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/mutex.h>
   9 #include <linux/ndctl.h>
  10 #include <linux/sysfs.h>
  11 #include <linux/delay.h>
  12 #include <linux/list.h>
  13 #include <linux/acpi.h>
  14 #include <linux/sort.h>
  15 #include <linux/io.h>
  16 #include <linux/nd.h>
  17 #include <asm/cacheflush.h>
  18 #include <acpi/nfit.h>
  19 #include "intel.h"
  20 #include "nfit.h"
  21
  22 /*
  23  * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
  24  * irrelevant.
  25  */
  26 #include <linux/io-64-nonatomic-hi-lo.h>
  27
  28 static bool force_enable_dimms;
  29 module_param(force_enable_dimms, bool, S_IRUGO|S_IWUSR);
  30 MODULE_PARM_DESC(force_enable_dimms, "Ignore _STA (ACPI DIMM device) status");
  31
  32 static bool disable_vendor_specific;
  33 module_param(disable_vendor_specific, bool, S_IRUGO);
  34 MODULE_PARM_DESC(disable_vendor_specific,
  35                 "Limit commands to the publicly specified set");
  36
  37 static unsigned long override_dsm_mask;
  38 module_param(override_dsm_mask, ulong, S_IRUGO);
  39 MODULE_PARM_DESC(override_dsm_mask, "Bitmask of allowed NVDIMM DSM functions");
  40
  41 static int default_dsm_family = -1;
  42 module_param(default_dsm_family, int, S_IRUGO);
  43 MODULE_PARM_DESC(default_dsm_family,
  44                 "Try this DSM type first when identifying NVDIMM family");
  45
  46 static bool no_init_ars;
  47 module_param(no_init_ars, bool, 0644);
  48 MODULE_PARM_DESC(no_init_ars, "Skip ARS run at nfit init time");
  49
  50 static bool force_labels;
  51 module_param(force_labels, bool, 0444);
  52 MODULE_PARM_DESC(force_labels, "Opt-in to labels despite missing methods");
  53
  54 LIST_HEAD(acpi_descs);
  55 DEFINE_MUTEX(acpi_desc_lock);
  56
  57 static struct workqueue_struct *nfit_wq;
  58
  59 struct nfit_table_prev {
  60         struct list_head spas;
  61         struct list_head memdevs;
  62         struct list_head dcrs;
  63         struct list_head bdws;
  64         struct list_head idts;
  65         struct list_head flushes;
  66 };
  67
  68 static guid_t nfit_uuid[NFIT_UUID_MAX];
  69
  70 const guid_t *to_nfit_uuid(enum nfit_uuids id)
  71 {
  72         return &nfit_uuid[id];
  73 }
  74 EXPORT_SYMBOL(to_nfit_uuid);
  75
  76 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
  77 {
  78         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
  79
  80         /*
  81          * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
  82          * acpi_device.
  83          */
  84         if (!nd_desc->provider_name
  85                         || strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
  86                 return NULL;
  87
  88         return to_acpi_device(acpi_desc->dev);
  89 }
  90
  91 static int xlat_bus_status(void *buf, unsigned int cmd, u32 status)
  92 {
  93         struct nd_cmd_clear_error *clear_err;
  94         struct nd_cmd_ars_status *ars_status;
  95         u16 flags;
  96
  97         switch (cmd) {
  98         case ND_CMD_ARS_CAP:
  99                 if ((status & 0xffff) == NFIT_ARS_CAP_NONE)
 100                         return -ENOTTY;
 101
 102                 /* Command failed */
 103                 if (status & 0xffff)
 104                         return -EIO;
 105
 106                 /* No supported scan types for this range */
 107                 flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
 108                 if ((status >> 16 & flags) == 0)
 109                         return -ENOTTY;
 110                 return 0;
 111         case ND_CMD_ARS_START:
 112                 /* ARS is in progress */
 113                 if ((status & 0xffff) == NFIT_ARS_START_BUSY)
 114                         return -EBUSY;
 115
 116                 /* Command failed */
 117                 if (status & 0xffff)
 118                         return -EIO;
 119                 return 0;
 120         case ND_CMD_ARS_STATUS:
 121                 ars_status = buf;
 122                 /* Command failed */
 123                 if (status & 0xffff)
 124                         return -EIO;
 125                 /* Check extended status (Upper two bytes) */
 126                 if (status == NFIT_ARS_STATUS_DONE)
 127                         return 0;
 128
 129                 /* ARS is in progress */
 130                 if (status == NFIT_ARS_STATUS_BUSY)
 131                         return -EBUSY;
 132
 133                 /* No ARS performed for the current boot */
 134                 if (status == NFIT_ARS_STATUS_NONE)
 135                         return -EAGAIN;
 136
 137                 /*
 138                  * ARS interrupted, either we overflowed or some other
 139                  * agent wants the scan to stop.  If we didn't overflow
 140                  * then just continue with the returned results.
 141                  */
 142                 if (status == NFIT_ARS_STATUS_INTR) {
 143                         if (ars_status->out_length >= 40 && (ars_status->flags
 144                                                 & NFIT_ARS_F_OVERFLOW))
 145                                 return -ENOSPC;
 146                         return 0;
 147                 }
 148
 149                 /* Unknown status */
 150                 if (status >> 16)
 151                         return -EIO;
 152                 return 0;
 153         case ND_CMD_CLEAR_ERROR:
 154                 clear_err = buf;
 155                 if (status & 0xffff)
 156                         return -EIO;
 157                 if (!clear_err->cleared)
 158                         return -EIO;
 159                 if (clear_err->length > clear_err->cleared)
 160                         return clear_err->cleared;
 161                 return 0;
 162         default:
 163                 break;
 164         }
 165
 166         /* all other non-zero status results in an error */
 167         if (status)
 168                 return -EIO;
 169         return 0;
 170 }
 171
 172 #define ACPI_LABELS_LOCKED 3
 173
 174 static int xlat_nvdimm_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
 175                 u32 status)
 176 {
 177         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 178
 179         switch (cmd) {
 180         case ND_CMD_GET_CONFIG_SIZE:
 181                 /*
 182                  * In the _LSI, _LSR, _LSW case the locked status is
 183                  * communicated via the read/write commands
 184                  */
 185                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags))
 186                         break;
 187
 188                 if (status >> 16 & ND_CONFIG_LOCKED)
 189                         return -EACCES;
 190                 break;
 191         case ND_CMD_GET_CONFIG_DATA:
 192                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
 193                                 && status == ACPI_LABELS_LOCKED)
 194                         return -EACCES;
 195                 break;
 196         case ND_CMD_SET_CONFIG_DATA:
 197                 if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags)
 198                                 && status == ACPI_LABELS_LOCKED)
 199                         return -EACCES;
 200                 break;
 201         default:
 202                 break;
 203         }
 204
 205         /* all other non-zero status results in an error */
 206         if (status)
 207                 return -EIO;
 208         return 0;
 209 }
 210
 211 static int xlat_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
 212                 u32 status)
 213 {
 214         if (!nvdimm)
 215                 return xlat_bus_status(buf, cmd, status);
 216         return xlat_nvdimm_status(nvdimm, buf, cmd, status);
 217 }
 218
 219 /* convert _LS{I,R} packages to the buffer object acpi_nfit_ctl expects */
 220 static union acpi_object *pkg_to_buf(union acpi_object *pkg)
 221 {
 222         int i;
 223         void *dst;
 224         size_t size = 0;
 225         union acpi_object *buf = NULL;
 226
 227         if (pkg->type != ACPI_TYPE_PACKAGE) {
 228                 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
 229                                 pkg->type);
 230                 goto err;
 231         }
 232
 233         for (i = 0; i < pkg->package.count; i++) {
 234                 union acpi_object *obj = &pkg->package.elements[i];
 235
 236                 if (obj->type == ACPI_TYPE_INTEGER)
 237                         size += 4;
 238                 else if (obj->type == ACPI_TYPE_BUFFER)
 239                         size += obj->buffer.length;
 240                 else {
 241                         WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
 242                                         obj->type);
 243                         goto err;
 244                 }
 245         }
 246
 247         buf = ACPI_ALLOCATE(sizeof(*buf) + size);
 248         if (!buf)
 249                 goto err;
 250
 251         dst = buf + 1;
 252         buf->type = ACPI_TYPE_BUFFER;
 253         buf->buffer.length = size;
 254         buf->buffer.pointer = dst;
 255         for (i = 0; i < pkg->package.count; i++) {
 256                 union acpi_object *obj = &pkg->package.elements[i];
 257
 258                 if (obj->type == ACPI_TYPE_INTEGER) {
 259                         memcpy(dst, &obj->integer.value, 4);
 260                         dst += 4;
 261                 } else if (obj->type == ACPI_TYPE_BUFFER) {
 262                         memcpy(dst, obj->buffer.pointer, obj->buffer.length);
 263                         dst += obj->buffer.length;
 264                 }
 265         }
 266 err:
 267         ACPI_FREE(pkg);
 268         return buf;
 269 }
 270
 271 static union acpi_object *int_to_buf(union acpi_object *integer)
 272 {
 273         union acpi_object *buf = ACPI_ALLOCATE(sizeof(*buf) + 4);
 274         void *dst = NULL;
 275
 276         if (!buf)
 277                 goto err;
 278
 279         if (integer->type != ACPI_TYPE_INTEGER) {
 280                 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
 281                                 integer->type);
 282                 goto err;
 283         }
 284
 285         dst = buf + 1;
 286         buf->type = ACPI_TYPE_BUFFER;
 287         buf->buffer.length = 4;
 288         buf->buffer.pointer = dst;
 289         memcpy(dst, &integer->integer.value, 4);
 290 err:
 291         ACPI_FREE(integer);
 292         return buf;
 293 }
 294
 295 static union acpi_object *acpi_label_write(acpi_handle handle, u32 offset,
 296                 u32 len, void *data)
 297 {
 298         acpi_status rc;
 299         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
 300         struct acpi_object_list input = {
 301                 .count = 3,
 302                 .pointer = (union acpi_object []) {
 303                         [0] = {
 304                                 .integer.type = ACPI_TYPE_INTEGER,
 305                                 .integer.value = offset,
 306                         },
 307                         [1] = {
 308                                 .integer.type = ACPI_TYPE_INTEGER,
 309                                 .integer.value = len,
 310                         },
 311                         [2] = {
 312                                 .buffer.type = ACPI_TYPE_BUFFER,
 313                                 .buffer.pointer = data,
 314                                 .buffer.length = len,
 315                         },
 316                 },
 317         };
 318
 319         rc = acpi_evaluate_object(handle, "_LSW", &input, &buf);
 320         if (ACPI_FAILURE(rc))
 321                 return NULL;
 322         return int_to_buf(buf.pointer);
 323 }
 324
 325 static union acpi_object *acpi_label_read(acpi_handle handle, u32 offset,
 326                 u32 len)
 327 {
 328         acpi_status rc;
 329         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
 330         struct acpi_object_list input = {
 331                 .count = 2,
 332                 .pointer = (union acpi_object []) {
 333                         [0] = {
 334                                 .integer.type = ACPI_TYPE_INTEGER,
 335                                 .integer.value = offset,
 336                         },
 337                         [1] = {
 338                                 .integer.type = ACPI_TYPE_INTEGER,
 339                                 .integer.value = len,
 340                         },
 341                 },
 342         };
 343
 344         rc = acpi_evaluate_object(handle, "_LSR", &input, &buf);
 345         if (ACPI_FAILURE(rc))
 346                 return NULL;
 347         return pkg_to_buf(buf.pointer);
 348 }
 349
 350 static union acpi_object *acpi_label_info(acpi_handle handle)
 351 {
 352         acpi_status rc;
 353         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
 354
 355         rc = acpi_evaluate_object(handle, "_LSI", NULL, &buf);
 356         if (ACPI_FAILURE(rc))
 357                 return NULL;
 358         return pkg_to_buf(buf.pointer);
 359 }
 360
 361 static u8 nfit_dsm_revid(unsigned family, unsigned func)
 362 {
 363         static const u8 revid_table[NVDIMM_FAMILY_MAX+1][32] = {
 364                 [NVDIMM_FAMILY_INTEL] = {
 365                         [NVDIMM_INTEL_GET_MODES] = 2,
 366                         [NVDIMM_INTEL_GET_FWINFO] = 2,
 367                         [NVDIMM_INTEL_START_FWUPDATE] = 2,
 368                         [NVDIMM_INTEL_SEND_FWUPDATE] = 2,
 369                         [NVDIMM_INTEL_FINISH_FWUPDATE] = 2,
 370                         [NVDIMM_INTEL_QUERY_FWUPDATE] = 2,
 371                         [NVDIMM_INTEL_SET_THRESHOLD] = 2,
 372                         [NVDIMM_INTEL_INJECT_ERROR] = 2,
 373                         [NVDIMM_INTEL_GET_SECURITY_STATE] = 2,
 374                         [NVDIMM_INTEL_SET_PASSPHRASE] = 2,
 375                         [NVDIMM_INTEL_DISABLE_PASSPHRASE] = 2,
 376                         [NVDIMM_INTEL_UNLOCK_UNIT] = 2,
 377                         [NVDIMM_INTEL_FREEZE_LOCK] = 2,
 378                         [NVDIMM_INTEL_SECURE_ERASE] = 2,
 379                         [NVDIMM_INTEL_OVERWRITE] = 2,
 380                         [NVDIMM_INTEL_QUERY_OVERWRITE] = 2,
 381                         [NVDIMM_INTEL_SET_MASTER_PASSPHRASE] = 2,
 382                         [NVDIMM_INTEL_MASTER_SECURE_ERASE] = 2,
 383                 },
 384         };
 385         u8 id;
 386
 387         if (family > NVDIMM_FAMILY_MAX)
 388                 return 0;
 389         if (func > 31)
 390                 return 0;
 391         id = revid_table[family][func];
 392         if (id == 0)
 393                 return 1; /* default */
 394         return id;
 395 }
 396
 397 static bool payload_dumpable(struct nvdimm *nvdimm, unsigned int func)
 398 {
 399         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 400
 401         if (nfit_mem && nfit_mem->family == NVDIMM_FAMILY_INTEL
 402                         && func >= NVDIMM_INTEL_GET_SECURITY_STATE
 403                         && func <= NVDIMM_INTEL_MASTER_SECURE_ERASE)
 404                 return IS_ENABLED(CONFIG_NFIT_SECURITY_DEBUG);
 405         return true;
 406 }
 407
 408 static int cmd_to_func(struct nfit_mem *nfit_mem, unsigned int cmd,
 409                 struct nd_cmd_pkg *call_pkg)
 410 {
 411         if (call_pkg) {
 412                 int i;
 413
 414                 if (nfit_mem && nfit_mem->family != call_pkg->nd_family)
 415                         return -ENOTTY;
 416
 417                 for (i = 0; i < ARRAY_SIZE(call_pkg->nd_reserved2); i++)
 418                         if (call_pkg->nd_reserved2[i])
 419                                 return -EINVAL;
 420                 return call_pkg->nd_command;
 421         }
 422
 423         /* In the !call_pkg case, bus commands == bus functions */
 424         if (!nfit_mem)
 425                 return cmd;
 426
 427         /* Linux ND commands == NVDIMM_FAMILY_INTEL function numbers */
 428         if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
 429                 return cmd;
 430
 431         /*
 432          * Force function number validation to fail since 0 is never
 433          * published as a valid function in dsm_mask.
 434          */
 435         return 0;
 436 }
 437
 438 int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm,
 439                 unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc)
 440 {
 441         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
 442         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 443         union acpi_object in_obj, in_buf, *out_obj;
 444         const struct nd_cmd_desc *desc = NULL;
 445         struct device *dev = acpi_desc->dev;
 446         struct nd_cmd_pkg *call_pkg = NULL;
 447         const char *cmd_name, *dimm_name;
 448         unsigned long cmd_mask, dsm_mask;
 449         u32 offset, fw_status = 0;
 450         acpi_handle handle;
 451         const guid_t *guid;
 452         int func, rc, i;
 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);
 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                 dsm_mask = nd_desc->bus_dsm_mask;
 482                 desc = nd_cmd_bus_desc(cmd);
 483                 guid = to_nfit_uuid(NFIT_DEV_BUS);
 484                 handle = adev->handle;
 485                 dimm_name = "bus";
 486         }
 487
 488         if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
 489                 return -ENOTTY;
 490
 491         /*
 492          * Check for a valid command.  For ND_CMD_CALL, we also have to
 493          * make sure that the DSM function is supported.
