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 * A platform may describe one error source for the handling of synchronous
104 * errors (e.g. MCE or SEA), or for handling asynchronous errors (e.g. SCI
105 * or External Interrupt). On x86, the HEST notifications are always
106 * asynchronous, so only SEA on ARM is delivered as a synchronous
109 static inline bool is_hest_sync_notify(struct ghes *ghes)
111 u8 notify_type = ghes->generic->notify.type;
113 return notify_type == ACPI_HEST_NOTIFY_SEA;
117 * This driver isn't really modular, however for the time being,
118 * continuing to use module_param is the easiest way to remain
119 * compatible with existing boot arg use cases.
122 module_param_named(disable, ghes_disable, bool, 0);
125 * All error sources notified with HED (Hardware Error Device) share a
126 * single notifier callback, so they need to be linked and checked one
127 * by one. This holds true for NMI too.
129 * RCU is used for these lists, so ghes_list_mutex is only used for
130 * list changing, not for traversing.
132 static LIST_HEAD(ghes_hed);
133 static DEFINE_MUTEX(ghes_list_mutex);
136 * Because the memory area used to transfer hardware error information
137 * from BIOS to Linux can be determined only in NMI, IRQ or timer
138 * handler, but general ioremap can not be used in atomic context, so
139 * the fixmap is used instead.
141 * This spinlock is used to prevent the fixmap entry from being used
144 static DEFINE_SPINLOCK(ghes_notify_lock_irq);
146 struct ghes_vendor_record_entry {
147 struct work_struct work;
149 char vendor_record[];
152 static struct gen_pool *ghes_estatus_pool;
153 static unsigned long ghes_estatus_pool_size_request;
155 static struct ghes_estatus_cache *ghes_estatus_caches[GHES_ESTATUS_CACHES_SIZE];
156 static atomic_t ghes_estatus_cache_alloced;
158 static int ghes_panic_timeout __read_mostly = 30;
160 static void __iomem *ghes_map(u64 pfn, enum fixed_addresses fixmap_idx)
165 paddr = PFN_PHYS(pfn);
166 prot = arch_apei_get_mem_attribute(paddr);
167 __set_fixmap(fixmap_idx, paddr, prot);
169 return (void __iomem *) __fix_to_virt(fixmap_idx);
172 static void ghes_unmap(void __iomem *vaddr, enum fixed_addresses fixmap_idx)
174 int _idx = virt_to_fix((unsigned long)vaddr);
176 WARN_ON_ONCE(fixmap_idx != _idx);
177 clear_fixmap(fixmap_idx);
180 int ghes_estatus_pool_init(unsigned int num_ghes)
182 unsigned long addr, len;
185 ghes_estatus_pool = gen_pool_create(GHES_ESTATUS_POOL_MIN_ALLOC_ORDER, -1);
186 if (!ghes_estatus_pool)
189 len = GHES_ESTATUS_CACHE_AVG_SIZE * GHES_ESTATUS_CACHE_ALLOCED_MAX;
190 len += (num_ghes * GHES_ESOURCE_PREALLOC_MAX_SIZE);
192 ghes_estatus_pool_size_request = PAGE_ALIGN(len);
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)) {
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)
492 flags = sync ? MF_ACTION_REQUIRED : 0;
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,
503 struct cper_sec_proc_arm *err = acpi_hest_get_payload(gdata);
504 int flags = sync ? MF_ACTION_REQUIRED : 0;
509 log_arm_hw_error(err);
511 sec_sev = ghes_severity(gdata->error_severity);
512 if (sev != GHES_SEV_RECOVERABLE || sec_sev != GHES_SEV_RECOVERABLE)
515 p = (char *)(err + 1);
516 for (i = 0; i < err->err_info_num; i++) {
517 struct cper_arm_err_info *err_info = (struct cper_arm_err_info *)p;
518 bool is_cache = (err_info->type == CPER_ARM_CACHE_ERROR);
519 bool has_pa = (err_info->validation_bits & CPER_ARM_INFO_VALID_PHYSICAL_ADDR);
520 const char *error_type = "unknown error";
523 * The field (err_info->error_info & BIT(26)) is fixed to set to
524 * 1 in some old firmware of HiSilicon Kunpeng920. We assume that
525 * firmware won't mix corrected errors in an uncorrected section,
526 * and don't filter out 'corrected' error here.
