2 * APEI Generic Hardware Error Source support
4 * Generic Hardware Error Source provides a way to report platform
5 * hardware errors (such as that from chipset). It works in so called
6 * "Firmware First" mode, that is, hardware errors are reported to
7 * firmware firstly, then reported to Linux by firmware. This way,
8 * some non-standard hardware error registers or non-standard hardware
9 * link can be checked by firmware to produce more hardware error
10 * information for Linux.
12 * For more information about Generic Hardware Error Source, please
13 * refer to ACPI Specification version 4.0, section 17.3.2.6
15 * Copyright 2010,2011 Intel Corp.
16 * Author: Huang Ying <ying.huang@intel.com>
18 * This program is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU General Public License version
20 * 2 as published by the Free Software Foundation;
22 * This program is distributed in the hope that it will be useful,
23 * but WITHOUT ANY WARRANTY; without even the implied warranty of
24 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
25 * GNU General Public License for more details.
28 #include <linux/kernel.h>
29 #include <linux/moduleparam.h>
30 #include <linux/init.h>
31 #include <linux/acpi.h>
33 #include <linux/interrupt.h>
34 #include <linux/timer.h>
35 #include <linux/cper.h>
36 #include <linux/kdebug.h>
37 #include <linux/platform_device.h>
38 #include <linux/mutex.h>
39 #include <linux/ratelimit.h>
40 #include <linux/vmalloc.h>
41 #include <linux/irq_work.h>
42 #include <linux/llist.h>
43 #include <linux/genalloc.h>
44 #include <linux/pci.h>
45 #include <linux/aer.h>
46 #include <linux/nmi.h>
47 #include <linux/sched/clock.h>
49 #include <acpi/ghes.h>
50 #include <acpi/apei.h>
51 #include <asm/tlbflush.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))
84 * This driver isn't really modular, however for the time being,
85 * continuing to use module_param is the easiest way to remain
86 * compatible with existing boot arg use cases.
89 module_param_named(disable, ghes_disable, bool, 0);
92 * All error sources notified with SCI shares one notifier function,
93 * so they need to be linked and checked one by one. This is applied
96 * RCU is used for these lists, so ghes_list_mutex is only used for
97 * list changing, not for traversing.
99 static LIST_HEAD(ghes_sci);
100 static DEFINE_MUTEX(ghes_list_mutex);
103 * Because the memory area used to transfer hardware error information
104 * from BIOS to Linux can be determined only in NMI, IRQ or timer
105 * handler, but general ioremap can not be used in atomic context, so
106 * a special version of atomic ioremap is implemented for that.
110 * Two virtual pages are used, one for IRQ/PROCESS context, the other for
111 * NMI context (optionally).
113 #ifdef CONFIG_HAVE_ACPI_APEI_NMI
114 #define GHES_IOREMAP_PAGES 2
116 #define GHES_IOREMAP_PAGES 1
118 #define GHES_IOREMAP_IRQ_PAGE(base) (base)
119 #define GHES_IOREMAP_NMI_PAGE(base) ((base) + PAGE_SIZE)
121 /* virtual memory area for atomic ioremap */
122 static struct vm_struct *ghes_ioremap_area;
124 * These 2 spinlock is used to prevent atomic ioremap virtual memory
125 * area from being mapped simultaneously.
