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/io.h>
37#include <linux/interrupt.h>
38#include <linux/timer.h>
39#include <linux/cper.h>
40#include <linux/kdebug.h>
41#include <linux/platform_device.h>
42#include <linux/mutex.h>
43#include <linux/ratelimit.h>
44#include <linux/vmalloc.h>
45#include <linux/irq_work.h>
46#include <linux/llist.h>
47#include <linux/genalloc.h>
48#include <linux/pci.h>
49#include <linux/aer.h>
50#include <linux/nmi.h>
51
52#include <acpi/ghes.h>
53#include <acpi/apei.h>
54#include <asm/tlbflush.h>
55
56#include "apei-internal.h"
57
58#define GHES_PFX	"GHES: "
59
60#define GHES_ESTATUS_MAX_SIZE		65536
61#define GHES_ESOURCE_PREALLOC_MAX_SIZE	65536
62
63#define GHES_ESTATUS_POOL_MIN_ALLOC_ORDER 3
64
65/* This is just an estimation for memory pool allocation */
66#define GHES_ESTATUS_CACHE_AVG_SIZE	512
67
68#define GHES_ESTATUS_CACHES_SIZE	4
69
70#define GHES_ESTATUS_IN_CACHE_MAX_NSEC	10000000000ULL
71/* Prevent too many caches are allocated because of RCU */
72#define GHES_ESTATUS_CACHE_ALLOCED_MAX	(GHES_ESTATUS_CACHES_SIZE * 3 / 2)
73
74#define GHES_ESTATUS_CACHE_LEN(estatus_len)			\
75	(sizeof(struct ghes_estatus_cache) + (estatus_len))
76#define GHES_ESTATUS_FROM_CACHE(estatus_cache)			\
77	((struct acpi_hest_generic_status *)				\
78	 ((struct ghes_estatus_cache *)(estatus_cache) + 1))
79
80#define GHES_ESTATUS_NODE_LEN(estatus_len)			\
81	(sizeof(struct ghes_estatus_node) + (estatus_len))
82#define GHES_ESTATUS_FROM_NODE(estatus_node)			\
83	((struct acpi_hest_generic_status *)				\
84	 ((struct ghes_estatus_node *)(estatus_node) + 1))
85
86bool ghes_disable;
87module_param_named(disable, ghes_disable, bool, 0);
88
89/*
90 * All error sources notified with SCI shares one notifier function,
91 * so they need to be linked and checked one by one.  This is applied
92 * to NMI too.
93 *
94 * RCU is used for these lists, so ghes_list_mutex is only used for
95 * list changing, not for traversing.
96 */
97static LIST_HEAD(ghes_sci);
98static DEFINE_MUTEX(ghes_list_mutex);
99
100/*
101 * Because the memory area used to transfer hardware error information
102 * from BIOS to Linux can be determined only in NMI, IRQ or timer
103 * handler, but general ioremap can not be used in atomic context, so
104 * a special version of atomic ioremap is implemented for that.
105 */
106
107/*
108 * Two virtual pages are used, one for IRQ/PROCESS context, the other for
109 * NMI context (optionally).
110 */
111#ifdef CONFIG_HAVE_ACPI_APEI_NMI
112#define GHES_IOREMAP_PAGES           2
113#else
114#define GHES_IOREMAP_PAGES           1
115#endif
116#define GHES_IOREMAP_IRQ_PAGE(base)	(base)
117#define GHES_IOREMAP_NMI_PAGE(base)	((base) + PAGE_SIZE)
118
119/* virtual memory area for atomic ioremap */
120static struct vm_struct *ghes_ioremap_area;
121/*
122 * These 2 spinlock is used to prevent atomic ioremap virtual memory
123 * area from being mapped simultaneously.
