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         guid_t guid;
 682         int i;
 683
 684         import_guid(&guid, spa->range_guid);
 685         for (i = 0; i < NFIT_UUID_MAX; i++)
 686                 if (guid_equal(to_nfit_uuid(i), &guid))
 687                         return i;
 688         return -1;
 689 }
 690
 691 static size_t sizeof_spa(struct acpi_nfit_system_address *spa)
 692 {
 693         if (spa->flags & ACPI_NFIT_LOCATION_COOKIE_VALID)
 694                 return sizeof(*spa);
 695         return sizeof(*spa) - 8;
 696 }
 697
 698 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
 699                 struct nfit_table_prev *prev,
 700                 struct acpi_nfit_system_address *spa)
 701 {
 702         struct device *dev = acpi_desc->dev;
 703         struct nfit_spa *nfit_spa;
 704
 705         if (spa->header.length != sizeof_spa(spa))
 706                 return false;
 707
 708         list_for_each_entry(nfit_spa, &prev->spas, list) {
 709                 if (memcmp(nfit_spa->spa, spa, sizeof_spa(spa)) == 0) {
 710                         list_move_tail(&nfit_spa->list, &acpi_desc->spas);
 711                         return true;
 712                 }
 713         }
 714
 715         nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof_spa(spa),
 716                         GFP_KERNEL);
 717         if (!nfit_spa)
 718                 return false;
 719         INIT_LIST_HEAD(&nfit_spa->list);
 720         memcpy(nfit_spa->spa, spa, sizeof_spa(spa));
 721         list_add_tail(&nfit_spa->list, &acpi_desc->spas);
 722         dev_dbg(dev, "spa index: %d type: %s\n",
 723                         spa->range_index,
 724                         spa_type_name(nfit_spa_type(spa)));
 725         return true;
 726 }
 727
 728 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
 729                 struct nfit_table_prev *prev,
 730                 struct acpi_nfit_memory_map *memdev)
 731 {
 732         struct device *dev = acpi_desc->dev;
 733         struct nfit_memdev *nfit_memdev;
 734
 735         if (memdev->header.length != sizeof(*memdev))
 736                 return false;
 737
 738         list_for_each_entry(nfit_memdev, &prev->memdevs, list)
 739                 if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
 740                         list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 741                         return true;
 742                 }
 743
 744         nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
 745                         GFP_KERNEL);
 746         if (!nfit_memdev)
 747                 return false;
 748         INIT_LIST_HEAD(&nfit_memdev->list);
 749         memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
 750         list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 751         dev_dbg(dev, "memdev handle: %#x spa: %d dcr: %d flags: %#x\n",
 752                         memdev->device_handle, memdev->range_index,
 753                         memdev->region_index, memdev->flags);
 754         return true;
 755 }
 756
 757 int nfit_get_smbios_id(u32 device_handle, u16 *flags)
 758 {
 759         struct acpi_nfit_memory_map *memdev;
 760         struct acpi_nfit_desc *acpi_desc;
 761         struct nfit_mem *nfit_mem;
 762         u16 physical_id;
 763
 764         mutex_lock(&acpi_desc_lock);
 765         list_for_each_entry(acpi_desc, &acpi_descs, list) {
 766                 mutex_lock(&acpi_desc->init_mutex);
 767                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
 768                         memdev = __to_nfit_memdev(nfit_mem);
 769                         if (memdev->device_handle == device_handle) {
 770                                 *flags = memdev->flags;
 771                                 physical_id = memdev->physical_id;
 772                                 mutex_unlock(&acpi_desc->init_mutex);
 773                                 mutex_unlock(&acpi_desc_lock);
 774                                 return physical_id;
 775                         }
 776                 }
 777                 mutex_unlock(&acpi_desc->init_mutex);
 778         }
 779         mutex_unlock(&acpi_desc_lock);
 780
 781         return -ENODEV;
 782 }
 783 EXPORT_SYMBOL_GPL(nfit_get_smbios_id);
 784
 785 /*
 786  * An implementation may provide a truncated control region if no block windows
 787  * are defined.
 788  */
 789 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
 790 {
 791         if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
 792                                 window_size))
 793                 return 0;
 794         if (dcr->windows)
 795                 return sizeof(*dcr);
 796         return offsetof(struct acpi_nfit_control_region, window_size);
 797 }
 798
 799 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
 800                 struct nfit_table_prev *prev,
 801                 struct acpi_nfit_control_region *dcr)
 802 {
 803         struct device *dev = acpi_desc->dev;
 804         struct nfit_dcr *nfit_dcr;
 805
 806         if (!sizeof_dcr(dcr))
 807                 return false;
 808
 809         list_for_each_entry(nfit_dcr, &prev->dcrs, list)
 810                 if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
 811                         list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 812                         return true;
 813                 }
 814
 815         nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
 816                         GFP_KERNEL);
 817         if (!nfit_dcr)
 818                 return false;
 819         INIT_LIST_HEAD(&nfit_dcr->list);
 820         memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
 821         list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 822         dev_dbg(dev, "dcr index: %d windows: %d\n",
 823                         dcr->region_index, dcr->windows);
 824         return true;
 825 }
 826
 827 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
 828                 struct nfit_table_prev *prev,
 829                 struct acpi_nfit_data_region *bdw)
 830 {
 831         struct device *dev = acpi_desc->dev;
 832         struct nfit_bdw *nfit_bdw;
 833
 834         if (bdw->header.length != sizeof(*bdw))
 835                 return false;
 836         list_for_each_entry(nfit_bdw, &prev->bdws, list)
 837                 if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
 838                         list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
 839                         return true;
 840                 }
 841
 842         nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
 843                         GFP_KERNEL);
 844         if (!nfit_bdw)
 845                 return false;
 846         INIT_LIST_HEAD(&nfit_bdw->list);
 847         memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
 848         list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
 849         dev_dbg(dev, "bdw dcr: %d windows: %d\n",
 850                         bdw->region_index, bdw->windows);
 851         return true;
 852 }
 853
 854 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
 855 {
 856         if (idt->header.length < sizeof(*idt))
 857                 return 0;
 858         return sizeof(*idt) + sizeof(u32) * (idt->line_count - 1);
 859 }
 860
 861 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
 862                 struct nfit_table_prev *prev,
 863                 struct acpi_nfit_interleave *idt)
 864 {
 865         struct device *dev = acpi_desc->dev;
 866         struct nfit_idt *nfit_idt;
 867
 868         if (!sizeof_idt(idt))
 869                 return false;
 870
 871         list_for_each_entry(nfit_idt, &prev->idts, list) {
 872                 if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
 873                         continue;
 874
 875                 if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
 876                         list_move_tail(&nfit_idt->list, &acpi_desc->idts);
 877                         return true;
 878                 }
 879         }
 880
 881         nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
 882                         GFP_KERNEL);
 883         if (!nfit_idt)
 884                 return false;
 885         INIT_LIST_HEAD(&nfit_idt->list);
 886         memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
 887         list_add_tail(&nfit_idt->list, &acpi_desc->idts);
 888         dev_dbg(dev, "idt index: %d num_lines: %d\n",
 889                         idt->interleave_index, idt->line_count);
 890         return true;
 891 }
 892
 893 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
 894 {
 895         if (flush->header.length < sizeof(*flush))
 896                 return 0;
 897         return sizeof(*flush) + sizeof(u64) * (flush->hint_count - 1);
 898 }
 899
 900 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
 901                 struct nfit_table_prev *prev,
 902                 struct acpi_nfit_flush_address *flush)
 903 {
 904         struct device *dev = acpi_desc->dev;
 905         struct nfit_flush *nfit_flush;
 906
 907         if (!sizeof_flush(flush))
 908                 return false;
 909
 910         list_for_each_entry(nfit_flush, &prev->flushes, list) {
 911                 if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
 912                         continue;
 913
 914                 if (memcmp(nfit_flush->flush, flush,
 915                                         sizeof_flush(flush)) == 0) {
 916                         list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
 917                         return true;
 918                 }
 919         }
 920
 921         nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
 922                         + sizeof_flush(flush), GFP_KERNEL);
 923         if (!nfit_flush)
 924                 return false;
 925         INIT_LIST_HEAD(&nfit_flush->list);
 926         memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
 927         list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
 928         dev_dbg(dev, "nfit_flush handle: %d hint_count: %d\n",
 929                         flush->device_handle, flush->hint_count);
 930         return true;
 931 }
 932
 933 static bool add_platform_cap(struct acpi_nfit_desc *acpi_desc,
 934                 struct acpi_nfit_capabilities *pcap)
 935 {
 936         struct device *dev = acpi_desc->dev;
 937         u32 mask;
 938
 939         mask = (1 << (pcap->highest_capability + 1)) - 1;
 940         acpi_desc->platform_cap = pcap->capabilities & mask;
 941         dev_dbg(dev, "cap: %#x\n", acpi_desc->platform_cap);
 942         return true;
 943 }
 944
 945 static void *add_table(struct acpi_nfit_desc *acpi_desc,
 946                 struct nfit_table_prev *prev, void *table, const void *end)
 947 {
 948         struct device *dev = acpi_desc->dev;
 949         struct acpi_nfit_header *hdr;
 950         void *err = ERR_PTR(-ENOMEM);
 951
 952         if (table >= end)
 953                 return NULL;
 954
 955         hdr = table;
 956         if (!hdr->length) {
 957                 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
 958                         hdr->type);
 959                 return NULL;
 960         }
 961
 962         switch (hdr->type) {
 963         case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
 964                 if (!add_spa(acpi_desc, prev, table))
 965                         return err;
 966                 break;
 967         case ACPI_NFIT_TYPE_MEMORY_MAP:
 968                 if (!add_memdev(acpi_desc, prev, table))
 969                         return err;
 970                 break;
 971         case ACPI_NFIT_TYPE_CONTROL_REGION:
 972                 if (!add_dcr(acpi_desc, prev, table))
 973                         return err;
 974                 break;
 975         case ACPI_NFIT_TYPE_DATA_REGION:
 976                 if (!add_bdw(acpi_desc, prev, table))
 977                         return err;
 978                 break;
 979         case ACPI_NFIT_TYPE_INTERLEAVE:
 980                 if (!add_idt(acpi_desc, prev, table))
 981                         return err;
 982                 break;
 983         case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
 984                 if (!add_flush(acpi_desc, prev, table))
 985                         return err;
 986                 break;
 987         case ACPI_NFIT_TYPE_SMBIOS:
 988                 dev_dbg(dev, "smbios\n");
 989                 break;
 990         case ACPI_NFIT_TYPE_CAPABILITIES:
 991                 if (!add_platform_cap(acpi_desc, table))
 992                         return err;
 993                 break;
 994         default:
 995                 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
 996                 break;
 997         }
 998
 999         return table + hdr->length;
1000 }
1001
1002 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
1003                 struct nfit_mem *nfit_mem)
1004 {
1005         u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1006         u16 dcr = nfit_mem->dcr->region_index;
1007         struct nfit_spa *nfit_spa;
1008
1009         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
1010                 u16 range_index = nfit_spa->spa->range_index;
1011                 int type = nfit_spa_type(nfit_spa->spa);
1012                 struct nfit_memdev *nfit_memdev;
1013
1014                 if (type != NFIT_SPA_BDW)
1015                         continue;
1016
1017                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1018                         if (nfit_memdev->memdev->range_index != range_index)
1019                                 continue;
1020                         if (nfit_memdev->memdev->device_handle != device_handle)
1021                                 continue;
1022                         if (nfit_memdev->memdev->region_index != dcr)
1023                                 continue;
1024
1025                         nfit_mem->spa_bdw = nfit_spa->spa;
1026                         return;
1027                 }
1028         }
1029
1030         dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
1031                         nfit_mem->spa_dcr->range_index);
1032         nfit_mem->bdw = NULL;
1033 }
1034
1035 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
1036                 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
1037 {
1038         u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
1039         struct nfit_memdev *nfit_memdev;
1040         struct nfit_bdw *nfit_bdw;
1041         struct nfit_idt *nfit_idt;
1042         u16 idt_idx, range_index;
1043
1044         list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
1045                 if (nfit_bdw->bdw->region_index != dcr)
1046                         continue;
1047                 nfit_mem->bdw = nfit_bdw->bdw;
1048                 break;
1049         }
1050
1051         if (!nfit_mem->bdw)
1052                 return;
1053
1054         nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
1055
1056         if (!nfit_mem->spa_bdw)
1057                 return;
1058
1059         range_index = nfit_mem->spa_bdw->range_index;
1060         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1061                 if (nfit_memdev->memdev->range_index != range_index ||
1062                                 nfit_memdev->memdev->region_index != dcr)
1063                         continue;
1064                 nfit_mem->memdev_bdw = nfit_memdev->memdev;
1065                 idt_idx = nfit_memdev->memdev->interleave_index;
1066                 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1067                         if (nfit_idt->idt->interleave_index != idt_idx)
1068                                 continue;
1069                         nfit_mem->idt_bdw = nfit_idt->idt;
1070                         break;
1071                 }
1072                 break;
1073         }
1074 }
1075
1076 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
1077                 struct acpi_nfit_system_address *spa)
1078 {
1079         struct nfit_mem *nfit_mem, *found;
1080         struct nfit_memdev *nfit_memdev;
1081         int type = spa ? nfit_spa_type(spa) : 0;
1082
1083         switch (type) {
1084         case NFIT_SPA_DCR:
1085         case NFIT_SPA_PM:
1086                 break;
1087         default:
1088                 if (spa)
1089                         return 0;
1090         }
1091
1092         /*
1093          * This loop runs in two modes, when a dimm is mapped the loop
1094          * adds memdev associations to an existing dimm, or creates a
1095          * dimm. In the unmapped dimm case this loop sweeps for memdev
1096          * instances with an invalid / zero range_index and adds those
1097          * dimms without spa associations.
