xref: /openbmc/linux/drivers/acpi/apei/ghes.c (revision 9d749629)
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 #include <acpi/apei.h>
52 #include <acpi/hed.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 /*
88  * One struct ghes is created for each generic hardware error source.
89  * It provides the context for APEI hardware error timer/IRQ/SCI/NMI
90  * handler.
91  *
92  * estatus: memory buffer for error status block, allocated during
93  * HEST parsing.
94  */
95 #define GHES_TO_CLEAR		0x0001
96 #define GHES_EXITING		0x0002
97 
98 struct ghes {
99 	struct acpi_hest_generic *generic;
100 	struct acpi_hest_generic_status *estatus;
101 	u64 buffer_paddr;
102 	unsigned long flags;
103 	union {
104 		struct list_head list;
105 		struct timer_list timer;
106 		unsigned int irq;
107 	};
108 };
109 
110 struct ghes_estatus_node {
111 	struct llist_node llnode;
112 	struct acpi_hest_generic *generic;
113 };
114 
115 struct ghes_estatus_cache {
116 	u32 estatus_len;
117 	atomic_t count;
118 	struct acpi_hest_generic *generic;
119 	unsigned long long time_in;
120 	struct rcu_head rcu;
121 };
122 
123 bool ghes_disable;
124 module_param_named(disable, ghes_disable, bool, 0);
125 
126 static int ghes_panic_timeout	__read_mostly = 30;
127 
128 /*
129  * All error sources notified with SCI shares one notifier function,
130  * so they need to be linked and checked one by one.  This is applied
131  * to NMI too.
132  *
133  * RCU is used for these lists, so ghes_list_mutex is only used for
134  * list changing, not for traversing.
135  */
136 static LIST_HEAD(ghes_sci);
137 static LIST_HEAD(ghes_nmi);
138 static DEFINE_MUTEX(ghes_list_mutex);
139 
140 /*
141  * NMI may be triggered on any CPU, so ghes_nmi_lock is used for
142  * mutual exclusion.
143  */
144 static DEFINE_RAW_SPINLOCK(ghes_nmi_lock);
145 
146 /*
147  * Because the memory area used to transfer hardware error information
148  * from BIOS to Linux can be determined only in NMI, IRQ or timer
149  * handler, but general ioremap can not be used in atomic context, so
150  * a special version of atomic ioremap is implemented for that.
151  */
152 
153 /*
154  * Two virtual pages are used, one for NMI context, the other for
155  * IRQ/PROCESS context
156  */
157 #define GHES_IOREMAP_PAGES		2
158 #define GHES_IOREMAP_NMI_PAGE(base)	(base)
159 #define GHES_IOREMAP_IRQ_PAGE(base)	((base) + PAGE_SIZE)
160 
161 /* virtual memory area for atomic ioremap */
162 static struct vm_struct *ghes_ioremap_area;
163 /*
164  * These 2 spinlock is used to prevent atomic ioremap virtual memory
165  * area from being mapped simultaneously.
166  */
167 static DEFINE_RAW_SPINLOCK(ghes_ioremap_lock_nmi);
168 static DEFINE_SPINLOCK(ghes_ioremap_lock_irq);
169 
170 /*
171  * printk is not safe in NMI context.  So in NMI handler, we allocate
172  * required memory from lock-less memory allocator
173  * (ghes_estatus_pool), save estatus into it, put them into lock-less
174  * list (ghes_estatus_llist), then delay printk into IRQ context via
175  * irq_work (ghes_proc_irq_work).  ghes_estatus_size_request record
176  * required pool size by all NMI error source.
177  */
178 static struct gen_pool *ghes_estatus_pool;
179 static unsigned long ghes_estatus_pool_size_request;
180 static struct llist_head ghes_estatus_llist;
181 static struct irq_work ghes_proc_irq_work;
182 
183 struct ghes_estatus_cache *ghes_estatus_caches[GHES_ESTATUS_CACHES_SIZE];
184 static atomic_t ghes_estatus_cache_alloced;
185 
186 static int ghes_ioremap_init(void)
187 {
188 	ghes_ioremap_area = __get_vm_area(PAGE_SIZE * GHES_IOREMAP_PAGES,
