
4487 *
4488 * code flow taken from tg3.c
4489 **/
4490static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
4491{
4492 struct net_device *netdev = adapter->netdev;
4493 struct e1000_hw *hw = &adapter->hw;
4494 int err;
4495
4496 /* poll_enable hasn't been called yet, so don't need disable */
4497 /* clear any pending events */
4498 er32(ICR);
4499
4500 /* free the real vector and request a test handler */
4501 e1000_free_irq(adapter);
4502 e1000e_reset_interrupt_capability(adapter);
4503
4504 /* Assume that the test fails, if it succeeds then the test
4505 * MSI irq handler will unset this flag
4506 */
4507 adapter->flags |= FLAG_MSI_TEST_FAILED;
4508
4509 err = pci_enable_msi(adapter->pdev);
4510 if (err)
4511 goto msi_test_failed;
4512
4513 err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
4514 netdev->name, netdev);
4515 if (err) {
4516 pci_disable_msi(adapter->pdev);
4517 goto msi_test_failed;
4518 }
4519
4520 /* Force memory writes to complete before enabling and firing an
4521 * interrupt.
4522 */
4523 wmb();
4524
4525 e1000_irq_enable(adapter);
4526
4527 /* fire an unusual interrupt on the test handler */
4528 ew32(ICS, E1000_ICS_RXSEQ);
4529 e1e_flush();
4530 msleep(100);
4531
4532 e1000_irq_disable(adapter);
4533
4534 rmb(); /* read flags after interrupt has been fired */
4535
4536 if (adapter->flags & FLAG_MSI_TEST_FAILED) {
4537 adapter->int_mode = E1000E_INT_MODE_LEGACY;
4538 e_info("MSI interrupt test failed, using legacy interrupt.\n");
4539 } else {
4540 e_dbg("MSI interrupt test succeeded!\n");
4541 }
4542
4543 free_irq(adapter->pdev->irq, netdev);
4544 pci_disable_msi(adapter->pdev);
4545
4546msi_test_failed:
4547 e1000e_set_interrupt_capability(adapter);
4548 return e1000_request_irq(adapter);
4549}
4550
4551/**
4552 * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
4553 * @adapter: board private struct
4554 *
4555 * code flow taken from tg3.c, called with e1000 interrupts disabled.
4556 **/
4557static int e1000_test_msi(struct e1000_adapter *adapter)
4558{
4559 int err;
4560 u16 pci_cmd;
4561
4562 if (!(adapter->flags & FLAG_MSI_ENABLED))
4563 return 0;
4564
4565 /* disable SERR in case the MSI write causes a master abort */
4566 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4567 if (pci_cmd & PCI_COMMAND_SERR)
4568 pci_write_config_word(adapter->pdev, PCI_COMMAND,
4569 pci_cmd & ~PCI_COMMAND_SERR);
4570
4571 err = e1000_test_msi_interrupt(adapter);
4572
4573 /* re-enable SERR */
4574 if (pci_cmd & PCI_COMMAND_SERR) {
4575 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4576 pci_cmd |= PCI_COMMAND_SERR;
4577 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
4578 }
4579
4580 return err;
4581}
4582
4583/**
4584 * e1000e_open - Called when a network interface is made active
4585 * @netdev: network interface device structure
4586 *
4587 * Returns 0 on success, negative value on failure
4588 *
4589 * The open entry point is called when a network interface is made
4590 * active by the system (IFF_UP). At this point all resources needed
4591 * for transmit and receive operations are allocated, the interrupt
4592 * handler is registered with the OS, the watchdog timer is started,
4593 * and the stack is notified that the interface is ready.
4594 **/
4595int e1000e_open(struct net_device *netdev)
4596{
4597 struct e1000_adapter *adapter = netdev_priv(netdev);
4598 struct e1000_hw *hw = &adapter->hw;
4599 struct pci_dev *pdev = adapter->pdev;
4600 int err;
4601
4602 /* disallow open during test */
4603 if (test_bit(__E1000_TESTING, &adapter->state))
4604 return -EBUSY;
4605
4606 pm_runtime_get_sync(&pdev->dev);
4607
4608 netif_carrier_off(netdev);
4609
4610 /* allocate transmit descriptors */
4611 err = e1000e_setup_tx_resources(adapter->tx_ring);
4612 if (err)
4613 goto err_setup_tx;
4614
4615 /* allocate receive descriptors */
4616 err = e1000e_setup_rx_resources(adapter->rx_ring);
4617 if (err)
4618 goto err_setup_rx;
4619
4620 /* If AMT is enabled, let the firmware know that the network
4621 * interface is now open and reset the part to a known state.
4622 */
4623 if (adapter->flags & FLAG_HAS_AMT) {
4624 e1000e_get_hw_control(adapter);
4625 e1000e_reset(adapter);
4626 }
4627
4628 e1000e_power_up_phy(adapter);
4629
4630 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4631 if ((adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
4632 e1000_update_mng_vlan(adapter);
4633
4634 /* DMA latency requirement to workaround jumbo issue */
4635 pm_qos_add_request(&adapter->pm_qos_req, PM_QOS_CPU_DMA_LATENCY,
4636 PM_QOS_DEFAULT_VALUE);
4637
4638 /* before we allocate an interrupt, we must be ready to handle it.
4639 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
4640 * as soon as we call pci_request_irq, so we have to setup our
4641 * clean_rx handler before we do so.
4642 */
4643 e1000_configure(adapter);
4644
4645 err = e1000_request_irq(adapter);
4646 if (err)
4647 goto err_req_irq;
4648
4649 /* Work around PCIe errata with MSI interrupts causing some chipsets to
4650 * ignore e1000e MSI messages, which means we need to test our MSI
4651 * interrupt now
4652 */
4653 if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
4654 err = e1000_test_msi(adapter);
4655 if (err) {
4656 e_err("Interrupt allocation failed\n");
4657 goto err_req_irq;
4658 }
4659 }
4660
4661 /* From here on the code is the same as e1000e_up() */
4662 clear_bit(__E1000_DOWN, &adapter->state);
4663
4664 napi_enable(&adapter->napi);
4665
4666 e1000_irq_enable(adapter);
4667
4668 adapter->tx_hang_recheck = false;
4669 netif_start_queue(netdev);
4670
4671 hw->mac.get_link_status = true;
4672 pm_runtime_put(&pdev->dev);
4673
4674 e1000e_trigger_lsc(adapter);
4675
4676 return 0;
4677
4678err_req_irq:
4679 pm_qos_remove_request(&adapter->pm_qos_req);
4680 e1000e_release_hw_control(adapter);
4681 e1000_power_down_phy(adapter);
4682 e1000e_free_rx_resources(adapter->rx_ring);
4683err_setup_rx:
4684 e1000e_free_tx_resources(adapter->tx_ring);
4685err_setup_tx:
4686 e1000e_reset(adapter);
4687 pm_runtime_put_sync(&pdev->dev);
4688
4689 return err;
4690}
4691
4692/**
4693 * e1000e_close - Disables a network interface
4694 * @netdev: network interface device structure
4695 *
4696 * Returns 0, this is not allowed to fail
4697 *
4698 * The close entry point is called when an interface is de-activated
4699 * by the OS. The hardware is still under the drivers control, but
4700 * needs to be disabled. A global MAC reset is issued to stop the
4701 * hardware, and all transmit and receive resources are freed.
4702 **/
4703int e1000e_close(struct net_device *netdev)
4704{
4705 struct e1000_adapter *adapter = netdev_priv(netdev);
4706 struct pci_dev *pdev = adapter->pdev;
4707 int count = E1000_CHECK_RESET_COUNT;
4708
4709 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
4710 usleep_range(10000, 20000);
4711
4712 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4713
4714 pm_runtime_get_sync(&pdev->dev);
4715
4716 if (!test_bit(__E1000_DOWN, &adapter->state)) {
4717 e1000e_down(adapter, true);
4718 e1000_free_irq(adapter);
4719
4720 /* Link status message must follow this format */
4721 pr_info("%s NIC Link is Down\n", adapter->netdev->name);
4722 }
4723
4724 napi_disable(&adapter->napi);
4725
4726 e1000e_free_tx_resources(adapter->tx_ring);
4727 e1000e_free_rx_resources(adapter->rx_ring);
4728
4729 /* kill manageability vlan ID if supported, but not if a vlan with
4730 * the same ID is registered on the host OS (let 8021q kill it)
4731 */
4732 if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
4733 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
4734 adapter->mng_vlan_id);
4735
4736 /* If AMT is enabled, let the firmware know that the network
4737 * interface is now closed
4738 */
4739 if ((adapter->flags & FLAG_HAS_AMT) &&
4740 !test_bit(__E1000_TESTING, &adapter->state))
4741 e1000e_release_hw_control(adapter);
4742
4743 pm_qos_remove_request(&adapter->pm_qos_req);
4744
4745 pm_runtime_put_sync(&pdev->dev);
4746
4747 return 0;
4748}
4749
4750/**
4751 * e1000_set_mac - Change the Ethernet Address of the NIC
4752 * @netdev: network interface device structure
4753 * @p: pointer to an address structure
4754 *
4755 * Returns 0 on success, negative on failure
4756 **/
4757static int e1000_set_mac(struct net_device *netdev, void *p)
4758{
4759 struct e1000_adapter *adapter = netdev_priv(netdev);
4760 struct e1000_hw *hw = &adapter->hw;
4761 struct sockaddr *addr = p;
4762
4763 if (!is_valid_ether_addr(addr->sa_data))
4764 return -EADDRNOTAVAIL;
4765
4766 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4767 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
4768
4769 hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
4770
4771 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
4772 /* activate the work around */
4773 e1000e_set_laa_state_82571(&adapter->hw, 1);
4774
4775 /* Hold a copy of the LAA in RAR[14] This is done so that
4776 * between the time RAR[0] gets clobbered and the time it
4777 * gets fixed (in e1000_watchdog), the actual LAA is in one
4778 * of the RARs and no incoming packets directed to this port
4779 * are dropped. Eventually the LAA will be in RAR[0] and
4780 * RAR[14]
4781 */
4782 hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr,
4783 adapter->hw.mac.rar_entry_count - 1);
4784 }
4785
4786 return 0;
4787}
4788
4789/**
4790 * e1000e_update_phy_task - work thread to update phy
4791 * @work: pointer to our work struct
4792 *
4793 * this worker thread exists because we must acquire a
4794 * semaphore to read the phy, which we could msleep while
4795 * waiting for it, and we can't msleep in a timer.
4796 **/
4797static void e1000e_update_phy_task(struct work_struct *work)
4798{
4799 struct e1000_adapter *adapter = container_of(work,
4800 struct e1000_adapter,
4801 update_phy_task);
4802 struct e1000_hw *hw = &adapter->hw;
4803
4804 if (test_bit(__E1000_DOWN, &adapter->state))
4805 return;
4806
4807 e1000_get_phy_info(hw);
4808
4809 /* Enable EEE on 82579 after link up */
4810 if (hw->phy.type >= e1000_phy_82579)
4811 e1000_set_eee_pchlan(hw);
4812}
4813
4814/**
4815 * e1000_update_phy_info - timre call-back to update PHY info
4816 * @data: pointer to adapter cast into an unsigned long
4817 *
4818 * Need to wait a few seconds after link up to get diagnostic information from
4819 * the phy
4820 **/
4821static void e1000_update_phy_info(struct timer_list *t)
4822{
4823 struct e1000_adapter *adapter = from_timer(adapter, t, phy_info_timer);
4824
4825 if (test_bit(__E1000_DOWN, &adapter->state))
4826 return;
4827
4828 schedule_work(&adapter->update_phy_task);
4829}
4830
4831/**
4832 * e1000e_update_phy_stats - Update the PHY statistics counters
4833 * @adapter: board private structure
4834 *
4835 * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4836 **/
4837static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4838{
4839 struct e1000_hw *hw = &adapter->hw;
4840 s32 ret_val;
4841 u16 phy_data;
4842
4843 ret_val = hw->phy.ops.acquire(hw);
4844 if (ret_val)
4845 return;
4846
4847 /* A page set is expensive so check if already on desired page.
4848 * If not, set to the page with the PHY status registers.
4849 */
4850 hw->phy.addr = 1;
4851 ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4852 &phy_data);
4853 if (ret_val)
4854 goto release;
4855 if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4856 ret_val = hw->phy.ops.set_page(hw,
4857 HV_STATS_PAGE << IGP_PAGE_SHIFT);
4858 if (ret_val)
4859 goto release;
4860 }
4861
4862 /* Single Collision Count */
4863 hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4864 ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4865 if (!ret_val)
4866 adapter->stats.scc += phy_data;
4867
4868 /* Excessive Collision Count */
4869 hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4870 ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4871 if (!ret_val)
4872 adapter->stats.ecol += phy_data;
4873
4874 /* Multiple Collision Count */
4875 hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4876 ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4877 if (!ret_val)
4878 adapter->stats.mcc += phy_data;
4879
4880 /* Late Collision Count */
4881 hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4882 ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4883 if (!ret_val)
4884 adapter->stats.latecol += phy_data;
4885
4886 /* Collision Count - also used for adaptive IFS */
4887 hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4888 ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4889 if (!ret_val)
4890 hw->mac.collision_delta = phy_data;
4891
4892 /* Defer Count */
4893 hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4894 ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4895 if (!ret_val)
4896 adapter->stats.dc += phy_data;
4897
4898 /* Transmit with no CRS */
4899 hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4900 ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4901 if (!ret_val)
4902 adapter->stats.tncrs += phy_data;
4903
4904release:
4905 hw->phy.ops.release(hw);
4906}
4907
4908/**
4909 * e1000e_update_stats - Update the board statistics counters
4910 * @adapter: board private structure
4911 **/
4912static void e1000e_update_stats(struct e1000_adapter *adapter)
4913{
4914 struct net_device *netdev = adapter->netdev;
4915 struct e1000_hw *hw = &adapter->hw;
4916 struct pci_dev *pdev = adapter->pdev;
4917
4918 /* Prevent stats update while adapter is being reset, or if the pci
4919 * connection is down.
4920 */
4921 if (adapter->link_speed == 0)
4922 return;
4923 if (pci_channel_offline(pdev))
4924 return;
4925
4926 adapter->stats.crcerrs += er32(CRCERRS);
4927 adapter->stats.gprc += er32(GPRC);
4928 adapter->stats.gorc += er32(GORCL);
4929 er32(GORCH); /* Clear gorc */
4930 adapter->stats.bprc += er32(BPRC);
4931 adapter->stats.mprc += er32(MPRC);
4932 adapter->stats.roc += er32(ROC);
4933
4934 adapter->stats.mpc += er32(MPC);
4935
4936 /* Half-duplex statistics */
4937 if (adapter->link_duplex == HALF_DUPLEX) {
4938 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4939 e1000e_update_phy_stats(adapter);
4940 } else {
4941 adapter->stats.scc += er32(SCC);
4942 adapter->stats.ecol += er32(ECOL);
4943 adapter->stats.mcc += er32(MCC);
4944 adapter->stats.latecol += er32(LATECOL);
4945 adapter->stats.dc += er32(DC);
4946
4947 hw->mac.collision_delta = er32(COLC);
4948
4949 if ((hw->mac.type != e1000_82574) &&
4950 (hw->mac.type != e1000_82583))
4951 adapter->stats.tncrs += er32(TNCRS);
4952 }
4953 adapter->stats.colc += hw->mac.collision_delta;
4954 }
4955
4956 adapter->stats.xonrxc += er32(XONRXC);
4957 adapter->stats.xontxc += er32(XONTXC);
4958 adapter->stats.xoffrxc += er32(XOFFRXC);
4959 adapter->stats.xofftxc += er32(XOFFTXC);
4960 adapter->stats.gptc += er32(GPTC);
4961 adapter->stats.gotc += er32(GOTCL);
4962 er32(GOTCH); /* Clear gotc */
4963 adapter->stats.rnbc += er32(RNBC);
4964 adapter->stats.ruc += er32(RUC);
4965
4966 adapter->stats.mptc += er32(MPTC);
4967 adapter->stats.bptc += er32(BPTC);
4968
4969 /* used for adaptive IFS */
4970
4971 hw->mac.tx_packet_delta = er32(TPT);
4972 adapter->stats.tpt += hw->mac.tx_packet_delta;
4973
4974 adapter->stats.algnerrc += er32(ALGNERRC);
4975 adapter->stats.rxerrc += er32(RXERRC);
4976 adapter->stats.cexterr += er32(CEXTERR);
4977 adapter->stats.tsctc += er32(TSCTC);
4978 adapter->stats.tsctfc += er32(TSCTFC);
4979
4980 /* Fill out the OS statistics structure */
4981 netdev->stats.multicast = adapter->stats.mprc;
4982 netdev->stats.collisions = adapter->stats.colc;
4983
4984 /* Rx Errors */
4985
4986 /* RLEC on some newer hardware can be incorrect so build
4987 * our own version based on RUC and ROC
4988 */
4989 netdev->stats.rx_errors = adapter->stats.rxerrc +
4990 adapter->stats.crcerrs + adapter->stats.algnerrc +
4991 adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
4992 netdev->stats.rx_length_errors = adapter->stats.ruc +
4993 adapter->stats.roc;
4994 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4995 netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4996 netdev->stats.rx_missed_errors = adapter->stats.mpc;
4997
4998 /* Tx Errors */
4999 netdev->stats.tx_errors = adapter->stats.ecol + adapter->stats.latecol;
5000 netdev->stats.tx_aborted_errors = adapter->stats.ecol;
5001 netdev->stats.tx_window_errors = adapter->stats.latecol;
5002 netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
5003
5004 /* Tx Dropped needs to be maintained elsewhere */
5005
5006 /* Management Stats */
5007 adapter->stats.mgptc += er32(MGTPTC);
5008 adapter->stats.mgprc += er32(MGTPRC);
5009 adapter->stats.mgpdc += er32(MGTPDC);
5010
5011 /* Correctable ECC Errors */
5012 if (hw->mac.type >= e1000_pch_lpt) {
5013 u32 pbeccsts = er32(PBECCSTS);
5014
5015 adapter->corr_errors +=
5016 pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
5017 adapter->uncorr_errors +=
5018 (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
5019 E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
5020 }
5021}
5022
5023/**
5024 * e1000_phy_read_status - Update the PHY register status snapshot
5025 * @adapter: board private structure
5026 **/
5027static void e1000_phy_read_status(struct e1000_adapter *adapter)
5028{
5029 struct e1000_hw *hw = &adapter->hw;
5030 struct e1000_phy_regs *phy = &adapter->phy_regs;
5031
5032 if (!pm_runtime_suspended((&adapter->pdev->dev)->parent) &&
5033 (er32(STATUS) & E1000_STATUS_LU) &&
5034 (adapter->hw.phy.media_type == e1000_media_type_copper)) {
5035 int ret_val;
5036
5037 ret_val = e1e_rphy(hw, MII_BMCR, &phy->bmcr);
5038 ret_val |= e1e_rphy(hw, MII_BMSR, &phy->bmsr);
5039 ret_val |= e1e_rphy(hw, MII_ADVERTISE, &phy->advertise);
5040 ret_val |= e1e_rphy(hw, MII_LPA, &phy->lpa);
5041 ret_val |= e1e_rphy(hw, MII_EXPANSION, &phy->expansion);
5042 ret_val |= e1e_rphy(hw, MII_CTRL1000, &phy->ctrl1000);
5043 ret_val |= e1e_rphy(hw, MII_STAT1000, &phy->stat1000);
5044 ret_val |= e1e_rphy(hw, MII_ESTATUS, &phy->estatus);
5045 if (ret_val)
5046 e_warn("Error reading PHY register\n");
5047 } else {
5048 /* Do not read PHY registers if link is not up
5049 * Set values to typical power-on defaults
5050 */
5051 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
5052 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
5053 BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
5054 BMSR_ERCAP);
5055 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
5056 ADVERTISE_ALL | ADVERTISE_CSMA);
5057 phy->lpa = 0;
5058 phy->expansion = EXPANSION_ENABLENPAGE;
5059 phy->ctrl1000 = ADVERTISE_1000FULL;
5060 phy->stat1000 = 0;
5061 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
5062 }
5063}
5064
5065static void e1000_print_link_info(struct e1000_adapter *adapter)
5066{
5067 struct e1000_hw *hw = &adapter->hw;
5068 u32 ctrl = er32(CTRL);
5069
5070 /* Link status message must follow this format for user tools */
5071 pr_info("%s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
5072 adapter->netdev->name, adapter->link_speed,
5073 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
5074 (ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
5075 (ctrl & E1000_CTRL_RFCE) ? "Rx" :
5076 (ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
5077}
5078
5079static bool e1000e_has_link(struct e1000_adapter *adapter)
5080{
5081 struct e1000_hw *hw = &adapter->hw;
5082 bool link_active = false;
5083 s32 ret_val = 0;
5084
5085 /* get_link_status is set on LSC (link status) interrupt or
5086 * Rx sequence error interrupt. get_link_status will stay
5087 * true until the check_for_link establishes link
5088 * for copper adapters ONLY
5089 */
5090 switch (hw->phy.media_type) {
5091 case e1000_media_type_copper:
5092 if (hw->mac.get_link_status) {
5093 ret_val = hw->mac.ops.check_for_link(hw);
5094 link_active = !hw->mac.get_link_status;
5095 } else {
5096 link_active = true;
5097 }
5098 break;
5099 case e1000_media_type_fiber:
5100 ret_val = hw->mac.ops.check_for_link(hw);
5101 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
5102 break;
5103 case e1000_media_type_internal_serdes:
5104 ret_val = hw->mac.ops.check_for_link(hw);
5105 link_active = hw->mac.serdes_has_link;
5106 break;
5107 default:
5108 case e1000_media_type_unknown:
5109 break;
5110 }
5111
5112 if ((ret_val == -E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
5113 (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
5114 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
5115 e_info("Gigabit has been disabled, downgrading speed\n");
5116 }
5117
5118 return link_active;
5119}
5120
5121static void e1000e_enable_receives(struct e1000_adapter *adapter)
5122{
5123 /* make sure the receive unit is started */
5124 if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5125 (adapter->flags & FLAG_RESTART_NOW)) {
5126 struct e1000_hw *hw = &adapter->hw;
5127 u32 rctl = er32(RCTL);
5128
5129 ew32(RCTL, rctl | E1000_RCTL_EN);
5130 adapter->flags &= ~FLAG_RESTART_NOW;
5131 }
5132}
5133
5134static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
5135{
5136 struct e1000_hw *hw = &adapter->hw;
5137
5138 /* With 82574 controllers, PHY needs to be checked periodically
5139 * for hung state and reset, if two calls return true
5140 */
5141 if (e1000_check_phy_82574(hw))
5142 adapter->phy_hang_count++;
5143 else
5144 adapter->phy_hang_count = 0;
5145
5146 if (adapter->phy_hang_count > 1) {
5147 adapter->phy_hang_count = 0;
5148 e_dbg("PHY appears hung - resetting\n");
5149 schedule_work(&adapter->reset_task);
5150 }
5151}
5152
5153/**
5154 * e1000_watchdog - Timer Call-back
5155 * @data: pointer to adapter cast into an unsigned long
5156 **/
5157static void e1000_watchdog(struct timer_list *t)
5158{
5159 struct e1000_adapter *adapter = from_timer(adapter, t, watchdog_timer);
5160
5161 /* Do the rest outside of interrupt context */
5162 schedule_work(&adapter->watchdog_task);
5163
5164 /* TODO: make this use queue_delayed_work() */
5165}
5166
5167static void e1000_watchdog_task(struct work_struct *work)
5168{
5169 struct e1000_adapter *adapter = container_of(work,
5170 struct e1000_adapter,
5171 watchdog_task);
5172 struct net_device *netdev = adapter->netdev;
5173 struct e1000_mac_info *mac = &adapter->hw.mac;
5174 struct e1000_phy_info *phy = &adapter->hw.phy;
5175 struct e1000_ring *tx_ring = adapter->tx_ring;
5176 struct e1000_hw *hw = &adapter->hw;
5177 u32 link, tctl;
5178
5179 if (test_bit(__E1000_DOWN, &adapter->state))
5180 return;
5181
5182 link = e1000e_has_link(adapter);
5183 if ((netif_carrier_ok(netdev)) && link) {
5184 /* Cancel scheduled suspend requests. */
5185 pm_runtime_resume(netdev->dev.parent);
5186
5187 e1000e_enable_receives(adapter);
5188 goto link_up;
5189 }
5190
5191 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
5192 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
5193 e1000_update_mng_vlan(adapter);
5194
5195 if (link) {
5196 if (!netif_carrier_ok(netdev)) {
5197 bool txb2b = true;
5198
5199 /* Cancel scheduled suspend requests. */
5200 pm_runtime_resume(netdev->dev.parent);
5201
5202 /* update snapshot of PHY registers on LSC */
5203 e1000_phy_read_status(adapter);
5204 mac->ops.get_link_up_info(&adapter->hw,
5205 &adapter->link_speed,
5206 &adapter->link_duplex);
5207 e1000_print_link_info(adapter);
5208
5209 /* check if SmartSpeed worked */
5210 e1000e_check_downshift(hw);
5211 if (phy->speed_downgraded)
5212 netdev_warn(netdev,
5213 "Link Speed was downgraded by SmartSpeed\n");
5214
5215 /* On supported PHYs, check for duplex mismatch only
5216 * if link has autonegotiated at 10/100 half
5217 */
5218 if ((hw->phy.type == e1000_phy_igp_3 ||
5219 hw->phy.type == e1000_phy_bm) &&
5220 hw->mac.autoneg &&
5221 (adapter->link_speed == SPEED_10 ||
5222 adapter->link_speed == SPEED_100) &&
5223 (adapter->link_duplex == HALF_DUPLEX)) {
5224 u16 autoneg_exp;
5225
5226 e1e_rphy(hw, MII_EXPANSION, &autoneg_exp);
5227
5228 if (!(autoneg_exp & EXPANSION_NWAY))
5229 e_info("Autonegotiated half duplex but link partner cannot autoneg. Try forcing full duplex if link gets many collisions.\n");
5230 }
5231
5232 /* adjust timeout factor according to speed/duplex */
5233 adapter->tx_timeout_factor = 1;
5234 switch (adapter->link_speed) {
5235 case SPEED_10:
5236 txb2b = false;
5237 adapter->tx_timeout_factor = 16;
5238 break;
5239 case SPEED_100:
5240 txb2b = false;
5241 adapter->tx_timeout_factor = 10;
5242 break;
5243 }
5244
5245 /* workaround: re-program speed mode bit after
5246 * link-up event
5247 */
5248 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
5249 !txb2b) {
5250 u32 tarc0;
5251
5252 tarc0 = er32(TARC(0));
5253 tarc0 &= ~SPEED_MODE_BIT;
5254 ew32(TARC(0), tarc0);
5255 }
5256
5257 /* disable TSO for pcie and 10/100 speeds, to avoid
5258 * some hardware issues
5259 */
5260 if (!(adapter->flags & FLAG_TSO_FORCE)) {
5261 switch (adapter->link_speed) {
5262 case SPEED_10:
5263 case SPEED_100:
5264 e_info("10/100 speed: disabling TSO\n");
5265 netdev->features &= ~NETIF_F_TSO;
5266 netdev->features &= ~NETIF_F_TSO6;
5267 break;
5268 case SPEED_1000:
5269 netdev->features |= NETIF_F_TSO;
5270 netdev->features |= NETIF_F_TSO6;
5271 break;
5272 default:
5273 /* oops */
5274 break;
5275 }
5276 }
5277
5278 /* enable transmits in the hardware, need to do this
5279 * after setting TARC(0)
5280 */
5281 tctl = er32(TCTL);
5282 tctl |= E1000_TCTL_EN;
5283 ew32(TCTL, tctl);
5284
5285 /* Perform any post-link-up configuration before
5286 * reporting link up.
5287 */
5288 if (phy->ops.cfg_on_link_up)
5289 phy->ops.cfg_on_link_up(hw);
5290
5291 netif_carrier_on(netdev);
5292
5293 if (!test_bit(__E1000_DOWN, &adapter->state))
5294 mod_timer(&adapter->phy_info_timer,
5295 round_jiffies(jiffies + 2 * HZ));
5296 }
5297 } else {
5298 if (netif_carrier_ok(netdev)) {
5299 adapter->link_speed = 0;
5300 adapter->link_duplex = 0;
5301 /* Link status message must follow this format */
5302 pr_info("%s NIC Link is Down\n", adapter->netdev->name);
5303 netif_carrier_off(netdev);
5304 if (!test_bit(__E1000_DOWN, &adapter->state))
5305 mod_timer(&adapter->phy_info_timer,
5306 round_jiffies(jiffies + 2 * HZ));
5307
5308 /* 8000ES2LAN requires a Rx packet buffer work-around
5309 * on link down event; reset the controller to flush
5310 * the Rx packet buffer.
5311 */
5312 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
5313 adapter->flags |= FLAG_RESTART_NOW;
5314 else
5315 pm_schedule_suspend(netdev->dev.parent,
5316 LINK_TIMEOUT);
5317 }
5318 }
5319
5320link_up:
5321 spin_lock(&adapter->stats64_lock);
5322 e1000e_update_stats(adapter);
5323
5324 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
5325 adapter->tpt_old = adapter->stats.tpt;
5326 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
5327 adapter->colc_old = adapter->stats.colc;
5328
5329 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
5330 adapter->gorc_old = adapter->stats.gorc;
5331 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
5332 adapter->gotc_old = adapter->stats.gotc;
5333 spin_unlock(&adapter->stats64_lock);
5334
5335 /* If the link is lost the controller stops DMA, but
5336 * if there is queued Tx work it cannot be done. So
5337 * reset the controller to flush the Tx packet buffers.
5338 */
5339 if (!netif_carrier_ok(netdev) &&
5340 (e1000_desc_unused(tx_ring) + 1 < tx_ring->count))
5341 adapter->flags |= FLAG_RESTART_NOW;
5342
5343 /* If reset is necessary, do it outside of interrupt context. */
5344 if (adapter->flags & FLAG_RESTART_NOW) {
5345 schedule_work(&adapter->reset_task);
5346 /* return immediately since reset is imminent */
5347 return;
5348 }
5349
5350 e1000e_update_adaptive(&adapter->hw);
5351
5352 /* Simple mode for Interrupt Throttle Rate (ITR) */
5353 if (adapter->itr_setting == 4) {
5354 /* Symmetric Tx/Rx gets a reduced ITR=2000;
5355 * Total asymmetrical Tx or Rx gets ITR=8000;
5356 * everyone else is between 2000-8000.
5357 */
5358 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
5359 u32 dif = (adapter->gotc > adapter->gorc ?
5360 adapter->gotc - adapter->gorc :
5361 adapter->gorc - adapter->gotc) / 10000;
5362 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
5363
5364 e1000e_write_itr(adapter, itr);
5365 }
5366
5367 /* Cause software interrupt to ensure Rx ring is cleaned */
5368 if (adapter->msix_entries)
5369 ew32(ICS, adapter->rx_ring->ims_val);
5370 else
5371 ew32(ICS, E1000_ICS_RXDMT0);
5372
5373 /* flush pending descriptors to memory before detecting Tx hang */
5374 e1000e_flush_descriptors(adapter);
5375
5376 /* Force detection of hung controller every watchdog period */
5377 adapter->detect_tx_hung = true;
5378
5379 /* With 82571 controllers, LAA may be overwritten due to controller
5380 * reset from the other port. Set the appropriate LAA in RAR[0]
5381 */
5382 if (e1000e_get_laa_state_82571(hw))
5383 hw->mac.ops.rar_set(hw, adapter->hw.mac.addr, 0);
5384
5385 if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
5386 e1000e_check_82574_phy_workaround(adapter);
5387
5388 /* Clear valid timestamp stuck in RXSTMPL/H due to a Rx error */
5389 if (adapter->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE) {
5390 if ((adapter->flags2 & FLAG2_CHECK_RX_HWTSTAMP) &&
5391 (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID)) {
5392 er32(RXSTMPH);
5393 adapter->rx_hwtstamp_cleared++;
5394 } else {
5395 adapter->flags2 |= FLAG2_CHECK_RX_HWTSTAMP;
5396 }
5397 }
5398
5399 /* Reset the timer */
5400 if (!test_bit(__E1000_DOWN, &adapter->state))
5401 mod_timer(&adapter->watchdog_timer,
5402 round_jiffies(jiffies + 2 * HZ));
5403}
5404
5405#define E1000_TX_FLAGS_CSUM 0x00000001
5406#define E1000_TX_FLAGS_VLAN 0x00000002
5407#define E1000_TX_FLAGS_TSO 0x00000004
5408#define E1000_TX_FLAGS_IPV4 0x00000008
5409#define E1000_TX_FLAGS_NO_FCS 0x00000010
5410#define E1000_TX_FLAGS_HWTSTAMP 0x00000020
5411#define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
5412#define E1000_TX_FLAGS_VLAN_SHIFT 16
5413
5414static int e1000_tso(struct e1000_ring *tx_ring, struct sk_buff *skb,
5415 __be16 protocol)
5416{
5417 struct e1000_context_desc *context_desc;
5418 struct e1000_buffer *buffer_info;
5419 unsigned int i;
5420 u32 cmd_length = 0;
5421 u16 ipcse = 0, mss;
5422 u8 ipcss, ipcso, tucss, tucso, hdr_len;
5423 int err;
5424
5425 if (!skb_is_gso(skb))
5426 return 0;
5427
5428 err = skb_cow_head(skb, 0);
5429 if (err < 0)
5430 return err;
5431
5432 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5433 mss = skb_shinfo(skb)->gso_size;
5434 if (protocol == htons(ETH_P_IP)) {
5435 struct iphdr *iph = ip_hdr(skb);
5436 iph->tot_len = 0;
5437 iph->check = 0;
5438 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
5439 0, IPPROTO_TCP, 0);
5440 cmd_length = E1000_TXD_CMD_IP;
5441 ipcse = skb_transport_offset(skb) - 1;
5442 } else if (skb_is_gso_v6(skb)) {
5443 ipv6_hdr(skb)->payload_len = 0;
5444 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5445 &ipv6_hdr(skb)->daddr,
5446 0, IPPROTO_TCP, 0);
5447 ipcse = 0;
5448 }
5449 ipcss = skb_network_offset(skb);
5450 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
5451 tucss = skb_transport_offset(skb);
5452 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
5453
5454 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
5455 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
5456
5457 i = tx_ring->next_to_use;
5458 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5459 buffer_info = &tx_ring->buffer_info[i];
5460
5461 context_desc->lower_setup.ip_fields.ipcss = ipcss;
5462 context_desc->lower_setup.ip_fields.ipcso = ipcso;
5463 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
5464 context_desc->upper_setup.tcp_fields.tucss = tucss;
5465 context_desc->upper_setup.tcp_fields.tucso = tucso;
5466 context_desc->upper_setup.tcp_fields.tucse = 0;
5467 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
5468 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
5469 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
5470
5471 buffer_info->time_stamp = jiffies;
5472 buffer_info->next_to_watch = i;
5473
5474 i++;
5475 if (i == tx_ring->count)
5476 i = 0;
5477 tx_ring->next_to_use = i;
5478
5479 return 1;
5480}
5481
5482static bool e1000_tx_csum(struct e1000_ring *tx_ring, struct sk_buff *skb,
5483 __be16 protocol)
5484{
5485 struct e1000_adapter *adapter = tx_ring->adapter;
5486 struct e1000_context_desc *context_desc;
5487 struct e1000_buffer *buffer_info;
5488 unsigned int i;
5489 u8 css;
5490 u32 cmd_len = E1000_TXD_CMD_DEXT;
5491
5492 if (skb->ip_summed != CHECKSUM_PARTIAL)
5493 return false;
5494
5495 switch (protocol) {
5496 case cpu_to_be16(ETH_P_IP):
5497 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5498 cmd_len |= E1000_TXD_CMD_TCP;
5499 break;
5500 case cpu_to_be16(ETH_P_IPV6):
5501 /* XXX not handling all IPV6 headers */
5502 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5503 cmd_len |= E1000_TXD_CMD_TCP;
5504 break;
5505 default:
5506 if (unlikely(net_ratelimit()))
5507 e_warn("checksum_partial proto=%x!\n",
5508 be16_to_cpu(protocol));
5509 break;
5510 }
5511
5512 css = skb_checksum_start_offset(skb);
5513
5514 i = tx_ring->next_to_use;
5515 buffer_info = &tx_ring->buffer_info[i];
5516 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5517
5518 context_desc->lower_setup.ip_config = 0;
5519 context_desc->upper_setup.tcp_fields.tucss = css;
5520 context_desc->upper_setup.tcp_fields.tucso = css + skb->csum_offset;
5521 context_desc->upper_setup.tcp_fields.tucse = 0;
5522 context_desc->tcp_seg_setup.data = 0;
5523 context_desc->cmd_and_length = cpu_to_le32(cmd_len);
5524
5525 buffer_info->time_stamp = jiffies;
5526 buffer_info->next_to_watch = i;
5527
5528 i++;
5529 if (i == tx_ring->count)
5530 i = 0;
5531 tx_ring->next_to_use = i;
5532
5533 return true;
5534}
5535
5536static int e1000_tx_map(struct e1000_ring *tx_ring, struct sk_buff *skb,
5537 unsigned int first, unsigned int max_per_txd,
5538 unsigned int nr_frags)
5539{
5540 struct e1000_adapter *adapter = tx_ring->adapter;
5541 struct pci_dev *pdev = adapter->pdev;
5542 struct e1000_buffer *buffer_info;
5543 unsigned int len = skb_headlen(skb);
5544 unsigned int offset = 0, size, count = 0, i;
5545 unsigned int f, bytecount, segs;
5546
5547 i = tx_ring->next_to_use;
5548
5549 while (len) {
5550 buffer_info = &tx_ring->buffer_info[i];
5551 size = min(len, max_per_txd);
5552
5553 buffer_info->length = size;
5554 buffer_info->time_stamp = jiffies;
5555 buffer_info->next_to_watch = i;
5556 buffer_info->dma = dma_map_single(&pdev->dev,
5557 skb->data + offset,
5558 size, DMA_TO_DEVICE);
5559 buffer_info->mapped_as_page = false;
5560 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5561 goto dma_error;
5562
5563 len -= size;
5564 offset += size;
5565 count++;
5566
5567 if (len) {
5568 i++;
5569 if (i == tx_ring->count)
5570 i = 0;
5571 }
5572 }
5573
5574 for (f = 0; f < nr_frags; f++) {
5575 const struct skb_frag_struct *frag;
5576
5577 frag = &skb_shinfo(skb)->frags[f];
5578 len = skb_frag_size(frag);
5579 offset = 0;
5580
5581 while (len) {
5582 i++;
5583 if (i == tx_ring->count)
5584 i = 0;
5585
5586 buffer_info = &tx_ring->buffer_info[i];
5587 size = min(len, max_per_txd);
5588
5589 buffer_info->length = size;
5590 buffer_info->time_stamp = jiffies;
5591 buffer_info->next_to_watch = i;
5592 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
5593 offset, size,
5594 DMA_TO_DEVICE);
5595 buffer_info->mapped_as_page = true;
5596 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5597 goto dma_error;
5598
5599 len -= size;
5600 offset += size;
5601 count++;
5602 }
5603 }
5604
5605 segs = skb_shinfo(skb)->gso_segs ? : 1;
5606 /* multiply data chunks by size of headers */
5607 bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
5608
5609 tx_ring->buffer_info[i].skb = skb;
5610 tx_ring->buffer_info[i].segs = segs;
5611 tx_ring->buffer_info[i].bytecount = bytecount;
5612 tx_ring->buffer_info[first].next_to_watch = i;
5613
5614 return count;
5615
5616dma_error:
5617 dev_err(&pdev->dev, "Tx DMA map failed\n");
5618 buffer_info->dma = 0;
5619 if (count)
5620 count--;
5621
5622 while (count--) {
5623 if (i == 0)
5624 i += tx_ring->count;
5625 i--;
5626 buffer_info = &tx_ring->buffer_info[i];
5627 e1000_put_txbuf(tx_ring, buffer_info, true);
5628 }
5629
5630 return 0;
5631}
5632
5633static void e1000_tx_queue(struct e1000_ring *tx_ring, int tx_flags, int count)
5634{
5635 struct e1000_adapter *adapter = tx_ring->adapter;
5636 struct e1000_tx_desc *tx_desc = NULL;
5637 struct e1000_buffer *buffer_info;
5638 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
5639 unsigned int i;
5640
5641 if (tx_flags & E1000_TX_FLAGS_TSO) {
5642 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
5643 E1000_TXD_CMD_TSE;
5644 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5645
5646 if (tx_flags & E1000_TX_FLAGS_IPV4)
5647 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
5648 }
5649
5650 if (tx_flags & E1000_TX_FLAGS_CSUM) {
5651 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5652 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5653 }
5654
5655 if (tx_flags & E1000_TX_FLAGS_VLAN) {
5656 txd_lower |= E1000_TXD_CMD_VLE;
5657 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
5658 }
5659
5660 if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5661 txd_lower &= ~(E1000_TXD_CMD_IFCS);
5662
5663 if (unlikely(tx_flags & E1000_TX_FLAGS_HWTSTAMP)) {
5664 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5665 txd_upper |= E1000_TXD_EXTCMD_TSTAMP;
5666 }
5667
5668 i = tx_ring->next_to_use;
5669
5670 do {
5671 buffer_info = &tx_ring->buffer_info[i];
5672 tx_desc = E1000_TX_DESC(*tx_ring, i);
5673 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
5674 tx_desc->lower.data = cpu_to_le32(txd_lower |
5675 buffer_info->length);
5676 tx_desc->upper.data = cpu_to_le32(txd_upper);
5677
5678 i++;
5679 if (i == tx_ring->count)
5680 i = 0;
5681 } while (--count > 0);
5682
5683 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
5684
5685 /* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
5686 if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5687 tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
5688
5689 /* Force memory writes to complete before letting h/w
5690 * know there are new descriptors to fetch. (Only
5691 * applicable for weak-ordered memory model archs,
5692 * such as IA-64).
