1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * File: mca_drv.c 4 * Purpose: Generic MCA handling layer 5 * 6 * Copyright (C) 2004 FUJITSU LIMITED 7 * Copyright (C) 2004 Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com> 8 * Copyright (C) 2005 Silicon Graphics, Inc 9 * Copyright (C) 2005 Keith Owens <kaos@sgi.com> 10 * Copyright (C) 2006 Russ Anderson <rja@sgi.com> 11 */ 12 #include <linux/types.h> 13 #include <linux/init.h> 14 #include <linux/sched.h> 15 #include <linux/interrupt.h> 16 #include <linux/irq.h> 17 #include <linux/kallsyms.h> 18 #include <linux/memblock.h> 19 #include <linux/acpi.h> 20 #include <linux/timer.h> 21 #include <linux/module.h> 22 #include <linux/kernel.h> 23 #include <linux/smp.h> 24 #include <linux/workqueue.h> 25 #include <linux/mm.h> 26 #include <linux/slab.h> 27 28 #include <asm/delay.h> 29 #include <asm/machvec.h> 30 #include <asm/page.h> 31 #include <asm/ptrace.h> 32 #include <asm/sal.h> 33 #include <asm/mca.h> 34 35 #include <asm/irq.h> 36 #include <asm/hw_irq.h> 37 38 #include "mca_drv.h" 39 40 /* max size of SAL error record (default) */ 41 static int sal_rec_max = 10000; 42 43 /* from mca_drv_asm.S */ 44 extern void *mca_handler_bhhook(void); 45 46 static DEFINE_SPINLOCK(mca_bh_lock); 47 48 typedef enum { 49 MCA_IS_LOCAL = 0, 50 MCA_IS_GLOBAL = 1 51 } mca_type_t; 52 53 #define MAX_PAGE_ISOLATE 1024 54 55 static struct page *page_isolate[MAX_PAGE_ISOLATE]; 56 static int num_page_isolate = 0; 57 58 typedef enum { 59 ISOLATE_NG, 60 ISOLATE_OK, 61 ISOLATE_NONE 62 } isolate_status_t; 63 64 typedef enum { 65 MCA_NOT_RECOVERED = 0, 66 MCA_RECOVERED = 1 67 } recovery_status_t; 68 69 /* 70 * This pool keeps pointers to the section part of SAL error record 71 */ 72 static struct { 73 slidx_list_t *buffer; /* section pointer list pool */ 74 int cur_idx; /* Current index of section pointer list pool */ 75 int max_idx; /* Maximum index of section pointer list pool */ 76 } slidx_pool; 77 78 static int 79 fatal_mca(const char *fmt, ...) 80 { 81 va_list args; 82 char buf[256]; 83 84 va_start(args, fmt); 85 vsnprintf(buf, sizeof(buf), fmt, args); 86 va_end(args); 87 ia64_mca_printk(KERN_ALERT "MCA: %s\n", buf); 88 89 return MCA_NOT_RECOVERED; 90 } 91 92 static int 93 mca_recovered(const char *fmt, ...) 94 { 95 va_list args; 96 char buf[256]; 97 98 va_start(args, fmt); 99 vsnprintf(buf, sizeof(buf), fmt, args); 100 va_end(args); 101 ia64_mca_printk(KERN_INFO "MCA: %s\n", buf); 102 103 return MCA_RECOVERED; 104 } 105 106 /** 107 * mca_page_isolate - isolate a poisoned page in order not to use it later 108 * @paddr: poisoned memory location 109 * 110 * Return value: 111 * one of isolate_status_t, ISOLATE_OK/NG/NONE. 112 */ 113 114 static isolate_status_t 115 mca_page_isolate(unsigned long paddr) 116 { 117 int i; 118 struct page *p; 119 120 /* whether physical address is valid or not */ 121 if (!ia64_phys_addr_valid(paddr)) 122 return ISOLATE_NONE; 123 124 if (!pfn_valid(paddr >> PAGE_SHIFT)) 125 return ISOLATE_NONE; 126 127 /* convert physical address to physical page number */ 128 p = pfn_to_page(paddr>>PAGE_SHIFT); 129 130 /* check whether a page number have been already registered or not */ 131 for (i = 0; i < num_page_isolate; i++) 132 if (page_isolate[i] == p) 133 return ISOLATE_OK; /* already listed */ 134 135 /* limitation check */ 136 if (num_page_isolate == MAX_PAGE_ISOLATE) 137 return