drivers/acpi/nfit/core.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
   4  */
   5 #include <linux/list_sort.h>
   6 #include <linux/libnvdimm.h>
   7 #include <linux/module.h>
   8 #include <linux/nospec.h>
   9 #include <linux/mutex.h>
  10 #include <linux/ndctl.h>
  11 #include <linux/sysfs.h>
  12 #include <linux/delay.h>
  13 #include <linux/list.h>
  14 #include <linux/acpi.h>
  15 #include <linux/sort.h>
  16 #include <linux/io.h>
  17 #include <linux/nd.h>
  18 #include <asm/cacheflush.h>
  19 #include <acpi/nfit.h>
  20 #include "intel.h"
  21 #include "nfit.h"
  22
  23 /*
  24  * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
  25  * irrelevant.
  26  */
  27 #include <linux/io-64-nonatomic-hi-lo.h>
  28
  29 static bool force_enable_dimms;
  30 module_param(force_enable_dimms, bool, S_IRUGO|S_IWUSR);
  31 MODULE_PARM_DESC(force_enable_dimms, "Ignore _STA (ACPI DIMM device) status");
  32
  33 static bool disable_vendor_specific;
  34 module_param(disable_vendor_specific, bool, S_IRUGO);
  35 MODULE_PARM_DESC(disable_vendor_specific,
  36                 "Limit commands to the publicly specified set");
  37
  38 static unsigned long override_dsm_mask;
  39 module_param(override_dsm_mask, ulong, S_IRUGO);
  40 MODULE_PARM_DESC(override_dsm_mask, "Bitmask of allowed NVDIMM DSM functions");
  41
  42 static int default_dsm_family = -1;
  43 module_param(default_dsm_family, int, S_IRUGO);
  44 MODULE_PARM_DESC(default_dsm_family,
  45                 "Try this DSM type first when identifying NVDIMM family");
  46
  47 static bool no_init_ars;
  48 module_param(no_init_ars, bool, 0644);
  49 MODULE_PARM_DESC(no_init_ars, "Skip ARS run at nfit init time");
  50
  51 static bool force_labels;
  52 module_param(force_labels, bool, 0444);
  53 MODULE_PARM_DESC(force_labels, "Opt-in to labels despite missing methods");
  54
  55 LIST_HEAD(acpi_descs);
  56 DEFINE_MUTEX(acpi_desc_lock);
  57
  58 static struct workqueue_struct *nfit_wq;
  59
  60 struct nfit_table_prev {
  61         struct list_head spas;
  62         struct list_head memdevs;
  63         struct list_head dcrs;
  64         struct list_head bdws;
  65         struct list_head idts;
  66         struct list_head flushes;
  67 };
  68
  69 static guid_t nfit_uuid[NFIT_UUID_MAX];
  70
  71 const guid_t *to_nfit_uuid(enum nfit_uuids id)
  72 {
  73         return &nfit_uuid[id];
  74 }
  75 EXPORT_SYMBOL(to_nfit_uuid);
  76
  77 static const guid_t *to_nfit_bus_uuid(int family)
  78 {
  79         if (WARN_ONCE(family == NVDIMM_BUS_FAMILY_NFIT,
  80                         "only secondary bus families can be translated\n"))
  81                 return NULL;
  82         /*
  83          * The index of bus UUIDs starts immediately following the last
  84          * NVDIMM/leaf family.
  85          */
  86         return to_nfit_uuid(family + NVDIMM_FAMILY_MAX);
  87 }
  88
  89 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
  90 {
  91         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
  92
  93         /*
  94          * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
  95          * acpi_device.
  96          */
  97         if (!nd_desc->provider_name
  98                         || strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
  99                 return NULL;
 100
 101         return to_acpi_device(acpi_desc->dev);
 102 }
 103
 104 static int xlat_bus_status(void *buf, unsigned int cmd, u32 status)
 105 {
 106         struct nd_cmd_clear_error *clear_err;
 107         struct nd_cmd_ars_status *ars_status;
 108         u16 flags;
 109
 110         switch (cmd) {
 111         case ND_CMD_ARS_CAP:
 112                 if ((status & 0xffff) == NFIT_ARS_CAP_NONE)
 113                         return -ENOTTY;
 114
 115                 /* Command failed */
 116                 if (status & 0xffff)
 117                         return -EIO;
 118
 119                 /* No supported scan types for this range */
 120                 flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
 121                 if ((status >> 16 & flags) == 0)
 122                         return -ENOTTY;
 123                 return 0;
 124         case ND_CMD_ARS_START:
 125                 /* ARS is in progress */
 126                 if ((status & 0xffff) == NFIT_ARS_START_BUSY)
 127                         return -EBUSY;
 128
 129                 /* Command failed */
 130                 if (status & 0xffff)
 131                         return -EIO;
 132                 return 0;
 133         case ND_CMD_ARS_STATUS:
 134                 ars_status = buf;
 135                 /* Command failed */
 136                 if (status & 0xffff)
 137                         return -EIO;
 138                 /* Check extended status (Upper two bytes) */
 139                 if (status == NFIT_ARS_STATUS_DONE)
 140                         return 0;
 141
 142                 /* ARS is in progress */
 143                 if (status == NFIT_ARS_STATUS_BUSY)
 144                         return -EBUSY;
 145
 146                 /* No ARS performed for the current boot */
 147                 if (status == NFIT_ARS_STATUS_NONE)
 148                         return -EAGAIN;
 149
 150                 /*
 151                  * ARS interrupted, either we overflowed or some other
 152                  * agent wants the scan to stop.  If we didn't overflow
 153                  * then just continue with the returned results.
 154                  */
 155                 if (status == NFIT_ARS_STATUS_INTR) {
 156                         if (ars_status->out_length >= 40 && (ars_status->flags
 157                                                 & NFIT_ARS_F_OVERFLOW))
 158                                 return -ENOSPC;
 159                         return 0;
 160                 }
 161
 162                 /* Unknown status */
 163                 if (status >> 16)
 164                         return -EIO;
 165                 return 0;
 166         case ND_CMD_CLEAR_ERROR:
 167                 clear_err = buf;
 168                 if (status & 0xffff)
 169                         return -EIO;
 170                 if (!clear_err->cleared)
 171                         return -EIO;
 172                 if (clear_err->length > clear_err->cleared)
 173                         return clear_err->cleared;
 174                 return 0;
 175         default:
 176                 break;
 177         }
 178
 179         /* all other non-zero status results in an error */
 180         if (status)
 181                 return -EIO;
 182         return 0;
 183 }
 184
 185 #define ACPI_LABELS_LOCKED 3
 186
 187 static int xlat_nvdimm_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
 188                 u32 status)
 189 {
 190         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 191
 192         switch (cmd) {
 193         case ND_CMD_GET_CONFIG_SIZE:
 194                 /*
 195                  * In the _LSI, _LSR, _LSW case the locked status is
 196                  * communicated via the read/write commands
 197                  */
 198                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags))
 199                         break;
 200
 201                 if (status >> 16 & ND_CONFIG_LOCKED)
 202                         return -EACCES;
 203                 break;
 204         case ND_CMD_GET_CONFIG_DATA:
 205                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
 206                                 && status == ACPI_LABELS_LOCKED)
 207                         return -EACCES;
 208                 break;
 209         case ND_CMD_SET_CONFIG_DATA:
 210                 if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags)
 211                                 && status == ACPI_LABELS_LOCKED)
 212                         return -EACCES;
 213                 break;
 214         default:
 215                 break;
 216         }
 217
 218         /* all other non-zero status results in an error */
 219         if (status)
 220                 return -EIO;
 221         return 0;
 222 }
 223
 224 static int xlat_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
 225                 u32 status)
 226 {
 227         if (!nvdimm)
 228                 return xlat_bus_status(buf, cmd, status);
 229         return xlat_nvdimm_status(nvdimm, buf, cmd, status);
 230 }
 231
 232 /* convert _LS{I,R} packages to the buffer object acpi_nfit_ctl expects */
 233 static union acpi_object *pkg_to_buf(union acpi_object *pkg)
 234 {
 235         int i;
 236         void *dst;
 237         size_t size = 0;
 238         union acpi_object *buf = NULL;
 239
 240         if (pkg->type != ACPI_TYPE_PACKAGE) {
 241                 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
 242                                 pkg->type);
 243                 goto err;
 244         }
 245
 246         for (i = 0; i < pkg->package.count; i++) {
 247                 union acpi_object *obj = &pkg->package.elements[i];
 248
 249                 if (obj->type == ACPI_TYPE_INTEGER)
 250                         size += 4;
 251                 else if (obj->type == ACPI_TYPE_BUFFER)
 252                         size += obj->buffer.length;
 253                 else {
 254                         WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
 255                                         obj->type);
 256                         goto err;
 257                 }
 258         }
 259
 260         buf = ACPI_ALLOCATE(sizeof(*buf) + size);
 261         if (!buf)
 262                 goto err;
 263
 264         dst = buf + 1;
 265         buf->type = ACPI_TYPE_BUFFER;
 266         buf->buffer.length = size;
 267         buf->buffer.pointer = dst;
 268         for (i = 0; i < pkg->package.count; i++) {
 269                 union acpi_object *obj = &pkg->package.elements[i];
 270
 271                 if (obj->type == ACPI_TYPE_INTEGER) {
 272                         memcpy(dst, &obj->integer.value, 4);
 273                         dst += 4;
 274                 } else if (obj->type == ACPI_TYPE_BUFFER) {
 275                         memcpy(dst, obj->buffer.pointer, obj->buffer.length);
 276                         dst += obj->buffer.length;
 277                 }
 278         }
 279 err:
 280         ACPI_FREE(pkg);
 281         return buf;
 282 }
 283
 284 static union acpi_object *int_to_buf(union acpi_object *integer)
 285 {
 286         union acpi_object *buf = ACPI_ALLOCATE(sizeof(*buf) + 4);
 287         void *dst = NULL;
 288
 289         if (!buf)
 290                 goto err;
 291
 292         if (integer->type != ACPI_TYPE_INTEGER) {
 293                 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
 294                                 integer->type);
 295                 goto err;
 296         }
 297
 298         dst = buf + 1;
 299         buf->type = ACPI_TYPE_BUFFER;
 300         buf->buffer.length = 4;
 301         buf->buffer.pointer = dst;
 302         memcpy(dst, &integer->integer.value, 4);
 303 err:
 304         ACPI_FREE(integer);
 305         return buf;
 306 }
 307
 308 static union acpi_object *acpi_label_write(acpi_handle handle, u32 offset,
 309                 u32 len, void *data)
 310 {
 311         acpi_status rc;
 312         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
 313         struct acpi_object_list input = {
 314                 .count = 3,
 315                 .pointer = (union acpi_object []) {
 316                         [0] = {
 317                                 .integer.type = ACPI_TYPE_INTEGER,
 318                                 .integer.value = offset,
 319                         },
 320                         [1] = {
 321                                 .integer.type = ACPI_TYPE_INTEGER,
 322                                 .integer.value = len,
 323                         },
 324                         [2] = {
 325                                 .buffer.type = ACPI_TYPE_BUFFER,
 326                                 .buffer.pointer = data,
 327                                 .buffer.length = len,
 328                         },
 329                 },
 330         };
 331
 332         rc = acpi_evaluate_object(handle, "_LSW", &input, &buf);
 333         if (ACPI_FAILURE(rc))
 334                 return NULL;
 335         return int_to_buf(buf.pointer);
 336 }
 337
 338 static union acpi_object *acpi_label_read(acpi_handle handle, u32 offset,
 339                 u32 len)
 340 {
 341         acpi_status rc;
 342         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
 343         struct acpi_object_list input = {
 344                 .count = 2,
 345                 .pointer = (union acpi_object []) {
 346                         [0] = {
 347                                 .integer.type = ACPI_TYPE_INTEGER,
 348                                 .integer.value = offset,
 349                         },
 350                         [1] = {
 351                                 .integer.type = ACPI_TYPE_INTEGER,
 352                                 .integer.value = len,
 353                         },
 354                 },
 355         };
 356
 357         rc = acpi_evaluate_object(handle, "_LSR", &input, &buf);
 358         if (ACPI_FAILURE(rc))
 359                 return NULL;
 360         return pkg_to_buf(buf.pointer);
 361 }
 362
 363 static union acpi_object *acpi_label_info(acpi_handle handle)
 364 {
 365         acpi_status rc;
 366         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
 367
 368         rc = acpi_evaluate_object(handle, "_LSI", NULL, &buf);
 369         if (ACPI_FAILURE(rc))
 370                 return NULL;
 371         return pkg_to_buf(buf.pointer);
 372 }
 373
 374 static u8 nfit_dsm_revid(unsigned family, unsigned func)
 375 {
 376         static const u8 revid_table[NVDIMM_FAMILY_MAX+1][NVDIMM_CMD_MAX+1] = {
 377                 [NVDIMM_FAMILY_INTEL] = {
 378                         [NVDIMM_INTEL_GET_MODES ...
 379                                 NVDIMM_INTEL_FW_ACTIVATE_ARM] = 2,
 380                 },
 381         };
 382         u8 id;
 383
 384         if (family > NVDIMM_FAMILY_MAX)
 385                 return 0;
 386         if (func > NVDIMM_CMD_MAX)
 387                 return 0;
 388         id = revid_table[family][func];
 389         if (id == 0)
 390                 return 1; /* default */
 391         return id;
 392 }
 393
 394 static bool payload_dumpable(struct nvdimm *nvdimm, unsigned int func)
 395 {
 396         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 397
 398         if (nfit_mem && nfit_mem->family == NVDIMM_FAMILY_INTEL
 399                         && func >= NVDIMM_INTEL_GET_SECURITY_STATE
 400                         && func <= NVDIMM_INTEL_MASTER_SECURE_ERASE)
 401                 return IS_ENABLED(CONFIG_NFIT_SECURITY_DEBUG);
 402         return true;
 403 }
 404
 405 static int cmd_to_func(struct nfit_mem *nfit_mem, unsigned int cmd,
 406                 struct nd_cmd_pkg *call_pkg, int *family)
 407 {
 408         if (call_pkg) {
 409                 int i;
 410
 411                 if (nfit_mem && nfit_mem->family != call_pkg->nd_family)
 412                         return -ENOTTY;
 413
 414                 for (i = 0; i < ARRAY_SIZE(call_pkg->nd_reserved2); i++)
 415                         if (call_pkg->nd_reserved2[i])
 416                                 return -EINVAL;
 417                 *family = call_pkg->nd_family;
 418                 return call_pkg->nd_command;
 419         }
 420
 421         /* In the !call_pkg case, bus commands == bus functions */
 422         if (!nfit_mem)
 423                 return cmd;
 424
 425         /* Linux ND commands == NVDIMM_FAMILY_INTEL function numbers */
 426         if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
 427                 return cmd;
 428
 429         /*
 430          * Force function number validation to fail since 0 is never
 431          * published as a valid function in dsm_mask.
 432          */
 433         return 0;
 434 }
 435
 436 int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm,
 437                 unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc)
 438 {
 439         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
 440         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 441         union acpi_object in_obj, in_buf, *out_obj;
 442         const struct nd_cmd_desc *desc = NULL;
 443         struct device *dev = acpi_desc->dev;
 444         struct nd_cmd_pkg *call_pkg = NULL;
 445         const char *cmd_name, *dimm_name;
 446         unsigned long cmd_mask, dsm_mask;
 447         u32 offset, fw_status = 0;
 448         acpi_handle handle;
 449         const guid_t *guid;
 450         int func, rc, i;
 451         int family = 0;
 452
 453         if (cmd_rc)
 454                 *cmd_rc = -EINVAL;
 455
 456         if (cmd == ND_CMD_CALL)
 457                 call_pkg = buf;
 458         func = cmd_to_func(nfit_mem, cmd, call_pkg, &family);
 459         if (func < 0)
 460                 return func;
 461
 462         if (nvdimm) {
 463                 struct acpi_device *adev = nfit_mem->adev;
 464
 465                 if (!adev)
 466                         return -ENOTTY;
 467
 468                 dimm_name = nvdimm_name(nvdimm);
 469                 cmd_name = nvdimm_cmd_name(cmd);
 470                 cmd_mask = nvdimm_cmd_mask(nvdimm);
 471                 dsm_mask = nfit_mem->dsm_mask;
 472                 desc = nd_cmd_dimm_desc(cmd);
 473                 guid = to_nfit_uuid(nfit_mem->family);
 474                 handle = adev->handle;
 475         } else {
 476                 struct acpi_device *adev = to_acpi_dev(acpi_desc);
 477
 478                 cmd_name = nvdimm_bus_cmd_name(cmd);
 479                 cmd_mask = nd_desc->cmd_mask;
 480                 if (cmd == ND_CMD_CALL && call_pkg->nd_family) {
 481                         family = call_pkg->nd_family;
 482                         if (family > NVDIMM_BUS_FAMILY_MAX ||
 483                             !test_bit(family, &nd_desc->bus_family_mask))
 484                                 return -EINVAL;
 485                         family = array_index_nospec(family,
 486                                                     NVDIMM_BUS_FAMILY_MAX + 1);
 487                         dsm_mask = acpi_desc->family_dsm_mask[family];
 488                         guid = to_nfit_bus_uuid(family);
 489                 } else {
 490                         dsm_mask = acpi_desc->bus_dsm_mask;
 491                         guid = to_nfit_uuid(NFIT_DEV_BUS);
 492                 }
 493                 desc = nd_cmd_bus_desc(cmd);
 494                 handle = adev->handle;
 495                 dimm_name = "bus";
 496         }
 497
 498         if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
 499                 return -ENOTTY;
 500
 501         /*
 502          * Check for a valid command.  For ND_CMD_CALL, we also have to
 503          * make sure that the DSM function is supported.
