1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * EDAC driver for Intel(R) Xeon(R) Skylake processors 4 * Copyright (c) 2016, Intel Corporation. 5 */ 6 7 #include <linux/kernel.h> 8 #include <linux/processor.h> 9 #include <asm/cpu_device_id.h> 10 #include <asm/intel-family.h> 11 #include <asm/mce.h> 12 13 #include "edac_module.h" 14 #include "skx_common.h" 15 16 #define EDAC_MOD_STR "skx_edac" 17 18 /* 19 * Debug macros 20 */ 21 #define skx_printk(level, fmt, arg...) \ 22 edac_printk(level, "skx", fmt, ##arg) 23 24 #define skx_mc_printk(mci, level, fmt, arg...) \ 25 edac_mc_chipset_printk(mci, level, "skx", fmt, ##arg) 26 27 static struct list_head *skx_edac_list; 28 29 static u64 skx_tolm, skx_tohm; 30 static int skx_num_sockets; 31 static unsigned int nvdimm_count; 32 33 #define MASK26 0x3FFFFFF /* Mask for 2^26 */ 34 #define MASK29 0x1FFFFFFF /* Mask for 2^29 */ 35 36 static struct skx_dev *get_skx_dev(struct pci_bus *bus, u8 idx) 37 { 38 struct skx_dev *d; 39 40 list_for_each_entry(d, skx_edac_list, list) { 41 if (d->seg == pci_domain_nr(bus) && d->bus[idx] == bus->number) 42 return d; 43 } 44 45 return NULL; 46 } 47 48 enum munittype { 49 CHAN0, CHAN1, CHAN2, SAD_ALL, UTIL_ALL, SAD, 50 ERRCHAN0, ERRCHAN1, ERRCHAN2, 51 }; 52 53 struct munit { 54 u16 did; 55 u16 devfn[SKX_NUM_IMC]; 56 u8 busidx; 57 u8 per_socket; 58 enum munittype mtype; 59 }; 60 61 /* 62 * List of PCI device ids that we need together with some device 63 * number and function numbers to tell which memory controller the 64 * device belongs to. 65 */ 66 static const struct munit skx_all_munits[] = { 67 { 0x2054, { }, 1, 1, SAD_ALL }, 68 { 0x2055, { }, 1, 1, UTIL_ALL }, 69 { 0x2040, { PCI_DEVFN(10, 0), PCI_DEVFN(12, 0) }, 2, 2, CHAN0 }, 70 { 0x2044, { PCI_DEVFN(10, 4), PCI_DEVFN(12, 4) }, 2, 2, CHAN1 }, 71 { 0x2048, { PCI_DEVFN(11, 0), PCI_DEVFN(13, 0) }, 2, 2, CHAN2 }, 72 { 0x2043, { PCI_DEVFN(10, 3), PCI_DEVFN(12, 3) }, 2, 2, ERRCHAN0 }, 73 { 0x2047, { PCI_DEVFN(10, 7), PCI_DEVFN(12, 7) }, 2, 2, ERRCHAN1 }, 74 { 0x204b, { PCI_DEVFN(11, 3), PCI_DEVFN(13, 3) }, 2, 2, ERRCHAN2 }, 75 { 0x208e, { }, 1, 0, SAD }, 76 { } 77 }; 78 79 static int get_all_munits(const struct munit *m) 80 { 81 struct pci_dev *pdev, *prev; 82 struct skx_dev *d; 83 u32 reg; 84 int i = 0, ndev = 0; 85 86 prev = NULL; 87 for (;;) { 88 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, m->did, prev); 89 if (!pdev) 90 break; 91 ndev++; 92 if (m->per_socket == SKX_NUM_IMC) { 93 for (i = 0; i < SKX_NUM_IMC; i++) 94 if (m->devfn[i] == pdev->devfn) 95 break; 96 if (i == SKX_NUM_IMC) 97 goto fail; 98 } 99 d = get_skx_dev(pdev->bus, m->busidx); 100 if (!d) 101 goto fail; 102 103 /* Be sure that the device is enabled */ 104 if (unlikely(pci_enable_device(pdev) < 