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>
48 #include <linux/uuid.h>
49 #include <linux/ras.h>
51 #include <acpi/actbl1.h>
52 #include <acpi/ghes.h>
53 #include <acpi/apei.h>
54 #include <asm/fixmap.h>
55 #include <asm/tlbflush.h>
56 #include <ras/ras_event.h>
58 #include "apei-internal.h"
60 #define GHES_PFX "GHES: "
62 #define GHES_ESTATUS_MAX_SIZE 65536
63 #define GHES_ESOURCE_PREALLOC_MAX_SIZE 65536
65 #define GHES_ESTATUS_POOL_MIN_ALLOC_ORDER 3
67 /* This is just an estimation for memory pool allocation */
68 #define GHES_ESTATUS_CACHE_AVG_SIZE 512
70 #define GHES_ESTATUS_CACHES_SIZE 4
72 #define GHES_ESTATUS_IN_CACHE_MAX_NSEC 10000000000ULL
73 /* Prevent too many caches are allocated because of RCU */
74 #define GHES_ESTATUS_CACHE_ALLOCED_MAX (GHES_ESTATUS_CACHES_SIZE * 3 / 2)
76 #define GHES_ESTATUS_CACHE_LEN(estatus_len) \
77 (sizeof(struct ghes_estatus_cache) + (estatus_len))
78 #define GHES_ESTATUS_FROM_CACHE(estatus_cache) \
79 ((struct acpi_hest_generic_status *) \
80 ((struct ghes_estatus_cache *)(estatus_cache) + 1))
82 #define GHES_ESTATUS_NODE_LEN(estatus_len) \
83 (sizeof(struct ghes_estatus_node) + (estatus_len))
84 #define GHES_ESTATUS_FROM_NODE(estatus_node) \
85 ((struct acpi_hest_generic_status *) \
86 ((struct ghes_estatus_node *)(estatus_node) + 1))
88 static inline bool is_hest_type_generic_v2(struct ghes
*ghes
)
90 return ghes
->generic
->header
.type
== ACPI_HEST_TYPE_GENERIC_ERROR_V2
;
94 * This driver isn't really modular, however for the time being,
95 * continuing to use module_param is the easiest way to remain
96 * compatible with existing boot arg use cases.
99 module_param_named(disable
, ghes_disable
, bool, 0);
102 * All error sources notified with HED (Hardware Error Device) share a
103 * single notifier callback, so they need to be linked and checked one
104 * by one. This holds true for NMI too.
106 * RCU is used for these lists, so ghes_list_mutex is only used for
107 * list changing, not for traversing.
109 static LIST_HEAD(ghes_hed
);
110 static DEFINE_MUTEX(ghes_list_mutex
);
113 * Because the memory area used to transfer hardware error information
114 * from BIOS to Linux can be determined only in NMI, IRQ or timer
115 * handler, but general ioremap can not be used in atomic context, so
116 * the fixmap is used instead.
118 * These 2 spinlocks are used to prevent the fixmap entries from being used
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_panic_timeout __read_mostly
= 30;
132 static void __iomem
*ghes_ioremap_pfn_nmi(u64 pfn
)
137 paddr
= pfn
<< PAGE_SHIFT
;
138 prot
= arch_apei_get_mem_attribute(paddr
);
139 __set_fixmap(FIX_APEI_GHES_NMI
, paddr
, prot
);
141 return (void __iomem
*) fix_to_virt(FIX_APEI_GHES_NMI
);
144 static void __iomem
*ghes_ioremap_pfn_irq(u64 pfn
)
149 paddr
= pfn
<< PAGE_SHIFT
;
150 prot
= arch_apei_get_mem_attribute(paddr
);
151 __set_fixmap(FIX_APEI_GHES_IRQ
, paddr
, prot
);
153 return (void __iomem
*) fix_to_virt(FIX_APEI_GHES_IRQ
);
156 static void ghes_iounmap_nmi(void)
158 clear_fixmap(FIX_APEI_GHES_NMI
);
161 static void ghes_iounmap_irq(void)
163 clear_fixmap(FIX_APEI_GHES_IRQ
);
166 static int ghes_estatus_pool_init(void)
168 ghes_estatus_pool
= gen_pool_create(GHES_ESTATUS_POOL_MIN_ALLOC_ORDER
, -1);
169 if (!ghes_estatus_pool
)
174 static void ghes_estatus_pool_free_chunk_page(struct gen_pool
*pool
,
175 struct gen_pool_chunk
*chunk
