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.
27 * You should have received a copy of the GNU General Public License
28 * along with this program; if not, write to the Free Software
29 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
32 #include <linux/kernel.h>
33 #include <linux/module.h>
34 #include <linux/init.h>
35 #include <linux/acpi.h>
37 #include <linux/interrupt.h>
38 #include <linux/timer.h>
39 #include <linux/cper.h>
40 #include <linux/kdebug.h>
41 #include <linux/platform_device.h>
42 #include <linux/mutex.h>
43 #include <linux/ratelimit.h>
44 #include <linux/vmalloc.h>
45 #include <linux/irq_work.h>
46 #include <linux/llist.h>
47 #include <linux/genalloc.h>
48 #include <acpi/apei.h>
49 #include <acpi/atomicio.h>
52 #include <asm/tlbflush.h>
54 #include "apei-internal.h"
56 #define GHES_PFX "GHES: "
58 #define GHES_ESTATUS_MAX_SIZE 65536
59 #define GHES_ESOURCE_PREALLOC_MAX_SIZE 65536
61 #define GHES_ESTATUS_POOL_MIN_ALLOC_ORDER 3
63 /* This is just an estimation for memory pool allocation */
64 #define GHES_ESTATUS_CACHE_AVG_SIZE 512
66 #define GHES_ESTATUS_CACHES_SIZE 4
68 #define GHES_ESTATUS_IN_CACHE_MAX_NSEC (10 * NSEC_PER_SEC)
69 /* Prevent too many caches are allocated because of RCU */
70 #define GHES_ESTATUS_CACHE_ALLOCED_MAX (GHES_ESTATUS_CACHES_SIZE * 3 / 2)
72 #define GHES_ESTATUS_CACHE_LEN(estatus_len) \
73 (sizeof(struct ghes_estatus_cache) + (estatus_len))
74 #define GHES_ESTATUS_FROM_CACHE(estatus_cache) \
75 ((struct acpi_hest_generic_status *) \
76 ((struct ghes_estatus_cache *)(estatus_cache) + 1))
78 #define GHES_ESTATUS_NODE_LEN(estatus_len) \
79 (sizeof(struct ghes_estatus_node) + (estatus_len))
80 #define GHES_ESTATUS_FROM_NODE(estatus_node) \
81 ((struct acpi_hest_generic_status *) \
82 ((struct ghes_estatus_node *)(estatus_node) + 1))
85 * One struct ghes is created for each generic hardware error source.
86 * It provides the context for APEI hardware error timer/IRQ/SCI/NMI
89 * estatus: memory buffer for error status block, allocated during
92 #define GHES_TO_CLEAR 0x0001
93 #define GHES_EXITING 0x0002
96 struct acpi_hest_generic *generic;
97 struct acpi_hest_generic_status *estatus;
101 struct list_head list;
102 struct timer_list timer;
107 struct ghes_estatus_node {
108 struct llist_node llnode;
109 struct acpi_hest_generic *generic;
112 struct ghes_estatus_cache {
115 struct acpi_hest_generic *generic;
116 unsigned long long time_in;
121 module_param_named(disable, ghes_disable, bool, 0);
123 static int ghes_panic_timeout __read_mostly = 30;
126 * All error sources notified with SCI shares one notifier function,
127 * so they need to be linked and checked one by one. This is applied
130 * RCU is used for these lists, so ghes_list_mutex is only used for
131 * list changing, not for traversing.
133 static LIST_HEAD(ghes_sci);
134 static LIST_HEAD(ghes_nmi);
135 static DEFINE_MUTEX(ghes_list_mutex);
138 * NMI may be triggered on any CPU, so ghes_nmi_lock is used for
141 static DEFINE_RAW_SPINLOCK(ghes_nmi_lock);
144 * Because the memory area used to transfer hardware error information
145 * from BIOS to Linux can be determined only in NMI, IRQ or timer
146 * handler, but general ioremap can not be used in atomic context, so
147 * a special version of atomic ioremap is implemented for that.