0)) { 105 skx_printk(KERN_ERR, "Couldn't enable device %04x:%04x\n", 106 PCI_VENDOR_ID_INTEL, m->did); 107 goto fail; 108 } 109 110 switch (m->mtype) { 111 case CHAN0: 112 case CHAN1: 113 case CHAN2: 114 pci_dev_get(pdev); 115 d->imc[i].chan[m->mtype].cdev = pdev; 116 break; 117 case ERRCHAN0: 118 case ERRCHAN1: 119 case ERRCHAN2: 120 pci_dev_get(pdev); 121 d->imc[i].chan[m->mtype - ERRCHAN0].edev = pdev; 122 break; 123 case SAD_ALL: 124 pci_dev_get(pdev); 125 d->sad_all = pdev; 126 break; 127 case UTIL_ALL: 128 pci_dev_get(pdev); 129 d->util_all = pdev; 130 break; 131 case SAD: 132 /* 133 * one of these devices per core, including cores 134 * that don't exist on this SKU. Ignore any that 135 * read a route table of zero, make sure all the 136 * non-zero values match. 137 */ 138 pci_read_config_dword(pdev, 0xB4, ®); 139 if (reg != 0) { 140 if (d->mcroute == 0) { 141 d->mcroute = reg; 142 } else if (d->mcroute != reg) { 143 skx_printk(KERN_ERR, "mcroute mismatch\n"); 144 goto fail; 145 } 146 } 147 ndev--; 148 break; 149 } 150 151 prev = pdev; 152 } 153 154 return ndev; 155 fail: 156 pci_dev_put(pdev); 157 return -ENODEV; 158 } 159 160 static struct res_config skx_cfg = { 161 .type = SKX, 162 .decs_did = 0x2016, 163 .busno_cfg_offset = 0xcc, 164 }; 165 166 static const struct x86_cpu_id skx_cpuids[] = { 167 X86_MATCH_VFM_STEPPINGS(INTEL_SKYLAKE_X, X86_STEPPINGS(0x0, 0xf), &skx_cfg), 168 { } 169 }; 170 MODULE_DEVICE_TABLE(x86cpu, skx_cpuids); 171 172 static bool skx_check_ecc(u32 mcmtr) 173 { 174 return !!GET_BITFIELD(mcmtr, 2, 2); 175 } 176 177 static int skx_get_dimm_config(struct mem_ctl_info *mci, struct res_config *cfg) 178 { 179 struct skx_pvt *pvt = mci->pvt_info; 180 u32 mtr, mcmtr, amap, mcddrtcfg; 181 struct skx_imc *imc = pvt->imc; 182 struct dimm_info *dimm; 183 int i, j; 184 int ndimms; 185 186 /* Only the mcmtr on the first channel is effective */ 187 pci_read_config_dword(imc->chan[0].cdev, 0x87c, &mcmtr); 188 189 for (i = 0; i < SKX_NUM_CHANNELS; i++) { 190 ndimms = 0; 191 pci_read_config_dword(imc->chan[i].cdev, 0x8C, &amap); 192 pci_read_config_dword(imc->chan[i].cdev, 0x400, &mcddrtcfg); 193 for (j = 0; j < SKX_NUM_DIMMS; j++) { 194 dimm = edac_get_dimm(mci, i, j, 0); 195 pci_read_config_dword(imc->chan[i].cdev, 196 0x80 + 4 * j, &mtr); 197 if (IS_DIMM_PRESENT(mtr)) { 198 ndimms += skx_get_dimm_info(mtr, mcmtr, amap, dimm, imc, i, j, cfg); 199 } else if (IS_NVDIMM_PRESENT(mcddrtcfg, j)) { 200 ndimms += skx_get_nvdimm_info(dimm, imc, i, j, 201 EDAC_MOD_STR); 202 nvdimm_count++; 203 } 204 } 205 if (ndimms && !skx_check_ecc(mcmtr)) { 206 skx_printk(KERN_ERR, "ECC is disabled on imc %d\n", imc->mc); 207 return -ENODEV; 208 } 209 } 210 211 return 0; 212 } 213 214 #define SKX_MAX_SAD 24 215 216 #define SKX_GET_SAD(d, i, reg) \ 217 pci_read_config_dword((d)->sad_all, 0x60 + 8 * (i), &(reg)) 218 #define SKX_GET_ILV(d, i, reg) \ 219 pci_read_config_dword((d)->sad_all, 0x64 + 8 * (i), &(reg)) 220 221 #define SKX_SAD_MOD3MODE(sad) GET_BITFIELD((sad), 30, 31) 222 #define SKX_SAD_MOD3(sad) GET_BITFIELD((sad), 27, 27) 223 #define SKX_SAD_LIMIT(sad) (((u64)GET_BITFIELD((sad), 7, 26) << 26) | MASK26) 224 #define SKX_SAD_MOD3ASMOD2(sad) GET_BITFIELD((sad), 5, 6) 225 #define SKX_SAD_ATTR(sad) GET_BITFIELD((sad), 3, 4) 226 #define SKX_SAD_INTERLEAVE(sad) GET_BITFIELD((sad), 1, 2) 227 #define SKX_SAD_ENABLE(sad) GET_BITFIELD((sad), 0, 0) 228 229 #define SKX_ILV_REMOTE(tgt) (((tgt) & 8) == 0) 230 #define SKX_ILV_TARGET(tgt) ((tgt) & 7) 231 232 static void skx_show_retry_rd_err_log(struct decoded_addr *res, 233 char *msg, int len, 234 bool scrub_err) 235 { 236 u32 log0, log1, log2, log3, log4; 237 u32 corr0, corr1, corr2, corr3; 238 struct pci_dev *edev; 239 int n; 240 241 edev = res->dev->imc[res->imc].chan[res->channel].edev; 242 243 pci_read_config_dword(edev, 0x154, &log0); 244 pci_read_config_dword(edev, 0x148, &log1); 245 pci_read_config_dword(edev, 0x150, &log2); 246 pci_read_config_dword(edev, 0x15c, &log3); 247 pci_read_config_dword(edev, 0x114, &log4); 248 249 n = snprintf(msg, len, " retry_rd_err_log[%.8x %.8x %.8x %.8x %.8x]", 250 log0, log1, log2, log3, log4); 251 252 pci_read_config_dword(edev, 0x104, &corr0); 253 pci_read_config_dword(edev, 0x108, &corr1); 254 pci_read_config_dword(edev, 0x10c, &corr2); 255 pci_read_config_dword(edev, 0x110, &corr3); 256 257 if (len - n > 0) 258 snprintf(msg + n, len - n, 259 " correrrcnt[%.4x %.4x %.4x %.4x %.4x %.4x %.4x %.4x]", 260 corr0 & 0xffff, corr0 >> 16, 261 corr1 & 0xffff, corr1 >> 16, 262 corr2 & 0xffff, corr2 >> 16, 263 corr3 & 0xffff, corr3 >> 16); 264 } 265 266 static bool skx_sad_decode(struct decoded_addr *res) 267 { 268 struct skx_dev *d = list_first_entry(skx_edac_list, typeof(*d), list); 269 u64 addr = res->addr; 270 int i, idx, tgt, lchan, shift; 271 u32 sad, ilv; 272 u64 limit, prev_limit; 273 int remote = 0; 274 275 /* Simple sanity check for I/O space or out of range */ 276 if (addr >= skx_tohm || (addr >= skx_tolm && addr < BIT_ULL(32))) { 277 edac_dbg(0, "Address 0x%llx out of range\n", addr); 278 return false; 279 } 280 281 restart: 282 prev_limit = 0; 283 for (i = 0; i < SKX_MAX_SAD; i++) { 284 SKX_GET_SAD(d, i, sad); 285 limit = SKX_SAD_LIMIT(sad); 286 if (SKX_SAD_ENABLE(sad)) { 287 if (addr >= prev_limit && addr <= limit) 288 goto sad_found; 289 } 290 prev_limit = limit + 1; 291 } 292 edac_dbg(0, "No SAD entry for 0x%llx\n", addr); 293 return false; 294 295 sad_found: 296 SKX_GET_ILV(d, i, ilv); 