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/module.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>
48 #include <acpi/ghes.h>
49 #include <acpi/apei.h>
50 #include <asm/tlbflush.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 module_param_named(disable, ghes_disable, bool, 0);
86 * All error sources notified with SCI shares one notifier function,
87 * so they need to be linked and checked one by one. This is applied
90 * RCU is used for these lists, so ghes_list_mutex is only used for
91 * list changing, not for traversing.
93 static LIST_HEAD(ghes_sci);
94 static DEFINE_MUTEX(ghes_list_mutex);
97 * Because the memory area used to transfer hardware error information
98 * from BIOS to Linux can be determined only in NMI, IRQ or timer
99 * handler, but general ioremap can not be used in atomic context, so
100 * a special version of atomic ioremap is implemented for that.
104 * Two virtual pages are used, one for IRQ/PROCESS context, the other for
105 * NMI context (optionally).
107 #ifdef CONFIG_HAVE_ACPI_APEI_NMI
108 #define GHES_IOREMAP_PAGES 2
110 #define GHES_IOREMAP_PAGES 1
112 #define GHES_IOREMAP_IRQ_PAGE(base) (base)
113 #define GHES_IOREMAP_NMI_PAGE(base) ((base) + PAGE_SIZE)
115 /* virtual memory area for atomic ioremap */
116 static struct vm_struct *ghes_ioremap_area;
118 * These 2 spinlock is used to prevent atomic ioremap virtual memory
119 * area from being mapped simultaneously.
121 static DEFINE_RAW_SPINLOCK(ghes_ioremap_lock_nmi);
122 static DEFINE_SPINLOCK(ghes_ioremap_lock_irq);
124 static struct gen_pool *ghes_estatus_pool;
125 static unsigned long ghes_estatus_pool_size_request;
127 static struct ghes_estatus_cache *ghes_estatus_caches[GHES_ESTATUS_CACHES_SIZE];
128 static atomic_t ghes_estatus_cache_alloced;
130 static int ghes_ioremap_init(void)
132 ghes_ioremap_area = __get_vm_area(PAGE_SIZE * GHES_IOREMAP_PAGES,
133 VM_IOREMAP, VMALLOC_START, VMALLOC_END);
134 if (!ghes_ioremap_area) {
135 pr_err(GHES_PFX "Failed to allocate virtual memory area for atomic ioremap.\n");
142 static void ghes_ioremap_exit(void)
144 free_vm_area(ghes_ioremap_area);
147 static void __iomem *ghes_ioremap_pfn_nmi(u64 pfn)
151 vaddr = (unsigned long)GHES_IOREMAP_NMI_PAGE(ghes_ioremap_area->addr);
152 ioremap_page_range(vaddr, vaddr + PAGE_SIZE,
153 pfn << PAGE_SHIFT, PAGE_KERNEL);
155 return (void __iomem *)vaddr;
158 static void __iomem *ghes_ioremap_pfn_irq(u64 pfn)
162 vaddr = (unsigned long)GHES_IOREMAP_IRQ_PAGE(ghes_ioremap_area->addr);
163 ioremap_page_range(vaddr, vaddr + PAGE_SIZE,
164 pfn << PAGE_SHIFT, PAGE_KERNEL);
166 return (void __iomem *)vaddr;
169 static void ghes_iounmap_nmi(void __iomem *vaddr_ptr)
171 unsigned long vaddr = (unsigned long __force)vaddr_ptr;
172 void *base = ghes_ioremap_area->addr;
174 BUG_ON(vaddr != (unsigned long)GHES_IOREMAP_NMI_PAGE(base));
175 unmap_kernel_range_noflush(vaddr, PAGE_SIZE);
176 arch_apei_flush_tlb_one(vaddr);
179 static void ghes_iounmap_irq(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_IRQ_PAGE(base));
185 unmap_kernel_range_noflush(vaddr, PAGE_SIZE);
186 arch_apei_flush_tlb_one(vaddr);
