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 /* CXL Event record UUIDs are formatted as GUIDs and reported in section type */
686 * General Media Event Record
687 * CXL rev 3.0 Section 8.2.9.2.1.1; Table 8-43
689 #define CPER_SEC_CXL_GEN_MEDIA_GUID \
690 GUID_INIT(0xfbcd0a77, 0xc260, 0x417f, \
691 0x85, 0xa9, 0x08, 0x8b, 0x16, 0x21, 0xeb, 0xa6)
695 * CXL rev 3.0 section 8.2.9.2.1.2; Table 8-44
697 #define CPER_SEC_CXL_DRAM_GUID \
698 GUID_INIT(0x601dcbb3, 0x9c06, 0x4eab, \
699 0xb8, 0xaf, 0x4e, 0x9b, 0xfb, 0x5c, 0x96, 0x24)
702 * Memory Module Event Record
703 * CXL rev 3.0 section 8.2.9.2.1.3; Table 8-45
705 #define CPER_SEC_CXL_MEM_MODULE_GUID \
706 GUID_INIT(0xfe927475, 0xdd59, 0x4339, \
707 0xa5, 0x86, 0x79, 0xba, 0xb1, 0x13, 0xb7, 0x74)
709 static void cxl_cper_post_event(enum cxl_event_type event_type,
710 struct cxl_cper_event_rec *rec)
712 if (rec->hdr.length <= sizeof(rec->hdr) ||
713 rec->hdr.length > sizeof(*rec)) {
714 pr_err(FW_WARN "CXL CPER Invalid section length (%u)\n",
719 if (!(rec->hdr.validation_bits & CPER_CXL_COMP_EVENT_LOG_VALID)) {
720 pr_err(FW_WARN "CXL CPER invalid event\n");
724 guard(rwsem_read)(&cxl_cper_rw_sem);
726 cper_callback(event_type, rec);
729 int cxl_cper_register_callback(cxl_cper_callback callback)
731 guard(rwsem_write)(&cxl_cper_rw_sem);
734 cper_callback = callback;
737 EXPORT_SYMBOL_NS_GPL(cxl_cper_register_callback, CXL);
739 int cxl_cper_unregister_callback(cxl_cper_callback callback)
741 guard(rwsem_write)(&cxl_cper_rw_sem);
742 if (callback != cper_callback)
744 cper_callback = NULL;
747 EXPORT_SYMBOL_NS_GPL(cxl_cper_unregister_callback, CXL);
749 static bool ghes_do_proc(struct ghes *ghes,
750 const struct acpi_hest_generic_status *estatus)
753 struct acpi_hest_generic_data *gdata;
755 const guid_t *fru_id = &guid_null;
758 bool sync = is_hest_sync_notify(ghes);
760 sev = ghes_severity(estatus->error_severity);
761 apei_estatus_for_each_section(estatus, gdata) {
762 sec_type = (guid_t *)gdata->section_type;
763 sec_sev = ghes_severity(gdata->error_severity);
764 if (gdata->validation_bits & CPER_SEC_VALID_FRU_ID)
765 fru_id = (guid_t *)gdata->fru_id;
767 if (gdata->validation_bits & CPER_SEC_VALID_FRU_TEXT)
768 fru_text = gdata->fru_text;
770 if (guid_equal(sec_type, &CPER_SEC_PLATFORM_MEM)) {
771 struct cper_sec_mem_err *mem_err = acpi_hest_get_payload(gdata);
773 atomic_notifier_call_chain(&ghes_report_chain, sev, mem_err);
775 arch_apei_report_mem_error(sev, mem_err);
776 queued = ghes_handle_memory_failure(gdata, sev, sync);
778 else if (guid_equal(sec_type, &CPER_SEC_PCIE)) {
779 ghes_handle_aer(gdata);
781 else if (guid_equal(sec_type, &CPER_SEC_PROC_ARM)) {
782 queued = ghes_handle_arm_hw_error(gdata, sev, sync);
783 } else if (guid_equal(sec_type, &CPER_SEC_CXL_GEN_MEDIA_GUID)) {
784 struct cxl_cper_event_rec *rec =
785 acpi_hest_get_payload(gdata);
787 cxl_cper_post_event(CXL_CPER_EVENT_GEN_MEDIA, rec);
788 } else if (guid_equal(sec_type, &CPER_SEC_CXL_DRAM_GUID)) {
789 struct cxl_cper_event_rec *rec =
790 acpi_hest_get_payload(gdata);
792 cxl_cper_post_event(CXL_CPER_EVENT_DRAM, rec);
793 } else if (guid_equal(sec_type,
794 &CPER_SEC_CXL_MEM_MODULE_GUID)) {
795 struct cxl_cper_event_rec *rec =
796 acpi_hest_get_payload(gdata);
798 cxl_cper_post_event(CXL_CPER_EVENT_MEM_MODULE, rec);
800 void *err = acpi_hest_get_payload(gdata);
802 ghes_defer_non_standard_event(gdata, sev);
803 log_non_standard_event(sec_type, fru_id, fru_text,
805 gdata->error_data_length);
812 static void __ghes_print_estatus(const char *pfx,
813 const struct acpi_hest_generic *generic,
814 const struct acpi_hest_generic_status *estatus)
816 static atomic_t seqno;
817 unsigned int curr_seqno;
