1 /* 2 * Copyright (C) 2001 Dave Engebretsen IBM Corporation 3 * 4 * This program is free software; you can redistribute it and/or modify 5 * it under the terms of the GNU General Public License as published by 6 * the Free Software Foundation; either version 2 of the License, or 7 * (at your option) any later version. 8 * 9 * This program is distributed in the hope that it will be useful, 10 * but WITHOUT ANY WARRANTY; without even the implied warranty of 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 12 * GNU General Public License for more details. 13 * 14 * You should have received a copy of the GNU General Public License 15 * along with this program; if not, write to the Free Software 16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA 17 */ 18 19 #include <linux/sched.h> 20 #include <linux/interrupt.h> 21 #include <linux/irq.h> 22 #include <linux/of.h> 23 #include <linux/fs.h> 24 #include <linux/reboot.h> 25 #include <linux/irq_work.h> 26 27 #include <asm/machdep.h> 28 #include <asm/rtas.h> 29 #include <asm/firmware.h> 30 #include <asm/mce.h> 31 32 #include "pseries.h" 33 34 static unsigned char ras_log_buf[RTAS_ERROR_LOG_MAX]; 35 static DEFINE_SPINLOCK(ras_log_buf_lock); 36 37 static int ras_check_exception_token; 38 39 static void mce_process_errlog_event(struct irq_work *work); 40 static struct irq_work mce_errlog_process_work = { 41 .func = mce_process_errlog_event, 42 }; 43 44 #define EPOW_SENSOR_TOKEN 9 45 #define EPOW_SENSOR_INDEX 0 46 47 /* EPOW events counter variable */ 48 static int num_epow_events; 49 50 static irqreturn_t ras_hotplug_interrupt(int irq, void *dev_id); 51 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id); 52 static irqreturn_t ras_error_interrupt(int irq, void *dev_id); 53 54 /* RTAS pseries MCE errorlog section. */ 55 struct pseries_mc_errorlog { 56 __be32 fru_id; 57 __be32 proc_id; 58 u8 error_type; 59 /* 60 * sub_err_type (1 byte). Bit fields depends on error_type 61 * 62 * MSB0 63 * | 64 * V 65 * 01234567 66 * XXXXXXXX 67 * 68 * For error_type == MC_ERROR_TYPE_UE 69 * XXXXXXXX 70 * X 1: Permanent or Transient UE. 71 * X 1: Effective address provided. 72 * X 1: Logical address provided. 73 * XX 2: Reserved. 74 * XXX 3: Type of UE error. 75 * 76 * For error_type != MC_ERROR_TYPE_UE 77 * XXXXXXXX 78 * X 1: Effective address provided. 79 * XXXXX 5: Reserved. 80 * XX 2: Type of SLB/ERAT/TLB error. 81 */ 82 u8 sub_err_type; 83 u8 reserved_1[6]; 84 __be64 effective_address; 85 __be64 logical_address; 86 } __packed; 87 88 /* RTAS pseries MCE error types */ 89 #define MC_ERROR_TYPE_UE 0x00 90 #define MC_ERROR_TYPE_SLB 0x01 91 #define MC_ERROR_TYPE_ERAT 0x02 92 #define MC_ERROR_TYPE_TLB 0x04 93 #define MC_ERROR_TYPE_D_CACHE 0x05 94 #define MC_ERROR_TYPE_I_CACHE 0x07 95 96 /* RTAS pseries MCE error sub types */ 97 #define MC_ERROR_UE_INDETERMINATE 0 98 #define MC_ERROR_UE_IFETCH 1 99 #define MC_ERROR_UE_PAGE_TABLE_WALK_IFETCH 2 100 #define MC_ERROR_UE_LOAD_STORE 3 101 #define MC_ERROR_UE_PAGE_TABLE_WALK_LOAD_STORE 4 102 103 #define MC_ERROR_SLB_PARITY 0 104 #define MC_ERROR_SLB_MULTIHIT 1 105 #define MC_ERROR_SLB_INDETERMINATE 2 106 107 #define MC_ERROR_ERAT_PARITY 1 108 #define MC_ERROR_ERAT_MULTIHIT 2 109 #define MC_ERROR_ERAT_INDETERMINATE 3 110 111 #define