xref: /openbmc/linux/drivers/acpi/apei/ghes.c (revision bc000245)
1 /*
2  * APEI Generic Hardware Error Source support
3  *
4  * Generic Hardware Error Source provides a way to report platform
5  * hardware errors (such as that from chipset). It works in so called
6  * "Firmware First" mode, that is, hardware errors are reported to
7  * firmware firstly, then reported to Linux by firmware. This way,
8  * some non-standard hardware error registers or non-standard hardware
9  * link can be checked by firmware to produce more hardware error
10  * information for Linux.
11  *
12  * For more information about Generic Hardware Error Source, please
13  * refer to ACPI Specification version 4.0, section 17.3.2.6
14  *
15  * Copyright 2010,2011 Intel Corp.
16  *   Author: Huang Ying <ying.huang@intel.com>
17  *
18  * This program is free software; you can redistribute it and/or
19  * modify it under the terms of the GNU General Public License version
20  * 2 as published by the Free Software Foundation;
21  *
22  * This program is distributed in the hope that it will be useful,
23  * but WITHOUT ANY WARRANTY; without even the implied warranty of
24  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25  * GNU General Public License for more details.
26  *
27  * You should have received a copy of the GNU General Public License
28  * along with this program; if not, write to the Free Software
29  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
30  */
31 
32 #include <linux/kernel.h>
33 #include <linux/module.h>
34 #include <linux/init.h>
35 #include <linux/acpi.h>
36 #include <linux/acpi_io.h>
37 #include <linux/io.h>
38 #include <linux/interrupt.h>
39 #include <linux/timer.h>
40 #include <linux/cper.h>
41 #include <linux/kdebug.h>
42 #include <linux/platform_device.h>
43 #include <linux/mutex.h>
44 #include <linux/ratelimit.h>
45 #include <linux/vmalloc.h>
46 #include <linux/irq_work.h>
47 #include <linux/llist.h>
48 #include <linux/genalloc.h>
49 #include <linux/pci.h>
50 #include <linux/aer.h>
51 
52 #include <acpi/ghes.h>
53 #include <asm/mce.h>
54 #include <asm/tlbflush.h>
55 #include <asm/nmi.h>
56 
57 #include "apei-internal.h"
58 
59 #define GHES_PFX	"GHES: "
60 
61 #define GHES_ESTATUS_MAX_SIZE		65536
62 #define GHES_ESOURCE_PREALLOC_MAX_SIZE	65536
63 
64 #define GHES_ESTATUS_POOL_MIN_ALLOC_ORDER 3
65 
66 /* This is just an estimation for memory pool allocation */
67 #define GHES_ESTATUS_CACHE_AVG_SIZE	512
68 
69 #define GHES_ESTATUS_CACHES_SIZE	4
70 
71 #define GHES_ESTATUS_IN_CACHE_MAX_NSEC	10000000000ULL
72 /* Prevent too many caches are allocated because of RCU */
73 #define GHES_ESTATUS_CACHE_ALLOCED_MAX	(GHES_ESTATUS_CACHES_SIZE * 3 / 2)
74 
75 #define GHES_ESTATUS_CACHE_LEN(estatus_len)			\
76 	(sizeof(struct ghes_estatus_cache) + (estatus_len))
77 #define GHES_ESTATUS_FROM_CACHE(estatus_cache)			\
78 	((struct acpi_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_generic_status *)				\
85 	 ((struct ghes_estatus_node *)(estatus_node) + 1))
86 
87 bool ghes_disable;
88 module_param_named(disable, ghes_disable, bool, 0);
89 
90 static int ghes_panic_timeout	__read_mostly = 30;
91 
92 /*
93  * All error sources notified with SCI shares one notifier function,
94  * so they need to be linked and checked one by one.  This is applied
95  * to NMI too.
96  *
97  * RCU is used for these lists, so ghes_list_mutex is only used for
98  * list changing, not for traversing.
99  */
100 static LIST_HEAD(ghes_sci);
101 static LIST_HEAD(ghes_nmi);
102 static DEFINE_MUTEX(ghes_list_mutex);
103 
104 /*
105  * NMI may be triggered on any CPU, so ghes_nmi_lock is used for
106  * mutual exclusion.
107  */
108 static DEFINE_RAW_SPINLOCK(ghes_nmi_lock);
109 
110 /*
111  * Because the memory area used to transfer hardware error information
112  * from BIOS to Linux can be determined only in NMI, IRQ or timer
113  * handler, but general ioremap can not be used in atomic context, so
114  * a special version of atomic ioremap is implemented for that.
115  */
116 
117 /*
118  * Two virtual pages are used, one for NMI context, the other for
119  * IRQ/PROCESS context
120  */
121 #define GHES_IOREMAP_PAGES		2
122 #define GHES_IOREMAP_NMI_PAGE(base)	(base)
123 #define GHES_IOREMAP_IRQ_PAGE(base)	((base) + PAGE_SIZE)
124 
125 /* virtual memory area for atomic ioremap */
126 static struct vm_struct *ghes_ioremap_area;
127 /*
128  * These 2 spinlock is used to prevent atomic ioremap virtual memory
129  * area from being mapped simultaneously.
130  */
131 static DEFINE_RAW_SPINLOCK(ghes_ioremap_lock_nmi);
132 static DEFINE_SPINLOCK(ghes_ioremap_lock_irq);
133 
134 /*
135  * printk is not safe in NMI context.  So in NMI handler, we allocate
136  * required memory from lock-less memory allocator
137  * (ghes_estatus_pool), save estatus into it, put them into lock-less
138  * list (ghes_estatus_llist), then delay printk into IRQ context via
139  * irq_work (ghes_proc_irq_work).  ghes_estatus_size_request record
140  * required pool size by all NMI error source.
