Linux Audio

Check our new training course

Loading...
v6.13.7
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * APEI Generic Hardware Error Source support
   4 *
   5 * Generic Hardware Error Source provides a way to report platform
   6 * hardware errors (such as that from chipset). It works in so called
   7 * "Firmware First" mode, that is, hardware errors are reported to
   8 * firmware firstly, then reported to Linux by firmware. This way,
   9 * some non-standard hardware error registers or non-standard hardware
  10 * link can be checked by firmware to produce more hardware error
  11 * information for Linux.
  12 *
  13 * For more information about Generic Hardware Error Source, please
  14 * refer to ACPI Specification version 4.0, section 17.3.2.6
  15 *
  16 * Copyright 2010,2011 Intel Corp.
  17 *   Author: Huang Ying <ying.huang@intel.com>
 
 
 
 
 
 
 
 
 
 
 
 
 
  18 */
  19
  20#include <linux/arm_sdei.h>
  21#include <linux/kernel.h>
  22#include <linux/moduleparam.h>
  23#include <linux/init.h>
  24#include <linux/acpi.h>
 
  25#include <linux/io.h>
  26#include <linux/interrupt.h>
  27#include <linux/timer.h>
  28#include <linux/cper.h>
  29#include <linux/cleanup.h>
  30#include <linux/platform_device.h>
  31#include <linux/mutex.h>
  32#include <linux/ratelimit.h>
  33#include <linux/vmalloc.h>
  34#include <linux/irq_work.h>
  35#include <linux/llist.h>
  36#include <linux/genalloc.h>
  37#include <linux/kfifo.h>
  38#include <linux/pci.h>
  39#include <linux/pfn.h>
  40#include <linux/aer.h>
  41#include <linux/nmi.h>
  42#include <linux/sched/clock.h>
  43#include <linux/uuid.h>
  44#include <linux/ras.h>
  45#include <linux/task_work.h>
  46
  47#include <acpi/actbl1.h>
  48#include <acpi/ghes.h>
  49#include <acpi/apei.h>
  50#include <asm/fixmap.h>
 
  51#include <asm/tlbflush.h>
  52#include <cxl/event.h>
  53#include <ras/ras_event.h>
  54
  55#include "apei-internal.h"
  56
  57#define GHES_PFX	"GHES: "
  58
  59#define GHES_ESTATUS_MAX_SIZE		65536
  60#define GHES_ESOURCE_PREALLOC_MAX_SIZE	65536
  61
  62#define GHES_ESTATUS_POOL_MIN_ALLOC_ORDER 3
  63
  64/* This is just an estimation for memory pool allocation */
  65#define GHES_ESTATUS_CACHE_AVG_SIZE	512
  66
  67#define GHES_ESTATUS_CACHES_SIZE	4
  68
  69#define GHES_ESTATUS_IN_CACHE_MAX_NSEC	10000000000ULL
  70/* Prevent too many caches are allocated because of RCU */
  71#define GHES_ESTATUS_CACHE_ALLOCED_MAX	(GHES_ESTATUS_CACHES_SIZE * 3 / 2)
  72
  73#define GHES_ESTATUS_CACHE_LEN(estatus_len)			\
  74	(sizeof(struct ghes_estatus_cache) + (estatus_len))
  75#define GHES_ESTATUS_FROM_CACHE(estatus_cache)			\
  76	((struct acpi_hest_generic_status *)				\
  77	 ((struct ghes_estatus_cache *)(estatus_cache) + 1))
  78
  79#define GHES_ESTATUS_NODE_LEN(estatus_len)			\
  80	(sizeof(struct ghes_estatus_node) + (estatus_len))
  81#define GHES_ESTATUS_FROM_NODE(estatus_node)			\
  82	((struct acpi_hest_generic_status *)				\
  83	 ((struct ghes_estatus_node *)(estatus_node) + 1))
  84
  85#define GHES_VENDOR_ENTRY_LEN(gdata_len)                               \
  86	(sizeof(struct ghes_vendor_record_entry) + (gdata_len))
  87#define GHES_GDATA_FROM_VENDOR_ENTRY(vendor_entry)                     \
  88	((struct acpi_hest_generic_data *)                              \
  89	((struct ghes_vendor_record_entry *)(vendor_entry) + 1))
  90
  91/*
  92 *  NMI-like notifications vary by architecture, before the compiler can prune
  93 *  unused static functions it needs a value for these enums.
 
 
 
 
  94 */
  95#ifndef CONFIG_ARM_SDE_INTERFACE
  96#define FIX_APEI_GHES_SDEI_NORMAL	__end_of_fixed_addresses
  97#define FIX_APEI_GHES_SDEI_CRITICAL	__end_of_fixed_addresses
  98#endif
  99
 100static ATOMIC_NOTIFIER_HEAD(ghes_report_chain);
 101
 102static inline bool is_hest_type_generic_v2(struct ghes *ghes)
 103{
 104	return ghes->generic->header.type == ACPI_HEST_TYPE_GENERIC_ERROR_V2;
 105}
 
 
 
 
 
 106
 107/*
 108 * A platform may describe one error source for the handling of synchronous
 109 * errors (e.g. MCE or SEA), or for handling asynchronous errors (e.g. SCI
 110 * or External Interrupt). On x86, the HEST notifications are always
 111 * asynchronous, so only SEA on ARM is delivered as a synchronous
 112 * notification.
 113 */
 114static inline bool is_hest_sync_notify(struct ghes *ghes)
 115{
 116	u8 notify_type = ghes->generic->notify.type;
 117
 118	return notify_type == ACPI_HEST_NOTIFY_SEA;
 119}
 
 
 
 
 
 120
 121/*
 122 * This driver isn't really modular, however for the time being,
 123 * continuing to use module_param is the easiest way to remain
 124 * compatible with existing boot arg use cases.
 125 */
 126bool ghes_disable;
 127module_param_named(disable, ghes_disable, bool, 0);
 128
 129/*
 130 * "ghes.edac_force_enable" forcibly enables ghes_edac and skips the platform
 131 * check.
 132 */
 133static bool ghes_edac_force_enable;
 134module_param_named(edac_force_enable, ghes_edac_force_enable, bool, 0);
 135
 136/*
 137 * All error sources notified with HED (Hardware Error Device) share a
 138 * single notifier callback, so they need to be linked and checked one
 139 * by one. This holds true for NMI too.
 140 *
 141 * RCU is used for these lists, so ghes_list_mutex is only used for
 142 * list changing, not for traversing.
 143 */
 144static LIST_HEAD(ghes_hed);
 
 145static DEFINE_MUTEX(ghes_list_mutex);
 146
 147/*
 148 * A list of GHES devices which are given to the corresponding EDAC driver
 149 * ghes_edac for further use.
 150 */
 151static LIST_HEAD(ghes_devs);
 152static DEFINE_MUTEX(ghes_devs_mutex);
 153
 154/*
 155 * Because the memory area used to transfer hardware error information
 156 * from BIOS to Linux can be determined only in NMI, IRQ or timer
 157 * handler, but general ioremap can not be used in atomic context, so
 158 * the fixmap is used instead.
 159 *
 160 * This spinlock is used to prevent the fixmap entry from being used
 161 * simultaneously.
 162 */
 163static DEFINE_SPINLOCK(ghes_notify_lock_irq);
 164
 165struct ghes_vendor_record_entry {
 166	struct work_struct work;
 167	int error_severity;
 168	char vendor_record[];
 169};
 
 
 
 
 
 
 
 
 
 
 
 170
 
 
 
 
 
 
 
 
 171static struct gen_pool *ghes_estatus_pool;
 
 
 
 172
 173static struct ghes_estatus_cache __rcu *ghes_estatus_caches[GHES_ESTATUS_CACHES_SIZE];
 174static atomic_t ghes_estatus_cache_alloced;
 175
 176static void __iomem *ghes_map(u64 pfn, enum fixed_addresses fixmap_idx)
 177{
 178	phys_addr_t paddr;
 179	pgprot_t prot;
 180
 181	paddr = PFN_PHYS(pfn);
 182	prot = arch_apei_get_mem_attribute(paddr);
 183	__set_fixmap(fixmap_idx, paddr, prot);
 184
 185	return (void __iomem *) __fix_to_virt(fixmap_idx);
 186}
 187
 188static void ghes_unmap(void __iomem *vaddr, enum fixed_addresses fixmap_idx)
 189{
 190	int _idx = virt_to_fix((unsigned long)vaddr);
 191
 192	WARN_ON_ONCE(fixmap_idx != _idx);
 193	clear_fixmap(fixmap_idx);
 194}
 195
 196int ghes_estatus_pool_init(unsigned int num_ghes)
 197{
 198	unsigned long addr, len;
 199	int rc;
 200
 201	ghes_estatus_pool = gen_pool_create(GHES_ESTATUS_POOL_MIN_ALLOC_ORDER, -1);
 202	if (!ghes_estatus_pool)
 203		return -ENOMEM;
 204
 205	len = GHES_ESTATUS_CACHE_AVG_SIZE * GHES_ESTATUS_CACHE_ALLOCED_MAX;
 206	len += (num_ghes * GHES_ESOURCE_PREALLOC_MAX_SIZE);
 207
 208	addr = (unsigned long)vmalloc(PAGE_ALIGN(len));
 209	if (!addr)
 210		goto err_pool_alloc;
 211
 212	rc = gen_pool_add(ghes_estatus_pool, addr, PAGE_ALIGN(len), -1);
 213	if (rc)
 214		goto err_pool_add;
 215
 216	return 0;
 
