drivers/acpi/nfit/core.c

   1 /*
   2  * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of version 2 of the GNU General Public License as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  */
  13 #include <linux/list_sort.h>
  14 #include <linux/libnvdimm.h>
  15 #include <linux/module.h>
  16 #include <linux/mutex.h>
  17 #include <linux/ndctl.h>
  18 #include <linux/sysfs.h>
  19 #include <linux/delay.h>
  20 #include <linux/list.h>
  21 #include <linux/acpi.h>
  22 #include <linux/sort.h>
  23 #include <linux/io.h>
  24 #include <linux/nd.h>
  25 #include <asm/cacheflush.h>
  26 #include <acpi/nfit.h>
  27 #include "intel.h"
  28 #include "nfit.h"
  29
  30 /*
  31  * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
  32  * irrelevant.
  33  */
  34 #include <linux/io-64-nonatomic-hi-lo.h>
  35
  36 static bool force_enable_dimms;
  37 module_param(force_enable_dimms, bool, S_IRUGO|S_IWUSR);
  38 MODULE_PARM_DESC(force_enable_dimms, "Ignore _STA (ACPI DIMM device) status");
  39
  40 static bool disable_vendor_specific;
  41 module_param(disable_vendor_specific, bool, S_IRUGO);
  42 MODULE_PARM_DESC(disable_vendor_specific,
  43                 "Limit commands to the publicly specified set");
  44
  45 static unsigned long override_dsm_mask;
  46 module_param(override_dsm_mask, ulong, S_IRUGO);
  47 MODULE_PARM_DESC(override_dsm_mask, "Bitmask of allowed NVDIMM DSM functions");
  48
  49 static int default_dsm_family = -1;
  50 module_param(default_dsm_family, int, S_IRUGO);
  51 MODULE_PARM_DESC(default_dsm_family,
  52                 "Try this DSM type first when identifying NVDIMM family");
  53
  54 static bool no_init_ars;
  55 module_param(no_init_ars, bool, 0644);
  56 MODULE_PARM_DESC(no_init_ars, "Skip ARS run at nfit init time");
  57
  58 LIST_HEAD(acpi_descs);
  59 DEFINE_MUTEX(acpi_desc_lock);
  60
  61 static struct workqueue_struct *nfit_wq;
  62
  63 struct nfit_table_prev {
  64         struct list_head spas;
  65         struct list_head memdevs;
  66         struct list_head dcrs;
  67         struct list_head bdws;
  68         struct list_head idts;
  69         struct list_head flushes;
  70 };
  71
  72 static guid_t nfit_uuid[NFIT_UUID_MAX];
  73
  74 const guid_t *to_nfit_uuid(enum nfit_uuids id)
  75 {
  76         return &nfit_uuid[id];
  77 }
  78 EXPORT_SYMBOL(to_nfit_uuid);
  79
  80 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
  81 {
  82         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
  83
  84         /*
  85          * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
  86          * acpi_device.
  87          */
  88         if (!nd_desc->provider_name
  89                         || strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
  90                 return NULL;
  91
  92         return to_acpi_device(acpi_desc->dev);
  93 }
  94
  95 static int xlat_bus_status(void *buf, unsigned int cmd, u32 status)
  96 {
  97         struct nd_cmd_clear_error *clear_err;
  98         struct nd_cmd_ars_status *ars_status;
  99         u16 flags;
 100
 101         switch (cmd) {
 102         case ND_CMD_ARS_CAP:
 103                 if ((status & 0xffff) == NFIT_ARS_CAP_NONE)
 104                         return -ENOTTY;
 105
 106                 /* Command failed */
 107                 if (status & 0xffff)
 108                         return -EIO;
 109
 110                 /* No supported scan types for this range */
 111                 flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
 112                 if ((status >> 16 & flags) == 0)
 113                         return -ENOTTY;
 114                 return 0;
 115         case ND_CMD_ARS_START:
 116                 /* ARS is in progress */
 117                 if ((status & 0xffff) == NFIT_ARS_START_BUSY)
 118                         return -EBUSY;
 119
 120                 /* Command failed */
 121                 if (status & 0xffff)
 122                         return -EIO;
 123                 return 0;
 124         case ND_CMD_ARS_STATUS:
 125                 ars_status = buf;
 126                 /* Command failed */
 127                 if (status & 0xffff)
 128                         return -EIO;
 129                 /* Check extended status (Upper two bytes) */
 130                 if (status == NFIT_ARS_STATUS_DONE)
 131                         return 0;
 132
 133                 /* ARS is in progress */
 134                 if (status == NFIT_ARS_STATUS_BUSY)
 135                         return -EBUSY;
 136
 137                 /* No ARS performed for the current boot */
 138                 if (status == NFIT_ARS_STATUS_NONE)
 139                         return -EAGAIN;
 140
 141                 /*
 142                  * ARS interrupted, either we overflowed or some other
 143                  * agent wants the scan to stop.  If we didn't overflow
 144                  * then just continue with the returned results.
 145                  */
 146                 if (status == NFIT_ARS_STATUS_INTR) {
 147                         if (ars_status->out_length >= 40 && (ars_status->flags
 148                                                 & NFIT_ARS_F_OVERFLOW))
 149                                 return -ENOSPC;
 150                         return 0;
 151                 }
 152
 153                 /* Unknown status */
 154                 if (status >> 16)
 155                         return -EIO;
 156                 return 0;
 157         case ND_CMD_CLEAR_ERROR:
 158                 clear_err = buf;
 159                 if (status & 0xffff)
 160                         return -EIO;
 161                 if (!clear_err->cleared)
 162                         return -EIO;
 163                 if (clear_err->length > clear_err->cleared)
 164                         return clear_err->cleared;
 165                 return 0;
 166         default:
 167                 break;
 168         }
 169
 170         /* all other non-zero status results in an error */
 171         if (status)
 172                 return -EIO;
 173         return 0;
 174 }
 175
 176 #define ACPI_LABELS_LOCKED 3
 177
 178 static int xlat_nvdimm_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
 179                 u32 status)
 180 {
 181         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 182
 183         switch (cmd) {
 184         case ND_CMD_GET_CONFIG_SIZE:
 185                 /*
 186                  * In the _LSI, _LSR, _LSW case the locked status is
 187                  * communicated via the read/write commands
 188                  */
 189                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags))
 190                         break;
 191
 192                 if (status >> 16 & ND_CONFIG_LOCKED)
 193                         return -EACCES;
 194                 break;
 195         case ND_CMD_GET_CONFIG_DATA:
 196                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
 197                                 && status == ACPI_LABELS_LOCKED)
 198                         return -EACCES;
 199                 break;
 200         case ND_CMD_SET_CONFIG_DATA:
 201                 if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags)
 202                                 && status == ACPI_LABELS_LOCKED)
 203                         return -EACCES;
 204                 break;
 205         default:
 206                 break;
 207         }
 208
 209         /* all other non-zero status results in an error */
 210         if (status)
 211                 return -EIO;
 212         return 0;
 213 }
 214
 215 static int xlat_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
 216                 u32 status)
 217 {
 218         if (!nvdimm)
 219                 return xlat_bus_status(buf, cmd, status);
 220         return xlat_nvdimm_status(nvdimm, buf, cmd, status);
 221 }
 222
 223 /* convert _LS{I,R} packages to the buffer object acpi_nfit_ctl expects */
 224 static union acpi_object *pkg_to_buf(union acpi_object *pkg)
 225 {
 226         int i;
 227         void *dst;
 228         size_t size = 0;
 229         union acpi_object *buf = NULL;
 230
 231         if (pkg->type != ACPI_TYPE_PACKAGE) {
 232                 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
 233                                 pkg->type);
 234                 goto err;
 235         }
 236
 237         for (i = 0; i < pkg->package.count; i++) {
 238                 union acpi_object *obj = &pkg->package.elements[i];
 239
 240                 if (obj->type == ACPI_TYPE_INTEGER)
 241                         size += 4;
 242                 else if (obj->type == ACPI_TYPE_BUFFER)
 243                         size += obj->buffer.length;
 244                 else {
 245                         WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
 246                                         obj->type);
 247                         goto err;
 248                 }
 249         }
 250
 251         buf = ACPI_ALLOCATE(sizeof(*buf) + size);
 252         if (!buf)
 253                 goto err;
 254
 255         dst = buf + 1;
 256         buf->type = ACPI_TYPE_BUFFER;
 257         buf->buffer.length = size;
 258         buf->buffer.pointer = dst;
 259         for (i = 0; i < pkg->package.count; i++) {
 260                 union acpi_object *obj = &pkg->package.elements[i];
 261
 262                 if (obj->type == ACPI_TYPE_INTEGER) {
 263                         memcpy(dst, &obj->integer.value, 4);
 264                         dst += 4;
 265                 } else if (obj->type == ACPI_TYPE_BUFFER) {
 266                         memcpy(dst, obj->buffer.pointer, obj->buffer.length);
 267                         dst += obj->buffer.length;
 268                 }
 269         }
 270 err:
 271         ACPI_FREE(pkg);
 272         return buf;
 273 }
 274
 275 static union acpi_object *int_to_buf(union acpi_object *integer)
 276 {
 277         union acpi_object *buf = ACPI_ALLOCATE(sizeof(*buf) + 4);
 278         void *dst = NULL;
 279
 280         if (!buf)
 281                 goto err;
 282
 283         if (integer->type != ACPI_TYPE_INTEGER) {
 284                 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
 285                                 integer->type);
 286                 goto err;
 287         }
 288
 289         dst = buf + 1;
 290         buf->type = ACPI_TYPE_BUFFER;
 291         buf->buffer.length = 4;
 292         buf->buffer.pointer = dst;
 293         memcpy(dst, &integer->integer.value, 4);
 294 err:
 295         ACPI_FREE(integer);
 296         return buf;
 297 }
 298
 299 static union acpi_object *acpi_label_write(acpi_handle handle, u32 offset,
 300                 u32 len, void *data)
 301 {
 302         acpi_status rc;
 303         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
 304         struct acpi_object_list input = {
 305                 .count = 3,
 306                 .pointer = (union acpi_object []) {
 307                         [0] = {
 308                                 .integer.type = ACPI_TYPE_INTEGER,
 309                                 .integer.value = offset,
 310                         },
 311                         [1] = {
 312                                 .integer.type = ACPI_TYPE_INTEGER,
 313                                 .integer.value = len,
 314                         },
 315                         [2] = {
 316                                 .buffer.type = ACPI_TYPE_BUFFER,
 317                                 .buffer.pointer = data,
 318                                 .buffer.length = len,
 319                         },
 320                 },
 321         };
 322
 323         rc = acpi_evaluate_object(handle, "_LSW", &input, &buf);
 324         if (ACPI_FAILURE(rc))
 325                 return NULL;
 326         return int_to_buf(buf.pointer);
 327 }
 328
 329 static union acpi_object *acpi_label_read(acpi_handle handle, u32 offset,
 330                 u32 len)
 331 {
 332         acpi_status rc;
 333         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
 334         struct acpi_object_list input = {
 335                 .count = 2,
 336                 .pointer = (union acpi_object []) {
 337                         [0] = {
 338                                 .integer.type = ACPI_TYPE_INTEGER,
 339                                 .integer.value = offset,
 340                         },
 341                         [1] = {
 342                                 .integer.type = ACPI_TYPE_INTEGER,
 343                                 .integer.value = len,
 344                         },
 345                 },
 346         };
 347
 348         rc = acpi_evaluate_object(handle, "_LSR", &input, &buf);
 349         if (ACPI_FAILURE(rc))
 350                 return NULL;
 351         return pkg_to_buf(buf.pointer);
 352 }
 353
 354 static union acpi_object *acpi_label_info(acpi_handle handle)
 355 {
 356         acpi_status rc;
 357         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
 358
 359         rc = acpi_evaluate_object(handle, "_LSI", NULL, &buf);
 360         if (ACPI_FAILURE(rc))
 361                 return NULL;
 362         return pkg_to_buf(buf.pointer);
 363 }
 364
 365 static u8 nfit_dsm_revid(unsigned family, unsigned func)
 366 {
 367         static const u8 revid_table[NVDIMM_FAMILY_MAX+1][32] = {
 368                 [NVDIMM_FAMILY_INTEL] = {
 369                         [NVDIMM_INTEL_GET_MODES] = 2,
 370                         [NVDIMM_INTEL_GET_FWINFO] = 2,
 371                         [NVDIMM_INTEL_START_FWUPDATE] = 2,
 372                         [NVDIMM_INTEL_SEND_FWUPDATE] = 2,
 373                         [NVDIMM_INTEL_FINISH_FWUPDATE] = 2,
 374                         [NVDIMM_INTEL_QUERY_FWUPDATE] = 2,
 375                         [NVDIMM_INTEL_SET_THRESHOLD] = 2,
 376                         [NVDIMM_INTEL_INJECT_ERROR] = 2,
 377                         [NVDIMM_INTEL_GET_SECURITY_STATE] = 2,
 378                         [NVDIMM_INTEL_SET_PASSPHRASE] = 2,
 379                         [NVDIMM_INTEL_DISABLE_PASSPHRASE] = 2,
 380                         [NVDIMM_INTEL_UNLOCK_UNIT] = 2,
 381                         [NVDIMM_INTEL_FREEZE_LOCK] = 2,
 382                         [NVDIMM_INTEL_SECURE_ERASE] = 2,
 383                         [NVDIMM_INTEL_OVERWRITE] = 2,
 384                         [NVDIMM_INTEL_QUERY_OVERWRITE] = 2,
 385                         [NVDIMM_INTEL_SET_MASTER_PASSPHRASE] = 2,
 386                         [NVDIMM_INTEL_MASTER_SECURE_ERASE] = 2,
 387                 },
 388         };
 389         u8 id;
 390
 391         if (family > NVDIMM_FAMILY_MAX)
 392                 return 0;
 393         if (func > 31)
 394                 return 0;
 395         id = revid_table[family][func];
 396         if (id == 0)
 397                 return 1; /* default */
 398         return id;
 399 }
 400
 401 static bool payload_dumpable(struct nvdimm *nvdimm, unsigned int func)
 402 {
 403         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 404
 405         if (nfit_mem && nfit_mem->family == NVDIMM_FAMILY_INTEL
 406                         && func >= NVDIMM_INTEL_GET_SECURITY_STATE
 407                         && func <= NVDIMM_INTEL_MASTER_SECURE_ERASE)
 408                 return IS_ENABLED(CONFIG_NFIT_SECURITY_DEBUG);
 409         return true;
 410 }
 411
 412 static int cmd_to_func(struct nfit_mem *nfit_mem, unsigned int cmd,
 413                 struct nd_cmd_pkg *call_pkg)
 414 {
 415         if (call_pkg) {
 416                 int i;
 417
 418                 if (nfit_mem->family != call_pkg->nd_family)
 419                         return -ENOTTY;
 420
 421                 for (i = 0; i < ARRAY_SIZE(call_pkg->nd_reserved2); i++)
 422                         if (call_pkg->nd_reserved2[i])
 423                                 return -EINVAL;
 424                 return call_pkg->nd_command;
 425         }
 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 (nvdimm) {
 458                 struct acpi_device *adev = nfit_mem->adev;
 459
 460                 if (!adev)
 461                         return -ENOTTY;
 462
 463                 if (cmd == ND_CMD_CALL)
 464                         call_pkg = buf;
 465                 func = cmd_to_func(nfit_mem, cmd, call_pkg);
 466                 if (func < 0)
 467                         return func;
 468                 dimm_name = nvdimm_name(nvdimm);
 469                 cmd_name = nvdimm_cmd_name(cmd);
 470                 cmd_mask = nvdimm_cmd_mask(nvdimm);
 471                 dsm_mask = nfit_mem->dsm_mask;
 472                 desc = nd_cmd_dimm_desc(cmd);
 473                 guid = to_nfit_uuid(nfit_mem->family);
 474                 handle = adev->handle;
 475         } else {
 476                 struct acpi_device *adev = to_acpi_dev(acpi_desc);
 477
 478                 func = cmd;
 479                 cmd_name = nvdimm_bus_cmd_name(cmd);
 480                 cmd_mask = nd_desc->cmd_mask;
 481                 dsm_mask = cmd_mask;
 482                 if (cmd == ND_CMD_CALL)
 483                         dsm_mask = nd_desc->bus_dsm_mask;
 484                 desc = nd_cmd_bus_desc(cmd);
 485                 guid = to_nfit_uuid(NFIT_DEV_BUS);
 486                 handle = adev->handle;
 487                 dimm_name = "bus";
 488         }
 489
 490         if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
 491                 return -ENOTTY;
 492
 493         /*
 494          * Check for a valid command.  For ND_CMD_CALL, we also have to
 495          * make sure that the DSM function is supported.
