xref: /openbmc/linux/drivers/acpi/apei/ghes.c (revision 6dfcd296)
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 
28 #include <linux/kernel.h>
29 #include <linux/moduleparam.h>
30 #include <linux/init.h>
31 #include <linux/acpi.h>
32 #include <linux/io.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 
48 #include <acpi/ghes.h>
49 #include <acpi/apei.h>
50 #include <asm/tlbflush.h>
51 
52 #include "apei-internal.h"
53 
54 #define GHES_PFX	"GHES: "
55 
56 #define GHES_ESTATUS_MAX_SIZE		65536
57 #define GHES_ESOURCE_PREALLOC_MAX_SIZE	65536
58 
59 #define GHES_ESTATUS_POOL_MIN_ALLOC_ORDER 3
60 
61 /* This is just an estimation for memory pool allocation */
62 #define GHES_ESTATUS_CACHE_AVG_SIZE	512
63 
64 #define GHES_ESTATUS_CACHES_SIZE	4
65 
66 #define GHES_ESTATUS_IN_CACHE_MAX_NSEC	10000000000ULL
67 /* Prevent too many caches are allocated because of RCU */
68 #define GHES_ESTATUS_CACHE_ALLOCED_MAX	(GHES_ESTATUS_CACHES_SIZE * 3 / 2)
69 
70 #define GHES_ESTATUS_CACHE_LEN(estatus_len)			\
71 	(sizeof(struct ghes_estatus_cache) + (estatus_len))
72 #define GHES_ESTATUS_FROM_CACHE(estatus_cache)			\
73 	((struct acpi_hest_generic_status *)				\
74 	 ((struct ghes_estatus_cache *)(estatus_cache) + 1))
75 
76 #define GHES_ESTATUS_NODE_LEN(estatus_len)			\
77 	(sizeof(struct ghes_estatus_node) + (estatus_len))
78 #define GHES_ESTATUS_FROM_NODE(estatus_node)			\
79 	((struct acpi_hest_generic_status *)				\
80 	 ((struct ghes_estatus_node *)(estatus_node) + 1))
81 
82 /*
83  * This driver isn't really modular, however for the time being,
84  * continuing to use module_param is the easiest way to remain
85  * compatible with existing boot arg use cases.
86  */
87 bool ghes_disable;
88 module_param_named(disable, ghes_disable, bool, 0);
89 
90 /*
91  * All error sources notified with SCI shares one notifier function,
92  * so they need to be linked and checked one by one.  This is applied
93  * to NMI too.
94  *
95  * RCU is used for these lists, so ghes_list_mutex is only used for
96  * list changing, not for traversing.
97  */
98 static LIST_HEAD(ghes_sci);
99 static DEFINE_MUTEX(ghes_list_mutex);
100 
101 /*
102  * Because the memory area used to transfer hardware error information
103  * from BIOS to Linux can be determined only in NMI, IRQ or timer
104  * handler, but general ioremap can not be used in atomic context, so
105  * a special version of atomic ioremap is implemented for that.
106  */
107 
108 /*
109  * Two virtual pages are used, one for IRQ/PROCESS context, the other for
110  * NMI context (optionally).
111  */
112 #ifdef CONFIG_HAVE_ACPI_APEI_NMI
113 #define GHES_IOREMAP_PAGES           2
114 #else
115 #define GHES_IOREMAP_PAGES           1
116 #endif
117 #define GHES_IOREMAP_IRQ_PAGE(base)	(base)
118 #define GHES_IOREMAP_NMI_PAGE(base)	((base) + PAGE_SIZE)
119 
120 /* virtual memory area for atomic ioremap */
121 static struct vm_struct *ghes_ioremap_area;
122 /*
123  * These 2 spinlock is used to prevent atomic ioremap virtual memory
124  * area from being mapped simultaneously.
125  */
126 static DEFINE_RAW_SPINLOCK(ghes_ioremap_lock_nmi);
127 static DEFINE_SPINLOCK(ghes_ioremap_lock_irq);
128 
129 static struct gen_pool *ghes_estatus_pool;
130 static unsigned long ghes_estatus_pool_size_request;
131 
132 static struct ghes_estatus_cache *ghes_estatus_caches[GHES_ESTATUS_CACHES_SIZE];
133 static atomic_t ghes_estatus_cache_alloced;
134 
135 static int ghes_ioremap_init(void)
136 {
137 	ghes_ioremap_area = __get_vm_area(PAGE_SIZE * GHES_IOREMAP_PAGES,
138 		VM_IOREMAP, VMALLOC_START, VMALLOC_END);
139 	if (!ghes_ioremap_area) {
140 		pr_err(GHES_PFX "Failed to allocate virtual memory area for atomic ioremap.\n");
141 		return -ENOMEM;
142 	}
143 
144 	return 0;
145 }
146 
147 static void ghes_ioremap_exit(void)
148 {
149 	free_vm_area(ghes_ioremap_area);
150 }
151 
152 static void __iomem *ghes_ioremap_pfn_nmi(u64 pfn)
153 {
154 	unsigned long vaddr;
155 
156 	vaddr = (unsigned long)GHES_IOREMAP_NMI_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 __iomem *ghes_ioremap_pfn_irq(u64 pfn)
164 {
165 	unsigned long vaddr, paddr;
166 	pgprot_t prot;
167 
168 	vaddr = (unsigned long)GHES_IOREMAP_IRQ_PAGE(ghes_ioremap_area->addr);
169 
170 	paddr = pfn << PAGE_SHIFT;
171 	prot = arch_apei_get_mem_attribute(paddr);
172 
173 	ioremap_page_range(vaddr, vaddr + PAGE_SIZE, paddr, prot);
174 
175 	return (void __iomem *)vaddr;
176 }
177 
