Loading...
1// SPDX-License-Identifier: GPL-2.0
2/* Copyright(c) 1999 - 2018 Intel Corporation. */
3
4#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
5
6#include <linux/module.h>
7#include <linux/types.h>
8#include <linux/init.h>
9#include <linux/pci.h>
10#include <linux/vmalloc.h>
11#include <linux/pagemap.h>
12#include <linux/delay.h>
13#include <linux/netdevice.h>
14#include <linux/interrupt.h>
15#include <linux/tcp.h>
16#include <linux/ipv6.h>
17#include <linux/slab.h>
18#include <net/checksum.h>
19#include <net/ip6_checksum.h>
20#include <linux/ethtool.h>
21#include <linux/if_vlan.h>
22#include <linux/cpu.h>
23#include <linux/smp.h>
24#include <linux/pm_qos.h>
25#include <linux/pm_runtime.h>
26#include <linux/prefetch.h>
27#include <linux/suspend.h>
28
29#include "e1000.h"
30#define CREATE_TRACE_POINTS
31#include "e1000e_trace.h"
32
33char e1000e_driver_name[] = "e1000e";
34
35#define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
36static int debug = -1;
37module_param(debug, int, 0);
38MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
39
40static const struct e1000_info *e1000_info_tbl[] = {
41 [board_82571] = &e1000_82571_info,
42 [board_82572] = &e1000_82572_info,
43 [board_82573] = &e1000_82573_info,
44 [board_82574] = &e1000_82574_info,
45 [board_82583] = &e1000_82583_info,
46 [board_80003es2lan] = &e1000_es2_info,
47 [board_ich8lan] = &e1000_ich8_info,
48 [board_ich9lan] = &e1000_ich9_info,
49 [board_ich10lan] = &e1000_ich10_info,
50 [board_pchlan] = &e1000_pch_info,
51 [board_pch2lan] = &e1000_pch2_info,
52 [board_pch_lpt] = &e1000_pch_lpt_info,
53 [board_pch_spt] = &e1000_pch_spt_info,
54 [board_pch_cnp] = &e1000_pch_cnp_info,
55 [board_pch_tgp] = &e1000_pch_tgp_info,
56 [board_pch_adp] = &e1000_pch_adp_info,
57 [board_pch_mtp] = &e1000_pch_mtp_info,
58};
59
60struct e1000_reg_info {
61 u32 ofs;
62 char *name;
63};
64
65static const struct e1000_reg_info e1000_reg_info_tbl[] = {
66 /* General Registers */
67 {E1000_CTRL, "CTRL"},
68 {E1000_STATUS, "STATUS"},
69 {E1000_CTRL_EXT, "CTRL_EXT"},
70
71 /* Interrupt Registers */
72 {E1000_ICR, "ICR"},
73
74 /* Rx Registers */
75 {E1000_RCTL, "RCTL"},
76 {E1000_RDLEN(0), "RDLEN"},
77 {E1000_RDH(0), "RDH"},
78 {E1000_RDT(0), "RDT"},
79 {E1000_RDTR, "RDTR"},
80 {E1000_RXDCTL(0), "RXDCTL"},
81 {E1000_ERT, "ERT"},
82 {E1000_RDBAL(0), "RDBAL"},
83 {E1000_RDBAH(0), "RDBAH"},
84 {E1000_RDFH, "RDFH"},
85 {E1000_RDFT, "RDFT"},
86 {E1000_RDFHS, "RDFHS"},
87 {E1000_RDFTS, "RDFTS"},
88 {E1000_RDFPC, "RDFPC"},
89
90 /* Tx Registers */
91 {E1000_TCTL, "TCTL"},
92 {E1000_TDBAL(0), "TDBAL"},
93 {E1000_TDBAH(0), "TDBAH"},
94 {E1000_TDLEN(0), "TDLEN"},
95 {E1000_TDH(0), "TDH"},
96 {E1000_TDT(0), "TDT"},
97 {E1000_TIDV, "TIDV"},
98 {E1000_TXDCTL(0), "TXDCTL"},
99 {E1000_TADV, "TADV"},
100 {E1000_TARC(0), "TARC"},
101 {E1000_TDFH, "TDFH"},
102 {E1000_TDFT, "TDFT"},
103 {E1000_TDFHS, "TDFHS"},
104 {E1000_TDFTS, "TDFTS"},
105 {E1000_TDFPC, "TDFPC"},
106
107 /* List Terminator */
108 {0, NULL}
109};
110
111/**
112 * __ew32_prepare - prepare to write to MAC CSR register on certain parts
113 * @hw: pointer to the HW structure
114 *
115 * When updating the MAC CSR registers, the Manageability Engine (ME) could
116 * be accessing the registers at the same time. Normally, this is handled in
117 * h/w by an arbiter but on some parts there is a bug that acknowledges Host
118 * accesses later than it should which could result in the register to have
119 * an incorrect value. Workaround this by checking the FWSM register which
120 * has bit 24 set while ME is accessing MAC CSR registers, wait if it is set
121 * and try again a number of times.
122 **/
123static void __ew32_prepare(struct e1000_hw *hw)
124{
125 s32 i = E1000_ICH_FWSM_PCIM2PCI_COUNT;
126
127 while ((er32(FWSM) & E1000_ICH_FWSM_PCIM2PCI) && --i)
128 udelay(50);
129}
130
131void __ew32(struct e1000_hw *hw, unsigned long reg, u32 val)
132{
133 if (hw->adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
134 __ew32_prepare(hw);
135
136 writel(val, hw->hw_addr + reg);
137}
138
139/**
140 * e1000_regdump - register printout routine
141 * @hw: pointer to the HW structure
142 * @reginfo: pointer to the register info table
143 **/
144static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
145{
146 int n = 0;
147 char rname[16];
148 u32 regs[8];
149
150 switch (reginfo->ofs) {
151 case E1000_RXDCTL(0):
152 for (n = 0; n < 2; n++)
153 regs[n] = __er32(hw, E1000_RXDCTL(n));
154 break;
155 case E1000_TXDCTL(0):
156 for (n = 0; n < 2; n++)
157 regs[n] = __er32(hw, E1000_TXDCTL(n));
158 break;
159 case E1000_TARC(0):
160 for (n = 0; n < 2; n++)
161 regs[n] = __er32(hw, E1000_TARC(n));
162 break;
163 default:
164 pr_info("%-15s %08x\n",
165 reginfo->name, __er32(hw, reginfo->ofs));
166 return;
167 }
168
169 snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
170 pr_info("%-15s %08x %08x\n", rname, regs[0], regs[1]);
171}
172
173static void e1000e_dump_ps_pages(struct e1000_adapter *adapter,
174 struct e1000_buffer *bi)
175{
176 int i;
177 struct e1000_ps_page *ps_page;
178
179 for (i = 0; i < adapter->rx_ps_pages; i++) {
180 ps_page = &bi->ps_pages[i];
181
182 if (ps_page->page) {
183 pr_info("packet dump for ps_page %d:\n", i);
184 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
185 16, 1, page_address(ps_page->page),
186 PAGE_SIZE, true);
187 }
188 }
189}
190
191/**
192 * e1000e_dump - Print registers, Tx-ring and Rx-ring
193 * @adapter: board private structure
194 **/
195static void e1000e_dump(struct e1000_adapter *adapter)
196{
197 struct net_device *netdev = adapter->netdev;
198 struct e1000_hw *hw = &adapter->hw;
199 struct e1000_reg_info *reginfo;
200 struct e1000_ring *tx_ring = adapter->tx_ring;
201 struct e1000_tx_desc *tx_desc;
202 struct my_u0 {
203 __le64 a;
204 __le64 b;
205 } *u0;
206 struct e1000_buffer *buffer_info;
207 struct e1000_ring *rx_ring = adapter->rx_ring;
208 union e1000_rx_desc_packet_split *rx_desc_ps;
209 union e1000_rx_desc_extended *rx_desc;
210 struct my_u1 {
211 __le64 a;
212 __le64 b;
213 __le64 c;
214 __le64 d;
215 } *u1;
216 u32 staterr;
217 int i = 0;
218
219 if (!netif_msg_hw(adapter))
220 return;
221
222 /* Print netdevice Info */
223 if (netdev) {
224 dev_info(&adapter->pdev->dev, "Net device Info\n");
225 pr_info("Device Name state trans_start\n");
226 pr_info("%-15s %016lX %016lX\n", netdev->name,
227 netdev->state, dev_trans_start(netdev));
228 }
229
230 /* Print Registers */
231 dev_info(&adapter->pdev->dev, "Register Dump\n");
232 pr_info(" Register Name Value\n");
233 for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
234 reginfo->name; reginfo++) {
235 e1000_regdump(hw, reginfo);
236 }
237
238 /* Print Tx Ring Summary */
239 if (!netdev || !netif_running(netdev))
240 return;
241
242 dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
243 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
244 buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
245 pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
246 0, tx_ring->next_to_use, tx_ring->next_to_clean,
247 (unsigned long long)buffer_info->dma,
248 buffer_info->length,
249 buffer_info->next_to_watch,
250 (unsigned long long)buffer_info->time_stamp);
251
252 /* Print Tx Ring */
253 if (!netif_msg_tx_done(adapter))
254 goto rx_ring_summary;
255
256 dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
257
258 /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
259 *
260 * Legacy Transmit Descriptor
261 * +--------------------------------------------------------------+
262 * 0 | Buffer Address [63:0] (Reserved on Write Back) |
263 * +--------------------------------------------------------------+
264 * 8 | Special | CSS | Status | CMD | CSO | Length |
265 * +--------------------------------------------------------------+
266 * 63 48 47 36 35 32 31 24 23 16 15 0
267 *
268 * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
269 * 63 48 47 40 39 32 31 16 15 8 7 0
270 * +----------------------------------------------------------------+
271 * 0 | TUCSE | TUCS0 | TUCSS | IPCSE | IPCS0 | IPCSS |
272 * +----------------------------------------------------------------+
273 * 8 | MSS | HDRLEN | RSV | STA | TUCMD | DTYP | PAYLEN |
274 * +----------------------------------------------------------------+
275 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
276 *
277 * Extended Data Descriptor (DTYP=0x1)
278 * +----------------------------------------------------------------+
279 * 0 | Buffer Address [63:0] |
280 * +----------------------------------------------------------------+
281 * 8 | VLAN tag | POPTS | Rsvd | Status | Command | DTYP | DTALEN |
282 * +----------------------------------------------------------------+
283 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
284 */
285 pr_info("Tl[desc] [address 63:0 ] [SpeCssSCmCsLen] [bi->dma ] leng ntw timestamp bi->skb <-- Legacy format\n");
286 pr_info("Tc[desc] [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma ] leng ntw timestamp bi->skb <-- Ext Context format\n");
287 pr_info("Td[desc] [address 63:0 ] [VlaPoRSCm1Dlen] [bi->dma ] leng ntw timestamp bi->skb <-- Ext Data format\n");
288 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
289 const char *next_desc;
290 tx_desc = E1000_TX_DESC(*tx_ring, i);
291 buffer_info = &tx_ring->buffer_info[i];
292 u0 = (struct my_u0 *)tx_desc;
293 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
294 next_desc = " NTC/U";
295 else if (i == tx_ring->next_to_use)
296 next_desc = " NTU";
297 else if (i == tx_ring->next_to_clean)
298 next_desc = " NTC";
299 else
300 next_desc = "";
301 pr_info("T%c[0x%03X] %016llX %016llX %016llX %04X %3X %016llX %p%s\n",
302 (!(le64_to_cpu(u0->b) & BIT(29)) ? 'l' :
303 ((le64_to_cpu(u0->b) & BIT(20)) ? 'd' : 'c')),
304 i,
305 (unsigned long long)le64_to_cpu(u0->a),
306 (unsigned long long)le64_to_cpu(u0->b),
307 (unsigned long long)buffer_info->dma,
308 buffer_info->length, buffer_info->next_to_watch,
309 (unsigned long long)buffer_info->time_stamp,
310 buffer_info->skb, next_desc);
311
312 if (netif_msg_pktdata(adapter) && buffer_info->skb)
313 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
314 16, 1, buffer_info->skb->data,
315 buffer_info->skb->len, true);
316 }
317
318 /* Print Rx Ring Summary */
319rx_ring_summary:
320 dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
321 pr_info("Queue [NTU] [NTC]\n");
322 pr_info(" %5d %5X %5X\n",
323 0, rx_ring->next_to_use, rx_ring->next_to_clean);
324
325 /* Print Rx Ring */
326 if (!netif_msg_rx_status(adapter))
327 return;
328
329 dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
330 switch (adapter->rx_ps_pages) {
331 case 1:
332 case 2:
333 case 3:
334 /* [Extended] Packet Split Receive Descriptor Format
335 *
336 * +-----------------------------------------------------+
337 * 0 | Buffer Address 0 [63:0] |
338 * +-----------------------------------------------------+
339 * 8 | Buffer Address 1 [63:0] |
340 * +-----------------------------------------------------+
341 * 16 | Buffer Address 2 [63:0] |
342 * +-----------------------------------------------------+
343 * 24 | Buffer Address 3 [63:0] |
344 * +-----------------------------------------------------+
345 */
346 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");
347 /* [Extended] Receive Descriptor (Write-Back) Format
348 *
349 * 63 48 47 32 31 13 12 8 7 4 3 0
350 * +------------------------------------------------------+
351 * 0 | Packet | IP | Rsvd | MRQ | Rsvd | MRQ RSS |
352 * | Checksum | Ident | | Queue | | Type |
353 * +------------------------------------------------------+
354 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
355 * +------------------------------------------------------+
356 * 63 48 47 32 31 20 19 0
357 */
358 pr_info("RWB[desc] [ck ipid mrqhsh] [vl l0 ee es] [ l3 l2 l1 hs] [reserved ] ---------------- [bi->skb] <-- Ext Rx Write-Back format\n");
359 for (i = 0; i < rx_ring->count; i++) {
360 const char *next_desc;
361 buffer_info = &rx_ring->buffer_info[i];
362 rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
363 u1 = (struct my_u1 *)rx_desc_ps;
364 staterr =
365 le32_to_cpu(rx_desc_ps->wb.middle.status_error);
366
367 if (i == rx_ring->next_to_use)
368 next_desc = " NTU";
369 else if (i == rx_ring->next_to_clean)
370 next_desc = " NTC";
371 else
372 next_desc = "";
373
374 if (staterr & E1000_RXD_STAT_DD) {
375 /* Descriptor Done */
376 pr_info("%s[0x%03X] %016llX %016llX %016llX %016llX ---------------- %p%s\n",
377 "RWB", i,
378 (unsigned long long)le64_to_cpu(u1->a),
379 (unsigned long long)le64_to_cpu(u1->b),
380 (unsigned long long)le64_to_cpu(u1->c),
381 (unsigned long long)le64_to_cpu(u1->d),
382 buffer_info->skb, next_desc);
383 } else {
384 pr_info("%s[0x%03X] %016llX %016llX %016llX %016llX %016llX %p%s\n",
385 "R ", i,
386 (unsigned long long)le64_to_cpu(u1->a),
387 (unsigned long long)le64_to_cpu(u1->b),
388 (unsigned long long)le64_to_cpu(u1->c),
389 (unsigned long long)le64_to_cpu(u1->d),
390 (unsigned long long)buffer_info->dma,
391 buffer_info->skb, next_desc);
392
393 if (netif_msg_pktdata(adapter))
394 e1000e_dump_ps_pages(adapter,
395 buffer_info);
396 }
397 }
398 break;
399 default:
400 case 0:
401 /* Extended Receive Descriptor (Read) Format
402 *
403 * +-----------------------------------------------------+
404 * 0 | Buffer Address [63:0] |
405 * +-----------------------------------------------------+
406 * 8 | Reserved |
407 * +-----------------------------------------------------+
408 */
409 pr_info("R [desc] [buf addr 63:0 ] [reserved 63:0 ] [bi->dma ] [bi->skb] <-- Ext (Read) format\n");
410 /* Extended Receive Descriptor (Write-Back) Format
411 *
412 * 63 48 47 32 31 24 23 4 3 0
413 * +------------------------------------------------------+
414 * | RSS Hash | | | |
415 * 0 +-------------------+ Rsvd | Reserved | MRQ RSS |
416 * | Packet | IP | | | Type |
417 * | Checksum | Ident | | | |
418 * +------------------------------------------------------+
419 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
420 * +------------------------------------------------------+
421 * 63 48 47 32 31 20 19 0
422 */
423 pr_info("RWB[desc] [cs ipid mrq] [vt ln xe xs] [bi->skb] <-- Ext (Write-Back) format\n");
424
425 for (i = 0; i < rx_ring->count; i++) {
426 const char *next_desc;
427
428 buffer_info = &rx_ring->buffer_info[i];
429 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
430 u1 = (struct my_u1 *)rx_desc;
431 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
432
433 if (i == rx_ring->next_to_use)
434 next_desc = " NTU";
435 else if (i == rx_ring->next_to_clean)
436 next_desc = " NTC";
437 else
438 next_desc = "";
439
440 if (staterr & E1000_RXD_STAT_DD) {
441 /* Descriptor Done */
442 pr_info("%s[0x%03X] %016llX %016llX ---------------- %p%s\n",
443 "RWB", i,
444 (unsigned long long)le64_to_cpu(u1->a),
445 (unsigned long long)le64_to_cpu(u1->b),
446 buffer_info->skb, next_desc);
447 } else {
448 pr_info("%s[0x%03X] %016llX %016llX %016llX %p%s\n",
449 "R ", i,
450 (unsigned long long)le64_to_cpu(u1->a),
451 (unsigned long long)le64_to_cpu(u1->b),
452 (unsigned long long)buffer_info->dma,
453 buffer_info->skb, next_desc);
454
455 if (netif_msg_pktdata(adapter) &&
456 buffer_info->skb)
457 print_hex_dump(KERN_INFO, "",
458 DUMP_PREFIX_ADDRESS, 16,
459 1,
460 buffer_info->skb->data,
461 adapter->rx_buffer_len,
462 true);
463 }
464 }
465 }
466}
467
468/**
469 * e1000_desc_unused - calculate if we have unused descriptors
470 * @ring: pointer to ring struct to perform calculation on
471 **/
472static int e1000_desc_unused(struct e1000_ring *ring)
473{
474 if (ring->next_to_clean > ring->next_to_use)
475 return ring->next_to_clean - ring->next_to_use - 1;
476
477 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
478}
479
480/**
481 * e1000e_systim_to_hwtstamp - convert system time value to hw time stamp
482 * @adapter: board private structure
483 * @hwtstamps: time stamp structure to update
484 * @systim: unsigned 64bit system time value.
485 *
486 * Convert the system time value stored in the RX/TXSTMP registers into a
487 * hwtstamp which can be used by the upper level time stamping functions.
488 *
489 * The 'systim_lock' spinlock is used to protect the consistency of the
490 * system time value. This is needed because reading the 64 bit time
491 * value involves reading two 32 bit registers. The first read latches the
492 * value.
493 **/
494static void e1000e_systim_to_hwtstamp(struct e1000_adapter *adapter,
495 struct skb_shared_hwtstamps *hwtstamps,
496 u64 systim)
497{
498 u64 ns;
499 unsigned long flags;
500
501 spin_lock_irqsave(&adapter->systim_lock, flags);
502 ns = timecounter_cyc2time(&adapter->tc, systim);
503 spin_unlock_irqrestore(&adapter->systim_lock, flags);
504
505 memset(hwtstamps, 0, sizeof(*hwtstamps));
506 hwtstamps->hwtstamp = ns_to_ktime(ns);
507}
508
509/**
510 * e1000e_rx_hwtstamp - utility function which checks for Rx time stamp
511 * @adapter: board private structure
512 * @status: descriptor extended error and status field
513 * @skb: particular skb to include time stamp
514 *
515 * If the time stamp is valid, convert it into the timecounter ns value
516 * and store that result into the shhwtstamps structure which is passed
517 * up the network stack.
518 **/
519static void e1000e_rx_hwtstamp(struct e1000_adapter *adapter, u32 status,
520 struct sk_buff *skb)
521{
522 struct e1000_hw *hw = &adapter->hw;
523 u64 rxstmp;
524
525 if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP) ||
526 !(status & E1000_RXDEXT_STATERR_TST) ||
527 !(er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID))
528 return;
529
530 /* The Rx time stamp registers contain the time stamp. No other
531 * received packet will be time stamped until the Rx time stamp
532 * registers are read. Because only one packet can be time stamped
533 * at a time, the register values must belong to this packet and
534 * therefore none of the other additional attributes need to be
535 * compared.
536 */
537 rxstmp = (u64)er32(RXSTMPL);
538 rxstmp |= (u64)er32(RXSTMPH) << 32;
539 e1000e_systim_to_hwtstamp(adapter, skb_hwtstamps(skb), rxstmp);
540
541 adapter->flags2 &= ~FLAG2_CHECK_RX_HWTSTAMP;
542}
543
544/**
545 * e1000_receive_skb - helper function to handle Rx indications
546 * @adapter: board private structure
547 * @netdev: pointer to netdev struct
548 * @staterr: descriptor extended error and status field as written by hardware
549 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
550 * @skb: pointer to sk_buff to be indicated to stack
551 **/
552static void e1000_receive_skb(struct e1000_adapter *adapter,
553 struct net_device *netdev, struct sk_buff *skb,
554 u32 staterr, __le16 vlan)
555{
556 u16 tag = le16_to_cpu(vlan);
557
558 e1000e_rx_hwtstamp(adapter, staterr, skb);
559
560 skb->protocol = eth_type_trans(skb, netdev);
561
562 if (staterr & E1000_RXD_STAT_VP)
563 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tag);
564
565 napi_gro_receive(&adapter->napi, skb);
566}
567
568/**
569 * e1000_rx_checksum - Receive Checksum Offload
570 * @adapter: board private structure
571 * @status_err: receive descriptor status and error fields
572 * @skb: socket buffer with received data
573 **/
574static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
575 struct sk_buff *skb)
576{
577 u16 status = (u16)status_err;
578 u8 errors = (u8)(status_err >> 24);
579
580 skb_checksum_none_assert(skb);
581
582 /* Rx checksum disabled */
583 if (!(adapter->netdev->features & NETIF_F_RXCSUM))
584 return;
585
586 /* Ignore Checksum bit is set */
587 if (status & E1000_RXD_STAT_IXSM)
588 return;
589
590 /* TCP/UDP checksum error bit or IP checksum error bit is set */
591 if (errors & (E1000_RXD_ERR_TCPE | E1000_RXD_ERR_IPE)) {
592 /* let the stack verify checksum errors */
593 adapter->hw_csum_err++;
594 return;
595 }
596
597 /* TCP/UDP Checksum has not been calculated */
598 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
599 return;
600
601 /* It must be a TCP or UDP packet with a valid checksum */
602 skb->ip_summed = CHECKSUM_UNNECESSARY;
603 adapter->hw_csum_good++;
604}
605
606static void e1000e_update_rdt_wa(struct e1000_ring *rx_ring, unsigned int i)
607{
608 struct e1000_adapter *adapter = rx_ring->adapter;
609 struct e1000_hw *hw = &adapter->hw;
610
611 __ew32_prepare(hw);
612 writel(i, rx_ring->tail);
613
614 if (unlikely(i != readl(rx_ring->tail))) {
615 u32 rctl = er32(RCTL);
616
617 ew32(RCTL, rctl & ~E1000_RCTL_EN);
618 e_err("ME firmware caused invalid RDT - resetting\n");
619 schedule_work(&adapter->reset_task);
620 }
621}
622
623static void e1000e_update_tdt_wa(struct e1000_ring *tx_ring, unsigned int i)
624{
625 struct e1000_adapter *adapter = tx_ring->adapter;
626 struct e1000_hw *hw = &adapter->hw;
627
628 __ew32_prepare(hw);
629 writel(i, tx_ring->tail);
630
631 if (unlikely(i != readl(tx_ring->tail))) {
632 u32 tctl = er32(TCTL);
633
634 ew32(TCTL, tctl & ~E1000_TCTL_EN);
635 e_err("ME firmware caused invalid TDT - resetting\n");
636 schedule_work(&adapter->reset_task);
637 }
638}
639
640/**
641 * e1000_alloc_rx_buffers - Replace used receive buffers
642 * @rx_ring: Rx descriptor ring
643 * @cleaned_count: number to reallocate
644 * @gfp: flags for allocation
645 **/
646static void e1000_alloc_rx_buffers(struct e1000_ring *rx_ring,
647 int cleaned_count, gfp_t gfp)
648{
649 struct e1000_adapter *adapter = rx_ring->adapter;
650 struct net_device *netdev = adapter->netdev;
651 struct pci_dev *pdev = adapter->pdev;
652 union e1000_rx_desc_extended *rx_desc;
653 struct e1000_buffer *buffer_info;
654 struct sk_buff *skb;
655 unsigned int i;
656 unsigned int bufsz = adapter->rx_buffer_len;
657
658 i = rx_ring->next_to_use;
659 buffer_info = &rx_ring->buffer_info[i];
660
661 while (cleaned_count--) {
662 skb = buffer_info->skb;
663 if (skb) {
664 skb_trim(skb, 0);
665 goto map_skb;
666 }
667
668 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
669 if (!skb) {
670 /* Better luck next round */
671 adapter->alloc_rx_buff_failed++;
672 break;
673 }
674
675 buffer_info->skb = skb;
676map_skb:
677 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
678 adapter->rx_buffer_len,
679 DMA_FROM_DEVICE);
680 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
681 dev_err(&pdev->dev, "Rx DMA map failed\n");
682 adapter->rx_dma_failed++;
683 break;
684 }
685
686 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
687 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
688
689 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
690 /* Force memory writes to complete before letting h/w
691 * know there are new descriptors to fetch. (Only
692 * applicable for weak-ordered memory model archs,
693 * such as IA-64).
694 */
695 wmb();
696 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
697 e1000e_update_rdt_wa(rx_ring, i);
698 else
699 writel(i, rx_ring->tail);
700 }
701 i++;
702 if (i == rx_ring->count)
703 i = 0;
704 buffer_info = &rx_ring->buffer_info[i];
705 }
706
707 rx_ring->next_to_use = i;
708}
709
710/**
711 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
712 * @rx_ring: Rx descriptor ring
713 * @cleaned_count: number to reallocate
714 * @gfp: flags for allocation
715 **/
716static void e1000_alloc_rx_buffers_ps(struct e1000_ring *rx_ring,
717 int cleaned_count, gfp_t gfp)
718{
719 struct e1000_adapter *adapter = rx_ring->adapter;
720 struct net_device *netdev = adapter->netdev;
721 struct pci_dev *pdev = adapter->pdev;
722 union e1000_rx_desc_packet_split *rx_desc;
723 struct e1000_buffer *buffer_info;
724 struct e1000_ps_page *ps_page;
725 struct sk_buff *skb;
726 unsigned int i, j;
727
728 i = rx_ring->next_to_use;
729 buffer_info = &rx_ring->buffer_info[i];
730
731 while (cleaned_count--) {
732 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
733
734 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
735 ps_page = &buffer_info->ps_pages[j];
736 if (j >= adapter->rx_ps_pages) {
737 /* all unused desc entries get hw null ptr */
738 rx_desc->read.buffer_addr[j + 1] =
739 ~cpu_to_le64(0);
740 continue;
741 }
742 if (!ps_page->page) {
743 ps_page->page = alloc_page(gfp);
744 if (!ps_page->page) {
745 adapter->alloc_rx_buff_failed++;
746 goto no_buffers;
747 }
748 ps_page->dma = dma_map_page(&pdev->dev,
749 ps_page->page,
750 0, PAGE_SIZE,
751 DMA_FROM_DEVICE);
752 if (dma_mapping_error(&pdev->dev,
753 ps_page->dma)) {
754 dev_err(&adapter->pdev->dev,
755 "Rx DMA page map failed\n");
756 adapter->rx_dma_failed++;
757 goto no_buffers;
758 }
759 }
760 /* Refresh the desc even if buffer_addrs
761 * didn't change because each write-back
762 * erases this info.
763 */
764 rx_desc->read.buffer_addr[j + 1] =
765 cpu_to_le64(ps_page->dma);
766 }
767
768 skb = __netdev_alloc_skb_ip_align(netdev, adapter->rx_ps_bsize0,
769 gfp);
770
771 if (!skb) {
772 adapter->alloc_rx_buff_failed++;
773 break;
774 }
775
776 buffer_info->skb = skb;
777 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
778 adapter->rx_ps_bsize0,
779 DMA_FROM_DEVICE);
780 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
781 dev_err(&pdev->dev, "Rx DMA map failed\n");
782 adapter->rx_dma_failed++;
783 /* cleanup skb */
784 dev_kfree_skb_any(skb);
785 buffer_info->skb = NULL;
786 break;
787 }
788
789 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
790
791 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
792 /* Force memory writes to complete before letting h/w
793 * know there are new descriptors to fetch. (Only
794 * applicable for weak-ordered memory model archs,
795 * such as IA-64).
796 */
797 wmb();
798 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
799 e1000e_update_rdt_wa(rx_ring, i << 1);
800 else
801 writel(i << 1, rx_ring->tail);
802 }
803
804 i++;
805 if (i == rx_ring->count)
806 i = 0;
807 buffer_info = &rx_ring->buffer_info[i];
808 }
809
810no_buffers:
811 rx_ring->next_to_use = i;
812}
813
814/**
815 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
816 * @rx_ring: Rx descriptor ring
817 * @cleaned_count: number of buffers to allocate this pass
818 * @gfp: flags for allocation
819 **/
820
821static void e1000_alloc_jumbo_rx_buffers(struct e1000_ring *rx_ring,
822 int cleaned_count, gfp_t gfp)
823{
824 struct e1000_adapter *adapter = rx_ring->adapter;
825 struct net_device *netdev = adapter->netdev;
826 struct pci_dev *pdev = adapter->pdev;
827 union e1000_rx_desc_extended *rx_desc;
828 struct e1000_buffer *buffer_info;
829 struct sk_buff *skb;
830 unsigned int i;
831 unsigned int bufsz = 256 - 16; /* for skb_reserve */
832
833 i = rx_ring->next_to_use;
834 buffer_info = &rx_ring->buffer_info[i];
835
836 while (cleaned_count--) {
837 skb = buffer_info->skb;
838 if (skb) {
839 skb_trim(skb, 0);
840 goto check_page;
841 }
842
843 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
844 if (unlikely(!skb)) {
845 /* Better luck next round */
846 adapter->alloc_rx_buff_failed++;
847 break;
848 }
849
850 buffer_info->skb = skb;
851check_page:
852 /* allocate a new page if necessary */
853 if (!buffer_info->page) {
854 buffer_info->page = alloc_page(gfp);
855 if (unlikely(!buffer_info->page)) {
856 adapter->alloc_rx_buff_failed++;
857 break;
858 }
859 }
860
861 if (!buffer_info->dma) {
862 buffer_info->dma = dma_map_page(&pdev->dev,
863 buffer_info->page, 0,
864 PAGE_SIZE,
865 DMA_FROM_DEVICE);
866 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
867 adapter->alloc_rx_buff_failed++;
868 break;
869 }
870 }
871
872 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
873 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
874
875 if (unlikely(++i == rx_ring->count))
876 i = 0;
877 buffer_info = &rx_ring->buffer_info[i];
878 }
879
880 if (likely(rx_ring->next_to_use != i)) {
881 rx_ring->next_to_use = i;
882 if (unlikely(i-- == 0))
883 i = (rx_ring->count - 1);
884
885 /* Force memory writes to complete before letting h/w
886 * know there are new descriptors to fetch. (Only
887 * applicable for weak-ordered memory model archs,
888 * such as IA-64).
889 */
890 wmb();
891 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
892 e1000e_update_rdt_wa(rx_ring, i);
893 else
894 writel(i, rx_ring->tail);
895 }
896}
897
898static inline void e1000_rx_hash(struct net_device *netdev, __le32 rss,
899 struct sk_buff *skb)
900{
901 if (netdev->features & NETIF_F_RXHASH)
902 skb_set_hash(skb, le32_to_cpu(rss), PKT_HASH_TYPE_L3);
903}
904
905/**
906 * e1000_clean_rx_irq - Send received data up the network stack
907 * @rx_ring: Rx descriptor ring
908 * @work_done: output parameter for indicating completed work
909 * @work_to_do: how many packets we can clean
910 *
911 * the return value indicates whether actual cleaning was done, there
912 * is no guarantee that everything was cleaned
913 **/
914static bool e1000_clean_rx_irq(struct e1000_ring *rx_ring, int *work_done,
915 int work_to_do)
916{
917 struct e1000_adapter *adapter = rx_ring->adapter;
918 struct net_device *netdev = adapter->netdev;
919 struct pci_dev *pdev = adapter->pdev;
920 struct e1000_hw *hw = &adapter->hw;
921 union e1000_rx_desc_extended *rx_desc, *next_rxd;
922 struct e1000_buffer *buffer_info, *next_buffer;
923 u32 length, staterr;
924 unsigned int i;
925 int cleaned_count = 0;
926 bool cleaned = false;
927 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
928
929 i = rx_ring->next_to_clean;
930 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
931 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
932 buffer_info = &rx_ring->buffer_info[i];
933
934 while (staterr & E1000_RXD_STAT_DD) {
935 struct sk_buff *skb;
936
937 if (*work_done >= work_to_do)
938 break;
939 (*work_done)++;
940 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
941
942 skb = buffer_info->skb;
943 buffer_info->skb = NULL;
944
945 prefetch(skb->data - NET_IP_ALIGN);
946
947 i++;
948 if (i == rx_ring->count)
949 i = 0;
950 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
951 prefetch(next_rxd);
952
953 next_buffer = &rx_ring->buffer_info[i];
954
955 cleaned = true;
956 cleaned_count++;
957 dma_unmap_single(&pdev->dev, buffer_info->dma,
958 adapter->rx_buffer_len, DMA_FROM_DEVICE);
959 buffer_info->dma = 0;
960
961 length = le16_to_cpu(rx_desc->wb.upper.length);
962
963 /* !EOP means multiple descriptors were used to store a single
964 * packet, if that's the case we need to toss it. In fact, we
965 * need to toss every packet with the EOP bit clear and the
966 * next frame that _does_ have the EOP bit set, as it is by
967 * definition only a frame fragment
968 */
969 if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
970 adapter->flags2 |= FLAG2_IS_DISCARDING;
971
972 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
973 /* All receives must fit into a single buffer */
974 e_dbg("Receive packet consumed multiple buffers\n");
975 /* recycle */
976 buffer_info->skb = skb;
977 if (staterr & E1000_RXD_STAT_EOP)
978 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
979 goto next_desc;
980 }
981
982 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
983 !(netdev->features & NETIF_F_RXALL))) {
984 /* recycle */
985 buffer_info->skb = skb;
986 goto next_desc;
987 }
988
989 /* adjust length to remove Ethernet CRC */
990 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
991 /* If configured to store CRC, don't subtract FCS,
992 * but keep the FCS bytes out of the total_rx_bytes
993 * counter
994 */
995 if (netdev->features & NETIF_F_RXFCS)
996 total_rx_bytes -= 4;
997 else
998 length -= 4;
999 }
1000
1001 total_rx_bytes += length;
1002 total_rx_packets++;
1003
1004 /* code added for copybreak, this should improve
1005 * performance for small packets with large amounts
1006 * of reassembly being done in the stack
1007 */
1008 if (length < copybreak) {
1009 struct sk_buff *new_skb =
1010 napi_alloc_skb(&adapter->napi, length);
1011 if (new_skb) {
1012 skb_copy_to_linear_data_offset(new_skb,
1013 -NET_IP_ALIGN,
1014 (skb->data -
1015 NET_IP_ALIGN),
1016 (length +
1017 NET_IP_ALIGN));
1018 /* save the skb in buffer_info as good */
1019 buffer_info->skb = skb;
1020 skb = new_skb;
1021 }
1022 /* else just continue with the old one */
1023 }
1024 /* end copybreak code */
1025 skb_put(skb, length);
1026
1027 /* Receive Checksum Offload */
1028 e1000_rx_checksum(adapter, staterr, skb);
1029
1030 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1031
1032 e1000_receive_skb(adapter, netdev, skb, staterr,
1033 rx_desc->wb.upper.vlan);
1034
1035next_desc:
1036 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1037
1038 /* return some buffers to hardware, one at a time is too slow */
1039 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1040 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1041 GFP_ATOMIC);
1042 cleaned_count = 0;
1043 }
1044
1045 /* use prefetched values */
1046 rx_desc = next_rxd;
1047 buffer_info = next_buffer;
1048
1049 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1050 }
1051 rx_ring->next_to_clean = i;
1052
1053 cleaned_count = e1000_desc_unused(rx_ring);
1054 if (cleaned_count)
1055 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1056
1057 adapter->total_rx_bytes += total_rx_bytes;
1058 adapter->total_rx_packets += total_rx_packets;
1059 return cleaned;
1060}
1061
1062static void e1000_put_txbuf(struct e1000_ring *tx_ring,
1063 struct e1000_buffer *buffer_info,
1064 bool drop)
1065{
1066 struct e1000_adapter *adapter = tx_ring->adapter;
1067
1068 if (buffer_info->dma) {
1069 if (buffer_info->mapped_as_page)
1070 dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1071 buffer_info->length, DMA_TO_DEVICE);
1072 else
1073 dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1074 buffer_info->length, DMA_TO_DEVICE);
1075 buffer_info->dma = 0;
1076 }
1077 if (buffer_info->skb) {
1078 if (drop)
1079 dev_kfree_skb_any(buffer_info->skb);
1080 else
1081 dev_consume_skb_any(buffer_info->skb);
1082 buffer_info->skb = NULL;
1083 }
1084 buffer_info->time_stamp = 0;
1085}
1086
1087static void e1000_print_hw_hang(struct work_struct *work)
1088{
1089 struct e1000_adapter *adapter = container_of(work,
1090 struct e1000_adapter,
1091 print_hang_task);
1092 struct net_device *netdev = adapter->netdev;
1093 struct e1000_ring *tx_ring = adapter->tx_ring;
1094 unsigned int i = tx_ring->next_to_clean;
1095 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1096 struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1097 struct e1000_hw *hw = &adapter->hw;
1098 u16 phy_status, phy_1000t_status, phy_ext_status;
1099 u16 pci_status;
1100
1101 if (test_bit(__E1000_DOWN, &adapter->state))
1102 return;
1103
1104 if (!adapter->tx_hang_recheck && (adapter->flags2 & FLAG2_DMA_BURST)) {
1105 /* May be block on write-back, flush and detect again
1106 * flush pending descriptor writebacks to memory
1107 */
1108 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1109 /* execute the writes immediately */
1110 e1e_flush();
1111 /* Due to rare timing issues, write to TIDV again to ensure
1112 * the write is successful
1113 */
1114 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1115 /* execute the writes immediately */
1116 e1e_flush();
1117 adapter->tx_hang_recheck = true;
1118 return;
1119 }
1120 adapter->tx_hang_recheck = false;
1121
1122 if (er32(TDH(0)) == er32(TDT(0))) {
1123 e_dbg("false hang detected, ignoring\n");
1124 return;
1125 }
1126
1127 /* Real hang detected */
1128 netif_stop_queue(netdev);
1129
1130 e1e_rphy(hw, MII_BMSR, &phy_status);
1131 e1e_rphy(hw, MII_STAT1000, &phy_1000t_status);
1132 e1e_rphy(hw, MII_ESTATUS, &phy_ext_status);
1133
1134 pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1135
1136 /* detected Hardware unit hang */
1137 e_err("Detected Hardware Unit Hang:\n"
1138 " TDH <%x>\n"
1139 " TDT <%x>\n"
1140 " next_to_use <%x>\n"
1141 " next_to_clean <%x>\n"
1142 "buffer_info[next_to_clean]:\n"
1143 " time_stamp <%lx>\n"
1144 " next_to_watch <%x>\n"
1145 " jiffies <%lx>\n"
1146 " next_to_watch.status <%x>\n"
1147 "MAC Status <%x>\n"
1148 "PHY Status <%x>\n"
1149 "PHY 1000BASE-T Status <%x>\n"
1150 "PHY Extended Status <%x>\n"
1151 "PCI Status <%x>\n",
1152 readl(tx_ring->head), readl(tx_ring->tail), tx_ring->next_to_use,
1153 tx_ring->next_to_clean, tx_ring->buffer_info[eop].time_stamp,
1154 eop, jiffies, eop_desc->upper.fields.status, er32(STATUS),
1155 phy_status, phy_1000t_status, phy_ext_status, pci_status);
1156
1157 e1000e_dump(adapter);
1158
1159 /* Suggest workaround for known h/w issue */
1160 if ((hw->mac.type == e1000_pchlan) && (er32(CTRL) & E1000_CTRL_TFCE))
1161 e_err("Try turning off Tx pause (flow control) via ethtool\n");
1162}
1163
1164/**
1165 * e1000e_tx_hwtstamp_work - check for Tx time stamp
1166 * @work: pointer to work struct
1167 *
1168 * This work function polls the TSYNCTXCTL valid bit to determine when a
1169 * timestamp has been taken for the current stored skb. The timestamp must
1170 * be for this skb because only one such packet is allowed in the queue.
1171 */
1172static void e1000e_tx_hwtstamp_work(struct work_struct *work)
1173{
1174 struct e1000_adapter *adapter = container_of(work, struct e1000_adapter,
1175 tx_hwtstamp_work);
1176 struct e1000_hw *hw = &adapter->hw;
1177
1178 if (er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID) {
1179 struct sk_buff *skb = adapter->tx_hwtstamp_skb;
1180 struct skb_shared_hwtstamps shhwtstamps;
1181 u64 txstmp;
1182
1183 txstmp = er32(TXSTMPL);
1184 txstmp |= (u64)er32(TXSTMPH) << 32;
1185
1186 e1000e_systim_to_hwtstamp(adapter, &shhwtstamps, txstmp);
1187
1188 /* Clear the global tx_hwtstamp_skb pointer and force writes
1189 * prior to notifying the stack of a Tx timestamp.
1190 */
1191 adapter->tx_hwtstamp_skb = NULL;
1192 wmb(); /* force write prior to skb_tstamp_tx */
1193
1194 skb_tstamp_tx(skb, &shhwtstamps);
1195 dev_consume_skb_any(skb);
1196 } else if (time_after(jiffies, adapter->tx_hwtstamp_start
1197 + adapter->tx_timeout_factor * HZ)) {
1198 dev_kfree_skb_any(adapter->tx_hwtstamp_skb);
1199 adapter->tx_hwtstamp_skb = NULL;
1200 adapter->tx_hwtstamp_timeouts++;
1201 e_warn("clearing Tx timestamp hang\n");
1202 } else {
1203 /* reschedule to check later */
1204 schedule_work(&adapter->tx_hwtstamp_work);
1205 }
1206}
1207
1208/**
1209 * e1000_clean_tx_irq - Reclaim resources after transmit completes
1210 * @tx_ring: Tx descriptor ring
1211 *
1212 * the return value indicates whether actual cleaning was done, there
1213 * is no guarantee that everything was cleaned
1214 **/
1215static bool e1000_clean_tx_irq(struct e1000_ring *tx_ring)
1216{
1217 struct e1000_adapter *adapter = tx_ring->adapter;
1218 struct net_device *netdev = adapter->netdev;
1219 struct e1000_hw *hw = &adapter->hw;
1220 struct e1000_tx_desc *tx_desc, *eop_desc;
1221 struct e1000_buffer *buffer_info;
1222 unsigned int i, eop;
1223 unsigned int count = 0;
1224 unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1225 unsigned int bytes_compl = 0, pkts_compl = 0;
1226
1227 i = tx_ring->next_to_clean;
1228 eop = tx_ring->buffer_info[i].next_to_watch;
1229 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1230
1231 while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1232 (count < tx_ring->count)) {
1233 bool cleaned = false;
1234
1235 dma_rmb(); /* read buffer_info after eop_desc */
1236 for (; !cleaned; count++) {
1237 tx_desc = E1000_TX_DESC(*tx_ring, i);
1238 buffer_info = &tx_ring->buffer_info[i];
1239 cleaned = (i == eop);
1240
1241 if (cleaned) {
1242 total_tx_packets += buffer_info->segs;
1243 total_tx_bytes += buffer_info->bytecount;
1244 if (buffer_info->skb) {
1245 bytes_compl += buffer_info->skb->len;
1246 pkts_compl++;
1247 }
1248 }
1249
1250 e1000_put_txbuf(tx_ring, buffer_info, false);
1251 tx_desc->upper.data = 0;
1252
1253 i++;
1254 if (i == tx_ring->count)
1255 i = 0;
1256 }
1257
1258 if (i == tx_ring->next_to_use)
1259 break;
1260 eop = tx_ring->buffer_info[i].next_to_watch;
1261 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1262 }
1263
1264 tx_ring->next_to_clean = i;
1265
1266 netdev_completed_queue(netdev, pkts_compl, bytes_compl);
1267
1268#define TX_WAKE_THRESHOLD 32
1269 if (count && netif_carrier_ok(netdev) &&
1270 e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1271 /* Make sure that anybody stopping the queue after this
1272 * sees the new next_to_clean.
1273 */
1274 smp_mb();
1275
1276 if (netif_queue_stopped(netdev) &&
1277 !(test_bit(__E1000_DOWN, &adapter->state))) {
1278 netif_wake_queue(netdev);
1279 ++adapter->restart_queue;
1280 }
1281 }
1282
1283 if (adapter->detect_tx_hung) {
1284 /* Detect a transmit hang in hardware, this serializes the
1285 * check with the clearing of time_stamp and movement of i
1286 */
1287 adapter->detect_tx_hung = false;
1288 if (tx_ring->buffer_info[i].time_stamp &&
1289 time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1290 + (adapter->tx_timeout_factor * HZ)) &&
1291 !(er32(STATUS) & E1000_STATUS_TXOFF))
1292 schedule_work(&adapter->print_hang_task);
1293 else
1294 adapter->tx_hang_recheck = false;
1295 }
1296 adapter->total_tx_bytes += total_tx_bytes;
1297 adapter->total_tx_packets += total_tx_packets;
1298 return count < tx_ring->count;
1299}
1300
1301/**
1302 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1303 * @rx_ring: Rx descriptor ring
1304 * @work_done: output parameter for indicating completed work
1305 * @work_to_do: how many packets we can clean
1306 *
1307 * the return value indicates whether actual cleaning was done, there
1308 * is no guarantee that everything was cleaned
1309 **/
1310static bool e1000_clean_rx_irq_ps(struct e1000_ring *rx_ring, int *work_done,
1311 int work_to_do)
1312{
1313 struct e1000_adapter *adapter = rx_ring->adapter;
1314 struct e1000_hw *hw = &adapter->hw;
1315 union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1316 struct net_device *netdev = adapter->netdev;
1317 struct pci_dev *pdev = adapter->pdev;
1318 struct e1000_buffer *buffer_info, *next_buffer;
1319 struct e1000_ps_page *ps_page;
1320 struct sk_buff *skb;
1321 unsigned int i, j;
1322 u32 length, staterr;
1323 int cleaned_count = 0;
1324 bool cleaned = false;
1325 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1326
1327 i = rx_ring->next_to_clean;
1328 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1329 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1330 buffer_info = &rx_ring->buffer_info[i];
1331
1332 while (staterr & E1000_RXD_STAT_DD) {
1333 if (*work_done >= work_to_do)
1334 break;
1335 (*work_done)++;
1336 skb = buffer_info->skb;
1337 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
1338
1339 /* in the packet split case this is header only */
1340 prefetch(skb->data - NET_IP_ALIGN);
1341
1342 i++;
1343 if (i == rx_ring->count)
1344 i = 0;
1345 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1346 prefetch(next_rxd);
1347
1348 next_buffer = &rx_ring->buffer_info[i];
1349
1350 cleaned = true;
1351 cleaned_count++;
1352 dma_unmap_single(&pdev->dev, buffer_info->dma,
1353 adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1354 buffer_info->dma = 0;
1355
1356 /* see !EOP comment in other Rx routine */
1357 if (!(staterr & E1000_RXD_STAT_EOP))
1358 adapter->flags2 |= FLAG2_IS_DISCARDING;
1359
1360 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1361 e_dbg("Packet Split buffers didn't pick up the full packet\n");
1362 dev_kfree_skb_irq(skb);
1363 if (staterr & E1000_RXD_STAT_EOP)
1364 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1365 goto next_desc;
1366 }
1367
1368 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1369 !(netdev->features & NETIF_F_RXALL))) {
1370 dev_kfree_skb_irq(skb);
1371 goto next_desc;
1372 }
1373
1374 length = le16_to_cpu(rx_desc->wb.middle.length0);
1375
1376 if (!length) {
1377 e_dbg("Last part of the packet spanning multiple descriptors\n");
1378 dev_kfree_skb_irq(skb);
1379 goto next_desc;
1380 }
1381
1382 /* Good Receive */
1383 skb_put(skb, length);
1384
1385 {
1386 /* this looks ugly, but it seems compiler issues make
1387 * it more efficient than reusing j
1388 */
1389 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1390
1391 /* page alloc/put takes too long and effects small
1392 * packet throughput, so unsplit small packets and
1393 * save the alloc/put
1394 */
1395 if (l1 && (l1 <= copybreak) &&
1396 ((length + l1) <= adapter->rx_ps_bsize0)) {
1397 ps_page = &buffer_info->ps_pages[0];
1398
1399 dma_sync_single_for_cpu(&pdev->dev,
1400 ps_page->dma,
1401 PAGE_SIZE,
1402 DMA_FROM_DEVICE);
1403 memcpy(skb_tail_pointer(skb),
1404 page_address(ps_page->page), l1);
1405 dma_sync_single_for_device(&pdev->dev,
1406 ps_page->dma,
1407 PAGE_SIZE,
1408 DMA_FROM_DEVICE);
1409
1410 /* remove the CRC */
1411 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1412 if (!(netdev->features & NETIF_F_RXFCS))
1413 l1 -= 4;
1414 }
1415
1416 skb_put(skb, l1);
1417 goto copydone;
1418 } /* if */
1419 }
1420
1421 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1422 length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1423 if (!length)
1424 break;
1425
1426 ps_page = &buffer_info->ps_pages[j];
1427 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1428 DMA_FROM_DEVICE);
1429 ps_page->dma = 0;
1430 skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1431 ps_page->page = NULL;
1432 skb->len += length;
1433 skb->data_len += length;
1434 skb->truesize += PAGE_SIZE;
1435 }
1436
1437 /* strip the ethernet crc, problem is we're using pages now so
1438 * this whole operation can get a little cpu intensive
1439 */
1440 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1441 if (!(netdev->features & NETIF_F_RXFCS))
1442 pskb_trim(skb, skb->len - 4);
1443 }
1444
1445copydone:
1446 total_rx_bytes += skb->len;
1447 total_rx_packets++;
1448
1449 e1000_rx_checksum(adapter, staterr, skb);
1450
1451 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1452
1453 if (rx_desc->wb.upper.header_status &
1454 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1455 adapter->rx_hdr_split++;
1456
1457 e1000_receive_skb(adapter, netdev, skb, staterr,
1458 rx_desc->wb.middle.vlan);
1459
1460next_desc:
1461 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1462 buffer_info->skb = NULL;
1463
1464 /* return some buffers to hardware, one at a time is too slow */
1465 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1466 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1467 GFP_ATOMIC);
1468 cleaned_count = 0;
1469 }
1470
1471 /* use prefetched values */
1472 rx_desc = next_rxd;
1473 buffer_info = next_buffer;
1474
1475 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1476 }
1477 rx_ring->next_to_clean = i;
1478
1479 cleaned_count = e1000_desc_unused(rx_ring);
1480 if (cleaned_count)
1481 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1482
1483 adapter->total_rx_bytes += total_rx_bytes;
1484 adapter->total_rx_packets += total_rx_packets;
1485 return cleaned;
1486}
1487
1488static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1489 u16 length)
1490{
1491 bi->page = NULL;
1492 skb->len += length;
1493 skb->data_len += length;
1494 skb->truesize += PAGE_SIZE;
1495}
1496
1497/**
1498 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1499 * @rx_ring: Rx descriptor ring
1500 * @work_done: output parameter for indicating completed work
1501 * @work_to_do: how many packets we can clean
1502 *
1503 * the return value indicates whether actual cleaning was done, there
1504 * is no guarantee that everything was cleaned
1505 **/
1506static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
1507 int work_to_do)
1508{
1509 struct e1000_adapter *adapter = rx_ring->adapter;
1510 struct net_device *netdev = adapter->netdev;
1511 struct pci_dev *pdev = adapter->pdev;
1512 union e1000_rx_desc_extended *rx_desc, *next_rxd;
1513 struct e1000_buffer *buffer_info, *next_buffer;
1514 u32 length, staterr;
1515 unsigned int i;
1516 int cleaned_count = 0;
1517 bool cleaned = false;
1518 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1519 struct skb_shared_info *shinfo;
1520
1521 i = rx_ring->next_to_clean;
1522 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1523 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1524 buffer_info = &rx_ring->buffer_info[i];
1525
1526 while (staterr & E1000_RXD_STAT_DD) {
1527 struct sk_buff *skb;
1528
1529 if (*work_done >= work_to_do)
1530 break;
1531 (*work_done)++;
1532 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
1533
1534 skb = buffer_info->skb;
1535 buffer_info->skb = NULL;
1536
1537 ++i;
1538 if (i == rx_ring->count)
1539 i = 0;
1540 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1541 prefetch(next_rxd);
1542
1543 next_buffer = &rx_ring->buffer_info[i];
1544
1545 cleaned = true;
1546 cleaned_count++;
1547 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1548 DMA_FROM_DEVICE);
1549 buffer_info->dma = 0;
1550
1551 length = le16_to_cpu(rx_desc->wb.upper.length);
1552
1553 /* errors is only valid for DD + EOP descriptors */
1554 if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1555 ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1556 !(netdev->features & NETIF_F_RXALL)))) {
1557 /* recycle both page and skb */
1558 buffer_info->skb = skb;
1559 /* an error means any chain goes out the window too */
1560 if (rx_ring->rx_skb_top)
1561 dev_kfree_skb_irq(rx_ring->rx_skb_top);
1562 rx_ring->rx_skb_top = NULL;
1563 goto next_desc;
1564 }
1565#define rxtop (rx_ring->rx_skb_top)
1566 if (!(staterr & E1000_RXD_STAT_EOP)) {
1567 /* this descriptor is only the beginning (or middle) */
1568 if (!rxtop) {
1569 /* this is the beginning of a chain */
1570 rxtop = skb;
1571 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1572 0, length);
1573 } else {
1574 /* this is the middle of a chain */
1575 shinfo = skb_shinfo(rxtop);
1576 skb_fill_page_desc(rxtop, shinfo->nr_frags,
1577 buffer_info->page, 0,
1578 length);
1579 /* re-use the skb, only consumed the page */
1580 buffer_info->skb = skb;
1581 }
1582 e1000_consume_page(buffer_info, rxtop, length);
1583 goto next_desc;
1584 } else {
1585 if (rxtop) {
1586 /* end of the chain */
1587 shinfo = skb_shinfo(rxtop);
1588 skb_fill_page_desc(rxtop, shinfo->nr_frags,
1589 buffer_info->page, 0,
1590 length);
1591 /* re-use the current skb, we only consumed the
1592 * page
1593 */
1594 buffer_info->skb = skb;
1595 skb = rxtop;
1596 rxtop = NULL;
1597 e1000_consume_page(buffer_info, skb, length);
1598 } else {
1599 /* no chain, got EOP, this buf is the packet
1600 * copybreak to save the put_page/alloc_page
1601 */
1602 if (length <= copybreak &&
1603 skb_tailroom(skb) >= length) {
1604 memcpy(skb_tail_pointer(skb),
1605 page_address(buffer_info->page),
1606 length);
1607 /* re-use the page, so don't erase
1608 * buffer_info->page
1609 */
1610 skb_put(skb, length);
1611 } else {
1612 skb_fill_page_desc(skb, 0,
1613 buffer_info->page, 0,
1614 length);
1615 e1000_consume_page(buffer_info, skb,
1616 length);
1617 }
1618 }
1619 }
1620
1621 /* Receive Checksum Offload */
1622 e1000_rx_checksum(adapter, staterr, skb);
1623
1624 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1625
1626 /* probably a little skewed due to removing CRC */
1627 total_rx_bytes += skb->len;
1628 total_rx_packets++;
1629
1630 /* eth type trans needs skb->data to point to something */
1631 if (!pskb_may_pull(skb, ETH_HLEN)) {
1632 e_err("pskb_may_pull failed.\n");
1633 dev_kfree_skb_irq(skb);
1634 goto next_desc;
1635 }
1636
1637 e1000_receive_skb(adapter, netdev, skb, staterr,
1638 rx_desc->wb.upper.vlan);
1639
1640next_desc:
1641 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1642
1643 /* return some buffers to hardware, one at a time is too slow */
1644 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1645 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1646 GFP_ATOMIC);
1647 cleaned_count = 0;
1648 }
1649
1650 /* use prefetched values */
1651 rx_desc = next_rxd;
1652 buffer_info = next_buffer;
1653
1654 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1655 }
1656 rx_ring->next_to_clean = i;
1657
1658 cleaned_count = e1000_desc_unused(rx_ring);
1659 if (cleaned_count)
1660 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1661
1662 adapter->total_rx_bytes += total_rx_bytes;
1663 adapter->total_rx_packets += total_rx_packets;
1664 return cleaned;
1665}
1666
1667/**
1668 * e1000_clean_rx_ring - Free Rx Buffers per Queue
1669 * @rx_ring: Rx descriptor ring
1670 **/
1671static void e1000_clean_rx_ring(struct e1000_ring *rx_ring)
1672{
1673 struct e1000_adapter *adapter = rx_ring->adapter;
1674 struct e1000_buffer *buffer_info;
1675 struct e1000_ps_page *ps_page;
1676 struct pci_dev *pdev = adapter->pdev;
1677 unsigned int i, j;
1678
1679 /* Free all the Rx ring sk_buffs */
1680 for (i = 0; i < rx_ring->count; i++) {
1681 buffer_info = &rx_ring->buffer_info[i];
1682 if (buffer_info->dma) {
1683 if (adapter->clean_rx == e1000_clean_rx_irq)
1684 dma_unmap_single(&pdev->dev, buffer_info->dma,
1685 adapter->rx_buffer_len,
1686 DMA_FROM_DEVICE);
1687 else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1688 dma_unmap_page(&pdev->dev, buffer_info->dma,
1689 PAGE_SIZE, DMA_FROM_DEVICE);
1690 else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1691 dma_unmap_single(&pdev->dev, buffer_info->dma,
1692 adapter->rx_ps_bsize0,
1693 DMA_FROM_DEVICE);
1694 buffer_info->dma = 0;
1695 }
1696
1697 if (buffer_info->page) {
1698 put_page(buffer_info->page);
1699 buffer_info->page = NULL;
1700 }
1701
1702 if (buffer_info->skb) {
1703 dev_kfree_skb(buffer_info->skb);
1704 buffer_info->skb = NULL;
1705 }
1706
1707 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1708 ps_page = &buffer_info->ps_pages[j];
1709 if (!ps_page->page)
1710 break;
1711 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1712 DMA_FROM_DEVICE);
1713 ps_page->dma = 0;
1714 put_page(ps_page->page);
1715 ps_page->page = NULL;
1716 }
1717 }
1718
1719 /* there also may be some cached data from a chained receive */
1720 if (rx_ring->rx_skb_top) {
1721 dev_kfree_skb(rx_ring->rx_skb_top);
1722 rx_ring->rx_skb_top = NULL;
1723 }
1724
1725 /* Zero out the descriptor ring */
1726 memset(rx_ring->desc, 0, rx_ring->size);
1727
1728 rx_ring->next_to_clean = 0;
1729 rx_ring->next_to_use = 0;
1730 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1731}
1732
1733static void e1000e_downshift_workaround(struct work_struct *work)
1734{
1735 struct e1000_adapter *adapter = container_of(work,
1736 struct e1000_adapter,
1737 downshift_task);
1738
1739 if (test_bit(__E1000_DOWN, &adapter->state))
1740 return;
1741
1742 e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1743}
1744
1745/**
1746 * e1000_intr_msi - Interrupt Handler
1747 * @irq: interrupt number
1748 * @data: pointer to a network interface device structure
1749 **/
1750static irqreturn_t e1000_intr_msi(int __always_unused irq, void *data)
1751{
1752 struct net_device *netdev = data;
1753 struct e1000_adapter *adapter = netdev_priv(netdev);
1754 struct e1000_hw *hw = &adapter->hw;
1755 u32 icr = er32(ICR);
1756
1757 /* read ICR disables interrupts using IAM */
1758 if (icr & E1000_ICR_LSC) {
1759 hw->mac.get_link_status = true;
1760 /* ICH8 workaround-- Call gig speed drop workaround on cable
1761 * disconnect (LSC) before accessing any PHY registers
1762 */
1763 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1764 (!(er32(STATUS) & E1000_STATUS_LU)))
1765 schedule_work(&adapter->downshift_task);
1766
1767 /* 80003ES2LAN workaround-- For packet buffer work-around on
1768 * link down event; disable receives here in the ISR and reset
1769 * adapter in watchdog
1770 */
1771 if (netif_carrier_ok(netdev) &&
1772 adapter->flags & FLAG_RX_NEEDS_RESTART) {
1773 /* disable receives */
1774 u32 rctl = er32(RCTL);
1775
1776 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1777 adapter->flags |= FLAG_RESTART_NOW;
1778 }
1779 /* guard against interrupt when we're going down */
1780 if (!test_bit(__E1000_DOWN, &adapter->state))
1781 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1782 }
1783
1784 /* Reset on uncorrectable ECC error */
1785 if ((icr & E1000_ICR_ECCER) && (hw->mac.type >= e1000_pch_lpt)) {
1786 u32 pbeccsts = er32(PBECCSTS);
1787
1788 adapter->corr_errors +=
1789 pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1790 adapter->uncorr_errors +=
1791 FIELD_GET(E1000_PBECCSTS_UNCORR_ERR_CNT_MASK, pbeccsts);
1792
1793 /* Do the reset outside of interrupt context */
1794 schedule_work(&adapter->reset_task);
1795
1796 /* return immediately since reset is imminent */
1797 return IRQ_HANDLED;
1798 }
1799
1800 if (napi_schedule_prep(&adapter->napi)) {
1801 adapter->total_tx_bytes = 0;
1802 adapter->total_tx_packets = 0;
1803 adapter->total_rx_bytes = 0;
1804 adapter->total_rx_packets = 0;
1805 __napi_schedule(&adapter->napi);
1806 }
1807
1808 return IRQ_HANDLED;
1809}
1810
1811/**
1812 * e1000_intr - Interrupt Handler
1813 * @irq: interrupt number
1814 * @data: pointer to a network interface device structure
1815 **/
1816static irqreturn_t e1000_intr(int __always_unused irq, void *data)
1817{
1818 struct net_device *netdev = data;
1819 struct e1000_adapter *adapter = netdev_priv(netdev);
1820 struct e1000_hw *hw = &adapter->hw;
1821 u32 rctl, icr = er32(ICR);
1822
1823 if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1824 return IRQ_NONE; /* Not our interrupt */
1825
1826 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1827 * not set, then the adapter didn't send an interrupt
1828 */
1829 if (!(icr & E1000_ICR_INT_ASSERTED))
1830 return IRQ_NONE;
1831
1832 /* Interrupt Auto-Mask...upon reading ICR,
1833 * interrupts are masked. No need for the
1834 * IMC write
1835 */
1836
1837 if (icr & E1000_ICR_LSC) {
1838 hw->mac.get_link_status = true;
1839 /* ICH8 workaround-- Call gig speed drop workaround on cable
1840 * disconnect (LSC) before accessing any PHY registers
1841 */
1842 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1843 (!(er32(STATUS) & E1000_STATUS_LU)))
1844 schedule_work(&adapter->downshift_task);
1845
1846 /* 80003ES2LAN workaround--
1847 * For packet buffer work-around on link down event;
1848 * disable receives here in the ISR and
1849 * reset adapter in watchdog
1850 */
1851 if (netif_carrier_ok(netdev) &&
1852 (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1853 /* disable receives */
1854 rctl = er32(RCTL);
1855 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1856 adapter->flags |= FLAG_RESTART_NOW;
1857 }
1858 /* guard against interrupt when we're going down */
1859 if (!test_bit(__E1000_DOWN, &adapter->state))
1860 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1861 }
1862
1863 /* Reset on uncorrectable ECC error */
1864 if ((icr & E1000_ICR_ECCER) && (hw->mac.type >= e1000_pch_lpt)) {
1865 u32 pbeccsts = er32(PBECCSTS);
1866
1867 adapter->corr_errors +=
1868 pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1869 adapter->uncorr_errors +=
1870 FIELD_GET(E1000_PBECCSTS_UNCORR_ERR_CNT_MASK, pbeccsts);
1871
1872 /* Do the reset outside of interrupt context */
1873 schedule_work(&adapter->reset_task);
1874
1875 /* return immediately since reset is imminent */
1876 return IRQ_HANDLED;
1877 }
1878
1879 if (napi_schedule_prep(&adapter->napi)) {
1880 adapter->total_tx_bytes = 0;
1881 adapter->total_tx_packets = 0;
1882 adapter->total_rx_bytes = 0;
1883 adapter->total_rx_packets = 0;
1884 __napi_schedule(&adapter->napi);
1885 }
1886
1887 return IRQ_HANDLED;
1888}
1889
1890static irqreturn_t e1000_msix_other(int __always_unused irq, void *data)
1891{
1892 struct net_device *netdev = data;
1893 struct e1000_adapter *adapter = netdev_priv(netdev);
1894 struct e1000_hw *hw = &adapter->hw;
1895 u32 icr = er32(ICR);
1896
1897 if (icr & adapter->eiac_mask)
1898 ew32(ICS, (icr & adapter->eiac_mask));
1899
1900 if (icr & E1000_ICR_LSC) {
1901 hw->mac.get_link_status = true;
1902 /* guard against interrupt when we're going down */
1903 if (!test_bit(__E1000_DOWN, &adapter->state))
1904 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1905 }
1906
1907 if (!test_bit(__E1000_DOWN, &adapter->state))
1908 ew32(IMS, E1000_IMS_OTHER | IMS_OTHER_MASK);
1909
1910 return IRQ_HANDLED;
1911}
1912
1913static irqreturn_t e1000_intr_msix_tx(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 struct e1000_ring *tx_ring = adapter->tx_ring;
1919
1920 adapter->total_tx_bytes = 0;
1921 adapter->total_tx_packets = 0;
1922
1923 if (!e1000_clean_tx_irq(tx_ring))
1924 /* Ring was not completely cleaned, so fire another interrupt */
1925 ew32(ICS, tx_ring->ims_val);
1926
1927 if (!test_bit(__E1000_DOWN, &adapter->state))
1928 ew32(IMS, adapter->tx_ring->ims_val);
1929
1930 return IRQ_HANDLED;
1931}
1932
1933static irqreturn_t e1000_intr_msix_rx(int __always_unused irq, void *data)
1934{
1935 struct net_device *netdev = data;
1936 struct e1000_adapter *adapter = netdev_priv(netdev);
1937 struct e1000_ring *rx_ring = adapter->rx_ring;
1938
1939 /* Write the ITR value calculated at the end of the
1940 * previous interrupt.
1941 */
1942 if (rx_ring->set_itr) {
1943 u32 itr = rx_ring->itr_val ?
1944 1000000000 / (rx_ring->itr_val * 256) : 0;
1945
1946 writel(itr, rx_ring->itr_register);
1947 rx_ring->set_itr = 0;
1948 }
1949
1950 if (napi_schedule_prep(&adapter->napi)) {
1951 adapter->total_rx_bytes = 0;
1952 adapter->total_rx_packets = 0;
1953 __napi_schedule(&adapter->napi);
1954 }
1955 return IRQ_HANDLED;
1956}
1957
1958/**
1959 * e1000_configure_msix - Configure MSI-X hardware
1960 * @adapter: board private structure
1961 *
1962 * e1000_configure_msix sets up the hardware to properly
1963 * generate MSI-X interrupts.
