projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'v3.10.108' into update
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / acpi / apei / ghes.c
CommitLineData
d334a491
HY		 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 *
67eb2e99 15 * Copyright 2010,2011 Intel Corp.
d334a491
HY		 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>
700130b4 36#include <linux/acpi_io.h>
d334a491
HY		 37#include <linux/io.h>
38#include <linux/interrupt.h>
81e88fdc 39#include <linux/timer.h>
d334a491
HY		 40#include <linux/cper.h>
41#include <linux/kdebug.h>
7ad6e943
HY		 42#include <linux/platform_device.h>
43#include <linux/mutex.h>
32c361f5 44#include <linux/ratelimit.h>
81e88fdc 45#include <linux/vmalloc.h>
67eb2e99
HY		 46#include <linux/irq_work.h>
47#include <linux/llist.h>
48#include <linux/genalloc.h>
a654e5ee
HY		 49#include <linux/pci.h>
50#include <linux/aer.h>
40e06415
MCC		 51
52#include <acpi/ghes.h>
d334a491 53#include <asm/mce.h>
81e88fdc 54#include <asm/tlbflush.h>
9c48f1c6 55#include <asm/nmi.h>
d334a491
HY		 56
57#include "apei-internal.h"
58
59#define GHES_PFX "GHES: "
60
61#define GHES_ESTATUS_MAX_SIZE 65536
67eb2e99
HY		 62#define GHES_ESOURCE_PREALLOC_MAX_SIZE 65536
63
64#define GHES_ESTATUS_POOL_MIN_ALLOC_ORDER 3
65
152cef40
HY		 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
70cb6e1d 71#define GHES_ESTATUS_IN_CACHE_MAX_NSEC 10000000000ULL
152cef40
HY		 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
67eb2e99
HY		 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))
d334a491 86
90ab5ee9 87bool ghes_disable;
b6a95016
HY		 88module_param_named(disable, ghes_disable, bool, 0);
89
81e88fdc
HY		 90static int ghes_panic_timeout __read_mostly = 30;
91
d334a491 92/*
81e88fdc
HY		 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.
d334a491 96 *
81e88fdc
HY		 97 * RCU is used for these lists, so ghes_list_mutex is only used for
98 * list changing, not for traversing.
d334a491
HY		 99 */
100static LIST_HEAD(ghes_sci);
81e88fdc 101static LIST_HEAD(ghes_nmi);
7ad6e943 102static DEFINE_MUTEX(ghes_list_mutex);
d334a491 103
81e88fdc
HY		 104/*
105 * NMI may be triggered on any CPU, so ghes_nmi_lock is used for
106 * mutual exclusion.
107 */
108static 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 */
126static 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 */
131static DEFINE_RAW_SPINLOCK(ghes_ioremap_lock_nmi);
132static DEFINE_SPINLOCK(ghes_ioremap_lock_irq);
133
67eb2e99
HY		 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 */
142static struct gen_pool *ghes_estatus_pool;
143static unsigned long ghes_estatus_pool_size_request;
144static struct llist_head ghes_estatus_llist;
145static struct irq_work ghes_proc_irq_work;
146
152cef40
HY		 147struct ghes_estatus_cache *ghes_estatus_caches[GHES_ESTATUS_CACHES_SIZE];
148static atomic_t ghes_estatus_cache_alloced;
149
81e88fdc
HY		 150static 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
162static void ghes_ioremap_exit(void)
163{
164 free_vm_area(ghes_ioremap_area);
165}
166
167static 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
178static 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
189static 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
199static 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
67eb2e99
HY		 209static 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
217static 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
224static 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
231static 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
253static void ghes_estatus_pool_shrink(unsigned long len)
254{
255 ghes_estatus_pool_size_request -= PAGE_ALIGN(len);
256}
257
d334a491
HY		 258static 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;
34ddeb03 268 rc = apei_map_generic_address(&generic->error_status_address);
d334a491