178 static void ghes_iounmap_nmi(void __iomem *vaddr_ptr)
179 {
180 	unsigned long vaddr = (unsigned long __force)vaddr_ptr;
181 	void *base = ghes_ioremap_area->addr;
182 
183 	BUG_ON(vaddr != (unsigned long)GHES_IOREMAP_NMI_PAGE(base));
184 	unmap_kernel_range_noflush(vaddr, PAGE_SIZE);
185 	arch_apei_flush_tlb_one(vaddr);
186 }
187 
188 static void ghes_iounmap_irq(void __iomem *vaddr_ptr)
189 {
190 	unsigned long vaddr = (unsigned long __force)vaddr_ptr;
191 	void *base = ghes_ioremap_area->addr;
192 
193 	BUG_ON(vaddr != (unsigned long)GHES_IOREMAP_IRQ_PAGE(base));
194 	unmap_kernel_range_noflush(vaddr, PAGE_SIZE);
195 	arch_apei_flush_tlb_one(vaddr);
196 }
197 
198 static int ghes_estatus_pool_init(void)
199 {
200 	ghes_estatus_pool = gen_pool_create(GHES_ESTATUS_POOL_MIN_ALLOC_ORDER, -1);
201 	if (!ghes_estatus_pool)
202 		return -ENOMEM;
203 	return 0;
204 }
205 
206 static void ghes_estatus_pool_free_chunk_page(struct gen_pool *pool,
207 					      struct gen_pool_chunk *chunk,
208 					      void *data)
209 {
210 	free_page(chunk->start_addr);
211 }
212 
213 static void ghes_estatus_pool_exit(void)
214 {
215 	gen_pool_for_each_chunk(ghes_estatus_pool,
216 				ghes_estatus_pool_free_chunk_page, NULL);
217 	gen_pool_destroy(ghes_estatus_pool);
218 }
219 
220 static int ghes_estatus_pool_expand(unsigned long len)
221 {
222 	unsigned long i, pages, size, addr;
223 	int ret;
224 
225 	ghes_estatus_pool_size_request += PAGE_ALIGN(len);
226 	size = gen_pool_size(ghes_estatus_pool);
227 	if (size >= ghes_estatus_pool_size_request)
228 		return 0;
229 	pages = (ghes_estatus_pool_size_request - size) / PAGE_SIZE;
230 	for (i = 0; i < pages; i++) {
231 		addr = __get_free_page(GFP_KERNEL);
232 		if (!addr)
233 			return -ENOMEM;
234 		ret = gen_pool_add(ghes_estatus_pool, addr, PAGE_SIZE, -1);
235 		if (ret)
236 			return ret;
237 	}
238 
239 	return 0;
240 }
241 
242 static struct ghes *ghes_new(struct acpi_hest_generic *generic)
243 {
244 	struct ghes *ghes;
245 	unsigned int error_block_length;
246 	int rc;
247 
248 	ghes = kzalloc(sizeof(*ghes), GFP_KERNEL);
249 	if (!ghes)
250 		return ERR_PTR(-ENOMEM);
251 	ghes->generic = generic;
252 	rc = apei_map_generic_address(&generic->error_status_address);
253 	if (rc)
254 		goto err_free;
255 	error_block_length = generic->error_block_length;
256 	if (error_block_length > GHES_ESTATUS_MAX_SIZE) {
257 		pr_warning(FW_WARN GHES_PFX
258 			   "Error status block length is too long: %u for "
259 			   "generic hardware error source: %d.\n",
260 			   error_block_length, generic->header.source_id);
261 		error_block_length = GHES_ESTATUS_MAX_SIZE;
262 	}
263 	ghes->estatus = kmalloc(error_block_length, GFP_KERNEL);
264 	if (!ghes->estatus) {
265 		rc = -ENOMEM;
266 		goto err_unmap;
267 	}
268 
269 	return ghes;
270 
271 err_unmap:
272 	apei_unmap_generic_address(&generic->error_status_address);
273 err_free:
274 	kfree(ghes);
275 	return ERR_PTR(rc);
276 }
277 
278 static void ghes_fini(struct ghes *ghes)
279 {
280 	kfree(ghes->estatus);
281 	apei_unmap_generic_address(&ghes->generic->error_status_address);
282 }
283 
284 static inline int ghes_severity(int severity)
285 {
286 	switch (severity) {
287 	case CPER_SEV_INFORMATIONAL:
288 		return GHES_SEV_NO;
289 	case CPER_SEV_CORRECTED:
290 		return GHES_SEV_CORRECTED;
291 	case CPER_SEV_RECOVERABLE:
292 		return GHES_SEV_RECOVERABLE;
293 	case CPER_SEV_FATAL:
294 		return GHES_SEV_PANIC;
295 	default:
296 		/* Unknown, go panic */
297 		return GHES_SEV_PANIC;
298 	}
299 }
300 
301 static void ghes_copy_tofrom_phys(void *buffer, u64 paddr, u32 len,
302 				  int from_phys)
303 {
304 	void __iomem *vaddr;
305 	unsigned long flags = 0;
306 	int in_nmi = in_nmi();
307 	u64 offset;
308 	u32 trunk;
309 
310 	while (len > 0) {
311 		offset = paddr - (paddr & PAGE_MASK);
312 		if (in_nmi) {
313 			raw_spin_lock(&ghes_ioremap_lock_nmi);
314 			vaddr = ghes_ioremap_pfn_nmi(paddr >> PAGE_SHIFT);
315 		} else {
316 			spin_lock_irqsave(&ghes_ioremap_lock_irq, flags);
317 			vaddr = ghes_ioremap_pfn_irq(paddr >> PAGE_SHIFT);
318 		}
319 		trunk = PAGE_SIZE - offset;
320 		trunk = min(trunk, len);
321 		if (from_phys)
322 			memcpy_fromio(buffer, vaddr + offset, trunk);
323 		else
324 			memcpy_toio(vaddr + offset, buffer, trunk);
325 		len -= trunk;
326 		paddr += trunk;
327 		buffer += trunk;
328 		if (in_nmi) {
329 			ghes_iounmap_nmi(vaddr);
330 			raw_spin_unlock(&ghes_ioremap_lock_nmi);