1964 **/
1965static void e1000_configure_msix(struct e1000_adapter *adapter)
1966{
1967 struct e1000_hw *hw = &adapter->hw;
1968 struct e1000_ring *rx_ring = adapter->rx_ring;
1969 struct e1000_ring *tx_ring = adapter->tx_ring;
1970 int vector = 0;
1971 u32 ctrl_ext, ivar = 0;
1972
1973 adapter->eiac_mask = 0;
1974
1975 /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1976 if (hw->mac.type == e1000_82574) {
1977 u32 rfctl = er32(RFCTL);
1978
1979 rfctl |= E1000_RFCTL_ACK_DIS;
1980 ew32(RFCTL, rfctl);
1981 }
1982
1983 /* Configure Rx vector */
1984 rx_ring->ims_val = E1000_IMS_RXQ0;
1985 adapter->eiac_mask |= rx_ring->ims_val;
1986 if (rx_ring->itr_val)
1987 writel(1000000000 / (rx_ring->itr_val * 256),
1988 rx_ring->itr_register);
1989 else
1990 writel(1, rx_ring->itr_register);
1991 ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1992
1993 /* Configure Tx vector */
1994 tx_ring->ims_val = E1000_IMS_TXQ0;
1995 vector++;
1996 if (tx_ring->itr_val)
1997 writel(1000000000 / (tx_ring->itr_val * 256),
1998 tx_ring->itr_register);
1999 else
2000 writel(1, tx_ring->itr_register);
2001 adapter->eiac_mask |= tx_ring->ims_val;
2002 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
2003
2004 /* set vector for Other Causes, e.g. link changes */
2005 vector++;
2006 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
2007 if (rx_ring->itr_val)
2008 writel(1000000000 / (rx_ring->itr_val * 256),
2009 hw->hw_addr + E1000_EITR_82574(vector));
2010 else
2011 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
2012
2013 /* Cause Tx interrupts on every write back */
2014 ivar |= BIT(31);
2015
2016 ew32(IVAR, ivar);
2017
2018 /* enable MSI-X PBA support */
2019 ctrl_ext = er32(CTRL_EXT) & ~E1000_CTRL_EXT_IAME;
2020 ctrl_ext |= E1000_CTRL_EXT_PBA_CLR | E1000_CTRL_EXT_EIAME;
2021 ew32(CTRL_EXT, ctrl_ext);
2022 e1e_flush();
2023}
2024
2025void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
2026{
2027 if (adapter->msix_entries) {
2028 pci_disable_msix(adapter->pdev);
2029 kfree(adapter->msix_entries);
2030 adapter->msix_entries = NULL;
2031 } else if (adapter->flags & FLAG_MSI_ENABLED) {
2032 pci_disable_msi(adapter->pdev);
2033 adapter->flags &= ~FLAG_MSI_ENABLED;
2034 }
2035}
2036
2037/**
2038 * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
2039 * @adapter: board private structure
2040 *
2041 * Attempt to configure interrupts using the best available
2042 * capabilities of the hardware and kernel.
2043 **/
2044void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
2045{
2046 int err;
2047 int i;
2048
2049 switch (adapter->int_mode) {
2050 case E1000E_INT_MODE_MSIX:
2051 if (adapter->flags & FLAG_HAS_MSIX) {
2052 adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
2053 adapter->msix_entries = kcalloc(adapter->num_vectors,
2054 sizeof(struct
2055 msix_entry),
2056 GFP_KERNEL);
2057 if (adapter->msix_entries) {
2058 struct e1000_adapter *a = adapter;
2059
2060 for (i = 0; i < adapter->num_vectors; i++)
2061 adapter->msix_entries[i].entry = i;
2062
2063 err = pci_enable_msix_range(a->pdev,
2064 a->msix_entries,
2065 a->num_vectors,
2066 a->num_vectors);
2067 if (err > 0)
2068 return;
2069 }
2070 /* MSI-X failed, so fall through and try MSI */
2071 e_err("Failed to initialize MSI-X interrupts. Falling back to MSI interrupts.\n");
2072 e1000e_reset_interrupt_capability(adapter);
2073 }
2074 adapter->int_mode = E1000E_INT_MODE_MSI;
2075 fallthrough;
2076 case E1000E_INT_MODE_MSI:
2077 if (!pci_enable_msi(adapter->pdev)) {
2078 adapter->flags |= FLAG_MSI_ENABLED;
2079 } else {
2080 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2081 e_err("Failed to initialize MSI interrupts. Falling back to legacy interrupts.\n");
2082 }
2083 fallthrough;
2084 case E1000E_INT_MODE_LEGACY:
2085 /* Don't do anything; this is the system default */
2086 break;
2087 }
2088
2089 /* store the number of vectors being used */
2090 adapter->num_vectors = 1;
2091}
2092
2093/**
2094 * e1000_request_msix - Initialize MSI-X interrupts
2095 * @adapter: board private structure
2096 *
2097 * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
2098 * kernel.
2099 **/
2100static int e1000_request_msix(struct e1000_adapter *adapter)
2101{
2102 struct net_device *netdev = adapter->netdev;
2103 int err = 0, vector = 0;
2104
2105 if (strlen(netdev->name) < (IFNAMSIZ - 5))
2106 snprintf(adapter->rx_ring->name,
2107 sizeof(adapter->rx_ring->name) - 1,
2108 "%.14s-rx-0", netdev->name);
2109 else
2110 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
2111 err = request_irq(adapter->msix_entries[vector].vector,
2112 e1000_intr_msix_rx, 0, adapter->rx_ring->name,
2113 netdev);
2114 if (err)
2115 return err;
2116 adapter->rx_ring->itr_register = adapter->hw.hw_addr +
2117 E1000_EITR_82574(vector);
2118 adapter->rx_ring->itr_val = adapter->itr;
2119 vector++;
2120
2121 if (strlen(netdev->name) < (IFNAMSIZ - 5))
2122 snprintf(adapter->tx_ring->name,
2123 sizeof(adapter->tx_ring->name) - 1,
2124 "%.14s-tx-0", netdev->name);
2125 else
2126 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
2127 err = request_irq(adapter->msix_entries[vector].vector,
2128 e1000_intr_msix_tx, 0, adapter->tx_ring->name,
2129 netdev);
2130 if (err)
2131 return err;
2132 adapter->tx_ring->itr_register = adapter->hw.hw_addr +
2133 E1000_EITR_82574(vector);
2134 adapter->tx_ring->itr_val = adapter->itr;
2135 vector++;
2136
2137 err = request_irq(adapter->msix_entries[vector].vector,
2138 e1000_msix_other, 0, netdev->name, netdev);
2139 if (err)
2140 return err;
2141
2142 e1000_configure_msix(adapter);
2143
2144 return 0;
2145}
2146
2147/**
2148 * e1000_request_irq - initialize interrupts
2149 * @adapter: board private structure
2150 *
2151 * Attempts to configure interrupts using the best available
2152 * capabilities of the hardware and kernel.
2153 **/
2154static int e1000_request_irq(struct e1000_adapter *adapter)
2155{
2156 struct net_device *netdev = adapter->netdev;
2157 int err;
2158
2159 if (adapter->msix_entries) {
2160 err = e1000_request_msix(adapter);
2161 if (!err)
2162 return err;
2163 /* fall back to MSI */
2164 e1000e_reset_interrupt_capability(adapter);
2165 adapter->int_mode = E1000E_INT_MODE_MSI;
2166 e1000e_set_interrupt_capability(adapter);
2167 }
2168 if (adapter->flags & FLAG_MSI_ENABLED) {
2169 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2170 netdev->name, netdev);
2171 if (!err)
2172 return err;
2173
2174 /* fall back to legacy interrupt */
2175 e1000e_reset_interrupt_capability(adapter);
2176 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2177 }
2178
2179 err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2180 netdev->name, netdev);
2181 if (err)
2182 e_err("Unable to allocate interrupt, Error: %d\n", err);
2183
2184 return err;
2185}
2186
2187static void e1000_free_irq(struct e1000_adapter *adapter)
2188{
2189 struct net_device *netdev = adapter->netdev;
2190
2191 if (adapter->msix_entries) {
2192 int vector = 0;
2193
2194 free_irq(adapter->msix_entries[vector].vector, netdev);
2195 vector++;
2196
2197 free_irq(adapter->msix_entries[vector].vector, netdev);
2198 vector++;
2199
2200 /* Other Causes interrupt vector */
2201 free_irq(adapter->msix_entries[vector].vector, netdev);
2202 return;
2203 }
2204
2205 free_irq(adapter->pdev->irq, netdev);
2206}
2207
2208/**
2209 * e1000_irq_disable - Mask off interrupt generation on the NIC
2210 * @adapter: board private structure
2211 **/
2212static void e1000_irq_disable(struct e1000_adapter *adapter)
2213{
2214 struct e1000_hw *hw = &adapter->hw;
2215
2216 ew32(IMC, ~0);
2217 if (adapter->msix_entries)
2218 ew32(EIAC_82574, 0);
2219 e1e_flush();
2220
2221 if (adapter->msix_entries) {
2222 int i;
2223
2224 for (i = 0; i < adapter->num_vectors; i++)
2225 synchronize_irq(adapter->msix_entries[i].vector);
2226 } else {
2227 synchronize_irq(adapter->pdev->irq);
2228 }
2229}
2230
2231/**
2232 * e1000_irq_enable - Enable default interrupt generation settings
2233 * @adapter: board private structure
2234 **/
2235static void e1000_irq_enable(struct e1000_adapter *adapter)
2236{
2237 struct e1000_hw *hw = &adapter->hw;
2238
2239 if (adapter->msix_entries) {
2240 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2241 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER |
2242 IMS_OTHER_MASK);
2243 } else if (hw->mac.type >= e1000_pch_lpt) {
2244 ew32(IMS, IMS_ENABLE_MASK | E1000_IMS_ECCER);
2245 } else {
2246 ew32(IMS, IMS_ENABLE_MASK);
2247 }
2248 e1e_flush();
2249}
2250
2251/**
2252 * e1000e_get_hw_control - get control of the h/w from f/w
2253 * @adapter: address of board private structure
2254 *
2255 * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2256 * For ASF and Pass Through versions of f/w this means that
2257 * the driver is loaded. For AMT version (only with 82573)
2258 * of the f/w this means that the network i/f is open.
2259 **/
2260void e1000e_get_hw_control(struct e1000_adapter *adapter)
2261{
2262 struct e1000_hw *hw = &adapter->hw;
2263 u32 ctrl_ext;
2264 u32 swsm;
2265
2266 /* Let firmware know the driver has taken over */
2267 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2268 swsm = er32(SWSM);
2269 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2270 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2271 ctrl_ext = er32(CTRL_EXT);
2272 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2273 }
2274}
2275
2276/**
2277 * e1000e_release_hw_control - release control of the h/w to f/w
2278 * @adapter: address of board private structure
2279 *
2280 * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2281 * For ASF and Pass Through versions of f/w this means that the
2282 * driver is no longer loaded. For AMT version (only with 82573) i
2283 * of the f/w this means that the network i/f is closed.
2284 *
2285 **/
2286void e1000e_release_hw_control(struct e1000_adapter *adapter)
2287{
2288 struct e1000_hw *hw = &adapter->hw;
2289 u32 ctrl_ext;
2290 u32 swsm;
2291
2292 /* Let firmware taken over control of h/w */
2293 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2294 swsm = er32(SWSM);
2295 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2296 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2297 ctrl_ext = er32(CTRL_EXT);
2298 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2299 }
2300}
2301
2302/**
2303 * e1000_alloc_ring_dma - allocate memory for a ring structure
2304 * @adapter: board private structure
2305 * @ring: ring struct for which to allocate dma
2306 **/
2307static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2308 struct e1000_ring *ring)
2309{
2310 struct pci_dev *pdev = adapter->pdev;
2311
2312 ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2313 GFP_KERNEL);
2314 if (!ring->desc)
2315 return -ENOMEM;
2316
2317 return 0;
2318}
2319
2320/**
2321 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2322 * @tx_ring: Tx descriptor ring
2323 *
2324 * Return 0 on success, negative on failure
2325 **/
2326int e1000e_setup_tx_resources(struct e1000_ring *tx_ring)
2327{
2328 struct e1000_adapter *adapter = tx_ring->adapter;
2329 int err = -ENOMEM, size;
2330
2331 size = sizeof(struct e1000_buffer) * tx_ring->count;
2332 tx_ring->buffer_info = vzalloc(size);
2333 if (!tx_ring->buffer_info)
2334 goto err;
2335
2336 /* round up to nearest 4K */
2337 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2338 tx_ring->size = ALIGN(tx_ring->size, 4096);
2339
2340 err = e1000_alloc_ring_dma(adapter, tx_ring);
2341 if (err)
2342 goto err;
2343
2344 tx_ring->next_to_use = 0;
2345 tx_ring->next_to_clean = 0;
2346
2347 return 0;
2348err:
2349 vfree(tx_ring->buffer_info);
2350 e_err("Unable to allocate memory for the transmit descriptor ring\n");
2351 return err;
2352}
2353
2354/**
2355 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2356 * @rx_ring: Rx descriptor ring
2357 *
2358 * Returns 0 on success, negative on failure
2359 **/
2360int e1000e_setup_rx_resources(struct e1000_ring *rx_ring)
2361{
2362 struct e1000_adapter *adapter = rx_ring->adapter;
2363 struct e1000_buffer *buffer_info;
2364 int i, size, desc_len, err = -ENOMEM;
2365
2366 size = sizeof(struct e1000_buffer) * rx_ring->count;
2367 rx_ring->buffer_info = vzalloc(size);
2368 if (!rx_ring->buffer_info)
2369 goto err;
2370
2371 for (i = 0; i < rx_ring->count; i++) {
2372 buffer_info = &rx_ring->buffer_info[i];
2373 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2374 sizeof(struct e1000_ps_page),
2375 GFP_KERNEL);
2376 if (!buffer_info->ps_pages)
2377 goto err_pages;
2378 }
2379
2380 desc_len = sizeof(union e1000_rx_desc_packet_split);
2381
2382 /* Round up to nearest 4K */
2383 rx_ring->size = rx_ring->count * desc_len;
2384 rx_ring->size = ALIGN(rx_ring->size, 4096);
2385
2386 err = e1000_alloc_ring_dma(adapter, rx_ring);
2387 if (err)
2388 goto err_pages;
2389
2390 rx_ring->next_to_clean = 0;
2391 rx_ring->next_to_use = 0;
2392 rx_ring->rx_skb_top = NULL;
2393
2394 return 0;
2395
2396err_pages:
2397 for (i = 0; i < rx_ring->count; i++) {
2398 buffer_info = &rx_ring->buffer_info[i];
2399 kfree(buffer_info->ps_pages);
2400 }
2401err:
2402 vfree(rx_ring->buffer_info);
2403 e_err("Unable to allocate memory for the receive descriptor ring\n");
2404 return err;
2405}
2406
2407/**
2408 * e1000_clean_tx_ring - Free Tx Buffers
2409 * @tx_ring: Tx descriptor ring
2410 **/
2411static void e1000_clean_tx_ring(struct e1000_ring *tx_ring)
2412{
2413 struct e1000_adapter *adapter = tx_ring->adapter;
2414 struct e1000_buffer *buffer_info;
2415 unsigned long size;
2416 unsigned int i;
2417
2418 for (i = 0; i < tx_ring->count; i++) {
2419 buffer_info = &tx_ring->buffer_info[i];
2420 e1000_put_txbuf(tx_ring, buffer_info, false);
2421 }
2422
2423 netdev_reset_queue(adapter->netdev);
2424 size = sizeof(struct e1000_buffer) * tx_ring->count;
2425 memset(tx_ring->buffer_info, 0, size);
2426
2427 memset(tx_ring->desc, 0, tx_ring->size);
2428
2429 tx_ring->next_to_use = 0;
2430 tx_ring->next_to_clean = 0;
2431}
2432
2433/**
2434 * e1000e_free_tx_resources - Free Tx Resources per Queue
2435 * @tx_ring: Tx descriptor ring
2436 *
2437 * Free all transmit software resources
2438 **/
2439void e1000e_free_tx_resources(struct e1000_ring *tx_ring)
2440{
2441 struct e1000_adapter *adapter = tx_ring->adapter;
2442 struct pci_dev *pdev = adapter->pdev;
2443
2444 e1000_clean_tx_ring(tx_ring);
2445
2446 vfree(tx_ring->buffer_info);
2447 tx_ring->buffer_info = NULL;
2448
2449 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2450 tx_ring->dma);
2451 tx_ring->desc = NULL;
2452}
2453
2454/**
2455 * e1000e_free_rx_resources - Free Rx Resources
2456 * @rx_ring: Rx descriptor ring
2457 *
2458 * Free all receive software resources
2459 **/
2460void e1000e_free_rx_resources(struct e1000_ring *rx_ring)
2461{
2462 struct e1000_adapter *adapter = rx_ring->adapter;
2463 struct pci_dev *pdev = adapter->pdev;
2464 int i;
2465
2466 e1000_clean_rx_ring(rx_ring);
2467
2468 for (i = 0; i < rx_ring->count; i++)
2469 kfree(rx_ring->buffer_info[i].ps_pages);
2470
2471 vfree(rx_ring->buffer_info);
2472 rx_ring->buffer_info = NULL;
2473
2474 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2475 rx_ring->dma);
2476 rx_ring->desc = NULL;
2477}
2478
2479/**
2480 * e1000_update_itr - update the dynamic ITR value based on statistics
2481 * @itr_setting: current adapter->itr
2482 * @packets: the number of packets during this measurement interval
2483 * @bytes: the number of bytes during this measurement interval
2484 *
2485 * Stores a new ITR value based on packets and byte
2486 * counts during the last interrupt. The advantage of per interrupt
2487 * computation is faster updates and more accurate ITR for the current
2488 * traffic pattern. Constants in this function were computed
2489 * based on theoretical maximum wire speed and thresholds were set based
2490 * on testing data as well as attempting to minimize response time
2491 * while increasing bulk throughput. This functionality is controlled
2492 * by the InterruptThrottleRate module parameter.
2493 **/
2494static unsigned int e1000_update_itr(u16 itr_setting, int packets, int bytes)
2495{
2496 unsigned int retval = itr_setting;
2497
2498 if (packets == 0)
2499 return itr_setting;
2500
2501 switch (itr_setting) {
2502 case lowest_latency:
2503 /* handle TSO and jumbo frames */
2504 if (bytes / packets > 8000)
2505 retval = bulk_latency;
2506 else if ((packets < 5) && (bytes > 512))
2507 retval = low_latency;
2508 break;
2509 case low_latency: /* 50 usec aka 20000 ints/s */
2510 if (bytes > 10000) {
2511 /* this if handles the TSO accounting */
2512 if (bytes / packets > 8000)
2513 retval = bulk_latency;
2514 else if ((packets < 10) || ((bytes / packets) > 1200))
2515 retval = bulk_latency;
2516 else if ((packets > 35))
2517 retval = lowest_latency;
2518 } else if (bytes / packets > 2000) {
2519 retval = bulk_latency;
2520 } else if (packets <= 2 && bytes < 512) {
2521 retval = lowest_latency;
2522 }
2523 break;
2524 case bulk_latency: /* 250 usec aka 4000 ints/s */
2525 if (bytes > 25000) {
2526 if (packets > 35)
2527 retval = low_latency;
2528 } else if (bytes < 6000) {
2529 retval = low_latency;
2530 }
2531 break;
2532 }
2533
2534 return retval;
2535}
2536
2537static void e1000_set_itr(struct e1000_adapter *adapter)
2538{
2539 u16 current_itr;
2540 u32 new_itr = adapter->itr;
2541
2542 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2543 if (adapter->link_speed != SPEED_1000) {
2544 new_itr = 4000;
2545 goto set_itr_now;
2546 }
2547
2548 if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2549 new_itr = 0;
2550 goto set_itr_now;
2551 }
2552
2553 adapter->tx_itr = e1000_update_itr(adapter->tx_itr,
2554 adapter->total_tx_packets,
2555 adapter->total_tx_bytes);
2556 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2557 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2558 adapter->tx_itr = low_latency;
2559
2560 adapter->rx_itr = e1000_update_itr(adapter->rx_itr,
2561 adapter->total_rx_packets,
2562 adapter->total_rx_bytes);
2563 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2564 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2565 adapter->rx_itr = low_latency;
2566
2567 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2568
2569 /* counts and packets in update_itr are dependent on these numbers */
2570 switch (current_itr) {
2571 case lowest_latency:
2572 new_itr = 70000;
2573 break;
2574 case low_latency:
2575 new_itr = 20000; /* aka hwitr = ~200 */
2576 break;
2577 case bulk_latency:
2578 new_itr = 4000;
2579 break;
2580 default:
2581 break;
2582 }
2583
2584set_itr_now:
2585 if (new_itr != adapter->itr) {
2586 /* this attempts to bias the interrupt rate towards Bulk
2587 * by adding intermediate steps when interrupt rate is
2588 * increasing
2589 */
2590 new_itr = new_itr > adapter->itr ?
2591 min(adapter->itr + (new_itr >> 2), new_itr) : new_itr;
2592 adapter->itr = new_itr;
2593 adapter->rx_ring->itr_val = new_itr;
2594 if (adapter->msix_entries)
2595 adapter->rx_ring->set_itr = 1;
2596 else
2597 e1000e_write_itr(adapter, new_itr);
2598 }
2599}
2600
2601/**
2602 * e1000e_write_itr - write the ITR value to the appropriate registers
2603 * @adapter: address of board private structure
2604 * @itr: new ITR value to program
2605 *
2606 * e1000e_write_itr determines if the adapter is in MSI-X mode
2607 * and, if so, writes the EITR registers with the ITR value.
2608 * Otherwise, it writes the ITR value into the ITR register.
2609 **/
2610void e1000e_write_itr(struct e1000_adapter *adapter, u32 itr)
2611{
2612 struct e1000_hw *hw = &adapter->hw;
2613 u32 new_itr = itr ? 1000000000 / (itr * 256) : 0;
2614
2615 if (adapter->msix_entries) {
2616 int vector;
2617
2618 for (vector = 0; vector < adapter->num_vectors; vector++)
2619 writel(new_itr, hw->hw_addr + E1000_EITR_82574(vector));
2620 } else {
2621 ew32(ITR, new_itr);
2622 }
2623}
2624
2625/**
2626 * e1000_alloc_queues - Allocate memory for all rings
2627 * @adapter: board private structure to initialize
2628 **/
2629static int e1000_alloc_queues(struct e1000_adapter *adapter)
2630{
2631 int size = sizeof(struct e1000_ring);
2632
2633 adapter->tx_ring = kzalloc(size, GFP_KERNEL);
2634 if (!adapter->tx_ring)
2635 goto err;
2636 adapter->tx_ring->count = adapter->tx_ring_count;
2637 adapter->tx_ring->adapter = adapter;
2638
2639 adapter->rx_ring = kzalloc(size, GFP_KERNEL);
2640 if (!adapter->rx_ring)
2641 goto err;
2642 adapter->rx_ring->count = adapter->rx_ring_count;
2643 adapter->rx_ring->adapter = adapter;
2644
2645 return 0;
2646err:
2647 e_err("Unable to allocate memory for queues\n");
2648 kfree(adapter->rx_ring);
2649 kfree(adapter->tx_ring);
2650 return -ENOMEM;
2651}
2652
2653/**
2654 * e1000e_poll - NAPI Rx polling callback
2655 * @napi: struct associated with this polling callback
2656 * @budget: number of packets driver is allowed to process this poll
2657 **/
2658static int e1000e_poll(struct napi_struct *napi, int budget)
2659{
2660 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter,
2661 napi);
2662 struct e1000_hw *hw = &adapter->hw;
2663 struct net_device *poll_dev = adapter->netdev;
2664 int tx_cleaned = 1, work_done = 0;
2665
2666 adapter = netdev_priv(poll_dev);
2667
2668 if (!adapter->msix_entries ||
2669 (adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2670 tx_cleaned = e1000_clean_tx_irq(adapter->tx_ring);
2671
2672 adapter->clean_rx(adapter->rx_ring, &work_done, budget);
2673
2674 if (!tx_cleaned || work_done == budget)
2675 return budget;
2676
2677 /* Exit the polling mode, but don't re-enable interrupts if stack might
2678 * poll us due to busy-polling
2679 */
2680 if (likely(napi_complete_done(napi, work_done))) {
2681 if (adapter->itr_setting & 3)
2682 e1000_set_itr(adapter);
2683 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2684 if (adapter->msix_entries)
2685 ew32(IMS, adapter->rx_ring->ims_val);
2686 else
2687 e1000_irq_enable(adapter);
2688 }
2689 }
2690
2691 return work_done;
2692}
2693
2694static int e1000_vlan_rx_add_vid(struct net_device *netdev,
2695 __always_unused __be16 proto, u16 vid)
2696{
2697 struct e1000_adapter *adapter = netdev_priv(netdev);
2698 struct e1000_hw *hw = &adapter->hw;
2699 u32 vfta, index;
2700
2701 /* don't update vlan cookie if already programmed */
2702 if ((adapter->hw.mng_cookie.status &
2703 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2704 (vid == adapter->mng_vlan_id))
2705 return 0;
2706
2707 /* add VID to filter table */
2708 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2709 index = (vid >> 5) & 0x7F;
2710 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2711 vfta |= BIT((vid & 0x1F));
2712 hw->mac.ops.write_vfta(hw, index, vfta);
2713 }
2714
2715 set_bit(vid, adapter->active_vlans);
2716
2717 return 0;
2718}
2719
2720static int e1000_vlan_rx_kill_vid(struct net_device *netdev,
2721 __always_unused __be16 proto, u16 vid)
2722{
2723 struct e1000_adapter *adapter = netdev_priv(netdev);
2724 struct e1000_hw *hw = &adapter->hw;
2725 u32 vfta, index;
2726
2727 if ((adapter->hw.mng_cookie.status &
2728 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2729 (vid == adapter->mng_vlan_id)) {
2730 /* release control to f/w */
2731 e1000e_release_hw_control(adapter);
2732 return 0;
2733 }
2734
2735 /* remove VID from filter table */
2736 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2737 index = (vid >> 5) & 0x7F;
2738 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2739 vfta &= ~BIT((vid & 0x1F));
2740 hw->mac.ops.write_vfta(hw, index, vfta);
2741 }
2742
2743 clear_bit(vid, adapter->active_vlans);
2744
2745 return 0;
2746}
2747
2748/**
2749 * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2750 * @adapter: board private structure to initialize
2751 **/
2752static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2753{
2754 struct net_device *netdev = adapter->netdev;
2755 struct e1000_hw *hw = &adapter->hw;
2756 u32 rctl;
2757
2758 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2759 /* disable VLAN receive filtering */
2760 rctl = er32(RCTL);
2761 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2762 ew32(RCTL, rctl);
2763
2764 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2765 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
2766 adapter->mng_vlan_id);
2767 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2768 }
2769 }
2770}
2771
2772/**
2773 * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2774 * @adapter: board private structure to initialize
2775 **/
2776static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2777{
2778 struct e1000_hw *hw = &adapter->hw;
2779 u32 rctl;
2780
2781 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2782 /* enable VLAN receive filtering */
2783 rctl = er32(RCTL);
2784 rctl |= E1000_RCTL_VFE;
2785 rctl &= ~E1000_RCTL_CFIEN;
2786 ew32(RCTL, rctl);
2787 }
2788}
2789
2790/**
2791 * e1000e_vlan_strip_disable - helper to disable HW VLAN stripping
2792 * @adapter: board private structure to initialize
2793 **/
2794static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2795{
2796 struct e1000_hw *hw = &adapter->hw;
2797 u32 ctrl;
2798
2799 /* disable VLAN tag insert/strip */
2800 ctrl = er32(CTRL);
2801 ctrl &= ~E1000_CTRL_VME;
2802 ew32(CTRL, ctrl);
2803}
2804
2805/**
2806 * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2807 * @adapter: board private structure to initialize
2808 **/
2809static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2810{
2811 struct e1000_hw *hw = &adapter->hw;
2812 u32 ctrl;
2813
2814 /* enable VLAN tag insert/strip */
2815 ctrl = er32(CTRL);
2816 ctrl |= E1000_CTRL_VME;
2817 ew32(CTRL, ctrl);
2818}
2819
2820static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2821{
2822 struct net_device *netdev = adapter->netdev;
2823 u16 vid = adapter->hw.mng_cookie.vlan_id;
2824 u16 old_vid = adapter->mng_vlan_id;
2825
2826 if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2827 e1000_vlan_rx_add_vid(netdev, htons(ETH_P_8021Q), vid);
2828 adapter->mng_vlan_id = vid;
2829 }
2830
2831 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2832 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q), old_vid);
2833}
2834
2835static void e1000_restore_vlan(struct e1000_adapter *adapter)
2836{
2837 u16 vid;
2838
2839 e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
2840
2841 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2842 e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
2843}
2844
2845static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2846{
2847 struct e1000_hw *hw = &adapter->hw;
2848 u32 manc, manc2h, mdef, i, j;
2849
2850 if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2851 return;
2852
2853 manc = er32(MANC);
2854
2855 /* enable receiving management packets to the host. this will probably
2856 * generate destination unreachable messages from the host OS, but
2857 * the packets will be handled on SMBUS
2858 */
2859 manc |= E1000_MANC_EN_MNG2HOST;
2860 manc2h = er32(MANC2H);
2861
2862 switch (hw->mac.type) {
2863 default:
2864 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2865 break;
2866 case e1000_82574:
2867 case e1000_82583:
2868 /* Check if IPMI pass-through decision filter already exists;
2869 * if so, enable it.
2870 */
2871 for (i = 0, j = 0; i < 8; i++) {
2872 mdef = er32(MDEF(i));
2873
2874 /* Ignore filters with anything other than IPMI ports */
2875 if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2876 continue;
2877
2878 /* Enable this decision filter in MANC2H */
2879 if (mdef)
2880 manc2h |= BIT(i);
2881
2882 j |= mdef;
2883 }
2884
2885 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2886 break;
2887
2888 /* Create new decision filter in an empty filter */
2889 for (i = 0, j = 0; i < 8; i++)
2890 if (er32(MDEF(i)) == 0) {
2891 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2892 E1000_MDEF_PORT_664));
2893 manc2h |= BIT(1);
2894 j++;
2895 break;
2896 }
2897
2898 if (!j)
2899 e_warn("Unable to create IPMI pass-through filter\n");
2900 break;
2901 }
2902
2903 ew32(MANC2H, manc2h);
2904 ew32(MANC, manc);
2905}
2906
2907/**
2908 * e1000_configure_tx - Configure Transmit Unit after Reset
2909 * @adapter: board private structure
2910 *
2911 * Configure the Tx unit of the MAC after a reset.
2912 **/
2913static void e1000_configure_tx(struct e1000_adapter *adapter)
2914{
2915 struct e1000_hw *hw = &adapter->hw;
2916 struct e1000_ring *tx_ring = adapter->tx_ring;
2917 u64 tdba;
2918 u32 tdlen, tctl, tarc;
2919
2920 /* Setup the HW Tx Head and Tail descriptor pointers */
2921 tdba = tx_ring->dma;
2922 tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2923 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
2924 ew32(TDBAH(0), (tdba >> 32));
2925 ew32(TDLEN(0), tdlen);
2926 ew32(TDH(0), 0);
2927 ew32(TDT(0), 0);
2928 tx_ring->head = adapter->hw.hw_addr + E1000_TDH(0);
2929 tx_ring->tail = adapter->hw.hw_addr + E1000_TDT(0);
2930
2931 writel(0, tx_ring->head);
2932 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
2933 e1000e_update_tdt_wa(tx_ring, 0);
2934 else
2935 writel(0, tx_ring->tail);
2936
2937 /* Set the Tx Interrupt Delay register */
2938 ew32(TIDV, adapter->tx_int_delay);
2939 /* Tx irq moderation */
2940 ew32(TADV, adapter->tx_abs_int_delay);
2941
2942 if (adapter->flags2 & FLAG2_DMA_BURST) {
2943 u32 txdctl = er32(TXDCTL(0));
2944
2945 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2946 E1000_TXDCTL_WTHRESH);
2947 /* set up some performance related parameters to encourage the
2948 * hardware to use the bus more efficiently in bursts, depends
2949 * on the tx_int_delay to be enabled,
2950 * wthresh = 1 ==> burst write is disabled to avoid Tx stalls
2951 * hthresh = 1 ==> prefetch when one or more available
2952 * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2953 * BEWARE: this seems to work but should be considered first if
2954 * there are Tx hangs or other Tx related bugs
2955 */
2956 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2957 ew32(TXDCTL(0), txdctl);
2958 }
2959 /* erratum work around: set txdctl the same for both queues */
2960 ew32(TXDCTL(1), er32(TXDCTL(0)));
2961
2962 /* Program the Transmit Control Register */
2963 tctl = er32(TCTL);
2964 tctl &= ~E1000_TCTL_CT;
2965 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2966 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2967
2968 if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2969 tarc = er32(TARC(0));
2970 /* set the speed mode bit, we'll clear it if we're not at
2971 * gigabit link later
2972 */
2973#define SPEED_MODE_BIT BIT(21)
2974 tarc |= SPEED_MODE_BIT;
2975 ew32(TARC(0), tarc);
2976 }
2977
2978 /* errata: program both queues to unweighted RR */
2979 if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2980 tarc = er32(TARC(0));
2981 tarc |= 1;
2982 ew32(TARC(0), tarc);
2983 tarc = er32(TARC(1));
2984 tarc |= 1;
2985 ew32(TARC(1), tarc);
2986 }
2987
2988 /* Setup Transmit Descriptor Settings for eop descriptor */
2989 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2990
2991 /* only set IDE if we are delaying interrupts using the timers */
2992 if (adapter->tx_int_delay)
2993 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2994
2995 /* enable Report Status bit */
2996 adapter->txd_cmd |= E1000_TXD_CMD_RS;
2997
2998 ew32(TCTL, tctl);
2999
3000 hw->mac.ops.config_collision_dist(hw);
3001
3002 /* SPT and KBL Si errata workaround to avoid data corruption */
3003 if (hw->mac.type == e1000_pch_spt) {
3004 u32 reg_val;
3005
3006 reg_val = er32(IOSFPC);
3007 reg_val |= E1000_RCTL_RDMTS_HEX;
3008 ew32(IOSFPC, reg_val);
3009
3010 reg_val = er32(TARC(0));
3011 /* SPT and KBL Si errata workaround to avoid Tx hang.
3012 * Dropping the number of outstanding requests from
3013 * 3 to 2 in order to avoid a buffer overrun.
3014 */
3015 reg_val &= ~E1000_TARC0_CB_MULTIQ_3_REQ;
3016 reg_val |= E1000_TARC0_CB_MULTIQ_2_REQ;
3017 ew32(TARC(0), reg_val);
3018 }
3019}
3020
3021#define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
3022 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
3023
3024/**
3025 * e1000_setup_rctl - configure the receive control registers
3026 * @adapter: Board private structure
3027 **/
3028static void e1000_setup_rctl(struct e1000_adapter *adapter)
3029{
3030 struct e1000_hw *hw = &adapter->hw;
3031 u32 rctl, rfctl;
3032 u32 pages = 0;
3033
3034 /* Workaround Si errata on PCHx - configure jumbo frame flow.
3035 * If jumbo frames not set, program related MAC/PHY registers
3036 * to h/w defaults
3037 */
3038 if (hw->mac.type >= e1000_pch2lan) {
3039 s32 ret_val;
3040
3041 if (adapter->netdev->mtu > ETH_DATA_LEN)
3042 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
3043 else
3044 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
3045
3046 if (ret_val)
3047 e_dbg("failed to enable|disable jumbo frame workaround mode\n");
3048 }
3049
3050 /* Program MC offset vector base */
3051 rctl = er32(RCTL);
3052 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3053 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
3054 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
3055 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3056
3057 /* Do not Store bad packets */
3058 rctl &= ~E1000_RCTL_SBP;
3059
3060 /* Enable Long Packet receive */
3061 if (adapter->netdev->mtu <= ETH_DATA_LEN)
3062 rctl &= ~E1000_RCTL_LPE;
3063 else
3064 rctl |= E1000_RCTL_LPE;
3065
3066 /* Some systems expect that the CRC is included in SMBUS traffic. The
3067 * hardware strips the CRC before sending to both SMBUS (BMC) and to
3068 * host memory when this is enabled
3069 */
3070 if (adapter->flags2 & FLAG2_CRC_STRIPPING)
3071 rctl |= E1000_RCTL_SECRC;
3072
3073 /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
3074 if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
3075 u16 phy_data;
3076
3077 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
3078 phy_data &= 0xfff8;
3079 phy_data |= BIT(2);
3080 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
3081
3082 e1e_rphy(hw, 22, &phy_data);
3083 phy_data &= 0x0fff;
3084 phy_data |= BIT(14);
3085 e1e_wphy(hw, 0x10, 0x2823);
3086 e1e_wphy(hw, 0x11, 0x0003);
3087 e1e_wphy(hw, 22, phy_data);
3088 }
3089
3090 /* Setup buffer sizes */
3091 rctl &= ~E1000_RCTL_SZ_4096;
3092 rctl |= E1000_RCTL_BSEX;
3093 switch (adapter->rx_buffer_len) {
3094 case 2048:
3095 default:
3096 rctl |= E1000_RCTL_SZ_2048;
3097 rctl &= ~E1000_RCTL_BSEX;
3098 break;
3099 case 4096:
3100 rctl |= E1000_RCTL_SZ_4096;
3101 break;
3102 case 8192:
3103 rctl |= E1000_RCTL_SZ_8192;
3104 break;
3105 case 16384:
3106 rctl |= E1000_RCTL_SZ_16384;
3107 break;
3108 }
3109
3110 /* Enable Extended Status in all Receive Descriptors */
3111 rfctl = er32(RFCTL);
3112 rfctl |= E1000_RFCTL_EXTEN;
3113 ew32(RFCTL, rfctl);
3114
3115 /* 82571 and greater support packet-split where the protocol
3116 * header is placed in skb->data and the packet data is
3117 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
3118 * In the case of a non-split, skb->data is linearly filled,
3119 * followed by the page buffers. Therefore, skb->data is
3120 * sized to hold the largest protocol header.
3121 *
3122 * allocations using alloc_page take too long for regular MTU
3123 * so only enable packet split for jumbo frames
3124 *
3125 * Using pages when the page size is greater than 16k wastes
3126 * a lot of memory, since we allocate 3 pages at all times
3127 * per packet.
3128 */
3129 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
3130 if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
3131 adapter->rx_ps_pages = pages;
3132 else
3133 adapter->rx_ps_pages = 0;
3134
3135 if (adapter->rx_ps_pages) {
3136 u32 psrctl = 0;
3137
3138 /* Enable Packet split descriptors */
3139 rctl |= E1000_RCTL_DTYP_PS;
3140
3141 psrctl |= adapter->rx_ps_bsize0 >> E1000_PSRCTL_BSIZE0_SHIFT;
3142
3143 switch (adapter->rx_ps_pages) {
3144 case 3:
3145 psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE3_SHIFT;
3146 fallthrough;
3147 case 2:
3148 psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE2_SHIFT;
3149 fallthrough;
3150 case 1:
3151 psrctl |= PAGE_SIZE >> E1000_PSRCTL_BSIZE1_SHIFT;
3152 break;
3153 }
3154
3155 ew32(PSRCTL, psrctl);
3156 }
3157
3158 /* This is useful for sniffing bad packets. */
3159 if (adapter->netdev->features & NETIF_F_RXALL) {
3160 /* UPE and MPE will be handled by normal PROMISC logic
3161 * in e1000e_set_rx_mode
3162 */
3163 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3164 E1000_RCTL_BAM | /* RX All Bcast Pkts */
3165 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3166
3167 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3168 E1000_RCTL_DPF | /* Allow filtered pause */
3169 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3170 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3171 * and that breaks VLANs.
3172 */
3173 }
3174
3175 ew32(RCTL, rctl);
3176 /* just started the receive unit, no need to restart */
3177 adapter->flags &= ~FLAG_RESTART_NOW;
3178}
3179
3180/**
3181 * e1000_configure_rx - Configure Receive Unit after Reset
3182 * @adapter: board private structure
3183 *
3184 * Configure the Rx unit of the MAC after a reset.
3185 **/
3186static void e1000_configure_rx(struct e1000_adapter *adapter)
3187{
3188 struct e1000_hw *hw = &adapter->hw;
3189 struct e1000_ring *rx_ring = adapter->rx_ring;
3190 u64 rdba;
3191 u32 rdlen, rctl, rxcsum, ctrl_ext;
3192
3193 if (adapter->rx_ps_pages) {
3194 /* this is a 32 byte descriptor */
3195 rdlen = rx_ring->count *
3196 sizeof(union e1000_rx_desc_packet_split);
3197 adapter->clean_rx = e1000_clean_rx_irq_ps;
3198 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3199 } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3200 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3201 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3202 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3203 } else {
3204 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3205 adapter->clean_rx = e1000_clean_rx_irq;
3206 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3207 }
3208
3209 /* disable receives while setting up the descriptors */
3210 rctl = er32(RCTL);
3211 if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3212 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3213 e1e_flush();
3214 usleep_range(10000, 11000);
3215
3216 if (adapter->flags2 & FLAG2_DMA_BURST) {
3217 /* set the writeback threshold (only takes effect if the RDTR
3218 * is set). set GRAN=1 and write back up to 0x4 worth, and
3219 * enable prefetching of 0x20 Rx descriptors
3220 * granularity = 01
3221 * wthresh = 04,
3222 * hthresh = 04,
3223 * pthresh = 0x20
3224 */
3225 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3226 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3227 }
3228
3229 /* set the Receive Delay Timer Register */
3230 ew32(RDTR, adapter->rx_int_delay);
3231
3232 /* irq moderation */
3233 ew32(RADV, adapter->rx_abs_int_delay);
3234 if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3235 e1000e_write_itr(adapter, adapter->itr);
3236
3237 ctrl_ext = er32(CTRL_EXT);
3238 /* Auto-Mask interrupts upon ICR access */
3239 ctrl_ext |= E1000_CTRL_EXT_IAME;
3240 ew32(IAM, 0xffffffff);
3241 ew32(CTRL_EXT, ctrl_ext);
3242 e1e_flush();
3243
3244 /* Setup the HW Rx Head and Tail Descriptor Pointers and
3245 * the Base and Length of the Rx Descriptor Ring
3246 */
3247 rdba = rx_ring->dma;
3248 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
3249 ew32(RDBAH(0), (rdba >> 32));
3250 ew32(RDLEN(0), rdlen);
3251 ew32(RDH(0), 0);
3252 ew32(RDT(0), 0);
3253 rx_ring->head = adapter->hw.hw_addr + E1000_RDH(0);
3254 rx_ring->tail = adapter->hw.hw_addr + E1000_RDT(0);
3255
3256 writel(0, rx_ring->head);
3257 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
3258 e1000e_update_rdt_wa(rx_ring, 0);
3259 else
3260 writel(0, rx_ring->tail);
3261
3262 /* Enable Receive Checksum Offload for TCP and UDP */
3263 rxcsum = er32(RXCSUM);
3264 if (adapter->netdev->features & NETIF_F_RXCSUM)
3265 rxcsum |= E1000_RXCSUM_TUOFL;
3266 else
3267 rxcsum &= ~E1000_RXCSUM_TUOFL;
3268 ew32(RXCSUM, rxcsum);
3269
3270 /* With jumbo frames, excessive C-state transition latencies result
3271 * in dropped transactions.
3272 */
3273 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3274 u32 lat =
3275 ((er32(PBA) & E1000_PBA_RXA_MASK) * 1024 -
3276 adapter->max_frame_size) * 8 / 1000;
3277
3278 if (adapter->flags & FLAG_IS_ICH) {
3279 u32 rxdctl = er32(RXDCTL(0));
3280
3281 ew32(RXDCTL(0), rxdctl | 0x3 | BIT(8));
3282 }
3283
3284 dev_info(&adapter->pdev->dev,
3285 "Some CPU C-states have been disabled in order to enable jumbo frames\n");
3286 cpu_latency_qos_update_request(&adapter->pm_qos_req, lat);
3287 } else {
3288 cpu_latency_qos_update_request(&adapter->pm_qos_req,
3289 PM_QOS_DEFAULT_VALUE);
3290 }
3291
3292 /* Enable Receives */
3293 ew32(RCTL, rctl);
3294}
3295
3296/**
3297 * e1000e_write_mc_addr_list - write multicast addresses to MTA
3298 * @netdev: network interface device structure
3299 *
3300 * Writes multicast address list to the MTA hash table.
3301 * Returns: -ENOMEM on failure
3302 * 0 on no addresses written
3303 * X on writing X addresses to MTA
3304 */
3305static int e1000e_write_mc_addr_list(struct net_device *netdev)
3306{
3307 struct e1000_adapter *adapter = netdev_priv(netdev);
3308 struct e1000_hw *hw = &adapter->hw;
3309 struct netdev_hw_addr *ha;
3310 u8 *mta_list;
3311 int i;
3312
3313 if (netdev_mc_empty(netdev)) {
3314 /* nothing to program, so clear mc list */
3315 hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3316 return 0;
3317 }
3318
3319 mta_list = kcalloc(netdev_mc_count(netdev), ETH_ALEN, GFP_ATOMIC);
3320 if (!mta_list)
3321 return -ENOMEM;
3322
3323 /* update_mc_addr_list expects a packed array of only addresses. */
3324 i = 0;
3325 netdev_for_each_mc_addr(ha, netdev)
3326 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3327
3328 hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3329 kfree(mta_list);
3330
3331 return netdev_mc_count(netdev);
3332}
3333
3334/**
3335 * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3336 * @netdev: network interface device structure
3337 *
3338 * Writes unicast address list to the RAR table.
3339 * Returns: -ENOMEM on failure/insufficient address space
3340 * 0 on no addresses written
3341 * X on writing X addresses to the RAR table
3342 **/
3343static int e1000e_write_uc_addr_list(struct net_device *netdev)
3344{
3345 struct e1000_adapter *adapter = netdev_priv(netdev);
3346 struct e1000_hw *hw = &adapter->hw;
3347 unsigned int rar_entries;
3348 int count = 0;
3349
3350 rar_entries = hw->mac.ops.rar_get_count(hw);
3351
3352 /* save a rar entry for our hardware address */
3353 rar_entries--;
3354
3355 /* save a rar entry for the LAA workaround */
3356 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3357 rar_entries--;
3358
3359 /* return ENOMEM indicating insufficient memory for addresses */
3360 if (netdev_uc_count(netdev) > rar_entries)
3361 return -ENOMEM;
3362
3363 if (!netdev_uc_empty(netdev) && rar_entries) {
3364 struct netdev_hw_addr *ha;
3365
3366 /* write the addresses in reverse order to avoid write
3367 * combining
3368 */
3369 netdev_for_each_uc_addr(ha, netdev) {
3370 int ret_val;
3371
3372 if (!rar_entries)
3373 break;
3374 ret_val = hw->mac.ops.rar_set(hw, ha->addr, rar_entries--);
3375 if (ret_val < 0)
3376 return -ENOMEM;
3377 count++;
3378 }
3379 }
3380
3381 /* zero out the remaining RAR entries not used above */
3382 for (; rar_entries > 0; rar_entries--) {
3383 ew32(RAH(rar_entries), 0);
3384 ew32(RAL(rar_entries), 0);
3385 }
3386 e1e_flush();
3387
3388 return count;
3389}
3390
3391/**
3392 * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3393 * @netdev: network interface device structure
3394 *
3395 * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3396 * address list or the network interface flags are updated. This routine is
3397 * responsible for configuring the hardware for proper unicast, multicast,
3398 * promiscuous mode, and all-multi behavior.
3399 **/
3400static void e1000e_set_rx_mode(struct net_device *netdev)
3401{
3402 struct e1000_adapter *adapter = netdev_priv(netdev);
3403 struct e1000_hw *hw = &adapter->hw;
3404 u32 rctl;
3405
3406 if (pm_runtime_suspended(netdev->dev.parent))
3407 return;
3408
3409 /* Check for Promiscuous and All Multicast modes */
3410 rctl = er32(RCTL);
3411
3412 /* clear the affected bits */
3413 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3414
3415 if (netdev->flags & IFF_PROMISC) {
3416 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3417 /* Do not hardware filter VLANs in promisc mode */
3418 e1000e_vlan_filter_disable(adapter);
3419 } else {
3420 int count;
3421
3422 if (netdev->flags & IFF_ALLMULTI) {
3423 rctl |= E1000_RCTL_MPE;
3424 } else {
3425 /* Write addresses to the MTA, if the attempt fails
3426 * then we should just turn on promiscuous mode so
3427 * that we can at least receive multicast traffic
3428 */
3429 count = e1000e_write_mc_addr_list(netdev);
3430 if (count < 0)
3431 rctl |= E1000_RCTL_MPE;
3432 }
3433 e1000e_vlan_filter_enable(adapter);
3434 /* Write addresses to available RAR registers, if there is not
3435 * sufficient space to store all the addresses then enable
3436 * unicast promiscuous mode
3437 */
3438 count = e1000e_write_uc_addr_list(netdev);
3439 if (count < 0)
3440 rctl |= E1000_RCTL_UPE;
3441 }
3442
3443 ew32(RCTL, rctl);
3444
3445 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
3446 e1000e_vlan_strip_enable(adapter);
3447 else
3448 e1000e_vlan_strip_disable(adapter);
3449}
3450
3451static void e1000e_setup_rss_hash(struct e1000_adapter *adapter)
3452{
3453 struct e1000_hw *hw = &adapter->hw;
3454 u32 mrqc, rxcsum;
3455 u32 rss_key[10];
3456 int i;
3457
3458 netdev_rss_key_fill(rss_key, sizeof(rss_key));
3459 for (i = 0; i < 10; i++)
3460 ew32(RSSRK(i), rss_key[i]);
3461
3462 /* Direct all traffic to queue 0 */
3463 for (i = 0; i < 32; i++)
3464 ew32(RETA(i), 0);
3465
3466 /* Disable raw packet checksumming so that RSS hash is placed in
3467 * descriptor on writeback.
3468 */
3469 rxcsum = er32(RXCSUM);
3470 rxcsum |= E1000_RXCSUM_PCSD;
3471
3472 ew32(RXCSUM, rxcsum);
3473
3474 mrqc = (E1000_MRQC_RSS_FIELD_IPV4 |
3475 E1000_MRQC_RSS_FIELD_IPV4_TCP |
3476 E1000_MRQC_RSS_FIELD_IPV6 |
3477 E1000_MRQC_RSS_FIELD_IPV6_TCP |
3478 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
3479
3480 ew32(MRQC, mrqc);
3481}
3482
3483/**
3484 * e1000e_get_base_timinca - get default SYSTIM time increment attributes
3485 * @adapter: board private structure
3486 * @timinca: pointer to returned time increment attributes
3487 *
3488 * Get attributes for incrementing the System Time Register SYSTIML/H at
3489 * the default base frequency, and set the cyclecounter shift value.
3490 **/
3491s32 e1000e_get_base_timinca(struct e1000_adapter *adapter, u32 *timinca)
3492{
3493 struct e1000_hw *hw = &adapter->hw;
3494 u32 incvalue, incperiod, shift;
3495
3496 /* Make sure clock is enabled on I217/I218/I219 before checking
3497 * the frequency
3498 */
3499 if ((hw->mac.type >= e1000_pch_lpt) &&
3500 !(er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) &&
3501 !(er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_ENABLED)) {
3502 u32 fextnvm7 = er32(FEXTNVM7);
3503
3504 if (!(fextnvm7 & BIT(0))) {
3505 ew32(FEXTNVM7, fextnvm7 | BIT(0));
3506 e1e_flush();
3507 }
3508 }
3509
3510 switch (hw->mac.type) {
3511 case e1000_pch2lan:
3512 /* Stable 96MHz frequency */
3513 incperiod = INCPERIOD_96MHZ;
3514 incvalue = INCVALUE_96MHZ;
3515 shift = INCVALUE_SHIFT_96MHZ;
3516 adapter->cc.shift = shift + INCPERIOD_SHIFT_96MHZ;
3517 break;
3518 case e1000_pch_lpt:
3519 if (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI) {
3520 /* Stable 96MHz frequency */
3521 incperiod = INCPERIOD_96MHZ;
3522 incvalue = INCVALUE_96MHZ;
3523 shift = INCVALUE_SHIFT_96MHZ;
3524 adapter->cc.shift = shift + INCPERIOD_SHIFT_96MHZ;
3525 } else {
3526 /* Stable 25MHz frequency */
3527 incperiod = INCPERIOD_25MHZ;
3528 incvalue = INCVALUE_25MHZ;
3529 shift = INCVALUE_SHIFT_25MHZ;
3530 adapter->cc.shift = shift;
3531 }
3532 break;
3533 case e1000_pch_spt:
3534 /* Stable 24MHz frequency */
3535 incperiod = INCPERIOD_24MHZ;
3536 incvalue = INCVALUE_24MHZ;
3537 shift = INCVALUE_SHIFT_24MHZ;
3538 adapter->cc.shift = shift;
3539 break;
3540 case e1000_pch_cnp:
3541 case e1000_pch_tgp:
3542 case e1000_pch_adp:
3543 case e1000_pch_mtp:
3544 case e1000_pch_lnp:
3545 case e1000_pch_ptp:
3546 case e1000_pch_nvp:
3547 if (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI) {
3548 /* Stable 24MHz frequency */
3549 incperiod = INCPERIOD_24MHZ;
3550 incvalue = INCVALUE_24MHZ;
3551 shift = INCVALUE_SHIFT_24MHZ;
3552 adapter->cc.shift = shift;
3553 } else {
3554 /* Stable 38400KHz frequency */
3555 incperiod = INCPERIOD_38400KHZ;
3556 incvalue = INCVALUE_38400KHZ;
3557 shift = INCVALUE_SHIFT_38400KHZ;
3558 adapter->cc.shift = shift;
3559 }
3560 break;
3561 case e1000_82574:
3562 case e1000_82583:
3563 /* Stable 25MHz frequency */
3564 incperiod = INCPERIOD_25MHZ;
3565 incvalue = INCVALUE_25MHZ;
3566 shift = INCVALUE_SHIFT_25MHZ;
3567 adapter->cc.shift = shift;
3568 break;
3569 default:
3570 return -EINVAL;
3571 }
3572
3573 *timinca = ((incperiod << E1000_TIMINCA_INCPERIOD_SHIFT) |
3574 ((incvalue << shift) & E1000_TIMINCA_INCVALUE_MASK));
3575
3576 return 0;
3577}
3578
3579/**
3580 * e1000e_config_hwtstamp - configure the hwtstamp registers and enable/disable
3581 * @adapter: board private structure
3582 * @config: timestamp configuration
3583 *
3584 * Outgoing time stamping can be enabled and disabled. Play nice and
3585 * disable it when requested, although it shouldn't cause any overhead
3586 * when no packet needs it. At most one packet in the queue may be
3587 * marked for time stamping, otherwise it would be impossible to tell
3588 * for sure to which packet the hardware time stamp belongs.
3589 *
3590 * Incoming time stamping has to be configured via the hardware filters.
3591 * Not all combinations are supported, in particular event type has to be
3592 * specified. Matching the kind of event packet is not supported, with the
3593 * exception of "all V2 events regardless of level 2 or 4".
3594 **/
3595static int e1000e_config_hwtstamp(struct e1000_adapter *adapter,
3596 struct hwtstamp_config *config)
3597{
3598 struct e1000_hw *hw = &adapter->hw;
3599 u32 tsync_tx_ctl = E1000_TSYNCTXCTL_ENABLED;
3600 u32 tsync_rx_ctl = E1000_TSYNCRXCTL_ENABLED;
3601 u32 rxmtrl = 0;
3602 u16 rxudp = 0;
3603 bool is_l4 = false;
3604 bool is_l2 = false;
3605 u32 regval;
3606
3607 if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP))
3608 return -EINVAL;
3609
3610 switch (config->tx_type) {
3611 case HWTSTAMP_TX_OFF:
3612 tsync_tx_ctl = 0;
3613 break;
3614 case HWTSTAMP_TX_ON:
3615 break;
3616 default:
3617 return -ERANGE;
3618 }
3619
3620 switch (config->rx_filter) {
3621 case HWTSTAMP_FILTER_NONE:
3622 tsync_rx_ctl = 0;
3623 break;
3624 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3625 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3626 rxmtrl = E1000_RXMTRL_PTP_V1_SYNC_MESSAGE;
3627 is_l4 = true;
3628 break;
3629 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3630 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3631 rxmtrl = E1000_RXMTRL_PTP_V1_DELAY_REQ_MESSAGE;
3632 is_l4 = true;
3633 break;
3634 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
3635 /* Also time stamps V2 L2 Path Delay Request/Response */
3636 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3637 rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3638 is_l2 = true;
3639 break;
3640 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
3641 /* Also time stamps V2 L2 Path Delay Request/Response. */
3642 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3643 rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3644 is_l2 = true;
3645 break;
3646 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3647 /* Hardware cannot filter just V2 L4 Sync messages */
3648 fallthrough;
3649 case HWTSTAMP_FILTER_PTP_V2_SYNC:
3650 /* Also time stamps V2 Path Delay Request/Response. */
3651 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3652 rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3653 is_l2 = true;
3654 is_l4 = true;
3655 break;
3656 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3657 /* Hardware cannot filter just V2 L4 Delay Request messages */
3658 fallthrough;
3659 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
3660 /* Also time stamps V2 Path Delay Request/Response. */
3661 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3662 rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3663 is_l2 = true;
3664 is_l4 = true;
3665 break;
3666 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3667 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
3668 /* Hardware cannot filter just V2 L4 or L2 Event messages */
3669 fallthrough;
3670 case HWTSTAMP_FILTER_PTP_V2_EVENT:
3671 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_EVENT_V2;
3672 config->rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
3673 is_l2 = true;
3674 is_l4 = true;
3675 break;
3676 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3677 /* For V1, the hardware can only filter Sync messages or
3678 * Delay Request messages but not both so fall-through to
3679 * time stamp all packets.
3680 */
3681 fallthrough;
3682 case HWTSTAMP_FILTER_NTP_ALL:
3683 case HWTSTAMP_FILTER_ALL:
3684 is_l2 = true;
3685 is_l4 = true;
3686 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_ALL;
3687 config->rx_filter = HWTSTAMP_FILTER_ALL;
3688 break;
3689 default:
3690 return -ERANGE;
3691 }
3692
3693 adapter->hwtstamp_config = *config;
3694
3695 /* enable/disable Tx h/w time stamping */
3696 regval = er32(TSYNCTXCTL);
3697 regval &= ~E1000_TSYNCTXCTL_ENABLED;
3698 regval |= tsync_tx_ctl;
3699 ew32(TSYNCTXCTL, regval);
3700 if ((er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) !=
3701 (regval & E1000_TSYNCTXCTL_ENABLED)) {
3702 e_err("Timesync Tx Control register not set as expected\n");
3703 return -EAGAIN;
3704 }
3705
3706 /* enable/disable Rx h/w time stamping */
3707 regval = er32(TSYNCRXCTL);
3708 regval &= ~(E1000_TSYNCRXCTL_ENABLED | E1000_TSYNCRXCTL_TYPE_MASK);
3709 regval |= tsync_rx_ctl;
3710 ew32(TSYNCRXCTL, regval);
3711 if ((er32(TSYNCRXCTL) & (E1000_TSYNCRXCTL_ENABLED |
3712 E1000_TSYNCRXCTL_TYPE_MASK)) !=
3713 (regval & (E1000_TSYNCRXCTL_ENABLED |
3714 E1000_TSYNCRXCTL_TYPE_MASK))) {
3715 e_err("Timesync Rx Control register not set as expected\n");
3716 return -EAGAIN;
3717 }
3718
3719 /* L2: define ethertype filter for time stamped packets */
3720 if (is_l2)
3721 rxmtrl |= ETH_P_1588;
3722
3723 /* define which PTP packets get time stamped */
3724 ew32(RXMTRL, rxmtrl);
3725
3726 /* Filter by destination port */
3727 if (is_l4) {
3728 rxudp = PTP_EV_PORT;
3729 cpu_to_be16s(&rxudp);
3730 }
3731 ew32(RXUDP, rxudp);
3732
3733 e1e_flush();
3734
3735 /* Clear TSYNCRXCTL_VALID & TSYNCTXCTL_VALID bit */
3736 er32(RXSTMPH);
3737 er32(TXSTMPH);
3738
3739 return 0;
3740}
3741
3742/**
3743 * e1000_configure - configure the hardware for Rx and Tx
3744 * @adapter: private board structure
3745 **/
3746static void e1000_configure(struct e1000_adapter *adapter)
3747{
3748 struct e1000_ring *rx_ring = adapter->rx_ring;
3749
3750 e1000e_set_rx_mode(adapter->netdev);
3751
3752 e1000_restore_vlan(adapter);
3753 e1000_init_manageability_pt(adapter);
3754
3755 e1000_configure_tx(adapter);
3756
3757 if (adapter->netdev->features & NETIF_F_RXHASH)
3758 e1000e_setup_rss_hash(adapter);
3759 e1000_setup_rctl(adapter);
3760 e1000_configure_rx(adapter);
3761 adapter->alloc_rx_buf(rx_ring, e1000_desc_unused(rx_ring), GFP_KERNEL);
3762}
3763
3764/**
3765 * e1000e_power_up_phy - restore link in case the phy was powered down
3766 * @adapter: address of board private structure
3767 *
3768 * The phy may be powered down to save power and turn off link when the
3769 * driver is unloaded and wake on lan is not enabled (among others)
3770 * *** this routine MUST be followed by a call to e1000e_reset ***
3771 **/
3772void e1000e_power_up_phy(struct e1000_adapter *adapter)
3773{
3774 if (adapter->hw.phy.ops.power_up)
3775 adapter->hw.phy.ops.power_up(&adapter->hw);
3776
3777 adapter->hw.mac.ops.setup_link(&adapter->hw);
3778}
3779
3780/**
3781 * e1000_power_down_phy - Power down the PHY
3782 * @adapter: board private structure
3783 *
3784 * Power down the PHY so no link is implied when interface is down.
3785 * The PHY cannot be powered down if management or WoL is active.
3786 */
3787static void e1000_power_down_phy(struct e1000_adapter *adapter)
3788{
3789 if (adapter->hw.phy.ops.power_down)
3790 adapter->hw.phy.ops.power_down(&adapter->hw);
3791}
3792
3793/**
3794 * e1000_flush_tx_ring - remove all descriptors from the tx_ring
3795 * @adapter: board private structure
3796 *
3797 * We want to clear all pending descriptors from the TX ring.
3798 * zeroing happens when the HW reads the regs. We assign the ring itself as
3799 * the data of the next descriptor. We don't care about the data we are about
3800 * to reset the HW.
3801 */
3802static void e1000_flush_tx_ring(struct e1000_adapter *adapter)
3803{
3804 struct e1000_hw *hw = &adapter->hw;
3805 struct e1000_ring *tx_ring = adapter->tx_ring;
3806 struct e1000_tx_desc *tx_desc = NULL;
3807 u32 tdt, tctl, txd_lower = E1000_TXD_CMD_IFCS;
3808 u16 size = 512;
3809
3810 tctl = er32(TCTL);
3811 ew32(TCTL, tctl | E1000_TCTL_EN);
3812 tdt = er32(TDT(0));
3813 BUG_ON(tdt != tx_ring->next_to_use);
3814 tx_desc = E1000_TX_DESC(*tx_ring, tx_ring->next_to_use);
3815 tx_desc->buffer_addr = cpu_to_le64(tx_ring->dma);
3816
3817 tx_desc->lower.data = cpu_to_le32(txd_lower | size);
3818 tx_desc->upper.data = 0;
3819 /* flush descriptors to memory before notifying the HW */
3820 wmb();
3821 tx_ring->next_to_use++;
3822 if (tx_ring->next_to_use == tx_ring->count)
3823 tx_ring->next_to_use = 0;
3824 ew32(TDT(0), tx_ring->next_to_use);
3825 usleep_range(200, 250);
3826}
3827
3828/**
3829 * e1000_flush_rx_ring - remove all descriptors from the rx_ring
3830 * @adapter: board private structure
3831 *
3832 * Mark all descriptors in the RX ring as consumed and disable the rx ring
3833 */
3834static void e1000_flush_rx_ring(struct e1000_adapter *adapter)
3835{
3836 u32 rctl, rxdctl;
3837 struct e1000_hw *hw = &adapter->hw;
3838
3839 rctl = er32(RCTL);
3840 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3841 e1e_flush();
3842 usleep_range(100, 150);
3843
3844 rxdctl = er32(RXDCTL(0));
3845 /* zero the lower 14 bits (prefetch and host thresholds) */
3846 rxdctl &= 0xffffc000;
3847
3848 /* update thresholds: prefetch threshold to 31, host threshold to 1
3849 * and make sure the granularity is "descriptors" and not "cache lines"
3850 */
3851 rxdctl |= (0x1F | BIT(8) | E1000_RXDCTL_THRESH_UNIT_DESC);
3852
3853 ew32(RXDCTL(0), rxdctl);
3854 /* momentarily enable the RX ring for the changes to take effect */
3855 ew32(RCTL, rctl | E1000_RCTL_EN);
3856 e1e_flush();
3857 usleep_range(100, 150);
3858 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3859}
3860
3861/**
3862 * e1000_flush_desc_rings - remove all descriptors from the descriptor rings
3863 * @adapter: board private structure
3864 *
3865 * In i219, the descriptor rings must be emptied before resetting the HW
3866 * or before changing the device state to D3 during runtime (runtime PM).
3867 *
3868 * Failure to do this will cause the HW to enter a unit hang state which can
3869 * only be released by PCI reset on the device
3870 *
3871 */
3872
3873static void e1000_flush_desc_rings(struct e1000_adapter *adapter)
3874{
3875 u16 hang_state;
3876 u32 fext_nvm11, tdlen;
3877 struct e1000_hw *hw = &adapter->hw;
3878
3879 /* First, disable MULR fix in FEXTNVM11 */
3880 fext_nvm11 = er32(FEXTNVM11);
3881 fext_nvm11 |= E1000_FEXTNVM11_DISABLE_MULR_FIX;
3882 ew32(FEXTNVM11, fext_nvm11);
3883 /* do nothing if we're not in faulty state, or if the queue is empty */
3884 tdlen = er32(TDLEN(0));
3885 pci_read_config_word(adapter->pdev, PCICFG_DESC_RING_STATUS,
3886 &hang_state);
3887 if (!(hang_state & FLUSH_DESC_REQUIRED) || !tdlen)
3888 return;
3889 e1000_flush_tx_ring(adapter);
3890 /* recheck, maybe the fault is caused by the rx ring */
3891 pci_read_config_word(adapter->pdev, PCICFG_DESC_RING_STATUS,
3892 &hang_state);
3893 if (hang_state & FLUSH_DESC_REQUIRED)
3894 e1000_flush_rx_ring(adapter);
3895}
3896
3897/**
3898 * e1000e_systim_reset - reset the timesync registers after a hardware reset
3899 * @adapter: board private structure
3900 *
3901 * When the MAC is reset, all hardware bits for timesync will be reset to the
3902 * default values. This function will restore the settings last in place.
3903 * Since the clock SYSTIME registers are reset, we will simply restore the
3904 * cyclecounter to the kernel real clock time.
3905 **/
3906static void e1000e_systim_reset(struct e1000_adapter *adapter)
3907{
3908 struct ptp_clock_info *info = &adapter->ptp_clock_info;
3909 struct e1000_hw *hw = &adapter->hw;
3910 unsigned long flags;
3911 u32 timinca;
3912 s32 ret_val;
3913
3914 if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP))
3915 return;
3916
3917 if (info->adjfine) {
3918 /* restore the previous ptp frequency delta */
3919 ret_val = info->adjfine(info, adapter->ptp_delta);
3920 } else {
3921 /* set the default base frequency if no adjustment possible */
3922 ret_val = e1000e_get_base_timinca(adapter, &timinca);
3923 if (!ret_val)
3924 ew32(TIMINCA, timinca);
3925 }
3926
3927 if (ret_val) {
3928 dev_warn(&adapter->pdev->dev,
3929 "Failed to restore TIMINCA clock rate delta: %d\n",
3930 ret_val);
3931 return;
3932 }
3933
3934 /* reset the systim ns time counter */
3935 spin_lock_irqsave(&adapter->systim_lock, flags);
3936 timecounter_init(&adapter->tc, &adapter->cc,
3937 ktime_to_ns(ktime_get_real()));
3938 spin_unlock_irqrestore(&adapter->systim_lock, flags);
3939
3940 /* restore the previous hwtstamp configuration settings */
3941 e1000e_config_hwtstamp(adapter, &adapter->hwtstamp_config);
3942}
3943
3944/**
3945 * e1000e_reset - bring the hardware into a known good state
3946 * @adapter: board private structure
3947 *
3948 * This function boots the hardware and enables some settings that
3949 * require a configuration cycle of the hardware - those cannot be
3950 * set/changed during runtime. After reset the device needs to be
3951 * properly configured for Rx, Tx etc.