ISOLATE_NG; 138 139 /* kick pages having attribute 'SLAB' or 'Reserved' */ 140 if (PageSlab(p) || PageReserved(p)) 141 return ISOLATE_NG; 142 143 /* add attribute 'Reserved' and register the page */ 144 get_page(p); 145 SetPageReserved(p); 146 page_isolate[num_page_isolate++] = p; 147 148 return ISOLATE_OK; 149 } 150 151 /** 152 * mca_hanlder_bh - Kill the process which occurred memory read error 153 * @paddr: poisoned address received from MCA Handler 154 */ 155 156 void 157 mca_handler_bh(unsigned long paddr, void *iip, unsigned long ipsr) 158 { 159 ia64_mlogbuf_dump(); 160 printk(KERN_ERR "OS_MCA: process [cpu %d, pid: %d, uid: %d, " 161 "iip: %p, psr: 0x%lx,paddr: 0x%lx](%s) encounters MCA.\n", 162 raw_smp_processor_id(), current->pid, 163 from_kuid(&init_user_ns, current_uid()), 164 iip, ipsr, paddr, current->comm); 165 166 spin_lock(&mca_bh_lock); 167 switch (mca_page_isolate(paddr)) { 168 case ISOLATE_OK: 169 printk(KERN_DEBUG "Page isolation: ( %lx ) success.\n", paddr); 170 break; 171 case ISOLATE_NG: 172 printk(KERN_CRIT "Page isolation: ( %lx ) failure.\n", paddr); 173 break; 174 default: 175 break; 176 } 177 spin_unlock(&mca_bh_lock); 178 179 /* This process is about to be killed itself */ 180 do_exit(SIGKILL); 181 } 182 183 /** 184 * mca_make_peidx - Make index of processor error section 185 * @slpi: pointer to record of processor error section 186 * @peidx: pointer to index of processor error section 187 */ 188 189 static void 190 mca_make_peidx(sal_log_processor_info_t *slpi, peidx_table_t *peidx) 191 { 192 /* 193 * calculate the start address of 194 * "struct cpuid_info" and "sal_processor_static_info_t". 195 */ 196 u64 total_check_num = slpi->valid.num_cache_check 197 + slpi->valid.num_tlb_check 198 + slpi->valid.num_bus_check 199 + slpi->valid.num_reg_file_check 200 + slpi->valid.num_ms_check; 201 u64 head_size = sizeof(sal_log_mod_error_info_t) * total_check_num 202 + sizeof(sal_log_processor_info_t); 203 u64 mid_size = slpi->valid.cpuid_info * sizeof(struct sal_cpuid_info); 204 205 peidx_head(peidx) = slpi; 206 peidx_mid(peidx) = (struct sal_cpuid_info *) 207 (slpi->valid.cpuid_info ? ((char*)slpi + head_size) : NULL); 208 peidx_bottom(peidx) = (sal_processor_static_info_t *) 209 (slpi->valid.psi_static_struct ? 210 ((char*)slpi + head_size + mid_size) : NULL); 211 } 212 213 /** 214 * mca_make_slidx - Make index of SAL error record 215 * @buffer: pointer to SAL error record 216 * @slidx: pointer to index of SAL error record 217 * 218 * Return value: 219 * 1 if record has platform error / 0 if not 220 */ 221 #define LOG_INDEX_ADD_SECT_PTR(sect, ptr) \ 222 {slidx_list_t *hl = &slidx_pool.buffer[slidx_pool.cur_idx]; \ 223 hl->hdr = ptr; \ 224 list_add(&hl->list, &(sect)); \ 225 slidx_pool.cur_idx = (slidx_pool.cur_idx + 1)%slidx_pool.max_idx; } 226 227 static int 228 mca_make_slidx(void *buffer, slidx_table_t *slidx) 229 { 230 int platform_err = 0; 231 int record_len = ((sal_log_record_header_t*)buffer)->len; 232 u32 ercd_pos; 233 int sects; 234 sal_log_section_hdr_t *sp; 235 236 /* 237 * Initialize index referring current record 238 */ 239 INIT_LIST_HEAD(&(slidx->proc_err)); 240 INIT_LIST_HEAD(&(slidx->mem_dev_err)); 241 INIT_LIST_HEAD(&(slidx->sel_dev_err)); 242 INIT_LIST_HEAD(&(slidx->pci_bus_err)); 243 INIT_LIST_HEAD(&(slidx->smbios_dev_err)); 244 INIT_LIST_HEAD(&(slidx->pci_comp_err)); 245 