297 298 switch (SKX_SAD_INTERLEAVE(sad)) { 299 case 0: 300 idx = GET_BITFIELD(addr, 6, 8); 301 break; 302 case 1: 303 idx = GET_BITFIELD(addr, 8, 10); 304 break; 305 case 2: 306 idx = GET_BITFIELD(addr, 12, 14); 307 break; 308 case 3: 309 idx = GET_BITFIELD(addr, 30, 32); 310 break; 311 } 312 313 tgt = GET_BITFIELD(ilv, 4 * idx, 4 * idx + 3); 314 315 /* If point to another node, find it and start over */ 316 if (SKX_ILV_REMOTE(tgt)) { 317 if (remote) { 318 edac_dbg(0, "Double remote!\n"); 319 return false; 320 } 321 remote = 1; 322 list_for_each_entry(d, skx_edac_list, list) { 323 if (d->imc[0].src_id == SKX_ILV_TARGET(tgt)) 324 goto restart; 325 } 326 edac_dbg(0, "Can't find node %d\n", SKX_ILV_TARGET(tgt)); 327 return false; 328 } 329 330 if (SKX_SAD_MOD3(sad) == 0) { 331 lchan = SKX_ILV_TARGET(tgt); 332 } else { 333 switch (SKX_SAD_MOD3MODE(sad)) { 334 case 0: 335 shift = 6; 336 break; 337 case 1: 338 shift = 8; 339 break; 340 case 2: 341 shift = 12; 342 break; 343 default: 344 edac_dbg(0, "illegal mod3mode\n"); 345 return false; 346 } 347 switch (SKX_SAD_MOD3ASMOD2(sad)) { 348 case 0: 349 lchan = (addr >> shift) % 3; 350 break; 351 case 1: 352 lchan = (addr >> shift) % 2; 353 break; 354 case 2: 355 lchan = (addr >> shift) % 2; 356 lchan = (lchan << 1) | !lchan; 357 break; 358 case 3: 359 lchan = ((addr >> shift) % 2) << 1; 360 break; 361 } 362 lchan = (lchan << 1) | (SKX_ILV_TARGET(tgt) & 1); 363 } 364 365 res->dev = d; 366 res->socket = d->imc[0].src_id; 367 res->imc = GET_BITFIELD(d->mcroute, lchan * 3, lchan * 3 + 2); 368 res->channel = GET_BITFIELD(d->mcroute, lchan * 2 + 18, lchan * 2 + 19); 369 370 edac_dbg(2, "0x%llx: socket=%d imc=%d channel=%d\n", 371 res->addr, res->socket, res->imc, res->channel); 372 return true; 373 } 374 375 #define SKX_MAX_TAD 8 376 377 #define SKX_GET_TADBASE(d, mc, i, reg) \ 378 pci_read_config_dword((d)->imc[mc].chan[0].cdev, 0x850 + 4 * (i), &(reg)) 379 #define SKX_GET_TADWAYNESS(d, mc, i, reg) \ 380 pci_read_config_dword((d)->imc[mc].chan[0].cdev, 0x880 + 4 * (i), &(reg)) 381 #define SKX_GET_TADCHNILVOFFSET(d, mc, ch, i, reg) \ 382 pci_read_config_dword((d)->imc[mc].chan[ch].cdev, 0x90 + 4 * (i), &(reg)) 383 384 #define SKX_TAD_BASE(b) ((u64)GET_BITFIELD((b), 12, 31) << 26) 385 #define SKX_TAD_SKT_GRAN(b) GET_BITFIELD((b), 4, 5) 386 #define SKX_TAD_CHN_GRAN(b) GET_BITFIELD((b), 6, 7) 387 #define SKX_TAD_LIMIT(b) (((u64)GET_BITFIELD((b), 12, 31) << 26) | MASK26) 388 #define SKX_TAD_OFFSET(b) ((u64)GET_BITFIELD((b), 4, 23) << 26) 389 #define SKX_TAD_SKTWAYS(b) (1 << GET_BITFIELD((b), 10, 11)) 390 #define SKX_TAD_CHNWAYS(b) (GET_BITFIELD((b), 8, 9) + 1) 391 392 /* which bit used for both socket and channel interleave */ 393 static int