821 if (ghes_severity(estatus->error_severity) <=
827 curr_seqno = atomic_inc_return(&seqno);
828 snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}" HW_ERR, pfx, curr_seqno);
829 printk("%s""Hardware error from APEI Generic Hardware Error Source: %d\n",
830 pfx_seq, generic->header.source_id);
831 cper_estatus_print(pfx_seq, estatus);
834 static int ghes_print_estatus(const char *pfx,
835 const struct acpi_hest_generic *generic,
836 const struct acpi_hest_generic_status *estatus)
838 /* Not more than 2 messages every 5 seconds */
839 static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2);
840 static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2);
841 struct ratelimit_state *ratelimit;
843 if (ghes_severity(estatus->error_severity) <= GHES_SEV_CORRECTED)
844 ratelimit = &ratelimit_corrected;
846 ratelimit = &ratelimit_uncorrected;
847 if (__ratelimit(ratelimit)) {
848 __ghes_print_estatus(pfx, generic, estatus);
855 * GHES error status reporting throttle, to report more kinds of
856 * errors, instead of just most frequently occurred errors.
858 static int ghes_estatus_cached(struct acpi_hest_generic_status *estatus)
862 unsigned long long now;
863 struct ghes_estatus_cache *cache;
864 struct acpi_hest_generic_status *cache_estatus;
866 len = cper_estatus_len(estatus);
868 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
869 cache = rcu_dereference(ghes_estatus_caches[i]);
872 if (len != cache->estatus_len)
874 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
875 if (memcmp(estatus, cache_estatus, len))
877 atomic_inc(&cache->count);
879 if (now - cache->time_in < GHES_ESTATUS_IN_CACHE_MAX_NSEC)
887 static struct ghes_estatus_cache *ghes_estatus_cache_alloc(
888 struct acpi_hest_generic *generic,
889 struct acpi_hest_generic_status *estatus)
893 struct ghes_estatus_cache *cache;
894 struct acpi_hest_generic_status *cache_estatus;
896 alloced = atomic_add_return(1, &ghes_estatus_cache_alloced);
897 if (alloced > GHES_ESTATUS_CACHE_ALLOCED_MAX) {
898 atomic_dec(&ghes_estatus_cache_alloced);
901 len = cper_estatus_len(estatus);
902 cache_len = GHES_ESTATUS_CACHE_LEN(len);
903 cache = (void *)gen_pool_alloc(ghes_estatus_pool, cache_len);
905 atomic_dec(&ghes_estatus_cache_alloced);
908 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
909 memcpy(cache_estatus, estatus, len);
910 cache->estatus_len = len;
911 atomic_set(&cache->count, 0);
912 cache->generic = generic;
913 cache->time_in = sched_clock();
917 static void ghes_estatus_cache_rcu_free(struct rcu_head *head)
919 struct ghes_estatus_cache *cache;
922 cache = container_of(head, struct ghes_estatus_cache, rcu);
923 len = cper_estatus_len(GHES_ESTATUS_FROM_CACHE(cache));
924 len = GHES_ESTATUS_CACHE_LEN(len);
925 gen_pool_free(ghes_estatus_pool, (unsigned long)cache, len);
926 atomic_dec(&ghes_estatus_cache_alloced);
930 ghes_estatus_cache_add(struct acpi_hest_generic *generic,
931 struct acpi_hest_generic_status *estatus)
933 unsigned long long now, duration, period, max_period = 0;
934 struct ghes_estatus_cache *cache, *new_cache;
935 struct ghes_estatus_cache __rcu *victim;
936 int i, slot = -1, count;
938 new_cache = ghes_estatus_cache_alloc(generic, estatus);
944 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
945 cache = rcu_dereference(ghes_estatus_caches[i]);
950 duration = now - cache->time_in;
951 if (duration >= GHES_ESTATUS_IN_CACHE_MAX_NSEC) {
955 count = atomic_read(&cache->count);
957 do_div(period, (count + 1));
958 if (period > max_period) {
967 * Use release semantics to ensure that ghes_estatus_cached()
968 * running on another CPU will see the updated cache fields if
969 * it can see the new value of the pointer.
