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 inline bool is_hest_type_generic_v2(struct ghes *ghes)
99 return ghes->generic->header.type == ACPI_HEST_TYPE_GENERIC_ERROR_V2;
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.
108 module_param_named(disable, ghes_disable, bool, 0);
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.
115 * RCU is used for these lists, so ghes_list_mutex is only used for
116 * list changing, not for traversing.
118 static LIST_HEAD(ghes_hed);
119 static DEFINE_MUTEX(ghes_list_mutex);
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.
127 * This spinlock is used to prevent the fixmap entry from being used
130 static DEFINE_SPINLOCK(ghes_notify_lock_irq);
132 struct ghes_vendor_record_entry {
133 struct work_struct work;
135 char vendor_record[];
138 static struct gen_pool *ghes_estatus_pool;
139 static unsigned long ghes_estatus_pool_size_request;
141 static struct ghes_estatus_cache *ghes_estatus_caches[GHES_ESTATUS_CACHES_SIZE];
142 static atomic_t ghes_estatus_cache_alloced;
144 static int ghes_panic_timeout __read_mostly = 30;
146 static void __iomem *ghes_map(u64 pfn, enum fixed_addresses fixmap_idx)
151 paddr = PFN_PHYS(pfn);
152 prot = arch_apei_get_mem_attribute(paddr);
153 __set_fixmap(fixmap_idx, paddr, prot);
155 return (void __iomem *) __fix_to_virt(fixmap_idx);
158 static void ghes_unmap(void __iomem *vaddr, enum fixed_addresses fixmap_idx)
160 int _idx = virt_to_fix((unsigned long)vaddr);
162 WARN_ON_ONCE(fixmap_idx != _idx);
163 clear_fixmap(fixmap_idx);
166 int ghes_estatus_pool_init(int num_ghes)
168 unsigned long addr, len;
171 ghes_estatus_pool = gen_pool_create(GHES_ESTATUS_POOL_MIN_ALLOC_ORDER, -1);
172 if (!ghes_estatus_pool)
175 len = GHES_ESTATUS_CACHE_AVG_SIZE * GHES_ESTATUS_CACHE_ALLOCED_MAX;
176 len += (num_ghes * GHES_ESOURCE_PREALLOC_MAX_SIZE);
178 ghes_estatus_pool_size_request = PAGE_ALIGN(len);
179 addr = (unsigned long)vmalloc(PAGE_ALIGN(len));
183 rc = gen_pool_add(ghes_estatus_pool, addr, PAGE_ALIGN(len), -1);
193 gen_pool_destroy(ghes_estatus_pool);
198 static int map_gen_v2(struct ghes *ghes)
200 return apei_map_generic_address(&ghes->generic_v2->read_ack_register);
203 static void unmap_gen_v2(struct ghes *ghes)
205 apei_unmap_generic_address(&ghes->generic_v2->read_ack_register);
208 static void ghes_ack_error(struct acpi_hest_generic_v2 *gv2)
213 rc = apei_read(&val, &gv2->read_ack_register);
217 val &= gv2->read_ack_preserve << gv2->read_ack_register.bit_offset;
218 val |= gv2->read_ack_write << gv2->read_ack_register.bit_offset;
220 apei_write(val, &gv2->read_ack_register);
223 static struct ghes *ghes_new(struct acpi_hest_generic *generic)
226 unsigned int error_block_length;
229 ghes = kzalloc(sizeof(*ghes), GFP_KERNEL);
231 return ERR_PTR(-ENOMEM);
233 ghes->generic = generic;
234 if (is_hest_type_generic_v2(ghes)) {
235 rc = map_gen_v2(ghes);
240 rc = apei_map_generic_address(&generic->error_status_address);
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;
251 ghes->estatus = kmalloc(error_block_length, GFP_KERNEL);
252 if (!ghes->estatus) {
