1 /*
2 * APEI Generic Hardware Error Source support
3 *
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.
11 *
12 * For more information about Generic Hardware Error Source, please
13 * refer to ACPI Specification version 4.0, section 17.3.2.6
14 *
15 * Copyright 2010,2011 Intel Corp.
16 * Author: Huang Ying <ying.huang@intel.com>
17 *
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;
21 *
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.
26 *
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
30 */
31
32 #include <linux/kernel.h>
33 #include <linux/module.h>
34 #include <linux/init.h>
35 #include <linux/acpi.h>
36 #include <linux/acpi_io.h>
37 #include <linux/io.h>
38 #include <linux/interrupt.h>
39 #include <linux/timer.h>
40 #include <linux/cper.h>
41 #include <linux/kdebug.h>
42 #include <linux/platform_device.h>
43 #include <linux/mutex.h>
44 #include <linux/ratelimit.h>
45 #include <linux/vmalloc.h>
46 #include <linux/irq_work.h>
47 #include <linux/llist.h>
48 #include <linux/genalloc.h>
49 #include <linux/pci.h>
50 #include <linux/aer.h>
51
52 #include <acpi/ghes.h>
53 #include <asm/mce.h>
54 #include <asm/tlbflush.h>
55 #include <asm/nmi.h>
56
57 #include "apei-internal.h"
58
59 #define GHES_PFX "GHES: "
60
61 #define GHES_ESTATUS_MAX_SIZE 65536
62 #define GHES_ESOURCE_PREALLOC_MAX_SIZE 65536
63
64 #define GHES_ESTATUS_POOL_MIN_ALLOC_ORDER 3
65
66 /* This is just an estimation for memory pool allocation */
67 #define GHES_ESTATUS_CACHE_AVG_SIZE 512
68
69 #define GHES_ESTATUS_CACHES_SIZE 4
70
71 #define GHES_ESTATUS_IN_CACHE_MAX_NSEC 10000000000ULL
72 /* Prevent too many caches are allocated because of RCU */
73 #define GHES_ESTATUS_CACHE_ALLOCED_MAX (GHES_ESTATUS_CACHES_SIZE * 3 / 2)
74
75 #define GHES_ESTATUS_CACHE_LEN(estatus_len) \
76 (sizeof(struct ghes_estatus_cache) + (estatus_len))
77 #define GHES_ESTATUS_FROM_CACHE(estatus_cache) \
78 ((struct acpi_hest_generic_status *) \
79 ((struct ghes_estatus_cache *)(estatus_cache) + 1))
80
81 #define GHES_ESTATUS_NODE_LEN(estatus_len) \
82 (sizeof(struct ghes_estatus_node) + (estatus_len))
83 #define GHES_ESTATUS_FROM_NODE(estatus_node) \
84 ((struct acpi_hest_generic_status *) \
85 ((struct ghes_estatus_node *)(estatus_node) + 1))
86
87 bool ghes_disable;
88 module_param_named(disable, ghes_disable, bool, 0);
89
90 static int ghes_panic_timeout __read_mostly = 30;
91
92 /*
93 * All error sources notified with SCI shares one notifier function,
94 * so they need to be linked and checked one by one. This is applied
95 * to NMI too.
96 *
97 * RCU is used for these lists, so ghes_list_mutex is only used for
98 * list changing, not for traversing.
99 */
100 static LIST_HEAD(ghes_sci);
101 static LIST_HEAD(ghes_nmi);
102 static DEFINE_MUTEX(ghes_list_mutex);
103
104 /*
105 * NMI may be triggered on any CPU, so ghes_nmi_lock is used for
106 * mutual exclusion.
107 */
108 static DEFINE_RAW_SPINLOCK(ghes_nmi_lock);
109
110 /*
111 * Because the memory area used to transfer hardware error information
112 * from BIOS to Linux can be determined only in NMI, IRQ or timer
113 * handler, but general ioremap can not be used in atomic context, so
114 * a special version of atomic ioremap is implemented for that.
115 */
116
117 /*
118 * Two virtual pages are used, one for NMI context, the other for
119 * IRQ/PROCESS context
120 */
121 #define GHES_IOREMAP_PAGES 2
122 #define GHES_IOREMAP_NMI_PAGE(base) (base)
123 #define GHES_IOREMAP_IRQ_PAGE(base) ((base) + PAGE_SIZE)
124
125 /* virtual memory area for atomic ioremap */
126 static struct vm_struct *ghes_ioremap_area;
127 /*
128 * These 2 spinlock is used to prevent atomic ioremap virtual memory
129 * area from being mapped simultaneously.
130 */
131 static DEFINE_RAW_SPINLOCK(ghes_ioremap_lock_nmi);
132 static DEFINE_SPINLOCK(ghes_ioremap_lock_irq);
133
134 /*
135 * printk is not safe in NMI context. So in NMI handler, we allocate
136 * required memory from lock-less memory allocator
137 * (ghes_estatus_pool), save estatus into it, put them into lock-less
138 * list (ghes_estatus_llist), then delay printk into IRQ context via
139 * irq_work (ghes_proc_irq_work). ghes_estatus_size_request record
140 * required pool size by all NMI error source.
