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/cxl-event.h>
30 #include <linux/platform_device.h>
31 #include <linux/mutex.h>
32 #include <linux/ratelimit.h>
33 #include <linux/vmalloc.h>
34 #include <linux/irq_work.h>
35 #include <linux/llist.h>
36 #include <linux/genalloc.h>
37 #include <linux/pci.h>
38 #include <linux/pfn.h>
39 #include <linux/aer.h>
40 #include <linux/nmi.h>
41 #include <linux/sched/clock.h>
42 #include <linux/uuid.h>
43 #include <linux/ras.h>
44 #include <linux/task_work.h>
46 #include <acpi/actbl1.h>
47 #include <acpi/ghes.h>
48 #include <acpi/apei.h>
49 #include <asm/fixmap.h>
50 #include <asm/tlbflush.h>
51 #include <ras/ras_event.h>
53 #include "apei-internal.h"
55 #define GHES_PFX "GHES: "
57 #define GHES_ESTATUS_MAX_SIZE 65536
58 #define GHES_ESOURCE_PREALLOC_MAX_SIZE 65536
60 #define GHES_ESTATUS_POOL_MIN_ALLOC_ORDER 3
62 /* This is just an estimation for memory pool allocation */
63 #define GHES_ESTATUS_CACHE_AVG_SIZE 512
65 #define GHES_ESTATUS_CACHES_SIZE 4
67 #define GHES_ESTATUS_IN_CACHE_MAX_NSEC 10000000000ULL
68 /* Prevent too many caches are allocated because of RCU */
69 #define GHES_ESTATUS_CACHE_ALLOCED_MAX (GHES_ESTATUS_CACHES_SIZE * 3 / 2)
71 #define GHES_ESTATUS_CACHE_LEN(estatus_len) \
72 (sizeof(struct ghes_estatus_cache) + (estatus_len))
73 #define GHES_ESTATUS_FROM_CACHE(estatus_cache) \
74 ((struct acpi_hest_generic_status *) \
75 ((struct ghes_estatus_cache *)(estatus_cache) + 1))
77 #define GHES_ESTATUS_NODE_LEN(estatus_len) \
78 (sizeof(struct ghes_estatus_node) + (estatus_len))
79 #define GHES_ESTATUS_FROM_NODE(estatus_node) \
80 ((struct acpi_hest_generic_status *) \
81 ((struct ghes_estatus_node *)(estatus_node) + 1))
83 #define GHES_VENDOR_ENTRY_LEN(gdata_len) \
84 (sizeof(struct ghes_vendor_record_entry) + (gdata_len))
85 #define GHES_GDATA_FROM_VENDOR_ENTRY(vendor_entry) \
86 ((struct acpi_hest_generic_data *) \
87 ((struct ghes_vendor_record_entry *)(vendor_entry) + 1))
90 * NMI-like notifications vary by architecture, before the compiler can prune
91 * unused static functions it needs a value for these enums.
93 #ifndef CONFIG_ARM_SDE_INTERFACE
94 #define FIX_APEI_GHES_SDEI_NORMAL __end_of_fixed_addresses
95 #define FIX_APEI_GHES_SDEI_CRITICAL __end_of_fixed_addresses
98 static ATOMIC_NOTIFIER_HEAD(ghes_report_chain);
100 static inline bool is_hest_type_generic_v2(struct ghes *ghes)
102 return ghes->generic->header.type == ACPI_HEST_TYPE_GENERIC_ERROR_V2;
106 * A platform may describe one error source for the handling of synchronous
107 * errors (e.g. MCE or SEA), or for handling asynchronous errors (e.g. SCI
108 * or External Interrupt). On x86, the HEST notifications are always
109 * asynchronous, so only SEA on ARM is delivered as a synchronous
112 static inline bool is_hest_sync_notify(struct ghes *ghes)
114 u8 notify_type = ghes->generic->notify.type;
116 return notify_type == ACPI_HEST_NOTIFY_SEA;
120 * This driver isn't really modular, however for the time being,
121 * continuing to use module_param is the easiest way to remain
122 * compatible with existing boot arg use cases.
125 module_param_named(disable, ghes_disable, bool, 0);
128 * "ghes.edac_force_enable" forcibly enables ghes_edac and skips the platform
131 static bool ghes_edac_force_enable;
132 module_param_named(edac_force_enable, ghes_edac_force_enable, bool, 0);
135 * All error sources notified with HED (Hardware Error Device) share a
136 * single notifier callback, so they need to be linked and checked one
137 * by one. This holds true for NMI too.
139 * RCU is used for these lists, so ghes_list_mutex is only used for
140 * list changing, not for traversing.
142 static LIST_HEAD(ghes_hed);
143 static DEFINE_MUTEX(ghes_list_mutex);
146 * A list of GHES devices which are given to the corresponding EDAC driver
147 * ghes_edac for further use.
149 static LIST_HEAD(ghes_devs);
150 static DEFINE_MUTEX(ghes_devs_mutex);
153 * Because the memory area used to transfer hardware error information
154 * from BIOS to Linux can be determined only in NMI, IRQ or timer
155 * handler, but general ioremap can not be used in atomic context, so
156 * the fixmap is used instead.
158 * This spinlock is used to prevent the fixmap entry from being used
161 static DEFINE_SPINLOCK(ghes_notify_lock_irq);
163 struct ghes_vendor_record_entry {
164 struct work_struct work;
166 char vendor_record[];
169 static struct gen_pool *ghes_estatus_pool;
171 static struct ghes_estatus_cache __rcu *ghes_estatus_caches[GHES_ESTATUS_CACHES_SIZE];
172 static atomic_t ghes_estatus_cache_alloced;
174 static int ghes_panic_timeout __read_mostly = 30;
176 static void __iomem *ghes_map(u64 pfn, enum fixed_addresses fixmap_idx)
181 paddr = PFN_PHYS(pfn);
182 prot = arch_apei_get_mem_attribute(paddr);
183 __set_fixmap(fixmap_idx, paddr, prot);
185 return (void __iomem *) __fix_to_virt(fixmap_idx);
188 static void ghes_unmap(void __iomem *vaddr, enum fixed_addresses fixmap_idx)
190 int _idx = virt_to_fix((unsigned long)vaddr);
192 WARN_ON_ONCE(fixmap_idx != _idx);
193 clear_fixmap(fixmap_idx);
196 int ghes_estatus_pool_init(unsigned int num_ghes)
198 unsigned long addr, len;
201 ghes_estatus_pool = gen_pool_create(GHES_ESTATUS_POOL_MIN_ALLOC_ORDER, -1);
202 if (!ghes_estatus_pool)
205 len = GHES_ESTATUS_CACHE_AVG_SIZE * GHES_ESTATUS_CACHE_ALLOCED_MAX;
206 len += (num_ghes * GHES_ESOURCE_PREALLOC_MAX_SIZE);
208 addr = (unsigned long)vmalloc(PAGE_ALIGN(len));
212 rc = gen_pool_add(ghes_estatus_pool, addr, PAGE_ALIGN(len), -1);
222 gen_pool_destroy(ghes_estatus_pool);
228 * ghes_estatus_pool_region_free - free previously allocated memory
229 * from the ghes_estatus_pool.