331 		} else {
332 			ghes_iounmap_irq(vaddr);
333 			spin_unlock_irqrestore(&ghes_ioremap_lock_irq, flags);
334 		}
335 	}
336 }
337 
338 static int ghes_read_estatus(struct ghes *ghes, int silent)
339 {
340 	struct acpi_hest_generic *g = ghes->generic;
341 	u64 buf_paddr;
342 	u32 len;
343 	int rc;
344 
345 	rc = apei_read(&buf_paddr, &g->error_status_address);
346 	if (rc) {
347 		if (!silent && printk_ratelimit())
348 			pr_warning(FW_WARN GHES_PFX
349 "Failed to read error status block address for hardware error source: %d.\n",
350 				   g->header.source_id);
351 		return -EIO;
352 	}
353 	if (!buf_paddr)
354 		return -ENOENT;
355 
356 	ghes_copy_tofrom_phys(ghes->estatus, buf_paddr,
357 			      sizeof(*ghes->estatus), 1);
358 	if (!ghes->estatus->block_status)
359 		return -ENOENT;
360 
361 	ghes->buffer_paddr = buf_paddr;
362 	ghes->flags |= GHES_TO_CLEAR;
363 
364 	rc = -EIO;
365 	len = cper_estatus_len(ghes->estatus);
366 	if (len < sizeof(*ghes->estatus))
367 		goto err_read_block;
368 	if (len > ghes->generic->error_block_length)
369 		goto err_read_block;
370 	if (cper_estatus_check_header(ghes->estatus))
371 		goto err_read_block;
372 	ghes_copy_tofrom_phys(ghes->estatus + 1,
373 			      buf_paddr + sizeof(*ghes->estatus),
374 			      len - sizeof(*ghes->estatus), 1);
375 	if (cper_estatus_check(ghes->estatus))
376 		goto err_read_block;
377 	rc = 0;
378 
379 err_read_block:
380 	if (rc && !silent && printk_ratelimit())
381 		pr_warning(FW_WARN GHES_PFX
382 			   "Failed to read error status block!\n");
383 	return rc;
384 }
385 
386 static void ghes_clear_estatus(struct ghes *ghes)
387 {
388 	ghes->estatus->block_status = 0;
389 	if (!(ghes->flags & GHES_TO_CLEAR))
390 		return;
391 	ghes_copy_tofrom_phys(ghes->estatus, ghes->buffer_paddr,
392 			      sizeof(ghes->estatus->block_status), 0);
393 	ghes->flags &= ~GHES_TO_CLEAR;
394 }
395 
396 static void ghes_handle_memory_failure(struct acpi_hest_generic_data *gdata, int sev)
397 {
398 #ifdef CONFIG_ACPI_APEI_MEMORY_FAILURE
399 	unsigned long pfn;
400 	int flags = -1;
401 	int sec_sev = ghes_severity(gdata->error_severity);
402 	struct cper_sec_mem_err *mem_err;
403 	mem_err = (struct cper_sec_mem_err *)(gdata + 1);
404 
405 	if (!(mem_err->validation_bits & CPER_MEM_VALID_PA))
406 		return;
407 
408 	pfn = mem_err->physical_addr >> PAGE_SHIFT;
409 	if (!pfn_valid(pfn)) {
410 		pr_warn_ratelimited(FW_WARN GHES_PFX
411 		"Invalid address in generic error data: %#llx\n",
412 		mem_err->physical_addr);
413 		return;
414 	}
415 
416 	/* iff following two events can be handled properly by now */
417 	if (sec_sev == GHES_SEV_CORRECTED &&
418 	    (gdata->flags & CPER_SEC_ERROR_THRESHOLD_EXCEEDED))
419 		flags = MF_SOFT_OFFLINE;
420 	if (sev == GHES_SEV_RECOVERABLE && sec_sev == GHES_SEV_RECOVERABLE)
421 		flags = 0;
422 
423 	if (flags != -1)
424 		memory_failure_queue(pfn, 0, flags);
425 #endif
426 }
427 
428 static void ghes_do_proc(struct ghes *ghes,
429 			 const struct acpi_hest_generic_status *estatus)
430 {
431 	int sev, sec_sev;
432 	struct acpi_hest_generic_data *gdata;
433 
434 	sev = ghes_severity(estatus->error_severity);
435 	apei_estatus_for_each_section(estatus, gdata) {
436 		sec_sev = ghes_severity(gdata->error_severity);
437 		if (!uuid_le_cmp(*(uuid_le *)gdata->section_type,
438 				 CPER_SEC_PLATFORM_MEM)) {
439 			struct cper_sec_mem_err *mem_err;
440 			mem_err = (struct cper_sec_mem_err *)(gdata+1);
441 			ghes_edac_report_mem_error(ghes, sev, mem_err);
442 
443 			arch_apei_report_mem_error(sev, mem_err);
444 			ghes_handle_memory_failure(gdata, sev);
445 		}
446 #ifdef CONFIG_ACPI_APEI_PCIEAER
447 		else if (!uuid_le_cmp(*(uuid_le *)gdata->section_type,
448 				      CPER_SEC_PCIE)) {
449 			struct cper_sec_pcie *pcie_err;
450 			pcie_err = (struct cper_sec_pcie *)(gdata+1);
451 			if (sev == GHES_SEV_RECOVERABLE &&
452 			    sec_sev == GHES_SEV_RECOVERABLE &&
453 			    pcie_err->validation_bits & CPER_PCIE_VALID_DEVICE_ID &&
454 			    pcie_err->validation_bits & CPER_PCIE_VALID_AER_INFO) {
455 				unsigned int devfn;
456 				int aer_severity;
457 
458 				devfn = PCI_DEVFN(pcie_err->device_id.device,
459 						  pcie_err->device_id.function);
460 				aer_severity = cper_severity_to_aer(gdata->error_severity);
461 
462 				/*
463 				 * If firmware reset the component to contain
464 				 * the error, we must reinitialize it before
465 				 * use, so treat it as a fatal AER error.