3952 */
3953void e1000e_reset(struct e1000_adapter *adapter)
3954{
3955 struct e1000_mac_info *mac = &adapter->hw.mac;
3956 struct e1000_fc_info *fc = &adapter->hw.fc;
3957 struct e1000_hw *hw = &adapter->hw;
3958 u32 tx_space, min_tx_space, min_rx_space;
3959 u32 pba = adapter->pba;
3960 u16 hwm;
3961
3962 /* reset Packet Buffer Allocation to default */
3963 ew32(PBA, pba);
3964
3965 if (adapter->max_frame_size > (VLAN_ETH_FRAME_LEN + ETH_FCS_LEN)) {
3966 /* To maintain wire speed transmits, the Tx FIFO should be
3967 * large enough to accommodate two full transmit packets,
3968 * rounded up to the next 1KB and expressed in KB. Likewise,
3969 * the Rx FIFO should be large enough to accommodate at least
3970 * one full receive packet and is similarly rounded up and
3971 * expressed in KB.
3972 */
3973 pba = er32(PBA);
3974 /* upper 16 bits has Tx packet buffer allocation size in KB */
3975 tx_space = pba >> 16;
3976 /* lower 16 bits has Rx packet buffer allocation size in KB */
3977 pba &= 0xffff;
3978 /* the Tx fifo also stores 16 bytes of information about the Tx
3979 * but don't include ethernet FCS because hardware appends it
3980 */
3981 min_tx_space = (adapter->max_frame_size +
3982 sizeof(struct e1000_tx_desc) - ETH_FCS_LEN) * 2;
3983 min_tx_space = ALIGN(min_tx_space, 1024);
3984 min_tx_space >>= 10;
3985 /* software strips receive CRC, so leave room for it */
3986 min_rx_space = adapter->max_frame_size;
3987 min_rx_space = ALIGN(min_rx_space, 1024);
3988 min_rx_space >>= 10;
3989
3990 /* If current Tx allocation is less than the min Tx FIFO size,
3991 * and the min Tx FIFO size is less than the current Rx FIFO
3992 * allocation, take space away from current Rx allocation
3993 */
3994 if ((tx_space < min_tx_space) &&
3995 ((min_tx_space - tx_space) < pba)) {
3996 pba -= min_tx_space - tx_space;
3997
3998 /* if short on Rx space, Rx wins and must trump Tx
3999 * adjustment
4000 */
4001 if (pba < min_rx_space)
4002 pba = min_rx_space;
4003 }
4004
4005 ew32(PBA, pba);
4006 }
4007
4008 /* flow control settings
4009 *
4010 * The high water mark must be low enough to fit one full frame
4011 * (or the size used for early receive) above it in the Rx FIFO.
4012 * Set it to the lower of:
4013 * - 90% of the Rx FIFO size, and
4014 * - the full Rx FIFO size minus one full frame
4015 */
4016 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
4017 fc->pause_time = 0xFFFF;
4018 else
4019 fc->pause_time = E1000_FC_PAUSE_TIME;
4020 fc->send_xon = true;
4021 fc->current_mode = fc->requested_mode;
4022
4023 switch (hw->mac.type) {
4024 case e1000_ich9lan:
4025 case e1000_ich10lan:
4026 if (adapter->netdev->mtu > ETH_DATA_LEN) {
4027 pba = 14;
4028 ew32(PBA, pba);
4029 fc->high_water = 0x2800;
4030 fc->low_water = fc->high_water - 8;
4031 break;
4032 }
4033 fallthrough;
4034 default:
4035 hwm = min(((pba << 10) * 9 / 10),
4036 ((pba << 10) - adapter->max_frame_size));
4037
4038 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
4039 fc->low_water = fc->high_water - 8;
4040 break;
4041 case e1000_pchlan:
4042 /* Workaround PCH LOM adapter hangs with certain network
4043 * loads. If hangs persist, try disabling Tx flow control.
4044 */
4045 if (adapter->netdev->mtu > ETH_DATA_LEN) {
4046 fc->high_water = 0x3500;
4047 fc->low_water = 0x1500;
4048 } else {
4049 fc->high_water = 0x5000;
4050 fc->low_water = 0x3000;
4051 }
4052 fc->refresh_time = 0x1000;
4053 break;
4054 case e1000_pch2lan:
4055 case e1000_pch_lpt:
4056 case e1000_pch_spt:
4057 case e1000_pch_cnp:
4058 case e1000_pch_tgp:
4059 case e1000_pch_adp:
4060 case e1000_pch_mtp:
4061 case e1000_pch_lnp:
4062 case e1000_pch_ptp:
4063 case e1000_pch_nvp:
4064 fc->refresh_time = 0xFFFF;
4065 fc->pause_time = 0xFFFF;
4066
4067 if (adapter->netdev->mtu <= ETH_DATA_LEN) {
4068 fc->high_water = 0x05C20;
4069 fc->low_water = 0x05048;
4070 break;
4071 }
4072
4073 pba = 14;
4074 ew32(PBA, pba);
4075 fc->high_water = ((pba << 10) * 9 / 10) & E1000_FCRTH_RTH;
4076 fc->low_water = ((pba << 10) * 8 / 10) & E1000_FCRTL_RTL;
4077 break;
4078 }
4079
4080 /* Alignment of Tx data is on an arbitrary byte boundary with the
4081 * maximum size per Tx descriptor limited only to the transmit
4082 * allocation of the packet buffer minus 96 bytes with an upper
4083 * limit of 24KB due to receive synchronization limitations.
4084 */
4085 adapter->tx_fifo_limit = min_t(u32, ((er32(PBA) >> 16) << 10) - 96,
4086 24 << 10);
4087
4088 /* Disable Adaptive Interrupt Moderation if 2 full packets cannot
4089 * fit in receive buffer.
4090 */
4091 if (adapter->itr_setting & 0x3) {
4092 if ((adapter->max_frame_size * 2) > (pba << 10)) {
4093 if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
4094 dev_info(&adapter->pdev->dev,
4095 "Interrupt Throttle Rate off\n");
4096 adapter->flags2 |= FLAG2_DISABLE_AIM;
4097 e1000e_write_itr(adapter, 0);
4098 }
4099 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
4100 dev_info(&adapter->pdev->dev,
4101 "Interrupt Throttle Rate on\n");
4102 adapter->flags2 &= ~FLAG2_DISABLE_AIM;
4103 adapter->itr = 20000;
4104 e1000e_write_itr(adapter, adapter->itr);
4105 }
4106 }
4107
4108 if (hw->mac.type >= e1000_pch_spt)
4109 e1000_flush_desc_rings(adapter);
4110 /* Allow time for pending master requests to run */
4111 mac->ops.reset_hw(hw);
4112
4113 /* For parts with AMT enabled, let the firmware know
4114 * that the network interface is in control
4115 */
4116 if (adapter->flags & FLAG_HAS_AMT)
4117 e1000e_get_hw_control(adapter);
4118
4119 ew32(WUC, 0);
4120
4121 if (mac->ops.init_hw(hw))
4122 e_err("Hardware Error\n");
4123
4124 e1000_update_mng_vlan(adapter);
4125
4126 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
4127 ew32(VET, ETH_P_8021Q);
4128
4129 e1000e_reset_adaptive(hw);
4130
4131 /* restore systim and hwtstamp settings */
4132 e1000e_systim_reset(adapter);
4133
4134 /* Set EEE advertisement as appropriate */
4135 if (adapter->flags2 & FLAG2_HAS_EEE) {
4136 s32 ret_val;
4137 u16 adv_addr;
4138
4139 switch (hw->phy.type) {
4140 case e1000_phy_82579:
4141 adv_addr = I82579_EEE_ADVERTISEMENT;
4142 break;
4143 case e1000_phy_i217:
4144 adv_addr = I217_EEE_ADVERTISEMENT;
4145 break;
4146 default:
4147 dev_err(&adapter->pdev->dev,
4148 "Invalid PHY type setting EEE advertisement\n");
4149 return;
4150 }
4151
4152 ret_val = hw->phy.ops.acquire(hw);
4153 if (ret_val) {
4154 dev_err(&adapter->pdev->dev,
4155 "EEE advertisement - unable to acquire PHY\n");
4156 return;
4157 }
4158
4159 e1000_write_emi_reg_locked(hw, adv_addr,
4160 hw->dev_spec.ich8lan.eee_disable ?
4161 0 : adapter->eee_advert);
4162
4163 hw->phy.ops.release(hw);
4164 }
4165
4166 if (!netif_running(adapter->netdev) &&
4167 !test_bit(__E1000_TESTING, &adapter->state))
4168 e1000_power_down_phy(adapter);
4169
4170 e1000_get_phy_info(hw);
4171
4172 if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
4173 !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
4174 u16 phy_data = 0;
4175 /* speed up time to link by disabling smart power down, ignore
4176 * the return value of this function because there is nothing
4177 * different we would do if it failed
4178 */
4179 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
4180 phy_data &= ~IGP02E1000_PM_SPD;
4181 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
4182 }
4183 if (hw->mac.type >= e1000_pch_spt && adapter->int_mode == 0) {
4184 u32 reg;
4185
4186 /* Fextnvm7 @ 0xe4[2] = 1 */
4187 reg = er32(FEXTNVM7);
4188 reg |= E1000_FEXTNVM7_SIDE_CLK_UNGATE;
4189 ew32(FEXTNVM7, reg);
4190 /* Fextnvm9 @ 0x5bb4[13:12] = 11 */
4191 reg = er32(FEXTNVM9);
4192 reg |= E1000_FEXTNVM9_IOSFSB_CLKGATE_DIS |
4193 E1000_FEXTNVM9_IOSFSB_CLKREQ_DIS;
4194 ew32(FEXTNVM9, reg);
4195 }
4196
4197}
4198
4199/**
4200 * e1000e_trigger_lsc - trigger an LSC interrupt
4201 * @adapter: board private structure
4202 *
4203 * Fire a link status change interrupt to start the watchdog.
4204 **/
4205static void e1000e_trigger_lsc(struct e1000_adapter *adapter)
4206{
4207 struct e1000_hw *hw = &adapter->hw;
4208
4209 if (adapter->msix_entries)
4210 ew32(ICS, E1000_ICS_LSC | E1000_ICS_OTHER);
4211 else
4212 ew32(ICS, E1000_ICS_LSC);
4213}
4214
4215void e1000e_up(struct e1000_adapter *adapter)
4216{
4217 /* hardware has been reset, we need to reload some things */
4218 e1000_configure(adapter);
4219
4220 clear_bit(__E1000_DOWN, &adapter->state);
4221
4222 if (adapter->msix_entries)
4223 e1000_configure_msix(adapter);
4224 e1000_irq_enable(adapter);
4225
4226 /* Tx queue started by watchdog timer when link is up */
4227
4228 e1000e_trigger_lsc(adapter);
4229}
4230
4231static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
4232{
4233 struct e1000_hw *hw = &adapter->hw;
4234
4235 if (!(adapter->flags2 & FLAG2_DMA_BURST))
4236 return;
4237
4238 /* flush pending descriptor writebacks to memory */
4239 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4240 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4241
4242 /* execute the writes immediately */
4243 e1e_flush();
4244
4245 /* due to rare timing issues, write to TIDV/RDTR again to ensure the
4246 * write is successful
4247 */
4248 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4249 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4250
4251 /* execute the writes immediately */
4252 e1e_flush();
4253}
4254
4255static void e1000e_update_stats(struct e1000_adapter *adapter);
4256
4257/**
4258 * e1000e_down - quiesce the device and optionally reset the hardware
4259 * @adapter: board private structure
4260 * @reset: boolean flag to reset the hardware or not
4261 */
4262void e1000e_down(struct e1000_adapter *adapter, bool reset)
4263{
4264 struct net_device *netdev = adapter->netdev;
4265 struct e1000_hw *hw = &adapter->hw;
4266 u32 tctl, rctl;
4267
4268 /* signal that we're down so the interrupt handler does not
4269 * reschedule our watchdog timer
4270 */
4271 set_bit(__E1000_DOWN, &adapter->state);
4272
4273 netif_carrier_off(netdev);
4274
4275 /* disable receives in the hardware */
4276 rctl = er32(RCTL);
4277 if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
4278 ew32(RCTL, rctl & ~E1000_RCTL_EN);
4279 /* flush and sleep below */
4280
4281 netif_stop_queue(netdev);
4282
4283 /* disable transmits in the hardware */
4284 tctl = er32(TCTL);
4285 tctl &= ~E1000_TCTL_EN;
4286 ew32(TCTL, tctl);
4287
4288 /* flush both disables and wait for them to finish */
4289 e1e_flush();
4290 usleep_range(10000, 11000);
4291
4292 e1000_irq_disable(adapter);
4293
4294 napi_synchronize(&adapter->napi);
4295
4296 del_timer_sync(&adapter->watchdog_timer);
4297 del_timer_sync(&adapter->phy_info_timer);
4298
4299 spin_lock(&adapter->stats64_lock);
4300 e1000e_update_stats(adapter);
4301 spin_unlock(&adapter->stats64_lock);
4302
4303 e1000e_flush_descriptors(adapter);
4304
4305 adapter->link_speed = 0;
4306 adapter->link_duplex = 0;
4307
4308 /* Disable Si errata workaround on PCHx for jumbo frame flow */
4309 if ((hw->mac.type >= e1000_pch2lan) &&
4310 (adapter->netdev->mtu > ETH_DATA_LEN) &&
4311 e1000_lv_jumbo_workaround_ich8lan(hw, false))
4312 e_dbg("failed to disable jumbo frame workaround mode\n");
4313
4314 if (!pci_channel_offline(adapter->pdev)) {
4315 if (reset)
4316 e1000e_reset(adapter);
4317 else if (hw->mac.type >= e1000_pch_spt)
4318 e1000_flush_desc_rings(adapter);
4319 }
4320 e1000_clean_tx_ring(adapter->tx_ring);
4321 e1000_clean_rx_ring(adapter->rx_ring);
4322}
4323
4324void e1000e_reinit_locked(struct e1000_adapter *adapter)
4325{
4326 might_sleep();
4327 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4328 usleep_range(1000, 1100);
4329 e1000e_down(adapter, true);
4330 e1000e_up(adapter);
4331 clear_bit(__E1000_RESETTING, &adapter->state);
4332}
4333
4334/**
4335 * e1000e_sanitize_systim - sanitize raw cycle counter reads
4336 * @hw: pointer to the HW structure
4337 * @systim: PHC time value read, sanitized and returned
4338 * @sts: structure to hold system time before and after reading SYSTIML,
4339 * may be NULL
4340 *
4341 * Errata for 82574/82583 possible bad bits read from SYSTIMH/L:
4342 * check to see that the time is incrementing at a reasonable
4343 * rate and is a multiple of incvalue.
4344 **/
4345static u64 e1000e_sanitize_systim(struct e1000_hw *hw, u64 systim,
4346 struct ptp_system_timestamp *sts)
4347{
4348 u64 time_delta, rem, temp;
4349 u64 systim_next;
4350 u32 incvalue;
4351 int i;
4352
4353 incvalue = er32(TIMINCA) & E1000_TIMINCA_INCVALUE_MASK;
4354 for (i = 0; i < E1000_MAX_82574_SYSTIM_REREADS; i++) {
4355 /* latch SYSTIMH on read of SYSTIML */
4356 ptp_read_system_prets(sts);
4357 systim_next = (u64)er32(SYSTIML);
4358 ptp_read_system_postts(sts);
4359 systim_next |= (u64)er32(SYSTIMH) << 32;
4360
4361 time_delta = systim_next - systim;
4362 temp = time_delta;
4363 /* VMWare users have seen incvalue of zero, don't div / 0 */
4364 rem = incvalue ? do_div(temp, incvalue) : (time_delta != 0);
4365
4366 systim = systim_next;
4367
4368 if ((time_delta < E1000_82574_SYSTIM_EPSILON) && (rem == 0))
4369 break;
4370 }
4371
4372 return systim;
4373}
4374
4375/**
4376 * e1000e_read_systim - read SYSTIM register
4377 * @adapter: board private structure
4378 * @sts: structure which will contain system time before and after reading
4379 * SYSTIML, may be NULL
4380 **/
4381u64 e1000e_read_systim(struct e1000_adapter *adapter,
4382 struct ptp_system_timestamp *sts)
4383{
4384 struct e1000_hw *hw = &adapter->hw;
4385 u32 systimel, systimel_2, systimeh;
4386 u64 systim;
4387 /* SYSTIMH latching upon SYSTIML read does not work well.
4388 * This means that if SYSTIML overflows after we read it but before
4389 * we read SYSTIMH, the value of SYSTIMH has been incremented and we
4390 * will experience a huge non linear increment in the systime value
4391 * to fix that we test for overflow and if true, we re-read systime.
4392 */
4393 ptp_read_system_prets(sts);
4394 systimel = er32(SYSTIML);
4395 ptp_read_system_postts(sts);
4396 systimeh = er32(SYSTIMH);
4397 /* Is systimel is so large that overflow is possible? */
4398 if (systimel >= (u32)0xffffffff - E1000_TIMINCA_INCVALUE_MASK) {
4399 ptp_read_system_prets(sts);
4400 systimel_2 = er32(SYSTIML);
4401 ptp_read_system_postts(sts);
4402 if (systimel > systimel_2) {
4403 /* There was an overflow, read again SYSTIMH, and use
4404 * systimel_2
4405 */
4406 systimeh = er32(SYSTIMH);
4407 systimel = systimel_2;
4408 }
4409 }
4410 systim = (u64)systimel;
4411 systim |= (u64)systimeh << 32;
4412
4413 if (adapter->flags2 & FLAG2_CHECK_SYSTIM_OVERFLOW)
4414 systim = e1000e_sanitize_systim(hw, systim, sts);
4415
4416 return systim;
4417}
4418
4419/**
4420 * e1000e_cyclecounter_read - read raw cycle counter (used by time counter)
4421 * @cc: cyclecounter structure
4422 **/
4423static u64 e1000e_cyclecounter_read(const struct cyclecounter *cc)
4424{
4425 struct e1000_adapter *adapter = container_of(cc, struct e1000_adapter,
4426 cc);
4427
4428 return e1000e_read_systim(adapter, NULL);
4429}
4430
4431/**
4432 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
4433 * @adapter: board private structure to initialize
4434 *
4435 * e1000_sw_init initializes the Adapter private data structure.
4436 * Fields are initialized based on PCI device information and
4437 * OS network device settings (MTU size).
4438 **/
4439static int e1000_sw_init(struct e1000_adapter *adapter)
4440{
4441 struct net_device *netdev = adapter->netdev;
4442
4443 adapter->rx_buffer_len = VLAN_ETH_FRAME_LEN + ETH_FCS_LEN;
4444 adapter->rx_ps_bsize0 = 128;
4445 adapter->max_frame_size = netdev->mtu + VLAN_ETH_HLEN + ETH_FCS_LEN;
4446 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
4447 adapter->tx_ring_count = E1000_DEFAULT_TXD;
4448 adapter->rx_ring_count = E1000_DEFAULT_RXD;
4449
4450 spin_lock_init(&adapter->stats64_lock);
4451
4452 e1000e_set_interrupt_capability(adapter);
4453
4454 if (e1000_alloc_queues(adapter))
4455 return -ENOMEM;
4456
4457 /* Setup hardware time stamping cyclecounter */
4458 if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
4459 adapter->cc.read = e1000e_cyclecounter_read;
4460 adapter->cc.mask = CYCLECOUNTER_MASK(64);
4461 adapter->cc.mult = 1;
4462 /* cc.shift set in e1000e_get_base_tininca() */
4463
4464 spin_lock_init(&adapter->systim_lock);
4465 INIT_WORK(&adapter->tx_hwtstamp_work, e1000e_tx_hwtstamp_work);
4466 }
4467
4468 /* Explicitly disable IRQ since the NIC can be in any state. */
4469 e1000_irq_disable(adapter);
4470
4471 set_bit(__E1000_DOWN, &adapter->state);
4472 return 0;
4473}
4474
4475/**
4476 * e1000_intr_msi_test - Interrupt Handler
4477 * @irq: interrupt number
4478 * @data: pointer to a network interface device structure
4479 **/
4480static irqreturn_t e1000_intr_msi_test(int __always_unused irq, void *data)
4481{
4482 struct net_device *netdev = data;
4483 struct e1000_adapter *adapter = netdev_priv(netdev);
4484 struct e1000_hw *hw = &adapter->hw;
4485 u32 icr = er32(ICR);
4486
4487 e_dbg("icr is %08X\n", icr);
4488 if (icr & E1000_ICR_RXSEQ) {
4489 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
4490 /* Force memory writes to complete before acknowledging the
4491 * interrupt is handled.
4492 */
4493 wmb();
4494 }
4495
4496 return IRQ_HANDLED;
4497}
4498
4499/**
4500 * e1000_test_msi_interrupt - Returns 0 for successful test
4501 * @adapter: board private struct
4502 *
4503 * code flow taken from tg3.c
4504 **/
4505static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
4506{
4507 struct net_device *netdev = adapter->netdev;
4508 struct e1000_hw *hw = &adapter->hw;
4509 int err;
4510
4511 /* poll_enable hasn't been called yet, so don't need disable */
4512 /* clear any pending events */
4513 er32(ICR);
4514
4515 /* free the real vector and request a test handler */
4516 e1000_free_irq(adapter);
4517 e1000e_reset_interrupt_capability(adapter);
4518
4519 /* Assume that the test fails, if it succeeds then the test
4520 * MSI irq handler will unset this flag
4521 */
4522 adapter->flags |= FLAG_MSI_TEST_FAILED;
4523
4524 err = pci_enable_msi(adapter->pdev);
4525 if (err)
4526 goto msi_test_failed;
4527
4528 err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
4529 netdev->name, netdev);
4530 if (err) {
4531 pci_disable_msi(adapter->pdev);
4532 goto msi_test_failed;
4533 }
4534
4535 /* Force memory writes to complete before enabling and firing an
4536 * interrupt.
4537 */
4538 wmb();
4539
4540 e1000_irq_enable(adapter);
4541
4542 /* fire an unusual interrupt on the test handler */
4543 ew32(ICS, E1000_ICS_RXSEQ);
4544 e1e_flush();
4545 msleep(100);
4546
4547 e1000_irq_disable(adapter);
4548
4549 rmb(); /* read flags after interrupt has been fired */
4550
4551 if (adapter->flags & FLAG_MSI_TEST_FAILED) {
4552 adapter->int_mode = E1000E_INT_MODE_LEGACY;
4553 e_info("MSI interrupt test failed, using legacy interrupt.\n");
4554 } else {
4555 e_dbg("MSI interrupt test succeeded!\n");
4556 }
4557
4558 free_irq(adapter->pdev->irq, netdev);
4559 pci_disable_msi(adapter->pdev);
4560
4561msi_test_failed:
4562 e1000e_set_interrupt_capability(adapter);
4563 return e1000_request_irq(adapter);
4564}
4565
4566/**
4567 * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
4568 * @adapter: board private struct
4569 *
4570 * code flow taken from tg3.c, called with e1000 interrupts disabled.
4571 **/
4572static int e1000_test_msi(struct e1000_adapter *adapter)
4573{
4574 int err;
4575 u16 pci_cmd;
4576
4577 if (!(adapter->flags & FLAG_MSI_ENABLED))
4578 return 0;
4579
4580 /* disable SERR in case the MSI write causes a master abort */
4581 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4582 if (pci_cmd & PCI_COMMAND_SERR)
4583 pci_write_config_word(adapter->pdev, PCI_COMMAND,
4584 pci_cmd & ~PCI_COMMAND_SERR);
4585
4586 err = e1000_test_msi_interrupt(adapter);
4587
4588 /* re-enable SERR */
4589 if (pci_cmd & PCI_COMMAND_SERR) {
4590 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4591 pci_cmd |= PCI_COMMAND_SERR;
4592 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
4593 }
4594
4595 return err;
4596}
4597
4598/**
4599 * e1000e_open - Called when a network interface is made active
4600 * @netdev: network interface device structure
4601 *
4602 * Returns 0 on success, negative value on failure
4603 *
4604 * The open entry point is called when a network interface is made
4605 * active by the system (IFF_UP). At this point all resources needed
4606 * for transmit and receive operations are allocated, the interrupt
4607 * handler is registered with the OS, the watchdog timer is started,
4608 * and the stack is notified that the interface is ready.
4609 **/
4610int e1000e_open(struct net_device *netdev)
4611{
4612 struct e1000_adapter *adapter = netdev_priv(netdev);
4613 struct e1000_hw *hw = &adapter->hw;
4614 struct pci_dev *pdev = adapter->pdev;
4615 int err;
4616
4617 /* disallow open during test */
4618 if (test_bit(__E1000_TESTING, &adapter->state))
4619 return -EBUSY;
4620
4621 pm_runtime_get_sync(&pdev->dev);
4622
4623 netif_carrier_off(netdev);
4624 netif_stop_queue(netdev);
4625
4626 /* allocate transmit descriptors */
4627 err = e1000e_setup_tx_resources(adapter->tx_ring);
4628 if (err)
4629 goto err_setup_tx;
4630
4631 /* allocate receive descriptors */
4632 err = e1000e_setup_rx_resources(adapter->rx_ring);
4633 if (err)
4634 goto err_setup_rx;
4635
4636 /* If AMT is enabled, let the firmware know that the network
4637 * interface is now open and reset the part to a known state.
4638 */
4639 if (adapter->flags & FLAG_HAS_AMT) {
4640 e1000e_get_hw_control(adapter);
4641 e1000e_reset(adapter);
4642 }
4643
4644 e1000e_power_up_phy(adapter);
4645
4646 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4647 if ((adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
4648 e1000_update_mng_vlan(adapter);
4649
4650 /* DMA latency requirement to workaround jumbo issue */
4651 cpu_latency_qos_add_request(&adapter->pm_qos_req, PM_QOS_DEFAULT_VALUE);
4652
4653 /* before we allocate an interrupt, we must be ready to handle it.
4654 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
4655 * as soon as we call pci_request_irq, so we have to setup our
4656 * clean_rx handler before we do so.
4657 */
4658 e1000_configure(adapter);
4659
4660 err = e1000_request_irq(adapter);
4661 if (err)
4662 goto err_req_irq;
4663
4664 /* Work around PCIe errata with MSI interrupts causing some chipsets to
4665 * ignore e1000e MSI messages, which means we need to test our MSI
4666 * interrupt now
4667 */
4668 if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
4669 err = e1000_test_msi(adapter);
4670 if (err) {
4671 e_err("Interrupt allocation failed\n");
4672 goto err_req_irq;
4673 }
4674 }
4675
4676 /* From here on the code is the same as e1000e_up() */
4677 clear_bit(__E1000_DOWN, &adapter->state);
4678
4679 napi_enable(&adapter->napi);
4680
4681 e1000_irq_enable(adapter);
4682
4683 adapter->tx_hang_recheck = false;
4684
4685 hw->mac.get_link_status = true;
4686 pm_runtime_put(&pdev->dev);
4687
4688 e1000e_trigger_lsc(adapter);
4689
4690 return 0;
4691
4692err_req_irq:
4693 cpu_latency_qos_remove_request(&adapter->pm_qos_req);
4694 e1000e_release_hw_control(adapter);
4695 e1000_power_down_phy(adapter);
4696 e1000e_free_rx_resources(adapter->rx_ring);
4697err_setup_rx:
4698 e1000e_free_tx_resources(adapter->tx_ring);
4699err_setup_tx:
4700 e1000e_reset(adapter);
4701 pm_runtime_put_sync(&pdev->dev);
4702
4703 return err;
4704}
4705
4706/**
4707 * e1000e_close - Disables a network interface
4708 * @netdev: network interface device structure
4709 *
4710 * Returns 0, this is not allowed to fail
4711 *
4712 * The close entry point is called when an interface is de-activated
4713 * by the OS. The hardware is still under the drivers control, but
4714 * needs to be disabled. A global MAC reset is issued to stop the
4715 * hardware, and all transmit and receive resources are freed.
4716 **/
4717int e1000e_close(struct net_device *netdev)
4718{
4719 struct e1000_adapter *adapter = netdev_priv(netdev);
4720 struct pci_dev *pdev = adapter->pdev;
4721 int count = E1000_CHECK_RESET_COUNT;
4722
4723 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
4724 usleep_range(10000, 11000);
4725
4726 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4727
4728 pm_runtime_get_sync(&pdev->dev);
4729
4730 if (netif_device_present(netdev)) {
4731 e1000e_down(adapter, true);
4732 e1000_free_irq(adapter);
4733
4734 /* Link status message must follow this format */
4735 netdev_info(netdev, "NIC Link is Down\n");
4736 }
4737
4738 napi_disable(&adapter->napi);
4739
4740 e1000e_free_tx_resources(adapter->tx_ring);
4741 e1000e_free_rx_resources(adapter->rx_ring);
4742
4743 /* kill manageability vlan ID if supported, but not if a vlan with
4744 * the same ID is registered on the host OS (let 8021q kill it)
4745 */
4746 if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
4747 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
4748 adapter->mng_vlan_id);
4749
4750 /* If AMT is enabled, let the firmware know that the network
4751 * interface is now closed
4752 */
4753 if ((adapter->flags & FLAG_HAS_AMT) &&
4754 !test_bit(__E1000_TESTING, &adapter->state))
4755 e1000e_release_hw_control(adapter);
4756
4757 cpu_latency_qos_remove_request(&adapter->pm_qos_req);
4758
4759 pm_runtime_put_sync(&pdev->dev);
4760
4761 return 0;
4762}
4763
4764/**
4765 * e1000_set_mac - Change the Ethernet Address of the NIC
4766 * @netdev: network interface device structure
4767 * @p: pointer to an address structure
4768 *
4769 * Returns 0 on success, negative on failure
4770 **/
4771static int e1000_set_mac(struct net_device *netdev, void *p)
4772{
4773 struct e1000_adapter *adapter = netdev_priv(netdev);
4774 struct e1000_hw *hw = &adapter->hw;
4775 struct sockaddr *addr = p;
4776
4777 if (!is_valid_ether_addr(addr->sa_data))
4778 return -EADDRNOTAVAIL;
4779
4780 eth_hw_addr_set(netdev, addr->sa_data);
4781 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
4782
4783 hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
4784
4785 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
4786 /* activate the work around */
4787 e1000e_set_laa_state_82571(&adapter->hw, 1);
4788
4789 /* Hold a copy of the LAA in RAR[14] This is done so that
4790 * between the time RAR[0] gets clobbered and the time it
4791 * gets fixed (in e1000_watchdog), the actual LAA is in one
4792 * of the RARs and no incoming packets directed to this port
4793 * are dropped. Eventually the LAA will be in RAR[0] and
4794 * RAR[14]
4795 */
4796 hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr,
4797 adapter->hw.mac.rar_entry_count - 1);
4798 }
4799
4800 return 0;
4801}
4802
4803/**
4804 * e1000e_update_phy_task - work thread to update phy
4805 * @work: pointer to our work struct
4806 *
4807 * this worker thread exists because we must acquire a
4808 * semaphore to read the phy, which we could msleep while
4809 * waiting for it, and we can't msleep in a timer.
4810 **/
4811static void e1000e_update_phy_task(struct work_struct *work)
4812{
4813 struct e1000_adapter *adapter = container_of(work,
4814 struct e1000_adapter,
4815 update_phy_task);
4816 struct e1000_hw *hw = &adapter->hw;
4817
4818 if (test_bit(__E1000_DOWN, &adapter->state))
4819 return;
4820
4821 e1000_get_phy_info(hw);
4822
4823 /* Enable EEE on 82579 after link up */
4824 if (hw->phy.type >= e1000_phy_82579)
4825 e1000_set_eee_pchlan(hw);
4826}
4827
4828/**
4829 * e1000_update_phy_info - timre call-back to update PHY info
4830 * @t: pointer to timer_list containing private info adapter
4831 *
4832 * Need to wait a few seconds after link up to get diagnostic information from
4833 * the phy
4834 **/
4835static void e1000_update_phy_info(struct timer_list *t)
4836{
4837 struct e1000_adapter *adapter = from_timer(adapter, t, phy_info_timer);
4838
4839 if (test_bit(__E1000_DOWN, &adapter->state))
4840 return;
4841
4842 schedule_work(&adapter->update_phy_task);
4843}
4844
4845/**
4846 * e1000e_update_phy_stats - Update the PHY statistics counters
4847 * @adapter: board private structure
4848 *
4849 * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4850 **/
4851static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4852{
4853 struct e1000_hw *hw = &adapter->hw;
4854 s32 ret_val;
4855 u16 phy_data;
4856
4857 ret_val = hw->phy.ops.acquire(hw);
4858 if (ret_val)
4859 return;
4860
4861 /* A page set is expensive so check if already on desired page.
4862 * If not, set to the page with the PHY status registers.
4863 */
4864 hw->phy.addr = 1;
4865 ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4866 &phy_data);
4867 if (ret_val)
4868 goto release;
4869 if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4870 ret_val = hw->phy.ops.set_page(hw,
4871 HV_STATS_PAGE << IGP_PAGE_SHIFT);
4872 if (ret_val)
4873 goto release;
4874 }
4875
4876 /* Single Collision Count */
4877 hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4878 ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4879 if (!ret_val)
4880 adapter->stats.scc += phy_data;
4881
4882 /* Excessive Collision Count */
4883 hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4884 ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4885 if (!ret_val)
4886 adapter->stats.ecol += phy_data;
4887
4888 /* Multiple Collision Count */
4889 hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4890 ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4891 if (!ret_val)
4892 adapter->stats.mcc += phy_data;
4893
4894 /* Late Collision Count */
4895 hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4896 ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4897 if (!ret_val)
4898 adapter->stats.latecol += phy_data;
4899
4900 /* Collision Count - also used for adaptive IFS */
4901 hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4902 ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4903 if (!ret_val)
4904 hw->mac.collision_delta = phy_data;
4905
4906 /* Defer Count */
4907 hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4908 ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4909 if (!ret_val)
4910 adapter->stats.dc += phy_data;
4911
4912 /* Transmit with no CRS */
4913 hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4914 ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4915 if (!ret_val)
4916 adapter->stats.tncrs += phy_data;
4917
4918release:
4919 hw->phy.ops.release(hw);
4920}
4921
4922/**
4923 * e1000e_update_stats - Update the board statistics counters
4924 * @adapter: board private structure
4925 **/
4926static void e1000e_update_stats(struct e1000_adapter *adapter)
4927{
4928 struct net_device *netdev = adapter->netdev;
4929 struct e1000_hw *hw = &adapter->hw;
4930 struct pci_dev *pdev = adapter->pdev;
4931
4932 /* Prevent stats update while adapter is being reset, or if the pci
4933 * connection is down.
4934 */
4935 if (adapter->link_speed == 0)
4936 return;
4937 if (pci_channel_offline(pdev))
4938 return;
4939
4940 adapter->stats.crcerrs += er32(CRCERRS);
4941 adapter->stats.gprc += er32(GPRC);
4942 adapter->stats.gorc += er32(GORCL);
4943 er32(GORCH); /* Clear gorc */
4944 adapter->stats.bprc += er32(BPRC);
4945 adapter->stats.mprc += er32(MPRC);
4946 adapter->stats.roc += er32(ROC);
4947
4948 adapter->stats.mpc += er32(MPC);
4949
4950 /* Half-duplex statistics */
4951 if (adapter->link_duplex == HALF_DUPLEX) {
4952 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4953 e1000e_update_phy_stats(adapter);
4954 } else {
4955 adapter->stats.scc += er32(SCC);
4956 adapter->stats.ecol += er32(ECOL);
4957 adapter->stats.mcc += er32(MCC);
4958 adapter->stats.latecol += er32(LATECOL);
4959 adapter->stats.dc += er32(DC);
4960
4961 hw->mac.collision_delta = er32(COLC);
4962
4963 if ((hw->mac.type != e1000_82574) &&
4964 (hw->mac.type != e1000_82583))
4965 adapter->stats.tncrs += er32(TNCRS);
4966 }
4967 adapter->stats.colc += hw->mac.collision_delta;
4968 }
4969
4970 adapter->stats.xonrxc += er32(XONRXC);
4971 adapter->stats.xontxc += er32(XONTXC);
4972 adapter->stats.xoffrxc += er32(XOFFRXC);
4973 adapter->stats.xofftxc += er32(XOFFTXC);
4974 adapter->stats.gptc += er32(GPTC);
4975 adapter->stats.gotc += er32(GOTCL);
4976 er32(GOTCH); /* Clear gotc */
4977 adapter->stats.rnbc += er32(RNBC);
4978 adapter->stats.ruc += er32(RUC);
4979
4980 adapter->stats.mptc += er32(MPTC);
4981 adapter->stats.bptc += er32(BPTC);
4982
4983 /* used for adaptive IFS */
4984
4985 hw->mac.tx_packet_delta = er32(TPT);
4986 adapter->stats.tpt += hw->mac.tx_packet_delta;
4987
4988 adapter->stats.algnerrc += er32(ALGNERRC);
4989 adapter->stats.rxerrc += er32(RXERRC);
4990 adapter->stats.cexterr += er32(CEXTERR);
4991 adapter->stats.tsctc += er32(TSCTC);
4992 adapter->stats.tsctfc += er32(TSCTFC);
4993
4994 /* Fill out the OS statistics structure */
4995 netdev->stats.multicast = adapter->stats.mprc;
4996 netdev->stats.collisions = adapter->stats.colc;
4997
4998 /* Rx Errors */
4999
5000 /* RLEC on some newer hardware can be incorrect so build
5001 * our own version based on RUC and ROC
5002 */
5003 netdev->stats.rx_errors = adapter->stats.rxerrc +
5004 adapter->stats.crcerrs + adapter->stats.algnerrc +
5005 adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
5006 netdev->stats.rx_length_errors = adapter->stats.ruc +
5007 adapter->stats.roc;
5008 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
5009 netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
5010 netdev->stats.rx_missed_errors = adapter->stats.mpc;
5011
5012 /* Tx Errors */
5013 netdev->stats.tx_errors = adapter->stats.ecol + adapter->stats.latecol;
5014 netdev->stats.tx_aborted_errors = adapter->stats.ecol;
5015 netdev->stats.tx_window_errors = adapter->stats.latecol;
5016 netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
5017
5018 /* Tx Dropped needs to be maintained elsewhere */
5019
5020 /* Management Stats */
5021 adapter->stats.mgptc += er32(MGTPTC);
5022 adapter->stats.mgprc += er32(MGTPRC);
5023 adapter->stats.mgpdc += er32(MGTPDC);
5024
5025 /* Correctable ECC Errors */
5026 if (hw->mac.type >= e1000_pch_lpt) {
5027 u32 pbeccsts = er32(PBECCSTS);
5028
5029 adapter->corr_errors +=
5030 pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
5031 adapter->uncorr_errors +=
5032 FIELD_GET(E1000_PBECCSTS_UNCORR_ERR_CNT_MASK, pbeccsts);
5033 }
5034}
5035
5036/**
5037 * e1000_phy_read_status - Update the PHY register status snapshot
5038 * @adapter: board private structure
5039 **/
5040static void e1000_phy_read_status(struct e1000_adapter *adapter)
5041{
5042 struct e1000_hw *hw = &adapter->hw;
5043 struct e1000_phy_regs *phy = &adapter->phy_regs;
5044
5045 if (!pm_runtime_suspended((&adapter->pdev->dev)->parent) &&
5046 (er32(STATUS) & E1000_STATUS_LU) &&
5047 (adapter->hw.phy.media_type == e1000_media_type_copper)) {
5048 int ret_val;
5049
5050 ret_val = e1e_rphy(hw, MII_BMCR, &phy->bmcr);
5051 ret_val |= e1e_rphy(hw, MII_BMSR, &phy->bmsr);
5052 ret_val |= e1e_rphy(hw, MII_ADVERTISE, &phy->advertise);
5053 ret_val |= e1e_rphy(hw, MII_LPA, &phy->lpa);
5054 ret_val |= e1e_rphy(hw, MII_EXPANSION, &phy->expansion);
5055 ret_val |= e1e_rphy(hw, MII_CTRL1000, &phy->ctrl1000);
5056 ret_val |= e1e_rphy(hw, MII_STAT1000, &phy->stat1000);
5057 ret_val |= e1e_rphy(hw, MII_ESTATUS, &phy->estatus);
5058 if (ret_val)
5059 e_warn("Error reading PHY register\n");
5060 } else {
5061 /* Do not read PHY registers if link is not up
5062 * Set values to typical power-on defaults
5063 */
5064 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
5065 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
5066 BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
5067 BMSR_ERCAP);
5068 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
5069 ADVERTISE_ALL | ADVERTISE_CSMA);
5070 phy->lpa = 0;
5071 phy->expansion = EXPANSION_ENABLENPAGE;
5072 phy->ctrl1000 = ADVERTISE_1000FULL;
5073 phy->stat1000 = 0;
5074 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
5075 }
5076}
5077
5078static void e1000_print_link_info(struct e1000_adapter *adapter)
5079{
5080 struct e1000_hw *hw = &adapter->hw;
5081 u32 ctrl = er32(CTRL);
5082
5083 /* Link status message must follow this format for user tools */
5084 netdev_info(adapter->netdev,
5085 "NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
5086 adapter->link_speed,
5087 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
5088 (ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
5089 (ctrl & E1000_CTRL_RFCE) ? "Rx" :
5090 (ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
5091}
5092
5093static bool e1000e_has_link(struct e1000_adapter *adapter)
5094{
5095 struct e1000_hw *hw = &adapter->hw;
5096 bool link_active = false;
5097 s32 ret_val = 0;
5098
5099 /* get_link_status is set on LSC (link status) interrupt or
5100 * Rx sequence error interrupt. get_link_status will stay
5101 * true until the check_for_link establishes link
5102 * for copper adapters ONLY
5103 */
5104 switch (hw->phy.media_type) {
5105 case e1000_media_type_copper:
5106 if (hw->mac.get_link_status) {
5107 ret_val = hw->mac.ops.check_for_link(hw);
5108 link_active = !hw->mac.get_link_status;
5109 } else {
5110 link_active = true;
5111 }
5112 break;
5113 case e1000_media_type_fiber:
5114 ret_val = hw->mac.ops.check_for_link(hw);
5115 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
5116 break;
5117 case e1000_media_type_internal_serdes:
5118 ret_val = hw->mac.ops.check_for_link(hw);
5119 link_active = hw->mac.serdes_has_link;
5120 break;
5121 default:
5122 case e1000_media_type_unknown:
5123 break;
5124 }
5125
5126 if ((ret_val == -E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
5127 (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
5128 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
5129 e_info("Gigabit has been disabled, downgrading speed\n");
5130 }
5131
5132 return link_active;
5133}
5134
5135static void e1000e_enable_receives(struct e1000_adapter *adapter)
5136{
5137 /* make sure the receive unit is started */
5138 if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5139 (adapter->flags & FLAG_RESTART_NOW)) {
5140 struct e1000_hw *hw = &adapter->hw;
5141 u32 rctl = er32(RCTL);
5142
5143 ew32(RCTL, rctl | E1000_RCTL_EN);
5144 adapter->flags &= ~FLAG_RESTART_NOW;
5145 }
5146}
5147
5148static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
5149{
5150 struct e1000_hw *hw = &adapter->hw;
5151
5152 /* With 82574 controllers, PHY needs to be checked periodically
5153 * for hung state and reset, if two calls return true
5154 */
5155 if (e1000_check_phy_82574(hw))
5156 adapter->phy_hang_count++;
5157 else
5158 adapter->phy_hang_count = 0;
5159
5160 if (adapter->phy_hang_count > 1) {
5161 adapter->phy_hang_count = 0;
5162 e_dbg("PHY appears hung - resetting\n");
5163 schedule_work(&adapter->reset_task);
5164 }
5165}
5166
5167/**
5168 * e1000_watchdog - Timer Call-back
5169 * @t: pointer to timer_list containing private info adapter
5170 **/
5171static void e1000_watchdog(struct timer_list *t)
5172{
5173 struct e1000_adapter *adapter = from_timer(adapter, t, watchdog_timer);
5174
5175 /* Do the rest outside of interrupt context */
5176 schedule_work(&adapter->watchdog_task);
5177
5178 /* TODO: make this use queue_delayed_work() */
5179}
5180
5181static void e1000_watchdog_task(struct work_struct *work)
5182{
5183 struct e1000_adapter *adapter = container_of(work,
5184 struct e1000_adapter,
5185 watchdog_task);
5186 struct net_device *netdev = adapter->netdev;
5187 struct e1000_mac_info *mac = &adapter->hw.mac;
5188 struct e1000_phy_info *phy = &adapter->hw.phy;
5189 struct e1000_ring *tx_ring = adapter->tx_ring;
5190 u32 dmoff_exit_timeout = 100, tries = 0;
5191 struct e1000_hw *hw = &adapter->hw;
5192 u32 link, tctl, pcim_state;
5193
5194 if (test_bit(__E1000_DOWN, &adapter->state))
5195 return;
5196
5197 link = e1000e_has_link(adapter);
5198 if ((netif_carrier_ok(netdev)) && link) {
5199 /* Cancel scheduled suspend requests. */
5200 pm_runtime_resume(netdev->dev.parent);
5201
5202 e1000e_enable_receives(adapter);
5203 goto link_up;
5204 }
5205
5206 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
5207 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
5208 e1000_update_mng_vlan(adapter);
5209
5210 if (link) {
5211 if (!netif_carrier_ok(netdev)) {
5212 bool txb2b = true;
5213
5214 /* Cancel scheduled suspend requests. */
5215 pm_runtime_resume(netdev->dev.parent);
5216
5217 /* Checking if MAC is in DMoff state*/
5218 if (er32(FWSM) & E1000_ICH_FWSM_FW_VALID) {
5219 pcim_state = er32(STATUS);
5220 while (pcim_state & E1000_STATUS_PCIM_STATE) {
5221 if (tries++ == dmoff_exit_timeout) {
5222 e_dbg("Error in exiting dmoff\n");
5223 break;
5224 }
5225 usleep_range(10000, 20000);
5226 pcim_state = er32(STATUS);
5227
5228 /* Checking if MAC exited DMoff state */
5229 if (!(pcim_state & E1000_STATUS_PCIM_STATE))
5230 e1000_phy_hw_reset(&adapter->hw);
5231 }
5232 }
5233
5234 /* update snapshot of PHY registers on LSC */
5235 e1000_phy_read_status(adapter);
5236 mac->ops.get_link_up_info(&adapter->hw,
5237 &adapter->link_speed,
5238 &adapter->link_duplex);
5239 e1000_print_link_info(adapter);
5240
5241 /* check if SmartSpeed worked */
5242 e1000e_check_downshift(hw);
5243 if (phy->speed_downgraded)
5244 netdev_warn(netdev,
5245 "Link Speed was downgraded by SmartSpeed\n");
5246
5247 /* On supported PHYs, check for duplex mismatch only
5248 * if link has autonegotiated at 10/100 half
5249 */
5250 if ((hw->phy.type == e1000_phy_igp_3 ||
5251 hw->phy.type == e1000_phy_bm) &&
5252 hw->mac.autoneg &&
5253 (adapter->link_speed == SPEED_10 ||
5254 adapter->link_speed == SPEED_100) &&
5255 (adapter->link_duplex == HALF_DUPLEX)) {
5256 u16 autoneg_exp;
5257
5258 e1e_rphy(hw, MII_EXPANSION, &autoneg_exp);
5259
5260 if (!(autoneg_exp & EXPANSION_NWAY))
5261 e_info("Autonegotiated half duplex but link partner cannot autoneg. Try forcing full duplex if link gets many collisions.\n");
5262 }
5263
5264 /* adjust timeout factor according to speed/duplex */
5265 adapter->tx_timeout_factor = 1;
5266 switch (adapter->link_speed) {
5267 case SPEED_10:
5268 txb2b = false;
5269 adapter->tx_timeout_factor = 16;
5270 break;
5271 case SPEED_100:
5272 txb2b = false;
5273 adapter->tx_timeout_factor = 10;
5274 break;
5275 }
5276
5277 /* workaround: re-program speed mode bit after
5278 * link-up event
5279 */
5280 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
5281 !txb2b) {
5282 u32 tarc0;
5283
5284 tarc0 = er32(TARC(0));
5285 tarc0 &= ~SPEED_MODE_BIT;
5286 ew32(TARC(0), tarc0);
5287 }
5288
5289 /* enable transmits in the hardware, need to do this
5290 * after setting TARC(0)
5291 */
5292 tctl = er32(TCTL);
5293 tctl |= E1000_TCTL_EN;
5294 ew32(TCTL, tctl);
5295
5296 /* Perform any post-link-up configuration before
5297 * reporting link up.
5298 */
5299 if (phy->ops.cfg_on_link_up)
5300 phy->ops.cfg_on_link_up(hw);
5301
5302 netif_wake_queue(netdev);
5303 netif_carrier_on(netdev);
5304
5305 if (!test_bit(__E1000_DOWN, &adapter->state))
5306 mod_timer(&adapter->phy_info_timer,
5307 round_jiffies(jiffies + 2 * HZ));
5308 }
5309 } else {
5310 if (netif_carrier_ok(netdev)) {
5311 adapter->link_speed = 0;
5312 adapter->link_duplex = 0;
5313 /* Link status message must follow this format */
5314 netdev_info(netdev, "NIC Link is Down\n");
5315 netif_carrier_off(netdev);
5316 netif_stop_queue(netdev);
5317 if (!test_bit(__E1000_DOWN, &adapter->state))
5318 mod_timer(&adapter->phy_info_timer,
5319 round_jiffies(jiffies + 2 * HZ));
5320
5321 /* 8000ES2LAN requires a Rx packet buffer work-around
5322 * on link down event; reset the controller to flush
5323 * the Rx packet buffer.
5324 */
5325 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
5326 adapter->flags |= FLAG_RESTART_NOW;
5327 else
5328 pm_schedule_suspend(netdev->dev.parent,
5329 LINK_TIMEOUT);
5330 }
5331 }
5332
5333link_up:
5334 spin_lock(&adapter->stats64_lock);
5335 e1000e_update_stats(adapter);
5336
5337 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
5338 adapter->tpt_old = adapter->stats.tpt;
5339 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
5340 adapter->colc_old = adapter->stats.colc;
5341
5342 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
5343 adapter->gorc_old = adapter->stats.gorc;
5344 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
5345 adapter->gotc_old = adapter->stats.gotc;
5346 spin_unlock(&adapter->stats64_lock);
5347
5348 /* If the link is lost the controller stops DMA, but
5349 * if there is queued Tx work it cannot be done. So
5350 * reset the controller to flush the Tx packet buffers.
5351 */
5352 if (!netif_carrier_ok(netdev) &&
5353 (e1000_desc_unused(tx_ring) + 1 < tx_ring->count))
5354 adapter->flags |= FLAG_RESTART_NOW;
5355
5356 /* If reset is necessary, do it outside of interrupt context. */
5357 if (adapter->flags & FLAG_RESTART_NOW) {
5358 schedule_work(&adapter->reset_task);
5359 /* return immediately since reset is imminent */
5360 return;
5361 }
5362
5363 e1000e_update_adaptive(&adapter->hw);
5364
5365 /* Simple mode for Interrupt Throttle Rate (ITR) */
5366 if (adapter->itr_setting == 4) {
5367 /* Symmetric Tx/Rx gets a reduced ITR=2000;
5368 * Total asymmetrical Tx or Rx gets ITR=8000;
5369 * everyone else is between 2000-8000.
5370 */
5371 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
5372 u32 dif = (adapter->gotc > adapter->gorc ?
5373 adapter->gotc - adapter->gorc :
5374 adapter->gorc - adapter->gotc) / 10000;
5375 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
5376
5377 e1000e_write_itr(adapter, itr);
5378 }
5379
5380 /* Cause software interrupt to ensure Rx ring is cleaned */
5381 if (adapter->msix_entries)
5382 ew32(ICS, adapter->rx_ring->ims_val);
5383 else
5384 ew32(ICS, E1000_ICS_RXDMT0);
5385
5386 /* flush pending descriptors to memory before detecting Tx hang */
5387 e1000e_flush_descriptors(adapter);
5388
5389 /* Force detection of hung controller every watchdog period */
5390 adapter->detect_tx_hung = true;
5391
5392 /* With 82571 controllers, LAA may be overwritten due to controller
5393 * reset from the other port. Set the appropriate LAA in RAR[0]
5394 */
5395 if (e1000e_get_laa_state_82571(hw))
5396 hw->mac.ops.rar_set(hw, adapter->hw.mac.addr, 0);
5397
5398 if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
5399 e1000e_check_82574_phy_workaround(adapter);
5400
5401 /* Clear valid timestamp stuck in RXSTMPL/H due to a Rx error */
5402 if (adapter->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE) {
5403 if ((adapter->flags2 & FLAG2_CHECK_RX_HWTSTAMP) &&
5404 (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID)) {
5405 er32(RXSTMPH);
5406 adapter->rx_hwtstamp_cleared++;
5407 } else {
5408 adapter->flags2 |= FLAG2_CHECK_RX_HWTSTAMP;
5409 }
5410 }
5411
5412 /* Reset the timer */
5413 if (!test_bit(__E1000_DOWN, &adapter->state))
5414 mod_timer(&adapter->watchdog_timer,
5415 round_jiffies(jiffies + 2 * HZ));
5416}
5417
5418#define E1000_TX_FLAGS_CSUM 0x00000001
5419#define E1000_TX_FLAGS_VLAN 0x00000002
5420#define E1000_TX_FLAGS_TSO 0x00000004
5421#define E1000_TX_FLAGS_IPV4 0x00000008
5422#define E1000_TX_FLAGS_NO_FCS 0x00000010
5423#define E1000_TX_FLAGS_HWTSTAMP 0x00000020
5424#define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
5425#define E1000_TX_FLAGS_VLAN_SHIFT 16
5426
5427static int e1000_tso(struct e1000_ring *tx_ring, struct sk_buff *skb,
5428 __be16 protocol)
5429{
5430 struct e1000_context_desc *context_desc;
5431 struct e1000_buffer *buffer_info;
5432 unsigned int i;
5433 u32 cmd_length = 0;
5434 u16 ipcse = 0, mss;
5435 u8 ipcss, ipcso, tucss, tucso, hdr_len;
5436 int err;
5437
5438 if (!skb_is_gso(skb))
5439 return 0;
5440
5441 err = skb_cow_head(skb, 0);
5442 if (err < 0)
5443 return err;
5444
5445 hdr_len = skb_tcp_all_headers(skb);
5446 mss = skb_shinfo(skb)->gso_size;
5447 if (protocol == htons(ETH_P_IP)) {
5448 struct iphdr *iph = ip_hdr(skb);
5449 iph->tot_len = 0;
5450 iph->check = 0;
5451 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
5452 0, IPPROTO_TCP, 0);
5453 cmd_length = E1000_TXD_CMD_IP;
5454 ipcse = skb_transport_offset(skb) - 1;
5455 } else if (skb_is_gso_v6(skb)) {
5456 tcp_v6_gso_csum_prep(skb);
5457 ipcse = 0;
5458 }
5459 ipcss = skb_network_offset(skb);
5460 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
5461 tucss = skb_transport_offset(skb);
5462 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
5463
5464 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
5465 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
5466
5467 i = tx_ring->next_to_use;
5468 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5469 buffer_info = &tx_ring->buffer_info[i];
5470
5471 context_desc->lower_setup.ip_fields.ipcss = ipcss;
5472 context_desc->lower_setup.ip_fields.ipcso = ipcso;
5473 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
5474 context_desc->upper_setup.tcp_fields.tucss = tucss;
5475 context_desc->upper_setup.tcp_fields.tucso = tucso;
5476 context_desc->upper_setup.tcp_fields.tucse = 0;
5477 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
5478 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
5479 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
5480
5481 buffer_info->time_stamp = jiffies;
5482 buffer_info->next_to_watch = i;
5483
5484 i++;
5485 if (i == tx_ring->count)
5486 i = 0;
5487 tx_ring->next_to_use = i;
5488
5489 return 1;
5490}
5491
5492static bool e1000_tx_csum(struct e1000_ring *tx_ring, struct sk_buff *skb,
5493 __be16 protocol)
5494{
5495 struct e1000_adapter *adapter = tx_ring->adapter;
5496 struct e1000_context_desc *context_desc;
5497 struct e1000_buffer *buffer_info;
5498 unsigned int i;
5499 u8 css;
5500 u32 cmd_len = E1000_TXD_CMD_DEXT;
5501
5502 if (skb->ip_summed != CHECKSUM_PARTIAL)
5503 return false;
5504
5505 switch (protocol) {
5506 case cpu_to_be16(ETH_P_IP):
5507 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5508 cmd_len |= E1000_TXD_CMD_TCP;
5509 break;
5510 case cpu_to_be16(ETH_P_IPV6):
5511 /* XXX not handling all IPV6 headers */
5512 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5513 cmd_len |= E1000_TXD_CMD_TCP;
5514 break;
5515 default:
5516 if (unlikely(net_ratelimit()))
5517 e_warn("checksum_partial proto=%x!\n",
5518 be16_to_cpu(protocol));
5519 break;
5520 }
5521
5522 css = skb_checksum_start_offset(skb);
5523
5524 i = tx_ring->next_to_use;
5525 buffer_info = &tx_ring->buffer_info[i];
5526 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5527
5528 context_desc->lower_setup.ip_config = 0;
5529 context_desc->upper_setup.tcp_fields.tucss = css;
5530 context_desc->upper_setup.tcp_fields.tucso = css + skb->csum_offset;
5531 context_desc->upper_setup.tcp_fields.tucse = 0;
5532 context_desc->tcp_seg_setup.data = 0;
5533 context_desc->cmd_and_length = cpu_to_le32(cmd_len);
5534
5535 buffer_info->time_stamp = jiffies;
5536 buffer_info->next_to_watch = i;
5537
5538 i++;
5539 if (i == tx_ring->count)
5540 i = 0;
5541 tx_ring->next_to_use = i;
5542
5543 return true;
5544}
5545
5546static int e1000_tx_map(struct e1000_ring *tx_ring, struct sk_buff *skb,
5547 unsigned int first, unsigned int max_per_txd,
5548 unsigned int nr_frags)
5549{
5550 struct e1000_adapter *adapter = tx_ring->adapter;
5551 struct pci_dev *pdev = adapter->pdev;
5552 struct e1000_buffer *buffer_info;
5553 unsigned int len = skb_headlen(skb);
5554 unsigned int offset = 0, size, count = 0, i;
5555 unsigned int f, bytecount, segs;
5556
5557 i = tx_ring->next_to_use;
5558
5559 while (len) {
5560 buffer_info = &tx_ring->buffer_info[i];
5561 size = min(len, max_per_txd);
5562
5563 buffer_info->length = size;
5564 buffer_info->time_stamp = jiffies;
5565 buffer_info->next_to_watch = i;
5566 buffer_info->dma = dma_map_single(&pdev->dev,
5567 skb->data + offset,
5568 size, DMA_TO_DEVICE);
5569 buffer_info->mapped_as_page = false;
5570 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5571 goto dma_error;
5572
5573 len -= size;
5574 offset += size;
5575 count++;
5576
5577 if (len) {
5578 i++;
5579 if (i == tx_ring->count)
5580 i = 0;
5581 }
5582 }
5583
5584 for (f = 0; f < nr_frags; f++) {
5585 const skb_frag_t *frag = &skb_shinfo(skb)->frags[f];
5586
5587 len = skb_frag_size(frag);
5588 offset = 0;
5589
5590 while (len) {
5591 i++;
5592 if (i == tx_ring->count)
5593 i = 0;
5594
5595 buffer_info = &tx_ring->buffer_info[i];
5596 size = min(len, max_per_txd);
5597
5598 buffer_info->length = size;
5599 buffer_info->time_stamp = jiffies;
5600 buffer_info->next_to_watch = i;
5601 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
5602 offset, size,
5603 DMA_TO_DEVICE);
5604 buffer_info->mapped_as_page = true;
5605 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5606 goto dma_error;
5607
5608 len -= size;
5609 offset += size;
5610 count++;
5611 }
5612 }
5613
5614 segs = skb_shinfo(skb)->gso_segs ? : 1;
5615 /* multiply data chunks by size of headers */
5616 bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
5617
5618 tx_ring->buffer_info[i].skb = skb;
5619 tx_ring->buffer_info[i].segs = segs;
5620 tx_ring->buffer_info[i].bytecount = bytecount;
5621 tx_ring->buffer_info[first].next_to_watch = i;
5622
5623 return count;
5624
5625dma_error:
5626 dev_err(&pdev->dev, "Tx DMA map failed\n");
5627 buffer_info->dma = 0;
5628 if (count)
5629 count--;
5630
5631 while (count--) {
5632 if (i == 0)
5633 i += tx_ring->count;
5634 i--;
5635 buffer_info = &tx_ring->buffer_info[i];
5636 e1000_put_txbuf(tx_ring, buffer_info, true);
5637 }
5638
5639 return 0;
5640}
5641
5642static void e1000_tx_queue(struct e1000_ring *tx_ring, int tx_flags, int count)
5643{
5644 struct e1000_adapter *adapter = tx_ring->adapter;
5645 struct e1000_tx_desc *tx_desc = NULL;
5646 struct e1000_buffer *buffer_info;
5647 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
5648 unsigned int i;
5649
5650 if (tx_flags & E1000_TX_FLAGS_TSO) {
5651 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
5652 E1000_TXD_CMD_TSE;
5653 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5654
5655 if (tx_flags & E1000_TX_FLAGS_IPV4)
5656 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
5657 }
5658
5659 if (tx_flags & E1000_TX_FLAGS_CSUM) {
5660 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5661 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5662 }
5663
5664 if (tx_flags & E1000_TX_FLAGS_VLAN) {
5665 txd_lower |= E1000_TXD_CMD_VLE;
5666 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
5667 }
5668
5669 if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5670 txd_lower &= ~(E1000_TXD_CMD_IFCS);
5671
5672 if (unlikely(tx_flags & E1000_TX_FLAGS_HWTSTAMP)) {
5673 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5674 txd_upper |= E1000_TXD_EXTCMD_TSTAMP;
5675 }
5676
5677 i = tx_ring->next_to_use;
5678
5679 do {
5680 buffer_info = &tx_ring->buffer_info[i];
5681 tx_desc = E1000_TX_DESC(*tx_ring, i);
5682 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
5683 tx_desc->lower.data = cpu_to_le32(txd_lower |
5684 buffer_info->length);
5685 tx_desc->upper.data = cpu_to_le32(txd_upper);
5686
5687 i++;
5688 if (i == tx_ring->count)
5689 i = 0;
5690 } while (--count > 0);
5691
5692 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
5693
5694 /* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
5695 if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5696 tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
5697
5698 /* Force memory writes to complete before letting h/w
5699 * know there are new descriptors to fetch. (Only
5700 * applicable for weak-ordered memory model archs,
5701 * such as IA-64).
5702 */
5703 wmb();
5704
5705 tx_ring->next_to_use = i;
5706}
5707
5708#define MINIMUM_DHCP_PACKET_SIZE 282
5709static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
5710 struct sk_buff *skb)
5711{
5712 struct e1000_hw *hw = &adapter->hw;
5713 u16 length, offset;
5714
5715 if (skb_vlan_tag_present(skb) &&
5716 !((skb_vlan_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
5717 (adapter->hw.mng_cookie.status &
5718 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
5719 return 0;
5720
5721 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
5722 return 0;
5723
5724 if (((struct ethhdr *)skb->data)->h_proto != htons(ETH_P_IP))
5725 return 0;
5726
5727 {
5728 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data + 14);
5729 struct udphdr *udp;
5730
5731 if (ip->protocol != IPPROTO_UDP)
5732 return 0;
5733
5734 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
5735 if (ntohs(udp->dest) != 67)
5736 return 0;
5737
5738 offset = (u8 *)udp + 8 - skb->data;
5739 length = skb->len - offset;
5740 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
5741 }
5742
5743 return 0;
5744}
5745
5746static int __e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5747{
5748 struct e1000_adapter *adapter = tx_ring->adapter;
5749
5750 netif_stop_queue(adapter->netdev);
5751 /* Herbert's original patch had:
5752 * smp_mb__after_netif_stop_queue();
5753 * but since that doesn't exist yet, just open code it.
5754 */
5755 smp_mb();
5756
5757 /* We need to check again in a case another CPU has just
5758 * made room available.
5759 */
5760 if (e1000_desc_unused(tx_ring) < size)
5761 return -EBUSY;
5762
5763 /* A reprieve! */
5764 netif_start_queue(adapter->netdev);
5765 ++adapter->restart_queue;
5766 return 0;
5767}
5768
5769static int e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5770{
5771 BUG_ON(size > tx_ring->count);
5772
5773 if (e1000_desc_unused(tx_ring) >= size)
5774 return 0;
5775 return __e1000_maybe_stop_tx(tx_ring, size);
5776}
5777
5778static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
5779 struct net_device *netdev)
5780{
5781 struct e1000_adapter *adapter = netdev_priv(netdev);
5782 struct e1000_ring *tx_ring = adapter->tx_ring;
5783 unsigned int first;
5784 unsigned int tx_flags = 0;
5785 unsigned int len = skb_headlen(skb);
5786 unsigned int nr_frags;
5787 unsigned int mss;
5788 int count = 0;
5789 int tso;
5790 unsigned int f;
5791 __be16 protocol = vlan_get_protocol(skb);
5792
5793 if (test_bit(__E1000_DOWN, &adapter->state)) {
5794 dev_kfree_skb_any(skb);
5795 return NETDEV_TX_OK;
5796 }
5797
5798 if (skb->len <= 0) {
5799 dev_kfree_skb_any(skb);
5800 return NETDEV_TX_OK;
5801 }
5802
5803 /* The minimum packet size with TCTL.PSP set is 17 bytes so
5804 * pad skb in order to meet this minimum size requirement
5805 */
5806 if (skb_put_padto(skb, 17))
5807 return NETDEV_TX_OK;
5808
5809 mss = skb_shinfo(skb)->gso_size;
5810 if (mss) {
5811 u8 hdr_len;
5812
5813 /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
5814 * points to just header, pull a few bytes of payload from
5815 * frags into skb->data
5816 */
5817 hdr_len = skb_tcp_all_headers(skb);
5818 /* we do this workaround for ES2LAN, but it is un-necessary,
5819 * avoiding it could save a lot of cycles
5820 */
5821 if (skb->data_len && (hdr_len == len)) {
5822 unsigned int pull_size;
5823
5824 pull_size = min_t(unsigned int, 4, skb->data_len);
5825 if (!__pskb_pull_tail(skb, pull_size)) {
5826 e_err("__pskb_pull_tail failed.\n");
5827 dev_kfree_skb_any(skb);
5828 return NETDEV_TX_OK;
5829 }
5830 len = skb_headlen(skb);
5831 }
5832 }
5833
5834 /* reserve a descriptor for the offload context */
5835 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5836 count++;
5837 count++;
5838
5839 count += DIV_ROUND_UP(len, adapter->tx_fifo_limit);
5840
5841 nr_frags = skb_shinfo(skb)->nr_frags;
5842 for (f = 0; f < nr_frags; f++)
5843 count += DIV_ROUND_UP(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5844 adapter->tx_fifo_limit);
5845
5846 if (adapter->hw.mac.tx_pkt_filtering)
5847 e1000_transfer_dhcp_info(adapter, skb);
5848
5849 /* need: count + 2 desc gap to keep tail from touching
5850 * head, otherwise try next time
5851 */
5852 if (e1000_maybe_stop_tx(tx_ring, count + 2))
5853 return NETDEV_TX_BUSY;
5854
5855 if (skb_vlan_tag_present(skb)) {
5856 tx_flags |= E1000_TX_FLAGS_VLAN;
5857 tx_flags |= (skb_vlan_tag_get(skb) <<
5858 E1000_TX_FLAGS_VLAN_SHIFT);
5859 }
5860
5861 first = tx_ring->next_to_use;
5862
5863 tso = e1000_tso(tx_ring, skb, protocol);
5864 if (tso < 0) {
5865 dev_kfree_skb_any(skb);
5866 return NETDEV_TX_OK;
5867 }
5868
5869 if (tso)
5870 tx_flags |= E1000_TX_FLAGS_TSO;
5871 else if (e1000_tx_csum(tx_ring, skb, protocol))
5872 tx_flags |= E1000_TX_FLAGS_CSUM;
5873
5874 /* Old method was to assume IPv4 packet by default if TSO was enabled.
