Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * APEI Generic Hardware Error Source support
4 *
5 * Generic Hardware Error Source provides a way to report platform
6 * hardware errors (such as that from chipset). It works in so called
7 * "Firmware First" mode, that is, hardware errors are reported to
8 * firmware firstly, then reported to Linux by firmware. This way,
9 * some non-standard hardware error registers or non-standard hardware
10 * link can be checked by firmware to produce more hardware error
11 * information for Linux.
12 *
13 * For more information about Generic Hardware Error Source, please
14 * refer to ACPI Specification version 4.0, section 17.3.2.6
15 *
16 * Copyright 2010,2011 Intel Corp.
17 * Author: Huang Ying <ying.huang@intel.com>
18 */
19
20#include <linux/arm_sdei.h>
21#include <linux/kernel.h>
22#include <linux/moduleparam.h>
23#include <linux/init.h>
24#include <linux/acpi.h>
25#include <linux/io.h>
26#include <linux/interrupt.h>
27#include <linux/timer.h>
28#include <linux/cper.h>
29#include <linux/cleanup.h>
30#include <linux/platform_device.h>
31#include <linux/mutex.h>
32#include <linux/ratelimit.h>
33#include <linux/vmalloc.h>
34#include <linux/irq_work.h>
35#include <linux/llist.h>
36#include <linux/genalloc.h>
37#include <linux/kfifo.h>
38#include <linux/pci.h>
39#include <linux/pfn.h>
40#include <linux/aer.h>
41#include <linux/nmi.h>
42#include <linux/sched/clock.h>
43#include <linux/uuid.h>
44#include <linux/ras.h>
45#include <linux/task_work.h>
46
47#include <acpi/actbl1.h>
48#include <acpi/ghes.h>
49#include <acpi/apei.h>
50#include <asm/fixmap.h>
51#include <asm/tlbflush.h>
52#include <cxl/event.h>
53#include <ras/ras_event.h>
54
55#include "apei-internal.h"
56
57#define GHES_PFX "GHES: "
58
59#define GHES_ESTATUS_MAX_SIZE 65536
60#define GHES_ESOURCE_PREALLOC_MAX_SIZE 65536
61
62#define GHES_ESTATUS_POOL_MIN_ALLOC_ORDER 3
63
64/* This is just an estimation for memory pool allocation */
65#define GHES_ESTATUS_CACHE_AVG_SIZE 512
66
67#define GHES_ESTATUS_CACHES_SIZE 4
68
69#define GHES_ESTATUS_IN_CACHE_MAX_NSEC 10000000000ULL
70/* Prevent too many caches are allocated because of RCU */
71#define GHES_ESTATUS_CACHE_ALLOCED_MAX (GHES_ESTATUS_CACHES_SIZE * 3 / 2)
72
73#define GHES_ESTATUS_CACHE_LEN(estatus_len) \
74 (sizeof(struct ghes_estatus_cache) + (estatus_len))
75#define GHES_ESTATUS_FROM_CACHE(estatus_cache) \
76 ((struct acpi_hest_generic_status *) \
77 ((struct ghes_estatus_cache *)(estatus_cache) + 1))
78
79#define GHES_ESTATUS_NODE_LEN(estatus_len) \
80 (sizeof(struct ghes_estatus_node) + (estatus_len))
81#define GHES_ESTATUS_FROM_NODE(estatus_node) \
82 ((struct acpi_hest_generic_status *) \
83 ((struct ghes_estatus_node *)(estatus_node) + 1))
84
85#define GHES_VENDOR_ENTRY_LEN(gdata_len) \
86 (sizeof(struct ghes_vendor_record_entry) + (gdata_len))
87#define GHES_GDATA_FROM_VENDOR_ENTRY(vendor_entry) \
88 ((struct acpi_hest_generic_data *) \
89 ((struct ghes_vendor_record_entry *)(vendor_entry) + 1))
90
91/*
92 * NMI-like notifications vary by architecture, before the compiler can prune
93 * unused static functions it needs a value for these enums.
94 */
95#ifndef CONFIG_ARM_SDE_INTERFACE
96#define FIX_APEI_GHES_SDEI_NORMAL __end_of_fixed_addresses
97#define FIX_APEI_GHES_SDEI_CRITICAL __end_of_fixed_addresses
98#endif
99
100static ATOMIC_NOTIFIER_HEAD(ghes_report_chain);
101
102static inline bool is_hest_type_generic_v2(struct ghes *ghes)
103{
104 return ghes->generic->header.type == ACPI_HEST_TYPE_GENERIC_ERROR_V2;
105}
106
107/*
108 * A platform may describe one error source for the handling of synchronous
109 * errors (e.g. MCE or SEA), or for handling asynchronous errors (e.g. SCI
110 * or External Interrupt). On x86, the HEST notifications are always
111 * asynchronous, so only SEA on ARM is delivered as a synchronous
112 * notification.
113 */
114static inline bool is_hest_sync_notify(struct ghes *ghes)
115{
116 u8 notify_type = ghes->generic->notify.type;
117
118 return notify_type == ACPI_HEST_NOTIFY_SEA;
119}
120
121/*
122 * This driver isn't really modular, however for the time being,
123 * continuing to use module_param is the easiest way to remain
124 * compatible with existing boot arg use cases.
125 */
126bool ghes_disable;
127module_param_named(disable, ghes_disable, bool, 0);
128
129/*
130 * "ghes.edac_force_enable" forcibly enables ghes_edac and skips the platform
131 * check.
132 */
133static bool ghes_edac_force_enable;
134module_param_named(edac_force_enable, ghes_edac_force_enable, bool, 0);
135
136/*
137 * All error sources notified with HED (Hardware Error Device) share a
138 * single notifier callback, so they need to be linked and checked one
139 * by one. This holds true for NMI too.
140 *
141 * RCU is used for these lists, so ghes_list_mutex is only used for
142 * list changing, not for traversing.
143 */
144static LIST_HEAD(ghes_hed);
145static DEFINE_MUTEX(ghes_list_mutex);
146
147/*
148 * A list of GHES devices which are given to the corresponding EDAC driver
149 * ghes_edac for further use.
150 */
151static LIST_HEAD(ghes_devs);
152static DEFINE_MUTEX(ghes_devs_mutex);
153
154/*
155 * Because the memory area used to transfer hardware error information
156 * from BIOS to Linux can be determined only in NMI, IRQ or timer
157 * handler, but general ioremap can not be used in atomic context, so
158 * the fixmap is used instead.
159 *
160 * This spinlock is used to prevent the fixmap entry from being used
161 * simultaneously.
162 */
163static DEFINE_SPINLOCK(ghes_notify_lock_irq);
164
165struct ghes_vendor_record_entry {
166 struct work_struct work;
167 int error_severity;
168 char vendor_record[];
169};
170
171static struct gen_pool *ghes_estatus_pool;
172
173static struct ghes_estatus_cache __rcu *ghes_estatus_caches[GHES_ESTATUS_CACHES_SIZE];
174static atomic_t ghes_estatus_cache_alloced;
175
176static void __iomem *ghes_map(u64 pfn, enum fixed_addresses fixmap_idx)
177{
178 phys_addr_t paddr;
179 pgprot_t prot;
180
181 paddr = PFN_PHYS(pfn);
182 prot = arch_apei_get_mem_attribute(paddr);
183 __set_fixmap(fixmap_idx, paddr, prot);
184
185 return (void __iomem *) __fix_to_virt(fixmap_idx);
186}
187
188static void ghes_unmap(void __iomem *vaddr, enum fixed_addresses fixmap_idx)
189{
190 int _idx = virt_to_fix((unsigned long)vaddr);
191
192 WARN_ON_ONCE(fixmap_idx != _idx);
193 clear_fixmap(fixmap_idx);
194}
195
196int ghes_estatus_pool_init(unsigned int num_ghes)
197{
198 unsigned long addr, len;
199 int rc;
200
201 ghes_estatus_pool = gen_pool_create(GHES_ESTATUS_POOL_MIN_ALLOC_ORDER, -1);
202 if (!ghes_estatus_pool)
203 return -ENOMEM;
204
205 len = GHES_ESTATUS_CACHE_AVG_SIZE * GHES_ESTATUS_CACHE_ALLOCED_MAX;
206 len += (num_ghes * GHES_ESOURCE_PREALLOC_MAX_SIZE);
207
208 addr = (unsigned long)vmalloc(PAGE_ALIGN(len));
209 if (!addr)
210 goto err_pool_alloc;
211
212 rc = gen_pool_add(ghes_estatus_pool, addr, PAGE_ALIGN(len), -1);
213 if (rc)
214 goto err_pool_add;
215
216 return 0;
217
218err_pool_add:
219 vfree((void *)addr);
220
221err_pool_alloc:
222 gen_pool_destroy(ghes_estatus_pool);
223
224 return -ENOMEM;
225}
226
227/**
228 * ghes_estatus_pool_region_free - free previously allocated memory
229 * from the ghes_estatus_pool.
230 * @addr: address of memory to free.
231 * @size: size of memory to free.
232 *
233 * Returns none.
234 */
235void ghes_estatus_pool_region_free(unsigned long addr, u32 size)
236{
237 gen_pool_free(ghes_estatus_pool, addr, size);
238}
239EXPORT_SYMBOL_GPL(ghes_estatus_pool_region_free);
240
241static int map_gen_v2(struct ghes *ghes)
242{
243 return apei_map_generic_address(&ghes->generic_v2->read_ack_register);
244}
245
246static void unmap_gen_v2(struct ghes *ghes)
247{
248 apei_unmap_generic_address(&ghes->generic_v2->read_ack_register);
249}
250
251static void ghes_ack_error(struct acpi_hest_generic_v2 *gv2)
252{
253 int rc;
254 u64 val = 0;
255
256 rc = apei_read(&val, &gv2->read_ack_register);
257 if (rc)
258 return;
259
260 val &= gv2->read_ack_preserve << gv2->read_ack_register.bit_offset;
261 val |= gv2->read_ack_write << gv2->read_ack_register.bit_offset;
262
263 apei_write(val, &gv2->read_ack_register);
264}
265
266static struct ghes *ghes_new(struct acpi_hest_generic *generic)
267{
268 struct ghes *ghes;
269 unsigned int error_block_length;
270 int rc;
271
272 ghes = kzalloc(sizeof(*ghes), GFP_KERNEL);
273 if (!ghes)
274 return ERR_PTR(-ENOMEM);
275
276 ghes->generic = generic;
277 if (is_hest_type_generic_v2(ghes)) {
278 rc = map_gen_v2(ghes);
279 if (rc)
280 goto err_free;
281 }
282
283 rc = apei_map_generic_address(&generic->error_status_address);
284 if (rc)
285 goto err_unmap_read_ack_addr;
286 error_block_length = generic->error_block_length;
287 if (error_block_length > GHES_ESTATUS_MAX_SIZE) {
288 pr_warn(FW_WARN GHES_PFX
289 "Error status block length is too long: %u for "
290 "generic hardware error source: %d.\n",
291 error_block_length, generic->header.source_id);
292 error_block_length = GHES_ESTATUS_MAX_SIZE;
293 }
294 ghes->estatus = kmalloc(error_block_length, GFP_KERNEL);
295 if (!ghes->estatus) {
296 rc = -ENOMEM;
297 goto err_unmap_status_addr;
298 }
299
300 return ghes;
301
302err_unmap_status_addr:
303 apei_unmap_generic_address(&generic->error_status_address);
304err_unmap_read_ack_addr:
305 if (is_hest_type_generic_v2(ghes))
306 unmap_gen_v2(ghes);
307err_free:
308 kfree(ghes);
309 return ERR_PTR(rc);
310}
311
312static void ghes_fini(struct ghes *ghes)
313{
314 kfree(ghes->estatus);
315 apei_unmap_generic_address(&ghes->generic->error_status_address);
316 if (is_hest_type_generic_v2(ghes))
317 unmap_gen_v2(ghes);
318}
319
320static inline int ghes_severity(int severity)
321{
322 switch (severity) {
323 case CPER_SEV_INFORMATIONAL:
324 return GHES_SEV_NO;
325 case CPER_SEV_CORRECTED:
326 return GHES_SEV_CORRECTED;
327 case CPER_SEV_RECOVERABLE:
328 return GHES_SEV_RECOVERABLE;
329 case CPER_SEV_FATAL:
330 return GHES_SEV_PANIC;
331 default:
332 /* Unknown, go panic */
333 return GHES_SEV_PANIC;
334 }
335}
336
337static void ghes_copy_tofrom_phys(void *buffer, u64 paddr, u32 len,
338 int from_phys,
339 enum fixed_addresses fixmap_idx)
340{
341 void __iomem *vaddr;
342 u64 offset;
343 u32 trunk;
344
345 while (len > 0) {
346 offset = paddr - (paddr & PAGE_MASK);
347 vaddr = ghes_map(PHYS_PFN(paddr), fixmap_idx);
348 trunk = PAGE_SIZE - offset;
349 trunk = min(trunk, len);
350 if (from_phys)
351 memcpy_fromio(buffer, vaddr + offset, trunk);
352 else
353 memcpy_toio(vaddr + offset, buffer, trunk);
354 len -= trunk;
355 paddr += trunk;
356 buffer += trunk;
357 ghes_unmap(vaddr, fixmap_idx);
358 }
359}
360
361/* Check the top-level record header has an appropriate size. */
362static int __ghes_check_estatus(struct ghes *ghes,
363 struct acpi_hest_generic_status *estatus)
364{
365 u32 len = cper_estatus_len(estatus);
366
367 if (len < sizeof(*estatus)) {
368 pr_warn_ratelimited(FW_WARN GHES_PFX "Truncated error status block!\n");
369 return -EIO;
370 }
371
372 if (len > ghes->generic->error_block_length) {
373 pr_warn_ratelimited(FW_WARN GHES_PFX "Invalid error status block length!\n");
374 return -EIO;
375 }
376
377 if (cper_estatus_check_header(estatus)) {
378 pr_warn_ratelimited(FW_WARN GHES_PFX "Invalid CPER header!\n");
379 return -EIO;
380 }
381
382 return 0;
383}
384
385/* Read the CPER block, returning its address, and header in estatus. */
386static int __ghes_peek_estatus(struct ghes *ghes,
387 struct acpi_hest_generic_status *estatus,
388 u64 *buf_paddr, enum fixed_addresses fixmap_idx)
389{
390 struct acpi_hest_generic *g = ghes->generic;
391 int rc;
392
393 rc = apei_read(buf_paddr, &g->error_status_address);
394 if (rc) {
395 *buf_paddr = 0;
396 pr_warn_ratelimited(FW_WARN GHES_PFX
397"Failed to read error status block address for hardware error source: %d.\n",
398 g->header.source_id);
399 return -EIO;
400 }
401 if (!*buf_paddr)
402 return -ENOENT;
403
404 ghes_copy_tofrom_phys(estatus, *buf_paddr, sizeof(*estatus), 1,
405 fixmap_idx);
406 if (!estatus->block_status) {
407 *buf_paddr = 0;
408 return -ENOENT;
409 }
410
411 return 0;
412}
413
414static int __ghes_read_estatus(struct acpi_hest_generic_status *estatus,
415 u64 buf_paddr, enum fixed_addresses fixmap_idx,
416 size_t buf_len)
417{
418 ghes_copy_tofrom_phys(estatus, buf_paddr, buf_len, 1, fixmap_idx);
419 if (cper_estatus_check(estatus)) {
420 pr_warn_ratelimited(FW_WARN GHES_PFX
421 "Failed to read error status block!\n");
422 return -EIO;
423 }
424
425 return 0;
426}
427
428static int ghes_read_estatus(struct ghes *ghes,
429 struct acpi_hest_generic_status *estatus,
430 u64 *buf_paddr, enum fixed_addresses fixmap_idx)
431{
432 int rc;
433
434 rc = __ghes_peek_estatus(ghes, estatus, buf_paddr, fixmap_idx);
435 if (rc)
436 return rc;
437
438 rc = __ghes_check_estatus(ghes, estatus);
439 if (rc)
440 return rc;
441
442 return __ghes_read_estatus(estatus, *buf_paddr, fixmap_idx,
443 cper_estatus_len(estatus));
444}
445
446static void ghes_clear_estatus(struct ghes *ghes,
447 struct acpi_hest_generic_status *estatus,
448 u64 buf_paddr, enum fixed_addresses fixmap_idx)
449{
450 estatus->block_status = 0;
451
452 if (!buf_paddr)
453 return;
454
455 ghes_copy_tofrom_phys(estatus, buf_paddr,
456 sizeof(estatus->block_status), 0,
457 fixmap_idx);
458
459 /*
460 * GHESv2 type HEST entries introduce support for error acknowledgment,
461 * so only acknowledge the error if this support is present.
