1/* 2 * File: mca_drv.c 3 * Purpose: Generic MCA handling layer 4 * 5 * Copyright (C) 2004 FUJITSU LIMITED 6 * Copyright (C) 2004 Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com> 7 * Copyright (C) 2005 Silicon Graphics, Inc 8 * Copyright (C) 2005 Keith Owens <kaos@sgi.com> 9 * Copyright (C) 2006 Russ Anderson <rja@sgi.com> 10 */ 11#include <linux/types.h> 12#include <linux/init.h> 13#include <linux/sched.h> 14#include <linux/interrupt.h> 15#include <linux/irq.h> 16#include <linux/kallsyms.h> 17#include <linux/bootmem.h> 18#include <linux/acpi.h> 19#include <linux/timer.h> 20#include <linux/module.h> 21#include <linux/kernel.h> 22#include <linux/smp.h> 23#include <linux/workqueue.h> 24#include <linux/mm.h> 25#include <linux/slab.h> 26 27#include <asm/delay.h> 28#include <asm/machvec.h> 29#include <asm/page.h> 30#include <asm/ptrace.h> 31#include <asm/sal.h> 32#include <asm/mca.h> 33 34#include <asm/irq.h> 35#include <asm/hw_irq.h> 36 37#include "mca_drv.h" 38 39/* max size of SAL error record (default) */ 40static int sal_rec_max = 10000; 41 42/* from mca_drv_asm.S */ 43extern void *mca_handler_bhhook(void); 44 45static DEFINE_SPINLOCK(mca_bh_lock); 46 47typedef enum { 48 MCA_IS_LOCAL = 0, 49 MCA_IS_GLOBAL = 1 50} mca_type_t; 51 52#define MAX_PAGE_ISOLATE 1024 53 54static struct page *page_isolate[MAX_PAGE_ISOLATE]; 55static int num_page_isolate = 0; 56 57typedef enum { 58 ISOLATE_NG, 59 ISOLATE_OK, 60 ISOLATE_NONE 61} isolate_status_t; 62 63typedef enum { 64 MCA_NOT_RECOVERED = 0, 65 MCA_RECOVERED = 1 66} recovery_status_t; 67 68/* 69 * This pool keeps pointers to the section part of SAL error record 70 */ 71static struct { 72 slidx_list_t *buffer; /* section pointer list pool */ 73 int cur_idx; /* Current index of section pointer list pool */ 74 int max_idx; /* Maximum index of section pointer list pool */ 75} slidx_pool; 76 77static int 78fatal_mca(const char *fmt, ...) 79{ 80 va_list args; 81 char buf[256]; 82 83 va_start(args, fmt); 84 vsnprintf(buf, sizeof(buf), fmt, args); 85 va_end(args); 86 ia64_mca_printk(KERN_ALERT "MCA: %s\n", buf); 87 88 return MCA_NOT_RECOVERED; 89} 90 91static int 92mca_recovered(const char *fmt, ...) 93{ 94 va_list args; 95 char buf[256]; 96 97 va_start(args, fmt); 98 vsnprintf(buf, sizeof(buf), fmt, args); 99 va_end(args); 100 ia64_mca_printk(KERN_INFO "MCA: %s\n", buf); 101 102 return MCA_RECOVERED; 103} 104 105/** 106 * mca_page_isolate - isolate a poisoned page in order not to use it later 107 * @paddr: poisoned memory location 108 * 109 * Return value: 110 * one of isolate_status_t, ISOLATE_OK/NG/NONE. 111 */ 112 113static isolate_status_t 114mca_page_isolate(unsigned long paddr) 115{ 116 int i; 117 struct page *p; 118 119 /* whether physical address is valid or not */ 120 if (!ia64_phys_addr_valid(paddr)) 121 return ISOLATE_NONE; 122 123 if (!pfn_valid(paddr >> PAGE_SHIFT)) 124 return ISOLATE_NONE; 125 126 /* convert physical address to physical page number */ 127 p = pfn_to_page(paddr>>PAGE_SHIFT); 128 129 /* check whether a page number have been already registered or not */ 130 for (i = 0; i < num_page_isolate; i++) 131 if (page_isolate[i] == p) 132 return ISOLATE_OK; /* already listed */ 133 134 /* limitation check */ 135 if (num_page_isolate == MAX_PAGE_ISOLATE) 136 return ISOLATE_NG; 137 138 /* kick pages having attribute 'SLAB' or 'Reserved' */ 139 