189 static int ghes_estatus_pool_init(void)
191 ghes_estatus_pool = gen_pool_create(GHES_ESTATUS_POOL_MIN_ALLOC_ORDER, -1);
192 if (!ghes_estatus_pool)
197 static void ghes_estatus_pool_free_chunk_page(struct gen_pool *pool,
198 struct gen_pool_chunk *chunk,
201 free_page(chunk->start_addr);
204 static void ghes_estatus_pool_exit(void)
206 gen_pool_for_each_chunk(ghes_estatus_pool,
207 ghes_estatus_pool_free_chunk_page, NULL);
208 gen_pool_destroy(ghes_estatus_pool);
211 static int ghes_estatus_pool_expand(unsigned long len)
213 unsigned long i, pages, size, addr;
216 ghes_estatus_pool_size_request += PAGE_ALIGN(len);
217 size = gen_pool_size(ghes_estatus_pool);
218 if (size >= ghes_estatus_pool_size_request)
220 pages = (ghes_estatus_pool_size_request - size) / PAGE_SIZE;
221 for (i = 0; i < pages; i++) {
222 addr = __get_free_page(GFP_KERNEL);
225 ret = gen_pool_add(ghes_estatus_pool, addr, PAGE_SIZE, -1);
233 static struct ghes *ghes_new(struct acpi_hest_generic *generic)
236 unsigned int error_block_length;
239 ghes = kzalloc(sizeof(*ghes), GFP_KERNEL);
241 return ERR_PTR(-ENOMEM);
242 ghes->generic = generic;
243 rc = apei_map_generic_address(&generic->error_status_address);
246 error_block_length = generic->error_block_length;
247 if (error_block_length > GHES_ESTATUS_MAX_SIZE) {
248 pr_warning(FW_WARN GHES_PFX
249 "Error status block length is too long: %u for "
250 "generic hardware error source: %d.\n",
251 error_block_length, generic->header.source_id);
252 error_block_length = GHES_ESTATUS_MAX_SIZE;
254 ghes->estatus = kmalloc(error_block_length, GFP_KERNEL);
255 if (!ghes->estatus) {
263 apei_unmap_generic_address(&generic->error_status_address);
269 static void ghes_fini(struct ghes *ghes)
271 kfree(ghes->estatus);
272 apei_unmap_generic_address(&ghes->generic->error_status_address);
275 static inline int ghes_severity(int severity)
278 case CPER_SEV_INFORMATIONAL:
280 case CPER_SEV_CORRECTED:
281 return GHES_SEV_CORRECTED;
282 case CPER_SEV_RECOVERABLE:
283 return GHES_SEV_RECOVERABLE;
285 return GHES_SEV_PANIC;
287 /* Unknown, go panic */
288 return GHES_SEV_PANIC;
292 static void ghes_copy_tofrom_phys(void *buffer, u64 paddr, u32 len,
296 unsigned long flags = 0;
297 int in_nmi = in_nmi();
302 offset = paddr - (paddr & PAGE_MASK);
304 raw_spin_lock(&ghes_ioremap_lock_nmi);
305 vaddr = ghes_ioremap_pfn_nmi(paddr >> PAGE_SHIFT);
307 spin_lock_irqsave(&ghes_ioremap_lock_irq, flags);
308 vaddr = ghes_ioremap_pfn_irq(paddr >> PAGE_SHIFT);
310 trunk = PAGE_SIZE - offset;
311 trunk = min(trunk, len);
313 memcpy_fromio(buffer, vaddr + offset, trunk);
315 memcpy_toio(vaddr + offset, buffer, trunk);
320 ghes_iounmap_nmi(vaddr);
321 raw_spin_unlock(&ghes_ioremap_lock_nmi);
323 ghes_iounmap_irq(vaddr);
324 spin_unlock_irqrestore(&ghes_ioremap_lock_irq, flags);
329 static int ghes_read_estatus(struct ghes *ghes, int silent)
331 struct acpi_hest_generic *g = ghes->generic;
336 rc = apei_read(&buf_paddr, &g->error_status_address);
338 if (!silent && printk_ratelimit())
339 pr_warning(FW_WARN GHES_PFX