 504          */
 505         if (cmd == ND_CMD_CALL &&
 506             (func > NVDIMM_CMD_MAX || !test_bit(func, &dsm_mask)))
 507                 return -ENOTTY;
 508         else if (!test_bit(cmd, &cmd_mask))
 509                 return -ENOTTY;
 510
 511         in_obj.type = ACPI_TYPE_PACKAGE;
 512         in_obj.package.count = 1;
 513         in_obj.package.elements = &in_buf;
 514         in_buf.type = ACPI_TYPE_BUFFER;
 515         in_buf.buffer.pointer = buf;
 516         in_buf.buffer.length = 0;
 517
 518         /* libnvdimm has already validated the input envelope */
 519         for (i = 0; i < desc->in_num; i++)
 520                 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
 521                                 i, buf);
 522
 523         if (call_pkg) {
 524                 /* skip over package wrapper */
 525                 in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
 526                 in_buf.buffer.length = call_pkg->nd_size_in;
 527         }
 528
 529         dev_dbg(dev, "%s cmd: %d: family: %d func: %d input length: %d\n",
 530                 dimm_name, cmd, family, func, in_buf.buffer.length);
 531         if (payload_dumpable(nvdimm, func))
 532                 print_hex_dump_debug("nvdimm in  ", DUMP_PREFIX_OFFSET, 4, 4,
 533                                 in_buf.buffer.pointer,
 534                                 min_t(u32, 256, in_buf.buffer.length), true);
 535
 536         /* call the BIOS, prefer the named methods over _DSM if available */
 537         if (nvdimm && cmd == ND_CMD_GET_CONFIG_SIZE
 538                         && test_bit(NFIT_MEM_LSR, &nfit_mem->flags))
 539                 out_obj = acpi_label_info(handle);
 540         else if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA
 541                         && test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
 542                 struct nd_cmd_get_config_data_hdr *p = buf;
 543
 544                 out_obj = acpi_label_read(handle, p->in_offset, p->in_length);
 545         } else if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA
 546                         && test_bit(NFIT_MEM_LSW, &nfit_mem->flags)) {
 547                 struct nd_cmd_set_config_hdr *p = buf;
 548
 549                 out_obj = acpi_label_write(handle, p->in_offset, p->in_length,
 550                                 p->in_buf);
 551         } else {
 552                 u8 revid;
 553
 554                 if (nvdimm)
 555                         revid = nfit_dsm_revid(nfit_mem->family, func);
 556                 else
 557                         revid = 1;
 558                 out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
 559         }
 560
 561         if (!out_obj) {
 562                 dev_dbg(dev, "%s _DSM failed cmd: %s\n", dimm_name, cmd_name);
 563                 return -EINVAL;
 564         }
 565
 566         if (out_obj->type != ACPI_TYPE_BUFFER) {
 567                 dev_dbg(dev, "%s unexpected output object type cmd: %s type: %d\n",
 568                                 dimm_name, cmd_name, out_obj->type);
 569                 rc = -EINVAL;
 570                 goto out;
 571         }
 572
 573         dev_dbg(dev, "%s cmd: %s output length: %d\n", dimm_name,
 574                         cmd_name, out_obj->buffer.length);
 575         print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4, 4,
 576                         out_obj->buffer.pointer,
 577                         min_t(u32, 128, out_obj->buffer.length), true);
 578
 579         if (call_pkg) {
 580                 call_pkg->nd_fw_size = out_obj->buffer.length;
 581                 memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
 582                         out_obj->buffer.pointer,
 583                         min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
 584
 585                 ACPI_FREE(out_obj);
 586                 /*
 587                  * Need to support FW function w/o known size in advance.
 588                  * Caller can determine required size based upon nd_fw_size.
 589                  * If we return an error (like elsewhere) then caller wouldn't
 590                  * be able to rely upon data returned to make calculation.
 591                  */
 592                 if (cmd_rc)
 593                         *cmd_rc = 0;
 594                 return 0;
 595         }
 596
 597         for (i = 0, offset = 0; i < desc->out_num; i++) {
 598                 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
 599                                 (u32 *) out_obj->buffer.pointer,
 600                                 out_obj->buffer.length - offset);
 601
 602                 if (offset + out_size > out_obj->buffer.length) {
 603                         dev_dbg(dev, "%s output object underflow cmd: %s field: %d\n",
 604                                         dimm_name, cmd_name, i);
 605                         break;
 606                 }
 607
 608                 if (in_buf.buffer.length + offset + out_size > buf_len) {
 609                         dev_dbg(dev, "%s output overrun cmd: %s field: %d\n",
 610                                         dimm_name, cmd_name, i);
 611                         rc = -ENXIO;
 612                         goto out;
 613                 }
 614                 memcpy(buf + in_buf.buffer.length + offset,
 615                                 out_obj->buffer.pointer + offset, out_size);
 616                 offset += out_size;
 617         }
 618
 619         /*
 620          * Set fw_status for all the commands with a known format to be
 621          * later interpreted by xlat_status().
 622          */
 623         if (i >= 1 && ((!nvdimm && cmd >= ND_CMD_ARS_CAP
 624                                         && cmd <= ND_CMD_CLEAR_ERROR)
 625                                 || (nvdimm && cmd >= ND_CMD_SMART
 626                                         && cmd <= ND_CMD_VENDOR)))
 627                 fw_status = *(u32 *) out_obj->buffer.pointer;
 628
 629         if (offset + in_buf.buffer.length < buf_len) {
 630                 if (i >= 1) {
 631                         /*
 632                          * status valid, return the number of bytes left
 633                          * unfilled in the output buffer
 634                          */
 635                         rc = buf_len - offset - in_buf.buffer.length;
 636                         if (cmd_rc)
 637                                 *cmd_rc = xlat_status(nvdimm, buf, cmd,
 638                                                 fw_status);
 639                 } else {
 640                         dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
 641                                         __func__, dimm_name, cmd_name, buf_len,
 642                                         offset);
 643                         rc = -ENXIO;
 644                 }
 645         } else {
 646                 rc = 0;
 647                 if (cmd_rc)
 648                         *cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status);
 649         }
 650
 651  out:
 652         ACPI_FREE(out_obj);
 653
 654         return rc;
 655 }
 656 EXPORT_SYMBOL_GPL(acpi_nfit_ctl);
 657
 658 static const char *spa_type_name(u16 type)
 659 {
 660         static const char *to_name[] = {
 661                 [NFIT_SPA_VOLATILE] = "volatile",
 662                 [NFIT_SPA_PM] = "pmem",
 663                 [NFIT_SPA_DCR] = "dimm-control-region",
 664                 [NFIT_SPA_BDW] = "block-data-window",
 665                 [NFIT_SPA_VDISK] = "volatile-disk",
 666                 [NFIT_SPA_VCD] = "volatile-cd",
 667                 [NFIT_SPA_PDISK] = "persistent-disk",
 668                 [NFIT_SPA_PCD] = "persistent-cd",
 669
 670         };
 671
 672         if (type > NFIT_SPA_PCD)
 673                 return "unknown";
 674
 675         return to_name[type];
 676 }
 677
 678 int nfit_spa_type(struct acpi_nfit_system_address *spa)
 679 {
 680         int i;
 681
 682         for (i = 0; i < NFIT_UUID_MAX; i++)
 683                 if (guid_equal(to_nfit_uuid(i), (guid_t *)&spa->range_guid))
 684                         return i;
 685         return -1;
 686 }
 687
 688 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
 689                 struct nfit_table_prev *prev,
 690                 struct acpi_nfit_system_address *spa)
 691 {
 692         struct device *dev = acpi_desc->dev;
 693         struct nfit_spa *nfit_spa;
 694
 695         if (spa->header.length != sizeof(*spa))
 696                 return false;
 697
 698         list_for_each_entry(nfit_spa, &prev->spas, list) {
 699                 if (memcmp(nfit_spa->spa, spa, sizeof(*spa)) == 0) {
 700                         list_move_tail(&nfit_spa->list, &acpi_desc->spas);
 701                         return true;
 702                 }
 703         }
 704
 705         nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof(*spa),
 706                         GFP_KERNEL);
 707         if (!nfit_spa)
 708                 return false;
 709         INIT_LIST_HEAD(&nfit_spa->list);
 710         memcpy(nfit_spa->spa, spa, sizeof(*spa));
 711         list_add_tail(&nfit_spa->list, &acpi_desc->spas);
 712         dev_dbg(dev, "spa index: %d type: %s\n",
 713                         spa->range_index,
 714                         spa_type_name(nfit_spa_type(spa)));
 715         return true;
 716 }
 717
 718 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
 719                 struct nfit_table_prev *prev,
 720                 struct acpi_nfit_memory_map *memdev)
 721 {
 722         struct device *dev = acpi_desc->dev;
 723         struct nfit_memdev *nfit_memdev;
 724
 725         if (memdev->header.length != sizeof(*memdev))
 726                 return false;
 727
 728         list_for_each_entry(nfit_memdev, &prev->memdevs, list)
 729                 if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
 730                         list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 731                         return true;
 732                 }
 733
 734         nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
 735                         GFP_KERNEL);
 736         if (!nfit_memdev)
 737                 return false;
 738         INIT_LIST_HEAD(&nfit_memdev->list);
 739         memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
 740         list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 741         dev_dbg(dev, "memdev handle: %#x spa: %d dcr: %d flags: %#x\n",
 742                         memdev->device_handle, memdev->range_index,
 743                         memdev->region_index, memdev->flags);
 744         return true;
 745 }
 746
 747 int nfit_get_smbios_id(u32 device_handle, u16 *flags)
 748 {
 749         struct acpi_nfit_memory_map *memdev;
 750         struct acpi_nfit_desc *acpi_desc;
 751         struct nfit_mem *nfit_mem;
 752         u16 physical_id;
 753
 754         mutex_lock(&acpi_desc_lock);
 755         list_for_each_entry(acpi_desc, &acpi_descs, list) {
 756                 mutex_lock(&acpi_desc->init_mutex);
 757                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
 758                         memdev = __to_nfit_memdev(nfit_mem);
 759                         if (memdev->device_handle == device_handle) {
 760                                 *flags = memdev->flags;
 761                                 physical_id = memdev->physical_id;
 762                                 mutex_unlock(&acpi_desc->init_mutex);
 763                                 mutex_unlock(&acpi_desc_lock);
 764                                 return physical_id;
 765                         }
 766                 }
 767                 mutex_unlock(&acpi_desc->init_mutex);
 768         }
 769         mutex_unlock(&acpi_desc_lock);
 770
 771         return -ENODEV;
 772 }
 773 EXPORT_SYMBOL_GPL(nfit_get_smbios_id);
 774
 775 /*
 776  * An implementation may provide a truncated control region if no block windows
 777  * are defined.
 778  */
 779 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
 780 {
 781         if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
 782                                 window_size))
 783                 return 0;
 784         if (dcr->windows)
 785                 return sizeof(*dcr);
 786         return offsetof(struct acpi_nfit_control_region, window_size);
 787 }
 788
 789 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
 790                 struct nfit_table_prev *prev,
 791                 struct acpi_nfit_control_region *dcr)
 792 {
 793         struct device *dev = acpi_desc->dev;
 794         struct nfit_dcr *nfit_dcr;
 795
 796         if (!sizeof_dcr(dcr))
 797                 return false;
 798
 799         list_for_each_entry(nfit_dcr, &prev->dcrs, list)
 800                 if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
 801                         list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 802                         return true;
 803                 }
 804
 805         nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
 806                         GFP_KERNEL);
 807         if (!nfit_dcr)
 808                 return false;
 809         INIT_LIST_HEAD(&nfit_dcr->list);
 810         memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
 811         list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 812         dev_dbg(dev, "dcr index: %d windows: %d\n",
 813                         dcr->region_index, dcr->windows);
 814         return true;
 815 }
 816
 817 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
 818                 struct nfit_table_prev *prev,
 819                 struct acpi_nfit_data_region *bdw)
 820 {
 821         struct device *dev = acpi_desc->dev;
 822         struct nfit_bdw *nfit_bdw;
 823
 824         if (bdw->header.length != sizeof(*bdw))
 825                 return false;
 826         list_for_each_entry(nfit_bdw, &prev->bdws, list)
 827                 if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
 828                         list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
 829                         return true;
 830                 }
 831
 832         nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
 833                         GFP_KERNEL);
 834         if (!nfit_bdw)
 835                 return false;
 836         INIT_LIST_HEAD(&nfit_bdw->list);
 837         memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
 838         list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
 839         dev_dbg(dev, "bdw dcr: %d windows: %d\n",
 840                         bdw->region_index, bdw->windows);
 841         return true;
 842 }
 843
 844 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
 845 {
 846         if (idt->header.length < sizeof(*idt))
 847                 return 0;
 848         return sizeof(*idt) + sizeof(u32) * (idt->line_count - 1);
 849 }
 850
 851 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
 852                 struct nfit_table_prev *prev,
 853                 struct acpi_nfit_interleave *idt)
 854 {
 855         struct device *dev = acpi_desc->dev;
 856         struct nfit_idt *nfit_idt;
 857
 858         if (!sizeof_idt(idt))
 859                 return false;
 860
 861         list_for_each_entry(nfit_idt, &prev->idts, list) {
 862                 if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
 863                         continue;
 864
 865                 if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
 866                         list_move_tail(&nfit_idt->list, &acpi_desc->idts);
 867                         return true;
 868                 }
 869         }
 870
 871         nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
 872                         GFP_KERNEL);
 873         if (!nfit_idt)
 874                 return false;
 875         INIT_LIST_HEAD(&nfit_idt->list);
 876         memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
 877         list_add_tail(&nfit_idt->list, &acpi_desc->idts);
 878         dev_dbg(dev, "idt index: %d num_lines: %d\n",
 879                         idt->interleave_index, idt->line_count);
 880         return true;
 881 }
 882
 883 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
 884 {
 885         if (flush->header.length < sizeof(*flush))
 886                 return 0;
 887         return sizeof(*flush) + sizeof(u64) * (flush->hint_count - 1);
 888 }
 889
 890 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
 891                 struct nfit_table_prev *prev,
 892                 struct acpi_nfit_flush_address *flush)
 893 {
 894         struct device *dev = acpi_desc->dev;
 895         struct nfit_flush *nfit_flush;
 896
 897         if (!sizeof_flush(flush))
 898                 return false;
 899
 900         list_for_each_entry(nfit_flush, &prev->flushes, list) {
 901                 if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
 902                         continue;
 903
 904                 if (memcmp(nfit_flush->flush, flush,
 905                                         sizeof_flush(flush)) == 0) {
 906                         list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
 907                         return true;
 908                 }
 909         }
 910
 911         nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
 912                         + sizeof_flush(flush), GFP_KERNEL);
 913         if (!nfit_flush)
 914                 return false;
 915         INIT_LIST_HEAD(&nfit_flush->list);
 916         memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
 917         list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
 918         dev_dbg(dev, "nfit_flush handle: %d hint_count: %d\n",
 919                         flush->device_handle, flush->hint_count);
 920         return true;
 921 }
 922
 923 static bool add_platform_cap(struct acpi_nfit_desc *acpi_desc,
 924                 struct acpi_nfit_capabilities *pcap)
 925 {
 926         struct device *dev = acpi_desc->dev;
 927         u32 mask;
 928
 929         mask = (1 << (pcap->highest_capability + 1)) - 1;
 930         acpi_desc->platform_cap = pcap->capabilities & mask;
 931         dev_dbg(dev, "cap: %#x\n", acpi_desc->platform_cap);
 932         return true;
 933 }
 934
 935 static void *add_table(struct acpi_nfit_desc *acpi_desc,
 936                 struct nfit_table_prev *prev, void *table, const void *end)
 937 {
 938         struct device *dev = acpi_desc->dev;
 939         struct acpi_nfit_header *hdr;
 940         void *err = ERR_PTR(-ENOMEM);
 941
 942         if (table >= end)
 943                 return NULL;
 944
 945         hdr = table;
 946         if (!hdr->length) {
 947                 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
 948                         hdr->type);
 949                 return NULL;
 950         }
 951
 952         switch (hdr->type) {
 953         case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
 954                 if (!add_spa(acpi_desc, prev, table))
 955                         return err;
 956                 break;
 957         case ACPI_NFIT_TYPE_MEMORY_MAP:
 958                 if (!add_memdev(acpi_desc, prev, table))
 959                         return err;
 960                 break;
 961         case ACPI_NFIT_TYPE_CONTROL_REGION:
 962                 if (!add_dcr(acpi_desc, prev, table))
 963                         return err;
 964                 break;
 965         case ACPI_NFIT_TYPE_DATA_REGION:
 966                 if (!add_bdw(acpi_desc, prev, table))
 967                         return err;
 968                 break;
 969         case ACPI_NFIT_TYPE_INTERLEAVE:
 970                 if (!add_idt(acpi_desc, prev, table))
 971                         return err;
 972                 break;
 973         case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
 974                 if (!add_flush(acpi_desc, prev, table))
 975                         return err;
 976                 break;
 977         case ACPI_NFIT_TYPE_SMBIOS:
 978                 dev_dbg(dev, "smbios\n");
 979                 break;
 980         case ACPI_NFIT_TYPE_CAPABILITIES:
 981                 if (!add_platform_cap(acpi_desc, table))
 982                         return err;
 983                 break;
 984         default:
 985                 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
 986                 break;
 987         }
 988
 989         return table + hdr->length;
 990 }
 991
 992 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
 993                 struct nfit_mem *nfit_mem)
 994 {
 995         u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
 996         u16 dcr = nfit_mem->dcr->region_index;
 997         struct nfit_spa *nfit_spa;
 998
 999         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
1000                 u16 range_index = nfit_spa->spa->range_index;
1001                 int type = nfit_spa_type(nfit_spa->spa);
1002                 struct nfit_memdev *nfit_memdev;
1003
1004                 if (type != NFIT_SPA_BDW)
1005                         continue;
1006
1007                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1008                         if (nfit_memdev->memdev->range_index != range_index)
1009                                 continue;
1010                         if (nfit_memdev->memdev->device_handle != device_handle)
1011                                 continue;
1012                         if (nfit_memdev->memdev->region_index != dcr)
1013                                 continue;
1014
1015                         nfit_mem->spa_bdw = nfit_spa->spa;
1016                         return;
1017                 }
1018         }
1019
1020         dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
1021                         nfit_mem->spa_dcr->range_index);
1022         nfit_mem->bdw = NULL;
1023 }
1024
1025 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
1026                 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
1027 {
1028         u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
1029         struct nfit_memdev *nfit_memdev;
1030         struct nfit_bdw *nfit_bdw;
1031         struct nfit_idt *nfit_idt;
1032         u16 idt_idx, range_index;
1033
1034         list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
1035                 if (nfit_bdw->bdw->region_index != dcr)
1036                         continue;
1037                 nfit_mem->bdw = nfit_bdw->bdw;
1038                 break;
1039         }
1040
1041         if (!nfit_mem->bdw)
1042                 return;
1043
1044         nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
1045
1046         if (!nfit_mem->spa_bdw)
1047                 return;
1048
1049         range_index = nfit_mem->spa_bdw->range_index;
1050         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1051                 if (nfit_memdev->memdev->range_index != range_index ||
1052                                 nfit_memdev->memdev->region_index != dcr)
1053                         continue;
1054                 nfit_mem->memdev_bdw = nfit_memdev->memdev;
1055                 idt_idx = nfit_memdev->memdev->interleave_index;
1056                 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1057                         if (nfit_idt->idt->interleave_index != idt_idx)
1058                                 continue;
1059                         nfit_mem->idt_bdw = nfit_idt->idt;
1060                         break;
1061                 }
1062                 break;
1063         }
1064 }
1065
1066 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
1067                 struct acpi_nfit_system_address *spa)
1068 {
1069         struct nfit_mem *nfit_mem, *found;
1070         struct nfit_memdev *nfit_memdev;
1071         int type = spa ? nfit_spa_type(spa) : 0;
1072
1073         switch (type) {
1074         case NFIT_SPA_DCR:
1075         case NFIT_SPA_PM:
1076                 break;
1077         default:
1078                 if (spa)
1079                         return 0;
1080         }
1081
1082         /*
1083          * This loop runs in two modes, when a dimm is mapped the loop
1084          * adds memdev associations to an existing dimm, or creates a
1085          * dimm. In the unmapped dimm case this loop sweeps for memdev
1086          * instances with an invalid / zero range_index and adds those
1087          * dimms without spa associations.