528 if (is_cache && has_pa) {
529 queued = ghes_do_memory_failure(err_info->physical_fault_addr, flags);
530 p += err_info->length;
534 if (err_info->type < ARRAY_SIZE(cper_proc_error_type_strs))
535 error_type = cper_proc_error_type_strs[err_info->type];
537 pr_warn_ratelimited(FW_WARN GHES_PFX
538 "Unhandled processor error type: %s\n",
540 p += err_info->length;
547 * PCIe AER errors need to be sent to the AER driver for reporting and
548 * recovery. The GHES severities map to the following AER severities and
549 * require the following handling:
551 * GHES_SEV_CORRECTABLE -> AER_CORRECTABLE
552 * These need to be reported by the AER driver but no recovery is
554 * GHES_SEV_RECOVERABLE -> AER_NONFATAL
555 * GHES_SEV_RECOVERABLE && CPER_SEC_RESET -> AER_FATAL
556 * These both need to be reported and recovered from by the AER driver.
557 * GHES_SEV_PANIC does not make it to this handling since the kernel must
560 static void ghes_handle_aer(struct acpi_hest_generic_data *gdata)
562 #ifdef CONFIG_ACPI_APEI_PCIEAER
563 struct cper_sec_pcie *pcie_err = acpi_hest_get_payload(gdata);
565 if (pcie_err->validation_bits & CPER_PCIE_VALID_DEVICE_ID &&
566 pcie_err->validation_bits & CPER_PCIE_VALID_AER_INFO) {
570 devfn = PCI_DEVFN(pcie_err->device_id.device,
571 pcie_err->device_id.function);
572 aer_severity = cper_severity_to_aer(gdata->error_severity);
575 * If firmware reset the component to contain
576 * the error, we must reinitialize it before
577 * use, so treat it as a fatal AER error.
579 if (gdata->flags & CPER_SEC_RESET)
580 aer_severity = AER_FATAL;
582 aer_recover_queue(pcie_err->device_id.segment,
583 pcie_err->device_id.bus,
585 (struct aer_capability_regs *)
591 static BLOCKING_NOTIFIER_HEAD(vendor_record_notify_list);
593 int ghes_register_vendor_record_notifier(struct notifier_block *nb)
595 return blocking_notifier_chain_register(&vendor_record_notify_list, nb);
597 EXPORT_SYMBOL_GPL(ghes_register_vendor_record_notifier);
599 void ghes_unregister_vendor_record_notifier(struct notifier_block *nb)
601 blocking_notifier_chain_unregister(&vendor_record_notify_list, nb);
603 EXPORT_SYMBOL_GPL(ghes_unregister_vendor_record_notifier);
605 static void ghes_vendor_record_work_func(struct work_struct *work)
607 struct ghes_vendor_record_entry *entry;
608 struct acpi_hest_generic_data *gdata;
611 entry = container_of(work, struct ghes_vendor_record_entry, work);
612 gdata = GHES_GDATA_FROM_VENDOR_ENTRY(entry);
614 blocking_notifier_call_chain(&vendor_record_notify_list,
615 entry->error_severity, gdata);
617 len = GHES_VENDOR_ENTRY_LEN(acpi_hest_get_record_size(gdata));
618 gen_pool_free(ghes_estatus_pool, (unsigned long)entry, len);
621 static void ghes_defer_non_standard_event(struct acpi_hest_generic_data *gdata,
624 struct acpi_hest_generic_data *copied_gdata;
625 struct ghes_vendor_record_entry *entry;
