1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Implement the AER root port service driver. The driver registers an IRQ
4 * handler. When a root port triggers an AER interrupt, the IRQ handler
5 * collects root port status and schedules work.
6 *
7 * Copyright (C) 2006 Intel Corp.
8 * Tom Long Nguyen (tom.l.nguyen@intel.com)
9 * Zhang Yanmin (yanmin.zhang@intel.com)
10 *
11 * (C) Copyright 2009 Hewlett-Packard Development Company, L.P.
12 * Andrew Patterson <andrew.patterson@hp.com>
13 */
14
15 #include <linux/cper.h>
16 #include <linux/pci.h>
17 #include <linux/pci-acpi.h>
18 #include <linux/sched.h>
19 #include <linux/kernel.h>
20 #include <linux/errno.h>
21 #include <linux/pm.h>
22 #include <linux/init.h>
23 #include <linux/interrupt.h>
24 #include <linux/delay.h>
25 #include <linux/kfifo.h>
26 #include <linux/slab.h>
27 #include <acpi/apei.h>
28 #include <ras/ras_event.h>
29
30 #include "../pci.h"
31 #include "portdrv.h"
32
33 #define AER_ERROR_SOURCES_MAX 100
34
35 #define AER_MAX_TYPEOF_COR_ERRS 16 /* as per PCI_ERR_COR_STATUS */
36 #define AER_MAX_TYPEOF_UNCOR_ERRS 26 /* as per PCI_ERR_UNCOR_STATUS*/
37
38 struct aer_err_source {
39 unsigned int status;
40 unsigned int id;
41 };
42
43 struct aer_rpc {
44 struct pci_dev *rpd; /* Root Port device */
45 struct work_struct dpc_handler;
46 struct aer_err_source e_sources[AER_ERROR_SOURCES_MAX];
47 struct aer_err_info e_info;
48 unsigned short prod_idx; /* Error Producer Index */
49 unsigned short cons_idx; /* Error Consumer Index */
50 int isr;
51 spinlock_t e_lock; /*
52 * Lock access to Error Status/ID Regs
53 * and error producer/consumer index
54 */
55 struct mutex rpc_mutex; /*
56 * only one thread could do
57 * recovery on the same
58 * root port hierarchy
59 */
60 };
61
62 /* AER stats for the device */
63 struct aer_stats {
64
65 /*
66 * Fields for all AER capable devices. They indicate the errors
67 * "as seen by this device". Note that this may mean that if an
68 * end point is causing problems, the AER counters may increment
69 * at its link partner (e.g. root port) because the errors will be
70 * "seen" by the link partner and not the the problematic end point
71 * itself (which may report all counters as 0 as it never saw any
72 * problems).
73 */
74 /* Counters for different type of correctable errors */
75 u64 dev_cor_errs[AER_MAX_TYPEOF_COR_ERRS];
76 /* Counters for different type of fatal uncorrectable errors */
77 u64 dev_fatal_errs[AER_MAX_TYPEOF_UNCOR_ERRS];
78 /* Counters for different type of nonfatal uncorrectable errors */
79 u64 dev_nonfatal_errs[AER_MAX_TYPEOF_UNCOR_ERRS];
80 /* Total number of ERR_COR sent by this device */
81 u64 dev_total_cor_errs;
82 /* Total number of ERR_FATAL sent by this device */
83 u64 dev_total_fatal_errs;
84 /* Total number of ERR_NONFATAL sent by this device */
85 u64 dev_total_nonfatal_errs;
86
87 /*
88 * Fields for Root ports & root complex event collectors only, these
89 * indicate the total number of ERR_COR, ERR_FATAL, and ERR_NONFATAL
90 * messages received by the root port / event collector, INCLUDING the
91 * ones that are generated internally (by the rootport itself)
92 */
93 u64 rootport_total_cor_errs;
94 u64 rootport_total_fatal_errs;
95 u64 rootport_total_nonfatal_errs;
96 };
97
98 #define AER_LOG_TLP_MASKS (PCI_ERR_UNC_POISON_TLP| \
99 PCI_ERR_UNC_ECRC| \
100 PCI_ERR_UNC_UNSUP| \
101 PCI_ERR_UNC_COMP_ABORT| \
102 PCI_ERR_UNC_UNX_COMP| \
103 PCI_ERR_UNC_MALF_TLP)
104
105 #define SYSTEM_ERROR_INTR_ON_MESG_MASK (PCI_EXP_RTCTL_SECEE| \
106 PCI_EXP_RTCTL_SENFEE| \
107 PCI_EXP_RTCTL_SEFEE)
108 #define ROOT_PORT_INTR_ON_MESG_MASK (PCI_ERR_ROOT_CMD_COR_EN| \
109 PCI_ERR_ROOT_CMD_NONFATAL_EN| \
110 PCI_ERR_ROOT_CMD_FATAL_EN)
111 #define ERR_COR_ID(d) (d & 0xffff)
112 #define ERR_UNCOR_ID(d) (d >> 16)
113
114 static int pcie_aer_disable;
115
pci_no_aer(void)116 void pci_no_aer(void)
117 {
118 pcie_aer_disable = 1;
119 }
120
pci_aer_available(void)121 bool pci_aer_available(void)
122 {
123 return !pcie_aer_disable && pci_msi_enabled();
124 }
125
126 #ifdef CONFIG_PCIE_ECRC
127
128 #define ECRC_POLICY_DEFAULT 0 /* ECRC set by BIOS */
129 #define ECRC_POLICY_OFF 1 /* ECRC off for performance */
130 #define ECRC_POLICY_ON 2 /* ECRC on for data integrity */
131
132 static int ecrc_policy = ECRC_POLICY_DEFAULT;
133
134 static const char *ecrc_policy_str[] = {
135 [ECRC_POLICY_DEFAULT] = "bios",
136 [ECRC_POLICY_OFF] = "off",
137 [ECRC_POLICY_ON] = "on"
138 };
139
140 /**
141 * enable_ercr_checking - enable PCIe ECRC checking for a device
142 * @dev: the PCI device
143 *
144 * Returns 0 on success, or negative on failure.
145 */
enable_ecrc_checking(struct pci_dev * dev)146 static int enable_ecrc_checking(struct pci_dev *dev)
147 {
148 int pos;
149 u32 reg32;
150
151 if (!pci_is_pcie(dev))
152 return -ENODEV;
153
154 pos = dev->aer_cap;
155 if (!pos)
156 return -ENODEV;
157
158 pci_read_config_dword(dev, pos + PCI_ERR_CAP, ®32);
159 if (reg32 & PCI_ERR_CAP_ECRC_GENC)
160 reg32 |= PCI_ERR_CAP_ECRC_GENE;
161 if (reg32 & PCI_ERR_CAP_ECRC_CHKC)
162 reg32 |= PCI_ERR_CAP_ECRC_CHKE;
163 pci_write_config_dword(dev, pos + PCI_ERR_CAP, reg32);
164
165 return 0;
166 }
167
168 /**
169 * disable_ercr_checking - disables PCIe ECRC checking for a device
170 * @dev: the PCI device
171 *
172 * Returns 0 on success, or negative on failure.
