| @@ -15,11 +15,13 @@ · 14 unmodified lines |
| 15 | #include "edac_module.h" | 15 | #include "edac_module.h" |
| 16 | #include <ras/ras_event.h> | 16 | #include <ras/ras_event.h> |
| 17 | | 17 | |
| | 18 | #define OTHER_DETAIL_LEN 400 |
| | 19 | |
| 18 | struct ghes_pvt { | 20 | struct ghes_pvt { |
| 19 | struct mem_ctl_info *mci; | 21 | struct mem_ctl_info *mci; |
| 20 | | 22 | |
| 21 | /* Buffers for the error handling routine */ | 23 | /* Buffers for the error handling routine */ |
| 22 | char other_detail[400]; | 24 | char other_detail[OTHER_DETAIL_LEN]; |
| 23 | char msg[80]; | 25 | char msg[80]; |
| 24 | }; | 26 | }; |
| 25 | | 27 | |
| @@ -235,8 +237,34 @@ static void ghes_scan_system(void) · 209 unmodified lines |
| 235 | system_scanned = true; | 237 | system_scanned = true; |
| 236 | } | 238 | } |
| 237 | | 239 | |
| | 240 | static int print_mem_error_other_detail(const struct cper_sec_mem_err *mem, char *msg, |
| | 241 | const char *location, unsigned int len) |
| | 242 | { |
| | 243 | u32 n; |
| | 244 | |
| | 245 | if (!msg) |
| | 246 | return 0; |
| | 247 | |
| | 248 | n = 0; |
| | 249 | len -= 1; |
| | 250 | |
| | 251 | n += scnprintf(msg + n, len - n, "APEI location: %s ", location); |
| | 252 | |
| | 253 | if (!(mem->validation_bits & CPER_MEM_VALID_ERROR_STATUS)) |
| | 254 | goto out; |
| | 255 | |
| | 256 | n += scnprintf(msg + n, len - n, "status(0x%016llx): ", mem->error_status); |
| | 257 | n += scnprintf(msg + n, len - n, "%s ", cper_mem_err_status_str(mem->error_status)); |
| | 258 | |
| | 259 | out: |
| | 260 | msg[n] = '\0'; |
| | 261 | |
| | 262 | return n; |
| | 263 | } |
| | 264 | |
| 238 | void ghes_edac_report_mem_error(int sev, struct cper_sec_mem_err *mem_err) | 265 | void ghes_edac_report_mem_error(int sev, struct cper_sec_mem_err *mem_err) |
| 239 | { | 266 | { |
| | 267 | struct cper_mem_err_compact cmem; |
| 240 | struct edac_raw_error_desc *e; | 268 | struct edac_raw_error_desc *e; |
| 241 | struct mem_ctl_info *mci; | 269 | struct mem_ctl_info *mci; |
| 242 | struct ghes_pvt *pvt; | 270 | struct ghes_pvt *pvt; |
| @@ -292,60 +320,10 @@ void ghes_edac_report_mem_error(int sev, struct cper_sec_mem_err *mem_err) · 49 unmodified lines |
| 292 | | 320 | |
| 293 | /* Error type, mapped on e->msg */ | 321 | /* Error type, mapped on e->msg */ |
| 294 | if (mem_err->validation_bits & CPER_MEM_VALID_ERROR_TYPE) { | 322 | if (mem_err->validation_bits & CPER_MEM_VALID_ERROR_TYPE) { |
| | 323 | u8 etype = mem_err->error_type; |
| | 324 | |
| 295 | p = pvt->msg; | 325 | p = pvt->msg; |
| 296 | switch (mem_err->error_type) { | 326 | p += snprintf(p, sizeof(pvt->msg), "%s", cper_mem_err_type_str(etype)); |
| 297 | case 0: | | |
| 298 | p += sprintf(p, "Unknown"); | | |
| 299 | break; | | |
| 300 | case 1: | | |
| 301 | p += sprintf(p, "No error"); | | |