124 */
125static DEFINE_RAW_SPINLOCK(ghes_ioremap_lock_nmi);
126static DEFINE_SPINLOCK(ghes_ioremap_lock_irq);
127
128static struct gen_pool *ghes_estatus_pool;
129static unsigned long ghes_estatus_pool_size_request;
130
131static struct ghes_estatus_cache *ghes_estatus_caches[GHES_ESTATUS_CACHES_SIZE];
132static atomic_t ghes_estatus_cache_alloced;
133
134static int ghes_ioremap_init(void)
135{
136	ghes_ioremap_area = __get_vm_area(PAGE_SIZE * GHES_IOREMAP_PAGES,
137		VM_IOREMAP, VMALLOC_START, VMALLOC_END);
138	if (!ghes_ioremap_area) {
139		pr_err(GHES_PFX "Failed to allocate virtual memory area for atomic ioremap.\n");
140		return -ENOMEM;
141	}
142
143	return 0;
144}
145
146static void ghes_ioremap_exit(void)
147{
148	free_vm_area(ghes_ioremap_area);
149}
150
151static void __iomem *ghes_ioremap_pfn_nmi(u64 pfn)
152{
153	unsigned long vaddr;
154
155	vaddr = (unsigned long)GHES_IOREMAP_NMI_PAGE(ghes_ioremap_area->addr);
156	ioremap_page_range(vaddr, vaddr + PAGE_SIZE,
157			   pfn << PAGE_SHIFT, PAGE_KERNEL);
158
159	return (void __iomem *)vaddr;
160}
161
162static void __iomem *ghes_ioremap_pfn_irq(u64 pfn)
163{
164	unsigned long vaddr;
165
166	vaddr = (unsigned long)GHES_IOREMAP_IRQ_PAGE(ghes_ioremap_area->addr);
167	ioremap_page_range(vaddr, vaddr + PAGE_SIZE,
168			   pfn << PAGE_SHIFT, PAGE_KERNEL);
169
170	return (void __iomem *)vaddr;
171}
172
173static void ghes_iounmap_nmi(void __iomem *vaddr_ptr)
174{
175	unsigned long vaddr = (unsigned long __force)vaddr_ptr;
176	void *base = ghes_ioremap_area->addr;
177
178	BUG_ON(vaddr != (unsigned long)GHES_IOREMAP_NMI_PAGE(base));
179	unmap_kernel_range_noflush(vaddr, PAGE_SIZE);
180	arch_apei_flush_tlb_one(vaddr);
181}
182
183static void ghes_iounmap_irq(void __iomem *vaddr_ptr)
184{
185	unsigned long vaddr = (unsigned long __force)vaddr_ptr;
186	void *base = ghes_ioremap_area->addr;
187
188	BUG_ON(vaddr != (unsigned long)GHES_IOREMAP_IRQ_PAGE(base));
189	unmap_kernel_range_noflush(vaddr, PAGE_SIZE);
190	arch_apei_flush_tlb_one(vaddr);
191}
192
193static int ghes_estatus_pool_init(void)
194{
195	ghes_estatus_pool = gen_pool_create(GHES_ESTATUS_POOL_MIN_ALLOC_ORDER, -1);
196	if (!ghes_estatus_pool)
197		return -ENOMEM;
198	return 0;
199}
200
201static void ghes_estatus_pool_free_chunk_page(struct gen_pool *pool,
202					      struct gen_pool_chunk *chunk,
203					      void *data)
204{
205	free_page(chunk->start_addr);
206}
207
208static void ghes_estatus_pool_exit(void)
209{
210	gen_pool_for_each_chunk(ghes_estatus_pool,
211				ghes_estatus_pool_free_chunk_page, NULL);
212	gen_pool_destroy(ghes_estatus_pool);
213}
214
215static int ghes_estatus_pool_expand(unsigned long len)
216{
217	unsigned long i, pages, size, addr;
218	int ret;
219
220	ghes_estatus_pool_size_request += PAGE_ALIGN(len);
221	size = gen_pool_size(ghes_estatus_pool);
222	if (size >= ghes_estatus_pool_size_request)
223		return 0;
224	pages = (ghes_estatus_pool_size_request - size) / PAGE_SIZE;
225	for (i = 0; i < pages; i++) {
226		addr = __get_free_page(GFP_KERNEL);
227		if (!addr)
228			return -ENOMEM;
229		ret = gen_pool_add(ghes_estatus_pool, addr, PAGE_SIZE, -1);
230		if (ret)
231			return ret;
232	}
233
234	return 0;
235}
236
237static struct ghes *ghes_new(struct acpi_hest_generic *generic)
238{
239	struct ghes *ghes;
240	unsigned int error_block_length;
241	int rc;
242
243	ghes = kzalloc(sizeof(*ghes), GFP_KERNEL);
244	if (!ghes)
245		return ERR_PTR(-ENOMEM);
246	ghes->generic = generic;
247	rc = apei_map_generic_address(&generic->error_status_address);
248	if (rc)
249		goto err_free;
250	error_block_length = generic->error_block_length;
251	if (error_block_length > GHES_ESTATUS_MAX_SIZE) {
252		pr_warning(FW_WARN GHES_PFX
253			   "Error status block length is too long: %u for "
254			   "generic hardware error source: %d.\n",
255			   error_block_length, generic->header.source_id);
256		error_block_length = GHES_ESTATUS_MAX_SIZE;
257	}
258	ghes->estatus = kmalloc(error_block_length, GFP_KERNEL);
259	if (!ghes->estatus) {
260		rc = -ENOMEM;
261		goto err_unmap;
262	}
263
264	return ghes;
265
266err_unmap:
267	apei_unmap_generic_address(&generic->error_status_address);
268err_free:
269	kfree(ghes);
270	return ERR_PTR(rc);
271}
272
273static void ghes_fini(struct ghes *ghes)
274{
275	kfree(ghes->estatus);
276	apei_unmap_generic_address(&ghes->generic->error_status_address);
277}
278
279static inline int ghes_severity(int severity)
280{
281	switch (severity) {
282	case CPER_SEV_INFORMATIONAL:
283		return GHES_SEV_NO;
284	case CPER_SEV_CORRECTED:
285		return GHES_SEV_CORRECTED;
286	case CPER_SEV_RECOVERABLE:
287		return GHES_SEV_RECOVERABLE;
288	case CPER_SEV_FATAL:
289		return GHES_SEV_PANIC;
290	default:
291		/* Unknown, go panic */
292		return GHES_SEV_PANIC;
293	}
294}
295
296static void ghes_copy_tofrom_phys(void *buffer, u64 paddr, u32 len,
297				  int from_phys)
298{
299	void __iomem *vaddr;
300	unsigned long flags = 0;
301	int in_nmi = in_nmi();
302	u64 offset;