462 */
463 if (is_hest_type_generic_v2(ghes))
464 ghes_ack_error(ghes->generic_v2);
465}
466
467/*
468 * Called as task_work before returning to user-space.
469 * Ensure any queued work has been done before we return to the context that
470 * triggered the notification.
471 */
472static void ghes_kick_task_work(struct callback_head *head)
473{
474 struct acpi_hest_generic_status *estatus;
475 struct ghes_estatus_node *estatus_node;
476 u32 node_len;
477
478 estatus_node = container_of(head, struct ghes_estatus_node, task_work);
479 if (IS_ENABLED(CONFIG_ACPI_APEI_MEMORY_FAILURE))
480 memory_failure_queue_kick(estatus_node->task_work_cpu);
481
482 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
483 node_len = GHES_ESTATUS_NODE_LEN(cper_estatus_len(estatus));
484 gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node, node_len);
485}
486
487static bool ghes_do_memory_failure(u64 physical_addr, int flags)
488{
489 unsigned long pfn;
490
491 if (!IS_ENABLED(CONFIG_ACPI_APEI_MEMORY_FAILURE))
492 return false;
493
494 pfn = PHYS_PFN(physical_addr);
495 if (!pfn_valid(pfn) && !arch_is_platform_page(physical_addr)) {
496 pr_warn_ratelimited(FW_WARN GHES_PFX
497 "Invalid address in generic error data: %#llx\n",
498 physical_addr);
499 return false;
500 }
501
502 memory_failure_queue(pfn, flags);
503 return true;
504}
505
506static bool ghes_handle_memory_failure(struct acpi_hest_generic_data *gdata,
507 int sev, bool sync)
508{
509 int flags = -1;
510 int sec_sev = ghes_severity(gdata->error_severity);
511 struct cper_sec_mem_err *mem_err = acpi_hest_get_payload(gdata);
512
513 if (!(mem_err->validation_bits & CPER_MEM_VALID_PA))
514 return false;
515
516 /* iff following two events can be handled properly by now */
517 if (sec_sev == GHES_SEV_CORRECTED &&
518 (gdata->flags & CPER_SEC_ERROR_THRESHOLD_EXCEEDED))
519 flags = MF_SOFT_OFFLINE;
520 if (sev == GHES_SEV_RECOVERABLE && sec_sev == GHES_SEV_RECOVERABLE)
521 flags = sync ? MF_ACTION_REQUIRED : 0;
522
523 if (flags != -1)
524 return ghes_do_memory_failure(mem_err->physical_addr, flags);
525
526 return false;
527}
528
529static bool ghes_handle_arm_hw_error(struct acpi_hest_generic_data *gdata,
530 int sev, bool sync)
531{
532 struct cper_sec_proc_arm *err = acpi_hest_get_payload(gdata);
533 int flags = sync ? MF_ACTION_REQUIRED : 0;
534 bool queued = false;
535 int sec_sev, i;
536 char *p;
537
538 log_arm_hw_error(err);
539
540 sec_sev = ghes_severity(gdata->error_severity);
541 if (sev != GHES_SEV_RECOVERABLE || sec_sev != GHES_SEV_RECOVERABLE)
542 return false;
543
544 p = (char *)(err + 1);
545 for (i = 0; i < err->err_info_num; i++) {
546 struct cper_arm_err_info *err_info = (struct cper_arm_err_info *)p;
547 bool is_cache = (err_info->type == CPER_ARM_CACHE_ERROR);
548 bool has_pa = (err_info->validation_bits & CPER_ARM_INFO_VALID_PHYSICAL_ADDR);
549 const char *error_type = "unknown error";
550
551 /*
552 * The field (err_info->error_info & BIT(26)) is fixed to set to
553 * 1 in some old firmware of HiSilicon Kunpeng920. We assume that
554 * firmware won't mix corrected errors in an uncorrected section,
555 * and don't filter out 'corrected' error here.
556 */
557 if (is_cache && has_pa) {
558 queued = ghes_do_memory_failure(err_info->physical_fault_addr, flags);
559 p += err_info->length;
560 continue;
561 }
562
563 if (err_info->type < ARRAY_SIZE(cper_proc_error_type_strs))
564 error_type = cper_proc_error_type_strs[err_info->type];
565
566 pr_warn_ratelimited(FW_WARN GHES_PFX
567 "Unhandled processor error type: %s\n",
568 error_type);
569 p += err_info->length;
570 }
571
572 return queued;
573}
574
575/*
576 * PCIe AER errors need to be sent to the AER driver for reporting and
577 * recovery. The GHES severities map to the following AER severities and
578 * require the following handling:
579 *
580 * GHES_SEV_CORRECTABLE -> AER_CORRECTABLE
581 * These need to be reported by the AER driver but no recovery is
582 * necessary.
583 * GHES_SEV_RECOVERABLE -> AER_NONFATAL
584 * GHES_SEV_RECOVERABLE && CPER_SEC_RESET -> AER_FATAL
585 * These both need to be reported and recovered from by the AER driver.
586 * GHES_SEV_PANIC does not make it to this handling since the kernel must
587 * panic.
588 */
589static void ghes_handle_aer(struct acpi_hest_generic_data *gdata)
590{
591#ifdef CONFIG_ACPI_APEI_PCIEAER
592 struct cper_sec_pcie *pcie_err = acpi_hest_get_payload(gdata);
593
594 if (pcie_err->validation_bits & CPER_PCIE_VALID_DEVICE_ID &&
595 pcie_err->validation_bits & CPER_PCIE_VALID_AER_INFO) {
596 unsigned int devfn;
597 int aer_severity;
598 u8 *aer_info;
599
600 devfn = PCI_DEVFN(pcie_err->device_id.device,
601 pcie_err->device_id.function);
602 aer_severity = cper_severity_to_aer(gdata->error_severity);
603
604 /*
605 * If firmware reset the component to contain
606 * the error, we must reinitialize it before
607 * use, so treat it as a fatal AER error.
608 */
609 if (gdata->flags & CPER_SEC_RESET)
610 aer_severity = AER_FATAL;
611
612 aer_info = (void *)gen_pool_alloc(ghes_estatus_pool,
613 sizeof(struct aer_capability_regs));
614 if (!aer_info)
615 return;
616 memcpy(aer_info, pcie_err->aer_info, sizeof(struct aer_capability_regs));
617
618 aer_recover_queue(pcie_err->device_id.segment,
619 pcie_err->device_id.bus,
620 devfn, aer_severity,
621 (struct aer_capability_regs *)
622 aer_info);
623 }
624#endif
625}
626
627static BLOCKING_NOTIFIER_HEAD(vendor_record_notify_list);
628
629int ghes_register_vendor_record_notifier(struct notifier_block *nb)
630{
631 return blocking_notifier_chain_register(&vendor_record_notify_list, nb);
632}
633EXPORT_SYMBOL_GPL(ghes_register_vendor_record_notifier);
634
635void ghes_unregister_vendor_record_notifier(struct notifier_block *nb)
636{
637 blocking_notifier_chain_unregister(&vendor_record_notify_list, nb);
638}
639EXPORT_SYMBOL_GPL(ghes_unregister_vendor_record_notifier);
640
641static void ghes_vendor_record_work_func(struct work_struct *work)
642{
643 struct ghes_vendor_record_entry *entry;
644 struct acpi_hest_generic_data *gdata;
645 u32 len;
646
647 entry = container_of(work, struct ghes_vendor_record_entry, work);
648 gdata = GHES_GDATA_FROM_VENDOR_ENTRY(entry);
649
650 blocking_notifier_call_chain(&vendor_record_notify_list,
651 entry->error_severity, gdata);
652
653 len = GHES_VENDOR_ENTRY_LEN(acpi_hest_get_record_size(gdata));
654 gen_pool_free(ghes_estatus_pool, (unsigned long)entry, len);
655}
656
657static void ghes_defer_non_standard_event(struct acpi_hest_generic_data *gdata,
658 int sev)
659{
660 struct acpi_hest_generic_data *copied_gdata;
661 struct ghes_vendor_record_entry *entry;
662 u32 len;
663
664 len = GHES_VENDOR_ENTRY_LEN(acpi_hest_get_record_size(gdata));
665 entry = (void *)gen_pool_alloc(ghes_estatus_pool, len);
666 if (!entry)
667 return;
668
669 copied_gdata = GHES_GDATA_FROM_VENDOR_ENTRY(entry);
670 memcpy(copied_gdata, gdata, acpi_hest_get_record_size(gdata));
671 entry->error_severity = sev;
672
673 INIT_WORK(&entry->work, ghes_vendor_record_work_func);
674 schedule_work(&entry->work);
675}
676
677/* Room for 8 entries for each of the 4 event log queues */
678#define CXL_CPER_FIFO_DEPTH 32
679DEFINE_KFIFO(cxl_cper_fifo, struct cxl_cper_work_data, CXL_CPER_FIFO_DEPTH);
680
681/* Synchronize schedule_work() with cxl_cper_work changes */
682static DEFINE_SPINLOCK(cxl_cper_work_lock);
683struct work_struct *cxl_cper_work;
684
685static void cxl_cper_post_event(enum cxl_event_type event_type,
686 struct cxl_cper_event_rec *rec)
687{
688 struct cxl_cper_work_data wd;
689
690 if (rec->hdr.length <= sizeof(rec->hdr) ||
691 rec->hdr.length > sizeof(*rec)) {
692 pr_err(FW_WARN "CXL CPER Invalid section length (%u)\n",
693 rec->hdr.length);
694 return;
695 }
696
697 if (!(rec->hdr.validation_bits & CPER_CXL_COMP_EVENT_LOG_VALID)) {
698 pr_err(FW_WARN "CXL CPER invalid event\n");
699 return;
700 }
701
702 guard(spinlock_irqsave)(&cxl_cper_work_lock);
703
704 if (!cxl_cper_work)
705 return;
706
707 wd.event_type = event_type;
708 memcpy(&wd.rec, rec, sizeof(wd.rec));
709
710 if (!kfifo_put(&cxl_cper_fifo, wd)) {
711 pr_err_ratelimited("CXL CPER kfifo overflow\n");
712 return;
713 }
714
715 schedule_work(cxl_cper_work);
716}
717
718int cxl_cper_register_work(struct work_struct *work)
719{
720 if (cxl_cper_work)
721 return -EINVAL;
722
723 guard(spinlock)(&cxl_cper_work_lock);
724 cxl_cper_work = work;
725 return 0;
726}
727EXPORT_SYMBOL_NS_GPL(cxl_cper_register_work, "CXL");
728
729int cxl_cper_unregister_work(struct work_struct *work)
730{
731 if (cxl_cper_work != work)
732 return -EINVAL;
733
734 guard(spinlock)(&cxl_cper_work_lock);
735 cxl_cper_work = NULL;
736 return 0;
737}
738EXPORT_SYMBOL_NS_GPL(cxl_cper_unregister_work, "CXL");
739
740int cxl_cper_kfifo_get(struct cxl_cper_work_data *wd)
741{
742 return kfifo_get(&cxl_cper_fifo, wd);
743}
744EXPORT_SYMBOL_NS_GPL(cxl_cper_kfifo_get, "CXL");
745
746static bool ghes_do_proc(struct ghes *ghes,
747 const struct acpi_hest_generic_status *estatus)
748{
749 int sev, sec_sev;
750 struct acpi_hest_generic_data *gdata;
751 guid_t *sec_type;
752 const guid_t *fru_id = &guid_null;
753 char *fru_text = "";
754 bool queued = false;
755 bool sync = is_hest_sync_notify(ghes);
756
757 sev = ghes_severity(estatus->error_severity);
758 apei_estatus_for_each_section(estatus, gdata) {
759 sec_type = (guid_t *)gdata->section_type;
760 sec_sev = ghes_severity(gdata->error_severity);
761 if (gdata->validation_bits & CPER_SEC_VALID_FRU_ID)
762 fru_id = (guid_t *)gdata->fru_id;
763
764 if (gdata->validation_bits & CPER_SEC_VALID_FRU_TEXT)
765 fru_text = gdata->fru_text;
766
767 if (guid_equal(sec_type, &CPER_SEC_PLATFORM_MEM)) {
768 struct cper_sec_mem_err *mem_err = acpi_hest_get_payload(gdata);
769
770 atomic_notifier_call_chain(&ghes_report_chain, sev, mem_err);
771
772 arch_apei_report_mem_error(sev, mem_err);
773 queued = ghes_handle_memory_failure(gdata, sev, sync);
774 }
775 else if (guid_equal(sec_type, &CPER_SEC_PCIE)) {
776 ghes_handle_aer(gdata);
777 }
778 else if (guid_equal(sec_type, &CPER_SEC_PROC_ARM)) {
779 queued = ghes_handle_arm_hw_error(gdata, sev, sync);
780 } else if (guid_equal(sec_type, &CPER_SEC_CXL_GEN_MEDIA_GUID)) {
781 struct cxl_cper_event_rec *rec = acpi_hest_get_payload(gdata);
782
783 cxl_cper_post_event(CXL_CPER_EVENT_GEN_MEDIA, rec);
784 } else if (guid_equal(sec_type, &CPER_SEC_CXL_DRAM_GUID)) {
785 struct cxl_cper_event_rec *rec = acpi_hest_get_payload(gdata);
786
787 cxl_cper_post_event(CXL_CPER_EVENT_DRAM, rec);
788 } else if (guid_equal(sec_type, &CPER_SEC_CXL_MEM_MODULE_GUID)) {
789 struct cxl_cper_event_rec *rec = acpi_hest_get_payload(gdata);
790
791 cxl_cper_post_event(CXL_CPER_EVENT_MEM_MODULE, rec);
792 } else {
793 void *err = acpi_hest_get_payload(gdata);
794
795 ghes_defer_non_standard_event(gdata, sev);
796 log_non_standard_event(sec_type, fru_id, fru_text,
797 sec_sev, err,
798 gdata->error_data_length);
799 }
800 }
801
802 return queued;
803}
804
805static void __ghes_print_estatus(const char *pfx,
806 const struct acpi_hest_generic *generic,
807 const struct acpi_hest_generic_status *estatus)
808{
809 static atomic_t seqno;
810 unsigned int curr_seqno;
811 char pfx_seq[64];
812
813 if (pfx == NULL) {
814 if (ghes_severity(estatus->error_severity) <=
815 GHES_SEV_CORRECTED)
816 pfx = KERN_WARNING;
817 else
818 pfx = KERN_ERR;
819 }
820 curr_seqno = atomic_inc_return(&seqno);
821 snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}" HW_ERR, pfx, curr_seqno);
822 printk("%s""Hardware error from APEI Generic Hardware Error Source: %d\n",
823 pfx_seq, generic->header.source_id);
824 cper_estatus_print(pfx_seq, estatus);
825}
826
827static int ghes_print_estatus(const char *pfx,
828 const struct acpi_hest_generic *generic,
829 const struct acpi_hest_generic_status *estatus)
830{
831 /* Not more than 2 messages every 5 seconds */
832 static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2);
833 static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2);
834 struct ratelimit_state *ratelimit;
835
836 if (ghes_severity(estatus->error_severity) <= GHES_SEV_CORRECTED)
837 ratelimit = &ratelimit_corrected;
838 else
839 ratelimit = &ratelimit_uncorrected;
840 if (__ratelimit(ratelimit)) {
841 __ghes_print_estatus(pfx, generic, estatus);
842 return 1;
843 }
844 return 0;
845}
846
847/*
848 * GHES error status reporting throttle, to report more kinds of
849 * errors, instead of just most frequently occurred errors.