HY		 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
287err_unmap:
34ddeb03 288 apei_unmap_generic_address(&generic->error_status_address);
d334a491
HY		 289err_free:
290 kfree(ghes);
291 return ERR_PTR(rc);
292}
293
294static void ghes_fini(struct ghes *ghes)
295{
296 kfree(ghes->estatus);
34ddeb03 297 apei_unmap_generic_address(&ghes->generic->error_status_address);
d334a491
HY		 298}
299
d334a491
HY		 300static inline int ghes_severity(int severity)
301{
302 switch (severity) {
ad4ecef2
HY		 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;
d334a491 311 default:
25985edc 312 /* Unknown, go panic */
ad4ecef2 313 return GHES_SEV_PANIC;
d334a491
HY		 314 }
315}
316
81e88fdc
HY		 317static void ghes_copy_tofrom_phys(void *buffer, u64 paddr, u32 len,
318 int from_phys)
d334a491 319{
81e88fdc
HY		 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 }
d334a491
HY		 352}
353
354static 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
700130b4 361 rc = apei_read(&buf_paddr, &g->error_status_address);
d334a491
HY		 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
81e88fdc
HY		 372 ghes_copy_tofrom_phys(ghes->estatus, buf_paddr,
373 sizeof(*ghes->estatus), 1);
d334a491
HY		 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;
81e88fdc
HY		 388 ghes_copy_tofrom_phys(ghes->estatus + 1,
389 buf_paddr + sizeof(*ghes->estatus),
390 len - sizeof(*ghes->estatus), 1);
d334a491
HY		 391 if (apei_estatus_check(ghes->estatus))
392 goto err_read_block;
393 rc = 0;
394
395err_read_block:
81e88fdc 396 if (rc && !silent && printk_ratelimit())
d334a491
HY		 397 pr_warning(FW_WARN GHES_PFX
398 "Failed to read error status block!\n");
399 return rc;
400}
401
402static 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
21480547
MCC		 412static void ghes_do_proc(struct ghes *ghes,
413 const struct acpi_hest_generic_status *estatus)
d334a491 414{
ba61ca4a 415 int sev, sec_sev;
d334a491
HY		 416 struct acpi_hest_generic_data *gdata;
417
67eb2e99
HY		 418 sev = ghes_severity(estatus->error_severity);
419 apei_estatus_for_each_section(estatus, gdata) {
ba61ca4a 420 sec_sev = ghes_severity(gdata->error_severity);
d334a491
HY		 421 if (!uuid_le_cmp(*(uuid_le *)gdata->section_type,
422 CPER_SEC_PLATFORM_MEM)) {
ba61ca4a
HY		 423 struct cper_sec_mem_err *mem_err;
424 mem_err = (struct cper_sec_mem_err *)(gdata+1);
21480547
MCC		 425 ghes_edac_report_mem_error(ghes, sev, mem_err);
426
ba61ca4a
HY		 427#ifdef CONFIG_X86_MCE
428 apei_mce_report_mem_error(sev == GHES_SEV_CORRECTED,
429 mem_err);
d334a491 430#endif
ba61ca4a
HY		 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 }
a654e5ee
HY		 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,
37448adf
LO		 457 devfn, aer_severity,
458 (struct aer_capability_regs *)
459 pcie_err->aer_info);
a654e5ee
HY		 460 }
461
462 }
463#endif
d334a491 464 }
32c361f5 465}
d334a491 466
67eb2e99
HY		 467static void __ghes_print_estatus(const char *pfx,
468 const struct acpi_hest_generic *generic,
469 const struct acpi_hest_generic_status *estatus)
32c361f5 470{
5ba82ab5
HY		 471 static atomic_t seqno;
472 unsigned int curr_seqno;
473 char pfx_seq[64];
474
32c361f5 475 if (pfx == NULL) {
67eb2e99 476 if (ghes_severity(estatus->error_severity) <=
32c361f5 477 GHES_SEV_CORRECTED)
5ba82ab5 478 pfx = KERN_WARNING;
32c361f5 479 else
5ba82ab5 480 pfx = KERN_ERR;
32c361f5 481 }
5ba82ab5
HY		 482 curr_seqno = atomic_inc_return(&seqno);
483 snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}" HW_ERR, pfx, curr_seqno);
5588340d 484 printk("%s""Hardware error from APEI Generic Hardware Error Source: %d\n",
5ba82ab5
HY		 485 pfx_seq, generic->header.source_id);
486 apei_estatus_print(pfx_seq, estatus);
5588340d
HY		 487}
488
152cef40
HY		 489static int ghes_print_estatus(const char *pfx,
490 const struct acpi_hest_generic *generic,
491 const struct acpi_hest_generic_status *estatus)