628 len = GHES_VENDOR_ENTRY_LEN(acpi_hest_get_record_size(gdata));
629 entry = (void *)gen_pool_alloc(ghes_estatus_pool, len);
633 copied_gdata = GHES_GDATA_FROM_VENDOR_ENTRY(entry);
634 memcpy(copied_gdata, gdata, acpi_hest_get_record_size(gdata));
635 entry->error_severity = sev;
637 INIT_WORK(&entry->work, ghes_vendor_record_work_func);
638 schedule_work(&entry->work);
641 static bool ghes_do_proc(struct ghes *ghes,
642 const struct acpi_hest_generic_status *estatus)
645 struct acpi_hest_generic_data *gdata;
647 const guid_t *fru_id = &guid_null;
650 bool sync = is_hest_sync_notify(ghes);
652 sev = ghes_severity(estatus->error_severity);
653 apei_estatus_for_each_section(estatus, gdata) {
654 sec_type = (guid_t *)gdata->section_type;
655 sec_sev = ghes_severity(gdata->error_severity);
656 if (gdata->validation_bits & CPER_SEC_VALID_FRU_ID)
657 fru_id = (guid_t *)gdata->fru_id;
659 if (gdata->validation_bits & CPER_SEC_VALID_FRU_TEXT)
660 fru_text = gdata->fru_text;
662 if (guid_equal(sec_type, &CPER_SEC_PLATFORM_MEM)) {
663 struct cper_sec_mem_err *mem_err = acpi_hest_get_payload(gdata);
665 ghes_edac_report_mem_error(sev, mem_err);
667 arch_apei_report_mem_error(sev, mem_err);
668 queued = ghes_handle_memory_failure(gdata, sev, sync);
670 else if (guid_equal(sec_type, &CPER_SEC_PCIE)) {
671 ghes_handle_aer(gdata);
673 else if (guid_equal(sec_type, &CPER_SEC_PROC_ARM)) {
674 queued = ghes_handle_arm_hw_error(gdata, sev, sync);
676 void *err = acpi_hest_get_payload(gdata);
678 ghes_defer_non_standard_event(gdata, sev);
679 log_non_standard_event(sec_type, fru_id, fru_text,
681 gdata->error_data_length);
688 static void __ghes_print_estatus(const char *pfx,
689 const struct acpi_hest_generic *generic,
690 const struct acpi_hest_generic_status *estatus)
692 static atomic_t seqno;
693 unsigned int curr_seqno;
697 if (ghes_severity(estatus->error_severity) <=
703 curr_seqno = atomic_inc_return(&seqno);
704 snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}" HW_ERR, pfx, curr_seqno);
705 printk("%s""Hardware error from APEI Generic Hardware Error Source: %d\n",
706 pfx_seq, generic->header.source_id);
707 cper_estatus_print(pfx_seq, estatus);
710 static int ghes_print_estatus(const char *pfx,
711 const struct acpi_hest_generic *generic,
712 const struct acpi_hest_generic_status *estatus)
714 /* Not more than 2 messages every 5 seconds */
715 static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2);
716 static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2);
717 struct ratelimit_state *ratelimit;
719 if (ghes_severity(estatus->error_severity) <= GHES_SEV_CORRECTED)
720 ratelimit = &ratelimit_corrected;
722 ratelimit = &ratelimit_uncorrected;
723 if (__ratelimit(ratelimit)) {
724 __ghes_print_estatus(pfx, generic, estatus);
731 * GHES error status reporting throttle, to report more kinds of
732 * errors, instead of just most frequently occurred errors.
