1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * 4 * Shared code by both skx_edac and i10nm_edac. Originally split out 5 * from the skx_edac driver. 6 * 7 * This file is linked into both skx_edac and i10nm_edac drivers. In 8 * order to avoid link errors, this file must be like a pure library 9 * without including symbols and defines which would otherwise conflict, 10 * when linked once into a module and into a built-in object, at the 11 * same time. For example, __this_module symbol references when that 12 * file is being linked into a built-in object. 13 * 14 * Copyright (c) 2018, Intel Corporation. 15 */ 16 17 #include <linux/acpi.h> 18 #include <linux/dmi.h> 19 #include <linux/adxl.h> 20 #include <acpi/nfit.h> 21 #include <asm/mce.h> 22 #include "edac_module.h" 23 #include "skx_common.h" 24 25 static const char * const component_names[] = { 26 [INDEX_SOCKET] = "ProcessorSocketId", 27 [INDEX_MEMCTRL] = "MemoryControllerId", 28 [INDEX_CHANNEL] = "ChannelId", 29 [INDEX_DIMM] = "DimmSlotId", 30 }; 31 32 static int component_indices[ARRAY_SIZE(component_names)]; 33 static int adxl_component_count; 34 static const char * const *adxl_component_names; 35 static u64 *adxl_values; 36 static char *adxl_msg; 37 38 static char skx_msg[MSG_SIZE]; 39 static skx_decode_f skx_decode; 40 static skx_show_retry_log_f skx_show_retry_rd_err_log; 41 static u64 skx_tolm, skx_tohm; 42 static LIST_HEAD(dev_edac_list); 43 44 int __init skx_adxl_get(void) 45 { 46 const char * const *names; 47 int i, j; 48 49 names = adxl_get_component_names(); 50 if (!names) { 51 skx_printk(KERN_NOTICE, "No firmware support for address translation.\n"); 52 return -ENODEV; 53 } 54 55 for (i = 0; i < INDEX_MAX; i++) { 56 for (j = 0; names[j]; j++) { 57 if (!strcmp(component_names[i], names[j])) { 58 component_indices[i] = j; 59 break; 60 } 61 } 62 63 if (!names[j]) 64 goto err; 65 } 66 67 adxl_component_names = names; 68 while (*names++) 69 adxl_component_count++; 70 71 adxl_values = kcalloc(adxl_component_count, sizeof(*adxl_values), 72 GFP_KERNEL); 73 if (!adxl_values) { 74 adxl_component_count = 0; 75 return -ENOMEM; 76 } 77 78 adxl_msg = kzalloc(MSG_SIZE, GFP_KERNEL); 79 if (!adxl_msg) { 80 adxl_component_count = 0; 81 kfree(adxl_values); 82 return -ENOMEM; 83 } 84 85 return 0; 86 err: 87 skx_printk(KERN_ERR, "'%s' is not matched from DSM parameters: ", 88 component_names[i]); 89 for (j = 0; names[j]; j++) 90 skx_printk(KERN_CONT, "%s ", names[j]); 91 skx_printk(KERN_CONT, "\n"); 92 93 return -ENODEV; 94 } 95 96 void __exit skx_adxl_put(void) 97 { 98 kfree(adxl_values); 99 kfree(adxl_msg); 100 } 101 102 static bool skx_adxl_decode(struct decoded_addr *res) 103 { 104 struct skx_dev *d; 105 int i, len = 0; 106 107 if (res->addr >= skx_tohm || (res->addr >= skx_tolm && 108 res->addr < BIT_ULL(32))) { 109 edac_dbg(0, "Address 0x%llx out of range\n", res->addr); 110 return false; 111 } 112 113 if (adxl_decode(res->addr, adxl_values)) { 114 edac_dbg(0, "Failed to decode 