drivers/acpi/nfit/core.c

   1 /*
   2  * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of version 2 of the GNU General Public License as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  */
  13 #include <linux/list_sort.h>
  14 #include <linux/libnvdimm.h>
  15 #include <linux/module.h>
  16 #include <linux/mutex.h>
  17 #include <linux/ndctl.h>
  18 #include <linux/sysfs.h>
  19 #include <linux/delay.h>
  20 #include <linux/list.h>
  21 #include <linux/acpi.h>
  22 #include <linux/sort.h>
  23 #include <linux/io.h>
  24 #include <linux/nd.h>
  25 #include <asm/cacheflush.h>
  26 #include <acpi/nfit.h>
  27 #include "nfit.h"
  28 #include "intel.h"
  29
  30 /*
  31  * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
  32  * irrelevant.
  33  */
  34 #include <linux/io-64-nonatomic-hi-lo.h>
  35
  36 static bool force_enable_dimms;
  37 module_param(force_enable_dimms, bool, S_IRUGO|S_IWUSR);
  38 MODULE_PARM_DESC(force_enable_dimms, "Ignore _STA (ACPI DIMM device) status");
  39
  40 static bool disable_vendor_specific;
  41 module_param(disable_vendor_specific, bool, S_IRUGO);
  42 MODULE_PARM_DESC(disable_vendor_specific,
  43                 "Limit commands to the publicly specified set");
  44
  45 static unsigned long override_dsm_mask;
  46 module_param(override_dsm_mask, ulong, S_IRUGO);
  47 MODULE_PARM_DESC(override_dsm_mask, "Bitmask of allowed NVDIMM DSM functions");
  48
  49 static int default_dsm_family = -1;
  50 module_param(default_dsm_family, int, S_IRUGO);
  51 MODULE_PARM_DESC(default_dsm_family,
  52                 "Try this DSM type first when identifying NVDIMM family");
  53
  54 static bool no_init_ars;
  55 module_param(no_init_ars, bool, 0644);
  56 MODULE_PARM_DESC(no_init_ars, "Skip ARS run at nfit init time");
  57
  58 LIST_HEAD(acpi_descs);
  59 DEFINE_MUTEX(acpi_desc_lock);
  60
  61 static struct workqueue_struct *nfit_wq;
  62
  63 struct nfit_table_prev {
  64         struct list_head spas;
  65         struct list_head memdevs;
  66         struct list_head dcrs;
  67         struct list_head bdws;
  68         struct list_head idts;
  69         struct list_head flushes;
  70 };
  71
  72 static guid_t nfit_uuid[NFIT_UUID_MAX];
  73
  74 const guid_t *to_nfit_uuid(enum nfit_uuids id)
  75 {
  76         return &nfit_uuid[id];
  77 }
  78 EXPORT_SYMBOL(to_nfit_uuid);
  79
  80 static struct acpi_nfit_desc *to_acpi_nfit_desc(
  81                 struct nvdimm_bus_descriptor *nd_desc)
  82 {
  83         return container_of(nd_desc, struct acpi_nfit_desc, nd_desc);
  84 }
  85
  86 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
  87 {
  88         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
  89
  90         /*
  91          * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
  92          * acpi_device.
  93          */
  94         if (!nd_desc->provider_name
  95                         || strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
  96                 return NULL;
  97
  98         return to_acpi_device(acpi_desc->dev);
  99 }
 100
 101 static int xlat_bus_status(void *buf, unsigned int cmd, u32 status)
 102 {
 103         struct nd_cmd_clear_error *clear_err;
 104         struct nd_cmd_ars_status *ars_status;
 105         u16 flags;
 106
 107         switch (cmd) {
 108         case ND_CMD_ARS_CAP:
 109                 if ((status & 0xffff) == NFIT_ARS_CAP_NONE)
 110                         return -ENOTTY;
 111
 112                 /* Command failed */
 113                 if (status & 0xffff)
 114                         return -EIO;
 115
 116                 /* No supported scan types for this range */
 117                 flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
 118                 if ((status >> 16 & flags) == 0)
 119                         return -ENOTTY;
 120                 return 0;
 121         case ND_CMD_ARS_START:
 122                 /* ARS is in progress */
 123                 if ((status & 0xffff) == NFIT_ARS_START_BUSY)
 124                         return -EBUSY;
 125
 126                 /* Command failed */
 127                 if (status & 0xffff)
 128                         return -EIO;
 129                 return 0;
 130         case ND_CMD_ARS_STATUS:
 131                 ars_status = buf;
 132                 /* Command failed */
 133                 if (status & 0xffff)
 134                         return -EIO;
 135                 /* Check extended status (Upper two bytes) */
 136                 if (status == NFIT_ARS_STATUS_DONE)
 137                         return 0;
 138
 139                 /* ARS is in progress */
 140                 if (status == NFIT_ARS_STATUS_BUSY)
 141                         return -EBUSY;
 142
 143                 /* No ARS performed for the current boot */
 144                 if (status == NFIT_ARS_STATUS_NONE)
 145                         return -EAGAIN;
 146
 147                 /*
 148                  * ARS interrupted, either we overflowed or some other
 149                  * agent wants the scan to stop.  If we didn't overflow
 150                  * then just continue with the returned results.
 151                  */
 152                 if (status == NFIT_ARS_STATUS_INTR) {
 153                         if (ars_status->out_length >= 40 && (ars_status->flags
 154                                                 & NFIT_ARS_F_OVERFLOW))
 155                                 return -ENOSPC;
 156                         return 0;
 157                 }
 158
 159                 /* Unknown status */
 160                 if (status >> 16)
 161                         return -EIO;
 162                 return 0;
 163         case ND_CMD_CLEAR_ERROR:
 164                 clear_err = buf;
 165                 if (status & 0xffff)
 166                         return -EIO;
 167                 if (!clear_err->cleared)
 168                         return -EIO;
 169                 if (clear_err->length > clear_err->cleared)
 170                         return clear_err->cleared;
 171                 return 0;
 172         default:
 173                 break;
 174         }
 175
 176         /* all other non-zero status results in an error */
 177         if (status)
 178                 return -EIO;
 179         return 0;
 180 }
 181
 182 #define ACPI_LABELS_LOCKED 3
 183
 184 static int xlat_nvdimm_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
 185                 u32 status)
 186 {
 187         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 188
 189         switch (cmd) {
 190         case ND_CMD_GET_CONFIG_SIZE:
 191                 /*
 192                  * In the _LSI, _LSR, _LSW case the locked status is
 193                  * communicated via the read/write commands
 194                  */
 195                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags))
 196                         break;
 197
 198                 if (status >> 16 & ND_CONFIG_LOCKED)
 199                         return -EACCES;
 200                 break;
 201         case ND_CMD_GET_CONFIG_DATA:
 202                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
 203                                 && status == ACPI_LABELS_LOCKED)
 204                         return -EACCES;
 205                 break;
 206         case ND_CMD_SET_CONFIG_DATA:
 207                 if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags)
 208                                 && status == ACPI_LABELS_LOCKED)
 209                         return -EACCES;
 210                 break;
 211         default:
 212                 break;
 213         }
 214
 215         /* all other non-zero status results in an error */
 216         if (status)
 217                 return -EIO;
 218         return 0;
 219 }
 220
 221 static int xlat_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
 222                 u32 status)
 223 {
 224         if (!nvdimm)
 225                 return xlat_bus_status(buf, cmd, status);
 226         return xlat_nvdimm_status(nvdimm, buf, cmd, status);
 227 }
 228
 229 /* convert _LS{I,R} packages to the buffer object acpi_nfit_ctl expects */
 230 static union acpi_object *pkg_to_buf(union acpi_object *pkg)
 231 {
 232         int i;
 233         void *dst;
 234         size_t size = 0;
 235         union acpi_object *buf = NULL;
 236
 237         if (pkg->type != ACPI_TYPE_PACKAGE) {
 238                 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
 239                                 pkg->type);
 240                 goto err;
 241         }
 242
 243         for (i = 0; i < pkg->package.count; i++) {
 244                 union acpi_object *obj = &pkg->package.elements[i];
 245
 246                 if (obj->type == ACPI_TYPE_INTEGER)
 247                         size += 4;
 248                 else if (obj->type == ACPI_TYPE_BUFFER)
 249                         size += obj->buffer.length;
 250                 else {
 251                         WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
 252                                         obj->type);
 253                         goto err;
 254                 }
 255         }
 256
 257         buf = ACPI_ALLOCATE(sizeof(*buf) + size);
 258         if (!buf)
 259                 goto err;
 260
 261         dst = buf + 1;
 262         buf->type = ACPI_TYPE_BUFFER;
 263         buf->buffer.length = size;
 264         buf->buffer.pointer = dst;
 265         for (i = 0; i < pkg->package.count; i++) {
 266                 union acpi_object *obj = &pkg->package.elements[i];
 267
 268                 if (obj->type == ACPI_TYPE_INTEGER) {
 269                         memcpy(dst, &obj->integer.value, 4);
 270                         dst += 4;
 271                 } else if (obj->type == ACPI_TYPE_BUFFER) {
 272                         memcpy(dst, obj->buffer.pointer, obj->buffer.length);
 273                         dst += obj->buffer.length;
 274                 }
 275         }
 276 err:
 277         ACPI_FREE(pkg);
 278         return buf;
 279 }
 280