5693 */
5694 wmb();
5695
5696 tx_ring->next_to_use = i;
5697}
5698
5699#define MINIMUM_DHCP_PACKET_SIZE 282
5700static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
5701 struct sk_buff *skb)
5702{
5703 struct e1000_hw *hw = &adapter->hw;
5704 u16 length, offset;
5705
5706 if (skb_vlan_tag_present(skb) &&
5707 !((skb_vlan_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
5708 (adapter->hw.mng_cookie.status &
5709 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
5710 return 0;
5711
5712 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
5713 return 0;
5714
5715 if (((struct ethhdr *)skb->data)->h_proto != htons(ETH_P_IP))
5716 return 0;
5717
5718 {
5719 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data + 14);
5720 struct udphdr *udp;
5721
5722 if (ip->protocol != IPPROTO_UDP)
5723 return 0;
5724
5725 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
5726 if (ntohs(udp->dest) != 67)
5727 return 0;
5728
5729 offset = (u8 *)udp + 8 - skb->data;
5730 length = skb->len - offset;
5731 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
5732 }
5733
5734 return 0;
5735}
5736
5737static int __e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5738{
5739 struct e1000_adapter *adapter = tx_ring->adapter;
5740
5741 netif_stop_queue(adapter->netdev);
5742 /* Herbert's original patch had:
5743 * smp_mb__after_netif_stop_queue();
5744 * but since that doesn't exist yet, just open code it.
5745 */
5746 smp_mb();
5747
5748 /* We need to check again in a case another CPU has just
5749 * made room available.
5750 */
5751 if (e1000_desc_unused(tx_ring) < size)
5752 return -EBUSY;
5753
5754 /* A reprieve! */
5755 netif_start_queue(adapter->netdev);
5756 ++adapter->restart_queue;
5757 return 0;
5758}
5759
5760static int e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5761{
5762 BUG_ON(size > tx_ring->count);
5763
5764 if (e1000_desc_unused(tx_ring) >= size)
5765 return 0;
5766 return __e1000_maybe_stop_tx(tx_ring, size);
5767}
5768
5769static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
5770 struct net_device *netdev)
5771{
5772 struct e1000_adapter *adapter = netdev_priv(netdev);
5773 struct e1000_ring *tx_ring = adapter->tx_ring;
5774 unsigned int first;
5775 unsigned int tx_flags = 0;
5776 unsigned int len = skb_headlen(skb);
5777 unsigned int nr_frags;
5778 unsigned int mss;
5779 int count = 0;
5780 int tso;
5781 unsigned int f;
5782 __be16 protocol = vlan_get_protocol(skb);
5783
5784 if (test_bit(__E1000_DOWN, &adapter->state)) {
5785 dev_kfree_skb_any(skb);
5786 return NETDEV_TX_OK;
5787 }
5788
5789 if (skb->len <= 0) {
5790 dev_kfree_skb_any(skb);
5791 return NETDEV_TX_OK;
5792 }
5793
5794 /* The minimum packet size with TCTL.PSP set is 17 bytes so
5795 * pad skb in order to meet this minimum size requirement
5796 */
5797 if (skb_put_padto(skb, 17))
5798 return NETDEV_TX_OK;
5799
5800 mss = skb_shinfo(skb)->gso_size;
5801 if (mss) {
5802 u8 hdr_len;
5803
5804 /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
5805 * points to just header, pull a few bytes of payload from
5806 * frags into skb->data
5807 */
5808 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5809 /* we do this workaround for ES2LAN, but it is un-necessary,
5810 * avoiding it could save a lot of cycles
5811 */
5812 if (skb->data_len && (hdr_len == len)) {
5813 unsigned int pull_size;
5814
5815 pull_size = min_t(unsigned int, 4, skb->data_len);
5816 if (!__pskb_pull_tail(skb, pull_size)) {
5817 e_err("__pskb_pull_tail failed.\n");
5818 dev_kfree_skb_any(skb);
5819 return NETDEV_TX_OK;
5820 }
5821 len = skb_headlen(skb);
5822 }
5823 }
5824
5825 /* reserve a descriptor for the offload context */
5826 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5827 count++;
5828 count++;
5829
5830 count += DIV_ROUND_UP(len, adapter->tx_fifo_limit);
5831
5832 nr_frags = skb_shinfo(skb)->nr_frags;
5833 for (f = 0; f < nr_frags; f++)
5834 count += DIV_ROUND_UP(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5835 adapter->tx_fifo_limit);
5836
5837 if (adapter->hw.mac.tx_pkt_filtering)
5838 e1000_transfer_dhcp_info(adapter, skb);
5839
5840 /* need: count + 2 desc gap to keep tail from touching
5841 * head, otherwise try next time
5842 */
5843 if (e1000_maybe_stop_tx(tx_ring, count + 2))
5844 return NETDEV_TX_BUSY;
5845
5846 if (skb_vlan_tag_present(skb)) {
5847 tx_flags |= E1000_TX_FLAGS_VLAN;
5848 tx_flags |= (skb_vlan_tag_get(skb) <<
5849 E1000_TX_FLAGS_VLAN_SHIFT);
5850 }
5851
5852 first = tx_ring->next_to_use;
5853
5854 tso = e1000_tso(tx_ring, skb, protocol);
5855 if (tso < 0) {
5856 dev_kfree_skb_any(skb);
5857 return NETDEV_TX_OK;
5858 }
5859
5860 if (tso)
5861 tx_flags |= E1000_TX_FLAGS_TSO;
5862 else if (e1000_tx_csum(tx_ring, skb, protocol))
5863 tx_flags |= E1000_TX_FLAGS_CSUM;
5864
5865 /* Old method was to assume IPv4 packet by default if TSO was enabled.
5866 * 82571 hardware supports TSO capabilities for IPv6 as well...
5867 * no longer assume, we must.
5868 */
5869 if (protocol == htons(ETH_P_IP))
5870 tx_flags |= E1000_TX_FLAGS_IPV4;
5871
5872 if (unlikely(skb->no_fcs))
5873 tx_flags |= E1000_TX_FLAGS_NO_FCS;
5874
5875 /* if count is 0 then mapping error has occurred */
5876 count = e1000_tx_map(tx_ring, skb, first, adapter->tx_fifo_limit,
5877 nr_frags);
5878 if (count) {
5879 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
5880 (adapter->flags & FLAG_HAS_HW_TIMESTAMP)) {
5881 if (!adapter->tx_hwtstamp_skb) {
5882 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
5883 tx_flags |= E1000_TX_FLAGS_HWTSTAMP;
5884 adapter->tx_hwtstamp_skb = skb_get(skb);
5885 adapter->tx_hwtstamp_start = jiffies;
5886 schedule_work(&adapter->tx_hwtstamp_work);
5887 } else {
5888 adapter->tx_hwtstamp_skipped++;
5889 }
5890 }
5891
5892 skb_tx_timestamp(skb);
5893
5894 netdev_sent_queue(netdev, skb->len);
5895 e1000_tx_queue(tx_ring, tx_flags, count);
5896 /* Make sure there is space in the ring for the next send. */
5897 e1000_maybe_stop_tx(tx_ring,
5898 (MAX_SKB_FRAGS *
5899 DIV_ROUND_UP(PAGE_SIZE,
5900 adapter->tx_fifo_limit) + 2));
5901
5902 if (!skb->xmit_more ||
5903 netif_xmit_stopped(netdev_get_tx_queue(netdev, 0))) {
5904 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
5905 e1000e_update_tdt_wa(tx_ring,
5906 tx_ring->next_to_use);
5907 else
5908 writel(tx_ring->next_to_use, tx_ring->tail);
5909
5910 /* we need this if more than one processor can write
5911 * to our tail at a time, it synchronizes IO on
5912 *IA64/Altix systems
5913 */
5914 mmiowb();
5915 }
5916 } else {
5917 dev_kfree_skb_any(skb);
5918 tx_ring->buffer_info[first].time_stamp = 0;
5919 tx_ring->next_to_use = first;
5920 }
5921
5922 return NETDEV_TX_OK;
5923}
5924
5925/**
5926 * e1000_tx_timeout - Respond to a Tx Hang
5927 * @netdev: network interface device structure
5928 **/
5929static void e1000_tx_timeout(struct net_device *netdev)
5930{
5931 struct e1000_adapter *adapter = netdev_priv(netdev);
5932
5933 /* Do the reset outside of interrupt context */
5934 adapter->tx_timeout_count++;
5935 schedule_work(&adapter->reset_task);
5936}
5937
5938static void e1000_reset_task(struct work_struct *work)
5939{
5940 struct e1000_adapter *adapter;
5941 adapter = container_of(work, struct e1000_adapter, reset_task);
5942
5943 /* don't run the task if already down */
5944 if (test_bit(__E1000_DOWN, &adapter->state))
5945 return;
5946
5947 if (!(adapter->flags & FLAG_RESTART_NOW)) {
5948 e1000e_dump(adapter);
5949 e_err("Reset adapter unexpectedly\n");
5950 }
5951 e1000e_reinit_locked(adapter);
5952}
5953
5954/**
5955 * e1000_get_stats64 - Get System Network Statistics
5956 * @netdev: network interface device structure
5957 * @stats: rtnl_link_stats64 pointer
5958 *
5959 * Returns the address of the device statistics structure.
5960 **/
5961void e1000e_get_stats64(struct net_device *netdev,
5962 struct rtnl_link_stats64 *stats)
5963{
5964 struct e1000_adapter *adapter = netdev_priv(netdev);
5965
5966 spin_lock(&adapter->stats64_lock);
5967 e1000e_update_stats(adapter);
5968 /* Fill out the OS statistics structure */
5969 stats->rx_bytes = adapter->stats.gorc;
5970 stats->rx_packets = adapter->stats.gprc;
5971 stats->tx_bytes = adapter->stats.gotc;
5972 stats->tx_packets = adapter->stats.gptc;
5973 stats->multicast = adapter->stats.mprc;
5974 stats->collisions = adapter->stats.colc;
5975
5976 /* Rx Errors */
5977
5978 /* RLEC on some newer hardware can be incorrect so build
5979 * our own version based on RUC and ROC
5980 */
5981 stats->rx_errors = adapter->stats.rxerrc +
5982 adapter->stats.crcerrs + adapter->stats.algnerrc +
5983 adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
5984 stats->rx_length_errors = adapter->stats.ruc + adapter->stats.roc;
5985 stats->rx_crc_errors = adapter->stats.crcerrs;
5986 stats->rx_frame_errors = adapter->stats.algnerrc;
5987 stats->rx_missed_errors = adapter->stats.mpc;
5988
5989 /* Tx Errors */
5990 stats->tx_errors = adapter->stats.ecol + adapter->stats.latecol;
5991 stats->tx_aborted_errors = adapter->stats.ecol;
5992 stats->tx_window_errors = adapter->stats.latecol;
5993 stats->tx_carrier_errors = adapter->stats.tncrs;
5994
5995 /* Tx Dropped needs to be maintained elsewhere */
5996
5997 spin_unlock(&adapter->stats64_lock);
5998}
5999
6000/**
6001 * e1000_change_mtu - Change the Maximum Transfer Unit
6002 * @netdev: network interface device structure
6003 * @new_mtu: new value for maximum frame size
6004 *
6005 * Returns 0 on success, negative on failure
6006 **/
6007static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
6008{
6009 struct e1000_adapter *adapter = netdev_priv(netdev);
6010 int max_frame = new_mtu + VLAN_ETH_HLEN + ETH_FCS_LEN;
6011
6012 /* Jumbo frame support */
6013 if ((new_mtu > ETH_DATA_LEN) &&
6014 !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
6015 e_err("Jumbo Frames not supported.\n");
6016 return -EINVAL;
6017 }
6018
6019 /* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
6020 if ((adapter->hw.mac.type >= e1000_pch2lan) &&
6021 !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
6022 (new_mtu > ETH_DATA_LEN)) {
6023 e_err("Jumbo Frames not supported on this device when CRC stripping is disabled.\n");
6024 return -EINVAL;
6025 }
6026
6027 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
6028 usleep_range(1000, 2000);
6029 /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
6030 adapter->max_frame_size = max_frame;
6031 e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
6032 netdev->mtu = new_mtu;
6033
6034 pm_runtime_get_sync(netdev->dev.parent);
6035
6036 if (netif_running(netdev))
6037 e1000e_down(adapter, true);
6038
6039 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
6040 * means we reserve 2 more, this pushes us to allocate from the next
6041 * larger slab size.
6042 * i.e. RXBUFFER_2048 --> size-4096 slab
6043 * However with the new *_jumbo_rx* routines, jumbo receives will use
6044 * fragmented skbs
6045 */
6046
6047 if (max_frame <= 2048)
6048 adapter->rx_buffer_len = 2048;
6049 else
6050 adapter->rx_buffer_len = 4096;
6051
6052 /* adjust allocation if LPE protects us, and we aren't using SBP */
6053 if (max_frame <= (VLAN_ETH_FRAME_LEN + ETH_FCS_LEN))
6054 adapter->rx_buffer_len = VLAN_ETH_FRAME_LEN + ETH_FCS_LEN;
6055
6056 if (netif_running(netdev))
6057 e1000e_up(adapter);
6058 else
6059 e1000e_reset(adapter);
6060
6061 pm_runtime_put_sync(netdev->dev.parent);
6062
6063 clear_bit(__E1000_RESETTING, &adapter->state);
6064
6065 return 0;
6066}
6067
6068static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
6069 int cmd)
6070{
6071 struct e1000_adapter *adapter = netdev_priv(netdev);
6072 struct mii_ioctl_data *data = if_mii(ifr);
6073
6074 if (adapter->hw.phy.media_type != e1000_media_type_copper)
6075 return -EOPNOTSUPP;
6076
6077 switch (cmd) {
6078 case SIOCGMIIPHY:
6079 data->phy_id = adapter->hw.phy.addr;
6080 break;
6081 case SIOCGMIIREG:
6082 e1000_phy_read_status(adapter);
6083
6084 switch (data->reg_num & 0x1F) {
6085 case MII_BMCR:
6086 data->val_out = adapter->phy_regs.bmcr;
6087 break;
6088 case MII_BMSR:
6089 data->val_out = adapter->phy_regs.bmsr;
6090 break;
6091 case MII_PHYSID1:
6092 data->val_out = (adapter->hw.phy.id >> 16);
6093 break;
6094 case MII_PHYSID2:
6095 data->val_out = (adapter->hw.phy.id & 0xFFFF);
6096 break;
6097 case MII_ADVERTISE:
6098 data->val_out = adapter->phy_regs.advertise;
6099 break;
6100 case MII_LPA:
6101 data->val_out = adapter->phy_regs.lpa;
6102 break;
6103 case MII_EXPANSION:
6104 data->val_out = adapter->phy_regs.expansion;
6105 break;
6106 case MII_CTRL1000:
6107 data->val_out = adapter->phy_regs.ctrl1000;
6108 break;
6109 case MII_STAT1000:
6110 data->val_out = adapter->phy_regs.stat1000;
6111 break;
6112 case MII_ESTATUS:
6113 data->val_out = adapter->phy_regs.estatus;
6114 break;
6115 default:
6116 return -EIO;
6117 }
6118 break;
6119 case SIOCSMIIREG:
6120 default:
6121 return -EOPNOTSUPP;
6122 }
6123 return 0;
6124}
6125
6126/**
6127 * e1000e_hwtstamp_ioctl - control hardware time stamping
6128 * @netdev: network interface device structure
6129 * @ifreq: interface request
6130 *
6131 * Outgoing time stamping can be enabled and disabled. Play nice and
6132 * disable it when requested, although it shouldn't cause any overhead
6133 * when no packet needs it. At most one packet in the queue may be
6134 * marked for time stamping, otherwise it would be impossible to tell
6135 * for sure to which packet the hardware time stamp belongs.
6136 *
6137 * Incoming time stamping has to be configured via the hardware filters.
6138 * Not all combinations are supported, in particular event type has to be
6139 * specified. Matching the kind of event packet is not supported, with the
6140 * exception of "all V2 events regardless of level 2 or 4".
6141 **/
6142static int e1000e_hwtstamp_set(struct net_device *netdev, struct ifreq *ifr)
6143{
6144 struct e1000_adapter *adapter = netdev_priv(netdev);
6145 struct hwtstamp_config config;
6146 int ret_val;
6147
6148 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
6149 return -EFAULT;
6150
6151 ret_val = e1000e_config_hwtstamp(adapter, &config);
6152 if (ret_val)
6153 return ret_val;
6154
6155 switch (config.rx_filter) {
6156 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
6157 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
6158 case HWTSTAMP_FILTER_PTP_V2_SYNC:
6159 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
6160 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
6161 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
6162 /* With V2 type filters which specify a Sync or Delay Request,
6163 * Path Delay Request/Response messages are also time stamped
6164 * by hardware so notify the caller the requested packets plus
6165 * some others are time stamped.
6166 */
6167 config.rx_filter = HWTSTAMP_FILTER_SOME;
6168 break;
6169 default:
6170 break;
6171 }
6172
6173 return copy_to_user(ifr->ifr_data, &config,
6174 sizeof(config)) ? -EFAULT : 0;
6175}
6176
6177static int e1000e_hwtstamp_get(struct net_device *netdev, struct ifreq *ifr)
6178{
6179 struct e1000_adapter *adapter = netdev_priv(netdev);
6180
6181 return copy_to_user(ifr->ifr_data, &adapter->hwtstamp_config,
6182 sizeof(adapter->hwtstamp_config)) ? -EFAULT : 0;
6183}
6184
6185static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6186{
6187 switch (cmd) {
6188 case SIOCGMIIPHY:
6189 case SIOCGMIIREG:
6190 case SIOCSMIIREG:
6191 return e1000_mii_ioctl(netdev, ifr, cmd);
6192 case SIOCSHWTSTAMP:
6193 return e1000e_hwtstamp_set(netdev, ifr);
6194 case SIOCGHWTSTAMP:
6195 return e1000e_hwtstamp_get(netdev, ifr);
6196 default:
6197 return -EOPNOTSUPP;
6198 }
6199}
6200
6201static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
6202{
6203 struct e1000_hw *hw = &adapter->hw;
6204 u32 i, mac_reg, wuc;
6205 u16 phy_reg, wuc_enable;
6206 int retval;
6207
6208 /* copy MAC RARs to PHY RARs */
6209 e1000_copy_rx_addrs_to_phy_ich8lan(hw);
6210
6211 retval = hw->phy.ops.acquire(hw);
6212 if (retval) {
6213 e_err("Could not acquire PHY\n");
6214 return retval;
6215 }
6216
6217 /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
6218 retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
6219 if (retval)
6220 goto release;
6221
6222 /* copy MAC MTA to PHY MTA - only needed for pchlan */
6223 for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
6224 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
6225 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
6226 (u16)(mac_reg & 0xFFFF));
6227 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
6228 (u16)((mac_reg >> 16) & 0xFFFF));
6229 }
6230
6231 /* configure PHY Rx Control register */
6232 hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
6233 mac_reg = er32(RCTL);
6234 if (mac_reg & E1000_RCTL_UPE)
6235 phy_reg |= BM_RCTL_UPE;
6236 if (mac_reg & E1000_RCTL_MPE)
6237 phy_reg |= BM_RCTL_MPE;
6238 phy_reg &= ~(BM_RCTL_MO_MASK);
6239 if (mac_reg & E1000_RCTL_MO_3)
6240 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
6241 << BM_RCTL_MO_SHIFT);
6242 if (mac_reg & E1000_RCTL_BAM)
6243 phy_reg |= BM_RCTL_BAM;
6244 if (mac_reg & E1000_RCTL_PMCF)
6245 phy_reg |= BM_RCTL_PMCF;
6246 mac_reg = er32(CTRL);
6247 if (mac_reg & E1000_CTRL_RFCE)
6248 phy_reg |= BM_RCTL_RFCE;
6249 hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
6250
6251 wuc = E1000_WUC_PME_EN;
6252 if (wufc & (E1000_WUFC_MAG | E1000_WUFC_LNKC))
6253 wuc |= E1000_WUC_APME;
6254
6255 /* enable PHY wakeup in MAC register */
6256 ew32(WUFC, wufc);
6257 ew32(WUC, (E1000_WUC_PHY_WAKE | E1000_WUC_APMPME |
6258 E1000_WUC_PME_STATUS | wuc));
6259
6260 /* configure and enable PHY wakeup in PHY registers */
6261 hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
6262 hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, wuc);
6263
6264 /* activate PHY wakeup */
6265 wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
6266 retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
6267 if (retval)
6268 e_err("Could not set PHY Host Wakeup bit\n");
6269release:
6270 hw->phy.ops.release(hw);
6271
6272 return retval;
6273}
6274
6275static void e1000e_flush_lpic(struct pci_dev *pdev)
6276{
6277 struct net_device *netdev = pci_get_drvdata(pdev);
6278 struct e1000_adapter *adapter = netdev_priv(netdev);
6279 struct e1000_hw *hw = &adapter->hw;
6280 u32 ret_val;
6281
6282 pm_runtime_get_sync(netdev->dev.parent);
6283
6284 ret_val = hw->phy.ops.acquire(hw);
6285 if (ret_val)
6286 goto fl_out;
6287
6288 pr_info("EEE TX LPI TIMER: %08X\n",
6289 er32(LPIC) >> E1000_LPIC_LPIET_SHIFT);
6290
6291 hw->phy.ops.release(hw);
6292
6293fl_out:
6294 pm_runtime_put_sync(netdev->dev.parent);
6295}
6296
6297static int e1000e_pm_freeze(struct device *dev)
6298{
6299 struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
6300 struct e1000_adapter *adapter = netdev_priv(netdev);
6301
6302 netif_device_detach(netdev);
6303
6304 if (netif_running(netdev)) {
6305 int count = E1000_CHECK_RESET_COUNT;
6306
6307 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
6308 usleep_range(10000, 20000);
6309
6310 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
6311
6312 /* Quiesce the device without resetting the hardware */
6313 e1000e_down(adapter, false);
6314 e1000_free_irq(adapter);
6315 }
6316 e1000e_reset_interrupt_capability(adapter);
6317
6318 /* Allow time for pending master requests to run */
6319 e1000e_disable_pcie_master(&adapter->hw);
6320
6321 return 0;
6322}
6323
6324static int __e1000_shutdown(struct pci_dev *pdev, bool runtime)
6325{
6326 struct net_device *netdev = pci_get_drvdata(pdev);
6327 struct e1000_adapter *adapter = netdev_priv(netdev);
6328 struct e1000_hw *hw = &adapter->hw;
6329 u32 ctrl, ctrl_ext, rctl, status;
6330 /* Runtime suspend should only enable wakeup for link changes */
6331 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
6332 int retval = 0;
6333
6334 status = er32(STATUS);
6335 if (status & E1000_STATUS_LU)
6336 wufc &= ~E1000_WUFC_LNKC;
6337
6338 if (wufc) {
6339 e1000_setup_rctl(adapter);
6340 e1000e_set_rx_mode(netdev);
6341
6342 /* turn on all-multi mode if wake on multicast is enabled */
6343 if (wufc & E1000_WUFC_MC) {
6344 rctl = er32(RCTL);
6345 rctl |= E1000_RCTL_MPE;
6346 ew32(RCTL, rctl);
6347 }
6348
6349 ctrl = er32(CTRL);
6350 ctrl |= E1000_CTRL_ADVD3WUC;
6351 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
6352 ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
6353 ew32(CTRL, ctrl);
6354
6355 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
6356 adapter->hw.phy.media_type ==
6357 e1000_media_type_internal_serdes) {
6358 /* keep the laser running in D3 */
6359 ctrl_ext = er32(CTRL_EXT);
6360 ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
6361 ew32(CTRL_EXT, ctrl_ext);
6362 }
6363
6364 if (!runtime)
6365 e1000e_power_up_phy(adapter);
6366
6367 if (adapter->flags & FLAG_IS_ICH)
6368 e1000_suspend_workarounds_ich8lan(&adapter->hw);
6369
6370 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
6371 /* enable wakeup by the PHY */
6372 retval = e1000_init_phy_wakeup(adapter, wufc);
6373 if (retval)
6374 return retval;
6375 } else {
6376 /* enable wakeup by the MAC */
6377 ew32(WUFC, wufc);
6378 ew32(WUC, E1000_WUC_PME_EN);
6379 }
6380 } else {
6381 ew32(WUC, 0);
6382 ew32(WUFC, 0);
6383
6384 e1000_power_down_phy(adapter);
6385 }
6386
6387 if (adapter->hw.phy.type == e1000_phy_igp_3) {
6388 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
6389 } else if (hw->mac.type >= e1000_pch_lpt) {
6390 if (!(wufc & (E1000_WUFC_EX | E1000_WUFC_MC | E1000_WUFC_BC)))
6391 /* ULP does not support wake from unicast, multicast
6392 * or broadcast.
6393 */
6394 retval = e1000_enable_ulp_lpt_lp(hw, !runtime);
6395
6396 if (retval)
6397 return retval;
6398 }
6399
6400 /* Ensure that the appropriate bits are set in LPI_CTRL
6401 * for EEE in Sx
6402 */
6403 if ((hw->phy.type >= e1000_phy_i217) &&
6404 adapter->eee_advert && hw->dev_spec.ich8lan.eee_lp_ability) {
6405 u16 lpi_ctrl = 0;
6406
6407 retval = hw->phy.ops.acquire(hw);
6408 if (!retval) {
6409 retval = e1e_rphy_locked(hw, I82579_LPI_CTRL,
6410 &lpi_ctrl);
6411 if (!retval) {
6412 if (adapter->eee_advert &
6413 hw->dev_spec.ich8lan.eee_lp_ability &
6414 I82579_EEE_100_SUPPORTED)
6415 lpi_ctrl |= I82579_LPI_CTRL_100_ENABLE;
6416 if (adapter->eee_advert &
6417 hw->dev_spec.ich8lan.eee_lp_ability &
6418 I82579_EEE_1000_SUPPORTED)
6419 lpi_ctrl |= I82579_LPI_CTRL_1000_ENABLE;
6420
6421 retval = e1e_wphy_locked(hw, I82579_LPI_CTRL,
6422 lpi_ctrl);
6423 }
6424 }
6425 hw->phy.ops.release(hw);
6426 }
6427
6428 /* Release control of h/w to f/w. If f/w is AMT enabled, this
6429 * would have already happened in close and is redundant.
6430 */
6431 e1000e_release_hw_control(adapter);
6432
6433 pci_clear_master(pdev);
6434
6435 /* The pci-e switch on some quad port adapters will report a
6436 * correctable error when the MAC transitions from D0 to D3. To
6437 * prevent this we need to mask off the correctable errors on the
6438 * downstream port of the pci-e switch.
6439 *
6440 * We don't have the associated upstream bridge while assigning
6441 * the PCI device into guest. For example, the KVM on power is
6442 * one of the cases.
6443 */
6444 if (adapter->flags & FLAG_IS_QUAD_PORT) {
6445 struct pci_dev *us_dev = pdev->bus->self;
6446 u16 devctl;
6447
6448 if (!us_dev)
6449 return 0;
6450
6451 pcie_capability_read_word(us_dev, PCI_EXP_DEVCTL, &devctl);
6452 pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL,
6453 (devctl & ~PCI_EXP_DEVCTL_CERE));
6454
6455 pci_save_state(pdev);
6456 pci_prepare_to_sleep(pdev);
6457
6458 pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL, devctl);
6459 }
6460
6461 return 0;
6462}
6463
6464/**
6465 * __e1000e_disable_aspm - Disable ASPM states
6466 * @pdev: pointer to PCI device struct
6467 * @state: bit-mask of ASPM states to disable
6468 * @locked: indication if this context holds pci_bus_sem locked.
6469 *
6470 * Some devices *must* have certain ASPM states disabled per hardware errata.
6471 **/
6472static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state, int locked)
6473{
6474 struct pci_dev *parent = pdev->bus->self;
6475 u16 aspm_dis_mask = 0;
6476 u16 pdev_aspmc, parent_aspmc;
6477
6478 switch (state) {
6479 case PCIE_LINK_STATE_L0S:
6480 case PCIE_LINK_STATE_L0S | PCIE_LINK_STATE_L1:
6481 aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L0S;
6482 /* fall-through - can't have L1 without L0s */
6483 case PCIE_LINK_STATE_L1:
6484 aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L1;
6485 break;
6486 default:
6487 return;
6488 }
6489
6490 pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6491 pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6492
6493 if (parent) {
6494 pcie_capability_read_word(parent, PCI_EXP_LNKCTL,
6495 &parent_aspmc);
6496 parent_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6497 }
6498
6499 /* Nothing to do if the ASPM states to be disabled already are */
6500 if (!(pdev_aspmc & aspm_dis_mask) &&
6501 (!parent || !(parent_aspmc & aspm_dis_mask)))
6502 return;
6503
6504 dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
6505 (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L0S) ?
6506 "L0s" : "",
6507 (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L1) ?
6508 "L1" : "");
6509
6510#ifdef CONFIG_PCIEASPM
6511 if (locked)
6512 pci_disable_link_state_locked(pdev, state);
6513 else
6514 pci_disable_link_state(pdev, state);
6515
6516 /* Double-check ASPM control. If not disabled by the above, the
6517 * BIOS is preventing that from happening (or CONFIG_PCIEASPM is
6518 * not enabled); override by writing PCI config space directly.
6519 */
6520 pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6521 pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6522
6523 if (!(aspm_dis_mask & pdev_aspmc))
6524 return;
6525#endif
6526
6527 /* Both device and parent should have the same ASPM setting.
6528 * Disable ASPM in downstream component first and then upstream.
6529 */
6530 pcie_capability_clear_word(pdev, PCI_EXP_LNKCTL, aspm_dis_mask);
6531
6532 if (parent)
6533 pcie_capability_clear_word(parent, PCI_EXP_LNKCTL,
6534 aspm_dis_mask);
6535}
6536
6537/**
6538 * e1000e_disable_aspm - Disable ASPM states.
6539 * @pdev: pointer to PCI device struct
6540 * @state: bit-mask of ASPM states to disable
6541 *
6542 * This function acquires the pci_bus_sem!
6543 * Some devices *must* have certain ASPM states disabled per hardware errata.
6544 **/
6545static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
6546{
6547 __e1000e_disable_aspm(pdev, state, 0);
6548}
6549
6550/**
6551 * e1000e_disable_aspm_locked Disable ASPM states.
6552 * @pdev: pointer to PCI device struct
6553 * @state: bit-mask of ASPM states to disable
6554 *
6555 * This function must be called with pci_bus_sem acquired!
6556 * Some devices *must* have certain ASPM states disabled per hardware errata.
6557 **/
6558static void e1000e_disable_aspm_locked(struct pci_dev *pdev, u16 state)
6559{
6560 __e1000e_disable_aspm(pdev, state, 1);
6561}
6562
6563#ifdef CONFIG_PM
6564static int __e1000_resume(struct pci_dev *pdev)
6565{
6566 struct net_device *netdev = pci_get_drvdata(pdev);
6567 struct e1000_adapter *adapter = netdev_priv(netdev);
6568 struct e1000_hw *hw = &adapter->hw;
6569 u16 aspm_disable_flag = 0;
6570
6571 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
6572 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6573 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
6574 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6575 if (aspm_disable_flag)
6576 e1000e_disable_aspm(pdev, aspm_disable_flag);
6577
6578 pci_set_master(pdev);
6579
6580 if (hw->mac.type >= e1000_pch2lan)
6581 e1000_resume_workarounds_pchlan(&adapter->hw);
6582
6583 e1000e_power_up_phy(adapter);
6584
6585 /* report the system wakeup cause from S3/S4 */
6586 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
6587 u16 phy_data;
6588
6589 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
6590 if (phy_data) {
6591 e_info("PHY Wakeup cause - %s\n",
6592 phy_data & E1000_WUS_EX ? "Unicast Packet" :
6593 phy_data & E1000_WUS_MC ? "Multicast Packet" :
6594 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
6595 phy_data & E1000_WUS_MAG ? "Magic Packet" :
6596 phy_data & E1000_WUS_LNKC ?
6597 "Link Status Change" : "other");
6598 }
6599 e1e_wphy(&adapter->hw, BM_WUS, ~0);
6600 } else {
6601 u32 wus = er32(WUS);
6602
6603 if (wus) {
6604 e_info("MAC Wakeup cause - %s\n",
6605 wus & E1000_WUS_EX ? "Unicast Packet" :
6606 wus & E1000_WUS_MC ? "Multicast Packet" :
6607 wus & E1000_WUS_BC ? "Broadcast Packet" :
6608 wus & E1000_WUS_MAG ? "Magic Packet" :
6609 wus & E1000_WUS_LNKC ? "Link Status Change" :
6610 "other");
6611 }
6612 ew32(WUS, ~0);
6613 }
6614
6615 e1000e_reset(adapter);
6616
6617 e1000_init_manageability_pt(adapter);
6618
6619 /* If the controller has AMT, do not set DRV_LOAD until the interface
6620 * is up. For all other cases, let the f/w know that the h/w is now
6621 * under the control of the driver.
6622 */
6623 if (!(adapter->flags & FLAG_HAS_AMT))
6624 e1000e_get_hw_control(adapter);
6625
6626 return 0;
6627}
6628
6629#ifdef CONFIG_PM_SLEEP
6630static int e1000e_pm_thaw(struct device *dev)
6631{
6632 struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
6633 struct e1000_adapter *adapter = netdev_priv(netdev);
6634
6635 e1000e_set_interrupt_capability(adapter);
6636 if (netif_running(netdev)) {
6637 u32 err = e1000_request_irq(adapter);
6638
6639 if (err)
6640 return err;
6641
6642 e1000e_up(adapter);
6643 }
6644
6645 netif_device_attach(netdev);
6646
6647 return 0;
6648}
6649
6650static int e1000e_pm_suspend(struct device *dev)
6651{
6652 struct pci_dev *pdev = to_pci_dev(dev);
6653 int rc;
6654
6655 e1000e_flush_lpic(pdev);
6656
6657 e1000e_pm_freeze(dev);
6658
6659 rc = __e1000_shutdown(pdev, false);
6660 if (rc)
6661 e1000e_pm_thaw(dev);
6662
6663 return rc;
6664}
6665
6666static int e1000e_pm_resume(struct device *dev)
6667{
6668 struct pci_dev *pdev = to_pci_dev(dev);
6669 int rc;
6670
6671 rc = __e1000_resume(pdev);
6672 if (rc)
6673 return rc;
6674
6675 return e1000e_pm_thaw(dev);
6676}
6677#endif /* CONFIG_PM_SLEEP */
6678
6679static int e1000e_pm_runtime_idle(struct device *dev)
6680{
6681 struct pci_dev *pdev = to_pci_dev(dev);
6682 struct net_device *netdev = pci_get_drvdata(pdev);
6683 struct e1000_adapter *adapter = netdev_priv(netdev);
6684 u16 eee_lp;
6685
6686 eee_lp = adapter->hw.dev_spec.ich8lan.eee_lp_ability;
6687
6688 if (!e1000e_has_link(adapter)) {
6689 adapter->hw.dev_spec.ich8lan.eee_lp_ability = eee_lp;
6690 pm_schedule_suspend(dev, 5 * MSEC_PER_SEC);
6691 }
6692
6693 return -EBUSY;
6694}
6695
6696static int e1000e_pm_runtime_resume(struct device *dev)
6697{
6698 struct pci_dev *pdev = to_pci_dev(dev);
6699 struct net_device *netdev = pci_get_drvdata(pdev);
6700 struct e1000_adapter *adapter = netdev_priv(netdev);
6701 int rc;
6702
6703 rc = __e1000_resume(pdev);
6704 if (rc)
6705 return rc;
6706
6707 if (netdev->flags & IFF_UP)
6708 e1000e_up(adapter);
6709
6710 return rc;
6711}
6712
6713static int e1000e_pm_runtime_suspend(struct device *dev)
6714{
6715 struct pci_dev *pdev = to_pci_dev(dev);
6716 struct net_device *netdev = pci_get_drvdata(pdev);
6717 struct e1000_adapter *adapter = netdev_priv(netdev);
6718
6719 if (netdev->flags & IFF_UP) {
6720 int count = E1000_CHECK_RESET_COUNT;
6721
6722 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
6723 usleep_range(10000, 20000);
6724
6725 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
6726
6727 /* Down the device without resetting the hardware */
6728 e1000e_down(adapter, false);
6729 }
6730
6731 if (__e1000_shutdown(pdev, true)) {
6732 e1000e_pm_runtime_resume(dev);
6733 return -EBUSY;
6734 }
6735
6736 return 0;
6737}
6738#endif /* CONFIG_PM */
6739
6740static void e1000_shutdown(struct pci_dev *pdev)
6741{
6742 e1000e_flush_lpic(pdev);
6743
6744 e1000e_pm_freeze(&pdev->dev);
6745
6746 __e1000_shutdown(pdev, false);
6747}
6748
6749#ifdef CONFIG_NET_POLL_CONTROLLER
6750
6751static irqreturn_t e1000_intr_msix(int __always_unused irq, void *data)
6752{
6753 struct net_device *netdev = data;
6754 struct e1000_adapter *adapter = netdev_priv(netdev);
6755
6756 if (adapter->msix_entries) {
6757 int vector, msix_irq;
6758
6759 vector = 0;
6760 msix_irq = adapter->msix_entries[vector].vector;
6761 if (disable_hardirq(msix_irq))
6762 e1000_intr_msix_rx(msix_irq, netdev);
6763 enable_irq(msix_irq);
6764
6765 vector++;
6766 msix_irq = adapter->msix_entries[vector].vector;
6767 if (disable_hardirq(msix_irq))
6768 e1000_intr_msix_tx(msix_irq, netdev);
6769 enable_irq(msix_irq);
6770
6771 vector++;
6772 msix_irq = adapter->msix_entries[vector].vector;
6773 if (disable_hardirq(msix_irq))
6774 e1000_msix_other(msix_irq, netdev);
6775 enable_irq(msix_irq);
6776 }
6777
6778 return IRQ_HANDLED;
6779}
6780
6781/**
6782 * e1000_netpoll
6783 * @netdev: network interface device structure
6784 *
6785 * Polling 'interrupt' - used by things like netconsole to send skbs
6786 * without having to re-enable interrupts. It's not called while
6787 * the interrupt routine is executing.
6788 */
6789static void e1000_netpoll(struct net_device *netdev)
6790{
6791 struct e1000_adapter *adapter = netdev_priv(netdev);
6792
6793 switch (adapter->int_mode) {
6794 case E1000E_INT_MODE_MSIX:
6795 e1000_intr_msix(adapter->pdev->irq, netdev);
6796 break;
6797 case E1000E_INT_MODE_MSI:
6798 if (disable_hardirq(adapter->pdev->irq))
6799 e1000_intr_msi(adapter->pdev->irq, netdev);
6800 enable_irq(adapter->pdev->irq);
6801 break;
6802 default: /* E1000E_INT_MODE_LEGACY */
6803 if (disable_hardirq(adapter->pdev->irq))
6804 e1000_intr(adapter->pdev->irq, netdev);
6805 enable_irq(adapter->pdev->irq);
6806 break;
6807 }
6808}
6809#endif
6810
6811/**
6812 * e1000_io_error_detected - called when PCI error is detected
6813 * @pdev: Pointer to PCI device
6814 * @state: The current pci connection state
6815 *
6816 * This function is called after a PCI bus error affecting
6817 * this device has been detected.
