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 #define pr_fmt(fmt) "AER: " fmt
16 #define dev_fmt pr_fmt
17
18 #include <linux/bitops.h>
19 #include <linux/cper.h>
20 #include <linux/pci.h>
21 #include <linux/pci-acpi.h>
22 #include <linux/sched.h>
23 #include <linux/kernel.h>
24 #include <linux/errno.h>
25 #include <linux/pm.h>
26 #include <linux/init.h>
27 #include <linux/interrupt.h>
28 #include <linux/delay.h>
29 #include <linux/kfifo.h>
30 #include <linux/slab.h>
31 #include <acpi/apei.h>
32 #include <ras/ras_event.h>
33
34 #include "../pci.h"
35 #include "portdrv.h"
36
37 #define AER_ERROR_SOURCES_MAX 128
38
39 #define AER_MAX_TYPEOF_COR_ERRS 16 /* as per PCI_ERR_COR_STATUS */
40 #define AER_MAX_TYPEOF_UNCOR_ERRS 27 /* as per PCI_ERR_UNCOR_STATUS*/
41
42 struct aer_err_source {
43 unsigned int status;
44 unsigned int id;
45 };
46
47 struct aer_rpc {
48 struct pci_dev *rpd; /* Root Port device */
49 DECLARE_KFIFO(aer_fifo, struct aer_err_source, AER_ERROR_SOURCES_MAX);
50 };
51
52 /* AER stats for the device */
53 struct aer_stats {
54
55 /*
56 * Fields for all AER capable devices. They indicate the errors
57 * "as seen by this device". Note that this may mean that if an
58 * end point is causing problems, the AER counters may increment
59 * at its link partner (e.g. root port) because the errors will be
60 * "seen" by the link partner and not the problematic end point
61 * itself (which may report all counters as 0 as it never saw any
62 * problems).
63 */
64 /* Counters for different type of correctable errors */
65 u64 dev_cor_errs[AER_MAX_TYPEOF_COR_ERRS];
66 /* Counters for different type of fatal uncorrectable errors */
67 u64 dev_fatal_errs[AER_MAX_TYPEOF_UNCOR_ERRS];
68 /* Counters for different type of nonfatal uncorrectable errors */
69 u64 dev_nonfatal_errs[AER_MAX_TYPEOF_UNCOR_ERRS];
70 /* Total number of ERR_COR sent by this device */
71 u64 dev_total_cor_errs;
72 /* Total number of ERR_FATAL sent by this device */
73 u64 dev_total_fatal_errs;
74 /* Total number of ERR_NONFATAL sent by this device */
75 u64 dev_total_nonfatal_errs;
76
77 /*
78 * Fields for Root ports & root complex event collectors only, these
79 * indicate the total number of ERR_COR, ERR_FATAL, and ERR_NONFATAL
80 * messages received by the root port / event collector, INCLUDING the
81 * ones that are generated internally (by the rootport itself)
82 */
83 u64 rootport_total_cor_errs;
84 u64 rootport_total_fatal_errs;
85 u64 rootport_total_nonfatal_errs;
86 };
87
88 #define AER_LOG_TLP_MASKS (PCI_ERR_UNC_POISON_TLP| \
89 PCI_ERR_UNC_ECRC| \
90 PCI_ERR_UNC_UNSUP| \
91 PCI_ERR_UNC_COMP_ABORT| \
92 PCI_ERR_UNC_UNX_COMP| \
93 PCI_ERR_UNC_MALF_TLP)
94
95 #define SYSTEM_ERROR_INTR_ON_MESG_MASK (PCI_EXP_RTCTL_SECEE| \
96 PCI_EXP_RTCTL_SENFEE| \
97 PCI_EXP_RTCTL_SEFEE)
98 #define ROOT_PORT_INTR_ON_MESG_MASK (PCI_ERR_ROOT_CMD_COR_EN| \
99 PCI_ERR_ROOT_CMD_NONFATAL_EN| \
100 PCI_ERR_ROOT_CMD_FATAL_EN)
101 #define ERR_COR_ID(d) (d & 0xffff)
102 #define ERR_UNCOR_ID(d) (d >> 16)
103
104 static int pcie_aer_disable;
105 static pci_ers_result_t aer_root_reset(struct pci_dev *dev);
106
pci_no_aer(void)107 void pci_no_aer(void)
108 {
109 pcie_aer_disable = 1;
110 }
111
pci_aer_available(void)112 bool pci_aer_available(void)
113 {
114 return !pcie_aer_disable && pci_msi_enabled();
115 }
116
117 #ifdef CONFIG_PCIE_ECRC
118
119 #define ECRC_POLICY_DEFAULT 0 /* ECRC set by BIOS */
120 #define ECRC_POLICY_OFF 1 /* ECRC off for performance */
121 #define ECRC_POLICY_ON 2 /* ECRC on for data integrity */
122
123 static int ecrc_policy = ECRC_POLICY_DEFAULT;
124
125 static const char * const ecrc_policy_str[] = {
126 [ECRC_POLICY_DEFAULT] = "bios",
127 [ECRC_POLICY_OFF] = "off",
128 [ECRC_POLICY_ON] = "on"
129 };
130
131 /**
132 * enable_ecrc_checking - enable PCIe ECRC checking for a device
133 * @dev: the PCI device
134 *
135 * Returns 0 on success, or negative on failure.
136 */
enable_ecrc_checking(struct pci_dev * dev)137 static int enable_ecrc_checking(struct pci_dev *dev)
138 {
139 int aer = dev->aer_cap;
140 u32 reg32;
141
142 if (!aer)
143 return -ENODEV;
144
145 pci_read_config_dword(dev, aer + PCI_ERR_CAP, ®32);
146 if (reg32 & PCI_ERR_CAP_ECRC_GENC)
147 reg32 |= PCI_ERR_CAP_ECRC_GENE;
148 if (reg32 & PCI_ERR_CAP_ECRC_CHKC)
149 reg32 |= PCI_ERR_CAP_ECRC_CHKE;
150 pci_write_config_dword(dev, aer + PCI_ERR_CAP, reg32);
151
152 return 0;
153 }
154
155 /**
156 * disable_ecrc_checking - disables PCIe ECRC checking for a device
157 * @dev: the PCI device
158 *
159 * Returns 0 on success, or negative on failure.