127 static DEFINE_RAW_SPINLOCK(ghes_ioremap_lock_nmi);
128 static DEFINE_SPINLOCK(ghes_ioremap_lock_irq);
130 static struct gen_pool *ghes_estatus_pool;
131 static unsigned long ghes_estatus_pool_size_request;
133 static struct ghes_estatus_cache *ghes_estatus_caches[GHES_ESTATUS_CACHES_SIZE];
134 static atomic_t ghes_estatus_cache_alloced;
136 static int ghes_ioremap_init(void)
138 ghes_ioremap_area = __get_vm_area(PAGE_SIZE * GHES_IOREMAP_PAGES,
139 VM_IOREMAP, VMALLOC_START, VMALLOC_END);
140 if (!ghes_ioremap_area) {
141 pr_err(GHES_PFX "Failed to allocate virtual memory area for atomic ioremap.\n");
148 static void ghes_ioremap_exit(void)
150 free_vm_area(ghes_ioremap_area);
153 static void __iomem *ghes_ioremap_pfn_nmi(u64 pfn)
157 vaddr = (unsigned long)GHES_IOREMAP_NMI_PAGE(ghes_ioremap_area->addr);
158 ioremap_page_range(vaddr, vaddr + PAGE_SIZE,
159 pfn << PAGE_SHIFT, PAGE_KERNEL);
161 return (void __iomem *)vaddr;
164 static void __iomem *ghes_ioremap_pfn_irq(u64 pfn)
166 unsigned long vaddr, paddr;
169 vaddr = (unsigned long)GHES_IOREMAP_IRQ_PAGE(ghes_ioremap_area->addr);
171 paddr = pfn << PAGE_SHIFT;
172 prot = arch_apei_get_mem_attribute(paddr);
174 ioremap_page_range(vaddr, vaddr + PAGE_SIZE, paddr, prot);
176 return (void __iomem *)vaddr;
179 static void ghes_iounmap_nmi(void __iomem *vaddr_ptr)
181 unsigned long vaddr = (unsigned long __force)vaddr_ptr;
182 void *base = ghes_ioremap_area->addr;
184 BUG_ON(vaddr != (unsigned long)GHES_IOREMAP_NMI_PAGE(base));
185 unmap_kernel_range_noflush(vaddr, PAGE_SIZE);
186 arch_apei_flush_tlb_one(vaddr);
189 static void ghes_iounmap_irq(void __iomem *vaddr_ptr)
191 unsigned long vaddr = (unsigned long __force)vaddr_ptr;
192 void *base = ghes_ioremap_area->addr;
194 BUG_ON(vaddr != (unsigned long)GHES_IOREMAP_IRQ_PAGE(base));
195 unmap_kernel_range_noflush(vaddr, PAGE_SIZE);
196 arch_apei_flush_tlb_one(vaddr);
199 static int ghes_estatus_pool_init(void)
201 ghes_estatus_pool = gen_pool_create(GHES_ESTATUS_POOL_MIN_ALLOC_ORDER, -1);
202 if (!ghes_estatus_pool)
207 static void ghes_estatus_pool_free_chunk_page(struct gen_pool *pool,
208 struct gen_pool_chunk *chunk,
211 free_page(chunk->start_addr);
214 static void ghes_estatus_pool_exit(void)
216 gen_pool_for_each_chunk(ghes_estatus_pool,
217 ghes_estatus_pool_free_chunk_page, NULL);
218 gen_pool_destroy(ghes_estatus_pool);
221 static int ghes_estatus_pool_expand(unsigned long len)
223 unsigned long i, pages, size, addr;
226 ghes_estatus_pool_size_request += PAGE_ALIGN(len);
227 size = gen_pool_size(ghes_estatus_pool);
228 if (size >= ghes_estatus_pool_size_request)
230 pages = (ghes_estatus_pool_size_request - size) / PAGE_SIZE;
231 for (i = 0; i < pages; i++) {
232 addr = __get_free_page(GFP_KERNEL);
235 ret = gen_pool_add(ghes_estatus_pool, addr, PAGE_SIZE, -1);
243 static struct ghes *ghes_new(struct acpi_hest_generic *generic)
246 unsigned int error_block_length;
249 ghes = kzalloc(sizeof(*ghes), GFP_KERNEL);
251 return ERR_PTR(-ENOMEM);
252 ghes->generic = generic;
253 rc = apei_map_generic_address(&generic->error_status_address);
256 error_block_length = generic->error_block_length;
257 if (error_block_length > GHES_ESTATUS_MAX_SIZE) {
