drivers/edac/ghes_edac.c

   1 /*
   2  * GHES/EDAC Linux driver
   3  *
   4  * This file may be distributed under the terms of the GNU General Public
   5  * License version 2.
   6  *
   7  * Copyright (c) 2013 by Mauro Carvalho Chehab <mchehab@redhat.com>
   8  *
   9  * Red Hat Inc. http://www.redhat.com
  10  */
  11
  12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  13
  14 #include <acpi/ghes.h>
  15 #include <linux/edac.h>
  16 #include <linux/dmi.h>
  17 #include "edac_core.h"
  18 #include <ras/ras_event.h>
  19
  20 #define GHES_EDAC_REVISION " Ver: 1.0.0"
  21
  22 struct ghes_edac_pvt {
  23         struct list_head list;
  24         struct ghes *ghes;
  25         struct mem_ctl_info *mci;
  26
  27         /* Buffers for the error handling routine */
  28         char detail_location[240];
  29         char other_detail[160];
  30         char msg[80];
  31 };
  32
  33 static LIST_HEAD(ghes_reglist);
  34 static DEFINE_MUTEX(ghes_edac_lock);
  35 static int ghes_edac_mc_num;
  36
  37
  38 /* Memory Device - Type 17 of SMBIOS spec */
  39 struct memdev_dmi_entry {
  40         u8 type;
  41         u8 length;
  42         u16 handle;
  43         u16 phys_mem_array_handle;
  44         u16 mem_err_info_handle;
  45         u16 total_width;
  46         u16 data_width;
  47         u16 size;
  48         u8 form_factor;
  49         u8 device_set;
  50         u8 device_locator;
  51         u8 bank_locator;
  52         u8 memory_type;
  53         u16 type_detail;
  54         u16 speed;
  55         u8 manufacturer;
  56         u8 serial_number;
  57         u8 asset_tag;
  58         u8 part_number;
  59         u8 attributes;
  60         u32 extended_size;
  61         u16 conf_mem_clk_speed;
  62 } __attribute__((__packed__));
  63
  64 struct ghes_edac_dimm_fill {
  65         struct mem_ctl_info *mci;
  66         unsigned count;
  67 };
  68
  69 char *memory_type[] = {
  70         [MEM_EMPTY] = "EMPTY",
  71         [MEM_RESERVED] = "RESERVED",
  72         [MEM_UNKNOWN] = "UNKNOWN",
  73         [MEM_FPM] = "FPM",
  74         [MEM_EDO] = "EDO",
  75         [MEM_BEDO] = "BEDO",
  76         [MEM_SDR] = "SDR",
  77         [MEM_RDR] = "RDR",
  78         [MEM_DDR] = "DDR",
  79         [MEM_RDDR] = "RDDR",
  80         [MEM_RMBS] = "RMBS",
  81         [MEM_DDR2] = "DDR2",
  82         [MEM_FB_DDR2] = "FB_DDR2",
  83         [MEM_RDDR2] = "RDDR2",
  84         [MEM_XDR] = "XDR",
  85         [MEM_DDR3] = "DDR3",
  86         [MEM_RDDR3] = "RDDR3",
  87 };
  88
  89 static void ghes_edac_count_dimms(const struct dmi_header *dh, void *arg)
  90 {
  91         int *num_dimm = arg;
  92
  93         if (dh->type == DMI_ENTRY_MEM_DEVICE)
  94                 (*num_dimm)++;
  95 }
  96
  97 static void ghes_edac_dmidecode(const struct dmi_header *dh, void *arg)
  98 {
  99         struct ghes_edac_dimm_fill *dimm_fill = arg;
 100         struct mem_ctl_info *mci = dimm_fill->mci;
 101
 102         if (dh->type == DMI_ENTRY_MEM_DEVICE) {
 103                 struct memdev_dmi_entry *entry = (struct memdev_dmi_entry *)dh;
 104                 struct dimm_info *dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms,
 105                                                        mci->n_layers,
 106                                                        dimm_fill->count, 0, 0);
 107
 108                 if (entry->size == 0xffff) {
 109                         pr_info("Can't get DIMM%i size\n",
 110                                 dimm_fill->count);
 111                         dimm->nr_pages = MiB_TO_PAGES(32);/* Unknown */
 112                 } else if (entry->size == 0x7fff) {
 113                         dimm->nr_pages = MiB_TO_PAGES(entry->extended_size);
 114                 } else {
 115                         if (entry->size & 1 << 15)
 116                                 dimm->nr_pages = MiB_TO_PAGES((entry->size &
 117                                                                0x7fff) << 10);
 118                         else
 119                                 dimm->nr_pages = MiB_TO_PAGES(entry->size);
 120                 }
 121
 122                 switch (entry->memory_type) {
 123                 case 0x12:
 124                         if (entry->type_detail & 1 << 13)
 125                                 dimm->mtype = MEM_RDDR;
 126                         else
 127                                 dimm->mtype = MEM_DDR;
 128                         break;
 129                 case 0x13:
 130                         if (entry->type_detail & 1 << 13)
 131                                 dimm->mtype = MEM_RDDR2;
 132                         else