1098          */
1099         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1100                 struct nfit_flush *nfit_flush;
1101                 struct nfit_dcr *nfit_dcr;
1102                 u32 device_handle;
1103                 u16 dcr;
1104
1105                 if (spa && nfit_memdev->memdev->range_index != spa->range_index)
1106                         continue;
1107                 if (!spa && nfit_memdev->memdev->range_index)
1108                         continue;
1109                 found = NULL;
1110                 dcr = nfit_memdev->memdev->region_index;
1111                 device_handle = nfit_memdev->memdev->device_handle;
1112                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1113                         if (__to_nfit_memdev(nfit_mem)->device_handle
1114                                         == device_handle) {
1115                                 found = nfit_mem;
1116                                 break;
1117                         }
1118
1119                 if (found)
1120                         nfit_mem = found;
1121                 else {
1122                         nfit_mem = devm_kzalloc(acpi_desc->dev,
1123                                         sizeof(*nfit_mem), GFP_KERNEL);
1124                         if (!nfit_mem)
1125                                 return -ENOMEM;
1126                         INIT_LIST_HEAD(&nfit_mem->list);
1127                         nfit_mem->acpi_desc = acpi_desc;
1128                         list_add(&nfit_mem->list, &acpi_desc->dimms);
1129                 }
1130
1131                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1132                         if (nfit_dcr->dcr->region_index != dcr)
1133                                 continue;
1134                         /*
1135                          * Record the control region for the dimm.  For
1136                          * the ACPI 6.1 case, where there are separate
1137                          * control regions for the pmem vs blk
1138                          * interfaces, be sure to record the extended
1139                          * blk details.
1140                          */
1141                         if (!nfit_mem->dcr)
1142                                 nfit_mem->dcr = nfit_dcr->dcr;
1143                         else if (nfit_mem->dcr->windows == 0
1144                                         && nfit_dcr->dcr->windows)
1145                                 nfit_mem->dcr = nfit_dcr->dcr;
1146                         break;
1147                 }
1148
1149                 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
1150                         struct acpi_nfit_flush_address *flush;
1151                         u16 i;
1152
1153                         if (nfit_flush->flush->device_handle != device_handle)
1154                                 continue;
1155                         nfit_mem->nfit_flush = nfit_flush;
1156                         flush = nfit_flush->flush;
1157                         nfit_mem->flush_wpq = devm_kcalloc(acpi_desc->dev,
1158                                         flush->hint_count,
1159                                         sizeof(struct resource),
1160                                         GFP_KERNEL);
1161                         if (!nfit_mem->flush_wpq)
1162                                 return -ENOMEM;
1163                         for (i = 0; i < flush->hint_count; i++) {
1164                                 struct resource *res = &nfit_mem->flush_wpq[i];
1165
1166                                 res->start = flush->hint_address[i];
1167                                 res->end = res->start + 8 - 1;
1168                         }
1169                         break;
1170                 }
1171
1172                 if (dcr && !nfit_mem->dcr) {
1173                         dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
1174                                         spa->range_index, dcr);
1175                         return -ENODEV;
1176                 }
1177
1178                 if (type == NFIT_SPA_DCR) {
1179                         struct nfit_idt *nfit_idt;
1180                         u16 idt_idx;
1181
1182                         /* multiple dimms may share a SPA when interleaved */
1183                         nfit_mem->spa_dcr = spa;
1184                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
1185                         idt_idx = nfit_memdev->memdev->interleave_index;
1186                         list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1187                                 if (nfit_idt->idt->interleave_index != idt_idx)
1188                                         continue;
1189                                 nfit_mem->idt_dcr = nfit_idt->idt;
1190                                 break;
1191                         }
1192                         nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
1193                 } else if (type == NFIT_SPA_PM) {
1194                         /*
1195                          * A single dimm may belong to multiple SPA-PM
1196                          * ranges, record at least one in addition to
1197                          * any SPA-DCR range.
1198                          */
1199                         nfit_mem->memdev_pmem = nfit_memdev->memdev;
1200                 } else
1201                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
1202         }
1203
1204         return 0;
1205 }
1206
1207 static int nfit_mem_cmp(void *priv, const struct list_head *_a,
1208                 const struct list_head *_b)
1209 {
1210         struct nfit_mem *a = container_of(_a, typeof(*a), list);
1211         struct nfit_mem *b = container_of(_b, typeof(*b), list);
1212         u32 handleA, handleB;
1213
1214         handleA = __to_nfit_memdev(a)->device_handle;
1215         handleB = __to_nfit_memdev(b)->device_handle;
1216         if (handleA < handleB)
1217                 return -1;
1218         else if (handleA > handleB)
1219                 return 1;
1220         return 0;
1221 }
1222
1223 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
1224 {
1225         struct nfit_spa *nfit_spa;
1226         int rc;
1227
1228
1229         /*
1230          * For each SPA-DCR or SPA-PMEM address range find its
1231          * corresponding MEMDEV(s).  From each MEMDEV find the
1232          * corresponding DCR.  Then, if we're operating on a SPA-DCR,
1233          * try to find a SPA-BDW and a corresponding BDW that references
1234          * the DCR.  Throw it all into an nfit_mem object.  Note, that
1235          * BDWs are optional.
1236          */
1237         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
1238                 rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
1239                 if (rc)
1240                         return rc;
1241         }
1242
1243         /*
1244          * If a DIMM has failed to be mapped into SPA there will be no
1245          * SPA entries above. Find and register all the unmapped DIMMs
1246          * for reporting and recovery purposes.
1247          */
1248         rc = __nfit_mem_init(acpi_desc, NULL);
1249         if (rc)
1250                 return rc;
1251
1252         list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
1253
1254         return 0;
1255 }
1256
1257 static ssize_t bus_dsm_mask_show(struct device *dev,
1258                 struct device_attribute *attr, char *buf)
1259 {
1260         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1261         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1262         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1263
1264         return sprintf(buf, "%#lx\n", acpi_desc->bus_dsm_mask);
1265 }
1266 static struct device_attribute dev_attr_bus_dsm_mask =
1267                 __ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
1268
1269 static ssize_t revision_show(struct device *dev,
1270                 struct device_attribute *attr, char *buf)
1271 {
1272         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1273         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1274         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1275
1276         return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
1277 }
1278 static DEVICE_ATTR_RO(revision);
1279
1280 static ssize_t hw_error_scrub_show(struct device *dev,
1281                 struct device_attribute *attr, char *buf)
1282 {
1283         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1284         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1285         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1286
1287         return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
1288 }
1289
1290 /*
1291  * The 'hw_error_scrub' attribute can have the following values written to it:
1292  * '0': Switch to the default mode where an exception will only insert
1293  *      the address of the memory error into the poison and badblocks lists.
1294  * '1': Enable a full scrub to happen if an exception for a memory error is
1295  *      received.
1296  */
1297 static ssize_t hw_error_scrub_store(struct device *dev,
1298                 struct device_attribute *attr, const char *buf, size_t size)
1299 {
1300         struct nvdimm_bus_descriptor *nd_desc;
1301         ssize_t rc;
1302         long val;
1303
1304         rc = kstrtol(buf, 0, &val);
1305         if (rc)
1306                 return rc;
1307
1308         nfit_device_lock(dev);
1309         nd_desc = dev_get_drvdata(dev);
1310         if (nd_desc) {
1311                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1312
1313                 switch (val) {
1314                 case HW_ERROR_SCRUB_ON:
1315                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
1316                         break;
1317                 case HW_ERROR_SCRUB_OFF:
1318                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
1319                         break;
1320                 default:
1321                         rc = -EINVAL;
1322                         break;
1323                 }
1324         }
1325         nfit_device_unlock(dev);
1326         if (rc)
1327                 return rc;
1328         return size;
1329 }
1330 static DEVICE_ATTR_RW(hw_error_scrub);
1331
1332 /*
1333  * This shows the number of full Address Range Scrubs that have been
1334  * completed since driver load time. Userspace can wait on this using
1335  * select/poll etc. A '+' at the end indicates an ARS is in progress
1336  */
1337 static ssize_t scrub_show(struct device *dev,
1338                 struct device_attribute *attr, char *buf)
1339 {
1340         struct nvdimm_bus_descriptor *nd_desc;
1341         struct acpi_nfit_desc *acpi_desc;
1342         ssize_t rc = -ENXIO;
1343         bool busy;
1344
1345         nfit_device_lock(dev);
1346         nd_desc = dev_get_drvdata(dev);
1347         if (!nd_desc) {
1348                 nfit_device_unlock(dev);
1349                 return rc;
1350         }
1351         acpi_desc = to_acpi_desc(nd_desc);
1352
1353         mutex_lock(&acpi_desc->init_mutex);
1354         busy = test_bit(ARS_BUSY, &acpi_desc->scrub_flags)
1355                 && !test_bit(ARS_CANCEL, &acpi_desc->scrub_flags);
1356         rc = sprintf(buf, "%d%s", acpi_desc->scrub_count, busy ? "+\n" : "\n");
1357         /* Allow an admin to poll the busy state at a higher rate */
1358         if (busy && capable(CAP_SYS_RAWIO) && !test_and_set_bit(ARS_POLL,
1359                                 &acpi_desc->scrub_flags)) {
1360                 acpi_desc->scrub_tmo = 1;
1361                 mod_delayed_work(nfit_wq, &acpi_desc->dwork, HZ);
1362         }
1363
1364         mutex_unlock(&acpi_desc->init_mutex);
1365         nfit_device_unlock(dev);
1366         return rc;
1367 }
1368
1369 static ssize_t scrub_store(struct device *dev,
1370                 struct device_attribute *attr, const char *buf, size_t size)
1371 {
1372         struct nvdimm_bus_descriptor *nd_desc;
1373         ssize_t rc;
1374         long val;
1375
1376         rc = kstrtol(buf, 0, &val);
1377         if (rc)
1378                 return rc;
1379         if (val != 1)
1380                 return -EINVAL;
1381
1382         nfit_device_lock(dev);
1383         nd_desc = dev_get_drvdata(dev);
1384         if (nd_desc) {
1385                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1386
1387                 rc = acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
1388         }
1389         nfit_device_unlock(dev);
1390         if (rc)
1391                 return rc;
1392         return size;
1393 }
1394 static DEVICE_ATTR_RW(scrub);
1395
1396 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1397 {
1398         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1399         const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1400                 | 1 << ND_CMD_ARS_STATUS;
1401
1402         return (nd_desc->cmd_mask & mask) == mask;
1403 }
1404
1405 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1406 {
1407         struct device *dev = kobj_to_dev(kobj);
1408         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1409
1410         if (a == &dev_attr_scrub.attr)
1411                 return ars_supported(nvdimm_bus) ? a->mode : 0;
1412
1413         if (a == &dev_attr_firmware_activate_noidle.attr)
1414                 return intel_fwa_supported(nvdimm_bus) ? a->mode : 0;
1415
1416         return a->mode;
1417 }
1418
1419 static struct attribute *acpi_nfit_attributes[] = {
1420         &dev_attr_revision.attr,
1421         &dev_attr_scrub.attr,
1422         &dev_attr_hw_error_scrub.attr,
1423         &dev_attr_bus_dsm_mask.attr,
1424         &dev_attr_firmware_activate_noidle.attr,
1425         NULL,
1426 };
1427
1428 static const struct attribute_group acpi_nfit_attribute_group = {
1429         .name = "nfit",
1430         .attrs = acpi_nfit_attributes,
1431         .is_visible = nfit_visible,
1432 };
1433
1434 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1435         &acpi_nfit_attribute_group,
1436         NULL,
1437 };
1438
1439 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1440 {
1441         struct nvdimm *nvdimm = to_nvdimm(dev);
1442         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1443
1444         return __to_nfit_memdev(nfit_mem);
1445 }
1446
1447 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1448 {
1449         struct nvdimm *nvdimm = to_nvdimm(dev);
1450         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1451
1452         return nfit_mem->dcr;
1453 }
1454
1455 static ssize_t handle_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->device_handle);
1461 }
1462 static DEVICE_ATTR_RO(handle);
1463
1464 static ssize_t phys_id_show(struct device *dev,
1465                 struct device_attribute *attr, char *buf)
1466 {
1467         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1468
1469         return sprintf(buf, "%#x\n", memdev->physical_id);
1470 }
1471 static DEVICE_ATTR_RO(phys_id);
1472
1473 static ssize_t vendor_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->vendor_id));
1479 }
1480 static DEVICE_ATTR_RO(vendor);
1481
1482 static ssize_t rev_id_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->revision_id));
1488 }
1489 static DEVICE_ATTR_RO(rev_id);
1490
1491 static ssize_t device_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->device_id));
1497 }
1498 static DEVICE_ATTR_RO(device);
1499
1500 static ssize_t subsystem_vendor_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", be16_to_cpu(dcr->subsystem_vendor_id));
1506 }
1507 static DEVICE_ATTR_RO(subsystem_vendor);
1508
1509 static ssize_t subsystem_rev_id_show(struct device *dev,
1510                 struct device_attribute *attr, char *buf)
1511 {
1512         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1513
1514         return sprintf(buf, "0x%04x\n",