340 "Failed to read error status block address for hardware error source: %d.\n",
341 g->header.source_id);
347 ghes_copy_tofrom_phys(ghes->estatus, buf_paddr,
348 sizeof(*ghes->estatus), 1);
349 if (!ghes->estatus->block_status)
352 ghes->buffer_paddr = buf_paddr;
353 ghes->flags |= GHES_TO_CLEAR;
356 len = cper_estatus_len(ghes->estatus);
357 if (len < sizeof(*ghes->estatus))
359 if (len > ghes->generic->error_block_length)
361 if (cper_estatus_check_header(ghes->estatus))
363 ghes_copy_tofrom_phys(ghes->estatus + 1,
364 buf_paddr + sizeof(*ghes->estatus),
365 len - sizeof(*ghes->estatus), 1);
366 if (cper_estatus_check(ghes->estatus))
371 if (rc && !silent && printk_ratelimit())
372 pr_warning(FW_WARN GHES_PFX
373 "Failed to read error status block!\n");
377 static void ghes_clear_estatus(struct ghes *ghes)
379 ghes->estatus->block_status = 0;
380 if (!(ghes->flags & GHES_TO_CLEAR))
382 ghes_copy_tofrom_phys(ghes->estatus, ghes->buffer_paddr,
383 sizeof(ghes->estatus->block_status), 0);
384 ghes->flags &= ~GHES_TO_CLEAR;
387 static void ghes_handle_memory_failure(struct acpi_hest_generic_data *gdata, int sev)
389 #ifdef CONFIG_ACPI_APEI_MEMORY_FAILURE
392 int sec_sev = ghes_severity(gdata->error_severity);
393 struct cper_sec_mem_err *mem_err;
394 mem_err = (struct cper_sec_mem_err *)(gdata + 1);
396 if (!(mem_err->validation_bits & CPER_MEM_VALID_PA))
399 pfn = mem_err->physical_addr >> PAGE_SHIFT;
400 if (!pfn_valid(pfn)) {
401 pr_warn_ratelimited(FW_WARN GHES_PFX
402 "Invalid address in generic error data: %#llx\n",
403 mem_err->physical_addr);
407 /* iff following two events can be handled properly by now */
408 if (sec_sev == GHES_SEV_CORRECTED &&
409 (gdata->flags & CPER_SEC_ERROR_THRESHOLD_EXCEEDED))
410 flags = MF_SOFT_OFFLINE;
411 if (sev == GHES_SEV_RECOVERABLE && sec_sev == GHES_SEV_RECOVERABLE)
415 memory_failure_queue(pfn, 0, flags);
419 static void ghes_do_proc(struct ghes *ghes,
420 const struct acpi_hest_generic_status *estatus)
423 struct acpi_hest_generic_data *gdata;
425 sev = ghes_severity(estatus->error_severity);
426 apei_estatus_for_each_section(estatus, gdata) {
427 sec_sev = ghes_severity(gdata->error_severity);
428 if (!uuid_le_cmp(*(uuid_le *)gdata->section_type,
429 CPER_SEC_PLATFORM_MEM)) {
430 struct cper_sec_mem_err *mem_err;
431 mem_err = (struct cper_sec_mem_err *)(gdata+1);
432 ghes_edac_report_mem_error(ghes, sev, mem_err);
434 arch_apei_report_mem_error(sev, mem_err);
435 ghes_handle_memory_failure(gdata, sev);
437 #ifdef CONFIG_ACPI_APEI_PCIEAER
438 else if (!uuid_le_cmp(*(uuid_le *)gdata->section_type,
440 struct cper_sec_pcie *pcie_err;
441 pcie_err = (struct cper_sec_pcie *)(gdata+1);
442 if (sev == GHES_SEV_RECOVERABLE &&
443 sec_sev == GHES_SEV_RECOVERABLE &&
444 pcie_err->validation_bits & CPER_PCIE_VALID_DEVICE_ID &&
445 pcie_err->validation_bits & CPER_PCIE_VALID_AER_INFO) {
449 devfn = PCI_DEVFN(pcie_err->device_id.device,
450 pcie_err->device_id.function);
451 aer_severity = cper_severity_to_aer(sev);
454 * If firmware reset the component to contain
455 * the error, we must reinitialize it before
456 * use, so treat it as a fatal AER error.