 217
 218err_pool_add:
 219	vfree((void *)addr);
 
 
 220
 221err_pool_alloc:
 222	gen_pool_destroy(ghes_estatus_pool);
 
 
 
 
 
 
 
 223
 224	return -ENOMEM;
 
 
 225}
 226
 227/**
 228 * ghes_estatus_pool_region_free - free previously allocated memory
 229 *				   from the ghes_estatus_pool.
 230 * @addr: address of memory to free.
 231 * @size: size of memory to free.
 232 *
 233 * Returns none.
 234 */
 235void ghes_estatus_pool_region_free(unsigned long addr, u32 size)
 236{
 237	gen_pool_free(ghes_estatus_pool, addr, size);
 
 
 
 238}
 239EXPORT_SYMBOL_GPL(ghes_estatus_pool_region_free);
 240
 241static int map_gen_v2(struct ghes *ghes)
 
 
 242{
 243	return apei_map_generic_address(&ghes->generic_v2->read_ack_register);
 244}
 245
 246static void unmap_gen_v2(struct ghes *ghes)
 247{
 248	apei_unmap_generic_address(&ghes->generic_v2->read_ack_register);
 
 
 249}
 250
 251static void ghes_ack_error(struct acpi_hest_generic_v2 *gv2)
 252{
 253	int rc;
 254	u64 val = 0;
 255
 256	rc = apei_read(&val, &gv2->read_ack_register);
 257	if (rc)
 258		return;
 
 
 
 
 
 
 
 
 
 
 259
 260	val &= gv2->read_ack_preserve << gv2->read_ack_register.bit_offset;
 261	val |= gv2->read_ack_write    << gv2->read_ack_register.bit_offset;
 262
 263	apei_write(val, &gv2->read_ack_register);
 
 
 264}
 265
 266static struct ghes *ghes_new(struct acpi_hest_generic *generic)
 267{
 268	struct ghes *ghes;
 269	unsigned int error_block_length;
 270	int rc;
 271
 272	ghes = kzalloc(sizeof(*ghes), GFP_KERNEL);
 273	if (!ghes)
 274		return ERR_PTR(-ENOMEM);
 275
 276	ghes->generic = generic;
 277	if (is_hest_type_generic_v2(ghes)) {
 278		rc = map_gen_v2(ghes);
 279		if (rc)
 280			goto err_free;
 281	}
 282
 283	rc = apei_map_generic_address(&generic->error_status_address);
 284	if (rc)
 285		goto err_unmap_read_ack_addr;
 286	error_block_length = generic->error_block_length;
 287	if (error_block_length > GHES_ESTATUS_MAX_SIZE) {
 288		pr_warn(FW_WARN GHES_PFX
 289			"Error status block length is too long: %u for "
 290			"generic hardware error source: %d.\n",
 291			error_block_length, generic->header.source_id);
 292		error_block_length = GHES_ESTATUS_MAX_SIZE;
 293	}
 294	ghes->estatus = kmalloc(error_block_length, GFP_KERNEL);
 295	if (!ghes->estatus) {
 296		rc = -ENOMEM;
 297		goto err_unmap_status_addr;
 298	}
 299
 300	return ghes;
 301
 302err_unmap_status_addr:
 303	apei_unmap_generic_address(&generic->error_status_address);
 304err_unmap_read_ack_addr:
 305	if (is_hest_type_generic_v2(ghes))
 306		unmap_gen_v2(ghes);
 307err_free:
 308	kfree(ghes);
 309	return ERR_PTR(rc);
 310}
 311
 312static void ghes_fini(struct ghes *ghes)
 313{
 314	kfree(ghes->estatus);
 315	apei_unmap_generic_address(&ghes->generic->error_status_address);
 316	if (is_hest_type_generic_v2(ghes))
 317		unmap_gen_v2(ghes);
 318}
 319
 
 
 
 
 
 
 
 320static inline int ghes_severity(int severity)
 321{
 322	switch (severity) {
 323	case CPER_SEV_INFORMATIONAL:
 324		return GHES_SEV_NO;
 325	case CPER_SEV_CORRECTED:
 326		return GHES_SEV_CORRECTED;
 327	case CPER_SEV_RECOVERABLE:
 328		return GHES_SEV_RECOVERABLE;
 329	case CPER_SEV_FATAL:
 330		return GHES_SEV_PANIC;
 331	default:
 332		/* Unknown, go panic */
 333		return GHES_SEV_PANIC;
 334	}
 335}
 336
 337static void ghes_copy_tofrom_phys(void *buffer, u64 paddr, u32 len,
 338				  int from_phys,
 339				  enum fixed_addresses fixmap_idx)
 340{
 341	void __iomem *vaddr;
 
 
 342	u64 offset;
 343	u32 trunk;
 344
 345	while (len > 0) {
 346		offset = paddr - (paddr & PAGE_MASK);
 347		vaddr = ghes_map(PHYS_PFN(paddr), fixmap_idx);
 
 
 
 
 
 
 348		trunk = PAGE_SIZE - offset;
 349		trunk = min(trunk, len);
 350		if (from_phys)
 351			memcpy_fromio(buffer, vaddr + offset, trunk);
 352		else
 353			memcpy_toio(vaddr + offset, buffer, trunk);
 354		len -= trunk;
 355		paddr += trunk;
 356		buffer += trunk;
 357		ghes_unmap(vaddr, fixmap_idx);
 358	}
 359}
 360
 361/* Check the top-level record header has an appropriate size. */
 362static int __ghes_check_estatus(struct ghes *ghes,
 363				struct acpi_hest_generic_status *estatus)
 364{
 365	u32 len = cper_estatus_len(estatus);
 366
 367	if (len < sizeof(*estatus)) {
 368		pr_warn_ratelimited(FW_WARN GHES_PFX "Truncated error status block!\n");
 369		return -EIO;
 370	}
 371
 372	if (len > ghes->generic->error_block_length) {
 373		pr_warn_ratelimited(FW_WARN GHES_PFX "Invalid error status block length!\n");
 374		return -EIO;
 375	}
 376
 377	if (cper_estatus_check_header(estatus)) {
 378		pr_warn_ratelimited(FW_WARN GHES_PFX "Invalid CPER header!\n");
 379		return -EIO;
 380	}
 381
 382	return 0;
 383}
 384
 385/* Read the CPER block, returning its address, and header in estatus. */
 386static int __ghes_peek_estatus(struct ghes *ghes,
 387			       struct acpi_hest_generic_status *estatus,
 388			       u64 *buf_paddr, enum fixed_addresses fixmap_idx)
 389{
 390	struct acpi_hest_generic *g = ghes->generic;
 