850 */
851static int ghes_estatus_cached(struct acpi_hest_generic_status *estatus)
852{
853 u32 len;
854 int i, cached = 0;
855 unsigned long long now;
856 struct ghes_estatus_cache *cache;
857 struct acpi_hest_generic_status *cache_estatus;
858
859 len = cper_estatus_len(estatus);
860 rcu_read_lock();
861 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
862 cache = rcu_dereference(ghes_estatus_caches[i]);
863 if (cache == NULL)
864 continue;
865 if (len != cache->estatus_len)
866 continue;
867 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
868 if (memcmp(estatus, cache_estatus, len))
869 continue;
870 atomic_inc(&cache->count);
871 now = sched_clock();
872 if (now - cache->time_in < GHES_ESTATUS_IN_CACHE_MAX_NSEC)
873 cached = 1;
874 break;
875 }
876 rcu_read_unlock();
877 return cached;
878}
879
880static struct ghes_estatus_cache *ghes_estatus_cache_alloc(
881 struct acpi_hest_generic *generic,
882 struct acpi_hest_generic_status *estatus)
883{
884 int alloced;
885 u32 len, cache_len;
886 struct ghes_estatus_cache *cache;
887 struct acpi_hest_generic_status *cache_estatus;
888
889 alloced = atomic_add_return(1, &ghes_estatus_cache_alloced);
890 if (alloced > GHES_ESTATUS_CACHE_ALLOCED_MAX) {
891 atomic_dec(&ghes_estatus_cache_alloced);
892 return NULL;
893 }
894 len = cper_estatus_len(estatus);
895 cache_len = GHES_ESTATUS_CACHE_LEN(len);
896 cache = (void *)gen_pool_alloc(ghes_estatus_pool, cache_len);
897 if (!cache) {
898 atomic_dec(&ghes_estatus_cache_alloced);
899 return NULL;
900 }
901 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
902 memcpy(cache_estatus, estatus, len);
903 cache->estatus_len = len;
904 atomic_set(&cache->count, 0);
905 cache->generic = generic;
906 cache->time_in = sched_clock();
907 return cache;
908}
909
910static void ghes_estatus_cache_rcu_free(struct rcu_head *head)
911{
912 struct ghes_estatus_cache *cache;
913 u32 len;
914
915 cache = container_of(head, struct ghes_estatus_cache, rcu);
916 len = cper_estatus_len(GHES_ESTATUS_FROM_CACHE(cache));
917 len = GHES_ESTATUS_CACHE_LEN(len);
918 gen_pool_free(ghes_estatus_pool, (unsigned long)cache, len);
919 atomic_dec(&ghes_estatus_cache_alloced);
920}
921
922static void
923ghes_estatus_cache_add(struct acpi_hest_generic *generic,
924 struct acpi_hest_generic_status *estatus)
925{
926 unsigned long long now, duration, period, max_period = 0;
927 struct ghes_estatus_cache *cache, *new_cache;
928 struct ghes_estatus_cache __rcu *victim;
929 int i, slot = -1, count;
930
931 new_cache = ghes_estatus_cache_alloc(generic, estatus);
932 if (!new_cache)
933 return;
934
935 rcu_read_lock();
936 now = sched_clock();
937 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
938 cache = rcu_dereference(ghes_estatus_caches[i]);
939 if (cache == NULL) {
940 slot = i;
941 break;
942 }
943 duration = now - cache->time_in;
944 if (duration >= GHES_ESTATUS_IN_CACHE_MAX_NSEC) {
945 slot = i;
946 break;
947 }
948 count = atomic_read(&cache->count);
949 period = duration;
950 do_div(period, (count + 1));
951 if (period > max_period) {
952 max_period = period;
953 slot = i;
954 }
955 }
956 rcu_read_unlock();
957
958 if (slot != -1) {
959 /*
960 * Use release semantics to ensure that ghes_estatus_cached()
961 * running on another CPU will see the updated cache fields if
962 * it can see the new value of the pointer.
963 */
964 victim = xchg_release(&ghes_estatus_caches[slot],
965 RCU_INITIALIZER(new_cache));
966
967 /*
968 * At this point, victim may point to a cached item different
969 * from the one based on which we selected the slot. Instead of
970 * going to the loop again to pick another slot, let's just
971 * drop the other item anyway: this may cause a false cache
972 * miss later on, but that won't cause any problems.
973 */
974 if (victim)
975 call_rcu(&unrcu_pointer(victim)->rcu,
976 ghes_estatus_cache_rcu_free);
977 }
978}
979
980static void __ghes_panic(struct ghes *ghes,
981 struct acpi_hest_generic_status *estatus,
982 u64 buf_paddr, enum fixed_addresses fixmap_idx)
983{
984 const char *msg = GHES_PFX "Fatal hardware error";
985
986 __ghes_print_estatus(KERN_EMERG, ghes->generic, estatus);
987
988 ghes_clear_estatus(ghes, estatus, buf_paddr, fixmap_idx);
989
990 if (!panic_timeout)
991 pr_emerg("%s but panic disabled\n", msg);
992
993 panic(msg);
994}
995
996static int ghes_proc(struct ghes *ghes)
997{
998 struct acpi_hest_generic_status *estatus = ghes->estatus;
999 u64 buf_paddr;
1000 int rc;
1001
1002 rc = ghes_read_estatus(ghes, estatus, &buf_paddr, FIX_APEI_GHES_IRQ);
1003 if (rc)
1004 goto out;
1005
1006 if (ghes_severity(estatus->error_severity) >= GHES_SEV_PANIC)
1007 __ghes_panic(ghes, estatus, buf_paddr, FIX_APEI_GHES_IRQ);
1008
1009 if (!ghes_estatus_cached(estatus)) {
1010 if (ghes_print_estatus(NULL, ghes->generic, estatus))
1011 ghes_estatus_cache_add(ghes->generic, estatus);
1012 }
1013 ghes_do_proc(ghes, estatus);
1014
1015out:
1016 ghes_clear_estatus(ghes, estatus, buf_paddr, FIX_APEI_GHES_IRQ);
1017
1018 return rc;
1019}
1020
1021static void ghes_add_timer(struct ghes *ghes)
1022{
1023 struct acpi_hest_generic *g = ghes->generic;
1024 unsigned long expire;
1025
1026 if (!g->notify.poll_interval) {
1027 pr_warn(FW_WARN GHES_PFX "Poll interval is 0 for generic hardware error source: %d, disabled.\n",
1028 g->header.source_id);
1029 return;
1030 }
1031 expire = jiffies + msecs_to_jiffies(g->notify.poll_interval);
1032 ghes->timer.expires = round_jiffies_relative(expire);
1033 add_timer(&ghes->timer);
1034}
1035
1036static void ghes_poll_func(struct timer_list *t)
1037{
1038 struct ghes *ghes = from_timer(ghes, t, timer);
1039 unsigned long flags;
1040
1041 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
1042 ghes_proc(ghes);
1043 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
1044 if (!(ghes->flags & GHES_EXITING))
1045 ghes_add_timer(ghes);
1046}
1047
1048static irqreturn_t ghes_irq_func(int irq, void *data)
1049{
1050 struct ghes *ghes = data;
1051 unsigned long flags;
1052 int rc;
1053
1054 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
1055 rc = ghes_proc(ghes);
1056 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
1057 if (rc)
1058 return IRQ_NONE;
1059
1060 return IRQ_HANDLED;
1061}
1062
1063static int ghes_notify_hed(struct notifier_block *this, unsigned long event,
1064 void *data)
1065{
1066 struct ghes *ghes;
1067 unsigned long flags;
1068 int ret = NOTIFY_DONE;
1069
1070 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
1071 rcu_read_lock();
1072 list_for_each_entry_rcu(ghes, &ghes_hed, list) {
1073 if (!ghes_proc(ghes))
1074 ret = NOTIFY_OK;
1075 }
1076 rcu_read_unlock();
1077 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
1078
1079 return ret;
1080}
1081
1082static struct notifier_block ghes_notifier_hed = {
1083 .notifier_call = ghes_notify_hed,
1084};
1085
1086/*
1087 * Handlers for CPER records may not be NMI safe. For example,
1088 * memory_failure_queue() takes spinlocks and calls schedule_work_on().
1089 * In any NMI-like handler, memory from ghes_estatus_pool is used to save
1090 * estatus, and added to the ghes_estatus_llist. irq_work_queue() causes
1091 * ghes_proc_in_irq() to run in IRQ context where each estatus in
1092 * ghes_estatus_llist is processed.
1093 *
1094 * Memory from the ghes_estatus_pool is also used with the ghes_estatus_cache
1095 * to suppress frequent messages.
1096 */
1097static struct llist_head ghes_estatus_llist;
1098static struct irq_work ghes_proc_irq_work;
1099
1100static void ghes_proc_in_irq(struct irq_work *irq_work)
1101{
1102 struct llist_node *llnode, *next;
1103 struct ghes_estatus_node *estatus_node;
1104 struct acpi_hest_generic *generic;
1105 struct acpi_hest_generic_status *estatus;
1106 bool task_work_pending;
1107 u32 len, node_len;
1108 int ret;
1109
1110 llnode = llist_del_all(&ghes_estatus_llist);
1111 /*
1112 * Because the time order of estatus in list is reversed,
1113 * revert it back to proper order.
1114 */
1115 llnode = llist_reverse_order(llnode);
1116 while (llnode) {
1117 next = llnode->next;
1118 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
1119 llnode);
1120 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
1121 len = cper_estatus_len(estatus);
1122 node_len = GHES_ESTATUS_NODE_LEN(len);
1123 task_work_pending = ghes_do_proc(estatus_node->ghes, estatus);
1124 if (!ghes_estatus_cached(estatus)) {
1125 generic = estatus_node->generic;
1126 if (ghes_print_estatus(NULL, generic, estatus))
1127 ghes_estatus_cache_add(generic, estatus);
1128 }
1129
1130 if (task_work_pending && current->mm) {
1131 estatus_node->task_work.func = ghes_kick_task_work;
1132 estatus_node->task_work_cpu = smp_processor_id();
1133 ret = task_work_add(current, &estatus_node->task_work,
1134 TWA_RESUME);
1135 if (ret)
1136 estatus_node->task_work.func = NULL;
1137 }
1138
1139 if (!estatus_node->task_work.func)
1140 gen_pool_free(ghes_estatus_pool,
1141 (unsigned long)estatus_node, node_len);
1142
1143 llnode = next;
1144 }
1145}
1146
1147static void ghes_print_queued_estatus(void)
1148{
1149 struct llist_node *llnode;
1150 struct ghes_estatus_node *estatus_node;
1151 struct acpi_hest_generic *generic;
1152 struct acpi_hest_generic_status *estatus;
1153
1154 llnode = llist_del_all(&ghes_estatus_llist);
1155 /*
1156 * Because the time order of estatus in list is reversed,
1157 * revert it back to proper order.