if (PageSlab(p) || PageReserved(p)) 140 return ISOLATE_NG; 141 142 /* add attribute 'Reserved' and register the page */ 143 get_page(p); 144 SetPageReserved(p); 145 page_isolate[num_page_isolate++] = p; 146 147 return ISOLATE_OK; 148} 149 150/** 151 * mca_hanlder_bh - Kill the process which occurred memory read error 152 * @paddr: poisoned address received from MCA Handler 153 */ 154 155void 156mca_handler_bh(unsigned long paddr, void *iip, unsigned long ipsr) 157{ 158 ia64_mlogbuf_dump(); 159 printk(KERN_ERR "OS_MCA: process [cpu %d, pid: %d, uid: %d, " 160 "iip: %p, psr: 0x%lx,paddr: 0x%lx](%s) encounters MCA.\n", 161 raw_smp_processor_id(), current->pid, 162 from_kuid(&init_user_ns, current_uid()), 163 iip, ipsr, paddr, current->comm); 164 165 spin_lock(&mca_bh_lock); 166 switch (mca_page_isolate(paddr)) { 167 case ISOLATE_OK: 168 printk(KERN_DEBUG "Page isolation: ( %lx ) success.\n", paddr); 169 break; 170 case ISOLATE_NG: 171 printk(KERN_CRIT "Page isolation: ( %lx ) failure.\n", paddr); 172 break; 173 default: 174 break; 175 } 176 spin_unlock(&mca_bh_lock); 177 178 /* This process is about to be killed itself */ 179 do_exit(SIGKILL); 180} 181 182/** 183 * mca_make_peidx - Make index of processor error section 184 * @slpi: pointer to record of processor error section 185 * @peidx: pointer to index of processor error section 186 */ 187 188static void 189mca_make_peidx(sal_log_processor_info_t *slpi, peidx_table_t *peidx) 190{ 191 /* 192 * calculate the start address of 193 * "struct cpuid_info" and "sal_processor_static_info_t". 194 */ 195 u64 total_check_num = slpi->valid.num_cache_check 196 + slpi->valid.num_tlb_check 197 + slpi->valid.num_bus_check 198 + slpi->valid.num_reg_file_check 199 + slpi->valid.num_ms_check; 200 u64 head_size = sizeof(sal_log_mod_error_info_t) * total_check_num 201 + sizeof(sal_log_processor_info_t); 202 u64 mid_size = slpi->valid.cpuid_info * sizeof(struct sal_cpuid_info); 203 204 peidx_head(peidx) = slpi; 205 peidx_mid(peidx) = (struct sal_cpuid_info *) 206 (slpi->valid.cpuid_info ? ((char*)slpi + head_size) : NULL); 207 peidx_bottom(peidx) = (sal_processor_static_info_t *) 208 (slpi->valid.psi_static_struct ? 209 ((char*)slpi + head_size + mid_size) : NULL); 210} 211 212/** 213 * mca_make_slidx - Make index of SAL error record 214 * @buffer: pointer to SAL error record 215 * @slidx: pointer to index of SAL error record 216 * 217 * Return value: 218 * 1 if record has platform error / 0 if not 219 */ 220#define LOG_INDEX_ADD_SECT_PTR(sect, ptr) \ 221 {slidx_list_t *hl = &slidx_pool.buffer[slidx_pool.cur_idx]; \ 222 hl->hdr = ptr; \ 223 list_add(&hl->list, &(sect)); \ 224 slidx_pool.cur_idx = (slidx_pool.cur_idx + 1)%slidx_pool.max_idx; } 225 226static int 227mca_make_slidx(void *buffer, slidx_table_t *slidx) 228{ 229 int platform_err = 0; 230 int record_len = ((sal_log_record_header_t*)buffer)->len; 231 u32 ercd_pos; 232 int sects; 233 sal_log_section_hdr_t *sp; 234 235 /* 236 * Initialize index referring current record 237 */ 238 INIT_LIST_HEAD(&(slidx->proc_err)); 239 INIT_LIST_HEAD(&(slidx->mem_dev_err)); 240 INIT_LIST_HEAD(&(slidx->sel_dev_err)); 241 INIT_LIST_HEAD(&(slidx->pci_bus_err)); 242 INIT_LIST_HEAD(&(slidx->smbios_dev_err)); 243 INIT_LIST_HEAD(&(slidx->pci_comp_err)); 244 INIT_LIST_HEAD(&(slidx->plat_specific_err)); 245 INIT_LIST_HEAD(&(slidx->host_ctlr_err)); 