303	u32 trunk;
304
305	while (len > 0) {
306		offset = paddr - (paddr & PAGE_MASK);
307		if (in_nmi) {
308			raw_spin_lock(&ghes_ioremap_lock_nmi);
309			vaddr = ghes_ioremap_pfn_nmi(paddr >> PAGE_SHIFT);
310		} else {
311			spin_lock_irqsave(&ghes_ioremap_lock_irq, flags);
312			vaddr = ghes_ioremap_pfn_irq(paddr >> PAGE_SHIFT);
313		}
314		trunk = PAGE_SIZE - offset;
315		trunk = min(trunk, len);
316		if (from_phys)
317			memcpy_fromio(buffer, vaddr + offset, trunk);
318		else
319			memcpy_toio(vaddr + offset, buffer, trunk);
320		len -= trunk;
321		paddr += trunk;
322		buffer += trunk;
323		if (in_nmi) {
324			ghes_iounmap_nmi(vaddr);
325			raw_spin_unlock(&ghes_ioremap_lock_nmi);
326		} else {
327			ghes_iounmap_irq(vaddr);
328			spin_unlock_irqrestore(&ghes_ioremap_lock_irq, flags);
329		}
330	}
331}
332
333static int ghes_read_estatus(struct ghes *ghes, int silent)
334{
335	struct acpi_hest_generic *g = ghes->generic;
336	u64 buf_paddr;
337	u32 len;
338	int rc;
339
340	rc = apei_read(&buf_paddr, &g->error_status_address);
341	if (rc) {
342		if (!silent && printk_ratelimit())
343			pr_warning(FW_WARN GHES_PFX
344"Failed to read error status block address for hardware error source: %d.\n",
345				   g->header.source_id);
346		return -EIO;
347	}
348	if (!buf_paddr)
349		return -ENOENT;
350
351	ghes_copy_tofrom_phys(ghes->estatus, buf_paddr,
352			      sizeof(*ghes->estatus), 1);
353	if (!ghes->estatus->block_status)
354		return -ENOENT;
355
356	ghes->buffer_paddr = buf_paddr;
357	ghes->flags |= GHES_TO_CLEAR;
358
359	rc = -EIO;
360	len = cper_estatus_len(ghes->estatus);
361	if (len < sizeof(*ghes->estatus))
362		goto err_read_block;
363	if (len > ghes->generic->error_block_length)
364		goto err_read_block;
365	if (cper_estatus_check_header(ghes->estatus))
366		goto err_read_block;
367	ghes_copy_tofrom_phys(ghes->estatus + 1,
368			      buf_paddr + sizeof(*ghes->estatus),
369			      len - sizeof(*ghes->estatus), 1);
370	if (cper_estatus_check(ghes->estatus))
371		goto err_read_block;
372	rc = 0;
373
374err_read_block:
375	if (rc && !silent && printk_ratelimit())
376		pr_warning(FW_WARN GHES_PFX
377			   "Failed to read error status block!\n");
378	return rc;
379}
380
381static void ghes_clear_estatus(struct ghes *ghes)
382{
383	ghes->estatus->block_status = 0;
384	if (!(ghes->flags & GHES_TO_CLEAR))
385		return;
386	ghes_copy_tofrom_phys(ghes->estatus, ghes->buffer_paddr,
387			      sizeof(ghes->estatus->block_status), 0);
388	ghes->flags &= ~GHES_TO_CLEAR;
389}
390
391static void ghes_handle_memory_failure(struct acpi_hest_generic_data *gdata, int sev)
392{
393#ifdef CONFIG_ACPI_APEI_MEMORY_FAILURE
394	unsigned long pfn;
395	int flags = -1;
396	int sec_sev = ghes_severity(gdata->error_severity);
397	struct cper_sec_mem_err *mem_err;
398	mem_err = (struct cper_sec_mem_err *)(gdata + 1);
399
400	if (!(mem_err->validation_bits & CPER_MEM_VALID_PA))
401		return;
402
403	pfn = mem_err->physical_addr >> PAGE_SHIFT;
404	if (!pfn_valid(pfn)) {
405		pr_warn_ratelimited(FW_WARN GHES_PFX
406		"Invalid address in generic error data: %#llx\n",
407		mem_err->physical_addr);
408		return;
409	}
410
411	/* iff following two events can be handled properly by now */
412	if (sec_sev == GHES_SEV_CORRECTED &&
413	    (gdata->flags & CPER_SEC_ERROR_THRESHOLD_EXCEEDED))
414		flags = MF_SOFT_OFFLINE;
415	if (sev == GHES_SEV_RECOVERABLE && sec_sev == GHES_SEV_RECOVERABLE)
416		flags = 0;
417
418	if (flags != -1)
419		memory_failure_queue(pfn, 0, flags);
420#endif
421}
422
423static void ghes_do_proc(struct ghes *ghes,
424			 const struct acpi_hest_generic_status *estatus)
425{
426	int sev, sec_sev;
427	struct acpi_hest_generic_data *gdata;
428
429	sev = ghes_severity(estatus->error_severity);
430	apei_estatus_for_each_section(estatus, gdata) {
431		sec_sev = ghes_severity(gdata->error_severity);
432		if (!uuid_le_cmp(*(uuid_le *)gdata->section_type,
433				 CPER_SEC_PLATFORM_MEM)) {
434			struct cper_sec_mem_err *mem_err;
435			mem_err = (struct cper_sec_mem_err *)(gdata+1);
436			ghes_edac_report_mem_error(ghes, sev, mem_err);
437
438			arch_apei_report_mem_error(sev, mem_err);
439			ghes_handle_memory_failure(gdata, sev);
440		}
441#ifdef CONFIG_ACPI_APEI_PCIEAER
442		else if (!uuid_le_cmp(*(uuid_le *)gdata->section_type,
443				      CPER_SEC_PCIE)) {
444			struct cper_sec_pcie *pcie_err;
445			pcie_err = (struct cper_sec_pcie *)(gdata+1);
446			if (sev == GHES_SEV_RECOVERABLE &&
447			    sec_sev == GHES_SEV_RECOVERABLE &&
448			    pcie_err->validation_bits & CPER_PCIE_VALID_DEVICE_ID &&
449			    pcie_err->validation_bits & CPER_PCIE_VALID_AER_INFO) {
450				unsigned int devfn;
451				int aer_severity;
452
453				devfn = PCI_DEVFN(pcie_err->device_id.device,
454						  pcie_err->device_id.function);
455				aer_severity = cper_severity_to_aer(sev);
456
457				/*
458				 * If firmware reset the component to contain
459				 * the error, we must reinitialize it before
460				 * use, so treat it as a fatal AER error.