141 */
142 static struct gen_pool *ghes_estatus_pool;
143 static unsigned long ghes_estatus_pool_size_request;
144 static struct llist_head ghes_estatus_llist;
145 static struct irq_work ghes_proc_irq_work;
146
147 struct ghes_estatus_cache *ghes_estatus_caches[GHES_ESTATUS_CACHES_SIZE];
148 static atomic_t ghes_estatus_cache_alloced;
149
ghes_ioremap_init(void)150 static int ghes_ioremap_init(void)
151 {
152 ghes_ioremap_area = __get_vm_area(PAGE_SIZE * GHES_IOREMAP_PAGES,
153 VM_IOREMAP, VMALLOC_START, VMALLOC_END);
154 if (!ghes_ioremap_area) {
155 pr_err(GHES_PFX "Failed to allocate virtual memory area for atomic ioremap.\n");
156 return -ENOMEM;
157 }
158
159 return 0;
160 }
161
ghes_ioremap_exit(void)162 static void ghes_ioremap_exit(void)
163 {
164 free_vm_area(ghes_ioremap_area);
165 }
166
ghes_ioremap_pfn_nmi(u64 pfn)167 static void __iomem *ghes_ioremap_pfn_nmi(u64 pfn)
168 {
169 unsigned long vaddr;
170
171 vaddr = (unsigned long)GHES_IOREMAP_NMI_PAGE(ghes_ioremap_area->addr);
172 ioremap_page_range(vaddr, vaddr + PAGE_SIZE,
173 pfn << PAGE_SHIFT, PAGE_KERNEL);
174
175 return (void __iomem *)vaddr;
176 }
177
ghes_ioremap_pfn_irq(u64 pfn)178 static void __iomem *ghes_ioremap_pfn_irq(u64 pfn)
179 {
180 unsigned long vaddr;
181
182 vaddr = (unsigned long)GHES_IOREMAP_IRQ_PAGE(ghes_ioremap_area->addr);
183 ioremap_page_range(vaddr, vaddr + PAGE_SIZE,
184 pfn << PAGE_SHIFT, PAGE_KERNEL);
185
186 return (void __iomem *)vaddr;
187 }
188
ghes_iounmap_nmi(void __iomem * vaddr_ptr)189 static void ghes_iounmap_nmi(void __iomem *vaddr_ptr)
190 {
191 unsigned long vaddr = (unsigned long __force)vaddr_ptr;
192 void *base = ghes_ioremap_area->addr;
193
194 BUG_ON(vaddr != (unsigned long)GHES_IOREMAP_NMI_PAGE(base));
195 unmap_kernel_range_noflush(vaddr, PAGE_SIZE);
196 __flush_tlb_one(vaddr);
197 }
198
ghes_iounmap_irq(void __iomem * vaddr_ptr)199 static void ghes_iounmap_irq(void __iomem *vaddr_ptr)
200 {
201 unsigned long vaddr = (unsigned long __force)vaddr_ptr;
202 void *base = ghes_ioremap_area->addr;
203
204 BUG_ON(vaddr != (unsigned long)GHES_IOREMAP_IRQ_PAGE(base));
205 unmap_kernel_range_noflush(vaddr, PAGE_SIZE);
206 __flush_tlb_one(vaddr);
207 }
208
ghes_estatus_pool_init(void)209 static int ghes_estatus_pool_init(void)
210 {
211 ghes_estatus_pool = gen_pool_create(GHES_ESTATUS_POOL_MIN_ALLOC_ORDER, -1);
212 if (!ghes_estatus_pool)
213 return -ENOMEM;
214 return 0;
215 }
216
ghes_estatus_pool_free_chunk_page(struct gen_pool * pool,struct gen_pool_chunk * chunk,void * data)217 static void ghes_estatus_pool_free_chunk_page(struct gen_pool *pool,
218 struct gen_pool_chunk *chunk,
219 void *data)
220 {
221 free_page(chunk->start_addr);
222 }
223
ghes_estatus_pool_exit(void)224 static void ghes_estatus_pool_exit(void)
225 {
226 gen_pool_for_each_chunk(ghes_estatus_pool,
227 ghes_estatus_pool_free_chunk_page, NULL);
228 gen_pool_destroy(ghes_estatus_pool);
229 }
230
ghes_estatus_pool_expand(unsigned long len)231 static int ghes_estatus_pool_expand(unsigned long len)
232 {
233 unsigned long i, pages, size, addr;
234 int ret;
235
236 ghes_estatus_pool_size_request += PAGE_ALIGN(len);
237 size = gen_pool_size(ghes_estatus_pool);
238 if (size >= ghes_estatus_pool_size_request)
239 return 0;
240 pages = (ghes_estatus_pool_size_request - size) / PAGE_SIZE;
241 for (i = 0; i < pages; i++) {
242 addr = __get_free_page(GFP_KERNEL);
243 if (!addr)
244 return -ENOMEM;
245 ret = gen_pool_add(ghes_estatus_pool, addr, PAGE_SIZE, -1);
246 if (ret)
247 return ret;
248 }
249
250 return 0;
251 }
252
ghes_estatus_pool_shrink(unsigned long len)253 static void ghes_estatus_pool_shrink(unsigned long len)
254 {
255 ghes_estatus_pool_size_request -= PAGE_ALIGN(len);
256 }
257
ghes_new(struct acpi_hest_generic * generic)258 static struct ghes *ghes_new(struct acpi_hest_generic *generic)
259 {
260 struct ghes *ghes;
261 unsigned int error_block_length;
262 int rc;
263