1515                         be16_to_cpu(dcr->subsystem_revision_id));
1516 }
1517 static DEVICE_ATTR_RO(subsystem_rev_id);
1518
1519 static ssize_t subsystem_device_show(struct device *dev,
1520                 struct device_attribute *attr, char *buf)
1521 {
1522         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1523
1524         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1525 }
1526 static DEVICE_ATTR_RO(subsystem_device);
1527
1528 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1529 {
1530         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1531         int formats = 0;
1532
1533         if (nfit_mem->memdev_pmem)
1534                 formats++;
1535         if (nfit_mem->memdev_bdw)
1536                 formats++;
1537         return formats;
1538 }
1539
1540 static ssize_t format_show(struct device *dev,
1541                 struct device_attribute *attr, char *buf)
1542 {
1543         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1544
1545         return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1546 }
1547 static DEVICE_ATTR_RO(format);
1548
1549 static ssize_t format1_show(struct device *dev,
1550                 struct device_attribute *attr, char *buf)
1551 {
1552         u32 handle;
1553         ssize_t rc = -ENXIO;
1554         struct nfit_mem *nfit_mem;
1555         struct nfit_memdev *nfit_memdev;
1556         struct acpi_nfit_desc *acpi_desc;
1557         struct nvdimm *nvdimm = to_nvdimm(dev);
1558         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1559
1560         nfit_mem = nvdimm_provider_data(nvdimm);
1561         acpi_desc = nfit_mem->acpi_desc;
1562         handle = to_nfit_memdev(dev)->device_handle;
1563
1564         /* assumes DIMMs have at most 2 published interface codes */
1565         mutex_lock(&acpi_desc->init_mutex);
1566         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1567                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1568                 struct nfit_dcr *nfit_dcr;
1569
1570                 if (memdev->device_handle != handle)
1571                         continue;
1572
1573                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1574                         if (nfit_dcr->dcr->region_index != memdev->region_index)
1575                                 continue;
1576                         if (nfit_dcr->dcr->code == dcr->code)
1577                                 continue;
1578                         rc = sprintf(buf, "0x%04x\n",
1579                                         le16_to_cpu(nfit_dcr->dcr->code));
1580                         break;
1581                 }
1582                 if (rc != -ENXIO)
1583                         break;
1584         }
1585         mutex_unlock(&acpi_desc->init_mutex);
1586         return rc;
1587 }
1588 static DEVICE_ATTR_RO(format1);
1589
1590 static ssize_t formats_show(struct device *dev,
1591                 struct device_attribute *attr, char *buf)
1592 {
1593         struct nvdimm *nvdimm = to_nvdimm(dev);
1594
1595         return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1596 }
1597 static DEVICE_ATTR_RO(formats);
1598
1599 static ssize_t serial_show(struct device *dev,
1600                 struct device_attribute *attr, char *buf)
1601 {
1602         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1603
1604         return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1605 }
1606 static DEVICE_ATTR_RO(serial);
1607
1608 static ssize_t family_show(struct device *dev,
1609                 struct device_attribute *attr, char *buf)
1610 {
1611         struct nvdimm *nvdimm = to_nvdimm(dev);
1612         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1613
1614         if (nfit_mem->family < 0)
1615                 return -ENXIO;
1616         return sprintf(buf, "%d\n", nfit_mem->family);
1617 }
1618 static DEVICE_ATTR_RO(family);
1619
1620 static ssize_t dsm_mask_show(struct device *dev,
1621                 struct device_attribute *attr, char *buf)
1622 {
1623         struct nvdimm *nvdimm = to_nvdimm(dev);
1624         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1625
1626         if (nfit_mem->family < 0)
1627                 return -ENXIO;
1628         return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1629 }
1630 static DEVICE_ATTR_RO(dsm_mask);
1631
1632 static ssize_t flags_show(struct device *dev,
1633                 struct device_attribute *attr, char *buf)
1634 {
1635         struct nvdimm *nvdimm = to_nvdimm(dev);
1636         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1637         u16 flags = __to_nfit_memdev(nfit_mem)->flags;
1638
1639         if (test_bit(NFIT_MEM_DIRTY, &nfit_mem->flags))
1640                 flags |= ACPI_NFIT_MEM_FLUSH_FAILED;
1641
1642         return sprintf(buf, "%s%s%s%s%s%s%s\n",
1643                 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1644                 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1645                 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1646                 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1647                 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1648                 flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1649                 flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1650 }
1651 static DEVICE_ATTR_RO(flags);
1652
1653 static ssize_t id_show(struct device *dev,
1654                 struct device_attribute *attr, char *buf)
1655 {
1656         struct nvdimm *nvdimm = to_nvdimm(dev);
1657         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1658
1659         return sprintf(buf, "%s\n", nfit_mem->id);
1660 }
1661 static DEVICE_ATTR_RO(id);
1662
1663 static ssize_t dirty_shutdown_show(struct device *dev,
1664                 struct device_attribute *attr, char *buf)
1665 {
1666         struct nvdimm *nvdimm = to_nvdimm(dev);
1667         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1668
1669         return sprintf(buf, "%d\n", nfit_mem->dirty_shutdown);
1670 }
1671 static DEVICE_ATTR_RO(dirty_shutdown);
1672
1673 static struct attribute *acpi_nfit_dimm_attributes[] = {
1674         &dev_attr_handle.attr,
1675         &dev_attr_phys_id.attr,
1676         &dev_attr_vendor.attr,
1677         &dev_attr_device.attr,
1678         &dev_attr_rev_id.attr,
1679         &dev_attr_subsystem_vendor.attr,
1680         &dev_attr_subsystem_device.attr,
1681         &dev_attr_subsystem_rev_id.attr,
1682         &dev_attr_format.attr,
1683         &dev_attr_formats.attr,
1684         &dev_attr_format1.attr,
1685         &dev_attr_serial.attr,
1686         &dev_attr_flags.attr,
1687         &dev_attr_id.attr,
1688         &dev_attr_family.attr,
1689         &dev_attr_dsm_mask.attr,
1690         &dev_attr_dirty_shutdown.attr,
1691         NULL,
1692 };
1693
1694 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1695                 struct attribute *a, int n)
1696 {
1697         struct device *dev = kobj_to_dev(kobj);
1698         struct nvdimm *nvdimm = to_nvdimm(dev);
1699         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1700
1701         if (!to_nfit_dcr(dev)) {
1702                 /* Without a dcr only the memdev attributes can be surfaced */
1703                 if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1704                                 || a == &dev_attr_flags.attr
1705                                 || a == &dev_attr_family.attr
1706                                 || a == &dev_attr_dsm_mask.attr)
1707                         return a->mode;
1708                 return 0;
1709         }
1710
1711         if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1712                 return 0;
1713
1714         if (!test_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags)
1715                         && a == &dev_attr_dirty_shutdown.attr)
1716                 return 0;
1717
1718         return a->mode;
1719 }
1720
1721 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1722         .name = "nfit",
1723         .attrs = acpi_nfit_dimm_attributes,
1724         .is_visible = acpi_nfit_dimm_attr_visible,
1725 };
1726
1727 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1728         &acpi_nfit_dimm_attribute_group,
1729         NULL,
1730 };
1731
1732 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1733                 u32 device_handle)
1734 {
1735         struct nfit_mem *nfit_mem;
1736
1737         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1738                 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1739                         return nfit_mem->nvdimm;
1740
1741         return NULL;
1742 }
1743
1744 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1745 {
1746         struct nfit_mem *nfit_mem;
1747         struct acpi_nfit_desc *acpi_desc;
1748
1749         dev_dbg(dev->parent, "%s: event: %d\n", dev_name(dev),
1750                         event);
1751
1752         if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1753                 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1754                                 event);
1755                 return;
1756         }
1757
1758         acpi_desc = dev_get_drvdata(dev->parent);
1759         if (!acpi_desc)
1760                 return;
1761
1762         /*
1763          * If we successfully retrieved acpi_desc, then we know nfit_mem data
1764          * is still valid.
1765          */
1766         nfit_mem = dev_get_drvdata(dev);
1767         if (nfit_mem && nfit_mem->flags_attr)
1768                 sysfs_notify_dirent(nfit_mem->flags_attr);
1769 }
1770 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1771
1772 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1773 {
1774         struct acpi_device *adev = data;
1775         struct device *dev = &adev->dev;
1776
1777         nfit_device_lock(dev->parent);
1778         __acpi_nvdimm_notify(dev, event);
1779         nfit_device_unlock(dev->parent);
1780 }
1781
1782 static bool acpi_nvdimm_has_method(struct acpi_device *adev, char *method)
1783 {
1784         acpi_handle handle;
1785         acpi_status status;
1786
1787         status = acpi_get_handle(adev->handle, method, &handle);
1788
1789         if (ACPI_SUCCESS(status))
1790                 return true;
1791         return false;
1792 }
1793
1794 __weak void nfit_intel_shutdown_status(struct nfit_mem *nfit_mem)
1795 {
1796         struct device *dev = &nfit_mem->adev->dev;
1797         struct nd_intel_smart smart = { 0 };
1798         union acpi_object in_buf = {
1799                 .buffer.type = ACPI_TYPE_BUFFER,
1800                 .buffer.length = 0,
1801         };
1802         union acpi_object in_obj = {
1803                 .package.type = ACPI_TYPE_PACKAGE,
1804                 .package.count = 1,
1805                 .package.elements = &in_buf,
1806         };
1807         const u8 func = ND_INTEL_SMART;
1808         const guid_t *guid = to_nfit_uuid(nfit_mem->family);
1809         u8 revid = nfit_dsm_revid(nfit_mem->family, func);
1810         struct acpi_device *adev = nfit_mem->adev;
1811         acpi_handle handle = adev->handle;
1812         union acpi_object *out_obj;
1813
1814         if ((nfit_mem->dsm_mask & (1 << func)) == 0)
1815                 return;
1816
1817         out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
1818         if (!out_obj || out_obj->type != ACPI_TYPE_BUFFER
1819                         || out_obj->buffer.length < sizeof(smart)) {
1820                 dev_dbg(dev->parent, "%s: failed to retrieve initial health\n",
1821                                 dev_name(dev));
1822                 ACPI_FREE(out_obj);
1823                 return;
1824         }
1825         memcpy(&smart, out_obj->buffer.pointer, sizeof(smart));
1826         ACPI_FREE(out_obj);
1827
1828         if (smart.flags & ND_INTEL_SMART_SHUTDOWN_VALID) {
1829                 if (smart.shutdown_state)
1830                         set_bit(NFIT_MEM_DIRTY, &nfit_mem->flags);
1831         }
1832
1833         if (smart.flags & ND_INTEL_SMART_SHUTDOWN_COUNT_VALID) {
1834                 set_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags);
1835                 nfit_mem->dirty_shutdown = smart.shutdown_count;
1836         }
1837 }
1838
1839 static void populate_shutdown_status(struct nfit_mem *nfit_mem)
1840 {
1841         /*
1842          * For DIMMs that provide a dynamic facility to retrieve a
1843          * dirty-shutdown status and/or a dirty-shutdown count, cache
1844          * these values in nfit_mem.
1845          */
1846         if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
1847                 nfit_intel_shutdown_status(nfit_mem);
1848 }
1849
1850 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1851                 struct nfit_mem *nfit_mem, u32 device_handle)
1852 {
1853         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1854         struct acpi_device *adev, *adev_dimm;
1855         struct device *dev = acpi_desc->dev;
1856         unsigned long dsm_mask, label_mask;
1857         const guid_t *guid;
1858         int i;
1859         int family = -1;
1860         struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
1861
1862         /* nfit test assumes 1:1 relationship between commands and dsms */
1863         nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1864         nfit_mem->family = NVDIMM_FAMILY_INTEL;
1865         set_bit(NVDIMM_FAMILY_INTEL, &nd_desc->dimm_family_mask);
1866
1867         if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1868                 sprintf(nfit_mem->id, "%04x-%02x-%04x-%08x",
1869                                 be16_to_cpu(dcr->vendor_id),
1870                                 dcr->manufacturing_location,
1871                                 be16_to_cpu(dcr->manufacturing_date),
1872                                 be32_to_cpu(dcr->serial_number));
1873         else
1874                 sprintf(nfit_mem->id, "%04x-%08x",
1875                                 be16_to_cpu(dcr->vendor_id),
1876                                 be32_to_cpu(dcr->serial_number));
1877
1878         adev = to_acpi_dev(acpi_desc);
1879         if (!adev) {
1880                 /* unit test case */
1881                 populate_shutdown_status(nfit_mem);
1882                 return 0;
1883         }
1884
1885         adev_dimm = acpi_find_child_device(adev, device_handle, false);
1886         nfit_mem->adev = adev_dimm;
1887         if (!adev_dimm) {
1888                 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1889                                 device_handle);
1890                 return force_enable_dimms ? 0 : -ENODEV;
1891         }
1892
1893         if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1894                 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1895                 dev_err(dev, "%s: notification registration failed\n",
1896                                 dev_name(&adev_dimm->dev));
1897                 return -ENXIO;
1898         }
1899         /*
1900          * Record nfit_mem for the notification path to track back to
1901          * the nfit sysfs attributes for this dimm device object.
1902          */
1903         dev_set_drvdata(&adev_dimm->dev, nfit_mem);
1904
1905         /*
1906          * There are 4 "legacy" NVDIMM command sets
1907          * (NVDIMM_FAMILY_{INTEL,MSFT,HPE1,HPE2}) that were created before
1908          * an EFI working group was established to constrain this
1909          * proliferation. The nfit driver probes for the supported command
1910          * set by GUID. Note, if you're a platform developer looking to add
1911          * a new command set to this probe, consider using an existing set,
1912          * or otherwise seek approval to publish the command set at
1913          * http://www.uefi.org/RFIC_LIST.
1914          *
1915          * Note, that checking for function0 (bit0) tells us if any commands
1916          * are reachable through this GUID.