461				 */
462				if (gdata->flags & CPER_SEC_RESET)
463					aer_severity = AER_FATAL;
464
465				aer_recover_queue(pcie_err->device_id.segment,
466						  pcie_err->device_id.bus,
467						  devfn, aer_severity,
468						  (struct aer_capability_regs *)
469						  pcie_err->aer_info);
470			}
471
472		}
473#endif
474	}
475}
476
477static void __ghes_print_estatus(const char *pfx,
478				 const struct acpi_hest_generic *generic,
479				 const struct acpi_hest_generic_status *estatus)
480{
481	static atomic_t seqno;
482	unsigned int curr_seqno;
483	char pfx_seq[64];
484
485	if (pfx == NULL) {
486		if (ghes_severity(estatus->error_severity) <=
487		    GHES_SEV_CORRECTED)
488			pfx = KERN_WARNING;
489		else
490			pfx = KERN_ERR;
491	}
492	curr_seqno = atomic_inc_return(&seqno);
493	snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}" HW_ERR, pfx, curr_seqno);
494	printk("%s""Hardware error from APEI Generic Hardware Error Source: %d\n",
495	       pfx_seq, generic->header.source_id);
496	cper_estatus_print(pfx_seq, estatus);
497}
498
499static int ghes_print_estatus(const char *pfx,
500			      const struct acpi_hest_generic *generic,
501			      const struct acpi_hest_generic_status *estatus)
502{
503	/* Not more than 2 messages every 5 seconds */
504	static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2);
505	static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2);
506	struct ratelimit_state *ratelimit;
507
508	if (ghes_severity(estatus->error_severity) <= GHES_SEV_CORRECTED)
509		ratelimit = &ratelimit_corrected;
510	else
511		ratelimit = &ratelimit_uncorrected;
512	if (__ratelimit(ratelimit)) {
513		__ghes_print_estatus(pfx, generic, estatus);
514		return 1;
515	}
516	return 0;
517}
518
519/*
520 * GHES error status reporting throttle, to report more kinds of
521 * errors, instead of just most frequently occurred errors.
522 */
523static int ghes_estatus_cached(struct acpi_hest_generic_status *estatus)
524{
525	u32 len;
526	int i, cached = 0;
527	unsigned long long now;
528	struct ghes_estatus_cache *cache;
529	struct acpi_hest_generic_status *cache_estatus;
530
531	len = cper_estatus_len(estatus);
532	rcu_read_lock();
533	for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
534		cache = rcu_dereference(ghes_estatus_caches[i]);
535		if (cache == NULL)
536			continue;
537		if (len != cache->estatus_len)
538			continue;
539		cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
540		if (memcmp(estatus, cache_estatus, len))
541			continue;
542		atomic_inc(&cache->count);
543		now = sched_clock();
544		if (now - cache->time_in < GHES_ESTATUS_IN_CACHE_MAX_NSEC)
545			cached = 1;
546		break;
547	}
548	rcu_read_unlock();
549	return cached;
550}
551
552static struct ghes_estatus_cache *ghes_estatus_cache_alloc(
553	struct acpi_hest_generic *generic,
554	struct acpi_hest_generic_status *estatus)
555{
556	int alloced;
557	u32 len, cache_len;
558	struct ghes_estatus_cache *cache;
559	struct acpi_hest_generic_status *cache_estatus;
560
561	alloced = atomic_add_return(1, &ghes_estatus_cache_alloced);
562	if (alloced > GHES_ESTATUS_CACHE_ALLOCED_MAX) {
563		atomic_dec(&ghes_estatus_cache_alloced);
564		return NULL;
565	}
566	len = cper_estatus_len(estatus);
567	cache_len = GHES_ESTATUS_CACHE_LEN(len);
568	cache = (void *)gen_pool_alloc(ghes_estatus_pool, cache_len);
569	if (!cache) {
570		atomic_dec(&ghes_estatus_cache_alloced);
571		return NULL;
572	}
573	cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
574	memcpy(cache_estatus, estatus, len);
575	cache->estatus_len = len;
576	atomic_set(&cache->count, 0);
577	cache->generic = generic;
578	cache->time_in = sched_clock();
579	return cache;
580}
581
582static void ghes_estatus_cache_free(struct ghes_estatus_cache *cache)
583{
584	u32 len;
585
586	len = cper_estatus_len(GHES_ESTATUS_FROM_CACHE(cache));
587	len = GHES_ESTATUS_CACHE_LEN(len);