264 ghes = kzalloc(sizeof(*ghes), GFP_KERNEL);
265 if (!ghes)
266 return ERR_PTR(-ENOMEM);
267 ghes->generic = generic;
268 rc = apei_map_generic_address(&generic->error_status_address);
269 if (rc)
270 goto err_free;
271 error_block_length = generic->error_block_length;
272 if (error_block_length > GHES_ESTATUS_MAX_SIZE) {
273 pr_warning(FW_WARN GHES_PFX
274 "Error status block length is too long: %u for "
275 "generic hardware error source: %d.\n",
276 error_block_length, generic->header.source_id);
277 error_block_length = GHES_ESTATUS_MAX_SIZE;
278 }
279 ghes->estatus = kmalloc(error_block_length, GFP_KERNEL);
280 if (!ghes->estatus) {
281 rc = -ENOMEM;
282 goto err_unmap;
283 }
284
285 return ghes;
286
287 err_unmap:
288 apei_unmap_generic_address(&generic->error_status_address);
289 err_free:
290 kfree(ghes);
291 return ERR_PTR(rc);
292 }
293
ghes_fini(struct ghes * ghes)294 static void ghes_fini(struct ghes *ghes)
295 {
296 kfree(ghes->estatus);
297 apei_unmap_generic_address(&ghes->generic->error_status_address);
298 }
299
ghes_severity(int severity)300 static inline int ghes_severity(int severity)
301 {
302 switch (severity) {
303 case CPER_SEV_INFORMATIONAL:
304 return GHES_SEV_NO;
305 case CPER_SEV_CORRECTED:
306 return GHES_SEV_CORRECTED;
307 case CPER_SEV_RECOVERABLE:
308 return GHES_SEV_RECOVERABLE;
309 case CPER_SEV_FATAL:
310 return GHES_SEV_PANIC;
311 default:
312 /* Unknown, go panic */
313 return GHES_SEV_PANIC;
314 }
315 }
316
ghes_copy_tofrom_phys(void * buffer,u64 paddr,u32 len,int from_phys)317 static void ghes_copy_tofrom_phys(void *buffer, u64 paddr, u32 len,
318 int from_phys)
319 {
320 void __iomem *vaddr;
321 unsigned long flags = 0;
322 int in_nmi = in_nmi();
323 u64 offset;
324 u32 trunk;
325
326 while (len > 0) {
327 offset = paddr - (paddr & PAGE_MASK);
328 if (in_nmi) {
329 raw_spin_lock(&ghes_ioremap_lock_nmi);
330 vaddr = ghes_ioremap_pfn_nmi(paddr >> PAGE_SHIFT);
331 } else {
332 spin_lock_irqsave(&ghes_ioremap_lock_irq, flags);
333 vaddr = ghes_ioremap_pfn_irq(paddr >> PAGE_SHIFT);
334 }
335 trunk = PAGE_SIZE - offset;
336 trunk = min(trunk, len);
337 if (from_phys)
338 memcpy_fromio(buffer, vaddr + offset, trunk);
339 else
340 memcpy_toio(vaddr + offset, buffer, trunk);
341 len -= trunk;
342 paddr += trunk;
343 buffer += trunk;
344 if (in_nmi) {
345 ghes_iounmap_nmi(vaddr);
346 raw_spin_unlock(&ghes_ioremap_lock_nmi);
347 } else {
348 ghes_iounmap_irq(vaddr);
349 spin_unlock_irqrestore(&ghes_ioremap_lock_irq, flags);
350 }
351 }
352 }
353
ghes_read_estatus(struct ghes * ghes,int silent)354 static int ghes_read_estatus(struct ghes *ghes, int silent)
355 {
356 struct acpi_hest_generic *g = ghes->generic;
357 u64 buf_paddr;
358 u32 len;
359 int rc;
360
361 rc = apei_read(&buf_paddr, &g->error_status_address);
362 if (rc) {
363 if (!silent && printk_ratelimit())
364 pr_warning(FW_WARN GHES_PFX
365 "Failed to read error status block address for hardware error source: %d.\n",
366 g->header.source_id);
367 return -EIO;
368 }
369 if (!buf_paddr)
370 return -ENOENT;
371
372 ghes_copy_tofrom_phys(ghes->estatus, buf_paddr,
373 sizeof(*ghes->estatus), 1);
374 if (!ghes->estatus->block_status)
375 return -ENOENT;
376
377 ghes->buffer_paddr = buf_paddr;
378 ghes->flags |= GHES_TO_CLEAR;
379
380 rc = -EIO;
381 len = apei_estatus_len(ghes->estatus);
382 if (len < sizeof(*ghes->estatus))
383 goto err_read_block;
384 if (len > ghes->generic->error_block_length)
385 goto err_read_block;
386 if (apei_estatus_check_header(ghes->estatus))
387 goto err_read_block;
388 ghes_copy_tofrom_phys(ghes->estatus + 1,
389 buf_paddr + sizeof(*ghes->estatus),
390 len - sizeof(*ghes->estatus), 1);
391 if (apei_estatus_check(ghes->estatus))
392 goto err_read_block;
393 rc = 0;
394
395 err_read_block:
396 if (rc && !silent && printk_ratelimit())
397 pr_warning(FW_WARN GHES_PFX
398 "Failed to read error status block!\n");
399 return rc;
400 }
401