258 pr_warning(FW_WARN GHES_PFX
259 "Error status block length is too long: %u for "
260 "generic hardware error source: %d.\n",
261 error_block_length, generic->header.source_id);
262 error_block_length = GHES_ESTATUS_MAX_SIZE;
264 ghes->estatus = kmalloc(error_block_length, GFP_KERNEL);
265 if (!ghes->estatus) {
273 apei_unmap_generic_address(&generic->error_status_address);
279 static void ghes_fini(struct ghes *ghes)
281 kfree(ghes->estatus);
282 apei_unmap_generic_address(&ghes->generic->error_status_address);
285 static inline int ghes_severity(int severity)
288 case CPER_SEV_INFORMATIONAL:
290 case CPER_SEV_CORRECTED:
291 return GHES_SEV_CORRECTED;
292 case CPER_SEV_RECOVERABLE:
293 return GHES_SEV_RECOVERABLE;
295 return GHES_SEV_PANIC;
297 /* Unknown, go panic */
298 return GHES_SEV_PANIC;
302 static void ghes_copy_tofrom_phys(void *buffer, u64 paddr, u32 len,
306 unsigned long flags = 0;
307 int in_nmi = in_nmi();
312 offset = paddr - (paddr & PAGE_MASK);
314 raw_spin_lock(&ghes_ioremap_lock_nmi);
315 vaddr = ghes_ioremap_pfn_nmi(paddr >> PAGE_SHIFT);
317 spin_lock_irqsave(&ghes_ioremap_lock_irq, flags);
318 vaddr = ghes_ioremap_pfn_irq(paddr >> PAGE_SHIFT);
320 trunk = PAGE_SIZE - offset;
321 trunk = min(trunk, len);
323 memcpy_fromio(buffer, vaddr + offset, trunk);
325 memcpy_toio(vaddr + offset, buffer, trunk);
330 ghes_iounmap_nmi(vaddr);
331 raw_spin_unlock(&ghes_ioremap_lock_nmi);
333 ghes_iounmap_irq(vaddr);
334 spin_unlock_irqrestore(&ghes_ioremap_lock_irq, flags);
339 static int ghes_read_estatus(struct ghes *ghes, int silent)
341 struct acpi_hest_generic *g = ghes->generic;
346 rc = apei_read(&buf_paddr, &g->error_status_address);
348 if (!silent && printk_ratelimit())
349 pr_warning(FW_WARN GHES_PFX
350 "Failed to read error status block address for hardware error source: %d.\n",
351 g->header.source_id);
357 ghes_copy_tofrom_phys(ghes->estatus, buf_paddr,
358 sizeof(*ghes->estatus), 1);
359 if (!ghes->estatus->block_status)
362 ghes->buffer_paddr = buf_paddr;
363 ghes->flags |= GHES_TO_CLEAR;
366 len = cper_estatus_len(ghes->estatus);
367 if (len < sizeof(*ghes->estatus))
369 if (len > ghes->generic->error_block_length)
371 if (cper_estatus_check_header(ghes->estatus))
373 ghes_copy_tofrom_phys(ghes->estatus + 1,
374 buf_paddr + sizeof(*ghes->estatus),
375 len - sizeof(*ghes->estatus), 1);
376 if (cper_estatus_check(ghes->estatus))
381 if (rc && !silent && printk_ratelimit())
382 pr_warning(FW_WARN GHES_PFX
383 "Failed to read error status block!\n");
387 static void ghes_clear_estatus(struct ghes *ghes)
389 ghes->estatus->block_status = 0;
390 if (!(ghes->flags & GHES_TO_CLEAR))
392 ghes_copy_tofrom_phys(ghes->estatus, ghes->buffer_paddr,
393 sizeof(ghes->estatus->block_status), 0);
394 ghes->flags &= ~GHES_TO_CLEAR;
397 static void ghes_handle_memory_failure(struct acpi_hest_generic_data *gdata, int sev)
399 #ifdef CONFIG_ACPI_APEI_MEMORY_FAILURE
402 int sec_sev = ghes_severity(gdata->error_severity);
403 struct cper_sec_mem_err *mem_err;
404 mem_err = (struct cper_sec_mem_err *)(gdata + 1);
406 if (!(mem_err->validation_bits & CPER_MEM_VALID_PA))
409 pfn = mem_err->physical_addr >> PAGE_SHIFT;