151 * Two virtual pages are used, one for NMI context, the other for
152 * IRQ/PROCESS context
154 #define GHES_IOREMAP_PAGES 2
155 #define GHES_IOREMAP_NMI_PAGE(base) (base)
156 #define GHES_IOREMAP_IRQ_PAGE(base) ((base) + PAGE_SIZE)
158 /* virtual memory area for atomic ioremap */
159 static struct vm_struct *ghes_ioremap_area;
161 * These 2 spinlock is used to prevent atomic ioremap virtual memory
162 * area from being mapped simultaneously.
164 static DEFINE_RAW_SPINLOCK(ghes_ioremap_lock_nmi);
165 static DEFINE_SPINLOCK(ghes_ioremap_lock_irq);
168 * printk is not safe in NMI context. So in NMI handler, we allocate
169 * required memory from lock-less memory allocator
170 * (ghes_estatus_pool), save estatus into it, put them into lock-less
171 * list (ghes_estatus_llist), then delay printk into IRQ context via
172 * irq_work (ghes_proc_irq_work). ghes_estatus_size_request record
173 * required pool size by all NMI error source.
175 static struct gen_pool *ghes_estatus_pool;
176 static unsigned long ghes_estatus_pool_size_request;
177 static struct llist_head ghes_estatus_llist;
178 static struct irq_work ghes_proc_irq_work;
180 struct ghes_estatus_cache *ghes_estatus_caches[GHES_ESTATUS_CACHES_SIZE];
181 static atomic_t ghes_estatus_cache_alloced;
183 static int ghes_ioremap_init(void)
185 ghes_ioremap_area = __get_vm_area(PAGE_SIZE * GHES_IOREMAP_PAGES,
186 VM_IOREMAP, VMALLOC_START, VMALLOC_END);
187 if (!ghes_ioremap_area) {
188 pr_err(GHES_PFX "Failed to allocate virtual memory area for atomic ioremap.\n");
195 static void ghes_ioremap_exit(void)
197 free_vm_area(ghes_ioremap_area);
200 static void __iomem *ghes_ioremap_pfn_nmi(u64 pfn)
204 vaddr = (unsigned long)GHES_IOREMAP_NMI_PAGE(ghes_ioremap_area->addr);
205 ioremap_page_range(vaddr, vaddr + PAGE_SIZE,
206 pfn << PAGE_SHIFT, PAGE_KERNEL);
208 return (void __iomem *)vaddr;
211 static void __iomem *ghes_ioremap_pfn_irq(u64 pfn)
215 vaddr = (unsigned long)GHES_IOREMAP_IRQ_PAGE(ghes_ioremap_area->addr);
216 ioremap_page_range(vaddr, vaddr + PAGE_SIZE,
217 pfn << PAGE_SHIFT, PAGE_KERNEL);
219 return (void __iomem *)vaddr;
222 static void ghes_iounmap_nmi(void __iomem *vaddr_ptr)
224 unsigned long vaddr = (unsigned long __force)vaddr_ptr;
225 void *base = ghes_ioremap_area->addr;
227 BUG_ON(vaddr != (unsigned long)GHES_IOREMAP_NMI_PAGE(base));
228 unmap_kernel_range_noflush(vaddr, PAGE_SIZE);
229 __flush_tlb_one(vaddr);
232 static void ghes_iounmap_irq(void __iomem *vaddr_ptr)
234 unsigned long vaddr = (unsigned long __force)vaddr_ptr;
235 void *base = ghes_ioremap_area->addr;
237 BUG_ON(vaddr != (unsigned long)GHES_IOREMAP_IRQ_PAGE(base));
238 unmap_kernel_range_noflush(vaddr, PAGE_SIZE);
239 __flush_tlb_one(vaddr);
242 static int ghes_estatus_pool_init(void)
244 ghes_estatus_pool = gen_pool_create(GHES_ESTATUS_POOL_MIN_ALLOC_ORDER, -1);
245 if (!ghes_estatus_pool)
250 static void ghes_estatus_pool_free_chunk_page(struct gen_pool *pool,
251 struct gen_pool_chunk *chunk,