,
178 free_page(chunk
->start_addr
);
181 static void ghes_estatus_pool_exit(void)
183 gen_pool_for_each_chunk(ghes_estatus_pool
,
184 ghes_estatus_pool_free_chunk_page
, NULL
);
185 gen_pool_destroy(ghes_estatus_pool
);
188 static int ghes_estatus_pool_expand(unsigned long len
)
190 unsigned long i
, pages
, size
, addr
;
193 ghes_estatus_pool_size_request
+= PAGE_ALIGN(len
);
194 size
= gen_pool_size(ghes_estatus_pool
);
195 if (size
>= ghes_estatus_pool_size_request
)
197 pages
= (ghes_estatus_pool_size_request
- size
) / PAGE_SIZE
;
198 for (i
= 0; i
< pages
; i
++) {
199 addr
= __get_free_page(GFP_KERNEL
);
202 ret
= gen_pool_add(ghes_estatus_pool
, addr
, PAGE_SIZE
, -1);
210 static int map_gen_v2(struct ghes
*ghes
)
212 return apei_map_generic_address(&ghes
->generic_v2
->read_ack_register
);
215 static void unmap_gen_v2(struct ghes
*ghes
)
217 apei_unmap_generic_address(&ghes
->generic_v2
->read_ack_register
);
220 static struct ghes
*ghes_new(struct acpi_hest_generic
*generic
)
223 unsigned int error_block_length
;
226 ghes
= kzalloc(sizeof(*ghes
), GFP_KERNEL
);
228 return ERR_PTR(-ENOMEM
);
230 ghes
->generic
= generic
;
231 if (is_hest_type_generic_v2(ghes
)) {
232 rc
= map_gen_v2(ghes
);
237 rc
= apei_map_generic_address(&generic
->error_status_address
);
239 goto err_unmap_read_ack_addr
;
240 error_block_length
= generic
->error_block_length
;
241 if (error_block_length
> GHES_ESTATUS_MAX_SIZE
) {
242 pr_warning(FW_WARN GHES_PFX
243 "Error status block length is too long: %u for "
244 "generic hardware error source: %d.\n",
245 error_block_length
, generic
->header
.source_id
);
246 error_block_length
= GHES_ESTATUS_MAX_SIZE
;
248 ghes
->estatus
= kmalloc(error_block_length
, GFP_KERNEL
);
249 if (!ghes
->estatus
) {
251 goto err_unmap_status_addr
;
256 err_unmap_status_addr
:
257 apei_unmap_generic_address(&generic
->error_status_address
);
258 err_unmap_read_ack_addr
:
259 if (is_hest_type_generic_v2(ghes
))
266 static void ghes_fini(struct ghes
*ghes
)
268 kfree(ghes
->estatus
);
269 apei_unmap_generic_address(&ghes
->generic
->error_status_address
);
270 if (is_hest_type_generic_v2(ghes
))
274 static inline int ghes_severity(int severity
)
277 case CPER_SEV_INFORMATIONAL
:
279 case CPER_SEV_CORRECTED
:
280 return GHES_SEV_CORRECTED
;
281 case CPER_SEV_RECOVERABLE
:
282 return GHES_SEV_RECOVERABLE
;
284 return GHES_SEV_PANIC
;
286 /* Unknown, go panic */
287 return GHES_SEV_PANIC
;
291 static void ghes_copy_tofrom_phys(void *buffer
, u64 paddr
, u32 len
,
295 unsigned long flags
= 0;
296 int in_nmi
= in_nmi();
301 offset
= paddr
- (paddr
& PAGE_MASK
);
303 raw_spin_lock(&ghes_ioremap_lock_nmi
);
304 vaddr
= ghes_ioremap_pfn_nmi(paddr
>> PAGE_SHIFT
);
306 spin_lock_irqsave(&ghes_ioremap_lock_irq
, flags
);
307 vaddr
= ghes_ioremap_pfn_irq(paddr
>> PAGE_SHIFT
);
309 trunk
= PAGE_SIZE
- offset
;
310 trunk
= min(trunk
, len
);
312 memcpy_fromio(buffer
, vaddr
+ offset
, trunk
);
314 memcpy_toio(vaddr
+ offset
, buffer
, trunk
);
320 raw_spin_unlock(&ghes_ioremap_lock_nmi
);
323 spin_unlock_irqrestore(&ghes_ioremap_lock_irq
, flags
);
328 static int ghes_read_estatus(struct ghes
*ghes
, int silent
)
330 struct acpi_hest_generic
*g
= ghes
->generic
;
335 rc
= apei_read(&buf_paddr
, &g
->error_status_address
);
337 if (!silent
&& printk_ratelimit())
338 pr_warning(FW_WARN GHES_PFX
339 "Failed to read error status block address for hardware error source: %d.\n",
