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))
83 * NMI-like notifications vary by architecture, before the compiler can prune
84 * unused static functions it needs a value for these enums.
86 #ifndef CONFIG_ARM_SDE_INTERFACE
87 #define FIX_APEI_GHES_SDEI_NORMAL __end_of_fixed_addresses
88 #define FIX_APEI_GHES_SDEI_CRITICAL __end_of_fixed_addresses
91 static inline bool is_hest_type_generic_v2(struct ghes *ghes)
93 return ghes->generic->header.type == ACPI_HEST_TYPE_GENERIC_ERROR_V2;
97 * This driver isn't really modular, however for the time being,
98 * continuing to use module_param is the easiest way to remain
99 * compatible with existing boot arg use cases.
102 module_param_named(disable, ghes_disable, bool, 0);
105 * All error sources notified with HED (Hardware Error Device) share a
106 * single notifier callback, so they need to be linked and checked one
107 * by one. This holds true for NMI too.
109 * RCU is used for these lists, so ghes_list_mutex is only used for
110 * list changing, not for traversing.
112 static LIST_HEAD(ghes_hed);
113 static DEFINE_MUTEX(ghes_list_mutex);
116 * Because the memory area used to transfer hardware error information
117 * from BIOS to Linux can be determined only in NMI, IRQ or timer
118 * handler, but general ioremap can not be used in atomic context, so
119 * the fixmap is used instead.
121 * This spinlock is used to prevent the fixmap entry from being used
124 static DEFINE_SPINLOCK(ghes_notify_lock_irq);
126 static struct gen_pool *ghes_estatus_pool;
127 static unsigned long ghes_estatus_pool_size_request;
129 static struct ghes_estatus_cache *ghes_estatus_caches[GHES_ESTATUS_CACHES_SIZE];
130 static atomic_t ghes_estatus_cache_alloced;
132 static int ghes_panic_timeout __read_mostly = 30;
134 static void __iomem *ghes_map(u64 pfn, enum fixed_addresses fixmap_idx)
139 paddr = PFN_PHYS(pfn);
140 prot = arch_apei_get_mem_attribute(paddr);
141 __set_fixmap(fixmap_idx, paddr, prot);
143 return (void __iomem *) __fix_to_virt(fixmap_idx);
146 static void ghes_unmap(void __iomem *vaddr, enum fixed_addresses fixmap_idx)
148 int _idx = virt_to_fix((unsigned long)vaddr);
150 WARN_ON_ONCE(fixmap_idx != _idx);
151 clear_fixmap(fixmap_idx);
154 int ghes_estatus_pool_init(int num_ghes)
156 unsigned long addr, len;
159 ghes_estatus_pool = gen_pool_create(GHES_ESTATUS_POOL_MIN_ALLOC_ORDER, -1);
160 if (!ghes_estatus_pool)
163 len = GHES_ESTATUS_CACHE_AVG_SIZE * GHES_ESTATUS_CACHE_ALLOCED_MAX;
164 len += (num_ghes * GHES_ESOURCE_PREALLOC_MAX_SIZE);
166 ghes_estatus_pool_size_request = PAGE_ALIGN(len);
167 addr = (unsigned long)vmalloc(PAGE_ALIGN(len));
171 rc = gen_pool_add(ghes_estatus_pool, addr, PAGE_ALIGN(len), -1);
181 gen_pool_destroy(ghes_estatus_pool);
186 static int map_gen_v2(struct ghes *ghes)
188 return apei_map_generic_address(&ghes->generic_v2->read_ack_register);
191 static void unmap_gen_v2(struct ghes *ghes)
193 apei_unmap_generic_address(&ghes->generic_v2->read_ack_register);
196 static void ghes_ack_error(struct acpi_hest_generic_v2 *gv2)
201 rc = apei_read(&val, &gv2->read_ack_register);
205 val &= gv2->read_ack_preserve << gv2->read_ack_register.bit_offset;
206 val |= gv2->read_ack_write << gv2->read_ack_register.bit_offset;
208 apei_write(val, &gv2->read_ack_register);
211 static struct ghes *ghes_new(struct acpi_hest_generic *generic)
214 unsigned int error_block_length;
217 ghes = kzalloc(sizeof(*ghes), GFP_KERNEL);
219 return ERR_PTR(-ENOMEM);
221 ghes->generic = generic;
222 if (is_hest_type_generic_v2(ghes)) {
223 rc = map_gen_v2(ghes);
228 rc = apei_map_generic_address(&generic->error_status_address);
230 goto err_unmap_read_ack_addr;
231 error_block_length = generic->error_block_length;
232 if (error_block_length > GHES_ESTATUS_MAX_SIZE) {
