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