254 goto err_unmap_status_addr;
259 err_unmap_status_addr:
260 apei_unmap_generic_address(&generic->error_status_address);
261 err_unmap_read_ack_addr:
262 if (is_hest_type_generic_v2(ghes))
269 static void ghes_fini(struct ghes *ghes)
271 kfree(ghes->estatus);
272 apei_unmap_generic_address(&ghes->generic->error_status_address);
273 if (is_hest_type_generic_v2(ghes))
277 static inline int ghes_severity(int severity)
280 case CPER_SEV_INFORMATIONAL:
282 case CPER_SEV_CORRECTED:
283 return GHES_SEV_CORRECTED;
284 case CPER_SEV_RECOVERABLE:
285 return GHES_SEV_RECOVERABLE;
287 return GHES_SEV_PANIC;
289 /* Unknown, go panic */
290 return GHES_SEV_PANIC;
294 static void ghes_copy_tofrom_phys(void *buffer, u64 paddr, u32 len,
296 enum fixed_addresses fixmap_idx)
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);
308 memcpy_fromio(buffer, vaddr + offset, trunk);
310 memcpy_toio(vaddr + offset, buffer, trunk);
314 ghes_unmap(vaddr, fixmap_idx);
318 /* Check the top-level record header has an appropriate size. */
319 static int __ghes_check_estatus(struct ghes *ghes,
320 struct acpi_hest_generic_status *estatus)
322 u32 len = cper_estatus_len(estatus);
324 if (len < sizeof(*estatus)) {
325 pr_warn_ratelimited(FW_WARN GHES_PFX "Truncated error status block!\n");
329 if (len > ghes->generic->error_block_length) {
330 pr_warn_ratelimited(FW_WARN GHES_PFX "Invalid error status block length!\n");
334 if (cper_estatus_check_header(estatus)) {
335 pr_warn_ratelimited(FW_WARN GHES_PFX "Invalid CPER header!\n");
342 /* Read the CPER block, returning its address, and header in estatus. */
343 static int __ghes_peek_estatus(struct ghes *ghes,
344 struct acpi_hest_generic_status *estatus,
345 u64 *buf_paddr, enum fixed_addresses fixmap_idx)
347 struct acpi_hest_generic *g = ghes->generic;
350 rc = apei_read(buf_paddr, &g->error_status_address);
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);
361 ghes_copy_tofrom_phys(estatus, *buf_paddr, sizeof(*estatus), 1,
363 if (!estatus->block_status) {
371 static int __ghes_read_estatus(struct acpi_hest_generic_status *estatus,
372 u64 buf_paddr, enum fixed_addresses fixmap_idx,
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");
385 static int ghes_read_estatus(struct ghes *ghes,
386 struct acpi_hest_generic_status *estatus,
387 u64 *buf_paddr, enum fixed_addresses fixmap_idx)
391 rc = __ghes_peek_estatus(ghes, estatus, buf_paddr, fixmap_idx);
395 rc = __ghes_check_estatus(ghes, estatus);
399 return __ghes_read_estatus(estatus, *buf_paddr, fixmap_idx,
400 cper_estatus_len(estatus));
403 static void ghes_clear_estatus(struct ghes *ghes,
404 struct acpi_hest_generic_status *estatus,
405 u64 buf_paddr, enum fixed_addresses fixmap_idx)
407 estatus->block_status = 0;
412 ghes_copy_tofrom_phys(estatus, buf_paddr,
413 sizeof(estatus->block_status), 0,
417 * GHESv2 type HEST entries introduce support for error acknowledgment,
418 * so only acknowledge the error if this support is present.
420 if (is_hest_type_generic_v2(ghes))
421 ghes_ack_error(ghes->generic_v2);
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.