233 pr_warn(FW_WARN GHES_PFX
234 "Error status block length is too long: %u for "
235 "generic hardware error source: %d.\n",
236 error_block_length, generic->header.source_id);
237 error_block_length = GHES_ESTATUS_MAX_SIZE;
239 ghes->estatus = kmalloc(error_block_length, GFP_KERNEL);
240 if (!ghes->estatus) {
242 goto err_unmap_status_addr;
247 err_unmap_status_addr:
248 apei_unmap_generic_address(&generic->error_status_address);
249 err_unmap_read_ack_addr:
250 if (is_hest_type_generic_v2(ghes))
257 static void ghes_fini(struct ghes *ghes)
259 kfree(ghes->estatus);
260 apei_unmap_generic_address(&ghes->generic->error_status_address);
261 if (is_hest_type_generic_v2(ghes))
265 static inline int ghes_severity(int severity)
268 case CPER_SEV_INFORMATIONAL:
270 case CPER_SEV_CORRECTED:
271 return GHES_SEV_CORRECTED;
272 case CPER_SEV_RECOVERABLE:
273 return GHES_SEV_RECOVERABLE;
275 return GHES_SEV_PANIC;
277 /* Unknown, go panic */
278 return GHES_SEV_PANIC;
282 static void ghes_copy_tofrom_phys(void *buffer, u64 paddr, u32 len,
284 enum fixed_addresses fixmap_idx)
291 offset = paddr - (paddr & PAGE_MASK);
292 vaddr = ghes_map(PHYS_PFN(paddr), fixmap_idx);
293 trunk = PAGE_SIZE - offset;
294 trunk = min(trunk, len);
296 memcpy_fromio(buffer, vaddr + offset, trunk);
298 memcpy_toio(vaddr + offset, buffer, trunk);
302 ghes_unmap(vaddr, fixmap_idx);
306 /* Check the top-level record header has an appropriate size. */
307 static int __ghes_check_estatus(struct ghes *ghes,
308 struct acpi_hest_generic_status *estatus)
310 u32 len = cper_estatus_len(estatus);
312 if (len < sizeof(*estatus)) {
313 pr_warn_ratelimited(FW_WARN GHES_PFX "Truncated error status block!\n");
317 if (len > ghes->generic->error_block_length) {
318 pr_warn_ratelimited(FW_WARN GHES_PFX "Invalid error status block length!\n");
322 if (cper_estatus_check_header(estatus)) {
323 pr_warn_ratelimited(FW_WARN GHES_PFX "Invalid CPER header!\n");
330 /* Read the CPER block, returning its address, and header in estatus. */
331 static int __ghes_peek_estatus(struct ghes *ghes,
332 struct acpi_hest_generic_status *estatus,
333 u64 *buf_paddr, enum fixed_addresses fixmap_idx)
335 struct acpi_hest_generic *g = ghes->generic;
338 rc = apei_read(buf_paddr, &g->error_status_address);
341 pr_warn_ratelimited(FW_WARN GHES_PFX
342 "Failed to read error status block address for hardware error source: %d.\n",
343 g->header.source_id);
349 ghes_copy_tofrom_phys(estatus, *buf_paddr, sizeof(*estatus), 1,
351 if (!estatus->block_status) {
359 static int __ghes_read_estatus(struct acpi_hest_generic_status *estatus,
360 u64 buf_paddr, enum fixed_addresses fixmap_idx,
363 ghes_copy_tofrom_phys(estatus, buf_paddr, buf_len, 1, fixmap_idx);
364 if (cper_estatus_check(estatus)) {
365 pr_warn_ratelimited(FW_WARN GHES_PFX
366 "Failed to read error status block!\n");
373 static int ghes_read_estatus(struct ghes *ghes,
374 struct acpi_hest_generic_status *estatus,
375 u64 *buf_paddr, enum fixed_addresses fixmap_idx)
379 rc = __ghes_peek_estatus(ghes, estatus, buf_paddr, fixmap_idx);
383 rc = __ghes_check_estatus(ghes, estatus);
387 return __ghes_read_estatus(estatus, *buf_paddr, fixmap_idx,
388 cper_estatus_len(estatus));
391 static void ghes_clear_estatus(struct ghes *ghes,
392 struct acpi_hest_generic_status *estatus,
393 u64 buf_paddr, enum fixed_addresses fixmap_idx)
395 estatus->block_status = 0;
400 ghes_copy_tofrom_phys(estatus, buf_paddr,
401 sizeof(estatus->block_status), 0,
405 * GHESv2 type HEST entries introduce support for error acknowledgment,
406 * so only acknowledge the error if this support is present.
408 if (is_hest_type_generic_v2(ghes))
409 ghes_ack_error(ghes->generic_v2);
413 * Called as task_work before returning to user-space.