skx_granularity[] = { 6, 8, 12, 30 }; 394 395 static u64 skx_do_interleave(u64 addr, int shift, int ways, u64 lowbits) 396 { 397 addr >>= shift; 398 addr /= ways; 399 addr <<= shift; 400 401 return addr | (lowbits & ((1ull << shift) - 1)); 402 } 403 404 static bool skx_tad_decode(struct decoded_addr *res) 405 { 406 int i; 407 u32 base, wayness, chnilvoffset; 408 int skt_interleave_bit, chn_interleave_bit; 409 u64 channel_addr; 410 411 for (i = 0; i < SKX_MAX_TAD; i++) { 412 SKX_GET_TADBASE(res->dev, res->imc, i, base); 413 SKX_GET_TADWAYNESS(res->dev, res->imc, i, wayness); 414 if (SKX_TAD_BASE(base) <= res->addr && res->addr <= SKX_TAD_LIMIT(wayness)) 415 goto tad_found; 416 } 417 edac_dbg(0, "No TAD entry for 0x%llx\n", res->addr); 418 return false; 419 420 tad_found: 421 res->sktways = SKX_TAD_SKTWAYS(wayness); 422 res->chanways = SKX_TAD_CHNWAYS(wayness); 423 skt_interleave_bit = skx_granularity[SKX_TAD_SKT_GRAN(base)]; 424 chn_interleave_bit = skx_granularity[SKX_TAD_CHN_GRAN(base)]; 425 426 SKX_GET_TADCHNILVOFFSET(res->dev, res->imc, res->channel, i, chnilvoffset); 427 channel_addr = res->addr - SKX_TAD_OFFSET(chnilvoffset); 428 429 if (res->chanways == 3 && skt_interleave_bit > chn_interleave_bit) { 430 /* Must handle channel first, then socket */ 431 channel_addr = skx_do_interleave(channel_addr, chn_interleave_bit, 432 res->chanways, channel_addr); 433 channel_addr = skx_do_interleave(channel_addr, skt_interleave_bit, 434 res->sktways, channel_addr); 435 } else { 436 /* Handle socket then channel. Preserve low bits from original address */ 437 channel_addr = skx_do_interleave(channel_addr, skt_interleave_bit, 438 res->sktways, res->addr); 439 channel_addr = skx_do_interleave(channel_addr, chn_interleave_bit, 440 res->chanways, res->addr); 441 } 442 443 res->chan_addr = channel_addr; 444 445 edac_dbg(2, "0x%llx: chan_addr=0x%llx sktways=%d chanways=%d\n", 446 res->addr, res->chan_addr, res->sktways, res->chanways); 447 return true; 448 } 449 450 #define SKX_MAX_RIR 4 451 452 #define SKX_GET_RIRWAYNESS(d, mc, ch, i, reg) \ 453 pci_read_config_dword((d)->imc[mc].chan[ch].cdev, \ 454 0x108 + 4 * (i), &(reg)) 455 #define SKX_GET_RIRILV(d, mc, ch, idx, i, reg) \ 456 pci_read_config_dword((d)->imc[mc].chan[ch].cdev, \ 457 0x120 + 16 * (idx) + 4 * (i), &(reg)) 458 459 #define SKX_RIR_VALID(b) GET_BITFIELD((b), 31, 31) 460 #define SKX_RIR_LIMIT(b) (((u64)GET_BITFIELD((b), 1, 11) << 29) | MASK29) 461 #define SKX_RIR_WAYS(b) (1 << GET_BITFIELD((b), 28, 29)) 462 #define SKX_RIR_CHAN_RANK(b) GET_BITFIELD((b), 16, 19) 463 #define SKX_RIR_OFFSET(b) ((u64)(GET_BITFIELD((b), 2, 15) << 26)) 464 465 static bool skx_rir_decode(struct decoded_addr *res) 466 { 467 int i, idx, chan_rank; 468 int shift; 