173 */
disable_ecrc_checking(struct pci_dev * dev)174 static int disable_ecrc_checking(struct pci_dev *dev)
175 {
176 int pos;
177 u32 reg32;
178
179 if (!pci_is_pcie(dev))
180 return -ENODEV;
181
182 pos = dev->aer_cap;
183 if (!pos)
184 return -ENODEV;
185
186 pci_read_config_dword(dev, pos + PCI_ERR_CAP, ®32);
187 reg32 &= ~(PCI_ERR_CAP_ECRC_GENE | PCI_ERR_CAP_ECRC_CHKE);
188 pci_write_config_dword(dev, pos + PCI_ERR_CAP, reg32);
189
190 return 0;
191 }
192
193 /**
194 * pcie_set_ecrc_checking - set/unset PCIe ECRC checking for a device based on global policy
195 * @dev: the PCI device
196 */
pcie_set_ecrc_checking(struct pci_dev * dev)197 void pcie_set_ecrc_checking(struct pci_dev *dev)
198 {
199 switch (ecrc_policy) {
200 case ECRC_POLICY_DEFAULT:
201 return;
202 case ECRC_POLICY_OFF:
203 disable_ecrc_checking(dev);
204 break;
205 case ECRC_POLICY_ON:
206 enable_ecrc_checking(dev);
207 break;
208 default:
209 return;
210 }
211 }
212
213 /**
214 * pcie_ecrc_get_policy - parse kernel command-line ecrc option
215 */
pcie_ecrc_get_policy(char * str)216 void pcie_ecrc_get_policy(char *str)
217 {
218 int i;
219
220 for (i = 0; i < ARRAY_SIZE(ecrc_policy_str); i++)
221 if (!strncmp(str, ecrc_policy_str[i],
222 strlen(ecrc_policy_str[i])))
223 break;
224 if (i >= ARRAY_SIZE(ecrc_policy_str))
225 return;
226
227 ecrc_policy = i;
228 }
229 #endif /* CONFIG_PCIE_ECRC */
230
231 #ifdef CONFIG_ACPI_APEI
hest_match_pci(struct acpi_hest_aer_common * p,struct pci_dev * pci)232 static inline int hest_match_pci(struct acpi_hest_aer_common *p,
233 struct pci_dev *pci)
234 {
235 return ACPI_HEST_SEGMENT(p->bus) == pci_domain_nr(pci->bus) &&
236 ACPI_HEST_BUS(p->bus) == pci->bus->number &&
237 p->device == PCI_SLOT(pci->devfn) &&
238 p->function == PCI_FUNC(pci->devfn);
239 }
240
hest_match_type(struct acpi_hest_header * hest_hdr,struct pci_dev * dev)241 static inline bool hest_match_type(struct acpi_hest_header *hest_hdr,
242 struct pci_dev *dev)
243 {
244 u16 hest_type = hest_hdr->type;
245 u8 pcie_type = pci_pcie_type(dev);
246
247 if ((hest_type == ACPI_HEST_TYPE_AER_ROOT_PORT &&
248 pcie_type == PCI_EXP_TYPE_ROOT_PORT) ||
249 (hest_type == ACPI_HEST_TYPE_AER_ENDPOINT &&
250 pcie_type == PCI_EXP_TYPE_ENDPOINT) ||
251 (hest_type == ACPI_HEST_TYPE_AER_BRIDGE &&
252 (dev->class >> 16) == PCI_BASE_CLASS_BRIDGE))
253 return true;
254 return false;
255 }
256
257 struct aer_hest_parse_info {
258 struct pci_dev *pci_dev;
259 int firmware_first;
260 };
261
hest_source_is_pcie_aer(struct acpi_hest_header * hest_hdr)262 static int hest_source_is_pcie_aer(struct acpi_hest_header *hest_hdr)
263 {
264 if (hest_hdr->type == ACPI_HEST_TYPE_AER_ROOT_PORT ||
265 hest_hdr->type == ACPI_HEST_TYPE_AER_ENDPOINT ||
266 hest_hdr->type == ACPI_HEST_TYPE_AER_BRIDGE)
267 return 1;
268 return 0;
269 }
270
aer_hest_parse(struct acpi_hest_header * hest_hdr,void * data)271 static int aer_hest_parse(struct acpi_hest_header *hest_hdr, void *data)
272 {
273 struct aer_hest_parse_info *info = data;
274 struct acpi_hest_aer_common *p;
275 int ff;
276
277 if (!hest_source_is_pcie_aer(hest_hdr))
278 return 0;
279
280 p = (struct acpi_hest_aer_common *)(hest_hdr + 1);
281 ff = !!(p->flags & ACPI_HEST_FIRMWARE_FIRST);
282
283 /*
284 * If no specific device is supplied, determine whether
285 * FIRMWARE_FIRST is set for *any* PCIe device.
286 */
287 if (!info->pci_dev) {
288 info->firmware_first |= ff;
289 return 0;
290 }
291
292 /* Otherwise, check the specific device */
293 if (p->flags & ACPI_HEST_GLOBAL) {
294 if (hest_match_type(hest_hdr, info->pci_dev))
295 info->firmware_first = ff;
296 } else
297 if (hest_match_pci(p, info->pci_dev))
298 info->firmware_first = ff;
299
300 return 0;
301 }
302
aer_set_firmware_first(struct pci_dev * pci_dev)303 static void aer_set_firmware_first(struct pci_dev *pci_dev)
304 {
305 int rc;
306 struct aer_hest_parse_info info = {
307 .pci_dev = pci_dev,
308 .firmware_first = 0,
309 };
310
311 rc = apei_hest_parse(aer_hest_parse, &info);
312
313 if (rc)
314 pci_dev->__aer_firmware_first = 0;
315 else
316 pci_dev->__aer_firmware_first = info.firmware_first;
317 pci_dev->__aer_firmware_first_valid = 1;
318 }
319
pcie_aer_get_firmware_first(struct pci_dev * dev)320 int pcie_aer_get_firmware_first(struct pci_dev *dev)
321 {
322 if (!pci_is_pcie(dev))
323 return 0;
324
325 if (pcie_ports_native)
326 return 0;
327
328 if (!dev->__aer_firmware_first_valid)
329 aer_set_firmware_first(dev);
330 return dev->__aer_firmware_first;
331 }
332
333 static bool aer_firmware_first;
334
335 /**
336 * aer_acpi_firmware_first - Check if APEI should control AER.
337 */
aer_acpi_firmware_first(void)338 bool aer_acpi_firmware_first(void)
339 {
340 static bool parsed = false;
341 struct aer_hest_parse_info info = {
342 .pci_dev = NULL, /* Check all PCIe devices */
343 .firmware_first = 0,
344 };
345
346 if (pcie_ports_native)
347 return false;
348
349 if (!parsed) {
350 apei_hest_parse(aer_hest_parse, &info);
351 aer_firmware_first = info.firmware_first;
352 parsed = true;
353 }
354 return aer_firmware_first;
355 }
356 #endif
357
358 #define PCI_EXP_AER_FLAGS (PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE | \
359 PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE)
360
pci_enable_pcie_error_reporting(struct pci_dev * dev)361 int pci_enable_pcie_error_reporting(struct pci_dev *dev)
362 {
363 if (pcie_aer_get_firmware_first(dev))
364 return -EIO;
365
366 if (!dev->aer_cap)
367 return -EIO;
368
369 return pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_AER_FLAGS);
370 }
371 EXPORT_SYMBOL_GPL(pci_enable_pcie_error_reporting);
372
pci_disable_pcie_error_reporting(struct pci_dev * dev)373 int pci_disable_pcie_error_reporting(struct pci_dev *dev)