414 * Ensure any queued work has been done before we return to the context that
415 * triggered the notification.
417 static void ghes_kick_task_work(struct callback_head *head)
419 struct acpi_hest_generic_status *estatus;
420 struct ghes_estatus_node *estatus_node;
423 estatus_node = container_of(head, struct ghes_estatus_node, task_work);
424 if (IS_ENABLED(CONFIG_ACPI_APEI_MEMORY_FAILURE))
425 memory_failure_queue_kick(estatus_node->task_work_cpu);
427 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
428 node_len = GHES_ESTATUS_NODE_LEN(cper_estatus_len(estatus));
429 gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node, node_len);
432 static bool ghes_handle_memory_failure(struct acpi_hest_generic_data *gdata,
437 int sec_sev = ghes_severity(gdata->error_severity);
438 struct cper_sec_mem_err *mem_err = acpi_hest_get_payload(gdata);
440 if (!IS_ENABLED(CONFIG_ACPI_APEI_MEMORY_FAILURE))
443 if (!(mem_err->validation_bits & CPER_MEM_VALID_PA))
446 pfn = mem_err->physical_addr >> PAGE_SHIFT;
447 if (!pfn_valid(pfn)) {
448 pr_warn_ratelimited(FW_WARN GHES_PFX
449 "Invalid address in generic error data: %#llx\n",
450 mem_err->physical_addr);
454 /* iff following two events can be handled properly by now */
455 if (sec_sev == GHES_SEV_CORRECTED &&
456 (gdata->flags & CPER_SEC_ERROR_THRESHOLD_EXCEEDED))
457 flags = MF_SOFT_OFFLINE;
458 if (sev == GHES_SEV_RECOVERABLE && sec_sev == GHES_SEV_RECOVERABLE)
462 memory_failure_queue(pfn, flags);
470 * PCIe AER errors need to be sent to the AER driver for reporting and
471 * recovery. The GHES severities map to the following AER severities and
472 * require the following handling:
474 * GHES_SEV_CORRECTABLE -> AER_CORRECTABLE
475 * These need to be reported by the AER driver but no recovery is
477 * GHES_SEV_RECOVERABLE -> AER_NONFATAL
478 * GHES_SEV_RECOVERABLE && CPER_SEC_RESET -> AER_FATAL
479 * These both need to be reported and recovered from by the AER driver.
480 * GHES_SEV_PANIC does not make it to this handling since the kernel must
483 static void ghes_handle_aer(struct acpi_hest_generic_data *gdata)
485 #ifdef CONFIG_ACPI_APEI_PCIEAER
486 struct cper_sec_pcie *pcie_err = acpi_hest_get_payload(gdata);
488 if (pcie_err->validation_bits & CPER_PCIE_VALID_DEVICE_ID &&
489 pcie_err->validation_bits & CPER_PCIE_VALID_AER_INFO) {
493 devfn = PCI_DEVFN(pcie_err->device_id.device,
494 pcie_err->device_id.function);
495 aer_severity = cper_severity_to_aer(gdata->error_severity);
498 * If firmware reset the component to contain
499 * the error, we must reinitialize it before
500 * use, so treat it as a fatal AER error.
502 if (gdata->flags & CPER_SEC_RESET)
503 aer_severity = AER_FATAL;
505 aer_recover_queue(pcie_err->device_id.segment,
506 pcie_err->device_id.bus,
508 (struct aer_capability_regs *)
514 static bool ghes_do_proc(struct ghes *ghes,
515 const struct acpi_hest_generic_status *estatus)
518 struct acpi_hest_generic_data *gdata;
520 const guid_t *fru_id = &guid_null;
524 sev = ghes_severity(estatus->error_severity);
525 apei_estatus_for_each_section(estatus, gdata) {
526 sec_type = (guid_t *)gdata->section_type;
527 sec_sev = ghes_severity(gdata->error_severity);
528 if (gdata->validation_bits & CPER_SEC_VALID_FRU_ID)
529 fru_id = (guid_t *)gdata->fru_id;
531 if (gdata->validation_bits & CPER_SEC_VALID_FRU_TEXT)
532 fru_text = gdata->fru_text;
534 if (guid_equal(sec_type, &CPER_SEC_PLATFORM_MEM)) {
535 struct cper_sec_mem_err *mem_err = acpi_hest_get_payload(gdata);
537 ghes_edac_report_mem_error(sev, mem_err);
539 arch_apei_report_mem_error(sev, mem_err);
540 queued = ghes_handle_memory_failure(gdata, sev);
542 else if (guid_equal(sec_type, &CPER_SEC_PCIE)) {
543 ghes_handle_aer(gdata);
545 else if (guid_equal(sec_type, &CPER_SEC_PROC_ARM)) {
546 struct cper_sec_proc_arm *err = acpi_hest_get_payload(gdata);
548 log_arm_hw_error(err);
550 void *err = acpi_hest_get_payload(gdata);
552 log_non_standard_event(sec_type, fru_id, fru_text,
554 gdata->error_data_length);
561 static void __ghes_print_estatus(const char *pfx,
562 const struct acpi_hest_generic *generic,
563 const struct acpi_hest_generic_status *estatus)
565 static atomic_t seqno;
566 unsigned int curr_seqno;