230 * @addr: address of memory to free.
231 * @size: size of memory to free.
235 void ghes_estatus_pool_region_free(unsigned long addr, u32 size)
237 gen_pool_free(ghes_estatus_pool, addr, size);
239 EXPORT_SYMBOL_GPL(ghes_estatus_pool_region_free);
241 static int map_gen_v2(struct ghes *ghes)
243 return apei_map_generic_address(&ghes->generic_v2->read_ack_register);
246 static void unmap_gen_v2(struct ghes *ghes)
248 apei_unmap_generic_address(&ghes->generic_v2->read_ack_register);
251 static void ghes_ack_error(struct acpi_hest_generic_v2 *gv2)
256 rc = apei_read(&val, &gv2->read_ack_register);
260 val &= gv2->read_ack_preserve << gv2->read_ack_register.bit_offset;
261 val |= gv2->read_ack_write << gv2->read_ack_register.bit_offset;
263 apei_write(val, &gv2->read_ack_register);
266 static struct ghes *ghes_new(struct acpi_hest_generic *generic)
269 unsigned int error_block_length;
272 ghes = kzalloc(sizeof(*ghes), GFP_KERNEL);
274 return ERR_PTR(-ENOMEM);
276 ghes->generic = generic;
277 if (is_hest_type_generic_v2(ghes)) {
278 rc = map_gen_v2(ghes);
283 rc = apei_map_generic_address(&generic->error_status_address);
285 goto err_unmap_read_ack_addr;
286 error_block_length = generic->error_block_length;
287 if (error_block_length > GHES_ESTATUS_MAX_SIZE) {
288 pr_warn(FW_WARN GHES_PFX
289 "Error status block length is too long: %u for "
290 "generic hardware error source: %d.\n",
291 error_block_length, generic->header.source_id);
292 error_block_length = GHES_ESTATUS_MAX_SIZE;
294 ghes->estatus = kmalloc(error_block_length, GFP_KERNEL);
295 if (!ghes->estatus) {
297 goto err_unmap_status_addr;
302 err_unmap_status_addr:
303 apei_unmap_generic_address(&generic->error_status_address);
304 err_unmap_read_ack_addr:
305 if (is_hest_type_generic_v2(ghes))
312 static void ghes_fini(struct ghes *ghes)
314 kfree(ghes->estatus);
315 apei_unmap_generic_address(&ghes->generic->error_status_address);
316 if (is_hest_type_generic_v2(ghes))
320 static inline int ghes_severity(int severity)
323 case CPER_SEV_INFORMATIONAL:
325 case CPER_SEV_CORRECTED:
326 return GHES_SEV_CORRECTED;
327 case CPER_SEV_RECOVERABLE:
328 return GHES_SEV_RECOVERABLE;
330 return GHES_SEV_PANIC;
332 /* Unknown, go panic */
333 return GHES_SEV_PANIC;
337 static void ghes_copy_tofrom_phys(void *buffer, u64 paddr, u32 len,
339 enum fixed_addresses fixmap_idx)
346 offset = paddr - (paddr & PAGE_MASK);
347 vaddr = ghes_map(PHYS_PFN(paddr), fixmap_idx);
348 trunk = PAGE_SIZE - offset;
349 trunk = min(trunk, len);
351 memcpy_fromio(buffer, vaddr + offset, trunk);
353 memcpy_toio(vaddr + offset, buffer, trunk);
357 ghes_unmap(vaddr, fixmap_idx);
361 /* Check the top-level record header has an appropriate size. */
362 static int __ghes_check_estatus(struct ghes *ghes,
363 struct acpi_hest_generic_status *estatus)
365 u32 len = cper_estatus_len(estatus);
367 if (len < sizeof(*estatus)) {
368 pr_warn_ratelimited(FW_WARN GHES_PFX "Truncated error status block!\n");
372 if (len > ghes->generic->error_block_length) {
373 pr_warn_ratelimited(FW_WARN GHES_PFX "Invalid error status block length!\n");
377 if (cper_estatus_check_header(estatus)) {
378 pr_warn_ratelimited(FW_WARN GHES_PFX "Invalid CPER header!\n");
385 /* Read the CPER block, returning its address, and header in estatus. */
386 static int __ghes_peek_estatus(struct ghes *ghes,
387 struct acpi_hest_generic_status *estatus,
388 u64 *buf_paddr, enum fixed_addresses fixmap_idx)
390 struct acpi_hest_generic *g = ghes->generic;
393 rc = apei_read(buf_paddr, &g->error_status_address);
396 pr_warn_ratelimited(FW_WARN GHES_PFX
397 "Failed to read error status block address for hardware error source: %d.\n",
398 g->header.source_id);
404 ghes_copy_tofrom_phys(estatus, *buf_paddr, sizeof(*estatus), 1,
406 if (!estatus->block_status) {
414 static int __ghes_read_estatus(struct acpi_hest_generic_status *estatus,
415 u64 buf_paddr, enum fixed_addresses fixmap_idx,
418 ghes_copy_tofrom_phys(estatus, buf_paddr, buf_len, 1, fixmap_idx);
419 if (cper_estatus_check(estatus)) {
420 pr_warn_ratelimited(FW_WARN GHES_PFX
421 "Failed to read error status block!\n");
428 static int ghes_read_estatus(struct ghes *ghes,
429 struct acpi_hest_generic_status *estatus,
430 u64 *buf_paddr, enum fixed_addresses fixmap_idx)
434 rc = __ghes_peek_estatus(ghes, estatus, buf_paddr, fixmap_idx);
438 rc = __ghes_check_estatus(ghes, estatus);
442 return __ghes_read_estatus(estatus, *buf_paddr, fixmap_idx,
443 cper_estatus_len(estatus));
446 static void ghes_clear_estatus(struct ghes *ghes,
447 struct acpi_hest_generic_status *estatus,
448 u64 buf_paddr, enum fixed_addresses fixmap_idx)
450 estatus->block_status = 0;
455 ghes_copy_tofrom_phys(estatus, buf_paddr,
456 sizeof(estatus->block_status), 0,
460 * GHESv2 type HEST entries introduce support for error acknowledgment,
461 * so only acknowledge the error if this support is present.
