drivers/acpi/nfit/core.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
   4  */
   5 #include <linux/list_sort.h>
   6 #include <linux/libnvdimm.h>
   7 #include <linux/module.h>
   8 #include <linux/mutex.h>
   9 #include <linux/ndctl.h>
  10 #include <linux/sysfs.h>
  11 #include <linux/delay.h>
  12 #include <linux/list.h>
  13 #include <linux/acpi.h>
  14 #include <linux/sort.h>
  15 #include <linux/io.h>
  16 #include <linux/nd.h>
  17 #include <asm/cacheflush.h>
  18 #include <acpi/nfit.h>
  19 #include "intel.h"
  20 #include "nfit.h"
  21
  22 /*
  23  * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
  24  * irrelevant.
  25  */
  26 #include <linux/io-64-nonatomic-hi-lo.h>
  27
  28 static bool force_enable_dimms;
  29 module_param(force_enable_dimms, bool, S_IRUGO|S_IWUSR);
  30 MODULE_PARM_DESC(force_enable_dimms, "Ignore _STA (ACPI DIMM device) status");
  31
  32 static bool disable_vendor_specific;
  33 module_param(disable_vendor_specific, bool, S_IRUGO);
  34 MODULE_PARM_DESC(disable_vendor_specific,
  35                 "Limit commands to the publicly specified set");
  36
  37 static unsigned long override_dsm_mask;
  38 module_param(override_dsm_mask, ulong, S_IRUGO);
  39 MODULE_PARM_DESC(override_dsm_mask, "Bitmask of allowed NVDIMM DSM functions");
  40
  41 static int default_dsm_family = -1;
  42 module_param(default_dsm_family, int, S_IRUGO);
  43 MODULE_PARM_DESC(default_dsm_family,
  44                 "Try this DSM type first when identifying NVDIMM family");
  45
  46 static bool no_init_ars;
  47 module_param(no_init_ars, bool, 0644);
  48 MODULE_PARM_DESC(no_init_ars, "Skip ARS run at nfit init time");
  49
  50 static bool force_labels;
  51 module_param(force_labels, bool, 0444);
  52 MODULE_PARM_DESC(force_labels, "Opt-in to labels despite missing methods");
  53
  54 LIST_HEAD(acpi_descs);
  55 DEFINE_MUTEX(acpi_desc_lock);
  56
  57 static struct workqueue_struct *nfit_wq;
  58
  59 struct nfit_table_prev {
  60         struct list_head spas;
  61         struct list_head memdevs;
  62         struct list_head dcrs;
  63         struct list_head bdws;
  64         struct list_head idts;
  65         struct list_head flushes;
  66 };
  67
  68 static guid_t nfit_uuid[NFIT_UUID_MAX];
  69
  70 const guid_t *to_nfit_uuid(enum nfit_uuids id)
  71 {
  72         return &nfit_uuid[id];
  73 }
  74 EXPORT_SYMBOL(to_nfit_uuid);
  75
  76 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
  77 {
  78         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
  79
  80         /*
  81          * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
  82          * acpi_device.
  83          */
  84         if (!nd_desc->provider_name
  85                         || strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
  86                 return NULL;
  87
  88         return to_acpi_device(acpi_desc->dev);
  89 }
  90
  91 static int xlat_bus_status(void *buf, unsigned int cmd, u32 status)
  92 {
  93         struct nd_cmd_clear_error *clear_err;
  94         struct nd_cmd_ars_status *ars_status;
  95         u16 flags;
  96
  97         switch (cmd) {
  98         case ND_CMD_ARS_CAP:
  99                 if ((status & 0xffff) == NFIT_ARS_CAP_NONE)
 100                         return -ENOTTY;
 101
 102                 /* Command failed */
 103                 if (status & 0xffff)
 104                         return -EIO;
 105
 106                 /* No supported scan types for this range */
 107                 flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
 108                 if ((status >> 16 & flags) == 0)
 109                         return -ENOTTY;
 110                 return 0;
 111         case ND_CMD_ARS_START:
 112                 /* ARS is in progress */
 113                 if ((status & 0xffff) == NFIT_ARS_START_BUSY)
 114                         return -EBUSY;
 115
 116                 /* Command failed */
 117                 if (status & 0xffff)
 118                         return -EIO;
 119                 return 0;
 120         case ND_CMD_ARS_STATUS:
 121                 ars_status = buf;
 122                 /* Command failed */
 123                 if (status & 0xffff)
 124                         return -EIO;
 125                 /* Check extended status (Upper two bytes) */
 126                 if (status == NFIT_ARS_STATUS_DONE)
 127                         return 0;
 128
 129                 /* ARS is in progress */
 130                 if (status == NFIT_ARS_STATUS_BUSY)
 131                         return -EBUSY;
 132
 133                 /* No ARS performed for the current boot */
 134                 if (status == NFIT_ARS_STATUS_NONE)
 135                         return -EAGAIN;
 136
 137                 /*
 138                  * ARS interrupted, either we overflowed or some other
 139                  * agent wants the scan to stop.  If we didn't overflow
 140                  * then just continue with the returned results.
 141                  */
 142                 if (status == NFIT_ARS_STATUS_INTR) {
 143                         if (ars_status->out_length >= 40 && (ars_status->flags
 144                                                 & NFIT_ARS_F_OVERFLOW))
 145                                 return -ENOSPC;
 146                         return 0;
 147                 }
 148
 149                 /* Unknown status */
 150                 if (status >> 16)
 151                         return -EIO;
 152                 return 0;
 153         case ND_CMD_CLEAR_ERROR:
 154                 clear_err = buf;
 155                 if (status & 0xffff)
 156                         return -EIO;
 157                 if (!clear_err->cleared)
 158                         return -EIO;
 159                 if (clear_err->length > clear_err->cleared)
 160                         return clear_err->cleared;
 161                 return 0;
 162         default:
 163                 break;
 164         }
 165
 166         /* all other non-zero status results in an error */
 167         if (status)
 168                 return -EIO;
 169         return 0;
 170 }
 171
 172 #define ACPI_LABELS_LOCKED 3
 173
 174 static int xlat_nvdimm_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
 175                 u32 status)
 176 {
 177         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 178
 179         switch (cmd) {
 180         case ND_CMD_GET_CONFIG_SIZE:
 181                 /*
 182                  * In the _LSI, _LSR, _LSW case the locked status is
 183                  * communicated via the read/write commands
 184                  */
 185                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags))
 186                         break;
 187
 188                 if (status >> 16 & ND_CONFIG_LOCKED)
 189                         return -EACCES;
 190                 break;
 191         case ND_CMD_GET_CONFIG_DATA:
 192                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
 193                                 && status == ACPI_LABELS_LOCKED)
 194                         return -EACCES;
 195                 break;
 196         case ND_CMD_SET_CONFIG_DATA:
 197                 if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags)
 198                                 && status == ACPI_LABELS_LOCKED)
 199                         return -EACCES;
 200                 break;
 201         default:
 202                 break;
 203         }
 204
 205         /* all other non-zero status results in an error */
 206         if (status)
 207                 return -EIO;
 208         return 0;
 209 }
 210
 211 static int xlat_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
 212                 u32 status)
 213 {
 214         if (!nvdimm)
 215                 return xlat_bus_status(buf, cmd, status);
 216         return xlat_nvdimm_status(nvdimm, buf, cmd, status);
 217 }
 218
 219 /* convert _LS{I,R} packages to the buffer object acpi_nfit_ctl expects */
 220 static union acpi_object *pkg_to_buf(union acpi_object *pkg)
 221 {
 222         int i;
 223         void *dst;
 224         size_t size = 0;
 225         union acpi_object *buf = NULL;
 226
 227         if (pkg->type != ACPI_TYPE_PACKAGE) {
 228                 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
 229                                 pkg->type);
 230                 goto err;
 231         }
 232
 233         for (i = 0; i < pkg->package.count; i++) {
 234                 union acpi_object *obj = &pkg->package.elements[i];
 235
 236                 if (obj->type == ACPI_TYPE_INTEGER)
 237                         size += 4;
 238                 else if (obj->type == ACPI_TYPE_BUFFER)
 239                         size += obj->buffer.length;
 240                 else {
 241                         WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
 242                                         obj->type);
 243                         goto err;
 244                 }
 245         }
 246
 247         buf = ACPI_ALLOCATE(sizeof(*buf) + size);
 248         if (!buf)
 249                 goto err;
 250
 251         dst = buf + 1;
 252         buf->type = ACPI_TYPE_BUFFER;
 253         buf->buffer.length = size;
 254         buf->buffer.pointer = dst;
 255         for (i = 0; i < pkg->package.count; i++) {
 256                 union acpi_object *obj = &pkg->package.elements[i];
 257
 258                 if (obj->type == ACPI_TYPE_INTEGER) {
 259                         memcpy(dst, &obj->integer.value, 4);
 260                         dst += 4;
 261                 } else if (obj->type == ACPI_TYPE_BUFFER) {
 262                         memcpy(dst, obj->buffer.pointer, obj->buffer.length);
 263                         dst += obj->buffer.length;
 264                 }
 265         }
 266 err:
 267         ACPI_FREE(pkg);
 268         return buf;
 269 }
 270
 271 static union acpi_object *int_to_buf(union acpi_object *integer)
 272 {
 273         union acpi_object *buf = ACPI_ALLOCATE(sizeof(*buf) + 4);
 274         void *dst = NULL;
 275
 276         if (!buf)
 277                 goto err;
 278
 279         if (integer->type != ACPI_TYPE_INTEGER) {
 280                 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
 281                                 integer->type);
 282                 goto err;
 283         }
 284
 285         dst = buf + 1;
 286         buf->type = ACPI_TYPE_BUFFER;
 287         buf->buffer.length = 4;
 288         buf->buffer.pointer = dst;
 289         memcpy(dst, &integer->integer.value, 4);
 290 err:
 291         ACPI_FREE(integer);
 292         return buf;
 293 }
 294
 295 static union acpi_object *acpi_label_write(acpi_handle handle, u32 offset,
 296                 u32 len, void *data)
 297 {
 298         acpi_status rc;
 299         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
 300         struct acpi_object_list input = {
 301                 .count = 3,
 302                 .pointer = (union acpi_object []) {
 303                         [0] = {
 304                                 .integer.type = ACPI_TYPE_INTEGER,
 305                                 .integer.value = offset,
 306                         },
 307                         [1] = {
 308                                 .integer.type = ACPI_TYPE_INTEGER,
 309                                 .integer.value = len,
 310                         },
 311                         [2] = {
 312                                 .buffer.type = ACPI_TYPE_BUFFER,
 313                                 .buffer.pointer = data,
 314                                 .buffer.length = len,
 315                         },
 316                 },
 317         };
 318
 319         rc = acpi_evaluate_object(handle, "_LSW", &input, &buf);
 320         if (ACPI_FAILURE(rc))
 321                 return NULL;
 322         return int_to_buf(buf.pointer);
 323 }
 324
 325 static union acpi_object *acpi_label_read(acpi_handle handle, u32 offset,
 326                 u32 len)
 327 {
 328         acpi_status rc;
 329         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
 330         struct acpi_object_list input = {
 331                 .count = 2,
 332                 .pointer = (union acpi_object []) {
 333                         [0] = {
 334                                 .integer.type = ACPI_TYPE_INTEGER,
 335                                 .integer.value = offset,
 336                         },
 337                         [1] = {
 338                                 .integer.type = ACPI_TYPE_INTEGER,
 339                                 .integer.value = len,
 340                         },
 341                 },
 342         };
 343
 344         rc = acpi_evaluate_object(handle, "_LSR", &input, &buf);
 345         if (ACPI_FAILURE(rc))
 346                 return NULL;
 347         return pkg_to_buf(buf.pointer);
 348 }
 349
 350 static union acpi_object *acpi_label_info(acpi_handle handle)
 351 {
 352         acpi_status rc;
 353         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
 354
 355         rc = acpi_evaluate_object(handle, "_LSI", NULL, &buf);
 356         if (ACPI_FAILURE(rc))
 357                 return NULL;
 358         return pkg_to_buf(buf.pointer);
 359 }
 360
 361 static u8 nfit_dsm_revid(unsigned family, unsigned func)
 362 {
 363         static const u8 revid_table[NVDIMM_FAMILY_MAX+1][NVDIMM_CMD_MAX+1] = {
 364                 [NVDIMM_FAMILY_INTEL] = {
 365                         [NVDIMM_INTEL_GET_MODES] = 2,
 366                         [NVDIMM_INTEL_GET_FWINFO] = 2,
 367                         [NVDIMM_INTEL_START_FWUPDATE] = 2,
 368                         [NVDIMM_INTEL_SEND_FWUPDATE] = 2,
 369                         [NVDIMM_INTEL_FINISH_FWUPDATE] = 2,
 370                         [NVDIMM_INTEL_QUERY_FWUPDATE] = 2,
 371                         [NVDIMM_INTEL_SET_THRESHOLD] = 2,
 372                         [NVDIMM_INTEL_INJECT_ERROR] = 2,
 373                         [NVDIMM_INTEL_GET_SECURITY_STATE] = 2,
 374                         [NVDIMM_INTEL_SET_PASSPHRASE] = 2,
 375                         [NVDIMM_INTEL_DISABLE_PASSPHRASE] = 2,
 376                         [NVDIMM_INTEL_UNLOCK_UNIT] = 2,
 377                         [NVDIMM_INTEL_FREEZE_LOCK] = 2,
 378                         [NVDIMM_INTEL_SECURE_ERASE] = 2,
 379                         [NVDIMM_INTEL_OVERWRITE] = 2,
 380                         [NVDIMM_INTEL_QUERY_OVERWRITE] = 2,
 381                         [NVDIMM_INTEL_SET_MASTER_PASSPHRASE] = 2,
 382                         [NVDIMM_INTEL_MASTER_SECURE_ERASE] = 2,
 383                 },
 384         };
 385         u8 id;
 386
 387         if (family > NVDIMM_FAMILY_MAX)
 388                 return 0;
 389         if (func > NVDIMM_CMD_MAX)
 390                 return 0;
 391         id = revid_table[family][func];
 392         if (id == 0)
 393                 return 1; /* default */
 394         return id;
 395 }
 396
 397 static bool payload_dumpable(struct nvdimm *nvdimm, unsigned int func)
 398 {
 399         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 400
 401         if (nfit_mem && nfit_mem->family == NVDIMM_FAMILY_INTEL
 402                         && func >= NVDIMM_INTEL_GET_SECURITY_STATE
 403                         && func <= NVDIMM_INTEL_MASTER_SECURE_ERASE)
 404                 return IS_ENABLED(CONFIG_NFIT_SECURITY_DEBUG);
 405         return true;
 406 }
 407
 408 static int cmd_to_func(struct nfit_mem *nfit_mem, unsigned int cmd,
 409                 struct nd_cmd_pkg *call_pkg)
 410 {
 411         if (call_pkg) {
 412                 int i;
 413
 414                 if (nfit_mem && nfit_mem->family != call_pkg->nd_family)
 415                         return -ENOTTY;
 416
 417                 for (i = 0; i < ARRAY_SIZE(call_pkg->nd_reserved2); i++)
 418                         if (call_pkg->nd_reserved2[i])
 419                                 return -EINVAL;
 420                 return call_pkg->nd_command;
 421         }
 422
 423         /* In the !call_pkg case, bus commands == bus functions */
 424         if (!nfit_mem)
 425                 return cmd;
 426
 427         /* Linux ND commands == NVDIMM_FAMILY_INTEL function numbers */
 428         if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
 429                 return cmd;
 430
 431         /*
 432          * Force function number validation to fail since 0 is never
 433          * published as a valid function in dsm_mask.