246 INIT_LIST_HEAD(&(slidx->plat_bus_err)); 247 INIT_LIST_HEAD(&(slidx->unsupported)); 248 249 /* 250 * Extract a Record Header 251 */ 252 slidx->header = buffer; 253 254 /* 255 * Extract each section records 256 * (arranged from "int ia64_log_platform_info_print()") 257 */ 258 for (ercd_pos = sizeof(sal_log_record_header_t), sects = 0; 259 ercd_pos < record_len; ercd_pos += sp->len, sects++) { 260 sp = (sal_log_section_hdr_t *)((char*)buffer + ercd_pos); 261 if (!efi_guidcmp(sp->guid, SAL_PROC_DEV_ERR_SECT_GUID)) { 262 LOG_INDEX_ADD_SECT_PTR(slidx->proc_err, sp); 263 } else if (!efi_guidcmp(sp->guid, 264 SAL_PLAT_MEM_DEV_ERR_SECT_GUID)) { 265 platform_err = 1; 266 LOG_INDEX_ADD_SECT_PTR(slidx->mem_dev_err, sp); 267 } else if (!efi_guidcmp(sp->guid, 268 SAL_PLAT_SEL_DEV_ERR_SECT_GUID)) { 269 platform_err = 1; 270 LOG_INDEX_ADD_SECT_PTR(slidx->sel_dev_err, sp); 271 } else if (!efi_guidcmp(sp->guid, 272 SAL_PLAT_PCI_BUS_ERR_SECT_GUID)) { 273 platform_err = 1; 274 LOG_INDEX_ADD_SECT_PTR(slidx->pci_bus_err, sp); 275 } else if (!efi_guidcmp(sp->guid, 276 SAL_PLAT_SMBIOS_DEV_ERR_SECT_GUID)) { 277 platform_err = 1; 278 LOG_INDEX_ADD_SECT_PTR(slidx->smbios_dev_err, sp); 279 } else if (!efi_guidcmp(sp->guid, 280 SAL_PLAT_PCI_COMP_ERR_SECT_GUID)) { 281 platform_err = 1; 282 LOG_INDEX_ADD_SECT_PTR(slidx->pci_comp_err, sp); 283 } else if (!efi_guidcmp(sp->guid, 284 SAL_PLAT_SPECIFIC_ERR_SECT_GUID)) { 285 platform_err = 1; 286 LOG_INDEX_ADD_SECT_PTR(slidx->plat_specific_err, sp); 287 } else if (!efi_guidcmp(sp->guid, 288 SAL_PLAT_HOST_CTLR_ERR_SECT_GUID)) { 289 platform_err = 1; 290 LOG_INDEX_ADD_SECT_PTR(slidx->host_ctlr_err, sp); 291 } else if (!efi_guidcmp(sp->guid, 292 SAL_PLAT_BUS_ERR_SECT_GUID)) { 293 platform_err = 1; 294 LOG_INDEX_ADD_SECT_PTR(slidx->plat_bus_err, sp); 295 } else { 296 LOG_INDEX_ADD_SECT_PTR(slidx->unsupported, sp); 297 } 298 } 299 slidx->n_sections = sects; 300 301 return platform_err; 302} 303 304/** 305 * init_record_index_pools - Initialize pool of lists for SAL record index 306 * 307 * Return value: 308 * 0 on Success / -ENOMEM on Failure 309 */ 310static int 311init_record_index_pools(void) 312{ 313 int i; 314 int rec_max_size; /* Maximum size of SAL error records */ 315 int sect_min_size; /* Minimum size of SAL error sections */ 316 /* minimum size table of each section */ 317 static int sal_log_sect_min_sizes[] = { 318 sizeof(sal_log_processor_info_t) 319 + sizeof(sal_processor_static_info_t), 320 sizeof(sal_log_mem_dev_err_info_t), 321 sizeof(sal_log_sel_dev_err_info_t), 322 sizeof(sal_log_pci_bus_err_info_t), 323 sizeof(sal_log_smbios_dev_err_info_t), 324 sizeof(sal_log_pci_comp_err_info_t), 325 sizeof(sal_log_plat_specific_err_info_t), 326 sizeof(sal_log_host_ctlr_err_info_t), 327 sizeof(sal_log_plat_bus_err_info_t), 328 }; 329 330 /* 331 * MCA handler cannot allocate new memory on flight, 332 * so we preallocate enough memory to handle a SAL record. 333 * 334 * Initialize a handling set of slidx_pool: 335 * 1. Pick up the max size of SAL error records 336 * 2. Pick up the min size of SAL error sections 337 * 3. Allocate the pool as enough to 2 SAL records 338 * (now we can estimate the maxinum of section in a record.) 