374 {
375 if (pcie_aer_get_firmware_first(dev))
376 return -EIO;
377
378 return pcie_capability_clear_word(dev, PCI_EXP_DEVCTL,
379 PCI_EXP_AER_FLAGS);
380 }
381 EXPORT_SYMBOL_GPL(pci_disable_pcie_error_reporting);
382
pci_aer_clear_device_status(struct pci_dev * dev)383 void pci_aer_clear_device_status(struct pci_dev *dev)
384 {
385 u16 sta;
386
387 pcie_capability_read_word(dev, PCI_EXP_DEVSTA, &sta);
388 pcie_capability_write_word(dev, PCI_EXP_DEVSTA, sta);
389 }
390
pci_cleanup_aer_uncorrect_error_status(struct pci_dev * dev)391 int pci_cleanup_aer_uncorrect_error_status(struct pci_dev *dev)
392 {
393 int pos;
394 u32 status, sev;
395
396 pos = dev->aer_cap;
397 if (!pos)
398 return -EIO;
399
400 if (pcie_aer_get_firmware_first(dev))
401 return -EIO;
402
403 /* Clear status bits for ERR_NONFATAL errors only */
404 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
405 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, &sev);
406 status &= ~sev;
407 if (status)
408 pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
409
410 return 0;
411 }
412 EXPORT_SYMBOL_GPL(pci_cleanup_aer_uncorrect_error_status);
413
pci_aer_clear_fatal_status(struct pci_dev * dev)414 void pci_aer_clear_fatal_status(struct pci_dev *dev)
415 {
416 int pos;
417 u32 status, sev;
418
419 pos = dev->aer_cap;
420 if (!pos)
421 return;
422
423 if (pcie_aer_get_firmware_first(dev))
424 return;
425
426 /* Clear status bits for ERR_FATAL errors only */
427 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
428 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, &sev);
429 status &= sev;
430 if (status)
431 pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
432 }
433
pci_cleanup_aer_error_status_regs(struct pci_dev * dev)434 int pci_cleanup_aer_error_status_regs(struct pci_dev *dev)
435 {
436 int pos;
437 u32 status;
438 int port_type;
439
440 if (!pci_is_pcie(dev))
441 return -ENODEV;
442
443 pos = dev->aer_cap;
444 if (!pos)
445 return -EIO;
446
447 if (pcie_aer_get_firmware_first(dev))
448 return -EIO;
449
450 port_type = pci_pcie_type(dev);
451 if (port_type == PCI_EXP_TYPE_ROOT_PORT) {
452 pci_read_config_dword(dev, pos + PCI_ERR_ROOT_STATUS, &status);
453 pci_write_config_dword(dev, pos + PCI_ERR_ROOT_STATUS, status);
454 }
455
456 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS, &status);
457 pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS, status);
458
459 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
460 pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
461
462 return 0;
463 }
464
pci_aer_init(struct pci_dev * dev)465 void pci_aer_init(struct pci_dev *dev)
466 {
467 dev->aer_cap = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
468
469 if (dev->aer_cap)
470 dev->aer_stats = kzalloc(sizeof(struct aer_stats), GFP_KERNEL);
471
472 pci_cleanup_aer_error_status_regs(dev);
473 }
474
pci_aer_exit(struct pci_dev * dev)475 void pci_aer_exit(struct pci_dev *dev)
476 {
477 kfree(dev->aer_stats);
478 dev->aer_stats = NULL;
479 }
480
481 #define AER_AGENT_RECEIVER 0
482 #define AER_AGENT_REQUESTER 1
483 #define AER_AGENT_COMPLETER 2
484 #define AER_AGENT_TRANSMITTER 3
485
486 #define AER_AGENT_REQUESTER_MASK(t) ((t == AER_CORRECTABLE) ? \
487 0 : (PCI_ERR_UNC_COMP_TIME|PCI_ERR_UNC_UNSUP))
488 #define AER_AGENT_COMPLETER_MASK(t) ((t == AER_CORRECTABLE) ? \
489 0 : PCI_ERR_UNC_COMP_ABORT)
490 #define AER_AGENT_TRANSMITTER_MASK(t) ((t == AER_CORRECTABLE) ? \
491 (PCI_ERR_COR_REP_ROLL|PCI_ERR_COR_REP_TIMER) : 0)
492
493 #define AER_GET_AGENT(t, e) \
494 ((e & AER_AGENT_COMPLETER_MASK(t)) ? AER_AGENT_COMPLETER : \
495 (e & AER_AGENT_REQUESTER_MASK(t)) ? AER_AGENT_REQUESTER : \
496 (e & AER_AGENT_TRANSMITTER_MASK(t)) ? AER_AGENT_TRANSMITTER : \
497 AER_AGENT_RECEIVER)
498
499 #define AER_PHYSICAL_LAYER_ERROR 0
500 #define AER_DATA_LINK_LAYER_ERROR 1
501 #define AER_TRANSACTION_LAYER_ERROR 2
502
503 #define AER_PHYSICAL_LAYER_ERROR_MASK(t) ((t == AER_CORRECTABLE) ? \
504 PCI_ERR_COR_RCVR : 0)
505 #define AER_DATA_LINK_LAYER_ERROR_MASK(t) ((t == AER_CORRECTABLE) ? \
506 (PCI_ERR_COR_BAD_TLP| \
507 PCI_ERR_COR_BAD_DLLP| \
508 PCI_ERR_COR_REP_ROLL| \
509 PCI_ERR_COR_REP_TIMER) : PCI_ERR_UNC_DLP)
510
511 #define AER_GET_LAYER_ERROR(t, e) \
512 ((e & AER_PHYSICAL_LAYER_ERROR_MASK(t)) ? AER_PHYSICAL_LAYER_ERROR : \
513 (e & AER_DATA_LINK_LAYER_ERROR_MASK(t)) ? AER_DATA_LINK_LAYER_ERROR : \
514 AER_TRANSACTION_LAYER_ERROR)
515
516 /*
517 * AER error strings
518 */
519 static const char *aer_error_severity_string[] = {
520 "Uncorrected (Non-Fatal)",
521 "Uncorrected (Fatal)",
522 "Corrected"
523 };
524
525 static const char *aer_error_layer[] = {
526 "Physical Layer",
527 "Data Link Layer",
528 "Transaction Layer"
529 };
530
531 static const char *aer_correctable_error_string[AER_MAX_TYPEOF_COR_ERRS] = {
532 "RxErr", /* Bit Position 0 */
533 NULL,
534 NULL,
535 NULL,
536 NULL,
537 NULL,
538 "BadTLP", /* Bit Position 6 */
539 "BadDLLP", /* Bit Position 7 */
540 "Rollover", /* Bit Position 8 */
541 NULL,
542 NULL,
543 NULL,
544 "Timeout", /* Bit Position 12 */
545 "NonFatalErr", /* Bit Position 13 */
546 "CorrIntErr", /* Bit Position 14 */
547 "HeaderOF", /* Bit Position 15 */
548 };
549
550 static const char *aer_uncorrectable_error_string[AER_MAX_TYPEOF_UNCOR_ERRS] = {
551 "Undefined", /* Bit Position 0 */
552 NULL,
553 NULL,
554 NULL,
555 "DLP", /* Bit Position 4 */
556 "SDES", /* Bit Position 5 */
557 NULL,
558 NULL,
559 NULL,
560 NULL,
561 NULL,
562 NULL,
563 "TLP", /* Bit Position 12 */
564 "FCP", /* Bit Position 13 */
565 "CmpltTO", /* Bit Position 14 */
566 "CmpltAbrt", /* Bit Position 15 */
567 "UnxCmplt", /* Bit Position 16 */
568 "RxOF", /* Bit Position 17 */