458 if (gdata->flags & CPER_SEC_RESET)
459 aer_severity = AER_FATAL;
461 aer_recover_queue(pcie_err->device_id.segment,
462 pcie_err->device_id.bus,
464 (struct aer_capability_regs *)
473 static void __ghes_print_estatus(const char *pfx,
474 const struct acpi_hest_generic *generic,
475 const struct acpi_hest_generic_status *estatus)
477 static atomic_t seqno;
478 unsigned int curr_seqno;
482 if (ghes_severity(estatus->error_severity) <=
488 curr_seqno = atomic_inc_return(&seqno);
489 snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}" HW_ERR, pfx, curr_seqno);
490 printk("%s""Hardware error from APEI Generic Hardware Error Source: %d\n",
491 pfx_seq, generic->header.source_id);
492 cper_estatus_print(pfx_seq, estatus);
495 static int ghes_print_estatus(const char *pfx,
496 const struct acpi_hest_generic *generic,
497 const struct acpi_hest_generic_status *estatus)
499 /* Not more than 2 messages every 5 seconds */
500 static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2);
501 static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2);
502 struct ratelimit_state *ratelimit;
504 if (ghes_severity(estatus->error_severity) <= GHES_SEV_CORRECTED)
505 ratelimit = &ratelimit_corrected;
507 ratelimit = &ratelimit_uncorrected;
508 if (__ratelimit(ratelimit)) {
509 __ghes_print_estatus(pfx, generic, estatus);
516 * GHES error status reporting throttle, to report more kinds of
517 * errors, instead of just most frequently occurred errors.
519 static int ghes_estatus_cached(struct acpi_hest_generic_status *estatus)
523 unsigned long long now;
524 struct ghes_estatus_cache *cache;
525 struct acpi_hest_generic_status *cache_estatus;
527 len = cper_estatus_len(estatus);
529 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
530 cache = rcu_dereference(ghes_estatus_caches[i]);
533 if (len != cache->estatus_len)
535 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
536 if (memcmp(estatus, cache_estatus, len))
538 atomic_inc(&cache->count);
540 if (now - cache->time_in < GHES_ESTATUS_IN_CACHE_MAX_NSEC)
548 static struct ghes_estatus_cache *ghes_estatus_cache_alloc(
549 struct acpi_hest_generic *generic,
550 struct acpi_hest_generic_status *estatus)
554 struct ghes_estatus_cache *cache;
555 struct acpi_hest_generic_status *cache_estatus;
557 alloced = atomic_add_return(1, &ghes_estatus_cache_alloced);
558 if (alloced > GHES_ESTATUS_CACHE_ALLOCED_MAX) {
559 atomic_dec(&ghes_estatus_cache_alloced);
562 len = cper_estatus_len(estatus);
563 cache_len = GHES_ESTATUS_CACHE_LEN(len);
564 cache = (void *)gen_pool_alloc(ghes_estatus_pool, cache_len);
566 atomic_dec(&ghes_estatus_cache_alloced);
569 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
570 memcpy(cache_estatus, estatus, len);
571 cache->estatus_len = len;
572 atomic_set(&cache->count, 0);
573 cache->generic = generic;
574 cache->time_in = sched_clock();
578 static void ghes_estatus_cache_free(struct ghes_estatus_cache *cache)
582 len = cper_estatus_len(GHES_ESTATUS_FROM_CACHE(cache));
583 len = GHES_ESTATUS_CACHE_LEN(len);
584 gen_pool_free(ghes_estatus_pool, (unsigned long)cache, len);
585 atomic_dec(&ghes_estatus_cache_alloced);
588 static void ghes_estatus_cache_rcu_free(struct rcu_head *head)
590 struct ghes_estatus_cache *cache;
592 cache = container_of(head, struct ghes_estatus_cache, rcu);
593 ghes_estatus_cache_free(cache);
596 static void ghes_estatus_cache_add(