 
 391	int rc;
 392
 393	rc = apei_read(buf_paddr, &g->error_status_address);
 394	if (rc) {
 395		*buf_paddr = 0;
 396		pr_warn_ratelimited(FW_WARN GHES_PFX
 397"Failed to read error status block address for hardware error source: %d.\n",
 398				   g->header.source_id);
 399		return -EIO;
 400	}
 401	if (!*buf_paddr)
 402		return -ENOENT;
 403
 404	ghes_copy_tofrom_phys(estatus, *buf_paddr, sizeof(*estatus), 1,
 405			      fixmap_idx);
 406	if (!estatus->block_status) {
 407		*buf_paddr = 0;
 408		return -ENOENT;
 409	}
 410
 411	return 0;
 412}
 413
 414static int __ghes_read_estatus(struct acpi_hest_generic_status *estatus,
 415			       u64 buf_paddr, enum fixed_addresses fixmap_idx,
 416			       size_t buf_len)
 417{
 418	ghes_copy_tofrom_phys(estatus, buf_paddr, buf_len, 1, fixmap_idx);
 419	if (cper_estatus_check(estatus)) {
 420		pr_warn_ratelimited(FW_WARN GHES_PFX
 421				    "Failed to read error status block!\n");
 422		return -EIO;
 423	}
 424
 425	return 0;
 426}
 427
 428static int ghes_read_estatus(struct ghes *ghes,
 429			     struct acpi_hest_generic_status *estatus,
 430			     u64 *buf_paddr, enum fixed_addresses fixmap_idx)
 431{
 432	int rc;
 433
 434	rc = __ghes_peek_estatus(ghes, estatus, buf_paddr, fixmap_idx);
 435	if (rc)
 436		return rc;
 437
 438	rc = __ghes_check_estatus(ghes, estatus);
 439	if (rc)
 440		return rc;
 441
 442	return __ghes_read_estatus(estatus, *buf_paddr, fixmap_idx,
 443				   cper_estatus_len(estatus));
 444}
 445
 446static void ghes_clear_estatus(struct ghes *ghes,
 447			       struct acpi_hest_generic_status *estatus,
 448			       u64 buf_paddr, enum fixed_addresses fixmap_idx)
 449{
 450	estatus->block_status = 0;
 451
 452	if (!buf_paddr)
 453		return;
 454
 455	ghes_copy_tofrom_phys(estatus, buf_paddr,
 456			      sizeof(estatus->block_status), 0,
 457			      fixmap_idx);
 458
 459	/*
 460	 * GHESv2 type HEST entries introduce support for error acknowledgment,
 461	 * so only acknowledge the error if this support is present.
 462	 */
 463	if (is_hest_type_generic_v2(ghes))
 464		ghes_ack_error(ghes->generic_v2);
 465}
 466
 467/*
 468 * Called as task_work before returning to user-space.
 469 * Ensure any queued work has been done before we return to the context that
 470 * triggered the notification.
 471 */
 472static void ghes_kick_task_work(struct callback_head *head)
 473{
 474	struct acpi_hest_generic_status *estatus;
 475	struct ghes_estatus_node *estatus_node;
 476	u32 node_len;
 477
 478	estatus_node = container_of(head, struct ghes_estatus_node, task_work);
 479	if (IS_ENABLED(CONFIG_ACPI_APEI_MEMORY_FAILURE))
 480		memory_failure_queue_kick(estatus_node->task_work_cpu);
 481
 482	estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
 483	node_len = GHES_ESTATUS_NODE_LEN(cper_estatus_len(estatus));
 484	gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node, node_len);
 485}
 486
 487static bool ghes_do_memory_failure(u64 physical_addr, int flags)
 488{
 489	unsigned long pfn;
 490
 491	if (!IS_ENABLED(CONFIG_ACPI_APEI_MEMORY_FAILURE))
 492		return false;
 493
 494	pfn = PHYS_PFN(physical_addr);
 495	if (!pfn_valid(pfn) && !arch_is_platform_page(physical_addr)) {
 496		pr_warn_ratelimited(FW_WARN GHES_PFX
 497		"Invalid address in generic error data: %#llx\n",
 498		physical_addr);
 499		return false;
 500	}
 501
 502	memory_failure_queue(pfn, flags);
 503	return true;
 504}
 505
 506static bool ghes_handle_memory_failure(struct acpi_hest_generic_data *gdata,
 507				       int sev, bool sync)
 508{
 509	int flags = -1;
 510	int sec_sev = ghes_severity(gdata->error_severity);
 511	struct cper_sec_mem_err *mem_err = acpi_hest_get_payload(gdata);
 512
 513	if (!(mem_err->validation_bits & CPER_MEM_VALID_PA))
 514		return false;
 515
 516	/* iff following two events can be handled properly by now */
 517	if (sec_sev == GHES_SEV_CORRECTED &&
 518	    (gdata->flags & CPER_SEC_ERROR_THRESHOLD_EXCEEDED))
 519		flags = MF_SOFT_OFFLINE;
 520	if (sev == GHES_SEV_RECOVERABLE && sec_sev == GHES_SEV_RECOVERABLE)
 521		flags = sync ? MF_ACTION_REQUIRED : 0;
 522
 523	if (flags != -1)
 524		return ghes_do_memory_failure(mem_err->physical_addr, flags);
 525
 526	return false;
 527}
 528
 529static bool ghes_handle_arm_hw_error(struct acpi_hest_generic_data *gdata,
 530				       int sev, bool sync)
 531{
 532	struct cper_sec_proc_arm *err = acpi_hest_get_payload(gdata);
 533	int flags = sync ? MF_ACTION_REQUIRED : 0;
 534	bool queued = false;
 535	int sec_sev, i;
 536	char *p;
 537
 538	log_arm_hw_error(err);
 539
 540	sec_sev = ghes_severity(gdata->error_severity);
 541	if (sev != GHES_SEV_RECOVERABLE || sec_sev != GHES_SEV_RECOVERABLE)
 542		return false;
 543
 544	p = (char *)(err + 1);
 545	for (i = 0; i < err->err_info_num; i++) {
 546		struct cper_arm_err_info *err_info = (struct cper_arm_err_info *)p;
 547		bool is_cache = (err_info->type == CPER_ARM_CACHE_ERROR);
 548		bool has_pa = (err_info->validation_bits & CPER_ARM_INFO_VALID_PHYSICAL_ADDR);
 549		const char *error_type = "unknown error";
 550
 551		/*
 552		 * The field (err_info->error_info & BIT(26)) is fixed to set to
 553		 * 1 in some old firmware of HiSilicon Kunpeng920. We assume that
 554		 * firmware won't mix corrected errors in an uncorrected section,
 555		 * and don't filter out 'corrected' error here.
 556		 */
 557		if (is_cache && has_pa) {
 558			queued = ghes_do_memory_failure(err_info->physical_fault_addr, flags);
 559			p += err_info->length;
 560			continue;
 561		}
 562
 563		if (err_info->type < ARRAY_SIZE(cper_proc_error_type_strs))
 564			error_type = cper_proc_error_type_strs[err_info->type];
 565
 566		pr_warn_ratelimited(FW_WARN GHES_PFX
 567				    "Unhandled processor error type: %s\n",
 568				    error_type);
 569		p += err_info->length;
 570	}
 571
 572	return queued;
 573}
 574
 575/*
 576 * PCIe AER errors need to be sent to the AER driver for reporting and
 577 * recovery. The GHES severities map to the following AER severities and
 578 * require the following handling:
 579 *
 580 * GHES_SEV_CORRECTABLE -> AER_CORRECTABLE
 581 *     These need to be reported by the AER driver but no recovery is
 582 *     necessary.
 583 * GHES_SEV_RECOVERABLE -> AER_NONFATAL
 584 * GHES_SEV_RECOVERABLE && CPER_SEC_RESET -> AER_FATAL
 585 *     These both need to be reported and recovered from by the AER driver.
 586 * GHES_SEV_PANIC does not make it to this handling since the kernel must
 587 *     panic.
 588 */
 589static void ghes_handle_aer(struct acpi_hest_generic_data *gdata)
 590{
 591#ifdef CONFIG_ACPI_APEI_PCIEAER
 592	struct cper_sec_pcie *pcie_err = acpi_hest_get_payload(gdata);
 593
 594	if (pcie_err->validation_bits & CPER_PCIE_VALID_DEVICE_ID &&
 595	    pcie_err->validation_bits & CPER_PCIE_VALID_AER_INFO) {
 596		unsigned int devfn;
 597		int aer_severity;
 598		u8 *aer_info;
 599
 600		devfn = PCI_DEVFN(pcie_err->device_id.device,
 601				  pcie_err->device_id.function);
 602		aer_severity = cper_severity_to_aer(gdata->error_severity);
 603
 604		/*
 605		 * If firmware reset the component to contain
 606		 * the error, we must reinitialize it before
 607		 * use, so treat it as a fatal AER error.
 608		 */
 609		if (gdata->flags & CPER_SEC_RESET)
 610			aer_severity = AER_FATAL;
 611
 612		aer_info = (void *)gen_pool_alloc(ghes_estatus_pool,
 613						  sizeof(struct aer_capability_regs));
 614		if (!aer_info)
 615			return;
 616		memcpy(aer_info, pcie_err->aer_info, sizeof(struct aer_capability_regs));
 617
 618		aer_recover_queue(pcie_err->device_id.segment,
 619				  pcie_err->device_id.bus,
 620				  devfn, aer_severity,
 621				  (struct aer_capability_regs *)
 622				  aer_info);
 623	}
 624#endif
 625}
 626
 627static BLOCKING_NOTIFIER_HEAD(vendor_record_notify_list);
 628
 629int ghes_register_vendor_record_notifier(struct notifier_block *nb)
 630{
 631	return blocking_notifier_chain_register(&vendor_record_notify_list, nb);
 632}
 633EXPORT_SYMBOL_GPL(ghes_register_vendor_record_notifier);
 634
 635void ghes_unregister_vendor_record_notifier(struct notifier_block *nb)
 636{
 637	blocking_notifier_chain_unregister(&vendor_record_notify_list, nb);
 638}
 639EXPORT_SYMBOL_GPL(ghes_unregister_vendor_record_notifier);
 640
 641static void ghes_vendor_record_work_func(struct work_struct *work)
 642{
 643	struct ghes_vendor_record_entry *entry;
 644	struct acpi_hest_generic_data *gdata;
 645	u32 len;
 646
 647	entry = container_of(work, struct ghes_vendor_record_entry, work);
 648	gdata = GHES_GDATA_FROM_VENDOR_ENTRY(entry);
 649
 650	blocking_notifier_call_chain(&vendor_record_notify_list,
 651				     entry->error_severity, gdata);
 652
 653	len = GHES_VENDOR_ENTRY_LEN(acpi_hest_get_record_size(gdata));
 654	gen_pool_free(ghes_estatus_pool, (unsigned long)entry, len);
 655}
 656
 657static void ghes_defer_non_standard_event(struct acpi_hest_generic_data *gdata,
 658					  int sev)
 659{
 660	struct acpi_hest_generic_data *copied_gdata;
 661	struct ghes_vendor_record_entry *entry;
 662	u32 len;
 663
 664	len = GHES_VENDOR_ENTRY_LEN(acpi_hest_get_record_size(gdata));
 665	entry = (void *)gen_pool_alloc(ghes_estatus_pool, len);
 666	if (!entry)
 667		return;
 668
 669	copied_gdata = GHES_GDATA_FROM_VENDOR_ENTRY(entry);
 670	memcpy(copied_gdata, gdata, acpi_hest_get_record_size(gdata));
 671	entry->error_severity = sev;
 672
 673	INIT_WORK(&entry->work, ghes_vendor_record_work_func);
 674	schedule_work(&entry->work);
 675}
 676
 677/* Room for 8 entries for each of the 4 event log queues */
 678#define CXL_CPER_FIFO_DEPTH 32
 679DEFINE_KFIFO(cxl_cper_fifo, struct cxl_cper_work_data, CXL_CPER_FIFO_DEPTH);
 680
 681/* Synchronize schedule_work() with cxl_cper_work changes */
 682static DEFINE_SPINLOCK(cxl_cper_work_lock);
 683struct work_struct *cxl_cper_work;
 684
 685static void cxl_cper_post_event(enum cxl_event_type event_type,
 686				struct cxl_cper_event_rec *rec)
 687{
 688	struct cxl_cper_work_data wd;
 689
 690	if (rec->hdr.length <= sizeof(rec->hdr) ||
 691	    rec->hdr.length > sizeof(*rec)) {
 692		pr_err(FW_WARN "CXL CPER Invalid section length (%u)\n",
 693		       rec->hdr.length);
 694		return;
 695	}
 696
 697	if (!(rec->hdr.validation_bits & CPER_CXL_COMP_EVENT_LOG_VALID)) {
 698		pr_err(FW_WARN "CXL CPER invalid event\n");
 699		return;
 700	}
 701
 702	guard(spinlock_irqsave)(&cxl_cper_work_lock);
 703
 704	if (!cxl_cper_work)
 705		return;
 706
 707	wd.event_type = event_type;
 708	memcpy(&wd.rec, rec, sizeof(wd.rec));
 709
 710	if (!kfifo_put(&cxl_cper_fifo, wd)) {
 711		pr_err_ratelimited("CXL CPER kfifo overflow\n");
 712		return;
 713	}
 714
 715	schedule_work(cxl_cper_work);
 716}
 717
 718int cxl_cper_register_work(struct work_struct *work)
 719{
 720	if (cxl_cper_work)
 721		return -EINVAL;
 722
 723	guard(spinlock)(&cxl_cper_work_lock);
 724	cxl_cper_work = work;
 725	return 0;
 726}
 727EXPORT_SYMBOL_NS_GPL(cxl_cper_register_work, "CXL");
 728
 729int cxl_cper_unregister_work(struct work_struct *work)
 730{
 731	if (cxl_cper_work != work)
 732		return -EINVAL;
 733
 734	guard(spinlock)(&cxl_cper_work_lock);
 735	cxl_cper_work = NULL;
 736	return 0;
 737}
 738EXPORT_SYMBOL_NS_GPL(cxl_cper_unregister_work, "CXL");
 739
 740int cxl_cper_kfifo_get(struct cxl_cper_work_data *wd)
 741{
 742	return kfifo_get(&cxl_cper_fifo, wd);
 743}
 744EXPORT_SYMBOL_NS_GPL(cxl_cper_kfifo_get, "CXL");
 745
 746static bool ghes_do_proc(struct ghes *ghes,
 747			 const struct acpi_hest_generic_status *estatus)
 748{
 749	int sev, sec_sev;
 750	struct acpi_hest_generic_data *gdata;
 751	guid_t *sec_type;
 752	const guid_t *fru_id = &guid_null;
 753	char *fru_text = "";
 754	bool queued = false;
 755	bool sync = is_hest_sync_notify(ghes);
 756
 757	sev = ghes_severity(estatus->error_severity);
 758	apei_estatus_for_each_section(estatus, gdata) {
 759		sec_type = (guid_t *)gdata->section_type;
 760		sec_sev = ghes_severity(gdata->error_severity);
 761		if (gdata->validation_bits & CPER_SEC_VALID_FRU_ID)
 762			fru_id = (guid_t *)gdata->fru_id;
 763
 764		if (gdata->validation_bits & CPER_SEC_VALID_FRU_TEXT)
 765			fru_text = gdata->fru_text;
 766
 767		if (guid_equal(sec_type, &CPER_SEC_PLATFORM_MEM)) {
 768			struct cper_sec_mem_err *mem_err = acpi_hest_get_payload(gdata);
 769
 770			atomic_notifier_call_chain(&ghes_report_chain, sev, mem_err);
 771
 772			arch_apei_report_mem_error(sev, mem_err);
 773			queued = ghes_handle_memory_failure(gdata, sev, sync);
 774		}
 775		else if (guid_equal(sec_type, &CPER_SEC_PCIE)) {
 776			ghes_handle_aer(gdata);
 