6818 */
6819static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
6820 pci_channel_state_t state)
6821{
6822 struct net_device *netdev = pci_get_drvdata(pdev);
6823 struct e1000_adapter *adapter = netdev_priv(netdev);
6824
6825 netif_device_detach(netdev);
6826
6827 if (state == pci_channel_io_perm_failure)
6828 return PCI_ERS_RESULT_DISCONNECT;
6829
6830 if (netif_running(netdev))
6831 e1000e_down(adapter, true);
6832 pci_disable_device(pdev);
6833
6834 /* Request a slot slot reset. */
6835 return PCI_ERS_RESULT_NEED_RESET;
6836}
6837
6838/**
6839 * e1000_io_slot_reset - called after the pci bus has been reset.
6840 * @pdev: Pointer to PCI device
6841 *
6842 * Restart the card from scratch, as if from a cold-boot. Implementation
6843 * resembles the first-half of the e1000e_pm_resume routine.
6844 */
6845static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
6846{
6847 struct net_device *netdev = pci_get_drvdata(pdev);
6848 struct e1000_adapter *adapter = netdev_priv(netdev);
6849 struct e1000_hw *hw = &adapter->hw;
6850 u16 aspm_disable_flag = 0;
6851 int err;
6852 pci_ers_result_t result;
6853
6854 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
6855 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6856 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
6857 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6858 if (aspm_disable_flag)
6859 e1000e_disable_aspm_locked(pdev, aspm_disable_flag);
6860
6861 err = pci_enable_device_mem(pdev);
6862 if (err) {
6863 dev_err(&pdev->dev,
6864 "Cannot re-enable PCI device after reset.\n");
6865 result = PCI_ERS_RESULT_DISCONNECT;
6866 } else {
6867 pdev->state_saved = true;
6868 pci_restore_state(pdev);
6869 pci_set_master(pdev);
6870
6871 pci_enable_wake(pdev, PCI_D3hot, 0);
6872 pci_enable_wake(pdev, PCI_D3cold, 0);
6873
6874 e1000e_reset(adapter);
6875 ew32(WUS, ~0);
6876 result = PCI_ERS_RESULT_RECOVERED;
6877 }
6878
6879 pci_cleanup_aer_uncorrect_error_status(pdev);
6880
6881 return result;
6882}
6883
6884/**
6885 * e1000_io_resume - called when traffic can start flowing again.
6886 * @pdev: Pointer to PCI device
6887 *
6888 * This callback is called when the error recovery driver tells us that
6889 * its OK to resume normal operation. Implementation resembles the
6890 * second-half of the e1000e_pm_resume routine.
6891 */
6892static void e1000_io_resume(struct pci_dev *pdev)
6893{
6894 struct net_device *netdev = pci_get_drvdata(pdev);
6895 struct e1000_adapter *adapter = netdev_priv(netdev);
6896
6897 e1000_init_manageability_pt(adapter);
6898
6899 if (netif_running(netdev))
6900 e1000e_up(adapter);
6901
6902 netif_device_attach(netdev);
6903
6904 /* If the controller has AMT, do not set DRV_LOAD until the interface
6905 * is up. For all other cases, let the f/w know that the h/w is now
6906 * under the control of the driver.
6907 */
6908 if (!(adapter->flags & FLAG_HAS_AMT))
6909 e1000e_get_hw_control(adapter);
6910}
6911
6912static void e1000_print_device_info(struct e1000_adapter *adapter)
6913{
6914 struct e1000_hw *hw = &adapter->hw;
6915 struct net_device *netdev = adapter->netdev;
6916 u32 ret_val;
6917 u8 pba_str[E1000_PBANUM_LENGTH];
6918
6919 /* print bus type/speed/width info */
6920 e_info("(PCI Express:2.5GT/s:%s) %pM\n",
6921 /* bus width */
6922 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
6923 "Width x1"),
6924 /* MAC address */
6925 netdev->dev_addr);
6926 e_info("Intel(R) PRO/%s Network Connection\n",
6927 (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
6928 ret_val = e1000_read_pba_string_generic(hw, pba_str,
6929 E1000_PBANUM_LENGTH);
6930 if (ret_val)
6931 strlcpy((char *)pba_str, "Unknown", sizeof(pba_str));
6932 e_info("MAC: %d, PHY: %d, PBA No: %s\n",
6933 hw->mac.type, hw->phy.type, pba_str);
6934}
6935
6936static void e1000_eeprom_checks(struct e1000_adapter *adapter)
6937{
6938 struct e1000_hw *hw = &adapter->hw;
6939 int ret_val;
6940 u16 buf = 0;
6941
6942 if (hw->mac.type != e1000_82573)
6943 return;
6944
6945 ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
6946 le16_to_cpus(&buf);
6947 if (!ret_val && (!(buf & BIT(0)))) {
6948 /* Deep Smart Power Down (DSPD) */
6949 dev_warn(&adapter->pdev->dev,
6950 "Warning: detected DSPD enabled in EEPROM\n");
6951 }
6952}
6953
6954static netdev_features_t e1000_fix_features(struct net_device *netdev,
6955 netdev_features_t features)
6956{
6957 struct e1000_adapter *adapter = netdev_priv(netdev);
6958 struct e1000_hw *hw = &adapter->hw;
6959
6960 /* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
6961 if ((hw->mac.type >= e1000_pch2lan) && (netdev->mtu > ETH_DATA_LEN))
6962 features &= ~NETIF_F_RXFCS;
6963
6964 /* Since there is no support for separate Rx/Tx vlan accel
6965 * enable/disable make sure Tx flag is always in same state as Rx.
6966 */
6967 if (features & NETIF_F_HW_VLAN_CTAG_RX)
6968 features |= NETIF_F_HW_VLAN_CTAG_TX;
6969 else
6970 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
6971
6972 return features;
6973}
6974
6975static int e1000_set_features(struct net_device *netdev,
6976 netdev_features_t features)
6977{
6978 struct e1000_adapter *adapter = netdev_priv(netdev);
6979 netdev_features_t changed = features ^ netdev->features;
6980
6981 if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
6982 adapter->flags |= FLAG_TSO_FORCE;
6983
6984 if (!(changed & (NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_TX |
6985 NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_RXFCS |
6986 NETIF_F_RXALL)))
6987 return 0;
6988
6989 if (changed & NETIF_F_RXFCS) {
6990 if (features & NETIF_F_RXFCS) {
6991 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6992 } else {
6993 /* We need to take it back to defaults, which might mean
6994 * stripping is still disabled at the adapter level.
6995 */
6996 if (adapter->flags2 & FLAG2_DFLT_CRC_STRIPPING)
6997 adapter->flags2 |= FLAG2_CRC_STRIPPING;
6998 else
6999 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
7000 }
7001 }
7002
7003 netdev->features = features;
7004
7005 if (netif_running(netdev))
7006 e1000e_reinit_locked(adapter);
7007 else
7008 e1000e_reset(adapter);
7009
7010 return 0;
7011}
7012
7013static const struct net_device_ops e1000e_netdev_ops = {
7014 .ndo_open = e1000e_open,
7015 .ndo_stop = e1000e_close,
7016 .ndo_start_xmit = e1000_xmit_frame,
7017 .ndo_get_stats64 = e1000e_get_stats64,
7018 .ndo_set_rx_mode = e1000e_set_rx_mode,
7019 .ndo_set_mac_address = e1000_set_mac,
7020 .ndo_change_mtu = e1000_change_mtu,
7021 .ndo_do_ioctl = e1000_ioctl,
7022 .ndo_tx_timeout = e1000_tx_timeout,
7023 .ndo_validate_addr = eth_validate_addr,
7024
7025 .ndo_vlan_rx_add_vid = e1000_vlan_rx_add_vid,
7026 .ndo_vlan_rx_kill_vid = e1000_vlan_rx_kill_vid,
7027#ifdef CONFIG_NET_POLL_CONTROLLER
7028 .ndo_poll_controller = e1000_netpoll,
7029#endif
7030 .ndo_set_features = e1000_set_features,
7031 .ndo_fix_features = e1000_fix_features,
7032 .ndo_features_check = passthru_features_check,
7033};
7034
7035/**
7036 * e1000_probe - Device Initialization Routine
7037 * @pdev: PCI device information struct
7038 * @ent: entry in e1000_pci_tbl
7039 *
7040 * Returns 0 on success, negative on failure
7041 *
7042 * e1000_probe initializes an adapter identified by a pci_dev structure.
7043 * The OS initialization, configuring of the adapter private structure,
7044 * and a hardware reset occur.
7045 **/
7046static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
7047{
7048 struct net_device *netdev;
7049 struct e1000_adapter *adapter;
7050 struct e1000_hw *hw;
7051 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
7052 resource_size_t mmio_start, mmio_len;
7053 resource_size_t flash_start, flash_len;
7054 static int cards_found;
7055 u16 aspm_disable_flag = 0;
7056 int bars, i, err, pci_using_dac;
7057 u16 eeprom_data = 0;
7058 u16 eeprom_apme_mask = E1000_EEPROM_APME;
7059 s32 ret_val = 0;
7060
7061 if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
7062 aspm_disable_flag = PCIE_LINK_STATE_L0S;
7063 if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
7064 aspm_disable_flag |= PCIE_LINK_STATE_L1;
7065 if (aspm_disable_flag)
7066 e1000e_disable_aspm(pdev, aspm_disable_flag);
7067
7068 err = pci_enable_device_mem(pdev);
7069 if (err)
7070 return err;
7071
7072 pci_using_dac = 0;
7073 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
7074 if (!err) {
7075 pci_using_dac = 1;
7076 } else {
7077 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
7078 if (err) {
7079 dev_err(&pdev->dev,
7080 "No usable DMA configuration, aborting\n");
7081 goto err_dma;
7082 }
7083 }
7084
7085 bars = pci_select_bars(pdev, IORESOURCE_MEM);
7086 err = pci_request_selected_regions_exclusive(pdev, bars,
7087 e1000e_driver_name);
7088 if (err)
7089 goto err_pci_reg;
7090
7091 /* AER (Advanced Error Reporting) hooks */
7092 pci_enable_pcie_error_reporting(pdev);
7093
7094 pci_set_master(pdev);
7095 /* PCI config space info */
7096 err = pci_save_state(pdev);
7097 if (err)
7098 goto err_alloc_etherdev;
7099
7100 err = -ENOMEM;
7101 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
7102 if (!netdev)
7103 goto err_alloc_etherdev;
7104
7105 SET_NETDEV_DEV(netdev, &pdev->dev);
7106
7107 netdev->irq = pdev->irq;
7108
7109 pci_set_drvdata(pdev, netdev);
7110 adapter = netdev_priv(netdev);
7111 hw = &adapter->hw;
7112 adapter->netdev = netdev;
7113 adapter->pdev = pdev;
7114 adapter->ei = ei;
7115 adapter->pba = ei->pba;
7116 adapter->flags = ei->flags;
7117 adapter->flags2 = ei->flags2;
7118 adapter->hw.adapter = adapter;
7119 adapter->hw.mac.type = ei->mac;
7120 adapter->max_hw_frame_size = ei->max_hw_frame_size;
7121 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
7122
7123 mmio_start = pci_resource_start(pdev, 0);
7124 mmio_len = pci_resource_len(pdev, 0);
7125
7126 err = -EIO;
7127 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
7128 if (!adapter->hw.hw_addr)
7129 goto err_ioremap;
7130
7131 if ((adapter->flags & FLAG_HAS_FLASH) &&
7132 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM) &&
7133 (hw->mac.type < e1000_pch_spt)) {
7134 flash_start = pci_resource_start(pdev, 1);
7135 flash_len = pci_resource_len(pdev, 1);
7136 adapter->hw.flash_address = ioremap(flash_start, flash_len);
7137 if (!adapter->hw.flash_address)
7138 goto err_flashmap;
7139 }
7140
7141 /* Set default EEE advertisement */
7142 if (adapter->flags2 & FLAG2_HAS_EEE)
7143 adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
7144
7145 /* construct the net_device struct */
7146 netdev->netdev_ops = &e1000e_netdev_ops;
7147 e1000e_set_ethtool_ops(netdev);
7148 netdev->watchdog_timeo = 5 * HZ;
7149 netif_napi_add(netdev, &adapter->napi, e1000e_poll, 64);
7150 strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
7151
7152 netdev->mem_start = mmio_start;
7153 netdev->mem_end = mmio_start + mmio_len;
7154
7155 adapter->bd_number = cards_found++;
7156
7157 e1000e_check_options(adapter);
7158
7159 /* setup adapter struct */
7160 err = e1000_sw_init(adapter);
7161 if (err)
7162 goto err_sw_init;
7163
7164 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
7165 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
7166 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
7167
7168 err = ei->get_variants(adapter);
7169 if (err)
7170 goto err_hw_init;
7171
7172 if ((adapter->flags & FLAG_IS_ICH) &&
7173 (adapter->flags & FLAG_READ_ONLY_NVM) &&
7174 (hw->mac.type < e1000_pch_spt))
7175 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
7176
7177 hw->mac.ops.get_bus_info(&adapter->hw);
7178
7179 adapter->hw.phy.autoneg_wait_to_complete = 0;
7180
7181 /* Copper options */
7182 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
7183 adapter->hw.phy.mdix = AUTO_ALL_MODES;
7184 adapter->hw.phy.disable_polarity_correction = 0;
7185 adapter->hw.phy.ms_type = e1000_ms_hw_default;
7186 }
7187
7188 if (hw->phy.ops.check_reset_block && hw->phy.ops.check_reset_block(hw))
7189 dev_info(&pdev->dev,
7190 "PHY reset is blocked due to SOL/IDER session.\n");
7191
7192 /* Set initial default active device features */
7193 netdev->features = (NETIF_F_SG |
7194 NETIF_F_HW_VLAN_CTAG_RX |
7195 NETIF_F_HW_VLAN_CTAG_TX |
7196 NETIF_F_TSO |
7197 NETIF_F_TSO6 |
7198 NETIF_F_RXHASH |
7199 NETIF_F_RXCSUM |
7200 NETIF_F_HW_CSUM);
7201
7202 /* Set user-changeable features (subset of all device features) */
7203 netdev->hw_features = netdev->features;
7204 netdev->hw_features |= NETIF_F_RXFCS;
7205 netdev->priv_flags |= IFF_SUPP_NOFCS;
7206 netdev->hw_features |= NETIF_F_RXALL;
7207
7208 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
7209 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
7210
7211 netdev->vlan_features |= (NETIF_F_SG |
7212 NETIF_F_TSO |
7213 NETIF_F_TSO6 |
7214 NETIF_F_HW_CSUM);
7215
7216 netdev->priv_flags |= IFF_UNICAST_FLT;
7217
7218 if (pci_using_dac) {
7219 netdev->features |= NETIF_F_HIGHDMA;
7220 netdev->vlan_features |= NETIF_F_HIGHDMA;
7221 }
7222
7223 /* MTU range: 68 - max_hw_frame_size */
7224 netdev->min_mtu = ETH_MIN_MTU;
7225 netdev->max_mtu = adapter->max_hw_frame_size -
7226 (VLAN_ETH_HLEN + ETH_FCS_LEN);
7227
7228 if (e1000e_enable_mng_pass_thru(&adapter->hw))
7229 adapter->flags |= FLAG_MNG_PT_ENABLED;
7230
7231 /* before reading the NVM, reset the controller to
7232 * put the device in a known good starting state
7233 */
7234 adapter->hw.mac.ops.reset_hw(&adapter->hw);
7235
7236 /* systems with ASPM and others may see the checksum fail on the first
7237 * attempt. Let's give it a few tries
7238 */
7239 for (i = 0;; i++) {
7240 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
7241 break;
7242 if (i == 2) {
7243 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
7244 err = -EIO;
7245 goto err_eeprom;
7246 }
7247 }
7248
7249 e1000_eeprom_checks(adapter);
7250
7251 /* copy the MAC address */
7252 if (e1000e_read_mac_addr(&adapter->hw))
7253 dev_err(&pdev->dev,
7254 "NVM Read Error while reading MAC address\n");
7255
7256 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
7257
7258 if (!is_valid_ether_addr(netdev->dev_addr)) {
7259 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
7260 netdev->dev_addr);
7261 err = -EIO;
7262 goto err_eeprom;
7263 }
7264
7265 timer_setup(&adapter->watchdog_timer, e1000_watchdog, 0);
7266 timer_setup(&adapter->phy_info_timer, e1000_update_phy_info, 0);
7267
7268 INIT_WORK(&adapter->reset_task, e1000_reset_task);
7269 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
7270 INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
7271 INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
7272 INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
7273
7274 /* Initialize link parameters. User can change them with ethtool */
7275 adapter->hw.mac.autoneg = 1;
7276 adapter->fc_autoneg = true;
7277 adapter->hw.fc.requested_mode = e1000_fc_default;
7278 adapter->hw.fc.current_mode = e1000_fc_default;
7279 adapter->hw.phy.autoneg_advertised = 0x2f;
7280
7281 /* Initial Wake on LAN setting - If APM wake is enabled in
7282 * the EEPROM, enable the ACPI Magic Packet filter
7283 */
7284 if (adapter->flags & FLAG_APME_IN_WUC) {
7285 /* APME bit in EEPROM is mapped to WUC.APME */
7286 eeprom_data = er32(WUC);
7287 eeprom_apme_mask = E1000_WUC_APME;
7288 if ((hw->mac.type > e1000_ich10lan) &&
7289 (eeprom_data & E1000_WUC_PHY_WAKE))
7290 adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
7291 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
7292 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
7293 (adapter->hw.bus.func == 1))
7294 ret_val = e1000_read_nvm(&adapter->hw,
7295 NVM_INIT_CONTROL3_PORT_B,
7296 1, &eeprom_data);
7297 else
7298 ret_val = e1000_read_nvm(&adapter->hw,
7299 NVM_INIT_CONTROL3_PORT_A,
7300 1, &eeprom_data);
7301 }
7302
7303 /* fetch WoL from EEPROM */
7304 if (ret_val)
7305 e_dbg("NVM read error getting WoL initial values: %d\n", ret_val);
7306 else if (eeprom_data & eeprom_apme_mask)
7307 adapter->eeprom_wol |= E1000_WUFC_MAG;
7308
7309 /* now that we have the eeprom settings, apply the special cases
7310 * where the eeprom may be wrong or the board simply won't support
7311 * wake on lan on a particular port
7312 */
7313 if (!(adapter->flags & FLAG_HAS_WOL))
7314 adapter->eeprom_wol = 0;
7315
7316 /* initialize the wol settings based on the eeprom settings */
7317 adapter->wol = adapter->eeprom_wol;
7318
7319 /* make sure adapter isn't asleep if manageability is enabled */
7320 if (adapter->wol || (adapter->flags & FLAG_MNG_PT_ENABLED) ||
7321 (hw->mac.ops.check_mng_mode(hw)))
7322 device_wakeup_enable(&pdev->dev);
7323
7324 /* save off EEPROM version number */
7325 ret_val = e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
7326
7327 if (ret_val) {
7328 e_dbg("NVM read error getting EEPROM version: %d\n", ret_val);
7329 adapter->eeprom_vers = 0;
7330 }
7331
7332 /* init PTP hardware clock */
7333 e1000e_ptp_init(adapter);
7334
7335 /* reset the hardware with the new settings */
7336 e1000e_reset(adapter);
7337
7338 /* If the controller has AMT, do not set DRV_LOAD until the interface
7339 * is up. For all other cases, let the f/w know that the h/w is now
7340 * under the control of the driver.
7341 */
7342 if (!(adapter->flags & FLAG_HAS_AMT))
7343 e1000e_get_hw_control(adapter);
7344
7345 strlcpy(netdev->name, "eth%d", sizeof(netdev->name));
7346 err = register_netdev(netdev);
7347 if (err)
7348 goto err_register;
7349
7350 /* carrier off reporting is important to ethtool even BEFORE open */
7351 netif_carrier_off(netdev);
7352
7353 e1000_print_device_info(adapter);
7354
7355 if (pci_dev_run_wake(pdev))
7356 pm_runtime_put_noidle(&pdev->dev);
7357
7358 return 0;
7359
7360err_register:
7361 if (!(adapter->flags & FLAG_HAS_AMT))
7362 e1000e_release_hw_control(adapter);
7363err_eeprom:
7364 if (hw->phy.ops.check_reset_block && !hw->phy.ops.check_reset_block(hw))
7365 e1000_phy_hw_reset(&adapter->hw);
7366err_hw_init:
7367 kfree(adapter->tx_ring);
7368 kfree(adapter->rx_ring);
7369err_sw_init:
7370 if ((adapter->hw.flash_address) && (hw->mac.type < e1000_pch_spt))
7371 iounmap(adapter->hw.flash_address);
7372 e1000e_reset_interrupt_capability(adapter);
7373err_flashmap:
7374 iounmap(adapter->hw.hw_addr);
7375err_ioremap:
7376 free_netdev(netdev);
7377err_alloc_etherdev:
7378 pci_release_mem_regions(pdev);
7379err_pci_reg:
7380err_dma:
7381 pci_disable_device(pdev);
7382 return err;
7383}
7384
7385/**
7386 * e1000_remove - Device Removal Routine
7387 * @pdev: PCI device information struct
7388 *
7389 * e1000_remove is called by the PCI subsystem to alert the driver
7390 * that it should release a PCI device. The could be caused by a
7391 * Hot-Plug event, or because the driver is going to be removed from
7392 * memory.
7393 **/
7394static void e1000_remove(struct pci_dev *pdev)
7395{
7396 struct net_device *netdev = pci_get_drvdata(pdev);
7397 struct e1000_adapter *adapter = netdev_priv(netdev);
7398 bool down = test_bit(__E1000_DOWN, &adapter->state);
7399
7400 e1000e_ptp_remove(adapter);
7401
7402 /* The timers may be rescheduled, so explicitly disable them
7403 * from being rescheduled.
7404 */
7405 if (!down)
7406 set_bit(__E1000_DOWN, &adapter->state);
7407 del_timer_sync(&adapter->watchdog_timer);
7408 del_timer_sync(&adapter->phy_info_timer);
7409
7410 cancel_work_sync(&adapter->reset_task);
7411 cancel_work_sync(&adapter->watchdog_task);
7412 cancel_work_sync(&adapter->downshift_task);
7413 cancel_work_sync(&adapter->update_phy_task);
7414 cancel_work_sync(&adapter->print_hang_task);
7415
7416 if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
7417 cancel_work_sync(&adapter->tx_hwtstamp_work);
7418 if (adapter->tx_hwtstamp_skb) {
7419 dev_consume_skb_any(adapter->tx_hwtstamp_skb);
7420 adapter->tx_hwtstamp_skb = NULL;
7421 }
7422 }
7423
7424 /* Don't lie to e1000_close() down the road. */
7425 if (!down)
7426 clear_bit(__E1000_DOWN, &adapter->state);
7427 unregister_netdev(netdev);
7428
7429 if (pci_dev_run_wake(pdev))
7430 pm_runtime_get_noresume(&pdev->dev);
7431
7432 /* Release control of h/w to f/w. If f/w is AMT enabled, this
7433 * would have already happened in close and is redundant.
7434 */
7435 e1000e_release_hw_control(adapter);
7436
7437 e1000e_reset_interrupt_capability(adapter);
7438 kfree(adapter->tx_ring);
7439 kfree(adapter->rx_ring);
7440
7441 iounmap(adapter->hw.hw_addr);
7442 if ((adapter->hw.flash_address) &&
7443 (adapter->hw.mac.type < e1000_pch_spt))
7444 iounmap(adapter->hw.flash_address);
7445 pci_release_mem_regions(pdev);
7446
7447 free_netdev(netdev);
7448
7449 /* AER disable */
7450 pci_disable_pcie_error_reporting(pdev);
7451
7452 pci_disable_device(pdev);
7453}
7454
7455/* PCI Error Recovery (ERS) */
7456static const struct pci_error_handlers e1000_err_handler = {
7457 .error_detected = e1000_io_error_detected,
7458 .slot_reset = e1000_io_slot_reset,
7459 .resume = e1000_io_resume,
7460};
7461
7462static const struct pci_device_id e1000_pci_tbl[] = {
7463 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
7464 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
7465 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
7466 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP),
7467 board_82571 },
7468 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
7469 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
7470 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
7471 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
7472 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
7473
7474 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
7475 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
7476 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
7477 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
7478
7479 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
7480 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
7481 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
7482
7483 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
7484 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
7485 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
7486
7487 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
7488 board_80003es2lan },
7489 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
7490 board_80003es2lan },
7491 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
7492 board_80003es2lan },
7493 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
7494 board_80003es2lan },
7495
7496 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
7497 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
7498 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
7499 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
7500 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
7501 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
7502 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
7503 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
7504
7505 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
7506 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
7507 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
7508 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
7509 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
7510 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
7511 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
7512 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
7513 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
7514
7515 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
7516 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
7517 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
7518
7519 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
7520 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
7521 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
7522
7523 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
7524 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
7525 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
7526 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
7527
7528 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
7529 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
7530
7531 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_LM), board_pch_lpt },
7532 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_V), board_pch_lpt },
7533 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_LM), board_pch_lpt },
7534 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_V), board_pch_lpt },
7535 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM2), board_pch_lpt },
7536 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V2), board_pch_lpt },
7537 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM3), board_pch_lpt },
7538 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V3), board_pch_lpt },
7539 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM), board_pch_spt },
7540 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V), board_pch_spt },
7541 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM2), board_pch_spt },
7542 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V2), board_pch_spt },
7543 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LBG_I219_LM3), board_pch_spt },
7544 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM4), board_pch_spt },
7545 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V4), board_pch_spt },
7546 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM5), board_pch_spt },
7547 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V5), board_pch_spt },
7548 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_LM6), board_pch_cnp },
7549 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_V6), board_pch_cnp },
7550 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_LM7), board_pch_cnp },
7551 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_V7), board_pch_cnp },
7552 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ICP_I219_LM8), board_pch_cnp },
7553 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ICP_I219_V8), board_pch_cnp },
7554 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ICP_I219_LM9), board_pch_cnp },
7555 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ICP_I219_V9), board_pch_cnp },
7556
7557 { 0, 0, 0, 0, 0, 0, 0 } /* terminate list */
7558};
7559MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
7560
7561static const struct dev_pm_ops e1000_pm_ops = {
7562#ifdef CONFIG_PM_SLEEP
7563 .suspend = e1000e_pm_suspend,
7564 .resume = e1000e_pm_resume,
7565 .freeze = e1000e_pm_freeze,
7566 .thaw = e1000e_pm_thaw,
7567 .poweroff = e1000e_pm_suspend,
7568 .restore = e1000e_pm_resume,
7569#endif
7570 SET_RUNTIME_PM_OPS(e1000e_pm_runtime_suspend, e1000e_pm_runtime_resume,
7571 e1000e_pm_runtime_idle)
7572};
7573
7574/* PCI Device API Driver */
7575static struct pci_driver e1000_driver = {
7576 .name = e1000e_driver_name,
7577 .id_table = e1000_pci_tbl,
7578 .probe = e1000_probe,
7579 .remove = e1000_remove,
7580 .driver = {
7581 .pm = &e1000_pm_ops,
7582 },
7583 .shutdown = e1000_shutdown,
7584 .err_handler = &e1000_err_handler
7585};
7586
7587/**
7588 * e1000_init_module - Driver Registration Routine
7589 *
7590 * e1000_init_module is the first routine called when the driver is
7591 * loaded. All it does is register with the PCI subsystem.
7592 **/
7593static int __init e1000_init_module(void)
7594{
7595 pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
7596 e1000e_driver_version);
7597 pr_info("Copyright(c) 1999 - 2015 Intel Corporation.\n");
7598
7599 return pci_register_driver(&e1000_driver);
7600}
7601module_init(e1000_init_module);
7602
7603/**
7604 * e1000_exit_module - Driver Exit Cleanup Routine
7605 *
7606 * e1000_exit_module is called just before the driver is removed
7607 * from memory.
7608 **/
7609static void __exit e1000_exit_module(void)
7610{
7611 pci_unregister_driver(&e1000_driver);
7612}
7613module_exit(e1000_exit_module);
7614
7615MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
7616MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
7617MODULE_LICENSE("GPL");
7618MODULE_VERSION(DRV_VERSION);
7619
7620/* netdev.c */
1/*******************************************************************************
2
3 Intel PRO/1000 Linux driver
4 Copyright(c) 1999 - 2012 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 "2.0.0" DRV_EXTRAVERSION
60char e1000e_driver_name[] = "e1000e";
61const char e1000e_driver_version[] = DRV_VERSION;
62
63#define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
64static int debug = -1;
65module_param(debug, int, 0);
66MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
67
68static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state);
69
70static const struct e1000_info *e1000_info_tbl[] = {
71 [board_82571] = &e1000_82571_info,
72 [board_82572] = &e1000_82572_info,
73 [board_82573] = &e1000_82573_info,
74 [board_82574] = &e1000_82574_info,
75 [board_82583] = &e1000_82583_info,
76 [board_80003es2lan] = &e1000_es2_info,
77 [board_ich8lan] = &e1000_ich8_info,
78 [board_ich9lan] = &e1000_ich9_info,
79 [board_ich10lan] = &e1000_ich10_info,
80 [board_pchlan] = &e1000_pch_info,
81 [board_pch2lan] = &e1000_pch2_info,
82 [board_pch_lpt] = &e1000_pch_lpt_info,
83};
84
85struct e1000_reg_info {
86 u32 ofs;
87 char *name;
88};
89
90#define E1000_RDFH 0x02410 /* Rx Data FIFO Head - RW */
91#define E1000_RDFT 0x02418 /* Rx Data FIFO Tail - RW */
92#define E1000_RDFHS 0x02420 /* Rx Data FIFO Head Saved - RW */
93#define E1000_RDFTS 0x02428 /* Rx Data FIFO Tail Saved - RW */
94#define E1000_RDFPC 0x02430 /* Rx Data FIFO Packet Count - RW */
95
96#define E1000_TDFH 0x03410 /* Tx Data FIFO Head - RW */
97#define E1000_TDFT 0x03418 /* Tx Data FIFO Tail - RW */
98#define E1000_TDFHS 0x03420 /* Tx Data FIFO Head Saved - RW */
99#define E1000_TDFTS 0x03428 /* Tx Data FIFO Tail Saved - RW */
100#define E1000_TDFPC 0x03430 /* Tx Data FIFO Packet Count - RW */
101
102static const struct e1000_reg_info e1000_reg_info_tbl[] = {
103
104 /* General Registers */
105 {E1000_CTRL, "CTRL"},
106 {E1000_STATUS, "STATUS"},
107 {E1000_CTRL_EXT, "CTRL_EXT"},
108
109 /* Interrupt Registers */
110 {E1000_ICR, "ICR"},
111
112 /* Rx Registers */
113 {E1000_RCTL, "RCTL"},
114 {E1000_RDLEN(0), "RDLEN"},
115 {E1000_RDH(0), "RDH"},
116 {E1000_RDT(0), "RDT"},
117 {E1000_RDTR, "RDTR"},
118 {E1000_RXDCTL(0), "RXDCTL"},
119 {E1000_ERT, "ERT"},
120 {E1000_RDBAL(0), "RDBAL"},
121 {E1000_RDBAH(0), "RDBAH"},
122 {E1000_RDFH, "RDFH"},
123 {E1000_RDFT, "RDFT"},
124 {E1000_RDFHS, "RDFHS"},
125 {E1000_RDFTS, "RDFTS"},
126 {E1000_RDFPC, "RDFPC"},
127
128 /* Tx Registers */
129 {E1000_TCTL, "TCTL"},
130 {E1000_TDBAL(0), "TDBAL"},
131 {E1000_TDBAH(0), "TDBAH"},
132 {E1000_TDLEN(0), "TDLEN"},
133 {E1000_TDH(0), "TDH"},
134 {E1000_TDT(0), "TDT"},
135 {E1000_TIDV, "TIDV"},
136 {E1000_TXDCTL(0), "TXDCTL"},
137 {E1000_TADV, "TADV"},
138 {E1000_TARC(0), "TARC"},
139 {E1000_TDFH, "TDFH"},
140 {E1000_TDFT, "TDFT"},
141 {E1000_TDFHS, "TDFHS"},
142 {E1000_TDFTS, "TDFTS"},
143 {E1000_TDFPC, "TDFPC"},
144
145 /* List Terminator */
146 {0, NULL}
147};
148
149/*
150 * e1000_regdump - register printout routine
151 */
152static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
153{
154 int n = 0;
155 char rname[16];
156 u32 regs[8];
157
158 switch (reginfo->ofs) {
159 case E1000_RXDCTL(0):
160 for (n = 0; n < 2; n++)
161 regs[n] = __er32(hw, E1000_RXDCTL(n));
162 break;
163 case E1000_TXDCTL(0):
164 for (n = 0; n < 2; n++)
165 regs[n] = __er32(hw, E1000_TXDCTL(n));
166 break;
167 case E1000_TARC(0):
168 for (n = 0; n < 2; n++)
169 regs[n] = __er32(hw, E1000_TARC(n));
170 break;
171 default:
172 pr_info("%-15s %08x\n",
173 reginfo->name, __er32(hw, reginfo->ofs));
174 return;
175 }
176
177 snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
178 pr_info("%-15s %08x %08x\n", rname, regs[0], regs[1]);
179}
180
181/*
182 * e1000e_dump - Print registers, Tx-ring and Rx-ring
183 */
184static void e1000e_dump(struct e1000_adapter *adapter)
185{
186 struct net_device *netdev = adapter->netdev;
187 struct e1000_hw *hw = &adapter->hw;
188 struct e1000_reg_info *reginfo;
189 struct e1000_ring *tx_ring = adapter->tx_ring;
190 struct e1000_tx_desc *tx_desc;
191 struct my_u0 {
192 __le64 a;
193 __le64 b;
194 } *u0;
195 struct e1000_buffer *buffer_info;
196 struct e1000_ring *rx_ring = adapter->rx_ring;
197 union e1000_rx_desc_packet_split *rx_desc_ps;
198 union e1000_rx_desc_extended *rx_desc;
199 struct my_u1 {
200 __le64 a;
201 __le64 b;
202 __le64 c;
203 __le64 d;
204 } *u1;
205 u32 staterr;
206 int i = 0;
207
208 if (!netif_msg_hw(adapter))
209 return;
210
211 /* Print netdevice Info */
212 if (netdev) {
213 dev_info(&adapter->pdev->dev, "Net device Info\n");
214 pr_info("Device Name state trans_start last_rx\n");
215 pr_info("%-15s %016lX %016lX %016lX\n",
216 netdev->name, netdev->state, netdev->trans_start,
217 netdev->last_rx);
218 }
219
220 /* Print Registers */
221 dev_info(&adapter->pdev->dev, "Register Dump\n");
222 pr_info(" Register Name Value\n");
223 for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
224 reginfo->name; reginfo++) {
225 e1000_regdump(hw, reginfo);
226 }
227
228 /* Print Tx Ring Summary */
229 if (!netdev || !netif_running(netdev))
230 return;
231
232 dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
233 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
234 buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
235 pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
236 0, tx_ring->next_to_use, tx_ring->next_to_clean,
237 (unsigned long long)buffer_info->dma,
238 buffer_info->length,
239 buffer_info->next_to_watch,
240 (unsigned long long)buffer_info->time_stamp);
241
242 /* Print Tx Ring */
243 if (!netif_msg_tx_done(adapter))
244 goto rx_ring_summary;
245
246 dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
247
248 /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
249 *
250 * Legacy Transmit Descriptor
251 * +--------------------------------------------------------------+
252 * 0 | Buffer Address [63:0] (Reserved on Write Back) |
253 * +--------------------------------------------------------------+
254 * 8 | Special | CSS | Status | CMD | CSO | Length |
255 * +--------------------------------------------------------------+
256 * 63 48 47 36 35 32 31 24 23 16 15 0
257 *
258 * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
259 * 63 48 47 40 39 32 31 16 15 8 7 0
260 * +----------------------------------------------------------------+
261 * 0 | TUCSE | TUCS0 | TUCSS | IPCSE | IPCS0 | IPCSS |
262 * +----------------------------------------------------------------+
263 * 8 | MSS | HDRLEN | RSV | STA | TUCMD | DTYP | PAYLEN |
264 * +----------------------------------------------------------------+
265 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
266 *
267 * Extended Data Descriptor (DTYP=0x1)
268 * +----------------------------------------------------------------+
269 * 0 | Buffer Address [63:0] |
270 * +----------------------------------------------------------------+
271 * 8 | VLAN tag | POPTS | Rsvd | Status | Command | DTYP | DTALEN |
272 * +----------------------------------------------------------------+
273 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
274 */
275 pr_info("Tl[desc] [address 63:0 ] [SpeCssSCmCsLen] [bi->dma ] leng ntw timestamp bi->skb <-- Legacy format\n");
276 pr_info("Tc[desc] [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma ] leng ntw timestamp bi->skb <-- Ext Context format\n");
277 pr_info("Td[desc] [address 63:0 ] [VlaPoRSCm1Dlen] [bi->dma ] leng ntw timestamp bi->skb <-- Ext Data format\n");
278 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
279 const char *next_desc;
280 tx_desc = E1000_TX_DESC(*tx_ring, i);
281 buffer_info = &tx_ring->buffer_info[i];
282 u0 = (struct my_u0 *)tx_desc;
283 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
284 next_desc = " NTC/U";
285 else if (i == tx_ring->next_to_use)
286 next_desc = " NTU";
287 else if (i == tx_ring->next_to_clean)
288 next_desc = " NTC";
289 else
290 next_desc = "";
291 pr_info("T%c[0x%03X] %016llX %016llX %016llX %04X %3X %016llX %p%s\n",
292 (!(le64_to_cpu(u0->b) & (1 << 29)) ? 'l' :
293 ((le64_to_cpu(u0->b) & (1 << 20)) ? 'd' : 'c')),
294 i,
295 (unsigned long long)le64_to_cpu(u0->a),
296 (unsigned long long)le64_to_cpu(u0->b),
297 (unsigned long long)buffer_info->dma,
298 buffer_info->length, buffer_info->next_to_watch,
299 (unsigned long long)buffer_info->time_stamp,
300 buffer_info->skb, next_desc);
301
302 if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
303 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
304 16, 1, phys_to_virt(buffer_info->dma),
305 buffer_info->length, true);
306 }
307
308 /* Print Rx Ring Summary */
309rx_ring_summary:
310 dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
311 pr_info("Queue [NTU] [NTC]\n");
312 pr_info(" %5d %5X %5X\n",
313 0, rx_ring->next_to_use, rx_ring->next_to_clean);
314
315 /* Print Rx Ring */
316 if (!netif_msg_rx_status(adapter))
317 return;
318
319 dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
320 switch (adapter->rx_ps_pages) {
321 case 1:
322 case 2:
323 case 3:
324 /* [Extended] Packet Split Receive Descriptor Format
325 *
326 * +-----------------------------------------------------+
327 * 0 | Buffer Address 0 [63:0] |
328 * +-----------------------------------------------------+
329 * 8 | Buffer Address 1 [63:0] |
330 * +-----------------------------------------------------+
331 * 16 | Buffer Address 2 [63:0] |
332 * +-----------------------------------------------------+
333 * 24 | Buffer Address 3 [63:0] |
334 * +-----------------------------------------------------+
335 */
336 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");
337 /* [Extended] Receive Descriptor (Write-Back) Format
338 *
339 * 63 48 47 32 31 13 12 8 7 4 3 0
340 * +------------------------------------------------------+
341 * 0 | Packet | IP | Rsvd | MRQ | Rsvd | MRQ RSS |
342 * | Checksum | Ident | | Queue | | Type |
343 * +------------------------------------------------------+
344 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
345 * +------------------------------------------------------+
346 * 63 48 47 32 31 20 19 0
347 */
348 pr_info("RWB[desc] [ck ipid mrqhsh] [vl l0 ee es] [ l3 l2 l1 hs] [reserved ] ---------------- [bi->skb] <-- Ext Rx Write-Back format\n");
349 for (i = 0; i < rx_ring->count; i++) {
350 const char *next_desc;
351 buffer_info = &rx_ring->buffer_info[i];
352 rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
353 u1 = (struct my_u1 *)rx_desc_ps;
354 staterr =
355 le32_to_cpu(rx_desc_ps->wb.middle.status_error);
356
357 if (i == rx_ring->next_to_use)
358 next_desc = " NTU";
359 else if (i == rx_ring->next_to_clean)
360 next_desc = " NTC";
361 else
362 next_desc = "";
363
364 if (staterr & E1000_RXD_STAT_DD) {
365 /* Descriptor Done */
366 pr_info("%s[0x%03X] %016llX %016llX %016llX %016llX ---------------- %p%s\n",
367 "RWB", i,
368 (unsigned long long)le64_to_cpu(u1->a),
369 (unsigned long long)le64_to_cpu(u1->b),
370 (unsigned long long)le64_to_cpu(u1->c),
371 (unsigned long long)le64_to_cpu(u1->d),
372 buffer_info->skb, next_desc);
373 } else {
374 pr_info("%s[0x%03X] %016llX %016llX %016llX %016llX %016llX %p%s\n",
375 "R ", i,
376 (unsigned long long)le64_to_cpu(u1->a),
377 (unsigned long long)le64_to_cpu(u1->b),
378 (unsigned long long)le64_to_cpu(u1->c),
379 (unsigned long long)le64_to_cpu(u1->d),
380 (unsigned long long)buffer_info->dma,
381 buffer_info->skb, next_desc);
382
383 if (netif_msg_pktdata(adapter))
384 print_hex_dump(KERN_INFO, "",
385 DUMP_PREFIX_ADDRESS, 16, 1,
386 phys_to_virt(buffer_info->dma),
387 adapter->rx_ps_bsize0, true);
388 }
389 }
390 break;
391 default:
392 case 0:
393 /* Extended Receive Descriptor (Read) Format
394 *
395 * +-----------------------------------------------------+
396 * 0 | Buffer Address [63:0] |
397 * +-----------------------------------------------------+
398 * 8 | Reserved |
399 * +-----------------------------------------------------+
400 */
401 pr_info("R [desc] [buf addr 63:0 ] [reserved 63:0 ] [bi->dma ] [bi->skb] <-- Ext (Read) format\n");
402 /* Extended Receive Descriptor (Write-Back) Format
403 *
404 * 63 48 47 32 31 24 23 4 3 0
405 * +------------------------------------------------------+
406 * | RSS Hash | | | |
407 * 0 +-------------------+ Rsvd | Reserved | MRQ RSS |
408 * | Packet | IP | | | Type |
409 * | Checksum | Ident | | | |
410 * +------------------------------------------------------+
411 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
412 * +------------------------------------------------------+
413 * 63 48 47 32 31 20 19 0
414 */
415 pr_info("RWB[desc] [cs ipid mrq] [vt ln xe xs] [bi->skb] <-- Ext (Write-Back) format\n");
416
417 for (i = 0; i < rx_ring->count; i++) {
418 const char *next_desc;
419
420 buffer_info = &rx_ring->buffer_info[i];
421 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
422 u1 = (struct my_u1 *)rx_desc;
423 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
424
425 if (i == rx_ring->next_to_use)
426 next_desc = " NTU";
427 else if (i == rx_ring->next_to_clean)
428 next_desc = " NTC";
429 else
430 next_desc = "";
431
432 if (staterr & E1000_RXD_STAT_DD) {
433 /* Descriptor Done */
434 pr_info("%s[0x%03X] %016llX %016llX ---------------- %p%s\n",
435 "RWB", i,
436 (unsigned long long)le64_to_cpu(u1->a),
437 (unsigned long long)le64_to_cpu(u1->b),
438 buffer_info->skb, next_desc);
439 } else {
440 pr_info("%s[0x%03X] %016llX %016llX %016llX %p%s\n",
441 "R ", i,
442 (unsigned long long)le64_to_cpu(u1->a),
443 (unsigned long long)le64_to_cpu(u1->b),
444 (unsigned long long)buffer_info->dma,
445 buffer_info->skb, next_desc);
446
447 if (netif_msg_pktdata(adapter))
448 print_hex_dump(KERN_INFO, "",
449 DUMP_PREFIX_ADDRESS, 16,
450 1,
451 phys_to_virt
452 (buffer_info->dma),
453 adapter->rx_buffer_len,
454 true);
455 }
456 }
457 }
458}
459
460/**
461 * e1000_desc_unused - calculate if we have unused descriptors
462 **/
463static int e1000_desc_unused(struct e1000_ring *ring)
464{
465 if (ring->next_to_clean > ring->next_to_use)
466 return ring->next_to_clean - ring->next_to_use - 1;
467
468 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
469}
470
471/**
472 * e1000_receive_skb - helper function to handle Rx indications
473 * @adapter: board private structure
474 * @status: descriptor status field as written by hardware
475 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
476 * @skb: pointer to sk_buff to be indicated to stack
477 **/
478static void e1000_receive_skb(struct e1000_adapter *adapter,
479 struct net_device *netdev, struct sk_buff *skb,
480 u8 status, __le16 vlan)
481{
482 u16 tag = le16_to_cpu(vlan);
483 skb->protocol = eth_type_trans(skb, netdev);
484
485 if (status & E1000_RXD_STAT_VP)
486 __vlan_hwaccel_put_tag(skb, tag);
487
488 napi_gro_receive(&adapter->napi, skb);
489}
490
491/**
492 * e1000_rx_checksum - Receive Checksum Offload
493 * @adapter: board private structure
494 * @status_err: receive descriptor status and error fields
495 * @csum: receive descriptor csum field
496 * @sk_buff: socket buffer with received data
497 **/
498static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
499 struct sk_buff *skb)
500{
501 u16 status = (u16)status_err;
502 u8 errors = (u8)(status_err >> 24);
503
504 skb_checksum_none_assert(skb);
505
506 /* Rx checksum disabled */
507 if (!(adapter->netdev->features & NETIF_F_RXCSUM))
508 return;
509
510 /* Ignore Checksum bit is set */
511 if (status & E1000_RXD_STAT_IXSM)
512 return;
513
514 /* TCP/UDP checksum error bit or IP checksum error bit is set */
515 if (errors & (E1000_RXD_ERR_TCPE | E1000_RXD_ERR_IPE)) {
516 /* let the stack verify checksum errors */
517 adapter->hw_csum_err++;
518 return;
519 }
520
521 /* TCP/UDP Checksum has not been calculated */
522 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
523 return;
524
525 /* It must be a TCP or UDP packet with a valid checksum */
526 skb->ip_summed = CHECKSUM_UNNECESSARY;
527 adapter->hw_csum_good++;
528}
529
530static void e1000e_update_rdt_wa(struct e1000_ring *rx_ring, unsigned int i)
531{
532 struct e1000_adapter *adapter = rx_ring->adapter;
533 struct e1000_hw *hw = &adapter->hw;
534 s32 ret_val = __ew32_prepare(hw);
535
536 writel(i, rx_ring->tail);
537
538 if (unlikely(!ret_val && (i != readl(rx_ring->tail)))) {
539 u32 rctl = er32(RCTL);
540 ew32(RCTL, rctl & ~E1000_RCTL_EN);
541 e_err("ME firmware caused invalid RDT - resetting\n");
542 schedule_work(&adapter->reset_task);
543 }
544}
545
546static void e1000e_update_tdt_wa(struct e1000_ring *tx_ring, unsigned int i)
547{
548 struct e1000_adapter *adapter = tx_ring->adapter;
549 struct e1000_hw *hw = &adapter->hw;
550 s32 ret_val = __ew32_prepare(hw);
551
552 writel(i, tx_ring->tail);
553
554 if (unlikely(!ret_val && (i != readl(tx_ring->tail)))) {
555 u32 tctl = er32(TCTL);
556 ew32(TCTL, tctl & ~E1000_TCTL_EN);
557 e_err("ME firmware caused invalid TDT - resetting\n");
558 schedule_work(&adapter->reset_task);
559 }
560}
561
562/**
563 * e1000_alloc_rx_buffers - Replace used receive buffers
564 * @rx_ring: Rx descriptor ring
565 **/
566static void e1000_alloc_rx_buffers(struct e1000_ring *rx_ring,
567 int cleaned_count, gfp_t gfp)
568{
569 struct e1000_adapter *adapter = rx_ring->adapter;
570 struct net_device *netdev = adapter->netdev;
571 struct pci_dev *pdev = adapter->pdev;
572 union e1000_rx_desc_extended *rx_desc;
573 struct e1000_buffer *buffer_info;
574 struct sk_buff *skb;
575 unsigned int i;
576 unsigned int bufsz = adapter->rx_buffer_len;
577
578 i = rx_ring->next_to_use;
579 buffer_info = &rx_ring->buffer_info[i];
580
581 while (cleaned_count--) {
582 skb = buffer_info->skb;
583 if (skb) {
584 skb_trim(skb, 0);
585 goto map_skb;
586 }
587
588 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
589 if (!skb) {
590 /* Better luck next round */
591 adapter->alloc_rx_buff_failed++;
592 break;
593 }
594
595 buffer_info->skb = skb;
596map_skb:
597 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
598 adapter->rx_buffer_len,
599 DMA_FROM_DEVICE);
600 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
601 dev_err(&pdev->dev, "Rx DMA map failed\n");
602 adapter->rx_dma_failed++;
603 break;
604 }
605
606 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
607 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
608
609 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
610 /*
611 * Force memory writes to complete before letting h/w
612 * know there are new descriptors to fetch. (Only
613 * applicable for weak-ordered memory model archs,
614 * such as IA-64).