1158 */
1159 llnode = llist_reverse_order(llnode);
1160 while (llnode) {
1161 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
1162 llnode);
1163 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
1164 generic = estatus_node->generic;
1165 ghes_print_estatus(NULL, generic, estatus);
1166 llnode = llnode->next;
1167 }
1168}
1169
1170static int ghes_in_nmi_queue_one_entry(struct ghes *ghes,
1171 enum fixed_addresses fixmap_idx)
1172{
1173 struct acpi_hest_generic_status *estatus, tmp_header;
1174 struct ghes_estatus_node *estatus_node;
1175 u32 len, node_len;
1176 u64 buf_paddr;
1177 int sev, rc;
1178
1179 if (!IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG))
1180 return -EOPNOTSUPP;
1181
1182 rc = __ghes_peek_estatus(ghes, &tmp_header, &buf_paddr, fixmap_idx);
1183 if (rc) {
1184 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1185 return rc;
1186 }
1187
1188 rc = __ghes_check_estatus(ghes, &tmp_header);
1189 if (rc) {
1190 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1191 return rc;
1192 }
1193
1194 len = cper_estatus_len(&tmp_header);
1195 node_len = GHES_ESTATUS_NODE_LEN(len);
1196 estatus_node = (void *)gen_pool_alloc(ghes_estatus_pool, node_len);
1197 if (!estatus_node)
1198 return -ENOMEM;
1199
1200 estatus_node->ghes = ghes;
1201 estatus_node->generic = ghes->generic;
1202 estatus_node->task_work.func = NULL;
1203 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
1204
1205 if (__ghes_read_estatus(estatus, buf_paddr, fixmap_idx, len)) {
1206 ghes_clear_estatus(ghes, estatus, buf_paddr, fixmap_idx);
1207 rc = -ENOENT;
1208 goto no_work;
1209 }
1210
1211 sev = ghes_severity(estatus->error_severity);
1212 if (sev >= GHES_SEV_PANIC) {
1213 ghes_print_queued_estatus();
1214 __ghes_panic(ghes, estatus, buf_paddr, fixmap_idx);
1215 }
1216
1217 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1218
1219 /* This error has been reported before, don't process it again. */
1220 if (ghes_estatus_cached(estatus))
1221 goto no_work;
1222
1223 llist_add(&estatus_node->llnode, &ghes_estatus_llist);
1224
1225 return rc;
1226
1227no_work:
1228 gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node,
1229 node_len);
1230
1231 return rc;
1232}
1233
1234static int ghes_in_nmi_spool_from_list(struct list_head *rcu_list,
1235 enum fixed_addresses fixmap_idx)
1236{
1237 int ret = -ENOENT;
1238 struct ghes *ghes;
1239
1240 rcu_read_lock();
1241 list_for_each_entry_rcu(ghes, rcu_list, list) {
1242 if (!ghes_in_nmi_queue_one_entry(ghes, fixmap_idx))
1243 ret = 0;
1244 }
1245 rcu_read_unlock();
1246
1247 if (IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG) && !ret)
1248 irq_work_queue(&ghes_proc_irq_work);
1249
1250 return ret;
1251}
1252
1253#ifdef CONFIG_ACPI_APEI_SEA
1254static LIST_HEAD(ghes_sea);
1255
1256/*
1257 * Return 0 only if one of the SEA error sources successfully reported an error
1258 * record sent from the firmware.
1259 */
1260int ghes_notify_sea(void)
1261{
1262 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sea);
1263 int rv;
1264
1265 raw_spin_lock(&ghes_notify_lock_sea);
1266 rv = ghes_in_nmi_spool_from_list(&ghes_sea, FIX_APEI_GHES_SEA);
1267 raw_spin_unlock(&ghes_notify_lock_sea);
1268
1269 return rv;
1270}
1271
1272static void ghes_sea_add(struct ghes *ghes)
1273{
1274 mutex_lock(&ghes_list_mutex);
1275 list_add_rcu(&ghes->list, &ghes_sea);
1276 mutex_unlock(&ghes_list_mutex);
1277}
1278
1279static void ghes_sea_remove(struct ghes *ghes)
1280{
1281 mutex_lock(&ghes_list_mutex);
1282 list_del_rcu(&ghes->list);
1283 mutex_unlock(&ghes_list_mutex);
1284 synchronize_rcu();
1285}
1286#else /* CONFIG_ACPI_APEI_SEA */
1287static inline void ghes_sea_add(struct ghes *ghes) { }
1288static inline void ghes_sea_remove(struct ghes *ghes) { }
1289#endif /* CONFIG_ACPI_APEI_SEA */
1290
1291#ifdef CONFIG_HAVE_ACPI_APEI_NMI
1292/*
1293 * NMI may be triggered on any CPU, so ghes_in_nmi is used for
1294 * having only one concurrent reader.
1295 */
1296static atomic_t ghes_in_nmi = ATOMIC_INIT(0);
1297
1298static LIST_HEAD(ghes_nmi);
1299
1300static int ghes_notify_nmi(unsigned int cmd, struct pt_regs *regs)
1301{
1302 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_nmi);
1303 int ret = NMI_DONE;
1304
1305 if (!atomic_add_unless(&ghes_in_nmi, 1, 1))
1306 return ret;
1307
1308 raw_spin_lock(&ghes_notify_lock_nmi);
1309 if (!ghes_in_nmi_spool_from_list(&ghes_nmi, FIX_APEI_GHES_NMI))
1310 ret = NMI_HANDLED;
1311 raw_spin_unlock(&ghes_notify_lock_nmi);
1312
1313 atomic_dec(&ghes_in_nmi);
1314 return ret;
1315}
1316
1317static void ghes_nmi_add(struct ghes *ghes)
1318{
1319 mutex_lock(&ghes_list_mutex);
1320 if (list_empty(&ghes_nmi))
1321 register_nmi_handler(NMI_LOCAL, ghes_notify_nmi, 0, "ghes");
1322 list_add_rcu(&ghes->list, &ghes_nmi);
1323 mutex_unlock(&ghes_list_mutex);
1324}
1325
1326static void ghes_nmi_remove(struct ghes *ghes)
1327{
1328 mutex_lock(&ghes_list_mutex);
1329 list_del_rcu(&ghes->list);
1330 if (list_empty(&ghes_nmi))
1331 unregister_nmi_handler(NMI_LOCAL, "ghes");
1332 mutex_unlock(&ghes_list_mutex);
1333 /*
1334 * To synchronize with NMI handler, ghes can only be
1335 * freed after NMI handler finishes.
1336 */
1337 synchronize_rcu();
1338}
1339#else /* CONFIG_HAVE_ACPI_APEI_NMI */
1340static inline void ghes_nmi_add(struct ghes *ghes) { }
1341static inline void ghes_nmi_remove(struct ghes *ghes) { }
1342#endif /* CONFIG_HAVE_ACPI_APEI_NMI */
1343
1344static void ghes_nmi_init_cxt(void)
1345{
1346 init_irq_work(&ghes_proc_irq_work, ghes_proc_in_irq);
1347}
1348
1349static int __ghes_sdei_callback(struct ghes *ghes,
1350 enum fixed_addresses fixmap_idx)
1351{
1352 if (!ghes_in_nmi_queue_one_entry(ghes, fixmap_idx)) {
1353 irq_work_queue(&ghes_proc_irq_work);
1354
1355 return 0;
1356 }
1357
1358 return -ENOENT;
1359}
1360
1361static int ghes_sdei_normal_callback(u32 event_num, struct pt_regs *regs,
1362 void *arg)
1363{
1364 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sdei_normal);
1365 struct ghes *ghes = arg;
1366 int err;
1367
1368 raw_spin_lock(&ghes_notify_lock_sdei_normal);
1369 err = __ghes_sdei_callback(ghes, FIX_APEI_GHES_SDEI_NORMAL);
1370 raw_spin_unlock(&ghes_notify_lock_sdei_normal);
1371
1372 return err;
1373}
1374
1375static int ghes_sdei_critical_callback(u32 event_num, struct pt_regs *regs,
1376 void *arg)
1377{
1378 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sdei_critical);
1379 struct ghes *ghes = arg;
1380 int err;
1381
1382 raw_spin_lock(&ghes_notify_lock_sdei_critical);
1383 err = __ghes_sdei_callback(ghes, FIX_APEI_GHES_SDEI_CRITICAL);
1384 raw_spin_unlock(&ghes_notify_lock_sdei_critical);
1385
1386 return err;
1387}
1388
1389static int apei_sdei_register_ghes(struct ghes *ghes)
1390{
1391 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE))
1392 return -EOPNOTSUPP;
1393
1394 return sdei_register_ghes(ghes, ghes_sdei_normal_callback,
1395 ghes_sdei_critical_callback);
1396}
1397
1398static int apei_sdei_unregister_ghes(struct ghes *ghes)
1399{
1400 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE))
1401 return -EOPNOTSUPP;
1402
1403 return sdei_unregister_ghes(ghes);
1404}
1405
1406static int ghes_probe(struct platform_device *ghes_dev)
1407{
1408 struct acpi_hest_generic *generic;
1409 struct ghes *ghes = NULL;
1410 unsigned long flags;
1411
1412 int rc = -EINVAL;
1413
1414 generic = *(struct acpi_hest_generic **)ghes_dev->dev.platform_data;
1415 if (!generic->enabled)
1416 return -ENODEV;
1417
1418 switch (generic->notify.type) {
1419 case ACPI_HEST_NOTIFY_POLLED:
1420 case ACPI_HEST_NOTIFY_EXTERNAL:
1421 case ACPI_HEST_NOTIFY_SCI:
1422 case ACPI_HEST_NOTIFY_GSIV:
1423 case ACPI_HEST_NOTIFY_GPIO:
1424 break;
1425
1426 case ACPI_HEST_NOTIFY_SEA:
1427 if (!IS_ENABLED(CONFIG_ACPI_APEI_SEA)) {
1428 pr_warn(GHES_PFX "Generic hardware error source: %d notified via SEA is not supported\n",
1429 generic->header.source_id);
1430 rc = -ENOTSUPP;
1431 goto err;
1432 }
1433 break;
1434 case ACPI_HEST_NOTIFY_NMI:
1435 if (!IS_ENABLED(CONFIG_HAVE_ACPI_APEI_NMI)) {
1436 pr_warn(GHES_PFX "Generic hardware error source: %d notified via NMI interrupt is not supported!\n",
1437 generic->header.source_id);
1438 goto err;
1439 }
1440 break;
1441 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1442 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE)) {
1443 pr_warn(GHES_PFX "Generic hardware error source: %d notified via SDE Interface is not supported!\n",
1444 generic->header.source_id);
1445 goto err;
1446 }
1447 break;
1448 case ACPI_HEST_NOTIFY_LOCAL:
1449 pr_warn(GHES_PFX "Generic hardware error source: %d notified via local interrupt is not supported!\n",
1450 generic->header.source_id);
1451 goto err;
1452 default:
1453 pr_warn(FW_WARN GHES_PFX "Unknown notification type: %u for generic hardware error source: %d\n",
1454 generic->notify.type, generic->header.source_id);
1455 goto err;
1456 }
1457
1458 rc = -EIO;
1459 if (generic->error_block_length <
1460 sizeof(struct acpi_hest_generic_status)) {
1461 pr_warn(FW_BUG GHES_PFX "Invalid error block length: %u for generic hardware error source: %d\n",
1462 generic->error_block_length, generic->header.source_id);
1463 goto err;
1464 }
1465 ghes = ghes_new(generic);
1466 if (IS_ERR(ghes)) {
1467 rc = PTR_ERR(ghes);
1468 ghes = NULL;
1469 goto err;
1470 }
1471
1472 switch (generic->notify.type) {
1473 case ACPI_HEST_NOTIFY_POLLED:
1474 timer_setup(&ghes->timer, ghes_poll_func, 0);
1475 ghes_add_timer(ghes);
1476 break;
1477 case ACPI_HEST_NOTIFY_EXTERNAL:
1478 /* External interrupt vector is GSI */
1479 rc = acpi_gsi_to_irq(generic->notify.vector, &ghes->irq);
1480 if (rc) {
1481 pr_err(GHES_PFX "Failed to map GSI to IRQ for generic hardware error source: %d\n",
1482 generic->header.source_id);
1483 goto err;
1484 }
1485 rc = request_irq(ghes->irq, ghes_irq_func, IRQF_SHARED,
1486 "GHES IRQ", ghes);
1487 if (rc) {
1488 pr_err(GHES_PFX "Failed to register IRQ for generic hardware error source: %d\n",
1489 generic->header.source_id);
1490 goto err;
1491 }
1492 break;
1493
1494 case ACPI_HEST_NOTIFY_SCI:
1495 case ACPI_HEST_NOTIFY_GSIV:
1496 case ACPI_HEST_NOTIFY_GPIO:
1497 mutex_lock(&ghes_list_mutex);
1498 if (list_empty(&ghes_hed))
1499 register_acpi_hed_notifier(&ghes_notifier_hed);
1500 list_add_rcu(&ghes->list, &ghes_hed);
1501 mutex_unlock(&ghes_list_mutex);
1502 break;
1503
1504 case ACPI_HEST_NOTIFY_SEA:
1505 ghes_sea_add(ghes);
1506 break;
1507 case ACPI_HEST_NOTIFY_NMI:
1508 ghes_nmi_add(ghes);
1509 break;
1510 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1511 rc = apei_sdei_register_ghes(ghes);
1512 if (rc)
1513 goto err;
1514 break;
1515 default:
1516 BUG();
1517 }
1518
1519 platform_set_drvdata(ghes_dev, ghes);
1520
1521 ghes->dev = &ghes_dev->dev;
1522
1523 mutex_lock(&ghes_devs_mutex);
1524 list_add_tail(&ghes->elist, &ghes_devs);
1525 mutex_unlock(&ghes_devs_mutex);
1526
1527 /* Handle any pending errors right away */
1528 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
1529 ghes_proc(ghes);
1530 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
1531
1532 return 0;
1533
1534err:
1535 if (ghes) {
1536 ghes_fini(ghes);
1537 kfree(ghes);
1538 }
1539 return rc;
1540}
1541
1542static void ghes_remove(struct platform_device *ghes_dev)
1543{
1544 int rc;
1545 struct ghes *ghes;
1546 struct acpi_hest_generic *generic;
1547
1548 ghes = platform_get_drvdata(ghes_dev);
1549 generic = ghes->generic;
1550
1551 ghes->flags |= GHES_EXITING;
1552 switch (generic->notify.type) {
1553 case ACPI_HEST_NOTIFY_POLLED:
1554 timer_shutdown_sync(&ghes->timer);
1555 break;
1556 case ACPI_HEST_NOTIFY_EXTERNAL:
1557 free_irq(ghes->irq, ghes);
1558 break;
1559
1560 case ACPI_HEST_NOTIFY_SCI:
1561 case ACPI_HEST_NOTIFY_GSIV:
1562 case ACPI_HEST_NOTIFY_GPIO:
1563 mutex_lock(&ghes_list_mutex);
1564 list_del_rcu(&ghes->list);
1565 if (list_empty(&ghes_hed))
1566 unregister_acpi_hed_notifier(&ghes_notifier_hed);
1567 mutex_unlock(&ghes_list_mutex);
1568 synchronize_rcu();
1569 break;
1570
1571 case ACPI_HEST_NOTIFY_SEA:
1572 ghes_sea_remove(ghes);
1573 break;
1574 case ACPI_HEST_NOTIFY_NMI:
1575 ghes_nmi_remove(ghes);
1576 break;
1577 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1578 rc = apei_sdei_unregister_ghes(ghes);
1579 if (rc) {
1580 /*
1581 * Returning early results in a resource leak, but we're
1582 * only here if stopping the hardware failed.
