drivers/acpi/nfit/core.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
   4  */
   5 #include <linux/list_sort.h>
   6 #include <linux/libnvdimm.h>
   7 #include <linux/module.h>
   8 #include <linux/nospec.h>
   9 #include <linux/mutex.h>
  10 #include <linux/ndctl.h>
  11 #include <linux/sysfs.h>
  12 #include <linux/delay.h>
  13 #include <linux/list.h>
  14 #include <linux/acpi.h>
  15 #include <linux/sort.h>
  16 #include <linux/io.h>
  17 #include <linux/nd.h>
  18 #include <asm/cacheflush.h>
  19 #include <acpi/nfit.h>
  20 #include "intel.h"
  21 #include "nfit.h"
  22
  23 /*
  24  * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
  25  * irrelevant.
  26  */
  27 #include <linux/io-64-nonatomic-hi-lo.h>
  28
  29 static bool force_enable_dimms;
  30 module_param(force_enable_dimms, bool, S_IRUGO|S_IWUSR);
  31 MODULE_PARM_DESC(force_enable_dimms, "Ignore _STA (ACPI DIMM device) status");
  32
  33 static bool disable_vendor_specific;
  34 module_param(disable_vendor_specific, bool, S_IRUGO);
  35 MODULE_PARM_DESC(disable_vendor_specific,
  36                 "Limit commands to the publicly specified set");
  37
  38 static unsigned long override_dsm_mask;
  39 module_param(override_dsm_mask, ulong, S_IRUGO);
  40 MODULE_PARM_DESC(override_dsm_mask, "Bitmask of allowed NVDIMM DSM functions");
  41
  42 static int default_dsm_family = -1;
  43 module_param(default_dsm_family, int, S_IRUGO);
  44 MODULE_PARM_DESC(default_dsm_family,
  45                 "Try this DSM type first when identifying NVDIMM family");
  46
  47 static bool no_init_ars;
  48 module_param(no_init_ars, bool, 0644);
  49 MODULE_PARM_DESC(no_init_ars, "Skip ARS run at nfit init time");
  50
  51 static bool force_labels;
  52 module_param(force_labels, bool, 0444);
  53 MODULE_PARM_DESC(force_labels, "Opt-in to labels despite missing methods");
  54
  55 LIST_HEAD(acpi_descs);
  56 DEFINE_MUTEX(acpi_desc_lock);
  57
  58 static struct workqueue_struct *nfit_wq;
  59
  60 struct nfit_table_prev {
  61         struct list_head spas;
  62         struct list_head memdevs;
  63         struct list_head dcrs;
  64         struct list_head bdws;
  65         struct list_head idts;
  66         struct list_head flushes;
  67 };
  68
  69 static guid_t nfit_uuid[NFIT_UUID_MAX];
  70
  71 const guid_t *to_nfit_uuid(enum nfit_uuids id)
  72 {
  73         return &nfit_uuid[id];
  74 }
  75 EXPORT_SYMBOL(to_nfit_uuid);
  76
  77 static const guid_t *to_nfit_bus_uuid(int family)
  78 {
  79         if (WARN_ONCE(family == NVDIMM_BUS_FAMILY_NFIT,
  80                         "only secondary bus families can be translated\n"))
  81                 return NULL;
  82         /*
  83          * The index of bus UUIDs starts immediately following the last
  84          * NVDIMM/leaf family.
  85          */
  86         return to_nfit_uuid(family + NVDIMM_FAMILY_MAX);
  87 }
  88
  89 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
  90 {
  91         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
  92
  93         /*
  94          * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
  95          * acpi_device.
  96          */
  97         if (!nd_desc->provider_name
  98                         || strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
  99                 return NULL;
 100
 101         return to_acpi_device(acpi_desc->dev);
 102 }
 103
 104 static int xlat_bus_status(void *buf, unsigned int cmd, u32 status)
 105 {
 106         struct nd_cmd_clear_error *clear_err;
 107         struct nd_cmd_ars_status *ars_status;
 108         u16 flags;
 109
 110         switch (cmd) {
 111         case ND_CMD_ARS_CAP:
 112                 if ((status & 0xffff) == NFIT_ARS_CAP_NONE)
 113                         return -ENOTTY;
 114
 115                 /* Command failed */
 116                 if (status & 0xffff)
 117                         return -EIO;
 118
 119                 /* No supported scan types for this range */
 120                 flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
 121                 if ((status >> 16 & flags) == 0)
 122                         return -ENOTTY;
 123                 return 0;
 124         case ND_CMD_ARS_START:
 125                 /* ARS is in progress */
 126                 if ((status & 0xffff) == NFIT_ARS_START_BUSY)
 127                         return -EBUSY;
 128
 129                 /* Command failed */
 130                 if (status & 0xffff)
 131                         return -EIO;
 132                 return 0;
 133         case ND_CMD_ARS_STATUS:
 134                 ars_status = buf;
 135                 /* Command failed */
 136                 if (status & 0xffff)
 137                         return -EIO;
 138                 /* Check extended status (Upper two bytes) */
 139                 if (status == NFIT_ARS_STATUS_DONE)
 140                         return 0;
 141
 142                 /* ARS is in progress */
 143                 if (status == NFIT_ARS_STATUS_BUSY)
 144                         return -EBUSY;
 145
 146                 /* No ARS performed for the current boot */
 147                 if (status == NFIT_ARS_STATUS_NONE)
 148                         return -EAGAIN;
 149
 150                 /*
 151                  * ARS interrupted, either we overflowed or some other
 152                  * agent wants the scan to stop.  If we didn't overflow
 153                  * then just continue with the returned results.
 154                  */
 155                 if (status == NFIT_ARS_STATUS_INTR) {
 156                         if (ars_status->out_length >= 40 && (ars_status->flags
 157                                                 & NFIT_ARS_F_OVERFLOW))
 158                                 return -ENOSPC;
 159                         return 0;
 160                 }
 161
 162                 /* Unknown status */
 163                 if (status >> 16)
 164                         return -EIO;
 165                 return 0;
 166         case ND_CMD_CLEAR_ERROR:
 167                 clear_err = buf;
 168                 if (status & 0xffff)
 169                         return -EIO;
 170                 if (!clear_err->cleared)
 171                         return -EIO;
 172                 if (clear_err->length > clear_err->cleared)
 173                         return clear_err->cleared;
 174                 return 0;
 175         default:
 176                 break;
 177         }
 178
 179         /* all other non-zero status results in an error */
 180         if (status)
 181                 return -EIO;
 182         return 0;
 183 }
 184
 185 #define ACPI_LABELS_LOCKED 3
 186
 187 static int xlat_nvdimm_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
 188                 u32 status)
 189 {
 190         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 191
 192         switch (cmd) {
 193         case ND_CMD_GET_CONFIG_SIZE:
 194                 /*
 195                  * In the _LSI, _LSR, _LSW case the locked status is
 196                  * communicated via the read/write commands
 197                  */
 198                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags))
 199                         break;
 200
 201                 if (status >> 16 & ND_CONFIG_LOCKED)
 202                         return -EACCES;
 203                 break;
 204         case ND_CMD_GET_CONFIG_DATA:
 205                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
 206                                 && status == ACPI_LABELS_LOCKED)
 207                         return -EACCES;
 208                 break;
 209         case ND_CMD_SET_CONFIG_DATA:
 210                 if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags)
 211                                 && status == ACPI_LABELS_LOCKED)
 212                         return -EACCES;
 213                 break;
 214         default:
 215                 break;
 216         }
 217
 218         /* all other non-zero status results in an error */
 219         if (status)
 220                 return -EIO;
 221         return 0;
 222 }
 223
 224 static int xlat_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
 225                 u32 status)
 226 {
 227         if (!nvdimm)
 228                 return xlat_bus_status(buf, cmd, status);
 229         return xlat_nvdimm_status(nvdimm, buf, cmd, status);
 230 }
 231
 232 /* convert _LS{I,R} packages to the buffer object acpi_nfit_ctl expects */
 233 static union acpi_object *pkg_to_buf(union acpi_object *pkg)
 234 {
 235         int i;
 236         void *dst;
 237         size_t size = 0;
 238         union acpi_object *buf = NULL;
 239
 240         if (pkg->type != ACPI_TYPE_PACKAGE) {
 241                 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
 242                                 pkg->type);
 243                 goto err;
 244         }
 245
 246         for (i = 0; i < pkg->package.count; i++) {
 247                 union acpi_object *obj = &pkg->package.elements[i];
 248
 249                 if (obj->type == ACPI_TYPE_INTEGER)
 250                         size += 4;
 251                 else if (obj->type == ACPI_TYPE_BUFFER)
 252                         size += obj->buffer.length;
 253                 else {
 254                         WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
 255                                         obj->type);
 256                         goto err;
 257                 }
 258         }
 259
 260         buf = ACPI_ALLOCATE(sizeof(*buf) + size);
 261         if (!buf)
 262                 goto err;
 263
 264         dst = buf + 1;
 265         buf->type = ACPI_TYPE_BUFFER;
 266         buf->buffer.length = size;
 267         buf->buffer.pointer = dst;
 268         for (i = 0; i < pkg->package.count; i++) {
 269                 union acpi_object *obj = &pkg->package.elements[i];
 270
 271                 if (obj->type == ACPI_TYPE_INTEGER) {
 272                         memcpy(dst, &obj->integer.value, 4);
 273                         dst += 4;
 274                 } else if (obj->type == ACPI_TYPE_BUFFER) {
 275                         memcpy(dst, obj->buffer.pointer, obj->buffer.length);
 276                         dst += obj->buffer.length;
 277                 }
 278         }
 279 err:
 280         ACPI_FREE(pkg);
 281         return buf;
 282 }
 283
 284 static union acpi_object *int_to_buf(union acpi_object *integer)
 285 {
 286         union acpi_object *buf = NULL;
 287         void *dst = NULL;
 288
 289         if (integer->type != ACPI_TYPE_INTEGER) {
 290                 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
 291                                 integer->type);
 292                 goto err;
 293         }
 294
 295         buf = ACPI_ALLOCATE(sizeof(*buf) + 4);
 296         if (!buf)
 297                 goto err;
 298
 299         dst = buf + 1;
 300         buf->type = ACPI_TYPE_BUFFER;
 301         buf->buffer.length = 4;
 302         buf->buffer.pointer = dst;
 303         memcpy(dst, &integer->integer.value, 4);
 304 err:
 305         ACPI_FREE(integer);
 306         return buf;
 307 }
 308
 309 static union acpi_object *acpi_label_write(acpi_handle handle, u32 offset,
 310                 u32 len, void *data)
 311 {
 312         acpi_status rc;
 313         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
 314         struct acpi_object_list input = {
 315                 .count = 3,
 316                 .pointer = (union acpi_object []) {
 317                         [0] = {
 318                                 .integer.type = ACPI_TYPE_INTEGER,
 319                                 .integer.value = offset,
 320                         },
 321                         [1] = {
 322                                 .integer.type = ACPI_TYPE_INTEGER,
 323                                 .integer.value = len,
 324                         },
 325                         [2] = {
 326                                 .buffer.type = ACPI_TYPE_BUFFER,
 327                                 .buffer.pointer = data,
 328                                 .buffer.length = len,
 329                         },
 330                 },
 331         };
 332
 333         rc = acpi_evaluate_object(handle, "_LSW", &input, &buf);
 334         if (ACPI_FAILURE(rc))
 335                 return NULL;
 336         return int_to_buf(buf.pointer);
 337 }
 338
 339 static union acpi_object *acpi_label_read(acpi_handle handle, u32 offset,
 340                 u32 len)
 341 {
 342         acpi_status rc;
 343         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
 344         struct acpi_object_list input = {
 345                 .count = 2,
 346                 .pointer = (union acpi_object []) {
 347                         [0] = {
 348                                 .integer.type = ACPI_TYPE_INTEGER,
 349                                 .integer.value = offset,
 350                         },
 351                         [1] = {
 352                                 .integer.type = ACPI_TYPE_INTEGER,
 353                                 .integer.value = len,
 354                         },
 355                 },
 356         };
 357
 358         rc = acpi_evaluate_object(handle, "_LSR", &input, &buf);
 359         if (ACPI_FAILURE(rc))
 360                 return NULL;
 361         return pkg_to_buf(buf.pointer);
 362 }
 363
 364 static union acpi_object *acpi_label_info(acpi_handle handle)
 365 {
 366         acpi_status rc;
 367         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
 368
 369         rc = acpi_evaluate_object(handle, "_LSI", NULL, &buf);
 370         if (ACPI_FAILURE(rc))
 371                 return NULL;
 372         return pkg_to_buf(buf.pointer);
 373 }
 374
 375 static u8 nfit_dsm_revid(unsigned family, unsigned func)
 376 {
 377         static const u8 revid_table[NVDIMM_FAMILY_MAX+1][NVDIMM_CMD_MAX+1] = {
 378                 [NVDIMM_FAMILY_INTEL] = {
 379                         [NVDIMM_INTEL_GET_MODES ...
 380                                 NVDIMM_INTEL_FW_ACTIVATE_ARM] = 2,
 381                 },
 382         };
 383         u8 id;
 384
 385         if (family > NVDIMM_FAMILY_MAX)
 386                 return 0;
 387         if (func > NVDIMM_CMD_MAX)
 388                 return 0;
 389         id = revid_table[family][func];
 390         if (id == 0)
 391                 return 1; /* default */
 392         return id;
 393 }
 394
 395 static bool payload_dumpable(struct nvdimm *nvdimm, unsigned int func)
 396 {
 397         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 398
 399         if (nfit_mem && nfit_mem->family == NVDIMM_FAMILY_INTEL
 400                         && func >= NVDIMM_INTEL_GET_SECURITY_STATE
 401                         && func <= NVDIMM_INTEL_MASTER_SECURE_ERASE)
 402                 return IS_ENABLED(CONFIG_NFIT_SECURITY_DEBUG);
 403         return true;
 404 }
 405
 406 static int cmd_to_func(struct nfit_mem *nfit_mem, unsigned int cmd,
 407                 struct nd_cmd_pkg *call_pkg, int *family)
 408 {
 409         if (call_pkg) {
 410                 int i;
 411
 412                 if (nfit_mem && nfit_mem->family != call_pkg->nd_family)
 413                         return -ENOTTY;
 414
 415                 for (i = 0; i < ARRAY_SIZE(call_pkg->nd_reserved2); i++)
 416                         if (call_pkg->nd_reserved2[i])
 417                                 return -EINVAL;
 418                 *family = call_pkg->nd_family;
 419                 return call_pkg->nd_command;
 420         }
 421
 422         /* In the !call_pkg case, bus commands == bus functions */
 423         if (!nfit_mem)
 424                 return cmd;
 425
 426         /* Linux ND commands == NVDIMM_FAMILY_INTEL function numbers */
 427         if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
 428                 return cmd;
 429
 430         /*
 431          * Force function number validation to fail since 0 is never
 432          * published as a valid function in dsm_mask.
 433          */
 434         return 0;
 435 }
 436
 437 int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm,
 438                 unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc)
 439 {
 440         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
 441         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 442         union acpi_object in_obj, in_buf, *out_obj;
 443         const struct nd_cmd_desc *desc = NULL;
 444         struct device *dev = acpi_desc->dev;
 445         struct nd_cmd_pkg *call_pkg = NULL;
 446         const char *cmd_name, *dimm_name;
 447         unsigned long cmd_mask, dsm_mask;
 448         u32 offset, fw_status = 0;
 449         acpi_handle handle;
 450         const guid_t *guid;
 451         int func, rc, i;
 452         int family = 0;
 453
 454         if (cmd_rc)
 455                 *cmd_rc = -EINVAL;
 456
 457         if (cmd == ND_CMD_CALL)
 458                 call_pkg = buf;
 459         func = cmd_to_func(nfit_mem, cmd, call_pkg, &family);
 460         if (func < 0)
 461                 return func;
 462
 463         if (nvdimm) {
 464                 struct acpi_device *adev = nfit_mem->adev;
 465
 466                 if (!adev)
 467                         return -ENOTTY;
 468
 469                 dimm_name = nvdimm_name(nvdimm);
 470                 cmd_name = nvdimm_cmd_name(cmd);
 471                 cmd_mask = nvdimm_cmd_mask(nvdimm);
 472                 dsm_mask = nfit_mem->dsm_mask;
 473                 desc = nd_cmd_dimm_desc(cmd);
 474                 guid = to_nfit_uuid(nfit_mem->family);
 475                 handle = adev->handle;
 476         } else {
 477                 struct acpi_device *adev = to_acpi_dev(acpi_desc);
 478
 479                 cmd_name = nvdimm_bus_cmd_name(cmd);
 480                 cmd_mask = nd_desc->cmd_mask;
 481                 if (cmd == ND_CMD_CALL && call_pkg->nd_family) {
 482                         family = call_pkg->nd_family;
 483                         if (family > NVDIMM_BUS_FAMILY_MAX ||
 484                             !test_bit(family, &nd_desc->bus_family_mask))
 485                                 return -EINVAL;
 486                         family = array_index_nospec(family,
 487                                                     NVDIMM_BUS_FAMILY_MAX + 1);
 488                         dsm_mask = acpi_desc->family_dsm_mask[family];
 489                         guid = to_nfit_bus_uuid(family);
 490                 } else {
 491                         dsm_mask = acpi_desc->bus_dsm_mask;
 492                         guid = to_nfit_uuid(NFIT_DEV_BUS);
 493                 }
 494                 desc = nd_cmd_bus_desc(cmd);
 495                 handle = adev->handle;
 496                 dimm_name = "bus";
 497         }
 498
 499         if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
 500                 return -ENOTTY;
 501
 502         /*
 503          * Check for a valid command.  For ND_CMD_CALL, we also have to
 504          * make sure that the DSM function is supported.
