1 // SPDX-License-Identifier: GPL-2.0-only
3 * APEI Generic Hardware Error Source support
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.
13 * For more information about Generic Hardware Error Source, please
14 * refer to ACPI Specification version 4.0, section 17.3.2.6
16 * Copyright 2010,2011 Intel Corp.
17 * Author: Huang Ying <ying.huang@intel.com>
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>
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>
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>
52 #include "apei-internal.h"
54 #define GHES_PFX "GHES: "
56 #define GHES_ESTATUS_MAX_SIZE 65536
57 #define GHES_ESOURCE_PREALLOC_MAX_SIZE 65536
59 #define GHES_ESTATUS_POOL_MIN_ALLOC_ORDER 3
61 /* This is just an estimation for memory pool allocation */
62 #define GHES_ESTATUS_CACHE_AVG_SIZE 512
64 #define GHES_ESTATUS_CACHES_SIZE 4
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)
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))
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))
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))
89 * NMI-like notifications vary by architecture, before the compiler can prune
90 * unused static functions it needs a value for these enums.
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
97 static ATOMIC_NOTIFIER_HEAD(ghes_report_chain);
99 static inline bool is_hest_type_generic_v2(struct ghes *ghes)
101 return ghes->generic->header.type == ACPI_HEST_TYPE_GENERIC_ERROR_V2;
105 * This driver isn't really modular, however for the time being,
106 * continuing to use module_param is the easiest way to remain
107 * compatible with existing boot arg use cases.
110 module_param_named(disable, ghes_disable, bool, 0);
113 * "ghes.edac_force_enable" forcibly enables ghes_edac and skips the platform
116 static bool ghes_edac_force_enable;
117 module_param_named(edac_force_enable, ghes_edac_force_enable, bool, 0);
120 * All error sources notified with HED (Hardware Error Device) share a
121 * single notifier callback, so they need to be linked and checked one
122 * by one. This holds true for NMI too.
124 * RCU is used for these lists, so ghes_list_mutex is only used for
125 * list changing, not for traversing.
127 static LIST_HEAD(ghes_hed);
128 static DEFINE_MUTEX(ghes_list_mutex);
131 * A list of GHES devices which are given to the corresponding EDAC driver
132 * ghes_edac for further use.
134 static LIST_HEAD(ghes_devs);
135 static DEFINE_MUTEX(ghes_devs_mutex);
138 * Because the memory area used to transfer hardware error information
139 * from BIOS to Linux can be determined only in NMI, IRQ or timer
140 * handler, but general ioremap can not be used in atomic context, so
141 * the fixmap is used instead.
143 * This spinlock is used to prevent the fixmap entry from being used
146 static DEFINE_SPINLOCK(ghes_notify_lock_irq);
148 struct ghes_vendor_record_entry {
149 struct work_struct work;
151 char vendor_record[];
154 static struct gen_pool *ghes_estatus_pool;
156 static struct ghes_estatus_cache __rcu *ghes_estatus_caches[GHES_ESTATUS_CACHES_SIZE];
157 static atomic_t ghes_estatus_cache_alloced;
159 static int ghes_panic_timeout __read_mostly = 30;
161 static void __iomem *ghes_map(u64 pfn, enum fixed_addresses fixmap_idx)
166 paddr = PFN_PHYS(pfn);
167 prot = arch_apei_get_mem_attribute(paddr);
168 __set_fixmap(fixmap_idx, paddr, prot);
170 return (void __iomem *) __fix_to_virt(fixmap_idx);
173 static void ghes_unmap(void __iomem *vaddr, enum fixed_addresses fixmap_idx)
175 int _idx = virt_to_fix((unsigned long)vaddr);
177 WARN_ON_ONCE(fixmap_idx != _idx);
178 clear_fixmap(fixmap_idx);
181 int ghes_estatus_pool_init(unsigned int num_ghes)
183 unsigned long addr, len;
186 ghes_estatus_pool = gen_pool_create(GHES_ESTATUS_POOL_MIN_ALLOC_ORDER, -1);
187 if (!ghes_estatus_pool)
190 len = GHES_ESTATUS_CACHE_AVG_SIZE * GHES_ESTATUS_CACHE_ALLOCED_MAX;
191 len += (num_ghes * GHES_ESOURCE_PREALLOC_MAX_SIZE);
193 addr = (unsigned long)vmalloc(PAGE_ALIGN(len));
197 rc = gen_pool_add(ghes_estatus_pool, addr, PAGE_ALIGN(len), -1);
207 gen_pool_destroy(ghes_estatus_pool);
212 static int map_gen_v2(struct ghes *ghes)
214 return apei_map_generic_address(&ghes->generic_v2->read_ack_register);
217 static void unmap_gen_v2(struct ghes *ghes)
219 apei_unmap_generic_address(&ghes->generic_v2->read_ack_register);
222 static void ghes_ack_error(struct acpi_hest_generic_v2 *gv2)
227 rc = apei_read(&val, &gv2->read_ack_register);
231 val &= gv2->read_ack_preserve << gv2->read_ack_register.bit_offset;
