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 if (!buf || buf_len < sizeof(*call_pkg)) 459 return -EINVAL; 460 461 call_pkg = buf; 462 } 463 464 func = cmd_to_func(nfit_mem, cmd, call_pkg, &family); 465 if (func < 0) 466 return func; 467 468 if (nvdimm) { 469 struct acpi_device *adev = nfit_mem->adev; 470 471 if (!adev) 472 return -ENOTTY; 473 474 dimm_name = nvdimm_name(nvdimm); 475 cmd_name = nvdimm_cmd_name(cmd); 476 cmd_mask = nvdimm_cmd_mask(nvdimm); 477 dsm_mask = nfit_mem->dsm_mask; 478 desc = nd_cmd_dimm_desc(cmd); 479 guid = to_nfit_uuid(nfit_mem->family); 480 handle = adev->handle; 481 } else { 482 struct acpi_device *adev = to_acpi_dev(acpi_desc); 483 484 cmd_name = nvdimm_bus_cmd_name(cmd); 485 cmd_mask = nd_desc->cmd_mask; 486 if (cmd == ND_CMD_CALL && call_pkg->nd_family) { 487 family = call_pkg->nd_family; 488 if (family > NVDIMM_BUS_FAMILY_MAX || 489 !test_bit(family, &nd_desc->bus_family_mask)) 490 return -EINVAL; 491 family = array_index_nospec(family, 492 NVDIMM_BUS_FAMILY_MAX + 1); 493 dsm_mask = acpi_desc->family_dsm_mask[family]; 494 guid = to_nfit_bus_uuid(family); 495 } else { 496 dsm_mask = acpi_desc->bus_dsm_mask; 497 guid = to_nfit_uuid(NFIT_DEV_BUS); 498 } 499 desc = nd_cmd_bus_desc(cmd); 500 handle = adev->handle; 501 dimm_name = "bus"; 502 } 503 504 if (!desc || (cmd && (desc->out_num + desc->in_num == 0))) 505 return -ENOTTY; 506 507 /* 508 * Check for a valid command. For ND_CMD_CALL, we also have to 509 * make sure that the DSM function is supported. 510 */ 511 if (cmd == ND_CMD_CALL && 512 (func > NVDIMM_CMD_MAX || !test_bit(func, &dsm_mask))) 513 return -ENOTTY; 514 else if (!test_bit(cmd, &cmd_mask)) 515 return -ENOTTY; 516 517 in_obj.type = ACPI_TYPE_PACKAGE; 518 in_obj.package.count = 1; 519 in_obj.package.elements = &in_buf; 520 in_buf.type = ACPI_TYPE_BUFFER; 521 in_buf.buffer.pointer = buf; 522 in_buf.buffer.length = 0; 523 524 /* libnvdimm has already validated the input envelope */ 525 for (i = 0; i < desc->in_num; i++) 526 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc, 527 i, buf); 528 529 if (call_pkg) { 530 /* skip over package wrapper */ 531 in_buf.buffer.pointer = (void *) &call_pkg->nd_payload; 532 in_buf.buffer.length = call_pkg->nd_size_in; 533 } 534 535 dev_dbg(dev, "%s cmd: %d: family: %d func: %d input length: %d\n", 536 dimm_name, cmd, family, func, in_buf.buffer.length); 537 if (payload_dumpable(nvdimm, func)) 538 print_hex_dump_debug("nvdimm in ", DUMP_PREFIX_OFFSET, 4, 4, 539 in_buf.buffer.pointer, 540 min_t(u32, 256, in_buf.buffer.length), true); 541 542 /* call the BIOS, prefer the named methods over _DSM if available */ 543 if (nvdimm && cmd == ND_CMD_GET_CONFIG_SIZE 544 && test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) 545 out_obj = acpi_label_info(handle); 546 else if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA 547 && test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) { 548 struct nd_cmd_get_config_data_hdr *p = buf; 549 550 out_obj = acpi_label_read(handle, p->in_offset, p->in_length); 551 } else if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA 552 && test_bit(NFIT_MEM_LSW, &nfit_mem->flags)) { 553 struct nd_cmd_set_config_hdr *p = buf; 554 555 out_obj = acpi_label_write(handle, p->in_offset, p->in_length, 556 p->in_buf); 557 } else { 558 u8 revid; 559 560 if (nvdimm) 561 revid = nfit_dsm_revid(nfit_mem->family, func); 562 else 563 revid = 1; 564 out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj); 565 } 566 567 if (!out_obj) { 568 dev_dbg(dev, "%s _DSM failed cmd: %s\n", dimm_name, cmd_name); 569 return -EINVAL; 570 } 571 572 if (out_obj->type != ACPI_TYPE_BUFFER) { 573 dev_dbg(dev, "%s unexpected output object type cmd: %s type: %d\n", 574 dimm_name, cmd_name, out_obj->type); 575 rc = -EINVAL; 576 goto out; 577 } 578 579 dev_dbg(dev, "%s cmd: %s output length: %d\n", dimm_name, 580 cmd_name, out_obj->buffer.length); 581 print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4, 4, 582 out_obj->buffer.pointer, 583 min_t(u32, 128, out_obj->buffer.length), true); 584 585 if (call_pkg) { 586 call_pkg->nd_fw_size = out_obj->buffer.length; 587 memcpy(call_pkg->nd_payload + call_pkg->nd_size_in, 588 out_obj->buffer.pointer, 589 min(call_pkg->nd_fw_size, call_pkg->nd_size_out)); 590 591 ACPI_FREE(out_obj); 592 /* 593 * Need to support FW function w/o known size in advance. 594 * Caller can determine required size based upon nd_fw_size. 595 * If we return an error (like elsewhere) then caller wouldn't 596 * be able to rely upon data returned to make calculation. 597 */ 598 if (cmd_rc) 599 *cmd_rc = 0; 600 return 0; 601 } 602 603 for (i = 0, offset = 0; i < desc->out_num; i++) { 604 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf, 605 (u32 *) out_obj->buffer.pointer, 606 out_obj->buffer.length - offset); 607 608 if (offset + out_size > out_obj->buffer.length) { 609 dev_dbg(dev, "%s output object underflow cmd: %s field: %d\n", 610 dimm_name, cmd_name, i); 611 break; 612 } 613 614 if (in_buf.buffer.length + offset + out_size > buf_len) { 615 dev_dbg(dev, "%s output overrun cmd: %s field: %d\n", 616 dimm_name, cmd_name, i); 617 rc = -ENXIO; 618 goto out; 619 } 620 memcpy(buf + in_buf.buffer.length + offset, 621 out_obj->buffer.pointer + offset, out_size); 622 offset += out_size; 623 } 624 625 /* 626 * Set fw_status for all the commands with a known format to be 627 * later interpreted by xlat_status(). 628 */ 629 if (i >= 1 && ((!nvdimm && cmd >= ND_CMD_ARS_CAP 630 && cmd <= ND_CMD_CLEAR_ERROR) 631 || (nvdimm && cmd >= ND_CMD_SMART 632 && cmd <= ND_CMD_VENDOR))) 633 fw_status = *(u32 *) out_obj->buffer.pointer; 634 635 if (offset + in_buf.buffer.length < buf_len) { 636 if (i >= 1) { 637 /* 638 * status valid, return the number of bytes left 639 * unfilled in the output buffer 640 */ 641 rc = buf_len - offset - in_buf.buffer.length; 642 if (cmd_rc) 643 *cmd_rc = xlat_status(nvdimm, buf, cmd, 644 fw_status); 645 } else { 646 dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n", 647 __func__, dimm_name, cmd_name, buf_len, 648 offset); 649 rc = -ENXIO; 650 } 651 } else { 652 rc = 0; 653 if (cmd_rc) 654 *cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status); 655 } 656 657 out: 658 ACPI_FREE(out_obj); 659 660 return rc; 661 } 662 EXPORT_SYMBOL_GPL(acpi_nfit_ctl); 663 664 static const char *spa_type_name(u16 type) 665 { 666 static const char *to_name[] = { 667 [NFIT_SPA_VOLATILE] = "volatile", 668 [NFIT_SPA_PM] = "pmem", 669 [NFIT_SPA_DCR] = "dimm-control-region", 670 [NFIT_SPA_BDW] = "block-data-window", 671 [NFIT_SPA_VDISK] = "volatile-disk", 672 [NFIT_SPA_VCD] = "volatile-cd", 673 [NFIT_SPA_PDISK] = "persistent-disk", 674 [NFIT_SPA_PCD] = "persistent-cd", 675 676 }; 677 678 if (type > NFIT_SPA_PCD) 679 return "unknown"; 680 681 return to_name[type]; 682 } 683 684 int nfit_spa_type(struct acpi_nfit_system_address *spa) 685 { 686 guid_t guid; 687 int i; 688 689 import_guid(&guid, spa->range_guid); 690 for (i = 0; i < NFIT_UUID_MAX; i++) 691 if (guid_equal(to_nfit_uuid(i), &guid)) 692 return i; 693 return -1; 694 } 695 696 static size_t sizeof_spa(struct acpi_nfit_system_address *spa) 697 { 698 if (spa->flags & ACPI_NFIT_LOCATION_COOKIE_VALID) 699 return sizeof(*spa); 700 return sizeof(*spa) - 8; 701 } 702 703 static bool add_spa(struct acpi_nfit_desc *acpi_desc, 704 struct nfit_table_prev *prev, 705 struct acpi_nfit_system_address *spa) 706 { 707 struct device *dev = acpi_desc->dev; 708 struct nfit_spa *nfit_spa; 709 710 if (spa->header.length != sizeof_spa(spa)) 711 return false; 712 713 list_for_each_entry(nfit_spa, &prev->spas, list) { 714 if (memcmp(nfit_spa->spa, spa, sizeof_spa(spa)) == 0) { 715 list_move_tail(&nfit_spa->list, &acpi_desc->spas); 716 return true; 717 } 718 } 719 720 nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof_spa(spa), 721 GFP_KERNEL); 722 if (!nfit_spa) 723 return false; 724 INIT_LIST_HEAD(&nfit_spa->list); 725 memcpy(nfit_spa->spa, spa, sizeof_spa(spa)); 726 list_add_tail(&nfit_spa->list, &acpi_desc->spas); 727 dev_dbg(dev, "spa index: %d type: %s\n", 728 spa->range_index, 729 spa_type_name(nfit_spa_type(spa))); 730 return true; 731 } 732 733 static bool add_memdev(struct acpi_nfit_desc *acpi_desc, 734 struct nfit_table_prev *prev, 735 struct acpi_nfit_memory_map *memdev) 736 { 737 struct device *dev = acpi_desc->dev; 738 struct nfit_memdev *nfit_memdev; 739 740 if (memdev->header.length != sizeof(*memdev)) 741 return false; 742 743 list_for_each_entry(nfit_memdev, &prev->memdevs, list) 744 if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) { 745 list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs); 746 return true; 747 } 748 749 nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev), 750 GFP_KERNEL); 751 if (!nfit_memdev) 752 return false; 753 INIT_LIST_HEAD(&nfit_memdev->list); 754 memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev)); 755 list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs); 756 dev_dbg(dev, "memdev handle: %#x spa: %d dcr: %d flags: %#x\n", 757 memdev->device_handle, memdev->range_index, 758 memdev->region_index, memdev->flags); 759 return true; 760 } 761 762 int nfit_get_smbios_id(u32 device_handle, u16 *flags) 763 { 764 struct acpi_nfit_memory_map *memdev; 765 struct acpi_nfit_desc *acpi_desc; 766 struct nfit_mem *nfit_mem; 767 u16 physical_id; 768 769 mutex_lock(&acpi_desc_lock); 770 list_for_each_entry(acpi_desc, &acpi_descs, list) { 771 mutex_lock(&acpi_desc->init_mutex); 772 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) { 773 memdev = __to_nfit_memdev(nfit_mem); 774 if (memdev->device_handle == device_handle) { 775 *flags = memdev->flags; 776 physical_id = memdev->physical_id; 777 mutex_unlock(&acpi_desc->init_mutex); 778 mutex_unlock(&acpi_desc_lock); 779 return physical_id; 780 } 781 } 782 mutex_unlock(&acpi_desc->init_mutex); 783 } 784 mutex_unlock(&acpi_desc_lock); 785 786 return -ENODEV; 787 } 788 EXPORT_SYMBOL_GPL(nfit_get_smbios_id); 789 790 /* 791 * An implementation may provide a truncated control region if no block windows 792 * are defined. 793 */ 794 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr) 795 { 796 if (dcr->header.length < offsetof(struct acpi_nfit_control_region, 797 window_size)) 798 return 0; 799 if (dcr->windows) 800 return sizeof(*dcr); 801 return offsetof(struct acpi_nfit_control_region, window_size); 802 } 803 804 static bool add_dcr(struct acpi_nfit_desc *acpi_desc, 805 struct nfit_table_prev *prev, 806 struct acpi_nfit_control_region *dcr) 807 { 808 struct device *dev = acpi_desc->dev; 809 struct nfit_dcr *nfit_dcr; 810 811 if (!sizeof_dcr(dcr)) 812 return false; 813 814 list_for_each_entry(nfit_dcr, &prev->dcrs, list) 815 if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) { 816 list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs); 817 return true; 818 } 819 820 nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr), 821 GFP_KERNEL); 822 if (!nfit_dcr) 823 return false; 824 INIT_LIST_HEAD(&nfit_dcr->list); 825 memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)); 826 list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs); 827 dev_dbg(dev, "dcr index: %d windows: %d\n", 828 dcr->region_index, dcr->windows); 829 return true; 830 } 831 832 static bool add_bdw(struct acpi_nfit_desc *acpi_desc, 833 struct nfit_table_prev *prev, 834 struct acpi_nfit_data_region *bdw) 835 { 836 struct device *dev = acpi_desc->dev; 837 struct nfit_bdw *nfit_bdw; 838 839 if (bdw->header.length != sizeof(*bdw)) 840 return false; 841 list_for_each_entry(nfit_bdw, &prev->bdws, list) 842 if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) { 843 list_move_tail(&nfit_bdw->list, &acpi_desc->bdws); 844 return true; 845 } 846 847 nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw), 848 GFP_KERNEL); 849 if (!nfit_bdw) 850 return false; 851 INIT_LIST_HEAD(&nfit_bdw->list); 852 memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw)); 853 list_add_tail(&nfit_bdw->list, &acpi_desc->bdws); 854 dev_dbg(dev, "bdw dcr: %d windows: %d\n", 855 bdw->region_index, bdw->windows); 856 return true; 857 } 858 859 static size_t sizeof_idt(struct acpi_nfit_interleave *idt) 860 { 861 if (idt->header.length < sizeof(*idt)) 862 return 0; 863 return sizeof(*idt) + sizeof(u32) * idt->line_count; 864 } 865 866 static bool add_idt(struct acpi_nfit_desc *acpi_desc, 867 struct nfit_table_prev *prev, 868 struct acpi_nfit_interleave *idt) 869 { 870 struct device *dev = acpi_desc->dev; 871 struct nfit_idt *nfit_idt; 872 873 if (!sizeof_idt(idt)) 874 return false; 875 876 list_for_each_entry(nfit_idt, &prev->idts, list) { 877 if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt)) 878 continue; 879 880 if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) { 881 list_move_tail(&nfit_idt->list, &acpi_desc->idts); 882 return true; 883 } 884 } 885 886 nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt), 887 GFP_KERNEL); 888 if (!nfit_idt) 889 return false; 890 INIT_LIST_HEAD(&nfit_idt->list); 891 memcpy(nfit_idt->idt, idt, sizeof_idt(idt)); 892 list_add_tail(&nfit_idt->list, &acpi_desc->idts); 893 dev_dbg(dev, "idt index: %d num_lines: %d\n", 894 idt->interleave_index, idt->line_count); 895 return true; 896 } 897 898 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush) 899 { 900 if (flush->header.length < sizeof(*flush)) 901 return 0; 902 return struct_size(flush, hint_address, flush->hint_count); 903 } 904 905 static bool add_flush(struct acpi_nfit_desc *acpi_desc, 906 struct nfit_table_prev *prev, 907 struct acpi_nfit_flush_address *flush) 908 { 909 struct device *dev = acpi_desc->dev; 910 struct nfit_flush *nfit_flush; 911 912 if (!sizeof_flush(flush)) 913 return false; 914 915 list_for_each_entry(nfit_flush, &prev->flushes, list) { 916 if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush)) 917 continue; 918 919 if (memcmp(nfit_flush->flush, flush, 920 sizeof_flush(flush)) == 0) { 921 list_move_tail(&nfit_flush->list, &acpi_desc->flushes); 922 return true; 923 } 924 } 925 926 nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush) 927 + sizeof_flush(flush), GFP_KERNEL); 928 if (!nfit_flush) 929 return false; 930 INIT_LIST_HEAD(&nfit_flush->list); 931 memcpy(nfit_flush->flush, flush, sizeof_flush(flush)); 932 list_add_tail(&nfit_flush->list, &acpi_desc->flushes); 933 dev_dbg(dev, "nfit_flush handle: %d hint_count: %d\n", 934 flush->device_handle, flush->hint_count); 935 return true; 936 } 937 938 static bool add_platform_cap(struct acpi_nfit_desc *acpi_desc, 939 struct acpi_nfit_capabilities *pcap) 940 { 941 struct device *dev = acpi_desc->dev; 942 u32 mask; 943 944 mask = (1 << (pcap->highest_capability + 1)) - 1; 945 acpi_desc->platform_cap = pcap->capabilities & mask; 946 dev_dbg(dev, "cap: %#x\n", acpi_desc->platform_cap); 947 return true; 948 } 949 950 static void *add_table(struct acpi_nfit_desc *acpi_desc, 951 struct nfit_table_prev *prev, void *table, const void *end) 952 { 953 struct device *dev = acpi_desc->dev; 954 struct acpi_nfit_header *hdr; 955 void *err = ERR_PTR(-ENOMEM); 956 957 if (table >= end) 958 return NULL; 959 960 hdr = table; 961 if (!hdr->length) { 962 dev_warn(dev, "found a zero length table '%d' parsing nfit\n", 963 hdr->type); 964 return NULL; 965 } 966 967 switch (hdr->type) { 968 case ACPI_NFIT_TYPE_SYSTEM_ADDRESS: 969 if (!add_spa(acpi_desc, prev, table)) 970 return err; 971 break; 972 case ACPI_NFIT_TYPE_MEMORY_MAP: 973 if (!add_memdev(acpi_desc, prev, table)) 974 return err; 975 break; 976 case ACPI_NFIT_TYPE_CONTROL_REGION: 977 if (!add_dcr(acpi_desc, prev, table)) 978 return err; 979 break; 980 case ACPI_NFIT_TYPE_DATA_REGION: 981 if (!add_bdw(acpi_desc, prev, table)) 982 return err; 983 break; 984 case ACPI_NFIT_TYPE_INTERLEAVE: 985 if (!add_idt(acpi_desc, prev, table)) 986 return err; 987 break; 988 case ACPI_NFIT_TYPE_FLUSH_ADDRESS: 989 if (!add_flush(acpi_desc, prev, table)) 990 return err; 991 break; 992 case ACPI_NFIT_TYPE_SMBIOS: 993 dev_dbg(dev, "smbios\n"); 994 break; 995 case ACPI_NFIT_TYPE_CAPABILITIES: 996 if (!add_platform_cap(acpi_desc, table)) 997 return err; 998 break; 999 default: 1000 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type); 1001 break; 1002 } 1003 1004 return table + hdr->length; 1005 } 1006 1007 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc, 1008 struct acpi_nfit_system_address *spa) 1009 { 1010 struct nfit_mem *nfit_mem, *found; 1011 struct nfit_memdev *nfit_memdev; 1012 int type = spa ? nfit_spa_type(spa) : 0; 1013 1014 switch (type) { 1015 case NFIT_SPA_DCR: 1016 case NFIT_SPA_PM: 1017 break; 1018 default: 1019 if (spa) 1020 return 0; 1021 } 1022 1023 /* 1024 * This loop runs in two modes, when a dimm is mapped the loop 1025 * adds memdev associations to an existing dimm, or creates a 1026 * dimm. In the unmapped dimm case this loop sweeps for memdev 1027 * instances with an invalid / zero range_index and adds those 1028 * dimms without spa associations. 1029 */ 1030 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) { 1031 struct nfit_flush *nfit_flush; 1032 struct nfit_dcr *nfit_dcr; 1033 u32 device_handle; 1034 u16 dcr; 1035 1036 if (spa && nfit_memdev->memdev->range_index != spa->range_index) 1037 continue; 1038 if (!spa && nfit_memdev->memdev->range_index) 1039 continue; 1040 found = NULL; 1041 dcr = nfit_memdev->memdev->region_index; 1042 device_handle = nfit_memdev->memdev->device_handle; 1043 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) 1044 if (__to_nfit_memdev(nfit_mem)->device_handle 1045 == device_handle) { 1046 found = nfit_mem; 1047 break; 1048 } 1049 1050 if (found) 1051 nfit_mem = found; 1052 else { 1053 nfit_mem = devm_kzalloc(acpi_desc->dev, 1054 sizeof(*nfit_mem), GFP_KERNEL); 1055 if (!nfit_mem) 1056 return -ENOMEM; 1057 INIT_LIST_HEAD(&nfit_mem->list); 1058 nfit_mem->acpi_desc = acpi_desc; 1059 list_add(&nfit_mem->list, &acpi_desc->dimms); 1060 } 1061 1062 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) { 1063 if (nfit_dcr->dcr->region_index != dcr) 1064 continue; 1065 /* 1066 * Record the control region for the dimm. For 1067 * the ACPI 6.1 case, where there are separate 1068 * control regions for the pmem vs blk 1069 * interfaces, be sure to record the extended 1070 * blk details. 1071 */ 1072 if (!nfit_mem->dcr) 1073 nfit_mem->dcr = nfit_dcr->dcr; 1074 else if (nfit_mem->dcr->windows == 0 1075 && nfit_dcr->dcr->windows) 1076 nfit_mem->dcr = nfit_dcr->dcr; 1077 break; 1078 } 1079 1080 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) { 1081 struct acpi_nfit_flush_address *flush; 1082 u16 i; 1083 1084 if (nfit_flush->flush->device_handle != device_handle) 1085 continue; 1086 nfit_mem->nfit_flush = nfit_flush; 1087 flush = nfit_flush->flush; 1088 nfit_mem->flush_wpq = devm_kcalloc(acpi_desc->dev, 1089 flush->hint_count, 1090 sizeof(struct resource), 1091 GFP_KERNEL); 1092 if (!nfit_mem->flush_wpq) 1093 return -ENOMEM; 1094 for (i = 0; i < flush->hint_count; i++) { 1095 struct resource *res = &nfit_mem->flush_wpq[i]; 1096 1097 res->start = flush->hint_address[i]; 1098 res->end = res->start + 8 - 1; 1099 } 1100 break; 1101 } 1102 1103 if (dcr && !nfit_mem->dcr) { 1104 dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n", 1105 spa->range_index, dcr); 1106 return -ENODEV; 1107 } 1108 1109 if (type == NFIT_SPA_DCR) { 1110 struct nfit_idt *nfit_idt; 1111 u16 idt_idx; 1112 1113 /* multiple dimms may share a SPA when interleaved */ 1114 nfit_mem->spa_dcr = spa; 1115 nfit_mem->memdev_dcr = nfit_memdev->memdev; 1116 idt_idx = nfit_memdev->memdev->interleave_index; 1117 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) { 1118 if (nfit_idt->idt->interleave_index != idt_idx) 1119 continue; 1120 nfit_mem->idt_dcr = nfit_idt->idt; 1121 break; 1122 } 1123 } else if (type == NFIT_SPA_PM) { 1124 /* 1125 * A single dimm may belong to multiple SPA-PM 1126 * ranges, record at least one in addition to 1127 * any SPA-DCR range. 1128 */ 1129 nfit_mem->memdev_pmem = nfit_memdev->memdev; 1130 } else 1131 nfit_mem->memdev_dcr = nfit_memdev->memdev; 1132 } 1133 1134 return 0; 1135 } 1136 1137 static int nfit_mem_cmp(void *priv, const struct list_head *_a, 1138 const struct list_head *_b) 1139 { 1140 struct nfit_mem *a = container_of(_a, typeof(*a), list); 1141 struct nfit_mem *b = container_of(_b, typeof(*b), list); 1142 u32 handleA, handleB; 1143 1144 handleA = __to_nfit_memdev(a)->device_handle; 1145 handleB = __to_nfit_memdev(b)->device_handle; 1146 if (handleA < handleB) 1147 return -1; 1148 else if (handleA > handleB) 1149 return 1; 1150 return 0; 1151 } 1152 1153 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc) 1154 { 1155 struct nfit_spa *nfit_spa; 1156 int rc; 1157 1158 1159 /* 1160 * For each SPA-DCR or SPA-PMEM address range find its 1161 * corresponding MEMDEV(s). From each MEMDEV find the 1162 * corresponding DCR. Then, if we're operating on a SPA-DCR, 1163 * try to find a SPA-BDW and a corresponding BDW that references 1164 * the DCR. Throw it all into an nfit_mem object. Note, that 1165 * BDWs are optional. 1166 */ 1167 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) { 1168 rc = __nfit_mem_init(acpi_desc, nfit_spa->spa); 1169 if (rc) 1170 return rc; 1171 } 1172 1173 /* 1174 * If a DIMM has failed to be mapped into SPA there will be no 1175 * SPA entries above. Find and register all the unmapped DIMMs 1176 * for reporting and recovery purposes. 1177 */ 1178 rc = __nfit_mem_init(acpi_desc, NULL); 1179 if (rc) 1180 return rc; 1181 1182 list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp); 1183 1184 return 0; 1185 } 1186 1187 static ssize_t bus_dsm_mask_show(struct device *dev, 1188 struct device_attribute *attr, char *buf) 1189 { 1190 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev); 1191 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus); 1192 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc); 1193 1194 return sysfs_emit(buf, "%#lx\n", acpi_desc->bus_dsm_mask); 1195 } 1196 static struct device_attribute dev_attr_bus_dsm_mask = 1197 __ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL); 1198 1199 static ssize_t revision_show(struct device *dev, 1200 struct device_attribute *attr, char *buf) 1201 { 1202 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev); 1203 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus); 1204 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc); 1205 1206 return sysfs_emit(buf, "%d\n", acpi_desc->acpi_header.revision); 1207 } 1208 static DEVICE_ATTR_RO(revision); 1209 1210 static ssize_t hw_error_scrub_show(struct device *dev, 1211 struct device_attribute *attr, char *buf) 1212 { 1213 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev); 1214 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus); 1215 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc); 1216 1217 return sysfs_emit(buf, "%d\n", acpi_desc->scrub_mode); 1218 } 1219 1220 /* 1221 * The 'hw_error_scrub' attribute can have the following values written to it: 1222 * '0': Switch to the default mode where an exception will only insert 1223 * the address of the memory error into the poison and badblocks lists. 1224 * '1': Enable a full scrub to happen if an exception for a memory error is 1225 * received. 