1917          */
1918         clear_bit(NVDIMM_FAMILY_INTEL, &nd_desc->dimm_family_mask);
1919         for (i = 0; i <= NVDIMM_FAMILY_MAX; i++)
1920                 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1)) {
1921                         set_bit(i, &nd_desc->dimm_family_mask);
1922                         if (family < 0 || i == default_dsm_family)
1923                                 family = i;
1924                 }
1925
1926         /* limit the supported commands to those that are publicly documented */
1927         nfit_mem->family = family;
1928         if (override_dsm_mask && !disable_vendor_specific)
1929                 dsm_mask = override_dsm_mask;
1930         else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1931                 dsm_mask = NVDIMM_INTEL_CMDMASK;
1932                 if (disable_vendor_specific)
1933                         dsm_mask &= ~(1 << ND_CMD_VENDOR);
1934         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1935                 dsm_mask = 0x1c3c76;
1936         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1937                 dsm_mask = 0x1fe;
1938                 if (disable_vendor_specific)
1939                         dsm_mask &= ~(1 << 8);
1940         } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1941                 dsm_mask = 0xffffffff;
1942         } else if (nfit_mem->family == NVDIMM_FAMILY_HYPERV) {
1943                 dsm_mask = 0x1f;
1944         } else {
1945                 dev_dbg(dev, "unknown dimm command family\n");
1946                 nfit_mem->family = -1;
1947                 /* DSMs are optional, continue loading the driver... */
1948                 return 0;
1949         }
1950
1951         /*
1952          * Function 0 is the command interrogation function, don't
1953          * export it to potential userspace use, and enable it to be
1954          * used as an error value in acpi_nfit_ctl().
1955          */
1956         dsm_mask &= ~1UL;
1957
1958         guid = to_nfit_uuid(nfit_mem->family);
1959         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1960                 if (acpi_check_dsm(adev_dimm->handle, guid,
1961                                         nfit_dsm_revid(nfit_mem->family, i),
1962                                         1ULL << i))
1963                         set_bit(i, &nfit_mem->dsm_mask);
1964
1965         /*
1966          * Prefer the NVDIMM_FAMILY_INTEL label read commands if present
1967          * due to their better semantics handling locked capacity.
1968          */
1969         label_mask = 1 << ND_CMD_GET_CONFIG_SIZE | 1 << ND_CMD_GET_CONFIG_DATA
1970                 | 1 << ND_CMD_SET_CONFIG_DATA;
1971         if (family == NVDIMM_FAMILY_INTEL
1972                         && (dsm_mask & label_mask) == label_mask)
1973                 /* skip _LS{I,R,W} enabling */;
1974         else {
1975                 if (acpi_nvdimm_has_method(adev_dimm, "_LSI")
1976                                 && acpi_nvdimm_has_method(adev_dimm, "_LSR")) {
1977                         dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev));
1978                         set_bit(NFIT_MEM_LSR, &nfit_mem->flags);
1979                 }
1980
1981                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
1982                                 && acpi_nvdimm_has_method(adev_dimm, "_LSW")) {
1983                         dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev));
1984                         set_bit(NFIT_MEM_LSW, &nfit_mem->flags);
1985                 }
1986
1987                 /*
1988                  * Quirk read-only label configurations to preserve
1989                  * access to label-less namespaces by default.
1990                  */
1991                 if (!test_bit(NFIT_MEM_LSW, &nfit_mem->flags)
1992                                 && !force_labels) {
1993                         dev_dbg(dev, "%s: No _LSW, disable labels\n",
1994                                         dev_name(&adev_dimm->dev));
1995                         clear_bit(NFIT_MEM_LSR, &nfit_mem->flags);
1996                 } else
1997                         dev_dbg(dev, "%s: Force enable labels\n",
1998                                         dev_name(&adev_dimm->dev));
1999         }
2000
2001         populate_shutdown_status(nfit_mem);
2002
2003         return 0;
2004 }
2005
2006 static void shutdown_dimm_notify(void *data)
2007 {
2008         struct acpi_nfit_desc *acpi_desc = data;
2009         struct nfit_mem *nfit_mem;
2010
2011         mutex_lock(&acpi_desc->init_mutex);
2012         /*
2013          * Clear out the nfit_mem->flags_attr and shut down dimm event
2014          * notifications.
2015          */
2016         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2017                 struct acpi_device *adev_dimm = nfit_mem->adev;
2018
2019                 if (nfit_mem->flags_attr) {
2020                         sysfs_put(nfit_mem->flags_attr);
2021                         nfit_mem->flags_attr = NULL;
2022                 }
2023                 if (adev_dimm) {
2024                         acpi_remove_notify_handler(adev_dimm->handle,
2025                                         ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
2026                         dev_set_drvdata(&adev_dimm->dev, NULL);
2027                 }
2028         }
2029         mutex_unlock(&acpi_desc->init_mutex);
2030 }
2031
2032 static const struct nvdimm_security_ops *acpi_nfit_get_security_ops(int family)
2033 {
2034         switch (family) {
2035         case NVDIMM_FAMILY_INTEL:
2036                 return intel_security_ops;
2037         default:
2038                 return NULL;
2039         }
2040 }
2041
2042 static const struct nvdimm_fw_ops *acpi_nfit_get_fw_ops(
2043                 struct nfit_mem *nfit_mem)
2044 {
2045         unsigned long mask;
2046         struct acpi_nfit_desc *acpi_desc = nfit_mem->acpi_desc;
2047         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2048
2049         if (!nd_desc->fw_ops)
2050                 return NULL;
2051
2052         if (nfit_mem->family != NVDIMM_FAMILY_INTEL)
2053                 return NULL;
2054
2055         mask = nfit_mem->dsm_mask & NVDIMM_INTEL_FW_ACTIVATE_CMDMASK;
2056         if (mask != NVDIMM_INTEL_FW_ACTIVATE_CMDMASK)
2057                 return NULL;
2058
2059         return intel_fw_ops;
2060 }
2061
2062 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
2063 {
2064         struct nfit_mem *nfit_mem;
2065         int dimm_count = 0, rc;
2066         struct nvdimm *nvdimm;
2067
2068         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2069                 struct acpi_nfit_flush_address *flush;
2070                 unsigned long flags = 0, cmd_mask;
2071                 struct nfit_memdev *nfit_memdev;
2072                 u32 device_handle;
2073                 u16 mem_flags;
2074
2075                 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
2076                 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
2077                 if (nvdimm) {
2078                         dimm_count++;
2079                         continue;
2080                 }
2081
2082                 if (nfit_mem->bdw && nfit_mem->memdev_pmem) {
2083                         set_bit(NDD_ALIASING, &flags);
2084                         set_bit(NDD_LABELING, &flags);
2085                 }
2086
2087                 /* collate flags across all memdevs for this dimm */
2088                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2089                         struct acpi_nfit_memory_map *dimm_memdev;
2090
2091                         dimm_memdev = __to_nfit_memdev(nfit_mem);
2092                         if (dimm_memdev->device_handle
2093                                         != nfit_memdev->memdev->device_handle)
2094                                 continue;
2095                         dimm_memdev->flags |= nfit_memdev->memdev->flags;
2096                 }
2097
2098                 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
2099                 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
2100                         set_bit(NDD_UNARMED, &flags);
2101
2102                 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
2103                 if (rc)
2104                         continue;
2105
2106                 /*
2107                  * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
2108                  * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
2109                  * userspace interface.
2110                  */
2111                 cmd_mask = 1UL << ND_CMD_CALL;
2112                 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
2113                         /*
2114                          * These commands have a 1:1 correspondence
2115                          * between DSM payload and libnvdimm ioctl
2116                          * payload format.
2117                          */
2118                         cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
2119                 }
2120
2121                 /* Quirk to ignore LOCAL for labels on HYPERV DIMMs */
2122                 if (nfit_mem->family == NVDIMM_FAMILY_HYPERV)
2123                         set_bit(NDD_NOBLK, &flags);
2124
2125                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
2126                         set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
2127                         set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
2128                 }
2129                 if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags))
2130                         set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
2131
2132                 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
2133                         : NULL;
2134                 nvdimm = __nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
2135                                 acpi_nfit_dimm_attribute_groups,
2136                                 flags, cmd_mask, flush ? flush->hint_count : 0,
2137                                 nfit_mem->flush_wpq, &nfit_mem->id[0],
2138                                 acpi_nfit_get_security_ops(nfit_mem->family),
2139                                 acpi_nfit_get_fw_ops(nfit_mem));
2140                 if (!nvdimm)
2141                         return -ENOMEM;
2142
2143                 nfit_mem->nvdimm = nvdimm;
2144                 dimm_count++;
2145
2146                 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
2147                         continue;
2148
2149                 dev_err(acpi_desc->dev, "Error found in NVDIMM %s flags:%s%s%s%s%s\n",
2150                                 nvdimm_name(nvdimm),
2151                   mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
2152                   mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
2153                   mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
2154                   mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
2155                   mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
2156
2157         }
2158
2159         rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
2160         if (rc)
2161                 return rc;
2162
2163         /*
2164          * Now that dimms are successfully registered, and async registration
2165          * is flushed, attempt to enable event notification.
2166          */
2167         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2168                 struct kernfs_node *nfit_kernfs;
2169
2170                 nvdimm = nfit_mem->nvdimm;
2171                 if (!nvdimm)
2172                         continue;
2173
2174                 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
2175                 if (nfit_kernfs)
2176                         nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
2177                                         "flags");
2178                 sysfs_put(nfit_kernfs);
2179                 if (!nfit_mem->flags_attr)
2180                         dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
2181                                         nvdimm_name(nvdimm));
2182         }
2183
2184         return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
2185                         acpi_desc);
2186 }
2187
2188 /*
2189  * These constants are private because there are no kernel consumers of
2190  * these commands.
2191  */
2192 enum nfit_aux_cmds {
2193         NFIT_CMD_TRANSLATE_SPA = 5,
2194         NFIT_CMD_ARS_INJECT_SET = 7,
2195         NFIT_CMD_ARS_INJECT_CLEAR = 8,
2196         NFIT_CMD_ARS_INJECT_GET = 9,
2197 };
2198
2199 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
2200 {
2201         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2202         const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
2203         unsigned long dsm_mask, *mask;
2204         struct acpi_device *adev;
2205         int i;
2206
2207         set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
2208         set_bit(NVDIMM_BUS_FAMILY_NFIT, &nd_desc->bus_family_mask);
2209
2210         /* enable nfit_test to inject bus command emulation */
2211         if (acpi_desc->bus_cmd_force_en) {
2212                 nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
2213                 mask = &nd_desc->bus_family_mask;
2214                 if (acpi_desc->family_dsm_mask[NVDIMM_BUS_FAMILY_INTEL]) {
2215                         set_bit(NVDIMM_BUS_FAMILY_INTEL, mask);
2216                         nd_desc->fw_ops = intel_bus_fw_ops;
2217                 }
2218         }
2219
2220         adev = to_acpi_dev(acpi_desc);
2221         if (!adev)
2222                 return;
2223
2224         for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
2225                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2226                         set_bit(i, &nd_desc->cmd_mask);
2227
2228         dsm_mask =
2229                 (1 << ND_CMD_ARS_CAP) |
2230                 (1 << ND_CMD_ARS_START) |
2231                 (1 << ND_CMD_ARS_STATUS) |
2232                 (1 << ND_CMD_CLEAR_ERROR) |
2233                 (1 << NFIT_CMD_TRANSLATE_SPA) |
2234                 (1 << NFIT_CMD_ARS_INJECT_SET) |
2235                 (1 << NFIT_CMD_ARS_INJECT_CLEAR) |
2236                 (1 << NFIT_CMD_ARS_INJECT_GET);
2237         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
2238                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2239                         set_bit(i, &acpi_desc->bus_dsm_mask);
2240
2241         /* Enumerate allowed NVDIMM_BUS_FAMILY_INTEL commands */
2242         dsm_mask = NVDIMM_BUS_INTEL_FW_ACTIVATE_CMDMASK;
2243         guid = to_nfit_bus_uuid(NVDIMM_BUS_FAMILY_INTEL);
2244         mask = &acpi_desc->family_dsm_mask[NVDIMM_BUS_FAMILY_INTEL];
2245         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
2246                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2247                         set_bit(i, mask);
2248
2249         if (*mask == dsm_mask) {
2250                 set_bit(NVDIMM_BUS_FAMILY_INTEL, &nd_desc->bus_family_mask);
2251                 nd_desc->fw_ops = intel_bus_fw_ops;
2252         }
2253 }
2254
2255 static ssize_t range_index_show(struct device *dev,
2256                 struct device_attribute *attr, char *buf)
2257 {
2258         struct nd_region *nd_region = to_nd_region(dev);
2259         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
2260
2261         return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
2262 }
2263 static DEVICE_ATTR_RO(range_index);
2264
2265 static struct attribute *acpi_nfit_region_attributes[] = {
2266         &dev_attr_range_index.attr,
2267         NULL,
2268 };
2269
2270 static const struct attribute_group acpi_nfit_region_attribute_group = {
2271         .name = "nfit",
2272         .attrs = acpi_nfit_region_attributes,
2273 };
2274
2275 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
2276         &acpi_nfit_region_attribute_group,
2277         NULL,
2278 };
2279
2280 /* enough info to uniquely specify an interleave set */
2281 struct nfit_set_info {
2282         u64 region_offset;
2283         u32 serial_number;
2284         u32 pad;
2285 };
2286
2287 struct nfit_set_info2 {
2288         u64 region_offset;
2289         u32 serial_number;
2290         u16 vendor_id;
2291         u16 manufacturing_date;
2292         u8 manufacturing_location;
2293         u8 reserved[31];
2294 };
2295
2296 static int cmp_map_compat(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         return memcmp(&map0->region_offset, &map1->region_offset,
2302                         sizeof(u64));
2303 }
2304
2305 static int cmp_map(const void *m0, const void *m1)
2306 {
2307         const struct nfit_set_info *map0 = m0;
2308         const struct nfit_set_info *map1 = m1;
2309
2310         if (map0->region_offset < map1->region_offset)
2311                 return -1;
2312         else if (map0->region_offset > map1->region_offset)
2313                 return 1;
2314         return 0;
2315 }
2316
2317 static int cmp_map2(const void *m0, const void *m1)
2318 {
2319         const struct nfit_set_info2 *map0 = m0;
2320         const struct nfit_set_info2 *map1 = m1;
2321
2322         if (map0->region_offset < map1->region_offset)
2323                 return -1;
2324         else if (map0->region_offset > map1->region_offset)
2325                 return 1;
2326         return 0;
2327 }
2328
2329 /* Retrieve the nth entry referencing this spa */
2330 static struct acpi_nfit_memory_map *memdev_from_spa(
2331                 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
2332 {
2333         struct nfit_memdev *nfit_memdev;
2334
2335         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
2336                 if (nfit_memdev->memdev->range_index == range_index)
2337                         if (n-- == 0)
2338                                 return nfit_memdev->memdev;
2339         return NULL;
2340 }
2341
2342 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
2343                 struct nd_region_desc *ndr_desc,
2344                 struct acpi_nfit_system_address *spa)
2345 {
2346         struct device *dev = acpi_desc->dev;
