1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved. 4 */ 5 #include <linux/list_sort.h> 6 #include <linux/libnvdimm.h> 7 #include <linux/module.h> 8 #include <linux/nospec.h> 9 #include <linux/mutex.h> 10 #include <linux/ndctl.h> 11 #include <linux/sysfs.h> 12 #include <linux/delay.h> 13 #include <linux/list.h> 14 #include <linux/acpi.h> 15 #include <linux/sort.h> 16 #include <linux/io.h> 17 #include <linux/nd.h> 18 #include <asm/cacheflush.h> 19 #include <acpi/nfit.h> 20 #include "intel.h" 21 #include "nfit.h" 22 23 /* 24 * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is 25 * irrelevant. 26 */ 27 #include <linux/io-64-nonatomic-hi-lo.h> 28 29 static bool force_enable_dimms; 30 module_param(force_enable_dimms, bool, S_IRUGO|S_IWUSR); 31 MODULE_PARM_DESC(force_enable_dimms, "Ignore _STA (ACPI DIMM device) status"); 32 33 static bool disable_vendor_specific; 34 module_param(disable_vendor_specific, bool, S_IRUGO); 35 MODULE_PARM_DESC(disable_vendor_specific, 36 "Limit commands to the publicly specified set"); 37 38 static unsigned long override_dsm_mask; 39 module_param(override_dsm_mask, ulong, S_IRUGO); 40 MODULE_PARM_DESC(override_dsm_mask, "Bitmask of allowed NVDIMM DSM functions"); 41 42 static int default_dsm_family = -1; 43 module_param(default_dsm_family, int, S_IRUGO); 44 MODULE_PARM_DESC(default_dsm_family, 45 "Try this DSM type first when identifying NVDIMM family"); 46 47 static bool no_init_ars; 48 module_param(no_init_ars, bool, 0644); 49 MODULE_PARM_DESC(no_init_ars, "Skip ARS run at nfit init time"); 50 51 static bool force_labels; 52 module_param(force_labels, bool, 0444); 53 MODULE_PARM_DESC(force_labels, "Opt-in to labels despite missing methods"); 54 55 LIST_HEAD(acpi_descs); 56 DEFINE_MUTEX(acpi_desc_lock); 57 58 static struct workqueue_struct *nfit_wq; 59 60 struct nfit_table_prev { 61 struct list_head spas; 62 struct list_head memdevs; 63 struct list_head dcrs; 64 struct list_head bdws; 65 struct list_head idts; 66 struct list_head flushes; 67 }; 68 69 static guid_t nfit_uuid[NFIT_UUID_MAX]; 70 71 const guid_t *to_nfit_uuid(enum nfit_uuids id) 72 { 73 return &nfit_uuid[id]; 74 } 75 EXPORT_SYMBOL(to_nfit_uuid); 76 77 static const guid_t *to_nfit_bus_uuid(int family) 78 { 79 if (WARN_ONCE(family == NVDIMM_BUS_FAMILY_NFIT, 80 "only secondary bus families can be translated\n")) 81 return NULL; 82 /* 83 * The index of bus UUIDs starts immediately following the last 84 * NVDIMM/leaf family. 85 */ 86 return to_nfit_uuid(family + NVDIMM_FAMILY_MAX); 87 } 88 89 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc) 90 { 91 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc; 92 93 /* 94 * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct 95 * acpi_device. 96 */ 97 if (!nd_desc->provider_name 98 || strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0) 99 return NULL; 100 101 return to_acpi_device(acpi_desc->dev); 102 } 103 104 static int xlat_bus_status(void *buf, unsigned int cmd, u32 status) 105 { 106 struct nd_cmd_clear_error *clear_err; 107 struct nd_cmd_ars_status *ars_status; 108 u16 flags; 109 110 switch (cmd) { 111 case ND_CMD_ARS_CAP: 112 if ((status & 0xffff) == NFIT_ARS_CAP_NONE) 113 return -ENOTTY; 114 115 /* Command failed */ 116 if (status & 0xffff) 117 return -EIO; 118 119 /* No supported scan types for this range */ 120 flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE; 121 if ((status >> 16 & flags) == 0) 122 return -ENOTTY; 123 return 0; 124 case ND_CMD_ARS_START: 125 /* ARS is in progress */ 126 if ((status & 0xffff) == NFIT_ARS_START_BUSY) 127 return -EBUSY; 128 129 /* Command failed */ 130 if (status & 0xffff) 131 return -EIO; 132 return 0; 133 case ND_CMD_ARS_STATUS: 134 ars_status = buf; 135 /* Command failed */ 136 if (status & 0xffff) 137 return -EIO; 138 /* Check extended status (Upper two bytes) */ 139 if (status == NFIT_ARS_STATUS_DONE) 140 return 0; 141 142 /* ARS is in progress */ 143 if (status == NFIT_ARS_STATUS_BUSY) 144 return -EBUSY; 145 146 /* No ARS performed for the current boot */ 147 if (status == NFIT_ARS_STATUS_NONE) 148 return -EAGAIN; 149 150 /* 151 * ARS interrupted, either we overflowed or some other 152 * agent wants the scan to stop. If we didn't overflow 153 * then just continue with the returned results. 154 */ 155 if (status == NFIT_ARS_STATUS_INTR) { 156 if (ars_status->out_length >= 40 && (ars_status->flags 157 & NFIT_ARS_F_OVERFLOW)) 158 return -ENOSPC; 159 return 0; 160 } 161 162 /* Unknown status */ 163 if (status >> 16) 164 return -EIO; 165 return 0; 166 case ND_CMD_CLEAR_ERROR: 167 clear_err = buf; 168 if (status & 0xffff) 169 return -EIO; 170 if (!clear_err->cleared) 171 return -EIO; 172 if (clear_err->length > clear_err->cleared) 173 return clear_err->cleared; 174 return 0; 175 default: 176 break; 177 } 178 179 /* all other non-zero status results in an error */ 180 if (status) 181 return -EIO; 182 return 0; 183 } 184 185 #define ACPI_LABELS_LOCKED 3 186 187 static int xlat_nvdimm_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd, 188 u32 status) 189 { 190 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 191 192 switch (cmd) { 193 case ND_CMD_GET_CONFIG_SIZE: 194 /* 195 * In the _LSI, _LSR, _LSW case the locked status is 196 * communicated via the read/write commands 197 */ 198 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) 199 break; 200 201 if (status >> 16 & ND_CONFIG_LOCKED) 202 return -EACCES; 203 break; 204 case ND_CMD_GET_CONFIG_DATA: 205 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags) 206 && status == ACPI_LABELS_LOCKED) 207 return -EACCES; 208 break; 209 case ND_CMD_SET_CONFIG_DATA: 210 if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags) 211 && status == ACPI_LABELS_LOCKED) 212 return -EACCES; 213 break; 214 default: 215 break; 216 } 217 218 /* all other non-zero status results in an error */ 219 if (status) 220 return -EIO; 221 return 0; 222 } 223 224 static int xlat_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd, 225 u32 status) 226 { 227 if (!nvdimm) 228 return xlat_bus_status(buf, cmd, status); 229 return xlat_nvdimm_status(nvdimm, buf, cmd, status); 230 } 231 232 /* convert _LS{I,R} packages to the buffer object acpi_nfit_ctl expects */ 233 static union acpi_object *pkg_to_buf(union acpi_object *pkg) 234 { 235 int i; 236 void *dst; 237 size_t size = 0; 238 union acpi_object *buf = NULL; 239 240 if (pkg->type != ACPI_TYPE_PACKAGE) { 241 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n", 242 pkg->type); 243 goto err; 244 } 245 246 for (i = 0; i < pkg->package.count; i++) { 247 union acpi_object *obj = &pkg->package.elements[i]; 248 249 if (obj->type == ACPI_TYPE_INTEGER) 250 size += 4; 251 else if (obj->type == ACPI_TYPE_BUFFER) 252 size += obj->buffer.length; 253 else { 254 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n", 255 obj->type); 256 goto err; 257 } 258 } 259 260 buf = ACPI_ALLOCATE(sizeof(*buf) + size); 261 if (!buf) 262 goto err; 263 264 dst = buf + 1; 265 buf->type = ACPI_TYPE_BUFFER; 266 buf->buffer.length = size; 267 buf->buffer.pointer = dst; 268 for (i = 0; i < pkg->package.count; i++) { 269 union acpi_object *obj = &pkg->package.elements[i]; 270 271 if (obj->type == ACPI_TYPE_INTEGER) { 272 memcpy(dst, &obj->integer.value, 4); 273 dst += 4; 274 } else if (obj->type == ACPI_TYPE_BUFFER) { 275 memcpy(dst, obj->buffer.pointer, obj->buffer.length); 276 dst += obj->buffer.length; 277 } 278 } 279 err: 280 ACPI_FREE(pkg); 281 return buf; 282 } 283 284 static union acpi_object *int_to_buf(union acpi_object *integer) 285 { 286 union acpi_object *buf = NULL; 287 void *dst = NULL; 288 289 if (integer->type != ACPI_TYPE_INTEGER) { 290 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n", 291 integer->type); 292 goto err; 293 } 294 295 buf = ACPI_ALLOCATE(sizeof(*buf) + 4); 296 if (!buf) 297 goto err; 298 299 dst = buf + 1; 300 buf->type = ACPI_TYPE_BUFFER; 301 buf->buffer.length = 4; 302 buf->buffer.pointer = dst; 303 memcpy(dst, &integer->integer.value, 4); 304 err: 305 ACPI_FREE(integer); 306 return buf; 307 } 308 309 static union acpi_object *acpi_label_write(acpi_handle handle, u32 offset, 310 u32 len, void *data) 311 { 312 acpi_status rc; 313 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL }; 314 struct acpi_object_list input = { 315 .count = 3, 316 .pointer = (union acpi_object []) { 317 [0] = { 318 .integer.type = ACPI_TYPE_INTEGER, 319 .integer.value = offset, 320 }, 321 [1] = { 322 .integer.type = ACPI_TYPE_INTEGER, 323 .integer.value = len, 324 }, 325 [2] = { 326 .buffer.type = ACPI_TYPE_BUFFER, 327 .buffer.pointer = data, 328 .buffer.length = len, 329 }, 330 }, 331 }; 332 333 rc = acpi_evaluate_object(handle, "_LSW", &input, &buf); 334 if (ACPI_FAILURE(rc)) 335 return NULL; 336 return int_to_buf(buf.pointer); 337 } 338 339 static union acpi_object *acpi_label_read(acpi_handle handle, u32 offset, 340 u32 len) 341 { 342 acpi_status rc; 343 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL }; 344 struct acpi_object_list input = { 345 .count = 2, 346 .pointer = (union acpi_object []) { 347 [0] = { 348 .integer.type = ACPI_TYPE_INTEGER, 349 .integer.value = offset, 350 }, 351 [1] = { 352 .integer.type = ACPI_TYPE_INTEGER, 353 .integer.value = len, 354 }, 355 }, 356 }; 357 358 rc = acpi_evaluate_object(handle, "_LSR", &input, &buf); 359 if (ACPI_FAILURE(rc)) 360 return NULL; 361 return pkg_to_buf(buf.pointer); 362 } 363 364 static union acpi_object *acpi_label_info(acpi_handle handle) 365 { 366 acpi_status rc; 367 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL }; 368 369 rc = acpi_evaluate_object(handle, "_LSI", NULL, &buf); 370 if (ACPI_FAILURE(rc)) 371 return NULL; 372 return pkg_to_buf(buf.pointer); 373 } 374 375 static u8 nfit_dsm_revid(unsigned family, unsigned func) 376 { 377 static const u8 revid_table[NVDIMM_FAMILY_MAX+1][NVDIMM_CMD_MAX+1] = { 378 [NVDIMM_FAMILY_INTEL] = { 379 [NVDIMM_INTEL_GET_MODES ... 380 NVDIMM_INTEL_FW_ACTIVATE_ARM] = 2, 381 }, 382 }; 383 u8 id; 384 385 if (family > NVDIMM_FAMILY_MAX) 386 return 0; 387 if (func > NVDIMM_CMD_MAX) 388 return 0; 389 id = revid_table[family][func]; 390 if (id == 0) 391 return 1; /* default */ 392 return id; 393 } 394 395 static bool payload_dumpable(struct nvdimm *nvdimm, unsigned int func) 396 { 397 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 398 399 if (nfit_mem && nfit_mem->family == NVDIMM_FAMILY_INTEL 400 && func >= NVDIMM_INTEL_GET_SECURITY_STATE 401 && func <= NVDIMM_INTEL_MASTER_SECURE_ERASE) 402 return IS_ENABLED(CONFIG_NFIT_SECURITY_DEBUG); 403 return true; 404 } 405 406 static int cmd_to_func(struct nfit_mem *nfit_mem, unsigned int cmd, 407 struct nd_cmd_pkg *call_pkg, int *family) 408 { 409 if (call_pkg) { 410 int i; 411 412 if (nfit_mem && nfit_mem->family != call_pkg->nd_family) 413 return -ENOTTY; 414 415 for (i = 0; i < ARRAY_SIZE(call_pkg->nd_reserved2); i++) 416 if (call_pkg->nd_reserved2[i]) 417 return -EINVAL; 418 *family = call_pkg->nd_family; 419 return call_pkg->nd_command; 420 } 421 422 /* In the !call_pkg case, bus commands == bus functions */ 423 if (!nfit_mem) 424 return cmd; 425 426 /* Linux ND commands == NVDIMM_FAMILY_INTEL function numbers */ 427 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) 428 return cmd; 429 430 /* 431 * Force function number validation to fail since 0 is never 432 * published as a valid function in dsm_mask. 433 */ 434 return 0; 435 } 436 437 int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm, 438 unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc) 439 { 440 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc); 441 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 442 union acpi_object in_obj, in_buf, *out_obj; 443 const struct nd_cmd_desc *desc = NULL; 444 struct device *dev = acpi_desc->dev; 445 struct nd_cmd_pkg *call_pkg = NULL; 446 const char *cmd_name, *dimm_name; 447 unsigned long cmd_mask, dsm_mask; 448 u32 offset, fw_status = 0; 449 acpi_handle handle; 450 const guid_t *guid; 451 int func, rc, i; 452 int family = 0; 453 454 if (cmd_rc) 455 *cmd_rc = -EINVAL; 456 457 if (cmd == ND_CMD_CALL) 458 call_pkg = buf; 459 func = cmd_to_func(nfit_mem, cmd, call_pkg, &family); 460 if (func < 0) 461 return func; 462 463 if (nvdimm) { 464 struct acpi_device *adev = nfit_mem->adev; 465 466 if (!adev) 467 return -ENOTTY; 468 469 dimm_name = nvdimm_name(nvdimm); 470 cmd_name = nvdimm_cmd_name(cmd); 471 cmd_mask = nvdimm_cmd_mask(nvdimm); 472 dsm_mask = nfit_mem->dsm_mask; 473 desc = nd_cmd_dimm_desc(cmd); 474 guid = to_nfit_uuid(nfit_mem->family); 475 handle = adev->handle; 476 } else { 477 struct acpi_device *adev = to_acpi_dev(acpi_desc); 478 479 cmd_name = nvdimm_bus_cmd_name(cmd); 480 cmd_mask = nd_desc->cmd_mask; 481 if (cmd == ND_CMD_CALL && call_pkg->nd_family) { 482 family = call_pkg->nd_family; 483 if (family > NVDIMM_BUS_FAMILY_MAX || 484 !test_bit(family, &nd_desc->bus_family_mask)) 485 return -EINVAL; 486 family = array_index_nospec(family, 487 NVDIMM_BUS_FAMILY_MAX + 1); 488 dsm_mask = acpi_desc->family_dsm_mask[family]; 489 guid = to_nfit_bus_uuid(family); 490 } else { 491 dsm_mask = acpi_desc->bus_dsm_mask; 492 guid = to_nfit_uuid(NFIT_DEV_BUS); 493 } 494 desc = nd_cmd_bus_desc(cmd); 495 handle = adev->handle; 496 dimm_name = "bus"; 497 } 498 499 if (!desc || (cmd && (desc->out_num + desc->in_num == 0))) 500 return -ENOTTY; 501 502 /* 503 * Check for a valid command. For ND_CMD_CALL, we also have to 504 * make sure that the DSM function is supported. 