ghes_clear_estatus(struct ghes * ghes)402 static void ghes_clear_estatus(struct ghes *ghes)
403 {
404 ghes->estatus->block_status = 0;
405 if (!(ghes->flags & GHES_TO_CLEAR))
406 return;
407 ghes_copy_tofrom_phys(ghes->estatus, ghes->buffer_paddr,
408 sizeof(ghes->estatus->block_status), 0);
409 ghes->flags &= ~GHES_TO_CLEAR;
410 }
411
ghes_do_proc(struct ghes * ghes,const struct acpi_hest_generic_status * estatus)412 static void ghes_do_proc(struct ghes *ghes,
413 const struct acpi_hest_generic_status *estatus)
414 {
415 int sev, sec_sev;
416 struct acpi_hest_generic_data *gdata;
417
418 sev = ghes_severity(estatus->error_severity);
419 apei_estatus_for_each_section(estatus, gdata) {
420 sec_sev = ghes_severity(gdata->error_severity);
421 if (!uuid_le_cmp(*(uuid_le *)gdata->section_type,
422 CPER_SEC_PLATFORM_MEM)) {
423 struct cper_sec_mem_err *mem_err;
424 mem_err = (struct cper_sec_mem_err *)(gdata+1);
425 ghes_edac_report_mem_error(ghes, sev, mem_err);
426
427 #ifdef CONFIG_X86_MCE
428 apei_mce_report_mem_error(sev == GHES_SEV_CORRECTED,
429 mem_err);
430 #endif
431 #ifdef CONFIG_ACPI_APEI_MEMORY_FAILURE
432 if (sev == GHES_SEV_RECOVERABLE &&
433 sec_sev == GHES_SEV_RECOVERABLE &&
434 mem_err->validation_bits & CPER_MEM_VALID_PHYSICAL_ADDRESS) {
435 unsigned long pfn;
436 pfn = mem_err->physical_addr >> PAGE_SHIFT;
437 memory_failure_queue(pfn, 0, 0);
438 }
439 #endif
440 }
441 #ifdef CONFIG_ACPI_APEI_PCIEAER
442 else if (!uuid_le_cmp(*(uuid_le *)gdata->section_type,
443 CPER_SEC_PCIE)) {
444 struct cper_sec_pcie *pcie_err;
445 pcie_err = (struct cper_sec_pcie *)(gdata+1);
446 if (sev == GHES_SEV_RECOVERABLE &&
447 sec_sev == GHES_SEV_RECOVERABLE &&
448 pcie_err->validation_bits & CPER_PCIE_VALID_DEVICE_ID &&
449 pcie_err->validation_bits & CPER_PCIE_VALID_AER_INFO) {
450 unsigned int devfn;
451 int aer_severity;
452 devfn = PCI_DEVFN(pcie_err->device_id.device,
453 pcie_err->device_id.function);
454 aer_severity = cper_severity_to_aer(sev);
455 aer_recover_queue(pcie_err->device_id.segment,
456 pcie_err->device_id.bus,
457 devfn, aer_severity,
458 (struct aer_capability_regs *)
459 pcie_err->aer_info);
460 }
461
462 }
463 #endif
464 }
465 }
466
__ghes_print_estatus(const char * pfx,const struct acpi_hest_generic * generic,const struct acpi_hest_generic_status * estatus)467 static void __ghes_print_estatus(const char *pfx,
468 const struct acpi_hest_generic *generic,
469 const struct acpi_hest_generic_status *estatus)
470 {
471 static atomic_t seqno;
472 unsigned int curr_seqno;
473 char pfx_seq[64];
474
475 if (pfx == NULL) {
476 if (ghes_severity(estatus->error_severity) <=
477 GHES_SEV_CORRECTED)
478 pfx = KERN_WARNING;
479 else
480 pfx = KERN_ERR;
481 }
482 curr_seqno = atomic_inc_return(&seqno);
483 snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}" HW_ERR, pfx, curr_seqno);
484 printk("%s""Hardware error from APEI Generic Hardware Error Source: %d\n",
485 pfx_seq, generic->header.source_id);
486 apei_estatus_print(pfx_seq, estatus);
487 }
488
ghes_print_estatus(const char * pfx,const struct acpi_hest_generic * generic,const struct acpi_hest_generic_status * estatus)489 static int ghes_print_estatus(const char *pfx,
490 const struct acpi_hest_generic *generic,
491 const struct acpi_hest_generic_status *estatus)
492 {
493 /* Not more than 2 messages every 5 seconds */
494 static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2);
495 static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2);
496 struct ratelimit_state *ratelimit;
497
498 if (ghes_severity(estatus->error_severity) <= GHES_SEV_CORRECTED)
499 ratelimit = &ratelimit_corrected;
500 else
501 ratelimit = &ratelimit_uncorrected;
502 if (__ratelimit(ratelimit)) {
503 __ghes_print_estatus(pfx, generic, estatus);
504 return 1;
505 }
506 return 0;
507 }
508
509 /*
510 * GHES error status reporting throttle, to report more kinds of
511 * errors, instead of just most frequently occurred errors.