254 free_page(chunk->start_addr);
257 static void ghes_estatus_pool_exit(void)
259 gen_pool_for_each_chunk(ghes_estatus_pool,
260 ghes_estatus_pool_free_chunk_page, NULL);
261 gen_pool_destroy(ghes_estatus_pool);
264 static int ghes_estatus_pool_expand(unsigned long len)
266 unsigned long i, pages, size, addr;
269 ghes_estatus_pool_size_request += PAGE_ALIGN(len);
270 size = gen_pool_size(ghes_estatus_pool);
271 if (size >= ghes_estatus_pool_size_request)
273 pages = (ghes_estatus_pool_size_request - size) / PAGE_SIZE;
274 for (i = 0; i < pages; i++) {
275 addr = __get_free_page(GFP_KERNEL);
278 ret = gen_pool_add(ghes_estatus_pool, addr, PAGE_SIZE, -1);
286 static void ghes_estatus_pool_shrink(unsigned long len)
288 ghes_estatus_pool_size_request -= PAGE_ALIGN(len);
291 static struct ghes *ghes_new(struct acpi_hest_generic *generic)
294 unsigned int error_block_length;
297 ghes = kzalloc(sizeof(*ghes), GFP_KERNEL);
299 return ERR_PTR(-ENOMEM);
300 ghes->generic = generic;
301 rc = acpi_pre_map_gar(&generic->error_status_address);
304 error_block_length = generic->error_block_length;
305 if (error_block_length > GHES_ESTATUS_MAX_SIZE) {
306 pr_warning(FW_WARN GHES_PFX
307 "Error status block length is too long: %u for "
308 "generic hardware error source: %d.\n",
309 error_block_length, generic->header.source_id);
310 error_block_length = GHES_ESTATUS_MAX_SIZE;
312 ghes->estatus = kmalloc(error_block_length, GFP_KERNEL);
313 if (!ghes->estatus) {
321 acpi_post_unmap_gar(&generic->error_status_address);
327 static void ghes_fini(struct ghes *ghes)
329 kfree(ghes->estatus);
330 acpi_post_unmap_gar(&ghes->generic->error_status_address);
335 GHES_SEV_CORRECTED = 0x1,
336 GHES_SEV_RECOVERABLE = 0x2,
337 GHES_SEV_PANIC = 0x3,
340 static inline int ghes_severity(int severity)
343 case CPER_SEV_INFORMATIONAL:
345 case CPER_SEV_CORRECTED:
346 return GHES_SEV_CORRECTED;
347 case CPER_SEV_RECOVERABLE:
348 return GHES_SEV_RECOVERABLE;
350 return GHES_SEV_PANIC;
352 /* Unknown, go panic */
353 return GHES_SEV_PANIC;
357 static void ghes_copy_tofrom_phys(void *buffer, u64 paddr, u32 len,
361 unsigned long flags = 0;
362 int in_nmi = in_nmi();
367 offset = paddr - (paddr & PAGE_MASK);
369 raw_spin_lock(&ghes_ioremap_lock_nmi);
370 vaddr = ghes_ioremap_pfn_nmi(paddr >> PAGE_SHIFT);
372 spin_lock_irqsave(&ghes_ioremap_lock_irq, flags);
373 vaddr = ghes_ioremap_pfn_irq(paddr >> PAGE_SHIFT);
375 trunk = PAGE_SIZE - offset;
376 trunk = min(trunk, len);
378 memcpy_fromio(buffer, vaddr + offset, trunk);
380 memcpy_toio(vaddr + offset, buffer, trunk);
385 ghes_iounmap_nmi(vaddr);
386 raw_spin_unlock(&ghes_ioremap_lock_nmi);
388 ghes_iounmap_irq(vaddr);
389 spin_unlock_irqrestore(&ghes_ioremap_lock_irq, flags);
394 static int ghes_read_estatus(struct ghes *ghes, int silent)
396 struct acpi_hest_generic *g = ghes->generic;
401 rc = acpi_atomic_read(&buf_paddr, &g->error_status_address);
403 if (!silent && printk_ratelimit())
404 pr_warning(FW_WARN GHES_PFX