569 "MalfTLP", /* Bit Position 18 */
570 "ECRC", /* Bit Position 19 */
571 "UnsupReq", /* Bit Position 20 */
572 "ACSViol", /* Bit Position 21 */
573 "UncorrIntErr", /* Bit Position 22 */
574 "BlockedTLP", /* Bit Position 23 */
575 "AtomicOpBlocked", /* Bit Position 24 */
576 "TLPBlockedErr", /* Bit Position 25 */
577 };
578
579 static const char *aer_agent_string[] = {
580 "Receiver ID",
581 "Requester ID",
582 "Completer ID",
583 "Transmitter ID"
584 };
585
586 #define aer_stats_dev_attr(name, stats_array, strings_array, \
587 total_string, total_field) \
588 static ssize_t \
589 name##_show(struct device *dev, struct device_attribute *attr, \
590 char *buf) \
591 { \
592 unsigned int i; \
593 char *str = buf; \
594 struct pci_dev *pdev = to_pci_dev(dev); \
595 u64 *stats = pdev->aer_stats->stats_array; \
596 \
597 for (i = 0; i < ARRAY_SIZE(strings_array); i++) { \
598 if (strings_array[i]) \
599 str += sprintf(str, "%s %llu\n", \
600 strings_array[i], stats[i]); \
601 else if (stats[i]) \
602 str += sprintf(str, #stats_array "_bit[%d] %llu\n",\
603 i, stats[i]); \
604 } \
605 str += sprintf(str, "TOTAL_%s %llu\n", total_string, \
606 pdev->aer_stats->total_field); \
607 return str-buf; \
608 } \
609 static DEVICE_ATTR_RO(name)
610
611 aer_stats_dev_attr(aer_dev_correctable, dev_cor_errs,
612 aer_correctable_error_string, "ERR_COR",
613 dev_total_cor_errs);
614 aer_stats_dev_attr(aer_dev_fatal, dev_fatal_errs,
615 aer_uncorrectable_error_string, "ERR_FATAL",
616 dev_total_fatal_errs);
617 aer_stats_dev_attr(aer_dev_nonfatal, dev_nonfatal_errs,
618 aer_uncorrectable_error_string, "ERR_NONFATAL",
619 dev_total_nonfatal_errs);
620
621 #define aer_stats_rootport_attr(name, field) \
622 static ssize_t \
623 name##_show(struct device *dev, struct device_attribute *attr, \
624 char *buf) \
625 { \
626 struct pci_dev *pdev = to_pci_dev(dev); \
627 return sprintf(buf, "%llu\n", pdev->aer_stats->field); \
628 } \
629 static DEVICE_ATTR_RO(name)
630
631 aer_stats_rootport_attr(aer_rootport_total_err_cor,
632 rootport_total_cor_errs);
633 aer_stats_rootport_attr(aer_rootport_total_err_fatal,
634 rootport_total_fatal_errs);
635 aer_stats_rootport_attr(aer_rootport_total_err_nonfatal,
636 rootport_total_nonfatal_errs);
637
638 static struct attribute *aer_stats_attrs[] __ro_after_init = {
639 &dev_attr_aer_dev_correctable.attr,
640 &dev_attr_aer_dev_fatal.attr,
641 &dev_attr_aer_dev_nonfatal.attr,
642 &dev_attr_aer_rootport_total_err_cor.attr,
643 &dev_attr_aer_rootport_total_err_fatal.attr,
644 &dev_attr_aer_rootport_total_err_nonfatal.attr,
645 NULL
646 };
647
aer_stats_attrs_are_visible(struct kobject * kobj,struct attribute * a,int n)648 static umode_t aer_stats_attrs_are_visible(struct kobject *kobj,
649 struct attribute *a, int n)
650 {
651 struct device *dev = kobj_to_dev(kobj);
652 struct pci_dev *pdev = to_pci_dev(dev);
653
654 if (!pdev->aer_stats)
655 return 0;
656
657 if ((a == &dev_attr_aer_rootport_total_err_cor.attr ||
658 a == &dev_attr_aer_rootport_total_err_fatal.attr ||
659 a == &dev_attr_aer_rootport_total_err_nonfatal.attr) &&
660 pci_pcie_type(pdev) != PCI_EXP_TYPE_ROOT_PORT)
661 return 0;
662
663 return a->mode;
664 }
665
666 const struct attribute_group aer_stats_attr_group = {
667 .attrs = aer_stats_attrs,
668 .is_visible = aer_stats_attrs_are_visible,
669 };
670
pci_dev_aer_stats_incr(struct pci_dev * pdev,struct aer_err_info * info)671 static void pci_dev_aer_stats_incr(struct pci_dev *pdev,
672 struct aer_err_info *info)
673 {
674 int status, i, max = -1;
675 u64 *counter = NULL;
676 struct aer_stats *aer_stats = pdev->aer_stats;
677
678 if (!aer_stats)
679 return;
680
681 switch (info->severity) {
682 case AER_CORRECTABLE:
683 aer_stats->dev_total_cor_errs++;
684 counter = &aer_stats->dev_cor_errs[0];
685 max = AER_MAX_TYPEOF_COR_ERRS;
686 break;
687 case AER_NONFATAL:
688 aer_stats->dev_total_nonfatal_errs++;
689 counter = &aer_stats->dev_nonfatal_errs[0];
690 max = AER_MAX_TYPEOF_UNCOR_ERRS;
691 break;
692 case AER_FATAL:
693 aer_stats->dev_total_fatal_errs++;
694 counter = &aer_stats->dev_fatal_errs[0];
695 max = AER_MAX_TYPEOF_UNCOR_ERRS;
696 break;
697 }
698
699 status = (info->status & ~info->mask);
700 for (i = 0; i < max; i++)
701 if (status & (1 << i))
702 counter[i]++;
703 }
704
pci_rootport_aer_stats_incr(struct pci_dev * pdev,struct aer_err_source * e_src)705 static void pci_rootport_aer_stats_incr(struct pci_dev *pdev,
706 struct aer_err_source *e_src)
707 {
708 struct aer_stats *aer_stats = pdev->aer_stats;
709
710 if (!aer_stats)
711 return;
712
713 if (e_src->status & PCI_ERR_ROOT_COR_RCV)
714 aer_stats->rootport_total_cor_errs++;
715
716 if (e_src->status & PCI_ERR_ROOT_UNCOR_RCV) {
717 if (e_src->status & PCI_ERR_ROOT_FATAL_RCV)
718 aer_stats->rootport_total_fatal_errs++;
719 else
720 aer_stats->rootport_total_nonfatal_errs++;
721 }
722 }
723
__print_tlp_header(struct pci_dev * dev,struct aer_header_log_regs * t)724 static void __print_tlp_header(struct pci_dev *dev,
725 struct aer_header_log_regs *t)
726 {
727 pci_err(dev, " TLP Header: %08x %08x %08x %08x\n",
728 t->dw0, t->dw1, t->dw2, t->dw3);
729 }
730
__aer_print_error(struct pci_dev * dev,struct aer_err_info * info)731 static void __aer_print_error(struct pci_dev *dev,
732 struct aer_err_info *info)
733 {
734 int i, status;
735 const char *errmsg = NULL;
736 status = (info->status & ~info->mask);
737
738 for (i = 0; i < 32; i++) {
739 if (!(status & (1 << i)))
740 continue;
741
742 if (info->severity == AER_CORRECTABLE)
743 errmsg = i < ARRAY_SIZE(aer_correctable_error_string) ?
744 aer_correctable_error_string[i] : NULL;
745 else
746 errmsg = i < ARRAY_SIZE(aer_uncorrectable_error_string) ?