232 val |= gv2->read_ack_write << gv2->read_ack_register.bit_offset;
234 apei_write(val, &gv2->read_ack_register);
237 static struct ghes *ghes_new(struct acpi_hest_generic *generic)
240 unsigned int error_block_length;
243 ghes = kzalloc(sizeof(*ghes), GFP_KERNEL);
245 return ERR_PTR(-ENOMEM);
247 ghes->generic = generic;
248 if (is_hest_type_generic_v2(ghes)) {
249 rc = map_gen_v2(ghes);
254 rc = apei_map_generic_address(&generic->error_status_address);
256 goto err_unmap_read_ack_addr;
257 error_block_length = generic->error_block_length;
258 if (error_block_length > GHES_ESTATUS_MAX_SIZE) {
259 pr_warn(FW_WARN GHES_PFX
260 "Error status block length is too long: %u for "
261 "generic hardware error source: %d.\n",
262 error_block_length, generic->header.source_id);
263 error_block_length = GHES_ESTATUS_MAX_SIZE;
265 ghes->estatus = kmalloc(error_block_length, GFP_KERNEL);
266 if (!ghes->estatus) {
268 goto err_unmap_status_addr;
273 err_unmap_status_addr:
274 apei_unmap_generic_address(&generic->error_status_address);
275 err_unmap_read_ack_addr:
276 if (is_hest_type_generic_v2(ghes))
283 static void ghes_fini(struct ghes *ghes)
285 kfree(ghes->estatus);
286 apei_unmap_generic_address(&ghes->generic->error_status_address);
287 if (is_hest_type_generic_v2(ghes))
291 static inline int ghes_severity(int severity)
294 case CPER_SEV_INFORMATIONAL:
296 case CPER_SEV_CORRECTED:
297 return GHES_SEV_CORRECTED;
298 case CPER_SEV_RECOVERABLE:
299 return GHES_SEV_RECOVERABLE;
301 return GHES_SEV_PANIC;
303 /* Unknown, go panic */
304 return GHES_SEV_PANIC;
308 static void ghes_copy_tofrom_phys(void *buffer, u64 paddr, u32 len,
310 enum fixed_addresses fixmap_idx)
317 offset = paddr - (paddr & PAGE_MASK);
318 vaddr = ghes_map(PHYS_PFN(paddr), fixmap_idx);
319 trunk = PAGE_SIZE - offset;
320 trunk = min(trunk, len);
322 memcpy_fromio(buffer, vaddr + offset, trunk);
324 memcpy_toio(vaddr + offset, buffer, trunk);
328 ghes_unmap(vaddr, fixmap_idx);
332 /* Check the top-level record header has an appropriate size. */
333 static int __ghes_check_estatus(struct ghes *ghes,
334 struct acpi_hest_generic_status *estatus)
336 u32 len = cper_estatus_len(estatus);
338 if (len < sizeof(*estatus)) {
339 pr_warn_ratelimited(FW_WARN GHES_PFX "Truncated error status block!\n");
343 if (len > ghes->generic->error_block_length) {
344 pr_warn_ratelimited(FW_WARN GHES_PFX "Invalid error status block length!\n");
348 if (cper_estatus_check_header(estatus)) {
349 pr_warn_ratelimited(FW_WARN GHES_PFX "Invalid CPER header!\n");
356 /* Read the CPER block, returning its address, and header in estatus. */
357 static int __ghes_peek_estatus(struct ghes *ghes,
358 struct acpi_hest_generic_status *estatus,
359 u64 *buf_paddr, enum fixed_addresses fixmap_idx)
361 struct acpi_hest_generic *g = ghes->generic;
364 rc = apei_read(buf_paddr, &g->error_status_address);
367 pr_warn_ratelimited(FW_WARN GHES_PFX
368 "Failed to read error status block address for hardware error source: %d.\n",
369 g->header.source_id);
375 ghes_copy_tofrom_phys(estatus, *buf_paddr, sizeof(*estatus), 1,
377 if (!estatus->block_status) {
385 static int __ghes_read_estatus(struct acpi_hest_generic_status *estatus,
386 u64 buf_paddr, enum fixed_addresses fixmap_idx,
389 ghes_copy_tofrom_phys(estatus, buf_paddr, buf_len, 1, fixmap_idx);
390 if (cper_estatus_check(estatus)) {
391 pr_warn_ratelimited(FW_WARN GHES_PFX
392 "Failed to read error status block!\n");
399 static int ghes_read_estatus(struct ghes *ghes,
400 struct acpi_hest_generic_status *estatus,
401 u64 *buf_paddr, enum fixed_addresses fixmap_idx)
405 rc = __ghes_peek_estatus(ghes, estatus, buf_paddr, fixmap_idx);
409 rc = __ghes_check_estatus(ghes, estatus);
413 return __ghes_read_estatus(estatus, *buf_paddr, fixmap_idx,
414 cper_estatus_len(estatus));
417 static void ghes_clear_estatus(struct ghes *ghes,
418 struct acpi_hest_generic_status *estatus,
419 u64 buf_paddr, enum fixed_addresses fixmap_idx)
421 estatus->block_status = 0;
426 ghes_copy_tofrom_phys(estatus, buf_paddr,
427 sizeof(estatus->block_status), 0,
431 * GHESv2 type HEST entries introduce support for error acknowledgment,
432 * so only acknowledge the error if this support is present.
434 if (is_hest_type_generic_v2(ghes))
435 ghes_ack_error(ghes->generic_v2);
439 * Called as task_work before returning to user-space.
440 * Ensure any queued work has been done before we return to the context that
441 * triggered the notification.