 133                                 dimm->mtype = MEM_DDR2;
 134                         break;
 135                 case 0x14:
 136                         dimm->mtype = MEM_FB_DDR2;
 137                         break;
 138                 case 0x18:
 139                         if (entry->type_detail & 1 << 13)
 140                                 dimm->mtype = MEM_RDDR3;
 141                         else
 142                                 dimm->mtype = MEM_DDR3;
 143                         break;
 144                 default:
 145                         if (entry->type_detail & 1 << 6)
 146                                 dimm->mtype = MEM_RMBS;
 147                         else if ((entry->type_detail & ((1 << 7) | (1 << 13)))
 148                                  == ((1 << 7) | (1 << 13)))
 149                                 dimm->mtype = MEM_RDR;
 150                         else if (entry->type_detail & 1 << 7)
 151                                 dimm->mtype = MEM_SDR;
 152                         else if (entry->type_detail & 1 << 9)
 153                                 dimm->mtype = MEM_EDO;
 154                         else
 155                                 dimm->mtype = MEM_UNKNOWN;
 156                 }
 157
 158                 /*
 159                  * Actually, we can only detect if the memory has bits for
 160                  * checksum or not
 161                  */
 162                 if (entry->total_width == entry->data_width)
 163                         dimm->edac_mode = EDAC_NONE;
 164                 else
 165                         dimm->edac_mode = EDAC_SECDED;
 166
 167                 dimm->dtype = DEV_UNKNOWN;
 168                 dimm->grain = 128;              /* Likely, worse case */
 169
 170                 /*
 171                  * FIXME: It shouldn't be hard to also fill the DIMM labels
 172                  */
 173
 174                 if (dimm->nr_pages) {
 175                         edac_dbg(1, "DIMM%i: %s size = %d MB%s\n",
 176                                 dimm_fill->count, memory_type[dimm->mtype],
 177                                 PAGES_TO_MiB(dimm->nr_pages),
 178                                 (dimm->edac_mode != EDAC_NONE) ? "(ECC)" : "");
 179                         edac_dbg(2, "\ttype %d, detail 0x%02x, width %d(total %d)\n",
 180                                 entry->memory_type, entry->type_detail,
 181                                 entry->total_width, entry->data_width);
 182                 }
 183
 184                 dimm_fill->count++;
 185         }
 186 }
 187
 188 void ghes_edac_report_mem_error(struct ghes *ghes, int sev,
 189                                 struct cper_sec_mem_err *mem_err)
 190 {
 191         enum hw_event_mc_err_type type;
 192         struct edac_raw_error_desc *e;
 193         struct mem_ctl_info *mci;
 194         struct ghes_edac_pvt *pvt = NULL;
 195         char *p;
 196         u8 grain_bits;
 197
 198         list_for_each_entry(pvt, &ghes_reglist, list) {
 199                 if (ghes == pvt->ghes)
 200                         break;
 201         }
 202         if (!pvt) {
 203                 pr_err("Internal error: Can't find EDAC structure\n");
 204                 return;
 205         }
 206         mci = pvt->mci;
 207         e = &mci->error_desc;
 208
 209         /* Cleans the error report buffer */
 210         memset(e, 0, sizeof (*e));
 211         e->error_count = 1;
 212         strcpy(e->label, "unknown label");
 213         e->msg = pvt->msg;
 214         e->other_detail = pvt->other_detail;
 215         e->top_layer = -1;
 216         e->mid_layer = -1;
 217         e->low_layer = -1;
 218         *pvt->other_detail = '\0';
 219         *pvt->msg = '\0';
 220
 221         switch (sev) {
 222         case GHES_SEV_CORRECTED:
 223                 type = HW_EVENT_ERR_CORRECTED;
 224                 break;
 225         case GHES_SEV_RECOVERABLE:
 226                 type = HW_EVENT_ERR_UNCORRECTED;
 227                 break;
 228         case GHES_SEV_PANIC:
 229                 type = HW_EVENT_ERR_FATAL;
 230                 break;
 231         default:
 232         case GHES_SEV_NO:
 233                 type = HW_EVENT_ERR_INFO;
 234         }
 235
 236         edac_dbg(1, "error validation_bits: 0x%08llx\n",
 237                  (long long)mem_err->validation_bits);
 238
 239         /* Error type, mapped on e->msg */
 240         if (mem_err->validation_bits & CPER_MEM_VALID_ERROR_TYPE) {
 241                 p = pvt->msg;
 242                 switch (mem_err->error_type) {
 243                 case 0:
 244                         p += sprintf(p, "Unknown");
 245                         break;