| 302 | break; | | |
| 303 | case 2: | | |
| 304 | p += sprintf(p, "Single-bit ECC"); | | |
| 305 | break; | | |
| 306 | case 3: | | |
| 307 | p += sprintf(p, "Multi-bit ECC"); | | |
| 308 | break; | | |
| 309 | case 4: | | |
| 310 | p += sprintf(p, "Single-symbol ChipKill ECC"); | | |
| 311 | break; | | |
| 312 | case 5: | | |
| 313 | p += sprintf(p, "Multi-symbol ChipKill ECC"); | | |
| 314 | break; | | |
| 315 | case 6: | | |
| 316 | p += sprintf(p, "Master abort"); | | |
| 317 | break; | | |
| 318 | case 7: | | |
| 319 | p += sprintf(p, "Target abort"); | | |
| 320 | break; | | |
| 321 | case 8: | | |
| 322 | p += sprintf(p, "Parity Error"); | | |
| 323 | break; | | |
| 324 | case 9: | | |
| 325 | p += sprintf(p, "Watchdog timeout"); | | |
| 326 | break; | | |
| 327 | case 10: | | |
| 328 | p += sprintf(p, "Invalid address"); | | |
| 329 | break; | | |
| 330 | case 11: | | |
| 331 | p += sprintf(p, "Mirror Broken"); | | |
| 332 | break; | | |
| 333 | case 12: | | |
| 334 | p += sprintf(p, "Memory Sparing"); | | |
| 335 | break; | | |
| 336 | case 13: | | |
| 337 | p += sprintf(p, "Scrub corrected error"); | | |
| 338 | break; | | |
| 339 | case 14: | | |
| 340 | p += sprintf(p, "Scrub uncorrected error"); | | |
| 341 | break; | | |
| 342 | case 15: | | |
| 343 | p += sprintf(p, "Physical Memory Map-out event"); | | |
| 344 | break; | | |
| 345 | default: | | |
| 346 | p += sprintf(p, "reserved error (%d)", | | |
| 347 | mem_err->error_type); | | |
| 348 | } | | |
| 349 | } else { | 327 | } else { |
| 350 | strcpy(pvt->msg, "unknown error"); | 328 | strcpy(pvt->msg, "unknown error"); |
| 351 | } | 329 | } |
| @@ -362,52 +340,19 @@ void ghes_edac_report_mem_error(int sev, struct cper_sec_mem_err *mem_err) · 10 unmodified lines |
| 362 | | 340 | |
| 363 | /* Memory error location, mapped on e->location */ | 341 | /* Memory error location, mapped on e->location */ |
| 364 | p = e->location; | 342 | p = e->location; |
| 365 | if (mem_err->validation_bits & CPER_MEM_VALID_NODE) | 343 | cper_mem_err_pack(mem_err, &cmem); |
| 366 | p += sprintf(p, "node:%d ", mem_err->node); | 344 | p += cper_mem_err_location(&cmem, p); |
| 367 | if (mem_err->validation_bits & CPER_MEM_VALID_CARD) | 345 | |
| 368 | p += sprintf(p, "card:%d ", mem_err->card); | | |
| 369 | if (mem_err->validation_bits & CPER_MEM_VALID_MODULE) | | |
| 370 | p += sprintf(p, "module:%d ", mem_err->module); | | |
| 371 | if (mem_err->validation_bits & CPER_MEM_VALID_RANK_NUMBER) | | |
| 372 | p += sprintf(p, "rank:%d ", mem_err->rank); | | |
| 373 | if (mem_err->validation_bits & CPER_MEM_VALID_BANK) | | |
| 374 | p += sprintf(p, "bank:%d ", mem_err->bank); | | |
| 375 | if (mem_err->validation_bits & CPER_MEM_VALID_BANK_GROUP) | | |
| 376 | p += sprintf(p, "bank_group:%d ", | | |