734 static int ghes_estatus_cached(struct acpi_hest_generic_status *estatus)
738 unsigned long long now;
739 struct ghes_estatus_cache *cache;
740 struct acpi_hest_generic_status *cache_estatus;
742 len = cper_estatus_len(estatus);
744 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
745 cache = rcu_dereference(ghes_estatus_caches[i]);
748 if (len != cache->estatus_len)
750 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
751 if (memcmp(estatus, cache_estatus, len))
753 atomic_inc(&cache->count);
755 if (now - cache->time_in < GHES_ESTATUS_IN_CACHE_MAX_NSEC)
763 static struct ghes_estatus_cache *ghes_estatus_cache_alloc(
764 struct acpi_hest_generic *generic,
765 struct acpi_hest_generic_status *estatus)
769 struct ghes_estatus_cache *cache;
770 struct acpi_hest_generic_status *cache_estatus;
772 alloced = atomic_add_return(1, &ghes_estatus_cache_alloced);
773 if (alloced > GHES_ESTATUS_CACHE_ALLOCED_MAX) {
774 atomic_dec(&ghes_estatus_cache_alloced);
777 len = cper_estatus_len(estatus);
778 cache_len = GHES_ESTATUS_CACHE_LEN(len);
779 cache = (void *)gen_pool_alloc(ghes_estatus_pool, cache_len);
781 atomic_dec(&ghes_estatus_cache_alloced);
784 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
785 memcpy(cache_estatus, estatus, len);
786 cache->estatus_len = len;
787 atomic_set(&cache->count, 0);
788 cache->generic = generic;
789 cache->time_in = sched_clock();
793 static void ghes_estatus_cache_free(struct ghes_estatus_cache *cache)
797 len = cper_estatus_len(GHES_ESTATUS_FROM_CACHE(cache));
798 len = GHES_ESTATUS_CACHE_LEN(len);
799 gen_pool_free(ghes_estatus_pool, (unsigned long)cache, len);
800 atomic_dec(&ghes_estatus_cache_alloced);
803 static void ghes_estatus_cache_rcu_free(struct rcu_head *head)
805 struct ghes_estatus_cache *cache;
807 cache = container_of(head, struct ghes_estatus_cache, rcu);
808 ghes_estatus_cache_free(cache);
811 static void ghes_estatus_cache_add(
812 struct acpi_hest_generic *generic,
813 struct acpi_hest_generic_status *estatus)
815 int i, slot = -1, count;
816 unsigned long long now, duration, period, max_period = 0;
817 struct ghes_estatus_cache *cache, *slot_cache = NULL, *new_cache;
819 new_cache = ghes_estatus_cache_alloc(generic, estatus);
820 if (new_cache == NULL)
824 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
825 cache = rcu_dereference(ghes_estatus_caches[i]);
831 duration = now - cache->time_in;
832 if (duration >= GHES_ESTATUS_IN_CACHE_MAX_NSEC) {
837 count = atomic_read(&cache->count);
839 do_div(period, (count + 1));
840 if (period > max_period) {
846 /* new_cache must be put into array after its contents are written */
848 if (slot != -1 && cmpxchg(ghes_estatus_caches + slot,
849 slot_cache, new_cache) == slot_cache) {
851 call_rcu(&slot_cache->rcu, ghes_estatus_cache_rcu_free);
853 ghes_estatus_cache_free(new_cache);
857 static void __ghes_panic(struct ghes *ghes,
858 struct acpi_hest_generic_status *estatus,
859 u64 buf_paddr, enum fixed_addresses fixmap_idx)
861 __ghes_print_estatus(KERN_EMERG, ghes->generic, estatus);
863 ghes_clear_estatus(ghes, estatus, buf_paddr, fixmap_idx);
865 /* reboot to log the error! */
867 panic_timeout = ghes_panic_timeout;
868 panic("Fatal hardware error!");
871 static int ghes_proc(struct ghes *ghes)
873 struct acpi_hest_generic_status *estatus = ghes->estatus;
877 rc = ghes_read_estatus(ghes, estatus, &buf_paddr, FIX_APEI_GHES_IRQ);
881 if (ghes_severity(estatus->error_severity) >= GHES_SEV_PANIC)
882 __ghes_panic(ghes, estatus, buf_paddr, FIX_APEI_GHES_IRQ);
884 if (!ghes_estatus_cached(estatus)) {
885 if (ghes_print_estatus(NULL, ghes->generic, estatus))
886 ghes_estatus_cache_add(ghes->generic, estatus);
888 ghes_do_proc(ghes, estatus);
891 ghes_clear_estatus(ghes, estatus, buf_paddr, FIX_APEI_GHES_IRQ);
896 static void ghes_add_timer(struct ghes *ghes)
898 struct acpi_hest_generic *g = ghes->generic;
899 unsigned long expire;
901 if (!g->notify.poll_interval) {
902 pr_warn(FW_WARN GHES_PFX "Poll interval is 0 for generic hardware error source: %d, disabled.\n",
903 g->header.source_id);
906 expire = jiffies + msecs_to_jiffies(g->notify.poll_interval);
907 ghes->timer.expires = round_jiffies_relative(expire);
908 add_timer(&ghes->timer);
911 static void ghes_poll_func(struct timer_list *t)
913 struct ghes *ghes = from_timer(ghes, t, timer);
916 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
918 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
919 if (!(ghes->flags & GHES_EXITING))
920 ghes_add_timer(ghes);
923 static irqreturn_t ghes_irq_func(int irq, void *data)
925 struct ghes *ghes = data;
929 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
930 rc = ghes_proc(ghes);
931 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
938 static int ghes_notify_hed(struct notifier_block *this, unsigned long event,
943 int ret = NOTIFY_DONE;
945 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
947 list_for_each_entry_rcu(ghes, &ghes_hed, list) {
948 if (!ghes_proc(ghes))
952 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
957 static struct notifier_block ghes_notifier_hed = {
958 .notifier_call = ghes_notify_hed,
962 * Handlers for CPER records may not be NMI safe. For example,
963 * memory_failure_queue() takes spinlocks and calls schedule_work_on().