588	gen_pool_free(ghes_estatus_pool, (unsigned long)cache, len);
589	atomic_dec(&ghes_estatus_cache_alloced);
590}
591
592static void ghes_estatus_cache_rcu_free(struct rcu_head *head)
593{
594	struct ghes_estatus_cache *cache;
595
596	cache = container_of(head, struct ghes_estatus_cache, rcu);
597	ghes_estatus_cache_free(cache);
598}
599
600static void ghes_estatus_cache_add(
601	struct acpi_hest_generic *generic,
602	struct acpi_hest_generic_status *estatus)
603{
604	int i, slot = -1, count;
605	unsigned long long now, duration, period, max_period = 0;
606	struct ghes_estatus_cache *cache, *slot_cache = NULL, *new_cache;
607
608	new_cache = ghes_estatus_cache_alloc(generic, estatus);
609	if (new_cache == NULL)
610		return;
611	rcu_read_lock();
612	now = sched_clock();
613	for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
614		cache = rcu_dereference(ghes_estatus_caches[i]);
615		if (cache == NULL) {
616			slot = i;
617			slot_cache = NULL;
618			break;
619		}
620		duration = now - cache->time_in;
621		if (duration >= GHES_ESTATUS_IN_CACHE_MAX_NSEC) {
622			slot = i;
623			slot_cache = cache;
624			break;
625		}
626		count = atomic_read(&cache->count);
627		period = duration;
628		do_div(period, (count + 1));
629		if (period > max_period) {
630			max_period = period;
631			slot = i;
632			slot_cache = cache;
633		}
634	}
635	/* new_cache must be put into array after its contents are written */
636	smp_wmb();
637	if (slot != -1 && cmpxchg(ghes_estatus_caches + slot,
638				  slot_cache, new_cache) == slot_cache) {
639		if (slot_cache)
640			call_rcu(&slot_cache->rcu, ghes_estatus_cache_rcu_free);
641	} else
642		ghes_estatus_cache_free(new_cache);
643	rcu_read_unlock();
644}
645
646static int ghes_proc(struct ghes *ghes)
647{
648	int rc;
649
650	rc = ghes_read_estatus(ghes, 0);
651	if (rc)
652		goto out;
653	if (!ghes_estatus_cached(ghes->estatus)) {
654		if (ghes_print_estatus(NULL, ghes->generic, ghes->estatus))
655			ghes_estatus_cache_add(ghes->generic, ghes->estatus);
656	}
657	ghes_do_proc(ghes, ghes->estatus);
658out:
659	ghes_clear_estatus(ghes);
660	return 0;
661}
662
663static void ghes_add_timer(struct ghes *ghes)
664{
665	struct acpi_hest_generic *g = ghes->generic;
666	unsigned long expire;
667
668	if (!g->notify.poll_interval) {
669		pr_warning(FW_WARN GHES_PFX "Poll interval is 0 for generic hardware error source: %d, disabled.\n",
670			   g->header.source_id);
671		return;
672	}
673	expire = jiffies + msecs_to_jiffies(g->notify.poll_interval);
674	ghes->timer.expires = round_jiffies_relative(expire);
675	add_timer(&ghes->timer);
676}
677
678static void ghes_poll_func(unsigned long data)
679{
680	struct ghes *ghes = (void *)data;
681
682	ghes_proc(ghes);
683	if (!(ghes->flags & GHES_EXITING))
684		ghes_add_timer(ghes);
685}
686
687static irqreturn_t ghes_irq_func(int irq, void *data)
688{
689	struct ghes *ghes = data;
690	int rc;
691
692	rc = ghes_proc(ghes);
693	if (rc)
694		return IRQ_NONE;
695
696	return IRQ_HANDLED;
697}
698
699static int ghes_notify_sci(struct notifier_block *this,
700				  unsigned long event, void *data)
701{
702	struct ghes *ghes;
703	int ret = NOTIFY_DONE;
704
705	rcu_read_lock();
706	list_for_each_entry_rcu(ghes, &ghes_sci, list) {
707		if (!ghes_proc(ghes))
708			ret = NOTIFY_OK;
709	}
710	rcu_read_unlock();
711
712	return ret;
713}
714
715static struct notifier_block ghes_notifier_sci = {
716	.notifier_call = ghes_notify_sci,
717};
718
719#ifdef CONFIG_HAVE_ACPI_APEI_NMI
720/*
721 * printk is not safe in NMI context.  So in NMI handler, we allocate
722 * required memory from lock-less memory allocator
723 * (ghes_estatus_pool), save estatus into it, put them into lock-less
724 * list (ghes_estatus_llist), then delay printk into IRQ context via
725 * irq_work (ghes_proc_irq_work).  ghes_estatus_size_request record
726 * required pool size by all NMI error source.