615 */
616 wmb();
617 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
618 e1000e_update_rdt_wa(rx_ring, i);
619 else
620 writel(i, rx_ring->tail);
621 }
622 i++;
623 if (i == rx_ring->count)
624 i = 0;
625 buffer_info = &rx_ring->buffer_info[i];
626 }
627
628 rx_ring->next_to_use = i;
629}
630
631/**
632 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
633 * @rx_ring: Rx descriptor ring
634 **/
635static void e1000_alloc_rx_buffers_ps(struct e1000_ring *rx_ring,
636 int cleaned_count, gfp_t gfp)
637{
638 struct e1000_adapter *adapter = rx_ring->adapter;
639 struct net_device *netdev = adapter->netdev;
640 struct pci_dev *pdev = adapter->pdev;
641 union e1000_rx_desc_packet_split *rx_desc;
642 struct e1000_buffer *buffer_info;
643 struct e1000_ps_page *ps_page;
644 struct sk_buff *skb;
645 unsigned int i, j;
646
647 i = rx_ring->next_to_use;
648 buffer_info = &rx_ring->buffer_info[i];
649
650 while (cleaned_count--) {
651 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
652
653 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
654 ps_page = &buffer_info->ps_pages[j];
655 if (j >= adapter->rx_ps_pages) {
656 /* all unused desc entries get hw null ptr */
657 rx_desc->read.buffer_addr[j + 1] =
658 ~cpu_to_le64(0);
659 continue;
660 }
661 if (!ps_page->page) {
662 ps_page->page = alloc_page(gfp);
663 if (!ps_page->page) {
664 adapter->alloc_rx_buff_failed++;
665 goto no_buffers;
666 }
667 ps_page->dma = dma_map_page(&pdev->dev,
668 ps_page->page,
669 0, PAGE_SIZE,
670 DMA_FROM_DEVICE);
671 if (dma_mapping_error(&pdev->dev,
672 ps_page->dma)) {
673 dev_err(&adapter->pdev->dev,
674 "Rx DMA page map failed\n");
675 adapter->rx_dma_failed++;
676 goto no_buffers;
677 }
678 }
679 /*
680 * Refresh the desc even if buffer_addrs
681 * didn't change because each write-back
682 * erases this info.
683 */
684 rx_desc->read.buffer_addr[j + 1] =
685 cpu_to_le64(ps_page->dma);
686 }
687
688 skb = __netdev_alloc_skb_ip_align(netdev,
689 adapter->rx_ps_bsize0,
690 gfp);
691
692 if (!skb) {
693 adapter->alloc_rx_buff_failed++;
694 break;
695 }
696
697 buffer_info->skb = skb;
698 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
699 adapter->rx_ps_bsize0,
700 DMA_FROM_DEVICE);
701 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
702 dev_err(&pdev->dev, "Rx DMA map failed\n");
703 adapter->rx_dma_failed++;
704 /* cleanup skb */
705 dev_kfree_skb_any(skb);
706 buffer_info->skb = NULL;
707 break;
708 }
709
710 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
711
712 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
713 /*
714 * Force memory writes to complete before letting h/w
715 * know there are new descriptors to fetch. (Only
716 * applicable for weak-ordered memory model archs,
717 * such as IA-64).
718 */
719 wmb();
720 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
721 e1000e_update_rdt_wa(rx_ring, i << 1);
722 else
723 writel(i << 1, rx_ring->tail);
724 }
725
726 i++;
727 if (i == rx_ring->count)
728 i = 0;
729 buffer_info = &rx_ring->buffer_info[i];
730 }
731
732no_buffers:
733 rx_ring->next_to_use = i;
734}
735
736/**
737 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
738 * @rx_ring: Rx descriptor ring
739 * @cleaned_count: number of buffers to allocate this pass
740 **/
741
742static void e1000_alloc_jumbo_rx_buffers(struct e1000_ring *rx_ring,
743 int cleaned_count, gfp_t gfp)
744{
745 struct e1000_adapter *adapter = rx_ring->adapter;
746 struct net_device *netdev = adapter->netdev;
747 struct pci_dev *pdev = adapter->pdev;
748 union e1000_rx_desc_extended *rx_desc;
749 struct e1000_buffer *buffer_info;
750 struct sk_buff *skb;
751 unsigned int i;
752 unsigned int bufsz = 256 - 16 /* for skb_reserve */;
753
754 i = rx_ring->next_to_use;
755 buffer_info = &rx_ring->buffer_info[i];
756
757 while (cleaned_count--) {
758 skb = buffer_info->skb;
759 if (skb) {
760 skb_trim(skb, 0);
761 goto check_page;
762 }
763
764 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
765 if (unlikely(!skb)) {
766 /* Better luck next round */
767 adapter->alloc_rx_buff_failed++;
768 break;
769 }
770
771 buffer_info->skb = skb;
772check_page:
773 /* allocate a new page if necessary */
774 if (!buffer_info->page) {
775 buffer_info->page = alloc_page(gfp);
776 if (unlikely(!buffer_info->page)) {
777 adapter->alloc_rx_buff_failed++;
778 break;
779 }
780 }
781
782 if (!buffer_info->dma)
783 buffer_info->dma = dma_map_page(&pdev->dev,
784 buffer_info->page, 0,
785 PAGE_SIZE,
786 DMA_FROM_DEVICE);
787
788 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
789 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
790
791 if (unlikely(++i == rx_ring->count))
792 i = 0;
793 buffer_info = &rx_ring->buffer_info[i];
794 }
795
796 if (likely(rx_ring->next_to_use != i)) {
797 rx_ring->next_to_use = i;
798 if (unlikely(i-- == 0))
799 i = (rx_ring->count - 1);
800
801 /* Force memory writes to complete before letting h/w
802 * know there are new descriptors to fetch. (Only
803 * applicable for weak-ordered memory model archs,
804 * such as IA-64). */
805 wmb();
806 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
807 e1000e_update_rdt_wa(rx_ring, i);
808 else
809 writel(i, rx_ring->tail);
810 }
811}
812
813static inline void e1000_rx_hash(struct net_device *netdev, __le32 rss,
814 struct sk_buff *skb)
815{
816 if (netdev->features & NETIF_F_RXHASH)
817 skb->rxhash = le32_to_cpu(rss);
818}
819
820/**
821 * e1000_clean_rx_irq - Send received data up the network stack
822 * @rx_ring: Rx descriptor ring
823 *
824 * the return value indicates whether actual cleaning was done, there
825 * is no guarantee that everything was cleaned
826 **/
827static bool e1000_clean_rx_irq(struct e1000_ring *rx_ring, int *work_done,
828 int work_to_do)
829{
830 struct e1000_adapter *adapter = rx_ring->adapter;
831 struct net_device *netdev = adapter->netdev;
832 struct pci_dev *pdev = adapter->pdev;
833 struct e1000_hw *hw = &adapter->hw;
834 union e1000_rx_desc_extended *rx_desc, *next_rxd;
835 struct e1000_buffer *buffer_info, *next_buffer;
836 u32 length, staterr;
837 unsigned int i;
838 int cleaned_count = 0;
839 bool cleaned = false;
840 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
841
842 i = rx_ring->next_to_clean;
843 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
844 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
845 buffer_info = &rx_ring->buffer_info[i];
846
847 while (staterr & E1000_RXD_STAT_DD) {
848 struct sk_buff *skb;
849
850 if (*work_done >= work_to_do)
851 break;
852 (*work_done)++;
853 rmb(); /* read descriptor and rx_buffer_info after status DD */
854
855 skb = buffer_info->skb;
856 buffer_info->skb = NULL;
857
858 prefetch(skb->data - NET_IP_ALIGN);
859
860 i++;
861 if (i == rx_ring->count)
862 i = 0;
863 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
864 prefetch(next_rxd);
865
866 next_buffer = &rx_ring->buffer_info[i];
867
868 cleaned = true;
869 cleaned_count++;
870 dma_unmap_single(&pdev->dev,
871 buffer_info->dma,
872 adapter->rx_buffer_len,
873 DMA_FROM_DEVICE);
874 buffer_info->dma = 0;
875
876 length = le16_to_cpu(rx_desc->wb.upper.length);
877
878 /*
879 * !EOP means multiple descriptors were used to store a single
880 * packet, if that's the case we need to toss it. In fact, we
881 * need to toss every packet with the EOP bit clear and the
882 * next frame that _does_ have the EOP bit set, as it is by
883 * definition only a frame fragment
884 */
885 if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
886 adapter->flags2 |= FLAG2_IS_DISCARDING;
887
888 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
889 /* All receives must fit into a single buffer */
890 e_dbg("Receive packet consumed multiple buffers\n");
891 /* recycle */
892 buffer_info->skb = skb;
893 if (staterr & E1000_RXD_STAT_EOP)
894 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
895 goto next_desc;
896 }
897
898 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
899 !(netdev->features & NETIF_F_RXALL))) {
900 /* recycle */
901 buffer_info->skb = skb;
902 goto next_desc;
903 }
904
905 /* adjust length to remove Ethernet CRC */
906 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
907 /* If configured to store CRC, don't subtract FCS,
908 * but keep the FCS bytes out of the total_rx_bytes
909 * counter
910 */
911 if (netdev->features & NETIF_F_RXFCS)
912 total_rx_bytes -= 4;
913 else
914 length -= 4;
915 }
916
917 total_rx_bytes += length;
918 total_rx_packets++;
919
920 /*
921 * code added for copybreak, this should improve
922 * performance for small packets with large amounts
923 * of reassembly being done in the stack
924 */
925 if (length < copybreak) {
926 struct sk_buff *new_skb =
927 netdev_alloc_skb_ip_align(netdev, length);
928 if (new_skb) {
929 skb_copy_to_linear_data_offset(new_skb,
930 -NET_IP_ALIGN,
931 (skb->data -
932 NET_IP_ALIGN),
933 (length +
934 NET_IP_ALIGN));
935 /* save the skb in buffer_info as good */
936 buffer_info->skb = skb;
937 skb = new_skb;
938 }
939 /* else just continue with the old one */
940 }
941 /* end copybreak code */
942 skb_put(skb, length);
943
944 /* Receive Checksum Offload */
945 e1000_rx_checksum(adapter, staterr, skb);
946
947 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
948
949 e1000_receive_skb(adapter, netdev, skb, staterr,
950 rx_desc->wb.upper.vlan);
951
952next_desc:
953 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
954
955 /* return some buffers to hardware, one at a time is too slow */
956 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
957 adapter->alloc_rx_buf(rx_ring, cleaned_count,
958 GFP_ATOMIC);
959 cleaned_count = 0;
960 }
961
962 /* use prefetched values */
963 rx_desc = next_rxd;
964 buffer_info = next_buffer;
965
966 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
967 }
968 rx_ring->next_to_clean = i;
969
970 cleaned_count = e1000_desc_unused(rx_ring);
971 if (cleaned_count)
972 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
973
974 adapter->total_rx_bytes += total_rx_bytes;
975 adapter->total_rx_packets += total_rx_packets;
976 return cleaned;
977}
978
979static void e1000_put_txbuf(struct e1000_ring *tx_ring,
980 struct e1000_buffer *buffer_info)
981{
982 struct e1000_adapter *adapter = tx_ring->adapter;
983
984 if (buffer_info->dma) {
985 if (buffer_info->mapped_as_page)
986 dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
987 buffer_info->length, DMA_TO_DEVICE);
988 else
989 dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
990 buffer_info->length, DMA_TO_DEVICE);
991 buffer_info->dma = 0;
992 }
993 if (buffer_info->skb) {
994 dev_kfree_skb_any(buffer_info->skb);
995 buffer_info->skb = NULL;
996 }
997 buffer_info->time_stamp = 0;
998}
999
1000static void e1000_print_hw_hang(struct work_struct *work)
1001{
1002 struct e1000_adapter *adapter = container_of(work,
1003 struct e1000_adapter,
1004 print_hang_task);
1005 struct net_device *netdev = adapter->netdev;
1006 struct e1000_ring *tx_ring = adapter->tx_ring;
1007 unsigned int i = tx_ring->next_to_clean;
1008 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1009 struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1010 struct e1000_hw *hw = &adapter->hw;
1011 u16 phy_status, phy_1000t_status, phy_ext_status;
1012 u16 pci_status;
1013
1014 if (test_bit(__E1000_DOWN, &adapter->state))
1015 return;
1016
1017 if (!adapter->tx_hang_recheck &&
1018 (adapter->flags2 & FLAG2_DMA_BURST)) {
1019 /*
1020 * May be block on write-back, flush and detect again
1021 * flush pending descriptor writebacks to memory
1022 */
1023 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1024 /* execute the writes immediately */
1025 e1e_flush();
1026 /*
1027 * Due to rare timing issues, write to TIDV again to ensure
1028 * the write is successful
1029 */
1030 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1031 /* execute the writes immediately */
1032 e1e_flush();
1033 adapter->tx_hang_recheck = true;
1034 return;
1035 }
1036 /* Real hang detected */
1037 adapter->tx_hang_recheck = false;
1038 netif_stop_queue(netdev);
1039
1040 e1e_rphy(hw, PHY_STATUS, &phy_status);
1041 e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
1042 e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
1043
1044 pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1045
1046 /* detected Hardware unit hang */
1047 e_err("Detected Hardware Unit Hang:\n"
1048 " TDH <%x>\n"
1049 " TDT <%x>\n"
1050 " next_to_use <%x>\n"
1051 " next_to_clean <%x>\n"
1052 "buffer_info[next_to_clean]:\n"
1053 " time_stamp <%lx>\n"
1054 " next_to_watch <%x>\n"
1055 " jiffies <%lx>\n"
1056 " next_to_watch.status <%x>\n"
1057 "MAC Status <%x>\n"
1058 "PHY Status <%x>\n"
1059 "PHY 1000BASE-T Status <%x>\n"
1060 "PHY Extended Status <%x>\n"
1061 "PCI Status <%x>\n",
1062 readl(tx_ring->head),
1063 readl(tx_ring->tail),
1064 tx_ring->next_to_use,
1065 tx_ring->next_to_clean,
1066 tx_ring->buffer_info[eop].time_stamp,
1067 eop,
1068 jiffies,
1069 eop_desc->upper.fields.status,
1070 er32(STATUS),
1071 phy_status,
1072 phy_1000t_status,
1073 phy_ext_status,
1074 pci_status);
1075
1076 /* Suggest workaround for known h/w issue */
1077 if ((hw->mac.type == e1000_pchlan) && (er32(CTRL) & E1000_CTRL_TFCE))
1078 e_err("Try turning off Tx pause (flow control) via ethtool\n");
1079}
1080
1081/**
1082 * e1000_clean_tx_irq - Reclaim resources after transmit completes
1083 * @tx_ring: Tx descriptor ring
1084 *
1085 * the return value indicates whether actual cleaning was done, there
1086 * is no guarantee that everything was cleaned
1087 **/
1088static bool e1000_clean_tx_irq(struct e1000_ring *tx_ring)
1089{
1090 struct e1000_adapter *adapter = tx_ring->adapter;
1091 struct net_device *netdev = adapter->netdev;
1092 struct e1000_hw *hw = &adapter->hw;
1093 struct e1000_tx_desc *tx_desc, *eop_desc;
1094 struct e1000_buffer *buffer_info;
1095 unsigned int i, eop;
1096 unsigned int count = 0;
1097 unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1098 unsigned int bytes_compl = 0, pkts_compl = 0;
1099
1100 i = tx_ring->next_to_clean;
1101 eop = tx_ring->buffer_info[i].next_to_watch;
1102 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1103
1104 while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1105 (count < tx_ring->count)) {
1106 bool cleaned = false;
1107 rmb(); /* read buffer_info after eop_desc */
1108 for (; !cleaned; count++) {
1109 tx_desc = E1000_TX_DESC(*tx_ring, i);
1110 buffer_info = &tx_ring->buffer_info[i];
1111 cleaned = (i == eop);
1112
1113 if (cleaned) {
1114 total_tx_packets += buffer_info->segs;
1115 total_tx_bytes += buffer_info->bytecount;
1116 if (buffer_info->skb) {
1117 bytes_compl += buffer_info->skb->len;
1118 pkts_compl++;
1119 }
1120 }
1121
1122 e1000_put_txbuf(tx_ring, buffer_info);
1123 tx_desc->upper.data = 0;
1124
1125 i++;
1126 if (i == tx_ring->count)
1127 i = 0;
1128 }
1129
1130 if (i == tx_ring->next_to_use)
1131 break;
1132 eop = tx_ring->buffer_info[i].next_to_watch;
1133 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1134 }
1135
1136 tx_ring->next_to_clean = i;
1137
1138 netdev_completed_queue(netdev, pkts_compl, bytes_compl);
1139
1140#define TX_WAKE_THRESHOLD 32
1141 if (count && netif_carrier_ok(netdev) &&
1142 e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1143 /* Make sure that anybody stopping the queue after this
1144 * sees the new next_to_clean.
1145 */
1146 smp_mb();
1147
1148 if (netif_queue_stopped(netdev) &&
1149 !(test_bit(__E1000_DOWN, &adapter->state))) {
1150 netif_wake_queue(netdev);
1151 ++adapter->restart_queue;
1152 }
1153 }
1154
1155 if (adapter->detect_tx_hung) {
1156 /*
1157 * Detect a transmit hang in hardware, this serializes the
1158 * check with the clearing of time_stamp and movement of i
1159 */
1160 adapter->detect_tx_hung = false;
1161 if (tx_ring->buffer_info[i].time_stamp &&
1162 time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1163 + (adapter->tx_timeout_factor * HZ)) &&
1164 !(er32(STATUS) & E1000_STATUS_TXOFF))
1165 schedule_work(&adapter->print_hang_task);
1166 else
1167 adapter->tx_hang_recheck = false;
1168 }
1169 adapter->total_tx_bytes += total_tx_bytes;
1170 adapter->total_tx_packets += total_tx_packets;
1171 return count < tx_ring->count;
1172}
1173
1174/**
1175 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1176 * @rx_ring: Rx descriptor ring
1177 *
1178 * the return value indicates whether actual cleaning was done, there
1179 * is no guarantee that everything was cleaned
1180 **/
1181static bool e1000_clean_rx_irq_ps(struct e1000_ring *rx_ring, int *work_done,
1182 int work_to_do)
1183{
1184 struct e1000_adapter *adapter = rx_ring->adapter;
1185 struct e1000_hw *hw = &adapter->hw;
1186 union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1187 struct net_device *netdev = adapter->netdev;
1188 struct pci_dev *pdev = adapter->pdev;
1189 struct e1000_buffer *buffer_info, *next_buffer;
1190 struct e1000_ps_page *ps_page;
1191 struct sk_buff *skb;
1192 unsigned int i, j;
1193 u32 length, staterr;
1194 int cleaned_count = 0;
1195 bool cleaned = false;
1196 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1197
1198 i = rx_ring->next_to_clean;
1199 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1200 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1201 buffer_info = &rx_ring->buffer_info[i];
1202
1203 while (staterr & E1000_RXD_STAT_DD) {
1204 if (*work_done >= work_to_do)
1205 break;
1206 (*work_done)++;
1207 skb = buffer_info->skb;
1208 rmb(); /* read descriptor and rx_buffer_info after status DD */
1209
1210 /* in the packet split case this is header only */
1211 prefetch(skb->data - NET_IP_ALIGN);
1212
1213 i++;
1214 if (i == rx_ring->count)
1215 i = 0;
1216 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1217 prefetch(next_rxd);
1218
1219 next_buffer = &rx_ring->buffer_info[i];
1220
1221 cleaned = true;
1222 cleaned_count++;
1223 dma_unmap_single(&pdev->dev, buffer_info->dma,
1224 adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1225 buffer_info->dma = 0;
1226
1227 /* see !EOP comment in other Rx routine */
1228 if (!(staterr & E1000_RXD_STAT_EOP))
1229 adapter->flags2 |= FLAG2_IS_DISCARDING;
1230
1231 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1232 e_dbg("Packet Split buffers didn't pick up the full packet\n");
1233 dev_kfree_skb_irq(skb);
1234 if (staterr & E1000_RXD_STAT_EOP)
1235 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1236 goto next_desc;
1237 }
1238
1239 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1240 !(netdev->features & NETIF_F_RXALL))) {
1241 dev_kfree_skb_irq(skb);
1242 goto next_desc;
1243 }
1244
1245 length = le16_to_cpu(rx_desc->wb.middle.length0);
1246
1247 if (!length) {
1248 e_dbg("Last part of the packet spanning multiple descriptors\n");
1249 dev_kfree_skb_irq(skb);
1250 goto next_desc;
1251 }
1252
1253 /* Good Receive */
1254 skb_put(skb, length);
1255
1256 {
1257 /*
1258 * this looks ugly, but it seems compiler issues make
1259 * it more efficient than reusing j
1260 */
1261 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1262
1263 /*
1264 * page alloc/put takes too long and effects small
1265 * packet throughput, so unsplit small packets and
1266 * save the alloc/put only valid in softirq (napi)
1267 * context to call kmap_*
1268 */
1269 if (l1 && (l1 <= copybreak) &&
1270 ((length + l1) <= adapter->rx_ps_bsize0)) {
1271 u8 *vaddr;
1272
1273 ps_page = &buffer_info->ps_pages[0];
1274
1275 /*
1276 * there is no documentation about how to call
1277 * kmap_atomic, so we can't hold the mapping
1278 * very long
1279 */
1280 dma_sync_single_for_cpu(&pdev->dev,
1281 ps_page->dma,
1282 PAGE_SIZE,
1283 DMA_FROM_DEVICE);
1284 vaddr = kmap_atomic(ps_page->page);
1285 memcpy(skb_tail_pointer(skb), vaddr, l1);
1286 kunmap_atomic(vaddr);
1287 dma_sync_single_for_device(&pdev->dev,
1288 ps_page->dma,
1289 PAGE_SIZE,
1290 DMA_FROM_DEVICE);
1291
1292 /* remove the CRC */
1293 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1294 if (!(netdev->features & NETIF_F_RXFCS))
1295 l1 -= 4;
1296 }
1297
1298 skb_put(skb, l1);
1299 goto copydone;
1300 } /* if */
1301 }
1302
1303 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1304 length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1305 if (!length)
1306 break;
1307
1308 ps_page = &buffer_info->ps_pages[j];
1309 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1310 DMA_FROM_DEVICE);
1311 ps_page->dma = 0;
1312 skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1313 ps_page->page = NULL;
1314 skb->len += length;
1315 skb->data_len += length;
1316 skb->truesize += PAGE_SIZE;
1317 }
1318
1319 /* strip the ethernet crc, problem is we're using pages now so
1320 * this whole operation can get a little cpu intensive
1321 */
1322 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1323 if (!(netdev->features & NETIF_F_RXFCS))
1324 pskb_trim(skb, skb->len - 4);
1325 }
1326
1327copydone:
1328 total_rx_bytes += skb->len;
1329 total_rx_packets++;
1330
1331 e1000_rx_checksum(adapter, staterr, skb);
1332
1333 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1334
1335 if (rx_desc->wb.upper.header_status &
1336 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1337 adapter->rx_hdr_split++;
1338
1339 e1000_receive_skb(adapter, netdev, skb,
1340 staterr, rx_desc->wb.middle.vlan);
1341
1342next_desc:
1343 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1344 buffer_info->skb = NULL;
1345
1346 /* return some buffers to hardware, one at a time is too slow */
1347 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1348 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1349 GFP_ATOMIC);
1350 cleaned_count = 0;
1351 }
1352
1353 /* use prefetched values */
1354 rx_desc = next_rxd;
1355 buffer_info = next_buffer;
1356
1357 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1358 }
1359 rx_ring->next_to_clean = i;
1360
1361 cleaned_count = e1000_desc_unused(rx_ring);
1362 if (cleaned_count)
1363 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1364
1365 adapter->total_rx_bytes += total_rx_bytes;
1366 adapter->total_rx_packets += total_rx_packets;
1367 return cleaned;
1368}
1369
1370/**
1371 * e1000_consume_page - helper function
1372 **/
1373static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1374 u16 length)
1375{
1376 bi->page = NULL;
1377 skb->len += length;
1378 skb->data_len += length;
1379 skb->truesize += PAGE_SIZE;
1380}
1381
1382/**
1383 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1384 * @adapter: board private structure
1385 *
1386 * the return value indicates whether actual cleaning was done, there
1387 * is no guarantee that everything was cleaned
1388 **/
1389static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
1390 int work_to_do)
1391{
1392 struct e1000_adapter *adapter = rx_ring->adapter;
1393 struct net_device *netdev = adapter->netdev;
1394 struct pci_dev *pdev = adapter->pdev;
1395 union e1000_rx_desc_extended *rx_desc, *next_rxd;
1396 struct e1000_buffer *buffer_info, *next_buffer;
1397 u32 length, staterr;
1398 unsigned int i;
1399 int cleaned_count = 0;
1400 bool cleaned = false;
1401 unsigned int total_rx_bytes=0, total_rx_packets=0;
1402
1403 i = rx_ring->next_to_clean;
1404 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1405 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1406 buffer_info = &rx_ring->buffer_info[i];
1407
1408 while (staterr & E1000_RXD_STAT_DD) {
1409 struct sk_buff *skb;
1410
1411 if (*work_done >= work_to_do)
1412 break;
1413 (*work_done)++;
1414 rmb(); /* read descriptor and rx_buffer_info after status DD */
1415
1416 skb = buffer_info->skb;
1417 buffer_info->skb = NULL;
1418
1419 ++i;
1420 if (i == rx_ring->count)
1421 i = 0;
1422 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1423 prefetch(next_rxd);
1424
1425 next_buffer = &rx_ring->buffer_info[i];
1426
1427 cleaned = true;
1428 cleaned_count++;
1429 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1430 DMA_FROM_DEVICE);
1431 buffer_info->dma = 0;
1432
1433 length = le16_to_cpu(rx_desc->wb.upper.length);
1434
1435 /* errors is only valid for DD + EOP descriptors */
1436 if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1437 ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1438 !(netdev->features & NETIF_F_RXALL)))) {
1439 /* recycle both page and skb */
1440 buffer_info->skb = skb;
1441 /* an error means any chain goes out the window too */
1442 if (rx_ring->rx_skb_top)
1443 dev_kfree_skb_irq(rx_ring->rx_skb_top);
1444 rx_ring->rx_skb_top = NULL;
1445 goto next_desc;
1446 }
1447
1448#define rxtop (rx_ring->rx_skb_top)
1449 if (!(staterr & E1000_RXD_STAT_EOP)) {
1450 /* this descriptor is only the beginning (or middle) */
1451 if (!rxtop) {
1452 /* this is the beginning of a chain */
1453 rxtop = skb;
1454 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1455 0, length);
1456 } else {
1457 /* this is the middle of a chain */
1458 skb_fill_page_desc(rxtop,
1459 skb_shinfo(rxtop)->nr_frags,
1460 buffer_info->page, 0, length);
1461 /* re-use the skb, only consumed the page */
1462 buffer_info->skb = skb;
1463 }
1464 e1000_consume_page(buffer_info, rxtop, length);
1465 goto next_desc;
1466 } else {
1467 if (rxtop) {
1468 /* end of the chain */
1469 skb_fill_page_desc(rxtop,
1470 skb_shinfo(rxtop)->nr_frags,
1471 buffer_info->page, 0, length);
1472 /* re-use the current skb, we only consumed the
1473 * page */
1474 buffer_info->skb = skb;
1475 skb = rxtop;
1476 rxtop = NULL;
1477 e1000_consume_page(buffer_info, skb, length);
1478 } else {
1479 /* no chain, got EOP, this buf is the packet
1480 * copybreak to save the put_page/alloc_page */
1481 if (length <= copybreak &&
1482 skb_tailroom(skb) >= length) {
1483 u8 *vaddr;
1484 vaddr = kmap_atomic(buffer_info->page);
1485 memcpy(skb_tail_pointer(skb), vaddr,
1486 length);
1487 kunmap_atomic(vaddr);
1488 /* re-use the page, so don't erase
1489 * buffer_info->page */
1490 skb_put(skb, length);
1491 } else {
1492 skb_fill_page_desc(skb, 0,
1493 buffer_info->page, 0,
1494 length);
1495 e1000_consume_page(buffer_info, skb,
1496 length);
1497 }
1498 }
1499 }
1500
1501 /* Receive Checksum Offload */
1502 e1000_rx_checksum(adapter, staterr, skb);
1503
1504 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1505
1506 /* probably a little skewed due to removing CRC */
1507 total_rx_bytes += skb->len;
1508 total_rx_packets++;
1509
1510 /* eth type trans needs skb->data to point to something */
1511 if (!pskb_may_pull(skb, ETH_HLEN)) {
1512 e_err("pskb_may_pull failed.\n");
1513 dev_kfree_skb_irq(skb);
1514 goto next_desc;
1515 }
1516
1517 e1000_receive_skb(adapter, netdev, skb, staterr,
1518 rx_desc->wb.upper.vlan);
1519
1520next_desc:
1521 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1522
1523 /* return some buffers to hardware, one at a time is too slow */
1524 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1525 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1526 GFP_ATOMIC);
1527 cleaned_count = 0;
1528 }
1529
1530 /* use prefetched values */
1531 rx_desc = next_rxd;
1532 buffer_info = next_buffer;
1533
1534 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1535 }
1536 rx_ring->next_to_clean = i;
1537
1538 cleaned_count = e1000_desc_unused(rx_ring);
1539 if (cleaned_count)
1540 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1541
1542 adapter->total_rx_bytes += total_rx_bytes;
1543 adapter->total_rx_packets += total_rx_packets;
1544 return cleaned;
1545}
1546
1547/**
1548 * e1000_clean_rx_ring - Free Rx Buffers per Queue
1549 * @rx_ring: Rx descriptor ring
1550 **/
1551static void e1000_clean_rx_ring(struct e1000_ring *rx_ring)
1552{
1553 struct e1000_adapter *adapter = rx_ring->adapter;
1554 struct e1000_buffer *buffer_info;
1555 struct e1000_ps_page *ps_page;
1556 struct pci_dev *pdev = adapter->pdev;
1557 unsigned int i, j;
1558
1559 /* Free all the Rx ring sk_buffs */
1560 for (i = 0; i < rx_ring->count; i++) {
1561 buffer_info = &rx_ring->buffer_info[i];
1562 if (buffer_info->dma) {
1563 if (adapter->clean_rx == e1000_clean_rx_irq)
1564 dma_unmap_single(&pdev->dev, buffer_info->dma,
1565 adapter->rx_buffer_len,
1566 DMA_FROM_DEVICE);
1567 else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1568 dma_unmap_page(&pdev->dev, buffer_info->dma,
1569 PAGE_SIZE,
1570 DMA_FROM_DEVICE);
1571 else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1572 dma_unmap_single(&pdev->dev, buffer_info->dma,
1573 adapter->rx_ps_bsize0,
1574 DMA_FROM_DEVICE);
1575 buffer_info->dma = 0;
1576 }
1577
1578 if (buffer_info->page) {
1579 put_page(buffer_info->page);
1580 buffer_info->page = NULL;
1581 }
1582
1583 if (buffer_info->skb) {
1584 dev_kfree_skb(buffer_info->skb);
1585 buffer_info->skb = NULL;
1586 }
1587
1588 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1589 ps_page = &buffer_info->ps_pages[j];
1590 if (!ps_page->page)
1591 break;
1592 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1593 DMA_FROM_DEVICE);
1594 ps_page->dma = 0;
1595 put_page(ps_page->page);
1596 ps_page->page = NULL;
1597 }
1598 }
1599
1600 /* there also may be some cached data from a chained receive */
1601 if (rx_ring->rx_skb_top) {
1602 dev_kfree_skb(rx_ring->rx_skb_top);
1603 rx_ring->rx_skb_top = NULL;
1604 }
1605
1606 /* Zero out the descriptor ring */
1607 memset(rx_ring->desc, 0, rx_ring->size);
1608
1609 rx_ring->next_to_clean = 0;
1610 rx_ring->next_to_use = 0;
1611 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1612
1613 writel(0, rx_ring->head);
1614 if (rx_ring->adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
1615 e1000e_update_rdt_wa(rx_ring, 0);
1616 else
1617 writel(0, rx_ring->tail);
1618}
1619
1620static void e1000e_downshift_workaround(struct work_struct *work)
1621{
1622 struct e1000_adapter *adapter = container_of(work,
1623 struct e1000_adapter, downshift_task);
1624
1625 if (test_bit(__E1000_DOWN, &adapter->state))
1626 return;
1627
1628 e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1629}
1630
1631/**
1632 * e1000_intr_msi - Interrupt Handler
1633 * @irq: interrupt number
1634 * @data: pointer to a network interface device structure
1635 **/
1636static irqreturn_t e1000_intr_msi(int irq, void *data)
1637{
1638 struct net_device *netdev = data;
1639 struct e1000_adapter *adapter = netdev_priv(netdev);
1640 struct e1000_hw *hw = &adapter->hw;
1641 u32 icr = er32(ICR);
1642
1643 /*
1644 * read ICR disables interrupts using IAM
1645 */
1646
1647 if (icr & E1000_ICR_LSC) {
1648 hw->mac.get_link_status = true;
1649 /*
1650 * ICH8 workaround-- Call gig speed drop workaround on cable
1651 * disconnect (LSC) before accessing any PHY registers
1652 */
1653 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1654 (!(er32(STATUS) & E1000_STATUS_LU)))
1655 schedule_work(&adapter->downshift_task);
1656
1657 /*
1658 * 80003ES2LAN workaround-- For packet buffer work-around on
1659 * link down event; disable receives here in the ISR and reset
1660 * adapter in watchdog
1661 */
1662 if (netif_carrier_ok(netdev) &&
1663 adapter->flags & FLAG_RX_NEEDS_RESTART) {
1664 /* disable receives */
1665 u32 rctl = er32(RCTL);
1666 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1667 adapter->flags |= FLAG_RX_RESTART_NOW;
1668 }
1669 /* guard against interrupt when we're going down */
1670 if (!test_bit(__E1000_DOWN, &adapter->state))
1671 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1672 }
1673
1674 if (napi_schedule_prep(&adapter->napi)) {
1675 adapter->total_tx_bytes = 0;
1676 adapter->total_tx_packets = 0;
1677 adapter->total_rx_bytes = 0;
1678 adapter->total_rx_packets = 0;
1679 __napi_schedule(&adapter->napi);
1680 }
1681
1682 return IRQ_HANDLED;
1683}
1684
1685/**
1686 * e1000_intr - Interrupt Handler
1687 * @irq: interrupt number
1688 * @data: pointer to a network interface device structure
1689 **/
1690static irqreturn_t e1000_intr(int irq, void *data)
1691{
1692 struct net_device *netdev = data;
1693 struct e1000_adapter *adapter = netdev_priv(netdev);
1694 struct e1000_hw *hw = &adapter->hw;
1695 u32 rctl, icr = er32(ICR);
1696
1697 if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1698 return IRQ_NONE; /* Not our interrupt */
1699
1700 /*
1701 * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1702 * not set, then the adapter didn't send an interrupt
1703 */
1704 if (!(icr & E1000_ICR_INT_ASSERTED))
1705 return IRQ_NONE;
1706
1707 /*
1708 * Interrupt Auto-Mask...upon reading ICR,
1709 * interrupts are masked. No need for the
1710 * IMC write
1711 */
1712
1713 if (icr & E1000_ICR_LSC) {
1714 hw->mac.get_link_status = true;
1715 /*
1716 * ICH8 workaround-- Call gig speed drop workaround on cable
1717 * disconnect (LSC) before accessing any PHY registers
1718 */
1719 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1720 (!(er32(STATUS) & E1000_STATUS_LU)))
1721 schedule_work(&adapter->downshift_task);
1722
1723 /*
1724 * 80003ES2LAN workaround--
1725 * For packet buffer work-around on link down event;
1726 * disable receives here in the ISR and
1727 * reset adapter in watchdog
1728 */
1729 if (netif_carrier_ok(netdev) &&
1730 (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1731 /* disable receives */
1732 rctl = er32(RCTL);
1733 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1734 adapter->flags |= FLAG_RX_RESTART_NOW;
1735 }
1736 /* guard against interrupt when we're going down */
1737 if (!test_bit(__E1000_DOWN, &adapter->state))
1738 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1739 }
1740
1741 if (napi_schedule_prep(&adapter->napi)) {
1742 adapter->total_tx_bytes = 0;
1743 adapter->total_tx_packets = 0;
1744 adapter->total_rx_bytes = 0;
1745 adapter->total_rx_packets = 0;
1746 __napi_schedule(&adapter->napi);
1747 }
1748
1749 return IRQ_HANDLED;
1750}
1751
1752static irqreturn_t e1000_msix_other(int irq, void *data)
1753{
1754 struct net_device *netdev = data;
1755 struct e1000_adapter *adapter = netdev_priv(netdev);
1756 struct e1000_hw *hw = &adapter->hw;
1757 u32 icr = er32(ICR);
1758
1759 if (!(icr & E1000_ICR_INT_ASSERTED)) {
1760 if (!test_bit(__E1000_DOWN, &adapter->state))
1761 ew32(IMS, E1000_IMS_OTHER);
1762 return IRQ_NONE;
1763 }
1764
1765 if (icr & adapter->eiac_mask)
1766 ew32(ICS, (icr & adapter->eiac_mask));
1767
1768 if (icr & E1000_ICR_OTHER) {
1769 if (!(icr & E1000_ICR_LSC))
1770 goto no_link_interrupt;
1771 hw->mac.get_link_status = true;
1772 /* guard against interrupt when we're going down */
1773 if (!test_bit(__E1000_DOWN, &adapter->state))
1774 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1775 }
1776
1777no_link_interrupt:
1778 if (!test_bit(__E1000_DOWN, &adapter->state))
1779 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1780
1781 return IRQ_HANDLED;
1782}
1783
1784
1785static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1786{
1787 struct net_device *netdev = data;
1788 struct e1000_adapter *adapter = netdev_priv(netdev);
1789 struct e1000_hw *hw = &adapter->hw;
1790 struct e1000_ring *tx_ring = adapter->tx_ring;
1791
1792
1793 adapter->total_tx_bytes = 0;
1794 adapter->total_tx_packets = 0;
1795
1796 if (!e1000_clean_tx_irq(tx_ring))
1797 /* Ring was not completely cleaned, so fire another interrupt */
1798 ew32(ICS, tx_ring->ims_val);
1799
1800 return IRQ_HANDLED;
1801}
1802
1803static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1804{
1805 struct net_device *netdev = data;
1806 struct e1000_adapter *adapter = netdev_priv(netdev);
1807 struct e1000_ring *rx_ring = adapter->rx_ring;
1808
1809 /* Write the ITR value calculated at the end of the
1810 * previous interrupt.