429 static void ghes_kick_task_work(struct callback_head *head)
431 struct acpi_hest_generic_status *estatus;
432 struct ghes_estatus_node *estatus_node;
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);
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);
444 static bool ghes_do_memory_failure(u64 physical_addr, int flags)
448 if (!IS_ENABLED(CONFIG_ACPI_APEI_MEMORY_FAILURE))
451 pfn = PHYS_PFN(physical_addr);
452 if (!pfn_valid(pfn) && !arch_is_platform_page(physical_addr)) {
453 pr_warn_ratelimited(FW_WARN GHES_PFX
454 "Invalid address in generic error data: %#llx\n",
459 memory_failure_queue(pfn, flags);
463 static bool ghes_handle_memory_failure(struct acpi_hest_generic_data *gdata,
467 int sec_sev = ghes_severity(gdata->error_severity);
468 struct cper_sec_mem_err *mem_err = acpi_hest_get_payload(gdata);
470 if (!(mem_err->validation_bits & CPER_MEM_VALID_PA))
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)
481 return ghes_do_memory_failure(mem_err->physical_addr, flags);
486 static bool ghes_handle_arm_hw_error(struct acpi_hest_generic_data *gdata, int sev)
488 struct cper_sec_proc_arm *err = acpi_hest_get_payload(gdata);
493 log_arm_hw_error(err);
495 sec_sev = ghes_severity(gdata->error_severity);
496 if (sev != GHES_SEV_RECOVERABLE || sec_sev != GHES_SEV_RECOVERABLE)
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";
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.
512 if (is_cache && has_pa) {
513 queued = ghes_do_memory_failure(err_info->physical_fault_addr, 0);
514 p += err_info->length;
518 if (err_info->type < ARRAY_SIZE(cper_proc_error_type_strs))
519 error_type = cper_proc_error_type_strs[err_info->type];
521 pr_warn_ratelimited(FW_WARN GHES_PFX
522 "Unhandled processor error type: %s\n",
524 p += err_info->length;
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:
535 * GHES_SEV_CORRECTABLE -> AER_CORRECTABLE
536 * These need to be reported by the AER driver but no recovery is
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
544 static void ghes_handle_aer(struct acpi_hest_generic_data *gdata)
546 #ifdef CONFIG_ACPI_APEI_PCIEAER
547 struct cper_sec_pcie *pcie_err = acpi_hest_get_payload(gdata);
549 if (pcie_err->validation_bits & CPER_PCIE_VALID_DEVICE_ID &&
550 pcie_err->validation_bits & CPER_PCIE_VALID_AER_INFO) {
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);
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.
563 if (gdata->flags & CPER_SEC_RESET)
564 aer_severity = AER_FATAL;
566 aer_recover_queue(pcie_err->device_id.segment,
567 pcie_err->device_id.bus,
569 (struct aer_capability_regs *)
575 static BLOCKING_NOTIFIER_HEAD(vendor_record_notify_list);
577 int ghes_register_vendor_record_notifier(struct notifier_block *nb)
579 return blocking_notifier_chain_register(&vendor_record_notify_list, nb);
581 EXPORT_SYMBOL_GPL(ghes_register_vendor_record_notifier);
583 void ghes_unregister_vendor_record_notifier(struct notifier_block *nb)
585 blocking_notifier_chain_unregister(&vendor_record_notify_list, nb);
587 EXPORT_SYMBOL_GPL(ghes_unregister_vendor_record_notifier);
589 static void ghes_vendor_record_work_func(struct work_struct *work)
591 struct ghes_vendor_record_entry *entry;
592 struct acpi_hest_generic_data *gdata;
595 entry = container_of(work, struct ghes_vendor_record_entry, work);
596 gdata = GHES_GDATA_FROM_VENDOR_ENTRY(entry);
598 blocking_notifier_call_chain(&vendor_record_notify_list,
599 entry->error_severity, gdata);
601 len = GHES_VENDOR_ENTRY_LEN(acpi_hest_get_record_size(gdata));
602 gen_pool_free(ghes_estatus_pool, (unsigned long)entry, len);
605 static void ghes_defer_non_standard_event(struct acpi_hest_generic_data *gdata,
608 struct acpi_hest_generic_data *copied_gdata;
609 struct ghes_vendor_record_entry *entry;
612 len = GHES_VENDOR_ENTRY_LEN(acpi_hest_get_record_size(gdata));