 494          */
 495         if (cmd == ND_CMD_CALL && !test_bit(func, &dsm_mask))
 496                 return -ENOTTY;
 497         else if (!test_bit(cmd, &cmd_mask))
 498                 return -ENOTTY;
 499
 500         in_obj.type = ACPI_TYPE_PACKAGE;
 501         in_obj.package.count = 1;
 502         in_obj.package.elements = &in_buf;
 503         in_buf.type = ACPI_TYPE_BUFFER;
 504         in_buf.buffer.pointer = buf;
 505         in_buf.buffer.length = 0;
 506
 507         /* libnvdimm has already validated the input envelope */
 508         for (i = 0; i < desc->in_num; i++)
 509                 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
 510                                 i, buf);
 511
 512         if (call_pkg) {
 513                 /* skip over package wrapper */
 514                 in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
 515                 in_buf.buffer.length = call_pkg->nd_size_in;
 516         }
 517
 518         dev_dbg(dev, "%s cmd: %d: func: %d input length: %d\n",
 519                 dimm_name, cmd, func, in_buf.buffer.length);
 520         if (payload_dumpable(nvdimm, func))
 521                 print_hex_dump_debug("nvdimm in  ", DUMP_PREFIX_OFFSET, 4, 4,
 522                                 in_buf.buffer.pointer,
 523                                 min_t(u32, 256, in_buf.buffer.length), true);
 524
 525         /* call the BIOS, prefer the named methods over _DSM if available */
 526         if (nvdimm && cmd == ND_CMD_GET_CONFIG_SIZE
 527                         && test_bit(NFIT_MEM_LSR, &nfit_mem->flags))
 528                 out_obj = acpi_label_info(handle);
 529         else if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA
 530                         && test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
 531                 struct nd_cmd_get_config_data_hdr *p = buf;
 532
 533                 out_obj = acpi_label_read(handle, p->in_offset, p->in_length);
 534         } else if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA
 535                         && test_bit(NFIT_MEM_LSW, &nfit_mem->flags)) {
 536                 struct nd_cmd_set_config_hdr *p = buf;
 537
 538                 out_obj = acpi_label_write(handle, p->in_offset, p->in_length,
 539                                 p->in_buf);
 540         } else {
 541                 u8 revid;
 542
 543                 if (nvdimm)
 544                         revid = nfit_dsm_revid(nfit_mem->family, func);
 545                 else
 546                         revid = 1;
 547                 out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
 548         }
 549
 550         if (!out_obj) {
 551                 dev_dbg(dev, "%s _DSM failed cmd: %s\n", dimm_name, cmd_name);
 552                 return -EINVAL;
 553         }
 554
 555         if (out_obj->type != ACPI_TYPE_BUFFER) {
 556                 dev_dbg(dev, "%s unexpected output object type cmd: %s type: %d\n",
 557                                 dimm_name, cmd_name, out_obj->type);
 558                 rc = -EINVAL;
 559                 goto out;
 560         }
 561
 562         dev_dbg(dev, "%s cmd: %s output length: %d\n", dimm_name,
 563                         cmd_name, out_obj->buffer.length);
 564         print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4, 4,
 565                         out_obj->buffer.pointer,
 566                         min_t(u32, 128, out_obj->buffer.length), true);
 567
 568         if (call_pkg) {
 569                 call_pkg->nd_fw_size = out_obj->buffer.length;
 570                 memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
 571                         out_obj->buffer.pointer,
 572                         min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
 573
 574                 ACPI_FREE(out_obj);
 575                 /*
 576                  * Need to support FW function w/o known size in advance.
 577                  * Caller can determine required size based upon nd_fw_size.
 578                  * If we return an error (like elsewhere) then caller wouldn't
 579                  * be able to rely upon data returned to make calculation.
 580                  */
 581                 if (cmd_rc)
 582                         *cmd_rc = 0;
 583                 return 0;
 584         }
 585
 586         for (i = 0, offset = 0; i < desc->out_num; i++) {
 587                 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
 588                                 (u32 *) out_obj->buffer.pointer,
 589                                 out_obj->buffer.length - offset);
 590
 591                 if (offset + out_size > out_obj->buffer.length) {
 592                         dev_dbg(dev, "%s output object underflow cmd: %s field: %d\n",
 593                                         dimm_name, cmd_name, i);
 594                         break;
 595                 }
 596
 597                 if (in_buf.buffer.length + offset + out_size > buf_len) {
 598                         dev_dbg(dev, "%s output overrun cmd: %s field: %d\n",
 599                                         dimm_name, cmd_name, i);
 600                         rc = -ENXIO;
 601                         goto out;
 602                 }
 603                 memcpy(buf + in_buf.buffer.length + offset,
 604                                 out_obj->buffer.pointer + offset, out_size);
 605                 offset += out_size;
 606         }
 607
 608         /*
 609          * Set fw_status for all the commands with a known format to be
 610          * later interpreted by xlat_status().
 611          */
 612         if (i >= 1 && ((!nvdimm && cmd >= ND_CMD_ARS_CAP
 613                                         && cmd <= ND_CMD_CLEAR_ERROR)
 614                                 || (nvdimm && cmd >= ND_CMD_SMART
 615                                         && cmd <= ND_CMD_VENDOR)))
 616                 fw_status = *(u32 *) out_obj->buffer.pointer;
 617
 618         if (offset + in_buf.buffer.length < buf_len) {
 619                 if (i >= 1) {
 620                         /*
 621                          * status valid, return the number of bytes left
 622                          * unfilled in the output buffer
 623                          */
 624                         rc = buf_len - offset - in_buf.buffer.length;
 625                         if (cmd_rc)
 626                                 *cmd_rc = xlat_status(nvdimm, buf, cmd,
 627                                                 fw_status);
 628                 } else {
 629                         dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
 630                                         __func__, dimm_name, cmd_name, buf_len,
 631                                         offset);
 632                         rc = -ENXIO;
 633                 }
 634         } else {
 635                 rc = 0;
 636                 if (cmd_rc)
 637                         *cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status);
 638         }
 639
 640  out:
 641         ACPI_FREE(out_obj);
 642
 643         return rc;
 644 }
 645 EXPORT_SYMBOL_GPL(acpi_nfit_ctl);
 646
 647 static const char *spa_type_name(u16 type)
 648 {
 649         static const char *to_name[] = {
 650                 [NFIT_SPA_VOLATILE] = "volatile",
 651                 [NFIT_SPA_PM] = "pmem",
 652                 [NFIT_SPA_DCR] = "dimm-control-region",
 653                 [NFIT_SPA_BDW] = "block-data-window",
 654                 [NFIT_SPA_VDISK] = "volatile-disk",
 655                 [NFIT_SPA_VCD] = "volatile-cd",
 656                 [NFIT_SPA_PDISK] = "persistent-disk",
 657                 [NFIT_SPA_PCD] = "persistent-cd",
 658
 659         };
 660
 661         if (type > NFIT_SPA_PCD)
 662                 return "unknown";
 663
 664         return to_name[type];
 665 }
 666
 667 int nfit_spa_type(struct acpi_nfit_system_address *spa)
 668 {
 669         int i;
 670
 671         for (i = 0; i < NFIT_UUID_MAX; i++)
 672                 if (guid_equal(to_nfit_uuid(i), (guid_t *)&spa->range_guid))
 673                         return i;
 674         return -1;
 675 }
 676
 677 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
 678                 struct nfit_table_prev *prev,
 679                 struct acpi_nfit_system_address *spa)
 680 {
 681         struct device *dev = acpi_desc->dev;
 682         struct nfit_spa *nfit_spa;
 683
 684         if (spa->header.length != sizeof(*spa))
 685                 return false;
 686
 687         list_for_each_entry(nfit_spa, &prev->spas, list) {
 688                 if (memcmp(nfit_spa->spa, spa, sizeof(*spa)) == 0) {
 689                         list_move_tail(&nfit_spa->list, &acpi_desc->spas);
 690                         return true;
 691                 }
 692         }
 693
 694         nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof(*spa),
 695                         GFP_KERNEL);
 696         if (!nfit_spa)
 697                 return false;
 698         INIT_LIST_HEAD(&nfit_spa->list);
 699         memcpy(nfit_spa->spa, spa, sizeof(*spa));
 700         list_add_tail(&nfit_spa->list, &acpi_desc->spas);
 701         dev_dbg(dev, "spa index: %d type: %s\n",
 702                         spa->range_index,
 703                         spa_type_name(nfit_spa_type(spa)));
 704         return true;
 705 }
 706
 707 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
 708                 struct nfit_table_prev *prev,
 709                 struct acpi_nfit_memory_map *memdev)
 710 {
 711         struct device *dev = acpi_desc->dev;
 712         struct nfit_memdev *nfit_memdev;
 713
 714         if (memdev->header.length != sizeof(*memdev))
 715                 return false;
 716
 717         list_for_each_entry(nfit_memdev, &prev->memdevs, list)
 718                 if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
 719                         list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 720                         return true;
 721                 }
 722
 723         nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
 724                         GFP_KERNEL);
 725         if (!nfit_memdev)
 726                 return false;
 727         INIT_LIST_HEAD(&nfit_memdev->list);
 728         memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
 729         list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 730         dev_dbg(dev, "memdev handle: %#x spa: %d dcr: %d flags: %#x\n",
 731                         memdev->device_handle, memdev->range_index,
 732                         memdev->region_index, memdev->flags);
 733         return true;
 734 }
 735
 736 int nfit_get_smbios_id(u32 device_handle, u16 *flags)
 737 {
 738         struct acpi_nfit_memory_map *memdev;
 739         struct acpi_nfit_desc *acpi_desc;
 740         struct nfit_mem *nfit_mem;
 741         u16 physical_id;
 742
 743         mutex_lock(&acpi_desc_lock);
 744         list_for_each_entry(acpi_desc, &acpi_descs, list) {
 745                 mutex_lock(&acpi_desc->init_mutex);
 746                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
 747                         memdev = __to_nfit_memdev(nfit_mem);
 748                         if (memdev->device_handle == device_handle) {
 749                                 *flags = memdev->flags;
 750                                 physical_id = memdev->physical_id;
 751                                 mutex_unlock(&acpi_desc->init_mutex);
 752                                 mutex_unlock(&acpi_desc_lock);
 753                                 return physical_id;
 754                         }
 755                 }
 756                 mutex_unlock(&acpi_desc->init_mutex);
 757         }
 758         mutex_unlock(&acpi_desc_lock);
 759
 760         return -ENODEV;
 761 }
 762 EXPORT_SYMBOL_GPL(nfit_get_smbios_id);
 763
 764 /*
 765  * An implementation may provide a truncated control region if no block windows
 766  * are defined.
 767  */
 768 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
 769 {
 770         if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
 771                                 window_size))
 772                 return 0;
 773         if (dcr->windows)
 774                 return sizeof(*dcr);
 775         return offsetof(struct acpi_nfit_control_region, window_size);
 776 }
 777
 778 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
 779                 struct nfit_table_prev *prev,
 780                 struct acpi_nfit_control_region *dcr)
 781 {
 782         struct device *dev = acpi_desc->dev;
 783         struct nfit_dcr *nfit_dcr;
 784
 785         if (!sizeof_dcr(dcr))
 786                 return false;
 787
 788         list_for_each_entry(nfit_dcr, &prev->dcrs, list)
 789                 if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
 790                         list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 791                         return true;
 792                 }
 793
 794         nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
 795                         GFP_KERNEL);
 796         if (!nfit_dcr)
 797                 return false;
 798         INIT_LIST_HEAD(&nfit_dcr->list);
 799         memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
 800         list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 801         dev_dbg(dev, "dcr index: %d windows: %d\n",
 802                         dcr->region_index, dcr->windows);
 803         return true;
 804 }
 805
 806 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
 807                 struct nfit_table_prev *prev,
 808                 struct acpi_nfit_data_region *bdw)
 809 {
 810         struct device *dev = acpi_desc->dev;
 811         struct nfit_bdw *nfit_bdw;
 812
 813         if (bdw->header.length != sizeof(*bdw))
 814                 return false;
 815         list_for_each_entry(nfit_bdw, &prev->bdws, list)
 816                 if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
 817                         list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
 818                         return true;
 819                 }
 820
 821         nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
 822                         GFP_KERNEL);
 823         if (!nfit_bdw)
 824                 return false;
 825         INIT_LIST_HEAD(&nfit_bdw->list);
 826         memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
 827         list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
 828         dev_dbg(dev, "bdw dcr: %d windows: %d\n",
 829                         bdw->region_index, bdw->windows);
 830         return true;
 831 }
 832
 833 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
 834 {
 835         if (idt->header.length < sizeof(*idt))
 836                 return 0;
 837         return sizeof(*idt) + sizeof(u32) * (idt->line_count - 1);
 838 }
 839
 840 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
 841                 struct nfit_table_prev *prev,
 842                 struct acpi_nfit_interleave *idt)
 843 {
 844         struct device *dev = acpi_desc->dev;
 845         struct nfit_idt *nfit_idt;
 846
 847         if (!sizeof_idt(idt))
 848                 return false;
 849
 850         list_for_each_entry(nfit_idt, &prev->idts, list) {
 851                 if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
 852                         continue;
 853
 854                 if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
 855                         list_move_tail(&nfit_idt->list, &acpi_desc->idts);
 856                         return true;
 857                 }
 858         }
 859
 860         nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
 861                         GFP_KERNEL);
 862         if (!nfit_idt)
 863                 return false;
 864         INIT_LIST_HEAD(&nfit_idt->list);
 865         memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
 866         list_add_tail(&nfit_idt->list, &acpi_desc->idts);
 867         dev_dbg(dev, "idt index: %d num_lines: %d\n",
 868                         idt->interleave_index, idt->line_count);
 869         return true;
 870 }
 871
 872 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
 873 {
 874         if (flush->header.length < sizeof(*flush))
 875                 return 0;
 876         return sizeof(*flush) + sizeof(u64) * (flush->hint_count - 1);
 877 }
 878
 879 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
 880                 struct nfit_table_prev *prev,
 881                 struct acpi_nfit_flush_address *flush)