5875 * 82571 hardware supports TSO capabilities for IPv6 as well...
5876 * no longer assume, we must.
5877 */
5878 if (protocol == htons(ETH_P_IP))
5879 tx_flags |= E1000_TX_FLAGS_IPV4;
5880
5881 if (unlikely(skb->no_fcs))
5882 tx_flags |= E1000_TX_FLAGS_NO_FCS;
5883
5884 /* if count is 0 then mapping error has occurred */
5885 count = e1000_tx_map(tx_ring, skb, first, adapter->tx_fifo_limit,
5886 nr_frags);
5887 if (count) {
5888 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
5889 (adapter->flags & FLAG_HAS_HW_TIMESTAMP)) {
5890 if (!adapter->tx_hwtstamp_skb) {
5891 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
5892 tx_flags |= E1000_TX_FLAGS_HWTSTAMP;
5893 adapter->tx_hwtstamp_skb = skb_get(skb);
5894 adapter->tx_hwtstamp_start = jiffies;
5895 schedule_work(&adapter->tx_hwtstamp_work);
5896 } else {
5897 adapter->tx_hwtstamp_skipped++;
5898 }
5899 }
5900
5901 skb_tx_timestamp(skb);
5902
5903 netdev_sent_queue(netdev, skb->len);
5904 e1000_tx_queue(tx_ring, tx_flags, count);
5905 /* Make sure there is space in the ring for the next send. */
5906 e1000_maybe_stop_tx(tx_ring,
5907 ((MAX_SKB_FRAGS + 1) *
5908 DIV_ROUND_UP(PAGE_SIZE,
5909 adapter->tx_fifo_limit) + 4));
5910
5911 if (!netdev_xmit_more() ||
5912 netif_xmit_stopped(netdev_get_tx_queue(netdev, 0))) {
5913 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
5914 e1000e_update_tdt_wa(tx_ring,
5915 tx_ring->next_to_use);
5916 else
5917 writel(tx_ring->next_to_use, tx_ring->tail);
5918 }
5919 } else {
5920 dev_kfree_skb_any(skb);
5921 tx_ring->buffer_info[first].time_stamp = 0;
5922 tx_ring->next_to_use = first;
5923 }
5924
5925 return NETDEV_TX_OK;
5926}
5927
5928/**
5929 * e1000_tx_timeout - Respond to a Tx Hang
5930 * @netdev: network interface device structure
5931 * @txqueue: index of the hung queue (unused)
5932 **/
5933static void e1000_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
5934{
5935 struct e1000_adapter *adapter = netdev_priv(netdev);
5936
5937 /* Do the reset outside of interrupt context */
5938 adapter->tx_timeout_count++;
5939 schedule_work(&adapter->reset_task);
5940}
5941
5942static void e1000_reset_task(struct work_struct *work)
5943{
5944 struct e1000_adapter *adapter;
5945 adapter = container_of(work, struct e1000_adapter, reset_task);
5946
5947 rtnl_lock();
5948 /* don't run the task if already down */
5949 if (test_bit(__E1000_DOWN, &adapter->state)) {
5950 rtnl_unlock();
5951 return;
5952 }
5953
5954 if (!(adapter->flags & FLAG_RESTART_NOW)) {
5955 e1000e_dump(adapter);
5956 e_err("Reset adapter unexpectedly\n");
5957 }
5958 e1000e_reinit_locked(adapter);
5959 rtnl_unlock();
5960}
5961
5962/**
5963 * e1000e_get_stats64 - Get System Network Statistics
5964 * @netdev: network interface device structure
5965 * @stats: rtnl_link_stats64 pointer
5966 *
5967 * Returns the address of the device statistics structure.
5968 **/
5969void e1000e_get_stats64(struct net_device *netdev,
5970 struct rtnl_link_stats64 *stats)
5971{
5972 struct e1000_adapter *adapter = netdev_priv(netdev);
5973
5974 spin_lock(&adapter->stats64_lock);
5975 e1000e_update_stats(adapter);
5976 /* Fill out the OS statistics structure */
5977 stats->rx_bytes = adapter->stats.gorc;
5978 stats->rx_packets = adapter->stats.gprc;
5979 stats->tx_bytes = adapter->stats.gotc;
5980 stats->tx_packets = adapter->stats.gptc;
5981 stats->multicast = adapter->stats.mprc;
5982 stats->collisions = adapter->stats.colc;
5983
5984 /* Rx Errors */
5985
5986 /* RLEC on some newer hardware can be incorrect so build
5987 * our own version based on RUC and ROC
5988 */
5989 stats->rx_errors = adapter->stats.rxerrc +
5990 adapter->stats.crcerrs + adapter->stats.algnerrc +
5991 adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
5992 stats->rx_length_errors = adapter->stats.ruc + adapter->stats.roc;
5993 stats->rx_crc_errors = adapter->stats.crcerrs;
5994 stats->rx_frame_errors = adapter->stats.algnerrc;
5995 stats->rx_missed_errors = adapter->stats.mpc;
5996
5997 /* Tx Errors */
5998 stats->tx_errors = adapter->stats.ecol + adapter->stats.latecol;
5999 stats->tx_aborted_errors = adapter->stats.ecol;
6000 stats->tx_window_errors = adapter->stats.latecol;
6001 stats->tx_carrier_errors = adapter->stats.tncrs;
6002
6003 /* Tx Dropped needs to be maintained elsewhere */
6004
6005 spin_unlock(&adapter->stats64_lock);
6006}
6007
6008/**
6009 * e1000_change_mtu - Change the Maximum Transfer Unit
6010 * @netdev: network interface device structure
6011 * @new_mtu: new value for maximum frame size
6012 *
6013 * Returns 0 on success, negative on failure
6014 **/
6015static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
6016{
6017 struct e1000_adapter *adapter = netdev_priv(netdev);
6018 int max_frame = new_mtu + VLAN_ETH_HLEN + ETH_FCS_LEN;
6019
6020 /* Jumbo frame support */
6021 if ((new_mtu > ETH_DATA_LEN) &&
6022 !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
6023 e_err("Jumbo Frames not supported.\n");
6024 return -EINVAL;
6025 }
6026
6027 /* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
6028 if ((adapter->hw.mac.type >= e1000_pch2lan) &&
6029 !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
6030 (new_mtu > ETH_DATA_LEN)) {
6031 e_err("Jumbo Frames not supported on this device when CRC stripping is disabled.\n");
6032 return -EINVAL;
6033 }
6034
6035 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
6036 usleep_range(1000, 1100);
6037 /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
6038 adapter->max_frame_size = max_frame;
6039 netdev_dbg(netdev, "changing MTU from %d to %d\n",
6040 netdev->mtu, new_mtu);
6041 netdev->mtu = new_mtu;
6042
6043 pm_runtime_get_sync(netdev->dev.parent);
6044
6045 if (netif_running(netdev))
6046 e1000e_down(adapter, true);
6047
6048 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
6049 * means we reserve 2 more, this pushes us to allocate from the next
6050 * larger slab size.
6051 * i.e. RXBUFFER_2048 --> size-4096 slab
6052 * However with the new *_jumbo_rx* routines, jumbo receives will use
6053 * fragmented skbs
6054 */
6055
6056 if (max_frame <= 2048)
6057 adapter->rx_buffer_len = 2048;
6058 else
6059 adapter->rx_buffer_len = 4096;
6060
6061 /* adjust allocation if LPE protects us, and we aren't using SBP */
6062 if (max_frame <= (VLAN_ETH_FRAME_LEN + ETH_FCS_LEN))
6063 adapter->rx_buffer_len = VLAN_ETH_FRAME_LEN + ETH_FCS_LEN;
6064
6065 if (netif_running(netdev))
6066 e1000e_up(adapter);
6067 else
6068 e1000e_reset(adapter);
6069
6070 pm_runtime_put_sync(netdev->dev.parent);
6071
6072 clear_bit(__E1000_RESETTING, &adapter->state);
6073
6074 return 0;
6075}
6076
6077static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
6078 int cmd)
6079{
6080 struct e1000_adapter *adapter = netdev_priv(netdev);
6081 struct mii_ioctl_data *data = if_mii(ifr);
6082
6083 if (adapter->hw.phy.media_type != e1000_media_type_copper)
6084 return -EOPNOTSUPP;
6085
6086 switch (cmd) {
6087 case SIOCGMIIPHY:
6088 data->phy_id = adapter->hw.phy.addr;
6089 break;
6090 case SIOCGMIIREG:
6091 e1000_phy_read_status(adapter);
6092
6093 switch (data->reg_num & 0x1F) {
6094 case MII_BMCR:
6095 data->val_out = adapter->phy_regs.bmcr;
6096 break;
6097 case MII_BMSR:
6098 data->val_out = adapter->phy_regs.bmsr;
6099 break;
6100 case MII_PHYSID1:
6101 data->val_out = (adapter->hw.phy.id >> 16);
6102 break;
6103 case MII_PHYSID2:
6104 data->val_out = (adapter->hw.phy.id & 0xFFFF);
6105 break;
6106 case MII_ADVERTISE:
6107 data->val_out = adapter->phy_regs.advertise;
6108 break;
6109 case MII_LPA:
6110 data->val_out = adapter->phy_regs.lpa;
6111 break;
6112 case MII_EXPANSION:
6113 data->val_out = adapter->phy_regs.expansion;
6114 break;
6115 case MII_CTRL1000:
6116 data->val_out = adapter->phy_regs.ctrl1000;
6117 break;
6118 case MII_STAT1000:
6119 data->val_out = adapter->phy_regs.stat1000;
6120 break;
6121 case MII_ESTATUS:
6122 data->val_out = adapter->phy_regs.estatus;
6123 break;
6124 default:
6125 return -EIO;
6126 }
6127 break;
6128 case SIOCSMIIREG:
6129 default:
6130 return -EOPNOTSUPP;
6131 }
6132 return 0;
6133}
6134
6135/**
6136 * e1000e_hwtstamp_set - control hardware time stamping
6137 * @netdev: network interface device structure
6138 * @ifr: interface request
6139 *
6140 * Outgoing time stamping can be enabled and disabled. Play nice and
6141 * disable it when requested, although it shouldn't cause any overhead
6142 * when no packet needs it. At most one packet in the queue may be
6143 * marked for time stamping, otherwise it would be impossible to tell
6144 * for sure to which packet the hardware time stamp belongs.
6145 *
6146 * Incoming time stamping has to be configured via the hardware filters.
6147 * Not all combinations are supported, in particular event type has to be
6148 * specified. Matching the kind of event packet is not supported, with the
6149 * exception of "all V2 events regardless of level 2 or 4".
6150 **/
6151static int e1000e_hwtstamp_set(struct net_device *netdev, struct ifreq *ifr)
6152{
6153 struct e1000_adapter *adapter = netdev_priv(netdev);
6154 struct hwtstamp_config config;
6155 int ret_val;
6156
6157 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
6158 return -EFAULT;
6159
6160 ret_val = e1000e_config_hwtstamp(adapter, &config);
6161 if (ret_val)
6162 return ret_val;
6163
6164 switch (config.rx_filter) {
6165 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
6166 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
6167 case HWTSTAMP_FILTER_PTP_V2_SYNC:
6168 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
6169 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
6170 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
6171 /* With V2 type filters which specify a Sync or Delay Request,
6172 * Path Delay Request/Response messages are also time stamped
6173 * by hardware so notify the caller the requested packets plus
6174 * some others are time stamped.
6175 */
6176 config.rx_filter = HWTSTAMP_FILTER_SOME;
6177 break;
6178 default:
6179 break;
6180 }
6181
6182 return copy_to_user(ifr->ifr_data, &config,
6183 sizeof(config)) ? -EFAULT : 0;
6184}
6185
6186static int e1000e_hwtstamp_get(struct net_device *netdev, struct ifreq *ifr)
6187{
6188 struct e1000_adapter *adapter = netdev_priv(netdev);
6189
6190 return copy_to_user(ifr->ifr_data, &adapter->hwtstamp_config,
6191 sizeof(adapter->hwtstamp_config)) ? -EFAULT : 0;
6192}
6193
6194static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6195{
6196 switch (cmd) {
6197 case SIOCGMIIPHY:
6198 case SIOCGMIIREG:
6199 case SIOCSMIIREG:
6200 return e1000_mii_ioctl(netdev, ifr, cmd);
6201 case SIOCSHWTSTAMP:
6202 return e1000e_hwtstamp_set(netdev, ifr);
6203 case SIOCGHWTSTAMP:
6204 return e1000e_hwtstamp_get(netdev, ifr);
6205 default:
6206 return -EOPNOTSUPP;
6207 }
6208}
6209
6210static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
6211{
6212 struct e1000_hw *hw = &adapter->hw;
6213 u32 i, mac_reg, wuc;
6214 u16 phy_reg, wuc_enable;
6215 int retval;
6216
6217 /* copy MAC RARs to PHY RARs */
6218 e1000_copy_rx_addrs_to_phy_ich8lan(hw);
6219
6220 retval = hw->phy.ops.acquire(hw);
6221 if (retval) {
6222 e_err("Could not acquire PHY\n");
6223 return retval;
6224 }
6225
6226 /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
6227 retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
6228 if (retval)
6229 goto release;
6230
6231 /* copy MAC MTA to PHY MTA - only needed for pchlan */
6232 for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
6233 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
6234 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
6235 (u16)(mac_reg & 0xFFFF));
6236 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
6237 (u16)((mac_reg >> 16) & 0xFFFF));
6238 }
6239
6240 /* configure PHY Rx Control register */
6241 hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
6242 mac_reg = er32(RCTL);
6243 if (mac_reg & E1000_RCTL_UPE)
6244 phy_reg |= BM_RCTL_UPE;
6245 if (mac_reg & E1000_RCTL_MPE)
6246 phy_reg |= BM_RCTL_MPE;
6247 phy_reg &= ~(BM_RCTL_MO_MASK);
6248 if (mac_reg & E1000_RCTL_MO_3)
6249 phy_reg |= (FIELD_GET(E1000_RCTL_MO_3, mac_reg)
6250 << BM_RCTL_MO_SHIFT);
6251 if (mac_reg & E1000_RCTL_BAM)
6252 phy_reg |= BM_RCTL_BAM;
6253 if (mac_reg & E1000_RCTL_PMCF)
6254 phy_reg |= BM_RCTL_PMCF;
6255 mac_reg = er32(CTRL);
6256 if (mac_reg & E1000_CTRL_RFCE)
6257 phy_reg |= BM_RCTL_RFCE;
6258 hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
6259
6260 wuc = E1000_WUC_PME_EN;
6261 if (wufc & (E1000_WUFC_MAG | E1000_WUFC_LNKC))
6262 wuc |= E1000_WUC_APME;
6263
6264 /* enable PHY wakeup in MAC register */
6265 ew32(WUFC, wufc);
6266 ew32(WUC, (E1000_WUC_PHY_WAKE | E1000_WUC_APMPME |
6267 E1000_WUC_PME_STATUS | wuc));
6268
6269 /* configure and enable PHY wakeup in PHY registers */
6270 hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
6271 hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, wuc);
6272
6273 /* activate PHY wakeup */
6274 wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
6275 retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
6276 if (retval)
6277 e_err("Could not set PHY Host Wakeup bit\n");
6278release:
6279 hw->phy.ops.release(hw);
6280
6281 return retval;
6282}
6283
6284static void e1000e_flush_lpic(struct pci_dev *pdev)
6285{
6286 struct net_device *netdev = pci_get_drvdata(pdev);
6287 struct e1000_adapter *adapter = netdev_priv(netdev);
6288 struct e1000_hw *hw = &adapter->hw;
6289 u32 ret_val;
6290
6291 pm_runtime_get_sync(netdev->dev.parent);
6292
6293 ret_val = hw->phy.ops.acquire(hw);
6294 if (ret_val)
6295 goto fl_out;
6296
6297 pr_info("EEE TX LPI TIMER: %08X\n",
6298 er32(LPIC) >> E1000_LPIC_LPIET_SHIFT);
6299
6300 hw->phy.ops.release(hw);
6301
6302fl_out:
6303 pm_runtime_put_sync(netdev->dev.parent);
6304}
6305
6306/* S0ix implementation */
6307static void e1000e_s0ix_entry_flow(struct e1000_adapter *adapter)
6308{
6309 struct e1000_hw *hw = &adapter->hw;
6310 u32 mac_data;
6311 u16 phy_data;
6312
6313 if (er32(FWSM) & E1000_ICH_FWSM_FW_VALID &&
6314 hw->mac.type >= e1000_pch_adp) {
6315 /* Request ME configure the device for S0ix */
6316 mac_data = er32(H2ME);
6317 mac_data |= E1000_H2ME_START_DPG;
6318 mac_data &= ~E1000_H2ME_EXIT_DPG;
6319 trace_e1000e_trace_mac_register(mac_data);
6320 ew32(H2ME, mac_data);
6321 } else {
6322 /* Request driver configure the device to S0ix */
6323 /* Disable the periodic inband message,
6324 * don't request PCIe clock in K1 page770_17[10:9] = 10b
6325 */
6326 e1e_rphy(hw, HV_PM_CTRL, &phy_data);
6327 phy_data &= ~HV_PM_CTRL_K1_CLK_REQ;
6328 phy_data |= BIT(10);
6329 e1e_wphy(hw, HV_PM_CTRL, phy_data);
6330
6331 /* Make sure we don't exit K1 every time a new packet arrives
6332 * 772_29[5] = 1 CS_Mode_Stay_In_K1
6333 */
6334 e1e_rphy(hw, I217_CGFREG, &phy_data);
6335 phy_data |= BIT(5);
6336 e1e_wphy(hw, I217_CGFREG, phy_data);
6337
6338 /* Change the MAC/PHY interface to SMBus
6339 * Force the SMBus in PHY page769_23[0] = 1
6340 * Force the SMBus in MAC CTRL_EXT[11] = 1
6341 */
6342 e1e_rphy(hw, CV_SMB_CTRL, &phy_data);
6343 phy_data |= CV_SMB_CTRL_FORCE_SMBUS;
6344 e1e_wphy(hw, CV_SMB_CTRL, phy_data);
6345 mac_data = er32(CTRL_EXT);
6346 mac_data |= E1000_CTRL_EXT_FORCE_SMBUS;
6347 ew32(CTRL_EXT, mac_data);
6348
6349 /* DFT control: PHY bit: page769_20[0] = 1
6350 * page769_20[7] - PHY PLL stop
6351 * page769_20[8] - PHY go to the electrical idle
6352 * page769_20[9] - PHY serdes disable
6353 * Gate PPW via EXTCNF_CTRL - set 0x0F00[7] = 1
6354 */
6355 e1e_rphy(hw, I82579_DFT_CTRL, &phy_data);
6356 phy_data |= BIT(0);
6357 phy_data |= BIT(7);
6358 phy_data |= BIT(8);
6359 phy_data |= BIT(9);
6360 e1e_wphy(hw, I82579_DFT_CTRL, phy_data);
6361
6362 mac_data = er32(EXTCNF_CTRL);
6363 mac_data |= E1000_EXTCNF_CTRL_GATE_PHY_CFG;
6364 ew32(EXTCNF_CTRL, mac_data);
6365
6366 /* Enable the Dynamic Power Gating in the MAC */
6367 mac_data = er32(FEXTNVM7);
6368 mac_data |= BIT(22);
6369 ew32(FEXTNVM7, mac_data);
6370
6371 /* Disable disconnected cable conditioning for Power Gating */
6372 mac_data = er32(DPGFR);
6373 mac_data |= BIT(2);
6374 ew32(DPGFR, mac_data);
6375
6376 /* Don't wake from dynamic Power Gating with clock request */
6377 mac_data = er32(FEXTNVM12);
6378 mac_data |= BIT(12);
6379 ew32(FEXTNVM12, mac_data);
6380
6381 /* Ungate PGCB clock */
6382 mac_data = er32(FEXTNVM9);
6383 mac_data &= ~BIT(28);
6384 ew32(FEXTNVM9, mac_data);
6385
6386 /* Enable K1 off to enable mPHY Power Gating */
6387 mac_data = er32(FEXTNVM6);
6388 mac_data |= BIT(31);
6389 ew32(FEXTNVM6, mac_data);
6390
6391 /* Enable mPHY power gating for any link and speed */
6392 mac_data = er32(FEXTNVM8);
6393 mac_data |= BIT(9);
6394 ew32(FEXTNVM8, mac_data);
6395
6396 /* Enable the Dynamic Clock Gating in the DMA and MAC */
6397 mac_data = er32(CTRL_EXT);
6398 mac_data |= E1000_CTRL_EXT_DMA_DYN_CLK_EN;
6399 ew32(CTRL_EXT, mac_data);
6400
6401 /* No MAC DPG gating SLP_S0 in modern standby
6402 * Switch the logic of the lanphypc to use PMC counter
6403 */
6404 mac_data = er32(FEXTNVM5);
6405 mac_data |= BIT(7);
6406 ew32(FEXTNVM5, mac_data);
6407 }
6408
6409 /* Disable the time synchronization clock */
6410 mac_data = er32(FEXTNVM7);
6411 mac_data |= BIT(31);
6412 mac_data &= ~BIT(0);
6413 ew32(FEXTNVM7, mac_data);
6414
6415 /* Dynamic Power Gating Enable */
6416 mac_data = er32(CTRL_EXT);
6417 mac_data |= BIT(3);
6418 ew32(CTRL_EXT, mac_data);
6419
6420 /* Check MAC Tx/Rx packet buffer pointers.
6421 * Reset MAC Tx/Rx packet buffer pointers to suppress any
6422 * pending traffic indication that would prevent power gating.
6423 */
6424 mac_data = er32(TDFH);
6425 if (mac_data)
6426 ew32(TDFH, 0);
6427 mac_data = er32(TDFT);
6428 if (mac_data)
6429 ew32(TDFT, 0);
6430 mac_data = er32(TDFHS);
6431 if (mac_data)
6432 ew32(TDFHS, 0);
6433 mac_data = er32(TDFTS);
6434 if (mac_data)
6435 ew32(TDFTS, 0);
6436 mac_data = er32(TDFPC);
6437 if (mac_data)
6438 ew32(TDFPC, 0);
6439 mac_data = er32(RDFH);
6440 if (mac_data)
6441 ew32(RDFH, 0);
6442 mac_data = er32(RDFT);
6443 if (mac_data)
6444 ew32(RDFT, 0);
6445 mac_data = er32(RDFHS);
6446 if (mac_data)
6447 ew32(RDFHS, 0);
6448 mac_data = er32(RDFTS);
6449 if (mac_data)
6450 ew32(RDFTS, 0);
6451 mac_data = er32(RDFPC);
6452 if (mac_data)
6453 ew32(RDFPC, 0);
6454}
6455
6456static void e1000e_s0ix_exit_flow(struct e1000_adapter *adapter)
6457{
6458 struct e1000_hw *hw = &adapter->hw;
6459 bool firmware_bug = false;
6460 u32 mac_data;
6461 u16 phy_data;
6462 u32 i = 0;
6463
6464 if (er32(FWSM) & E1000_ICH_FWSM_FW_VALID &&
6465 hw->mac.type >= e1000_pch_adp) {
6466 /* Keep the GPT clock enabled for CSME */
6467 mac_data = er32(FEXTNVM);
6468 mac_data |= BIT(3);
6469 ew32(FEXTNVM, mac_data);
6470 /* Request ME unconfigure the device from S0ix */
6471 mac_data = er32(H2ME);
6472 mac_data &= ~E1000_H2ME_START_DPG;
6473 mac_data |= E1000_H2ME_EXIT_DPG;
6474 trace_e1000e_trace_mac_register(mac_data);
6475 ew32(H2ME, mac_data);
6476
6477 /* Poll up to 2.5 seconds for ME to unconfigure DPG.
6478 * If this takes more than 1 second, show a warning indicating a
6479 * firmware bug
6480 */
6481 while (!(er32(EXFWSM) & E1000_EXFWSM_DPG_EXIT_DONE)) {
6482 if (i > 100 && !firmware_bug)
6483 firmware_bug = true;
6484
6485 if (i++ == 250) {
6486 e_dbg("Timeout (firmware bug): %d msec\n",
6487 i * 10);
6488 break;
6489 }
6490
6491 usleep_range(10000, 11000);
6492 }
6493 if (firmware_bug)
6494 e_warn("DPG_EXIT_DONE took %d msec. This is a firmware bug\n",
6495 i * 10);
6496 else
6497 e_dbg("DPG_EXIT_DONE cleared after %d msec\n", i * 10);
6498 } else {
6499 /* Request driver unconfigure the device from S0ix */
6500
6501 /* Disable the Dynamic Power Gating in the MAC */
6502 mac_data = er32(FEXTNVM7);
6503 mac_data &= 0xFFBFFFFF;
6504 ew32(FEXTNVM7, mac_data);
6505
6506 /* Disable mPHY power gating for any link and speed */
6507 mac_data = er32(FEXTNVM8);
6508 mac_data &= ~BIT(9);
6509 ew32(FEXTNVM8, mac_data);
6510
6511 /* Disable K1 off */
6512 mac_data = er32(FEXTNVM6);
6513 mac_data &= ~BIT(31);
6514 ew32(FEXTNVM6, mac_data);
6515
6516 /* Disable Ungate PGCB clock */
6517 mac_data = er32(FEXTNVM9);
6518 mac_data |= BIT(28);
6519 ew32(FEXTNVM9, mac_data);
6520
6521 /* Cancel not waking from dynamic
6522 * Power Gating with clock request
6523 */
6524 mac_data = er32(FEXTNVM12);
6525 mac_data &= ~BIT(12);
6526 ew32(FEXTNVM12, mac_data);
6527
6528 /* Cancel disable disconnected cable conditioning
6529 * for Power Gating
6530 */
6531 mac_data = er32(DPGFR);
6532 mac_data &= ~BIT(2);
6533 ew32(DPGFR, mac_data);
6534
6535 /* Disable the Dynamic Clock Gating in the DMA and MAC */
6536 mac_data = er32(CTRL_EXT);
6537 mac_data &= 0xFFF7FFFF;
6538 ew32(CTRL_EXT, mac_data);
6539
6540 /* Revert the lanphypc logic to use the internal Gbe counter
6541 * and not the PMC counter
6542 */
6543 mac_data = er32(FEXTNVM5);
6544 mac_data &= 0xFFFFFF7F;
6545 ew32(FEXTNVM5, mac_data);
6546
6547 /* Enable the periodic inband message,
6548 * Request PCIe clock in K1 page770_17[10:9] =01b
6549 */
6550 e1e_rphy(hw, HV_PM_CTRL, &phy_data);
6551 phy_data &= 0xFBFF;
6552 phy_data |= HV_PM_CTRL_K1_CLK_REQ;
6553 e1e_wphy(hw, HV_PM_CTRL, phy_data);
6554
6555 /* Return back configuration
6556 * 772_29[5] = 0 CS_Mode_Stay_In_K1
6557 */
6558 e1e_rphy(hw, I217_CGFREG, &phy_data);
6559 phy_data &= 0xFFDF;
6560 e1e_wphy(hw, I217_CGFREG, phy_data);
6561
6562 /* Change the MAC/PHY interface to Kumeran
6563 * Unforce the SMBus in PHY page769_23[0] = 0
6564 * Unforce the SMBus in MAC CTRL_EXT[11] = 0
6565 */
6566 e1e_rphy(hw, CV_SMB_CTRL, &phy_data);
6567 phy_data &= ~CV_SMB_CTRL_FORCE_SMBUS;
6568 e1e_wphy(hw, CV_SMB_CTRL, phy_data);
6569 mac_data = er32(CTRL_EXT);
6570 mac_data &= ~E1000_CTRL_EXT_FORCE_SMBUS;
6571 ew32(CTRL_EXT, mac_data);
6572 }
6573
6574 /* Disable Dynamic Power Gating */
6575 mac_data = er32(CTRL_EXT);
6576 mac_data &= 0xFFFFFFF7;
6577 ew32(CTRL_EXT, mac_data);
6578
6579 /* Enable the time synchronization clock */
6580 mac_data = er32(FEXTNVM7);
6581 mac_data &= ~BIT(31);
6582 mac_data |= BIT(0);
6583 ew32(FEXTNVM7, mac_data);
6584}
6585
6586static int e1000e_pm_freeze(struct device *dev)
6587{
6588 struct net_device *netdev = dev_get_drvdata(dev);
6589 struct e1000_adapter *adapter = netdev_priv(netdev);
6590 bool present;
6591
6592 rtnl_lock();
6593
6594 present = netif_device_present(netdev);
6595 netif_device_detach(netdev);
6596
6597 if (present && netif_running(netdev)) {
6598 int count = E1000_CHECK_RESET_COUNT;
6599
6600 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
6601 usleep_range(10000, 11000);
6602
6603 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
6604
6605 /* Quiesce the device without resetting the hardware */
6606 e1000e_down(adapter, false);
6607 e1000_free_irq(adapter);
6608 }
6609 rtnl_unlock();
6610
6611 e1000e_reset_interrupt_capability(adapter);
6612
6613 /* Allow time for pending master requests to run */
6614 e1000e_disable_pcie_master(&adapter->hw);
6615
6616 return 0;
6617}
6618
6619static int __e1000_shutdown(struct pci_dev *pdev, bool runtime)
6620{
6621 struct net_device *netdev = pci_get_drvdata(pdev);
6622 struct e1000_adapter *adapter = netdev_priv(netdev);
6623 struct e1000_hw *hw = &adapter->hw;
6624 u32 ctrl, ctrl_ext, rctl, status, wufc;
6625 int retval = 0;
6626 u16 smb_ctrl;
6627
6628 /* Runtime suspend should only enable wakeup for link changes */
6629 if (runtime)
6630 wufc = E1000_WUFC_LNKC;
6631 else if (device_may_wakeup(&pdev->dev))
6632 wufc = adapter->wol;
6633 else
6634 wufc = 0;
6635
6636 status = er32(STATUS);
6637 if (status & E1000_STATUS_LU)
6638 wufc &= ~E1000_WUFC_LNKC;
6639
6640 if (wufc) {
6641 e1000_setup_rctl(adapter);
6642 e1000e_set_rx_mode(netdev);
6643
6644 /* turn on all-multi mode if wake on multicast is enabled */
6645 if (wufc & E1000_WUFC_MC) {
6646 rctl = er32(RCTL);
6647 rctl |= E1000_RCTL_MPE;
6648 ew32(RCTL, rctl);
6649 }
6650
6651 ctrl = er32(CTRL);
6652 ctrl |= E1000_CTRL_ADVD3WUC;
6653 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
6654 ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
6655 ew32(CTRL, ctrl);
6656
6657 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
6658 adapter->hw.phy.media_type ==
6659 e1000_media_type_internal_serdes) {
6660 /* keep the laser running in D3 */
6661 ctrl_ext = er32(CTRL_EXT);
6662 ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
6663 ew32(CTRL_EXT, ctrl_ext);
6664 }
6665
6666 if (!runtime)
6667 e1000e_power_up_phy(adapter);
6668
6669 if (adapter->flags & FLAG_IS_ICH)
6670 e1000_suspend_workarounds_ich8lan(&adapter->hw);
6671
6672 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
6673 /* enable wakeup by the PHY */
6674 retval = e1000_init_phy_wakeup(adapter, wufc);
6675 if (retval)
6676 return retval;
6677 } else {
6678 /* enable wakeup by the MAC */
6679 ew32(WUFC, wufc);
6680 ew32(WUC, E1000_WUC_PME_EN);
6681 }
6682 } else {
6683 ew32(WUC, 0);
6684 ew32(WUFC, 0);
6685
6686 e1000_power_down_phy(adapter);
6687 }
6688
6689 if (adapter->hw.phy.type == e1000_phy_igp_3) {
6690 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
6691 } else if (hw->mac.type >= e1000_pch_lpt) {
6692 if (wufc && !(wufc & (E1000_WUFC_EX | E1000_WUFC_MC | E1000_WUFC_BC))) {
6693 /* ULP does not support wake from unicast, multicast
6694 * or broadcast.
6695 */
6696 retval = e1000_enable_ulp_lpt_lp(hw, !runtime);
6697 if (retval)
6698 return retval;
6699 }
6700
6701 /* Force SMBUS to allow WOL */
6702 /* Switching PHY interface always returns MDI error
6703 * so disable retry mechanism to avoid wasting time
6704 */
6705 e1000e_disable_phy_retry(hw);
6706
6707 e1e_rphy(hw, CV_SMB_CTRL, &smb_ctrl);
6708 smb_ctrl |= CV_SMB_CTRL_FORCE_SMBUS;
6709 e1e_wphy(hw, CV_SMB_CTRL, smb_ctrl);
6710
6711 e1000e_enable_phy_retry(hw);
6712
6713 /* Force SMBus mode in MAC */
6714 ctrl_ext = er32(CTRL_EXT);
6715 ctrl_ext |= E1000_CTRL_EXT_FORCE_SMBUS;
6716 ew32(CTRL_EXT, ctrl_ext);
6717 }
6718
6719 /* Ensure that the appropriate bits are set in LPI_CTRL
6720 * for EEE in Sx
6721 */
6722 if ((hw->phy.type >= e1000_phy_i217) &&
6723 adapter->eee_advert && hw->dev_spec.ich8lan.eee_lp_ability) {
6724 u16 lpi_ctrl = 0;
6725
6726 retval = hw->phy.ops.acquire(hw);
6727 if (!retval) {
6728 retval = e1e_rphy_locked(hw, I82579_LPI_CTRL,
6729 &lpi_ctrl);
6730 if (!retval) {
6731 if (adapter->eee_advert &
6732 hw->dev_spec.ich8lan.eee_lp_ability &
6733 I82579_EEE_100_SUPPORTED)
6734 lpi_ctrl |= I82579_LPI_CTRL_100_ENABLE;
6735 if (adapter->eee_advert &
6736 hw->dev_spec.ich8lan.eee_lp_ability &
6737 I82579_EEE_1000_SUPPORTED)
6738 lpi_ctrl |= I82579_LPI_CTRL_1000_ENABLE;
6739
6740 retval = e1e_wphy_locked(hw, I82579_LPI_CTRL,
6741 lpi_ctrl);
6742 }
6743 }
6744 hw->phy.ops.release(hw);
6745 }
6746
6747 /* Release control of h/w to f/w. If f/w is AMT enabled, this
6748 * would have already happened in close and is redundant.
6749 */
6750 e1000e_release_hw_control(adapter);
6751
6752 pci_clear_master(pdev);
6753
6754 /* The pci-e switch on some quad port adapters will report a
6755 * correctable error when the MAC transitions from D0 to D3. To
6756 * prevent this we need to mask off the correctable errors on the
6757 * downstream port of the pci-e switch.
6758 *
6759 * We don't have the associated upstream bridge while assigning
6760 * the PCI device into guest. For example, the KVM on power is
6761 * one of the cases.
6762 */
6763 if (adapter->flags & FLAG_IS_QUAD_PORT) {
6764 struct pci_dev *us_dev = pdev->bus->self;
6765 u16 devctl;
6766
6767 if (!us_dev)
6768 return 0;
6769
6770 pcie_capability_read_word(us_dev, PCI_EXP_DEVCTL, &devctl);
6771 pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL,
6772 (devctl & ~PCI_EXP_DEVCTL_CERE));
6773
6774 pci_save_state(pdev);
6775 pci_prepare_to_sleep(pdev);
6776
6777 pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL, devctl);
6778 }
6779
6780 return 0;
6781}
6782
6783/**
6784 * __e1000e_disable_aspm - Disable ASPM states
6785 * @pdev: pointer to PCI device struct
6786 * @state: bit-mask of ASPM states to disable
6787 * @locked: indication if this context holds pci_bus_sem locked.
6788 *
6789 * Some devices *must* have certain ASPM states disabled per hardware errata.
6790 **/
6791static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state, int locked)
6792{
6793 struct pci_dev *parent = pdev->bus->self;
6794 u16 aspm_dis_mask = 0;
6795 u16 pdev_aspmc, parent_aspmc;
6796
6797 switch (state) {
6798 case PCIE_LINK_STATE_L0S:
6799 case PCIE_LINK_STATE_L0S | PCIE_LINK_STATE_L1:
6800 aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L0S;
6801 fallthrough; /* can't have L1 without L0s */
6802 case PCIE_LINK_STATE_L1:
6803 aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L1;
6804 break;
6805 default:
6806 return;
6807 }
6808
6809 pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6810 pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6811
6812 if (parent) {
6813 pcie_capability_read_word(parent, PCI_EXP_LNKCTL,
6814 &parent_aspmc);
6815 parent_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6816 }
6817
6818 /* Nothing to do if the ASPM states to be disabled already are */
6819 if (!(pdev_aspmc & aspm_dis_mask) &&
6820 (!parent || !(parent_aspmc & aspm_dis_mask)))
6821 return;
6822
6823 dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
6824 (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L0S) ?
6825 "L0s" : "",
6826 (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L1) ?
6827 "L1" : "");
6828
6829#ifdef CONFIG_PCIEASPM
6830 if (locked)
6831 pci_disable_link_state_locked(pdev, state);
6832 else
6833 pci_disable_link_state(pdev, state);
6834
6835 /* Double-check ASPM control. If not disabled by the above, the
6836 * BIOS is preventing that from happening (or CONFIG_PCIEASPM is
6837 * not enabled); override by writing PCI config space directly.
6838 */
6839 pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6840 pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6841
6842 if (!(aspm_dis_mask & pdev_aspmc))
6843 return;
6844#endif
6845
6846 /* Both device and parent should have the same ASPM setting.
6847 * Disable ASPM in downstream component first and then upstream.
6848 */
6849 pcie_capability_clear_word(pdev, PCI_EXP_LNKCTL, aspm_dis_mask);
6850
6851 if (parent)
6852 pcie_capability_clear_word(parent, PCI_EXP_LNKCTL,
6853 aspm_dis_mask);
6854}
6855
6856/**
6857 * e1000e_disable_aspm - Disable ASPM states.
6858 * @pdev: pointer to PCI device struct
6859 * @state: bit-mask of ASPM states to disable
6860 *
6861 * This function acquires the pci_bus_sem!
6862 * Some devices *must* have certain ASPM states disabled per hardware errata.
6863 **/
6864static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
6865{
6866 __e1000e_disable_aspm(pdev, state, 0);
6867}
6868
6869/**
6870 * e1000e_disable_aspm_locked - Disable ASPM states.
6871 * @pdev: pointer to PCI device struct
6872 * @state: bit-mask of ASPM states to disable
6873 *
6874 * This function must be called with pci_bus_sem acquired!
6875 * Some devices *must* have certain ASPM states disabled per hardware errata.
6876 **/
6877static void e1000e_disable_aspm_locked(struct pci_dev *pdev, u16 state)
6878{
6879 __e1000e_disable_aspm(pdev, state, 1);
6880}
6881
6882static int e1000e_pm_thaw(struct device *dev)
6883{
6884 struct net_device *netdev = dev_get_drvdata(dev);
6885 struct e1000_adapter *adapter = netdev_priv(netdev);
6886 int rc = 0;
6887
6888 e1000e_set_interrupt_capability(adapter);
6889
6890 rtnl_lock();
6891 if (netif_running(netdev)) {
6892 rc = e1000_request_irq(adapter);
6893 if (rc)
6894 goto err_irq;
6895
6896 e1000e_up(adapter);
6897 }
6898
6899 netif_device_attach(netdev);
6900err_irq:
6901 rtnl_unlock();
6902
6903 return rc;
6904}
6905
6906static int __e1000_resume(struct pci_dev *pdev)
6907{
6908 struct net_device *netdev = pci_get_drvdata(pdev);
6909 struct e1000_adapter *adapter = netdev_priv(netdev);
6910 struct e1000_hw *hw = &adapter->hw;
6911 u16 aspm_disable_flag = 0;
6912
6913 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
6914 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6915 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
6916 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6917 if (aspm_disable_flag)
6918 e1000e_disable_aspm(pdev, aspm_disable_flag);
6919
6920 pci_set_master(pdev);
6921
6922 if (hw->mac.type >= e1000_pch2lan)
6923 e1000_resume_workarounds_pchlan(&adapter->hw);
6924
6925 e1000e_power_up_phy(adapter);
6926
6927 /* report the system wakeup cause from S3/S4 */
6928 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
6929 u16 phy_data;
6930
6931 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
6932 if (phy_data) {
6933 e_info("PHY Wakeup cause - %s\n",
6934 phy_data & E1000_WUS_EX ? "Unicast Packet" :
6935 phy_data & E1000_WUS_MC ? "Multicast Packet" :
6936 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
6937 phy_data & E1000_WUS_MAG ? "Magic Packet" :
6938 phy_data & E1000_WUS_LNKC ?
6939 "Link Status Change" : "other");
6940 }
6941 e1e_wphy(&adapter->hw, BM_WUS, ~0);
6942 } else {
6943 u32 wus = er32(WUS);
6944
6945 if (wus) {
6946 e_info("MAC Wakeup cause - %s\n",
6947 wus & E1000_WUS_EX ? "Unicast Packet" :
6948 wus & E1000_WUS_MC ? "Multicast Packet" :
6949 wus & E1000_WUS_BC ? "Broadcast Packet" :
6950 wus & E1000_WUS_MAG ? "Magic Packet" :
6951 wus & E1000_WUS_LNKC ? "Link Status Change" :
6952 "other");
6953 }
6954 ew32(WUS, ~0);
6955 }
6956
6957 e1000e_reset(adapter);
6958
6959 e1000_init_manageability_pt(adapter);
6960
6961 /* If the controller has AMT, do not set DRV_LOAD until the interface
6962 * is up. For all other cases, let the f/w know that the h/w is now
6963 * under the control of the driver.
6964 */
6965 if (!(adapter->flags & FLAG_HAS_AMT))
6966 e1000e_get_hw_control(adapter);
6967
6968 return 0;
6969}
6970
6971static __maybe_unused int e1000e_pm_prepare(struct device *dev)
6972{
6973 return pm_runtime_suspended(dev) &&
6974 pm_suspend_via_firmware();
6975}
6976
6977static __maybe_unused int e1000e_pm_suspend(struct device *dev)
6978{
6979 struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
6980 struct e1000_adapter *adapter = netdev_priv(netdev);
6981 struct pci_dev *pdev = to_pci_dev(dev);
6982 int rc;
6983
6984 e1000e_flush_lpic(pdev);
6985
6986 e1000e_pm_freeze(dev);
6987
6988 rc = __e1000_shutdown(pdev, false);
6989 if (rc) {
6990 e1000e_pm_thaw(dev);
6991 } else {
6992 /* Introduce S0ix implementation */
6993 if (adapter->flags2 & FLAG2_ENABLE_S0IX_FLOWS)
6994 e1000e_s0ix_entry_flow(adapter);
6995 }
6996
6997 return rc;
6998}
6999
7000static __maybe_unused int e1000e_pm_resume(struct device *dev)
7001{
7002 struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
7003 struct e1000_adapter *adapter = netdev_priv(netdev);
7004 struct pci_dev *pdev = to_pci_dev(dev);
7005 int rc;
7006
7007 /* Introduce S0ix implementation */
7008 if (adapter->flags2 & FLAG2_ENABLE_S0IX_FLOWS)
7009 e1000e_s0ix_exit_flow(adapter);
7010
7011 rc = __e1000_resume(pdev);
7012 if (rc)
7013 return rc;
7014
7015 return e1000e_pm_thaw(dev);
7016}
7017
7018static __maybe_unused int e1000e_pm_runtime_idle(struct device *dev)
7019{
7020 struct net_device *netdev = dev_get_drvdata(dev);
7021 struct e1000_adapter *adapter = netdev_priv(netdev);
7022 u16 eee_lp;
7023
7024 eee_lp = adapter->hw.dev_spec.ich8lan.eee_lp_ability;
7025
7026 if (!e1000e_has_link(adapter)) {
7027 adapter->hw.dev_spec.ich8lan.eee_lp_ability = eee_lp;
7028 pm_schedule_suspend(dev, 5 * MSEC_PER_SEC);
7029 }
7030
7031 return -EBUSY;
7032}
7033
7034static __maybe_unused int e1000e_pm_runtime_resume(struct device *dev)
7035{
7036 struct pci_dev *pdev = to_pci_dev(dev);
7037 struct net_device *netdev = pci_get_drvdata(pdev);
7038 struct e1000_adapter *adapter = netdev_priv(netdev);
7039 int rc;
7040
7041 pdev->pme_poll = true;
7042
7043 rc = __e1000_resume(pdev);
7044 if (rc)
7045 return rc;
7046
7047 if (netdev->flags & IFF_UP)
7048 e1000e_up(adapter);
7049
7050 return rc;
7051}
7052
7053static __maybe_unused int e1000e_pm_runtime_suspend(struct device *dev)
7054{
7055 struct pci_dev *pdev = to_pci_dev(dev);
7056 struct net_device *netdev = pci_get_drvdata(pdev);
7057 struct e1000_adapter *adapter = netdev_priv(netdev);
7058
7059 if (netdev->flags & IFF_UP) {
7060 int count = E1000_CHECK_RESET_COUNT;
7061
7062 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
7063 usleep_range(10000, 11000);
7064
7065 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
7066
7067 /* Down the device without resetting the hardware */
7068 e1000e_down(adapter, false);
7069 }
7070
7071 if (__e1000_shutdown(pdev, true)) {
7072 e1000e_pm_runtime_resume(dev);
7073 return -EBUSY;
7074 }
7075
7076 return 0;
7077}
7078
7079static void e1000_shutdown(struct pci_dev *pdev)
7080{
7081 e1000e_flush_lpic(pdev);
7082
7083 e1000e_pm_freeze(&pdev->dev);
7084
7085 __e1000_shutdown(pdev, false);
7086}
7087
7088#ifdef CONFIG_NET_POLL_CONTROLLER
7089
7090static irqreturn_t e1000_intr_msix(int __always_unused irq, void *data)
7091{
7092 struct net_device *netdev = data;
7093 struct e1000_adapter *adapter = netdev_priv(netdev);
7094
7095 if (adapter->msix_entries) {
7096 int vector, msix_irq;
7097
7098 vector = 0;
7099 msix_irq = adapter->msix_entries[vector].vector;
7100 if (disable_hardirq(msix_irq))
7101 e1000_intr_msix_rx(msix_irq, netdev);
7102 enable_irq(msix_irq);
7103
7104 vector++;
7105 msix_irq = adapter->msix_entries[vector].vector;
7106 if (disable_hardirq(msix_irq))
7107 e1000_intr_msix_tx(msix_irq, netdev);
7108 enable_irq(msix_irq);
7109
7110 vector++;
7111 msix_irq = adapter->msix_entries[vector].vector;
7112 if (disable_hardirq(msix_irq))
7113 e1000_msix_other(msix_irq, netdev);
7114 enable_irq(msix_irq);
7115 }
7116
7117 return IRQ_HANDLED;
7118}
7119
7120/**
7121 * e1000_netpoll
7122 * @netdev: network interface device structure
7123 *
7124 * Polling 'interrupt' - used by things like netconsole to send skbs
7125 * without having to re-enable interrupts. It's not called while
7126 * the interrupt routine is executing.
7127 */
7128static void e1000_netpoll(struct net_device *netdev)
7129{
7130 struct e1000_adapter *adapter = netdev_priv(netdev);
7131
7132 switch (adapter->int_mode) {
7133 case E1000E_INT_MODE_MSIX:
7134 e1000_intr_msix(adapter->pdev->irq, netdev);
7135 break;
7136 case E1000E_INT_MODE_MSI:
7137 if (disable_hardirq(adapter->pdev->irq))
7138 e1000_intr_msi(adapter->pdev->irq, netdev);
7139 enable_irq(adapter->pdev->irq);
7140 break;
7141 default: /* E1000E_INT_MODE_LEGACY */
7142 if (disable_hardirq(adapter->pdev->irq))
7143 e1000_intr(adapter->pdev->irq, netdev);
7144 enable_irq(adapter->pdev->irq);
7145 break;
7146 }
7147}
7148#endif
7149
7150/**
7151 * e1000_io_error_detected - called when PCI error is detected
7152 * @pdev: Pointer to PCI device
7153 * @state: The current pci connection state
7154 *
7155 * This function is called after a PCI bus error affecting
7156 * this device has been detected.
7157 */
7158static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
7159 pci_channel_state_t state)
7160{
7161 e1000e_pm_freeze(&pdev->dev);
7162
7163 if (state == pci_channel_io_perm_failure)
7164 return PCI_ERS_RESULT_DISCONNECT;
7165
7166 pci_disable_device(pdev);
7167
7168 /* Request a slot reset. */
7169 return PCI_ERS_RESULT_NEED_RESET;
7170}
7171
7172/**
7173 * e1000_io_slot_reset - called after the pci bus has been reset.
7174 * @pdev: Pointer to PCI device
7175 *
7176 * Restart the card from scratch, as if from a cold-boot. Implementation
7177 * resembles the first-half of the e1000e_pm_resume routine.
7178 */
7179static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
7180{
7181 struct net_device *netdev = pci_get_drvdata(pdev);
7182 struct e1000_adapter *adapter = netdev_priv(netdev);
7183 struct e1000_hw *hw = &adapter->hw;
7184 u16 aspm_disable_flag = 0;
7185 int err;
7186 pci_ers_result_t result;
7187
7188 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
7189 aspm_disable_flag = PCIE_LINK_STATE_L0S;
7190 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
7191 aspm_disable_flag |= PCIE_LINK_STATE_L1;
7192 if (aspm_disable_flag)
7193 e1000e_disable_aspm_locked(pdev, aspm_disable_flag);
7194
7195 err = pci_enable_device_mem(pdev);
7196 if (err) {
7197 dev_err(&pdev->dev,
7198 "Cannot re-enable PCI device after reset.\n");
7199 result = PCI_ERS_RESULT_DISCONNECT;
7200 } else {
7201 pdev->state_saved = true;
7202 pci_restore_state(pdev);
7203 pci_set_master(pdev);
7204
7205 pci_enable_wake(pdev, PCI_D3hot, 0);
7206 pci_enable_wake(pdev, PCI_D3cold, 0);
7207
7208 e1000e_reset(adapter);
7209 ew32(WUS, ~0);
7210 result = PCI_ERS_RESULT_RECOVERED;
7211 }
7212
7213 return result;
7214}
7215
7216/**
7217 * e1000_io_resume - called when traffic can start flowing again.
7218 * @pdev: Pointer to PCI device
7219 *
7220 * This callback is called when the error recovery driver tells us that
7221 * its OK to resume normal operation. Implementation resembles the
7222 * second-half of the e1000e_pm_resume routine.
7223 */
7224static void e1000_io_resume(struct pci_dev *pdev)
7225{
7226 struct net_device *netdev = pci_get_drvdata(pdev);
7227 struct e1000_adapter *adapter = netdev_priv(netdev);
7228
7229 e1000_init_manageability_pt(adapter);
7230
7231 e1000e_pm_thaw(&pdev->dev);
7232
7233 /* If the controller has AMT, do not set DRV_LOAD until the interface
7234 * is up. For all other cases, let the f/w know that the h/w is now
7235 * under the control of the driver.
7236 */
7237 if (!(adapter->flags & FLAG_HAS_AMT))
7238 e1000e_get_hw_control(adapter);
7239}
7240
7241static void e1000_print_device_info(struct e1000_adapter *adapter)
7242{
7243 struct e1000_hw *hw = &adapter->hw;
7244 struct net_device *netdev = adapter->netdev;
7245 u32 ret_val;
7246 u8 pba_str[E1000_PBANUM_LENGTH];
7247
7248 /* print bus type/speed/width info */
7249 e_info("(PCI Express:2.5GT/s:%s) %pM\n",
7250 /* bus width */
7251 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
7252 "Width x1"),
7253 /* MAC address */
7254 netdev->dev_addr);
7255 e_info("Intel(R) PRO/%s Network Connection\n",
7256 (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
7257 ret_val = e1000_read_pba_string_generic(hw, pba_str,
7258 E1000_PBANUM_LENGTH);
7259 if (ret_val)
7260 strscpy((char *)pba_str, "Unknown", sizeof(pba_str));
7261 e_info("MAC: %d, PHY: %d, PBA No: %s\n",
7262 hw->mac.type, hw->phy.type, pba_str);
7263}
7264
7265static void e1000_eeprom_checks(struct e1000_adapter *adapter)
7266{
7267 struct e1000_hw *hw = &adapter->hw;
7268 int ret_val;
7269 u16 buf = 0;
7270
7271 if (hw->mac.type != e1000_82573)
7272 return;
7273
7274 ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
7275 le16_to_cpus(&buf);
7276 if (!ret_val && (!(buf & BIT(0)))) {
7277 /* Deep Smart Power Down (DSPD) */
7278 dev_warn(&adapter->pdev->dev,
7279 "Warning: detected DSPD enabled in EEPROM\n");
7280 }
7281}
7282
7283static netdev_features_t e1000_fix_features(struct net_device *netdev,
7284 netdev_features_t features)
7285{
7286 struct e1000_adapter *adapter = netdev_priv(netdev);
7287 struct e1000_hw *hw = &adapter->hw;
7288
7289 /* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
7290 if ((hw->mac.type >= e1000_pch2lan) && (netdev->mtu > ETH_DATA_LEN))
7291 features &= ~NETIF_F_RXFCS;
7292
7293 /* Since there is no support for separate Rx/Tx vlan accel
7294 * enable/disable make sure Tx flag is always in same state as Rx.
7295 */
7296 if (features & NETIF_F_HW_VLAN_CTAG_RX)
7297 features |= NETIF_F_HW_VLAN_CTAG_TX;
7298 else
7299 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
7300
7301 return features;
7302}
7303
7304static int e1000_set_features(struct net_device *netdev,
7305 netdev_features_t features)
7306{
7307 struct e1000_adapter *adapter = netdev_priv(netdev);
7308 netdev_features_t changed = features ^ netdev->features;
7309
7310 if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
7311 adapter->flags |= FLAG_TSO_FORCE;
7312
7313 if (!(changed & (NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_TX |
7314 NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_RXFCS |
7315 NETIF_F_RXALL)))
7316 return 0;
7317
7318 if (changed & NETIF_F_RXFCS) {
7319 if (features & NETIF_F_RXFCS) {
7320 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
7321 } else {
7322 /* We need to take it back to defaults, which might mean
7323 * stripping is still disabled at the adapter level.
7324 */
7325 if (adapter->flags2 & FLAG2_DFLT_CRC_STRIPPING)
7326 adapter->flags2 |= FLAG2_CRC_STRIPPING;
7327 else
7328 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
7329 }
7330 }
7331
7332 netdev->features = features;
7333
7334 if (netif_running(netdev))
7335 e1000e_reinit_locked(adapter);
7336 else
7337 e1000e_reset(adapter);
7338
7339 return 1;
7340}
7341
7342static const struct net_device_ops e1000e_netdev_ops = {
7343 .ndo_open = e1000e_open,
7344 .ndo_stop = e1000e_close,
7345 .ndo_start_xmit = e1000_xmit_frame,
7346 .ndo_get_stats64 = e1000e_get_stats64,
7347 .ndo_set_rx_mode = e1000e_set_rx_mode,
7348 .ndo_set_mac_address = e1000_set_mac,
7349 .ndo_change_mtu = e1000_change_mtu,
7350 .ndo_eth_ioctl = e1000_ioctl,
7351 .ndo_tx_timeout = e1000_tx_timeout,
7352 .ndo_validate_addr = eth_validate_addr,
7353
7354 .ndo_vlan_rx_add_vid = e1000_vlan_rx_add_vid,
7355 .ndo_vlan_rx_kill_vid = e1000_vlan_rx_kill_vid,
7356#ifdef CONFIG_NET_POLL_CONTROLLER
7357 .ndo_poll_controller = e1000_netpoll,
7358#endif
7359 .ndo_set_features = e1000_set_features,
7360 .ndo_fix_features = e1000_fix_features,
7361 .ndo_features_check = passthru_features_check,
7362};
7363
7364/**
7365 * e1000_probe - Device Initialization Routine
7366 * @pdev: PCI device information struct
7367 * @ent: entry in e1000_pci_tbl
7368 *
7369 * Returns 0 on success, negative on failure
7370 *
7371 * e1000_probe initializes an adapter identified by a pci_dev structure.
7372 * The OS initialization, configuring of the adapter private structure,
7373 * and a hardware reset occur.
7374 **/
7375static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
7376{
7377 struct net_device *netdev;
7378 struct e1000_adapter *adapter;
7379 struct e1000_hw *hw;
7380 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
7381 resource_size_t mmio_start, mmio_len;
7382 resource_size_t flash_start, flash_len;
7383 static int cards_found;
7384 u16 aspm_disable_flag = 0;
7385 u16 eeprom_data = 0;
7386 u16 eeprom_apme_mask = E1000_EEPROM_APME;
7387 int bars, i, err;
7388 s32 ret_val = 0;
7389
7390 if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
7391 aspm_disable_flag = PCIE_LINK_STATE_L0S;
7392 if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
7393 aspm_disable_flag |= PCIE_LINK_STATE_L1;
7394 if (aspm_disable_flag)
7395 e1000e_disable_aspm(pdev, aspm_disable_flag);
7396
7397 err = pci_enable_device_mem(pdev);
7398 if (err)
7399 return err;
7400
7401 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
7402 if (err) {
7403 dev_err(&pdev->dev,
7404 "No usable DMA configuration, aborting\n");
7405 goto err_dma;
7406 }
7407
7408 bars = pci_select_bars(pdev, IORESOURCE_MEM);
7409 err = pci_request_selected_regions_exclusive(pdev, bars,
7410 e1000e_driver_name);
7411 if (err)
7412 goto err_pci_reg;
7413
7414 pci_set_master(pdev);
7415 /* PCI config space info */
7416 err = pci_save_state(pdev);
7417 if (err)
7418 goto err_alloc_etherdev;
7419
7420 err = -ENOMEM;
7421 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
7422 if (!netdev)
7423 goto err_alloc_etherdev;
7424
7425 SET_NETDEV_DEV(netdev, &pdev->dev);
7426
7427 netdev->irq = pdev->irq;
7428
7429 pci_set_drvdata(pdev, netdev);
7430 adapter = netdev_priv(netdev);
7431 hw = &adapter->hw;
7432 adapter->netdev = netdev;
7433 adapter->pdev = pdev;
7434 adapter->ei = ei;
7435 adapter->pba = ei->pba;
7436 adapter->flags = ei->flags;
7437 adapter->flags2 = ei->flags2;
7438 adapter->hw.adapter = adapter;
7439 adapter->hw.mac.type = ei->mac;
7440 adapter->max_hw_frame_size = ei->max_hw_frame_size;
7441 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
7442
7443 mmio_start = pci_resource_start(pdev, 0);
7444 mmio_len = pci_resource_len(pdev, 0);
7445
7446 err = -EIO;
7447 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
7448 if (!adapter->hw.hw_addr)
7449 goto err_ioremap;
7450
7451 if ((adapter->flags & FLAG_HAS_FLASH) &&
7452 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM) &&
7453 (hw->mac.type < e1000_pch_spt)) {
7454 flash_start = pci_resource_start(pdev, 1);
7455 flash_len = pci_resource_len(pdev, 1);
7456 adapter->hw.flash_address = ioremap(flash_start, flash_len);
7457 if (!adapter->hw.flash_address)
7458 goto err_flashmap;
7459 }
7460
7461 /* Set default EEE advertisement */
7462 if (adapter->flags2 & FLAG2_HAS_EEE)
7463 adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
7464
7465 /* construct the net_device struct */
7466 netdev->netdev_ops = &e1000e_netdev_ops;
7467 e1000e_set_ethtool_ops(netdev);
7468 netdev->watchdog_timeo = 5 * HZ;
7469 netif_napi_add(netdev, &adapter->napi, e1000e_poll);
7470 strscpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
7471
7472 netdev->mem_start = mmio_start;
7473 netdev->mem_end = mmio_start + mmio_len;
7474
7475 adapter->bd_number = cards_found++;
7476
7477 e1000e_check_options(adapter);
7478
7479 /* setup adapter struct */
7480 err = e1000_sw_init(adapter);
7481 if (err)
7482 goto err_sw_init;
7483
7484 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
7485 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
7486 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
7487
7488 err = ei->get_variants(adapter);
7489 if (err)
7490 goto err_hw_init;
7491
7492 if ((adapter->flags & FLAG_IS_ICH) &&
7493 (adapter->flags & FLAG_READ_ONLY_NVM) &&
7494 (hw->mac.type < e1000_pch_spt))
7495 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
7496
7497 hw->mac.ops.get_bus_info(&adapter->hw);
7498
7499 adapter->hw.phy.autoneg_wait_to_complete = 0;
7500
7501 /* Copper options */
7502 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
7503 adapter->hw.phy.mdix = AUTO_ALL_MODES;
7504 adapter->hw.phy.disable_polarity_correction = 0;
7505 adapter->hw.phy.ms_type = e1000_ms_hw_default;
7506 }
7507
7508 if (hw->phy.ops.check_reset_block && hw->phy.ops.check_reset_block(hw))
7509 dev_info(&pdev->dev,
7510 "PHY reset is blocked due to SOL/IDER session.\n");
7511
7512 /* Set initial default active device features */
7513 netdev->features = (NETIF_F_SG |
7514 NETIF_F_HW_VLAN_CTAG_RX |
7515 NETIF_F_HW_VLAN_CTAG_TX |
7516 NETIF_F_TSO |
7517 NETIF_F_TSO6 |
7518 NETIF_F_RXHASH |
7519 NETIF_F_RXCSUM |
7520 NETIF_F_HW_CSUM);
7521
7522 /* disable TSO for pcie and 10/100 speeds to avoid
7523 * some hardware issues and for i219 to fix transfer
7524 * speed being capped at 60%
7525 */
7526 if (!(adapter->flags & FLAG_TSO_FORCE)) {
7527 switch (adapter->link_speed) {
7528 case SPEED_10:
7529 case SPEED_100:
7530 e_info("10/100 speed: disabling TSO\n");
7531 netdev->features &= ~NETIF_F_TSO;
7532 netdev->features &= ~NETIF_F_TSO6;
7533 break;
7534 case SPEED_1000:
7535 netdev->features |= NETIF_F_TSO;
7536 netdev->features |= NETIF_F_TSO6;
7537 break;
7538 default:
7539 /* oops */
7540 break;
7541 }
7542 if (hw->mac.type == e1000_pch_spt) {
7543 netdev->features &= ~NETIF_F_TSO;
7544 netdev->features &= ~NETIF_F_TSO6;
7545 }
7546 }
7547
7548 /* Set user-changeable features (subset of all device features) */
7549 netdev->hw_features = netdev->features;
7550 netdev->hw_features |= NETIF_F_RXFCS;
7551 netdev->priv_flags |= IFF_SUPP_NOFCS;
7552 netdev->hw_features |= NETIF_F_RXALL;
7553
7554 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
7555 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
7556
7557 netdev->vlan_features |= (NETIF_F_SG |
7558 NETIF_F_TSO |
7559 NETIF_F_TSO6 |
7560 NETIF_F_HW_CSUM);
7561
7562 netdev->priv_flags |= IFF_UNICAST_FLT;
7563
7564 netdev->features |= NETIF_F_HIGHDMA;
7565 netdev->vlan_features |= NETIF_F_HIGHDMA;
7566
7567 /* MTU range: 68 - max_hw_frame_size */
7568 netdev->min_mtu = ETH_MIN_MTU;
7569 netdev->max_mtu = adapter->max_hw_frame_size -
7570 (VLAN_ETH_HLEN + ETH_FCS_LEN);
7571
7572 if (e1000e_enable_mng_pass_thru(&adapter->hw))
7573 adapter->flags |= FLAG_MNG_PT_ENABLED;
7574
7575 /* before reading the NVM, reset the controller to
7576 * put the device in a known good starting state
7577 */
7578 adapter->hw.mac.ops.reset_hw(&adapter->hw);
7579
7580 /* systems with ASPM and others may see the checksum fail on the first
7581 * attempt. Let's give it a few tries
7582 */
7583 for (i = 0;; i++) {
7584 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
7585 break;
7586 if (i == 2) {
7587 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
7588 err = -EIO;
7589 goto err_eeprom;
7590 }
7591 }
7592
7593 e1000_eeprom_checks(adapter);
7594
7595 /* copy the MAC address */
7596 if (e1000e_read_mac_addr(&adapter->hw))
7597 dev_err(&pdev->dev,
7598 "NVM Read Error while reading MAC address\n");
7599
7600 eth_hw_addr_set(netdev, adapter->hw.mac.addr);
7601
7602 if (!is_valid_ether_addr(netdev->dev_addr)) {
7603 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
7604 netdev->dev_addr);
7605 err = -EIO;
7606 goto err_eeprom;
7607 }
7608
7609 timer_setup(&adapter->watchdog_timer, e1000_watchdog, 0);
7610 timer_setup(&adapter->phy_info_timer, e1000_update_phy_info, 0);
7611
7612 INIT_WORK(&adapter->reset_task, e1000_reset_task);
7613 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
7614 INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
7615 INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
7616 INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
7617
7618 /* Initialize link parameters. User can change them with ethtool */
7619 adapter->hw.mac.autoneg = 1;
7620 adapter->fc_autoneg = true;
7621 adapter->hw.fc.requested_mode = e1000_fc_default;
7622 adapter->hw.fc.current_mode = e1000_fc_default;
7623 adapter->hw.phy.autoneg_advertised = 0x2f;
7624
7625 /* Initial Wake on LAN setting - If APM wake is enabled in
7626 * the EEPROM, enable the ACPI Magic Packet filter
7627 */
7628 if (adapter->flags & FLAG_APME_IN_WUC) {
7629 /* APME bit in EEPROM is mapped to WUC.APME */
7630 eeprom_data = er32(WUC);
7631 eeprom_apme_mask = E1000_WUC_APME;
7632 if ((hw->mac.type > e1000_ich10lan) &&
7633 (eeprom_data & E1000_WUC_PHY_WAKE))
7634 adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
7635 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
7636 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
7637 (adapter->hw.bus.func == 1))
7638 ret_val = e1000_read_nvm(&adapter->hw,
7639 NVM_INIT_CONTROL3_PORT_B,
7640 1, &eeprom_data);
7641 else
7642 ret_val = e1000_read_nvm(&adapter->hw,
7643 NVM_INIT_CONTROL3_PORT_A,
7644 1, &eeprom_data);
7645 }
7646
7647 /* fetch WoL from EEPROM */
7648 if (ret_val)
7649 e_dbg("NVM read error getting WoL initial values: %d\n", ret_val);
7650 else if (eeprom_data & eeprom_apme_mask)
7651 adapter->eeprom_wol |= E1000_WUFC_MAG;
7652
7653 /* now that we have the eeprom settings, apply the special cases
7654 * where the eeprom may be wrong or the board simply won't support
7655 * wake on lan on a particular port
7656 */
7657 if (!(adapter->flags & FLAG_HAS_WOL))
7658 adapter->eeprom_wol = 0;
7659
7660 /* initialize the wol settings based on the eeprom settings */
7661 adapter->wol = adapter->eeprom_wol;
7662
7663 /* make sure adapter isn't asleep if manageability is enabled */
7664 if (adapter->wol || (adapter->flags & FLAG_MNG_PT_ENABLED) ||
7665 (hw->mac.ops.check_mng_mode(hw)))
7666 device_wakeup_enable(&pdev->dev);
7667
7668 /* save off EEPROM version number */
7669 ret_val = e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
7670
7671 if (ret_val) {
7672 e_dbg("NVM read error getting EEPROM version: %d\n", ret_val);
7673 adapter->eeprom_vers = 0;
7674 }
7675
7676 /* init PTP hardware clock */
7677 e1000e_ptp_init(adapter);
7678
7679 /* reset the hardware with the new settings */
7680 e1000e_reset(adapter);
7681
7682 /* If the controller has AMT, do not set DRV_LOAD until the interface
7683 * is up. For all other cases, let the f/w know that the h/w is now
7684 * under the control of the driver.