2347         struct nd_interleave_set *nd_set;
2348         u16 nr = ndr_desc->num_mappings;
2349         struct nfit_set_info2 *info2;
2350         struct nfit_set_info *info;
2351         int i;
2352
2353         nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
2354         if (!nd_set)
2355                 return -ENOMEM;
2356         import_guid(&nd_set->type_guid, spa->range_guid);
2357
2358         info = devm_kcalloc(dev, nr, sizeof(*info), GFP_KERNEL);
2359         if (!info)
2360                 return -ENOMEM;
2361
2362         info2 = devm_kcalloc(dev, nr, sizeof(*info2), GFP_KERNEL);
2363         if (!info2)
2364                 return -ENOMEM;
2365
2366         for (i = 0; i < nr; i++) {
2367                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
2368                 struct nvdimm *nvdimm = mapping->nvdimm;
2369                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2370                 struct nfit_set_info *map = &info[i];
2371                 struct nfit_set_info2 *map2 = &info2[i];
2372                 struct acpi_nfit_memory_map *memdev =
2373                         memdev_from_spa(acpi_desc, spa->range_index, i);
2374                 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2375
2376                 if (!memdev || !nfit_mem->dcr) {
2377                         dev_err(dev, "%s: failed to find DCR\n", __func__);
2378                         return -ENODEV;
2379                 }
2380
2381                 map->region_offset = memdev->region_offset;
2382                 map->serial_number = dcr->serial_number;
2383
2384                 map2->region_offset = memdev->region_offset;
2385                 map2->serial_number = dcr->serial_number;
2386                 map2->vendor_id = dcr->vendor_id;
2387                 map2->manufacturing_date = dcr->manufacturing_date;
2388                 map2->manufacturing_location = dcr->manufacturing_location;
2389         }
2390
2391         /* v1.1 namespaces */
2392         sort(info, nr, sizeof(*info), cmp_map, NULL);
2393         nd_set->cookie1 = nd_fletcher64(info, sizeof(*info) * nr, 0);
2394
2395         /* v1.2 namespaces */
2396         sort(info2, nr, sizeof(*info2), cmp_map2, NULL);
2397         nd_set->cookie2 = nd_fletcher64(info2, sizeof(*info2) * nr, 0);
2398
2399         /* support v1.1 namespaces created with the wrong sort order */
2400         sort(info, nr, sizeof(*info), cmp_map_compat, NULL);
2401         nd_set->altcookie = nd_fletcher64(info, sizeof(*info) * nr, 0);
2402
2403         /* record the result of the sort for the mapping position */
2404         for (i = 0; i < nr; i++) {
2405                 struct nfit_set_info2 *map2 = &info2[i];
2406                 int j;
2407
2408                 for (j = 0; j < nr; j++) {
2409                         struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
2410                         struct nvdimm *nvdimm = mapping->nvdimm;
2411                         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2412                         struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2413
2414                         if (map2->serial_number == dcr->serial_number &&
2415                             map2->vendor_id == dcr->vendor_id &&
2416                             map2->manufacturing_date == dcr->manufacturing_date &&
2417                             map2->manufacturing_location
2418                                     == dcr->manufacturing_location) {
2419                                 mapping->position = i;
2420                                 break;
2421                         }
2422                 }
2423         }
2424
2425         ndr_desc->nd_set = nd_set;
2426         devm_kfree(dev, info);
2427         devm_kfree(dev, info2);
2428
2429         return 0;
2430 }
2431
2432 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
2433 {
2434         struct acpi_nfit_interleave *idt = mmio->idt;
2435         u32 sub_line_offset, line_index, line_offset;
2436         u64 line_no, table_skip_count, table_offset;
2437
2438         line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
2439         table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
2440         line_offset = idt->line_offset[line_index]
2441                 * mmio->line_size;
2442         table_offset = table_skip_count * mmio->table_size;
2443
2444         return mmio->base_offset + line_offset + table_offset + sub_line_offset;
2445 }
2446
2447 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
2448 {
2449         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2450         u64 offset = nfit_blk->stat_offset + mmio->size * bw;
2451         const u32 STATUS_MASK = 0x80000037;
2452
2453         if (mmio->num_lines)
2454                 offset = to_interleave_offset(offset, mmio);
2455
2456         return readl(mmio->addr.base + offset) & STATUS_MASK;
2457 }
2458
2459 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
2460                 resource_size_t dpa, unsigned int len, unsigned int write)
2461 {
2462         u64 cmd, offset;
2463         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2464
2465         enum {
2466                 BCW_OFFSET_MASK = (1ULL << 48)-1,
2467                 BCW_LEN_SHIFT = 48,
2468                 BCW_LEN_MASK = (1ULL << 8) - 1,
2469                 BCW_CMD_SHIFT = 56,
2470         };
2471
2472         cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
2473         len = len >> L1_CACHE_SHIFT;
2474         cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
2475         cmd |= ((u64) write) << BCW_CMD_SHIFT;
2476
2477         offset = nfit_blk->cmd_offset + mmio->size * bw;
2478         if (mmio->num_lines)
2479                 offset = to_interleave_offset(offset, mmio);
2480
2481         writeq(cmd, mmio->addr.base + offset);
2482         nvdimm_flush(nfit_blk->nd_region, NULL);
2483
2484         if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
2485                 readq(mmio->addr.base + offset);
2486 }
2487
2488 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
2489                 resource_size_t dpa, void *iobuf, size_t len, int rw,
2490                 unsigned int lane)
2491 {
2492         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2493         unsigned int copied = 0;
2494         u64 base_offset;
2495         int rc;
2496
2497         base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
2498                 + lane * mmio->size;
2499         write_blk_ctl(nfit_blk, lane, dpa, len, rw);
2500         while (len) {
2501                 unsigned int c;
2502                 u64 offset;
2503
2504                 if (mmio->num_lines) {
2505                         u32 line_offset;
2506
2507                         offset = to_interleave_offset(base_offset + copied,
2508                                         mmio);
2509                         div_u64_rem(offset, mmio->line_size, &line_offset);
2510                         c = min_t(size_t, len, mmio->line_size - line_offset);
2511                 } else {
2512                         offset = base_offset + nfit_blk->bdw_offset;
2513                         c = len;
2514                 }
2515
2516                 if (rw)
2517                         memcpy_flushcache(mmio->addr.aperture + offset, iobuf + copied, c);
2518                 else {
2519                         if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
2520                                 arch_invalidate_pmem((void __force *)
2521                                         mmio->addr.aperture + offset, c);
2522
2523                         memcpy(iobuf + copied, mmio->addr.aperture + offset, c);
2524                 }
2525
2526                 copied += c;
2527                 len -= c;
2528         }
2529
2530         if (rw)
2531                 nvdimm_flush(nfit_blk->nd_region, NULL);
2532
2533         rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
2534         return rc;
2535 }
2536
2537 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
2538                 resource_size_t dpa, void *iobuf, u64 len, int rw)
2539 {
2540         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
2541         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2542         struct nd_region *nd_region = nfit_blk->nd_region;
2543         unsigned int lane, copied = 0;
2544         int rc = 0;
2545
2546         lane = nd_region_acquire_lane(nd_region);
2547         while (len) {
2548                 u64 c = min(len, mmio->size);
2549
2550                 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
2551                                 iobuf + copied, c, rw, lane);
2552                 if (rc)
2553                         break;
2554
2555                 copied += c;
2556                 len -= c;
2557         }
2558         nd_region_release_lane(nd_region, lane);
2559
2560         return rc;
2561 }
2562
2563 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
2564                 struct acpi_nfit_interleave *idt, u16 interleave_ways)
2565 {
2566         if (idt) {
2567                 mmio->num_lines = idt->line_count;
2568                 mmio->line_size = idt->line_size;
2569                 if (interleave_ways == 0)
2570                         return -ENXIO;
2571                 mmio->table_size = mmio->num_lines * interleave_ways
2572                         * mmio->line_size;
2573         }
2574
2575         return 0;
2576 }
2577
2578 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
2579                 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
2580 {
2581         struct nd_cmd_dimm_flags flags;
2582         int rc;
2583
2584         memset(&flags, 0, sizeof(flags));
2585         rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
2586                         sizeof(flags), NULL);
2587
2588         if (rc >= 0 && flags.status == 0)
2589                 nfit_blk->dimm_flags = flags.flags;
2590         else if (rc == -ENOTTY) {
2591                 /* fall back to a conservative default */
2592                 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
2593                 rc = 0;
2594         } else
2595                 rc = -ENXIO;
2596
2597         return rc;
2598 }
2599
2600 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
2601                 struct device *dev)
2602 {
2603         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
2604         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
2605         struct nfit_blk_mmio *mmio;
2606         struct nfit_blk *nfit_blk;
2607         struct nfit_mem *nfit_mem;
2608         struct nvdimm *nvdimm;
2609         int rc;
2610
2611         nvdimm = nd_blk_region_to_dimm(ndbr);
2612         nfit_mem = nvdimm_provider_data(nvdimm);
2613         if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
2614                 dev_dbg(dev, "missing%s%s%s\n",
2615                                 nfit_mem ? "" : " nfit_mem",
2616                                 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
2617                                 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
2618                 return -ENXIO;
2619         }
2620
2621         nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
2622         if (!nfit_blk)
2623                 return -ENOMEM;
2624         nd_blk_region_set_provider_data(ndbr, nfit_blk);
2625         nfit_blk->nd_region = to_nd_region(dev);
2626
2627         /* map block aperture memory */
2628         nfit_blk->bdw_offset = nfit_mem->bdw->offset;
2629         mmio = &nfit_blk->mmio[BDW];
2630         mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
2631                         nfit_mem->spa_bdw->length, nd_blk_memremap_flags(ndbr));
2632         if (!mmio->addr.base) {
2633                 dev_dbg(dev, "%s failed to map bdw\n",
2634                                 nvdimm_name(nvdimm));
2635                 return -ENOMEM;
2636         }
2637         mmio->size = nfit_mem->bdw->size;
2638         mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
2639         mmio->idt = nfit_mem->idt_bdw;
2640         mmio->spa = nfit_mem->spa_bdw;
2641         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
2642                         nfit_mem->memdev_bdw->interleave_ways);
2643         if (rc) {
2644                 dev_dbg(dev, "%s failed to init bdw interleave\n",
2645                                 nvdimm_name(nvdimm));
2646                 return rc;
2647         }
2648
2649         /* map block control memory */
2650         nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
2651         nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
2652         mmio = &nfit_blk->mmio[DCR];
2653         mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
2654                         nfit_mem->spa_dcr->length);
2655         if (!mmio->addr.base) {
2656                 dev_dbg(dev, "%s failed to map dcr\n",
2657                                 nvdimm_name(nvdimm));
2658                 return -ENOMEM;
2659         }
2660         mmio->size = nfit_mem->dcr->window_size;
2661         mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
2662         mmio->idt = nfit_mem->idt_dcr;
2663         mmio->spa = nfit_mem->spa_dcr;
2664         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
2665                         nfit_mem->memdev_dcr->interleave_ways);
2666         if (rc) {
2667                 dev_dbg(dev, "%s failed to init dcr interleave\n",
2668                                 nvdimm_name(nvdimm));
2669                 return rc;
2670         }
2671
2672         rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
2673         if (rc < 0) {
2674                 dev_dbg(dev, "%s failed get DIMM flags\n",
2675                                 nvdimm_name(nvdimm));
2676                 return rc;
2677         }
2678
2679         if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
2680                 dev_warn(dev, "unable to guarantee persistence of writes\n");
2681
2682         if (mmio->line_size == 0)
2683                 return 0;
2684
2685         if ((u32) nfit_blk->cmd_offset % mmio->line_size
2686                         + 8 > mmio->line_size) {
2687                 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
2688                 return -ENXIO;
2689         } else if ((u32) nfit_blk->stat_offset % mmio->line_size
2690                         + 8 > mmio->line_size) {
2691                 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
2692                 return -ENXIO;
2693         }
2694
2695         return 0;
2696 }
2697
2698 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2699                 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2700 {
2701         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2702         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2703         int cmd_rc, rc;
2704
2705         cmd->address = spa->address;
2706         cmd->length = spa->length;
2707         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2708                         sizeof(*cmd), &cmd_rc);
2709         if (rc < 0)
2710                 return rc;
2711         return cmd_rc;
2712 }
2713
2714 static int ars_start(struct acpi_nfit_desc *acpi_desc,
2715                 struct nfit_spa *nfit_spa, enum nfit_ars_state req_type)
2716 {
2717         int rc;
2718         int cmd_rc;
2719         struct nd_cmd_ars_start ars_start;
2720         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2721         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2722
2723         memset(&ars_start, 0, sizeof(ars_start));
2724         ars_start.address = spa->address;
2725         ars_start.length = spa->length;
2726         if (req_type == ARS_REQ_SHORT)
2727                 ars_start.flags = ND_ARS_RETURN_PREV_DATA;
2728         if (nfit_spa_type(spa) == NFIT_SPA_PM)
2729                 ars_start.type = ND_ARS_PERSISTENT;
2730         else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2731                 ars_start.type = ND_ARS_VOLATILE;
2732         else
2733                 return -ENOTTY;
2734
2735         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2736                         sizeof(ars_start), &cmd_rc);
2737
2738         if (rc < 0)
2739                 return rc;
2740         if (cmd_rc < 0)
2741                 return cmd_rc;
2742         set_bit(ARS_VALID, &acpi_desc->scrub_flags);
2743         return 0;
2744 }
2745
2746 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2747 {
2748         int rc, cmd_rc;
2749         struct nd_cmd_ars_start ars_start;
2750         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2751         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2752
2753         ars_start = (struct nd_cmd_ars_start) {
2754                 .address = ars_status->restart_address,
2755                 .length = ars_status->restart_length,
2756                 .type = ars_status->type,
2757         };
2758         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2759                         sizeof(ars_start), &cmd_rc);
2760         if (rc < 0)
2761                 return rc;
2762         return cmd_rc;
2763 }
2764
2765 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2766 {
2767         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2768         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2769         int rc, cmd_rc;
2770
2771         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2772                         acpi_desc->max_ars, &cmd_rc);
2773         if (rc < 0)
2774                 return rc;
2775         return cmd_rc;
2776 }
2777
2778 static void ars_complete(struct acpi_nfit_desc *acpi_desc,
2779                 struct nfit_spa *nfit_spa)
2780 {
2781         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2782         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2783         struct nd_region *nd_region = nfit_spa->nd_region;
2784         struct device *dev;
2785
2786         lockdep_assert_held(&acpi_desc->init_mutex);
2787         /*
2788          * Only advance the ARS state for ARS runs initiated by the
2789          * kernel, ignore ARS results from BIOS initiated runs for scrub
2790          * completion tracking.