 882 {
 883         struct device *dev = acpi_desc->dev;
 884         struct nfit_flush *nfit_flush;
 885
 886         if (!sizeof_flush(flush))
 887                 return false;
 888
 889         list_for_each_entry(nfit_flush, &prev->flushes, list) {
 890                 if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
 891                         continue;
 892
 893                 if (memcmp(nfit_flush->flush, flush,
 894                                         sizeof_flush(flush)) == 0) {
 895                         list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
 896                         return true;
 897                 }
 898         }
 899
 900         nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
 901                         + sizeof_flush(flush), GFP_KERNEL);
 902         if (!nfit_flush)
 903                 return false;
 904         INIT_LIST_HEAD(&nfit_flush->list);
 905         memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
 906         list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
 907         dev_dbg(dev, "nfit_flush handle: %d hint_count: %d\n",
 908                         flush->device_handle, flush->hint_count);
 909         return true;
 910 }
 911
 912 static bool add_platform_cap(struct acpi_nfit_desc *acpi_desc,
 913                 struct acpi_nfit_capabilities *pcap)
 914 {
 915         struct device *dev = acpi_desc->dev;
 916         u32 mask;
 917
 918         mask = (1 << (pcap->highest_capability + 1)) - 1;
 919         acpi_desc->platform_cap = pcap->capabilities & mask;
 920         dev_dbg(dev, "cap: %#x\n", acpi_desc->platform_cap);
 921         return true;
 922 }
 923
 924 static void *add_table(struct acpi_nfit_desc *acpi_desc,
 925                 struct nfit_table_prev *prev, void *table, const void *end)
 926 {
 927         struct device *dev = acpi_desc->dev;
 928         struct acpi_nfit_header *hdr;
 929         void *err = ERR_PTR(-ENOMEM);
 930
 931         if (table >= end)
 932                 return NULL;
 933
 934         hdr = table;
 935         if (!hdr->length) {
 936                 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
 937                         hdr->type);
 938                 return NULL;
 939         }
 940
 941         switch (hdr->type) {
 942         case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
 943                 if (!add_spa(acpi_desc, prev, table))
 944                         return err;
 945                 break;
 946         case ACPI_NFIT_TYPE_MEMORY_MAP:
 947                 if (!add_memdev(acpi_desc, prev, table))
 948                         return err;
 949                 break;
 950         case ACPI_NFIT_TYPE_CONTROL_REGION:
 951                 if (!add_dcr(acpi_desc, prev, table))
 952                         return err;
 953                 break;
 954         case ACPI_NFIT_TYPE_DATA_REGION:
 955                 if (!add_bdw(acpi_desc, prev, table))
 956                         return err;
 957                 break;
 958         case ACPI_NFIT_TYPE_INTERLEAVE:
 959                 if (!add_idt(acpi_desc, prev, table))
 960                         return err;
 961                 break;
 962         case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
 963                 if (!add_flush(acpi_desc, prev, table))
 964                         return err;
 965                 break;
 966         case ACPI_NFIT_TYPE_SMBIOS:
 967                 dev_dbg(dev, "smbios\n");
 968                 break;
 969         case ACPI_NFIT_TYPE_CAPABILITIES:
 970                 if (!add_platform_cap(acpi_desc, table))
 971                         return err;
 972                 break;
 973         default:
 974                 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
 975                 break;
 976         }
 977
 978         return table + hdr->length;
 979 }
 980
 981 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
 982                 struct nfit_mem *nfit_mem)
 983 {
 984         u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
 985         u16 dcr = nfit_mem->dcr->region_index;
 986         struct nfit_spa *nfit_spa;
 987
 988         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
 989                 u16 range_index = nfit_spa->spa->range_index;
 990                 int type = nfit_spa_type(nfit_spa->spa);
 991                 struct nfit_memdev *nfit_memdev;
 992
 993                 if (type != NFIT_SPA_BDW)
 994                         continue;
 995
 996                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 997                         if (nfit_memdev->memdev->range_index != range_index)
 998                                 continue;
 999                         if (nfit_memdev->memdev->device_handle != device_handle)
1000                                 continue;
1001                         if (nfit_memdev->memdev->region_index != dcr)
1002                                 continue;
1003
1004                         nfit_mem->spa_bdw = nfit_spa->spa;
1005                         return;
1006                 }
1007         }
1008
1009         dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
1010                         nfit_mem->spa_dcr->range_index);
1011         nfit_mem->bdw = NULL;
1012 }
1013
1014 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
1015                 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
1016 {
1017         u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
1018         struct nfit_memdev *nfit_memdev;
1019         struct nfit_bdw *nfit_bdw;
1020         struct nfit_idt *nfit_idt;
1021         u16 idt_idx, range_index;
1022
1023         list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
1024                 if (nfit_bdw->bdw->region_index != dcr)
1025                         continue;
1026                 nfit_mem->bdw = nfit_bdw->bdw;
1027                 break;
1028         }
1029
1030         if (!nfit_mem->bdw)
1031                 return;
1032
1033         nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
1034
1035         if (!nfit_mem->spa_bdw)
1036                 return;
1037
1038         range_index = nfit_mem->spa_bdw->range_index;
1039         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1040                 if (nfit_memdev->memdev->range_index != range_index ||
1041                                 nfit_memdev->memdev->region_index != dcr)
1042                         continue;
1043                 nfit_mem->memdev_bdw = nfit_memdev->memdev;
1044                 idt_idx = nfit_memdev->memdev->interleave_index;
1045                 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1046                         if (nfit_idt->idt->interleave_index != idt_idx)
1047                                 continue;
1048                         nfit_mem->idt_bdw = nfit_idt->idt;
1049                         break;
1050                 }
1051                 break;
1052         }
1053 }
1054
1055 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
1056                 struct acpi_nfit_system_address *spa)
1057 {
1058         struct nfit_mem *nfit_mem, *found;
1059         struct nfit_memdev *nfit_memdev;
1060         int type = spa ? nfit_spa_type(spa) : 0;
1061
1062         switch (type) {
1063         case NFIT_SPA_DCR:
1064         case NFIT_SPA_PM:
1065                 break;
1066         default:
1067                 if (spa)
1068                         return 0;
1069         }
1070
1071         /*
1072          * This loop runs in two modes, when a dimm is mapped the loop
1073          * adds memdev associations to an existing dimm, or creates a
1074          * dimm. In the unmapped dimm case this loop sweeps for memdev
1075          * instances with an invalid / zero range_index and adds those
1076          * dimms without spa associations.
1077          */
1078         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1079                 struct nfit_flush *nfit_flush;
1080                 struct nfit_dcr *nfit_dcr;
1081                 u32 device_handle;
1082                 u16 dcr;
1083
1084                 if (spa && nfit_memdev->memdev->range_index != spa->range_index)
1085                         continue;
1086                 if (!spa && nfit_memdev->memdev->range_index)
1087                         continue;
1088                 found = NULL;
1089                 dcr = nfit_memdev->memdev->region_index;
1090                 device_handle = nfit_memdev->memdev->device_handle;
1091                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1092                         if (__to_nfit_memdev(nfit_mem)->device_handle
1093                                         == device_handle) {
1094                                 found = nfit_mem;
1095                                 break;
1096                         }
1097
1098                 if (found)
1099                         nfit_mem = found;
1100                 else {
1101                         nfit_mem = devm_kzalloc(acpi_desc->dev,
1102                                         sizeof(*nfit_mem), GFP_KERNEL);
1103                         if (!nfit_mem)
1104                                 return -ENOMEM;
1105                         INIT_LIST_HEAD(&nfit_mem->list);
1106                         nfit_mem->acpi_desc = acpi_desc;
1107                         list_add(&nfit_mem->list, &acpi_desc->dimms);
1108                 }
1109
1110                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1111                         if (nfit_dcr->dcr->region_index != dcr)
1112                                 continue;
1113                         /*
1114                          * Record the control region for the dimm.  For
1115                          * the ACPI 6.1 case, where there are separate
1116                          * control regions for the pmem vs blk
1117                          * interfaces, be sure to record the extended
1118                          * blk details.
1119                          */
1120                         if (!nfit_mem->dcr)
1121                                 nfit_mem->dcr = nfit_dcr->dcr;
1122                         else if (nfit_mem->dcr->windows == 0
1123                                         && nfit_dcr->dcr->windows)
1124                                 nfit_mem->dcr = nfit_dcr->dcr;
1125                         break;
1126                 }
1127
1128                 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
1129                         struct acpi_nfit_flush_address *flush;
1130                         u16 i;
1131
1132                         if (nfit_flush->flush->device_handle != device_handle)
1133                                 continue;
1134                         nfit_mem->nfit_flush = nfit_flush;
1135                         flush = nfit_flush->flush;
1136                         nfit_mem->flush_wpq = devm_kcalloc(acpi_desc->dev,
1137                                         flush->hint_count,
1138                                         sizeof(struct resource),
1139                                         GFP_KERNEL);
1140                         if (!nfit_mem->flush_wpq)
1141                                 return -ENOMEM;
1142                         for (i = 0; i < flush->hint_count; i++) {
1143                                 struct resource *res = &nfit_mem->flush_wpq[i];
1144
1145                                 res->start = flush->hint_address[i];
1146                                 res->end = res->start + 8 - 1;
1147                         }
1148                         break;
1149                 }
1150
1151                 if (dcr && !nfit_mem->dcr) {
1152                         dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
1153                                         spa->range_index, dcr);
1154                         return -ENODEV;
1155                 }
1156
1157                 if (type == NFIT_SPA_DCR) {
1158                         struct nfit_idt *nfit_idt;
1159                         u16 idt_idx;
1160
1161                         /* multiple dimms may share a SPA when interleaved */
1162                         nfit_mem->spa_dcr = spa;
1163                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
1164                         idt_idx = nfit_memdev->memdev->interleave_index;
1165                         list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1166                                 if (nfit_idt->idt->interleave_index != idt_idx)
1167                                         continue;
1168                                 nfit_mem->idt_dcr = nfit_idt->idt;
1169                                 break;
1170                         }
1171                         nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
1172                 } else if (type == NFIT_SPA_PM) {
1173                         /*
1174                          * A single dimm may belong to multiple SPA-PM
1175                          * ranges, record at least one in addition to
1176                          * any SPA-DCR range.
1177                          */
1178                         nfit_mem->memdev_pmem = nfit_memdev->memdev;
1179                 } else
1180                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
1181         }
1182
1183         return 0;
1184 }
1185
1186 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
1187 {
1188         struct nfit_mem *a = container_of(_a, typeof(*a), list);
1189         struct nfit_mem *b = container_of(_b, typeof(*b), list);
1190         u32 handleA, handleB;
1191
1192         handleA = __to_nfit_memdev(a)->device_handle;
1193         handleB = __to_nfit_memdev(b)->device_handle;
1194         if (handleA < handleB)
1195                 return -1;
1196         else if (handleA > handleB)
1197                 return 1;
1198         return 0;
1199 }
1200
1201 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
1202 {
1203         struct nfit_spa *nfit_spa;
1204         int rc;
1205
1206
1207         /*
1208          * For each SPA-DCR or SPA-PMEM address range find its
1209          * corresponding MEMDEV(s).  From each MEMDEV find the
1210          * corresponding DCR.  Then, if we're operating on a SPA-DCR,
1211          * try to find a SPA-BDW and a corresponding BDW that references
1212          * the DCR.  Throw it all into an nfit_mem object.  Note, that
1213          * BDWs are optional.
1214          */
1215         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
1216                 rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
1217                 if (rc)
1218                         return rc;
1219         }
1220
1221         /*
1222          * If a DIMM has failed to be mapped into SPA there will be no
1223          * SPA entries above. Find and register all the unmapped DIMMs
1224          * for reporting and recovery purposes.
1225          */
1226         rc = __nfit_mem_init(acpi_desc, NULL);
1227         if (rc)
1228                 return rc;
1229
1230         list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
1231
1232         return 0;
1233 }
1234
1235 static ssize_t bus_dsm_mask_show(struct device *dev,
1236                 struct device_attribute *attr, char *buf)
1237 {
1238         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1239         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1240
1241         return sprintf(buf, "%#lx\n", nd_desc->bus_dsm_mask);
1242 }
1243 static struct device_attribute dev_attr_bus_dsm_mask =
1244                 __ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
1245
1246 static ssize_t revision_show(struct device *dev,
1247                 struct device_attribute *attr, char *buf)
1248 {
1249         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1250         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1251         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1252
1253         return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
1254 }
1255 static DEVICE_ATTR_RO(revision);
1256
1257 static ssize_t hw_error_scrub_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, "%d\n", acpi_desc->scrub_mode);
1265 }
1266
1267 /*
1268  * The 'hw_error_scrub' attribute can have the following values written to it:
1269  * '0': Switch to the default mode where an exception will only insert
1270  *      the address of the memory error into the poison and badblocks lists.
1271  * '1': Enable a full scrub to happen if an exception for a memory error is
1272  *      received.
