1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Copyright (C) 2001 Dave Engebretsen IBM Corporation 4 */ 5 6 #include <linux/sched.h> 7 #include <linux/interrupt.h> 8 #include <linux/irq.h> 9 #include <linux/of.h> 10 #include <linux/fs.h> 11 #include <linux/reboot.h> 12 #include <linux/irq_work.h> 13 14 #include <asm/machdep.h> 15 #include <asm/rtas.h> 16 #include <asm/firmware.h> 17 #include <asm/mce.h> 18 19 #include "pseries.h" 20 21 static unsigned char ras_log_buf[RTAS_ERROR_LOG_MAX]; 22 static DEFINE_SPINLOCK(ras_log_buf_lock); 23 24 static int ras_check_exception_token; 25 26 static void mce_process_errlog_event(struct irq_work *work); 27 static struct irq_work mce_errlog_process_work = { 28 .func = mce_process_errlog_event, 29 }; 30 31 #define EPOW_SENSOR_TOKEN 9 32 #define EPOW_SENSOR_INDEX 0 33 34 /* EPOW events counter variable */ 35 static int num_epow_events; 36 37 static irqreturn_t ras_hotplug_interrupt(int irq, void *dev_id); 38 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id); 39 static irqreturn_t ras_error_interrupt(int irq, void *dev_id); 40 41 /* RTAS pseries MCE errorlog section. */ 42 struct pseries_mc_errorlog { 43 __be32 fru_id; 44 __be32 proc_id; 45 u8 error_type; 46 /* 47 * sub_err_type (1 byte). Bit fields depends on error_type 48 * 49 * MSB0 50 * | 51 * V 52 * 01234567 53 * XXXXXXXX 54 * 55 * For error_type == MC_ERROR_TYPE_UE 56 * XXXXXXXX 57 * X 1: Permanent or Transient UE. 58 * X 1: Effective address provided. 59 * X 1: Logical address provided. 60 * XX 2: Reserved. 61 * XXX 3: Type of UE error. 62 * 63 * For error_type == MC_ERROR_TYPE_SLB/ERAT/TLB 64 * XXXXXXXX 65 * X 1: Effective address provided. 66 * XXXXX 5: Reserved. 67 * XX 2: Type of SLB/ERAT/TLB error. 68 * 69 * For error_type == MC_ERROR_TYPE_CTRL_MEM_ACCESS 70 * XXXXXXXX 71 * X 1: Error causing address provided. 72 * XXX 3: Type of error. 73 * XXXX 4: Reserved. 74 */ 75 u8 sub_err_type; 76 u8 reserved_1[6]; 77 __be64 effective_address; 78 __be64 logical_address; 79 } __packed; 80 81 /* RTAS pseries MCE error types */ 82 #define MC_ERROR_TYPE_UE 0x00 83 #define MC_ERROR_TYPE_SLB 0x01 84 #define MC_ERROR_TYPE_ERAT 0x02 85 #define MC_ERROR_TYPE_UNKNOWN 0x03 86 #define MC_ERROR_TYPE_TLB 0x04 87 #define MC_ERROR_TYPE_D_CACHE 0x05 88 #define MC_ERROR_TYPE_I_CACHE 0x07 89 #define MC_ERROR_TYPE_CTRL_MEM_ACCESS 0x08 90 91 /* RTAS pseries MCE error sub types */ 92 #define MC_ERROR_UE_INDETERMINATE 0 93 #define MC_ERROR_UE_IFETCH 1 94 #define MC_ERROR_UE_PAGE_TABLE_WALK_IFETCH 2 95 #define MC_ERROR_UE_LOAD_STORE 3 96 #define MC_ERROR_UE_PAGE_TABLE_WALK_LOAD_STORE 4 97 98 #define UE_EFFECTIVE_ADDR_PROVIDED 0x40 99 #define UE_LOGICAL_ADDR_PROVIDED 0x20 100 #define MC_EFFECTIVE_ADDR_PROVIDED 0x80 101 102 #define MC_ERROR_SLB_PARITY 0 103 #define MC_ERROR_SLB_MULTIHIT 1 104 #define MC_ERROR_SLB_INDETERMINATE 2 105 106 #define MC_ERROR_ERAT_PARITY 1 107 #define MC_ERROR_ERAT_MULTIHIT 2 108 #define MC_ERROR_ERAT_INDETERMINATE 3 109 110 #define MC_ERROR_TLB_PARITY 1 111 #define MC_ERROR_TLB_MULTIHIT 2 112 #define MC_ERROR_TLB_INDETERMINATE 3 113 114 #define MC_ERROR_CTRL_MEM_ACCESS_PTABLE_WALK 0 115 #define MC_ERROR_CTRL_MEM_ACCESS_OP_ACCESS 1 116 117 static inline u8 rtas_mc_error_sub_type(const struct pseries_mc_errorlog *mlog) 118 { 119 switch (mlog->error_type) { 120 case MC_ERROR_TYPE_UE: 121 return (mlog->sub_err_type & 0x07); 122 case MC_ERROR_TYPE_SLB: 123 case MC_ERROR_TYPE_ERAT: 124 case MC_ERROR_TYPE_TLB: 125 return (mlog->sub_err_type & 0x03); 126 case MC_ERROR_TYPE_CTRL_MEM_ACCESS: 127 return (mlog->sub_err_type & 0x70) >> 4; 128 default: 129 return 0; 130 } 131 } 132 133 /* 134 * Enable the hotplug interrupt late because processing them may touch other 135 * devices or systems (e.g. hugepages) that have not been initialized at the 136 * subsys stage. 