 496          */
 497         if (cmd == ND_CMD_CALL && !test_bit(func, &dsm_mask))
 498                 return -ENOTTY;
 499         else if (!test_bit(cmd, &cmd_mask))
 500                 return -ENOTTY;
 501
 502         in_obj.type = ACPI_TYPE_PACKAGE;
 503         in_obj.package.count = 1;
 504         in_obj.package.elements = &in_buf;
 505         in_buf.type = ACPI_TYPE_BUFFER;
 506         in_buf.buffer.pointer = buf;
 507         in_buf.buffer.length = 0;
 508
 509         /* libnvdimm has already validated the input envelope */
 510         for (i = 0; i < desc->in_num; i++)
 511                 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
 512                                 i, buf);
 513
 514         if (call_pkg) {
 515                 /* skip over package wrapper */
 516                 in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
 517                 in_buf.buffer.length = call_pkg->nd_size_in;
 518         }
 519
 520         dev_dbg(dev, "%s cmd: %d: func: %d input length: %d\n",
 521                 dimm_name, cmd, func, in_buf.buffer.length);
 522         if (payload_dumpable(nvdimm, func))
 523                 print_hex_dump_debug("nvdimm in  ", DUMP_PREFIX_OFFSET, 4, 4,
 524                                 in_buf.buffer.pointer,
 525                                 min_t(u32, 256, in_buf.buffer.length), true);
 526
 527         /* call the BIOS, prefer the named methods over _DSM if available */
 528         if (nvdimm && cmd == ND_CMD_GET_CONFIG_SIZE
 529                         && test_bit(NFIT_MEM_LSR, &nfit_mem->flags))
 530                 out_obj = acpi_label_info(handle);
 531         else if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA
 532                         && test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
 533                 struct nd_cmd_get_config_data_hdr *p = buf;
 534
 535                 out_obj = acpi_label_read(handle, p->in_offset, p->in_length);
 536         } else if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA
 537                         && test_bit(NFIT_MEM_LSW, &nfit_mem->flags)) {
 538                 struct nd_cmd_set_config_hdr *p = buf;
 539
 540                 out_obj = acpi_label_write(handle, p->in_offset, p->in_length,
 541                                 p->in_buf);
 542         } else {
 543                 u8 revid;
 544
 545                 if (nvdimm)
 546                         revid = nfit_dsm_revid(nfit_mem->family, func);
 547                 else
 548                         revid = 1;
 549                 out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
 550         }
 551
 552         if (!out_obj) {
 553                 dev_dbg(dev, "%s _DSM failed cmd: %s\n", dimm_name, cmd_name);
 554                 return -EINVAL;
 555         }
 556
 557         if (call_pkg) {
 558                 call_pkg->nd_fw_size = out_obj->buffer.length;
 559                 memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
 560                         out_obj->buffer.pointer,
 561                         min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
 562
 563                 ACPI_FREE(out_obj);
 564                 /*
 565                  * Need to support FW function w/o known size in advance.
 566                  * Caller can determine required size based upon nd_fw_size.
 567                  * If we return an error (like elsewhere) then caller wouldn't
 568                  * be able to rely upon data returned to make calculation.
 569                  */
 570                 if (cmd_rc)
 571                         *cmd_rc = 0;
 572                 return 0;
 573         }
 574
 575         if (out_obj->package.type != ACPI_TYPE_BUFFER) {
 576                 dev_dbg(dev, "%s unexpected output object type cmd: %s type: %d\n",
 577                                 dimm_name, cmd_name, out_obj->type);
 578                 rc = -EINVAL;
 579                 goto out;
 580         }
 581
 582         dev_dbg(dev, "%s cmd: %s output length: %d\n", dimm_name,
 583                         cmd_name, out_obj->buffer.length);
 584         print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4, 4,
 585                         out_obj->buffer.pointer,
 586                         min_t(u32, 128, out_obj->buffer.length), true);
 587
 588         for (i = 0, offset = 0; i < desc->out_num; i++) {
 589                 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
 590                                 (u32 *) out_obj->buffer.pointer,
 591                                 out_obj->buffer.length - offset);
 592
 593                 if (offset + out_size > out_obj->buffer.length) {
 594                         dev_dbg(dev, "%s output object underflow cmd: %s field: %d\n",
 595                                         dimm_name, cmd_name, i);
 596                         break;
 597                 }
 598
 599                 if (in_buf.buffer.length + offset + out_size > buf_len) {
 600                         dev_dbg(dev, "%s output overrun cmd: %s field: %d\n",
 601                                         dimm_name, cmd_name, i);
 602                         rc = -ENXIO;
 603                         goto out;
 604                 }
 605                 memcpy(buf + in_buf.buffer.length + offset,
 606                                 out_obj->buffer.pointer + offset, out_size);
 607                 offset += out_size;
 608         }
 609
 610         /*
 611          * Set fw_status for all the commands with a known format to be
 612          * later interpreted by xlat_status().
 613          */
 614         if (i >= 1 && ((!nvdimm && cmd >= ND_CMD_ARS_CAP
 615                                         && cmd <= ND_CMD_CLEAR_ERROR)
 616                                 || (nvdimm && cmd >= ND_CMD_SMART
 617                                         && cmd <= ND_CMD_VENDOR)))
 618                 fw_status = *(u32 *) out_obj->buffer.pointer;
 619
 620         if (offset + in_buf.buffer.length < buf_len) {
 621                 if (i >= 1) {
 622                         /*
 623                          * status valid, return the number of bytes left
 624                          * unfilled in the output buffer
 625                          */
 626                         rc = buf_len - offset - in_buf.buffer.length;
 627                         if (cmd_rc)
 628                                 *cmd_rc = xlat_status(nvdimm, buf, cmd,
 629                                                 fw_status);
 630                 } else {
 631                         dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
 632                                         __func__, dimm_name, cmd_name, buf_len,
 633                                         offset);
 634                         rc = -ENXIO;
 635                 }
 636         } else {
 637                 rc = 0;
 638                 if (cmd_rc)
 639                         *cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status);
 640         }
 641
 642  out:
 643         ACPI_FREE(out_obj);
 644
 645         return rc;
 646 }
 647 EXPORT_SYMBOL_GPL(acpi_nfit_ctl);
 648
 649 static const char *spa_type_name(u16 type)
 650 {
 651         static const char *to_name[] = {
 652                 [NFIT_SPA_VOLATILE] = "volatile",
 653                 [NFIT_SPA_PM] = "pmem",
 654                 [NFIT_SPA_DCR] = "dimm-control-region",
 655                 [NFIT_SPA_BDW] = "block-data-window",
 656                 [NFIT_SPA_VDISK] = "volatile-disk",
 657                 [NFIT_SPA_VCD] = "volatile-cd",
 658                 [NFIT_SPA_PDISK] = "persistent-disk",
 659                 [NFIT_SPA_PCD] = "persistent-cd",
 660
 661         };
 662
 663         if (type > NFIT_SPA_PCD)
 664                 return "unknown";
 665
 666         return to_name[type];
 667 }
 668
 669 int nfit_spa_type(struct acpi_nfit_system_address *spa)
 670 {
 671         int i;
 672
 673         for (i = 0; i < NFIT_UUID_MAX; i++)
 674                 if (guid_equal(to_nfit_uuid(i), (guid_t *)&spa->range_guid))
 675                         return i;
 676         return -1;
 677 }
 678
 679 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
 680                 struct nfit_table_prev *prev,
 681                 struct acpi_nfit_system_address *spa)
 682 {
 683         struct device *dev = acpi_desc->dev;
 684         struct nfit_spa *nfit_spa;
 685
 686         if (spa->header.length != sizeof(*spa))
 687                 return false;
 688
 689         list_for_each_entry(nfit_spa, &prev->spas, list) {
 690                 if (memcmp(nfit_spa->spa, spa, sizeof(*spa)) == 0) {
 691                         list_move_tail(&nfit_spa->list, &acpi_desc->spas);
 692                         return true;
 693                 }
 694         }
 695
 696         nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof(*spa),
 697                         GFP_KERNEL);
 698         if (!nfit_spa)
 699                 return false;
 700         INIT_LIST_HEAD(&nfit_spa->list);
 701         memcpy(nfit_spa->spa, spa, sizeof(*spa));
 702         list_add_tail(&nfit_spa->list, &acpi_desc->spas);
 703         dev_dbg(dev, "spa index: %d type: %s\n",
 704                         spa->range_index,
 705                         spa_type_name(nfit_spa_type(spa)));
 706         return true;
 707 }
 708
 709 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
 710                 struct nfit_table_prev *prev,
 711                 struct acpi_nfit_memory_map *memdev)
 712 {
 713         struct device *dev = acpi_desc->dev;
 714         struct nfit_memdev *nfit_memdev;
 715
 716         if (memdev->header.length != sizeof(*memdev))
 717                 return false;
 718
 719         list_for_each_entry(nfit_memdev, &prev->memdevs, list)
 720                 if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
 721                         list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 722                         return true;
 723                 }
 724
 725         nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
 726                         GFP_KERNEL);
 727         if (!nfit_memdev)
 728                 return false;
 729         INIT_LIST_HEAD(&nfit_memdev->list);
 730         memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
 731         list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 732         dev_dbg(dev, "memdev handle: %#x spa: %d dcr: %d flags: %#x\n",
 733                         memdev->device_handle, memdev->range_index,
 734                         memdev->region_index, memdev->flags);
 735         return true;
 736 }
 737
 738 int nfit_get_smbios_id(u32 device_handle, u16 *flags)
 739 {
 740         struct acpi_nfit_memory_map *memdev;
 741         struct acpi_nfit_desc *acpi_desc;
 742         struct nfit_mem *nfit_mem;
 743         u16 physical_id;
 744
 745         mutex_lock(&acpi_desc_lock);
 746         list_for_each_entry(acpi_desc, &acpi_descs, list) {
 747                 mutex_lock(&acpi_desc->init_mutex);
 748                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
 749                         memdev = __to_nfit_memdev(nfit_mem);
 750                         if (memdev->device_handle == device_handle) {
 751                                 *flags = memdev->flags;
 752                                 physical_id = memdev->physical_id;
 753                                 mutex_unlock(&acpi_desc->init_mutex);
 754                                 mutex_unlock(&acpi_desc_lock);
 755                                 return physical_id;
 756                         }
 757                 }
 758                 mutex_unlock(&acpi_desc->init_mutex);
 759         }
 760         mutex_unlock(&acpi_desc_lock);
 761
 762         return -ENODEV;
 763 }
 764 EXPORT_SYMBOL_GPL(nfit_get_smbios_id);
 765
 766 /*
 767  * An implementation may provide a truncated control region if no block windows
 768  * are defined.