 777		}
 778		else if (guid_equal(sec_type, &CPER_SEC_PROC_ARM)) {
 779			queued = ghes_handle_arm_hw_error(gdata, sev, sync);
 780		} else if (guid_equal(sec_type, &CPER_SEC_CXL_GEN_MEDIA_GUID)) {
 781			struct cxl_cper_event_rec *rec = acpi_hest_get_payload(gdata);
 782
 783			cxl_cper_post_event(CXL_CPER_EVENT_GEN_MEDIA, rec);
 784		} else if (guid_equal(sec_type, &CPER_SEC_CXL_DRAM_GUID)) {
 785			struct cxl_cper_event_rec *rec = acpi_hest_get_payload(gdata);
 786
 787			cxl_cper_post_event(CXL_CPER_EVENT_DRAM, rec);
 788		} else if (guid_equal(sec_type, &CPER_SEC_CXL_MEM_MODULE_GUID)) {
 789			struct cxl_cper_event_rec *rec = acpi_hest_get_payload(gdata);
 790
 791			cxl_cper_post_event(CXL_CPER_EVENT_MEM_MODULE, rec);
 792		} else {
 793			void *err = acpi_hest_get_payload(gdata);
 
 
 794
 795			ghes_defer_non_standard_event(gdata, sev);
 796			log_non_standard_event(sec_type, fru_id, fru_text,
 797					       sec_sev, err,
 798					       gdata->error_data_length);
 799		}
 