2791          */
2792         if (acpi_desc->scrub_spa != nfit_spa)
2793                 return;
2794
2795         if ((ars_status->address >= spa->address && ars_status->address
2796                                 < spa->address + spa->length)
2797                         || (ars_status->address < spa->address)) {
2798                 /*
2799                  * Assume that if a scrub starts at an offset from the
2800                  * start of nfit_spa that we are in the continuation
2801                  * case.
2802                  *
2803                  * Otherwise, if the scrub covers the spa range, mark
2804                  * any pending request complete.
2805                  */
2806                 if (ars_status->address + ars_status->length
2807                                 >= spa->address + spa->length)
2808                                 /* complete */;
2809                 else
2810                         return;
2811         } else
2812                 return;
2813
2814         acpi_desc->scrub_spa = NULL;
2815         if (nd_region) {
2816                 dev = nd_region_dev(nd_region);
2817                 nvdimm_region_notify(nd_region, NVDIMM_REVALIDATE_POISON);
2818         } else
2819                 dev = acpi_desc->dev;
2820         dev_dbg(dev, "ARS: range %d complete\n", spa->range_index);
2821 }
2822
2823 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc)
2824 {
2825         struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2826         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2827         int rc;
2828         u32 i;
2829
2830         /*
2831          * First record starts at 44 byte offset from the start of the
2832          * payload.
2833          */
2834         if (ars_status->out_length < 44)
2835                 return 0;
2836
2837         /*
2838          * Ignore potentially stale results that are only refreshed
2839          * after a start-ARS event.
2840          */
2841         if (!test_and_clear_bit(ARS_VALID, &acpi_desc->scrub_flags)) {
2842                 dev_dbg(acpi_desc->dev, "skip %d stale records\n",
2843                                 ars_status->num_records);
2844                 return 0;
2845         }
2846
2847         for (i = 0; i < ars_status->num_records; i++) {
2848                 /* only process full records */
2849                 if (ars_status->out_length
2850                                 < 44 + sizeof(struct nd_ars_record) * (i + 1))
2851                         break;
2852                 rc = nvdimm_bus_add_badrange(nvdimm_bus,
2853                                 ars_status->records[i].err_address,
2854                                 ars_status->records[i].length);
2855                 if (rc)
2856                         return rc;
2857         }
2858         if (i < ars_status->num_records)
2859                 dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2860
2861         return 0;
2862 }
2863
2864 static void acpi_nfit_remove_resource(void *data)
2865 {
2866         struct resource *res = data;
2867
2868         remove_resource(res);
2869 }
2870
2871 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2872                 struct nd_region_desc *ndr_desc)
2873 {
2874         struct resource *res, *nd_res = ndr_desc->res;
2875         int is_pmem, ret;
2876
2877         /* No operation if the region is already registered as PMEM */
2878         is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2879                                 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2880         if (is_pmem == REGION_INTERSECTS)
2881                 return 0;
2882
2883         res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2884         if (!res)
2885                 return -ENOMEM;
2886
2887         res->name = "Persistent Memory";
2888         res->start = nd_res->start;
2889         res->end = nd_res->end;
2890         res->flags = IORESOURCE_MEM;
2891         res->desc = IORES_DESC_PERSISTENT_MEMORY;
2892
2893         ret = insert_resource(&iomem_resource, res);
2894         if (ret)
2895                 return ret;
2896
2897         ret = devm_add_action_or_reset(acpi_desc->dev,
2898                                         acpi_nfit_remove_resource,
2899                                         res);
2900         if (ret)
2901                 return ret;
2902
2903         return 0;
2904 }
2905
2906 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2907                 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2908                 struct acpi_nfit_memory_map *memdev,
2909                 struct nfit_spa *nfit_spa)
2910 {
2911         struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2912                         memdev->device_handle);
2913         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2914         struct nd_blk_region_desc *ndbr_desc;
2915         struct nfit_mem *nfit_mem;
2916         int rc;
2917
2918         if (!nvdimm) {
2919                 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2920                                 spa->range_index, memdev->device_handle);
2921                 return -ENODEV;
2922         }
2923
2924         mapping->nvdimm = nvdimm;
2925         switch (nfit_spa_type(spa)) {
2926         case NFIT_SPA_PM:
2927         case NFIT_SPA_VOLATILE:
2928                 mapping->start = memdev->address;
2929                 mapping->size = memdev->region_size;
2930                 break;
2931         case NFIT_SPA_DCR:
2932                 nfit_mem = nvdimm_provider_data(nvdimm);
2933                 if (!nfit_mem || !nfit_mem->bdw) {
2934                         dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2935                                         spa->range_index, nvdimm_name(nvdimm));
2936                         break;
2937                 }
2938
2939                 mapping->size = nfit_mem->bdw->capacity;
2940                 mapping->start = nfit_mem->bdw->start_address;
2941                 ndr_desc->num_lanes = nfit_mem->bdw->windows;
2942                 ndr_desc->mapping = mapping;
2943                 ndr_desc->num_mappings = 1;
2944                 ndbr_desc = to_blk_region_desc(ndr_desc);
2945                 ndbr_desc->enable = acpi_nfit_blk_region_enable;
2946                 ndbr_desc->do_io = acpi_desc->blk_do_io;
2947                 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2948                 if (rc)
2949                         return rc;
2950                 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2951                                 ndr_desc);
2952                 if (!nfit_spa->nd_region)
2953                         return -ENOMEM;
2954                 break;
2955         }
2956
2957         return 0;
2958 }
2959
2960 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2961 {
2962         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2963                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2964                 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2965                 nfit_spa_type(spa) == NFIT_SPA_PCD);
2966 }
2967
2968 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
2969 {
2970         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2971                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2972                 nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
2973 }
2974
2975 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2976                 struct nfit_spa *nfit_spa)
2977 {
2978         static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2979         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2980         struct nd_blk_region_desc ndbr_desc;
2981         struct nd_region_desc *ndr_desc;
2982         struct nfit_memdev *nfit_memdev;
2983         struct nvdimm_bus *nvdimm_bus;
2984         struct resource res;
2985         int count = 0, rc;
2986
2987         if (nfit_spa->nd_region)
2988                 return 0;
2989
2990         if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2991                 dev_dbg(acpi_desc->dev, "detected invalid spa index\n");
2992                 return 0;
2993         }
2994
2995         memset(&res, 0, sizeof(res));
2996         memset(&mappings, 0, sizeof(mappings));
2997         memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2998         res.start = spa->address;
2999         res.end = res.start + spa->length - 1;
3000         ndr_desc = &ndbr_desc.ndr_desc;
3001         ndr_desc->res = &res;
3002         ndr_desc->provider_data = nfit_spa;
3003         ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
3004         if (spa->flags & ACPI_NFIT_PROXIMITY_VALID) {
3005                 ndr_desc->numa_node = pxm_to_online_node(spa->proximity_domain);
3006                 ndr_desc->target_node = pxm_to_node(spa->proximity_domain);
3007         } else {
3008                 ndr_desc->numa_node = NUMA_NO_NODE;
3009                 ndr_desc->target_node = NUMA_NO_NODE;
3010         }
3011
3012         /* Fallback to address based numa information if node lookup failed */
3013         if (ndr_desc->numa_node == NUMA_NO_NODE) {
3014                 ndr_desc->numa_node = memory_add_physaddr_to_nid(spa->address);
3015                 dev_info(acpi_desc->dev, "changing numa node from %d to %d for nfit region [%pa-%pa]",
3016                         NUMA_NO_NODE, ndr_desc->numa_node, &res.start, &res.end);
3017         }
3018         if (ndr_desc->target_node == NUMA_NO_NODE) {
3019                 ndr_desc->target_node = phys_to_target_node(spa->address);
3020                 dev_info(acpi_desc->dev, "changing target node from %d to %d for nfit region [%pa-%pa]",
3021                         NUMA_NO_NODE, ndr_desc->numa_node, &res.start, &res.end);
3022         }
3023
3024         /*
3025          * Persistence domain bits are hierarchical, if
3026          * ACPI_NFIT_CAPABILITY_CACHE_FLUSH is set then
3027          * ACPI_NFIT_CAPABILITY_MEM_FLUSH is implied.
3028          */
3029         if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_CACHE_FLUSH)
3030                 set_bit(ND_REGION_PERSIST_CACHE, &ndr_desc->flags);
3031         else if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_MEM_FLUSH)
3032                 set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc->flags);
3033
3034         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
3035                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
3036                 struct nd_mapping_desc *mapping;
3037
3038                 /* range index 0 == unmapped in SPA or invalid-SPA */
3039                 if (memdev->range_index == 0 || spa->range_index == 0)
3040                         continue;
3041                 if (memdev->range_index != spa->range_index)
3042                         continue;
3043                 if (count >= ND_MAX_MAPPINGS) {
3044                         dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
3045                                         spa->range_index, ND_MAX_MAPPINGS);
3046                         return -ENXIO;
3047                 }
3048                 mapping = &mappings[count++];
3049                 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
3050                                 memdev, nfit_spa);
3051                 if (rc)
3052                         goto out;
3053         }
3054
3055         ndr_desc->mapping = mappings;
3056         ndr_desc->num_mappings = count;
3057         rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
3058         if (rc)
3059                 goto out;
3060
3061         nvdimm_bus = acpi_desc->nvdimm_bus;
3062         if (nfit_spa_type(spa) == NFIT_SPA_PM) {
3063                 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
3064                 if (rc) {
3065                         dev_warn(acpi_desc->dev,
3066                                 "failed to insert pmem resource to iomem: %d\n",
3067                                 rc);
3068                         goto out;
3069                 }
3070
3071                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
3072                                 ndr_desc);
3073                 if (!nfit_spa->nd_region)
3074                         rc = -ENOMEM;
3075         } else if (nfit_spa_is_volatile(spa)) {
3076                 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
3077                                 ndr_desc);
3078                 if (!nfit_spa->nd_region)
3079                         rc = -ENOMEM;
3080         } else if (nfit_spa_is_virtual(spa)) {
3081                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
3082                                 ndr_desc);
3083                 if (!nfit_spa->nd_region)
3084                         rc = -ENOMEM;
3085         }
3086
3087  out:
3088         if (rc)
3089                 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
3090                                 nfit_spa->spa->range_index);
3091         return rc;
3092 }
3093
3094 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc)
3095 {
3096         struct device *dev = acpi_desc->dev;
3097         struct nd_cmd_ars_status *ars_status;
3098
3099         if (acpi_desc->ars_status) {
3100                 memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
3101                 return 0;
3102         }
3103
3104         ars_status = devm_kzalloc(dev, acpi_desc->max_ars, GFP_KERNEL);
3105         if (!ars_status)
3106                 return -ENOMEM;
3107         acpi_desc->ars_status = ars_status;
3108         return 0;
3109 }
3110
3111 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc)
3112 {
3113         int rc;
3114
3115         if (ars_status_alloc(acpi_desc))
3116                 return -ENOMEM;
3117
3118         rc = ars_get_status(acpi_desc);
3119
3120         if (rc < 0 && rc != -ENOSPC)
3121                 return rc;
3122
3123         if (ars_status_process_records(acpi_desc))
3124                 dev_err(acpi_desc->dev, "Failed to process ARS records\n");
3125
3126         return rc;
3127 }
3128
3129 static int ars_register(struct acpi_nfit_desc *acpi_desc,
3130                 struct nfit_spa *nfit_spa)
3131 {
3132         int rc;
3133
3134         if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3135                 return acpi_nfit_register_region(acpi_desc, nfit_spa);
3136
3137         set_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
3138         if (!no_init_ars)
3139                 set_bit(ARS_REQ_LONG, &nfit_spa->ars_state);
3140
3141         switch (acpi_nfit_query_poison(acpi_desc)) {
3142         case 0:
3143         case -ENOSPC:
3144         case -EAGAIN:
3145                 rc = ars_start(acpi_desc, nfit_spa, ARS_REQ_SHORT);
3146                 /* shouldn't happen, try again later */
3147                 if (rc == -EBUSY)
3148                         break;
3149                 if (rc) {
3150                         set_bit(ARS_FAILED, &nfit_spa->ars_state);
3151                         break;
3152                 }
3153                 clear_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
3154                 rc = acpi_nfit_query_poison(acpi_desc);
3155                 if (rc)
3156                         break;
3157                 acpi_desc->scrub_spa = nfit_spa;
3158                 ars_complete(acpi_desc, nfit_spa);
3159                 /*
3160                  * If ars_complete() says we didn't complete the
3161                  * short scrub, we'll try again with a long
3162                  * request.