466 				 */
467 				if (gdata->flags & CPER_SEC_RESET)
468 					aer_severity = AER_FATAL;
469 
470 				aer_recover_queue(pcie_err->device_id.segment,
471 						  pcie_err->device_id.bus,
472 						  devfn, aer_severity,
473 						  (struct aer_capability_regs *)
474 						  pcie_err->aer_info);
475 			}
476 
477 		}
478 #endif
479 	}
480 }
481 
482 static void __ghes_print_estatus(const char *pfx,
483 				 const struct acpi_hest_generic *generic,
484 				 const struct acpi_hest_generic_status *estatus)
485 {
486 	static atomic_t seqno;
487 	unsigned int curr_seqno;
488 	char pfx_seq[64];
489 
490 	if (pfx == NULL) {
491 		if (ghes_severity(estatus->error_severity) <=
492 		    GHES_SEV_CORRECTED)
493 			pfx = KERN_WARNING;
494 		else
495 			pfx = KERN_ERR;
496 	}
497 	curr_seqno = atomic_inc_return(&seqno);
498 	snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}" HW_ERR, pfx, curr_seqno);
499 	printk("%s""Hardware error from APEI Generic Hardware Error Source: %d\n",
500 	       pfx_seq, generic->header.source_id);
501 	cper_estatus_print(pfx_seq, estatus);
502 }
503 
504 static int ghes_print_estatus(const char *pfx,
505 			      const struct acpi_hest_generic *generic,
506 			      const struct acpi_hest_generic_status *estatus)
507 {
508 	/* Not more than 2 messages every 5 seconds */
509 	static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2);
510 	static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2);
511 	struct ratelimit_state *ratelimit;
512 
513 	if (ghes_severity(estatus->error_severity) <= GHES_SEV_CORRECTED)
514 		ratelimit = &ratelimit_corrected;
515 	else
516 		ratelimit = &ratelimit_uncorrected;
517 	if (__ratelimit(ratelimit)) {
518 		__ghes_print_estatus(pfx, generic, estatus);
519 		return 1;
520 	}
521 	return 0;
522 }
523 
524 /*
525  * GHES error status reporting throttle, to report more kinds of
526  * errors, instead of just most frequently occurred errors.
527  */
528 static int ghes_estatus_cached(struct acpi_hest_generic_status *estatus)
529 {
530 	u32 len;
531 	int i, cached = 0;
532 	unsigned long long now;
533 	struct ghes_estatus_cache *cache;
534 	struct acpi_hest_generic_status *cache_estatus;
535 
536 	len = cper_estatus_len(estatus);
537 	rcu_read_lock();
538 	for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
539 		cache = rcu_dereference(ghes_estatus_caches[i]);
540 		if (cache == NULL)
541 			continue;
542 		if (len != cache->estatus_len)
543 			continue;
544 		cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
545 		if (memcmp(estatus, cache_estatus, len))
546 			continue;
547 		atomic_inc(&cache->count);
548 		now = sched_clock();
549 		if (now - cache->time_in < GHES_ESTATUS_IN_CACHE_MAX_NSEC)
550 			cached = 1;
551 		break;
552 	}
553 	rcu_read_unlock();
554 	return cached;
555 }
556 
557 static struct ghes_estatus_cache *ghes_estatus_cache_alloc(
558 	struct acpi_hest_generic *generic,
559 	struct acpi_hest_generic_status *estatus)
560 {
561 	int alloced;
562 	u32 len, cache_len;
563 	struct ghes_estatus_cache *cache;
564 	struct acpi_hest_generic_status *cache_estatus;
565 
566 	alloced = atomic_add_return(1, &ghes_estatus_cache_alloced);
567 	if (alloced > GHES_ESTATUS_CACHE_ALLOCED_MAX) {
568 		atomic_dec(&ghes_estatus_cache_alloced);
569 		return NULL;
570 	}
571 	len = cper_estatus_len(estatus);
572 	cache_len = GHES_ESTATUS_CACHE_LEN(len);
573 	cache = (void *)gen_pool_alloc(ghes_estatus_pool, cache_len);
574 	if (!cache) {
575 		atomic_dec(&ghes_estatus_cache_alloced);
576 		return NULL;
577 	}
578 	cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
579 	memcpy(cache_estatus, estatus, len);
580 	cache->estatus_len = len;
581 	atomic_set(&cache->count, 0);
582 	cache->generic = generic;
583 	cache->time_in = sched_clock();
584 	return cache;
585 }
586 
587 static void ghes_estatus_cache_free(struct ghes_estatus_cache *cache)
588 {
589 	u32 len;
590 
591 	len = cper_estatus_len(GHES_ESTATUS_FROM_CACHE(cache));
592 	len = GHES_ESTATUS_CACHE_LEN(len);