570 if (ghes_severity(estatus->error_severity) <=
576 curr_seqno = atomic_inc_return(&seqno);
577 snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}" HW_ERR, pfx, curr_seqno);
578 printk("%s""Hardware error from APEI Generic Hardware Error Source: %d\n",
579 pfx_seq, generic->header.source_id);
580 cper_estatus_print(pfx_seq, estatus);
583 static int ghes_print_estatus(const char *pfx,
584 const struct acpi_hest_generic *generic,
585 const struct acpi_hest_generic_status *estatus)
587 /* Not more than 2 messages every 5 seconds */
588 static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2);
589 static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2);
590 struct ratelimit_state *ratelimit;
592 if (ghes_severity(estatus->error_severity) <= GHES_SEV_CORRECTED)
593 ratelimit = &ratelimit_corrected;
595 ratelimit = &ratelimit_uncorrected;
596 if (__ratelimit(ratelimit)) {
597 __ghes_print_estatus(pfx, generic, estatus);
604 * GHES error status reporting throttle, to report more kinds of
605 * errors, instead of just most frequently occurred errors.
607 static int ghes_estatus_cached(struct acpi_hest_generic_status *estatus)
611 unsigned long long now;
612 struct ghes_estatus_cache *cache;
613 struct acpi_hest_generic_status *cache_estatus;
615 len = cper_estatus_len(estatus);
617 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
618 cache = rcu_dereference(ghes_estatus_caches[i]);
621 if (len != cache->estatus_len)
623 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
624 if (memcmp(estatus, cache_estatus, len))
626 atomic_inc(&cache->count);
628 if (now - cache->time_in < GHES_ESTATUS_IN_CACHE_MAX_NSEC)
636 static struct ghes_estatus_cache *ghes_estatus_cache_alloc(
637 struct acpi_hest_generic *generic,
638 struct acpi_hest_generic_status *estatus)
642 struct ghes_estatus_cache *cache;
643 struct acpi_hest_generic_status *cache_estatus;
645 alloced = atomic_add_return(1, &ghes_estatus_cache_alloced);
646 if (alloced > GHES_ESTATUS_CACHE_ALLOCED_MAX) {
647 atomic_dec(&ghes_estatus_cache_alloced);
650 len = cper_estatus_len(estatus);
651 cache_len = GHES_ESTATUS_CACHE_LEN(len);
652 cache = (void *)gen_pool_alloc(ghes_estatus_pool, cache_len);
654 atomic_dec(&ghes_estatus_cache_alloced);
657 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
658 memcpy(cache_estatus, estatus, len);
659 cache->estatus_len = len;
660 atomic_set(&cache->count, 0);
661 cache->generic = generic;
662 cache->time_in = sched_clock();
666 static void ghes_estatus_cache_free(struct ghes_estatus_cache *cache)
670 len = cper_estatus_len(GHES_ESTATUS_FROM_CACHE(cache));
671 len = GHES_ESTATUS_CACHE_LEN(len);
672 gen_pool_free(ghes_estatus_pool, (unsigned long)cache, len);
673 atomic_dec(&ghes_estatus_cache_alloced);
676 static void ghes_estatus_cache_rcu_free(struct rcu_head *head)
678 struct ghes_estatus_cache *cache;
680 cache = container_of(head, struct ghes_estatus_cache, rcu);
681 ghes_estatus_cache_free(cache);
684 static void ghes_estatus_cache_add(
685 struct acpi_hest_generic *generic,
686 struct acpi_hest_generic_status *estatus)
688 int i, slot = -1, count;
689 unsigned long long now, duration, period, max_period = 0;
690 struct ghes_estatus_cache *cache, *slot_cache = NULL, *new_cache;
692 new_cache = ghes_estatus_cache_alloc(generic, estatus);
693 if (new_cache == NULL)
697 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
698 cache = rcu_dereference(ghes_estatus_caches[i]);
704 duration = now - cache->time_in;
705 if (duration >= GHES_ESTATUS_IN_CACHE_MAX_NSEC) {
710 count = atomic_read(&cache->count);
712 do_div(period, (count + 1));
713 if (period > max_period) {
719 /* new_cache must be put into array after its contents are written */
721 if (slot != -1 && cmpxchg(ghes_estatus_caches + slot,
722 slot_cache, new_cache) == slot_cache) {
724 call_rcu(&slot_cache->rcu, ghes_estatus_cache_rcu_free);
726 ghes_estatus_cache_free(new_cache);
730 static void __ghes_panic(struct ghes *ghes,
731 struct acpi_hest_generic_status *estatus,
732 u64 buf_paddr, enum fixed_addresses fixmap_idx)
734 __ghes_print_estatus(KERN_EMERG, ghes->generic, estatus);
736 ghes_clear_estatus(ghes, estatus, buf_paddr, fixmap_idx);
738 /* reboot to log the error! */