| 377 | mem_err->bank >> CPER_MEM_BANK_GROUP_SHIFT); | | |
| 378 | if (mem_err->validation_bits & CPER_MEM_VALID_BANK_ADDRESS) | | |
| 379 | p += sprintf(p, "bank_address:%d ", | | |
| 380 | mem_err->bank & CPER_MEM_BANK_ADDRESS_MASK); | | |
| 381 | if (mem_err->validation_bits & (CPER_MEM_VALID_ROW | CPER_MEM_VALID_ROW_EXT)) { | | |
| 382 | u32 row = mem_err->row; | | |
| 383 | | | |
| 384 | row |= cper_get_mem_extension(mem_err->validation_bits, mem_err->extended); | | |
| 385 | p += sprintf(p, "row:%d ", row); | | |
| 386 | } | | |
| 387 | if (mem_err->validation_bits & CPER_MEM_VALID_COLUMN) | | |
| 388 | p += sprintf(p, "col:%d ", mem_err->column); | | |
| 389 | if (mem_err->validation_bits & CPER_MEM_VALID_BIT_POSITION) | | |
| 390 | p += sprintf(p, "bit_pos:%d ", mem_err->bit_pos); | | |
| 391 | if (mem_err->validation_bits & CPER_MEM_VALID_MODULE_HANDLE) { | 346 | if (mem_err->validation_bits & CPER_MEM_VALID_MODULE_HANDLE) { |
| 392 | const char *bank = NULL, *device = NULL; | | |
| 393 | struct dimm_info *dimm; | 347 | struct dimm_info *dimm; |
| 394 | | 348 | |
| 395 | dmi_memdev_name(mem_err->mem_dev_handle, &bank, &device); | 349 | p += cper_dimm_err_location(&cmem, p); |
| 396 | if (bank != NULL && device != NULL) | | |
| 397 | p += sprintf(p, "DIMM location:%s %s ", bank, device); | | |
| 398 | else | | |
| 399 | p += sprintf(p, "DIMM DMI handle: 0x%.4x ", | | |
| 400 | mem_err->mem_dev_handle); | | |
| 401 | | | |
| 402 | dimm = find_dimm_by_handle(mci, mem_err->mem_dev_handle); | 350 | dimm = find_dimm_by_handle(mci, mem_err->mem_dev_handle); |
| 403 | if (dimm) { | 351 | if (dimm) { |
| 404 | e->top_layer = dimm->idx; | 352 | e->top_layer = dimm->idx; |
| 405 | strcpy(e->label, dimm->label); | 353 | strcpy(e->label, dimm->label); |
| 406 | } | 354 | } |
| 407 | } | 355 | } |
| 408 | if (mem_err->validation_bits & CPER_MEM_VALID_CHIP_ID) | | |
| 409 | p += sprintf(p, "chipID: %d ", | | |
| 410 | mem_err->extended >> CPER_MEM_CHIP_ID_SHIFT); | | |
| 411 | if (p > e->location) | 356 | if (p > e->location) |
| 412 | *(p - 1) = '\0'; | 357 | *(p - 1) = '\0'; |
| 413 | | 358 | |
| @@ -416,78 +361,7 @@ void ghes_edac_report_mem_error(int sev, struct cper_sec_mem_err *mem_err) · 2 unmodified lines |
| 416 | | 361 | |
| 417 | /* All other fields are mapped on e->other_detail */ | 362 | /* All other fields are mapped on e->other_detail */ |
| 418 | p = pvt->other_detail; | 363 | p = pvt->other_detail; |
| 419 | p += snprintf(p, sizeof(pvt->other_detail), | 364 | p += print_mem_error_other_detail(mem_err, p, e->location, OTHER_DETAIL_LEN); |
| 420 | "APEI location: %s ", e->location); | | |
| 421 | if (mem_err->validation_bits & CPER_MEM_VALID_ERROR_STATUS) { | | |
| 422 | u64 status = mem_err->error_status; | | |
| 423 | | | |