463 if (is_hest_type_generic_v2(ghes))
464 ghes_ack_error(ghes->generic_v2);
468 * Called as task_work before returning to user-space.
469 * Ensure any queued work has been done before we return to the context that
470 * triggered the notification.
472 static void ghes_kick_task_work(struct callback_head *head)
474 struct acpi_hest_generic_status *estatus;
475 struct ghes_estatus_node *estatus_node;
478 estatus_node = container_of(head, struct ghes_estatus_node, task_work);
479 if (IS_ENABLED(CONFIG_ACPI_APEI_MEMORY_FAILURE))
480 memory_failure_queue_kick(estatus_node->task_work_cpu);
482 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
483 node_len = GHES_ESTATUS_NODE_LEN(cper_estatus_len(estatus));
484 gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node, node_len);
487 static bool ghes_do_memory_failure(u64 physical_addr, int flags)
491 if (!IS_ENABLED(CONFIG_ACPI_APEI_MEMORY_FAILURE))
494 pfn = PHYS_PFN(physical_addr);
495 if (!pfn_valid(pfn) && !arch_is_platform_page(physical_addr)) {
496 pr_warn_ratelimited(FW_WARN GHES_PFX
497 "Invalid address in generic error data: %#llx\n",
502 memory_failure_queue(pfn, flags);
506 static bool ghes_handle_memory_failure(struct acpi_hest_generic_data *gdata,
510 int sec_sev = ghes_severity(gdata->error_severity);
511 struct cper_sec_mem_err *mem_err = acpi_hest_get_payload(gdata);
513 if (!(mem_err->validation_bits & CPER_MEM_VALID_PA))
516 /* iff following two events can be handled properly by now */
517 if (sec_sev == GHES_SEV_CORRECTED &&
518 (gdata->flags & CPER_SEC_ERROR_THRESHOLD_EXCEEDED))
519 flags = MF_SOFT_OFFLINE;
520 if (sev == GHES_SEV_RECOVERABLE && sec_sev == GHES_SEV_RECOVERABLE)
521 flags = sync ? MF_ACTION_REQUIRED : 0;
524 return ghes_do_memory_failure(mem_err->physical_addr, flags);
529 static bool ghes_handle_arm_hw_error(struct acpi_hest_generic_data *gdata,
532 struct cper_sec_proc_arm *err = acpi_hest_get_payload(gdata);
533 int flags = sync ? MF_ACTION_REQUIRED : 0;
538 log_arm_hw_error(err);
540 sec_sev = ghes_severity(gdata->error_severity);
541 if (sev != GHES_SEV_RECOVERABLE || sec_sev != GHES_SEV_RECOVERABLE)
544 p = (char *)(err + 1);
545 for (i = 0; i < err->err_info_num; i++) {
546 struct cper_arm_err_info *err_info = (struct cper_arm_err_info *)p;
547 bool is_cache = (err_info->type == CPER_ARM_CACHE_ERROR);
548 bool has_pa = (err_info->validation_bits & CPER_ARM_INFO_VALID_PHYSICAL_ADDR);
549 const char *error_type = "unknown error";
552 * The field (err_info->error_info & BIT(26)) is fixed to set to
553 * 1 in some old firmware of HiSilicon Kunpeng920. We assume that
554 * firmware won't mix corrected errors in an uncorrected section,
555 * and don't filter out 'corrected' error here.
557 if (is_cache && has_pa) {
558 queued = ghes_do_memory_failure(err_info->physical_fault_addr, flags);
559 p += err_info->length;
563 if (err_info->type < ARRAY_SIZE(cper_proc_error_type_strs))
564 error_type = cper_proc_error_type_strs[err_info->type];
566 pr_warn_ratelimited(FW_WARN GHES_PFX
567 "Unhandled processor error type: %s\n",
569 p += err_info->length;
576 * PCIe AER errors need to be sent to the AER driver for reporting and
577 * recovery. The GHES severities map to the following AER severities and
578 * require the following handling:
580 * GHES_SEV_CORRECTABLE -> AER_CORRECTABLE
581 * These need to be reported by the AER driver but no recovery is
583 * GHES_SEV_RECOVERABLE -> AER_NONFATAL
584 * GHES_SEV_RECOVERABLE && CPER_SEC_RESET -> AER_FATAL
585 * These both need to be reported and recovered from by the AER driver.
586 * GHES_SEV_PANIC does not make it to this handling since the kernel must
589 static void ghes_handle_aer(struct acpi_hest_generic_data *gdata)
591 #ifdef CONFIG_ACPI_APEI_PCIEAER
592 struct cper_sec_pcie *pcie_err = acpi_hest_get_payload(gdata);
594 if (pcie_err->validation_bits & CPER_PCIE_VALID_DEVICE_ID &&
595 pcie_err->validation_bits & CPER_PCIE_VALID_AER_INFO) {
600 devfn = PCI_DEVFN(pcie_err->device_id.device,
601 pcie_err->device_id.function);
602 aer_severity = cper_severity_to_aer(gdata->error_severity);
605 * If firmware reset the component to contain
606 * the error, we must reinitialize it before
607 * use, so treat it as a fatal AER error.