405 "Failed to read error status block address for hardware error source: %d.\n",
406 g->header.source_id);
412 ghes_copy_tofrom_phys(ghes->estatus, buf_paddr,
413 sizeof(*ghes->estatus), 1);
414 if (!ghes->estatus->block_status)
417 ghes->buffer_paddr = buf_paddr;
418 ghes->flags |= GHES_TO_CLEAR;
421 len = apei_estatus_len(ghes->estatus);
422 if (len < sizeof(*ghes->estatus))
424 if (len > ghes->generic->error_block_length)
426 if (apei_estatus_check_header(ghes->estatus))
428 ghes_copy_tofrom_phys(ghes->estatus + 1,
429 buf_paddr + sizeof(*ghes->estatus),
430 len - sizeof(*ghes->estatus), 1);
431 if (apei_estatus_check(ghes->estatus))
436 if (rc && !silent && printk_ratelimit())
437 pr_warning(FW_WARN GHES_PFX
438 "Failed to read error status block!\n");
442 static void ghes_clear_estatus(struct ghes *ghes)
444 ghes->estatus->block_status = 0;
445 if (!(ghes->flags & GHES_TO_CLEAR))
447 ghes_copy_tofrom_phys(ghes->estatus, ghes->buffer_paddr,
448 sizeof(ghes->estatus->block_status), 0);
449 ghes->flags &= ~GHES_TO_CLEAR;
452 static void ghes_do_proc(const struct acpi_hest_generic_status *estatus)
455 struct acpi_hest_generic_data *gdata;
457 sev = ghes_severity(estatus->error_severity);
458 apei_estatus_for_each_section(estatus, gdata) {
459 sec_sev = ghes_severity(gdata->error_severity);
460 if (!uuid_le_cmp(*(uuid_le *)gdata->section_type,
461 CPER_SEC_PLATFORM_MEM)) {
462 struct cper_sec_mem_err *mem_err;
463 mem_err = (struct cper_sec_mem_err *)(gdata+1);
464 #ifdef CONFIG_X86_MCE
465 apei_mce_report_mem_error(sev == GHES_SEV_CORRECTED,
468 #ifdef CONFIG_ACPI_APEI_MEMORY_FAILURE
469 if (sev == GHES_SEV_RECOVERABLE &&
470 sec_sev == GHES_SEV_RECOVERABLE &&
471 mem_err->validation_bits & CPER_MEM_VALID_PHYSICAL_ADDRESS) {
473 pfn = mem_err->physical_addr >> PAGE_SHIFT;
474 memory_failure_queue(pfn, 0, 0);
481 static void __ghes_print_estatus(const char *pfx,
482 const struct acpi_hest_generic *generic,
483 const struct acpi_hest_generic_status *estatus)
486 if (ghes_severity(estatus->error_severity) <=
488 pfx = KERN_WARNING HW_ERR;
490 pfx = KERN_ERR HW_ERR;
492 printk("%s""Hardware error from APEI Generic Hardware Error Source: %d\n",
493 pfx, generic->header.source_id);
494 apei_estatus_print(pfx, estatus);
497 static int ghes_print_estatus(const char *pfx,
498 const struct acpi_hest_generic *generic,
499 const struct acpi_hest_generic_status *estatus)
501 /* Not more than 2 messages every 5 seconds */
502 static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2);
503 static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2);
504 struct ratelimit_state *ratelimit;
506 if (ghes_severity(estatus->error_severity) <= GHES_SEV_CORRECTED)
507 ratelimit = &ratelimit_corrected;
509 ratelimit = &ratelimit_uncorrected;
510 if (__ratelimit(ratelimit)) {
511 __ghes_print_estatus(pfx, generic, estatus);
518 * GHES error status reporting throttle, to report more kinds of
519 * errors, instead of just most frequently occurred errors.