 281 static union acpi_object *int_to_buf(union acpi_object *integer)
 282 {
 283         union acpi_object *buf = ACPI_ALLOCATE(sizeof(*buf) + 4);
 284         void *dst = NULL;
 285
 286         if (!buf)
 287                 goto err;
 288
 289         if (integer->type != ACPI_TYPE_INTEGER) {
 290                 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
 291                                 integer->type);
 292                 goto err;
 293         }
 294
 295         dst = buf + 1;
 296         buf->type = ACPI_TYPE_BUFFER;
 297         buf->buffer.length = 4;
 298         buf->buffer.pointer = dst;
 299         memcpy(dst, &integer->integer.value, 4);
 300 err:
 301         ACPI_FREE(integer);
 302         return buf;
 303 }
 304
 305 static union acpi_object *acpi_label_write(acpi_handle handle, u32 offset,
 306                 u32 len, void *data)
 307 {
 308         acpi_status rc;
 309         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
 310         struct acpi_object_list input = {
 311                 .count = 3,
 312                 .pointer = (union acpi_object []) {
 313                         [0] = {
 314                                 .integer.type = ACPI_TYPE_INTEGER,
 315                                 .integer.value = offset,
 316                         },
 317                         [1] = {
 318                                 .integer.type = ACPI_TYPE_INTEGER,
 319                                 .integer.value = len,
 320                         },
 321                         [2] = {
 322                                 .buffer.type = ACPI_TYPE_BUFFER,
 323                                 .buffer.pointer = data,
 324                                 .buffer.length = len,
 325                         },
 326                 },
 327         };
 328
 329         rc = acpi_evaluate_object(handle, "_LSW", &input, &buf);
 330         if (ACPI_FAILURE(rc))
 331                 return NULL;
 332         return int_to_buf(buf.pointer);
 333 }
 334
 335 static union acpi_object *acpi_label_read(acpi_handle handle, u32 offset,
 336                 u32 len)
 337 {
 338         acpi_status rc;
 339         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
 340         struct acpi_object_list input = {
 341                 .count = 2,
 342                 .pointer = (union acpi_object []) {
 343                         [0] = {
 344                                 .integer.type = ACPI_TYPE_INTEGER,
 345                                 .integer.value = offset,
 346                         },
 347                         [1] = {
 348                                 .integer.type = ACPI_TYPE_INTEGER,
 349                                 .integer.value = len,
 350                         },
 351                 },
 352         };
 353
 354         rc = acpi_evaluate_object(handle, "_LSR", &input, &buf);
 355         if (ACPI_FAILURE(rc))
 356                 return NULL;
 357         return pkg_to_buf(buf.pointer);
 358 }
 359
 360 static union acpi_object *acpi_label_info(acpi_handle handle)
 361 {
 362         acpi_status rc;
 363         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
 364
 365         rc = acpi_evaluate_object(handle, "_LSI", NULL, &buf);
 366         if (ACPI_FAILURE(rc))
 367                 return NULL;
 368         return pkg_to_buf(buf.pointer);
 369 }
 370
 371 static u8 nfit_dsm_revid(unsigned family, unsigned func)
 372 {
 373         static const u8 revid_table[NVDIMM_FAMILY_MAX+1][32] = {
 374                 [NVDIMM_FAMILY_INTEL] = {
 375                         [NVDIMM_INTEL_GET_MODES] = 2,
 376                         [NVDIMM_INTEL_GET_FWINFO] = 2,
 377                         [NVDIMM_INTEL_START_FWUPDATE] = 2,
 378                         [NVDIMM_INTEL_SEND_FWUPDATE] = 2,
 379                         [NVDIMM_INTEL_FINISH_FWUPDATE] = 2,
 380                         [NVDIMM_INTEL_QUERY_FWUPDATE] = 2,
 381                         [NVDIMM_INTEL_SET_THRESHOLD] = 2,
 382                         [NVDIMM_INTEL_INJECT_ERROR] = 2,
 383                 },
 384         };
 385         u8 id;
 386
 387         if (family > NVDIMM_FAMILY_MAX)
 388                 return 0;
 389         if (func > 31)
 390                 return 0;
 391         id = revid_table[family][func];
 392         if (id == 0)
 393                 return 1; /* default */
 394         return id;
 395 }
 396
 397 int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm,
 398                 unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc)
 399 {
 400         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
 401         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 402         union acpi_object in_obj, in_buf, *out_obj;
 403         const struct nd_cmd_desc *desc = NULL;
 404         struct device *dev = acpi_desc->dev;
 405         struct nd_cmd_pkg *call_pkg = NULL;
 406         const char *cmd_name, *dimm_name;
 407         unsigned long cmd_mask, dsm_mask;
 408         u32 offset, fw_status = 0;
 409         acpi_handle handle;
 410         unsigned int func;
 411         const guid_t *guid;
 412         int rc, i;
 413
 414         if (cmd_rc)
 415                 *cmd_rc = -EINVAL;
 416         func = cmd;
 417         if (cmd == ND_CMD_CALL) {
 418                 call_pkg = buf;
 419                 func = call_pkg->nd_command;
 420
 421                 for (i = 0; i < ARRAY_SIZE(call_pkg->nd_reserved2); i++)
 422                         if (call_pkg->nd_reserved2[i])
 423                                 return -EINVAL;
 424         }
 425
 426         if (nvdimm) {
 427                 struct acpi_device *adev = nfit_mem->adev;
 428
 429                 if (!adev)
 430                         return -ENOTTY;
 431                 if (call_pkg && nfit_mem->family != call_pkg->nd_family)
 432                         return -ENOTTY;
 433
 434                 dimm_name = nvdimm_name(nvdimm);
 435                 cmd_name = nvdimm_cmd_name(cmd);
 436                 cmd_mask = nvdimm_cmd_mask(nvdimm);
 437                 dsm_mask = nfit_mem->dsm_mask;
 438                 desc = nd_cmd_dimm_desc(cmd);
 439                 guid = to_nfit_uuid(nfit_mem->family);
 440                 handle = adev->handle;
 441         } else {
 442                 struct acpi_device *adev = to_acpi_dev(acpi_desc);
 443
 444                 cmd_name = nvdimm_bus_cmd_name(cmd);
 445                 cmd_mask = nd_desc->cmd_mask;
 446                 dsm_mask = cmd_mask;
 447                 if (cmd == ND_CMD_CALL)
 448                         dsm_mask = nd_desc->bus_dsm_mask;
 449                 desc = nd_cmd_bus_desc(cmd);
 450                 guid = to_nfit_uuid(NFIT_DEV_BUS);
 451                 handle = adev->handle;
 452                 dimm_name = "bus";
 453         }
 454
 455         if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
 456                 return -ENOTTY;
 457
 458         if (!test_bit(cmd, &cmd_mask) || !test_bit(func, &dsm_mask))
 459                 return -ENOTTY;
 460
 461         in_obj.type = ACPI_TYPE_PACKAGE;
 462         in_obj.package.count = 1;
 463         in_obj.package.elements = &in_buf;
 464         in_buf.type = ACPI_TYPE_BUFFER;
 465         in_buf.buffer.pointer = buf;
 466         in_buf.buffer.length = 0;
 467
 468         /* libnvdimm has already validated the input envelope */
 469         for (i = 0; i < desc->in_num; i++)
 470                 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
 471                                 i, buf);
 472
 473         if (call_pkg) {
 474                 /* skip over package wrapper */
 475                 in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
 476                 in_buf.buffer.length = call_pkg->nd_size_in;
 477         }
 478
 479         dev_dbg(dev, "%s cmd: %d: func: %d input length: %d\n",
 480                 dimm_name, cmd, func, in_buf.buffer.length);
 481         print_hex_dump_debug("nvdimm in  ", DUMP_PREFIX_OFFSET, 4, 4,
 482                         in_buf.buffer.pointer,
 483                         min_t(u32, 256, in_buf.buffer.length), true);
 484
 485         /* call the BIOS, prefer the named methods over _DSM if available */
 486         if (nvdimm && cmd == ND_CMD_GET_CONFIG_SIZE
 487                         && test_bit(NFIT_MEM_LSR, &nfit_mem->flags))
 488                 out_obj = acpi_label_info(handle);
 489         else if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA
 490                         && test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
 491                 struct nd_cmd_get_config_data_hdr *p = buf;
 492
 493                 out_obj = acpi_label_read(handle, p->in_offset, p->in_length);
 494         } else if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA
 495                         && test_bit(NFIT_MEM_LSW, &nfit_mem->flags)) {
 496                 struct nd_cmd_set_config_hdr *p = buf;
 497
 498                 out_obj = acpi_label_write(handle, p->in_offset, p->in_length,
 499                                 p->in_buf);
 500         } else {
 501                 u8 revid;
 502
 503                 if (nvdimm)
 504                         revid = nfit_dsm_revid(nfit_mem->family, func);
 505                 else
 506                         revid = 1;
 507                 out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
 508         }
 509
 510         if (!out_obj) {
 511                 dev_dbg(dev, "%s _DSM failed cmd: %s\n", dimm_name, cmd_name);
 512                 return -EINVAL;
 513         }
 514
 515         if (call_pkg) {
 516                 call_pkg->nd_fw_size = out_obj->buffer.length;
 517                 memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
 518                         out_obj->buffer.pointer,
 519                         min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
 520
 521                 ACPI_FREE(out_obj);
 522                 /*
 523                  * Need to support FW function w/o known size in advance.