1226 */ 1227 static ssize_t hw_error_scrub_store(struct device *dev, 1228 struct device_attribute *attr, const char *buf, size_t size) 1229 { 1230 struct nvdimm_bus_descriptor *nd_desc; 1231 ssize_t rc; 1232 long val; 1233 1234 rc = kstrtol(buf, 0, &val); 1235 if (rc) 1236 return rc; 1237 1238 device_lock(dev); 1239 nd_desc = dev_get_drvdata(dev); 1240 if (nd_desc) { 1241 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc); 1242 1243 switch (val) { 1244 case HW_ERROR_SCRUB_ON: 1245 acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON; 1246 break; 1247 case HW_ERROR_SCRUB_OFF: 1248 acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF; 1249 break; 1250 default: 1251 rc = -EINVAL; 1252 break; 1253 } 1254 } 1255 device_unlock(dev); 1256 if (rc) 1257 return rc; 1258 return size; 1259 } 1260 static DEVICE_ATTR_RW(hw_error_scrub); 1261 1262 /* 1263 * This shows the number of full Address Range Scrubs that have been 1264 * completed since driver load time. Userspace can wait on this using 1265 * select/poll etc. A '+' at the end indicates an ARS is in progress 1266 */ 1267 static ssize_t scrub_show(struct device *dev, 1268 struct device_attribute *attr, char *buf) 1269 { 1270 struct nvdimm_bus_descriptor *nd_desc; 1271 struct acpi_nfit_desc *acpi_desc; 1272 ssize_t rc = -ENXIO; 1273 bool busy; 1274 1275 device_lock(dev); 1276 nd_desc = dev_get_drvdata(dev); 1277 if (!nd_desc) { 1278 device_unlock(dev); 1279 return rc; 1280 } 1281 acpi_desc = to_acpi_desc(nd_desc); 1282 1283 mutex_lock(&acpi_desc->init_mutex); 1284 busy = test_bit(ARS_BUSY, &acpi_desc->scrub_flags) 1285 && !test_bit(ARS_CANCEL, &acpi_desc->scrub_flags); 1286 rc = sysfs_emit(buf, "%d%s", acpi_desc->scrub_count, busy ? "+\n" : "\n"); 1287 /* Allow an admin to poll the busy state at a higher rate */ 1288 if (busy && capable(CAP_SYS_RAWIO) && !test_and_set_bit(ARS_POLL, 1289 &acpi_desc->scrub_flags)) { 1290 acpi_desc->scrub_tmo = 1; 1291 mod_delayed_work(nfit_wq, &acpi_desc->dwork, HZ); 1292 } 1293 1294 mutex_unlock(&acpi_desc->init_mutex); 1295 device_unlock(dev); 1296 return rc; 1297 } 1298 1299 static ssize_t scrub_store(struct device *dev, 1300 struct device_attribute *attr, const char *buf, size_t size) 1301 { 1302 struct nvdimm_bus_descriptor *nd_desc; 1303 ssize_t rc; 1304 long val; 1305 1306 rc = kstrtol(buf, 0, &val); 1307 if (rc) 1308 return rc; 1309 if (val != 1) 1310 return -EINVAL; 1311 1312 device_lock(dev); 1313 nd_desc = dev_get_drvdata(dev); 1314 if (nd_desc) { 1315 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc); 1316 1317 rc = acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG); 1318 } 1319 device_unlock(dev); 1320 if (rc) 1321 return rc; 1322 return size; 1323 } 1324 static DEVICE_ATTR_RW(scrub); 1325 1326 static bool ars_supported(struct nvdimm_bus *nvdimm_bus) 1327 { 1328 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus); 1329 const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START 1330 | 1 << ND_CMD_ARS_STATUS; 1331 1332 return (nd_desc->cmd_mask & mask) == mask; 1333 } 1334 1335 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n) 1336 { 1337 struct device *dev = kobj_to_dev(kobj); 1338 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev); 1339 1340 if (a == &dev_attr_scrub.attr) 1341 return ars_supported(nvdimm_bus) ? a->mode : 0; 1342 1343 if (a == &dev_attr_firmware_activate_noidle.attr) 1344 return intel_fwa_supported(nvdimm_bus) ? a->mode : 0; 1345 1346 return a->mode; 1347 } 1348 1349 static struct attribute *acpi_nfit_attributes[] = { 1350 &dev_attr_revision.attr, 1351 &dev_attr_scrub.attr, 1352 &dev_attr_hw_error_scrub.attr, 1353 &dev_attr_bus_dsm_mask.attr, 1354 &dev_attr_firmware_activate_noidle.attr, 1355 NULL, 1356 }; 1357 1358 static const struct attribute_group acpi_nfit_attribute_group = { 1359 .name = "nfit", 1360 .attrs = acpi_nfit_attributes, 1361 .is_visible = nfit_visible, 1362 }; 1363 1364 static const struct attribute_group *acpi_nfit_attribute_groups[] = { 1365 &acpi_nfit_attribute_group, 1366 NULL, 1367 }; 1368 1369 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev) 1370 { 1371 struct nvdimm *nvdimm = to_nvdimm(dev); 1372 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 1373 1374 return __to_nfit_memdev(nfit_mem); 1375 } 1376 1377 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev) 1378 { 1379 struct nvdimm *nvdimm = to_nvdimm(dev); 1380 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 1381 1382 return nfit_mem->dcr; 1383 } 1384 1385 static ssize_t handle_show(struct device *dev, 1386 struct device_attribute *attr, char *buf) 1387 { 1388 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev); 1389 1390 return sysfs_emit(buf, "%#x\n", memdev->device_handle); 1391 } 1392 static DEVICE_ATTR_RO(handle); 1393 1394 static ssize_t phys_id_show(struct device *dev, 1395 struct device_attribute *attr, char *buf) 1396 { 1397 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev); 1398 1399 return sysfs_emit(buf, "%#x\n", memdev->physical_id); 1400 } 1401 static DEVICE_ATTR_RO(phys_id); 1402 1403 static ssize_t vendor_show(struct device *dev, 1404 struct device_attribute *attr, char *buf) 1405 { 1406 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev); 1407 1408 return sysfs_emit(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id)); 1409 } 1410 static DEVICE_ATTR_RO(vendor); 1411 1412 static ssize_t rev_id_show(struct device *dev, 1413 struct device_attribute *attr, char *buf) 1414 { 1415 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev); 1416 1417 return sysfs_emit(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id)); 1418 } 1419 static DEVICE_ATTR_RO(rev_id); 1420 1421 static ssize_t device_show(struct device *dev, 1422 struct device_attribute *attr, char *buf) 1423 { 1424 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev); 1425 1426 return sysfs_emit(buf, "0x%04x\n", be16_to_cpu(dcr->device_id)); 1427 } 1428 static DEVICE_ATTR_RO(device); 1429 1430 static ssize_t subsystem_vendor_show(struct device *dev, 1431 struct device_attribute *attr, char *buf) 1432 { 1433 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev); 1434 1435 return sysfs_emit(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id)); 1436 } 1437 static DEVICE_ATTR_RO(subsystem_vendor); 1438 1439 static ssize_t subsystem_rev_id_show(struct device *dev, 1440 struct device_attribute *attr, char *buf) 1441 { 1442 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev); 1443 1444 return sysfs_emit(buf, "0x%04x\n", 1445 be16_to_cpu(dcr->subsystem_revision_id)); 1446 } 1447 static DEVICE_ATTR_RO(subsystem_rev_id); 1448 1449 static ssize_t subsystem_device_show(struct device *dev, 1450 struct device_attribute *attr, char *buf) 1451 { 1452 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev); 1453 1454 return sysfs_emit(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id)); 1455 } 1456 static DEVICE_ATTR_RO(subsystem_device); 1457 1458 static int num_nvdimm_formats(struct nvdimm *nvdimm) 1459 { 1460 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 1461 int formats = 0; 1462 1463 if (nfit_mem->memdev_pmem) 1464 formats++; 1465 return formats; 1466 } 1467 1468 static ssize_t format_show(struct device *dev, 1469 struct device_attribute *attr, char *buf) 1470 { 1471 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev); 1472 1473 return sysfs_emit(buf, "0x%04x\n", le16_to_cpu(dcr->code)); 1474 } 1475 static DEVICE_ATTR_RO(format); 1476 1477 static ssize_t format1_show(struct device *dev, 1478 struct device_attribute *attr, char *buf) 1479 { 1480 u32 handle; 1481 ssize_t rc = -ENXIO; 1482 struct nfit_mem *nfit_mem; 1483 struct nfit_memdev *nfit_memdev; 1484 struct acpi_nfit_desc *acpi_desc; 1485 struct nvdimm *nvdimm = to_nvdimm(dev); 1486 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev); 1487 1488 nfit_mem = nvdimm_provider_data(nvdimm); 1489 acpi_desc = nfit_mem->acpi_desc; 1490 handle = to_nfit_memdev(dev)->device_handle; 1491 1492 /* assumes DIMMs have at most 2 published interface codes */ 1493 mutex_lock(&acpi_desc->init_mutex); 1494 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) { 1495 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev; 1496 struct nfit_dcr *nfit_dcr; 1497 1498 if (memdev->device_handle != handle) 1499 continue; 1500 1501 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) { 1502 if (nfit_dcr->dcr->region_index != memdev->region_index) 1503 continue; 1504 if (nfit_dcr->dcr->code == dcr->code) 1505 continue; 1506 rc = sysfs_emit(buf, "0x%04x\n", 1507 le16_to_cpu(nfit_dcr->dcr->code)); 1508 break; 1509 } 1510 if (rc != -ENXIO) 1511 break; 1512 } 1513 mutex_unlock(&acpi_desc->init_mutex); 1514 return rc; 1515 } 1516 static DEVICE_ATTR_RO(format1); 1517 1518 static ssize_t formats_show(struct device *dev, 1519 struct device_attribute *attr, char *buf) 1520 { 1521 struct nvdimm *nvdimm = to_nvdimm(dev); 1522 1523 return sysfs_emit(buf, "%d\n", num_nvdimm_formats(nvdimm)); 1524 } 1525 static DEVICE_ATTR_RO(formats); 1526 1527 static ssize_t serial_show(struct device *dev, 1528 struct device_attribute *attr, char *buf) 1529 { 1530 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev); 1531 1532 return sysfs_emit(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number)); 1533 } 1534 static DEVICE_ATTR_RO(serial); 1535 1536 static ssize_t family_show(struct device *dev, 1537 struct device_attribute *attr, char *buf) 1538 { 1539 struct nvdimm *nvdimm = to_nvdimm(dev); 1540 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 1541 1542 if (nfit_mem->family < 0) 1543 return -ENXIO; 1544 return sysfs_emit(buf, "%d\n", nfit_mem->family); 1545 } 1546 static DEVICE_ATTR_RO(family); 1547 1548 static ssize_t dsm_mask_show(struct device *dev, 1549 struct device_attribute *attr, char *buf) 1550 { 1551 struct nvdimm *nvdimm = to_nvdimm(dev); 1552 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 1553 1554 if (nfit_mem->family < 0) 1555 return -ENXIO; 1556 return sysfs_emit(buf, "%#lx\n", nfit_mem->dsm_mask); 1557 } 1558 static DEVICE_ATTR_RO(dsm_mask); 1559 1560 static ssize_t flags_show(struct device *dev, 1561 struct device_attribute *attr, char *buf) 1562 { 1563 struct nvdimm *nvdimm = to_nvdimm(dev); 1564 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 1565 u16 flags = __to_nfit_memdev(nfit_mem)->flags; 1566 1567 if (test_bit(NFIT_MEM_DIRTY, &nfit_mem->flags)) 1568 flags |= ACPI_NFIT_MEM_FLUSH_FAILED; 1569 1570 return sysfs_emit(buf, "%s%s%s%s%s%s%s\n", 1571 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "", 1572 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "", 1573 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "", 1574 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "", 1575 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "", 1576 flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "", 1577 flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : ""); 1578 } 1579 static DEVICE_ATTR_RO(flags); 1580 1581 static ssize_t id_show(struct device *dev, 1582 struct device_attribute *attr, char *buf) 1583 { 1584 struct nvdimm *nvdimm = to_nvdimm(dev); 1585 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 1586 1587 return sysfs_emit(buf, "%s\n", nfit_mem->id); 1588 } 1589 static DEVICE_ATTR_RO(id); 1590 1591 static ssize_t dirty_shutdown_show(struct device *dev, 1592 struct device_attribute *attr, char *buf) 1593 { 1594 struct nvdimm *nvdimm = to_nvdimm(dev); 1595 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 1596 1597 return sysfs_emit(buf, "%d\n", nfit_mem->dirty_shutdown); 1598 } 1599 static DEVICE_ATTR_RO(dirty_shutdown); 1600 1601 static struct attribute *acpi_nfit_dimm_attributes[] = { 1602 &dev_attr_handle.attr, 1603 &dev_attr_phys_id.attr, 1604 &dev_attr_vendor.attr, 1605 &dev_attr_device.attr, 1606 &dev_attr_rev_id.attr, 1607 &dev_attr_subsystem_vendor.attr, 1608 &dev_attr_subsystem_device.attr, 1609 &dev_attr_subsystem_rev_id.attr, 1610 &dev_attr_format.attr, 1611 &dev_attr_formats.attr, 1612 &dev_attr_format1.attr, 1613 &dev_attr_serial.attr, 1614 &dev_attr_flags.attr, 1615 &dev_attr_id.attr, 1616 &dev_attr_family.attr, 1617 &dev_attr_dsm_mask.attr, 1618 &dev_attr_dirty_shutdown.attr, 1619 NULL, 1620 }; 1621 1622 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj, 1623 struct attribute *a, int n) 1624 { 1625 struct device *dev = kobj_to_dev(kobj); 1626 struct nvdimm *nvdimm = to_nvdimm(dev); 1627 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 1628 1629 if (!to_nfit_dcr(dev)) { 1630 /* Without a dcr only the memdev attributes can be surfaced */ 1631 if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr 1632 || a == &dev_attr_flags.attr 1633 || a == &dev_attr_family.attr 1634 || a == &dev_attr_dsm_mask.attr) 1635 return a->mode; 1636 return 0; 1637 } 1638 1639 if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1) 1640 return 0; 1641 1642 if (!test_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags) 1643 && a == &dev_attr_dirty_shutdown.attr) 1644 return 0; 1645 1646 return a->mode; 1647 } 1648 1649 static const struct attribute_group acpi_nfit_dimm_attribute_group = { 1650 .name = "nfit", 1651 .attrs = acpi_nfit_dimm_attributes, 1652 .is_visible = acpi_nfit_dimm_attr_visible, 1653 }; 1654 1655 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = { 1656 &acpi_nfit_dimm_attribute_group, 1657 NULL, 1658 }; 1659 1660 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc, 1661 u32 device_handle) 1662 { 1663 struct nfit_mem *nfit_mem; 1664 1665 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) 1666 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle) 1667 return nfit_mem->nvdimm; 1668 1669 return NULL; 1670 } 1671 1672 void __acpi_nvdimm_notify(struct device *dev, u32 event) 1673 { 1674 struct nfit_mem *nfit_mem; 1675 struct acpi_nfit_desc *acpi_desc; 1676 1677 dev_dbg(dev->parent, "%s: event: %d\n", dev_name(dev), 1678 event); 1679 1680 if (event != NFIT_NOTIFY_DIMM_HEALTH) { 1681 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev), 1682 event); 1683 return; 1684 } 1685 1686 acpi_desc = dev_get_drvdata(dev->parent); 1687 if (!acpi_desc) 1688 return; 1689 1690 /* 1691 * If we successfully retrieved acpi_desc, then we know nfit_mem data 1692 * is still valid. 