1811 */
1812 if (rx_ring->set_itr) {
1813 writel(1000000000 / (rx_ring->itr_val * 256),
1814 rx_ring->itr_register);
1815 rx_ring->set_itr = 0;
1816 }
1817
1818 if (napi_schedule_prep(&adapter->napi)) {
1819 adapter->total_rx_bytes = 0;
1820 adapter->total_rx_packets = 0;
1821 __napi_schedule(&adapter->napi);
1822 }
1823 return IRQ_HANDLED;
1824}
1825
1826/**
1827 * e1000_configure_msix - Configure MSI-X hardware
1828 *
1829 * e1000_configure_msix sets up the hardware to properly
1830 * generate MSI-X interrupts.
1831 **/
1832static void e1000_configure_msix(struct e1000_adapter *adapter)
1833{
1834 struct e1000_hw *hw = &adapter->hw;
1835 struct e1000_ring *rx_ring = adapter->rx_ring;
1836 struct e1000_ring *tx_ring = adapter->tx_ring;
1837 int vector = 0;
1838 u32 ctrl_ext, ivar = 0;
1839
1840 adapter->eiac_mask = 0;
1841
1842 /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1843 if (hw->mac.type == e1000_82574) {
1844 u32 rfctl = er32(RFCTL);
1845 rfctl |= E1000_RFCTL_ACK_DIS;
1846 ew32(RFCTL, rfctl);
1847 }
1848
1849#define E1000_IVAR_INT_ALLOC_VALID 0x8
1850 /* Configure Rx vector */
1851 rx_ring->ims_val = E1000_IMS_RXQ0;
1852 adapter->eiac_mask |= rx_ring->ims_val;
1853 if (rx_ring->itr_val)
1854 writel(1000000000 / (rx_ring->itr_val * 256),
1855 rx_ring->itr_register);
1856 else
1857 writel(1, rx_ring->itr_register);
1858 ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1859
1860 /* Configure Tx vector */
1861 tx_ring->ims_val = E1000_IMS_TXQ0;
1862 vector++;
1863 if (tx_ring->itr_val)
1864 writel(1000000000 / (tx_ring->itr_val * 256),
1865 tx_ring->itr_register);
1866 else
1867 writel(1, tx_ring->itr_register);
1868 adapter->eiac_mask |= tx_ring->ims_val;
1869 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1870
1871 /* set vector for Other Causes, e.g. link changes */
1872 vector++;
1873 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1874 if (rx_ring->itr_val)
1875 writel(1000000000 / (rx_ring->itr_val * 256),
1876 hw->hw_addr + E1000_EITR_82574(vector));
1877 else
1878 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1879
1880 /* Cause Tx interrupts on every write back */
1881 ivar |= (1 << 31);
1882
1883 ew32(IVAR, ivar);
1884
1885 /* enable MSI-X PBA support */
1886 ctrl_ext = er32(CTRL_EXT);
1887 ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1888
1889 /* Auto-Mask Other interrupts upon ICR read */
1890#define E1000_EIAC_MASK_82574 0x01F00000
1891 ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1892 ctrl_ext |= E1000_CTRL_EXT_EIAME;
1893 ew32(CTRL_EXT, ctrl_ext);
1894 e1e_flush();
1895}
1896
1897void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1898{
1899 if (adapter->msix_entries) {
1900 pci_disable_msix(adapter->pdev);
1901 kfree(adapter->msix_entries);
1902 adapter->msix_entries = NULL;
1903 } else if (adapter->flags & FLAG_MSI_ENABLED) {
1904 pci_disable_msi(adapter->pdev);
1905 adapter->flags &= ~FLAG_MSI_ENABLED;
1906 }
1907}
1908
1909/**
1910 * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1911 *
1912 * Attempt to configure interrupts using the best available
1913 * capabilities of the hardware and kernel.
1914 **/
1915void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1916{
1917 int err;
1918 int i;
1919
1920 switch (adapter->int_mode) {
1921 case E1000E_INT_MODE_MSIX:
1922 if (adapter->flags & FLAG_HAS_MSIX) {
1923 adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1924 adapter->msix_entries = kcalloc(adapter->num_vectors,
1925 sizeof(struct msix_entry),
1926 GFP_KERNEL);
1927 if (adapter->msix_entries) {
1928 for (i = 0; i < adapter->num_vectors; i++)
1929 adapter->msix_entries[i].entry = i;
1930
1931 err = pci_enable_msix(adapter->pdev,
1932 adapter->msix_entries,
1933 adapter->num_vectors);
1934 if (err == 0)
1935 return;
1936 }
1937 /* MSI-X failed, so fall through and try MSI */
1938 e_err("Failed to initialize MSI-X interrupts. Falling back to MSI interrupts.\n");
1939 e1000e_reset_interrupt_capability(adapter);
1940 }
1941 adapter->int_mode = E1000E_INT_MODE_MSI;
1942 /* Fall through */
1943 case E1000E_INT_MODE_MSI:
1944 if (!pci_enable_msi(adapter->pdev)) {
1945 adapter->flags |= FLAG_MSI_ENABLED;
1946 } else {
1947 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1948 e_err("Failed to initialize MSI interrupts. Falling back to legacy interrupts.\n");
1949 }
1950 /* Fall through */
1951 case E1000E_INT_MODE_LEGACY:
1952 /* Don't do anything; this is the system default */
1953 break;
1954 }
1955
1956 /* store the number of vectors being used */
1957 adapter->num_vectors = 1;
1958}
1959
1960/**
1961 * e1000_request_msix - Initialize MSI-X interrupts
1962 *
1963 * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1964 * kernel.
1965 **/
1966static int e1000_request_msix(struct e1000_adapter *adapter)
1967{
1968 struct net_device *netdev = adapter->netdev;
1969 int err = 0, vector = 0;
1970
1971 if (strlen(netdev->name) < (IFNAMSIZ - 5))
1972 snprintf(adapter->rx_ring->name,
1973 sizeof(adapter->rx_ring->name) - 1,
1974 "%s-rx-0", netdev->name);
1975 else
1976 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1977 err = request_irq(adapter->msix_entries[vector].vector,
1978 e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1979 netdev);
1980 if (err)
1981 return err;
1982 adapter->rx_ring->itr_register = adapter->hw.hw_addr +
1983 E1000_EITR_82574(vector);
1984 adapter->rx_ring->itr_val = adapter->itr;
1985 vector++;
1986
1987 if (strlen(netdev->name) < (IFNAMSIZ - 5))
1988 snprintf(adapter->tx_ring->name,
1989 sizeof(adapter->tx_ring->name) - 1,
1990 "%s-tx-0", netdev->name);
1991 else
1992 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1993 err = request_irq(adapter->msix_entries[vector].vector,
1994 e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1995 netdev);
1996 if (err)
1997 return err;
1998 adapter->tx_ring->itr_register = adapter->hw.hw_addr +
1999 E1000_EITR_82574(vector);
2000 adapter->tx_ring->itr_val = adapter->itr;
2001 vector++;
2002
2003 err = request_irq(adapter->msix_entries[vector].vector,
2004 e1000_msix_other, 0, netdev->name, netdev);
2005 if (err)
2006 return err;
2007
2008 e1000_configure_msix(adapter);
2009
2010 return 0;
2011}
2012
2013/**
2014 * e1000_request_irq - initialize interrupts
2015 *
2016 * Attempts to configure interrupts using the best available
2017 * capabilities of the hardware and kernel.
2018 **/
2019static int e1000_request_irq(struct e1000_adapter *adapter)
2020{
2021 struct net_device *netdev = adapter->netdev;
2022 int err;
2023
2024 if (adapter->msix_entries) {
2025 err = e1000_request_msix(adapter);
2026 if (!err)
2027 return err;
2028 /* fall back to MSI */
2029 e1000e_reset_interrupt_capability(adapter);
2030 adapter->int_mode = E1000E_INT_MODE_MSI;
2031 e1000e_set_interrupt_capability(adapter);
2032 }
2033 if (adapter->flags & FLAG_MSI_ENABLED) {
2034 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2035 netdev->name, netdev);
2036 if (!err)
2037 return err;
2038
2039 /* fall back to legacy interrupt */
2040 e1000e_reset_interrupt_capability(adapter);
2041 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2042 }
2043
2044 err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2045 netdev->name, netdev);
2046 if (err)
2047 e_err("Unable to allocate interrupt, Error: %d\n", err);
2048
2049 return err;
2050}
2051
2052static void e1000_free_irq(struct e1000_adapter *adapter)
2053{
2054 struct net_device *netdev = adapter->netdev;
2055
2056 if (adapter->msix_entries) {
2057 int vector = 0;
2058
2059 free_irq(adapter->msix_entries[vector].vector, netdev);
2060 vector++;
2061
2062 free_irq(adapter->msix_entries[vector].vector, netdev);
2063 vector++;
2064
2065 /* Other Causes interrupt vector */
2066 free_irq(adapter->msix_entries[vector].vector, netdev);
2067 return;
2068 }
2069
2070 free_irq(adapter->pdev->irq, netdev);
2071}
2072
2073/**
2074 * e1000_irq_disable - Mask off interrupt generation on the NIC
2075 **/
2076static void e1000_irq_disable(struct e1000_adapter *adapter)
2077{
2078 struct e1000_hw *hw = &adapter->hw;
2079
2080 ew32(IMC, ~0);
2081 if (adapter->msix_entries)
2082 ew32(EIAC_82574, 0);
2083 e1e_flush();
2084
2085 if (adapter->msix_entries) {
2086 int i;
2087 for (i = 0; i < adapter->num_vectors; i++)
2088 synchronize_irq(adapter->msix_entries[i].vector);
2089 } else {
2090 synchronize_irq(adapter->pdev->irq);
2091 }
2092}
2093
2094/**
2095 * e1000_irq_enable - Enable default interrupt generation settings
2096 **/
2097static void e1000_irq_enable(struct e1000_adapter *adapter)
2098{
2099 struct e1000_hw *hw = &adapter->hw;
2100
2101 if (adapter->msix_entries) {
2102 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2103 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
2104 } else {
2105 ew32(IMS, IMS_ENABLE_MASK);
2106 }
2107 e1e_flush();
2108}
2109
2110/**
2111 * e1000e_get_hw_control - get control of the h/w from f/w
2112 * @adapter: address of board private structure
2113 *
2114 * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2115 * For ASF and Pass Through versions of f/w this means that
2116 * the driver is loaded. For AMT version (only with 82573)
2117 * of the f/w this means that the network i/f is open.
2118 **/
2119void e1000e_get_hw_control(struct e1000_adapter *adapter)
2120{
2121 struct e1000_hw *hw = &adapter->hw;
2122 u32 ctrl_ext;
2123 u32 swsm;
2124
2125 /* Let firmware know the driver has taken over */
2126 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2127 swsm = er32(SWSM);
2128 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2129 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2130 ctrl_ext = er32(CTRL_EXT);
2131 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2132 }
2133}
2134
2135/**
2136 * e1000e_release_hw_control - release control of the h/w to f/w
2137 * @adapter: address of board private structure
2138 *
2139 * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2140 * For ASF and Pass Through versions of f/w this means that the
2141 * driver is no longer loaded. For AMT version (only with 82573) i
2142 * of the f/w this means that the network i/f is closed.
2143 *
2144 **/
2145void e1000e_release_hw_control(struct e1000_adapter *adapter)
2146{
2147 struct e1000_hw *hw = &adapter->hw;
2148 u32 ctrl_ext;
2149 u32 swsm;
2150
2151 /* Let firmware taken over control of h/w */
2152 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2153 swsm = er32(SWSM);
2154 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2155 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2156 ctrl_ext = er32(CTRL_EXT);
2157 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2158 }
2159}
2160
2161/**
2162 * @e1000_alloc_ring - allocate memory for a ring structure
2163 **/
2164static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2165 struct e1000_ring *ring)
2166{
2167 struct pci_dev *pdev = adapter->pdev;
2168
2169 ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2170 GFP_KERNEL);
2171 if (!ring->desc)
2172 return -ENOMEM;
2173
2174 return 0;
2175}
2176
2177/**
2178 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2179 * @tx_ring: Tx descriptor ring
2180 *
2181 * Return 0 on success, negative on failure
2182 **/
2183int e1000e_setup_tx_resources(struct e1000_ring *tx_ring)
2184{
2185 struct e1000_adapter *adapter = tx_ring->adapter;
2186 int err = -ENOMEM, size;
2187
2188 size = sizeof(struct e1000_buffer) * tx_ring->count;
2189 tx_ring->buffer_info = vzalloc(size);
2190 if (!tx_ring->buffer_info)
2191 goto err;
2192
2193 /* round up to nearest 4K */
2194 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2195 tx_ring->size = ALIGN(tx_ring->size, 4096);
2196
2197 err = e1000_alloc_ring_dma(adapter, tx_ring);
2198 if (err)
2199 goto err;
2200
2201 tx_ring->next_to_use = 0;
2202 tx_ring->next_to_clean = 0;
2203
2204 return 0;
2205err:
2206 vfree(tx_ring->buffer_info);
2207 e_err("Unable to allocate memory for the transmit descriptor ring\n");
2208 return err;
2209}
2210
2211/**
2212 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2213 * @rx_ring: Rx descriptor ring
2214 *
2215 * Returns 0 on success, negative on failure
2216 **/
2217int e1000e_setup_rx_resources(struct e1000_ring *rx_ring)
2218{
2219 struct e1000_adapter *adapter = rx_ring->adapter;
2220 struct e1000_buffer *buffer_info;
2221 int i, size, desc_len, err = -ENOMEM;
2222
2223 size = sizeof(struct e1000_buffer) * rx_ring->count;
2224 rx_ring->buffer_info = vzalloc(size);
2225 if (!rx_ring->buffer_info)
2226 goto err;
2227
2228 for (i = 0; i < rx_ring->count; i++) {
2229 buffer_info = &rx_ring->buffer_info[i];
2230 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2231 sizeof(struct e1000_ps_page),
2232 GFP_KERNEL);
2233 if (!buffer_info->ps_pages)
2234 goto err_pages;
2235 }
2236
2237 desc_len = sizeof(union e1000_rx_desc_packet_split);
2238
2239 /* Round up to nearest 4K */
2240 rx_ring->size = rx_ring->count * desc_len;
2241 rx_ring->size = ALIGN(rx_ring->size, 4096);
2242
2243 err = e1000_alloc_ring_dma(adapter, rx_ring);
2244 if (err)
2245 goto err_pages;
2246
2247 rx_ring->next_to_clean = 0;
2248 rx_ring->next_to_use = 0;
2249 rx_ring->rx_skb_top = NULL;
2250
2251 return 0;
2252
2253err_pages:
2254 for (i = 0; i < rx_ring->count; i++) {
2255 buffer_info = &rx_ring->buffer_info[i];
2256 kfree(buffer_info->ps_pages);
2257 }
2258err:
2259 vfree(rx_ring->buffer_info);
2260 e_err("Unable to allocate memory for the receive descriptor ring\n");
2261 return err;
2262}
2263
2264/**
2265 * e1000_clean_tx_ring - Free Tx Buffers
2266 * @tx_ring: Tx descriptor ring
2267 **/
2268static void e1000_clean_tx_ring(struct e1000_ring *tx_ring)
2269{
2270 struct e1000_adapter *adapter = tx_ring->adapter;
2271 struct e1000_buffer *buffer_info;
2272 unsigned long size;
2273 unsigned int i;
2274
2275 for (i = 0; i < tx_ring->count; i++) {
2276 buffer_info = &tx_ring->buffer_info[i];
2277 e1000_put_txbuf(tx_ring, buffer_info);
2278 }
2279
2280 netdev_reset_queue(adapter->netdev);
2281 size = sizeof(struct e1000_buffer) * tx_ring->count;
2282 memset(tx_ring->buffer_info, 0, size);
2283
2284 memset(tx_ring->desc, 0, tx_ring->size);
2285
2286 tx_ring->next_to_use = 0;
2287 tx_ring->next_to_clean = 0;
2288
2289 writel(0, tx_ring->head);
2290 if (tx_ring->adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
2291 e1000e_update_tdt_wa(tx_ring, 0);
2292 else
2293 writel(0, tx_ring->tail);
2294}
2295
2296/**
2297 * e1000e_free_tx_resources - Free Tx Resources per Queue
2298 * @tx_ring: Tx descriptor ring
2299 *
2300 * Free all transmit software resources
2301 **/
2302void e1000e_free_tx_resources(struct e1000_ring *tx_ring)
2303{
2304 struct e1000_adapter *adapter = tx_ring->adapter;
2305 struct pci_dev *pdev = adapter->pdev;
2306
2307 e1000_clean_tx_ring(tx_ring);
2308
2309 vfree(tx_ring->buffer_info);
2310 tx_ring->buffer_info = NULL;
2311
2312 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2313 tx_ring->dma);
2314 tx_ring->desc = NULL;
2315}
2316
2317/**
2318 * e1000e_free_rx_resources - Free Rx Resources
2319 * @rx_ring: Rx descriptor ring
2320 *
2321 * Free all receive software resources
2322 **/
2323void e1000e_free_rx_resources(struct e1000_ring *rx_ring)
2324{
2325 struct e1000_adapter *adapter = rx_ring->adapter;
2326 struct pci_dev *pdev = adapter->pdev;
2327 int i;
2328
2329 e1000_clean_rx_ring(rx_ring);
2330
2331 for (i = 0; i < rx_ring->count; i++)
2332 kfree(rx_ring->buffer_info[i].ps_pages);
2333
2334 vfree(rx_ring->buffer_info);
2335 rx_ring->buffer_info = NULL;
2336
2337 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2338 rx_ring->dma);
2339 rx_ring->desc = NULL;
2340}
2341
2342/**
2343 * e1000_update_itr - update the dynamic ITR value based on statistics
2344 * @adapter: pointer to adapter
2345 * @itr_setting: current adapter->itr
2346 * @packets: the number of packets during this measurement interval
2347 * @bytes: the number of bytes during this measurement interval
2348 *
2349 * Stores a new ITR value based on packets and byte
2350 * counts during the last interrupt. The advantage of per interrupt
2351 * computation is faster updates and more accurate ITR for the current
2352 * traffic pattern. Constants in this function were computed
2353 * based on theoretical maximum wire speed and thresholds were set based
2354 * on testing data as well as attempting to minimize response time
2355 * while increasing bulk throughput. This functionality is controlled
2356 * by the InterruptThrottleRate module parameter.
2357 **/
2358static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2359 u16 itr_setting, int packets,
2360 int bytes)
2361{
2362 unsigned int retval = itr_setting;
2363
2364 if (packets == 0)
2365 return itr_setting;
2366
2367 switch (itr_setting) {
2368 case lowest_latency:
2369 /* handle TSO and jumbo frames */
2370 if (bytes/packets > 8000)
2371 retval = bulk_latency;
2372 else if ((packets < 5) && (bytes > 512))
2373 retval = low_latency;
2374 break;
2375 case low_latency: /* 50 usec aka 20000 ints/s */
2376 if (bytes > 10000) {
2377 /* this if handles the TSO accounting */
2378 if (bytes/packets > 8000)
2379 retval = bulk_latency;
2380 else if ((packets < 10) || ((bytes/packets) > 1200))
2381 retval = bulk_latency;
2382 else if ((packets > 35))
2383 retval = lowest_latency;
2384 } else if (bytes/packets > 2000) {
2385 retval = bulk_latency;
2386 } else if (packets <= 2 && bytes < 512) {
2387 retval = lowest_latency;
2388 }
2389 break;
2390 case bulk_latency: /* 250 usec aka 4000 ints/s */
2391 if (bytes > 25000) {
2392 if (packets > 35)
2393 retval = low_latency;
2394 } else if (bytes < 6000) {
2395 retval = low_latency;
2396 }
2397 break;
2398 }
2399
2400 return retval;
2401}
2402
2403static void e1000_set_itr(struct e1000_adapter *adapter)
2404{
2405 struct e1000_hw *hw = &adapter->hw;
2406 u16 current_itr;
2407 u32 new_itr = adapter->itr;
2408
2409 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2410 if (adapter->link_speed != SPEED_1000) {
2411 current_itr = 0;
2412 new_itr = 4000;
2413 goto set_itr_now;
2414 }
2415
2416 if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2417 new_itr = 0;
2418 goto set_itr_now;
2419 }
2420
2421 adapter->tx_itr = e1000_update_itr(adapter,
2422 adapter->tx_itr,
2423 adapter->total_tx_packets,
2424 adapter->total_tx_bytes);
2425 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2426 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2427 adapter->tx_itr = low_latency;
2428
2429 adapter->rx_itr = e1000_update_itr(adapter,
2430 adapter->rx_itr,
2431 adapter->total_rx_packets,
2432 adapter->total_rx_bytes);
2433 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2434 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2435 adapter->rx_itr = low_latency;
2436
2437 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2438
2439 switch (current_itr) {
2440 /* counts and packets in update_itr are dependent on these numbers */
2441 case lowest_latency:
2442 new_itr = 70000;
2443 break;
2444 case low_latency:
2445 new_itr = 20000; /* aka hwitr = ~200 */
2446 break;
2447 case bulk_latency:
2448 new_itr = 4000;
2449 break;
2450 default:
2451 break;
2452 }
2453
2454set_itr_now:
2455 if (new_itr != adapter->itr) {
2456 /*
2457 * this attempts to bias the interrupt rate towards Bulk
2458 * by adding intermediate steps when interrupt rate is
2459 * increasing
2460 */
2461 new_itr = new_itr > adapter->itr ?
2462 min(adapter->itr + (new_itr >> 2), new_itr) :
2463 new_itr;
2464 adapter->itr = new_itr;
2465 adapter->rx_ring->itr_val = new_itr;
2466 if (adapter->msix_entries)
2467 adapter->rx_ring->set_itr = 1;
2468 else
2469 if (new_itr)
2470 ew32(ITR, 1000000000 / (new_itr * 256));
2471 else
2472 ew32(ITR, 0);
2473 }
2474}
2475
2476/**
2477 * e1000_alloc_queues - Allocate memory for all rings
2478 * @adapter: board private structure to initialize
2479 **/
2480static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2481{
2482 int size = sizeof(struct e1000_ring);
2483
2484 adapter->tx_ring = kzalloc(size, GFP_KERNEL);
2485 if (!adapter->tx_ring)
2486 goto err;
2487 adapter->tx_ring->count = adapter->tx_ring_count;
2488 adapter->tx_ring->adapter = adapter;
2489
2490 adapter->rx_ring = kzalloc(size, GFP_KERNEL);
2491 if (!adapter->rx_ring)
2492 goto err;
2493 adapter->rx_ring->count = adapter->rx_ring_count;
2494 adapter->rx_ring->adapter = adapter;
2495
2496 return 0;
2497err:
2498 e_err("Unable to allocate memory for queues\n");
2499 kfree(adapter->rx_ring);
2500 kfree(adapter->tx_ring);
2501 return -ENOMEM;
2502}
2503
2504/**
2505 * e1000e_poll - NAPI Rx polling callback
2506 * @napi: struct associated with this polling callback
2507 * @weight: number of packets driver is allowed to process this poll
2508 **/
2509static int e1000e_poll(struct napi_struct *napi, int weight)
2510{
2511 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter,
2512 napi);
2513 struct e1000_hw *hw = &adapter->hw;
2514 struct net_device *poll_dev = adapter->netdev;
2515 int tx_cleaned = 1, work_done = 0;
2516
2517 adapter = netdev_priv(poll_dev);
2518
2519 if (!adapter->msix_entries ||
2520 (adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2521 tx_cleaned = e1000_clean_tx_irq(adapter->tx_ring);
2522
2523 adapter->clean_rx(adapter->rx_ring, &work_done, weight);
2524
2525 if (!tx_cleaned)
2526 work_done = weight;
2527
2528 /* If weight not fully consumed, exit the polling mode */
2529 if (work_done < weight) {
2530 if (adapter->itr_setting & 3)
2531 e1000_set_itr(adapter);
2532 napi_complete(napi);
2533 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2534 if (adapter->msix_entries)
2535 ew32(IMS, adapter->rx_ring->ims_val);
2536 else
2537 e1000_irq_enable(adapter);
2538 }
2539 }
2540
2541 return work_done;
2542}
2543
2544static int e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2545{
2546 struct e1000_adapter *adapter = netdev_priv(netdev);
2547 struct e1000_hw *hw = &adapter->hw;
2548 u32 vfta, index;
2549
2550 /* don't update vlan cookie if already programmed */
2551 if ((adapter->hw.mng_cookie.status &
2552 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2553 (vid == adapter->mng_vlan_id))
2554 return 0;
2555
2556 /* add VID to filter table */
2557 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2558 index = (vid >> 5) & 0x7F;
2559 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2560 vfta |= (1 << (vid & 0x1F));
2561 hw->mac.ops.write_vfta(hw, index, vfta);
2562 }
2563
2564 set_bit(vid, adapter->active_vlans);
2565
2566 return 0;
2567}
2568
2569static int e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2570{
2571 struct e1000_adapter *adapter = netdev_priv(netdev);
2572 struct e1000_hw *hw = &adapter->hw;
2573 u32 vfta, index;
2574
2575 if ((adapter->hw.mng_cookie.status &
2576 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2577 (vid == adapter->mng_vlan_id)) {
2578 /* release control to f/w */
2579 e1000e_release_hw_control(adapter);
2580 return 0;
2581 }
2582
2583 /* remove VID from filter table */
2584 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2585 index = (vid >> 5) & 0x7F;
2586 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2587 vfta &= ~(1 << (vid & 0x1F));
2588 hw->mac.ops.write_vfta(hw, index, vfta);
2589 }
2590
2591 clear_bit(vid, adapter->active_vlans);
2592
2593 return 0;
2594}
2595
2596/**
2597 * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2598 * @adapter: board private structure to initialize
2599 **/
2600static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2601{
2602 struct net_device *netdev = adapter->netdev;
2603 struct e1000_hw *hw = &adapter->hw;
2604 u32 rctl;
2605
2606 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2607 /* disable VLAN receive filtering */
2608 rctl = er32(RCTL);
2609 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2610 ew32(RCTL, rctl);
2611
2612 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2613 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2614 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2615 }
2616 }
2617}
2618
2619/**
2620 * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2621 * @adapter: board private structure to initialize
2622 **/
2623static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2624{
2625 struct e1000_hw *hw = &adapter->hw;
2626 u32 rctl;
2627
2628 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2629 /* enable VLAN receive filtering */
2630 rctl = er32(RCTL);
2631 rctl |= E1000_RCTL_VFE;
2632 rctl &= ~E1000_RCTL_CFIEN;
2633 ew32(RCTL, rctl);
2634 }
2635}
2636
2637/**
2638 * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2639 * @adapter: board private structure to initialize
2640 **/
2641static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2642{
2643 struct e1000_hw *hw = &adapter->hw;
2644 u32 ctrl;
2645
2646 /* disable VLAN tag insert/strip */
2647 ctrl = er32(CTRL);
2648 ctrl &= ~E1000_CTRL_VME;
2649 ew32(CTRL, ctrl);
2650}
2651
2652/**
2653 * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2654 * @adapter: board private structure to initialize
2655 **/
2656static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2657{
2658 struct e1000_hw *hw = &adapter->hw;
2659 u32 ctrl;
2660
2661 /* enable VLAN tag insert/strip */
2662 ctrl = er32(CTRL);
2663 ctrl |= E1000_CTRL_VME;
2664 ew32(CTRL, ctrl);
2665}
2666
2667static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2668{
2669 struct net_device *netdev = adapter->netdev;
2670 u16 vid = adapter->hw.mng_cookie.vlan_id;
2671 u16 old_vid = adapter->mng_vlan_id;
2672
2673 if (adapter->hw.mng_cookie.status &
2674 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2675 e1000_vlan_rx_add_vid(netdev, vid);
2676 adapter->mng_vlan_id = vid;
2677 }
2678
2679 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2680 e1000_vlan_rx_kill_vid(netdev, old_vid);
2681}
2682
2683static void e1000_restore_vlan(struct e1000_adapter *adapter)
2684{
2685 u16 vid;
2686
2687 e1000_vlan_rx_add_vid(adapter->netdev, 0);
2688
2689 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2690 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2691}
2692
2693static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2694{
2695 struct e1000_hw *hw = &adapter->hw;
2696 u32 manc, manc2h, mdef, i, j;
2697
2698 if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2699 return;
2700
2701 manc = er32(MANC);
2702
2703 /*
2704 * enable receiving management packets to the host. this will probably
2705 * generate destination unreachable messages from the host OS, but
2706 * the packets will be handled on SMBUS
2707 */
2708 manc |= E1000_MANC_EN_MNG2HOST;
2709 manc2h = er32(MANC2H);
2710
2711 switch (hw->mac.type) {
2712 default:
2713 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2714 break;
2715 case e1000_82574:
2716 case e1000_82583:
2717 /*
2718 * Check if IPMI pass-through decision filter already exists;
2719 * if so, enable it.
2720 */
2721 for (i = 0, j = 0; i < 8; i++) {
2722 mdef = er32(MDEF(i));
2723
2724 /* Ignore filters with anything other than IPMI ports */
2725 if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2726 continue;
2727
2728 /* Enable this decision filter in MANC2H */
2729 if (mdef)
2730 manc2h |= (1 << i);
2731
2732 j |= mdef;
2733 }
2734
2735 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2736 break;
2737
2738 /* Create new decision filter in an empty filter */
2739 for (i = 0, j = 0; i < 8; i++)
2740 if (er32(MDEF(i)) == 0) {
2741 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2742 E1000_MDEF_PORT_664));
2743 manc2h |= (1 << 1);
2744 j++;
2745 break;
2746 }
2747
2748 if (!j)
2749 e_warn("Unable to create IPMI pass-through filter\n");
2750 break;
2751 }
2752
2753 ew32(MANC2H, manc2h);
2754 ew32(MANC, manc);
2755}
2756
2757/**
2758 * e1000_configure_tx - Configure Transmit Unit after Reset
2759 * @adapter: board private structure
2760 *
2761 * Configure the Tx unit of the MAC after a reset.
2762 **/
2763static void e1000_configure_tx(struct e1000_adapter *adapter)
2764{
2765 struct e1000_hw *hw = &adapter->hw;
2766 struct e1000_ring *tx_ring = adapter->tx_ring;
2767 u64 tdba;
2768 u32 tdlen, tarc;
2769
2770 /* Setup the HW Tx Head and Tail descriptor pointers */
2771 tdba = tx_ring->dma;
2772 tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2773 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
2774 ew32(TDBAH(0), (tdba >> 32));
2775 ew32(TDLEN(0), tdlen);
2776 ew32(TDH(0), 0);
2777 ew32(TDT(0), 0);
2778 tx_ring->head = adapter->hw.hw_addr + E1000_TDH(0);
2779 tx_ring->tail = adapter->hw.hw_addr + E1000_TDT(0);
2780
2781 /* Set the Tx Interrupt Delay register */
2782 ew32(TIDV, adapter->tx_int_delay);
2783 /* Tx irq moderation */
2784 ew32(TADV, adapter->tx_abs_int_delay);
2785
2786 if (adapter->flags2 & FLAG2_DMA_BURST) {
2787 u32 txdctl = er32(TXDCTL(0));
2788 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2789 E1000_TXDCTL_WTHRESH);
2790 /*
2791 * set up some performance related parameters to encourage the
2792 * hardware to use the bus more efficiently in bursts, depends
2793 * on the tx_int_delay to be enabled,
2794 * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2795 * hthresh = 1 ==> prefetch when one or more available
2796 * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2797 * BEWARE: this seems to work but should be considered first if
2798 * there are Tx hangs or other Tx related bugs
2799 */
2800 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2801 ew32(TXDCTL(0), txdctl);
2802 }
2803 /* erratum work around: set txdctl the same for both queues */
2804 ew32(TXDCTL(1), er32(TXDCTL(0)));
2805
2806 if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2807 tarc = er32(TARC(0));
2808 /*
2809 * set the speed mode bit, we'll clear it if we're not at
2810 * gigabit link later
2811 */
2812#define SPEED_MODE_BIT (1 << 21)
2813 tarc |= SPEED_MODE_BIT;
2814 ew32(TARC(0), tarc);
2815 }
2816
2817 /* errata: program both queues to unweighted RR */
2818 if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2819 tarc = er32(TARC(0));
2820 tarc |= 1;
2821 ew32(TARC(0), tarc);
2822 tarc = er32(TARC(1));
2823 tarc |= 1;
2824 ew32(TARC(1), tarc);
2825 }
2826
2827 /* Setup Transmit Descriptor Settings for eop descriptor */
2828 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2829
2830 /* only set IDE if we are delaying interrupts using the timers */
2831 if (adapter->tx_int_delay)
2832 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2833
2834 /* enable Report Status bit */
2835 adapter->txd_cmd |= E1000_TXD_CMD_RS;
2836
2837 hw->mac.ops.config_collision_dist(hw);
2838}
2839
2840/**
2841 * e1000_setup_rctl - configure the receive control registers
2842 * @adapter: Board private structure
2843 **/
2844#define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2845 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2846static void e1000_setup_rctl(struct e1000_adapter *adapter)
2847{
2848 struct e1000_hw *hw = &adapter->hw;
2849 u32 rctl, rfctl;
2850 u32 pages = 0;
2851
2852 /* Workaround Si errata on PCHx - configure jumbo frame flow */
2853 if (hw->mac.type >= e1000_pch2lan) {
2854 s32 ret_val;
2855
2856 if (adapter->netdev->mtu > ETH_DATA_LEN)
2857 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2858 else
2859 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2860
2861 if (ret_val)
2862 e_dbg("failed to enable jumbo frame workaround mode\n");
2863 }
2864
2865 /* Program MC offset vector base */
2866 rctl = er32(RCTL);
2867 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2868 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2869 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2870 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2871
2872 /* Do not Store bad packets */
2873 rctl &= ~E1000_RCTL_SBP;
2874
2875 /* Enable Long Packet receive */
2876 if (adapter->netdev->mtu <= ETH_DATA_LEN)
2877 rctl &= ~E1000_RCTL_LPE;
2878 else
2879 rctl |= E1000_RCTL_LPE;
2880
2881 /* Some systems expect that the CRC is included in SMBUS traffic. The
2882 * hardware strips the CRC before sending to both SMBUS (BMC) and to
2883 * host memory when this is enabled
2884 */
2885 if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2886 rctl |= E1000_RCTL_SECRC;
2887
2888 /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2889 if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2890 u16 phy_data;
2891
2892 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2893 phy_data &= 0xfff8;
2894 phy_data |= (1 << 2);
2895 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2896
2897 e1e_rphy(hw, 22, &phy_data);
2898 phy_data &= 0x0fff;
2899 phy_data |= (1 << 14);
2900 e1e_wphy(hw, 0x10, 0x2823);
2901 e1e_wphy(hw, 0x11, 0x0003);
2902 e1e_wphy(hw, 22, phy_data);
2903 }
2904
2905 /* Setup buffer sizes */
2906 rctl &= ~E1000_RCTL_SZ_4096;
2907 rctl |= E1000_RCTL_BSEX;
2908 switch (adapter->rx_buffer_len) {
2909 case 2048:
2910 default:
2911 rctl |= E1000_RCTL_SZ_2048;
2912 rctl &= ~E1000_RCTL_BSEX;
2913 break;
2914 case 4096:
2915 rctl |= E1000_RCTL_SZ_4096;
2916 break;
2917 case 8192:
2918 rctl |= E1000_RCTL_SZ_8192;
2919 break;
2920 case 16384:
2921 rctl |= E1000_RCTL_SZ_16384;
2922 break;
2923 }
2924
2925 /* Enable Extended Status in all Receive Descriptors */
2926 rfctl = er32(RFCTL);
2927 rfctl |= E1000_RFCTL_EXTEN;
2928 ew32(RFCTL, rfctl);
2929
2930 /*
2931 * 82571 and greater support packet-split where the protocol
2932 * header is placed in skb->data and the packet data is
2933 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2934 * In the case of a non-split, skb->data is linearly filled,
2935 * followed by the page buffers. Therefore, skb->data is
2936 * sized to hold the largest protocol header.