160 */
disable_ecrc_checking(struct pci_dev * dev)161 static int disable_ecrc_checking(struct pci_dev *dev)
162 {
163 int aer = dev->aer_cap;
164 u32 reg32;
165
166 if (!aer)
167 return -ENODEV;
168
169 pci_read_config_dword(dev, aer + PCI_ERR_CAP, ®32);
170 reg32 &= ~(PCI_ERR_CAP_ECRC_GENE | PCI_ERR_CAP_ECRC_CHKE);
171 pci_write_config_dword(dev, aer + PCI_ERR_CAP, reg32);
172
173 return 0;
174 }
175
176 /**
177 * pcie_set_ecrc_checking - set/unset PCIe ECRC checking for a device based on global policy
178 * @dev: the PCI device
179 */
pcie_set_ecrc_checking(struct pci_dev * dev)180 void pcie_set_ecrc_checking(struct pci_dev *dev)
181 {
182 switch (ecrc_policy) {
183 case ECRC_POLICY_DEFAULT:
184 return;
185 case ECRC_POLICY_OFF:
186 disable_ecrc_checking(dev);
187 break;
188 case ECRC_POLICY_ON:
189 enable_ecrc_checking(dev);
190 break;
191 default:
192 return;
193 }
194 }
195
196 /**
197 * pcie_ecrc_get_policy - parse kernel command-line ecrc option
198 * @str: ECRC policy from kernel command line to use
199 */
pcie_ecrc_get_policy(char * str)200 void pcie_ecrc_get_policy(char *str)
201 {
202 int i;
203
204 i = match_string(ecrc_policy_str, ARRAY_SIZE(ecrc_policy_str), str);
205 if (i < 0)
206 return;
207
208 ecrc_policy = i;
209 }
210 #endif /* CONFIG_PCIE_ECRC */
211
212 #define PCI_EXP_AER_FLAGS (PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE | \
213 PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE)
214
pcie_aer_is_native(struct pci_dev * dev)215 int pcie_aer_is_native(struct pci_dev *dev)
216 {
217 struct pci_host_bridge *host = pci_find_host_bridge(dev->bus);
218
219 if (!dev->aer_cap)
220 return 0;
221
222 return pcie_ports_native || host->native_aer;
223 }
224
pci_enable_pcie_error_reporting(struct pci_dev * dev)225 int pci_enable_pcie_error_reporting(struct pci_dev *dev)
226 {
227 int rc;
228
229 if (!pcie_aer_is_native(dev))
230 return -EIO;
231
232 rc = pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_AER_FLAGS);
233 return pcibios_err_to_errno(rc);
234 }
235 EXPORT_SYMBOL_GPL(pci_enable_pcie_error_reporting);
236
pci_disable_pcie_error_reporting(struct pci_dev * dev)237 int pci_disable_pcie_error_reporting(struct pci_dev *dev)
238 {
239 int rc;
240
241 if (!pcie_aer_is_native(dev))
242 return -EIO;
243
244 rc = pcie_capability_clear_word(dev, PCI_EXP_DEVCTL, PCI_EXP_AER_FLAGS);
245 return pcibios_err_to_errno(rc);
246 }
247 EXPORT_SYMBOL_GPL(pci_disable_pcie_error_reporting);
248
pci_aer_clear_nonfatal_status(struct pci_dev * dev)249 int pci_aer_clear_nonfatal_status(struct pci_dev *dev)
250 {
251 int aer = dev->aer_cap;
252 u32 status, sev;
253
254 if (!pcie_aer_is_native(dev))
255 return -EIO;
256
257 /* Clear status bits for ERR_NONFATAL errors only */
258 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, &status);
259 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_SEVER, &sev);
260 status &= ~sev;
261 if (status)
262 pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, status);
263
264 return 0;
265 }
266 EXPORT_SYMBOL_GPL(pci_aer_clear_nonfatal_status);
267
pci_aer_clear_fatal_status(struct pci_dev * dev)268 void pci_aer_clear_fatal_status(struct pci_dev *dev)
269 {
270 int aer = dev->aer_cap;
271 u32 status, sev;
272
273 if (!pcie_aer_is_native(dev))
274 return;
275
276 /* Clear status bits for ERR_FATAL errors only */
277 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, &status);
278 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_SEVER, &sev);
279 status &= sev;
280 if (status)
281 pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, status);
282 }
283
284 /**
285 * pci_aer_raw_clear_status - Clear AER error registers.
286 * @dev: the PCI device
287 *
288 * Clearing AER error status registers unconditionally, regardless of
289 * whether they're owned by firmware or the OS.
290 *
291 * Returns 0 on success, or negative on failure.
292 */
pci_aer_raw_clear_status(struct pci_dev * dev)293 int pci_aer_raw_clear_status(struct pci_dev *dev)
294 {
295 int aer = dev->aer_cap;
296 u32 status;
297 int port_type;
298
299 if (!aer)
300 return -EIO;
301
302 port_type = pci_pcie_type(dev);
303 if (port_type == PCI_EXP_TYPE_ROOT_PORT ||
304 port_type == PCI_EXP_TYPE_RC_EC) {
305 pci_read_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, &status);
306 pci_write_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, status);
307 }
308
309 pci_read_config_dword(dev, aer + PCI_ERR_COR_STATUS, &status);
310 pci_write_config_dword(dev, aer + PCI_ERR_COR_STATUS, status);
311
312 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, &status);
313 pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, status);
314
315 return 0;
316 }
317
pci_aer_clear_status(struct pci_dev * dev)318 int pci_aer_clear_status(struct pci_dev *dev)
319 {
320 if (!pcie_aer_is_native(dev))
321 return -EIO;
322
323 return pci_aer_raw_clear_status(dev);
324 }
325
pci_save_aer_state(struct pci_dev * dev)326 void pci_save_aer_state(struct pci_dev *dev)
327 {
328 int aer = dev->aer_cap;
329 struct pci_cap_saved_state *save_state;
330 u32 *cap;
331
332 if (!aer)
333 return;
334
335 save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_ERR);
336 if (!save_state)
337 return;
338
339 cap = &save_state->cap.data[0];
340 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, cap++);
341 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_SEVER, cap++);
342 pci_read_config_dword(dev, aer + PCI_ERR_COR_MASK, cap++);
343 pci_read_config_dword(dev, aer + PCI_ERR_CAP, cap++);
344 if (pcie_cap_has_rtctl(dev))
345 pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, cap++);
346 }
347
pci_restore_aer_state(struct pci_dev * dev)348 void pci_restore_aer_state(struct pci_dev *dev)
349 {
350 int aer = dev->aer_cap;
351 struct pci_cap_saved_state *save_state;
352 u32 *cap;
353
354 if (!aer)
355 return;
356
357 save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_ERR);
358 if (!save_state)
359 return;
360
361 cap = &save_state->cap.data[0];
362 pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, *cap++);
363 pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_SEVER, *cap++);
364 pci_write_config_dword(dev, aer + PCI_ERR_COR_MASK, *cap++);
365 pci_write_config_dword(dev, aer + PCI_ERR_CAP, *cap++);
366 if (pcie_cap_has_rtctl(dev))
367 pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, *cap++);
368 }
369
pci_aer_init(struct pci_dev * dev)370 void pci_aer_init(struct pci_dev *dev)
371 {
372 int n;
373
374 dev->aer_cap = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
375 if (!dev->aer_cap)
376 return;
377
378 dev->aer_stats = kzalloc(sizeof(struct aer_stats), GFP_KERNEL);
379
380 /*
381 * We save/restore PCI_ERR_UNCOR_MASK, PCI_ERR_UNCOR_SEVER,
382 * PCI_ERR_COR_MASK, and PCI_ERR_CAP. Root and Root Complex Event
383 * Collectors also implement PCI_ERR_ROOT_COMMAND (PCIe r5.0, sec
384 * 7.8.4).