 434          */
 435         return 0;
 436 }
 437
 438 int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm,
 439                 unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc)
 440 {
 441         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
 442         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 443         union acpi_object in_obj, in_buf, *out_obj;
 444         const struct nd_cmd_desc *desc = NULL;
 445         struct device *dev = acpi_desc->dev;
 446         struct nd_cmd_pkg *call_pkg = NULL;
 447         const char *cmd_name, *dimm_name;
 448         unsigned long cmd_mask, dsm_mask;
 449         u32 offset, fw_status = 0;
 450         acpi_handle handle;
 451         const guid_t *guid;
 452         int func, rc, i;
 453
 454         if (cmd_rc)
 455                 *cmd_rc = -EINVAL;
 456
 457         if (cmd == ND_CMD_CALL)
 458                 call_pkg = buf;
 459         func = cmd_to_func(nfit_mem, cmd, call_pkg);
 460         if (func < 0)
 461                 return func;
 462
 463         if (nvdimm) {
 464                 struct acpi_device *adev = nfit_mem->adev;
 465
 466                 if (!adev)
 467                         return -ENOTTY;
 468
 469                 dimm_name = nvdimm_name(nvdimm);
 470                 cmd_name = nvdimm_cmd_name(cmd);
 471                 cmd_mask = nvdimm_cmd_mask(nvdimm);
 472                 dsm_mask = nfit_mem->dsm_mask;
 473                 desc = nd_cmd_dimm_desc(cmd);
 474                 guid = to_nfit_uuid(nfit_mem->family);
 475                 handle = adev->handle;
 476         } else {
 477                 struct acpi_device *adev = to_acpi_dev(acpi_desc);
 478
 479                 cmd_name = nvdimm_bus_cmd_name(cmd);
 480                 cmd_mask = nd_desc->cmd_mask;
 481                 dsm_mask = nd_desc->bus_dsm_mask;
 482                 desc = nd_cmd_bus_desc(cmd);
 483                 guid = to_nfit_uuid(NFIT_DEV_BUS);
 484                 handle = adev->handle;
 485                 dimm_name = "bus";
 486         }
 487
 488         if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
 489                 return -ENOTTY;
 490
 491         /*
 492          * Check for a valid command.  For ND_CMD_CALL, we also have to
 493          * make sure that the DSM function is supported.
 494          */
 495         if (cmd == ND_CMD_CALL &&
 496             (func > NVDIMM_CMD_MAX || !test_bit(func, &dsm_mask)))
 497                 return -ENOTTY;
 498         else if (!test_bit(cmd, &cmd_mask))
 499                 return -ENOTTY;
 500
 501         in_obj.type = ACPI_TYPE_PACKAGE;
 502         in_obj.package.count = 1;
 503         in_obj.package.elements = &in_buf;
 504         in_buf.type = ACPI_TYPE_BUFFER;
 505         in_buf.buffer.pointer = buf;
 506         in_buf.buffer.length = 0;
 507
 508         /* libnvdimm has already validated the input envelope */
 509         for (i = 0; i < desc->in_num; i++)
 510                 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
 511                                 i, buf);
 512
 513         if (call_pkg) {
 514                 /* skip over package wrapper */
 515                 in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
 516                 in_buf.buffer.length = call_pkg->nd_size_in;
 517         }
 518
 519         dev_dbg(dev, "%s cmd: %d: func: %d input length: %d\n",
 520                 dimm_name, cmd, func, in_buf.buffer.length);
 521         if (payload_dumpable(nvdimm, func))
 522                 print_hex_dump_debug("nvdimm in  ", DUMP_PREFIX_OFFSET, 4, 4,
 523                                 in_buf.buffer.pointer,
 524                                 min_t(u32, 256, in_buf.buffer.length), true);
 525
 526         /* call the BIOS, prefer the named methods over _DSM if available */
 527         if (nvdimm && cmd == ND_CMD_GET_CONFIG_SIZE
 528                         && test_bit(NFIT_MEM_LSR, &nfit_mem->flags))
 529                 out_obj = acpi_label_info(handle);
 530         else if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA
 531                         && test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
 532                 struct nd_cmd_get_config_data_hdr *p = buf;
 533
 534                 out_obj = acpi_label_read(handle, p->in_offset, p->in_length);
 535         } else if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA
 536                         && test_bit(NFIT_MEM_LSW, &nfit_mem->flags)) {
 537                 struct nd_cmd_set_config_hdr *p = buf;
 538
 539                 out_obj = acpi_label_write(handle, p->in_offset, p->in_length,
 540                                 p->in_buf);
 541         } else {
 542                 u8 revid;
 543
 544                 if (nvdimm)
 545                         revid = nfit_dsm_revid(nfit_mem->family, func);
 546                 else
 547                         revid = 1;
 548                 out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
 549         }
 550
 551         if (!out_obj) {
 552                 dev_dbg(dev, "%s _DSM failed cmd: %s\n", dimm_name, cmd_name);
 553                 return -EINVAL;
 554         }
 555
 556         if (out_obj->type != ACPI_TYPE_BUFFER) {
 557                 dev_dbg(dev, "%s unexpected output object type cmd: %s type: %d\n",
 558                                 dimm_name, cmd_name, out_obj->type);
 559                 rc = -EINVAL;
 560                 goto out;
 561         }
 562
 563         dev_dbg(dev, "%s cmd: %s output length: %d\n", dimm_name,
 564                         cmd_name, out_obj->buffer.length);
 565         print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4, 4,
 566                         out_obj->buffer.pointer,
 567                         min_t(u32, 128, out_obj->buffer.length), true);
 568
 569         if (call_pkg) {
 570                 call_pkg->nd_fw_size = out_obj->buffer.length;
 571                 memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
 572                         out_obj->buffer.pointer,
 573                         min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
 574
 575                 ACPI_FREE(out_obj);
 576                 /*
 577                  * Need to support FW function w/o known size in advance.
 578                  * Caller can determine required size based upon nd_fw_size.
 579                  * If we return an error (like elsewhere) then caller wouldn't
 580                  * be able to rely upon data returned to make calculation.
 581                  */
 582                 if (cmd_rc)
 583                         *cmd_rc = 0;
 584                 return 0;
 585         }
 586
 587         for (i = 0, offset = 0; i < desc->out_num; i++) {
 588                 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
 589                                 (u32 *) out_obj->buffer.pointer,
 590                                 out_obj->buffer.length - offset);
 591
 592                 if (offset + out_size > out_obj->buffer.length) {
 593                         dev_dbg(dev, "%s output object underflow cmd: %s field: %d\n",
 594                                         dimm_name, cmd_name, i);
 595                         break;
 596                 }
 597
 598                 if (in_buf.buffer.length + offset + out_size > buf_len) {
 599                         dev_dbg(dev, "%s output overrun cmd: %s field: %d\n",
 600                                         dimm_name, cmd_name, i);
 601                         rc = -ENXIO;
 602                         goto out;
 603                 }
 604                 memcpy(buf + in_buf.buffer.length + offset,
 605                                 out_obj->buffer.pointer + offset, out_size);
 606                 offset += out_size;
 607         }
 608
 609         /*
 610          * Set fw_status for all the commands with a known format to be
 611          * later interpreted by xlat_status().
 612          */
 613         if (i >= 1 && ((!nvdimm && cmd >= ND_CMD_ARS_CAP
 614                                         && cmd <= ND_CMD_CLEAR_ERROR)
 615                                 || (nvdimm && cmd >= ND_CMD_SMART
 616                                         && cmd <= ND_CMD_VENDOR)))
 617                 fw_status = *(u32 *) out_obj->buffer.pointer;
 618
 619         if (offset + in_buf.buffer.length < buf_len) {
 620                 if (i >= 1) {
 621                         /*
 622                          * status valid, return the number of bytes left
 623                          * unfilled in the output buffer
 624                          */
 625                         rc = buf_len - offset - in_buf.buffer.length;
 626                         if (cmd_rc)
 627                                 *cmd_rc = xlat_status(nvdimm, buf, cmd,
 628                                                 fw_status);
 629                 } else {
 630                         dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
 631                                         __func__, dimm_name, cmd_name, buf_len,
 632                                         offset);
 633                         rc = -ENXIO;
 634                 }
 635         } else {
 636                 rc = 0;
 637                 if (cmd_rc)
 638                         *cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status);
 639         }
 640
 641  out:
 642         ACPI_FREE(out_obj);
 643
 644         return rc;
 645 }
 646 EXPORT_SYMBOL_GPL(acpi_nfit_ctl);
 647
 648 static const char *spa_type_name(u16 type)
 649 {
 650         static const char *to_name[] = {
 651                 [NFIT_SPA_VOLATILE] = "volatile",
 652                 [NFIT_SPA_PM] = "pmem",
 653                 [NFIT_SPA_DCR] = "dimm-control-region",
 654                 [NFIT_SPA_BDW] = "block-data-window",
 655                 [NFIT_SPA_VDISK] = "volatile-disk",
 656                 [NFIT_SPA_VCD] = "volatile-cd",
 657                 [NFIT_SPA_PDISK] = "persistent-disk",
 658                 [NFIT_SPA_PCD] = "persistent-cd",
 659
 660         };
 661
 662         if (type > NFIT_SPA_PCD)
 663                 return "unknown";
 664
 665         return to_name[type];
 666 }
 667
 668 int nfit_spa_type(struct acpi_nfit_system_address *spa)
 669 {
 670         int i;
 671
 672         for (i = 0; i < NFIT_UUID_MAX; i++)
 673                 if (guid_equal(to_nfit_uuid(i), (guid_t *)&spa->range_guid))
 674                         return i;
 675         return -1;
 676 }
 677
 678 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
 679                 struct nfit_table_prev *prev,
 680                 struct acpi_nfit_system_address *spa)
 681 {
 682         struct device *dev = acpi_desc->dev;
 683         struct nfit_spa *nfit_spa;
 684
 685         if (spa->header.length != sizeof(*spa))
 686                 return false;
 687
 688         list_for_each_entry(nfit_spa, &prev->spas, list) {
 689                 if (memcmp(nfit_spa->spa, spa, sizeof(*spa)) == 0) {
 690                         list_move_tail(&nfit_spa->list, &acpi_desc->spas);
 691                         return true;
 692                 }
 693         }
 694
 695         nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof(*spa),
 696                         GFP_KERNEL);
 697         if (!nfit_spa)
 698                 return false;
 699         INIT_LIST_HEAD(&nfit_spa->list);
 700         memcpy(nfit_spa->spa, spa, sizeof(*spa));
 701         list_add_tail(&nfit_spa->list, &acpi_desc->spas);
 702         dev_dbg(dev, "spa index: %d type: %s\n",
 703                         spa->range_index,
 704                         spa_type_name(nfit_spa_type(spa)));
 705         return true;
 706 }
 707
 708 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
 709                 struct nfit_table_prev *prev,
 710                 struct acpi_nfit_memory_map *memdev)
 711 {
 712         struct device *dev = acpi_desc->dev;
 713         struct nfit_memdev *nfit_memdev;
 714
 715         if (memdev->header.length != sizeof(*memdev))
 716                 return false;
 717
 718         list_for_each_entry(nfit_memdev, &prev->memdevs, list)
 719                 if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
 720                         list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 721                         return true;
 722                 }
 723
 724         nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
 725                         GFP_KERNEL);
 726         if (!nfit_memdev)
 727                 return false;
 728         INIT_LIST_HEAD(&nfit_memdev->list);
 729         memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
 730         list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 731         dev_dbg(dev, "memdev handle: %#x spa: %d dcr: %d flags: %#x\n",
 732                         memdev->device_handle, memdev->range_index,
 733                         memdev->region_index, memdev->flags);
 734         return true;
 735 }
 736
 737 int nfit_get_smbios_id(u32 device_handle, u16 *flags)
 738 {
 739         struct acpi_nfit_memory_map *memdev;
 740         struct acpi_nfit_desc *acpi_desc;
 741         struct nfit_mem *nfit_mem;
 742         u16 physical_id;
 743
 744         mutex_lock(&acpi_desc_lock);
 745         list_for_each_entry(acpi_desc, &acpi_descs, list) {
 746                 mutex_lock(&acpi_desc->init_mutex);
 747                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
 748                         memdev = __to_nfit_memdev(nfit_mem);
 749                         if (memdev->device_handle == device_handle) {
 750                                 *flags = memdev->flags;
 751                                 physical_id = memdev->physical_id;
 752                                 mutex_unlock(&acpi_desc->init_mutex);
 753                                 mutex_unlock(&acpi_desc_lock);
 754                                 return physical_id;
 755                         }
 756                 }
 757                 mutex_unlock(&acpi_desc->init_mutex);
 758         }
 759         mutex_unlock(&acpi_desc_lock);
 760
 761         return -ENODEV;
 762 }
 763 EXPORT_SYMBOL_GPL(nfit_get_smbios_id);
 764
 765 /*
 766  * An implementation may provide a truncated control region if no block windows
 767  * are defined.