964 * In any NMI-like handler, memory from ghes_estatus_pool is used to save
965 * estatus, and added to the ghes_estatus_llist. irq_work_queue() causes
966 * ghes_proc_in_irq() to run in IRQ context where each estatus in
967 * ghes_estatus_llist is processed.
969 * Memory from the ghes_estatus_pool is also used with the ghes_estatus_cache
970 * to suppress frequent messages.
972 static struct llist_head ghes_estatus_llist;
973 static struct irq_work ghes_proc_irq_work;
975 static void ghes_proc_in_irq(struct irq_work *irq_work)
977 struct llist_node *llnode, *next;
978 struct ghes_estatus_node *estatus_node;
979 struct acpi_hest_generic *generic;
980 struct acpi_hest_generic_status *estatus;
981 bool task_work_pending;
985 llnode = llist_del_all(&ghes_estatus_llist);
987 * Because the time order of estatus in list is reversed,
988 * revert it back to proper order.
990 llnode = llist_reverse_order(llnode);
993 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
995 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
996 len = cper_estatus_len(estatus);
997 node_len = GHES_ESTATUS_NODE_LEN(len);
998 task_work_pending = ghes_do_proc(estatus_node->ghes, estatus);
999 if (!ghes_estatus_cached(estatus)) {
1000 generic = estatus_node->generic;
1001 if (ghes_print_estatus(NULL, generic, estatus))
1002 ghes_estatus_cache_add(generic, estatus);
1005 if (task_work_pending && current->mm) {
1006 estatus_node->task_work.func = ghes_kick_task_work;
1007 estatus_node->task_work_cpu = smp_processor_id();
1008 ret = task_work_add(current, &estatus_node->task_work,
1011 estatus_node->task_work.func = NULL;
1014 if (!estatus_node->task_work.func)
1015 gen_pool_free(ghes_estatus_pool,
1016 (unsigned long)estatus_node, node_len);
1022 static void ghes_print_queued_estatus(void)
1024 struct llist_node *llnode;
1025 struct ghes_estatus_node *estatus_node;
1026 struct acpi_hest_generic *generic;
1027 struct acpi_hest_generic_status *estatus;
1029 llnode = llist_del_all(&ghes_estatus_llist);
1031 * Because the time order of estatus in list is reversed,
1032 * revert it back to proper order.