| 424 | p += sprintf(p, "status(0x%016llx): ", (long long)status); | | |
| 425 | switch ((status >> 8) & 0xff) { | | |
| 426 | case 1: | | |
| 427 | p += sprintf(p, "Error detected internal to the component "); | | |
| 428 | break; | | |
| 429 | case 16: | | |
| 430 | p += sprintf(p, "Error detected in the bus "); | | |
| 431 | break; | | |
| 432 | case 4: | | |
| 433 | p += sprintf(p, "Storage error in DRAM memory "); | | |
| 434 | break; | | |
| 435 | case 5: | | |
| 436 | p += sprintf(p, "Storage error in TLB "); | | |
| 437 | break; | | |
| 438 | case 6: | | |
| 439 | p += sprintf(p, "Storage error in cache "); | | |
| 440 | break; | | |
| 441 | case 7: | | |
| 442 | p += sprintf(p, "Error in one or more functional units "); | | |
| 443 | break; | | |
| 444 | case 8: | | |
| 445 | p += sprintf(p, "component failed self test "); | | |
| 446 | break; | | |
| 447 | case 9: | | |
| 448 | p += sprintf(p, "Overflow or undervalue of internal queue "); | | |
| 449 | break; | | |
| 450 | case 17: | | |
| 451 | p += sprintf(p, "Virtual address not found on IO-TLB or IO-PDIR "); | | |
| 452 | break; | | |
| 453 | case 18: | | |
| 454 | p += sprintf(p, "Improper access error "); | | |
| 455 | break; | | |
| 456 | case 19: | | |
| 457 | p += sprintf(p, "Access to a memory address which is not mapped to any component "); | | |
| 458 | break; | | |
| 459 | case 20: | | |
| 460 | p += sprintf(p, "Loss of Lockstep "); | | |
| 461 | break; | | |
| 462 | case 21: | | |
| 463 | p += sprintf(p, "Response not associated with a request "); | | |
| 464 | break; | | |
| 465 | case 22: | | |
| 466 | p += sprintf(p, "Bus parity error - must also set the A, C, or D Bits "); | | |
| 467 | break; | | |
| 468 | case 23: | | |
| 469 | p += sprintf(p, "Detection of a PATH_ERROR "); | | |
| 470 | break; | | |
| 471 | case 25: | | |
| 472 | p += sprintf(p, "Bus operation timeout "); | | |
| 473 | break; | | |
| 474 | case 26: | | |
| 475 | p += sprintf(p, "A read was issued to data that has been poisoned "); | | |
| 476 | break; | | |
| 477 | default: | | |
| 478 | p += sprintf(p, "reserved "); | | |
| 479 | break; | | |
| 480 | } | | |
| 481 | } | | |
| 482 | if (mem_err->validation_bits & CPER_MEM_VALID_REQUESTOR_ID) | | |
| 483 | p += sprintf(p, "requestorID: 0x%016llx ", | | |
| 484 | (long long)mem_err->requestor_id); | | |
| 485 | if (mem_err->validation_bits & CPER_MEM_VALID_RESPONDER_ID) | | |
| 486 | p += sprintf(p, "responderID: 0x%016llx ", | | |
| 487 | (long long)mem_err->responder_id); | | |
| 488 | if (mem_err->validation_bits & CPER_MEM_VALID_TARGET_ID) | | |
| 489 | p += sprintf(p, "targetID: 0x%016llx ", | | |
| 490 | (long long)mem_err->responder_id); | | |
| 491 | if (p > pvt->other_detail) | 365 | if (p > pvt->other_detail) |
| 492 | *(p - 1) = '\0'; | 366 | *(p - 1) = '\0'; |
| 493 | | 367 | |