1088          */
1089         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1090                 struct nfit_flush *nfit_flush;
1091                 struct nfit_dcr *nfit_dcr;
1092                 u32 device_handle;
1093                 u16 dcr;
1094
1095                 if (spa && nfit_memdev->memdev->range_index != spa->range_index)
1096                         continue;
1097                 if (!spa && nfit_memdev->memdev->range_index)
1098                         continue;
1099                 found = NULL;
1100                 dcr = nfit_memdev->memdev->region_index;
1101                 device_handle = nfit_memdev->memdev->device_handle;
1102                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1103                         if (__to_nfit_memdev(nfit_mem)->device_handle
1104                                         == device_handle) {
1105                                 found = nfit_mem;
1106                                 break;
1107                         }
1108
1109                 if (found)
1110                         nfit_mem = found;
1111                 else {
1112                         nfit_mem = devm_kzalloc(acpi_desc->dev,
1113                                         sizeof(*nfit_mem), GFP_KERNEL);
1114                         if (!nfit_mem)
1115                                 return -ENOMEM;
1116                         INIT_LIST_HEAD(&nfit_mem->list);
1117                         nfit_mem->acpi_desc = acpi_desc;
1118                         list_add(&nfit_mem->list, &acpi_desc->dimms);
1119                 }
1120
1121                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1122                         if (nfit_dcr->dcr->region_index != dcr)
1123                                 continue;
1124                         /*
1125                          * Record the control region for the dimm.  For
1126                          * the ACPI 6.1 case, where there are separate
1127                          * control regions for the pmem vs blk
1128                          * interfaces, be sure to record the extended
1129                          * blk details.
1130                          */
1131                         if (!nfit_mem->dcr)
1132                                 nfit_mem->dcr = nfit_dcr->dcr;
1133                         else if (nfit_mem->dcr->windows == 0
1134                                         && nfit_dcr->dcr->windows)
1135                                 nfit_mem->dcr = nfit_dcr->dcr;
1136                         break;
1137                 }
1138
1139                 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
1140                         struct acpi_nfit_flush_address *flush;
1141                         u16 i;
1142
1143                         if (nfit_flush->flush->device_handle != device_handle)
1144                                 continue;
1145                         nfit_mem->nfit_flush = nfit_flush;
1146                         flush = nfit_flush->flush;
1147                         nfit_mem->flush_wpq = devm_kcalloc(acpi_desc->dev,
1148                                         flush->hint_count,
1149                                         sizeof(struct resource),
1150                                         GFP_KERNEL);
1151                         if (!nfit_mem->flush_wpq)
1152                                 return -ENOMEM;
1153                         for (i = 0; i < flush->hint_count; i++) {
1154                                 struct resource *res = &nfit_mem->flush_wpq[i];
1155
1156                                 res->start = flush->hint_address[i];
1157                                 res->end = res->start + 8 - 1;
1158                         }
1159                         break;
1160                 }
1161
1162                 if (dcr && !nfit_mem->dcr) {
1163                         dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
1164                                         spa->range_index, dcr);
1165                         return -ENODEV;
1166                 }
1167
1168                 if (type == NFIT_SPA_DCR) {
1169                         struct nfit_idt *nfit_idt;
1170                         u16 idt_idx;
1171
1172                         /* multiple dimms may share a SPA when interleaved */
1173                         nfit_mem->spa_dcr = spa;
1174                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
1175                         idt_idx = nfit_memdev->memdev->interleave_index;
1176                         list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1177                                 if (nfit_idt->idt->interleave_index != idt_idx)
1178                                         continue;
1179                                 nfit_mem->idt_dcr = nfit_idt->idt;
1180                                 break;
1181                         }
1182                         nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
1183                 } else if (type == NFIT_SPA_PM) {
1184                         /*
1185                          * A single dimm may belong to multiple SPA-PM
1186                          * ranges, record at least one in addition to
1187                          * any SPA-DCR range.
1188                          */
1189                         nfit_mem->memdev_pmem = nfit_memdev->memdev;
1190                 } else
1191                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
1192         }
1193
1194         return 0;
1195 }
1196
1197 static int nfit_mem_cmp(void *priv, const struct list_head *_a,
1198                 const struct list_head *_b)
1199 {
1200         struct nfit_mem *a = container_of(_a, typeof(*a), list);
1201         struct nfit_mem *b = container_of(_b, typeof(*b), list);
1202         u32 handleA, handleB;
1203
1204         handleA = __to_nfit_memdev(a)->device_handle;
1205         handleB = __to_nfit_memdev(b)->device_handle;
1206         if (handleA < handleB)
1207                 return -1;
1208         else if (handleA > handleB)
1209                 return 1;
1210         return 0;
1211 }
1212
1213 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
1214 {
1215         struct nfit_spa *nfit_spa;
1216         int rc;
1217
1218
1219         /*
1220          * For each SPA-DCR or SPA-PMEM address range find its
1221          * corresponding MEMDEV(s).  From each MEMDEV find the
1222          * corresponding DCR.  Then, if we're operating on a SPA-DCR,
1223          * try to find a SPA-BDW and a corresponding BDW that references
1224          * the DCR.  Throw it all into an nfit_mem object.  Note, that
1225          * BDWs are optional.
1226          */
1227         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
1228                 rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
1229                 if (rc)
1230                         return rc;
1231         }
1232
1233         /*
1234          * If a DIMM has failed to be mapped into SPA there will be no
1235          * SPA entries above. Find and register all the unmapped DIMMs
1236          * for reporting and recovery purposes.
1237          */
1238         rc = __nfit_mem_init(acpi_desc, NULL);
1239         if (rc)
1240                 return rc;
1241
1242         list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
1243
1244         return 0;
1245 }
1246
1247 static ssize_t bus_dsm_mask_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, "%#lx\n", acpi_desc->bus_dsm_mask);
1255 }
1256 static struct device_attribute dev_attr_bus_dsm_mask =
1257                 __ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
1258
1259 static ssize_t revision_show(struct device *dev,
1260                 struct device_attribute *attr, char *buf)
1261 {
1262         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1263         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1264         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1265
1266         return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
1267 }
1268 static DEVICE_ATTR_RO(revision);
1269
1270 static ssize_t hw_error_scrub_show(struct device *dev,
1271                 struct device_attribute *attr, char *buf)
1272 {
1273         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1274         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1275         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1276
1277         return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
1278 }
1279
1280 /*
1281  * The 'hw_error_scrub' attribute can have the following values written to it:
1282  * '0': Switch to the default mode where an exception will only insert
1283  *      the address of the memory error into the poison and badblocks lists.
1284  * '1': Enable a full scrub to happen if an exception for a memory error is
1285  *      received.
1286  */
1287 static ssize_t hw_error_scrub_store(struct device *dev,
1288                 struct device_attribute *attr, const char *buf, size_t size)
1289 {
1290         struct nvdimm_bus_descriptor *nd_desc;
1291         ssize_t rc;
1292         long val;
1293
1294         rc = kstrtol(buf, 0, &val);
1295         if (rc)
1296                 return rc;
1297
1298         nfit_device_lock(dev);
1299         nd_desc = dev_get_drvdata(dev);
1300         if (nd_desc) {
1301                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1302
1303                 switch (val) {
1304                 case HW_ERROR_SCRUB_ON:
1305                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
1306                         break;
1307                 case HW_ERROR_SCRUB_OFF:
1308                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
1309                         break;
1310                 default:
1311                         rc = -EINVAL;
1312                         break;
1313                 }
1314         }
1315         nfit_device_unlock(dev);
1316         if (rc)
1317                 return rc;
1318         return size;
1319 }
1320 static DEVICE_ATTR_RW(hw_error_scrub);
1321
1322 /*
1323  * This shows the number of full Address Range Scrubs that have been
1324  * completed since driver load time. Userspace can wait on this using
1325  * select/poll etc. A '+' at the end indicates an ARS is in progress
1326  */
1327 static ssize_t scrub_show(struct device *dev,
1328                 struct device_attribute *attr, char *buf)
1329 {
1330         struct nvdimm_bus_descriptor *nd_desc;
1331         struct acpi_nfit_desc *acpi_desc;
1332         ssize_t rc = -ENXIO;
1333         bool busy;
1334
1335         nfit_device_lock(dev);
1336         nd_desc = dev_get_drvdata(dev);
1337         if (!nd_desc) {
1338                 nfit_device_unlock(dev);
1339                 return rc;
1340         }
1341         acpi_desc = to_acpi_desc(nd_desc);
1342
1343         mutex_lock(&acpi_desc->init_mutex);
1344         busy = test_bit(ARS_BUSY, &acpi_desc->scrub_flags)
1345                 && !test_bit(ARS_CANCEL, &acpi_desc->scrub_flags);
1346         rc = sprintf(buf, "%d%s", acpi_desc->scrub_count, busy ? "+\n" : "\n");
1347         /* Allow an admin to poll the busy state at a higher rate */
1348         if (busy && capable(CAP_SYS_RAWIO) && !test_and_set_bit(ARS_POLL,
1349                                 &acpi_desc->scrub_flags)) {
1350                 acpi_desc->scrub_tmo = 1;
1351                 mod_delayed_work(nfit_wq, &acpi_desc->dwork, HZ);
1352         }
1353
1354         mutex_unlock(&acpi_desc->init_mutex);
1355         nfit_device_unlock(dev);
1356         return rc;
1357 }
1358
1359 static ssize_t scrub_store(struct device *dev,
1360                 struct device_attribute *attr, const char *buf, size_t size)
1361 {
1362         struct nvdimm_bus_descriptor *nd_desc;
1363         ssize_t rc;
1364         long val;
1365
1366         rc = kstrtol(buf, 0, &val);
1367         if (rc)
1368                 return rc;
1369         if (val != 1)
1370                 return -EINVAL;
1371
1372         nfit_device_lock(dev);
1373         nd_desc = dev_get_drvdata(dev);
1374         if (nd_desc) {
1375                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1376
1377                 rc = acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
1378         }
1379         nfit_device_unlock(dev);
1380         if (rc)
1381                 return rc;
1382         return size;
1383 }
1384 static DEVICE_ATTR_RW(scrub);
1385
1386 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1387 {
1388         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1389         const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1390                 | 1 << ND_CMD_ARS_STATUS;
1391
1392         return (nd_desc->cmd_mask & mask) == mask;
1393 }
1394
1395 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1396 {
1397         struct device *dev = kobj_to_dev(kobj);
1398         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1399
1400         if (a == &dev_attr_scrub.attr)
1401                 return ars_supported(nvdimm_bus) ? a->mode : 0;
1402
1403         if (a == &dev_attr_firmware_activate_noidle.attr)
1404                 return intel_fwa_supported(nvdimm_bus) ? a->mode : 0;
1405
1406         return a->mode;
1407 }
1408
1409 static struct attribute *acpi_nfit_attributes[] = {
1410         &dev_attr_revision.attr,
1411         &dev_attr_scrub.attr,
1412         &dev_attr_hw_error_scrub.attr,
1413         &dev_attr_bus_dsm_mask.attr,
1414         &dev_attr_firmware_activate_noidle.attr,
1415         NULL,
1416 };
1417
1418 static const struct attribute_group acpi_nfit_attribute_group = {
1419         .name = "nfit",
1420         .attrs = acpi_nfit_attributes,
1421         .is_visible = nfit_visible,
1422 };
1423
1424 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1425         &acpi_nfit_attribute_group,
1426         NULL,
1427 };
1428
1429 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1430 {
1431         struct nvdimm *nvdimm = to_nvdimm(dev);
1432         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1433
1434         return __to_nfit_memdev(nfit_mem);
1435 }
1436
1437 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1438 {
1439         struct nvdimm *nvdimm = to_nvdimm(dev);
1440         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1441
1442         return nfit_mem->dcr;
1443 }
1444
1445 static ssize_t handle_show(struct device *dev,
1446                 struct device_attribute *attr, char *buf)
1447 {
1448         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1449
1450         return sprintf(buf, "%#x\n", memdev->device_handle);
1451 }
1452 static DEVICE_ATTR_RO(handle);
1453
1454 static ssize_t phys_id_show(struct device *dev,
1455                 struct device_attribute *attr, char *buf)
1456 {
1457         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1458
1459         return sprintf(buf, "%#x\n", memdev->physical_id);
1460 }
1461 static DEVICE_ATTR_RO(phys_id);
1462
1463 static ssize_t vendor_show(struct device *dev,
1464                 struct device_attribute *attr, char *buf)
1465 {
1466         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1467
1468         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1469 }
1470 static DEVICE_ATTR_RO(vendor);
1471
1472 static ssize_t rev_id_show(struct device *dev,
1473                 struct device_attribute *attr, char *buf)
1474 {
1475         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1476
1477         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1478 }
1479 static DEVICE_ATTR_RO(rev_id);
1480
1481 static ssize_t device_show(struct device *dev,
1482                 struct device_attribute *attr, char *buf)
1483 {
1484         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1485
1486         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1487 }
1488 static DEVICE_ATTR_RO(device);
1489
1490 static ssize_t subsystem_vendor_show(struct device *dev,
1491                 struct device_attribute *attr, char *buf)
1492 {
1493         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1494
1495         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1496 }
1497 static DEVICE_ATTR_RO(subsystem_vendor);
1498
1499 static ssize_t subsystem_rev_id_show(struct device *dev,
1500                 struct device_attribute *attr, char *buf)
1501 {
1502         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1503
1504         return sprintf(buf, "0x%04x\n",
1505                         be16_to_cpu(dcr->subsystem_revision_id));
1506 }
1507 static DEVICE_ATTR_RO(subsystem_rev_id);
1508
1509 static ssize_t subsystem_device_show(struct device *dev,
1510                 struct device_attribute *attr, char *buf)
1511 {
1512         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1513
1514         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1515 }
1516 static DEVICE_ATTR_RO(subsystem_device);
1517
1518 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1519 {
1520         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1521         int formats = 0;
1522
1523         if (nfit_mem->memdev_pmem)
1524                 formats++;
1525         if (nfit_mem->memdev_bdw)