1693 */ 1694 nfit_mem = dev_get_drvdata(dev); 1695 if (nfit_mem && nfit_mem->flags_attr) 1696 sysfs_notify_dirent(nfit_mem->flags_attr); 1697 } 1698 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify); 1699 1700 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data) 1701 { 1702 struct acpi_device *adev = data; 1703 struct device *dev = &adev->dev; 1704 1705 device_lock(dev->parent); 1706 __acpi_nvdimm_notify(dev, event); 1707 device_unlock(dev->parent); 1708 } 1709 1710 static bool acpi_nvdimm_has_method(struct acpi_device *adev, char *method) 1711 { 1712 acpi_handle handle; 1713 acpi_status status; 1714 1715 status = acpi_get_handle(adev->handle, method, &handle); 1716 1717 if (ACPI_SUCCESS(status)) 1718 return true; 1719 return false; 1720 } 1721 1722 __weak void nfit_intel_shutdown_status(struct nfit_mem *nfit_mem) 1723 { 1724 struct device *dev = &nfit_mem->adev->dev; 1725 struct nd_intel_smart smart = { 0 }; 1726 union acpi_object in_buf = { 1727 .buffer.type = ACPI_TYPE_BUFFER, 1728 .buffer.length = 0, 1729 }; 1730 union acpi_object in_obj = { 1731 .package.type = ACPI_TYPE_PACKAGE, 1732 .package.count = 1, 1733 .package.elements = &in_buf, 1734 }; 1735 const u8 func = ND_INTEL_SMART; 1736 const guid_t *guid = to_nfit_uuid(nfit_mem->family); 1737 u8 revid = nfit_dsm_revid(nfit_mem->family, func); 1738 struct acpi_device *adev = nfit_mem->adev; 1739 acpi_handle handle = adev->handle; 1740 union acpi_object *out_obj; 1741 1742 if ((nfit_mem->dsm_mask & (1 << func)) == 0) 1743 return; 1744 1745 out_obj = acpi_evaluate_dsm_typed(handle, guid, revid, func, &in_obj, ACPI_TYPE_BUFFER); 1746 if (!out_obj || out_obj->buffer.length < sizeof(smart)) { 1747 dev_dbg(dev->parent, "%s: failed to retrieve initial health\n", 1748 dev_name(dev)); 1749 ACPI_FREE(out_obj); 1750 return; 1751 } 1752 memcpy(&smart, out_obj->buffer.pointer, sizeof(smart)); 1753 ACPI_FREE(out_obj); 1754 1755 if (smart.flags & ND_INTEL_SMART_SHUTDOWN_VALID) { 1756 if (smart.shutdown_state) 1757 set_bit(NFIT_MEM_DIRTY, &nfit_mem->flags); 1758 } 1759 1760 if (smart.flags & ND_INTEL_SMART_SHUTDOWN_COUNT_VALID) { 1761 set_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags); 1762 nfit_mem->dirty_shutdown = smart.shutdown_count; 1763 } 1764 } 1765 1766 static void populate_shutdown_status(struct nfit_mem *nfit_mem) 1767 { 1768 /* 1769 * For DIMMs that provide a dynamic facility to retrieve a 1770 * dirty-shutdown status and/or a dirty-shutdown count, cache 1771 * these values in nfit_mem. 1772 */ 1773 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) 1774 nfit_intel_shutdown_status(nfit_mem); 1775 } 1776 1777 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc, 1778 struct nfit_mem *nfit_mem, u32 device_handle) 1779 { 1780 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc; 1781 struct acpi_device *adev, *adev_dimm; 1782 struct device *dev = acpi_desc->dev; 1783 unsigned long dsm_mask, label_mask; 1784 const guid_t *guid; 1785 int i; 1786 int family = -1; 1787 struct acpi_nfit_control_region *dcr = nfit_mem->dcr; 1788 1789 /* nfit test assumes 1:1 relationship between commands and dsms */ 1790 nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en; 1791 nfit_mem->family = NVDIMM_FAMILY_INTEL; 1792 set_bit(NVDIMM_FAMILY_INTEL, &nd_desc->dimm_family_mask); 1793 1794 if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID) 1795 sprintf(nfit_mem->id, "%04x-%02x-%04x-%08x", 1796 be16_to_cpu(dcr->vendor_id), 1797 dcr->manufacturing_location, 1798 be16_to_cpu(dcr->manufacturing_date), 1799 be32_to_cpu(dcr->serial_number)); 1800 else 1801 sprintf(nfit_mem->id, "%04x-%08x", 1802 be16_to_cpu(dcr->vendor_id), 1803 be32_to_cpu(dcr->serial_number)); 1804 1805 adev = to_acpi_dev(acpi_desc); 1806 if (!adev) { 1807 /* unit test case */ 1808 populate_shutdown_status(nfit_mem); 1809 return 0; 1810 } 1811 1812 adev_dimm = acpi_find_child_device(adev, device_handle, false); 1813 nfit_mem->adev = adev_dimm; 1814 if (!adev_dimm) { 1815 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n", 1816 device_handle); 1817 return force_enable_dimms ? 0 : -ENODEV; 1818 } 1819 1820 if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle, 1821 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) { 1822 dev_err(dev, "%s: notification registration failed\n", 1823 dev_name(&adev_dimm->dev)); 1824 return -ENXIO; 1825 } 1826 /* 1827 * Record nfit_mem for the notification path to track back to 1828 * the nfit sysfs attributes for this dimm device object. 1829 */ 1830 dev_set_drvdata(&adev_dimm->dev, nfit_mem); 1831 1832 /* 1833 * There are 4 "legacy" NVDIMM command sets 1834 * (NVDIMM_FAMILY_{INTEL,MSFT,HPE1,HPE2}) that were created before 1835 * an EFI working group was established to constrain this 1836 * proliferation. The nfit driver probes for the supported command 1837 * set by GUID. Note, if you're a platform developer looking to add 1838 * a new command set to this probe, consider using an existing set, 1839 * or otherwise seek approval to publish the command set at 1840 * http://www.uefi.org/RFIC_LIST. 1841 * 1842 * Note, that checking for function0 (bit0) tells us if any commands 1843 * are reachable through this GUID. 1844 */ 1845 clear_bit(NVDIMM_FAMILY_INTEL, &nd_desc->dimm_family_mask); 1846 for (i = 0; i <= NVDIMM_FAMILY_MAX; i++) 1847 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1)) { 1848 set_bit(i, &nd_desc->dimm_family_mask); 1849 if (family < 0 || i == default_dsm_family) 1850 family = i; 1851 } 1852 1853 /* limit the supported commands to those that are publicly documented */ 1854 nfit_mem->family = family; 1855 if (override_dsm_mask && !disable_vendor_specific) 1856 dsm_mask = override_dsm_mask; 1857 else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) { 1858 dsm_mask = NVDIMM_INTEL_CMDMASK; 1859 if (disable_vendor_specific) 1860 dsm_mask &= ~(1 << ND_CMD_VENDOR); 1861 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) { 1862 dsm_mask = 0x1c3c76; 1863 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) { 1864 dsm_mask = 0x1fe; 1865 if (disable_vendor_specific) 1866 dsm_mask &= ~(1 << 8); 1867 } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) { 1868 dsm_mask = 0xffffffff; 1869 } else if (nfit_mem->family == NVDIMM_FAMILY_HYPERV) { 1870 dsm_mask = 0x1f; 1871 } else { 1872 dev_dbg(dev, "unknown dimm command family\n"); 1873 nfit_mem->family = -1; 1874 /* DSMs are optional, continue loading the driver... */ 1875 return 0; 1876 } 1877 1878 /* 1879 * Function 0 is the command interrogation function, don't 1880 * export it to potential userspace use, and enable it to be 1881 * used as an error value in acpi_nfit_ctl(). 1882 */ 1883 dsm_mask &= ~1UL; 1884 1885 guid = to_nfit_uuid(nfit_mem->family); 1886 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG) 1887 if (acpi_check_dsm(adev_dimm->handle, guid, 1888 nfit_dsm_revid(nfit_mem->family, i), 1889 1ULL << i)) 1890 set_bit(i, &nfit_mem->dsm_mask); 1891 1892 /* 1893 * Prefer the NVDIMM_FAMILY_INTEL label read commands if present 1894 * due to their better semantics handling locked capacity. 1895 */ 1896 label_mask = 1 << ND_CMD_GET_CONFIG_SIZE | 1 << ND_CMD_GET_CONFIG_DATA 1897 | 1 << ND_CMD_SET_CONFIG_DATA; 1898 if (family == NVDIMM_FAMILY_INTEL 1899 && (dsm_mask & label_mask) == label_mask) 1900 /* skip _LS{I,R,W} enabling */; 1901 else { 1902 if (acpi_nvdimm_has_method(adev_dimm, "_LSI") 1903 && acpi_nvdimm_has_method(adev_dimm, "_LSR")) { 1904 dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev)); 1905 set_bit(NFIT_MEM_LSR, &nfit_mem->flags); 1906 } 1907 1908 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags) 1909 && acpi_nvdimm_has_method(adev_dimm, "_LSW")) { 1910 dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev)); 1911 set_bit(NFIT_MEM_LSW, &nfit_mem->flags); 1912 } 1913 1914 /* 1915 * Quirk read-only label configurations to preserve 1916 * access to label-less namespaces by default. 1917 */ 1918 if (!test_bit(NFIT_MEM_LSW, &nfit_mem->flags) 1919 && !force_labels) { 1920 dev_dbg(dev, "%s: No _LSW, disable labels\n", 1921 dev_name(&adev_dimm->dev)); 1922 clear_bit(NFIT_MEM_LSR, &nfit_mem->flags); 1923 } else 1924 dev_dbg(dev, "%s: Force enable labels\n", 1925 dev_name(&adev_dimm->dev)); 1926 } 1927 1928 populate_shutdown_status(nfit_mem); 1929 1930 return 0; 1931 } 1932 1933 static void shutdown_dimm_notify(void *data) 1934 { 1935 struct acpi_nfit_desc *acpi_desc = data; 1936 struct nfit_mem *nfit_mem; 1937 1938 mutex_lock(&acpi_desc->init_mutex); 1939 /* 1940 * Clear out the nfit_mem->flags_attr and shut down dimm event 1941 * notifications. 1942 */ 1943 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) { 1944 struct acpi_device *adev_dimm = nfit_mem->adev; 1945 1946 if (nfit_mem->flags_attr) { 1947 sysfs_put(nfit_mem->flags_attr); 1948 nfit_mem->flags_attr = NULL; 1949 } 1950 if (adev_dimm) { 1951 acpi_remove_notify_handler(adev_dimm->handle, 1952 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify); 1953 dev_set_drvdata(&adev_dimm->dev, NULL); 1954 } 1955 } 1956 mutex_unlock(&acpi_desc->init_mutex); 1957 } 1958 1959 static const struct nvdimm_security_ops *acpi_nfit_get_security_ops(int family) 1960 { 1961 switch (family) { 1962 case NVDIMM_FAMILY_INTEL: 1963 return intel_security_ops; 1964 default: 1965 return NULL; 1966 } 1967 } 1968 1969 static const struct nvdimm_fw_ops *acpi_nfit_get_fw_ops( 1970 struct nfit_mem *nfit_mem) 1971 { 1972 unsigned long mask; 1973 struct acpi_nfit_desc *acpi_desc = nfit_mem->acpi_desc; 1974 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc; 1975 1976 if (!nd_desc->fw_ops) 1977 return NULL; 1978 1979 if (nfit_mem->family != NVDIMM_FAMILY_INTEL) 1980 return NULL; 1981 1982 mask = nfit_mem->dsm_mask & NVDIMM_INTEL_FW_ACTIVATE_CMDMASK; 1983 if (mask != NVDIMM_INTEL_FW_ACTIVATE_CMDMASK) 1984 return NULL; 1985 1986 return intel_fw_ops; 1987 } 1988 1989 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc) 1990 { 1991 struct nfit_mem *nfit_mem; 1992 int dimm_count = 0, rc; 1993 struct nvdimm *nvdimm; 1994 1995 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) { 1996 struct acpi_nfit_flush_address *flush; 1997 unsigned long flags = 0, cmd_mask; 1998 struct nfit_memdev *nfit_memdev; 1999 u32 device_handle; 2000 u16 mem_flags; 2001 2002 device_handle = __to_nfit_memdev(nfit_mem)->device_handle; 2003 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle); 2004 if (nvdimm) { 2005 dimm_count++; 2006 continue; 2007 } 2008 2009 /* collate flags across all memdevs for this dimm */ 2010 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) { 2011 struct acpi_nfit_memory_map *dimm_memdev; 2012 2013 dimm_memdev = __to_nfit_memdev(nfit_mem); 2014 if (dimm_memdev->device_handle 2015 != nfit_memdev->memdev->device_handle) 2016 continue; 2017 dimm_memdev->flags |= nfit_memdev->memdev->flags; 2018 } 2019 2020 mem_flags = __to_nfit_memdev(nfit_mem)->flags; 2021 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED) 2022 set_bit(NDD_UNARMED, &flags); 2023 2024 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle); 2025 if (rc) 2026 continue; 2027 2028 /* 2029 * TODO: provide translation for non-NVDIMM_FAMILY_INTEL 2030 * devices (i.e. from nd_cmd to acpi_dsm) to standardize the 2031 * userspace interface. 