 769  */
 770 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
 771 {
 772         if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
 773                                 window_size))
 774                 return 0;
 775         if (dcr->windows)
 776                 return sizeof(*dcr);
 777         return offsetof(struct acpi_nfit_control_region, window_size);
 778 }
 779
 780 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
 781                 struct nfit_table_prev *prev,
 782                 struct acpi_nfit_control_region *dcr)
 783 {
 784         struct device *dev = acpi_desc->dev;
 785         struct nfit_dcr *nfit_dcr;
 786
 787         if (!sizeof_dcr(dcr))
 788                 return false;
 789
 790         list_for_each_entry(nfit_dcr, &prev->dcrs, list)
 791                 if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
 792                         list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 793                         return true;
 794                 }
 795
 796         nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
 797                         GFP_KERNEL);
 798         if (!nfit_dcr)
 799                 return false;
 800         INIT_LIST_HEAD(&nfit_dcr->list);
 801         memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
 802         list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 803         dev_dbg(dev, "dcr index: %d windows: %d\n",
 804                         dcr->region_index, dcr->windows);
 805         return true;
 806 }
 807
 808 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
 809                 struct nfit_table_prev *prev,
 810                 struct acpi_nfit_data_region *bdw)
 811 {
 812         struct device *dev = acpi_desc->dev;
 813         struct nfit_bdw *nfit_bdw;
 814
 815         if (bdw->header.length != sizeof(*bdw))
 816                 return false;
 817         list_for_each_entry(nfit_bdw, &prev->bdws, list)
 818                 if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
 819                         list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
 820                         return true;
 821                 }
 822
 823         nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
 824                         GFP_KERNEL);
 825         if (!nfit_bdw)
 826                 return false;
 827         INIT_LIST_HEAD(&nfit_bdw->list);
 828         memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
 829         list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
 830         dev_dbg(dev, "bdw dcr: %d windows: %d\n",
 831                         bdw->region_index, bdw->windows);
 832         return true;
 833 }
 834
 835 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
 836 {
 837         if (idt->header.length < sizeof(*idt))
 838                 return 0;
 839         return sizeof(*idt) + sizeof(u32) * (idt->line_count - 1);
 840 }
 841
 842 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
 843                 struct nfit_table_prev *prev,
 844                 struct acpi_nfit_interleave *idt)
 845 {
 846         struct device *dev = acpi_desc->dev;
 847         struct nfit_idt *nfit_idt;
 848
 849         if (!sizeof_idt(idt))
 850                 return false;
 851
 852         list_for_each_entry(nfit_idt, &prev->idts, list) {
 853                 if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
 854                         continue;
 855
 856                 if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
 857                         list_move_tail(&nfit_idt->list, &acpi_desc->idts);
 858                         return true;
 859                 }
 860         }
 861
 862         nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
 863                         GFP_KERNEL);
 864         if (!nfit_idt)
 865                 return false;
 866         INIT_LIST_HEAD(&nfit_idt->list);
 867         memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
 868         list_add_tail(&nfit_idt->list, &acpi_desc->idts);
 869         dev_dbg(dev, "idt index: %d num_lines: %d\n",
 870                         idt->interleave_index, idt->line_count);
 871         return true;
 872 }
 873
 874 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
 875 {
 876         if (flush->header.length < sizeof(*flush))
 877                 return 0;
 878         return sizeof(*flush) + sizeof(u64) * (flush->hint_count - 1);
 879 }
 880
 881 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
 882                 struct nfit_table_prev *prev,
 883                 struct acpi_nfit_flush_address *flush)
 884 {
 885         struct device *dev = acpi_desc->dev;
 886         struct nfit_flush *nfit_flush;
 887
 888         if (!sizeof_flush(flush))
 889                 return false;
 890
 891         list_for_each_entry(nfit_flush, &prev->flushes, list) {
 892                 if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
 893                         continue;
 894
 895                 if (memcmp(nfit_flush->flush, flush,
 896                                         sizeof_flush(flush)) == 0) {
 897                         list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
 898                         return true;
 899                 }
 900         }
 901
 902         nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
 903                         + sizeof_flush(flush), GFP_KERNEL);
 904         if (!nfit_flush)
 905                 return false;
 906         INIT_LIST_HEAD(&nfit_flush->list);
 907         memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
 908         list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
 909         dev_dbg(dev, "nfit_flush handle: %d hint_count: %d\n",
 910                         flush->device_handle, flush->hint_count);
 911         return true;
 912 }
 913
 914 static bool add_platform_cap(struct acpi_nfit_desc *acpi_desc,
 915                 struct acpi_nfit_capabilities *pcap)
 916 {
 917         struct device *dev = acpi_desc->dev;
 918         u32 mask;
 919
 920         mask = (1 << (pcap->highest_capability + 1)) - 1;
 921         acpi_desc->platform_cap = pcap->capabilities & mask;
 922         dev_dbg(dev, "cap: %#x\n", acpi_desc->platform_cap);
 923         return true;
 924 }
 925
 926 static void *add_table(struct acpi_nfit_desc *acpi_desc,
 927                 struct nfit_table_prev *prev, void *table, const void *end)
 928 {
 929         struct device *dev = acpi_desc->dev;
 930         struct acpi_nfit_header *hdr;
 931         void *err = ERR_PTR(-ENOMEM);
 932
 933         if (table >= end)
 934                 return NULL;
 935
 936         hdr = table;
 937         if (!hdr->length) {
 938                 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
 939                         hdr->type);
 940                 return NULL;
 941         }
 942
 943         switch (hdr->type) {
 944         case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
 945                 if (!add_spa(acpi_desc, prev, table))
 946                         return err;
 947                 break;
 948         case ACPI_NFIT_TYPE_MEMORY_MAP:
 949                 if (!add_memdev(acpi_desc, prev, table))
 950                         return err;
 951                 break;
 952         case ACPI_NFIT_TYPE_CONTROL_REGION:
 953                 if (!add_dcr(acpi_desc, prev, table))
 954                         return err;
 955                 break;
 956         case ACPI_NFIT_TYPE_DATA_REGION:
 957                 if (!add_bdw(acpi_desc, prev, table))
 958                         return err;
 959                 break;
 960         case ACPI_NFIT_TYPE_INTERLEAVE:
 961                 if (!add_idt(acpi_desc, prev, table))
 962                         return err;
 963                 break;
 964         case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
 965                 if (!add_flush(acpi_desc, prev, table))
 966                         return err;
 967                 break;
 968         case ACPI_NFIT_TYPE_SMBIOS:
 969                 dev_dbg(dev, "smbios\n");
 970                 break;
 971         case ACPI_NFIT_TYPE_CAPABILITIES:
 972                 if (!add_platform_cap(acpi_desc, table))
 973                         return err;
 974                 break;
 975         default:
 976                 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
 977                 break;
 978         }
 979
 980         return table + hdr->length;
 981 }
 982
 983 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
 984                 struct nfit_mem *nfit_mem)
 985 {
 986         u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
 987         u16 dcr = nfit_mem->dcr->region_index;
 988         struct nfit_spa *nfit_spa;
 989
 990         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
 991                 u16 range_index = nfit_spa->spa->range_index;
 992                 int type = nfit_spa_type(nfit_spa->spa);
 993                 struct nfit_memdev *nfit_memdev;
 994
 995                 if (type != NFIT_SPA_BDW)
 996                         continue;
 997
 998                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 999                         if (nfit_memdev->memdev->range_index != range_index)
1000                                 continue;
1001                         if (nfit_memdev->memdev->device_handle != device_handle)
1002                                 continue;
1003                         if (nfit_memdev->memdev->region_index != dcr)
1004                                 continue;
1005
1006                         nfit_mem->spa_bdw = nfit_spa->spa;
1007                         return;
1008                 }
1009         }
1010
1011         dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
1012                         nfit_mem->spa_dcr->range_index);
1013         nfit_mem->bdw = NULL;
1014 }
1015
1016 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
1017                 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
1018 {
1019         u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
1020         struct nfit_memdev *nfit_memdev;
1021         struct nfit_bdw *nfit_bdw;
1022         struct nfit_idt *nfit_idt;
1023         u16 idt_idx, range_index;
1024
1025         list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
1026                 if (nfit_bdw->bdw->region_index != dcr)
1027                         continue;
1028                 nfit_mem->bdw = nfit_bdw->bdw;
1029                 break;
1030         }
1031
1032         if (!nfit_mem->bdw)
1033                 return;
1034
1035         nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
1036
1037         if (!nfit_mem->spa_bdw)
1038                 return;
1039
1040         range_index = nfit_mem->spa_bdw->range_index;
1041         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1042                 if (nfit_memdev->memdev->range_index != range_index ||
1043                                 nfit_memdev->memdev->region_index != dcr)
1044                         continue;
1045                 nfit_mem->memdev_bdw = nfit_memdev->memdev;
1046                 idt_idx = nfit_memdev->memdev->interleave_index;
1047                 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1048                         if (nfit_idt->idt->interleave_index != idt_idx)
1049                                 continue;
1050                         nfit_mem->idt_bdw = nfit_idt->idt;
1051                         break;
1052                 }
1053                 break;
1054         }
1055 }
1056
1057 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
1058                 struct acpi_nfit_system_address *spa)
1059 {
1060         struct nfit_mem *nfit_mem, *found;
1061         struct nfit_memdev *nfit_memdev;
1062         int type = spa ? nfit_spa_type(spa) : 0;
1063
1064         switch (type) {
1065         case NFIT_SPA_DCR:
1066         case NFIT_SPA_PM:
1067                 break;
1068         default:
1069                 if (spa)
1070                         return 0;
1071         }
1072
1073         /*
1074          * This loop runs in two modes, when a dimm is mapped the loop
1075          * adds memdev associations to an existing dimm, or creates a
1076          * dimm. In the unmapped dimm case this loop sweeps for memdev
1077          * instances with an invalid / zero range_index and adds those
1078          * dimms without spa associations.
1079          */
1080         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1081                 struct nfit_flush *nfit_flush;
1082                 struct nfit_dcr *nfit_dcr;
1083                 u32 device_handle;
1084                 u16 dcr;
1085
1086                 if (spa && nfit_memdev->memdev->range_index != spa->range_index)
1087                         continue;
1088                 if (!spa && nfit_memdev->memdev->range_index)
1089                         continue;
1090                 found = NULL;
1091                 dcr = nfit_memdev->memdev->region_index;
1092                 device_handle = nfit_memdev->memdev->device_handle;
1093                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1094                         if (__to_nfit_memdev(nfit_mem)->device_handle
1095                                         == device_handle) {
1096                                 found = nfit_mem;
1097                                 break;
1098                         }
1099
1100                 if (found)
1101                         nfit_mem = found;
1102                 else {
1103                         nfit_mem = devm_kzalloc(acpi_desc->dev,
1104                                         sizeof(*nfit_mem), GFP_KERNEL);
1105                         if (!nfit_mem)
1106                                 return -ENOMEM;
1107                         INIT_LIST_HEAD(&nfit_mem->list);
1108                         nfit_mem->acpi_desc = acpi_desc;
1109                         list_add(&nfit_mem->list, &acpi_desc->dimms);
1110                 }
1111
1112                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1113                         if (nfit_dcr->dcr->region_index != dcr)
1114                                 continue;
1115                         /*
1116                          * Record the control region for the dimm.  For
1117                          * the ACPI 6.1 case, where there are separate
1118                          * control regions for the pmem vs blk
1119                          * interfaces, be sure to record the extended
1120                          * blk details.
1121                          */
1122                         if (!nfit_mem->dcr)
1123                                 nfit_mem->dcr = nfit_dcr->dcr;
1124                         else if (nfit_mem->dcr->windows == 0
1125                                         && nfit_dcr->dcr->windows)
1126                                 nfit_mem->dcr = nfit_dcr->dcr;
1127                         break;
1128                 }
1129
1130                 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
1131                         struct acpi_nfit_flush_address *flush;
1132                         u16 i;
1133
1134                         if (nfit_flush->flush->device_handle != device_handle)
1135                                 continue;
1136                         nfit_mem->nfit_flush = nfit_flush;
1137                         flush = nfit_flush->flush;
1138                         nfit_mem->flush_wpq = devm_kcalloc(acpi_desc->dev,
1139                                         flush->hint_count,
1140                                         sizeof(struct resource),
1141                                         GFP_KERNEL);
1142                         if (!nfit_mem->flush_wpq)
1143                                 return -ENOMEM;
1144                         for (i = 0; i < flush->hint_count; i++) {
1145                                 struct resource *res = &nfit_mem->flush_wpq[i];
1146
1147                                 res->start = flush->hint_address[i];
1148                                 res->end = res->start + 8 - 1;
1149                         }
1150                         break;
1151                 }
1152
1153                 if (dcr && !nfit_mem->dcr) {
1154                         dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
1155                                         spa->range_index, dcr);
1156                         return -ENODEV;
1157                 }
1158
1159                 if (type == NFIT_SPA_DCR) {
1160                         struct nfit_idt *nfit_idt;
1161                         u16 idt_idx;
1162
1163                         /* multiple dimms may share a SPA when interleaved */
1164                         nfit_mem->spa_dcr = spa;
1165                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
1166                         idt_idx = nfit_memdev->memdev->interleave_index;
1167                         list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1168                                 if (nfit_idt->idt->interleave_index != idt_idx)
1169                                         continue;
1170                                 nfit_mem->idt_dcr = nfit_idt->idt;
1171                                 break;
1172                         }
1173                         nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
1174                 } else if (type == NFIT_SPA_PM) {
1175                         /*
1176                          * A single dimm may belong to multiple SPA-PM
1177                          * ranges, record at least one in addition to
1178                          * any SPA-DCR range.
1179                          */
1180                         nfit_mem->memdev_pmem = nfit_memdev->memdev;
1181                 } else
1182                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
1183         }
1184
1185         return 0;
1186 }
1187
1188 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
1189 {
1190         struct nfit_mem *a = container_of(_a, typeof(*a), list);
1191         struct nfit_mem *b = container_of(_b, typeof(*b), list);
1192         u32 handleA, handleB;
1193
1194         handleA = __to_nfit_memdev(a)->device_handle;
1195         handleB = __to_nfit_memdev(b)->device_handle;
1196         if (handleA < handleB)
1197                 return -1;
1198         else if (handleA > handleB)
1199                 return 1;
1200         return 0;
1201 }
1202
1203 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
1204 {
1205         struct nfit_spa *nfit_spa;
1206         int rc;
1207
1208
1209         /*
1210          * For each SPA-DCR or SPA-PMEM address range find its
1211          * corresponding MEMDEV(s).  From each MEMDEV find the
1212          * corresponding DCR.  Then, if we're operating on a SPA-DCR,
1213          * try to find a SPA-BDW and a corresponding BDW that references
1214          * the DCR.  Throw it all into an nfit_mem object.  Note, that
1215          * BDWs are optional.
1216          */
1217         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
1218                 rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
1219                 if (rc)
1220                         return rc;
1221         }
1222
1223         /*
1224          * If a DIMM has failed to be mapped into SPA there will be no
1225          * SPA entries above. Find and register all the unmapped DIMMs
1226          * for reporting and recovery purposes.
1227          */
1228         rc = __nfit_mem_init(acpi_desc, NULL);
1229         if (rc)
1230                 return rc;
1231
1232         list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
1233
1234         return 0;
1235 }
1236
1237 static ssize_t bus_dsm_mask_show(struct device *dev,
1238                 struct device_attribute *attr, char *buf)
1239 {
1240         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1241         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1242
1243         return sprintf(buf, "%#lx\n", nd_desc->bus_dsm_mask);
1244 }
1245 static struct device_attribute dev_attr_bus_dsm_mask =
1246                 __ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
1247
1248 static ssize_t revision_show(struct device *dev,
1249                 struct device_attribute *attr, char *buf)
1250 {
1251         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1252         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1253         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1254
1255         return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
1256 }
1257 static DEVICE_ATTR_RO(revision);
1258
1259 static ssize_t hw_error_scrub_show(struct device *dev,
1260                 struct device_attribute *attr, char *buf)
1261 {
1262         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1263         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1264         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1265
1266         return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
1267 }
1268
1269 /*
1270  * The 'hw_error_scrub' attribute can have the following values written to it:
1271  * '0': Switch to the default mode where an exception will only insert
1272  *      the address of the memory error into the poison and badblocks lists.
1273  * '1': Enable a full scrub to happen if an exception for a memory error is
1274  *      received.