385 */
386 n = pcie_cap_has_rtctl(dev) ? 5 : 4;
387 pci_add_ext_cap_save_buffer(dev, PCI_EXT_CAP_ID_ERR, sizeof(u32) * n);
388
389 pci_aer_clear_status(dev);
390 }
391
pci_aer_exit(struct pci_dev * dev)392 void pci_aer_exit(struct pci_dev *dev)
393 {
394 kfree(dev->aer_stats);
395 dev->aer_stats = NULL;
396 }
397
398 #define AER_AGENT_RECEIVER 0
399 #define AER_AGENT_REQUESTER 1
400 #define AER_AGENT_COMPLETER 2
401 #define AER_AGENT_TRANSMITTER 3
402
403 #define AER_AGENT_REQUESTER_MASK(t) ((t == AER_CORRECTABLE) ? \
404 0 : (PCI_ERR_UNC_COMP_TIME|PCI_ERR_UNC_UNSUP))
405 #define AER_AGENT_COMPLETER_MASK(t) ((t == AER_CORRECTABLE) ? \
406 0 : PCI_ERR_UNC_COMP_ABORT)
407 #define AER_AGENT_TRANSMITTER_MASK(t) ((t == AER_CORRECTABLE) ? \
408 (PCI_ERR_COR_REP_ROLL|PCI_ERR_COR_REP_TIMER) : 0)
409
410 #define AER_GET_AGENT(t, e) \
411 ((e & AER_AGENT_COMPLETER_MASK(t)) ? AER_AGENT_COMPLETER : \
412 (e & AER_AGENT_REQUESTER_MASK(t)) ? AER_AGENT_REQUESTER : \
413 (e & AER_AGENT_TRANSMITTER_MASK(t)) ? AER_AGENT_TRANSMITTER : \
414 AER_AGENT_RECEIVER)
415
416 #define AER_PHYSICAL_LAYER_ERROR 0
417 #define AER_DATA_LINK_LAYER_ERROR 1
418 #define AER_TRANSACTION_LAYER_ERROR 2
419
420 #define AER_PHYSICAL_LAYER_ERROR_MASK(t) ((t == AER_CORRECTABLE) ? \
421 PCI_ERR_COR_RCVR : 0)
422 #define AER_DATA_LINK_LAYER_ERROR_MASK(t) ((t == AER_CORRECTABLE) ? \
423 (PCI_ERR_COR_BAD_TLP| \
424 PCI_ERR_COR_BAD_DLLP| \
425 PCI_ERR_COR_REP_ROLL| \
426 PCI_ERR_COR_REP_TIMER) : PCI_ERR_UNC_DLP)
427
428 #define AER_GET_LAYER_ERROR(t, e) \
429 ((e & AER_PHYSICAL_LAYER_ERROR_MASK(t)) ? AER_PHYSICAL_LAYER_ERROR : \
430 (e & AER_DATA_LINK_LAYER_ERROR_MASK(t)) ? AER_DATA_LINK_LAYER_ERROR : \
431 AER_TRANSACTION_LAYER_ERROR)
432
433 /*
434 * AER error strings
435 */
436 static const char *aer_error_severity_string[] = {
437 "Uncorrected (Non-Fatal)",
438 "Uncorrected (Fatal)",
439 "Corrected"
440 };
441
442 static const char *aer_error_layer[] = {
443 "Physical Layer",
444 "Data Link Layer",
445 "Transaction Layer"
446 };
447
448 static const char *aer_correctable_error_string[] = {
449 "RxErr", /* Bit Position 0 */
450 NULL,
451 NULL,
452 NULL,
453 NULL,
454 NULL,
455 "BadTLP", /* Bit Position 6 */
456 "BadDLLP", /* Bit Position 7 */
457 "Rollover", /* Bit Position 8 */
458 NULL,
459 NULL,
460 NULL,
461 "Timeout", /* Bit Position 12 */
462 "NonFatalErr", /* Bit Position 13 */
463 "CorrIntErr", /* Bit Position 14 */
464 "HeaderOF", /* Bit Position 15 */
465 NULL, /* Bit Position 16 */
466 NULL, /* Bit Position 17 */
467 NULL, /* Bit Position 18 */
468 NULL, /* Bit Position 19 */
469 NULL, /* Bit Position 20 */
470 NULL, /* Bit Position 21 */
471 NULL, /* Bit Position 22 */
472 NULL, /* Bit Position 23 */
473 NULL, /* Bit Position 24 */
474 NULL, /* Bit Position 25 */
475 NULL, /* Bit Position 26 */
476 NULL, /* Bit Position 27 */
477 NULL, /* Bit Position 28 */
478 NULL, /* Bit Position 29 */
479 NULL, /* Bit Position 30 */
480 NULL, /* Bit Position 31 */
481 };
482
483 static const char *aer_uncorrectable_error_string[] = {
484 "Undefined", /* Bit Position 0 */
485 NULL,
486 NULL,
487 NULL,
488 "DLP", /* Bit Position 4 */
489 "SDES", /* Bit Position 5 */
490 NULL,
491 NULL,
492 NULL,
493 NULL,
494 NULL,
495 NULL,
496 "TLP", /* Bit Position 12 */
497 "FCP", /* Bit Position 13 */
498 "CmpltTO", /* Bit Position 14 */
499 "CmpltAbrt", /* Bit Position 15 */
500 "UnxCmplt", /* Bit Position 16 */
501 "RxOF", /* Bit Position 17 */
502 "MalfTLP", /* Bit Position 18 */
503 "ECRC", /* Bit Position 19 */
504 "UnsupReq", /* Bit Position 20 */
505 "ACSViol", /* Bit Position 21 */
506 "UncorrIntErr", /* Bit Position 22 */
507 "BlockedTLP", /* Bit Position 23 */
508 "AtomicOpBlocked", /* Bit Position 24 */
509 "TLPBlockedErr", /* Bit Position 25 */
510 "PoisonTLPBlocked", /* Bit Position 26 */
511 NULL, /* Bit Position 27 */
512 NULL, /* Bit Position 28 */
513 NULL, /* Bit Position 29 */
514 NULL, /* Bit Position 30 */
515 NULL, /* Bit Position 31 */
516 };
517
518 static const char *aer_agent_string[] = {
519 "Receiver ID",
520 "Requester ID",
521 "Completer ID",
522 "Transmitter ID"
523 };
524
525 #define aer_stats_dev_attr(name, stats_array, strings_array, \
526 total_string, total_field) \
527 static ssize_t \
528 name##_show(struct device *dev, struct device_attribute *attr, \
529 char *buf) \
530 { \
531 unsigned int i; \
532 struct pci_dev *pdev = to_pci_dev(dev); \
533 u64 *stats = pdev->aer_stats->stats_array; \
534 size_t len = 0; \
535 \
536 for (i = 0; i < ARRAY_SIZE(strings_array); i++) { \
537 if (strings_array[i]) \
538 len += sysfs_emit_at(buf, len, "%s %llu\n", \
539 strings_array[i], \
540 stats[i]); \
541 else if (stats[i]) \
542 len += sysfs_emit_at(buf, len, \
543 #stats_array "_bit[%d] %llu\n",\
544 i, stats[i]); \
545 } \
546 len += sysfs_emit_at(buf, len, "TOTAL_%s %llu\n", total_string, \
547 pdev->aer_stats->total_field); \
548 return len; \
549 } \
550 static DEVICE_ATTR_RO(name)
551
552 aer_stats_dev_attr(aer_dev_correctable, dev_cor_errs,
553 aer_correctable_error_string, "ERR_COR",
554 dev_total_cor_errs);
555 aer_stats_dev_attr(aer_dev_fatal, dev_fatal_errs,
556 aer_uncorrectable_error_string, "ERR_FATAL",
557 dev_total_fatal_errs);
558 aer_stats_dev_attr(aer_dev_nonfatal, dev_nonfatal_errs,
559 aer_uncorrectable_error_string, "ERR_NONFATAL",
560 dev_total_nonfatal_errs);
561
562 #define aer_stats_rootport_attr(name, field) \
563 static ssize_t \
564 name##_show(struct device *dev, struct device_attribute *attr, \
565 char *buf) \
566 { \
567 struct pci_dev *pdev = to_pci_dev(dev); \
568 return sysfs_emit(buf, "%llu\n", pdev->aer_stats->field); \
569 } \
570 static DEVICE_ATTR_RO(name)
571
572 aer_stats_rootport_attr(aer_rootport_total_err_cor,
573 rootport_total_cor_errs);
574 aer_stats_rootport_attr(aer_rootport_total_err_fatal,
575 rootport_total_fatal_errs);
576 aer_stats_rootport_attr(aer_rootport_total_err_nonfatal,