597 struct acpi_hest_generic *generic,
598 struct acpi_hest_generic_status *estatus)
600 int i, slot = -1, count;
601 unsigned long long now, duration, period, max_period = 0;
602 struct ghes_estatus_cache *cache, *slot_cache = NULL, *new_cache;
604 new_cache = ghes_estatus_cache_alloc(generic, estatus);
605 if (new_cache == NULL)
609 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
610 cache = rcu_dereference(ghes_estatus_caches[i]);
616 duration = now - cache->time_in;
617 if (duration >= GHES_ESTATUS_IN_CACHE_MAX_NSEC) {
622 count = atomic_read(&cache->count);
624 do_div(period, (count + 1));
625 if (period > max_period) {
631 /* new_cache must be put into array after its contents are written */
633 if (slot != -1 && cmpxchg(ghes_estatus_caches + slot,
634 slot_cache, new_cache) == slot_cache) {
636 call_rcu(&slot_cache->rcu, ghes_estatus_cache_rcu_free);
638 ghes_estatus_cache_free(new_cache);
642 static int ghes_proc(struct ghes *ghes)
646 rc = ghes_read_estatus(ghes, 0);
649 if (!ghes_estatus_cached(ghes->estatus)) {
650 if (ghes_print_estatus(NULL, ghes->generic, ghes->estatus))
651 ghes_estatus_cache_add(ghes->generic, ghes->estatus);
653 ghes_do_proc(ghes, ghes->estatus);
655 ghes_clear_estatus(ghes);
659 static void ghes_add_timer(struct ghes *ghes)
661 struct acpi_hest_generic *g = ghes->generic;
662 unsigned long expire;
664 if (!g->notify.poll_interval) {
665 pr_warning(FW_WARN GHES_PFX "Poll interval is 0 for generic hardware error source: %d, disabled.\n",
666 g->header.source_id);
669 expire = jiffies + msecs_to_jiffies(g->notify.poll_interval);
670 ghes->timer.expires = round_jiffies_relative(expire);
671 add_timer(&ghes->timer);
674 static void ghes_poll_func(unsigned long data)
676 struct ghes *ghes = (void *)data;
679 if (!(ghes->flags & GHES_EXITING))
680 ghes_add_timer(ghes);
683 static irqreturn_t ghes_irq_func(int irq, void *data)
685 struct ghes *ghes = data;
688 rc = ghes_proc(ghes);
695 static int ghes_notify_sci(struct notifier_block *this,
696 unsigned long event, void *data)
699 int ret = NOTIFY_DONE;
702 list_for_each_entry_rcu(ghes, &ghes_sci, list) {
703 if (!ghes_proc(ghes))
711 static struct notifier_block ghes_notifier_sci = {
712 .notifier_call = ghes_notify_sci,
715 #ifdef CONFIG_HAVE_ACPI_APEI_NMI
717 * printk is not safe in NMI context. So in NMI handler, we allocate
718 * required memory from lock-less memory allocator
719 * (ghes_estatus_pool), save estatus into it, put them into lock-less
720 * list (ghes_estatus_llist), then delay printk into IRQ context via
721 * irq_work (ghes_proc_irq_work). ghes_estatus_size_request record
722 * required pool size by all NMI error source.
724 static struct llist_head ghes_estatus_llist;
725 static struct irq_work ghes_proc_irq_work;
728 * NMI may be triggered on any CPU, so ghes_in_nmi is used for
729 * having only one concurrent reader.
731 static atomic_t ghes_in_nmi = ATOMIC_INIT(0);
733 static LIST_HEAD(ghes_nmi);
735 static int ghes_panic_timeout __read_mostly = 30;
737 static void ghes_proc_in_irq(struct irq_work *irq_work)
739 struct llist_node *llnode, *next;
740 struct ghes_estatus_node *estatus_node;
741 struct acpi_hest_generic *generic;
742 struct acpi_hest_generic_status *estatus;
745 llnode = llist_del_all(&ghes_estatus_llist);
747 * Because the time order of estatus in list is reversed,
748 * revert it back to proper order.