7685 */
7686 if (!(adapter->flags & FLAG_HAS_AMT))
7687 e1000e_get_hw_control(adapter);
7688
7689 if (hw->mac.type >= e1000_pch_cnp)
7690 adapter->flags2 |= FLAG2_ENABLE_S0IX_FLOWS;
7691
7692 strscpy(netdev->name, "eth%d", sizeof(netdev->name));
7693 err = register_netdev(netdev);
7694 if (err)
7695 goto err_register;
7696
7697 /* carrier off reporting is important to ethtool even BEFORE open */
7698 netif_carrier_off(netdev);
7699
7700 e1000_print_device_info(adapter);
7701
7702 dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_SMART_PREPARE);
7703
7704 if (pci_dev_run_wake(pdev))
7705 pm_runtime_put_noidle(&pdev->dev);
7706
7707 return 0;
7708
7709err_register:
7710 if (!(adapter->flags & FLAG_HAS_AMT))
7711 e1000e_release_hw_control(adapter);
7712err_eeprom:
7713 if (hw->phy.ops.check_reset_block && !hw->phy.ops.check_reset_block(hw))
7714 e1000_phy_hw_reset(&adapter->hw);
7715err_hw_init:
7716 kfree(adapter->tx_ring);
7717 kfree(adapter->rx_ring);
7718err_sw_init:
7719 if ((adapter->hw.flash_address) && (hw->mac.type < e1000_pch_spt))
7720 iounmap(adapter->hw.flash_address);
7721 e1000e_reset_interrupt_capability(adapter);
7722err_flashmap:
7723 iounmap(adapter->hw.hw_addr);
7724err_ioremap:
7725 free_netdev(netdev);
7726err_alloc_etherdev:
7727 pci_release_mem_regions(pdev);
7728err_pci_reg:
7729err_dma:
7730 pci_disable_device(pdev);
7731 return err;
7732}
7733
7734/**
7735 * e1000_remove - Device Removal Routine
7736 * @pdev: PCI device information struct
7737 *
7738 * e1000_remove is called by the PCI subsystem to alert the driver
7739 * that it should release a PCI device. This could be caused by a
7740 * Hot-Plug event, or because the driver is going to be removed from
7741 * memory.
7742 **/
7743static void e1000_remove(struct pci_dev *pdev)
7744{
7745 struct net_device *netdev = pci_get_drvdata(pdev);
7746 struct e1000_adapter *adapter = netdev_priv(netdev);
7747
7748 e1000e_ptp_remove(adapter);
7749
7750 /* The timers may be rescheduled, so explicitly disable them
7751 * from being rescheduled.
7752 */
7753 set_bit(__E1000_DOWN, &adapter->state);
7754 del_timer_sync(&adapter->watchdog_timer);
7755 del_timer_sync(&adapter->phy_info_timer);
7756
7757 cancel_work_sync(&adapter->reset_task);
7758 cancel_work_sync(&adapter->watchdog_task);
7759 cancel_work_sync(&adapter->downshift_task);
7760 cancel_work_sync(&adapter->update_phy_task);
7761 cancel_work_sync(&adapter->print_hang_task);
7762
7763 if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
7764 cancel_work_sync(&adapter->tx_hwtstamp_work);
7765 if (adapter->tx_hwtstamp_skb) {
7766 dev_consume_skb_any(adapter->tx_hwtstamp_skb);
7767 adapter->tx_hwtstamp_skb = NULL;
7768 }
7769 }
7770
7771 unregister_netdev(netdev);
7772
7773 if (pci_dev_run_wake(pdev))
7774 pm_runtime_get_noresume(&pdev->dev);
7775
7776 /* Release control of h/w to f/w. If f/w is AMT enabled, this
7777 * would have already happened in close and is redundant.
7778 */
7779 e1000e_release_hw_control(adapter);
7780
7781 e1000e_reset_interrupt_capability(adapter);
7782 kfree(adapter->tx_ring);
7783 kfree(adapter->rx_ring);
7784
7785 iounmap(adapter->hw.hw_addr);
7786 if ((adapter->hw.flash_address) &&
7787 (adapter->hw.mac.type < e1000_pch_spt))
7788 iounmap(adapter->hw.flash_address);
7789 pci_release_mem_regions(pdev);
7790
7791 free_netdev(netdev);
7792
7793 pci_disable_device(pdev);
7794}
7795
7796/* PCI Error Recovery (ERS) */
7797static const struct pci_error_handlers e1000_err_handler = {
7798 .error_detected = e1000_io_error_detected,
7799 .slot_reset = e1000_io_slot_reset,
7800 .resume = e1000_io_resume,
7801};
7802
7803static const struct pci_device_id e1000_pci_tbl[] = {
7804 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
7805 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
7806 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
7807 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP),
7808 board_82571 },
7809 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
7810 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
7811 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
7812 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
7813 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
7814
7815 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
7816 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
7817 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
7818 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
7819
7820 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
7821 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
7822 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
7823
7824 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
7825 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
7826 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
7827
7828 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
7829 board_80003es2lan },
7830 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
7831 board_80003es2lan },
7832 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
7833 board_80003es2lan },
7834 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
7835 board_80003es2lan },
7836
7837 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
7838 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
7839 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
7840 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
7841 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
7842 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
7843 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
7844 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
7845
7846 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
7847 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
7848 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
7849 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
7850 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
7851 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
7852 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
7853 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
7854 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
7855
7856 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
7857 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
7858 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
7859
7860 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
7861 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
7862 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
7863
7864 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
7865 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
7866 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
7867 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
7868
7869 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
7870 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
7871
7872 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_LM), board_pch_lpt },
7873 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_V), board_pch_lpt },
7874 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_LM), board_pch_lpt },
7875 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_V), board_pch_lpt },
7876 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM2), board_pch_lpt },
7877 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V2), board_pch_lpt },
7878 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM3), board_pch_lpt },
7879 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V3), board_pch_lpt },
7880 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM), board_pch_spt },
7881 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V), board_pch_spt },
7882 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM2), board_pch_spt },
7883 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V2), board_pch_spt },
7884 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LBG_I219_LM3), board_pch_spt },
7885 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM4), board_pch_spt },
7886 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V4), board_pch_spt },
7887 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM5), board_pch_spt },
7888 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V5), board_pch_spt },
7889 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_LM6), board_pch_cnp },
7890 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_V6), board_pch_cnp },
7891 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_LM7), board_pch_cnp },
7892 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_V7), board_pch_cnp },
7893 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ICP_I219_LM8), board_pch_cnp },
7894 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ICP_I219_V8), board_pch_cnp },
7895 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ICP_I219_LM9), board_pch_cnp },
7896 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ICP_I219_V9), board_pch_cnp },
7897 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CMP_I219_LM10), board_pch_cnp },
7898 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CMP_I219_V10), board_pch_cnp },
7899 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CMP_I219_LM11), board_pch_cnp },
7900 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CMP_I219_V11), board_pch_cnp },
7901 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CMP_I219_LM12), board_pch_spt },
7902 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CMP_I219_V12), board_pch_spt },
7903 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_TGP_I219_LM13), board_pch_tgp },
7904 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_TGP_I219_V13), board_pch_tgp },
7905 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_TGP_I219_LM14), board_pch_tgp },
7906 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_TGP_I219_V14), board_pch_tgp },
7907 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_TGP_I219_LM15), board_pch_tgp },
7908 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_TGP_I219_V15), board_pch_tgp },
7909 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_RPL_I219_LM23), board_pch_adp },
7910 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_RPL_I219_V23), board_pch_adp },
7911 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ADP_I219_LM16), board_pch_adp },
7912 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ADP_I219_V16), board_pch_adp },
7913 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ADP_I219_LM17), board_pch_adp },
7914 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ADP_I219_V17), board_pch_adp },
7915 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_RPL_I219_LM22), board_pch_adp },
7916 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_RPL_I219_V22), board_pch_adp },
7917 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_MTP_I219_LM18), board_pch_mtp },
7918 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_MTP_I219_V18), board_pch_mtp },
7919 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_MTP_I219_LM19), board_pch_mtp },
7920 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_MTP_I219_V19), board_pch_mtp },
7921 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LNP_I219_LM20), board_pch_mtp },
7922 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LNP_I219_V20), board_pch_mtp },
7923 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LNP_I219_LM21), board_pch_mtp },
7924 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LNP_I219_V21), board_pch_mtp },
7925 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ARL_I219_LM24), board_pch_mtp },
7926 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ARL_I219_V24), board_pch_mtp },
7927 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_PTP_I219_LM25), board_pch_mtp },
7928 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_PTP_I219_V25), board_pch_mtp },
7929 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_PTP_I219_LM26), board_pch_mtp },
7930 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_PTP_I219_V26), board_pch_mtp },
7931 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_PTP_I219_LM27), board_pch_mtp },
7932 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_PTP_I219_V27), board_pch_mtp },
7933 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_NVL_I219_LM29), board_pch_mtp },
7934 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_NVL_I219_V29), board_pch_mtp },
7935
7936 { 0, 0, 0, 0, 0, 0, 0 } /* terminate list */
7937};
7938MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
7939
7940static const struct dev_pm_ops e1000_pm_ops = {
7941#ifdef CONFIG_PM_SLEEP
7942 .prepare = e1000e_pm_prepare,
7943 .suspend = e1000e_pm_suspend,
7944 .resume = e1000e_pm_resume,
7945 .freeze = e1000e_pm_freeze,
7946 .thaw = e1000e_pm_thaw,
7947 .poweroff = e1000e_pm_suspend,
7948 .restore = e1000e_pm_resume,
7949#endif
7950 SET_RUNTIME_PM_OPS(e1000e_pm_runtime_suspend, e1000e_pm_runtime_resume,
7951 e1000e_pm_runtime_idle)
7952};
7953
7954/* PCI Device API Driver */
7955static struct pci_driver e1000_driver = {
7956 .name = e1000e_driver_name,
7957 .id_table = e1000_pci_tbl,
7958 .probe = e1000_probe,
7959 .remove = e1000_remove,
7960 .driver = {
7961 .pm = &e1000_pm_ops,
7962 },
7963 .shutdown = e1000_shutdown,
7964 .err_handler = &e1000_err_handler
7965};
7966
7967/**
7968 * e1000_init_module - Driver Registration Routine
7969 *
7970 * e1000_init_module is the first routine called when the driver is
7971 * loaded. All it does is register with the PCI subsystem.
7972 **/
7973static int __init e1000_init_module(void)
7974{
7975 pr_info("Intel(R) PRO/1000 Network Driver\n");
7976 pr_info("Copyright(c) 1999 - 2015 Intel Corporation.\n");
7977
7978 return pci_register_driver(&e1000_driver);
7979}
7980module_init(e1000_init_module);
7981
7982/**
7983 * e1000_exit_module - Driver Exit Cleanup Routine
7984 *
7985 * e1000_exit_module is called just before the driver is removed
7986 * from memory.
7987 **/
7988static void __exit e1000_exit_module(void)
7989{
7990 pci_unregister_driver(&e1000_driver);
7991}
7992module_exit(e1000_exit_module);
7993
7994MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
7995MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
7996MODULE_LICENSE("GPL v2");
7997
7998/* netdev.c */
1/* Intel PRO/1000 Linux driver
2 * Copyright(c) 1999 - 2015 Intel Corporation.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
11 * more details.
12 *
13 * The full GNU General Public License is included in this distribution in
14 * the file called "COPYING".
15 *
16 * Contact Information:
17 * Linux NICS <linux.nics@intel.com>
18 * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
19 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
20 */
21
22#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
23
24#include <linux/module.h>
25#include <linux/types.h>
26#include <linux/init.h>
27#include <linux/pci.h>
28#include <linux/vmalloc.h>
29#include <linux/pagemap.h>
30#include <linux/delay.h>
31#include <linux/netdevice.h>
32#include <linux/interrupt.h>
33#include <linux/tcp.h>
34#include <linux/ipv6.h>
35#include <linux/slab.h>
36#include <net/checksum.h>
37#include <net/ip6_checksum.h>
38#include <linux/ethtool.h>
39#include <linux/if_vlan.h>
40#include <linux/cpu.h>
41#include <linux/smp.h>
42#include <linux/pm_qos.h>
43#include <linux/pm_runtime.h>
44#include <linux/aer.h>
45#include <linux/prefetch.h>
46
47#include "e1000.h"
48
49#define DRV_EXTRAVERSION "-k"
50
51#define DRV_VERSION "3.2.6" DRV_EXTRAVERSION
52char e1000e_driver_name[] = "e1000e";
53const char e1000e_driver_version[] = DRV_VERSION;
54
55#define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
56static int debug = -1;
57module_param(debug, int, 0);
58MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
59
60static const struct e1000_info *e1000_info_tbl[] = {
61 [board_82571] = &e1000_82571_info,
62 [board_82572] = &e1000_82572_info,
63 [board_82573] = &e1000_82573_info,
64 [board_82574] = &e1000_82574_info,
65 [board_82583] = &e1000_82583_info,
66 [board_80003es2lan] = &e1000_es2_info,
67 [board_ich8lan] = &e1000_ich8_info,
68 [board_ich9lan] = &e1000_ich9_info,
69 [board_ich10lan] = &e1000_ich10_info,
70 [board_pchlan] = &e1000_pch_info,
71 [board_pch2lan] = &e1000_pch2_info,
72 [board_pch_lpt] = &e1000_pch_lpt_info,
73 [board_pch_spt] = &e1000_pch_spt_info,
74};
75
76struct e1000_reg_info {
77 u32 ofs;
78 char *name;
79};
80
81static const struct e1000_reg_info e1000_reg_info_tbl[] = {
82 /* General Registers */
83 {E1000_CTRL, "CTRL"},
84 {E1000_STATUS, "STATUS"},
85 {E1000_CTRL_EXT, "CTRL_EXT"},
86
87 /* Interrupt Registers */
88 {E1000_ICR, "ICR"},
89
90 /* Rx Registers */
91 {E1000_RCTL, "RCTL"},
92 {E1000_RDLEN(0), "RDLEN"},
93 {E1000_RDH(0), "RDH"},
94 {E1000_RDT(0), "RDT"},
95 {E1000_RDTR, "RDTR"},
96 {E1000_RXDCTL(0), "RXDCTL"},
97 {E1000_ERT, "ERT"},
98 {E1000_RDBAL(0), "RDBAL"},
99 {E1000_RDBAH(0), "RDBAH"},
100 {E1000_RDFH, "RDFH"},
101 {E1000_RDFT, "RDFT"},
102 {E1000_RDFHS, "RDFHS"},
103 {E1000_RDFTS, "RDFTS"},
104 {E1000_RDFPC, "RDFPC"},
105
106 /* Tx Registers */
107 {E1000_TCTL, "TCTL"},
108 {E1000_TDBAL(0), "TDBAL"},
109 {E1000_TDBAH(0), "TDBAH"},
110 {E1000_TDLEN(0), "TDLEN"},
111 {E1000_TDH(0), "TDH"},
112 {E1000_TDT(0), "TDT"},
113 {E1000_TIDV, "TIDV"},
114 {E1000_TXDCTL(0), "TXDCTL"},
115 {E1000_TADV, "TADV"},
116 {E1000_TARC(0), "TARC"},
117 {E1000_TDFH, "TDFH"},
118 {E1000_TDFT, "TDFT"},
119 {E1000_TDFHS, "TDFHS"},
120 {E1000_TDFTS, "TDFTS"},
121 {E1000_TDFPC, "TDFPC"},
122
123 /* List Terminator */
124 {0, NULL}
125};
126
127/**
128 * __ew32_prepare - prepare to write to MAC CSR register on certain parts
129 * @hw: pointer to the HW structure
130 *
131 * When updating the MAC CSR registers, the Manageability Engine (ME) could
132 * be accessing the registers at the same time. Normally, this is handled in
133 * h/w by an arbiter but on some parts there is a bug that acknowledges Host
134 * accesses later than it should which could result in the register to have
135 * an incorrect value. Workaround this by checking the FWSM register which
136 * has bit 24 set while ME is accessing MAC CSR registers, wait if it is set
137 * and try again a number of times.
138 **/
139s32 __ew32_prepare(struct e1000_hw *hw)
140{
141 s32 i = E1000_ICH_FWSM_PCIM2PCI_COUNT;
142
143 while ((er32(FWSM) & E1000_ICH_FWSM_PCIM2PCI) && --i)
144 udelay(50);
145
146 return i;
147}
148
149void __ew32(struct e1000_hw *hw, unsigned long reg, u32 val)
150{
151 if (hw->adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
152 __ew32_prepare(hw);
153
154 writel(val, hw->hw_addr + reg);
155}
156
157/**
158 * e1000_regdump - register printout routine
159 * @hw: pointer to the HW structure
160 * @reginfo: pointer to the register info table
161 **/
162static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
163{
164 int n = 0;
165 char rname[16];
166 u32 regs[8];
167
168 switch (reginfo->ofs) {
169 case E1000_RXDCTL(0):
170 for (n = 0; n < 2; n++)
171 regs[n] = __er32(hw, E1000_RXDCTL(n));
172 break;
173 case E1000_TXDCTL(0):
174 for (n = 0; n < 2; n++)
175 regs[n] = __er32(hw, E1000_TXDCTL(n));
176 break;
177 case E1000_TARC(0):
178 for (n = 0; n < 2; n++)
179 regs[n] = __er32(hw, E1000_TARC(n));
180 break;
181 default:
182 pr_info("%-15s %08x\n",
183 reginfo->name, __er32(hw, reginfo->ofs));
184 return;
185 }
186
187 snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
188 pr_info("%-15s %08x %08x\n", rname, regs[0], regs[1]);
189}
190
191static void e1000e_dump_ps_pages(struct e1000_adapter *adapter,
192 struct e1000_buffer *bi)
193{
194 int i;
195 struct e1000_ps_page *ps_page;
196
197 for (i = 0; i < adapter->rx_ps_pages; i++) {
198 ps_page = &bi->ps_pages[i];
199
200 if (ps_page->page) {
201 pr_info("packet dump for ps_page %d:\n", i);
202 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
203 16, 1, page_address(ps_page->page),
204 PAGE_SIZE, true);
205 }
206 }
207}
208
209/**
210 * e1000e_dump - Print registers, Tx-ring and Rx-ring
211 * @adapter: board private structure
212 **/
213static void e1000e_dump(struct e1000_adapter *adapter)
214{
215 struct net_device *netdev = adapter->netdev;
216 struct e1000_hw *hw = &adapter->hw;
217 struct e1000_reg_info *reginfo;
218 struct e1000_ring *tx_ring = adapter->tx_ring;
219 struct e1000_tx_desc *tx_desc;
220 struct my_u0 {
221 __le64 a;
222 __le64 b;
223 } *u0;
224 struct e1000_buffer *buffer_info;
225 struct e1000_ring *rx_ring = adapter->rx_ring;
226 union e1000_rx_desc_packet_split *rx_desc_ps;
227 union e1000_rx_desc_extended *rx_desc;
228 struct my_u1 {
229 __le64 a;
230 __le64 b;
231 __le64 c;
232 __le64 d;
233 } *u1;
234 u32 staterr;
235 int i = 0;
236
237 if (!netif_msg_hw(adapter))
238 return;
239
240 /* Print netdevice Info */
241 if (netdev) {
242 dev_info(&adapter->pdev->dev, "Net device Info\n");
243 pr_info("Device Name state trans_start last_rx\n");
244 pr_info("%-15s %016lX %016lX %016lX\n", netdev->name,
245 netdev->state, dev_trans_start(netdev), netdev->last_rx);
246 }
247
248 /* Print Registers */
249 dev_info(&adapter->pdev->dev, "Register Dump\n");
250 pr_info(" Register Name Value\n");
251 for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
252 reginfo->name; reginfo++) {
253 e1000_regdump(hw, reginfo);
254 }
255
256 /* Print Tx Ring Summary */
257 if (!netdev || !netif_running(netdev))
258 return;
259
260 dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
261 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
262 buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
263 pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
264 0, tx_ring->next_to_use, tx_ring->next_to_clean,
265 (unsigned long long)buffer_info->dma,
266 buffer_info->length,
267 buffer_info->next_to_watch,
268 (unsigned long long)buffer_info->time_stamp);
269
270 /* Print Tx Ring */
271 if (!netif_msg_tx_done(adapter))
272 goto rx_ring_summary;
273
274 dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
275
276 /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
277 *
278 * Legacy Transmit Descriptor
279 * +--------------------------------------------------------------+
280 * 0 | Buffer Address [63:0] (Reserved on Write Back) |
281 * +--------------------------------------------------------------+
282 * 8 | Special | CSS | Status | CMD | CSO | Length |
283 * +--------------------------------------------------------------+
284 * 63 48 47 36 35 32 31 24 23 16 15 0
285 *
286 * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
287 * 63 48 47 40 39 32 31 16 15 8 7 0
288 * +----------------------------------------------------------------+
289 * 0 | TUCSE | TUCS0 | TUCSS | IPCSE | IPCS0 | IPCSS |
290 * +----------------------------------------------------------------+
291 * 8 | MSS | HDRLEN | RSV | STA | TUCMD | DTYP | PAYLEN |
292 * +----------------------------------------------------------------+
293 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
294 *
295 * Extended Data Descriptor (DTYP=0x1)
296 * +----------------------------------------------------------------+
297 * 0 | Buffer Address [63:0] |
298 * +----------------------------------------------------------------+
299 * 8 | VLAN tag | POPTS | Rsvd | Status | Command | DTYP | DTALEN |
300 * +----------------------------------------------------------------+
301 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
302 */
303 pr_info("Tl[desc] [address 63:0 ] [SpeCssSCmCsLen] [bi->dma ] leng ntw timestamp bi->skb <-- Legacy format\n");
304 pr_info("Tc[desc] [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma ] leng ntw timestamp bi->skb <-- Ext Context format\n");
305 pr_info("Td[desc] [address 63:0 ] [VlaPoRSCm1Dlen] [bi->dma ] leng ntw timestamp bi->skb <-- Ext Data format\n");
306 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
307 const char *next_desc;
308 tx_desc = E1000_TX_DESC(*tx_ring, i);
309 buffer_info = &tx_ring->buffer_info[i];
310 u0 = (struct my_u0 *)tx_desc;
311 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
312 next_desc = " NTC/U";
313 else if (i == tx_ring->next_to_use)
314 next_desc = " NTU";
315 else if (i == tx_ring->next_to_clean)
316 next_desc = " NTC";
317 else
318 next_desc = "";
319 pr_info("T%c[0x%03X] %016llX %016llX %016llX %04X %3X %016llX %p%s\n",
320 (!(le64_to_cpu(u0->b) & BIT(29)) ? 'l' :
321 ((le64_to_cpu(u0->b) & BIT(20)) ? 'd' : 'c')),
322 i,
323 (unsigned long long)le64_to_cpu(u0->a),
324 (unsigned long long)le64_to_cpu(u0->b),
325 (unsigned long long)buffer_info->dma,
326 buffer_info->length, buffer_info->next_to_watch,
327 (unsigned long long)buffer_info->time_stamp,
328 buffer_info->skb, next_desc);
329
330 if (netif_msg_pktdata(adapter) && buffer_info->skb)
331 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
332 16, 1, buffer_info->skb->data,
333 buffer_info->skb->len, true);
334 }
335
336 /* Print Rx Ring Summary */
337rx_ring_summary:
338 dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
339 pr_info("Queue [NTU] [NTC]\n");
340 pr_info(" %5d %5X %5X\n",
341 0, rx_ring->next_to_use, rx_ring->next_to_clean);
342
343 /* Print Rx Ring */
344 if (!netif_msg_rx_status(adapter))
345 return;
346
347 dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
348 switch (adapter->rx_ps_pages) {
349 case 1:
350 case 2:
351 case 3:
352 /* [Extended] Packet Split Receive Descriptor Format
353 *
354 * +-----------------------------------------------------+
355 * 0 | Buffer Address 0 [63:0] |
356 * +-----------------------------------------------------+
357 * 8 | Buffer Address 1 [63:0] |
358 * +-----------------------------------------------------+
359 * 16 | Buffer Address 2 [63:0] |
360 * +-----------------------------------------------------+
361 * 24 | Buffer Address 3 [63:0] |
362 * +-----------------------------------------------------+
363 */
364 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");
365 /* [Extended] Receive Descriptor (Write-Back) Format
366 *
367 * 63 48 47 32 31 13 12 8 7 4 3 0
368 * +------------------------------------------------------+
369 * 0 | Packet | IP | Rsvd | MRQ | Rsvd | MRQ RSS |
370 * | Checksum | Ident | | Queue | | Type |
371 * +------------------------------------------------------+
372 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
373 * +------------------------------------------------------+
374 * 63 48 47 32 31 20 19 0
375 */
376 pr_info("RWB[desc] [ck ipid mrqhsh] [vl l0 ee es] [ l3 l2 l1 hs] [reserved ] ---------------- [bi->skb] <-- Ext Rx Write-Back format\n");
377 for (i = 0; i < rx_ring->count; i++) {
378 const char *next_desc;
379 buffer_info = &rx_ring->buffer_info[i];
380 rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
381 u1 = (struct my_u1 *)rx_desc_ps;
382 staterr =
383 le32_to_cpu(rx_desc_ps->wb.middle.status_error);
384
385 if (i == rx_ring->next_to_use)
386 next_desc = " NTU";
387 else if (i == rx_ring->next_to_clean)
388 next_desc = " NTC";
389 else
390 next_desc = "";
391
392 if (staterr & E1000_RXD_STAT_DD) {
393 /* Descriptor Done */
394 pr_info("%s[0x%03X] %016llX %016llX %016llX %016llX ---------------- %p%s\n",
395 "RWB", i,
396 (unsigned long long)le64_to_cpu(u1->a),
397 (unsigned long long)le64_to_cpu(u1->b),
398 (unsigned long long)le64_to_cpu(u1->c),
399 (unsigned long long)le64_to_cpu(u1->d),
400 buffer_info->skb, next_desc);
401 } else {
402 pr_info("%s[0x%03X] %016llX %016llX %016llX %016llX %016llX %p%s\n",
403 "R ", i,
404 (unsigned long long)le64_to_cpu(u1->a),
405 (unsigned long long)le64_to_cpu(u1->b),
406 (unsigned long long)le64_to_cpu(u1->c),
407 (unsigned long long)le64_to_cpu(u1->d),
408 (unsigned long long)buffer_info->dma,
409 buffer_info->skb, next_desc);
410
411 if (netif_msg_pktdata(adapter))
412 e1000e_dump_ps_pages(adapter,
413 buffer_info);
414 }
415 }
416 break;
417 default:
418 case 0:
419 /* Extended Receive Descriptor (Read) Format
420 *
421 * +-----------------------------------------------------+
422 * 0 | Buffer Address [63:0] |
423 * +-----------------------------------------------------+
424 * 8 | Reserved |
425 * +-----------------------------------------------------+
426 */
427 pr_info("R [desc] [buf addr 63:0 ] [reserved 63:0 ] [bi->dma ] [bi->skb] <-- Ext (Read) format\n");
428 /* Extended Receive Descriptor (Write-Back) Format
429 *
430 * 63 48 47 32 31 24 23 4 3 0
431 * +------------------------------------------------------+
432 * | RSS Hash | | | |
433 * 0 +-------------------+ Rsvd | Reserved | MRQ RSS |
434 * | Packet | IP | | | Type |
435 * | Checksum | Ident | | | |
436 * +------------------------------------------------------+
437 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
438 * +------------------------------------------------------+
439 * 63 48 47 32 31 20 19 0
440 */
441 pr_info("RWB[desc] [cs ipid mrq] [vt ln xe xs] [bi->skb] <-- Ext (Write-Back) format\n");
442
443 for (i = 0; i < rx_ring->count; i++) {
444 const char *next_desc;
445
446 buffer_info = &rx_ring->buffer_info[i];
447 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
448 u1 = (struct my_u1 *)rx_desc;
449 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
450
451 if (i == rx_ring->next_to_use)
452 next_desc = " NTU";
453 else if (i == rx_ring->next_to_clean)
454 next_desc = " NTC";
455 else
456 next_desc = "";
457
458 if (staterr & E1000_RXD_STAT_DD) {
459 /* Descriptor Done */
460 pr_info("%s[0x%03X] %016llX %016llX ---------------- %p%s\n",
461 "RWB", i,
462 (unsigned long long)le64_to_cpu(u1->a),
463 (unsigned long long)le64_to_cpu(u1->b),
464 buffer_info->skb, next_desc);
465 } else {
466 pr_info("%s[0x%03X] %016llX %016llX %016llX %p%s\n",
467 "R ", i,
468 (unsigned long long)le64_to_cpu(u1->a),
469 (unsigned long long)le64_to_cpu(u1->b),
470 (unsigned long long)buffer_info->dma,
471 buffer_info->skb, next_desc);
472
473 if (netif_msg_pktdata(adapter) &&
474 buffer_info->skb)
475 print_hex_dump(KERN_INFO, "",
476 DUMP_PREFIX_ADDRESS, 16,
477 1,
478 buffer_info->skb->data,
479 adapter->rx_buffer_len,
480 true);
481 }
482 }
483 }
484}
485
486/**
487 * e1000_desc_unused - calculate if we have unused descriptors
488 **/
489static int e1000_desc_unused(struct e1000_ring *ring)
490{
491 if (ring->next_to_clean > ring->next_to_use)
492 return ring->next_to_clean - ring->next_to_use - 1;
493
494 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
495}
496
497/**
498 * e1000e_systim_to_hwtstamp - convert system time value to hw time stamp
499 * @adapter: board private structure
500 * @hwtstamps: time stamp structure to update
501 * @systim: unsigned 64bit system time value.
502 *
503 * Convert the system time value stored in the RX/TXSTMP registers into a
504 * hwtstamp which can be used by the upper level time stamping functions.
505 *
506 * The 'systim_lock' spinlock is used to protect the consistency of the
507 * system time value. This is needed because reading the 64 bit time
508 * value involves reading two 32 bit registers. The first read latches the
509 * value.
510 **/
511static void e1000e_systim_to_hwtstamp(struct e1000_adapter *adapter,
512 struct skb_shared_hwtstamps *hwtstamps,
513 u64 systim)
514{
515 u64 ns;
516 unsigned long flags;
517
518 spin_lock_irqsave(&adapter->systim_lock, flags);
519 ns = timecounter_cyc2time(&adapter->tc, systim);
520 spin_unlock_irqrestore(&adapter->systim_lock, flags);
521
522 memset(hwtstamps, 0, sizeof(*hwtstamps));
523 hwtstamps->hwtstamp = ns_to_ktime(ns);
524}
525
526/**
527 * e1000e_rx_hwtstamp - utility function which checks for Rx time stamp
528 * @adapter: board private structure
529 * @status: descriptor extended error and status field
530 * @skb: particular skb to include time stamp
531 *
532 * If the time stamp is valid, convert it into the timecounter ns value
533 * and store that result into the shhwtstamps structure which is passed
534 * up the network stack.
535 **/
536static void e1000e_rx_hwtstamp(struct e1000_adapter *adapter, u32 status,
537 struct sk_buff *skb)
538{
539 struct e1000_hw *hw = &adapter->hw;
540 u64 rxstmp;
541
542 if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP) ||
543 !(status & E1000_RXDEXT_STATERR_TST) ||
544 !(er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID))
545 return;
546
547 /* The Rx time stamp registers contain the time stamp. No other
548 * received packet will be time stamped until the Rx time stamp
549 * registers are read. Because only one packet can be time stamped
550 * at a time, the register values must belong to this packet and
551 * therefore none of the other additional attributes need to be
552 * compared.
553 */
554 rxstmp = (u64)er32(RXSTMPL);
555 rxstmp |= (u64)er32(RXSTMPH) << 32;
556 e1000e_systim_to_hwtstamp(adapter, skb_hwtstamps(skb), rxstmp);
557
558 adapter->flags2 &= ~FLAG2_CHECK_RX_HWTSTAMP;
559}
560
561/**
562 * e1000_receive_skb - helper function to handle Rx indications
563 * @adapter: board private structure
564 * @staterr: descriptor extended error and status field as written by hardware
565 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
566 * @skb: pointer to sk_buff to be indicated to stack
567 **/
568static void e1000_receive_skb(struct e1000_adapter *adapter,
569 struct net_device *netdev, struct sk_buff *skb,
570 u32 staterr, __le16 vlan)
571{
572 u16 tag = le16_to_cpu(vlan);
573
574 e1000e_rx_hwtstamp(adapter, staterr, skb);
575
576 skb->protocol = eth_type_trans(skb, netdev);
577
578 if (staterr & E1000_RXD_STAT_VP)
579 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tag);
580
581 napi_gro_receive(&adapter->napi, skb);
582}
583
584/**
585 * e1000_rx_checksum - Receive Checksum Offload
586 * @adapter: board private structure
587 * @status_err: receive descriptor status and error fields
588 * @csum: receive descriptor csum field
589 * @sk_buff: socket buffer with received data
590 **/
591static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
592 struct sk_buff *skb)
593{
594 u16 status = (u16)status_err;
595 u8 errors = (u8)(status_err >> 24);
596
597 skb_checksum_none_assert(skb);
598
599 /* Rx checksum disabled */
600 if (!(adapter->netdev->features & NETIF_F_RXCSUM))
601 return;
602
603 /* Ignore Checksum bit is set */
604 if (status & E1000_RXD_STAT_IXSM)
605 return;
606
607 /* TCP/UDP checksum error bit or IP checksum error bit is set */
608 if (errors & (E1000_RXD_ERR_TCPE | E1000_RXD_ERR_IPE)) {
609 /* let the stack verify checksum errors */
610 adapter->hw_csum_err++;
611 return;
612 }
613
614 /* TCP/UDP Checksum has not been calculated */
615 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
616 return;
617
618 /* It must be a TCP or UDP packet with a valid checksum */
619 skb->ip_summed = CHECKSUM_UNNECESSARY;
620 adapter->hw_csum_good++;
621}
622
623static void e1000e_update_rdt_wa(struct e1000_ring *rx_ring, unsigned int i)
624{
625 struct e1000_adapter *adapter = rx_ring->adapter;
626 struct e1000_hw *hw = &adapter->hw;
627 s32 ret_val = __ew32_prepare(hw);
628
629 writel(i, rx_ring->tail);
630
631 if (unlikely(!ret_val && (i != readl(rx_ring->tail)))) {
632 u32 rctl = er32(RCTL);
633
634 ew32(RCTL, rctl & ~E1000_RCTL_EN);
635 e_err("ME firmware caused invalid RDT - resetting\n");
636 schedule_work(&adapter->reset_task);
637 }
638}
639
640static void e1000e_update_tdt_wa(struct e1000_ring *tx_ring, unsigned int i)
641{
642 struct e1000_adapter *adapter = tx_ring->adapter;
643 struct e1000_hw *hw = &adapter->hw;
644 s32 ret_val = __ew32_prepare(hw);
645
646 writel(i, tx_ring->tail);
647
648 if (unlikely(!ret_val && (i != readl(tx_ring->tail)))) {
649 u32 tctl = er32(TCTL);
650
651 ew32(TCTL, tctl & ~E1000_TCTL_EN);
652 e_err("ME firmware caused invalid TDT - resetting\n");
653 schedule_work(&adapter->reset_task);
654 }
655}
656
657/**
658 * e1000_alloc_rx_buffers - Replace used receive buffers
659 * @rx_ring: Rx descriptor ring
660 **/
661static void e1000_alloc_rx_buffers(struct e1000_ring *rx_ring,
662 int cleaned_count, gfp_t gfp)
663{
664 struct e1000_adapter *adapter = rx_ring->adapter;
665 struct net_device *netdev = adapter->netdev;
666 struct pci_dev *pdev = adapter->pdev;
667 union e1000_rx_desc_extended *rx_desc;
668 struct e1000_buffer *buffer_info;
669 struct sk_buff *skb;
670 unsigned int i;
671 unsigned int bufsz = adapter->rx_buffer_len;
672
673 i = rx_ring->next_to_use;
674 buffer_info = &rx_ring->buffer_info[i];
675
676 while (cleaned_count--) {
677 skb = buffer_info->skb;
678 if (skb) {
679 skb_trim(skb, 0);
680 goto map_skb;
681 }
682
683 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
684 if (!skb) {
685 /* Better luck next round */
686 adapter->alloc_rx_buff_failed++;
687 break;
688 }
689
690 buffer_info->skb = skb;
691map_skb:
692 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
693 adapter->rx_buffer_len,
694 DMA_FROM_DEVICE);
695 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
696 dev_err(&pdev->dev, "Rx DMA map failed\n");
697 adapter->rx_dma_failed++;
698 break;
699 }
700
701 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
702 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
703
704 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
705 /* Force memory writes to complete before letting h/w
706 * know there are new descriptors to fetch. (Only
707 * applicable for weak-ordered memory model archs,
708 * such as IA-64).
709 */
710 wmb();
711 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
712 e1000e_update_rdt_wa(rx_ring, i);
713 else
714 writel(i, rx_ring->tail);
715 }
716 i++;
717 if (i == rx_ring->count)
718 i = 0;
719 buffer_info = &rx_ring->buffer_info[i];
720 }
721
722 rx_ring->next_to_use = i;
723}
724
725/**
726 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
727 * @rx_ring: Rx descriptor ring
728 **/
729static void e1000_alloc_rx_buffers_ps(struct e1000_ring *rx_ring,
730 int cleaned_count, gfp_t gfp)
731{
732 struct e1000_adapter *adapter = rx_ring->adapter;
733 struct net_device *netdev = adapter->netdev;
734 struct pci_dev *pdev = adapter->pdev;
735 union e1000_rx_desc_packet_split *rx_desc;
736 struct e1000_buffer *buffer_info;
737 struct e1000_ps_page *ps_page;
738 struct sk_buff *skb;
739 unsigned int i, j;
740
741 i = rx_ring->next_to_use;
742 buffer_info = &rx_ring->buffer_info[i];
743
744 while (cleaned_count--) {
745 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
746
747 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
748 ps_page = &buffer_info->ps_pages[j];
749 if (j >= adapter->rx_ps_pages) {
750 /* all unused desc entries get hw null ptr */
751 rx_desc->read.buffer_addr[j + 1] =
752 ~cpu_to_le64(0);
753 continue;
754 }
755 if (!ps_page->page) {
756 ps_page->page = alloc_page(gfp);
757 if (!ps_page->page) {
758 adapter->alloc_rx_buff_failed++;
759 goto no_buffers;
760 }
761 ps_page->dma = dma_map_page(&pdev->dev,
762 ps_page->page,
763 0, PAGE_SIZE,
764 DMA_FROM_DEVICE);
765 if (dma_mapping_error(&pdev->dev,
766 ps_page->dma)) {
767 dev_err(&adapter->pdev->dev,
768 "Rx DMA page map failed\n");
769 adapter->rx_dma_failed++;
770 goto no_buffers;
771 }
772 }
773 /* Refresh the desc even if buffer_addrs
774 * didn't change because each write-back
775 * erases this info.
776 */
777 rx_desc->read.buffer_addr[j + 1] =
778 cpu_to_le64(ps_page->dma);
779 }
780
781 skb = __netdev_alloc_skb_ip_align(netdev, adapter->rx_ps_bsize0,
782 gfp);
783
784 if (!skb) {
785 adapter->alloc_rx_buff_failed++;
786 break;
787 }
788
789 buffer_info->skb = skb;
790 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
791 adapter->rx_ps_bsize0,
792 DMA_FROM_DEVICE);
793 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
794 dev_err(&pdev->dev, "Rx DMA map failed\n");
795 adapter->rx_dma_failed++;
796 /* cleanup skb */
797 dev_kfree_skb_any(skb);
798 buffer_info->skb = NULL;
799 break;
800 }
801
802 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
803
804 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
805 /* Force memory writes to complete before letting h/w
806 * know there are new descriptors to fetch. (Only
807 * applicable for weak-ordered memory model archs,
808 * such as IA-64).
809 */
810 wmb();
811 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
812 e1000e_update_rdt_wa(rx_ring, i << 1);
813 else
814 writel(i << 1, rx_ring->tail);
815 }
816
817 i++;
818 if (i == rx_ring->count)
819 i = 0;
820 buffer_info = &rx_ring->buffer_info[i];
821 }
822
823no_buffers:
824 rx_ring->next_to_use = i;
825}
826
827/**
828 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
829 * @rx_ring: Rx descriptor ring
830 * @cleaned_count: number of buffers to allocate this pass
831 **/
832
833static void e1000_alloc_jumbo_rx_buffers(struct e1000_ring *rx_ring,
834 int cleaned_count, gfp_t gfp)
835{
836 struct e1000_adapter *adapter = rx_ring->adapter;
837 struct net_device *netdev = adapter->netdev;
838 struct pci_dev *pdev = adapter->pdev;
839 union e1000_rx_desc_extended *rx_desc;
840 struct e1000_buffer *buffer_info;
841 struct sk_buff *skb;
842 unsigned int i;
843 unsigned int bufsz = 256 - 16; /* for skb_reserve */
844
845 i = rx_ring->next_to_use;
846 buffer_info = &rx_ring->buffer_info[i];
847
848 while (cleaned_count--) {
849 skb = buffer_info->skb;
850 if (skb) {
851 skb_trim(skb, 0);
852 goto check_page;
853 }
854
855 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
856 if (unlikely(!skb)) {
857 /* Better luck next round */
858 adapter->alloc_rx_buff_failed++;
859 break;
860 }
861
862 buffer_info->skb = skb;
863check_page:
864 /* allocate a new page if necessary */
865 if (!buffer_info->page) {
866 buffer_info->page = alloc_page(gfp);
867 if (unlikely(!buffer_info->page)) {
868 adapter->alloc_rx_buff_failed++;
869 break;
870 }
871 }
872
873 if (!buffer_info->dma) {
874 buffer_info->dma = dma_map_page(&pdev->dev,
875 buffer_info->page, 0,
876 PAGE_SIZE,
877 DMA_FROM_DEVICE);
878 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
879 adapter->alloc_rx_buff_failed++;
880 break;
881 }
882 }
883
884 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
885 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
886
887 if (unlikely(++i == rx_ring->count))
888 i = 0;
889 buffer_info = &rx_ring->buffer_info[i];
890 }
891
892 if (likely(rx_ring->next_to_use != i)) {
893 rx_ring->next_to_use = i;
894 if (unlikely(i-- == 0))
895 i = (rx_ring->count - 1);
896
897 /* Force memory writes to complete before letting h/w
898 * know there are new descriptors to fetch. (Only
899 * applicable for weak-ordered memory model archs,
900 * such as IA-64).
901 */
902 wmb();
903 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
904 e1000e_update_rdt_wa(rx_ring, i);
905 else
906 writel(i, rx_ring->tail);
907 }
908}
909
910static inline void e1000_rx_hash(struct net_device *netdev, __le32 rss,
911 struct sk_buff *skb)
912{
913 if (netdev->features & NETIF_F_RXHASH)
914 skb_set_hash(skb, le32_to_cpu(rss), PKT_HASH_TYPE_L3);
915}
916
917/**
918 * e1000_clean_rx_irq - Send received data up the network stack
919 * @rx_ring: Rx descriptor ring
920 *
921 * the return value indicates whether actual cleaning was done, there
922 * is no guarantee that everything was cleaned
923 **/
924static bool e1000_clean_rx_irq(struct e1000_ring *rx_ring, int *work_done,
925 int work_to_do)
926{
927 struct e1000_adapter *adapter = rx_ring->adapter;
928 struct net_device *netdev = adapter->netdev;
929 struct pci_dev *pdev = adapter->pdev;
930 struct e1000_hw *hw = &adapter->hw;
931 union e1000_rx_desc_extended *rx_desc, *next_rxd;
932 struct e1000_buffer *buffer_info, *next_buffer;
933 u32 length, staterr;
934 unsigned int i;
935 int cleaned_count = 0;
936 bool cleaned = false;
937 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
938
939 i = rx_ring->next_to_clean;
940 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
941 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
942 buffer_info = &rx_ring->buffer_info[i];
943
944 while (staterr & E1000_RXD_STAT_DD) {
945 struct sk_buff *skb;
946
947 if (*work_done >= work_to_do)
948 break;
949 (*work_done)++;
950 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
951
952 skb = buffer_info->skb;
953 buffer_info->skb = NULL;
954
955 prefetch(skb->data - NET_IP_ALIGN);
956
957 i++;
958 if (i == rx_ring->count)
959 i = 0;
960 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
961 prefetch(next_rxd);
962
963 next_buffer = &rx_ring->buffer_info[i];
964
965 cleaned = true;
966 cleaned_count++;
967 dma_unmap_single(&pdev->dev, buffer_info->dma,
968 adapter->rx_buffer_len, DMA_FROM_DEVICE);
969 buffer_info->dma = 0;
970
971 length = le16_to_cpu(rx_desc->wb.upper.length);
972
973 /* !EOP means multiple descriptors were used to store a single
974 * packet, if that's the case we need to toss it. In fact, we
975 * need to toss every packet with the EOP bit clear and the
976 * next frame that _does_ have the EOP bit set, as it is by
977 * definition only a frame fragment
978 */
979 if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
980 adapter->flags2 |= FLAG2_IS_DISCARDING;
981
982 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
983 /* All receives must fit into a single buffer */
984 e_dbg("Receive packet consumed multiple buffers\n");
985 /* recycle */
986 buffer_info->skb = skb;
987 if (staterr & E1000_RXD_STAT_EOP)
988 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
989 goto next_desc;
990 }
991
992 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
993 !(netdev->features & NETIF_F_RXALL))) {
994 /* recycle */
995 buffer_info->skb = skb;
996 goto next_desc;
997 }
998
999 /* adjust length to remove Ethernet CRC */
1000 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1001 /* If configured to store CRC, don't subtract FCS,
1002 * but keep the FCS bytes out of the total_rx_bytes
1003 * counter
1004 */
1005 if (netdev->features & NETIF_F_RXFCS)
1006 total_rx_bytes -= 4;
1007 else
1008 length -= 4;
1009 }
1010
1011 total_rx_bytes += length;
1012 total_rx_packets++;
1013
1014 /* code added for copybreak, this should improve
1015 * performance for small packets with large amounts
1016 * of reassembly being done in the stack
1017 */
1018 if (length < copybreak) {
1019 struct sk_buff *new_skb =
1020 napi_alloc_skb(&adapter->napi, length);
1021 if (new_skb) {
1022 skb_copy_to_linear_data_offset(new_skb,
1023 -NET_IP_ALIGN,
1024 (skb->data -
1025 NET_IP_ALIGN),
1026 (length +
1027 NET_IP_ALIGN));
1028 /* save the skb in buffer_info as good */
1029 buffer_info->skb = skb;
1030 skb = new_skb;
1031 }
1032 /* else just continue with the old one */
1033 }
1034 /* end copybreak code */
1035 skb_put(skb, length);
1036
1037 /* Receive Checksum Offload */
1038 e1000_rx_checksum(adapter, staterr, skb);
1039
1040 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1041
1042 e1000_receive_skb(adapter, netdev, skb, staterr,
1043 rx_desc->wb.upper.vlan);
1044
1045next_desc:
1046 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1047
1048 /* return some buffers to hardware, one at a time is too slow */
1049 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1050 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1051 GFP_ATOMIC);
1052 cleaned_count = 0;
1053 }
1054
1055 /* use prefetched values */
1056 rx_desc = next_rxd;
1057 buffer_info = next_buffer;
1058
1059 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1060 }
1061 rx_ring->next_to_clean = i;
1062
1063 cleaned_count = e1000_desc_unused(rx_ring);
1064 if (cleaned_count)
1065 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1066
1067 adapter->total_rx_bytes += total_rx_bytes;
1068 adapter->total_rx_packets += total_rx_packets;
1069 return cleaned;
1070}
1071
1072static void e1000_put_txbuf(struct e1000_ring *tx_ring,
1073 struct e1000_buffer *buffer_info)
1074{
1075 struct e1000_adapter *adapter = tx_ring->adapter;
1076
1077 if (buffer_info->dma) {
1078 if (buffer_info->mapped_as_page)
1079 dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1080 buffer_info->length, DMA_TO_DEVICE);
1081 else
1082 dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1083 buffer_info->length, DMA_TO_DEVICE);
1084 buffer_info->dma = 0;
1085 }
1086 if (buffer_info->skb) {
1087 dev_kfree_skb_any(buffer_info->skb);
1088 buffer_info->skb = NULL;
1089 }
1090 buffer_info->time_stamp = 0;
1091}
1092
1093static void e1000_print_hw_hang(struct work_struct *work)
1094{
1095 struct e1000_adapter *adapter = container_of(work,
1096 struct e1000_adapter,
1097 print_hang_task);
1098 struct net_device *netdev = adapter->netdev;
1099 struct e1000_ring *tx_ring = adapter->tx_ring;
1100 unsigned int i = tx_ring->next_to_clean;
1101 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1102 struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1103 struct e1000_hw *hw = &adapter->hw;
1104 u16 phy_status, phy_1000t_status, phy_ext_status;
1105 u16 pci_status;
1106
1107 if (test_bit(__E1000_DOWN, &adapter->state))
1108 return;
1109
1110 if (!adapter->tx_hang_recheck && (adapter->flags2 & FLAG2_DMA_BURST)) {
1111 /* May be block on write-back, flush and detect again
1112 * flush pending descriptor writebacks to memory
1113 */
1114 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1115 /* execute the writes immediately */
1116 e1e_flush();
1117 /* Due to rare timing issues, write to TIDV again to ensure
1118 * the write is successful
1119 */
1120 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1121 /* execute the writes immediately */
1122 e1e_flush();
1123 adapter->tx_hang_recheck = true;
1124 return;
1125 }
1126 adapter->tx_hang_recheck = false;
1127
1128 if (er32(TDH(0)) == er32(TDT(0))) {
1129 e_dbg("false hang detected, ignoring\n");
1130 return;
1131 }
1132
1133 /* Real hang detected */
1134 netif_stop_queue(netdev);
1135
1136 e1e_rphy(hw, MII_BMSR, &phy_status);
1137 e1e_rphy(hw, MII_STAT1000, &phy_1000t_status);
1138 e1e_rphy(hw, MII_ESTATUS, &phy_ext_status);
1139
1140 pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1141
1142 /* detected Hardware unit hang */
1143 e_err("Detected Hardware Unit Hang:\n"
1144 " TDH <%x>\n"
1145 " TDT <%x>\n"
1146 " next_to_use <%x>\n"
1147 " next_to_clean <%x>\n"
1148 "buffer_info[next_to_clean]:\n"
1149 " time_stamp <%lx>\n"
1150 " next_to_watch <%x>\n"
1151 " jiffies <%lx>\n"
1152 " next_to_watch.status <%x>\n"
1153 "MAC Status <%x>\n"
1154 "PHY Status <%x>\n"
1155 "PHY 1000BASE-T Status <%x>\n"
1156 "PHY Extended Status <%x>\n"
1157 "PCI Status <%x>\n",
1158 readl(tx_ring->head), readl(tx_ring->tail), tx_ring->next_to_use,
1159 tx_ring->next_to_clean, tx_ring->buffer_info[eop].time_stamp,
1160 eop, jiffies, eop_desc->upper.fields.status, er32(STATUS),
1161 phy_status, phy_1000t_status, phy_ext_status, pci_status);
1162
1163 e1000e_dump(adapter);
1164
1165 /* Suggest workaround for known h/w issue */
1166 if ((hw->mac.type == e1000_pchlan) && (er32(CTRL) & E1000_CTRL_TFCE))
1167 e_err("Try turning off Tx pause (flow control) via ethtool\n");
1168}
1169
1170/**
1171 * e1000e_tx_hwtstamp_work - check for Tx time stamp
1172 * @work: pointer to work struct
1173 *
1174 * This work function polls the TSYNCTXCTL valid bit to determine when a
1175 * timestamp has been taken for the current stored skb. The timestamp must
1176 * be for this skb because only one such packet is allowed in the queue.
1177 */
1178static void e1000e_tx_hwtstamp_work(struct work_struct *work)
1179{
1180 struct e1000_adapter *adapter = container_of(work, struct e1000_adapter,
1181 tx_hwtstamp_work);
1182 struct e1000_hw *hw = &adapter->hw;
1183
1184 if (er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID) {
1185 struct skb_shared_hwtstamps shhwtstamps;
1186 u64 txstmp;
1187
1188 txstmp = er32(TXSTMPL);
1189 txstmp |= (u64)er32(TXSTMPH) << 32;
1190
1191 e1000e_systim_to_hwtstamp(adapter, &shhwtstamps, txstmp);
1192
1193 skb_tstamp_tx(adapter->tx_hwtstamp_skb, &shhwtstamps);
1194 dev_kfree_skb_any(adapter->tx_hwtstamp_skb);
1195 adapter->tx_hwtstamp_skb = NULL;
1196 } else if (time_after(jiffies, adapter->tx_hwtstamp_start
1197 + adapter->tx_timeout_factor * HZ)) {
1198 dev_kfree_skb_any(adapter->tx_hwtstamp_skb);
1199 adapter->tx_hwtstamp_skb = NULL;
1200 adapter->tx_hwtstamp_timeouts++;
1201 e_warn("clearing Tx timestamp hang\n");
1202 } else {
1203 /* reschedule to check later */
1204 schedule_work(&adapter->tx_hwtstamp_work);
1205 }
1206}
1207
1208/**
1209 * e1000_clean_tx_irq - Reclaim resources after transmit completes
1210 * @tx_ring: Tx descriptor ring
1211 *
1212 * the return value indicates whether actual cleaning was done, there
1213 * is no guarantee that everything was cleaned
1214 **/
1215static bool e1000_clean_tx_irq(struct e1000_ring *tx_ring)
1216{
1217 struct e1000_adapter *adapter = tx_ring->adapter;
1218 struct net_device *netdev = adapter->netdev;
1219 struct e1000_hw *hw = &adapter->hw;
1220 struct e1000_tx_desc *tx_desc, *eop_desc;
1221 struct e1000_buffer *buffer_info;
1222 unsigned int i, eop;
1223 unsigned int count = 0;
1224 unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1225 unsigned int bytes_compl = 0, pkts_compl = 0;
1226
1227 i = tx_ring->next_to_clean;
1228 eop = tx_ring->buffer_info[i].next_to_watch;
1229 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1230
1231 while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1232 (count < tx_ring->count)) {
1233 bool cleaned = false;
1234
1235 dma_rmb(); /* read buffer_info after eop_desc */
1236 for (; !cleaned; count++) {
1237 tx_desc = E1000_TX_DESC(*tx_ring, i);
1238 buffer_info = &tx_ring->buffer_info[i];
1239 cleaned = (i == eop);
1240
1241 if (cleaned) {
1242 total_tx_packets += buffer_info->segs;
1243 total_tx_bytes += buffer_info->bytecount;
1244 if (buffer_info->skb) {
1245 bytes_compl += buffer_info->skb->len;
1246 pkts_compl++;
1247 }
1248 }
1249
1250 e1000_put_txbuf(tx_ring, buffer_info);
1251 tx_desc->upper.data = 0;
1252
1253 i++;
1254 if (i == tx_ring->count)
1255 i = 0;
1256 }
1257
1258 if (i == tx_ring->next_to_use)
1259 break;
1260 eop = tx_ring->buffer_info[i].next_to_watch;
1261 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1262 }
1263
1264 tx_ring->next_to_clean = i;
1265
1266 netdev_completed_queue(netdev, pkts_compl, bytes_compl);
1267
1268#define TX_WAKE_THRESHOLD 32
1269 if (count && netif_carrier_ok(netdev) &&
1270 e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1271 /* Make sure that anybody stopping the queue after this
1272 * sees the new next_to_clean.
1273 */
1274 smp_mb();
1275
1276 if (netif_queue_stopped(netdev) &&
1277 !(test_bit(__E1000_DOWN, &adapter->state))) {
1278 netif_wake_queue(netdev);
1279 ++adapter->restart_queue;
1280 }
1281 }
1282
1283 if (adapter->detect_tx_hung) {
1284 /* Detect a transmit hang in hardware, this serializes the
1285 * check with the clearing of time_stamp and movement of i
1286 */
1287 adapter->detect_tx_hung = false;
1288 if (tx_ring->buffer_info[i].time_stamp &&
1289 time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1290 + (adapter->tx_timeout_factor * HZ)) &&
1291 !(er32(STATUS) & E1000_STATUS_TXOFF))
1292 schedule_work(&adapter->print_hang_task);
1293 else
1294 adapter->tx_hang_recheck = false;
1295 }
1296 adapter->total_tx_bytes += total_tx_bytes;
1297 adapter->total_tx_packets += total_tx_packets;
1298 return count < tx_ring->count;
1299}
1300
1301/**
1302 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1303 * @rx_ring: Rx descriptor ring
1304 *
1305 * the return value indicates whether actual cleaning was done, there
1306 * is no guarantee that everything was cleaned
1307 **/
1308static bool e1000_clean_rx_irq_ps(struct e1000_ring *rx_ring, int *work_done,
1309 int work_to_do)
1310{
1311 struct e1000_adapter *adapter = rx_ring->adapter;
1312 struct e1000_hw *hw = &adapter->hw;
1313 union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1314 struct net_device *netdev = adapter->netdev;
1315 struct pci_dev *pdev = adapter->pdev;
1316 struct e1000_buffer *buffer_info, *next_buffer;
1317 struct e1000_ps_page *ps_page;
1318 struct sk_buff *skb;
1319 unsigned int i, j;
1320 u32 length, staterr;
1321 int cleaned_count = 0;
1322 bool cleaned = false;
1323 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1324
1325 i = rx_ring->next_to_clean;
1326 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1327 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1328 buffer_info = &rx_ring->buffer_info[i];
1329
1330 while (staterr & E1000_RXD_STAT_DD) {
1331 if (*work_done >= work_to_do)
1332 break;
1333 (*work_done)++;
1334 skb = buffer_info->skb;
1335 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
1336
1337 /* in the packet split case this is header only */
1338 prefetch(skb->data - NET_IP_ALIGN);
1339
1340 i++;
1341 if (i == rx_ring->count)
1342 i = 0;
1343 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1344 prefetch(next_rxd);
1345
1346 next_buffer = &rx_ring->buffer_info[i];
1347
1348 cleaned = true;
1349 cleaned_count++;
1350 dma_unmap_single(&pdev->dev, buffer_info->dma,
1351 adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1352 buffer_info->dma = 0;
1353
1354 /* see !EOP comment in other Rx routine */
1355 if (!(staterr & E1000_RXD_STAT_EOP))
1356 adapter->flags2 |= FLAG2_IS_DISCARDING;
1357
1358 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1359 e_dbg("Packet Split buffers didn't pick up the full packet\n");
1360 dev_kfree_skb_irq(skb);
1361 if (staterr & E1000_RXD_STAT_EOP)
1362 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1363 goto next_desc;
1364 }
1365
1366 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1367 !(netdev->features & NETIF_F_RXALL))) {
1368 dev_kfree_skb_irq(skb);
1369 goto next_desc;
1370 }
1371
1372 length = le16_to_cpu(rx_desc->wb.middle.length0);
1373
1374 if (!length) {
1375 e_dbg("Last part of the packet spanning multiple descriptors\n");
1376 dev_kfree_skb_irq(skb);
1377 goto next_desc;
1378 }
1379
1380 /* Good Receive */
1381 skb_put(skb, length);
1382
1383 {
1384 /* this looks ugly, but it seems compiler issues make
1385 * it more efficient than reusing j
1386 */
1387 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1388
1389 /* page alloc/put takes too long and effects small
1390 * packet throughput, so unsplit small packets and
1391 * save the alloc/put only valid in softirq (napi)
1392 * context to call kmap_*
1393 */
1394 if (l1 && (l1 <= copybreak) &&
1395 ((length + l1) <= adapter->rx_ps_bsize0)) {
1396 u8 *vaddr;
1397
1398 ps_page = &buffer_info->ps_pages[0];
1399
1400 /* there is no documentation about how to call
1401 * kmap_atomic, so we can't hold the mapping
1402 * very long
1403 */
1404 dma_sync_single_for_cpu(&pdev->dev,
1405 ps_page->dma,
1406 PAGE_SIZE,
1407 DMA_FROM_DEVICE);
1408 vaddr = kmap_atomic(ps_page->page);
1409 memcpy(skb_tail_pointer(skb), vaddr, l1);
1410 kunmap_atomic(vaddr);
1411 dma_sync_single_for_device(&pdev->dev,
1412 ps_page->dma,
1413 PAGE_SIZE,
1414 DMA_FROM_DEVICE);
1415
1416 /* remove the CRC */
1417 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1418 if (!(netdev->features & NETIF_F_RXFCS))
1419 l1 -= 4;
1420 }
1421
1422 skb_put(skb, l1);
1423 goto copydone;
1424 } /* if */
1425 }
1426
1427 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1428 length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1429 if (!length)
1430 break;
1431
1432 ps_page = &buffer_info->ps_pages[j];
1433 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1434 DMA_FROM_DEVICE);
1435 ps_page->dma = 0;
1436 skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1437 ps_page->page = NULL;
1438 skb->len += length;
1439 skb->data_len += length;
1440 skb->truesize += PAGE_SIZE;
1441 }
1442
1443 /* strip the ethernet crc, problem is we're using pages now so
1444 * this whole operation can get a little cpu intensive
1445 */
1446 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1447 if (!(netdev->features & NETIF_F_RXFCS))
1448 pskb_trim(skb, skb->len - 4);
1449 }
1450
1451copydone:
1452 total_rx_bytes += skb->len;
1453 total_rx_packets++;
1454
1455 e1000_rx_checksum(adapter, staterr, skb);
1456
1457 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1458
1459 if (rx_desc->wb.upper.header_status &
1460 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1461 adapter->rx_hdr_split++;
1462
1463 e1000_receive_skb(adapter, netdev, skb, staterr,
1464 rx_desc->wb.middle.vlan);
1465
1466next_desc:
1467 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1468 buffer_info->skb = NULL;
1469
1470 /* return some buffers to hardware, one at a time is too slow */
1471 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1472 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1473 GFP_ATOMIC);
1474 cleaned_count = 0;
1475 }
1476
1477 /* use prefetched values */
1478 rx_desc = next_rxd;
1479 buffer_info = next_buffer;
1480
1481 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1482 }
1483 rx_ring->next_to_clean = i;
1484
1485 cleaned_count = e1000_desc_unused(rx_ring);
1486 if (cleaned_count)
1487 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1488
1489 adapter->total_rx_bytes += total_rx_bytes;
1490 adapter->total_rx_packets += total_rx_packets;
1491 return cleaned;
1492}
1493
1494/**
1495 * e1000_consume_page - helper function
1496 **/
1497static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1498 u16 length)
1499{
1500 bi->page = NULL;
1501 skb->len += length;
1502 skb->data_len += length;
1503 skb->truesize += PAGE_SIZE;
1504}
1505
1506/**
1507 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1508 * @adapter: board private structure
1509 *
1510 * the return value indicates whether actual cleaning was done, there
1511 * is no guarantee that everything was cleaned
1512 **/
1513static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
1514 int work_to_do)
1515{
1516 struct e1000_adapter *adapter = rx_ring->adapter;
1517 struct net_device *netdev = adapter->netdev;
1518 struct pci_dev *pdev = adapter->pdev;
1519 union e1000_rx_desc_extended *rx_desc, *next_rxd;
1520 struct e1000_buffer *buffer_info, *next_buffer;
1521 u32 length, staterr;
1522 unsigned int i;
1523 int cleaned_count = 0;
1524 bool cleaned = false;
1525 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1526 struct skb_shared_info *shinfo;
1527
1528 i = rx_ring->next_to_clean;
1529 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1530 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1531 buffer_info = &rx_ring->buffer_info[i];
1532
1533 while (staterr & E1000_RXD_STAT_DD) {
1534 struct sk_buff *skb;
1535
1536 if (*work_done >= work_to_do)
1537 break;
1538 (*work_done)++;
1539 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
1540
1541 skb = buffer_info->skb;
1542 buffer_info->skb = NULL;
1543
1544 ++i;
1545 if (i == rx_ring->count)
1546 i = 0;
1547 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1548 prefetch(next_rxd);
1549
1550 next_buffer = &rx_ring->buffer_info[i];
1551
1552 cleaned = true;
1553 cleaned_count++;
1554 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1555 DMA_FROM_DEVICE);
1556 buffer_info->dma = 0;
1557
1558 length = le16_to_cpu(rx_desc->wb.upper.length);
1559
1560 /* errors is only valid for DD + EOP descriptors */
1561 if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1562 ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1563 !(netdev->features & NETIF_F_RXALL)))) {
1564 /* recycle both page and skb */
1565 buffer_info->skb = skb;
1566 /* an error means any chain goes out the window too */
1567 if (rx_ring->rx_skb_top)
1568 dev_kfree_skb_irq(rx_ring->rx_skb_top);
1569 rx_ring->rx_skb_top = NULL;
1570 goto next_desc;
1571 }
1572#define rxtop (rx_ring->rx_skb_top)
1573 if (!(staterr & E1000_RXD_STAT_EOP)) {
1574 /* this descriptor is only the beginning (or middle) */
1575 if (!rxtop) {
1576 /* this is the beginning of a chain */
1577 rxtop = skb;
1578 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1579 0, length);
1580 } else {
1581 /* this is the middle of a chain */
1582 shinfo = skb_shinfo(rxtop);
1583 skb_fill_page_desc(rxtop, shinfo->nr_frags,
1584 buffer_info->page, 0,
1585 length);
1586 /* re-use the skb, only consumed the page */
1587 buffer_info->skb = skb;
1588 }
1589 e1000_consume_page(buffer_info, rxtop, length);
1590 goto next_desc;
1591 } else {
1592 if (rxtop) {
1593 /* end of the 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 current skb, we only consumed the
1599 * page
1600 */
1601 buffer_info->skb = skb;
1602 skb = rxtop;
1603 rxtop = NULL;
1604 e1000_consume_page(buffer_info, skb, length);
1605 } else {
1606 /* no chain, got EOP, this buf is the packet
1607 * copybreak to save the put_page/alloc_page
1608 */
1609 if (length <= copybreak &&
1610 skb_tailroom(skb) >= length) {
1611 u8 *vaddr;
1612 vaddr = kmap_atomic(buffer_info->page);
1613 memcpy(skb_tail_pointer(skb), vaddr,
1614 length);
1615 kunmap_atomic(vaddr);
1616 /* re-use the page, so don't erase
1617 * buffer_info->page
1618 */
1619 skb_put(skb, length);
1620 } else {
1621 skb_fill_page_desc(skb, 0,
1622 buffer_info->page, 0,
1623 length);
1624 e1000_consume_page(buffer_info, skb,
1625 length);
1626 }
1627 }
1628 }
1629
1630 /* Receive Checksum Offload */
1631 e1000_rx_checksum(adapter, staterr, skb);
1632
1633 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1634
1635 /* probably a little skewed due to removing CRC */
1636 total_rx_bytes += skb->len;
1637 total_rx_packets++;
1638
1639 /* eth type trans needs skb->data to point to something */
1640 if (!pskb_may_pull(skb, ETH_HLEN)) {
1641 e_err("pskb_may_pull failed.\n");
1642 dev_kfree_skb_irq(skb);
1643 goto next_desc;
1644 }
1645
1646 e1000_receive_skb(adapter, netdev, skb, staterr,
1647 rx_desc->wb.upper.vlan);
1648
1649next_desc:
1650 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1651
1652 /* return some buffers to hardware, one at a time is too slow */
1653 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1654 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1655 GFP_ATOMIC);
1656 cleaned_count = 0;
1657 }
1658
1659 /* use prefetched values */
1660 rx_desc = next_rxd;
1661 buffer_info = next_buffer;
1662
1663 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1664 }
1665 rx_ring->next_to_clean = i;
1666
1667 cleaned_count = e1000_desc_unused(rx_ring);
1668 if (cleaned_count)
1669 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1670
1671 adapter->total_rx_bytes += total_rx_bytes;
1672 adapter->total_rx_packets += total_rx_packets;
1673 return cleaned;
1674}
1675
1676/**
1677 * e1000_clean_rx_ring - Free Rx Buffers per Queue
1678 * @rx_ring: Rx descriptor ring
1679 **/
1680static void e1000_clean_rx_ring(struct e1000_ring *rx_ring)
1681{
1682 struct e1000_adapter *adapter = rx_ring->adapter;
1683 struct e1000_buffer *buffer_info;
1684 struct e1000_ps_page *ps_page;
1685 struct pci_dev *pdev = adapter->pdev;
1686 unsigned int i, j;
1687
1688 /* Free all the Rx ring sk_buffs */
1689 for (i = 0; i < rx_ring->count; i++) {
1690 buffer_info = &rx_ring->buffer_info[i];
1691 if (buffer_info->dma) {
1692 if (adapter->clean_rx == e1000_clean_rx_irq)
1693 dma_unmap_single(&pdev->dev, buffer_info->dma,
1694 adapter->rx_buffer_len,
1695 DMA_FROM_DEVICE);
1696 else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1697 dma_unmap_page(&pdev->dev, buffer_info->dma,
1698 PAGE_SIZE, DMA_FROM_DEVICE);
1699 else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1700 dma_unmap_single(&pdev->dev, buffer_info->dma,
1701 adapter->rx_ps_bsize0,
1702 DMA_FROM_DEVICE);
1703 buffer_info->dma = 0;
1704 }
1705
1706 if (buffer_info->page) {
1707 put_page(buffer_info->page);
1708 buffer_info->page = NULL;
1709 }
1710
1711 if (buffer_info->skb) {
1712 dev_kfree_skb(buffer_info->skb);
1713 buffer_info->skb = NULL;
1714 }
1715
1716 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1717 ps_page = &buffer_info->ps_pages[j];
1718 if (!ps_page->page)
1719 break;
1720 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1721 DMA_FROM_DEVICE);
1722 ps_page->dma = 0;
1723 put_page(ps_page->page);
1724 ps_page->page = NULL;
1725 }
1726 }
1727
1728 /* there also may be some cached data from a chained receive */
1729 if (rx_ring->rx_skb_top) {
1730 dev_kfree_skb(rx_ring->rx_skb_top);
1731 rx_ring->rx_skb_top = NULL;
1732 }
1733
1734 /* Zero out the descriptor ring */
1735 memset(rx_ring->desc, 0, rx_ring->size);
1736
1737 rx_ring->next_to_clean = 0;
1738 rx_ring->next_to_use = 0;
1739 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1740}
1741
1742static void e1000e_downshift_workaround(struct work_struct *work)
1743{
1744 struct e1000_adapter *adapter = container_of(work,
1745 struct e1000_adapter,
1746 downshift_task);
1747
1748 if (test_bit(__E1000_DOWN, &adapter->state))
1749 return;
1750
1751 e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1752}
1753
1754/**
1755 * e1000_intr_msi - Interrupt Handler
1756 * @irq: interrupt number
1757 * @data: pointer to a network interface device structure
1758 **/
1759static irqreturn_t e1000_intr_msi(int __always_unused irq, void *data)
1760{
1761 struct net_device *netdev = data;
1762 struct e1000_adapter *adapter = netdev_priv(netdev);
1763 struct e1000_hw *hw = &adapter->hw;
1764 u32 icr = er32(ICR);
1765
1766 /* read ICR disables interrupts using IAM */
1767 if (icr & E1000_ICR_LSC) {
1768 hw->mac.get_link_status = true;
1769 /* ICH8 workaround-- Call gig speed drop workaround on cable
1770 * disconnect (LSC) before accessing any PHY registers
1771 */
1772 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1773 (!(er32(STATUS) & E1000_STATUS_LU)))
1774 schedule_work(&adapter->downshift_task);
1775
1776 /* 80003ES2LAN workaround-- For packet buffer work-around on
1777 * link down event; disable receives here in the ISR and reset
1778 * adapter in watchdog
1779 */
1780 if (netif_carrier_ok(netdev) &&
1781 adapter->flags & FLAG_RX_NEEDS_RESTART) {
1782 /* disable receives */
1783 u32 rctl = er32(RCTL);
1784
1785 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1786 adapter->flags |= FLAG_RESTART_NOW;
1787 }
1788 /* guard against interrupt when we're going down */
1789 if (!test_bit(__E1000_DOWN, &adapter->state))
1790 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1791 }
1792
1793 /* Reset on uncorrectable ECC error */
1794 if ((icr & E1000_ICR_ECCER) && ((hw->mac.type == e1000_pch_lpt) ||
1795 (hw->mac.type == e1000_pch_spt))) {
1796 u32 pbeccsts = er32(PBECCSTS);
1797
1798 adapter->corr_errors +=
1799 pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1800 adapter->uncorr_errors +=
1801 (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
1802 E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
1803
1804 /* Do the reset outside of interrupt context */
1805 schedule_work(&adapter->reset_task);
1806
1807 /* return immediately since reset is imminent */
1808 return IRQ_HANDLED;
1809 }
1810
1811 if (napi_schedule_prep(&adapter->napi)) {
1812 adapter->total_tx_bytes = 0;
1813 adapter->total_tx_packets = 0;
1814 adapter->total_rx_bytes = 0;
1815 adapter->total_rx_packets = 0;
1816 __napi_schedule(&adapter->napi);
1817 }
1818
1819 return IRQ_HANDLED;
1820}
1821
1822/**
1823 * e1000_intr - Interrupt Handler
1824 * @irq: interrupt number
1825 * @data: pointer to a network interface device structure
1826 **/
1827static irqreturn_t e1000_intr(int __always_unused irq, void *data)
1828{
1829 struct net_device *netdev = data;
1830 struct e1000_adapter *adapter = netdev_priv(netdev);
1831 struct e1000_hw *hw = &adapter->hw;
1832 u32 rctl, icr = er32(ICR);
1833
1834 if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1835 return IRQ_NONE; /* Not our interrupt */
1836
1837 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1838 * not set, then the adapter didn't send an interrupt
1839 */
1840 if (!(icr & E1000_ICR_INT_ASSERTED))
1841 return IRQ_NONE;
1842
1843 /* Interrupt Auto-Mask...upon reading ICR,
1844 * interrupts are masked. No need for the
1845 * IMC write
1846 */
1847
1848 if (icr & E1000_ICR_LSC) {
1849 hw->mac.get_link_status = true;
1850 /* ICH8 workaround-- Call gig speed drop workaround on cable
1851 * disconnect (LSC) before accessing any PHY registers
1852 */
1853 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1854 (!(er32(STATUS) & E1000_STATUS_LU)))
1855 schedule_work(&adapter->downshift_task);
1856
1857 /* 80003ES2LAN workaround--
1858 * For packet buffer work-around on link down event;
1859 * disable receives here in the ISR and
1860 * reset adapter in watchdog
1861 */
1862 if (netif_carrier_ok(netdev) &&
1863 (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1864 /* disable receives */
1865 rctl = er32(RCTL);
1866 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1867 adapter->flags |= FLAG_RESTART_NOW;
1868 }
1869 /* guard against interrupt when we're going down */
1870 if (!test_bit(__E1000_DOWN, &adapter->state))
1871 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1872 }
1873
1874 /* Reset on uncorrectable ECC error */
1875 if ((icr & E1000_ICR_ECCER) && ((hw->mac.type == e1000_pch_lpt) ||
1876 (hw->mac.type == e1000_pch_spt))) {
1877 u32 pbeccsts = er32(PBECCSTS);
1878
1879 adapter->corr_errors +=
1880 pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1881 adapter->uncorr_errors +=
1882 (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
1883 E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
1884
1885 /* Do the reset outside of interrupt context */
1886 schedule_work(&adapter->reset_task);
1887
1888 /* return immediately since reset is imminent */
1889 return IRQ_HANDLED;
1890 }
1891
1892 if (napi_schedule_prep(&adapter->napi)) {
1893 adapter->total_tx_bytes = 0;
1894 adapter->total_tx_packets = 0;
1895 adapter->total_rx_bytes = 0;
1896 adapter->total_rx_packets = 0;
1897 __napi_schedule(&adapter->napi);
1898 }
1899
1900 return IRQ_HANDLED;
1901}
1902
1903static irqreturn_t e1000_msix_other(int __always_unused irq, void *data)
1904{
1905 struct net_device *netdev = data;
1906 struct e1000_adapter *adapter = netdev_priv(netdev);
1907 struct e1000_hw *hw = &adapter->hw;
1908
1909 hw->mac.get_link_status = true;
1910
1911 /* guard against interrupt when we're going down */
1912 if (!test_bit(__E1000_DOWN, &adapter->state)) {
1913 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1914 ew32(IMS, E1000_IMS_OTHER);
1915 }
1916
1917 return IRQ_HANDLED;
1918}
1919
1920static irqreturn_t e1000_intr_msix_tx(int __always_unused irq, void *data)
1921{
1922 struct net_device *netdev = data;
1923 struct e1000_adapter *adapter = netdev_priv(netdev);
1924 struct e1000_hw *hw = &adapter->hw;
1925 struct e1000_ring *tx_ring = adapter->tx_ring;
1926
1927 adapter->total_tx_bytes = 0;
1928 adapter->total_tx_packets = 0;
1929
1930 if (!e1000_clean_tx_irq(tx_ring))
1931 /* Ring was not completely cleaned, so fire another interrupt */
1932 ew32(ICS, tx_ring->ims_val);
1933
1934 if (!test_bit(__E1000_DOWN, &adapter->state))
1935 ew32(IMS, adapter->tx_ring->ims_val);
1936
1937 return IRQ_HANDLED;
1938}
1939
1940static irqreturn_t e1000_intr_msix_rx(int __always_unused irq, void *data)
1941{
1942 struct net_device *netdev = data;
1943 struct e1000_adapter *adapter = netdev_priv(netdev);
1944 struct e1000_ring *rx_ring = adapter->rx_ring;
1945
1946 /* Write the ITR value calculated at the end of the
1947 * previous interrupt.