747 aer_uncorrectable_error_string[i] : NULL;
748
749 if (errmsg)
750 pci_err(dev, " [%2d] %-22s%s\n", i, errmsg,
751 info->first_error == i ? " (First)" : "");
752 else
753 pci_err(dev, " [%2d] Unknown Error Bit%s\n",
754 i, info->first_error == i ? " (First)" : "");
755 }
756 pci_dev_aer_stats_incr(dev, info);
757 }
758
aer_print_error(struct pci_dev * dev,struct aer_err_info * info)759 void aer_print_error(struct pci_dev *dev, struct aer_err_info *info)
760 {
761 int layer, agent;
762 int id = ((dev->bus->number << 8) | dev->devfn);
763
764 if (!info->status) {
765 pci_err(dev, "PCIe Bus Error: severity=%s, type=Inaccessible, (Unregistered Agent ID)\n",
766 aer_error_severity_string[info->severity]);
767 goto out;
768 }
769
770 layer = AER_GET_LAYER_ERROR(info->severity, info->status);
771 agent = AER_GET_AGENT(info->severity, info->status);
772
773 pci_err(dev, "PCIe Bus Error: severity=%s, type=%s, (%s)\n",
774 aer_error_severity_string[info->severity],
775 aer_error_layer[layer], aer_agent_string[agent]);
776
777 pci_err(dev, " device [%04x:%04x] error status/mask=%08x/%08x\n",
778 dev->vendor, dev->device,
779 info->status, info->mask);
780
781 __aer_print_error(dev, info);
782
783 if (info->tlp_header_valid)
784 __print_tlp_header(dev, &info->tlp);
785
786 out:
787 if (info->id && info->error_dev_num > 1 && info->id == id)
788 pci_err(dev, " Error of this Agent is reported first\n");
789
790 trace_aer_event(dev_name(&dev->dev), (info->status & ~info->mask),
791 info->severity, info->tlp_header_valid, &info->tlp);
792 }
793
aer_print_port_info(struct pci_dev * dev,struct aer_err_info * info)794 static void aer_print_port_info(struct pci_dev *dev, struct aer_err_info *info)
795 {
796 u8 bus = info->id >> 8;
797 u8 devfn = info->id & 0xff;
798
799 pci_info(dev, "AER: %s%s error received: %04x:%02x:%02x.%d\n",
800 info->multi_error_valid ? "Multiple " : "",
801 aer_error_severity_string[info->severity],
802 pci_domain_nr(dev->bus), bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
803 }
804
805 #ifdef CONFIG_ACPI_APEI_PCIEAER
cper_severity_to_aer(int cper_severity)806 int cper_severity_to_aer(int cper_severity)
807 {
808 switch (cper_severity) {
809 case CPER_SEV_RECOVERABLE:
810 return AER_NONFATAL;
811 case CPER_SEV_FATAL:
812 return AER_FATAL;
813 default:
814 return AER_CORRECTABLE;
815 }
816 }
817 EXPORT_SYMBOL_GPL(cper_severity_to_aer);
818
cper_print_aer(struct pci_dev * dev,int aer_severity,struct aer_capability_regs * aer)819 void cper_print_aer(struct pci_dev *dev, int aer_severity,
820 struct aer_capability_regs *aer)
821 {
822 int layer, agent, tlp_header_valid = 0;
823 u32 status, mask;
824 struct aer_err_info info;
825
826 if (aer_severity == AER_CORRECTABLE) {
827 status = aer->cor_status;
828 mask = aer->cor_mask;
829 } else {
830 status = aer->uncor_status;
831 mask = aer->uncor_mask;
832 tlp_header_valid = status & AER_LOG_TLP_MASKS;
833 }
834
835 layer = AER_GET_LAYER_ERROR(aer_severity, status);
836 agent = AER_GET_AGENT(aer_severity, status);
837
838 memset(&info, 0, sizeof(info));
839 info.severity = aer_severity;
840 info.status = status;
841 info.mask = mask;
842 info.first_error = PCI_ERR_CAP_FEP(aer->cap_control);
843
844 pci_err(dev, "aer_status: 0x%08x, aer_mask: 0x%08x\n", status, mask);
845 __aer_print_error(dev, &info);
846 pci_err(dev, "aer_layer=%s, aer_agent=%s\n",
847 aer_error_layer[layer], aer_agent_string[agent]);
848
849 if (aer_severity != AER_CORRECTABLE)
850 pci_err(dev, "aer_uncor_severity: 0x%08x\n",
851 aer->uncor_severity);
852
853 if (tlp_header_valid)
854 __print_tlp_header(dev, &aer->header_log);
855
856 trace_aer_event(dev_name(&dev->dev), (status & ~mask),
857 aer_severity, tlp_header_valid, &aer->header_log);
858 }
859 #endif
860
861 /**
862 * add_error_device - list device to be handled
863 * @e_info: pointer to error info
864 * @dev: pointer to pci_dev to be added
865 */
add_error_device(struct aer_err_info * e_info,struct pci_dev * dev)866 static int add_error_device(struct aer_err_info *e_info, struct pci_dev *dev)
867 {
868 if (e_info->error_dev_num < AER_MAX_MULTI_ERR_DEVICES) {
869 e_info->dev[e_info->error_dev_num] = pci_dev_get(dev);
870 e_info->error_dev_num++;
871 return 0;
872 }
873 return -ENOSPC;
874 }
875
876 /**
877 * is_error_source - check whether the device is source of reported error
878 * @dev: pointer to pci_dev to be checked
879 * @e_info: pointer to reported error info
880 */
is_error_source(struct pci_dev * dev,struct aer_err_info * e_info)881 static bool is_error_source(struct pci_dev *dev, struct aer_err_info *e_info)
882 {
883 int pos;
884 u32 status, mask;
885 u16 reg16;
886
887 /*
888 * When bus id is equal to 0, it might be a bad id
889 * reported by root port.
890 */
891 if ((PCI_BUS_NUM(e_info->id) != 0) &&
892 !(dev->bus->bus_flags & PCI_BUS_FLAGS_NO_AERSID)) {
893 /* Device ID match? */
894 if (e_info->id == ((dev->bus->number << 8) | dev->devfn))
895 return true;
896
897 /* Continue id comparing if there is no multiple error */
898 if (!e_info->multi_error_valid)
899 return false;
900 }
901
902 /*
903 * When either
904 * 1) bus id is equal to 0. Some ports might lose the bus
905 * id of error source id;
906 * 2) bus flag PCI_BUS_FLAGS_NO_AERSID is set
907 * 3) There are multiple errors and prior ID comparing fails;
908 * We check AER status registers to find possible reporter.
909 */
910 if (atomic_read(&dev->enable_cnt) == 0)
911 return false;
912
913 /* Check if AER is enabled */
914 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, ®16);
915 if (!(reg16 & PCI_EXP_AER_FLAGS))
916 return false;
917
918 pos = dev->aer_cap;
919 if (!pos)
920 return false;
921
922 /* Check if error is recorded */
923 if (e_info->severity == AER_CORRECTABLE) {
924 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS, &status);
925 pci_read_config_dword(dev, pos + PCI_ERR_COR_MASK, &mask);
926 } else {
927 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
928 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK, &mask);
929 }
930 if (status & ~mask)
931 return true;
932
933 return false;
934 }
935
find_device_iter(struct pci_dev * dev,void * data)936 static int find_device_iter(struct pci_dev *dev, void *data)
937 {
938 struct aer_err_info *e_info = (struct aer_err_info *)data;
939
940 if (is_error_source(dev, e_info)) {
941 /* List this device */
942 if (add_error_device(e_info, dev)) {
943 /* We cannot handle more... Stop iteration */
944 /* TODO: Should print error message here? */
945 return 1;
946 }
947
948 /* If there is only a single error, stop iteration */
949 if (!e_info->multi_error_valid)
950 return 1;
951 }
952 return 0;
953 }
954
955 /**
956 * find_source_device - search through device hierarchy for source device
957 * @parent: pointer to Root Port pci_dev data structure
958 * @e_info: including detailed error information such like id
959 *
960 * Return true if found.
961 *
962 * Invoked by DPC when error is detected at the Root Port.
963 * Caller of this function must set id, severity, and multi_error_valid of
964 * struct aer_err_info pointed by @e_info properly. This function must fill
965 * e_info->error_dev_num and e_info->dev[], based on the given information.
966 */
find_source_device(struct pci_dev * parent,struct aer_err_info * e_info)967 static bool find_source_device(struct pci_dev *parent,
968 struct aer_err_info *e_info)
969 {
970 struct pci_dev *dev = parent;
971 int result;
972
973 /* Must reset in this function */
974 e_info->error_dev_num = 0;
975
976 /* Is Root Port an agent that sends error message? */
977 result = find_device_iter(dev, e_info);
978 if (result)
979 return true;
980
981 pci_walk_bus(parent->subordinate, find_device_iter, e_info);
982
983 if (!e_info->error_dev_num) {
984 pci_printk(KERN_DEBUG, parent, "can't find device of ID%04x\n",
985 e_info->id);
986 return false;
987 }
988 return true;
989 }
990
991 /**
992 * handle_error_source - handle logging error into an event log
993 * @dev: pointer to pci_dev data structure of error source device
994 * @info: comprehensive error information
995 *
996 * Invoked when an error being detected by Root Port.
997 */
handle_error_source(struct pci_dev * dev,struct aer_err_info * info)998 static void handle_error_source(struct pci_dev *dev, struct aer_err_info *info)
999 {
1000 int pos;
1001
1002 if (info->severity == AER_CORRECTABLE) {
1003 /*
1004 * Correctable error does not need software intervention.
1005 * No need to go through error recovery process.