593 	gen_pool_free(ghes_estatus_pool, (unsigned long)cache, len);
594 	atomic_dec(&ghes_estatus_cache_alloced);
595 }
596 
597 static void ghes_estatus_cache_rcu_free(struct rcu_head *head)
598 {
599 	struct ghes_estatus_cache *cache;
600 
601 	cache = container_of(head, struct ghes_estatus_cache, rcu);
602 	ghes_estatus_cache_free(cache);
603 }
604 
605 static void ghes_estatus_cache_add(
606 	struct acpi_hest_generic *generic,
607 	struct acpi_hest_generic_status *estatus)
608 {
609 	int i, slot = -1, count;
610 	unsigned long long now, duration, period, max_period = 0;
611 	struct ghes_estatus_cache *cache, *slot_cache = NULL, *new_cache;
612 
613 	new_cache = ghes_estatus_cache_alloc(generic, estatus);
614 	if (new_cache == NULL)
615 		return;
616 	rcu_read_lock();
617 	now = sched_clock();
618 	for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
619 		cache = rcu_dereference(ghes_estatus_caches[i]);
620 		if (cache == NULL) {
621 			slot = i;
622 			slot_cache = NULL;
623 			break;
624 		}
625 		duration = now - cache->time_in;
626 		if (duration >= GHES_ESTATUS_IN_CACHE_MAX_NSEC) {
627 			slot = i;
628 			slot_cache = cache;
629 			break;
630 		}
631 		count = atomic_read(&cache->count);
632 		period = duration;
633 		do_div(period, (count + 1));
634 		if (period > max_period) {
635 			max_period = period;
636 			slot = i;
637 			slot_cache = cache;
638 		}
639 	}
640 	/* new_cache must be put into array after its contents are written */
641 	smp_wmb();
642 	if (slot != -1 && cmpxchg(ghes_estatus_caches + slot,
643 				  slot_cache, new_cache) == slot_cache) {
644 		if (slot_cache)
645 			call_rcu(&slot_cache->rcu, ghes_estatus_cache_rcu_free);
646 	} else
647 		ghes_estatus_cache_free(new_cache);
648 	rcu_read_unlock();
649 }
650 
651 static int ghes_proc(struct ghes *ghes)
652 {
653 	int rc;
654 
655 	rc = ghes_read_estatus(ghes, 0);
656 	if (rc)
657 		goto out;
658 	if (!ghes_estatus_cached(ghes->estatus)) {
659 		if (ghes_print_estatus(NULL, ghes->generic, ghes->estatus))
660 			ghes_estatus_cache_add(ghes->generic, ghes->estatus);
661 	}
662 	ghes_do_proc(ghes, ghes->estatus);
663 out:
664 	ghes_clear_estatus(ghes);
665 	return 0;
666 }
667 
668 static void ghes_add_timer(struct ghes *ghes)
669 {
670 	struct acpi_hest_generic *g = ghes->generic;
671 	unsigned long expire;
672 
673 	if (!g->notify.poll_interval) {
674 		pr_warning(FW_WARN GHES_PFX "Poll interval is 0 for generic hardware error source: %d, disabled.\n",
675 			   g->header.source_id);
676 		return;
677 	}
678 	expire = jiffies + msecs_to_jiffies(g->notify.poll_interval);
679 	ghes->timer.expires = round_jiffies_relative(expire);
680 	add_timer(&ghes->timer);
681 }
682 
683 static void ghes_poll_func(unsigned long data)
684 {
685 	struct ghes *ghes = (void *)data;
686 
687 	ghes_proc(ghes);
688 	if (!(ghes->flags & GHES_EXITING))
689 		ghes_add_timer(ghes);
690 }
691 
692 static irqreturn_t ghes_irq_func(int irq, void *data)
693 {
694 	struct ghes *ghes = data;
695 	int rc;
696 
697 	rc = ghes_proc(ghes);
698 	if (rc)
699 		return IRQ_NONE;
700 
701 	return IRQ_HANDLED;
702 }
703 
704 static int ghes_notify_sci(struct notifier_block *this,
705 				  unsigned long event, void *data)
706 {
707 	struct ghes *ghes;
708 	int ret = NOTIFY_DONE;
709 
710 	rcu_read_lock();
711 	list_for_each_entry_rcu(ghes, &ghes_sci, list) {
712 		if (!ghes_proc(ghes))
713 			ret = NOTIFY_OK;
714 	}
715 	rcu_read_unlock();
716 
717 	return ret;
718 }
719 
720 static struct notifier_block ghes_notifier_sci = {
721 	.notifier_call = ghes_notify_sci,
722 };
723 
724 #ifdef CONFIG_HAVE_ACPI_APEI_NMI
725 /*
726  * printk is not safe in NMI context.  So in NMI handler, we allocate
727  * required memory from lock-less memory allocator
728  * (ghes_estatus_pool), save estatus into it, put them into lock-less
729  * list (ghes_estatus_llist), then delay printk into IRQ context via
730  * irq_work (ghes_proc_irq_work).  ghes_estatus_size_request record
731  * required pool size by all NMI error source.