141  */
142 static struct gen_pool *ghes_estatus_pool;
143 static unsigned long ghes_estatus_pool_size_request;
144 static struct llist_head ghes_estatus_llist;
145 static struct irq_work ghes_proc_irq_work;
146 
147 struct ghes_estatus_cache *ghes_estatus_caches[GHES_ESTATUS_CACHES_SIZE];
148 static atomic_t ghes_estatus_cache_alloced;
149 
150 static int ghes_ioremap_init(void)
151 {
152 	ghes_ioremap_area = __get_vm_area(PAGE_SIZE * GHES_IOREMAP_PAGES,
153 		VM_IOREMAP, VMALLOC_START, VMALLOC_END);
154 	if (!ghes_ioremap_area) {
155 		pr_err(GHES_PFX "Failed to allocate virtual memory area for atomic ioremap.\n");
156 		return -ENOMEM;
157 	}
158 
159 	return 0;
160 }
161 
162 static void ghes_ioremap_exit(void)
163 {
164 	free_vm_area(ghes_ioremap_area);
165 }
166 
167 static void __iomem *ghes_ioremap_pfn_nmi(u64 pfn)
168 {
169 	unsigned long vaddr;
170 
171 	vaddr = (unsigned long)GHES_IOREMAP_NMI_PAGE(ghes_ioremap_area->addr);
172 	ioremap_page_range(vaddr, vaddr + PAGE_SIZE,
173 			   pfn << PAGE_SHIFT, PAGE_KERNEL);
174 
175 	return (void __iomem *)vaddr;
176 }
177 
178 static void __iomem *ghes_ioremap_pfn_irq(u64 pfn)
179 {
180 	unsigned long vaddr;
181 
182 	vaddr = (unsigned long)GHES_IOREMAP_IRQ_PAGE(ghes_ioremap_area->addr);
183 	ioremap_page_range(vaddr, vaddr + PAGE_SIZE,
184 			   pfn << PAGE_SHIFT, PAGE_KERNEL);
185 
186 	return (void __iomem *)vaddr;
187 }
188 
189 static void ghes_iounmap_nmi(void __iomem *vaddr_ptr)
190 {
191 	unsigned long vaddr = (unsigned long __force)vaddr_ptr;
192 	void *base = ghes_ioremap_area->addr;
193 
194 	BUG_ON(vaddr != (unsigned long)GHES_IOREMAP_NMI_PAGE(base));
195 	unmap_kernel_range_noflush(vaddr, PAGE_SIZE);
196 	__flush_tlb_one(vaddr);
197 }
198 
199 static void ghes_iounmap_irq(void __iomem *vaddr_ptr)
200 {
201 	unsigned long vaddr = (unsigned long __force)vaddr_ptr;
202 	void *base = ghes_ioremap_area->addr;
203 
204 	BUG_ON(vaddr != (unsigned long)GHES_IOREMAP_IRQ_PAGE(base));
205 	unmap_kernel_range_noflush(vaddr, PAGE_SIZE);
206 	__flush_tlb_one(vaddr);
207 }
208 
209 static int ghes_estatus_pool_init(void)
210 {
211 	ghes_estatus_pool = gen_pool_create(GHES_ESTATUS_POOL_MIN_ALLOC_ORDER, -1);
212 	if (!ghes_estatus_pool)
213 		return -ENOMEM;
214 	return 0;
215 }
216 
217 static void ghes_estatus_pool_free_chunk_page(struct gen_pool *pool,
218 					      struct gen_pool_chunk *chunk,
219 					      void *data)
220 {
221 	free_page(chunk->start_addr);
222 }
223 
224 static void ghes_estatus_pool_exit(void)
225 {
226 	gen_pool_for_each_chunk(ghes_estatus_pool,
227 				ghes_estatus_pool_free_chunk_page, NULL);
228 	gen_pool_destroy(ghes_estatus_pool);
229 }
230 
231 static int ghes_estatus_pool_expand(unsigned long len)
232 {
233 	unsigned long i, pages, size, addr;
234 	int ret;
235 
236 	ghes_estatus_pool_size_request += PAGE_ALIGN(len);
237 	size = gen_pool_size(ghes_estatus_pool);
238 	if (size >= ghes_estatus_pool_size_request)
239 		return 0;
240 	pages = (ghes_estatus_pool_size_request - size) / PAGE_SIZE;
241 	for (i = 0; i < pages; i++) {
242 		addr = __get_free_page(GFP_KERNEL);
243 		if (!addr)
244 			return -ENOMEM;
245 		ret = gen_pool_add(ghes_estatus_pool, addr, PAGE_SIZE, -1);
246 		if (ret)
247 			return ret;
248 	}
249 
250 	return 0;
251 }
252 
253 static void ghes_estatus_pool_shrink(unsigned long len)
254 {
255 	ghes_estatus_pool_size_request -= PAGE_ALIGN(len);
256 }
257 
258 static struct ghes *ghes_new(struct acpi_hest_generic *generic)
259 {
260 	struct ghes *ghes;
261 	unsigned int error_block_length;
262 	int rc;
263 
264 	ghes = kzalloc(sizeof(*ghes), GFP_KERNEL);
265 	if (!ghes)
266 		return ERR_PTR(-ENOMEM);
267 	ghes->generic = generic;
268 	rc = apei_map_generic_address(&generic->error_status_address);
269 	if (rc)
270 		goto err_free;
271 	error_block_length = generic->error_block_length;
272 	if (error_block_length > GHES_ESTATUS_MAX_SIZE) {
273 		pr_warning(FW_WARN GHES_PFX
274 			   "Error status block length is too long: %u for "
275 			   "generic hardware error source: %d.\n",
276 			   error_block_length, generic->header.source_id);