1273  */
1274 static ssize_t hw_error_scrub_store(struct device *dev,
1275                 struct device_attribute *attr, const char *buf, size_t size)
1276 {
1277         struct nvdimm_bus_descriptor *nd_desc;
1278         ssize_t rc;
1279         long val;
1280
1281         rc = kstrtol(buf, 0, &val);
1282         if (rc)
1283                 return rc;
1284
1285         nfit_device_lock(dev);
1286         nd_desc = dev_get_drvdata(dev);
1287         if (nd_desc) {
1288                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1289
1290                 switch (val) {
1291                 case HW_ERROR_SCRUB_ON:
1292                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
1293                         break;
1294                 case HW_ERROR_SCRUB_OFF:
1295                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
1296                         break;
1297                 default:
1298                         rc = -EINVAL;
1299                         break;
1300                 }
1301         }
1302         nfit_device_unlock(dev);
1303         if (rc)
1304                 return rc;
1305         return size;
1306 }
1307 static DEVICE_ATTR_RW(hw_error_scrub);
1308
1309 /*
1310  * This shows the number of full Address Range Scrubs that have been
1311  * completed since driver load time. Userspace can wait on this using
1312  * select/poll etc. A '+' at the end indicates an ARS is in progress
1313  */
1314 static ssize_t scrub_show(struct device *dev,
1315                 struct device_attribute *attr, char *buf)
1316 {
1317         struct nvdimm_bus_descriptor *nd_desc;
1318         struct acpi_nfit_desc *acpi_desc;
1319         ssize_t rc = -ENXIO;
1320         bool busy;
1321
1322         nfit_device_lock(dev);
1323         nd_desc = dev_get_drvdata(dev);
1324         if (!nd_desc) {
1325                 nfit_device_unlock(dev);
1326                 return rc;
1327         }
1328         acpi_desc = to_acpi_desc(nd_desc);
1329
1330         mutex_lock(&acpi_desc->init_mutex);
1331         busy = test_bit(ARS_BUSY, &acpi_desc->scrub_flags)
1332                 && !test_bit(ARS_CANCEL, &acpi_desc->scrub_flags);
1333         rc = sprintf(buf, "%d%s", acpi_desc->scrub_count, busy ? "+\n" : "\n");
1334         /* Allow an admin to poll the busy state at a higher rate */
1335         if (busy && capable(CAP_SYS_RAWIO) && !test_and_set_bit(ARS_POLL,
1336                                 &acpi_desc->scrub_flags)) {
1337                 acpi_desc->scrub_tmo = 1;
1338                 mod_delayed_work(nfit_wq, &acpi_desc->dwork, HZ);
1339         }
1340
1341         mutex_unlock(&acpi_desc->init_mutex);
1342         nfit_device_unlock(dev);
1343         return rc;
1344 }
1345
1346 static ssize_t scrub_store(struct device *dev,
1347                 struct device_attribute *attr, const char *buf, size_t size)
1348 {
1349         struct nvdimm_bus_descriptor *nd_desc;
1350         ssize_t rc;
1351         long val;
1352
1353         rc = kstrtol(buf, 0, &val);
1354         if (rc)
1355                 return rc;
1356         if (val != 1)
1357                 return -EINVAL;
1358
1359         nfit_device_lock(dev);
1360         nd_desc = dev_get_drvdata(dev);
1361         if (nd_desc) {
1362                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1363
1364                 rc = acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
1365         }
1366         nfit_device_unlock(dev);
1367         if (rc)
1368                 return rc;
1369         return size;
1370 }
1371 static DEVICE_ATTR_RW(scrub);
1372
1373 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1374 {
1375         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1376         const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1377                 | 1 << ND_CMD_ARS_STATUS;
1378
1379         return (nd_desc->cmd_mask & mask) == mask;
1380 }
1381
1382 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1383 {
1384         struct device *dev = container_of(kobj, struct device, kobj);
1385         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1386
1387         if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
1388                 return 0;
1389         return a->mode;
1390 }
1391
1392 static struct attribute *acpi_nfit_attributes[] = {
1393         &dev_attr_revision.attr,
1394         &dev_attr_scrub.attr,
1395         &dev_attr_hw_error_scrub.attr,
1396         &dev_attr_bus_dsm_mask.attr,
1397         NULL,
1398 };
1399
1400 static const struct attribute_group acpi_nfit_attribute_group = {
1401         .name = "nfit",
1402         .attrs = acpi_nfit_attributes,
1403         .is_visible = nfit_visible,
1404 };
1405
1406 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1407         &acpi_nfit_attribute_group,
1408         NULL,
1409 };
1410
1411 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1412 {
1413         struct nvdimm *nvdimm = to_nvdimm(dev);
1414         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1415
1416         return __to_nfit_memdev(nfit_mem);
1417 }
1418
1419 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1420 {
1421         struct nvdimm *nvdimm = to_nvdimm(dev);
1422         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1423
1424         return nfit_mem->dcr;
1425 }
1426
1427 static ssize_t handle_show(struct device *dev,
1428                 struct device_attribute *attr, char *buf)
1429 {
1430         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1431
1432         return sprintf(buf, "%#x\n", memdev->device_handle);
1433 }
1434 static DEVICE_ATTR_RO(handle);
1435
1436 static ssize_t phys_id_show(struct device *dev,
1437                 struct device_attribute *attr, char *buf)
1438 {
1439         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1440
1441         return sprintf(buf, "%#x\n", memdev->physical_id);
1442 }
1443 static DEVICE_ATTR_RO(phys_id);
1444
1445 static ssize_t vendor_show(struct device *dev,
1446                 struct device_attribute *attr, char *buf)
1447 {
1448         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1449
1450         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1451 }
1452 static DEVICE_ATTR_RO(vendor);
1453
1454 static ssize_t rev_id_show(struct device *dev,
1455                 struct device_attribute *attr, char *buf)
1456 {
1457         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1458
1459         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1460 }
1461 static DEVICE_ATTR_RO(rev_id);
1462
1463 static ssize_t device_show(struct device *dev,
1464                 struct device_attribute *attr, char *buf)
1465 {
1466         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1467
1468         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1469 }
1470 static DEVICE_ATTR_RO(device);
1471
1472 static ssize_t subsystem_vendor_show(struct device *dev,
1473                 struct device_attribute *attr, char *buf)
1474 {
1475         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1476
1477         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1478 }
1479 static DEVICE_ATTR_RO(subsystem_vendor);
1480
1481 static ssize_t subsystem_rev_id_show(struct device *dev,
1482                 struct device_attribute *attr, char *buf)
1483 {
1484         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1485
1486         return sprintf(buf, "0x%04x\n",
1487                         be16_to_cpu(dcr->subsystem_revision_id));
1488 }
1489 static DEVICE_ATTR_RO(subsystem_rev_id);
1490
1491 static ssize_t subsystem_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->subsystem_device_id));
1497 }
1498 static DEVICE_ATTR_RO(subsystem_device);
1499
1500 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1501 {
1502         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1503         int formats = 0;
1504
1505         if (nfit_mem->memdev_pmem)
1506                 formats++;
1507         if (nfit_mem->memdev_bdw)
1508                 formats++;
1509         return formats;
1510 }
1511
1512 static ssize_t format_show(struct device *dev,
1513                 struct device_attribute *attr, char *buf)
1514 {
1515         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1516
1517         return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1518 }
1519 static DEVICE_ATTR_RO(format);
1520
1521 static ssize_t format1_show(struct device *dev,
1522                 struct device_attribute *attr, char *buf)
1523 {
1524         u32 handle;
1525         ssize_t rc = -ENXIO;
1526         struct nfit_mem *nfit_mem;
1527         struct nfit_memdev *nfit_memdev;
1528         struct acpi_nfit_desc *acpi_desc;
1529         struct nvdimm *nvdimm = to_nvdimm(dev);
1530         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1531
1532         nfit_mem = nvdimm_provider_data(nvdimm);
1533         acpi_desc = nfit_mem->acpi_desc;
1534         handle = to_nfit_memdev(dev)->device_handle;
1535
1536         /* assumes DIMMs have at most 2 published interface codes */
1537         mutex_lock(&acpi_desc->init_mutex);
1538         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1539                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1540                 struct nfit_dcr *nfit_dcr;
1541
1542                 if (memdev->device_handle != handle)
1543                         continue;
1544
1545                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1546                         if (nfit_dcr->dcr->region_index != memdev->region_index)
1547                                 continue;
1548                         if (nfit_dcr->dcr->code == dcr->code)
1549                                 continue;
1550                         rc = sprintf(buf, "0x%04x\n",
1551                                         le16_to_cpu(nfit_dcr->dcr->code));
1552                         break;
1553                 }
1554                 if (rc != ENXIO)
1555                         break;
1556         }
1557         mutex_unlock(&acpi_desc->init_mutex);
1558         return rc;
1559 }
1560 static DEVICE_ATTR_RO(format1);
1561
1562 static ssize_t formats_show(struct device *dev,
1563                 struct device_attribute *attr, char *buf)
1564 {
1565         struct nvdimm *nvdimm = to_nvdimm(dev);
1566
1567         return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1568 }
1569 static DEVICE_ATTR_RO(formats);
1570
1571 static ssize_t serial_show(struct device *dev,
1572                 struct device_attribute *attr, char *buf)
1573 {
1574         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1575
1576         return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1577 }
1578 static DEVICE_ATTR_RO(serial);
1579
1580 static ssize_t family_show(struct device *dev,
1581                 struct device_attribute *attr, char *buf)
1582 {
1583         struct nvdimm *nvdimm = to_nvdimm(dev);
1584         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1585
1586         if (nfit_mem->family < 0)
1587                 return -ENXIO;
1588         return sprintf(buf, "%d\n", nfit_mem->family);
1589 }
1590 static DEVICE_ATTR_RO(family);
1591
1592 static ssize_t dsm_mask_show(struct device *dev,
1593                 struct device_attribute *attr, char *buf)
1594 {
1595         struct nvdimm *nvdimm = to_nvdimm(dev);
1596         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1597
1598         if (nfit_mem->family < 0)
1599                 return -ENXIO;
1600         return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1601 }
1602 static DEVICE_ATTR_RO(dsm_mask);
1603
1604 static ssize_t flags_show(struct device *dev,
1605                 struct device_attribute *attr, char *buf)
1606 {
1607         struct nvdimm *nvdimm = to_nvdimm(dev);
1608         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1609         u16 flags = __to_nfit_memdev(nfit_mem)->flags;
1610
1611         if (test_bit(NFIT_MEM_DIRTY, &nfit_mem->flags))
1612                 flags |= ACPI_NFIT_MEM_FLUSH_FAILED;
1613
1614         return sprintf(buf, "%s%s%s%s%s%s%s\n",
1615                 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1616                 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1617                 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1618                 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1619                 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1620                 flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1621                 flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1622 }
1623 static DEVICE_ATTR_RO(flags);
1624
1625 static ssize_t id_show(struct device *dev,
1626                 struct device_attribute *attr, char *buf)
1627 {
1628         struct nvdimm *nvdimm = to_nvdimm(dev);
1629         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1630
1631         return sprintf(buf, "%s\n", nfit_mem->id);
1632 }
1633 static DEVICE_ATTR_RO(id);
1634
1635 static ssize_t dirty_shutdown_show(struct device *dev,
1636                 struct device_attribute *attr, char *buf)
1637 {
1638         struct nvdimm *nvdimm = to_nvdimm(dev);
1639         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1640
1641         return sprintf(buf, "%d\n", nfit_mem->dirty_shutdown);
1642 }
1643 static DEVICE_ATTR_RO(dirty_shutdown);
1644
1645 static struct attribute *acpi_nfit_dimm_attributes[] = {
1646         &dev_attr_handle.attr,
1647         &dev_attr_phys_id.attr,
1648         &dev_attr_vendor.attr,
1649         &dev_attr_device.attr,
1650         &dev_attr_rev_id.attr,
1651         &dev_attr_subsystem_vendor.attr,
1652         &dev_attr_subsystem_device.attr,
1653         &dev_attr_subsystem_rev_id.attr,
1654         &dev_attr_format.attr,
1655         &dev_attr_formats.attr,
1656         &dev_attr_format1.attr,
1657         &dev_attr_serial.attr,
1658         &dev_attr_flags.attr,
1659         &dev_attr_id.attr,
1660         &dev_attr_family.attr,
1661         &dev_attr_dsm_mask.attr,
1662         &dev_attr_dirty_shutdown.attr,
1663         NULL,
1664 };
1665
1666 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1667                 struct attribute *a, int n)
1668 {
1669         struct device *dev = container_of(kobj, struct device, kobj);
1670         struct nvdimm *nvdimm = to_nvdimm(dev);
1671         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1672
1673         if (!to_nfit_dcr(dev)) {
1674                 /* Without a dcr only the memdev attributes can be surfaced */
1675                 if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1676                                 || a == &dev_attr_flags.attr
1677                                 || a == &dev_attr_family.attr
1678                                 || a == &dev_attr_dsm_mask.attr)
1679                         return a->mode;
1680                 return 0;
1681         }
1682
1683         if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1684                 return 0;
1685
1686         if (!test_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags)
1687                         && a == &dev_attr_dirty_shutdown.attr)
1688                 return 0;
1689
1690         return a->mode;
1691 }
1692
1693 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1694         .name = "nfit",
1695         .attrs = acpi_nfit_dimm_attributes,
1696         .is_visible = acpi_nfit_dimm_attr_visible,
1697 };
1698
1699 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1700         &acpi_nfit_dimm_attribute_group,
1701         NULL,
1702 };
1703
1704 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1705                 u32 device_handle)
1706 {
1707         struct nfit_mem *nfit_mem;
1708
1709         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1710                 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1711                         return nfit_mem->nvdimm;
1712
1713         return NULL;
1714 }
1715
1716 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1717 {
1718         struct nfit_mem *nfit_mem;
1719         struct acpi_nfit_desc *acpi_desc;
1720
1721         dev_dbg(dev->parent, "%s: event: %d\n", dev_name(dev),
1722                         event);
1723
1724         if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1725                 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1726                                 event);
1727                 return;
1728         }
1729
1730         acpi_desc = dev_get_drvdata(dev->parent);
1731         if (!acpi_desc)
1732                 return;
1733
1734         /*
1735          * If we successfully retrieved acpi_desc, then we know nfit_mem data
1736          * is still valid.
1737          */
1738         nfit_mem = dev_get_drvdata(dev);
1739         if (nfit_mem && nfit_mem->flags_attr)
1740                 sysfs_notify_dirent(nfit_mem->flags_attr);
1741 }
1742 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1743
1744 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1745 {
1746         struct acpi_device *adev = data;
1747         struct device *dev = &adev->dev;
1748
1749         nfit_device_lock(dev->parent);
1750         __acpi_nvdimm_notify(dev, event);
1751         nfit_device_unlock(dev->parent);
1752 }
1753
1754 static bool acpi_nvdimm_has_method(struct acpi_device *adev, char *method)
1755 {
1756         acpi_handle handle;
1757         acpi_status status;
1758
1759         status = acpi_get_handle(adev->handle, method, &handle);
1760
1761         if (ACPI_SUCCESS(status))
1762                 return true;
1763         return false;
1764 }
1765
1766 __weak void nfit_intel_shutdown_status(struct nfit_mem *nfit_mem)
1767 {
1768         struct device *dev = &nfit_mem->adev->dev;
1769         struct nd_intel_smart smart = { 0 };
1770         union acpi_object in_buf = {
1771                 .buffer.type = ACPI_TYPE_BUFFER,
1772                 .buffer.length = 0,
1773         };
1774         union acpi_object in_obj = {
1775                 .package.type = ACPI_TYPE_PACKAGE,
1776                 .package.count = 1,
1777                 .package.elements = &in_buf,
1778         };
1779         const u8 func = ND_INTEL_SMART;
1780         const guid_t *guid = to_nfit_uuid(nfit_mem->family);
1781         u8 revid = nfit_dsm_revid(nfit_mem->family, func);
1782         struct acpi_device *adev = nfit_mem->adev;
1783         acpi_handle handle = adev->handle;
1784         union acpi_object *out_obj;
1785
1786         if ((nfit_mem->dsm_mask & (1 << func)) == 0)
1787                 return;
1788
1789         out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
1790         if (!out_obj || out_obj->type != ACPI_TYPE_BUFFER
1791                         || out_obj->buffer.length < sizeof(smart)) {
1792                 dev_dbg(dev->parent, "%s: failed to retrieve initial health\n",
1793                                 dev_name(dev));
1794                 ACPI_FREE(out_obj);
1795                 return;
1796         }
1797         memcpy(&smart, out_obj->buffer.pointer, sizeof(smart));
1798         ACPI_FREE(out_obj);
1799
1800         if (smart.flags & ND_INTEL_SMART_SHUTDOWN_VALID) {
1801                 if (smart.shutdown_state)
1802                         set_bit(NFIT_MEM_DIRTY, &nfit_mem->flags);
1803         }
1804
1805         if (smart.flags & ND_INTEL_SMART_SHUTDOWN_COUNT_VALID) {
1806                 set_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags);
1807                 nfit_mem->dirty_shutdown = smart.shutdown_count;
1808         }
1809 }
1810
1811 static void populate_shutdown_status(struct nfit_mem *nfit_mem)
1812 {
1813         /*
1814          * For DIMMs that provide a dynamic facility to retrieve a
1815          * dirty-shutdown status and/or a dirty-shutdown count, cache
1816          * these values in nfit_mem.
1817          */
1818         if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
1819                 nfit_intel_shutdown_status(nfit_mem);
1820 }
1821
1822 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1823                 struct nfit_mem *nfit_mem, u32 device_handle)
1824 {
1825         struct acpi_device *adev, *adev_dimm;
1826         struct device *dev = acpi_desc->dev;
1827         unsigned long dsm_mask, label_mask;
1828         const guid_t *guid;
1829         int i;
1830         int family = -1;
1831         struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
1832
1833         /* nfit test assumes 1:1 relationship between commands and dsms */
1834         nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1835         nfit_mem->family = NVDIMM_FAMILY_INTEL;
1836
1837         if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1838                 sprintf(nfit_mem->id, "%04x-%02x-%04x-%08x",
1839                                 be16_to_cpu(dcr->vendor_id),
1840                                 dcr->manufacturing_location,
1841                                 be16_to_cpu(dcr->manufacturing_date),
1842                                 be32_to_cpu(dcr->serial_number));
1843         else
1844                 sprintf(nfit_mem->id, "%04x-%08x",
1845                                 be16_to_cpu(dcr->vendor_id),
1846                                 be32_to_cpu(dcr->serial_number));
1847
1848         adev = to_acpi_dev(acpi_desc);
1849         if (!adev) {
1850                 /* unit test case */
1851                 populate_shutdown_status(nfit_mem);
1852                 return 0;
1853         }
1854
1855         adev_dimm = acpi_find_child_device(adev, device_handle, false);
1856         nfit_mem->adev = adev_dimm;
1857         if (!adev_dimm) {
1858                 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1859                                 device_handle);
1860                 return force_enable_dimms ? 0 : -ENODEV;
1861         }
1862
1863         if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1864                 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1865                 dev_err(dev, "%s: notification registration failed\n",
1866                                 dev_name(&adev_dimm->dev));
1867                 return -ENXIO;
1868         }
1869         /*
1870          * Record nfit_mem for the notification path to track back to
1871          * the nfit sysfs attributes for this dimm device object.
1872          */
1873         dev_set_drvdata(&adev_dimm->dev, nfit_mem);
1874
1875         /*
1876          * There are 4 "legacy" NVDIMM command sets
1877          * (NVDIMM_FAMILY_{INTEL,MSFT,HPE1,HPE2}) that were created before
1878          * an EFI working group was established to constrain this
1879          * proliferation. The nfit driver probes for the supported command
1880          * set by GUID. Note, if you're a platform developer looking to add
1881          * a new command set to this probe, consider using an existing set,
1882          * or otherwise seek approval to publish the command set at
1883          * http://www.uefi.org/RFIC_LIST.
1884          *
1885          * Note, that checking for function0 (bit0) tells us if any commands
1886          * are reachable through this GUID.
