ACPI, APEI, Generic Hardware Error Source POLL/IRQ/NMI notification type support
[deliverable/linux.git] / drivers / acpi / apei / ghes.c
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 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/io.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 <acpi/apei.h>
46 #include <acpi/atomicio.h>
47 #include <acpi/hed.h>
48 #include <asm/mce.h>
49 #include <asm/tlbflush.h>
50
51 #include "apei-internal.h"
52
53 #define GHES_PFX "GHES: "
54
55 #define GHES_ESTATUS_MAX_SIZE 65536
56
57 /*
58 * One struct ghes is created for each generic hardware error source.
59 * It provides the context for APEI hardware error timer/IRQ/SCI/NMI
60 * handler.
61 *
62 * estatus: memory buffer for error status block, allocated during
63 * HEST parsing.
64 */
65 #define GHES_TO_CLEAR 0x0001
66 #define GHES_EXITING 0x0002
67
68 struct ghes {
69 struct acpi_hest_generic *generic;
70 struct acpi_hest_generic_status *estatus;
71 u64 buffer_paddr;
72 unsigned long flags;
73 union {
74 struct list_head list;
75 struct timer_list timer;
76 unsigned int irq;
77 };
78 };
79
80 static int ghes_panic_timeout __read_mostly = 30;
81
82 /*
83 * All error sources notified with SCI shares one notifier function,
84 * so they need to be linked and checked one by one. This is applied
85 * to NMI too.
86 *
87 * RCU is used for these lists, so ghes_list_mutex is only used for
88 * list changing, not for traversing.
89 */
90 static LIST_HEAD(ghes_sci);
91 static LIST_HEAD(ghes_nmi);
92 static DEFINE_MUTEX(ghes_list_mutex);
93
94 /*
95 * NMI may be triggered on any CPU, so ghes_nmi_lock is used for
96 * mutual exclusion.
97 */
98 static DEFINE_RAW_SPINLOCK(ghes_nmi_lock);
99
100 /*
101 * Because the memory area used to transfer hardware error information
102 * from BIOS to Linux can be determined only in NMI, IRQ or timer
103 * handler, but general ioremap can not be used in atomic context, so
104 * a special version of atomic ioremap is implemented for that.
105 */
106
107 /*
108 * Two virtual pages are used, one for NMI context, the other for
109 * IRQ/PROCESS context
110 */
111 #define GHES_IOREMAP_PAGES 2
112 #define GHES_IOREMAP_NMI_PAGE(base) (base)
113 #define GHES_IOREMAP_IRQ_PAGE(base) ((base) + PAGE_SIZE)
114
115 /* virtual memory area for atomic ioremap */
116 static struct vm_struct *ghes_ioremap_area;
117 /*
118 * These 2 spinlock is used to prevent atomic ioremap virtual memory
119 * area from being mapped simultaneously.