277 		error_block_length = GHES_ESTATUS_MAX_SIZE;
278 	}
279 	ghes->estatus = kmalloc(error_block_length, GFP_KERNEL);
280 	if (!ghes->estatus) {
281 		rc = -ENOMEM;
282 		goto err_unmap;
283 	}
284 
285 	return ghes;
286 
287 err_unmap:
288 	apei_unmap_generic_address(&generic->error_status_address);
289 err_free:
290 	kfree(ghes);
291 	return ERR_PTR(rc);
292 }
293 
294 static void ghes_fini(struct ghes *ghes)
295 {
296 	kfree(ghes->estatus);
297 	apei_unmap_generic_address(&ghes->generic->error_status_address);
298 }
299 
300 static inline int ghes_severity(int severity)
301 {
302 	switch (severity) {
303 	case CPER_SEV_INFORMATIONAL:
304 		return GHES_SEV_NO;
305 	case CPER_SEV_CORRECTED:
306 		return GHES_SEV_CORRECTED;
307 	case CPER_SEV_RECOVERABLE:
308 		return GHES_SEV_RECOVERABLE;
309 	case CPER_SEV_FATAL:
310 		return GHES_SEV_PANIC;
311 	default:
312 		/* Unknown, go panic */
313 		return GHES_SEV_PANIC;
314 	}
315 }
316 
317 static void ghes_copy_tofrom_phys(void *buffer, u64 paddr, u32 len,
318 				  int from_phys)
319 {
320 	void __iomem *vaddr;
321 	unsigned long flags = 0;
322 	int in_nmi = in_nmi();
323 	u64 offset;
324 	u32 trunk;
325 
326 	while (len > 0) {
327 		offset = paddr - (paddr & PAGE_MASK);
328 		if (in_nmi) {
329 			raw_spin_lock(&ghes_ioremap_lock_nmi);
330 			vaddr = ghes_ioremap_pfn_nmi(paddr >> PAGE_SHIFT);
331 		} else {
332 			spin_lock_irqsave(&ghes_ioremap_lock_irq, flags);
333 			vaddr = ghes_ioremap_pfn_irq(paddr >> PAGE_SHIFT);
334 		}
335 		trunk = PAGE_SIZE - offset;
336 		trunk = min(trunk, len);
337 		if (from_phys)
338 			memcpy_fromio(buffer, vaddr + offset, trunk);
339 		else
340 			memcpy_toio(vaddr + offset, buffer, trunk);
341 		len -= trunk;
342 		paddr += trunk;
343 		buffer += trunk;
344 		if (in_nmi) {
345 			ghes_iounmap_nmi(vaddr);
346 			raw_spin_unlock(&ghes_ioremap_lock_nmi);
347 		} else {
348 			ghes_iounmap_irq(vaddr);
349 			spin_unlock_irqrestore(&ghes_ioremap_lock_irq, flags);
350 		}
351 	}
352 }
353 
354 static int ghes_read_estatus(struct ghes *ghes, int silent)
355 {
356 	struct acpi_hest_generic *g = ghes->generic;
357 	u64 buf_paddr;
358 	u32 len;
359 	int rc;
360 
361 	rc = apei_read(&buf_paddr, &g->error_status_address);
362 	if (rc) {
363 		if (!silent && printk_ratelimit())
364 			pr_warning(FW_WARN GHES_PFX
365 "Failed to read error status block address for hardware error source: %d.\n",
366 				   g->header.source_id);
367 		return -EIO;
368 	}
369 	if (!buf_paddr)
370 		return -ENOENT;
371 
372 	ghes_copy_tofrom_phys(ghes->estatus, buf_paddr,
373 			      sizeof(*ghes->estatus), 1);
374 	if (!ghes->estatus->block_status)
375 		return -ENOENT;
376 
377 	ghes->buffer_paddr = buf_paddr;
378 	ghes->flags |= GHES_TO_CLEAR;
379 
380 	rc = -EIO;
381 	len = cper_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 (cper_estatus_check_header(ghes->estatus))
387 		goto err_read_block;
388 	ghes_copy_tofrom_phys(ghes->estatus + 1,
389 			      buf_paddr + sizeof(*ghes->estatus),
390 			      len - sizeof(*ghes->estatus), 1);
391 	if (cper_estatus_check(ghes->estatus))
392 		goto err_read_block;
393 	rc = 0;
394 
395 err_read_block:
396 	if (rc && !silent && printk_ratelimit())
397 		pr_warning(FW_WARN GHES_PFX
398 			   "Failed to read error status block!\n");
399 	return rc;
400 }
401 
402 static void ghes_clear_estatus(struct ghes *ghes)
403 {
404 	ghes->estatus->block_status = 0;
405 	if (!(ghes->flags & GHES_TO_CLEAR))
406 		return;
407 	ghes_copy_tofrom_phys(ghes->estatus, ghes->buffer_paddr,
408 			      sizeof(ghes->estatus->block_status), 0);
409 	ghes->flags &= ~GHES_TO_CLEAR;
410 }
411 
412 static void ghes_handle_memory_failure(struct acpi_generic_data *gdata, int sev)
413 {
414 #ifdef CONFIG_ACPI_APEI_MEMORY_FAILURE
415 	unsigned long pfn;
416 	int sec_sev = ghes_severity(gdata->error_severity);
417 	struct cper_sec_mem_err *mem_err;
418 	mem_err = (struct cper_sec_mem_err *)(gdata + 1);
419 
420 	if (sec_sev == GHES_SEV_CORRECTED &&
421 	    (gdata->flags & CPER_SEC_ERROR_THRESHOLD_EXCEEDED) &&