971 victim = xchg_release(&ghes_estatus_caches[slot],
972 RCU_INITIALIZER(new_cache));
975 * At this point, victim may point to a cached item different
976 * from the one based on which we selected the slot. Instead of
977 * going to the loop again to pick another slot, let's just
978 * drop the other item anyway: this may cause a false cache
979 * miss later on, but that won't cause any problems.
982 call_rcu(&unrcu_pointer(victim)->rcu,
983 ghes_estatus_cache_rcu_free);
987 static void __ghes_panic(struct ghes *ghes,
988 struct acpi_hest_generic_status *estatus,
989 u64 buf_paddr, enum fixed_addresses fixmap_idx)
991 __ghes_print_estatus(KERN_EMERG, ghes->generic, estatus);
993 ghes_clear_estatus(ghes, estatus, buf_paddr, fixmap_idx);
995 /* reboot to log the error! */
997 panic_timeout = ghes_panic_timeout;
998 panic("Fatal hardware error!");
1001 static int ghes_proc(struct ghes *ghes)
1003 struct acpi_hest_generic_status *estatus = ghes->estatus;
1007 rc = ghes_read_estatus(ghes, estatus, &buf_paddr, FIX_APEI_GHES_IRQ);
1011 if (ghes_severity(estatus->error_severity) >= GHES_SEV_PANIC)
1012 __ghes_panic(ghes, estatus, buf_paddr, FIX_APEI_GHES_IRQ);
1014 if (!ghes_estatus_cached(estatus)) {
1015 if (ghes_print_estatus(NULL, ghes->generic, estatus))
1016 ghes_estatus_cache_add(ghes->generic, estatus);
1018 ghes_do_proc(ghes, estatus);
1021 ghes_clear_estatus(ghes, estatus, buf_paddr, FIX_APEI_GHES_IRQ);
1026 static void ghes_add_timer(struct ghes *ghes)
1028 struct acpi_hest_generic *g = ghes->generic;
1029 unsigned long expire;
1031 if (!g->notify.poll_interval) {
1032 pr_warn(FW_WARN GHES_PFX "Poll interval is 0 for generic hardware error source: %d, disabled.\n",
1033 g->header.source_id);
1036 expire = jiffies + msecs_to_jiffies(g->notify.poll_interval);
1037 ghes->timer.expires = round_jiffies_relative(expire);
1038 add_timer(&ghes->timer);
1041 static void ghes_poll_func(struct timer_list *t)
1043 struct ghes *ghes = from_timer(ghes, t, timer);
1044 unsigned long flags;
1046 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
1048 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
1049 if (!(ghes->flags & GHES_EXITING))
1050 ghes_add_timer(ghes);
1053 static irqreturn_t ghes_irq_func(int irq, void *data)
1055 struct ghes *ghes = data;
1056 unsigned long flags;
1059 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
1060 rc = ghes_proc(ghes);
1061 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
1068 static int ghes_notify_hed(struct notifier_block *this, unsigned long event,
1072 unsigned long flags;
1073 int ret = NOTIFY_DONE;
1075 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
1077 list_for_each_entry_rcu(ghes, &ghes_hed, list) {
1078 if (!ghes_proc(ghes))
1082 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
1087 static struct notifier_block ghes_notifier_hed = {
1088 .notifier_call = ghes_notify_hed,
1092 * Handlers for CPER records may not be NMI safe. For example,
1093 * memory_failure_queue() takes spinlocks and calls schedule_work_on().