137 */ 138 static int __init init_ras_hotplug_IRQ(void) 139 { 140 struct device_node *np; 141 142 /* Hotplug Events */ 143 np = of_find_node_by_path("/event-sources/hot-plug-events"); 144 if (np != NULL) { 145 if (dlpar_workqueue_init() == 0) 146 request_event_sources_irqs(np, ras_hotplug_interrupt, 147 "RAS_HOTPLUG"); 148 of_node_put(np); 149 } 150 151 return 0; 152 } 153 machine_late_initcall(pseries, init_ras_hotplug_IRQ); 154 155 /* 156 * Initialize handlers for the set of interrupts caused by hardware errors 157 * and power system events. 158 */ 159 static int __init init_ras_IRQ(void) 160 { 161 struct device_node *np; 162 163 ras_check_exception_token = rtas_token("check-exception"); 164 165 /* Internal Errors */ 166 np = of_find_node_by_path("/event-sources/internal-errors"); 167 if (np != NULL) { 168 request_event_sources_irqs(np, ras_error_interrupt, 169 "RAS_ERROR"); 170 of_node_put(np); 171 } 172 173 /* EPOW Events */ 174 np = of_find_node_by_path("/event-sources/epow-events"); 175 if (np != NULL) { 176 request_event_sources_irqs(np, ras_epow_interrupt, "RAS_EPOW"); 177 of_node_put(np); 178 } 179 180 return 0; 181 } 182 machine_subsys_initcall(pseries, init_ras_IRQ); 183 184 #define EPOW_SHUTDOWN_NORMAL 1 185 #define EPOW_SHUTDOWN_ON_UPS 2 186 #define EPOW_SHUTDOWN_LOSS_OF_CRITICAL_FUNCTIONS 3 187 #define EPOW_SHUTDOWN_AMBIENT_TEMPERATURE_TOO_HIGH 4 188 189 static void handle_system_shutdown(char event_modifier) 190 { 191 switch (event_modifier) { 192 case EPOW_SHUTDOWN_NORMAL: 193 pr_emerg("Power off requested\n"); 194 orderly_poweroff(true); 195 break; 196 197 case EPOW_SHUTDOWN_ON_UPS: 198 pr_emerg("Loss of system power detected. System is running on" 199 " UPS/battery. Check RTAS error log for details\n"); 200 break; 201 202 case EPOW_SHUTDOWN_LOSS_OF_CRITICAL_FUNCTIONS: 203 pr_emerg("Loss of system critical functions detected. Check" 204 " RTAS error log for details\n"); 205 orderly_poweroff(true); 206 break; 207 208 case EPOW_SHUTDOWN_AMBIENT_TEMPERATURE_TOO_HIGH: 209 pr_emerg("High ambient temperature detected. Check RTAS" 210 " error log for details\n"); 211 orderly_poweroff(true); 212 break; 213 214 default: 215 pr_err("Unknown power/cooling shutdown event (modifier = %d)\n", 216 event_modifier); 217 } 218 } 219 220 struct epow_errorlog { 221 unsigned char sensor_value; 222 unsigned char event_modifier; 223 unsigned char extended_modifier; 224 unsigned char reserved; 225 unsigned char platform_reason; 226 }; 227 228 #define EPOW_RESET 0 229 #define EPOW_WARN_COOLING 1 230 #define EPOW_WARN_POWER 2 231 #define EPOW_SYSTEM_SHUTDOWN 3 232 #define EPOW_SYSTEM_HALT 4 233 #define EPOW_MAIN_ENCLOSURE 5 234 #define EPOW_POWER_OFF 7 235 236 static void rtas_parse_epow_errlog(struct rtas_error_log *log) 237 { 238 struct pseries_errorlog *pseries_log; 239 struct epow_errorlog *epow_log; 240 char action_code; 241 char modifier; 242 243 pseries_log = get_pseries_errorlog(log, PSERIES_ELOG_SECT_ID_EPOW); 244 if (pseries_log == NULL) 245 return; 246 247 epow_log = (struct epow_errorlog *)pseries_log->data; 248 action_code = epow_log->sensor_value & 0xF; /* bottom 4 bits */ 249 modifier = epow_log->event_modifier & 0xF; /* bottom 4 bits */ 250 251 switch (action_code) { 252 case EPOW_RESET: 253 if (num_epow_events) { 254 pr_info("Non critical power/cooling issue cleared\n"); 255 num_epow_events--; 256 } 257 break; 258 259 case EPOW_WARN_COOLING: 260 pr_info("Non-critical cooling issue detected. Check RTAS error" 261 " log for details\n"); 262 break; 263 264 case EPOW_WARN_POWER: 265 pr_info("Non-critical power issue detected. Check RTAS error" 266 " log for details\n"); 267 break; 268 269 case EPOW_SYSTEM_SHUTDOWN: 270 handle_system_shutdown(modifier); 271 break; 272 273 case EPOW_SYSTEM_HALT: 274 pr_emerg("Critical power/cooling issue detected. Check RTAS" 275 " error log for details. Powering off.\n"); 276 orderly_poweroff(true); 277 break; 278 279 case EPOW_MAIN_ENCLOSURE: 280 case EPOW_POWER_OFF: 281 pr_emerg("System about to lose power. Check RTAS error log " 282 " for details. Powering off immediately.\n"); 283 emergency_sync(); 284 kernel_power_off(); 285 break; 286 287 default: 288 pr_err("Unknown power/cooling event (action code = %d)\n", 289 action_code); 290 } 291 292 /* Increment epow events counter variable */ 293 if (action_code != EPOW_RESET) 294 num_epow_events++; 295 } 296 297 static irqreturn_t ras_hotplug_interrupt(int irq, void *dev_id) 298 { 299 struct pseries_errorlog *pseries_log; 300 struct pseries_hp_errorlog *hp_elog; 301 302 spin_lock(&ras_log_buf_lock); 303 304 rtas_call(ras_check_exception_token, 6, 1, NULL, 305 RTAS_VECTOR_EXTERNAL_INTERRUPT, virq_to_hw(irq), 306 RTAS_HOTPLUG_EVENTS, 0, __pa(&ras_log_buf), 307 rtas_get_error_log_max()); 308 309 pseries_log = get_pseries_errorlog((struct rtas_error_log *)ras_log_buf, 310 PSERIES_ELOG_SECT_ID_HOTPLUG); 311 hp_elog = (struct pseries_hp_errorlog *)pseries_log->data; 312 313 /* 314 * Since PCI hotplug is not currently supported on pseries, put PCI 315 * hotplug events on the ras_log_buf to be handled by rtas_errd. 316 */ 317 if (hp_elog->resource == PSERIES_HP_ELOG_RESOURCE_MEM || 318 hp_elog->resource == PSERIES_HP_ELOG_RESOURCE_CPU || 319 hp_elog->resource == PSERIES_HP_ELOG_RESOURCE_PMEM) 320 queue_hotplug_event(hp_elog); 321 else 322 log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, 0); 323 324 spin_unlock(&ras_log_buf_lock); 325 return IRQ_HANDLED; 326 } 327 328 /* Handle environmental and power warning (EPOW) interrupts. */ 329 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id) 330 { 331 int state; 332 int critical; 333 334 rtas_get_sensor_fast(EPOW_SENSOR_TOKEN, EPOW_SENSOR_INDEX, &state); 335 336 if (state > 3) 337 critical = 1; /* Time Critical */ 338 else 339 critical = 0; 340 341 spin_lock(&ras_log_buf_lock); 342 343 rtas_call(ras_check_exception_token, 6, 1, NULL, RTAS_VECTOR_EXTERNAL_INTERRUPT, 344 virq_to_hw(irq), RTAS_EPOW_WARNING, critical, __pa(&ras_log_buf), 345 rtas_get_error_log_max()); 346 347 log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, 0); 348 349 rtas_parse_epow_errlog((struct rtas_error_log *)ras_log_buf); 350 351 spin_unlock(&ras_log_buf_lock); 352 return IRQ_HANDLED; 353 } 354 355 /* 356 * Handle hardware error interrupts. 357 * 358 * RTAS check-exception is called to collect data on the exception. If 359 * the error is deemed recoverable, we log a warning and return. 