1526                 formats++;
1527         return formats;
1528 }
1529
1530 static ssize_t format_show(struct device *dev,
1531                 struct device_attribute *attr, char *buf)
1532 {
1533         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1534
1535         return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1536 }
1537 static DEVICE_ATTR_RO(format);
1538
1539 static ssize_t format1_show(struct device *dev,
1540                 struct device_attribute *attr, char *buf)
1541 {
1542         u32 handle;
1543         ssize_t rc = -ENXIO;
1544         struct nfit_mem *nfit_mem;
1545         struct nfit_memdev *nfit_memdev;
1546         struct acpi_nfit_desc *acpi_desc;
1547         struct nvdimm *nvdimm = to_nvdimm(dev);
1548         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1549
1550         nfit_mem = nvdimm_provider_data(nvdimm);
1551         acpi_desc = nfit_mem->acpi_desc;
1552         handle = to_nfit_memdev(dev)->device_handle;
1553
1554         /* assumes DIMMs have at most 2 published interface codes */
1555         mutex_lock(&acpi_desc->init_mutex);
1556         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1557                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1558                 struct nfit_dcr *nfit_dcr;
1559
1560                 if (memdev->device_handle != handle)
1561                         continue;
1562
1563                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1564                         if (nfit_dcr->dcr->region_index != memdev->region_index)
1565                                 continue;
1566                         if (nfit_dcr->dcr->code == dcr->code)
1567                                 continue;
1568                         rc = sprintf(buf, "0x%04x\n",
1569                                         le16_to_cpu(nfit_dcr->dcr->code));
1570                         break;
1571                 }
1572                 if (rc != -ENXIO)
1573                         break;
1574         }
1575         mutex_unlock(&acpi_desc->init_mutex);
1576         return rc;
1577 }
1578 static DEVICE_ATTR_RO(format1);
1579
1580 static ssize_t formats_show(struct device *dev,
1581                 struct device_attribute *attr, char *buf)
1582 {
1583         struct nvdimm *nvdimm = to_nvdimm(dev);
1584
1585         return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1586 }
1587 static DEVICE_ATTR_RO(formats);
1588
1589 static ssize_t serial_show(struct device *dev,
1590                 struct device_attribute *attr, char *buf)
1591 {
1592         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1593
1594         return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1595 }
1596 static DEVICE_ATTR_RO(serial);
1597
1598 static ssize_t family_show(struct device *dev,
1599                 struct device_attribute *attr, char *buf)
1600 {
1601         struct nvdimm *nvdimm = to_nvdimm(dev);
1602         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1603
1604         if (nfit_mem->family < 0)
1605                 return -ENXIO;
1606         return sprintf(buf, "%d\n", nfit_mem->family);
1607 }
1608 static DEVICE_ATTR_RO(family);
1609
1610 static ssize_t dsm_mask_show(struct device *dev,
1611                 struct device_attribute *attr, char *buf)
1612 {
1613         struct nvdimm *nvdimm = to_nvdimm(dev);
1614         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1615
1616         if (nfit_mem->family < 0)
1617                 return -ENXIO;
1618         return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1619 }
1620 static DEVICE_ATTR_RO(dsm_mask);
1621
1622 static ssize_t flags_show(struct device *dev,
1623                 struct device_attribute *attr, char *buf)
1624 {
1625         struct nvdimm *nvdimm = to_nvdimm(dev);
1626         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1627         u16 flags = __to_nfit_memdev(nfit_mem)->flags;
1628
1629         if (test_bit(NFIT_MEM_DIRTY, &nfit_mem->flags))
1630                 flags |= ACPI_NFIT_MEM_FLUSH_FAILED;
1631
1632         return sprintf(buf, "%s%s%s%s%s%s%s\n",
1633                 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1634                 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1635                 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1636                 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1637                 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1638                 flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1639                 flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1640 }
1641 static DEVICE_ATTR_RO(flags);
1642
1643 static ssize_t id_show(struct device *dev,
1644                 struct device_attribute *attr, char *buf)
1645 {
1646         struct nvdimm *nvdimm = to_nvdimm(dev);
1647         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1648
1649         return sprintf(buf, "%s\n", nfit_mem->id);
1650 }
1651 static DEVICE_ATTR_RO(id);
1652
1653 static ssize_t dirty_shutdown_show(struct device *dev,
1654                 struct device_attribute *attr, char *buf)
1655 {
1656         struct nvdimm *nvdimm = to_nvdimm(dev);
1657         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1658
1659         return sprintf(buf, "%d\n", nfit_mem->dirty_shutdown);
1660 }
1661 static DEVICE_ATTR_RO(dirty_shutdown);
1662
1663 static struct attribute *acpi_nfit_dimm_attributes[] = {
1664         &dev_attr_handle.attr,
1665         &dev_attr_phys_id.attr,
1666         &dev_attr_vendor.attr,
1667         &dev_attr_device.attr,
1668         &dev_attr_rev_id.attr,
1669         &dev_attr_subsystem_vendor.attr,
1670         &dev_attr_subsystem_device.attr,
1671         &dev_attr_subsystem_rev_id.attr,
1672         &dev_attr_format.attr,
1673         &dev_attr_formats.attr,
1674         &dev_attr_format1.attr,
1675         &dev_attr_serial.attr,
1676         &dev_attr_flags.attr,
1677         &dev_attr_id.attr,
1678         &dev_attr_family.attr,
1679         &dev_attr_dsm_mask.attr,
1680         &dev_attr_dirty_shutdown.attr,
1681         NULL,
1682 };
1683
1684 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1685                 struct attribute *a, int n)
1686 {
1687         struct device *dev = kobj_to_dev(kobj);
1688         struct nvdimm *nvdimm = to_nvdimm(dev);
1689         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1690
1691         if (!to_nfit_dcr(dev)) {
1692                 /* Without a dcr only the memdev attributes can be surfaced */
1693                 if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1694                                 || a == &dev_attr_flags.attr
1695                                 || a == &dev_attr_family.attr
1696                                 || a == &dev_attr_dsm_mask.attr)
1697                         return a->mode;
1698                 return 0;
1699         }
1700
1701         if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1702                 return 0;
1703
1704         if (!test_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags)
1705                         && a == &dev_attr_dirty_shutdown.attr)
1706                 return 0;
1707
1708         return a->mode;
1709 }
1710
1711 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1712         .name = "nfit",
1713         .attrs = acpi_nfit_dimm_attributes,
1714         .is_visible = acpi_nfit_dimm_attr_visible,
1715 };
1716
1717 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1718         &acpi_nfit_dimm_attribute_group,
1719         NULL,
1720 };
1721
1722 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1723                 u32 device_handle)
1724 {
1725         struct nfit_mem *nfit_mem;
1726
1727         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1728                 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1729                         return nfit_mem->nvdimm;
1730
1731         return NULL;
1732 }
1733
1734 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1735 {
1736         struct nfit_mem *nfit_mem;
1737         struct acpi_nfit_desc *acpi_desc;
1738
1739         dev_dbg(dev->parent, "%s: event: %d\n", dev_name(dev),
1740                         event);
1741
1742         if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1743                 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1744                                 event);
1745                 return;
1746         }
1747
1748         acpi_desc = dev_get_drvdata(dev->parent);
1749         if (!acpi_desc)
1750                 return;
1751
1752         /*
1753          * If we successfully retrieved acpi_desc, then we know nfit_mem data
1754          * is still valid.
1755          */
1756         nfit_mem = dev_get_drvdata(dev);
1757         if (nfit_mem && nfit_mem->flags_attr)
1758                 sysfs_notify_dirent(nfit_mem->flags_attr);
1759 }
1760 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1761
1762 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1763 {
1764         struct acpi_device *adev = data;
1765         struct device *dev = &adev->dev;
1766
1767         nfit_device_lock(dev->parent);
1768         __acpi_nvdimm_notify(dev, event);
1769         nfit_device_unlock(dev->parent);
1770 }
1771
1772 static bool acpi_nvdimm_has_method(struct acpi_device *adev, char *method)
1773 {
1774         acpi_handle handle;
1775         acpi_status status;
1776
1777         status = acpi_get_handle(adev->handle, method, &handle);
1778
1779         if (ACPI_SUCCESS(status))
1780                 return true;
1781         return false;
1782 }
1783
1784 __weak void nfit_intel_shutdown_status(struct nfit_mem *nfit_mem)
1785 {
1786         struct device *dev = &nfit_mem->adev->dev;
1787         struct nd_intel_smart smart = { 0 };
1788         union acpi_object in_buf = {
1789                 .buffer.type = ACPI_TYPE_BUFFER,
1790                 .buffer.length = 0,
1791         };
1792         union acpi_object in_obj = {
1793                 .package.type = ACPI_TYPE_PACKAGE,
1794                 .package.count = 1,
1795                 .package.elements = &in_buf,
1796         };
1797         const u8 func = ND_INTEL_SMART;
1798         const guid_t *guid = to_nfit_uuid(nfit_mem->family);
1799         u8 revid = nfit_dsm_revid(nfit_mem->family, func);
1800         struct acpi_device *adev = nfit_mem->adev;
1801         acpi_handle handle = adev->handle;
1802         union acpi_object *out_obj;
1803
1804         if ((nfit_mem->dsm_mask & (1 << func)) == 0)
1805                 return;
1806
1807         out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
1808         if (!out_obj || out_obj->type != ACPI_TYPE_BUFFER
1809                         || out_obj->buffer.length < sizeof(smart)) {
1810                 dev_dbg(dev->parent, "%s: failed to retrieve initial health\n",
1811                                 dev_name(dev));
1812                 ACPI_FREE(out_obj);
1813                 return;
1814         }
1815         memcpy(&smart, out_obj->buffer.pointer, sizeof(smart));
1816         ACPI_FREE(out_obj);
1817
1818         if (smart.flags & ND_INTEL_SMART_SHUTDOWN_VALID) {
1819                 if (smart.shutdown_state)
1820                         set_bit(NFIT_MEM_DIRTY, &nfit_mem->flags);
1821         }
1822
1823         if (smart.flags & ND_INTEL_SMART_SHUTDOWN_COUNT_VALID) {
1824                 set_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags);
1825                 nfit_mem->dirty_shutdown = smart.shutdown_count;
1826         }
1827 }
1828
1829 static void populate_shutdown_status(struct nfit_mem *nfit_mem)
1830 {
1831         /*
1832          * For DIMMs that provide a dynamic facility to retrieve a
1833          * dirty-shutdown status and/or a dirty-shutdown count, cache
1834          * these values in nfit_mem.
1835          */
1836         if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
1837                 nfit_intel_shutdown_status(nfit_mem);
1838 }
1839
1840 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1841                 struct nfit_mem *nfit_mem, u32 device_handle)
1842 {
1843         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1844         struct acpi_device *adev, *adev_dimm;
1845         struct device *dev = acpi_desc->dev;
1846         unsigned long dsm_mask, label_mask;
1847         const guid_t *guid;
1848         int i;
1849         int family = -1;
1850         struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
1851
1852         /* nfit test assumes 1:1 relationship between commands and dsms */
1853         nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1854         nfit_mem->family = NVDIMM_FAMILY_INTEL;
1855         set_bit(NVDIMM_FAMILY_INTEL, &nd_desc->dimm_family_mask);
1856
1857         if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1858                 sprintf(nfit_mem->id, "%04x-%02x-%04x-%08x",
1859                                 be16_to_cpu(dcr->vendor_id),
1860                                 dcr->manufacturing_location,
1861                                 be16_to_cpu(dcr->manufacturing_date),
1862                                 be32_to_cpu(dcr->serial_number));
1863         else
1864                 sprintf(nfit_mem->id, "%04x-%08x",
1865                                 be16_to_cpu(dcr->vendor_id),
1866                                 be32_to_cpu(dcr->serial_number));
1867
1868         adev = to_acpi_dev(acpi_desc);
1869         if (!adev) {
1870                 /* unit test case */
1871                 populate_shutdown_status(nfit_mem);
1872                 return 0;
1873         }
1874
1875         adev_dimm = acpi_find_child_device(adev, device_handle, false);
1876         nfit_mem->adev = adev_dimm;
1877         if (!adev_dimm) {
1878                 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1879                                 device_handle);
1880                 return force_enable_dimms ? 0 : -ENODEV;
1881         }
1882
1883         if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1884                 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1885                 dev_err(dev, "%s: notification registration failed\n",
1886                                 dev_name(&adev_dimm->dev));
1887                 return -ENXIO;
1888         }
1889         /*
1890          * Record nfit_mem for the notification path to track back to
1891          * the nfit sysfs attributes for this dimm device object.
1892          */
1893         dev_set_drvdata(&adev_dimm->dev, nfit_mem);
1894
1895         /*
1896          * There are 4 "legacy" NVDIMM command sets
1897          * (NVDIMM_FAMILY_{INTEL,MSFT,HPE1,HPE2}) that were created before
1898          * an EFI working group was established to constrain this
1899          * proliferation. The nfit driver probes for the supported command
1900          * set by GUID. Note, if you're a platform developer looking to add
1901          * a new command set to this probe, consider using an existing set,
1902          * or otherwise seek approval to publish the command set at
1903          * http://www.uefi.org/RFIC_LIST.
1904          *
1905          * Note, that checking for function0 (bit0) tells us if any commands
1906          * are reachable through this GUID.
1907          */
1908         clear_bit(NVDIMM_FAMILY_INTEL, &nd_desc->dimm_family_mask);
1909         for (i = 0; i <= NVDIMM_FAMILY_MAX; i++)
1910                 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1)) {
1911                         set_bit(i, &nd_desc->dimm_family_mask);
1912                         if (family < 0 || i == default_dsm_family)
1913                                 family = i;
1914                 }
1915
1916         /* limit the supported commands to those that are publicly documented */
1917         nfit_mem->family = family;
1918         if (override_dsm_mask && !disable_vendor_specific)
1919                 dsm_mask = override_dsm_mask;
1920         else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1921                 dsm_mask = NVDIMM_INTEL_CMDMASK;
1922                 if (disable_vendor_specific)
1923                         dsm_mask &= ~(1 << ND_CMD_VENDOR);
1924         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1925                 dsm_mask = 0x1c3c76;
1926         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1927                 dsm_mask = 0x1fe;
1928                 if (disable_vendor_specific)
1929                         dsm_mask &= ~(1 << 8);
1930         } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1931                 dsm_mask = 0xffffffff;
1932         } else if (nfit_mem->family == NVDIMM_FAMILY_HYPERV) {
1933                 dsm_mask = 0x1f;
1934         } else {
1935                 dev_dbg(dev, "unknown dimm command family\n");
1936                 nfit_mem->family = -1;
1937                 /* DSMs are optional, continue loading the driver... */
1938                 return 0;
1939         }
1940
1941         /*
1942          * Function 0 is the command interrogation function, don't
1943          * export it to potential userspace use, and enable it to be
1944          * used as an error value in acpi_nfit_ctl().