 800	}
 801
 802	return queued;
 803}
 804
 805static void __ghes_print_estatus(const char *pfx,
 806				 const struct acpi_hest_generic *generic,
 807				 const struct acpi_hest_generic_status *estatus)
 808{
 809	static atomic_t seqno;
 810	unsigned int curr_seqno;
 811	char pfx_seq[64];
 812
 813	if (pfx == NULL) {
 814		if (ghes_severity(estatus->error_severity) <=
 815		    GHES_SEV_CORRECTED)
 816			pfx = KERN_WARNING;
 817		else
 818			pfx = KERN_ERR;
 819	}
 820	curr_seqno = atomic_inc_return(&seqno);
 821	snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}" HW_ERR, pfx, curr_seqno);
 822	printk("%s""Hardware error from APEI Generic Hardware Error Source: %d\n",
 823	       pfx_seq, generic->header.source_id);
 824	cper_estatus_print(pfx_seq, estatus);
 825}
 826
 827static int ghes_print_estatus(const char *pfx,
 828			      const struct acpi_hest_generic *generic,
 829			      const struct acpi_hest_generic_status *estatus)
 830{
 831	/* Not more than 2 messages every 5 seconds */
 832	static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2);
 833	static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2);
 834	struct ratelimit_state *ratelimit;
 835
 836	if (ghes_severity(estatus->error_severity) <= GHES_SEV_CORRECTED)
 837		ratelimit = &ratelimit_corrected;
 838	else
 839		ratelimit = &ratelimit_uncorrected;
 840	if (__ratelimit(ratelimit)) {
 841		__ghes_print_estatus(pfx, generic, estatus);
 842		return 1;
 843	}
 844	return 0;
 845}
 846
 847/*
 848 * GHES error status reporting throttle, to report more kinds of
 849 * errors, instead of just most frequently occurred errors.
 850 */
 851static int ghes_estatus_cached(struct acpi_hest_generic_status *estatus)
 852{
 853	u32 len;
 854	int i, cached = 0;
 855	unsigned long long now;
 856	struct ghes_estatus_cache *cache;
 857	struct acpi_hest_generic_status *cache_estatus;
 858
 859	len = cper_estatus_len(estatus);
 860	rcu_read_lock();
 861	for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
 862		cache = rcu_dereference(ghes_estatus_caches[i]);
 863		if (cache == NULL)
 864			continue;
 865		if (len != cache->estatus_len)
 866			continue;
 867		cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
 868		if (memcmp(estatus, cache_estatus, len))
 869			continue;
 870		atomic_inc(&cache->count);
 871		now = sched_clock();
 872		if (now - cache->time_in < GHES_ESTATUS_IN_CACHE_MAX_NSEC)
 873			cached = 1;
 874		break;
 875	}
 876	rcu_read_unlock();
 877	return cached;
 878}
 879
 880static struct ghes_estatus_cache *ghes_estatus_cache_alloc(
 881	struct acpi_hest_generic *generic,
 882	struct acpi_hest_generic_status *estatus)
 883{
 884	int alloced;
 885	u32 len, cache_len;
 886	struct ghes_estatus_cache *cache;
 887	struct acpi_hest_generic_status *cache_estatus;
 888
 889	alloced = atomic_add_return(1, &ghes_estatus_cache_alloced);
 890	if (alloced > GHES_ESTATUS_CACHE_ALLOCED_MAX) {
 891		atomic_dec(&ghes_estatus_cache_alloced);
 892		return NULL;
 893	}
 894	len = cper_estatus_len(estatus);
 895	cache_len = GHES_ESTATUS_CACHE_LEN(len);
 896	cache = (void *)gen_pool_alloc(ghes_estatus_pool, cache_len);
 897	if (!cache) {
 898		atomic_dec(&ghes_estatus_cache_alloced);
 899		return NULL;
 900	}
 901	cache_estatus = GHES_ESTATUS_FROM_CACHE(cache);
 902	memcpy(cache_estatus, estatus, len);
 903	cache->estatus_len = len;
 904	atomic_set(&cache->count, 0);
 905	cache->generic = generic;
 906	cache->time_in = sched_clock();
 907	return cache;
 908}
 909
 910static void ghes_estatus_cache_rcu_free(struct rcu_head *head)
 911{
 912	struct ghes_estatus_cache *cache;
 913	u32 len;
 914
 915	cache = container_of(head, struct ghes_estatus_cache, rcu);
 916	len = cper_estatus_len(GHES_ESTATUS_FROM_CACHE(cache));
 917	len = GHES_ESTATUS_CACHE_LEN(len);
 918	gen_pool_free(ghes_estatus_pool, (unsigned long)cache, len);
 919	atomic_dec(&ghes_estatus_cache_alloced);
 920}
 921
 922static void
 923ghes_estatus_cache_add(struct acpi_hest_generic *generic,
 924		       struct acpi_hest_generic_status *estatus)
 
 
 
 
 
 
 
 
 925{
 926	unsigned long long now, duration, period, max_period = 0;
 927	struct ghes_estatus_cache *cache, *new_cache;
 928	struct ghes_estatus_cache __rcu *victim;
 929	int i, slot = -1, count;
 
 
 930
 931	new_cache = ghes_estatus_cache_alloc(generic, estatus);
 932	if (!new_cache)
 933		return;
 934
 935	rcu_read_lock();
 936	now = sched_clock();
 937	for (i = 0; i < GHES_ESTATUS_CACHES_SIZE; i++) {
 938		cache = rcu_dereference(ghes_estatus_caches[i]);
 939		if (cache == NULL) {
 940			slot = i;
 
 941			break;
 942		}
 943		duration = now - cache->time_in;
 944		if (duration >= GHES_ESTATUS_IN_CACHE_MAX_NSEC) {
 945			slot = i;
 
 946			break;
 947		}
 948		count = atomic_read(&cache->count);
 949		period = duration;
 950		do_div(period, (count + 1));
 951		if (period > max_period) {
 952			max_period = period;
 953			slot = i;
 
 954		}
 955	}
 
 
 
 
 
 
 
 
 956	rcu_read_unlock();
 957
 958	if (slot != -1) {
 959		/*
 960		 * Use release semantics to ensure that ghes_estatus_cached()
 961		 * running on another CPU will see the updated cache fields if
 962		 * it can see the new value of the pointer.
 963		 */
 964		victim = xchg_release(&ghes_estatus_caches[slot],
 965				      RCU_INITIALIZER(new_cache));
 966
 967		/*
 968		 * At this point, victim may point to a cached item different
 969		 * from the one based on which we selected the slot. Instead of
 970		 * going to the loop again to pick another slot, let's just
 971		 * drop the other item anyway: this may cause a false cache
 972		 * miss later on, but that won't cause any problems.
 973		 */
 974		if (victim)
 975			call_rcu(&unrcu_pointer(victim)->rcu,
 976				 ghes_estatus_cache_rcu_free);
 977	}
 978}
 979
 980static void __ghes_panic(struct ghes *ghes,
 981			 struct acpi_hest_generic_status *estatus,
 982			 u64 buf_paddr, enum fixed_addresses fixmap_idx)
 983{
 984	const char *msg = GHES_PFX "Fatal hardware error";
 985
 986	__ghes_print_estatus(KERN_EMERG, ghes->generic, estatus);
 987
 988	ghes_clear_estatus(ghes, estatus, buf_paddr, fixmap_idx);
 989
 990	if (!panic_timeout)
 991		pr_emerg("%s but panic disabled\n", msg);
 992
 993	panic(msg);
 994}
 995
 996static int ghes_proc(struct ghes *ghes)
 997{
 998	struct acpi_hest_generic_status *estatus = ghes->estatus;
 999	u64 buf_paddr;
1000	int rc;
1001
1002	rc = ghes_read_estatus(ghes, estatus, &buf_paddr, FIX_APEI_GHES_IRQ);
1003	if (rc)
1004		goto out;
1005
1006	if (ghes_severity(estatus->error_severity) >= GHES_SEV_PANIC)
1007		__ghes_panic(ghes, estatus, buf_paddr, FIX_APEI_GHES_IRQ);
1008
1009	if (!ghes_estatus_cached(estatus)) {
1010		if (ghes_print_estatus(NULL, ghes->generic, estatus))
1011			ghes_estatus_cache_add(ghes->generic, estatus);
1012	}
1013	ghes_do_proc(ghes, estatus);
1014
1015out:
1016	ghes_clear_estatus(ghes, estatus, buf_paddr, FIX_APEI_GHES_IRQ);
1017
1018	return rc;
1019}
1020
1021static void ghes_add_timer(struct ghes *ghes)
1022{
1023	struct acpi_hest_generic *g = ghes->generic;
1024	unsigned long expire;
1025
1026	if (!g->notify.poll_interval) {
1027		pr_warn(FW_WARN GHES_PFX "Poll interval is 0 for generic hardware error source: %d, disabled.\n",
1028			g->header.source_id);
1029		return;
1030	}
1031	expire = jiffies + msecs_to_jiffies(g->notify.poll_interval);
1032	ghes->timer.expires = round_jiffies_relative(expire);
1033	add_timer(&ghes->timer);
1034}
1035
1036static void ghes_poll_func(struct timer_list *t)
1037{
1038	struct ghes *ghes = from_timer(ghes, t, timer);
1039	unsigned long flags;
1040
1041	spin_lock_irqsave(&ghes_notify_lock_irq, flags);
1042	ghes_proc(ghes);
1043	spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
1044	if (!(ghes->flags & GHES_EXITING))
1045		ghes_add_timer(ghes);
1046}
1047
1048static irqreturn_t ghes_irq_func(int irq, void *data)
1049{
1050	struct ghes *ghes = data;
1051	unsigned long flags;
1052	int rc;
1053
1054	spin_lock_irqsave(&ghes_notify_lock_irq, flags);
1055	rc = ghes_proc(ghes);
1056	spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
1057	if (rc)
1058		return IRQ_NONE;
1059
1060	return IRQ_HANDLED;
1061}
1062
1063static int ghes_notify_hed(struct notifier_block *this, unsigned long event,
1064			   void *data)
1065{
1066	struct ghes *ghes;
1067	unsigned long flags;
1068	int ret = NOTIFY_DONE;
1069
1070	spin_lock_irqsave(&ghes_notify_lock_irq, flags);
1071	rcu_read_lock();
1072	list_for_each_entry_rcu(ghes, &ghes_hed, list) {
1073		if (!ghes_proc(ghes))
1074			ret = NOTIFY_OK;
1075	}
1076	rcu_read_unlock();
1077	spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
1078
1079	return ret;
1080}
1081
1082static struct notifier_block ghes_notifier_hed = {
1083	.notifier_call = ghes_notify_hed,
1084};
1085
1086/*
1087 * Handlers for CPER records may not be NMI safe. For example,
1088 * memory_failure_queue() takes spinlocks and calls schedule_work_on().
1089 * In any NMI-like handler, memory from ghes_estatus_pool is used to save
1090 * estatus, and added to the ghes_estatus_llist. irq_work_queue() causes
1091 * ghes_proc_in_irq() to run in IRQ context where each estatus in
1092 * ghes_estatus_llist is processed.
1093 *
1094 * Memory from the ghes_estatus_pool is also used with the ghes_estatus_cache
1095 * to suppress frequent messages.
1096 */
1097static struct llist_head ghes_estatus_llist;
1098static struct irq_work ghes_proc_irq_work;
1099
1100static void ghes_proc_in_irq(struct irq_work *irq_work)
1101{
1102	struct llist_node *llnode, *next;
1103	struct ghes_estatus_node *estatus_node;
1104	struct acpi_hest_generic *generic;
1105	struct acpi_hest_generic_status *estatus;
1106	bool task_work_pending;
1107	u32 len, node_len;
1108	int ret;
1109
1110	llnode = llist_del_all(&ghes_estatus_llist);
1111	/*
1112	 * Because the time order of estatus in list is reversed,
1113	 * revert it back to proper order.
1114	 */
1115	llnode = llist_reverse_order(llnode);
1116	while (llnode) {
1117		next = llnode->next;
1118		estatus_node = llist_entry(llnode, struct ghes_estatus_node,
1119					   llnode);
1120		estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
1121		len = cper_estatus_len(estatus);
1122		node_len = GHES_ESTATUS_NODE_LEN(len);
1123		task_work_pending = ghes_do_proc(estatus_node->ghes, estatus);
1124		if (!ghes_estatus_cached(estatus)) {
1125			generic = estatus_node->generic;
1126			if (ghes_print_estatus(NULL, generic, estatus))
1127				ghes_estatus_cache_add(generic, estatus);
1128		}
1129
1130		if (task_work_pending && current->mm) {
1131			estatus_node->task_work.func = ghes_kick_task_work;
1132			estatus_node->task_work_cpu = smp_processor_id();
1133			ret = task_work_add(current, &estatus_node->task_work,
1134					    TWA_RESUME);
1135			if (ret)
1136				estatus_node->task_work.func = NULL;
1137		}
1138
1139		if (!estatus_node->task_work.func)
1140			gen_pool_free(ghes_estatus_pool,
1141				      (unsigned long)estatus_node, node_len);
1142
1143		llnode = next;
1144	}
1145}
1146
1147static void ghes_print_queued_estatus(void)
1148{
1149	struct llist_node *llnode;
1150	struct ghes_estatus_node *estatus_node;
1151	struct acpi_hest_generic *generic;
1152	struct acpi_hest_generic_status *estatus;
 