1887          */
1888         for (i = 0; i <= NVDIMM_FAMILY_MAX; i++)
1889                 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1890                         if (family < 0 || i == default_dsm_family)
1891                                 family = i;
1892
1893         /* limit the supported commands to those that are publicly documented */
1894         nfit_mem->family = family;
1895         if (override_dsm_mask && !disable_vendor_specific)
1896                 dsm_mask = override_dsm_mask;
1897         else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1898                 dsm_mask = NVDIMM_INTEL_CMDMASK;
1899                 if (disable_vendor_specific)
1900                         dsm_mask &= ~(1 << ND_CMD_VENDOR);
1901         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1902                 dsm_mask = 0x1c3c76;
1903         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1904                 dsm_mask = 0x1fe;
1905                 if (disable_vendor_specific)
1906                         dsm_mask &= ~(1 << 8);
1907         } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1908                 dsm_mask = 0xffffffff;
1909         } else if (nfit_mem->family == NVDIMM_FAMILY_HYPERV) {
1910                 dsm_mask = 0x1f;
1911         } else {
1912                 dev_dbg(dev, "unknown dimm command family\n");
1913                 nfit_mem->family = -1;
1914                 /* DSMs are optional, continue loading the driver... */
1915                 return 0;
1916         }
1917
1918         /*
1919          * Function 0 is the command interrogation function, don't
1920          * export it to potential userspace use, and enable it to be
1921          * used as an error value in acpi_nfit_ctl().
1922          */
1923         dsm_mask &= ~1UL;
1924
1925         guid = to_nfit_uuid(nfit_mem->family);
1926         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1927                 if (acpi_check_dsm(adev_dimm->handle, guid,
1928                                         nfit_dsm_revid(nfit_mem->family, i),
1929                                         1ULL << i))
1930                         set_bit(i, &nfit_mem->dsm_mask);
1931
1932         /*
1933          * Prefer the NVDIMM_FAMILY_INTEL label read commands if present
1934          * due to their better semantics handling locked capacity.
1935          */
1936         label_mask = 1 << ND_CMD_GET_CONFIG_SIZE | 1 << ND_CMD_GET_CONFIG_DATA
1937                 | 1 << ND_CMD_SET_CONFIG_DATA;
1938         if (family == NVDIMM_FAMILY_INTEL
1939                         && (dsm_mask & label_mask) == label_mask)
1940                 /* skip _LS{I,R,W} enabling */;
1941         else {
1942                 if (acpi_nvdimm_has_method(adev_dimm, "_LSI")
1943                                 && acpi_nvdimm_has_method(adev_dimm, "_LSR")) {
1944                         dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev));
1945                         set_bit(NFIT_MEM_LSR, &nfit_mem->flags);
1946                 }
1947
1948                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
1949                                 && acpi_nvdimm_has_method(adev_dimm, "_LSW")) {
1950                         dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev));
1951                         set_bit(NFIT_MEM_LSW, &nfit_mem->flags);
1952                 }
1953
1954                 /*
1955                  * Quirk read-only label configurations to preserve
1956                  * access to label-less namespaces by default.
1957                  */
1958                 if (!test_bit(NFIT_MEM_LSW, &nfit_mem->flags)
1959                                 && !force_labels) {
1960                         dev_dbg(dev, "%s: No _LSW, disable labels\n",
1961                                         dev_name(&adev_dimm->dev));
1962                         clear_bit(NFIT_MEM_LSR, &nfit_mem->flags);
1963                 } else
1964                         dev_dbg(dev, "%s: Force enable labels\n",
1965                                         dev_name(&adev_dimm->dev));
1966         }
1967
1968         populate_shutdown_status(nfit_mem);
1969
1970         return 0;
1971 }
1972
1973 static void shutdown_dimm_notify(void *data)
1974 {
1975         struct acpi_nfit_desc *acpi_desc = data;
1976         struct nfit_mem *nfit_mem;
1977
1978         mutex_lock(&acpi_desc->init_mutex);
1979         /*
1980          * Clear out the nfit_mem->flags_attr and shut down dimm event
1981          * notifications.
1982          */
1983         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1984                 struct acpi_device *adev_dimm = nfit_mem->adev;
1985
1986                 if (nfit_mem->flags_attr) {
1987                         sysfs_put(nfit_mem->flags_attr);
1988                         nfit_mem->flags_attr = NULL;
1989                 }
1990                 if (adev_dimm) {
1991                         acpi_remove_notify_handler(adev_dimm->handle,
1992                                         ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
1993                         dev_set_drvdata(&adev_dimm->dev, NULL);
1994                 }
1995         }
1996         mutex_unlock(&acpi_desc->init_mutex);
1997 }
1998
1999 static const struct nvdimm_security_ops *acpi_nfit_get_security_ops(int family)
2000 {
2001         switch (family) {
2002         case NVDIMM_FAMILY_INTEL:
2003                 return intel_security_ops;
2004         default:
2005                 return NULL;
2006         }
2007 }
2008
2009 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
2010 {
2011         struct nfit_mem *nfit_mem;
2012         int dimm_count = 0, rc;
2013         struct nvdimm *nvdimm;
2014
2015         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2016                 struct acpi_nfit_flush_address *flush;
2017                 unsigned long flags = 0, cmd_mask;
2018                 struct nfit_memdev *nfit_memdev;
2019                 u32 device_handle;
2020                 u16 mem_flags;
2021
2022                 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
2023                 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
2024                 if (nvdimm) {
2025                         dimm_count++;
2026                         continue;
2027                 }
2028
2029                 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
2030                         set_bit(NDD_ALIASING, &flags);
2031
2032                 /* collate flags across all memdevs for this dimm */
2033                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2034                         struct acpi_nfit_memory_map *dimm_memdev;
2035
2036                         dimm_memdev = __to_nfit_memdev(nfit_mem);
2037                         if (dimm_memdev->device_handle
2038                                         != nfit_memdev->memdev->device_handle)
2039                                 continue;
2040                         dimm_memdev->flags |= nfit_memdev->memdev->flags;
2041                 }
2042
2043                 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
2044                 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
2045                         set_bit(NDD_UNARMED, &flags);
2046
2047                 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
2048                 if (rc)
2049                         continue;
2050
2051                 /*
2052                  * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
2053                  * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
2054                  * userspace interface.
2055                  */
2056                 cmd_mask = 1UL << ND_CMD_CALL;
2057                 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
2058                         /*
2059                          * These commands have a 1:1 correspondence
2060                          * between DSM payload and libnvdimm ioctl
2061                          * payload format.
2062                          */
2063                         cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
2064                 }
2065
2066                 /* Quirk to ignore LOCAL for labels on HYPERV DIMMs */
2067                 if (nfit_mem->family == NVDIMM_FAMILY_HYPERV)
2068                         set_bit(NDD_NOBLK, &flags);
2069
2070                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
2071                         set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
2072                         set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
2073                 }
2074                 if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags))
2075                         set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
2076
2077                 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
2078                         : NULL;
2079                 nvdimm = __nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
2080                                 acpi_nfit_dimm_attribute_groups,
2081                                 flags, cmd_mask, flush ? flush->hint_count : 0,
2082                                 nfit_mem->flush_wpq, &nfit_mem->id[0],
2083                                 acpi_nfit_get_security_ops(nfit_mem->family));
2084                 if (!nvdimm)
2085                         return -ENOMEM;
2086
2087                 nfit_mem->nvdimm = nvdimm;
2088                 dimm_count++;
2089
2090                 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
2091                         continue;
2092
2093                 dev_err(acpi_desc->dev, "Error found in NVDIMM %s flags:%s%s%s%s%s\n",
2094                                 nvdimm_name(nvdimm),
2095                   mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
2096                   mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
2097                   mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
2098                   mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
2099                   mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
2100
2101         }
2102
2103         rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
2104         if (rc)
2105                 return rc;
2106
2107         /*
2108          * Now that dimms are successfully registered, and async registration
2109          * is flushed, attempt to enable event notification.
2110          */
2111         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2112                 struct kernfs_node *nfit_kernfs;
2113
2114                 nvdimm = nfit_mem->nvdimm;
2115                 if (!nvdimm)
2116                         continue;
2117
2118                 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
2119                 if (nfit_kernfs)
2120                         nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
2121                                         "flags");
2122                 sysfs_put(nfit_kernfs);
2123                 if (!nfit_mem->flags_attr)
2124                         dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
2125                                         nvdimm_name(nvdimm));
2126         }
2127
2128         return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
2129                         acpi_desc);
2130 }
2131
2132 /*
2133  * These constants are private because there are no kernel consumers of
2134  * these commands.
2135  */
2136 enum nfit_aux_cmds {
2137         NFIT_CMD_TRANSLATE_SPA = 5,
2138         NFIT_CMD_ARS_INJECT_SET = 7,
2139         NFIT_CMD_ARS_INJECT_CLEAR = 8,
2140         NFIT_CMD_ARS_INJECT_GET = 9,
2141 };
2142
2143 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
2144 {
2145         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2146         const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
2147         struct acpi_device *adev;
2148         unsigned long dsm_mask;
2149         int i;
2150
2151         nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
2152         nd_desc->bus_dsm_mask = acpi_desc->bus_nfit_cmd_force_en;
2153         adev = to_acpi_dev(acpi_desc);
2154         if (!adev)
2155                 return;
2156
2157         for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
2158                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2159                         set_bit(i, &nd_desc->cmd_mask);
2160         set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
2161
2162         dsm_mask =
2163                 (1 << ND_CMD_ARS_CAP) |
2164                 (1 << ND_CMD_ARS_START) |
2165                 (1 << ND_CMD_ARS_STATUS) |
2166                 (1 << ND_CMD_CLEAR_ERROR) |
2167                 (1 << NFIT_CMD_TRANSLATE_SPA) |
2168                 (1 << NFIT_CMD_ARS_INJECT_SET) |
2169                 (1 << NFIT_CMD_ARS_INJECT_CLEAR) |
2170                 (1 << NFIT_CMD_ARS_INJECT_GET);
2171         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
2172                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2173                         set_bit(i, &nd_desc->bus_dsm_mask);
2174 }
2175
2176 static ssize_t range_index_show(struct device *dev,
2177                 struct device_attribute *attr, char *buf)
2178 {
2179         struct nd_region *nd_region = to_nd_region(dev);
2180         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
2181
2182         return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
2183 }
2184 static DEVICE_ATTR_RO(range_index);
2185
2186 static struct attribute *acpi_nfit_region_attributes[] = {
2187         &dev_attr_range_index.attr,
2188         NULL,
2189 };
2190
2191 static const struct attribute_group acpi_nfit_region_attribute_group = {
2192         .name = "nfit",
2193         .attrs = acpi_nfit_region_attributes,
2194 };
2195
2196 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
2197         &acpi_nfit_region_attribute_group,
2198         NULL,
2199 };
2200
2201 /* enough info to uniquely specify an interleave set */
2202 struct nfit_set_info {
2203         struct nfit_set_info_map {
2204                 u64 region_offset;
2205                 u32 serial_number;
2206                 u32 pad;
2207         } mapping[0];
2208 };
2209
2210 struct nfit_set_info2 {
2211         struct nfit_set_info_map2 {
2212                 u64 region_offset;
2213                 u32 serial_number;
2214                 u16 vendor_id;
2215                 u16 manufacturing_date;
2216                 u8  manufacturing_location;
2217                 u8  reserved[31];
2218         } mapping[0];
2219 };
2220
2221 static size_t sizeof_nfit_set_info(int num_mappings)
2222 {
2223         return sizeof(struct nfit_set_info)
2224                 + num_mappings * sizeof(struct nfit_set_info_map);
2225 }
2226
2227 static size_t sizeof_nfit_set_info2(int num_mappings)
2228 {
2229         return sizeof(struct nfit_set_info2)
2230                 + num_mappings * sizeof(struct nfit_set_info_map2);
2231 }
2232
2233 static int cmp_map_compat(const void *m0, const void *m1)
2234 {
2235         const struct nfit_set_info_map *map0 = m0;
2236         const struct nfit_set_info_map *map1 = m1;
2237
2238         return memcmp(&map0->region_offset, &map1->region_offset,
2239                         sizeof(u64));
2240 }
2241
2242 static int cmp_map(const void *m0, const void *m1)
2243 {
2244         const struct nfit_set_info_map *map0 = m0;
2245         const struct nfit_set_info_map *map1 = m1;
2246
2247         if (map0->region_offset < map1->region_offset)
2248                 return -1;
2249         else if (map0->region_offset > map1->region_offset)
2250                 return 1;
2251         return 0;
2252 }
2253
2254 static int cmp_map2(const void *m0, const void *m1)
2255 {
2256         const struct nfit_set_info_map2 *map0 = m0;
2257         const struct nfit_set_info_map2 *map1 = m1;
2258
2259         if (map0->region_offset < map1->region_offset)
2260                 return -1;
2261         else if (map0->region_offset > map1->region_offset)
2262                 return 1;
2263         return 0;
2264 }
2265
2266 /* Retrieve the nth entry referencing this spa */
2267 static struct acpi_nfit_memory_map *memdev_from_spa(
2268                 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
2269 {
2270         struct nfit_memdev *nfit_memdev;
2271
2272         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
2273                 if (nfit_memdev->memdev->range_index == range_index)
2274                         if (n-- == 0)
2275                                 return nfit_memdev->memdev;
2276         return NULL;
2277 }
2278
2279 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
2280                 struct nd_region_desc *ndr_desc,
2281                 struct acpi_nfit_system_address *spa)
2282 {
2283         struct device *dev = acpi_desc->dev;
2284         struct nd_interleave_set *nd_set;
2285         u16 nr = ndr_desc->num_mappings;
2286         struct nfit_set_info2 *info2;
2287         struct nfit_set_info *info;
2288         int i;
2289
2290         nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
2291         if (!nd_set)
2292                 return -ENOMEM;
2293         guid_copy(&nd_set->type_guid, (guid_t *) spa->range_guid);
2294
2295         info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
2296         if (!info)
2297                 return -ENOMEM;
2298
2299         info2 = devm_kzalloc(dev, sizeof_nfit_set_info2(nr), GFP_KERNEL);
2300         if (!info2)
2301                 return -ENOMEM;
2302
2303         for (i = 0; i < nr; i++) {
2304                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
2305                 struct nfit_set_info_map *map = &info->mapping[i];
2306                 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2307                 struct nvdimm *nvdimm = mapping->nvdimm;
2308                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2309                 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
2310                                 spa->range_index, i);
2311                 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2312
2313                 if (!memdev || !nfit_mem->dcr) {
2314                         dev_err(dev, "%s: failed to find DCR\n", __func__);
2315                         return -ENODEV;
2316                 }
2317
2318                 map->region_offset = memdev->region_offset;
2319                 map->serial_number = dcr->serial_number;
2320
2321                 map2->region_offset = memdev->region_offset;
2322                 map2->serial_number = dcr->serial_number;
2323                 map2->vendor_id = dcr->vendor_id;
2324                 map2->manufacturing_date = dcr->manufacturing_date;
2325                 map2->manufacturing_location = dcr->manufacturing_location;
2326         }
2327
2328         /* v1.1 namespaces */
2329         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2330                         cmp_map, NULL);
2331         nd_set->cookie1 = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2332
2333         /* v1.2 namespaces */
2334         sort(&info2->mapping[0], nr, sizeof(struct nfit_set_info_map2),
2335                         cmp_map2, NULL);
2336         nd_set->cookie2 = nd_fletcher64(info2, sizeof_nfit_set_info2(nr), 0);
2337
2338         /* support v1.1 namespaces created with the wrong sort order */
2339         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2340                         cmp_map_compat, NULL);
2341         nd_set->altcookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2342
2343         /* record the result of the sort for the mapping position */
2344         for (i = 0; i < nr; i++) {
2345                 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2346                 int j;
2347
2348                 for (j = 0; j < nr; j++) {
2349                         struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
2350                         struct nvdimm *nvdimm = mapping->nvdimm;
2351                         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2352                         struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2353
2354                         if (map2->serial_number == dcr->serial_number &&
2355                             map2->vendor_id == dcr->vendor_id &&
2356                             map2->manufacturing_date == dcr->manufacturing_date &&
2357                             map2->manufacturing_location
2358                                     == dcr->manufacturing_location) {
2359                                 mapping->position = i;
2360                                 break;
2361                         }
2362                 }
2363         }
2364
2365         ndr_desc->nd_set = nd_set;
2366         devm_kfree(dev, info);
2367         devm_kfree(dev, info2);
2368
2369         return 0;
2370 }
2371
2372 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
2373 {
2374         struct acpi_nfit_interleave *idt = mmio->idt;
2375         u32 sub_line_offset, line_index, line_offset;
2376         u64 line_no, table_skip_count, table_offset;
2377
2378         line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
2379         table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
2380         line_offset = idt->line_offset[line_index]
2381                 * mmio->line_size;
2382         table_offset = table_skip_count * mmio->table_size;
2383
2384         return mmio->base_offset + line_offset + table_offset + sub_line_offset;
2385 }
2386
2387 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
2388 {
2389         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2390         u64 offset = nfit_blk->stat_offset + mmio->size * bw;
2391         const u32 STATUS_MASK = 0x80000037;
2392
2393         if (mmio->num_lines)
2394                 offset = to_interleave_offset(offset, mmio);
2395
2396         return readl(mmio->addr.base + offset) & STATUS_MASK;
2397 }
2398
2399 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
2400                 resource_size_t dpa, unsigned int len, unsigned int write)
2401 {
2402         u64 cmd, offset;
2403         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2404
2405         enum {
2406                 BCW_OFFSET_MASK = (1ULL << 48)-1,