189 		VM_IOREMAP, VMALLOC_START, VMALLOC_END);
190 	if (!ghes_ioremap_area) {
191 		pr_err(GHES_PFX "Failed to allocate virtual memory area for atomic ioremap.\n");
192 		return -ENOMEM;
193 	}
194 
195 	return 0;
196 }
197 
198 static void ghes_ioremap_exit(void)
199 {
200 	free_vm_area(ghes_ioremap_area);
201 }
202 
203 static void __iomem *ghes_ioremap_pfn_nmi(u64 pfn)
204 {
205 	unsigned long vaddr;
206 
207 	vaddr = (unsigned long)GHES_IOREMAP_NMI_PAGE(ghes_ioremap_area->addr);
208 	ioremap_page_range(vaddr, vaddr + PAGE_SIZE,
209 			   pfn << PAGE_SHIFT, PAGE_KERNEL);
210 
211 	return (void __iomem *)vaddr;
212 }
213 
214 static void __iomem *ghes_ioremap_pfn_irq(u64 pfn)
215 {
216 	unsigned long vaddr;
217 
218 	vaddr = (unsigned long)GHES_IOREMAP_IRQ_PAGE(ghes_ioremap_area->addr);
219 	ioremap_page_range(vaddr, vaddr + PAGE_SIZE,
220 			   pfn << PAGE_SHIFT, PAGE_KERNEL);
221 
222 	return (void __iomem *)vaddr;
223 }
224 
225 static void ghes_iounmap_nmi(void __iomem *vaddr_ptr)
226 {
227 	unsigned long vaddr = (unsigned long __force)vaddr_ptr;
228 	void *base = ghes_ioremap_area->addr;
229 
230 	BUG_ON(vaddr != (unsigned long)GHES_IOREMAP_NMI_PAGE(base));
231 	unmap_kernel_range_noflush(vaddr, PAGE_SIZE);
232 	__flush_tlb_one(vaddr);
233 }
234 
235 static void ghes_iounmap_irq(void __iomem *vaddr_ptr)
236 {
237 	unsigned long vaddr = (unsigned long __force)vaddr_ptr;
238 	void *base = ghes_ioremap_area->addr;
239 
240 	BUG_ON(vaddr != (unsigned long)GHES_IOREMAP_IRQ_PAGE(base));
241 	unmap_kernel_range_noflush(vaddr, PAGE_SIZE);
242 	__flush_tlb_one(vaddr);
243 }
244 
245 static int ghes_estatus_pool_init(void)
246 {
247 	ghes_estatus_pool = gen_pool_create(GHES_ESTATUS_POOL_MIN_ALLOC_ORDER, -1);
248 	if (!ghes_estatus_pool)
249 		return -ENOMEM;
250 	return 0;
251 }
252 
253 static void ghes_estatus_pool_free_chunk_page(struct gen_pool *pool,
254 					      struct gen_pool_chunk *chunk,
255 					      void *data)
256 {
257 	free_page(chunk->start_addr);
258 }
259 
260 static void ghes_estatus_pool_exit(void)
261 {
262 	gen_pool_for_each_chunk(ghes_estatus_pool,
263 				ghes_estatus_pool_free_chunk_page, NULL);
264 	gen_pool_destroy(ghes_estatus_pool);
265 }
266 
267 static int ghes_estatus_pool_expand(unsigned long len)
268 {
269 	unsigned long i, pages, size, addr;
270 	int ret;
271 
272 	ghes_estatus_pool_size_request += PAGE_ALIGN(len);
273 	size = gen_pool_size(ghes_estatus_pool);
274 	if (size >= ghes_estatus_pool_size_request)
275 		return 0;
276 	pages = (ghes_estatus_pool_size_request - size) / PAGE_SIZE;
277 	for (i = 0; i < pages; i++) {
278 		addr = __get_free_page(GFP_KERNEL);
279 		if (!addr)
280 			return -ENOMEM;
281 		ret = gen_pool_add(ghes_estatus_pool, addr, PAGE_SIZE, -1);
282 		if (ret)
283 			return ret;
284 	}
285 
286 	return 0;
287 }
288 
289 static void ghes_estatus_pool_shrink(unsigned long len)
290 {
291 	ghes_estatus_pool_size_request -= PAGE_ALIGN(len);
292 }
293 
294 static struct ghes *ghes_new(struct acpi_hest_generic *generic)
295 {
296 	struct ghes *ghes;
297 	unsigned int error_block_length;
298 	int rc;
299 
300 	ghes = kzalloc(sizeof(*ghes), GFP_KERNEL);
301 	if (!ghes)
302 		return ERR_PTR(-ENOMEM);
303 	ghes->generic = generic;
304 	rc = apei_map_generic_address(&generic->error_status_address);
305 	if (rc)
306 		goto err_free;
307 	error_block_length = generic->error_block_length;
308 	if (error_block_length > GHES_ESTATUS_MAX_SIZE) {
309 		pr_warning(FW_WARN GHES_PFX
310 			   "Error status block length is too long: %u for "
311 			   "generic hardware error source: %d.\n",
312 			   error_block_length, generic->header.source_id);