512 */
ghes_estatus_cached(struct acpi_hest_generic_status * estatus)513 static int ghes_estatus_cached(struct acpi_hest_generic_status *estatus)
514 {
515 u32 len;
516 int i, cached = 0;
517 unsigned long long now;
518 struct ghes_estatus_cache *cache;
519 struct acpi_hest_generic_status *cache_estatus;
520
521 len = apei_estatus_len(estatus);
522 rcu_read_lock();
523 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
524 cache = rcu_dereference(ghes_estatus_caches[i]);
525 if (cache == NULL)
526 continue;
527 if (len != cache->estatus_len)
528 continue;
529 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
530 if (memcmp(estatus, cache_estatus, len))
531 continue;
532 atomic_inc(&cache->count);
533 now = sched_clock();
534 if (now - cache->time_in < GHES_ESTATUS_IN_CACHE_MAX_NSEC)
535 cached = 1;
536 break;
537 }
538 rcu_read_unlock();
539 return cached;
540 }
541
ghes_estatus_cache_alloc(struct acpi_hest_generic * generic,struct acpi_hest_generic_status * estatus)542 static struct ghes_estatus_cache *ghes_estatus_cache_alloc(
543 struct acpi_hest_generic *generic,
544 struct acpi_hest_generic_status *estatus)
545 {
546 int alloced;
547 u32 len, cache_len;
548 struct ghes_estatus_cache *cache;
549 struct acpi_hest_generic_status *cache_estatus;
550
551 alloced = atomic_add_return(1, &ghes_estatus_cache_alloced);
552 if (alloced > GHES_ESTATUS_CACHE_ALLOCED_MAX) {
553 atomic_dec(&ghes_estatus_cache_alloced);
554 return NULL;
555 }
556 len = apei_estatus_len(estatus);
557 cache_len = GHES_ESTATUS_CACHE_LEN(len);
558 cache = (void *)gen_pool_alloc(ghes_estatus_pool, cache_len);
559 if (!cache) {
560 atomic_dec(&ghes_estatus_cache_alloced);
561 return NULL;
562 }
563 cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
564 memcpy(cache_estatus, estatus, len);
565 cache->estatus_len = len;
566 atomic_set(&cache->count, 0);
567 cache->generic = generic;
568 cache->time_in = sched_clock();
569 return cache;
570 }
571
ghes_estatus_cache_free(struct ghes_estatus_cache * cache)572 static void ghes_estatus_cache_free(struct ghes_estatus_cache *cache)
573 {
574 u32 len;
575
576 len = apei_estatus_len(GHES_ESTATUS_FROM_CACHE(cache));
577 len = GHES_ESTATUS_CACHE_LEN(len);
578 gen_pool_free(ghes_estatus_pool, (unsigned long)cache, len);
579 atomic_dec(&ghes_estatus_cache_alloced);
580 }
581
ghes_estatus_cache_rcu_free(struct rcu_head * head)582 static void ghes_estatus_cache_rcu_free(struct rcu_head *head)
583 {
584 struct ghes_estatus_cache *cache;
585
586 cache = container_of(head, struct ghes_estatus_cache, rcu);
587 ghes_estatus_cache_free(cache);
588 }
589
ghes_estatus_cache_add(struct acpi_hest_generic * generic,struct acpi_hest_generic_status * estatus)590 static void ghes_estatus_cache_add(
591 struct acpi_hest_generic *generic,
592 struct acpi_hest_generic_status *estatus)
593 {
594 int i, slot = -1, count;
595 unsigned long long now, duration, period, max_period = 0;
596 struct ghes_estatus_cache *cache, *slot_cache = NULL, *new_cache;
597
598 new_cache = ghes_estatus_cache_alloc(generic, estatus);
599 if (new_cache == NULL)
600 return;
601 rcu_read_lock();
602 now = sched_clock();
603 for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
604 cache = rcu_dereference(ghes_estatus_caches[i]);
605 if (cache == NULL) {
606 slot = i;
607 slot_cache = NULL;
608 break;
609 }
610 duration = now - cache->time_in;
611 if (duration >= GHES_ESTATUS_IN_CACHE_MAX_NSEC) {
612 slot = i;
613 slot_cache = cache;
614 break;
615 }
616 count = atomic_read(&cache->count);
617 period = duration;
618 do_div(period, (count + 1));
619 if (period > max_period) {
620 max_period = period;
621 slot = i;
622 slot_cache = cache;
623 }
624 }
625 /* new_cache must be put into array after its contents are written */
626 smp_wmb();
627 if (slot != -1 && cmpxchg(ghes_estatus_caches + slot,
628 slot_cache, new_cache) == slot_cache) {
629 if (slot_cache)
630 call_rcu(&slot_cache->rcu, ghes_estatus_cache_rcu_free);
631 } else
632 ghes_estatus_cache_free(new_cache);
633 rcu_read_unlock();
634 }
635
ghes_proc(struct ghes * ghes)636 static int ghes_proc(struct ghes *ghes)
637 {
638 int rc;
639
640 rc = ghes_read_estatus(ghes, 0);
641 if (rc)
642 goto out;
643 if (!ghes_estatus_cached(ghes->estatus)) {
644 if (ghes_print_estatus(NULL, ghes->generic, ghes->estatus))
645 ghes_estatus_cache_add(ghes->generic, ghes->estatus);