 246                 case 1:
 247                         p += sprintf(p, "No error");
 248                         break;
 249                 case 2:
 250                         p += sprintf(p, "Single-bit ECC");
 251                         break;
 252                 case 3:
 253                         p += sprintf(p, "Multi-bit ECC");
 254                         break;
 255                 case 4:
 256                         p += sprintf(p, "Single-symbol ChipKill ECC");
 257                         break;
 258                 case 5:
 259                         p += sprintf(p, "Multi-symbol ChipKill ECC");
 260                         break;
 261                 case 6:
 262                         p += sprintf(p, "Master abort");
 263                         break;
 264                 case 7:
 265                         p += sprintf(p, "Target abort");
 266                         break;
 267                 case 8:
 268                         p += sprintf(p, "Parity Error");
 269                         break;
 270                 case 9:
 271                         p += sprintf(p, "Watchdog timeout");
 272                         break;
 273                 case 10:
 274                         p += sprintf(p, "Invalid address");
 275                         break;
 276                 case 11:
 277                         p += sprintf(p, "Mirror Broken");
 278                         break;
 279                 case 12:
 280                         p += sprintf(p, "Memory Sparing");
 281                         break;
 282                 case 13:
 283                         p += sprintf(p, "Scrub corrected error");
 284                         break;
 285                 case 14:
 286                         p += sprintf(p, "Scrub uncorrected error");
 287                         break;
 288                 case 15:
 289                         p += sprintf(p, "Physical Memory Map-out event");
 290                         break;
 291                 default:
 292                         p += sprintf(p, "reserved error (%d)",
 293                                      mem_err->error_type);
 294                 }
 295         } else {
 296                 strcpy(pvt->msg, "unknown error");
 297         }
 298
 299         /* Error address */
 300         if (mem_err->validation_bits & CPER_MEM_VALID_PA) {
 301                 e->page_frame_number = mem_err->physical_addr >> PAGE_SHIFT;
 302                 e->offset_in_page = mem_err->physical_addr & ~PAGE_MASK;
 303         }
 304
 305         /* Error grain */
 306         if (mem_err->validation_bits & CPER_MEM_VALID_PA_MASK)
 307                 e->grain = ~(mem_err->physical_addr_mask & ~PAGE_MASK);
 308
 309         /* Memory error location, mapped on e->location */
 310         p = e->location;
 311         if (mem_err->validation_bits & CPER_MEM_VALID_NODE)
 312                 p += sprintf(p, "node:%d ", mem_err->node);
 313         if (mem_err->validation_bits & CPER_MEM_VALID_CARD)
 314                 p += sprintf(p, "card:%d ", mem_err->card);
 315         if (mem_err->validation_bits & CPER_MEM_VALID_MODULE)
 316                 p += sprintf(p, "module:%d ", mem_err->module);
 317         if (mem_err->validation_bits & CPER_MEM_VALID_RANK_NUMBER)
 318                 p += sprintf(p, "rank:%d ", mem_err->rank);
 319         if (mem_err->validation_bits & CPER_MEM_VALID_BANK)
 320                 p += sprintf(p, "bank:%d ", mem_err->bank);
 321         if (mem_err->validation_bits & CPER_MEM_VALID_ROW)
 322                 p += sprintf(p, "row:%d ", mem_err->row);
 323         if (mem_err->validation_bits & CPER_MEM_VALID_COLUMN)
 324                 p += sprintf(p, "col:%d ", mem_err->column);
 325         if (mem_err->validation_bits & CPER_MEM_VALID_BIT_POSITION)
 326                 p += sprintf(p, "bit_pos:%d ", mem_err->bit_pos);
 327         if (mem_err->validation_bits & CPER_MEM_VALID_MODULE_HANDLE) {
 328                 const char *bank = NULL, *device = NULL;
 329                 dmi_memdev_name(mem_err->mem_dev_handle, &bank, &device);
 330                 if (bank != NULL && device != NULL)
 331                         p += sprintf(p, "DIMM location:%s %s ", bank, device);
 332                 else
 333                         p += sprintf(p, "DIMM DMI handle: 0x%.4x ",
 334                                      mem_err->mem_dev_handle);
 335         }
 336         if (p > e->location)
 337                 *(p - 1) = '\0';
 338
 339         /* All other fields are mapped on e->other_detail */
 340         p = pvt->other_detail;
 341         if (mem_err->validation_bits & CPER_MEM_VALID_ERROR_STATUS) {