1583 */
1584 dev_err(&ghes_dev->dev, "Failed to unregister ghes (%pe)\n",
1585 ERR_PTR(rc));
1586 return;
1587 }
1588 break;
1589 default:
1590 BUG();
1591 break;
1592 }
1593
1594 ghes_fini(ghes);
1595
1596 mutex_lock(&ghes_devs_mutex);
1597 list_del(&ghes->elist);
1598 mutex_unlock(&ghes_devs_mutex);
1599
1600 kfree(ghes);
1601}
1602
1603static struct platform_driver ghes_platform_driver = {
1604 .driver = {
1605 .name = "GHES",
1606 },
1607 .probe = ghes_probe,
1608 .remove = ghes_remove,
1609};
1610
1611void __init acpi_ghes_init(void)
1612{
1613 int rc;
1614
1615 sdei_init();
1616
1617 if (acpi_disabled)
1618 return;
1619
1620 switch (hest_disable) {
1621 case HEST_NOT_FOUND:
1622 return;
1623 case HEST_DISABLED:
1624 pr_info(GHES_PFX "HEST is not enabled!\n");
1625 return;
1626 default:
1627 break;
1628 }
1629
1630 if (ghes_disable) {
1631 pr_info(GHES_PFX "GHES is not enabled!\n");
1632 return;
1633 }
1634
1635 ghes_nmi_init_cxt();
1636
1637 rc = platform_driver_register(&ghes_platform_driver);
1638 if (rc)
1639 return;
1640
1641 rc = apei_osc_setup();
1642 if (rc == 0 && osc_sb_apei_support_acked)
1643 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit and WHEA _OSC.\n");
1644 else if (rc == 0 && !osc_sb_apei_support_acked)
1645 pr_info(GHES_PFX "APEI firmware first mode is enabled by WHEA _OSC.\n");
1646 else if (rc && osc_sb_apei_support_acked)
1647 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit.\n");
1648 else
1649 pr_info(GHES_PFX "Failed to enable APEI firmware first mode.\n");
1650}
1651
1652/*
1653 * Known x86 systems that prefer GHES error reporting:
1654 */
1655static struct acpi_platform_list plat_list[] = {
1656 {"HPE ", "Server ", 0, ACPI_SIG_FADT, all_versions},
1657 { } /* End */
1658};
1659
1660struct list_head *ghes_get_devices(void)
1661{
1662 int idx = -1;
1663
1664 if (IS_ENABLED(CONFIG_X86)) {
1665 idx = acpi_match_platform_list(plat_list);
1666 if (idx < 0) {
1667 if (!ghes_edac_force_enable)
1668 return NULL;
1669
1670 pr_warn_once("Force-loading ghes_edac on an unsupported platform. You're on your own!\n");
1671 }
1672 } else if (list_empty(&ghes_devs)) {
1673 return NULL;
1674 }
1675
1676 return &ghes_devs;
1677}
1678EXPORT_SYMBOL_GPL(ghes_get_devices);
1679
1680void ghes_register_report_chain(struct notifier_block *nb)
1681{
1682 atomic_notifier_chain_register(&ghes_report_chain, nb);
1683}
1684EXPORT_SYMBOL_GPL(ghes_register_report_chain);
1685
1686void ghes_unregister_report_chain(struct notifier_block *nb)
1687{
1688 atomic_notifier_chain_unregister(&ghes_report_chain, nb);
1689}
1690EXPORT_SYMBOL_GPL(ghes_unregister_report_chain);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * APEI Generic Hardware Error Source support
4 *
5 * Generic Hardware Error Source provides a way to report platform
6 * hardware errors (such as that from chipset). It works in so called
7 * "Firmware First" mode, that is, hardware errors are reported to
8 * firmware firstly, then reported to Linux by firmware. This way,
9 * some non-standard hardware error registers or non-standard hardware
10 * link can be checked by firmware to produce more hardware error
11 * information for Linux.
12 *
13 * For more information about Generic Hardware Error Source, please
14 * refer to ACPI Specification version 4.0, section 17.3.2.6
15 *
16 * Copyright 2010,2011 Intel Corp.
17 * Author: Huang Ying <ying.huang@intel.com>
18 */
19
20#include <linux/arm_sdei.h>
21#include <linux/kernel.h>
22#include <linux/moduleparam.h>
23#include <linux/init.h>
24#include <linux/acpi.h>
25#include <linux/io.h>
26#include <linux/interrupt.h>
27#include <linux/timer.h>
28#include <linux/cper.h>
29#include <linux/platform_device.h>
30#include <linux/mutex.h>
31#include <linux/ratelimit.h>
32#include <linux/vmalloc.h>
33#include <linux/irq_work.h>
34#include <linux/llist.h>
35#include <linux/genalloc.h>
36#include <linux/pci.h>
37#include <linux/pfn.h>
38#include <linux/aer.h>
39#include <linux/nmi.h>
40#include <linux/sched/clock.h>
41#include <linux/uuid.h>
42#include <linux/ras.h>
43#include <linux/task_work.h>
44
45#include <acpi/actbl1.h>
46#include <acpi/ghes.h>
47#include <acpi/apei.h>
48#include <asm/fixmap.h>
49#include <asm/tlbflush.h>
50#include <ras/ras_event.h>
51
52#include "apei-internal.h"
53
54#define GHES_PFX "GHES: "
55
56#define GHES_ESTATUS_MAX_SIZE 65536
57#define GHES_ESOURCE_PREALLOC_MAX_SIZE 65536
58
59#define GHES_ESTATUS_POOL_MIN_ALLOC_ORDER 3
60
61/* This is just an estimation for memory pool allocation */
62#define GHES_ESTATUS_CACHE_AVG_SIZE 512
63
64#define GHES_ESTATUS_CACHES_SIZE 4
65
66#define GHES_ESTATUS_IN_CACHE_MAX_NSEC 10000000000ULL
67/* Prevent too many caches are allocated because of RCU */
68#define GHES_ESTATUS_CACHE_ALLOCED_MAX (GHES_ESTATUS_CACHES_SIZE * 3 / 2)
69
70#define GHES_ESTATUS_CACHE_LEN(estatus_len) \
71 (sizeof(struct ghes_estatus_cache) + (estatus_len))
72#define GHES_ESTATUS_FROM_CACHE(estatus_cache) \
73 ((struct acpi_hest_generic_status *) \
74 ((struct ghes_estatus_cache *)(estatus_cache) + 1))
75
76#define GHES_ESTATUS_NODE_LEN(estatus_len) \
77 (sizeof(struct ghes_estatus_node) + (estatus_len))
78#define GHES_ESTATUS_FROM_NODE(estatus_node) \
79 ((struct acpi_hest_generic_status *) \
80 ((struct ghes_estatus_node *)(estatus_node) + 1))
81
82#define GHES_VENDOR_ENTRY_LEN(gdata_len) \
83 (sizeof(struct ghes_vendor_record_entry) + (gdata_len))
84#define GHES_GDATA_FROM_VENDOR_ENTRY(vendor_entry) \
85 ((struct acpi_hest_generic_data *) \
86 ((struct ghes_vendor_record_entry *)(vendor_entry) + 1))
87
88/*
89 * NMI-like notifications vary by architecture, before the compiler can prune
90 * unused static functions it needs a value for these enums.
91 */
92#ifndef CONFIG_ARM_SDE_INTERFACE
93#define FIX_APEI_GHES_SDEI_NORMAL __end_of_fixed_addresses
94#define FIX_APEI_GHES_SDEI_CRITICAL __end_of_fixed_addresses
95#endif
96
97static inline bool is_hest_type_generic_v2(struct ghes *ghes)
98{
99 return ghes->generic->header.type == ACPI_HEST_TYPE_GENERIC_ERROR_V2;
100}
101
102/*
103 * This driver isn't really modular, however for the time being,
104 * continuing to use module_param is the easiest way to remain
105 * compatible with existing boot arg use cases.
106 */
107bool ghes_disable;
108module_param_named(disable, ghes_disable, bool, 0);
109
110/*
111 * All error sources notified with HED (Hardware Error Device) share a
112 * single notifier callback, so they need to be linked and checked one
113 * by one. This holds true for NMI too.
114 *
115 * RCU is used for these lists, so ghes_list_mutex is only used for
116 * list changing, not for traversing.
117 */
118static LIST_HEAD(ghes_hed);
119static DEFINE_MUTEX(ghes_list_mutex);
120
121/*
122 * Because the memory area used to transfer hardware error information
123 * from BIOS to Linux can be determined only in NMI, IRQ or timer
124 * handler, but general ioremap can not be used in atomic context, so
125 * the fixmap is used instead.
126 *
127 * This spinlock is used to prevent the fixmap entry from being used
128 * simultaneously.
129 */
130static DEFINE_SPINLOCK(ghes_notify_lock_irq);
131
132struct ghes_vendor_record_entry {
133 struct work_struct work;
134 int error_severity;
135 char vendor_record[];
136};
137
138static struct gen_pool *ghes_estatus_pool;
139static unsigned long ghes_estatus_pool_size_request;
140
141static struct ghes_estatus_cache *ghes_estatus_caches[GHES_ESTATUS_CACHES_SIZE];
142static atomic_t ghes_estatus_cache_alloced;
143
144static int ghes_panic_timeout __read_mostly = 30;
145
146static void __iomem *ghes_map(u64 pfn, enum fixed_addresses fixmap_idx)
147{
148 phys_addr_t paddr;
149 pgprot_t prot;
150
151 paddr = PFN_PHYS(pfn);
152 prot = arch_apei_get_mem_attribute(paddr);
153 __set_fixmap(fixmap_idx, paddr, prot);
154
155 return (void __iomem *) __fix_to_virt(fixmap_idx);
156}
157
158static void ghes_unmap(void __iomem *vaddr, enum fixed_addresses fixmap_idx)
159{
160 int _idx = virt_to_fix((unsigned long)vaddr);
161
162 WARN_ON_ONCE(fixmap_idx != _idx);
163 clear_fixmap(fixmap_idx);
164}
165
166int ghes_estatus_pool_init(int num_ghes)
167{
168 unsigned long addr, len;
169 int rc;
170
171 ghes_estatus_pool = gen_pool_create(GHES_ESTATUS_POOL_MIN_ALLOC_ORDER, -1);
172 if (!ghes_estatus_pool)
173 return -ENOMEM;
174
175 len = GHES_ESTATUS_CACHE_AVG_SIZE * GHES_ESTATUS_CACHE_ALLOCED_MAX;
176 len += (num_ghes * GHES_ESOURCE_PREALLOC_MAX_SIZE);
177
178 ghes_estatus_pool_size_request = PAGE_ALIGN(len);
179 addr = (unsigned long)vmalloc(PAGE_ALIGN(len));
180 if (!addr)
181 goto err_pool_alloc;
182
183 rc = gen_pool_add(ghes_estatus_pool, addr, PAGE_ALIGN(len), -1);
184 if (rc)
185 goto err_pool_add;
186
187 return 0;
188
189err_pool_add:
190 vfree((void *)addr);
191
192err_pool_alloc:
193 gen_pool_destroy(ghes_estatus_pool);
194
195 return -ENOMEM;
196}
197
198static int map_gen_v2(struct ghes *ghes)
199{
200 return apei_map_generic_address(&ghes->generic_v2->read_ack_register);
201}
202
203static void unmap_gen_v2(struct ghes *ghes)
204{
205 apei_unmap_generic_address(&ghes->generic_v2->read_ack_register);
206}
207
208static void ghes_ack_error(struct acpi_hest_generic_v2 *gv2)
209{
210 int rc;
211 u64 val = 0;
212
213 rc = apei_read(&val, &gv2->read_ack_register);
214 if (rc)
215 return;
216
217 val &= gv2->read_ack_preserve << gv2->read_ack_register.bit_offset;
218 val |= gv2->read_ack_write << gv2->read_ack_register.bit_offset;
219
220 apei_write(val, &gv2->read_ack_register);
221}
222
223static struct ghes *ghes_new(struct acpi_hest_generic *generic)
224{
225 struct ghes *ghes;
226 unsigned int error_block_length;
227 int rc;
228
229 ghes = kzalloc(sizeof(*ghes), GFP_KERNEL);
230 if (!ghes)
231 return ERR_PTR(-ENOMEM);
232
233 ghes->generic = generic;
234 if (is_hest_type_generic_v2(ghes)) {
235 rc = map_gen_v2(ghes);
236 if (rc)
237 goto err_free;
238 }
239
240 rc = apei_map_generic_address(&generic->error_status_address);
241 if (rc)
242 goto err_unmap_read_ack_addr;
243 error_block_length = generic->error_block_length;
244 if (error_block_length > GHES_ESTATUS_MAX_SIZE) {
245 pr_warn(FW_WARN GHES_PFX
246 "Error status block length is too long: %u for "
247 "generic hardware error source: %d.\n",
248 error_block_length, generic->header.source_id);
249 error_block_length = GHES_ESTATUS_MAX_SIZE;
250 }
251 ghes->estatus = kmalloc(error_block_length, GFP_KERNEL);
252 if (!ghes->estatus) {
253 rc = -ENOMEM;
254 goto err_unmap_status_addr;
255 }
256
257 return ghes;
258
259err_unmap_status_addr:
260 apei_unmap_generic_address(&generic->error_status_address);
261err_unmap_read_ack_addr:
262 if (is_hest_type_generic_v2(ghes))
263 unmap_gen_v2(ghes);
264err_free:
265 kfree(ghes);
266 return ERR_PTR(rc);
267}
268
269static void ghes_fini(struct ghes *ghes)
270{
271 kfree(ghes->estatus);
272 apei_unmap_generic_address(&ghes->generic->error_status_address);
273 if (is_hest_type_generic_v2(ghes))
274 unmap_gen_v2(ghes);
275}
276
277static inline int ghes_severity(int severity)
278{
279 switch (severity) {
280 case CPER_SEV_INFORMATIONAL:
281 return GHES_SEV_NO;
282 case CPER_SEV_CORRECTED:
283 return GHES_SEV_CORRECTED;
284 case CPER_SEV_RECOVERABLE:
285 return GHES_SEV_RECOVERABLE;
286 case CPER_SEV_FATAL:
287 return GHES_SEV_PANIC;
288 default:
289 /* Unknown, go panic */
290 return GHES_SEV_PANIC;
291 }
292}
293
294static void ghes_copy_tofrom_phys(void *buffer, u64 paddr, u32 len,
295 int from_phys,
296 enum fixed_addresses fixmap_idx)
297{
298 void __iomem *vaddr;
299 u64 offset;
300 u32 trunk;
301
302 while (len > 0) {
303 offset = paddr - (paddr & PAGE_MASK);
304 vaddr = ghes_map(PHYS_PFN(paddr), fixmap_idx);
305 trunk = PAGE_SIZE - offset;
306 trunk = min(trunk, len);
307 if (from_phys)
308 memcpy_fromio(buffer, vaddr + offset, trunk);
309 else
310 memcpy_toio(vaddr + offset, buffer, trunk);
311 len -= trunk;
312 paddr += trunk;
313 buffer += trunk;
314 ghes_unmap(vaddr, fixmap_idx);
315 }
316}
317
318/* Check the top-level record header has an appropriate size. */
319static int __ghes_check_estatus(struct ghes *ghes,
320 struct acpi_hest_generic_status *estatus)
321{
322 u32 len = cper_estatus_len(estatus);
323