313 		error_block_length = GHES_ESTATUS_MAX_SIZE;
314 	}
315 	ghes->estatus = kmalloc(error_block_length, GFP_KERNEL);
316 	if (!ghes->estatus) {
317 		rc = -ENOMEM;
318 		goto err_unmap;
319 	}
320 
321 	return ghes;
322 
323 err_unmap:
324 	apei_unmap_generic_address(&generic->error_status_address);
325 err_free:
326 	kfree(ghes);
327 	return ERR_PTR(rc);
328 }
329 
330 static void ghes_fini(struct ghes *ghes)
331 {
332 	kfree(ghes->estatus);
333 	apei_unmap_generic_address(&ghes->generic->error_status_address);
334 }
335 
336 enum {
337 	GHES_SEV_NO = 0x0,
338 	GHES_SEV_CORRECTED = 0x1,
339 	GHES_SEV_RECOVERABLE = 0x2,
340 	GHES_SEV_PANIC = 0x3,
341 };
342 
343 static inline int ghes_severity(int severity)
344 {
345 	switch (severity) {
346 	case CPER_SEV_INFORMATIONAL:
347 		return GHES_SEV_NO;
348 	case CPER_SEV_CORRECTED:
349 		return GHES_SEV_CORRECTED;
350 	case CPER_SEV_RECOVERABLE:
351 		return GHES_SEV_RECOVERABLE;
352 	case CPER_SEV_FATAL:
353 		return GHES_SEV_PANIC;
354 	default:
355 		/* Unknown, go panic */
356 		return GHES_SEV_PANIC;
357 	}
358 }
359 
360 static void ghes_copy_tofrom_phys(void *buffer, u64 paddr, u32 len,
361 				  int from_phys)
362 {
363 	void __iomem *vaddr;
364 	unsigned long flags = 0;
365 	int in_nmi = in_nmi();
366 	u64 offset;
367 	u32 trunk;
368 
369 	while (len > 0) {
370 		offset = paddr - (paddr & PAGE_MASK);
371 		if (in_nmi) {
372 			raw_spin_lock(&ghes_ioremap_lock_nmi);
373 			vaddr = ghes_ioremap_pfn_nmi(paddr >> PAGE_SHIFT);
374 		} else {
375 			spin_lock_irqsave(&ghes_ioremap_lock_irq, flags);
376 			vaddr = ghes_ioremap_pfn_irq(paddr >> PAGE_SHIFT);
377 		}
378 		trunk = PAGE_SIZE - offset;
379 		trunk = min(trunk, len);
380 		if (from_phys)
381 			memcpy_fromio(buffer, vaddr + offset, trunk);
382 		else
383 			memcpy_toio(vaddr + offset, buffer, trunk);
384 		len -= trunk;
385 		paddr += trunk;
386 		buffer += trunk;
387 		if (in_nmi) {
388 			ghes_iounmap_nmi(vaddr);
389 			raw_spin_unlock(&ghes_ioremap_lock_nmi);
390 		} else {
391 			ghes_iounmap_irq(vaddr);
392 			spin_unlock_irqrestore(&ghes_ioremap_lock_irq, flags);
393 		}
394 	}
395 }
396 
397 static int ghes_read_estatus(struct ghes *ghes, int silent)
398 {
399 	struct acpi_hest_generic *g = ghes->generic;
400 	u64 buf_paddr;
401 	u32 len;
402 	int rc;
403 
404 	rc = apei_read(&buf_paddr, &g->error_status_address);
405 	if (rc) {
406 		if (!silent && printk_ratelimit())
407 			pr_warning(FW_WARN GHES_PFX
408 "Failed to read error status block address for hardware error source: %d.\n",
409 				   g->header.source_id);
410 		return -EIO;
411 	}
412 	if (!buf_paddr)
413 		return -ENOENT;
414 
415 	ghes_copy_tofrom_phys(ghes->estatus, buf_paddr,
416 			      sizeof(*ghes->estatus), 1);
417 	if (!ghes->estatus->block_status)
418 		return -ENOENT;
419 
420 	ghes->buffer_paddr = buf_paddr;
421 	ghes->flags |= GHES_TO_CLEAR;
422 
423 	rc = -EIO;
424 	len = apei_estatus_len(ghes->estatus);
425 	if (len < sizeof(*ghes->estatus))
426 		goto err_read_block;
427 	if (len > ghes->generic->error_block_length)
428 		goto err_read_block;
429 	if (apei_estatus_check_header(ghes->estatus))
430 		goto err_read_block;
431 	ghes_copy_tofrom_phys(ghes->estatus + 1,
432 			      buf_paddr + sizeof(*ghes->estatus),
433 			      len - sizeof(*ghes->estatus), 1);
434 	if (apei_estatus_check(ghes->estatus))
435 		goto err_read_block;
436 	rc = 0;
437 
438 err_read_block:
439 	if (rc && !silent && printk_ratelimit())
440 		pr_warning(FW_WARN GHES_PFX
441 			   "Failed to read error status block!\n");
442 	return rc;
443 }
444 
445 static void ghes_clear_estatus(struct ghes *ghes)
446 {