 342                 u64 status = mem_err->error_status;
 343
 344                 p += sprintf(p, "status(0x%016llx): ", (long long)status);
 345                 switch ((status >> 8) & 0xff) {
 346                 case 1:
 347                         p += sprintf(p, "Error detected internal to the component ");
 348                         break;
 349                 case 16:
 350                         p += sprintf(p, "Error detected in the bus ");
 351                         break;
 352                 case 4:
 353                         p += sprintf(p, "Storage error in DRAM memory ");
 354                         break;
 355                 case 5:
 356                         p += sprintf(p, "Storage error in TLB ");
 357                         break;
 358                 case 6:
 359                         p += sprintf(p, "Storage error in cache ");
 360                         break;
 361                 case 7:
 362                         p += sprintf(p, "Error in one or more functional units ");
 363                         break;
 364                 case 8:
 365                         p += sprintf(p, "component failed self test ");
 366                         break;
 367                 case 9:
 368                         p += sprintf(p, "Overflow or undervalue of internal queue ");
 369                         break;
 370                 case 17:
 371                         p += sprintf(p, "Virtual address not found on IO-TLB or IO-PDIR ");
 372                         break;
 373                 case 18:
 374                         p += sprintf(p, "Improper access error ");
 375                         break;
 376                 case 19:
 377                         p += sprintf(p, "Access to a memory address which is not mapped to any component ");
 378                         break;
 379                 case 20:
 380                         p += sprintf(p, "Loss of Lockstep ");
 381                         break;
 382                 case 21:
 383                         p += sprintf(p, "Response not associated with a request ");
 384                         break;
 385                 case 22:
 386                         p += sprintf(p, "Bus parity error - must also set the A, C, or D Bits ");
 387                         break;
 388                 case 23:
 389                         p += sprintf(p, "Detection of a PATH_ERROR ");
 390                         break;
 391                 case 25:
 392                         p += sprintf(p, "Bus operation timeout ");
 393                         break;
 394                 case 26:
 395                         p += sprintf(p, "A read was issued to data that has been poisoned ");
 396                         break;
 397                 default:
 398                         p += sprintf(p, "reserved ");
 399                         break;
 400                 }
 401         }
 402         if (mem_err->validation_bits & CPER_MEM_VALID_REQUESTOR_ID)
 403                 p += sprintf(p, "requestorID: 0x%016llx ",
 404                              (long long)mem_err->requestor_id);
 405         if (mem_err->validation_bits & CPER_MEM_VALID_RESPONDER_ID)
 406                 p += sprintf(p, "responderID: 0x%016llx ",
 407                              (long long)mem_err->responder_id);
 408         if (mem_err->validation_bits & CPER_MEM_VALID_TARGET_ID)
 409                 p += sprintf(p, "targetID: 0x%016llx ",
 410                              (long long)mem_err->responder_id);
 411         if (p > pvt->other_detail)
 412                 *(p - 1) = '\0';
 413
 414         /* Generate the trace event */
 415         grain_bits = fls_long(e->grain);
 416         sprintf(pvt->detail_location, "APEI location: %s %s",
 417                 e->location, e->other_detail);
 418         trace_mc_event(type, e->msg, e->label, e->error_count,
 419                        mci->mc_idx, e->top_layer, e->mid_layer, e->low_layer,
 420                        PAGES_TO_MiB(e->page_frame_number) | e->offset_in_page,
 421                        grain_bits, e->syndrome, pvt->detail_location);
 422
 423         /* Report the error via EDAC API */
 424         edac_raw_mc_handle_error(type, mci, e);
 425 }
 426 EXPORT_SYMBOL_GPL(ghes_edac_report_mem_error);
 427
 428 int ghes_edac_register(struct ghes *ghes, struct device *dev)
 429 {
 430         bool fake = false;
 431         int rc, num_dimm = 0;
 432         struct mem_ctl_info *mci;
 433         struct edac_mc_layer layers[1];
 434         struct ghes_edac_pvt *pvt;
 435         struct ghes_edac_dimm_fill dimm_fill;
 436
 437         /* Get the number of DIMMs */
 438         dmi_walk(ghes_edac_count_dimms, &num_dimm);