 768  */
 769 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
 770 {
 771         if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
 772                                 window_size))
 773                 return 0;
 774         if (dcr->windows)
 775                 return sizeof(*dcr);
 776         return offsetof(struct acpi_nfit_control_region, window_size);
 777 }
 778
 779 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
 780                 struct nfit_table_prev *prev,
 781                 struct acpi_nfit_control_region *dcr)
 782 {
 783         struct device *dev = acpi_desc->dev;
 784         struct nfit_dcr *nfit_dcr;
 785
 786         if (!sizeof_dcr(dcr))
 787                 return false;
 788
 789         list_for_each_entry(nfit_dcr, &prev->dcrs, list)
 790                 if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
 791                         list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 792                         return true;
 793                 }
 794
 795         nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
 796                         GFP_KERNEL);
 797         if (!nfit_dcr)
 798                 return false;
 799         INIT_LIST_HEAD(&nfit_dcr->list);
 800         memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
 801         list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 802         dev_dbg(dev, "dcr index: %d windows: %d\n",
 803                         dcr->region_index, dcr->windows);
 804         return true;
 805 }
 806
 807 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
 808                 struct nfit_table_prev *prev,
 809                 struct acpi_nfit_data_region *bdw)
 810 {
 811         struct device *dev = acpi_desc->dev;
 812         struct nfit_bdw *nfit_bdw;
 813
 814         if (bdw->header.length != sizeof(*bdw))
 815                 return false;
 816         list_for_each_entry(nfit_bdw, &prev->bdws, list)
 817                 if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
 818                         list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
 819                         return true;
 820                 }
 821
 822         nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
 823                         GFP_KERNEL);
 824         if (!nfit_bdw)
 825                 return false;
 826         INIT_LIST_HEAD(&nfit_bdw->list);
 827         memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
 828         list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
 829         dev_dbg(dev, "bdw dcr: %d windows: %d\n",
 830                         bdw->region_index, bdw->windows);
 831         return true;
 832 }
 833
 834 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
 835 {
 836         if (idt->header.length < sizeof(*idt))
 837                 return 0;
 838         return sizeof(*idt) + sizeof(u32) * (idt->line_count - 1);
 839 }
 840
 841 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
 842                 struct nfit_table_prev *prev,
 843                 struct acpi_nfit_interleave *idt)
 844 {
 845         struct device *dev = acpi_desc->dev;
 846         struct nfit_idt *nfit_idt;
 847
 848         if (!sizeof_idt(idt))
 849                 return false;
 850
 851         list_for_each_entry(nfit_idt, &prev->idts, list) {
 852                 if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
 853                         continue;
 854
 855                 if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
 856                         list_move_tail(&nfit_idt->list, &acpi_desc->idts);
 857                         return true;
 858                 }
 859         }
 860
 861         nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
 862                         GFP_KERNEL);
 863         if (!nfit_idt)
 864                 return false;
 865         INIT_LIST_HEAD(&nfit_idt->list);
 866         memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
 867         list_add_tail(&nfit_idt->list, &acpi_desc->idts);
 868         dev_dbg(dev, "idt index: %d num_lines: %d\n",
 869                         idt->interleave_index, idt->line_count);
 870         return true;
 871 }
 872
 873 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
 874 {
 875         if (flush->header.length < sizeof(*flush))
 876                 return 0;
 877         return sizeof(*flush) + sizeof(u64) * (flush->hint_count - 1);
 878 }
 879
 880 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
 881                 struct nfit_table_prev *prev,
 882                 struct acpi_nfit_flush_address *flush)
 883 {
 884         struct device *dev = acpi_desc->dev;
 885         struct nfit_flush *nfit_flush;
 886
 887         if (!sizeof_flush(flush))
 888                 return false;
 889
 890         list_for_each_entry(nfit_flush, &prev->flushes, list) {
 891                 if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
 892                         continue;
 893
 894                 if (memcmp(nfit_flush->flush, flush,
 895                                         sizeof_flush(flush)) == 0) {
 896                         list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
 897                         return true;
 898                 }
 899         }
 900
 901         nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
 902                         + sizeof_flush(flush), GFP_KERNEL);
 903         if (!nfit_flush)
 904                 return false;
 905         INIT_LIST_HEAD(&nfit_flush->list);
 906         memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
 907         list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
 908         dev_dbg(dev, "nfit_flush handle: %d hint_count: %d\n",
 909                         flush->device_handle, flush->hint_count);
 910         return true;
 911 }
 912
 913 static bool add_platform_cap(struct acpi_nfit_desc *acpi_desc,
 914                 struct acpi_nfit_capabilities *pcap)
 915 {
 916         struct device *dev = acpi_desc->dev;
 917         u32 mask;
 918
 919         mask = (1 << (pcap->highest_capability + 1)) - 1;
 920         acpi_desc->platform_cap = pcap->capabilities & mask;
 921         dev_dbg(dev, "cap: %#x\n", acpi_desc->platform_cap);
 922         return true;
 923 }
 924
 925 static void *add_table(struct acpi_nfit_desc *acpi_desc,
 926                 struct nfit_table_prev *prev, void *table, const void *end)
 927 {
 928         struct device *dev = acpi_desc->dev;
 929         struct acpi_nfit_header *hdr;
 930         void *err = ERR_PTR(-ENOMEM);
 931
 932         if (table >= end)
 933                 return NULL;
 934
 935         hdr = table;
 936         if (!hdr->length) {
 937                 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
 938                         hdr->type);
 939                 return NULL;
 940         }
 941
 942         switch (hdr->type) {
 943         case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
 944                 if (!add_spa(acpi_desc, prev, table))
 945                         return err;
 946                 break;
 947         case ACPI_NFIT_TYPE_MEMORY_MAP:
 948                 if (!add_memdev(acpi_desc, prev, table))
 949                         return err;
 950                 break;
 951         case ACPI_NFIT_TYPE_CONTROL_REGION:
 952                 if (!add_dcr(acpi_desc, prev, table))
 953                         return err;
 954                 break;
 955         case ACPI_NFIT_TYPE_DATA_REGION:
 956                 if (!add_bdw(acpi_desc, prev, table))
 957                         return err;
 958                 break;
 959         case ACPI_NFIT_TYPE_INTERLEAVE:
 960                 if (!add_idt(acpi_desc, prev, table))
 961                         return err;
 962                 break;
 963         case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
 964                 if (!add_flush(acpi_desc, prev, table))
 965                         return err;
 966                 break;
 967         case ACPI_NFIT_TYPE_SMBIOS:
 968                 dev_dbg(dev, "smbios\n");
 969                 break;
 970         case ACPI_NFIT_TYPE_CAPABILITIES:
 971                 if (!add_platform_cap(acpi_desc, table))
 972                         return err;
 973                 break;
 974         default:
 975                 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
 976                 break;
 977         }
 978
 979         return table + hdr->length;
 980 }
 981
 982 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
 983                 struct nfit_mem *nfit_mem)
 984 {
 985         u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
 986         u16 dcr = nfit_mem->dcr->region_index;
 987         struct nfit_spa *nfit_spa;
 988
 989         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
 990                 u16 range_index = nfit_spa->spa->range_index;
 991                 int type = nfit_spa_type(nfit_spa->spa);
 992                 struct nfit_memdev *nfit_memdev;
 993
 994                 if (type != NFIT_SPA_BDW)
 995                         continue;
 996
 997                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 998                         if (nfit_memdev->memdev->range_index != range_index)
 999                                 continue;
1000                         if (nfit_memdev->memdev->device_handle != device_handle)
1001                                 continue;
1002                         if (nfit_memdev->memdev->region_index != dcr)
1003                                 continue;
1004
1005                         nfit_mem->spa_bdw = nfit_spa->spa;
1006                         return;
1007                 }
1008         }
1009
1010         dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
1011                         nfit_mem->spa_dcr->range_index);
1012         nfit_mem->bdw = NULL;
1013 }
1014
1015 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
1016                 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
1017 {
1018         u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
1019         struct nfit_memdev *nfit_memdev;
1020         struct nfit_bdw *nfit_bdw;
1021         struct nfit_idt *nfit_idt;
1022         u16 idt_idx, range_index;
1023
1024         list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
1025                 if (nfit_bdw->bdw->region_index != dcr)
1026                         continue;
1027                 nfit_mem->bdw = nfit_bdw->bdw;
1028                 break;
1029         }
1030
1031         if (!nfit_mem->bdw)
1032                 return;
1033
1034         nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
1035
1036         if (!nfit_mem->spa_bdw)
1037                 return;
1038
1039         range_index = nfit_mem->spa_bdw->range_index;
1040         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1041                 if (nfit_memdev->memdev->range_index != range_index ||
1042                                 nfit_memdev->memdev->region_index != dcr)
1043                         continue;
1044                 nfit_mem->memdev_bdw = nfit_memdev->memdev;
1045                 idt_idx = nfit_memdev->memdev->interleave_index;
1046                 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1047                         if (nfit_idt->idt->interleave_index != idt_idx)
1048                                 continue;
1049                         nfit_mem->idt_bdw = nfit_idt->idt;
1050                         break;
1051                 }
1052                 break;
1053         }
1054 }
1055
1056 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
1057                 struct acpi_nfit_system_address *spa)
1058 {
1059         struct nfit_mem *nfit_mem, *found;
1060         struct nfit_memdev *nfit_memdev;
1061         int type = spa ? nfit_spa_type(spa) : 0;
1062
1063         switch (type) {
1064         case NFIT_SPA_DCR:
1065         case NFIT_SPA_PM:
1066                 break;
1067         default:
1068                 if (spa)
1069                         return 0;
1070         }
1071
1072         /*
1073          * This loop runs in two modes, when a dimm is mapped the loop
1074          * adds memdev associations to an existing dimm, or creates a
1075          * dimm. In the unmapped dimm case this loop sweeps for memdev
1076          * instances with an invalid / zero range_index and adds those
1077          * dimms without spa associations.
1078          */
1079         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1080                 struct nfit_flush *nfit_flush;
1081                 struct nfit_dcr *nfit_dcr;
1082                 u32 device_handle;
1083                 u16 dcr;
1084
1085                 if (spa && nfit_memdev->memdev->range_index != spa->range_index)
1086                         continue;
1087                 if (!spa && nfit_memdev->memdev->range_index)
1088                         continue;
1089                 found = NULL;
1090                 dcr = nfit_memdev->memdev->region_index;
1091                 device_handle = nfit_memdev->memdev->device_handle;
1092                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1093                         if (__to_nfit_memdev(nfit_mem)->device_handle
1094                                         == device_handle) {
1095                                 found = nfit_mem;
1096                                 break;
1097                         }
1098
1099                 if (found)
1100                         nfit_mem = found;
1101                 else {
1102                         nfit_mem = devm_kzalloc(acpi_desc->dev,
1103                                         sizeof(*nfit_mem), GFP_KERNEL);
1104                         if (!nfit_mem)
1105                                 return -ENOMEM;
1106                         INIT_LIST_HEAD(&nfit_mem->list);
1107                         nfit_mem->acpi_desc = acpi_desc;
1108                         list_add(&nfit_mem->list, &acpi_desc->dimms);
1109                 }
1110
1111                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1112                         if (nfit_dcr->dcr->region_index != dcr)
1113                                 continue;
1114                         /*
1115                          * Record the control region for the dimm.  For
1116                          * the ACPI 6.1 case, where there are separate
1117                          * control regions for the pmem vs blk
1118                          * interfaces, be sure to record the extended
1119                          * blk details.
1120                          */
1121                         if (!nfit_mem->dcr)
1122                                 nfit_mem->dcr = nfit_dcr->dcr;
1123                         else if (nfit_mem->dcr->windows == 0
1124                                         && nfit_dcr->dcr->windows)
1125                                 nfit_mem->dcr = nfit_dcr->dcr;
1126                         break;
1127                 }
1128
1129                 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
1130                         struct acpi_nfit_flush_address *flush;
1131                         u16 i;
1132
1133                         if (nfit_flush->flush->device_handle != device_handle)
1134                                 continue;
1135                         nfit_mem->nfit_flush = nfit_flush;
1136                         flush = nfit_flush->flush;
1137                         nfit_mem->flush_wpq = devm_kcalloc(acpi_desc->dev,
1138                                         flush->hint_count,
1139                                         sizeof(struct resource),
1140                                         GFP_KERNEL);
1141                         if (!nfit_mem->flush_wpq)
1142                                 return -ENOMEM;
1143                         for (i = 0; i < flush->hint_count; i++) {
1144                                 struct resource *res = &nfit_mem->flush_wpq[i];
1145
1146                                 res->start = flush->hint_address[i];
1147                                 res->end = res->start + 8 - 1;
1148                         }
1149                         break;
1150                 }
1151
1152                 if (dcr && !nfit_mem->dcr) {
1153                         dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
1154                                         spa->range_index, dcr);
1155                         return -ENODEV;
1156                 }
1157
1158                 if (type == NFIT_SPA_DCR) {
1159                         struct nfit_idt *nfit_idt;
1160                         u16 idt_idx;
1161
1162                         /* multiple dimms may share a SPA when interleaved */
1163                         nfit_mem->spa_dcr = spa;
1164                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
1165                         idt_idx = nfit_memdev->memdev->interleave_index;
1166                         list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1167                                 if (nfit_idt->idt->interleave_index != idt_idx)
1168                                         continue;
1169                                 nfit_mem->idt_dcr = nfit_idt->idt;
1170                                 break;
1171                         }
1172                         nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
1173                 } else if (type == NFIT_SPA_PM) {
1174                         /*
1175                          * A single dimm may belong to multiple SPA-PM
1176                          * ranges, record at least one in addition to
1177                          * any SPA-DCR range.
1178                          */
1179                         nfit_mem->memdev_pmem = nfit_memdev->memdev;
1180                 } else
1181                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
1182         }
1183
1184         return 0;
1185 }
1186
1187 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
1188 {
1189         struct nfit_mem *a = container_of(_a, typeof(*a), list);
1190         struct nfit_mem *b = container_of(_b, typeof(*b), list);
1191         u32 handleA, handleB;
1192
1193         handleA = __to_nfit_memdev(a)->device_handle;
1194         handleB = __to_nfit_memdev(b)->device_handle;
1195         if (handleA < handleB)
1196                 return -1;
1197         else if (handleA > handleB)
1198                 return 1;
1199         return 0;
1200 }
1201
1202 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
1203 {
1204         struct nfit_spa *nfit_spa;
1205         int rc;
1206
1207
1208         /*
1209          * For each SPA-DCR or SPA-PMEM address range find its
1210          * corresponding MEMDEV(s).  From each MEMDEV find the
1211          * corresponding DCR.  Then, if we're operating on a SPA-DCR,
1212          * try to find a SPA-BDW and a corresponding BDW that references
1213          * the DCR.  Throw it all into an nfit_mem object.  Note, that
1214          * BDWs are optional.
1215          */
1216         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
1217                 rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
1218                 if (rc)
1219                         return rc;
1220         }
1221
1222         /*
1223          * If a DIMM has failed to be mapped into SPA there will be no
1224          * SPA entries above. Find and register all the unmapped DIMMs
1225          * for reporting and recovery purposes.
1226          */
1227         rc = __nfit_mem_init(acpi_desc, NULL);
1228         if (rc)
1229                 return rc;
1230
1231         list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
1232
1233         return 0;
1234 }
1235
1236 static ssize_t bus_dsm_mask_show(struct device *dev,
1237                 struct device_attribute *attr, char *buf)
1238 {
1239         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1240         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1241
1242         return sprintf(buf, "%#lx\n", nd_desc->bus_dsm_mask);
1243 }
1244 static struct device_attribute dev_attr_bus_dsm_mask =
1245                 __ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
1246
1247 static ssize_t revision_show(struct device *dev,
1248                 struct device_attribute *attr, char *buf)
1249 {
1250         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1251         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1252         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1253
1254         return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
1255 }
1256 static DEVICE_ATTR_RO(revision);
1257
1258 static ssize_t hw_error_scrub_show(struct device *dev,
1259                 struct device_attribute *attr, char *buf)
1260 {
1261         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1262         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1263         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1264
1265         return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
1266 }
1267
1268 /*
1269  * The 'hw_error_scrub' attribute can have the following values written to it:
1270  * '0': Switch to the default mode where an exception will only insert
1271  *      the address of the memory error into the poison and badblocks lists.
1272  * '1': Enable a full scrub to happen if an exception for a memory error is
1273  *      received.