324 if (len < sizeof(*estatus)) {
325 pr_warn_ratelimited(FW_WARN GHES_PFX "Truncated error status block!\n");
326 return -EIO;
327 }
328
329 if (len > ghes->generic->error_block_length) {
330 pr_warn_ratelimited(FW_WARN GHES_PFX "Invalid error status block length!\n");
331 return -EIO;
332 }
333
334 if (cper_estatus_check_header(estatus)) {
335 pr_warn_ratelimited(FW_WARN GHES_PFX "Invalid CPER header!\n");
336 return -EIO;
337 }
338
339 return 0;
340}
341
342/* Read the CPER block, returning its address, and header in estatus. */
343static int __ghes_peek_estatus(struct ghes *ghes,
344 struct acpi_hest_generic_status *estatus,
345 u64 *buf_paddr, enum fixed_addresses fixmap_idx)
346{
347 struct acpi_hest_generic *g = ghes->generic;
348 int rc;
349
350 rc = apei_read(buf_paddr, &g->error_status_address);
351 if (rc) {
352 *buf_paddr = 0;
353 pr_warn_ratelimited(FW_WARN GHES_PFX
354"Failed to read error status block address for hardware error source: %d.\n",
355 g->header.source_id);
356 return -EIO;
357 }
358 if (!*buf_paddr)
359 return -ENOENT;
360
361 ghes_copy_tofrom_phys(estatus, *buf_paddr, sizeof(*estatus), 1,
362 fixmap_idx);
363 if (!estatus->block_status) {
364 *buf_paddr = 0;
365 return -ENOENT;
366 }
367
368 return 0;
369}
370
371static int __ghes_read_estatus(struct acpi_hest_generic_status *estatus,
372 u64 buf_paddr, enum fixed_addresses fixmap_idx,
373 size_t buf_len)
374{
375 ghes_copy_tofrom_phys(estatus, buf_paddr, buf_len, 1, fixmap_idx);
376 if (cper_estatus_check(estatus)) {
377 pr_warn_ratelimited(FW_WARN GHES_PFX
378 "Failed to read error status block!\n");
379 return -EIO;
380 }
381
382 return 0;
383}
384
385static int ghes_read_estatus(struct ghes *ghes,
386 struct acpi_hest_generic_status *estatus,
387 u64 *buf_paddr, enum fixed_addresses fixmap_idx)
388{
389 int rc;
390
391 rc = __ghes_peek_estatus(ghes, estatus, buf_paddr, fixmap_idx);
392 if (rc)
393 return rc;
394
395 rc = __ghes_check_estatus(ghes, estatus);
396 if (rc)
397 return rc;
398
399 return __ghes_read_estatus(estatus, *buf_paddr, fixmap_idx,
400 cper_estatus_len(estatus));
401}
402
403static void ghes_clear_estatus(struct ghes *ghes,
404 struct acpi_hest_generic_status *estatus,
405 u64 buf_paddr, enum fixed_addresses fixmap_idx)
406{
407 estatus->block_status = 0;
408
409 if (!buf_paddr)
410 return;
411
412 ghes_copy_tofrom_phys(estatus, buf_paddr,
413 sizeof(estatus->block_status), 0,
414 fixmap_idx);
415
416 /*
417 * GHESv2 type HEST entries introduce support for error acknowledgment,
418 * so only acknowledge the error if this support is present.
419 */
420 if (is_hest_type_generic_v2(ghes))
421 ghes_ack_error(ghes->generic_v2);
422}
423
424/*
425 * Called as task_work before returning to user-space.
426 * Ensure any queued work has been done before we return to the context that
427 * triggered the notification.
428 */
429static void ghes_kick_task_work(struct callback_head *head)
430{
431 struct acpi_hest_generic_status *estatus;
432 struct ghes_estatus_node *estatus_node;
433 u32 node_len;
434
435 estatus_node = container_of(head, struct ghes_estatus_node, task_work);
436 if (IS_ENABLED(CONFIG_ACPI_APEI_MEMORY_FAILURE))
437 memory_failure_queue_kick(estatus_node->task_work_cpu);
438
439 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
440 node_len = GHES_ESTATUS_NODE_LEN(cper_estatus_len(estatus));
441 gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node, node_len);
442}
443
444static bool ghes_do_memory_failure(u64 physical_addr, int flags)
445{
446 unsigned long pfn;
447
448 if (!IS_ENABLED(CONFIG_ACPI_APEI_MEMORY_FAILURE))
449 return false;
450
451 pfn = PHYS_PFN(physical_addr);
452 if (!pfn_valid(pfn)) {
453 pr_warn_ratelimited(FW_WARN GHES_PFX
454 "Invalid address in generic error data: %#llx\n",
455 physical_addr);
456 return false;
457 }
458
459 memory_failure_queue(pfn, flags);
460 return true;
461}
462
463static bool ghes_handle_memory_failure(struct acpi_hest_generic_data *gdata,
464 int sev)
465{
466 int flags = -1;
467 int sec_sev = ghes_severity(gdata->error_severity);
468 struct cper_sec_mem_err *mem_err = acpi_hest_get_payload(gdata);
469
470 if (!(mem_err->validation_bits & CPER_MEM_VALID_PA))
471 return false;
472
473 /* iff following two events can be handled properly by now */
474 if (sec_sev == GHES_SEV_CORRECTED &&
475 (gdata->flags & CPER_SEC_ERROR_THRESHOLD_EXCEEDED))
476 flags = MF_SOFT_OFFLINE;
477 if (sev == GHES_SEV_RECOVERABLE && sec_sev == GHES_SEV_RECOVERABLE)
478 flags = 0;
479
480 if (flags != -1)
481 return ghes_do_memory_failure(mem_err->physical_addr, flags);
482
483 return false;
484}
485
486static bool ghes_handle_arm_hw_error(struct acpi_hest_generic_data *gdata, int sev)
487{
488 struct cper_sec_proc_arm *err = acpi_hest_get_payload(gdata);
489 bool queued = false;
490 int sec_sev, i;
491 char *p;
492
493 log_arm_hw_error(err);
494
495 sec_sev = ghes_severity(gdata->error_severity);
496 if (sev != GHES_SEV_RECOVERABLE || sec_sev != GHES_SEV_RECOVERABLE)
497 return false;
498
499 p = (char *)(err + 1);
500 for (i = 0; i < err->err_info_num; i++) {
501 struct cper_arm_err_info *err_info = (struct cper_arm_err_info *)p;
502 bool is_cache = (err_info->type == CPER_ARM_CACHE_ERROR);
503 bool has_pa = (err_info->validation_bits & CPER_ARM_INFO_VALID_PHYSICAL_ADDR);
504 const char *error_type = "unknown error";
505
506 /*
507 * The field (err_info->error_info & BIT(26)) is fixed to set to
508 * 1 in some old firmware of HiSilicon Kunpeng920. We assume that
509 * firmware won't mix corrected errors in an uncorrected section,
510 * and don't filter out 'corrected' error here.
511 */
512 if (is_cache && has_pa) {
513 queued = ghes_do_memory_failure(err_info->physical_fault_addr, 0);
514 p += err_info->length;
515 continue;
516 }
517
518 if (err_info->type < ARRAY_SIZE(cper_proc_error_type_strs))
519 error_type = cper_proc_error_type_strs[err_info->type];
520
521 pr_warn_ratelimited(FW_WARN GHES_PFX
522 "Unhandled processor error type: %s\n",
523 error_type);
524 p += err_info->length;
525 }
526
527 return queued;
528}
529
530/*
531 * PCIe AER errors need to be sent to the AER driver for reporting and
532 * recovery. The GHES severities map to the following AER severities and
533 * require the following handling:
534 *
535 * GHES_SEV_CORRECTABLE -> AER_CORRECTABLE
536 * These need to be reported by the AER driver but no recovery is
537 * necessary.
538 * GHES_SEV_RECOVERABLE -> AER_NONFATAL
539 * GHES_SEV_RECOVERABLE && CPER_SEC_RESET -> AER_FATAL
540 * These both need to be reported and recovered from by the AER driver.
541 * GHES_SEV_PANIC does not make it to this handling since the kernel must
542 * panic.
543 */
544static void ghes_handle_aer(struct acpi_hest_generic_data *gdata)
545{
546#ifdef CONFIG_ACPI_APEI_PCIEAER
547 struct cper_sec_pcie *pcie_err = acpi_hest_get_payload(gdata);
548
549 if (pcie_err->validation_bits & CPER_PCIE_VALID_DEVICE_ID &&
550 pcie_err->validation_bits & CPER_PCIE_VALID_AER_INFO) {
551 unsigned int devfn;
552 int aer_severity;
553
554 devfn = PCI_DEVFN(pcie_err->device_id.device,
555 pcie_err->device_id.function);
556 aer_severity = cper_severity_to_aer(gdata->error_severity);
557
558 /*
559 * If firmware reset the component to contain
560 * the error, we must reinitialize it before
561 * use, so treat it as a fatal AER error.
562 */
563 if (gdata->flags & CPER_SEC_RESET)
564 aer_severity = AER_FATAL;
565
566 aer_recover_queue(pcie_err->device_id.segment,
567 pcie_err->device_id.bus,
568 devfn, aer_severity,
569 (struct aer_capability_regs *)
570 pcie_err->aer_info);
571 }
572#endif
573}
574
575static BLOCKING_NOTIFIER_HEAD(vendor_record_notify_list);
576
577int ghes_register_vendor_record_notifier(struct notifier_block *nb)
578{
579 return blocking_notifier_chain_register(&vendor_record_notify_list, nb);
580}
581EXPORT_SYMBOL_GPL(ghes_register_vendor_record_notifier);
582
583void ghes_unregister_vendor_record_notifier(struct notifier_block *nb)
584{
585 blocking_notifier_chain_unregister(&vendor_record_notify_list, nb);
586}
587EXPORT_SYMBOL_GPL(ghes_unregister_vendor_record_notifier);
588
589static void ghes_vendor_record_work_func(struct work_struct *work)
590{
591 struct ghes_vendor_record_entry *entry;
592 struct acpi_hest_generic_data *gdata;
593 u32 len;
594
595 entry = container_of(work, struct ghes_vendor_record_entry, work);
596 gdata = GHES_GDATA_FROM_VENDOR_ENTRY(entry);
597
598 blocking_notifier_call_chain(&vendor_record_notify_list,
599 entry->error_severity, gdata);
600
601 len = GHES_VENDOR_ENTRY_LEN(acpi_hest_get_record_size(gdata));
602 gen_pool_free(ghes_estatus_pool, (unsigned long)entry, len);
603}
604
605static void ghes_defer_non_standard_event(struct acpi_hest_generic_data *gdata,
606 int sev)
607{
608 struct acpi_hest_generic_data *copied_gdata;
609 struct ghes_vendor_record_entry *entry;
610 u32 len;
611
612 len = GHES_VENDOR_ENTRY_LEN(acpi_hest_get_record_size(gdata));
613 entry = (void *)gen_pool_alloc(ghes_estatus_pool, len);
614 if (!entry)
615 return;
616
617 copied_gdata = GHES_GDATA_FROM_VENDOR_ENTRY(entry);
618 memcpy(copied_gdata, gdata, acpi_hest_get_record_size(gdata));
619 entry->error_severity = sev;
620
621 INIT_WORK(&entry->work, ghes_vendor_record_work_func);
622 schedule_work(&entry->work);
623}
624
625static bool ghes_do_proc(struct ghes *ghes,
626 const struct acpi_hest_generic_status *estatus)
627{
628 int sev, sec_sev;
629 struct acpi_hest_generic_data *gdata;
630 guid_t *sec_type;
631 const guid_t *fru_id = &guid_null;
632 char *fru_text = "";
633 bool queued = false;
634
635 sev = ghes_severity(estatus->error_severity);
636 apei_estatus_for_each_section(estatus, gdata) {
637 sec_type = (guid_t *)gdata->section_type;
638 sec_sev = ghes_severity(gdata->error_severity);
639 if (gdata->validation_bits & CPER_SEC_VALID_FRU_ID)
640 fru_id = (guid_t *)gdata->fru_id;
641
642 if (gdata->validation_bits & CPER_SEC_VALID_FRU_TEXT)
643 fru_text = gdata->fru_text;
644
645 if (guid_equal(sec_type, &CPER_SEC_PLATFORM_MEM)) {
646 struct cper_sec_mem_err *mem_err = acpi_hest_get_payload(gdata);
647
648 ghes_edac_report_mem_error(sev, mem_err);
649
650 arch_apei_report_mem_error(sev, mem_err);
651 queued = ghes_handle_memory_failure(gdata, sev);
652 }
653 else if (guid_equal(sec_type, &CPER_SEC_PCIE)) {
654 ghes_handle_aer(gdata);
655 }
656 else if (guid_equal(sec_type, &CPER_SEC_PROC_ARM)) {
657 queued = ghes_handle_arm_hw_error(gdata, sev);
658 } else {
659 void *err = acpi_hest_get_payload(gdata);
660
661 ghes_defer_non_standard_event(gdata, sev);
662 log_non_standard_event(sec_type, fru_id, fru_text,
663 sec_sev, err,
664 gdata->error_data_length);
665 }
666 }
667
668 return queued;
669}
670
671static void __ghes_print_estatus(const char *pfx,
672 const struct acpi_hest_generic *generic,
673 const struct acpi_hest_generic_status *estatus)
674{
675 static atomic_t seqno;
676 unsigned int curr_seqno;
677 char pfx_seq[64];
678
679 if (pfx == NULL) {
680 if (ghes_severity(estatus->error_severity) <=
681 GHES_SEV_CORRECTED)
682 pfx = KERN_WARNING;
683 else
684 pfx = KERN_ERR;
685 }
686 curr_seqno = atomic_inc_return(&seqno);
687 snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}" HW_ERR, pfx, curr_seqno);
688 printk("%s""Hardware error from APEI Generic Hardware Error Source: %d\n",
689 pfx_seq, generic->header.source_id);
690 cper_estatus_print(pfx_seq, estatus);
691}
692
693static int ghes_print_estatus(const char *pfx,
694 const struct acpi_hest_generic *generic,
695 const struct acpi_hest_generic_status *estatus)
696{
697 /* Not more than 2 messages every 5 seconds */
698 static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2);
699 static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2);
700 struct ratelimit_state *ratelimit;
701
702 if (ghes_severity(estatus->error_severity) <= GHES_SEV_CORRECTED)
703 ratelimit = &ratelimit_corrected;
704 else
705 ratelimit = &ratelimit_uncorrected;
706 if (__ratelimit(ratelimit)) {
707 __ghes_print_estatus(pfx, generic, estatus);
708 return 1;
709 }
710 return 0;
711}
712
713/*
714 * GHES error status reporting throttle, to report more kinds of
715 * errors, instead of just most frequently occurred errors.