2032 */ 2033 cmd_mask = 1UL << ND_CMD_CALL; 2034 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) { 2035 /* 2036 * These commands have a 1:1 correspondence 2037 * between DSM payload and libnvdimm ioctl 2038 * payload format. 2039 */ 2040 cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK; 2041 } 2042 2043 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) { 2044 set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask); 2045 set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask); 2046 } 2047 if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags)) 2048 set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask); 2049 2050 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush 2051 : NULL; 2052 nvdimm = __nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem, 2053 acpi_nfit_dimm_attribute_groups, 2054 flags, cmd_mask, flush ? flush->hint_count : 0, 2055 nfit_mem->flush_wpq, &nfit_mem->id[0], 2056 acpi_nfit_get_security_ops(nfit_mem->family), 2057 acpi_nfit_get_fw_ops(nfit_mem)); 2058 if (!nvdimm) 2059 return -ENOMEM; 2060 2061 nfit_mem->nvdimm = nvdimm; 2062 dimm_count++; 2063 2064 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0) 2065 continue; 2066 2067 dev_err(acpi_desc->dev, "Error found in NVDIMM %s flags:%s%s%s%s%s\n", 2068 nvdimm_name(nvdimm), 2069 mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "", 2070 mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"", 2071 mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "", 2072 mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "", 2073 mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : ""); 2074 2075 } 2076 2077 rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count); 2078 if (rc) 2079 return rc; 2080 2081 /* 2082 * Now that dimms are successfully registered, and async registration 2083 * is flushed, attempt to enable event notification. 2084 */ 2085 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) { 2086 struct kernfs_node *nfit_kernfs; 2087 2088 nvdimm = nfit_mem->nvdimm; 2089 if (!nvdimm) 2090 continue; 2091 2092 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit"); 2093 if (nfit_kernfs) 2094 nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs, 2095 "flags"); 2096 sysfs_put(nfit_kernfs); 2097 if (!nfit_mem->flags_attr) 2098 dev_warn(acpi_desc->dev, "%s: notifications disabled\n", 2099 nvdimm_name(nvdimm)); 2100 } 2101 2102 return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify, 2103 acpi_desc); 2104 } 2105 2106 /* 2107 * These constants are private because there are no kernel consumers of 2108 * these commands. 2109 */ 2110 enum nfit_aux_cmds { 2111 NFIT_CMD_TRANSLATE_SPA = 5, 2112 NFIT_CMD_ARS_INJECT_SET = 7, 2113 NFIT_CMD_ARS_INJECT_CLEAR = 8, 2114 NFIT_CMD_ARS_INJECT_GET = 9, 2115 }; 2116 2117 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc) 2118 { 2119 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc; 2120 const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS); 2121 unsigned long dsm_mask, *mask; 2122 struct acpi_device *adev; 2123 int i; 2124 2125 set_bit(ND_CMD_CALL, &nd_desc->cmd_mask); 2126 set_bit(NVDIMM_BUS_FAMILY_NFIT, &nd_desc->bus_family_mask); 2127 2128 /* enable nfit_test to inject bus command emulation */ 2129 if (acpi_desc->bus_cmd_force_en) { 2130 nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en; 2131 mask = &nd_desc->bus_family_mask; 2132 if (acpi_desc->family_dsm_mask[NVDIMM_BUS_FAMILY_INTEL]) { 2133 set_bit(NVDIMM_BUS_FAMILY_INTEL, mask); 2134 nd_desc->fw_ops = intel_bus_fw_ops; 2135 } 2136 } 2137 2138 adev = to_acpi_dev(acpi_desc); 2139 if (!adev) 2140 return; 2141 2142 for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++) 2143 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i)) 2144 set_bit(i, &nd_desc->cmd_mask); 2145 2146 dsm_mask = 2147 (1 << ND_CMD_ARS_CAP) | 2148 (1 << ND_CMD_ARS_START) | 2149 (1 << ND_CMD_ARS_STATUS) | 2150 (1 << ND_CMD_CLEAR_ERROR) | 2151 (1 << NFIT_CMD_TRANSLATE_SPA) | 2152 (1 << NFIT_CMD_ARS_INJECT_SET) | 2153 (1 << NFIT_CMD_ARS_INJECT_CLEAR) | 2154 (1 << NFIT_CMD_ARS_INJECT_GET); 2155 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG) 2156 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i)) 2157 set_bit(i, &acpi_desc->bus_dsm_mask); 2158 2159 /* Enumerate allowed NVDIMM_BUS_FAMILY_INTEL commands */ 2160 dsm_mask = NVDIMM_BUS_INTEL_FW_ACTIVATE_CMDMASK; 2161 guid = to_nfit_bus_uuid(NVDIMM_BUS_FAMILY_INTEL); 2162 mask = &acpi_desc->family_dsm_mask[NVDIMM_BUS_FAMILY_INTEL]; 2163 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG) 2164 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i)) 2165 set_bit(i, mask); 2166 2167 if (*mask == dsm_mask) { 2168 set_bit(NVDIMM_BUS_FAMILY_INTEL, &nd_desc->bus_family_mask); 2169 nd_desc->fw_ops = intel_bus_fw_ops; 2170 } 2171 } 2172 2173 static ssize_t range_index_show(struct device *dev, 2174 struct device_attribute *attr, char *buf) 2175 { 2176 struct nd_region *nd_region = to_nd_region(dev); 2177 struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region); 2178 2179 return sysfs_emit(buf, "%d\n", nfit_spa->spa->range_index); 2180 } 2181 static DEVICE_ATTR_RO(range_index); 2182 2183 static struct attribute *acpi_nfit_region_attributes[] = { 2184 &dev_attr_range_index.attr, 2185 NULL, 2186 }; 2187 2188 static const struct attribute_group acpi_nfit_region_attribute_group = { 2189 .name = "nfit", 2190 .attrs = acpi_nfit_region_attributes, 2191 }; 2192 2193 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = { 2194 &acpi_nfit_region_attribute_group, 2195 NULL, 2196 }; 2197 2198 /* enough info to uniquely specify an interleave set */ 2199 struct nfit_set_info { 2200 u64 region_offset; 2201 u32 serial_number; 2202 u32 pad; 2203 }; 2204 2205 struct nfit_set_info2 { 2206 u64 region_offset; 2207 u32 serial_number; 2208 u16 vendor_id; 2209 u16 manufacturing_date; 2210 u8 manufacturing_location; 2211 u8 reserved[31]; 2212 }; 2213 2214 static int cmp_map_compat(const void *m0, const void *m1) 2215 { 2216 const struct nfit_set_info *map0 = m0; 2217 const struct nfit_set_info *map1 = m1; 2218 2219 return memcmp(&map0->region_offset, &map1->region_offset, 2220 sizeof(u64)); 2221 } 2222 2223 static int cmp_map(const void *m0, const void *m1) 2224 { 2225 const struct nfit_set_info *map0 = m0; 2226 const struct nfit_set_info *map1 = m1; 2227 2228 if (map0->region_offset < map1->region_offset) 2229 return -1; 2230 else if (map0->region_offset > map1->region_offset) 2231 return 1; 2232 return 0; 2233 } 2234 2235 static int cmp_map2(const void *m0, const void *m1) 2236 { 2237 const struct nfit_set_info2 *map0 = m0; 2238 const struct nfit_set_info2 *map1 = m1; 2239 2240 if (map0->region_offset < map1->region_offset) 2241 return -1; 2242 else if (map0->region_offset > map1->region_offset) 2243 return 1; 2244 return 0; 2245 } 2246 2247 /* Retrieve the nth entry referencing this spa */ 2248 static struct acpi_nfit_memory_map *memdev_from_spa( 2249 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n) 2250 { 2251 struct nfit_memdev *nfit_memdev; 2252 2253 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) 2254 if (nfit_memdev->memdev->range_index == range_index) 2255 if (n-- == 0) 2256 return nfit_memdev->memdev; 2257 return NULL; 2258 } 2259 2260 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc, 2261 struct nd_region_desc *ndr_desc, 2262 struct acpi_nfit_system_address *spa) 2263 { 2264 u16 nr = ndr_desc->num_mappings; 2265 struct nfit_set_info2 *info2 __free(kfree) = 2266 kcalloc(nr, sizeof(*info2), GFP_KERNEL); 2267 struct nfit_set_info *info __free(kfree) = 2268 kcalloc(nr, sizeof(*info), GFP_KERNEL); 2269 struct device *dev = acpi_desc->dev; 2270 struct nd_interleave_set *nd_set; 2271 int i; 2272 2273 if (!info || !info2) 2274 return -ENOMEM; 2275 2276 nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL); 2277 if (!nd_set) 2278 return -ENOMEM; 2279 import_guid(&nd_set->type_guid, spa->range_guid); 2280 2281 for (i = 0; i < nr; i++) { 2282 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i]; 2283 struct nvdimm *nvdimm = mapping->nvdimm; 2284 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 2285 struct nfit_set_info *map = &info[i]; 2286 struct nfit_set_info2 *map2 = &info2[i]; 2287 struct acpi_nfit_memory_map *memdev = 2288 memdev_from_spa(acpi_desc, spa->range_index, i); 2289 struct acpi_nfit_control_region *dcr = nfit_mem->dcr; 2290 2291 if (!memdev || !nfit_mem->dcr) { 2292 dev_err(dev, "%s: failed to find DCR\n", __func__); 2293 return -ENODEV; 2294 } 2295 2296 map->region_offset = memdev->region_offset; 2297 map->serial_number = dcr->serial_number; 2298 2299 map2->region_offset = memdev->region_offset; 2300 map2->serial_number = dcr->serial_number; 2301 map2->vendor_id = dcr->vendor_id; 2302 map2->manufacturing_date = dcr->manufacturing_date; 2303 map2->manufacturing_location = dcr->manufacturing_location; 2304 } 2305 2306 /* v1.1 namespaces */ 2307 sort(info, nr, sizeof(*info), cmp_map, NULL); 2308 nd_set->cookie1 = nd_fletcher64(info, sizeof(*info) * nr, 0); 2309 2310 /* v1.2 namespaces */ 2311 sort(info2, nr, sizeof(*info2), cmp_map2, NULL); 2312 nd_set->cookie2 = nd_fletcher64(info2, sizeof(*info2) * nr, 0); 2313 2314 /* support v1.1 namespaces created with the wrong sort order */ 2315 sort(info, nr, sizeof(*info), cmp_map_compat, NULL); 2316 nd_set->altcookie = nd_fletcher64(info, sizeof(*info) * nr, 0); 2317 2318 /* record the result of the sort for the mapping position */ 2319 for (i = 0; i < nr; i++) { 2320 struct nfit_set_info2 *map2 = &info2[i]; 2321 int j; 2322 2323 for (j = 0; j < nr; j++) { 2324 struct nd_mapping_desc *mapping = &ndr_desc->mapping[j]; 2325 struct nvdimm *nvdimm = mapping->nvdimm; 2326 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm); 2327 struct acpi_nfit_control_region *dcr = nfit_mem->dcr; 2328 2329 if (map2->serial_number == dcr->serial_number && 2330 map2->vendor_id == dcr->vendor_id && 2331 map2->manufacturing_date == dcr->manufacturing_date && 2332 map2->manufacturing_location 2333 == dcr->manufacturing_location) { 2334 mapping->position = i; 2335 break; 2336 } 2337 } 2338 } 2339 2340 ndr_desc->nd_set = nd_set; 2341 2342 return 0; 2343 } 2344 2345 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc, 2346 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa) 2347 { 2348 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc; 2349 struct acpi_nfit_system_address *spa = nfit_spa->spa; 2350 int cmd_rc, rc; 2351 2352 cmd->address = spa->address; 2353 cmd->length = spa->length; 2354 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd, 2355 sizeof(*cmd), &cmd_rc); 2356 if (rc < 0) 2357 return rc; 2358 return cmd_rc; 2359 } 2360 2361 static int ars_start(struct acpi_nfit_desc *acpi_desc, 2362 struct nfit_spa *nfit_spa, enum nfit_ars_state