340 g
->header
.source_id
);
346 ghes_copy_tofrom_phys(ghes
->estatus
, buf_paddr
,
347 sizeof(*ghes
->estatus
), 1);
348 if (!ghes
->estatus
->block_status
)
351 ghes
->buffer_paddr
= buf_paddr
;
352 ghes
->flags
|= GHES_TO_CLEAR
;
355 len
= cper_estatus_len(ghes
->estatus
);
356 if (len
< sizeof(*ghes
->estatus
))
358 if (len
> ghes
->generic
->error_block_length
)
360 if (cper_estatus_check_header(ghes
->estatus
))
362 ghes_copy_tofrom_phys(ghes
->estatus
+ 1,
363 buf_paddr
+ sizeof(*ghes
->estatus
),
364 len
- sizeof(*ghes
->estatus
), 1);
365 if (cper_estatus_check(ghes
->estatus
))
370 if (rc
&& !silent
&& printk_ratelimit())
371 pr_warning(FW_WARN GHES_PFX
372 "Failed to read error status block!\n");
376 static void ghes_clear_estatus(struct ghes
*ghes
)
378 ghes
->estatus
->block_status
= 0;
379 if (!(ghes
->flags
& GHES_TO_CLEAR
))
381 ghes_copy_tofrom_phys(ghes
->estatus
, ghes
->buffer_paddr
,
382 sizeof(ghes
->estatus
->block_status
), 0);
383 ghes
->flags
&= ~GHES_TO_CLEAR
;
386 static void ghes_handle_memory_failure(struct acpi_hest_generic_data
*gdata
, int sev
)
388 #ifdef CONFIG_ACPI_APEI_MEMORY_FAILURE
391 int sec_sev
= ghes_severity(gdata
->error_severity
);
392 struct cper_sec_mem_err
*mem_err
= acpi_hest_get_payload(gdata
);
394 if (!(mem_err
->validation_bits
& CPER_MEM_VALID_PA
))
397 pfn
= mem_err
->physical_addr
>> PAGE_SHIFT
;
398 if (!pfn_valid(pfn
)) {
399 pr_warn_ratelimited(FW_WARN GHES_PFX
400 "Invalid address in generic error data: %#llx\n",
401 mem_err
->physical_addr
);
405 /* iff following two events can be handled properly by now */
406 if (sec_sev
== GHES_SEV_CORRECTED
&&
407 (gdata
->flags
& CPER_SEC_ERROR_THRESHOLD_EXCEEDED
))
408 flags
= MF_SOFT_OFFLINE
;
409 if (sev
== GHES_SEV_RECOVERABLE
&& sec_sev
== GHES_SEV_RECOVERABLE
)
413 memory_failure_queue(pfn
, 0, flags
);
417 static void ghes_do_proc(struct ghes
*ghes
,
418 const struct acpi_hest_generic_status
*estatus
)
421 struct acpi_hest_generic_data
*gdata
;
423 guid_t
*fru_id
= &NULL_UUID_LE
;
426 sev
= ghes_severity(estatus
->error_severity
);
427 apei_estatus_for_each_section(estatus
, gdata
) {
428 sec_type
= (guid_t
*)gdata
->section_type
;
429 sec_sev
= ghes_severity(gdata
->error_severity
);
430 if (gdata
->validation_bits
& CPER_SEC_VALID_FRU_ID
)
431 fru_id
= (guid_t
*)gdata
->fru_id
;
433 if (gdata
->validation_bits
& CPER_SEC_VALID_FRU_TEXT
)
434 fru_text
= gdata
->fru_text
;
436 if (guid_equal(sec_type
, &CPER_SEC_PLATFORM_MEM
)) {
437 struct cper_sec_mem_err
*mem_err
= acpi_hest_get_payload(gdata
);
439 ghes_edac_report_mem_error(ghes
, sev
, mem_err
);
441 arch_apei_report_mem_error(sev
, mem_err
);
442 ghes_handle_memory_failure(gdata
, sev
);
444 #ifdef CONFIG_ACPI_APEI_PCIEAER
445 else if (guid_equal(sec_type
, &CPER_SEC_PCIE
)) {
446 struct cper_sec_pcie
*pcie_err
= acpi_hest_get_payload(gdata
);
448 if (sev
== GHES_SEV_RECOVERABLE
&&
449 sec_sev
== GHES_SEV_RECOVERABLE
&&
450 pcie_err
->validation_bits
& CPER_PCIE_VALID_DEVICE_ID
&&
451 pcie_err
->validation_bits
& CPER_PCIE_VALID_AER_INFO
) {
455 devfn
= PCI_DEVFN(pcie_err
->device_id
.device
,
456 pcie_err
->device_id
.function
);
457 aer_severity
= cper_severity_to_aer(gdata
->error_severity
);
460 * If firmware reset the component to contain
461 * the error, we must reinitialize it before
462 * use, so treat it as a fatal AER error.