716 */
717static int ghes_estatus_cached(struct acpi_hest_generic_status *estatus)
718{
719 u32 len;
720 int i, cached = 0;
721 unsigned long long now;
722 struct ghes_estatus_cache *cache;
723 struct acpi_hest_generic_status *cache_estatus;
724
725 len = cper_estatus_len(estatus);
726 rcu_read_lock();
727 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
728 cache = rcu_dereference(ghes_estatus_caches[i]);
729 if (cache == NULL)
730 continue;
731 if (len != cache->estatus_len)
732 continue;
733 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
734 if (memcmp(estatus, cache_estatus, len))
735 continue;
736 atomic_inc(&cache->count);
737 now = sched_clock();
738 if (now - cache->time_in < GHES_ESTATUS_IN_CACHE_MAX_NSEC)
739 cached = 1;
740 break;
741 }
742 rcu_read_unlock();
743 return cached;
744}
745
746static struct ghes_estatus_cache *ghes_estatus_cache_alloc(
747 struct acpi_hest_generic *generic,
748 struct acpi_hest_generic_status *estatus)
749{
750 int alloced;
751 u32 len, cache_len;
752 struct ghes_estatus_cache *cache;
753 struct acpi_hest_generic_status *cache_estatus;
754
755 alloced = atomic_add_return(1, &ghes_estatus_cache_alloced);
756 if (alloced > GHES_ESTATUS_CACHE_ALLOCED_MAX) {
757 atomic_dec(&ghes_estatus_cache_alloced);
758 return NULL;
759 }
760 len = cper_estatus_len(estatus);
761 cache_len = GHES_ESTATUS_CACHE_LEN(len);
762 cache = (void *)gen_pool_alloc(ghes_estatus_pool, cache_len);
763 if (!cache) {
764 atomic_dec(&ghes_estatus_cache_alloced);
765 return NULL;
766 }
767 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
768 memcpy(cache_estatus, estatus, len);
769 cache->estatus_len = len;
770 atomic_set(&cache->count, 0);
771 cache->generic = generic;
772 cache->time_in = sched_clock();
773 return cache;
774}
775
776static void ghes_estatus_cache_free(struct ghes_estatus_cache *cache)
777{
778 u32 len;
779
780 len = cper_estatus_len(GHES_ESTATUS_FROM_CACHE(cache));
781 len = GHES_ESTATUS_CACHE_LEN(len);
782 gen_pool_free(ghes_estatus_pool, (unsigned long)cache, len);
783 atomic_dec(&ghes_estatus_cache_alloced);
784}
785
786static void ghes_estatus_cache_rcu_free(struct rcu_head *head)
787{
788 struct ghes_estatus_cache *cache;
789
790 cache = container_of(head, struct ghes_estatus_cache, rcu);
791 ghes_estatus_cache_free(cache);
792}
793
794static void ghes_estatus_cache_add(
795 struct acpi_hest_generic *generic,
796 struct acpi_hest_generic_status *estatus)
797{
798 int i, slot = -1, count;
799 unsigned long long now, duration, period, max_period = 0;
800 struct ghes_estatus_cache *cache, *slot_cache = NULL, *new_cache;
801
802 new_cache = ghes_estatus_cache_alloc(generic, estatus);
803 if (new_cache == NULL)
804 return;
805 rcu_read_lock();
806 now = sched_clock();
807 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
808 cache = rcu_dereference(ghes_estatus_caches[i]);
809 if (cache == NULL) {
810 slot = i;
811 slot_cache = NULL;
812 break;
813 }
814 duration = now - cache->time_in;
815 if (duration >= GHES_ESTATUS_IN_CACHE_MAX_NSEC) {
816 slot = i;
817 slot_cache = cache;
818 break;
819 }
820 count = atomic_read(&cache->count);
821 period = duration;
822 do_div(period, (count + 1));
823 if (period > max_period) {
824 max_period = period;
825 slot = i;
826 slot_cache = cache;
827 }
828 }
829 /* new_cache must be put into array after its contents are written */
830 smp_wmb();
831 if (slot != -1 && cmpxchg(ghes_estatus_caches + slot,
832 slot_cache, new_cache) == slot_cache) {
833 if (slot_cache)
834 call_rcu(&slot_cache->rcu, ghes_estatus_cache_rcu_free);
835 } else
836 ghes_estatus_cache_free(new_cache);
837 rcu_read_unlock();
838}
839
840static void __ghes_panic(struct ghes *ghes,
841 struct acpi_hest_generic_status *estatus,
842 u64 buf_paddr, enum fixed_addresses fixmap_idx)
843{
844 __ghes_print_estatus(KERN_EMERG, ghes->generic, estatus);
845
846 ghes_clear_estatus(ghes, estatus, buf_paddr, fixmap_idx);
847
848 /* reboot to log the error! */
849 if (!panic_timeout)
850 panic_timeout = ghes_panic_timeout;
851 panic("Fatal hardware error!");
852}
853
854static int ghes_proc(struct ghes *ghes)
855{
856 struct acpi_hest_generic_status *estatus = ghes->estatus;
857 u64 buf_paddr;
858 int rc;
859
860 rc = ghes_read_estatus(ghes, estatus, &buf_paddr, FIX_APEI_GHES_IRQ);
861 if (rc)
862 goto out;
863
864 if (ghes_severity(estatus->error_severity) >= GHES_SEV_PANIC)
865 __ghes_panic(ghes, estatus, buf_paddr, FIX_APEI_GHES_IRQ);
866
867 if (!ghes_estatus_cached(estatus)) {
868 if (ghes_print_estatus(NULL, ghes->generic, estatus))
869 ghes_estatus_cache_add(ghes->generic, estatus);
870 }
871 ghes_do_proc(ghes, estatus);
872
873out:
874 ghes_clear_estatus(ghes, estatus, buf_paddr, FIX_APEI_GHES_IRQ);
875
876 return rc;
877}
878
879static void ghes_add_timer(struct ghes *ghes)
880{
881 struct acpi_hest_generic *g = ghes->generic;
882 unsigned long expire;
883
884 if (!g->notify.poll_interval) {
885 pr_warn(FW_WARN GHES_PFX "Poll interval is 0 for generic hardware error source: %d, disabled.\n",
886 g->header.source_id);
887 return;
888 }
889 expire = jiffies + msecs_to_jiffies(g->notify.poll_interval);
890 ghes->timer.expires = round_jiffies_relative(expire);
891 add_timer(&ghes->timer);
892}
893
894static void ghes_poll_func(struct timer_list *t)
895{
896 struct ghes *ghes = from_timer(ghes, t, timer);
897 unsigned long flags;
898
899 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
900 ghes_proc(ghes);
901 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
902 if (!(ghes->flags & GHES_EXITING))
903 ghes_add_timer(ghes);
904}
905
906static irqreturn_t ghes_irq_func(int irq, void *data)
907{
908 struct ghes *ghes = data;
909 unsigned long flags;
910 int rc;
911
912 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
913 rc = ghes_proc(ghes);
914 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
915 if (rc)
916 return IRQ_NONE;
917
918 return IRQ_HANDLED;
919}
920
921static int ghes_notify_hed(struct notifier_block *this, unsigned long event,
922 void *data)
923{
924 struct ghes *ghes;
925 unsigned long flags;
926 int ret = NOTIFY_DONE;
927
928 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
929 rcu_read_lock();
930 list_for_each_entry_rcu(ghes, &ghes_hed, list) {
931 if (!ghes_proc(ghes))
932 ret = NOTIFY_OK;
933 }
934 rcu_read_unlock();
935 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
936
937 return ret;
938}
939
940static struct notifier_block ghes_notifier_hed = {
941 .notifier_call = ghes_notify_hed,
942};
943
944/*
945 * Handlers for CPER records may not be NMI safe. For example,
946 * memory_failure_queue() takes spinlocks and calls schedule_work_on().
947 * In any NMI-like handler, memory from ghes_estatus_pool is used to save
948 * estatus, and added to the ghes_estatus_llist. irq_work_queue() causes
949 * ghes_proc_in_irq() to run in IRQ context where each estatus in
950 * ghes_estatus_llist is processed.
951 *
952 * Memory from the ghes_estatus_pool is also used with the ghes_estatus_cache
953 * to suppress frequent messages.
954 */
955static struct llist_head ghes_estatus_llist;
956static struct irq_work ghes_proc_irq_work;
957
958static void ghes_proc_in_irq(struct irq_work *irq_work)
959{
960 struct llist_node *llnode, *next;
961 struct ghes_estatus_node *estatus_node;
962 struct acpi_hest_generic *generic;
963 struct acpi_hest_generic_status *estatus;
964 bool task_work_pending;
965 u32 len, node_len;
966 int ret;
967
968 llnode = llist_del_all(&ghes_estatus_llist);
969 /*
970 * Because the time order of estatus in list is reversed,
971 * revert it back to proper order.
972 */
973 llnode = llist_reverse_order(llnode);
974 while (llnode) {
975 next = llnode->next;
976 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
977 llnode);
978 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
979 len = cper_estatus_len(estatus);
980 node_len = GHES_ESTATUS_NODE_LEN(len);
981 task_work_pending = ghes_do_proc(estatus_node->ghes, estatus);
982 if (!ghes_estatus_cached(estatus)) {
983 generic = estatus_node->generic;
984 if (ghes_print_estatus(NULL, generic, estatus))
985 ghes_estatus_cache_add(generic, estatus);
986 }
987
988 if (task_work_pending && current->mm != &init_mm) {
989 estatus_node->task_work.func = ghes_kick_task_work;
990 estatus_node->task_work_cpu = smp_processor_id();
991 ret = task_work_add(current, &estatus_node->task_work,
992 TWA_RESUME);
993 if (ret)
994 estatus_node->task_work.func = NULL;
995 }
996
997 if (!estatus_node->task_work.func)
998 gen_pool_free(ghes_estatus_pool,
999 (unsigned long)estatus_node, node_len);
1000
1001 llnode = next;
1002 }
1003}
1004
1005static void ghes_print_queued_estatus(void)
1006{
1007 struct llist_node *llnode;
1008 struct ghes_estatus_node *estatus_node;
1009 struct acpi_hest_generic *generic;
1010 struct acpi_hest_generic_status *estatus;
1011
1012 llnode = llist_del_all(&ghes_estatus_llist);
1013 /*
1014 * Because the time order of estatus in list is reversed,
1015 * revert it back to proper order.
1016 */
1017 llnode = llist_reverse_order(llnode);
1018 while (llnode) {
1019 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
1020 llnode);
1021 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
1022 generic = estatus_node->generic;
1023 ghes_print_estatus(NULL, generic, estatus);
1024 llnode = llnode->next;
1025 }
1026}
1027
1028static int ghes_in_nmi_queue_one_entry(struct ghes *ghes,
1029 enum fixed_addresses fixmap_idx)
1030{
1031 struct acpi_hest_generic_status *estatus, tmp_header;
1032 struct ghes_estatus_node *estatus_node;
1033 u32 len, node_len;
1034 u64 buf_paddr;
1035 int sev, rc;
1036
1037 if (!IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG))
1038 return -EOPNOTSUPP;
1039
1040 rc = __ghes_peek_estatus(ghes, &tmp_header, &buf_paddr, fixmap_idx);
1041 if (rc) {
1042 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1043 return rc;
1044 }
1045
1046 rc = __ghes_check_estatus(ghes, &tmp_header);
1047 if (rc) {
1048 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1049 return rc;
1050 }
1051
1052 len = cper_estatus_len(&tmp_header);
1053 node_len = GHES_ESTATUS_NODE_LEN(len);
1054 estatus_node = (void *)gen_pool_alloc(ghes_estatus_pool, node_len);
1055 if (!estatus_node)
1056 return -ENOMEM;
1057
1058 estatus_node->ghes = ghes;
1059 estatus_node->generic = ghes->generic;
1060 estatus_node->task_work.func = NULL;
1061 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
1062
1063 if (__ghes_read_estatus(estatus, buf_paddr, fixmap_idx, len)) {
1064 ghes_clear_estatus(ghes, estatus, buf_paddr, fixmap_idx);
1065 rc = -ENOENT;
1066 goto no_work;
1067 }
1068
1069 sev = ghes_severity(estatus->error_severity);
1070 if (sev >= GHES_SEV_PANIC) {
1071 ghes_print_queued_estatus();
1072 __ghes_panic(ghes, estatus, buf_paddr, fixmap_idx);
1073 }
1074
1075 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1076
1077 /* This error has been reported before, don't process it again. */
1078 if (ghes_estatus_cached(estatus))
1079 goto no_work;
1080
1081 llist_add(&estatus_node->llnode, &ghes_estatus_llist);
1082
1083 return rc;
1084
1085no_work:
1086 gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node,
1087 node_len);
1088
1089 return rc;
1090}
1091
1092static int ghes_in_nmi_spool_from_list(struct list_head *rcu_list,
1093 enum fixed_addresses fixmap_idx)
1094{
1095 int ret = -ENOENT;
1096 struct ghes *ghes;
1097
1098 rcu_read_lock();
1099 list_for_each_entry_rcu(ghes, rcu_list, list) {
1100 if (!ghes_in_nmi_queue_one_entry(ghes, fixmap_idx))
1101 ret = 0;
1102 }
1103 rcu_read_unlock();
1104
1105 if (IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG) && !ret)
1106 irq_work_queue(&ghes_proc_irq_work);
1107
1108 return ret;
1109}
1110
1111#ifdef CONFIG_ACPI_APEI_SEA
1112static LIST_HEAD(ghes_sea);
1113
1114/*
1115 * Return 0 only if one of the SEA error sources successfully reported an error
1116 * record sent from the firmware.