1006 */
1007 pos = dev->aer_cap;
1008 if (pos)
1009 pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS,
1010 info->status);
1011 pci_aer_clear_device_status(dev);
1012 } else if (info->severity == AER_NONFATAL)
1013 pcie_do_nonfatal_recovery(dev);
1014 else if (info->severity == AER_FATAL)
1015 pcie_do_fatal_recovery(dev, PCIE_PORT_SERVICE_AER);
1016 pci_dev_put(dev);
1017 }
1018
1019 #ifdef CONFIG_ACPI_APEI_PCIEAER
1020
1021 #define AER_RECOVER_RING_ORDER 4
1022 #define AER_RECOVER_RING_SIZE (1 << AER_RECOVER_RING_ORDER)
1023
1024 struct aer_recover_entry {
1025 u8 bus;
1026 u8 devfn;
1027 u16 domain;
1028 int severity;
1029 struct aer_capability_regs *regs;
1030 };
1031
1032 static DEFINE_KFIFO(aer_recover_ring, struct aer_recover_entry,
1033 AER_RECOVER_RING_SIZE);
1034
aer_recover_work_func(struct work_struct * work)1035 static void aer_recover_work_func(struct work_struct *work)
1036 {
1037 struct aer_recover_entry entry;
1038 struct pci_dev *pdev;
1039
1040 while (kfifo_get(&aer_recover_ring, &entry)) {
1041 pdev = pci_get_domain_bus_and_slot(entry.domain, entry.bus,
1042 entry.devfn);
1043 if (!pdev) {
1044 pr_err("AER recover: Can not find pci_dev for %04x:%02x:%02x:%x\n",
1045 entry.domain, entry.bus,
1046 PCI_SLOT(entry.devfn), PCI_FUNC(entry.devfn));
1047 continue;
1048 }
1049 cper_print_aer(pdev, entry.severity, entry.regs);
1050 if (entry.severity == AER_NONFATAL)
1051 pcie_do_nonfatal_recovery(pdev);
1052 else if (entry.severity == AER_FATAL)
1053 pcie_do_fatal_recovery(pdev, PCIE_PORT_SERVICE_AER);
1054 pci_dev_put(pdev);
1055 }
1056 }
1057
1058 /*
1059 * Mutual exclusion for writers of aer_recover_ring, reader side don't
1060 * need lock, because there is only one reader and lock is not needed
1061 * between reader and writer.
1062 */
1063 static DEFINE_SPINLOCK(aer_recover_ring_lock);
1064 static DECLARE_WORK(aer_recover_work, aer_recover_work_func);
1065
aer_recover_queue(int domain,unsigned int bus,unsigned int devfn,int severity,struct aer_capability_regs * aer_regs)1066 void aer_recover_queue(int domain, unsigned int bus, unsigned int devfn,
1067 int severity, struct aer_capability_regs *aer_regs)
1068 {
1069 unsigned long flags;
1070 struct aer_recover_entry entry = {
1071 .bus = bus,
1072 .devfn = devfn,
1073 .domain = domain,
1074 .severity = severity,
1075 .regs = aer_regs,
1076 };
1077
1078 spin_lock_irqsave(&aer_recover_ring_lock, flags);
1079 if (kfifo_put(&aer_recover_ring, entry))
1080 schedule_work(&aer_recover_work);
1081 else
1082 pr_err("AER recover: Buffer overflow when recovering AER for %04x:%02x:%02x:%x\n",
1083 domain, bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
1084 spin_unlock_irqrestore(&aer_recover_ring_lock, flags);
1085 }
1086 EXPORT_SYMBOL_GPL(aer_recover_queue);
1087 #endif
1088
1089 /**
1090 * aer_get_device_error_info - read error status from dev and store it to info
1091 * @dev: pointer to the device expected to have a error record
1092 * @info: pointer to structure to store the error record
1093 *
1094 * Return 1 on success, 0 on error.
1095 *
1096 * Note that @info is reused among all error devices. Clear fields properly.
1097 */
aer_get_device_error_info(struct pci_dev * dev,struct aer_err_info * info)1098 int aer_get_device_error_info(struct pci_dev *dev, struct aer_err_info *info)
1099 {
1100 int pos, temp;
1101
1102 /* Must reset in this function */
1103 info->status = 0;
1104 info->tlp_header_valid = 0;
1105
1106 pos = dev->aer_cap;
1107
1108 /* The device might not support AER */
1109 if (!pos)
1110 return 0;
1111
1112 if (info->severity == AER_CORRECTABLE) {
1113 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS,
1114 &info->status);
1115 pci_read_config_dword(dev, pos + PCI_ERR_COR_MASK,
1116 &info->mask);
1117 if (!(info->status & ~info->mask))
1118 return 0;
1119 } else if (pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT ||
1120 pci_pcie_type(dev) == PCI_EXP_TYPE_DOWNSTREAM ||
1121 info->severity == AER_NONFATAL) {
1122
1123 /* Link is still healthy for IO reads */
1124 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS,
1125 &info->status);
1126 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK,
1127 &info->mask);
1128 if (!(info->status & ~info->mask))
1129 return 0;
1130
1131 /* Get First Error Pointer */
1132 pci_read_config_dword(dev, pos + PCI_ERR_CAP, &temp);
1133 info->first_error = PCI_ERR_CAP_FEP(temp);
1134
1135 if (info->status & AER_LOG_TLP_MASKS) {
1136 info->tlp_header_valid = 1;
1137 pci_read_config_dword(dev,
1138 pos + PCI_ERR_HEADER_LOG, &info->tlp.dw0);
1139 pci_read_config_dword(dev,
1140 pos + PCI_ERR_HEADER_LOG + 4, &info->tlp.dw1);
1141 pci_read_config_dword(dev,
1142 pos + PCI_ERR_HEADER_LOG + 8, &info->tlp.dw2);
1143 pci_read_config_dword(dev,
1144 pos + PCI_ERR_HEADER_LOG + 12, &info->tlp.dw3);
1145 }
1146 }
1147
1148 return 1;
1149 }
1150
aer_process_err_devices(struct aer_err_info * e_info)1151 static inline void aer_process_err_devices(struct aer_err_info *e_info)
1152 {
1153 int i;
1154
1155 /* Report all before handle them, not to lost records by reset etc. */
1156 for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
1157 if (aer_get_device_error_info(e_info->dev[i], e_info))
1158 aer_print_error(e_info->dev[i], e_info);
1159 }
1160 for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
1161 if (aer_get_device_error_info(e_info->dev[i], e_info))
1162 handle_error_source(e_info->dev[i], e_info);
1163 }
1164 }
1165
1166 /**
1167 * aer_isr_one_error - consume an error detected by root port
1168 * @rpc: pointer to the root port which holds an error
1169 * @e_src: pointer to an error source
1170 */
aer_isr_one_error(struct aer_rpc * rpc,struct aer_err_source * e_src)1171 static void aer_isr_one_error(struct aer_rpc *rpc,
1172 struct aer_err_source *e_src)
1173 {
1174 struct pci_dev *pdev = rpc->rpd;
1175 struct aer_err_info *e_info = &rpc->e_info;
1176
1177 pci_rootport_aer_stats_incr(pdev, e_src);
1178
1179 /*
1180 * There is a possibility that both correctable error and
1181 * uncorrectable error being logged. Report correctable error first.