410 if (!pfn_valid(pfn)) {
411 pr_warn_ratelimited(FW_WARN GHES_PFX
412 "Invalid address in generic error data: %#llx\n",
413 mem_err->physical_addr);
417 /* iff following two events can be handled properly by now */
418 if (sec_sev == GHES_SEV_CORRECTED &&
419 (gdata->flags & CPER_SEC_ERROR_THRESHOLD_EXCEEDED))
420 flags = MF_SOFT_OFFLINE;
421 if (sev == GHES_SEV_RECOVERABLE && sec_sev == GHES_SEV_RECOVERABLE)
425 memory_failure_queue(pfn, 0, flags);
429 static void ghes_do_proc(struct ghes *ghes,
430 const struct acpi_hest_generic_status *estatus)
433 struct acpi_hest_generic_data *gdata;
435 sev = ghes_severity(estatus->error_severity);
436 apei_estatus_for_each_section(estatus, gdata) {
437 sec_sev = ghes_severity(gdata->error_severity);
438 if (!uuid_le_cmp(*(uuid_le *)gdata->section_type,
439 CPER_SEC_PLATFORM_MEM)) {
440 struct cper_sec_mem_err *mem_err;
441 mem_err = (struct cper_sec_mem_err *)(gdata+1);
442 ghes_edac_report_mem_error(ghes, sev, mem_err);
444 arch_apei_report_mem_error(sev, mem_err);
445 ghes_handle_memory_failure(gdata, sev);
447 #ifdef CONFIG_ACPI_APEI_PCIEAER
448 else if (!uuid_le_cmp(*(uuid_le *)gdata->section_type,
450 struct cper_sec_pcie *pcie_err;
451 pcie_err = (struct cper_sec_pcie *)(gdata+1);
452 if (sev == GHES_SEV_RECOVERABLE &&
453 sec_sev == GHES_SEV_RECOVERABLE &&
454 pcie_err->validation_bits & CPER_PCIE_VALID_DEVICE_ID &&
455 pcie_err->validation_bits & CPER_PCIE_VALID_AER_INFO) {
459 devfn = PCI_DEVFN(pcie_err->device_id.device,
460 pcie_err->device_id.function);
461 aer_severity = cper_severity_to_aer(gdata->error_severity);
464 * If firmware reset the component to contain
465 * the error, we must reinitialize it before
466 * use, so treat it as a fatal AER error.
468 if (gdata->flags & CPER_SEC_RESET)
469 aer_severity = AER_FATAL;
471 aer_recover_queue(pcie_err->device_id.segment,
472 pcie_err->device_id.bus,
474 (struct aer_capability_regs *)
483 static void __ghes_print_estatus(const char *pfx,
484 const struct acpi_hest_generic *generic,
485 const struct acpi_hest_generic_status *estatus)
487 static atomic_t seqno;
488 unsigned int curr_seqno;
492 if (ghes_severity(estatus->error_severity) <=
498 curr_seqno = atomic_inc_return(&seqno);
499 snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}" HW_ERR, pfx, curr_seqno);
500 printk("%s""Hardware error from APEI Generic Hardware Error Source: %d\n",
501 pfx_seq, generic->header.source_id);
502 cper_estatus_print(pfx_seq, estatus);
505 static int ghes_print_estatus(const char *pfx,
506 const struct acpi_hest_generic *generic,
507 const struct acpi_hest_generic_status *estatus)
509 /* Not more than 2 messages every 5 seconds */
510 static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2);
511 static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2);
512 struct ratelimit_state *ratelimit;
514 if (ghes_severity(estatus->error_severity) <= GHES_SEV_CORRECTED)
515 ratelimit = &ratelimit_corrected;
517 ratelimit = &ratelimit_uncorrected;
518 if (__ratelimit(ratelimit)) {
519 __ghes_print_estatus(pfx, generic, estatus);
526 * GHES error status reporting throttle, to report more kinds of
527 * errors, instead of just most frequently occurred errors.