732  */
733 static struct llist_head ghes_estatus_llist;
734 static struct irq_work ghes_proc_irq_work;
735 
736 /*
737  * NMI may be triggered on any CPU, so ghes_in_nmi is used for
738  * having only one concurrent reader.
739  */
740 static atomic_t ghes_in_nmi = ATOMIC_INIT(0);
741 
742 static LIST_HEAD(ghes_nmi);
743 
744 static int ghes_panic_timeout	__read_mostly = 30;
745 
746 static void ghes_proc_in_irq(struct irq_work *irq_work)
747 {
748 	struct llist_node *llnode, *next;
749 	struct ghes_estatus_node *estatus_node;
750 	struct acpi_hest_generic *generic;
751 	struct acpi_hest_generic_status *estatus;
752 	u32 len, node_len;
753 
754 	llnode = llist_del_all(&ghes_estatus_llist);
755 	/*
756 	 * Because the time order of estatus in list is reversed,
757 	 * revert it back to proper order.
758 	 */
759 	llnode = llist_reverse_order(llnode);
760 	while (llnode) {
761 		next = llnode->next;
762 		estatus_node = llist_entry(llnode, struct ghes_estatus_node,
763 					   llnode);
764 		estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
765 		len = cper_estatus_len(estatus);
766 		node_len = GHES_ESTATUS_NODE_LEN(len);
767 		ghes_do_proc(estatus_node->ghes, estatus);
768 		if (!ghes_estatus_cached(estatus)) {
769 			generic = estatus_node->generic;
770 			if (ghes_print_estatus(NULL, generic, estatus))
771 				ghes_estatus_cache_add(generic, estatus);
772 		}
773 		gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node,
774 			      node_len);
775 		llnode = next;
776 	}
777 }
778 
779 static void ghes_print_queued_estatus(void)
780 {
781 	struct llist_node *llnode;
782 	struct ghes_estatus_node *estatus_node;
783 	struct acpi_hest_generic *generic;
784 	struct acpi_hest_generic_status *estatus;
785 	u32 len, node_len;
786 
787 	llnode = llist_del_all(&ghes_estatus_llist);
788 	/*
789 	 * Because the time order of estatus in list is reversed,
790 	 * revert it back to proper order.
791 	 */
792 	llnode = llist_reverse_order(llnode);
793 	while (llnode) {
794 		estatus_node = llist_entry(llnode, struct ghes_estatus_node,
795 					   llnode);
796 		estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
797 		len = cper_estatus_len(estatus);
798 		node_len = GHES_ESTATUS_NODE_LEN(len);
799 		generic = estatus_node->generic;
800 		ghes_print_estatus(NULL, generic, estatus);
801 		llnode = llnode->next;
802 	}
803 }
804 
805 /* Save estatus for further processing in IRQ context */
806 static void __process_error(struct ghes *ghes)
807 {
808 #ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
809 	u32 len, node_len;
810 	struct ghes_estatus_node *estatus_node;
811 	struct acpi_hest_generic_status *estatus;
812 
813 	if (ghes_estatus_cached(ghes->estatus))
814 		return;
815 
816 	len = cper_estatus_len(ghes->estatus);
817 	node_len = GHES_ESTATUS_NODE_LEN(len);
818 
819 	estatus_node = (void *)gen_pool_alloc(ghes_estatus_pool, node_len);
820 	if (!estatus_node)
821 		return;
822 
823 	estatus_node->ghes = ghes;
824 	estatus_node->generic = ghes->generic;
825 	estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
826 	memcpy(estatus, ghes->estatus, len);
827 	llist_add(&estatus_node->llnode, &ghes_estatus_llist);
828 #endif
829 }
830 
831 static void __ghes_panic(struct ghes *ghes)
832 {
833 	oops_begin();
834 	ghes_print_queued_estatus();
835 	__ghes_print_estatus(KERN_EMERG, ghes->generic, ghes->estatus);
836 
837 	/* reboot to log the error! */
838 	if (panic_timeout == 0)
839 		panic_timeout = ghes_panic_timeout;
840 	panic("Fatal hardware error!");
841 }
842 
843 static int ghes_notify_nmi(unsigned int cmd, struct pt_regs *regs)
844 {
845 	struct ghes *ghes;
846 	int sev, ret = NMI_DONE;
847 
848 	if (!atomic_add_unless(&ghes_in_nmi, 1, 1))
849 		return ret;
850 
851 	list_for_each_entry_rcu(ghes, &ghes_nmi, list) {
852 		if (ghes_read_estatus(ghes, 1)) {
853 			ghes_clear_estatus(ghes);
854 			continue;
855 		}
856 
857 		sev = ghes_severity(ghes->estatus->error_severity);
858 		if (sev >= GHES_SEV_PANIC)
859 			__ghes_panic(ghes);
860 
861 		if (!(ghes->flags & GHES_TO_CLEAR))
862 			continue;
863 
864 		__process_error(ghes);
865 		ghes_clear_estatus(ghes);
866 