521 static int ghes_estatus_cached(struct acpi_hest_generic_status *estatus)
525 unsigned long long now;
526 struct ghes_estatus_cache *cache;
527 struct acpi_hest_generic_status *cache_estatus;
529 len = apei_estatus_len(estatus);
531 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
532 cache = rcu_dereference(ghes_estatus_caches[i]);
535 if (len != cache->estatus_len)
537 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
538 if (memcmp(estatus, cache_estatus, len))
540 atomic_inc(&cache->count);
542 if (now - cache->time_in < GHES_ESTATUS_IN_CACHE_MAX_NSEC)
550 static struct ghes_estatus_cache *ghes_estatus_cache_alloc(
551 struct acpi_hest_generic *generic,
552 struct acpi_hest_generic_status *estatus)
556 struct ghes_estatus_cache *cache;
557 struct acpi_hest_generic_status *cache_estatus;
559 alloced = atomic_add_return(1, &ghes_estatus_cache_alloced);
560 if (alloced > GHES_ESTATUS_CACHE_ALLOCED_MAX) {
561 atomic_dec(&ghes_estatus_cache_alloced);
564 len = apei_estatus_len(estatus);
565 cache_len = GHES_ESTATUS_CACHE_LEN(len);
566 cache = (void *)gen_pool_alloc(ghes_estatus_pool, cache_len);
568 atomic_dec(&ghes_estatus_cache_alloced);
571 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
572 memcpy(cache_estatus, estatus, len);
573 cache->estatus_len = len;
574 atomic_set(&cache->count, 0);
575 cache->generic = generic;
576 cache->time_in = sched_clock();
580 static void ghes_estatus_cache_free(struct ghes_estatus_cache *cache)
584 len = apei_estatus_len(GHES_ESTATUS_FROM_CACHE(cache));
585 len = GHES_ESTATUS_CACHE_LEN(len);
586 gen_pool_free(ghes_estatus_pool, (unsigned long)cache, len);
587 atomic_dec(&ghes_estatus_cache_alloced);
590 static void ghes_estatus_cache_rcu_free(struct rcu_head *head)
592 struct ghes_estatus_cache *cache;
594 cache = container_of(head, struct ghes_estatus_cache, rcu);
595 ghes_estatus_cache_free(cache);
598 static void ghes_estatus_cache_add(
599 struct acpi_hest_generic *generic,
600 struct acpi_hest_generic_status *estatus)
602 int i, slot = -1, count;
603 unsigned long long now, duration, period, max_period = 0;
604 struct ghes_estatus_cache *cache, *slot_cache = NULL, *new_cache;
606 new_cache = ghes_estatus_cache_alloc(generic, estatus);
607 if (new_cache == NULL)
611 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
612 cache = rcu_dereference(ghes_estatus_caches[i]);
618 duration = now - cache->time_in;
619 if (duration >= GHES_ESTATUS_IN_CACHE_MAX_NSEC) {
624 count = atomic_read(&cache->count);
625 period = duration / (count + 1);
626 if (period > max_period) {
632 /* new_cache must be put into array after its contents are written */
634 if (slot != -1 && cmpxchg(ghes_estatus_caches + slot,
635 slot_cache, new_cache) == slot_cache) {
637 call_rcu(&slot_cache->rcu, ghes_estatus_cache_rcu_free);
639 ghes_estatus_cache_free(new_cache);
643 static int ghes_proc(struct ghes *ghes)
647 rc = ghes_read_estatus(ghes, 0);
650 if (!ghes_estatus_cached(ghes->estatus)) {
651 if (ghes_print_estatus(NULL, ghes->generic, ghes->estatus))