464 if (gdata
->flags
& CPER_SEC_RESET
)
465 aer_severity
= AER_FATAL
;
467 aer_recover_queue(pcie_err
->device_id
.segment
,
468 pcie_err
->device_id
.bus
,
470 (struct aer_capability_regs
*)
476 else if (guid_equal(sec_type
, &CPER_SEC_PROC_ARM
)) {
477 struct cper_sec_proc_arm
*err
= acpi_hest_get_payload(gdata
);
479 log_arm_hw_error(err
);
481 void *err
= acpi_hest_get_payload(gdata
);
483 log_non_standard_event(sec_type
, fru_id
, fru_text
,
485 gdata
->error_data_length
);
490 static void __ghes_print_estatus(const char *pfx
,
491 const struct acpi_hest_generic
*generic
,
492 const struct acpi_hest_generic_status
*estatus
)
494 static atomic_t seqno
;
495 unsigned int curr_seqno
;
499 if (ghes_severity(estatus
->error_severity
) <=
505 curr_seqno
= atomic_inc_return(&seqno
);
506 snprintf(pfx_seq
, sizeof(pfx_seq
), "%s{%u}" HW_ERR
, pfx
, curr_seqno
);
507 printk("%s""Hardware error from APEI Generic Hardware Error Source: %d\n",
508 pfx_seq
, generic
->header
.source_id
);
509 cper_estatus_print(pfx_seq
, estatus
);
512 static int ghes_print_estatus(const char *pfx
,
513 const struct acpi_hest_generic
*generic
,
514 const struct acpi_hest_generic_status
*estatus
)
516 /* Not more than 2 messages every 5 seconds */
517 static DEFINE_RATELIMIT_STATE(ratelimit_corrected
, 5*HZ
, 2);
518 static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected
, 5*HZ
, 2);
519 struct ratelimit_state
*ratelimit
;
521 if (ghes_severity(estatus
->error_severity
) <= GHES_SEV_CORRECTED
)
522 ratelimit
= &ratelimit_corrected
;
524 ratelimit
= &ratelimit_uncorrected
;
525 if (__ratelimit(ratelimit
)) {
526 __ghes_print_estatus(pfx
, generic
, estatus
);
533 * GHES error status reporting throttle, to report more kinds of
534 * errors, instead of just most frequently occurred errors.
536 static int ghes_estatus_cached(struct acpi_hest_generic_status
*estatus
)
540 unsigned long long now
;
541 struct ghes_estatus_cache
*cache
;
542 struct acpi_hest_generic_status
*cache_estatus
;
544 len
= cper_estatus_len(estatus
);
546 for (i
= 0; i
< GHES_ESTATUS_CACHES_SIZE
; i
++) {
547 cache
= rcu_dereference(ghes_estatus_caches
[i
]);
550 if (len
!= cache
->estatus_len
)
552 cache_estatus
= GHES_ESTATUS_FROM_CACHE(cache
);
553 if (memcmp(estatus
, cache_estatus
, len
))
555 atomic_inc(&cache
->count
);
557 if (now
- cache
->time_in
< GHES_ESTATUS_IN_CACHE_MAX_NSEC
)
565 static struct ghes_estatus_cache
*ghes_estatus_cache_alloc(
566 struct acpi_hest_generic
*generic
,
567 struct acpi_hest_generic_status
*estatus
)
571 struct ghes_estatus_cache
*cache
;
572 struct acpi_hest_generic_status
*cache_estatus
;
574 alloced
= atomic_add_return(1, &ghes_estatus_cache_alloced
);
575 if (alloced
> GHES_ESTATUS_CACHE_ALLOCED_MAX
) {
576 atomic_dec(&ghes_estatus_cache_alloced
);
579 len
= cper_estatus_len(estatus
);
580 cache_len
= GHES_ESTATUS_CACHE_LEN(len
);
581 cache
= (void *)gen_pool_alloc(ghes_estatus_pool
, cache_len
);
583 atomic_dec(&ghes_estatus_cache_alloced
);
586 cache_estatus
= GHES_ESTATUS_FROM_CACHE(cache
);
587 memcpy(cache_estatus
, estatus
, len
);
588 cache
->estatus_len
= len
;
589 atomic_set(&cache
->count
, 0);
590 cache
->generic
= generic
;
591 cache
->time_in
= sched_clock();
595 static void ghes_estatus_cache_free(struct ghes_estatus_cache
*cache
)
599 len
= cper_estatus_len(GHES_ESTATUS_FROM_CACHE(cache
));
600 len
= GHES_ESTATUS_CACHE_LEN(len
);
601 gen_pool_free(ghes_estatus_pool
, (unsigned long)cache
, len
);
602 atomic_dec(&ghes_estatus_cache_alloced
);
605 static void ghes_estatus_cache_rcu_free(struct rcu_head