529 static int ghes_estatus_cached(struct acpi_hest_generic_status *estatus)
533 unsigned long long now;
534 struct ghes_estatus_cache *cache;
535 struct acpi_hest_generic_status *cache_estatus;
537 len = cper_estatus_len(estatus);
539 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
540 cache = rcu_dereference(ghes_estatus_caches[i]);
543 if (len != cache->estatus_len)
545 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
546 if (memcmp(estatus, cache_estatus, len))
548 atomic_inc(&cache->count);
550 if (now - cache->time_in < GHES_ESTATUS_IN_CACHE_MAX_NSEC)
558 static struct ghes_estatus_cache *ghes_estatus_cache_alloc(
559 struct acpi_hest_generic *generic,
560 struct acpi_hest_generic_status *estatus)
564 struct ghes_estatus_cache *cache;
565 struct acpi_hest_generic_status *cache_estatus;
567 alloced = atomic_add_return(1, &ghes_estatus_cache_alloced);
568 if (alloced > GHES_ESTATUS_CACHE_ALLOCED_MAX) {
569 atomic_dec(&ghes_estatus_cache_alloced);
572 len = cper_estatus_len(estatus);
573 cache_len = GHES_ESTATUS_CACHE_LEN(len);
574 cache = (void *)gen_pool_alloc(ghes_estatus_pool, cache_len);
576 atomic_dec(&ghes_estatus_cache_alloced);
579 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
580 memcpy(cache_estatus, estatus, len);
581 cache->estatus_len = len;
582 atomic_set(&cache->count, 0);
583 cache->generic = generic;
584 cache->time_in = sched_clock();
588 static void ghes_estatus_cache_free(struct ghes_estatus_cache *cache)
592 len = cper_estatus_len(GHES_ESTATUS_FROM_CACHE(cache));
593 len = GHES_ESTATUS_CACHE_LEN(len);
594 gen_pool_free(ghes_estatus_pool, (unsigned long)cache, len);
595 atomic_dec(&ghes_estatus_cache_alloced);
598 static void ghes_estatus_cache_rcu_free(struct rcu_head *head)
600 struct ghes_estatus_cache *cache;
602 cache = container_of(head, struct ghes_estatus_cache, rcu);
603 ghes_estatus_cache_free(cache);
606 static void ghes_estatus_cache_add(
607 struct acpi_hest_generic *generic,
608 struct acpi_hest_generic_status *estatus)
610 int i, slot = -1, count;
611 unsigned long long now, duration, period, max_period = 0;
612 struct ghes_estatus_cache *cache, *slot_cache = NULL, *new_cache;
614 new_cache = ghes_estatus_cache_alloc(generic, estatus);
615 if (new_cache == NULL)
619 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
620 cache = rcu_dereference(ghes_estatus_caches[i]);
626 duration = now - cache->time_in;
627 if (duration >= GHES_ESTATUS_IN_CACHE_MAX_NSEC) {
632 count = atomic_read(&cache->count);
634 do_div(period, (count + 1));
635 if (period > max_period) {
641 /* new_cache must be put into array after its contents are written */
643 if (slot != -1 && cmpxchg(ghes_estatus_caches + slot,
644 slot_cache, new_cache) == slot_cache) {
646 call_rcu(&slot_cache->rcu, ghes_estatus_cache_rcu_free);
648 ghes_estatus_cache_free(new_cache);
652 static int ghes_proc(struct ghes *ghes)
656 rc = ghes_read_estatus(ghes, 0);
659 if (!ghes_estatus_cached(ghes->estatus)) {
660 if (ghes_print_estatus(NULL, ghes->generic, ghes->estatus))
661 ghes_estatus_cache_add(ghes->generic, ghes->estatus);
663 ghes_do_proc(ghes, ghes->estatus);
665 ghes_clear_estatus(ghes);
669 static void ghes_add_timer(struct ghes *ghes)
671 struct acpi_hest_generic *g = ghes->generic;
672 unsigned long expire;
674 if (!g->notify.poll_interval) {
675 pr_warning(FW_WARN GHES_PFX "Poll interval is 0 for generic hardware error source: %d, disabled.\n",
676 g->header.source_id);
679 expire = jiffies + msecs_to_jiffies(g->notify.poll_interval);
680 ghes->timer.expires = round_jiffies_relative(expire);
681 add_timer(&ghes->timer);
684 static void ghes_poll_func(unsigned long data)
686 struct ghes *ghes = (void *)data;
689 if (!(ghes->flags & GHES_EXITING))
690 ghes_add_timer(ghes);
693 static irqreturn_t ghes_irq_func(int irq, void *data)
695 struct ghes *ghes = data;
698 rc = ghes_proc(ghes);
705 static int ghes_notify_sci(struct notifier_block *this,
706 unsigned long event, void *data)
709 int ret = NOTIFY_DONE;
712 list_for_each_entry_rcu(ghes, &ghes_sci, list) {
713 if (!ghes_proc(ghes))
721 static struct notifier_block ghes_notifier_sci = {
722 .notifier_call = ghes_notify_sci,
725 #ifdef CONFIG_HAVE_ACPI_APEI_NMI
727 * printk is not safe in NMI context. So in NMI handler, we allocate
728 * required memory from lock-less memory allocator
729 * (ghes_estatus_pool), save estatus into it, put them into lock-less
730 * list (ghes_estatus_llist), then delay printk into IRQ context via
731 * irq_work (ghes_proc_irq_work). ghes_estatus_size_request record
732 * required pool size by all NMI error source.