469 u32 rirway, rirlv; 470 u64 rank_addr, prev_limit = 0, limit; 471 472 if (res->dev->imc[res->imc].chan[res->channel].dimms[0].close_pg) 473 shift = 6; 474 else 475 shift = 13; 476 477 for (i = 0; i < SKX_MAX_RIR; i++) { 478 SKX_GET_RIRWAYNESS(res->dev, res->imc, res->channel, i, rirway); 479 limit = SKX_RIR_LIMIT(rirway); 480 if (SKX_RIR_VALID(rirway)) { 481 if (prev_limit <= res->chan_addr && 482 res->chan_addr <= limit) 483 goto rir_found; 484 } 485 prev_limit = limit; 486 } 487 edac_dbg(0, "No RIR entry for 0x%llx\n", res->addr); 488 return false; 489 490 rir_found: 491 rank_addr = res->chan_addr >> shift; 492 rank_addr /= SKX_RIR_WAYS(rirway); 493 rank_addr <<= shift; 494 rank_addr |= res->chan_addr & GENMASK_ULL(shift - 1, 0); 495 496 res->rank_address = rank_addr; 497 idx = (res->chan_addr >> shift) % SKX_RIR_WAYS(rirway); 498 499 SKX_GET_RIRILV(res->dev, res->imc, res->channel, idx, i, rirlv); 500 res->rank_address = rank_addr - SKX_RIR_OFFSET(rirlv); 501 chan_rank = SKX_RIR_CHAN_RANK(rirlv); 502 res->channel_rank = chan_rank; 503 res->dimm = chan_rank / 4; 504 res->rank = chan_rank % 4; 505 506 edac_dbg(2, "0x%llx: dimm=%d rank=%d chan_rank=%d rank_addr=0x%llx\n", 507 res->addr, res->dimm, res->rank, 508 res->channel_rank, res->rank_address); 509 return true; 510 } 511 512 static u8 skx_close_row[] = { 513 15, 16, 17, 18, 20, 21, 22, 28, 10, 11, 12, 13, 29, 30, 31, 32, 33, 34 514 }; 515 516 static u8 skx_close_column[] = { 517 3, 4, 5, 14, 19, 23, 24, 25, 26, 27 518 }; 519 520 static u8 skx_open_row[] = { 521 14, 15, 16, 20, 28, 21, 22, 23, 24, 25, 26, 27, 29, 30, 31, 32, 33, 34 522 }; 523 524 static u8 skx_open_column[] = { 525 3, 4, 5, 6, 7, 8, 9, 10, 11, 12 526 }; 527 528 static u8 skx_open_fine_column[] = { 529 3, 4, 5, 7, 8, 9, 10, 11, 12, 13 530 }; 531 532 static int skx_bits(u64 addr, int nbits, u8 *bits) 533 { 534 int i, res = 0; 535 536 for (i = 0; i < nbits; i++) 537 res |= ((addr >> bits[i]) & 1) << i; 538 return res; 539 } 540 541 static int skx_bank_bits(u64 addr, int b0, int b1, int do_xor, int x0, int x1) 542 { 543 int ret = GET_BITFIELD(addr, b0, b0) | (GET_BITFIELD(addr, b1, b1) << 1); 544 545 if (do_xor) 546 ret ^= GET_BITFIELD(addr, x0, x0) | (GET_BITFIELD(addr, x1, x1) << 1); 547 548 return ret; 549 } 550 551 static bool skx_mad_decode(struct decoded_addr *r) 552 { 553 struct skx_dimm *dimm = &r->dev->imc[r->imc].chan[r->channel].dimms[r->dimm]; 554 int bg0 = dimm->fine_grain_bank ? 