505 */ 506 if (cmd == ND_CMD_CALL && 507 (func > NVDIMM_CMD_MAX || !test_bit(func, &dsm_mask))) 508 return -ENOTTY; 509 else if (!test_bit(cmd, &cmd_mask)) 510 return -ENOTTY; 511 512 in_obj.type = ACPI_TYPE_PACKAGE; 513 in_obj.package.count = 1; 514 in_obj.package.elements = &in_buf; 515 in_buf.type = ACPI_TYPE_BUFFER; 516 in_buf.buffer.pointer = buf; 517 in_buf.buffer.length = 0; 518 519 /* libnvdimm has already validated the input envelope */ 520 for (i = 0; i < desc->in_num; i++) 521 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc, 522 i, buf); 523 524 if (call_pkg) { 525 /* skip over package wrapper */ 526 in_buf.buffer.pointer = (void *) &call_pkg->nd_payload; 527 in_buf.buffer.length = call_pkg->nd_size_in; 528 } 529 530 dev_dbg(dev, "%s cmd: %d: family: %d func: %d input length: %d\n", 531 dimm_name, cmd, family, func, in_buf.buffer.length); 532 if (payload_dumpable(nvdimm, func)) 533 print_hex_dump_debug("nvdimm in ", DUMP_PREFIX_OFFSET, 4, 4, 534 in_buf.buffer.pointer, 535 min_t(u32, 256, in_buf.buffer.length), true); 536 537 /* call the BIOS, prefer the named methods over _DSM if available */ 538 if (nvdimm && cmd == ND_CMD_GET_CONFIG_SIZE 539 && test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) 540 out_obj = acpi_label_info(handle); 541 else if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA 542 && test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) { 543 struct nd_cmd_get_config_data_hdr *p = buf; 544 545 out_obj = acpi_label_read(handle, p->in_offset, p->in_length); 546 } else if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA 547 && test_bit(NFIT_MEM_LSW, &nfit_mem->flags)) { 548 struct nd_cmd_set_config_hdr *p = buf; 549 550 out_obj = acpi_label_write(handle, p->in_offset, p->in_length, 551 p->in_buf); 552 } else { 553 u8 revid; 554 555 if (nvdimm) 556 revid = nfit_dsm_revid(nfit_mem->family, func); 557 else 558 revid = 1; 559 out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj); 560 } 561 562 if (!out_obj) { 563 dev_dbg(dev, "%s _DSM failed cmd: %s\n", dimm_name, cmd_name); 564 return -EINVAL; 565 } 566 567 if (out_obj->type != ACPI_TYPE_BUFFER) { 568 dev_dbg(dev, "%s unexpected output object type cmd: %s type: %d\n", 569 dimm_name, cmd_name, out_obj->type); 570 rc = -EINVAL; 571 goto out; 572 } 573 574 dev_dbg(dev, "%s cmd: %s output length: %d\n", dimm_name, 575 cmd_name, out_obj->buffer.length); 576 print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4, 4, 577 out_obj->buffer.pointer, 578 min_t(u32, 128, out_obj->buffer.length), true); 579 580 if (call_pkg) { 581 call_pkg->nd_fw_size = out_obj->buffer.length; 582 memcpy(call_pkg->nd_payload + call_pkg->nd_size_in, 583 out_obj->buffer.pointer, 584 min(call_pkg->nd_fw_size, call_pkg->nd_size_out)); 585 586 ACPI_FREE(out_obj); 587 /* 588 * Need to support FW function w/o known size in advance. 589 * Caller can determine required size based upon nd_fw_size. 590 * If we return an error (like elsewhere) then caller wouldn't 591 * be able to rely upon data returned to make calculation. 592 */ 593 if (cmd_rc) 594 *cmd_rc = 0; 595 return 0; 596 } 597 598 for (i = 0, offset = 0; i < desc->out_num; i++) { 599 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf, 600 (u32 *) out_obj->buffer.pointer, 601 out_obj->buffer.length - offset); 602 603 if (offset + out_size > out_obj->buffer.length) { 604 dev_dbg(dev, "%s output object underflow cmd: %s field: %d\n", 605 dimm_name, cmd_name, i); 606 break; 607 } 608 609 if (in_buf.buffer.length + offset + out_size > buf_len) { 610 dev_dbg(dev, "%s output overrun cmd: %s field: %d\n", 611 dimm_name, cmd_name, i); 612 rc = -ENXIO; 613 goto out; 614 } 615 memcpy(buf + in_buf.buffer.length + offset, 616 out_obj->buffer.pointer + offset, out_size); 617 offset += out_size; 618 } 619 620 /* 621 * Set fw_status for all the commands with a known format to be 622 * later interpreted by xlat_status(). 623 */ 624 if (i >= 1 && ((!nvdimm && cmd >= ND_CMD_ARS_CAP 625 && cmd <= ND_CMD_CLEAR_ERROR) 626 || (nvdimm && cmd >= ND_CMD_SMART 627 && cmd <= ND_CMD_VENDOR))) 628 fw_status = *(u32 *) out_obj->buffer.pointer; 629 630 if (offset + in_buf.buffer.length < buf_len) { 631 if (i >= 1) { 632 /* 633 * status valid, return the number of bytes left 634 * unfilled in the output buffer 635 */ 636 rc = buf_len - offset - in_buf.buffer.length; 637 if (cmd_rc) 638 *cmd_rc = xlat_status(nvdimm, buf, cmd, 639 fw_status); 640 } else { 641 dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n", 642 __func__, dimm_name, cmd_name, buf_len, 643 offset); 644 rc = -ENXIO; 645 } 646 } else { 647 rc = 0; 648 if (cmd_rc) 649 *cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status); 650 } 651 652 out: 653 ACPI_FREE(out_obj); 654 655 return rc; 656 } 657 EXPORT_SYMBOL_GPL(acpi_nfit_ctl); 658 659 static const char *spa_type_name(u16 type) 660 { 661 static const char *to_name[] = { 662 [NFIT_SPA_VOLATILE] = "volatile", 663 [NFIT_SPA_PM] = "pmem", 664 [NFIT_SPA_DCR] = "dimm-control-region", 665 [NFIT_SPA_BDW] = "block-data-window", 666 [NFIT_SPA_VDISK] = "volatile-disk", 667 [NFIT_SPA_VCD] = "volatile-cd", 668 [NFIT_SPA_PDISK] = "persistent-disk", 669 [NFIT_SPA_PCD] = "persistent-cd", 670 671 }; 672 673 if (type > NFIT_SPA_PCD) 674 return "unknown"; 675 676 return to_name[type]; 677 } 678 679 int nfit_spa_type(struct acpi_nfit_system_address *spa) 680 { 681 guid_t guid; 682 int i; 683 684 import_guid(&guid, spa->range_guid); 685 for (i = 0; i < NFIT_UUID_MAX; i++) 686 if (guid_equal(to_nfit_uuid(i), &guid)) 687 return i; 688 return -1; 689 } 690 691 static size_t sizeof_spa(struct acpi_nfit_system_address *spa) 692 { 693 if (spa->flags & ACPI_NFIT_LOCATION_COOKIE_VALID) 694 return sizeof(*spa); 695 return sizeof(*spa) - 8; 696 } 697 698 static bool add_spa(struct acpi_nfit_desc *acpi_desc, 699 struct nfit_table_prev *prev, 700 struct acpi_nfit_system_address *spa) 701 { 702 struct device *dev = acpi_desc->dev; 703 struct nfit_spa *nfit_spa; 704 705 if (spa->header.length != sizeof_spa(spa)) 706 return false; 707 708 list_for_each_entry(nfit_spa, &prev->spas, list) { 709 if (memcmp(nfit_spa->spa, spa, sizeof_spa(spa)) == 0) { 710 list_move_tail(&nfit_spa->list, &acpi_desc->spas); 711 return true; 712 } 713 } 714 715 nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof_spa(spa), 716 GFP_KERNEL); 717 if (!nfit_spa) 718 return false; 719 INIT_LIST_HEAD(&nfit_spa->list); 720 memcpy(nfit_spa->spa, spa, sizeof_spa(spa)); 721 list_add_tail(&nfit_spa->list, &acpi_desc->spas); 722 dev_dbg(dev, "spa index: %d type: %s\n", 723 spa->range_index, 724 spa_type_name(nfit_spa_type(spa))); 725 return true; 726 } 727 728 static bool add_memdev(struct acpi_nfit_desc *acpi_desc, 729 struct nfit_table_prev *prev, 730 struct acpi_nfit_memory_map *memdev) 731 { 732 struct device *dev = acpi_desc->dev; 733 struct nfit_memdev *nfit_memdev; 734 735 if (memdev->header.length != sizeof(*memdev)) 736 return false; 737 738 list_for_each_entry(nfit_memdev, &prev->memdevs, list) 739 if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) { 740 list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs); 741 return true; 742 } 743 744 nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev), 745 GFP_KERNEL); 746 if (!nfit_memdev) 747 return false; 748 INIT_LIST_HEAD(&nfit_memdev->list); 749 memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev)); 750 list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs); 751 dev_dbg(dev, "memdev handle: %#x spa: %d dcr: %d flags: %#x\n", 752 memdev->device_handle, memdev->range_index, 753 memdev->region_index, memdev->flags); 754 return true; 755 } 756 757 int nfit_get_smbios_id(u32 device_handle, u16 *flags) 758 { 759 struct acpi_nfit_memory_map *memdev; 760 struct acpi_nfit_desc *acpi_desc; 761 struct nfit_mem *nfit_mem; 762 u16 physical_id; 763 764 mutex_lock(&acpi_desc_lock); 765 list_for_each_entry(acpi_desc, &acpi_descs, list) { 766 mutex_lock(&acpi_desc->init_mutex); 767 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) { 768 memdev = __to_nfit_memdev(nfit_mem); 769 if (memdev->device_handle == device_handle) { 770 *flags = memdev->flags; 771 physical_id = memdev->physical_id; 772 mutex_unlock(&acpi_desc->init_mutex); 773 mutex_unlock(&acpi_desc_lock); 774 return physical_id; 775 } 776 } 777 mutex_unlock(&acpi_desc->init_mutex); 778 } 779 mutex_unlock(&acpi_desc_lock); 780 781 return -ENODEV; 782 } 783 EXPORT_SYMBOL_GPL(nfit_get_smbios_id); 784 785 /* 786 * An implementation may provide a truncated control region if no block windows 787 * are defined. 788 */ 789 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr) 790 { 791 if (dcr->header.length < offsetof(struct acpi_nfit_control_region, 792 window_size)) 793 return 0; 794 if (dcr->windows) 795 return sizeof(*dcr); 796 return offsetof(struct acpi_nfit_control_region, window_size); 797 } 798 799 static bool add_dcr(struct acpi_nfit_desc *acpi_desc, 800 struct nfit_table_prev *prev, 801 struct acpi_nfit_control_region *dcr) 802 { 803 struct device *dev = acpi_desc->dev; 804 struct nfit_dcr *nfit_dcr; 805 806 if (!sizeof_dcr(dcr)) 807 return false; 808 809 list_for_each_entry(nfit_dcr, &prev->dcrs, list) 810 if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) { 811 list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs); 812 return true; 813 } 814 815 nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr), 816 GFP_KERNEL); 817 if (!nfit_dcr) 818 return false; 819 INIT_LIST_HEAD(&nfit_dcr->list); 820 memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)); 821 list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs); 822 dev_dbg(dev, "dcr index: %d windows: %d\n", 823 dcr->region_index, dcr->windows); 824 return true; 825 } 826 827 static bool add_bdw(struct acpi_nfit_desc *acpi_desc, 828 struct nfit_table_prev *prev, 829 struct acpi_nfit_data_region *bdw) 830 { 831 struct device *dev = acpi_desc->dev; 832 struct nfit_bdw *nfit_bdw; 833 834 if (bdw->header.length != sizeof(*bdw)) 835 return false; 836 list_for_each_entry(nfit_bdw, &prev->bdws, list) 837 if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) { 838 list_move_tail(&nfit_bdw->list, &acpi_desc->bdws); 839 return true; 840 } 841 842 nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw), 843 GFP_KERNEL); 844 if (!nfit_bdw) 845 return false; 846 INIT_LIST_HEAD(&nfit_bdw->list); 847 memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw)); 848 list_add_tail(&nfit_bdw->list, &acpi_desc->bdws); 849 dev_dbg(dev, "bdw dcr: %d windows: %d\n", 850 bdw->region_index, bdw->windows); 851 return true; 852 } 853 854 static size_t sizeof_idt(struct acpi_nfit_interleave *idt) 855 { 856 if (idt->header.length < sizeof(*idt)) 857 return 0; 858 return sizeof(*idt) + sizeof(u32) * idt->line_count; 859 } 860 861 static bool add_idt(struct acpi_nfit_desc *acpi_desc, 862 struct nfit_table_prev *prev, 863 struct acpi_nfit_interleave *idt) 864 { 865 struct device *dev = acpi_desc->dev; 866 struct nfit_idt *nfit_idt; 867 868 if (!sizeof_idt(idt)) 869 return false; 870 871 list_for_each_entry(nfit_idt, &prev->idts, list) { 872 if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt)) 873 continue; 874 875 if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) { 876 list_move_tail(&nfit_idt->list, &acpi_desc->idts); 877 return true; 878 } 879 } 880 881 nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt), 882 GFP_KERNEL); 883 if (!nfit_idt) 884 return false; 885 INIT_LIST_HEAD(&nfit_idt->list); 886 memcpy(nfit_idt->idt, idt, sizeof_idt(idt)); 887 list_add_tail(&nfit_idt->list, &acpi_desc->idts); 888 dev_dbg(dev, "idt index: %d num_lines: %d\n", 889 idt->interleave_index, idt->line_count); 890 return true; 891 } 892 893 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush) 894 { 895 if (flush->header.length < sizeof(*flush)) 896 return 0; 897 return struct_size(flush, hint_address, flush->hint_count); 898 } 899 900 static bool add_flush(struct acpi_nfit_desc *acpi_desc, 901 struct nfit_table_prev *prev, 902 struct acpi_nfit_flush_address *flush) 903 { 904 struct device *dev = acpi_desc->dev; 905 struct nfit_flush *nfit_flush; 906 907 if (!sizeof_flush(flush)) 908 return false; 909 910 list_for_each_entry(nfit_flush, &prev->flushes, list) { 911 if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush)) 912 continue; 913 914 if (memcmp(nfit_flush->flush, flush, 915 sizeof_flush(flush)) == 0) { 916 list_move_tail(&nfit_flush->list, &acpi_desc->flushes); 917 return true; 918 } 919 } 920 921 nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush) 922 + sizeof_flush(flush), GFP_KERNEL); 923 if (!nfit_flush) 924 return false; 925 INIT_LIST_HEAD(&nfit_flush->list); 926 memcpy(nfit_flush->flush, flush, sizeof_flush(flush)); 927 list_add_tail(&nfit_flush->list, &acpi_desc->flushes); 928 dev_dbg(dev, "nfit_flush handle: %d hint_count: %d\n", 929 flush->device_handle, flush->hint_count); 930 return true; 931 } 932 933 static bool add_platform_cap(struct acpi_nfit_desc *acpi_desc, 934 struct acpi_nfit_capabilities *pcap) 935 { 936 struct device *dev = acpi_desc->dev; 937 u32 mask; 938 939 mask = (1 << (pcap->highest_capability + 1)) - 1; 940 acpi_desc->platform_cap = pcap->capabilities & mask; 941 dev_dbg(dev, "cap: %#x\n", acpi_desc->platform_cap); 942 return