734 static struct llist_head ghes_estatus_llist;
735 static struct irq_work ghes_proc_irq_work;
738 * NMI may be triggered on any CPU, so ghes_in_nmi is used for
739 * having only one concurrent reader.
741 static atomic_t ghes_in_nmi = ATOMIC_INIT(0);
743 static LIST_HEAD(ghes_nmi);
745 static int ghes_panic_timeout __read_mostly = 30;
747 static void ghes_proc_in_irq(struct irq_work *irq_work)
749 struct llist_node *llnode, *next;
750 struct ghes_estatus_node *estatus_node;
751 struct acpi_hest_generic *generic;
752 struct acpi_hest_generic_status *estatus;
755 llnode = llist_del_all(&ghes_estatus_llist);
757 * Because the time order of estatus in list is reversed,
758 * revert it back to proper order.
760 llnode = llist_reverse_order(llnode);
763 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
765 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
766 len = cper_estatus_len(estatus);
767 node_len = GHES_ESTATUS_NODE_LEN(len);
768 ghes_do_proc(estatus_node->ghes, estatus);
769 if (!ghes_estatus_cached(estatus)) {
770 generic = estatus_node->generic;
771 if (ghes_print_estatus(NULL, generic, estatus))
772 ghes_estatus_cache_add(generic, estatus);
774 gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node,
780 static void ghes_print_queued_estatus(void)
782 struct llist_node *llnode;
783 struct ghes_estatus_node *estatus_node;
784 struct acpi_hest_generic *generic;
785 struct acpi_hest_generic_status *estatus;
788 llnode = llist_del_all(&ghes_estatus_llist);
790 * Because the time order of estatus in list is reversed,
791 * revert it back to proper order.
793 llnode = llist_reverse_order(llnode);
795 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
797 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
798 len = cper_estatus_len(estatus);
799 node_len = GHES_ESTATUS_NODE_LEN(len);
800 generic = estatus_node->generic;
801 ghes_print_estatus(NULL, generic, estatus);
802 llnode = llnode->next;
806 /* Save estatus for further processing in IRQ context */
807 static void __process_error(struct ghes *ghes)
809 #ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
811 struct ghes_estatus_node *estatus_node;
812 struct acpi_hest_generic_status *estatus;
814 if (ghes_estatus_cached(ghes->estatus))
817 len = cper_estatus_len(ghes->estatus);
818 node_len = GHES_ESTATUS_NODE_LEN(len);
820 estatus_node = (void *)gen_pool_alloc(ghes_estatus_pool, node_len);
824 estatus_node->ghes = ghes;
825 estatus_node->generic = ghes->generic;
826 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
827 memcpy(estatus, ghes->estatus, len);
828 llist_add(&estatus_node->llnode, &ghes_estatus_llist);
832 static void __ghes_panic(struct ghes *ghes)
835 ghes_print_queued_estatus();
836 __ghes_print_estatus(KERN_EMERG, ghes->generic, ghes->estatus);
838 /* reboot to log the error! */
839 if (panic_timeout == 0)
840 panic_timeout = ghes_panic_timeout;
841 panic("Fatal hardware error!");
844 static int ghes_notify_nmi(unsigned int cmd, struct pt_regs *regs)
847 int sev, ret = NMI_DONE;
849 if (!atomic_add_unless(&ghes_in_nmi, 1, 1))
852 list_for_each_entry_rcu(ghes, &ghes_nmi, list) {
853 if (ghes_read_estatus(ghes, 1)) {
854 ghes_clear_estatus(ghes);