6 : 13; 555 556 if (dimm->close_pg) { 557 r->row = skx_bits(r->rank_address, dimm->rowbits, skx_close_row); 558 r->column = skx_bits(r->rank_address, dimm->colbits, skx_close_column); 559 r->column |= 0x400; /* C10 is autoprecharge, always set */ 560 r->bank_address = skx_bank_bits(r->rank_address, 8, 9, dimm->bank_xor_enable, 22, 28); 561 r->bank_group = skx_bank_bits(r->rank_address, 6, 7, dimm->bank_xor_enable, 20, 21); 562 } else { 563 r->row = skx_bits(r->rank_address, dimm->rowbits, skx_open_row); 564 if (dimm->fine_grain_bank) 565 r->column = skx_bits(r->rank_address, dimm->colbits, skx_open_fine_column); 566 else 567 r->column = skx_bits(r->rank_address, dimm->colbits, skx_open_column); 568 r->bank_address = skx_bank_bits(r->rank_address, 18, 19, dimm->bank_xor_enable, 22, 23); 569 r->bank_group = skx_bank_bits(r->rank_address, bg0, 17, dimm->bank_xor_enable, 20, 21); 570 } 571 r->row &= (1u << dimm->rowbits) - 1; 572 573 edac_dbg(2, "0x%llx: row=0x%x col=0x%x bank_addr=%d bank_group=%d\n", 574 r->addr, r->row, r->column, r->bank_address, 575 r->bank_group); 576 return true; 577 } 578 579 static bool skx_decode(struct decoded_addr *res) 580 { 581 return skx_sad_decode(res) && skx_tad_decode(res) && 582 skx_rir_decode(res) && skx_mad_decode(res); 583 } 584 585 static struct notifier_block skx_mce_dec = { 586 .notifier_call = skx_mce_check_error, 587 .priority = MCE_PRIO_EDAC, 588 }; 589 590 #ifdef CONFIG_EDAC_DEBUG 591 /* 592 * Debug feature. 593 * Exercise the address decode logic by writing an address to 594 * /sys/kernel/debug/edac/skx_test/addr. 595 */ 596 static struct dentry *skx_test; 597 598 static int debugfs_u64_set(void *data, u64 val) 599 { 600 struct mce m; 601 602 pr_warn_once("Fake error to 0x%llx injected via debugfs\n", val); 603 604 memset(&m, 0, sizeof(m)); 605 /* ADDRV + MemRd + Unknown channel */ 606 m.status = MCI_STATUS_ADDRV + 0x90; 607 /* One corrected error */ 608 m.status |= BIT_ULL(MCI_STATUS_CEC_SHIFT); 609 m.addr = val; 610 skx_mce_check_error(NULL, 0, &m); 611 612 return 0; 613 } 614 DEFINE_SIMPLE_ATTRIBUTE(fops_u64_wo, NULL, debugfs_u64_set, "%llu\n"); 615 616 static void setup_skx_debug(void) 617 { 618 skx_test = edac_debugfs_create_dir("skx_test"); 619 if (!skx_test) 620 return; 621 622 if (!edac_debugfs_create_file("addr", 0200, skx_test, 623 NULL, &fops_u64_wo)) { 624 debugfs_remove(skx_test); 625 skx_test = NULL; 626 } 627 } 628 629 static void teardown_skx_debug(void) 630 { 631 debugfs_remove_recursive(skx_test); 632 } 633 #else 634 static inline void setup_skx_debug(void) {} 635 static inline void teardown_skx_debug(void) {} 636 #endif /*CONFIG_EDAC_DEBUG*/ 637 638 /* 639 * skx_init: 640 * make sure we are running on the correct cpu model 641 * search for all the devices we need 642 * check which DIMMs are present. 