*head
)
607 struct ghes_estatus_cache
*cache
;
609 cache
= container_of(head
, struct ghes_estatus_cache
, rcu
);
610 ghes_estatus_cache_free(cache
);
613 static void ghes_estatus_cache_add(
614 struct acpi_hest_generic
*generic
,
615 struct acpi_hest_generic_status
*estatus
)
617 int i
, slot
= -1, count
;
618 unsigned long long now
, duration
, period
, max_period
= 0;
619 struct ghes_estatus_cache
*cache
, *slot_cache
= NULL
, *new_cache
;
621 new_cache
= ghes_estatus_cache_alloc(generic
, estatus
);
622 if (new_cache
== NULL
)
626 for (i
= 0; i
< GHES_ESTATUS_CACHES_SIZE
; i
++) {
627 cache
= rcu_dereference(ghes_estatus_caches
[i
]);
633 duration
= now
- cache
->time_in
;
634 if (duration
>= GHES_ESTATUS_IN_CACHE_MAX_NSEC
) {
639 count
= atomic_read(&cache
->count
);
641 do_div(period
, (count
+ 1));
642 if (period
> max_period
) {
648 /* new_cache must be put into array after its contents are written */
650 if (slot
!= -1 && cmpxchg(ghes_estatus_caches
+ slot
,
651 slot_cache
, new_cache
) == slot_cache
) {
653 call_rcu(&slot_cache
->rcu
, ghes_estatus_cache_rcu_free
);
655 ghes_estatus_cache_free(new_cache
);
659 static int ghes_ack_error(struct acpi_hest_generic_v2
*gv2
)
664 rc
= apei_read(&val
, &gv2
->read_ack_register
);
668 val
&= gv2
->read_ack_preserve
<< gv2
->read_ack_register
.bit_offset
;
669 val
|= gv2
->read_ack_write
<< gv2
->read_ack_register
.bit_offset
;
671 return apei_write(val
, &gv2
->read_ack_register
);
674 static void __ghes_panic(struct ghes
*ghes
)
676 __ghes_print_estatus(KERN_EMERG
, ghes
->generic
, ghes
->estatus
);
678 /* reboot to log the error! */
680 panic_timeout
= ghes_panic_timeout
;
681 panic("Fatal hardware error!");
684 static int ghes_proc(struct ghes
*ghes
)
688 rc
= ghes_read_estatus(ghes
, 0);
692 if (ghes_severity(ghes
->estatus
->error_severity
) >= GHES_SEV_PANIC
) {
696 if (!ghes_estatus_cached(ghes
->estatus
)) {
697 if (ghes_print_estatus(NULL
, ghes
->generic
, ghes
->estatus
))
698 ghes_estatus_cache_add(ghes
->generic
, ghes
->estatus
);
700 ghes_do_proc(ghes
, ghes
->estatus
);
703 ghes_clear_estatus(ghes
);
709 * GHESv2 type HEST entries introduce support for error acknowledgment,
710 * so only acknowledge the error if this support is present.
712 if (is_hest_type_generic_v2(ghes
))
713 return ghes_ack_error(ghes
->generic_v2
);
718 static void ghes_add_timer(struct ghes
*ghes
)
720 struct acpi_hest_generic
*g
= ghes
->generic
;
721 unsigned long expire
;
723 if (!g
->notify
.poll_interval
) {
724 pr_warning(FW_WARN GHES_PFX
"Poll interval is 0 for generic hardware error source: %d, disabled.\n",
725 g
->header
.source_id
);
728 expire
= jiffies
+ msecs_to_jiffies(g
->notify
.poll_interval
);
729 ghes
->timer
.expires
= round_jiffies_relative(expire
);
730 add_timer(&ghes
->timer
);
733 static void ghes_poll_func(struct timer_list
*t
)
735 struct ghes
*ghes
= from_timer(ghes
, t
, timer
);
738 if (!(ghes
->flags
& GHES_EXITING
))
739 ghes_add_timer(ghes
);
742 static irqreturn_t
ghes_irq_func(int irq
, void *data
)
744 struct ghes
*ghes
= data
;
747 rc
= ghes_proc(ghes
);
754 static int ghes_notify_hed(struct notifier_block
*this, unsigned long event
,
758 int ret
= NOTIFY_DONE
;
761 list_for_each_entry_rcu(ghes
, &ghes_hed
, list
) {
762 if (!ghes_proc(ghes
))
770 static struct notifier_block ghes_notifier_hed
= {
771 .notifier_call
= ghes_notify_hed
,
774 #ifdef CONFIG_ACPI_APEI_SEA
775 static LIST_HEAD(ghes_sea
);
778 * Return 0 only if one of the SEA error sources successfully reported an error
779 * record sent from the firmware.
