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