447 	ghes->estatus->block_status = 0;
448 	if (!(ghes->flags & GHES_TO_CLEAR))
449 		return;
450 	ghes_copy_tofrom_phys(ghes->estatus, ghes->buffer_paddr,
451 			      sizeof(ghes->estatus->block_status), 0);
452 	ghes->flags &= ~GHES_TO_CLEAR;
453 }
454 
455 static void ghes_do_proc(const struct acpi_hest_generic_status *estatus)
456 {
457 	int sev, sec_sev;
458 	struct acpi_hest_generic_data *gdata;
459 
460 	sev = ghes_severity(estatus->error_severity);
461 	apei_estatus_for_each_section(estatus, gdata) {
462 		sec_sev = ghes_severity(gdata->error_severity);
463 		if (!uuid_le_cmp(*(uuid_le *)gdata->section_type,
464 				 CPER_SEC_PLATFORM_MEM)) {
465 			struct cper_sec_mem_err *mem_err;
466 			mem_err = (struct cper_sec_mem_err *)(gdata+1);
467 #ifdef CONFIG_X86_MCE
468 			apei_mce_report_mem_error(sev == GHES_SEV_CORRECTED,
469 						  mem_err);
470 #endif
471 #ifdef CONFIG_ACPI_APEI_MEMORY_FAILURE
472 			if (sev == GHES_SEV_RECOVERABLE &&
473 			    sec_sev == GHES_SEV_RECOVERABLE &&
474 			    mem_err->validation_bits & CPER_MEM_VALID_PHYSICAL_ADDRESS) {
475 				unsigned long pfn;
476 				pfn = mem_err->physical_addr >> PAGE_SHIFT;
477 				memory_failure_queue(pfn, 0, 0);
478 			}
479 #endif
480 		}
481 #ifdef CONFIG_ACPI_APEI_PCIEAER
482 		else if (!uuid_le_cmp(*(uuid_le *)gdata->section_type,
483 				      CPER_SEC_PCIE)) {
484 			struct cper_sec_pcie *pcie_err;
485 			pcie_err = (struct cper_sec_pcie *)(gdata+1);
486 			if (sev == GHES_SEV_RECOVERABLE &&
487 			    sec_sev == GHES_SEV_RECOVERABLE &&
488 			    pcie_err->validation_bits & CPER_PCIE_VALID_DEVICE_ID &&
489 			    pcie_err->validation_bits & CPER_PCIE_VALID_AER_INFO) {
490 				unsigned int devfn;
491 				int aer_severity;
492 				devfn = PCI_DEVFN(pcie_err->device_id.device,
493 						  pcie_err->device_id.function);
494 				aer_severity = cper_severity_to_aer(sev);
495 				aer_recover_queue(pcie_err->device_id.segment,
496 						  pcie_err->device_id.bus,
497 						  devfn, aer_severity);
498 			}
499 
500 		}
501 #endif
502 	}
503 }
504 
505 static void __ghes_print_estatus(const char *pfx,
506 				 const struct acpi_hest_generic *generic,
507 				 const struct acpi_hest_generic_status *estatus)
508 {
509 	static atomic_t seqno;
510 	unsigned int curr_seqno;
511 	char pfx_seq[64];
512 
513 	if (pfx == NULL) {
514 		if (ghes_severity(estatus->error_severity) <=
515 		    GHES_SEV_CORRECTED)
516 			pfx = KERN_WARNING;
517 		else
518 			pfx = KERN_ERR;
519 	}
520 	curr_seqno = atomic_inc_return(&seqno);
521 	snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}" HW_ERR, pfx, curr_seqno);
522 	printk("%s""Hardware error from APEI Generic Hardware Error Source: %d\n",
523 	       pfx_seq, generic->header.source_id);
524 	apei_estatus_print(pfx_seq, estatus);
525 }
526 
527 static int ghes_print_estatus(const char *pfx,
528 			      const struct acpi_hest_generic *generic,
529 			      const struct acpi_hest_generic_status *estatus)
530 {
531 	/* Not more than 2 messages every 5 seconds */
532 	static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2);
533 	static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2);
534 	struct ratelimit_state *ratelimit;
535 
536 	if (ghes_severity(estatus->error_severity) <= GHES_SEV_CORRECTED)
537 		ratelimit = &ratelimit_corrected;
538 	else
539 		ratelimit = &ratelimit_uncorrected;
540 	if (__ratelimit(ratelimit)) {
541 		__ghes_print_estatus(pfx, generic, estatus);
542 		return 1;
543 	}
544 	return 0;
545 }
546 
547 /*
548  * GHES error status reporting throttle, to report more kinds of
549  * errors, instead of just most frequently occurred errors.