750 llnode = llist_reverse_order(llnode);
753 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
755 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
756 len = cper_estatus_len(estatus);
757 node_len = GHES_ESTATUS_NODE_LEN(len);
758 ghes_do_proc(estatus_node->ghes, estatus);
759 if (!ghes_estatus_cached(estatus)) {
760 generic = estatus_node->generic;
761 if (ghes_print_estatus(NULL, generic, estatus))
762 ghes_estatus_cache_add(generic, estatus);
764 gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node,
770 static void ghes_print_queued_estatus(void)
772 struct llist_node *llnode;
773 struct ghes_estatus_node *estatus_node;
774 struct acpi_hest_generic *generic;
775 struct acpi_hest_generic_status *estatus;
778 llnode = llist_del_all(&ghes_estatus_llist);
780 * Because the time order of estatus in list is reversed,
781 * revert it back to proper order.
783 llnode = llist_reverse_order(llnode);
785 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
787 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
788 len = cper_estatus_len(estatus);
789 node_len = GHES_ESTATUS_NODE_LEN(len);
790 generic = estatus_node->generic;
791 ghes_print_estatus(NULL, generic, estatus);
792 llnode = llnode->next;
796 /* Save estatus for further processing in IRQ context */
797 static void __process_error(struct ghes *ghes)
799 #ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
801 struct ghes_estatus_node *estatus_node;
802 struct acpi_hest_generic_status *estatus;
804 if (ghes_estatus_cached(ghes->estatus))
807 len = cper_estatus_len(ghes->estatus);
808 node_len = GHES_ESTATUS_NODE_LEN(len);
810 estatus_node = (void *)gen_pool_alloc(ghes_estatus_pool, node_len);
814 estatus_node->ghes = ghes;
815 estatus_node->generic = ghes->generic;
816 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
817 memcpy(estatus, ghes->estatus, len);
818 llist_add(&estatus_node->llnode, &ghes_estatus_llist);
822 static void __ghes_panic(struct ghes *ghes)
825 ghes_print_queued_estatus();
826 __ghes_print_estatus(KERN_EMERG, ghes->generic, ghes->estatus);
828 /* reboot to log the error! */
829 if (panic_timeout == 0)
830 panic_timeout = ghes_panic_timeout;
831 panic("Fatal hardware error!");
834 static int ghes_notify_nmi(unsigned int cmd, struct pt_regs *regs)
837 int sev, ret = NMI_DONE;
839 if (!atomic_add_unless(&ghes_in_nmi, 1, 1))
842 list_for_each_entry_rcu(ghes, &ghes_nmi, list) {
843 if (ghes_read_estatus(ghes, 1)) {
844 ghes_clear_estatus(ghes);
848 sev = ghes_severity(ghes->estatus->error_severity);
849 if (sev >= GHES_SEV_PANIC)
852 if (!(ghes->flags & GHES_TO_CLEAR))
855 __process_error(ghes);
856 ghes_clear_estatus(ghes);
861 #ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
862 irq_work_queue(&ghes_proc_irq_work);
864 atomic_dec(&ghes_in_nmi);
868 static unsigned long ghes_esource_prealloc_size(
869 const struct acpi_hest_generic *generic)
871 unsigned long block_length, prealloc_records, prealloc_size;
873 block_length = min_t(unsigned long, generic->error_block_length,
874 GHES_ESTATUS_MAX_SIZE);
875 prealloc_records = max_t(unsigned long,
876 generic->records_to_preallocate, 1);
877 prealloc_size = min_t(unsigned long, block_length * prealloc_records,
878 GHES_ESOURCE_PREALLOC_MAX_SIZE);
880 return prealloc_size;
883 static void ghes_estatus_pool_shrink(unsigned long len)
885 ghes_estatus_pool_size_request -= PAGE_ALIGN(len);
888 static void ghes_nmi_add(struct ghes *ghes)
892 len = ghes_esource_prealloc_size(ghes->generic);
893 ghes_estatus_pool_expand(len);
894 mutex_lock(&ghes_list_mutex);
895 if (list_empty(&ghes_nmi))
896 register_nmi_handler(NMI_LOCAL, ghes_notify_nmi, 0, "ghes");
897 list_add_rcu(&ghes->list, &ghes_nmi);
898 mutex_unlock(&ghes_list_mutex);
901 static void ghes_nmi_remove(struct ghes *ghes)
905 mutex_lock(&ghes_list_mutex);
906 list_del_rcu(&ghes->list);
907 if (list_empty(&ghes_nmi))
908 unregister_nmi_handler(NMI_LOCAL, "ghes");
909 mutex_unlock(&ghes_list_mutex);
911 * To synchronize with NMI handler, ghes can only be
912 * freed after NMI handler finishes.