MC_ERROR_TLB_PARITY 1 112 #define MC_ERROR_TLB_MULTIHIT 2 113 #define MC_ERROR_TLB_INDETERMINATE 3 114 115 static inline u8 rtas_mc_error_sub_type(const struct pseries_mc_errorlog *mlog) 116 { 117 switch (mlog->error_type) { 118 case MC_ERROR_TYPE_UE: 119 return (mlog->sub_err_type & 0x07); 120 case MC_ERROR_TYPE_SLB: 121 case MC_ERROR_TYPE_ERAT: 122 case MC_ERROR_TYPE_TLB: 123 return (mlog->sub_err_type & 0x03); 124 default: 125 return 0; 126 } 127 } 128 129 static 130 inline u64 rtas_mc_get_effective_addr(const struct pseries_mc_errorlog *mlog) 131 { 132 __be64 addr = 0; 133 134 switch (mlog->error_type) { 135 case MC_ERROR_TYPE_UE: 136 if (mlog->sub_err_type & 0x40) 137 addr = mlog->effective_address; 138 break; 139 case MC_ERROR_TYPE_SLB: 140 case MC_ERROR_TYPE_ERAT: 141 case MC_ERROR_TYPE_TLB: 142 if (mlog->sub_err_type & 0x80) 143 addr = mlog->effective_address; 144 default: 145 break; 146 } 147 return be64_to_cpu(addr); 148 } 149 150 /* 151 * Enable the hotplug interrupt late because processing them may touch other 152 * devices or systems (e.g. hugepages) that have not been initialized at the 153 * subsys stage. 154 */ 155 int __init init_ras_hotplug_IRQ(void) 156 { 157 struct device_node *np; 158 159 /* Hotplug Events */ 160 np = of_find_node_by_path("/event-sources/hot-plug-events"); 161 if (np != NULL) { 162 if (dlpar_workqueue_init() == 0) 163 request_event_sources_irqs(np, ras_hotplug_interrupt, 164 "RAS_HOTPLUG"); 165 of_node_put(np); 166 } 167 168 return 0; 169 } 170 machine_late_initcall(pseries, init_ras_hotplug_IRQ); 171 172 /* 173 * Initialize handlers for the set of interrupts caused by hardware errors 174 * and power system events. 175 */ 176 static int __init init_ras_IRQ(void) 177 { 178 struct device_node *np; 179 180 ras_check_exception_token = rtas_token("check-exception"); 181 182 /* Internal Errors */ 183 np = of_find_node_by_path("/event-sources/internal-errors"); 184 if (np != NULL) { 185 request_event_sources_irqs(np, ras_error_interrupt, 186 "RAS_ERROR"); 187 of_node_put(np); 188 } 189 190 /* EPOW Events */ 191 np = of_find_node_by_path("/event-sources/epow-events"); 192 if (np != NULL) { 193 request_event_sources_irqs(np, ras_epow_interrupt, "RAS_EPOW"); 194 of_node_put(np); 195 } 196 197 return 0; 198 } 199 machine_subsys_initcall(pseries, init_ras_IRQ); 200 201 #define EPOW_SHUTDOWN_NORMAL 1 202 #define EPOW_SHUTDOWN_ON_UPS 2 203 #define EPOW_SHUTDOWN_LOSS_OF_CRITICAL_FUNCTIONS 3 204 #define EPOW_SHUTDOWN_AMBIENT_TEMPERATURE_TOO_HIGH 4 205 206 static void handle_system_shutdown(char event_modifier) 207 { 208 switch (event_modifier) { 209 case EPOW_SHUTDOWN_NORMAL: 210 pr_emerg("Power off requested\n"); 211 orderly_poweroff(true); 212 break; 213 214 case EPOW_SHUTDOWN_ON_UPS: 215 pr_emerg("Loss of system power detected. System is running on" 216 " UPS/battery. Check RTAS error log for details\n"); 217 orderly_poweroff(true); 218 break; 219 220 case EPOW_SHUTDOWN_LOSS_OF_CRITICAL_FUNCTIONS: 221 pr_emerg("Loss of system critical functions detected. Check" 222 " RTAS error log for details\n"); 223 orderly_poweroff(true); 224 break; 225 226 case EPOW_SHUTDOWN_AMBIENT_TEMPERATURE_TOO_HIGH: 227 pr_emerg("High ambient temperature detected. Check RTAS" 228 " error log for details\n"); 