1153
1154	llnode = llist_del_all(&ghes_estatus_llist);
1155	/*
1156	 * Because the time order of estatus in list is reversed,
1157	 * revert it back to proper order.
1158	 */
1159	llnode = llist_reverse_order(llnode);
1160	while (llnode) {
1161		estatus_node = llist_entry(llnode, struct ghes_estatus_node,
1162					   llnode);
1163		estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
 
 
1164		generic = estatus_node->generic;
1165		ghes_print_estatus(NULL, generic, estatus);
1166		llnode = llnode->next;
1167	}
1168}
1169
1170static int ghes_in_nmi_queue_one_entry(struct ghes *ghes,
1171				       enum fixed_addresses fixmap_idx)
1172{
1173	struct acpi_hest_generic_status *estatus, tmp_header;
1174	struct ghes_estatus_node *estatus_node;
1175	u32 len, node_len;
1176	u64 buf_paddr;
1177	int sev, rc;
1178
1179	if (!IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG))
1180		return -EOPNOTSUPP;
1181
1182	rc = __ghes_peek_estatus(ghes, &tmp_header, &buf_paddr, fixmap_idx);
1183	if (rc) {
1184		ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1185		return rc;
1186	}
1187
1188	rc = __ghes_check_estatus(ghes, &tmp_header);
1189	if (rc) {
1190		ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1191		return rc;
 
 
 
 
 
 
 
 
1192	}
1193
1194	len = cper_estatus_len(&tmp_header);
1195	node_len = GHES_ESTATUS_NODE_LEN(len);
1196	estatus_node = (void *)gen_pool_alloc(ghes_estatus_pool, node_len);
1197	if (!estatus_node)
1198		return -ENOMEM;
1199
1200	estatus_node->ghes = ghes;
1201	estatus_node->generic = ghes->generic;
1202	estatus_node->task_work.func = NULL;
1203	estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
1204
1205	if (__ghes_read_estatus(estatus, buf_paddr, fixmap_idx, len)) {
1206		ghes_clear_estatus(ghes, estatus, buf_paddr, fixmap_idx);
1207		rc = -ENOENT;
1208		goto no_work;
1209	}
1210
1211	sev = ghes_severity(estatus->error_severity);
1212	if (sev >= GHES_SEV_PANIC) {
1213		ghes_print_queued_estatus();
1214		__ghes_panic(ghes, estatus, buf_paddr, fixmap_idx);
 
 
 
 
 
1215	}
1216
1217	ghes_clear_estatus(ghes, &tmp_header, buf_paddr, fixmap_idx);
1218
1219	/* This error has been reported before, don't process it again. */
1220	if (ghes_estatus_cached(estatus))
1221		goto no_work;
1222
1223	llist_add(&estatus_node->llnode, &ghes_estatus_llist);
1224
1225	return rc;
1226
1227no_work:
1228	gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node,
1229		      node_len);
1230
1231	return rc;
1232}
1233
1234static int ghes_in_nmi_spool_from_list(struct list_head *rcu_list,
1235				       enum fixed_addresses fixmap_idx)
1236{
1237	int ret = -ENOENT;
1238	struct ghes *ghes;
1239
1240	rcu_read_lock();
1241	list_for_each_entry_rcu(ghes, rcu_list, list) {
1242		if (!ghes_in_nmi_queue_one_entry(ghes, fixmap_idx))
1243			ret = 0;
1244	}
1245	rcu_read_unlock();
1246
1247	if (IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG) && !ret)
1248		irq_work_queue(&ghes_proc_irq_work);
1249
1250	return ret;
1251}
1252
1253#ifdef CONFIG_ACPI_APEI_SEA
1254static LIST_HEAD(ghes_sea);
1255
1256/*
1257 * Return 0 only if one of the SEA error sources successfully reported an error
1258 * record sent from the firmware.
1259 */
1260int ghes_notify_sea(void)
1261{
1262	static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sea);
1263	int rv;
1264
1265	raw_spin_lock(&ghes_notify_lock_sea);
1266	rv = ghes_in_nmi_spool_from_list(&ghes_sea, FIX_APEI_GHES_SEA);
1267	raw_spin_unlock(&ghes_notify_lock_sea);
1268
1269	return rv;
1270}
1271
1272static void ghes_sea_add(struct ghes *ghes)
1273{
1274	mutex_lock(&ghes_list_mutex);
1275	list_add_rcu(&ghes->list, &ghes_sea);
1276	mutex_unlock(&ghes_list_mutex);
1277}
1278
1279static void ghes_sea_remove(struct ghes *ghes)
1280{
1281	mutex_lock(&ghes_list_mutex);
1282	list_del_rcu(&ghes->list);
1283	mutex_unlock(&ghes_list_mutex);
1284	synchronize_rcu();
1285}
1286#else /* CONFIG_ACPI_APEI_SEA */
1287static inline void ghes_sea_add(struct ghes *ghes) { }
1288static inline void ghes_sea_remove(struct ghes *ghes) { }
1289#endif /* CONFIG_ACPI_APEI_SEA */
1290
1291#ifdef CONFIG_HAVE_ACPI_APEI_NMI
1292/*
1293 * NMI may be triggered on any CPU, so ghes_in_nmi is used for
1294 * having only one concurrent reader.
1295 */
1296static atomic_t ghes_in_nmi = ATOMIC_INIT(0);
1297
1298static LIST_HEAD(ghes_nmi);
1299
1300static int ghes_notify_nmi(unsigned int cmd, struct pt_regs *regs)
1301{
1302	static DEFINE_RAW_SPINLOCK(ghes_notify_lock_nmi);
1303	int ret = NMI_DONE;
1304
1305	if (!atomic_add_unless(&ghes_in_nmi, 1, 1))
1306		return ret;
1307
1308	raw_spin_lock(&ghes_notify_lock_nmi);
1309	if (!ghes_in_nmi_spool_from_list(&ghes_nmi, FIX_APEI_GHES_NMI))
1310		ret = NMI_HANDLED;
1311	raw_spin_unlock(&ghes_notify_lock_nmi);
1312
1313	atomic_dec(&ghes_in_nmi);
 