1274  */
1275 static ssize_t hw_error_scrub_store(struct device *dev,
1276                 struct device_attribute *attr, const char *buf, size_t size)
1277 {
1278         struct nvdimm_bus_descriptor *nd_desc;
1279         ssize_t rc;
1280         long val;
1281
1282         rc = kstrtol(buf, 0, &val);
1283         if (rc)
1284                 return rc;
1285
1286         nfit_device_lock(dev);
1287         nd_desc = dev_get_drvdata(dev);
1288         if (nd_desc) {
1289                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1290
1291                 switch (val) {
1292                 case HW_ERROR_SCRUB_ON:
1293                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
1294                         break;
1295                 case HW_ERROR_SCRUB_OFF:
1296                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
1297                         break;
1298                 default:
1299                         rc = -EINVAL;
1300                         break;
1301                 }
1302         }
1303         nfit_device_unlock(dev);
1304         if (rc)
1305                 return rc;
1306         return size;
1307 }
1308 static DEVICE_ATTR_RW(hw_error_scrub);
1309
1310 /*
1311  * This shows the number of full Address Range Scrubs that have been
1312  * completed since driver load time. Userspace can wait on this using
1313  * select/poll etc. A '+' at the end indicates an ARS is in progress
1314  */
1315 static ssize_t scrub_show(struct device *dev,
1316                 struct device_attribute *attr, char *buf)
1317 {
1318         struct nvdimm_bus_descriptor *nd_desc;
1319         struct acpi_nfit_desc *acpi_desc;
1320         ssize_t rc = -ENXIO;
1321         bool busy;
1322
1323         nfit_device_lock(dev);
1324         nd_desc = dev_get_drvdata(dev);
1325         if (!nd_desc) {
1326                 nfit_device_unlock(dev);
1327                 return rc;
1328         }
1329         acpi_desc = to_acpi_desc(nd_desc);
1330
1331         mutex_lock(&acpi_desc->init_mutex);
1332         busy = test_bit(ARS_BUSY, &acpi_desc->scrub_flags)
1333                 && !test_bit(ARS_CANCEL, &acpi_desc->scrub_flags);
1334         rc = sprintf(buf, "%d%s", acpi_desc->scrub_count, busy ? "+\n" : "\n");
1335         /* Allow an admin to poll the busy state at a higher rate */
1336         if (busy && capable(CAP_SYS_RAWIO) && !test_and_set_bit(ARS_POLL,
1337                                 &acpi_desc->scrub_flags)) {
1338                 acpi_desc->scrub_tmo = 1;
1339                 mod_delayed_work(nfit_wq, &acpi_desc->dwork, HZ);
1340         }
1341
1342         mutex_unlock(&acpi_desc->init_mutex);
1343         nfit_device_unlock(dev);
1344         return rc;
1345 }
1346
1347 static ssize_t scrub_store(struct device *dev,
1348                 struct device_attribute *attr, const char *buf, size_t size)
1349 {
1350         struct nvdimm_bus_descriptor *nd_desc;
1351         ssize_t rc;
1352         long val;
1353
1354         rc = kstrtol(buf, 0, &val);
1355         if (rc)
1356                 return rc;
1357         if (val != 1)
1358                 return -EINVAL;
1359
1360         nfit_device_lock(dev);
1361         nd_desc = dev_get_drvdata(dev);
1362         if (nd_desc) {
1363                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1364
1365                 rc = acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
1366         }
1367         nfit_device_unlock(dev);
1368         if (rc)
1369                 return rc;
1370         return size;
1371 }
1372 static DEVICE_ATTR_RW(scrub);
1373
1374 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1375 {
1376         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1377         const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1378                 | 1 << ND_CMD_ARS_STATUS;
1379
1380         return (nd_desc->cmd_mask & mask) == mask;
1381 }
1382
1383 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1384 {
1385         struct device *dev = container_of(kobj, struct device, kobj);
1386         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1387
1388         if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
1389                 return 0;
1390         return a->mode;
1391 }
1392
1393 static struct attribute *acpi_nfit_attributes[] = {
1394         &dev_attr_revision.attr,
1395         &dev_attr_scrub.attr,
1396         &dev_attr_hw_error_scrub.attr,
1397         &dev_attr_bus_dsm_mask.attr,
1398         NULL,
1399 };
1400
1401 static const struct attribute_group acpi_nfit_attribute_group = {
1402         .name = "nfit",
1403         .attrs = acpi_nfit_attributes,
1404         .is_visible = nfit_visible,
1405 };
1406
1407 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1408         &acpi_nfit_attribute_group,
1409         NULL,
1410 };
1411
1412 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1413 {
1414         struct nvdimm *nvdimm = to_nvdimm(dev);
1415         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1416
1417         return __to_nfit_memdev(nfit_mem);
1418 }
1419
1420 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1421 {
1422         struct nvdimm *nvdimm = to_nvdimm(dev);
1423         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1424
1425         return nfit_mem->dcr;
1426 }
1427
1428 static ssize_t handle_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->device_handle);
1434 }
1435 static DEVICE_ATTR_RO(handle);
1436
1437 static ssize_t phys_id_show(struct device *dev,
1438                 struct device_attribute *attr, char *buf)
1439 {
1440         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1441
1442         return sprintf(buf, "%#x\n", memdev->physical_id);
1443 }
1444 static DEVICE_ATTR_RO(phys_id);
1445
1446 static ssize_t vendor_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->vendor_id));
1452 }
1453 static DEVICE_ATTR_RO(vendor);
1454
1455 static ssize_t rev_id_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->revision_id));
1461 }
1462 static DEVICE_ATTR_RO(rev_id);
1463
1464 static ssize_t device_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->device_id));
1470 }
1471 static DEVICE_ATTR_RO(device);
1472
1473 static ssize_t subsystem_vendor_show(struct device *dev,
1474                 struct device_attribute *attr, char *buf)
1475 {
1476         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1477
1478         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1479 }
1480 static DEVICE_ATTR_RO(subsystem_vendor);
1481
1482 static ssize_t subsystem_rev_id_show(struct device *dev,
1483                 struct device_attribute *attr, char *buf)
1484 {
1485         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1486
1487         return sprintf(buf, "0x%04x\n",
1488                         be16_to_cpu(dcr->subsystem_revision_id));
1489 }
1490 static DEVICE_ATTR_RO(subsystem_rev_id);
1491
1492 static ssize_t subsystem_device_show(struct device *dev,
1493                 struct device_attribute *attr, char *buf)
1494 {
1495         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1496
1497         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1498 }
1499 static DEVICE_ATTR_RO(subsystem_device);
1500
1501 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1502 {
1503         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1504         int formats = 0;
1505
1506         if (nfit_mem->memdev_pmem)
1507                 formats++;
1508         if (nfit_mem->memdev_bdw)
1509                 formats++;
1510         return formats;
1511 }
1512
1513 static ssize_t format_show(struct device *dev,
1514                 struct device_attribute *attr, char *buf)
1515 {
1516         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1517
1518         return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1519 }
1520 static DEVICE_ATTR_RO(format);
1521
1522 static ssize_t format1_show(struct device *dev,
1523                 struct device_attribute *attr, char *buf)
1524 {
1525         u32 handle;
1526         ssize_t rc = -ENXIO;
1527         struct nfit_mem *nfit_mem;
1528         struct nfit_memdev *nfit_memdev;
1529         struct acpi_nfit_desc *acpi_desc;
1530         struct nvdimm *nvdimm = to_nvdimm(dev);
1531         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1532
1533         nfit_mem = nvdimm_provider_data(nvdimm);
1534         acpi_desc = nfit_mem->acpi_desc;
1535         handle = to_nfit_memdev(dev)->device_handle;
1536
1537         /* assumes DIMMs have at most 2 published interface codes */
1538         mutex_lock(&acpi_desc->init_mutex);
1539         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1540                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1541                 struct nfit_dcr *nfit_dcr;
1542
1543                 if (memdev->device_handle != handle)
1544                         continue;
1545
1546                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1547                         if (nfit_dcr->dcr->region_index != memdev->region_index)
1548                                 continue;
1549                         if (nfit_dcr->dcr->code == dcr->code)
1550                                 continue;
1551                         rc = sprintf(buf, "0x%04x\n",
1552                                         le16_to_cpu(nfit_dcr->dcr->code));
1553                         break;
1554                 }
1555                 if (rc != ENXIO)
1556                         break;
1557         }
1558         mutex_unlock(&acpi_desc->init_mutex);
1559         return rc;
1560 }
1561 static DEVICE_ATTR_RO(format1);
1562
1563 static ssize_t formats_show(struct device *dev,
1564                 struct device_attribute *attr, char *buf)
1565 {
1566         struct nvdimm *nvdimm = to_nvdimm(dev);
1567
1568         return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1569 }
1570 static DEVICE_ATTR_RO(formats);
1571
1572 static ssize_t serial_show(struct device *dev,
1573                 struct device_attribute *attr, char *buf)
1574 {
1575         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1576
1577         return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1578 }
1579 static DEVICE_ATTR_RO(serial);
1580
1581 static ssize_t family_show(struct device *dev,
1582                 struct device_attribute *attr, char *buf)
1583 {
1584         struct nvdimm *nvdimm = to_nvdimm(dev);
1585         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1586
1587         if (nfit_mem->family < 0)
1588                 return -ENXIO;
1589         return sprintf(buf, "%d\n", nfit_mem->family);
1590 }
1591 static DEVICE_ATTR_RO(family);
1592
1593 static ssize_t dsm_mask_show(struct device *dev,
1594                 struct device_attribute *attr, char *buf)
1595 {
1596         struct nvdimm *nvdimm = to_nvdimm(dev);
1597         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1598
1599         if (nfit_mem->family < 0)
1600                 return -ENXIO;
1601         return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1602 }
1603 static DEVICE_ATTR_RO(dsm_mask);
1604
1605 static ssize_t flags_show(struct device *dev,
1606                 struct device_attribute *attr, char *buf)
1607 {
1608         struct nvdimm *nvdimm = to_nvdimm(dev);
1609         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1610         u16 flags = __to_nfit_memdev(nfit_mem)->flags;
1611
1612         if (test_bit(NFIT_MEM_DIRTY, &nfit_mem->flags))
1613                 flags |= ACPI_NFIT_MEM_FLUSH_FAILED;
1614
1615         return sprintf(buf, "%s%s%s%s%s%s%s\n",
1616                 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1617                 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1618                 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1619                 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1620                 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1621                 flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1622                 flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1623 }
1624 static DEVICE_ATTR_RO(flags);
1625
1626 static ssize_t id_show(struct device *dev,
1627                 struct device_attribute *attr, char *buf)
1628 {
1629         struct nvdimm *nvdimm = to_nvdimm(dev);
1630         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1631
1632         return sprintf(buf, "%s\n", nfit_mem->id);
1633 }
1634 static DEVICE_ATTR_RO(id);
1635
1636 static ssize_t dirty_shutdown_show(struct device *dev,
1637                 struct device_attribute *attr, char *buf)
1638 {
1639         struct nvdimm *nvdimm = to_nvdimm(dev);
1640         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1641
1642         return sprintf(buf, "%d\n", nfit_mem->dirty_shutdown);
1643 }
1644 static DEVICE_ATTR_RO(dirty_shutdown);
1645
1646 static struct attribute *acpi_nfit_dimm_attributes[] = {
1647         &dev_attr_handle.attr,
1648         &dev_attr_phys_id.attr,
1649         &dev_attr_vendor.attr,
1650         &dev_attr_device.attr,
1651         &dev_attr_rev_id.attr,
1652         &dev_attr_subsystem_vendor.attr,
1653         &dev_attr_subsystem_device.attr,
1654         &dev_attr_subsystem_rev_id.attr,
1655         &dev_attr_format.attr,
1656         &dev_attr_formats.attr,
1657         &dev_attr_format1.attr,
1658         &dev_attr_serial.attr,
1659         &dev_attr_flags.attr,
1660         &dev_attr_id.attr,
1661         &dev_attr_family.attr,
1662         &dev_attr_dsm_mask.attr,
1663         &dev_attr_dirty_shutdown.attr,
1664         NULL,
1665 };
1666
1667 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1668                 struct attribute *a, int n)
1669 {
1670         struct device *dev = container_of(kobj, struct device, kobj);
1671         struct nvdimm *nvdimm = to_nvdimm(dev);
1672         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1673
1674         if (!to_nfit_dcr(dev)) {
1675                 /* Without a dcr only the memdev attributes can be surfaced */
1676                 if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1677                                 || a == &dev_attr_flags.attr
1678                                 || a == &dev_attr_family.attr
1679                                 || a == &dev_attr_dsm_mask.attr)
1680                         return a->mode;
1681                 return 0;
1682         }
1683
1684         if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1685                 return 0;
1686
1687         if (!test_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags)
1688                         && a == &dev_attr_dirty_shutdown.attr)
1689                 return 0;
1690
1691         return a->mode;
1692 }
1693
1694 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1695         .name = "nfit",
1696         .attrs = acpi_nfit_dimm_attributes,
1697         .is_visible = acpi_nfit_dimm_attr_visible,
1698 };
1699
1700 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1701         &acpi_nfit_dimm_attribute_group,
1702         NULL,
1703 };
1704
1705 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1706                 u32 device_handle)
1707 {
1708         struct nfit_mem *nfit_mem;
1709
1710         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1711                 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1712                         return nfit_mem->nvdimm;
1713
1714         return NULL;
1715 }
1716
1717 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1718 {
1719         struct nfit_mem *nfit_mem;
1720         struct acpi_nfit_desc *acpi_desc;
1721
1722         dev_dbg(dev->parent, "%s: event: %d\n", dev_name(dev),
1723                         event);
1724
1725         if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1726                 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1727                                 event);
1728                 return;
1729         }
1730
1731         acpi_desc = dev_get_drvdata(dev->parent);
1732         if (!acpi_desc)
1733                 return;
1734
1735         /*
1736          * If we successfully retrieved acpi_desc, then we know nfit_mem data
1737          * is still valid.
1738          */
1739         nfit_mem = dev_get_drvdata(dev);
1740         if (nfit_mem && nfit_mem->flags_attr)
1741                 sysfs_notify_dirent(nfit_mem->flags_attr);
1742 }
1743 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1744
1745 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1746 {
1747         struct acpi_device *adev = data;
1748         struct device *dev = &adev->dev;
1749
1750         nfit_device_lock(dev->parent);
1751         __acpi_nvdimm_notify(dev, event);
1752         nfit_device_unlock(dev->parent);
1753 }
1754
1755 static bool acpi_nvdimm_has_method(struct acpi_device *adev, char *method)
1756 {
1757         acpi_handle handle;
1758         acpi_status status;
1759
1760         status = acpi_get_handle(adev->handle, method, &handle);
1761
1762         if (ACPI_SUCCESS(status))
1763                 return true;
1764         return false;
1765 }
1766
1767 __weak void nfit_intel_shutdown_status(struct nfit_mem *nfit_mem)
1768 {
1769         struct device *dev = &nfit_mem->adev->dev;
1770         struct nd_intel_smart smart = { 0 };
1771         union acpi_object in_buf = {
1772                 .buffer.type = ACPI_TYPE_BUFFER,
1773                 .buffer.length = 0,
1774         };
1775         union acpi_object in_obj = {
1776                 .package.type = ACPI_TYPE_PACKAGE,
1777                 .package.count = 1,
1778                 .package.elements = &in_buf,
1779         };
1780         const u8 func = ND_INTEL_SMART;
1781         const guid_t *guid = to_nfit_uuid(nfit_mem->family);
1782         u8 revid = nfit_dsm_revid(nfit_mem->family, func);
1783         struct acpi_device *adev = nfit_mem->adev;
1784         acpi_handle handle = adev->handle;
1785         union acpi_object *out_obj;
1786
1787         if ((nfit_mem->dsm_mask & (1 << func)) == 0)
1788                 return;
1789
1790         out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
1791         if (!out_obj || out_obj->type != ACPI_TYPE_BUFFER
1792                         || out_obj->buffer.length < sizeof(smart)) {
1793                 dev_dbg(dev->parent, "%s: failed to retrieve initial health\n",
1794                                 dev_name(dev));
1795                 ACPI_FREE(out_obj);
1796                 return;
1797         }
1798         memcpy(&smart, out_obj->buffer.pointer, sizeof(smart));
1799         ACPI_FREE(out_obj);
1800
1801         if (smart.flags & ND_INTEL_SMART_SHUTDOWN_VALID) {
1802                 if (smart.shutdown_state)
1803                         set_bit(NFIT_MEM_DIRTY, &nfit_mem->flags);
1804         }
1805
1806         if (smart.flags & ND_INTEL_SMART_SHUTDOWN_COUNT_VALID) {
1807                 set_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags);
1808                 nfit_mem->dirty_shutdown = smart.shutdown_count;
1809         }
1810 }
1811
1812 static void populate_shutdown_status(struct nfit_mem *nfit_mem)
1813 {
1814         /*
1815          * For DIMMs that provide a dynamic facility to retrieve a
1816          * dirty-shutdown status and/or a dirty-shutdown count, cache
1817          * these values in nfit_mem.
1818          */
1819         if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
1820                 nfit_intel_shutdown_status(nfit_mem);
1821 }
1822
1823 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1824                 struct nfit_mem *nfit_mem, u32 device_handle)
1825 {
1826         struct acpi_device *adev, *adev_dimm;
1827         struct device *dev = acpi_desc->dev;
1828         unsigned long dsm_mask, label_mask;
1829         const guid_t *guid;
1830         int i;
1831         int family = -1;
1832         struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
1833
1834         /* nfit test assumes 1:1 relationship between commands and dsms */
1835         nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1836         nfit_mem->family = NVDIMM_FAMILY_INTEL;
1837
1838         if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1839                 sprintf(nfit_mem->id, "%04x-%02x-%04x-%08x",
1840                                 be16_to_cpu(dcr->vendor_id),
1841                                 dcr->manufacturing_location,
1842                                 be16_to_cpu(dcr->manufacturing_date),
1843                                 be32_to_cpu(dcr->serial_number));
1844         else
1845                 sprintf(nfit_mem->id, "%04x-%08x",
1846                                 be16_to_cpu(dcr->vendor_id),
1847                                 be32_to_cpu(dcr->serial_number));
1848
1849         adev = to_acpi_dev(acpi_desc);
1850         if (!adev) {
1851                 /* unit test case */
1852                 populate_shutdown_status(nfit_mem);
1853                 return 0;
1854         }
1855
1856         adev_dimm = acpi_find_child_device(adev, device_handle, false);
1857         nfit_mem->adev = adev_dimm;
1858         if (!adev_dimm) {
1859                 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1860                                 device_handle);
1861                 return force_enable_dimms ? 0 : -ENODEV;
1862         }
1863
1864         if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1865                 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1866                 dev_err(dev, "%s: notification registration failed\n",
1867                                 dev_name(&adev_dimm->dev));
1868                 return -ENXIO;
1869         }
1870         /*
1871          * Record nfit_mem for the notification path to track back to
1872          * the nfit sysfs attributes for this dimm device object.
1873          */
1874         dev_set_drvdata(&adev_dimm->dev, nfit_mem);
1875
1876         /*
1877          * There are 4 "legacy" NVDIMM command sets
1878          * (NVDIMM_FAMILY_{INTEL,MSFT,HPE1,HPE2}) that were created before
1879          * an EFI working group was established to constrain this
1880          * proliferation. The nfit driver probes for the supported command
1881          * set by GUID. Note, if you're a platform developer looking to add
1882          * a new command set to this probe, consider using an existing set,
1883          * or otherwise seek approval to publish the command set at
1884          * http://www.uefi.org/RFIC_LIST.
1885          *
1886          * Note, that checking for function0 (bit0) tells us if any commands
1887          * are reachable through this GUID.