req_type) 2363 { 2364 int rc; 2365 int cmd_rc; 2366 struct nd_cmd_ars_start ars_start; 2367 struct acpi_nfit_system_address *spa = nfit_spa->spa; 2368 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc; 2369 2370 memset(&ars_start, 0, sizeof(ars_start)); 2371 ars_start.address = spa->address; 2372 ars_start.length = spa->length; 2373 if (req_type == ARS_REQ_SHORT) 2374 ars_start.flags = ND_ARS_RETURN_PREV_DATA; 2375 if (nfit_spa_type(spa) == NFIT_SPA_PM) 2376 ars_start.type = ND_ARS_PERSISTENT; 2377 else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE) 2378 ars_start.type = ND_ARS_VOLATILE; 2379 else 2380 return -ENOTTY; 2381 2382 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start, 2383 sizeof(ars_start), &cmd_rc); 2384 2385 if (rc < 0) 2386 return rc; 2387 if (cmd_rc < 0) 2388 return cmd_rc; 2389 set_bit(ARS_VALID, &acpi_desc->scrub_flags); 2390 return 0; 2391 } 2392 2393 static int ars_continue(struct acpi_nfit_desc *acpi_desc) 2394 { 2395 int rc, cmd_rc; 2396 struct nd_cmd_ars_start ars_start; 2397 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc; 2398 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status; 2399 2400 ars_start = (struct nd_cmd_ars_start) { 2401 .address = ars_status->restart_address, 2402 .length = ars_status->restart_length, 2403 .type = ars_status->type, 2404 }; 2405 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start, 2406 sizeof(ars_start), &cmd_rc); 2407 if (rc < 0) 2408 return rc; 2409 return cmd_rc; 2410 } 2411 2412 static int ars_get_status(struct acpi_nfit_desc *acpi_desc) 2413 { 2414 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc; 2415 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status; 2416 int rc, cmd_rc; 2417 2418 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status, 2419 acpi_desc->max_ars, &cmd_rc); 2420 if (rc < 0) 2421 return rc; 2422 return cmd_rc; 2423 } 2424 2425 static void ars_complete(struct acpi_nfit_desc *acpi_desc, 2426 struct nfit_spa *nfit_spa) 2427 { 2428 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status; 2429 struct acpi_nfit_system_address *spa = nfit_spa->spa; 2430 struct nd_region *nd_region = nfit_spa->nd_region; 2431 struct device *dev; 2432 2433 lockdep_assert_held(&acpi_desc->init_mutex); 2434 /* 2435 * Only advance the ARS state for ARS runs initiated by the 2436 * kernel, ignore ARS results from BIOS initiated runs for scrub 2437 * completion tracking. 2438 */ 2439 if (acpi_desc->scrub_spa != nfit_spa) 2440 return; 2441 2442 if ((ars_status->address >= spa->address && ars_status->address 2443 < spa->address + spa->length) 2444 || (ars_status->address < spa->address)) { 2445 /* 2446 * Assume that if a scrub starts at an offset from the 2447 * start of nfit_spa that we are in the continuation 2448 * case. 2449 * 2450 * Otherwise, if the scrub covers the spa range, mark 2451 * any pending request complete. 2452 */ 2453 if (ars_status->address + ars_status->length 2454 >= spa->address + spa->length) 2455 /* complete */; 2456 else 2457 return; 2458 } else 2459 return; 2460 2461 acpi_desc->scrub_spa = NULL; 2462 if (nd_region) { 2463 dev = nd_region_dev(nd_region); 2464 nvdimm_region_notify(nd_region, NVDIMM_REVALIDATE_POISON); 2465 } else 2466 dev = acpi_desc->dev; 2467 dev_dbg(dev, "ARS: range %d complete\n", spa->range_index); 2468 } 2469 2470 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc) 2471 { 2472 struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus; 2473 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status; 2474 int rc; 2475 u32 i; 2476 2477 /* 2478 * First record starts at 44 byte offset from the start of the 2479 * payload. 2480 */ 2481 if (ars_status->out_length < 44) 2482 return 0; 2483 2484 /* 2485 * Ignore potentially stale results that are only refreshed 2486 * after a start-ARS event. 2487 */ 2488 if (!test_and_clear_bit(ARS_VALID, &acpi_desc->scrub_flags)) { 2489 dev_dbg(acpi_desc->dev, "skip %d stale records\n", 2490 ars_status->num_records); 2491 return 0; 2492 } 2493 2494 for (i = 0; i < ars_status->num_records; i++) { 2495 /* only process full records */ 2496 if (ars_status->out_length 2497 < 44 + sizeof(struct nd_ars_record) * (i + 1)) 2498 break; 2499 rc = nvdimm_bus_add_badrange(nvdimm_bus, 2500 ars_status->records[i].err_address, 2501 ars_status->records[i].length); 2502 if (rc) 2503 return rc; 2504 } 2505 if (i < ars_status->num_records) 2506 dev_warn(acpi_desc->dev, "detected truncated ars results\n"); 2507 2508 return 0; 2509 } 2510 2511 static void acpi_nfit_remove_resource(void *data) 2512 { 2513 struct resource *res = data; 2514 2515 remove_resource(res); 2516 } 2517 2518 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc, 2519 struct nd_region_desc *ndr_desc) 2520 { 2521 struct resource *res, *nd_res = ndr_desc->res; 2522 int is_pmem, ret; 2523 2524 /* No operation if the region is already registered as PMEM */ 2525 is_pmem = region_intersects(nd_res->start, resource_size(nd_res), 2526 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY); 2527 if (is_pmem == REGION_INTERSECTS) 2528 return 0; 2529 2530 res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL); 2531 if (!res) 2532 return -ENOMEM; 2533 2534 res->name = "Persistent Memory"; 2535 res->start = nd_res->start; 2536 res->end = nd_res->end; 2537 res->flags = IORESOURCE_MEM; 2538 res->desc = IORES_DESC_PERSISTENT_MEMORY; 2539 2540 ret = insert_resource(&iomem_resource, res); 2541 if (ret) 2542 return ret; 2543 2544 ret = devm_add_action_or_reset(acpi_desc->dev, 2545 acpi_nfit_remove_resource, 2546 res); 2547 if (ret) 2548 return ret; 2549 2550 return 0; 2551 } 2552 2553 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc, 2554 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc, 2555 struct acpi_nfit_memory_map *memdev, 2556 struct nfit_spa *nfit_spa) 2557 { 2558 struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, 2559 memdev->device_handle); 2560 struct acpi_nfit_system_address *spa = nfit_spa->spa; 2561 2562 if (!nvdimm) { 2563 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n", 2564 spa->range_index, memdev->device_handle); 2565 return -ENODEV; 2566 } 2567 2568 mapping->nvdimm = nvdimm; 2569 switch (nfit_spa_type(spa)) { 2570 case NFIT_SPA_PM: 2571 case NFIT_SPA_VOLATILE: 2572 mapping->start = memdev->address; 2573 mapping->size = memdev->region_size; 2574 break; 2575 } 2576 2577 return 0; 2578 } 2579 2580 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa) 2581 { 2582 return (nfit_spa_type(spa) == NFIT_SPA_VDISK || 2583 nfit_spa_type(spa) == NFIT_SPA_VCD || 2584 nfit_spa_type(spa) == NFIT_SPA_PDISK || 2585 nfit_spa_type(spa) == NFIT_SPA_PCD); 2586 } 2587 2588 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa) 2589 { 2590 return (nfit_spa_type(spa) == NFIT_SPA_VDISK || 2591 nfit_spa_type(spa) == NFIT_SPA_VCD || 2592 nfit_spa_type(spa) == NFIT_SPA_VOLATILE); 2593 } 2594 2595 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc, 2596 struct nfit_spa *nfit_spa) 2597 { 2598 static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS]; 2599 struct acpi_nfit_system_address *spa = nfit_spa->spa; 2600 struct nd_region_desc *ndr_desc, _ndr_desc; 2601 struct nfit_memdev *nfit_memdev; 2602 struct nvdimm_bus *nvdimm_bus; 2603 struct resource res; 2604 int count = 0, rc; 2605 2606 if (nfit_spa->nd_region) 2607 return 0; 2608 2609 if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) { 2610 dev_dbg(acpi_desc->dev, "detected invalid spa index\n"); 2611 return 0; 2612 } 2613 2614 memset(&res, 0, sizeof(res)); 2615 memset(&mappings, 0, sizeof(mappings)); 2616 memset(&_ndr_desc, 0, sizeof(_ndr_desc)); 2617 res.start = spa->address; 2618 res.end = res.start + spa->length - 1; 2619 ndr_desc = &_ndr_desc; 2620 ndr_desc->res = &res; 2621 ndr_desc->provider_data = nfit_spa; 2622 ndr_desc->attr_groups = acpi_nfit_region_attribute_groups; 2623 if (spa->flags & ACPI_NFIT_PROXIMITY_VALID) { 2624 ndr_desc->numa_node = pxm_to_online_node(spa->proximity_domain); 2625 ndr_desc->target_node = pxm_to_node(spa->proximity_domain); 2626 } else { 2627 ndr_desc->numa_node = NUMA_NO_NODE; 2628 ndr_desc->target_node = NUMA_NO_NODE; 2629 } 2630 2631 /* Fallback to address based numa information if node lookup failed */ 2632 if (ndr_desc->numa_node == NUMA_NO_NODE) { 2633 ndr_desc->numa_node = memory_add_physaddr_to_nid(spa->address); 2634 dev_info(acpi_desc->dev, "changing numa node from %d to %d for nfit region [%pa-%pa]", 2635 NUMA_NO_NODE, ndr_desc->numa_node, &res.start, &res.end); 2636 } 2637 if (ndr_desc->target_node == NUMA_NO_NODE) { 2638 ndr_desc->target_node = phys_to_target_node(spa->address); 2639 dev_info(acpi_desc->dev, "changing target node from %d to %d for nfit region [%pa-%pa]", 2640 NUMA_NO_NODE, ndr_desc->numa_node, &res.start, &res.end); 2641 } 2642 2643 /* 2644 * Persistence domain bits are hierarchical, if 2645 * ACPI_NFIT_CAPABILITY_CACHE_FLUSH is set then 2646 * ACPI_NFIT_CAPABILITY_MEM_FLUSH is implied. 2647 */ 2648 if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_CACHE_FLUSH) 2649 set_bit(ND_REGION_PERSIST_CACHE, &ndr_desc->flags); 2650 else if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_MEM_FLUSH) 2651 set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc->flags); 2652 2653 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) { 2654 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev; 2655 struct nd_mapping_desc *mapping; 2656 2657 /* range index 0 == unmapped in SPA or invalid-SPA */ 2658 if (memdev->range_index == 0 || spa->range_index == 0) 2659 continue; 2660 if (memdev->range_index != spa->range_index) 2661 continue; 2662 if (count >= ND_MAX_MAPPINGS) { 2663 dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n", 2664 spa->range_index, ND_MAX_MAPPINGS); 2665 return -ENXIO; 2666 } 2667 mapping = &mappings[count++]; 2668 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc, 2669 memdev, nfit_spa); 2670 if (rc) 2671 goto out; 2672 } 2673 2674 ndr_desc->mapping = mappings; 2675 ndr_desc->num_mappings = count; 2676 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa); 2677 if (rc) 2678 goto out; 2679 2680 nvdimm_bus = acpi_desc->nvdimm_bus; 2681 if (nfit_spa_type(spa) == NFIT_SPA_PM) { 2682 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc); 2683 if (rc) { 2684 dev_warn(acpi_desc->dev, 2685 "failed to insert pmem resource to iomem: %d\n", 2686 rc); 2687 goto out; 2688 } 2689 2690 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus, 2691 ndr_desc); 2692 if (!nfit_spa->nd_region) 2693 rc = -ENOMEM; 2694 } else if (nfit_spa_is_volatile(spa)) { 2695 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus, 2696 ndr_desc); 2697 if (!nfit_spa->nd_region) 2698 rc = -ENOMEM; 2699 } else if (nfit_spa_is_virtual(spa)) { 2700 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus, 2701 ndr_desc); 2702 if (!nfit_spa->nd_region) 2703 rc = -ENOMEM; 2704 } 2705 2706 out: 2707 if (rc) 2708 dev_err(acpi_desc->dev, "failed to register spa range %d\n", 2709 nfit_spa->spa->range_index); 2710 return rc; 2711 } 2712 2713 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc) 2714 { 2715 struct device *dev = acpi_desc->dev; 2716 struct nd_cmd_ars_status *ars_status; 2717 2718 if (acpi_desc->ars_status) { 2719 memset(acpi_desc->ars_status, 0, acpi_desc->max_ars); 2720 return 0; 2721 } 2722 2723 ars_status = devm_kzalloc(dev, acpi_desc->max_ars, GFP_KERNEL); 2724 if (!ars_status) 2725 return -ENOMEM; 2726 acpi_desc->ars_status = ars_status; 2727 return 0; 2728 } 2729 2730 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc) 2731 { 2732 int rc; 2733 2734 if (ars_status_alloc(acpi_desc)) 2735 return -ENOMEM; 2736 2737 rc = ars_get_status(acpi_desc); 2738 2739 if (rc < 0 && rc != -ENOSPC) 2740 return rc; 2741 2742 if (ars_status_process_records(acpi_desc)) 2743 dev_err(acpi_desc->dev, "Failed to process ARS records\n"); 2744 2745 return rc; 2746 } 2747 2748 static int ars_register(struct acpi_nfit_desc *acpi_desc, 2749 struct nfit_spa *nfit_spa) 2750 { 2751 int rc; 2752 2753 if (test_bit(ARS_FAILED, &nfit_spa->ars_state)) 2754 return acpi_nfit_register_region(acpi_desc, nfit_spa); 2755 2756 set_bit(ARS_REQ_SHORT, &nfit_spa->ars_state); 2757 if (!no_init_ars) 2758 set_bit(ARS_REQ_LONG, &nfit_spa->ars_state); 2759 2760 switch (acpi_nfit_query_poison(acpi_desc)) { 2761 case 0: 2762 case -ENOSPC: 2763 case -EAGAIN: 2764 rc = ars_start(acpi_desc, nfit_spa, ARS_REQ_SHORT); 2765 /* shouldn't happen, try again later */ 2766 if (rc == -EBUSY) 2767 break; 2768 if (rc) { 2769 set_bit(ARS_FAILED, &nfit_spa->ars_state); 2770 break; 2771 } 2772 clear_bit(ARS_REQ_SHORT, &nfit_spa->ars_state); 2773 rc = acpi_nfit_query_poison(acpi_desc); 2774 if (rc) 2775 break; 2776 acpi_desc->scrub_spa = nfit_spa; 2777 ars_complete(acpi_desc, nfit_spa); 2778 /* 2779 * If ars_complete() says we didn't complete the 2780 * short scrub, we'll try again with a long 2781 * request. 