1314	return ret;
1315}
1316
1317static void ghes_nmi_add(struct ghes *ghes)
1318{
1319	mutex_lock(&ghes_list_mutex);
1320	if (list_empty(&ghes_nmi))
1321		register_nmi_handler(NMI_LOCAL, ghes_notify_nmi, 0, "ghes");
1322	list_add_rcu(&ghes->list, &ghes_nmi);
1323	mutex_unlock(&ghes_list_mutex);
1324}
1325
1326static void ghes_nmi_remove(struct ghes *ghes)
1327{
1328	mutex_lock(&ghes_list_mutex);
1329	list_del_rcu(&ghes->list);
1330	if (list_empty(&ghes_nmi))
1331		unregister_nmi_handler(NMI_LOCAL, "ghes");
1332	mutex_unlock(&ghes_list_mutex);
1333	/*
1334	 * To synchronize with NMI handler, ghes can only be
1335	 * freed after NMI handler finishes.
1336	 */
1337	synchronize_rcu();
1338}
1339#else /* CONFIG_HAVE_ACPI_APEI_NMI */
1340static inline void ghes_nmi_add(struct ghes *ghes) { }
1341static inline void ghes_nmi_remove(struct ghes *ghes) { }
1342#endif /* CONFIG_HAVE_ACPI_APEI_NMI */
1343
1344static void ghes_nmi_init_cxt(void)
1345{
1346	init_irq_work(&ghes_proc_irq_work, ghes_proc_in_irq);
1347}
1348
1349static int __ghes_sdei_callback(struct ghes *ghes,
1350				enum fixed_addresses fixmap_idx)
1351{
1352	if (!ghes_in_nmi_queue_one_entry(ghes, fixmap_idx)) {
1353		irq_work_queue(&ghes_proc_irq_work);
1354
1355		return 0;
1356	}
1357
1358	return -ENOENT;
1359}
1360
1361static int ghes_sdei_normal_callback(u32 event_num, struct pt_regs *regs,
1362				      void *arg)
1363{
1364	static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sdei_normal);
1365	struct ghes *ghes = arg;
1366	int err;
1367
1368	raw_spin_lock(&ghes_notify_lock_sdei_normal);
1369	err = __ghes_sdei_callback(ghes, FIX_APEI_GHES_SDEI_NORMAL);
1370	raw_spin_unlock(&ghes_notify_lock_sdei_normal);
1371
1372	return err;
1373}
1374
1375static int ghes_sdei_critical_callback(u32 event_num, struct pt_regs *regs,
1376				       void *arg)
1377{
1378	static DEFINE_RAW_SPINLOCK(ghes_notify_lock_sdei_critical);
1379	struct ghes *ghes = arg;
1380	int err;
1381
1382	raw_spin_lock(&ghes_notify_lock_sdei_critical);
1383	err = __ghes_sdei_callback(ghes, FIX_APEI_GHES_SDEI_CRITICAL);
1384	raw_spin_unlock(&ghes_notify_lock_sdei_critical);
1385
1386	return err;
1387}
1388
1389static int apei_sdei_register_ghes(struct ghes *ghes)
 
1390{
1391	if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE))
1392		return -EOPNOTSUPP;
1393
1394	return sdei_register_ghes(ghes, ghes_sdei_normal_callback,
1395				 ghes_sdei_critical_callback);
1396}
1397
1398static int apei_sdei_unregister_ghes(struct ghes *ghes)
1399{
1400	if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE))
1401		return -EOPNOTSUPP;
1402
1403	return sdei_unregister_ghes(ghes);
1404}
1405
1406static int ghes_probe(struct platform_device *ghes_dev)
1407{
1408	struct acpi_hest_generic *generic;
1409	struct ghes *ghes = NULL;
1410	unsigned long flags;
1411
1412	int rc = -EINVAL;
1413
1414	generic = *(struct acpi_hest_generic **)ghes_dev->dev.platform_data;
1415	if (!generic->enabled)
1416		return -ENODEV;
1417
1418	switch (generic->notify.type) {
1419	case ACPI_HEST_NOTIFY_POLLED:
1420	case ACPI_HEST_NOTIFY_EXTERNAL:
1421	case ACPI_HEST_NOTIFY_SCI:
1422	case ACPI_HEST_NOTIFY_GSIV:
1423	case ACPI_HEST_NOTIFY_GPIO:
1424		break;
1425
1426	case ACPI_HEST_NOTIFY_SEA:
1427		if (!IS_ENABLED(CONFIG_ACPI_APEI_SEA)) {
1428			pr_warn(GHES_PFX "Generic hardware error source: %d notified via SEA is not supported\n",
1429				generic->header.source_id);
1430			rc = -ENOTSUPP;
1431			goto err;
1432		}
1433		break;
1434	case ACPI_HEST_NOTIFY_NMI:
1435		if (!IS_ENABLED(CONFIG_HAVE_ACPI_APEI_NMI)) {
1436			pr_warn(GHES_PFX "Generic hardware error source: %d notified via NMI interrupt is not supported!\n",
1437				generic->header.source_id);
1438			goto err;
1439		}
1440		break;
1441	case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1442		if (!IS_ENABLED(CONFIG_ARM_SDE_INTERFACE)) {
1443			pr_warn(GHES_PFX "Generic hardware error source: %d notified via SDE Interface is not supported!\n",
1444				generic->header.source_id);
1445			goto err;
1446		}
1447		break;
1448	case ACPI_HEST_NOTIFY_LOCAL:
1449		pr_warn(GHES_PFX "Generic hardware error source: %d notified via local interrupt is not supported!\n",
1450			generic->header.source_id);
1451		goto err;
1452	default:
1453		pr_warn(FW_WARN GHES_PFX "Unknown notification type: %u for generic hardware error source: %d\n",
1454			generic->notify.type, generic->header.source_id);
1455		goto err;
1456	}
1457
1458	rc = -EIO;
1459	if (generic->error_block_length <
1460	    sizeof(struct acpi_hest_generic_status)) {
1461		pr_warn(FW_BUG GHES_PFX "Invalid error block length: %u for generic hardware error source: %d\n",
1462			generic->error_block_length, generic->header.source_id);
 
1463		goto err;
1464	}
1465	ghes = ghes_new(generic);
1466	if (IS_ERR(ghes)) {
1467		rc = PTR_ERR(ghes);
1468		ghes = NULL;
1469		goto err;
1470	}
1471
1472	switch (generic->notify.type) {
1473	case ACPI_HEST_NOTIFY_POLLED:
1474		timer_setup(&ghes->timer, ghes_poll_func, 0);
 
 
1475		ghes_add_timer(ghes);
1476		break;
1477	case ACPI_HEST_NOTIFY_EXTERNAL:
1478		/* External interrupt vector is GSI */
1479		rc = acpi_gsi_to_irq(generic->notify.vector, &ghes->irq);
1480		if (rc) {
1481			pr_err(GHES_PFX "Failed to map GSI to IRQ for generic hardware error source: %d\n",
1482			       generic->header.source_id);
1483			goto err;
1484		}
1485		rc = request_irq(ghes->irq, ghes_irq_func, IRQF_SHARED,
1486				 "GHES IRQ", ghes);
1487		if (rc) {
1488			pr_err(GHES_PFX "Failed to register IRQ for generic hardware error source: %d\n",
1489			       generic->header.source_id);
1490			goto err;
1491		}
1492		break;
1493
1494	case ACPI_HEST_NOTIFY_SCI:
1495	case ACPI_HEST_NOTIFY_GSIV:
1496	case ACPI_HEST_NOTIFY_GPIO:
1497		mutex_lock(&ghes_list_mutex);
1498		if (list_empty(&ghes_hed))
1499			register_acpi_hed_notifier(&ghes_notifier_hed);
1500		list_add_rcu(&ghes->list, &ghes_hed);
1501		mutex_unlock(&ghes_list_mutex);
1502		break;
1503
1504	case ACPI_HEST_NOTIFY_SEA:
1505		ghes_sea_add(ghes);
1506		break;
1507	case ACPI_HEST_NOTIFY_NMI:
1508		ghes_nmi_add(ghes);
1509		break;
1510	case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1511		rc = apei_sdei_register_ghes(ghes);
1512		if (rc)
1513			goto err;
 
 
1514		break;
1515	default:
1516		BUG();
1517	}
1518
1519	platform_set_drvdata(ghes_dev, ghes);
1520
1521	ghes->dev = &ghes_dev->dev;
1522
1523	mutex_lock(&ghes_devs_mutex);
1524	list_add_tail(&ghes->elist, &ghes_devs);
1525	mutex_unlock(&ghes_devs_mutex);
1526
1527	/* Handle any pending errors right away */
1528	spin_lock_irqsave(&ghes_notify_lock_irq, flags);
1529	ghes_proc(ghes);
1530	spin_unlock_irqrestore(&ghes_notify_lock_irq, flags);
1531
1532	return 0;
1533
1534err:
1535	if (ghes) {
1536		ghes_fini(ghes);
1537		kfree(ghes);
1538	}
1539	return rc;
1540}
1541
1542static void ghes_remove(struct platform_device *ghes_dev)
1543{
1544	int rc;
1545	struct ghes *ghes;
1546	struct acpi_hest_generic *generic;
 