1117 */
1118int ghes_notify_sea(void)
1119{
1120 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sea);
1121 int rv;
1122
1123 raw_spin_lock(&ghes_notify_lock_sea);
1124 rv = ghes_in_nmi_spool_from_list(&ghes_sea, FIX_APEI_GHES_SEA);
1125 raw_spin_unlock(&ghes_notify_lock_sea);
1126
1127 return rv;
1128}
1129
1130static void ghes_sea_add(struct ghes *ghes)
1131{
1132 mutex_lock(&ghes_list_mutex);
1133 list_add_rcu(&ghes->list, &ghes_sea);
1134 mutex_unlock(&ghes_list_mutex);
1135}
1136
1137static void ghes_sea_remove(struct ghes *ghes)
1138{
1139 mutex_lock(&ghes_list_mutex);
1140 list_del_rcu(&ghes->list);
1141 mutex_unlock(&ghes_list_mutex);
1142 synchronize_rcu();
1143}
1144#else /* CONFIG_ACPI_APEI_SEA */
1145static inline void ghes_sea_add(struct ghes *ghes) { }
1146static inline void ghes_sea_remove(struct ghes *ghes) { }
1147#endif /* CONFIG_ACPI_APEI_SEA */
1148
1149#ifdef CONFIG_HAVE_ACPI_APEI_NMI
1150/*
1151 * NMI may be triggered on any CPU, so ghes_in_nmi is used for
1152 * having only one concurrent reader.
1153 */
1154static atomic_t ghes_in_nmi = ATOMIC_INIT(0);
1155
1156static LIST_HEAD(ghes_nmi);
1157
1158static int ghes_notify_nmi(unsigned int cmd, struct pt_regs *regs)
1159{
1160 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_nmi);
1161 int ret = NMI_DONE;
1162
1163 if (!atomic_add_unless(&ghes_in_nmi, 1, 1))
1164 return ret;
1165
1166 raw_spin_lock(&ghes_notify_lock_nmi);
1167 if (!ghes_in_nmi_spool_from_list(&ghes_nmi, FIX_APEI_GHES_NMI))
1168 ret = NMI_HANDLED;
1169 raw_spin_unlock(&ghes_notify_lock_nmi);
1170
1171 atomic_dec(&ghes_in_nmi);
1172 return ret;
1173}
1174
1175static void ghes_nmi_add(struct ghes *ghes)
1176{
1177 mutex_lock(&ghes_list_mutex);
1178 if (list_empty(&ghes_nmi))
1179 register_nmi_handler(NMI_LOCAL, ghes_notify_nmi, 0, "ghes");
1180 list_add_rcu(&ghes->list, &ghes_nmi);
1181 mutex_unlock(&ghes_list_mutex);
1182}
1183
1184static void ghes_nmi_remove(struct ghes *ghes)
1185{
1186 mutex_lock(&ghes_list_mutex);
1187 list_del_rcu(&ghes->list);
1188 if (list_empty(&ghes_nmi))
1189 unregister_nmi_handler(NMI_LOCAL, "ghes");
1190 mutex_unlock(&ghes_list_mutex);
1191 /*
1192 * To synchronize with NMI handler, ghes can only be
1193 * freed after NMI handler finishes.
1194 */
1195 synchronize_rcu();
1196}
1197#else /* CONFIG_HAVE_ACPI_APEI_NMI */
1198static inline void ghes_nmi_add(struct ghes *ghes) { }
1199static inline void ghes_nmi_remove(struct ghes *ghes) { }
1200#endif /* CONFIG_HAVE_ACPI_APEI_NMI */
1201
1202static void ghes_nmi_init_cxt(void)
1203{
1204 init_irq_work(&ghes_proc_irq_work, ghes_proc_in_irq);
1205}
1206
1207static int __ghes_sdei_callback(struct ghes *ghes,
1208 enum fixed_addresses fixmap_idx)
1209{
1210 if (!ghes_in_nmi_queue_one_entry(ghes, fixmap_idx)) {
1211 irq_work_queue(&ghes_proc_irq_work);
1212
1213 return 0;
1214 }
1215
1216 return -ENOENT;
1217}
1218
1219static int ghes_sdei_normal_callback(u32 event_num, struct pt_regs *regs,
1220 void *arg)
1221{
1222 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sdei_normal);
1223 struct ghes *ghes = arg;
1224 int err;
1225
1226 raw_spin_lock(&ghes_notify_lock_sdei_normal);
1227 err = __ghes_sdei_callback(ghes, FIX_APEI_GHES_SDEI_NORMAL);
1228 raw_spin_unlock(&ghes_notify_lock_sdei_normal);
1229
1230 return err;
1231}
1232
1233static int ghes_sdei_critical_callback(u32 event_num, struct pt_regs *regs,
1234 void *arg)
1235{
1236 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sdei_critical);
1237 struct ghes *ghes = arg;
1238 int err;
1239
1240 raw_spin_lock(&ghes_notify_lock_sdei_critical);
1241 err = __ghes_sdei_callback(ghes, FIX_APEI_GHES_SDEI_CRITICAL);
1242 raw_spin_unlock(&ghes_notify_lock_sdei_critical);
1243
1244 return err;
1245}
1246
1247static int apei_sdei_register_ghes(struct ghes *ghes)
1248{
1249 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE))
1250 return -EOPNOTSUPP;
1251
1252 return sdei_register_ghes(ghes, ghes_sdei_normal_callback,
1253 ghes_sdei_critical_callback);
1254}
1255
1256static int apei_sdei_unregister_ghes(struct ghes *ghes)
1257{
1258 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE))
1259 return -EOPNOTSUPP;
1260
1261 return sdei_unregister_ghes(ghes);
1262}
1263
1264static int ghes_probe(struct platform_device *ghes_dev)
1265{
1266 struct acpi_hest_generic *generic;
1267 struct ghes *ghes = NULL;
1268 unsigned long flags;
1269
1270 int rc = -EINVAL;
1271
1272 generic = *(struct acpi_hest_generic **)ghes_dev->dev.platform_data;
1273 if (!generic->enabled)
1274 return -ENODEV;
1275
1276 switch (generic->notify.type) {
1277 case ACPI_HEST_NOTIFY_POLLED:
1278 case ACPI_HEST_NOTIFY_EXTERNAL:
1279 case ACPI_HEST_NOTIFY_SCI:
1280 case ACPI_HEST_NOTIFY_GSIV:
1281 case ACPI_HEST_NOTIFY_GPIO:
1282 break;
1283
1284 case ACPI_HEST_NOTIFY_SEA:
1285 if (!IS_ENABLED(CONFIG_ACPI_APEI_SEA)) {
1286 pr_warn(GHES_PFX "Generic hardware error source: %d notified via SEA is not supported\n",
1287 generic->header.source_id);
1288 rc = -ENOTSUPP;
1289 goto err;
1290 }
1291 break;
1292 case ACPI_HEST_NOTIFY_NMI:
1293 if (!IS_ENABLED(CONFIG_HAVE_ACPI_APEI_NMI)) {
1294 pr_warn(GHES_PFX "Generic hardware error source: %d notified via NMI interrupt is not supported!\n",
1295 generic->header.source_id);
1296 goto err;
1297 }
1298 break;
1299 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1300 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE)) {
1301 pr_warn(GHES_PFX "Generic hardware error source: %d notified via SDE Interface is not supported!\n",
1302 generic->header.source_id);
1303 goto err;
1304 }
1305 break;
1306 case ACPI_HEST_NOTIFY_LOCAL:
1307 pr_warn(GHES_PFX "Generic hardware error source: %d notified via local interrupt is not supported!\n",
1308 generic->header.source_id);
1309 goto err;
1310 default:
1311 pr_warn(FW_WARN GHES_PFX "Unknown notification type: %u for generic hardware error source: %d\n",
1312 generic->notify.type, generic->header.source_id);
1313 goto err;
1314 }
1315
1316 rc = -EIO;
1317 if (generic->error_block_length <
1318 sizeof(struct acpi_hest_generic_status)) {
1319 pr_warn(FW_BUG GHES_PFX "Invalid error block length: %u for generic hardware error source: %d\n",
1320 generic->error_block_length, generic->header.source_id);
1321 goto err;
1322 }
1323 ghes = ghes_new(generic);
1324 if (IS_ERR(ghes)) {
1325 rc = PTR_ERR(ghes);
1326 ghes = NULL;
1327 goto err;
1328 }
1329
1330 switch (generic->notify.type) {
1331 case ACPI_HEST_NOTIFY_POLLED:
1332 timer_setup(&ghes->timer, ghes_poll_func, 0);
1333 ghes_add_timer(ghes);
1334 break;
1335 case ACPI_HEST_NOTIFY_EXTERNAL:
1336 /* External interrupt vector is GSI */
1337 rc = acpi_gsi_to_irq(generic->notify.vector, &ghes->irq);
1338 if (rc) {
1339 pr_err(GHES_PFX "Failed to map GSI to IRQ for generic hardware error source: %d\n",
1340 generic->header.source_id);
1341 goto err;
1342 }
1343 rc = request_irq(ghes->irq, ghes_irq_func, IRQF_SHARED,
1344 "GHES IRQ", ghes);
1345 if (rc) {
1346 pr_err(GHES_PFX "Failed to register IRQ for generic hardware error source: %d\n",
1347 generic->header.source_id);
1348 goto err;
1349 }
1350 break;
1351
1352 case ACPI_HEST_NOTIFY_SCI:
1353 case ACPI_HEST_NOTIFY_GSIV:
1354 case ACPI_HEST_NOTIFY_GPIO:
1355 mutex_lock(&ghes_list_mutex);
1356 if (list_empty(&ghes_hed))
1357 register_acpi_hed_notifier(&ghes_notifier_hed);
1358 list_add_rcu(&ghes->list, &ghes_hed);
1359 mutex_unlock(&ghes_list_mutex);
1360 break;
1361
1362 case ACPI_HEST_NOTIFY_SEA:
1363 ghes_sea_add(ghes);
1364 break;
1365 case ACPI_HEST_NOTIFY_NMI:
1366 ghes_nmi_add(ghes);
1367 break;
1368 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1369 rc = apei_sdei_register_ghes(ghes);
1370 if (rc)
1371 goto err;
1372 break;
1373 default:
1374 BUG();
1375 }
1376
1377 platform_set_drvdata(ghes_dev, ghes);
1378
1379 ghes_edac_register(ghes, &ghes_dev->dev);
1380
1381 /* Handle any pending errors right away */
1382 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
1383 ghes_proc(ghes);
1384 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
1385
1386 return 0;
1387
1388err:
1389 if (ghes) {
1390 ghes_fini(ghes);
1391 kfree(ghes);
1392 }
1393 return rc;
1394}
1395
1396static int ghes_remove(struct platform_device *ghes_dev)
1397{
1398 int rc;
1399 struct ghes *ghes;
1400 struct acpi_hest_generic *generic;
1401
1402 ghes = platform_get_drvdata(ghes_dev);
1403 generic = ghes->generic;
1404
1405 ghes->flags |= GHES_EXITING;
1406 switch (generic->notify.type) {
1407 case ACPI_HEST_NOTIFY_POLLED:
1408 del_timer_sync(&ghes->timer);
1409 break;
1410 case ACPI_HEST_NOTIFY_EXTERNAL:
1411 free_irq(ghes->irq, ghes);
1412 break;
1413
1414 case ACPI_HEST_NOTIFY_SCI:
1415 case ACPI_HEST_NOTIFY_GSIV:
1416 case ACPI_HEST_NOTIFY_GPIO:
1417 mutex_lock(&ghes_list_mutex);
1418 list_del_rcu(&ghes->list);
1419 if (list_empty(&ghes_hed))
1420 unregister_acpi_hed_notifier(&ghes_notifier_hed);
1421 mutex_unlock(&ghes_list_mutex);
1422 synchronize_rcu();
1423 break;
1424
1425 case ACPI_HEST_NOTIFY_SEA:
1426 ghes_sea_remove(ghes);
1427 break;
1428 case ACPI_HEST_NOTIFY_NMI:
1429 ghes_nmi_remove(ghes);
1430 break;
1431 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1432 rc = apei_sdei_unregister_ghes(ghes);
1433 if (rc)
1434 return rc;
1435 break;
1436 default:
1437 BUG();
1438 break;
1439 }
1440
1441 ghes_fini(ghes);
1442
1443 ghes_edac_unregister(ghes);
1444
1445 kfree(ghes);
1446
1447 platform_set_drvdata(ghes_dev, NULL);
1448
1449 return 0;
1450}
1451
1452static struct platform_driver ghes_platform_driver = {
1453 .driver = {
1454 .name = "GHES",
1455 },
1456 .probe = ghes_probe,
1457 .remove = ghes_remove,
1458};
1459
1460static int __init ghes_init(void)
1461{
1462 int rc;
1463
1464 if (acpi_disabled)
1465 return -ENODEV;
1466
1467 switch (hest_disable) {
1468 case HEST_NOT_FOUND:
1469 return -ENODEV;
1470 case HEST_DISABLED:
1471 pr_info(GHES_PFX "HEST is not enabled!\n");
1472 return -EINVAL;
1473 default:
1474 break;
1475 }
1476
1477 if (ghes_disable) {
1478 pr_info(GHES_PFX "GHES is not enabled!\n");
1479 return -EINVAL;
1480 }
1481
1482 ghes_nmi_init_cxt();
1483
1484 rc = platform_driver_register(&ghes_platform_driver);
1485 if (rc)
1486 goto err;
1487
1488 rc = apei_osc_setup();
1489 if (rc == 0 && osc_sb_apei_support_acked)
1490 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit and WHEA _OSC.\n");
1491 else if (rc == 0 && !osc_sb_apei_support_acked)
1492 pr_info(GHES_PFX "APEI firmware first mode is enabled by WHEA _OSC.\n");
1493 else if (rc && osc_sb_apei_support_acked)
1494 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit.\n");
1495 else
1496 pr_info(GHES_PFX "Failed to enable APEI firmware first mode.\n");
1497
1498 return 0;
1499err:
1500 return rc;
1501}
1502device_initcall(ghes_init);