1275  */
1276 static ssize_t hw_error_scrub_store(struct device *dev,
1277                 struct device_attribute *attr, const char *buf, size_t size)
1278 {
1279         struct nvdimm_bus_descriptor *nd_desc;
1280         ssize_t rc;
1281         long val;
1282
1283         rc = kstrtol(buf, 0, &val);
1284         if (rc)
1285                 return rc;
1286
1287         device_lock(dev);
1288         nd_desc = dev_get_drvdata(dev);
1289         if (nd_desc) {
1290                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1291
1292                 switch (val) {
1293                 case HW_ERROR_SCRUB_ON:
1294                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
1295                         break;
1296                 case HW_ERROR_SCRUB_OFF:
1297                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
1298                         break;
1299                 default:
1300                         rc = -EINVAL;
1301                         break;
1302                 }
1303         }
1304         device_unlock(dev);
1305         if (rc)
1306                 return rc;
1307         return size;
1308 }
1309 static DEVICE_ATTR_RW(hw_error_scrub);
1310
1311 /*
1312  * This shows the number of full Address Range Scrubs that have been
1313  * completed since driver load time. Userspace can wait on this using
1314  * select/poll etc. A '+' at the end indicates an ARS is in progress
1315  */
1316 static ssize_t scrub_show(struct device *dev,
1317                 struct device_attribute *attr, char *buf)
1318 {
1319         struct nvdimm_bus_descriptor *nd_desc;
1320         ssize_t rc = -ENXIO;
1321
1322         device_lock(dev);
1323         nd_desc = dev_get_drvdata(dev);
1324         if (nd_desc) {
1325                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1326
1327                 mutex_lock(&acpi_desc->init_mutex);
1328                 rc = sprintf(buf, "%d%s", acpi_desc->scrub_count,
1329                                 acpi_desc->scrub_busy
1330                                 && !acpi_desc->cancel ? "+\n" : "\n");
1331                 mutex_unlock(&acpi_desc->init_mutex);
1332         }
1333         device_unlock(dev);
1334         return rc;
1335 }
1336
1337 static ssize_t scrub_store(struct device *dev,
1338                 struct device_attribute *attr, const char *buf, size_t size)
1339 {
1340         struct nvdimm_bus_descriptor *nd_desc;
1341         ssize_t rc;
1342         long val;
1343
1344         rc = kstrtol(buf, 0, &val);
1345         if (rc)
1346                 return rc;
1347         if (val != 1)
1348                 return -EINVAL;
1349
1350         device_lock(dev);
1351         nd_desc = dev_get_drvdata(dev);
1352         if (nd_desc) {
1353                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1354
1355                 rc = acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
1356         }
1357         device_unlock(dev);
1358         if (rc)
1359                 return rc;
1360         return size;
1361 }
1362 static DEVICE_ATTR_RW(scrub);
1363
1364 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1365 {
1366         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1367         const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1368                 | 1 << ND_CMD_ARS_STATUS;
1369
1370         return (nd_desc->cmd_mask & mask) == mask;
1371 }
1372
1373 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1374 {
1375         struct device *dev = container_of(kobj, struct device, kobj);
1376         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1377
1378         if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
1379                 return 0;
1380         return a->mode;
1381 }
1382
1383 static struct attribute *acpi_nfit_attributes[] = {
1384         &dev_attr_revision.attr,
1385         &dev_attr_scrub.attr,
1386         &dev_attr_hw_error_scrub.attr,
1387         &dev_attr_bus_dsm_mask.attr,
1388         NULL,
1389 };
1390
1391 static const struct attribute_group acpi_nfit_attribute_group = {
1392         .name = "nfit",
1393         .attrs = acpi_nfit_attributes,
1394         .is_visible = nfit_visible,
1395 };
1396
1397 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1398         &nvdimm_bus_attribute_group,
1399         &acpi_nfit_attribute_group,
1400         NULL,
1401 };
1402
1403 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1404 {
1405         struct nvdimm *nvdimm = to_nvdimm(dev);
1406         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1407
1408         return __to_nfit_memdev(nfit_mem);
1409 }
1410
1411 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1412 {
1413         struct nvdimm *nvdimm = to_nvdimm(dev);
1414         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1415
1416         return nfit_mem->dcr;
1417 }
1418
1419 static ssize_t handle_show(struct device *dev,
1420                 struct device_attribute *attr, char *buf)
1421 {
1422         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1423
1424         return sprintf(buf, "%#x\n", memdev->device_handle);
1425 }
1426 static DEVICE_ATTR_RO(handle);
1427
1428 static ssize_t phys_id_show(struct device *dev,
1429                 struct device_attribute *attr, char *buf)
1430 {
1431         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1432
1433         return sprintf(buf, "%#x\n", memdev->physical_id);
1434 }
1435 static DEVICE_ATTR_RO(phys_id);
1436
1437 static ssize_t vendor_show(struct device *dev,
1438                 struct device_attribute *attr, char *buf)
1439 {
1440         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1441
1442         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1443 }
1444 static DEVICE_ATTR_RO(vendor);
1445
1446 static ssize_t rev_id_show(struct device *dev,
1447                 struct device_attribute *attr, char *buf)
1448 {
1449         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1450
1451         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1452 }
1453 static DEVICE_ATTR_RO(rev_id);
1454
1455 static ssize_t device_show(struct device *dev,
1456                 struct device_attribute *attr, char *buf)
1457 {
1458         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1459
1460         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1461 }
1462 static DEVICE_ATTR_RO(device);
1463
1464 static ssize_t subsystem_vendor_show(struct device *dev,
1465                 struct device_attribute *attr, char *buf)
1466 {
1467         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1468
1469         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1470 }
1471 static DEVICE_ATTR_RO(subsystem_vendor);
1472
1473 static ssize_t subsystem_rev_id_show(struct device *dev,
1474                 struct device_attribute *attr, char *buf)
1475 {
1476         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1477
1478         return sprintf(buf, "0x%04x\n",
1479                         be16_to_cpu(dcr->subsystem_revision_id));
1480 }
1481 static DEVICE_ATTR_RO(subsystem_rev_id);
1482
1483 static ssize_t subsystem_device_show(struct device *dev,
1484                 struct device_attribute *attr, char *buf)
1485 {
1486         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1487
1488         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1489 }
1490 static DEVICE_ATTR_RO(subsystem_device);
1491
1492 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1493 {
1494         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1495         int formats = 0;
1496
1497         if (nfit_mem->memdev_pmem)
1498                 formats++;
1499         if (nfit_mem->memdev_bdw)
1500                 formats++;
1501         return formats;
1502 }
1503
1504 static ssize_t format_show(struct device *dev,
1505                 struct device_attribute *attr, char *buf)
1506 {
1507         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1508
1509         return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1510 }
1511 static DEVICE_ATTR_RO(format);
1512
1513 static ssize_t format1_show(struct device *dev,
1514                 struct device_attribute *attr, char *buf)
1515 {
1516         u32 handle;
1517         ssize_t rc = -ENXIO;
1518         struct nfit_mem *nfit_mem;
1519         struct nfit_memdev *nfit_memdev;
1520         struct acpi_nfit_desc *acpi_desc;
1521         struct nvdimm *nvdimm = to_nvdimm(dev);
1522         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1523
1524         nfit_mem = nvdimm_provider_data(nvdimm);
1525         acpi_desc = nfit_mem->acpi_desc;
1526         handle = to_nfit_memdev(dev)->device_handle;
1527
1528         /* assumes DIMMs have at most 2 published interface codes */
1529         mutex_lock(&acpi_desc->init_mutex);
1530         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1531                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1532                 struct nfit_dcr *nfit_dcr;
1533
1534                 if (memdev->device_handle != handle)
1535                         continue;
1536
1537                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1538                         if (nfit_dcr->dcr->region_index != memdev->region_index)
1539                                 continue;
1540                         if (nfit_dcr->dcr->code == dcr->code)
1541                                 continue;
1542                         rc = sprintf(buf, "0x%04x\n",
1543                                         le16_to_cpu(nfit_dcr->dcr->code));
1544                         break;
1545                 }
1546                 if (rc != ENXIO)
1547                         break;
1548         }
1549         mutex_unlock(&acpi_desc->init_mutex);
1550         return rc;
1551 }
1552 static DEVICE_ATTR_RO(format1);
1553
1554 static ssize_t formats_show(struct device *dev,
1555                 struct device_attribute *attr, char *buf)
1556 {
1557         struct nvdimm *nvdimm = to_nvdimm(dev);
1558
1559         return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1560 }
1561 static DEVICE_ATTR_RO(formats);
1562
1563 static ssize_t serial_show(struct device *dev,
1564                 struct device_attribute *attr, char *buf)
1565 {
1566         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1567
1568         return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1569 }
1570 static DEVICE_ATTR_RO(serial);
1571
1572 static ssize_t family_show(struct device *dev,
1573                 struct device_attribute *attr, char *buf)
1574 {
1575         struct nvdimm *nvdimm = to_nvdimm(dev);
1576         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1577
1578         if (nfit_mem->family < 0)
1579                 return -ENXIO;
1580         return sprintf(buf, "%d\n", nfit_mem->family);
1581 }
1582 static DEVICE_ATTR_RO(family);
1583
1584 static ssize_t dsm_mask_show(struct device *dev,
1585                 struct device_attribute *attr, char *buf)
1586 {
1587         struct nvdimm *nvdimm = to_nvdimm(dev);
1588         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1589
1590         if (nfit_mem->family < 0)
1591                 return -ENXIO;
1592         return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1593 }
1594 static DEVICE_ATTR_RO(dsm_mask);
1595
1596 static ssize_t flags_show(struct device *dev,
1597                 struct device_attribute *attr, char *buf)
1598 {
1599         struct nvdimm *nvdimm = to_nvdimm(dev);
1600         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1601         u16 flags = __to_nfit_memdev(nfit_mem)->flags;
1602
1603         if (test_bit(NFIT_MEM_DIRTY, &nfit_mem->flags))
1604                 flags |= ACPI_NFIT_MEM_FLUSH_FAILED;
1605
1606         return sprintf(buf, "%s%s%s%s%s%s%s\n",
1607                 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1608                 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1609                 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1610                 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1611                 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1612                 flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1613                 flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1614 }
1615 static DEVICE_ATTR_RO(flags);
1616
1617 static ssize_t id_show(struct device *dev,
1618                 struct device_attribute *attr, char *buf)
1619 {
1620         struct nvdimm *nvdimm = to_nvdimm(dev);
1621         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1622
1623         return sprintf(buf, "%s\n", nfit_mem->id);
1624 }
1625 static DEVICE_ATTR_RO(id);
1626
1627 static ssize_t dirty_shutdown_show(struct device *dev,
1628                 struct device_attribute *attr, char *buf)
1629 {
1630         struct nvdimm *nvdimm = to_nvdimm(dev);
1631         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1632
1633         return sprintf(buf, "%d\n", nfit_mem->dirty_shutdown);
1634 }
1635 static DEVICE_ATTR_RO(dirty_shutdown);
1636
1637 static struct attribute *acpi_nfit_dimm_attributes[] = {
1638         &dev_attr_handle.attr,
1639         &dev_attr_phys_id.attr,
1640         &dev_attr_vendor.attr,
1641         &dev_attr_device.attr,
1642         &dev_attr_rev_id.attr,
1643         &dev_attr_subsystem_vendor.attr,
1644         &dev_attr_subsystem_device.attr,
1645         &dev_attr_subsystem_rev_id.attr,
1646         &dev_attr_format.attr,
1647         &dev_attr_formats.attr,
1648         &dev_attr_format1.attr,
1649         &dev_attr_serial.attr,
1650         &dev_attr_flags.attr,
1651         &dev_attr_id.attr,
1652         &dev_attr_family.attr,
1653         &dev_attr_dsm_mask.attr,
1654         &dev_attr_dirty_shutdown.attr,
1655         NULL,
1656 };
1657
1658 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1659                 struct attribute *a, int n)
1660 {
1661         struct device *dev = container_of(kobj, struct device, kobj);
1662         struct nvdimm *nvdimm = to_nvdimm(dev);
1663         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1664
1665         if (!to_nfit_dcr(dev)) {
1666                 /* Without a dcr only the memdev attributes can be surfaced */
1667                 if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1668                                 || a == &dev_attr_flags.attr
1669                                 || a == &dev_attr_family.attr
1670                                 || a == &dev_attr_dsm_mask.attr)
1671                         return a->mode;
1672                 return 0;
1673         }
1674
1675         if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1676                 return 0;
1677
1678         if (!test_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags)
1679                         && a == &dev_attr_dirty_shutdown.attr)
1680                 return 0;
1681
1682         return a->mode;
1683 }
1684
1685 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1686         .name = "nfit",
1687         .attrs = acpi_nfit_dimm_attributes,
1688         .is_visible = acpi_nfit_dimm_attr_visible,
1689 };
1690
1691 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1692         &nvdimm_attribute_group,
1693         &nd_device_attribute_group,
1694         &acpi_nfit_dimm_attribute_group,
1695         NULL,
1696 };
1697
1698 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1699                 u32 device_handle)
1700 {
1701         struct nfit_mem *nfit_mem;
1702
1703         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1704                 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1705                         return nfit_mem->nvdimm;
1706
1707         return NULL;
1708 }
1709
1710 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1711 {
1712         struct nfit_mem *nfit_mem;
1713         struct acpi_nfit_desc *acpi_desc;
1714
1715         dev_dbg(dev->parent, "%s: event: %d\n", dev_name(dev),
1716                         event);
1717
1718         if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1719                 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1720                                 event);
1721                 return;
1722         }
1723
1724         acpi_desc = dev_get_drvdata(dev->parent);
1725         if (!acpi_desc)
1726                 return;
1727
1728         /*
1729          * If we successfully retrieved acpi_desc, then we know nfit_mem data
1730          * is still valid.
1731          */
1732         nfit_mem = dev_get_drvdata(dev);
1733         if (nfit_mem && nfit_mem->flags_attr)
1734                 sysfs_notify_dirent(nfit_mem->flags_attr);
1735 }
1736 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1737
1738 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1739 {
1740         struct acpi_device *adev = data;
1741         struct device *dev = &adev->dev;
1742
1743         device_lock(dev->parent);
1744         __acpi_nvdimm_notify(dev, event);
1745         device_unlock(dev->parent);
1746 }
1747
1748 static bool acpi_nvdimm_has_method(struct acpi_device *adev, char *method)
1749 {
1750         acpi_handle handle;
1751         acpi_status status;
1752
1753         status = acpi_get_handle(adev->handle, method, &handle);
1754
1755         if (ACPI_SUCCESS(status))
1756                 return true;
1757         return false;
1758 }
1759
1760 __weak void nfit_intel_shutdown_status(struct nfit_mem *nfit_mem)
1761 {
1762         struct nd_intel_smart smart = { 0 };
1763         union acpi_object in_buf = {
1764                 .type = ACPI_TYPE_BUFFER,
1765                 .buffer.pointer = (char *) &smart,
1766                 .buffer.length = sizeof(smart),
1767         };
1768         union acpi_object in_obj = {
1769                 .type = ACPI_TYPE_PACKAGE,
1770                 .package.count = 1,
1771                 .package.elements = &in_buf,
1772         };
1773         const u8 func = ND_INTEL_SMART;
1774         const guid_t *guid = to_nfit_uuid(nfit_mem->family);
1775         u8 revid = nfit_dsm_revid(nfit_mem->family, func);
1776         struct acpi_device *adev = nfit_mem->adev;
1777         acpi_handle handle = adev->handle;
1778         union acpi_object *out_obj;
1779
1780         if ((nfit_mem->dsm_mask & (1 << func)) == 0)
1781                 return;
1782
1783         out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
1784         if (!out_obj)
1785                 return;
1786
1787         if (smart.flags & ND_INTEL_SMART_SHUTDOWN_VALID) {
1788                 if (smart.shutdown_state)
1789                         set_bit(NFIT_MEM_DIRTY, &nfit_mem->flags);
1790         }
1791
1792         if (smart.flags & ND_INTEL_SMART_SHUTDOWN_COUNT_VALID) {
1793                 set_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags);
1794                 nfit_mem->dirty_shutdown = smart.shutdown_count;
1795         }
1796         ACPI_FREE(out_obj);
1797 }
1798
1799 static void populate_shutdown_status(struct nfit_mem *nfit_mem)
1800 {
1801         /*
1802          * For DIMMs that provide a dynamic facility to retrieve a
1803          * dirty-shutdown status and/or a dirty-shutdown count, cache
1804          * these values in nfit_mem.
1805          */
1806         if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
1807                 nfit_intel_shutdown_status(nfit_mem);
1808 }
1809
1810 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1811                 struct nfit_mem *nfit_mem, u32 device_handle)
1812 {
1813         struct acpi_device *adev, *adev_dimm;
1814         struct device *dev = acpi_desc->dev;
1815         unsigned long dsm_mask, label_mask;
1816         const guid_t *guid;
1817         int i;
1818         int family = -1;
1819         struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
1820
1821         /* nfit test assumes 1:1 relationship between commands and dsms */
1822         nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1823         nfit_mem->family = NVDIMM_FAMILY_INTEL;
1824
1825         if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1826                 sprintf(nfit_mem->id, "%04x-%02x-%04x-%08x",
1827                                 be16_to_cpu(dcr->vendor_id),
1828                                 dcr->manufacturing_location,
1829                                 be16_to_cpu(dcr->manufacturing_date),
1830                                 be32_to_cpu(dcr->serial_number));
1831         else
1832                 sprintf(nfit_mem->id, "%04x-%08x",
1833                                 be16_to_cpu(dcr->vendor_id),
1834                                 be32_to_cpu(dcr->serial_number));
1835
1836         adev = to_acpi_dev(acpi_desc);
1837         if (!adev) {
1838                 /* unit test case */
1839                 populate_shutdown_status(nfit_mem);
1840                 return 0;
1841         }
1842
1843         adev_dimm = acpi_find_child_device(adev, device_handle, false);
1844         nfit_mem->adev = adev_dimm;
1845         if (!adev_dimm) {
1846                 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1847                                 device_handle);
1848                 return force_enable_dimms ? 0 : -ENODEV;
1849         }
1850
1851         if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1852                 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1853                 dev_err(dev, "%s: notification registration failed\n",
1854                                 dev_name(&adev_dimm->dev));
1855                 return -ENXIO;
1856         }
1857         /*
1858          * Record nfit_mem for the notification path to track back to
1859          * the nfit sysfs attributes for this dimm device object.
1860          */
1861         dev_set_drvdata(&adev_dimm->dev, nfit_mem);
1862
1863         /*
1864          * Until standardization materializes we need to consider 4
1865          * different command sets.  Note, that checking for function0 (bit0)
1866          * tells us if any commands are reachable through this GUID.