1888          */
1889         for (i = 0; i <= NVDIMM_FAMILY_MAX; i++)
1890                 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1891                         if (family < 0 || i == default_dsm_family)
1892                                 family = i;
1893
1894         /* limit the supported commands to those that are publicly documented */
1895         nfit_mem->family = family;
1896         if (override_dsm_mask && !disable_vendor_specific)
1897                 dsm_mask = override_dsm_mask;
1898         else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1899                 dsm_mask = NVDIMM_INTEL_CMDMASK;
1900                 if (disable_vendor_specific)
1901                         dsm_mask &= ~(1 << ND_CMD_VENDOR);
1902         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1903                 dsm_mask = 0x1c3c76;
1904         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1905                 dsm_mask = 0x1fe;
1906                 if (disable_vendor_specific)
1907                         dsm_mask &= ~(1 << 8);
1908         } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1909                 dsm_mask = 0xffffffff;
1910         } else if (nfit_mem->family == NVDIMM_FAMILY_HYPERV) {
1911                 dsm_mask = 0x1f;
1912         } else {
1913                 dev_dbg(dev, "unknown dimm command family\n");
1914                 nfit_mem->family = -1;
1915                 /* DSMs are optional, continue loading the driver... */
1916                 return 0;
1917         }
1918
1919         /*
1920          * Function 0 is the command interrogation function, don't
1921          * export it to potential userspace use, and enable it to be
1922          * used as an error value in acpi_nfit_ctl().
1923          */
1924         dsm_mask &= ~1UL;
1925
1926         guid = to_nfit_uuid(nfit_mem->family);
1927         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1928                 if (acpi_check_dsm(adev_dimm->handle, guid,
1929                                         nfit_dsm_revid(nfit_mem->family, i),
1930                                         1ULL << i))
1931                         set_bit(i, &nfit_mem->dsm_mask);
1932
1933         /*
1934          * Prefer the NVDIMM_FAMILY_INTEL label read commands if present
1935          * due to their better semantics handling locked capacity.
1936          */
1937         label_mask = 1 << ND_CMD_GET_CONFIG_SIZE | 1 << ND_CMD_GET_CONFIG_DATA
1938                 | 1 << ND_CMD_SET_CONFIG_DATA;
1939         if (family == NVDIMM_FAMILY_INTEL
1940                         && (dsm_mask & label_mask) == label_mask)
1941                 /* skip _LS{I,R,W} enabling */;
1942         else {
1943                 if (acpi_nvdimm_has_method(adev_dimm, "_LSI")
1944                                 && acpi_nvdimm_has_method(adev_dimm, "_LSR")) {
1945                         dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev));
1946                         set_bit(NFIT_MEM_LSR, &nfit_mem->flags);
1947                 }
1948
1949                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
1950                                 && acpi_nvdimm_has_method(adev_dimm, "_LSW")) {
1951                         dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev));
1952                         set_bit(NFIT_MEM_LSW, &nfit_mem->flags);
1953                 }
1954
1955                 /*
1956                  * Quirk read-only label configurations to preserve
1957                  * access to label-less namespaces by default.
1958                  */
1959                 if (!test_bit(NFIT_MEM_LSW, &nfit_mem->flags)
1960                                 && !force_labels) {
1961                         dev_dbg(dev, "%s: No _LSW, disable labels\n",
1962                                         dev_name(&adev_dimm->dev));
1963                         clear_bit(NFIT_MEM_LSR, &nfit_mem->flags);
1964                 } else
1965                         dev_dbg(dev, "%s: Force enable labels\n",
1966                                         dev_name(&adev_dimm->dev));
1967         }
1968
1969         populate_shutdown_status(nfit_mem);
1970
1971         return 0;
1972 }
1973
1974 static void shutdown_dimm_notify(void *data)
1975 {
1976         struct acpi_nfit_desc *acpi_desc = data;
1977         struct nfit_mem *nfit_mem;
1978
1979         mutex_lock(&acpi_desc->init_mutex);
1980         /*
1981          * Clear out the nfit_mem->flags_attr and shut down dimm event
1982          * notifications.
1983          */
1984         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1985                 struct acpi_device *adev_dimm = nfit_mem->adev;
1986
1987                 if (nfit_mem->flags_attr) {
1988                         sysfs_put(nfit_mem->flags_attr);
1989                         nfit_mem->flags_attr = NULL;
1990                 }
1991                 if (adev_dimm) {
1992                         acpi_remove_notify_handler(adev_dimm->handle,
1993                                         ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
1994                         dev_set_drvdata(&adev_dimm->dev, NULL);
1995                 }
1996         }
1997         mutex_unlock(&acpi_desc->init_mutex);
1998 }
1999
2000 static const struct nvdimm_security_ops *acpi_nfit_get_security_ops(int family)
2001 {
2002         switch (family) {
2003         case NVDIMM_FAMILY_INTEL:
2004                 return intel_security_ops;
2005         default:
2006                 return NULL;
2007         }
2008 }
2009
2010 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
2011 {
2012         struct nfit_mem *nfit_mem;
2013         int dimm_count = 0, rc;
2014         struct nvdimm *nvdimm;
2015
2016         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2017                 struct acpi_nfit_flush_address *flush;
2018                 unsigned long flags = 0, cmd_mask;
2019                 struct nfit_memdev *nfit_memdev;
2020                 u32 device_handle;
2021                 u16 mem_flags;
2022
2023                 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
2024                 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
2025                 if (nvdimm) {
2026                         dimm_count++;
2027                         continue;
2028                 }
2029
2030                 if (nfit_mem->bdw && nfit_mem->memdev_pmem) {
2031                         set_bit(NDD_ALIASING, &flags);
2032                         set_bit(NDD_LABELING, &flags);
2033                 }
2034
2035                 /* collate flags across all memdevs for this dimm */
2036                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2037                         struct acpi_nfit_memory_map *dimm_memdev;
2038
2039                         dimm_memdev = __to_nfit_memdev(nfit_mem);
2040                         if (dimm_memdev->device_handle
2041                                         != nfit_memdev->memdev->device_handle)
2042                                 continue;
2043                         dimm_memdev->flags |= nfit_memdev->memdev->flags;
2044                 }
2045
2046                 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
2047                 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
2048                         set_bit(NDD_UNARMED, &flags);
2049
2050                 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
2051                 if (rc)
2052                         continue;
2053
2054                 /*
2055                  * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
2056                  * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
2057                  * userspace interface.
2058                  */
2059                 cmd_mask = 1UL << ND_CMD_CALL;
2060                 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
2061                         /*
2062                          * These commands have a 1:1 correspondence
2063                          * between DSM payload and libnvdimm ioctl
2064                          * payload format.
2065                          */
2066                         cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
2067                 }
2068
2069                 /* Quirk to ignore LOCAL for labels on HYPERV DIMMs */
2070                 if (nfit_mem->family == NVDIMM_FAMILY_HYPERV)
2071                         set_bit(NDD_NOBLK, &flags);
2072
2073                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
2074                         set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
2075                         set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
2076                 }
2077                 if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags))
2078                         set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
2079
2080                 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
2081                         : NULL;
2082                 nvdimm = __nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
2083                                 acpi_nfit_dimm_attribute_groups,
2084                                 flags, cmd_mask, flush ? flush->hint_count : 0,
2085                                 nfit_mem->flush_wpq, &nfit_mem->id[0],
2086                                 acpi_nfit_get_security_ops(nfit_mem->family));
2087                 if (!nvdimm)
2088                         return -ENOMEM;
2089
2090                 nfit_mem->nvdimm = nvdimm;
2091                 dimm_count++;
2092
2093                 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
2094                         continue;
2095
2096                 dev_err(acpi_desc->dev, "Error found in NVDIMM %s flags:%s%s%s%s%s\n",
2097                                 nvdimm_name(nvdimm),
2098                   mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
2099                   mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
2100                   mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
2101                   mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
2102                   mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
2103
2104         }
2105
2106         rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
2107         if (rc)
2108                 return rc;
2109
2110         /*
2111          * Now that dimms are successfully registered, and async registration
2112          * is flushed, attempt to enable event notification.
2113          */
2114         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2115                 struct kernfs_node *nfit_kernfs;
2116
2117                 nvdimm = nfit_mem->nvdimm;
2118                 if (!nvdimm)
2119                         continue;
2120
2121                 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
2122                 if (nfit_kernfs)
2123                         nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
2124                                         "flags");
2125                 sysfs_put(nfit_kernfs);
2126                 if (!nfit_mem->flags_attr)
2127                         dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
2128                                         nvdimm_name(nvdimm));
2129         }
2130
2131         return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
2132                         acpi_desc);
2133 }
2134
2135 /*
2136  * These constants are private because there are no kernel consumers of
2137  * these commands.
2138  */
2139 enum nfit_aux_cmds {
2140         NFIT_CMD_TRANSLATE_SPA = 5,
2141         NFIT_CMD_ARS_INJECT_SET = 7,
2142         NFIT_CMD_ARS_INJECT_CLEAR = 8,
2143         NFIT_CMD_ARS_INJECT_GET = 9,
2144 };
2145
2146 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
2147 {
2148         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2149         const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
2150         struct acpi_device *adev;
2151         unsigned long dsm_mask;
2152         int i;
2153
2154         nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
2155         nd_desc->bus_dsm_mask = acpi_desc->bus_nfit_cmd_force_en;
2156         adev = to_acpi_dev(acpi_desc);
2157         if (!adev)
2158                 return;
2159
2160         for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
2161                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2162                         set_bit(i, &nd_desc->cmd_mask);
2163         set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
2164
2165         dsm_mask =
2166                 (1 << ND_CMD_ARS_CAP) |
2167                 (1 << ND_CMD_ARS_START) |
2168                 (1 << ND_CMD_ARS_STATUS) |
2169                 (1 << ND_CMD_CLEAR_ERROR) |
2170                 (1 << NFIT_CMD_TRANSLATE_SPA) |
2171                 (1 << NFIT_CMD_ARS_INJECT_SET) |
2172                 (1 << NFIT_CMD_ARS_INJECT_CLEAR) |
2173                 (1 << NFIT_CMD_ARS_INJECT_GET);
2174         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
2175                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2176                         set_bit(i, &nd_desc->bus_dsm_mask);
2177 }
2178
2179 static ssize_t range_index_show(struct device *dev,
2180                 struct device_attribute *attr, char *buf)
2181 {
2182         struct nd_region *nd_region = to_nd_region(dev);
2183         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
2184
2185         return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
2186 }
2187 static DEVICE_ATTR_RO(range_index);
2188
2189 static struct attribute *acpi_nfit_region_attributes[] = {
2190         &dev_attr_range_index.attr,
2191         NULL,
2192 };
2193
2194 static const struct attribute_group acpi_nfit_region_attribute_group = {
2195         .name = "nfit",
2196         .attrs = acpi_nfit_region_attributes,
2197 };
2198
2199 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
2200         &acpi_nfit_region_attribute_group,
2201         NULL,
2202 };
2203
2204 /* enough info to uniquely specify an interleave set */
2205 struct nfit_set_info {
2206         struct nfit_set_info_map {
2207                 u64 region_offset;
2208                 u32 serial_number;
2209                 u32 pad;
2210         } mapping[0];
2211 };
2212
2213 struct nfit_set_info2 {
2214         struct nfit_set_info_map2 {
2215                 u64 region_offset;
2216                 u32 serial_number;
2217                 u16 vendor_id;
2218                 u16 manufacturing_date;
2219                 u8  manufacturing_location;
2220                 u8  reserved[31];
2221         } mapping[0];
2222 };
2223
2224 static size_t sizeof_nfit_set_info(int num_mappings)
2225 {
2226         return sizeof(struct nfit_set_info)
2227                 + num_mappings * sizeof(struct nfit_set_info_map);
2228 }
2229
2230 static size_t sizeof_nfit_set_info2(int num_mappings)
2231 {
2232         return sizeof(struct nfit_set_info2)
2233                 + num_mappings * sizeof(struct nfit_set_info_map2);
2234 }
2235
2236 static int cmp_map_compat(const void *m0, const void *m1)
2237 {
2238         const struct nfit_set_info_map *map0 = m0;
2239         const struct nfit_set_info_map *map1 = m1;
2240
2241         return memcmp(&map0->region_offset, &map1->region_offset,
2242                         sizeof(u64));
2243 }
2244
2245 static int cmp_map(const void *m0, const void *m1)
2246 {
2247         const struct nfit_set_info_map *map0 = m0;
2248         const struct nfit_set_info_map *map1 = m1;
2249
2250         if (map0->region_offset < map1->region_offset)
2251                 return -1;
2252         else if (map0->region_offset > map1->region_offset)
2253                 return 1;
2254         return 0;
2255 }
2256
2257 static int cmp_map2(const void *m0, const void *m1)
2258 {
2259         const struct nfit_set_info_map2 *map0 = m0;
2260         const struct nfit_set_info_map2 *map1 = m1;
2261
2262         if (map0->region_offset < map1->region_offset)
2263                 return -1;
2264         else if (map0->region_offset > map1->region_offset)
2265                 return 1;
2266         return 0;
2267 }
2268
2269 /* Retrieve the nth entry referencing this spa */
2270 static struct acpi_nfit_memory_map *memdev_from_spa(
2271                 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
2272 {
2273         struct nfit_memdev *nfit_memdev;
2274
2275         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
2276                 if (nfit_memdev->memdev->range_index == range_index)
2277                         if (n-- == 0)
2278                                 return nfit_memdev->memdev;
2279         return NULL;
2280 }
2281
2282 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
2283                 struct nd_region_desc *ndr_desc,
2284                 struct acpi_nfit_system_address *spa)
2285 {
2286         struct device *dev = acpi_desc->dev;
2287         struct nd_interleave_set *nd_set;
2288         u16 nr = ndr_desc->num_mappings;
2289         struct nfit_set_info2 *info2;
2290         struct nfit_set_info *info;
2291         int i;
2292
2293         nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
2294         if (!nd_set)
2295                 return -ENOMEM;
2296         import_guid(&nd_set->type_guid, spa->range_guid);
2297
2298         info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
2299         if (!info)
2300                 return -ENOMEM;
2301
2302         info2 = devm_kzalloc(dev, sizeof_nfit_set_info2(nr), GFP_KERNEL);
2303         if (!info2)
2304                 return -ENOMEM;
2305
2306         for (i = 0; i < nr; i++) {
2307                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
2308                 struct nfit_set_info_map *map = &info->mapping[i];
2309                 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2310                 struct nvdimm *nvdimm = mapping->nvdimm;
2311                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2312                 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
2313                                 spa->range_index, i);
2314                 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2315
2316                 if (!memdev || !nfit_mem->dcr) {
2317                         dev_err(dev, "%s: failed to find DCR\n", __func__);
2318                         return -ENODEV;
2319                 }
2320
2321                 map->region_offset = memdev->region_offset;
2322                 map->serial_number = dcr->serial_number;
2323
2324                 map2->region_offset = memdev->region_offset;
2325                 map2->serial_number = dcr->serial_number;