1547
1548	ghes = platform_get_drvdata(ghes_dev);
1549	generic = ghes->generic;
1550
1551	ghes->flags |= GHES_EXITING;
1552	switch (generic->notify.type) {
1553	case ACPI_HEST_NOTIFY_POLLED:
1554		timer_shutdown_sync(&ghes->timer);
1555		break;
1556	case ACPI_HEST_NOTIFY_EXTERNAL:
1557		free_irq(ghes->irq, ghes);
1558		break;
1559
1560	case ACPI_HEST_NOTIFY_SCI:
1561	case ACPI_HEST_NOTIFY_GSIV:
1562	case ACPI_HEST_NOTIFY_GPIO:
1563		mutex_lock(&ghes_list_mutex);
1564		list_del_rcu(&ghes->list);
1565		if (list_empty(&ghes_hed))
1566			unregister_acpi_hed_notifier(&ghes_notifier_hed);
1567		mutex_unlock(&ghes_list_mutex);
1568		synchronize_rcu();
1569		break;
1570
1571	case ACPI_HEST_NOTIFY_SEA:
1572		ghes_sea_remove(ghes);
1573		break;
1574	case ACPI_HEST_NOTIFY_NMI:
1575		ghes_nmi_remove(ghes);
1576		break;
1577	case ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED:
1578		rc = apei_sdei_unregister_ghes(ghes);
1579		if (rc) {
1580			/*
1581			 * Returning early results in a resource leak, but we're
1582			 * only here if stopping the hardware failed.
1583			 */
1584			dev_err(&ghes_dev->dev, "Failed to unregister ghes (%pe)\n",
1585				ERR_PTR(rc));
1586			return;
1587		}
1588		break;
1589	default:
1590		BUG();
1591		break;
1592	}
1593
1594	ghes_fini(ghes);
 
1595
1596	mutex_lock(&ghes_devs_mutex);
1597	list_del(&ghes->elist);
1598	mutex_unlock(&ghes_devs_mutex);
1599
1600	kfree(ghes);
1601}
1602
1603static struct platform_driver ghes_platform_driver = {
1604	.driver		= {
1605		.name	= "GHES",
 
1606	},
1607	.probe		= ghes_probe,
1608	.remove		= ghes_remove,
1609};
1610
1611void __init acpi_ghes_init(void)
1612{
1613	int rc;
1614
1615	sdei_init();
1616
1617	if (acpi_disabled)
1618		return;
1619
1620	switch (hest_disable) {
1621	case HEST_NOT_FOUND:
1622		return;
1623	case HEST_DISABLED:
1624		pr_info(GHES_PFX "HEST is not enabled!\n");
1625		return;
1626	default:
1627		break;
1628	}
1629
1630	if (ghes_disable) {
1631		pr_info(GHES_PFX "GHES is not enabled!\n");
1632		return;
1633	}
1634
1635	ghes_nmi_init_cxt();
 
 
 
 
 
 
 
 
 
 
 
 
 
1636
1637	rc = platform_driver_register(&ghes_platform_driver);
1638	if (rc)
1639		return;
1640
1641	rc = apei_osc_setup();
1642	if (rc == 0 && osc_sb_apei_support_acked)
1643		pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit and WHEA _OSC.\n");
1644	else if (rc == 0 && !osc_sb_apei_support_acked)
1645		pr_info(GHES_PFX "APEI firmware first mode is enabled by WHEA _OSC.\n");
1646	else if (rc && osc_sb_apei_support_acked)
1647		pr_info(GHES_PFX "APEI firmware first mode is enabled by APEI bit.\n");
1648	else
1649		pr_info(GHES_PFX "Failed to enable APEI firmware first mode.\n");
1650}
1651
1652/*
1653 * Known x86 systems that prefer GHES error reporting:
1654 */
1655static struct acpi_platform_list plat_list[] = {
1656	{"HPE   ", "Server  ", 0, ACPI_SIG_FADT, all_versions},
1657	{ } /* End */
1658};
1659
1660struct list_head *ghes_get_devices(void)
1661{
1662	int idx = -1;
1663
1664	if (IS_ENABLED(CONFIG_X86)) {
1665		idx = acpi_match_platform_list(plat_list);
1666		if (idx < 0) {
1667			if (!ghes_edac_force_enable)
1668				return NULL;
1669
1670			pr_warn_once("Force-loading ghes_edac on an unsupported platform. You're on your own!\n");
1671		}
1672	} else if (list_empty(&ghes_devs)) {
1673		return NULL;
1674	}
1675
1676	return &ghes_devs;
1677}
1678EXPORT_SYMBOL_GPL(ghes_get_devices);
1679
1680void ghes_register_report_chain(struct notifier_block *nb)
1681{
1682	atomic_notifier_chain_register(&ghes_report_chain, nb);
 
 
1683}
1684EXPORT_SYMBOL_GPL(ghes_register_report_chain);
1685
1686void ghes_unregister_report_chain(struct notifier_block *nb)
1687{
1688	atomic_notifier_chain_unregister(&ghes_report_chain, nb);
1689}
1690EXPORT_SYMBOL_GPL(ghes_unregister_report_chain);
 
 
v3.5.6
 
   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
  98struct 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
 110struct ghes_estatus_node {
 111	struct llist_node llnode;
 112	struct acpi_hest_generic *generic;
 113};
 
 
 
 
 
 
 114
 115struct 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
 
 
 
 
 
 123bool ghes_disable;
 124module_param_named(disable, ghes_disable, bool, 0);
 125
 126static 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 */
 136static LIST_HEAD(ghes_sci);
 137static LIST_HEAD(ghes_nmi);
 138static 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 */
 144static 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 */
 162static 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 */
 167static DEFINE_RAW_SPINLOCK(ghes_ioremap_lock_nmi);
 168static 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 */
 178static struct gen_pool *ghes_estatus_pool;
 179static unsigned long ghes_estatus_pool_size_request;
 180static struct llist_head ghes_estatus_llist;
 181static struct irq_work ghes_proc_irq_work;
 182
 183struct ghes_estatus_cache *ghes_estatus_caches[GHES_ESTATUS_CACHES_SIZE];
 184static atomic_t ghes_estatus_cache_alloced;
 185
 186static 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
 198static void ghes_ioremap_exit(void)
 199{
 200	free_vm_area(ghes_ioremap_area);
 
 
 
 201}
 202
 203static 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
 214static 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
 225static 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
 235static 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
 245static 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
 253static 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
 260static 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
 267static 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
 289static void ghes_estatus_pool_shrink(unsigned long len)
 290{
 291	ghes_estatus_pool_size_request -= PAGE_ALIGN(len);
 292}
 293
 294static 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
 323err_unmap:
 324	apei_unmap_generic_address(&generic->error_status_address);
 
 
 
 325err_free:
 326	kfree(ghes);
 327	return ERR_PTR(rc);
 328}
 329
 330static void ghes_fini(struct ghes *ghes)
 331{
 332	kfree(ghes->estatus);
 333	apei_unmap_generic_address(&ghes->generic->error_status_address);
 
 
 334}
 335
 336enum {
 337	GHES_SEV_NO = 0x0,
 338	GHES_SEV_CORRECTED = 0x1,
 339	GHES_SEV_RECOVERABLE = 0x2,
 340	GHES_SEV_PANIC = 0x3,
 341};
 342
 343static 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
 360static 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
 397static 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
 438err_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
 445static 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
 455static 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
 505static 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
 527static 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 */
 551static 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
 580static 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
 610static 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
 620static 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
 628static 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
 674static 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);
 
 686out:
 687	ghes_clear_estatus(ghes);
 688	return 0;
 
 689}
 690
 691static 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
 706static 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
 715static 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
 727static 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
 743static 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
 757static 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
 790static 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
 816static 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		}
 872next:
 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
 880out:
 881	raw_spin_unlock(&ghes_nmi_lock);
 882	return ret;
 883}
 884
 885static struct notifier_block ghes_notifier_sci = {
 886	.notifier_call = ghes_notify_sci,
 887};
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 888
 889static 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
 904static int __devinit 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;
 
 989err:
 990	if (ghes) {
 991		ghes_fini(ghes);
 992		kfree(ghes);
 993	}
 994	return rc;
 995}
 996
 997static int __devexit 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
1048static 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
1057static 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;
1104err_pool_exit:
1105	ghes_estatus_pool_exit();
1106err_ioremap_exit:
1107	ghes_ioremap_exit();
1108err:
1109	return rc;
 
 
 
 
 
 
1110}
 
1111
1112static void __exit ghes_exit(void)
1113{
1114	platform_driver_unregister(&ghes_platform_driver);
1115	ghes_estatus_pool_exit();
1116	ghes_ioremap_exit();
1117}
 
1118
1119module_init(ghes_init);
1120module_exit(ghes_exit);
1121
1122MODULE_AUTHOR("Huang Ying");
1123MODULE_DESCRIPTION("APEI Generic Hardware Error Source support");
1124MODULE_LICENSE("GPL");
1125MODULE_ALIAS("platform:GHES");