1182 */
1183 if (e_src->status & PCI_ERR_ROOT_COR_RCV) {
1184 e_info->id = ERR_COR_ID(e_src->id);
1185 e_info->severity = AER_CORRECTABLE;
1186
1187 if (e_src->status & PCI_ERR_ROOT_MULTI_COR_RCV)
1188 e_info->multi_error_valid = 1;
1189 else
1190 e_info->multi_error_valid = 0;
1191 aer_print_port_info(pdev, e_info);
1192
1193 if (find_source_device(pdev, e_info))
1194 aer_process_err_devices(e_info);
1195 }
1196
1197 if (e_src->status & PCI_ERR_ROOT_UNCOR_RCV) {
1198 e_info->id = ERR_UNCOR_ID(e_src->id);
1199
1200 if (e_src->status & PCI_ERR_ROOT_FATAL_RCV)
1201 e_info->severity = AER_FATAL;
1202 else
1203 e_info->severity = AER_NONFATAL;
1204
1205 if (e_src->status & PCI_ERR_ROOT_MULTI_UNCOR_RCV)
1206 e_info->multi_error_valid = 1;
1207 else
1208 e_info->multi_error_valid = 0;
1209
1210 aer_print_port_info(pdev, e_info);
1211
1212 if (find_source_device(pdev, e_info))
1213 aer_process_err_devices(e_info);
1214 }
1215 }
1216
1217 /**
1218 * get_e_source - retrieve an error source
1219 * @rpc: pointer to the root port which holds an error
1220 * @e_src: pointer to store retrieved error source
1221 *
1222 * Return 1 if an error source is retrieved, otherwise 0.
1223 *
1224 * Invoked by DPC handler to consume an error.
1225 */
get_e_source(struct aer_rpc * rpc,struct aer_err_source * e_src)1226 static int get_e_source(struct aer_rpc *rpc, struct aer_err_source *e_src)
1227 {
1228 unsigned long flags;
1229
1230 /* Lock access to Root error producer/consumer index */
1231 spin_lock_irqsave(&rpc->e_lock, flags);
1232 if (rpc->prod_idx == rpc->cons_idx) {
1233 spin_unlock_irqrestore(&rpc->e_lock, flags);
1234 return 0;
1235 }
1236
1237 *e_src = rpc->e_sources[rpc->cons_idx];
1238 rpc->cons_idx++;
1239 if (rpc->cons_idx == AER_ERROR_SOURCES_MAX)
1240 rpc->cons_idx = 0;
1241 spin_unlock_irqrestore(&rpc->e_lock, flags);
1242
1243 return 1;
1244 }
1245
1246 /**
1247 * aer_isr - consume errors detected by root port
1248 * @work: definition of this work item
1249 *
1250 * Invoked, as DPC, when root port records new detected error
1251 */
aer_isr(struct work_struct * work)1252 static void aer_isr(struct work_struct *work)
1253 {
1254 struct aer_rpc *rpc = container_of(work, struct aer_rpc, dpc_handler);
1255 struct aer_err_source uninitialized_var(e_src);
1256
1257 mutex_lock(&rpc->rpc_mutex);
1258 while (get_e_source(rpc, &e_src))
1259 aer_isr_one_error(rpc, &e_src);
1260 mutex_unlock(&rpc->rpc_mutex);
1261 }
1262
1263 /**
1264 * aer_irq - Root Port's ISR
1265 * @irq: IRQ assigned to Root Port
1266 * @context: pointer to Root Port data structure
1267 *
1268 * Invoked when Root Port detects AER messages.
1269 */
aer_irq(int irq,void * context)1270 irqreturn_t aer_irq(int irq, void *context)
1271 {
1272 unsigned int status, id;
1273 struct pcie_device *pdev = (struct pcie_device *)context;
1274 struct aer_rpc *rpc = get_service_data(pdev);
1275 int next_prod_idx;
1276 unsigned long flags;
1277 int pos;
1278
1279 pos = pdev->port->aer_cap;
1280 /*
1281 * Must lock access to Root Error Status Reg, Root Error ID Reg,
1282 * and Root error producer/consumer index
1283 */
1284 spin_lock_irqsave(&rpc->e_lock, flags);
1285
1286 /* Read error status */
1287 pci_read_config_dword(pdev->port, pos + PCI_ERR_ROOT_STATUS, &status);
1288 if (!(status & (PCI_ERR_ROOT_UNCOR_RCV|PCI_ERR_ROOT_COR_RCV))) {
1289 spin_unlock_irqrestore(&rpc->e_lock, flags);
1290 return IRQ_NONE;
1291 }
1292
1293 /* Read error source and clear error status */
1294 pci_read_config_dword(pdev->port, pos + PCI_ERR_ROOT_ERR_SRC, &id);
1295 pci_write_config_dword(pdev->port, pos + PCI_ERR_ROOT_STATUS, status);
1296
1297 /* Store error source for later DPC handler */
1298 next_prod_idx = rpc->prod_idx + 1;
1299 if (next_prod_idx == AER_ERROR_SOURCES_MAX)
1300 next_prod_idx = 0;
1301 if (next_prod_idx == rpc->cons_idx) {
1302 /*
1303 * Error Storm Condition - possibly the same error occurred.
1304 * Drop the error.
1305 */
1306 spin_unlock_irqrestore(&rpc->e_lock, flags);
1307 return IRQ_HANDLED;
1308 }
1309 rpc->e_sources[rpc->prod_idx].status = status;
1310 rpc->e_sources[rpc->prod_idx].id = id;
1311 rpc->prod_idx = next_prod_idx;
1312 spin_unlock_irqrestore(&rpc->e_lock, flags);
1313
1314 /* Invoke DPC handler */
1315 schedule_work(&rpc->dpc_handler);
1316
1317 return IRQ_HANDLED;
1318 }
1319 EXPORT_SYMBOL_GPL(aer_irq);
1320
set_device_error_reporting(struct pci_dev * dev,void * data)1321 static int set_device_error_reporting(struct pci_dev *dev, void *data)
1322 {
1323 bool enable = *((bool *)data);
1324 int type = pci_pcie_type(dev);
1325
1326 if ((type == PCI_EXP_TYPE_ROOT_PORT) ||
1327 (type == PCI_EXP_TYPE_UPSTREAM) ||
1328 (type == PCI_EXP_TYPE_DOWNSTREAM)) {
1329 if (enable)
1330 pci_enable_pcie_error_reporting(dev);
1331 else
1332 pci_disable_pcie_error_reporting(dev);
1333 }
1334
1335 if (enable)
1336 pcie_set_ecrc_checking(dev);
1337
1338 return 0;
1339 }
1340
1341 /**
1342 * set_downstream_devices_error_reporting - enable/disable the error reporting bits on the root port and its downstream ports.
1343 * @dev: pointer to root port's pci_dev data structure
1344 * @enable: true = enable error reporting, false = disable error reporting.
1345 */
set_downstream_devices_error_reporting(struct pci_dev * dev,bool enable)1346 static void set_downstream_devices_error_reporting(struct pci_dev *dev,
1347 bool enable)
1348 {
1349 set_device_error_reporting(dev, &enable);
1350
1351 if (!dev->subordinate)
1352 return;
1353 pci_walk_bus(dev->subordinate, set_device_error_reporting, &enable);
1354 }
1355
1356 /**
1357 * aer_enable_rootport - enable Root Port's interrupts when receiving messages
1358 * @rpc: pointer to a Root Port data structure
1359 *
1360 * Invoked when PCIe bus loads AER service driver.
1361 */
aer_enable_rootport(struct aer_rpc * rpc)1362 static void aer_enable_rootport(struct aer_rpc *rpc)
1363 {
1364 struct pci_dev *pdev = rpc->rpd;
1365 int aer_pos;
1366 u16 reg16;
1367 u32 reg32;
1368
1369 /* Clear PCIe Capability's Device Status */
1370 pcie_capability_read_word(pdev, PCI_EXP_DEVSTA, ®16);
1371 pcie_capability_write_word(pdev, PCI_EXP_DEVSTA, reg16);
1372
1373 /* Disable system error generation in response to error messages */
1374 pcie_capability_clear_word(pdev, PCI_EXP_RTCTL,
1375 SYSTEM_ERROR_INTR_ON_MESG_MASK);
1376
1377 aer_pos = pdev->aer_cap;
1378 /* Clear error status */
1379 pci_read_config_dword(pdev, aer_pos + PCI_ERR_ROOT_STATUS, ®32);
1380 pci_write_config_dword(pdev, aer_pos + PCI_ERR_ROOT_STATUS, reg32);
1381 pci_read_config_dword(pdev, aer_pos + PCI_ERR_COR_STATUS, ®32);
1382 pci_write_config_dword(pdev, aer_pos + PCI_ERR_COR_STATUS, reg32);
1383 pci_read_config_dword(pdev, aer_pos + PCI_ERR_UNCOR_STATUS, ®32);
1384 pci_write_config_dword(pdev, aer_pos + PCI_ERR_UNCOR_STATUS, reg32);
1385
1386 /*
1387 * Enable error reporting for the root port device and downstream port
1388 * devices.