 505          */
 506         if (cmd == ND_CMD_CALL &&
 507             (func > NVDIMM_CMD_MAX || !test_bit(func, &dsm_mask)))
 508                 return -ENOTTY;
 509         else if (!test_bit(cmd, &cmd_mask))
 510                 return -ENOTTY;
 511
 512         in_obj.type = ACPI_TYPE_PACKAGE;
 513         in_obj.package.count = 1;
 514         in_obj.package.elements = &in_buf;
 515         in_buf.type = ACPI_TYPE_BUFFER;
 516         in_buf.buffer.pointer = buf;
 517         in_buf.buffer.length = 0;
 518
 519         /* libnvdimm has already validated the input envelope */
 520         for (i = 0; i < desc->in_num; i++)
 521                 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
 522                                 i, buf);
 523
 524         if (call_pkg) {
 525                 /* skip over package wrapper */
 526                 in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
 527                 in_buf.buffer.length = call_pkg->nd_size_in;
 528         }
 529
 530         dev_dbg(dev, "%s cmd: %d: family: %d func: %d input length: %d\n",
 531                 dimm_name, cmd, family, func, in_buf.buffer.length);
 532         if (payload_dumpable(nvdimm, func))
 533                 print_hex_dump_debug("nvdimm in  ", DUMP_PREFIX_OFFSET, 4, 4,
 534                                 in_buf.buffer.pointer,
 535                                 min_t(u32, 256, in_buf.buffer.length), true);
 536
 537         /* call the BIOS, prefer the named methods over _DSM if available */
 538         if (nvdimm && cmd == ND_CMD_GET_CONFIG_SIZE
 539                         && test_bit(NFIT_MEM_LSR, &nfit_mem->flags))
 540                 out_obj = acpi_label_info(handle);
 541         else if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA
 542                         && test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
 543                 struct nd_cmd_get_config_data_hdr *p = buf;
 544
 545                 out_obj = acpi_label_read(handle, p->in_offset, p->in_length);
 546         } else if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA
 547                         && test_bit(NFIT_MEM_LSW, &nfit_mem->flags)) {
 548                 struct nd_cmd_set_config_hdr *p = buf;
 549
 550                 out_obj = acpi_label_write(handle, p->in_offset, p->in_length,
 551                                 p->in_buf);
 552         } else {
 553                 u8 revid;
 554
 555                 if (nvdimm)
 556                         revid = nfit_dsm_revid(nfit_mem->family, func);
 557                 else
 558                         revid = 1;
 559                 out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
 560         }
 561
 562         if (!out_obj) {
 563                 dev_dbg(dev, "%s _DSM failed cmd: %s\n", dimm_name, cmd_name);
 564                 return -EINVAL;
 565         }
 566
 567         if (out_obj->type != ACPI_TYPE_BUFFER) {
 568                 dev_dbg(dev, "%s unexpected output object type cmd: %s type: %d\n",
 569                                 dimm_name, cmd_name, out_obj->type);
 570                 rc = -EINVAL;
 571                 goto out;
 572         }
 573
 574         dev_dbg(dev, "%s cmd: %s output length: %d\n", dimm_name,
 575                         cmd_name, out_obj->buffer.length);
 576         print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4, 4,
 577                         out_obj->buffer.pointer,
 578                         min_t(u32, 128, out_obj->buffer.length), true);
 579
 580         if (call_pkg) {
 581                 call_pkg->nd_fw_size = out_obj->buffer.length;
 582                 memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
 583                         out_obj->buffer.pointer,
 584                         min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
 585
 586                 ACPI_FREE(out_obj);
 587                 /*
 588                  * Need to support FW function w/o known size in advance.
 589                  * Caller can determine required size based upon nd_fw_size.
 590                  * If we return an error (like elsewhere) then caller wouldn't
 591                  * be able to rely upon data returned to make calculation.
 592                  */
 593                 if (cmd_rc)
 594                         *cmd_rc = 0;
 595                 return 0;
 596         }
 597
 598         for (i = 0, offset = 0; i < desc->out_num; i++) {
 599                 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
 600                                 (u32 *) out_obj->buffer.pointer,
 601                                 out_obj->buffer.length - offset);
 602
 603                 if (offset + out_size > out_obj->buffer.length) {
 604                         dev_dbg(dev, "%s output object underflow cmd: %s field: %d\n",
 605                                         dimm_name, cmd_name, i);
 606                         break;
 607                 }
 608
 609                 if (in_buf.buffer.length + offset + out_size > buf_len) {
 610                         dev_dbg(dev, "%s output overrun cmd: %s field: %d\n",
 611                                         dimm_name, cmd_name, i);
 612                         rc = -ENXIO;
 613                         goto out;
 614                 }
 615                 memcpy(buf + in_buf.buffer.length + offset,
 616                                 out_obj->buffer.pointer + offset, out_size);
 617                 offset += out_size;
 618         }
 619
 620         /*
 621          * Set fw_status for all the commands with a known format to be
 622          * later interpreted by xlat_status().
 623          */
 624         if (i >= 1 && ((!nvdimm && cmd >= ND_CMD_ARS_CAP
 625                                         && cmd <= ND_CMD_CLEAR_ERROR)
 626                                 || (nvdimm && cmd >= ND_CMD_SMART
 627                                         && cmd <= ND_CMD_VENDOR)))
 628                 fw_status = *(u32 *) out_obj->buffer.pointer;
 629
 630         if (offset + in_buf.buffer.length < buf_len) {
 631                 if (i >= 1) {
 632                         /*
 633                          * status valid, return the number of bytes left
 634                          * unfilled in the output buffer
 635                          */
 636                         rc = buf_len - offset - in_buf.buffer.length;
 637                         if (cmd_rc)
 638                                 *cmd_rc = xlat_status(nvdimm, buf, cmd,
 639                                                 fw_status);
 640                 } else {
 641                         dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
 642                                         __func__, dimm_name, cmd_name, buf_len,
 643                                         offset);
 644                         rc = -ENXIO;
 645                 }
 646         } else {
 647                 rc = 0;
 648                 if (cmd_rc)
 649                         *cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status);
 650         }
 651
 652  out:
 653         ACPI_FREE(out_obj);
 654
 655         return rc;
 656 }
 657 EXPORT_SYMBOL_GPL(acpi_nfit_ctl);
 658
 659 static const char *spa_type_name(u16 type)
 660 {
 661         static const char *to_name[] = {
 662                 [NFIT_SPA_VOLATILE] = "volatile",
 663                 [NFIT_SPA_PM] = "pmem",
 664                 [NFIT_SPA_DCR] = "dimm-control-region",
 665                 [NFIT_SPA_BDW] = "block-data-window",
 666                 [NFIT_SPA_VDISK] = "volatile-disk",
 667                 [NFIT_SPA_VCD] = "volatile-cd",
 668                 [NFIT_SPA_PDISK] = "persistent-disk",
 669                 [NFIT_SPA_PCD] = "persistent-cd",
 670
 671         };
 672
 673         if (type > NFIT_SPA_PCD)
 674                 return "unknown";
 675
 676         return to_name[type];
 677 }
 678
 679 int nfit_spa_type(struct acpi_nfit_system_address *spa)
 680 {
 681         int i;
 682
 683         for (i = 0; i < NFIT_UUID_MAX; i++)
 684                 if (guid_equal(to_nfit_uuid(i), (guid_t *)&spa->range_guid))
 685                         return i;
 686         return -1;
 687 }
 688
 689 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
 690                 struct nfit_table_prev *prev,
 691                 struct acpi_nfit_system_address *spa)
 692 {
 693         struct device *dev = acpi_desc->dev;
 694         struct nfit_spa *nfit_spa;
 695
 696         if (spa->header.length != sizeof(*spa))
 697                 return false;
 698
 699         list_for_each_entry(nfit_spa, &prev->spas, list) {
 700                 if (memcmp(nfit_spa->spa, spa, sizeof(*spa)) == 0) {
 701                         list_move_tail(&nfit_spa->list, &acpi_desc->spas);
 702                         return true;
 703                 }
 704         }
 705
 706         nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof(*spa),
 707                         GFP_KERNEL);
 708         if (!nfit_spa)
 709                 return false;
 710         INIT_LIST_HEAD(&nfit_spa->list);
 711         memcpy(nfit_spa->spa, spa, sizeof(*spa));
 712         list_add_tail(&nfit_spa->list, &acpi_desc->spas);
 713         dev_dbg(dev, "spa index: %d type: %s\n",
 714                         spa->range_index,
 715                         spa_type_name(nfit_spa_type(spa)));
 716         return true;
 717 }
 718
 719 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
 720                 struct nfit_table_prev *prev,
 721                 struct acpi_nfit_memory_map *memdev)
 722 {
 723         struct device *dev = acpi_desc->dev;
 724         struct nfit_memdev *nfit_memdev;
 725
 726         if (memdev->header.length != sizeof(*memdev))
 727                 return false;
 728
 729         list_for_each_entry(nfit_memdev, &prev->memdevs, list)
 730                 if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
 731                         list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 732                         return true;
 733                 }
 734
 735         nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
 736                         GFP_KERNEL);
 737         if (!nfit_memdev)
 738                 return false;
 739         INIT_LIST_HEAD(&nfit_memdev->list);
 740         memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
 741         list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 742         dev_dbg(dev, "memdev handle: %#x spa: %d dcr: %d flags: %#x\n",
 743                         memdev->device_handle, memdev->range_index,
 744                         memdev->region_index, memdev->flags);
 745         return true;
 746 }
 747
 748 int nfit_get_smbios_id(u32 device_handle, u16 *flags)
 749 {
 750         struct acpi_nfit_memory_map *memdev;
 751         struct acpi_nfit_desc *acpi_desc;
 752         struct nfit_mem *nfit_mem;
 753         u16 physical_id;
 754
 755         mutex_lock(&acpi_desc_lock);
 756         list_for_each_entry(acpi_desc, &acpi_descs, list) {
 757                 mutex_lock(&acpi_desc->init_mutex);
 758                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
 759                         memdev = __to_nfit_memdev(nfit_mem);
 760                         if (memdev->device_handle == device_handle) {
 761                                 *flags = memdev->flags;
 762                                 physical_id = memdev->physical_id;
 763                                 mutex_unlock(&acpi_desc->init_mutex);
 764                                 mutex_unlock(&acpi_desc_lock);
 765                                 return physical_id;
 766                         }
 767                 }
 768                 mutex_unlock(&acpi_desc->init_mutex);
 769         }
 770         mutex_unlock(&acpi_desc_lock);
 771
 772         return -ENODEV;
 773 }
 774 EXPORT_SYMBOL_GPL(nfit_get_smbios_id);
 775
 776 /*
 777  * An implementation may provide a truncated control region if no block windows
 778  * are defined.