577 rootport_total_nonfatal_errs);
578
579 static struct attribute *aer_stats_attrs[] __ro_after_init = {
580 &dev_attr_aer_dev_correctable.attr,
581 &dev_attr_aer_dev_fatal.attr,
582 &dev_attr_aer_dev_nonfatal.attr,
583 &dev_attr_aer_rootport_total_err_cor.attr,
584 &dev_attr_aer_rootport_total_err_fatal.attr,
585 &dev_attr_aer_rootport_total_err_nonfatal.attr,
586 NULL
587 };
588
aer_stats_attrs_are_visible(struct kobject * kobj,struct attribute * a,int n)589 static umode_t aer_stats_attrs_are_visible(struct kobject *kobj,
590 struct attribute *a, int n)
591 {
592 struct device *dev = kobj_to_dev(kobj);
593 struct pci_dev *pdev = to_pci_dev(dev);
594
595 if (!pdev->aer_stats)
596 return 0;
597
598 if ((a == &dev_attr_aer_rootport_total_err_cor.attr ||
599 a == &dev_attr_aer_rootport_total_err_fatal.attr ||
600 a == &dev_attr_aer_rootport_total_err_nonfatal.attr) &&
601 ((pci_pcie_type(pdev) != PCI_EXP_TYPE_ROOT_PORT) &&
602 (pci_pcie_type(pdev) != PCI_EXP_TYPE_RC_EC)))
603 return 0;
604
605 return a->mode;
606 }
607
608 const struct attribute_group aer_stats_attr_group = {
609 .attrs = aer_stats_attrs,
610 .is_visible = aer_stats_attrs_are_visible,
611 };
612
pci_dev_aer_stats_incr(struct pci_dev * pdev,struct aer_err_info * info)613 static void pci_dev_aer_stats_incr(struct pci_dev *pdev,
614 struct aer_err_info *info)
615 {
616 unsigned long status = info->status & ~info->mask;
617 int i, max = -1;
618 u64 *counter = NULL;
619 struct aer_stats *aer_stats = pdev->aer_stats;
620
621 if (!aer_stats)
622 return;
623
624 switch (info->severity) {
625 case AER_CORRECTABLE:
626 aer_stats->dev_total_cor_errs++;
627 counter = &aer_stats->dev_cor_errs[0];
628 max = AER_MAX_TYPEOF_COR_ERRS;
629 break;
630 case AER_NONFATAL:
631 aer_stats->dev_total_nonfatal_errs++;
632 counter = &aer_stats->dev_nonfatal_errs[0];
633 max = AER_MAX_TYPEOF_UNCOR_ERRS;
634 break;
635 case AER_FATAL:
636 aer_stats->dev_total_fatal_errs++;
637 counter = &aer_stats->dev_fatal_errs[0];
638 max = AER_MAX_TYPEOF_UNCOR_ERRS;
639 break;
640 }
641
642 for_each_set_bit(i, &status, max)
643 counter[i]++;
644 }
645
pci_rootport_aer_stats_incr(struct pci_dev * pdev,struct aer_err_source * e_src)646 static void pci_rootport_aer_stats_incr(struct pci_dev *pdev,
647 struct aer_err_source *e_src)
648 {
649 struct aer_stats *aer_stats = pdev->aer_stats;
650
651 if (!aer_stats)
652 return;
653
654 if (e_src->status & PCI_ERR_ROOT_COR_RCV)
655 aer_stats->rootport_total_cor_errs++;
656
657 if (e_src->status & PCI_ERR_ROOT_UNCOR_RCV) {
658 if (e_src->status & PCI_ERR_ROOT_FATAL_RCV)
659 aer_stats->rootport_total_fatal_errs++;
660 else
661 aer_stats->rootport_total_nonfatal_errs++;
662 }
663 }
664
__print_tlp_header(struct pci_dev * dev,struct aer_header_log_regs * t)665 static void __print_tlp_header(struct pci_dev *dev,
666 struct aer_header_log_regs *t)
667 {
668 pci_err(dev, " TLP Header: %08x %08x %08x %08x\n",
669 t->dw0, t->dw1, t->dw2, t->dw3);
670 }
671
__aer_print_error(struct pci_dev * dev,struct aer_err_info * info)672 static void __aer_print_error(struct pci_dev *dev,
673 struct aer_err_info *info)
674 {
675 const char **strings;
676 unsigned long status = info->status & ~info->mask;
677 const char *level, *errmsg;
678 int i;
679
680 if (info->severity == AER_CORRECTABLE) {
681 strings = aer_correctable_error_string;
682 level = KERN_WARNING;
683 } else {
684 strings = aer_uncorrectable_error_string;
685 level = KERN_ERR;
686 }
687
688 for_each_set_bit(i, &status, 32) {
689 errmsg = strings[i];
690 if (!errmsg)
691 errmsg = "Unknown Error Bit";
692
693 pci_printk(level, dev, " [%2d] %-22s%s\n", i, errmsg,
694 info->first_error == i ? " (First)" : "");
695 }
696 pci_dev_aer_stats_incr(dev, info);
697 }
698
aer_print_error(struct pci_dev * dev,struct aer_err_info * info)699 void aer_print_error(struct pci_dev *dev, struct aer_err_info *info)
700 {
701 int layer, agent;
702 int id = ((dev->bus->number << 8) | dev->devfn);
703 const char *level;
704
705 if (!info->status) {
706 pci_err(dev, "PCIe Bus Error: severity=%s, type=Inaccessible, (Unregistered Agent ID)\n",
707 aer_error_severity_string[info->severity]);
708 goto out;
709 }
710
711 layer = AER_GET_LAYER_ERROR(info->severity, info->status);
712 agent = AER_GET_AGENT(info->severity, info->status);
713
714 level = (info->severity == AER_CORRECTABLE) ? KERN_WARNING : KERN_ERR;
715
716 pci_printk(level, dev, "PCIe Bus Error: severity=%s, type=%s, (%s)\n",
717 aer_error_severity_string[info->severity],
718 aer_error_layer[layer], aer_agent_string[agent]);
719
720 pci_printk(level, dev, " device [%04x:%04x] error status/mask=%08x/%08x\n",
721 dev->vendor, dev->device, info->status, info->mask);
722
723 __aer_print_error(dev, info);
724
725 if (info->tlp_header_valid)
726 __print_tlp_header(dev, &info->tlp);
727
728 out:
729 if (info->id && info->error_dev_num > 1 && info->id == id)
730 pci_err(dev, " Error of this Agent is reported first\n");
731
732 trace_aer_event(dev_name(&dev->dev), (info->status & ~info->mask),
733 info->severity, info->tlp_header_valid, &info->tlp);
734 }
735
aer_print_port_info(struct pci_dev * dev,struct aer_err_info * info)736 static void aer_print_port_info(struct pci_dev *dev, struct aer_err_info *info)
737 {
738 u8 bus = info->id >> 8;
739 u8 devfn = info->id & 0xff;
740
741 pci_info(dev, "%s%s error received: %04x:%02x:%02x.%d\n",
742 info->multi_error_valid ? "Multiple " : "",
743 aer_error_severity_string[info->severity],
744 pci_domain_nr(dev->bus), bus, PCI_SLOT(devfn),
745 PCI_FUNC(devfn));
746 }
747
748 #ifdef CONFIG_ACPI_APEI_PCIEAER
cper_severity_to_aer(int cper_severity)749 int cper_severity_to_aer(int cper_severity)
750 {
751 switch (cper_severity) {
752 case CPER_SEV_RECOVERABLE:
753 return AER_NONFATAL;
754 case CPER_SEV_FATAL:
755 return AER_FATAL;
756 default:
757 return AER_CORRECTABLE;
758 }
759 }