229 orderly_poweroff(true); 230 break; 231 232 default: 233 pr_err("Unknown power/cooling shutdown event (modifier = %d)\n", 234 event_modifier); 235 } 236 } 237 238 struct epow_errorlog { 239 unsigned char sensor_value; 240 unsigned char event_modifier; 241 unsigned char extended_modifier; 242 unsigned char reserved; 243 unsigned char platform_reason; 244 }; 245 246 #define EPOW_RESET 0 247 #define EPOW_WARN_COOLING 1 248 #define EPOW_WARN_POWER 2 249 #define EPOW_SYSTEM_SHUTDOWN 3 250 #define EPOW_SYSTEM_HALT 4 251 #define EPOW_MAIN_ENCLOSURE 5 252 #define EPOW_POWER_OFF 7 253 254 static void rtas_parse_epow_errlog(struct rtas_error_log *log) 255 { 256 struct pseries_errorlog *pseries_log; 257 struct epow_errorlog *epow_log; 258 char action_code; 259 char modifier; 260 261 pseries_log = get_pseries_errorlog(log, PSERIES_ELOG_SECT_ID_EPOW); 262 if (pseries_log == NULL) 263 return; 264 265 epow_log = (struct epow_errorlog *)pseries_log->data; 266 action_code = epow_log->sensor_value & 0xF; /* bottom 4 bits */ 267 modifier = epow_log->event_modifier & 0xF; /* bottom 4 bits */ 268 269 switch (action_code) { 270 case EPOW_RESET: 271 if (num_epow_events) { 272 pr_info("Non critical power/cooling issue cleared\n"); 273 num_epow_events--; 274 } 275 break; 276 277 case EPOW_WARN_COOLING: 278 pr_info("Non-critical cooling issue detected. Check RTAS error" 279 " log for details\n"); 280 break; 281 282 case EPOW_WARN_POWER: 283 pr_info("Non-critical power issue detected. Check RTAS error" 284 " log for details\n"); 285 break; 286 287 case EPOW_SYSTEM_SHUTDOWN: 288 handle_system_shutdown(epow_log->event_modifier); 289 break; 290 291 case EPOW_SYSTEM_HALT: 292 pr_emerg("Critical power/cooling issue detected. Check RTAS" 293 " error log for details. Powering off.\n"); 294 orderly_poweroff(true); 295 break; 296 297 case EPOW_MAIN_ENCLOSURE: 298 case EPOW_POWER_OFF: 299 pr_emerg("System about to lose power. Check RTAS error log " 300 " for details. Powering off immediately.\n"); 301 emergency_sync(); 302 kernel_power_off(); 303 break; 304 305 default: 306 pr_err("Unknown power/cooling event (action code = %d)\n", 307 action_code); 308 } 309 310 /* Increment epow events counter variable */ 311 if (action_code != EPOW_RESET) 312 num_epow_events++; 313 } 314 315 static irqreturn_t ras_hotplug_interrupt(int irq, void *dev_id) 316 { 317 struct pseries_errorlog *pseries_log; 318 struct pseries_hp_errorlog *hp_elog; 319 320 spin_lock(&ras_log_buf_lock); 321 322 rtas_call(ras_check_exception_token, 6, 1, NULL, 323 RTAS_VECTOR_EXTERNAL_INTERRUPT, virq_to_hw(irq), 324 RTAS_HOTPLUG_EVENTS, 0, __pa(&ras_log_buf), 325 rtas_get_error_log_max()); 326 327 pseries_log = get_pseries_errorlog((struct rtas_error_log *)ras_log_buf, 328 PSERIES_ELOG_SECT_ID_HOTPLUG); 329 hp_elog = (struct pseries_hp_errorlog *)pseries_log->data; 330 331 /* 332 * Since PCI hotplug is not currently supported on pseries, put PCI 333 * hotplug events on the ras_log_buf to be handled by rtas_errd. 334 */ 335 if (hp_elog->resource == PSERIES_HP_ELOG_RESOURCE_MEM || 336 hp_elog->resource == PSERIES_HP_ELOG_RESOURCE_CPU || 337 hp_elog->resource == PSERIES_HP_ELOG_RESOURCE_PMEM) 338 queue_hotplug_event(hp_elog); 339 else 340 log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, 