5588340d
HY		 492{
493 /* Not more than 2 messages every 5 seconds */
67eb2e99
HY		 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;
5588340d 497
67eb2e99
HY		 498 if (ghes_severity(estatus->error_severity) <= GHES_SEV_CORRECTED)
499 ratelimit = &ratelimit_corrected;
500 else
501 ratelimit = &ratelimit_uncorrected;
152cef40 502 if (__ratelimit(ratelimit)) {
67eb2e99 503 __ghes_print_estatus(pfx, generic, estatus);
152cef40
HY		 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 */
513static 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
542static 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
572static 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
582static 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
590static 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);
70cb6e1d
LB		 617 period = duration;
618 do_div(period, (count + 1));
152cef40
HY		 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();
d334a491
HY		 634}
635
636static int ghes_proc(struct ghes *ghes)
637{
638 int rc;
639
640 rc = ghes_read_estatus(ghes, 0);
641 if (rc)
642 goto out;
152cef40
HY		 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 }
21480547 647 ghes_do_proc(ghes, ghes->estatus);
d334a491
HY		 648out:
649 ghes_clear_estatus(ghes);
bdb29c7b 650 return rc;
d334a491
HY		 651}
652
81e88fdc
HY		 653static 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
668static 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
677static 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
d334a491
HY		 689static 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
46d12f0b
HY		 705static 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
67eb2e99
HY		 719static void ghes_proc_in_irq(struct irq_work *irq_work)
720{
46d12f0b 721 struct llist_node *llnode, *next;
67eb2e99 722 struct ghes_estatus_node *estatus_node;
152cef40 723 struct acpi_hest_generic *generic;
67eb2e99
HY		 724 struct acpi_hest_generic_status *estatus;
725 u32 len, node_len;
726
46d12f0b 727 llnode = llist_del_all(&ghes_estatus_llist);
67eb2e99
HY		 728 /*
729 * Because the time order of estatus in list is reversed,
730 * revert it back to proper order.
731 */
46d12f0b 732 llnode = llist_nodes_reverse(llnode);
67eb2e99
HY		 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);
21480547 740 ghes_do_proc(estatus_node->ghes, estatus);
152cef40
HY		 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 }
67eb2e99
HY		 746 gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node,
747 node_len);
748 llnode = next;
749 }
750}
751
46d12f0b
HY		 752static 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
9c48f1c6 778static int ghes_notify_nmi(unsigned int cmd, struct pt_regs *regs)
81e88fdc
HY		 779{
780 struct ghes *ghes, *ghes_global = NULL;
781 int sev, sev_global = -1;
9c48f1c6 782 int ret = NMI_DONE;
81e88fdc
HY		 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 }
9c48f1c6 795 ret = NMI_HANDLED;
81e88fdc
HY		 796 }
797
9c48f1c6 798 if (ret == NMI_DONE)
81e88fdc
HY		 799 goto out;
800
801 if (sev_global >= GHES_SEV_PANIC) {
802 oops_begin();
46d12f0b 803 ghes_print_queued_estatus();
5ba82ab5 804 __ghes_print_estatus(KERN_EMERG, ghes_global->generic,
67eb2e99 805 ghes_global->estatus);
81e88fdc
HY		 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) {
67eb2e99
HY		 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
81e88fdc
HY		 818 if (!(ghes->flags & GHES_TO_CLEAR))
819 continue;
67eb2e99 820#ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
152cef40
HY		 821 if (ghes_estatus_cached(ghes->estatus))
822 goto next;
67eb2e99
HY		 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) {
21480547 829 estatus_node->ghes = ghes;
67eb2e99
HY		 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 }
152cef40 835next:
67eb2e99 836#endif
81e88fdc