3163                  */
3164                 acpi_desc->scrub_spa = NULL;
3165                 break;
3166         case -EBUSY:
3167         case -ENOMEM:
3168                 /*
3169                  * BIOS was using ARS, wait for it to complete (or
3170                  * resources to become available) and then perform our
3171                  * own scrubs.
3172                  */
3173                 break;
3174         default:
3175                 set_bit(ARS_FAILED, &nfit_spa->ars_state);
3176                 break;
3177         }
3178
3179         return acpi_nfit_register_region(acpi_desc, nfit_spa);
3180 }
3181
3182 static void ars_complete_all(struct acpi_nfit_desc *acpi_desc)
3183 {
3184         struct nfit_spa *nfit_spa;
3185
3186         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3187                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3188                         continue;
3189                 ars_complete(acpi_desc, nfit_spa);
3190         }
3191 }
3192
3193 static unsigned int __acpi_nfit_scrub(struct acpi_nfit_desc *acpi_desc,
3194                 int query_rc)
3195 {
3196         unsigned int tmo = acpi_desc->scrub_tmo;
3197         struct device *dev = acpi_desc->dev;
3198         struct nfit_spa *nfit_spa;
3199
3200         lockdep_assert_held(&acpi_desc->init_mutex);
3201
3202         if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags))
3203                 return 0;
3204
3205         if (query_rc == -EBUSY) {
3206                 dev_dbg(dev, "ARS: ARS busy\n");
3207                 return min(30U * 60U, tmo * 2);
3208         }
3209         if (query_rc == -ENOSPC) {
3210                 dev_dbg(dev, "ARS: ARS continue\n");
3211                 ars_continue(acpi_desc);
3212                 return 1;
3213         }
3214         if (query_rc && query_rc != -EAGAIN) {
3215                 unsigned long long addr, end;
3216
3217                 addr = acpi_desc->ars_status->address;
3218                 end = addr + acpi_desc->ars_status->length;
3219                 dev_dbg(dev, "ARS: %llx-%llx failed (%d)\n", addr, end,
3220                                 query_rc);
3221         }
3222
3223         ars_complete_all(acpi_desc);
3224         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3225                 enum nfit_ars_state req_type;
3226                 int rc;
3227
3228                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3229                         continue;
3230
3231                 /* prefer short ARS requests first */
3232                 if (test_bit(ARS_REQ_SHORT, &nfit_spa->ars_state))
3233                         req_type = ARS_REQ_SHORT;
3234                 else if (test_bit(ARS_REQ_LONG, &nfit_spa->ars_state))
3235                         req_type = ARS_REQ_LONG;
3236                 else
3237                         continue;
3238                 rc = ars_start(acpi_desc, nfit_spa, req_type);
3239
3240                 dev = nd_region_dev(nfit_spa->nd_region);
3241                 dev_dbg(dev, "ARS: range %d ARS start %s (%d)\n",
3242                                 nfit_spa->spa->range_index,
3243                                 req_type == ARS_REQ_SHORT ? "short" : "long",
3244                                 rc);
3245                 /*
3246                  * Hmm, we raced someone else starting ARS? Try again in
3247                  * a bit.
3248                  */
3249                 if (rc == -EBUSY)
3250                         return 1;
3251                 if (rc == 0) {
3252                         dev_WARN_ONCE(dev, acpi_desc->scrub_spa,
3253                                         "scrub start while range %d active\n",
3254                                         acpi_desc->scrub_spa->spa->range_index);
3255                         clear_bit(req_type, &nfit_spa->ars_state);
3256                         acpi_desc->scrub_spa = nfit_spa;
3257                         /*
3258                          * Consider this spa last for future scrub
3259                          * requests
3260                          */
3261                         list_move_tail(&nfit_spa->list, &acpi_desc->spas);
3262                         return 1;
3263                 }
3264
3265                 dev_err(dev, "ARS: range %d ARS failed (%d)\n",
3266                                 nfit_spa->spa->range_index, rc);
3267                 set_bit(ARS_FAILED, &nfit_spa->ars_state);
3268         }
3269         return 0;
3270 }
3271
3272 static void __sched_ars(struct acpi_nfit_desc *acpi_desc, unsigned int tmo)
3273 {
3274         lockdep_assert_held(&acpi_desc->init_mutex);
3275
3276         set_bit(ARS_BUSY, &acpi_desc->scrub_flags);
3277         /* note this should only be set from within the workqueue */
3278         if (tmo)
3279                 acpi_desc->scrub_tmo = tmo;
3280         queue_delayed_work(nfit_wq, &acpi_desc->dwork, tmo * HZ);
3281 }
3282
3283 static void sched_ars(struct acpi_nfit_desc *acpi_desc)
3284 {
3285         __sched_ars(acpi_desc, 0);
3286 }
3287
3288 static void notify_ars_done(struct acpi_nfit_desc *acpi_desc)
3289 {
3290         lockdep_assert_held(&acpi_desc->init_mutex);
3291
3292         clear_bit(ARS_BUSY, &acpi_desc->scrub_flags);
3293         acpi_desc->scrub_count++;
3294         if (acpi_desc->scrub_count_state)
3295                 sysfs_notify_dirent(acpi_desc->scrub_count_state);
3296 }
3297
3298 static void acpi_nfit_scrub(struct work_struct *work)
3299 {
3300         struct acpi_nfit_desc *acpi_desc;
3301         unsigned int tmo;
3302         int query_rc;
3303
3304         acpi_desc = container_of(work, typeof(*acpi_desc), dwork.work);
3305         mutex_lock(&acpi_desc->init_mutex);
3306         query_rc = acpi_nfit_query_poison(acpi_desc);
3307         tmo = __acpi_nfit_scrub(acpi_desc, query_rc);
3308         if (tmo)
3309                 __sched_ars(acpi_desc, tmo);
3310         else
3311                 notify_ars_done(acpi_desc);
3312         memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
3313         clear_bit(ARS_POLL, &acpi_desc->scrub_flags);
3314         mutex_unlock(&acpi_desc->init_mutex);
3315 }
3316
3317 static void acpi_nfit_init_ars(struct acpi_nfit_desc *acpi_desc,
3318                 struct nfit_spa *nfit_spa)
3319 {
3320         int type = nfit_spa_type(nfit_spa->spa);
3321         struct nd_cmd_ars_cap ars_cap;
3322         int rc;
3323
3324         set_bit(ARS_FAILED, &nfit_spa->ars_state);
3325         memset(&ars_cap, 0, sizeof(ars_cap));
3326         rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
3327         if (rc < 0)
3328                 return;
3329         /* check that the supported scrub types match the spa type */
3330         if (type == NFIT_SPA_VOLATILE && ((ars_cap.status >> 16)
3331                                 & ND_ARS_VOLATILE) == 0)
3332                 return;
3333         if (type == NFIT_SPA_PM && ((ars_cap.status >> 16)
3334                                 & ND_ARS_PERSISTENT) == 0)
3335                 return;
3336
3337         nfit_spa->max_ars = ars_cap.max_ars_out;
3338         nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
3339         acpi_desc->max_ars = max(nfit_spa->max_ars, acpi_desc->max_ars);
3340         clear_bit(ARS_FAILED, &nfit_spa->ars_state);
3341 }
3342
3343 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
3344 {
3345         struct nfit_spa *nfit_spa;
3346         int rc, do_sched_ars = 0;
3347
3348         set_bit(ARS_VALID, &acpi_desc->scrub_flags);
3349         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3350                 switch (nfit_spa_type(nfit_spa->spa)) {
3351                 case NFIT_SPA_VOLATILE:
3352                 case NFIT_SPA_PM:
3353                         acpi_nfit_init_ars(acpi_desc, nfit_spa);
3354                         break;
3355                 }
3356         }
3357
3358         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3359                 switch (nfit_spa_type(nfit_spa->spa)) {
3360                 case NFIT_SPA_VOLATILE:
3361                 case NFIT_SPA_PM:
3362                         /* register regions and kick off initial ARS run */
3363                         rc = ars_register(acpi_desc, nfit_spa);
3364                         if (rc)
3365                                 return rc;
3366
3367                         /*
3368                          * Kick off background ARS if at least one
3369                          * region successfully registered ARS
3370                          */
3371                         if (!test_bit(ARS_FAILED, &nfit_spa->ars_state))
3372                                 do_sched_ars++;
3373                         break;
3374                 case NFIT_SPA_BDW:
3375                         /* nothing to register */
3376                         break;
3377                 case NFIT_SPA_DCR:
3378                 case NFIT_SPA_VDISK:
3379                 case NFIT_SPA_VCD:
3380                 case NFIT_SPA_PDISK:
3381                 case NFIT_SPA_PCD:
3382                         /* register known regions that don't support ARS */
3383                         rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
3384                         if (rc)
3385                                 return rc;
3386                         break;
3387                 default:
3388                         /* don't register unknown regions */
3389                         break;
3390                 }
3391         }
3392
3393         if (do_sched_ars)
3394                 sched_ars(acpi_desc);
3395         return 0;
3396 }
3397
3398 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
3399                 struct nfit_table_prev *prev)
3400 {
3401         struct device *dev = acpi_desc->dev;
3402
3403         if (!list_empty(&prev->spas) ||
3404                         !list_empty(&prev->memdevs) ||
3405                         !list_empty(&prev->dcrs) ||
3406                         !list_empty(&prev->bdws) ||
3407                         !list_empty(&prev->idts) ||
3408                         !list_empty(&prev->flushes)) {
3409                 dev_err(dev, "new nfit deletes entries (unsupported)\n");
3410                 return -ENXIO;
3411         }
3412         return 0;
3413 }
3414
3415 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
3416 {
3417         struct device *dev = acpi_desc->dev;
3418         struct kernfs_node *nfit;
3419         struct device *bus_dev;
3420
3421         if (!ars_supported(acpi_desc->nvdimm_bus))
3422                 return 0;
3423
3424         bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3425         nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
3426         if (!nfit) {
3427                 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
3428                 return -ENODEV;
3429         }
3430         acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
3431         sysfs_put(nfit);
3432         if (!acpi_desc->scrub_count_state) {
3433                 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
3434                 return -ENODEV;
3435         }
3436
3437         return 0;
3438 }
3439
3440 static void acpi_nfit_unregister(void *data)
3441 {
3442         struct acpi_nfit_desc *acpi_desc = data;
3443
3444         nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
3445 }
3446
3447 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
3448 {
3449         struct device *dev = acpi_desc->dev;
3450         struct nfit_table_prev prev;
3451         const void *end;
3452         int rc;
3453
3454         if (!acpi_desc->nvdimm_bus) {
3455                 acpi_nfit_init_dsms(acpi_desc);
3456
3457                 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
3458                                 &acpi_desc->nd_desc);
3459                 if (!acpi_desc->nvdimm_bus)
3460                         return -ENOMEM;
3461
3462                 rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
3463                                 acpi_desc);
3464                 if (rc)
3465                         return rc;
3466
3467                 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
3468                 if (rc)
3469                         return rc;
3470
3471                 /* register this acpi_desc for mce notifications */
3472                 mutex_lock(&acpi_desc_lock);
3473                 list_add_tail(&acpi_desc->list, &acpi_descs);
3474                 mutex_unlock(&acpi_desc_lock);
3475         }
3476
3477         mutex_lock(&acpi_desc->init_mutex);
3478
3479         INIT_LIST_HEAD(&prev.spas);
3480         INIT_LIST_HEAD(&prev.memdevs);
3481         INIT_LIST_HEAD(&prev.dcrs);
3482         INIT_LIST_HEAD(&prev.bdws);
3483         INIT_LIST_HEAD(&prev.idts);
3484         INIT_LIST_HEAD(&prev.flushes);
3485
3486         list_cut_position(&prev.spas, &acpi_desc->spas,
3487                                 acpi_desc->spas.prev);
3488         list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
3489                                 acpi_desc->memdevs.prev);
3490         list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
3491                                 acpi_desc->dcrs.prev);
3492         list_cut_position(&prev.bdws, &acpi_desc->bdws,
3493                                 acpi_desc->bdws.prev);
3494         list_cut_position(&prev.idts, &acpi_desc->idts,
3495                                 acpi_desc->idts.prev);
3496         list_cut_position(&prev.flushes, &acpi_desc->flushes,
3497                                 acpi_desc->flushes.prev);
3498
3499         end = data + sz;
3500         while (!IS_ERR_OR_NULL(data))
3501                 data = add_table(acpi_desc, &prev, data, end);
3502
3503         if (IS_ERR(data)) {
3504                 dev_dbg(dev, "nfit table parsing error: %ld\n", PTR_ERR(data));
3505                 rc = PTR_ERR(data);
3506                 goto out_unlock;
3507         }
3508
3509         rc = acpi_nfit_check_deletions(acpi_desc, &prev);
3510         if (rc)
3511                 goto out_unlock;
3512
3513         rc = nfit_mem_init(acpi_desc);
3514         if (rc)
3515                 goto out_unlock;
3516
3517         rc = acpi_nfit_register_dimms(acpi_desc);
3518         if (rc)
3519                 goto out_unlock;
3520
3521         rc = acpi_nfit_register_regions(acpi_desc);
3522
3523  out_unlock:
3524         mutex_unlock(&acpi_desc->init_mutex);
3525         return rc;
3526 }
3527 EXPORT_SYMBOL_GPL(acpi_nfit_init);
3528
3529 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
3530 {
3531         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
3532         struct device *dev = acpi_desc->dev;
3533
3534         /* Bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
3535         nfit_device_lock(dev);
3536         nfit_device_unlock(dev);
3537
3538         /* Bounce the init_mutex to complete initial registration */
3539         mutex_lock(&acpi_desc->init_mutex);
3540         mutex_unlock(&acpi_desc->init_mutex);
3541
3542         return 0;
3543 }
3544
3545 static int __acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3546                 struct nvdimm *nvdimm, unsigned int cmd)
3547 {
3548         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
3549
3550         if (nvdimm)
3551                 return 0;
3552         if (cmd != ND_CMD_ARS_START)
3553                 return 0;
3554
3555         /*
3556          * The kernel and userspace may race to initiate a scrub, but
3557          * the scrub thread is prepared to lose that initial race.  It
3558          * just needs guarantees that any ARS it initiates are not
3559          * interrupted by any intervening start requests from userspace.