443 static void ghes_kick_task_work(struct callback_head *head)
445 struct acpi_hest_generic_status *estatus;
446 struct ghes_estatus_node *estatus_node;
449 estatus_node = container_of(head, struct ghes_estatus_node, task_work);
450 if (IS_ENABLED(CONFIG_ACPI_APEI_MEMORY_FAILURE))
451 memory_failure_queue_kick(estatus_node->task_work_cpu);
453 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
454 node_len = GHES_ESTATUS_NODE_LEN(cper_estatus_len(estatus));
455 gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node, node_len);
458 static bool ghes_do_memory_failure(u64 physical_addr, int flags)
462 if (!IS_ENABLED(CONFIG_ACPI_APEI_MEMORY_FAILURE))
465 pfn = PHYS_PFN(physical_addr);
466 if (!pfn_valid(pfn) && !arch_is_platform_page(physical_addr)) {
467 pr_warn_ratelimited(FW_WARN GHES_PFX
468 "Invalid address in generic error data: %#llx\n",
473 memory_failure_queue(pfn, flags);
477 static bool ghes_handle_memory_failure(struct acpi_hest_generic_data *gdata,
481 int sec_sev = ghes_severity(gdata->error_severity);
482 struct cper_sec_mem_err *mem_err = acpi_hest_get_payload(gdata);
484 if (!(mem_err->validation_bits & CPER_MEM_VALID_PA))
487 /* iff following two events can be handled properly by now */
488 if (sec_sev == GHES_SEV_CORRECTED &&
489 (gdata->flags & CPER_SEC_ERROR_THRESHOLD_EXCEEDED))
490 flags = MF_SOFT_OFFLINE;
491 if (sev == GHES_SEV_RECOVERABLE && sec_sev == GHES_SEV_RECOVERABLE)
495 return ghes_do_memory_failure(mem_err->physical_addr, flags);
500 static bool ghes_handle_arm_hw_error(struct acpi_hest_generic_data *gdata, int sev)
502 struct cper_sec_proc_arm *err = acpi_hest_get_payload(gdata);
507 log_arm_hw_error(err);
509 sec_sev = ghes_severity(gdata->error_severity);
510 if (sev != GHES_SEV_RECOVERABLE || sec_sev != GHES_SEV_RECOVERABLE)
513 p = (char *)(err + 1);
514 for (i = 0; i < err->err_info_num; i++) {
515 struct cper_arm_err_info *err_info = (struct cper_arm_err_info *)p;
516 bool is_cache = (err_info->type == CPER_ARM_CACHE_ERROR);
517 bool has_pa = (err_info->validation_bits & CPER_ARM_INFO_VALID_PHYSICAL_ADDR);
518 const char *error_type = "unknown error";
521 * The field (err_info->error_info & BIT(26)) is fixed to set to
522 * 1 in some old firmware of HiSilicon Kunpeng920. We assume that
523 * firmware won't mix corrected errors in an uncorrected section,
524 * and don't filter out 'corrected' error here.
526 if (is_cache && has_pa) {
527 queued = ghes_do_memory_failure(err_info->physical_fault_addr, 0);
528 p += err_info->length;
532 if (err_info->type < ARRAY_SIZE(cper_proc_error_type_strs))
533 error_type = cper_proc_error_type_strs[err_info->type];
535 pr_warn_ratelimited(FW_WARN GHES_PFX
536 "Unhandled processor error type: %s\n",
538 p += err_info->length;
545 * PCIe AER errors need to be sent to the AER driver for reporting and
546 * recovery. The GHES severities map to the following AER severities and
547 * require the following handling:
549 * GHES_SEV_CORRECTABLE -> AER_CORRECTABLE
550 * These need to be reported by the AER driver but no recovery is
552 * GHES_SEV_RECOVERABLE -> AER_NONFATAL
553 * GHES_SEV_RECOVERABLE && CPER_SEC_RESET -> AER_FATAL
554 * These both need to be reported and recovered from by the AER driver.
555 * GHES_SEV_PANIC does not make it to this handling since the kernel must
558 static void ghes_handle_aer(struct acpi_hest_generic_data *gdata)
560 #ifdef CONFIG_ACPI_APEI_PCIEAER
561 struct cper_sec_pcie *pcie_err = acpi_hest_get_payload(gdata);
563 if (pcie_err->validation_bits & CPER_PCIE_VALID_DEVICE_ID &&
564 pcie_err->validation_bits & CPER_PCIE_VALID_AER_INFO) {
568 devfn = PCI_DEVFN(pcie_err->device_id.device,
569 pcie_err->device_id.function);
570 aer_severity = cper_severity_to_aer(gdata->error_severity);
573 * If firmware reset the component to contain
574 * the error, we must reinitialize it before
575 * use, so treat it as a fatal AER error.