HY		 837 ghes_clear_estatus(ghes);
838 }
67eb2e99
HY		 839#ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
840 irq_work_queue(&ghes_proc_irq_work);
841#endif
81e88fdc
HY		 842
843out:
844 raw_spin_unlock(&ghes_nmi_lock);
845 return ret;
846}
847
d334a491
HY		 848static struct notifier_block ghes_notifier_sci = {
849 .notifier_call = ghes_notify_sci,
850};
851
67eb2e99
HY		 852static 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
da095fd3 867static int ghes_probe(struct platform_device *ghes_dev)
d334a491
HY		 868{
869 struct acpi_hest_generic *generic;
870 struct ghes *ghes = NULL;
67eb2e99 871 unsigned long len;
7ad6e943 872 int rc = -EINVAL;
d334a491 873
1dd6b20e 874 generic = *(struct acpi_hest_generic **)ghes_dev->dev.platform_data;
d334a491 875 if (!generic->enabled)
7ad6e943 876 return -ENODEV;
d334a491 877
81e88fdc
HY		 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",
d334a491
HY		 886 generic->header.source_id);
887 goto err;
81e88fdc
HY		 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;
d334a491 892 }
81e88fdc
HY		 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,
d334a491
HY		 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 }
21480547
MCC		 908
909 rc = ghes_edac_register(ghes, &ghes_dev->dev);
910 if (rc < 0)
911 goto err;
912
81e88fdc
HY		 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 */
a98d4f64
WY		 922 rc = acpi_gsi_to_irq(generic->notify.vector, &ghes->irq);
923 if (rc) {
81e88fdc
HY		 924 pr_err(GHES_PFX "Failed to map GSI to IRQ for generic hardware error source: %d\n",
925 generic->header.source_id);
21480547 926 goto err_edac_unreg;
81e88fdc 927 }
a98d4f64
WY		 928 rc = request_irq(ghes->irq, ghes_irq_func, 0, "GHES IRQ", ghes);
929 if (rc) {
81e88fdc
HY		 930 pr_err(GHES_PFX "Failed to register IRQ for generic hardware error source: %d\n",
931 generic->header.source_id);
21480547 932 goto err_edac_unreg;
81e88fdc
HY		 933 }
934 break;
935 case ACPI_HEST_NOTIFY_SCI:
7ad6e943 936 mutex_lock(&ghes_list_mutex);
d334a491
HY		 937 if (list_empty(&ghes_sci))
938 register_acpi_hed_notifier(&ghes_notifier_sci);
939 list_add_rcu(&ghes->list, &ghes_sci);
7ad6e943 940 mutex_unlock(&ghes_list_mutex);
81e88fdc
HY		 941 break;
942 case ACPI_HEST_NOTIFY_NMI:
67eb2e99
HY		 943 len = ghes_esource_prealloc_size(generic);
944 ghes_estatus_pool_expand(len);
81e88fdc
HY		 945 mutex_lock(&ghes_list_mutex);
946 if (list_empty(&ghes_nmi))
9c48f1c6
DZ		 947 register_nmi_handler(NMI_LOCAL, ghes_notify_nmi, 0,
948 "ghes");
81e88fdc
HY		 949 list_add_rcu(&ghes->list, &ghes_nmi);
950 mutex_unlock(&ghes_list_mutex);
951 break;
952 default:
953 BUG();
d334a491 954 }
7ad6e943 955 platform_set_drvdata(ghes_dev, ghes);
d334a491
HY		 956
957 return 0;
21480547
MCC		 958err_edac_unreg:
959 ghes_edac_unregister(ghes);
d334a491 960err:
7ad6e943 961 if (ghes) {
d334a491 962 ghes_fini(ghes);
7ad6e943
HY		 963 kfree(ghes);
964 }
d334a491
HY		 965 return rc;
966}
967
b59bc2fb 968static int ghes_remove(struct platform_device *ghes_dev)
d334a491 969{
7ad6e943
HY		 970 struct ghes *ghes;
971 struct acpi_hest_generic *generic;
67eb2e99 972 unsigned long len;
d334a491 973
7ad6e943
HY		 974 ghes = platform_get_drvdata(ghes_dev);
975 generic = ghes->generic;
976
81e88fdc 977 ghes->flags |= GHES_EXITING;
7ad6e943 978 switch (generic->notify.type) {
81e88fdc
HY		 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;
7ad6e943
HY		 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);
71bec91e 991 synchronize_rcu();
7ad6e943 992 break;
81e88fdc
HY		 993 case ACPI_HEST_NOTIFY_NMI:
994 mutex_lock(&ghes_list_mutex);
995 list_del_rcu(&ghes->list);
996 if (list_empty(&ghes_nmi))
9c48f1c6 997 unregister_nmi_handler(NMI_LOCAL, "ghes");
81e88fdc
HY		 998 mutex_unlock(&ghes_list_mutex);
999 /*
1000 * To synchronize with NMI handler, ghes can only be
1001 * freed after NMI handler finishes.