652 ghes_estatus_cache_add(ghes->generic, ghes->estatus);
654 ghes_do_proc(ghes->estatus);
656 ghes_clear_estatus(ghes);
660 static void ghes_add_timer(struct ghes *ghes)
662 struct acpi_hest_generic *g = ghes->generic;
663 unsigned long expire;
665 if (!g->notify.poll_interval) {
666 pr_warning(FW_WARN GHES_PFX "Poll interval is 0 for generic hardware error source: %d, disabled.\n",
667 g->header.source_id);
670 expire = jiffies + msecs_to_jiffies(g->notify.poll_interval);
671 ghes->timer.expires = round_jiffies_relative(expire);
672 add_timer(&ghes->timer);
675 static void ghes_poll_func(unsigned long data)
677 struct ghes *ghes = (void *)data;
680 if (!(ghes->flags & GHES_EXITING))
681 ghes_add_timer(ghes);
684 static irqreturn_t ghes_irq_func(int irq, void *data)
686 struct ghes *ghes = data;
689 rc = ghes_proc(ghes);
696 static int ghes_notify_sci(struct notifier_block *this,
697 unsigned long event, void *data)
700 int ret = NOTIFY_DONE;
703 list_for_each_entry_rcu(ghes, &ghes_sci, list) {
704 if (!ghes_proc(ghes))
712 static void ghes_proc_in_irq(struct irq_work *irq_work)
714 struct llist_node *llnode, *next, *tail = NULL;
715 struct ghes_estatus_node *estatus_node;
716 struct acpi_hest_generic *generic;
717 struct acpi_hest_generic_status *estatus;
721 * Because the time order of estatus in list is reversed,
722 * revert it back to proper order.
724 llnode = llist_del_all(&ghes_estatus_llist);
734 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
736 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
737 len = apei_estatus_len(estatus);
738 node_len = GHES_ESTATUS_NODE_LEN(len);
739 ghes_do_proc(estatus);
740 if (!ghes_estatus_cached(estatus)) {
741 generic = estatus_node->generic;
742 if (ghes_print_estatus(NULL, generic, estatus))
743 ghes_estatus_cache_add(generic, estatus);
745 gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node,
751 static int ghes_notify_nmi(struct notifier_block *this,
752 unsigned long cmd, void *data)
754 struct ghes *ghes, *ghes_global = NULL;
755 int sev, sev_global = -1;
756 int ret = NOTIFY_DONE;
761 raw_spin_lock(&ghes_nmi_lock);
762 list_for_each_entry_rcu(ghes, &ghes_nmi, list) {
763 if (ghes_read_estatus(ghes, 1)) {
764 ghes_clear_estatus(ghes);
767 sev = ghes_severity(ghes->estatus->error_severity);
768 if (sev > sev_global) {
775 if (ret == NOTIFY_DONE)
778 if (sev_global >= GHES_SEV_PANIC) {
780 __ghes_print_estatus(KERN_EMERG HW_ERR, ghes_global->generic,
781 ghes_global->estatus);
782 /* reboot to log the error! */
783 if (panic_timeout == 0)
784 panic_timeout = ghes_panic_timeout;
785 panic("Fatal hardware error!");
788 list_for_each_entry_rcu(ghes, &ghes_nmi, list) {
789 #ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
791 struct ghes_estatus_node *estatus_node;
792 struct acpi_hest_generic_status *estatus;
794 if (!(ghes->flags & GHES_TO_CLEAR))
796 #ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
797 if (ghes_estatus_cached(ghes->estatus))