0x%llx\n", res->addr); 115 return false; 116 } 117 118 res->socket = (int)adxl_values[component_indices[INDEX_SOCKET]]; 119 res->imc = (int)adxl_values[component_indices[INDEX_MEMCTRL]]; 120 res->channel = (int)adxl_values[component_indices[INDEX_CHANNEL]]; 121 res->dimm = (int)adxl_values[component_indices[INDEX_DIMM]]; 122 123 if (res->imc > NUM_IMC - 1) { 124 skx_printk(KERN_ERR, "Bad imc %d\n", res->imc); 125 return false; 126 } 127 128 list_for_each_entry(d, &dev_edac_list, list) { 129 if (d->imc[0].src_id == res->socket) { 130 res->dev = d; 131 break; 132 } 133 } 134 135 if (!res->dev) { 136 skx_printk(KERN_ERR, "No device for src_id %d imc %d\n", 137 res->socket, res->imc); 138 return false; 139 } 140 141 for (i = 0; i < adxl_component_count; i++) { 142 if (adxl_values[i] == ~0x0ull) 143 continue; 144 145 len += snprintf(adxl_msg + len, MSG_SIZE - len, " %s:0x%llx", 146 adxl_component_names[i], adxl_values[i]); 147 if (MSG_SIZE - len <= 0) 148 break; 149 } 150 151 return true; 152 } 153 154 void skx_set_decode(skx_decode_f decode, skx_show_retry_log_f show_retry_log) 155 { 156 skx_decode = decode; 157 skx_show_retry_rd_err_log = show_retry_log; 158 } 159 160 int skx_get_src_id(struct skx_dev *d, int off, u8 *id) 161 { 162 u32 reg; 163 164 if (pci_read_config_dword(d->util_all, off, ®)) { 165 skx_printk(KERN_ERR, "Failed to read src id\n"); 166 return -ENODEV; 167 } 168 169 *id = GET_BITFIELD(reg, 12, 14); 170 return 0; 171 } 172 173 int skx_get_node_id(struct skx_dev *d, u8 *id) 174 { 175 u32 reg; 176 177 if (pci_read_config_dword(d->util_all, 0xf4, ®)) { 178 skx_printk(KERN_ERR, "Failed to read node id\n"); 179 return -ENODEV; 180 } 181 182 *id = GET_BITFIELD(reg, 0, 2); 183 return 0; 184 } 185 186 static int get_width(u32 mtr) 187 { 188 switch (GET_BITFIELD(mtr, 8, 9)) { 189 case 0: 190 return DEV_X4; 191 case 1: 192 return DEV_X8; 193 case 2: 194 return DEV_X16; 195 } 196 return DEV_UNKNOWN; 197 } 198 199 /* 200 * We use the per-socket device @did to count how many sockets are present, 201 * and to detemine which PCI buses are associated with each socket. Allocate 202 * and build the full list of all the skx_dev structures that we need here. 203 */ 204 int skx_get_all_bus_mappings(unsigned int did, int off, enum type type, 205 struct list_head **list) 206 { 207 struct pci_dev *pdev, *prev; 208 struct skx_dev *d; 209 u32 reg; 210 int ndev = 0; 211 212 prev = NULL; 213 for (;;) { 214 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, did, prev); 215 if (!pdev) 216 break; 217 ndev++; 218 d = kzalloc(sizeof(*d), GFP_KERNEL); 219 if (!d) { 220 pci_dev_put(pdev); 221 return -ENOMEM; 222 } 223 224 if (pci_read_config_dword(pdev, off, ®)) { 225 kfree(d); 226 pci_dev_put(pdev); 227 skx_printk(KERN_ERR, "Failed to read bus idx\n"); 228 return -ENODEV; 229 } 230 231 d->bus[0] = GET_BITFIELD(reg, 0, 7); 232 d->bus[1] = GET_BITFIELD(reg, 8, 15); 233 if (type == SKX) { 234 d->seg = pci_domain_nr(pdev->bus); 235 d->bus[2] = GET_BITFIELD(reg, 16, 