2782 */ 2783 acpi_desc->scrub_spa = NULL; 2784 break; 2785 case -EBUSY: 2786 case -ENOMEM: 2787 /* 2788 * BIOS was using ARS, wait for it to complete (or 2789 * resources to become available) and then perform our 2790 * own scrubs. 2791 */ 2792 break; 2793 default: 2794 set_bit(ARS_FAILED, &nfit_spa->ars_state); 2795 break; 2796 } 2797 2798 return acpi_nfit_register_region(acpi_desc, nfit_spa); 2799 } 2800 2801 static void ars_complete_all(struct acpi_nfit_desc *acpi_desc) 2802 { 2803 struct nfit_spa *nfit_spa; 2804 2805 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) { 2806 if (test_bit(ARS_FAILED, &nfit_spa->ars_state)) 2807 continue; 2808 ars_complete(acpi_desc, nfit_spa); 2809 } 2810 } 2811 2812 static unsigned int __acpi_nfit_scrub(struct acpi_nfit_desc *acpi_desc, 2813 int query_rc) 2814 { 2815 unsigned int tmo = acpi_desc->scrub_tmo; 2816 struct device *dev = acpi_desc->dev; 2817 struct nfit_spa *nfit_spa; 2818 2819 lockdep_assert_held(&acpi_desc->init_mutex); 2820 2821 if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags)) 2822 return 0; 2823 2824 if (query_rc == -EBUSY) { 2825 dev_dbg(dev, "ARS: ARS busy\n"); 2826 return min(30U * 60U, tmo * 2); 2827 } 2828 if (query_rc == -ENOSPC) { 2829 dev_dbg(dev, "ARS: ARS continue\n"); 2830 ars_continue(acpi_desc); 2831 return 1; 2832 } 2833 if (query_rc && query_rc != -EAGAIN) { 2834 unsigned long long addr, end; 2835 2836 addr = acpi_desc->ars_status->address; 2837 end = addr + acpi_desc->ars_status->length; 2838 dev_dbg(dev, "ARS: %llx-%llx failed (%d)\n", addr, end, 2839 query_rc); 2840 } 2841 2842 ars_complete_all(acpi_desc); 2843 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) { 2844 enum nfit_ars_state req_type; 2845 int rc; 2846 2847 if (test_bit(ARS_FAILED, &nfit_spa->ars_state)) 2848 continue; 2849 2850 /* prefer short ARS requests first */ 2851 if (test_bit(ARS_REQ_SHORT, &nfit_spa->ars_state)) 2852 req_type = ARS_REQ_SHORT; 2853 else if (test_bit(ARS_REQ_LONG, &nfit_spa->ars_state)) 2854 req_type = ARS_REQ_LONG; 2855 else 2856 continue; 2857 rc = ars_start(acpi_desc, nfit_spa, req_type); 2858 2859 dev = nd_region_dev(nfit_spa->nd_region); 2860 dev_dbg(dev, "ARS: range %d ARS start %s (%d)\n", 2861 nfit_spa->spa->range_index, 2862 req_type == ARS_REQ_SHORT ? "short" : "long", 2863 rc); 2864 /* 2865 * Hmm, we raced someone else starting ARS? Try again in 2866 * a bit. 2867 */ 2868 if (rc == -EBUSY) 2869 return 1; 2870 if (rc == 0) { 2871 dev_WARN_ONCE(dev, acpi_desc->scrub_spa, 2872 "scrub start while range %d active\n", 2873 acpi_desc->scrub_spa->spa->range_index); 2874 clear_bit(req_type, &nfit_spa->ars_state); 2875 acpi_desc->scrub_spa = nfit_spa; 2876 /* 2877 * Consider this spa last for future scrub 2878 * requests 2879 */ 2880 list_move_tail(&nfit_spa->list, &acpi_desc->spas); 2881 return 1; 2882 } 2883 2884 dev_err(dev, "ARS: range %d ARS failed (%d)\n", 2885 nfit_spa->spa->range_index, rc); 2886 set_bit(ARS_FAILED, &nfit_spa->ars_state); 2887 } 2888 return 0; 2889 } 2890 2891 static void __sched_ars(struct acpi_nfit_desc *acpi_desc, unsigned int tmo) 2892 { 2893 lockdep_assert_held(&acpi_desc->init_mutex); 2894 2895 set_bit(ARS_BUSY, &acpi_desc->scrub_flags); 2896 /* note this should only be set from within the workqueue */ 2897 if (tmo) 2898 acpi_desc->scrub_tmo = tmo; 2899 queue_delayed_work(nfit_wq, &acpi_desc->dwork, tmo * HZ); 2900 } 2901 2902 static void sched_ars(struct acpi_nfit_desc *acpi_desc) 2903 { 2904 __sched_ars(acpi_desc, 0); 2905 } 2906 2907 static void notify_ars_done(struct acpi_nfit_desc *acpi_desc) 2908 { 2909 lockdep_assert_held(&acpi_desc->init_mutex); 2910 2911 clear_bit(ARS_BUSY, &acpi_desc->scrub_flags); 2912 acpi_desc->scrub_count++; 2913 if (acpi_desc->scrub_count_state) 2914 sysfs_notify_dirent(acpi_desc->scrub_count_state); 2915 } 2916 2917 static void acpi_nfit_scrub(struct work_struct *work) 2918 { 2919 struct acpi_nfit_desc *acpi_desc; 2920 unsigned int tmo; 2921 int query_rc; 2922 2923 acpi_desc = container_of(work, typeof(*acpi_desc), dwork.work); 2924 mutex_lock(&acpi_desc->init_mutex); 2925 query_rc = acpi_nfit_query_poison(acpi_desc); 2926 tmo = __acpi_nfit_scrub(acpi_desc, query_rc); 2927 if (tmo) 2928 __sched_ars(acpi_desc, tmo); 2929 else 2930 notify_ars_done(acpi_desc); 2931 memset(acpi_desc->ars_status, 0, acpi_desc->max_ars); 2932 clear_bit(ARS_POLL, &acpi_desc->scrub_flags); 2933 mutex_unlock(&acpi_desc->init_mutex); 2934 } 2935 2936 static void acpi_nfit_init_ars(struct acpi_nfit_desc *acpi_desc, 2937 struct nfit_spa *nfit_spa) 2938 { 2939 int type = nfit_spa_type(nfit_spa->spa); 2940 struct nd_cmd_ars_cap ars_cap; 2941 int rc; 2942 2943 set_bit(ARS_FAILED, &nfit_spa->ars_state); 2944 memset(&ars_cap, 0, sizeof(ars_cap)); 2945 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa); 2946 if (rc < 0) 2947 return; 2948 /* check that the supported scrub types match the spa type */ 2949 if (type == NFIT_SPA_VOLATILE && ((ars_cap.status >> 16) 2950 & ND_ARS_VOLATILE) == 0) 2951 return; 2952 if (type == NFIT_SPA_PM && ((ars_cap.status >> 16) 2953 & ND_ARS_PERSISTENT) == 0) 2954 return; 2955 2956 nfit_spa->max_ars = ars_cap.max_ars_out; 2957 nfit_spa->clear_err_unit = ars_cap.clear_err_unit; 2958 acpi_desc->max_ars = max(nfit_spa->max_ars, acpi_desc->max_ars); 2959 clear_bit(ARS_FAILED, &nfit_spa->ars_state); 2960 } 2961 2962 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc) 2963 { 2964 struct nfit_spa *nfit_spa; 2965 int rc, do_sched_ars = 0; 2966 2967 set_bit(ARS_VALID, &acpi_desc->scrub_flags); 2968 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) { 2969 switch (nfit_spa_type(nfit_spa->spa)) { 2970 case NFIT_SPA_VOLATILE: 2971 case NFIT_SPA_PM: 2972 acpi_nfit_init_ars(acpi_desc, nfit_spa); 2973 break; 2974 } 2975 } 2976 2977 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) { 2978 switch (nfit_spa_type(nfit_spa->spa)) { 2979 case NFIT_SPA_VOLATILE: 2980 case NFIT_SPA_PM: 2981 /* register regions and kick off initial ARS run */ 2982 rc = ars_register(acpi_desc, nfit_spa); 2983 if (rc) 2984 return rc; 2985 2986 /* 2987 * Kick off background ARS if at least one 2988 * region successfully registered ARS 2989 */ 2990 if (!test_bit(ARS_FAILED, &nfit_spa->ars_state)) 2991 do_sched_ars++; 2992 break; 2993 case NFIT_SPA_BDW: 2994 /* nothing to register */ 2995 break; 2996 case NFIT_SPA_DCR: 2997 case NFIT_SPA_VDISK: 2998 case NFIT_SPA_VCD: 2999 case NFIT_SPA_PDISK: 3000 case NFIT_SPA_PCD: 3001 /* register known regions that don't support ARS */ 3002 rc = acpi_nfit_register_region(acpi_desc, nfit_spa); 3003 if (rc) 3004 return rc; 3005 break; 3006 default: 3007 /* don't register unknown regions */ 3008 break; 3009 } 3010 } 3011 3012 if (do_sched_ars) 3013 sched_ars(acpi_desc); 3014 return 0; 3015 } 3016 3017 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc, 3018 struct nfit_table_prev *prev) 3019 { 3020 struct device *dev = acpi_desc->dev; 3021 3022 if (!list_empty(&prev->spas) || 3023 !list_empty(&prev->memdevs) || 3024 !list_empty(&prev->dcrs) || 3025 !list_empty(&prev->bdws) || 3026 !list_empty(&prev->idts) || 3027 !list_empty(&prev->flushes)) { 3028 dev_err(dev, "new nfit deletes entries (unsupported)\n"); 3029 return -ENXIO; 3030 } 3031 return 0; 3032 } 3033 3034 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc) 3035 { 3036 struct device *dev = acpi_desc->dev; 3037 struct kernfs_node *nfit; 3038 struct device *bus_dev; 3039 3040 if (!ars_supported(acpi_desc->nvdimm_bus)) 3041 return 0; 3042 3043 bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus); 3044 nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit"); 3045 if (!nfit) { 3046 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n"); 3047 return -ENODEV; 3048 } 3049 acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub"); 3050 sysfs_put(nfit); 3051 if (!acpi_desc->scrub_count_state) { 3052 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n"); 3053 return -ENODEV; 3054 } 3055 3056 return 0; 3057 } 3058 3059 static void acpi_nfit_unregister(void *data) 3060 { 3061 struct acpi_nfit_desc *acpi_desc = data; 3062 3063 nvdimm_bus_unregister(acpi_desc->nvdimm_bus); 3064 } 3065 3066 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz) 3067 { 3068 struct device *dev = acpi_desc->dev; 3069 struct nfit_table_prev prev; 3070 const void *end; 3071 int rc; 3072 3073 if (!acpi_desc->nvdimm_bus) { 3074 acpi_nfit_init_dsms(acpi_desc); 3075 3076 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev, 3077 &acpi_desc->nd_desc); 3078 if (!acpi_desc->nvdimm_bus) 3079 return -ENOMEM; 3080 3081 rc = devm_add_action_or_reset(dev, acpi_nfit_unregister, 3082 acpi_desc); 3083 if (rc) 3084 return rc; 3085 3086 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc); 3087 if (rc) 3088 return rc; 3089 3090 /* register this acpi_desc for mce notifications */ 3091 mutex_lock(&acpi_desc_lock); 3092 list_add_tail(&acpi_desc->list, &acpi_descs); 3093 mutex_unlock(&acpi_desc_lock); 3094 } 3095 3096 mutex_lock(&acpi_desc->init_mutex); 3097 3098 INIT_LIST_HEAD(&prev.spas); 3099 INIT_LIST_HEAD(&prev.memdevs); 3100 INIT_LIST_HEAD(&prev.dcrs); 3101 INIT_LIST_HEAD(&prev.bdws); 3102 INIT_LIST_HEAD(&prev.idts); 3103 INIT_LIST_HEAD(&prev.flushes); 3104 3105 list_cut_position(&prev.spas, &acpi_desc->spas, 3106 acpi_desc->spas.prev); 3107 list_cut_position(&prev.memdevs, &acpi_desc->memdevs, 3108 acpi_desc->memdevs.prev); 3109 list_cut_position(&prev.dcrs, &acpi_desc->dcrs, 3110 acpi_desc->dcrs.prev); 3111 list_cut_position(&prev.bdws, &acpi_desc->bdws, 3112 acpi_desc->bdws.prev); 3113 list_cut_position(&prev.idts, &acpi_desc->idts, 3114 acpi_desc->idts.prev); 3115 list_cut_position(&prev.flushes, &acpi_desc->flushes, 3116 acpi_desc->flushes.prev); 3117 3118 end = data + sz; 3119 while (!IS_ERR_OR_NULL(data)) 3120 data = add_table(acpi_desc, &prev, data, end); 3121 3122 if (IS_ERR(data)) { 3123 dev_dbg(dev, "nfit table parsing error: %ld\n", PTR_ERR(data)); 3124 rc = PTR_ERR(data); 3125 goto out_unlock; 3126 } 3127 3128 rc = acpi_nfit_check_deletions(acpi_desc, &prev); 3129 if (rc) 3130 goto out_unlock; 3131 3132 rc = nfit_mem_init(acpi_desc); 3133 if (rc) 3134 goto out_unlock; 3135 3136 rc = acpi_nfit_register_dimms(acpi_desc); 3137 if (rc) 3138 goto out_unlock; 3139 3140 rc = acpi_nfit_register_regions(acpi_desc); 3141 3142 out_unlock: 3143 mutex_unlock(&acpi_desc->init_mutex); 3144 return rc; 3145 } 3146 EXPORT_SYMBOL_GPL(acpi_nfit_init); 3147 3148 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc) 3149 { 3150 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc); 3151 struct device *dev = acpi_desc->dev; 3152 3153 /* Bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */ 3154 device_lock(dev); 3155 device_unlock(dev); 3156 3157 /* Bounce the init_mutex to complete initial registration */ 3158 mutex_lock(&acpi_desc->init_mutex); 3159 mutex_unlock(&acpi_desc->init_mutex); 3160 3161 return 0; 3162 } 3163 3164 static int __acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc, 3165 struct nvdimm *nvdimm, unsigned int cmd) 3166 { 3167 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc); 3168 3169 if (nvdimm) 3170 return 0; 3171 if (cmd != ND_CMD_ARS_START) 3172 return 0; 3173 3174 /* 3175 * The kernel and userspace may race to initiate a scrub, but 3176 * the scrub thread is prepared to lose that initial race. It 3177 * just needs guarantees that any ARS it initiates are not 3178 * interrupted by any intervening start requests from userspace. 