727 */
728static struct llist_head ghes_estatus_llist;
729static struct irq_work ghes_proc_irq_work;
730
731/*
732 * NMI may be triggered on any CPU, so ghes_nmi_lock is used for
733 * mutual exclusion.
734 */
735static DEFINE_RAW_SPINLOCK(ghes_nmi_lock);
736
737static LIST_HEAD(ghes_nmi);
738
739static int ghes_panic_timeout	__read_mostly = 30;
740
741static void ghes_proc_in_irq(struct irq_work *irq_work)
742{
743	struct llist_node *llnode, *next;
744	struct ghes_estatus_node *estatus_node;
745	struct acpi_hest_generic *generic;
746	struct acpi_hest_generic_status *estatus;
747	u32 len, node_len;
748
749	llnode = llist_del_all(&ghes_estatus_llist);
750	/*
751	 * Because the time order of estatus in list is reversed,
752	 * revert it back to proper order.
753	 */
754	llnode = llist_reverse_order(llnode);
755	while (llnode) {
756		next = llnode->next;
757		estatus_node = llist_entry(llnode, struct ghes_estatus_node,
758					   llnode);
759		estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
760		len = cper_estatus_len(estatus);
761		node_len = GHES_ESTATUS_NODE_LEN(len);
762		ghes_do_proc(estatus_node->ghes, estatus);
763		if (!ghes_estatus_cached(estatus)) {
764			generic = estatus_node->generic;
765			if (ghes_print_estatus(NULL, generic, estatus))
766				ghes_estatus_cache_add(generic, estatus);
767		}
768		gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node,
769			      node_len);
770		llnode = next;
771	}
772}
773
774static void ghes_print_queued_estatus(void)
775{
776	struct llist_node *llnode;
777	struct ghes_estatus_node *estatus_node;
778	struct acpi_hest_generic *generic;
779	struct acpi_hest_generic_status *estatus;
780	u32 len, node_len;
781
782	llnode = llist_del_all(&ghes_estatus_llist);
783	/*
784	 * Because the time order of estatus in list is reversed,
785	 * revert it back to proper order.
786	 */
787	llnode = llist_reverse_order(llnode);
788	while (llnode) {
789		estatus_node = llist_entry(llnode, struct ghes_estatus_node,
790					   llnode);
791		estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
792		len = cper_estatus_len(estatus);
793		node_len = GHES_ESTATUS_NODE_LEN(len);
794		generic = estatus_node->generic;
795		ghes_print_estatus(NULL, generic, estatus);
796		llnode = llnode->next;
797	}
798}
799
800static int ghes_notify_nmi(unsigned int cmd, struct pt_regs *regs)
801{
802	struct ghes *ghes, *ghes_global = NULL;
803	int sev, sev_global = -1;
804	int ret = NMI_DONE;
805
806	raw_spin_lock(&ghes_nmi_lock);
807	list_for_each_entry_rcu(ghes, &ghes_nmi, list) {
808		if (ghes_read_estatus(ghes, 1)) {
809			ghes_clear_estatus(ghes);
810			continue;
811		}
812		sev = ghes_severity(ghes->estatus->error_severity);
813		if (sev > sev_global) {
814			sev_global = sev;
815			ghes_global = ghes;
816		}
817		ret = NMI_HANDLED;
818	}
819
820	if (ret == NMI_DONE)
821		goto out;
822
823	if (sev_global >= GHES_SEV_PANIC) {
824		oops_begin();
825		ghes_print_queued_estatus();
826		__ghes_print_estatus(KERN_EMERG, ghes_global->generic,
827				     ghes_global->estatus);
828		/* reboot to log the error! */
829		if (panic_timeout == 0)
830			panic_timeout = ghes_panic_timeout;
831		panic("Fatal hardware error!");
832	}
833
834	list_for_each_entry_rcu(ghes, &ghes_nmi, list) {
835#ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
836		u32 len, node_len;
837		struct ghes_estatus_node *estatus_node;
838		struct acpi_hest_generic_status *estatus;
839#endif
840		if (!(ghes->flags & GHES_TO_CLEAR))
841			continue;
842#ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
843		if (ghes_estatus_cached(ghes->estatus))
844			goto next;
845		/* Save estatus for further processing in IRQ context */
846		len = cper_estatus_len(ghes->estatus);
847		node_len = GHES_ESTATUS_NODE_LEN(len);
848		estatus_node = (void *)gen_pool_alloc(ghes_estatus_pool,
849						      node_len);
850		if (estatus_node) {
851			estatus_node->ghes = ghes;
852			estatus_node->generic = ghes->generic;
853			estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
854			memcpy(estatus, ghes->estatus, len);
855			llist_add(&estatus_node->llnode, &ghes_estatus_llist);
856		}
857next:
858#endif
859		ghes_clear_estatus(ghes);
860	}
861#ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
862	irq_work_queue(&ghes_proc_irq_work);
863#endif
864
865out:
866	raw_spin_unlock(&ghes_nmi_lock);
867	return ret;
868}
869
870static unsigned long ghes_esource_prealloc_size(
871	const struct acpi_hest_generic *generic)
872{
873	unsigned long block_length, prealloc_records, prealloc_size;
874
875	block_length = min_t(unsigned long, generic->error_block_length,
876			     GHES_ESTATUS_MAX_SIZE);
877	prealloc_records = max_t(unsigned long,
878				 generic->records_to_preallocate, 1);
879	prealloc_size = min_t(unsigned long, block_length * prealloc_records,
880			      GHES_ESOURCE_PREALLOC_MAX_SIZE);
881
882	return prealloc_size;
883}
884
885static void ghes_estatus_pool_shrink(unsigned long len)
886{
887	ghes_estatus_pool_size_request -= PAGE_ALIGN(len);
888}
889
890static void ghes_nmi_add(struct ghes *ghes)
891{
892	unsigned long len;
893
894	len = ghes_esource_prealloc_size(ghes->generic);
895	ghes_estatus_pool_expand(len);
896	mutex_lock(&ghes_list_mutex);
897	if (list_empty(&ghes_nmi))
898		register_nmi_handler(NMI_LOCAL, ghes_notify_nmi, 0, "ghes");
899	list_add_rcu(&ghes->list, &ghes_nmi);
900	mutex_unlock(&ghes_list_mutex);
901}
902
903static void ghes_nmi_remove(struct ghes *ghes)
904{
905	unsigned long len;
906
907	mutex_lock(&ghes_list_mutex);
908	list_del_rcu(&ghes->list);
909	if (list_empty(&ghes_nmi))
910		unregister_nmi_handler(NMI_LOCAL, "ghes");
911	mutex_unlock(&ghes_list_mutex);
912	/*
913	 * To synchronize with NMI handler, ghes can only be
914	 * freed after NMI handler finishes.