1034 llnode = llist_reverse_order(llnode);
1036 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
1038 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
1039 generic = estatus_node->generic;
1040 ghes_print_estatus(NULL, generic, estatus);
1041 llnode = llnode->next;
1045 static int ghes_in_nmi_queue_one_entry(struct ghes *ghes,
1046 enum fixed_addresses fixmap_idx)
1048 struct acpi_hest_generic_status *estatus, tmp_header;
1049 struct ghes_estatus_node *estatus_node;
1054 if (!IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG))
1057 rc = __ghes_peek_estatus(ghes, &tmp_header, &buf_paddr, fixmap_idx);
1059 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1063 rc = __ghes_check_estatus(ghes, &tmp_header);
1065 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1069 len = cper_estatus_len(&tmp_header);
1070 node_len = GHES_ESTATUS_NODE_LEN(len);
1071 estatus_node = (void *)gen_pool_alloc(ghes_estatus_pool, node_len);
1075 estatus_node->ghes = ghes;
1076 estatus_node->generic = ghes->generic;
1077 estatus_node->task_work.func = NULL;
1078 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
1080 if (__ghes_read_estatus(estatus, buf_paddr, fixmap_idx, len)) {
1081 ghes_clear_estatus(ghes, estatus, buf_paddr, fixmap_idx);
1086 sev = ghes_severity(estatus->error_severity);
1087 if (sev >= GHES_SEV_PANIC) {
1088 ghes_print_queued_estatus();
1089 __ghes_panic(ghes, estatus, buf_paddr, fixmap_idx);
1092 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1094 /* This error has been reported before, don't process it again. */
1095 if (ghes_estatus_cached(estatus))
1098 llist_add(&estatus_node->llnode, &ghes_estatus_llist);
1103 gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node,
1109 static int ghes_in_nmi_spool_from_list(struct list_head *rcu_list,
1110 enum fixed_addresses fixmap_idx)
1116 list_for_each_entry_rcu(ghes, rcu_list, list) {
1117 if (!ghes_in_nmi_queue_one_entry(ghes, fixmap_idx))
1122 if (IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG) && !ret)
1123 irq_work_queue(&ghes_proc_irq_work);
1128 #ifdef CONFIG_ACPI_APEI_SEA
1129 static LIST_HEAD(ghes_sea);
1132 * Return 0 only if one of the SEA error sources successfully reported an error
1133 * record sent from the firmware.
1135 int ghes_notify_sea(void)
1137 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sea);
1140 raw_spin_lock(&ghes_notify_lock_sea);
1141 rv = ghes_in_nmi_spool_from_list(&ghes_sea, FIX_APEI_GHES_SEA);
1142 raw_spin_unlock(&ghes_notify_lock_sea);
1147 static void ghes_sea_add(struct ghes *ghes)
1149 mutex_lock(&ghes_list_mutex);
1150 list_add_rcu(&ghes->list, &ghes_sea);
1151 mutex_unlock(&ghes_list_mutex);
1154 static void ghes_sea_remove(struct ghes *ghes)
1156 mutex_lock(&ghes_list_mutex);
1157 list_del_rcu(&ghes->list);
1158 mutex_unlock(&ghes_list_mutex);
1161 #else /* CONFIG_ACPI_APEI_SEA */
1162 static inline void ghes_sea_add(struct ghes *ghes) { }
1163 static inline void ghes_sea_remove(struct ghes *ghes) { }
1164 #endif /* CONFIG_ACPI_APEI_SEA */
1166 #ifdef CONFIG_HAVE_ACPI_APEI_NMI
1168 * NMI may be triggered on any CPU, so ghes_in_nmi is used for
1169 * having only one concurrent reader.
1171 static atomic_t ghes_in_nmi = ATOMIC_INIT(0);
1173 static LIST_HEAD(ghes_nmi);
1175 static int ghes_notify_nmi(unsigned int cmd, struct pt_regs *regs)
1177 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_nmi);
1180 if (!atomic_add_unless(&ghes_in_nmi, 1, 1))
1183 raw_spin_lock(&ghes_notify_lock_nmi);
1184 if (!ghes_in_nmi_spool_from_list(&ghes_nmi, FIX_APEI_GHES_NMI))
1186 raw_spin_unlock(&ghes_notify_lock_nmi);
1188 atomic_dec(&ghes_in_nmi);
1192 static void ghes_nmi_add(struct ghes *ghes)
1194 mutex_lock(&ghes_list_mutex);
1195 if (list_empty(&ghes_nmi))
1196 register_nmi_handler(NMI_LOCAL, ghes_notify_nmi, 0, "ghes");
1197 list_add_rcu(&ghes->list, &ghes_nmi);
1198 mutex_unlock(&ghes_list_mutex);
1201 static void ghes_nmi_remove(struct ghes *ghes)
1203 mutex_lock(&ghes_list_mutex);
1204 list_del_rcu(&ghes->list);
1205 if (list_empty(&ghes_nmi))
1206 unregister_nmi_handler(NMI_LOCAL, "ghes");
1207 mutex_unlock(&ghes_list_mutex);
1209 * To synchronize with NMI handler, ghes can only be
1210 * freed after NMI handler finishes.