1094 * In any NMI-like handler, memory from ghes_estatus_pool is used to save
1095 * estatus, and added to the ghes_estatus_llist. irq_work_queue() causes
1096 * ghes_proc_in_irq() to run in IRQ context where each estatus in
1097 * ghes_estatus_llist is processed.
1099 * Memory from the ghes_estatus_pool is also used with the ghes_estatus_cache
1100 * to suppress frequent messages.
1102 static struct llist_head ghes_estatus_llist;
1103 static struct irq_work ghes_proc_irq_work;
1105 static void ghes_proc_in_irq(struct irq_work *irq_work)
1107 struct llist_node *llnode, *next;
1108 struct ghes_estatus_node *estatus_node;
1109 struct acpi_hest_generic *generic;
1110 struct acpi_hest_generic_status *estatus;
1111 bool task_work_pending;
1115 llnode = llist_del_all(&ghes_estatus_llist);
1117 * Because the time order of estatus in list is reversed,
1118 * revert it back to proper order.
1120 llnode = llist_reverse_order(llnode);
1122 next = llnode->next;
1123 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
1125 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
1126 len = cper_estatus_len(estatus);
1127 node_len = GHES_ESTATUS_NODE_LEN(len);
1128 task_work_pending = ghes_do_proc(estatus_node->ghes, estatus);
1129 if (!ghes_estatus_cached(estatus)) {
1130 generic = estatus_node->generic;
1131 if (ghes_print_estatus(NULL, generic, estatus))
1132 ghes_estatus_cache_add(generic, estatus);
1135 if (task_work_pending && current->mm) {
1136 estatus_node->task_work.func = ghes_kick_task_work;
1137 estatus_node->task_work_cpu = smp_processor_id();
1138 ret = task_work_add(current, &estatus_node->task_work,
1141 estatus_node->task_work.func = NULL;
1144 if (!estatus_node->task_work.func)
1145 gen_pool_free(ghes_estatus_pool,
1146 (unsigned long)estatus_node, node_len);
1152 static void ghes_print_queued_estatus(void)
1154 struct llist_node *llnode;
1155 struct ghes_estatus_node *estatus_node;
1156 struct acpi_hest_generic *generic;
1157 struct acpi_hest_generic_status *estatus;
1159 llnode = llist_del_all(&ghes_estatus_llist);
1161 * Because the time order of estatus in list is reversed,
1162 * revert it back to proper order.
1164 llnode = llist_reverse_order(llnode);
1166 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
1168 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
1169 generic = estatus_node->generic;
1170 ghes_print_estatus(NULL, generic, estatus);
1171 llnode = llnode->next;
1175 static int ghes_in_nmi_queue_one_entry(struct ghes *ghes,
1176 enum fixed_addresses fixmap_idx)
1178 struct acpi_hest_generic_status *estatus, tmp_header;
1179 struct ghes_estatus_node *estatus_node;
1184 if (!IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG))
1187 rc = __ghes_peek_estatus(ghes, &tmp_header, &buf_paddr, fixmap_idx);
1189 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1193 rc = __ghes_check_estatus(ghes, &tmp_header);
1195 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1199 len = cper_estatus_len(&tmp_header);
1200 node_len = GHES_ESTATUS_NODE_LEN(len);
1201 estatus_node = (void *)gen_pool_alloc(ghes_estatus_pool, node_len);
1205 estatus_node->ghes = ghes;
1206 estatus_node->generic = ghes->generic;
1207 estatus_node->task_work.func = NULL;
1208 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
1210 if (__ghes_read_estatus(estatus, buf_paddr, fixmap_idx, len)) {
1211 ghes_clear_estatus(ghes, estatus, buf_paddr, fixmap_idx);
1216 sev = ghes_severity(estatus->error_severity);
1217 if (sev >= GHES_SEV_PANIC) {
1218 ghes_print_queued_estatus();
1219 __ghes_panic(ghes, estatus, buf_paddr, fixmap_idx);
1222 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1224 /* This error has been reported before, don't process it again. */
1225 if (ghes_estatus_cached(estatus))
1228 llist_add(&estatus_node->llnode, &ghes_estatus_llist);
1233 gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node,
1239 static int ghes_in_nmi_spool_from_list(struct list_head *rcu_list,
1240 enum fixed_addresses fixmap_idx)
1246 list_for_each_entry_rcu(ghes, rcu_list, list) {
1247 if (!ghes_in_nmi_queue_one_entry(ghes, fixmap_idx))
1252 if (IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG) && !ret)
1253 irq_work_queue(&ghes_proc_irq_work);
1258 #ifdef CONFIG_ACPI_APEI_SEA
1259 static LIST_HEAD(ghes_sea);
1262 * Return 0 only if one of the SEA error sources successfully reported an error
1263 * record sent from the firmware.