577 if (gdata->flags & CPER_SEC_RESET)
578 aer_severity = AER_FATAL;
580 aer_recover_queue(pcie_err->device_id.segment,
581 pcie_err->device_id.bus,
583 (struct aer_capability_regs *)
589 static BLOCKING_NOTIFIER_HEAD(vendor_record_notify_list);
591 int ghes_register_vendor_record_notifier(struct notifier_block *nb)
593 return blocking_notifier_chain_register(&vendor_record_notify_list, nb);
595 EXPORT_SYMBOL_GPL(ghes_register_vendor_record_notifier);
597 void ghes_unregister_vendor_record_notifier(struct notifier_block *nb)
599 blocking_notifier_chain_unregister(&vendor_record_notify_list, nb);
601 EXPORT_SYMBOL_GPL(ghes_unregister_vendor_record_notifier);
603 static void ghes_vendor_record_work_func(struct work_struct *work)
605 struct ghes_vendor_record_entry *entry;
606 struct acpi_hest_generic_data *gdata;
609 entry = container_of(work, struct ghes_vendor_record_entry, work);
610 gdata = GHES_GDATA_FROM_VENDOR_ENTRY(entry);
612 blocking_notifier_call_chain(&vendor_record_notify_list,
613 entry->error_severity, gdata);
615 len = GHES_VENDOR_ENTRY_LEN(acpi_hest_get_record_size(gdata));
616 gen_pool_free(ghes_estatus_pool, (unsigned long)entry, len);
619 static void ghes_defer_non_standard_event(struct acpi_hest_generic_data *gdata,
622 struct acpi_hest_generic_data *copied_gdata;
623 struct ghes_vendor_record_entry *entry;
626 len = GHES_VENDOR_ENTRY_LEN(acpi_hest_get_record_size(gdata));
627 entry = (void *)gen_pool_alloc(ghes_estatus_pool, len);
631 copied_gdata = GHES_GDATA_FROM_VENDOR_ENTRY(entry);
632 memcpy(copied_gdata, gdata, acpi_hest_get_record_size(gdata));
633 entry->error_severity = sev;
635 INIT_WORK(&entry->work, ghes_vendor_record_work_func);
636 schedule_work(&entry->work);
639 static bool ghes_do_proc(struct ghes *ghes,
640 const struct acpi_hest_generic_status *estatus)
643 struct acpi_hest_generic_data *gdata;
645 const guid_t *fru_id = &guid_null;
649 sev = ghes_severity(estatus->error_severity);
650 apei_estatus_for_each_section(estatus, gdata) {
651 sec_type = (guid_t *)gdata->section_type;
652 sec_sev = ghes_severity(gdata->error_severity);
653 if (gdata->validation_bits & CPER_SEC_VALID_FRU_ID)
654 fru_id = (guid_t *)gdata->fru_id;
656 if (gdata->validation_bits & CPER_SEC_VALID_FRU_TEXT)
657 fru_text = gdata->fru_text;
659 if (guid_equal(sec_type, &CPER_SEC_PLATFORM_MEM)) {
660 struct cper_sec_mem_err *mem_err = acpi_hest_get_payload(gdata);
662 atomic_notifier_call_chain(&ghes_report_chain, sev, mem_err);
664 arch_apei_report_mem_error(sev, mem_err);
665 queued = ghes_handle_memory_failure(gdata, sev);
667 else if (guid_equal(sec_type, &CPER_SEC_PCIE)) {
668 ghes_handle_aer(gdata);
670 else if (guid_equal(sec_type, &CPER_SEC_PROC_ARM)) {
671 queued = ghes_handle_arm_hw_error(gdata, sev);
673 void *err = acpi_hest_get_payload(gdata);
675 ghes_defer_non_standard_event(gdata, sev);
676 log_non_standard_event(sec_type, fru_id, fru_text,
678 gdata->error_data_length);
685 static void __ghes_print_estatus(const char *pfx,
686 const struct acpi_hest_generic *generic,
687 const struct acpi_hest_generic_status *estatus)
689 static atomic_t seqno;
690 unsigned int curr_seqno;
694 if (ghes_severity(estatus->error_severity) <=
700 curr_seqno = atomic_inc_return(&seqno);
701 snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}" HW_ERR, pfx, curr_seqno);
702 printk("%s""Hardware error from APEI Generic Hardware Error Source: %d\n",
703 pfx_seq, generic->header.source_id);
704 cper_estatus_print(pfx_seq, estatus);
707 static int ghes_print_estatus(const char *pfx,
708 const struct acpi_hest_generic *generic,
709 const struct acpi_hest_generic_status *estatus)
711 /* Not more than 2 messages every 5 seconds */
712 static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2);
713 static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2);
714 struct ratelimit_state *ratelimit;
716 if (ghes_severity(estatus->error_severity) <= GHES_SEV_CORRECTED)
717 ratelimit = &ratelimit_corrected;
719 ratelimit = &ratelimit_uncorrected;
720 if (__ratelimit(ratelimit)) {
721 __ghes_print_estatus(pfx, generic, estatus);
728 * GHES error status reporting throttle, to report more kinds of
729 * errors, instead of just most frequently occurred errors.