1867          */
1868         for (i = 0; i <= NVDIMM_FAMILY_MAX; i++)
1869                 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1870                         if (family < 0 || i == default_dsm_family)
1871                                 family = i;
1872
1873         /* limit the supported commands to those that are publicly documented */
1874         nfit_mem->family = family;
1875         if (override_dsm_mask && !disable_vendor_specific)
1876                 dsm_mask = override_dsm_mask;
1877         else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1878                 dsm_mask = NVDIMM_INTEL_CMDMASK;
1879                 if (disable_vendor_specific)
1880                         dsm_mask &= ~(1 << ND_CMD_VENDOR);
1881         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1882                 dsm_mask = 0x1c3c76;
1883         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1884                 dsm_mask = 0x1fe;
1885                 if (disable_vendor_specific)
1886                         dsm_mask &= ~(1 << 8);
1887         } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1888                 dsm_mask = 0xffffffff;
1889         } else {
1890                 dev_dbg(dev, "unknown dimm command family\n");
1891                 nfit_mem->family = -1;
1892                 /* DSMs are optional, continue loading the driver... */
1893                 return 0;
1894         }
1895
1896         /*
1897          * Function 0 is the command interrogation function, don't
1898          * export it to potential userspace use, and enable it to be
1899          * used as an error value in acpi_nfit_ctl().
1900          */
1901         dsm_mask &= ~1UL;
1902
1903         guid = to_nfit_uuid(nfit_mem->family);
1904         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1905                 if (acpi_check_dsm(adev_dimm->handle, guid,
1906                                         nfit_dsm_revid(nfit_mem->family, i),
1907                                         1ULL << i))
1908                         set_bit(i, &nfit_mem->dsm_mask);
1909
1910         /*
1911          * Prefer the NVDIMM_FAMILY_INTEL label read commands if present
1912          * due to their better semantics handling locked capacity.
1913          */
1914         label_mask = 1 << ND_CMD_GET_CONFIG_SIZE | 1 << ND_CMD_GET_CONFIG_DATA
1915                 | 1 << ND_CMD_SET_CONFIG_DATA;
1916         if (family == NVDIMM_FAMILY_INTEL
1917                         && (dsm_mask & label_mask) == label_mask)
1918                 return 0;
1919
1920         if (acpi_nvdimm_has_method(adev_dimm, "_LSI")
1921                         && acpi_nvdimm_has_method(adev_dimm, "_LSR")) {
1922                 dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev));
1923                 set_bit(NFIT_MEM_LSR, &nfit_mem->flags);
1924         }
1925
1926         if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
1927                         && acpi_nvdimm_has_method(adev_dimm, "_LSW")) {
1928                 dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev));
1929                 set_bit(NFIT_MEM_LSW, &nfit_mem->flags);
1930         }
1931
1932         populate_shutdown_status(nfit_mem);
1933
1934         return 0;
1935 }
1936
1937 static void shutdown_dimm_notify(void *data)
1938 {
1939         struct acpi_nfit_desc *acpi_desc = data;
1940         struct nfit_mem *nfit_mem;
1941
1942         mutex_lock(&acpi_desc->init_mutex);
1943         /*
1944          * Clear out the nfit_mem->flags_attr and shut down dimm event
1945          * notifications.
1946          */
1947         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1948                 struct acpi_device *adev_dimm = nfit_mem->adev;
1949
1950                 if (nfit_mem->flags_attr) {
1951                         sysfs_put(nfit_mem->flags_attr);
1952                         nfit_mem->flags_attr = NULL;
1953                 }
1954                 if (adev_dimm) {
1955                         acpi_remove_notify_handler(adev_dimm->handle,
1956                                         ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
1957                         dev_set_drvdata(&adev_dimm->dev, NULL);
1958                 }
1959         }
1960         mutex_unlock(&acpi_desc->init_mutex);
1961 }
1962
1963 static const struct nvdimm_security_ops *acpi_nfit_get_security_ops(int family)
1964 {
1965         switch (family) {
1966         case NVDIMM_FAMILY_INTEL:
1967                 return intel_security_ops;
1968         default:
1969                 return NULL;
1970         }
1971 }
1972
1973 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1974 {
1975         struct nfit_mem *nfit_mem;
1976         int dimm_count = 0, rc;
1977         struct nvdimm *nvdimm;
1978
1979         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1980                 struct acpi_nfit_flush_address *flush;
1981                 unsigned long flags = 0, cmd_mask;
1982                 struct nfit_memdev *nfit_memdev;
1983                 u32 device_handle;
1984                 u16 mem_flags;
1985
1986                 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1987                 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1988                 if (nvdimm) {
1989                         dimm_count++;
1990                         continue;
1991                 }
1992
1993                 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1994                         set_bit(NDD_ALIASING, &flags);
1995
1996                 /* collate flags across all memdevs for this dimm */
1997                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1998                         struct acpi_nfit_memory_map *dimm_memdev;
1999
2000                         dimm_memdev = __to_nfit_memdev(nfit_mem);
2001                         if (dimm_memdev->device_handle
2002                                         != nfit_memdev->memdev->device_handle)
2003                                 continue;
2004                         dimm_memdev->flags |= nfit_memdev->memdev->flags;
2005                 }
2006
2007                 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
2008                 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
2009                         set_bit(NDD_UNARMED, &flags);
2010
2011                 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
2012                 if (rc)
2013                         continue;
2014
2015                 /*
2016                  * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
2017                  * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
2018                  * userspace interface.
2019                  */
2020                 cmd_mask = 1UL << ND_CMD_CALL;
2021                 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
2022                         /*
2023                          * These commands have a 1:1 correspondence
2024                          * between DSM payload and libnvdimm ioctl
2025                          * payload format.
2026                          */
2027                         cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
2028                 }
2029
2030                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
2031                         set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
2032                         set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
2033                 }
2034                 if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags))
2035                         set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
2036
2037                 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
2038                         : NULL;
2039                 nvdimm = __nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
2040                                 acpi_nfit_dimm_attribute_groups,
2041                                 flags, cmd_mask, flush ? flush->hint_count : 0,
2042                                 nfit_mem->flush_wpq, &nfit_mem->id[0],
2043                                 acpi_nfit_get_security_ops(nfit_mem->family));
2044                 if (!nvdimm)
2045                         return -ENOMEM;
2046
2047                 nfit_mem->nvdimm = nvdimm;
2048                 dimm_count++;
2049
2050                 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
2051                         continue;
2052
2053                 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s%s\n",
2054                                 nvdimm_name(nvdimm),
2055                   mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
2056                   mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
2057                   mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
2058                   mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
2059                   mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
2060
2061         }
2062
2063         rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
2064         if (rc)
2065                 return rc;
2066
2067         /*
2068          * Now that dimms are successfully registered, and async registration
2069          * is flushed, attempt to enable event notification.
2070          */
2071         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2072                 struct kernfs_node *nfit_kernfs;
2073
2074                 nvdimm = nfit_mem->nvdimm;
2075                 if (!nvdimm)
2076                         continue;
2077
2078                 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
2079                 if (nfit_kernfs)
2080                         nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
2081                                         "flags");
2082                 sysfs_put(nfit_kernfs);
2083                 if (!nfit_mem->flags_attr)
2084                         dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
2085                                         nvdimm_name(nvdimm));
2086         }
2087
2088         return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
2089                         acpi_desc);
2090 }
2091
2092 /*
2093  * These constants are private because there are no kernel consumers of
2094  * these commands.
2095  */
2096 enum nfit_aux_cmds {
2097         NFIT_CMD_TRANSLATE_SPA = 5,
2098         NFIT_CMD_ARS_INJECT_SET = 7,
2099         NFIT_CMD_ARS_INJECT_CLEAR = 8,
2100         NFIT_CMD_ARS_INJECT_GET = 9,
2101 };
2102
2103 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
2104 {
2105         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2106         const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
2107         struct acpi_device *adev;
2108         unsigned long dsm_mask;
2109         int i;
2110
2111         nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
2112         nd_desc->bus_dsm_mask = acpi_desc->bus_nfit_cmd_force_en;
2113         adev = to_acpi_dev(acpi_desc);
2114         if (!adev)
2115                 return;
2116
2117         for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
2118                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2119                         set_bit(i, &nd_desc->cmd_mask);
2120         set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
2121
2122         dsm_mask =
2123                 (1 << ND_CMD_ARS_CAP) |
2124                 (1 << ND_CMD_ARS_START) |
2125                 (1 << ND_CMD_ARS_STATUS) |
2126                 (1 << ND_CMD_CLEAR_ERROR) |
2127                 (1 << NFIT_CMD_TRANSLATE_SPA) |
2128                 (1 << NFIT_CMD_ARS_INJECT_SET) |
2129                 (1 << NFIT_CMD_ARS_INJECT_CLEAR) |
2130                 (1 << NFIT_CMD_ARS_INJECT_GET);
2131         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
2132                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2133                         set_bit(i, &nd_desc->bus_dsm_mask);
2134 }
2135
2136 static ssize_t range_index_show(struct device *dev,
2137                 struct device_attribute *attr, char *buf)
2138 {
2139         struct nd_region *nd_region = to_nd_region(dev);
2140         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
2141
2142         return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
2143 }
2144 static DEVICE_ATTR_RO(range_index);
2145
2146 static struct attribute *acpi_nfit_region_attributes[] = {
2147         &dev_attr_range_index.attr,
2148         NULL,
2149 };
2150
2151 static const struct attribute_group acpi_nfit_region_attribute_group = {
2152         .name = "nfit",
2153         .attrs = acpi_nfit_region_attributes,
2154 };
2155
2156 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
2157         &nd_region_attribute_group,
2158         &nd_mapping_attribute_group,
2159         &nd_device_attribute_group,
2160         &nd_numa_attribute_group,
2161         &acpi_nfit_region_attribute_group,
2162         NULL,
2163 };
2164
2165 /* enough info to uniquely specify an interleave set */
2166 struct nfit_set_info {
2167         struct nfit_set_info_map {
2168                 u64 region_offset;
2169                 u32 serial_number;
2170                 u32 pad;
2171         } mapping[0];
2172 };
2173
2174 struct nfit_set_info2 {
2175         struct nfit_set_info_map2 {
2176                 u64 region_offset;
2177                 u32 serial_number;
2178                 u16 vendor_id;
2179                 u16 manufacturing_date;
2180                 u8  manufacturing_location;
2181                 u8  reserved[31];
2182         } mapping[0];
2183 };
2184
2185 static size_t sizeof_nfit_set_info(int num_mappings)
2186 {
2187         return sizeof(struct nfit_set_info)
2188                 + num_mappings * sizeof(struct nfit_set_info_map);
2189 }
2190
2191 static size_t sizeof_nfit_set_info2(int num_mappings)
2192 {
2193         return sizeof(struct nfit_set_info2)
2194                 + num_mappings * sizeof(struct nfit_set_info_map2);
2195 }
2196
2197 static int cmp_map_compat(const void *m0, const void *m1)
2198 {
2199         const struct nfit_set_info_map *map0 = m0;
2200         const struct nfit_set_info_map *map1 = m1;
2201
2202         return memcmp(&map0->region_offset, &map1->region_offset,
2203                         sizeof(u64));
2204 }
2205
2206 static int cmp_map(const void *m0, const void *m1)
2207 {
2208         const struct nfit_set_info_map *map0 = m0;
2209         const struct nfit_set_info_map *map1 = m1;
2210
2211         if (map0->region_offset < map1->region_offset)
2212                 return -1;
2213         else if (map0->region_offset > map1->region_offset)
2214                 return 1;
2215         return 0;
2216 }
2217
2218 static int cmp_map2(const void *m0, const void *m1)
2219 {
2220         const struct nfit_set_info_map2 *map0 = m0;
2221         const struct nfit_set_info_map2 *map1 = m1;
2222
2223         if (map0->region_offset < map1->region_offset)
2224                 return -1;
2225         else if (map0->region_offset > map1->region_offset)
2226                 return 1;
2227         return 0;
2228 }
2229
2230 /* Retrieve the nth entry referencing this spa */
2231 static struct acpi_nfit_memory_map *memdev_from_spa(
2232                 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
2233 {
2234         struct nfit_memdev *nfit_memdev;
2235
2236         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
2237                 if (nfit_memdev->memdev->range_index == range_index)
2238                         if (n-- == 0)
2239                                 return nfit_memdev->memdev;
2240         return NULL;
2241 }
2242
2243 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
2244                 struct nd_region_desc *ndr_desc,
2245                 struct acpi_nfit_system_address *spa)
2246 {
2247         struct device *dev = acpi_desc->dev;
2248         struct nd_interleave_set *nd_set;
2249         u16 nr = ndr_desc->num_mappings;
2250         struct nfit_set_info2 *info2;
2251         struct nfit_set_info *info;
2252         int i;
2253
2254         nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
2255         if (!nd_set)
2256                 return -ENOMEM;
2257         guid_copy(&nd_set->type_guid, (guid_t *) spa->range_guid);
2258
2259         info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
2260         if (!info)
2261                 return -ENOMEM;
2262
2263         info2 = devm_kzalloc(dev, sizeof_nfit_set_info2(nr), GFP_KERNEL);
2264         if (!info2)
2265                 return -ENOMEM;
2266
2267         for (i = 0; i < nr; i++) {
2268                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
2269                 struct nfit_set_info_map *map = &info->mapping[i];
2270                 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2271                 struct nvdimm *nvdimm = mapping->nvdimm;
2272                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2273                 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
2274                                 spa->range_index, i);
2275                 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2276
2277                 if (!memdev || !nfit_mem->dcr) {
2278                         dev_err(dev, "%s: failed to find DCR\n", __func__);
2279                         return -ENODEV;
2280                 }
2281
2282                 map->region_offset = memdev->region_offset;
2283                 map->serial_number = dcr->serial_number;
2284
2285                 map2->region_offset = memdev->region_offset;
2286                 map2->serial_number = dcr->serial_number;
2287                 map2->vendor_id = dcr->vendor_id;
2288                 map2->manufacturing_date = dcr->manufacturing_date;
2289                 map2->manufacturing_location = dcr->manufacturing_location;
2290         }
2291
2292         /* v1.1 namespaces */
2293         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2294                         cmp_map, NULL);
2295         nd_set->cookie1 = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2296
2297         /* v1.2 namespaces */
2298         sort(&info2->mapping[0], nr, sizeof(struct nfit_set_info_map2),
2299                         cmp_map2, NULL);
2300         nd_set->cookie2 = nd_fletcher64(info2, sizeof_nfit_set_info2(nr), 0);
2301
2302         /* support v1.1 namespaces created with the wrong sort order */
2303         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2304                         cmp_map_compat, NULL);
2305         nd_set->altcookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2306
2307         /* record the result of the sort for the mapping position */
2308         for (i = 0; i < nr; i++) {
2309                 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2310                 int j;
2311
2312                 for (j = 0; j < nr; j++) {
2313                         struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
2314                         struct nvdimm *nvdimm = mapping->nvdimm;
2315                         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2316                         struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2317
2318                         if (map2->serial_number == dcr->serial_number &&
2319                             map2->vendor_id == dcr->vendor_id &&
2320                             map2->manufacturing_date == dcr->manufacturing_date &&
2321                             map2->manufacturing_location
2322                                     == dcr->manufacturing_location) {
2323                                 mapping->position = i;
2324                                 break;
2325                         }
2326                 }
2327         }
2328
2329         ndr_desc->nd_set = nd_set;
2330         devm_kfree(dev, info);
2331         devm_kfree(dev, info2);
2332
2333         return 0;
2334 }
2335
2336 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
2337 {
2338         struct acpi_nfit_interleave *idt = mmio->idt;
2339         u32 sub_line_offset, line_index, line_offset;
2340         u64 line_no, table_skip_count, table_offset;
2341
2342         line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
2343         table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
2344         line_offset = idt->line_offset[line_index]
2345                 * mmio->line_size;
2346         table_offset = table_skip_count * mmio->table_size;
2347
2348         return mmio->base_offset + line_offset + table_offset + sub_line_offset;
2349 }
2350
2351 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
2352 {
2353         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2354         u64 offset = nfit_blk->stat_offset + mmio->size * bw;
2355         const u32 STATUS_MASK = 0x80000037;
2356
2357         if (mmio->num_lines)
2358                 offset = to_interleave_offset(offset, mmio);
2359
2360         return readl(mmio->addr.base + offset) & STATUS_MASK;
2361 }
2362
2363 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
2364                 resource_size_t dpa, unsigned int len, unsigned int write)
2365 {
2366         u64 cmd, offset;
2367         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2368
2369         enum {
2370                 BCW_OFFSET_MASK = (1ULL << 48)-1,
2371                 BCW_LEN_SHIFT = 48,
2372                 BCW_LEN_MASK = (1ULL << 8) - 1,
2373                 BCW_CMD_SHIFT = 56,
2374         };
2375