422 	    (mem_err->validation_bits & CPER_MEM_VALID_PA)) {
423 		pfn = mem_err->physical_addr >> PAGE_SHIFT;
424 		if (pfn_valid(pfn))
425 			memory_failure_queue(pfn, 0, MF_SOFT_OFFLINE);
426 		else if (printk_ratelimit())
427 			pr_warn(FW_WARN GHES_PFX
428 			"Invalid address in generic error data: %#llx\n",
429 			mem_err->physical_addr);
430 	}
431 	if (sev == GHES_SEV_RECOVERABLE &&
432 	    sec_sev == GHES_SEV_RECOVERABLE &&
433 	    mem_err->validation_bits & CPER_MEM_VALID_PA) {
434 		pfn = mem_err->physical_addr >> PAGE_SHIFT;
435 		memory_failure_queue(pfn, 0, 0);
436 	}
437 #endif
438 }
439 
440 static void ghes_do_proc(struct ghes *ghes,
441 			 const struct acpi_generic_status *estatus)
442 {
443 	int sev, sec_sev;
444 	struct acpi_generic_data *gdata;
445 
446 	sev = ghes_severity(estatus->error_severity);
447 	apei_estatus_for_each_section(estatus, gdata) {
448 		sec_sev = ghes_severity(gdata->error_severity);
449 		if (!uuid_le_cmp(*(uuid_le *)gdata->section_type,
450 				 CPER_SEC_PLATFORM_MEM)) {
451 			struct cper_sec_mem_err *mem_err;
452 			mem_err = (struct cper_sec_mem_err *)(gdata+1);
453 			ghes_edac_report_mem_error(ghes, sev, mem_err);
454 
455 #ifdef CONFIG_X86_MCE
456 			apei_mce_report_mem_error(sev == GHES_SEV_CORRECTED,
457 						  mem_err);
458 #endif
459 			ghes_handle_memory_failure(gdata, sev);
460 		}
461 #ifdef CONFIG_ACPI_APEI_PCIEAER
462 		else if (!uuid_le_cmp(*(uuid_le *)gdata->section_type,
463 				      CPER_SEC_PCIE)) {
464 			struct cper_sec_pcie *pcie_err;
465 			pcie_err = (struct cper_sec_pcie *)(gdata+1);
466 			if (sev == GHES_SEV_RECOVERABLE &&
467 			    sec_sev == GHES_SEV_RECOVERABLE &&
468 			    pcie_err->validation_bits & CPER_PCIE_VALID_DEVICE_ID &&
469 			    pcie_err->validation_bits & CPER_PCIE_VALID_AER_INFO) {
470 				unsigned int devfn;
471 				int aer_severity;
472 
473 				devfn = PCI_DEVFN(pcie_err->device_id.device,
474 						  pcie_err->device_id.function);
475 				aer_severity = cper_severity_to_aer(sev);
476 
477 				/*
478 				 * If firmware reset the component to contain
479 				 * the error, we must reinitialize it before
480 				 * use, so treat it as a fatal AER error.
481 				 */
482 				if (gdata->flags & CPER_SEC_RESET)
483 					aer_severity = AER_FATAL;
484 
485 				aer_recover_queue(pcie_err->device_id.segment,
486 						  pcie_err->device_id.bus,
487 						  devfn, aer_severity,
488 						  (struct aer_capability_regs *)
489 						  pcie_err->aer_info);
490 			}
491 
492 		}
493 #endif
494 	}
495 }
496 
497 static void __ghes_print_estatus(const char *pfx,
498 				 const struct acpi_hest_generic *generic,
499 				 const struct acpi_generic_status *estatus)
500 {
501 	static atomic_t seqno;
502 	unsigned int curr_seqno;
503 	char pfx_seq[64];
504 
505 	if (pfx == NULL) {
506 		if (ghes_severity(estatus->error_severity) <=
507 		    GHES_SEV_CORRECTED)
508 			pfx = KERN_WARNING;
509 		else
510 			pfx = KERN_ERR;
511 	}
512 	curr_seqno = atomic_inc_return(&seqno);
513 	snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}" HW_ERR, pfx, curr_seqno);
514 	printk("%s""Hardware error from APEI Generic Hardware Error Source: %d\n",
515 	       pfx_seq, generic->header.source_id);
516 	cper_estatus_print(pfx_seq, estatus);
517 }
518 
519 static int ghes_print_estatus(const char *pfx,
520 			      const struct acpi_hest_generic *generic,
521 			      const struct acpi_generic_status *estatus)
522 {
523 	/* Not more than 2 messages every 5 seconds */
524 	static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2);
525 	static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2);
526 	struct ratelimit_state *ratelimit;
527 
528 	if (ghes_severity(estatus->error_severity) <= GHES_SEV_CORRECTED)
529 		ratelimit = &ratelimit_corrected;
530 	else
531 		ratelimit = &ratelimit_uncorrected;
532 	if (__ratelimit(ratelimit)) {
533 		__ghes_print_estatus(pfx, generic, estatus);
534 		return 1;
535 	}
536 	return 0;
537 }
538 
539 /*
540  * GHES error status reporting throttle, to report more kinds of
541  * errors, instead of just most frequently occurred errors.