INIT_LIST_HEAD(&(slidx->plat_specific_err)); 246 INIT_LIST_HEAD(&(slidx->host_ctlr_err)); 247 INIT_LIST_HEAD(&(slidx->plat_bus_err)); 248 INIT_LIST_HEAD(&(slidx->unsupported)); 249 250 /* 251 * Extract a Record Header 252 */ 253 slidx->header = buffer; 254 255 /* 256 * Extract each section records 257 * (arranged from "int ia64_log_platform_info_print()") 258 */ 259 for (ercd_pos = sizeof(sal_log_record_header_t), sects = 0; 260 ercd_pos < record_len; ercd_pos += sp->len, sects++) { 261 sp = (sal_log_section_hdr_t *)((char*)buffer + ercd_pos); 262 if (!efi_guidcmp(sp->guid, SAL_PROC_DEV_ERR_SECT_GUID)) { 263 LOG_INDEX_ADD_SECT_PTR(slidx->proc_err, sp); 264 } else if (!efi_guidcmp(sp->guid, 265 SAL_PLAT_MEM_DEV_ERR_SECT_GUID)) { 266 platform_err = 1; 267 LOG_INDEX_ADD_SECT_PTR(slidx->mem_dev_err, sp); 268 } else if (!efi_guidcmp(sp->guid, 269 SAL_PLAT_SEL_DEV_ERR_SECT_GUID)) { 270 platform_err = 1; 271 LOG_INDEX_ADD_SECT_PTR(slidx->sel_dev_err, sp); 272 } else if (!efi_guidcmp(sp->guid, 273 SAL_PLAT_PCI_BUS_ERR_SECT_GUID)) { 274 platform_err = 1; 275 LOG_INDEX_ADD_SECT_PTR(slidx->pci_bus_err, sp); 276 } else if (!efi_guidcmp(sp->guid, 277 SAL_PLAT_SMBIOS_DEV_ERR_SECT_GUID)) { 278 platform_err = 1; 279 LOG_INDEX_ADD_SECT_PTR(slidx->smbios_dev_err, sp); 280 } else if (!efi_guidcmp(sp->guid, 281 SAL_PLAT_PCI_COMP_ERR_SECT_GUID)) { 282 platform_err = 1; 283 LOG_INDEX_ADD_SECT_PTR(slidx->pci_comp_err, sp); 284 } else if (!efi_guidcmp(sp->guid, 285 SAL_PLAT_SPECIFIC_ERR_SECT_GUID)) { 286 platform_err = 1; 287 LOG_INDEX_ADD_SECT_PTR(slidx->plat_specific_err, sp); 288 } else if (!efi_guidcmp(sp->guid, 289 SAL_PLAT_HOST_CTLR_ERR_SECT_GUID)) { 290 platform_err = 1; 291 LOG_INDEX_ADD_SECT_PTR(slidx->host_ctlr_err, sp); 292 } else if (!efi_guidcmp(sp->guid, 293 SAL_PLAT_BUS_ERR_SECT_GUID)) { 294 platform_err = 1; 295 LOG_INDEX_ADD_SECT_PTR(slidx->plat_bus_err, sp); 296 } else { 297 LOG_INDEX_ADD_SECT_PTR(slidx->unsupported, sp); 298 } 299 } 300 slidx->n_sections = sects; 301 302 return platform_err; 303 } 304 305 /** 306 * init_record_index_pools - Initialize pool of lists for SAL record index 307 * 308 * Return value: 309 * 0 on Success / -ENOMEM on Failure 310 */ 311 static int 312 init_record_index_pools(void) 313 { 314 int i; 315 int rec_max_size; /* Maximum size of SAL error records */ 316 int sect_min_size; /* Minimum size of SAL error sections */ 317 /* minimum size table of each section */ 318 static int sal_log_sect_min_sizes[] = { 319 sizeof(sal_log_processor_info_t) 320 + sizeof(sal_processor_static_info_t), 321 sizeof(sal_log_mem_dev_err_info_t), 322 sizeof(sal_log_sel_dev_err_info_t), 323 sizeof(sal_log_pci_bus_err_info_t), 324 sizeof(sal_log_smbios_dev_err_info_t), 325 sizeof(sal_log_pci_comp_err_info_t), 326 sizeof(sal_log_plat_specific_err_info_t), 327 sizeof(sal_log_host_ctlr_err_info_t), 328 sizeof(sal_log_plat_bus_err_info_t), 329 }; 330 331 /* 332 * MCA handler cannot allocate new memory on flight, 333 * so we preallocate enough memory to handle a SAL record. 334 * 335 * Initialize a handling set of slidx_pool: 336 * 1. Pick up the max size of SAL error records 337 * 2. Pick up the min size of SAL error sections 338 * 3. Allocate the pool as enough to 2 SAL records 339 * (now we can estimate the maxinum of section in a record.) 