1214 #else /* CONFIG_HAVE_ACPI_APEI_NMI */
1215 static inline void ghes_nmi_add(struct ghes *ghes) { }
1216 static inline void ghes_nmi_remove(struct ghes *ghes) { }
1217 #endif /* CONFIG_HAVE_ACPI_APEI_NMI */
1219 static void ghes_nmi_init_cxt(void)
1221 init_irq_work(&ghes_proc_irq_work, ghes_proc_in_irq);
1224 static int __ghes_sdei_callback(struct ghes *ghes,
1225 enum fixed_addresses fixmap_idx)
1227 if (!ghes_in_nmi_queue_one_entry(ghes, fixmap_idx)) {
1228 irq_work_queue(&ghes_proc_irq_work);
1236 static int ghes_sdei_normal_callback(u32 event_num, struct pt_regs *regs,
1239 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sdei_normal);
1240 struct ghes *ghes = arg;
1243 raw_spin_lock(&ghes_notify_lock_sdei_normal);
1244 err = __ghes_sdei_callback(ghes, FIX_APEI_GHES_SDEI_NORMAL);
1245 raw_spin_unlock(&ghes_notify_lock_sdei_normal);
1250 static int ghes_sdei_critical_callback(u32 event_num, struct pt_regs *regs,
1253 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sdei_critical);
1254 struct ghes *ghes = arg;
1257 raw_spin_lock(&ghes_notify_lock_sdei_critical);
1258 err = __ghes_sdei_callback(ghes, FIX_APEI_GHES_SDEI_CRITICAL);
1259 raw_spin_unlock(&ghes_notify_lock_sdei_critical);
1264 static int apei_sdei_register_ghes(struct ghes *ghes)
1266 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE))
1269 return sdei_register_ghes(ghes, ghes_sdei_normal_callback,
1270 ghes_sdei_critical_callback);
1273 static int apei_sdei_unregister_ghes(struct ghes *ghes)
1275 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE))
1278 return sdei_unregister_ghes(ghes);
1281 static int ghes_probe(struct platform_device *ghes_dev)
1283 struct acpi_hest_generic *generic;
1284 struct ghes *ghes = NULL;
1285 unsigned long flags;
1289 generic = *(struct acpi_hest_generic **)ghes_dev->dev.platform_data;
1290 if (!generic->enabled)
1293 switch (generic->notify.type) {
1294 case ACPI_HEST_NOTIFY_POLLED:
1295 case ACPI_HEST_NOTIFY_EXTERNAL:
1296 case ACPI_HEST_NOTIFY_SCI:
1297 case ACPI_HEST_NOTIFY_GSIV:
1298 case ACPI_HEST_NOTIFY_GPIO:
1301 case ACPI_HEST_NOTIFY_SEA:
1302 if (!IS_ENABLED(CONFIG_ACPI_APEI_SEA)) {
1303 pr_warn(GHES_PFX "Generic hardware error source: %d notified via SEA is not supported\n",
1304 generic->header.source_id);
1309 case ACPI_HEST_NOTIFY_NMI:
1310 if (!IS_ENABLED(CONFIG_HAVE_ACPI_APEI_NMI)) {
1311 pr_warn(GHES_PFX "Generic hardware error source: %d notified via NMI interrupt is not supported!\n",
1312 generic->header.source_id);
1316 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1317 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE)) {
1318 pr_warn(GHES_PFX "Generic hardware error source: %d notified via SDE Interface is not supported!\n",
1319 generic->header.source_id);
1323 case ACPI_HEST_NOTIFY_LOCAL:
1324 pr_warn(GHES_PFX "Generic hardware error source: %d notified via local interrupt is not supported!\n",
1325 generic->header.source_id);
1328 pr_warn(FW_WARN GHES_PFX "Unknown notification type: %u for generic hardware error source: %d\n",
1329 generic->notify.type, generic->header.source_id);
1334 if (generic->error_block_length <
1335 sizeof(struct acpi_hest_generic_status)) {
1336 pr_warn(FW_BUG GHES_PFX "Invalid error block length: %u for generic hardware error source: %d\n",