781 int ghes_notify_sea(void)
787 list_for_each_entry_rcu(ghes
, &ghes_sea
, list
) {
788 if (!ghes_proc(ghes
))
795 static void ghes_sea_add(struct ghes
*ghes
)
797 mutex_lock(&ghes_list_mutex
);
798 list_add_rcu(&ghes
->list
, &ghes_sea
);
799 mutex_unlock(&ghes_list_mutex
);
802 static void ghes_sea_remove(struct ghes
*ghes
)
804 mutex_lock(&ghes_list_mutex
);
805 list_del_rcu(&ghes
->list
);
806 mutex_unlock(&ghes_list_mutex
);
809 #else /* CONFIG_ACPI_APEI_SEA */
810 static inline void ghes_sea_add(struct ghes
*ghes
) { }
811 static inline void ghes_sea_remove(struct ghes
*ghes
) { }
812 #endif /* CONFIG_ACPI_APEI_SEA */
814 #ifdef CONFIG_HAVE_ACPI_APEI_NMI
816 * printk is not safe in NMI context. So in NMI handler, we allocate
817 * required memory from lock-less memory allocator
818 * (ghes_estatus_pool), save estatus into it, put them into lock-less
819 * list (ghes_estatus_llist), then delay printk into IRQ context via
820 * irq_work (ghes_proc_irq_work). ghes_estatus_size_request record
821 * required pool size by all NMI error source.
823 static struct llist_head ghes_estatus_llist
;
824 static struct irq_work ghes_proc_irq_work
;
827 * NMI may be triggered on any CPU, so ghes_in_nmi is used for
828 * having only one concurrent reader.
830 static atomic_t ghes_in_nmi
= ATOMIC_INIT(0);
832 static LIST_HEAD(ghes_nmi
);
834 static void ghes_proc_in_irq(struct irq_work
*irq_work
)
836 struct llist_node
*llnode
, *next
;
837 struct ghes_estatus_node
*estatus_node
;
838 struct acpi_hest_generic
*generic
;
839 struct acpi_hest_generic_status
*estatus
;
842 llnode
= llist_del_all(&ghes_estatus_llist
);
844 * Because the time order of estatus in list is reversed,
845 * revert it back to proper order.
847 llnode
= llist_reverse_order(llnode
);
850 estatus_node
= llist_entry(llnode
, struct ghes_estatus_node
,
852 estatus
= GHES_ESTATUS_FROM_NODE(estatus_node
);
853 len
= cper_estatus_len(estatus
);
854 node_len
= GHES_ESTATUS_NODE_LEN(len
);
855 ghes_do_proc(estatus_node
->ghes
, estatus
);
856 if (!ghes_estatus_cached(estatus
)) {
857 generic
= estatus_node
->generic
;
858 if (ghes_print_estatus(NULL
, generic
, estatus
))
859 ghes_estatus_cache_add(generic
, estatus
);
861 gen_pool_free(ghes_estatus_pool
, (unsigned long)estatus_node
,
867 static void ghes_print_queued_estatus(void)
869 struct llist_node
*llnode
;
870 struct ghes_estatus_node
*estatus_node
;
871 struct acpi_hest_generic
*generic
;
872 struct acpi_hest_generic_status
*estatus
;
875 llnode
= llist_del_all(&ghes_estatus_llist
);
877 * Because the time order of estatus in list is reversed,
878 * revert it back to proper order.
880 llnode
= llist_reverse_order(llnode
);
882 estatus_node
= llist_entry(llnode
, struct ghes_estatus_node
,
884 estatus
= GHES_ESTATUS_FROM_NODE(estatus_node
);
885 len
= cper_estatus_len(estatus
);
886 node_len
= GHES_ESTATUS_NODE_LEN(len
);
887 generic
= estatus_node
->generic
;
888 ghes_print_estatus(NULL
, generic
, estatus
);
889 llnode
= llnode
->next
;
893 /* Save estatus for further processing in IRQ context */
894 static void __process_error(struct ghes
*ghes
)
896 #ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
898 struct ghes_estatus_node
*estatus_node
;
899 struct acpi_hest_generic_status
*estatus
;
901 if (ghes_estatus_cached(ghes
->estatus
))
904 len
= cper_estatus_len(ghes
->estatus
);
905 node_len
= GHES_ESTATUS_NODE_LEN(len
);
907 estatus_node
= (void *)gen_pool_alloc(ghes_estatus_pool
, node_len
);
911 estatus_node
->ghes
= ghes
;
912 estatus_node
->generic
= ghes
->generic
;
913 estatus
= GHES_ESTATUS_FROM_NODE(estatus_node
);
914 memcpy(estatus
, ghes
->estatus