3560          */
3561         if (work_busy(&acpi_desc->dwork.work))
3562                 return -EBUSY;
3563
3564         return 0;
3565 }
3566
3567 /*
3568  * Prevent security and firmware activate commands from being issued via
3569  * ioctl.
3570  */
3571 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3572                 struct nvdimm *nvdimm, unsigned int cmd, void *buf)
3573 {
3574         struct nd_cmd_pkg *call_pkg = buf;
3575         unsigned int func;
3576
3577         if (nvdimm && cmd == ND_CMD_CALL &&
3578                         call_pkg->nd_family == NVDIMM_FAMILY_INTEL) {
3579                 func = call_pkg->nd_command;
3580                 if (func > NVDIMM_CMD_MAX ||
3581                     (1 << func) & NVDIMM_INTEL_DENY_CMDMASK)
3582                         return -EOPNOTSUPP;
3583         }
3584
3585         /* block all non-nfit bus commands */
3586         if (!nvdimm && cmd == ND_CMD_CALL &&
3587                         call_pkg->nd_family != NVDIMM_BUS_FAMILY_NFIT)
3588                 return -EOPNOTSUPP;
3589
3590         return __acpi_nfit_clear_to_send(nd_desc, nvdimm, cmd);
3591 }
3592
3593 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc,
3594                 enum nfit_ars_state req_type)
3595 {
3596         struct device *dev = acpi_desc->dev;
3597         int scheduled = 0, busy = 0;
3598         struct nfit_spa *nfit_spa;
3599
3600         mutex_lock(&acpi_desc->init_mutex);
3601         if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags)) {
3602                 mutex_unlock(&acpi_desc->init_mutex);
3603                 return 0;
3604         }
3605
3606         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3607                 int type = nfit_spa_type(nfit_spa->spa);
3608
3609                 if (type != NFIT_SPA_PM && type != NFIT_SPA_VOLATILE)
3610                         continue;
3611                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3612                         continue;
3613
3614                 if (test_and_set_bit(req_type, &nfit_spa->ars_state))
3615                         busy++;
3616                 else
3617                         scheduled++;
3618         }
3619         if (scheduled) {
3620                 sched_ars(acpi_desc);
3621                 dev_dbg(dev, "ars_scan triggered\n");
3622         }
3623         mutex_unlock(&acpi_desc->init_mutex);
3624
3625         if (scheduled)
3626                 return 0;
3627         if (busy)
3628                 return -EBUSY;
3629         return -ENOTTY;
3630 }
3631
3632 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
3633 {
3634         struct nvdimm_bus_descriptor *nd_desc;
3635
3636         dev_set_drvdata(dev, acpi_desc);
3637         acpi_desc->dev = dev;
3638         acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
3639         nd_desc = &acpi_desc->nd_desc;
3640         nd_desc->provider_name = "ACPI.NFIT";
3641         nd_desc->module = THIS_MODULE;
3642         nd_desc->ndctl = acpi_nfit_ctl;
3643         nd_desc->flush_probe = acpi_nfit_flush_probe;
3644         nd_desc->clear_to_send = acpi_nfit_clear_to_send;
3645         nd_desc->attr_groups = acpi_nfit_attribute_groups;
3646
3647         INIT_LIST_HEAD(&acpi_desc->spas);
3648         INIT_LIST_HEAD(&acpi_desc->dcrs);
3649         INIT_LIST_HEAD(&acpi_desc->bdws);
3650         INIT_LIST_HEAD(&acpi_desc->idts);
3651         INIT_LIST_HEAD(&acpi_desc->flushes);
3652         INIT_LIST_HEAD(&acpi_desc->memdevs);
3653         INIT_LIST_HEAD(&acpi_desc->dimms);
3654         INIT_LIST_HEAD(&acpi_desc->list);
3655         mutex_init(&acpi_desc->init_mutex);
3656         acpi_desc->scrub_tmo = 1;
3657         INIT_DELAYED_WORK(&acpi_desc->dwork, acpi_nfit_scrub);
3658 }
3659 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
3660
3661 static void acpi_nfit_put_table(void *table)
3662 {
3663         acpi_put_table(table);
3664 }
3665
3666 void acpi_nfit_shutdown(void *data)
3667 {
3668         struct acpi_nfit_desc *acpi_desc = data;
3669         struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3670
3671         /*
3672          * Destruct under acpi_desc_lock so that nfit_handle_mce does not
3673          * race teardown
3674          */
3675         mutex_lock(&acpi_desc_lock);
3676         list_del(&acpi_desc->list);
3677         mutex_unlock(&acpi_desc_lock);
3678
3679         mutex_lock(&acpi_desc->init_mutex);
3680         set_bit(ARS_CANCEL, &acpi_desc->scrub_flags);
3681         cancel_delayed_work_sync(&acpi_desc->dwork);
3682         mutex_unlock(&acpi_desc->init_mutex);
3683
3684         /*
3685          * Bounce the nvdimm bus lock to make sure any in-flight
3686          * acpi_nfit_ars_rescan() submissions have had a chance to
3687          * either submit or see ->cancel set.
3688          */
3689         nfit_device_lock(bus_dev);
3690         nfit_device_unlock(bus_dev);
3691
3692         flush_workqueue(nfit_wq);
3693 }
3694 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown);
3695
3696 static int acpi_nfit_add(struct acpi_device *adev)
3697 {
3698         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3699         struct acpi_nfit_desc *acpi_desc;
3700         struct device *dev = &adev->dev;
3701         struct acpi_table_header *tbl;
3702         acpi_status status = AE_OK;
3703         acpi_size sz;
3704         int rc = 0;
3705
3706         status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
3707         if (ACPI_FAILURE(status)) {
3708                 /* The NVDIMM root device allows OS to trigger enumeration of
3709                  * NVDIMMs through NFIT at boot time and re-enumeration at
3710                  * root level via the _FIT method during runtime.
3711                  * This is ok to return 0 here, we could have an nvdimm
3712                  * hotplugged later and evaluate _FIT method which returns
3713                  * data in the format of a series of NFIT Structures.
3714                  */
3715                 dev_dbg(dev, "failed to find NFIT at startup\n");
3716                 return 0;
3717         }
3718
3719         rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl);
3720         if (rc)
3721                 return rc;
3722         sz = tbl->length;
3723
3724         acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3725         if (!acpi_desc)
3726                 return -ENOMEM;
3727         acpi_nfit_desc_init(acpi_desc, &adev->dev);
3728
3729         /* Save the acpi header for exporting the revision via sysfs */
3730         acpi_desc->acpi_header = *tbl;
3731
3732         /* Evaluate _FIT and override with that if present */
3733         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
3734         if (ACPI_SUCCESS(status) && buf.length > 0) {
3735                 union acpi_object *obj = buf.pointer;
3736
3737                 if (obj->type == ACPI_TYPE_BUFFER)
3738                         rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3739                                         obj->buffer.length);
3740                 else
3741                         dev_dbg(dev, "invalid type %d, ignoring _FIT\n",
3742                                 (int) obj->type);
3743                 kfree(buf.pointer);
3744         } else
3745                 /* skip over the lead-in header table */
3746                 rc = acpi_nfit_init(acpi_desc, (void *) tbl
3747                                 + sizeof(struct acpi_table_nfit),
3748                                 sz - sizeof(struct acpi_table_nfit));
3749
3750         if (rc)
3751                 return rc;
3752         return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc);
3753 }
3754
3755 static int acpi_nfit_remove(struct acpi_device *adev)
3756 {
3757         /* see acpi_nfit_unregister */
3758         return 0;
3759 }
3760
3761 static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle)
3762 {
3763         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3764         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3765         union acpi_object *obj;
3766         acpi_status status;
3767         int ret;
3768
3769         if (!dev->driver) {
3770                 /* dev->driver may be null if we're being removed */
3771                 dev_dbg(dev, "no driver found for dev\n");
3772                 return;
3773         }
3774
3775         if (!acpi_desc) {
3776                 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3777                 if (!acpi_desc)
3778                         return;
3779                 acpi_nfit_desc_init(acpi_desc, dev);
3780         } else {
3781                 /*
3782                  * Finish previous registration before considering new
3783                  * regions.
3784                  */
3785                 flush_workqueue(nfit_wq);
3786         }
3787
3788         /* Evaluate _FIT */
3789         status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
3790         if (ACPI_FAILURE(status)) {
3791                 dev_err(dev, "failed to evaluate _FIT\n");
3792                 return;
3793         }
3794
3795         obj = buf.pointer;
3796         if (obj->type == ACPI_TYPE_BUFFER) {
3797                 ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3798                                 obj->buffer.length);
3799                 if (ret)
3800                         dev_err(dev, "failed to merge updated NFIT\n");
3801         } else
3802                 dev_err(dev, "Invalid _FIT\n");
3803         kfree(buf.pointer);
3804 }
3805
3806 static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle)
3807 {
3808         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3809
3810         if (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON)
3811                 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
3812         else
3813                 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_SHORT);
3814 }
3815
3816 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
3817 {
3818         dev_dbg(dev, "event: 0x%x\n", event);
3819
3820         switch (event) {
3821         case NFIT_NOTIFY_UPDATE:
3822                 return acpi_nfit_update_notify(dev, handle);
3823         case NFIT_NOTIFY_UC_MEMORY_ERROR:
3824                 return acpi_nfit_uc_error_notify(dev, handle);
3825         default:
3826                 return;
3827         }
3828 }
3829 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
3830
3831 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
3832 {
3833         nfit_device_lock(&adev->dev);
3834         __acpi_nfit_notify(&adev->dev, adev->handle, event);
3835         nfit_device_unlock(&adev->dev);
3836 }
3837
3838 static const struct acpi_device_id acpi_nfit_ids[] = {
3839         { "ACPI0012", 0 },
3840         { "", 0 },
3841 };
3842 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
3843
3844 static struct acpi_driver acpi_nfit_driver = {
3845         .name = KBUILD_MODNAME,
3846         .ids = acpi_nfit_ids,
3847         .ops = {
3848                 .add = acpi_nfit_add,
3849                 .remove = acpi_nfit_remove,
3850                 .notify = acpi_nfit_notify,
3851         },
3852 };
3853
3854 static __init int nfit_init(void)
3855 {
3856         int ret;
3857
3858         BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
3859         BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 64);
3860         BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
3861         BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
3862         BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
3863         BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
3864         BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
3865         BUILD_BUG_ON(sizeof(struct acpi_nfit_capabilities) != 16);
3866
3867         guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]);
3868         guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]);
3869         guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]);
3870         guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]);
3871         guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]);
3872         guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]);
3873         guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]);
3874         guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]);
3875         guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]);
3876         guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]);
3877         guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
3878         guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
3879         guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
3880         guid_parse(UUID_NFIT_DIMM_N_HYPERV, &nfit_uuid[NFIT_DEV_DIMM_N_HYPERV]);
3881         guid_parse(UUID_INTEL_BUS, &nfit_uuid[NFIT_BUS_INTEL]);
3882
3883         nfit_wq = create_singlethread_workqueue("nfit");
3884         if (!nfit_wq)
3885                 return -ENOMEM;
3886
3887         nfit_mce_register();
3888         ret = acpi_bus_register_driver(&acpi_nfit_driver);
3889         if (ret) {
3890                 nfit_mce_unregister();
3891                 destroy_workqueue(nfit_wq);
3892         }
3893
3894         return ret;
3895
3896 }
3897
3898 static __exit void nfit_exit(void)
3899 {
3900         nfit_mce_unregister();
3901         acpi_bus_unregister_driver(&acpi_nfit_driver);
3902         destroy_workqueue(nfit_wq);
3903         WARN_ON(!list_empty(&acpi_descs));
3904 }
3905
3906 module_init(nfit_init);
3907 module_exit(nfit_exit);
3908 MODULE_LICENSE("GPL v2");
3909 MODULE_AUTHOR("Intel Corporation");