542  */
543 static int ghes_estatus_cached(struct acpi_generic_status *estatus)
544 {
545 	u32 len;
546 	int i, cached = 0;
547 	unsigned long long now;
548 	struct ghes_estatus_cache *cache;
549 	struct acpi_generic_status *cache_estatus;
550 
551 	len = cper_estatus_len(estatus);
552 	rcu_read_lock();
553 	for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
554 		cache = rcu_dereference(ghes_estatus_caches[i]);
555 		if (cache == NULL)
556 			continue;
557 		if (len != cache->estatus_len)
558 			continue;
559 		cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
560 		if (memcmp(estatus, cache_estatus, len))
561 			continue;
562 		atomic_inc(&cache->count);
563 		now = sched_clock();
564 		if (now - cache->time_in < GHES_ESTATUS_IN_CACHE_MAX_NSEC)
565 			cached = 1;
566 		break;
567 	}
568 	rcu_read_unlock();
569 	return cached;
570 }
571 
572 static struct ghes_estatus_cache *ghes_estatus_cache_alloc(
573 	struct acpi_hest_generic *generic,
574 	struct acpi_generic_status *estatus)
575 {
576 	int alloced;
577 	u32 len, cache_len;
578 	struct ghes_estatus_cache *cache;
579 	struct acpi_generic_status *cache_estatus;
580 
581 	alloced = atomic_add_return(1, &ghes_estatus_cache_alloced);
582 	if (alloced > GHES_ESTATUS_CACHE_ALLOCED_MAX) {
583 		atomic_dec(&ghes_estatus_cache_alloced);
584 		return NULL;
585 	}
586 	len = cper_estatus_len(estatus);
587 	cache_len = GHES_ESTATUS_CACHE_LEN(len);
588 	cache = (void *)gen_pool_alloc(ghes_estatus_pool, cache_len);
589 	if (!cache) {
590 		atomic_dec(&ghes_estatus_cache_alloced);
591 		return NULL;
592 	}
593 	cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
594 	memcpy(cache_estatus, estatus, len);
595 	cache->estatus_len = len;
596 	atomic_set(&cache->count, 0);
597 	cache->generic = generic;
598 	cache->time_in = sched_clock();
599 	return cache;
600 }
601 
602 static void ghes_estatus_cache_free(struct ghes_estatus_cache *cache)
603 {
604 	u32 len;
605 
606 	len = cper_estatus_len(GHES_ESTATUS_FROM_CACHE(cache));
607 	len = GHES_ESTATUS_CACHE_LEN(len);
608 	gen_pool_free(ghes_estatus_pool, (unsigned long)cache, len);
609 	atomic_dec(&ghes_estatus_cache_alloced);
610 }
611 
612 static void ghes_estatus_cache_rcu_free(struct rcu_head *head)
613 {
614 	struct ghes_estatus_cache *cache;
615 
616 	cache = container_of(head, struct ghes_estatus_cache, rcu);
617 	ghes_estatus_cache_free(cache);
618 }
619 
620 static void ghes_estatus_cache_add(
621 	struct acpi_hest_generic *generic,
622 	struct acpi_generic_status *estatus)
623 {
624 	int i, slot = -1, count;
625 	unsigned long long now, duration, period, max_period = 0;
626 	struct ghes_estatus_cache *cache, *slot_cache = NULL, *new_cache;
627 
628 	new_cache = ghes_estatus_cache_alloc(generic, estatus);
629 	if (new_cache == NULL)
630 		return;
631 	rcu_read_lock();
632 	now = sched_clock();
633 	for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
634 		cache = rcu_dereference(ghes_estatus_caches[i]);
635 		if (cache == NULL) {
636 			slot = i;
637 			slot_cache = NULL;
638 			break;
639 		}
640 		duration = now - cache->time_in;
641 		if (duration >= GHES_ESTATUS_IN_CACHE_MAX_NSEC) {
642 			slot = i;
643 			slot_cache = cache;
644 			break;
645 		}
646 		count = atomic_read(&cache->count);
647 		period = duration;
648 		do_div(period, (count + 1));
649 		if (period > max_period) {
650 			max_period = period;
651 			slot = i;
652 			slot_cache = cache;
653 		}
654 	}
655 	/* new_cache must be put into array after its contents are written */
656 	smp_wmb();
657 	if (slot != -1 && cmpxchg(ghes_estatus_caches + slot,
658 				  slot_cache, new_cache) == slot_cache) {
659 		if (slot_cache)
660 			call_rcu(&slot_cache->rcu, ghes_estatus_cache_rcu_free);
661 	} else
662 		ghes_estatus_cache_free(new_cache);
663 	rcu_read_unlock();
664 }
665 
666 static int ghes_proc(struct ghes *ghes)
667 {
668 	int rc;
669 
670 	rc = ghes_read_estatus(ghes, 0);
671 	if (rc)
672 		goto out;
673 	if (!ghes_estatus_cached(ghes->estatus)) {
674 		if (ghes_print_estatus(NULL, ghes->generic, ghes->estatus))
675 			ghes_estatus_cache_add(ghes->generic, ghes->estatus);
676 	}
677 	ghes_do_proc(ghes, ghes->estatus);
678 out:
679 	ghes_clear_estatus(ghes);
680 	return 0;
681 }
682 
683 static void ghes_add_timer(struct ghes *ghes)
684 {
685 	struct acpi_hest_generic *g = ghes->generic;