true; 943 } 944 945 static void *add_table(struct acpi_nfit_desc *acpi_desc, 946 struct nfit_table_prev *prev, void *table, const void *end) 947 { 948 struct device *dev = acpi_desc->dev; 949 struct acpi_nfit_header *hdr; 950 void *err = ERR_PTR(-ENOMEM); 951 952 if (table >= end) 953 return NULL; 954 955 hdr = table; 956 if (!hdr->length) { 957 dev_warn(dev, "found a zero length table '%d' parsing nfit\n", 958 hdr->type); 959 return NULL; 960 } 961 962 switch (hdr->type) { 963 case ACPI_NFIT_TYPE_SYSTEM_ADDRESS: 964 if (!add_spa(acpi_desc, prev, table)) 965 return err; 966 break; 967 case ACPI_NFIT_TYPE_MEMORY_MAP: 968 if (!add_memdev(acpi_desc, prev, table)) 969 return err; 970 break; 971 case ACPI_NFIT_TYPE_CONTROL_REGION: 972 if (!add_dcr(acpi_desc, prev, table)) 973 return err; 974 break; 975 case ACPI_NFIT_TYPE_DATA_REGION: 976 if (!add_bdw(acpi_desc, prev, table)) 977 return err; 978 break; 979 case ACPI_NFIT_TYPE_INTERLEAVE: 980 if (!add_idt(acpi_desc, prev, table)) 981 return err; 982 break; 983 case ACPI_NFIT_TYPE_FLUSH_ADDRESS: 984 if (!add_flush(acpi_desc, prev, table)) 985 return err; 986 break; 987 case ACPI_NFIT_TYPE_SMBIOS: 988 dev_dbg(dev, "smbios\n"); 989 break; 990 case ACPI_NFIT_TYPE_CAPABILITIES: 991 if (!add_platform_cap(acpi_desc, table)) 992 return err; 993 break; 994 default: 995 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type); 996 break; 997 } 998 999 return table + hdr->length; 1000 } 1001 1002 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc, 1003 struct acpi_nfit_system_address *spa) 1004 { 1005 struct nfit_mem *nfit_mem, *found; 1006 struct nfit_memdev *nfit_memdev; 1007 int type = spa ? nfit_spa_type(spa) : 0; 1008 1009 switch (type) { 1010 case NFIT_SPA_DCR: 1011 case NFIT_SPA_PM: 1012 break; 1013 default: 1014 if (spa) 1015 return 0; 1016 } 1017 1018 /* 1019 * This loop runs in two modes, when a dimm is mapped the loop 1020 * adds memdev associations to an existing dimm, or creates a 1021 * dimm. In the unmapped dimm case this loop sweeps for memdev 1022 * instances with an invalid / zero range_index and adds those 1023 * dimms without spa associations. 1024 */ 1025 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) { 1026 struct nfit_flush *nfit_flush; 1027 struct nfit_dcr *nfit_dcr; 1028 u32 device_handle; 1029 u16 dcr; 1030 1031 if (spa && nfit_memdev->memdev->range_index != spa->range_index) 1032 continue; 1033 if (!spa && nfit_memdev->memdev->range_index) 1034 continue; 1035 found = NULL; 1036 dcr = nfit_memdev->memdev->region_index; 1037 device_handle = nfit_memdev->memdev->device_handle; 1038 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) 1039 if (__to_nfit_memdev(nfit_mem)->device_handle 1040 == device_handle) { 1041 found = nfit_mem; 1042 break; 1043 } 1044 1045 if (found) 1046 nfit_mem = found; 1047 else { 1048 nfit_mem = devm_kzalloc(acpi_desc->dev, 1049 sizeof(*nfit_mem), GFP_KERNEL); 1050 if (!nfit_mem) 1051 return -ENOMEM; 1052 INIT_LIST_HEAD(&nfit_mem->list); 1053 nfit_mem->acpi_desc = acpi_desc; 1054 list_add(&nfit_mem->list, &acpi_desc->dimms); 1055 } 1056 1057 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) { 1058 if (nfit_dcr->dcr->region_index != dcr) 1059 continue; 1060 /* 1061 * Record the control region for the dimm. For 1062 * the ACPI 6.1 case, where there are separate 1063 * control regions for the pmem vs blk 1064 * interfaces, be sure to record the extended 1065 * blk details. 1066 */ 1067 if (!nfit_mem->dcr) 1068 nfit_mem->dcr = nfit_dcr->dcr; 1069 else if (nfit_mem->dcr->windows == 0 1070 && nfit_dcr->dcr->windows) 1071 nfit_mem->dcr = nfit_dcr->dcr; 1072 break; 1073 } 1074 1075 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) { 1076 struct acpi_nfit_flush_address *flush; 1077 u16 i; 1078 1079 if (nfit_flush->flush->device_handle != device_handle) 1080 continue; 1081 nfit_mem->nfit_flush = nfit_flush; 1082 flush = nfit_flush->flush; 1083 nfit_mem->flush_wpq = devm_kcalloc(acpi_desc->dev, 1084 flush->hint_count, 1085 sizeof(struct resource), 1086 GFP_KERNEL); 1087 if (!nfit_mem->flush_wpq) 1088 return -ENOMEM; 1089 for (i = 0; i < flush->hint_count; i++) { 1090 struct resource *res = &nfit_mem->flush_wpq[i]; 1091 1092 res->start = flush->hint_address[i]; 1093 res->end = res->start + 8 - 1; 1094 } 1095 break; 1096 } 1097 1098 if (dcr && !nfit_mem->dcr) { 1099 dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n", 1100 spa->range_index, dcr); 1101 return -ENODEV; 1102 } 1103 1104 if (type == NFIT_SPA_DCR) { 1105 struct nfit_idt *nfit_idt; 1106 u16 idt_idx; 1107 1108 /* multiple dimms may share a SPA when interleaved */ 1109 nfit_mem->spa_dcr = spa; 1110 nfit_mem->memdev_dcr = nfit_memdev->memdev; 1111 idt_idx = nfit_memdev->memdev->interleave_index; 1112 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) { 1113 if (nfit_idt->idt->interleave_index != idt_idx) 1114 continue; 1115 nfit_mem->idt_dcr = nfit_idt->idt; 1116 break; 1117 } 1118 } else if (type == NFIT_SPA_PM) { 1119 /* 1120 * A single dimm may belong to multiple SPA-PM 1121 * ranges, record at least one in addition to 1122 * any SPA-DCR range. 1123 */ 1124 nfit_mem->memdev_pmem = nfit_memdev->memdev; 1125 } else 1126 nfit_mem->memdev_dcr = nfit_memdev->memdev; 1127 } 1128 1129 return 0; 1130 } 1131 1132 static int nfit_mem_cmp(void *priv, const struct list_head *_a, 1133 const struct list_head *_b) 1134 { 1135 struct nfit_mem *a = container_of(_a, typeof(*a), list); 1136 struct nfit_mem *b = container_of(_b, typeof(*b), list); 1137 u32 handleA, handleB; 1138 1139 handleA = __to_nfit_memdev(a)->device_handle; 1140 handleB = __to_nfit_memdev(b)->device_handle; 1141 if (handleA < handleB) 1142 return -1; 1143 else if (handleA > handleB) 1144 return 1; 1145 return 0; 1146 } 1147 1148 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc) 1149 { 1150 struct nfit_spa *nfit_spa; 1151 int rc; 1152 1153 1154 /* 1155 * For each SPA-DCR or SPA-PMEM address range find its 1156 * corresponding MEMDEV(s). From each MEMDEV find the 1157 * corresponding DCR. Then, if we're operating on a SPA-DCR, 1158 * try to find a SPA-BDW and a corresponding BDW that references 1159 * the DCR. Throw it all into an nfit_mem object. Note, that 1160 * BDWs are optional. 1161 */ 1162 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) { 1163 rc = __nfit_mem_init(acpi_desc, nfit_spa->spa); 1164 if (rc) 1165 return rc; 1166 } 1167 1168 /* 1169 * If a DIMM has failed to be mapped into SPA there will be no 1170 * SPA entries above. Find and register all the unmapped DIMMs 1171 * for reporting and recovery purposes. 1172 */ 1173 rc = __nfit_mem_init(acpi_desc, NULL); 1174 if (rc) 1175 return rc; 1176 1177 list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp); 1178 1179 return 0; 1180 } 1181 1182 static ssize_t bus_dsm_mask_show(struct device *dev, 1183 struct device_attribute *attr, char *buf) 1184 { 1185 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev); 1186 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus); 1187 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc); 1188 1189 return sysfs_emit(buf, "%#lx\n", acpi_desc->bus_dsm_mask); 1190 } 1191 static struct device_attribute dev_attr_bus_dsm_mask = 1192 __ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL); 1193 1194 static ssize_t revision_show(struct device *dev, 1195 struct device_attribute *attr, char *buf) 1196 { 1197 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev); 1198 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus); 1199 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc); 1200 1201 return sysfs_emit(buf, "%d\n", acpi_desc->acpi_header.revision); 1202 } 1203 static DEVICE_ATTR_RO(revision); 1204 1205 static ssize_t hw_error_scrub_show(struct device *dev, 1206 struct device_attribute *attr, char *buf) 1207 { 1208 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev); 1209 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus); 1210 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc); 1211 1212 return sysfs_emit(buf, "%d\n", acpi_desc->scrub_mode); 1213 } 1214 1215 /* 1216 * The 'hw_error_scrub' attribute can have the following values written to it: 1217 * '0': Switch to the default mode where an exception will only insert 1218 * the address of the memory error into the poison and badblocks lists. 1219 * '1': Enable a full scrub to happen if an exception for a memory error is 1220 * received. 1221 */ 1222 static ssize_t hw_error_scrub_store(struct device *dev, 1223 struct device_attribute *attr, const char *buf, size_t size) 1224 { 1225 struct nvdimm_bus_descriptor *nd_desc; 1226 ssize_t rc; 1227 long val; 1228 1229 rc = kstrtol(buf, 0, &val); 1230 if (rc) 1231 return rc; 1232 1233 device_lock(dev); 1234 nd_desc = dev_get_drvdata(dev); 1235 if (nd_desc) { 1236 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc); 1237 1238 switch (val) { 1239 case HW_ERROR_SCRUB_ON: 1240 acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON; 1241 break; 1242 case HW_ERROR_SCRUB_OFF: 1243 acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF; 1244 break; 1245 default: 1246 rc = -EINVAL; 1247 break; 1248 } 1249 } 1250 device_unlock(dev); 1251 if (rc) 1252 return rc; 1253 return size; 1254 } 1255 static DEVICE_ATTR_RW(hw_error_scrub); 1256 1257 /* 1258 * This shows the number of full Address Range Scrubs that have been 1259 * completed since driver load time. Userspace can wait on this using 1260 * select/poll etc. A '+' at the end indicates an ARS is in progress 1261 */ 1262 static ssize_t scrub_show(struct device *dev, 1263 struct device_attribute *attr, char *buf) 1264 { 1265 struct nvdimm_bus_descriptor *nd_desc; 1266 struct acpi_nfit_desc *acpi_desc; 1267 ssize_t rc = -ENXIO; 1268 bool busy; 1269 1270 device_lock(dev); 1271 nd_desc = dev_get_drvdata(dev); 1272 if (!nd_desc) { 1273 device_unlock(dev); 1274 return rc; 1275 } 1276 acpi_desc = to_acpi_desc(nd_desc); 1277 1278 mutex_lock(&acpi_desc->init_mutex); 1279 busy = test_bit(ARS_BUSY, &acpi_desc->scrub_flags) 1280 && !test_bit(ARS_CANCEL, &acpi_desc->scrub_flags); 1281 rc = sysfs_emit(buf, "%d%s", acpi_desc->scrub_count, busy ? "+\n" : "\n"); 1282 /* Allow an admin to poll the busy state at a higher rate */ 1283 if (busy && capable(CAP_SYS_RAWIO) && !test_and_set_bit(ARS_POLL, 1284 &acpi_desc->scrub_flags)) { 1285 acpi_desc->scrub_tmo = 1; 1286 mod_delayed_work(nfit_wq, &acpi_desc->dwork, HZ); 1287 } 1288 1289 mutex_unlock(&acpi_desc->init_mutex); 1290 device_unlock(dev); 1291 return rc; 1292 } 1293 1294 static ssize_t scrub_store(struct device *dev, 1295 struct device_attribute *attr, const char *buf, size_t size) 1296 { 1297 struct nvdimm_bus_descriptor *nd_desc; 1298 ssize_t rc; 1299 long val; 1300 1301 rc = kstrtol(buf, 0, &val); 1302 if (rc) 1303 return rc; 1304 if (val != 1) 1305 return -EINVAL; 1306 1307 device_lock(dev); 1308 nd_desc = dev_get_drvdata(dev); 1309 if (nd_desc) { 1310 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc); 1311 1312 rc = acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG); 1313 } 1314 device_unlock(dev); 1315 if (rc) 1316 return rc; 1317 return size; 1318 } 1319 static DEVICE_ATTR_RW(scrub); 1320 1321 static bool ars_supported(struct nvdimm_bus *nvdimm_bus) 1322 { 1323 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus); 1324 const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START 1325 | 1 << ND_CMD_ARS_STATUS; 1326 1327 return (nd_desc->cmd_mask & mask) == mask; 1328 } 1329 1330 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n) 1331 { 1332 struct device *dev = kobj_to_dev(kobj); 1333 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev); 1334 1335 if (a == &dev_attr_scrub.attr) 1336 return ars_supported(nvdimm_bus) ? a->mode : 0; 1337 1338 if (a == &dev_attr_firmware_activate_noidle.attr) 1339 return intel_fwa_supported(nvdimm_bus) ? a->mode : 0; 1340 1341 return a->mode; 1342 } 1343 1344 static struct attribute *acpi_nfit_attributes[] = { 1345 &dev_attr_revision.attr, 1346 &dev_attr_scrub.attr, 1347 &dev_attr_hw_error_scrub.attr, 1348 &dev_attr_bus_dsm_mask.attr, 1349 &dev_attr_firmware_activate_noidle.attr, 1350 NULL, 1351 }; 1352 1353 static const struct attribute_group acpi_nfit_attribute_group = { 1354 .name = "nfit", 1355 .attrs = acpi_nfit_attributes, 1356 .is_visible = nfit_visible, 1357 }; 1358 1359 static const struct attribute_group *acpi_nfit_attribute_groups[] = { 1360 &acpi_nfit_attribute_group, 1361 NULL, 1362 }; 1363 1364 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev) 1365 { 1366 struct nvdimm *nvdimm = to_nvdimm(dev); 1367 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 1368 1369 return __to_nfit_memdev(nfit_mem); 1370 } 1371 1372 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev) 1373 { 1374 struct nvdimm *nvdimm = to_nvdimm(dev); 1375 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 1376 1377 return nfit_mem->dcr; 1378 } 1379 1380 static ssize_t handle_show(struct device *dev, 1381 struct device_attribute *attr, char *buf) 1382 { 1383 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev); 1384 1385 return sysfs_emit(buf, "%#x\n", memdev->device_handle); 1386 } 1387 static DEVICE_ATTR_RO(handle); 1388 1389 static ssize_t phys_id_show(struct device *dev, 1390 struct device_attribute *attr, char *buf) 1391 { 1392 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev); 1393 1394 return