613 entry = (void *)gen_pool_alloc(ghes_estatus_pool, len);
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;
621 INIT_WORK(&entry->work, ghes_vendor_record_work_func);
622 schedule_work(&entry->work);
625 static bool ghes_do_proc(struct ghes *ghes,
626 const struct acpi_hest_generic_status *estatus)
629 struct acpi_hest_generic_data *gdata;
631 const guid_t *fru_id = &guid_null;
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;
642 if (gdata->validation_bits & CPER_SEC_VALID_FRU_TEXT)
643 fru_text = gdata->fru_text;
645 if (guid_equal(sec_type, &CPER_SEC_PLATFORM_MEM)) {
646 struct cper_sec_mem_err *mem_err = acpi_hest_get_payload(gdata);
648 ghes_edac_report_mem_error(sev, mem_err);
650 arch_apei_report_mem_error(sev, mem_err);
651 queued = ghes_handle_memory_failure(gdata, sev);
653 else if (guid_equal(sec_type, &CPER_SEC_PCIE)) {
654 ghes_handle_aer(gdata);
656 else if (guid_equal(sec_type, &CPER_SEC_PROC_ARM)) {
657 queued = ghes_handle_arm_hw_error(gdata, sev);
659 void *err = acpi_hest_get_payload(gdata);
661 ghes_defer_non_standard_event(gdata, sev);
662 log_non_standard_event(sec_type, fru_id, fru_text,
664 gdata->error_data_length);
671 static void __ghes_print_estatus(const char *pfx,
672 const struct acpi_hest_generic *generic,
673 const struct acpi_hest_generic_status *estatus)
675 static atomic_t seqno;
676 unsigned int curr_seqno;
680 if (ghes_severity(estatus->error_severity) <=
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);
693 static int ghes_print_estatus(const char *pfx,
694 const struct acpi_hest_generic *generic,
695 const struct acpi_hest_generic_status *estatus)
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;
702 if (ghes_severity(estatus->error_severity) <= GHES_SEV_CORRECTED)
703 ratelimit = &ratelimit_corrected;
705 ratelimit = &ratelimit_uncorrected;
706 if (__ratelimit(ratelimit)) {
707 __ghes_print_estatus(pfx, generic, estatus);
714 * GHES error status reporting throttle, to report more kinds of
715 * errors, instead of just most frequently occurred errors.
717 static int ghes_estatus_cached(struct acpi_hest_generic_status *estatus)
721 unsigned long long now;
722 struct ghes_estatus_cache *cache;
723 struct acpi_hest_generic_status *cache_estatus;
725 len = cper_estatus_len(estatus);
727 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
728 cache = rcu_dereference(ghes_estatus_caches[i]);
731 if (len != cache->estatus_len)
733 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
734 if (memcmp(estatus, cache_estatus, len))
736 atomic_inc(&cache->count);
738 if (now - cache->time_in < GHES_ESTATUS_IN_CACHE_MAX_NSEC)
746 static struct ghes_estatus_cache *ghes_estatus_cache_alloc(
747 struct acpi_hest_generic *generic,
748 struct acpi_hest_generic_status *estatus)
752 struct ghes_estatus_cache *cache;
753 struct acpi_hest_generic_status *cache_estatus;
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);
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);
764 atomic_dec(&ghes_estatus_cache_alloced);
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();
776 static void ghes_estatus_cache_free(struct ghes_estatus_cache *cache)
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);
786 static void ghes_estatus_cache_rcu_free(struct rcu_head *head)
788 struct ghes_estatus_cache *cache;
790 cache = container_of(head, struct ghes_estatus_cache, rcu);
791 ghes_estatus_cache_free(cache);
794 static void ghes_estatus_cache_add(
795 struct acpi_hest_generic *generic,
796 struct acpi_hest_generic_status *estatus)
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;
802 new_cache = ghes_estatus_cache_alloc(generic, estatus);
803 if (new_cache == NULL)
807 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
808 cache = rcu_dereference(ghes_estatus_caches[i]);