2407                 BCW_LEN_SHIFT = 48,
2408                 BCW_LEN_MASK = (1ULL << 8) - 1,
2409                 BCW_CMD_SHIFT = 56,
2410         };
2411
2412         cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
2413         len = len >> L1_CACHE_SHIFT;
2414         cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
2415         cmd |= ((u64) write) << BCW_CMD_SHIFT;
2416
2417         offset = nfit_blk->cmd_offset + mmio->size * bw;
2418         if (mmio->num_lines)
2419                 offset = to_interleave_offset(offset, mmio);
2420
2421         writeq(cmd, mmio->addr.base + offset);
2422         nvdimm_flush(nfit_blk->nd_region, NULL);
2423
2424         if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
2425                 readq(mmio->addr.base + offset);
2426 }
2427
2428 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
2429                 resource_size_t dpa, void *iobuf, size_t len, int rw,
2430                 unsigned int lane)
2431 {
2432         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2433         unsigned int copied = 0;
2434         u64 base_offset;
2435         int rc;
2436
2437         base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
2438                 + lane * mmio->size;
2439         write_blk_ctl(nfit_blk, lane, dpa, len, rw);
2440         while (len) {
2441                 unsigned int c;
2442                 u64 offset;
2443
2444                 if (mmio->num_lines) {
2445                         u32 line_offset;
2446
2447                         offset = to_interleave_offset(base_offset + copied,
2448                                         mmio);
2449                         div_u64_rem(offset, mmio->line_size, &line_offset);
2450                         c = min_t(size_t, len, mmio->line_size - line_offset);
2451                 } else {
2452                         offset = base_offset + nfit_blk->bdw_offset;
2453                         c = len;
2454                 }
2455
2456                 if (rw)
2457                         memcpy_flushcache(mmio->addr.aperture + offset, iobuf + copied, c);
2458                 else {
2459                         if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
2460                                 arch_invalidate_pmem((void __force *)
2461                                         mmio->addr.aperture + offset, c);
2462
2463                         memcpy(iobuf + copied, mmio->addr.aperture + offset, c);
2464                 }
2465
2466                 copied += c;
2467                 len -= c;
2468         }
2469
2470         if (rw)
2471                 nvdimm_flush(nfit_blk->nd_region, NULL);
2472
2473         rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
2474         return rc;
2475 }
2476
2477 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
2478                 resource_size_t dpa, void *iobuf, u64 len, int rw)
2479 {
2480         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
2481         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2482         struct nd_region *nd_region = nfit_blk->nd_region;
2483         unsigned int lane, copied = 0;
2484         int rc = 0;
2485
2486         lane = nd_region_acquire_lane(nd_region);
2487         while (len) {
2488                 u64 c = min(len, mmio->size);
2489
2490                 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
2491                                 iobuf + copied, c, rw, lane);
2492                 if (rc)
2493                         break;
2494
2495                 copied += c;
2496                 len -= c;
2497         }
2498         nd_region_release_lane(nd_region, lane);
2499
2500         return rc;
2501 }
2502
2503 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
2504                 struct acpi_nfit_interleave *idt, u16 interleave_ways)
2505 {
2506         if (idt) {
2507                 mmio->num_lines = idt->line_count;
2508                 mmio->line_size = idt->line_size;
2509                 if (interleave_ways == 0)
2510                         return -ENXIO;
2511                 mmio->table_size = mmio->num_lines * interleave_ways
2512                         * mmio->line_size;
2513         }
2514
2515         return 0;
2516 }
2517
2518 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
2519                 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
2520 {
2521         struct nd_cmd_dimm_flags flags;
2522         int rc;
2523
2524         memset(&flags, 0, sizeof(flags));
2525         rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
2526                         sizeof(flags), NULL);
2527
2528         if (rc >= 0 && flags.status == 0)
2529                 nfit_blk->dimm_flags = flags.flags;
2530         else if (rc == -ENOTTY) {
2531                 /* fall back to a conservative default */
2532                 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
2533                 rc = 0;
2534         } else
2535                 rc = -ENXIO;
2536
2537         return rc;
2538 }
2539
2540 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
2541                 struct device *dev)
2542 {
2543         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
2544         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
2545         struct nfit_blk_mmio *mmio;
2546         struct nfit_blk *nfit_blk;
2547         struct nfit_mem *nfit_mem;
2548         struct nvdimm *nvdimm;
2549         int rc;
2550
2551         nvdimm = nd_blk_region_to_dimm(ndbr);
2552         nfit_mem = nvdimm_provider_data(nvdimm);
2553         if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
2554                 dev_dbg(dev, "missing%s%s%s\n",
2555                                 nfit_mem ? "" : " nfit_mem",
2556                                 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
2557                                 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
2558                 return -ENXIO;
2559         }
2560
2561         nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
2562         if (!nfit_blk)
2563                 return -ENOMEM;
2564         nd_blk_region_set_provider_data(ndbr, nfit_blk);
2565         nfit_blk->nd_region = to_nd_region(dev);
2566
2567         /* map block aperture memory */
2568         nfit_blk->bdw_offset = nfit_mem->bdw->offset;
2569         mmio = &nfit_blk->mmio[BDW];
2570         mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
2571                         nfit_mem->spa_bdw->length, nd_blk_memremap_flags(ndbr));
2572         if (!mmio->addr.base) {
2573                 dev_dbg(dev, "%s failed to map bdw\n",
2574                                 nvdimm_name(nvdimm));
2575                 return -ENOMEM;
2576         }
2577         mmio->size = nfit_mem->bdw->size;
2578         mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
2579         mmio->idt = nfit_mem->idt_bdw;
2580         mmio->spa = nfit_mem->spa_bdw;
2581         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
2582                         nfit_mem->memdev_bdw->interleave_ways);
2583         if (rc) {
2584                 dev_dbg(dev, "%s failed to init bdw interleave\n",
2585                                 nvdimm_name(nvdimm));
2586                 return rc;
2587         }
2588
2589         /* map block control memory */
2590         nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
2591         nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
2592         mmio = &nfit_blk->mmio[DCR];
2593         mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
2594                         nfit_mem->spa_dcr->length);
2595         if (!mmio->addr.base) {
2596                 dev_dbg(dev, "%s failed to map dcr\n",
2597                                 nvdimm_name(nvdimm));
2598                 return -ENOMEM;
2599         }
2600         mmio->size = nfit_mem->dcr->window_size;
2601         mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
2602         mmio->idt = nfit_mem->idt_dcr;
2603         mmio->spa = nfit_mem->spa_dcr;
2604         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
2605                         nfit_mem->memdev_dcr->interleave_ways);
2606         if (rc) {
2607                 dev_dbg(dev, "%s failed to init dcr interleave\n",
2608                                 nvdimm_name(nvdimm));
2609                 return rc;
2610         }
2611
2612         rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
2613         if (rc < 0) {
2614                 dev_dbg(dev, "%s failed get DIMM flags\n",
2615                                 nvdimm_name(nvdimm));
2616                 return rc;
2617         }
2618
2619         if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
2620                 dev_warn(dev, "unable to guarantee persistence of writes\n");
2621
2622         if (mmio->line_size == 0)
2623                 return 0;
2624
2625         if ((u32) nfit_blk->cmd_offset % mmio->line_size
2626                         + 8 > mmio->line_size) {
2627                 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
2628                 return -ENXIO;
2629         } else if ((u32) nfit_blk->stat_offset % mmio->line_size
2630                         + 8 > mmio->line_size) {
2631                 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
2632                 return -ENXIO;
2633         }
2634
2635         return 0;
2636 }
2637
2638 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2639                 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2640 {
2641         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2642         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2643         int cmd_rc, rc;
2644
2645         cmd->address = spa->address;
2646         cmd->length = spa->length;
2647         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2648                         sizeof(*cmd), &cmd_rc);
2649         if (rc < 0)
2650                 return rc;
2651         return cmd_rc;
2652 }
2653
2654 static int ars_start(struct acpi_nfit_desc *acpi_desc,
2655                 struct nfit_spa *nfit_spa, enum nfit_ars_state req_type)
2656 {
2657         int rc;
2658         int cmd_rc;
2659         struct nd_cmd_ars_start ars_start;
2660         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2661         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2662
2663         memset(&ars_start, 0, sizeof(ars_start));
2664         ars_start.address = spa->address;
2665         ars_start.length = spa->length;
2666         if (req_type == ARS_REQ_SHORT)
2667                 ars_start.flags = ND_ARS_RETURN_PREV_DATA;
2668         if (nfit_spa_type(spa) == NFIT_SPA_PM)
2669                 ars_start.type = ND_ARS_PERSISTENT;
2670         else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2671                 ars_start.type = ND_ARS_VOLATILE;
2672         else
2673                 return -ENOTTY;
2674
2675         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2676                         sizeof(ars_start), &cmd_rc);
2677
2678         if (rc < 0)
2679                 return rc;
2680         if (cmd_rc < 0)
2681                 return cmd_rc;
2682         set_bit(ARS_VALID, &acpi_desc->scrub_flags);
2683         return 0;
2684 }
2685
2686 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2687 {
2688         int rc, cmd_rc;
2689         struct nd_cmd_ars_start ars_start;
2690         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2691         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2692
2693         ars_start = (struct nd_cmd_ars_start) {
2694                 .address = ars_status->restart_address,
2695                 .length = ars_status->restart_length,
2696                 .type = ars_status->type,
2697         };
2698         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2699                         sizeof(ars_start), &cmd_rc);
2700         if (rc < 0)
2701                 return rc;
2702         return cmd_rc;
2703 }
2704
2705 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2706 {
2707         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2708         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2709         int rc, cmd_rc;
2710
2711         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2712                         acpi_desc->max_ars, &cmd_rc);
2713         if (rc < 0)
2714                 return rc;
2715         return cmd_rc;
2716 }
2717
2718 static void ars_complete(struct acpi_nfit_desc *acpi_desc,
2719                 struct nfit_spa *nfit_spa)
2720 {
2721         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2722         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2723         struct nd_region *nd_region = nfit_spa->nd_region;
2724         struct device *dev;
2725
2726         lockdep_assert_held(&acpi_desc->init_mutex);
2727         /*
2728          * Only advance the ARS state for ARS runs initiated by the
2729          * kernel, ignore ARS results from BIOS initiated runs for scrub
2730          * completion tracking.
2731          */
2732         if (acpi_desc->scrub_spa != nfit_spa)
2733                 return;
2734
2735         if ((ars_status->address >= spa->address && ars_status->address
2736                                 < spa->address + spa->length)
2737                         || (ars_status->address < spa->address)) {
2738                 /*
2739                  * Assume that if a scrub starts at an offset from the
2740                  * start of nfit_spa that we are in the continuation
2741                  * case.
2742                  *
2743                  * Otherwise, if the scrub covers the spa range, mark
2744                  * any pending request complete.
2745                  */
2746                 if (ars_status->address + ars_status->length
2747                                 >= spa->address + spa->length)
2748                                 /* complete */;
2749                 else
2750                         return;
2751         } else
2752                 return;
2753
2754         acpi_desc->scrub_spa = NULL;
2755         if (nd_region) {
2756                 dev = nd_region_dev(nd_region);
2757                 nvdimm_region_notify(nd_region, NVDIMM_REVALIDATE_POISON);
2758         } else
2759                 dev = acpi_desc->dev;
2760         dev_dbg(dev, "ARS: range %d complete\n", spa->range_index);
2761 }
2762
2763 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc)
2764 {
2765         struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2766         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2767         int rc;
2768         u32 i;
2769
2770         /*
2771          * First record starts at 44 byte offset from the start of the
2772          * payload.
2773          */
2774         if (ars_status->out_length < 44)
2775                 return 0;
2776
2777         /*
2778          * Ignore potentially stale results that are only refreshed
2779          * after a start-ARS event.
2780          */
2781         if (!test_and_clear_bit(ARS_VALID, &acpi_desc->scrub_flags)) {
2782                 dev_dbg(acpi_desc->dev, "skip %d stale records\n",
2783                                 ars_status->num_records);
2784                 return 0;
2785         }
2786
2787         for (i = 0; i < ars_status->num_records; i++) {
2788                 /* only process full records */
2789                 if (ars_status->out_length
2790                                 < 44 + sizeof(struct nd_ars_record) * (i + 1))
2791                         break;
2792                 rc = nvdimm_bus_add_badrange(nvdimm_bus,
2793                                 ars_status->records[i].err_address,
2794                                 ars_status->records[i].length);
2795                 if (rc)
2796                         return rc;
2797         }
2798         if (i < ars_status->num_records)
2799                 dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2800
2801         return 0;
2802 }
2803
2804 static void acpi_nfit_remove_resource(void *data)
2805 {
2806         struct resource *res = data;
2807
2808         remove_resource(res);
2809 }
2810
2811 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2812                 struct nd_region_desc *ndr_desc)
2813 {
2814         struct resource *res, *nd_res = ndr_desc->res;
2815         int is_pmem, ret;
2816
2817         /* No operation if the region is already registered as PMEM */
2818         is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2819                                 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2820         if (is_pmem == REGION_INTERSECTS)
2821                 return 0;
2822
2823         res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2824         if (!res)
2825                 return -ENOMEM;
2826
2827         res->name = "Persistent Memory";
2828         res->start = nd_res->start;
2829         res->end = nd_res->end;
2830         res->flags = IORESOURCE_MEM;
2831         res->desc = IORES_DESC_PERSISTENT_MEMORY;
2832
2833         ret = insert_resource(&iomem_resource, res);
2834         if (ret)
2835                 return ret;
2836
2837         ret = devm_add_action_or_reset(acpi_desc->dev,
2838                                         acpi_nfit_remove_resource,
2839                                         res);
2840         if (ret)
2841                 return ret;
2842
2843         return 0;
2844 }
2845
2846 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2847                 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2848                 struct acpi_nfit_memory_map *memdev,
2849                 struct nfit_spa *nfit_spa)
2850 {
2851         struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2852                         memdev->device_handle);
2853         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2854         struct nd_blk_region_desc *ndbr_desc;
2855         struct nfit_mem *nfit_mem;
2856         int rc;
2857
2858         if (!nvdimm) {
2859                 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2860                                 spa->range_index, memdev->device_handle);
2861                 return -ENODEV;
2862         }
2863
2864         mapping->nvdimm = nvdimm;
2865         switch (nfit_spa_type(spa)) {
2866         case NFIT_SPA_PM:
2867         case NFIT_SPA_VOLATILE:
2868                 mapping->start = memdev->address;
2869                 mapping->size = memdev->region_size;
2870                 break;
2871         case NFIT_SPA_DCR:
2872                 nfit_mem = nvdimm_provider_data(nvdimm);
2873                 if (!nfit_mem || !nfit_mem->bdw) {
2874                         dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2875                                         spa->range_index, nvdimm_name(nvdimm));
2876                         break;
2877                 }
2878
2879                 mapping->size = nfit_mem->bdw->capacity;
2880                 mapping->start = nfit_mem->bdw->start_address;
2881                 ndr_desc->num_lanes = nfit_mem->bdw->windows;
2882                 ndr_desc->mapping = mapping;
2883                 ndr_desc->num_mappings = 1;
2884                 ndbr_desc = to_blk_region_desc(ndr_desc);
2885                 ndbr_desc->enable = acpi_nfit_blk_region_enable;
2886                 ndbr_desc->do_io = acpi_desc->blk_do_io;
2887                 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2888                 if (rc)
2889                         return rc;
2890                 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2891                                 ndr_desc);
2892                 if (!nfit_spa->nd_region)
2893                         return -ENOMEM;
2894                 break;
2895         }
2896
2897         return 0;
2898 }
2899
2900 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2901 {
2902         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2903                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2904                 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2905                 nfit_spa_type(spa) == NFIT_SPA_PCD);
2906 }
2907
2908 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
2909 {
2910         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2911                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2912                 nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
2913 }
2914
2915 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2916                 struct nfit_spa *nfit_spa)
2917 {
2918         static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2919         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2920         struct nd_blk_region_desc ndbr_desc;
2921         struct nd_region_desc *ndr_desc;
2922         struct nfit_memdev *nfit_memdev;
2923         struct nvdimm_bus *nvdimm_bus;
2924         struct resource res;
2925         int count = 0, rc;
2926
2927         if (nfit_spa->nd_region)
2928                 return 0;
2929
2930         if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2931                 dev_dbg(acpi_desc->dev, "detected invalid spa index\n");
2932                 return 0;
2933         }
2934
2935         memset(&res, 0, sizeof(res));
2936         memset(&mappings, 0, sizeof(mappings));
2937         memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2938         res.start = spa->address;
2939         res.end = res.start + spa->length - 1;
2940         ndr_desc = &ndbr_desc.ndr_desc;
2941         ndr_desc->res = &res;
2942         ndr_desc->provider_data = nfit_spa;
2943         ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2944         if (spa->flags & ACPI_NFIT_PROXIMITY_VALID) {
2945                 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2946                                                 spa->proximity_domain);
2947                 ndr_desc->target_node = acpi_map_pxm_to_node(
2948                                 spa->proximity_domain);
2949         } else {
2950                 ndr_desc->numa_node = NUMA_NO_NODE;
2951                 ndr_desc->target_node = NUMA_NO_NODE;
2952         }
2953
2954         /*
2955          * Persistence domain bits are hierarchical, if
2956          * ACPI_NFIT_CAPABILITY_CACHE_FLUSH is set then
2957          * ACPI_NFIT_CAPABILITY_MEM_FLUSH is implied.