2937 *
2938 * allocations using alloc_page take too long for regular MTU
2939 * so only enable packet split for jumbo frames
2940 *
2941 * Using pages when the page size is greater than 16k wastes
2942 * a lot of memory, since we allocate 3 pages at all times
2943 * per packet.
2944 */
2945 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2946 if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2947 adapter->rx_ps_pages = pages;
2948 else
2949 adapter->rx_ps_pages = 0;
2950
2951 if (adapter->rx_ps_pages) {
2952 u32 psrctl = 0;
2953
2954 /* Enable Packet split descriptors */
2955 rctl |= E1000_RCTL_DTYP_PS;
2956
2957 psrctl |= adapter->rx_ps_bsize0 >>
2958 E1000_PSRCTL_BSIZE0_SHIFT;
2959
2960 switch (adapter->rx_ps_pages) {
2961 case 3:
2962 psrctl |= PAGE_SIZE <<
2963 E1000_PSRCTL_BSIZE3_SHIFT;
2964 case 2:
2965 psrctl |= PAGE_SIZE <<
2966 E1000_PSRCTL_BSIZE2_SHIFT;
2967 case 1:
2968 psrctl |= PAGE_SIZE >>
2969 E1000_PSRCTL_BSIZE1_SHIFT;
2970 break;
2971 }
2972
2973 ew32(PSRCTL, psrctl);
2974 }
2975
2976 /* This is useful for sniffing bad packets. */
2977 if (adapter->netdev->features & NETIF_F_RXALL) {
2978 /* UPE and MPE will be handled by normal PROMISC logic
2979 * in e1000e_set_rx_mode */
2980 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
2981 E1000_RCTL_BAM | /* RX All Bcast Pkts */
2982 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
2983
2984 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
2985 E1000_RCTL_DPF | /* Allow filtered pause */
2986 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
2987 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
2988 * and that breaks VLANs.
2989 */
2990 }
2991
2992 ew32(RCTL, rctl);
2993 /* just started the receive unit, no need to restart */
2994 adapter->flags &= ~FLAG_RX_RESTART_NOW;
2995}
2996
2997/**
2998 * e1000_configure_rx - Configure Receive Unit after Reset
2999 * @adapter: board private structure
3000 *
3001 * Configure the Rx unit of the MAC after a reset.
3002 **/
3003static void e1000_configure_rx(struct e1000_adapter *adapter)
3004{
3005 struct e1000_hw *hw = &adapter->hw;
3006 struct e1000_ring *rx_ring = adapter->rx_ring;
3007 u64 rdba;
3008 u32 rdlen, rctl, rxcsum, ctrl_ext;
3009
3010 if (adapter->rx_ps_pages) {
3011 /* this is a 32 byte descriptor */
3012 rdlen = rx_ring->count *
3013 sizeof(union e1000_rx_desc_packet_split);
3014 adapter->clean_rx = e1000_clean_rx_irq_ps;
3015 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3016 } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3017 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3018 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3019 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3020 } else {
3021 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3022 adapter->clean_rx = e1000_clean_rx_irq;
3023 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3024 }
3025
3026 /* disable receives while setting up the descriptors */
3027 rctl = er32(RCTL);
3028 if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3029 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3030 e1e_flush();
3031 usleep_range(10000, 20000);
3032
3033 if (adapter->flags2 & FLAG2_DMA_BURST) {
3034 /*
3035 * set the writeback threshold (only takes effect if the RDTR
3036 * is set). set GRAN=1 and write back up to 0x4 worth, and
3037 * enable prefetching of 0x20 Rx descriptors
3038 * granularity = 01
3039 * wthresh = 04,
3040 * hthresh = 04,
3041 * pthresh = 0x20
3042 */
3043 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3044 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3045
3046 /*
3047 * override the delay timers for enabling bursting, only if
3048 * the value was not set by the user via module options
3049 */
3050 if (adapter->rx_int_delay == DEFAULT_RDTR)
3051 adapter->rx_int_delay = BURST_RDTR;
3052 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
3053 adapter->rx_abs_int_delay = BURST_RADV;
3054 }
3055
3056 /* set the Receive Delay Timer Register */
3057 ew32(RDTR, adapter->rx_int_delay);
3058
3059 /* irq moderation */
3060 ew32(RADV, adapter->rx_abs_int_delay);
3061 if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3062 ew32(ITR, 1000000000 / (adapter->itr * 256));
3063
3064 ctrl_ext = er32(CTRL_EXT);
3065 /* Auto-Mask interrupts upon ICR access */
3066 ctrl_ext |= E1000_CTRL_EXT_IAME;
3067 ew32(IAM, 0xffffffff);
3068 ew32(CTRL_EXT, ctrl_ext);
3069 e1e_flush();
3070
3071 /*
3072 * Setup the HW Rx Head and Tail Descriptor Pointers and
3073 * the Base and Length of the Rx Descriptor Ring
3074 */
3075 rdba = rx_ring->dma;
3076 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
3077 ew32(RDBAH(0), (rdba >> 32));
3078 ew32(RDLEN(0), rdlen);
3079 ew32(RDH(0), 0);
3080 ew32(RDT(0), 0);
3081 rx_ring->head = adapter->hw.hw_addr + E1000_RDH(0);
3082 rx_ring->tail = adapter->hw.hw_addr + E1000_RDT(0);
3083
3084 /* Enable Receive Checksum Offload for TCP and UDP */
3085 rxcsum = er32(RXCSUM);
3086 if (adapter->netdev->features & NETIF_F_RXCSUM)
3087 rxcsum |= E1000_RXCSUM_TUOFL;
3088 else
3089 rxcsum &= ~E1000_RXCSUM_TUOFL;
3090 ew32(RXCSUM, rxcsum);
3091
3092 if (adapter->hw.mac.type == e1000_pch2lan) {
3093 /*
3094 * With jumbo frames, excessive C-state transition
3095 * latencies result in dropped transactions.
3096 */
3097 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3098 u32 rxdctl = er32(RXDCTL(0));
3099 ew32(RXDCTL(0), rxdctl | 0x3);
3100 pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
3101 } else {
3102 pm_qos_update_request(&adapter->netdev->pm_qos_req,
3103 PM_QOS_DEFAULT_VALUE);
3104 }
3105 }
3106
3107 /* Enable Receives */
3108 ew32(RCTL, rctl);
3109}
3110
3111/**
3112 * e1000e_write_mc_addr_list - write multicast addresses to MTA
3113 * @netdev: network interface device structure
3114 *
3115 * Writes multicast address list to the MTA hash table.
3116 * Returns: -ENOMEM on failure
3117 * 0 on no addresses written
3118 * X on writing X addresses to MTA
3119 */
3120static int e1000e_write_mc_addr_list(struct net_device *netdev)
3121{
3122 struct e1000_adapter *adapter = netdev_priv(netdev);
3123 struct e1000_hw *hw = &adapter->hw;
3124 struct netdev_hw_addr *ha;
3125 u8 *mta_list;
3126 int i;
3127
3128 if (netdev_mc_empty(netdev)) {
3129 /* nothing to program, so clear mc list */
3130 hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3131 return 0;
3132 }
3133
3134 mta_list = kzalloc(netdev_mc_count(netdev) * ETH_ALEN, GFP_ATOMIC);
3135 if (!mta_list)
3136 return -ENOMEM;
3137
3138 /* update_mc_addr_list expects a packed array of only addresses. */
3139 i = 0;
3140 netdev_for_each_mc_addr(ha, netdev)
3141 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3142
3143 hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3144 kfree(mta_list);
3145
3146 return netdev_mc_count(netdev);
3147}
3148
3149/**
3150 * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3151 * @netdev: network interface device structure
3152 *
3153 * Writes unicast address list to the RAR table.
3154 * Returns: -ENOMEM on failure/insufficient address space
3155 * 0 on no addresses written
3156 * X on writing X addresses to the RAR table
3157 **/
3158static int e1000e_write_uc_addr_list(struct net_device *netdev)
3159{
3160 struct e1000_adapter *adapter = netdev_priv(netdev);
3161 struct e1000_hw *hw = &adapter->hw;
3162 unsigned int rar_entries = hw->mac.rar_entry_count;
3163 int count = 0;
3164
3165 /* save a rar entry for our hardware address */
3166 rar_entries--;
3167
3168 /* save a rar entry for the LAA workaround */
3169 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3170 rar_entries--;
3171
3172 /* return ENOMEM indicating insufficient memory for addresses */
3173 if (netdev_uc_count(netdev) > rar_entries)
3174 return -ENOMEM;
3175
3176 if (!netdev_uc_empty(netdev) && rar_entries) {
3177 struct netdev_hw_addr *ha;
3178
3179 /*
3180 * write the addresses in reverse order to avoid write
3181 * combining
3182 */
3183 netdev_for_each_uc_addr(ha, netdev) {
3184 if (!rar_entries)
3185 break;
3186 hw->mac.ops.rar_set(hw, ha->addr, rar_entries--);
3187 count++;
3188 }
3189 }
3190
3191 /* zero out the remaining RAR entries not used above */
3192 for (; rar_entries > 0; rar_entries--) {
3193 ew32(RAH(rar_entries), 0);
3194 ew32(RAL(rar_entries), 0);
3195 }
3196 e1e_flush();
3197
3198 return count;
3199}
3200
3201/**
3202 * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3203 * @netdev: network interface device structure
3204 *
3205 * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3206 * address list or the network interface flags are updated. This routine is
3207 * responsible for configuring the hardware for proper unicast, multicast,
3208 * promiscuous mode, and all-multi behavior.
3209 **/
3210static void e1000e_set_rx_mode(struct net_device *netdev)
3211{
3212 struct e1000_adapter *adapter = netdev_priv(netdev);
3213 struct e1000_hw *hw = &adapter->hw;
3214 u32 rctl;
3215
3216 /* Check for Promiscuous and All Multicast modes */
3217 rctl = er32(RCTL);
3218
3219 /* clear the affected bits */
3220 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3221
3222 if (netdev->flags & IFF_PROMISC) {
3223 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3224 /* Do not hardware filter VLANs in promisc mode */
3225 e1000e_vlan_filter_disable(adapter);
3226 } else {
3227 int count;
3228
3229 if (netdev->flags & IFF_ALLMULTI) {
3230 rctl |= E1000_RCTL_MPE;
3231 } else {
3232 /*
3233 * Write addresses to the MTA, if the attempt fails
3234 * then we should just turn on promiscuous mode so
3235 * that we can at least receive multicast traffic
3236 */
3237 count = e1000e_write_mc_addr_list(netdev);
3238 if (count < 0)
3239 rctl |= E1000_RCTL_MPE;
3240 }
3241 e1000e_vlan_filter_enable(adapter);
3242 /*
3243 * Write addresses to available RAR registers, if there is not
3244 * sufficient space to store all the addresses then enable
3245 * unicast promiscuous mode
3246 */
3247 count = e1000e_write_uc_addr_list(netdev);
3248 if (count < 0)
3249 rctl |= E1000_RCTL_UPE;
3250 }
3251
3252 ew32(RCTL, rctl);
3253
3254 if (netdev->features & NETIF_F_HW_VLAN_RX)
3255 e1000e_vlan_strip_enable(adapter);
3256 else
3257 e1000e_vlan_strip_disable(adapter);
3258}
3259
3260static void e1000e_setup_rss_hash(struct e1000_adapter *adapter)
3261{
3262 struct e1000_hw *hw = &adapter->hw;
3263 u32 mrqc, rxcsum;
3264 int i;
3265 static const u32 rsskey[10] = {
3266 0xda565a6d, 0xc20e5b25, 0x3d256741, 0xb08fa343, 0xcb2bcad0,
3267 0xb4307bae, 0xa32dcb77, 0x0cf23080, 0x3bb7426a, 0xfa01acbe
3268 };
3269
3270 /* Fill out hash function seed */
3271 for (i = 0; i < 10; i++)
3272 ew32(RSSRK(i), rsskey[i]);
3273
3274 /* Direct all traffic to queue 0 */
3275 for (i = 0; i < 32; i++)
3276 ew32(RETA(i), 0);
3277
3278 /*
3279 * Disable raw packet checksumming so that RSS hash is placed in
3280 * descriptor on writeback.
3281 */
3282 rxcsum = er32(RXCSUM);
3283 rxcsum |= E1000_RXCSUM_PCSD;
3284
3285 ew32(RXCSUM, rxcsum);
3286
3287 mrqc = (E1000_MRQC_RSS_FIELD_IPV4 |
3288 E1000_MRQC_RSS_FIELD_IPV4_TCP |
3289 E1000_MRQC_RSS_FIELD_IPV6 |
3290 E1000_MRQC_RSS_FIELD_IPV6_TCP |
3291 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
3292
3293 ew32(MRQC, mrqc);
3294}
3295
3296/**
3297 * e1000_configure - configure the hardware for Rx and Tx
3298 * @adapter: private board structure
3299 **/
3300static void e1000_configure(struct e1000_adapter *adapter)
3301{
3302 struct e1000_ring *rx_ring = adapter->rx_ring;
3303
3304 e1000e_set_rx_mode(adapter->netdev);
3305
3306 e1000_restore_vlan(adapter);
3307 e1000_init_manageability_pt(adapter);
3308
3309 e1000_configure_tx(adapter);
3310
3311 if (adapter->netdev->features & NETIF_F_RXHASH)
3312 e1000e_setup_rss_hash(adapter);
3313 e1000_setup_rctl(adapter);
3314 e1000_configure_rx(adapter);
3315 adapter->alloc_rx_buf(rx_ring, e1000_desc_unused(rx_ring), GFP_KERNEL);
3316}
3317
3318/**
3319 * e1000e_power_up_phy - restore link in case the phy was powered down
3320 * @adapter: address of board private structure
3321 *
3322 * The phy may be powered down to save power and turn off link when the
3323 * driver is unloaded and wake on lan is not enabled (among others)
3324 * *** this routine MUST be followed by a call to e1000e_reset ***
3325 **/
3326void e1000e_power_up_phy(struct e1000_adapter *adapter)
3327{
3328 if (adapter->hw.phy.ops.power_up)
3329 adapter->hw.phy.ops.power_up(&adapter->hw);
3330
3331 adapter->hw.mac.ops.setup_link(&adapter->hw);
3332}
3333
3334/**
3335 * e1000_power_down_phy - Power down the PHY
3336 *
3337 * Power down the PHY so no link is implied when interface is down.
3338 * The PHY cannot be powered down if management or WoL is active.
3339 */
3340static void e1000_power_down_phy(struct e1000_adapter *adapter)
3341{
3342 /* WoL is enabled */
3343 if (adapter->wol)
3344 return;
3345
3346 if (adapter->hw.phy.ops.power_down)
3347 adapter->hw.phy.ops.power_down(&adapter->hw);
3348}
3349
3350/**
3351 * e1000e_reset - bring the hardware into a known good state
3352 *
3353 * This function boots the hardware and enables some settings that
3354 * require a configuration cycle of the hardware - those cannot be
3355 * set/changed during runtime. After reset the device needs to be
3356 * properly configured for Rx, Tx etc.
3357 */
3358void e1000e_reset(struct e1000_adapter *adapter)
3359{
3360 struct e1000_mac_info *mac = &adapter->hw.mac;
3361 struct e1000_fc_info *fc = &adapter->hw.fc;
3362 struct e1000_hw *hw = &adapter->hw;
3363 u32 tx_space, min_tx_space, min_rx_space;
3364 u32 pba = adapter->pba;
3365 u16 hwm;
3366
3367 /* reset Packet Buffer Allocation to default */
3368 ew32(PBA, pba);
3369
3370 if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3371 /*
3372 * To maintain wire speed transmits, the Tx FIFO should be
3373 * large enough to accommodate two full transmit packets,
3374 * rounded up to the next 1KB and expressed in KB. Likewise,
3375 * the Rx FIFO should be large enough to accommodate at least
3376 * one full receive packet and is similarly rounded up and
3377 * expressed in KB.
3378 */
3379 pba = er32(PBA);
3380 /* upper 16 bits has Tx packet buffer allocation size in KB */
3381 tx_space = pba >> 16;
3382 /* lower 16 bits has Rx packet buffer allocation size in KB */
3383 pba &= 0xffff;
3384 /*
3385 * the Tx fifo also stores 16 bytes of information about the Tx
3386 * but don't include ethernet FCS because hardware appends it
3387 */
3388 min_tx_space = (adapter->max_frame_size +
3389 sizeof(struct e1000_tx_desc) -
3390 ETH_FCS_LEN) * 2;
3391 min_tx_space = ALIGN(min_tx_space, 1024);
3392 min_tx_space >>= 10;
3393 /* software strips receive CRC, so leave room for it */
3394 min_rx_space = adapter->max_frame_size;
3395 min_rx_space = ALIGN(min_rx_space, 1024);
3396 min_rx_space >>= 10;
3397
3398 /*
3399 * If current Tx allocation is less than the min Tx FIFO size,
3400 * and the min Tx FIFO size is less than the current Rx FIFO
3401 * allocation, take space away from current Rx allocation
3402 */
3403 if ((tx_space < min_tx_space) &&
3404 ((min_tx_space - tx_space) < pba)) {
3405 pba -= min_tx_space - tx_space;
3406
3407 /*
3408 * if short on Rx space, Rx wins and must trump Tx
3409 * adjustment or use Early Receive if available
3410 */
3411 if (pba < min_rx_space)
3412 pba = min_rx_space;
3413 }
3414
3415 ew32(PBA, pba);
3416 }
3417
3418 /*
3419 * flow control settings
3420 *
3421 * The high water mark must be low enough to fit one full frame
3422 * (or the size used for early receive) above it in the Rx FIFO.
3423 * Set it to the lower of:
3424 * - 90% of the Rx FIFO size, and
3425 * - the full Rx FIFO size minus one full frame
3426 */
3427 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3428 fc->pause_time = 0xFFFF;
3429 else
3430 fc->pause_time = E1000_FC_PAUSE_TIME;
3431 fc->send_xon = true;
3432 fc->current_mode = fc->requested_mode;
3433
3434 switch (hw->mac.type) {
3435 case e1000_ich9lan:
3436 case e1000_ich10lan:
3437 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3438 pba = 14;
3439 ew32(PBA, pba);
3440 fc->high_water = 0x2800;
3441 fc->low_water = fc->high_water - 8;
3442 break;
3443 }
3444 /* fall-through */
3445 default:
3446 hwm = min(((pba << 10) * 9 / 10),
3447 ((pba << 10) - adapter->max_frame_size));
3448
3449 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3450 fc->low_water = fc->high_water - 8;
3451 break;
3452 case e1000_pchlan:
3453 /*
3454 * Workaround PCH LOM adapter hangs with certain network
3455 * loads. If hangs persist, try disabling Tx flow control.
3456 */
3457 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3458 fc->high_water = 0x3500;
3459 fc->low_water = 0x1500;
3460 } else {
3461 fc->high_water = 0x5000;
3462 fc->low_water = 0x3000;
3463 }
3464 fc->refresh_time = 0x1000;
3465 break;
3466 case e1000_pch2lan:
3467 case e1000_pch_lpt:
3468 fc->high_water = 0x05C20;
3469 fc->low_water = 0x05048;
3470 fc->pause_time = 0x0650;
3471 fc->refresh_time = 0x0400;
3472 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3473 pba = 14;
3474 ew32(PBA, pba);
3475 }
3476 break;
3477 }
3478
3479 /*
3480 * Alignment of Tx data is on an arbitrary byte boundary with the
3481 * maximum size per Tx descriptor limited only to the transmit
3482 * allocation of the packet buffer minus 96 bytes with an upper
3483 * limit of 24KB due to receive synchronization limitations.
3484 */
3485 adapter->tx_fifo_limit = min_t(u32, ((er32(PBA) >> 16) << 10) - 96,
3486 24 << 10);
3487
3488 /*
3489 * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3490 * fit in receive buffer.
3491 */
3492 if (adapter->itr_setting & 0x3) {
3493 if ((adapter->max_frame_size * 2) > (pba << 10)) {
3494 if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3495 dev_info(&adapter->pdev->dev,
3496 "Interrupt Throttle Rate turned off\n");
3497 adapter->flags2 |= FLAG2_DISABLE_AIM;
3498 ew32(ITR, 0);
3499 }
3500 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3501 dev_info(&adapter->pdev->dev,
3502 "Interrupt Throttle Rate turned on\n");
3503 adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3504 adapter->itr = 20000;
3505 ew32(ITR, 1000000000 / (adapter->itr * 256));
3506 }
3507 }
3508
3509 /* Allow time for pending master requests to run */
3510 mac->ops.reset_hw(hw);
3511
3512 /*
3513 * For parts with AMT enabled, let the firmware know
3514 * that the network interface is in control
3515 */
3516 if (adapter->flags & FLAG_HAS_AMT)
3517 e1000e_get_hw_control(adapter);
3518
3519 ew32(WUC, 0);
3520
3521 if (mac->ops.init_hw(hw))
3522 e_err("Hardware Error\n");
3523
3524 e1000_update_mng_vlan(adapter);
3525
3526 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3527 ew32(VET, ETH_P_8021Q);
3528
3529 e1000e_reset_adaptive(hw);
3530
3531 if (!netif_running(adapter->netdev) &&
3532 !test_bit(__E1000_TESTING, &adapter->state)) {
3533 e1000_power_down_phy(adapter);
3534 return;
3535 }
3536
3537 e1000_get_phy_info(hw);
3538
3539 if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3540 !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3541 u16 phy_data = 0;
3542 /*
3543 * speed up time to link by disabling smart power down, ignore
3544 * the return value of this function because there is nothing
3545 * different we would do if it failed
3546 */
3547 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3548 phy_data &= ~IGP02E1000_PM_SPD;
3549 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3550 }
3551}
3552
3553int e1000e_up(struct e1000_adapter *adapter)
3554{
3555 struct e1000_hw *hw = &adapter->hw;
3556
3557 /* hardware has been reset, we need to reload some things */
3558 e1000_configure(adapter);
3559
3560 clear_bit(__E1000_DOWN, &adapter->state);
3561
3562 if (adapter->msix_entries)
3563 e1000_configure_msix(adapter);
3564 e1000_irq_enable(adapter);
3565
3566 netif_start_queue(adapter->netdev);
3567
3568 /* fire a link change interrupt to start the watchdog */
3569 if (adapter->msix_entries)
3570 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3571 else
3572 ew32(ICS, E1000_ICS_LSC);
3573
3574 return 0;
3575}
3576
3577static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3578{
3579 struct e1000_hw *hw = &adapter->hw;
3580
3581 if (!(adapter->flags2 & FLAG2_DMA_BURST))
3582 return;
3583
3584 /* flush pending descriptor writebacks to memory */
3585 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3586 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3587
3588 /* execute the writes immediately */
3589 e1e_flush();
3590
3591 /*
3592 * due to rare timing issues, write to TIDV/RDTR again to ensure the
3593 * write is successful
3594 */
3595 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3596 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3597
3598 /* execute the writes immediately */
3599 e1e_flush();
3600}
3601
3602static void e1000e_update_stats(struct e1000_adapter *adapter);
3603
3604void e1000e_down(struct e1000_adapter *adapter)
3605{
3606 struct net_device *netdev = adapter->netdev;
3607 struct e1000_hw *hw = &adapter->hw;
3608 u32 tctl, rctl;
3609
3610 /*
3611 * signal that we're down so the interrupt handler does not
3612 * reschedule our watchdog timer
3613 */
3614 set_bit(__E1000_DOWN, &adapter->state);
3615
3616 /* disable receives in the hardware */
3617 rctl = er32(RCTL);
3618 if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3619 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3620 /* flush and sleep below */
3621
3622 netif_stop_queue(netdev);
3623
3624 /* disable transmits in the hardware */
3625 tctl = er32(TCTL);
3626 tctl &= ~E1000_TCTL_EN;
3627 ew32(TCTL, tctl);
3628
3629 /* flush both disables and wait for them to finish */
3630 e1e_flush();
3631 usleep_range(10000, 20000);
3632
3633 e1000_irq_disable(adapter);
3634
3635 del_timer_sync(&adapter->watchdog_timer);
3636 del_timer_sync(&adapter->phy_info_timer);
3637
3638 netif_carrier_off(netdev);
3639
3640 spin_lock(&adapter->stats64_lock);
3641 e1000e_update_stats(adapter);
3642 spin_unlock(&adapter->stats64_lock);
3643
3644 e1000e_flush_descriptors(adapter);
3645 e1000_clean_tx_ring(adapter->tx_ring);
3646 e1000_clean_rx_ring(adapter->rx_ring);
3647
3648 adapter->link_speed = 0;
3649 adapter->link_duplex = 0;
3650
3651 if (!pci_channel_offline(adapter->pdev))
3652 e1000e_reset(adapter);
3653
3654 /*
3655 * TODO: for power management, we could drop the link and
3656 * pci_disable_device here.
3657 */
3658}
3659
3660void e1000e_reinit_locked(struct e1000_adapter *adapter)
3661{
3662 might_sleep();
3663 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3664 usleep_range(1000, 2000);
3665 e1000e_down(adapter);
3666 e1000e_up(adapter);
3667 clear_bit(__E1000_RESETTING, &adapter->state);
3668}
3669
3670/**
3671 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3672 * @adapter: board private structure to initialize
3673 *
3674 * e1000_sw_init initializes the Adapter private data structure.
3675 * Fields are initialized based on PCI device information and
3676 * OS network device settings (MTU size).
3677 **/
3678static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3679{
3680 struct net_device *netdev = adapter->netdev;
3681
3682 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3683 adapter->rx_ps_bsize0 = 128;
3684 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3685 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3686 adapter->tx_ring_count = E1000_DEFAULT_TXD;
3687 adapter->rx_ring_count = E1000_DEFAULT_RXD;
3688
3689 spin_lock_init(&adapter->stats64_lock);
3690
3691 e1000e_set_interrupt_capability(adapter);
3692
3693 if (e1000_alloc_queues(adapter))
3694 return -ENOMEM;
3695
3696 /* Explicitly disable IRQ since the NIC can be in any state. */
3697 e1000_irq_disable(adapter);
3698
3699 set_bit(__E1000_DOWN, &adapter->state);
3700 return 0;
3701}
3702
3703/**
3704 * e1000_intr_msi_test - Interrupt Handler
3705 * @irq: interrupt number
3706 * @data: pointer to a network interface device structure
3707 **/
3708static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3709{
3710 struct net_device *netdev = data;
3711 struct e1000_adapter *adapter = netdev_priv(netdev);
3712 struct e1000_hw *hw = &adapter->hw;
3713 u32 icr = er32(ICR);
3714
3715 e_dbg("icr is %08X\n", icr);
3716 if (icr & E1000_ICR_RXSEQ) {
3717 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3718 wmb();
3719 }
3720
3721 return IRQ_HANDLED;
3722}
3723
3724/**
3725 * e1000_test_msi_interrupt - Returns 0 for successful test
3726 * @adapter: board private struct
3727 *
3728 * code flow taken from tg3.c
3729 **/
3730static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3731{
3732 struct net_device *netdev = adapter->netdev;
3733 struct e1000_hw *hw = &adapter->hw;
3734 int err;
3735
3736 /* poll_enable hasn't been called yet, so don't need disable */
3737 /* clear any pending events */
3738 er32(ICR);
3739
3740 /* free the real vector and request a test handler */
3741 e1000_free_irq(adapter);
3742 e1000e_reset_interrupt_capability(adapter);
3743
3744 /* Assume that the test fails, if it succeeds then the test
3745 * MSI irq handler will unset this flag */
3746 adapter->flags |= FLAG_MSI_TEST_FAILED;
3747
3748 err = pci_enable_msi(adapter->pdev);
3749 if (err)
3750 goto msi_test_failed;
3751
3752 err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3753 netdev->name, netdev);
3754 if (err) {
3755 pci_disable_msi(adapter->pdev);
3756 goto msi_test_failed;
3757 }
3758
3759 wmb();
3760
3761 e1000_irq_enable(adapter);
3762
3763 /* fire an unusual interrupt on the test handler */
3764 ew32(ICS, E1000_ICS_RXSEQ);
3765 e1e_flush();
3766 msleep(100);
3767
3768 e1000_irq_disable(adapter);
3769
3770 rmb();
3771
3772 if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3773 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3774 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3775 } else {
3776 e_dbg("MSI interrupt test succeeded!\n");
3777 }
3778
3779 free_irq(adapter->pdev->irq, netdev);
3780 pci_disable_msi(adapter->pdev);
3781
3782msi_test_failed:
3783 e1000e_set_interrupt_capability(adapter);
3784 return e1000_request_irq(adapter);
3785}
3786
3787/**
3788 * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3789 * @adapter: board private struct
3790 *
3791 * code flow taken from tg3.c, called with e1000 interrupts disabled.
3792 **/
3793static int e1000_test_msi(struct e1000_adapter *adapter)
3794{
3795 int err;
3796 u16 pci_cmd;
3797
3798 if (!(adapter->flags & FLAG_MSI_ENABLED))
3799 return 0;
3800
3801 /* disable SERR in case the MSI write causes a master abort */
3802 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3803 if (pci_cmd & PCI_COMMAND_SERR)
3804 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3805 pci_cmd & ~PCI_COMMAND_SERR);
3806
3807 err = e1000_test_msi_interrupt(adapter);
3808
3809 /* re-enable SERR */
3810 if (pci_cmd & PCI_COMMAND_SERR) {
3811 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3812 pci_cmd |= PCI_COMMAND_SERR;
3813 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3814 }
3815
3816 return err;
3817}
3818
3819/**
3820 * e1000_open - Called when a network interface is made active
3821 * @netdev: network interface device structure
3822 *
3823 * Returns 0 on success, negative value on failure
3824 *
3825 * The open entry point is called when a network interface is made
3826 * active by the system (IFF_UP). At this point all resources needed
3827 * for transmit and receive operations are allocated, the interrupt
3828 * handler is registered with the OS, the watchdog timer is started,
3829 * and the stack is notified that the interface is ready.
3830 **/
3831static int e1000_open(struct net_device *netdev)
3832{
3833 struct e1000_adapter *adapter = netdev_priv(netdev);
3834 struct e1000_hw *hw = &adapter->hw;
3835 struct pci_dev *pdev = adapter->pdev;
3836 int err;
3837
3838 /* disallow open during test */
3839 if (test_bit(__E1000_TESTING, &adapter->state))
3840 return -EBUSY;
3841
3842 pm_runtime_get_sync(&pdev->dev);
3843
3844 netif_carrier_off(netdev);
3845
3846 /* allocate transmit descriptors */
3847 err = e1000e_setup_tx_resources(adapter->tx_ring);
3848 if (err)
3849 goto err_setup_tx;
3850
3851 /* allocate receive descriptors */
3852 err = e1000e_setup_rx_resources(adapter->rx_ring);
3853 if (err)
3854 goto err_setup_rx;
3855
3856 /*
3857 * If AMT is enabled, let the firmware know that the network
3858 * interface is now open and reset the part to a known state.
3859 */
3860 if (adapter->flags & FLAG_HAS_AMT) {
3861 e1000e_get_hw_control(adapter);
3862 e1000e_reset(adapter);
3863 }
3864
3865 e1000e_power_up_phy(adapter);
3866
3867 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3868 if ((adapter->hw.mng_cookie.status &
3869 E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3870 e1000_update_mng_vlan(adapter);
3871
3872 /* DMA latency requirement to workaround jumbo issue */
3873 if (adapter->hw.mac.type == e1000_pch2lan)
3874 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3875 PM_QOS_CPU_DMA_LATENCY,
3876 PM_QOS_DEFAULT_VALUE);
3877
3878 /*
3879 * before we allocate an interrupt, we must be ready to handle it.
3880 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3881 * as soon as we call pci_request_irq, so we have to setup our
3882 * clean_rx handler before we do so.
3883 */
3884 e1000_configure(adapter);
3885
3886 err = e1000_request_irq(adapter);
3887 if (err)
3888 goto err_req_irq;
3889
3890 /*
3891 * Work around PCIe errata with MSI interrupts causing some chipsets to
3892 * ignore e1000e MSI messages, which means we need to test our MSI
3893 * interrupt now
3894 */
3895 if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3896 err = e1000_test_msi(adapter);
3897 if (err) {
3898 e_err("Interrupt allocation failed\n");
3899 goto err_req_irq;
3900 }
3901 }
3902
3903 /* From here on the code is the same as e1000e_up() */
3904 clear_bit(__E1000_DOWN, &adapter->state);
3905
3906 napi_enable(&adapter->napi);
3907
3908 e1000_irq_enable(adapter);
3909
3910 adapter->tx_hang_recheck = false;
3911 netif_start_queue(netdev);
3912
3913 adapter->idle_check = true;
3914 pm_runtime_put(&pdev->dev);
3915
3916 /* fire a link status change interrupt to start the watchdog */
3917 if (adapter->msix_entries)
3918 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3919 else
3920 ew32(ICS, E1000_ICS_LSC);
3921
3922 return 0;
3923
3924err_req_irq:
3925 e1000e_release_hw_control(adapter);
3926 e1000_power_down_phy(adapter);
3927 e1000e_free_rx_resources(adapter->rx_ring);
3928err_setup_rx:
3929 e1000e_free_tx_resources(adapter->tx_ring);
3930err_setup_tx:
3931 e1000e_reset(adapter);
3932 pm_runtime_put_sync(&pdev->dev);
3933
3934 return err;
3935}
3936
3937/**
3938 * e1000_close - Disables a network interface
3939 * @netdev: network interface device structure
3940 *
3941 * Returns 0, this is not allowed to fail
3942 *
3943 * The close entry point is called when an interface is de-activated
3944 * by the OS. The hardware is still under the drivers control, but
3945 * needs to be disabled. A global MAC reset is issued to stop the
3946 * hardware, and all transmit and receive resources are freed.
3947 **/
3948static int e1000_close(struct net_device *netdev)
3949{
3950 struct e1000_adapter *adapter = netdev_priv(netdev);
3951 struct pci_dev *pdev = adapter->pdev;
3952 int count = E1000_CHECK_RESET_COUNT;
3953
3954 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
3955 usleep_range(10000, 20000);
3956
3957 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3958
3959 pm_runtime_get_sync(&pdev->dev);
3960
3961 napi_disable(&adapter->napi);
3962
3963 if (!test_bit(__E1000_DOWN, &adapter->state)) {
3964 e1000e_down(adapter);
3965 e1000_free_irq(adapter);
3966 }
3967 e1000_power_down_phy(adapter);
3968
3969 e1000e_free_tx_resources(adapter->tx_ring);
3970 e1000e_free_rx_resources(adapter->rx_ring);
3971
3972 /*
3973 * kill manageability vlan ID if supported, but not if a vlan with
3974 * the same ID is registered on the host OS (let 8021q kill it)
3975 */
3976 if (adapter->hw.mng_cookie.status &
3977 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
3978 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3979
3980 /*
3981 * If AMT is enabled, let the firmware know that the network
3982 * interface is now closed
3983 */
3984 if ((adapter->flags & FLAG_HAS_AMT) &&
3985 !test_bit(__E1000_TESTING, &adapter->state))
3986 e1000e_release_hw_control(adapter);
3987
3988 if (adapter->hw.mac.type == e1000_pch2lan)
3989 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3990
3991 pm_runtime_put_sync(&pdev->dev);
3992
3993 return 0;
3994}
3995/**
3996 * e1000_set_mac - Change the Ethernet Address of the NIC
3997 * @netdev: network interface device structure
3998 * @p: pointer to an address structure
3999 *
4000 * Returns 0 on success, negative on failure
4001 **/
4002static int e1000_set_mac(struct net_device *netdev, void *p)
4003{
4004 struct e1000_adapter *adapter = netdev_priv(netdev);
4005 struct e1000_hw *hw = &adapter->hw;
4006 struct sockaddr *addr = p;
4007
4008 if (!is_valid_ether_addr(addr->sa_data))
4009 return -EADDRNOTAVAIL;
4010
4011 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4012 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
4013
4014 hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
4015
4016 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
4017 /* activate the work around */
4018 e1000e_set_laa_state_82571(&adapter->hw, 1);
4019
4020 /*
4021 * Hold a copy of the LAA in RAR[14] This is done so that
4022 * between the time RAR[0] gets clobbered and the time it
4023 * gets fixed (in e1000_watchdog), the actual LAA is in one
4024 * of the RARs and no incoming packets directed to this port
4025 * are dropped. Eventually the LAA will be in RAR[0] and
4026 * RAR[14]
4027 */
4028 hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr,
4029 adapter->hw.mac.rar_entry_count - 1);
4030 }
4031
4032 return 0;
4033}
4034
4035/**
4036 * e1000e_update_phy_task - work thread to update phy
4037 * @work: pointer to our work struct
4038 *
4039 * this worker thread exists because we must acquire a
4040 * semaphore to read the phy, which we could msleep while
4041 * waiting for it, and we can't msleep in a timer.
4042 **/
4043static void e1000e_update_phy_task(struct work_struct *work)
4044{
4045 struct e1000_adapter *adapter = container_of(work,
4046 struct e1000_adapter, update_phy_task);
4047
4048 if (test_bit(__E1000_DOWN, &adapter->state))
4049 return;
4050
4051 e1000_get_phy_info(&adapter->hw);
4052}
4053
4054/*
4055 * Need to wait a few seconds after link up to get diagnostic information from
4056 * the phy
4057 */
4058static void e1000_update_phy_info(unsigned long data)
4059{
4060 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4061
4062 if (test_bit(__E1000_DOWN, &adapter->state))
4063 return;
4064
4065 schedule_work(&adapter->update_phy_task);
4066}
4067
4068/**
4069 * e1000e_update_phy_stats - Update the PHY statistics counters
4070 * @adapter: board private structure
4071 *
4072 * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4073 **/
4074static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4075{
4076 struct e1000_hw *hw = &adapter->hw;
4077 s32 ret_val;
4078 u16 phy_data;
4079
4080 ret_val = hw->phy.ops.acquire(hw);
4081 if (ret_val)
4082 return;
4083
4084 /*
4085 * A page set is expensive so check if already on desired page.
4086 * If not, set to the page with the PHY status registers.