3179 */ 3180 if (work_busy(&acpi_desc->dwork.work)) 3181 return -EBUSY; 3182 3183 return 0; 3184 } 3185 3186 /* 3187 * Prevent security and firmware activate commands from being issued via 3188 * ioctl. 3189 */ 3190 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc, 3191 struct nvdimm *nvdimm, unsigned int cmd, void *buf) 3192 { 3193 struct nd_cmd_pkg *call_pkg = buf; 3194 unsigned int func; 3195 3196 if (nvdimm && cmd == ND_CMD_CALL && 3197 call_pkg->nd_family == NVDIMM_FAMILY_INTEL) { 3198 func = call_pkg->nd_command; 3199 if (func > NVDIMM_CMD_MAX || 3200 (1 << func) & NVDIMM_INTEL_DENY_CMDMASK) 3201 return -EOPNOTSUPP; 3202 } 3203 3204 /* block all non-nfit bus commands */ 3205 if (!nvdimm && cmd == ND_CMD_CALL && 3206 call_pkg->nd_family != NVDIMM_BUS_FAMILY_NFIT) 3207 return -EOPNOTSUPP; 3208 3209 return __acpi_nfit_clear_to_send(nd_desc, nvdimm, cmd); 3210 } 3211 3212 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc, 3213 enum nfit_ars_state req_type) 3214 { 3215 struct device *dev = acpi_desc->dev; 3216 int scheduled = 0, busy = 0; 3217 struct nfit_spa *nfit_spa; 3218 3219 mutex_lock(&acpi_desc->init_mutex); 3220 if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags)) { 3221 mutex_unlock(&acpi_desc->init_mutex); 3222 return 0; 3223 } 3224 3225 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) { 3226 int type = nfit_spa_type(nfit_spa->spa); 3227 3228 if (type != NFIT_SPA_PM && type != NFIT_SPA_VOLATILE) 3229 continue; 3230 if (test_bit(ARS_FAILED, &nfit_spa->ars_state)) 3231 continue; 3232 3233 if (test_and_set_bit(req_type, &nfit_spa->ars_state)) 3234 busy++; 3235 else 3236 scheduled++; 3237 } 3238 if (scheduled) { 3239 sched_ars(acpi_desc); 3240 dev_dbg(dev, "ars_scan triggered\n"); 3241 } 3242 mutex_unlock(&acpi_desc->init_mutex); 3243 3244 if (scheduled) 3245 return 0; 3246 if (busy) 3247 return -EBUSY; 3248 return -ENOTTY; 3249 } 3250 3251 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev) 3252 { 3253 struct nvdimm_bus_descriptor *nd_desc; 3254 3255 dev_set_drvdata(dev, acpi_desc); 3256 acpi_desc->dev = dev; 3257 nd_desc = &acpi_desc->nd_desc; 3258 nd_desc->provider_name = "ACPI.NFIT"; 3259 nd_desc->module = THIS_MODULE; 3260 nd_desc->ndctl = acpi_nfit_ctl; 3261 nd_desc->flush_probe = acpi_nfit_flush_probe; 3262 nd_desc->clear_to_send = acpi_nfit_clear_to_send; 3263 nd_desc->attr_groups = acpi_nfit_attribute_groups; 3264 3265 INIT_LIST_HEAD(&acpi_desc->spas); 3266 INIT_LIST_HEAD(&acpi_desc->dcrs); 3267 INIT_LIST_HEAD(&acpi_desc->bdws); 3268 INIT_LIST_HEAD(&acpi_desc->idts); 3269 INIT_LIST_HEAD(&acpi_desc->flushes); 3270 INIT_LIST_HEAD(&acpi_desc->memdevs); 3271 INIT_LIST_HEAD(&acpi_desc->dimms); 3272 INIT_LIST_HEAD(&acpi_desc->list); 3273 mutex_init(&acpi_desc->init_mutex); 3274 acpi_desc->scrub_tmo = 1; 3275 INIT_DELAYED_WORK(&acpi_desc->dwork, acpi_nfit_scrub); 3276 } 3277 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init); 3278 3279 static void acpi_nfit_put_table(void *table) 3280 { 3281 acpi_put_table(table); 3282 } 3283 3284 static void acpi_nfit_notify(acpi_handle handle, u32 event, void *data) 3285 { 3286 struct acpi_device *adev = data; 3287 3288 device_lock(&adev->dev); 3289 __acpi_nfit_notify(&adev->dev, handle, event); 3290 device_unlock(&adev->dev); 3291 } 3292 3293 static void acpi_nfit_remove_notify_handler(void *data) 3294 { 3295 struct acpi_device *adev = data; 3296 3297 acpi_dev_remove_notify_handler(adev, ACPI_DEVICE_NOTIFY, 3298 acpi_nfit_notify); 3299 } 3300 3301 void acpi_nfit_shutdown(void *data) 3302 { 3303 struct acpi_nfit_desc *acpi_desc = data; 3304 struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus); 3305 3306 /* 3307 * Destruct under acpi_desc_lock so that nfit_handle_mce does not 3308 * race teardown 3309 */ 3310 mutex_lock(&acpi_desc_lock); 3311 list_del(&acpi_desc->list); 3312 mutex_unlock(&acpi_desc_lock); 3313 3314 mutex_lock(&acpi_desc->init_mutex); 3315 set_bit(ARS_CANCEL, &acpi_desc->scrub_flags); 3316 mutex_unlock(&acpi_desc->init_mutex); 3317 cancel_delayed_work_sync(&acpi_desc->dwork); 3318 3319 /* 3320 * Bounce the nvdimm bus lock to make sure any in-flight 3321 * acpi_nfit_ars_rescan() submissions have had a chance to 3322 * either submit or see ->cancel set. 3323 */ 3324 device_lock(bus_dev); 3325 device_unlock(bus_dev); 3326 3327 flush_workqueue(nfit_wq); 3328 } 3329 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown); 3330 3331 static int acpi_nfit_add(struct acpi_device *adev) 3332 { 3333 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL }; 3334 struct acpi_nfit_desc *acpi_desc; 3335 struct device *dev = &adev->dev; 3336 struct acpi_table_header *tbl; 3337 acpi_status status = AE_OK; 3338 acpi_size sz; 3339 int rc = 0; 3340 3341 rc = acpi_dev_install_notify_handler(adev, ACPI_DEVICE_NOTIFY, 3342 acpi_nfit_notify, adev); 3343 if (rc) 3344 return rc; 3345 3346 rc = devm_add_action_or_reset(dev, acpi_nfit_remove_notify_handler, 3347 adev); 3348 if (rc) 3349 return rc; 3350 3351 status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl); 3352 if (ACPI_FAILURE(status)) { 3353 /* The NVDIMM root device allows OS to trigger enumeration of 3354 * NVDIMMs through NFIT at boot time and re-enumeration at 3355 * root level via the _FIT method during runtime. 3356 * This is ok to return 0 here, we could have an nvdimm 3357 * hotplugged later and evaluate _FIT method which returns 3358 * data in the format of a series of NFIT Structures. 3359 */ 3360 dev_dbg(dev, "failed to find NFIT at startup\n"); 3361 return 0; 3362 } 3363 3364 rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl); 3365 if (rc) 3366 return rc; 3367 sz = tbl->length; 3368 3369 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL); 3370 if (!acpi_desc) 3371 return -ENOMEM; 3372 acpi_nfit_desc_init(acpi_desc, &adev->dev); 3373 3374 /* Save the acpi header for exporting the revision via sysfs */ 3375 acpi_desc->acpi_header = *tbl; 3376 3377 /* Evaluate _FIT and override with that if present */ 3378 status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf); 3379 if (ACPI_SUCCESS(status) && buf.length > 0) { 3380 union acpi_object *obj = buf.pointer; 3381 3382 if (obj->type == ACPI_TYPE_BUFFER) 3383 rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer, 3384 obj->buffer.length); 3385 else 3386 dev_dbg(dev, "invalid type %d, ignoring _FIT\n", 3387 (int) obj->type); 3388 kfree(buf.pointer); 3389 } else 3390 /* skip over the lead-in header table */ 3391 rc = acpi_nfit_init(acpi_desc, (void *) tbl 3392 + sizeof(struct acpi_table_nfit), 3393 sz - sizeof(struct acpi_table_nfit)); 3394 3395 if (rc) 3396 return rc; 3397 3398 return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc); 3399 } 3400 3401 static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle) 3402 { 3403 struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev); 3404 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL }; 3405 union acpi_object *obj; 3406 acpi_status status; 3407 int ret; 3408 3409 if (!dev->driver) { 3410 /* dev->driver may be null if we're being removed */ 3411 dev_dbg(dev, "no driver found for dev\n"); 3412 return; 3413 } 3414 3415 if (!acpi_desc) { 3416 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL); 3417 if (!acpi_desc) 3418 return; 3419 acpi_nfit_desc_init(acpi_desc, dev); 3420 } else { 3421 /* 3422 * Finish previous registration before considering new 3423 * regions. 3424 */ 3425 flush_workqueue(nfit_wq); 3426 } 3427 3428 /* Evaluate _FIT */ 3429 status = acpi_evaluate_object(handle, "_FIT", NULL, &buf); 3430 if (ACPI_FAILURE(status)) { 3431 dev_err(dev, "failed to evaluate _FIT\n"); 3432 return; 3433 } 3434 3435 obj = buf.pointer; 3436 if (obj->type == ACPI_TYPE_BUFFER) { 3437 ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer, 3438 obj->buffer.length); 3439 if (ret) 3440 dev_err(dev, "failed to merge updated NFIT\n"); 3441 } else 3442 dev_err(dev, "Invalid _FIT\n"); 3443 kfree(buf.pointer); 3444 } 3445 3446 static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle) 3447 { 3448 struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev); 3449 3450 if (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON) 3451 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG); 3452 else 3453 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_SHORT); 3454 } 3455 3456 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event) 3457 { 3458 dev_dbg(dev, "event: 0x%x\n", event); 3459 3460 switch (event) { 3461 case NFIT_NOTIFY_UPDATE: 3462 return acpi_nfit_update_notify(dev, handle); 3463 case NFIT_NOTIFY_UC_MEMORY_ERROR: 3464 return acpi_nfit_uc_error_notify(dev, handle); 3465 default: 3466 return; 3467 } 3468 } 3469 EXPORT_SYMBOL_GPL(__acpi_nfit_notify); 3470 3471 static const struct acpi_device_id acpi_nfit_ids[] = { 3472 { "ACPI0012", 0 }, 3473 { "", 0 }, 3474 }; 3475 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids); 3476 3477 static struct acpi_driver acpi_nfit_driver = { 3478 .name = KBUILD_MODNAME, 3479 .ids = acpi_nfit_ids, 3480 .ops = { 3481 .add = acpi_nfit_add, 3482 }, 3483 }; 3484 3485 static __init int nfit_init(void) 3486 { 3487 int ret; 3488 3489 BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40); 3490 BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 64); 3491 BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48); 3492 BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 16); 3493 BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 8); 3494 BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80); 3495 BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40); 3496 BUILD_BUG_ON(sizeof(struct acpi_nfit_capabilities) != 16); 3497 3498 guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]); 3499 guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]); 3500 guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]); 3501 guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]); 3502 guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]); 3503 guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]); 3504 guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]); 3505 guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]); 3506 guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]); 3507 guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]); 3508 guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]); 3509 guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]); 3510 guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]); 3511 guid_parse(UUID_NFIT_DIMM_N_HYPERV, &nfit_uuid[NFIT_DEV_DIMM_N_HYPERV]); 3512 guid_parse(UUID_INTEL_BUS, &nfit_uuid[NFIT_BUS_INTEL]); 3513 3514 nfit_wq = create_singlethread_workqueue("nfit"); 3515 if (!nfit_wq) 3516 return -ENOMEM; 3517 3518 nfit_mce_register(); 3519 ret = acpi_bus_register_driver(&acpi_nfit_driver); 3520 if (ret) { 3521 nfit_mce_unregister(); 3522 destroy_workqueue(nfit_wq); 3523 } 3524 3525 return ret; 3526 3527 } 3528 3529 static __exit void nfit_exit(void) 3530 { 3531 nfit_mce_unregister(); 3532 acpi_bus_unregister_driver(&acpi_nfit_driver); 3533 destroy_workqueue(nfit_wq); 3534 WARN_ON(!list_empty(&acpi_descs)); 3535 } 3536 3537 module_init(nfit_init); 3538 module_exit(nfit_exit); 3539 MODULE_DESCRIPTION("ACPI NVDIMM Firmware Interface Table (NFIT) driver"); 3540 MODULE_LICENSE("GPL v2"); 3541 MODULE_AUTHOR("Intel Corporation"); 3542