550  */
551 static int ghes_estatus_cached(struct acpi_hest_generic_status *estatus)
552 {
553 	u32 len;
554 	int i, cached = 0;
555 	unsigned long long now;
556 	struct ghes_estatus_cache *cache;
557 	struct acpi_hest_generic_status *cache_estatus;
558 
559 	len = apei_estatus_len(estatus);
560 	rcu_read_lock();
561 	for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
562 		cache = rcu_dereference(ghes_estatus_caches[i]);
563 		if (cache == NULL)
564 			continue;
565 		if (len != cache->estatus_len)
566 			continue;
567 		cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
568 		if (memcmp(estatus, cache_estatus, len))
569 			continue;
570 		atomic_inc(&cache->count);
571 		now = sched_clock();
572 		if (now - cache->time_in < GHES_ESTATUS_IN_CACHE_MAX_NSEC)
573 			cached = 1;
574 		break;
575 	}
576 	rcu_read_unlock();
577 	return cached;
578 }
579 
580 static struct ghes_estatus_cache *ghes_estatus_cache_alloc(
581 	struct acpi_hest_generic *generic,
582 	struct acpi_hest_generic_status *estatus)
583 {
584 	int alloced;
585 	u32 len, cache_len;
586 	struct ghes_estatus_cache *cache;
587 	struct acpi_hest_generic_status *cache_estatus;
588 
589 	alloced = atomic_add_return(1, &ghes_estatus_cache_alloced);
590 	if (alloced > GHES_ESTATUS_CACHE_ALLOCED_MAX) {
591 		atomic_dec(&ghes_estatus_cache_alloced);
592 		return NULL;
593 	}
594 	len = apei_estatus_len(estatus);
595 	cache_len = GHES_ESTATUS_CACHE_LEN(len);
596 	cache = (void *)gen_pool_alloc(ghes_estatus_pool, cache_len);
597 	if (!cache) {
598 		atomic_dec(&ghes_estatus_cache_alloced);
599 		return NULL;
600 	}
601 	cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
602 	memcpy(cache_estatus, estatus, len);
603 	cache->estatus_len = len;
604 	atomic_set(&cache->count, 0);
605 	cache->generic = generic;
606 	cache->time_in = sched_clock();
607 	return cache;
608 }
609 
610 static void ghes_estatus_cache_free(struct ghes_estatus_cache *cache)
611 {
612 	u32 len;
613 
614 	len = apei_estatus_len(GHES_ESTATUS_FROM_CACHE(cache));
615 	len = GHES_ESTATUS_CACHE_LEN(len);
616 	gen_pool_free(ghes_estatus_pool, (unsigned long)cache, len);
617 	atomic_dec(&ghes_estatus_cache_alloced);
618 }
619 
620 static void ghes_estatus_cache_rcu_free(struct rcu_head *head)
621 {
622 	struct ghes_estatus_cache *cache;
623 
624 	cache = container_of(head, struct ghes_estatus_cache, rcu);
625 	ghes_estatus_cache_free(cache);
626 }
627 
628 static void ghes_estatus_cache_add(
629 	struct acpi_hest_generic *generic,
630 	struct acpi_hest_generic_status *estatus)
631 {
632 	int i, slot = -1, count;
633 	unsigned long long now, duration, period, max_period = 0;
634 	struct ghes_estatus_cache *cache, *slot_cache = NULL, *new_cache;
635 
636 	new_cache = ghes_estatus_cache_alloc(generic, estatus);
637 	if (new_cache == NULL)
638 		return;
639 	rcu_read_lock();
640 	now = sched_clock();
641 	for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
642 		cache = rcu_dereference(ghes_estatus_caches[i]);
643 		if (cache == NULL) {
644 			slot = i;
645 			slot_cache = NULL;
646 			break;
647 		}
648 		duration = now - cache->time_in;
649 		if (duration >= GHES_ESTATUS_IN_CACHE_MAX_NSEC) {
650 			slot = i;
651 			slot_cache = cache;
652 			break;
653 		}
654 		count = atomic_read(&cache->count);
655 		period = duration;
656 		do_div(period, (count + 1));
657 		if (period > max_period) {
658 			max_period = period;
659 			slot = i;
660 			slot_cache = cache;
661 		}
662 	}
663 	/* new_cache must be put into array after its contents are written */
664 	smp_wmb();
665 	if (slot != -1 && cmpxchg(ghes_estatus_caches + slot,
666 				  slot_cache, new_cache) == slot_cache) {
667 		if (slot_cache)
668 			call_rcu(&slot_cache->rcu, ghes_estatus_cache_rcu_free);
669 	} else
670 		ghes_estatus_cache_free(new_cache);
671 	rcu_read_unlock();
672 }
673 
674 static int ghes_proc(struct ghes *ghes)
675 {
676 	int rc;
677 
678 	rc = ghes_read_estatus(ghes, 0);
679 	if (rc)
680 		goto out;
681 	if (!ghes_estatus_cached(ghes->estatus)) {
682 		if (ghes_print_estatus(NULL, ghes->generic, ghes->estatus))
683 			ghes_estatus_cache_add(ghes->generic, ghes->estatus);
684 	}
685 	ghes_do_proc(ghes->estatus);
686 out:
687 	ghes_clear_estatus(ghes);
688 	return 0;
689 }
690 
691 static void ghes_add_timer(struct ghes *ghes)
692 {
693 	struct acpi_hest_generic *g = ghes->generic;
694 	unsigned long expire;
695 
696 	if (!g->notify.poll_interval) {
697 		pr_warning(FW_WARN GHES_PFX "Poll interval is 0 for generic hardware error source: %d, disabled.\n",
698 			   g->header.source_id);
699 		return;
700 	}
701 	expire = jiffies + msecs_to_jiffies(g->notify.poll_interval);
702 	ghes->timer.expires = round_jiffies_relative(expire);
703 	add_timer(&ghes->timer);
704 }
705 
706 static void ghes_poll_func(unsigned long data)
707 {
708 	struct ghes *ghes = (void *)data;
709 
710 	ghes_proc(ghes);
711 	if (!(ghes->flags & GHES_EXITING))
712 		ghes_add_timer(ghes);
713 }
714 
715 static irqreturn_t ghes_irq_func(int irq, void *data)
716 {
717 	struct ghes *ghes = data;
718 	int rc;
719 
720 	rc = ghes_proc(ghes);
721 	if (rc)
722 		return IRQ_NONE;
723 
724 	return IRQ_HANDLED;
725 }
726 
727 static int ghes_notify_sci(struct notifier_block *this,
728 				  unsigned long event, void *data)
729 {
730 	struct ghes *ghes;
731 	int ret = NOTIFY_DONE;
732 
733 	rcu_read_lock();
734 	list_for_each_entry_rcu(ghes, &ghes_sci, list) {
735 		if (!ghes_proc(ghes))
736 			ret = NOTIFY_OK;
737 	}
738 	rcu_read_unlock();
739 
740 	return ret;
741 }
742 
743 static struct llist_node *llist_nodes_reverse(struct llist_node *llnode)
744 {
745 	struct llist_node *next, *tail = NULL;
746 
747 	while (llnode) {
748 		next = llnode->next;
749 		llnode->next = tail;
750 		tail = llnode;
751 		llnode = next;
752 	}
753 
754 	return tail;
755 }
756 
757 static void ghes_proc_in_irq(struct irq_work *irq_work)
758 {
759 	struct llist_node *llnode, *next;
760 	struct ghes_estatus_node *estatus_node;
761 	struct acpi_hest_generic *generic;
762 	struct acpi_hest_generic_status *estatus;
763 	u32 len, node_len;
764 
765 	llnode = llist_del_all(&ghes_estatus_llist);
766 	/*
767 	 * Because the time order of estatus in list is reversed,
768 	 * revert it back to proper order.