 524                  * Caller can determine required size based upon nd_fw_size.
 525                  * If we return an error (like elsewhere) then caller wouldn't
 526                  * be able to rely upon data returned to make calculation.
 527                  */
 528                 if (cmd_rc)
 529                         *cmd_rc = 0;
 530                 return 0;
 531         }
 532
 533         if (out_obj->package.type != ACPI_TYPE_BUFFER) {
 534                 dev_dbg(dev, "%s unexpected output object type cmd: %s type: %d\n",
 535                                 dimm_name, cmd_name, out_obj->type);
 536                 rc = -EINVAL;
 537                 goto out;
 538         }
 539
 540         dev_dbg(dev, "%s cmd: %s output length: %d\n", dimm_name,
 541                         cmd_name, out_obj->buffer.length);
 542         print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4, 4,
 543                         out_obj->buffer.pointer,
 544                         min_t(u32, 128, out_obj->buffer.length), true);
 545
 546         for (i = 0, offset = 0; i < desc->out_num; i++) {
 547                 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
 548                                 (u32 *) out_obj->buffer.pointer,
 549                                 out_obj->buffer.length - offset);
 550
 551                 if (offset + out_size > out_obj->buffer.length) {
 552                         dev_dbg(dev, "%s output object underflow cmd: %s field: %d\n",
 553                                         dimm_name, cmd_name, i);
 554                         break;
 555                 }
 556
 557                 if (in_buf.buffer.length + offset + out_size > buf_len) {
 558                         dev_dbg(dev, "%s output overrun cmd: %s field: %d\n",
 559                                         dimm_name, cmd_name, i);
 560                         rc = -ENXIO;
 561                         goto out;
 562                 }
 563                 memcpy(buf + in_buf.buffer.length + offset,
 564                                 out_obj->buffer.pointer + offset, out_size);
 565                 offset += out_size;
 566         }
 567
 568         /*
 569          * Set fw_status for all the commands with a known format to be
 570          * later interpreted by xlat_status().
 571          */
 572         if (i >= 1 && ((!nvdimm && cmd >= ND_CMD_ARS_CAP
 573                                         && cmd <= ND_CMD_CLEAR_ERROR)
 574                                 || (nvdimm && cmd >= ND_CMD_SMART
 575                                         && cmd <= ND_CMD_VENDOR)))
 576                 fw_status = *(u32 *) out_obj->buffer.pointer;
 577
 578         if (offset + in_buf.buffer.length < buf_len) {
 579                 if (i >= 1) {
 580                         /*
 581                          * status valid, return the number of bytes left
 582                          * unfilled in the output buffer
 583                          */
 584                         rc = buf_len - offset - in_buf.buffer.length;
 585                         if (cmd_rc)
 586                                 *cmd_rc = xlat_status(nvdimm, buf, cmd,
 587                                                 fw_status);
 588                 } else {
 589                         dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
 590                                         __func__, dimm_name, cmd_name, buf_len,
 591                                         offset);
 592                         rc = -ENXIO;
 593                 }
 594         } else {
 595                 rc = 0;
 596                 if (cmd_rc)
 597                         *cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status);
 598         }
 599
 600  out:
 601         ACPI_FREE(out_obj);
 602
 603         return rc;
 604 }
 605 EXPORT_SYMBOL_GPL(acpi_nfit_ctl);
 606
 607 static const char *spa_type_name(u16 type)
 608 {
 609         static const char *to_name[] = {
 610                 [NFIT_SPA_VOLATILE] = "volatile",
 611                 [NFIT_SPA_PM] = "pmem",
 612                 [NFIT_SPA_DCR] = "dimm-control-region",
 613                 [NFIT_SPA_BDW] = "block-data-window",
 614                 [NFIT_SPA_VDISK] = "volatile-disk",
 615                 [NFIT_SPA_VCD] = "volatile-cd",
 616                 [NFIT_SPA_PDISK] = "persistent-disk",
 617                 [NFIT_SPA_PCD] = "persistent-cd",
 618
 619         };
 620
 621         if (type > NFIT_SPA_PCD)
 622                 return "unknown";
 623
 624         return to_name[type];
 625 }
 626
 627 int nfit_spa_type(struct acpi_nfit_system_address *spa)
 628 {
 629         int i;
 630
 631         for (i = 0; i < NFIT_UUID_MAX; i++)
 632                 if (guid_equal(to_nfit_uuid(i), (guid_t *)&spa->range_guid))
 633                         return i;
 634         return -1;
 635 }
 636
 637 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
 638                 struct nfit_table_prev *prev,
 639                 struct acpi_nfit_system_address *spa)
 640 {
 641         struct device *dev = acpi_desc->dev;
 642         struct nfit_spa *nfit_spa;
 643
 644         if (spa->header.length != sizeof(*spa))
 645                 return false;
 646
 647         list_for_each_entry(nfit_spa, &prev->spas, list) {
 648                 if (memcmp(nfit_spa->spa, spa, sizeof(*spa)) == 0) {
 649                         list_move_tail(&nfit_spa->list, &acpi_desc->spas);
 650                         return true;
 651                 }
 652         }
 653
 654         nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof(*spa),
 655                         GFP_KERNEL);
 656         if (!nfit_spa)
 657                 return false;
 658         INIT_LIST_HEAD(&nfit_spa->list);
 659         memcpy(nfit_spa->spa, spa, sizeof(*spa));
 660         list_add_tail(&nfit_spa->list, &acpi_desc->spas);
 661         dev_dbg(dev, "spa index: %d type: %s\n",
 662                         spa->range_index,
 663                         spa_type_name(nfit_spa_type(spa)));
 664         return true;
 665 }
 666
 667 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
 668                 struct nfit_table_prev *prev,
 669                 struct acpi_nfit_memory_map *memdev)
 670 {
 671         struct device *dev = acpi_desc->dev;
 672         struct nfit_memdev *nfit_memdev;
 673
 674         if (memdev->header.length != sizeof(*memdev))
 675                 return false;
 676
 677         list_for_each_entry(nfit_memdev, &prev->memdevs, list)
 678                 if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
 679                         list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 680                         return true;
 681                 }
 682
 683         nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
 684                         GFP_KERNEL);
 685         if (!nfit_memdev)
 686                 return false;
 687         INIT_LIST_HEAD(&nfit_memdev->list);
 688         memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
 689         list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 690         dev_dbg(dev, "memdev handle: %#x spa: %d dcr: %d flags: %#x\n",
 691                         memdev->device_handle, memdev->range_index,
 692                         memdev->region_index, memdev->flags);
 693         return true;
 694 }
 695
 696 int nfit_get_smbios_id(u32 device_handle, u16 *flags)
 697 {
 698         struct acpi_nfit_memory_map *memdev;
 699         struct acpi_nfit_desc *acpi_desc;
 700         struct nfit_mem *nfit_mem;
 701
 702         mutex_lock(&acpi_desc_lock);
 703         list_for_each_entry(acpi_desc, &acpi_descs, list) {
 704                 mutex_lock(&acpi_desc->init_mutex);
 705                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
 706                         memdev = __to_nfit_memdev(nfit_mem);
 707                         if (memdev->device_handle == device_handle) {
 708                                 mutex_unlock(&acpi_desc->init_mutex);
 709                                 mutex_unlock(&acpi_desc_lock);
 710                                 *flags = memdev->flags;
 711                                 return memdev->physical_id;
 712                         }
 713                 }
 714                 mutex_unlock(&acpi_desc->init_mutex);
 715         }
 716         mutex_unlock(&acpi_desc_lock);
 717
 718         return -ENODEV;
 719 }
 720 EXPORT_SYMBOL_GPL(nfit_get_smbios_id);
 721
 722 /*
 723  * An implementation may provide a truncated control region if no block windows
 724  * are defined.