2326                 map2->vendor_id = dcr->vendor_id;
2327                 map2->manufacturing_date = dcr->manufacturing_date;
2328                 map2->manufacturing_location = dcr->manufacturing_location;
2329         }
2330
2331         /* v1.1 namespaces */
2332         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2333                         cmp_map, NULL);
2334         nd_set->cookie1 = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2335
2336         /* v1.2 namespaces */
2337         sort(&info2->mapping[0], nr, sizeof(struct nfit_set_info_map2),
2338                         cmp_map2, NULL);
2339         nd_set->cookie2 = nd_fletcher64(info2, sizeof_nfit_set_info2(nr), 0);
2340
2341         /* support v1.1 namespaces created with the wrong sort order */
2342         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2343                         cmp_map_compat, NULL);
2344         nd_set->altcookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2345
2346         /* record the result of the sort for the mapping position */
2347         for (i = 0; i < nr; i++) {
2348                 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2349                 int j;
2350
2351                 for (j = 0; j < nr; j++) {
2352                         struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
2353                         struct nvdimm *nvdimm = mapping->nvdimm;
2354                         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2355                         struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2356
2357                         if (map2->serial_number == dcr->serial_number &&
2358                             map2->vendor_id == dcr->vendor_id &&
2359                             map2->manufacturing_date == dcr->manufacturing_date &&
2360                             map2->manufacturing_location
2361                                     == dcr->manufacturing_location) {
2362                                 mapping->position = i;
2363                                 break;
2364                         }
2365                 }
2366         }
2367
2368         ndr_desc->nd_set = nd_set;
2369         devm_kfree(dev, info);
2370         devm_kfree(dev, info2);
2371
2372         return 0;
2373 }
2374
2375 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
2376 {
2377         struct acpi_nfit_interleave *idt = mmio->idt;
2378         u32 sub_line_offset, line_index, line_offset;
2379         u64 line_no, table_skip_count, table_offset;
2380
2381         line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
2382         table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
2383         line_offset = idt->line_offset[line_index]
2384                 * mmio->line_size;
2385         table_offset = table_skip_count * mmio->table_size;
2386
2387         return mmio->base_offset + line_offset + table_offset + sub_line_offset;
2388 }
2389
2390 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
2391 {
2392         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2393         u64 offset = nfit_blk->stat_offset + mmio->size * bw;
2394         const u32 STATUS_MASK = 0x80000037;
2395
2396         if (mmio->num_lines)
2397                 offset = to_interleave_offset(offset, mmio);
2398
2399         return readl(mmio->addr.base + offset) & STATUS_MASK;
2400 }
2401
2402 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
2403                 resource_size_t dpa, unsigned int len, unsigned int write)
2404 {
2405         u64 cmd, offset;
2406         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2407
2408         enum {
2409                 BCW_OFFSET_MASK = (1ULL << 48)-1,
2410                 BCW_LEN_SHIFT = 48,
2411                 BCW_LEN_MASK = (1ULL << 8) - 1,
2412                 BCW_CMD_SHIFT = 56,
2413         };
2414
2415         cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
2416         len = len >> L1_CACHE_SHIFT;
2417         cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
2418         cmd |= ((u64) write) << BCW_CMD_SHIFT;
2419
2420         offset = nfit_blk->cmd_offset + mmio->size * bw;
2421         if (mmio->num_lines)
2422                 offset = to_interleave_offset(offset, mmio);
2423
2424         writeq(cmd, mmio->addr.base + offset);
2425         nvdimm_flush(nfit_blk->nd_region, NULL);
2426
2427         if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
2428                 readq(mmio->addr.base + offset);
2429 }
2430
2431 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
2432                 resource_size_t dpa, void *iobuf, size_t len, int rw,
2433                 unsigned int lane)
2434 {
2435         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2436         unsigned int copied = 0;
2437         u64 base_offset;
2438         int rc;
2439
2440         base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
2441                 + lane * mmio->size;
2442         write_blk_ctl(nfit_blk, lane, dpa, len, rw);
2443         while (len) {
2444                 unsigned int c;
2445                 u64 offset;
2446
2447                 if (mmio->num_lines) {
2448                         u32 line_offset;
2449
2450                         offset = to_interleave_offset(base_offset + copied,
2451                                         mmio);
2452                         div_u64_rem(offset, mmio->line_size, &line_offset);
2453                         c = min_t(size_t, len, mmio->line_size - line_offset);
2454                 } else {
2455                         offset = base_offset + nfit_blk->bdw_offset;
2456                         c = len;
2457                 }
2458
2459                 if (rw)
2460                         memcpy_flushcache(mmio->addr.aperture + offset, iobuf + copied, c);
2461                 else {
2462                         if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
2463                                 arch_invalidate_pmem((void __force *)
2464                                         mmio->addr.aperture + offset, c);
2465
2466                         memcpy(iobuf + copied, mmio->addr.aperture + offset, c);
2467                 }
2468
2469                 copied += c;
2470                 len -= c;
2471         }
2472
2473         if (rw)
2474                 nvdimm_flush(nfit_blk->nd_region, NULL);
2475
2476         rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
2477         return rc;
2478 }
2479
2480 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
2481                 resource_size_t dpa, void *iobuf, u64 len, int rw)
2482 {
2483         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
2484         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2485         struct nd_region *nd_region = nfit_blk->nd_region;
2486         unsigned int lane, copied = 0;
2487         int rc = 0;
2488
2489         lane = nd_region_acquire_lane(nd_region);
2490         while (len) {
2491                 u64 c = min(len, mmio->size);
2492
2493                 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
2494                                 iobuf + copied, c, rw, lane);
2495                 if (rc)
2496                         break;
2497
2498                 copied += c;
2499                 len -= c;
2500         }
2501         nd_region_release_lane(nd_region, lane);
2502
2503         return rc;
2504 }
2505
2506 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
2507                 struct acpi_nfit_interleave *idt, u16 interleave_ways)
2508 {
2509         if (idt) {
2510                 mmio->num_lines = idt->line_count;
2511                 mmio->line_size = idt->line_size;
2512                 if (interleave_ways == 0)
2513                         return -ENXIO;
2514                 mmio->table_size = mmio->num_lines * interleave_ways
2515                         * mmio->line_size;
2516         }
2517
2518         return 0;
2519 }
2520
2521 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
2522                 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
2523 {
2524         struct nd_cmd_dimm_flags flags;
2525         int rc;
2526
2527         memset(&flags, 0, sizeof(flags));
2528         rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
2529                         sizeof(flags), NULL);
2530
2531         if (rc >= 0 && flags.status == 0)
2532                 nfit_blk->dimm_flags = flags.flags;
2533         else if (rc == -ENOTTY) {
2534                 /* fall back to a conservative default */
2535                 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
2536                 rc = 0;
2537         } else
2538                 rc = -ENXIO;
2539
2540         return rc;
2541 }
2542
2543 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
2544                 struct device *dev)
2545 {
2546         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
2547         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
2548         struct nfit_blk_mmio *mmio;
2549         struct nfit_blk *nfit_blk;
2550         struct nfit_mem *nfit_mem;
2551         struct nvdimm *nvdimm;
2552         int rc;
2553
2554         nvdimm = nd_blk_region_to_dimm(ndbr);
2555         nfit_mem = nvdimm_provider_data(nvdimm);
2556         if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
2557                 dev_dbg(dev, "missing%s%s%s\n",
2558                                 nfit_mem ? "" : " nfit_mem",
2559                                 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
2560                                 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
2561                 return -ENXIO;
2562         }
2563
2564         nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
2565         if (!nfit_blk)
2566                 return -ENOMEM;
2567         nd_blk_region_set_provider_data(ndbr, nfit_blk);
2568         nfit_blk->nd_region = to_nd_region(dev);
2569
2570         /* map block aperture memory */
2571         nfit_blk->bdw_offset = nfit_mem->bdw->offset;
2572         mmio = &nfit_blk->mmio[BDW];
2573         mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
2574                         nfit_mem->spa_bdw->length, nd_blk_memremap_flags(ndbr));
2575         if (!mmio->addr.base) {
2576                 dev_dbg(dev, "%s failed to map bdw\n",
2577                                 nvdimm_name(nvdimm));
2578                 return -ENOMEM;
2579         }
2580         mmio->size = nfit_mem->bdw->size;
2581         mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
2582         mmio->idt = nfit_mem->idt_bdw;
2583         mmio->spa = nfit_mem->spa_bdw;
2584         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
2585                         nfit_mem->memdev_bdw->interleave_ways);
2586         if (rc) {
2587                 dev_dbg(dev, "%s failed to init bdw interleave\n",
2588                                 nvdimm_name(nvdimm));
2589                 return rc;
2590         }
2591
2592         /* map block control memory */
2593         nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
2594         nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
2595         mmio = &nfit_blk->mmio[DCR];
2596         mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
2597                         nfit_mem->spa_dcr->length);
2598         if (!mmio->addr.base) {
2599                 dev_dbg(dev, "%s failed to map dcr\n",
2600                                 nvdimm_name(nvdimm));
2601                 return -ENOMEM;
2602         }
2603         mmio->size = nfit_mem->dcr->window_size;
2604         mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
2605         mmio->idt = nfit_mem->idt_dcr;
2606         mmio->spa = nfit_mem->spa_dcr;
2607         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
2608                         nfit_mem->memdev_dcr->interleave_ways);
2609         if (rc) {
2610                 dev_dbg(dev, "%s failed to init dcr interleave\n",
2611                                 nvdimm_name(nvdimm));
2612                 return rc;
2613         }
2614
2615         rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
2616         if (rc < 0) {
2617                 dev_dbg(dev, "%s failed get DIMM flags\n",
2618                                 nvdimm_name(nvdimm));
2619                 return rc;
2620         }
2621
2622         if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
2623                 dev_warn(dev, "unable to guarantee persistence of writes\n");
2624
2625         if (mmio->line_size == 0)
2626                 return 0;
2627
2628         if ((u32) nfit_blk->cmd_offset % mmio->line_size
2629                         + 8 > mmio->line_size) {
2630                 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
2631                 return -ENXIO;
2632         } else if ((u32) nfit_blk->stat_offset % mmio->line_size
2633                         + 8 > mmio->line_size) {
2634                 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
2635                 return -ENXIO;
2636         }
2637
2638         return 0;
2639 }
2640
2641 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2642                 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2643 {
2644         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2645         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2646         int cmd_rc, rc;
2647
2648         cmd->address = spa->address;
2649         cmd->length = spa->length;
2650         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2651                         sizeof(*cmd), &cmd_rc);
2652         if (rc < 0)
2653                 return rc;
2654         return cmd_rc;
2655 }
2656
2657 static int ars_start(struct acpi_nfit_desc *acpi_desc,
2658                 struct nfit_spa *nfit_spa, enum nfit_ars_state req_type)
2659 {
2660         int rc;
2661         int cmd_rc;
2662         struct nd_cmd_ars_start ars_start;
2663         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2664         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2665
2666         memset(&ars_start, 0, sizeof(ars_start));
2667         ars_start.address = spa->address;
2668         ars_start.length = spa->length;
2669         if (req_type == ARS_REQ_SHORT)
2670                 ars_start.flags = ND_ARS_RETURN_PREV_DATA;
2671         if (nfit_spa_type(spa) == NFIT_SPA_PM)
2672                 ars_start.type = ND_ARS_PERSISTENT;
2673         else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2674                 ars_start.type = ND_ARS_VOLATILE;
2675         else
2676                 return -ENOTTY;
2677
2678         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2679                         sizeof(ars_start), &cmd_rc);
2680
2681         if (rc < 0)
2682                 return rc;
2683         if (cmd_rc < 0)
2684                 return cmd_rc;
2685         set_bit(ARS_VALID, &acpi_desc->scrub_flags);
2686         return 0;
2687 }
2688
2689 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2690 {
2691         int rc, cmd_rc;
2692         struct nd_cmd_ars_start ars_start;
2693         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2694         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2695
2696         ars_start = (struct nd_cmd_ars_start) {
2697                 .address = ars_status->restart_address,
2698                 .length = ars_status->restart_length,
2699                 .type = ars_status->type,
2700         };
2701         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2702                         sizeof(ars_start), &cmd_rc);
2703         if (rc < 0)
2704                 return rc;
2705         return cmd_rc;
2706 }
2707
2708 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2709 {
2710         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2711         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2712         int rc, cmd_rc;
2713
2714         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2715                         acpi_desc->max_ars, &cmd_rc);
2716         if (rc < 0)
2717                 return rc;
2718         return cmd_rc;
2719 }
2720
2721 static void ars_complete(struct acpi_nfit_desc *acpi_desc,
2722                 struct nfit_spa *nfit_spa)
2723 {
2724         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2725         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2726         struct nd_region *nd_region = nfit_spa->nd_region;
2727         struct device *dev;
2728
2729         lockdep_assert_held(&acpi_desc->init_mutex);
2730         /*
2731          * Only advance the ARS state for ARS runs initiated by the
2732          * kernel, ignore ARS results from BIOS initiated runs for scrub
2733          * completion tracking.
2734          */
2735         if (acpi_desc->scrub_spa != nfit_spa)
2736                 return;
2737
2738         if ((ars_status->address >= spa->address && ars_status->address
2739                                 < spa->address + spa->length)
2740                         || (ars_status->address < spa->address)) {
2741                 /*
2742                  * Assume that if a scrub starts at an offset from the
2743                  * start of nfit_spa that we are in the continuation
2744                  * case.
2745                  *
2746                  * Otherwise, if the scrub covers the spa range, mark
2747                  * any pending request complete.
2748                  */
2749                 if (ars_status->address + ars_status->length
2750                                 >= spa->address + spa->length)
2751                                 /* complete */;
2752                 else
2753                         return;
2754         } else
2755                 return;
2756
2757         acpi_desc->scrub_spa = NULL;
2758         if (nd_region) {
2759                 dev = nd_region_dev(nd_region);
2760                 nvdimm_region_notify(nd_region, NVDIMM_REVALIDATE_POISON);
2761         } else
2762                 dev = acpi_desc->dev;
2763         dev_dbg(dev, "ARS: range %d complete\n", spa->range_index);
2764 }
2765
2766 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc)
2767 {
2768         struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2769         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2770         int rc;
2771         u32 i;
2772
2773         /*
2774          * First record starts at 44 byte offset from the start of the
2775          * payload.
2776          */
2777         if (ars_status->out_length < 44)
2778                 return 0;
2779
2780         /*
2781          * Ignore potentially stale results that are only refreshed
2782          * after a start-ARS event.