0); 341 342 spin_unlock(&ras_log_buf_lock); 343 return IRQ_HANDLED; 344 } 345 346 /* Handle environmental and power warning (EPOW) interrupts. */ 347 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id) 348 { 349 int status; 350 int state; 351 int critical; 352 353 status = rtas_get_sensor_fast(EPOW_SENSOR_TOKEN, EPOW_SENSOR_INDEX, 354 &state); 355 356 if (state > 3) 357 critical = 1; /* Time Critical */ 358 else 359 critical = 0; 360 361 spin_lock(&ras_log_buf_lock); 362 363 status = rtas_call(ras_check_exception_token, 6, 1, NULL, 364 RTAS_VECTOR_EXTERNAL_INTERRUPT, 365 virq_to_hw(irq), 366 RTAS_EPOW_WARNING, 367 critical, __pa(&ras_log_buf), 368 rtas_get_error_log_max()); 369 370 log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, 0); 371 372 rtas_parse_epow_errlog((struct rtas_error_log *)ras_log_buf); 373 374 spin_unlock(&ras_log_buf_lock); 375 return IRQ_HANDLED; 376 } 377 378 /* 379 * Handle hardware error interrupts. 380 * 381 * RTAS check-exception is called to collect data on the exception. If 382 * the error is deemed recoverable, we log a warning and return. 383 * For nonrecoverable errors, an error is logged and we stop all processing 384 * as quickly as possible in order to prevent propagation of the failure. 385 */ 386 static irqreturn_t ras_error_interrupt(int irq, void *dev_id) 387 { 388 struct rtas_error_log *rtas_elog; 389 int status; 390 int fatal; 391 392 spin_lock(&ras_log_buf_lock); 393 394 status = rtas_call(ras_check_exception_token, 6, 1, NULL, 395 RTAS_VECTOR_EXTERNAL_INTERRUPT, 396 virq_to_hw(irq), 397 RTAS_INTERNAL_ERROR, 1 /* Time Critical */, 398 __pa(&ras_log_buf), 399 rtas_get_error_log_max()); 400 401 rtas_elog = (struct rtas_error_log *)ras_log_buf; 402 403 if (status == 0 && 404 rtas_error_severity(rtas_elog) >= RTAS_SEVERITY_ERROR_SYNC) 405 fatal = 1; 406 else 407 fatal = 0; 408 409 /* format and print the extended information */ 410 log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, fatal); 411 412 if (fatal) { 413 pr_emerg("Fatal hardware error detected. Check RTAS error" 414 " log for details. Powering off immediately\n"); 415 emergency_sync(); 416 kernel_power_off(); 417 } else { 418 pr_err("Recoverable hardware error detected\n"); 419 } 420 421 spin_unlock(&ras_log_buf_lock); 422 return IRQ_HANDLED; 423 } 424 425 /* 426 * Some versions of FWNMI place the buffer inside the 4kB page starting at 427 * 0x7000. Other versions place it inside the rtas buffer. We check both. 428 */ 429 #define VALID_FWNMI_BUFFER(A) \ 430 ((((A) >= 0x7000) && ((A) < 0x7ff0)) || \ 431 (((A) >= rtas.base) && ((A) < (rtas.base + rtas.size - 16)))) 432 433 static inline struct rtas_error_log *fwnmi_get_errlog(void) 434 { 435 return (struct rtas_error_log *)local_paca->mce_data_buf; 436 } 437 438 /* 439 * Get the error information for errors coming through the 440 * FWNMI vectors. The pt_regs' r3 will be updated to reflect 441 * the actual r3 if possible, and a ptr to the error log entry 442 * will be returned if found. 443 * 444 * Use one buffer mce_data_buf per cpu to store RTAS error. 