sysfs_emit(buf, "%#x\n", memdev->physical_id); 1395 } 1396 static DEVICE_ATTR_RO(phys_id); 1397 1398 static ssize_t vendor_show(struct device *dev, 1399 struct device_attribute *attr, char *buf) 1400 { 1401 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev); 1402 1403 return sysfs_emit(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id)); 1404 } 1405 static DEVICE_ATTR_RO(vendor); 1406 1407 static ssize_t rev_id_show(struct device *dev, 1408 struct device_attribute *attr, char *buf) 1409 { 1410 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev); 1411 1412 return sysfs_emit(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id)); 1413 } 1414 static DEVICE_ATTR_RO(rev_id); 1415 1416 static ssize_t device_show(struct device *dev, 1417 struct device_attribute *attr, char *buf) 1418 { 1419 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev); 1420 1421 return sysfs_emit(buf, "0x%04x\n", be16_to_cpu(dcr->device_id)); 1422 } 1423 static DEVICE_ATTR_RO(device); 1424 1425 static ssize_t subsystem_vendor_show(struct device *dev, 1426 struct device_attribute *attr, char *buf) 1427 { 1428 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev); 1429 1430 return sysfs_emit(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id)); 1431 } 1432 static DEVICE_ATTR_RO(subsystem_vendor); 1433 1434 static ssize_t subsystem_rev_id_show(struct device *dev, 1435 struct device_attribute *attr, char *buf) 1436 { 1437 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev); 1438 1439 return sysfs_emit(buf, "0x%04x\n", 1440 be16_to_cpu(dcr->subsystem_revision_id)); 1441 } 1442 static DEVICE_ATTR_RO(subsystem_rev_id); 1443 1444 static ssize_t subsystem_device_show(struct device *dev, 1445 struct device_attribute *attr, char *buf) 1446 { 1447 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev); 1448 1449 return sysfs_emit(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id)); 1450 } 1451 static DEVICE_ATTR_RO(subsystem_device); 1452 1453 static int num_nvdimm_formats(struct nvdimm *nvdimm) 1454 { 1455 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 1456 int formats = 0; 1457 1458 if (nfit_mem->memdev_pmem) 1459 formats++; 1460 return formats; 1461 } 1462 1463 static ssize_t format_show(struct device *dev, 1464 struct device_attribute *attr, char *buf) 1465 { 1466 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev); 1467 1468 return sysfs_emit(buf, "0x%04x\n", le16_to_cpu(dcr->code)); 1469 } 1470 static DEVICE_ATTR_RO(format); 1471 1472 static ssize_t format1_show(struct device *dev, 1473 struct device_attribute *attr, char *buf) 1474 { 1475 u32 handle; 1476 ssize_t rc = -ENXIO; 1477 struct nfit_mem *nfit_mem; 1478 struct nfit_memdev *nfit_memdev; 1479 struct acpi_nfit_desc *acpi_desc; 1480 struct nvdimm *nvdimm = to_nvdimm(dev); 1481 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev); 1482 1483 nfit_mem = nvdimm_provider_data(nvdimm); 1484 acpi_desc = nfit_mem->acpi_desc; 1485 handle = to_nfit_memdev(dev)->device_handle; 1486 1487 /* assumes DIMMs have at most 2 published interface codes */ 1488 mutex_lock(&acpi_desc->init_mutex); 1489 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) { 1490 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev; 1491 struct nfit_dcr *nfit_dcr; 1492 1493 if (memdev->device_handle != handle) 1494 continue; 1495 1496 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) { 1497 if (nfit_dcr->dcr->region_index != memdev->region_index) 1498 continue; 1499 if (nfit_dcr->dcr->code == dcr->code) 1500 continue; 1501 rc = sysfs_emit(buf, "0x%04x\n", 1502 le16_to_cpu(nfit_dcr->dcr->code)); 1503 break; 1504 } 1505 if (rc != -ENXIO) 1506 break; 1507 } 1508 mutex_unlock(&acpi_desc->init_mutex); 1509 return rc; 1510 } 1511 static DEVICE_ATTR_RO(format1); 1512 1513 static ssize_t formats_show(struct device *dev, 1514 struct device_attribute *attr, char *buf) 1515 { 1516 struct nvdimm *nvdimm = to_nvdimm(dev); 1517 1518 return sysfs_emit(buf, "%d\n", num_nvdimm_formats(nvdimm)); 1519 } 1520 static DEVICE_ATTR_RO(formats); 1521 1522 static ssize_t serial_show(struct device *dev, 1523 struct device_attribute *attr, char *buf) 1524 { 1525 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev); 1526 1527 return sysfs_emit(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number)); 1528 } 1529 static DEVICE_ATTR_RO(serial); 1530 1531 static ssize_t family_show(struct device *dev, 1532 struct device_attribute *attr, char *buf) 1533 { 1534 struct nvdimm *nvdimm = to_nvdimm(dev); 1535 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 1536 1537 if (nfit_mem->family < 0) 1538 return -ENXIO; 1539 return sysfs_emit(buf, "%d\n", nfit_mem->family); 1540 } 1541 static DEVICE_ATTR_RO(family); 1542 1543 static ssize_t dsm_mask_show(struct device *dev, 1544 struct device_attribute *attr, char *buf) 1545 { 1546 struct nvdimm *nvdimm = to_nvdimm(dev); 1547 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 1548 1549 if (nfit_mem->family < 0) 1550 return -ENXIO; 1551 return sysfs_emit(buf, "%#lx\n", nfit_mem->dsm_mask); 1552 } 1553 static DEVICE_ATTR_RO(dsm_mask); 1554 1555 static ssize_t flags_show(struct device *dev, 1556 struct device_attribute *attr, char *buf) 1557 { 1558 struct nvdimm *nvdimm = to_nvdimm(dev); 1559 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 1560 u16 flags = __to_nfit_memdev(nfit_mem)->flags; 1561 1562 if (test_bit(NFIT_MEM_DIRTY, &nfit_mem->flags)) 1563 flags |= ACPI_NFIT_MEM_FLUSH_FAILED; 1564 1565 return sysfs_emit(buf, "%s%s%s%s%s%s%s\n", 1566 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "", 1567 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "", 1568 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "", 1569 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "", 1570 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "", 1571 flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "", 1572 flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : ""); 1573 } 1574 static DEVICE_ATTR_RO(flags); 1575 1576 static ssize_t id_show(struct device *dev, 1577 struct device_attribute *attr, char *buf) 1578 { 1579 struct nvdimm *nvdimm = to_nvdimm(dev); 1580 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 1581 1582 return sysfs_emit(buf, "%s\n", nfit_mem->id); 1583 } 1584 static DEVICE_ATTR_RO(id); 1585 1586 static ssize_t dirty_shutdown_show(struct device *dev, 1587 struct device_attribute *attr, char *buf) 1588 { 1589 struct nvdimm *nvdimm = to_nvdimm(dev); 1590 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 1591 1592 return sysfs_emit(buf, "%d\n", nfit_mem->dirty_shutdown); 1593 } 1594 static DEVICE_ATTR_RO(dirty_shutdown); 1595 1596 static struct attribute *acpi_nfit_dimm_attributes[] = { 1597 &dev_attr_handle.attr, 1598 &dev_attr_phys_id.attr, 1599 &dev_attr_vendor.attr, 1600 &dev_attr_device.attr, 1601 &dev_attr_rev_id.attr, 1602 &dev_attr_subsystem_vendor.attr, 1603 &dev_attr_subsystem_device.attr, 1604 &dev_attr_subsystem_rev_id.attr, 1605 &dev_attr_format.attr, 1606 &dev_attr_formats.attr, 1607 &dev_attr_format1.attr, 1608 &dev_attr_serial.attr, 1609 &dev_attr_flags.attr, 1610 &dev_attr_id.attr, 1611 &dev_attr_family.attr, 1612 &dev_attr_dsm_mask.attr, 1613 &dev_attr_dirty_shutdown.attr, 1614 NULL, 1615 }; 1616 1617 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj, 1618 struct attribute *a, int n) 1619 { 1620 struct device *dev = kobj_to_dev(kobj); 1621 struct nvdimm *nvdimm = to_nvdimm(dev); 1622 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 1623 1624 if (!to_nfit_dcr(dev)) { 1625 /* Without a dcr only the memdev attributes can be surfaced */ 1626 if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr 1627 || a == &dev_attr_flags.attr 1628 || a == &dev_attr_family.attr 1629 || a == &dev_attr_dsm_mask.attr) 1630 return a->mode; 1631 return 0; 1632 } 1633 1634 if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1) 1635 return 0; 1636 1637 if (!test_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags) 1638 && a == &dev_attr_dirty_shutdown.attr) 1639 return 0; 1640 1641 return a->mode; 1642 } 1643 1644 static const struct attribute_group acpi_nfit_dimm_attribute_group = { 1645 .name = "nfit", 1646 .attrs = acpi_nfit_dimm_attributes, 1647 .is_visible = acpi_nfit_dimm_attr_visible, 1648 }; 1649 1650 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = { 1651 &acpi_nfit_dimm_attribute_group, 1652 NULL, 1653 }; 1654 1655 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc, 1656 u32 device_handle) 1657 { 1658 struct nfit_mem *nfit_mem; 1659 1660 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) 1661 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle) 1662 return nfit_mem->nvdimm; 1663 1664 return NULL; 1665 } 1666 1667 void __acpi_nvdimm_notify(struct device *dev, u32 event) 1668 { 1669 struct nfit_mem *nfit_mem; 1670 struct acpi_nfit_desc *acpi_desc; 1671 1672 dev_dbg(dev->parent, "%s: event: %d\n", dev_name(dev), 1673 event); 1674 1675 if (event != NFIT_NOTIFY_DIMM_HEALTH) { 1676 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev), 1677 event); 1678 return; 1679 } 1680 1681 acpi_desc = dev_get_drvdata(dev->parent); 1682 if (!acpi_desc) 1683 return; 1684 1685 /* 1686 * If we successfully retrieved acpi_desc, then we know nfit_mem data 1687 * is still valid. 1688 */ 1689 nfit_mem = dev_get_drvdata(dev); 1690 if (nfit_mem && nfit_mem->flags_attr) 1691 sysfs_notify_dirent(nfit_mem->flags_attr); 1692 } 1693 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify); 1694 1695 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data) 1696 { 1697 struct acpi_device *adev = data; 1698 struct device *dev = &adev->dev; 1699 1700 device_lock(dev->parent); 1701 __acpi_nvdimm_notify(dev, event); 1702 device_unlock(dev->parent); 1703 } 1704 1705 static bool acpi_nvdimm_has_method(struct acpi_device *adev, char *method) 1706 { 1707 acpi_handle handle; 1708 acpi_status status; 1709 1710 status = acpi_get_handle(adev->handle, method, &handle); 1711 1712 if (ACPI_SUCCESS(status)) 1713 return true; 1714 return false; 1715 } 1716 1717 __weak void nfit_intel_shutdown_status(struct nfit_mem *nfit_mem) 1718 { 1719 struct device *dev = &nfit_mem->adev->dev; 1720 struct nd_intel_smart smart = { 0 }; 1721 union acpi_object in_buf = { 1722 .buffer.type = ACPI_TYPE_BUFFER, 1723 .buffer.length = 0, 1724 }; 1725 union acpi_object in_obj = { 1726 .package.type = ACPI_TYPE_PACKAGE, 1727 .package.count = 1, 1728 .package.elements = &in_buf, 1729 }; 1730 const u8 func = ND_INTEL_SMART; 1731 const guid_t *guid = to_nfit_uuid(nfit_mem->family); 1732 u8 revid = nfit_dsm_revid(nfit_mem->family, func); 1733 struct acpi_device *adev = nfit_mem->adev; 1734 acpi_handle handle = adev->handle; 1735 union acpi_object *out_obj; 1736 1737 if ((nfit_mem->dsm_mask & (1 << func)) == 0) 1738 return; 1739 1740 out_obj = acpi_evaluate_dsm_typed(handle, guid, revid, func, &in_obj, ACPI_TYPE_BUFFER); 1741 if (!out_obj || out_obj->buffer.length < sizeof(smart)) { 1742 dev_dbg(dev->parent, "%s: failed to retrieve initial health\n", 1743 dev_name(dev)); 1744 ACPI_FREE(out_obj); 1745 return; 1746 } 1747 memcpy(&smart, out_obj->buffer.pointer, sizeof(smart)); 1748 ACPI_FREE(out_obj); 1749 1750 if (smart.flags & ND_INTEL_SMART_SHUTDOWN_VALID) { 1751 if (smart.shutdown_state) 1752 set_bit(NFIT_MEM_DIRTY, &nfit_mem->flags); 1753 } 1754 1755 if (smart.flags & ND_INTEL_SMART_SHUTDOWN_COUNT_VALID) { 1756 set_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags); 1757 nfit_mem->dirty_shutdown = smart.shutdown_count; 1758 } 1759 } 1760 1761 static void populate_shutdown_status(struct nfit_mem *nfit_mem) 1762 { 1763 /* 1764 * For DIMMs that provide a dynamic facility to retrieve a 1765 * dirty-shutdown status and/or a dirty-shutdown count, cache 1766 * these values in nfit_mem. 1767 */ 1768 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) 1769 nfit_intel_shutdown_status(nfit_mem); 1770 } 1771 1772 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc, 1773 struct nfit_mem *nfit_mem, u32 device_handle) 1774 { 1775 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc; 1776 struct acpi_device *adev, *adev_dimm; 1777 struct device *dev = acpi_desc->dev; 1778 unsigned long dsm_mask, label_mask; 1779 const guid_t *guid; 1780 int i; 1781 int family = -1; 1782 struct acpi_nfit_control_region *dcr = nfit_mem->dcr; 1783 1784 /* nfit test assumes 1:1 relationship between commands and dsms */ 1785 nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en; 1786 nfit_mem->family = NVDIMM_FAMILY_INTEL; 1787 set_bit(NVDIMM_FAMILY_INTEL, &nd_desc->dimm_family_mask); 1788 1789 if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID) 1790 sprintf(nfit_mem->id, "%04x-%02x-%04x-%08x", 1791 be16_to_cpu(dcr->vendor_id), 1792 dcr->manufacturing_location, 1793 be16_to_cpu(dcr->manufacturing_date), 1794 be32_to_cpu(dcr->serial_number)); 1795 else 1796 sprintf(nfit_mem->id, "%04x-%08x", 1797 be16_to_cpu(dcr->vendor_id), 1798 be32_to_cpu(dcr->serial_number)); 1799 1800 adev = to_acpi_dev(acpi_desc); 1801 if (!adev) { 1802 /* unit test case */ 1803 populate_shutdown_status(nfit_mem); 1804 return 0; 1805 } 1806 1807 adev_dimm = acpi_find_child_device(adev, device_handle, false); 1808 nfit_mem->adev = adev_dimm; 1809 if (!adev_dimm) { 1810 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n", 1811 device_handle); 1812 return force_enable_dimms ? 