686 	unsigned long expire;
687 
688 	if (!g->notify.poll_interval) {
689 		pr_warning(FW_WARN GHES_PFX "Poll interval is 0 for generic hardware error source: %d, disabled.\n",
690 			   g->header.source_id);
691 		return;
692 	}
693 	expire = jiffies + msecs_to_jiffies(g->notify.poll_interval);
694 	ghes->timer.expires = round_jiffies_relative(expire);
695 	add_timer(&ghes->timer);
696 }
697 
698 static void ghes_poll_func(unsigned long data)
699 {
700 	struct ghes *ghes = (void *)data;
701 
702 	ghes_proc(ghes);
703 	if (!(ghes->flags & GHES_EXITING))
704 		ghes_add_timer(ghes);
705 }
706 
707 static irqreturn_t ghes_irq_func(int irq, void *data)
708 {
709 	struct ghes *ghes = data;
710 	int rc;
711 
712 	rc = ghes_proc(ghes);
713 	if (rc)
714 		return IRQ_NONE;
715 
716 	return IRQ_HANDLED;
717 }
718 
719 static int ghes_notify_sci(struct notifier_block *this,
720 				  unsigned long event, void *data)
721 {
722 	struct ghes *ghes;
723 	int ret = NOTIFY_DONE;
724 
725 	rcu_read_lock();
726 	list_for_each_entry_rcu(ghes, &ghes_sci, list) {
727 		if (!ghes_proc(ghes))
728 			ret = NOTIFY_OK;
729 	}
730 	rcu_read_unlock();
731 
732 	return ret;
733 }
734 
735 static struct llist_node *llist_nodes_reverse(struct llist_node *llnode)
736 {
737 	struct llist_node *next, *tail = NULL;
738 
739 	while (llnode) {
740 		next = llnode->next;
741 		llnode->next = tail;
742 		tail = llnode;
743 		llnode = next;
744 	}
745 
746 	return tail;
747 }
748 
749 static void ghes_proc_in_irq(struct irq_work *irq_work)
750 {
751 	struct llist_node *llnode, *next;
752 	struct ghes_estatus_node *estatus_node;
753 	struct acpi_hest_generic *generic;
754 	struct acpi_generic_status *estatus;
755 	u32 len, node_len;
756 
757 	llnode = llist_del_all(&ghes_estatus_llist);
758 	/*
759 	 * Because the time order of estatus in list is reversed,
760 	 * revert it back to proper order.
761 	 */
762 	llnode = llist_nodes_reverse(llnode);
763 	while (llnode) {
764 		next = llnode->next;
765 		estatus_node = llist_entry(llnode, struct ghes_estatus_node,
766 					   llnode);
767 		estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
768 		len = cper_estatus_len(estatus);
769 		node_len = GHES_ESTATUS_NODE_LEN(len);
770 		ghes_do_proc(estatus_node->ghes, estatus);
771 		if (!ghes_estatus_cached(estatus)) {
772 			generic = estatus_node->generic;
773 			if (ghes_print_estatus(NULL, generic, estatus))
774 				ghes_estatus_cache_add(generic, estatus);
775 		}
776 		gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node,
777 			      node_len);
778 		llnode = next;
779 	}
780 }
781 
782 static void ghes_print_queued_estatus(void)
783 {
784 	struct llist_node *llnode;
785 	struct ghes_estatus_node *estatus_node;
786 	struct acpi_hest_generic *generic;
787 	struct acpi_generic_status *estatus;
788 	u32 len, node_len;
789 
790 	llnode = llist_del_all(&ghes_estatus_llist);
791 	/*
792 	 * Because the time order of estatus in list is reversed,
793 	 * revert it back to proper order.
794 	 */
795 	llnode = llist_nodes_reverse(llnode);
796 	while (llnode) {
797 		estatus_node = llist_entry(llnode, struct ghes_estatus_node,
798 					   llnode);
799 		estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
800 		len = cper_estatus_len(estatus);
801 		node_len = GHES_ESTATUS_NODE_LEN(len);
802 		generic = estatus_node->generic;
803 		ghes_print_estatus(NULL, generic, estatus);
804 		llnode = llnode->next;
805 	}
806 }
807 
808 static int ghes_notify_nmi(unsigned int cmd, struct pt_regs *regs)
809 {
810 	struct ghes *ghes, *ghes_global = NULL;
811 	int sev, sev_global = -1;
812 	int ret = NMI_DONE;
813 
814 	raw_spin_lock(&ghes_nmi_lock);
815 	list_for_each_entry_rcu(ghes, &ghes_nmi, list) {
816 		if (ghes_read_estatus(ghes, 1)) {
817 			ghes_clear_estatus(ghes);
818 			continue;
819 		}
820 		sev = ghes_severity(ghes->estatus->error_severity);
821 		if (sev > sev_global) {
822 			sev_global = sev;
823 			ghes_global = ghes;
824 		}
825 		ret = NMI_HANDLED;
826 	}
827 
828 	if (ret == NMI_DONE)
829 		goto out;
830 
831 	if (sev_global >= GHES_SEV_PANIC) {
832 		oops_begin();
833 		ghes_print_queued_estatus();
834 		__ghes_print_estatus(KERN_EMERG, ghes_global->generic,
835 				     ghes_global->estatus);
836 		/* reboot to log the error! */
837 		if (panic_timeout == 0)