 779  */
 780 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
 781 {
 782         if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
 783                                 window_size))
 784                 return 0;
 785         if (dcr->windows)
 786                 return sizeof(*dcr);
 787         return offsetof(struct acpi_nfit_control_region, window_size);
 788 }
 789
 790 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
 791                 struct nfit_table_prev *prev,
 792                 struct acpi_nfit_control_region *dcr)
 793 {
 794         struct device *dev = acpi_desc->dev;
 795         struct nfit_dcr *nfit_dcr;
 796
 797         if (!sizeof_dcr(dcr))
 798                 return false;
 799
 800         list_for_each_entry(nfit_dcr, &prev->dcrs, list)
 801                 if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
 802                         list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 803                         return true;
 804                 }
 805
 806         nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
 807                         GFP_KERNEL);
 808         if (!nfit_dcr)
 809                 return false;
 810         INIT_LIST_HEAD(&nfit_dcr->list);
 811         memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
 812         list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 813         dev_dbg(dev, "dcr index: %d windows: %d\n",
 814                         dcr->region_index, dcr->windows);
 815         return true;
 816 }
 817
 818 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
 819                 struct nfit_table_prev *prev,
 820                 struct acpi_nfit_data_region *bdw)
 821 {
 822         struct device *dev = acpi_desc->dev;
 823         struct nfit_bdw *nfit_bdw;
 824
 825         if (bdw->header.length != sizeof(*bdw))
 826                 return false;
 827         list_for_each_entry(nfit_bdw, &prev->bdws, list)
 828                 if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
 829                         list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
 830                         return true;
 831                 }
 832
 833         nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
 834                         GFP_KERNEL);
 835         if (!nfit_bdw)
 836                 return false;
 837         INIT_LIST_HEAD(&nfit_bdw->list);
 838         memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
 839         list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
 840         dev_dbg(dev, "bdw dcr: %d windows: %d\n",
 841                         bdw->region_index, bdw->windows);
 842         return true;
 843 }
 844
 845 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
 846 {
 847         if (idt->header.length < sizeof(*idt))
 848                 return 0;
 849         return sizeof(*idt) + sizeof(u32) * (idt->line_count - 1);
 850 }
 851
 852 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
 853                 struct nfit_table_prev *prev,
 854                 struct acpi_nfit_interleave *idt)
 855 {
 856         struct device *dev = acpi_desc->dev;
 857         struct nfit_idt *nfit_idt;
 858
 859         if (!sizeof_idt(idt))
 860                 return false;
 861
 862         list_for_each_entry(nfit_idt, &prev->idts, list) {
 863                 if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
 864                         continue;
 865
 866                 if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
 867                         list_move_tail(&nfit_idt->list, &acpi_desc->idts);
 868                         return true;
 869                 }
 870         }
 871
 872         nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
 873                         GFP_KERNEL);
 874         if (!nfit_idt)
 875                 return false;
 876         INIT_LIST_HEAD(&nfit_idt->list);
 877         memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
 878         list_add_tail(&nfit_idt->list, &acpi_desc->idts);
 879         dev_dbg(dev, "idt index: %d num_lines: %d\n",
 880                         idt->interleave_index, idt->line_count);
 881         return true;
 882 }
 883
 884 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
 885 {
 886         if (flush->header.length < sizeof(*flush))
 887                 return 0;
 888         return sizeof(*flush) + sizeof(u64) * (flush->hint_count - 1);
 889 }
 890
 891 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
 892                 struct nfit_table_prev *prev,
 893                 struct acpi_nfit_flush_address *flush)
 894 {
 895         struct device *dev = acpi_desc->dev;
 896         struct nfit_flush *nfit_flush;
 897
 898         if (!sizeof_flush(flush))
 899                 return false;
 900
 901         list_for_each_entry(nfit_flush, &prev->flushes, list) {
 902                 if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
 903                         continue;
 904
 905                 if (memcmp(nfit_flush->flush, flush,
 906                                         sizeof_flush(flush)) == 0) {
 907                         list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
 908                         return true;
 909                 }
 910         }
 911
 912         nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
 913                         + sizeof_flush(flush), GFP_KERNEL);
 914         if (!nfit_flush)
 915                 return false;
 916         INIT_LIST_HEAD(&nfit_flush->list);
 917         memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
 918         list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
 919         dev_dbg(dev, "nfit_flush handle: %d hint_count: %d\n",
 920                         flush->device_handle, flush->hint_count);
 921         return true;
 922 }
 923
 924 static bool add_platform_cap(struct acpi_nfit_desc *acpi_desc,
 925                 struct acpi_nfit_capabilities *pcap)
 926 {
 927         struct device *dev = acpi_desc->dev;
 928         u32 mask;
 929
 930         mask = (1 << (pcap->highest_capability + 1)) - 1;
 931         acpi_desc->platform_cap = pcap->capabilities & mask;
 932         dev_dbg(dev, "cap: %#x\n", acpi_desc->platform_cap);
 933         return true;
 934 }
 935
 936 static void *add_table(struct acpi_nfit_desc *acpi_desc,
 937                 struct nfit_table_prev *prev, void *table, const void *end)
 938 {
 939         struct device *dev = acpi_desc->dev;
 940         struct acpi_nfit_header *hdr;
 941         void *err = ERR_PTR(-ENOMEM);
 942
 943         if (table >= end)
 944                 return NULL;
 945
 946         hdr = table;
 947         if (!hdr->length) {
 948                 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
 949                         hdr->type);
 950                 return NULL;
 951         }
 952
 953         switch (hdr->type) {
 954         case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
 955                 if (!add_spa(acpi_desc, prev, table))
 956                         return err;
 957                 break;
 958         case ACPI_NFIT_TYPE_MEMORY_MAP:
 959                 if (!add_memdev(acpi_desc, prev, table))
 960                         return err;
 961                 break;
 962         case ACPI_NFIT_TYPE_CONTROL_REGION:
 963                 if (!add_dcr(acpi_desc, prev, table))
 964                         return err;
 965                 break;
 966         case ACPI_NFIT_TYPE_DATA_REGION:
 967                 if (!add_bdw(acpi_desc, prev, table))
 968                         return err;
 969                 break;
 970         case ACPI_NFIT_TYPE_INTERLEAVE:
 971                 if (!add_idt(acpi_desc, prev, table))
 972                         return err;
 973                 break;
 974         case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
 975                 if (!add_flush(acpi_desc, prev, table))
 976                         return err;
 977                 break;
 978         case ACPI_NFIT_TYPE_SMBIOS:
 979                 dev_dbg(dev, "smbios\n");
 980                 break;
 981         case ACPI_NFIT_TYPE_CAPABILITIES:
 982                 if (!add_platform_cap(acpi_desc, table))
 983                         return err;
 984                 break;
 985         default:
 986                 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
 987                 break;
 988         }
 989
 990         return table + hdr->length;
 991 }
 992
 993 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
 994                 struct nfit_mem *nfit_mem)
 995 {
 996         u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
 997         u16 dcr = nfit_mem->dcr->region_index;
 998         struct nfit_spa *nfit_spa;
 999
1000         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
1001                 u16 range_index = nfit_spa->spa->range_index;
1002                 int type = nfit_spa_type(nfit_spa->spa);
1003                 struct nfit_memdev *nfit_memdev;
1004
1005                 if (type != NFIT_SPA_BDW)
1006                         continue;
1007
1008                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1009                         if (nfit_memdev->memdev->range_index != range_index)
1010                                 continue;
1011                         if (nfit_memdev->memdev->device_handle != device_handle)
1012                                 continue;
1013                         if (nfit_memdev->memdev->region_index != dcr)
1014                                 continue;
1015
1016                         nfit_mem->spa_bdw = nfit_spa->spa;
1017                         return;
1018                 }
1019         }
1020
1021         dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
1022                         nfit_mem->spa_dcr->range_index);
1023         nfit_mem->bdw = NULL;
1024 }
1025
1026 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
1027                 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
1028 {
1029         u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
1030         struct nfit_memdev *nfit_memdev;
1031         struct nfit_bdw *nfit_bdw;
1032         struct nfit_idt *nfit_idt;
1033         u16 idt_idx, range_index;
1034
1035         list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
1036                 if (nfit_bdw->bdw->region_index != dcr)
1037                         continue;
1038                 nfit_mem->bdw = nfit_bdw->bdw;
1039                 break;
1040         }
1041
1042         if (!nfit_mem->bdw)
1043                 return;
1044
1045         nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
1046
1047         if (!nfit_mem->spa_bdw)
1048                 return;
1049
1050         range_index = nfit_mem->spa_bdw->range_index;
1051         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1052                 if (nfit_memdev->memdev->range_index != range_index ||
1053                                 nfit_memdev->memdev->region_index != dcr)
1054                         continue;
1055                 nfit_mem->memdev_bdw = nfit_memdev->memdev;
1056                 idt_idx = nfit_memdev->memdev->interleave_index;
1057                 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1058                         if (nfit_idt->idt->interleave_index != idt_idx)
1059                                 continue;
1060                         nfit_mem->idt_bdw = nfit_idt->idt;
1061                         break;
1062                 }
1063                 break;
1064         }
1065 }
1066
1067 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
1068                 struct acpi_nfit_system_address *spa)
1069 {
1070         struct nfit_mem *nfit_mem, *found;
1071         struct nfit_memdev *nfit_memdev;
1072         int type = spa ? nfit_spa_type(spa) : 0;
1073
1074         switch (type) {
1075         case NFIT_SPA_DCR:
1076         case NFIT_SPA_PM:
1077                 break;
1078         default:
1079                 if (spa)
1080                         return 0;
1081         }
1082
1083         /*
1084          * This loop runs in two modes, when a dimm is mapped the loop
1085          * adds memdev associations to an existing dimm, or creates a
1086          * dimm. In the unmapped dimm case this loop sweeps for memdev
1087          * instances with an invalid / zero range_index and adds those
1088          * dimms without spa associations.
1089          */
1090         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1091                 struct nfit_flush *nfit_flush;
1092                 struct nfit_dcr *nfit_dcr;
1093                 u32 device_handle;
1094                 u16 dcr;
1095
1096                 if (spa && nfit_memdev->memdev->range_index != spa->range_index)
1097                         continue;
1098                 if (!spa && nfit_memdev->memdev->range_index)
1099                         continue;
1100                 found = NULL;
1101                 dcr = nfit_memdev->memdev->region_index;
1102                 device_handle = nfit_memdev->memdev->device_handle;
1103                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1104                         if (__to_nfit_memdev(nfit_mem)->device_handle
1105                                         == device_handle) {
1106                                 found = nfit_mem;
1107                                 break;
1108                         }
1109
1110                 if (found)
1111                         nfit_mem = found;
1112                 else {
1113                         nfit_mem = devm_kzalloc(acpi_desc->dev,
1114                                         sizeof(*nfit_mem), GFP_KERNEL);
1115                         if (!nfit_mem)
1116                                 return -ENOMEM;
1117                         INIT_LIST_HEAD(&nfit_mem->list);
1118                         nfit_mem->acpi_desc = acpi_desc;
1119                         list_add(&nfit_mem->list, &acpi_desc->dimms);
1120                 }
1121
1122                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1123                         if (nfit_dcr->dcr->region_index != dcr)
1124                                 continue;
1125                         /*
1126                          * Record the control region for the dimm.  For
1127                          * the ACPI 6.1 case, where there are separate
1128                          * control regions for the pmem vs blk
1129                          * interfaces, be sure to record the extended
1130                          * blk details.
1131                          */
1132                         if (!nfit_mem->dcr)
1133                                 nfit_mem->dcr = nfit_dcr->dcr;
1134                         else if (nfit_mem->dcr->windows == 0
1135                                         && nfit_dcr->dcr->windows)
1136                                 nfit_mem->dcr = nfit_dcr->dcr;
1137                         break;
1138                 }
1139
1140                 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
1141                         struct acpi_nfit_flush_address *flush;
1142                         u16 i;
1143
1144                         if (nfit_flush->flush->device_handle != device_handle)
1145                                 continue;
1146                         nfit_mem->nfit_flush = nfit_flush;
1147                         flush = nfit_flush->flush;
1148                         nfit_mem->flush_wpq = devm_kcalloc(acpi_desc->dev,
1149                                         flush->hint_count,
1150                                         sizeof(struct resource),
1151                                         GFP_KERNEL);
1152                         if (!nfit_mem->flush_wpq)
1153                                 return -ENOMEM;
1154                         for (i = 0; i < flush->hint_count; i++) {
1155                                 struct resource *res = &nfit_mem->flush_wpq[i];
1156
1157                                 res->start = flush->hint_address[i];
1158                                 res->end = res->start + 8 - 1;
1159                         }
1160                         break;
1161                 }
1162
1163                 if (dcr && !nfit_mem->dcr) {
1164                         dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
1165                                         spa->range_index, dcr);
1166                         return -ENODEV;
1167                 }
1168
1169                 if (type == NFIT_SPA_DCR) {
1170                         struct nfit_idt *nfit_idt;
1171                         u16 idt_idx;
1172
1173                         /* multiple dimms may share a SPA when interleaved */
1174                         nfit_mem->spa_dcr = spa;
1175                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
1176                         idt_idx = nfit_memdev->memdev->interleave_index;
1177                         list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1178                                 if (nfit_idt->idt->interleave_index != idt_idx)
1179                                         continue;
1180                                 nfit_mem->idt_dcr = nfit_idt->idt;
1181                                 break;
1182                         }
1183                         nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
1184                 } else if (type == NFIT_SPA_PM) {
1185                         /*
1186                          * A single dimm may belong to multiple SPA-PM
1187                          * ranges, record at least one in addition to
1188                          * any SPA-DCR range.
1189                          */
1190                         nfit_mem->memdev_pmem = nfit_memdev->memdev;
1191                 } else
1192                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
1193         }
1194
1195         return 0;
1196 }
1197
1198 static int nfit_mem_cmp(void *priv, const struct list_head *_a,
1199                 const struct list_head *_b)
1200 {
1201         struct nfit_mem *a = container_of(_a, typeof(*a), list);
1202         struct nfit_mem *b = container_of(_b, typeof(*b), list);
1203         u32 handleA, handleB;
1204
1205         handleA = __to_nfit_memdev(a)->device_handle;
1206         handleB = __to_nfit_memdev(b)->device_handle;
1207         if (handleA < handleB)
1208                 return -1;
1209         else if (handleA > handleB)
1210                 return 1;
1211         return 0;
1212 }
1213
1214 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
1215 {
1216         struct nfit_spa *nfit_spa;
1217         int rc;
1218
1219
1220         /*
1221          * For each SPA-DCR or SPA-PMEM address range find its
1222          * corresponding MEMDEV(s).  From each MEMDEV find the
1223          * corresponding DCR.  Then, if we're operating on a SPA-DCR,
1224          * try to find a SPA-BDW and a corresponding BDW that references
1225          * the DCR.  Throw it all into an nfit_mem object.  Note, that
1226          * BDWs are optional.
1227          */
1228         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
1229                 rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
1230                 if (rc)
1231                         return rc;
1232         }
1233
1234         /*
1235          * If a DIMM has failed to be mapped into SPA there will be no
1236          * SPA entries above. Find and register all the unmapped DIMMs
1237          * for reporting and recovery purposes.
1238          */
1239         rc = __nfit_mem_init(acpi_desc, NULL);
1240         if (rc)
1241                 return rc;
1242
1243         list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
1244
1245         return 0;
1246 }
1247
1248 static ssize_t bus_dsm_mask_show(struct device *dev,
1249                 struct device_attribute *attr, char *buf)
1250 {
1251         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1252         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1253         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1254
1255         return sprintf(buf, "%#lx\n", acpi_desc->bus_dsm_mask);
1256 }
1257 static struct device_attribute dev_attr_bus_dsm_mask =
1258                 __ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
1259
1260 static ssize_t revision_show(struct device *dev,
1261                 struct device_attribute *attr, char *buf)
1262 {
1263         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1264         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1265         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1266
1267         return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
1268 }
1269 static DEVICE_ATTR_RO(revision);
1270
1271 static ssize_t hw_error_scrub_show(struct device *dev,
1272                 struct device_attribute *attr, char *buf)
1273 {
1274         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1275         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1276         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1277
1278         return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
1279 }
1280
1281 /*
1282  * The 'hw_error_scrub' attribute can have the following values written to it:
1283  * '0': Switch to the default mode where an exception will only insert
1284  *      the address of the memory error into the poison and badblocks lists.
1285  * '1': Enable a full scrub to happen if an exception for a memory error is
1286  *      received.