0 : -ENODEV; 1813 } 1814 1815 if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle, 1816 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) { 1817 dev_err(dev, "%s: notification registration failed\n", 1818 dev_name(&adev_dimm->dev)); 1819 return -ENXIO; 1820 } 1821 /* 1822 * Record nfit_mem for the notification path to track back to 1823 * the nfit sysfs attributes for this dimm device object. 1824 */ 1825 dev_set_drvdata(&adev_dimm->dev, nfit_mem); 1826 1827 /* 1828 * There are 4 "legacy" NVDIMM command sets 1829 * (NVDIMM_FAMILY_{INTEL,MSFT,HPE1,HPE2}) that were created before 1830 * an EFI working group was established to constrain this 1831 * proliferation. The nfit driver probes for the supported command 1832 * set by GUID. Note, if you're a platform developer looking to add 1833 * a new command set to this probe, consider using an existing set, 1834 * or otherwise seek approval to publish the command set at 1835 * http://www.uefi.org/RFIC_LIST. 1836 * 1837 * Note, that checking for function0 (bit0) tells us if any commands 1838 * are reachable through this GUID. 1839 */ 1840 clear_bit(NVDIMM_FAMILY_INTEL, &nd_desc->dimm_family_mask); 1841 for (i = 0; i <= NVDIMM_FAMILY_MAX; i++) 1842 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1)) { 1843 set_bit(i, &nd_desc->dimm_family_mask); 1844 if (family < 0 || i == default_dsm_family) 1845 family = i; 1846 } 1847 1848 /* limit the supported commands to those that are publicly documented */ 1849 nfit_mem->family = family; 1850 if (override_dsm_mask && !disable_vendor_specific) 1851 dsm_mask = override_dsm_mask; 1852 else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) { 1853 dsm_mask = NVDIMM_INTEL_CMDMASK; 1854 if (disable_vendor_specific) 1855 dsm_mask &= ~(1 << ND_CMD_VENDOR); 1856 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) { 1857 dsm_mask = 0x1c3c76; 1858 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) { 1859 dsm_mask = 0x1fe; 1860 if (disable_vendor_specific) 1861 dsm_mask &= ~(1 << 8); 1862 } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) { 1863 dsm_mask = 0xffffffff; 1864 } else if (nfit_mem->family == NVDIMM_FAMILY_HYPERV) { 1865 dsm_mask = 0x1f; 1866 } else { 1867 dev_dbg(dev, "unknown dimm command family\n"); 1868 nfit_mem->family = -1; 1869 /* DSMs are optional, continue loading the driver... */ 1870 return 0; 1871 } 1872 1873 /* 1874 * Function 0 is the command interrogation function, don't 1875 * export it to potential userspace use, and enable it to be 1876 * used as an error value in acpi_nfit_ctl(). 1877 */ 1878 dsm_mask &= ~1UL; 1879 1880 guid = to_nfit_uuid(nfit_mem->family); 1881 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG) 1882 if (acpi_check_dsm(adev_dimm->handle, guid, 1883 nfit_dsm_revid(nfit_mem->family, i), 1884 1ULL << i)) 1885 set_bit(i, &nfit_mem->dsm_mask); 1886 1887 /* 1888 * Prefer the NVDIMM_FAMILY_INTEL label read commands if present 1889 * due to their better semantics handling locked capacity. 1890 */ 1891 label_mask = 1 << ND_CMD_GET_CONFIG_SIZE | 1 << ND_CMD_GET_CONFIG_DATA 1892 | 1 << ND_CMD_SET_CONFIG_DATA; 1893 if (family == NVDIMM_FAMILY_INTEL 1894 && (dsm_mask & label_mask) == label_mask) 1895 /* skip _LS{I,R,W} enabling */; 1896 else { 1897 if (acpi_nvdimm_has_method(adev_dimm, "_LSI") 1898 && acpi_nvdimm_has_method(adev_dimm, "_LSR")) { 1899 dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev)); 1900 set_bit(NFIT_MEM_LSR, &nfit_mem->flags); 1901 } 1902 1903 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags) 1904 && acpi_nvdimm_has_method(adev_dimm, "_LSW")) { 1905 dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev)); 1906 set_bit(NFIT_MEM_LSW, &nfit_mem->flags); 1907 } 1908 1909 /* 1910 * Quirk read-only label configurations to preserve 1911 * access to label-less namespaces by default. 1912 */ 1913 if (!test_bit(NFIT_MEM_LSW, &nfit_mem->flags) 1914 && !force_labels) { 1915 dev_dbg(dev, "%s: No _LSW, disable labels\n", 1916 dev_name(&adev_dimm->dev)); 1917 clear_bit(NFIT_MEM_LSR, &nfit_mem->flags); 1918 } else 1919 dev_dbg(dev, "%s: Force enable labels\n", 1920 dev_name(&adev_dimm->dev)); 1921 } 1922 1923 populate_shutdown_status(nfit_mem); 1924 1925 return 0; 1926 } 1927 1928 static void shutdown_dimm_notify(void *data) 1929 { 1930 struct acpi_nfit_desc *acpi_desc = data; 1931 struct nfit_mem *nfit_mem; 1932 1933 mutex_lock(&acpi_desc->init_mutex); 1934 /* 1935 * Clear out the nfit_mem->flags_attr and shut down dimm event 1936 * notifications. 1937 */ 1938 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) { 1939 struct acpi_device *adev_dimm = nfit_mem->adev; 1940 1941 if (nfit_mem->flags_attr) { 1942 sysfs_put(nfit_mem->flags_attr); 1943 nfit_mem->flags_attr = NULL; 1944 } 1945 if (adev_dimm) { 1946 acpi_remove_notify_handler(adev_dimm->handle, 1947 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify); 1948 dev_set_drvdata(&adev_dimm->dev, NULL); 1949 } 1950 } 1951 mutex_unlock(&acpi_desc->init_mutex); 1952 } 1953 1954 static const struct nvdimm_security_ops *acpi_nfit_get_security_ops(int family) 1955 { 1956 switch (family) { 1957 case NVDIMM_FAMILY_INTEL: 1958 return intel_security_ops; 1959 default: 1960 return NULL; 1961 } 1962 } 1963 1964 static const struct nvdimm_fw_ops *acpi_nfit_get_fw_ops( 1965 struct nfit_mem *nfit_mem) 1966 { 1967 unsigned long mask; 1968 struct acpi_nfit_desc *acpi_desc = nfit_mem->acpi_desc; 1969 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc; 1970 1971 if (!nd_desc->fw_ops) 1972 return NULL; 1973 1974 if (nfit_mem->family != NVDIMM_FAMILY_INTEL) 1975 return NULL; 1976 1977 mask = nfit_mem->dsm_mask & NVDIMM_INTEL_FW_ACTIVATE_CMDMASK; 1978 if (mask != NVDIMM_INTEL_FW_ACTIVATE_CMDMASK) 1979 return NULL; 1980 1981 return intel_fw_ops; 1982 } 1983 1984 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc) 1985 { 1986 struct nfit_mem *nfit_mem; 1987 int dimm_count = 0, rc; 1988 struct nvdimm *nvdimm; 1989 1990 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) { 1991 struct acpi_nfit_flush_address *flush; 1992 unsigned long flags = 0, cmd_mask; 1993 struct nfit_memdev *nfit_memdev; 1994 u32 device_handle; 1995 u16 mem_flags; 1996 1997 device_handle = __to_nfit_memdev(nfit_mem)->device_handle; 1998 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle); 1999 if (nvdimm) { 2000 dimm_count++; 2001 continue; 2002 } 2003 2004 /* collate flags across all memdevs for this dimm */ 2005 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) { 2006 struct acpi_nfit_memory_map *dimm_memdev; 2007 2008 dimm_memdev = __to_nfit_memdev(nfit_mem); 2009 if (dimm_memdev->device_handle 2010 != nfit_memdev->memdev->device_handle) 2011 continue; 2012 dimm_memdev->flags |= nfit_memdev->memdev->flags; 2013 } 2014 2015 mem_flags = __to_nfit_memdev(nfit_mem)->flags; 2016 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED) 2017 set_bit(NDD_UNARMED, &flags); 2018 2019 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle); 2020 if (rc) 2021 continue; 2022 2023 /* 2024 * TODO: provide translation for non-NVDIMM_FAMILY_INTEL 2025 * devices (i.e. from nd_cmd to acpi_dsm) to standardize the 2026 * userspace interface. 2027 */ 2028 cmd_mask = 1UL << ND_CMD_CALL; 2029 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) { 2030 /* 2031 * These commands have a 1:1 correspondence 2032 * between DSM payload and libnvdimm ioctl 2033 * payload format. 2034 */ 2035 cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK; 2036 } 2037 2038 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) { 2039 set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask); 2040 set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask); 2041 } 2042 if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags)) 2043 set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask); 2044 2045 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush 2046 : NULL; 2047 nvdimm = __nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem, 2048 acpi_nfit_dimm_attribute_groups, 2049 flags, cmd_mask, flush ? flush->hint_count : 0, 2050 nfit_mem->flush_wpq, &nfit_mem->id[0], 2051 acpi_nfit_get_security_ops(nfit_mem->family), 2052 acpi_nfit_get_fw_ops(nfit_mem)); 2053 if (!nvdimm) 2054 return -ENOMEM; 2055 2056 nfit_mem->nvdimm = nvdimm; 2057 dimm_count++; 2058 2059 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0) 2060 continue; 2061 2062 dev_err(acpi_desc->dev, "Error found in NVDIMM %s flags:%s%s%s%s%s\n", 2063 nvdimm_name(nvdimm), 2064 mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "", 2065 mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"", 2066 mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "", 2067 mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "", 2068 mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : ""); 2069 2070 } 2071 2072 rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count); 2073 if (rc) 2074 return rc; 2075 2076 /* 2077 * Now that dimms are successfully registered, and async registration 2078 * is flushed, attempt to enable event notification. 2079 */ 2080 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) { 2081 struct kernfs_node *nfit_kernfs; 2082 2083 nvdimm = nfit_mem->nvdimm; 2084 if (!nvdimm) 2085 continue; 2086 2087 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit"); 2088 if (nfit_kernfs) 2089 nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs, 2090 "flags"); 2091 sysfs_put(nfit_kernfs); 2092 if (!nfit_mem->flags_attr) 2093 dev_warn(acpi_desc->dev, "%s: notifications disabled\n", 2094 nvdimm_name(nvdimm)); 2095 } 2096 2097 return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify, 2098 acpi_desc); 2099 } 2100 2101 /* 2102 * These constants are private because there are no kernel consumers of 2103 * these commands. 2104 */ 2105 enum nfit_aux_cmds { 2106 NFIT_CMD_TRANSLATE_SPA = 5, 2107 NFIT_CMD_ARS_INJECT_SET = 7, 2108 NFIT_CMD_ARS_INJECT_CLEAR = 8, 2109 NFIT_CMD_ARS_INJECT_GET = 9, 2110 }; 2111 2112 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc) 2113 { 2114 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc; 2115 const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS); 2116 unsigned long dsm_mask, *mask; 2117 struct acpi_device *adev; 2118 int i; 2119 2120 set_bit(ND_CMD_CALL, &nd_desc->cmd_mask); 2121 set_bit(NVDIMM_BUS_FAMILY_NFIT, &nd_desc->bus_family_mask); 2122 2123 /* enable nfit_test to inject bus command emulation */ 2124 if (acpi_desc->bus_cmd_force_en) { 2125 nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en; 2126 mask = &nd_desc->bus_family_mask; 2127 if (acpi_desc->family_dsm_mask[NVDIMM_BUS_FAMILY_INTEL]) { 2128 set_bit(NVDIMM_BUS_FAMILY_INTEL, mask); 2129 nd_desc->fw_ops = intel_bus_fw_ops; 2130 } 2131 } 2132 2133 adev = to_acpi_dev(acpi_desc); 2134 if (!adev) 2135 return; 2136 2137 for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++) 2138 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i)) 2139 set_bit(i, &nd_desc->cmd_mask); 2140 2141 dsm_mask = 2142 (1 << ND_CMD_ARS_CAP) | 2143 (1 << ND_CMD_ARS_START) | 2144 (1 << ND_CMD_ARS_STATUS) | 2145 (1 << ND_CMD_CLEAR_ERROR) | 2146 (1 << NFIT_CMD_TRANSLATE_SPA) | 2147 (1 << NFIT_CMD_ARS_INJECT_SET) | 2148 (1 << NFIT_CMD_ARS_INJECT_CLEAR) | 2149 (1 << NFIT_CMD_ARS_INJECT_GET); 2150 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG) 2151 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i)) 2152 set_bit(i, &acpi_desc->bus_dsm_mask); 2153 2154 /* Enumerate allowed NVDIMM_BUS_FAMILY_INTEL commands */ 2155 dsm_mask = NVDIMM_BUS_INTEL_FW_ACTIVATE_CMDMASK; 2156 guid = to_nfit_bus_uuid(NVDIMM_BUS_FAMILY_INTEL); 2157 mask = &acpi_desc->family_dsm_mask[NVDIMM_BUS_FAMILY_INTEL]; 2158 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG) 2159 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i)) 2160 set_bit(i, mask); 2161 2162 if (*mask == dsm_mask) { 2163 set_bit(NVDIMM_BUS_FAMILY_INTEL, &nd_desc->bus_family_mask); 2164 nd_desc->fw_ops = intel_bus_fw_ops; 2165 } 2166 } 2167 2168 static ssize_t range_index_show(struct device *dev, 2169 struct device_attribute *attr, char *buf) 2170 { 2171 struct nd_region *nd_region = to_nd_region(dev); 2172 struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region); 2173 2174 return sysfs_emit(buf, "%d\n", nfit_spa->spa->range_index); 2175 } 2176 static DEVICE_ATTR_RO(range_index); 2177 2178 static struct attribute *acpi_nfit_region_attributes[] = { 2179 &dev_attr_range_index.attr, 2180 NULL, 2181 }; 2182 2183 static const struct attribute_group acpi_nfit_region_attribute_group = { 2184 .name = "nfit", 2185 .attrs = acpi_nfit_region_attributes, 2186 }; 2187 2188 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = { 2189 &acpi_nfit_region_attribute_group, 2190 NULL, 2191 }; 2192 2193 /* enough info to uniquely specify an interleave set */ 2194 struct nfit_set_info { 2195 u64 region_offset; 2196 u32 serial_number; 2197 u32 pad; 2198 }; 2199 2200 struct nfit_set_info2 { 2201 u64 region_offset; 2202 u32 serial_number; 2203 u16 vendor_id; 2204 u16 manufacturing_date; 2205 u8 manufacturing_location; 2206 u8 reserved[31]; 2207 }; 2208 2209 static int cmp_map_compat(const void *m0, const void *m1) 2210 { 2211 const struct nfit_set_info *map0 = m0; 2212 const struct nfit_set_info *map1 = m1; 2213 2214 return memcmp(&map0->region_offset, &map1->region_offset, 2215 sizeof(u64)); 2216 } 2217 2218 static int cmp_map(const void *m0, const void *m1) 2219 { 2220 const struct