23); 236 d->bus[3] = GET_BITFIELD(reg, 24, 31); 237 } else { 238 d->seg = GET_BITFIELD(reg, 16, 23); 239 } 240 241 edac_dbg(2, "busses: 0x%x, 0x%x, 0x%x, 0x%x\n", 242 d->bus[0], d->bus[1], d->bus[2], d->bus[3]); 243 list_add_tail(&d->list, &dev_edac_list); 244 prev = pdev; 245 } 246 247 if (list) 248 *list = &dev_edac_list; 249 return ndev; 250 } 251 252 int skx_get_hi_lo(unsigned int did, int off[], u64 *tolm, u64 *tohm) 253 { 254 struct pci_dev *pdev; 255 u32 reg; 256 257 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, did, NULL); 258 if (!pdev) { 259 skx_printk(KERN_ERR, "Can't get tolm/tohm\n"); 260 return -ENODEV; 261 } 262 263 if (pci_read_config_dword(pdev, off[0], ®)) { 264 skx_printk(KERN_ERR, "Failed to read tolm\n"); 265 goto fail; 266 } 267 skx_tolm = reg; 268 269 if (pci_read_config_dword(pdev, off[1], ®)) { 270 skx_printk(KERN_ERR, "Failed to read lower tohm\n"); 271 goto fail; 272 } 273 skx_tohm = reg; 274 275 if (pci_read_config_dword(pdev, off[2], ®)) { 276 skx_printk(KERN_ERR, "Failed to read upper tohm\n"); 277 goto fail; 278 } 279 skx_tohm |= (u64)reg << 32; 280 281 pci_dev_put(pdev); 282 *tolm = skx_tolm; 283 *tohm = skx_tohm; 284 edac_dbg(2, "tolm = 0x%llx tohm = 0x%llx\n", skx_tolm, skx_tohm); 285 return 0; 286 fail: 287 pci_dev_put(pdev); 288 return -ENODEV; 289 } 290 291 static int skx_get_dimm_attr(u32 reg, int lobit, int hibit, int add, 292 int minval, int maxval, const char *name) 293 { 294 u32 val = GET_BITFIELD(reg, lobit, hibit); 295 296 if (val < minval || val > maxval) { 297 edac_dbg(2, "bad %s = %d (raw=0x%x)\n", name, val, reg); 298 return -EINVAL; 299 } 300 return val + add; 301 } 302 303 #define numrank(reg) skx_get_dimm_attr(reg, 12, 13, 0, 0, 2, "ranks") 304 #define numrow(reg) skx_get_dimm_attr(reg, 2, 4, 12, 1, 6, "rows") 305 #define numcol(reg) skx_get_dimm_attr(reg, 0, 1, 10, 0, 2, "cols") 306 307 int skx_get_dimm_info(u32 mtr, u32 amap, struct dimm_info *dimm, 308 struct skx_imc *imc, int chan, int dimmno) 309 { 310 int banks = 16, ranks, rows, cols, npages; 311 u64 size; 312 313 ranks = numrank(mtr); 314 rows = numrow(mtr); 315 cols = numcol(mtr); 316 317 /* 318 * Compute size in 8-byte (2^3) words, then shift to MiB (2^20) 319 */ 320 size = ((1ull << (rows + cols + ranks)) * banks) >> (20 - 3); 321 npages = MiB_TO_PAGES(size); 322 323 edac_dbg(0, "mc#%d: channel %d, dimm %d, %lld MiB (%d pages) bank: %d, rank: %d, row: 0x%x, col: 0x%x\n", 324 imc->mc, chan, dimmno, size, npages, 325 banks, 1 << ranks, rows, cols); 326 327 imc->chan[chan].dimms[dimmno].close_pg = GET_BITFIELD(mtr, 0, 0); 328 imc->chan[chan].dimms[dimmno].bank_xor_enable = GET_BITFIELD(mtr, 9, 9); 329 imc->chan[chan].dimms[dimmno].fine_grain_bank = GET_BITFIELD(amap, 0, 0); 330 imc->chan[chan].dimms[dimmno].rowbits = rows; 331 imc->chan[chan].dimms[dimmno].colbits = cols; 332 333 dimm->nr_pages = npages; 334 dimm->grain = 32; 335 dimm->dtype = get_width(mtr); 336 dimm->mtype = MEM_DDR4; 337 dimm->edac_mode = EDAC_SECDED; /* likely