760 EXPORT_SYMBOL_GPL(cper_severity_to_aer);
761
cper_print_aer(struct pci_dev * dev,int aer_severity,struct aer_capability_regs * aer)762 void cper_print_aer(struct pci_dev *dev, int aer_severity,
763 struct aer_capability_regs *aer)
764 {
765 int layer, agent, tlp_header_valid = 0;
766 u32 status, mask;
767 struct aer_err_info info;
768
769 if (aer_severity == AER_CORRECTABLE) {
770 status = aer->cor_status;
771 mask = aer->cor_mask;
772 } else {
773 status = aer->uncor_status;
774 mask = aer->uncor_mask;
775 tlp_header_valid = status & AER_LOG_TLP_MASKS;
776 }
777
778 layer = AER_GET_LAYER_ERROR(aer_severity, status);
779 agent = AER_GET_AGENT(aer_severity, status);
780
781 memset(&info, 0, sizeof(info));
782 info.severity = aer_severity;
783 info.status = status;
784 info.mask = mask;
785 info.first_error = PCI_ERR_CAP_FEP(aer->cap_control);
786
787 pci_err(dev, "aer_status: 0x%08x, aer_mask: 0x%08x\n", status, mask);
788 __aer_print_error(dev, &info);
789 pci_err(dev, "aer_layer=%s, aer_agent=%s\n",
790 aer_error_layer[layer], aer_agent_string[agent]);
791
792 if (aer_severity != AER_CORRECTABLE)
793 pci_err(dev, "aer_uncor_severity: 0x%08x\n",
794 aer->uncor_severity);
795
796 if (tlp_header_valid)
797 __print_tlp_header(dev, &aer->header_log);
798
799 trace_aer_event(dev_name(&dev->dev), (status & ~mask),
800 aer_severity, tlp_header_valid, &aer->header_log);
801 }
802 #endif
803
804 /**
805 * add_error_device - list device to be handled
806 * @e_info: pointer to error info
807 * @dev: pointer to pci_dev to be added
808 */
add_error_device(struct aer_err_info * e_info,struct pci_dev * dev)809 static int add_error_device(struct aer_err_info *e_info, struct pci_dev *dev)
810 {
811 if (e_info->error_dev_num < AER_MAX_MULTI_ERR_DEVICES) {
812 e_info->dev[e_info->error_dev_num] = pci_dev_get(dev);
813 e_info->error_dev_num++;
814 return 0;
815 }
816 return -ENOSPC;
817 }
818
819 /**
820 * is_error_source - check whether the device is source of reported error
821 * @dev: pointer to pci_dev to be checked
822 * @e_info: pointer to reported error info
823 */
is_error_source(struct pci_dev * dev,struct aer_err_info * e_info)824 static bool is_error_source(struct pci_dev *dev, struct aer_err_info *e_info)
825 {
826 int aer = dev->aer_cap;
827 u32 status, mask;
828 u16 reg16;
829
830 /*
831 * When bus id is equal to 0, it might be a bad id
832 * reported by root port.
833 */
834 if ((PCI_BUS_NUM(e_info->id) != 0) &&
835 !(dev->bus->bus_flags & PCI_BUS_FLAGS_NO_AERSID)) {
836 /* Device ID match? */
837 if (e_info->id == ((dev->bus->number << 8) | dev->devfn))
838 return true;
839
840 /* Continue id comparing if there is no multiple error */
841 if (!e_info->multi_error_valid)
842 return false;
843 }
844
845 /*
846 * When either
847 * 1) bus id is equal to 0. Some ports might lose the bus
848 * id of error source id;
849 * 2) bus flag PCI_BUS_FLAGS_NO_AERSID is set
850 * 3) There are multiple errors and prior ID comparing fails;
851 * We check AER status registers to find possible reporter.
852 */
853 if (atomic_read(&dev->enable_cnt) == 0)
854 return false;
855
856 /* Check if AER is enabled */
857 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, ®16);
858 if (!(reg16 & PCI_EXP_AER_FLAGS))
859 return false;
860
861 if (!aer)
862 return false;
863
864 /* Check if error is recorded */
865 if (e_info->severity == AER_CORRECTABLE) {
866 pci_read_config_dword(dev, aer + PCI_ERR_COR_STATUS, &status);
867 pci_read_config_dword(dev, aer + PCI_ERR_COR_MASK, &mask);
868 } else {
869 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, &status);
870 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, &mask);
871 }
872 if (status & ~mask)
873 return true;
874
875 return false;
876 }
877
find_device_iter(struct pci_dev * dev,void * data)878 static int find_device_iter(struct pci_dev *dev, void *data)
879 {
880 struct aer_err_info *e_info = (struct aer_err_info *)data;
881
882 if (is_error_source(dev, e_info)) {
883 /* List this device */
884 if (add_error_device(e_info, dev)) {
885 /* We cannot handle more... Stop iteration */
886 /* TODO: Should print error message here? */
887 return 1;
888 }
889
890 /* If there is only a single error, stop iteration */
891 if (!e_info->multi_error_valid)
892 return 1;
893 }
894 return 0;
895 }
896
897 /**
898 * find_source_device - search through device hierarchy for source device
899 * @parent: pointer to Root Port pci_dev data structure
900 * @e_info: including detailed error information such like id
901 *
902 * Return true if found.
903 *
904 * Invoked by DPC when error is detected at the Root Port.
905 * Caller of this function must set id, severity, and multi_error_valid of
906 * struct aer_err_info pointed by @e_info properly. This function must fill
907 * e_info->error_dev_num and e_info->dev[], based on the given information.
908 */
find_source_device(struct pci_dev * parent,struct aer_err_info * e_info)909 static bool find_source_device(struct pci_dev *parent,
910 struct aer_err_info *e_info)
911 {
912 struct pci_dev *dev = parent;
913 int result;
914
915 /* Must reset in this function */
916 e_info->error_dev_num = 0;
917
918 /* Is Root Port an agent that sends error message? */
919 result = find_device_iter(dev, e_info);
920 if (result)
921 return true;
922
923 if (pci_pcie_type(parent) == PCI_EXP_TYPE_RC_EC)
924 pcie_walk_rcec(parent, find_device_iter, e_info);
925 else
926 pci_walk_bus(parent->subordinate, find_device_iter, e_info);
927
928 if (!e_info->error_dev_num) {
929 pci_info(parent, "can't find device of ID%04x\n", e_info->id);
930 return false;
931 }
932 return true;
933 }
934
935 /**
936 * handle_error_source - handle logging error into an event log
937 * @dev: pointer to pci_dev data structure of error source device
938 * @info: comprehensive error information
939 *
940 * Invoked when an error being detected by Root Port.