340 */ 341 342 /* - 1 - */ 343 rec_max_size = sal_rec_max; 344 345 /* - 2 - */ 346 sect_min_size = sal_log_sect_min_sizes[0]; 347 for (i = 1; i < sizeof sal_log_sect_min_sizes/sizeof(size_t); i++) 348 if (sect_min_size > sal_log_sect_min_sizes[i]) 349 sect_min_size = sal_log_sect_min_sizes[i]; 350 351 /* - 3 - */ 352 slidx_pool.max_idx = (rec_max_size/sect_min_size) * 2 + 1; 353 slidx_pool.buffer = 354 kmalloc_array(slidx_pool.max_idx, sizeof(slidx_list_t), 355 GFP_KERNEL); 356 357 return slidx_pool.buffer ? 0 : -ENOMEM; 358 } 359 360 361 /***************************************************************************** 362 * Recovery functions * 363 *****************************************************************************/ 364 365 /** 366 * is_mca_global - Check whether this MCA is global or not 367 * @peidx: pointer of index of processor error section 368 * @pbci: pointer to pal_bus_check_info_t 369 * @sos: pointer to hand off struct between SAL and OS 370 * 371 * Return value: 372 * MCA_IS_LOCAL / MCA_IS_GLOBAL 373 */ 374 375 static mca_type_t 376 is_mca_global(peidx_table_t *peidx, pal_bus_check_info_t *pbci, 377 struct ia64_sal_os_state *sos) 378 { 379 pal_processor_state_info_t *psp = 380 (pal_processor_state_info_t*)peidx_psp(peidx); 381 382 /* 383 * PAL can request a rendezvous, if the MCA has a global scope. 384 * If "rz_always" flag is set, SAL requests MCA rendezvous 385 * in spite of global MCA. 386 * Therefore it is local MCA when rendezvous has not been requested. 387 * Failed to rendezvous, the system must be down. 388 */ 389 switch (sos->rv_rc) { 390 case -1: /* SAL rendezvous unsuccessful */ 391 return MCA_IS_GLOBAL; 392 case 0: /* SAL rendezvous not required */ 393 return MCA_IS_LOCAL; 394 case 1: /* SAL rendezvous successful int */ 395 case 2: /* SAL rendezvous successful int with init */ 396 default: 397 break; 398 } 399 400 /* 401 * If One or more Cache/TLB/Reg_File/Uarch_Check is here, 402 * it would be a local MCA. (i.e. processor internal error) 403 */ 404 if (psp->tc || psp->cc || psp->rc || psp->uc) 405 return MCA_IS_LOCAL; 406 407 /* 408 * Bus_Check structure with Bus_Check.ib (internal bus error) flag set 409 * would be a global MCA. (e.g. a system bus address parity error) 410 */ 411 if (!pbci || pbci->ib) 412 return MCA_IS_GLOBAL; 413 414 /* 415 * Bus_Check structure with Bus_Check.eb (external bus error) flag set 416 * could be either a local MCA or a global MCA. 417 * 418 * Referring Bus_Check.bsi: 419 * 0: Unknown/unclassified 420 * 1: BERR# 421 * 2: BINIT# 422 * 3: Hard Fail 423 * (FIXME: Are these SGI specific or generic bsi values?) 424 */ 425 if (pbci->eb) 426 switch (pbci->bsi) { 427 case 0: 428 /* e.g. a load from poisoned memory */ 429 return MCA_IS_LOCAL; 430 case 1: 431 case 2: 432 case 3: 433 return MCA_IS_GLOBAL; 434 } 435 436 return MCA_IS_GLOBAL; 437 } 438 439 /** 440 * get_target_identifier - Get the valid Cache or Bus check target identifier. 441 * @peidx: pointer of index of processor error section 442 * 443 * Return value: 444 * target address on Success / 0 on Failure 445 */ 446 static u64 447 get_target_identifier(peidx_table_t *peidx) 448 { 449 u64 target_address = 0; 450 sal_log_mod_error_info_t *smei; 451 pal_cache_check_info_t *pcci; 452 int i, level = 9; 453 454 /* 455 * Look through the cache checks for a valid target identifier 456 * If more than one valid target identifier, return the one 457 * with the lowest cache level. 