1287  */
1288 static ssize_t hw_error_scrub_store(struct device *dev,
1289                 struct device_attribute *attr, const char *buf, size_t size)
1290 {
1291         struct nvdimm_bus_descriptor *nd_desc;
1292         ssize_t rc;
1293         long val;
1294
1295         rc = kstrtol(buf, 0, &val);
1296         if (rc)
1297                 return rc;
1298
1299         nfit_device_lock(dev);
1300         nd_desc = dev_get_drvdata(dev);
1301         if (nd_desc) {
1302                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1303
1304                 switch (val) {
1305                 case HW_ERROR_SCRUB_ON:
1306                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
1307                         break;
1308                 case HW_ERROR_SCRUB_OFF:
1309                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
1310                         break;
1311                 default:
1312                         rc = -EINVAL;
1313                         break;
1314                 }
1315         }
1316         nfit_device_unlock(dev);
1317         if (rc)
1318                 return rc;
1319         return size;
1320 }
1321 static DEVICE_ATTR_RW(hw_error_scrub);
1322
1323 /*
1324  * This shows the number of full Address Range Scrubs that have been
1325  * completed since driver load time. Userspace can wait on this using
1326  * select/poll etc. A '+' at the end indicates an ARS is in progress
1327  */
1328 static ssize_t scrub_show(struct device *dev,
1329                 struct device_attribute *attr, char *buf)
1330 {
1331         struct nvdimm_bus_descriptor *nd_desc;
1332         struct acpi_nfit_desc *acpi_desc;
1333         ssize_t rc = -ENXIO;
1334         bool busy;
1335
1336         nfit_device_lock(dev);
1337         nd_desc = dev_get_drvdata(dev);
1338         if (!nd_desc) {
1339                 nfit_device_unlock(dev);
1340                 return rc;
1341         }
1342         acpi_desc = to_acpi_desc(nd_desc);
1343
1344         mutex_lock(&acpi_desc->init_mutex);
1345         busy = test_bit(ARS_BUSY, &acpi_desc->scrub_flags)
1346                 && !test_bit(ARS_CANCEL, &acpi_desc->scrub_flags);
1347         rc = sprintf(buf, "%d%s", acpi_desc->scrub_count, busy ? "+\n" : "\n");
1348         /* Allow an admin to poll the busy state at a higher rate */
1349         if (busy && capable(CAP_SYS_RAWIO) && !test_and_set_bit(ARS_POLL,
1350                                 &acpi_desc->scrub_flags)) {
1351                 acpi_desc->scrub_tmo = 1;
1352                 mod_delayed_work(nfit_wq, &acpi_desc->dwork, HZ);
1353         }
1354
1355         mutex_unlock(&acpi_desc->init_mutex);
1356         nfit_device_unlock(dev);
1357         return rc;
1358 }
1359
1360 static ssize_t scrub_store(struct device *dev,
1361                 struct device_attribute *attr, const char *buf, size_t size)
1362 {
1363         struct nvdimm_bus_descriptor *nd_desc;
1364         ssize_t rc;
1365         long val;
1366
1367         rc = kstrtol(buf, 0, &val);
1368         if (rc)
1369                 return rc;
1370         if (val != 1)
1371                 return -EINVAL;
1372
1373         nfit_device_lock(dev);
1374         nd_desc = dev_get_drvdata(dev);
1375         if (nd_desc) {
1376                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1377
1378                 rc = acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
1379         }
1380         nfit_device_unlock(dev);
1381         if (rc)
1382                 return rc;
1383         return size;
1384 }
1385 static DEVICE_ATTR_RW(scrub);
1386
1387 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1388 {
1389         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1390         const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1391                 | 1 << ND_CMD_ARS_STATUS;
1392
1393         return (nd_desc->cmd_mask & mask) == mask;
1394 }
1395
1396 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1397 {
1398         struct device *dev = kobj_to_dev(kobj);
1399         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1400
1401         if (a == &dev_attr_scrub.attr)
1402                 return ars_supported(nvdimm_bus) ? a->mode : 0;
1403
1404         if (a == &dev_attr_firmware_activate_noidle.attr)
1405                 return intel_fwa_supported(nvdimm_bus) ? a->mode : 0;
1406
1407         return a->mode;
1408 }
1409
1410 static struct attribute *acpi_nfit_attributes[] = {
1411         &dev_attr_revision.attr,
1412         &dev_attr_scrub.attr,
1413         &dev_attr_hw_error_scrub.attr,
1414         &dev_attr_bus_dsm_mask.attr,
1415         &dev_attr_firmware_activate_noidle.attr,
1416         NULL,
1417 };
1418
1419 static const struct attribute_group acpi_nfit_attribute_group = {
1420         .name = "nfit",
1421         .attrs = acpi_nfit_attributes,
1422         .is_visible = nfit_visible,
1423 };
1424
1425 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1426         &acpi_nfit_attribute_group,
1427         NULL,
1428 };
1429
1430 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1431 {
1432         struct nvdimm *nvdimm = to_nvdimm(dev);
1433         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1434
1435         return __to_nfit_memdev(nfit_mem);
1436 }
1437
1438 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1439 {
1440         struct nvdimm *nvdimm = to_nvdimm(dev);
1441         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1442
1443         return nfit_mem->dcr;
1444 }
1445
1446 static ssize_t handle_show(struct device *dev,
1447                 struct device_attribute *attr, char *buf)
1448 {
1449         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1450
1451         return sprintf(buf, "%#x\n", memdev->device_handle);
1452 }
1453 static DEVICE_ATTR_RO(handle);
1454
1455 static ssize_t phys_id_show(struct device *dev,
1456                 struct device_attribute *attr, char *buf)
1457 {
1458         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1459
1460         return sprintf(buf, "%#x\n", memdev->physical_id);
1461 }
1462 static DEVICE_ATTR_RO(phys_id);
1463
1464 static ssize_t vendor_show(struct device *dev,
1465                 struct device_attribute *attr, char *buf)
1466 {
1467         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1468
1469         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1470 }
1471 static DEVICE_ATTR_RO(vendor);
1472
1473 static ssize_t rev_id_show(struct device *dev,
1474                 struct device_attribute *attr, char *buf)
1475 {
1476         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1477
1478         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1479 }
1480 static DEVICE_ATTR_RO(rev_id);
1481
1482 static ssize_t device_show(struct device *dev,
1483                 struct device_attribute *attr, char *buf)
1484 {
1485         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1486
1487         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1488 }
1489 static DEVICE_ATTR_RO(device);
1490
1491 static ssize_t subsystem_vendor_show(struct device *dev,
1492                 struct device_attribute *attr, char *buf)
1493 {
1494         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1495
1496         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1497 }
1498 static DEVICE_ATTR_RO(subsystem_vendor);
1499
1500 static ssize_t subsystem_rev_id_show(struct device *dev,
1501                 struct device_attribute *attr, char *buf)
1502 {
1503         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1504
1505         return sprintf(buf, "0x%04x\n",
1506                         be16_to_cpu(dcr->subsystem_revision_id));
1507 }
1508 static DEVICE_ATTR_RO(subsystem_rev_id);
1509
1510 static ssize_t subsystem_device_show(struct device *dev,
1511                 struct device_attribute *attr, char *buf)
1512 {
1513         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1514
1515         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1516 }
1517 static DEVICE_ATTR_RO(subsystem_device);
1518
1519 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1520 {
1521         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1522         int formats = 0;
1523
1524         if (nfit_mem->memdev_pmem)
1525                 formats++;
1526         if (nfit_mem->memdev_bdw)
1527                 formats++;
1528         return formats;
1529 }
1530
1531 static ssize_t format_show(struct device *dev,
1532                 struct device_attribute *attr, char *buf)
1533 {
1534         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1535
1536         return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1537 }
1538 static DEVICE_ATTR_RO(format);
1539
1540 static ssize_t format1_show(struct device *dev,
1541                 struct device_attribute *attr, char *buf)
1542 {
1543         u32 handle;
1544         ssize_t rc = -ENXIO;
1545         struct nfit_mem *nfit_mem;
1546         struct nfit_memdev *nfit_memdev;
1547         struct acpi_nfit_desc *acpi_desc;
1548         struct nvdimm *nvdimm = to_nvdimm(dev);
1549         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1550
1551         nfit_mem = nvdimm_provider_data(nvdimm);
1552         acpi_desc = nfit_mem->acpi_desc;
1553         handle = to_nfit_memdev(dev)->device_handle;
1554
1555         /* assumes DIMMs have at most 2 published interface codes */
1556         mutex_lock(&acpi_desc->init_mutex);
1557         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1558                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1559                 struct nfit_dcr *nfit_dcr;
1560
1561                 if (memdev->device_handle != handle)
1562                         continue;
1563
1564                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1565                         if (nfit_dcr->dcr->region_index != memdev->region_index)
1566                                 continue;
1567                         if (nfit_dcr->dcr->code == dcr->code)
1568                                 continue;
1569                         rc = sprintf(buf, "0x%04x\n",
1570                                         le16_to_cpu(nfit_dcr->dcr->code));
1571                         break;
1572                 }
1573                 if (rc != -ENXIO)
1574                         break;
1575         }
1576         mutex_unlock(&acpi_desc->init_mutex);
1577         return rc;
1578 }
1579 static DEVICE_ATTR_RO(format1);
1580
1581 static ssize_t formats_show(struct device *dev,
1582                 struct device_attribute *attr, char *buf)
1583 {
1584         struct nvdimm *nvdimm = to_nvdimm(dev);
1585
1586         return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1587 }
1588 static DEVICE_ATTR_RO(formats);
1589
1590 static ssize_t serial_show(struct device *dev,
1591                 struct device_attribute *attr, char *buf)
1592 {
1593         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1594
1595         return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1596 }
1597 static DEVICE_ATTR_RO(serial);
1598
1599 static ssize_t family_show(struct device *dev,
1600                 struct device_attribute *attr, char *buf)
1601 {
1602         struct nvdimm *nvdimm = to_nvdimm(dev);
1603         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1604
1605         if (nfit_mem->family < 0)
1606                 return -ENXIO;
1607         return sprintf(buf, "%d\n", nfit_mem->family);
1608 }
1609 static DEVICE_ATTR_RO(family);
1610
1611 static ssize_t dsm_mask_show(struct device *dev,
1612                 struct device_attribute *attr, char *buf)
1613 {
1614         struct nvdimm *nvdimm = to_nvdimm(dev);
1615         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1616
1617         if (nfit_mem->family < 0)
1618                 return -ENXIO;
1619         return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1620 }
1621 static DEVICE_ATTR_RO(dsm_mask);
1622
1623 static ssize_t flags_show(struct device *dev,
1624                 struct device_attribute *attr, char *buf)
1625 {
1626         struct nvdimm *nvdimm = to_nvdimm(dev);
1627         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1628         u16 flags = __to_nfit_memdev(nfit_mem)->flags;
1629
1630         if (test_bit(NFIT_MEM_DIRTY, &nfit_mem->flags))
1631                 flags |= ACPI_NFIT_MEM_FLUSH_FAILED;
1632
1633         return sprintf(buf, "%s%s%s%s%s%s%s\n",
1634                 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1635                 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1636                 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1637                 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1638                 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1639                 flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1640                 flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1641 }
1642 static DEVICE_ATTR_RO(flags);
1643
1644 static ssize_t id_show(struct device *dev,
1645                 struct device_attribute *attr, char *buf)
1646 {
1647         struct nvdimm *nvdimm = to_nvdimm(dev);
1648         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1649
1650         return sprintf(buf, "%s\n", nfit_mem->id);
1651 }
1652 static DEVICE_ATTR_RO(id);
1653
1654 static ssize_t dirty_shutdown_show(struct device *dev,
1655                 struct device_attribute *attr, char *buf)
1656 {
1657         struct nvdimm *nvdimm = to_nvdimm(dev);
1658         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1659
1660         return sprintf(buf, "%d\n", nfit_mem->dirty_shutdown);
1661 }
1662 static DEVICE_ATTR_RO(dirty_shutdown);
1663
1664 static struct attribute *acpi_nfit_dimm_attributes[] = {
1665         &dev_attr_handle.attr,
1666         &dev_attr_phys_id.attr,
1667         &dev_attr_vendor.attr,
1668         &dev_attr_device.attr,
1669         &dev_attr_rev_id.attr,
1670         &dev_attr_subsystem_vendor.attr,
1671         &dev_attr_subsystem_device.attr,
1672         &dev_attr_subsystem_rev_id.attr,
1673         &dev_attr_format.attr,
1674         &dev_attr_formats.attr,
1675         &dev_attr_format1.attr,
1676         &dev_attr_serial.attr,
1677         &dev_attr_flags.attr,
1678         &dev_attr_id.attr,
1679         &dev_attr_family.attr,
1680         &dev_attr_dsm_mask.attr,
1681         &dev_attr_dirty_shutdown.attr,
1682         NULL,
1683 };
1684
1685 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1686                 struct attribute *a, int n)
1687 {
1688         struct device *dev = kobj_to_dev(kobj);
1689         struct nvdimm *nvdimm = to_nvdimm(dev);
1690         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1691
1692         if (!to_nfit_dcr(dev)) {
1693                 /* Without a dcr only the memdev attributes can be surfaced */
1694                 if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1695                                 || a == &dev_attr_flags.attr
1696                                 || a == &dev_attr_family.attr
1697                                 || a == &dev_attr_dsm_mask.attr)
1698                         return a->mode;
1699                 return 0;
1700         }
1701
1702         if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1703                 return 0;
1704
1705         if (!test_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags)
1706                         && a == &dev_attr_dirty_shutdown.attr)
1707                 return 0;
1708
1709         return a->mode;
1710 }
1711
1712 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1713         .name = "nfit",
1714         .attrs = acpi_nfit_dimm_attributes,
1715         .is_visible = acpi_nfit_dimm_attr_visible,
1716 };
1717
1718 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1719         &acpi_nfit_dimm_attribute_group,
1720         NULL,
1721 };
1722
1723 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1724                 u32 device_handle)
1725 {
1726         struct nfit_mem *nfit_mem;
1727
1728         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1729                 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1730                         return nfit_mem->nvdimm;
1731
1732         return NULL;
1733 }
1734
1735 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1736 {
1737         struct nfit_mem *nfit_mem;
1738         struct acpi_nfit_desc *acpi_desc;
1739
1740         dev_dbg(dev->parent, "%s: event: %d\n", dev_name(dev),
1741                         event);
1742
1743         if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1744                 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1745                                 event);
1746                 return;
1747         }
1748
1749         acpi_desc = dev_get_drvdata(dev->parent);
1750         if (!acpi_desc)
1751                 return;
1752
1753         /*
1754          * If we successfully retrieved acpi_desc, then we know nfit_mem data
1755          * is still valid.
1756          */
1757         nfit_mem = dev_get_drvdata(dev);
1758         if (nfit_mem && nfit_mem->flags_attr)
1759                 sysfs_notify_dirent(nfit_mem->flags_attr);
1760 }
1761 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1762
1763 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1764 {
1765         struct acpi_device *adev = data;
1766         struct device *dev = &adev->dev;
1767
1768         nfit_device_lock(dev->parent);
1769         __acpi_nvdimm_notify(dev, event);
1770         nfit_device_unlock(dev->parent);
1771 }
1772
1773 static bool acpi_nvdimm_has_method(struct acpi_device *adev, char *method)
1774 {
1775         acpi_handle handle;
1776         acpi_status status;
1777
1778         status = acpi_get_handle(adev->handle, method, &handle);
1779
1780         if (ACPI_SUCCESS(status))
1781                 return true;
1782         return false;
1783 }
1784
1785 __weak void nfit_intel_shutdown_status(struct nfit_mem *nfit_mem)
1786 {
1787         struct device *dev = &nfit_mem->adev->dev;
1788         struct nd_intel_smart smart = { 0 };
1789         union acpi_object in_buf = {
1790                 .buffer.type = ACPI_TYPE_BUFFER,
1791                 .buffer.length = 0,
1792         };
1793         union acpi_object in_obj = {
1794                 .package.type = ACPI_TYPE_PACKAGE,
1795                 .package.count = 1,
1796                 .package.elements = &in_buf,
1797         };
1798         const u8 func = ND_INTEL_SMART;
1799         const guid_t *guid = to_nfit_uuid(nfit_mem->family);
1800         u8 revid = nfit_dsm_revid(nfit_mem->family, func);
1801         struct acpi_device *adev = nfit_mem->adev;
1802         acpi_handle handle = adev->handle;
1803         union acpi_object *out_obj;
1804
1805         if ((nfit_mem->dsm_mask & (1 << func)) == 0)
1806                 return;
1807
1808         out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
1809         if (!out_obj || out_obj->type != ACPI_TYPE_BUFFER
1810                         || out_obj->buffer.length < sizeof(smart)) {
1811                 dev_dbg(dev->parent, "%s: failed to retrieve initial health\n",
1812                                 dev_name(dev));
1813                 ACPI_FREE(out_obj);
1814                 return;
1815         }
1816         memcpy(&smart, out_obj->buffer.pointer, sizeof(smart));
1817         ACPI_FREE(out_obj);
1818
1819         if (smart.flags & ND_INTEL_SMART_SHUTDOWN_VALID) {
1820                 if (smart.shutdown_state)
1821                         set_bit(NFIT_MEM_DIRTY, &nfit_mem->flags);
1822         }
1823
1824         if (smart.flags & ND_INTEL_SMART_SHUTDOWN_COUNT_VALID) {
1825                 set_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags);
1826                 nfit_mem->dirty_shutdown = smart.shutdown_count;
1827         }
1828 }
1829
1830 static void populate_shutdown_status(struct nfit_mem *nfit_mem)
1831 {
1832         /*
1833          * For DIMMs that provide a dynamic facility to retrieve a
1834          * dirty-shutdown status and/or a dirty-shutdown count, cache
1835          * these values in nfit_mem.
1836          */
1837         if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
1838                 nfit_intel_shutdown_status(nfit_mem);
1839 }
1840
1841 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1842                 struct nfit_mem *nfit_mem, u32 device_handle)
1843 {
1844         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1845         struct acpi_device *adev, *adev_dimm;
1846         struct device *dev = acpi_desc->dev;
1847         unsigned long dsm_mask, label_mask;
1848         const guid_t *guid;
1849         int i;
1850         int family = -1;
1851         struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
1852
1853         /* nfit test assumes 1:1 relationship between commands and dsms */
1854         nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1855         nfit_mem->family = NVDIMM_FAMILY_INTEL;
1856         set_bit(NVDIMM_FAMILY_INTEL, &nd_desc->dimm_family_mask);
1857
1858         if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1859                 sprintf(nfit_mem->id, "%04x-%02x-%04x-%08x",
1860                                 be16_to_cpu(dcr->vendor_id),
1861                                 dcr->manufacturing_location,
1862                                 be16_to_cpu(dcr->manufacturing_date),
1863                                 be32_to_cpu(dcr->serial_number));
1864         else
1865                 sprintf(nfit_mem->id, "%04x-%08x",
1866                                 be16_to_cpu(dcr->vendor_id),
1867                                 be32_to_cpu(dcr->serial_number));
1868
1869         adev = to_acpi_dev(acpi_desc);
1870         if (!adev) {
1871                 /* unit test case */
1872                 populate_shutdown_status(nfit_mem);
1873                 return 0;
1874         }
1875
1876         adev_dimm = acpi_find_child_device(adev, device_handle, false);
1877         nfit_mem->adev = adev_dimm;
1878         if (!adev_dimm) {
1879                 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1880                                 device_handle);
1881                 return force_enable_dimms ? 0 : -ENODEV;
1882         }
1883
1884         if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1885                 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1886                 dev_err(dev, "%s: notification registration failed\n",
1887                                 dev_name(&adev_dimm->dev));
1888                 return -ENXIO;
1889         }
1890         /*
1891          * Record nfit_mem for the notification path to track back to
1892          * the nfit sysfs attributes for this dimm device object.
1893          */
1894         dev_set_drvdata(&adev_dimm->dev, nfit_mem);
1895
1896         /*
1897          * There are 4 "legacy" NVDIMM command sets
1898          * (NVDIMM_FAMILY_{INTEL,MSFT,HPE1,HPE2}) that were created before
1899          * an EFI working group was established to constrain this
1900          * proliferation. The nfit driver probes for the supported command
1901          * set by GUID. Note, if you're a platform developer looking to add
1902          * a new command set to this probe, consider using an existing set,
1903          * or otherwise seek approval to publish the command set at
1904          * http://www.uefi.org/RFIC_LIST.
1905          *
1906          * Note, that checking for function0 (bit0) tells us if any commands
1907          * are reachable through this GUID.
1908          */
1909         clear_bit(NVDIMM_FAMILY_INTEL, &nd_desc->dimm_family_mask);
1910         for (i = 0; i <= NVDIMM_FAMILY_MAX; i++)
1911                 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1)) {
1912                         set_bit(i, &nd_desc->dimm_family_mask);
1913                         if (family < 0 || i == default_dsm_family)
1914                                 family = i;
1915                 }
1916
1917         /* limit the supported commands to those that are publicly documented */
1918         nfit_mem->family = family;
1919         if (override_dsm_mask && !disable_vendor_specific)
1920                 dsm_mask = override_dsm_mask;
1921         else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1922                 dsm_mask = NVDIMM_INTEL_CMDMASK;
1923                 if (disable_vendor_specific)
1924                         dsm_mask &= ~(1 << ND_CMD_VENDOR);
1925         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1926                 dsm_mask = 0x1c3c76;
1927         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1928                 dsm_mask = 0x1fe;
1929                 if (disable_vendor_specific)
1930                         dsm_mask &= ~(1 << 8);
1931         } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1932                 dsm_mask = 0xffffffff;
1933         } else if (nfit_mem->family == NVDIMM_FAMILY_HYPERV) {
1934                 dsm_mask = 0x1f;
1935         } else {
1936                 dev_dbg(dev, "unknown dimm command family\n");
1937                 nfit_mem->family = -1;
1938                 /* DSMs are optional, continue loading the driver... */
1939                 return 0;
1940         }
1941
1942         /*
1943          * Function 0 is the command interrogation function, don't
1944          * export it to potential userspace use, and enable it to be
1945          * used as an error value in acpi_nfit_ctl().