609 if (gdata->flags & CPER_SEC_RESET)
610 aer_severity = AER_FATAL;
612 aer_info = (void *)gen_pool_alloc(ghes_estatus_pool,
613 sizeof(struct aer_capability_regs));
616 memcpy(aer_info, pcie_err->aer_info, sizeof(struct aer_capability_regs));
618 aer_recover_queue(pcie_err->device_id.segment,
619 pcie_err->device_id.bus,
621 (struct aer_capability_regs *)
627 static BLOCKING_NOTIFIER_HEAD(vendor_record_notify_list);
629 int ghes_register_vendor_record_notifier(struct notifier_block *nb)
631 return blocking_notifier_chain_register(&vendor_record_notify_list, nb);
633 EXPORT_SYMBOL_GPL(ghes_register_vendor_record_notifier);
635 void ghes_unregister_vendor_record_notifier(struct notifier_block *nb)
637 blocking_notifier_chain_unregister(&vendor_record_notify_list, nb);
639 EXPORT_SYMBOL_GPL(ghes_unregister_vendor_record_notifier);
641 static void ghes_vendor_record_work_func(struct work_struct *work)
643 struct ghes_vendor_record_entry *entry;
644 struct acpi_hest_generic_data *gdata;
647 entry = container_of(work, struct ghes_vendor_record_entry, work);
648 gdata = GHES_GDATA_FROM_VENDOR_ENTRY(entry);
650 blocking_notifier_call_chain(&vendor_record_notify_list,
651 entry->error_severity, gdata);
653 len = GHES_VENDOR_ENTRY_LEN(acpi_hest_get_record_size(gdata));
654 gen_pool_free(ghes_estatus_pool, (unsigned long)entry, len);
657 static void ghes_defer_non_standard_event(struct acpi_hest_generic_data *gdata,
660 struct acpi_hest_generic_data *copied_gdata;
661 struct ghes_vendor_record_entry *entry;
664 len = GHES_VENDOR_ENTRY_LEN(acpi_hest_get_record_size(gdata));
665 entry = (void *)gen_pool_alloc(ghes_estatus_pool, len);
669 copied_gdata = GHES_GDATA_FROM_VENDOR_ENTRY(entry);
670 memcpy(copied_gdata, gdata, acpi_hest_get_record_size(gdata));
671 entry->error_severity = sev;
673 INIT_WORK(&entry->work, ghes_vendor_record_work_func);
674 schedule_work(&entry->work);
678 * Only a single callback can be registered for CXL CPER events.
680 static DECLARE_RWSEM(cxl_cper_rw_sem);
681 static cxl_cper_callback cper_callback;
683 static void cxl_cper_post_event(enum cxl_event_type event_type,
684 struct cxl_cper_event_rec *rec)
686 if (rec->hdr.length <= sizeof(rec->hdr) ||
687 rec->hdr.length > sizeof(*rec)) {
688 pr_err(FW_WARN "CXL CPER Invalid section length (%u)\n",
693 if (!(rec->hdr.validation_bits & CPER_CXL_COMP_EVENT_LOG_VALID)) {
694 pr_err(FW_WARN "CXL CPER invalid event\n");
698 guard(rwsem_read)(&cxl_cper_rw_sem);
700 cper_callback(event_type, rec);
703 int cxl_cper_register_callback(cxl_cper_callback callback)
705 guard(rwsem_write)(&cxl_cper_rw_sem);
708 cper_callback = callback;
711 EXPORT_SYMBOL_NS_GPL(cxl_cper_register_callback, CXL);
713 int cxl_cper_unregister_callback(cxl_cper_callback callback)
715 guard(rwsem_write)(&cxl_cper_rw_sem);
716 if (callback != cper_callback)
718 cper_callback = NULL;
721 EXPORT_SYMBOL_NS_GPL(cxl_cper_unregister_callback, CXL);
723 static bool ghes_do_proc(struct ghes *ghes,
724 const struct acpi_hest_generic_status *estatus)
727 struct acpi_hest_generic_data *gdata;
729 const guid_t *fru_id = &guid_null;
732 bool sync = is_hest_sync_notify(ghes);
734 sev = ghes_severity(estatus->error_severity);
735 apei_estatus_for_each_section(estatus, gdata) {
736 sec_type = (guid_t *)gdata->section_type;
737 sec_sev = ghes_severity(gdata->error_severity);
738 if (gdata->validation_bits & CPER_SEC_VALID_FRU_ID)
739 fru_id = (guid_t *)gdata->fru_id;
741 if (gdata->validation_bits & CPER_SEC_VALID_FRU_TEXT)
742 fru_text = gdata->fru_text;
744 if (guid_equal(sec_type, &CPER_SEC_PLATFORM_MEM)) {
745 struct cper_sec_mem_err *mem_err = acpi_hest_get_payload(gdata);
747 atomic_notifier_call_chain(&ghes_report_chain, sev, mem_err);
749 arch_apei_report_mem_error(sev, mem_err);
750 queued = ghes_handle_memory_failure(gdata, sev, sync);
752 else if (guid_equal(sec_type, &CPER_SEC_PCIE)) {
753 ghes_handle_aer(gdata);
755 else if (guid_equal(sec_type, &CPER_SEC_PROC_ARM)) {
756 queued = ghes_handle_arm_hw_error(gdata, sev, sync);
757 } else if (guid_equal(sec_type, &CPER_SEC_CXL_GEN_MEDIA_GUID)) {
758 struct cxl_cper_event_rec *rec =
759 acpi_hest_get_payload(gdata);
761 cxl_cper_post_event(CXL_CPER_EVENT_GEN_MEDIA, rec);
762 } else if (guid_equal(sec_type, &CPER_SEC_CXL_DRAM_GUID)) {
763 struct cxl_cper_event_rec *rec =
764 acpi_hest_get_payload(gdata);
766 cxl_cper_post_event(CXL_CPER_EVENT_DRAM, rec);
767 } else if (guid_equal(sec_type,
768 &CPER_SEC_CXL_MEM_MODULE_GUID)) {
769 struct cxl_cper_event_rec *rec =
770 acpi_hest_get_payload(gdata);
772 cxl_cper_post_event(CXL_CPER_EVENT_MEM_MODULE, rec);
774 void *err = acpi_hest_get_payload(gdata);
776 ghes_defer_non_standard_event(gdata, sev);
777 log_non_standard_event(sec_type, fru_id, fru_text,
779 gdata->error_data_length);
786 static void __ghes_print_estatus(const char *pfx,
787 const struct acpi_hest_generic *generic,
788 const struct acpi_hest_generic_status *estatus)
790 static atomic_t seqno;
791 unsigned int curr_seqno;
795 if (ghes_severity(estatus->error_severity) <=
801 curr_seqno = atomic_inc_return(&seqno);
802 snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}" HW_ERR, pfx, curr_seqno);
803 printk("%s""Hardware error from APEI Generic Hardware Error Source: %d\n",
804 pfx_seq, generic->header.source_id);
805 cper_estatus_print(pfx_seq, estatus);
808 static int ghes_print_estatus(const char *pfx,
809 const struct acpi_hest_generic *generic,
810 const struct acpi_hest_generic_status *estatus)
812 /* Not more than 2 messages every 5 seconds */
813 static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2);
814 static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2);
815 struct ratelimit_state *ratelimit;
817 if (ghes_severity(estatus->error_severity) <= GHES_SEV_CORRECTED)
818 ratelimit = &ratelimit_corrected;
820 ratelimit = &ratelimit_uncorrected;
821 if (__ratelimit(ratelimit)) {
822 __ghes_print_estatus(pfx, generic, estatus);
829 * GHES error status reporting throttle, to report more kinds of
830 * errors, instead of just most frequently occurred errors.