915 len = ghes_esource_prealloc_size(ghes->generic);
916 ghes_estatus_pool_shrink(len);
919 static void ghes_nmi_init_cxt(void)
921 init_irq_work(&ghes_proc_irq_work, ghes_proc_in_irq);
923 #else /* CONFIG_HAVE_ACPI_APEI_NMI */
924 static inline void ghes_nmi_add(struct ghes *ghes)
926 pr_err(GHES_PFX "ID: %d, trying to add NMI notification which is not supported!\n",
927 ghes->generic->header.source_id);
931 static inline void ghes_nmi_remove(struct ghes *ghes)
933 pr_err(GHES_PFX "ID: %d, trying to remove NMI notification which is not supported!\n",
934 ghes->generic->header.source_id);
938 static inline void ghes_nmi_init_cxt(void)
941 #endif /* CONFIG_HAVE_ACPI_APEI_NMI */
943 static int ghes_probe(struct platform_device *ghes_dev)
945 struct acpi_hest_generic *generic;
946 struct ghes *ghes = NULL;
950 generic = *(struct acpi_hest_generic **)ghes_dev->dev.platform_data;
951 if (!generic->enabled)
954 switch (generic->notify.type) {
955 case ACPI_HEST_NOTIFY_POLLED:
956 case ACPI_HEST_NOTIFY_EXTERNAL:
957 case ACPI_HEST_NOTIFY_SCI:
959 case ACPI_HEST_NOTIFY_NMI:
960 if (!IS_ENABLED(CONFIG_HAVE_ACPI_APEI_NMI)) {
961 pr_warn(GHES_PFX "Generic hardware error source: %d notified via NMI interrupt is not supported!\n",
962 generic->header.source_id);
966 case ACPI_HEST_NOTIFY_LOCAL:
967 pr_warning(GHES_PFX "Generic hardware error source: %d notified via local interrupt is not supported!\n",
968 generic->header.source_id);
971 pr_warning(FW_WARN GHES_PFX "Unknown notification type: %u for generic hardware error source: %d\n",
972 generic->notify.type, generic->header.source_id);
977 if (generic->error_block_length <
978 sizeof(struct acpi_hest_generic_status)) {
979 pr_warning(FW_BUG GHES_PFX "Invalid error block length: %u for generic hardware error source: %d\n",
980 generic->error_block_length,
981 generic->header.source_id);
984 ghes = ghes_new(generic);
991 rc = ghes_edac_register(ghes, &ghes_dev->dev);
995 switch (generic->notify.type) {
996 case ACPI_HEST_NOTIFY_POLLED:
997 ghes->timer.function = ghes_poll_func;
998 ghes->timer.data = (unsigned long)ghes;
999 init_timer_deferrable(&ghes->timer);
1000 ghes_add_timer(ghes);
1002 case ACPI_HEST_NOTIFY_EXTERNAL:
1003 /* External interrupt vector is GSI */
1004 rc = acpi_gsi_to_irq(generic->notify.vector, &ghes->irq);
1006 pr_err(GHES_PFX "Failed to map GSI to IRQ for generic hardware error source: %d\n",
1007 generic->header.source_id);
1008 goto err_edac_unreg;
1010 rc = request_irq(ghes->irq, ghes_irq_func, 0, "GHES IRQ", ghes);
1012 pr_err(GHES_PFX "Failed to register IRQ for generic hardware error source: %d\n",