799 /* Save estatus for further processing in IRQ context */
800 len = apei_estatus_len(ghes->estatus);
801 node_len = GHES_ESTATUS_NODE_LEN(len);
802 estatus_node = (void *)gen_pool_alloc(ghes_estatus_pool,
805 estatus_node->generic = ghes->generic;
806 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
807 memcpy(estatus, ghes->estatus, len);
808 llist_add(&estatus_node->llnode, &ghes_estatus_llist);
812 ghes_clear_estatus(ghes);
814 #ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
815 irq_work_queue(&ghes_proc_irq_work);
819 raw_spin_unlock(&ghes_nmi_lock);
823 static struct notifier_block ghes_notifier_sci = {
824 .notifier_call = ghes_notify_sci,
827 static struct notifier_block ghes_notifier_nmi = {
828 .notifier_call = ghes_notify_nmi,
831 static unsigned long ghes_esource_prealloc_size(
832 const struct acpi_hest_generic *generic)
834 unsigned long block_length, prealloc_records, prealloc_size;
836 block_length = min_t(unsigned long, generic->error_block_length,
837 GHES_ESTATUS_MAX_SIZE);
838 prealloc_records = max_t(unsigned long,
839 generic->records_to_preallocate, 1);
840 prealloc_size = min_t(unsigned long, block_length * prealloc_records,
841 GHES_ESOURCE_PREALLOC_MAX_SIZE);
843 return prealloc_size;
846 static int __devinit ghes_probe(struct platform_device *ghes_dev)
848 struct acpi_hest_generic *generic;
849 struct ghes *ghes = NULL;
853 generic = *(struct acpi_hest_generic **)ghes_dev->dev.platform_data;
854 if (!generic->enabled)
857 switch (generic->notify.type) {
858 case ACPI_HEST_NOTIFY_POLLED:
859 case ACPI_HEST_NOTIFY_EXTERNAL:
860 case ACPI_HEST_NOTIFY_SCI:
861 case ACPI_HEST_NOTIFY_NMI:
863 case ACPI_HEST_NOTIFY_LOCAL:
864 pr_warning(GHES_PFX "Generic hardware error source: %d notified via local interrupt is not supported!\n",
865 generic->header.source_id);
868 pr_warning(FW_WARN GHES_PFX "Unknown notification type: %u for generic hardware error source: %d\n",
869 generic->notify.type, generic->header.source_id);
874 if (generic->error_block_length <
875 sizeof(struct acpi_hest_generic_status)) {
876 pr_warning(FW_BUG GHES_PFX "Invalid error block length: %u for generic hardware error source: %d\n",
877 generic->error_block_length,
878 generic->header.source_id);
881 ghes = ghes_new(generic);
887 switch (generic->notify.type) {
888 case ACPI_HEST_NOTIFY_POLLED:
889 ghes->timer.function = ghes_poll_func;
890 ghes->timer.data = (unsigned long)ghes;
891 init_timer_deferrable(&ghes->timer);
892 ghes_add_timer(ghes);
894 case ACPI_HEST_NOTIFY_EXTERNAL:
895 /* External interrupt vector is GSI */
896 if (acpi_gsi_to_irq(generic->notify.vector, &ghes->irq)) {
897 pr_err(GHES_PFX "Failed to map GSI to IRQ for generic hardware error source: %d\n",
898 generic->header.source_id);
901 if (request_irq(ghes->irq, ghes_irq_func,
902 0, "GHES IRQ", ghes)) {
903 pr_err(GHES_PFX "Failed to register IRQ for generic hardware error source: %d\n",
904 generic->header.source_id);
908 case ACPI_HEST_NOTIFY_SCI:
909 mutex_lock(&ghes_list_mutex);