838 			panic_timeout = ghes_panic_timeout;
839 		panic("Fatal hardware error!");
840 	}
841 
842 	list_for_each_entry_rcu(ghes, &ghes_nmi, list) {
843 #ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
844 		u32 len, node_len;
845 		struct ghes_estatus_node *estatus_node;
846 		struct acpi_generic_status *estatus;
847 #endif
848 		if (!(ghes->flags & GHES_TO_CLEAR))
849 			continue;
850 #ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
851 		if (ghes_estatus_cached(ghes->estatus))
852 			goto next;
853 		/* Save estatus for further processing in IRQ context */
854 		len = cper_estatus_len(ghes->estatus);
855 		node_len = GHES_ESTATUS_NODE_LEN(len);
856 		estatus_node = (void *)gen_pool_alloc(ghes_estatus_pool,
857 						      node_len);
858 		if (estatus_node) {
859 			estatus_node->ghes = ghes;
860 			estatus_node->generic = ghes->generic;
861 			estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
862 			memcpy(estatus, ghes->estatus, len);
863 			llist_add(&estatus_node->llnode, &ghes_estatus_llist);
864 		}
865 next:
866 #endif
867 		ghes_clear_estatus(ghes);
868 	}
869 #ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
870 	irq_work_queue(&ghes_proc_irq_work);
871 #endif
872 
873 out:
874 	raw_spin_unlock(&ghes_nmi_lock);
875 	return ret;
876 }
877 
878 static struct notifier_block ghes_notifier_sci = {
879 	.notifier_call = ghes_notify_sci,
880 };
881 
882 static unsigned long ghes_esource_prealloc_size(
883 	const struct acpi_hest_generic *generic)
884 {
885 	unsigned long block_length, prealloc_records, prealloc_size;
886 
887 	block_length = min_t(unsigned long, generic->error_block_length,
888 			     GHES_ESTATUS_MAX_SIZE);
889 	prealloc_records = max_t(unsigned long,
890 				 generic->records_to_preallocate, 1);
891 	prealloc_size = min_t(unsigned long, block_length * prealloc_records,
892 			      GHES_ESOURCE_PREALLOC_MAX_SIZE);
893 
894 	return prealloc_size;
895 }
896 
897 static int ghes_probe(struct platform_device *ghes_dev)
898 {
899 	struct acpi_hest_generic *generic;
900 	struct ghes *ghes = NULL;
901 	unsigned long len;
902 	int rc = -EINVAL;
903 
904 	generic = *(struct acpi_hest_generic **)ghes_dev->dev.platform_data;
905 	if (!generic->enabled)
906 		return -ENODEV;
907 
908 	switch (generic->notify.type) {
909 	case ACPI_HEST_NOTIFY_POLLED:
910 	case ACPI_HEST_NOTIFY_EXTERNAL:
911 	case ACPI_HEST_NOTIFY_SCI:
912 	case ACPI_HEST_NOTIFY_NMI:
913 		break;
914 	case ACPI_HEST_NOTIFY_LOCAL:
915 		pr_warning(GHES_PFX "Generic hardware error source: %d notified via local interrupt is not supported!\n",
916 			   generic->header.source_id);
917 		goto err;
918 	default:
919 		pr_warning(FW_WARN GHES_PFX "Unknown notification type: %u for generic hardware error source: %d\n",
920 			   generic->notify.type, generic->header.source_id);
921 		goto err;
922 	}
923 
924 	rc = -EIO;
925 	if (generic->error_block_length <
926 	    sizeof(struct acpi_generic_status)) {
927 		pr_warning(FW_BUG GHES_PFX "Invalid error block length: %u for generic hardware error source: %d\n",
928 			   generic->error_block_length,
929 			   generic->header.source_id);
930 		goto err;
931 	}
932 	ghes = ghes_new(generic);
933 	if (IS_ERR(ghes)) {
934 		rc = PTR_ERR(ghes);
935 		ghes = NULL;
936 		goto err;
937 	}
938 
939 	rc = ghes_edac_register(ghes, &ghes_dev->dev);
940 	if (rc < 0)
941 		goto err;
942 
943 	switch (generic->notify.type) {
944 	case ACPI_HEST_NOTIFY_POLLED:
945 		ghes->timer.function = ghes_poll_func;
946 		ghes->timer.data = (unsigned long)ghes;
947 		init_timer_deferrable(&ghes->timer);
948 		ghes_add_timer(ghes);
949 		break;
950 	case ACPI_HEST_NOTIFY_EXTERNAL:
951 		/* External interrupt vector is GSI */
952 		rc = acpi_gsi_to_irq(generic->notify.vector, &ghes->irq);
953 		if (rc) {
954 			pr_err(GHES_PFX "Failed to map GSI to IRQ for generic hardware error source: %d\n",
955 			       generic->header.source_id);
956 			goto err_edac_unreg;
957 		}
958 		rc = request_irq(ghes->irq, ghes_irq_func, 0, "GHES IRQ", ghes);
959 		if (rc) {
960 			pr_err(GHES_PFX "Failed to register IRQ for generic hardware error source: %d\n",
961 			       generic->header.source_id);
962 			goto err_edac_unreg;
963 		}
964 		break;
965 	case ACPI_HEST_NOTIFY_SCI:
966 		mutex_lock(&ghes_list_mutex);
967 		if (list_empty(&ghes_sci))