120 */
121 static DEFINE_RAW_SPINLOCK(ghes_ioremap_lock_nmi);
122 static DEFINE_SPINLOCK(ghes_ioremap_lock_irq);
123
124 static int ghes_ioremap_init(void)
125 {
126 ghes_ioremap_area = __get_vm_area(PAGE_SIZE * GHES_IOREMAP_PAGES,
127 VM_IOREMAP, VMALLOC_START, VMALLOC_END);
128 if (!ghes_ioremap_area) {
129 pr_err(GHES_PFX "Failed to allocate virtual memory area for atomic ioremap.\n");
130 return -ENOMEM;
131 }
132
133 return 0;
134 }
135
136 static void ghes_ioremap_exit(void)
137 {
138 free_vm_area(ghes_ioremap_area);
139 }
140
141 static void __iomem *ghes_ioremap_pfn_nmi(u64 pfn)
142 {
143 unsigned long vaddr;
144
145 vaddr = (unsigned long)GHES_IOREMAP_NMI_PAGE(ghes_ioremap_area->addr);
146 ioremap_page_range(vaddr, vaddr + PAGE_SIZE,
147 pfn << PAGE_SHIFT, PAGE_KERNEL);
148
149 return (void __iomem *)vaddr;
150 }
151
152 static void __iomem *ghes_ioremap_pfn_irq(u64 pfn)
153 {
154 unsigned long vaddr;
155
156 vaddr = (unsigned long)GHES_IOREMAP_IRQ_PAGE(ghes_ioremap_area->addr);
157 ioremap_page_range(vaddr, vaddr + PAGE_SIZE,
158 pfn << PAGE_SHIFT, PAGE_KERNEL);
159
160 return (void __iomem *)vaddr;
161 }
162
163 static void ghes_iounmap_nmi(void __iomem *vaddr_ptr)
164 {
165 unsigned long vaddr = (unsigned long __force)vaddr_ptr;
166 void *base = ghes_ioremap_area->addr;
167
168 BUG_ON(vaddr != (unsigned long)GHES_IOREMAP_NMI_PAGE(base));
169 unmap_kernel_range_noflush(vaddr, PAGE_SIZE);
170 __flush_tlb_one(vaddr);
171 }
172
173 static void ghes_iounmap_irq(void __iomem *vaddr_ptr)
174 {
175 unsigned long vaddr = (unsigned long __force)vaddr_ptr;
176 void *base = ghes_ioremap_area->addr;
177
178 BUG_ON(vaddr != (unsigned long)GHES_IOREMAP_IRQ_PAGE(base));
179 unmap_kernel_range_noflush(vaddr, PAGE_SIZE);
180 __flush_tlb_one(vaddr);
181 }
182
183 static struct ghes *ghes_new(struct acpi_hest_generic *generic)
184 {
185 struct ghes *ghes;
186 unsigned int error_block_length;
187 int rc;
188
189 ghes = kzalloc(sizeof(*ghes), GFP_KERNEL);
190 if (!ghes)
191 return ERR_PTR(-ENOMEM);
192 ghes->generic = generic;
193 rc = acpi_pre_map_gar(&generic->error_status_address);
194 if (rc)
195 goto err_free;
196 error_block_length = generic->error_block_length;
197 if (error_block_length > GHES_ESTATUS_MAX_SIZE) {
198 pr_warning(FW_WARN GHES_PFX
199 "Error status block length is too long: %u for "
200 "generic hardware error source: %d.\n",
201 error_block_length, generic->header.source_id);
202 error_block_length = GHES_ESTATUS_MAX_SIZE;
203 }
204 ghes->estatus = kmalloc(error_block_length, GFP_KERNEL);
205 if (!ghes->estatus) {
206 rc = -ENOMEM;
207 goto err_unmap;
208 }
209
210 return ghes;
211
212 err_unmap:
213 acpi_post_unmap_gar(&generic->error_status_address);
214 err_free:
215 kfree(ghes);
216 return ERR_PTR(rc);
217 }
218
219 static void ghes_fini(struct ghes *ghes)
220 {
221 kfree(ghes->estatus);
222 acpi_post_unmap_gar(&ghes->generic->error_status_address);
223 }
224
225 enum {
226 GHES_SEV_NO = 0x0,
227 GHES_SEV_CORRECTED = 0x1,