731 static int ghes_estatus_cached(struct acpi_hest_generic_status *estatus)
735 unsigned long long now;
736 struct ghes_estatus_cache *cache;
737 struct acpi_hest_generic_status *cache_estatus;
739 len = cper_estatus_len(estatus);
741 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
742 cache = rcu_dereference(ghes_estatus_caches[i]);
745 if (len != cache->estatus_len)
747 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
748 if (memcmp(estatus, cache_estatus, len))
750 atomic_inc(&cache->count);
752 if (now - cache->time_in < GHES_ESTATUS_IN_CACHE_MAX_NSEC)
760 static struct ghes_estatus_cache *ghes_estatus_cache_alloc(
761 struct acpi_hest_generic *generic,
762 struct acpi_hest_generic_status *estatus)
766 struct ghes_estatus_cache *cache;
767 struct acpi_hest_generic_status *cache_estatus;
769 alloced = atomic_add_return(1, &ghes_estatus_cache_alloced);
770 if (alloced > GHES_ESTATUS_CACHE_ALLOCED_MAX) {
771 atomic_dec(&ghes_estatus_cache_alloced);
774 len = cper_estatus_len(estatus);
775 cache_len = GHES_ESTATUS_CACHE_LEN(len);
776 cache = (void *)gen_pool_alloc(ghes_estatus_pool, cache_len);
778 atomic_dec(&ghes_estatus_cache_alloced);
781 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
782 memcpy(cache_estatus, estatus, len);
783 cache->estatus_len = len;
784 atomic_set(&cache->count, 0);
785 cache->generic = generic;
786 cache->time_in = sched_clock();
790 static void ghes_estatus_cache_rcu_free(struct rcu_head *head)
792 struct ghes_estatus_cache *cache;
795 cache = container_of(head, struct ghes_estatus_cache, rcu);
796 len = cper_estatus_len(GHES_ESTATUS_FROM_CACHE(cache));
797 len = GHES_ESTATUS_CACHE_LEN(len);
798 gen_pool_free(ghes_estatus_pool, (unsigned long)cache, len);
799 atomic_dec(&ghes_estatus_cache_alloced);
803 ghes_estatus_cache_add(struct acpi_hest_generic *generic,
804 struct acpi_hest_generic_status *estatus)
806 unsigned long long now, duration, period, max_period = 0;
807 struct ghes_estatus_cache *cache, *new_cache;
808 struct ghes_estatus_cache __rcu *victim;
809 int i, slot = -1, count;
811 new_cache = ghes_estatus_cache_alloc(generic, estatus);
817 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
818 cache = rcu_dereference(ghes_estatus_caches[i]);
823 duration = now - cache->time_in;
824 if (duration >= GHES_ESTATUS_IN_CACHE_MAX_NSEC) {
828 count = atomic_read(&cache->count);
830 do_div(period, (count + 1));
831 if (period > max_period) {
840 * Use release semantics to ensure that ghes_estatus_cached()
841 * running on another CPU will see the updated cache fields if
842 * it can see the new value of the pointer.
844 victim = xchg_release(&ghes_estatus_caches[slot],
845 RCU_INITIALIZER(new_cache));
848 * At this point, victim may point to a cached item different
849 * from the one based on which we selected the slot. Instead of
850 * going to the loop again to pick another slot, let's just
851 * drop the other item anyway: this may cause a false cache
852 * miss later on, but that won't cause any problems.
855 call_rcu(&unrcu_pointer(victim)->rcu,
856 ghes_estatus_cache_rcu_free);
860 static void __ghes_panic(struct ghes *ghes,
861 struct acpi_hest_generic_status *estatus,
862 u64 buf_paddr, enum fixed_addresses fixmap_idx)
864 __ghes_print_estatus(KERN_EMERG, ghes->generic, estatus);
866 ghes_clear_estatus(ghes, estatus, buf_paddr, fixmap_idx);
868 /* reboot to log the error! */
870 panic_timeout = ghes_panic_timeout;
871 panic("Fatal hardware error!");
874 static int ghes_proc(struct ghes *ghes)
876 struct acpi_hest_generic_status *estatus = ghes->estatus;
880 rc = ghes_read_estatus(ghes, estatus, &buf_paddr, FIX_APEI_GHES_IRQ);
884 if (ghes_severity(estatus->error_severity) >= GHES_SEV_PANIC)
885 __ghes_panic(ghes, estatus, buf_paddr, FIX_APEI_GHES_IRQ);
887 if (!ghes_estatus_cached(estatus)) {
888 if (ghes_print_estatus(NULL, ghes->generic, estatus))
889 ghes_estatus_cache_add(ghes->generic, estatus);
891 ghes_do_proc(ghes, estatus);
894 ghes_clear_estatus(ghes, estatus, buf_paddr, FIX_APEI_GHES_IRQ);
899 static void ghes_add_timer(struct ghes *ghes)
901 struct acpi_hest_generic *g = ghes->generic;
902 unsigned long expire;
904 if (!g->notify.poll_interval) {
905 pr_warn(FW_WARN GHES_PFX "Poll interval is 0 for generic hardware error source: %d, disabled.\n",
906 g->header.source_id);
909 expire = jiffies + msecs_to_jiffies(g->notify.poll_interval);
910 ghes->timer.expires = round_jiffies_relative(expire);
911 add_timer(&ghes->timer);
914 static void ghes_poll_func(struct timer_list *t)
916 struct ghes *ghes = from_timer(ghes, t, timer);
919 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
921 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
922 if (!(ghes->flags & GHES_EXITING))
923 ghes_add_timer(ghes);
926 static irqreturn_t ghes_irq_func(int irq, void *data)
928 struct ghes *ghes = data;
932 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
933 rc = ghes_proc(ghes);
934 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
941 static int ghes_notify_hed(struct notifier_block *this, unsigned long event,
946 int ret = NOTIFY_DONE;
948 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
950 list_for_each_entry_rcu(ghes, &ghes_hed, list) {
951 if (!ghes_proc(ghes))
955 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
960 static struct notifier_block ghes_notifier_hed = {
961 .notifier_call = ghes_notify_hed,
965 * Handlers for CPER records may not be NMI safe. For example,
966 * memory_failure_queue() takes spinlocks and calls schedule_work_on().
967 * In any NMI-like handler, memory from ghes_estatus_pool is used to save
968 * estatus, and added to the ghes_estatus_llist. irq_work_queue() causes
969 * ghes_proc_in_irq() to run in IRQ context where each estatus in
970 * ghes_estatus_llist is processed.
972 * Memory from the ghes_estatus_pool is also used with the ghes_estatus_cache
973 * to suppress frequent messages.