360 * For nonrecoverable errors, an error is logged and we stop all processing 361 * as quickly as possible in order to prevent propagation of the failure. 362 */ 363 static irqreturn_t ras_error_interrupt(int irq, void *dev_id) 364 { 365 struct rtas_error_log *rtas_elog; 366 int status; 367 int fatal; 368 369 spin_lock(&ras_log_buf_lock); 370 371 status = rtas_call(ras_check_exception_token, 6, 1, NULL, 372 RTAS_VECTOR_EXTERNAL_INTERRUPT, 373 virq_to_hw(irq), 374 RTAS_INTERNAL_ERROR, 1 /* Time Critical */, 375 __pa(&ras_log_buf), 376 rtas_get_error_log_max()); 377 378 rtas_elog = (struct rtas_error_log *)ras_log_buf; 379 380 if (status == 0 && 381 rtas_error_severity(rtas_elog) >= RTAS_SEVERITY_ERROR_SYNC) 382 fatal = 1; 383 else 384 fatal = 0; 385 386 /* format and print the extended information */ 387 log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, fatal); 388 389 if (fatal) { 390 pr_emerg("Fatal hardware error detected. Check RTAS error" 391 " log for details. Powering off immediately\n"); 392 emergency_sync(); 393 kernel_power_off(); 394 } else { 395 pr_err("Recoverable hardware error detected\n"); 396 } 397 398 spin_unlock(&ras_log_buf_lock); 399 return IRQ_HANDLED; 400 } 401 402 /* 403 * Some versions of FWNMI place the buffer inside the 4kB page starting at 404 * 0x7000. Other versions place it inside the rtas buffer. We check both. 405 * Minimum size of the buffer is 16 bytes. 406 */ 407 #define VALID_FWNMI_BUFFER(A) \ 408 ((((A) >= 0x7000) && ((A) <= 0x8000 - 16)) || \ 409 (((A) >= rtas.base) && ((A) <= (rtas.base + rtas.size - 16)))) 410 411 static inline struct rtas_error_log *fwnmi_get_errlog(void) 412 { 413 return (struct rtas_error_log *)local_paca->mce_data_buf; 414 } 415 416 static __be64 *fwnmi_get_savep(struct pt_regs *regs) 417 { 418 unsigned long savep_ra; 419 420 /* Mask top two bits */ 421 savep_ra = regs->gpr[3] & ~(0x3UL << 62); 422 if (!VALID_FWNMI_BUFFER(savep_ra)) { 423 printk(KERN_ERR "FWNMI: corrupt r3 0x%016lx\n", regs->gpr[3]); 424 return NULL; 425 } 426 427 return __va(savep_ra); 428 } 429 430 /* 431 * Get the error information for errors coming through the 432 * FWNMI vectors. The pt_regs' r3 will be updated to reflect 433 * the actual r3 if possible, and a ptr to the error log entry 434 * will be returned if found. 435 * 436 * Use one buffer mce_data_buf per cpu to store RTAS error. 437 * 438 * The mce_data_buf does not have any locks or protection around it, 439 * if a second machine check comes in, or a system reset is done 440 * before we have logged the error, then we will get corruption in the 441 * error log. This is preferable over holding off on calling 442 * ibm,nmi-interlock which would result in us checkstopping if a 443 * second machine check did come in. 444 */ 445 static struct rtas_error_log *fwnmi_get_errinfo(struct pt_regs *regs) 446 { 447 struct rtas_error_log *h; 448 __be64 *savep; 449 450 savep = fwnmi_get_savep(regs); 451 if (!savep) 452 return NULL; 453 454 regs->gpr[3] = be64_to_cpu(savep[0]); /* restore original r3 */ 455 456 h = (struct rtas_error_log *)&savep[1]; 457 /* Use the per cpu buffer from paca to store rtas error log */ 458 memset(local_paca->mce_data_buf, 0, RTAS_ERROR_LOG_MAX); 459 if (!rtas_error_extended(h)) { 460 memcpy(local_paca->mce_data_buf, h, sizeof(__u64)); 461 } else { 462 int len, error_log_length; 463 464 error_log_length = 8 + rtas_error_extended_log_length(h); 465 len = min_t(int, error_log_length, RTAS_ERROR_LOG_MAX); 466 memcpy(local_paca->mce_data_buf, h, len); 467 } 468 469 return (struct rtas_error_log *)local_paca->mce_data_buf; 470 } 471 472 /* Call this when done with the data returned by FWNMI_get_errinfo. 