740 panic_timeout = ghes_panic_timeout;
741 panic("Fatal hardware error!");
744 static int ghes_proc(struct ghes *ghes)
746 struct acpi_hest_generic_status *estatus = ghes->estatus;
750 rc = ghes_read_estatus(ghes, estatus, &buf_paddr, FIX_APEI_GHES_IRQ);
754 if (ghes_severity(estatus->error_severity) >= GHES_SEV_PANIC)
755 __ghes_panic(ghes, estatus, buf_paddr, FIX_APEI_GHES_IRQ);
757 if (!ghes_estatus_cached(estatus)) {
758 if (ghes_print_estatus(NULL, ghes->generic, estatus))
759 ghes_estatus_cache_add(ghes->generic, estatus);
761 ghes_do_proc(ghes, estatus);
764 ghes_clear_estatus(ghes, estatus, buf_paddr, FIX_APEI_GHES_IRQ);
769 static void ghes_add_timer(struct ghes *ghes)
771 struct acpi_hest_generic *g = ghes->generic;
772 unsigned long expire;
774 if (!g->notify.poll_interval) {
775 pr_warn(FW_WARN GHES_PFX "Poll interval is 0 for generic hardware error source: %d, disabled.\n",
776 g->header.source_id);
779 expire = jiffies + msecs_to_jiffies(g->notify.poll_interval);
780 ghes->timer.expires = round_jiffies_relative(expire);
781 add_timer(&ghes->timer);
784 static void ghes_poll_func(struct timer_list *t)
786 struct ghes *ghes = from_timer(ghes, t, timer);
789 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
791 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
792 if (!(ghes->flags & GHES_EXITING))
793 ghes_add_timer(ghes);
796 static irqreturn_t ghes_irq_func(int irq, void *data)
798 struct ghes *ghes = data;
802 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
803 rc = ghes_proc(ghes);
804 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
811 static int ghes_notify_hed(struct notifier_block *this, unsigned long event,
816 int ret = NOTIFY_DONE;
818 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
820 list_for_each_entry_rcu(ghes, &ghes_hed, list) {
821 if (!ghes_proc(ghes))
825 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
830 static struct notifier_block ghes_notifier_hed = {
831 .notifier_call = ghes_notify_hed,
835 * Handlers for CPER records may not be NMI safe. For example,
836 * memory_failure_queue() takes spinlocks and calls schedule_work_on().
837 * In any NMI-like handler, memory from ghes_estatus_pool is used to save
838 * estatus, and added to the ghes_estatus_llist. irq_work_queue() causes
839 * ghes_proc_in_irq() to run in IRQ context where each estatus in
840 * ghes_estatus_llist is processed.
842 * Memory from the ghes_estatus_pool is also used with the ghes_estatus_cache
843 * to suppress frequent messages.
845 static struct llist_head ghes_estatus_llist;
846 static struct irq_work ghes_proc_irq_work;
848 static void ghes_proc_in_irq(struct irq_work *irq_work)
850 struct llist_node *llnode, *next;
851 struct ghes_estatus_node *estatus_node;
852 struct acpi_hest_generic *generic;
853 struct acpi_hest_generic_status *estatus;
854 bool task_work_pending;
858 llnode = llist_del_all(&ghes_estatus_llist);
860 * Because the time order of estatus in list is reversed,
861 * revert it back to proper order.
863 llnode = llist_reverse_order(llnode);
866 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
868 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
869 len = cper_estatus_len(estatus);
870 node_len = GHES_ESTATUS_NODE_LEN(len);
871 task_work_pending = ghes_do_proc(estatus_node->ghes, estatus);
872 if (!ghes_estatus_cached(estatus)) {
873 generic = estatus_node->generic;
874 if (ghes_print_estatus(NULL, generic, estatus))
875 ghes_estatus_cache_add(generic, estatus);
878 if (task_work_pending && current->mm != &init_mm) {
879 estatus_node->task_work.func = ghes_kick_task_work;
880 estatus_node->task_work_cpu = smp_processor_id();
881 ret = task_work_add(current, &estatus_node->task_work,
884 estatus_node->task_work.func = NULL;
887 if (!estatus_node->task_work.func)
888 gen_pool_free(ghes_estatus_pool,
889 (unsigned long)estatus_node, node_len);
895 static void ghes_print_queued_estatus(void)
897 struct llist_node *llnode;
898 struct ghes_estatus_node *estatus_node;
899 struct acpi_hest_generic *generic;
900 struct acpi_hest_generic_status *estatus;
902 llnode = llist_del_all(&ghes_estatus_llist);
904 * Because the time order of estatus in list is reversed,
905 * revert it back to proper order.