1946          */
1947         dsm_mask &= ~1UL;
1948
1949         guid = to_nfit_uuid(nfit_mem->family);
1950         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1951                 if (acpi_check_dsm(adev_dimm->handle, guid,
1952                                         nfit_dsm_revid(nfit_mem->family, i),
1953                                         1ULL << i))
1954                         set_bit(i, &nfit_mem->dsm_mask);
1955
1956         /*
1957          * Prefer the NVDIMM_FAMILY_INTEL label read commands if present
1958          * due to their better semantics handling locked capacity.
1959          */
1960         label_mask = 1 << ND_CMD_GET_CONFIG_SIZE | 1 << ND_CMD_GET_CONFIG_DATA
1961                 | 1 << ND_CMD_SET_CONFIG_DATA;
1962         if (family == NVDIMM_FAMILY_INTEL
1963                         && (dsm_mask & label_mask) == label_mask)
1964                 /* skip _LS{I,R,W} enabling */;
1965         else {
1966                 if (acpi_nvdimm_has_method(adev_dimm, "_LSI")
1967                                 && acpi_nvdimm_has_method(adev_dimm, "_LSR")) {
1968                         dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev));
1969                         set_bit(NFIT_MEM_LSR, &nfit_mem->flags);
1970                 }
1971
1972                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
1973                                 && acpi_nvdimm_has_method(adev_dimm, "_LSW")) {
1974                         dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev));
1975                         set_bit(NFIT_MEM_LSW, &nfit_mem->flags);
1976                 }
1977
1978                 /*
1979                  * Quirk read-only label configurations to preserve
1980                  * access to label-less namespaces by default.
1981                  */
1982                 if (!test_bit(NFIT_MEM_LSW, &nfit_mem->flags)
1983                                 && !force_labels) {
1984                         dev_dbg(dev, "%s: No _LSW, disable labels\n",
1985                                         dev_name(&adev_dimm->dev));
1986                         clear_bit(NFIT_MEM_LSR, &nfit_mem->flags);
1987                 } else
1988                         dev_dbg(dev, "%s: Force enable labels\n",
1989                                         dev_name(&adev_dimm->dev));
1990         }
1991
1992         populate_shutdown_status(nfit_mem);
1993
1994         return 0;
1995 }
1996
1997 static void shutdown_dimm_notify(void *data)
1998 {
1999         struct acpi_nfit_desc *acpi_desc = data;
2000         struct nfit_mem *nfit_mem;
2001
2002         mutex_lock(&acpi_desc->init_mutex);
2003         /*
2004          * Clear out the nfit_mem->flags_attr and shut down dimm event
2005          * notifications.
2006          */
2007         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2008                 struct acpi_device *adev_dimm = nfit_mem->adev;
2009
2010                 if (nfit_mem->flags_attr) {
2011                         sysfs_put(nfit_mem->flags_attr);
2012                         nfit_mem->flags_attr = NULL;
2013                 }
2014                 if (adev_dimm) {
2015                         acpi_remove_notify_handler(adev_dimm->handle,
2016                                         ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
2017                         dev_set_drvdata(&adev_dimm->dev, NULL);
2018                 }
2019         }
2020         mutex_unlock(&acpi_desc->init_mutex);
2021 }
2022
2023 static const struct nvdimm_security_ops *acpi_nfit_get_security_ops(int family)
2024 {
2025         switch (family) {
2026         case NVDIMM_FAMILY_INTEL:
2027                 return intel_security_ops;
2028         default:
2029                 return NULL;
2030         }
2031 }
2032
2033 static const struct nvdimm_fw_ops *acpi_nfit_get_fw_ops(
2034                 struct nfit_mem *nfit_mem)
2035 {
2036         unsigned long mask;
2037         struct acpi_nfit_desc *acpi_desc = nfit_mem->acpi_desc;
2038         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2039
2040         if (!nd_desc->fw_ops)
2041                 return NULL;
2042
2043         if (nfit_mem->family != NVDIMM_FAMILY_INTEL)
2044                 return NULL;
2045
2046         mask = nfit_mem->dsm_mask & NVDIMM_INTEL_FW_ACTIVATE_CMDMASK;
2047         if (mask != NVDIMM_INTEL_FW_ACTIVATE_CMDMASK)
2048                 return NULL;
2049
2050         return intel_fw_ops;
2051 }
2052
2053 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
2054 {
2055         struct nfit_mem *nfit_mem;
2056         int dimm_count = 0, rc;
2057         struct nvdimm *nvdimm;
2058
2059         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2060                 struct acpi_nfit_flush_address *flush;
2061                 unsigned long flags = 0, cmd_mask;
2062                 struct nfit_memdev *nfit_memdev;
2063                 u32 device_handle;
2064                 u16 mem_flags;
2065
2066                 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
2067                 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
2068                 if (nvdimm) {
2069                         dimm_count++;
2070                         continue;
2071                 }
2072
2073                 if (nfit_mem->bdw && nfit_mem->memdev_pmem) {
2074                         set_bit(NDD_ALIASING, &flags);
2075                         set_bit(NDD_LABELING, &flags);
2076                 }
2077
2078                 /* collate flags across all memdevs for this dimm */
2079                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2080                         struct acpi_nfit_memory_map *dimm_memdev;
2081
2082                         dimm_memdev = __to_nfit_memdev(nfit_mem);
2083                         if (dimm_memdev->device_handle
2084                                         != nfit_memdev->memdev->device_handle)
2085                                 continue;
2086                         dimm_memdev->flags |= nfit_memdev->memdev->flags;
2087                 }
2088
2089                 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
2090                 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
2091                         set_bit(NDD_UNARMED, &flags);
2092
2093                 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
2094                 if (rc)
2095                         continue;
2096
2097                 /*
2098                  * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
2099                  * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
2100                  * userspace interface.
2101                  */
2102                 cmd_mask = 1UL << ND_CMD_CALL;
2103                 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
2104                         /*
2105                          * These commands have a 1:1 correspondence
2106                          * between DSM payload and libnvdimm ioctl
2107                          * payload format.
2108                          */
2109                         cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
2110                 }
2111
2112                 /* Quirk to ignore LOCAL for labels on HYPERV DIMMs */
2113                 if (nfit_mem->family == NVDIMM_FAMILY_HYPERV)
2114                         set_bit(NDD_NOBLK, &flags);
2115
2116                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
2117                         set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
2118                         set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
2119                 }
2120                 if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags))
2121                         set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
2122
2123                 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
2124                         : NULL;
2125                 nvdimm = __nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
2126                                 acpi_nfit_dimm_attribute_groups,
2127                                 flags, cmd_mask, flush ? flush->hint_count : 0,
2128                                 nfit_mem->flush_wpq, &nfit_mem->id[0],
2129                                 acpi_nfit_get_security_ops(nfit_mem->family),
2130                                 acpi_nfit_get_fw_ops(nfit_mem));
2131                 if (!nvdimm)
2132                         return -ENOMEM;
2133
2134                 nfit_mem->nvdimm = nvdimm;
2135                 dimm_count++;
2136
2137                 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
2138                         continue;
2139
2140                 dev_err(acpi_desc->dev, "Error found in NVDIMM %s flags:%s%s%s%s%s\n",
2141                                 nvdimm_name(nvdimm),
2142                   mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
2143                   mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
2144                   mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
2145                   mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
2146                   mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
2147
2148         }
2149
2150         rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
2151         if (rc)
2152                 return rc;
2153
2154         /*
2155          * Now that dimms are successfully registered, and async registration
2156          * is flushed, attempt to enable event notification.
2157          */
2158         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2159                 struct kernfs_node *nfit_kernfs;
2160
2161                 nvdimm = nfit_mem->nvdimm;
2162                 if (!nvdimm)
2163                         continue;
2164
2165                 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
2166                 if (nfit_kernfs)
2167                         nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
2168                                         "flags");
2169                 sysfs_put(nfit_kernfs);
2170                 if (!nfit_mem->flags_attr)
2171                         dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
2172                                         nvdimm_name(nvdimm));
2173         }
2174
2175         return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
2176                         acpi_desc);
2177 }
2178
2179 /*
2180  * These constants are private because there are no kernel consumers of
2181  * these commands.
2182  */
2183 enum nfit_aux_cmds {
2184         NFIT_CMD_TRANSLATE_SPA = 5,
2185         NFIT_CMD_ARS_INJECT_SET = 7,
2186         NFIT_CMD_ARS_INJECT_CLEAR = 8,
2187         NFIT_CMD_ARS_INJECT_GET = 9,
2188 };
2189
2190 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
2191 {
2192         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2193         const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
2194         unsigned long dsm_mask, *mask;
2195         struct acpi_device *adev;
2196         int i;
2197
2198         set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
2199         set_bit(NVDIMM_BUS_FAMILY_NFIT, &nd_desc->bus_family_mask);
2200
2201         /* enable nfit_test to inject bus command emulation */
2202         if (acpi_desc->bus_cmd_force_en) {
2203                 nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
2204                 mask = &nd_desc->bus_family_mask;
2205                 if (acpi_desc->family_dsm_mask[NVDIMM_BUS_FAMILY_INTEL]) {
2206                         set_bit(NVDIMM_BUS_FAMILY_INTEL, mask);
2207                         nd_desc->fw_ops = intel_bus_fw_ops;
2208                 }
2209         }
2210
2211         adev = to_acpi_dev(acpi_desc);
2212         if (!adev)
2213                 return;
2214
2215         for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
2216                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2217                         set_bit(i, &nd_desc->cmd_mask);
2218
2219         dsm_mask =
2220                 (1 << ND_CMD_ARS_CAP) |
2221                 (1 << ND_CMD_ARS_START) |
2222                 (1 << ND_CMD_ARS_STATUS) |
2223                 (1 << ND_CMD_CLEAR_ERROR) |
2224                 (1 << NFIT_CMD_TRANSLATE_SPA) |
2225                 (1 << NFIT_CMD_ARS_INJECT_SET) |
2226                 (1 << NFIT_CMD_ARS_INJECT_CLEAR) |
2227                 (1 << NFIT_CMD_ARS_INJECT_GET);
2228         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
2229                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2230                         set_bit(i, &acpi_desc->bus_dsm_mask);
2231
2232         /* Enumerate allowed NVDIMM_BUS_FAMILY_INTEL commands */
2233         dsm_mask = NVDIMM_BUS_INTEL_FW_ACTIVATE_CMDMASK;
2234         guid = to_nfit_bus_uuid(NVDIMM_BUS_FAMILY_INTEL);
2235         mask = &acpi_desc->family_dsm_mask[NVDIMM_BUS_FAMILY_INTEL];
2236         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
2237                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2238                         set_bit(i, mask);
2239
2240         if (*mask == dsm_mask) {
2241                 set_bit(NVDIMM_BUS_FAMILY_INTEL, &nd_desc->bus_family_mask);
2242                 nd_desc->fw_ops = intel_bus_fw_ops;
2243         }
2244 }
2245
2246 static ssize_t range_index_show(struct device *dev,
2247                 struct device_attribute *attr, char *buf)
2248 {
2249         struct nd_region *nd_region = to_nd_region(dev);
2250         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
2251
2252         return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
2253 }
2254 static DEVICE_ATTR_RO(range_index);
2255
2256 static struct attribute *acpi_nfit_region_attributes[] = {
2257         &dev_attr_range_index.attr,
2258         NULL,
2259 };
2260
2261 static const struct attribute_group acpi_nfit_region_attribute_group = {
2262         .name = "nfit",
2263         .attrs = acpi_nfit_region_attributes,
2264 };
2265
2266 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
2267         &acpi_nfit_region_attribute_group,
2268         NULL,
2269 };
2270
2271 /* enough info to uniquely specify an interleave set */
2272 struct nfit_set_info {
2273         u64 region_offset;
2274         u32 serial_number;
2275         u32 pad;
2276 };
2277
2278 struct nfit_set_info2 {
2279         u64 region_offset;
2280         u32 serial_number;
2281         u16 vendor_id;
2282         u16 manufacturing_date;
2283         u8 manufacturing_location;
2284         u8 reserved[31];
2285 };
2286
2287 static int cmp_map_compat(const void *m0, const void *m1)
2288 {
2289         const struct nfit_set_info *map0 = m0;
2290         const struct nfit_set_info *map1 = m1;
2291
2292         return memcmp(&map0->region_offset, &map1->region_offset,
2293                         sizeof(u64));
2294 }
2295
2296 static int cmp_map(const void *m0, const void *m1)
2297 {
2298         const struct nfit_set_info *map0 = m0;
2299         const struct nfit_set_info *map1 = m1;
2300
2301         if (map0->region_offset < map1->region_offset)
2302                 return -1;
2303         else if (map0->region_offset > map1->region_offset)
2304                 return 1;
2305         return 0;
2306 }
2307
2308 static int cmp_map2(const void *m0, const void *m1)
2309 {
2310         const struct nfit_set_info2 *map0 = m0;
2311         const struct nfit_set_info2 *map1 = m1;
2312
2313         if (map0->region_offset < map1->region_offset)
2314                 return -1;
2315         else if (map0->region_offset > map1->region_offset)
2316                 return 1;
2317         return 0;
2318 }
2319
2320 /* Retrieve the nth entry referencing this spa */
2321 static struct acpi_nfit_memory_map *memdev_from_spa(
2322                 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
2323 {
2324         struct nfit_memdev *nfit_memdev;
2325
2326         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
2327                 if (nfit_memdev->memdev->range_index == range_index)
2328                         if (n-- == 0)
2329                                 return nfit_memdev->memdev;
2330         return NULL;
2331 }
2332
2333 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
2334                 struct nd_region_desc *ndr_desc,
2335                 struct acpi_nfit_system_address *spa)
2336 {
2337         struct device *dev = acpi_desc->dev;
2338         struct nd_interleave_set *nd_set;
2339         u16 nr = ndr_desc->num_mappings;
2340         struct nfit_set_info2 *info2;
2341         struct nfit_set_info *info;
2342         int i;
2343
2344         nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
2345         if (!nd_set)
2346                 return -ENOMEM;
2347         import_guid(&nd_set->type_guid, spa->range_guid);
2348
2349         info = devm_kcalloc(dev, nr, sizeof(*info), GFP_KERNEL);
2350         if (!info)
2351                 return -ENOMEM;
2352
2353         info2 = devm_kcalloc(dev, nr, sizeof(*info2), GFP_KERNEL);
2354         if (!info2)
2355                 return -ENOMEM;
2356
2357         for (i = 0; i < nr; i++) {
2358                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
2359                 struct nvdimm *nvdimm = mapping->nvdimm;
2360                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2361                 struct nfit_set_info *map = &info[i];
2362                 struct nfit_set_info2 *map2 = &info2[i];
2363                 struct acpi_nfit_memory_map *memdev =
2364                         memdev_from_spa(acpi_desc, spa->range_index, i);
2365                 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2366
2367                 if (!memdev || !nfit_mem->dcr) {
2368                         dev_err(dev, "%s: failed to find DCR\n", __func__);
2369                         return -ENODEV;
2370                 }
2371
2372                 map->region_offset = memdev->region_offset;
2373                 map->serial_number = dcr->serial_number;
2374
2375                 map2->region_offset = memdev->region_offset;
2376                 map2->serial_number = dcr->serial_number;
2377                 map2->vendor_id = dcr->vendor_id;
2378                 map2->manufacturing_date = dcr->manufacturing_date;