228 GHES_SEV_RECOVERABLE = 0x2,
229 GHES_SEV_PANIC = 0x3,
230 };
231
232 static inline int ghes_severity(int severity)
233 {
234 switch (severity) {
235 case CPER_SEV_INFORMATIONAL:
236 return GHES_SEV_NO;
237 case CPER_SEV_CORRECTED:
238 return GHES_SEV_CORRECTED;
239 case CPER_SEV_RECOVERABLE:
240 return GHES_SEV_RECOVERABLE;
241 case CPER_SEV_FATAL:
242 return GHES_SEV_PANIC;
243 default:
244 /* Unkown, go panic */
245 return GHES_SEV_PANIC;
246 }
247 }
248
249 static void ghes_copy_tofrom_phys(void *buffer, u64 paddr, u32 len,
250 int from_phys)
251 {
252 void __iomem *vaddr;
253 unsigned long flags = 0;
254 int in_nmi = in_nmi();
255 u64 offset;
256 u32 trunk;
257
258 while (len > 0) {
259 offset = paddr - (paddr & PAGE_MASK);
260 if (in_nmi) {
261 raw_spin_lock(&ghes_ioremap_lock_nmi);
262 vaddr = ghes_ioremap_pfn_nmi(paddr >> PAGE_SHIFT);
263 } else {
264 spin_lock_irqsave(&ghes_ioremap_lock_irq, flags);
265 vaddr = ghes_ioremap_pfn_irq(paddr >> PAGE_SHIFT);
266 }
267 trunk = PAGE_SIZE - offset;
268 trunk = min(trunk, len);
269 if (from_phys)
270 memcpy_fromio(buffer, vaddr + offset, trunk);
271 else
272 memcpy_toio(vaddr + offset, buffer, trunk);
273 len -= trunk;
274 paddr += trunk;
275 buffer += trunk;
276 if (in_nmi) {
277 ghes_iounmap_nmi(vaddr);
278 raw_spin_unlock(&ghes_ioremap_lock_nmi);
279 } else {
280 ghes_iounmap_irq(vaddr);
281 spin_unlock_irqrestore(&ghes_ioremap_lock_irq, flags);
282 }
283 }
284 }
285
286 static int ghes_read_estatus(struct ghes *ghes, int silent)
287 {
288 struct acpi_hest_generic *g = ghes->generic;
289 u64 buf_paddr;
290 u32 len;
291 int rc;
292
293 rc = acpi_atomic_read(&buf_paddr, &g->error_status_address);
294 if (rc) {
295 if (!silent && printk_ratelimit())
296 pr_warning(FW_WARN GHES_PFX
297 "Failed to read error status block address for hardware error source: %d.\n",
298 g->header.source_id);
299 return -EIO;
300 }
301 if (!buf_paddr)
302 return -ENOENT;
303
304 ghes_copy_tofrom_phys(ghes->estatus, buf_paddr,
305 sizeof(*ghes->estatus), 1);
306 if (!ghes->estatus->block_status)
307 return -ENOENT;
308
309 ghes->buffer_paddr = buf_paddr;
310 ghes->flags |= GHES_TO_CLEAR;
311
312 rc = -EIO;
313 len = apei_estatus_len(ghes->estatus);
314 if (len < sizeof(*ghes->estatus))
315 goto err_read_block;
316 if (len > ghes->generic->error_block_length)
317 goto err_read_block;
318 if (apei_estatus_check_header(ghes->estatus))
319 goto err_read_block;
320 ghes_copy_tofrom_phys(ghes->estatus + 1,
321 buf_paddr + sizeof(*ghes->estatus),
322 len - sizeof(*ghes->estatus), 1);
323 if (apei_estatus_check(ghes->estatus))
324 goto err_read_block;
325 rc = 0;
326
327 err_read_block:
328 if (rc && !silent && printk_ratelimit())
329 pr_warning(FW_WARN GHES_PFX
330 "Failed to read error status block!\n");
331 return rc;
332 }
333
334 static void ghes_clear_estatus(struct ghes *ghes)
335 {
336 ghes->estatus->block_status = 0;
337 if (!(ghes->flags & GHES_TO_CLEAR))
338 return;
339 ghes_copy_tofrom_phys(ghes->estatus, ghes->buffer_paddr,