975 static struct llist_head ghes_estatus_llist;
976 static struct irq_work ghes_proc_irq_work;
978 static void ghes_proc_in_irq(struct irq_work *irq_work)
980 struct llist_node *llnode, *next;
981 struct ghes_estatus_node *estatus_node;
982 struct acpi_hest_generic *generic;
983 struct acpi_hest_generic_status *estatus;
984 bool task_work_pending;
988 llnode = llist_del_all(&ghes_estatus_llist);
990 * Because the time order of estatus in list is reversed,
991 * revert it back to proper order.
993 llnode = llist_reverse_order(llnode);
996 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
998 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
999 len = cper_estatus_len(estatus);
1000 node_len = GHES_ESTATUS_NODE_LEN(len);
1001 task_work_pending = ghes_do_proc(estatus_node->ghes, estatus);
1002 if (!ghes_estatus_cached(estatus)) {
1003 generic = estatus_node->generic;
1004 if (ghes_print_estatus(NULL, generic, estatus))
1005 ghes_estatus_cache_add(generic, estatus);
1008 if (task_work_pending && current->mm) {
1009 estatus_node->task_work.func = ghes_kick_task_work;
1010 estatus_node->task_work_cpu = smp_processor_id();
1011 ret = task_work_add(current, &estatus_node->task_work,
1014 estatus_node->task_work.func = NULL;
1017 if (!estatus_node->task_work.func)
1018 gen_pool_free(ghes_estatus_pool,
1019 (unsigned long)estatus_node, node_len);
1025 static void ghes_print_queued_estatus(void)
1027 struct llist_node *llnode;
1028 struct ghes_estatus_node *estatus_node;
1029 struct acpi_hest_generic *generic;
1030 struct acpi_hest_generic_status *estatus;
1032 llnode = llist_del_all(&ghes_estatus_llist);
1034 * Because the time order of estatus in list is reversed,
1035 * revert it back to proper order.
1037 llnode = llist_reverse_order(llnode);
1039 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
1041 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
1042 generic = estatus_node->generic;
1043 ghes_print_estatus(NULL, generic, estatus);
1044 llnode = llnode->next;
1048 static int ghes_in_nmi_queue_one_entry(struct ghes *ghes,
1049 enum fixed_addresses fixmap_idx)
1051 struct acpi_hest_generic_status *estatus, tmp_header;
1052 struct ghes_estatus_node *estatus_node;
1057 if (!IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG))
1060 rc = __ghes_peek_estatus(ghes, &tmp_header, &buf_paddr, fixmap_idx);
1062 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1066 rc = __ghes_check_estatus(ghes, &tmp_header);
1068 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1072 len = cper_estatus_len(&tmp_header);
1073 node_len = GHES_ESTATUS_NODE_LEN(len);
1074 estatus_node = (void *)gen_pool_alloc(ghes_estatus_pool, node_len);
1078 estatus_node->ghes = ghes;
1079 estatus_node->generic = ghes->generic;
1080 estatus_node->task_work.func = NULL;
1081 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
1083 if (__ghes_read_estatus(estatus, buf_paddr, fixmap_idx, len)) {
1084 ghes_clear_estatus(ghes, estatus, buf_paddr, fixmap_idx);
1089 sev = ghes_severity(estatus->error_severity);
1090 if (sev >= GHES_SEV_PANIC) {
1091 ghes_print_queued_estatus();
1092 __ghes_panic(ghes, estatus, buf_paddr, fixmap_idx);
1095 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1097 /* This error has been reported before, don't process it again. */
1098 if (ghes_estatus_cached(estatus))
1101 llist_add(&estatus_node->llnode, &ghes_estatus_llist);
1106 gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node,
1112 static int ghes_in_nmi_spool_from_list(struct list_head *rcu_list,
1113 enum fixed_addresses fixmap_idx)
1119 list_for_each_entry_rcu(ghes, rcu_list, list) {
1120 if (!ghes_in_nmi_queue_one_entry(ghes, fixmap_idx))
1125 if (IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG) && !ret)
1126 irq_work_queue(&ghes_proc_irq_work);
1131 #ifdef CONFIG_ACPI_APEI_SEA
1132 static LIST_HEAD(ghes_sea);
1135 * Return 0 only if one of the SEA error sources successfully reported an error
1136 * record sent from the firmware.
1138 int ghes_notify_sea(void)
1140 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sea);
1143 raw_spin_lock(&ghes_notify_lock_sea);
1144 rv = ghes_in_nmi_spool_from_list(&ghes_sea, FIX_APEI_GHES_SEA);
1145 raw_spin_unlock(&ghes_notify_lock_sea);
1150 static void ghes_sea_add(struct ghes *ghes)
1152 mutex_lock(&ghes_list_mutex);
1153 list_add_rcu(&ghes->list, &ghes_sea);
1154 mutex_unlock(&ghes_list_mutex);
1157 static void ghes_sea_remove(struct ghes *ghes)
1159 mutex_lock(&ghes_list_mutex);
1160 list_del_rcu(&ghes->list);
1161 mutex_unlock(&ghes_list_mutex);
1164 #else /* CONFIG_ACPI_APEI_SEA */
1165 static inline void ghes_sea_add(struct ghes *ghes) { }
1166 static inline void ghes_sea_remove(struct ghes *ghes) { }
1167 #endif /* CONFIG_ACPI_APEI_SEA */
1169 #ifdef CONFIG_HAVE_ACPI_APEI_NMI
1171 * NMI may be triggered on any CPU, so ghes_in_nmi is used for
1172 * having only one concurrent reader.