907 llnode = llist_reverse_order(llnode);
909 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
911 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
912 generic = estatus_node->generic;
913 ghes_print_estatus(NULL, generic, estatus);
914 llnode = llnode->next;
918 static int ghes_in_nmi_queue_one_entry(struct ghes *ghes,
919 enum fixed_addresses fixmap_idx)
921 struct acpi_hest_generic_status *estatus, tmp_header;
922 struct ghes_estatus_node *estatus_node;
927 if (!IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG))
930 rc = __ghes_peek_estatus(ghes, &tmp_header, &buf_paddr, fixmap_idx);
932 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
936 rc = __ghes_check_estatus(ghes, &tmp_header);
938 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
942 len = cper_estatus_len(&tmp_header);
943 node_len = GHES_ESTATUS_NODE_LEN(len);
944 estatus_node = (void *)gen_pool_alloc(ghes_estatus_pool, node_len);
948 estatus_node->ghes = ghes;
949 estatus_node->generic = ghes->generic;
950 estatus_node->task_work.func = NULL;
951 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
953 if (__ghes_read_estatus(estatus, buf_paddr, fixmap_idx, len)) {
954 ghes_clear_estatus(ghes, estatus, buf_paddr, fixmap_idx);
959 sev = ghes_severity(estatus->error_severity);
960 if (sev >= GHES_SEV_PANIC) {
961 ghes_print_queued_estatus();
962 __ghes_panic(ghes, estatus, buf_paddr, fixmap_idx);
965 ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
967 /* This error has been reported before, don't process it again. */
968 if (ghes_estatus_cached(estatus))
971 llist_add(&estatus_node->llnode, &ghes_estatus_llist);
976 gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node,
982 static int ghes_in_nmi_spool_from_list(struct list_head *rcu_list,
983 enum fixed_addresses fixmap_idx)
989 list_for_each_entry_rcu(ghes, rcu_list, list) {
990 if (!ghes_in_nmi_queue_one_entry(ghes, fixmap_idx))
995 if (IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG) && !ret)
996 irq_work_queue(&ghes_proc_irq_work);
1001 #ifdef CONFIG_ACPI_APEI_SEA
1002 static LIST_HEAD(ghes_sea);
1005 * Return 0 only if one of the SEA error sources successfully reported an error
1006 * record sent from the firmware.
1008 int ghes_notify_sea(void)
1010 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sea);
1013 raw_spin_lock(&ghes_notify_lock_sea);
1014 rv = ghes_in_nmi_spool_from_list(&ghes_sea, FIX_APEI_GHES_SEA);
1015 raw_spin_unlock(&ghes_notify_lock_sea);
1020 static void ghes_sea_add(struct ghes *ghes)
1022 mutex_lock(&ghes_list_mutex);
1023 list_add_rcu(&ghes->list, &ghes_sea);
1024 mutex_unlock(&ghes_list_mutex);
1027 static void ghes_sea_remove(struct ghes *ghes)
1029 mutex_lock(&ghes_list_mutex);
1030 list_del_rcu(&ghes->list);
1031 mutex_unlock(&ghes_list_mutex);
1034 #else /* CONFIG_ACPI_APEI_SEA */
1035 static inline void ghes_sea_add(struct ghes *ghes) { }
1036 static inline void ghes_sea_remove(struct ghes *ghes) { }
1037 #endif /* CONFIG_ACPI_APEI_SEA */
1039 #ifdef CONFIG_HAVE_ACPI_APEI_NMI
1041 * NMI may be triggered on any CPU, so ghes_in_nmi is used for
1042 * having only one concurrent reader.