941 */
handle_error_source(struct pci_dev * dev,struct aer_err_info * info)942 static void handle_error_source(struct pci_dev *dev, struct aer_err_info *info)
943 {
944 int aer = dev->aer_cap;
945
946 if (info->severity == AER_CORRECTABLE) {
947 /*
948 * Correctable error does not need software intervention.
949 * No need to go through error recovery process.
950 */
951 if (aer)
952 pci_write_config_dword(dev, aer + PCI_ERR_COR_STATUS,
953 info->status);
954 if (pcie_aer_is_native(dev))
955 pcie_clear_device_status(dev);
956 } else if (info->severity == AER_NONFATAL)
957 pcie_do_recovery(dev, pci_channel_io_normal, aer_root_reset);
958 else if (info->severity == AER_FATAL)
959 pcie_do_recovery(dev, pci_channel_io_frozen, aer_root_reset);
960 pci_dev_put(dev);
961 }
962
963 #ifdef CONFIG_ACPI_APEI_PCIEAER
964
965 #define AER_RECOVER_RING_ORDER 4
966 #define AER_RECOVER_RING_SIZE (1 << AER_RECOVER_RING_ORDER)
967
968 struct aer_recover_entry {
969 u8 bus;
970 u8 devfn;
971 u16 domain;
972 int severity;
973 struct aer_capability_regs *regs;
974 };
975
976 static DEFINE_KFIFO(aer_recover_ring, struct aer_recover_entry,
977 AER_RECOVER_RING_SIZE);
978
aer_recover_work_func(struct work_struct * work)979 static void aer_recover_work_func(struct work_struct *work)
980 {
981 struct aer_recover_entry entry;
982 struct pci_dev *pdev;
983
984 while (kfifo_get(&aer_recover_ring, &entry)) {
985 pdev = pci_get_domain_bus_and_slot(entry.domain, entry.bus,
986 entry.devfn);
987 if (!pdev) {
988 pr_err("no pci_dev for %04x:%02x:%02x.%x\n",
989 entry.domain, entry.bus,
990 PCI_SLOT(entry.devfn), PCI_FUNC(entry.devfn));
991 continue;
992 }
993 cper_print_aer(pdev, entry.severity, entry.regs);
994 if (entry.severity == AER_NONFATAL)
995 pcie_do_recovery(pdev, pci_channel_io_normal,
996 aer_root_reset);
997 else if (entry.severity == AER_FATAL)
998 pcie_do_recovery(pdev, pci_channel_io_frozen,
999 aer_root_reset);
1000 pci_dev_put(pdev);
1001 }
1002 }
1003
1004 /*
1005 * Mutual exclusion for writers of aer_recover_ring, reader side don't
1006 * need lock, because there is only one reader and lock is not needed
1007 * between reader and writer.
1008 */
1009 static DEFINE_SPINLOCK(aer_recover_ring_lock);
1010 static DECLARE_WORK(aer_recover_work, aer_recover_work_func);
1011
aer_recover_queue(int domain,unsigned int bus,unsigned int devfn,int severity,struct aer_capability_regs * aer_regs)1012 void aer_recover_queue(int domain, unsigned int bus, unsigned int devfn,
1013 int severity, struct aer_capability_regs *aer_regs)
1014 {
1015 struct aer_recover_entry entry = {
1016 .bus = bus,
1017 .devfn = devfn,
1018 .domain = domain,
1019 .severity = severity,
1020 .regs = aer_regs,
1021 };
1022
1023 if (kfifo_in_spinlocked(&aer_recover_ring, &entry, 1,
1024 &aer_recover_ring_lock))
1025 schedule_work(&aer_recover_work);
1026 else
1027 pr_err("buffer overflow in recovery for %04x:%02x:%02x.%x\n",
1028 domain, bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
1029 }
1030 EXPORT_SYMBOL_GPL(aer_recover_queue);
1031 #endif
1032
1033 /**
1034 * aer_get_device_error_info - read error status from dev and store it to info
1035 * @dev: pointer to the device expected to have a error record
1036 * @info: pointer to structure to store the error record
1037 *
1038 * Return 1 on success, 0 on error.
1039 *
1040 * Note that @info is reused among all error devices. Clear fields properly.
1041 */
aer_get_device_error_info(struct pci_dev * dev,struct aer_err_info * info)1042 int aer_get_device_error_info(struct pci_dev *dev, struct aer_err_info *info)
1043 {
1044 int type = pci_pcie_type(dev);
1045 int aer = dev->aer_cap;
1046 int temp;
1047
1048 /* Must reset in this function */
1049 info->status = 0;
1050 info->tlp_header_valid = 0;
1051
1052 /* The device might not support AER */
1053 if (!aer)
1054 return 0;
1055
1056 if (info->severity == AER_CORRECTABLE) {
1057 pci_read_config_dword(dev, aer + PCI_ERR_COR_STATUS,
1058 &info->status);
1059 pci_read_config_dword(dev, aer + PCI_ERR_COR_MASK,
1060 &info->mask);
1061 if (!(info->status & ~info->mask))
1062 return 0;
1063 } else if (type == PCI_EXP_TYPE_ROOT_PORT ||
1064 type == PCI_EXP_TYPE_RC_EC ||
1065 type == PCI_EXP_TYPE_DOWNSTREAM ||
1066 info->severity == AER_NONFATAL) {
1067
1068 /* Link is still healthy for IO reads */
1069 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS,
1070 &info->status);
1071 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_MASK,
1072 &info->mask);
1073 if (!(info->status & ~info->mask))
1074 return 0;
1075
1076 /* Get First Error Pointer */
1077 pci_read_config_dword(dev, aer + PCI_ERR_CAP, &temp);
1078 info->first_error = PCI_ERR_CAP_FEP(temp);
1079
1080 if (info->status & AER_LOG_TLP_MASKS) {
1081 info->tlp_header_valid = 1;
1082 pci_read_config_dword(dev,
1083 aer + PCI_ERR_HEADER_LOG, &info->tlp.dw0);
1084 pci_read_config_dword(dev,
1085 aer + PCI_ERR_HEADER_LOG + 4, &info->tlp.dw1);
1086 pci_read_config_dword(dev,
1087 aer + PCI_ERR_HEADER_LOG + 8, &info->tlp.dw2);
1088 pci_read_config_dword(dev,
1089 aer + PCI_ERR_HEADER_LOG + 12, &info->tlp.dw3);
1090 }
1091 }
1092
1093 return 1;
1094 }
1095
aer_process_err_devices(struct aer_err_info * e_info)1096 static inline void aer_process_err_devices(struct aer_err_info *e_info)
1097 {
1098 int i;
1099
1100 /* Report all before handle them, not to lost records by reset etc. */
1101 for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
1102 if (aer_get_device_error_info(e_info->dev[i], e_info))
1103 aer_print_error(e_info->dev[i], e_info);
1104 }
1105 for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
1106 if (aer_get_device_error_info(e_info->dev[i], e_info))
1107 handle_error_source(e_info->dev[i], e_info);
1108 }
1109 }
1110
1111 /**
1112 * aer_isr_one_error - consume an error detected by root port
1113 * @rpc: pointer to the root port which holds an error
1114 * @e_src: pointer to an error source
1115 */
aer_isr_one_error(struct aer_rpc * rpc,struct aer_err_source * e_src)1116 static void aer_isr_one_error(struct aer_rpc *rpc,
1117 struct aer_err_source *e_src)
1118 {
1119 struct pci_dev *pdev = rpc->rpd;
1120 struct aer_err_info e_info;
1121
1122 pci_rootport_aer_stats_incr(pdev, e_src);
1123
1124 /*
1125 * There is a possibility that both correctable error and
1126 * uncorrectable error being logged. Report correctable error first.