910 if (list_empty(&ghes_sci))
911 register_acpi_hed_notifier(&ghes_notifier_sci);
912 list_add_rcu(&ghes->list, &ghes_sci);
913 mutex_unlock(&ghes_list_mutex);
915 case ACPI_HEST_NOTIFY_NMI:
916 len = ghes_esource_prealloc_size(generic);
917 ghes_estatus_pool_expand(len);
918 mutex_lock(&ghes_list_mutex);
919 if (list_empty(&ghes_nmi))
920 register_die_notifier(&ghes_notifier_nmi);
921 list_add_rcu(&ghes->list, &ghes_nmi);
922 mutex_unlock(&ghes_list_mutex);
927 platform_set_drvdata(ghes_dev, ghes);
938 static int __devexit ghes_remove(struct platform_device *ghes_dev)
941 struct acpi_hest_generic *generic;
944 ghes = platform_get_drvdata(ghes_dev);
945 generic = ghes->generic;
947 ghes->flags |= GHES_EXITING;
948 switch (generic->notify.type) {
949 case ACPI_HEST_NOTIFY_POLLED:
950 del_timer_sync(&ghes->timer);
952 case ACPI_HEST_NOTIFY_EXTERNAL:
953 free_irq(ghes->irq, ghes);
955 case ACPI_HEST_NOTIFY_SCI:
956 mutex_lock(&ghes_list_mutex);
957 list_del_rcu(&ghes->list);
958 if (list_empty(&ghes_sci))
959 unregister_acpi_hed_notifier(&ghes_notifier_sci);
960 mutex_unlock(&ghes_list_mutex);
962 case ACPI_HEST_NOTIFY_NMI:
963 mutex_lock(&ghes_list_mutex);
964 list_del_rcu(&ghes->list);
965 if (list_empty(&ghes_nmi))
966 unregister_die_notifier(&ghes_notifier_nmi);
967 mutex_unlock(&ghes_list_mutex);
969 * To synchronize with NMI handler, ghes can only be
970 * freed after NMI handler finishes.
973 len = ghes_esource_prealloc_size(generic);
974 ghes_estatus_pool_shrink(len);
984 platform_set_drvdata(ghes_dev, NULL);
989 static struct platform_driver ghes_platform_driver = {
992 .owner = THIS_MODULE,
995 .remove = ghes_remove,
998 static int __init ghes_init(void)
1006 pr_info(GHES_PFX "HEST is not enabled!\n");
1011 pr_info(GHES_PFX "GHES is not enabled!\n");
1015 init_irq_work(&ghes_proc_irq_work, ghes_proc_in_irq);
1017 rc = ghes_ioremap_init();
1021 rc = ghes_estatus_pool_init();
1023 goto err_ioremap_exit;
1025 rc = ghes_estatus_pool_expand(GHES_ESTATUS_CACHE_AVG_SIZE *
1026 GHES_ESTATUS_CACHE_ALLOCED_MAX);
1030 rc = platform_driver_register(&ghes_platform_driver);
1034 rc = apei_osc_setup();
1035 if (rc == 0 && osc_sb_apei_support_acked)
1036 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit and WHEA _OSC.\n");
1037 else if (rc == 0 && !osc_sb_apei_support_acked)
1038 pr_info(GHES_PFX "APEI firmware first mode is enabled by WHEA _OSC.\n");
1039 else if (rc && osc_sb_apei_support_acked)
1040 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit.\n");
1042 pr_info(GHES_PFX "Failed to enable APEI firmware first mode.\n");
1046 ghes_estatus_pool_exit();
1048 ghes_ioremap_exit();
1053 static void __exit ghes_exit(void)
1055 platform_driver_unregister(&ghes_platform_driver);
1056 ghes_estatus_pool_exit();
1057 ghes_ioremap_exit();
1060 module_init(ghes_init);
1061 module_exit(ghes_exit);
1063 MODULE_AUTHOR("Huang Ying");
1064 MODULE_DESCRIPTION("APEI Generic Hardware Error Source support");
1065 MODULE_LICENSE("GPL");
1066 MODULE_ALIAS("platform:GHES");