339 */ 340 341 /* - 1 - */ 342 rec_max_size = sal_rec_max; 343 344 /* - 2 - */ 345 sect_min_size = sal_log_sect_min_sizes[0]; 346 for (i = 1; i < sizeof sal_log_sect_min_sizes/sizeof(size_t); i++) 347 if (sect_min_size > sal_log_sect_min_sizes[i]) 348 sect_min_size = sal_log_sect_min_sizes[i]; 349 350 /* - 3 - */ 351 slidx_pool.max_idx = (rec_max_size/sect_min_size) * 2 + 1; 352 slidx_pool.buffer = 353 kmalloc(slidx_pool.max_idx * sizeof(slidx_list_t), GFP_KERNEL); 354 355 return slidx_pool.buffer ? 0 : -ENOMEM; 356} 357 358 359/***************************************************************************** 360 * Recovery functions * 361 *****************************************************************************/ 362 363/** 364 * is_mca_global - Check whether this MCA is global or not 365 * @peidx: pointer of index of processor error section 366 * @pbci: pointer to pal_bus_check_info_t 367 * @sos: pointer to hand off struct between SAL and OS 368 * 369 * Return value: 370 * MCA_IS_LOCAL / MCA_IS_GLOBAL 371 */ 372 373static mca_type_t 374is_mca_global(peidx_table_t *peidx, pal_bus_check_info_t *pbci, 375 struct ia64_sal_os_state *sos) 376{ 377 pal_processor_state_info_t *psp = 378 (pal_processor_state_info_t*)peidx_psp(peidx); 379 380 /* 381 * PAL can request a rendezvous, if the MCA has a global scope. 382 * If "rz_always" flag is set, SAL requests MCA rendezvous 383 * in spite of global MCA. 384 * Therefore it is local MCA when rendezvous has not been requested. 385 * Failed to rendezvous, the system must be down. 386 */ 387 switch (sos->rv_rc) { 388 case -1: /* SAL rendezvous unsuccessful */ 389 return MCA_IS_GLOBAL; 390 case 0: /* SAL rendezvous not required */ 391 return MCA_IS_LOCAL; 392 case 1: /* SAL rendezvous successful int */ 393 case 2: /* SAL rendezvous successful int with init */ 394 default: 395 break; 396 } 397 398 /* 399 * If One or more Cache/TLB/Reg_File/Uarch_Check is here, 400 * it would be a local MCA. (i.e. processor internal error) 401 */ 402 if (psp->tc || psp->cc || psp->rc || psp->uc) 403 return MCA_IS_LOCAL; 404 405 /* 406 * Bus_Check structure with Bus_Check.ib (internal bus error) flag set 407 * would be a global MCA. (e.g. a system bus address parity error) 408 */ 409 if (!pbci || pbci->ib) 410 return MCA_IS_GLOBAL; 411 412 /* 413 * Bus_Check structure with Bus_Check.eb (external bus error) flag set 414 * could be either a local MCA or a global MCA. 415 * 416 * Referring Bus_Check.bsi: 417 * 0: Unknown/unclassified 418 * 1: BERR# 419 * 2: BINIT# 420 * 3: Hard Fail 421 * (FIXME: Are these SGI specific or generic bsi values?) 422 */ 423 if (pbci->eb) 424 switch (pbci->bsi) { 425 case 0: 426 /* e.g. a load from poisoned memory */ 427 return MCA_IS_LOCAL; 428 case 1: 429 case 2: 430 case 3: 431 return MCA_IS_GLOBAL; 432 } 433 434 return MCA_IS_GLOBAL; 435} 436 437/** 438 * get_target_identifier - Get the valid Cache or Bus check target identifier. 439 * @peidx: pointer of index of processor error section 440 * 441 * Return value: 442 * target address on Success / 0 on Failure 443 */ 444static u64 445get_target_identifier(peidx_table_t *peidx) 446{ 447 u64 target_address = 0; 448 sal_log_mod_error_info_t *smei; 449 pal_cache_check_info_t *pcci; 450 int i, level = 9; 451 452 /* 453 * Look through the cache checks for a valid target identifier 454 * If more than one valid target identifier, return the one 455 * with the lowest cache level. 