 725  */
 726 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
 727 {
 728         if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
 729                                 window_size))
 730                 return 0;
 731         if (dcr->windows)
 732                 return sizeof(*dcr);
 733         return offsetof(struct acpi_nfit_control_region, window_size);
 734 }
 735
 736 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
 737                 struct nfit_table_prev *prev,
 738                 struct acpi_nfit_control_region *dcr)
 739 {
 740         struct device *dev = acpi_desc->dev;
 741         struct nfit_dcr *nfit_dcr;
 742
 743         if (!sizeof_dcr(dcr))
 744                 return false;
 745
 746         list_for_each_entry(nfit_dcr, &prev->dcrs, list)
 747                 if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
 748                         list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 749                         return true;
 750                 }
 751
 752         nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
 753                         GFP_KERNEL);
 754         if (!nfit_dcr)
 755                 return false;
 756         INIT_LIST_HEAD(&nfit_dcr->list);
 757         memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
 758         list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 759         dev_dbg(dev, "dcr index: %d windows: %d\n",
 760                         dcr->region_index, dcr->windows);
 761         return true;
 762 }
 763
 764 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
 765                 struct nfit_table_prev *prev,
 766                 struct acpi_nfit_data_region *bdw)
 767 {
 768         struct device *dev = acpi_desc->dev;
 769         struct nfit_bdw *nfit_bdw;
 770
 771         if (bdw->header.length != sizeof(*bdw))
 772                 return false;
 773         list_for_each_entry(nfit_bdw, &prev->bdws, list)
 774                 if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
 775                         list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
 776                         return true;
 777                 }
 778
 779         nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
 780                         GFP_KERNEL);
 781         if (!nfit_bdw)
 782                 return false;
 783         INIT_LIST_HEAD(&nfit_bdw->list);
 784         memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
 785         list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
 786         dev_dbg(dev, "bdw dcr: %d windows: %d\n",
 787                         bdw->region_index, bdw->windows);
 788         return true;
 789 }
 790
 791 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
 792 {
 793         if (idt->header.length < sizeof(*idt))
 794                 return 0;
 795         return sizeof(*idt) + sizeof(u32) * (idt->line_count - 1);
 796 }
 797
 798 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
 799                 struct nfit_table_prev *prev,
 800                 struct acpi_nfit_interleave *idt)
 801 {
 802         struct device *dev = acpi_desc->dev;
 803         struct nfit_idt *nfit_idt;
 804
 805         if (!sizeof_idt(idt))
 806                 return false;
 807
 808         list_for_each_entry(nfit_idt, &prev->idts, list) {
 809                 if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
 810                         continue;
 811
 812                 if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
 813                         list_move_tail(&nfit_idt->list, &acpi_desc->idts);
 814                         return true;
 815                 }
 816         }
 817
 818         nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
 819                         GFP_KERNEL);
 820         if (!nfit_idt)
 821                 return false;
 822         INIT_LIST_HEAD(&nfit_idt->list);
 823         memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
 824         list_add_tail(&nfit_idt->list, &acpi_desc->idts);
 825         dev_dbg(dev, "idt index: %d num_lines: %d\n",
 826                         idt->interleave_index, idt->line_count);
 827         return true;
 828 }
 829
 830 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
 831 {
 832         if (flush->header.length < sizeof(*flush))
 833                 return 0;
 834         return sizeof(*flush) + sizeof(u64) * (flush->hint_count - 1);
 835 }
 836
 837 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
 838                 struct nfit_table_prev *prev,
 839                 struct acpi_nfit_flush_address *flush)
 840 {
 841         struct device *dev = acpi_desc->dev;
 842         struct nfit_flush *nfit_flush;
 843
 844         if (!sizeof_flush(flush))
 845                 return false;
 846
 847         list_for_each_entry(nfit_flush, &prev->flushes, list) {
 848                 if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
 849                         continue;
 850
 851                 if (memcmp(nfit_flush->flush, flush,
 852                                         sizeof_flush(flush)) == 0) {
 853                         list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
 854                         return true;
 855                 }
 856         }
 857
 858         nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
 859                         + sizeof_flush(flush), GFP_KERNEL);
 860         if (!nfit_flush)
 861                 return false;
 862         INIT_LIST_HEAD(&nfit_flush->list);
 863         memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
 864         list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
 865         dev_dbg(dev, "nfit_flush handle: %d hint_count: %d\n",
 866                         flush->device_handle, flush->hint_count);
 867         return true;
 868 }
 869
 870 static bool add_platform_cap(struct acpi_nfit_desc *acpi_desc,
 871                 struct acpi_nfit_capabilities *pcap)
 872 {
 873         struct device *dev = acpi_desc->dev;
 874         u32 mask;
 875
 876         mask = (1 << (pcap->highest_capability + 1)) - 1;
 877         acpi_desc->platform_cap = pcap->capabilities & mask;
 878         dev_dbg(dev, "cap: %#x\n", acpi_desc->platform_cap);
 879         return true;
 880 }
 881
 882 static void *add_table(struct acpi_nfit_desc *acpi_desc,
 883                 struct nfit_table_prev *prev, void *table, const void *end)
 884 {
 885         struct device *dev = acpi_desc->dev;
 886         struct acpi_nfit_header *hdr;
 887         void *err = ERR_PTR(-ENOMEM);
 888
 889         if (table >= end)
 890                 return NULL;
 891
 892         hdr = table;
 893         if (!hdr->length) {
 894                 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
 895                         hdr->type);
 896                 return NULL;
 897         }
 898
 899         switch (hdr->type) {
 900         case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
 901                 if (!add_spa(acpi_desc, prev, table))
 902                         return err;
 903                 break;
 904         case ACPI_NFIT_TYPE_MEMORY_MAP:
 905                 if (!add_memdev(acpi_desc, prev, table))
 906                         return err;
 907                 break;
 908         case ACPI_NFIT_TYPE_CONTROL_REGION:
 909                 if (!add_dcr(acpi_desc, prev, table))
 910                         return err;
 911                 break;
 912         case ACPI_NFIT_TYPE_DATA_REGION:
 913                 if (!add_bdw(acpi_desc, prev, table))
 914                         return err;
 915                 break;
 916         case ACPI_NFIT_TYPE_INTERLEAVE:
 917                 if (!add_idt(acpi_desc, prev, table))
 918                         return err;
 919                 break;
 920         case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
 921                 if (!add_flush(acpi_desc, prev, table))
 922                         return err;
 923                 break;
 924         case ACPI_NFIT_TYPE_SMBIOS:
 925                 dev_dbg(dev, "smbios\n");
 926                 break;
 927         case ACPI_NFIT_TYPE_CAPABILITIES:
 928                 if (!add_platform_cap(acpi_desc, table))
 929                         return err;
 930                 break;
 931         default:
 932                 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
 933                 break;
 934         }
 935
 936         return table + hdr->length;
 937 }
 938
 939 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
 940                 struct nfit_mem *nfit_mem)
 941 {
 942         u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
 943         u16 dcr = nfit_mem->dcr->region_index;
 944         struct nfit_spa *nfit_spa;
 945
 946         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
 947                 u16 range_index = nfit_spa->spa->range_index;
 948                 int type = nfit_spa_type(nfit_spa->spa);
 949                 struct nfit_memdev *nfit_memdev;
 950
 951                 if (type != NFIT_SPA_BDW)
 952                         continue;
 953
 954                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 955                         if (nfit_memdev->memdev->range_index != range_index)
 956                                 continue;
 957                         if (nfit_memdev->memdev->device_handle != device_handle)
 958                                 continue;
 959                         if (nfit_memdev->memdev->region_index != dcr)
 960                                 continue;
 961
 962                         nfit_mem->spa_bdw = nfit_spa->spa;
 963                         return;
 964                 }
 965         }
 966
 967         dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
 968                         nfit_mem->spa_dcr->range_index);
 969         nfit_mem->bdw = NULL;
 970 }
 971
 972 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
 973                 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
 974 {
 975         u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
 976         struct nfit_memdev *nfit_memdev;
 977         struct nfit_bdw *nfit_bdw;
 978         struct nfit_idt *nfit_idt;
 979         u16 idt_idx, range_index;
 980
 981         list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
 982                 if (nfit_bdw->bdw->region_index != dcr)
 983                         continue;
 984                 nfit_mem->bdw = nfit_bdw->bdw;
 985                 break;
 986         }
 987
 988         if (!nfit_mem->bdw)
 989                 return;
 990
 991         nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
 992
 993         if (!nfit_mem->spa_bdw)
 994                 return;
 995
 996         range_index = nfit_mem->spa_bdw->range_index;
 997         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 998                 if (nfit_memdev->memdev->range_index != range_index ||
 999                                 nfit_memdev->memdev->region_index != dcr)
1000                         continue;
1001                 nfit_mem->memdev_bdw = nfit_memdev->memdev;
1002                 idt_idx = nfit_memdev->memdev->interleave_index;
1003                 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1004                         if (nfit_idt->idt->interleave_index != idt_idx)
1005                                 continue;
1006                         nfit_mem->idt_bdw = nfit_idt->idt;
1007                         break;
1008                 }
1009                 break;
1010         }
1011 }
1012
1013 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
1014                 struct acpi_nfit_system_address *spa)
1015 {
1016         struct nfit_mem *nfit_mem, *found;
1017         struct nfit_memdev *nfit_memdev;
1018         int type = spa ? nfit_spa_type(spa) : 0;
1019
1020         switch (type) {
1021         case NFIT_SPA_DCR:
1022         case NFIT_SPA_PM:
1023                 break;
1024         default:
1025                 if (spa)
1026                         return 0;
1027         }
1028
1029         /*
1030          * This loop runs in two modes, when a dimm is mapped the loop
1031          * adds memdev associations to an existing dimm, or creates a
1032          * dimm. In the unmapped dimm case this loop sweeps for memdev
1033          * instances with an invalid / zero range_index and adds those
1034          * dimms without spa associations.