, len
);
915 llist_add(&estatus_node
->llnode
, &ghes_estatus_llist
);
919 static int ghes_notify_nmi(unsigned int cmd
, struct pt_regs
*regs
)
922 int sev
, ret
= NMI_DONE
;
924 if (!atomic_add_unless(&ghes_in_nmi
, 1, 1))
927 list_for_each_entry_rcu(ghes
, &ghes_nmi
, list
) {
928 if (ghes_read_estatus(ghes
, 1)) {
929 ghes_clear_estatus(ghes
);
935 sev
= ghes_severity(ghes
->estatus
->error_severity
);
936 if (sev
>= GHES_SEV_PANIC
) {
938 ghes_print_queued_estatus();
942 if (!(ghes
->flags
& GHES_TO_CLEAR
))
945 __process_error(ghes
);
946 ghes_clear_estatus(ghes
);
949 #ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
950 if (ret
== NMI_HANDLED
)
951 irq_work_queue(&ghes_proc_irq_work
);
953 atomic_dec(&ghes_in_nmi
);
957 static unsigned long ghes_esource_prealloc_size(
958 const struct acpi_hest_generic
*generic
)
960 unsigned long block_length
, prealloc_records
, prealloc_size
;
962 block_length
= min_t(unsigned long, generic
->error_block_length
,
963 GHES_ESTATUS_MAX_SIZE
);
964 prealloc_records
= max_t(unsigned long,
965 generic
->records_to_preallocate
, 1);
966 prealloc_size
= min_t(unsigned long, block_length
* prealloc_records
,
967 GHES_ESOURCE_PREALLOC_MAX_SIZE
);
969 return prealloc_size
;
972 static void ghes_estatus_pool_shrink(unsigned long len
)
974 ghes_estatus_pool_size_request
-= PAGE_ALIGN(len
);
977 static void ghes_nmi_add(struct ghes
*ghes
)
981 len
= ghes_esource_prealloc_size(ghes
->generic
);
982 ghes_estatus_pool_expand(len
);
983 mutex_lock(&ghes_list_mutex
);
984 if (list_empty(&ghes_nmi
))
985 register_nmi_handler(NMI_LOCAL
, ghes_notify_nmi
, 0, "ghes");
986 list_add_rcu(&ghes
->list
, &ghes_nmi
);
987 mutex_unlock(&ghes_list_mutex
);
990 static void ghes_nmi_remove(struct ghes
*ghes
)
994 mutex_lock(&ghes_list_mutex
);
995 list_del_rcu(&ghes
->list
);
996 if (list_empty(&ghes_nmi
))
997 unregister_nmi_handler(NMI_LOCAL
, "ghes");
998 mutex_unlock(&ghes_list_mutex
);
1000 * To synchronize with NMI handler, ghes can only be
1001 * freed after NMI handler finishes.
1004 len
= ghes_esource_prealloc_size(ghes
->generic
);
1005 ghes_estatus_pool_shrink(len
);
1008 static void ghes_nmi_init_cxt(void)
1010 init_irq_work(&ghes_proc_irq_work
, ghes_proc_in_irq
);
1012 #else /* CONFIG_HAVE_ACPI_APEI_NMI */
1013 static inline void ghes_nmi_add(struct ghes
*ghes
) { }
1014 static inline void ghes_nmi_remove(struct ghes
*ghes
) { }
1015 static inline void ghes_nmi_init_cxt(void) { }
1016 #endif /* CONFIG_HAVE_ACPI_APEI_NMI */
1018 static int ghes_probe(struct platform_device
*ghes_dev
)
1020 struct acpi_hest_generic
*generic
;
1021 struct ghes
*ghes
= NULL
;
1025 generic
= *(struct acpi_hest_generic
**)ghes_dev
->dev
.platform_data
;
1026 if (!generic
->enabled
)
1029 switch (generic
->notify
.type
) {
1030 case ACPI_HEST_NOTIFY_POLLED
:
1031 case ACPI_HEST_NOTIFY_EXTERNAL
:
1032 case ACPI_HEST_NOTIFY_SCI
:
1033 case ACPI_HEST_NOTIFY_GSIV
:
1034 case ACPI_HEST_NOTIFY_GPIO
:
1037 case ACPI_HEST_NOTIFY_SEA
:
1038 if (!IS_ENABLED(CONFIG_ACPI_APEI_SEA
)) {
1039 pr_warn(GHES_PFX
"Generic hardware error source: %d notified via SEA is not supported\n",
1040 generic
->header
.source_id
);
1045 case ACPI_HEST_NOTIFY_NMI
:
1046 if (!IS_ENABLED(CONFIG_HAVE_ACPI_APEI_NMI
)) {
1047 pr_warn(GHES_PFX
"Generic hardware error source: %d notified via NMI interrupt is not supported!\n",
1048 generic
->header
.source_id
);
1052 case ACPI_HEST_NOTIFY_LOCAL
:
1053 pr_warning(GHES_PFX
"Generic hardware error source: %d notified via local interrupt is not supported!\n",
1054 generic
->header
.source_id
);
1057 pr_warning(FW_WARN GHES_PFX