340 sizeof(ghes->estatus->block_status), 0);
341 ghes->flags &= ~GHES_TO_CLEAR;
342 }
343
344 static void ghes_do_proc(struct ghes *ghes)
345 {
346 int sev, processed = 0;
347 struct acpi_hest_generic_data *gdata;
348
349 sev = ghes_severity(ghes->estatus->error_severity);
350 apei_estatus_for_each_section(ghes->estatus, gdata) {
351 #ifdef CONFIG_X86_MCE
352 if (!uuid_le_cmp(*(uuid_le *)gdata->section_type,
353 CPER_SEC_PLATFORM_MEM)) {
354 apei_mce_report_mem_error(
355 sev == GHES_SEV_CORRECTED,
356 (struct cper_sec_mem_err *)(gdata+1));
357 processed = 1;
358 }
359 #endif
360 }
361 }
362
363 static void ghes_print_estatus(const char *pfx, struct ghes *ghes)
364 {
365 /* Not more than 2 messages every 5 seconds */
366 static DEFINE_RATELIMIT_STATE(ratelimit, 5*HZ, 2);
367
368 if (pfx == NULL) {
369 if (ghes_severity(ghes->estatus->error_severity) <=
370 GHES_SEV_CORRECTED)
371 pfx = KERN_WARNING HW_ERR;
372 else
373 pfx = KERN_ERR HW_ERR;
374 }
375 if (__ratelimit(&ratelimit)) {
376 printk(
377 "%s""Hardware error from APEI Generic Hardware Error Source: %d\n",
378 pfx, ghes->generic->header.source_id);
379 apei_estatus_print(pfx, ghes->estatus);
380 }
381 }
382
383 static int ghes_proc(struct ghes *ghes)
384 {
385 int rc;
386
387 rc = ghes_read_estatus(ghes, 0);
388 if (rc)
389 goto out;
390 ghes_print_estatus(NULL, ghes);
391 ghes_do_proc(ghes);
392
393 out:
394 ghes_clear_estatus(ghes);
395 return 0;
396 }
397
398 static void ghes_add_timer(struct ghes *ghes)
399 {
400 struct acpi_hest_generic *g = ghes->generic;
401 unsigned long expire;
402
403 if (!g->notify.poll_interval) {
404 pr_warning(FW_WARN GHES_PFX "Poll interval is 0 for generic hardware error source: %d, disabled.\n",
405 g->header.source_id);
406 return;
407 }
408 expire = jiffies + msecs_to_jiffies(g->notify.poll_interval);
409 ghes->timer.expires = round_jiffies_relative(expire);
410 add_timer(&ghes->timer);
411 }
412
413 static void ghes_poll_func(unsigned long data)
414 {
415 struct ghes *ghes = (void *)data;
416
417 ghes_proc(ghes);
418 if (!(ghes->flags & GHES_EXITING))
419 ghes_add_timer(ghes);
420 }
421
422 static irqreturn_t ghes_irq_func(int irq, void *data)
423 {
424 struct ghes *ghes = data;
425 int rc;
426
427 rc = ghes_proc(ghes);
428 if (rc)
429 return IRQ_NONE;
430
431 return IRQ_HANDLED;
432 }
433
434 static int ghes_notify_sci(struct notifier_block *this,
435 unsigned long event, void *data)
436 {
437 struct ghes *ghes;
438 int ret = NOTIFY_DONE;
439
440 rcu_read_lock();
441 list_for_each_entry_rcu(ghes, &ghes_sci, list) {
442 if (!ghes_proc(ghes))
443 ret = NOTIFY_OK;
444 }
445 rcu_read_unlock();
446
447 return ret;
448 }
449
450 static int ghes_notify_nmi(struct notifier_block *this,
451 unsigned long cmd, void *data)
452 {
453 struct ghes *ghes, *ghes_global = NULL;
454 int sev, sev_global = -1;
455 int ret = NOTIFY_DONE;
456
457 if (cmd != DIE_NMI)
458 return ret;
459
460 raw_spin_lock(&ghes_nmi_lock);
461 list_for_each_entry_rcu(ghes, &ghes_nmi, list) {