473 * It will release the saved data area for other CPUs in the 474 * partition to receive FWNMI errors. 475 */ 476 static void fwnmi_release_errinfo(void) 477 { 478 struct rtas_args rtas_args; 479 int ret; 480 481 /* 482 * On pseries, the machine check stack is limited to under 4GB, so 483 * args can be on-stack. 484 */ 485 rtas_call_unlocked(&rtas_args, ibm_nmi_interlock_token, 0, 1, NULL); 486 ret = be32_to_cpu(rtas_args.rets[0]); 487 if (ret != 0) 488 printk(KERN_ERR "FWNMI: nmi-interlock failed: %d\n", ret); 489 } 490 491 int pSeries_system_reset_exception(struct pt_regs *regs) 492 { 493 #ifdef __LITTLE_ENDIAN__ 494 /* 495 * Some firmware byteswaps SRR registers and gives incorrect SRR1. Try 496 * to detect the bad SRR1 pattern here. Flip the NIP back to correct 497 * endian for reporting purposes. Unfortunately the MSR can't be fixed, 498 * so clear it. It will be missing MSR_RI so we won't try to recover. 499 */ 500 if ((be64_to_cpu(regs->msr) & 501 (MSR_LE|MSR_RI|MSR_DR|MSR_IR|MSR_ME|MSR_PR| 502 MSR_ILE|MSR_HV|MSR_SF)) == (MSR_DR|MSR_SF)) { 503 regs_set_return_ip(regs, be64_to_cpu((__be64)regs->nip)); 504 regs_set_return_msr(regs, 0); 505 } 506 #endif 507 508 if (fwnmi_active) { 509 __be64 *savep; 510 511 /* 512 * Firmware (PowerVM and KVM) saves r3 to a save area like 513 * machine check, which is not exactly what PAPR (2.9) 514 * suggests but there is no way to detect otherwise, so this 515 * is the interface now. 516 * 517 * System resets do not save any error log or require an 518 * "ibm,nmi-interlock" rtas call to release. 519 */ 520 521 savep = fwnmi_get_savep(regs); 522 if (savep) 523 regs->gpr[3] = be64_to_cpu(savep[0]); /* restore original r3 */ 524 } 525 526 if (smp_handle_nmi_ipi(regs)) 527 return 1; 528 529 return 0; /* need to perform reset */ 530 } 531 532 static int mce_handle_err_realmode(int disposition, u8 error_type) 533 { 534 #ifdef CONFIG_PPC_BOOK3S_64 535 if (disposition == RTAS_DISP_NOT_RECOVERED) { 536 switch (error_type) { 537 case MC_ERROR_TYPE_ERAT: 538 flush_erat(); 539 disposition = RTAS_DISP_FULLY_RECOVERED; 540 break; 541 case MC_ERROR_TYPE_SLB: 542 #ifdef CONFIG_PPC_64S_HASH_MMU 543 /* 544 * Store the old slb content in paca before flushing. 545 * Print this when we go to virtual mode. 546 * There are chances that we may hit MCE again if there 547 * is a parity error on the SLB entry we trying to read 548 * for saving. Hence limit the slb saving to single 549 * level of recursion. 550 */ 551 if (local_paca->in_mce == 1) 552 slb_save_contents(local_paca->mce_faulty_slbs); 553 flush_and_reload_slb(); 554 disposition = RTAS_DISP_FULLY_RECOVERED; 555 #endif 556 break; 557 default: 558 break; 559 } 560 } else if (disposition == RTAS_DISP_LIMITED_RECOVERY) { 561 /* Platform corrected itself but could be degraded */ 562 pr_err("MCE: limited recovery, system may be degraded\n"); 563 disposition = RTAS_DISP_FULLY_RECOVERED; 564 } 565 #endif 566 return disposition; 567 } 568 569 static int mce_handle_err_virtmode(struct pt_regs *regs, 570 struct rtas_error_log *errp, 571 struct pseries_mc_errorlog *mce_log, 572 int disposition) 573 { 574 struct mce_error_info mce_err = { 0 }; 575 int initiator = rtas_error_initiator(errp); 576 int severity = rtas_error_severity(errp); 577 unsigned