458 */ 459 for (i = 0; i < peidx_cache_check_num(peidx); i++) { 460 smei = (sal_log_mod_error_info_t *)peidx_cache_check(peidx, i); 461 if (smei->valid.target_identifier && smei->target_identifier) { 462 pcci = (pal_cache_check_info_t *)&(smei->check_info); 463 if (!target_address || (pcci->level < level)) { 464 target_address = smei->target_identifier; 465 level = pcci->level; 466 continue; 467 } 468 } 469 } 470 if (target_address) 471 return target_address; 472 473 /* 474 * Look at the bus check for a valid target identifier 475 */ 476 smei = peidx_bus_check(peidx, 0); 477 if (smei && smei->valid.target_identifier) 478 return smei->target_identifier; 479 480 return 0; 481 } 482 483 /** 484 * recover_from_read_error - Try to recover the errors which type are "read"s. 485 * @slidx: pointer of index of SAL error record 486 * @peidx: pointer of index of processor error section 487 * @pbci: pointer of pal_bus_check_info 488 * @sos: pointer to hand off struct between SAL and OS 489 * 490 * Return value: 491 * 1 on Success / 0 on Failure 492 */ 493 494 static int 495 recover_from_read_error(slidx_table_t *slidx, 496 peidx_table_t *peidx, pal_bus_check_info_t *pbci, 497 struct ia64_sal_os_state *sos) 498 { 499 u64 target_identifier; 500 pal_min_state_area_t *pmsa; 501 struct ia64_psr *psr1, *psr2; 502 ia64_fptr_t *mca_hdlr_bh = (ia64_fptr_t*)mca_handler_bhhook; 503 504 /* Is target address valid? */ 505 target_identifier = get_target_identifier(peidx); 506 if (!target_identifier) 507 return fatal_mca("target address not valid"); 508 509 /* 510 * cpu read or memory-mapped io read 511 * 512 * offending process affected process OS MCA do 513 * kernel mode kernel mode down system 514 * kernel mode user mode kill the process 515 * user mode kernel mode down system (*) 516 * user mode user mode kill the process 517 * 518 * (*) You could terminate offending user-mode process 519 * if (pbci->pv && pbci->pl != 0) *and* if you sure 520 * the process not have any locks of kernel. 521 */ 522 523 /* Is minstate valid? */ 524 if (!peidx_bottom(peidx) || !(peidx_bottom(peidx)->valid.minstate)) 525 return fatal_mca("minstate not valid"); 526 psr1 =(struct ia64_psr *)&(peidx_minstate_area(peidx)->pmsa_ipsr); 527 psr2 =(struct ia64_psr *)&(peidx_minstate_area(peidx)->pmsa_xpsr); 528 529 /* 530 * Check the privilege level of interrupted context. 531 * If it is user-mode, then terminate affected process. 532 */ 533 534 pmsa = sos->pal_min_state; 535 if (psr1->cpl != 0 || 536 ((psr2->cpl != 0) && mca_recover_range(pmsa->pmsa_iip))) { 537 /* 538 * setup for resume to bottom half of MCA, 539 * "mca_handler_bhhook" 540 */ 541 /* pass to bhhook as argument (gr8, ...) */ 542 pmsa->pmsa_gr[8-1] = target_identifier; 543 pmsa->pmsa_gr[9-1] = pmsa->pmsa_iip; 544 pmsa->pmsa_gr[10-1] = pmsa->pmsa_ipsr; 545 /* set interrupted return address (but no use) */ 546 pmsa->pmsa_br0 = pmsa->pmsa_iip; 547 /* change resume address to bottom half */ 548 pmsa->pmsa_iip = mca_hdlr_bh->fp; 549 pmsa->pmsa_gr[1-1] = mca_hdlr_bh->gp; 550 /* set cpl with kernel mode */ 551 psr2 = (struct ia64_psr *)&pmsa->pmsa_ipsr; 552 psr2->cpl = 0; 553 psr2->ri = 0; 554 psr2->bn = 1; 555 psr2->i = 0; 556 557 return mca_recovered("user memory corruption. " 558 "kill affected process - recovered."); 559 } 560 561 return fatal_mca("kernel context not recovered, iip 0x%lx\n", 562 pmsa->pmsa_iip); 563 } 564 565 /** 566 * recover_from_platform_error - Recover from platform error. 