462 if (ghes_read_estatus(ghes, 1)) {
463 ghes_clear_estatus(ghes);
464 continue;
465 }
466 sev = ghes_severity(ghes->estatus->error_severity);
467 if (sev > sev_global) {
468 sev_global = sev;
469 ghes_global = ghes;
470 }
471 ret = NOTIFY_STOP;
472 }
473
474 if (ret == NOTIFY_DONE)
475 goto out;
476
477 if (sev_global >= GHES_SEV_PANIC) {
478 oops_begin();
479 ghes_print_estatus(KERN_EMERG HW_ERR, ghes_global);
480 /* reboot to log the error! */
481 if (panic_timeout == 0)
482 panic_timeout = ghes_panic_timeout;
483 panic("Fatal hardware error!");
484 }
485
486 list_for_each_entry_rcu(ghes, &ghes_nmi, list) {
487 if (!(ghes->flags & GHES_TO_CLEAR))
488 continue;
489 /* Do not print estatus because printk is not NMI safe */
490 ghes_do_proc(ghes);
491 ghes_clear_estatus(ghes);
492 }
493
494 out:
495 raw_spin_unlock(&ghes_nmi_lock);
496 return ret;
497 }
498
499 static struct notifier_block ghes_notifier_sci = {
500 .notifier_call = ghes_notify_sci,
501 };
502
503 static struct notifier_block ghes_notifier_nmi = {
504 .notifier_call = ghes_notify_nmi,
505 };
506
507 static int __devinit ghes_probe(struct platform_device *ghes_dev)
508 {
509 struct acpi_hest_generic *generic;
510 struct ghes *ghes = NULL;
511 int rc = -EINVAL;
512
513 generic = *(struct acpi_hest_generic **)ghes_dev->dev.platform_data;
514 if (!generic->enabled)
515 return -ENODEV;
516
517 switch (generic->notify.type) {
518 case ACPI_HEST_NOTIFY_POLLED:
519 case ACPI_HEST_NOTIFY_EXTERNAL:
520 case ACPI_HEST_NOTIFY_SCI:
521 case ACPI_HEST_NOTIFY_NMI:
522 break;
523 case ACPI_HEST_NOTIFY_LOCAL:
524 pr_warning(GHES_PFX "Generic hardware error source: %d notified via local interrupt is not supported!\n",
525 generic->header.source_id);
526 goto err;
527 default:
528 pr_warning(FW_WARN GHES_PFX "Unknown notification type: %u for generic hardware error source: %d\n",
529 generic->notify.type, generic->header.source_id);
530 goto err;
531 }
532
533 rc = -EIO;
534 if (generic->error_block_length <
535 sizeof(struct acpi_hest_generic_status)) {
536 pr_warning(FW_BUG GHES_PFX "Invalid error block length: %u for generic hardware error source: %d\n",
537 generic->error_block_length,
538 generic->header.source_id);
539 goto err;
540 }
541 ghes = ghes_new(generic);
542 if (IS_ERR(ghes)) {
543 rc = PTR_ERR(ghes);
544 ghes = NULL;
545 goto err;
546 }
547 switch (generic->notify.type) {
548 case ACPI_HEST_NOTIFY_POLLED:
549 ghes->timer.function = ghes_poll_func;
550 ghes->timer.data = (unsigned long)ghes;
551 init_timer_deferrable(&ghes->timer);
552 ghes_add_timer(ghes);
553 break;
554 case ACPI_HEST_NOTIFY_EXTERNAL:
555 /* External interrupt vector is GSI */
556 if (acpi_gsi_to_irq(generic->notify.vector, &ghes->irq)) {
557 pr_err(GHES_PFX "Failed to map GSI to IRQ for generic hardware error source: %d\n",
558 generic->header.source_id);
559 goto err;
560 }
561 if (request_irq(ghes->irq, ghes_irq_func,
562 0, "GHES IRQ", ghes)) {
563 pr_err(GHES_PFX "Failed to register IRQ for generic hardware error source: %d\n",