445 * 446 * The mce_data_buf does not have any locks or protection around it, 447 * if a second machine check comes in, or a system reset is done 448 * before we have logged the error, then we will get corruption in the 449 * error log. This is preferable over holding off on calling 450 * ibm,nmi-interlock which would result in us checkstopping if a 451 * second machine check did come in. 452 */ 453 static struct rtas_error_log *fwnmi_get_errinfo(struct pt_regs *regs) 454 { 455 unsigned long *savep; 456 struct rtas_error_log *h; 457 458 /* Mask top two bits */ 459 regs->gpr[3] &= ~(0x3UL << 62); 460 461 if (!VALID_FWNMI_BUFFER(regs->gpr[3])) { 462 printk(KERN_ERR "FWNMI: corrupt r3 0x%016lx\n", regs->gpr[3]); 463 return NULL; 464 } 465 466 savep = __va(regs->gpr[3]); 467 regs->gpr[3] = be64_to_cpu(savep[0]); /* restore original r3 */ 468 469 h = (struct rtas_error_log *)&savep[1]; 470 /* Use the per cpu buffer from paca to store rtas error log */ 471 memset(local_paca->mce_data_buf, 0, RTAS_ERROR_LOG_MAX); 472 if (!rtas_error_extended(h)) { 473 memcpy(local_paca->mce_data_buf, h, sizeof(__u64)); 474 } else { 475 int len, error_log_length; 476 477 error_log_length = 8 + rtas_error_extended_log_length(h); 478 len = min_t(int, error_log_length, RTAS_ERROR_LOG_MAX); 479 memcpy(local_paca->mce_data_buf, h, len); 480 } 481 482 return (struct rtas_error_log *)local_paca->mce_data_buf; 483 } 484 485 /* Call this when done with the data returned by FWNMI_get_errinfo. 486 * It will release the saved data area for other CPUs in the 487 * partition to receive FWNMI errors. 488 */ 489 static void fwnmi_release_errinfo(void) 490 { 491 int ret = rtas_call(rtas_token("ibm,nmi-interlock"), 0, 1, NULL); 492 if (ret != 0) 493 printk(KERN_ERR "FWNMI: nmi-interlock failed: %d\n", ret); 494 } 495 496 int pSeries_system_reset_exception(struct pt_regs *regs) 497 { 498 #ifdef __LITTLE_ENDIAN__ 499 /* 500 * Some firmware byteswaps SRR registers and gives incorrect SRR1. Try 501 * to detect the bad SRR1 pattern here. Flip the NIP back to correct 502 * endian for reporting purposes. Unfortunately the MSR can't be fixed, 503 * so clear it. It will be missing MSR_RI so we won't try to recover. 504 */ 505 if ((be64_to_cpu(regs->msr) & 506 (MSR_LE|MSR_RI|MSR_DR|MSR_IR|MSR_ME|MSR_PR| 507 MSR_ILE|MSR_HV|MSR_SF)) == (MSR_DR|MSR_SF)) { 508 regs->nip = be64_to_cpu((__be64)regs->nip); 509 regs->msr = 0; 510 } 511 #endif 512 513 if (fwnmi_active) { 514 struct rtas_error_log *errhdr = fwnmi_get_errinfo(regs); 515 if (errhdr) { 516 /* XXX Should look at FWNMI information */ 517 } 518 fwnmi_release_errinfo(); 519 } 520 521 if (smp_handle_nmi_ipi(regs)) 522 return 1; 523 524 return 0; /* need to perform reset */ 525 } 526 527 #define VAL_TO_STRING(ar, val) \ 528 (((val) < ARRAY_SIZE(ar)) ? ar[(val)] : "Unknown") 529 530 static void pseries_print_mce_info(struct pt_regs *regs, 531 struct rtas_error_log *errp) 532 { 533 const char *level, *sevstr; 534 struct pseries_errorlog *pseries_log; 535 struct pseries_mc_errorlog *mce_log; 536 u8 error_type, err_sub_type; 537 u64 addr; 538 u8 initiator = rtas_error_initiator(errp); 539 int disposition = rtas_error_disposition(errp); 540 541 static const char * const initiators[] = { 542 [0] = "Unknown", 543 [1] = "CPU", 544 [2] = "PCI", 545 [3] = "ISA", 546 [4] = "Memory", 547 [5] = "Power Mgmt", 548 }; 549 static const char * const mc_err_types[] = { 550 [0] = "UE", 551 [1] = "SLB", 552 [2] = "ERAT", 553 [3] = "Unknown", 554 [4] = "TLB", 555 [5] = "D-Cache", 556 [6] = "Unknown", 557 [7] = "I-Cache", 558 }; 559 static const char * const mc_ue_types[] = { 560 [0] = "Indeterminate", 561 [1] = "Instruction fetch", 562 [2] = "Page table walk ifetch", 563 [3] = "Load/Store", 564 [4] = "Page table walk Load/Store", 565 }; 566 567 /* SLB sub errors valid values are 0x0, 0x1, 0x2 */ 568 static const char * const mc_slb_types[] = { 569 [0] = "Parity", 570 [1] = "Multihit", 571 [2] = "Indeterminate", 572 }; 573 574 /* TLB and ERAT sub errors valid values are 0x1, 0x2, 0x3 */ 575 static const char * const mc_soft_types[] = { 576 [0] = "Unknown", 577 [1] = "Parity", 578 [2] = "Multihit", 579 [3] = "Indeterminate", 580 }; 581 582 if (!rtas_error_extended(errp)) { 583 pr_err("Machine check interrupt: Missing extended error log\n"); 584 return; 585 } 586 587 pseries_log = get_pseries_errorlog(errp, PSERIES_ELOG_SECT_ID_MCE); 588 if (pseries_log == NULL) 589 return; 590 591 mce_log = (struct pseries_mc_errorlog *)pseries_log->data; 592 593 error_type = mce_log->error_type; 594 err_sub_type = rtas_mc_error_sub_type(mce_log); 595 596 switch (rtas_error_severity(errp)) { 597 case RTAS_SEVERITY_NO_ERROR: 598 level = KERN_INFO; 599 sevstr = "Harmless"; 600 break; 601 case RTAS_SEVERITY_WARNING: 602 level = KERN_WARNING; 603 sevstr = ""; 604 break; 605 case RTAS_SEVERITY_ERROR: 606 case RTAS_SEVERITY_ERROR_SYNC: 607 level = KERN_ERR; 608 sevstr = "Severe"; 609 break; 610 case RTAS_SEVERITY_FATAL: 611 default: 612 level = KERN_ERR; 613 sevstr = "Fatal"; 614 break; 615 } 616 617 #ifdef CONFIG_PPC_BOOK3S_64 618 /* Display faulty slb contents for SLB errors. */ 619 if (error_type == MC_ERROR_TYPE_SLB) 620 slb_dump_contents(local_paca->mce_faulty_slbs); 621 #endif 622 623 printk("%s%s Machine check interrupt [%s]\n", level, sevstr, 624 disposition == RTAS_DISP_FULLY_RECOVERED ? 625 "Recovered" : "Not recovered"); 626 if (user_mode(regs)) { 627 printk("%s NIP: [%016lx] PID: %d Comm: %s\n", level, 628 regs->nip, current->pid, current->comm); 629 } else { 630 printk("%s NIP [%016lx]: %pS\n", level, regs->nip, 631 (void *)regs->nip); 632 } 633 printk("%s Initiator: %s\n", level, 634 VAL_TO_STRING(initiators, initiator)); 635 636 switch (error_type) { 637 case MC_ERROR_TYPE_UE: 638 printk("%s Error type: %s [%s]\n", level, 639 VAL_TO_STRING(mc_err_types, error_type), 640 VAL_TO_STRING(mc_ue_types, err_sub_type)); 641 break; 642 case MC_ERROR_TYPE_SLB: 643 printk("%s Error type: %s [%s]\n", level, 644 VAL_TO_STRING(mc_err_types, error_type), 645 VAL_TO_STRING(mc_slb_types, err_sub_type)); 646 break; 647 case MC_ERROR_TYPE_ERAT: 648 case MC_ERROR_TYPE_TLB: 649 printk("%s Error type: %s [%s]\n", level, 650 VAL_TO_STRING(mc_err_types, error_type), 651 VAL_TO_STRING(mc_soft_types, err_sub_type)); 652 break; 653 default: 654 printk("%s Error type: %s\n", level, 655 VAL_TO_STRING(mc_err_types, error_type)); 656 break; 657 } 658 659 addr = rtas_mc_get_effective_addr(mce_log); 660 if (addr) 661 printk("%s Effective address: %016llx\n", level, addr); 662 } 663 664 static int mce_handle_error(struct rtas_error_log *errp) 665 { 666 struct pseries_errorlog *pseries_log; 667 struct pseries_mc_errorlog *mce_log; 668 int disposition = rtas_error_disposition(errp); 669 u8 error_type; 670 671 if (!rtas_error_extended(errp)) 672 goto out; 673 674 pseries_log = get_pseries_errorlog(errp, PSERIES_ELOG_SECT_ID_MCE); 675 if (pseries_log == NULL) 676 goto out; 677 678 mce_log = (struct pseries_mc_errorlog *)pseries_log->data; 679 error_type = mce_log->error_type; 680 681 #ifdef CONFIG_PPC_BOOK3S_64 682 if (disposition == RTAS_DISP_NOT_RECOVERED) { 683 switch (error_type) { 684 case MC_ERROR_TYPE_SLB: 685 case MC_ERROR_TYPE_ERAT: 686 /* 687 * Store the old slb content in paca before flushing. 688 * Print this when we go to virtual mode. 689 * There are chances that we may hit MCE again if there 690 * is a parity error on the SLB entry we trying to read 691 * for saving. Hence limit the slb saving to single 692 * level of recursion. 693 */ 694 if (local_paca->in_mce == 1) 695 slb_save_contents(local_paca->mce_faulty_slbs); 696 flush_and_reload_slb(); 697 disposition = RTAS_DISP_FULLY_RECOVERED; 698 rtas_set_disposition_recovered(errp); 699 break; 700 default: 701 break; 702 } 703 } 704 #endif 705 706 out: 707 return disposition; 708 } 709 710 #ifdef CONFIG_MEMORY_FAILURE 711 712 static DEFINE_PER_CPU(int, rtas_ue_count); 713 static DEFINE_PER_CPU(unsigned long, rtas_ue_paddr[MAX_MC_EVT]); 714 715 #define UE_EFFECTIVE_ADDR_PROVIDED 0x40 716 #define UE_LOGICAL_ADDR_PROVIDED 0x20 717 718 719 static void pseries_hwpoison_work_fn(struct work_struct *work) 720 { 721 unsigned long paddr; 722 int index; 723 724 while (__this_cpu_read(rtas_ue_count) > 0) { 725 index = __this_cpu_read(rtas_ue_count) - 1; 726 paddr = __this_cpu_read(rtas_ue_paddr[index]); 727 memory_failure(paddr >> PAGE_SHIFT, 0); 728 __this_cpu_dec(rtas_ue_count); 729 } 730 } 731 732 static DECLARE_WORK(hwpoison_work, pseries_hwpoison_work_fn); 733 734 static void queue_ue_paddr(unsigned long paddr) 735 { 736 int index; 737 738 index = __this_cpu_inc_return(rtas_ue_count) - 1; 739 if (index >= MAX_MC_EVT) { 740 __this_cpu_dec(rtas_ue_count); 741 return; 742 } 743 this_cpu_write(rtas_ue_paddr[index], paddr); 744 schedule_work(&hwpoison_work); 745 } 746 747 static void pseries_do_memory_failure(struct pt_regs *regs, 748 struct pseries_mc_errorlog *mce_log) 749 { 750 unsigned long paddr; 751 752 if (mce_log->sub_err_type & UE_LOGICAL_ADDR_PROVIDED) { 753 paddr = be64_to_cpu(mce_log->logical_address); 754 } else if (mce_log->sub_err_type & UE_EFFECTIVE_ADDR_PROVIDED) { 755 unsigned long pfn; 756 757 pfn = addr_to_pfn(regs, 758 be64_to_cpu(mce_log->effective_address)); 759 if (pfn == ULONG_MAX) 760 return; 761 paddr = pfn << PAGE_SHIFT; 762 } else { 763 return; 764 } 765 queue_ue_paddr(paddr); 766 } 767 768 static void pseries_process_ue(struct pt_regs *regs, 769 struct rtas_error_log *errp) 770 { 771 struct pseries_errorlog *pseries_log; 772 struct pseries_mc_errorlog *mce_log; 773 774 if (!rtas_error_extended(errp)) 775 return; 776 777 pseries_log = get_pseries_errorlog(errp, PSERIES_ELOG_SECT_ID_MCE); 778 if (!pseries_log) 779 return; 780 781 mce_log = (struct pseries_mc_errorlog *)pseries_log->data; 782 783 if (mce_log->error_type == MC_ERROR_TYPE_UE) 784 pseries_do_memory_failure(regs, mce_log); 785 } 786 #else 787 static inline void pseries_process_ue(struct pt_regs *regs, 788 struct rtas_error_log *errp) { } 789 #endif /*CONFIG_MEMORY_FAILURE */ 790 791 /* 792 * Process MCE rtas errlog event. 