860 sev = ghes_severity(ghes->estatus->error_severity);
861 if (sev >= GHES_SEV_PANIC)
864 if (!(ghes->flags & GHES_TO_CLEAR))
867 __process_error(ghes);
868 ghes_clear_estatus(ghes);
871 #ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
872 if (ret == NMI_HANDLED)
873 irq_work_queue(&ghes_proc_irq_work);
875 atomic_dec(&ghes_in_nmi);
879 static unsigned long ghes_esource_prealloc_size(
880 const struct acpi_hest_generic *generic)
882 unsigned long block_length, prealloc_records, prealloc_size;
884 block_length = min_t(unsigned long, generic->error_block_length,
885 GHES_ESTATUS_MAX_SIZE);
886 prealloc_records = max_t(unsigned long,
887 generic->records_to_preallocate, 1);
888 prealloc_size = min_t(unsigned long, block_length * prealloc_records,
889 GHES_ESOURCE_PREALLOC_MAX_SIZE);
891 return prealloc_size;
894 static void ghes_estatus_pool_shrink(unsigned long len)
896 ghes_estatus_pool_size_request -= PAGE_ALIGN(len);
899 static void ghes_nmi_add(struct ghes *ghes)
903 len = ghes_esource_prealloc_size(ghes->generic);
904 ghes_estatus_pool_expand(len);
905 mutex_lock(&ghes_list_mutex);
906 if (list_empty(&ghes_nmi))
907 register_nmi_handler(NMI_LOCAL, ghes_notify_nmi, 0, "ghes");
908 list_add_rcu(&ghes->list, &ghes_nmi);
909 mutex_unlock(&ghes_list_mutex);
912 static void ghes_nmi_remove(struct ghes *ghes)
916 mutex_lock(&ghes_list_mutex);
917 list_del_rcu(&ghes->list);
918 if (list_empty(&ghes_nmi))
919 unregister_nmi_handler(NMI_LOCAL, "ghes");
920 mutex_unlock(&ghes_list_mutex);
922 * To synchronize with NMI handler, ghes can only be
923 * freed after NMI handler finishes.
926 len = ghes_esource_prealloc_size(ghes->generic);
927 ghes_estatus_pool_shrink(len);
930 static void ghes_nmi_init_cxt(void)
932 init_irq_work(&ghes_proc_irq_work, ghes_proc_in_irq);
934 #else /* CONFIG_HAVE_ACPI_APEI_NMI */
935 static inline void ghes_nmi_add(struct ghes *ghes)
937 pr_err(GHES_PFX "ID: %d, trying to add NMI notification which is not supported!\n",
938 ghes->generic->header.source_id);
942 static inline void ghes_nmi_remove(struct ghes *ghes)
944 pr_err(GHES_PFX "ID: %d, trying to remove NMI notification which is not supported!\n",
945 ghes->generic->header.source_id);
949 static inline void ghes_nmi_init_cxt(void)
952 #endif /* CONFIG_HAVE_ACPI_APEI_NMI */
954 static int ghes_probe(struct platform_device *ghes_dev)
956 struct acpi_hest_generic *generic;
957 struct ghes *ghes = NULL;
961 generic = *(struct acpi_hest_generic **)ghes_dev->dev.platform_data;
962 if (!generic->enabled)
965 switch (generic->notify.type) {
966 case ACPI_HEST_NOTIFY_POLLED:
967 case ACPI_HEST_NOTIFY_EXTERNAL:
968 case ACPI_HEST_NOTIFY_SCI:
970 case ACPI_HEST_NOTIFY_NMI:
971 if (!IS_ENABLED(CONFIG_HAVE_ACPI_APEI_NMI)) {
972 pr_warn(GHES_PFX "Generic hardware error source: %d notified via NMI interrupt is not supported!\n",
973 generic->header.source_id);
977 case ACPI_HEST_NOTIFY_LOCAL:
978 pr_warning(GHES_PFX "Generic hardware error source: %d notified via local interrupt is not supported!\n",
979 generic->header.source_id);
982 pr_warning(FW_WARN GHES_PFX "Unknown notification type: %u for generic hardware error source: %d\n",
983 generic->notify.type, generic->header.source_id);