646 }
647 ghes_do_proc(ghes, ghes->estatus);
648 out:
649 ghes_clear_estatus(ghes);
650 return 0;
651 }
652
ghes_add_timer(struct ghes * ghes)653 static void ghes_add_timer(struct ghes *ghes)
654 {
655 struct acpi_hest_generic *g = ghes->generic;
656 unsigned long expire;
657
658 if (!g->notify.poll_interval) {
659 pr_warning(FW_WARN GHES_PFX "Poll interval is 0 for generic hardware error source: %d, disabled.\n",
660 g->header.source_id);
661 return;
662 }
663 expire = jiffies + msecs_to_jiffies(g->notify.poll_interval);
664 ghes->timer.expires = round_jiffies_relative(expire);
665 add_timer(&ghes->timer);
666 }
667
ghes_poll_func(unsigned long data)668 static void ghes_poll_func(unsigned long data)
669 {
670 struct ghes *ghes = (void *)data;
671
672 ghes_proc(ghes);
673 if (!(ghes->flags & GHES_EXITING))
674 ghes_add_timer(ghes);
675 }
676
ghes_irq_func(int irq,void * data)677 static irqreturn_t ghes_irq_func(int irq, void *data)
678 {
679 struct ghes *ghes = data;
680 int rc;
681
682 rc = ghes_proc(ghes);
683 if (rc)
684 return IRQ_NONE;
685
686 return IRQ_HANDLED;
687 }
688
ghes_notify_sci(struct notifier_block * this,unsigned long event,void * data)689 static int ghes_notify_sci(struct notifier_block *this,
690 unsigned long event, void *data)
691 {
692 struct ghes *ghes;
693 int ret = NOTIFY_DONE;
694
695 rcu_read_lock();
696 list_for_each_entry_rcu(ghes, &ghes_sci, list) {
697 if (!ghes_proc(ghes))
698 ret = NOTIFY_OK;
699 }
700 rcu_read_unlock();
701
702 return ret;
703 }
704
llist_nodes_reverse(struct llist_node * llnode)705 static struct llist_node *llist_nodes_reverse(struct llist_node *llnode)
706 {
707 struct llist_node *next, *tail = NULL;
708
709 while (llnode) {
710 next = llnode->next;
711 llnode->next = tail;
712 tail = llnode;
713 llnode = next;
714 }
715
716 return tail;
717 }
718
ghes_proc_in_irq(struct irq_work * irq_work)719 static void ghes_proc_in_irq(struct irq_work *irq_work)
720 {
721 struct llist_node *llnode, *next;
722 struct ghes_estatus_node *estatus_node;
723 struct acpi_hest_generic *generic;
724 struct acpi_hest_generic_status *estatus;
725 u32 len, node_len;
726
727 llnode = llist_del_all(&ghes_estatus_llist);
728 /*
729 * Because the time order of estatus in list is reversed,
730 * revert it back to proper order.
731 */
732 llnode = llist_nodes_reverse(llnode);
733 while (llnode) {
734 next = llnode->next;
735 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
736 llnode);
737 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
738 len = apei_estatus_len(estatus);
739 node_len = GHES_ESTATUS_NODE_LEN(len);
740 ghes_do_proc(estatus_node->ghes, estatus);
741 if (!ghes_estatus_cached(estatus)) {
742 generic = estatus_node->generic;
743 if (ghes_print_estatus(NULL, generic, estatus))
744 ghes_estatus_cache_add(generic, estatus);
745 }
746 gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node,
747 node_len);
748 llnode = next;
749 }
750 }
751
ghes_print_queued_estatus(void)752 static void ghes_print_queued_estatus(void)
753 {
754 struct llist_node *llnode;
755 struct ghes_estatus_node *estatus_node;
756 struct acpi_hest_generic *generic;
757 struct acpi_hest_generic_status *estatus;
758 u32 len, node_len;
759
760 llnode = llist_del_all(&ghes_estatus_llist);
761 /*
762 * Because the time order of estatus in list is reversed,
763 * revert it back to proper order.
764 */
765 llnode = llist_nodes_reverse(llnode);
766 while (llnode) {
767 estatus_node = llist_entry(llnode, struct ghes_estatus_node,
768 llnode);
769 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
770 len = apei_estatus_len(estatus);
771 node_len = GHES_ESTATUS_NODE_LEN(len);
772 generic = estatus_node->generic;
773 ghes_print_estatus(NULL, generic, estatus);
774 llnode = llnode->next;
775 }
776 }
777
ghes_notify_nmi(unsigned int cmd,struct pt_regs * regs)778 static int ghes_notify_nmi(unsigned int cmd, struct pt_regs *regs)
779 {
780 struct ghes *ghes, *ghes_global = NULL;
781 int sev, sev_global = -1;
782 int ret = NMI_DONE;
783
784 raw_spin_lock(&ghes_nmi_lock);
785 list_for_each_entry_rcu(ghes, &ghes_nmi, list) {
786 if (ghes_read_estatus(ghes, 1)) {
787 ghes_clear_estatus(ghes);
788 continue;
789 }
790 sev = ghes_severity(ghes->estatus->error_severity);
791 if (sev > sev_global) {
792 sev_global = sev;
793 ghes_global = ghes;