832 static int ghes_estatus_cached(struct acpi_hest_generic_status *estatus)
836 unsigned long long now;
837 struct ghes_estatus_cache *cache;
838 struct acpi_hest_generic_status *cache_estatus;
840 len = cper_estatus_len(estatus);
842 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
843 cache = rcu_dereference(ghes_estatus_caches[i]);
846 if (len != cache->estatus_len)
848 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
849 if (memcmp(estatus, cache_estatus, len))
851 atomic_inc(&cache->count);
853 if (now - cache->time_in < GHES_ESTATUS_IN_CACHE_MAX_NSEC)
861 static struct ghes_estatus_cache *ghes_estatus_cache_alloc(
862 struct acpi_hest_generic *generic,
863 struct acpi_hest_generic_status *estatus)
867 struct ghes_estatus_cache *cache;
868 struct acpi_hest_generic_status *cache_estatus;
870 alloced = atomic_add_return(1, &ghes_estatus_cache_alloced);
871 if (alloced > GHES_ESTATUS_CACHE_ALLOCED_MAX) {
872 atomic_dec(&ghes_estatus_cache_alloced);
875 len = cper_estatus_len(estatus);
876 cache_len = GHES_ESTATUS_CACHE_LEN(len);
877 cache = (void *)gen_pool_alloc(ghes_estatus_pool, cache_len);
879 atomic_dec(&ghes_estatus_cache_alloced);
882 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
883 memcpy(cache_estatus, estatus, len);
884 cache->estatus_len = len;
885 atomic_set(&cache->count, 0);
886 cache->generic = generic;
887 cache->time_in = sched_clock();
891 static void ghes_estatus_cache_rcu_free(struct rcu_head *head)
893 struct ghes_estatus_cache *cache;
896 cache = container_of(head, struct ghes_estatus_cache, rcu);
897 len = cper_estatus_len(GHES_ESTATUS_FROM_CACHE(cache));
898 len = GHES_ESTATUS_CACHE_LEN(len);
899 gen_pool_free(ghes_estatus_pool, (unsigned long)cache, len);
900 atomic_dec(&ghes_estatus_cache_alloced);
904 ghes_estatus_cache_add(struct acpi_hest_generic *generic,
905 struct acpi_hest_generic_status *estatus)
907 unsigned long long now, duration, period, max_period = 0;
908 struct ghes_estatus_cache *cache, *new_cache;
909 struct ghes_estatus_cache __rcu *victim;
910 int i, slot = -1, count;
912 new_cache = ghes_estatus_cache_alloc(generic, estatus);
918 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
919 cache = rcu_dereference(ghes_estatus_caches[i]);
924 duration = now - cache->time_in;
925 if (duration >= GHES_ESTATUS_IN_CACHE_MAX_NSEC) {
929 count = atomic_read(&cache->count);
931 do_div(period, (count + 1));
932 if (period > max_period) {
941 * Use release semantics to ensure that ghes_estatus_cached()
942 * running on another CPU will see the updated cache fields if
943 * it can see the new value of the pointer.
945 victim = xchg_release(&ghes_estatus_caches[slot],
946 RCU_INITIALIZER(new_cache));
949 * At this point, victim may point to a cached item different
950 * from the one based on which we selected the slot. Instead of
951 * going to the loop again to pick another slot, let's just
952 * drop the other item anyway: this may cause a false cache
953 * miss later on, but that won't cause any problems.