1035          */
1036         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1037                 struct nfit_flush *nfit_flush;
1038                 struct nfit_dcr *nfit_dcr;
1039                 u32 device_handle;
1040                 u16 dcr;
1041
1042                 if (spa && nfit_memdev->memdev->range_index != spa->range_index)
1043                         continue;
1044                 if (!spa && nfit_memdev->memdev->range_index)
1045                         continue;
1046                 found = NULL;
1047                 dcr = nfit_memdev->memdev->region_index;
1048                 device_handle = nfit_memdev->memdev->device_handle;
1049                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1050                         if (__to_nfit_memdev(nfit_mem)->device_handle
1051                                         == device_handle) {
1052                                 found = nfit_mem;
1053                                 break;
1054                         }
1055
1056                 if (found)
1057                         nfit_mem = found;
1058                 else {
1059                         nfit_mem = devm_kzalloc(acpi_desc->dev,
1060                                         sizeof(*nfit_mem), GFP_KERNEL);
1061                         if (!nfit_mem)
1062                                 return -ENOMEM;
1063                         INIT_LIST_HEAD(&nfit_mem->list);
1064                         nfit_mem->acpi_desc = acpi_desc;
1065                         list_add(&nfit_mem->list, &acpi_desc->dimms);
1066                 }
1067
1068                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1069                         if (nfit_dcr->dcr->region_index != dcr)
1070                                 continue;
1071                         /*
1072                          * Record the control region for the dimm.  For
1073                          * the ACPI 6.1 case, where there are separate
1074                          * control regions for the pmem vs blk
1075                          * interfaces, be sure to record the extended
1076                          * blk details.
1077                          */
1078                         if (!nfit_mem->dcr)
1079                                 nfit_mem->dcr = nfit_dcr->dcr;
1080                         else if (nfit_mem->dcr->windows == 0
1081                                         && nfit_dcr->dcr->windows)
1082                                 nfit_mem->dcr = nfit_dcr->dcr;
1083                         break;
1084                 }
1085
1086                 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
1087                         struct acpi_nfit_flush_address *flush;
1088                         u16 i;
1089
1090                         if (nfit_flush->flush->device_handle != device_handle)
1091                                 continue;
1092                         nfit_mem->nfit_flush = nfit_flush;
1093                         flush = nfit_flush->flush;
1094                         nfit_mem->flush_wpq = devm_kcalloc(acpi_desc->dev,
1095                                         flush->hint_count,
1096                                         sizeof(struct resource),
1097                                         GFP_KERNEL);
1098                         if (!nfit_mem->flush_wpq)
1099                                 return -ENOMEM;
1100                         for (i = 0; i < flush->hint_count; i++) {
1101                                 struct resource *res = &nfit_mem->flush_wpq[i];
1102
1103                                 res->start = flush->hint_address[i];
1104                                 res->end = res->start + 8 - 1;
1105                         }
1106                         break;
1107                 }
1108
1109                 if (dcr && !nfit_mem->dcr) {
1110                         dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
1111                                         spa->range_index, dcr);
1112                         return -ENODEV;
1113                 }
1114
1115                 if (type == NFIT_SPA_DCR) {
1116                         struct nfit_idt *nfit_idt;
1117                         u16 idt_idx;
1118
1119                         /* multiple dimms may share a SPA when interleaved */
1120                         nfit_mem->spa_dcr = spa;
1121                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
1122                         idt_idx = nfit_memdev->memdev->interleave_index;
1123                         list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1124                                 if (nfit_idt->idt->interleave_index != idt_idx)
1125                                         continue;
1126                                 nfit_mem->idt_dcr = nfit_idt->idt;
1127                                 break;
1128                         }
1129                         nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
1130                 } else if (type == NFIT_SPA_PM) {
1131                         /*
1132                          * A single dimm may belong to multiple SPA-PM
1133                          * ranges, record at least one in addition to
1134                          * any SPA-DCR range.
1135                          */
1136                         nfit_mem->memdev_pmem = nfit_memdev->memdev;
1137                 } else
1138                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
1139         }
1140
1141         return 0;
1142 }
1143
1144 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
1145 {
1146         struct nfit_mem *a = container_of(_a, typeof(*a), list);
1147         struct nfit_mem *b = container_of(_b, typeof(*b), list);
1148         u32 handleA, handleB;
1149
1150         handleA = __to_nfit_memdev(a)->device_handle;
1151         handleB = __to_nfit_memdev(b)->device_handle;
1152         if (handleA < handleB)
1153                 return -1;
1154         else if (handleA > handleB)
1155                 return 1;
1156         return 0;
1157 }
1158
1159 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
1160 {
1161         struct nfit_spa *nfit_spa;
1162         int rc;
1163
1164
1165         /*
1166          * For each SPA-DCR or SPA-PMEM address range find its
1167          * corresponding MEMDEV(s).  From each MEMDEV find the
1168          * corresponding DCR.  Then, if we're operating on a SPA-DCR,
1169          * try to find a SPA-BDW and a corresponding BDW that references
1170          * the DCR.  Throw it all into an nfit_mem object.  Note, that
1171          * BDWs are optional.
1172          */
1173         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
1174                 rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
1175                 if (rc)
1176                         return rc;
1177         }
1178
1179         /*
1180          * If a DIMM has failed to be mapped into SPA there will be no
1181          * SPA entries above. Find and register all the unmapped DIMMs
1182          * for reporting and recovery purposes.
1183          */
1184         rc = __nfit_mem_init(acpi_desc, NULL);
1185         if (rc)
1186                 return rc;
1187
1188         list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
1189
1190         return 0;
1191 }
1192
1193 static ssize_t bus_dsm_mask_show(struct device *dev,
1194                 struct device_attribute *attr, char *buf)
1195 {
1196         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1197         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1198
1199         return sprintf(buf, "%#lx\n", nd_desc->bus_dsm_mask);
1200 }
1201 static struct device_attribute dev_attr_bus_dsm_mask =
1202                 __ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
1203
1204 static ssize_t revision_show(struct device *dev,
1205                 struct device_attribute *attr, char *buf)
1206 {
1207         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1208         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1209         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1210
1211         return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
1212 }
1213 static DEVICE_ATTR_RO(revision);
1214
1215 static ssize_t hw_error_scrub_show(struct device *dev,
1216                 struct device_attribute *attr, char *buf)
1217 {
1218         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1219         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1220         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1221
1222         return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
1223 }
1224
1225 /*
1226  * The 'hw_error_scrub' attribute can have the following values written to it:
1227  * '0': Switch to the default mode where an exception will only insert
1228  *      the address of the memory error into the poison and badblocks lists.
1229  * '1': Enable a full scrub to happen if an exception for a memory error is
1230  *      received.