better than this */ 338 snprintf(dimm->label, sizeof(dimm->label), "CPU_SrcID#%u_MC#%u_Chan#%u_DIMM#%u", 339 imc->src_id, imc->lmc, chan, dimmno); 340 341 return 1; 342 } 343 344 int skx_get_nvdimm_info(struct dimm_info *dimm, struct skx_imc *imc, 345 int chan, int dimmno, const char *mod_str) 346 { 347 int smbios_handle; 348 u32 dev_handle; 349 u16 flags; 350 u64 size = 0; 351 352 dev_handle = ACPI_NFIT_BUILD_DEVICE_HANDLE(dimmno, chan, imc->lmc, 353 imc->src_id, 0); 354 355 smbios_handle = nfit_get_smbios_id(dev_handle, &flags); 356 if (smbios_handle == -EOPNOTSUPP) { 357 pr_warn_once("%s: Can't find size of NVDIMM. Try enabling CONFIG_ACPI_NFIT\n", mod_str); 358 goto unknown_size; 359 } 360 361 if (smbios_handle < 0) { 362 skx_printk(KERN_ERR, "Can't find handle for NVDIMM ADR=0x%x\n", dev_handle); 363 goto unknown_size; 364 } 365 366 if (flags & ACPI_NFIT_MEM_MAP_FAILED) { 367 skx_printk(KERN_ERR, "NVDIMM ADR=0x%x is not mapped\n", dev_handle); 368 goto unknown_size; 369 } 370 371 size = dmi_memdev_size(smbios_handle); 372 if (size == ~0ull) 373 skx_printk(KERN_ERR, "Can't find size for NVDIMM ADR=0x%x/SMBIOS=0x%x\n", 374 dev_handle, smbios_handle); 375 376 unknown_size: 377 dimm->nr_pages = size >> PAGE_SHIFT; 378 dimm->grain = 32; 379 dimm->dtype = DEV_UNKNOWN; 380 dimm->mtype = MEM_NVDIMM; 381 dimm->edac_mode = EDAC_SECDED; /* likely better than this */ 382 383 edac_dbg(0, "mc#%d: channel %d, dimm %d, %llu MiB (%u pages)\n", 384 imc->mc, chan, dimmno, size >> 20, dimm->nr_pages); 385 386 snprintf(dimm->label, sizeof(dimm->label), "CPU_SrcID#%u_MC#%u_Chan#%u_DIMM#%u", 387 imc->src_id, imc->lmc, chan, dimmno); 388 389 return (size == 0 || size == ~0ull) ? 0 : 1; 390 } 391 392 int skx_register_mci(struct skx_imc *imc, struct pci_dev *pdev, 393 const char *ctl_name, const char *mod_str, 394 get_dimm_config_f get_dimm_config) 395 { 396 struct mem_ctl_info *mci; 397 struct edac_mc_layer layers[2]; 398 struct skx_pvt *pvt; 399 int rc; 400 401 /* Allocate a new MC control structure */ 402 layers[0].type = EDAC_MC_LAYER_CHANNEL; 403 layers[0].size = NUM_CHANNELS; 404 layers[0].is_virt_csrow = false; 405 layers[1].type = EDAC_MC_LAYER_SLOT; 406 layers[1].size = NUM_DIMMS; 407 layers[1].is_virt_csrow = true; 408 mci = edac_mc_alloc(imc->mc, ARRAY_SIZE(layers), layers, 409 sizeof(struct skx_pvt)); 410 411 if (unlikely(!mci)) 412 return -ENOMEM; 413 414 edac_dbg(0, "MC#%d: mci = %p\n", imc->mc, mci); 415 416 /* Associate skx_dev and mci for future usage */ 417 imc->mci = mci; 418 pvt = mci->pvt_info; 419 pvt->imc = imc; 420 421 mci->ctl_name = kasprintf(GFP_KERNEL, "%s#%d IMC#%d", ctl_name, 422 imc->node_id, imc->lmc); 423 if (!mci->ctl_name) { 424 rc = -ENOMEM; 425 goto fail0; 426 } 427 428 mci->mtype_cap = MEM_FLAG_DDR4 | MEM_FLAG_NVDIMM; 429 mci->edac_ctl_cap = EDAC_FLAG_NONE; 430 mci->edac_cap = EDAC_FLAG_NONE; 431 mci->mod_name = mod_str; 432 mci->dev_name = pci_name(pdev); 433 mci->ctl_page_to_phys = NULL; 434 435 rc = get_dimm_config(mci); 436 if (rc < 0) 437 goto fail; 438 439 /* Record ptr to the generic device */ 440 mci->pdev = &pdev->dev; 441 442 /* Add this new MC control structure to EDAC's list of MCs */ 443 if (unlikely(edac_mc_add_mc(mci))) { 444 edac_dbg(0, "MC: failed edac_mc_add_mc()\n"); 445 rc = -EINVAL; 446 goto fail; 447 } 448 449 return 0; 450 451 fail: 452 kfree(mci->ctl_name); 453 fail0: 454 edac_mc_free(mci); 455 imc->mci = NULL; 456 return rc; 457 } 458 459 static void skx_unregister_mci(struct skx_imc *imc) 460 { 461 struct mem_ctl_info *mci = imc->mci; 462 463 if (!mci) 464 return; 465 466 edac_dbg(0, "MC%d: mci = %p\n", imc->mc, mci); 467 468 /* Remove MC sysfs nodes */ 469 edac_mc_del_mc(mci->pdev); 470 471 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name); 472 kfree(mci->ctl_name); 473 edac_mc_free(mci); 474 } 475 476 static void skx_mce_output_error(struct mem_ctl_info *mci, 477 const struct mce *m, 478 struct decoded_addr *res) 479 { 480 enum hw_event_mc_err_type tp_event; 481 char *optype; 482 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0); 483 bool overflow = GET_BITFIELD(m->status, 62, 62); 484 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61); 485 bool recoverable; 486 int len; 487 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52); 488 u32 mscod = GET_BITFIELD(m->status, 16, 31); 489 u32 errcode = GET_BITFIELD(m->status, 0, 15); 490 u32 optypenum = GET_BITFIELD(m->status, 4, 6); 491 492 recoverable = GET_BITFIELD(m->status, 56, 56); 493 494 if (uncorrected_error) { 495 core_err_cnt = 1; 496 if (ripv) { 497 tp_event = HW_EVENT_ERR_FATAL; 498 } else { 499 tp_event = HW_EVENT_ERR_UNCORRECTED; 500 } 501 } else { 502 tp_event = HW_EVENT_ERR_CORRECTED; 503 } 504 505 /* 506 * According to Intel Architecture spec vol 3B, 507 * Table 15-10 "IA32_MCi_Status [15:0] Compound Error Code Encoding" 508 * memory errors should fit one of these masks: 509 * 000f 0000 1mmm cccc (binary) 510 * 000f 0010 1mmm cccc (binary) [RAM used as cache] 511 * where: 512 * f = Correction Report Filtering Bit. If 1, subsequent errors 513 * won't be shown 514 * mmm = error type 515 * cccc = channel 516 * If the mask doesn't match, report an error to the parsing logic 517 */ 518 if (!((errcode & 0xef80) == 0x80 || (errcode & 0xef80) == 0x280)) { 519 optype = "Can't parse: it is not a mem"; 520 } else { 521 switch (optypenum) { 522 case 0: 523 optype = "generic undef request error"; 524 break; 525 case 1: 526 optype = "memory read error"; 527 break; 528 case 2: 529 optype = "memory write error"; 530 break; 531 case 3: 532 optype = "addr/cmd error"; 533 break; 534 case 4: 535 optype = "memory scrubbing error"; 536 break; 537 default: 538 optype = "reserved"; 539 break; 540 } 541 } 542 if (adxl_component_count) { 543 len = snprintf(skx_msg, MSG_SIZE, "%s%s err_code:0x%04x:0x%04x %s", 544 overflow ? " OVERFLOW" : "", 545 (uncorrected_error && recoverable) ? " recoverable" : "", 546 mscod, errcode, adxl_msg); 547 } else { 548 len = snprintf(skx_msg, MSG_SIZE, 549 "%s%s err_code:0x%04x:0x%04x socket:%d imc:%d rank:%d bg:%d ba:%d row:0x%x col:0x%x", 550 overflow ? " OVERFLOW" : "", 551 (uncorrected_error && recoverable) ? " recoverable" : "", 552 mscod, errcode, 553 res->socket, res->imc, res->rank, 554 res->bank_group, res->bank_address, res->row, res->column); 555 } 556 557 if (skx_show_retry_rd_err_log) 558 skx_show_retry_rd_err_log(res, skx_msg + len, MSG_SIZE - len); 559 560 edac_dbg(0, "%s\n", skx_msg); 561 562 /* Call the helper to output message */ 563 edac_mc_handle_error(tp_event, mci, core_err_cnt, 564 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0, 565 res->channel, res->dimm, -1, 566 optype, skx_msg); 567 } 568 569 int skx_mce_check_error(struct notifier_block *nb, unsigned long val, 570 void *data) 571 { 572 struct mce *mce = (struct mce *)data; 573 struct decoded_addr res; 574 struct mem_ctl_info *mci; 575 char *type; 576 577 if (edac_get_report_status() == EDAC_REPORTING_DISABLED) 578 return NOTIFY_DONE; 579 580 /* ignore unless this is memory related with an address */ 581 if ((mce->status & 0xefff) >> 7 != 1 || !(mce->status & MCI_STATUS_ADDRV)) 582 return NOTIFY_DONE; 583 584 memset(&res, 0, sizeof(res)); 585 res.addr = mce->addr; 586 587 if (adxl_component_count) { 588 if (!skx_adxl_decode(&res)) 589 return NOTIFY_DONE; 590 } else if (!skx_decode || !skx_decode(&res)) { 591 return NOTIFY_DONE; 592 } 593 594 mci = res.dev->imc[res.imc].mci; 595 596 if (!mci) 597 return NOTIFY_DONE; 598 599 if (mce->mcgstatus & MCG_STATUS_MCIP) 600 type = "Exception"; 601 else 602 type = "Event"; 603 604 skx_mc_printk(mci, KERN_DEBUG, "HANDLING MCE MEMORY ERROR\n"); 605 606 skx_mc_printk(mci, KERN_DEBUG, "CPU %d: Machine Check %s: 0x%llx " 607 "Bank %d: 0x%llx\n", mce->extcpu, type, 608 mce->mcgstatus, mce->bank, mce->status); 609 skx_mc_printk(mci, KERN_DEBUG, "TSC 0x%llx ", mce->tsc); 610 skx_mc_printk(mci, KERN_DEBUG, "ADDR 0x%llx ", mce->addr); 611 skx_mc_printk(mci, KERN_DEBUG, "MISC 0x%llx ", mce->misc); 612 613 skx_mc_printk(mci, KERN_DEBUG, "PROCESSOR %u:0x%x TIME %llu SOCKET " 614 "%u APIC 0x%x\n", mce->cpuvendor, mce->cpuid, 615 mce->time, mce->socketid, mce->apicid); 616 617 skx_mce_output_error(mci, mce, &res); 618 619 return NOTIFY_DONE; 620 } 621 622 void skx_remove(void) 623 { 624 int i, j; 625 struct skx_dev *d, *tmp; 626 627 edac_dbg(0, "\n"); 628 629 list_for_each_entry_safe(d, tmp, &dev_edac_list, list) { 630 list_del(&d->list); 631 for (i = 0; i < NUM_IMC; i++) { 632 if (d->imc[i].mci) 633 skx_unregister_mci(&d->imc[i]); 634 635 if (d->imc[i].mdev) 636 pci_dev_put(d->imc[i].mdev); 637 638 if (d->imc[i].mbase) 639 iounmap(d->imc[i].mbase); 640 641 for (j = 0; j < NUM_CHANNELS; j++) { 642 if (d->imc[i].chan[j].cdev) 643 pci_dev_put(d->imc[i].chan[j].cdev); 644 } 645 } 646 if (d->util_all) 647 pci_dev_put(d->util_all); 648 if (d->sad_all) 649 pci_dev_put(d->sad_all); 650 if (d->uracu) 651 pci_dev_put(d->uracu); 652 653 kfree(d); 654 } 655 } 656