1127 */
1128 if (e_src->status & PCI_ERR_ROOT_COR_RCV) {
1129 e_info.id = ERR_COR_ID(e_src->id);
1130 e_info.severity = AER_CORRECTABLE;
1131
1132 if (e_src->status & PCI_ERR_ROOT_MULTI_COR_RCV)
1133 e_info.multi_error_valid = 1;
1134 else
1135 e_info.multi_error_valid = 0;
1136 aer_print_port_info(pdev, &e_info);
1137
1138 if (find_source_device(pdev, &e_info))
1139 aer_process_err_devices(&e_info);
1140 }
1141
1142 if (e_src->status & PCI_ERR_ROOT_UNCOR_RCV) {
1143 e_info.id = ERR_UNCOR_ID(e_src->id);
1144
1145 if (e_src->status & PCI_ERR_ROOT_FATAL_RCV)
1146 e_info.severity = AER_FATAL;
1147 else
1148 e_info.severity = AER_NONFATAL;
1149
1150 if (e_src->status & PCI_ERR_ROOT_MULTI_UNCOR_RCV)
1151 e_info.multi_error_valid = 1;
1152 else
1153 e_info.multi_error_valid = 0;
1154
1155 aer_print_port_info(pdev, &e_info);
1156
1157 if (find_source_device(pdev, &e_info))
1158 aer_process_err_devices(&e_info);
1159 }
1160 }
1161
1162 /**
1163 * aer_isr - consume errors detected by root port
1164 * @irq: IRQ assigned to Root Port
1165 * @context: pointer to Root Port data structure
1166 *
1167 * Invoked, as DPC, when root port records new detected error
1168 */
aer_isr(int irq,void * context)1169 static irqreturn_t aer_isr(int irq, void *context)
1170 {
1171 struct pcie_device *dev = (struct pcie_device *)context;
1172 struct aer_rpc *rpc = get_service_data(dev);
1173 struct aer_err_source e_src;
1174
1175 if (kfifo_is_empty(&rpc->aer_fifo))
1176 return IRQ_NONE;
1177
1178 while (kfifo_get(&rpc->aer_fifo, &e_src))
1179 aer_isr_one_error(rpc, &e_src);
1180 return IRQ_HANDLED;
1181 }
1182
1183 /**
1184 * aer_irq - Root Port's ISR
1185 * @irq: IRQ assigned to Root Port
1186 * @context: pointer to Root Port data structure
1187 *
1188 * Invoked when Root Port detects AER messages.
1189 */
aer_irq(int irq,void * context)1190 static irqreturn_t aer_irq(int irq, void *context)
1191 {
1192 struct pcie_device *pdev = (struct pcie_device *)context;
1193 struct aer_rpc *rpc = get_service_data(pdev);
1194 struct pci_dev *rp = rpc->rpd;
1195 int aer = rp->aer_cap;
1196 struct aer_err_source e_src = {};
1197
1198 pci_read_config_dword(rp, aer + PCI_ERR_ROOT_STATUS, &e_src.status);
1199 if (!(e_src.status & (PCI_ERR_ROOT_UNCOR_RCV|PCI_ERR_ROOT_COR_RCV)))
1200 return IRQ_NONE;
1201
1202 pci_read_config_dword(rp, aer + PCI_ERR_ROOT_ERR_SRC, &e_src.id);
1203 pci_write_config_dword(rp, aer + PCI_ERR_ROOT_STATUS, e_src.status);
1204
1205 if (!kfifo_put(&rpc->aer_fifo, e_src))
1206 return IRQ_HANDLED;
1207
1208 return IRQ_WAKE_THREAD;
1209 }
1210
set_device_error_reporting(struct pci_dev * dev,void * data)1211 static int set_device_error_reporting(struct pci_dev *dev, void *data)
1212 {
1213 bool enable = *((bool *)data);
1214 int type = pci_pcie_type(dev);
1215
1216 if ((type == PCI_EXP_TYPE_ROOT_PORT) ||
1217 (type == PCI_EXP_TYPE_RC_EC) ||
1218 (type == PCI_EXP_TYPE_UPSTREAM) ||
1219 (type == PCI_EXP_TYPE_DOWNSTREAM)) {
1220 if (enable)
1221 pci_enable_pcie_error_reporting(dev);
1222 else
1223 pci_disable_pcie_error_reporting(dev);
1224 }
1225
1226 if (enable)
1227 pcie_set_ecrc_checking(dev);
1228
1229 return 0;
1230 }
1231
1232 /**
1233 * set_downstream_devices_error_reporting - enable/disable the error reporting bits on the root port and its downstream ports.
1234 * @dev: pointer to root port's pci_dev data structure
1235 * @enable: true = enable error reporting, false = disable error reporting.
1236 */
set_downstream_devices_error_reporting(struct pci_dev * dev,bool enable)1237 static void set_downstream_devices_error_reporting(struct pci_dev *dev,
1238 bool enable)
1239 {
1240 set_device_error_reporting(dev, &enable);
1241
1242 if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_EC)
1243 pcie_walk_rcec(dev, set_device_error_reporting, &enable);
1244 else if (dev->subordinate)
1245 pci_walk_bus(dev->subordinate, set_device_error_reporting,
1246 &enable);
1247
1248 }
1249
1250 /**
1251 * aer_enable_rootport - enable Root Port's interrupts when receiving messages
1252 * @rpc: pointer to a Root Port data structure
1253 *
1254 * Invoked when PCIe bus loads AER service driver.
1255 */
aer_enable_rootport(struct aer_rpc * rpc)1256 static void aer_enable_rootport(struct aer_rpc *rpc)
1257 {
1258 struct pci_dev *pdev = rpc->rpd;
1259 int aer = pdev->aer_cap;
1260 u16 reg16;
1261 u32 reg32;
1262
1263 /* Clear PCIe Capability's Device Status */
1264 pcie_capability_read_word(pdev, PCI_EXP_DEVSTA, ®16);
1265 pcie_capability_write_word(pdev, PCI_EXP_DEVSTA, reg16);
1266
1267 /* Disable system error generation in response to error messages */
1268 pcie_capability_clear_word(pdev, PCI_EXP_RTCTL,
1269 SYSTEM_ERROR_INTR_ON_MESG_MASK);
1270
1271 /* Clear error status */
1272 pci_read_config_dword(pdev, aer + PCI_ERR_ROOT_STATUS, ®32);
1273 pci_write_config_dword(pdev, aer + PCI_ERR_ROOT_STATUS, reg32);
1274 pci_read_config_dword(pdev, aer + PCI_ERR_COR_STATUS, ®32);
1275 pci_write_config_dword(pdev, aer + PCI_ERR_COR_STATUS, reg32);
1276 pci_read_config_dword(pdev, aer + PCI_ERR_UNCOR_STATUS, ®32);
1277 pci_write_config_dword(pdev, aer + PCI_ERR_UNCOR_STATUS, reg32);
1278
1279 /*
1280 * Enable error reporting for the root port device and downstream port
1281 * devices.