"Unknown notification type: %u for generic hardware error source: %d\n",
1058 generic
->notify
.type
, generic
->header
.source_id
);
1063 if (generic
->error_block_length
<
1064 sizeof(struct acpi_hest_generic_status
)) {
1065 pr_warning(FW_BUG GHES_PFX
"Invalid error block length: %u for generic hardware error source: %d\n",
1066 generic
->error_block_length
,
1067 generic
->header
.source_id
);
1070 ghes
= ghes_new(generic
);
1077 rc
= ghes_edac_register(ghes
, &ghes_dev
->dev
);
1081 switch (generic
->notify
.type
) {
1082 case ACPI_HEST_NOTIFY_POLLED
:
1083 timer_setup(&ghes
->timer
, ghes_poll_func
, TIMER_DEFERRABLE
);
1084 ghes_add_timer(ghes
);
1086 case ACPI_HEST_NOTIFY_EXTERNAL
:
1087 /* External interrupt vector is GSI */
1088 rc
= acpi_gsi_to_irq(generic
->notify
.vector
, &ghes
->irq
);
1090 pr_err(GHES_PFX
"Failed to map GSI to IRQ for generic hardware error source: %d\n",
1091 generic
->header
.source_id
);
1092 goto err_edac_unreg
;
1094 rc
= request_irq(ghes
->irq
, ghes_irq_func
, IRQF_SHARED
,
1097 pr_err(GHES_PFX
"Failed to register IRQ for generic hardware error source: %d\n",
1098 generic
->header
.source_id
);
1099 goto err_edac_unreg
;
1103 case ACPI_HEST_NOTIFY_SCI
:
1104 case ACPI_HEST_NOTIFY_GSIV
:
1105 case ACPI_HEST_NOTIFY_GPIO
:
1106 mutex_lock(&ghes_list_mutex
);
1107 if (list_empty(&ghes_hed
))
1108 register_acpi_hed_notifier(&ghes_notifier_hed
);
1109 list_add_rcu(&ghes
->list
, &ghes_hed
);
1110 mutex_unlock(&ghes_list_mutex
);
1113 case ACPI_HEST_NOTIFY_SEA
:
1116 case ACPI_HEST_NOTIFY_NMI
:
1122 platform_set_drvdata(ghes_dev
, ghes
);
1124 /* Handle any pending errors right away */
1129 ghes_edac_unregister(ghes
);
1138 static int ghes_remove(struct platform_device
*ghes_dev
)
1141 struct acpi_hest_generic
*generic
;
1143 ghes
= platform_get_drvdata(ghes_dev
);
1144 generic
= ghes
->generic
;
1146 ghes
->flags
|= GHES_EXITING
;
1147 switch (generic
->notify
.type
) {
1148 case ACPI_HEST_NOTIFY_POLLED
:
1149 del_timer_sync(&ghes
->timer
);
1151 case ACPI_HEST_NOTIFY_EXTERNAL
:
1152 free_irq(ghes
->irq
, ghes
);
1155 case ACPI_HEST_NOTIFY_SCI
:
1156 case ACPI_HEST_NOTIFY_GSIV
:
1157 case ACPI_HEST_NOTIFY_GPIO
:
1158 mutex_lock(&ghes_list_mutex
);
1159 list_del_rcu(&ghes
->list
);
1160 if (list_empty(&ghes_hed
))
1161 unregister_acpi_hed_notifier(&ghes_notifier_hed
);
1162 mutex_unlock(&ghes_list_mutex
);
1166 case ACPI_HEST_NOTIFY_SEA
:
1167 ghes_sea_remove(ghes
);
1169 case ACPI_HEST_NOTIFY_NMI
:
1170 ghes_nmi_remove(ghes
);
1179 ghes_edac_unregister(ghes
);
1183 platform_set_drvdata(ghes_dev
, NULL
);
1188 static struct platform_driver ghes_platform_driver
= {
1192 .probe
= ghes_probe
,
1193 .remove
= ghes_remove
,
1196 static int __init
ghes_init(void)
1203 switch (hest_disable
) {
1204 case HEST_NOT_FOUND
:
1207 pr_info(GHES_PFX
"HEST is not enabled!\n");
1214 pr_info(GHES_PFX
"GHES is not enabled!\n");
1218 ghes_nmi_init_cxt();
1220 rc
= ghes_estatus_pool_init();
1224 rc
= ghes_estatus_pool_expand(GHES_ESTATUS_CACHE_AVG_SIZE
*
1225 GHES_ESTATUS_CACHE_ALLOCED_MAX
);
1229 rc
= platform_driver_register(&ghes_platform_driver
);
1233 rc
= apei_osc_setup();
1234 if (rc
== 0 && osc_sb_apei_support_acked
)
1235 pr_info(GHES_PFX
"APEI firmware first mode is enabled by APEI bit and WHEA _OSC.\n");
1236 else if (rc
== 0 && !osc_sb_apei_support_acked
)
1237 pr_info(GHES_PFX
"APEI firmware first mode is enabled by WHEA _OSC.\n");
1238 else if (rc
&& osc_sb_apei_support_acked
)
1239 pr_info(GHES_PFX
"APEI firmware first mode is enabled by APEI bit.\n");
1241 pr_info(GHES_PFX
"Failed to enable APEI firmware first mode.\n");
1245 ghes_estatus_pool_exit();
1249 device_initcall(ghes_init
);