1948 */
1949 if (rx_ring->set_itr) {
1950 u32 itr = rx_ring->itr_val ?
1951 1000000000 / (rx_ring->itr_val * 256) : 0;
1952
1953 writel(itr, rx_ring->itr_register);
1954 rx_ring->set_itr = 0;
1955 }
1956
1957 if (napi_schedule_prep(&adapter->napi)) {
1958 adapter->total_rx_bytes = 0;
1959 adapter->total_rx_packets = 0;
1960 __napi_schedule(&adapter->napi);
1961 }
1962 return IRQ_HANDLED;
1963}
1964
1965/**
1966 * e1000_configure_msix - Configure MSI-X hardware
1967 *
1968 * e1000_configure_msix sets up the hardware to properly
1969 * generate MSI-X interrupts.
1970 **/
1971static void e1000_configure_msix(struct e1000_adapter *adapter)
1972{
1973 struct e1000_hw *hw = &adapter->hw;
1974 struct e1000_ring *rx_ring = adapter->rx_ring;
1975 struct e1000_ring *tx_ring = adapter->tx_ring;
1976 int vector = 0;
1977 u32 ctrl_ext, ivar = 0;
1978
1979 adapter->eiac_mask = 0;
1980
1981 /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1982 if (hw->mac.type == e1000_82574) {
1983 u32 rfctl = er32(RFCTL);
1984
1985 rfctl |= E1000_RFCTL_ACK_DIS;
1986 ew32(RFCTL, rfctl);
1987 }
1988
1989 /* Configure Rx vector */
1990 rx_ring->ims_val = E1000_IMS_RXQ0;
1991 adapter->eiac_mask |= rx_ring->ims_val;
1992 if (rx_ring->itr_val)
1993 writel(1000000000 / (rx_ring->itr_val * 256),
1994 rx_ring->itr_register);
1995 else
1996 writel(1, rx_ring->itr_register);
1997 ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1998
1999 /* Configure Tx vector */
2000 tx_ring->ims_val = E1000_IMS_TXQ0;
2001 vector++;
2002 if (tx_ring->itr_val)
2003 writel(1000000000 / (tx_ring->itr_val * 256),
2004 tx_ring->itr_register);
2005 else
2006 writel(1, tx_ring->itr_register);
2007 adapter->eiac_mask |= tx_ring->ims_val;
2008 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
2009
2010 /* set vector for Other Causes, e.g. link changes */
2011 vector++;
2012 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
2013 if (rx_ring->itr_val)
2014 writel(1000000000 / (rx_ring->itr_val * 256),
2015 hw->hw_addr + E1000_EITR_82574(vector));
2016 else
2017 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
2018 adapter->eiac_mask |= E1000_IMS_OTHER;
2019
2020 /* Cause Tx interrupts on every write back */
2021 ivar |= BIT(31);
2022
2023 ew32(IVAR, ivar);
2024
2025 /* enable MSI-X PBA support */
2026 ctrl_ext = er32(CTRL_EXT) & ~E1000_CTRL_EXT_IAME;
2027 ctrl_ext |= E1000_CTRL_EXT_PBA_CLR | E1000_CTRL_EXT_EIAME;
2028 ew32(CTRL_EXT, ctrl_ext);
2029 e1e_flush();
2030}
2031
2032void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
2033{
2034 if (adapter->msix_entries) {
2035 pci_disable_msix(adapter->pdev);
2036 kfree(adapter->msix_entries);
2037 adapter->msix_entries = NULL;
2038 } else if (adapter->flags & FLAG_MSI_ENABLED) {
2039 pci_disable_msi(adapter->pdev);
2040 adapter->flags &= ~FLAG_MSI_ENABLED;
2041 }
2042}
2043
2044/**
2045 * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
2046 *
2047 * Attempt to configure interrupts using the best available
2048 * capabilities of the hardware and kernel.
2049 **/
2050void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
2051{
2052 int err;
2053 int i;
2054
2055 switch (adapter->int_mode) {
2056 case E1000E_INT_MODE_MSIX:
2057 if (adapter->flags & FLAG_HAS_MSIX) {
2058 adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
2059 adapter->msix_entries = kcalloc(adapter->num_vectors,
2060 sizeof(struct
2061 msix_entry),
2062 GFP_KERNEL);
2063 if (adapter->msix_entries) {
2064 struct e1000_adapter *a = adapter;
2065
2066 for (i = 0; i < adapter->num_vectors; i++)
2067 adapter->msix_entries[i].entry = i;
2068
2069 err = pci_enable_msix_range(a->pdev,
2070 a->msix_entries,
2071 a->num_vectors,
2072 a->num_vectors);
2073 if (err > 0)
2074 return;
2075 }
2076 /* MSI-X failed, so fall through and try MSI */
2077 e_err("Failed to initialize MSI-X interrupts. Falling back to MSI interrupts.\n");
2078 e1000e_reset_interrupt_capability(adapter);
2079 }
2080 adapter->int_mode = E1000E_INT_MODE_MSI;
2081 /* Fall through */
2082 case E1000E_INT_MODE_MSI:
2083 if (!pci_enable_msi(adapter->pdev)) {
2084 adapter->flags |= FLAG_MSI_ENABLED;
2085 } else {
2086 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2087 e_err("Failed to initialize MSI interrupts. Falling back to legacy interrupts.\n");
2088 }
2089 /* Fall through */
2090 case E1000E_INT_MODE_LEGACY:
2091 /* Don't do anything; this is the system default */
2092 break;
2093 }
2094
2095 /* store the number of vectors being used */
2096 adapter->num_vectors = 1;
2097}
2098
2099/**
2100 * e1000_request_msix - Initialize MSI-X interrupts
2101 *
2102 * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
2103 * kernel.
2104 **/
2105static int e1000_request_msix(struct e1000_adapter *adapter)
2106{
2107 struct net_device *netdev = adapter->netdev;
2108 int err = 0, vector = 0;
2109
2110 if (strlen(netdev->name) < (IFNAMSIZ - 5))
2111 snprintf(adapter->rx_ring->name,
2112 sizeof(adapter->rx_ring->name) - 1,
2113 "%s-rx-0", netdev->name);
2114 else
2115 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
2116 err = request_irq(adapter->msix_entries[vector].vector,
2117 e1000_intr_msix_rx, 0, adapter->rx_ring->name,
2118 netdev);
2119 if (err)
2120 return err;
2121 adapter->rx_ring->itr_register = adapter->hw.hw_addr +
2122 E1000_EITR_82574(vector);
2123 adapter->rx_ring->itr_val = adapter->itr;
2124 vector++;
2125
2126 if (strlen(netdev->name) < (IFNAMSIZ - 5))
2127 snprintf(adapter->tx_ring->name,
2128 sizeof(adapter->tx_ring->name) - 1,
2129 "%s-tx-0", netdev->name);
2130 else
2131 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
2132 err = request_irq(adapter->msix_entries[vector].vector,
2133 e1000_intr_msix_tx, 0, adapter->tx_ring->name,
2134 netdev);
2135 if (err)
2136 return err;
2137 adapter->tx_ring->itr_register = adapter->hw.hw_addr +
2138 E1000_EITR_82574(vector);
2139 adapter->tx_ring->itr_val = adapter->itr;
2140 vector++;
2141
2142 err = request_irq(adapter->msix_entries[vector].vector,
2143 e1000_msix_other, 0, netdev->name, netdev);
2144 if (err)
2145 return err;
2146
2147 e1000_configure_msix(adapter);
2148
2149 return 0;
2150}
2151
2152/**
2153 * e1000_request_irq - initialize interrupts
2154 *
2155 * Attempts to configure interrupts using the best available
2156 * capabilities of the hardware and kernel.
2157 **/
2158static int e1000_request_irq(struct e1000_adapter *adapter)
2159{
2160 struct net_device *netdev = adapter->netdev;
2161 int err;
2162
2163 if (adapter->msix_entries) {
2164 err = e1000_request_msix(adapter);
2165 if (!err)
2166 return err;
2167 /* fall back to MSI */
2168 e1000e_reset_interrupt_capability(adapter);
2169 adapter->int_mode = E1000E_INT_MODE_MSI;
2170 e1000e_set_interrupt_capability(adapter);
2171 }
2172 if (adapter->flags & FLAG_MSI_ENABLED) {
2173 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2174 netdev->name, netdev);
2175 if (!err)
2176 return err;
2177
2178 /* fall back to legacy interrupt */
2179 e1000e_reset_interrupt_capability(adapter);
2180 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2181 }
2182
2183 err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2184 netdev->name, netdev);
2185 if (err)
2186 e_err("Unable to allocate interrupt, Error: %d\n", err);
2187
2188 return err;
2189}
2190
2191static void e1000_free_irq(struct e1000_adapter *adapter)
2192{
2193 struct net_device *netdev = adapter->netdev;
2194
2195 if (adapter->msix_entries) {
2196 int vector = 0;
2197
2198 free_irq(adapter->msix_entries[vector].vector, netdev);
2199 vector++;
2200
2201 free_irq(adapter->msix_entries[vector].vector, netdev);
2202 vector++;
2203
2204 /* Other Causes interrupt vector */
2205 free_irq(adapter->msix_entries[vector].vector, netdev);
2206 return;
2207 }
2208
2209 free_irq(adapter->pdev->irq, netdev);
2210}
2211
2212/**
2213 * e1000_irq_disable - Mask off interrupt generation on the NIC
2214 **/
2215static void e1000_irq_disable(struct e1000_adapter *adapter)
2216{
2217 struct e1000_hw *hw = &adapter->hw;
2218
2219 ew32(IMC, ~0);
2220 if (adapter->msix_entries)
2221 ew32(EIAC_82574, 0);
2222 e1e_flush();
2223
2224 if (adapter->msix_entries) {
2225 int i;
2226
2227 for (i = 0; i < adapter->num_vectors; i++)
2228 synchronize_irq(adapter->msix_entries[i].vector);
2229 } else {
2230 synchronize_irq(adapter->pdev->irq);
2231 }
2232}
2233
2234/**
2235 * e1000_irq_enable - Enable default interrupt generation settings
2236 **/
2237static void e1000_irq_enable(struct e1000_adapter *adapter)
2238{
2239 struct e1000_hw *hw = &adapter->hw;
2240
2241 if (adapter->msix_entries) {
2242 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2243 ew32(IMS, adapter->eiac_mask | E1000_IMS_LSC);
2244 } else if ((hw->mac.type == e1000_pch_lpt) ||
2245 (hw->mac.type == e1000_pch_spt)) {
2246 ew32(IMS, IMS_ENABLE_MASK | E1000_IMS_ECCER);
2247 } else {
2248 ew32(IMS, IMS_ENABLE_MASK);
2249 }
2250 e1e_flush();
2251}
2252
2253/**
2254 * e1000e_get_hw_control - get control of the h/w from f/w
2255 * @adapter: address of board private structure
2256 *
2257 * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2258 * For ASF and Pass Through versions of f/w this means that
2259 * the driver is loaded. For AMT version (only with 82573)
2260 * of the f/w this means that the network i/f is open.
2261 **/
2262void e1000e_get_hw_control(struct e1000_adapter *adapter)
2263{
2264 struct e1000_hw *hw = &adapter->hw;
2265 u32 ctrl_ext;
2266 u32 swsm;
2267
2268 /* Let firmware know the driver has taken over */
2269 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2270 swsm = er32(SWSM);
2271 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2272 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2273 ctrl_ext = er32(CTRL_EXT);
2274 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2275 }
2276}
2277
2278/**
2279 * e1000e_release_hw_control - release control of the h/w to f/w
2280 * @adapter: address of board private structure
2281 *
2282 * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2283 * For ASF and Pass Through versions of f/w this means that the
2284 * driver is no longer loaded. For AMT version (only with 82573) i
2285 * of the f/w this means that the network i/f is closed.
2286 *
2287 **/
2288void e1000e_release_hw_control(struct e1000_adapter *adapter)
2289{
2290 struct e1000_hw *hw = &adapter->hw;
2291 u32 ctrl_ext;
2292 u32 swsm;
2293
2294 /* Let firmware taken over control of h/w */
2295 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2296 swsm = er32(SWSM);
2297 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2298 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2299 ctrl_ext = er32(CTRL_EXT);
2300 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2301 }
2302}
2303
2304/**
2305 * e1000_alloc_ring_dma - allocate memory for a ring structure
2306 **/
2307static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2308 struct e1000_ring *ring)
2309{
2310 struct pci_dev *pdev = adapter->pdev;
2311
2312 ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2313 GFP_KERNEL);
2314 if (!ring->desc)
2315 return -ENOMEM;
2316
2317 return 0;
2318}
2319
2320/**
2321 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2322 * @tx_ring: Tx descriptor ring
2323 *
2324 * Return 0 on success, negative on failure
2325 **/
2326int e1000e_setup_tx_resources(struct e1000_ring *tx_ring)
2327{
2328 struct e1000_adapter *adapter = tx_ring->adapter;
2329 int err = -ENOMEM, size;
2330
2331 size = sizeof(struct e1000_buffer) * tx_ring->count;
2332 tx_ring->buffer_info = vzalloc(size);
2333 if (!tx_ring->buffer_info)
2334 goto err;
2335
2336 /* round up to nearest 4K */
2337 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2338 tx_ring->size = ALIGN(tx_ring->size, 4096);
2339
2340 err = e1000_alloc_ring_dma(adapter, tx_ring);
2341 if (err)
2342 goto err;
2343
2344 tx_ring->next_to_use = 0;
2345 tx_ring->next_to_clean = 0;
2346
2347 return 0;
2348err:
2349 vfree(tx_ring->buffer_info);
2350 e_err("Unable to allocate memory for the transmit descriptor ring\n");
2351 return err;
2352}
2353
2354/**
2355 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2356 * @rx_ring: Rx descriptor ring
2357 *
2358 * Returns 0 on success, negative on failure
2359 **/
2360int e1000e_setup_rx_resources(struct e1000_ring *rx_ring)
2361{
2362 struct e1000_adapter *adapter = rx_ring->adapter;
2363 struct e1000_buffer *buffer_info;
2364 int i, size, desc_len, err = -ENOMEM;
2365
2366 size = sizeof(struct e1000_buffer) * rx_ring->count;
2367 rx_ring->buffer_info = vzalloc(size);
2368 if (!rx_ring->buffer_info)
2369 goto err;
2370
2371 for (i = 0; i < rx_ring->count; i++) {
2372 buffer_info = &rx_ring->buffer_info[i];
2373 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2374 sizeof(struct e1000_ps_page),
2375 GFP_KERNEL);
2376 if (!buffer_info->ps_pages)
2377 goto err_pages;
2378 }
2379
2380 desc_len = sizeof(union e1000_rx_desc_packet_split);
2381
2382 /* Round up to nearest 4K */
2383 rx_ring->size = rx_ring->count * desc_len;
2384 rx_ring->size = ALIGN(rx_ring->size, 4096);
2385
2386 err = e1000_alloc_ring_dma(adapter, rx_ring);
2387 if (err)
2388 goto err_pages;
2389
2390 rx_ring->next_to_clean = 0;
2391 rx_ring->next_to_use = 0;
2392 rx_ring->rx_skb_top = NULL;
2393
2394 return 0;
2395
2396err_pages:
2397 for (i = 0; i < rx_ring->count; i++) {
2398 buffer_info = &rx_ring->buffer_info[i];
2399 kfree(buffer_info->ps_pages);
2400 }
2401err:
2402 vfree(rx_ring->buffer_info);
2403 e_err("Unable to allocate memory for the receive descriptor ring\n");
2404 return err;
2405}
2406
2407/**
2408 * e1000_clean_tx_ring - Free Tx Buffers
2409 * @tx_ring: Tx descriptor ring
2410 **/
2411static void e1000_clean_tx_ring(struct e1000_ring *tx_ring)
2412{
2413 struct e1000_adapter *adapter = tx_ring->adapter;
2414 struct e1000_buffer *buffer_info;
2415 unsigned long size;
2416 unsigned int i;
2417
2418 for (i = 0; i < tx_ring->count; i++) {
2419 buffer_info = &tx_ring->buffer_info[i];
2420 e1000_put_txbuf(tx_ring, buffer_info);
2421 }
2422
2423 netdev_reset_queue(adapter->netdev);
2424 size = sizeof(struct e1000_buffer) * tx_ring->count;
2425 memset(tx_ring->buffer_info, 0, size);
2426
2427 memset(tx_ring->desc, 0, tx_ring->size);
2428
2429 tx_ring->next_to_use = 0;
2430 tx_ring->next_to_clean = 0;
2431}
2432
2433/**
2434 * e1000e_free_tx_resources - Free Tx Resources per Queue
2435 * @tx_ring: Tx descriptor ring
2436 *
2437 * Free all transmit software resources
2438 **/
2439void e1000e_free_tx_resources(struct e1000_ring *tx_ring)
2440{
2441 struct e1000_adapter *adapter = tx_ring->adapter;
2442 struct pci_dev *pdev = adapter->pdev;
2443
2444 e1000_clean_tx_ring(tx_ring);
2445
2446 vfree(tx_ring->buffer_info);
2447 tx_ring->buffer_info = NULL;
2448
2449 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2450 tx_ring->dma);
2451 tx_ring->desc = NULL;
2452}
2453
2454/**
2455 * e1000e_free_rx_resources - Free Rx Resources
2456 * @rx_ring: Rx descriptor ring
2457 *
2458 * Free all receive software resources
2459 **/
2460void e1000e_free_rx_resources(struct e1000_ring *rx_ring)
2461{
2462 struct e1000_adapter *adapter = rx_ring->adapter;
2463 struct pci_dev *pdev = adapter->pdev;
2464 int i;
2465
2466 e1000_clean_rx_ring(rx_ring);
2467
2468 for (i = 0; i < rx_ring->count; i++)
2469 kfree(rx_ring->buffer_info[i].ps_pages);
2470
2471 vfree(rx_ring->buffer_info);
2472 rx_ring->buffer_info = NULL;
2473
2474 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2475 rx_ring->dma);
2476 rx_ring->desc = NULL;
2477}
2478
2479/**
2480 * e1000_update_itr - update the dynamic ITR value based on statistics
2481 * @adapter: pointer to adapter
2482 * @itr_setting: current adapter->itr
2483 * @packets: the number of packets during this measurement interval
2484 * @bytes: the number of bytes during this measurement interval
2485 *
2486 * Stores a new ITR value based on packets and byte
2487 * counts during the last interrupt. The advantage of per interrupt
2488 * computation is faster updates and more accurate ITR for the current
2489 * traffic pattern. Constants in this function were computed
2490 * based on theoretical maximum wire speed and thresholds were set based
2491 * on testing data as well as attempting to minimize response time
2492 * while increasing bulk throughput. This functionality is controlled
2493 * by the InterruptThrottleRate module parameter.
2494 **/
2495static unsigned int e1000_update_itr(u16 itr_setting, int packets, int bytes)
2496{
2497 unsigned int retval = itr_setting;
2498
2499 if (packets == 0)
2500 return itr_setting;
2501
2502 switch (itr_setting) {
2503 case lowest_latency:
2504 /* handle TSO and jumbo frames */
2505 if (bytes / packets > 8000)
2506 retval = bulk_latency;
2507 else if ((packets < 5) && (bytes > 512))
2508 retval = low_latency;
2509 break;
2510 case low_latency: /* 50 usec aka 20000 ints/s */
2511 if (bytes > 10000) {
2512 /* this if handles the TSO accounting */
2513 if (bytes / packets > 8000)
2514 retval = bulk_latency;
2515 else if ((packets < 10) || ((bytes / packets) > 1200))
2516 retval = bulk_latency;
2517 else if ((packets > 35))
2518 retval = lowest_latency;
2519 } else if (bytes / packets > 2000) {
2520 retval = bulk_latency;
2521 } else if (packets <= 2 && bytes < 512) {
2522 retval = lowest_latency;
2523 }
2524 break;
2525 case bulk_latency: /* 250 usec aka 4000 ints/s */
2526 if (bytes > 25000) {
2527 if (packets > 35)
2528 retval = low_latency;
2529 } else if (bytes < 6000) {
2530 retval = low_latency;
2531 }
2532 break;
2533 }
2534
2535 return retval;
2536}
2537
2538static void e1000_set_itr(struct e1000_adapter *adapter)
2539{
2540 u16 current_itr;
2541 u32 new_itr = adapter->itr;
2542
2543 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2544 if (adapter->link_speed != SPEED_1000) {
2545 current_itr = 0;
2546 new_itr = 4000;
2547 goto set_itr_now;
2548 }
2549
2550 if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2551 new_itr = 0;
2552 goto set_itr_now;
2553 }
2554
2555 adapter->tx_itr = e1000_update_itr(adapter->tx_itr,
2556 adapter->total_tx_packets,
2557 adapter->total_tx_bytes);
2558 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2559 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2560 adapter->tx_itr = low_latency;
2561
2562 adapter->rx_itr = e1000_update_itr(adapter->rx_itr,
2563 adapter->total_rx_packets,
2564 adapter->total_rx_bytes);
2565 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2566 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2567 adapter->rx_itr = low_latency;
2568
2569 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2570
2571 /* counts and packets in update_itr are dependent on these numbers */
2572 switch (current_itr) {
2573 case lowest_latency:
2574 new_itr = 70000;
2575 break;
2576 case low_latency:
2577 new_itr = 20000; /* aka hwitr = ~200 */
2578 break;
2579 case bulk_latency:
2580 new_itr = 4000;
2581 break;
2582 default:
2583 break;
2584 }
2585
2586set_itr_now:
2587 if (new_itr != adapter->itr) {
2588 /* this attempts to bias the interrupt rate towards Bulk
2589 * by adding intermediate steps when interrupt rate is
2590 * increasing
2591 */
2592 new_itr = new_itr > adapter->itr ?
2593 min(adapter->itr + (new_itr >> 2), new_itr) : new_itr;
2594 adapter->itr = new_itr;
2595 adapter->rx_ring->itr_val = new_itr;
2596 if (adapter->msix_entries)
2597 adapter->rx_ring->set_itr = 1;
2598 else
2599 e1000e_write_itr(adapter, new_itr);
2600 }
2601}
2602
2603/**
2604 * e1000e_write_itr - write the ITR value to the appropriate registers
2605 * @adapter: address of board private structure
2606 * @itr: new ITR value to program
2607 *
2608 * e1000e_write_itr determines if the adapter is in MSI-X mode
2609 * and, if so, writes the EITR registers with the ITR value.
2610 * Otherwise, it writes the ITR value into the ITR register.
2611 **/
2612void e1000e_write_itr(struct e1000_adapter *adapter, u32 itr)
2613{
2614 struct e1000_hw *hw = &adapter->hw;
2615 u32 new_itr = itr ? 1000000000 / (itr * 256) : 0;
2616
2617 if (adapter->msix_entries) {
2618 int vector;
2619
2620 for (vector = 0; vector < adapter->num_vectors; vector++)
2621 writel(new_itr, hw->hw_addr + E1000_EITR_82574(vector));
2622 } else {
2623 ew32(ITR, new_itr);
2624 }
2625}
2626
2627/**
2628 * e1000_alloc_queues - Allocate memory for all rings
2629 * @adapter: board private structure to initialize
2630 **/
2631static int e1000_alloc_queues(struct e1000_adapter *adapter)
2632{
2633 int size = sizeof(struct e1000_ring);
2634
2635 adapter->tx_ring = kzalloc(size, GFP_KERNEL);
2636 if (!adapter->tx_ring)
2637 goto err;
2638 adapter->tx_ring->count = adapter->tx_ring_count;
2639 adapter->tx_ring->adapter = adapter;
2640
2641 adapter->rx_ring = kzalloc(size, GFP_KERNEL);
2642 if (!adapter->rx_ring)
2643 goto err;
2644 adapter->rx_ring->count = adapter->rx_ring_count;
2645 adapter->rx_ring->adapter = adapter;
2646
2647 return 0;
2648err:
2649 e_err("Unable to allocate memory for queues\n");
2650 kfree(adapter->rx_ring);
2651 kfree(adapter->tx_ring);
2652 return -ENOMEM;
2653}
2654
2655/**
2656 * e1000e_poll - NAPI Rx polling callback
2657 * @napi: struct associated with this polling callback
2658 * @weight: number of packets driver is allowed to process this poll
2659 **/
2660static int e1000e_poll(struct napi_struct *napi, int weight)
2661{
2662 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter,
2663 napi);
2664 struct e1000_hw *hw = &adapter->hw;
2665 struct net_device *poll_dev = adapter->netdev;
2666 int tx_cleaned = 1, work_done = 0;
2667
2668 adapter = netdev_priv(poll_dev);
2669
2670 if (!adapter->msix_entries ||
2671 (adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2672 tx_cleaned = e1000_clean_tx_irq(adapter->tx_ring);
2673
2674 adapter->clean_rx(adapter->rx_ring, &work_done, weight);
2675
2676 if (!tx_cleaned)
2677 work_done = weight;
2678
2679 /* If weight not fully consumed, exit the polling mode */
2680 if (work_done < weight) {
2681 if (adapter->itr_setting & 3)
2682 e1000_set_itr(adapter);
2683 napi_complete_done(napi, work_done);
2684 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2685 if (adapter->msix_entries)
2686 ew32(IMS, adapter->rx_ring->ims_val);
2687 else
2688 e1000_irq_enable(adapter);
2689 }
2690 }
2691
2692 return work_done;
2693}
2694
2695static int e1000_vlan_rx_add_vid(struct net_device *netdev,
2696 __always_unused __be16 proto, u16 vid)
2697{
2698 struct e1000_adapter *adapter = netdev_priv(netdev);
2699 struct e1000_hw *hw = &adapter->hw;
2700 u32 vfta, index;
2701
2702 /* don't update vlan cookie if already programmed */
2703 if ((adapter->hw.mng_cookie.status &
2704 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2705 (vid == adapter->mng_vlan_id))
2706 return 0;
2707
2708 /* add VID to filter table */
2709 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2710 index = (vid >> 5) & 0x7F;
2711 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2712 vfta |= BIT((vid & 0x1F));
2713 hw->mac.ops.write_vfta(hw, index, vfta);
2714 }
2715
2716 set_bit(vid, adapter->active_vlans);
2717
2718 return 0;
2719}
2720
2721static int e1000_vlan_rx_kill_vid(struct net_device *netdev,
2722 __always_unused __be16 proto, u16 vid)
2723{
2724 struct e1000_adapter *adapter = netdev_priv(netdev);
2725 struct e1000_hw *hw = &adapter->hw;
2726 u32 vfta, index;
2727
2728 if ((adapter->hw.mng_cookie.status &
2729 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2730 (vid == adapter->mng_vlan_id)) {
2731 /* release control to f/w */
2732 e1000e_release_hw_control(adapter);
2733 return 0;
2734 }
2735
2736 /* remove VID from filter table */
2737 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2738 index = (vid >> 5) & 0x7F;
2739 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2740 vfta &= ~BIT((vid & 0x1F));
2741 hw->mac.ops.write_vfta(hw, index, vfta);
2742 }
2743
2744 clear_bit(vid, adapter->active_vlans);
2745
2746 return 0;
2747}
2748
2749/**
2750 * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2751 * @adapter: board private structure to initialize
2752 **/
2753static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2754{
2755 struct net_device *netdev = adapter->netdev;
2756 struct e1000_hw *hw = &adapter->hw;
2757 u32 rctl;
2758
2759 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2760 /* disable VLAN receive filtering */
2761 rctl = er32(RCTL);
2762 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2763 ew32(RCTL, rctl);
2764
2765 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2766 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
2767 adapter->mng_vlan_id);
2768 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2769 }
2770 }
2771}
2772
2773/**
2774 * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2775 * @adapter: board private structure to initialize
2776 **/
2777static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2778{
2779 struct e1000_hw *hw = &adapter->hw;
2780 u32 rctl;
2781
2782 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2783 /* enable VLAN receive filtering */
2784 rctl = er32(RCTL);
2785 rctl |= E1000_RCTL_VFE;
2786 rctl &= ~E1000_RCTL_CFIEN;
2787 ew32(RCTL, rctl);
2788 }
2789}
2790
2791/**
2792 * e1000e_vlan_strip_disable - helper to disable HW VLAN stripping
2793 * @adapter: board private structure to initialize
2794 **/
2795static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2796{
2797 struct e1000_hw *hw = &adapter->hw;
2798 u32 ctrl;
2799
2800 /* disable VLAN tag insert/strip */
2801 ctrl = er32(CTRL);
2802 ctrl &= ~E1000_CTRL_VME;
2803 ew32(CTRL, ctrl);
2804}
2805
2806/**
2807 * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2808 * @adapter: board private structure to initialize
2809 **/
2810static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2811{
2812 struct e1000_hw *hw = &adapter->hw;
2813 u32 ctrl;
2814
2815 /* enable VLAN tag insert/strip */
2816 ctrl = er32(CTRL);
2817 ctrl |= E1000_CTRL_VME;
2818 ew32(CTRL, ctrl);
2819}
2820
2821static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2822{
2823 struct net_device *netdev = adapter->netdev;
2824 u16 vid = adapter->hw.mng_cookie.vlan_id;
2825 u16 old_vid = adapter->mng_vlan_id;
2826
2827 if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2828 e1000_vlan_rx_add_vid(netdev, htons(ETH_P_8021Q), vid);
2829 adapter->mng_vlan_id = vid;
2830 }
2831
2832 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2833 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q), old_vid);
2834}
2835
2836static void e1000_restore_vlan(struct e1000_adapter *adapter)
2837{
2838 u16 vid;
2839
2840 e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
2841
2842 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2843 e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
2844}
2845
2846static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2847{
2848 struct e1000_hw *hw = &adapter->hw;
2849 u32 manc, manc2h, mdef, i, j;
2850
2851 if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2852 return;
2853
2854 manc = er32(MANC);
2855
2856 /* enable receiving management packets to the host. this will probably
2857 * generate destination unreachable messages from the host OS, but
2858 * the packets will be handled on SMBUS
2859 */
2860 manc |= E1000_MANC_EN_MNG2HOST;
2861 manc2h = er32(MANC2H);
2862
2863 switch (hw->mac.type) {
2864 default:
2865 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2866 break;
2867 case e1000_82574:
2868 case e1000_82583:
2869 /* Check if IPMI pass-through decision filter already exists;
2870 * if so, enable it.
2871 */
2872 for (i = 0, j = 0; i < 8; i++) {
2873 mdef = er32(MDEF(i));
2874
2875 /* Ignore filters with anything other than IPMI ports */
2876 if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2877 continue;
2878
2879 /* Enable this decision filter in MANC2H */
2880 if (mdef)
2881 manc2h |= BIT(i);
2882
2883 j |= mdef;
2884 }
2885
2886 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2887 break;
2888
2889 /* Create new decision filter in an empty filter */
2890 for (i = 0, j = 0; i < 8; i++)
2891 if (er32(MDEF(i)) == 0) {
2892 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2893 E1000_MDEF_PORT_664));
2894 manc2h |= BIT(1);
2895 j++;
2896 break;
2897 }
2898
2899 if (!j)
2900 e_warn("Unable to create IPMI pass-through filter\n");
2901 break;
2902 }
2903
2904 ew32(MANC2H, manc2h);
2905 ew32(MANC, manc);
2906}
2907
2908/**
2909 * e1000_configure_tx - Configure Transmit Unit after Reset
2910 * @adapter: board private structure
2911 *
2912 * Configure the Tx unit of the MAC after a reset.
2913 **/
2914static void e1000_configure_tx(struct e1000_adapter *adapter)
2915{
2916 struct e1000_hw *hw = &adapter->hw;
2917 struct e1000_ring *tx_ring = adapter->tx_ring;
2918 u64 tdba;
2919 u32 tdlen, tctl, tarc;
2920
2921 /* Setup the HW Tx Head and Tail descriptor pointers */
2922 tdba = tx_ring->dma;
2923 tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2924 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
2925 ew32(TDBAH(0), (tdba >> 32));
2926 ew32(TDLEN(0), tdlen);
2927 ew32(TDH(0), 0);
2928 ew32(TDT(0), 0);
2929 tx_ring->head = adapter->hw.hw_addr + E1000_TDH(0);
2930 tx_ring->tail = adapter->hw.hw_addr + E1000_TDT(0);
2931
2932 writel(0, tx_ring->head);
2933 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
2934 e1000e_update_tdt_wa(tx_ring, 0);
2935 else
2936 writel(0, tx_ring->tail);
2937
2938 /* Set the Tx Interrupt Delay register */
2939 ew32(TIDV, adapter->tx_int_delay);
2940 /* Tx irq moderation */
2941 ew32(TADV, adapter->tx_abs_int_delay);
2942
2943 if (adapter->flags2 & FLAG2_DMA_BURST) {
2944 u32 txdctl = er32(TXDCTL(0));
2945
2946 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2947 E1000_TXDCTL_WTHRESH);
2948 /* set up some performance related parameters to encourage the
2949 * hardware to use the bus more efficiently in bursts, depends
2950 * on the tx_int_delay to be enabled,
2951 * wthresh = 1 ==> burst write is disabled to avoid Tx stalls
2952 * hthresh = 1 ==> prefetch when one or more available
2953 * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2954 * BEWARE: this seems to work but should be considered first if
2955 * there are Tx hangs or other Tx related bugs
2956 */
2957 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2958 ew32(TXDCTL(0), txdctl);
2959 }
2960 /* erratum work around: set txdctl the same for both queues */
2961 ew32(TXDCTL(1), er32(TXDCTL(0)));
2962
2963 /* Program the Transmit Control Register */
2964 tctl = er32(TCTL);
2965 tctl &= ~E1000_TCTL_CT;
2966 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2967 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2968
2969 if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2970 tarc = er32(TARC(0));
2971 /* set the speed mode bit, we'll clear it if we're not at
2972 * gigabit link later
2973 */
2974#define SPEED_MODE_BIT BIT(21)
2975 tarc |= SPEED_MODE_BIT;
2976 ew32(TARC(0), tarc);
2977 }
2978
2979 /* errata: program both queues to unweighted RR */
2980 if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2981 tarc = er32(TARC(0));
2982 tarc |= 1;
2983 ew32(TARC(0), tarc);
2984 tarc = er32(TARC(1));
2985 tarc |= 1;
2986 ew32(TARC(1), tarc);
2987 }
2988
2989 /* Setup Transmit Descriptor Settings for eop descriptor */
2990 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2991
2992 /* only set IDE if we are delaying interrupts using the timers */
2993 if (adapter->tx_int_delay)
2994 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2995
2996 /* enable Report Status bit */
2997 adapter->txd_cmd |= E1000_TXD_CMD_RS;
2998
2999 ew32(TCTL, tctl);
3000
3001 hw->mac.ops.config_collision_dist(hw);
3002
3003 /* SPT Si errata workaround to avoid data corruption */
3004 if (hw->mac.type == e1000_pch_spt) {
3005 u32 reg_val;
3006
3007 reg_val = er32(IOSFPC);
3008 reg_val |= E1000_RCTL_RDMTS_HEX;
3009 ew32(IOSFPC, reg_val);
3010
3011 reg_val = er32(TARC(0));
3012 reg_val |= E1000_TARC0_CB_MULTIQ_3_REQ;
3013 ew32(TARC(0), reg_val);
3014 }
3015}
3016
3017/**
3018 * e1000_setup_rctl - configure the receive control registers
3019 * @adapter: Board private structure
3020 **/
3021#define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
3022 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
3023static void e1000_setup_rctl(struct e1000_adapter *adapter)
3024{
3025 struct e1000_hw *hw = &adapter->hw;
3026 u32 rctl, rfctl;
3027 u32 pages = 0;
3028
3029 /* Workaround Si errata on PCHx - configure jumbo frame flow.
3030 * If jumbo frames not set, program related MAC/PHY registers
3031 * to h/w defaults
3032 */
3033 if (hw->mac.type >= e1000_pch2lan) {
3034 s32 ret_val;
3035
3036 if (adapter->netdev->mtu > ETH_DATA_LEN)
3037 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
3038 else
3039 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
3040
3041 if (ret_val)
3042 e_dbg("failed to enable|disable jumbo frame workaround mode\n");
3043 }
3044
3045 /* Program MC offset vector base */
3046 rctl = er32(RCTL);
3047 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3048 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
3049 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
3050 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3051
3052 /* Do not Store bad packets */
3053 rctl &= ~E1000_RCTL_SBP;
3054
3055 /* Enable Long Packet receive */
3056 if (adapter->netdev->mtu <= ETH_DATA_LEN)
3057 rctl &= ~E1000_RCTL_LPE;
3058 else
3059 rctl |= E1000_RCTL_LPE;
3060
3061 /* Some systems expect that the CRC is included in SMBUS traffic. The
3062 * hardware strips the CRC before sending to both SMBUS (BMC) and to
3063 * host memory when this is enabled
3064 */
3065 if (adapter->flags2 & FLAG2_CRC_STRIPPING)
3066 rctl |= E1000_RCTL_SECRC;
3067
3068 /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
3069 if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
3070 u16 phy_data;
3071
3072 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
3073 phy_data &= 0xfff8;
3074 phy_data |= BIT(2);
3075 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
3076
3077 e1e_rphy(hw, 22, &phy_data);
3078 phy_data &= 0x0fff;
3079 phy_data |= BIT(14);
3080 e1e_wphy(hw, 0x10, 0x2823);
3081 e1e_wphy(hw, 0x11, 0x0003);
3082 e1e_wphy(hw, 22, phy_data);
3083 }
3084
3085 /* Setup buffer sizes */
3086 rctl &= ~E1000_RCTL_SZ_4096;
3087 rctl |= E1000_RCTL_BSEX;
3088 switch (adapter->rx_buffer_len) {
3089 case 2048:
3090 default:
3091 rctl |= E1000_RCTL_SZ_2048;
3092 rctl &= ~E1000_RCTL_BSEX;
3093 break;
3094 case 4096:
3095 rctl |= E1000_RCTL_SZ_4096;
3096 break;
3097 case 8192:
3098 rctl |= E1000_RCTL_SZ_8192;
3099 break;
3100 case 16384:
3101 rctl |= E1000_RCTL_SZ_16384;
3102 break;
3103 }
3104
3105 /* Enable Extended Status in all Receive Descriptors */
3106 rfctl = er32(RFCTL);
3107 rfctl |= E1000_RFCTL_EXTEN;
3108 ew32(RFCTL, rfctl);
3109
3110 /* 82571 and greater support packet-split where the protocol
3111 * header is placed in skb->data and the packet data is
3112 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
3113 * In the case of a non-split, skb->data is linearly filled,
3114 * followed by the page buffers. Therefore, skb->data is
3115 * sized to hold the largest protocol header.
3116 *
3117 * allocations using alloc_page take too long for regular MTU
3118 * so only enable packet split for jumbo frames
3119 *
3120 * Using pages when the page size is greater than 16k wastes
3121 * a lot of memory, since we allocate 3 pages at all times
3122 * per packet.
3123 */
3124 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
3125 if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
3126 adapter->rx_ps_pages = pages;
3127 else
3128 adapter->rx_ps_pages = 0;
3129
3130 if (adapter->rx_ps_pages) {
3131 u32 psrctl = 0;
3132
3133 /* Enable Packet split descriptors */
3134 rctl |= E1000_RCTL_DTYP_PS;
3135
3136 psrctl |= adapter->rx_ps_bsize0 >> E1000_PSRCTL_BSIZE0_SHIFT;
3137
3138 switch (adapter->rx_ps_pages) {
3139 case 3:
3140 psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE3_SHIFT;
3141 /* fall-through */
3142 case 2:
3143 psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE2_SHIFT;
3144 /* fall-through */
3145 case 1:
3146 psrctl |= PAGE_SIZE >> E1000_PSRCTL_BSIZE1_SHIFT;
3147 break;
3148 }
3149
3150 ew32(PSRCTL, psrctl);
3151 }
3152
3153 /* This is useful for sniffing bad packets. */
3154 if (adapter->netdev->features & NETIF_F_RXALL) {
3155 /* UPE and MPE will be handled by normal PROMISC logic
3156 * in e1000e_set_rx_mode
3157 */
3158 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3159 E1000_RCTL_BAM | /* RX All Bcast Pkts */
3160 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3161
3162 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3163 E1000_RCTL_DPF | /* Allow filtered pause */
3164 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3165 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3166 * and that breaks VLANs.
3167 */
3168 }
3169
3170 ew32(RCTL, rctl);
3171 /* just started the receive unit, no need to restart */
3172 adapter->flags &= ~FLAG_RESTART_NOW;
3173}
3174
3175/**
3176 * e1000_configure_rx - Configure Receive Unit after Reset
3177 * @adapter: board private structure
3178 *
3179 * Configure the Rx unit of the MAC after a reset.
3180 **/
3181static void e1000_configure_rx(struct e1000_adapter *adapter)
3182{
3183 struct e1000_hw *hw = &adapter->hw;
3184 struct e1000_ring *rx_ring = adapter->rx_ring;
3185 u64 rdba;
3186 u32 rdlen, rctl, rxcsum, ctrl_ext;
3187
3188 if (adapter->rx_ps_pages) {
3189 /* this is a 32 byte descriptor */
3190 rdlen = rx_ring->count *
3191 sizeof(union e1000_rx_desc_packet_split);
3192 adapter->clean_rx = e1000_clean_rx_irq_ps;
3193 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3194 } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3195 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3196 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3197 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3198 } else {
3199 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3200 adapter->clean_rx = e1000_clean_rx_irq;
3201 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3202 }
3203
3204 /* disable receives while setting up the descriptors */
3205 rctl = er32(RCTL);
3206 if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3207 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3208 e1e_flush();
3209 usleep_range(10000, 20000);
3210
3211 if (adapter->flags2 & FLAG2_DMA_BURST) {
3212 /* set the writeback threshold (only takes effect if the RDTR
3213 * is set). set GRAN=1 and write back up to 0x4 worth, and
3214 * enable prefetching of 0x20 Rx descriptors
3215 * granularity = 01
3216 * wthresh = 04,
3217 * hthresh = 04,
3218 * pthresh = 0x20
3219 */
3220 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3221 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3222
3223 /* override the delay timers for enabling bursting, only if
3224 * the value was not set by the user via module options
3225 */
3226 if (adapter->rx_int_delay == DEFAULT_RDTR)
3227 adapter->rx_int_delay = BURST_RDTR;
3228 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
3229 adapter->rx_abs_int_delay = BURST_RADV;
3230 }
3231
3232 /* set the Receive Delay Timer Register */
3233 ew32(RDTR, adapter->rx_int_delay);
3234
3235 /* irq moderation */
3236 ew32(RADV, adapter->rx_abs_int_delay);
3237 if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3238 e1000e_write_itr(adapter, adapter->itr);
3239
3240 ctrl_ext = er32(CTRL_EXT);
3241 /* Auto-Mask interrupts upon ICR access */
3242 ctrl_ext |= E1000_CTRL_EXT_IAME;
3243 ew32(IAM, 0xffffffff);
3244 ew32(CTRL_EXT, ctrl_ext);
3245 e1e_flush();
3246
3247 /* Setup the HW Rx Head and Tail Descriptor Pointers and
3248 * the Base and Length of the Rx Descriptor Ring
3249 */
3250 rdba = rx_ring->dma;
3251 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
3252 ew32(RDBAH(0), (rdba >> 32));
3253 ew32(RDLEN(0), rdlen);
3254 ew32(RDH(0), 0);
3255 ew32(RDT(0), 0);
3256 rx_ring->head = adapter->hw.hw_addr + E1000_RDH(0);
3257 rx_ring->tail = adapter->hw.hw_addr + E1000_RDT(0);
3258
3259 writel(0, rx_ring->head);
3260 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
3261 e1000e_update_rdt_wa(rx_ring, 0);
3262 else
3263 writel(0, rx_ring->tail);
3264
3265 /* Enable Receive Checksum Offload for TCP and UDP */
3266 rxcsum = er32(RXCSUM);
3267 if (adapter->netdev->features & NETIF_F_RXCSUM)
3268 rxcsum |= E1000_RXCSUM_TUOFL;
3269 else
3270 rxcsum &= ~E1000_RXCSUM_TUOFL;
3271 ew32(RXCSUM, rxcsum);
3272
3273 /* With jumbo frames, excessive C-state transition latencies result
3274 * in dropped transactions.
3275 */
3276 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3277 u32 lat =
3278 ((er32(PBA) & E1000_PBA_RXA_MASK) * 1024 -
3279 adapter->max_frame_size) * 8 / 1000;
3280
3281 if (adapter->flags & FLAG_IS_ICH) {
3282 u32 rxdctl = er32(RXDCTL(0));
3283
3284 ew32(RXDCTL(0), rxdctl | 0x3);
3285 }
3286
3287 pm_qos_update_request(&adapter->pm_qos_req, lat);
3288 } else {
3289 pm_qos_update_request(&adapter->pm_qos_req,
3290 PM_QOS_DEFAULT_VALUE);
3291 }
3292
3293 /* Enable Receives */
3294 ew32(RCTL, rctl);
3295}
3296
3297/**
3298 * e1000e_write_mc_addr_list - write multicast addresses to MTA
3299 * @netdev: network interface device structure
3300 *
3301 * Writes multicast address list to the MTA hash table.
3302 * Returns: -ENOMEM on failure
3303 * 0 on no addresses written
3304 * X on writing X addresses to MTA
3305 */
3306static int e1000e_write_mc_addr_list(struct net_device *netdev)
3307{
3308 struct e1000_adapter *adapter = netdev_priv(netdev);
3309 struct e1000_hw *hw = &adapter->hw;
3310 struct netdev_hw_addr *ha;
3311 u8 *mta_list;
3312 int i;
3313
3314 if (netdev_mc_empty(netdev)) {
3315 /* nothing to program, so clear mc list */
3316 hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3317 return 0;
3318 }
3319
3320 mta_list = kzalloc(netdev_mc_count(netdev) * ETH_ALEN, GFP_ATOMIC);
3321 if (!mta_list)
3322 return -ENOMEM;
3323
3324 /* update_mc_addr_list expects a packed array of only addresses. */
3325 i = 0;
3326 netdev_for_each_mc_addr(ha, netdev)
3327 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3328
3329 hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3330 kfree(mta_list);
3331
3332 return netdev_mc_count(netdev);
3333}
3334
3335/**
3336 * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3337 * @netdev: network interface device structure
3338 *
3339 * Writes unicast address list to the RAR table.
3340 * Returns: -ENOMEM on failure/insufficient address space
3341 * 0 on no addresses written
3342 * X on writing X addresses to the RAR table
3343 **/
3344static int e1000e_write_uc_addr_list(struct net_device *netdev)
3345{
3346 struct e1000_adapter *adapter = netdev_priv(netdev);
3347 struct e1000_hw *hw = &adapter->hw;
3348 unsigned int rar_entries;
3349 int count = 0;
3350
3351 rar_entries = hw->mac.ops.rar_get_count(hw);
3352
3353 /* save a rar entry for our hardware address */
3354 rar_entries--;
3355
3356 /* save a rar entry for the LAA workaround */
3357 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3358 rar_entries--;
3359
3360 /* return ENOMEM indicating insufficient memory for addresses */
3361 if (netdev_uc_count(netdev) > rar_entries)
3362 return -ENOMEM;
3363
3364 if (!netdev_uc_empty(netdev) && rar_entries) {
3365 struct netdev_hw_addr *ha;
3366
3367 /* write the addresses in reverse order to avoid write
3368 * combining
3369 */
3370 netdev_for_each_uc_addr(ha, netdev) {
3371 int ret_val;
3372
3373 if (!rar_entries)
3374 break;
3375 ret_val = hw->mac.ops.rar_set(hw, ha->addr, rar_entries--);
3376 if (ret_val < 0)
3377 return -ENOMEM;
3378 count++;
3379 }
3380 }
3381
3382 /* zero out the remaining RAR entries not used above */
3383 for (; rar_entries > 0; rar_entries--) {
3384 ew32(RAH(rar_entries), 0);
3385 ew32(RAL(rar_entries), 0);
3386 }
3387 e1e_flush();
3388
3389 return count;
3390}
3391
3392/**
3393 * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3394 * @netdev: network interface device structure
3395 *
3396 * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3397 * address list or the network interface flags are updated. This routine is
3398 * responsible for configuring the hardware for proper unicast, multicast,
3399 * promiscuous mode, and all-multi behavior.
3400 **/
3401static void e1000e_set_rx_mode(struct net_device *netdev)
3402{
3403 struct e1000_adapter *adapter = netdev_priv(netdev);
3404 struct e1000_hw *hw = &adapter->hw;
3405 u32 rctl;
3406
3407 if (pm_runtime_suspended(netdev->dev.parent))
3408 return;
3409
3410 /* Check for Promiscuous and All Multicast modes */
3411 rctl = er32(RCTL);
3412
3413 /* clear the affected bits */
3414 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3415
3416 if (netdev->flags & IFF_PROMISC) {
3417 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3418 /* Do not hardware filter VLANs in promisc mode */
3419 e1000e_vlan_filter_disable(adapter);
3420 } else {
3421 int count;
3422
3423 if (netdev->flags & IFF_ALLMULTI) {
3424 rctl |= E1000_RCTL_MPE;
3425 } else {
3426 /* Write addresses to the MTA, if the attempt fails
3427 * then we should just turn on promiscuous mode so
3428 * that we can at least receive multicast traffic
3429 */
3430 count = e1000e_write_mc_addr_list(netdev);
3431 if (count < 0)
3432 rctl |= E1000_RCTL_MPE;
3433 }
3434 e1000e_vlan_filter_enable(adapter);
3435 /* Write addresses to available RAR registers, if there is not
3436 * sufficient space to store all the addresses then enable
3437 * unicast promiscuous mode
3438 */
3439 count = e1000e_write_uc_addr_list(netdev);
3440 if (count < 0)
3441 rctl |= E1000_RCTL_UPE;
3442 }
3443
3444 ew32(RCTL, rctl);
3445
3446 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
3447 e1000e_vlan_strip_enable(adapter);
3448 else
3449 e1000e_vlan_strip_disable(adapter);
3450}
3451
3452static void e1000e_setup_rss_hash(struct e1000_adapter *adapter)
3453{
3454 struct e1000_hw *hw = &adapter->hw;
3455 u32 mrqc, rxcsum;
3456 u32 rss_key[10];
3457 int i;
3458
3459 netdev_rss_key_fill(rss_key, sizeof(rss_key));
3460 for (i = 0; i < 10; i++)
3461 ew32(RSSRK(i), rss_key[i]);
3462
3463 /* Direct all traffic to queue 0 */
3464 for (i = 0; i < 32; i++)
3465 ew32(RETA(i), 0);
3466
3467 /* Disable raw packet checksumming so that RSS hash is placed in
3468 * descriptor on writeback.
3469 */
3470 rxcsum = er32(RXCSUM);
3471 rxcsum |= E1000_RXCSUM_PCSD;
3472
3473 ew32(RXCSUM, rxcsum);
3474
3475 mrqc = (E1000_MRQC_RSS_FIELD_IPV4 |
3476 E1000_MRQC_RSS_FIELD_IPV4_TCP |
3477 E1000_MRQC_RSS_FIELD_IPV6 |
3478 E1000_MRQC_RSS_FIELD_IPV6_TCP |
3479 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
3480
3481 ew32(MRQC, mrqc);
3482}
3483
3484/**
3485 * e1000e_get_base_timinca - get default SYSTIM time increment attributes
3486 * @adapter: board private structure
3487 * @timinca: pointer to returned time increment attributes
3488 *
3489 * Get attributes for incrementing the System Time Register SYSTIML/H at
3490 * the default base frequency, and set the cyclecounter shift value.
3491 **/
3492s32 e1000e_get_base_timinca(struct e1000_adapter *adapter, u32 *timinca)
3493{
3494 struct e1000_hw *hw = &adapter->hw;
3495 u32 incvalue, incperiod, shift;
3496
3497 /* Make sure clock is enabled on I217/I218/I219 before checking
3498 * the frequency
3499 */
3500 if (((hw->mac.type == e1000_pch_lpt) ||
3501 (hw->mac.type == e1000_pch_spt)) &&
3502 !(er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) &&
3503 !(er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_ENABLED)) {
3504 u32 fextnvm7 = er32(FEXTNVM7);
3505
3506 if (!(fextnvm7 & BIT(0))) {
3507 ew32(FEXTNVM7, fextnvm7 | BIT(0));
3508 e1e_flush();
3509 }
3510 }
3511
3512 switch (hw->mac.type) {
3513 case e1000_pch2lan:
3514 case e1000_pch_lpt:
3515 if (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI) {
3516 /* Stable 96MHz frequency */
3517 incperiod = INCPERIOD_96MHz;
3518 incvalue = INCVALUE_96MHz;
3519 shift = INCVALUE_SHIFT_96MHz;
3520 adapter->cc.shift = shift + INCPERIOD_SHIFT_96MHz;
3521 } else {
3522 /* Stable 25MHz frequency */
3523 incperiod = INCPERIOD_25MHz;
3524 incvalue = INCVALUE_25MHz;
3525 shift = INCVALUE_SHIFT_25MHz;
3526 adapter->cc.shift = shift;
3527 }
3528 break;
3529 case e1000_pch_spt:
3530 if (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI) {
3531 /* Stable 24MHz frequency */
3532 incperiod = INCPERIOD_24MHz;
3533 incvalue = INCVALUE_24MHz;
3534 shift = INCVALUE_SHIFT_24MHz;
3535 adapter->cc.shift = shift;
3536 break;
3537 }
3538 return -EINVAL;
3539 case e1000_82574:
3540 case e1000_82583:
3541 /* Stable 25MHz frequency */
3542 incperiod = INCPERIOD_25MHz;
3543 incvalue = INCVALUE_25MHz;
3544 shift = INCVALUE_SHIFT_25MHz;
3545 adapter->cc.shift = shift;
3546 break;
3547 default:
3548 return -EINVAL;
3549 }
3550
3551 *timinca = ((incperiod << E1000_TIMINCA_INCPERIOD_SHIFT) |
3552 ((incvalue << shift) & E1000_TIMINCA_INCVALUE_MASK));
3553
3554 return 0;
3555}
3556
3557/**
3558 * e1000e_config_hwtstamp - configure the hwtstamp registers and enable/disable
3559 * @adapter: board private structure
3560 *
3561 * Outgoing time stamping can be enabled and disabled. Play nice and
3562 * disable it when requested, although it shouldn't cause any overhead
3563 * when no packet needs it. At most one packet in the queue may be
3564 * marked for time stamping, otherwise it would be impossible to tell
3565 * for sure to which packet the hardware time stamp belongs.
3566 *
3567 * Incoming time stamping has to be configured via the hardware filters.
3568 * Not all combinations are supported, in particular event type has to be
3569 * specified. Matching the kind of event packet is not supported, with the
3570 * exception of "all V2 events regardless of level 2 or 4".
3571 **/
3572static int e1000e_config_hwtstamp(struct e1000_adapter *adapter,
3573 struct hwtstamp_config *config)
3574{
3575 struct e1000_hw *hw = &adapter->hw;
3576 u32 tsync_tx_ctl = E1000_TSYNCTXCTL_ENABLED;
3577 u32 tsync_rx_ctl = E1000_TSYNCRXCTL_ENABLED;
3578 u32 rxmtrl = 0;
3579 u16 rxudp = 0;
3580 bool is_l4 = false;
3581 bool is_l2 = false;
3582 u32 regval;
3583
3584 if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP))
3585 return -EINVAL;
3586
3587 /* flags reserved for future extensions - must be zero */
3588 if (config->flags)
3589 return -EINVAL;
3590
3591 switch (config->tx_type) {
3592 case HWTSTAMP_TX_OFF:
3593 tsync_tx_ctl = 0;
3594 break;
3595 case HWTSTAMP_TX_ON:
3596 break;
3597 default:
3598 return -ERANGE;
3599 }
3600
3601 switch (config->rx_filter) {
3602 case HWTSTAMP_FILTER_NONE:
3603 tsync_rx_ctl = 0;
3604 break;
3605 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3606 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3607 rxmtrl = E1000_RXMTRL_PTP_V1_SYNC_MESSAGE;
3608 is_l4 = true;
3609 break;
3610 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3611 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3612 rxmtrl = E1000_RXMTRL_PTP_V1_DELAY_REQ_MESSAGE;
3613 is_l4 = true;
3614 break;
3615 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
3616 /* Also time stamps V2 L2 Path Delay Request/Response */
3617 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3618 rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3619 is_l2 = true;
3620 break;
3621 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
3622 /* Also time stamps V2 L2 Path Delay Request/Response. */
3623 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3624 rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3625 is_l2 = true;
3626 break;
3627 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3628 /* Hardware cannot filter just V2 L4 Sync messages;
3629 * fall-through to V2 (both L2 and L4) Sync.
3630 */
3631 case HWTSTAMP_FILTER_PTP_V2_SYNC:
3632 /* Also time stamps V2 Path Delay Request/Response. */
3633 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3634 rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3635 is_l2 = true;
3636 is_l4 = true;
3637 break;
3638 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3639 /* Hardware cannot filter just V2 L4 Delay Request messages;
3640 * fall-through to V2 (both L2 and L4) Delay Request.
3641 */
3642 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
3643 /* Also time stamps V2 Path Delay Request/Response. */
3644 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3645 rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3646 is_l2 = true;
3647 is_l4 = true;
3648 break;
3649 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3650 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
3651 /* Hardware cannot filter just V2 L4 or L2 Event messages;
3652 * fall-through to all V2 (both L2 and L4) Events.
3653 */
3654 case HWTSTAMP_FILTER_PTP_V2_EVENT:
3655 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_EVENT_V2;
3656 config->rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
3657 is_l2 = true;
3658 is_l4 = true;
3659 break;
3660 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3661 /* For V1, the hardware can only filter Sync messages or
3662 * Delay Request messages but not both so fall-through to
3663 * time stamp all packets.
3664 */
3665 case HWTSTAMP_FILTER_ALL:
3666 is_l2 = true;
3667 is_l4 = true;
3668 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_ALL;
3669 config->rx_filter = HWTSTAMP_FILTER_ALL;
3670 break;
3671 default:
3672 return -ERANGE;
3673 }
3674
3675 adapter->hwtstamp_config = *config;
3676
3677 /* enable/disable Tx h/w time stamping */
3678 regval = er32(TSYNCTXCTL);
3679 regval &= ~E1000_TSYNCTXCTL_ENABLED;
3680 regval |= tsync_tx_ctl;
3681 ew32(TSYNCTXCTL, regval);
3682 if ((er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) !=
3683 (regval & E1000_TSYNCTXCTL_ENABLED)) {
3684 e_err("Timesync Tx Control register not set as expected\n");
3685 return -EAGAIN;
3686 }
3687
3688 /* enable/disable Rx h/w time stamping */
3689 regval = er32(TSYNCRXCTL);
3690 regval &= ~(E1000_TSYNCRXCTL_ENABLED | E1000_TSYNCRXCTL_TYPE_MASK);
3691 regval |= tsync_rx_ctl;
3692 ew32(TSYNCRXCTL, regval);
3693 if ((er32(TSYNCRXCTL) & (E1000_TSYNCRXCTL_ENABLED |
3694 E1000_TSYNCRXCTL_TYPE_MASK)) !=
3695 (regval & (E1000_TSYNCRXCTL_ENABLED |
3696 E1000_TSYNCRXCTL_TYPE_MASK))) {
3697 e_err("Timesync Rx Control register not set as expected\n");
3698 return -EAGAIN;
3699 }
3700
3701 /* L2: define ethertype filter for time stamped packets */
3702 if (is_l2)
3703 rxmtrl |= ETH_P_1588;
3704
3705 /* define which PTP packets get time stamped */
3706 ew32(RXMTRL, rxmtrl);
3707
3708 /* Filter by destination port */
3709 if (is_l4) {
3710 rxudp = PTP_EV_PORT;
3711 cpu_to_be16s(&rxudp);
3712 }
3713 ew32(RXUDP, rxudp);
3714
3715 e1e_flush();
3716
3717 /* Clear TSYNCRXCTL_VALID & TSYNCTXCTL_VALID bit */
3718 er32(RXSTMPH);
3719 er32(TXSTMPH);
3720
3721 return 0;
3722}
3723
3724/**
3725 * e1000_configure - configure the hardware for Rx and Tx
3726 * @adapter: private board structure
3727 **/
3728static void e1000_configure(struct e1000_adapter *adapter)
3729{
3730 struct e1000_ring *rx_ring = adapter->rx_ring;
3731
3732 e1000e_set_rx_mode(adapter->netdev);
3733
3734 e1000_restore_vlan(adapter);
3735 e1000_init_manageability_pt(adapter);
3736
3737 e1000_configure_tx(adapter);
3738
3739 if (adapter->netdev->features & NETIF_F_RXHASH)
3740 e1000e_setup_rss_hash(adapter);
3741 e1000_setup_rctl(adapter);
3742 e1000_configure_rx(adapter);
3743 adapter->alloc_rx_buf(rx_ring, e1000_desc_unused(rx_ring), GFP_KERNEL);
3744}
3745
3746/**
3747 * e1000e_power_up_phy - restore link in case the phy was powered down
3748 * @adapter: address of board private structure
3749 *
3750 * The phy may be powered down to save power and turn off link when the
3751 * driver is unloaded and wake on lan is not enabled (among others)
3752 * *** this routine MUST be followed by a call to e1000e_reset ***
3753 **/
3754void e1000e_power_up_phy(struct e1000_adapter *adapter)
3755{
3756 if (adapter->hw.phy.ops.power_up)
3757 adapter->hw.phy.ops.power_up(&adapter->hw);
3758
3759 adapter->hw.mac.ops.setup_link(&adapter->hw);
3760}
3761
3762/**
3763 * e1000_power_down_phy - Power down the PHY
3764 *
3765 * Power down the PHY so no link is implied when interface is down.