1231  */
1232 static ssize_t hw_error_scrub_store(struct device *dev,
1233                 struct device_attribute *attr, const char *buf, size_t size)
1234 {
1235         struct nvdimm_bus_descriptor *nd_desc;
1236         ssize_t rc;
1237         long val;
1238
1239         rc = kstrtol(buf, 0, &val);
1240         if (rc)
1241                 return rc;
1242
1243         device_lock(dev);
1244         nd_desc = dev_get_drvdata(dev);
1245         if (nd_desc) {
1246                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1247
1248                 switch (val) {
1249                 case HW_ERROR_SCRUB_ON:
1250                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
1251                         break;
1252                 case HW_ERROR_SCRUB_OFF:
1253                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
1254                         break;
1255                 default:
1256                         rc = -EINVAL;
1257                         break;
1258                 }
1259         }
1260         device_unlock(dev);
1261         if (rc)
1262                 return rc;
1263         return size;
1264 }
1265 static DEVICE_ATTR_RW(hw_error_scrub);
1266
1267 /*
1268  * This shows the number of full Address Range Scrubs that have been
1269  * completed since driver load time. Userspace can wait on this using
1270  * select/poll etc. A '+' at the end indicates an ARS is in progress
1271  */
1272 static ssize_t scrub_show(struct device *dev,
1273                 struct device_attribute *attr, char *buf)
1274 {
1275         struct nvdimm_bus_descriptor *nd_desc;
1276         ssize_t rc = -ENXIO;
1277
1278         device_lock(dev);
1279         nd_desc = dev_get_drvdata(dev);
1280         if (nd_desc) {
1281                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1282
1283                 mutex_lock(&acpi_desc->init_mutex);
1284                 rc = sprintf(buf, "%d%s", acpi_desc->scrub_count,
1285                                 acpi_desc->scrub_busy
1286                                 && !acpi_desc->cancel ? "+\n" : "\n");
1287                 mutex_unlock(&acpi_desc->init_mutex);
1288         }
1289         device_unlock(dev);
1290         return rc;
1291 }
1292
1293 static ssize_t scrub_store(struct device *dev,
1294                 struct device_attribute *attr, const char *buf, size_t size)
1295 {
1296         struct nvdimm_bus_descriptor *nd_desc;
1297         ssize_t rc;
1298         long val;
1299
1300         rc = kstrtol(buf, 0, &val);
1301         if (rc)
1302                 return rc;
1303         if (val != 1)
1304                 return -EINVAL;
1305
1306         device_lock(dev);
1307         nd_desc = dev_get_drvdata(dev);
1308         if (nd_desc) {
1309                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1310
1311                 rc = acpi_nfit_ars_rescan(acpi_desc, 0);
1312         }
1313         device_unlock(dev);
1314         if (rc)
1315                 return rc;
1316         return size;
1317 }
1318 static DEVICE_ATTR_RW(scrub);
1319
1320 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1321 {
1322         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1323         const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1324                 | 1 << ND_CMD_ARS_STATUS;
1325
1326         return (nd_desc->cmd_mask & mask) == mask;
1327 }
1328
1329 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1330 {
1331         struct device *dev = container_of(kobj, struct device, kobj);
1332         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1333
1334         if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
1335                 return 0;
1336         return a->mode;
1337 }
1338
1339 static struct attribute *acpi_nfit_attributes[] = {
1340         &dev_attr_revision.attr,
1341         &dev_attr_scrub.attr,
1342         &dev_attr_hw_error_scrub.attr,
1343         &dev_attr_bus_dsm_mask.attr,
1344         NULL,
1345 };
1346
1347 static const struct attribute_group acpi_nfit_attribute_group = {
1348         .name = "nfit",
1349         .attrs = acpi_nfit_attributes,
1350         .is_visible = nfit_visible,
1351 };
1352
1353 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1354         &nvdimm_bus_attribute_group,
1355         &acpi_nfit_attribute_group,
1356         NULL,
1357 };
1358
1359 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1360 {
1361         struct nvdimm *nvdimm = to_nvdimm(dev);
1362         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1363
1364         return __to_nfit_memdev(nfit_mem);
1365 }
1366
1367 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1368 {
1369         struct nvdimm *nvdimm = to_nvdimm(dev);
1370         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1371
1372         return nfit_mem->dcr;
1373 }
1374
1375 static ssize_t handle_show(struct device *dev,
1376                 struct device_attribute *attr, char *buf)
1377 {
1378         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1379
1380         return sprintf(buf, "%#x\n", memdev->device_handle);
1381 }
1382 static DEVICE_ATTR_RO(handle);
1383
1384 static ssize_t phys_id_show(struct device *dev,
1385                 struct device_attribute *attr, char *buf)
1386 {
1387         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1388
1389         return sprintf(buf, "%#x\n", memdev->physical_id);
1390 }
1391 static DEVICE_ATTR_RO(phys_id);
1392
1393 static ssize_t vendor_show(struct device *dev,
1394                 struct device_attribute *attr, char *buf)
1395 {
1396         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1397
1398         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1399 }
1400 static DEVICE_ATTR_RO(vendor);
1401
1402 static ssize_t rev_id_show(struct device *dev,
1403                 struct device_attribute *attr, char *buf)
1404 {
1405         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1406
1407         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1408 }
1409 static DEVICE_ATTR_RO(rev_id);
1410
1411 static ssize_t device_show(struct device *dev,
1412                 struct device_attribute *attr, char *buf)
1413 {
1414         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1415
1416         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1417 }
1418 static DEVICE_ATTR_RO(device);
1419
1420 static ssize_t subsystem_vendor_show(struct device *dev,
1421                 struct device_attribute *attr, char *buf)
1422 {
1423         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1424
1425         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1426 }
1427 static DEVICE_ATTR_RO(subsystem_vendor);
1428
1429 static ssize_t subsystem_rev_id_show(struct device *dev,
1430                 struct device_attribute *attr, char *buf)
1431 {
1432         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1433
1434         return sprintf(buf, "0x%04x\n",
1435                         be16_to_cpu(dcr->subsystem_revision_id));
1436 }
1437 static DEVICE_ATTR_RO(subsystem_rev_id);
1438
1439 static ssize_t subsystem_device_show(struct device *dev,
1440                 struct device_attribute *attr, char *buf)
1441 {
1442         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1443
1444         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1445 }
1446 static DEVICE_ATTR_RO(subsystem_device);
1447
1448 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1449 {
1450         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1451         int formats = 0;
1452
1453         if (nfit_mem->memdev_pmem)
1454                 formats++;
1455         if (nfit_mem->memdev_bdw)
1456                 formats++;
1457         return formats;
1458 }
1459
1460 static ssize_t format_show(struct device *dev,
1461                 struct device_attribute *attr, char *buf)
1462 {
1463         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1464
1465         return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1466 }
1467 static DEVICE_ATTR_RO(format);
1468
1469 static ssize_t format1_show(struct device *dev,
1470                 struct device_attribute *attr, char *buf)
1471 {
1472         u32 handle;
1473         ssize_t rc = -ENXIO;
1474         struct nfit_mem *nfit_mem;
1475         struct nfit_memdev *nfit_memdev;
1476         struct acpi_nfit_desc *acpi_desc;
1477         struct nvdimm *nvdimm = to_nvdimm(dev);
1478         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1479
1480         nfit_mem = nvdimm_provider_data(nvdimm);
1481         acpi_desc = nfit_mem->acpi_desc;
1482         handle = to_nfit_memdev(dev)->device_handle;
1483
1484         /* assumes DIMMs have at most 2 published interface codes */
1485         mutex_lock(&acpi_desc->init_mutex);
1486         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1487                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1488                 struct nfit_dcr *nfit_dcr;
1489
1490                 if (memdev->device_handle != handle)
1491                         continue;
1492
1493                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1494                         if (nfit_dcr->dcr->region_index != memdev->region_index)
1495                                 continue;
1496                         if (nfit_dcr->dcr->code == dcr->code)
1497                                 continue;
1498                         rc = sprintf(buf, "0x%04x\n",
1499                                         le16_to_cpu(nfit_dcr->dcr->code));
1500                         break;
1501                 }
1502                 if (rc != ENXIO)
1503                         break;
1504         }
1505         mutex_unlock(&acpi_desc->init_mutex);
1506         return rc;
1507 }
1508 static DEVICE_ATTR_RO(format1);
1509
1510 static ssize_t formats_show(struct device *dev,