2376         cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
2377         len = len >> L1_CACHE_SHIFT;
2378         cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
2379         cmd |= ((u64) write) << BCW_CMD_SHIFT;
2380
2381         offset = nfit_blk->cmd_offset + mmio->size * bw;
2382         if (mmio->num_lines)
2383                 offset = to_interleave_offset(offset, mmio);
2384
2385         writeq(cmd, mmio->addr.base + offset);
2386         nvdimm_flush(nfit_blk->nd_region);
2387
2388         if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
2389                 readq(mmio->addr.base + offset);
2390 }
2391
2392 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
2393                 resource_size_t dpa, void *iobuf, size_t len, int rw,
2394                 unsigned int lane)
2395 {
2396         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2397         unsigned int copied = 0;
2398         u64 base_offset;
2399         int rc;
2400
2401         base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
2402                 + lane * mmio->size;
2403         write_blk_ctl(nfit_blk, lane, dpa, len, rw);
2404         while (len) {
2405                 unsigned int c;
2406                 u64 offset;
2407
2408                 if (mmio->num_lines) {
2409                         u32 line_offset;
2410
2411                         offset = to_interleave_offset(base_offset + copied,
2412                                         mmio);
2413                         div_u64_rem(offset, mmio->line_size, &line_offset);
2414                         c = min_t(size_t, len, mmio->line_size - line_offset);
2415                 } else {
2416                         offset = base_offset + nfit_blk->bdw_offset;
2417                         c = len;
2418                 }
2419
2420                 if (rw)
2421                         memcpy_flushcache(mmio->addr.aperture + offset, iobuf + copied, c);
2422                 else {
2423                         if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
2424                                 arch_invalidate_pmem((void __force *)
2425                                         mmio->addr.aperture + offset, c);
2426
2427                         memcpy(iobuf + copied, mmio->addr.aperture + offset, c);
2428                 }
2429
2430                 copied += c;
2431                 len -= c;
2432         }
2433
2434         if (rw)
2435                 nvdimm_flush(nfit_blk->nd_region);
2436
2437         rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
2438         return rc;
2439 }
2440
2441 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
2442                 resource_size_t dpa, void *iobuf, u64 len, int rw)
2443 {
2444         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
2445         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2446         struct nd_region *nd_region = nfit_blk->nd_region;
2447         unsigned int lane, copied = 0;
2448         int rc = 0;
2449
2450         lane = nd_region_acquire_lane(nd_region);
2451         while (len) {
2452                 u64 c = min(len, mmio->size);
2453
2454                 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
2455                                 iobuf + copied, c, rw, lane);
2456                 if (rc)
2457                         break;
2458
2459                 copied += c;
2460                 len -= c;
2461         }
2462         nd_region_release_lane(nd_region, lane);
2463
2464         return rc;
2465 }
2466
2467 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
2468                 struct acpi_nfit_interleave *idt, u16 interleave_ways)
2469 {
2470         if (idt) {
2471                 mmio->num_lines = idt->line_count;
2472                 mmio->line_size = idt->line_size;
2473                 if (interleave_ways == 0)
2474                         return -ENXIO;
2475                 mmio->table_size = mmio->num_lines * interleave_ways
2476                         * mmio->line_size;
2477         }
2478
2479         return 0;
2480 }
2481
2482 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
2483                 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
2484 {
2485         struct nd_cmd_dimm_flags flags;
2486         int rc;
2487
2488         memset(&flags, 0, sizeof(flags));
2489         rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
2490                         sizeof(flags), NULL);
2491
2492         if (rc >= 0 && flags.status == 0)
2493                 nfit_blk->dimm_flags = flags.flags;
2494         else if (rc == -ENOTTY) {
2495                 /* fall back to a conservative default */
2496                 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
2497                 rc = 0;
2498         } else
2499                 rc = -ENXIO;
2500
2501         return rc;
2502 }
2503
2504 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
2505                 struct device *dev)
2506 {
2507         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
2508         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
2509         struct nfit_blk_mmio *mmio;
2510         struct nfit_blk *nfit_blk;
2511         struct nfit_mem *nfit_mem;
2512         struct nvdimm *nvdimm;
2513         int rc;
2514
2515         nvdimm = nd_blk_region_to_dimm(ndbr);
2516         nfit_mem = nvdimm_provider_data(nvdimm);
2517         if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
2518                 dev_dbg(dev, "missing%s%s%s\n",
2519                                 nfit_mem ? "" : " nfit_mem",
2520                                 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
2521                                 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
2522                 return -ENXIO;
2523         }
2524
2525         nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
2526         if (!nfit_blk)
2527                 return -ENOMEM;
2528         nd_blk_region_set_provider_data(ndbr, nfit_blk);
2529         nfit_blk->nd_region = to_nd_region(dev);
2530
2531         /* map block aperture memory */
2532         nfit_blk->bdw_offset = nfit_mem->bdw->offset;
2533         mmio = &nfit_blk->mmio[BDW];
2534         mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
2535                         nfit_mem->spa_bdw->length, nd_blk_memremap_flags(ndbr));
2536         if (!mmio->addr.base) {
2537                 dev_dbg(dev, "%s failed to map bdw\n",
2538                                 nvdimm_name(nvdimm));
2539                 return -ENOMEM;
2540         }
2541         mmio->size = nfit_mem->bdw->size;
2542         mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
2543         mmio->idt = nfit_mem->idt_bdw;
2544         mmio->spa = nfit_mem->spa_bdw;
2545         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
2546                         nfit_mem->memdev_bdw->interleave_ways);
2547         if (rc) {
2548                 dev_dbg(dev, "%s failed to init bdw interleave\n",
2549                                 nvdimm_name(nvdimm));
2550                 return rc;
2551         }
2552
2553         /* map block control memory */
2554         nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
2555         nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
2556         mmio = &nfit_blk->mmio[DCR];
2557         mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
2558                         nfit_mem->spa_dcr->length);
2559         if (!mmio->addr.base) {
2560                 dev_dbg(dev, "%s failed to map dcr\n",
2561                                 nvdimm_name(nvdimm));
2562                 return -ENOMEM;
2563         }
2564         mmio->size = nfit_mem->dcr->window_size;
2565         mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
2566         mmio->idt = nfit_mem->idt_dcr;
2567         mmio->spa = nfit_mem->spa_dcr;
2568         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
2569                         nfit_mem->memdev_dcr->interleave_ways);
2570         if (rc) {
2571                 dev_dbg(dev, "%s failed to init dcr interleave\n",
2572                                 nvdimm_name(nvdimm));
2573                 return rc;
2574         }
2575
2576         rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
2577         if (rc < 0) {
2578                 dev_dbg(dev, "%s failed get DIMM flags\n",
2579                                 nvdimm_name(nvdimm));
2580                 return rc;
2581         }
2582
2583         if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
2584                 dev_warn(dev, "unable to guarantee persistence of writes\n");
2585
2586         if (mmio->line_size == 0)
2587                 return 0;
2588
2589         if ((u32) nfit_blk->cmd_offset % mmio->line_size
2590                         + 8 > mmio->line_size) {
2591                 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
2592                 return -ENXIO;
2593         } else if ((u32) nfit_blk->stat_offset % mmio->line_size
2594                         + 8 > mmio->line_size) {
2595                 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
2596                 return -ENXIO;
2597         }
2598
2599         return 0;
2600 }
2601
2602 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2603                 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2604 {
2605         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2606         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2607         int cmd_rc, rc;
2608
2609         cmd->address = spa->address;
2610         cmd->length = spa->length;
2611         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2612                         sizeof(*cmd), &cmd_rc);
2613         if (rc < 0)
2614                 return rc;
2615         return cmd_rc;
2616 }
2617
2618 static int ars_start(struct acpi_nfit_desc *acpi_desc,
2619                 struct nfit_spa *nfit_spa, enum nfit_ars_state req_type)
2620 {
2621         int rc;
2622         int cmd_rc;
2623         struct nd_cmd_ars_start ars_start;
2624         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2625         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2626
2627         memset(&ars_start, 0, sizeof(ars_start));
2628         ars_start.address = spa->address;
2629         ars_start.length = spa->length;
2630         if (req_type == ARS_REQ_SHORT)
2631                 ars_start.flags = ND_ARS_RETURN_PREV_DATA;
2632         if (nfit_spa_type(spa) == NFIT_SPA_PM)
2633                 ars_start.type = ND_ARS_PERSISTENT;
2634         else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2635                 ars_start.type = ND_ARS_VOLATILE;
2636         else
2637                 return -ENOTTY;
2638
2639         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2640                         sizeof(ars_start), &cmd_rc);
2641
2642         if (rc < 0)
2643                 return rc;
2644         return cmd_rc;
2645 }
2646
2647 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2648 {
2649         int rc, cmd_rc;
2650         struct nd_cmd_ars_start ars_start;
2651         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2652         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2653
2654         memset(&ars_start, 0, sizeof(ars_start));
2655         ars_start.address = ars_status->restart_address;
2656         ars_start.length = ars_status->restart_length;
2657         ars_start.type = ars_status->type;
2658         ars_start.flags = acpi_desc->ars_start_flags;
2659         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2660                         sizeof(ars_start), &cmd_rc);
2661         if (rc < 0)
2662                 return rc;
2663         return cmd_rc;
2664 }
2665
2666 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2667 {
2668         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2669         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2670         int rc, cmd_rc;
2671
2672         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2673                         acpi_desc->max_ars, &cmd_rc);
2674         if (rc < 0)
2675                 return rc;
2676         return cmd_rc;
2677 }
2678
2679 static void ars_complete(struct acpi_nfit_desc *acpi_desc,
2680                 struct nfit_spa *nfit_spa)
2681 {
2682         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2683         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2684         struct nd_region *nd_region = nfit_spa->nd_region;
2685         struct device *dev;
2686
2687         lockdep_assert_held(&acpi_desc->init_mutex);
2688         /*
2689          * Only advance the ARS state for ARS runs initiated by the
2690          * kernel, ignore ARS results from BIOS initiated runs for scrub
2691          * completion tracking.
2692          */
2693         if (acpi_desc->scrub_spa != nfit_spa)
2694                 return;
2695
2696         if ((ars_status->address >= spa->address && ars_status->address
2697                                 < spa->address + spa->length)
2698                         || (ars_status->address < spa->address)) {
2699                 /*
2700                  * Assume that if a scrub starts at an offset from the
2701                  * start of nfit_spa that we are in the continuation
2702                  * case.
2703                  *
2704                  * Otherwise, if the scrub covers the spa range, mark
2705                  * any pending request complete.
2706                  */
2707                 if (ars_status->address + ars_status->length
2708                                 >= spa->address + spa->length)
2709                                 /* complete */;
2710                 else
2711                         return;
2712         } else
2713                 return;
2714
2715         acpi_desc->scrub_spa = NULL;
2716         if (nd_region) {
2717                 dev = nd_region_dev(nd_region);
2718                 nvdimm_region_notify(nd_region, NVDIMM_REVALIDATE_POISON);
2719         } else
2720                 dev = acpi_desc->dev;
2721         dev_dbg(dev, "ARS: range %d complete\n", spa->range_index);
2722 }
2723
2724 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc)
2725 {
2726         struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2727         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2728         int rc;
2729         u32 i;
2730
2731         /*
2732          * First record starts at 44 byte offset from the start of the
2733          * payload.
2734          */
2735         if (ars_status->out_length < 44)
2736                 return 0;
2737         for (i = 0; i < ars_status->num_records; i++) {
2738                 /* only process full records */
2739                 if (ars_status->out_length
2740                                 < 44 + sizeof(struct nd_ars_record) * (i + 1))
2741                         break;
2742                 rc = nvdimm_bus_add_badrange(nvdimm_bus,
2743                                 ars_status->records[i].err_address,
2744                                 ars_status->records[i].length);
2745                 if (rc)
2746                         return rc;
2747         }
2748         if (i < ars_status->num_records)
2749                 dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2750
2751         return 0;
2752 }
2753
2754 static void acpi_nfit_remove_resource(void *data)
2755 {
2756         struct resource *res = data;
2757
2758         remove_resource(res);
2759 }
2760
2761 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2762                 struct nd_region_desc *ndr_desc)
2763 {
2764         struct resource *res, *nd_res = ndr_desc->res;
2765         int is_pmem, ret;
2766
2767         /* No operation if the region is already registered as PMEM */
2768         is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2769                                 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2770         if (is_pmem == REGION_INTERSECTS)
2771                 return 0;
2772
2773         res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2774         if (!res)
2775                 return -ENOMEM;
2776
2777         res->name = "Persistent Memory";
2778         res->start = nd_res->start;
2779         res->end = nd_res->end;
2780         res->flags = IORESOURCE_MEM;
2781         res->desc = IORES_DESC_PERSISTENT_MEMORY;
2782
2783         ret = insert_resource(&iomem_resource, res);
2784         if (ret)
2785                 return ret;
2786
2787         ret = devm_add_action_or_reset(acpi_desc->dev,
2788                                         acpi_nfit_remove_resource,
2789                                         res);
2790         if (ret)
2791                 return ret;
2792
2793         return 0;
2794 }
2795
2796 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2797                 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2798                 struct acpi_nfit_memory_map *memdev,
2799                 struct nfit_spa *nfit_spa)
2800 {
2801         struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2802                         memdev->device_handle);
2803         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2804         struct nd_blk_region_desc *ndbr_desc;
2805         struct nfit_mem *nfit_mem;
2806         int rc;
2807
2808         if (!nvdimm) {
2809                 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2810                                 spa->range_index, memdev->device_handle);
2811                 return -ENODEV;
2812         }
2813
2814         mapping->nvdimm = nvdimm;
2815         switch (nfit_spa_type(spa)) {
2816         case NFIT_SPA_PM:
2817         case NFIT_SPA_VOLATILE:
2818                 mapping->start = memdev->address;
2819                 mapping->size = memdev->region_size;
2820                 break;
2821         case NFIT_SPA_DCR:
2822                 nfit_mem = nvdimm_provider_data(nvdimm);
2823                 if (!nfit_mem || !nfit_mem->bdw) {
2824                         dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2825                                         spa->range_index, nvdimm_name(nvdimm));
2826                         break;
2827                 }
2828
2829                 mapping->size = nfit_mem->bdw->capacity;
2830                 mapping->start = nfit_mem->bdw->start_address;
2831                 ndr_desc->num_lanes = nfit_mem->bdw->windows;
2832                 ndr_desc->mapping = mapping;
2833                 ndr_desc->num_mappings = 1;
2834                 ndbr_desc = to_blk_region_desc(ndr_desc);
2835                 ndbr_desc->enable = acpi_nfit_blk_region_enable;
2836                 ndbr_desc->do_io = acpi_desc->blk_do_io;
2837                 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2838                 if (rc)
2839                         return rc;
2840                 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2841                                 ndr_desc);
2842                 if (!nfit_spa->nd_region)
2843                         return -ENOMEM;
2844                 break;
2845         }
2846
2847         return 0;
2848 }
2849
2850 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2851 {
2852         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2853                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2854                 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2855                 nfit_spa_type(spa) == NFIT_SPA_PCD);
2856 }
2857
2858 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
2859 {
2860         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2861                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2862                 nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
2863 }
2864
2865 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2866                 struct nfit_spa *nfit_spa)
2867 {
2868         static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2869         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2870         struct nd_blk_region_desc ndbr_desc;
2871         struct nd_region_desc *ndr_desc;
2872         struct nfit_memdev *nfit_memdev;
2873         struct nvdimm_bus *nvdimm_bus;
2874         struct resource res;
2875         int count = 0, rc;
2876
2877         if (nfit_spa->nd_region)
2878                 return 0;
2879
2880         if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2881                 dev_dbg(acpi_desc->dev, "detected invalid spa index\n");
2882                 return 0;
2883         }
2884
2885         memset(&res, 0, sizeof(res));
2886         memset(&mappings, 0, sizeof(mappings));
2887         memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2888         res.start = spa->address;
2889         res.end = res.start + spa->length - 1;
2890         ndr_desc = &ndbr_desc.ndr_desc;
2891         ndr_desc->res = &res;
2892         ndr_desc->provider_data = nfit_spa;
2893         ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2894         if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
2895                 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2896                                                 spa->proximity_domain);
2897         else
2898                 ndr_desc->numa_node = NUMA_NO_NODE;
2899
2900         /*
2901          * Persistence domain bits are hierarchical, if
2902          * ACPI_NFIT_CAPABILITY_CACHE_FLUSH is set then
2903          * ACPI_NFIT_CAPABILITY_MEM_FLUSH is implied.