794 }
795 ret = NMI_HANDLED;
796 }
797
798 if (ret == NMI_DONE)
799 goto out;
800
801 if (sev_global >= GHES_SEV_PANIC) {
802 oops_begin();
803 ghes_print_queued_estatus();
804 __ghes_print_estatus(KERN_EMERG, ghes_global->generic,
805 ghes_global->estatus);
806 /* reboot to log the error! */
807 if (panic_timeout == 0)
808 panic_timeout = ghes_panic_timeout;
809 panic("Fatal hardware error!");
810 }
811
812 list_for_each_entry_rcu(ghes, &ghes_nmi, list) {
813 #ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
814 u32 len, node_len;
815 struct ghes_estatus_node *estatus_node;
816 struct acpi_hest_generic_status *estatus;
817 #endif
818 if (!(ghes->flags & GHES_TO_CLEAR))
819 continue;
820 #ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
821 if (ghes_estatus_cached(ghes->estatus))
822 goto next;
823 /* Save estatus for further processing in IRQ context */
824 len = apei_estatus_len(ghes->estatus);
825 node_len = GHES_ESTATUS_NODE_LEN(len);
826 estatus_node = (void *)gen_pool_alloc(ghes_estatus_pool,
827 node_len);
828 if (estatus_node) {
829 estatus_node->ghes = ghes;
830 estatus_node->generic = ghes->generic;
831 estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
832 memcpy(estatus, ghes->estatus, len);
833 llist_add(&estatus_node->llnode, &ghes_estatus_llist);
834 }
835 next:
836 #endif
837 ghes_clear_estatus(ghes);
838 }
839 #ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
840 irq_work_queue(&ghes_proc_irq_work);
841 #endif
842
843 out:
844 raw_spin_unlock(&ghes_nmi_lock);
845 return ret;
846 }
847
848 static struct notifier_block ghes_notifier_sci = {
849 .notifier_call = ghes_notify_sci,
850 };
851
ghes_esource_prealloc_size(const struct acpi_hest_generic * generic)852 static unsigned long ghes_esource_prealloc_size(
853 const struct acpi_hest_generic *generic)
854 {
855 unsigned long block_length, prealloc_records, prealloc_size;
856
857 block_length = min_t(unsigned long, generic->error_block_length,
858 GHES_ESTATUS_MAX_SIZE);
859 prealloc_records = max_t(unsigned long,
860 generic->records_to_preallocate, 1);
861 prealloc_size = min_t(unsigned long, block_length * prealloc_records,
862 GHES_ESOURCE_PREALLOC_MAX_SIZE);
863
864 return prealloc_size;
865 }
866
ghes_probe(struct platform_device * ghes_dev)867 static int ghes_probe(struct platform_device *ghes_dev)
868 {
869 struct acpi_hest_generic *generic;
870 struct ghes *ghes = NULL;
871 unsigned long len;
872 int rc = -EINVAL;
873
874 generic = *(struct acpi_hest_generic **)ghes_dev->dev.platform_data;
875 if (!generic->enabled)
876 return -ENODEV;
877
878 switch (generic->notify.type) {
879 case ACPI_HEST_NOTIFY_POLLED:
880 case ACPI_HEST_NOTIFY_EXTERNAL:
881 case ACPI_HEST_NOTIFY_SCI:
882 case ACPI_HEST_NOTIFY_NMI:
883 break;
884 case ACPI_HEST_NOTIFY_LOCAL:
885 pr_warning(GHES_PFX "Generic hardware error source: %d notified via local interrupt is not supported!\n",
886 generic->header.source_id);
887 goto err;
888 default:
889 pr_warning(FW_WARN GHES_PFX "Unknown notification type: %u for generic hardware error source: %d\n",
890 generic->notify.type, generic->header.source_id);
891 goto err;
892 }
893
894 rc = -EIO;
895 if (generic->error_block_length <
896 sizeof(struct acpi_hest_generic_status)) {
897 pr_warning(FW_BUG GHES_PFX "Invalid error block length: %u for generic hardware error source: %d\n",
898 generic->error_block_length,
899 generic->header.source_id);
900 goto err;
901 }
902 ghes = ghes_new(generic);
903 if (IS_ERR(ghes)) {
904 rc = PTR_ERR(ghes);
905 ghes = NULL;
906 goto err;
907 }
908
909 rc = ghes_edac_register(ghes, &ghes_dev->dev);
910 if (rc < 0)
911 goto err;
912
913 switch (generic->notify.type) {
914 case ACPI_HEST_NOTIFY_POLLED:
915 ghes->timer.function = ghes_poll_func;
916 ghes->timer.data = (unsigned long)ghes;
917 init_timer_deferrable(&ghes->timer);
918 ghes_add_timer(ghes);
919 break;
920 case ACPI_HEST_NOTIFY_EXTERNAL:
921 /* External interrupt vector is GSI */
922 rc = acpi_gsi_to_irq(generic->notify.vector, &ghes->irq);
923 if (rc) {
924 pr_err(GHES_PFX "Failed to map GSI to IRQ for generic hardware error source: %d\n",
925 generic->header.source_id);
926 goto err_edac_unreg;
927 }
928 rc = request_irq(ghes->irq, ghes_irq_func, 0, "GHES IRQ", ghes);
929 if (rc) {
930 pr_err(GHES_PFX "Failed to register IRQ for generic hardware error source: %d\n",