956 call_rcu(&unrcu_pointer(victim)->rcu,
957 ghes_estatus_cache_rcu_free);
961 static void __ghes_panic(struct ghes *ghes,
962 struct acpi_hest_generic_status *estatus,
963 u64 buf_paddr, enum fixed_addresses fixmap_idx)
965 __ghes_print_estatus(KERN_EMERG, ghes->generic, estatus);
967 ghes_clear_estatus(ghes, estatus, buf_paddr, fixmap_idx);
969 /* reboot to log the error! */
971 panic_timeout = ghes_panic_timeout;
972 panic("Fatal hardware error!");
975 static int ghes_proc(struct ghes *ghes)
977 struct acpi_hest_generic_status *estatus = ghes->estatus;
981 rc = ghes_read_estatus(ghes, estatus, &buf_paddr, FIX_APEI_GHES_IRQ);
985 if (ghes_severity(estatus->error_severity) >= GHES_SEV_PANIC)
986 __ghes_panic(ghes, estatus, buf_paddr, FIX_APEI_GHES_IRQ);
988 if (!ghes_estatus_cached(estatus)) {
989 if (ghes_print_estatus(NULL, ghes->generic, estatus))
990 ghes_estatus_cache_add(ghes->generic, estatus);
992 ghes_do_proc(ghes, estatus);
995 ghes_clear_estatus(ghes, estatus, buf_paddr, FIX_APEI_GHES_IRQ);
1000 static void ghes_add_timer(struct ghes *ghes)
1002 struct acpi_hest_generic *g = ghes->generic;
1003 unsigned long expire;
1005 if (!g->notify.poll_interval) {
1006 pr_warn(FW_WARN GHES_PFX "Poll interval is 0 for generic hardware error source: %d, disabled.\n",
1007 g->header.source_id);
1010 expire = jiffies + msecs_to_jiffies(g->notify.poll_interval);
1011 ghes->timer.expires = round_jiffies_relative(expire);
1012 add_timer(&ghes->timer);
1015 static void ghes_poll_func(struct timer_list *t)
1017 struct ghes *ghes = from_timer(ghes, t, timer);
1018 unsigned long flags;
1020 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
1022 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
1023 if (!(ghes->flags & GHES_EXITING))
1024 ghes_add_timer(ghes);
1027 static irqreturn_t ghes_irq_func(int irq, void *data)
1029 struct ghes *ghes = data;
1030 unsigned long flags;
1033 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
1034 rc = ghes_proc(ghes);
1035 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
1042 static int ghes_notify_hed(struct notifier_block *this, unsigned long event,
1046 unsigned long flags;
1047 int ret = NOTIFY_DONE;
1049 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
1051 list_for_each_entry_rcu(ghes, &ghes_hed, list) {
1052 if (!ghes_proc(ghes))
1056 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
1061 static struct notifier_block ghes_notifier_hed = {
1062 .notifier_call = ghes_notify_hed,
1066 * Handlers for CPER records may not be NMI safe. For example,
1067 * memory_failure_queue() takes spinlocks and calls schedule_work_on().
1068 * In any NMI-like handler, memory from ghes_estatus_pool is used to save
1069 * estatus, and added to the ghes_estatus_llist. irq_work_queue() causes
1070 * ghes_proc_in_irq() to run in IRQ context where each estatus in
1071 * ghes_estatus_llist is processed.
1073 * Memory from the ghes_estatus_pool is also used with the ghes_estatus_cache
1074 * to suppress frequent messages.
1076 static struct llist_head ghes_estatus_llist;
1077 static struct irq_work ghes_proc_irq_work;
1079 static void ghes_proc_in_irq(struct irq_work *irq_work)
1081 struct llist_node *llnode, *next;
1082 struct ghes_estatus_node *estatus_node;
1083 struct acpi_hest_generic *generic;
1084 struct acpi_hest_generic_status *estatus;
1085 bool task_work_pending;
1089 llnode = llist_del_all(&ghes_estatus_llist);
1091 * Because the time order of estatus in list is reversed,
1092 * revert it back to proper order.
1094 llnode = llist_reverse_order(llnode);
1096 next = llnode->next;
1097 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
1099 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
1100 len = cper_estatus_len(estatus);
1101 node_len = GHES_ESTATUS_NODE_LEN(len);
1102 task_work_pending = ghes_do_proc(estatus_node->ghes, estatus);
1103 if (!ghes_estatus_cached(estatus)) {
1104 generic = estatus_node->generic;
1105 if (ghes_print_estatus(NULL, generic, estatus))
1106 ghes_estatus_cache_add(generic, estatus);
1109 if (task_work_pending && current->mm) {
1110 estatus_node->task_work.func = ghes_kick_task_work;
1111 estatus_node->task_work_cpu = smp_processor_id();
1112 ret = task_work_add(current, &estatus_node->task_work,
1115 estatus_node->task_work.func = NULL;
1118 if (!estatus_node->task_work.func)
1119 gen_pool_free(ghes_estatus_pool,
1120 (unsigned long)estatus_node, node_len);
1126 static void ghes_print_queued_estatus(void)
1128 struct llist_node *llnode;
1129 struct ghes_estatus_node *estatus_node;
1130 struct acpi_hest_generic *generic;
1131 struct acpi_hest_generic_status *estatus;
1133 llnode = llist_del_all(&ghes_estatus_llist);
1135 * Because the time order of estatus in list is reversed,
1136 * revert it back to proper order.
1138 llnode = llist_reverse_order(llnode);
1140 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
1142 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
1143 generic = estatus_node->generic;
1144 ghes_print_estatus(NULL, generic, estatus);
1145 llnode = llnode->next;
1149 static int ghes_in_nmi_queue_one_entry(struct ghes *ghes,
1150 enum fixed_addresses fixmap_idx)
1152 struct acpi_hest_generic_status *estatus, tmp_header;
1153 struct ghes_estatus_node *estatus_node;
1158 if (!IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG))
1161 rc = __ghes_peek_estatus(ghes, &tmp_header, &buf_paddr, fixmap_idx);
1163 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1167 rc = __ghes_check_estatus(ghes, &tmp_header);
1169 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1173 len = cper_estatus_len(&tmp_header);
1174 node_len = GHES_ESTATUS_NODE_LEN(len);
1175 estatus_node = (void *)gen_pool_alloc(ghes_estatus_pool, node_len);
1179 estatus_node->ghes = ghes;
1180 estatus_node->generic = ghes->generic;
1181 estatus_node->task_work.func = NULL;
1182 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
1184 if (__ghes_read_estatus(estatus, buf_paddr, fixmap_idx, len)) {
1185 ghes_clear_estatus(ghes, estatus, buf_paddr, fixmap_idx);
1190 sev = ghes_severity(estatus->error_severity);
1191 if (sev >= GHES_SEV_PANIC) {
1192 ghes_print_queued_estatus();
1193 __ghes_panic(ghes, estatus, buf_paddr, fixmap_idx);
1196 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1198 /* This error has been reported before, don't process it again. */
1199 if (ghes_estatus_cached(estatus))
1202 llist_add(&estatus_node->llnode, &ghes_estatus_llist);
1207 gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node,
1213 static int ghes_in_nmi_spool_from_list(struct list_head *rcu_list,
1214 enum fixed_addresses fixmap_idx)
1220 list_for_each_entry_rcu(ghes, rcu_list, list) {
1221 if (!ghes_in_nmi_queue_one_entry(ghes, fixmap_idx))
1226 if (IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG) && !ret)
1227 irq_work_queue(&ghes_proc_irq_work);
1232 #ifdef CONFIG_ACPI_APEI_SEA
1233 static LIST_HEAD(ghes_sea);
1236 * Return 0 only if one of the SEA error sources successfully reported an error
1237 * record sent from the firmware.