1265 int ghes_notify_sea(void)
1267 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sea);
1270 raw_spin_lock(&ghes_notify_lock_sea);
1271 rv = ghes_in_nmi_spool_from_list(&ghes_sea, FIX_APEI_GHES_SEA);
1272 raw_spin_unlock(&ghes_notify_lock_sea);
1277 static void ghes_sea_add(struct ghes *ghes)
1279 mutex_lock(&ghes_list_mutex);
1280 list_add_rcu(&ghes->list, &ghes_sea);
1281 mutex_unlock(&ghes_list_mutex);
1284 static void ghes_sea_remove(struct ghes *ghes)
1286 mutex_lock(&ghes_list_mutex);
1287 list_del_rcu(&ghes->list);
1288 mutex_unlock(&ghes_list_mutex);
1291 #else /* CONFIG_ACPI_APEI_SEA */
1292 static inline void ghes_sea_add(struct ghes *ghes) { }
1293 static inline void ghes_sea_remove(struct ghes *ghes) { }
1294 #endif /* CONFIG_ACPI_APEI_SEA */
1296 #ifdef CONFIG_HAVE_ACPI_APEI_NMI
1298 * NMI may be triggered on any CPU, so ghes_in_nmi is used for
1299 * having only one concurrent reader.
1301 static atomic_t ghes_in_nmi = ATOMIC_INIT(0);
1303 static LIST_HEAD(ghes_nmi);
1305 static int ghes_notify_nmi(unsigned int cmd, struct pt_regs *regs)
1307 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_nmi);
1310 if (!atomic_add_unless(&ghes_in_nmi, 1, 1))
1313 raw_spin_lock(&ghes_notify_lock_nmi);
1314 if (!ghes_in_nmi_spool_from_list(&ghes_nmi, FIX_APEI_GHES_NMI))
1316 raw_spin_unlock(&ghes_notify_lock_nmi);
1318 atomic_dec(&ghes_in_nmi);
1322 static void ghes_nmi_add(struct ghes *ghes)
1324 mutex_lock(&ghes_list_mutex);
1325 if (list_empty(&ghes_nmi))
1326 register_nmi_handler(NMI_LOCAL, ghes_notify_nmi, 0, "ghes");
1327 list_add_rcu(&ghes->list, &ghes_nmi);
1328 mutex_unlock(&ghes_list_mutex);
1331 static void ghes_nmi_remove(struct ghes *ghes)
1333 mutex_lock(&ghes_list_mutex);
1334 list_del_rcu(&ghes->list);
1335 if (list_empty(&ghes_nmi))
1336 unregister_nmi_handler(NMI_LOCAL, "ghes");
1337 mutex_unlock(&ghes_list_mutex);
1339 * To synchronize with NMI handler, ghes can only be
1340 * freed after NMI handler finishes.