1945          */
1946         dsm_mask &= ~1UL;
1947
1948         guid = to_nfit_uuid(nfit_mem->family);
1949         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1950                 if (acpi_check_dsm(adev_dimm->handle, guid,
1951                                         nfit_dsm_revid(nfit_mem->family, i),
1952                                         1ULL << i))
1953                         set_bit(i, &nfit_mem->dsm_mask);
1954
1955         /*
1956          * Prefer the NVDIMM_FAMILY_INTEL label read commands if present
1957          * due to their better semantics handling locked capacity.
1958          */
1959         label_mask = 1 << ND_CMD_GET_CONFIG_SIZE | 1 << ND_CMD_GET_CONFIG_DATA
1960                 | 1 << ND_CMD_SET_CONFIG_DATA;
1961         if (family == NVDIMM_FAMILY_INTEL
1962                         && (dsm_mask & label_mask) == label_mask)
1963                 /* skip _LS{I,R,W} enabling */;
1964         else {
1965                 if (acpi_nvdimm_has_method(adev_dimm, "_LSI")
1966                                 && acpi_nvdimm_has_method(adev_dimm, "_LSR")) {
1967                         dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev));
1968                         set_bit(NFIT_MEM_LSR, &nfit_mem->flags);
1969                 }
1970
1971                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
1972                                 && acpi_nvdimm_has_method(adev_dimm, "_LSW")) {
1973                         dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev));
1974                         set_bit(NFIT_MEM_LSW, &nfit_mem->flags);
1975                 }
1976
1977                 /*
1978                  * Quirk read-only label configurations to preserve
1979                  * access to label-less namespaces by default.
1980                  */
1981                 if (!test_bit(NFIT_MEM_LSW, &nfit_mem->flags)
1982                                 && !force_labels) {
1983                         dev_dbg(dev, "%s: No _LSW, disable labels\n",
1984                                         dev_name(&adev_dimm->dev));
1985                         clear_bit(NFIT_MEM_LSR, &nfit_mem->flags);
1986                 } else
1987                         dev_dbg(dev, "%s: Force enable labels\n",
1988                                         dev_name(&adev_dimm->dev));
1989         }
1990
1991         populate_shutdown_status(nfit_mem);
1992
1993         return 0;
1994 }
1995
1996 static void shutdown_dimm_notify(void *data)
1997 {
1998         struct acpi_nfit_desc *acpi_desc = data;
1999         struct nfit_mem *nfit_mem;
2000
2001         mutex_lock(&acpi_desc->init_mutex);
2002         /*
2003          * Clear out the nfit_mem->flags_attr and shut down dimm event
2004          * notifications.
2005          */
2006         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2007                 struct acpi_device *adev_dimm = nfit_mem->adev;
2008
2009                 if (nfit_mem->flags_attr) {
2010                         sysfs_put(nfit_mem->flags_attr);
2011                         nfit_mem->flags_attr = NULL;
2012                 }
2013                 if (adev_dimm) {
2014                         acpi_remove_notify_handler(adev_dimm->handle,
2015                                         ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
2016                         dev_set_drvdata(&adev_dimm->dev, NULL);
2017                 }
2018         }
2019         mutex_unlock(&acpi_desc->init_mutex);
2020 }
2021
2022 static const struct nvdimm_security_ops *acpi_nfit_get_security_ops(int family)
2023 {
2024         switch (family) {
2025         case NVDIMM_FAMILY_INTEL:
2026                 return intel_security_ops;
2027         default:
2028                 return NULL;
2029         }
2030 }
2031
2032 static const struct nvdimm_fw_ops *acpi_nfit_get_fw_ops(
2033                 struct nfit_mem *nfit_mem)
2034 {
2035         unsigned long mask;
2036         struct acpi_nfit_desc *acpi_desc = nfit_mem->acpi_desc;
2037         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2038
2039         if (!nd_desc->fw_ops)
2040                 return NULL;
2041
2042         if (nfit_mem->family != NVDIMM_FAMILY_INTEL)
2043                 return NULL;
2044
2045         mask = nfit_mem->dsm_mask & NVDIMM_INTEL_FW_ACTIVATE_CMDMASK;
2046         if (mask != NVDIMM_INTEL_FW_ACTIVATE_CMDMASK)
2047                 return NULL;
2048
2049         return intel_fw_ops;
2050 }
2051
2052 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
2053 {
2054         struct nfit_mem *nfit_mem;
2055         int dimm_count = 0, rc;
2056         struct nvdimm *nvdimm;
2057
2058         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2059                 struct acpi_nfit_flush_address *flush;
2060                 unsigned long flags = 0, cmd_mask;
2061                 struct nfit_memdev *nfit_memdev;
2062                 u32 device_handle;
2063                 u16 mem_flags;
2064
2065                 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
2066                 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
2067                 if (nvdimm) {
2068                         dimm_count++;
2069                         continue;
2070                 }
2071
2072                 if (nfit_mem->bdw && nfit_mem->memdev_pmem) {
2073                         set_bit(NDD_ALIASING, &flags);
2074                         set_bit(NDD_LABELING, &flags);
2075                 }
2076
2077                 /* collate flags across all memdevs for this dimm */
2078                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2079                         struct acpi_nfit_memory_map *dimm_memdev;
2080
2081                         dimm_memdev = __to_nfit_memdev(nfit_mem);
2082                         if (dimm_memdev->device_handle
2083                                         != nfit_memdev->memdev->device_handle)
2084                                 continue;
2085                         dimm_memdev->flags |= nfit_memdev->memdev->flags;
2086                 }
2087
2088                 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
2089                 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
2090                         set_bit(NDD_UNARMED, &flags);
2091
2092                 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
2093                 if (rc)
2094                         continue;
2095
2096                 /*
2097                  * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
2098                  * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
2099                  * userspace interface.
2100                  */
2101                 cmd_mask = 1UL << ND_CMD_CALL;
2102                 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
2103                         /*
2104                          * These commands have a 1:1 correspondence
2105                          * between DSM payload and libnvdimm ioctl
2106                          * payload format.
2107                          */
2108                         cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
2109                 }
2110
2111                 /* Quirk to ignore LOCAL for labels on HYPERV DIMMs */
2112                 if (nfit_mem->family == NVDIMM_FAMILY_HYPERV)
2113                         set_bit(NDD_NOBLK, &flags);
2114
2115                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
2116                         set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
2117                         set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
2118                 }
2119                 if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags))
2120                         set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
2121
2122                 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
2123                         : NULL;
2124                 nvdimm = __nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
2125                                 acpi_nfit_dimm_attribute_groups,
2126                                 flags, cmd_mask, flush ? flush->hint_count : 0,
2127                                 nfit_mem->flush_wpq, &nfit_mem->id[0],
2128                                 acpi_nfit_get_security_ops(nfit_mem->family),
2129                                 acpi_nfit_get_fw_ops(nfit_mem));
2130                 if (!nvdimm)
2131                         return -ENOMEM;
2132
2133                 nfit_mem->nvdimm = nvdimm;
2134                 dimm_count++;
2135
2136                 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
2137                         continue;
2138
2139                 dev_err(acpi_desc->dev, "Error found in NVDIMM %s flags:%s%s%s%s%s\n",
2140                                 nvdimm_name(nvdimm),
2141                   mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
2142                   mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
2143                   mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
2144                   mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
2145                   mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
2146
2147         }
2148
2149         rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
2150         if (rc)
2151                 return rc;
2152
2153         /*
2154          * Now that dimms are successfully registered, and async registration
2155          * is flushed, attempt to enable event notification.
2156          */
2157         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2158                 struct kernfs_node *nfit_kernfs;
2159
2160                 nvdimm = nfit_mem->nvdimm;
2161                 if (!nvdimm)
2162                         continue;
2163
2164                 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
2165                 if (nfit_kernfs)
2166                         nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
2167                                         "flags");
2168                 sysfs_put(nfit_kernfs);
2169                 if (!nfit_mem->flags_attr)
2170                         dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
2171                                         nvdimm_name(nvdimm));
2172         }
2173
2174         return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
2175                         acpi_desc);
2176 }
2177
2178 /*
2179  * These constants are private because there are no kernel consumers of
2180  * these commands.
2181  */
2182 enum nfit_aux_cmds {
2183         NFIT_CMD_TRANSLATE_SPA = 5,
2184         NFIT_CMD_ARS_INJECT_SET = 7,
2185         NFIT_CMD_ARS_INJECT_CLEAR = 8,
2186         NFIT_CMD_ARS_INJECT_GET = 9,
2187 };
2188
2189 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
2190 {
2191         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2192         const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
2193         unsigned long dsm_mask, *mask;
2194         struct acpi_device *adev;
2195         int i;
2196
2197         set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
2198         set_bit(NVDIMM_BUS_FAMILY_NFIT, &nd_desc->bus_family_mask);
2199
2200         /* enable nfit_test to inject bus command emulation */
2201         if (acpi_desc->bus_cmd_force_en) {
2202                 nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
2203                 mask = &nd_desc->bus_family_mask;
2204                 if (acpi_desc->family_dsm_mask[NVDIMM_BUS_FAMILY_INTEL]) {
2205                         set_bit(NVDIMM_BUS_FAMILY_INTEL, mask);
2206                         nd_desc->fw_ops = intel_bus_fw_ops;
2207                 }
2208         }
2209
2210         adev = to_acpi_dev(acpi_desc);
2211         if (!adev)
2212                 return;
2213
2214         for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
2215                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2216                         set_bit(i, &nd_desc->cmd_mask);
2217
2218         dsm_mask =
2219                 (1 << ND_CMD_ARS_CAP) |
2220                 (1 << ND_CMD_ARS_START) |
2221                 (1 << ND_CMD_ARS_STATUS) |
2222                 (1 << ND_CMD_CLEAR_ERROR) |
2223                 (1 << NFIT_CMD_TRANSLATE_SPA) |
2224                 (1 << NFIT_CMD_ARS_INJECT_SET) |
2225                 (1 << NFIT_CMD_ARS_INJECT_CLEAR) |
2226                 (1 << NFIT_CMD_ARS_INJECT_GET);
2227         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
2228                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2229                         set_bit(i, &acpi_desc->bus_dsm_mask);
2230
2231         /* Enumerate allowed NVDIMM_BUS_FAMILY_INTEL commands */
2232         dsm_mask = NVDIMM_BUS_INTEL_FW_ACTIVATE_CMDMASK;
2233         guid = to_nfit_bus_uuid(NVDIMM_BUS_FAMILY_INTEL);
2234         mask = &acpi_desc->family_dsm_mask[NVDIMM_BUS_FAMILY_INTEL];
2235         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
2236                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2237                         set_bit(i, mask);
2238
2239         if (*mask == dsm_mask) {
2240                 set_bit(NVDIMM_BUS_FAMILY_INTEL, &nd_desc->bus_family_mask);
2241                 nd_desc->fw_ops = intel_bus_fw_ops;
2242         }
2243 }
2244
2245 static ssize_t range_index_show(struct device *dev,
2246                 struct device_attribute *attr, char *buf)
2247 {
2248         struct nd_region *nd_region = to_nd_region(dev);
2249         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
2250
2251         return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
2252 }
2253 static DEVICE_ATTR_RO(range_index);
2254
2255 static struct attribute *acpi_nfit_region_attributes[] = {
2256         &dev_attr_range_index.attr,
2257         NULL,
2258 };
2259
2260 static const struct attribute_group acpi_nfit_region_attribute_group = {
2261         .name = "nfit",
2262         .attrs = acpi_nfit_region_attributes,
2263 };
2264
2265 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
2266         &acpi_nfit_region_attribute_group,
2267         NULL,
2268 };
2269
2270 /* enough info to uniquely specify an interleave set */
2271 struct nfit_set_info {
2272         struct nfit_set_info_map {
2273                 u64 region_offset;
2274                 u32 serial_number;
2275                 u32 pad;
2276         } mapping[0];
2277 };
2278
2279 struct nfit_set_info2 {
2280         struct nfit_set_info_map2 {
2281                 u64 region_offset;
2282                 u32 serial_number;
2283                 u16 vendor_id;
2284                 u16 manufacturing_date;
2285                 u8  manufacturing_location;
2286                 u8  reserved[31];
2287         } mapping[0];
2288 };
2289
2290 static size_t sizeof_nfit_set_info(int num_mappings)
2291 {
2292         return sizeof(struct nfit_set_info)
2293                 + num_mappings * sizeof(struct nfit_set_info_map);
2294 }
2295
2296 static size_t sizeof_nfit_set_info2(int num_mappings)
2297 {
2298         return sizeof(struct nfit_set_info2)
2299                 + num_mappings * sizeof(struct nfit_set_info_map2);
2300 }
2301
2302 static int cmp_map_compat(const void *m0, const void *m1)
2303 {
2304         const struct nfit_set_info_map *map0 = m0;
2305         const struct nfit_set_info_map *map1 = m1;
2306
2307         return memcmp(&map0->region_offset, &map1->region_offset,
2308                         sizeof(u64));
2309 }
2310
2311 static int cmp_map(const void *m0, const void *m1)
2312 {
2313         const struct nfit_set_info_map *map0 = m0;
2314         const struct nfit_set_info_map *map1 = m1;
2315
2316         if (map0->region_offset < map1->region_offset)
2317                 return -1;
2318         else if (map0->region_offset > map1->region_offset)
2319                 return 1;
2320         return 0;
2321 }
2322
2323 static int cmp_map2(const void *m0, const void *m1)
2324 {
2325         const struct nfit_set_info_map2 *map0 = m0;
2326         const struct nfit_set_info_map2 *map1 = m1;
2327
2328         if (map0->region_offset < map1->region_offset)
2329                 return -1;
2330         else if (map0->region_offset > map1->region_offset)
2331                 return 1;
2332         return 0;
2333 }
2334
2335 /* Retrieve the nth entry referencing this spa */
2336 static struct acpi_nfit_memory_map *memdev_from_spa(
2337                 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
2338 {
2339         struct nfit_memdev *nfit_memdev;
2340
2341         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
2342                 if (nfit_memdev->memdev->range_index == range_index)
2343                         if (n-- == 0)
2344                                 return nfit_memdev->memdev;
2345         return NULL;
2346 }
2347
2348 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
2349                 struct nd_region_desc *ndr_desc,
2350                 struct acpi_nfit_system_address *spa)
2351 {
2352         struct device *dev = acpi_desc->dev;
2353         struct nd_interleave_set *nd_set;
2354         u16 nr = ndr_desc->num_mappings;
2355         struct nfit_set_info2 *info2;
2356         struct nfit_set_info *info;
2357         int i;
2358
2359         nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
2360         if (!nd_set)
2361                 return -ENOMEM;
2362         import_guid(&nd_set->type_guid, spa->range_guid);
2363
2364         info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
2365         if (!info)
2366                 return -ENOMEM;
2367
2368         info2 = devm_kzalloc(dev, sizeof_nfit_set_info2(nr), GFP_KERNEL);
2369         if (!info2)
2370                 return -ENOMEM;
2371
2372         for (i = 0; i < nr; i++) {
2373                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
2374                 struct nfit_set_info_map *map = &info->mapping[i];
2375                 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2376                 struct nvdimm *nvdimm = mapping->nvdimm;
2377                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2378                 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
2379                                 spa->range_index, i);
2380                 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2381
2382                 if (!memdev || !nfit_mem->dcr) {
2383                         dev_err(dev, "%s: failed to find DCR\n", __func__);
2384                         return -ENODEV;
2385                 }
2386
2387                 map->region_offset = memdev->region_offset;
2388                 map->serial_number = dcr->serial_number;
2389
2390                 map2->region_offset = memdev->region_offset;
2391                 map2->serial_number = dcr->serial_number;
2392                 map2->vendor_id = dcr->vendor_id;
2393                 map2->manufacturing_date = dcr->manufacturing_date;
2394                 map2->manufacturing_location = dcr->manufacturing_location;
2395         }
2396
2397         /* v1.1 namespaces */
2398         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2399                         cmp_map, NULL);
2400         nd_set->cookie1 = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2401
2402         /* v1.2 namespaces */
2403         sort(&info2->mapping[0], nr, sizeof(struct nfit_set_info_map2),
2404                         cmp_map2, NULL);
2405         nd_set->cookie2 = nd_fletcher64(info2, sizeof_nfit_set_info2(nr), 0);
2406
2407         /* support v1.1 namespaces created with the wrong sort order */
2408         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2409                         cmp_map_compat, NULL);
2410         nd_set->altcookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2411
2412         /* record the result of the sort for the mapping position */
2413         for (i = 0; i < nr; i++) {
2414                 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2415                 int j;
2416
2417                 for (j = 0; j < nr; j++) {
2418                         struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
2419                         struct nvdimm *nvdimm = mapping->nvdimm;
2420                         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2421                         struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2422
2423                         if (map2->serial_number == dcr->serial_number &&
2424                             map2->vendor_id == dcr->vendor_id &&
2425                             map2->manufacturing_date == dcr->manufacturing_date &&
2426                             map2->manufacturing_location
2427                                     == dcr->manufacturing_location) {
2428                                 mapping->position = i;
2429                                 break;
2430                         }
2431                 }
2432         }
2433
2434         ndr_desc->nd_set = nd_set;
2435         devm_kfree(dev, info);
2436         devm_kfree(dev, info2);
2437
2438         return 0;
2439 }
2440
2441 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
2442 {
2443         struct acpi_nfit_interleave *idt = mmio->idt;
2444         u32 sub_line_offset, line_index, line_offset;
2445         u64 line_no, table_skip_count, table_offset;
2446
2447         line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
2448         table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
2449         line_offset = idt->line_offset[line_index]
2450                 * mmio->line_size;
2451         table_offset = table_skip_count * mmio->table_size;
2452
2453         return mmio->base_offset + line_offset + table_offset + sub_line_offset;
2454 }
2455
2456 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
2457 {
2458         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2459         u64 offset = nfit_blk->stat_offset + mmio->size * bw;
2460         const u32 STATUS_MASK = 0x80000037;
2461
2462         if (mmio->num_lines)
2463                 offset = to_interleave_offset(offset, mmio);
2464
2465         return readl(mmio->addr.base + offset) & STATUS_MASK;
2466 }
2467
2468 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
2469                 resource_size_t dpa, unsigned int len, unsigned int write)
2470 {
2471         u64 cmd, offset;
2472         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2473
2474         enum {
2475                 BCW_OFFSET_MASK = (1ULL << 48)-1,
2476                 BCW_LEN_SHIFT = 48,
2477                 BCW_LEN_MASK = (1ULL << 8) - 1,
2478                 BCW_CMD_SHIFT = 56,
2479         };
2480
2481         cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
2482         len = len >> L1_CACHE_SHIFT;