1337 generic->error_block_length, generic->header.source_id);
1340 ghes = ghes_new(generic);
1347 switch (generic->notify.type) {
1348 case ACPI_HEST_NOTIFY_POLLED:
1349 timer_setup(&ghes->timer, ghes_poll_func, 0);
1350 ghes_add_timer(ghes);
1352 case ACPI_HEST_NOTIFY_EXTERNAL:
1353 /* External interrupt vector is GSI */
1354 rc = acpi_gsi_to_irq(generic->notify.vector, &ghes->irq);
1356 pr_err(GHES_PFX "Failed to map GSI to IRQ for generic hardware error source: %d\n",
1357 generic->header.source_id);
1360 rc = request_irq(ghes->irq, ghes_irq_func, IRQF_SHARED,
1363 pr_err(GHES_PFX "Failed to register IRQ for generic hardware error source: %d\n",
1364 generic->header.source_id);
1369 case ACPI_HEST_NOTIFY_SCI:
1370 case ACPI_HEST_NOTIFY_GSIV:
1371 case ACPI_HEST_NOTIFY_GPIO:
1372 mutex_lock(&ghes_list_mutex);
1373 if (list_empty(&ghes_hed))
1374 register_acpi_hed_notifier(&ghes_notifier_hed);
1375 list_add_rcu(&ghes->list, &ghes_hed);
1376 mutex_unlock(&ghes_list_mutex);
1379 case ACPI_HEST_NOTIFY_SEA:
1382 case ACPI_HEST_NOTIFY_NMI:
1385 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1386 rc = apei_sdei_register_ghes(ghes);
1394 platform_set_drvdata(ghes_dev, ghes);
1396 ghes_edac_register(ghes, &ghes_dev->dev);
1398 /* Handle any pending errors right away */
1399 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
1401 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
1413 static int ghes_remove(struct platform_device *ghes_dev)
1417 struct acpi_hest_generic *generic;
1419 ghes = platform_get_drvdata(ghes_dev);
1420 generic = ghes->generic;
1422 ghes->flags |= GHES_EXITING;
1423 switch (generic->notify.type) {
1424 case ACPI_HEST_NOTIFY_POLLED:
1425 del_timer_sync(&ghes->timer);
1427 case ACPI_HEST_NOTIFY_EXTERNAL:
1428 free_irq(ghes->irq, ghes);
1431 case ACPI_HEST_NOTIFY_SCI:
1432 case ACPI_HEST_NOTIFY_GSIV:
1433 case ACPI_HEST_NOTIFY_GPIO:
1434 mutex_lock(&ghes_list_mutex);
1435 list_del_rcu(&ghes->list);
1436 if (list_empty(&ghes_hed))
1437 unregister_acpi_hed_notifier(&ghes_notifier_hed);
1438 mutex_unlock(&ghes_list_mutex);
1442 case ACPI_HEST_NOTIFY_SEA:
1443 ghes_sea_remove(ghes);
1445 case ACPI_HEST_NOTIFY_NMI:
1446 ghes_nmi_remove(ghes);
1448 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1449 rc = apei_sdei_unregister_ghes(ghes);
1460 ghes_edac_unregister(ghes);
1464 platform_set_drvdata(ghes_dev, NULL);
1469 static struct platform_driver ghes_platform_driver = {
1473 .probe = ghes_probe,
1474 .remove = ghes_remove,
1477 void __init ghes_init(void)
1486 switch (hest_disable) {
1487 case HEST_NOT_FOUND:
1490 pr_info(GHES_PFX "HEST is not enabled!\n");
1497 pr_info(GHES_PFX "GHES is not enabled!\n");
1501 ghes_nmi_init_cxt();
1503 rc = platform_driver_register(&ghes_platform_driver);
1507 rc = apei_osc_setup();
1508 if (rc == 0 && osc_sb_apei_support_acked)
1509 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit and WHEA _OSC.\n");
1510 else if (rc == 0 && !osc_sb_apei_support_acked)
1511 pr_info(GHES_PFX "APEI firmware first mode is enabled by WHEA _OSC.\n");
1512 else if (rc && osc_sb_apei_support_acked)
1513 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit.\n");
1515 pr_info(GHES_PFX "Failed to enable APEI firmware first mode.\n");