1044 static atomic_t ghes_in_nmi = ATOMIC_INIT(0);
1046 static LIST_HEAD(ghes_nmi);
1048 static int ghes_notify_nmi(unsigned int cmd, struct pt_regs *regs)
1050 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_nmi);
1053 if (!atomic_add_unless(&ghes_in_nmi, 1, 1))
1056 raw_spin_lock(&ghes_notify_lock_nmi);
1057 if (!ghes_in_nmi_spool_from_list(&ghes_nmi, FIX_APEI_GHES_NMI))
1059 raw_spin_unlock(&ghes_notify_lock_nmi);
1061 atomic_dec(&ghes_in_nmi);
1065 static void ghes_nmi_add(struct ghes *ghes)
1067 mutex_lock(&ghes_list_mutex);
1068 if (list_empty(&ghes_nmi))
1069 register_nmi_handler(NMI_LOCAL, ghes_notify_nmi, 0, "ghes");
1070 list_add_rcu(&ghes->list, &ghes_nmi);
1071 mutex_unlock(&ghes_list_mutex);
1074 static void ghes_nmi_remove(struct ghes *ghes)
1076 mutex_lock(&ghes_list_mutex);
1077 list_del_rcu(&ghes->list);
1078 if (list_empty(&ghes_nmi))
1079 unregister_nmi_handler(NMI_LOCAL, "ghes");
1080 mutex_unlock(&ghes_list_mutex);
1082 * To synchronize with NMI handler, ghes can only be
1083 * freed after NMI handler finishes.
1087 #else /* CONFIG_HAVE_ACPI_APEI_NMI */
1088 static inline void ghes_nmi_add(struct ghes *ghes) { }
1089 static inline void ghes_nmi_remove(struct ghes *ghes) { }
1090 #endif /* CONFIG_HAVE_ACPI_APEI_NMI */
1092 static void ghes_nmi_init_cxt(void)
1094 init_irq_work(&ghes_proc_irq_work, ghes_proc_in_irq);
1097 static int __ghes_sdei_callback(struct ghes *ghes,
1098 enum fixed_addresses fixmap_idx)
1100 if (!ghes_in_nmi_queue_one_entry(ghes, fixmap_idx)) {
1101 irq_work_queue(&ghes_proc_irq_work);
1109 static int ghes_sdei_normal_callback(u32 event_num, struct pt_regs *regs,
1112 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sdei_normal);
1113 struct ghes *ghes = arg;
1116 raw_spin_lock(&ghes_notify_lock_sdei_normal);
1117 err = __ghes_sdei_callback(ghes, FIX_APEI_GHES_SDEI_NORMAL);
1118 raw_spin_unlock(&ghes_notify_lock_sdei_normal);
1123 static int ghes_sdei_critical_callback(u32 event_num, struct pt_regs *regs,
1126 static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sdei_critical);
1127 struct ghes *ghes = arg;
1130 raw_spin_lock(&ghes_notify_lock_sdei_critical);
1131 err = __ghes_sdei_callback(ghes, FIX_APEI_GHES_SDEI_CRITICAL);
1132 raw_spin_unlock(&ghes_notify_lock_sdei_critical);
1137 static int apei_sdei_register_ghes(struct ghes *ghes)
1139 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE))
1142 return sdei_register_ghes(ghes, ghes_sdei_normal_callback,
1143 ghes_sdei_critical_callback);
1146 static int apei_sdei_unregister_ghes(struct ghes *ghes)
1148 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE))
1151 return sdei_unregister_ghes(ghes);
1154 static int ghes_probe(struct platform_device *ghes_dev)
1156 struct acpi_hest_generic *generic;
1157 struct ghes *ghes = NULL;
1158 unsigned long flags;
1162 generic = *(struct acpi_hest_generic **)ghes_dev->dev.platform_data;
1163 if (!generic->enabled)
1166 switch (generic->notify.type) {
1167 case ACPI_HEST_NOTIFY_POLLED:
1168 case ACPI_HEST_NOTIFY_EXTERNAL:
1169 case ACPI_HEST_NOTIFY_SCI:
1170 case ACPI_HEST_NOTIFY_GSIV:
1171 case ACPI_HEST_NOTIFY_GPIO:
1174 case ACPI_HEST_NOTIFY_SEA:
1175 if (!IS_ENABLED(CONFIG_ACPI_APEI_SEA)) {
1176 pr_warn(GHES_PFX "Generic hardware error source: %d notified via SEA is not supported\n",
1177 generic->header.source_id);
1182 case ACPI_HEST_NOTIFY_NMI:
1183 if (!IS_ENABLED(CONFIG_HAVE_ACPI_APEI_NMI)) {
1184 pr_warn(GHES_PFX "Generic hardware error source: %d notified via NMI interrupt is not supported!\n",
1185 generic->header.source_id);
1189 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1190 if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE)) {
1191 pr_warn(GHES_PFX "Generic hardware error source: %d notified via SDE Interface is not supported!\n",
1192 generic->header.source_id);