nfit_set_info *map0 = m0; 2221 const struct nfit_set_info *map1 = m1; 2222 2223 if (map0->region_offset < map1->region_offset) 2224 return -1; 2225 else if (map0->region_offset > map1->region_offset) 2226 return 1; 2227 return 0; 2228 } 2229 2230 static int cmp_map2(const void *m0, const void *m1) 2231 { 2232 const struct nfit_set_info2 *map0 = m0; 2233 const struct nfit_set_info2 *map1 = m1; 2234 2235 if (map0->region_offset < map1->region_offset) 2236 return -1; 2237 else if (map0->region_offset > map1->region_offset) 2238 return 1; 2239 return 0; 2240 } 2241 2242 /* Retrieve the nth entry referencing this spa */ 2243 static struct acpi_nfit_memory_map *memdev_from_spa( 2244 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n) 2245 { 2246 struct nfit_memdev *nfit_memdev; 2247 2248 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) 2249 if (nfit_memdev->memdev->range_index == range_index) 2250 if (n-- == 0) 2251 return nfit_memdev->memdev; 2252 return NULL; 2253 } 2254 2255 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc, 2256 struct nd_region_desc *ndr_desc, 2257 struct acpi_nfit_system_address *spa) 2258 { 2259 u16 nr = ndr_desc->num_mappings; 2260 struct nfit_set_info2 *info2 __free(kfree) = 2261 kcalloc(nr, sizeof(*info2), GFP_KERNEL); 2262 struct nfit_set_info *info __free(kfree) = 2263 kcalloc(nr, sizeof(*info), GFP_KERNEL); 2264 struct device *dev = acpi_desc->dev; 2265 struct nd_interleave_set *nd_set; 2266 int i; 2267 2268 if (!info || !info2) 2269 return -ENOMEM; 2270 2271 nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL); 2272 if (!nd_set) 2273 return -ENOMEM; 2274 import_guid(&nd_set->type_guid, spa->range_guid); 2275 2276 for (i = 0; i < nr; i++) { 2277 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i]; 2278 struct nvdimm *nvdimm = mapping->nvdimm; 2279 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 2280 struct nfit_set_info *map = &info[i]; 2281 struct nfit_set_info2 *map2 = &info2[i]; 2282 struct acpi_nfit_memory_map *memdev = 2283 memdev_from_spa(acpi_desc, spa->range_index, i); 2284 struct acpi_nfit_control_region *dcr = nfit_mem->dcr; 2285 2286 if (!memdev || !nfit_mem->dcr) { 2287 dev_err(dev, "%s: failed to find DCR\n", __func__); 2288 return -ENODEV; 2289 } 2290 2291 map->region_offset = memdev->region_offset; 2292 map->serial_number = dcr->serial_number; 2293 2294 map2->region_offset = memdev->region_offset; 2295 map2->serial_number = dcr->serial_number; 2296 map2->vendor_id = dcr->vendor_id; 2297 map2->manufacturing_date = dcr->manufacturing_date; 2298 map2->manufacturing_location = dcr->manufacturing_location; 2299 } 2300 2301 /* v1.1 namespaces */ 2302 sort(info, nr, sizeof(*info), cmp_map, NULL); 2303 nd_set->cookie1 = nd_fletcher64(info, sizeof(*info) * nr, 0); 2304 2305 /* v1.2 namespaces */ 2306 sort(info2, nr, sizeof(*info2), cmp_map2, NULL); 2307 nd_set->cookie2 = nd_fletcher64(info2, sizeof(*info2) * nr, 0); 2308 2309 /* support v1.1 namespaces created with the wrong sort order */ 2310 sort(info, nr, sizeof(*info), cmp_map_compat, NULL); 2311 nd_set->altcookie = nd_fletcher64(info, sizeof(*info) * nr, 0); 2312 2313 /* record the result of the sort for the mapping position */ 2314 for (i = 0; i < nr; i++) { 2315 struct nfit_set_info2 *map2 = &info2[i]; 2316 int j; 2317 2318 for (j = 0; j < nr; j++) { 2319 struct nd_mapping_desc *mapping = &ndr_desc->mapping[j]; 2320 struct nvdimm *nvdimm = mapping->nvdimm; 2321 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 2322 struct acpi_nfit_control_region *dcr = nfit_mem->dcr; 2323 2324 if (map2->serial_number == dcr->serial_number && 2325 map2->vendor_id == dcr->vendor_id && 2326 map2->manufacturing_date == dcr->manufacturing_date && 2327 map2->manufacturing_location 2328 == dcr->manufacturing_location) { 2329 mapping->position = i; 2330 break; 2331 } 2332 } 2333 } 2334 2335 ndr_desc->nd_set = nd_set; 2336 2337 return 0; 2338 } 2339 2340 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc, 2341 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa) 2342 { 2343 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc; 2344 struct acpi_nfit_system_address *spa = nfit_spa->spa; 2345 int cmd_rc, rc; 2346 2347 cmd->address = spa->address; 2348 cmd->length = spa->length; 2349 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd, 2350 sizeof(*cmd), &cmd_rc); 2351 if (rc < 0) 2352 return rc; 2353 return cmd_rc; 2354 } 2355 2356 static int ars_start(struct acpi_nfit_desc *acpi_desc, 2357 struct nfit_spa *nfit_spa, enum nfit_ars_state req_type) 2358 { 2359 int rc; 2360 int cmd_rc; 2361 struct nd_cmd_ars_start ars_start; 2362 struct acpi_nfit_system_address *spa = nfit_spa->spa; 2363 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc; 2364 2365 memset(&ars_start, 0, sizeof(ars_start)); 2366 ars_start.address = spa->address; 2367 ars_start.length = spa->length; 2368 if (req_type == ARS_REQ_SHORT) 2369 ars_start.flags = ND_ARS_RETURN_PREV_DATA; 2370 if (nfit_spa_type(spa) == NFIT_SPA_PM) 2371 ars_start.type = ND_ARS_PERSISTENT; 2372 else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE) 2373 ars_start.type = ND_ARS_VOLATILE; 2374 else 2375 return -ENOTTY; 2376 2377 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start, 2378 sizeof(ars_start), &cmd_rc); 2379 2380 if (rc < 0) 2381 return rc; 2382 if (cmd_rc < 0) 2383 return cmd_rc; 2384 set_bit(ARS_VALID, &acpi_desc->scrub_flags); 2385 return 0; 2386 } 2387 2388 static int ars_continue(struct acpi_nfit_desc *acpi_desc) 2389 { 2390 int rc, cmd_rc; 2391 struct nd_cmd_ars_start ars_start; 2392 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc; 2393 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status; 2394 2395 ars_start = (struct nd_cmd_ars_start) { 2396 .address = ars_status->restart_address, 2397 .length = ars_status->restart_length, 2398 .type = ars_status->type, 2399 }; 2400 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start, 2401 sizeof(ars_start), &cmd_rc); 2402 if (rc < 0) 2403 return rc; 2404 return cmd_rc; 2405 } 2406 2407 static int ars_get_status(struct acpi_nfit_desc *acpi_desc) 2408 { 2409 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc; 2410 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status; 2411 int rc, cmd_rc; 2412 2413 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status, 2414 acpi_desc->max_ars, &cmd_rc); 2415 if (rc < 0) 2416 return rc; 2417 return cmd_rc; 2418 } 2419 2420 static void ars_complete(struct acpi_nfit_desc *acpi_desc, 2421 struct nfit_spa *nfit_spa) 2422 { 2423 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status; 2424 struct acpi_nfit_system_address *spa = nfit_spa->spa; 2425 struct nd_region *nd_region = nfit_spa->nd_region; 2426 struct device *dev; 2427 2428 lockdep_assert_held(&acpi_desc->init_mutex); 2429 /* 2430 * Only advance the ARS state for ARS runs initiated by the 2431 * kernel, ignore ARS results from BIOS initiated runs for scrub 2432 * completion tracking. 2433 */ 2434 if (acpi_desc->scrub_spa != nfit_spa) 2435 return; 2436 2437 if ((ars_status->address >= spa->address && ars_status->address 2438 < spa->address + spa->length) 2439 || (ars_status->address < spa->address)) { 2440 /* 2441 * Assume that if a scrub starts at an offset from the 2442 * start of nfit_spa that we are in the continuation 2443 * case. 2444 * 2445 * Otherwise, if the scrub covers the spa range, mark 2446 * any pending request complete. 2447 */ 2448 if (ars_status->address + ars_status->length 2449 >= spa->address + spa->length) 2450 /* complete */; 2451 else 2452 return; 2453 } else 2454 return; 2455 2456 acpi_desc->scrub_spa = NULL; 2457 if (nd_region) { 2458 dev = nd_region_dev(nd_region); 2459 nvdimm_region_notify(nd_region, NVDIMM_REVALIDATE_POISON); 2460 } else 2461 dev = acpi_desc->dev; 2462 dev_dbg(dev, "ARS: range %d complete\n", spa->range_index); 2463 } 2464 2465 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc) 2466 { 2467 struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus; 2468 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status; 2469 int rc; 2470 u32 i; 2471 2472 /* 2473 * First record starts at 44 byte offset from the start of the 2474 * payload. 2475 */ 2476 if (ars_status->out_length < 44) 2477 return 0; 2478 2479 /* 2480 * Ignore potentially stale results that are only refreshed 2481 * after a start-ARS event. 2482 */ 2483 if (!test_and_clear_bit(ARS_VALID, &acpi_desc->scrub_flags)) { 2484 dev_dbg(acpi_desc->dev, "skip %d stale records\n", 2485 ars_status->num_records); 2486 return 0; 2487 } 2488 2489 for (i = 0; i < ars_status->num_records; i++) { 2490 /* only process full records */ 2491 if (ars_status->out_length 2492 < 44 + sizeof(struct nd_ars_record) * (i + 1)) 2493 break; 2494 rc = nvdimm_bus_add_badrange(nvdimm_bus, 2495 ars_status->records[i].err_address, 2496 ars_status->records[i].length); 2497 if (rc) 2498 return rc; 2499 } 2500 if (i < ars_status->num_records) 2501 dev_warn(acpi_desc->dev, "detected truncated ars results\n"); 2502 2503 return 0; 2504 } 2505 2506 static void acpi_nfit_remove_resource(void *data) 2507 { 2508 struct resource *res = data; 2509 2510 remove_resource(res); 2511 } 2512 2513 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc, 2514 struct nd_region_desc *ndr_desc) 2515 { 2516 struct resource *res, *nd_res = ndr_desc->res; 2517 int is_pmem, ret; 2518 2519 /* No operation if the region is already registered as PMEM */ 2520 is_pmem = region_intersects(nd_res->start, resource_size(nd_res), 2521 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY); 2522 if (is_pmem == REGION_INTERSECTS) 2523 return 0; 2524 2525 res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL); 2526 if (!res) 2527 return -ENOMEM; 2528 2529 res->name = "Persistent Memory"; 2530 res->start = nd_res->start; 2531 res->end = nd_res->end; 2532 res->flags = IORESOURCE_MEM; 2533 res->desc = IORES_DESC_PERSISTENT_MEMORY; 2534 2535 ret = insert_resource(&iomem_resource, res); 2536 if (ret) 2537 return ret; 2538 2539 ret = devm_add_action_or_reset(acpi_desc->dev, 2540 acpi_nfit_remove_resource, 2541 res); 2542 if (ret) 2543 return ret; 2544 2545 return 0; 2546 } 2547 2548 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc, 2549 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc, 2550 struct acpi_nfit_memory_map *memdev, 2551 struct nfit_spa *nfit_spa) 2552 { 2553 struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, 2554 memdev->device_handle); 2555 struct acpi_nfit_system_address *spa = nfit_spa->spa; 2556 2557 if (!nvdimm) { 2558 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n", 2559 spa->range_index, memdev->device_handle); 2560 return -ENODEV; 2561 } 2562 2563 mapping->nvdimm = nvdimm; 2564 switch (nfit_spa_type(spa)) { 2565 case NFIT_SPA_PM: 2566 case NFIT_SPA_VOLATILE: 2567 mapping->start = memdev->address; 2568 mapping->size = memdev->region_size; 2569 break; 2570 } 2571 2572 return 0; 2573 } 2574 2575 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa) 2576 { 2577 return (nfit_spa_type(spa) == NFIT_SPA_VDISK || 2578 nfit_spa_type(spa) == NFIT_SPA_VCD || 2579 nfit_spa_type(spa) == NFIT_SPA_PDISK || 2580 nfit_spa_type(spa) == NFIT_SPA_PCD); 2581 } 2582 2583 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa) 2584 { 2585 return (nfit_spa_type(spa) == NFIT_SPA_VDISK || 2586 nfit_spa_type(spa) == NFIT_SPA_VCD || 2587 nfit_spa_type(spa) == NFIT_SPA_VOLATILE); 2588 } 2589 2590 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc, 2591 struct nfit_spa *nfit_spa) 2592 { 2593 static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS]; 2594 struct acpi_nfit_system_address *spa = nfit_spa->spa; 2595 struct nd_region_desc *ndr_desc, _ndr_desc; 2596 struct nfit_memdev *nfit_memdev; 2597 struct nvdimm_bus *nvdimm_bus; 2598 struct resource res; 2599 int count = 0, rc; 2600 2601 if (nfit_spa->nd_region) 2602 return 0; 2603 2604 if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) { 2605 dev_dbg(acpi_desc->dev, "detected invalid spa index\n"); 2606 return 0; 2607 } 2608 2609 memset(&res, 0, sizeof(res)); 2610 memset(&mappings, 0, sizeof(mappings)); 2611 memset(&_ndr_desc, 0, sizeof(_ndr_desc)); 2612 res.start = spa->address; 2613 res.end = res.start + spa->length - 1; 2614 ndr_desc = &_ndr_desc; 2615 ndr_desc->res = &res; 2616 ndr_desc->provider_data = nfit_spa; 2617 ndr_desc->attr_groups = acpi_nfit_region_attribute_groups; 2618 if (spa->flags & ACPI_NFIT_PROXIMITY_VALID) { 2619 ndr_desc->numa_node = pxm_to_online_node(spa->proximity_domain); 2620 ndr_desc->target_node = pxm_to_node(spa->proximity_domain); 2621 } else { 2622 ndr_desc->numa_node = NUMA_NO_NODE; 2623 ndr_desc->target_node = NUMA_NO_NODE; 2624 } 2625 2626 /* Fallback to address based numa information if node lookup failed */ 2627 if (ndr_desc->numa_node == NUMA_NO_NODE) { 2628 ndr_desc->numa_node = memory_add_physaddr_to_nid(spa->address); 2629 dev_info(acpi_desc->dev, "changing numa node from %d to %d for nfit region [%pa-%pa]", 2630 NUMA_NO_NODE, ndr_desc->numa_node, &res.start, &res.end); 2631 } 2632 if (ndr_desc->target_node == NUMA_NO_NODE) { 2633 ndr_desc->target_node = phys_to_target_node(spa->address); 2634 dev_info(acpi_desc->dev, "changing target node from %d to %d for nfit region [%pa-%pa]", 2635 NUMA_NO_NODE, ndr_desc->numa_node, &res.start, &res.end); 2636 } 2637 2638 /* 2639 * Persistence domain bits are hierarchical, if 2640 * ACPI_NFIT_CAPABILITY_CACHE_FLUSH is set then 2641 * ACPI_NFIT_CAPABILITY_MEM_FLUSH is implied. 