2483         cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
2484         cmd |= ((u64) write) << BCW_CMD_SHIFT;
2485
2486         offset = nfit_blk->cmd_offset + mmio->size * bw;
2487         if (mmio->num_lines)
2488                 offset = to_interleave_offset(offset, mmio);
2489
2490         writeq(cmd, mmio->addr.base + offset);
2491         nvdimm_flush(nfit_blk->nd_region, NULL);
2492
2493         if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
2494                 readq(mmio->addr.base + offset);
2495 }
2496
2497 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
2498                 resource_size_t dpa, void *iobuf, size_t len, int rw,
2499                 unsigned int lane)
2500 {
2501         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2502         unsigned int copied = 0;
2503         u64 base_offset;
2504         int rc;
2505
2506         base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
2507                 + lane * mmio->size;
2508         write_blk_ctl(nfit_blk, lane, dpa, len, rw);
2509         while (len) {
2510                 unsigned int c;
2511                 u64 offset;
2512
2513                 if (mmio->num_lines) {
2514                         u32 line_offset;
2515
2516                         offset = to_interleave_offset(base_offset + copied,
2517                                         mmio);
2518                         div_u64_rem(offset, mmio->line_size, &line_offset);
2519                         c = min_t(size_t, len, mmio->line_size - line_offset);
2520                 } else {
2521                         offset = base_offset + nfit_blk->bdw_offset;
2522                         c = len;
2523                 }
2524
2525                 if (rw)
2526                         memcpy_flushcache(mmio->addr.aperture + offset, iobuf + copied, c);
2527                 else {
2528                         if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
2529                                 arch_invalidate_pmem((void __force *)
2530                                         mmio->addr.aperture + offset, c);
2531
2532                         memcpy(iobuf + copied, mmio->addr.aperture + offset, c);
2533                 }
2534
2535                 copied += c;
2536                 len -= c;
2537         }
2538
2539         if (rw)
2540                 nvdimm_flush(nfit_blk->nd_region, NULL);
2541
2542         rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
2543         return rc;
2544 }
2545
2546 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
2547                 resource_size_t dpa, void *iobuf, u64 len, int rw)
2548 {
2549         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
2550         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2551         struct nd_region *nd_region = nfit_blk->nd_region;
2552         unsigned int lane, copied = 0;
2553         int rc = 0;
2554
2555         lane = nd_region_acquire_lane(nd_region);
2556         while (len) {
2557                 u64 c = min(len, mmio->size);
2558
2559                 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
2560                                 iobuf + copied, c, rw, lane);
2561                 if (rc)
2562                         break;
2563
2564                 copied += c;
2565                 len -= c;
2566         }
2567         nd_region_release_lane(nd_region, lane);
2568
2569         return rc;
2570 }
2571
2572 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
2573                 struct acpi_nfit_interleave *idt, u16 interleave_ways)
2574 {
2575         if (idt) {
2576                 mmio->num_lines = idt->line_count;
2577                 mmio->line_size = idt->line_size;
2578                 if (interleave_ways == 0)
2579                         return -ENXIO;
2580                 mmio->table_size = mmio->num_lines * interleave_ways
2581                         * mmio->line_size;
2582         }
2583
2584         return 0;
2585 }
2586
2587 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
2588                 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
2589 {
2590         struct nd_cmd_dimm_flags flags;
2591         int rc;
2592
2593         memset(&flags, 0, sizeof(flags));
2594         rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
2595                         sizeof(flags), NULL);
2596
2597         if (rc >= 0 && flags.status == 0)
2598                 nfit_blk->dimm_flags = flags.flags;
2599         else if (rc == -ENOTTY) {
2600                 /* fall back to a conservative default */
2601                 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
2602                 rc = 0;
2603         } else
2604                 rc = -ENXIO;
2605
2606         return rc;
2607 }
2608
2609 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
2610                 struct device *dev)
2611 {
2612         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
2613         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
2614         struct nfit_blk_mmio *mmio;
2615         struct nfit_blk *nfit_blk;
2616         struct nfit_mem *nfit_mem;
2617         struct nvdimm *nvdimm;
2618         int rc;
2619
2620         nvdimm = nd_blk_region_to_dimm(ndbr);
2621         nfit_mem = nvdimm_provider_data(nvdimm);
2622         if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
2623                 dev_dbg(dev, "missing%s%s%s\n",
2624                                 nfit_mem ? "" : " nfit_mem",
2625                                 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
2626                                 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
2627                 return -ENXIO;
2628         }
2629
2630         nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
2631         if (!nfit_blk)
2632                 return -ENOMEM;
2633         nd_blk_region_set_provider_data(ndbr, nfit_blk);
2634         nfit_blk->nd_region = to_nd_region(dev);
2635
2636         /* map block aperture memory */
2637         nfit_blk->bdw_offset = nfit_mem->bdw->offset;
2638         mmio = &nfit_blk->mmio[BDW];
2639         mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
2640                         nfit_mem->spa_bdw->length, nd_blk_memremap_flags(ndbr));
2641         if (!mmio->addr.base) {
2642                 dev_dbg(dev, "%s failed to map bdw\n",
2643                                 nvdimm_name(nvdimm));
2644                 return -ENOMEM;
2645         }
2646         mmio->size = nfit_mem->bdw->size;
2647         mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
2648         mmio->idt = nfit_mem->idt_bdw;
2649         mmio->spa = nfit_mem->spa_bdw;
2650         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
2651                         nfit_mem->memdev_bdw->interleave_ways);
2652         if (rc) {
2653                 dev_dbg(dev, "%s failed to init bdw interleave\n",
2654                                 nvdimm_name(nvdimm));
2655                 return rc;
2656         }
2657
2658         /* map block control memory */
2659         nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
2660         nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
2661         mmio = &nfit_blk->mmio[DCR];
2662         mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
2663                         nfit_mem->spa_dcr->length);
2664         if (!mmio->addr.base) {
2665                 dev_dbg(dev, "%s failed to map dcr\n",
2666                                 nvdimm_name(nvdimm));
2667                 return -ENOMEM;
2668         }
2669         mmio->size = nfit_mem->dcr->window_size;
2670         mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
2671         mmio->idt = nfit_mem->idt_dcr;
2672         mmio->spa = nfit_mem->spa_dcr;
2673         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
2674                         nfit_mem->memdev_dcr->interleave_ways);
2675         if (rc) {
2676                 dev_dbg(dev, "%s failed to init dcr interleave\n",
2677                                 nvdimm_name(nvdimm));
2678                 return rc;
2679         }
2680
2681         rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
2682         if (rc < 0) {
2683                 dev_dbg(dev, "%s failed get DIMM flags\n",
2684                                 nvdimm_name(nvdimm));
2685                 return rc;
2686         }
2687
2688         if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
2689                 dev_warn(dev, "unable to guarantee persistence of writes\n");
2690
2691         if (mmio->line_size == 0)
2692                 return 0;
2693
2694         if ((u32) nfit_blk->cmd_offset % mmio->line_size
2695                         + 8 > mmio->line_size) {
2696                 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
2697                 return -ENXIO;
2698         } else if ((u32) nfit_blk->stat_offset % mmio->line_size
2699                         + 8 > mmio->line_size) {
2700                 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
2701                 return -ENXIO;
2702         }
2703
2704         return 0;
2705 }
2706
2707 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2708                 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2709 {
2710         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2711         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2712         int cmd_rc, rc;
2713
2714         cmd->address = spa->address;
2715         cmd->length = spa->length;
2716         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2717                         sizeof(*cmd), &cmd_rc);
2718         if (rc < 0)
2719                 return rc;
2720         return cmd_rc;
2721 }
2722
2723 static int ars_start(struct acpi_nfit_desc *acpi_desc,
2724                 struct nfit_spa *nfit_spa, enum nfit_ars_state req_type)
2725 {
2726         int rc;
2727         int cmd_rc;
2728         struct nd_cmd_ars_start ars_start;
2729         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2730         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2731
2732         memset(&ars_start, 0, sizeof(ars_start));
2733         ars_start.address = spa->address;
2734         ars_start.length = spa->length;
2735         if (req_type == ARS_REQ_SHORT)
2736                 ars_start.flags = ND_ARS_RETURN_PREV_DATA;
2737         if (nfit_spa_type(spa) == NFIT_SPA_PM)
2738                 ars_start.type = ND_ARS_PERSISTENT;
2739         else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2740                 ars_start.type = ND_ARS_VOLATILE;
2741         else
2742                 return -ENOTTY;
2743
2744         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2745                         sizeof(ars_start), &cmd_rc);
2746
2747         if (rc < 0)
2748                 return rc;
2749         if (cmd_rc < 0)
2750                 return cmd_rc;
2751         set_bit(ARS_VALID, &acpi_desc->scrub_flags);
2752         return 0;
2753 }
2754
2755 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2756 {
2757         int rc, cmd_rc;
2758         struct nd_cmd_ars_start ars_start;
2759         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2760         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2761
2762         ars_start = (struct nd_cmd_ars_start) {
2763                 .address = ars_status->restart_address,
2764                 .length = ars_status->restart_length,
2765                 .type = ars_status->type,
2766         };
2767         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2768                         sizeof(ars_start), &cmd_rc);
2769         if (rc < 0)
2770                 return rc;
2771         return cmd_rc;
2772 }
2773
2774 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2775 {
2776         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2777         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2778         int rc, cmd_rc;
2779
2780         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2781                         acpi_desc->max_ars, &cmd_rc);
2782         if (rc < 0)
2783                 return rc;
2784         return cmd_rc;
2785 }
2786
2787 static void ars_complete(struct acpi_nfit_desc *acpi_desc,
2788                 struct nfit_spa *nfit_spa)
2789 {
2790         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2791         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2792         struct nd_region *nd_region = nfit_spa->nd_region;
2793         struct device *dev;
2794
2795         lockdep_assert_held(&acpi_desc->init_mutex);
2796         /*
2797          * Only advance the ARS state for ARS runs initiated by the
2798          * kernel, ignore ARS results from BIOS initiated runs for scrub
2799          * completion tracking.
2800          */
2801         if (acpi_desc->scrub_spa != nfit_spa)
2802                 return;
2803
2804         if ((ars_status->address >= spa->address && ars_status->address
2805                                 < spa->address + spa->length)
2806                         || (ars_status->address < spa->address)) {
2807                 /*
2808                  * Assume that if a scrub starts at an offset from the
2809                  * start of nfit_spa that we are in the continuation
2810                  * case.
2811                  *
2812                  * Otherwise, if the scrub covers the spa range, mark
2813                  * any pending request complete.
2814                  */
2815                 if (ars_status->address + ars_status->length
2816                                 >= spa->address + spa->length)
2817                                 /* complete */;
2818                 else
2819                         return;
2820         } else
2821                 return;
2822
2823         acpi_desc->scrub_spa = NULL;
2824         if (nd_region) {
2825                 dev = nd_region_dev(nd_region);
2826                 nvdimm_region_notify(nd_region, NVDIMM_REVALIDATE_POISON);
2827         } else
2828                 dev = acpi_desc->dev;
2829         dev_dbg(dev, "ARS: range %d complete\n", spa->range_index);
2830 }
2831
2832 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc)
2833 {
2834         struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2835         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2836         int rc;
2837         u32 i;
2838
2839         /*
2840          * First record starts at 44 byte offset from the start of the
2841          * payload.
2842          */
2843         if (ars_status->out_length < 44)
2844                 return 0;
2845
2846         /*
2847          * Ignore potentially stale results that are only refreshed
2848          * after a start-ARS event.
2849          */
2850         if (!test_and_clear_bit(ARS_VALID, &acpi_desc->scrub_flags)) {
2851                 dev_dbg(acpi_desc->dev, "skip %d stale records\n",
2852                                 ars_status->num_records);
2853                 return 0;
2854         }
2855
2856         for (i = 0; i < ars_status->num_records; i++) {
2857                 /* only process full records */
2858                 if (ars_status->out_length
2859                                 < 44 + sizeof(struct nd_ars_record) * (i + 1))
2860                         break;
2861                 rc = nvdimm_bus_add_badrange(nvdimm_bus,
2862                                 ars_status->records[i].err_address,
2863                                 ars_status->records[i].length);
2864                 if (rc)
2865                         return rc;
2866         }
2867         if (i < ars_status->num_records)
2868                 dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2869
2870         return 0;
2871 }
2872
2873 static void acpi_nfit_remove_resource(void *data)
2874 {
2875         struct resource *res = data;
2876
2877         remove_resource(res);
2878 }
2879
2880 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2881                 struct nd_region_desc *ndr_desc)
2882 {
2883         struct resource *res, *nd_res = ndr_desc->res;
2884         int is_pmem, ret;
2885
2886         /* No operation if the region is already registered as PMEM */
2887         is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2888                                 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2889         if (is_pmem == REGION_INTERSECTS)
2890                 return 0;
2891
2892         res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2893         if (!res)
2894                 return -ENOMEM;
2895
2896         res->name = "Persistent Memory";
2897         res->start = nd_res->start;
2898         res->end = nd_res->end;
2899         res->flags = IORESOURCE_MEM;
2900         res->desc = IORES_DESC_PERSISTENT_MEMORY;
2901
2902         ret = insert_resource(&iomem_resource, res);
2903         if (ret)
2904                 return ret;
2905
2906         ret = devm_add_action_or_reset(acpi_desc->dev,
2907                                         acpi_nfit_remove_resource,
2908                                         res);
2909         if (ret)
2910                 return ret;
2911
2912         return 0;
2913 }
2914
2915 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2916                 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2917                 struct acpi_nfit_memory_map *memdev,
2918                 struct nfit_spa *nfit_spa)
2919 {
2920         struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2921                         memdev->device_handle);
2922         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2923         struct nd_blk_region_desc *ndbr_desc;
2924         struct nfit_mem *nfit_mem;
2925         int rc;
2926
2927         if (!nvdimm) {
2928                 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2929                                 spa->range_index, memdev->device_handle);
2930                 return -ENODEV;
2931         }
2932
2933         mapping->nvdimm = nvdimm;
2934         switch (nfit_spa_type(spa)) {
2935         case NFIT_SPA_PM:
2936         case NFIT_SPA_VOLATILE:
2937                 mapping->start = memdev->address;
2938                 mapping->size = memdev->region_size;
2939                 break;
2940         case NFIT_SPA_DCR:
2941                 nfit_mem = nvdimm_provider_data(nvdimm);
2942                 if (!nfit_mem || !nfit_mem->bdw) {
2943                         dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2944                                         spa->range_index, nvdimm_name(nvdimm));
2945                         break;
2946                 }
2947
2948                 mapping->size = nfit_mem->bdw->capacity;
2949                 mapping->start = nfit_mem->bdw->start_address;
2950                 ndr_desc->num_lanes = nfit_mem->bdw->windows;
2951                 ndr_desc->mapping = mapping;
2952                 ndr_desc->num_mappings = 1;
2953                 ndbr_desc = to_blk_region_desc(ndr_desc);
2954                 ndbr_desc->enable = acpi_nfit_blk_region_enable;
2955                 ndbr_desc->do_io = acpi_desc->blk_do_io;
2956                 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2957                 if (rc)
2958                         return rc;
2959                 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2960                                 ndr_desc);
2961                 if (!nfit_spa->nd_region)
2962                         return -ENOMEM;
2963                 break;
2964         }
2965
2966         return 0;
2967 }
2968
2969 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2970 {
2971         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2972                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2973                 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2974                 nfit_spa_type(spa) == NFIT_SPA_PCD);
2975 }
2976
2977 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
2978 {
2979         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2980                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2981                 nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
2982 }
2983
2984 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2985                 struct nfit_spa *nfit_spa)
2986 {
2987         static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2988         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2989         struct nd_blk_region_desc ndbr_desc;
2990         struct nd_region_desc *ndr_desc;
2991         struct nfit_memdev *nfit_memdev;
2992         struct nvdimm_bus *nvdimm_bus;
2993         struct resource res;
2994         int count = 0, rc;
2995
2996         if (nfit_spa->nd_region)
2997                 return 0;
2998
2999         if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
3000                 dev_dbg(acpi_desc->dev, "detected invalid spa index\n");
3001                 return 0;
3002         }
3003
3004         memset(&res, 0, sizeof(res));
3005         memset(&mappings, 0, sizeof(mappings));
3006         memset(&ndbr_desc, 0, sizeof(ndbr_desc));
3007         res.start = spa->address;
3008         res.end = res.start + spa->length - 1;
3009         ndr_desc = &ndbr_desc.ndr_desc;
3010         ndr_desc->res = &res;
3011         ndr_desc->provider_data = nfit_spa;
3012         ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
3013         if (spa->flags & ACPI_NFIT_PROXIMITY_VALID) {
3014                 ndr_desc->numa_node = pxm_to_online_node(spa->proximity_domain);
3015                 ndr_desc->target_node = pxm_to_node(spa->proximity_domain);
3016         } else {
3017                 ndr_desc->numa_node = NUMA_NO_NODE;
3018                 ndr_desc->target_node = NUMA_NO_NODE;
3019         }
3020
3021         /* Fallback to address based numa information if node lookup failed */
3022         if (ndr_desc->numa_node == NUMA_NO_NODE) {
3023                 ndr_desc->numa_node = memory_add_physaddr_to_nid(spa->address);
3024                 dev_info(acpi_desc->dev, "changing numa node from %d to %d for nfit region [%pa-%pa]",
3025                         NUMA_NO_NODE, ndr_desc->numa_node, &res.start, &res.end);
3026         }
3027         if (ndr_desc->target_node == NUMA_NO_NODE) {
3028                 ndr_desc->target_node = phys_to_target_node(spa->address);
3029                 dev_info(acpi_desc->dev, "changing target node from %d to %d for nfit region [%pa-%pa]",
3030                         NUMA_NO_NODE, ndr_desc->numa_node, &res.start, &res.end);
3031         }
3032
3033         /*
3034          * Persistence domain bits are hierarchical, if
3035          * ACPI_NFIT_CAPABILITY_CACHE_FLUSH is set then
3036          * ACPI_NFIT_CAPABILITY_MEM_FLUSH is implied.