2904          */
2905         if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_CACHE_FLUSH)
2906                 set_bit(ND_REGION_PERSIST_CACHE, &ndr_desc->flags);
2907         else if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_MEM_FLUSH)
2908                 set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc->flags);
2909
2910         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2911                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2912                 struct nd_mapping_desc *mapping;
2913
2914                 if (memdev->range_index != spa->range_index)
2915                         continue;
2916                 if (count >= ND_MAX_MAPPINGS) {
2917                         dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2918                                         spa->range_index, ND_MAX_MAPPINGS);
2919                         return -ENXIO;
2920                 }
2921                 mapping = &mappings[count++];
2922                 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
2923                                 memdev, nfit_spa);
2924                 if (rc)
2925                         goto out;
2926         }
2927
2928         ndr_desc->mapping = mappings;
2929         ndr_desc->num_mappings = count;
2930         rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2931         if (rc)
2932                 goto out;
2933
2934         nvdimm_bus = acpi_desc->nvdimm_bus;
2935         if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2936                 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2937                 if (rc) {
2938                         dev_warn(acpi_desc->dev,
2939                                 "failed to insert pmem resource to iomem: %d\n",
2940                                 rc);
2941                         goto out;
2942                 }
2943
2944                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2945                                 ndr_desc);
2946                 if (!nfit_spa->nd_region)
2947                         rc = -ENOMEM;
2948         } else if (nfit_spa_is_volatile(spa)) {
2949                 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
2950                                 ndr_desc);
2951                 if (!nfit_spa->nd_region)
2952                         rc = -ENOMEM;
2953         } else if (nfit_spa_is_virtual(spa)) {
2954                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2955                                 ndr_desc);
2956                 if (!nfit_spa->nd_region)
2957                         rc = -ENOMEM;
2958         }
2959
2960  out:
2961         if (rc)
2962                 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
2963                                 nfit_spa->spa->range_index);
2964         return rc;
2965 }
2966
2967 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc)
2968 {
2969         struct device *dev = acpi_desc->dev;
2970         struct nd_cmd_ars_status *ars_status;
2971
2972         if (acpi_desc->ars_status) {
2973                 memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
2974                 return 0;
2975         }
2976
2977         ars_status = devm_kzalloc(dev, acpi_desc->max_ars, GFP_KERNEL);
2978         if (!ars_status)
2979                 return -ENOMEM;
2980         acpi_desc->ars_status = ars_status;
2981         return 0;
2982 }
2983
2984 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc)
2985 {
2986         int rc;
2987
2988         if (ars_status_alloc(acpi_desc))
2989                 return -ENOMEM;
2990
2991         rc = ars_get_status(acpi_desc);
2992
2993         if (rc < 0 && rc != -ENOSPC)
2994                 return rc;
2995
2996         if (ars_status_process_records(acpi_desc))
2997                 dev_err(acpi_desc->dev, "Failed to process ARS records\n");
2998
2999         return rc;
3000 }
3001
3002 static int ars_register(struct acpi_nfit_desc *acpi_desc,
3003                 struct nfit_spa *nfit_spa)
3004 {
3005         int rc;
3006
3007         if (no_init_ars || test_bit(ARS_FAILED, &nfit_spa->ars_state))
3008                 return acpi_nfit_register_region(acpi_desc, nfit_spa);
3009
3010         set_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
3011         set_bit(ARS_REQ_LONG, &nfit_spa->ars_state);
3012
3013         switch (acpi_nfit_query_poison(acpi_desc)) {
3014         case 0:
3015         case -EAGAIN:
3016                 rc = ars_start(acpi_desc, nfit_spa, ARS_REQ_SHORT);
3017                 /* shouldn't happen, try again later */
3018                 if (rc == -EBUSY)
3019                         break;
3020                 if (rc) {
3021                         set_bit(ARS_FAILED, &nfit_spa->ars_state);
3022                         break;
3023                 }
3024                 clear_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
3025                 rc = acpi_nfit_query_poison(acpi_desc);
3026                 if (rc)
3027                         break;
3028                 acpi_desc->scrub_spa = nfit_spa;
3029                 ars_complete(acpi_desc, nfit_spa);
3030                 /*
3031                  * If ars_complete() says we didn't complete the
3032                  * short scrub, we'll try again with a long
3033                  * request.
3034                  */
3035                 acpi_desc->scrub_spa = NULL;
3036                 break;
3037         case -EBUSY:
3038         case -ENOMEM:
3039         case -ENOSPC:
3040                 /*
3041                  * BIOS was using ARS, wait for it to complete (or
3042                  * resources to become available) and then perform our
3043                  * own scrubs.
3044                  */
3045                 break;
3046         default:
3047                 set_bit(ARS_FAILED, &nfit_spa->ars_state);
3048                 break;
3049         }
3050
3051         return acpi_nfit_register_region(acpi_desc, nfit_spa);
3052 }
3053
3054 static void ars_complete_all(struct acpi_nfit_desc *acpi_desc)
3055 {
3056         struct nfit_spa *nfit_spa;
3057
3058         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3059                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3060                         continue;
3061                 ars_complete(acpi_desc, nfit_spa);
3062         }
3063 }
3064
3065 static unsigned int __acpi_nfit_scrub(struct acpi_nfit_desc *acpi_desc,
3066                 int query_rc)
3067 {
3068         unsigned int tmo = acpi_desc->scrub_tmo;
3069         struct device *dev = acpi_desc->dev;
3070         struct nfit_spa *nfit_spa;
3071
3072         lockdep_assert_held(&acpi_desc->init_mutex);
3073
3074         if (acpi_desc->cancel)
3075                 return 0;
3076
3077         if (query_rc == -EBUSY) {
3078                 dev_dbg(dev, "ARS: ARS busy\n");
3079                 return min(30U * 60U, tmo * 2);
3080         }
3081         if (query_rc == -ENOSPC) {
3082                 dev_dbg(dev, "ARS: ARS continue\n");
3083                 ars_continue(acpi_desc);
3084                 return 1;
3085         }
3086         if (query_rc && query_rc != -EAGAIN) {
3087                 unsigned long long addr, end;
3088
3089                 addr = acpi_desc->ars_status->address;
3090                 end = addr + acpi_desc->ars_status->length;
3091                 dev_dbg(dev, "ARS: %llx-%llx failed (%d)\n", addr, end,
3092                                 query_rc);
3093         }
3094
3095         ars_complete_all(acpi_desc);
3096         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3097                 enum nfit_ars_state req_type;
3098                 int rc;
3099
3100                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3101                         continue;
3102
3103                 /* prefer short ARS requests first */
3104                 if (test_bit(ARS_REQ_SHORT, &nfit_spa->ars_state))
3105                         req_type = ARS_REQ_SHORT;
3106                 else if (test_bit(ARS_REQ_LONG, &nfit_spa->ars_state))
3107                         req_type = ARS_REQ_LONG;
3108                 else
3109                         continue;
3110                 rc = ars_start(acpi_desc, nfit_spa, req_type);
3111
3112                 dev = nd_region_dev(nfit_spa->nd_region);
3113                 dev_dbg(dev, "ARS: range %d ARS start %s (%d)\n",
3114                                 nfit_spa->spa->range_index,
3115                                 req_type == ARS_REQ_SHORT ? "short" : "long",
3116                                 rc);
3117                 /*
3118                  * Hmm, we raced someone else starting ARS? Try again in
3119                  * a bit.
3120                  */
3121                 if (rc == -EBUSY)
3122                         return 1;
3123                 if (rc == 0) {
3124                         dev_WARN_ONCE(dev, acpi_desc->scrub_spa,
3125                                         "scrub start while range %d active\n",
3126                                         acpi_desc->scrub_spa->spa->range_index);
3127                         clear_bit(req_type, &nfit_spa->ars_state);
3128                         acpi_desc->scrub_spa = nfit_spa;
3129                         /*
3130                          * Consider this spa last for future scrub
3131                          * requests
3132                          */
3133                         list_move_tail(&nfit_spa->list, &acpi_desc->spas);
3134                         return 1;
3135                 }
3136
3137                 dev_err(dev, "ARS: range %d ARS failed (%d)\n",
3138                                 nfit_spa->spa->range_index, rc);
3139                 set_bit(ARS_FAILED, &nfit_spa->ars_state);
3140         }
3141         return 0;
3142 }
3143
3144 static void __sched_ars(struct acpi_nfit_desc *acpi_desc, unsigned int tmo)
3145 {
3146         lockdep_assert_held(&acpi_desc->init_mutex);
3147
3148         acpi_desc->scrub_busy = 1;
3149         /* note this should only be set from within the workqueue */
3150         if (tmo)
3151                 acpi_desc->scrub_tmo = tmo;
3152         queue_delayed_work(nfit_wq, &acpi_desc->dwork, tmo * HZ);
3153 }
3154
3155 static void sched_ars(struct acpi_nfit_desc *acpi_desc)
3156 {
3157         __sched_ars(acpi_desc, 0);
3158 }
3159
3160 static void notify_ars_done(struct acpi_nfit_desc *acpi_desc)
3161 {
3162         lockdep_assert_held(&acpi_desc->init_mutex);
3163
3164         acpi_desc->scrub_busy = 0;
3165         acpi_desc->scrub_count++;
3166         if (acpi_desc->scrub_count_state)
3167                 sysfs_notify_dirent(acpi_desc->scrub_count_state);
3168 }
3169
3170 static void acpi_nfit_scrub(struct work_struct *work)
3171 {
3172         struct acpi_nfit_desc *acpi_desc;
3173         unsigned int tmo;
3174         int query_rc;
3175
3176         acpi_desc = container_of(work, typeof(*acpi_desc), dwork.work);
3177         mutex_lock(&acpi_desc->init_mutex);
3178         query_rc = acpi_nfit_query_poison(acpi_desc);
3179         tmo = __acpi_nfit_scrub(acpi_desc, query_rc);
3180         if (tmo)
3181                 __sched_ars(acpi_desc, tmo);
3182         else
3183                 notify_ars_done(acpi_desc);
3184         memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
3185         mutex_unlock(&acpi_desc->init_mutex);
3186 }
3187
3188 static void acpi_nfit_init_ars(struct acpi_nfit_desc *acpi_desc,
3189                 struct nfit_spa *nfit_spa)
3190 {
3191         int type = nfit_spa_type(nfit_spa->spa);
3192         struct nd_cmd_ars_cap ars_cap;
3193         int rc;
3194
3195         set_bit(ARS_FAILED, &nfit_spa->ars_state);
3196         memset(&ars_cap, 0, sizeof(ars_cap));
3197         rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
3198         if (rc < 0)
3199                 return;
3200         /* check that the supported scrub types match the spa type */
3201         if (type == NFIT_SPA_VOLATILE && ((ars_cap.status >> 16)
3202                                 & ND_ARS_VOLATILE) == 0)
3203                 return;
3204         if (type == NFIT_SPA_PM && ((ars_cap.status >> 16)
3205                                 & ND_ARS_PERSISTENT) == 0)
3206                 return;
3207
3208         nfit_spa->max_ars = ars_cap.max_ars_out;
3209         nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
3210         acpi_desc->max_ars = max(nfit_spa->max_ars, acpi_desc->max_ars);
3211         clear_bit(ARS_FAILED, &nfit_spa->ars_state);
3212 }
3213
3214 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
3215 {
3216         struct nfit_spa *nfit_spa;
3217         int rc;
3218
3219         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3220                 switch (nfit_spa_type(nfit_spa->spa)) {
3221                 case NFIT_SPA_VOLATILE:
3222                 case NFIT_SPA_PM:
3223                         acpi_nfit_init_ars(acpi_desc, nfit_spa);
3224                         break;
3225                 }
3226         }
3227
3228         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
3229                 switch (nfit_spa_type(nfit_spa->spa)) {
3230                 case NFIT_SPA_VOLATILE:
3231                 case NFIT_SPA_PM:
3232                         /* register regions and kick off initial ARS run */
3233                         rc = ars_register(acpi_desc, nfit_spa);
3234                         if (rc)
3235                                 return rc;
3236                         break;
3237                 case NFIT_SPA_BDW:
3238                         /* nothing to register */
3239                         break;
3240                 case NFIT_SPA_DCR:
3241                 case NFIT_SPA_VDISK:
3242                 case NFIT_SPA_VCD:
3243                 case NFIT_SPA_PDISK:
3244                 case NFIT_SPA_PCD:
3245                         /* register known regions that don't support ARS */
3246                         rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
3247                         if (rc)
3248                                 return rc;
3249                         break;
3250                 default:
3251                         /* don't register unknown regions */
3252                         break;
3253                 }
3254
3255         sched_ars(acpi_desc);
3256         return 0;
3257 }
3258
3259 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
3260                 struct nfit_table_prev *prev)
3261 {
3262         struct device *dev = acpi_desc->dev;
3263
3264         if (!list_empty(&prev->spas) ||
3265                         !list_empty(&prev->memdevs) ||
3266                         !list_empty(&prev->dcrs) ||
3267                         !list_empty(&prev->bdws) ||
3268                         !list_empty(&prev->idts) ||
3269                         !list_empty(&prev->flushes)) {
3270                 dev_err(dev, "new nfit deletes entries (unsupported)\n");
3271                 return -ENXIO;
3272         }
3273         return 0;
3274 }
3275
3276 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
3277 {
3278         struct device *dev = acpi_desc->dev;
3279         struct kernfs_node *nfit;
3280         struct device *bus_dev;
3281
3282         if (!ars_supported(acpi_desc->nvdimm_bus))
3283                 return 0;
3284
3285         bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3286         nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
3287         if (!nfit) {
3288                 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
3289                 return -ENODEV;
3290         }
3291         acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
3292         sysfs_put(nfit);
3293         if (!acpi_desc->scrub_count_state) {
3294                 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
3295                 return -ENODEV;
3296         }
3297
3298         return 0;
3299 }
3300
3301 static void acpi_nfit_unregister(void *data)
3302 {
3303         struct acpi_nfit_desc *acpi_desc = data;
3304
3305         nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
3306 }
3307
3308 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
3309 {
3310         struct device *dev = acpi_desc->dev;
3311         struct nfit_table_prev prev;
3312         const void *end;
3313         int rc;
3314
3315         if (!acpi_desc->nvdimm_bus) {
3316                 acpi_nfit_init_dsms(acpi_desc);
3317
3318                 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
3319                                 &acpi_desc->nd_desc);
3320                 if (!acpi_desc->nvdimm_bus)
3321                         return -ENOMEM;
3322
3323                 rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
3324                                 acpi_desc);
3325                 if (rc)
3326                         return rc;
3327
3328                 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
3329                 if (rc)
3330                         return rc;
3331
3332                 /* register this acpi_desc for mce notifications */
3333                 mutex_lock(&acpi_desc_lock);
3334                 list_add_tail(&acpi_desc->list, &acpi_descs);
3335                 mutex_unlock(&acpi_desc_lock);
3336         }
3337
3338         mutex_lock(&acpi_desc->init_mutex);
3339
3340         INIT_LIST_HEAD(&prev.spas);
3341         INIT_LIST_HEAD(&prev.memdevs);
3342         INIT_LIST_HEAD(&prev.dcrs);
3343         INIT_LIST_HEAD(&prev.bdws);
3344         INIT_LIST_HEAD(&prev.idts);
3345         INIT_LIST_HEAD(&prev.flushes);
3346
3347         list_cut_position(&prev.spas, &acpi_desc->spas,
3348                                 acpi_desc->spas.prev);
3349         list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
3350                                 acpi_desc->memdevs.prev);
3351         list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
3352                                 acpi_desc->dcrs.prev);
3353         list_cut_position(&prev.bdws, &acpi_desc->bdws,
3354                                 acpi_desc->bdws.prev);
3355         list_cut_position(&prev.idts, &acpi_desc->idts,
3356                                 acpi_desc->idts.prev);
3357         list_cut_position(&prev.flushes, &acpi_desc->flushes,
3358                                 acpi_desc->flushes.prev);
3359
3360         end = data + sz;
3361         while (!IS_ERR_OR_NULL(data))
3362                 data = add_table(acpi_desc, &prev, data, end);
3363
3364         if (IS_ERR(data)) {
3365                 dev_dbg(dev, "nfit table parsing error: %ld\n", PTR_ERR(data));
3366                 rc = PTR_ERR(data);
3367                 goto out_unlock;
3368         }
3369
3370         rc = acpi_nfit_check_deletions(acpi_desc, &prev);
3371         if (rc)
3372                 goto out_unlock;
3373
3374         rc = nfit_mem_init(acpi_desc);
3375         if (rc)
3376                 goto out_unlock;
3377
3378         rc = acpi_nfit_register_dimms(acpi_desc);
3379         if (rc)
3380                 goto out_unlock;
3381
3382         rc = acpi_nfit_register_regions(acpi_desc);
3383
3384  out_unlock:
3385         mutex_unlock(&acpi_desc->init_mutex);
3386         return rc;
3387 }
3388 EXPORT_SYMBOL_GPL(acpi_nfit_init);
3389
3390 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
3391 {
3392         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
3393         struct device *dev = acpi_desc->dev;
3394
3395         /* Bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
3396         device_lock(dev);
3397         device_unlock(dev);
3398
3399         /* Bounce the init_mutex to complete initial registration */
3400         mutex_lock(&acpi_desc->init_mutex);
3401         mutex_unlock(&acpi_desc->init_mutex);
3402
3403         return 0;
3404 }
3405
3406 static int __acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3407                 struct nvdimm *nvdimm, unsigned int cmd)
3408 {
3409         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
3410
3411         if (nvdimm)
3412                 return 0;
3413         if (cmd != ND_CMD_ARS_START)
3414                 return 0;
3415
3416         /*
3417          * The kernel and userspace may race to initiate a scrub, but
3418          * the scrub thread is prepared to lose that initial race.  It
3419          * just needs guarantees that any ARS it initiates are not
3420          * interrupted by any intervening start requests from userspace.