1511                 struct device_attribute *attr, char *buf)
1512 {
1513         struct nvdimm *nvdimm = to_nvdimm(dev);
1514
1515         return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1516 }
1517 static DEVICE_ATTR_RO(formats);
1518
1519 static ssize_t serial_show(struct device *dev,
1520                 struct device_attribute *attr, char *buf)
1521 {
1522         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1523
1524         return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1525 }
1526 static DEVICE_ATTR_RO(serial);
1527
1528 static ssize_t family_show(struct device *dev,
1529                 struct device_attribute *attr, char *buf)
1530 {
1531         struct nvdimm *nvdimm = to_nvdimm(dev);
1532         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1533
1534         if (nfit_mem->family < 0)
1535                 return -ENXIO;
1536         return sprintf(buf, "%d\n", nfit_mem->family);
1537 }
1538 static DEVICE_ATTR_RO(family);
1539
1540 static ssize_t dsm_mask_show(struct device *dev,
1541                 struct device_attribute *attr, char *buf)
1542 {
1543         struct nvdimm *nvdimm = to_nvdimm(dev);
1544         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1545
1546         if (nfit_mem->family < 0)
1547                 return -ENXIO;
1548         return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1549 }
1550 static DEVICE_ATTR_RO(dsm_mask);
1551
1552 static ssize_t flags_show(struct device *dev,
1553                 struct device_attribute *attr, char *buf)
1554 {
1555         struct nvdimm *nvdimm = to_nvdimm(dev);
1556         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1557         u16 flags = __to_nfit_memdev(nfit_mem)->flags;
1558
1559         if (test_bit(NFIT_MEM_DIRTY, &nfit_mem->flags))
1560                 flags |= ACPI_NFIT_MEM_FLUSH_FAILED;
1561
1562         return sprintf(buf, "%s%s%s%s%s%s%s\n",
1563                 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1564                 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1565                 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1566                 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1567                 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1568                 flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1569                 flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1570 }
1571 static DEVICE_ATTR_RO(flags);
1572
1573 static ssize_t id_show(struct device *dev,
1574                 struct device_attribute *attr, char *buf)
1575 {
1576         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1577
1578         if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1579                 return sprintf(buf, "%04x-%02x-%04x-%08x\n",
1580                                 be16_to_cpu(dcr->vendor_id),
1581                                 dcr->manufacturing_location,
1582                                 be16_to_cpu(dcr->manufacturing_date),
1583                                 be32_to_cpu(dcr->serial_number));
1584         else
1585                 return sprintf(buf, "%04x-%08x\n",
1586                                 be16_to_cpu(dcr->vendor_id),
1587                                 be32_to_cpu(dcr->serial_number));
1588 }
1589 static DEVICE_ATTR_RO(id);
1590
1591 static ssize_t dirty_shutdown_show(struct device *dev,
1592                 struct device_attribute *attr, char *buf)
1593 {
1594         struct nvdimm *nvdimm = to_nvdimm(dev);
1595         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1596
1597         return sprintf(buf, "%d\n", nfit_mem->dirty_shutdown);
1598 }
1599 static DEVICE_ATTR_RO(dirty_shutdown);
1600
1601 static struct attribute *acpi_nfit_dimm_attributes[] = {
1602         &dev_attr_handle.attr,
1603         &dev_attr_phys_id.attr,
1604         &dev_attr_vendor.attr,
1605         &dev_attr_device.attr,
1606         &dev_attr_rev_id.attr,
1607         &dev_attr_subsystem_vendor.attr,
1608         &dev_attr_subsystem_device.attr,
1609         &dev_attr_subsystem_rev_id.attr,
1610         &dev_attr_format.attr,
1611         &dev_attr_formats.attr,
1612         &dev_attr_format1.attr,
1613         &dev_attr_serial.attr,
1614         &dev_attr_flags.attr,
1615         &dev_attr_id.attr,
1616         &dev_attr_family.attr,
1617         &dev_attr_dsm_mask.attr,
1618         &dev_attr_dirty_shutdown.attr,
1619         NULL,
1620 };
1621
1622 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1623                 struct attribute *a, int n)
1624 {
1625         struct device *dev = container_of(kobj, struct device, kobj);
1626         struct nvdimm *nvdimm = to_nvdimm(dev);
1627         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1628
1629         if (!to_nfit_dcr(dev)) {
1630                 /* Without a dcr only the memdev attributes can be surfaced */
1631                 if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1632                                 || a == &dev_attr_flags.attr
1633                                 || a == &dev_attr_family.attr
1634                                 || a == &dev_attr_dsm_mask.attr)
1635                         return a->mode;
1636                 return 0;
1637         }
1638
1639         if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1640                 return 0;
1641
1642         if (!test_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags)
1643                         && a == &dev_attr_dirty_shutdown.attr)
1644                 return 0;
1645
1646         return a->mode;
1647 }
1648
1649 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1650         .name = "nfit",
1651         .attrs = acpi_nfit_dimm_attributes,
1652         .is_visible = acpi_nfit_dimm_attr_visible,
1653 };
1654
1655 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1656         &nvdimm_attribute_group,
1657         &nd_device_attribute_group,
1658         &acpi_nfit_dimm_attribute_group,
1659         NULL,
1660 };
1661
1662 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1663                 u32 device_handle)
1664 {
1665         struct nfit_mem *nfit_mem;
1666
1667         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1668                 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1669                         return nfit_mem->nvdimm;
1670
1671         return NULL;
1672 }
1673
1674 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1675 {
1676         struct nfit_mem *nfit_mem;
1677         struct acpi_nfit_desc *acpi_desc;
1678
1679         dev_dbg(dev->parent, "%s: event: %d\n", dev_name(dev),
1680                         event);
1681
1682         if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1683                 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1684                                 event);
1685                 return;
1686         }
1687
1688         acpi_desc = dev_get_drvdata(dev->parent);
1689         if (!acpi_desc)
1690                 return;
1691
1692         /*
1693          * If we successfully retrieved acpi_desc, then we know nfit_mem data
1694          * is still valid.
1695          */
1696         nfit_mem = dev_get_drvdata(dev);
1697         if (nfit_mem && nfit_mem->flags_attr)
1698                 sysfs_notify_dirent(nfit_mem->flags_attr);
1699 }
1700 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1701
1702 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1703 {
1704         struct acpi_device *adev = data;
1705         struct device *dev = &adev->dev;
1706
1707         device_lock(dev->parent);
1708         __acpi_nvdimm_notify(dev, event);
1709         device_unlock(dev->parent);
1710 }
1711
1712 static bool acpi_nvdimm_has_method(struct acpi_device *adev, char *method)
1713 {
1714         acpi_handle handle;
1715         acpi_status status;
1716
1717         status = acpi_get_handle(adev->handle, method, &handle);
1718
1719         if (ACPI_SUCCESS(status))
1720                 return true;
1721         return false;
1722 }
1723
1724 __weak void nfit_intel_shutdown_status(struct nfit_mem *nfit_mem)
1725 {
1726         struct nd_intel_smart smart = { 0 };
1727         union acpi_object in_buf = {
1728                 .type = ACPI_TYPE_BUFFER,
1729                 .buffer.pointer = (char *) &smart,
1730                 .buffer.length = sizeof(smart),
1731         };
1732         union acpi_object in_obj = {
1733                 .type = ACPI_TYPE_PACKAGE,
1734                 .package.count = 1,
1735                 .package.elements = &in_buf,
1736         };
1737         const u8 func = ND_INTEL_SMART;
1738         const guid_t *guid = to_nfit_uuid(nfit_mem->family);
1739         u8 revid = nfit_dsm_revid(nfit_mem->family, func);
1740         struct acpi_device *adev = nfit_mem->adev;
1741         acpi_handle handle = adev->handle;
1742         union acpi_object *out_obj;
1743
1744         if ((nfit_mem->dsm_mask & (1 << func)) == 0)
1745                 return;
1746
1747         out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
1748         if (!out_obj)
1749                 return;
1750
1751         if (smart.flags & ND_INTEL_SMART_SHUTDOWN_VALID) {
1752                 if (smart.shutdown_state)
1753                         set_bit(NFIT_MEM_DIRTY, &nfit_mem->flags);
1754         }
1755
1756         if (smart.flags & ND_INTEL_SMART_SHUTDOWN_COUNT_VALID) {
1757                 set_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags);
1758                 nfit_mem->dirty_shutdown = smart.shutdown_count;
1759         }
1760         ACPI_FREE(out_obj);
1761 }
1762
1763 static void populate_shutdown_status(struct nfit_mem *nfit_mem)
1764 {
1765         /*
1766          * For DIMMs that provide a dynamic facility to retrieve a
1767          * dirty-shutdown status and/or a dirty-shutdown count, cache
1768          * these values in nfit_mem.