3037          */
3038         if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_CACHE_FLUSH)
3039                 set_bit(ND_REGION_PERSIST_CACHE, &ndr_desc->flags);
3040         else if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_MEM_FLUSH)
3041                 set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc->flags);
3042
3043         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
3044                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
3045                 struct nd_mapping_desc *mapping;
3046
3047                 /* range index 0 == unmapped in SPA or invalid-SPA */
3048                 if (memdev->range_index == 0 || spa->range_index == 0)
3049                         continue;
3050                 if (memdev->range_index != spa->range_index)
3051                         continue;
3052                 if (count >= ND_MAX_MAPPINGS) {
3053                         dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
3054                                         spa->range_index, ND_MAX_MAPPINGS);
3055                         return -ENXIO;
3056                 }
3057                 mapping = &mappings[count++];
3058                 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
3059                                 memdev, nfit_spa);
3060                 if (rc)
3061                         goto out;
3062         }
3063
3064         ndr_desc->mapping = mappings;
3065         ndr_desc->num_mappings = count;
3066         rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
3067         if (rc)
3068                 goto out;
3069
3070         nvdimm_bus = acpi_desc->nvdimm_bus;
3071         if (nfit_spa_type(spa) == NFIT_SPA_PM) {
3072                 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
3073                 if (rc) {
3074                         dev_warn(acpi_desc->dev,
3075                                 "failed to insert pmem resource to iomem: %d\n",
3076                                 rc);
3077                         goto out;
3078                 }
3079
3080                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
3081                                 ndr_desc);
3082                 if (!nfit_spa->nd_region)
3083                         rc = -ENOMEM;
3084         } else if (nfit_spa_is_volatile(spa)) {
3085                 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
3086                                 ndr_desc);
3087                 if (!nfit_spa->nd_region)
3088                         rc = -ENOMEM;
3089         } else if (nfit_spa_is_virtual(spa)) {
3090                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
3091                                 ndr_desc);
3092                 if (!nfit_spa->nd_region)
3093                         rc = -ENOMEM;
3094         }
3095
3096  out:
3097         if (rc)
3098                 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
3099                                 nfit_spa->spa->range_index);
3100         return rc;
3101 }
3102
3103 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc)
3104 {
3105         struct device *dev = acpi_desc->dev;
3106         struct nd_cmd_ars_status *ars_status;
3107
3108         if (acpi_desc->ars_status) {
3109                 memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
3110                 return 0;
3111         }
3112
3113         ars_status = devm_kzalloc(dev, acpi_desc->max_ars, GFP_KERNEL);
3114         if (!ars_status)
3115                 return -ENOMEM;
3116         acpi_desc->ars_status = ars_status;
3117         return 0;
3118 }
3119
3120 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc)
3121 {
3122         int rc;
3123
3124         if (ars_status_alloc(acpi_desc))
3125                 return -ENOMEM;
3126
3127         rc = ars_get_status(acpi_desc);
3128
3129         if (rc < 0 && rc != -ENOSPC)
3130                 return rc;
3131
3132         if (ars_status_process_records(acpi_desc))
3133                 dev_err(acpi_desc->dev, "Failed to process ARS records\n");
3134
3135         return rc;
3136 }
3137
3138 static int ars_register(struct acpi_nfit_desc *acpi_desc,
3139                 struct nfit_spa *nfit_spa)
3140 {
3141         int rc;
3142
3143         if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3144                 return acpi_nfit_register_region(acpi_desc, nfit_spa);
3145
3146         set_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
3147         if (!no_init_ars)
3148                 set_bit(ARS_REQ_LONG, &nfit_spa->ars_state);
3149
3150         switch (acpi_nfit_query_poison(acpi_desc)) {
3151         case 0:
3152         case -ENOSPC:
3153         case -EAGAIN:
3154                 rc = ars_start(acpi_desc, nfit_spa, ARS_REQ_SHORT);
3155                 /* shouldn't happen, try again later */
3156                 if (rc == -EBUSY)
3157                         break;
3158                 if (rc) {
3159                         set_bit(ARS_FAILED, &nfit_spa->ars_state);
3160                         break;
3161                 }
3162                 clear_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
3163                 rc = acpi_nfit_query_poison(acpi_desc);
3164                 if (rc)
3165                         break;
3166                 acpi_desc->scrub_spa = nfit_spa;
3167                 ars_complete(acpi_desc, nfit_spa);
3168                 /*
3169                  * If ars_complete() says we didn't complete the
3170                  * short scrub, we'll try again with a long
3171                  * request.
3172                  */
3173                 acpi_desc->scrub_spa = NULL;
3174                 break;
3175         case -EBUSY:
3176         case -ENOMEM:
3177                 /*
3178                  * BIOS was using ARS, wait for it to complete (or
3179                  * resources to become available) and then perform our
3180                  * own scrubs.
3181                  */
3182                 break;
3183         default:
3184                 set_bit(ARS_FAILED, &nfit_spa->ars_state);
3185                 break;
3186         }
3187
3188         return acpi_nfit_register_region(acpi_desc, nfit_spa);
3189 }
3190
3191 static void ars_complete_all(struct acpi_nfit_desc *acpi_desc)
3192 {
3193         struct nfit_spa *nfit_spa;
3194
3195         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3196                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3197                         continue;
3198                 ars_complete(acpi_desc, nfit_spa);
3199         }
3200 }
3201
3202 static unsigned int __acpi_nfit_scrub(struct acpi_nfit_desc *acpi_desc,
3203                 int query_rc)
3204 {
3205         unsigned int tmo = acpi_desc->scrub_tmo;
3206         struct device *dev = acpi_desc->dev;
3207         struct nfit_spa *nfit_spa;
3208
3209         lockdep_assert_held(&acpi_desc->init_mutex);
3210
3211         if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags))
3212                 return 0;
3213
3214         if (query_rc == -EBUSY) {
3215                 dev_dbg(dev, "ARS: ARS busy\n");
3216                 return min(30U * 60U, tmo * 2);
3217         }
3218         if (query_rc == -ENOSPC) {
3219                 dev_dbg(dev, "ARS: ARS continue\n");
3220                 ars_continue(acpi_desc);
3221                 return 1;
3222         }
3223         if (query_rc && query_rc != -EAGAIN) {
3224                 unsigned long long addr, end;
3225
3226                 addr = acpi_desc->ars_status->address;
3227                 end = addr + acpi_desc->ars_status->length;
3228                 dev_dbg(dev, "ARS: %llx-%llx failed (%d)\n", addr, end,
3229                                 query_rc);
3230         }
3231
3232         ars_complete_all(acpi_desc);
3233         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3234                 enum nfit_ars_state req_type;
3235                 int rc;
3236
3237                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3238                         continue;
3239
3240                 /* prefer short ARS requests first */
3241                 if (test_bit(ARS_REQ_SHORT, &nfit_spa->ars_state))
3242                         req_type = ARS_REQ_SHORT;
3243                 else if (test_bit(ARS_REQ_LONG, &nfit_spa->ars_state))
3244                         req_type = ARS_REQ_LONG;
3245                 else
3246                         continue;
3247                 rc = ars_start(acpi_desc, nfit_spa, req_type);
3248
3249                 dev = nd_region_dev(nfit_spa->nd_region);
3250                 dev_dbg(dev, "ARS: range %d ARS start %s (%d)\n",
3251                                 nfit_spa->spa->range_index,
3252                                 req_type == ARS_REQ_SHORT ? "short" : "long",
3253                                 rc);
3254                 /*
3255                  * Hmm, we raced someone else starting ARS? Try again in
3256                  * a bit.
3257                  */
3258                 if (rc == -EBUSY)
3259                         return 1;
3260                 if (rc == 0) {
3261                         dev_WARN_ONCE(dev, acpi_desc->scrub_spa,
3262                                         "scrub start while range %d active\n",
3263                                         acpi_desc->scrub_spa->spa->range_index);
3264                         clear_bit(req_type, &nfit_spa->ars_state);
3265                         acpi_desc->scrub_spa = nfit_spa;
3266                         /*
3267                          * Consider this spa last for future scrub
3268                          * requests
3269                          */
3270                         list_move_tail(&nfit_spa->list, &acpi_desc->spas);
3271                         return 1;
3272                 }
3273
3274                 dev_err(dev, "ARS: range %d ARS failed (%d)\n",
3275                                 nfit_spa->spa->range_index, rc);
3276                 set_bit(ARS_FAILED, &nfit_spa->ars_state);
3277         }
3278         return 0;
3279 }
3280
3281 static void __sched_ars(struct acpi_nfit_desc *acpi_desc, unsigned int tmo)
3282 {
3283         lockdep_assert_held(&acpi_desc->init_mutex);
3284
3285         set_bit(ARS_BUSY, &acpi_desc->scrub_flags);
3286         /* note this should only be set from within the workqueue */
3287         if (tmo)
3288                 acpi_desc->scrub_tmo = tmo;
3289         queue_delayed_work(nfit_wq, &acpi_desc->dwork, tmo * HZ);
3290 }
3291
3292 static void sched_ars(struct acpi_nfit_desc *acpi_desc)
3293 {
3294         __sched_ars(acpi_desc, 0);
3295 }
3296
3297 static void notify_ars_done(struct acpi_nfit_desc *acpi_desc)
3298 {
3299         lockdep_assert_held(&acpi_desc->init_mutex);
3300
3301         clear_bit(ARS_BUSY, &acpi_desc->scrub_flags);
3302         acpi_desc->scrub_count++;
3303         if (acpi_desc->scrub_count_state)
3304                 sysfs_notify_dirent(acpi_desc->scrub_count_state);
3305 }
3306
3307 static void acpi_nfit_scrub(struct work_struct *work)
3308 {
3309         struct acpi_nfit_desc *acpi_desc;
3310         unsigned int tmo;
3311         int query_rc;
3312
3313         acpi_desc = container_of(work, typeof(*acpi_desc), dwork.work);
3314         mutex_lock(&acpi_desc->init_mutex);
3315         query_rc = acpi_nfit_query_poison(acpi_desc);
3316         tmo = __acpi_nfit_scrub(acpi_desc, query_rc);
3317         if (tmo)
3318                 __sched_ars(acpi_desc, tmo);
3319         else
3320                 notify_ars_done(acpi_desc);
3321         memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
3322         clear_bit(ARS_POLL, &acpi_desc->scrub_flags);
3323         mutex_unlock(&acpi_desc->init_mutex);
3324 }
3325
3326 static void acpi_nfit_init_ars(struct acpi_nfit_desc *acpi_desc,
3327                 struct nfit_spa *nfit_spa)
3328 {
3329         int type = nfit_spa_type(nfit_spa->spa);
3330         struct nd_cmd_ars_cap ars_cap;
3331         int rc;
3332
3333         set_bit(ARS_FAILED, &nfit_spa->ars_state);
3334         memset(&ars_cap, 0, sizeof(ars_cap));
3335         rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
3336         if (rc < 0)
3337                 return;
3338         /* check that the supported scrub types match the spa type */
3339         if (type == NFIT_SPA_VOLATILE && ((ars_cap.status >> 16)
3340                                 & ND_ARS_VOLATILE) == 0)
3341                 return;
3342         if (type == NFIT_SPA_PM && ((ars_cap.status >> 16)
3343                                 & ND_ARS_PERSISTENT) == 0)
3344                 return;
3345
3346         nfit_spa->max_ars = ars_cap.max_ars_out;
3347         nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
3348         acpi_desc->max_ars = max(nfit_spa->max_ars, acpi_desc->max_ars);
3349         clear_bit(ARS_FAILED, &nfit_spa->ars_state);
3350 }
3351
3352 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
3353 {
3354         struct nfit_spa *nfit_spa;
3355         int rc, do_sched_ars = 0;
3356
3357         set_bit(ARS_VALID, &acpi_desc->scrub_flags);
3358         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3359                 switch (nfit_spa_type(nfit_spa->spa)) {
3360                 case NFIT_SPA_VOLATILE:
3361                 case NFIT_SPA_PM:
3362                         acpi_nfit_init_ars(acpi_desc, nfit_spa);
3363                         break;
3364                 }
3365         }
3366
3367         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3368                 switch (nfit_spa_type(nfit_spa->spa)) {
3369                 case NFIT_SPA_VOLATILE:
3370                 case NFIT_SPA_PM:
3371                         /* register regions and kick off initial ARS run */
3372                         rc = ars_register(acpi_desc, nfit_spa);
3373                         if (rc)
3374                                 return rc;
3375
3376                         /*
3377                          * Kick off background ARS if at least one
3378                          * region successfully registered ARS
3379                          */
3380                         if (!test_bit(ARS_FAILED, &nfit_spa->ars_state))
3381                                 do_sched_ars++;
3382                         break;
3383                 case NFIT_SPA_BDW:
3384                         /* nothing to register */
3385                         break;
3386                 case NFIT_SPA_DCR:
3387                 case NFIT_SPA_VDISK:
3388                 case NFIT_SPA_VCD:
3389                 case NFIT_SPA_PDISK:
3390                 case NFIT_SPA_PCD:
3391                         /* register known regions that don't support ARS */
3392                         rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
3393                         if (rc)
3394                                 return rc;
3395                         break;
3396                 default:
3397                         /* don't register unknown regions */
3398                         break;
3399                 }
3400         }
3401
3402         if (do_sched_ars)
3403                 sched_ars(acpi_desc);
3404         return 0;
3405 }
3406
3407 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
3408                 struct nfit_table_prev *prev)
3409 {
3410         struct device *dev = acpi_desc->dev;
3411
3412         if (!list_empty(&prev->spas) ||
3413                         !list_empty(&prev->memdevs) ||
3414                         !list_empty(&prev->dcrs) ||
3415                         !list_empty(&prev->bdws) ||
3416                         !list_empty(&prev->idts) ||
3417                         !list_empty(&prev->flushes)) {
3418                 dev_err(dev, "new nfit deletes entries (unsupported)\n");
3419                 return -ENXIO;
3420         }
3421         return 0;
3422 }
3423
3424 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
3425 {
3426         struct device *dev = acpi_desc->dev;
3427         struct kernfs_node *nfit;
3428         struct device *bus_dev;
3429
3430         if (!ars_supported(acpi_desc->nvdimm_bus))
3431                 return 0;
3432
3433         bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3434         nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
3435         if (!nfit) {
3436                 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
3437                 return -ENODEV;
3438         }
3439         acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
3440         sysfs_put(nfit);
3441         if (!acpi_desc->scrub_count_state) {
3442                 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
3443                 return -ENODEV;
3444         }
3445
3446         return 0;
3447 }
3448
3449 static void acpi_nfit_unregister(void *data)
3450 {
3451         struct acpi_nfit_desc *acpi_desc = data;
3452
3453         nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
3454 }
3455
3456 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
3457 {
3458         struct device *dev = acpi_desc->dev;
3459         struct nfit_table_prev prev;
3460         const void *end;
3461         int rc;
3462
3463         if (!acpi_desc->nvdimm_bus) {
3464                 acpi_nfit_init_dsms(acpi_desc);
3465
3466                 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
3467                                 &acpi_desc->nd_desc);
3468                 if (!acpi_desc->nvdimm_bus)
3469                         return -ENOMEM;
3470
3471                 rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
3472                                 acpi_desc);
3473                 if (rc)
3474                         return rc;
3475
3476                 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
3477                 if (rc)
3478                         return rc;
3479
3480                 /* register this acpi_desc for mce notifications */
3481                 mutex_lock(&acpi_desc_lock);
3482                 list_add_tail(&acpi_desc->list, &acpi_descs);
3483                 mutex_unlock(&acpi_desc_lock);
3484         }
3485
3486         mutex_lock(&acpi_desc->init_mutex);
3487
3488         INIT_LIST_HEAD(&prev.spas);
3489         INIT_LIST_HEAD(&prev.memdevs);
3490         INIT_LIST_HEAD(&prev.dcrs);
3491         INIT_LIST_HEAD(&prev.bdws);
3492         INIT_LIST_HEAD(&prev.idts);
3493         INIT_LIST_HEAD(&prev.flushes);
3494
3495         list_cut_position(&prev.spas, &acpi_desc->spas,
3496                                 acpi_desc->spas.prev);
3497         list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
3498                                 acpi_desc->memdevs.prev);
3499         list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
3500                                 acpi_desc->dcrs.prev);
3501         list_cut_position(&prev.bdws, &acpi_desc->bdws,
3502                                 acpi_desc->bdws.prev);
3503         list_cut_position(&prev.idts, &acpi_desc->idts,
3504                                 acpi_desc->idts.prev);
3505         list_cut_position(&prev.flushes, &acpi_desc->flushes,
3506                                 acpi_desc->flushes.prev);
3507
3508         end = data + sz;
3509         while (!IS_ERR_OR_NULL(data))
3510                 data = add_table(acpi_desc, &prev, data, end);
3511
3512         if (IS_ERR(data)) {
3513                 dev_dbg(dev, "nfit table parsing error: %ld\n", PTR_ERR(data));
3514                 rc = PTR_ERR(data);
3515                 goto out_unlock;
3516         }
3517
3518         rc = acpi_nfit_check_deletions(acpi_desc, &prev);
3519         if (rc)
3520                 goto out_unlock;
3521
3522         rc = nfit_mem_init(acpi_desc);
3523         if (rc)
3524                 goto out_unlock;
3525
3526         rc = acpi_nfit_register_dimms(acpi_desc);
3527         if (rc)
3528                 goto out_unlock;
3529
3530         rc = acpi_nfit_register_regions(acpi_desc);
3531
3532  out_unlock:
3533         mutex_unlock(&acpi_desc->init_mutex);
3534         return rc;
3535 }
3536 EXPORT_SYMBOL_GPL(acpi_nfit_init);
3537
3538 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
3539 {
3540         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
3541         struct device *dev = acpi_desc->dev;
3542
3543         /* Bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
3544         nfit_device_lock(dev);
3545         nfit_device_unlock(dev);
3546
3547         /* Bounce the init_mutex to complete initial registration */
3548         mutex_lock(&acpi_desc->init_mutex);
3549         mutex_unlock(&acpi_desc->init_mutex);
3550
3551         return 0;
3552 }
3553
3554 static int __acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3555                 struct nvdimm *nvdimm, unsigned int cmd)
3556 {
3557         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
3558
3559         if (nvdimm)
3560                 return 0;
3561         if (cmd != ND_CMD_ARS_START)
3562                 return 0;
3563
3564         /*
3565          * The kernel and userspace may race to initiate a scrub, but
3566          * the scrub thread is prepared to lose that initial race.  It
3567          * just needs guarantees that any ARS it initiates are not
3568          * interrupted by any intervening start requests from userspace.