3421          */
3422         if (work_busy(&acpi_desc->dwork.work))
3423                 return -EBUSY;
3424
3425         return 0;
3426 }
3427
3428 /* prevent security commands from being issued via ioctl */
3429 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3430                 struct nvdimm *nvdimm, unsigned int cmd, void *buf)
3431 {
3432         struct nd_cmd_pkg *call_pkg = buf;
3433         unsigned int func;
3434
3435         if (nvdimm && cmd == ND_CMD_CALL &&
3436                         call_pkg->nd_family == NVDIMM_FAMILY_INTEL) {
3437                 func = call_pkg->nd_command;
3438                 if ((1 << func) & NVDIMM_INTEL_SECURITY_CMDMASK)
3439                         return -EOPNOTSUPP;
3440         }
3441
3442         return __acpi_nfit_clear_to_send(nd_desc, nvdimm, cmd);
3443 }
3444
3445 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc,
3446                 enum nfit_ars_state req_type)
3447 {
3448         struct device *dev = acpi_desc->dev;
3449         int scheduled = 0, busy = 0;
3450         struct nfit_spa *nfit_spa;
3451
3452         mutex_lock(&acpi_desc->init_mutex);
3453         if (acpi_desc->cancel) {
3454                 mutex_unlock(&acpi_desc->init_mutex);
3455                 return 0;
3456         }
3457
3458         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3459                 int type = nfit_spa_type(nfit_spa->spa);
3460
3461                 if (type != NFIT_SPA_PM && type != NFIT_SPA_VOLATILE)
3462                         continue;
3463                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3464                         continue;
3465
3466                 if (test_and_set_bit(req_type, &nfit_spa->ars_state))
3467                         busy++;
3468                 else
3469                         scheduled++;
3470         }
3471         if (scheduled) {
3472                 sched_ars(acpi_desc);
3473                 dev_dbg(dev, "ars_scan triggered\n");
3474         }
3475         mutex_unlock(&acpi_desc->init_mutex);
3476
3477         if (scheduled)
3478                 return 0;
3479         if (busy)
3480                 return -EBUSY;
3481         return -ENOTTY;
3482 }
3483
3484 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
3485 {
3486         struct nvdimm_bus_descriptor *nd_desc;
3487
3488         dev_set_drvdata(dev, acpi_desc);
3489         acpi_desc->dev = dev;
3490         acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
3491         nd_desc = &acpi_desc->nd_desc;
3492         nd_desc->provider_name = "ACPI.NFIT";
3493         nd_desc->module = THIS_MODULE;
3494         nd_desc->ndctl = acpi_nfit_ctl;
3495         nd_desc->flush_probe = acpi_nfit_flush_probe;
3496         nd_desc->clear_to_send = acpi_nfit_clear_to_send;
3497         nd_desc->attr_groups = acpi_nfit_attribute_groups;
3498
3499         INIT_LIST_HEAD(&acpi_desc->spas);
3500         INIT_LIST_HEAD(&acpi_desc->dcrs);
3501         INIT_LIST_HEAD(&acpi_desc->bdws);
3502         INIT_LIST_HEAD(&acpi_desc->idts);
3503         INIT_LIST_HEAD(&acpi_desc->flushes);
3504         INIT_LIST_HEAD(&acpi_desc->memdevs);
3505         INIT_LIST_HEAD(&acpi_desc->dimms);
3506         INIT_LIST_HEAD(&acpi_desc->list);
3507         mutex_init(&acpi_desc->init_mutex);
3508         acpi_desc->scrub_tmo = 1;
3509         INIT_DELAYED_WORK(&acpi_desc->dwork, acpi_nfit_scrub);
3510 }
3511 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
3512
3513 static void acpi_nfit_put_table(void *table)
3514 {
3515         acpi_put_table(table);
3516 }
3517
3518 void acpi_nfit_shutdown(void *data)
3519 {
3520         struct acpi_nfit_desc *acpi_desc = data;
3521         struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3522
3523         /*
3524          * Destruct under acpi_desc_lock so that nfit_handle_mce does not
3525          * race teardown
3526          */
3527         mutex_lock(&acpi_desc_lock);
3528         list_del(&acpi_desc->list);
3529         mutex_unlock(&acpi_desc_lock);
3530
3531         mutex_lock(&acpi_desc->init_mutex);
3532         acpi_desc->cancel = 1;
3533         cancel_delayed_work_sync(&acpi_desc->dwork);
3534         mutex_unlock(&acpi_desc->init_mutex);
3535
3536         /*
3537          * Bounce the nvdimm bus lock to make sure any in-flight
3538          * acpi_nfit_ars_rescan() submissions have had a chance to
3539          * either submit or see ->cancel set.
3540          */
3541         device_lock(bus_dev);
3542         device_unlock(bus_dev);
3543
3544         flush_workqueue(nfit_wq);
3545 }
3546 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown);
3547
3548 static int acpi_nfit_add(struct acpi_device *adev)
3549 {
3550         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3551         struct acpi_nfit_desc *acpi_desc;
3552         struct device *dev = &adev->dev;
3553         struct acpi_table_header *tbl;
3554         acpi_status status = AE_OK;
3555         acpi_size sz;
3556         int rc = 0;
3557
3558         status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
3559         if (ACPI_FAILURE(status)) {
3560                 /* The NVDIMM root device allows OS to trigger enumeration of
3561                  * NVDIMMs through NFIT at boot time and re-enumeration at
3562                  * root level via the _FIT method during runtime.
3563                  * This is ok to return 0 here, we could have an nvdimm
3564                  * hotplugged later and evaluate _FIT method which returns
3565                  * data in the format of a series of NFIT Structures.
3566                  */
3567                 dev_dbg(dev, "failed to find NFIT at startup\n");
3568                 return 0;
3569         }
3570
3571         rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl);
3572         if (rc)
3573                 return rc;
3574         sz = tbl->length;
3575
3576         acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3577         if (!acpi_desc)
3578                 return -ENOMEM;
3579         acpi_nfit_desc_init(acpi_desc, &adev->dev);
3580
3581         /* Save the acpi header for exporting the revision via sysfs */
3582         acpi_desc->acpi_header = *tbl;
3583
3584         /* Evaluate _FIT and override with that if present */
3585         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
3586         if (ACPI_SUCCESS(status) && buf.length > 0) {
3587                 union acpi_object *obj = buf.pointer;
3588
3589                 if (obj->type == ACPI_TYPE_BUFFER)
3590                         rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3591                                         obj->buffer.length);
3592                 else
3593                         dev_dbg(dev, "invalid type %d, ignoring _FIT\n",
3594                                 (int) obj->type);
3595                 kfree(buf.pointer);
3596         } else
3597                 /* skip over the lead-in header table */
3598                 rc = acpi_nfit_init(acpi_desc, (void *) tbl
3599                                 + sizeof(struct acpi_table_nfit),
3600                                 sz - sizeof(struct acpi_table_nfit));
3601
3602         if (rc)
3603                 return rc;
3604         return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc);
3605 }
3606
3607 static int acpi_nfit_remove(struct acpi_device *adev)
3608 {
3609         /* see acpi_nfit_unregister */
3610         return 0;
3611 }
3612
3613 static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle)
3614 {
3615         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3616         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3617         union acpi_object *obj;
3618         acpi_status status;
3619         int ret;
3620
3621         if (!dev->driver) {
3622                 /* dev->driver may be null if we're being removed */
3623                 dev_dbg(dev, "no driver found for dev\n");
3624                 return;
3625         }
3626
3627         if (!acpi_desc) {
3628                 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3629                 if (!acpi_desc)
3630                         return;
3631                 acpi_nfit_desc_init(acpi_desc, dev);
3632         } else {
3633                 /*
3634                  * Finish previous registration before considering new
3635                  * regions.
3636                  */
3637                 flush_workqueue(nfit_wq);
3638         }
3639
3640         /* Evaluate _FIT */
3641         status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
3642         if (ACPI_FAILURE(status)) {
3643                 dev_err(dev, "failed to evaluate _FIT\n");
3644                 return;
3645         }
3646
3647         obj = buf.pointer;
3648         if (obj->type == ACPI_TYPE_BUFFER) {
3649                 ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3650                                 obj->buffer.length);
3651                 if (ret)
3652                         dev_err(dev, "failed to merge updated NFIT\n");
3653         } else
3654                 dev_err(dev, "Invalid _FIT\n");
3655         kfree(buf.pointer);
3656 }
3657
3658 static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle)
3659 {
3660         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3661
3662         if (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON)
3663                 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
3664         else
3665                 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_SHORT);
3666 }
3667
3668 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
3669 {
3670         dev_dbg(dev, "event: 0x%x\n", event);
3671
3672         switch (event) {
3673         case NFIT_NOTIFY_UPDATE:
3674                 return acpi_nfit_update_notify(dev, handle);
3675         case NFIT_NOTIFY_UC_MEMORY_ERROR:
3676                 return acpi_nfit_uc_error_notify(dev, handle);
3677         default:
3678                 return;
3679         }
3680 }
3681 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
3682
3683 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
3684 {
3685         device_lock(&adev->dev);
3686         __acpi_nfit_notify(&adev->dev, adev->handle, event);
3687         device_unlock(&adev->dev);
3688 }
3689
3690 static const struct acpi_device_id acpi_nfit_ids[] = {
3691         { "ACPI0012", 0 },
3692         { "", 0 },
3693 };
3694 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
3695
3696 static struct acpi_driver acpi_nfit_driver = {
3697         .name = KBUILD_MODNAME,
3698         .ids = acpi_nfit_ids,
3699         .ops = {
3700                 .add = acpi_nfit_add,
3701                 .remove = acpi_nfit_remove,
3702                 .notify = acpi_nfit_notify,
3703         },
3704 };
3705
3706 static __init int nfit_init(void)
3707 {
3708         int ret;
3709
3710         BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
3711         BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
3712         BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
3713         BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
3714         BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
3715         BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
3716         BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
3717         BUILD_BUG_ON(sizeof(struct acpi_nfit_capabilities) != 16);
3718
3719         guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]);
3720         guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]);
3721         guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]);
3722         guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]);
3723         guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]);
3724         guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]);
3725         guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]);
3726         guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]);
3727         guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]);
3728         guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]);
3729         guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
3730         guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
3731         guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
3732
3733         nfit_wq = create_singlethread_workqueue("nfit");
3734         if (!nfit_wq)
3735                 return -ENOMEM;
3736
3737         nfit_mce_register();
3738         ret = acpi_bus_register_driver(&acpi_nfit_driver);
3739         if (ret) {
3740                 nfit_mce_unregister();
3741                 destroy_workqueue(nfit_wq);
3742         }
3743
3744         return ret;
3745
3746 }
3747
3748 static __exit void nfit_exit(void)
3749 {
3750         nfit_mce_unregister();
3751         acpi_bus_unregister_driver(&acpi_nfit_driver);
3752         destroy_workqueue(nfit_wq);
3753         WARN_ON(!list_empty(&acpi_descs));
3754 }
3755
3756 module_init(nfit_init);
3757 module_exit(nfit_exit);
3758 MODULE_LICENSE("GPL v2");
3759 MODULE_AUTHOR("Intel Corporation");