567 * @slidx: pointer of index of SAL error record 568 * @peidx: pointer of index of processor error section 569 * @pbci: pointer of pal_bus_check_info 570 * @sos: pointer to hand off struct between SAL and OS 571 * 572 * Return value: 573 * 1 on Success / 0 on Failure 574 */ 575 576 static int 577 recover_from_platform_error(slidx_table_t *slidx, peidx_table_t *peidx, 578 pal_bus_check_info_t *pbci, 579 struct ia64_sal_os_state *sos) 580 { 581 int status = 0; 582 pal_processor_state_info_t *psp = 583 (pal_processor_state_info_t*)peidx_psp(peidx); 584 585 if (psp->bc && pbci->eb && pbci->bsi == 0) { 586 switch(pbci->type) { 587 case 1: /* partial read */ 588 case 3: /* full line(cpu) read */ 589 case 9: /* I/O space read */ 590 status = recover_from_read_error(slidx, peidx, pbci, 591 sos); 592 break; 593 case 0: /* unknown */ 594 case 2: /* partial write */ 595 case 4: /* full line write */ 596 case 5: /* implicit or explicit write-back operation */ 597 case 6: /* snoop probe */ 598 case 7: /* incoming or outgoing ptc.g */ 599 case 8: /* write coalescing transactions */ 600 case 10: /* I/O space write */ 601 case 11: /* inter-processor interrupt message(IPI) */ 602 case 12: /* interrupt acknowledge or 603 external task priority cycle */ 604 default: 605 break; 606 } 607 } else if (psp->cc && !psp->bc) { /* Cache error */ 608 status = recover_from_read_error(slidx, peidx, pbci, sos); 609 } 610 611 return status; 612 } 613 614 /* 615 * recover_from_tlb_check 616 * @peidx: pointer of index of processor error section 617 * 618 * Return value: 619 * 1 on Success / 0 on Failure 620 */ 621 static int 622 recover_from_tlb_check(peidx_table_t *peidx) 623 { 624 sal_log_mod_error_info_t *smei; 625 pal_tlb_check_info_t *ptci; 626 627 smei = (sal_log_mod_error_info_t *)peidx_tlb_check(peidx, 0); 628 ptci = (pal_tlb_check_info_t *)&(smei->check_info); 629 630 /* 631 * Look for signature of a duplicate TLB DTC entry, which is 632 * a SW bug and always fatal. 633 */ 634 if (ptci->op == PAL_TLB_CHECK_OP_PURGE 635 && !(ptci->itr || ptci->dtc || ptci->itc)) 636 return fatal_mca("Duplicate TLB entry"); 637 638 return mca_recovered("TLB check recovered"); 639 } 640 641 /** 642 * recover_from_processor_error 643 * @platform: whether there are some platform error section or not 644 * @slidx: pointer of index of SAL error record 645 * @peidx: pointer of index of processor error section 646 * @pbci: pointer of pal_bus_check_info 647 * @sos: pointer to hand off struct between SAL and OS 648 * 649 * Return value: 650 * 1 on Success / 0 on Failure 651 */ 652 653 static int 654 recover_from_processor_error(int platform, slidx_table_t *slidx, 655 peidx_table_t *peidx, pal_bus_check_info_t *pbci, 656 struct ia64_sal_os_state *sos) 657 { 658 pal_processor_state_info_t *psp = 659 (pal_processor_state_info_t*)peidx_psp(peidx); 660 661 /* 662 * Processor recovery status must key off of the PAL recovery 663 * status in the Processor State Parameter. 664 */ 665 666 /* 667 * The machine check is corrected. 668 */ 669 if (psp->cm == 1) 670 return mca_recovered("machine check is already corrected."); 671 672 /* 673 * The error was not contained. Software must be reset. 674 */ 675 if (psp->us || psp->ci == 0) 676 return fatal_mca("error not contained"); 677 678 /* 679 * Look for recoverable TLB check 680 */ 681 if (psp->tc && !