814 duration = now - cache->time_in;
815 if (duration >= GHES_ESTATUS_IN_CACHE_MAX_NSEC) {
820 count = atomic_read(&cache->count);
822 do_div(period, (count + 1));
823 if (period > max_period) {
829 /* new_cache must be put into array after its contents are written */
831 if (slot != -1 && cmpxchg(ghes_estatus_caches + slot,
832 slot_cache, new_cache) == slot_cache) {
834 call_rcu(&slot_cache->rcu, ghes_estatus_cache_rcu_free);
836 ghes_estatus_cache_free(new_cache);
840 static void __ghes_panic(struct ghes *ghes,
841 struct acpi_hest_generic_status *estatus,
842 u64 buf_paddr, enum fixed_addresses fixmap_idx)
844 __ghes_print_estatus(KERN_EMERG, ghes->generic, estatus);
846 ghes_clear_estatus(ghes, estatus, buf_paddr, fixmap_idx);
848 /* reboot to log the error! */
850 panic_timeout = ghes_panic_timeout;
851 panic("Fatal hardware error!");
854 static int ghes_proc(struct ghes *ghes)
856 struct acpi_hest_generic_status *estatus = ghes->estatus;
860 rc = ghes_read_estatus(ghes, estatus, &buf_paddr, FIX_APEI_GHES_IRQ);
864 if (ghes_severity(estatus->error_severity) >= GHES_SEV_PANIC)
865 __ghes_panic(ghes, estatus, buf_paddr, FIX_APEI_GHES_IRQ);
867 if (!ghes_estatus_cached(estatus)) {
868 if (ghes_print_estatus(NULL, ghes->generic, estatus))
869 ghes_estatus_cache_add(ghes->generic, estatus);
871 ghes_do_proc(ghes, estatus);
874 ghes_clear_estatus(ghes, estatus, buf_paddr, FIX_APEI_GHES_IRQ);
879 static void ghes_add_timer(struct ghes *ghes)
881 struct acpi_hest_generic *g = ghes->generic;
882 unsigned long expire;
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);
889 expire = jiffies + msecs_to_jiffies(g->notify.poll_interval);
890 ghes->timer.expires = round_jiffies_relative(expire);
891 add_timer(&ghes->timer);
894 static void ghes_poll_func(struct timer_list *t)
896 struct ghes *ghes = from_timer(ghes, t, timer);
899 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
901 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
902 if (!(ghes->flags & GHES_EXITING))
903 ghes_add_timer(ghes);
906 static irqreturn_t ghes_irq_func(int irq, void *data)
908 struct ghes *ghes = data;
912 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
913 rc = ghes_proc(ghes);
914 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
921 static int ghes_notify_hed(struct notifier_block *this, unsigned long event,
926 int ret = NOTIFY_DONE;
928 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
930 list_for_each_entry_rcu(ghes, &ghes_hed, list) {
931 if (!ghes_proc(ghes))
935 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
940 static struct notifier_block ghes_notifier_hed = {
941 .notifier_call = ghes_notify_hed,
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.
952 * Memory from the ghes_estatus_pool is also used with the ghes_estatus_cache
953 * to suppress frequent messages.
955 static struct llist_head ghes_estatus_llist;
956 static struct irq_work ghes_proc_irq_work;
958 static void ghes_proc_in_irq(struct irq_work *irq_work)
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;
968 llnode = llist_del_all(&ghes_estatus_llist);
970 * Because the time order of estatus in list is reversed,
971 * revert it back to proper order.
973 llnode = llist_reverse_order(llnode);
976 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
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);
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,
994 estatus_node->task_work.func = NULL;
997 if (!estatus_node->task_work.func)
998 gen_pool_free(ghes_estatus_pool,
999 (unsigned long)estatus_node, node_len);
1005 static void ghes_print_queued_estatus(void)
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;
1012 llnode = llist_del_all(&ghes_estatus_llist);
1014 * Because the time order of estatus in list is reversed,
1015 * revert it back to proper order.