1174 static atomic_t ghes_in_nmi = ATOMIC_INIT(0);
1176 static LIST_HEAD(ghes_nmi);
1178 static int ghes_notify_nmi(unsigned int cmd, struct pt_regs *regs)
1180 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_nmi);
1183 if (!atomic_add_unless(&ghes_in_nmi, 1, 1))
1186 raw_spin_lock(&ghes_notify_lock_nmi);
1187 if (!ghes_in_nmi_spool_from_list(&ghes_nmi, FIX_APEI_GHES_NMI))
1189 raw_spin_unlock(&ghes_notify_lock_nmi);
1191 atomic_dec(&ghes_in_nmi);
1195 static void ghes_nmi_add(struct ghes *ghes)
1197 mutex_lock(&ghes_list_mutex);
1198 if (list_empty(&ghes_nmi))
1199 register_nmi_handler(NMI_LOCAL, ghes_notify_nmi, 0, "ghes");
1200 list_add_rcu(&ghes->list, &ghes_nmi);
1201 mutex_unlock(&ghes_list_mutex);
1204 static void ghes_nmi_remove(struct ghes *ghes)
1206 mutex_lock(&ghes_list_mutex);
1207 list_del_rcu(&ghes->list);
1208 if (list_empty(&ghes_nmi))
1209 unregister_nmi_handler(NMI_LOCAL, "ghes");
1210 mutex_unlock(&ghes_list_mutex);
1212 * To synchronize with NMI handler, ghes can only be
1213 * freed after NMI handler finishes.
1217 #else /* CONFIG_HAVE_ACPI_APEI_NMI */
1218 static inline void ghes_nmi_add(struct ghes *ghes) { }
1219 static inline void ghes_nmi_remove(struct ghes *ghes) { }
1220 #endif /* CONFIG_HAVE_ACPI_APEI_NMI */
1222 static void ghes_nmi_init_cxt(void)
1224 init_irq_work(&ghes_proc_irq_work, ghes_proc_in_irq);
1227 static int __ghes_sdei_callback(struct ghes *ghes,
1228 enum fixed_addresses fixmap_idx)
1230 if (!ghes_in_nmi_queue_one_entry(ghes, fixmap_idx)) {
1231 irq_work_queue(&ghes_proc_irq_work);
1239 static int ghes_sdei_normal_callback(u32 event_num, struct pt_regs *regs,
1242 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sdei_normal);
1243 struct ghes *ghes = arg;
1246 raw_spin_lock(&ghes_notify_lock_sdei_normal);
1247 err = __ghes_sdei_callback(ghes, FIX_APEI_GHES_SDEI_NORMAL);
1248 raw_spin_unlock(&ghes_notify_lock_sdei_normal);
1253 static int ghes_sdei_critical_callback(u32 event_num, struct pt_regs *regs,
1256 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sdei_critical);
1257 struct ghes *ghes = arg;
1260 raw_spin_lock(&ghes_notify_lock_sdei_critical);
1261 err = __ghes_sdei_callback(ghes, FIX_APEI_GHES_SDEI_CRITICAL);
1262 raw_spin_unlock(&ghes_notify_lock_sdei_critical);
1267 static int apei_sdei_register_ghes(struct ghes *ghes)
1269 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE))
1272 return sdei_register_ghes(ghes, ghes_sdei_normal_callback,
1273 ghes_sdei_critical_callback);
1276 static int apei_sdei_unregister_ghes(struct ghes *ghes)
1278 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE))
1281 return sdei_unregister_ghes(ghes);
1284 static int ghes_probe(struct platform_device *ghes_dev)
1286 struct acpi_hest_generic *generic;
1287 struct ghes *ghes = NULL;
1288 unsigned long flags;
1292 generic = *(struct acpi_hest_generic **)ghes_dev->dev.platform_data;
1293 if (!generic->enabled)
1296 switch (generic->notify.type) {
1297 case ACPI_HEST_NOTIFY_POLLED:
1298 case ACPI_HEST_NOTIFY_EXTERNAL:
1299 case ACPI_HEST_NOTIFY_SCI:
1300 case ACPI_HEST_NOTIFY_GSIV:
1301 case ACPI_HEST_NOTIFY_GPIO:
1304 case ACPI_HEST_NOTIFY_SEA:
1305 if (!IS_ENABLED(CONFIG_ACPI_APEI_SEA)) {
1306 pr_warn(GHES_PFX "Generic hardware error source: %d notified via SEA is not supported\n",
1307 generic->header.source_id);
1312 case ACPI_HEST_NOTIFY_NMI:
1313 if (!IS_ENABLED(CONFIG_HAVE_ACPI_APEI_NMI)) {
1314 pr_warn(GHES_PFX "Generic hardware error source: %d notified via NMI interrupt is not supported!\n",
1315 generic->header.source_id);
1319 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1320 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE)) {
1321 pr_warn(GHES_PFX "Generic hardware error source: %d notified via SDE Interface is not supported!\n",
1322 generic->header.source_id);
1326 case ACPI_HEST_NOTIFY_LOCAL:
1327 pr_warn(GHES_PFX "Generic hardware error source: %d notified via local interrupt is not supported!\n",
1328 generic->header.source_id);
1331 pr_warn(FW_WARN GHES_PFX "Unknown notification type: %u for generic hardware error source: %d\n",
1332 generic->notify.type, generic->header.source_id);
1337 if (generic->error_block_length <
1338 sizeof(struct acpi_hest_generic_status)) {
1339 pr_warn(FW_BUG GHES_PFX "Invalid error block length: %u for generic hardware error source: %d\n",