 439
 440         /* Check if we've got a bogus BIOS */
 441         if (num_dimm == 0) {
 442                 fake = true;
 443                 num_dimm = 1;
 444         }
 445
 446         layers[0].type = EDAC_MC_LAYER_ALL_MEM;
 447         layers[0].size = num_dimm;
 448         layers[0].is_virt_csrow = true;
 449
 450         /*
 451          * We need to serialize edac_mc_alloc() and edac_mc_add_mc(),
 452          * to avoid duplicated memory controller numbers
 453          */
 454         mutex_lock(&ghes_edac_lock);
 455         mci = edac_mc_alloc(ghes_edac_mc_num, ARRAY_SIZE(layers), layers,
 456                             sizeof(*pvt));
 457         if (!mci) {
 458                 pr_info("Can't allocate memory for EDAC data\n");
 459                 mutex_unlock(&ghes_edac_lock);
 460                 return -ENOMEM;
 461         }
 462
 463         pvt = mci->pvt_info;
 464         memset(pvt, 0, sizeof(*pvt));
 465         list_add_tail(&pvt->list, &ghes_reglist);
 466         pvt->ghes = ghes;
 467         pvt->mci  = mci;
 468         mci->pdev = dev;
 469
 470         mci->mtype_cap = MEM_FLAG_EMPTY;
 471         mci->edac_ctl_cap = EDAC_FLAG_NONE;
 472         mci->edac_cap = EDAC_FLAG_NONE;
 473         mci->mod_name = "ghes_edac.c";
 474         mci->mod_ver = GHES_EDAC_REVISION;
 475         mci->ctl_name = "ghes_edac";
 476         mci->dev_name = "ghes";
 477
 478         if (!ghes_edac_mc_num) {
 479                 if (!fake) {
 480                         pr_info("This EDAC driver relies on BIOS to enumerate memory and get error reports.\n");
 481                         pr_info("Unfortunately, not all BIOSes reflect the memory layout correctly.\n");
 482                         pr_info("So, the end result of using this driver varies from vendor to vendor.\n");
 483                         pr_info("If you find incorrect reports, please contact your hardware vendor\n");
 484                         pr_info("to correct its BIOS.\n");
 485                         pr_info("This system has %d DIMM sockets.\n",
 486                                 num_dimm);
 487                 } else {
 488                         pr_info("This system has a very crappy BIOS: It doesn't even list the DIMMS.\n");
 489                         pr_info("Its SMBIOS info is wrong. It is doubtful that the error report would\n");
 490                         pr_info("work on such system. Use this driver with caution\n");
 491                 }
 492         }
 493
 494         if (!fake) {
 495                 /*
 496                  * Fill DIMM info from DMI for the memory controller #0
 497                  *
 498                  * Keep it in blank for the other memory controllers, as
 499                  * there's no reliable way to properly credit each DIMM to
 500                  * the memory controller, as different BIOSes fill the
 501                  * DMI bank location fields on different ways
 502                  */
 503                 if (!ghes_edac_mc_num) {
 504                         dimm_fill.count = 0;
 505                         dimm_fill.mci = mci;
 506                         dmi_walk(ghes_edac_dmidecode, &dimm_fill);
 507                 }
 508         } else {
 509                 struct dimm_info *dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms,
 510                                                        mci->n_layers, 0, 0, 0);
 511
 512                 dimm->nr_pages = 1;
 513                 dimm->grain = 128;
 514                 dimm->mtype = MEM_UNKNOWN;
 515                 dimm->dtype = DEV_UNKNOWN;
 516                 dimm->edac_mode = EDAC_SECDED;
 517         }
 518
 519         rc = edac_mc_add_mc(mci);
 520         if (rc < 0) {
 521                 pr_info("Can't register at EDAC core\n");
 522                 edac_mc_free(mci);
 523                 mutex_unlock(&ghes_edac_lock);
 524                 return -ENODEV;
 525         }
 526
 527         ghes_edac_mc_num++;
 528         mutex_unlock(&ghes_edac_lock);
 529         return 0;
 530 }
 531 EXPORT_SYMBOL_GPL(ghes_edac_register);
 532
 533 void ghes_edac_unregister(struct ghes *ghes)
 534 {
 535         struct mem_ctl_info *mci;
 536         struct ghes_edac_pvt *pvt, *tmp;
 537
 538         list_for_each_entry_safe(pvt, tmp, &ghes_reglist, list) {
 539                 if (ghes == pvt->ghes) {
 540                         mci = pvt->mci;
 541                         edac_mc_del_mc(mci->pdev);
 542                         edac_mc_free(mci);
 543                         list_del(&pvt->list);
 544                 }
 545         }
 546 }
 547 EXPORT_SYMBOL_GPL(ghes_edac_unregister);