564 generic->header.source_id);
565 goto err;
566 }
567 break;
568 case ACPI_HEST_NOTIFY_SCI:
569 mutex_lock(&ghes_list_mutex);
570 if (list_empty(&ghes_sci))
571 register_acpi_hed_notifier(&ghes_notifier_sci);
572 list_add_rcu(&ghes->list, &ghes_sci);
573 mutex_unlock(&ghes_list_mutex);
574 break;
575 case ACPI_HEST_NOTIFY_NMI:
576 mutex_lock(&ghes_list_mutex);
577 if (list_empty(&ghes_nmi))
578 register_die_notifier(&ghes_notifier_nmi);
579 list_add_rcu(&ghes->list, &ghes_nmi);
580 mutex_unlock(&ghes_list_mutex);
581 break;
582 default:
583 BUG();
584 }
585 platform_set_drvdata(ghes_dev, ghes);
586
587 return 0;
588 err:
589 if (ghes) {
590 ghes_fini(ghes);
591 kfree(ghes);
592 }
593 return rc;
594 }
595
596 static int __devexit ghes_remove(struct platform_device *ghes_dev)
597 {
598 struct ghes *ghes;
599 struct acpi_hest_generic *generic;
600
601 ghes = platform_get_drvdata(ghes_dev);
602 generic = ghes->generic;
603
604 ghes->flags |= GHES_EXITING;
605 switch (generic->notify.type) {
606 case ACPI_HEST_NOTIFY_POLLED:
607 del_timer_sync(&ghes->timer);
608 break;
609 case ACPI_HEST_NOTIFY_EXTERNAL:
610 free_irq(ghes->irq, ghes);
611 break;
612 case ACPI_HEST_NOTIFY_SCI:
613 mutex_lock(&ghes_list_mutex);
614 list_del_rcu(&ghes->list);
615 if (list_empty(&ghes_sci))
616 unregister_acpi_hed_notifier(&ghes_notifier_sci);
617 mutex_unlock(&ghes_list_mutex);
618 break;
619 case ACPI_HEST_NOTIFY_NMI:
620 mutex_lock(&ghes_list_mutex);
621 list_del_rcu(&ghes->list);
622 if (list_empty(&ghes_nmi))
623 unregister_die_notifier(&ghes_notifier_nmi);
624 mutex_unlock(&ghes_list_mutex);
625 /*
626 * To synchronize with NMI handler, ghes can only be
627 * freed after NMI handler finishes.
628 */
629 synchronize_rcu();
630 break;
631 default:
632 BUG();
633 break;
634 }
635
636 ghes_fini(ghes);
637 kfree(ghes);
638
639 platform_set_drvdata(ghes_dev, NULL);
640
641 return 0;
642 }
643
644 static struct platform_driver ghes_platform_driver = {
645 .driver = {
646 .name = "GHES",
647 .owner = THIS_MODULE,
648 },
649 .probe = ghes_probe,
650 .remove = ghes_remove,
651 };
652
653 static int __init ghes_init(void)
654 {
655 int rc;
656
657 if (acpi_disabled)
658 return -ENODEV;
659
660 if (hest_disable) {
661 pr_info(GHES_PFX "HEST is not enabled!\n");
662 return -EINVAL;
663 }
664
665 rc = ghes_ioremap_init();
666 if (rc)
667 goto err;
668
669 rc = platform_driver_register(&ghes_platform_driver);
670 if (rc)
671 goto err_ioremap_exit;
672
673 return 0;
674 err_ioremap_exit:
675 ghes_ioremap_exit();
676 err:
677 return rc;
678 }
679
680 static void __exit ghes_exit(void)
681 {
682 platform_driver_unregister(&ghes_platform_driver);
683 ghes_ioremap_exit();
684 }
685
686 module_init(ghes_init);
687 module_exit(ghes_exit);
688
689 MODULE_AUTHOR("Huang Ying");
690 MODULE_DESCRIPTION("APEI Generic Hardware Error Source support");
691 MODULE_LICENSE("GPL");
692 MODULE_ALIAS("platform:GHES");
This page took 0.056329 seconds and 5 git commands to generate.