3766 * The PHY cannot be powered down if management or WoL is active.
3767 */
3768static void e1000_power_down_phy(struct e1000_adapter *adapter)
3769{
3770 if (adapter->hw.phy.ops.power_down)
3771 adapter->hw.phy.ops.power_down(&adapter->hw);
3772}
3773
3774/**
3775 * e1000_flush_tx_ring - remove all descriptors from the tx_ring
3776 *
3777 * We want to clear all pending descriptors from the TX ring.
3778 * zeroing happens when the HW reads the regs. We assign the ring itself as
3779 * the data of the next descriptor. We don't care about the data we are about
3780 * to reset the HW.
3781 */
3782static void e1000_flush_tx_ring(struct e1000_adapter *adapter)
3783{
3784 struct e1000_hw *hw = &adapter->hw;
3785 struct e1000_ring *tx_ring = adapter->tx_ring;
3786 struct e1000_tx_desc *tx_desc = NULL;
3787 u32 tdt, tctl, txd_lower = E1000_TXD_CMD_IFCS;
3788 u16 size = 512;
3789
3790 tctl = er32(TCTL);
3791 ew32(TCTL, tctl | E1000_TCTL_EN);
3792 tdt = er32(TDT(0));
3793 BUG_ON(tdt != tx_ring->next_to_use);
3794 tx_desc = E1000_TX_DESC(*tx_ring, tx_ring->next_to_use);
3795 tx_desc->buffer_addr = tx_ring->dma;
3796
3797 tx_desc->lower.data = cpu_to_le32(txd_lower | size);
3798 tx_desc->upper.data = 0;
3799 /* flush descriptors to memory before notifying the HW */
3800 wmb();
3801 tx_ring->next_to_use++;
3802 if (tx_ring->next_to_use == tx_ring->count)
3803 tx_ring->next_to_use = 0;
3804 ew32(TDT(0), tx_ring->next_to_use);
3805 mmiowb();
3806 usleep_range(200, 250);
3807}
3808
3809/**
3810 * e1000_flush_rx_ring - remove all descriptors from the rx_ring
3811 *
3812 * Mark all descriptors in the RX ring as consumed and disable the rx ring
3813 */
3814static void e1000_flush_rx_ring(struct e1000_adapter *adapter)
3815{
3816 u32 rctl, rxdctl;
3817 struct e1000_hw *hw = &adapter->hw;
3818
3819 rctl = er32(RCTL);
3820 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3821 e1e_flush();
3822 usleep_range(100, 150);
3823
3824 rxdctl = er32(RXDCTL(0));
3825 /* zero the lower 14 bits (prefetch and host thresholds) */
3826 rxdctl &= 0xffffc000;
3827
3828 /* update thresholds: prefetch threshold to 31, host threshold to 1
3829 * and make sure the granularity is "descriptors" and not "cache lines"
3830 */
3831 rxdctl |= (0x1F | BIT(8) | E1000_RXDCTL_THRESH_UNIT_DESC);
3832
3833 ew32(RXDCTL(0), rxdctl);
3834 /* momentarily enable the RX ring for the changes to take effect */
3835 ew32(RCTL, rctl | E1000_RCTL_EN);
3836 e1e_flush();
3837 usleep_range(100, 150);
3838 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3839}
3840
3841/**
3842 * e1000_flush_desc_rings - remove all descriptors from the descriptor rings
3843 *
3844 * In i219, the descriptor rings must be emptied before resetting the HW
3845 * or before changing the device state to D3 during runtime (runtime PM).
3846 *
3847 * Failure to do this will cause the HW to enter a unit hang state which can
3848 * only be released by PCI reset on the device
3849 *
3850 */
3851
3852static void e1000_flush_desc_rings(struct e1000_adapter *adapter)
3853{
3854 u16 hang_state;
3855 u32 fext_nvm11, tdlen;
3856 struct e1000_hw *hw = &adapter->hw;
3857
3858 /* First, disable MULR fix in FEXTNVM11 */
3859 fext_nvm11 = er32(FEXTNVM11);
3860 fext_nvm11 |= E1000_FEXTNVM11_DISABLE_MULR_FIX;
3861 ew32(FEXTNVM11, fext_nvm11);
3862 /* do nothing if we're not in faulty state, or if the queue is empty */
3863 tdlen = er32(TDLEN(0));
3864 pci_read_config_word(adapter->pdev, PCICFG_DESC_RING_STATUS,
3865 &hang_state);
3866 if (!(hang_state & FLUSH_DESC_REQUIRED) || !tdlen)
3867 return;
3868 e1000_flush_tx_ring(adapter);
3869 /* recheck, maybe the fault is caused by the rx ring */
3870 pci_read_config_word(adapter->pdev, PCICFG_DESC_RING_STATUS,
3871 &hang_state);
3872 if (hang_state & FLUSH_DESC_REQUIRED)
3873 e1000_flush_rx_ring(adapter);
3874}
3875
3876/**
3877 * e1000e_systim_reset - reset the timesync registers after a hardware reset
3878 * @adapter: board private structure
3879 *
3880 * When the MAC is reset, all hardware bits for timesync will be reset to the
3881 * default values. This function will restore the settings last in place.
3882 * Since the clock SYSTIME registers are reset, we will simply restore the
3883 * cyclecounter to the kernel real clock time.
3884 **/
3885static void e1000e_systim_reset(struct e1000_adapter *adapter)
3886{
3887 struct ptp_clock_info *info = &adapter->ptp_clock_info;
3888 struct e1000_hw *hw = &adapter->hw;
3889 unsigned long flags;
3890 u32 timinca;
3891 s32 ret_val;
3892
3893 if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP))
3894 return;
3895
3896 if (info->adjfreq) {
3897 /* restore the previous ptp frequency delta */
3898 ret_val = info->adjfreq(info, adapter->ptp_delta);
3899 } else {
3900 /* set the default base frequency if no adjustment possible */
3901 ret_val = e1000e_get_base_timinca(adapter, &timinca);
3902 if (!ret_val)
3903 ew32(TIMINCA, timinca);
3904 }
3905
3906 if (ret_val) {
3907 dev_warn(&adapter->pdev->dev,
3908 "Failed to restore TIMINCA clock rate delta: %d\n",
3909 ret_val);
3910 return;
3911 }
3912
3913 /* reset the systim ns time counter */
3914 spin_lock_irqsave(&adapter->systim_lock, flags);
3915 timecounter_init(&adapter->tc, &adapter->cc,
3916 ktime_to_ns(ktime_get_real()));
3917 spin_unlock_irqrestore(&adapter->systim_lock, flags);
3918
3919 /* restore the previous hwtstamp configuration settings */
3920 e1000e_config_hwtstamp(adapter, &adapter->hwtstamp_config);
3921}
3922
3923/**
3924 * e1000e_reset - bring the hardware into a known good state
3925 *
3926 * This function boots the hardware and enables some settings that
3927 * require a configuration cycle of the hardware - those cannot be
3928 * set/changed during runtime. After reset the device needs to be
3929 * properly configured for Rx, Tx etc.
3930 */
3931void e1000e_reset(struct e1000_adapter *adapter)
3932{
3933 struct e1000_mac_info *mac = &adapter->hw.mac;
3934 struct e1000_fc_info *fc = &adapter->hw.fc;
3935 struct e1000_hw *hw = &adapter->hw;
3936 u32 tx_space, min_tx_space, min_rx_space;
3937 u32 pba = adapter->pba;
3938 u16 hwm;
3939
3940 /* reset Packet Buffer Allocation to default */
3941 ew32(PBA, pba);
3942
3943 if (adapter->max_frame_size > (VLAN_ETH_FRAME_LEN + ETH_FCS_LEN)) {
3944 /* To maintain wire speed transmits, the Tx FIFO should be
3945 * large enough to accommodate two full transmit packets,
3946 * rounded up to the next 1KB and expressed in KB. Likewise,
3947 * the Rx FIFO should be large enough to accommodate at least
3948 * one full receive packet and is similarly rounded up and
3949 * expressed in KB.
3950 */
3951 pba = er32(PBA);
3952 /* upper 16 bits has Tx packet buffer allocation size in KB */
3953 tx_space = pba >> 16;
3954 /* lower 16 bits has Rx packet buffer allocation size in KB */
3955 pba &= 0xffff;
3956 /* the Tx fifo also stores 16 bytes of information about the Tx
3957 * but don't include ethernet FCS because hardware appends it
3958 */
3959 min_tx_space = (adapter->max_frame_size +
3960 sizeof(struct e1000_tx_desc) - ETH_FCS_LEN) * 2;
3961 min_tx_space = ALIGN(min_tx_space, 1024);
3962 min_tx_space >>= 10;
3963 /* software strips receive CRC, so leave room for it */
3964 min_rx_space = adapter->max_frame_size;
3965 min_rx_space = ALIGN(min_rx_space, 1024);
3966 min_rx_space >>= 10;
3967
3968 /* If current Tx allocation is less than the min Tx FIFO size,
3969 * and the min Tx FIFO size is less than the current Rx FIFO
3970 * allocation, take space away from current Rx allocation
3971 */
3972 if ((tx_space < min_tx_space) &&
3973 ((min_tx_space - tx_space) < pba)) {
3974 pba -= min_tx_space - tx_space;
3975
3976 /* if short on Rx space, Rx wins and must trump Tx
3977 * adjustment
3978 */
3979 if (pba < min_rx_space)
3980 pba = min_rx_space;
3981 }
3982
3983 ew32(PBA, pba);
3984 }
3985
3986 /* flow control settings
3987 *
3988 * The high water mark must be low enough to fit one full frame
3989 * (or the size used for early receive) above it in the Rx FIFO.
3990 * Set it to the lower of:
3991 * - 90% of the Rx FIFO size, and
3992 * - the full Rx FIFO size minus one full frame
3993 */
3994 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3995 fc->pause_time = 0xFFFF;
3996 else
3997 fc->pause_time = E1000_FC_PAUSE_TIME;
3998 fc->send_xon = true;
3999 fc->current_mode = fc->requested_mode;
4000
4001 switch (hw->mac.type) {
4002 case e1000_ich9lan:
4003 case e1000_ich10lan:
4004 if (adapter->netdev->mtu > ETH_DATA_LEN) {
4005 pba = 14;
4006 ew32(PBA, pba);
4007 fc->high_water = 0x2800;
4008 fc->low_water = fc->high_water - 8;
4009 break;
4010 }
4011 /* fall-through */
4012 default:
4013 hwm = min(((pba << 10) * 9 / 10),
4014 ((pba << 10) - adapter->max_frame_size));
4015
4016 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
4017 fc->low_water = fc->high_water - 8;
4018 break;
4019 case e1000_pchlan:
4020 /* Workaround PCH LOM adapter hangs with certain network
4021 * loads. If hangs persist, try disabling Tx flow control.
4022 */
4023 if (adapter->netdev->mtu > ETH_DATA_LEN) {
4024 fc->high_water = 0x3500;
4025 fc->low_water = 0x1500;
4026 } else {
4027 fc->high_water = 0x5000;
4028 fc->low_water = 0x3000;
4029 }
4030 fc->refresh_time = 0x1000;
4031 break;
4032 case e1000_pch2lan:
4033 case e1000_pch_lpt:
4034 case e1000_pch_spt:
4035 fc->refresh_time = 0x0400;
4036
4037 if (adapter->netdev->mtu <= ETH_DATA_LEN) {
4038 fc->high_water = 0x05C20;
4039 fc->low_water = 0x05048;
4040 fc->pause_time = 0x0650;
4041 break;
4042 }
4043
4044 pba = 14;
4045 ew32(PBA, pba);
4046 fc->high_water = ((pba << 10) * 9 / 10) & E1000_FCRTH_RTH;
4047 fc->low_water = ((pba << 10) * 8 / 10) & E1000_FCRTL_RTL;
4048 break;
4049 }
4050
4051 /* Alignment of Tx data is on an arbitrary byte boundary with the
4052 * maximum size per Tx descriptor limited only to the transmit
4053 * allocation of the packet buffer minus 96 bytes with an upper
4054 * limit of 24KB due to receive synchronization limitations.
4055 */
4056 adapter->tx_fifo_limit = min_t(u32, ((er32(PBA) >> 16) << 10) - 96,
4057 24 << 10);
4058
4059 /* Disable Adaptive Interrupt Moderation if 2 full packets cannot
4060 * fit in receive buffer.
4061 */
4062 if (adapter->itr_setting & 0x3) {
4063 if ((adapter->max_frame_size * 2) > (pba << 10)) {
4064 if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
4065 dev_info(&adapter->pdev->dev,
4066 "Interrupt Throttle Rate off\n");
4067 adapter->flags2 |= FLAG2_DISABLE_AIM;
4068 e1000e_write_itr(adapter, 0);
4069 }
4070 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
4071 dev_info(&adapter->pdev->dev,
4072 "Interrupt Throttle Rate on\n");
4073 adapter->flags2 &= ~FLAG2_DISABLE_AIM;
4074 adapter->itr = 20000;
4075 e1000e_write_itr(adapter, adapter->itr);
4076 }
4077 }
4078
4079 if (hw->mac.type == e1000_pch_spt)
4080 e1000_flush_desc_rings(adapter);
4081 /* Allow time for pending master requests to run */
4082 mac->ops.reset_hw(hw);
4083
4084 /* For parts with AMT enabled, let the firmware know
4085 * that the network interface is in control
4086 */
4087 if (adapter->flags & FLAG_HAS_AMT)
4088 e1000e_get_hw_control(adapter);
4089
4090 ew32(WUC, 0);
4091
4092 if (mac->ops.init_hw(hw))
4093 e_err("Hardware Error\n");
4094
4095 e1000_update_mng_vlan(adapter);
4096
4097 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
4098 ew32(VET, ETH_P_8021Q);
4099
4100 e1000e_reset_adaptive(hw);
4101
4102 /* restore systim and hwtstamp settings */
4103 e1000e_systim_reset(adapter);
4104
4105 /* Set EEE advertisement as appropriate */
4106 if (adapter->flags2 & FLAG2_HAS_EEE) {
4107 s32 ret_val;
4108 u16 adv_addr;
4109
4110 switch (hw->phy.type) {
4111 case e1000_phy_82579:
4112 adv_addr = I82579_EEE_ADVERTISEMENT;
4113 break;
4114 case e1000_phy_i217:
4115 adv_addr = I217_EEE_ADVERTISEMENT;
4116 break;
4117 default:
4118 dev_err(&adapter->pdev->dev,
4119 "Invalid PHY type setting EEE advertisement\n");
4120 return;
4121 }
4122
4123 ret_val = hw->phy.ops.acquire(hw);
4124 if (ret_val) {
4125 dev_err(&adapter->pdev->dev,
4126 "EEE advertisement - unable to acquire PHY\n");
4127 return;
4128 }
4129
4130 e1000_write_emi_reg_locked(hw, adv_addr,
4131 hw->dev_spec.ich8lan.eee_disable ?
4132 0 : adapter->eee_advert);
4133
4134 hw->phy.ops.release(hw);
4135 }
4136
4137 if (!netif_running(adapter->netdev) &&
4138 !test_bit(__E1000_TESTING, &adapter->state))
4139 e1000_power_down_phy(adapter);
4140
4141 e1000_get_phy_info(hw);
4142
4143 if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
4144 !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
4145 u16 phy_data = 0;
4146 /* speed up time to link by disabling smart power down, ignore
4147 * the return value of this function because there is nothing
4148 * different we would do if it failed
4149 */
4150 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
4151 phy_data &= ~IGP02E1000_PM_SPD;
4152 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
4153 }
4154 if (hw->mac.type == e1000_pch_spt && adapter->int_mode == 0) {
4155 u32 reg;
4156
4157 /* Fextnvm7 @ 0xe4[2] = 1 */
4158 reg = er32(FEXTNVM7);
4159 reg |= E1000_FEXTNVM7_SIDE_CLK_UNGATE;
4160 ew32(FEXTNVM7, reg);
4161 /* Fextnvm9 @ 0x5bb4[13:12] = 11 */
4162 reg = er32(FEXTNVM9);
4163 reg |= E1000_FEXTNVM9_IOSFSB_CLKGATE_DIS |
4164 E1000_FEXTNVM9_IOSFSB_CLKREQ_DIS;
4165 ew32(FEXTNVM9, reg);
4166 }
4167
4168}
4169
4170/**
4171 * e1000e_trigger_lsc - trigger an LSC interrupt
4172 * @adapter:
4173 *
4174 * Fire a link status change interrupt to start the watchdog.
4175 **/
4176static void e1000e_trigger_lsc(struct e1000_adapter *adapter)
4177{
4178 struct e1000_hw *hw = &adapter->hw;
4179
4180 if (adapter->msix_entries)
4181 ew32(ICS, E1000_ICS_OTHER);
4182 else
4183 ew32(ICS, E1000_ICS_LSC);
4184}
4185
4186void e1000e_up(struct e1000_adapter *adapter)
4187{
4188 /* hardware has been reset, we need to reload some things */
4189 e1000_configure(adapter);
4190
4191 clear_bit(__E1000_DOWN, &adapter->state);
4192
4193 if (adapter->msix_entries)
4194 e1000_configure_msix(adapter);
4195 e1000_irq_enable(adapter);
4196
4197 netif_start_queue(adapter->netdev);
4198
4199 e1000e_trigger_lsc(adapter);
4200}
4201
4202static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
4203{
4204 struct e1000_hw *hw = &adapter->hw;
4205
4206 if (!(adapter->flags2 & FLAG2_DMA_BURST))
4207 return;
4208
4209 /* flush pending descriptor writebacks to memory */
4210 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4211 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4212
4213 /* execute the writes immediately */
4214 e1e_flush();
4215
4216 /* due to rare timing issues, write to TIDV/RDTR again to ensure the
4217 * write is successful
4218 */
4219 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4220 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4221
4222 /* execute the writes immediately */
4223 e1e_flush();
4224}
4225
4226static void e1000e_update_stats(struct e1000_adapter *adapter);
4227
4228/**
4229 * e1000e_down - quiesce the device and optionally reset the hardware
4230 * @adapter: board private structure
4231 * @reset: boolean flag to reset the hardware or not
4232 */
4233void e1000e_down(struct e1000_adapter *adapter, bool reset)
4234{
4235 struct net_device *netdev = adapter->netdev;
4236 struct e1000_hw *hw = &adapter->hw;
4237 u32 tctl, rctl;
4238
4239 /* signal that we're down so the interrupt handler does not
4240 * reschedule our watchdog timer
4241 */
4242 set_bit(__E1000_DOWN, &adapter->state);
4243
4244 netif_carrier_off(netdev);
4245
4246 /* disable receives in the hardware */
4247 rctl = er32(RCTL);
4248 if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
4249 ew32(RCTL, rctl & ~E1000_RCTL_EN);
4250 /* flush and sleep below */
4251
4252 netif_stop_queue(netdev);
4253
4254 /* disable transmits in the hardware */
4255 tctl = er32(TCTL);
4256 tctl &= ~E1000_TCTL_EN;
4257 ew32(TCTL, tctl);
4258
4259 /* flush both disables and wait for them to finish */
4260 e1e_flush();
4261 usleep_range(10000, 20000);
4262
4263 e1000_irq_disable(adapter);
4264
4265 napi_synchronize(&adapter->napi);
4266
4267 del_timer_sync(&adapter->watchdog_timer);
4268 del_timer_sync(&adapter->phy_info_timer);
4269
4270 spin_lock(&adapter->stats64_lock);
4271 e1000e_update_stats(adapter);
4272 spin_unlock(&adapter->stats64_lock);
4273
4274 e1000e_flush_descriptors(adapter);
4275
4276 adapter->link_speed = 0;
4277 adapter->link_duplex = 0;
4278
4279 /* Disable Si errata workaround on PCHx for jumbo frame flow */
4280 if ((hw->mac.type >= e1000_pch2lan) &&
4281 (adapter->netdev->mtu > ETH_DATA_LEN) &&
4282 e1000_lv_jumbo_workaround_ich8lan(hw, false))
4283 e_dbg("failed to disable jumbo frame workaround mode\n");
4284
4285 if (!pci_channel_offline(adapter->pdev)) {
4286 if (reset)
4287 e1000e_reset(adapter);
4288 else if (hw->mac.type == e1000_pch_spt)
4289 e1000_flush_desc_rings(adapter);
4290 }
4291 e1000_clean_tx_ring(adapter->tx_ring);
4292 e1000_clean_rx_ring(adapter->rx_ring);
4293}
4294
4295void e1000e_reinit_locked(struct e1000_adapter *adapter)
4296{
4297 might_sleep();
4298 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4299 usleep_range(1000, 2000);
4300 e1000e_down(adapter, true);
4301 e1000e_up(adapter);
4302 clear_bit(__E1000_RESETTING, &adapter->state);
4303}
4304
4305/**
4306 * e1000e_sanitize_systim - sanitize raw cycle counter reads
4307 * @hw: pointer to the HW structure
4308 * @systim: time value read, sanitized and returned
4309 *
4310 * Errata for 82574/82583 possible bad bits read from SYSTIMH/L:
4311 * check to see that the time is incrementing at a reasonable
4312 * rate and is a multiple of incvalue.
4313 **/
4314static u64 e1000e_sanitize_systim(struct e1000_hw *hw, u64 systim)
4315{
4316 u64 time_delta, rem, temp;
4317 u64 systim_next;
4318 u32 incvalue;
4319 int i;
4320
4321 incvalue = er32(TIMINCA) & E1000_TIMINCA_INCVALUE_MASK;
4322 for (i = 0; i < E1000_MAX_82574_SYSTIM_REREADS; i++) {
4323 /* latch SYSTIMH on read of SYSTIML */
4324 systim_next = (u64)er32(SYSTIML);
4325 systim_next |= (u64)er32(SYSTIMH) << 32;
4326
4327 time_delta = systim_next - systim;
4328 temp = time_delta;
4329 /* VMWare users have seen incvalue of zero, don't div / 0 */
4330 rem = incvalue ? do_div(temp, incvalue) : (time_delta != 0);
4331
4332 systim = systim_next;
4333
4334 if ((time_delta < E1000_82574_SYSTIM_EPSILON) && (rem == 0))
4335 break;
4336 }
4337
4338 return systim;
4339}
4340
4341/**
4342 * e1000e_cyclecounter_read - read raw cycle counter (used by time counter)
4343 * @cc: cyclecounter structure
4344 **/
4345static u64 e1000e_cyclecounter_read(const struct cyclecounter *cc)
4346{
4347 struct e1000_adapter *adapter = container_of(cc, struct e1000_adapter,
4348 cc);
4349 struct e1000_hw *hw = &adapter->hw;
4350 u32 systimel, systimeh;
4351 u64 systim;
4352 /* SYSTIMH latching upon SYSTIML read does not work well.
4353 * This means that if SYSTIML overflows after we read it but before
4354 * we read SYSTIMH, the value of SYSTIMH has been incremented and we
4355 * will experience a huge non linear increment in the systime value
4356 * to fix that we test for overflow and if true, we re-read systime.
4357 */
4358 systimel = er32(SYSTIML);
4359 systimeh = er32(SYSTIMH);
4360 /* Is systimel is so large that overflow is possible? */
4361 if (systimel >= (u32)0xffffffff - E1000_TIMINCA_INCVALUE_MASK) {
4362 u32 systimel_2 = er32(SYSTIML);
4363 if (systimel > systimel_2) {
4364 /* There was an overflow, read again SYSTIMH, and use
4365 * systimel_2
4366 */
4367 systimeh = er32(SYSTIMH);
4368 systimel = systimel_2;
4369 }
4370 }
4371 systim = (u64)systimel;
4372 systim |= (u64)systimeh << 32;
4373
4374 if (adapter->flags2 & FLAG2_CHECK_SYSTIM_OVERFLOW)
4375 systim = e1000e_sanitize_systim(hw, systim);
4376
4377 return systim;
4378}
4379
4380/**
4381 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
4382 * @adapter: board private structure to initialize
4383 *
4384 * e1000_sw_init initializes the Adapter private data structure.
4385 * Fields are initialized based on PCI device information and
4386 * OS network device settings (MTU size).
4387 **/
4388static int e1000_sw_init(struct e1000_adapter *adapter)
4389{
4390 struct net_device *netdev = adapter->netdev;
4391
4392 adapter->rx_buffer_len = VLAN_ETH_FRAME_LEN + ETH_FCS_LEN;
4393 adapter->rx_ps_bsize0 = 128;
4394 adapter->max_frame_size = netdev->mtu + VLAN_ETH_HLEN + ETH_FCS_LEN;
4395 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
4396 adapter->tx_ring_count = E1000_DEFAULT_TXD;
4397 adapter->rx_ring_count = E1000_DEFAULT_RXD;
4398
4399 spin_lock_init(&adapter->stats64_lock);
4400
4401 e1000e_set_interrupt_capability(adapter);
4402
4403 if (e1000_alloc_queues(adapter))
4404 return -ENOMEM;
4405
4406 /* Setup hardware time stamping cyclecounter */
4407 if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
4408 adapter->cc.read = e1000e_cyclecounter_read;
4409 adapter->cc.mask = CYCLECOUNTER_MASK(64);
4410 adapter->cc.mult = 1;
4411 /* cc.shift set in e1000e_get_base_tininca() */
4412
4413 spin_lock_init(&adapter->systim_lock);
4414 INIT_WORK(&adapter->tx_hwtstamp_work, e1000e_tx_hwtstamp_work);
4415 }
4416
4417 /* Explicitly disable IRQ since the NIC can be in any state. */
4418 e1000_irq_disable(adapter);
4419
4420 set_bit(__E1000_DOWN, &adapter->state);
4421 return 0;
4422}
4423
4424/**
4425 * e1000_intr_msi_test - Interrupt Handler
4426 * @irq: interrupt number
4427 * @data: pointer to a network interface device structure
4428 **/
4429static irqreturn_t e1000_intr_msi_test(int __always_unused irq, void *data)
4430{
4431 struct net_device *netdev = data;
4432 struct e1000_adapter *adapter = netdev_priv(netdev);
4433 struct e1000_hw *hw = &adapter->hw;
4434 u32 icr = er32(ICR);
4435
4436 e_dbg("icr is %08X\n", icr);
4437 if (icr & E1000_ICR_RXSEQ) {
4438 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
4439 /* Force memory writes to complete before acknowledging the
4440 * interrupt is handled.
4441 */
4442 wmb();
4443 }
4444
4445 return IRQ_HANDLED;
4446}
4447
4448/**
4449 * e1000_test_msi_interrupt - Returns 0 for successful test
4450 * @adapter: board private struct
4451 *
4452 * code flow taken from tg3.c
4453 **/
4454static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
4455{
4456 struct net_device *netdev = adapter->netdev;
4457 struct e1000_hw *hw = &adapter->hw;
4458 int err;
4459
4460 /* poll_enable hasn't been called yet, so don't need disable */
4461 /* clear any pending events */
4462 er32(ICR);
4463
4464 /* free the real vector and request a test handler */
4465 e1000_free_irq(adapter);
4466 e1000e_reset_interrupt_capability(adapter);
4467
4468 /* Assume that the test fails, if it succeeds then the test
4469 * MSI irq handler will unset this flag
4470 */
4471 adapter->flags |= FLAG_MSI_TEST_FAILED;
4472
4473 err = pci_enable_msi(adapter->pdev);
4474 if (err)
4475 goto msi_test_failed;
4476
4477 err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
4478 netdev->name, netdev);
4479 if (err) {
4480 pci_disable_msi(adapter->pdev);
4481 goto msi_test_failed;
4482 }
4483
4484 /* Force memory writes to complete before enabling and firing an
4485 * interrupt.
4486 */
4487 wmb();
4488
4489 e1000_irq_enable(adapter);
4490
4491 /* fire an unusual interrupt on the test handler */
4492 ew32(ICS, E1000_ICS_RXSEQ);
4493 e1e_flush();
4494 msleep(100);
4495
4496 e1000_irq_disable(adapter);
4497
4498 rmb(); /* read flags after interrupt has been fired */
4499
4500 if (adapter->flags & FLAG_MSI_TEST_FAILED) {
4501 adapter->int_mode = E1000E_INT_MODE_LEGACY;
4502 e_info("MSI interrupt test failed, using legacy interrupt.\n");
4503 } else {
4504 e_dbg("MSI interrupt test succeeded!\n");
4505 }
4506
4507 free_irq(adapter->pdev->irq, netdev);
4508 pci_disable_msi(adapter->pdev);
4509
4510msi_test_failed:
4511 e1000e_set_interrupt_capability(adapter);
4512 return e1000_request_irq(adapter);
4513}
4514
4515/**
4516 * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
4517 * @adapter: board private struct
4518 *
4519 * code flow taken from tg3.c, called with e1000 interrupts disabled.
4520 **/
4521static int e1000_test_msi(struct e1000_adapter *adapter)
4522{
4523 int err;
4524 u16 pci_cmd;
4525
4526 if (!(adapter->flags & FLAG_MSI_ENABLED))
4527 return 0;
4528
4529 /* disable SERR in case the MSI write causes a master abort */
4530 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4531 if (pci_cmd & PCI_COMMAND_SERR)
4532 pci_write_config_word(adapter->pdev, PCI_COMMAND,
4533 pci_cmd & ~PCI_COMMAND_SERR);
4534
4535 err = e1000_test_msi_interrupt(adapter);
4536
4537 /* re-enable SERR */
4538 if (pci_cmd & PCI_COMMAND_SERR) {
4539 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4540 pci_cmd |= PCI_COMMAND_SERR;
4541 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
4542 }
4543
4544 return err;
4545}
4546
4547/**
4548 * e1000e_open - Called when a network interface is made active
4549 * @netdev: network interface device structure
4550 *
4551 * Returns 0 on success, negative value on failure
4552 *
4553 * The open entry point is called when a network interface is made
4554 * active by the system (IFF_UP). At this point all resources needed
4555 * for transmit and receive operations are allocated, the interrupt
4556 * handler is registered with the OS, the watchdog timer is started,
4557 * and the stack is notified that the interface is ready.
4558 **/
4559int e1000e_open(struct net_device *netdev)
4560{
4561 struct e1000_adapter *adapter = netdev_priv(netdev);
4562 struct e1000_hw *hw = &adapter->hw;
4563 struct pci_dev *pdev = adapter->pdev;
4564 int err;
4565
4566 /* disallow open during test */
4567 if (test_bit(__E1000_TESTING, &adapter->state))
4568 return -EBUSY;
4569
4570 pm_runtime_get_sync(&pdev->dev);
4571
4572 netif_carrier_off(netdev);
4573
4574 /* allocate transmit descriptors */
4575 err = e1000e_setup_tx_resources(adapter->tx_ring);
4576 if (err)
4577 goto err_setup_tx;
4578
4579 /* allocate receive descriptors */
4580 err = e1000e_setup_rx_resources(adapter->rx_ring);
4581 if (err)
4582 goto err_setup_rx;
4583
4584 /* If AMT is enabled, let the firmware know that the network
4585 * interface is now open and reset the part to a known state.
4586 */
4587 if (adapter->flags & FLAG_HAS_AMT) {
4588 e1000e_get_hw_control(adapter);
4589 e1000e_reset(adapter);
4590 }
4591
4592 e1000e_power_up_phy(adapter);
4593
4594 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4595 if ((adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
4596 e1000_update_mng_vlan(adapter);
4597
4598 /* DMA latency requirement to workaround jumbo issue */
4599 pm_qos_add_request(&adapter->pm_qos_req, PM_QOS_CPU_DMA_LATENCY,
4600 PM_QOS_DEFAULT_VALUE);
4601
4602 /* before we allocate an interrupt, we must be ready to handle it.
4603 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
4604 * as soon as we call pci_request_irq, so we have to setup our
4605 * clean_rx handler before we do so.
4606 */
4607 e1000_configure(adapter);
4608
4609 err = e1000_request_irq(adapter);
4610 if (err)
4611 goto err_req_irq;
4612
4613 /* Work around PCIe errata with MSI interrupts causing some chipsets to
4614 * ignore e1000e MSI messages, which means we need to test our MSI
4615 * interrupt now
4616 */
4617 if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
4618 err = e1000_test_msi(adapter);
4619 if (err) {
4620 e_err("Interrupt allocation failed\n");
4621 goto err_req_irq;
4622 }
4623 }
4624
4625 /* From here on the code is the same as e1000e_up() */
4626 clear_bit(__E1000_DOWN, &adapter->state);
4627
4628 napi_enable(&adapter->napi);
4629
4630 e1000_irq_enable(adapter);
4631
4632 adapter->tx_hang_recheck = false;
4633 netif_start_queue(netdev);
4634
4635 hw->mac.get_link_status = true;
4636 pm_runtime_put(&pdev->dev);
4637
4638 e1000e_trigger_lsc(adapter);
4639
4640 return 0;
4641
4642err_req_irq:
4643 pm_qos_remove_request(&adapter->pm_qos_req);
4644 e1000e_release_hw_control(adapter);
4645 e1000_power_down_phy(adapter);
4646 e1000e_free_rx_resources(adapter->rx_ring);
4647err_setup_rx:
4648 e1000e_free_tx_resources(adapter->tx_ring);
4649err_setup_tx:
4650 e1000e_reset(adapter);
4651 pm_runtime_put_sync(&pdev->dev);
4652
4653 return err;
4654}
4655
4656/**
4657 * e1000e_close - Disables a network interface
4658 * @netdev: network interface device structure
4659 *
4660 * Returns 0, this is not allowed to fail
4661 *
4662 * The close entry point is called when an interface is de-activated
4663 * by the OS. The hardware is still under the drivers control, but
4664 * needs to be disabled. A global MAC reset is issued to stop the
4665 * hardware, and all transmit and receive resources are freed.
4666 **/
4667int e1000e_close(struct net_device *netdev)
4668{
4669 struct e1000_adapter *adapter = netdev_priv(netdev);
4670 struct pci_dev *pdev = adapter->pdev;
4671 int count = E1000_CHECK_RESET_COUNT;
4672
4673 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
4674 usleep_range(10000, 20000);
4675
4676 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4677
4678 pm_runtime_get_sync(&pdev->dev);
4679
4680 if (!test_bit(__E1000_DOWN, &adapter->state)) {
4681 e1000e_down(adapter, true);
4682 e1000_free_irq(adapter);
4683
4684 /* Link status message must follow this format */
4685 pr_info("%s NIC Link is Down\n", adapter->netdev->name);
4686 }
4687
4688 napi_disable(&adapter->napi);
4689
4690 e1000e_free_tx_resources(adapter->tx_ring);
4691 e1000e_free_rx_resources(adapter->rx_ring);
4692
4693 /* kill manageability vlan ID if supported, but not if a vlan with
4694 * the same ID is registered on the host OS (let 8021q kill it)
4695 */
4696 if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
4697 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
4698 adapter->mng_vlan_id);
4699
4700 /* If AMT is enabled, let the firmware know that the network
4701 * interface is now closed
4702 */
4703 if ((adapter->flags & FLAG_HAS_AMT) &&
4704 !test_bit(__E1000_TESTING, &adapter->state))
4705 e1000e_release_hw_control(adapter);
4706
4707 pm_qos_remove_request(&adapter->pm_qos_req);
4708
4709 pm_runtime_put_sync(&pdev->dev);
4710
4711 return 0;
4712}
4713
4714/**
4715 * e1000_set_mac - Change the Ethernet Address of the NIC
4716 * @netdev: network interface device structure
4717 * @p: pointer to an address structure
4718 *
4719 * Returns 0 on success, negative on failure
4720 **/
4721static int e1000_set_mac(struct net_device *netdev, void *p)
4722{
4723 struct e1000_adapter *adapter = netdev_priv(netdev);
4724 struct e1000_hw *hw = &adapter->hw;
4725 struct sockaddr *addr = p;
4726
4727 if (!is_valid_ether_addr(addr->sa_data))
4728 return -EADDRNOTAVAIL;
4729
4730 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4731 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
4732
4733 hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
4734
4735 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
4736 /* activate the work around */
4737 e1000e_set_laa_state_82571(&adapter->hw, 1);
4738
4739 /* Hold a copy of the LAA in RAR[14] This is done so that
4740 * between the time RAR[0] gets clobbered and the time it
4741 * gets fixed (in e1000_watchdog), the actual LAA is in one
4742 * of the RARs and no incoming packets directed to this port
4743 * are dropped. Eventually the LAA will be in RAR[0] and
4744 * RAR[14]
4745 */
4746 hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr,
4747 adapter->hw.mac.rar_entry_count - 1);
4748 }
4749
4750 return 0;
4751}
4752
4753/**
4754 * e1000e_update_phy_task - work thread to update phy
4755 * @work: pointer to our work struct
4756 *
4757 * this worker thread exists because we must acquire a
4758 * semaphore to read the phy, which we could msleep while
4759 * waiting for it, and we can't msleep in a timer.
4760 **/
4761static void e1000e_update_phy_task(struct work_struct *work)
4762{
4763 struct e1000_adapter *adapter = container_of(work,
4764 struct e1000_adapter,
4765 update_phy_task);
4766 struct e1000_hw *hw = &adapter->hw;
4767
4768 if (test_bit(__E1000_DOWN, &adapter->state))
4769 return;
4770
4771 e1000_get_phy_info(hw);
4772
4773 /* Enable EEE on 82579 after link up */
4774 if (hw->phy.type >= e1000_phy_82579)
4775 e1000_set_eee_pchlan(hw);
4776}
4777
4778/**
4779 * e1000_update_phy_info - timre call-back to update PHY info
4780 * @data: pointer to adapter cast into an unsigned long
4781 *
4782 * Need to wait a few seconds after link up to get diagnostic information from
4783 * the phy
4784 **/
4785static void e1000_update_phy_info(unsigned long data)
4786{
4787 struct e1000_adapter *adapter = (struct e1000_adapter *)data;
4788
4789 if (test_bit(__E1000_DOWN, &adapter->state))
4790 return;
4791
4792 schedule_work(&adapter->update_phy_task);
4793}
4794
4795/**
4796 * e1000e_update_phy_stats - Update the PHY statistics counters
4797 * @adapter: board private structure
4798 *
4799 * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4800 **/
4801static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4802{
4803 struct e1000_hw *hw = &adapter->hw;
4804 s32 ret_val;
4805 u16 phy_data;
4806
4807 ret_val = hw->phy.ops.acquire(hw);
4808 if (ret_val)
4809 return;
4810
4811 /* A page set is expensive so check if already on desired page.
4812 * If not, set to the page with the PHY status registers.
4813 */
4814 hw->phy.addr = 1;
4815 ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4816 &phy_data);
4817 if (ret_val)
4818 goto release;
4819 if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4820 ret_val = hw->phy.ops.set_page(hw,
4821 HV_STATS_PAGE << IGP_PAGE_SHIFT);
4822 if (ret_val)
4823 goto release;
4824 }
4825
4826 /* Single Collision Count */
4827 hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4828 ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4829 if (!ret_val)
4830 adapter->stats.scc += phy_data;
4831
4832 /* Excessive Collision Count */
4833 hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4834 ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4835 if (!ret_val)
4836 adapter->stats.ecol += phy_data;
4837
4838 /* Multiple Collision Count */
4839 hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4840 ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4841 if (!ret_val)
4842 adapter->stats.mcc += phy_data;
4843
4844 /* Late Collision Count */
4845 hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4846 ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4847 if (!ret_val)
4848 adapter->stats.latecol += phy_data;
4849
4850 /* Collision Count - also used for adaptive IFS */
4851 hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4852 ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4853 if (!ret_val)
4854 hw->mac.collision_delta = phy_data;
4855
4856 /* Defer Count */
4857 hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4858 ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4859 if (!ret_val)
4860 adapter->stats.dc += phy_data;
4861
4862 /* Transmit with no CRS */
4863 hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4864 ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4865 if (!ret_val)
4866 adapter->stats.tncrs += phy_data;
4867
4868release:
4869 hw->phy.ops.release(hw);
4870}
4871
4872/**
4873 * e1000e_update_stats - Update the board statistics counters
4874 * @adapter: board private structure
4875 **/
4876static void e1000e_update_stats(struct e1000_adapter *adapter)
4877{
4878 struct net_device *netdev = adapter->netdev;
4879 struct e1000_hw *hw = &adapter->hw;
4880 struct pci_dev *pdev = adapter->pdev;
4881
4882 /* Prevent stats update while adapter is being reset, or if the pci
4883 * connection is down.
4884 */
4885 if (adapter->link_speed == 0)
4886 return;
4887 if (pci_channel_offline(pdev))
4888 return;
4889
4890 adapter->stats.crcerrs += er32(CRCERRS);
4891 adapter->stats.gprc += er32(GPRC);
4892 adapter->stats.gorc += er32(GORCL);
4893 er32(GORCH); /* Clear gorc */
4894 adapter->stats.bprc += er32(BPRC);
4895 adapter->stats.mprc += er32(MPRC);
4896 adapter->stats.roc += er32(ROC);
4897
4898 adapter->stats.mpc += er32(MPC);
4899
4900 /* Half-duplex statistics */
4901 if (adapter->link_duplex == HALF_DUPLEX) {
4902 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4903 e1000e_update_phy_stats(adapter);
4904 } else {
4905 adapter->stats.scc += er32(SCC);
4906 adapter->stats.ecol += er32(ECOL);
4907 adapter->stats.mcc += er32(MCC);
4908 adapter->stats.latecol += er32(LATECOL);
4909 adapter->stats.dc += er32(DC);
4910
4911 hw->mac.collision_delta = er32(COLC);
4912
4913 if ((hw->mac.type != e1000_82574) &&
4914 (hw->mac.type != e1000_82583))
4915 adapter->stats.tncrs += er32(TNCRS);
4916 }
4917 adapter->stats.colc += hw->mac.collision_delta;
4918 }
4919
4920 adapter->stats.xonrxc += er32(XONRXC);
4921 adapter->stats.xontxc += er32(XONTXC);
4922 adapter->stats.xoffrxc += er32(XOFFRXC);
4923 adapter->stats.xofftxc += er32(XOFFTXC);
4924 adapter->stats.gptc += er32(GPTC);
4925 adapter->stats.gotc += er32(GOTCL);
4926 er32(GOTCH); /* Clear gotc */
4927 adapter->stats.rnbc += er32(RNBC);
4928 adapter->stats.ruc += er32(RUC);
4929
4930 adapter->stats.mptc += er32(MPTC);
4931 adapter->stats.bptc += er32(BPTC);
4932
4933 /* used for adaptive IFS */
4934
4935 hw->mac.tx_packet_delta = er32(TPT);
4936 adapter->stats.tpt += hw->mac.tx_packet_delta;
4937
4938 adapter->stats.algnerrc += er32(ALGNERRC);
4939 adapter->stats.rxerrc += er32(RXERRC);
4940 adapter->stats.cexterr += er32(CEXTERR);
4941 adapter->stats.tsctc += er32(TSCTC);
4942 adapter->stats.tsctfc += er32(TSCTFC);
4943
4944 /* Fill out the OS statistics structure */
4945 netdev->stats.multicast = adapter->stats.mprc;
4946 netdev->stats.collisions = adapter->stats.colc;
4947
4948 /* Rx Errors */
4949
4950 /* RLEC on some newer hardware can be incorrect so build
4951 * our own version based on RUC and ROC
4952 */
4953 netdev->stats.rx_errors = adapter->stats.rxerrc +
4954 adapter->stats.crcerrs + adapter->stats.algnerrc +
4955 adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
4956 netdev->stats.rx_length_errors = adapter->stats.ruc +
4957 adapter->stats.roc;
4958 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4959 netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4960 netdev->stats.rx_missed_errors = adapter->stats.mpc;
4961
4962 /* Tx Errors */
4963 netdev->stats.tx_errors = adapter->stats.ecol + adapter->stats.latecol;
4964 netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4965 netdev->stats.tx_window_errors = adapter->stats.latecol;
4966 netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4967
4968 /* Tx Dropped needs to be maintained elsewhere */
4969
4970 /* Management Stats */
4971 adapter->stats.mgptc += er32(MGTPTC);
4972 adapter->stats.mgprc += er32(MGTPRC);
4973 adapter->stats.mgpdc += er32(MGTPDC);
4974
4975 /* Correctable ECC Errors */
4976 if ((hw->mac.type == e1000_pch_lpt) ||
4977 (hw->mac.type == e1000_pch_spt)) {
4978 u32 pbeccsts = er32(PBECCSTS);
4979
4980 adapter->corr_errors +=
4981 pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
4982 adapter->uncorr_errors +=
4983 (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
4984 E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
4985 }
4986}
4987
4988/**
4989 * e1000_phy_read_status - Update the PHY register status snapshot
4990 * @adapter: board private structure
4991 **/
4992static void e1000_phy_read_status(struct e1000_adapter *adapter)
4993{
4994 struct e1000_hw *hw = &adapter->hw;
4995 struct e1000_phy_regs *phy = &adapter->phy_regs;
4996
4997 if (!pm_runtime_suspended((&adapter->pdev->dev)->parent) &&
4998 (er32(STATUS) & E1000_STATUS_LU) &&
4999 (adapter->hw.phy.media_type == e1000_media_type_copper)) {
5000 int ret_val;
5001
5002 ret_val = e1e_rphy(hw, MII_BMCR, &phy->bmcr);
5003 ret_val |= e1e_rphy(hw, MII_BMSR, &phy->bmsr);
5004 ret_val |= e1e_rphy(hw, MII_ADVERTISE, &phy->advertise);
5005 ret_val |= e1e_rphy(hw, MII_LPA, &phy->lpa);
5006 ret_val |= e1e_rphy(hw, MII_EXPANSION, &phy->expansion);
5007 ret_val |= e1e_rphy(hw, MII_CTRL1000, &phy->ctrl1000);
5008 ret_val |= e1e_rphy(hw, MII_STAT1000, &phy->stat1000);
5009 ret_val |= e1e_rphy(hw, MII_ESTATUS, &phy->estatus);
5010 if (ret_val)
5011 e_warn("Error reading PHY register\n");
5012 } else {
5013 /* Do not read PHY registers if link is not up
5014 * Set values to typical power-on defaults
5015 */
5016 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
5017 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
5018 BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
5019 BMSR_ERCAP);
5020 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
5021 ADVERTISE_ALL | ADVERTISE_CSMA);
5022 phy->lpa = 0;
5023 phy->expansion = EXPANSION_ENABLENPAGE;
5024 phy->ctrl1000 = ADVERTISE_1000FULL;
5025 phy->stat1000 = 0;
5026 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
5027 }
5028}
5029
5030static void e1000_print_link_info(struct e1000_adapter *adapter)
5031{
5032 struct e1000_hw *hw = &adapter->hw;
5033 u32 ctrl = er32(CTRL);
5034
5035 /* Link status message must follow this format for user tools */
5036 pr_info("%s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
5037 adapter->netdev->name, adapter->link_speed,
5038 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
5039 (ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
5040 (ctrl & E1000_CTRL_RFCE) ? "Rx" :
5041 (ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
5042}
5043
5044static bool e1000e_has_link(struct e1000_adapter *adapter)
5045{
5046 struct e1000_hw *hw = &adapter->hw;
5047 bool link_active = false;
5048 s32 ret_val = 0;
5049
5050 /* get_link_status is set on LSC (link status) interrupt or
5051 * Rx sequence error interrupt. get_link_status will stay
5052 * false until the check_for_link establishes link
5053 * for copper adapters ONLY
5054 */
5055 switch (hw->phy.media_type) {
5056 case e1000_media_type_copper:
5057 if (hw->mac.get_link_status) {
5058 ret_val = hw->mac.ops.check_for_link(hw);
5059 link_active = !hw->mac.get_link_status;
5060 } else {
5061 link_active = true;
5062 }
5063 break;
5064 case e1000_media_type_fiber:
5065 ret_val = hw->mac.ops.check_for_link(hw);
5066 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
5067 break;
5068 case e1000_media_type_internal_serdes:
5069 ret_val = hw->mac.ops.check_for_link(hw);
5070 link_active = adapter->hw.mac.serdes_has_link;
5071 break;
5072 default:
5073 case e1000_media_type_unknown:
5074 break;
5075 }
5076
5077 if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
5078 (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
5079 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
5080 e_info("Gigabit has been disabled, downgrading speed\n");
5081 }
5082
5083 return link_active;
5084}
5085
5086static void e1000e_enable_receives(struct e1000_adapter *adapter)
5087{
5088 /* make sure the receive unit is started */
5089 if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5090 (adapter->flags & FLAG_RESTART_NOW)) {
5091 struct e1000_hw *hw = &adapter->hw;
5092 u32 rctl = er32(RCTL);
5093
5094 ew32(RCTL, rctl | E1000_RCTL_EN);
5095 adapter->flags &= ~FLAG_RESTART_NOW;
5096 }
5097}
5098
5099static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
5100{
5101 struct e1000_hw *hw = &adapter->hw;
5102
5103 /* With 82574 controllers, PHY needs to be checked periodically
5104 * for hung state and reset, if two calls return true
5105 */
5106 if (e1000_check_phy_82574(hw))
5107 adapter->phy_hang_count++;
5108 else
5109 adapter->phy_hang_count = 0;
5110
5111 if (adapter->phy_hang_count > 1) {
5112 adapter->phy_hang_count = 0;
5113 e_dbg("PHY appears hung - resetting\n");
5114 schedule_work(&adapter->reset_task);
5115 }
5116}
5117
5118/**
5119 * e1000_watchdog - Timer Call-back
5120 * @data: pointer to adapter cast into an unsigned long
5121 **/
5122static void e1000_watchdog(unsigned long data)
5123{
5124 struct e1000_adapter *adapter = (struct e1000_adapter *)data;
5125
5126 /* Do the rest outside of interrupt context */
5127 schedule_work(&adapter->watchdog_task);
5128
5129 /* TODO: make this use queue_delayed_work() */
5130}
5131
5132static void e1000_watchdog_task(struct work_struct *work)
5133{
5134 struct e1000_adapter *adapter = container_of(work,
5135 struct e1000_adapter,
5136 watchdog_task);
5137 struct net_device *netdev = adapter->netdev;
5138 struct e1000_mac_info *mac = &adapter->hw.mac;
5139 struct e1000_phy_info *phy = &adapter->hw.phy;
5140 struct e1000_ring *tx_ring = adapter->tx_ring;
5141 struct e1000_hw *hw = &adapter->hw;
5142 u32 link, tctl;
5143
5144 if (test_bit(__E1000_DOWN, &adapter->state))
5145 return;
5146
5147 link = e1000e_has_link(adapter);
5148 if ((netif_carrier_ok(netdev)) && link) {
5149 /* Cancel scheduled suspend requests. */
5150 pm_runtime_resume(netdev->dev.parent);
5151
5152 e1000e_enable_receives(adapter);
5153 goto link_up;
5154 }
5155
5156 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
5157 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
5158 e1000_update_mng_vlan(adapter);
5159
5160 if (link) {
5161 if (!netif_carrier_ok(netdev)) {
5162 bool txb2b = true;
5163
5164 /* Cancel scheduled suspend requests. */
5165 pm_runtime_resume(netdev->dev.parent);
5166
5167 /* update snapshot of PHY registers on LSC */
5168 e1000_phy_read_status(adapter);
5169 mac->ops.get_link_up_info(&adapter->hw,
5170 &adapter->link_speed,
5171 &adapter->link_duplex);
5172 e1000_print_link_info(adapter);
5173
5174 /* check if SmartSpeed worked */
5175 e1000e_check_downshift(hw);
5176 if (phy->speed_downgraded)
5177 netdev_warn(netdev,
5178 "Link Speed was downgraded by SmartSpeed\n");
5179
5180 /* On supported PHYs, check for duplex mismatch only
5181 * if link has autonegotiated at 10/100 half
5182 */
5183 if ((hw->phy.type == e1000_phy_igp_3 ||
5184 hw->phy.type == e1000_phy_bm) &&
5185 hw->mac.autoneg &&
5186 (adapter->link_speed == SPEED_10 ||
5187 adapter->link_speed == SPEED_100) &&
5188 (adapter->link_duplex == HALF_DUPLEX)) {
5189 u16 autoneg_exp;
5190
5191 e1e_rphy(hw, MII_EXPANSION, &autoneg_exp);
5192
5193 if (!(autoneg_exp & EXPANSION_NWAY))
5194 e_info("Autonegotiated half duplex but link partner cannot autoneg. Try forcing full duplex if link gets many collisions.\n");
5195 }
5196
5197 /* adjust timeout factor according to speed/duplex */
5198 adapter->tx_timeout_factor = 1;
5199 switch (adapter->link_speed) {
5200 case SPEED_10:
5201 txb2b = false;
5202 adapter->tx_timeout_factor = 16;
5203 break;
5204 case SPEED_100:
5205 txb2b = false;
5206 adapter->tx_timeout_factor = 10;
5207 break;
5208 }
5209
5210 /* workaround: re-program speed mode bit after
5211 * link-up event
5212 */
5213 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
5214 !txb2b) {
5215 u32 tarc0;
5216
5217 tarc0 = er32(TARC(0));
5218 tarc0 &= ~SPEED_MODE_BIT;
5219 ew32(TARC(0), tarc0);
5220 }
5221
5222 /* disable TSO for pcie and 10/100 speeds, to avoid
5223 * some hardware issues
5224 */
5225 if (!(adapter->flags & FLAG_TSO_FORCE)) {
5226 switch (adapter->link_speed) {
5227 case SPEED_10:
5228 case SPEED_100:
5229 e_info("10/100 speed: disabling TSO\n");
5230 netdev->features &= ~NETIF_F_TSO;
5231 netdev->features &= ~NETIF_F_TSO6;
5232 break;
5233 case SPEED_1000:
5234 netdev->features |= NETIF_F_TSO;
5235 netdev->features |= NETIF_F_TSO6;
5236 break;
5237 default:
5238 /* oops */
5239 break;
5240 }
5241 }
5242
5243 /* enable transmits in the hardware, need to do this
5244 * after setting TARC(0)
5245 */
5246 tctl = er32(TCTL);
5247 tctl |= E1000_TCTL_EN;
5248 ew32(TCTL, tctl);
5249
5250 /* Perform any post-link-up configuration before
5251 * reporting link up.
5252 */
5253 if (phy->ops.cfg_on_link_up)
5254 phy->ops.cfg_on_link_up(hw);
5255
5256 netif_carrier_on(netdev);
5257
5258 if (!test_bit(__E1000_DOWN, &adapter->state))
5259 mod_timer(&adapter->phy_info_timer,
5260 round_jiffies(jiffies + 2 * HZ));
5261 }
5262 } else {
5263 if (netif_carrier_ok(netdev)) {
5264 adapter->link_speed = 0;
5265 adapter->link_duplex = 0;
5266 /* Link status message must follow this format */
5267 pr_info("%s NIC Link is Down\n", adapter->netdev->name);
5268 netif_carrier_off(netdev);
5269 if (!test_bit(__E1000_DOWN, &adapter->state))
5270 mod_timer(&adapter->phy_info_timer,
5271 round_jiffies(jiffies + 2 * HZ));
5272
5273 /* 8000ES2LAN requires a Rx packet buffer work-around
5274 * on link down event; reset the controller to flush
5275 * the Rx packet buffer.
5276 */
5277 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
5278 adapter->flags |= FLAG_RESTART_NOW;
5279 else
5280 pm_schedule_suspend(netdev->dev.parent,
5281 LINK_TIMEOUT);
5282 }
5283 }
5284
5285link_up:
5286 spin_lock(&adapter->stats64_lock);
5287 e1000e_update_stats(adapter);
5288
5289 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
5290 adapter->tpt_old = adapter->stats.tpt;
5291 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
5292 adapter->colc_old = adapter->stats.colc;
5293
5294 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
5295 adapter->gorc_old = adapter->stats.gorc;
5296 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
5297 adapter->gotc_old = adapter->stats.gotc;
5298 spin_unlock(&adapter->stats64_lock);
5299
5300 /* If the link is lost the controller stops DMA, but
5301 * if there is queued Tx work it cannot be done. So
5302 * reset the controller to flush the Tx packet buffers.
5303 */
5304 if (!netif_carrier_ok(netdev) &&
5305 (e1000_desc_unused(tx_ring) + 1 < tx_ring->count))
5306 adapter->flags |= FLAG_RESTART_NOW;
5307
5308 /* If reset is necessary, do it outside of interrupt context. */
5309 if (adapter->flags & FLAG_RESTART_NOW) {
5310 schedule_work(&adapter->reset_task);
5311 /* return immediately since reset is imminent */
5312 return;
5313 }
5314
5315 e1000e_update_adaptive(&adapter->hw);
5316
5317 /* Simple mode for Interrupt Throttle Rate (ITR) */
5318 if (adapter->itr_setting == 4) {
5319 /* Symmetric Tx/Rx gets a reduced ITR=2000;
5320 * Total asymmetrical Tx or Rx gets ITR=8000;
5321 * everyone else is between 2000-8000.
5322 */
5323 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
5324 u32 dif = (adapter->gotc > adapter->gorc ?
5325 adapter->gotc - adapter->gorc :
5326 adapter->gorc - adapter->gotc) / 10000;
5327 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
5328
5329 e1000e_write_itr(adapter, itr);
5330 }
5331
5332 /* Cause software interrupt to ensure Rx ring is cleaned */
5333 if (adapter->msix_entries)
5334 ew32(ICS, adapter->rx_ring->ims_val);
5335 else
5336 ew32(ICS, E1000_ICS_RXDMT0);
5337
5338 /* flush pending descriptors to memory before detecting Tx hang */
5339 e1000e_flush_descriptors(adapter);
5340
5341 /* Force detection of hung controller every watchdog period */
5342 adapter->detect_tx_hung = true;
5343
5344 /* With 82571 controllers, LAA may be overwritten due to controller
5345 * reset from the other port. Set the appropriate LAA in RAR[0]
5346 */
5347 if (e1000e_get_laa_state_82571(hw))
5348 hw->mac.ops.rar_set(hw, adapter->hw.mac.addr, 0);
5349
5350 if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
5351 e1000e_check_82574_phy_workaround(adapter);
5352
5353 /* Clear valid timestamp stuck in RXSTMPL/H due to a Rx error */
5354 if (adapter->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE) {
5355 if ((adapter->flags2 & FLAG2_CHECK_RX_HWTSTAMP) &&
5356 (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID)) {
5357 er32(RXSTMPH);
5358 adapter->rx_hwtstamp_cleared++;
5359 } else {
5360 adapter->flags2 |= FLAG2_CHECK_RX_HWTSTAMP;
5361 }
5362 }
5363
5364 /* Reset the timer */
5365 if (!test_bit(__E1000_DOWN, &adapter->state))
5366 mod_timer(&adapter->watchdog_timer,
5367 round_jiffies(jiffies + 2 * HZ));
5368}
5369
5370#define E1000_TX_FLAGS_CSUM 0x00000001
5371#define E1000_TX_FLAGS_VLAN 0x00000002
5372#define E1000_TX_FLAGS_TSO 0x00000004
5373#define E1000_TX_FLAGS_IPV4 0x00000008
5374#define E1000_TX_FLAGS_NO_FCS 0x00000010
5375#define E1000_TX_FLAGS_HWTSTAMP 0x00000020
5376#define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
5377#define E1000_TX_FLAGS_VLAN_SHIFT 16
5378
5379static int e1000_tso(struct e1000_ring *tx_ring, struct sk_buff *skb,
5380 __be16 protocol)
5381{
5382 struct e1000_context_desc *context_desc;
5383 struct e1000_buffer *buffer_info;
5384 unsigned int i;
5385 u32 cmd_length = 0;
5386 u16 ipcse = 0, mss;
5387 u8 ipcss, ipcso, tucss, tucso, hdr_len;
5388 int err;
5389
5390 if (!skb_is_gso(skb))
5391 return 0;
5392
5393 err = skb_cow_head(skb, 0);
5394 if (err < 0)
5395 return err;
5396
5397 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5398 mss = skb_shinfo(skb)->gso_size;
5399 if (protocol == htons(ETH_P_IP)) {
5400 struct iphdr *iph = ip_hdr(skb);
5401 iph->tot_len = 0;
5402 iph->check = 0;
5403 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
5404 0, IPPROTO_TCP, 0);
5405 cmd_length = E1000_TXD_CMD_IP;
5406 ipcse = skb_transport_offset(skb) - 1;
5407 } else if (skb_is_gso_v6(skb)) {
5408 ipv6_hdr(skb)->payload_len = 0;
5409 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5410 &ipv6_hdr(skb)->daddr,
5411 0, IPPROTO_TCP, 0);
5412 ipcse = 0;
5413 }
5414 ipcss = skb_network_offset(skb);
5415 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
5416 tucss = skb_transport_offset(skb);
5417 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
5418
5419 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
5420 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
5421
5422 i = tx_ring->next_to_use;
5423 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5424 buffer_info = &tx_ring->buffer_info[i];
5425
5426 context_desc->lower_setup.ip_fields.ipcss = ipcss;
5427 context_desc->lower_setup.ip_fields.ipcso = ipcso;
5428 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
5429 context_desc->upper_setup.tcp_fields.tucss = tucss;
5430 context_desc->upper_setup.tcp_fields.tucso = tucso;
5431 context_desc->upper_setup.tcp_fields.tucse = 0;
5432 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
5433 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
5434 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
5435
5436 buffer_info->time_stamp = jiffies;
5437 buffer_info->next_to_watch = i;
5438
5439 i++;
5440 if (i == tx_ring->count)
5441 i = 0;
5442 tx_ring->next_to_use = i;
5443
5444 return 1;
5445}
5446
5447static bool e1000_tx_csum(struct e1000_ring *tx_ring, struct sk_buff *skb,
5448 __be16 protocol)
5449{
5450 struct e1000_adapter *adapter = tx_ring->adapter;
5451 struct e1000_context_desc *context_desc;
5452 struct e1000_buffer *buffer_info;
5453 unsigned int i;
5454 u8 css;
5455 u32 cmd_len = E1000_TXD_CMD_DEXT;
5456
5457 if (skb->ip_summed != CHECKSUM_PARTIAL)
5458 return false;
5459
5460 switch (protocol) {
5461 case cpu_to_be16(ETH_P_IP):
5462 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5463 cmd_len |= E1000_TXD_CMD_TCP;
5464 break;
5465 case cpu_to_be16(ETH_P_IPV6):
5466 /* XXX not handling all IPV6 headers */
5467 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5468 cmd_len |= E1000_TXD_CMD_TCP;
5469 break;
5470 default:
5471 if (unlikely(net_ratelimit()))
5472 e_warn("checksum_partial proto=%x!\n",
5473 be16_to_cpu(protocol));
5474 break;
5475 }
5476
5477 css = skb_checksum_start_offset(skb);
5478
5479 i = tx_ring->next_to_use;
5480 buffer_info = &tx_ring->buffer_info[i];
5481 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5482
5483 context_desc->lower_setup.ip_config = 0;
5484 context_desc->upper_setup.tcp_fields.tucss = css;
5485 context_desc->upper_setup.tcp_fields.tucso = css + skb->csum_offset;
5486 context_desc->upper_setup.tcp_fields.tucse = 0;
5487 context_desc->tcp_seg_setup.data = 0;
5488 context_desc->cmd_and_length = cpu_to_le32(cmd_len);
5489
5490 buffer_info->time_stamp = jiffies;
5491 buffer_info->next_to_watch = i;
5492
5493 i++;
5494 if (i == tx_ring->count)
5495 i = 0;
5496 tx_ring->next_to_use = i;
5497
5498 return true;
5499}
5500
5501static int e1000_tx_map(struct e1000_ring *tx_ring, struct sk_buff *skb,
5502 unsigned int first, unsigned int max_per_txd,
5503 unsigned int nr_frags)
5504{
5505 struct e1000_adapter *adapter = tx_ring->adapter;
5506 struct pci_dev *pdev = adapter->pdev;
5507 struct e1000_buffer *buffer_info;
5508 unsigned int len = skb_headlen(skb);
5509 unsigned int offset = 0, size, count = 0, i;
5510 unsigned int f, bytecount, segs;
5511
5512 i = tx_ring->next_to_use;
5513
5514 while (len) {
5515 buffer_info = &tx_ring->buffer_info[i];
5516 size = min(len, max_per_txd);
5517
5518 buffer_info->length = size;
5519 buffer_info->time_stamp = jiffies;
5520 buffer_info->next_to_watch = i;
5521 buffer_info->dma = dma_map_single(&pdev->dev,
5522 skb->data + offset,
5523 size, DMA_TO_DEVICE);
5524 buffer_info->mapped_as_page = false;
5525 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5526 goto dma_error;
5527
5528 len -= size;
5529 offset += size;
5530 count++;
5531
5532 if (len) {
5533 i++;
5534 if (i == tx_ring->count)
5535 i = 0;
5536 }
5537 }
5538
5539 for (f = 0; f < nr_frags; f++) {
5540 const struct skb_frag_struct *frag;
5541
5542 frag = &skb_shinfo(skb)->frags[f];
5543 len = skb_frag_size(frag);
5544 offset = 0;
5545
5546 while (len) {
5547 i++;
5548 if (i == tx_ring->count)
5549 i = 0;
5550
5551 buffer_info = &tx_ring->buffer_info[i];
5552 size = min(len, max_per_txd);
5553
5554 buffer_info->length = size;
5555 buffer_info->time_stamp = jiffies;
5556 buffer_info->next_to_watch = i;
5557 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
5558 offset, size,
5559 DMA_TO_DEVICE);
5560 buffer_info->mapped_as_page = true;
5561 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5562 goto dma_error;
5563
5564 len -= size;
5565 offset += size;
5566 count++;
5567 }
5568 }
5569
5570 segs = skb_shinfo(skb)->gso_segs ? : 1;
5571 /* multiply data chunks by size of headers */
5572 bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
5573
5574 tx_ring->buffer_info[i].skb = skb;
5575 tx_ring->buffer_info[i].segs = segs;
5576 tx_ring->buffer_info[i].bytecount = bytecount;
5577 tx_ring->buffer_info[first].next_to_watch = i;
5578
5579 return count;
5580
5581dma_error:
5582 dev_err(&pdev->dev, "Tx DMA map failed\n");
5583 buffer_info->dma = 0;
5584 if (count)
5585 count--;
5586
5587 while (count--) {
5588 if (i == 0)
5589 i += tx_ring->count;
5590 i--;
5591 buffer_info = &tx_ring->buffer_info[i];
5592 e1000_put_txbuf(tx_ring, buffer_info);
5593 }
5594
5595 return 0;
5596}
5597
5598static void e1000_tx_queue(struct e1000_ring *tx_ring, int tx_flags, int count)
5599{
5600 struct e1000_adapter *adapter = tx_ring->adapter;
5601 struct e1000_tx_desc *tx_desc = NULL;
5602 struct e1000_buffer *buffer_info;
5603 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
5604 unsigned int i;
5605
5606 if (tx_flags & E1000_TX_FLAGS_TSO) {
5607 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
5608 E1000_TXD_CMD_TSE;
5609 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5610
5611 if (tx_flags & E1000_TX_FLAGS_IPV4)
5612 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
5613 }
5614
5615 if (tx_flags & E1000_TX_FLAGS_CSUM) {
5616 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5617 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5618 }
5619
5620 if (tx_flags & E1000_TX_FLAGS_VLAN) {
5621 txd_lower |= E1000_TXD_CMD_VLE;
5622 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
5623 }
5624
5625 if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5626 txd_lower &= ~(E1000_TXD_CMD_IFCS);
5627
5628 if (unlikely(tx_flags & E1000_TX_FLAGS_HWTSTAMP)) {
5629 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5630 txd_upper |= E1000_TXD_EXTCMD_TSTAMP;
5631 }
5632
5633 i = tx_ring->next_to_use;
5634
5635 do {
5636 buffer_info = &tx_ring->buffer_info[i];
5637 tx_desc = E1000_TX_DESC(*tx_ring, i);
5638 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
5639 tx_desc->lower.data = cpu_to_le32(txd_lower |
5640 buffer_info->length);
5641 tx_desc->upper.data = cpu_to_le32(txd_upper);
5642
5643 i++;
5644 if (i == tx_ring->count)
5645 i = 0;
5646 } while (--count > 0);
5647
5648 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
5649
5650 /* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
5651 if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5652 tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
5653
5654 /* Force memory writes to complete before letting h/w
5655 * know there are new descriptors to fetch. (Only
5656 * applicable for weak-ordered memory model archs,
5657 * such as IA-64).