2958          */
2959         if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_CACHE_FLUSH)
2960                 set_bit(ND_REGION_PERSIST_CACHE, &ndr_desc->flags);
2961         else if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_MEM_FLUSH)
2962                 set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc->flags);
2963
2964         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2965                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2966                 struct nd_mapping_desc *mapping;
2967
2968                 if (memdev->range_index != spa->range_index)
2969                         continue;
2970                 if (count >= ND_MAX_MAPPINGS) {
2971                         dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2972                                         spa->range_index, ND_MAX_MAPPINGS);
2973                         return -ENXIO;
2974                 }
2975                 mapping = &mappings[count++];
2976                 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
2977                                 memdev, nfit_spa);
2978                 if (rc)
2979                         goto out;
2980         }
2981
2982         ndr_desc->mapping = mappings;
2983         ndr_desc->num_mappings = count;
2984         rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2985         if (rc)
2986                 goto out;
2987
2988         nvdimm_bus = acpi_desc->nvdimm_bus;
2989         if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2990                 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2991                 if (rc) {
2992                         dev_warn(acpi_desc->dev,
2993                                 "failed to insert pmem resource to iomem: %d\n",
2994                                 rc);
2995                         goto out;
2996                 }
2997
2998                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2999                                 ndr_desc);
3000                 if (!nfit_spa->nd_region)
3001                         rc = -ENOMEM;
3002         } else if (nfit_spa_is_volatile(spa)) {
3003                 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
3004                                 ndr_desc);
3005                 if (!nfit_spa->nd_region)
3006                         rc = -ENOMEM;
3007         } else if (nfit_spa_is_virtual(spa)) {
3008                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
3009                                 ndr_desc);
3010                 if (!nfit_spa->nd_region)
3011                         rc = -ENOMEM;
3012         }
3013
3014  out:
3015         if (rc)
3016                 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
3017                                 nfit_spa->spa->range_index);
3018         return rc;
3019 }
3020
3021 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc)
3022 {
3023         struct device *dev = acpi_desc->dev;
3024         struct nd_cmd_ars_status *ars_status;
3025
3026         if (acpi_desc->ars_status) {
3027                 memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
3028                 return 0;
3029         }
3030
3031         ars_status = devm_kzalloc(dev, acpi_desc->max_ars, GFP_KERNEL);
3032         if (!ars_status)
3033                 return -ENOMEM;
3034         acpi_desc->ars_status = ars_status;
3035         return 0;
3036 }
3037
3038 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc)
3039 {
3040         int rc;
3041
3042         if (ars_status_alloc(acpi_desc))
3043                 return -ENOMEM;
3044
3045         rc = ars_get_status(acpi_desc);
3046
3047         if (rc < 0 && rc != -ENOSPC)
3048                 return rc;
3049
3050         if (ars_status_process_records(acpi_desc))
3051                 dev_err(acpi_desc->dev, "Failed to process ARS records\n");
3052
3053         return rc;
3054 }
3055
3056 static int ars_register(struct acpi_nfit_desc *acpi_desc,
3057                 struct nfit_spa *nfit_spa)
3058 {
3059         int rc;
3060
3061         if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3062                 return acpi_nfit_register_region(acpi_desc, nfit_spa);
3063
3064         set_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
3065         if (!no_init_ars)
3066                 set_bit(ARS_REQ_LONG, &nfit_spa->ars_state);
3067
3068         switch (acpi_nfit_query_poison(acpi_desc)) {
3069         case 0:
3070         case -ENOSPC:
3071         case -EAGAIN:
3072                 rc = ars_start(acpi_desc, nfit_spa, ARS_REQ_SHORT);
3073                 /* shouldn't happen, try again later */
3074                 if (rc == -EBUSY)
3075                         break;
3076                 if (rc) {
3077                         set_bit(ARS_FAILED, &nfit_spa->ars_state);
3078                         break;
3079                 }
3080                 clear_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
3081                 rc = acpi_nfit_query_poison(acpi_desc);
3082                 if (rc)
3083                         break;
3084                 acpi_desc->scrub_spa = nfit_spa;
3085                 ars_complete(acpi_desc, nfit_spa);
3086                 /*
3087                  * If ars_complete() says we didn't complete the
3088                  * short scrub, we'll try again with a long
3089                  * request.
3090                  */
3091                 acpi_desc->scrub_spa = NULL;
3092                 break;
3093         case -EBUSY:
3094         case -ENOMEM:
3095                 /*
3096                  * BIOS was using ARS, wait for it to complete (or
3097                  * resources to become available) and then perform our
3098                  * own scrubs.
3099                  */
3100                 break;
3101         default:
3102                 set_bit(ARS_FAILED, &nfit_spa->ars_state);
3103                 break;
3104         }
3105
3106         return acpi_nfit_register_region(acpi_desc, nfit_spa);
3107 }
3108
3109 static void ars_complete_all(struct acpi_nfit_desc *acpi_desc)
3110 {
3111         struct nfit_spa *nfit_spa;
3112
3113         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3114                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3115                         continue;
3116                 ars_complete(acpi_desc, nfit_spa);
3117         }
3118 }
3119
3120 static unsigned int __acpi_nfit_scrub(struct acpi_nfit_desc *acpi_desc,
3121                 int query_rc)
3122 {
3123         unsigned int tmo = acpi_desc->scrub_tmo;
3124         struct device *dev = acpi_desc->dev;
3125         struct nfit_spa *nfit_spa;
3126
3127         lockdep_assert_held(&acpi_desc->init_mutex);
3128
3129         if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags))
3130                 return 0;
3131
3132         if (query_rc == -EBUSY) {
3133                 dev_dbg(dev, "ARS: ARS busy\n");
3134                 return min(30U * 60U, tmo * 2);
3135         }
3136         if (query_rc == -ENOSPC) {
3137                 dev_dbg(dev, "ARS: ARS continue\n");
3138                 ars_continue(acpi_desc);
3139                 return 1;
3140         }
3141         if (query_rc && query_rc != -EAGAIN) {
3142                 unsigned long long addr, end;
3143
3144                 addr = acpi_desc->ars_status->address;
3145                 end = addr + acpi_desc->ars_status->length;
3146                 dev_dbg(dev, "ARS: %llx-%llx failed (%d)\n", addr, end,
3147                                 query_rc);
3148         }
3149
3150         ars_complete_all(acpi_desc);
3151         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3152                 enum nfit_ars_state req_type;
3153                 int rc;
3154
3155                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3156                         continue;
3157
3158                 /* prefer short ARS requests first */
3159                 if (test_bit(ARS_REQ_SHORT, &nfit_spa->ars_state))
3160                         req_type = ARS_REQ_SHORT;
3161                 else if (test_bit(ARS_REQ_LONG, &nfit_spa->ars_state))
3162                         req_type = ARS_REQ_LONG;
3163                 else
3164                         continue;
3165                 rc = ars_start(acpi_desc, nfit_spa, req_type);
3166
3167                 dev = nd_region_dev(nfit_spa->nd_region);
3168                 dev_dbg(dev, "ARS: range %d ARS start %s (%d)\n",
3169                                 nfit_spa->spa->range_index,
3170                                 req_type == ARS_REQ_SHORT ? "short" : "long",
3171                                 rc);
3172                 /*
3173                  * Hmm, we raced someone else starting ARS? Try again in
3174                  * a bit.
3175                  */
3176                 if (rc == -EBUSY)
3177                         return 1;
3178                 if (rc == 0) {
3179                         dev_WARN_ONCE(dev, acpi_desc->scrub_spa,
3180                                         "scrub start while range %d active\n",
3181                                         acpi_desc->scrub_spa->spa->range_index);
3182                         clear_bit(req_type, &nfit_spa->ars_state);
3183                         acpi_desc->scrub_spa = nfit_spa;
3184                         /*
3185                          * Consider this spa last for future scrub
3186                          * requests
3187                          */
3188                         list_move_tail(&nfit_spa->list, &acpi_desc->spas);
3189                         return 1;
3190                 }
3191
3192                 dev_err(dev, "ARS: range %d ARS failed (%d)\n",
3193                                 nfit_spa->spa->range_index, rc);
3194                 set_bit(ARS_FAILED, &nfit_spa->ars_state);
3195         }
3196         return 0;
3197 }
3198
3199 static void __sched_ars(struct acpi_nfit_desc *acpi_desc, unsigned int tmo)
3200 {
3201         lockdep_assert_held(&acpi_desc->init_mutex);
3202
3203         set_bit(ARS_BUSY, &acpi_desc->scrub_flags);
3204         /* note this should only be set from within the workqueue */
3205         if (tmo)
3206                 acpi_desc->scrub_tmo = tmo;
3207         queue_delayed_work(nfit_wq, &acpi_desc->dwork, tmo * HZ);
3208 }
3209
3210 static void sched_ars(struct acpi_nfit_desc *acpi_desc)
3211 {
3212         __sched_ars(acpi_desc, 0);
3213 }
3214
3215 static void notify_ars_done(struct acpi_nfit_desc *acpi_desc)
3216 {
3217         lockdep_assert_held(&acpi_desc->init_mutex);
3218
3219         clear_bit(ARS_BUSY, &acpi_desc->scrub_flags);
3220         acpi_desc->scrub_count++;
3221         if (acpi_desc->scrub_count_state)
3222                 sysfs_notify_dirent(acpi_desc->scrub_count_state);
3223 }
3224
3225 static void acpi_nfit_scrub(struct work_struct *work)
3226 {
3227         struct acpi_nfit_desc *acpi_desc;
3228         unsigned int tmo;
3229         int query_rc;
3230
3231         acpi_desc = container_of(work, typeof(*acpi_desc), dwork.work);
3232         mutex_lock(&acpi_desc->init_mutex);
3233         query_rc = acpi_nfit_query_poison(acpi_desc);
3234         tmo = __acpi_nfit_scrub(acpi_desc, query_rc);
3235         if (tmo)
3236                 __sched_ars(acpi_desc, tmo);
3237         else
3238                 notify_ars_done(acpi_desc);
3239         memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
3240         clear_bit(ARS_POLL, &acpi_desc->scrub_flags);
3241         mutex_unlock(&acpi_desc->init_mutex);
3242 }
3243
3244 static void acpi_nfit_init_ars(struct acpi_nfit_desc *acpi_desc,
3245                 struct nfit_spa *nfit_spa)
3246 {
3247         int type = nfit_spa_type(nfit_spa->spa);
3248         struct nd_cmd_ars_cap ars_cap;
3249         int rc;
3250
3251         set_bit(ARS_FAILED, &nfit_spa->ars_state);
3252         memset(&ars_cap, 0, sizeof(ars_cap));
3253         rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
3254         if (rc < 0)
3255                 return;
3256         /* check that the supported scrub types match the spa type */
3257         if (type == NFIT_SPA_VOLATILE && ((ars_cap.status >> 16)
3258                                 & ND_ARS_VOLATILE) == 0)
3259                 return;
3260         if (type == NFIT_SPA_PM && ((ars_cap.status >> 16)
3261                                 & ND_ARS_PERSISTENT) == 0)
3262                 return;
3263
3264         nfit_spa->max_ars = ars_cap.max_ars_out;
3265         nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
3266         acpi_desc->max_ars = max(nfit_spa->max_ars, acpi_desc->max_ars);
3267         clear_bit(ARS_FAILED, &nfit_spa->ars_state);
3268 }
3269
3270 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
3271 {
3272         struct nfit_spa *nfit_spa;
3273         int rc;
3274
3275         set_bit(ARS_VALID, &acpi_desc->scrub_flags);
3276         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3277                 switch (nfit_spa_type(nfit_spa->spa)) {
3278                 case NFIT_SPA_VOLATILE:
3279                 case NFIT_SPA_PM:
3280                         acpi_nfit_init_ars(acpi_desc, nfit_spa);
3281                         break;
3282                 }
3283         }
3284
3285         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
3286                 switch (nfit_spa_type(nfit_spa->spa)) {
3287                 case NFIT_SPA_VOLATILE:
3288                 case NFIT_SPA_PM:
3289                         /* register regions and kick off initial ARS run */
3290                         rc = ars_register(acpi_desc, nfit_spa);
3291                         if (rc)
3292                                 return rc;
3293                         break;
3294                 case NFIT_SPA_BDW:
3295                         /* nothing to register */
3296                         break;
3297                 case NFIT_SPA_DCR:
3298                 case NFIT_SPA_VDISK:
3299                 case NFIT_SPA_VCD:
3300                 case NFIT_SPA_PDISK:
3301                 case NFIT_SPA_PCD:
3302                         /* register known regions that don't support ARS */
3303                         rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
3304                         if (rc)
3305                                 return rc;
3306                         break;
3307                 default:
3308                         /* don't register unknown regions */
3309                         break;
3310                 }
3311
3312         sched_ars(acpi_desc);
3313         return 0;
3314 }
3315
3316 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
3317                 struct nfit_table_prev *prev)
3318 {
3319         struct device *dev = acpi_desc->dev;
3320
3321         if (!list_empty(&prev->spas) ||
3322                         !list_empty(&prev->memdevs) ||
3323                         !list_empty(&prev->dcrs) ||
3324                         !list_empty(&prev->bdws) ||
3325                         !list_empty(&prev->idts) ||
3326                         !list_empty(&prev->flushes)) {
3327                 dev_err(dev, "new nfit deletes entries (unsupported)\n");
3328                 return -ENXIO;
3329         }
3330         return 0;
3331 }
3332
3333 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
3334 {
3335         struct device *dev = acpi_desc->dev;
3336         struct kernfs_node *nfit;
3337         struct device *bus_dev;
3338
3339         if (!ars_supported(acpi_desc->nvdimm_bus))
3340                 return 0;
3341
3342         bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3343         nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
3344         if (!nfit) {
3345                 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
3346                 return -ENODEV;
3347         }
3348         acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
3349         sysfs_put(nfit);
3350         if (!acpi_desc->scrub_count_state) {
3351                 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
3352                 return -ENODEV;
3353         }
3354
3355         return 0;
3356 }
3357
3358 static void acpi_nfit_unregister(void *data)
3359 {
3360         struct acpi_nfit_desc *acpi_desc = data;
3361
3362         nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
3363 }
3364
3365 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
3366 {
3367         struct device *dev = acpi_desc->dev;
3368         struct nfit_table_prev prev;
3369         const void *end;
3370         int rc;
3371
3372         if (!acpi_desc->nvdimm_bus) {
3373                 acpi_nfit_init_dsms(acpi_desc);
3374
3375                 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
3376                                 &acpi_desc->nd_desc);
3377                 if (!acpi_desc->nvdimm_bus)
3378                         return -ENOMEM;
3379
3380                 rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
3381                                 acpi_desc);
3382                 if (rc)
3383                         return rc;
3384
3385                 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
3386                 if (rc)
3387                         return rc;
3388
3389                 /* register this acpi_desc for mce notifications */
3390                 mutex_lock(&acpi_desc_lock);
3391                 list_add_tail(&acpi_desc->list, &acpi_descs);
3392                 mutex_unlock(&acpi_desc_lock);
3393         }
3394
3395         mutex_lock(&acpi_desc->init_mutex);
3396
3397         INIT_LIST_HEAD(&prev.spas);
3398         INIT_LIST_HEAD(&prev.memdevs);
3399         INIT_LIST_HEAD(&prev.dcrs);
3400         INIT_LIST_HEAD(&prev.bdws);
3401         INIT_LIST_HEAD(&prev.idts);
3402         INIT_LIST_HEAD(&prev.flushes);
3403
3404         list_cut_position(&prev.spas, &acpi_desc->spas,
3405                                 acpi_desc->spas.prev);
3406         list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
3407                                 acpi_desc->memdevs.prev);
3408         list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
3409                                 acpi_desc->dcrs.prev);
3410         list_cut_position(&prev.bdws, &acpi_desc->bdws,
3411                                 acpi_desc->bdws.prev);
3412         list_cut_position(&prev.idts, &acpi_desc->idts,
3413                                 acpi_desc->idts.prev);
3414         list_cut_position(&prev.flushes, &acpi_desc->flushes,
3415                                 acpi_desc->flushes.prev);
3416
3417         end = data + sz;
3418         while (!IS_ERR_OR_NULL(data))
3419                 data = add_table(acpi_desc, &prev, data, end);
3420
3421         if (IS_ERR(data)) {
3422                 dev_dbg(dev, "nfit table parsing error: %ld\n", PTR_ERR(data));
3423                 rc = PTR_ERR(data);
3424                 goto out_unlock;
3425         }
3426
3427         rc = acpi_nfit_check_deletions(acpi_desc, &prev);
3428         if (rc)
3429                 goto out_unlock;
3430
3431         rc = nfit_mem_init(acpi_desc);
3432         if (rc)
3433                 goto out_unlock;
3434
3435         rc = acpi_nfit_register_dimms(acpi_desc);
3436         if (rc)
3437                 goto out_unlock;
3438
3439         rc = acpi_nfit_register_regions(acpi_desc);
3440
3441  out_unlock:
3442         mutex_unlock(&acpi_desc->init_mutex);
3443         return rc;
3444 }
3445 EXPORT_SYMBOL_GPL(acpi_nfit_init);
3446
3447 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
3448 {
3449         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
3450         struct device *dev = acpi_desc->dev;
3451
3452         /* Bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
3453         nfit_device_lock(dev);
3454         nfit_device_unlock(dev);
3455
3456         /* Bounce the init_mutex to complete initial registration */
3457         mutex_lock(&acpi_desc->init_mutex);
3458         mutex_unlock(&acpi_desc->init_mutex);
3459
3460         return 0;
3461 }
3462
3463 static int __acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3464                 struct nvdimm *nvdimm, unsigned int cmd)
3465 {
3466         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
3467
3468         if (nvdimm)
3469                 return 0;
3470         if (cmd != ND_CMD_ARS_START)
3471                 return 0;
3472
3473         /*
3474          * The kernel and userspace may race to initiate a scrub, but
3475          * the scrub thread is prepared to lose that initial race.  It
3476          * just needs guarantees that any ARS it initiates are not
3477          * interrupted by any intervening start requests from userspace.