4087 */
4088 hw->phy.addr = 1;
4089 ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4090 &phy_data);
4091 if (ret_val)
4092 goto release;
4093 if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4094 ret_val = hw->phy.ops.set_page(hw,
4095 HV_STATS_PAGE << IGP_PAGE_SHIFT);
4096 if (ret_val)
4097 goto release;
4098 }
4099
4100 /* Single Collision Count */
4101 hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4102 ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4103 if (!ret_val)
4104 adapter->stats.scc += phy_data;
4105
4106 /* Excessive Collision Count */
4107 hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4108 ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4109 if (!ret_val)
4110 adapter->stats.ecol += phy_data;
4111
4112 /* Multiple Collision Count */
4113 hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4114 ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4115 if (!ret_val)
4116 adapter->stats.mcc += phy_data;
4117
4118 /* Late Collision Count */
4119 hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4120 ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4121 if (!ret_val)
4122 adapter->stats.latecol += phy_data;
4123
4124 /* Collision Count - also used for adaptive IFS */
4125 hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4126 ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4127 if (!ret_val)
4128 hw->mac.collision_delta = phy_data;
4129
4130 /* Defer Count */
4131 hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4132 ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4133 if (!ret_val)
4134 adapter->stats.dc += phy_data;
4135
4136 /* Transmit with no CRS */
4137 hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4138 ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4139 if (!ret_val)
4140 adapter->stats.tncrs += phy_data;
4141
4142release:
4143 hw->phy.ops.release(hw);
4144}
4145
4146/**
4147 * e1000e_update_stats - Update the board statistics counters
4148 * @adapter: board private structure
4149 **/
4150static void e1000e_update_stats(struct e1000_adapter *adapter)
4151{
4152 struct net_device *netdev = adapter->netdev;
4153 struct e1000_hw *hw = &adapter->hw;
4154 struct pci_dev *pdev = adapter->pdev;
4155
4156 /*
4157 * Prevent stats update while adapter is being reset, or if the pci
4158 * connection is down.
4159 */
4160 if (adapter->link_speed == 0)
4161 return;
4162 if (pci_channel_offline(pdev))
4163 return;
4164
4165 adapter->stats.crcerrs += er32(CRCERRS);
4166 adapter->stats.gprc += er32(GPRC);
4167 adapter->stats.gorc += er32(GORCL);
4168 er32(GORCH); /* Clear gorc */
4169 adapter->stats.bprc += er32(BPRC);
4170 adapter->stats.mprc += er32(MPRC);
4171 adapter->stats.roc += er32(ROC);
4172
4173 adapter->stats.mpc += er32(MPC);
4174
4175 /* Half-duplex statistics */
4176 if (adapter->link_duplex == HALF_DUPLEX) {
4177 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4178 e1000e_update_phy_stats(adapter);
4179 } else {
4180 adapter->stats.scc += er32(SCC);
4181 adapter->stats.ecol += er32(ECOL);
4182 adapter->stats.mcc += er32(MCC);
4183 adapter->stats.latecol += er32(LATECOL);
4184 adapter->stats.dc += er32(DC);
4185
4186 hw->mac.collision_delta = er32(COLC);
4187
4188 if ((hw->mac.type != e1000_82574) &&
4189 (hw->mac.type != e1000_82583))
4190 adapter->stats.tncrs += er32(TNCRS);
4191 }
4192 adapter->stats.colc += hw->mac.collision_delta;
4193 }
4194
4195 adapter->stats.xonrxc += er32(XONRXC);
4196 adapter->stats.xontxc += er32(XONTXC);
4197 adapter->stats.xoffrxc += er32(XOFFRXC);
4198 adapter->stats.xofftxc += er32(XOFFTXC);
4199 adapter->stats.gptc += er32(GPTC);
4200 adapter->stats.gotc += er32(GOTCL);
4201 er32(GOTCH); /* Clear gotc */
4202 adapter->stats.rnbc += er32(RNBC);
4203 adapter->stats.ruc += er32(RUC);
4204
4205 adapter->stats.mptc += er32(MPTC);
4206 adapter->stats.bptc += er32(BPTC);
4207
4208 /* used for adaptive IFS */
4209
4210 hw->mac.tx_packet_delta = er32(TPT);
4211 adapter->stats.tpt += hw->mac.tx_packet_delta;
4212
4213 adapter->stats.algnerrc += er32(ALGNERRC);
4214 adapter->stats.rxerrc += er32(RXERRC);
4215 adapter->stats.cexterr += er32(CEXTERR);
4216 adapter->stats.tsctc += er32(TSCTC);
4217 adapter->stats.tsctfc += er32(TSCTFC);
4218
4219 /* Fill out the OS statistics structure */
4220 netdev->stats.multicast = adapter->stats.mprc;
4221 netdev->stats.collisions = adapter->stats.colc;
4222
4223 /* Rx Errors */
4224
4225 /*
4226 * RLEC on some newer hardware can be incorrect so build
4227 * our own version based on RUC and ROC
4228 */
4229 netdev->stats.rx_errors = adapter->stats.rxerrc +
4230 adapter->stats.crcerrs + adapter->stats.algnerrc +
4231 adapter->stats.ruc + adapter->stats.roc +
4232 adapter->stats.cexterr;
4233 netdev->stats.rx_length_errors = adapter->stats.ruc +
4234 adapter->stats.roc;
4235 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4236 netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4237 netdev->stats.rx_missed_errors = adapter->stats.mpc;
4238
4239 /* Tx Errors */
4240 netdev->stats.tx_errors = adapter->stats.ecol +
4241 adapter->stats.latecol;
4242 netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4243 netdev->stats.tx_window_errors = adapter->stats.latecol;
4244 netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4245
4246 /* Tx Dropped needs to be maintained elsewhere */
4247
4248 /* Management Stats */
4249 adapter->stats.mgptc += er32(MGTPTC);
4250 adapter->stats.mgprc += er32(MGTPRC);
4251 adapter->stats.mgpdc += er32(MGTPDC);
4252}
4253
4254/**
4255 * e1000_phy_read_status - Update the PHY register status snapshot
4256 * @adapter: board private structure
4257 **/
4258static void e1000_phy_read_status(struct e1000_adapter *adapter)
4259{
4260 struct e1000_hw *hw = &adapter->hw;
4261 struct e1000_phy_regs *phy = &adapter->phy_regs;
4262
4263 if ((er32(STATUS) & E1000_STATUS_LU) &&
4264 (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4265 int ret_val;
4266
4267 ret_val = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4268 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4269 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4270 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4271 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4272 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4273 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4274 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4275 if (ret_val)
4276 e_warn("Error reading PHY register\n");
4277 } else {
4278 /*
4279 * Do not read PHY registers if link is not up
4280 * Set values to typical power-on defaults
4281 */
4282 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4283 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4284 BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4285 BMSR_ERCAP);
4286 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4287 ADVERTISE_ALL | ADVERTISE_CSMA);
4288 phy->lpa = 0;
4289 phy->expansion = EXPANSION_ENABLENPAGE;
4290 phy->ctrl1000 = ADVERTISE_1000FULL;
4291 phy->stat1000 = 0;
4292 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4293 }
4294}
4295
4296static void e1000_print_link_info(struct e1000_adapter *adapter)
4297{
4298 struct e1000_hw *hw = &adapter->hw;
4299 u32 ctrl = er32(CTRL);
4300
4301 /* Link status message must follow this format for user tools */
4302 printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4303 adapter->netdev->name,
4304 adapter->link_speed,
4305 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
4306 (ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
4307 (ctrl & E1000_CTRL_RFCE) ? "Rx" :
4308 (ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
4309}
4310
4311static bool e1000e_has_link(struct e1000_adapter *adapter)
4312{
4313 struct e1000_hw *hw = &adapter->hw;
4314 bool link_active = false;
4315 s32 ret_val = 0;
4316
4317 /*
4318 * get_link_status is set on LSC (link status) interrupt or
4319 * Rx sequence error interrupt. get_link_status will stay
4320 * false until the check_for_link establishes link
4321 * for copper adapters ONLY
4322 */
4323 switch (hw->phy.media_type) {
4324 case e1000_media_type_copper:
4325 if (hw->mac.get_link_status) {
4326 ret_val = hw->mac.ops.check_for_link(hw);
4327 link_active = !hw->mac.get_link_status;
4328 } else {
4329 link_active = true;
4330 }
4331 break;
4332 case e1000_media_type_fiber:
4333 ret_val = hw->mac.ops.check_for_link(hw);
4334 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4335 break;
4336 case e1000_media_type_internal_serdes:
4337 ret_val = hw->mac.ops.check_for_link(hw);
4338 link_active = adapter->hw.mac.serdes_has_link;
4339 break;
4340 default:
4341 case e1000_media_type_unknown:
4342 break;
4343 }
4344
4345 if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4346 (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4347 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4348 e_info("Gigabit has been disabled, downgrading speed\n");
4349 }
4350
4351 return link_active;
4352}
4353
4354static void e1000e_enable_receives(struct e1000_adapter *adapter)
4355{
4356 /* make sure the receive unit is started */
4357 if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4358 (adapter->flags & FLAG_RX_RESTART_NOW)) {
4359 struct e1000_hw *hw = &adapter->hw;
4360 u32 rctl = er32(RCTL);
4361 ew32(RCTL, rctl | E1000_RCTL_EN);
4362 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4363 }
4364}
4365
4366static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4367{
4368 struct e1000_hw *hw = &adapter->hw;
4369
4370 /*
4371 * With 82574 controllers, PHY needs to be checked periodically
4372 * for hung state and reset, if two calls return true
4373 */
4374 if (e1000_check_phy_82574(hw))
4375 adapter->phy_hang_count++;
4376 else
4377 adapter->phy_hang_count = 0;
4378
4379 if (adapter->phy_hang_count > 1) {
4380 adapter->phy_hang_count = 0;
4381 schedule_work(&adapter->reset_task);
4382 }
4383}
4384
4385/**
4386 * e1000_watchdog - Timer Call-back
4387 * @data: pointer to adapter cast into an unsigned long
4388 **/
4389static void e1000_watchdog(unsigned long data)
4390{
4391 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4392
4393 /* Do the rest outside of interrupt context */
4394 schedule_work(&adapter->watchdog_task);
4395
4396 /* TODO: make this use queue_delayed_work() */
4397}
4398
4399static void e1000_watchdog_task(struct work_struct *work)
4400{
4401 struct e1000_adapter *adapter = container_of(work,
4402 struct e1000_adapter, watchdog_task);
4403 struct net_device *netdev = adapter->netdev;
4404 struct e1000_mac_info *mac = &adapter->hw.mac;
4405 struct e1000_phy_info *phy = &adapter->hw.phy;
4406 struct e1000_ring *tx_ring = adapter->tx_ring;
4407 struct e1000_hw *hw = &adapter->hw;
4408 u32 link, tctl;
4409
4410 if (test_bit(__E1000_DOWN, &adapter->state))
4411 return;
4412
4413 link = e1000e_has_link(adapter);
4414 if ((netif_carrier_ok(netdev)) && link) {
4415 /* Cancel scheduled suspend requests. */
4416 pm_runtime_resume(netdev->dev.parent);
4417
4418 e1000e_enable_receives(adapter);
4419 goto link_up;
4420 }
4421
4422 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4423 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4424 e1000_update_mng_vlan(adapter);
4425
4426 if (link) {
4427 if (!netif_carrier_ok(netdev)) {
4428 bool txb2b = true;
4429
4430 /* Cancel scheduled suspend requests. */
4431 pm_runtime_resume(netdev->dev.parent);
4432
4433 /* update snapshot of PHY registers on LSC */
4434 e1000_phy_read_status(adapter);
4435 mac->ops.get_link_up_info(&adapter->hw,
4436 &adapter->link_speed,
4437 &adapter->link_duplex);
4438 e1000_print_link_info(adapter);
4439 /*
4440 * On supported PHYs, check for duplex mismatch only
4441 * if link has autonegotiated at 10/100 half
4442 */
4443 if ((hw->phy.type == e1000_phy_igp_3 ||
4444 hw->phy.type == e1000_phy_bm) &&
4445 (hw->mac.autoneg == true) &&
4446 (adapter->link_speed == SPEED_10 ||
4447 adapter->link_speed == SPEED_100) &&
4448 (adapter->link_duplex == HALF_DUPLEX)) {
4449 u16 autoneg_exp;
4450
4451 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4452
4453 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4454 e_info("Autonegotiated half duplex but link partner cannot autoneg. Try forcing full duplex if link gets many collisions.\n");
4455 }
4456
4457 /* adjust timeout factor according to speed/duplex */
4458 adapter->tx_timeout_factor = 1;
4459 switch (adapter->link_speed) {
4460 case SPEED_10:
4461 txb2b = false;
4462 adapter->tx_timeout_factor = 16;
4463 break;
4464 case SPEED_100:
4465 txb2b = false;
4466 adapter->tx_timeout_factor = 10;
4467 break;
4468 }
4469
4470 /*
4471 * workaround: re-program speed mode bit after
4472 * link-up event
4473 */
4474 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4475 !txb2b) {
4476 u32 tarc0;
4477 tarc0 = er32(TARC(0));
4478 tarc0 &= ~SPEED_MODE_BIT;
4479 ew32(TARC(0), tarc0);
4480 }
4481
4482 /*
4483 * disable TSO for pcie and 10/100 speeds, to avoid
4484 * some hardware issues
4485 */
4486 if (!(adapter->flags & FLAG_TSO_FORCE)) {
4487 switch (adapter->link_speed) {
4488 case SPEED_10:
4489 case SPEED_100:
4490 e_info("10/100 speed: disabling TSO\n");
4491 netdev->features &= ~NETIF_F_TSO;
4492 netdev->features &= ~NETIF_F_TSO6;
4493 break;
4494 case SPEED_1000:
4495 netdev->features |= NETIF_F_TSO;
4496 netdev->features |= NETIF_F_TSO6;
4497 break;
4498 default:
4499 /* oops */
4500 break;
4501 }
4502 }
4503
4504 /*
4505 * enable transmits in the hardware, need to do this
4506 * after setting TARC(0)
4507 */
4508 tctl = er32(TCTL);
4509 tctl |= E1000_TCTL_EN;
4510 ew32(TCTL, tctl);
4511
4512 /*
4513 * Perform any post-link-up configuration before
4514 * reporting link up.
4515 */
4516 if (phy->ops.cfg_on_link_up)
4517 phy->ops.cfg_on_link_up(hw);
4518
4519 netif_carrier_on(netdev);
4520
4521 if (!test_bit(__E1000_DOWN, &adapter->state))
4522 mod_timer(&adapter->phy_info_timer,
4523 round_jiffies(jiffies + 2 * HZ));
4524 }
4525 } else {
4526 if (netif_carrier_ok(netdev)) {
4527 adapter->link_speed = 0;
4528 adapter->link_duplex = 0;
4529 /* Link status message must follow this format */
4530 printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4531 adapter->netdev->name);
4532 netif_carrier_off(netdev);
4533 if (!test_bit(__E1000_DOWN, &adapter->state))
4534 mod_timer(&adapter->phy_info_timer,
4535 round_jiffies(jiffies + 2 * HZ));
4536
4537 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4538 schedule_work(&adapter->reset_task);
4539 else
4540 pm_schedule_suspend(netdev->dev.parent,
4541 LINK_TIMEOUT);
4542 }
4543 }
4544
4545link_up:
4546 spin_lock(&adapter->stats64_lock);
4547 e1000e_update_stats(adapter);
4548
4549 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4550 adapter->tpt_old = adapter->stats.tpt;
4551 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4552 adapter->colc_old = adapter->stats.colc;
4553
4554 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4555 adapter->gorc_old = adapter->stats.gorc;
4556 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4557 adapter->gotc_old = adapter->stats.gotc;
4558 spin_unlock(&adapter->stats64_lock);
4559
4560 e1000e_update_adaptive(&adapter->hw);
4561
4562 if (!netif_carrier_ok(netdev) &&
4563 (e1000_desc_unused(tx_ring) + 1 < tx_ring->count)) {
4564 /*
4565 * We've lost link, so the controller stops DMA,
4566 * but we've got queued Tx work that's never going
4567 * to get done, so reset controller to flush Tx.
4568 * (Do the reset outside of interrupt context).
4569 */
4570 schedule_work(&adapter->reset_task);
4571 /* return immediately since reset is imminent */
4572 return;
4573 }
4574
4575 /* Simple mode for Interrupt Throttle Rate (ITR) */
4576 if (adapter->itr_setting == 4) {
4577 /*
4578 * Symmetric Tx/Rx gets a reduced ITR=2000;
4579 * Total asymmetrical Tx or Rx gets ITR=8000;
4580 * everyone else is between 2000-8000.
4581 */
4582 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4583 u32 dif = (adapter->gotc > adapter->gorc ?
4584 adapter->gotc - adapter->gorc :
4585 adapter->gorc - adapter->gotc) / 10000;
4586 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4587
4588 ew32(ITR, 1000000000 / (itr * 256));
4589 }
4590
4591 /* Cause software interrupt to ensure Rx ring is cleaned */
4592 if (adapter->msix_entries)
4593 ew32(ICS, adapter->rx_ring->ims_val);
4594 else
4595 ew32(ICS, E1000_ICS_RXDMT0);
4596
4597 /* flush pending descriptors to memory before detecting Tx hang */
4598 e1000e_flush_descriptors(adapter);
4599
4600 /* Force detection of hung controller every watchdog period */
4601 adapter->detect_tx_hung = true;
4602
4603 /*
4604 * With 82571 controllers, LAA may be overwritten due to controller
4605 * reset from the other port. Set the appropriate LAA in RAR[0]
4606 */
4607 if (e1000e_get_laa_state_82571(hw))
4608 hw->mac.ops.rar_set(hw, adapter->hw.mac.addr, 0);
4609
4610 if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4611 e1000e_check_82574_phy_workaround(adapter);
4612
4613 /* Reset the timer */
4614 if (!test_bit(__E1000_DOWN, &adapter->state))
4615 mod_timer(&adapter->watchdog_timer,
4616 round_jiffies(jiffies + 2 * HZ));
4617}
4618
4619#define E1000_TX_FLAGS_CSUM 0x00000001
4620#define E1000_TX_FLAGS_VLAN 0x00000002
4621#define E1000_TX_FLAGS_TSO 0x00000004
4622#define E1000_TX_FLAGS_IPV4 0x00000008
4623#define E1000_TX_FLAGS_NO_FCS 0x00000010
4624#define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
4625#define E1000_TX_FLAGS_VLAN_SHIFT 16
4626
4627static int e1000_tso(struct e1000_ring *tx_ring, struct sk_buff *skb)
4628{
4629 struct e1000_context_desc *context_desc;
4630 struct e1000_buffer *buffer_info;
4631 unsigned int i;
4632 u32 cmd_length = 0;
4633 u16 ipcse = 0, tucse, mss;
4634 u8 ipcss, ipcso, tucss, tucso, hdr_len;
4635
4636 if (!skb_is_gso(skb))
4637 return 0;
4638
4639 if (skb_header_cloned(skb)) {
4640 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4641
4642 if (err)
4643 return err;
4644 }
4645
4646 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4647 mss = skb_shinfo(skb)->gso_size;
4648 if (skb->protocol == htons(ETH_P_IP)) {
4649 struct iphdr *iph = ip_hdr(skb);
4650 iph->tot_len = 0;
4651 iph->check = 0;
4652 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4653 0, IPPROTO_TCP, 0);
4654 cmd_length = E1000_TXD_CMD_IP;
4655 ipcse = skb_transport_offset(skb) - 1;
4656 } else if (skb_is_gso_v6(skb)) {
4657 ipv6_hdr(skb)->payload_len = 0;
4658 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4659 &ipv6_hdr(skb)->daddr,
4660 0, IPPROTO_TCP, 0);
4661 ipcse = 0;
4662 }
4663 ipcss = skb_network_offset(skb);
4664 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4665 tucss = skb_transport_offset(skb);
4666 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4667 tucse = 0;
4668
4669 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4670 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4671
4672 i = tx_ring->next_to_use;
4673 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4674 buffer_info = &tx_ring->buffer_info[i];
4675
4676 context_desc->lower_setup.ip_fields.ipcss = ipcss;
4677 context_desc->lower_setup.ip_fields.ipcso = ipcso;
4678 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
4679 context_desc->upper_setup.tcp_fields.tucss = tucss;
4680 context_desc->upper_setup.tcp_fields.tucso = tucso;
4681 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4682 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
4683 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4684 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4685
4686 buffer_info->time_stamp = jiffies;
4687 buffer_info->next_to_watch = i;
4688
4689 i++;
4690 if (i == tx_ring->count)
4691 i = 0;
4692 tx_ring->next_to_use = i;
4693
4694 return 1;
4695}
4696
4697static bool e1000_tx_csum(struct e1000_ring *tx_ring, struct sk_buff *skb)
4698{
4699 struct e1000_adapter *adapter = tx_ring->adapter;
4700 struct e1000_context_desc *context_desc;
4701 struct e1000_buffer *buffer_info;
4702 unsigned int i;
4703 u8 css;
4704 u32 cmd_len = E1000_TXD_CMD_DEXT;
4705 __be16 protocol;
4706
4707 if (skb->ip_summed != CHECKSUM_PARTIAL)
4708 return 0;
4709
4710 if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4711 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4712 else
4713 protocol = skb->protocol;
4714
4715 switch (protocol) {
4716 case cpu_to_be16(ETH_P_IP):
4717 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4718 cmd_len |= E1000_TXD_CMD_TCP;
4719 break;
4720 case cpu_to_be16(ETH_P_IPV6):
4721 /* XXX not handling all IPV6 headers */
4722 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4723 cmd_len |= E1000_TXD_CMD_TCP;
4724 break;
4725 default:
4726 if (unlikely(net_ratelimit()))
4727 e_warn("checksum_partial proto=%x!\n",
4728 be16_to_cpu(protocol));
4729 break;
4730 }
4731
4732 css = skb_checksum_start_offset(skb);
4733
4734 i = tx_ring->next_to_use;
4735 buffer_info = &tx_ring->buffer_info[i];
4736 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4737
4738 context_desc->lower_setup.ip_config = 0;
4739 context_desc->upper_setup.tcp_fields.tucss = css;
4740 context_desc->upper_setup.tcp_fields.tucso =
4741 css + skb->csum_offset;
4742 context_desc->upper_setup.tcp_fields.tucse = 0;
4743 context_desc->tcp_seg_setup.data = 0;
4744 context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4745
4746 buffer_info->time_stamp = jiffies;
4747 buffer_info->next_to_watch = i;
4748
4749 i++;
4750 if (i == tx_ring->count)
4751 i = 0;
4752 tx_ring->next_to_use = i;
4753
4754 return 1;
4755}
4756
4757static int e1000_tx_map(struct e1000_ring *tx_ring, struct sk_buff *skb,
4758 unsigned int first, unsigned int max_per_txd,
4759 unsigned int nr_frags)
4760{
4761 struct e1000_adapter *adapter = tx_ring->adapter;
4762 struct pci_dev *pdev = adapter->pdev;
4763 struct e1000_buffer *buffer_info;
4764 unsigned int len = skb_headlen(skb);
4765 unsigned int offset = 0, size, count = 0, i;
4766 unsigned int f, bytecount, segs;
4767
4768 i = tx_ring->next_to_use;
4769
4770 while (len) {
4771 buffer_info = &tx_ring->buffer_info[i];
4772 size = min(len, max_per_txd);
4773
4774 buffer_info->length = size;
4775 buffer_info->time_stamp = jiffies;
4776 buffer_info->next_to_watch = i;
4777 buffer_info->dma = dma_map_single(&pdev->dev,
4778 skb->data + offset,
4779 size, DMA_TO_DEVICE);
4780 buffer_info->mapped_as_page = false;
4781 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4782 goto dma_error;
4783
4784 len -= size;
4785 offset += size;
4786 count++;
4787
4788 if (len) {
4789 i++;
4790 if (i == tx_ring->count)
4791 i = 0;
4792 }
4793 }
4794
4795 for (f = 0; f < nr_frags; f++) {
4796 const struct skb_frag_struct *frag;
4797
4798 frag = &skb_shinfo(skb)->frags[f];
4799 len = skb_frag_size(frag);
4800 offset = 0;
4801
4802 while (len) {
4803 i++;
4804 if (i == tx_ring->count)
4805 i = 0;
4806
4807 buffer_info = &tx_ring->buffer_info[i];
4808 size = min(len, max_per_txd);
4809
4810 buffer_info->length = size;
4811 buffer_info->time_stamp = jiffies;
4812 buffer_info->next_to_watch = i;
4813 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
4814 offset, size, DMA_TO_DEVICE);
4815 buffer_info->mapped_as_page = true;
4816 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4817 goto dma_error;
4818
4819 len -= size;
4820 offset += size;
4821 count++;
4822 }
4823 }
4824
4825 segs = skb_shinfo(skb)->gso_segs ? : 1;
4826 /* multiply data chunks by size of headers */
4827 bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4828
4829 tx_ring->buffer_info[i].skb = skb;
4830 tx_ring->buffer_info[i].segs = segs;
4831 tx_ring->buffer_info[i].bytecount = bytecount;
4832 tx_ring->buffer_info[first].next_to_watch = i;
4833
4834 return count;
4835
4836dma_error:
4837 dev_err(&pdev->dev, "Tx DMA map failed\n");
4838 buffer_info->dma = 0;
4839 if (count)
4840 count--;
4841
4842 while (count--) {
4843 if (i == 0)
4844 i += tx_ring->count;
4845 i--;
4846 buffer_info = &tx_ring->buffer_info[i];
4847 e1000_put_txbuf(tx_ring, buffer_info);
4848 }
4849
4850 return 0;
4851}
4852
4853static void e1000_tx_queue(struct e1000_ring *tx_ring, int tx_flags, int count)
4854{
4855 struct e1000_adapter *adapter = tx_ring->adapter;
4856 struct e1000_tx_desc *tx_desc = NULL;
4857 struct e1000_buffer *buffer_info;
4858 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4859 unsigned int i;
4860
4861 if (tx_flags & E1000_TX_FLAGS_TSO) {
4862 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4863 E1000_TXD_CMD_TSE;
4864 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4865
4866 if (tx_flags & E1000_TX_FLAGS_IPV4)
4867 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4868 }
4869
4870 if (tx_flags & E1000_TX_FLAGS_CSUM) {
4871 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4872 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4873 }
4874
4875 if (tx_flags & E1000_TX_FLAGS_VLAN) {
4876 txd_lower |= E1000_TXD_CMD_VLE;
4877 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4878 }
4879
4880 if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
4881 txd_lower &= ~(E1000_TXD_CMD_IFCS);
4882
4883 i = tx_ring->next_to_use;
4884
4885 do {
4886 buffer_info = &tx_ring->buffer_info[i];
4887 tx_desc = E1000_TX_DESC(*tx_ring, i);
4888 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4889 tx_desc->lower.data =
4890 cpu_to_le32(txd_lower | buffer_info->length);
4891 tx_desc->upper.data = cpu_to_le32(txd_upper);
4892
4893 i++;
4894 if (i == tx_ring->count)
4895 i = 0;
4896 } while (--count > 0);
4897
4898 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4899
4900 /* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
4901 if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
4902 tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
4903
4904 /*
4905 * Force memory writes to complete before letting h/w
4906 * know there are new descriptors to fetch. (Only
4907 * applicable for weak-ordered memory model archs,
4908 * such as IA-64).
4909 */
4910 wmb();
4911
4912 tx_ring->next_to_use = i;
4913
4914 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
4915 e1000e_update_tdt_wa(tx_ring, i);
4916 else
4917 writel(i, tx_ring->tail);
4918
4919 /*
4920 * we need this if more than one processor can write to our tail
4921 * at a time, it synchronizes IO on IA64/Altix systems
4922 */
4923 mmiowb();
4924}
4925
4926#define MINIMUM_DHCP_PACKET_SIZE 282
4927static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4928 struct sk_buff *skb)
4929{
4930 struct e1000_hw *hw = &adapter->hw;
4931 u16 length, offset;
4932
4933 if (vlan_tx_tag_present(skb)) {
4934 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4935 (adapter->hw.mng_cookie.status &
4936 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4937 return 0;
4938 }
4939
4940 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4941 return 0;
4942
4943 if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4944 return 0;
4945
4946 {
4947 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4948 struct udphdr *udp;
4949
4950 if (ip->protocol != IPPROTO_UDP)
4951 return 0;
4952
4953 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4954 if (ntohs(udp->dest) != 67)
4955 return 0;
4956
4957 offset = (u8 *)udp + 8 - skb->data;
4958 length = skb->len - offset;
4959 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4960 }
4961
4962 return 0;
4963}
4964
4965static int __e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
4966{
4967 struct e1000_adapter *adapter = tx_ring->adapter;
4968
4969 netif_stop_queue(adapter->netdev);
4970 /*
4971 * Herbert's original patch had:
4972 * smp_mb__after_netif_stop_queue();
4973 * but since that doesn't exist yet, just open code it.
4974 */
4975 smp_mb();
4976
4977 /*
4978 * We need to check again in a case another CPU has just
4979 * made room available.
4980 */
4981 if (e1000_desc_unused(tx_ring) < size)
4982 return -EBUSY;
4983
4984 /* A reprieve! */
4985 netif_start_queue(adapter->netdev);
4986 ++adapter->restart_queue;
4987 return 0;
4988}
4989
4990static int e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
4991{
4992 BUG_ON(size > tx_ring->count);
4993
4994 if (e1000_desc_unused(tx_ring) >= size)
4995 return 0;
4996 return __e1000_maybe_stop_tx(tx_ring, size);
4997}
4998
4999static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
5000 struct net_device *netdev)
5001{
5002 struct e1000_adapter *adapter = netdev_priv(netdev);
5003 struct e1000_ring *tx_ring = adapter->tx_ring;
5004 unsigned int first;
5005 unsigned int tx_flags = 0;
5006 unsigned int len = skb_headlen(skb);
5007 unsigned int nr_frags;
5008 unsigned int mss;
5009 int count = 0;
5010 int tso;
5011 unsigned int f;
5012
5013 if (test_bit(__E1000_DOWN, &adapter->state)) {
5014 dev_kfree_skb_any(skb);
5015 return NETDEV_TX_OK;
5016 }
5017
5018 if (skb->len <= 0) {
5019 dev_kfree_skb_any(skb);
5020 return NETDEV_TX_OK;
5021 }
5022
5023 mss = skb_shinfo(skb)->gso_size;
5024 if (mss) {
5025 u8 hdr_len;
5026
5027 /*
5028 * TSO Workaround for 82571/2/3 Controllers -- if skb->data
5029 * points to just header, pull a few bytes of payload from
5030 * frags into skb->data
5031 */
5032 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5033 /*
5034 * we do this workaround for ES2LAN, but it is un-necessary,
5035 * avoiding it could save a lot of cycles
5036 */
5037 if (skb->data_len && (hdr_len == len)) {
5038 unsigned int pull_size;
5039
5040 pull_size = min_t(unsigned int, 4, skb->data_len);
5041 if (!__pskb_pull_tail(skb, pull_size)) {
5042 e_err("__pskb_pull_tail failed.\n");
5043 dev_kfree_skb_any(skb);
5044 return NETDEV_TX_OK;
5045 }
5046 len = skb_headlen(skb);
5047 }
5048 }
5049
5050 /* reserve a descriptor for the offload context */
5051 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5052 count++;
5053 count++;
5054
5055 count += DIV_ROUND_UP(len, adapter->tx_fifo_limit);
5056
5057 nr_frags = skb_shinfo(skb)->nr_frags;
5058 for (f = 0; f < nr_frags; f++)
5059 count += DIV_ROUND_UP(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5060 adapter->tx_fifo_limit);
5061
5062 if (adapter->hw.mac.tx_pkt_filtering)
5063 e1000_transfer_dhcp_info(adapter, skb);
5064
5065 /*
5066 * need: count + 2 desc gap to keep tail from touching
5067 * head, otherwise try next time
5068 */
5069 if (e1000_maybe_stop_tx(tx_ring, count + 2))
5070 return NETDEV_TX_BUSY;
5071
5072 if (vlan_tx_tag_present(skb)) {
5073 tx_flags |= E1000_TX_FLAGS_VLAN;
5074 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
5075 }
5076
5077 first = tx_ring->next_to_use;
5078
5079 tso = e1000_tso(tx_ring, skb);
5080 if (tso < 0) {
5081 dev_kfree_skb_any(skb);
5082 return NETDEV_TX_OK;
5083 }
5084
5085 if (tso)
5086 tx_flags |= E1000_TX_FLAGS_TSO;
5087 else if (e1000_tx_csum(tx_ring, skb))
5088 tx_flags |= E1000_TX_FLAGS_CSUM;
5089
5090 /*
5091 * Old method was to assume IPv4 packet by default if TSO was enabled.
5092 * 82571 hardware supports TSO capabilities for IPv6 as well...
5093 * no longer assume, we must.
5094 */
5095 if (skb->protocol == htons(ETH_P_IP))
5096 tx_flags |= E1000_TX_FLAGS_IPV4;
5097
5098 if (unlikely(skb->no_fcs))
5099 tx_flags |= E1000_TX_FLAGS_NO_FCS;
5100
5101 /* if count is 0 then mapping error has occurred */
5102 count = e1000_tx_map(tx_ring, skb, first, adapter->tx_fifo_limit,
5103 nr_frags);
5104 if (count) {
5105 skb_tx_timestamp(skb);
5106
5107 netdev_sent_queue(netdev, skb->len);
5108 e1000_tx_queue(tx_ring, tx_flags, count);
5109 /* Make sure there is space in the ring for the next send. */
5110 e1000_maybe_stop_tx(tx_ring,
5111 (MAX_SKB_FRAGS *
5112 DIV_ROUND_UP(PAGE_SIZE,
5113 adapter->tx_fifo_limit) + 2));
5114 } else {
5115 dev_kfree_skb_any(skb);
5116 tx_ring->buffer_info[first].time_stamp = 0;
5117 tx_ring->next_to_use = first;
5118 }
5119
5120 return NETDEV_TX_OK;
5121}
5122
5123/**
5124 * e1000_tx_timeout - Respond to a Tx Hang
5125 * @netdev: network interface device structure
5126 **/
5127static void e1000_tx_timeout(struct net_device *netdev)
5128{
5129 struct e1000_adapter *adapter = netdev_priv(netdev);
5130
5131 /* Do the reset outside of interrupt context */
5132 adapter->tx_timeout_count++;
5133 schedule_work(&adapter->reset_task);
5134}
5135
5136static void e1000_reset_task(struct work_struct *work)
5137{
5138 struct e1000_adapter *adapter;
5139 adapter = container_of(work, struct e1000_adapter, reset_task);
5140
5141 /* don't run the task if already down */
5142 if (test_bit(__E1000_DOWN, &adapter->state))
5143 return;
5144
5145 if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5146 (adapter->flags & FLAG_RX_RESTART_NOW))) {
5147 e1000e_dump(adapter);
5148 e_err("Reset adapter\n");
5149 }
5150 e1000e_reinit_locked(adapter);
5151}
5152
5153/**
5154 * e1000_get_stats64 - Get System Network Statistics
5155 * @netdev: network interface device structure
5156 * @stats: rtnl_link_stats64 pointer
5157 *
5158 * Returns the address of the device statistics structure.
5159 **/
5160struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
5161 struct rtnl_link_stats64 *stats)
5162{
5163 struct e1000_adapter *adapter = netdev_priv(netdev);
5164
5165 memset(stats, 0, sizeof(struct rtnl_link_stats64));
5166 spin_lock(&adapter->stats64_lock);
5167 e1000e_update_stats(adapter);
5168 /* Fill out the OS statistics structure */
5169 stats->rx_bytes = adapter->stats.gorc;
5170 stats->rx_packets = adapter->stats.gprc;
5171 stats->tx_bytes = adapter->stats.gotc;
5172 stats->tx_packets = adapter->stats.gptc;
5173 stats->multicast = adapter->stats.mprc;
5174 stats->collisions = adapter->stats.colc;
5175
5176 /* Rx Errors */
5177
5178 /*
5179 * RLEC on some newer hardware can be incorrect so build
5180 * our own version based on RUC and ROC
5181 */
5182 stats->rx_errors = adapter->stats.rxerrc +
5183 adapter->stats.crcerrs + adapter->stats.algnerrc +
5184 adapter->stats.ruc + adapter->stats.roc +
5185 adapter->stats.cexterr;
5186 stats->rx_length_errors = adapter->stats.ruc +
5187 adapter->stats.roc;
5188 stats->rx_crc_errors = adapter->stats.crcerrs;
5189 stats->rx_frame_errors = adapter->stats.algnerrc;
5190 stats->rx_missed_errors = adapter->stats.mpc;
5191
5192 /* Tx Errors */
5193 stats->tx_errors = adapter->stats.ecol +
5194 adapter->stats.latecol;
5195 stats->tx_aborted_errors = adapter->stats.ecol;
5196 stats->tx_window_errors = adapter->stats.latecol;
5197 stats->tx_carrier_errors = adapter->stats.tncrs;
5198
5199 /* Tx Dropped needs to be maintained elsewhere */
5200
5201 spin_unlock(&adapter->stats64_lock);
5202 return stats;
5203}
5204
5205/**
5206 * e1000_change_mtu - Change the Maximum Transfer Unit
5207 * @netdev: network interface device structure
5208 * @new_mtu: new value for maximum frame size
5209 *
5210 * Returns 0 on success, negative on failure
5211 **/
5212static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5213{
5214 struct e1000_adapter *adapter = netdev_priv(netdev);
5215 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5216
5217 /* Jumbo frame support */
5218 if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
5219 !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5220 e_err("Jumbo Frames not supported.\n");
5221 return -EINVAL;
5222 }
5223
5224 /* Supported frame sizes */
5225 if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
5226 (max_frame > adapter->max_hw_frame_size)) {
5227 e_err("Unsupported MTU setting\n");
5228 return -EINVAL;
5229 }
5230
5231 /* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
5232 if ((adapter->hw.mac.type >= e1000_pch2lan) &&
5233 !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5234 (new_mtu > ETH_DATA_LEN)) {
5235 e_err("Jumbo Frames not supported on this device when CRC stripping is disabled.\n");
5236 return -EINVAL;
5237 }
5238
5239 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5240 usleep_range(1000, 2000);
5241 /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5242 adapter->max_frame_size = max_frame;
5243 e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5244 netdev->mtu = new_mtu;
5245 if (netif_running(netdev))
5246 e1000e_down(adapter);
5247
5248 /*
5249 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5250 * means we reserve 2 more, this pushes us to allocate from the next
5251 * larger slab size.