1239 int ghes_notify_sea(void)
1241 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sea);
1244 raw_spin_lock(&ghes_notify_lock_sea);
1245 rv = ghes_in_nmi_spool_from_list(&ghes_sea, FIX_APEI_GHES_SEA);
1246 raw_spin_unlock(&ghes_notify_lock_sea);
1251 static void ghes_sea_add(struct ghes *ghes)
1253 mutex_lock(&ghes_list_mutex);
1254 list_add_rcu(&ghes->list, &ghes_sea);
1255 mutex_unlock(&ghes_list_mutex);
1258 static void ghes_sea_remove(struct ghes *ghes)
1260 mutex_lock(&ghes_list_mutex);
1261 list_del_rcu(&ghes->list);
1262 mutex_unlock(&ghes_list_mutex);
1265 #else /* CONFIG_ACPI_APEI_SEA */
1266 static inline void ghes_sea_add(struct ghes *ghes) { }
1267 static inline void ghes_sea_remove(struct ghes *ghes) { }
1268 #endif /* CONFIG_ACPI_APEI_SEA */
1270 #ifdef CONFIG_HAVE_ACPI_APEI_NMI
1272 * NMI may be triggered on any CPU, so ghes_in_nmi is used for
1273 * having only one concurrent reader.
1275 static atomic_t ghes_in_nmi = ATOMIC_INIT(0);
1277 static LIST_HEAD(ghes_nmi);
1279 static int ghes_notify_nmi(unsigned int cmd, struct pt_regs *regs)
1281 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_nmi);
1284 if (!atomic_add_unless(&ghes_in_nmi, 1, 1))
1287 raw_spin_lock(&ghes_notify_lock_nmi);
1288 if (!ghes_in_nmi_spool_from_list(&ghes_nmi, FIX_APEI_GHES_NMI))
1290 raw_spin_unlock(&ghes_notify_lock_nmi);
1292 atomic_dec(&ghes_in_nmi);
1296 static void ghes_nmi_add(struct ghes *ghes)
1298 mutex_lock(&ghes_list_mutex);
1299 if (list_empty(&ghes_nmi))
1300 register_nmi_handler(NMI_LOCAL, ghes_notify_nmi, 0, "ghes");
1301 list_add_rcu(&ghes->list, &ghes_nmi);
1302 mutex_unlock(&ghes_list_mutex);
1305 static void ghes_nmi_remove(struct ghes *ghes)
1307 mutex_lock(&ghes_list_mutex);
1308 list_del_rcu(&ghes->list);
1309 if (list_empty(&ghes_nmi))
1310 unregister_nmi_handler(NMI_LOCAL, "ghes");
1311 mutex_unlock(&ghes_list_mutex);
1313 * To synchronize with NMI handler, ghes can only be
1314 * freed after NMI handler finishes.
1318 #else /* CONFIG_HAVE_ACPI_APEI_NMI */
1319 static inline void ghes_nmi_add(struct ghes *ghes) { }
1320 static inline void ghes_nmi_remove(struct ghes *ghes) { }
1321 #endif /* CONFIG_HAVE_ACPI_APEI_NMI */
1323 static void ghes_nmi_init_cxt(void)
1325 init_irq_work(&ghes_proc_irq_work, ghes_proc_in_irq);
1328 static int __ghes_sdei_callback(struct ghes *ghes,
1329 enum fixed_addresses fixmap_idx)
1331 if (!ghes_in_nmi_queue_one_entry(ghes, fixmap_idx)) {
1332 irq_work_queue(&ghes_proc_irq_work);
1340 static int ghes_sdei_normal_callback(u32 event_num, struct pt_regs *regs,
1343 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sdei_normal);
1344 struct ghes *ghes = arg;
1347 raw_spin_lock(&ghes_notify_lock_sdei_normal);
1348 err = __ghes_sdei_callback(ghes, FIX_APEI_GHES_SDEI_NORMAL);
1349 raw_spin_unlock(&ghes_notify_lock_sdei_normal);
1354 static int ghes_sdei_critical_callback(u32 event_num, struct pt_regs *regs,
1357 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sdei_critical);
1358 struct ghes *ghes = arg;
1361 raw_spin_lock(&ghes_notify_lock_sdei_critical);
1362 err = __ghes_sdei_callback(ghes, FIX_APEI_GHES_SDEI_CRITICAL);
1363 raw_spin_unlock(&ghes_notify_lock_sdei_critical);
1368 static int apei_sdei_register_ghes(struct ghes *ghes)
1370 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE))
1373 return sdei_register_ghes(ghes, ghes_sdei_normal_callback,
1374 ghes_sdei_critical_callback);
1377 static int apei_sdei_unregister_ghes(struct ghes *ghes)
1379 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE))
1382 return sdei_unregister_ghes(ghes);
1385 static int ghes_probe(struct platform_device *ghes_dev)
1387 struct acpi_hest_generic *generic;
1388 struct ghes *ghes = NULL;
1389 unsigned long flags;
1393 generic = *(struct acpi_hest_generic **)ghes_dev->dev.platform_data;
1394 if (!generic->enabled)
1397 switch (generic->notify.type) {
1398 case ACPI_HEST_NOTIFY_POLLED:
1399 case ACPI_HEST_NOTIFY_EXTERNAL:
1400 case ACPI_HEST_NOTIFY_SCI:
1401 case ACPI_HEST_NOTIFY_GSIV:
1402 case ACPI_HEST_NOTIFY_GPIO:
1405 case ACPI_HEST_NOTIFY_SEA:
1406 if (!IS_ENABLED(CONFIG_ACPI_APEI_SEA)) {
1407 pr_warn(GHES_PFX "Generic hardware error source: %d notified via SEA is not supported\n",
1408 generic->header.source_id);
1413 case ACPI_HEST_NOTIFY_NMI:
1414 if (!IS_ENABLED(CONFIG_HAVE_ACPI_APEI_NMI)) {
1415 pr_warn(GHES_PFX "Generic hardware error source: %d notified via NMI interrupt is not supported!\n",
1416 generic->header.source_id);
1420 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1421 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE)) {
1422 pr_warn(GHES_PFX "Generic hardware error source: %d notified via SDE Interface is not supported!\n",
1423 generic->header.source_id);
1427 case ACPI_HEST_NOTIFY_LOCAL:
1428 pr_warn(GHES_PFX "Generic hardware error source: %d notified via local interrupt is not supported!\n",
1429 generic->header.source_id);
1432 pr_warn(FW_WARN GHES_PFX "Unknown notification type: %u for generic hardware error source: %d\n",
1433 generic->notify.type, generic->header.source_id);
1438 if (generic->error_block_length <
1439 sizeof(struct acpi_hest_generic_status)) {
1440 pr_warn(FW_BUG GHES_PFX "Invalid error block length: %u for generic hardware error source: %d\n",
1441 generic->error_block_length, generic->header.source_id);
1444 ghes = ghes_new(generic);
1451 switch (generic->notify.type) {
1452 case ACPI_HEST_NOTIFY_POLLED:
1453 timer_setup(&ghes->timer, ghes_poll_func, 0);
1454 ghes_add_timer(ghes);
1456 case ACPI_HEST_NOTIFY_EXTERNAL:
1457 /* External interrupt vector is GSI */
1458 rc = acpi_gsi_to_irq(generic->notify.vector, &ghes->irq);
1460 pr_err(GHES_PFX "Failed to map GSI to IRQ for generic hardware error source: %d\n",
1461 generic->header.source_id);
1464 rc = request_irq(ghes->irq, ghes_irq_func, IRQF_SHARED,
1467 pr_err(GHES_PFX "Failed to register IRQ for generic hardware error source: %d\n",
1468 generic->header.source_id);
1473 case ACPI_HEST_NOTIFY_SCI:
1474 case ACPI_HEST_NOTIFY_GSIV:
1475 case ACPI_HEST_NOTIFY_GPIO:
1476 mutex_lock(&ghes_list_mutex);
1477 if (list_empty(&ghes_hed))
1478 register_acpi_hed_notifier(&ghes_notifier_hed);
1479 list_add_rcu(&ghes->list, &ghes_hed);
1480 mutex_unlock(&ghes_list_mutex);
1483 case ACPI_HEST_NOTIFY_SEA:
1486 case ACPI_HEST_NOTIFY_NMI:
1489 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1490 rc = apei_sdei_register_ghes(ghes);
1498 platform_set_drvdata(ghes_dev, ghes);
1500 ghes->dev = &ghes_dev->dev;
1502 mutex_lock(&ghes_devs_mutex);
1503 list_add_tail(&ghes->elist, &ghes_devs);
1504 mutex_unlock(&ghes_devs_mutex);
1506 /* Handle any pending errors right away */
1507 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
1509 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
1521 static int ghes_remove(struct platform_device *ghes_dev)
1525 struct acpi_hest_generic *generic;
1527 ghes = platform_get_drvdata(ghes_dev);
1528 generic = ghes->generic;
1530 ghes->flags |= GHES_EXITING;
1531 switch (generic->notify.type) {
1532 case ACPI_HEST_NOTIFY_POLLED:
1533 timer_shutdown_sync(&ghes->timer);
1535 case ACPI_HEST_NOTIFY_EXTERNAL:
1536 free_irq(ghes->irq, ghes);
1539 case ACPI_HEST_NOTIFY_SCI:
1540 case ACPI_HEST_NOTIFY_GSIV:
1541 case ACPI_HEST_NOTIFY_GPIO:
1542 mutex_lock(&ghes_list_mutex);
1543 list_del_rcu(&ghes->list);
1544 if (list_empty(&ghes_hed))
1545 unregister_acpi_hed_notifier(&ghes_notifier_hed);
1546 mutex_unlock(&ghes_list_mutex);
1550 case ACPI_HEST_NOTIFY_SEA:
1551 ghes_sea_remove(ghes);
1553 case ACPI_HEST_NOTIFY_NMI:
1554 ghes_nmi_remove(ghes);
1556 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1557 rc = apei_sdei_unregister_ghes(ghes);
1568 mutex_lock(&ghes_devs_mutex);
1569 list_del(&ghes->elist);
1570 mutex_unlock(&ghes_devs_mutex);
1577 static struct platform_driver ghes_platform_driver = {
1581 .probe = ghes_probe,
1582 .remove = ghes_remove,
1585 void __init acpi_ghes_init(void)
1594 switch (hest_disable) {
1595 case HEST_NOT_FOUND:
1598 pr_info(GHES_PFX "HEST is not enabled!\n");
1605 pr_info(GHES_PFX "GHES is not enabled!\n");
1609 ghes_nmi_init_cxt();
1611 rc = platform_driver_register(&ghes_platform_driver);
1615 rc = apei_osc_setup();
1616 if (rc == 0 && osc_sb_apei_support_acked)
1617 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit and WHEA _OSC.\n");
1618 else if (rc == 0 && !osc_sb_apei_support_acked)
1619 pr_info(GHES_PFX "APEI firmware first mode is enabled by WHEA _OSC.\n");
1620 else if (rc && osc_sb_apei_support_acked)
1621 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit.\n");
1623 pr_info(GHES_PFX "Failed to enable APEI firmware first mode.\n");
1627 * Known x86 systems that prefer GHES error reporting:
1629 static struct acpi_platform_list plat_list[] = {
1630 {"HPE ", "Server ", 0, ACPI_SIG_FADT, all_versions},
1634 struct list_head *ghes_get_devices(void)
1638 if (IS_ENABLED(CONFIG_X86)) {
1639 idx = acpi_match_platform_list(plat_list);
1641 if (!ghes_edac_force_enable)
1644 pr_warn_once("Force-loading ghes_edac on an unsupported platform. You're on your own!\n");
1646 } else if (list_empty(&ghes_devs)) {
1652 EXPORT_SYMBOL_GPL(ghes_get_devices);
1654 void ghes_register_report_chain(struct notifier_block *nb)
1656 atomic_notifier_chain_register(&ghes_report_chain, nb);
1658 EXPORT_SYMBOL_GPL(ghes_register_report_chain);
1660 void ghes_unregister_report_chain(struct notifier_block *nb)
1662 atomic_notifier_chain_unregister(&ghes_report_chain, nb);
1664 EXPORT_SYMBOL_GPL(ghes_unregister_report_chain);