931 generic->header.source_id);
932 goto err_edac_unreg;
933 }
934 break;
935 case ACPI_HEST_NOTIFY_SCI:
936 mutex_lock(&ghes_list_mutex);
937 if (list_empty(&ghes_sci))
938 register_acpi_hed_notifier(&ghes_notifier_sci);
939 list_add_rcu(&ghes->list, &ghes_sci);
940 mutex_unlock(&ghes_list_mutex);
941 break;
942 case ACPI_HEST_NOTIFY_NMI:
943 len = ghes_esource_prealloc_size(generic);
944 ghes_estatus_pool_expand(len);
945 mutex_lock(&ghes_list_mutex);
946 if (list_empty(&ghes_nmi))
947 register_nmi_handler(NMI_LOCAL, ghes_notify_nmi, 0,
948 "ghes");
949 list_add_rcu(&ghes->list, &ghes_nmi);
950 mutex_unlock(&ghes_list_mutex);
951 break;
952 default:
953 BUG();
954 }
955 platform_set_drvdata(ghes_dev, ghes);
956
957 return 0;
958 err_edac_unreg:
959 ghes_edac_unregister(ghes);
960 err:
961 if (ghes) {
962 ghes_fini(ghes);
963 kfree(ghes);
964 }
965 return rc;
966 }
967
ghes_remove(struct platform_device * ghes_dev)968 static int ghes_remove(struct platform_device *ghes_dev)
969 {
970 struct ghes *ghes;
971 struct acpi_hest_generic *generic;
972 unsigned long len;
973
974 ghes = platform_get_drvdata(ghes_dev);
975 generic = ghes->generic;
976
977 ghes->flags |= GHES_EXITING;
978 switch (generic->notify.type) {
979 case ACPI_HEST_NOTIFY_POLLED:
980 del_timer_sync(&ghes->timer);
981 break;
982 case ACPI_HEST_NOTIFY_EXTERNAL:
983 free_irq(ghes->irq, ghes);
984 break;
985 case ACPI_HEST_NOTIFY_SCI:
986 mutex_lock(&ghes_list_mutex);
987 list_del_rcu(&ghes->list);
988 if (list_empty(&ghes_sci))
989 unregister_acpi_hed_notifier(&ghes_notifier_sci);
990 mutex_unlock(&ghes_list_mutex);
991 break;
992 case ACPI_HEST_NOTIFY_NMI:
993 mutex_lock(&ghes_list_mutex);
994 list_del_rcu(&ghes->list);
995 if (list_empty(&ghes_nmi))
996 unregister_nmi_handler(NMI_LOCAL, "ghes");
997 mutex_unlock(&ghes_list_mutex);
998 /*
999 * To synchronize with NMI handler, ghes can only be
1000 * freed after NMI handler finishes.
1001 */
1002 synchronize_rcu();
1003 len = ghes_esource_prealloc_size(generic);
1004 ghes_estatus_pool_shrink(len);
1005 break;
1006 default:
1007 BUG();
1008 break;
1009 }
1010
1011 ghes_fini(ghes);
1012
1013 ghes_edac_unregister(ghes);
1014
1015 kfree(ghes);
1016
1017 platform_set_drvdata(ghes_dev, NULL);
1018
1019 return 0;
1020 }
1021
1022 static struct platform_driver ghes_platform_driver = {
1023 .driver = {
1024 .name = "GHES",
1025 .owner = THIS_MODULE,
1026 },
1027 .probe = ghes_probe,
1028 .remove = ghes_remove,
1029 };
1030
ghes_init(void)1031 static int __init ghes_init(void)
1032 {
1033 int rc;
1034
1035 if (acpi_disabled)
1036 return -ENODEV;
1037
1038 if (hest_disable) {
1039 pr_info(GHES_PFX "HEST is not enabled!\n");
1040 return -EINVAL;
1041 }
1042
1043 if (ghes_disable) {
1044 pr_info(GHES_PFX "GHES is not enabled!\n");
1045 return -EINVAL;
1046 }
1047
1048 init_irq_work(&ghes_proc_irq_work, ghes_proc_in_irq);
1049
1050 rc = ghes_ioremap_init();
1051 if (rc)
1052 goto err;
1053
1054 rc = ghes_estatus_pool_init();
1055 if (rc)
1056 goto err_ioremap_exit;
1057
1058 rc = ghes_estatus_pool_expand(GHES_ESTATUS_CACHE_AVG_SIZE *
1059 GHES_ESTATUS_CACHE_ALLOCED_MAX);
1060 if (rc)
1061 goto err_pool_exit;
1062
1063 rc = platform_driver_register(&ghes_platform_driver);
1064 if (rc)
1065 goto err_pool_exit;
1066
1067 rc = apei_osc_setup();
1068 if (rc == 0 && osc_sb_apei_support_acked)
1069 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit and WHEA _OSC.\n");
1070 else if (rc == 0 && !osc_sb_apei_support_acked)
1071 pr_info(GHES_PFX "APEI firmware first mode is enabled by WHEA _OSC.\n");
1072 else if (rc && osc_sb_apei_support_acked)
1073 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit.\n");
1074 else
1075 pr_info(GHES_PFX "Failed to enable APEI firmware first mode.\n");
1076
1077 return 0;
1078 err_pool_exit:
1079 ghes_estatus_pool_exit();
1080 err_ioremap_exit:
1081 ghes_ioremap_exit();
1082 err:
1083 return rc;
1084 }
1085
ghes_exit(void)1086 static void __exit ghes_exit(void)
1087 {
1088 platform_driver_unregister(&ghes_platform_driver);
1089 ghes_estatus_pool_exit();
1090 ghes_ioremap_exit();
1091 }
1092
1093 module_init(ghes_init);
1094 module_exit(ghes_exit);
1095
1096 MODULE_AUTHOR("Huang Ying");
1097 MODULE_DESCRIPTION("APEI Generic Hardware Error Source support");
1098 MODULE_LICENSE("GPL");
1099 MODULE_ALIAS("platform:GHES");
1100