1340 generic->error_block_length, generic->header.source_id);
1343 ghes = ghes_new(generic);
1350 switch (generic->notify.type) {
1351 case ACPI_HEST_NOTIFY_POLLED:
1352 timer_setup(&ghes->timer, ghes_poll_func, 0);
1353 ghes_add_timer(ghes);
1355 case ACPI_HEST_NOTIFY_EXTERNAL:
1356 /* External interrupt vector is GSI */
1357 rc = acpi_gsi_to_irq(generic->notify.vector, &ghes->irq);
1359 pr_err(GHES_PFX "Failed to map GSI to IRQ for generic hardware error source: %d\n",
1360 generic->header.source_id);
1363 rc = request_irq(ghes->irq, ghes_irq_func, IRQF_SHARED,
1366 pr_err(GHES_PFX "Failed to register IRQ for generic hardware error source: %d\n",
1367 generic->header.source_id);
1372 case ACPI_HEST_NOTIFY_SCI:
1373 case ACPI_HEST_NOTIFY_GSIV:
1374 case ACPI_HEST_NOTIFY_GPIO:
1375 mutex_lock(&ghes_list_mutex);
1376 if (list_empty(&ghes_hed))
1377 register_acpi_hed_notifier(&ghes_notifier_hed);
1378 list_add_rcu(&ghes->list, &ghes_hed);
1379 mutex_unlock(&ghes_list_mutex);
1382 case ACPI_HEST_NOTIFY_SEA:
1385 case ACPI_HEST_NOTIFY_NMI:
1388 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1389 rc = apei_sdei_register_ghes(ghes);
1397 platform_set_drvdata(ghes_dev, ghes);
1399 ghes->dev = &ghes_dev->dev;
1401 mutex_lock(&ghes_devs_mutex);
1402 list_add_tail(&ghes->elist, &ghes_devs);
1403 mutex_unlock(&ghes_devs_mutex);
1405 /* Handle any pending errors right away */
1406 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
1408 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
1420 static int ghes_remove(struct platform_device *ghes_dev)
1424 struct acpi_hest_generic *generic;
1426 ghes = platform_get_drvdata(ghes_dev);
1427 generic = ghes->generic;
1429 ghes->flags |= GHES_EXITING;
1430 switch (generic->notify.type) {
1431 case ACPI_HEST_NOTIFY_POLLED:
1432 timer_shutdown_sync(&ghes->timer);
1434 case ACPI_HEST_NOTIFY_EXTERNAL:
1435 free_irq(ghes->irq, ghes);
1438 case ACPI_HEST_NOTIFY_SCI:
1439 case ACPI_HEST_NOTIFY_GSIV:
1440 case ACPI_HEST_NOTIFY_GPIO:
1441 mutex_lock(&ghes_list_mutex);
1442 list_del_rcu(&ghes->list);
1443 if (list_empty(&ghes_hed))
1444 unregister_acpi_hed_notifier(&ghes_notifier_hed);
1445 mutex_unlock(&ghes_list_mutex);
1449 case ACPI_HEST_NOTIFY_SEA:
1450 ghes_sea_remove(ghes);
1452 case ACPI_HEST_NOTIFY_NMI:
1453 ghes_nmi_remove(ghes);
1455 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1456 rc = apei_sdei_unregister_ghes(ghes);
1467 mutex_lock(&ghes_devs_mutex);
1468 list_del(&ghes->elist);
1469 mutex_unlock(&ghes_devs_mutex);
1476 static struct platform_driver ghes_platform_driver = {
1480 .probe = ghes_probe,
1481 .remove = ghes_remove,
1484 void __init acpi_ghes_init(void)
1493 switch (hest_disable) {
1494 case HEST_NOT_FOUND:
1497 pr_info(GHES_PFX "HEST is not enabled!\n");
1504 pr_info(GHES_PFX "GHES is not enabled!\n");
1508 ghes_nmi_init_cxt();
1510 rc = platform_driver_register(&ghes_platform_driver);
1514 rc = apei_osc_setup();
1515 if (rc == 0 && osc_sb_apei_support_acked)
1516 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit and WHEA _OSC.\n");
1517 else if (rc == 0 && !osc_sb_apei_support_acked)
1518 pr_info(GHES_PFX "APEI firmware first mode is enabled by WHEA _OSC.\n");
1519 else if (rc && osc_sb_apei_support_acked)
1520 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit.\n");
1522 pr_info(GHES_PFX "Failed to enable APEI firmware first mode.\n");
1526 * Known x86 systems that prefer GHES error reporting:
1528 static struct acpi_platform_list plat_list[] = {
1529 {"HPE ", "Server ", 0, ACPI_SIG_FADT, all_versions},
1533 struct list_head *ghes_get_devices(void)
1537 if (IS_ENABLED(CONFIG_X86)) {
1538 idx = acpi_match_platform_list(plat_list);
1540 if (!ghes_edac_force_enable)
1543 pr_warn_once("Force-loading ghes_edac on an unsupported platform. You're on your own!\n");
1545 } else if (list_empty(&ghes_devs)) {
1551 EXPORT_SYMBOL_GPL(ghes_get_devices);
1553 void ghes_register_report_chain(struct notifier_block *nb)
1555 atomic_notifier_chain_register(&ghes_report_chain, nb);
1557 EXPORT_SYMBOL_GPL(ghes_register_report_chain);
1559 void ghes_unregister_report_chain(struct notifier_block *nb)
1561 atomic_notifier_chain_unregister(&ghes_report_chain, nb);
1563 EXPORT_SYMBOL_GPL(ghes_unregister_report_chain);