456 */ 457 for (i = 0; i < peidx_cache_check_num(peidx); i++) { 458 smei = (sal_log_mod_error_info_t *)peidx_cache_check(peidx, i); 459 if (smei->valid.target_identifier && smei->target_identifier) { 460 pcci = (pal_cache_check_info_t *)&(smei->check_info); 461 if (!target_address || (pcci->level < level)) { 462 target_address = smei->target_identifier; 463 level = pcci->level; 464 continue; 465 } 466 } 467 } 468 if (target_address) 469 return target_address; 470 471 /* 472 * Look at the bus check for a valid target identifier 473 */ 474 smei = peidx_bus_check(peidx, 0); 475 if (smei && smei->valid.target_identifier) 476 return smei->target_identifier; 477 478 return 0; 479} 480 481/** 482 * recover_from_read_error - Try to recover the errors which type are "read"s. 483 * @slidx: pointer of index of SAL error record 484 * @peidx: pointer of index of processor error section 485 * @pbci: pointer of pal_bus_check_info 486 * @sos: pointer to hand off struct between SAL and OS 487 * 488 * Return value: 489 * 1 on Success / 0 on Failure 490 */ 491 492static int 493recover_from_read_error(slidx_table_t *slidx, 494 peidx_table_t *peidx, pal_bus_check_info_t *pbci, 495 struct ia64_sal_os_state *sos) 496{ 497 u64 target_identifier; 498 pal_min_state_area_t *pmsa; 499 struct ia64_psr *psr1, *psr2; 500 ia64_fptr_t *mca_hdlr_bh = (ia64_fptr_t*)mca_handler_bhhook; 501 502 /* Is target address valid? */ 503 target_identifier = get_target_identifier(peidx); 504 if (!target_identifier) 505 return fatal_mca("target address not valid"); 506 507 /* 508 * cpu read or memory-mapped io read 509 * 510 * offending process affected process OS MCA do 511 * kernel mode kernel mode down system 512 * kernel mode user mode kill the process 513 * user mode kernel mode down system (*) 514 * user mode user mode kill the process 515 * 516 * (*) You could terminate offending user-mode process 517 * if (pbci->pv && pbci->pl != 0) *and* if you sure 518 * the process not have any locks of kernel. 519 */ 520 521 /* Is minstate valid? */ 522 if (!peidx_bottom(peidx) || !(peidx_bottom(peidx)->valid.minstate)) 523 return fatal_mca("minstate not valid"); 524 psr1 =(struct ia64_psr *)&(peidx_minstate_area(peidx)->pmsa_ipsr); 525 psr2 =(struct ia64_psr *)&(peidx_minstate_area(peidx)->pmsa_xpsr); 526 527 /* 528 * Check the privilege level of interrupted context. 529 * If it is user-mode, then terminate affected process. 530 */ 531 532 pmsa = sos->pal_min_state; 533 if (psr1->cpl != 0 || 534 ((psr2->cpl != 0) && mca_recover_range(pmsa->pmsa_iip))) { 535 /* 536 * setup for resume to bottom half of MCA, 537 * "mca_handler_bhhook" 538 */ 539 /* pass to bhhook as argument (gr8, ...) */ 540 pmsa->pmsa_gr[8-1] = target_identifier; 541 pmsa->pmsa_gr[9-1] = pmsa->pmsa_iip; 542 pmsa->pmsa_gr[10-1] = pmsa->pmsa_ipsr; 543 /* set interrupted return address (but no use) */ 544 pmsa->pmsa_br0 = pmsa->pmsa_iip; 545 /* change resume address to bottom half */ 546 pmsa->pmsa_iip = mca_hdlr_bh->fp; 547 pmsa->pmsa_gr[1-1] = mca_hdlr_bh->gp; 548 /* set cpl with kernel mode */ 549 psr2 = (struct ia64_psr *)&pmsa->pmsa_ipsr; 550 psr2->cpl = 0; 551 psr2->ri = 0; 552 psr2->bn = 1; 553 psr2->i = 0; 554 555 return mca_recovered("user memory corruption. " 556 "kill affected process - recovered."); 557 } 558 559 return fatal_mca("kernel context not recovered, iip 0x%lx\n", 560 pmsa->pmsa_iip); 561} 562 563/** 564 * recover_from_platform_error - Recover from platform error. 