1282 */
1283 set_downstream_devices_error_reporting(pdev, true);
1284
1285 /* Enable Root Port's interrupt in response to error messages */
1286 pci_read_config_dword(pdev, aer + PCI_ERR_ROOT_COMMAND, ®32);
1287 reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
1288 pci_write_config_dword(pdev, aer + PCI_ERR_ROOT_COMMAND, reg32);
1289 }
1290
1291 /**
1292 * aer_disable_rootport - disable Root Port's interrupts when receiving messages
1293 * @rpc: pointer to a Root Port data structure
1294 *
1295 * Invoked when PCIe bus unloads AER service driver.
1296 */
aer_disable_rootport(struct aer_rpc * rpc)1297 static void aer_disable_rootport(struct aer_rpc *rpc)
1298 {
1299 struct pci_dev *pdev = rpc->rpd;
1300 int aer = pdev->aer_cap;
1301 u32 reg32;
1302
1303 /*
1304 * Disable error reporting for the root port device and downstream port
1305 * devices.
1306 */
1307 set_downstream_devices_error_reporting(pdev, false);
1308
1309 /* Disable Root's interrupt in response to error messages */
1310 pci_read_config_dword(pdev, aer + PCI_ERR_ROOT_COMMAND, ®32);
1311 reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
1312 pci_write_config_dword(pdev, aer + PCI_ERR_ROOT_COMMAND, reg32);
1313
1314 /* Clear Root's error status reg */
1315 pci_read_config_dword(pdev, aer + PCI_ERR_ROOT_STATUS, ®32);
1316 pci_write_config_dword(pdev, aer + PCI_ERR_ROOT_STATUS, reg32);
1317 }
1318
1319 /**
1320 * aer_remove - clean up resources
1321 * @dev: pointer to the pcie_dev data structure
1322 *
1323 * Invoked when PCI Express bus unloads or AER probe fails.
1324 */
aer_remove(struct pcie_device * dev)1325 static void aer_remove(struct pcie_device *dev)
1326 {
1327 struct aer_rpc *rpc = get_service_data(dev);
1328
1329 aer_disable_rootport(rpc);
1330 }
1331
1332 /**
1333 * aer_probe - initialize resources
1334 * @dev: pointer to the pcie_dev data structure
1335 *
1336 * Invoked when PCI Express bus loads AER service driver.
1337 */
aer_probe(struct pcie_device * dev)1338 static int aer_probe(struct pcie_device *dev)
1339 {
1340 int status;
1341 struct aer_rpc *rpc;
1342 struct device *device = &dev->device;
1343 struct pci_dev *port = dev->port;
1344
1345 /* Limit to Root Ports or Root Complex Event Collectors */
1346 if ((pci_pcie_type(port) != PCI_EXP_TYPE_RC_EC) &&
1347 (pci_pcie_type(port) != PCI_EXP_TYPE_ROOT_PORT))
1348 return -ENODEV;
1349
1350 rpc = devm_kzalloc(device, sizeof(struct aer_rpc), GFP_KERNEL);
1351 if (!rpc)
1352 return -ENOMEM;
1353
1354 rpc->rpd = port;
1355 INIT_KFIFO(rpc->aer_fifo);
1356 set_service_data(dev, rpc);
1357
1358 status = devm_request_threaded_irq(device, dev->irq, aer_irq, aer_isr,
1359 IRQF_SHARED, "aerdrv", dev);
1360 if (status) {
1361 pci_err(port, "request AER IRQ %d failed\n", dev->irq);
1362 return status;
1363 }
1364
1365 aer_enable_rootport(rpc);
1366 pci_info(port, "enabled with IRQ %d\n", dev->irq);
1367 return 0;
1368 }
1369
1370 /**
1371 * aer_root_reset - reset Root Port hierarchy, RCEC, or RCiEP
1372 * @dev: pointer to Root Port, RCEC, or RCiEP
1373 *
1374 * Invoked by Port Bus driver when performing reset.
1375 */
aer_root_reset(struct pci_dev * dev)1376 static pci_ers_result_t aer_root_reset(struct pci_dev *dev)
1377 {
1378 int type = pci_pcie_type(dev);
1379 struct pci_dev *root;
1380 int aer;
1381 struct pci_host_bridge *host = pci_find_host_bridge(dev->bus);
1382 u32 reg32;
1383 int rc;
1384
1385 /*
1386 * Only Root Ports and RCECs have AER Root Command and Root Status
1387 * registers. If "dev" is an RCiEP, the relevant registers are in
1388 * the RCEC.
1389 */
1390 if (type == PCI_EXP_TYPE_RC_END)
1391 root = dev->rcec;
1392 else
1393 root = pcie_find_root_port(dev);
1394
1395 /*
1396 * If the platform retained control of AER, an RCiEP may not have
1397 * an RCEC visible to us, so dev->rcec ("root") may be NULL. In
1398 * that case, firmware is responsible for these registers.
1399 */
1400 aer = root ? root->aer_cap : 0;
1401
1402 if ((host->native_aer || pcie_ports_native) && aer) {
1403 /* Disable Root's interrupt in response to error messages */
1404 pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, ®32);
1405 reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
1406 pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, reg32);
1407 }
1408
1409 if (type == PCI_EXP_TYPE_RC_EC || type == PCI_EXP_TYPE_RC_END) {
1410 rc = pcie_reset_flr(dev, PCI_RESET_DO_RESET);
1411 if (!rc)
1412 pci_info(dev, "has been reset\n");
1413 else
1414 pci_info(dev, "not reset (no FLR support: %d)\n", rc);
1415 } else {
1416 rc = pci_bus_error_reset(dev);
1417 pci_info(dev, "%s Port link has been reset (%d)\n",
1418 pci_is_root_bus(dev->bus) ? "Root" : "Downstream", rc);
1419 }
1420
1421 if ((host->native_aer || pcie_ports_native) && aer) {
1422 /* Clear Root Error Status */
1423 pci_read_config_dword(root, aer + PCI_ERR_ROOT_STATUS, ®32);
1424 pci_write_config_dword(root, aer + PCI_ERR_ROOT_STATUS, reg32);
1425
1426 /* Enable Root Port's interrupt in response to error messages */
1427 pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, ®32);
1428 reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
1429 pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, reg32);
1430 }
1431
1432 return rc ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
1433 }
1434
1435 static struct pcie_port_service_driver aerdriver = {
1436 .name = "aer",
1437 .port_type = PCIE_ANY_PORT,
1438 .service = PCIE_PORT_SERVICE_AER,
1439
1440 .probe = aer_probe,
1441 .remove = aer_remove,
1442 };
1443
1444 /**
1445 * pcie_aer_init - register AER root service driver
1446 *
1447 * Invoked when AER root service driver is loaded.
1448 */
pcie_aer_init(void)1449 int __init pcie_aer_init(void)
1450 {
1451 if (!pci_aer_available())
1452 return -ENXIO;
1453 return pcie_port_service_register(&aerdriver);
1454 }
1455