5252 * i.e. RXBUFFER_2048 --> size-4096 slab
5253 * However with the new *_jumbo_rx* routines, jumbo receives will use
5254 * fragmented skbs
5255 */
5256
5257 if (max_frame <= 2048)
5258 adapter->rx_buffer_len = 2048;
5259 else
5260 adapter->rx_buffer_len = 4096;
5261
5262 /* adjust allocation if LPE protects us, and we aren't using SBP */
5263 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5264 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5265 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5266 + ETH_FCS_LEN;
5267
5268 if (netif_running(netdev))
5269 e1000e_up(adapter);
5270 else
5271 e1000e_reset(adapter);
5272
5273 clear_bit(__E1000_RESETTING, &adapter->state);
5274
5275 return 0;
5276}
5277
5278static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5279 int cmd)
5280{
5281 struct e1000_adapter *adapter = netdev_priv(netdev);
5282 struct mii_ioctl_data *data = if_mii(ifr);
5283
5284 if (adapter->hw.phy.media_type != e1000_media_type_copper)
5285 return -EOPNOTSUPP;
5286
5287 switch (cmd) {
5288 case SIOCGMIIPHY:
5289 data->phy_id = adapter->hw.phy.addr;
5290 break;
5291 case SIOCGMIIREG:
5292 e1000_phy_read_status(adapter);
5293
5294 switch (data->reg_num & 0x1F) {
5295 case MII_BMCR:
5296 data->val_out = adapter->phy_regs.bmcr;
5297 break;
5298 case MII_BMSR:
5299 data->val_out = adapter->phy_regs.bmsr;
5300 break;
5301 case MII_PHYSID1:
5302 data->val_out = (adapter->hw.phy.id >> 16);
5303 break;
5304 case MII_PHYSID2:
5305 data->val_out = (adapter->hw.phy.id & 0xFFFF);
5306 break;
5307 case MII_ADVERTISE:
5308 data->val_out = adapter->phy_regs.advertise;
5309 break;
5310 case MII_LPA:
5311 data->val_out = adapter->phy_regs.lpa;
5312 break;
5313 case MII_EXPANSION:
5314 data->val_out = adapter->phy_regs.expansion;
5315 break;
5316 case MII_CTRL1000:
5317 data->val_out = adapter->phy_regs.ctrl1000;
5318 break;
5319 case MII_STAT1000:
5320 data->val_out = adapter->phy_regs.stat1000;
5321 break;
5322 case MII_ESTATUS:
5323 data->val_out = adapter->phy_regs.estatus;
5324 break;
5325 default:
5326 return -EIO;
5327 }
5328 break;
5329 case SIOCSMIIREG:
5330 default:
5331 return -EOPNOTSUPP;
5332 }
5333 return 0;
5334}
5335
5336static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5337{
5338 switch (cmd) {
5339 case SIOCGMIIPHY:
5340 case SIOCGMIIREG:
5341 case SIOCSMIIREG:
5342 return e1000_mii_ioctl(netdev, ifr, cmd);
5343 default:
5344 return -EOPNOTSUPP;
5345 }
5346}
5347
5348static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5349{
5350 struct e1000_hw *hw = &adapter->hw;
5351 u32 i, mac_reg;
5352 u16 phy_reg, wuc_enable;
5353 int retval = 0;
5354
5355 /* copy MAC RARs to PHY RARs */
5356 e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5357
5358 retval = hw->phy.ops.acquire(hw);
5359 if (retval) {
5360 e_err("Could not acquire PHY\n");
5361 return retval;
5362 }
5363
5364 /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
5365 retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5366 if (retval)
5367 goto release;
5368
5369 /* copy MAC MTA to PHY MTA - only needed for pchlan */
5370 for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5371 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5372 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
5373 (u16)(mac_reg & 0xFFFF));
5374 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
5375 (u16)((mac_reg >> 16) & 0xFFFF));
5376 }
5377
5378 /* configure PHY Rx Control register */
5379 hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
5380 mac_reg = er32(RCTL);
5381 if (mac_reg & E1000_RCTL_UPE)
5382 phy_reg |= BM_RCTL_UPE;
5383 if (mac_reg & E1000_RCTL_MPE)
5384 phy_reg |= BM_RCTL_MPE;
5385 phy_reg &= ~(BM_RCTL_MO_MASK);
5386 if (mac_reg & E1000_RCTL_MO_3)
5387 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5388 << BM_RCTL_MO_SHIFT);
5389 if (mac_reg & E1000_RCTL_BAM)
5390 phy_reg |= BM_RCTL_BAM;
5391 if (mac_reg & E1000_RCTL_PMCF)
5392 phy_reg |= BM_RCTL_PMCF;
5393 mac_reg = er32(CTRL);
5394 if (mac_reg & E1000_CTRL_RFCE)
5395 phy_reg |= BM_RCTL_RFCE;
5396 hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
5397
5398 /* enable PHY wakeup in MAC register */
5399 ew32(WUFC, wufc);
5400 ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5401
5402 /* configure and enable PHY wakeup in PHY registers */
5403 hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
5404 hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5405
5406 /* activate PHY wakeup */
5407 wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5408 retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5409 if (retval)
5410 e_err("Could not set PHY Host Wakeup bit\n");
5411release:
5412 hw->phy.ops.release(hw);
5413
5414 return retval;
5415}
5416
5417static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5418 bool runtime)
5419{
5420 struct net_device *netdev = pci_get_drvdata(pdev);
5421 struct e1000_adapter *adapter = netdev_priv(netdev);
5422 struct e1000_hw *hw = &adapter->hw;
5423 u32 ctrl, ctrl_ext, rctl, status;
5424 /* Runtime suspend should only enable wakeup for link changes */
5425 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5426 int retval = 0;
5427
5428 netif_device_detach(netdev);
5429
5430 if (netif_running(netdev)) {
5431 int count = E1000_CHECK_RESET_COUNT;
5432
5433 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
5434 usleep_range(10000, 20000);
5435
5436 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5437 e1000e_down(adapter);
5438 e1000_free_irq(adapter);
5439 }
5440 e1000e_reset_interrupt_capability(adapter);
5441
5442 retval = pci_save_state(pdev);
5443 if (retval)
5444 return retval;
5445
5446 status = er32(STATUS);
5447 if (status & E1000_STATUS_LU)
5448 wufc &= ~E1000_WUFC_LNKC;
5449
5450 if (wufc) {
5451 e1000_setup_rctl(adapter);
5452 e1000e_set_rx_mode(netdev);
5453
5454 /* turn on all-multi mode if wake on multicast is enabled */
5455 if (wufc & E1000_WUFC_MC) {
5456 rctl = er32(RCTL);
5457 rctl |= E1000_RCTL_MPE;
5458 ew32(RCTL, rctl);
5459 }
5460
5461 ctrl = er32(CTRL);
5462 /* advertise wake from D3Cold */
5463 #define E1000_CTRL_ADVD3WUC 0x00100000
5464 /* phy power management enable */
5465 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5466 ctrl |= E1000_CTRL_ADVD3WUC;
5467 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5468 ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5469 ew32(CTRL, ctrl);
5470
5471 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5472 adapter->hw.phy.media_type ==
5473 e1000_media_type_internal_serdes) {
5474 /* keep the laser running in D3 */
5475 ctrl_ext = er32(CTRL_EXT);
5476 ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5477 ew32(CTRL_EXT, ctrl_ext);
5478 }
5479
5480 if (adapter->flags & FLAG_IS_ICH)
5481 e1000_suspend_workarounds_ich8lan(&adapter->hw);
5482
5483 /* Allow time for pending master requests to run */
5484 e1000e_disable_pcie_master(&adapter->hw);
5485
5486 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5487 /* enable wakeup by the PHY */
5488 retval = e1000_init_phy_wakeup(adapter, wufc);
5489 if (retval)
5490 return retval;
5491 } else {
5492 /* enable wakeup by the MAC */
5493 ew32(WUFC, wufc);
5494 ew32(WUC, E1000_WUC_PME_EN);
5495 }
5496 } else {
5497 ew32(WUC, 0);
5498 ew32(WUFC, 0);
5499 }
5500
5501 *enable_wake = !!wufc;
5502
5503 /* make sure adapter isn't asleep if manageability is enabled */
5504 if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5505 (hw->mac.ops.check_mng_mode(hw)))
5506 *enable_wake = true;
5507
5508 if (adapter->hw.phy.type == e1000_phy_igp_3)
5509 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5510
5511 /*
5512 * Release control of h/w to f/w. If f/w is AMT enabled, this
5513 * would have already happened in close and is redundant.
5514 */
5515 e1000e_release_hw_control(adapter);
5516
5517 pci_disable_device(pdev);
5518
5519 return 0;
5520}
5521
5522static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5523{
5524 if (sleep && wake) {
5525 pci_prepare_to_sleep(pdev);
5526 return;
5527 }
5528
5529 pci_wake_from_d3(pdev, wake);
5530 pci_set_power_state(pdev, PCI_D3hot);
5531}
5532
5533static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5534 bool wake)
5535{
5536 struct net_device *netdev = pci_get_drvdata(pdev);
5537 struct e1000_adapter *adapter = netdev_priv(netdev);
5538
5539 /*
5540 * The pci-e switch on some quad port adapters will report a
5541 * correctable error when the MAC transitions from D0 to D3. To
5542 * prevent this we need to mask off the correctable errors on the
5543 * downstream port of the pci-e switch.
5544 */
5545 if (adapter->flags & FLAG_IS_QUAD_PORT) {
5546 struct pci_dev *us_dev = pdev->bus->self;
5547 int pos = pci_pcie_cap(us_dev);
5548 u16 devctl;
5549
5550 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5551 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5552 (devctl & ~PCI_EXP_DEVCTL_CERE));
5553
5554 e1000_power_off(pdev, sleep, wake);
5555
5556 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5557 } else {
5558 e1000_power_off(pdev, sleep, wake);
5559 }
5560}
5561
5562#ifdef CONFIG_PCIEASPM
5563static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5564{
5565 pci_disable_link_state_locked(pdev, state);
5566}
5567#else
5568static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5569{
5570 int pos;
5571 u16 reg16;
5572
5573 /*
5574 * Both device and parent should have the same ASPM setting.
5575 * Disable ASPM in downstream component first and then upstream.
5576 */
5577 pos = pci_pcie_cap(pdev);
5578 pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, ®16);
5579 reg16 &= ~state;
5580 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5581
5582 if (!pdev->bus->self)
5583 return;
5584
5585 pos = pci_pcie_cap(pdev->bus->self);
5586 pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, ®16);
5587 reg16 &= ~state;
5588 pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5589}
5590#endif
5591static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5592{
5593 dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5594 (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5595 (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5596
5597 __e1000e_disable_aspm(pdev, state);
5598}
5599
5600#ifdef CONFIG_PM
5601static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5602{
5603 return !!adapter->tx_ring->buffer_info;
5604}
5605
5606static int __e1000_resume(struct pci_dev *pdev)
5607{
5608 struct net_device *netdev = pci_get_drvdata(pdev);
5609 struct e1000_adapter *adapter = netdev_priv(netdev);
5610 struct e1000_hw *hw = &adapter->hw;
5611 u16 aspm_disable_flag = 0;
5612 u32 err;
5613
5614 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5615 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5616 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5617 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5618 if (aspm_disable_flag)
5619 e1000e_disable_aspm(pdev, aspm_disable_flag);
5620
5621 pci_set_power_state(pdev, PCI_D0);
5622 pci_restore_state(pdev);
5623 pci_save_state(pdev);
5624
5625 e1000e_set_interrupt_capability(adapter);
5626 if (netif_running(netdev)) {
5627 err = e1000_request_irq(adapter);
5628 if (err)
5629 return err;
5630 }
5631
5632 if (hw->mac.type >= e1000_pch2lan)
5633 e1000_resume_workarounds_pchlan(&adapter->hw);
5634
5635 e1000e_power_up_phy(adapter);
5636
5637 /* report the system wakeup cause from S3/S4 */
5638 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5639 u16 phy_data;
5640
5641 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5642 if (phy_data) {
5643 e_info("PHY Wakeup cause - %s\n",
5644 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5645 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5646 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5647 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5648 phy_data & E1000_WUS_LNKC ?
5649 "Link Status Change" : "other");
5650 }
5651 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5652 } else {
5653 u32 wus = er32(WUS);
5654 if (wus) {
5655 e_info("MAC Wakeup cause - %s\n",
5656 wus & E1000_WUS_EX ? "Unicast Packet" :
5657 wus & E1000_WUS_MC ? "Multicast Packet" :
5658 wus & E1000_WUS_BC ? "Broadcast Packet" :
5659 wus & E1000_WUS_MAG ? "Magic Packet" :
5660 wus & E1000_WUS_LNKC ? "Link Status Change" :
5661 "other");
5662 }
5663 ew32(WUS, ~0);
5664 }
5665
5666 e1000e_reset(adapter);
5667
5668 e1000_init_manageability_pt(adapter);
5669
5670 if (netif_running(netdev))
5671 e1000e_up(adapter);
5672
5673 netif_device_attach(netdev);
5674
5675 /*
5676 * If the controller has AMT, do not set DRV_LOAD until the interface
5677 * is up. For all other cases, let the f/w know that the h/w is now
5678 * under the control of the driver.
5679 */
5680 if (!(adapter->flags & FLAG_HAS_AMT))
5681 e1000e_get_hw_control(adapter);
5682
5683 return 0;
5684}
5685
5686#ifdef CONFIG_PM_SLEEP
5687static int e1000_suspend(struct device *dev)
5688{
5689 struct pci_dev *pdev = to_pci_dev(dev);
5690 int retval;
5691 bool wake;
5692
5693 retval = __e1000_shutdown(pdev, &wake, false);
5694 if (!retval)
5695 e1000_complete_shutdown(pdev, true, wake);
5696
5697 return retval;
5698}
5699
5700static int e1000_resume(struct device *dev)
5701{
5702 struct pci_dev *pdev = to_pci_dev(dev);
5703 struct net_device *netdev = pci_get_drvdata(pdev);
5704 struct e1000_adapter *adapter = netdev_priv(netdev);
5705
5706 if (e1000e_pm_ready(adapter))
5707 adapter->idle_check = true;
5708
5709 return __e1000_resume(pdev);
5710}
5711#endif /* CONFIG_PM_SLEEP */
5712
5713#ifdef CONFIG_PM_RUNTIME
5714static int e1000_runtime_suspend(struct device *dev)
5715{
5716 struct pci_dev *pdev = to_pci_dev(dev);
5717 struct net_device *netdev = pci_get_drvdata(pdev);
5718 struct e1000_adapter *adapter = netdev_priv(netdev);
5719
5720 if (e1000e_pm_ready(adapter)) {
5721 bool wake;
5722
5723 __e1000_shutdown(pdev, &wake, true);
5724 }
5725
5726 return 0;
5727}
5728
5729static int e1000_idle(struct device *dev)
5730{
5731 struct pci_dev *pdev = to_pci_dev(dev);
5732 struct net_device *netdev = pci_get_drvdata(pdev);
5733 struct e1000_adapter *adapter = netdev_priv(netdev);
5734
5735 if (!e1000e_pm_ready(adapter))
5736 return 0;
5737
5738 if (adapter->idle_check) {
5739 adapter->idle_check = false;
5740 if (!e1000e_has_link(adapter))
5741 pm_schedule_suspend(dev, MSEC_PER_SEC);
5742 }
5743
5744 return -EBUSY;
5745}
5746
5747static int e1000_runtime_resume(struct device *dev)
5748{
5749 struct pci_dev *pdev = to_pci_dev(dev);
5750 struct net_device *netdev = pci_get_drvdata(pdev);
5751 struct e1000_adapter *adapter = netdev_priv(netdev);
5752
5753 if (!e1000e_pm_ready(adapter))
5754 return 0;
5755
5756 adapter->idle_check = !dev->power.runtime_auto;
5757 return __e1000_resume(pdev);
5758}
5759#endif /* CONFIG_PM_RUNTIME */
5760#endif /* CONFIG_PM */
5761
5762static void e1000_shutdown(struct pci_dev *pdev)
5763{
5764 bool wake = false;
5765
5766 __e1000_shutdown(pdev, &wake, false);
5767
5768 if (system_state == SYSTEM_POWER_OFF)
5769 e1000_complete_shutdown(pdev, false, wake);
5770}
5771
5772#ifdef CONFIG_NET_POLL_CONTROLLER
5773
5774static irqreturn_t e1000_intr_msix(int irq, void *data)
5775{
5776 struct net_device *netdev = data;
5777 struct e1000_adapter *adapter = netdev_priv(netdev);
5778
5779 if (adapter->msix_entries) {
5780 int vector, msix_irq;
5781
5782 vector = 0;
5783 msix_irq = adapter->msix_entries[vector].vector;
5784 disable_irq(msix_irq);
5785 e1000_intr_msix_rx(msix_irq, netdev);
5786 enable_irq(msix_irq);
5787
5788 vector++;
5789 msix_irq = adapter->msix_entries[vector].vector;
5790 disable_irq(msix_irq);
5791 e1000_intr_msix_tx(msix_irq, netdev);
5792 enable_irq(msix_irq);
5793
5794 vector++;
5795 msix_irq = adapter->msix_entries[vector].vector;
5796 disable_irq(msix_irq);
5797 e1000_msix_other(msix_irq, netdev);
5798 enable_irq(msix_irq);
5799 }
5800
5801 return IRQ_HANDLED;
5802}
5803
5804/*
5805 * Polling 'interrupt' - used by things like netconsole to send skbs
5806 * without having to re-enable interrupts. It's not called while
5807 * the interrupt routine is executing.
5808 */
5809static void e1000_netpoll(struct net_device *netdev)
5810{
5811 struct e1000_adapter *adapter = netdev_priv(netdev);
5812
5813 switch (adapter->int_mode) {
5814 case E1000E_INT_MODE_MSIX:
5815 e1000_intr_msix(adapter->pdev->irq, netdev);
5816 break;
5817 case E1000E_INT_MODE_MSI:
5818 disable_irq(adapter->pdev->irq);
5819 e1000_intr_msi(adapter->pdev->irq, netdev);
5820 enable_irq(adapter->pdev->irq);
5821 break;
5822 default: /* E1000E_INT_MODE_LEGACY */
5823 disable_irq(adapter->pdev->irq);
5824 e1000_intr(adapter->pdev->irq, netdev);
5825 enable_irq(adapter->pdev->irq);
5826 break;
5827 }
5828}
5829#endif
5830
5831/**
5832 * e1000_io_error_detected - called when PCI error is detected
5833 * @pdev: Pointer to PCI device
5834 * @state: The current pci connection state
5835 *
5836 * This function is called after a PCI bus error affecting
5837 * this device has been detected.
5838 */
5839static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5840 pci_channel_state_t state)
5841{
5842 struct net_device *netdev = pci_get_drvdata(pdev);
5843 struct e1000_adapter *adapter = netdev_priv(netdev);
5844
5845 netif_device_detach(netdev);
5846
5847 if (state == pci_channel_io_perm_failure)
5848 return PCI_ERS_RESULT_DISCONNECT;
5849
5850 if (netif_running(netdev))
5851 e1000e_down(adapter);
5852 pci_disable_device(pdev);
5853
5854 /* Request a slot slot reset. */
5855 return PCI_ERS_RESULT_NEED_RESET;
5856}
5857
5858/**
5859 * e1000_io_slot_reset - called after the pci bus has been reset.
5860 * @pdev: Pointer to PCI device
5861 *
5862 * Restart the card from scratch, as if from a cold-boot. Implementation
5863 * resembles the first-half of the e1000_resume routine.
5864 */
5865static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5866{
5867 struct net_device *netdev = pci_get_drvdata(pdev);
5868 struct e1000_adapter *adapter = netdev_priv(netdev);
5869 struct e1000_hw *hw = &adapter->hw;
5870 u16 aspm_disable_flag = 0;
5871 int err;
5872 pci_ers_result_t result;
5873
5874 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5875 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5876 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5877 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5878 if (aspm_disable_flag)
5879 e1000e_disable_aspm(pdev, aspm_disable_flag);
5880
5881 err = pci_enable_device_mem(pdev);
5882 if (err) {
5883 dev_err(&pdev->dev,
5884 "Cannot re-enable PCI device after reset.\n");
5885 result = PCI_ERS_RESULT_DISCONNECT;
5886 } else {
5887 pci_set_master(pdev);
5888 pdev->state_saved = true;
5889 pci_restore_state(pdev);
5890
5891 pci_enable_wake(pdev, PCI_D3hot, 0);
5892 pci_enable_wake(pdev, PCI_D3cold, 0);
5893
5894 e1000e_reset(adapter);
5895 ew32(WUS, ~0);
5896 result = PCI_ERS_RESULT_RECOVERED;
5897 }
5898
5899 pci_cleanup_aer_uncorrect_error_status(pdev);
5900
5901 return result;
5902}
5903
5904/**
5905 * e1000_io_resume - called when traffic can start flowing again.
5906 * @pdev: Pointer to PCI device
5907 *
5908 * This callback is called when the error recovery driver tells us that
5909 * its OK to resume normal operation. Implementation resembles the
5910 * second-half of the e1000_resume routine.
5911 */
5912static void e1000_io_resume(struct pci_dev *pdev)
5913{
5914 struct net_device *netdev = pci_get_drvdata(pdev);
5915 struct e1000_adapter *adapter = netdev_priv(netdev);
5916
5917 e1000_init_manageability_pt(adapter);
5918
5919 if (netif_running(netdev)) {
5920 if (e1000e_up(adapter)) {
5921 dev_err(&pdev->dev,
5922 "can't bring device back up after reset\n");
5923 return;
5924 }
5925 }
5926
5927 netif_device_attach(netdev);
5928
5929 /*
5930 * If the controller has AMT, do not set DRV_LOAD until the interface
5931 * is up. For all other cases, let the f/w know that the h/w is now
5932 * under the control of the driver.
5933 */
5934 if (!(adapter->flags & FLAG_HAS_AMT))
5935 e1000e_get_hw_control(adapter);
5936
5937}
5938
5939static void e1000_print_device_info(struct e1000_adapter *adapter)
5940{
5941 struct e1000_hw *hw = &adapter->hw;
5942 struct net_device *netdev = adapter->netdev;
5943 u32 ret_val;
5944 u8 pba_str[E1000_PBANUM_LENGTH];
5945
5946 /* print bus type/speed/width info */
5947 e_info("(PCI Express:2.5GT/s:%s) %pM\n",
5948 /* bus width */
5949 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5950 "Width x1"),
5951 /* MAC address */
5952 netdev->dev_addr);
5953 e_info("Intel(R) PRO/%s Network Connection\n",
5954 (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5955 ret_val = e1000_read_pba_string_generic(hw, pba_str,
5956 E1000_PBANUM_LENGTH);
5957 if (ret_val)
5958 strlcpy((char *)pba_str, "Unknown", sizeof(pba_str));
5959 e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5960 hw->mac.type, hw->phy.type, pba_str);
5961}
5962
5963static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5964{
5965 struct e1000_hw *hw = &adapter->hw;
5966 int ret_val;
5967 u16 buf = 0;
5968
5969 if (hw->mac.type != e1000_82573)
5970 return;
5971
5972 ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5973 le16_to_cpus(&buf);
5974 if (!ret_val && (!(buf & (1 << 0)))) {
5975 /* Deep Smart Power Down (DSPD) */
5976 dev_warn(&adapter->pdev->dev,
5977 "Warning: detected DSPD enabled in EEPROM\n");
5978 }
5979}
5980
5981static int e1000_set_features(struct net_device *netdev,
5982 netdev_features_t features)
5983{
5984 struct e1000_adapter *adapter = netdev_priv(netdev);
5985 netdev_features_t changed = features ^ netdev->features;
5986
5987 if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
5988 adapter->flags |= FLAG_TSO_FORCE;
5989
5990 if (!(changed & (NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX |
5991 NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_RXFCS |
5992 NETIF_F_RXALL)))
5993 return 0;
5994
5995 if (changed & NETIF_F_RXFCS) {
5996 if (features & NETIF_F_RXFCS) {
5997 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
5998 } else {
5999 /* We need to take it back to defaults, which might mean
6000 * stripping is still disabled at the adapter level.
6001 */
6002 if (adapter->flags2 & FLAG2_DFLT_CRC_STRIPPING)
6003 adapter->flags2 |= FLAG2_CRC_STRIPPING;
6004 else
6005 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6006 }
6007 }
6008
6009 netdev->features = features;
6010
6011 if (netif_running(netdev))
6012 e1000e_reinit_locked(adapter);
6013 else
6014 e1000e_reset(adapter);
6015
6016 return 0;
6017}
6018
6019static const struct net_device_ops e1000e_netdev_ops = {
6020 .ndo_open = e1000_open,
6021 .ndo_stop = e1000_close,
6022 .ndo_start_xmit = e1000_xmit_frame,
6023 .ndo_get_stats64 = e1000e_get_stats64,
6024 .ndo_set_rx_mode = e1000e_set_rx_mode,
6025 .ndo_set_mac_address = e1000_set_mac,
6026 .ndo_change_mtu = e1000_change_mtu,
6027 .ndo_do_ioctl = e1000_ioctl,
6028 .ndo_tx_timeout = e1000_tx_timeout,
6029 .ndo_validate_addr = eth_validate_addr,
6030
6031 .ndo_vlan_rx_add_vid = e1000_vlan_rx_add_vid,
6032 .ndo_vlan_rx_kill_vid = e1000_vlan_rx_kill_vid,
6033#ifdef CONFIG_NET_POLL_CONTROLLER
6034 .ndo_poll_controller = e1000_netpoll,
6035#endif
6036 .ndo_set_features = e1000_set_features,
6037};
6038
6039/**
6040 * e1000_probe - Device Initialization Routine
6041 * @pdev: PCI device information struct
6042 * @ent: entry in e1000_pci_tbl
6043 *
6044 * Returns 0 on success, negative on failure
6045 *
6046 * e1000_probe initializes an adapter identified by a pci_dev structure.
6047 * The OS initialization, configuring of the adapter private structure,
6048 * and a hardware reset occur.
6049 **/
6050static int __devinit e1000_probe(struct pci_dev *pdev,
6051 const struct pci_device_id *ent)
6052{
6053 struct net_device *netdev;
6054 struct e1000_adapter *adapter;
6055 struct e1000_hw *hw;
6056 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
6057 resource_size_t mmio_start, mmio_len;
6058 resource_size_t flash_start, flash_len;
6059 static int cards_found;
6060 u16 aspm_disable_flag = 0;
6061 int i, err, pci_using_dac;
6062 u16 eeprom_data = 0;
6063 u16 eeprom_apme_mask = E1000_EEPROM_APME;
6064
6065 if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
6066 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6067 if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
6068 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6069 if (aspm_disable_flag)
6070 e1000e_disable_aspm(pdev, aspm_disable_flag);
6071
6072 err = pci_enable_device_mem(pdev);
6073 if (err)
6074 return err;
6075
6076 pci_using_dac = 0;
6077 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
6078 if (!err) {
6079 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
6080 if (!err)
6081 pci_using_dac = 1;
6082 } else {
6083 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6084 if (err) {
6085 err = dma_set_coherent_mask(&pdev->dev,
6086 DMA_BIT_MASK(32));
6087 if (err) {
6088 dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
6089 goto err_dma;
6090 }
6091 }
6092 }
6093
6094 err = pci_request_selected_regions_exclusive(pdev,
6095 pci_select_bars(pdev, IORESOURCE_MEM),
6096 e1000e_driver_name);
6097 if (err)
6098 goto err_pci_reg;
6099
6100 /* AER (Advanced Error Reporting) hooks */
6101 pci_enable_pcie_error_reporting(pdev);
6102
6103 pci_set_master(pdev);
6104 /* PCI config space info */
6105 err = pci_save_state(pdev);
6106 if (err)
6107 goto err_alloc_etherdev;
6108
6109 err = -ENOMEM;
6110 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
6111 if (!netdev)
6112 goto err_alloc_etherdev;
6113
6114 SET_NETDEV_DEV(netdev, &pdev->dev);
6115
6116 netdev->irq = pdev->irq;
6117
6118 pci_set_drvdata(pdev, netdev);
6119 adapter = netdev_priv(netdev);
6120 hw = &adapter->hw;
6121 adapter->netdev = netdev;
6122 adapter->pdev = pdev;
6123 adapter->ei = ei;
6124 adapter->pba = ei->pba;
6125 adapter->flags = ei->flags;
6126 adapter->flags2 = ei->flags2;
6127 adapter->hw.adapter = adapter;
6128 adapter->hw.mac.type = ei->mac;
6129 adapter->max_hw_frame_size = ei->max_hw_frame_size;
6130 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
6131
6132 mmio_start = pci_resource_start(pdev, 0);
6133 mmio_len = pci_resource_len(pdev, 0);
6134
6135 err = -EIO;
6136 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
6137 if (!adapter->hw.hw_addr)
6138 goto err_ioremap;
6139
6140 if ((adapter->flags & FLAG_HAS_FLASH) &&
6141 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
6142 flash_start = pci_resource_start(pdev, 1);
6143 flash_len = pci_resource_len(pdev, 1);
6144 adapter->hw.flash_address = ioremap(flash_start, flash_len);
6145 if (!adapter->hw.flash_address)
6146 goto err_flashmap;
6147 }
6148
6149 /* construct the net_device struct */
6150 netdev->netdev_ops = &e1000e_netdev_ops;
6151 e1000e_set_ethtool_ops(netdev);
6152 netdev->watchdog_timeo = 5 * HZ;
6153 netif_napi_add(netdev, &adapter->napi, e1000e_poll, 64);
6154 strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
6155
6156 netdev->mem_start = mmio_start;
6157 netdev->mem_end = mmio_start + mmio_len;
6158
6159 adapter->bd_number = cards_found++;
6160
6161 e1000e_check_options(adapter);
6162
6163 /* setup adapter struct */
6164 err = e1000_sw_init(adapter);
6165 if (err)
6166 goto err_sw_init;
6167
6168 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
6169 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
6170 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
6171
6172 err = ei->get_variants(adapter);
6173 if (err)
6174 goto err_hw_init;
6175
6176 if ((adapter->flags & FLAG_IS_ICH) &&
6177 (adapter->flags & FLAG_READ_ONLY_NVM))
6178 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
6179
6180 hw->mac.ops.get_bus_info(&adapter->hw);
6181
6182 adapter->hw.phy.autoneg_wait_to_complete = 0;
6183
6184 /* Copper options */
6185 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
6186 adapter->hw.phy.mdix = AUTO_ALL_MODES;
6187 adapter->hw.phy.disable_polarity_correction = 0;
6188 adapter->hw.phy.ms_type = e1000_ms_hw_default;
6189 }
6190
6191 if (hw->phy.ops.check_reset_block && hw->phy.ops.check_reset_block(hw))
6192 e_info("PHY reset is blocked due to SOL/IDER session.\n");
6193
6194 /* Set initial default active device features */
6195 netdev->features = (NETIF_F_SG |
6196 NETIF_F_HW_VLAN_RX |
6197 NETIF_F_HW_VLAN_TX |
6198 NETIF_F_TSO |
6199 NETIF_F_TSO6 |
6200 NETIF_F_RXHASH |
6201 NETIF_F_RXCSUM |
6202 NETIF_F_HW_CSUM);
6203
6204 /* Set user-changeable features (subset of all device features) */
6205 netdev->hw_features = netdev->features;
6206 netdev->hw_features |= NETIF_F_RXFCS;
6207 netdev->priv_flags |= IFF_SUPP_NOFCS;
6208 netdev->hw_features |= NETIF_F_RXALL;
6209
6210 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
6211 netdev->features |= NETIF_F_HW_VLAN_FILTER;
6212
6213 netdev->vlan_features |= (NETIF_F_SG |
6214 NETIF_F_TSO |
6215 NETIF_F_TSO6 |
6216 NETIF_F_HW_CSUM);
6217
6218 netdev->priv_flags |= IFF_UNICAST_FLT;
6219
6220 if (pci_using_dac) {
6221 netdev->features |= NETIF_F_HIGHDMA;
6222 netdev->vlan_features |= NETIF_F_HIGHDMA;
6223 }
6224
6225 if (e1000e_enable_mng_pass_thru(&adapter->hw))
6226 adapter->flags |= FLAG_MNG_PT_ENABLED;
6227
6228 /*
6229 * before reading the NVM, reset the controller to
6230 * put the device in a known good starting state
6231 */
6232 adapter->hw.mac.ops.reset_hw(&adapter->hw);
6233
6234 /*
6235 * systems with ASPM and others may see the checksum fail on the first
6236 * attempt. Let's give it a few tries
6237 */
6238 for (i = 0;; i++) {
6239 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
6240 break;
6241 if (i == 2) {
6242 e_err("The NVM Checksum Is Not Valid\n");
6243 err = -EIO;
6244 goto err_eeprom;
6245 }
6246 }
6247
6248 e1000_eeprom_checks(adapter);
6249
6250 /* copy the MAC address */
6251 if (e1000e_read_mac_addr(&adapter->hw))
6252 e_err("NVM Read Error while reading MAC address\n");
6253
6254 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
6255 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
6256
6257 if (!is_valid_ether_addr(netdev->perm_addr)) {
6258 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
6259 err = -EIO;
6260 goto err_eeprom;
6261 }
6262
6263 init_timer(&adapter->watchdog_timer);
6264 adapter->watchdog_timer.function = e1000_watchdog;
6265 adapter->watchdog_timer.data = (unsigned long) adapter;
6266
6267 init_timer(&adapter->phy_info_timer);
6268 adapter->phy_info_timer.function = e1000_update_phy_info;
6269 adapter->phy_info_timer.data = (unsigned long) adapter;
6270
6271 INIT_WORK(&adapter->reset_task, e1000_reset_task);
6272 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
6273 INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
6274 INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
6275 INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
6276
6277 /* Initialize link parameters. User can change them with ethtool */
6278 adapter->hw.mac.autoneg = 1;
6279 adapter->fc_autoneg = true;
6280 adapter->hw.fc.requested_mode = e1000_fc_default;
6281 adapter->hw.fc.current_mode = e1000_fc_default;
6282 adapter->hw.phy.autoneg_advertised = 0x2f;
6283
6284 /* ring size defaults */
6285 adapter->rx_ring->count = E1000_DEFAULT_RXD;
6286 adapter->tx_ring->count = E1000_DEFAULT_TXD;
6287
6288 /*
6289 * Initial Wake on LAN setting - If APM wake is enabled in
6290 * the EEPROM, enable the ACPI Magic Packet filter
6291 */
6292 if (adapter->flags & FLAG_APME_IN_WUC) {
6293 /* APME bit in EEPROM is mapped to WUC.APME */
6294 eeprom_data = er32(WUC);
6295 eeprom_apme_mask = E1000_WUC_APME;
6296 if ((hw->mac.type > e1000_ich10lan) &&
6297 (eeprom_data & E1000_WUC_PHY_WAKE))
6298 adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
6299 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
6300 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
6301 (adapter->hw.bus.func == 1))
6302 e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_B,
6303 1, &eeprom_data);
6304 else
6305 e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_A,
6306 1, &eeprom_data);
6307 }
6308
6309 /* fetch WoL from EEPROM */
6310 if (eeprom_data & eeprom_apme_mask)
6311 adapter->eeprom_wol |= E1000_WUFC_MAG;
6312
6313 /*
6314 * now that we have the eeprom settings, apply the special cases
6315 * where the eeprom may be wrong or the board simply won't support
6316 * wake on lan on a particular port
6317 */
6318 if (!(adapter->flags & FLAG_HAS_WOL))
6319 adapter->eeprom_wol = 0;
6320
6321 /* initialize the wol settings based on the eeprom settings */
6322 adapter->wol = adapter->eeprom_wol;
6323 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6324
6325 /* save off EEPROM version number */
6326 e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6327
6328 /* reset the hardware with the new settings */
6329 e1000e_reset(adapter);
6330
6331 /*
6332 * If the controller has AMT, do not set DRV_LOAD until the interface
6333 * is up. For all other cases, let the f/w know that the h/w is now
6334 * under the control of the driver.
6335 */
6336 if (!(adapter->flags & FLAG_HAS_AMT))
6337 e1000e_get_hw_control(adapter);
6338
6339 strlcpy(netdev->name, "eth%d", sizeof(netdev->name));
6340 err = register_netdev(netdev);
6341 if (err)
6342 goto err_register;
6343
6344 /* carrier off reporting is important to ethtool even BEFORE open */
6345 netif_carrier_off(netdev);
6346
6347 e1000_print_device_info(adapter);
6348
6349 if (pci_dev_run_wake(pdev))
6350 pm_runtime_put_noidle(&pdev->dev);
6351
6352 return 0;
6353
6354err_register:
6355 if (!(adapter->flags & FLAG_HAS_AMT))
6356 e1000e_release_hw_control(adapter);
6357err_eeprom:
6358 if (hw->phy.ops.check_reset_block && !hw->phy.ops.check_reset_block(hw))
6359 e1000_phy_hw_reset(&adapter->hw);
6360err_hw_init:
6361 kfree(adapter->tx_ring);
6362 kfree(adapter->rx_ring);
6363err_sw_init:
6364 if (adapter->hw.flash_address)
6365 iounmap(adapter->hw.flash_address);
6366 e1000e_reset_interrupt_capability(adapter);
6367err_flashmap:
6368 iounmap(adapter->hw.hw_addr);
6369err_ioremap:
6370 free_netdev(netdev);
6371err_alloc_etherdev:
6372 pci_release_selected_regions(pdev,
6373 pci_select_bars(pdev, IORESOURCE_MEM));
6374err_pci_reg:
6375err_dma:
6376 pci_disable_device(pdev);
6377 return err;
6378}
6379
6380/**
6381 * e1000_remove - Device Removal Routine
6382 * @pdev: PCI device information struct
6383 *
6384 * e1000_remove is called by the PCI subsystem to alert the driver
6385 * that it should release a PCI device. The could be caused by a
6386 * Hot-Plug event, or because the driver is going to be removed from
6387 * memory.
6388 **/
6389static void __devexit e1000_remove(struct pci_dev *pdev)
6390{
6391 struct net_device *netdev = pci_get_drvdata(pdev);
6392 struct e1000_adapter *adapter = netdev_priv(netdev);
6393 bool down = test_bit(__E1000_DOWN, &adapter->state);
6394
6395 /*
6396 * The timers may be rescheduled, so explicitly disable them
6397 * from being rescheduled.
6398 */
6399 if (!down)
6400 set_bit(__E1000_DOWN, &adapter->state);
6401 del_timer_sync(&adapter->watchdog_timer);
6402 del_timer_sync(&adapter->phy_info_timer);
6403
6404 cancel_work_sync(&adapter->reset_task);
6405 cancel_work_sync(&adapter->watchdog_task);
6406 cancel_work_sync(&adapter->downshift_task);
6407 cancel_work_sync(&adapter->update_phy_task);
6408 cancel_work_sync(&adapter->print_hang_task);
6409
6410 if (!(netdev->flags & IFF_UP))
6411 e1000_power_down_phy(adapter);
6412
6413 /* Don't lie to e1000_close() down the road. */
6414 if (!down)
6415 clear_bit(__E1000_DOWN, &adapter->state);
6416 unregister_netdev(netdev);
6417
6418 if (pci_dev_run_wake(pdev))
6419 pm_runtime_get_noresume(&pdev->dev);
6420
6421 /*
6422 * Release control of h/w to f/w. If f/w is AMT enabled, this
6423 * would have already happened in close and is redundant.
6424 */
6425 e1000e_release_hw_control(adapter);
6426
6427 e1000e_reset_interrupt_capability(adapter);
6428 kfree(adapter->tx_ring);
6429 kfree(adapter->rx_ring);
6430
6431 iounmap(adapter->hw.hw_addr);
6432 if (adapter->hw.flash_address)
6433 iounmap(adapter->hw.flash_address);
6434 pci_release_selected_regions(pdev,
6435 pci_select_bars(pdev, IORESOURCE_MEM));
6436
6437 free_netdev(netdev);
6438
6439 /* AER disable */
6440 pci_disable_pcie_error_reporting(pdev);
6441
6442 pci_disable_device(pdev);
6443}
6444
6445/* PCI Error Recovery (ERS) */
6446static struct pci_error_handlers e1000_err_handler = {
6447 .error_detected = e1000_io_error_detected,
6448 .slot_reset = e1000_io_slot_reset,
6449 .resume = e1000_io_resume,
6450};
6451
6452static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6453 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6454 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6455 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6456 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6457 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6458 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6459 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6460 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6461 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6462
6463 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6464 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6465 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6466 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6467
6468 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6469 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6470 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6471
6472 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6473 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6474 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6475
6476 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6477 board_80003es2lan },
6478 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6479 board_80003es2lan },
6480 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6481 board_80003es2lan },
6482 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6483 board_80003es2lan },
6484
6485 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6486 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6487 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6488 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6489 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6490 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6491 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6492 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6493
6494 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6495 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6496 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6497 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6498 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6499 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6500 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6501 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6502 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6503
6504 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6505 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6506 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6507
6508 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6509 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6510 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6511
6512 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6513 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6514 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6515 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6516
6517 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6518 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6519
6520 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_LM), board_pch_lpt },
6521 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_V), board_pch_lpt },
6522
6523 { 0, 0, 0, 0, 0, 0, 0 } /* terminate list */
6524};
6525MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6526
6527#ifdef CONFIG_PM
6528static const struct dev_pm_ops e1000_pm_ops = {
6529 SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6530 SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6531 e1000_runtime_resume, e1000_idle)
6532};
6533#endif
6534
6535/* PCI Device API Driver */
6536static struct pci_driver e1000_driver = {
6537 .name = e1000e_driver_name,
6538 .id_table = e1000_pci_tbl,
6539 .probe = e1000_probe,
6540 .remove = __devexit_p(e1000_remove),
6541#ifdef CONFIG_PM
6542 .driver = {
6543 .pm = &e1000_pm_ops,
6544 },
6545#endif
6546 .shutdown = e1000_shutdown,
6547 .err_handler = &e1000_err_handler
6548};
6549
6550/**
6551 * e1000_init_module - Driver Registration Routine
6552 *
6553 * e1000_init_module is the first routine called when the driver is
6554 * loaded. All it does is register with the PCI subsystem.
6555 **/
6556static int __init e1000_init_module(void)
6557{
6558 int ret;
6559 pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6560 e1000e_driver_version);
6561 pr_info("Copyright(c) 1999 - 2012 Intel Corporation.\n");
6562 ret = pci_register_driver(&e1000_driver);
6563
6564 return ret;
6565}
6566module_init(e1000_init_module);
6567
6568/**
6569 * e1000_exit_module - Driver Exit Cleanup Routine
6570 *
6571 * e1000_exit_module is called just before the driver is removed
6572 * from memory.
6573 **/
6574static void __exit e1000_exit_module(void)
6575{
6576 pci_unregister_driver(&e1000_driver);
6577}
6578module_exit(e1000_exit_module);
6579
6580
6581MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6582MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6583MODULE_LICENSE("GPL");
6584MODULE_VERSION(DRV_VERSION);
6585
6586/* netdev.c */