1017 llnode = llist_reverse_order(llnode);
1019 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
1021 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
1022 generic = estatus_node->generic;
1023 ghes_print_estatus(NULL, generic, estatus);
1024 llnode = llnode->next;
1028 static int ghes_in_nmi_queue_one_entry(struct ghes *ghes,
1029 enum fixed_addresses fixmap_idx)
1031 struct acpi_hest_generic_status *estatus, tmp_header;
1032 struct ghes_estatus_node *estatus_node;
1037 if (!IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG))
1040 rc = __ghes_peek_estatus(ghes, &tmp_header, &buf_paddr, fixmap_idx);
1042 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1046 rc = __ghes_check_estatus(ghes, &tmp_header);
1048 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
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);
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);
1063 if (__ghes_read_estatus(estatus, buf_paddr, fixmap_idx, len)) {
1064 ghes_clear_estatus(ghes, estatus, buf_paddr, fixmap_idx);
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);
1075 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1077 /* This error has been reported before, don't process it again. */
1078 if (ghes_estatus_cached(estatus))
1081 llist_add(&estatus_node->llnode, &ghes_estatus_llist);
1086 gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node,
1092 static int ghes_in_nmi_spool_from_list(struct list_head *rcu_list,
1093 enum fixed_addresses fixmap_idx)
1099 list_for_each_entry_rcu(ghes, rcu_list, list) {
1100 if (!ghes_in_nmi_queue_one_entry(ghes, fixmap_idx))
1105 if (IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG) && !ret)
1106 irq_work_queue(&ghes_proc_irq_work);
1111 #ifdef CONFIG_ACPI_APEI_SEA
1112 static LIST_HEAD(ghes_sea);
1115 * Return 0 only if one of the SEA error sources successfully reported an error
1116 * record sent from the firmware.
1118 int ghes_notify_sea(void)
1120 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sea);
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);
1130 static void ghes_sea_add(struct ghes *ghes)
1132 mutex_lock(&ghes_list_mutex);
1133 list_add_rcu(&ghes->list, &ghes_sea);
1134 mutex_unlock(&ghes_list_mutex);
1137 static void ghes_sea_remove(struct ghes *ghes)
1139 mutex_lock(&ghes_list_mutex);
1140 list_del_rcu(&ghes->list);
1141 mutex_unlock(&ghes_list_mutex);
1144 #else /* CONFIG_ACPI_APEI_SEA */
1145 static inline void ghes_sea_add(struct ghes *ghes) { }
1146 static inline void ghes_sea_remove(struct ghes *ghes) { }
1147 #endif /* CONFIG_ACPI_APEI_SEA */
1149 #ifdef CONFIG_HAVE_ACPI_APEI_NMI
1151 * NMI may be triggered on any CPU, so ghes_in_nmi is used for
1152 * having only one concurrent reader.
1154 static atomic_t ghes_in_nmi = ATOMIC_INIT(0);
1156 static LIST_HEAD(ghes_nmi);
1158 static int ghes_notify_nmi(unsigned int cmd, struct pt_regs *regs)
1160 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_nmi);
1163 if (!atomic_add_unless(&ghes_in_nmi, 1, 1))
1166 raw_spin_lock(&ghes_notify_lock_nmi);
1167 if (!ghes_in_nmi_spool_from_list(&ghes_nmi, FIX_APEI_GHES_NMI))
1169 raw_spin_unlock(&ghes_notify_lock_nmi);
1171 atomic_dec(&ghes_in_nmi);
1175 static void ghes_nmi_add(struct ghes *ghes)
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);
1184 static void ghes_nmi_remove(struct ghes *ghes)
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);
1192 * To synchronize with NMI handler, ghes can only be
1193 * freed after NMI handler finishes.