565 * @slidx: pointer of index of SAL error record 566 * @peidx: pointer of index of processor error section 567 * @pbci: pointer of pal_bus_check_info 568 * @sos: pointer to hand off struct between SAL and OS 569 * 570 * Return value: 571 * 1 on Success / 0 on Failure 572 */ 573 574static int 575recover_from_platform_error(slidx_table_t *slidx, peidx_table_t *peidx, 576 pal_bus_check_info_t *pbci, 577 struct ia64_sal_os_state *sos) 578{ 579 int status = 0; 580 pal_processor_state_info_t *psp = 581 (pal_processor_state_info_t*)peidx_psp(peidx); 582 583 if (psp->bc && pbci->eb && pbci->bsi == 0) { 584 switch(pbci->type) { 585 case 1: /* partial read */ 586 case 3: /* full line(cpu) read */ 587 case 9: /* I/O space read */ 588 status = recover_from_read_error(slidx, peidx, pbci, 589 sos); 590 break; 591 case 0: /* unknown */ 592 case 2: /* partial write */ 593 case 4: /* full line write */ 594 case 5: /* implicit or explicit write-back operation */ 595 case 6: /* snoop probe */ 596 case 7: /* incoming or outgoing ptc.g */ 597 case 8: /* write coalescing transactions */ 598 case 10: /* I/O space write */ 599 case 11: /* inter-processor interrupt message(IPI) */ 600 case 12: /* interrupt acknowledge or 601 external task priority cycle */ 602 default: 603 break; 604 } 605 } else if (psp->cc && !psp->bc) { /* Cache error */ 606 status = recover_from_read_error(slidx, peidx, pbci, sos); 607 } 608 609 return status; 610} 611 612/* 613 * recover_from_tlb_check 614 * @peidx: pointer of index of processor error section 615 * 616 * Return value: 617 * 1 on Success / 0 on Failure 618 */ 619static int 620recover_from_tlb_check(peidx_table_t *peidx) 621{ 622 sal_log_mod_error_info_t *smei; 623 pal_tlb_check_info_t *ptci; 624 625 smei = (sal_log_mod_error_info_t *)peidx_tlb_check(peidx, 0); 626 ptci = (pal_tlb_check_info_t *)&(smei->check_info); 627 628 /* 629 * Look for signature of a duplicate TLB DTC entry, which is 630 * a SW bug and always fatal. 631 */ 632 if (ptci->op == PAL_TLB_CHECK_OP_PURGE 633 && !(ptci->itr || ptci->dtc || ptci->itc)) 634 return fatal_mca("Duplicate TLB entry"); 635 636 return mca_recovered("TLB check recovered"); 637} 638 639/** 640 * recover_from_processor_error 641 * @platform: whether there are some platform error section or not 642 * @slidx: pointer of index of SAL error record 643 * @peidx: pointer of index of processor error section 644 * @pbci: pointer of pal_bus_check_info 645 * @sos: pointer to hand off struct between SAL and OS 646 * 647 * Return value: 648 * 1 on Success / 0 on Failure 649 */ 650 651static int 652recover_from_processor_error(int platform, slidx_table_t *slidx, 653 peidx_table_t *peidx, pal_bus_check_info_t *pbci, 654 struct ia64_sal_os_state *sos) 655{ 656 pal_processor_state_info_t *psp = 657 (pal_processor_state_info_t*)peidx_psp(peidx); 658 659 /* 660 * Processor recovery status must key off of the PAL recovery 661 * status in the Processor State Parameter. 662 */ 663 664 /* 665 * The machine check is corrected. 666 */ 667 if (psp->cm == 1) 668 return mca_recovered("machine check is already corrected."); 669 670 /* 671 * The error was not contained. Software must be reset. 672 */ 673 if (psp->us || psp->ci == 0) 674 return fatal_mca("error not contained"); 675 676 /* 677 * Look for recoverable TLB check 678 */ 679 if (psp->tc && !