2783          */
2784         if (!test_and_clear_bit(ARS_VALID, &acpi_desc->scrub_flags)) {
2785                 dev_dbg(acpi_desc->dev, "skip %d stale records\n",
2786                                 ars_status->num_records);
2787                 return 0;
2788         }
2789
2790         for (i = 0; i < ars_status->num_records; i++) {
2791                 /* only process full records */
2792                 if (ars_status->out_length
2793                                 < 44 + sizeof(struct nd_ars_record) * (i + 1))
2794                         break;
2795                 rc = nvdimm_bus_add_badrange(nvdimm_bus,
2796                                 ars_status->records[i].err_address,
2797                                 ars_status->records[i].length);
2798                 if (rc)
2799                         return rc;
2800         }
2801         if (i < ars_status->num_records)
2802                 dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2803
2804         return 0;
2805 }
2806
2807 static void acpi_nfit_remove_resource(void *data)
2808 {
2809         struct resource *res = data;
2810
2811         remove_resource(res);
2812 }
2813
2814 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2815                 struct nd_region_desc *ndr_desc)
2816 {
2817         struct resource *res, *nd_res = ndr_desc->res;
2818         int is_pmem, ret;
2819
2820         /* No operation if the region is already registered as PMEM */
2821         is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2822                                 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2823         if (is_pmem == REGION_INTERSECTS)
2824                 return 0;
2825
2826         res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2827         if (!res)
2828                 return -ENOMEM;
2829
2830         res->name = "Persistent Memory";
2831         res->start = nd_res->start;
2832         res->end = nd_res->end;
2833         res->flags = IORESOURCE_MEM;
2834         res->desc = IORES_DESC_PERSISTENT_MEMORY;
2835
2836         ret = insert_resource(&iomem_resource, res);
2837         if (ret)
2838                 return ret;
2839
2840         ret = devm_add_action_or_reset(acpi_desc->dev,
2841                                         acpi_nfit_remove_resource,
2842                                         res);
2843         if (ret)
2844                 return ret;
2845
2846         return 0;
2847 }
2848
2849 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2850                 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2851                 struct acpi_nfit_memory_map *memdev,
2852                 struct nfit_spa *nfit_spa)
2853 {
2854         struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2855                         memdev->device_handle);
2856         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2857         struct nd_blk_region_desc *ndbr_desc;
2858         struct nfit_mem *nfit_mem;
2859         int rc;
2860
2861         if (!nvdimm) {
2862                 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2863                                 spa->range_index, memdev->device_handle);
2864                 return -ENODEV;
2865         }
2866
2867         mapping->nvdimm = nvdimm;
2868         switch (nfit_spa_type(spa)) {
2869         case NFIT_SPA_PM:
2870         case NFIT_SPA_VOLATILE:
2871                 mapping->start = memdev->address;
2872                 mapping->size = memdev->region_size;
2873                 break;
2874         case NFIT_SPA_DCR:
2875                 nfit_mem = nvdimm_provider_data(nvdimm);
2876                 if (!nfit_mem || !nfit_mem->bdw) {
2877                         dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2878                                         spa->range_index, nvdimm_name(nvdimm));
2879                         break;
2880                 }
2881
2882                 mapping->size = nfit_mem->bdw->capacity;
2883                 mapping->start = nfit_mem->bdw->start_address;
2884                 ndr_desc->num_lanes = nfit_mem->bdw->windows;
2885                 ndr_desc->mapping = mapping;
2886                 ndr_desc->num_mappings = 1;
2887                 ndbr_desc = to_blk_region_desc(ndr_desc);
2888                 ndbr_desc->enable = acpi_nfit_blk_region_enable;
2889                 ndbr_desc->do_io = acpi_desc->blk_do_io;
2890                 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2891                 if (rc)
2892                         return rc;
2893                 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2894                                 ndr_desc);
2895                 if (!nfit_spa->nd_region)
2896                         return -ENOMEM;
2897                 break;
2898         }
2899
2900         return 0;
2901 }
2902
2903 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2904 {
2905         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2906                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2907                 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2908                 nfit_spa_type(spa) == NFIT_SPA_PCD);
2909 }
2910
2911 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
2912 {
2913         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2914                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2915                 nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
2916 }
2917
2918 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2919                 struct nfit_spa *nfit_spa)
2920 {
2921         static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2922         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2923         struct nd_blk_region_desc ndbr_desc;
2924         struct nd_region_desc *ndr_desc;
2925         struct nfit_memdev *nfit_memdev;
2926         struct nvdimm_bus *nvdimm_bus;
2927         struct resource res;
2928         int count = 0, rc;
2929
2930         if (nfit_spa->nd_region)
2931                 return 0;
2932
2933         if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2934                 dev_dbg(acpi_desc->dev, "detected invalid spa index\n");
2935                 return 0;
2936         }
2937
2938         memset(&res, 0, sizeof(res));
2939         memset(&mappings, 0, sizeof(mappings));
2940         memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2941         res.start = spa->address;
2942         res.end = res.start + spa->length - 1;
2943         ndr_desc = &ndbr_desc.ndr_desc;
2944         ndr_desc->res = &res;
2945         ndr_desc->provider_data = nfit_spa;
2946         ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2947         if (spa->flags & ACPI_NFIT_PROXIMITY_VALID) {
2948                 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2949                                                 spa->proximity_domain);
2950                 ndr_desc->target_node = acpi_map_pxm_to_node(
2951                                 spa->proximity_domain);
2952         } else {
2953                 ndr_desc->numa_node = NUMA_NO_NODE;
2954                 ndr_desc->target_node = NUMA_NO_NODE;
2955         }
2956
2957         /*
2958          * Persistence domain bits are hierarchical, if
2959          * ACPI_NFIT_CAPABILITY_CACHE_FLUSH is set then
2960          * ACPI_NFIT_CAPABILITY_MEM_FLUSH is implied.
2961          */
2962         if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_CACHE_FLUSH)
2963                 set_bit(ND_REGION_PERSIST_CACHE, &ndr_desc->flags);
2964         else if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_MEM_FLUSH)
2965                 set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc->flags);
2966
2967         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2968                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2969                 struct nd_mapping_desc *mapping;
2970
2971                 if (memdev->range_index != spa->range_index)
2972                         continue;
2973                 if (count >= ND_MAX_MAPPINGS) {
2974                         dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2975                                         spa->range_index, ND_MAX_MAPPINGS);
2976                         return -ENXIO;
2977                 }
2978                 mapping = &mappings[count++];
2979                 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
2980                                 memdev, nfit_spa);
2981                 if (rc)
2982                         goto out;
2983         }
2984
2985         ndr_desc->mapping = mappings;
2986         ndr_desc->num_mappings = count;
2987         rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2988         if (rc)
2989                 goto out;
2990
2991         nvdimm_bus = acpi_desc->nvdimm_bus;
2992         if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2993                 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2994                 if (rc) {
2995                         dev_warn(acpi_desc->dev,
2996                                 "failed to insert pmem resource to iomem: %d\n",
2997                                 rc);
2998                         goto out;
2999                 }
3000
3001                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
3002                                 ndr_desc);
3003                 if (!nfit_spa->nd_region)
3004                         rc = -ENOMEM;
3005         } else if (nfit_spa_is_volatile(spa)) {
3006                 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
3007                                 ndr_desc);
3008                 if (!nfit_spa->nd_region)
3009                         rc = -ENOMEM;
3010         } else if (nfit_spa_is_virtual(spa)) {
3011                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
3012                                 ndr_desc);
3013                 if (!nfit_spa->nd_region)
3014                         rc = -ENOMEM;
3015         }
3016
3017  out:
3018         if (rc)
3019                 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
3020                                 nfit_spa->spa->range_index);
3021         return rc;
3022 }
3023
3024 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc)
3025 {
3026         struct device *dev = acpi_desc->dev;
3027         struct nd_cmd_ars_status *ars_status;
3028
3029         if (acpi_desc->ars_status) {
3030                 memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
3031                 return 0;
3032         }
3033
3034         ars_status = devm_kzalloc(dev, acpi_desc->max_ars, GFP_KERNEL);
3035         if (!ars_status)
3036                 return -ENOMEM;
3037         acpi_desc->ars_status = ars_status;
3038         return 0;
3039 }
3040
3041 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc)
3042 {
3043         int rc;
3044
3045         if (ars_status_alloc(acpi_desc))
3046                 return -ENOMEM;
3047
3048         rc = ars_get_status(acpi_desc);
3049
3050         if (rc < 0 && rc != -ENOSPC)
3051                 return rc;
3052
3053         if (ars_status_process_records(acpi_desc))
3054                 dev_err(acpi_desc->dev, "Failed to process ARS records\n");
3055
3056         return rc;
3057 }
3058
3059 static int ars_register(struct acpi_nfit_desc *acpi_desc,
3060                 struct nfit_spa *nfit_spa)
3061 {
3062         int rc;
3063
3064         if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3065                 return acpi_nfit_register_region(acpi_desc, nfit_spa);
3066
3067         set_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
3068         if (!no_init_ars)
3069                 set_bit(ARS_REQ_LONG, &nfit_spa->ars_state);
3070
3071         switch (acpi_nfit_query_poison(acpi_desc)) {
3072         case 0:
3073         case -ENOSPC:
3074         case -EAGAIN:
3075                 rc = ars_start(acpi_desc, nfit_spa, ARS_REQ_SHORT);
3076                 /* shouldn't happen, try again later */
3077                 if (rc == -EBUSY)
3078                         break;
3079                 if (rc) {
3080                         set_bit(ARS_FAILED, &nfit_spa->ars_state);
3081                         break;
3082                 }
3083                 clear_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
3084                 rc = acpi_nfit_query_poison(acpi_desc);
3085                 if (rc)
3086                         break;
3087                 acpi_desc->scrub_spa = nfit_spa;
3088                 ars_complete(acpi_desc, nfit_spa);
3089                 /*
3090                  * If ars_complete() says we didn't complete the
3091                  * short scrub, we'll try again with a long
3092                  * request.
3093                  */
3094                 acpi_desc->scrub_spa = NULL;
3095                 break;
3096         case -EBUSY:
3097         case -ENOMEM:
3098                 /*
3099                  * BIOS was using ARS, wait for it to complete (or
3100                  * resources to become available) and then perform our
3101                  * own scrubs.
3102                  */
3103                 break;
3104         default:
3105                 set_bit(ARS_FAILED, &nfit_spa->ars_state);
3106                 break;
3107         }
3108
3109         return acpi_nfit_register_region(acpi_desc, nfit_spa);
3110 }
3111
3112 static void ars_complete_all(struct acpi_nfit_desc *acpi_desc)
3113 {
3114         struct nfit_spa *nfit_spa;
3115
3116         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3117                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3118                         continue;
3119                 ars_complete(acpi_desc, nfit_spa);
3120         }
3121 }
3122
3123 static unsigned int __acpi_nfit_scrub(struct acpi_nfit_desc *acpi_desc,
3124                 int query_rc)
3125 {
3126         unsigned int tmo = acpi_desc->scrub_tmo;
3127         struct device *dev = acpi_desc->dev;
3128         struct nfit_spa *nfit_spa;
3129
3130         lockdep_assert_held(&acpi_desc->init_mutex);
3131
3132         if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags))
3133                 return 0;
3134
3135         if (query_rc == -EBUSY) {
3136                 dev_dbg(dev, "ARS: ARS busy\n");
3137                 return min(30U * 60U, tmo * 2);
3138         }
3139         if (query_rc == -ENOSPC) {
3140                 dev_dbg(dev, "ARS: ARS continue\n");
3141                 ars_continue(acpi_desc);
3142                 return 1;
3143         }
3144         if (query_rc && query_rc != -EAGAIN) {
3145                 unsigned long long addr, end;
3146
3147                 addr = acpi_desc->ars_status->address;
3148                 end = addr + acpi_desc->ars_status->length;
3149                 dev_dbg(dev, "ARS: %llx-%llx failed (%d)\n", addr, end,
3150                                 query_rc);
3151         }
3152
3153         ars_complete_all(acpi_desc);
3154         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3155                 enum nfit_ars_state req_type;
3156                 int rc;
3157
3158                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3159                         continue;
3160
3161                 /* prefer short ARS requests first */
3162                 if (test_bit(ARS_REQ_SHORT, &nfit_spa->ars_state))
3163                         req_type = ARS_REQ_SHORT;
3164                 else if (test_bit(ARS_REQ_LONG, &nfit_spa->ars_state))
3165                         req_type = ARS_REQ_LONG;
3166                 else
3167                         continue;
3168                 rc = ars_start(acpi_desc, nfit_spa, req_type);
3169
3170                 dev = nd_region_dev(nfit_spa->nd_region);
3171                 dev_dbg(dev, "ARS: range %d ARS start %s (%d)\n",
3172                                 nfit_spa->spa->range_index,
3173                                 req_type == ARS_REQ_SHORT ? "short" : "long",
3174                                 rc);
3175                 /*
3176                  * Hmm, we raced someone else starting ARS? Try again in
3177                  * a bit.
3178                  */
3179                 if (rc == -EBUSY)
3180                         return 1;
3181                 if (rc == 0) {
3182                         dev_WARN_ONCE(dev, acpi_desc->scrub_spa,
3183                                         "scrub start while range %d active\n",
3184                                         acpi_desc->scrub_spa->spa->range_index);
3185                         clear_bit(req_type, &nfit_spa->ars_state);
3186                         acpi_desc->scrub_spa = nfit_spa;
3187                         /*
3188                          * Consider this spa last for future scrub
3189                          * requests
3190                          */
3191                         list_move_tail(&nfit_spa->list, &acpi_desc->spas);
3192                         return 1;
3193                 }
3194
3195                 dev_err(dev, "ARS: range %d ARS failed (%d)\n",
3196                                 nfit_spa->spa->range_index, rc);
3197                 set_bit(ARS_FAILED, &nfit_spa->ars_state);
3198         }
3199         return 0;
3200 }
3201
3202 static void __sched_ars(struct acpi_nfit_desc *acpi_desc, unsigned int tmo)
3203 {
3204         lockdep_assert_held(&acpi_desc->init_mutex);
3205
3206         set_bit(ARS_BUSY, &acpi_desc->scrub_flags);
3207         /* note this should only be set from within the workqueue */
3208         if (tmo)
3209                 acpi_desc->scrub_tmo = tmo;
3210         queue_delayed_work(nfit_wq, &acpi_desc->dwork, tmo * HZ);
3211 }
3212
3213 static void sched_ars(struct acpi_nfit_desc *acpi_desc)
3214 {
3215         __sched_ars(acpi_desc, 0);
3216 }
3217
3218 static void notify_ars_done(struct acpi_nfit_desc *acpi_desc)
3219 {
3220         lockdep_assert_held(&acpi_desc->init_mutex);
3221
3222         clear_bit(ARS_BUSY, &acpi_desc->scrub_flags);
3223         acpi_desc->scrub_count++;
3224         if (acpi_desc->scrub_count_state)
3225                 sysfs_notify_dirent(acpi_desc->scrub_count_state);
3226 }
3227
3228 static void acpi_nfit_scrub(struct work_struct *work)
3229 {
3230         struct acpi_nfit_desc *acpi_desc;
3231         unsigned int tmo;
3232         int query_rc;
3233
3234         acpi_desc = container_of(work, typeof(*acpi_desc), dwork.work);
3235         mutex_lock(&acpi_desc->init_mutex);
3236         query_rc = acpi_nfit_query_poison(acpi_desc);
3237         tmo = __acpi_nfit_scrub(acpi_desc, query_rc);
3238         if (tmo)
3239                 __sched_ars(acpi_desc, tmo);
3240         else
3241                 notify_ars_done(acpi_desc);
3242         memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
3243         clear_bit(ARS_POLL, &acpi_desc->scrub_flags);
3244         mutex_unlock(&acpi_desc->init_mutex);
3245 }
3246
3247 static void acpi_nfit_init_ars(struct acpi_nfit_desc *acpi_desc,
3248                 struct nfit_spa *nfit_spa)
3249 {
3250         int type = nfit_spa_type(nfit_spa->spa);
3251         struct nd_cmd_ars_cap ars_cap;
3252         int rc;
3253
3254         set_bit(ARS_FAILED, &nfit_spa->ars_state);
3255         memset(&ars_cap, 0, sizeof(ars_cap));
3256         rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
3257         if (rc < 0)
3258                 return;
3259         /* check that the supported scrub types match the spa type */
3260         if (type == NFIT_SPA_VOLATILE && ((ars_cap.status >> 16)
3261                                 & ND_ARS_VOLATILE) == 0)
3262                 return;
3263         if (type == NFIT_SPA_PM && ((ars_cap.status >> 16)
3264                                 & ND_ARS_PERSISTENT) == 0)
3265                 return;
3266
3267         nfit_spa->max_ars = ars_cap.max_ars_out;
3268         nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
3269         acpi_desc->max_ars = max(nfit_spa->max_ars, acpi_desc->max_ars);
3270         clear_bit(ARS_FAILED, &nfit_spa->ars_state);
3271 }
3272
3273 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
3274 {
3275         struct nfit_spa *nfit_spa;
3276         int rc;
3277
3278         set_bit(ARS_VALID, &acpi_desc->scrub_flags);
3279         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3280                 switch (nfit_spa_type(nfit_spa->spa)) {
3281                 case NFIT_SPA_VOLATILE:
3282                 case NFIT_SPA_PM:
3283                         acpi_nfit_init_ars(acpi_desc, nfit_spa);
3284                         break;
3285                 }
3286         }
3287
3288         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
3289                 switch (nfit_spa_type(nfit_spa->spa)) {
3290                 case NFIT_SPA_VOLATILE:
3291                 case NFIT_SPA_PM:
3292                         /* register regions and kick off initial ARS run */
3293                         rc = ars_register(acpi_desc, nfit_spa);
3294                         if (rc)
3295                                 return rc;
3296                         break;
3297                 case NFIT_SPA_BDW:
3298                         /* nothing to register */
3299                         break;
3300                 case NFIT_SPA_DCR:
3301                 case NFIT_SPA_VDISK:
3302                 case NFIT_SPA_VCD:
3303                 case NFIT_SPA_PDISK:
3304                 case NFIT_SPA_PCD:
3305                         /* register known regions that don't support ARS */
3306                         rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
3307                         if (rc)
3308                                 return rc;
3309                         break;
3310                 default:
3311                         /* don't register unknown regions */
3312                         break;
3313                 }
3314
3315         sched_ars(acpi_desc);
3316         return 0;
3317 }
3318
3319 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
3320                 struct nfit_table_prev *prev)
3321 {
3322         struct device *dev = acpi_desc->dev;
3323
3324         if (!list_empty(&prev->spas) ||
3325                         !list_empty(&prev->memdevs) ||
3326                         !list_empty(&prev->dcrs) ||
3327                         !list_empty(&prev->bdws) ||
3328                         !list_empty(&prev->idts) ||
3329                         !list_empty(&prev->flushes)) {
3330                 dev_err(dev, "new nfit deletes entries (unsupported)\n");
3331                 return -ENXIO;
3332         }
3333         return 0;
3334 }
3335
3336 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
3337 {
3338         struct device *dev = acpi_desc->dev;
3339         struct kernfs_node *nfit;
3340         struct device *bus_dev;
3341
3342         if (!ars_supported(acpi_desc->nvdimm_bus))
3343                 return 0;
3344
3345         bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3346         nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
3347         if (!nfit) {
3348                 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
3349                 return -ENODEV;
3350         }
3351         acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
3352         sysfs_put(nfit);
3353         if (!acpi_desc->scrub_count_state) {
3354                 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
3355                 return -ENODEV;
3356         }
3357
3358         return 0;
3359 }
3360
3361 static void acpi_nfit_unregister(void *data)
3362 {
3363         struct acpi_nfit_desc *acpi_desc = data;
3364
3365         nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
3366 }
3367
3368 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
3369 {
3370         struct device *dev = acpi_desc->dev;
3371         struct nfit_table_prev prev;
3372         const void *end;
3373         int rc;
3374
3375         if (!acpi_desc->nvdimm_bus) {
3376                 acpi_nfit_init_dsms(acpi_desc);
3377
3378                 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
3379                                 &acpi_desc->nd_desc);
3380                 if (!acpi_desc->nvdimm_bus)
3381                         return -ENOMEM;
3382
3383                 rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
3384                                 acpi_desc);
3385                 if (rc)
3386                         return rc;
3387
3388                 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
3389                 if (rc)
3390                         return rc;
3391
3392                 /* register this acpi_desc for mce notifications */
3393                 mutex_lock(&acpi_desc_lock);
3394                 list_add_tail(&acpi_desc->list, &acpi_descs);
3395                 mutex_unlock(&acpi_desc_lock);
3396         }
3397
3398         mutex_lock(&acpi_desc->init_mutex);
3399
3400         INIT_LIST_HEAD(&prev.spas);
3401         INIT_LIST_HEAD(&prev.memdevs);
3402         INIT_LIST_HEAD(&prev.dcrs);
3403         INIT_LIST_HEAD(&prev.bdws);
3404         INIT_LIST_HEAD(&prev.idts);
3405         INIT_LIST_HEAD(&prev.flushes);
3406
3407         list_cut_position(&prev.spas, &acpi_desc->spas,
3408                                 acpi_desc->spas.prev);
3409         list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
3410                                 acpi_desc->memdevs.prev);
3411         list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
3412                                 acpi_desc->dcrs.prev);
3413         list_cut_position(&prev.bdws, &acpi_desc->bdws,
3414                                 acpi_desc->bdws.prev);
3415         list_cut_position(&prev.idts, &acpi_desc->idts,
3416                                 acpi_desc->idts.prev);
3417         list_cut_position(&prev.flushes, &acpi_desc->flushes,
3418                                 acpi_desc->flushes.prev);
3419
3420         end = data + sz;
3421         while (!IS_ERR_OR_NULL(data))
3422                 data = add_table(acpi_desc, &prev, data, end);
3423
3424         if (IS_ERR(data)) {
3425                 dev_dbg(dev, "nfit table parsing error: %ld\n", PTR_ERR(data));
3426                 rc = PTR_ERR(data);
3427                 goto out_unlock;
3428         }
3429
3430         rc = acpi_nfit_check_deletions(acpi_desc, &prev);
3431         if (rc)
3432                 goto out_unlock;
3433
3434         rc = nfit_mem_init(acpi_desc);
3435         if (rc)
3436                 goto out_unlock;
3437
3438         rc = acpi_nfit_register_dimms(acpi_desc);
3439         if (rc)
3440                 goto out_unlock;
3441
3442         rc = acpi_nfit_register_regions(acpi_desc);
3443
3444  out_unlock:
3445         mutex_unlock(&acpi_desc->init_mutex);
3446         return rc;
3447 }
3448 EXPORT_SYMBOL_GPL(acpi_nfit_init);
3449
3450 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
3451 {
3452         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
3453         struct device *dev = acpi_desc->dev;
3454
3455         /* Bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
3456         nfit_device_lock(dev);
3457         nfit_device_unlock(dev);
3458
3459         /* Bounce the init_mutex to complete initial registration */
3460         mutex_lock(&acpi_desc->init_mutex);
3461         mutex_unlock(&acpi_desc->init_mutex);
3462
3463         return 0;
3464 }
3465
3466 static int __acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3467                 struct nvdimm *nvdimm, unsigned int cmd)
3468 {
3469         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
3470
3471         if (nvdimm)
3472                 return 0;
3473         if (cmd != ND_CMD_ARS_START)
3474                 return 0;
3475
3476         /*
3477          * The kernel and userspace may race to initiate a scrub, but
3478          * the scrub thread is prepared to lose that initial race.  It
3479          * just needs guarantees that any ARS it initiates are not
3480          * interrupted by any intervening start requests from userspace.