1389 */
1390 set_downstream_devices_error_reporting(pdev, true);
1391
1392 /* Enable Root Port's interrupt in response to error messages */
1393 pci_read_config_dword(pdev, aer_pos + PCI_ERR_ROOT_COMMAND, ®32);
1394 reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
1395 pci_write_config_dword(pdev, aer_pos + PCI_ERR_ROOT_COMMAND, reg32);
1396 }
1397
1398 /**
1399 * aer_disable_rootport - disable Root Port's interrupts when receiving messages
1400 * @rpc: pointer to a Root Port data structure
1401 *
1402 * Invoked when PCIe bus unloads AER service driver.
1403 */
aer_disable_rootport(struct aer_rpc * rpc)1404 static void aer_disable_rootport(struct aer_rpc *rpc)
1405 {
1406 struct pci_dev *pdev = rpc->rpd;
1407 u32 reg32;
1408 int pos;
1409
1410 /*
1411 * Disable error reporting for the root port device and downstream port
1412 * devices.
1413 */
1414 set_downstream_devices_error_reporting(pdev, false);
1415
1416 pos = pdev->aer_cap;
1417 /* Disable Root's interrupt in response to error messages */
1418 pci_read_config_dword(pdev, pos + PCI_ERR_ROOT_COMMAND, ®32);
1419 reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
1420 pci_write_config_dword(pdev, pos + PCI_ERR_ROOT_COMMAND, reg32);
1421
1422 /* Clear Root's error status reg */
1423 pci_read_config_dword(pdev, pos + PCI_ERR_ROOT_STATUS, ®32);
1424 pci_write_config_dword(pdev, pos + PCI_ERR_ROOT_STATUS, reg32);
1425 }
1426
1427 /**
1428 * aer_alloc_rpc - allocate Root Port data structure
1429 * @dev: pointer to the pcie_dev data structure
1430 *
1431 * Invoked when Root Port's AER service is loaded.
1432 */
aer_alloc_rpc(struct pcie_device * dev)1433 static struct aer_rpc *aer_alloc_rpc(struct pcie_device *dev)
1434 {
1435 struct aer_rpc *rpc;
1436
1437 rpc = kzalloc(sizeof(struct aer_rpc), GFP_KERNEL);
1438 if (!rpc)
1439 return NULL;
1440
1441 /* Initialize Root lock access, e_lock, to Root Error Status Reg */
1442 spin_lock_init(&rpc->e_lock);
1443
1444 rpc->rpd = dev->port;
1445 INIT_WORK(&rpc->dpc_handler, aer_isr);
1446 mutex_init(&rpc->rpc_mutex);
1447
1448 /* Use PCIe bus function to store rpc into PCIe device */
1449 set_service_data(dev, rpc);
1450
1451 return rpc;
1452 }
1453
1454 /**
1455 * aer_remove - clean up resources
1456 * @dev: pointer to the pcie_dev data structure
1457 *
1458 * Invoked when PCI Express bus unloads or AER probe fails.
1459 */
aer_remove(struct pcie_device * dev)1460 static void aer_remove(struct pcie_device *dev)
1461 {
1462 struct aer_rpc *rpc = get_service_data(dev);
1463
1464 if (rpc) {
1465 /* If register interrupt service, it must be free. */
1466 if (rpc->isr)
1467 free_irq(dev->irq, dev);
1468
1469 flush_work(&rpc->dpc_handler);
1470 aer_disable_rootport(rpc);
1471 kfree(rpc);
1472 set_service_data(dev, NULL);
1473 }
1474 }
1475
1476 /**
1477 * aer_probe - initialize resources
1478 * @dev: pointer to the pcie_dev data structure
1479 *
1480 * Invoked when PCI Express bus loads AER service driver.
1481 */
aer_probe(struct pcie_device * dev)1482 static int aer_probe(struct pcie_device *dev)
1483 {
1484 int status;
1485 struct aer_rpc *rpc;
1486 struct device *device = &dev->port->dev;
1487
1488 /* Alloc rpc data structure */
1489 rpc = aer_alloc_rpc(dev);
1490 if (!rpc) {
1491 dev_printk(KERN_DEBUG, device, "alloc AER rpc failed\n");
1492 aer_remove(dev);
1493 return -ENOMEM;
1494 }
1495
1496 /* Request IRQ ISR */
1497 status = request_irq(dev->irq, aer_irq, IRQF_SHARED, "aerdrv", dev);
1498 if (status) {
1499 dev_printk(KERN_DEBUG, device, "request AER IRQ %d failed\n",
1500 dev->irq);
1501 aer_remove(dev);
1502 return status;
1503 }
1504
1505 rpc->isr = 1;
1506
1507 aer_enable_rootport(rpc);
1508 dev_info(device, "AER enabled with IRQ %d\n", dev->irq);
1509 return 0;
1510 }
1511
1512 /**
1513 * aer_root_reset - reset link on Root Port
1514 * @dev: pointer to Root Port's pci_dev data structure
1515 *
1516 * Invoked by Port Bus driver when performing link reset at Root Port.
1517 */
aer_root_reset(struct pci_dev * dev)1518 static pci_ers_result_t aer_root_reset(struct pci_dev *dev)
1519 {
1520 u32 reg32;
1521 int pos;
1522 int rc;
1523
1524 pos = dev->aer_cap;
1525
1526 /* Disable Root's interrupt in response to error messages */
1527 pci_read_config_dword(dev, pos + PCI_ERR_ROOT_COMMAND, ®32);
1528 reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
1529 pci_write_config_dword(dev, pos + PCI_ERR_ROOT_COMMAND, reg32);
1530
1531 rc = pci_bus_error_reset(dev);
1532 pci_printk(KERN_DEBUG, dev, "Root Port link has been reset\n");
1533
1534 /* Clear Root Error Status */
1535 pci_read_config_dword(dev, pos + PCI_ERR_ROOT_STATUS, ®32);
1536 pci_write_config_dword(dev, pos + PCI_ERR_ROOT_STATUS, reg32);
1537
1538 /* Enable Root Port's interrupt in response to error messages */
1539 pci_read_config_dword(dev, pos + PCI_ERR_ROOT_COMMAND, ®32);
1540 reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
1541 pci_write_config_dword(dev, pos + PCI_ERR_ROOT_COMMAND, reg32);
1542
1543 return rc ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
1544 }
1545
1546 /**
1547 * aer_error_resume - clean up corresponding error status bits
1548 * @dev: pointer to Root Port's pci_dev data structure
1549 *
1550 * Invoked by Port Bus driver during nonfatal recovery.
1551 */
aer_error_resume(struct pci_dev * dev)1552 static void aer_error_resume(struct pci_dev *dev)
1553 {
1554 pci_aer_clear_device_status(dev);
1555 pci_cleanup_aer_uncorrect_error_status(dev);
1556 }
1557
1558 static struct pcie_port_service_driver aerdriver = {
1559 .name = "aer",
1560 .port_type = PCI_EXP_TYPE_ROOT_PORT,
1561 .service = PCIE_PORT_SERVICE_AER,
1562
1563 .probe = aer_probe,
1564 .remove = aer_remove,
1565 .error_resume = aer_error_resume,
1566 .reset_link = aer_root_reset,
1567 };
1568
1569 /**
1570 * aer_service_init - register AER root service driver
1571 *
1572 * Invoked when AER root service driver is loaded.
1573 */
pcie_aer_init(void)1574 int __init pcie_aer_init(void)
1575 {
1576 if (!pci_aer_available() || aer_acpi_firmware_first())
1577 return -ENXIO;
1578 return pcie_port_service_register(&aerdriver);
1579 }
1580