5658 */
5659 wmb();
5660
5661 tx_ring->next_to_use = i;
5662}
5663
5664#define MINIMUM_DHCP_PACKET_SIZE 282
5665static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
5666 struct sk_buff *skb)
5667{
5668 struct e1000_hw *hw = &adapter->hw;
5669 u16 length, offset;
5670
5671 if (skb_vlan_tag_present(skb) &&
5672 !((skb_vlan_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
5673 (adapter->hw.mng_cookie.status &
5674 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
5675 return 0;
5676
5677 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
5678 return 0;
5679
5680 if (((struct ethhdr *)skb->data)->h_proto != htons(ETH_P_IP))
5681 return 0;
5682
5683 {
5684 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data + 14);
5685 struct udphdr *udp;
5686
5687 if (ip->protocol != IPPROTO_UDP)
5688 return 0;
5689
5690 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
5691 if (ntohs(udp->dest) != 67)
5692 return 0;
5693
5694 offset = (u8 *)udp + 8 - skb->data;
5695 length = skb->len - offset;
5696 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
5697 }
5698
5699 return 0;
5700}
5701
5702static int __e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5703{
5704 struct e1000_adapter *adapter = tx_ring->adapter;
5705
5706 netif_stop_queue(adapter->netdev);
5707 /* Herbert's original patch had:
5708 * smp_mb__after_netif_stop_queue();
5709 * but since that doesn't exist yet, just open code it.
5710 */
5711 smp_mb();
5712
5713 /* We need to check again in a case another CPU has just
5714 * made room available.
5715 */
5716 if (e1000_desc_unused(tx_ring) < size)
5717 return -EBUSY;
5718
5719 /* A reprieve! */
5720 netif_start_queue(adapter->netdev);
5721 ++adapter->restart_queue;
5722 return 0;
5723}
5724
5725static int e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5726{
5727 BUG_ON(size > tx_ring->count);
5728
5729 if (e1000_desc_unused(tx_ring) >= size)
5730 return 0;
5731 return __e1000_maybe_stop_tx(tx_ring, size);
5732}
5733
5734static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
5735 struct net_device *netdev)
5736{
5737 struct e1000_adapter *adapter = netdev_priv(netdev);
5738 struct e1000_ring *tx_ring = adapter->tx_ring;
5739 unsigned int first;
5740 unsigned int tx_flags = 0;
5741 unsigned int len = skb_headlen(skb);
5742 unsigned int nr_frags;
5743 unsigned int mss;
5744 int count = 0;
5745 int tso;
5746 unsigned int f;
5747 __be16 protocol = vlan_get_protocol(skb);
5748
5749 if (test_bit(__E1000_DOWN, &adapter->state)) {
5750 dev_kfree_skb_any(skb);
5751 return NETDEV_TX_OK;
5752 }
5753
5754 if (skb->len <= 0) {
5755 dev_kfree_skb_any(skb);
5756 return NETDEV_TX_OK;
5757 }
5758
5759 /* The minimum packet size with TCTL.PSP set is 17 bytes so
5760 * pad skb in order to meet this minimum size requirement
5761 */
5762 if (skb_put_padto(skb, 17))
5763 return NETDEV_TX_OK;
5764
5765 mss = skb_shinfo(skb)->gso_size;
5766 if (mss) {
5767 u8 hdr_len;
5768
5769 /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
5770 * points to just header, pull a few bytes of payload from
5771 * frags into skb->data
5772 */
5773 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5774 /* we do this workaround for ES2LAN, but it is un-necessary,
5775 * avoiding it could save a lot of cycles
5776 */
5777 if (skb->data_len && (hdr_len == len)) {
5778 unsigned int pull_size;
5779
5780 pull_size = min_t(unsigned int, 4, skb->data_len);
5781 if (!__pskb_pull_tail(skb, pull_size)) {
5782 e_err("__pskb_pull_tail failed.\n");
5783 dev_kfree_skb_any(skb);
5784 return NETDEV_TX_OK;
5785 }
5786 len = skb_headlen(skb);
5787 }
5788 }
5789
5790 /* reserve a descriptor for the offload context */
5791 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5792 count++;
5793 count++;
5794
5795 count += DIV_ROUND_UP(len, adapter->tx_fifo_limit);
5796
5797 nr_frags = skb_shinfo(skb)->nr_frags;
5798 for (f = 0; f < nr_frags; f++)
5799 count += DIV_ROUND_UP(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5800 adapter->tx_fifo_limit);
5801
5802 if (adapter->hw.mac.tx_pkt_filtering)
5803 e1000_transfer_dhcp_info(adapter, skb);
5804
5805 /* need: count + 2 desc gap to keep tail from touching
5806 * head, otherwise try next time
5807 */
5808 if (e1000_maybe_stop_tx(tx_ring, count + 2))
5809 return NETDEV_TX_BUSY;
5810
5811 if (skb_vlan_tag_present(skb)) {
5812 tx_flags |= E1000_TX_FLAGS_VLAN;
5813 tx_flags |= (skb_vlan_tag_get(skb) <<
5814 E1000_TX_FLAGS_VLAN_SHIFT);
5815 }
5816
5817 first = tx_ring->next_to_use;
5818
5819 tso = e1000_tso(tx_ring, skb, protocol);
5820 if (tso < 0) {
5821 dev_kfree_skb_any(skb);
5822 return NETDEV_TX_OK;
5823 }
5824
5825 if (tso)
5826 tx_flags |= E1000_TX_FLAGS_TSO;
5827 else if (e1000_tx_csum(tx_ring, skb, protocol))
5828 tx_flags |= E1000_TX_FLAGS_CSUM;
5829
5830 /* Old method was to assume IPv4 packet by default if TSO was enabled.
5831 * 82571 hardware supports TSO capabilities for IPv6 as well...
5832 * no longer assume, we must.
5833 */
5834 if (protocol == htons(ETH_P_IP))
5835 tx_flags |= E1000_TX_FLAGS_IPV4;
5836
5837 if (unlikely(skb->no_fcs))
5838 tx_flags |= E1000_TX_FLAGS_NO_FCS;
5839
5840 /* if count is 0 then mapping error has occurred */
5841 count = e1000_tx_map(tx_ring, skb, first, adapter->tx_fifo_limit,
5842 nr_frags);
5843 if (count) {
5844 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
5845 (adapter->flags & FLAG_HAS_HW_TIMESTAMP) &&
5846 !adapter->tx_hwtstamp_skb) {
5847 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
5848 tx_flags |= E1000_TX_FLAGS_HWTSTAMP;
5849 adapter->tx_hwtstamp_skb = skb_get(skb);
5850 adapter->tx_hwtstamp_start = jiffies;
5851 schedule_work(&adapter->tx_hwtstamp_work);
5852 } else {
5853 skb_tx_timestamp(skb);
5854 }
5855
5856 netdev_sent_queue(netdev, skb->len);
5857 e1000_tx_queue(tx_ring, tx_flags, count);
5858 /* Make sure there is space in the ring for the next send. */
5859 e1000_maybe_stop_tx(tx_ring,
5860 (MAX_SKB_FRAGS *
5861 DIV_ROUND_UP(PAGE_SIZE,
5862 adapter->tx_fifo_limit) + 2));
5863
5864 if (!skb->xmit_more ||
5865 netif_xmit_stopped(netdev_get_tx_queue(netdev, 0))) {
5866 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
5867 e1000e_update_tdt_wa(tx_ring,
5868 tx_ring->next_to_use);
5869 else
5870 writel(tx_ring->next_to_use, tx_ring->tail);
5871
5872 /* we need this if more than one processor can write
5873 * to our tail at a time, it synchronizes IO on
5874 *IA64/Altix systems
5875 */
5876 mmiowb();
5877 }
5878 } else {
5879 dev_kfree_skb_any(skb);
5880 tx_ring->buffer_info[first].time_stamp = 0;
5881 tx_ring->next_to_use = first;
5882 }
5883
5884 return NETDEV_TX_OK;
5885}
5886
5887/**
5888 * e1000_tx_timeout - Respond to a Tx Hang
5889 * @netdev: network interface device structure
5890 **/
5891static void e1000_tx_timeout(struct net_device *netdev)
5892{
5893 struct e1000_adapter *adapter = netdev_priv(netdev);
5894
5895 /* Do the reset outside of interrupt context */
5896 adapter->tx_timeout_count++;
5897 schedule_work(&adapter->reset_task);
5898}
5899
5900static void e1000_reset_task(struct work_struct *work)
5901{
5902 struct e1000_adapter *adapter;
5903 adapter = container_of(work, struct e1000_adapter, reset_task);
5904
5905 /* don't run the task if already down */
5906 if (test_bit(__E1000_DOWN, &adapter->state))
5907 return;
5908
5909 if (!(adapter->flags & FLAG_RESTART_NOW)) {
5910 e1000e_dump(adapter);
5911 e_err("Reset adapter unexpectedly\n");
5912 }
5913 e1000e_reinit_locked(adapter);
5914}
5915
5916/**
5917 * e1000_get_stats64 - Get System Network Statistics
5918 * @netdev: network interface device structure
5919 * @stats: rtnl_link_stats64 pointer
5920 *
5921 * Returns the address of the device statistics structure.
5922 **/
5923struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
5924 struct rtnl_link_stats64 *stats)
5925{
5926 struct e1000_adapter *adapter = netdev_priv(netdev);
5927
5928 memset(stats, 0, sizeof(struct rtnl_link_stats64));
5929 spin_lock(&adapter->stats64_lock);
5930 e1000e_update_stats(adapter);
5931 /* Fill out the OS statistics structure */
5932 stats->rx_bytes = adapter->stats.gorc;
5933 stats->rx_packets = adapter->stats.gprc;
5934 stats->tx_bytes = adapter->stats.gotc;
5935 stats->tx_packets = adapter->stats.gptc;
5936 stats->multicast = adapter->stats.mprc;
5937 stats->collisions = adapter->stats.colc;
5938
5939 /* Rx Errors */
5940
5941 /* RLEC on some newer hardware can be incorrect so build
5942 * our own version based on RUC and ROC
5943 */
5944 stats->rx_errors = adapter->stats.rxerrc +
5945 adapter->stats.crcerrs + adapter->stats.algnerrc +
5946 adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
5947 stats->rx_length_errors = adapter->stats.ruc + adapter->stats.roc;
5948 stats->rx_crc_errors = adapter->stats.crcerrs;
5949 stats->rx_frame_errors = adapter->stats.algnerrc;
5950 stats->rx_missed_errors = adapter->stats.mpc;
5951
5952 /* Tx Errors */
5953 stats->tx_errors = adapter->stats.ecol + adapter->stats.latecol;
5954 stats->tx_aborted_errors = adapter->stats.ecol;
5955 stats->tx_window_errors = adapter->stats.latecol;
5956 stats->tx_carrier_errors = adapter->stats.tncrs;
5957
5958 /* Tx Dropped needs to be maintained elsewhere */
5959
5960 spin_unlock(&adapter->stats64_lock);
5961 return stats;
5962}
5963
5964/**
5965 * e1000_change_mtu - Change the Maximum Transfer Unit
5966 * @netdev: network interface device structure
5967 * @new_mtu: new value for maximum frame size
5968 *
5969 * Returns 0 on success, negative on failure
5970 **/
5971static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5972{
5973 struct e1000_adapter *adapter = netdev_priv(netdev);
5974 int max_frame = new_mtu + VLAN_ETH_HLEN + ETH_FCS_LEN;
5975
5976 /* Jumbo frame support */
5977 if ((new_mtu > ETH_DATA_LEN) &&
5978 !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5979 e_err("Jumbo Frames not supported.\n");
5980 return -EINVAL;
5981 }
5982
5983 /* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
5984 if ((adapter->hw.mac.type >= e1000_pch2lan) &&
5985 !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5986 (new_mtu > ETH_DATA_LEN)) {
5987 e_err("Jumbo Frames not supported on this device when CRC stripping is disabled.\n");
5988 return -EINVAL;
5989 }
5990
5991 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5992 usleep_range(1000, 2000);
5993 /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5994 adapter->max_frame_size = max_frame;
5995 e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5996 netdev->mtu = new_mtu;
5997
5998 pm_runtime_get_sync(netdev->dev.parent);
5999
6000 if (netif_running(netdev))
6001 e1000e_down(adapter, true);
6002
6003 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
6004 * means we reserve 2 more, this pushes us to allocate from the next
6005 * larger slab size.
6006 * i.e. RXBUFFER_2048 --> size-4096 slab
6007 * However with the new *_jumbo_rx* routines, jumbo receives will use
6008 * fragmented skbs
6009 */
6010
6011 if (max_frame <= 2048)
6012 adapter->rx_buffer_len = 2048;
6013 else
6014 adapter->rx_buffer_len = 4096;
6015
6016 /* adjust allocation if LPE protects us, and we aren't using SBP */
6017 if (max_frame <= (VLAN_ETH_FRAME_LEN + ETH_FCS_LEN))
6018 adapter->rx_buffer_len = VLAN_ETH_FRAME_LEN + ETH_FCS_LEN;
6019
6020 if (netif_running(netdev))
6021 e1000e_up(adapter);
6022 else
6023 e1000e_reset(adapter);
6024
6025 pm_runtime_put_sync(netdev->dev.parent);
6026
6027 clear_bit(__E1000_RESETTING, &adapter->state);
6028
6029 return 0;
6030}
6031
6032static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
6033 int cmd)
6034{
6035 struct e1000_adapter *adapter = netdev_priv(netdev);
6036 struct mii_ioctl_data *data = if_mii(ifr);
6037
6038 if (adapter->hw.phy.media_type != e1000_media_type_copper)
6039 return -EOPNOTSUPP;
6040
6041 switch (cmd) {
6042 case SIOCGMIIPHY:
6043 data->phy_id = adapter->hw.phy.addr;
6044 break;
6045 case SIOCGMIIREG:
6046 e1000_phy_read_status(adapter);
6047
6048 switch (data->reg_num & 0x1F) {
6049 case MII_BMCR:
6050 data->val_out = adapter->phy_regs.bmcr;
6051 break;
6052 case MII_BMSR:
6053 data->val_out = adapter->phy_regs.bmsr;
6054 break;
6055 case MII_PHYSID1:
6056 data->val_out = (adapter->hw.phy.id >> 16);
6057 break;
6058 case MII_PHYSID2:
6059 data->val_out = (adapter->hw.phy.id & 0xFFFF);
6060 break;
6061 case MII_ADVERTISE:
6062 data->val_out = adapter->phy_regs.advertise;
6063 break;
6064 case MII_LPA:
6065 data->val_out = adapter->phy_regs.lpa;
6066 break;
6067 case MII_EXPANSION:
6068 data->val_out = adapter->phy_regs.expansion;
6069 break;
6070 case MII_CTRL1000:
6071 data->val_out = adapter->phy_regs.ctrl1000;
6072 break;
6073 case MII_STAT1000:
6074 data->val_out = adapter->phy_regs.stat1000;
6075 break;
6076 case MII_ESTATUS:
6077 data->val_out = adapter->phy_regs.estatus;
6078 break;
6079 default:
6080 return -EIO;
6081 }
6082 break;
6083 case SIOCSMIIREG:
6084 default:
6085 return -EOPNOTSUPP;
6086 }
6087 return 0;
6088}
6089
6090/**
6091 * e1000e_hwtstamp_ioctl - control hardware time stamping
6092 * @netdev: network interface device structure
6093 * @ifreq: interface request
6094 *
6095 * Outgoing time stamping can be enabled and disabled. Play nice and
6096 * disable it when requested, although it shouldn't cause any overhead
6097 * when no packet needs it. At most one packet in the queue may be
6098 * marked for time stamping, otherwise it would be impossible to tell
6099 * for sure to which packet the hardware time stamp belongs.
6100 *
6101 * Incoming time stamping has to be configured via the hardware filters.
6102 * Not all combinations are supported, in particular event type has to be
6103 * specified. Matching the kind of event packet is not supported, with the
6104 * exception of "all V2 events regardless of level 2 or 4".
6105 **/
6106static int e1000e_hwtstamp_set(struct net_device *netdev, struct ifreq *ifr)
6107{
6108 struct e1000_adapter *adapter = netdev_priv(netdev);
6109 struct hwtstamp_config config;
6110 int ret_val;
6111
6112 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
6113 return -EFAULT;
6114
6115 ret_val = e1000e_config_hwtstamp(adapter, &config);
6116 if (ret_val)
6117 return ret_val;
6118
6119 switch (config.rx_filter) {
6120 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
6121 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
6122 case HWTSTAMP_FILTER_PTP_V2_SYNC:
6123 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
6124 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
6125 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
6126 /* With V2 type filters which specify a Sync or Delay Request,
6127 * Path Delay Request/Response messages are also time stamped
6128 * by hardware so notify the caller the requested packets plus
6129 * some others are time stamped.
6130 */
6131 config.rx_filter = HWTSTAMP_FILTER_SOME;
6132 break;
6133 default:
6134 break;
6135 }
6136
6137 return copy_to_user(ifr->ifr_data, &config,
6138 sizeof(config)) ? -EFAULT : 0;
6139}
6140
6141static int e1000e_hwtstamp_get(struct net_device *netdev, struct ifreq *ifr)
6142{
6143 struct e1000_adapter *adapter = netdev_priv(netdev);
6144
6145 return copy_to_user(ifr->ifr_data, &adapter->hwtstamp_config,
6146 sizeof(adapter->hwtstamp_config)) ? -EFAULT : 0;
6147}
6148
6149static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6150{
6151 switch (cmd) {
6152 case SIOCGMIIPHY:
6153 case SIOCGMIIREG:
6154 case SIOCSMIIREG:
6155 return e1000_mii_ioctl(netdev, ifr, cmd);
6156 case SIOCSHWTSTAMP:
6157 return e1000e_hwtstamp_set(netdev, ifr);
6158 case SIOCGHWTSTAMP:
6159 return e1000e_hwtstamp_get(netdev, ifr);
6160 default:
6161 return -EOPNOTSUPP;
6162 }
6163}
6164
6165static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
6166{
6167 struct e1000_hw *hw = &adapter->hw;
6168 u32 i, mac_reg, wuc;
6169 u16 phy_reg, wuc_enable;
6170 int retval;
6171
6172 /* copy MAC RARs to PHY RARs */
6173 e1000_copy_rx_addrs_to_phy_ich8lan(hw);
6174
6175 retval = hw->phy.ops.acquire(hw);
6176 if (retval) {
6177 e_err("Could not acquire PHY\n");
6178 return retval;
6179 }
6180
6181 /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
6182 retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
6183 if (retval)
6184 goto release;
6185
6186 /* copy MAC MTA to PHY MTA - only needed for pchlan */
6187 for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
6188 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
6189 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
6190 (u16)(mac_reg & 0xFFFF));
6191 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
6192 (u16)((mac_reg >> 16) & 0xFFFF));
6193 }
6194
6195 /* configure PHY Rx Control register */
6196 hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
6197 mac_reg = er32(RCTL);
6198 if (mac_reg & E1000_RCTL_UPE)
6199 phy_reg |= BM_RCTL_UPE;
6200 if (mac_reg & E1000_RCTL_MPE)
6201 phy_reg |= BM_RCTL_MPE;
6202 phy_reg &= ~(BM_RCTL_MO_MASK);
6203 if (mac_reg & E1000_RCTL_MO_3)
6204 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
6205 << BM_RCTL_MO_SHIFT);
6206 if (mac_reg & E1000_RCTL_BAM)
6207 phy_reg |= BM_RCTL_BAM;
6208 if (mac_reg & E1000_RCTL_PMCF)
6209 phy_reg |= BM_RCTL_PMCF;
6210 mac_reg = er32(CTRL);
6211 if (mac_reg & E1000_CTRL_RFCE)
6212 phy_reg |= BM_RCTL_RFCE;
6213 hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
6214
6215 wuc = E1000_WUC_PME_EN;
6216 if (wufc & (E1000_WUFC_MAG | E1000_WUFC_LNKC))
6217 wuc |= E1000_WUC_APME;
6218
6219 /* enable PHY wakeup in MAC register */
6220 ew32(WUFC, wufc);
6221 ew32(WUC, (E1000_WUC_PHY_WAKE | E1000_WUC_APMPME |
6222 E1000_WUC_PME_STATUS | wuc));
6223
6224 /* configure and enable PHY wakeup in PHY registers */
6225 hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
6226 hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, wuc);
6227
6228 /* activate PHY wakeup */
6229 wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
6230 retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
6231 if (retval)
6232 e_err("Could not set PHY Host Wakeup bit\n");
6233release:
6234 hw->phy.ops.release(hw);
6235
6236 return retval;
6237}
6238
6239static void e1000e_flush_lpic(struct pci_dev *pdev)
6240{
6241 struct net_device *netdev = pci_get_drvdata(pdev);
6242 struct e1000_adapter *adapter = netdev_priv(netdev);
6243 struct e1000_hw *hw = &adapter->hw;
6244 u32 ret_val;
6245
6246 pm_runtime_get_sync(netdev->dev.parent);
6247
6248 ret_val = hw->phy.ops.acquire(hw);
6249 if (ret_val)
6250 goto fl_out;
6251
6252 pr_info("EEE TX LPI TIMER: %08X\n",
6253 er32(LPIC) >> E1000_LPIC_LPIET_SHIFT);
6254
6255 hw->phy.ops.release(hw);
6256
6257fl_out:
6258 pm_runtime_put_sync(netdev->dev.parent);
6259}
6260
6261static int e1000e_pm_freeze(struct device *dev)
6262{
6263 struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
6264 struct e1000_adapter *adapter = netdev_priv(netdev);
6265
6266 netif_device_detach(netdev);
6267
6268 if (netif_running(netdev)) {
6269 int count = E1000_CHECK_RESET_COUNT;
6270
6271 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
6272 usleep_range(10000, 20000);
6273
6274 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
6275
6276 /* Quiesce the device without resetting the hardware */
6277 e1000e_down(adapter, false);
6278 e1000_free_irq(adapter);
6279 }
6280 e1000e_reset_interrupt_capability(adapter);
6281
6282 /* Allow time for pending master requests to run */
6283 e1000e_disable_pcie_master(&adapter->hw);
6284
6285 return 0;
6286}
6287
6288static int __e1000_shutdown(struct pci_dev *pdev, bool runtime)
6289{
6290 struct net_device *netdev = pci_get_drvdata(pdev);
6291 struct e1000_adapter *adapter = netdev_priv(netdev);
6292 struct e1000_hw *hw = &adapter->hw;
6293 u32 ctrl, ctrl_ext, rctl, status;
6294 /* Runtime suspend should only enable wakeup for link changes */
6295 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
6296 int retval = 0;
6297
6298 status = er32(STATUS);
6299 if (status & E1000_STATUS_LU)
6300 wufc &= ~E1000_WUFC_LNKC;
6301
6302 if (wufc) {
6303 e1000_setup_rctl(adapter);
6304 e1000e_set_rx_mode(netdev);
6305
6306 /* turn on all-multi mode if wake on multicast is enabled */
6307 if (wufc & E1000_WUFC_MC) {
6308 rctl = er32(RCTL);
6309 rctl |= E1000_RCTL_MPE;
6310 ew32(RCTL, rctl);
6311 }
6312
6313 ctrl = er32(CTRL);
6314 ctrl |= E1000_CTRL_ADVD3WUC;
6315 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
6316 ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
6317 ew32(CTRL, ctrl);
6318
6319 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
6320 adapter->hw.phy.media_type ==
6321 e1000_media_type_internal_serdes) {
6322 /* keep the laser running in D3 */
6323 ctrl_ext = er32(CTRL_EXT);
6324 ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
6325 ew32(CTRL_EXT, ctrl_ext);
6326 }
6327
6328 if (!runtime)
6329 e1000e_power_up_phy(adapter);
6330
6331 if (adapter->flags & FLAG_IS_ICH)
6332 e1000_suspend_workarounds_ich8lan(&adapter->hw);
6333
6334 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
6335 /* enable wakeup by the PHY */
6336 retval = e1000_init_phy_wakeup(adapter, wufc);
6337 if (retval)
6338 return retval;
6339 } else {
6340 /* enable wakeup by the MAC */
6341 ew32(WUFC, wufc);
6342 ew32(WUC, E1000_WUC_PME_EN);
6343 }
6344 } else {
6345 ew32(WUC, 0);
6346 ew32(WUFC, 0);
6347
6348 e1000_power_down_phy(adapter);
6349 }
6350
6351 if (adapter->hw.phy.type == e1000_phy_igp_3) {
6352 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
6353 } else if ((hw->mac.type == e1000_pch_lpt) ||
6354 (hw->mac.type == e1000_pch_spt)) {
6355 if (!(wufc & (E1000_WUFC_EX | E1000_WUFC_MC | E1000_WUFC_BC)))
6356 /* ULP does not support wake from unicast, multicast
6357 * or broadcast.
6358 */
6359 retval = e1000_enable_ulp_lpt_lp(hw, !runtime);
6360
6361 if (retval)
6362 return retval;
6363 }
6364
6365 /* Ensure that the appropriate bits are set in LPI_CTRL
6366 * for EEE in Sx
6367 */
6368 if ((hw->phy.type >= e1000_phy_i217) &&
6369 adapter->eee_advert && hw->dev_spec.ich8lan.eee_lp_ability) {
6370 u16 lpi_ctrl = 0;
6371
6372 retval = hw->phy.ops.acquire(hw);
6373 if (!retval) {
6374 retval = e1e_rphy_locked(hw, I82579_LPI_CTRL,
6375 &lpi_ctrl);
6376 if (!retval) {
6377 if (adapter->eee_advert &
6378 hw->dev_spec.ich8lan.eee_lp_ability &
6379 I82579_EEE_100_SUPPORTED)
6380 lpi_ctrl |= I82579_LPI_CTRL_100_ENABLE;
6381 if (adapter->eee_advert &
6382 hw->dev_spec.ich8lan.eee_lp_ability &
6383 I82579_EEE_1000_SUPPORTED)
6384 lpi_ctrl |= I82579_LPI_CTRL_1000_ENABLE;
6385
6386 retval = e1e_wphy_locked(hw, I82579_LPI_CTRL,
6387 lpi_ctrl);
6388 }
6389 }
6390 hw->phy.ops.release(hw);
6391 }
6392
6393 /* Release control of h/w to f/w. If f/w is AMT enabled, this
6394 * would have already happened in close and is redundant.
6395 */
6396 e1000e_release_hw_control(adapter);
6397
6398 pci_clear_master(pdev);
6399
6400 /* The pci-e switch on some quad port adapters will report a
6401 * correctable error when the MAC transitions from D0 to D3. To
6402 * prevent this we need to mask off the correctable errors on the
6403 * downstream port of the pci-e switch.
6404 *
6405 * We don't have the associated upstream bridge while assigning
6406 * the PCI device into guest. For example, the KVM on power is
6407 * one of the cases.
6408 */
6409 if (adapter->flags & FLAG_IS_QUAD_PORT) {
6410 struct pci_dev *us_dev = pdev->bus->self;
6411 u16 devctl;
6412
6413 if (!us_dev)
6414 return 0;
6415
6416 pcie_capability_read_word(us_dev, PCI_EXP_DEVCTL, &devctl);
6417 pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL,
6418 (devctl & ~PCI_EXP_DEVCTL_CERE));
6419
6420 pci_save_state(pdev);
6421 pci_prepare_to_sleep(pdev);
6422
6423 pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL, devctl);
6424 }
6425
6426 return 0;
6427}
6428
6429/**
6430 * __e1000e_disable_aspm - Disable ASPM states
6431 * @pdev: pointer to PCI device struct
6432 * @state: bit-mask of ASPM states to disable
6433 * @locked: indication if this context holds pci_bus_sem locked.
6434 *
6435 * Some devices *must* have certain ASPM states disabled per hardware errata.
6436 **/
6437static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state, int locked)
6438{
6439 struct pci_dev *parent = pdev->bus->self;
6440 u16 aspm_dis_mask = 0;
6441 u16 pdev_aspmc, parent_aspmc;
6442
6443 switch (state) {
6444 case PCIE_LINK_STATE_L0S:
6445 case PCIE_LINK_STATE_L0S | PCIE_LINK_STATE_L1:
6446 aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L0S;
6447 /* fall-through - can't have L1 without L0s */
6448 case PCIE_LINK_STATE_L1:
6449 aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L1;
6450 break;
6451 default:
6452 return;
6453 }
6454
6455 pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6456 pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6457
6458 if (parent) {
6459 pcie_capability_read_word(parent, PCI_EXP_LNKCTL,
6460 &parent_aspmc);
6461 parent_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6462 }
6463
6464 /* Nothing to do if the ASPM states to be disabled already are */
6465 if (!(pdev_aspmc & aspm_dis_mask) &&
6466 (!parent || !(parent_aspmc & aspm_dis_mask)))
6467 return;
6468
6469 dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
6470 (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L0S) ?
6471 "L0s" : "",
6472 (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L1) ?
6473 "L1" : "");
6474
6475#ifdef CONFIG_PCIEASPM
6476 if (locked)
6477 pci_disable_link_state_locked(pdev, state);
6478 else
6479 pci_disable_link_state(pdev, state);
6480
6481 /* Double-check ASPM control. If not disabled by the above, the
6482 * BIOS is preventing that from happening (or CONFIG_PCIEASPM is
6483 * not enabled); override by writing PCI config space directly.
6484 */
6485 pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6486 pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6487
6488 if (!(aspm_dis_mask & pdev_aspmc))
6489 return;
6490#endif
6491
6492 /* Both device and parent should have the same ASPM setting.
6493 * Disable ASPM in downstream component first and then upstream.
6494 */
6495 pcie_capability_clear_word(pdev, PCI_EXP_LNKCTL, aspm_dis_mask);
6496
6497 if (parent)
6498 pcie_capability_clear_word(parent, PCI_EXP_LNKCTL,
6499 aspm_dis_mask);
6500}
6501
6502/**
6503 * e1000e_disable_aspm - Disable ASPM states.
6504 * @pdev: pointer to PCI device struct
6505 * @state: bit-mask of ASPM states to disable
6506 *
6507 * This function acquires the pci_bus_sem!
6508 * Some devices *must* have certain ASPM states disabled per hardware errata.
6509 **/
6510static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
6511{
6512 __e1000e_disable_aspm(pdev, state, 0);
6513}
6514
6515/**
6516 * e1000e_disable_aspm_locked Disable ASPM states.
6517 * @pdev: pointer to PCI device struct
6518 * @state: bit-mask of ASPM states to disable
6519 *
6520 * This function must be called with pci_bus_sem acquired!
6521 * Some devices *must* have certain ASPM states disabled per hardware errata.
6522 **/
6523static void e1000e_disable_aspm_locked(struct pci_dev *pdev, u16 state)
6524{
6525 __e1000e_disable_aspm(pdev, state, 1);
6526}
6527
6528#ifdef CONFIG_PM
6529static int __e1000_resume(struct pci_dev *pdev)
6530{
6531 struct net_device *netdev = pci_get_drvdata(pdev);
6532 struct e1000_adapter *adapter = netdev_priv(netdev);
6533 struct e1000_hw *hw = &adapter->hw;
6534 u16 aspm_disable_flag = 0;
6535
6536 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
6537 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6538 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
6539 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6540 if (aspm_disable_flag)
6541 e1000e_disable_aspm(pdev, aspm_disable_flag);
6542
6543 pci_set_master(pdev);
6544
6545 if (hw->mac.type >= e1000_pch2lan)
6546 e1000_resume_workarounds_pchlan(&adapter->hw);
6547
6548 e1000e_power_up_phy(adapter);
6549
6550 /* report the system wakeup cause from S3/S4 */
6551 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
6552 u16 phy_data;
6553
6554 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
6555 if (phy_data) {
6556 e_info("PHY Wakeup cause - %s\n",
6557 phy_data & E1000_WUS_EX ? "Unicast Packet" :
6558 phy_data & E1000_WUS_MC ? "Multicast Packet" :
6559 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
6560 phy_data & E1000_WUS_MAG ? "Magic Packet" :
6561 phy_data & E1000_WUS_LNKC ?
6562 "Link Status Change" : "other");
6563 }
6564 e1e_wphy(&adapter->hw, BM_WUS, ~0);
6565 } else {
6566 u32 wus = er32(WUS);
6567
6568 if (wus) {
6569 e_info("MAC Wakeup cause - %s\n",
6570 wus & E1000_WUS_EX ? "Unicast Packet" :
6571 wus & E1000_WUS_MC ? "Multicast Packet" :
6572 wus & E1000_WUS_BC ? "Broadcast Packet" :
6573 wus & E1000_WUS_MAG ? "Magic Packet" :
6574 wus & E1000_WUS_LNKC ? "Link Status Change" :
6575 "other");
6576 }
6577 ew32(WUS, ~0);
6578 }
6579
6580 e1000e_reset(adapter);
6581
6582 e1000_init_manageability_pt(adapter);
6583
6584 /* If the controller has AMT, do not set DRV_LOAD until the interface
6585 * is up. For all other cases, let the f/w know that the h/w is now
6586 * under the control of the driver.
6587 */
6588 if (!(adapter->flags & FLAG_HAS_AMT))
6589 e1000e_get_hw_control(adapter);
6590
6591 return 0;
6592}
6593
6594#ifdef CONFIG_PM_SLEEP
6595static int e1000e_pm_thaw(struct device *dev)
6596{
6597 struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
6598 struct e1000_adapter *adapter = netdev_priv(netdev);
6599
6600 e1000e_set_interrupt_capability(adapter);
6601 if (netif_running(netdev)) {
6602 u32 err = e1000_request_irq(adapter);
6603
6604 if (err)
6605 return err;
6606
6607 e1000e_up(adapter);
6608 }
6609
6610 netif_device_attach(netdev);
6611
6612 return 0;
6613}
6614
6615static int e1000e_pm_suspend(struct device *dev)
6616{
6617 struct pci_dev *pdev = to_pci_dev(dev);
6618
6619 e1000e_flush_lpic(pdev);
6620
6621 e1000e_pm_freeze(dev);
6622
6623 return __e1000_shutdown(pdev, false);
6624}
6625
6626static int e1000e_pm_resume(struct device *dev)
6627{
6628 struct pci_dev *pdev = to_pci_dev(dev);
6629 int rc;
6630
6631 rc = __e1000_resume(pdev);
6632 if (rc)
6633 return rc;
6634
6635 return e1000e_pm_thaw(dev);
6636}
6637#endif /* CONFIG_PM_SLEEP */
6638
6639static int e1000e_pm_runtime_idle(struct device *dev)
6640{
6641 struct pci_dev *pdev = to_pci_dev(dev);
6642 struct net_device *netdev = pci_get_drvdata(pdev);
6643 struct e1000_adapter *adapter = netdev_priv(netdev);
6644 u16 eee_lp;
6645
6646 eee_lp = adapter->hw.dev_spec.ich8lan.eee_lp_ability;
6647
6648 if (!e1000e_has_link(adapter)) {
6649 adapter->hw.dev_spec.ich8lan.eee_lp_ability = eee_lp;
6650 pm_schedule_suspend(dev, 5 * MSEC_PER_SEC);
6651 }
6652
6653 return -EBUSY;
6654}
6655
6656static int e1000e_pm_runtime_resume(struct device *dev)
6657{
6658 struct pci_dev *pdev = to_pci_dev(dev);
6659 struct net_device *netdev = pci_get_drvdata(pdev);
6660 struct e1000_adapter *adapter = netdev_priv(netdev);
6661 int rc;
6662
6663 rc = __e1000_resume(pdev);
6664 if (rc)
6665 return rc;
6666
6667 if (netdev->flags & IFF_UP)
6668 e1000e_up(adapter);
6669
6670 return rc;
6671}
6672
6673static int e1000e_pm_runtime_suspend(struct device *dev)
6674{
6675 struct pci_dev *pdev = to_pci_dev(dev);
6676 struct net_device *netdev = pci_get_drvdata(pdev);
6677 struct e1000_adapter *adapter = netdev_priv(netdev);
6678
6679 if (netdev->flags & IFF_UP) {
6680 int count = E1000_CHECK_RESET_COUNT;
6681
6682 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
6683 usleep_range(10000, 20000);
6684
6685 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
6686
6687 /* Down the device without resetting the hardware */
6688 e1000e_down(adapter, false);
6689 }
6690
6691 if (__e1000_shutdown(pdev, true)) {
6692 e1000e_pm_runtime_resume(dev);
6693 return -EBUSY;
6694 }
6695
6696 return 0;
6697}
6698#endif /* CONFIG_PM */
6699
6700static void e1000_shutdown(struct pci_dev *pdev)
6701{
6702 e1000e_flush_lpic(pdev);
6703
6704 e1000e_pm_freeze(&pdev->dev);
6705
6706 __e1000_shutdown(pdev, false);
6707}
6708
6709#ifdef CONFIG_NET_POLL_CONTROLLER
6710
6711static irqreturn_t e1000_intr_msix(int __always_unused irq, void *data)
6712{
6713 struct net_device *netdev = data;
6714 struct e1000_adapter *adapter = netdev_priv(netdev);
6715
6716 if (adapter->msix_entries) {
6717 int vector, msix_irq;
6718
6719 vector = 0;
6720 msix_irq = adapter->msix_entries[vector].vector;
6721 disable_irq(msix_irq);
6722 e1000_intr_msix_rx(msix_irq, netdev);
6723 enable_irq(msix_irq);
6724
6725 vector++;
6726 msix_irq = adapter->msix_entries[vector].vector;
6727 disable_irq(msix_irq);
6728 e1000_intr_msix_tx(msix_irq, netdev);
6729 enable_irq(msix_irq);
6730
6731 vector++;
6732 msix_irq = adapter->msix_entries[vector].vector;
6733 disable_irq(msix_irq);
6734 e1000_msix_other(msix_irq, netdev);
6735 enable_irq(msix_irq);
6736 }
6737
6738 return IRQ_HANDLED;
6739}
6740
6741/**
6742 * e1000_netpoll
6743 * @netdev: network interface device structure
6744 *
6745 * Polling 'interrupt' - used by things like netconsole to send skbs
6746 * without having to re-enable interrupts. It's not called while
6747 * the interrupt routine is executing.
6748 */
6749static void e1000_netpoll(struct net_device *netdev)
6750{
6751 struct e1000_adapter *adapter = netdev_priv(netdev);
6752
6753 switch (adapter->int_mode) {
6754 case E1000E_INT_MODE_MSIX:
6755 e1000_intr_msix(adapter->pdev->irq, netdev);
6756 break;
6757 case E1000E_INT_MODE_MSI:
6758 if (disable_hardirq(adapter->pdev->irq))
6759 e1000_intr_msi(adapter->pdev->irq, netdev);
6760 enable_irq(adapter->pdev->irq);
6761 break;
6762 default: /* E1000E_INT_MODE_LEGACY */
6763 if (disable_hardirq(adapter->pdev->irq))
6764 e1000_intr(adapter->pdev->irq, netdev);
6765 enable_irq(adapter->pdev->irq);
6766 break;
6767 }
6768}
6769#endif
6770
6771/**
6772 * e1000_io_error_detected - called when PCI error is detected
6773 * @pdev: Pointer to PCI device
6774 * @state: The current pci connection state
6775 *
6776 * This function is called after a PCI bus error affecting
6777 * this device has been detected.
6778 */
6779static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
6780 pci_channel_state_t state)
6781{
6782 struct net_device *netdev = pci_get_drvdata(pdev);
6783 struct e1000_adapter *adapter = netdev_priv(netdev);
6784
6785 netif_device_detach(netdev);
6786
6787 if (state == pci_channel_io_perm_failure)
6788 return PCI_ERS_RESULT_DISCONNECT;
6789
6790 if (netif_running(netdev))
6791 e1000e_down(adapter, true);
6792 pci_disable_device(pdev);
6793
6794 /* Request a slot slot reset. */
6795 return PCI_ERS_RESULT_NEED_RESET;
6796}
6797
6798/**
6799 * e1000_io_slot_reset - called after the pci bus has been reset.
6800 * @pdev: Pointer to PCI device
6801 *
6802 * Restart the card from scratch, as if from a cold-boot. Implementation
6803 * resembles the first-half of the e1000e_pm_resume routine.
6804 */
6805static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
6806{
6807 struct net_device *netdev = pci_get_drvdata(pdev);
6808 struct e1000_adapter *adapter = netdev_priv(netdev);
6809 struct e1000_hw *hw = &adapter->hw;
6810 u16 aspm_disable_flag = 0;
6811 int err;
6812 pci_ers_result_t result;
6813
6814 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
6815 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6816 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
6817 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6818 if (aspm_disable_flag)
6819 e1000e_disable_aspm_locked(pdev, aspm_disable_flag);
6820
6821 err = pci_enable_device_mem(pdev);
6822 if (err) {
6823 dev_err(&pdev->dev,
6824 "Cannot re-enable PCI device after reset.\n");
6825 result = PCI_ERS_RESULT_DISCONNECT;
6826 } else {
6827 pdev->state_saved = true;
6828 pci_restore_state(pdev);
6829 pci_set_master(pdev);
6830
6831 pci_enable_wake(pdev, PCI_D3hot, 0);
6832 pci_enable_wake(pdev, PCI_D3cold, 0);
6833
6834 e1000e_reset(adapter);
6835 ew32(WUS, ~0);
6836 result = PCI_ERS_RESULT_RECOVERED;
6837 }
6838
6839 pci_cleanup_aer_uncorrect_error_status(pdev);
6840
6841 return result;
6842}
6843
6844/**
6845 * e1000_io_resume - called when traffic can start flowing again.
6846 * @pdev: Pointer to PCI device
6847 *
6848 * This callback is called when the error recovery driver tells us that
6849 * its OK to resume normal operation. Implementation resembles the
6850 * second-half of the e1000e_pm_resume routine.
6851 */
6852static void e1000_io_resume(struct pci_dev *pdev)
6853{
6854 struct net_device *netdev = pci_get_drvdata(pdev);
6855 struct e1000_adapter *adapter = netdev_priv(netdev);
6856
6857 e1000_init_manageability_pt(adapter);
6858
6859 if (netif_running(netdev))
6860 e1000e_up(adapter);
6861
6862 netif_device_attach(netdev);
6863
6864 /* If the controller has AMT, do not set DRV_LOAD until the interface
6865 * is up. For all other cases, let the f/w know that the h/w is now
6866 * under the control of the driver.
6867 */
6868 if (!(adapter->flags & FLAG_HAS_AMT))
6869 e1000e_get_hw_control(adapter);
6870}
6871
6872static void e1000_print_device_info(struct e1000_adapter *adapter)
6873{
6874 struct e1000_hw *hw = &adapter->hw;
6875 struct net_device *netdev = adapter->netdev;
6876 u32 ret_val;
6877 u8 pba_str[E1000_PBANUM_LENGTH];
6878
6879 /* print bus type/speed/width info */
6880 e_info("(PCI Express:2.5GT/s:%s) %pM\n",
6881 /* bus width */
6882 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
6883 "Width x1"),
6884 /* MAC address */
6885 netdev->dev_addr);
6886 e_info("Intel(R) PRO/%s Network Connection\n",
6887 (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
6888 ret_val = e1000_read_pba_string_generic(hw, pba_str,
6889 E1000_PBANUM_LENGTH);
6890 if (ret_val)
6891 strlcpy((char *)pba_str, "Unknown", sizeof(pba_str));
6892 e_info("MAC: %d, PHY: %d, PBA No: %s\n",
6893 hw->mac.type, hw->phy.type, pba_str);
6894}
6895
6896static void e1000_eeprom_checks(struct e1000_adapter *adapter)
6897{
6898 struct e1000_hw *hw = &adapter->hw;
6899 int ret_val;
6900 u16 buf = 0;
6901
6902 if (hw->mac.type != e1000_82573)
6903 return;
6904
6905 ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
6906 le16_to_cpus(&buf);
6907 if (!ret_val && (!(buf & BIT(0)))) {
6908 /* Deep Smart Power Down (DSPD) */
6909 dev_warn(&adapter->pdev->dev,
6910 "Warning: detected DSPD enabled in EEPROM\n");
6911 }
6912}
6913
6914static netdev_features_t e1000_fix_features(struct net_device *netdev,
6915 netdev_features_t features)
6916{
6917 struct e1000_adapter *adapter = netdev_priv(netdev);
6918 struct e1000_hw *hw = &adapter->hw;
6919
6920 /* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
6921 if ((hw->mac.type >= e1000_pch2lan) && (netdev->mtu > ETH_DATA_LEN))
6922 features &= ~NETIF_F_RXFCS;
6923
6924 /* Since there is no support for separate Rx/Tx vlan accel
6925 * enable/disable make sure Tx flag is always in same state as Rx.
6926 */
6927 if (features & NETIF_F_HW_VLAN_CTAG_RX)
6928 features |= NETIF_F_HW_VLAN_CTAG_TX;
6929 else
6930 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
6931
6932 return features;
6933}
6934
6935static int e1000_set_features(struct net_device *netdev,
6936 netdev_features_t features)
6937{
6938 struct e1000_adapter *adapter = netdev_priv(netdev);
6939 netdev_features_t changed = features ^ netdev->features;
6940
6941 if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
6942 adapter->flags |= FLAG_TSO_FORCE;
6943
6944 if (!(changed & (NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_TX |
6945 NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_RXFCS |
6946 NETIF_F_RXALL)))
6947 return 0;
6948
6949 if (changed & NETIF_F_RXFCS) {
6950 if (features & NETIF_F_RXFCS) {
6951 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6952 } else {
6953 /* We need to take it back to defaults, which might mean
6954 * stripping is still disabled at the adapter level.
6955 */
6956 if (adapter->flags2 & FLAG2_DFLT_CRC_STRIPPING)
6957 adapter->flags2 |= FLAG2_CRC_STRIPPING;
6958 else
6959 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6960 }
6961 }
6962
6963 netdev->features = features;
6964
6965 if (netif_running(netdev))
6966 e1000e_reinit_locked(adapter);
6967 else
6968 e1000e_reset(adapter);
6969
6970 return 0;
6971}
6972
6973static const struct net_device_ops e1000e_netdev_ops = {
6974 .ndo_open = e1000e_open,
6975 .ndo_stop = e1000e_close,
6976 .ndo_start_xmit = e1000_xmit_frame,
6977 .ndo_get_stats64 = e1000e_get_stats64,
6978 .ndo_set_rx_mode = e1000e_set_rx_mode,
6979 .ndo_set_mac_address = e1000_set_mac,
6980 .ndo_change_mtu = e1000_change_mtu,
6981 .ndo_do_ioctl = e1000_ioctl,
6982 .ndo_tx_timeout = e1000_tx_timeout,
6983 .ndo_validate_addr = eth_validate_addr,
6984
6985 .ndo_vlan_rx_add_vid = e1000_vlan_rx_add_vid,
6986 .ndo_vlan_rx_kill_vid = e1000_vlan_rx_kill_vid,
6987#ifdef CONFIG_NET_POLL_CONTROLLER
6988 .ndo_poll_controller = e1000_netpoll,
6989#endif
6990 .ndo_set_features = e1000_set_features,
6991 .ndo_fix_features = e1000_fix_features,
6992 .ndo_features_check = passthru_features_check,
6993};
6994
6995/**
6996 * e1000_probe - Device Initialization Routine
6997 * @pdev: PCI device information struct
6998 * @ent: entry in e1000_pci_tbl
6999 *
7000 * Returns 0 on success, negative on failure
7001 *
7002 * e1000_probe initializes an adapter identified by a pci_dev structure.
7003 * The OS initialization, configuring of the adapter private structure,
7004 * and a hardware reset occur.
7005 **/
7006static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
7007{
7008 struct net_device *netdev;
7009 struct e1000_adapter *adapter;
7010 struct e1000_hw *hw;
7011 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
7012 resource_size_t mmio_start, mmio_len;
7013 resource_size_t flash_start, flash_len;
7014 static int cards_found;
7015 u16 aspm_disable_flag = 0;
7016 int bars, i, err, pci_using_dac;
7017 u16 eeprom_data = 0;
7018 u16 eeprom_apme_mask = E1000_EEPROM_APME;
7019 s32 ret_val = 0;
7020
7021 if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
7022 aspm_disable_flag = PCIE_LINK_STATE_L0S;
7023 if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
7024 aspm_disable_flag |= PCIE_LINK_STATE_L1;
7025 if (aspm_disable_flag)
7026 e1000e_disable_aspm(pdev, aspm_disable_flag);
7027
7028 err = pci_enable_device_mem(pdev);
7029 if (err)
7030 return err;
7031
7032 pci_using_dac = 0;
7033 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
7034 if (!err) {
7035 pci_using_dac = 1;
7036 } else {
7037 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
7038 if (err) {
7039 dev_err(&pdev->dev,
7040 "No usable DMA configuration, aborting\n");
7041 goto err_dma;
7042 }
7043 }
7044
7045 bars = pci_select_bars(pdev, IORESOURCE_MEM);
7046 err = pci_request_selected_regions_exclusive(pdev, bars,
7047 e1000e_driver_name);
7048 if (err)
7049 goto err_pci_reg;
7050
7051 /* AER (Advanced Error Reporting) hooks */
7052 pci_enable_pcie_error_reporting(pdev);
7053
7054 pci_set_master(pdev);
7055 /* PCI config space info */
7056 err = pci_save_state(pdev);
7057 if (err)
7058 goto err_alloc_etherdev;
7059
7060 err = -ENOMEM;
7061 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
7062 if (!netdev)
7063 goto err_alloc_etherdev;
7064
7065 SET_NETDEV_DEV(netdev, &pdev->dev);
7066
7067 netdev->irq = pdev->irq;
7068
7069 pci_set_drvdata(pdev, netdev);
7070 adapter = netdev_priv(netdev);
7071 hw = &adapter->hw;
7072 adapter->netdev = netdev;
7073 adapter->pdev = pdev;
7074 adapter->ei = ei;
7075 adapter->pba = ei->pba;
7076 adapter->flags = ei->flags;
7077 adapter->flags2 = ei->flags2;
7078 adapter->hw.adapter = adapter;
7079 adapter->hw.mac.type = ei->mac;
7080 adapter->max_hw_frame_size = ei->max_hw_frame_size;
7081 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
7082
7083 mmio_start = pci_resource_start(pdev, 0);
7084 mmio_len = pci_resource_len(pdev, 0);
7085
7086 err = -EIO;
7087 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
7088 if (!adapter->hw.hw_addr)
7089 goto err_ioremap;
7090
7091 if ((adapter->flags & FLAG_HAS_FLASH) &&
7092 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM) &&
7093 (hw->mac.type < e1000_pch_spt)) {
7094 flash_start = pci_resource_start(pdev, 1);
7095 flash_len = pci_resource_len(pdev, 1);
7096 adapter->hw.flash_address = ioremap(flash_start, flash_len);
7097 if (!adapter->hw.flash_address)
7098 goto err_flashmap;
7099 }
7100
7101 /* Set default EEE advertisement */
7102 if (adapter->flags2 & FLAG2_HAS_EEE)
7103 adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
7104
7105 /* construct the net_device struct */
7106 netdev->netdev_ops = &e1000e_netdev_ops;
7107 e1000e_set_ethtool_ops(netdev);
7108 netdev->watchdog_timeo = 5 * HZ;
7109 netif_napi_add(netdev, &adapter->napi, e1000e_poll, 64);
7110 strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
7111
7112 netdev->mem_start = mmio_start;
7113 netdev->mem_end = mmio_start + mmio_len;
7114
7115 adapter->bd_number = cards_found++;
7116
7117 e1000e_check_options(adapter);
7118
7119 /* setup adapter struct */
7120 err = e1000_sw_init(adapter);
7121 if (err)
7122 goto err_sw_init;
7123
7124 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
7125 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
7126 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
7127
7128 err = ei->get_variants(adapter);
7129 if (err)
7130 goto err_hw_init;
7131
7132 if ((adapter->flags & FLAG_IS_ICH) &&
7133 (adapter->flags & FLAG_READ_ONLY_NVM) &&
7134 (hw->mac.type < e1000_pch_spt))
7135 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
7136
7137 hw->mac.ops.get_bus_info(&adapter->hw);
7138
7139 adapter->hw.phy.autoneg_wait_to_complete = 0;
7140
7141 /* Copper options */
7142 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
7143 adapter->hw.phy.mdix = AUTO_ALL_MODES;
7144 adapter->hw.phy.disable_polarity_correction = 0;
7145 adapter->hw.phy.ms_type = e1000_ms_hw_default;
7146 }
7147
7148 if (hw->phy.ops.check_reset_block && hw->phy.ops.check_reset_block(hw))
7149 dev_info(&pdev->dev,
7150 "PHY reset is blocked due to SOL/IDER session.\n");
7151
7152 /* Set initial default active device features */
7153 netdev->features = (NETIF_F_SG |
7154 NETIF_F_HW_VLAN_CTAG_RX |
7155 NETIF_F_HW_VLAN_CTAG_TX |
7156 NETIF_F_TSO |
7157 NETIF_F_TSO6 |
7158 NETIF_F_RXHASH |
7159 NETIF_F_RXCSUM |
7160 NETIF_F_HW_CSUM);
7161
7162 /* Set user-changeable features (subset of all device features) */
7163 netdev->hw_features = netdev->features;
7164 netdev->hw_features |= NETIF_F_RXFCS;
7165 netdev->priv_flags |= IFF_SUPP_NOFCS;
7166 netdev->hw_features |= NETIF_F_RXALL;
7167
7168 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
7169 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
7170
7171 netdev->vlan_features |= (NETIF_F_SG |
7172 NETIF_F_TSO |
7173 NETIF_F_TSO6 |
7174 NETIF_F_HW_CSUM);
7175
7176 netdev->priv_flags |= IFF_UNICAST_FLT;
7177
7178 if (pci_using_dac) {
7179 netdev->features |= NETIF_F_HIGHDMA;
7180 netdev->vlan_features |= NETIF_F_HIGHDMA;
7181 }
7182
7183 /* MTU range: 68 - max_hw_frame_size */
7184 netdev->min_mtu = ETH_MIN_MTU;
7185 netdev->max_mtu = adapter->max_hw_frame_size -
7186 (VLAN_ETH_HLEN + ETH_FCS_LEN);
7187
7188 if (e1000e_enable_mng_pass_thru(&adapter->hw))
7189 adapter->flags |= FLAG_MNG_PT_ENABLED;
7190
7191 /* before reading the NVM, reset the controller to
7192 * put the device in a known good starting state
7193 */
7194 adapter->hw.mac.ops.reset_hw(&adapter->hw);
7195
7196 /* systems with ASPM and others may see the checksum fail on the first
7197 * attempt. Let's give it a few tries
7198 */
7199 for (i = 0;; i++) {
7200 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
7201 break;
7202 if (i == 2) {
7203 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
7204 err = -EIO;
7205 goto err_eeprom;
7206 }
7207 }
7208
7209 e1000_eeprom_checks(adapter);
7210
7211 /* copy the MAC address */
7212 if (e1000e_read_mac_addr(&adapter->hw))
7213 dev_err(&pdev->dev,
7214 "NVM Read Error while reading MAC address\n");
7215
7216 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
7217
7218 if (!is_valid_ether_addr(netdev->dev_addr)) {
7219 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
7220 netdev->dev_addr);
7221 err = -EIO;
7222 goto err_eeprom;
7223 }
7224
7225 init_timer(&adapter->watchdog_timer);
7226 adapter->watchdog_timer.function = e1000_watchdog;
7227 adapter->watchdog_timer.data = (unsigned long)adapter;
7228
7229 init_timer(&adapter->phy_info_timer);
7230 adapter->phy_info_timer.function = e1000_update_phy_info;
7231 adapter->phy_info_timer.data = (unsigned long)adapter;
7232
7233 INIT_WORK(&adapter->reset_task, e1000_reset_task);
7234 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
7235 INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
7236 INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
7237 INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
7238
7239 /* Initialize link parameters. User can change them with ethtool */
7240 adapter->hw.mac.autoneg = 1;
7241 adapter->fc_autoneg = true;
7242 adapter->hw.fc.requested_mode = e1000_fc_default;
7243 adapter->hw.fc.current_mode = e1000_fc_default;
7244 adapter->hw.phy.autoneg_advertised = 0x2f;
7245
7246 /* Initial Wake on LAN setting - If APM wake is enabled in
7247 * the EEPROM, enable the ACPI Magic Packet filter
7248 */
7249 if (adapter->flags & FLAG_APME_IN_WUC) {
7250 /* APME bit in EEPROM is mapped to WUC.APME */
7251 eeprom_data = er32(WUC);
7252 eeprom_apme_mask = E1000_WUC_APME;
7253 if ((hw->mac.type > e1000_ich10lan) &&
7254 (eeprom_data & E1000_WUC_PHY_WAKE))
7255 adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
7256 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
7257 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
7258 (adapter->hw.bus.func == 1))
7259 ret_val = e1000_read_nvm(&adapter->hw,
7260 NVM_INIT_CONTROL3_PORT_B,
7261 1, &eeprom_data);
7262 else
7263 ret_val = e1000_read_nvm(&adapter->hw,
7264 NVM_INIT_CONTROL3_PORT_A,
7265 1, &eeprom_data);
7266 }
7267
7268 /* fetch WoL from EEPROM */
7269 if (ret_val)
7270 e_dbg("NVM read error getting WoL initial values: %d\n", ret_val);
7271 else if (eeprom_data & eeprom_apme_mask)
7272 adapter->eeprom_wol |= E1000_WUFC_MAG;
7273
7274 /* now that we have the eeprom settings, apply the special cases
7275 * where the eeprom may be wrong or the board simply won't support
7276 * wake on lan on a particular port
7277 */
7278 if (!(adapter->flags & FLAG_HAS_WOL))
7279 adapter->eeprom_wol = 0;
7280
7281 /* initialize the wol settings based on the eeprom settings */
7282 adapter->wol = adapter->eeprom_wol;
7283
7284 /* make sure adapter isn't asleep if manageability is enabled */
7285 if (adapter->wol || (adapter->flags & FLAG_MNG_PT_ENABLED) ||
7286 (hw->mac.ops.check_mng_mode(hw)))
7287 device_wakeup_enable(&pdev->dev);
7288
7289 /* save off EEPROM version number */
7290 ret_val = e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
7291
7292 if (ret_val) {
7293 e_dbg("NVM read error getting EEPROM version: %d\n", ret_val);
7294 adapter->eeprom_vers = 0;
7295 }
7296
7297 /* init PTP hardware clock */
7298 e1000e_ptp_init(adapter);
7299
7300 /* reset the hardware with the new settings */
7301 e1000e_reset(adapter);
7302
7303 /* If the controller has AMT, do not set DRV_LOAD until the interface
7304 * is up. For all other cases, let the f/w know that the h/w is now
7305 * under the control of the driver.
7306 */
7307 if (!(adapter->flags & FLAG_HAS_AMT))
7308 e1000e_get_hw_control(adapter);
7309
7310 strlcpy(netdev->name, "eth%d", sizeof(netdev->name));
7311 err = register_netdev(netdev);
7312 if (err)
7313 goto err_register;
7314
7315 /* carrier off reporting is important to ethtool even BEFORE open */
7316 netif_carrier_off(netdev);
7317
7318 e1000_print_device_info(adapter);
7319
7320 if (pci_dev_run_wake(pdev))
7321 pm_runtime_put_noidle(&pdev->dev);
7322
7323 return 0;
7324
7325err_register:
7326 if (!(adapter->flags & FLAG_HAS_AMT))
7327 e1000e_release_hw_control(adapter);
7328err_eeprom:
7329 if (hw->phy.ops.check_reset_block && !hw->phy.ops.check_reset_block(hw))
7330 e1000_phy_hw_reset(&adapter->hw);
7331err_hw_init:
7332 kfree(adapter->tx_ring);
7333 kfree(adapter->rx_ring);
7334err_sw_init:
7335 if ((adapter->hw.flash_address) && (hw->mac.type < e1000_pch_spt))
7336 iounmap(adapter->hw.flash_address);
7337 e1000e_reset_interrupt_capability(adapter);
7338err_flashmap:
7339 iounmap(adapter->hw.hw_addr);
7340err_ioremap:
7341 free_netdev(netdev);
7342err_alloc_etherdev:
7343 pci_release_mem_regions(pdev);
7344err_pci_reg:
7345err_dma:
7346 pci_disable_device(pdev);
7347 return err;
7348}
7349
7350/**
7351 * e1000_remove - Device Removal Routine
7352 * @pdev: PCI device information struct
7353 *
7354 * e1000_remove is called by the PCI subsystem to alert the driver
7355 * that it should release a PCI device. The could be caused by a
7356 * Hot-Plug event, or because the driver is going to be removed from
7357 * memory.
7358 **/
7359static void e1000_remove(struct pci_dev *pdev)
7360{
7361 struct net_device *netdev = pci_get_drvdata(pdev);
7362 struct e1000_adapter *adapter = netdev_priv(netdev);
7363 bool down = test_bit(__E1000_DOWN, &adapter->state);
7364
7365 e1000e_ptp_remove(adapter);
7366
7367 /* The timers may be rescheduled, so explicitly disable them
7368 * from being rescheduled.
7369 */
7370 if (!down)
7371 set_bit(__E1000_DOWN, &adapter->state);
7372 del_timer_sync(&adapter->watchdog_timer);
7373 del_timer_sync(&adapter->phy_info_timer);
7374
7375 cancel_work_sync(&adapter->reset_task);
7376 cancel_work_sync(&adapter->watchdog_task);
7377 cancel_work_sync(&adapter->downshift_task);
7378 cancel_work_sync(&adapter->update_phy_task);
7379 cancel_work_sync(&adapter->print_hang_task);
7380
7381 if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
7382 cancel_work_sync(&adapter->tx_hwtstamp_work);
7383 if (adapter->tx_hwtstamp_skb) {
7384 dev_kfree_skb_any(adapter->tx_hwtstamp_skb);
7385 adapter->tx_hwtstamp_skb = NULL;
7386 }
7387 }
7388
7389 /* Don't lie to e1000_close() down the road. */
7390 if (!down)
7391 clear_bit(__E1000_DOWN, &adapter->state);
7392 unregister_netdev(netdev);
7393
7394 if (pci_dev_run_wake(pdev))
7395 pm_runtime_get_noresume(&pdev->dev);
7396
7397 /* Release control of h/w to f/w. If f/w is AMT enabled, this
7398 * would have already happened in close and is redundant.
7399 */
7400 e1000e_release_hw_control(adapter);
7401
7402 e1000e_reset_interrupt_capability(adapter);
7403 kfree(adapter->tx_ring);
7404 kfree(adapter->rx_ring);
7405
7406 iounmap(adapter->hw.hw_addr);
7407 if ((adapter->hw.flash_address) &&
7408 (adapter->hw.mac.type < e1000_pch_spt))
7409 iounmap(adapter->hw.flash_address);
7410 pci_release_mem_regions(pdev);
7411
7412 free_netdev(netdev);
7413
7414 /* AER disable */
7415 pci_disable_pcie_error_reporting(pdev);
7416
7417 pci_disable_device(pdev);
7418}
7419
7420/* PCI Error Recovery (ERS) */
7421static const struct pci_error_handlers e1000_err_handler = {
7422 .error_detected = e1000_io_error_detected,
7423 .slot_reset = e1000_io_slot_reset,
7424 .resume = e1000_io_resume,
7425};
7426
7427static const struct pci_device_id e1000_pci_tbl[] = {
7428 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
7429 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
7430 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
7431 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP),
7432 board_82571 },
7433 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
7434 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
7435 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
7436 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
7437 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
7438
7439 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
7440 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
7441 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
7442 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
7443
7444 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
7445 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
7446 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
7447
7448 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
7449 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
7450 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
7451
7452 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
7453 board_80003es2lan },
7454 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
7455 board_80003es2lan },
7456 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
7457 board_80003es2lan },
7458 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
7459 board_80003es2lan },
7460
7461 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
7462 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
7463 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
7464 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
7465 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
7466 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
7467 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
7468 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
7469
7470 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
7471 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
7472 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
7473 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
7474 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
7475 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
7476 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
7477 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
7478 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
7479
7480 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
7481 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
7482 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
7483
7484 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
7485 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
7486 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
7487
7488 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
7489 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
7490 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
7491 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
7492
7493 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
7494 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
7495
7496 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_LM), board_pch_lpt },
7497 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_V), board_pch_lpt },
7498 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_LM), board_pch_lpt },
7499 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_V), board_pch_lpt },
7500 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM2), board_pch_lpt },
7501 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V2), board_pch_lpt },
7502 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM3), board_pch_lpt },
7503 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V3), board_pch_lpt },
7504 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM), board_pch_spt },
7505 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V), board_pch_spt },
7506 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM2), board_pch_spt },
7507 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V2), board_pch_spt },
7508 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LBG_I219_LM3), board_pch_spt },
7509 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM4), board_pch_spt },
7510 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V4), board_pch_spt },
7511 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM5), board_pch_spt },
7512 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V5), board_pch_spt },
7513
7514 { 0, 0, 0, 0, 0, 0, 0 } /* terminate list */
7515};
7516MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
7517
7518static const struct dev_pm_ops e1000_pm_ops = {
7519#ifdef CONFIG_PM_SLEEP
7520 .suspend = e1000e_pm_suspend,
7521 .resume = e1000e_pm_resume,
7522 .freeze = e1000e_pm_freeze,
7523 .thaw = e1000e_pm_thaw,
7524 .poweroff = e1000e_pm_suspend,
7525 .restore = e1000e_pm_resume,
7526#endif
7527 SET_RUNTIME_PM_OPS(e1000e_pm_runtime_suspend, e1000e_pm_runtime_resume,
7528 e1000e_pm_runtime_idle)
7529};
7530
7531/* PCI Device API Driver */
7532static struct pci_driver e1000_driver = {
7533 .name = e1000e_driver_name,
7534 .id_table = e1000_pci_tbl,
7535 .probe = e1000_probe,
7536 .remove = e1000_remove,
7537 .driver = {
7538 .pm = &e1000_pm_ops,
7539 },
7540 .shutdown = e1000_shutdown,
7541 .err_handler = &e1000_err_handler
7542};
7543
7544/**
7545 * e1000_init_module - Driver Registration Routine
7546 *
7547 * e1000_init_module is the first routine called when the driver is
7548 * loaded. All it does is register with the PCI subsystem.
7549 **/
7550static int __init e1000_init_module(void)
7551{
7552 pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
7553 e1000e_driver_version);
7554 pr_info("Copyright(c) 1999 - 2015 Intel Corporation.\n");
7555
7556 return pci_register_driver(&e1000_driver);
7557}
7558module_init(e1000_init_module);
7559
7560/**
7561 * e1000_exit_module - Driver Exit Cleanup Routine
7562 *
7563 * e1000_exit_module is called just before the driver is removed
7564 * from memory.
7565 **/
7566static void __exit e1000_exit_module(void)
7567{
7568 pci_unregister_driver(&e1000_driver);
7569}
7570module_exit(e1000_exit_module);
7571
7572MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
7573MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
7574MODULE_LICENSE("GPL");
7575MODULE_VERSION(DRV_VERSION);
7576
7577/* netdev.c */