1344 #else /* CONFIG_HAVE_ACPI_APEI_NMI */
1345 static inline void ghes_nmi_add(struct ghes *ghes) { }
1346 static inline void ghes_nmi_remove(struct ghes *ghes) { }
1347 #endif /* CONFIG_HAVE_ACPI_APEI_NMI */
1349 static void ghes_nmi_init_cxt(void)
1351 init_irq_work(&ghes_proc_irq_work, ghes_proc_in_irq);
1354 static int __ghes_sdei_callback(struct ghes *ghes,
1355 enum fixed_addresses fixmap_idx)
1357 if (!ghes_in_nmi_queue_one_entry(ghes, fixmap_idx)) {
1358 irq_work_queue(&ghes_proc_irq_work);
1366 static int ghes_sdei_normal_callback(u32 event_num, struct pt_regs *regs,
1369 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sdei_normal);
1370 struct ghes *ghes = arg;
1373 raw_spin_lock(&ghes_notify_lock_sdei_normal);
1374 err = __ghes_sdei_callback(ghes, FIX_APEI_GHES_SDEI_NORMAL);
1375 raw_spin_unlock(&ghes_notify_lock_sdei_normal);
1380 static int ghes_sdei_critical_callback(u32 event_num, struct pt_regs *regs,
1383 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sdei_critical);
1384 struct ghes *ghes = arg;
1387 raw_spin_lock(&ghes_notify_lock_sdei_critical);
1388 err = __ghes_sdei_callback(ghes, FIX_APEI_GHES_SDEI_CRITICAL);
1389 raw_spin_unlock(&ghes_notify_lock_sdei_critical);
1394 static int apei_sdei_register_ghes(struct ghes *ghes)
1396 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE))
1399 return sdei_register_ghes(ghes, ghes_sdei_normal_callback,
1400 ghes_sdei_critical_callback);
1403 static int apei_sdei_unregister_ghes(struct ghes *ghes)
1405 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE))
1408 return sdei_unregister_ghes(ghes);
1411 static int ghes_probe(struct platform_device *ghes_dev)
1413 struct acpi_hest_generic *generic;
1414 struct ghes *ghes = NULL;
1415 unsigned long flags;
1419 generic = *(struct acpi_hest_generic **)ghes_dev->dev.platform_data;
1420 if (!generic->enabled)
1423 switch (generic->notify.type) {
1424 case ACPI_HEST_NOTIFY_POLLED:
1425 case ACPI_HEST_NOTIFY_EXTERNAL:
1426 case ACPI_HEST_NOTIFY_SCI:
1427 case ACPI_HEST_NOTIFY_GSIV:
1428 case ACPI_HEST_NOTIFY_GPIO:
1431 case ACPI_HEST_NOTIFY_SEA:
1432 if (!IS_ENABLED(CONFIG_ACPI_APEI_SEA)) {
1433 pr_warn(GHES_PFX "Generic hardware error source: %d notified via SEA is not supported\n",
1434 generic->header.source_id);
1439 case ACPI_HEST_NOTIFY_NMI:
1440 if (!IS_ENABLED(CONFIG_HAVE_ACPI_APEI_NMI)) {
1441 pr_warn(GHES_PFX "Generic hardware error source: %d notified via NMI interrupt is not supported!\n",
1442 generic->header.source_id);
1446 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1447 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE)) {
1448 pr_warn(GHES_PFX "Generic hardware error source: %d notified via SDE Interface is not supported!\n",
1449 generic->header.source_id);
1453 case ACPI_HEST_NOTIFY_LOCAL:
1454 pr_warn(GHES_PFX "Generic hardware error source: %d notified via local interrupt is not supported!\n",
1455 generic->header.source_id);
1458 pr_warn(FW_WARN GHES_PFX "Unknown notification type: %u for generic hardware error source: %d\n",
1459 generic->notify.type, generic->header.source_id);
1464 if (generic->error_block_length <
1465 sizeof(struct acpi_hest_generic_status)) {
1466 pr_warn(FW_BUG GHES_PFX "Invalid error block length: %u for generic hardware error source: %d\n",
1467 generic->error_block_length, generic->header.source_id);
1470 ghes = ghes_new(generic);
1477 switch (generic->notify.type) {
1478 case ACPI_HEST_NOTIFY_POLLED:
1479 timer_setup(&ghes->timer, ghes_poll_func, 0);
1480 ghes_add_timer(ghes);
1482 case ACPI_HEST_NOTIFY_EXTERNAL:
1483 /* External interrupt vector is GSI */
1484 rc = acpi_gsi_to_irq(generic->notify.vector, &ghes->irq);
1486 pr_err(GHES_PFX "Failed to map GSI to IRQ for generic hardware error source: %d\n",