long eaddr = 0, paddr = 0; 578 u8 error_type, err_sub_type; 579 580 if (!mce_log) 581 goto out; 582 583 error_type = mce_log->error_type; 584 err_sub_type = rtas_mc_error_sub_type(mce_log); 585 586 if (initiator == RTAS_INITIATOR_UNKNOWN) 587 mce_err.initiator = MCE_INITIATOR_UNKNOWN; 588 else if (initiator == RTAS_INITIATOR_CPU) 589 mce_err.initiator = MCE_INITIATOR_CPU; 590 else if (initiator == RTAS_INITIATOR_PCI) 591 mce_err.initiator = MCE_INITIATOR_PCI; 592 else if (initiator == RTAS_INITIATOR_ISA) 593 mce_err.initiator = MCE_INITIATOR_ISA; 594 else if (initiator == RTAS_INITIATOR_MEMORY) 595 mce_err.initiator = MCE_INITIATOR_MEMORY; 596 else if (initiator == RTAS_INITIATOR_POWERMGM) 597 mce_err.initiator = MCE_INITIATOR_POWERMGM; 598 else 599 mce_err.initiator = MCE_INITIATOR_UNKNOWN; 600 601 if (severity == RTAS_SEVERITY_NO_ERROR) 602 mce_err.severity = MCE_SEV_NO_ERROR; 603 else if (severity == RTAS_SEVERITY_EVENT) 604 mce_err.severity = MCE_SEV_WARNING; 605 else if (severity == RTAS_SEVERITY_WARNING) 606 mce_err.severity = MCE_SEV_WARNING; 607 else if (severity == RTAS_SEVERITY_ERROR_SYNC) 608 mce_err.severity = MCE_SEV_SEVERE; 609 else if (severity == RTAS_SEVERITY_ERROR) 610 mce_err.severity = MCE_SEV_SEVERE; 611 else 612 mce_err.severity = MCE_SEV_FATAL; 613 614 if (severity <= RTAS_SEVERITY_ERROR_SYNC) 615 mce_err.sync_error = true; 616 else 617 mce_err.sync_error = false; 618 619 mce_err.error_type = MCE_ERROR_TYPE_UNKNOWN; 620 mce_err.error_class = MCE_ECLASS_UNKNOWN; 621 622 switch (error_type) { 623 case MC_ERROR_TYPE_UE: 624 mce_err.error_type = MCE_ERROR_TYPE_UE; 625 mce_common_process_ue(regs, &mce_err); 626 if (mce_err.ignore_event) 627 disposition = RTAS_DISP_FULLY_RECOVERED; 628 switch (err_sub_type) { 629 case MC_ERROR_UE_IFETCH: 630 mce_err.u.ue_error_type = MCE_UE_ERROR_IFETCH; 631 break; 632 case MC_ERROR_UE_PAGE_TABLE_WALK_IFETCH: 633 mce_err.u.ue_error_type = MCE_UE_ERROR_PAGE_TABLE_WALK_IFETCH; 634 break; 635 case MC_ERROR_UE_LOAD_STORE: 636 mce_err.u.ue_error_type = MCE_UE_ERROR_LOAD_STORE; 637 break; 638 case MC_ERROR_UE_PAGE_TABLE_WALK_LOAD_STORE: 639 mce_err.u.ue_error_type = MCE_UE_ERROR_PAGE_TABLE_WALK_LOAD_STORE; 640 break; 641 case MC_ERROR_UE_INDETERMINATE: 642 default: 643 mce_err.u.ue_error_type = MCE_UE_ERROR_INDETERMINATE; 644 break; 645 } 646 if (mce_log->sub_err_type & UE_EFFECTIVE_ADDR_PROVIDED) 647 eaddr = be64_to_cpu(mce_log->effective_address); 648 649 if (mce_log->sub_err_type & UE_LOGICAL_ADDR_PROVIDED) { 650 paddr = be64_to_cpu(mce_log->logical_address); 651 } else if (mce_log->sub_err_type & UE_EFFECTIVE_ADDR_PROVIDED) { 652 unsigned long pfn; 653 654 pfn = addr_to_pfn(regs, eaddr); 655 if (pfn != ULONG_MAX) 656 paddr = pfn << PAGE_SHIFT; 657 } 658 659 break; 660 case MC_ERROR_TYPE_SLB: 661 mce_err.error_type = MCE_ERROR_TYPE_SLB; 662 switch (err_sub_type) { 663 case MC_ERROR_SLB_PARITY: 664 mce_err.u.slb_error_type = MCE_SLB_ERROR_PARITY; 665 break; 666 case MC_ERROR_SLB_MULTIHIT: 667 mce_err.u.slb_error_type = MCE_SLB_ERROR_MULTIHIT; 668 break; 669 case MC_ERROR_SLB_INDETERMINATE: 670 default: 671 mce_err.u.slb_error_type = MCE_SLB_ERROR_INDETERMINATE; 672 break; 673 } 674 if (mce_log->sub_err_type & MC_EFFECTIVE_ADDR_PROVIDED) 675 eaddr = be64_to_cpu(mce_log->effective_address); 676 break; 677 case MC_ERROR_TYPE_ERAT: 678 mce_err.error_type = MCE_ERROR_TYPE_ERAT; 679 switch (err_sub_type) { 680 case MC_ERROR_ERAT_PARITY: 