(psp->cc || psp->bc || psp->rc || psp->uc)) 680 return recover_from_tlb_check(peidx); 681 682 /* 683 * The cache check and bus check bits have four possible states 684 * cc bc 685 * 1 1 Memory error, attempt recovery 686 * 1 0 Cache error, attempt recovery 687 * 0 1 I/O error, attempt recovery 688 * 0 0 Other error type, not recovered 689 */ 690 if (psp->cc == 0 && (psp->bc == 0 || pbci == NULL)) 691 return fatal_mca("No cache or bus check"); 692 693 /* 694 * Cannot handle more than one bus check. 695 */ 696 if (peidx_bus_check_num(peidx) > 1) 697 return fatal_mca("Too many bus checks"); 698 699 if (pbci->ib) 700 return fatal_mca("Internal Bus error"); 701 if (pbci->eb && pbci->bsi > 0) 702 return fatal_mca("External bus check fatal status"); 703 704 /* 705 * This is a local MCA and estimated as a recoverable error. 706 */ 707 if (platform) 708 return recover_from_platform_error(slidx, peidx, pbci, sos); 709 710 /* 711 * On account of strange SAL error record, we cannot recover. 712 */ 713 return fatal_mca("Strange SAL record"); 714} 715 716/** 717 * mca_try_to_recover - Try to recover from MCA 718 * @rec: pointer to a SAL error record 719 * @sos: pointer to hand off struct between SAL and OS 720 * 721 * Return value: 722 * 1 on Success / 0 on Failure 723 */ 724 725static int 726mca_try_to_recover(void *rec, struct ia64_sal_os_state *sos) 727{ 728 int platform_err; 729 int n_proc_err; 730 slidx_table_t slidx; 731 peidx_table_t peidx; 732 pal_bus_check_info_t pbci; 733 734 /* Make index of SAL error record */ 735 platform_err = mca_make_slidx(rec, &slidx); 736 737 /* Count processor error sections */ 738 n_proc_err = slidx_count(&slidx, proc_err); 739 740 /* Now, OS can recover when there is one processor error section */ 741 if (n_proc_err > 1) 742 return fatal_mca("Too Many Errors"); 743 else if (n_proc_err == 0) 744 /* Weird SAL record ... We can't do anything */ 745 return fatal_mca("Weird SAL record"); 746 747 /* Make index of processor error section */ 748 mca_make_peidx((sal_log_processor_info_t*) 749 slidx_first_entry(&slidx.proc_err)->hdr, &peidx); 750 751 /* Extract Processor BUS_CHECK[0] */ 752 *((u64*)&pbci) = peidx_check_info(&peidx, bus_check, 0); 753 754 /* Check whether MCA is global or not */ 755 if (is_mca_global(&peidx, &pbci, sos)) 756 return fatal_mca("global MCA"); 757 758 /* Try to recover a processor error */ 759 return recover_from_processor_error(platform_err, &slidx, &peidx, 760 &pbci, sos); 761} 762 763/* 764 * ============================================================================= 765 */ 766 767int __init mca_external_handler_init(void) 768{ 769 if (init_record_index_pools()) 770 return -ENOMEM; 771 772 /* register external mca handlers */ 773 if (ia64_reg_MCA_extension(mca_try_to_recover)) { 774 printk(KERN_ERR "ia64_reg_MCA_extension failed.\n"); 775 kfree(slidx_pool.buffer); 776 return -EFAULT; 777 } 778 return 0; 779} 780 781void __exit mca_external_handler_exit(void) 782{ 783 /* unregister external mca handlers */ 784 ia64_unreg_MCA_extension(); 785 kfree(slidx_pool.buffer); 786} 787 788module_init(mca_external_handler_init); 789module_exit(mca_external_handler_exit); 790 791module_param(sal_rec_max, int, 0644); 792MODULE_PARM_DESC(sal_rec_max, "Max size of SAL error record"); 793 794MODULE_DESCRIPTION("ia64 platform dependent mca handler driver"); 795MODULE_LICENSE("GPL"); 796