769 	 */
770 	llnode = llist_nodes_reverse(llnode);
771 	while (llnode) {
772 		next = llnode->next;
773 		estatus_node = llist_entry(llnode, struct ghes_estatus_node,
774 					   llnode);
775 		estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
776 		len = apei_estatus_len(estatus);
777 		node_len = GHES_ESTATUS_NODE_LEN(len);
778 		ghes_do_proc(estatus);
779 		if (!ghes_estatus_cached(estatus)) {
780 			generic = estatus_node->generic;
781 			if (ghes_print_estatus(NULL, generic, estatus))
782 				ghes_estatus_cache_add(generic, estatus);
783 		}
784 		gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node,
785 			      node_len);
786 		llnode = next;
787 	}
788 }
789 
790 static void ghes_print_queued_estatus(void)
791 {
792 	struct llist_node *llnode;
793 	struct ghes_estatus_node *estatus_node;
794 	struct acpi_hest_generic *generic;
795 	struct acpi_hest_generic_status *estatus;
796 	u32 len, node_len;
797 
798 	llnode = llist_del_all(&ghes_estatus_llist);
799 	/*
800 	 * Because the time order of estatus in list is reversed,
801 	 * revert it back to proper order.
802 	 */
803 	llnode = llist_nodes_reverse(llnode);
804 	while (llnode) {
805 		estatus_node = llist_entry(llnode, struct ghes_estatus_node,
806 					   llnode);
807 		estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
808 		len = apei_estatus_len(estatus);
809 		node_len = GHES_ESTATUS_NODE_LEN(len);
810 		generic = estatus_node->generic;
811 		ghes_print_estatus(NULL, generic, estatus);
812 		llnode = llnode->next;
813 	}
814 }
815 
816 static int ghes_notify_nmi(unsigned int cmd, struct pt_regs *regs)
817 {
818 	struct ghes *ghes, *ghes_global = NULL;
819 	int sev, sev_global = -1;
820 	int ret = NMI_DONE;
821 
822 	raw_spin_lock(&ghes_nmi_lock);
823 	list_for_each_entry_rcu(ghes, &ghes_nmi, list) {
824 		if (ghes_read_estatus(ghes, 1)) {
825 			ghes_clear_estatus(ghes);
826 			continue;
827 		}
828 		sev = ghes_severity(ghes->estatus->error_severity);
829 		if (sev > sev_global) {
830 			sev_global = sev;
831 			ghes_global = ghes;
832 		}
833 		ret = NMI_HANDLED;
834 	}
835 
836 	if (ret == NMI_DONE)
837 		goto out;
838 
839 	if (sev_global >= GHES_SEV_PANIC) {
840 		oops_begin();
841 		ghes_print_queued_estatus();
842 		__ghes_print_estatus(KERN_EMERG, ghes_global->generic,
843 				     ghes_global->estatus);
844 		/* reboot to log the error! */
845 		if (panic_timeout == 0)
846 			panic_timeout = ghes_panic_timeout;
847 		panic("Fatal hardware error!");
848 	}
849 
850 	list_for_each_entry_rcu(ghes, &ghes_nmi, list) {
851 #ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
852 		u32 len, node_len;
853 		struct ghes_estatus_node *estatus_node;
854 		struct acpi_hest_generic_status *estatus;
855 #endif
856 		if (!(ghes->flags & GHES_TO_CLEAR))
857 			continue;
858 #ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
859 		if (ghes_estatus_cached(ghes->estatus))
860 			goto next;
861 		/* Save estatus for further processing in IRQ context */
862 		len = apei_estatus_len(ghes->estatus);
863 		node_len = GHES_ESTATUS_NODE_LEN(len);
864 		estatus_node = (void *)gen_pool_alloc(ghes_estatus_pool,
865 						      node_len);
866 		if (estatus_node) {
867 			estatus_node->generic = ghes->generic;
868 			estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
869 			memcpy(estatus, ghes->estatus, len);
870 			llist_add(&estatus_node->llnode, &ghes_estatus_llist);
871 		}
872 next:
873 #endif
874 		ghes_clear_estatus(ghes);
875 	}
876 #ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
877 	irq_work_queue(&ghes_proc_irq_work);
878 #endif
879 
880 out:
881 	raw_spin_unlock(&ghes_nmi_lock);
882 	return ret;
883 }
884 
885 static struct notifier_block ghes_notifier_sci = {
886 	.notifier_call = ghes_notify_sci,
887 };
888 
889 static unsigned long ghes_esource_prealloc_size(
890 	const struct acpi_hest_generic *generic)
891 {
892 	unsigned long block_length, prealloc_records, prealloc_size;
893 
894 	block_length = min_t(unsigned long, generic->error_block_length,
895 			     GHES_ESTATUS_MAX_SIZE);
896 	prealloc_records = max_t(unsigned long,
897 				 generic->records_to_preallocate, 1);
898 	prealloc_size = min_t(unsigned long, block_length * prealloc_records,
899 			      GHES_ESOURCE_PREALLOC_MAX_SIZE);
900 
901 	return prealloc_size;
902 }
903 
904 static int ghes_probe(struct platform_device *ghes_dev)
905 {
906 	struct acpi_hest_generic *generic;
907 	struct ghes *ghes = NULL;
908 	unsigned long len;
909 	int rc = -EINVAL;
910 
911 	generic = *(struct acpi_hest_generic **)ghes_dev->dev.platform_data;
912 	if (!generic->enabled)
913 		return -ENODEV;
914 
915 	switch (generic->notify.type) {
916 	case ACPI_HEST_NOTIFY_POLLED:
917 	case ACPI_HEST_NOTIFY_EXTERNAL:
918 	case ACPI_HEST_NOTIFY_SCI:
919 	case ACPI_HEST_NOTIFY_NMI:
920 		break;
921 	case ACPI_HEST_NOTIFY_LOCAL:
922 		pr_warning(GHES_PFX "Generic hardware error source: %d notified via local interrupt is not supported!\n",
923 			   generic->header.source_id);
924 		goto err;
925 	default:
926 		pr_warning(FW_WARN GHES_PFX "Unknown notification type: %u for generic hardware error source: %d\n",
927 			   generic->notify.type, generic->header.source_id);
928 		goto err;
929 	}
930 
931 	rc = -EIO;
932 	if (generic->error_block_length <
933 	    sizeof(struct acpi_hest_generic_status)) {
934 		pr_warning(FW_BUG GHES_PFX "Invalid error block length: %u for generic hardware error source: %d\n",
935 			   generic->error_block_length,
936 			   generic->header.source_id);
937 		goto err;
938 	}
939 	ghes = ghes_new(generic);
940 	if (IS_ERR(ghes)) {
941 		rc = PTR_ERR(ghes);
942 		ghes = NULL;
943 		goto err;
944 	}
945 	switch (generic->notify.type) {
946 	case ACPI_HEST_NOTIFY_POLLED:
947 		ghes->timer.function = ghes_poll_func;
948 		ghes->timer.data = (unsigned long)ghes;
949 		init_timer_deferrable(&ghes->timer);
950 		ghes_add_timer(ghes);
951 		break;
952 	case ACPI_HEST_NOTIFY_EXTERNAL:
953 		/* External interrupt vector is GSI */
954 		if (acpi_gsi_to_irq(generic->notify.vector, &ghes->irq)) {
955 			pr_err(GHES_PFX "Failed to map GSI to IRQ for generic hardware error source: %d\n",
956 			       generic->header.source_id);
957 			goto err;
958 		}
959 		if (request_irq(ghes->irq, ghes_irq_func,
960 				0, "GHES IRQ", ghes)) {
961 			pr_err(GHES_PFX "Failed to register IRQ for generic hardware error source: %d\n",
962 			       generic->header.source_id);
963 			goto err;
964 		}
965 		break;
966 	case ACPI_HEST_NOTIFY_SCI:
967 		mutex_lock(&ghes_list_mutex);
968 		if (list_empty(&ghes_sci))
969 			register_acpi_hed_notifier(&ghes_notifier_sci);
970 		list_add_rcu(&ghes->list, &ghes_sci);
971 		mutex_unlock(&ghes_list_mutex);
972 		break;
973 	case ACPI_HEST_NOTIFY_NMI:
974 		len = ghes_esource_prealloc_size(generic);
975 		ghes_estatus_pool_expand(len);
976 		mutex_lock(&ghes_list_mutex);
977 		if (list_empty(&ghes_nmi))
978 			register_nmi_handler(NMI_LOCAL, ghes_notify_nmi, 0,
979 						"ghes");
980 		list_add_rcu(&ghes->list, &ghes_nmi);
981 		mutex_unlock(&ghes_list_mutex);
982 		break;
983 	default:
984 		BUG();
985 	}
986 	platform_set_drvdata(ghes_dev, ghes);
987 
988 	return 0;
989 err:
990 	if (ghes) {