1013 generic->header.source_id);
1014 goto err_edac_unreg;
1017 case ACPI_HEST_NOTIFY_SCI:
1018 mutex_lock(&ghes_list_mutex);
1019 if (list_empty(&ghes_sci))
1020 register_acpi_hed_notifier(&ghes_notifier_sci);
1021 list_add_rcu(&ghes->list, &ghes_sci);
1022 mutex_unlock(&ghes_list_mutex);
1024 case ACPI_HEST_NOTIFY_NMI:
1030 platform_set_drvdata(ghes_dev, ghes);
1034 ghes_edac_unregister(ghes);
1043 static int ghes_remove(struct platform_device *ghes_dev)
1046 struct acpi_hest_generic *generic;
1048 ghes = platform_get_drvdata(ghes_dev);
1049 generic = ghes->generic;
1051 ghes->flags |= GHES_EXITING;
1052 switch (generic->notify.type) {
1053 case ACPI_HEST_NOTIFY_POLLED:
1054 del_timer_sync(&ghes->timer);
1056 case ACPI_HEST_NOTIFY_EXTERNAL:
1057 free_irq(ghes->irq, ghes);
1059 case ACPI_HEST_NOTIFY_SCI:
1060 mutex_lock(&ghes_list_mutex);
1061 list_del_rcu(&ghes->list);
1062 if (list_empty(&ghes_sci))
1063 unregister_acpi_hed_notifier(&ghes_notifier_sci);
1064 mutex_unlock(&ghes_list_mutex);
1066 case ACPI_HEST_NOTIFY_NMI:
1067 ghes_nmi_remove(ghes);
1076 ghes_edac_unregister(ghes);
1080 platform_set_drvdata(ghes_dev, NULL);
1085 static struct platform_driver ghes_platform_driver = {
1089 .probe = ghes_probe,
1090 .remove = ghes_remove,
1093 static int __init ghes_init(void)
1101 pr_info(GHES_PFX "HEST is not enabled!\n");
1106 pr_info(GHES_PFX "GHES is not enabled!\n");
1110 ghes_nmi_init_cxt();
1112 rc = ghes_ioremap_init();
1116 rc = ghes_estatus_pool_init();
1118 goto err_ioremap_exit;
1120 rc = ghes_estatus_pool_expand(GHES_ESTATUS_CACHE_AVG_SIZE *
1121 GHES_ESTATUS_CACHE_ALLOCED_MAX);
1125 rc = platform_driver_register(&ghes_platform_driver);
1129 rc = apei_osc_setup();
1130 if (rc == 0 && osc_sb_apei_support_acked)
1131 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit and WHEA _OSC.\n");
1132 else if (rc == 0 && !osc_sb_apei_support_acked)
1133 pr_info(GHES_PFX "APEI firmware first mode is enabled by WHEA _OSC.\n");
1134 else if (rc && osc_sb_apei_support_acked)
1135 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit.\n");
1137 pr_info(GHES_PFX "Failed to enable APEI firmware first mode.\n");
1141 ghes_estatus_pool_exit();
1143 ghes_ioremap_exit();
1148 static void __exit ghes_exit(void)
1150 platform_driver_unregister(&ghes_platform_driver);
1151 ghes_estatus_pool_exit();
1152 ghes_ioremap_exit();
1155 module_init(ghes_init);
1156 module_exit(ghes_exit);
1158 MODULE_AUTHOR("Huang Ying");
1159 MODULE_DESCRIPTION("APEI Generic Hardware Error Source support");
1160 MODULE_LICENSE("GPL");
1161 MODULE_ALIAS("platform:GHES");