1196 case ACPI_HEST_NOTIFY_LOCAL:
1197 pr_warn(GHES_PFX "Generic hardware error source: %d notified via local interrupt is not supported!\n",
1198 generic->header.source_id);
1201 pr_warn(FW_WARN GHES_PFX "Unknown notification type: %u for generic hardware error source: %d\n",
1202 generic->notify.type, generic->header.source_id);
1207 if (generic->error_block_length <
1208 sizeof(struct acpi_hest_generic_status)) {
1209 pr_warn(FW_BUG GHES_PFX "Invalid error block length: %u for generic hardware error source: %d\n",
1210 generic->error_block_length, generic->header.source_id);
1213 ghes = ghes_new(generic);
1220 switch (generic->notify.type) {
1221 case ACPI_HEST_NOTIFY_POLLED:
1222 timer_setup(&ghes->timer, ghes_poll_func, 0);
1223 ghes_add_timer(ghes);
1225 case ACPI_HEST_NOTIFY_EXTERNAL:
1226 /* External interrupt vector is GSI */
1227 rc = acpi_gsi_to_irq(generic->notify.vector, &ghes->irq);
1229 pr_err(GHES_PFX "Failed to map GSI to IRQ for generic hardware error source: %d\n",
1230 generic->header.source_id);
1233 rc = request_irq(ghes->irq, ghes_irq_func, IRQF_SHARED,
1236 pr_err(GHES_PFX "Failed to register IRQ for generic hardware error source: %d\n",
1237 generic->header.source_id);
1242 case ACPI_HEST_NOTIFY_SCI:
1243 case ACPI_HEST_NOTIFY_GSIV:
1244 case ACPI_HEST_NOTIFY_GPIO:
1245 mutex_lock(&ghes_list_mutex);
1246 if (list_empty(&ghes_hed))
1247 register_acpi_hed_notifier(&ghes_notifier_hed);
1248 list_add_rcu(&ghes->list, &ghes_hed);
1249 mutex_unlock(&ghes_list_mutex);
1252 case ACPI_HEST_NOTIFY_SEA:
1255 case ACPI_HEST_NOTIFY_NMI:
1258 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1259 rc = apei_sdei_register_ghes(ghes);
1267 platform_set_drvdata(ghes_dev, ghes);
1269 ghes_edac_register(ghes, &ghes_dev->dev);
1271 /* Handle any pending errors right away */
1272 spin_lock_irqsave(&ghes_notify_lock_irq, flags);
1274 spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
1286 static int ghes_remove(struct platform_device *ghes_dev)
1290 struct acpi_hest_generic *generic;
1292 ghes = platform_get_drvdata(ghes_dev);
1293 generic = ghes->generic;
1295 ghes->flags |= GHES_EXITING;
1296 switch (generic->notify.type) {
1297 case ACPI_HEST_NOTIFY_POLLED:
1298 del_timer_sync(&ghes->timer);
1300 case ACPI_HEST_NOTIFY_EXTERNAL:
1301 free_irq(ghes->irq, ghes);
1304 case ACPI_HEST_NOTIFY_SCI:
1305 case ACPI_HEST_NOTIFY_GSIV:
1306 case ACPI_HEST_NOTIFY_GPIO:
1307 mutex_lock(&ghes_list_mutex);
1308 list_del_rcu(&ghes->list);
1309 if (list_empty(&ghes_hed))
1310 unregister_acpi_hed_notifier(&ghes_notifier_hed);
1311 mutex_unlock(&ghes_list_mutex);
1315 case ACPI_HEST_NOTIFY_SEA:
1316 ghes_sea_remove(ghes);
1318 case ACPI_HEST_NOTIFY_NMI:
1319 ghes_nmi_remove(ghes);
1321 case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1322 rc = apei_sdei_unregister_ghes(ghes);
1333 ghes_edac_unregister(ghes);
1337 platform_set_drvdata(ghes_dev, NULL);
1342 static struct platform_driver ghes_platform_driver = {
1346 .probe = ghes_probe,
1347 .remove = ghes_remove,
1350 static int __init ghes_init(void)
1357 switch (hest_disable) {
1358 case HEST_NOT_FOUND:
1361 pr_info(GHES_PFX "HEST is not enabled!\n");
1368 pr_info(GHES_PFX "GHES is not enabled!\n");
1372 ghes_nmi_init_cxt();
1374 rc = platform_driver_register(&ghes_platform_driver);
1378 rc = apei_osc_setup();
1379 if (rc == 0 && osc_sb_apei_support_acked)
1380 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit and WHEA _OSC.\n");
1381 else if (rc == 0 && !osc_sb_apei_support_acked)
1382 pr_info(GHES_PFX "APEI firmware first mode is enabled by WHEA _OSC.\n");
1383 else if (rc && osc_sb_apei_support_acked)
1384 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit.\n");
1386 pr_info(GHES_PFX "Failed to enable APEI firmware first mode.\n");
1392 device_initcall(ghes_init);