915	 */
916	synchronize_rcu();
917	len = ghes_esource_prealloc_size(ghes->generic);
918	ghes_estatus_pool_shrink(len);
919}
920
921static void ghes_nmi_init_cxt(void)
922{
923	init_irq_work(&ghes_proc_irq_work, ghes_proc_in_irq);
924}
925#else /* CONFIG_HAVE_ACPI_APEI_NMI */
926static inline void ghes_nmi_add(struct ghes *ghes)
927{
928	pr_err(GHES_PFX "ID: %d, trying to add NMI notification which is not supported!\n",
929	       ghes->generic->header.source_id);
930	BUG();
931}
932
933static inline void ghes_nmi_remove(struct ghes *ghes)
934{
935	pr_err(GHES_PFX "ID: %d, trying to remove NMI notification which is not supported!\n",
936	       ghes->generic->header.source_id);
937	BUG();
938}
939
940static inline void ghes_nmi_init_cxt(void)
941{
942}
943#endif /* CONFIG_HAVE_ACPI_APEI_NMI */
944
945static int ghes_probe(struct platform_device *ghes_dev)
946{
947	struct acpi_hest_generic *generic;
948	struct ghes *ghes = NULL;
949
950	int rc = -EINVAL;
951
952	generic = *(struct acpi_hest_generic **)ghes_dev->dev.platform_data;
953	if (!generic->enabled)
954		return -ENODEV;
955
956	switch (generic->notify.type) {
957	case ACPI_HEST_NOTIFY_POLLED:
958	case ACPI_HEST_NOTIFY_EXTERNAL:
959	case ACPI_HEST_NOTIFY_SCI:
960		break;
961	case ACPI_HEST_NOTIFY_NMI:
962		if (!IS_ENABLED(CONFIG_HAVE_ACPI_APEI_NMI)) {
963			pr_warn(GHES_PFX "Generic hardware error source: %d notified via NMI interrupt is not supported!\n",
964				generic->header.source_id);
965			goto err;
966		}
967		break;
968	case ACPI_HEST_NOTIFY_LOCAL:
969		pr_warning(GHES_PFX "Generic hardware error source: %d notified via local interrupt is not supported!\n",
970			   generic->header.source_id);
971		goto err;
972	default:
973		pr_warning(FW_WARN GHES_PFX "Unknown notification type: %u for generic hardware error source: %d\n",
974			   generic->notify.type, generic->header.source_id);
975		goto err;
976	}
977
978	rc = -EIO;
979	if (generic->error_block_length <
980	    sizeof(struct acpi_hest_generic_status)) {
981		pr_warning(FW_BUG GHES_PFX "Invalid error block length: %u for generic hardware error source: %d\n",
982			   generic->error_block_length,
983			   generic->header.source_id);
984		goto err;
985	}
986	ghes = ghes_new(generic);
987	if (IS_ERR(ghes)) {
988		rc = PTR_ERR(ghes);
989		ghes = NULL;
990		goto err;
991	}
992
993	rc = ghes_edac_register(ghes, &ghes_dev->dev);
994	if (rc < 0)
995		goto err;
996
997	switch (generic->notify.type) {
998	case ACPI_HEST_NOTIFY_POLLED:
999		ghes->timer.function = ghes_poll_func;
1000		ghes->timer.data = (unsigned long)ghes;
1001		init_timer_deferrable(&ghes->timer);
1002		ghes_add_timer(ghes);
1003		break;
1004	case ACPI_HEST_NOTIFY_EXTERNAL:
1005		/* External interrupt vector is GSI */
1006		rc = acpi_gsi_to_irq(generic->notify.vector, &ghes->irq);
1007		if (rc) {
1008			pr_err(GHES_PFX "Failed to map GSI to IRQ for generic hardware error source: %d\n",
1009			       generic->header.source_id);
1010			goto err_edac_unreg;
1011		}
1012		rc = request_irq(ghes->irq, ghes_irq_func, 0, "GHES IRQ", ghes);
1013		if (rc) {
1014			pr_err(GHES_PFX "Failed to register IRQ for generic hardware error source: %d\n",
1015			       generic->header.source_id);
1016			goto err_edac_unreg;
1017		}
1018		break;
1019	case ACPI_HEST_NOTIFY_SCI:
1020		mutex_lock(&ghes_list_mutex);
1021		if (list_empty(&ghes_sci))
1022			register_acpi_hed_notifier(&ghes_notifier_sci);