(psp->cc || psp->bc || psp->rc || psp->uc)) 682 return recover_from_tlb_check(peidx); 683 684 /* 685 * The cache check and bus check bits have four possible states 686 * cc bc 687 * 1 1 Memory error, attempt recovery 688 * 1 0 Cache error, attempt recovery 689 * 0 1 I/O error, attempt recovery 690 * 0 0 Other error type, not recovered 691 */ 692 if (psp->cc == 0 && (psp->bc == 0 || pbci == NULL)) 693 return fatal_mca("No cache or bus check"); 694 695 /* 696 * Cannot handle more than one bus check. 697 */ 698 if (peidx_bus_check_num(peidx) > 1) 699 return fatal_mca("Too many bus checks"); 700 701 if (pbci->ib) 702 return fatal_mca("Internal Bus error"); 703 if (pbci->eb && pbci->bsi > 0) 704 return fatal_mca("External bus check fatal status"); 705 706 /* 707 * This is a local MCA and estimated as a recoverable error. 708 */ 709 if (platform) 710 return recover_from_platform_error(slidx, peidx, pbci, sos); 711 712 /* 713 * On account of strange SAL error record, we cannot recover. 714 */ 715 return fatal_mca("Strange SAL record"); 716 } 717 718 /** 719 * mca_try_to_recover - Try to recover from MCA 720 * @rec: pointer to a SAL error record 721 * @sos: pointer to hand off struct between SAL and OS 722 * 723 * Return value: 724 * 1 on Success / 0 on Failure 725 */ 726 727 static int 728 mca_try_to_recover(void *rec, struct ia64_sal_os_state *sos) 729 { 730 int platform_err; 731 int n_proc_err; 732 slidx_table_t slidx; 733 peidx_table_t peidx; 734 pal_bus_check_info_t pbci; 735 736 /* Make index of SAL error record */ 737 platform_err = mca_make_slidx(rec, &slidx); 738 739 /* Count processor error sections */ 740 n_proc_err = slidx_count(&slidx, proc_err); 741 742 /* Now, OS can recover when there is one processor error section */ 743 if (n_proc_err > 1) 744 return fatal_mca("Too Many Errors"); 745 else if (n_proc_err == 0) 746 /* Weird SAL record ... We can't do anything */ 747 return fatal_mca("Weird SAL record"); 748 749 /* Make index of processor error section */ 750 mca_make_peidx((sal_log_processor_info_t*) 751 slidx_first_entry(&slidx.proc_err)->hdr, &peidx); 752 753 /* Extract Processor BUS_CHECK[0] */ 754 *((u64*)&pbci) = peidx_check_info(&peidx, bus_check, 0); 755 756 /* Check whether MCA is global or not */ 757 if (is_mca_global(&peidx, &pbci, sos)) 758 return fatal_mca("global MCA"); 759 760 /* Try to recover a processor error */ 761 return recover_from_processor_error(platform_err, &slidx, &peidx, 762 &pbci, sos); 763 } 764 765 /* 766 * ============================================================================= 767 */ 768 769 int __init mca_external_handler_init(void) 770 { 771 if (init_record_index_pools()) 772 return -ENOMEM; 773 774 /* register external mca handlers */ 775 if (ia64_reg_MCA_extension(mca_try_to_recover)) { 776 printk(KERN_ERR "ia64_reg_MCA_extension failed.\n"); 777 kfree(slidx_pool.buffer); 778 return -EFAULT; 779 } 780 return 0; 781 } 782 783 void __exit mca_external_handler_exit(void) 784 { 785 /* unregister external mca handlers */ 786 ia64_unreg_MCA_extension(); 787 kfree(slidx_pool.buffer); 788 } 789 790 module_init(mca_external_handler_init); 791 module_exit(mca_external_handler_exit); 792 793 module_param(sal_rec_max, int, 0644); 794 MODULE_PARM_DESC(sal_rec_max, "Max size of SAL error record"); 795 796 MODULE_DESCRIPTION("ia64 platform dependent mca handler driver"); 797 MODULE_LICENSE("GPL"); 798