2379                 map2->manufacturing_location = dcr->manufacturing_location;
2380         }
2381
2382         /* v1.1 namespaces */
2383         sort(info, nr, sizeof(*info), cmp_map, NULL);
2384         nd_set->cookie1 = nd_fletcher64(info, sizeof(*info) * nr, 0);
2385
2386         /* v1.2 namespaces */
2387         sort(info2, nr, sizeof(*info2), cmp_map2, NULL);
2388         nd_set->cookie2 = nd_fletcher64(info2, sizeof(*info2) * nr, 0);
2389
2390         /* support v1.1 namespaces created with the wrong sort order */
2391         sort(info, nr, sizeof(*info), cmp_map_compat, NULL);
2392         nd_set->altcookie = nd_fletcher64(info, sizeof(*info) * nr, 0);
2393
2394         /* record the result of the sort for the mapping position */
2395         for (i = 0; i < nr; i++) {
2396                 struct nfit_set_info2 *map2 = &info2[i];
2397                 int j;
2398
2399                 for (j = 0; j < nr; j++) {
2400                         struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
2401                         struct nvdimm *nvdimm = mapping->nvdimm;
2402                         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2403                         struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2404
2405                         if (map2->serial_number == dcr->serial_number &&
2406                             map2->vendor_id == dcr->vendor_id &&
2407                             map2->manufacturing_date == dcr->manufacturing_date &&
2408                             map2->manufacturing_location
2409                                     == dcr->manufacturing_location) {
2410                                 mapping->position = i;
2411                                 break;
2412                         }
2413                 }
2414         }
2415
2416         ndr_desc->nd_set = nd_set;
2417         devm_kfree(dev, info);
2418         devm_kfree(dev, info2);
2419
2420         return 0;
2421 }
2422
2423 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
2424 {
2425         struct acpi_nfit_interleave *idt = mmio->idt;
2426         u32 sub_line_offset, line_index, line_offset;
2427         u64 line_no, table_skip_count, table_offset;
2428
2429         line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
2430         table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
2431         line_offset = idt->line_offset[line_index]
2432                 * mmio->line_size;
2433         table_offset = table_skip_count * mmio->table_size;
2434
2435         return mmio->base_offset + line_offset + table_offset + sub_line_offset;
2436 }
2437
2438 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
2439 {
2440         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2441         u64 offset = nfit_blk->stat_offset + mmio->size * bw;
2442         const u32 STATUS_MASK = 0x80000037;
2443
2444         if (mmio->num_lines)
2445                 offset = to_interleave_offset(offset, mmio);
2446
2447         return readl(mmio->addr.base + offset) & STATUS_MASK;
2448 }
2449
2450 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
2451                 resource_size_t dpa, unsigned int len, unsigned int write)
2452 {
2453         u64 cmd, offset;
2454         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2455
2456         enum {
2457                 BCW_OFFSET_MASK = (1ULL << 48)-1,
2458                 BCW_LEN_SHIFT = 48,
2459                 BCW_LEN_MASK = (1ULL << 8) - 1,
2460                 BCW_CMD_SHIFT = 56,
2461         };
2462
2463         cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
2464         len = len >> L1_CACHE_SHIFT;
2465         cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
2466         cmd |= ((u64) write) << BCW_CMD_SHIFT;
2467
2468         offset = nfit_blk->cmd_offset + mmio->size * bw;
2469         if (mmio->num_lines)
2470                 offset = to_interleave_offset(offset, mmio);
2471
2472         writeq(cmd, mmio->addr.base + offset);
2473         nvdimm_flush(nfit_blk->nd_region, NULL);
2474
2475         if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
2476                 readq(mmio->addr.base + offset);
2477 }
2478
2479 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
2480                 resource_size_t dpa, void *iobuf, size_t len, int rw,
2481                 unsigned int lane)
2482 {
2483         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2484         unsigned int copied = 0;
2485         u64 base_offset;
2486         int rc;
2487
2488         base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
2489                 + lane * mmio->size;
2490         write_blk_ctl(nfit_blk, lane, dpa, len, rw);
2491         while (len) {
2492                 unsigned int c;
2493                 u64 offset;
2494
2495                 if (mmio->num_lines) {
2496                         u32 line_offset;
2497
2498                         offset = to_interleave_offset(base_offset + copied,
2499                                         mmio);
2500                         div_u64_rem(offset, mmio->line_size, &line_offset);
2501                         c = min_t(size_t, len, mmio->line_size - line_offset);
2502                 } else {
2503                         offset = base_offset + nfit_blk->bdw_offset;
2504                         c = len;
2505                 }
2506
2507                 if (rw)
2508                         memcpy_flushcache(mmio->addr.aperture + offset, iobuf + copied, c);
2509                 else {
2510                         if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
2511                                 arch_invalidate_pmem((void __force *)
2512                                         mmio->addr.aperture + offset, c);
2513
2514                         memcpy(iobuf + copied, mmio->addr.aperture + offset, c);
2515                 }
2516
2517                 copied += c;
2518                 len -= c;
2519         }
2520
2521         if (rw)
2522                 nvdimm_flush(nfit_blk->nd_region, NULL);
2523
2524         rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
2525         return rc;
2526 }
2527
2528 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
2529                 resource_size_t dpa, void *iobuf, u64 len, int rw)
2530 {
2531         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
2532         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2533         struct nd_region *nd_region = nfit_blk->nd_region;
2534         unsigned int lane, copied = 0;
2535         int rc = 0;
2536
2537         lane = nd_region_acquire_lane(nd_region);
2538         while (len) {
2539                 u64 c = min(len, mmio->size);
2540
2541                 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
2542                                 iobuf + copied, c, rw, lane);
2543                 if (rc)
2544                         break;
2545
2546                 copied += c;
2547                 len -= c;
2548         }
2549         nd_region_release_lane(nd_region, lane);
2550
2551         return rc;
2552 }
2553
2554 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
2555                 struct acpi_nfit_interleave *idt, u16 interleave_ways)
2556 {
2557         if (idt) {
2558                 mmio->num_lines = idt->line_count;
2559                 mmio->line_size = idt->line_size;
2560                 if (interleave_ways == 0)
2561                         return -ENXIO;
2562                 mmio->table_size = mmio->num_lines * interleave_ways
2563                         * mmio->line_size;
2564         }
2565
2566         return 0;
2567 }
2568
2569 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
2570                 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
2571 {
2572         struct nd_cmd_dimm_flags flags;
2573         int rc;
2574
2575         memset(&flags, 0, sizeof(flags));
2576         rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
2577                         sizeof(flags), NULL);
2578
2579         if (rc >= 0 && flags.status == 0)
2580                 nfit_blk->dimm_flags = flags.flags;
2581         else if (rc == -ENOTTY) {
2582                 /* fall back to a conservative default */
2583                 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
2584                 rc = 0;
2585         } else
2586                 rc = -ENXIO;
2587
2588         return rc;
2589 }
2590
2591 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
2592                 struct device *dev)
2593 {
2594         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
2595         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
2596         struct nfit_blk_mmio *mmio;
2597         struct nfit_blk *nfit_blk;
2598         struct nfit_mem *nfit_mem;
2599         struct nvdimm *nvdimm;
2600         int rc;
2601
2602         nvdimm = nd_blk_region_to_dimm(ndbr);
2603         nfit_mem = nvdimm_provider_data(nvdimm);
2604         if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
2605                 dev_dbg(dev, "missing%s%s%s\n",
2606                                 nfit_mem ? "" : " nfit_mem",
2607                                 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
2608                                 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
2609                 return -ENXIO;
2610         }
2611
2612         nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
2613         if (!nfit_blk)
2614                 return -ENOMEM;
2615         nd_blk_region_set_provider_data(ndbr, nfit_blk);
2616         nfit_blk->nd_region = to_nd_region(dev);
2617
2618         /* map block aperture memory */
2619         nfit_blk->bdw_offset = nfit_mem->bdw->offset;
2620         mmio = &nfit_blk->mmio[BDW];
2621         mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
2622                         nfit_mem->spa_bdw->length, nd_blk_memremap_flags(ndbr));
2623         if (!mmio->addr.base) {
2624                 dev_dbg(dev, "%s failed to map bdw\n",
2625                                 nvdimm_name(nvdimm));
2626                 return -ENOMEM;
2627         }
2628         mmio->size = nfit_mem->bdw->size;
2629         mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
2630         mmio->idt = nfit_mem->idt_bdw;
2631         mmio->spa = nfit_mem->spa_bdw;
2632         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
2633                         nfit_mem->memdev_bdw->interleave_ways);
2634         if (rc) {
2635                 dev_dbg(dev, "%s failed to init bdw interleave\n",
2636                                 nvdimm_name(nvdimm));
2637                 return rc;
2638         }
2639
2640         /* map block control memory */
2641         nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
2642         nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
2643         mmio = &nfit_blk->mmio[DCR];
2644         mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
2645                         nfit_mem->spa_dcr->length);
2646         if (!mmio->addr.base) {
2647                 dev_dbg(dev, "%s failed to map dcr\n",
2648                                 nvdimm_name(nvdimm));
2649                 return -ENOMEM;
2650         }
2651         mmio->size = nfit_mem->dcr->window_size;
2652         mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
2653         mmio->idt = nfit_mem->idt_dcr;
2654         mmio->spa = nfit_mem->spa_dcr;
2655         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
2656                         nfit_mem->memdev_dcr->interleave_ways);
2657         if (rc) {
2658                 dev_dbg(dev, "%s failed to init dcr interleave\n",
2659                                 nvdimm_name(nvdimm));
2660                 return rc;
2661         }
2662
2663         rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
2664         if (rc < 0) {
2665                 dev_dbg(dev, "%s failed get DIMM flags\n",
2666                                 nvdimm_name(nvdimm));
2667                 return rc;
2668         }
2669
2670         if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
2671                 dev_warn(dev, "unable to guarantee persistence of writes\n");
2672
2673         if (mmio->line_size == 0)
2674                 return 0;
2675
2676         if ((u32) nfit_blk->cmd_offset % mmio->line_size
2677                         + 8 > mmio->line_size) {
2678                 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
2679                 return -ENXIO;
2680         } else if ((u32) nfit_blk->stat_offset % mmio->line_size
2681                         + 8 > mmio->line_size) {
2682                 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
2683                 return -ENXIO;
2684         }
2685
2686         return 0;
2687 }
2688
2689 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2690                 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2691 {
2692         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2693         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2694         int cmd_rc, rc;
2695
2696         cmd->address = spa->address;
2697         cmd->length = spa->length;
2698         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2699                         sizeof(*cmd), &cmd_rc);
2700         if (rc < 0)
2701                 return rc;
2702         return cmd_rc;
2703 }
2704
2705 static int ars_start(struct acpi_nfit_desc *acpi_desc,
2706                 struct nfit_spa *nfit_spa, enum nfit_ars_state req_type)
2707 {
2708         int rc;
2709         int cmd_rc;
2710         struct nd_cmd_ars_start ars_start;
2711         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2712         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2713
2714         memset(&ars_start, 0, sizeof(ars_start));
2715         ars_start.address = spa->address;
2716         ars_start.length = spa->length;
2717         if (req_type == ARS_REQ_SHORT)
2718                 ars_start.flags = ND_ARS_RETURN_PREV_DATA;
2719         if (nfit_spa_type(spa) == NFIT_SPA_PM)
2720                 ars_start.type = ND_ARS_PERSISTENT;
2721         else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2722                 ars_start.type = ND_ARS_VOLATILE;
2723         else
2724                 return -ENOTTY;
2725
2726         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2727                         sizeof(ars_start), &cmd_rc);
2728
2729         if (rc < 0)
2730                 return rc;
2731         if (cmd_rc < 0)
2732                 return cmd_rc;
2733         set_bit(ARS_VALID, &acpi_desc->scrub_flags);
2734         return 0;
2735 }
2736
2737 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2738 {
2739         int rc, cmd_rc;
2740         struct nd_cmd_ars_start ars_start;
2741         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2742         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2743
2744         ars_start = (struct nd_cmd_ars_start) {
2745                 .address = ars_status->restart_address,
2746                 .length = ars_status->restart_length,
2747                 .type = ars_status->type,
2748         };
2749         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2750                         sizeof(ars_start), &cmd_rc);
2751         if (rc < 0)
2752                 return rc;
2753         return cmd_rc;
2754 }
2755
2756 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2757 {
2758         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2759         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2760         int rc, cmd_rc;
2761
2762         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2763                         acpi_desc->max_ars, &cmd_rc);
2764         if (rc < 0)
2765                 return rc;
2766         return cmd_rc;
2767 }
2768
2769 static void ars_complete(struct acpi_nfit_desc *acpi_desc,
2770                 struct nfit_spa *nfit_spa)
2771 {
2772         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2773         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2774         struct nd_region *nd_region = nfit_spa->nd_region;
2775         struct device *dev;
2776
2777         lockdep_assert_held(&acpi_desc->init_mutex);
2778         /*
2779          * Only advance the ARS state for ARS runs initiated by the
2780          * kernel, ignore ARS results from BIOS initiated runs for scrub
2781          * completion tracking.
2782          */
2783         if (acpi_desc->scrub_spa != nfit_spa)
2784                 return;
2785
2786         if ((ars_status->address >= spa->address && ars_status->address
2787                                 < spa->address + spa->length)
2788                         || (ars_status->address < spa->address)) {
2789                 /*
2790                  * Assume that if a scrub starts at an offset from the
2791                  * start of nfit_spa that we are in the continuation
2792                  * case.
2793                  *
2794                  * Otherwise, if the scrub covers the spa range, mark
2795                  * any pending request complete.
2796                  */
2797                 if (ars_status->address + ars_status->length
2798                                 >= spa->address + spa->length)
2799                                 /* complete */;
2800                 else
2801                         return;
2802         } else
2803                 return;
2804
2805         acpi_desc->scrub_spa = NULL;
2806         if (nd_region) {
2807                 dev = nd_region_dev(nd_region);
2808                 nvdimm_region_notify(nd_region, NVDIMM_REVALIDATE_POISON);
2809         } else
2810                 dev = acpi_desc->dev;
2811         dev_dbg(dev, "ARS: range %d complete\n", spa->range_index);
2812 }
2813
2814 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc)
2815 {
2816         struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2817         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2818         int rc;
2819         u32 i;
2820
2821         /*
2822          * First record starts at 44 byte offset from the start of the
2823          * payload.
2824          */
2825         if (ars_status->out_length < 44)
2826                 return 0;
2827
2828         /*
2829          * Ignore potentially stale results that are only refreshed
2830          * after a start-ARS event.