968 			register_acpi_hed_notifier(&ghes_notifier_sci);
969 		list_add_rcu(&ghes->list, &ghes_sci);
970 		mutex_unlock(&ghes_list_mutex);
971 		break;
972 	case ACPI_HEST_NOTIFY_NMI:
973 		len = ghes_esource_prealloc_size(generic);
974 		ghes_estatus_pool_expand(len);
975 		mutex_lock(&ghes_list_mutex);
976 		if (list_empty(&ghes_nmi))
977 			register_nmi_handler(NMI_LOCAL, ghes_notify_nmi, 0,
978 						"ghes");
979 		list_add_rcu(&ghes->list, &ghes_nmi);
980 		mutex_unlock(&ghes_list_mutex);
981 		break;
982 	default:
983 		BUG();
984 	}
985 	platform_set_drvdata(ghes_dev, ghes);
986 
987 	return 0;
988 err_edac_unreg:
989 	ghes_edac_unregister(ghes);
990 err:
991 	if (ghes) {
992 		ghes_fini(ghes);
993 		kfree(ghes);
994 	}
995 	return rc;
996 }
997 
998 static int ghes_remove(struct platform_device *ghes_dev)
999 {
1000 	struct ghes *ghes;
1001 	struct acpi_hest_generic *generic;
1002 	unsigned long len;
1003 
1004 	ghes = platform_get_drvdata(ghes_dev);
1005 	generic = ghes->generic;
1006 
1007 	ghes->flags |= GHES_EXITING;
1008 	switch (generic->notify.type) {
1009 	case ACPI_HEST_NOTIFY_POLLED:
1010 		del_timer_sync(&ghes->timer);
1011 		break;
1012 	case ACPI_HEST_NOTIFY_EXTERNAL:
1013 		free_irq(ghes->irq, ghes);
1014 		break;
1015 	case ACPI_HEST_NOTIFY_SCI:
1016 		mutex_lock(&ghes_list_mutex);
1017 		list_del_rcu(&ghes->list);
1018 		if (list_empty(&ghes_sci))
1019 			unregister_acpi_hed_notifier(&ghes_notifier_sci);
1020 		mutex_unlock(&ghes_list_mutex);
1021 		break;
1022 	case ACPI_HEST_NOTIFY_NMI:
1023 		mutex_lock(&ghes_list_mutex);
1024 		list_del_rcu(&ghes->list);
1025 		if (list_empty(&ghes_nmi))
1026 			unregister_nmi_handler(NMI_LOCAL, "ghes");
1027 		mutex_unlock(&ghes_list_mutex);
1028 		/*
1029 		 * To synchronize with NMI handler, ghes can only be
1030 		 * freed after NMI handler finishes.
1031 		 */
1032 		synchronize_rcu();
1033 		len = ghes_esource_prealloc_size(generic);
1034 		ghes_estatus_pool_shrink(len);
1035 		break;
1036 	default:
1037 		BUG();
1038 		break;
1039 	}
1040 
1041 	ghes_fini(ghes);
1042 
1043 	ghes_edac_unregister(ghes);
1044 
1045 	kfree(ghes);
1046 
1047 	platform_set_drvdata(ghes_dev, NULL);
1048 
1049 	return 0;
1050 }
1051 
1052 static struct platform_driver ghes_platform_driver = {
1053 	.driver		= {
1054 		.name	= "GHES",
1055 		.owner	= THIS_MODULE,
1056 	},
1057 	.probe		= ghes_probe,
1058 	.remove		= ghes_remove,
1059 };
1060 
1061 static int __init ghes_init(void)
1062 {
1063 	int rc;
1064 
1065 	if (acpi_disabled)
1066 		return -ENODEV;
1067 
1068 	if (hest_disable) {
1069 		pr_info(GHES_PFX "HEST is not enabled!\n");
1070 		return -EINVAL;
1071 	}
1072 
1073 	if (ghes_disable) {
1074 		pr_info(GHES_PFX "GHES is not enabled!\n");
1075 		return -EINVAL;
1076 	}
1077 
1078 	init_irq_work(&ghes_proc_irq_work, ghes_proc_in_irq);
1079 
1080 	rc = ghes_ioremap_init();
1081 	if (rc)
1082 		goto err;
1083 
1084 	rc = ghes_estatus_pool_init();
1085 	if (rc)
1086 		goto err_ioremap_exit;
1087 
1088 	rc = ghes_estatus_pool_expand(GHES_ESTATUS_CACHE_AVG_SIZE *
1089 				      GHES_ESTATUS_CACHE_ALLOCED_MAX);
1090 	if (rc)
1091 		goto err_pool_exit;
1092 
1093 	rc = platform_driver_register(&ghes_platform_driver);
1094 	if (rc)
1095 		goto err_pool_exit;
1096 
1097 	rc = apei_osc_setup();
1098 	if (rc == 0 && osc_sb_apei_support_acked)
1099 		pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit and WHEA _OSC.\n");
1100 	else if (rc == 0 && !osc_sb_apei_support_acked)
1101 		pr_info(GHES_PFX "APEI firmware first mode is enabled by WHEA _OSC.\n");
1102 	else if (rc && osc_sb_apei_support_acked)
1103 		pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit.\n");
1104 	else
1105 		pr_info(GHES_PFX "Failed to enable APEI firmware first mode.\n");
1106 
1107 	return 0;
1108 err_pool_exit:
1109 	ghes_estatus_pool_exit();
1110 err_ioremap_exit:
1111 	ghes_ioremap_exit();
1112 err:
1113 	return rc;
1114 }
1115 
1116 static void __exit ghes_exit(void)
1117 {
1118 	platform_driver_unregister(&ghes_platform_driver);
1119 	ghes_estatus_pool_exit();
1120 	ghes_ioremap_exit();
1121 }
1122 
1123 module_init(ghes_init);
1124 module_exit(ghes_exit);
1125 
1126 MODULE_AUTHOR("Huang Ying");
1127 MODULE_DESCRIPTION("APEI Generic Hardware Error Source support");
1128 MODULE_LICENSE("GPL");
1129 MODULE_ALIAS("platform:GHES");
1130