681 mce_err.u.erat_error_type = MCE_ERAT_ERROR_PARITY; 682 break; 683 case MC_ERROR_ERAT_MULTIHIT: 684 mce_err.u.erat_error_type = MCE_ERAT_ERROR_MULTIHIT; 685 break; 686 case MC_ERROR_ERAT_INDETERMINATE: 687 default: 688 mce_err.u.erat_error_type = MCE_ERAT_ERROR_INDETERMINATE; 689 break; 690 } 691 if (mce_log->sub_err_type & MC_EFFECTIVE_ADDR_PROVIDED) 692 eaddr = be64_to_cpu(mce_log->effective_address); 693 break; 694 case MC_ERROR_TYPE_TLB: 695 mce_err.error_type = MCE_ERROR_TYPE_TLB; 696 switch (err_sub_type) { 697 case MC_ERROR_TLB_PARITY: 698 mce_err.u.tlb_error_type = MCE_TLB_ERROR_PARITY; 699 break; 700 case MC_ERROR_TLB_MULTIHIT: 701 mce_err.u.tlb_error_type = MCE_TLB_ERROR_MULTIHIT; 702 break; 703 case MC_ERROR_TLB_INDETERMINATE: 704 default: 705 mce_err.u.tlb_error_type = MCE_TLB_ERROR_INDETERMINATE; 706 break; 707 } 708 if (mce_log->sub_err_type & MC_EFFECTIVE_ADDR_PROVIDED) 709 eaddr = be64_to_cpu(mce_log->effective_address); 710 break; 711 case MC_ERROR_TYPE_D_CACHE: 712 mce_err.error_type = MCE_ERROR_TYPE_DCACHE; 713 break; 714 case MC_ERROR_TYPE_I_CACHE: 715 mce_err.error_type = MCE_ERROR_TYPE_ICACHE; 716 break; 717 case MC_ERROR_TYPE_CTRL_MEM_ACCESS: 718 mce_err.error_type = MCE_ERROR_TYPE_RA; 719 switch (err_sub_type) { 720 case MC_ERROR_CTRL_MEM_ACCESS_PTABLE_WALK: 721 mce_err.u.ra_error_type = 722 MCE_RA_ERROR_PAGE_TABLE_WALK_LOAD_STORE_FOREIGN; 723 break; 724 case MC_ERROR_CTRL_MEM_ACCESS_OP_ACCESS: 725 mce_err.u.ra_error_type = 726 MCE_RA_ERROR_LOAD_STORE_FOREIGN; 727 break; 728 } 729 if (mce_log->sub_err_type & MC_EFFECTIVE_ADDR_PROVIDED) 730 eaddr = be64_to_cpu(mce_log->effective_address); 731 break; 732 case MC_ERROR_TYPE_UNKNOWN: 733 default: 734 mce_err.error_type = MCE_ERROR_TYPE_UNKNOWN; 735 break; 736 } 737 out: 738 save_mce_event(regs, disposition == RTAS_DISP_FULLY_RECOVERED, 739 &mce_err, regs->nip, eaddr, paddr); 740 return disposition; 741 } 742 743 static int mce_handle_error(struct pt_regs *regs, struct rtas_error_log *errp) 744 { 745 struct pseries_errorlog *pseries_log; 746 struct pseries_mc_errorlog *mce_log = NULL; 747 int disposition = rtas_error_disposition(errp); 748 unsigned long msr; 749 u8 error_type; 750 751 if (!rtas_error_extended(errp)) 752 goto out; 753 754 pseries_log = get_pseries_errorlog(errp, PSERIES_ELOG_SECT_ID_MCE); 755 if (!pseries_log) 756 goto out; 757 758 mce_log = (struct pseries_mc_errorlog *)pseries_log->data; 759 error_type = mce_log->error_type; 760 761 disposition = mce_handle_err_realmode(disposition, error_type); 762 763 /* 764 * Enable translation as we will be accessing per-cpu variables 765 * in save_mce_event() which may fall outside RMO region, also 766 * leave it enabled because subsequently we will be queuing work 767 * to workqueues where again per-cpu variables accessed, besides 768 * fwnmi_release_errinfo() crashes when called in realmode on 769 * pseries. 770 * Note: All the realmode handling like flushing SLB entries for 771 * SLB multihit is done by now. 772 */ 773 out: 774 msr = mfmsr(); 775 mtmsr(msr | MSR_IR | MSR_DR); 776 777 disposition = mce_handle_err_virtmode(regs, errp, mce_log, 778 disposition); 779 780 /* 781 * Queue irq work to log this rtas event later. 782 * irq_work_queue uses per-cpu variables, so do this in virt 783 * mode as well. 784 */ 785 irq_work_queue(&mce_errlog_process_work); 786 787 mtmsr(msr); 788 789 return disposition; 790 } 791 792 /* 793 * Process MCE rtas errlog event. 