3481          */
3482         if (work_busy(&acpi_desc->dwork.work))
3483                 return -EBUSY;
3484
3485         return 0;
3486 }
3487
3488 /* prevent security commands from being issued via ioctl */
3489 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3490                 struct nvdimm *nvdimm, unsigned int cmd, void *buf)
3491 {
3492         struct nd_cmd_pkg *call_pkg = buf;
3493         unsigned int func;
3494
3495         if (nvdimm && cmd == ND_CMD_CALL &&
3496                         call_pkg->nd_family == NVDIMM_FAMILY_INTEL) {
3497                 func = call_pkg->nd_command;
3498                 if (func > NVDIMM_CMD_MAX ||
3499                     (1 << func) & NVDIMM_INTEL_SECURITY_CMDMASK)
3500                         return -EOPNOTSUPP;
3501         }
3502
3503         return __acpi_nfit_clear_to_send(nd_desc, nvdimm, cmd);
3504 }
3505
3506 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc,
3507                 enum nfit_ars_state req_type)
3508 {
3509         struct device *dev = acpi_desc->dev;
3510         int scheduled = 0, busy = 0;
3511         struct nfit_spa *nfit_spa;
3512
3513         mutex_lock(&acpi_desc->init_mutex);
3514         if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags)) {
3515                 mutex_unlock(&acpi_desc->init_mutex);
3516                 return 0;
3517         }
3518
3519         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3520                 int type = nfit_spa_type(nfit_spa->spa);
3521
3522                 if (type != NFIT_SPA_PM && type != NFIT_SPA_VOLATILE)
3523                         continue;
3524                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3525                         continue;
3526
3527                 if (test_and_set_bit(req_type, &nfit_spa->ars_state))
3528                         busy++;
3529                 else
3530                         scheduled++;
3531         }
3532         if (scheduled) {
3533                 sched_ars(acpi_desc);
3534                 dev_dbg(dev, "ars_scan triggered\n");
3535         }
3536         mutex_unlock(&acpi_desc->init_mutex);
3537
3538         if (scheduled)
3539                 return 0;
3540         if (busy)
3541                 return -EBUSY;
3542         return -ENOTTY;
3543 }
3544
3545 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
3546 {
3547         struct nvdimm_bus_descriptor *nd_desc;
3548
3549         dev_set_drvdata(dev, acpi_desc);
3550         acpi_desc->dev = dev;
3551         acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
3552         nd_desc = &acpi_desc->nd_desc;
3553         nd_desc->provider_name = "ACPI.NFIT";
3554         nd_desc->module = THIS_MODULE;
3555         nd_desc->ndctl = acpi_nfit_ctl;
3556         nd_desc->flush_probe = acpi_nfit_flush_probe;
3557         nd_desc->clear_to_send = acpi_nfit_clear_to_send;
3558         nd_desc->attr_groups = acpi_nfit_attribute_groups;
3559
3560         INIT_LIST_HEAD(&acpi_desc->spas);
3561         INIT_LIST_HEAD(&acpi_desc->dcrs);
3562         INIT_LIST_HEAD(&acpi_desc->bdws);
3563         INIT_LIST_HEAD(&acpi_desc->idts);
3564         INIT_LIST_HEAD(&acpi_desc->flushes);
3565         INIT_LIST_HEAD(&acpi_desc->memdevs);
3566         INIT_LIST_HEAD(&acpi_desc->dimms);
3567         INIT_LIST_HEAD(&acpi_desc->list);
3568         mutex_init(&acpi_desc->init_mutex);
3569         acpi_desc->scrub_tmo = 1;
3570         INIT_DELAYED_WORK(&acpi_desc->dwork, acpi_nfit_scrub);
3571 }
3572 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
3573
3574 static void acpi_nfit_put_table(void *table)
3575 {
3576         acpi_put_table(table);
3577 }
3578
3579 void acpi_nfit_shutdown(void *data)
3580 {
3581         struct acpi_nfit_desc *acpi_desc = data;
3582         struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3583
3584         /*
3585          * Destruct under acpi_desc_lock so that nfit_handle_mce does not
3586          * race teardown
3587          */
3588         mutex_lock(&acpi_desc_lock);
3589         list_del(&acpi_desc->list);
3590         mutex_unlock(&acpi_desc_lock);
3591
3592         mutex_lock(&acpi_desc->init_mutex);
3593         set_bit(ARS_CANCEL, &acpi_desc->scrub_flags);
3594         cancel_delayed_work_sync(&acpi_desc->dwork);
3595         mutex_unlock(&acpi_desc->init_mutex);
3596
3597         /*
3598          * Bounce the nvdimm bus lock to make sure any in-flight
3599          * acpi_nfit_ars_rescan() submissions have had a chance to
3600          * either submit or see ->cancel set.
3601          */
3602         nfit_device_lock(bus_dev);
3603         nfit_device_unlock(bus_dev);
3604
3605         flush_workqueue(nfit_wq);
3606 }
3607 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown);
3608
3609 static int acpi_nfit_add(struct acpi_device *adev)
3610 {
3611         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3612         struct acpi_nfit_desc *acpi_desc;
3613         struct device *dev = &adev->dev;
3614         struct acpi_table_header *tbl;
3615         acpi_status status = AE_OK;
3616         acpi_size sz;
3617         int rc = 0;
3618
3619         status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
3620         if (ACPI_FAILURE(status)) {
3621                 /* The NVDIMM root device allows OS to trigger enumeration of
3622                  * NVDIMMs through NFIT at boot time and re-enumeration at
3623                  * root level via the _FIT method during runtime.
3624                  * This is ok to return 0 here, we could have an nvdimm
3625                  * hotplugged later and evaluate _FIT method which returns
3626                  * data in the format of a series of NFIT Structures.
3627                  */
3628                 dev_dbg(dev, "failed to find NFIT at startup\n");
3629                 return 0;
3630         }
3631
3632         rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl);
3633         if (rc)
3634                 return rc;
3635         sz = tbl->length;
3636
3637         acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3638         if (!acpi_desc)
3639                 return -ENOMEM;
3640         acpi_nfit_desc_init(acpi_desc, &adev->dev);
3641
3642         /* Save the acpi header for exporting the revision via sysfs */
3643         acpi_desc->acpi_header = *tbl;
3644
3645         /* Evaluate _FIT and override with that if present */
3646         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
3647         if (ACPI_SUCCESS(status) && buf.length > 0) {
3648                 union acpi_object *obj = buf.pointer;
3649
3650                 if (obj->type == ACPI_TYPE_BUFFER)
3651                         rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3652                                         obj->buffer.length);
3653                 else
3654                         dev_dbg(dev, "invalid type %d, ignoring _FIT\n",
3655                                 (int) obj->type);
3656                 kfree(buf.pointer);
3657         } else
3658                 /* skip over the lead-in header table */
3659                 rc = acpi_nfit_init(acpi_desc, (void *) tbl
3660                                 + sizeof(struct acpi_table_nfit),
3661                                 sz - sizeof(struct acpi_table_nfit));
3662
3663         if (rc)
3664                 return rc;
3665         return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc);
3666 }
3667
3668 static int acpi_nfit_remove(struct acpi_device *adev)
3669 {
3670         /* see acpi_nfit_unregister */
3671         return 0;
3672 }
3673
3674 static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle)
3675 {
3676         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3677         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3678         union acpi_object *obj;
3679         acpi_status status;
3680         int ret;
3681
3682         if (!dev->driver) {
3683                 /* dev->driver may be null if we're being removed */
3684                 dev_dbg(dev, "no driver found for dev\n");
3685                 return;
3686         }
3687
3688         if (!acpi_desc) {
3689                 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3690                 if (!acpi_desc)
3691                         return;
3692                 acpi_nfit_desc_init(acpi_desc, dev);
3693         } else {
3694                 /*
3695                  * Finish previous registration before considering new
3696                  * regions.
3697                  */
3698                 flush_workqueue(nfit_wq);
3699         }
3700
3701         /* Evaluate _FIT */
3702         status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
3703         if (ACPI_FAILURE(status)) {
3704                 dev_err(dev, "failed to evaluate _FIT\n");
3705                 return;
3706         }
3707
3708         obj = buf.pointer;
3709         if (obj->type == ACPI_TYPE_BUFFER) {
3710                 ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3711                                 obj->buffer.length);
3712                 if (ret)
3713                         dev_err(dev, "failed to merge updated NFIT\n");
3714         } else
3715                 dev_err(dev, "Invalid _FIT\n");
3716         kfree(buf.pointer);
3717 }
3718
3719 static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle)
3720 {
3721         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3722
3723         if (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON)
3724                 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
3725         else
3726                 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_SHORT);
3727 }
3728
3729 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
3730 {
3731         dev_dbg(dev, "event: 0x%x\n", event);
3732
3733         switch (event) {
3734         case NFIT_NOTIFY_UPDATE:
3735                 return acpi_nfit_update_notify(dev, handle);
3736         case NFIT_NOTIFY_UC_MEMORY_ERROR:
3737                 return acpi_nfit_uc_error_notify(dev, handle);
3738         default:
3739                 return;
3740         }
3741 }
3742 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
3743
3744 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
3745 {
3746         nfit_device_lock(&adev->dev);
3747         __acpi_nfit_notify(&adev->dev, adev->handle, event);
3748         nfit_device_unlock(&adev->dev);
3749 }
3750
3751 static const struct acpi_device_id acpi_nfit_ids[] = {
3752         { "ACPI0012", 0 },
3753         { "", 0 },
3754 };
3755 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
3756
3757 static struct acpi_driver acpi_nfit_driver = {
3758         .name = KBUILD_MODNAME,
3759         .ids = acpi_nfit_ids,
3760         .ops = {
3761                 .add = acpi_nfit_add,
3762                 .remove = acpi_nfit_remove,
3763                 .notify = acpi_nfit_notify,
3764         },
3765 };
3766
3767 static __init int nfit_init(void)
3768 {
3769         int ret;
3770
3771         BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
3772         BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
3773         BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
3774         BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
3775         BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
3776         BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
3777         BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
3778         BUILD_BUG_ON(sizeof(struct acpi_nfit_capabilities) != 16);
3779
3780         guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]);
3781         guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]);
3782         guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]);
3783         guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]);
3784         guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]);
3785         guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]);
3786         guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]);
3787         guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]);
3788         guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]);
3789         guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]);
3790         guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
3791         guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
3792         guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
3793         guid_parse(UUID_NFIT_DIMM_N_HYPERV, &nfit_uuid[NFIT_DEV_DIMM_N_HYPERV]);
3794
3795         nfit_wq = create_singlethread_workqueue("nfit");
3796         if (!nfit_wq)
3797                 return -ENOMEM;
3798
3799         nfit_mce_register();
3800         ret = acpi_bus_register_driver(&acpi_nfit_driver);
3801         if (ret) {
3802                 nfit_mce_unregister();
3803                 destroy_workqueue(nfit_wq);
3804         }
3805
3806         return ret;
3807
3808 }
3809
3810 static __exit void nfit_exit(void)
3811 {
3812         nfit_mce_unregister();
3813         acpi_bus_unregister_driver(&acpi_nfit_driver);
3814         destroy_workqueue(nfit_wq);
3815         WARN_ON(!list_empty(&acpi_descs));
3816 }
3817
3818 module_init(nfit_init);
3819 module_exit(nfit_exit);
3820 MODULE_LICENSE("GPL v2");
3821 MODULE_AUTHOR("Intel Corporation");