3569          */
3570         if (work_busy(&acpi_desc->dwork.work))
3571                 return -EBUSY;
3572
3573         return 0;
3574 }
3575
3576 /*
3577  * Prevent security and firmware activate commands from being issued via
3578  * ioctl.
3579  */
3580 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3581                 struct nvdimm *nvdimm, unsigned int cmd, void *buf)
3582 {
3583         struct nd_cmd_pkg *call_pkg = buf;
3584         unsigned int func;
3585
3586         if (nvdimm && cmd == ND_CMD_CALL &&
3587                         call_pkg->nd_family == NVDIMM_FAMILY_INTEL) {
3588                 func = call_pkg->nd_command;
3589                 if (func > NVDIMM_CMD_MAX ||
3590                     (1 << func) & NVDIMM_INTEL_DENY_CMDMASK)
3591                         return -EOPNOTSUPP;
3592         }
3593
3594         /* block all non-nfit bus commands */
3595         if (!nvdimm && cmd == ND_CMD_CALL &&
3596                         call_pkg->nd_family != NVDIMM_BUS_FAMILY_NFIT)
3597                 return -EOPNOTSUPP;
3598
3599         return __acpi_nfit_clear_to_send(nd_desc, nvdimm, cmd);
3600 }
3601
3602 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc,
3603                 enum nfit_ars_state req_type)
3604 {
3605         struct device *dev = acpi_desc->dev;
3606         int scheduled = 0, busy = 0;
3607         struct nfit_spa *nfit_spa;
3608
3609         mutex_lock(&acpi_desc->init_mutex);
3610         if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags)) {
3611                 mutex_unlock(&acpi_desc->init_mutex);
3612                 return 0;
3613         }
3614
3615         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3616                 int type = nfit_spa_type(nfit_spa->spa);
3617
3618                 if (type != NFIT_SPA_PM && type != NFIT_SPA_VOLATILE)
3619                         continue;
3620                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3621                         continue;
3622
3623                 if (test_and_set_bit(req_type, &nfit_spa->ars_state))
3624                         busy++;
3625                 else
3626                         scheduled++;
3627         }
3628         if (scheduled) {
3629                 sched_ars(acpi_desc);
3630                 dev_dbg(dev, "ars_scan triggered\n");
3631         }
3632         mutex_unlock(&acpi_desc->init_mutex);
3633
3634         if (scheduled)
3635                 return 0;
3636         if (busy)
3637                 return -EBUSY;
3638         return -ENOTTY;
3639 }
3640
3641 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
3642 {
3643         struct nvdimm_bus_descriptor *nd_desc;
3644
3645         dev_set_drvdata(dev, acpi_desc);
3646         acpi_desc->dev = dev;
3647         acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
3648         nd_desc = &acpi_desc->nd_desc;
3649         nd_desc->provider_name = "ACPI.NFIT";
3650         nd_desc->module = THIS_MODULE;
3651         nd_desc->ndctl = acpi_nfit_ctl;
3652         nd_desc->flush_probe = acpi_nfit_flush_probe;
3653         nd_desc->clear_to_send = acpi_nfit_clear_to_send;
3654         nd_desc->attr_groups = acpi_nfit_attribute_groups;
3655
3656         INIT_LIST_HEAD(&acpi_desc->spas);
3657         INIT_LIST_HEAD(&acpi_desc->dcrs);
3658         INIT_LIST_HEAD(&acpi_desc->bdws);
3659         INIT_LIST_HEAD(&acpi_desc->idts);
3660         INIT_LIST_HEAD(&acpi_desc->flushes);
3661         INIT_LIST_HEAD(&acpi_desc->memdevs);
3662         INIT_LIST_HEAD(&acpi_desc->dimms);
3663         INIT_LIST_HEAD(&acpi_desc->list);
3664         mutex_init(&acpi_desc->init_mutex);
3665         acpi_desc->scrub_tmo = 1;
3666         INIT_DELAYED_WORK(&acpi_desc->dwork, acpi_nfit_scrub);
3667 }
3668 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
3669
3670 static void acpi_nfit_put_table(void *table)
3671 {
3672         acpi_put_table(table);
3673 }
3674
3675 void acpi_nfit_shutdown(void *data)
3676 {
3677         struct acpi_nfit_desc *acpi_desc = data;
3678         struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3679
3680         /*
3681          * Destruct under acpi_desc_lock so that nfit_handle_mce does not
3682          * race teardown
3683          */
3684         mutex_lock(&acpi_desc_lock);
3685         list_del(&acpi_desc->list);
3686         mutex_unlock(&acpi_desc_lock);
3687
3688         mutex_lock(&acpi_desc->init_mutex);
3689         set_bit(ARS_CANCEL, &acpi_desc->scrub_flags);
3690         mutex_unlock(&acpi_desc->init_mutex);
3691         cancel_delayed_work_sync(&acpi_desc->dwork);
3692
3693         /*
3694          * Bounce the nvdimm bus lock to make sure any in-flight
3695          * acpi_nfit_ars_rescan() submissions have had a chance to
3696          * either submit or see ->cancel set.
3697          */
3698         nfit_device_lock(bus_dev);
3699         nfit_device_unlock(bus_dev);
3700
3701         flush_workqueue(nfit_wq);
3702 }
3703 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown);
3704
3705 static int acpi_nfit_add(struct acpi_device *adev)
3706 {
3707         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3708         struct acpi_nfit_desc *acpi_desc;
3709         struct device *dev = &adev->dev;
3710         struct acpi_table_header *tbl;
3711         acpi_status status = AE_OK;
3712         acpi_size sz;
3713         int rc = 0;
3714
3715         status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
3716         if (ACPI_FAILURE(status)) {
3717                 /* The NVDIMM root device allows OS to trigger enumeration of
3718                  * NVDIMMs through NFIT at boot time and re-enumeration at
3719                  * root level via the _FIT method during runtime.
3720                  * This is ok to return 0 here, we could have an nvdimm
3721                  * hotplugged later and evaluate _FIT method which returns
3722                  * data in the format of a series of NFIT Structures.
3723                  */
3724                 dev_dbg(dev, "failed to find NFIT at startup\n");
3725                 return 0;
3726         }
3727
3728         rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl);
3729         if (rc)
3730                 return rc;
3731         sz = tbl->length;
3732
3733         acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3734         if (!acpi_desc)
3735                 return -ENOMEM;
3736         acpi_nfit_desc_init(acpi_desc, &adev->dev);
3737
3738         /* Save the acpi header for exporting the revision via sysfs */
3739         acpi_desc->acpi_header = *tbl;
3740
3741         /* Evaluate _FIT and override with that if present */
3742         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
3743         if (ACPI_SUCCESS(status) && buf.length > 0) {
3744                 union acpi_object *obj = buf.pointer;
3745
3746                 if (obj->type == ACPI_TYPE_BUFFER)
3747                         rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3748                                         obj->buffer.length);
3749                 else
3750                         dev_dbg(dev, "invalid type %d, ignoring _FIT\n",
3751                                 (int) obj->type);
3752                 kfree(buf.pointer);
3753         } else
3754                 /* skip over the lead-in header table */
3755                 rc = acpi_nfit_init(acpi_desc, (void *) tbl
3756                                 + sizeof(struct acpi_table_nfit),
3757                                 sz - sizeof(struct acpi_table_nfit));
3758
3759         if (rc)
3760                 return rc;
3761         return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc);
3762 }
3763
3764 static int acpi_nfit_remove(struct acpi_device *adev)
3765 {
3766         /* see acpi_nfit_unregister */
3767         return 0;
3768 }
3769
3770 static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle)
3771 {
3772         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3773         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3774         union acpi_object *obj;
3775         acpi_status status;
3776         int ret;
3777
3778         if (!dev->driver) {
3779                 /* dev->driver may be null if we're being removed */
3780                 dev_dbg(dev, "no driver found for dev\n");
3781                 return;
3782         }
3783
3784         if (!acpi_desc) {
3785                 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3786                 if (!acpi_desc)
3787                         return;
3788                 acpi_nfit_desc_init(acpi_desc, dev);
3789         } else {
3790                 /*
3791                  * Finish previous registration before considering new
3792                  * regions.
3793                  */
3794                 flush_workqueue(nfit_wq);
3795         }
3796
3797         /* Evaluate _FIT */
3798         status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
3799         if (ACPI_FAILURE(status)) {
3800                 dev_err(dev, "failed to evaluate _FIT\n");
3801                 return;
3802         }
3803
3804         obj = buf.pointer;
3805         if (obj->type == ACPI_TYPE_BUFFER) {
3806                 ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3807                                 obj->buffer.length);
3808                 if (ret)
3809                         dev_err(dev, "failed to merge updated NFIT\n");
3810         } else
3811                 dev_err(dev, "Invalid _FIT\n");
3812         kfree(buf.pointer);
3813 }
3814
3815 static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle)
3816 {
3817         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3818
3819         if (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON)
3820                 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
3821         else
3822                 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_SHORT);
3823 }
3824
3825 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
3826 {
3827         dev_dbg(dev, "event: 0x%x\n", event);
3828
3829         switch (event) {
3830         case NFIT_NOTIFY_UPDATE:
3831                 return acpi_nfit_update_notify(dev, handle);
3832         case NFIT_NOTIFY_UC_MEMORY_ERROR:
3833                 return acpi_nfit_uc_error_notify(dev, handle);
3834         default:
3835                 return;
3836         }
3837 }
3838 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
3839
3840 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
3841 {
3842         nfit_device_lock(&adev->dev);
3843         __acpi_nfit_notify(&adev->dev, adev->handle, event);
3844         nfit_device_unlock(&adev->dev);
3845 }
3846
3847 static const struct acpi_device_id acpi_nfit_ids[] = {
3848         { "ACPI0012", 0 },
3849         { "", 0 },
3850 };
3851 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
3852
3853 static struct acpi_driver acpi_nfit_driver = {
3854         .name = KBUILD_MODNAME,
3855         .ids = acpi_nfit_ids,
3856         .ops = {
3857                 .add = acpi_nfit_add,
3858                 .remove = acpi_nfit_remove,
3859                 .notify = acpi_nfit_notify,
3860         },
3861 };
3862
3863 static __init int nfit_init(void)
3864 {
3865         int ret;
3866
3867         BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
3868         BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
3869         BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
3870         BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
3871         BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
3872         BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
3873         BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
3874         BUILD_BUG_ON(sizeof(struct acpi_nfit_capabilities) != 16);
3875
3876         guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]);
3877         guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]);
3878         guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]);
3879         guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]);
3880         guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]);
3881         guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]);
3882         guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]);
3883         guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]);
3884         guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]);
3885         guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]);
3886         guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
3887         guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
3888         guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
3889         guid_parse(UUID_NFIT_DIMM_N_HYPERV, &nfit_uuid[NFIT_DEV_DIMM_N_HYPERV]);
3890         guid_parse(UUID_INTEL_BUS, &nfit_uuid[NFIT_BUS_INTEL]);
3891
3892         nfit_wq = create_singlethread_workqueue("nfit");
3893         if (!nfit_wq)
3894                 return -ENOMEM;
3895
3896         nfit_mce_register();
3897         ret = acpi_bus_register_driver(&acpi_nfit_driver);
3898         if (ret) {
3899                 nfit_mce_unregister();
3900                 destroy_workqueue(nfit_wq);
3901         }
3902
3903         return ret;
3904
3905 }
3906
3907 static __exit void nfit_exit(void)
3908 {
3909         nfit_mce_unregister();
3910         acpi_bus_unregister_driver(&acpi_nfit_driver);
3911         destroy_workqueue(nfit_wq);
3912         WARN_ON(!list_empty(&acpi_descs));
3913 }
3914
3915 module_init(nfit_init);
3916 module_exit(nfit_exit);
3917 MODULE_LICENSE("GPL v2");
3918 MODULE_AUTHOR("Intel Corporation");