3478          */
3479         if (work_busy(&acpi_desc->dwork.work))
3480                 return -EBUSY;
3481
3482         return 0;
3483 }
3484
3485 /* prevent security commands from being issued via ioctl */
3486 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3487                 struct nvdimm *nvdimm, unsigned int cmd, void *buf)
3488 {
3489         struct nd_cmd_pkg *call_pkg = buf;
3490         unsigned int func;
3491
3492         if (nvdimm && cmd == ND_CMD_CALL &&
3493                         call_pkg->nd_family == NVDIMM_FAMILY_INTEL) {
3494                 func = call_pkg->nd_command;
3495                 if ((1 << func) & NVDIMM_INTEL_SECURITY_CMDMASK)
3496                         return -EOPNOTSUPP;
3497         }
3498
3499         return __acpi_nfit_clear_to_send(nd_desc, nvdimm, cmd);
3500 }
3501
3502 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc,
3503                 enum nfit_ars_state req_type)
3504 {
3505         struct device *dev = acpi_desc->dev;
3506         int scheduled = 0, busy = 0;
3507         struct nfit_spa *nfit_spa;
3508
3509         mutex_lock(&acpi_desc->init_mutex);
3510         if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags)) {
3511                 mutex_unlock(&acpi_desc->init_mutex);
3512                 return 0;
3513         }
3514
3515         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3516                 int type = nfit_spa_type(nfit_spa->spa);
3517
3518                 if (type != NFIT_SPA_PM && type != NFIT_SPA_VOLATILE)
3519                         continue;
3520                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3521                         continue;
3522
3523                 if (test_and_set_bit(req_type, &nfit_spa->ars_state))
3524                         busy++;
3525                 else
3526                         scheduled++;
3527         }
3528         if (scheduled) {
3529                 sched_ars(acpi_desc);
3530                 dev_dbg(dev, "ars_scan triggered\n");
3531         }
3532         mutex_unlock(&acpi_desc->init_mutex);
3533
3534         if (scheduled)
3535                 return 0;
3536         if (busy)
3537                 return -EBUSY;
3538         return -ENOTTY;
3539 }
3540
3541 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
3542 {
3543         struct nvdimm_bus_descriptor *nd_desc;
3544
3545         dev_set_drvdata(dev, acpi_desc);
3546         acpi_desc->dev = dev;
3547         acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
3548         nd_desc = &acpi_desc->nd_desc;
3549         nd_desc->provider_name = "ACPI.NFIT";
3550         nd_desc->module = THIS_MODULE;
3551         nd_desc->ndctl = acpi_nfit_ctl;
3552         nd_desc->flush_probe = acpi_nfit_flush_probe;
3553         nd_desc->clear_to_send = acpi_nfit_clear_to_send;
3554         nd_desc->attr_groups = acpi_nfit_attribute_groups;
3555
3556         INIT_LIST_HEAD(&acpi_desc->spas);
3557         INIT_LIST_HEAD(&acpi_desc->dcrs);
3558         INIT_LIST_HEAD(&acpi_desc->bdws);
3559         INIT_LIST_HEAD(&acpi_desc->idts);
3560         INIT_LIST_HEAD(&acpi_desc->flushes);
3561         INIT_LIST_HEAD(&acpi_desc->memdevs);
3562         INIT_LIST_HEAD(&acpi_desc->dimms);
3563         INIT_LIST_HEAD(&acpi_desc->list);
3564         mutex_init(&acpi_desc->init_mutex);
3565         acpi_desc->scrub_tmo = 1;
3566         INIT_DELAYED_WORK(&acpi_desc->dwork, acpi_nfit_scrub);
3567 }
3568 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
3569
3570 static void acpi_nfit_put_table(void *table)
3571 {
3572         acpi_put_table(table);
3573 }
3574
3575 void acpi_nfit_shutdown(void *data)
3576 {
3577         struct acpi_nfit_desc *acpi_desc = data;
3578         struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3579
3580         /*
3581          * Destruct under acpi_desc_lock so that nfit_handle_mce does not
3582          * race teardown
3583          */
3584         mutex_lock(&acpi_desc_lock);
3585         list_del(&acpi_desc->list);
3586         mutex_unlock(&acpi_desc_lock);
3587
3588         mutex_lock(&acpi_desc->init_mutex);
3589         set_bit(ARS_CANCEL, &acpi_desc->scrub_flags);
3590         cancel_delayed_work_sync(&acpi_desc->dwork);
3591         mutex_unlock(&acpi_desc->init_mutex);
3592
3593         /*
3594          * Bounce the nvdimm bus lock to make sure any in-flight
3595          * acpi_nfit_ars_rescan() submissions have had a chance to
3596          * either submit or see ->cancel set.
3597          */
3598         nfit_device_lock(bus_dev);
3599         nfit_device_unlock(bus_dev);
3600
3601         flush_workqueue(nfit_wq);
3602 }
3603 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown);
3604
3605 static int acpi_nfit_add(struct acpi_device *adev)
3606 {
3607         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3608         struct acpi_nfit_desc *acpi_desc;
3609         struct device *dev = &adev->dev;
3610         struct acpi_table_header *tbl;
3611         acpi_status status = AE_OK;
3612         acpi_size sz;
3613         int rc = 0;
3614
3615         status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
3616         if (ACPI_FAILURE(status)) {
3617                 /* The NVDIMM root device allows OS to trigger enumeration of
3618                  * NVDIMMs through NFIT at boot time and re-enumeration at
3619                  * root level via the _FIT method during runtime.
3620                  * This is ok to return 0 here, we could have an nvdimm
3621                  * hotplugged later and evaluate _FIT method which returns
3622                  * data in the format of a series of NFIT Structures.
3623                  */
3624                 dev_dbg(dev, "failed to find NFIT at startup\n");
3625                 return 0;
3626         }
3627
3628         rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl);
3629         if (rc)
3630                 return rc;
3631         sz = tbl->length;
3632
3633         acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3634         if (!acpi_desc)
3635                 return -ENOMEM;
3636         acpi_nfit_desc_init(acpi_desc, &adev->dev);
3637
3638         /* Save the acpi header for exporting the revision via sysfs */
3639         acpi_desc->acpi_header = *tbl;
3640
3641         /* Evaluate _FIT and override with that if present */
3642         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
3643         if (ACPI_SUCCESS(status) && buf.length > 0) {
3644                 union acpi_object *obj = buf.pointer;
3645
3646                 if (obj->type == ACPI_TYPE_BUFFER)
3647                         rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3648                                         obj->buffer.length);
3649                 else
3650                         dev_dbg(dev, "invalid type %d, ignoring _FIT\n",
3651                                 (int) obj->type);
3652                 kfree(buf.pointer);
3653         } else
3654                 /* skip over the lead-in header table */
3655                 rc = acpi_nfit_init(acpi_desc, (void *) tbl
3656                                 + sizeof(struct acpi_table_nfit),
3657                                 sz - sizeof(struct acpi_table_nfit));
3658
3659         if (rc)
3660                 return rc;
3661         return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc);
3662 }
3663
3664 static int acpi_nfit_remove(struct acpi_device *adev)
3665 {
3666         /* see acpi_nfit_unregister */
3667         return 0;
3668 }
3669
3670 static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle)
3671 {
3672         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3673         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3674         union acpi_object *obj;
3675         acpi_status status;
3676         int ret;
3677
3678         if (!dev->driver) {
3679                 /* dev->driver may be null if we're being removed */
3680                 dev_dbg(dev, "no driver found for dev\n");
3681                 return;
3682         }
3683
3684         if (!acpi_desc) {
3685                 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3686                 if (!acpi_desc)
3687                         return;
3688                 acpi_nfit_desc_init(acpi_desc, dev);
3689         } else {
3690                 /*
3691                  * Finish previous registration before considering new
3692                  * regions.
3693                  */
3694                 flush_workqueue(nfit_wq);
3695         }
3696
3697         /* Evaluate _FIT */
3698         status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
3699         if (ACPI_FAILURE(status)) {
3700                 dev_err(dev, "failed to evaluate _FIT\n");
3701                 return;
3702         }
3703
3704         obj = buf.pointer;
3705         if (obj->type == ACPI_TYPE_BUFFER) {
3706                 ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3707                                 obj->buffer.length);
3708                 if (ret)
3709                         dev_err(dev, "failed to merge updated NFIT\n");
3710         } else
3711                 dev_err(dev, "Invalid _FIT\n");
3712         kfree(buf.pointer);
3713 }
3714
3715 static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle)
3716 {
3717         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3718
3719         if (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON)
3720                 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
3721         else
3722                 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_SHORT);
3723 }
3724
3725 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
3726 {
3727         dev_dbg(dev, "event: 0x%x\n", event);
3728
3729         switch (event) {
3730         case NFIT_NOTIFY_UPDATE:
3731                 return acpi_nfit_update_notify(dev, handle);
3732         case NFIT_NOTIFY_UC_MEMORY_ERROR:
3733                 return acpi_nfit_uc_error_notify(dev, handle);
3734         default:
3735                 return;
3736         }
3737 }
3738 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
3739
3740 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
3741 {
3742         nfit_device_lock(&adev->dev);
3743         __acpi_nfit_notify(&adev->dev, adev->handle, event);
3744         nfit_device_unlock(&adev->dev);
3745 }
3746
3747 static const struct acpi_device_id acpi_nfit_ids[] = {
3748         { "ACPI0012", 0 },
3749         { "", 0 },
3750 };
3751 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
3752
3753 static struct acpi_driver acpi_nfit_driver = {
3754         .name = KBUILD_MODNAME,
3755         .ids = acpi_nfit_ids,
3756         .ops = {
3757                 .add = acpi_nfit_add,
3758                 .remove = acpi_nfit_remove,
3759                 .notify = acpi_nfit_notify,
3760         },
3761 };
3762
3763 static __init int nfit_init(void)
3764 {
3765         int ret;
3766
3767         BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
3768         BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
3769         BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
3770         BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
3771         BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
3772         BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
3773         BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
3774         BUILD_BUG_ON(sizeof(struct acpi_nfit_capabilities) != 16);
3775
3776         guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]);
3777         guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]);
3778         guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]);
3779         guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]);
3780         guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]);
3781         guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]);
3782         guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]);
3783         guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]);
3784         guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]);
3785         guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]);
3786         guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
3787         guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
3788         guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
3789         guid_parse(UUID_NFIT_DIMM_N_HYPERV, &nfit_uuid[NFIT_DEV_DIMM_N_HYPERV]);
3790
3791         nfit_wq = create_singlethread_workqueue("nfit");
3792         if (!nfit_wq)
3793                 return -ENOMEM;
3794
3795         nfit_mce_register();
3796         ret = acpi_bus_register_driver(&acpi_nfit_driver);
3797         if (ret) {
3798                 nfit_mce_unregister();
3799                 destroy_workqueue(nfit_wq);
3800         }
3801
3802         return ret;
3803
3804 }
3805
3806 static __exit void nfit_exit(void)
3807 {
3808         nfit_mce_unregister();
3809         acpi_bus_unregister_driver(&acpi_nfit_driver);
3810         destroy_workqueue(nfit_wq);
3811         WARN_ON(!list_empty(&acpi_descs));
3812 }
3813
3814 module_init(nfit_init);
3815 module_exit(nfit_exit);
3816 MODULE_LICENSE("GPL v2");
3817 MODULE_AUTHOR("Intel Corporation");