988 if (generic->error_block_length <
989 sizeof(struct acpi_hest_generic_status)) {
990 pr_warning(FW_BUG GHES_PFX "Invalid error block length: %u for generic hardware error source: %d\n",
991 generic->error_block_length,
992 generic->header.source_id);
995 ghes = ghes_new(generic);
1002 rc = ghes_edac_register(ghes, &ghes_dev->dev);
1006 switch (generic->notify.type) {
1007 case ACPI_HEST_NOTIFY_POLLED:
1008 setup_deferrable_timer(&ghes->timer, ghes_poll_func,
1009 (unsigned long)ghes);
1010 ghes_add_timer(ghes);
1012 case ACPI_HEST_NOTIFY_EXTERNAL:
1013 /* External interrupt vector is GSI */
1014 rc = acpi_gsi_to_irq(generic->notify.vector, &ghes->irq);
1016 pr_err(GHES_PFX "Failed to map GSI to IRQ for generic hardware error source: %d\n",
1017 generic->header.source_id);
1018 goto err_edac_unreg;
1020 rc = request_irq(ghes->irq, ghes_irq_func, 0, "GHES IRQ", ghes);
1022 pr_err(GHES_PFX "Failed to register IRQ for generic hardware error source: %d\n",
1023 generic->header.source_id);
1024 goto err_edac_unreg;
1027 case ACPI_HEST_NOTIFY_SCI:
1028 mutex_lock(&ghes_list_mutex);
1029 if (list_empty(&ghes_sci))
1030 register_acpi_hed_notifier(&ghes_notifier_sci);
1031 list_add_rcu(&ghes->list, &ghes_sci);
1032 mutex_unlock(&ghes_list_mutex);
1034 case ACPI_HEST_NOTIFY_NMI:
1040 platform_set_drvdata(ghes_dev, ghes);
1044 ghes_edac_unregister(ghes);
1053 static int ghes_remove(struct platform_device *ghes_dev)
1056 struct acpi_hest_generic *generic;
1058 ghes = platform_get_drvdata(ghes_dev);
1059 generic = ghes->generic;
1061 ghes->flags |= GHES_EXITING;
1062 switch (generic->notify.type) {
1063 case ACPI_HEST_NOTIFY_POLLED:
1064 del_timer_sync(&ghes->timer);
1066 case ACPI_HEST_NOTIFY_EXTERNAL:
1067 free_irq(ghes->irq, ghes);
1069 case ACPI_HEST_NOTIFY_SCI:
1070 mutex_lock(&ghes_list_mutex);
1071 list_del_rcu(&ghes->list);
1072 if (list_empty(&ghes_sci))
1073 unregister_acpi_hed_notifier(&ghes_notifier_sci);
1074 mutex_unlock(&ghes_list_mutex);
1077 case ACPI_HEST_NOTIFY_NMI:
1078 ghes_nmi_remove(ghes);
1087 ghes_edac_unregister(ghes);
1091 platform_set_drvdata(ghes_dev, NULL);
1096 static struct platform_driver ghes_platform_driver = {
1100 .probe = ghes_probe,
1101 .remove = ghes_remove,
1104 static int __init ghes_init(void)
1112 pr_info(GHES_PFX "HEST is not enabled!\n");
1117 pr_info(GHES_PFX "GHES is not enabled!\n");
1121 ghes_nmi_init_cxt();
1123 rc = ghes_ioremap_init();
1127 rc = ghes_estatus_pool_init();
1129 goto err_ioremap_exit;
1131 rc = ghes_estatus_pool_expand(GHES_ESTATUS_CACHE_AVG_SIZE *
1132 GHES_ESTATUS_CACHE_ALLOCED_MAX);
1136 rc = platform_driver_register(&ghes_platform_driver);
1140 rc = apei_osc_setup();
1141 if (rc == 0 && osc_sb_apei_support_acked)
1142 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit and WHEA _OSC.\n");
1143 else if (rc == 0 && !osc_sb_apei_support_acked)
1144 pr_info(GHES_PFX "APEI firmware first mode is enabled by WHEA _OSC.\n");
1145 else if (rc && osc_sb_apei_support_acked)
1146 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit.\n");
1148 pr_info(GHES_PFX "Failed to enable APEI firmware first mode.\n");
1152 ghes_estatus_pool_exit();
1154 ghes_ioremap_exit();
1158 device_initcall(ghes_init);