1487 generic->header.source_id);
1490 rc = request_irq(ghes->irq, ghes_irq_func, IRQF_SHARED,
1493 pr_err(GHES_PFX "Failed to register IRQ for generic hardware error source: %d\n",
1494 generic->header.source_id);
1499 case ACPI_HEST_NOTIFY_SCI:
1500 case ACPI_HEST_NOTIFY_GSIV:
1501 case ACPI_HEST_NOTIFY_GPIO:
1502 mutex_lock(&ghes_list_mutex);
1503 if (list_empty(&ghes_hed))
1504 register_acpi_hed_notifier(&ghes_notifier_hed);
1505 list_add_rcu(&ghes->list, &ghes_hed);
1506 mutex_unlock(&ghes_list_mutex);
1509 case ACPI_HEST_NOTIFY_SEA:
1512 case ACPI_HEST_NOTIFY_NMI:
1515 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1516 rc = apei_sdei_register_ghes(ghes);
1524 platform_set_drvdata(ghes_dev, ghes);
1526 ghes->dev = &ghes_dev->dev;
1528 mutex_lock(&ghes_devs_mutex);
1529 list_add_tail(&ghes->elist, &ghes_devs);
1530 mutex_unlock(&ghes_devs_mutex);
1532 /* Handle any pending errors right away */
1533 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
1535 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
1547 static int ghes_remove(struct platform_device *ghes_dev)
1551 struct acpi_hest_generic *generic;
1553 ghes = platform_get_drvdata(ghes_dev);
1554 generic = ghes->generic;
1556 ghes->flags |= GHES_EXITING;
1557 switch (generic->notify.type) {
1558 case ACPI_HEST_NOTIFY_POLLED:
1559 timer_shutdown_sync(&ghes->timer);
1561 case ACPI_HEST_NOTIFY_EXTERNAL:
1562 free_irq(ghes->irq, ghes);
1565 case ACPI_HEST_NOTIFY_SCI:
1566 case ACPI_HEST_NOTIFY_GSIV:
1567 case ACPI_HEST_NOTIFY_GPIO:
1568 mutex_lock(&ghes_list_mutex);
1569 list_del_rcu(&ghes->list);
1570 if (list_empty(&ghes_hed))
1571 unregister_acpi_hed_notifier(&ghes_notifier_hed);
1572 mutex_unlock(&ghes_list_mutex);
1576 case ACPI_HEST_NOTIFY_SEA:
1577 ghes_sea_remove(ghes);
1579 case ACPI_HEST_NOTIFY_NMI:
1580 ghes_nmi_remove(ghes);
1582 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1583 rc = apei_sdei_unregister_ghes(ghes);
1594 mutex_lock(&ghes_devs_mutex);
1595 list_del(&ghes->elist);
1596 mutex_unlock(&ghes_devs_mutex);
1603 static struct platform_driver ghes_platform_driver = {
1607 .probe = ghes_probe,
1608 .remove = ghes_remove,
1611 void __init acpi_ghes_init(void)
1620 switch (hest_disable) {
1621 case HEST_NOT_FOUND:
1624 pr_info(GHES_PFX "HEST is not enabled!\n");
1631 pr_info(GHES_PFX "GHES is not enabled!\n");
1635 ghes_nmi_init_cxt();
1637 rc = platform_driver_register(&ghes_platform_driver);
1641 rc = apei_osc_setup();
1642 if (rc == 0 && osc_sb_apei_support_acked)
1643 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit and WHEA _OSC.\n");
1644 else if (rc == 0 && !osc_sb_apei_support_acked)
1645 pr_info(GHES_PFX "APEI firmware first mode is enabled by WHEA _OSC.\n");
1646 else if (rc && osc_sb_apei_support_acked)
1647 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit.\n");
1649 pr_info(GHES_PFX "Failed to enable APEI firmware first mode.\n");
1653 * Known x86 systems that prefer GHES error reporting:
1655 static struct acpi_platform_list plat_list[] = {
1656 {"HPE ", "Server ", 0, ACPI_SIG_FADT, all_versions},
1660 struct list_head *ghes_get_devices(void)
1664 if (IS_ENABLED(CONFIG_X86)) {
1665 idx = acpi_match_platform_list(plat_list);
1667 if (!ghes_edac_force_enable)
1670 pr_warn_once("Force-loading ghes_edac on an unsupported platform. You're on your own!\n");
1672 } else if (list_empty(&ghes_devs)) {
1678 EXPORT_SYMBOL_GPL(ghes_get_devices);
1680 void ghes_register_report_chain(struct notifier_block *nb)
1682 atomic_notifier_chain_register(&ghes_report_chain, nb);
1684 EXPORT_SYMBOL_GPL(ghes_register_report_chain);
1686 void ghes_unregister_report_chain(struct notifier_block *nb)
1688 atomic_notifier_chain_unregister(&ghes_report_chain, nb);
1690 EXPORT_SYMBOL_GPL(ghes_unregister_report_chain);