1002 */
1003 synchronize_rcu();
67eb2e99
HY		 1004 len = ghes_esource_prealloc_size(generic);
1005 ghes_estatus_pool_shrink(len);
81e88fdc 1006 break;
7ad6e943
HY		 1007 default:
1008 BUG();
1009 break;
1010 }
d334a491 1011
7ad6e943 1012 ghes_fini(ghes);
21480547
MCC		 1013
1014 ghes_edac_unregister(ghes);
1015
7ad6e943 1016 kfree(ghes);
d334a491 1017
7ad6e943
HY		 1018 platform_set_drvdata(ghes_dev, NULL);
1019
1020 return 0;
d334a491
HY		 1021}
1022
7ad6e943
HY		 1023static struct platform_driver ghes_platform_driver = {
1024 .driver = {
1025 .name = "GHES",
1026 .owner = THIS_MODULE,
1027 },
1028 .probe = ghes_probe,
1029 .remove = ghes_remove,
1030};
1031
d334a491
HY		 1032static int __init ghes_init(void)
1033{
81e88fdc
HY		 1034 int rc;
1035
d334a491
HY		 1036 if (acpi_disabled)
1037 return -ENODEV;
1038
1039 if (hest_disable) {
1040 pr_info(GHES_PFX "HEST is not enabled!\n");
1041 return -EINVAL;
1042 }
1043
b6a95016
HY		 1044 if (ghes_disable) {
1045 pr_info(GHES_PFX "GHES is not enabled!\n");
1046 return -EINVAL;
1047 }
1048
67eb2e99
HY		 1049 init_irq_work(&ghes_proc_irq_work, ghes_proc_in_irq);
1050
81e88fdc
HY		 1051 rc = ghes_ioremap_init();
1052 if (rc)
1053 goto err;
1054
67eb2e99 1055 rc = ghes_estatus_pool_init();
81e88fdc
HY		 1056 if (rc)
1057 goto err_ioremap_exit;
1058
152cef40
HY		 1059 rc = ghes_estatus_pool_expand(GHES_ESTATUS_CACHE_AVG_SIZE *
1060 GHES_ESTATUS_CACHE_ALLOCED_MAX);
1061 if (rc)
1062 goto err_pool_exit;
1063
67eb2e99
HY		 1064 rc = platform_driver_register(&ghes_platform_driver);
1065 if (rc)
1066 goto err_pool_exit;
1067
9fb0bfe1
HY		 1068 rc = apei_osc_setup();
1069 if (rc == 0 && osc_sb_apei_support_acked)
1070 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit and WHEA _OSC.\n");
1071 else if (rc == 0 && !osc_sb_apei_support_acked)
1072 pr_info(GHES_PFX "APEI firmware first mode is enabled by WHEA _OSC.\n");
1073 else if (rc && osc_sb_apei_support_acked)
1074 pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit.\n");
1075 else
1076 pr_info(GHES_PFX "Failed to enable APEI firmware first mode.\n");
1077
81e88fdc 1078 return 0;
67eb2e99
HY		 1079err_pool_exit:
1080 ghes_estatus_pool_exit();
81e88fdc
HY		 1081err_ioremap_exit:
1082 ghes_ioremap_exit();
1083err:
1084 return rc;
d334a491
HY		 1085}
1086
1087static void __exit ghes_exit(void)
1088{
7ad6e943 1089 platform_driver_unregister(&ghes_platform_driver);
67eb2e99 1090 ghes_estatus_pool_exit();
81e88fdc 1091 ghes_ioremap_exit();
d334a491
HY		 1092}
1093
1094module_init(ghes_init);
1095module_exit(ghes_exit);
1096
1097MODULE_AUTHOR("Huang Ying");
1098MODULE_DESCRIPTION("APEI Generic Hardware Error Source support");
1099MODULE_LICENSE("GPL");
7ad6e943 1100MODULE_ALIAS("platform:GHES");
kernel source for telekom puls (alcatel/mediatek)