867 		ret = NMI_HANDLED;
868 	}
869 
870 #ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
871 	irq_work_queue(&ghes_proc_irq_work);
872 #endif
873 	atomic_dec(&ghes_in_nmi);
874 	return ret;
875 }
876 
877 static unsigned long ghes_esource_prealloc_size(
878 	const struct acpi_hest_generic *generic)
879 {
880 	unsigned long block_length, prealloc_records, prealloc_size;
881 
882 	block_length = min_t(unsigned long, generic->error_block_length,
883 			     GHES_ESTATUS_MAX_SIZE);
884 	prealloc_records = max_t(unsigned long,
885 				 generic->records_to_preallocate, 1);
886 	prealloc_size = min_t(unsigned long, block_length * prealloc_records,
887 			      GHES_ESOURCE_PREALLOC_MAX_SIZE);
888 
889 	return prealloc_size;
890 }
891 
892 static void ghes_estatus_pool_shrink(unsigned long len)
893 {
894 	ghes_estatus_pool_size_request -= PAGE_ALIGN(len);
895 }
896 
897 static void ghes_nmi_add(struct ghes *ghes)
898 {
899 	unsigned long len;
900 
901 	len = ghes_esource_prealloc_size(ghes->generic);
902 	ghes_estatus_pool_expand(len);
903 	mutex_lock(&ghes_list_mutex);
904 	if (list_empty(&ghes_nmi))
905 		register_nmi_handler(NMI_LOCAL, ghes_notify_nmi, 0, "ghes");
906 	list_add_rcu(&ghes->list, &ghes_nmi);
907 	mutex_unlock(&ghes_list_mutex);
908 }
909 
910 static void ghes_nmi_remove(struct ghes *ghes)
911 {
912 	unsigned long len;
913 
914 	mutex_lock(&ghes_list_mutex);
915 	list_del_rcu(&ghes->list);
916 	if (list_empty(&ghes_nmi))
917 		unregister_nmi_handler(NMI_LOCAL, "ghes");
918 	mutex_unlock(&ghes_list_mutex);
919 	/*
920 	 * To synchronize with NMI handler, ghes can only be
921 	 * freed after NMI handler finishes.
922 	 */
923 	synchronize_rcu();
924 	len = ghes_esource_prealloc_size(ghes->generic);
925 	ghes_estatus_pool_shrink(len);
926 }
927 
928 static void ghes_nmi_init_cxt(void)
929 {
930 	init_irq_work(&ghes_proc_irq_work, ghes_proc_in_irq);
931 }
932 #else /* CONFIG_HAVE_ACPI_APEI_NMI */
933 static inline void ghes_nmi_add(struct ghes *ghes)
934 {
935 	pr_err(GHES_PFX "ID: %d, trying to add NMI notification which is not supported!\n",
936 	       ghes->generic->header.source_id);
937 	BUG();
938 }
939 
940 static inline void ghes_nmi_remove(struct ghes *ghes)
941 {
942 	pr_err(GHES_PFX "ID: %d, trying to remove NMI notification which is not supported!\n",
943 	       ghes->generic->header.source_id);
944 	BUG();
945 }
946 
947 static inline void ghes_nmi_init_cxt(void)
948 {
949 }
950 #endif /* CONFIG_HAVE_ACPI_APEI_NMI */
951 
952 static int ghes_probe(struct platform_device *ghes_dev)
953 {
954 	struct acpi_hest_generic *generic;
955 	struct ghes *ghes = NULL;
956 
957 	int rc = -EINVAL;
958 
959 	generic = *(struct acpi_hest_generic **)ghes_dev->dev.platform_data;
960 	if (!generic->enabled)
961 		return -ENODEV;
962 
963 	switch (generic->notify.type) {
964 	case ACPI_HEST_NOTIFY_POLLED:
965 	case ACPI_HEST_NOTIFY_EXTERNAL:
966 	case ACPI_HEST_NOTIFY_SCI:
967 		break;
968 	case ACPI_HEST_NOTIFY_NMI:
969 		if (!IS_ENABLED(CONFIG_HAVE_ACPI_APEI_NMI)) {
970 			pr_warn(GHES_PFX "Generic hardware error source: %d notified via NMI interrupt is not supported!\n",
971 				generic->header.source_id);
972 			goto err;
973 		}
974 		break;
975 	case ACPI_HEST_NOTIFY_LOCAL:
976 		pr_warning(GHES_PFX "Generic hardware error source: %d notified via local interrupt is not supported!\n",
977 			   generic->header.source_id);
978 		goto err;
979 	default:
980 		pr_warning(FW_WARN GHES_PFX "Unknown notification type: %u for generic hardware error source: %d\n",
981 			   generic->notify.type, generic->header.source_id);
982 		goto err;
983 	}
984 
985 	rc = -EIO;
986 	if (generic->error_block_length <
987 	    sizeof(struct acpi_hest_generic_status)) {
988 		pr_warning(FW_BUG GHES_PFX "Invalid error block length: %u for generic hardware error source: %d\n",
989 			   generic->error_block_length,
990 			   generic->header.source_id);
991 		goto err;
992 	}
993 	ghes = ghes_new(generic);
994 	if (IS_ERR(ghes)) {
995 		rc = PTR_ERR(ghes);
996 		ghes = NULL;
997 		goto err;
998 	}
999 
1000 	rc = ghes_edac_register(ghes, &ghes_dev->dev);
1001 	if (rc < 0)
1002 		goto err;
1003 