794 */ 795 static void mce_process_errlog_event(struct irq_work *work) 796 { 797 struct rtas_error_log *err; 798 799 err = fwnmi_get_errlog(); 800 log_error((char *)err, ERR_TYPE_RTAS_LOG, 0); 801 } 802 803 /* 804 * See if we can recover from a machine check exception. 805 * This is only called on power4 (or above) and only via 806 * the Firmware Non-Maskable Interrupts (fwnmi) handler 807 * which provides the error analysis for us. 808 * 809 * Return 1 if corrected (or delivered a signal). 810 * Return 0 if there is nothing we can do. 811 */ 812 static int recover_mce(struct pt_regs *regs, struct machine_check_event *evt) 813 { 814 int recovered = 0; 815 816 if (regs_is_unrecoverable(regs)) { 817 /* If MSR_RI isn't set, we cannot recover */ 818 pr_err("Machine check interrupt unrecoverable: MSR(RI=0)\n"); 819 recovered = 0; 820 } else if (evt->disposition == MCE_DISPOSITION_RECOVERED) { 821 /* Platform corrected itself */ 822 recovered = 1; 823 } else if (evt->severity == MCE_SEV_FATAL) { 824 /* Fatal machine check */ 825 pr_err("Machine check interrupt is fatal\n"); 826 recovered = 0; 827 } 828 829 if (!recovered && evt->sync_error) { 830 /* 831 * Try to kill processes if we get a synchronous machine check 832 * (e.g., one caused by execution of this instruction). This 833 * will devolve into a panic if we try to kill init or are in 834 * an interrupt etc. 835 * 836 * TODO: Queue up this address for hwpoisioning later. 837 * TODO: This is not quite right for d-side machine 838 * checks ->nip is not necessarily the important 839 * address. 840 */ 841 if ((user_mode(regs))) { 842 _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip); 843 recovered = 1; 844 } else if (die_will_crash()) { 845 /* 846 * die() would kill the kernel, so better to go via 847 * the platform reboot code that will log the 848 * machine check. 849 */ 850 recovered = 0; 851 } else { 852 die_mce("Machine check", regs, SIGBUS); 853 recovered = 1; 854 } 855 } 856 857 return recovered; 858 } 859 860 /* 861 * Handle a machine check. 862 * 863 * Note that on Power 4 and beyond Firmware Non-Maskable Interrupts (fwnmi) 864 * should be present. If so the handler which called us tells us if the 865 * error was recovered (never true if RI=0). 866 * 867 * On hardware prior to Power 4 these exceptions were asynchronous which 868 * means we can't tell exactly where it occurred and so we can't recover. 869 */ 870 int pSeries_machine_check_exception(struct pt_regs *regs) 871 { 872 struct machine_check_event evt; 873 874 if (!get_mce_event(&evt, MCE_EVENT_RELEASE)) 875 return 0; 876 877 /* Print things out */ 878 if (evt.version != MCE_V1) { 879 pr_err("Machine Check Exception, Unknown event version %d !\n", 880 evt.version); 881 return 0; 882 } 883 machine_check_print_event_info(&evt, user_mode(regs), false); 884 885 if (recover_mce(regs, &evt)) 886 return 1; 887 888 return 0; 889 } 890 891 long pseries_machine_check_realmode(struct pt_regs *regs) 892 { 893 struct rtas_error_log *errp; 894 int disposition; 895 896 if (fwnmi_active) { 897 errp = fwnmi_get_errinfo(regs); 898 /* 899 * Call to fwnmi_release_errinfo() in real mode causes kernel 900 * to panic. Hence we will call it as soon as we go into 901 * virtual mode. 902 */ 903 disposition = mce_handle_error(regs, errp); 904 905 fwnmi_release_errinfo(); 906 907 if (disposition == RTAS_DISP_FULLY_RECOVERED) 908 return 1; 909 } 910 911 return 0; 912 } 913