1197 #else /* CONFIG_HAVE_ACPI_APEI_NMI */
1198 static inline void ghes_nmi_add(struct ghes *ghes) { }
1199 static inline void ghes_nmi_remove(struct ghes *ghes) { }
1200 #endif /* CONFIG_HAVE_ACPI_APEI_NMI */
1202 static void ghes_nmi_init_cxt(void)
1204 init_irq_work(&ghes_proc_irq_work, ghes_proc_in_irq);
1207 static int __ghes_sdei_callback(struct ghes *ghes,
1208 enum fixed_addresses fixmap_idx)
1210 if (!ghes_in_nmi_queue_one_entry(ghes, fixmap_idx)) {
1211 irq_work_queue(&ghes_proc_irq_work);
1219 static int ghes_sdei_normal_callback(u32 event_num, struct pt_regs *regs,
1222 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sdei_normal);
1223 struct ghes *ghes = arg;
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);
1233 static int ghes_sdei_critical_callback(u32 event_num, struct pt_regs *regs,
1236 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sdei_critical);
1237 struct ghes *ghes = arg;
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);
1247 static int apei_sdei_register_ghes(struct ghes *ghes)
1249 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE))
1252 return sdei_register_ghes(ghes, ghes_sdei_normal_callback,
1253 ghes_sdei_critical_callback);
1256 static int apei_sdei_unregister_ghes(struct ghes *ghes)
1258 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE))
1261 return sdei_unregister_ghes(ghes);
1264 static int ghes_probe(struct platform_device *ghes_dev)
1266 struct acpi_hest_generic *generic;
1267 struct ghes *ghes = NULL;
1268 unsigned long flags;
1272 generic = *(struct acpi_hest_generic **)ghes_dev->dev.platform_data;
1273 if (!generic->enabled)
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:
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);
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);
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);
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);
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);
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);
1323 ghes = ghes_new(generic);
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);
1335 case ACPI_HEST_NOTIFY_EXTERNAL:
1336 /* External interrupt vector is GSI */
1337 rc = acpi_gsi_to_irq(generic->notify.vector, &ghes->irq);
1339 pr_err(GHES_PFX "Failed to map GSI to IRQ for generic hardware error source: %d\n",
1340 generic->header.source_id);
1343 rc = request_irq(ghes->irq, ghes_irq_func, IRQF_SHARED,
1346 pr_err(GHES_PFX "Failed to register IRQ for generic hardware error source: %d\n",
1347 generic->header.source_id);
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);
1362 case ACPI_HEST_NOTIFY_SEA:
1365 case ACPI_HEST_NOTIFY_NMI:
1368 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1369 rc = apei_sdei_register_ghes(ghes);
1377 platform_set_drvdata(ghes_dev, ghes);
1379 ghes_edac_register(ghes, &ghes_dev->dev);
1381 /* Handle any pending errors right away */
1382 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
1384 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
1396 static int ghes_remove(struct platform_device *ghes_dev)
1400 struct acpi_hest_generic *generic;
1402 ghes = platform_get_drvdata(ghes_dev);
1403 generic = ghes->generic;
1405 ghes->flags |= GHES_EXITING;
1406 switch (generic->notify.type) {
1407 case ACPI_HEST_NOTIFY_POLLED:
1408 del_timer_sync(&ghes->timer);
1410 case ACPI_HEST_NOTIFY_EXTERNAL:
1411 free_irq(ghes->irq, ghes);
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);
1425 case ACPI_HEST_NOTIFY_SEA:
1426 ghes_sea_remove(ghes);
1428 case ACPI_HEST_NOTIFY_NMI:
1429 ghes_nmi_remove(ghes);
1431 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1432 rc = apei_sdei_unregister_ghes(ghes);
1443 ghes_edac_unregister(ghes);
1447 platform_set_drvdata(ghes_dev, NULL);
1452 static struct platform_driver ghes_platform_driver = {
1456 .probe = ghes_probe,
1457 .remove = ghes_remove,
1460 void __init acpi_ghes_init(void)
1469 switch (hest_disable) {
1470 case HEST_NOT_FOUND:
1473 pr_info(GHES_PFX "HEST is not enabled!\n");
1480 pr_info(GHES_PFX "GHES is not enabled!\n");
1484 ghes_nmi_init_cxt();
1486 rc = platform_driver_register(&ghes_platform_driver);
1490 rc = apei_osc_setup();
1491 if (rc == 0 && osc_sb_apei_support_acked)
1492 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit and WHEA _OSC.\n");
1493 else if (rc == 0 && !osc_sb_apei_support_acked)
1494 pr_info(GHES_PFX "APEI firmware first mode is enabled by WHEA _OSC.\n");
1495 else if (rc && osc_sb_apei_support_acked)
1496 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit.\n");
1498 pr_info(GHES_PFX "Failed to enable APEI firmware first mode.\n");