793 */ 794 static void mce_process_errlog_event(struct irq_work *work) 795 { 796 struct rtas_error_log *err; 797 798 err = fwnmi_get_errlog(); 799 log_error((char *)err, ERR_TYPE_RTAS_LOG, 0); 800 } 801 802 /* 803 * See if we can recover from a machine check exception. 804 * This is only called on power4 (or above) and only via 805 * the Firmware Non-Maskable Interrupts (fwnmi) handler 806 * which provides the error analysis for us. 807 * 808 * Return 1 if corrected (or delivered a signal). 809 * Return 0 if there is nothing we can do. 810 */ 811 static int recover_mce(struct pt_regs *regs, struct rtas_error_log *err) 812 { 813 int recovered = 0; 814 int disposition = rtas_error_disposition(err); 815 816 pseries_print_mce_info(regs, err); 817 818 if (!(regs->msr & MSR_RI)) { 819 /* If MSR_RI isn't set, we cannot recover */ 820 pr_err("Machine check interrupt unrecoverable: MSR(RI=0)\n"); 821 recovered = 0; 822 823 } else if (disposition == RTAS_DISP_FULLY_RECOVERED) { 824 /* Platform corrected itself */ 825 recovered = 1; 826 827 } else if (disposition == RTAS_DISP_LIMITED_RECOVERY) { 828 /* Platform corrected itself but could be degraded */ 829 printk(KERN_ERR "MCE: limited recovery, system may " 830 "be degraded\n"); 831 recovered = 1; 832 833 } else if (user_mode(regs) && !is_global_init(current) && 834 rtas_error_severity(err) == RTAS_SEVERITY_ERROR_SYNC) { 835 836 /* 837 * If we received a synchronous error when in userspace 838 * kill the task. Firmware may report details of the fail 839 * asynchronously, so we can't rely on the target and type 840 * fields being valid here. 841 */ 842 printk(KERN_ERR "MCE: uncorrectable error, killing task " 843 "%s:%d\n", current->comm, current->pid); 844 845 _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip); 846 recovered = 1; 847 } 848 849 pseries_process_ue(regs, err); 850 851 /* Queue irq work to log this rtas event later. */ 852 irq_work_queue(&mce_errlog_process_work); 853 854 return recovered; 855 } 856 857 /* 858 * Handle a machine check. 859 * 860 * Note that on Power 4 and beyond Firmware Non-Maskable Interrupts (fwnmi) 861 * should be present. If so the handler which called us tells us if the 862 * error was recovered (never true if RI=0). 863 * 864 * On hardware prior to Power 4 these exceptions were asynchronous which 865 * means we can't tell exactly where it occurred and so we can't recover. 866 */ 867 int pSeries_machine_check_exception(struct pt_regs *regs) 868 { 869 struct rtas_error_log *errp; 870 871 if (fwnmi_active) { 872 fwnmi_release_errinfo(); 873 errp = fwnmi_get_errlog(); 874 if (errp && recover_mce(regs, errp)) 875 return 1; 876 } 877 878 return 0; 879 } 880 881 long pseries_machine_check_realmode(struct pt_regs *regs) 882 { 883 struct rtas_error_log *errp; 884 int disposition; 885 886 if (fwnmi_active) { 887 errp = fwnmi_get_errinfo(regs); 888 /* 889 * Call to fwnmi_release_errinfo() in real mode causes kernel 890 * to panic. Hence we will call it as soon as we go into 891 * virtual mode. 892 */ 893 disposition = mce_handle_error(errp); 894 if (disposition == RTAS_DISP_FULLY_RECOVERED) 895 return 1; 896 } 897 898 return 0; 899 } 900