991 		ghes_fini(ghes);
992 		kfree(ghes);
993 	}
994 	return rc;
995 }
996 
997 static int ghes_remove(struct platform_device *ghes_dev)
998 {
999 	struct ghes *ghes;
1000 	struct acpi_hest_generic *generic;
1001 	unsigned long len;
1002 
1003 	ghes = platform_get_drvdata(ghes_dev);
1004 	generic = ghes->generic;
1005 
1006 	ghes->flags |= GHES_EXITING;
1007 	switch (generic->notify.type) {
1008 	case ACPI_HEST_NOTIFY_POLLED:
1009 		del_timer_sync(&ghes->timer);
1010 		break;
1011 	case ACPI_HEST_NOTIFY_EXTERNAL:
1012 		free_irq(ghes->irq, ghes);
1013 		break;
1014 	case ACPI_HEST_NOTIFY_SCI:
1015 		mutex_lock(&ghes_list_mutex);
1016 		list_del_rcu(&ghes->list);
1017 		if (list_empty(&ghes_sci))
1018 			unregister_acpi_hed_notifier(&ghes_notifier_sci);
1019 		mutex_unlock(&ghes_list_mutex);
1020 		break;
1021 	case ACPI_HEST_NOTIFY_NMI:
1022 		mutex_lock(&ghes_list_mutex);
1023 		list_del_rcu(&ghes->list);
1024 		if (list_empty(&ghes_nmi))
1025 			unregister_nmi_handler(NMI_LOCAL, "ghes");
1026 		mutex_unlock(&ghes_list_mutex);
1027 		/*
1028 		 * To synchronize with NMI handler, ghes can only be
1029 		 * freed after NMI handler finishes.
1030 		 */
1031 		synchronize_rcu();
1032 		len = ghes_esource_prealloc_size(generic);
1033 		ghes_estatus_pool_shrink(len);
1034 		break;
1035 	default:
1036 		BUG();
1037 		break;
1038 	}
1039 
1040 	ghes_fini(ghes);
1041 	kfree(ghes);
1042 
1043 	platform_set_drvdata(ghes_dev, NULL);
1044 
1045 	return 0;
1046 }
1047 
1048 static struct platform_driver ghes_platform_driver = {
1049 	.driver		= {
1050 		.name	= "GHES",
1051 		.owner	= THIS_MODULE,
1052 	},
1053 	.probe		= ghes_probe,
1054 	.remove		= ghes_remove,
1055 };
1056 
1057 static int __init ghes_init(void)
1058 {
1059 	int rc;
1060 
1061 	if (acpi_disabled)
1062 		return -ENODEV;
1063 
1064 	if (hest_disable) {
1065 		pr_info(GHES_PFX "HEST is not enabled!\n");
1066 		return -EINVAL;
1067 	}
1068 
1069 	if (ghes_disable) {
1070 		pr_info(GHES_PFX "GHES is not enabled!\n");
1071 		return -EINVAL;
1072 	}
1073 
1074 	init_irq_work(&ghes_proc_irq_work, ghes_proc_in_irq);
1075 
1076 	rc = ghes_ioremap_init();
1077 	if (rc)
1078 		goto err;
1079 
1080 	rc = ghes_estatus_pool_init();
1081 	if (rc)
1082 		goto err_ioremap_exit;
1083 
1084 	rc = ghes_estatus_pool_expand(GHES_ESTATUS_CACHE_AVG_SIZE *
1085 				      GHES_ESTATUS_CACHE_ALLOCED_MAX);
1086 	if (rc)
1087 		goto err_pool_exit;
1088 
1089 	rc = platform_driver_register(&ghes_platform_driver);
1090 	if (rc)
1091 		goto err_pool_exit;
1092 
1093 	rc = apei_osc_setup();
1094 	if (rc == 0 && osc_sb_apei_support_acked)
1095 		pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit and WHEA _OSC.\n");
1096 	else if (rc == 0 && !osc_sb_apei_support_acked)
1097 		pr_info(GHES_PFX "APEI firmware first mode is enabled by WHEA _OSC.\n");
1098 	else if (rc && osc_sb_apei_support_acked)
1099 		pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit.\n");
1100 	else
1101 		pr_info(GHES_PFX "Failed to enable APEI firmware first mode.\n");
1102 
1103 	return 0;
1104 err_pool_exit:
1105 	ghes_estatus_pool_exit();
1106 err_ioremap_exit:
1107 	ghes_ioremap_exit();
1108 err:
1109 	return rc;
1110 }
1111 
1112 static void __exit ghes_exit(void)
1113 {
1114 	platform_driver_unregister(&ghes_platform_driver);
1115 	ghes_estatus_pool_exit();
1116 	ghes_ioremap_exit();
1117 }
1118 
1119 module_init(ghes_init);
1120 module_exit(ghes_exit);
1121 
1122 MODULE_AUTHOR("Huang Ying");
1123 MODULE_DESCRIPTION("APEI Generic Hardware Error Source support");
1124 MODULE_LICENSE("GPL");
1125 MODULE_ALIAS("platform:GHES");
1126