643 */ 644 static int __init skx_init(void) 645 { 646 const struct x86_cpu_id *id; 647 struct res_config *cfg; 648 const struct munit *m; 649 const char *owner; 650 int rc = 0, i, off[3] = {0xd0, 0xd4, 0xd8}; 651 u8 mc = 0, src_id, node_id; 652 struct skx_dev *d; 653 654 edac_dbg(2, "\n"); 655 656 if (ghes_get_devices()) 657 return -EBUSY; 658 659 owner = edac_get_owner(); 660 if (owner && strncmp(owner, EDAC_MOD_STR, sizeof(EDAC_MOD_STR))) 661 return -EBUSY; 662 663 if (cpu_feature_enabled(X86_FEATURE_HYPERVISOR)) 664 return -ENODEV; 665 666 id = x86_match_cpu(skx_cpuids); 667 if (!id) 668 return -ENODEV; 669 670 cfg = (struct res_config *)id->driver_data; 671 672 rc = skx_get_hi_lo(0x2034, off, &skx_tolm, &skx_tohm); 673 if (rc) 674 return rc; 675 676 rc = skx_get_all_bus_mappings(cfg, &skx_edac_list); 677 if (rc < 0) 678 goto fail; 679 if (rc == 0) { 680 edac_dbg(2, "No memory controllers found\n"); 681 return -ENODEV; 682 } 683 skx_num_sockets = rc; 684 685 for (m = skx_all_munits; m->did; m++) { 686 rc = get_all_munits(m); 687 if (rc < 0) 688 goto fail; 689 if (rc != m->per_socket * skx_num_sockets) { 690 edac_dbg(2, "Expected %d, got %d of 0x%x\n", 691 m->per_socket * skx_num_sockets, rc, m->did); 692 rc = -ENODEV; 693 goto fail; 694 } 695 } 696 697 list_for_each_entry(d, skx_edac_list, list) { 698 rc = skx_get_src_id(d, 0xf0, &src_id); 699 if (rc < 0) 700 goto fail; 701 rc = skx_get_node_id(d, &node_id); 702 if (rc < 0) 703 goto fail; 704 edac_dbg(2, "src_id=%d node_id=%d\n", src_id, node_id); 705 for (i = 0; i < SKX_NUM_IMC; i++) { 706 d->imc[i].mc = mc++; 707 d->imc[i].lmc = i; 708 d->imc[i].src_id = src_id; 709 d->imc[i].node_id = node_id; 710 rc = skx_register_mci(&d->imc[i], d->imc[i].chan[0].cdev, 711 "Skylake Socket", EDAC_MOD_STR, 712 skx_get_dimm_config, cfg); 713 if (rc < 0) 714 goto fail; 715 } 716 } 717 718 skx_set_decode(skx_decode, skx_show_retry_rd_err_log); 719 720 if (nvdimm_count && skx_adxl_get() != -ENODEV) { 721 skx_set_decode(NULL, skx_show_retry_rd_err_log); 722 } else { 723 if (nvdimm_count) 724 skx_printk(KERN_NOTICE, "Only decoding DDR4 address!\n"); 725 skx_set_decode(skx_decode, skx_show_retry_rd_err_log); 726 } 727 728 /* Ensure that the OPSTATE is set correctly for POLL or NMI */ 729 opstate_init(); 730 731 setup_skx_debug(); 732 733 mce_register_decode_chain(&skx_mce_dec); 734 735 return 0; 736 fail: 737 skx_remove(); 738 return rc; 739 } 740 741 static void __exit skx_exit(void) 742 { 743 edac_dbg(2, "\n"); 744 mce_unregister_decode_chain(&skx_mce_dec); 745 teardown_skx_debug(); 746 if (nvdimm_count) 747 skx_adxl_put(); 748 skx_remove(); 749 } 750 751 module_init(skx_init); 752 module_exit(skx_exit); 753 754 module_param(edac_op_state, int, 0444); 755 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI"); 756 757 MODULE_LICENSE("GPL v2"); 758 MODULE_AUTHOR("Tony Luck"); 759 MODULE_DESCRIPTION("MC Driver for Intel Skylake server processors"); 760