2642 */ 2643 if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_CACHE_FLUSH) 2644 set_bit(ND_REGION_PERSIST_CACHE, &ndr_desc->flags); 2645 else if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_MEM_FLUSH) 2646 set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc->flags); 2647 2648 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) { 2649 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev; 2650 struct nd_mapping_desc *mapping; 2651 2652 /* range index 0 == unmapped in SPA or invalid-SPA */ 2653 if (memdev->range_index == 0 || spa->range_index == 0) 2654 continue; 2655 if (memdev->range_index != spa->range_index) 2656 continue; 2657 if (count >= ND_MAX_MAPPINGS) { 2658 dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n", 2659 spa->range_index, ND_MAX_MAPPINGS); 2660 return -ENXIO; 2661 } 2662 mapping = &mappings[count++]; 2663 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc, 2664 memdev, nfit_spa); 2665 if (rc) 2666 goto out; 2667 } 2668 2669 ndr_desc->mapping = mappings; 2670 ndr_desc->num_mappings = count; 2671 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa); 2672 if (rc) 2673 goto out; 2674 2675 nvdimm_bus = acpi_desc->nvdimm_bus; 2676 if (nfit_spa_type(spa) == NFIT_SPA_PM) { 2677 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc); 2678 if (rc) { 2679 dev_warn(acpi_desc->dev, 2680 "failed to insert pmem resource to iomem: %d\n", 2681 rc); 2682 goto out; 2683 } 2684 2685 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus, 2686 ndr_desc); 2687 if (!nfit_spa->nd_region) 2688 rc = -ENOMEM; 2689 } else if (nfit_spa_is_volatile(spa)) { 2690 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus, 2691 ndr_desc); 2692 if (!nfit_spa->nd_region) 2693 rc = -ENOMEM; 2694 } else if (nfit_spa_is_virtual(spa)) { 2695 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus, 2696 ndr_desc); 2697 if (!nfit_spa->nd_region) 2698 rc = -ENOMEM; 2699 } 2700 2701 out: 2702 if (rc) 2703 dev_err(acpi_desc->dev, "failed to register spa range %d\n", 2704 nfit_spa->spa->range_index); 2705 return rc; 2706 } 2707 2708 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc) 2709 { 2710 struct device *dev = acpi_desc->dev; 2711 struct nd_cmd_ars_status *ars_status; 2712 2713 if (acpi_desc->ars_status) { 2714 memset(acpi_desc->ars_status, 0, acpi_desc->max_ars); 2715 return 0; 2716 } 2717 2718 ars_status = devm_kzalloc(dev, acpi_desc->max_ars, GFP_KERNEL); 2719 if (!ars_status) 2720 return -ENOMEM; 2721 acpi_desc->ars_status = ars_status; 2722 return 0; 2723 } 2724 2725 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc) 2726 { 2727 int rc; 2728 2729 if (ars_status_alloc(acpi_desc)) 2730 return -ENOMEM; 2731 2732 rc = ars_get_status(acpi_desc); 2733 2734 if (rc < 0 && rc != -ENOSPC) 2735 return rc; 2736 2737 if (ars_status_process_records(acpi_desc)) 2738 dev_err(acpi_desc->dev, "Failed to process ARS records\n"); 2739 2740 return rc; 2741 } 2742 2743 static int ars_register(struct acpi_nfit_desc *acpi_desc, 2744 struct nfit_spa *nfit_spa) 2745 { 2746 int rc; 2747 2748 if (test_bit(ARS_FAILED, &nfit_spa->ars_state)) 2749 return acpi_nfit_register_region(acpi_desc, nfit_spa); 2750 2751 set_bit(ARS_REQ_SHORT, &nfit_spa->ars_state); 2752 if (!no_init_ars) 2753 set_bit(ARS_REQ_LONG, &nfit_spa->ars_state); 2754 2755 switch (acpi_nfit_query_poison(acpi_desc)) { 2756 case 0: 2757 case -ENOSPC: 2758 case -EAGAIN: 2759 rc = ars_start(acpi_desc, nfit_spa, ARS_REQ_SHORT); 2760 /* shouldn't happen, try again later */ 2761 if (rc == -EBUSY) 2762 break; 2763 if (rc) { 2764 set_bit(ARS_FAILED, &nfit_spa->ars_state); 2765 break; 2766 } 2767 clear_bit(ARS_REQ_SHORT, &nfit_spa->ars_state); 2768 rc = acpi_nfit_query_poison(acpi_desc); 2769 if (rc) 2770 break; 2771 acpi_desc->scrub_spa = nfit_spa; 2772 ars_complete(acpi_desc, nfit_spa); 2773 /* 2774 * If ars_complete() says we didn't complete the 2775 * short scrub, we'll try again with a long 2776 * request. 2777 */ 2778 acpi_desc->scrub_spa = NULL; 2779 break; 2780 case -EBUSY: 2781 case -ENOMEM: 2782 /* 2783 * BIOS was using ARS, wait for it to complete (or 2784 * resources to become available) and then perform our 2785 * own scrubs. 2786 */ 2787 break; 2788 default: 2789 set_bit(ARS_FAILED, &nfit_spa->ars_state); 2790 break; 2791 } 2792 2793 return acpi_nfit_register_region(acpi_desc, nfit_spa); 2794 } 2795 2796 static void ars_complete_all(struct acpi_nfit_desc *acpi_desc) 2797 { 2798 struct nfit_spa *nfit_spa; 2799 2800 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) { 2801 if (test_bit(ARS_FAILED, &nfit_spa->ars_state)) 2802 continue; 2803 ars_complete(acpi_desc, nfit_spa); 2804 } 2805 } 2806 2807 static unsigned int __acpi_nfit_scrub(struct acpi_nfit_desc *acpi_desc, 2808 int query_rc) 2809 { 2810 unsigned int tmo = acpi_desc->scrub_tmo; 2811 struct device *dev = acpi_desc->dev; 2812 struct nfit_spa *nfit_spa; 2813 2814 lockdep_assert_held(&acpi_desc->init_mutex); 2815 2816 if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags)) 2817 return 0; 2818 2819 if (query_rc == -EBUSY) { 2820 dev_dbg(dev, "ARS: ARS busy\n"); 2821 return min(30U * 60U, tmo * 2); 2822 } 2823 if (query_rc == -ENOSPC) { 2824 dev_dbg(dev, "ARS: ARS continue\n"); 2825 ars_continue(acpi_desc); 2826 return 1; 2827 } 2828 if (query_rc && query_rc != -EAGAIN) { 2829 unsigned long long addr, end; 2830 2831 addr = acpi_desc->ars_status->address; 2832 end = addr + acpi_desc->ars_status->length; 2833 dev_dbg(dev, "ARS: %llx-%llx failed (%d)\n", addr, end, 2834 query_rc); 2835 } 2836 2837 ars_complete_all(acpi_desc); 2838 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) { 2839 enum nfit_ars_state req_type; 2840 int rc; 2841 2842 if (test_bit(ARS_FAILED, &nfit_spa->ars_state)) 2843 continue; 2844 2845 /* prefer short ARS requests first */ 2846 if (test_bit(ARS_REQ_SHORT, &nfit_spa->ars_state)) 2847 req_type = ARS_REQ_SHORT; 2848 else if (test_bit(ARS_REQ_LONG, &nfit_spa->ars_state)) 2849 req_type = ARS_REQ_LONG; 2850 else 2851 continue; 2852 rc = ars_start(acpi_desc, nfit_spa, req_type); 2853 2854 dev = nd_region_dev(nfit_spa->nd_region); 2855 dev_dbg(dev, "ARS: range %d ARS start %s (%d)\n", 2856 nfit_spa->spa->range_index, 2857 req_type == ARS_REQ_SHORT ? "short" : "long", 2858 rc); 2859 /* 2860 * Hmm, we raced someone else starting ARS? Try again in 2861 * a bit. 2862 */ 2863 if (rc == -EBUSY) 2864 return 1; 2865 if (rc == 0) { 2866 dev_WARN_ONCE(dev, acpi_desc->scrub_spa, 2867 "scrub start while range %d active\n", 2868 acpi_desc->scrub_spa->spa->range_index); 2869 clear_bit(req_type, &nfit_spa->ars_state); 2870 acpi_desc->scrub_spa = nfit_spa; 2871 /* 2872 * Consider this spa last for future scrub 2873 * requests 2874 */ 2875 list_move_tail(&nfit_spa->list, &acpi_desc->spas); 2876 return 1; 2877 } 2878 2879 dev_err(dev, "ARS: range %d ARS failed (%d)\n", 2880 nfit_spa->spa->range_index, rc); 2881 set_bit(ARS_FAILED, &nfit_spa->ars_state); 2882 } 2883 return 0; 2884 } 2885 2886 static void __sched_ars(struct acpi_nfit_desc *acpi_desc, unsigned int tmo) 2887 { 2888 lockdep_assert_held(&acpi_desc->init_mutex); 2889 2890 set_bit(ARS_BUSY, &acpi_desc->scrub_flags); 2891 /* note this should only be set from within the workqueue */ 2892 if (tmo) 2893 acpi_desc->scrub_tmo = tmo; 2894 queue_delayed_work(nfit_wq, &acpi_desc->dwork, tmo * HZ); 2895 } 2896 2897 static void sched_ars(struct acpi_nfit_desc *acpi_desc) 2898 { 2899 __sched_ars(acpi_desc, 0); 2900 } 2901 2902 static void notify_ars_done(struct acpi_nfit_desc *acpi_desc) 2903 { 2904 lockdep_assert_held(&acpi_desc->init_mutex); 2905 2906 clear_bit(ARS_BUSY, &acpi_desc->scrub_flags); 2907 acpi_desc->scrub_count++; 2908 if (acpi_desc->scrub_count_state) 2909 sysfs_notify_dirent(acpi_desc->scrub_count_state); 2910 } 2911 2912 static void acpi_nfit_scrub(struct work_struct *work) 2913 { 2914 struct acpi_nfit_desc *acpi_desc; 2915 unsigned int tmo; 2916 int query_rc; 2917 2918 acpi_desc = container_of(work, typeof(*acpi_desc), dwork.work); 2919 mutex_lock(&acpi_desc->init_mutex); 2920 query_rc = acpi_nfit_query_poison(acpi_desc); 2921 tmo = __acpi_nfit_scrub(acpi_desc, query_rc); 2922 if (tmo) 2923 __sched_ars(acpi_desc, tmo); 2924 else 2925 notify_ars_done(acpi_desc); 2926 memset(acpi_desc->ars_status, 0, acpi_desc->max_ars); 2927 clear_bit(ARS_POLL, &acpi_desc->scrub_flags); 2928 mutex_unlock(&acpi_desc->init_mutex); 2929 } 2930 2931 static void acpi_nfit_init_ars(struct acpi_nfit_desc *acpi_desc, 2932 struct nfit_spa *nfit_spa) 2933 { 2934 int type = nfit_spa_type(nfit_spa->spa); 2935 struct nd_cmd_ars_cap ars_cap; 2936 int rc; 2937 2938 set_bit(ARS_FAILED, &nfit_spa->ars_state); 2939 memset(&ars_cap, 0, sizeof(ars_cap)); 2940 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa); 2941 if (rc < 0) 2942 return; 2943 /* check that the supported scrub types match the spa type */ 2944 if (type == NFIT_SPA_VOLATILE && ((ars_cap.status >> 16) 2945 & ND_ARS_VOLATILE) == 0) 2946 return; 2947 if (type == NFIT_SPA_PM && ((ars_cap.status >> 16) 2948 & ND_ARS_PERSISTENT) == 0) 2949 return; 2950 2951 nfit_spa->max_ars = ars_cap.max_ars_out; 2952 nfit_spa->clear_err_unit = ars_cap.clear_err_unit; 2953 acpi_desc->max_ars = max(nfit_spa->max_ars, acpi_desc->max_ars); 2954 clear_bit(ARS_FAILED, &nfit_spa->ars_state); 2955 } 2956 2957 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc) 2958 { 2959 struct nfit_spa *nfit_spa; 2960 int rc, do_sched_ars = 0; 2961 2962 set_bit(ARS_VALID, &acpi_desc->scrub_flags); 2963 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) { 2964 switch (nfit_spa_type(nfit_spa->spa)) { 2965 case NFIT_SPA_VOLATILE: 2966 case NFIT_SPA_PM: 2967 acpi_nfit_init_ars(acpi_desc, nfit_spa); 2968 break; 2969 } 2970 } 2971 2972 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) { 2973 switch (nfit_spa_type(nfit_spa->spa)) { 2974 case NFIT_SPA_VOLATILE: 2975 case NFIT_SPA_PM: 2976 /* register regions and kick off initial ARS run */ 2977 rc = ars_register(acpi_desc, nfit_spa); 2978 if (rc) 2979 return rc; 2980 2981 /* 2982 * Kick off background ARS if at least one 2983 * region successfully registered ARS 2984 */ 2985 if (!test_bit(ARS_FAILED, &nfit_spa->ars_state)) 2986 do_sched_ars++; 2987 break; 2988 case NFIT_SPA_BDW: 2989 /* nothing to register */ 2990 break; 2991 case NFIT_SPA_DCR: 2992 case NFIT_SPA_VDISK: 2993 case NFIT_SPA_VCD: 2994 case NFIT_SPA_PDISK: 2995 case NFIT_SPA_PCD: 2996 /* register known regions that don't support ARS */ 2997 rc = acpi_nfit_register_region(acpi_desc, nfit_spa); 2998 if (rc) 2999 return rc; 3000 break; 3001 default: 3002 /* don't register unknown regions */ 3003 break; 3004 } 3005 } 3006 3007 if (do_sched_ars) 3008 sched_ars(acpi_desc); 3009 return 0; 3010 } 3011 3012 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc, 3013 struct nfit_table_prev *prev) 3014 { 3015 struct device *dev = acpi_desc->dev; 3016 3017 if (!list_empty(&prev->spas) || 3018 !list_empty(&prev->memdevs) || 3019 !list_empty(&prev->dcrs) || 3020 !list_empty(&prev->bdws) || 3021 !list_empty(&prev->idts) || 3022 !list_empty(&prev->flushes)) { 3023 dev_err(dev, "new nfit deletes entries (unsupported)\n"); 3024 return -ENXIO; 3025 } 3026 return 0; 3027 } 3028 3029 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc) 3030 { 3031 struct device *dev = acpi_desc->dev; 3032 struct kernfs_node *nfit; 3033 struct device *bus_dev; 3034 3035 if (!ars_supported(acpi_desc->nvdimm_bus)) 3036 return 0; 3037 3038 bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus); 3039 nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit"); 3040 if (!nfit) { 3041 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n"); 3042 return -ENODEV; 3043 } 3044 acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub"); 3045 sysfs_put(nfit); 3046 if (!acpi_desc->scrub_count_state) { 3047 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n"); 3048 return -ENODEV; 3049 } 3050 3051 return 0; 3052 } 3053 3054 static void acpi_nfit_unregister(void *data) 3055 { 3056 struct acpi_nfit_desc *acpi_desc = data; 3057 3058 nvdimm_bus_unregister(acpi_desc->nvdimm_bus); 3059 } 3060 3061 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz) 3062 { 3063 struct device *dev = acpi_desc->dev; 3064 struct nfit_table_prev prev; 3065 const void *end; 3066 int rc; 3067 3068 if (!acpi_desc->nvdimm_bus) { 3069 acpi_nfit_init_dsms(acpi_desc); 3070 3071 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev, 3072 &acpi_desc->nd_desc); 3073 if (!acpi_desc->nvdimm_bus) 3074 return -ENOMEM; 3075 3076 rc = devm_add_action_or_reset(dev, acpi_nfit_unregister, 3077 acpi_desc); 3078 if (rc) 3079 return rc; 3080 3081 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc); 3082 if (rc) 3083 return rc; 3084 3085 /* register this acpi_desc for mce notifications */ 3086 mutex_lock(&acpi_desc_lock); 3087 list_add_tail(&acpi_desc->list, &acpi_descs); 3088 mutex_unlock(&acpi_desc_lock); 3089 } 3090 3091 mutex_lock(&acpi_desc->init_mutex); 3092 3093 INIT_LIST_HEAD(&prev.spas); 3094 INIT_LIST_HEAD(&prev.memdevs); 3095 INIT_LIST_HEAD(&prev.dcrs); 3096 INIT_LIST_HEAD(&prev.bdws); 3097 INIT_LIST_HEAD(&prev.idts); 3098 INIT_LIST_HEAD(&prev.flushes); 3099 3100 list_cut_position(&prev.spas, &acpi_desc->spas, 3101 acpi_desc->spas.prev); 3102 list_cut_position(&prev.memdevs, &acpi_desc->memdevs, 