2831          */
2832         if (!test_and_clear_bit(ARS_VALID, &acpi_desc->scrub_flags)) {
2833                 dev_dbg(acpi_desc->dev, "skip %d stale records\n",
2834                                 ars_status->num_records);
2835                 return 0;
2836         }
2837
2838         for (i = 0; i < ars_status->num_records; i++) {
2839                 /* only process full records */
2840                 if (ars_status->out_length
2841                                 < 44 + sizeof(struct nd_ars_record) * (i + 1))
2842                         break;
2843                 rc = nvdimm_bus_add_badrange(nvdimm_bus,
2844                                 ars_status->records[i].err_address,
2845                                 ars_status->records[i].length);
2846                 if (rc)
2847                         return rc;
2848         }
2849         if (i < ars_status->num_records)
2850                 dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2851
2852         return 0;
2853 }
2854
2855 static void acpi_nfit_remove_resource(void *data)
2856 {
2857         struct resource *res = data;
2858
2859         remove_resource(res);
2860 }
2861
2862 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2863                 struct nd_region_desc *ndr_desc)
2864 {
2865         struct resource *res, *nd_res = ndr_desc->res;
2866         int is_pmem, ret;
2867
2868         /* No operation if the region is already registered as PMEM */
2869         is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2870                                 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2871         if (is_pmem == REGION_INTERSECTS)
2872                 return 0;
2873
2874         res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2875         if (!res)
2876                 return -ENOMEM;
2877
2878         res->name = "Persistent Memory";
2879         res->start = nd_res->start;
2880         res->end = nd_res->end;
2881         res->flags = IORESOURCE_MEM;
2882         res->desc = IORES_DESC_PERSISTENT_MEMORY;
2883
2884         ret = insert_resource(&iomem_resource, res);
2885         if (ret)
2886                 return ret;
2887
2888         ret = devm_add_action_or_reset(acpi_desc->dev,
2889                                         acpi_nfit_remove_resource,
2890                                         res);
2891         if (ret)
2892                 return ret;
2893
2894         return 0;
2895 }
2896
2897 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2898                 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2899                 struct acpi_nfit_memory_map *memdev,
2900                 struct nfit_spa *nfit_spa)
2901 {
2902         struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2903                         memdev->device_handle);
2904         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2905         struct nd_blk_region_desc *ndbr_desc;
2906         struct nfit_mem *nfit_mem;
2907         int rc;
2908
2909         if (!nvdimm) {
2910                 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2911                                 spa->range_index, memdev->device_handle);
2912                 return -ENODEV;
2913         }
2914
2915         mapping->nvdimm = nvdimm;
2916         switch (nfit_spa_type(spa)) {
2917         case NFIT_SPA_PM:
2918         case NFIT_SPA_VOLATILE:
2919                 mapping->start = memdev->address;
2920                 mapping->size = memdev->region_size;
2921                 break;
2922         case NFIT_SPA_DCR:
2923                 nfit_mem = nvdimm_provider_data(nvdimm);
2924                 if (!nfit_mem || !nfit_mem->bdw) {
2925                         dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2926                                         spa->range_index, nvdimm_name(nvdimm));
2927                         break;
2928                 }
2929
2930                 mapping->size = nfit_mem->bdw->capacity;
2931                 mapping->start = nfit_mem->bdw->start_address;
2932                 ndr_desc->num_lanes = nfit_mem->bdw->windows;
2933                 ndr_desc->mapping = mapping;
2934                 ndr_desc->num_mappings = 1;
2935                 ndbr_desc = to_blk_region_desc(ndr_desc);
2936                 ndbr_desc->enable = acpi_nfit_blk_region_enable;
2937                 ndbr_desc->do_io = acpi_desc->blk_do_io;
2938                 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2939                 if (rc)
2940                         return rc;
2941                 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2942                                 ndr_desc);
2943                 if (!nfit_spa->nd_region)
2944                         return -ENOMEM;
2945                 break;
2946         }
2947
2948         return 0;
2949 }
2950
2951 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2952 {
2953         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2954                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2955                 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2956                 nfit_spa_type(spa) == NFIT_SPA_PCD);
2957 }
2958
2959 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
2960 {
2961         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2962                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2963                 nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
2964 }
2965
2966 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2967                 struct nfit_spa *nfit_spa)
2968 {
2969         static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2970         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2971         struct nd_blk_region_desc ndbr_desc;
2972         struct nd_region_desc *ndr_desc;
2973         struct nfit_memdev *nfit_memdev;
2974         struct nvdimm_bus *nvdimm_bus;
2975         struct resource res;
2976         int count = 0, rc;
2977
2978         if (nfit_spa->nd_region)
2979                 return 0;
2980
2981         if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2982                 dev_dbg(acpi_desc->dev, "detected invalid spa index\n");
2983                 return 0;
2984         }
2985
2986         memset(&res, 0, sizeof(res));
2987         memset(&mappings, 0, sizeof(mappings));
2988         memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2989         res.start = spa->address;
2990         res.end = res.start + spa->length - 1;
2991         ndr_desc = &ndbr_desc.ndr_desc;
2992         ndr_desc->res = &res;
2993         ndr_desc->provider_data = nfit_spa;
2994         ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2995         if (spa->flags & ACPI_NFIT_PROXIMITY_VALID) {
2996                 ndr_desc->numa_node = pxm_to_online_node(spa->proximity_domain);
2997                 ndr_desc->target_node = pxm_to_node(spa->proximity_domain);
2998         } else {
2999                 ndr_desc->numa_node = NUMA_NO_NODE;
3000                 ndr_desc->target_node = NUMA_NO_NODE;
3001         }
3002
3003         /*
3004          * Persistence domain bits are hierarchical, if
3005          * ACPI_NFIT_CAPABILITY_CACHE_FLUSH is set then
3006          * ACPI_NFIT_CAPABILITY_MEM_FLUSH is implied.
3007          */
3008         if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_CACHE_FLUSH)
3009                 set_bit(ND_REGION_PERSIST_CACHE, &ndr_desc->flags);
3010         else if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_MEM_FLUSH)
3011                 set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc->flags);
3012
3013         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
3014                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
3015                 struct nd_mapping_desc *mapping;
3016
3017                 if (memdev->range_index != spa->range_index)
3018                         continue;
3019                 if (count >= ND_MAX_MAPPINGS) {
3020                         dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
3021                                         spa->range_index, ND_MAX_MAPPINGS);
3022                         return -ENXIO;
3023                 }
3024                 mapping = &mappings[count++];
3025                 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
3026                                 memdev, nfit_spa);
3027                 if (rc)
3028                         goto out;
3029         }
3030
3031         ndr_desc->mapping = mappings;
3032         ndr_desc->num_mappings = count;
3033         rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
3034         if (rc)
3035                 goto out;
3036
3037         nvdimm_bus = acpi_desc->nvdimm_bus;
3038         if (nfit_spa_type(spa) == NFIT_SPA_PM) {
3039                 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
3040                 if (rc) {
3041                         dev_warn(acpi_desc->dev,
3042                                 "failed to insert pmem resource to iomem: %d\n",
3043                                 rc);
3044                         goto out;
3045                 }
3046
3047                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
3048                                 ndr_desc);
3049                 if (!nfit_spa->nd_region)
3050                         rc = -ENOMEM;
3051         } else if (nfit_spa_is_volatile(spa)) {
3052                 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
3053                                 ndr_desc);
3054                 if (!nfit_spa->nd_region)
3055                         rc = -ENOMEM;
3056         } else if (nfit_spa_is_virtual(spa)) {
3057                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
3058                                 ndr_desc);
3059                 if (!nfit_spa->nd_region)
3060                         rc = -ENOMEM;
3061         }
3062
3063  out:
3064         if (rc)
3065                 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
3066                                 nfit_spa->spa->range_index);
3067         return rc;
3068 }
3069
3070 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc)
3071 {
3072         struct device *dev = acpi_desc->dev;
3073         struct nd_cmd_ars_status *ars_status;
3074
3075         if (acpi_desc->ars_status) {
3076                 memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
3077                 return 0;
3078         }
3079
3080         ars_status = devm_kzalloc(dev, acpi_desc->max_ars, GFP_KERNEL);
3081         if (!ars_status)
3082                 return -ENOMEM;
3083         acpi_desc->ars_status = ars_status;
3084         return 0;
3085 }
3086
3087 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc)
3088 {
3089         int rc;
3090
3091         if (ars_status_alloc(acpi_desc))
3092                 return -ENOMEM;
3093
3094         rc = ars_get_status(acpi_desc);
3095
3096         if (rc < 0 && rc != -ENOSPC)
3097                 return rc;
3098
3099         if (ars_status_process_records(acpi_desc))
3100                 dev_err(acpi_desc->dev, "Failed to process ARS records\n");
3101
3102         return rc;
3103 }
3104
3105 static int ars_register(struct acpi_nfit_desc *acpi_desc,
3106                 struct nfit_spa *nfit_spa)
3107 {
3108         int rc;
3109
3110         if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3111                 return acpi_nfit_register_region(acpi_desc, nfit_spa);
3112
3113         set_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
3114         if (!no_init_ars)
3115                 set_bit(ARS_REQ_LONG, &nfit_spa->ars_state);
3116
3117         switch (acpi_nfit_query_poison(acpi_desc)) {
3118         case 0:
3119         case -ENOSPC:
3120         case -EAGAIN:
3121                 rc = ars_start(acpi_desc, nfit_spa, ARS_REQ_SHORT);
3122                 /* shouldn't happen, try again later */
3123                 if (rc == -EBUSY)
3124                         break;
3125                 if (rc) {
3126                         set_bit(ARS_FAILED, &nfit_spa->ars_state);
3127                         break;
3128                 }
3129                 clear_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
3130                 rc = acpi_nfit_query_poison(acpi_desc);
3131                 if (rc)
3132                         break;
3133                 acpi_desc->scrub_spa = nfit_spa;
3134                 ars_complete(acpi_desc, nfit_spa);
3135                 /*
3136                  * If ars_complete() says we didn't complete the
3137                  * short scrub, we'll try again with a long
3138                  * request.
3139                  */
3140                 acpi_desc->scrub_spa = NULL;
3141                 break;
3142         case -EBUSY:
3143         case -ENOMEM:
3144                 /*
3145                  * BIOS was using ARS, wait for it to complete (or
3146                  * resources to become available) and then perform our
3147                  * own scrubs.
3148                  */
3149                 break;
3150         default:
3151                 set_bit(ARS_FAILED, &nfit_spa->ars_state);
3152                 break;
3153         }
3154
3155         return acpi_nfit_register_region(acpi_desc, nfit_spa);
3156 }
3157
3158 static void ars_complete_all(struct acpi_nfit_desc *acpi_desc)
3159 {
3160         struct nfit_spa *nfit_spa;
3161
3162         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3163                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3164                         continue;
3165                 ars_complete(acpi_desc, nfit_spa);
3166         }
3167 }
3168
3169 static unsigned int __acpi_nfit_scrub(struct acpi_nfit_desc *acpi_desc,
3170                 int query_rc)
3171 {
3172         unsigned int tmo = acpi_desc->scrub_tmo;
3173         struct device *dev = acpi_desc->dev;
3174         struct nfit_spa *nfit_spa;
3175
3176         lockdep_assert_held(&acpi_desc->init_mutex);
3177
3178         if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags))
3179                 return 0;
3180
3181         if (query_rc == -EBUSY) {
3182                 dev_dbg(dev, "ARS: ARS busy\n");
3183                 return min(30U * 60U, tmo * 2);
3184         }
3185         if (query_rc == -ENOSPC) {
3186                 dev_dbg(dev, "ARS: ARS continue\n");
3187                 ars_continue(acpi_desc);
3188                 return 1;
3189         }
3190         if (query_rc && query_rc != -EAGAIN) {
3191                 unsigned long long addr, end;
3192
3193                 addr = acpi_desc->ars_status->address;
3194                 end = addr + acpi_desc->ars_status->length;
3195                 dev_dbg(dev, "ARS: %llx-%llx failed (%d)\n", addr, end,
3196                                 query_rc);
3197         }
3198
3199         ars_complete_all(acpi_desc);
3200         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3201                 enum nfit_ars_state req_type;
3202                 int rc;
3203
3204                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3205                         continue;
3206
3207                 /* prefer short ARS requests first */
3208                 if (test_bit(ARS_REQ_SHORT, &nfit_spa->ars_state))
3209                         req_type = ARS_REQ_SHORT;
3210                 else if (test_bit(ARS_REQ_LONG, &nfit_spa->ars_state))
3211                         req_type = ARS_REQ_LONG;
3212                 else
3213                         continue;
3214                 rc = ars_start(acpi_desc, nfit_spa, req_type);
3215
3216                 dev = nd_region_dev(nfit_spa->nd_region);
3217                 dev_dbg(dev, "ARS: range %d ARS start %s (%d)\n",
3218                                 nfit_spa->spa->range_index,
3219                                 req_type == ARS_REQ_SHORT ? "short" : "long",
3220                                 rc);
3221                 /*
3222                  * Hmm, we raced someone else starting ARS? Try again in
3223                  * a bit.
3224                  */
3225                 if (rc == -EBUSY)
3226                         return 1;
3227                 if (rc == 0) {
3228                         dev_WARN_ONCE(dev, acpi_desc->scrub_spa,
3229                                         "scrub start while range %d active\n",
3230                                         acpi_desc->scrub_spa->spa->range_index);
3231                         clear_bit(req_type, &nfit_spa->ars_state);
3232                         acpi_desc->scrub_spa = nfit_spa;
3233                         /*
3234                          * Consider this spa last for future scrub
3235                          * requests
3236                          */
3237                         list_move_tail(&nfit_spa->list, &acpi_desc->spas);
3238                         return 1;
3239                 }
3240
3241                 dev_err(dev, "ARS: range %d ARS failed (%d)\n",
3242                                 nfit_spa->spa->range_index, rc);
3243                 set_bit(ARS_FAILED, &nfit_spa->ars_state);
3244         }
3245         return 0;
3246 }
3247
3248 static void __sched_ars(struct acpi_nfit_desc *acpi_desc, unsigned int tmo)
3249 {
3250         lockdep_assert_held(&acpi_desc->init_mutex);
3251
3252         set_bit(ARS_BUSY, &acpi_desc->scrub_flags);
3253         /* note this should only be set from within the workqueue */
3254         if (tmo)
3255                 acpi_desc->scrub_tmo = tmo;
3256         queue_delayed_work(nfit_wq, &acpi_desc->dwork, tmo * HZ);
3257 }
3258
3259 static void sched_ars(struct acpi_nfit_desc *acpi_desc)
3260 {
3261         __sched_ars(acpi_desc, 0);
3262 }
3263
3264 static void notify_ars_done(struct acpi_nfit_desc *acpi_desc)
3265 {
3266         lockdep_assert_held(&acpi_desc->init_mutex);
3267
3268         clear_bit(ARS_BUSY, &acpi_desc->scrub_flags);
3269         acpi_desc->scrub_count++;
3270         if (acpi_desc->scrub_count_state)
3271                 sysfs_notify_dirent(acpi_desc->scrub_count_state);
3272 }
3273
3274 static void acpi_nfit_scrub(struct work_struct *work)
3275 {
3276         struct acpi_nfit_desc *acpi_desc;
3277         unsigned int tmo;
3278         int query_rc;
3279
3280         acpi_desc = container_of(work, typeof(*acpi_desc), dwork.work);
3281         mutex_lock(&acpi_desc->init_mutex);
3282         query_rc = acpi_nfit_query_poison(acpi_desc);
3283         tmo = __acpi_nfit_scrub(acpi_desc, query_rc);
3284         if (tmo)
3285                 __sched_ars(acpi_desc, tmo);
3286         else
3287                 notify_ars_done(acpi_desc);
3288         memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
3289         clear_bit(ARS_POLL, &acpi_desc->scrub_flags);
3290         mutex_unlock(&acpi_desc->init_mutex);
3291 }
3292
3293 static void acpi_nfit_init_ars(struct acpi_nfit_desc *acpi_desc,
3294                 struct nfit_spa *nfit_spa)
3295 {
3296         int type = nfit_spa_type(nfit_spa->spa);
3297         struct nd_cmd_ars_cap ars_cap;
3298         int rc;
3299
3300         set_bit(ARS_FAILED, &nfit_spa->ars_state);
3301         memset(&ars_cap, 0, sizeof(ars_cap));
3302         rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
3303         if (rc < 0)
3304                 return;
3305         /* check that the supported scrub types match the spa type */
3306         if (type == NFIT_SPA_VOLATILE && ((ars_cap.status >> 16)
3307                                 & ND_ARS_VOLATILE) == 0)
3308                 return;
3309         if (type == NFIT_SPA_PM && ((ars_cap.status >> 16)
3310                                 & ND_ARS_PERSISTENT) == 0)
3311                 return;
3312
3313         nfit_spa->max_ars = ars_cap.max_ars_out;
3314         nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
3315         acpi_desc->max_ars = max(nfit_spa->max_ars, acpi_desc->max_ars);
3316         clear_bit(ARS_FAILED, &nfit_spa->ars_state);
3317 }
3318
3319 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
3320 {
3321         struct nfit_spa *nfit_spa;
3322         int rc, do_sched_ars = 0;
3323
3324         set_bit(ARS_VALID, &acpi_desc->scrub_flags);
3325         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3326                 switch (nfit_spa_type(nfit_spa->spa)) {
3327                 case NFIT_SPA_VOLATILE:
3328                 case NFIT_SPA_PM:
3329                         acpi_nfit_init_ars(acpi_desc, nfit_spa);
3330                         break;
3331                 }
3332         }
3333
3334         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3335                 switch (nfit_spa_type(nfit_spa->spa)) {
3336                 case NFIT_SPA_VOLATILE:
3337                 case NFIT_SPA_PM:
3338                         /* register regions and kick off initial ARS run */
3339                         rc = ars_register(acpi_desc, nfit_spa);
3340                         if (rc)
3341                                 return rc;
3342
3343                         /*
3344                          * Kick off background ARS if at least one
3345                          * region successfully registered ARS
3346                          */
3347                         if (!test_bit(ARS_FAILED, &nfit_spa->ars_state))
3348                                 do_sched_ars++;
3349                         break;
3350                 case NFIT_SPA_BDW:
3351                         /* nothing to register */
3352                         break;
3353                 case NFIT_SPA_DCR:
3354                 case NFIT_SPA_VDISK:
3355                 case NFIT_SPA_VCD:
3356                 case NFIT_SPA_PDISK:
3357                 case NFIT_SPA_PCD:
3358                         /* register known regions that don't support ARS */
3359                         rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
3360                         if (rc)
3361                                 return rc;
3362                         break;
3363                 default:
3364                         /* don't register unknown regions */
3365                         break;
3366                 }
3367         }
3368
3369         if (do_sched_ars)
3370                 sched_ars(acpi_desc);
3371         return 0;
3372 }
3373
3374 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
3375                 struct nfit_table_prev *prev)
3376 {
3377         struct device *dev = acpi_desc->dev;
3378
3379         if (!list_empty(&prev->spas) ||
3380                         !list_empty(&prev->memdevs) ||
3381                         !list_empty(&prev->dcrs) ||
3382                         !list_empty(&prev->bdws) ||
3383                         !list_empty(&prev->idts) ||
3384                         !list_empty(&prev->flushes)) {
3385                 dev_err(dev, "new nfit deletes entries (unsupported)\n");
3386                 return -ENXIO;
3387         }
3388         return 0;
3389 }
3390
3391 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
3392 {
3393         struct device *dev = acpi_desc->dev;
3394         struct kernfs_node *nfit;
3395         struct device *bus_dev;
3396
3397         if (!ars_supported(acpi_desc->nvdimm_bus))
3398                 return 0;
3399
3400         bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3401         nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
3402         if (!nfit) {
3403                 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
3404                 return -ENODEV;
3405         }
3406         acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
3407         sysfs_put(nfit);
3408         if (!acpi_desc->scrub_count_state) {
3409                 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
3410                 return -ENODEV;
3411         }
3412
3413         return 0;
3414 }
3415
3416 static void acpi_nfit_unregister(void *data)
3417 {
3418         struct acpi_nfit_desc *acpi_desc = data;
3419
3420         nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
3421 }
3422
3423 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
3424 {
3425         struct device *dev = acpi_desc->dev;
3426         struct nfit_table_prev prev;
3427         const void *end;
3428         int rc;
3429
3430         if (!acpi_desc->nvdimm_bus) {
3431                 acpi_nfit_init_dsms(acpi_desc);
3432
3433                 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
3434                                 &acpi_desc->nd_desc);
3435                 if (!acpi_desc->nvdimm_bus)
3436                         return -ENOMEM;
3437
3438                 rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
3439                                 acpi_desc);
3440                 if (rc)
3441                         return rc;
3442
3443                 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
3444                 if (rc)
3445                         return rc;
3446
3447                 /* register this acpi_desc for mce notifications */
3448                 mutex_lock(&acpi_desc_lock);
3449                 list_add_tail(&acpi_desc->list, &acpi_descs);
3450                 mutex_unlock(&acpi_desc_lock);
3451         }
3452
3453         mutex_lock(&acpi_desc->init_mutex);
3454
3455         INIT_LIST_HEAD(&prev.spas);
3456         INIT_LIST_HEAD(&prev.memdevs);
3457         INIT_LIST_HEAD(&prev.dcrs);
3458         INIT_LIST_HEAD(&prev.bdws);
3459         INIT_LIST_HEAD(&prev.idts);
3460         INIT_LIST_HEAD(&prev.flushes);
3461
3462         list_cut_position(&prev.spas, &acpi_desc->spas,
3463                                 acpi_desc->spas.prev);
3464         list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
3465                                 acpi_desc->memdevs.prev);
3466         list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
3467                                 acpi_desc->dcrs.prev);
3468         list_cut_position(&prev.bdws, &acpi_desc->bdws,
3469                                 acpi_desc->bdws.prev);
3470         list_cut_position(&prev.idts, &acpi_desc->idts,
3471                                 acpi_desc->idts.prev);
3472         list_cut_position(&prev.flushes, &acpi_desc->flushes,
3473                                 acpi_desc->flushes.prev);
3474
3475         end = data + sz;
3476         while (!IS_ERR_OR_NULL(data))
3477                 data = add_table(acpi_desc, &prev, data, end);
3478
3479         if (IS_ERR(data)) {
3480                 dev_dbg(dev, "nfit table parsing error: %ld\n", PTR_ERR(data));
3481                 rc = PTR_ERR(data);
3482                 goto out_unlock;
3483         }
3484
3485         rc = acpi_nfit_check_deletions(acpi_desc, &prev);
3486         if (rc)
3487                 goto out_unlock;
3488
3489         rc = nfit_mem_init(acpi_desc);
3490         if (rc)
3491                 goto out_unlock;
3492
3493         rc = acpi_nfit_register_dimms(acpi_desc);
3494         if (rc)
3495                 goto out_unlock;
3496
3497         rc = acpi_nfit_register_regions(acpi_desc);
3498
3499  out_unlock:
3500         mutex_unlock(&acpi_desc->init_mutex);
3501         return rc;
3502 }
3503 EXPORT_SYMBOL_GPL(acpi_nfit_init);
3504
3505 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
3506 {
3507         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
3508         struct device *dev = acpi_desc->dev;
3509
3510         /* Bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
3511         nfit_device_lock(dev);
3512         nfit_device_unlock(dev);
3513
3514         /* Bounce the init_mutex to complete initial registration */
3515         mutex_lock(&acpi_desc->init_mutex);
3516         mutex_unlock(&acpi_desc->init_mutex);
3517
3518         return 0;
3519 }
3520
3521 static int __acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3522                 struct nvdimm *nvdimm, unsigned int cmd)
3523 {
3524         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
3525
3526         if (nvdimm)
3527                 return 0;
3528         if (cmd != ND_CMD_ARS_START)
3529                 return 0;
3530
3531         /*
3532          * The kernel and userspace may race to initiate a scrub, but
3533          * the scrub thread is prepared to lose that initial race.  It
3534          * just needs guarantees that any ARS it initiates are not
3535          * interrupted by any intervening start requests from userspace.