1004 	switch (generic->notify.type) {
1005 	case ACPI_HEST_NOTIFY_POLLED:
1006 		ghes->timer.function = ghes_poll_func;
1007 		ghes->timer.data = (unsigned long)ghes;
1008 		init_timer_deferrable(&ghes->timer);
1009 		ghes_add_timer(ghes);
1010 		break;
1011 	case ACPI_HEST_NOTIFY_EXTERNAL:
1012 		/* External interrupt vector is GSI */
1013 		rc = acpi_gsi_to_irq(generic->notify.vector, &ghes->irq);
1014 		if (rc) {
1015 			pr_err(GHES_PFX "Failed to map GSI to IRQ for generic hardware error source: %d\n",
1016 			       generic->header.source_id);
1017 			goto err_edac_unreg;
1018 		}
1019 		rc = request_irq(ghes->irq, ghes_irq_func, 0, "GHES IRQ", ghes);
1020 		if (rc) {
1021 			pr_err(GHES_PFX "Failed to register IRQ for generic hardware error source: %d\n",
1022 			       generic->header.source_id);
1023 			goto err_edac_unreg;
1024 		}
1025 		break;
1026 	case ACPI_HEST_NOTIFY_SCI:
1027 		mutex_lock(&ghes_list_mutex);
1028 		if (list_empty(&ghes_sci))
1029 			register_acpi_hed_notifier(&ghes_notifier_sci);
1030 		list_add_rcu(&ghes->list, &ghes_sci);
1031 		mutex_unlock(&ghes_list_mutex);
1032 		break;
1033 	case ACPI_HEST_NOTIFY_NMI:
1034 		ghes_nmi_add(ghes);
1035 		break;
1036 	default:
1037 		BUG();
1038 	}
1039 	platform_set_drvdata(ghes_dev, ghes);
1040 
1041 	return 0;
1042 err_edac_unreg:
1043 	ghes_edac_unregister(ghes);
1044 err:
1045 	if (ghes) {
1046 		ghes_fini(ghes);
1047 		kfree(ghes);
1048 	}
1049 	return rc;
1050 }
1051 
1052 static int ghes_remove(struct platform_device *ghes_dev)
1053 {
1054 	struct ghes *ghes;
1055 	struct acpi_hest_generic *generic;
1056 
1057 	ghes = platform_get_drvdata(ghes_dev);
1058 	generic = ghes->generic;
1059 
1060 	ghes->flags |= GHES_EXITING;
1061 	switch (generic->notify.type) {
1062 	case ACPI_HEST_NOTIFY_POLLED:
1063 		del_timer_sync(&ghes->timer);
1064 		break;
1065 	case ACPI_HEST_NOTIFY_EXTERNAL:
1066 		free_irq(ghes->irq, ghes);
1067 		break;
1068 	case ACPI_HEST_NOTIFY_SCI:
1069 		mutex_lock(&ghes_list_mutex);
1070 		list_del_rcu(&ghes->list);
1071 		if (list_empty(&ghes_sci))
1072 			unregister_acpi_hed_notifier(&ghes_notifier_sci);
1073 		mutex_unlock(&ghes_list_mutex);
1074 		break;
1075 	case ACPI_HEST_NOTIFY_NMI:
1076 		ghes_nmi_remove(ghes);
1077 		break;
1078 	default:
1079 		BUG();
1080 		break;
1081 	}
1082 
1083 	ghes_fini(ghes);
1084 
1085 	ghes_edac_unregister(ghes);
1086 
1087 	kfree(ghes);
1088 
1089 	platform_set_drvdata(ghes_dev, NULL);
1090 
1091 	return 0;
1092 }
1093 
1094 static struct platform_driver ghes_platform_driver = {
1095 	.driver		= {
1096 		.name	= "GHES",
1097 	},
1098 	.probe		= ghes_probe,
1099 	.remove		= ghes_remove,
1100 };
1101 
1102 static int __init ghes_init(void)
1103 {
1104 	int rc;
1105 
1106 	if (acpi_disabled)
1107 		return -ENODEV;
1108 
1109 	if (hest_disable) {
1110 		pr_info(GHES_PFX "HEST is not enabled!\n");
1111 		return -EINVAL;
1112 	}
1113 
1114 	if (ghes_disable) {
1115 		pr_info(GHES_PFX "GHES is not enabled!\n");
1116 		return -EINVAL;
1117 	}
1118 
1119 	ghes_nmi_init_cxt();
1120 
1121 	rc = ghes_ioremap_init();
1122 	if (rc)
1123 		goto err;
1124 
1125 	rc = ghes_estatus_pool_init();
1126 	if (rc)
1127 		goto err_ioremap_exit;
1128 
1129 	rc = ghes_estatus_pool_expand(GHES_ESTATUS_CACHE_AVG_SIZE *
1130 				      GHES_ESTATUS_CACHE_ALLOCED_MAX);
1131 	if (rc)
1132 		goto err_pool_exit;
1133 
1134 	rc = platform_driver_register(&ghes_platform_driver);
1135 	if (rc)
1136 		goto err_pool_exit;
1137 
1138 	rc = apei_osc_setup();
1139 	if (rc == 0 && osc_sb_apei_support_acked)
1140 		pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit and WHEA _OSC.\n");
1141 	else if (rc == 0 && !osc_sb_apei_support_acked)
1142 		pr_info(GHES_PFX "APEI firmware first mode is enabled by WHEA _OSC.\n");
1143 	else if (rc && osc_sb_apei_support_acked)
1144 		pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit.\n");
1145 	else
1146 		pr_info(GHES_PFX "Failed to enable APEI firmware first mode.\n");
1147 
1148 	return 0;
1149 err_pool_exit:
1150 	ghes_estatus_pool_exit();
1151 err_ioremap_exit:
1152 	ghes_ioremap_exit();
1153 err:
1154 	return rc;
1155 }
1156 device_initcall(ghes_init);
1157