1023		list_add_rcu(&ghes->list, &ghes_sci);
1024		mutex_unlock(&ghes_list_mutex);
1025		break;
1026	case ACPI_HEST_NOTIFY_NMI:
1027		ghes_nmi_add(ghes);
1028		break;
1029	default:
1030		BUG();
1031	}
1032	platform_set_drvdata(ghes_dev, ghes);
1033
1034	return 0;
1035err_edac_unreg:
1036	ghes_edac_unregister(ghes);
1037err:
1038	if (ghes) {
1039		ghes_fini(ghes);
1040		kfree(ghes);
1041	}
1042	return rc;
1043}
1044
1045static int ghes_remove(struct platform_device *ghes_dev)
1046{
1047	struct ghes *ghes;
1048	struct acpi_hest_generic *generic;
1049
1050	ghes = platform_get_drvdata(ghes_dev);
1051	generic = ghes->generic;
1052
1053	ghes->flags |= GHES_EXITING;
1054	switch (generic->notify.type) {
1055	case ACPI_HEST_NOTIFY_POLLED:
1056		del_timer_sync(&ghes->timer);
1057		break;
1058	case ACPI_HEST_NOTIFY_EXTERNAL:
1059		free_irq(ghes->irq, ghes);
1060		break;
1061	case ACPI_HEST_NOTIFY_SCI:
1062		mutex_lock(&ghes_list_mutex);
1063		list_del_rcu(&ghes->list);
1064		if (list_empty(&ghes_sci))
1065			unregister_acpi_hed_notifier(&ghes_notifier_sci);
1066		mutex_unlock(&ghes_list_mutex);
1067		break;
1068	case ACPI_HEST_NOTIFY_NMI:
1069		ghes_nmi_remove(ghes);
1070		break;
1071	default:
1072		BUG();
1073		break;
1074	}
1075
1076	ghes_fini(ghes);
1077
1078	ghes_edac_unregister(ghes);
1079
1080	kfree(ghes);
1081
1082	platform_set_drvdata(ghes_dev, NULL);
1083
1084	return 0;
1085}
1086
1087static struct platform_driver ghes_platform_driver = {
1088	.driver		= {
1089		.name	= "GHES",
1090	},
1091	.probe		= ghes_probe,
1092	.remove		= ghes_remove,
1093};
1094
1095static int __init ghes_init(void)
1096{
1097	int rc;
1098
1099	if (acpi_disabled)
1100		return -ENODEV;
1101
1102	if (hest_disable) {
1103		pr_info(GHES_PFX "HEST is not enabled!\n");
1104		return -EINVAL;
1105	}
1106
1107	if (ghes_disable) {
1108		pr_info(GHES_PFX "GHES is not enabled!\n");
1109		return -EINVAL;
1110	}
1111
1112	ghes_nmi_init_cxt();
1113
1114	rc = ghes_ioremap_init();
1115	if (rc)
1116		goto err;
1117
1118	rc = ghes_estatus_pool_init();
1119	if (rc)
1120		goto err_ioremap_exit;
1121
1122	rc = ghes_estatus_pool_expand(GHES_ESTATUS_CACHE_AVG_SIZE *
1123				      GHES_ESTATUS_CACHE_ALLOCED_MAX);
1124	if (rc)
1125		goto err_pool_exit;
1126
1127	rc = platform_driver_register(&ghes_platform_driver);
1128	if (rc)
1129		goto err_pool_exit;
1130
1131	rc = apei_osc_setup();
1132	if (rc == 0 && osc_sb_apei_support_acked)
1133		pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit and WHEA _OSC.\n");
1134	else if (rc == 0 && !osc_sb_apei_support_acked)
1135		pr_info(GHES_PFX "APEI firmware first mode is enabled by WHEA _OSC.\n");
1136	else if (rc && osc_sb_apei_support_acked)
1137		pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit.\n");
1138	else
1139		pr_info(GHES_PFX "Failed to enable APEI firmware first mode.\n");
1140
1141	return 0;
1142err_pool_exit:
1143	ghes_estatus_pool_exit();
1144err_ioremap_exit:
1145	ghes_ioremap_exit();
1146err:
1147	return rc;
1148}
1149
1150static void __exit ghes_exit(void)
1151{
1152	platform_driver_unregister(&ghes_platform_driver);
1153	ghes_estatus_pool_exit();
1154	ghes_ioremap_exit();
1155}
1156
1157module_init(ghes_init);
1158module_exit(ghes_exit);
1159
1160MODULE_AUTHOR("Huang Ying");
1161MODULE_DESCRIPTION("APEI Generic Hardware Error Source support");
1162MODULE_LICENSE("GPL");
1163MODULE_ALIAS("platform:GHES");
1164