3536          */
3537         if (work_busy(&acpi_desc->dwork.work))
3538                 return -EBUSY;
3539
3540         return 0;
3541 }
3542
3543 /*
3544  * Prevent security and firmware activate commands from being issued via
3545  * ioctl.
3546  */
3547 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3548                 struct nvdimm *nvdimm, unsigned int cmd, void *buf)
3549 {
3550         struct nd_cmd_pkg *call_pkg = buf;
3551         unsigned int func;
3552
3553         if (nvdimm && cmd == ND_CMD_CALL &&
3554                         call_pkg->nd_family == NVDIMM_FAMILY_INTEL) {
3555                 func = call_pkg->nd_command;
3556                 if (func > NVDIMM_CMD_MAX ||
3557                     (1 << func) & NVDIMM_INTEL_DENY_CMDMASK)
3558                         return -EOPNOTSUPP;
3559         }
3560
3561         /* block all non-nfit bus commands */
3562         if (!nvdimm && cmd == ND_CMD_CALL &&
3563                         call_pkg->nd_family != NVDIMM_BUS_FAMILY_NFIT)
3564                 return -EOPNOTSUPP;
3565
3566         return __acpi_nfit_clear_to_send(nd_desc, nvdimm, cmd);
3567 }
3568
3569 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc,
3570                 enum nfit_ars_state req_type)
3571 {
3572         struct device *dev = acpi_desc->dev;
3573         int scheduled = 0, busy = 0;
3574         struct nfit_spa *nfit_spa;
3575
3576         mutex_lock(&acpi_desc->init_mutex);
3577         if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags)) {
3578                 mutex_unlock(&acpi_desc->init_mutex);
3579                 return 0;
3580         }
3581
3582         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3583                 int type = nfit_spa_type(nfit_spa->spa);
3584
3585                 if (type != NFIT_SPA_PM && type != NFIT_SPA_VOLATILE)
3586                         continue;
3587                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3588                         continue;
3589
3590                 if (test_and_set_bit(req_type, &nfit_spa->ars_state))
3591                         busy++;
3592                 else
3593                         scheduled++;
3594         }
3595         if (scheduled) {
3596                 sched_ars(acpi_desc);
3597                 dev_dbg(dev, "ars_scan triggered\n");
3598         }
3599         mutex_unlock(&acpi_desc->init_mutex);
3600
3601         if (scheduled)
3602                 return 0;
3603         if (busy)
3604                 return -EBUSY;
3605         return -ENOTTY;
3606 }
3607
3608 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
3609 {
3610         struct nvdimm_bus_descriptor *nd_desc;
3611
3612         dev_set_drvdata(dev, acpi_desc);
3613         acpi_desc->dev = dev;
3614         acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
3615         nd_desc = &acpi_desc->nd_desc;
3616         nd_desc->provider_name = "ACPI.NFIT";
3617         nd_desc->module = THIS_MODULE;
3618         nd_desc->ndctl = acpi_nfit_ctl;
3619         nd_desc->flush_probe = acpi_nfit_flush_probe;
3620         nd_desc->clear_to_send = acpi_nfit_clear_to_send;
3621         nd_desc->attr_groups = acpi_nfit_attribute_groups;
3622
3623         INIT_LIST_HEAD(&acpi_desc->spas);
3624         INIT_LIST_HEAD(&acpi_desc->dcrs);
3625         INIT_LIST_HEAD(&acpi_desc->bdws);
3626         INIT_LIST_HEAD(&acpi_desc->idts);
3627         INIT_LIST_HEAD(&acpi_desc->flushes);
3628         INIT_LIST_HEAD(&acpi_desc->memdevs);
3629         INIT_LIST_HEAD(&acpi_desc->dimms);
3630         INIT_LIST_HEAD(&acpi_desc->list);
3631         mutex_init(&acpi_desc->init_mutex);
3632         acpi_desc->scrub_tmo = 1;
3633         INIT_DELAYED_WORK(&acpi_desc->dwork, acpi_nfit_scrub);
3634 }
3635 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
3636
3637 static void acpi_nfit_put_table(void *table)
3638 {
3639         acpi_put_table(table);
3640 }
3641
3642 void acpi_nfit_shutdown(void *data)
3643 {
3644         struct acpi_nfit_desc *acpi_desc = data;
3645         struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3646
3647         /*
3648          * Destruct under acpi_desc_lock so that nfit_handle_mce does not
3649          * race teardown
3650          */
3651         mutex_lock(&acpi_desc_lock);
3652         list_del(&acpi_desc->list);
3653         mutex_unlock(&acpi_desc_lock);
3654
3655         mutex_lock(&acpi_desc->init_mutex);
3656         set_bit(ARS_CANCEL, &acpi_desc->scrub_flags);
3657         cancel_delayed_work_sync(&acpi_desc->dwork);
3658         mutex_unlock(&acpi_desc->init_mutex);
3659
3660         /*
3661          * Bounce the nvdimm bus lock to make sure any in-flight
3662          * acpi_nfit_ars_rescan() submissions have had a chance to
3663          * either submit or see ->cancel set.
3664          */
3665         nfit_device_lock(bus_dev);
3666         nfit_device_unlock(bus_dev);
3667
3668         flush_workqueue(nfit_wq);
3669 }
3670 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown);
3671
3672 static int acpi_nfit_add(struct acpi_device *adev)
3673 {
3674         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3675         struct acpi_nfit_desc *acpi_desc;
3676         struct device *dev = &adev->dev;
3677         struct acpi_table_header *tbl;
3678         acpi_status status = AE_OK;
3679         acpi_size sz;
3680         int rc = 0;
3681
3682         status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
3683         if (ACPI_FAILURE(status)) {
3684                 /* The NVDIMM root device allows OS to trigger enumeration of
3685                  * NVDIMMs through NFIT at boot time and re-enumeration at
3686                  * root level via the _FIT method during runtime.
3687                  * This is ok to return 0 here, we could have an nvdimm
3688                  * hotplugged later and evaluate _FIT method which returns
3689                  * data in the format of a series of NFIT Structures.
3690                  */
3691                 dev_dbg(dev, "failed to find NFIT at startup\n");
3692                 return 0;
3693         }
3694
3695         rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl);
3696         if (rc)
3697                 return rc;
3698         sz = tbl->length;
3699
3700         acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3701         if (!acpi_desc)
3702                 return -ENOMEM;
3703         acpi_nfit_desc_init(acpi_desc, &adev->dev);
3704
3705         /* Save the acpi header for exporting the revision via sysfs */
3706         acpi_desc->acpi_header = *tbl;
3707
3708         /* Evaluate _FIT and override with that if present */
3709         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
3710         if (ACPI_SUCCESS(status) && buf.length > 0) {
3711                 union acpi_object *obj = buf.pointer;
3712
3713                 if (obj->type == ACPI_TYPE_BUFFER)
3714                         rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3715                                         obj->buffer.length);
3716                 else
3717                         dev_dbg(dev, "invalid type %d, ignoring _FIT\n",
3718                                 (int) obj->type);
3719                 kfree(buf.pointer);
3720         } else
3721                 /* skip over the lead-in header table */
3722                 rc = acpi_nfit_init(acpi_desc, (void *) tbl
3723                                 + sizeof(struct acpi_table_nfit),
3724                                 sz - sizeof(struct acpi_table_nfit));
3725
3726         if (rc)
3727                 return rc;
3728         return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc);
3729 }
3730
3731 static int acpi_nfit_remove(struct acpi_device *adev)
3732 {
3733         /* see acpi_nfit_unregister */
3734         return 0;
3735 }
3736
3737 static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle)
3738 {
3739         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3740         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3741         union acpi_object *obj;
3742         acpi_status status;
3743         int ret;
3744
3745         if (!dev->driver) {
3746                 /* dev->driver may be null if we're being removed */
3747                 dev_dbg(dev, "no driver found for dev\n");
3748                 return;
3749         }
3750
3751         if (!acpi_desc) {
3752                 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3753                 if (!acpi_desc)
3754                         return;
3755                 acpi_nfit_desc_init(acpi_desc, dev);
3756         } else {
3757                 /*
3758                  * Finish previous registration before considering new
3759                  * regions.
3760                  */
3761                 flush_workqueue(nfit_wq);
3762         }
3763
3764         /* Evaluate _FIT */
3765         status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
3766         if (ACPI_FAILURE(status)) {
3767                 dev_err(dev, "failed to evaluate _FIT\n");
3768                 return;
3769         }
3770
3771         obj = buf.pointer;
3772         if (obj->type == ACPI_TYPE_BUFFER) {
3773                 ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3774                                 obj->buffer.length);
3775                 if (ret)
3776                         dev_err(dev, "failed to merge updated NFIT\n");
3777         } else
3778                 dev_err(dev, "Invalid _FIT\n");
3779         kfree(buf.pointer);
3780 }
3781
3782 static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle)
3783 {
3784         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3785
3786         if (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON)
3787                 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
3788         else
3789                 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_SHORT);
3790 }
3791
3792 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
3793 {
3794         dev_dbg(dev, "event: 0x%x\n", event);
3795
3796         switch (event) {
3797         case NFIT_NOTIFY_UPDATE:
3798                 return acpi_nfit_update_notify(dev, handle);
3799         case NFIT_NOTIFY_UC_MEMORY_ERROR:
3800                 return acpi_nfit_uc_error_notify(dev, handle);
3801         default:
3802                 return;
3803         }
3804 }
3805 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
3806
3807 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
3808 {
3809         nfit_device_lock(&adev->dev);
3810         __acpi_nfit_notify(&adev->dev, adev->handle, event);
3811         nfit_device_unlock(&adev->dev);
3812 }
3813
3814 static const struct acpi_device_id acpi_nfit_ids[] = {
3815         { "ACPI0012", 0 },
3816         { "", 0 },
3817 };
3818 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
3819
3820 static struct acpi_driver acpi_nfit_driver = {
3821         .name = KBUILD_MODNAME,
3822         .ids = acpi_nfit_ids,
3823         .ops = {
3824                 .add = acpi_nfit_add,
3825                 .remove = acpi_nfit_remove,
3826                 .notify = acpi_nfit_notify,
3827         },
3828 };
3829
3830 static __init int nfit_init(void)
3831 {
3832         int ret;
3833
3834         BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
3835         BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 64);
3836         BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
3837         BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
3838         BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
3839         BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
3840         BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
3841         BUILD_BUG_ON(sizeof(struct acpi_nfit_capabilities) != 16);
3842
3843         guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]);
3844         guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]);
3845         guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]);
3846         guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]);
3847         guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]);
3848         guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]);
3849         guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]);
3850         guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]);
3851         guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]);
3852         guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]);
3853         guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
3854         guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
3855         guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
3856         guid_parse(UUID_NFIT_DIMM_N_HYPERV, &nfit_uuid[NFIT_DEV_DIMM_N_HYPERV]);
3857         guid_parse(UUID_INTEL_BUS, &nfit_uuid[NFIT_BUS_INTEL]);
3858
3859         nfit_wq = create_singlethread_workqueue("nfit");
3860         if (!nfit_wq)
3861                 return -ENOMEM;
3862
3863         nfit_mce_register();
3864         ret = acpi_bus_register_driver(&acpi_nfit_driver);
3865         if (ret) {
3866                 nfit_mce_unregister();
3867                 destroy_workqueue(nfit_wq);
3868         }
3869
3870         return ret;
3871
3872 }
3873
3874 static __exit void nfit_exit(void)
3875 {
3876         nfit_mce_unregister();
3877         acpi_bus_unregister_driver(&acpi_nfit_driver);
3878         destroy_workqueue(nfit_wq);
3879         WARN_ON(!list_empty(&acpi_descs));
3880 }
3881
3882 module_init(nfit_init);
3883 module_exit(nfit_exit);
3884 MODULE_LICENSE("GPL v2");
3885 MODULE_AUTHOR("Intel Corporation");