3103 acpi_desc->memdevs.prev); 3104 list_cut_position(&prev.dcrs, &acpi_desc->dcrs, 3105 acpi_desc->dcrs.prev); 3106 list_cut_position(&prev.bdws, &acpi_desc->bdws, 3107 acpi_desc->bdws.prev); 3108 list_cut_position(&prev.idts, &acpi_desc->idts, 3109 acpi_desc->idts.prev); 3110 list_cut_position(&prev.flushes, &acpi_desc->flushes, 3111 acpi_desc->flushes.prev); 3112 3113 end = data + sz; 3114 while (!IS_ERR_OR_NULL(data)) 3115 data = add_table(acpi_desc, &prev, data, end); 3116 3117 if (IS_ERR(data)) { 3118 dev_dbg(dev, "nfit table parsing error: %ld\n", PTR_ERR(data)); 3119 rc = PTR_ERR(data); 3120 goto out_unlock; 3121 } 3122 3123 rc = acpi_nfit_check_deletions(acpi_desc, &prev); 3124 if (rc) 3125 goto out_unlock; 3126 3127 rc = nfit_mem_init(acpi_desc); 3128 if (rc) 3129 goto out_unlock; 3130 3131 rc = acpi_nfit_register_dimms(acpi_desc); 3132 if (rc) 3133 goto out_unlock; 3134 3135 rc = acpi_nfit_register_regions(acpi_desc); 3136 3137 out_unlock: 3138 mutex_unlock(&acpi_desc->init_mutex); 3139 return rc; 3140 } 3141 EXPORT_SYMBOL_GPL(acpi_nfit_init); 3142 3143 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc) 3144 { 3145 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc); 3146 struct device *dev = acpi_desc->dev; 3147 3148 /* Bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */ 3149 device_lock(dev); 3150 device_unlock(dev); 3151 3152 /* Bounce the init_mutex to complete initial registration */ 3153 mutex_lock(&acpi_desc->init_mutex); 3154 mutex_unlock(&acpi_desc->init_mutex); 3155 3156 return 0; 3157 } 3158 3159 static int __acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc, 3160 struct nvdimm *nvdimm, unsigned int cmd) 3161 { 3162 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc); 3163 3164 if (nvdimm) 3165 return 0; 3166 if (cmd != ND_CMD_ARS_START) 3167 return 0; 3168 3169 /* 3170 * The kernel and userspace may race to initiate a scrub, but 3171 * the scrub thread is prepared to lose that initial race. It 3172 * just needs guarantees that any ARS it initiates are not 3173 * interrupted by any intervening start requests from userspace. 3174 */ 3175 if (work_busy(&acpi_desc->dwork.work)) 3176 return -EBUSY; 3177 3178 return 0; 3179 } 3180 3181 /* 3182 * Prevent security and firmware activate commands from being issued via 3183 * ioctl. 3184 */ 3185 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc, 3186 struct nvdimm *nvdimm, unsigned int cmd, void *buf) 3187 { 3188 struct nd_cmd_pkg *call_pkg = buf; 3189 unsigned int func; 3190 3191 if (nvdimm && cmd == ND_CMD_CALL && 3192 call_pkg->nd_family == NVDIMM_FAMILY_INTEL) { 3193 func = call_pkg->nd_command; 3194 if (func > NVDIMM_CMD_MAX || 3195 (1 << func) & NVDIMM_INTEL_DENY_CMDMASK) 3196 return -EOPNOTSUPP; 3197 } 3198 3199 /* block all non-nfit bus commands */ 3200 if (!nvdimm && cmd == ND_CMD_CALL && 3201 call_pkg->nd_family != NVDIMM_BUS_FAMILY_NFIT) 3202 return -EOPNOTSUPP; 3203 3204 return __acpi_nfit_clear_to_send(nd_desc, nvdimm, cmd); 3205 } 3206 3207 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc, 3208 enum nfit_ars_state req_type) 3209 { 3210 struct device *dev = acpi_desc->dev; 3211 int scheduled = 0, busy = 0; 3212 struct nfit_spa *nfit_spa; 3213 3214 mutex_lock(&acpi_desc->init_mutex); 3215 if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags)) { 3216 mutex_unlock(&acpi_desc->init_mutex); 3217 return 0; 3218 } 3219 3220 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) { 3221 int type = nfit_spa_type(nfit_spa->spa); 3222 3223 if (type != NFIT_SPA_PM && type != NFIT_SPA_VOLATILE) 3224 continue; 3225 if (test_bit(ARS_FAILED, &nfit_spa->ars_state)) 3226 continue; 3227 3228 if (test_and_set_bit(req_type, &nfit_spa->ars_state)) 3229 busy++; 3230 else 3231 scheduled++; 3232 } 3233 if (scheduled) { 3234 sched_ars(acpi_desc); 3235 dev_dbg(dev, "ars_scan triggered\n"); 3236 } 3237 mutex_unlock(&acpi_desc->init_mutex); 3238 3239 if (scheduled) 3240 return 0; 3241 if (busy) 3242 return -EBUSY; 3243 return -ENOTTY; 3244 } 3245 3246 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev) 3247 { 3248 struct nvdimm_bus_descriptor *nd_desc; 3249 3250 dev_set_drvdata(dev, acpi_desc); 3251 acpi_desc->dev = dev; 3252 nd_desc = &acpi_desc->nd_desc; 3253 nd_desc->provider_name = "ACPI.NFIT"; 3254 nd_desc->module = THIS_MODULE; 3255 nd_desc->ndctl = acpi_nfit_ctl; 3256 nd_desc->flush_probe = acpi_nfit_flush_probe; 3257 nd_desc->clear_to_send = acpi_nfit_clear_to_send; 3258 nd_desc->attr_groups = acpi_nfit_attribute_groups; 3259 3260 INIT_LIST_HEAD(&acpi_desc->spas); 3261 INIT_LIST_HEAD(&acpi_desc->dcrs); 3262 INIT_LIST_HEAD(&acpi_desc->bdws); 3263 INIT_LIST_HEAD(&acpi_desc->idts); 3264 INIT_LIST_HEAD(&acpi_desc->flushes); 3265 INIT_LIST_HEAD(&acpi_desc->memdevs); 3266 INIT_LIST_HEAD(&acpi_desc->dimms); 3267 INIT_LIST_HEAD(&acpi_desc->list); 3268 mutex_init(&acpi_desc->init_mutex); 3269 acpi_desc->scrub_tmo = 1; 3270 INIT_DELAYED_WORK(&acpi_desc->dwork, acpi_nfit_scrub); 3271 } 3272 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init); 3273 3274 static void acpi_nfit_put_table(void *table) 3275 { 3276 acpi_put_table(table); 3277 } 3278 3279 static void acpi_nfit_notify(acpi_handle handle, u32 event, void *data) 3280 { 3281 struct acpi_device *adev = data; 3282 3283 device_lock(&adev->dev); 3284 __acpi_nfit_notify(&adev->dev, handle, event); 3285 device_unlock(&adev->dev); 3286 } 3287 3288 static void acpi_nfit_remove_notify_handler(void *data) 3289 { 3290 struct acpi_device *adev = data; 3291 3292 acpi_dev_remove_notify_handler(adev, ACPI_DEVICE_NOTIFY, 3293 acpi_nfit_notify); 3294 } 3295 3296 void acpi_nfit_shutdown(void *data) 3297 { 3298 struct acpi_nfit_desc *acpi_desc = data; 3299 struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus); 3300 3301 /* 3302 * Destruct under acpi_desc_lock so that nfit_handle_mce does not 3303 * race teardown 3304 */ 3305 mutex_lock(&acpi_desc_lock); 3306 list_del(&acpi_desc->list); 3307 mutex_unlock(&acpi_desc_lock); 3308 3309 mutex_lock(&acpi_desc->init_mutex); 3310 set_bit(ARS_CANCEL, &acpi_desc->scrub_flags); 3311 mutex_unlock(&acpi_desc->init_mutex); 3312 cancel_delayed_work_sync(&acpi_desc->dwork); 3313 3314 /* 3315 * Bounce the nvdimm bus lock to make sure any in-flight 3316 * acpi_nfit_ars_rescan() submissions have had a chance to 3317 * either submit or see ->cancel set. 3318 */ 3319 device_lock(bus_dev); 3320 device_unlock(bus_dev); 3321 3322 flush_workqueue(nfit_wq); 3323 } 3324 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown); 3325 3326 static int acpi_nfit_add(struct acpi_device *adev) 3327 { 3328 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL }; 3329 struct acpi_nfit_desc *acpi_desc; 3330 struct device *dev = &adev->dev; 3331 struct acpi_table_header *tbl; 3332 acpi_status status = AE_OK; 3333 acpi_size sz; 3334 int rc = 0; 3335 3336 rc = acpi_dev_install_notify_handler(adev, ACPI_DEVICE_NOTIFY, 3337 acpi_nfit_notify, adev); 3338 if (rc) 3339 return rc; 3340 3341 rc = devm_add_action_or_reset(dev, acpi_nfit_remove_notify_handler, 3342 adev); 3343 if (rc) 3344 return rc; 3345 3346 status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl); 3347 if (ACPI_FAILURE(status)) { 3348 /* The NVDIMM root device allows OS to trigger enumeration of 3349 * NVDIMMs through NFIT at boot time and re-enumeration at 3350 * root level via the _FIT method during runtime. 3351 * This is ok to return 0 here, we could have an nvdimm 3352 * hotplugged later and evaluate _FIT method which returns 3353 * data in the format of a series of NFIT Structures. 3354 */ 3355 dev_dbg(dev, "failed to find NFIT at startup\n"); 3356 return 0; 3357 } 3358 3359 rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl); 3360 if (rc) 3361 return rc; 3362 sz = tbl->length; 3363 3364 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL); 3365 if (!acpi_desc) 3366 return -ENOMEM; 3367 acpi_nfit_desc_init(acpi_desc, &adev->dev); 3368 3369 /* Save the acpi header for exporting the revision via sysfs */ 3370 acpi_desc->acpi_header = *tbl; 3371 3372 /* Evaluate _FIT and override with that if present */ 3373 status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf); 3374 if (ACPI_SUCCESS(status) && buf.length > 0) { 3375 union acpi_object *obj = buf.pointer; 3376 3377 if (obj->type == ACPI_TYPE_BUFFER) 3378 rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer, 3379 obj->buffer.length); 3380 else 3381 dev_dbg(dev, "invalid type %d, ignoring _FIT\n", 3382 (int) obj->type); 3383 kfree(buf.pointer); 3384 } else 3385 /* skip over the lead-in header table */ 3386 rc = acpi_nfit_init(acpi_desc, (void *) tbl 3387 + sizeof(struct acpi_table_nfit), 3388 sz - sizeof(struct acpi_table_nfit)); 3389 3390 if (rc) 3391 return rc; 3392 3393 return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc); 3394 } 3395 3396 static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle) 3397 { 3398 struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev); 3399 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL }; 3400 union acpi_object *obj; 3401 acpi_status status; 3402 int ret; 3403 3404 if (!dev->driver) { 3405 /* dev->driver may be null if we're being removed */ 3406 dev_dbg(dev, "no driver found for dev\n"); 3407 return; 3408 } 3409 3410 if (!acpi_desc) { 3411 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL); 3412 if (!acpi_desc) 3413 return; 3414 acpi_nfit_desc_init(acpi_desc, dev); 3415 } else { 3416 /* 3417 * Finish previous registration before considering new 3418 * regions. 3419 */ 3420 flush_workqueue(nfit_wq); 3421 } 3422 3423 /* Evaluate _FIT */ 3424 status = acpi_evaluate_object(handle, "_FIT", NULL, &buf); 3425 if (ACPI_FAILURE(status)) { 3426 dev_err(dev, "failed to evaluate _FIT\n"); 3427 return; 3428 } 3429 3430 obj = buf.pointer; 3431 if (obj->type == ACPI_TYPE_BUFFER) { 3432 ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer, 3433 obj->buffer.length); 3434 if (ret) 3435 dev_err(dev, "failed to merge updated NFIT\n"); 3436 } else 3437 dev_err(dev, "Invalid _FIT\n"); 3438 kfree(buf.pointer); 3439 } 3440 3441 static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle) 3442 { 3443 struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev); 3444 3445 if (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON) 3446 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG); 3447 else 3448 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_SHORT); 3449 } 3450 3451 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event) 3452 { 3453 dev_dbg(dev, "event: 0x%x\n", event); 3454 3455 switch (event) { 3456 case NFIT_NOTIFY_UPDATE: 3457 return acpi_nfit_update_notify(dev, handle); 3458 case NFIT_NOTIFY_UC_MEMORY_ERROR: 3459 return acpi_nfit_uc_error_notify(dev, handle); 3460 default: 3461 return; 3462 } 3463 } 3464 EXPORT_SYMBOL_GPL(__acpi_nfit_notify); 3465 3466 static const struct acpi_device_id acpi_nfit_ids[] = { 3467 { "ACPI0012", 0 }, 3468 { "", 0 }, 3469 }; 3470 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids); 3471 3472 static struct acpi_driver acpi_nfit_driver = { 3473 .name = KBUILD_MODNAME, 3474 .ids = acpi_nfit_ids, 3475 .ops = { 3476 .add = acpi_nfit_add, 3477 }, 3478 }; 3479 3480 static __init int nfit_init(void) 3481 { 3482 int ret; 3483 3484 BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40); 3485 BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 64); 3486 BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48); 3487 BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 16); 3488 BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 8); 3489 BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80); 3490 BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40); 3491 BUILD_BUG_ON(sizeof(struct acpi_nfit_capabilities) != 16); 3492 3493 guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]); 3494 guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]); 3495 guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]); 3496 guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]); 3497 guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]); 3498 guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]); 3499 guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]); 3500 guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]); 3501 guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]); 3502 guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]); 3503 guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]); 3504 guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]); 3505 guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]); 3506 guid_parse(UUID_NFIT_DIMM_N_HYPERV, &nfit_uuid[NFIT_DEV_DIMM_N_HYPERV]); 3507 guid_parse(UUID_INTEL_BUS, &nfit_uuid[NFIT_BUS_INTEL]); 3508 3509 nfit_wq = create_singlethread_workqueue("nfit"); 3510 if (!nfit_wq) 3511 return -ENOMEM; 3512 3513 nfit_mce_register(); 3514 ret = acpi_bus_register_driver(&acpi_nfit_driver); 3515 if (ret) { 3516 nfit_mce_unregister(); 3517 destroy_workqueue(nfit_wq); 3518 } 3519 3520 return ret; 3521 3522 } 3523 3524 static __exit void nfit_exit(void) 3525 { 3526 nfit_mce_unregister(); 3527 acpi_bus_unregister_driver(&acpi_nfit_driver); 3528 destroy_workqueue(nfit_wq); 3529 WARN_ON(!list_empty(&acpi_descs)); 3530 } 3531 3532 module_init(nfit_init); 3533 module_exit(nfit_exit); 3534 MODULE_LICENSE("GPL v2"); 3535 MODULE_AUTHOR("Intel Corporation"); 3536