1769          */
1770         if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
1771                 nfit_intel_shutdown_status(nfit_mem);
1772 }
1773
1774 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1775                 struct nfit_mem *nfit_mem, u32 device_handle)
1776 {
1777         struct acpi_device *adev, *adev_dimm;
1778         struct device *dev = acpi_desc->dev;
1779         unsigned long dsm_mask, label_mask;
1780         const guid_t *guid;
1781         int i;
1782         int family = -1;
1783
1784         /* nfit test assumes 1:1 relationship between commands and dsms */
1785         nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1786         nfit_mem->family = NVDIMM_FAMILY_INTEL;
1787         adev = to_acpi_dev(acpi_desc);
1788         if (!adev) {
1789                 /* unit test case */
1790                 populate_shutdown_status(nfit_mem);
1791                 return 0;
1792         }
1793
1794         adev_dimm = acpi_find_child_device(adev, device_handle, false);
1795         nfit_mem->adev = adev_dimm;
1796         if (!adev_dimm) {
1797                 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1798                                 device_handle);
1799                 return force_enable_dimms ? 0 : -ENODEV;
1800         }
1801
1802         if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1803                 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1804                 dev_err(dev, "%s: notification registration failed\n",
1805                                 dev_name(&adev_dimm->dev));
1806                 return -ENXIO;
1807         }
1808         /*
1809          * Record nfit_mem for the notification path to track back to
1810          * the nfit sysfs attributes for this dimm device object.
1811          */
1812         dev_set_drvdata(&adev_dimm->dev, nfit_mem);
1813
1814         /*
1815          * Until standardization materializes we need to consider 4
1816          * different command sets.  Note, that checking for function0 (bit0)
1817          * tells us if any commands are reachable through this GUID.
1818          */
1819         for (i = 0; i <= NVDIMM_FAMILY_MAX; i++)
1820                 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1821                         if (family < 0 || i == default_dsm_family)
1822                                 family = i;
1823
1824         /* limit the supported commands to those that are publicly documented */
1825         nfit_mem->family = family;
1826         if (override_dsm_mask && !disable_vendor_specific)
1827                 dsm_mask = override_dsm_mask;
1828         else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1829                 dsm_mask = NVDIMM_INTEL_CMDMASK;
1830                 if (disable_vendor_specific)
1831                         dsm_mask &= ~(1 << ND_CMD_VENDOR);
1832         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1833                 dsm_mask = 0x1c3c76;
1834         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1835                 dsm_mask = 0x1fe;
1836                 if (disable_vendor_specific)
1837                         dsm_mask &= ~(1 << 8);
1838         } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1839                 dsm_mask = 0xffffffff;
1840         } else {
1841                 dev_dbg(dev, "unknown dimm command family\n");
1842                 nfit_mem->family = -1;
1843                 /* DSMs are optional, continue loading the driver... */
1844                 return 0;
1845         }
1846
1847         guid = to_nfit_uuid(nfit_mem->family);
1848         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1849                 if (acpi_check_dsm(adev_dimm->handle, guid,
1850                                         nfit_dsm_revid(nfit_mem->family, i),
1851                                         1ULL << i))
1852                         set_bit(i, &nfit_mem->dsm_mask);
1853
1854         /*
1855          * Prefer the NVDIMM_FAMILY_INTEL label read commands if present
1856          * due to their better semantics handling locked capacity.
1857          */
1858         label_mask = 1 << ND_CMD_GET_CONFIG_SIZE | 1 << ND_CMD_GET_CONFIG_DATA
1859                 | 1 << ND_CMD_SET_CONFIG_DATA;
1860         if (family == NVDIMM_FAMILY_INTEL
1861                         && (dsm_mask & label_mask) == label_mask)
1862                 return 0;
1863
1864         if (acpi_nvdimm_has_method(adev_dimm, "_LSI")
1865                         && acpi_nvdimm_has_method(adev_dimm, "_LSR")) {
1866                 dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev));
1867                 set_bit(NFIT_MEM_LSR, &nfit_mem->flags);
1868         }
1869
1870         if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
1871                         && acpi_nvdimm_has_method(adev_dimm, "_LSW")) {
1872                 dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev));
1873                 set_bit(NFIT_MEM_LSW, &nfit_mem->flags);
1874         }
1875
1876         populate_shutdown_status(nfit_mem);
1877
1878         return 0;
1879 }
1880
1881 static void shutdown_dimm_notify(void *data)
1882 {
1883         struct acpi_nfit_desc *acpi_desc = data;
1884         struct nfit_mem *nfit_mem;
1885
1886         mutex_lock(&acpi_desc->init_mutex);
1887         /*
1888          * Clear out the nfit_mem->flags_attr and shut down dimm event
1889          * notifications.
1890          */
1891         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1892                 struct acpi_device *adev_dimm = nfit_mem->adev;
1893
1894                 if (nfit_mem->flags_attr) {
1895                         sysfs_put(nfit_mem->flags_attr);
1896                         nfit_mem->flags_attr = NULL;
1897                 }
1898                 if (adev_dimm) {
1899                         acpi_remove_notify_handler(adev_dimm->handle,
1900                                         ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
1901                         dev_set_drvdata(&adev_dimm->dev, NULL);
1902                 }
1903         }
1904         mutex_unlock(&acpi_desc->init_mutex);
1905 }
1906
1907 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1908 {
1909         struct nfit_mem *nfit_mem;
1910         int dimm_count = 0, rc;
1911         struct nvdimm *nvdimm;
1912
1913         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1914                 struct acpi_nfit_flush_address *flush;
1915                 unsigned long flags = 0, cmd_mask;
1916                 struct nfit_memdev *nfit_memdev;
1917                 u32 device_handle;
1918                 u16 mem_flags;
1919
1920                 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1921                 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1922                 if (nvdimm) {
1923                         dimm_count++;
1924                         continue;
1925                 }
1926
1927                 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1928                         set_bit(NDD_ALIASING, &flags);
1929
1930                 /* collate flags across all memdevs for this dimm */
1931                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1932                         struct acpi_nfit_memory_map *dimm_memdev;
1933
1934                         dimm_memdev = __to_nfit_memdev(nfit_mem);
1935                         if (dimm_memdev->device_handle
1936                                         != nfit_memdev->memdev->device_handle)
1937                                 continue;
1938                         dimm_memdev->flags |= nfit_memdev->memdev->flags;
1939                 }
1940
1941                 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1942                 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1943                         set_bit(NDD_UNARMED, &flags);
1944
1945                 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1946                 if (rc)
1947                         continue;
1948
1949                 /*
1950                  * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
1951                  * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
1952                  * userspace interface.
1953                  */
1954                 cmd_mask = 1UL << ND_CMD_CALL;
1955                 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1956                         /*
1957                          * These commands have a 1:1 correspondence
1958                          * between DSM payload and libnvdimm ioctl
1959                          * payload format.
1960                          */
1961                         cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
1962                 }
1963
1964                 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
1965                         set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
1966                         set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
1967                 }
1968                 if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags))
1969                         set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
1970
1971                 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
1972                         : NULL;
1973                 nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
1974                                 acpi_nfit_dimm_attribute_groups,
1975                                 flags, cmd_mask, flush ? flush->hint_count : 0,
1976                                 nfit_mem->flush_wpq);
1977                 if (!nvdimm)
1978                         return -ENOMEM;
1979
1980                 nfit_mem->nvdimm = nvdimm;
1981                 dimm_count++;
1982
1983                 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
1984                         continue;
1985
1986                 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s%s\n",
1987                                 nvdimm_name(nvdimm),
1988                   mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
1989                   mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
1990                   mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
1991                   mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
1992                   mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
1993
1994         }
1995
1996         rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1997         if (rc)
1998                 return rc;
1999
2000         /*
2001          * Now that dimms are successfully registered, and async registration
2002          * is flushed, attempt to enable event notification.
2003          */
2004         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2005                 struct kernfs_node *nfit_kernfs;
2006
2007                 nvdimm = nfit_mem->nvdimm;
2008                 if (!nvdimm)
2009                         continue;
2010
2011                 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
2012                 if (nfit_kernfs)
2013                         nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
2014                                         "flags");
2015                 sysfs_put(nfit_kernfs);
2016                 if (!nfit_mem->flags_attr)
2017                         dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
2018                                         nvdimm_name(nvdimm));
2019         }
2020
2021         return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
2022                         acpi_desc);
2023 }
2024
2025 /*
2026  * These constants are private because there are no kernel consumers of
2027  * these commands.
2028  */
2029 enum nfit_aux_cmds {
2030         NFIT_CMD_TRANSLATE_SPA = 5,
2031         NFIT_CMD_ARS_INJECT_SET = 7,
2032         NFIT_CMD_ARS_INJECT_CLEAR = 8,
2033         NFIT_CMD_ARS_INJECT_GET = 9,
2034 };
2035
2036 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
2037 {
2038         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2039         const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
2040         struct acpi_device *adev;
2041         unsigned long dsm_mask;
2042         int i;
2043
2044         nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
2045         nd_desc->bus_dsm_mask = acpi_desc->bus_nfit_cmd_force_en;
2046         adev = to_acpi_dev(acpi_desc);
2047         if (!adev)
2048                 return;
2049
2050         for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
2051                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2052                         set_bit(i, &nd_desc->cmd_mask);
2053         set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
2054
2055         dsm_mask =
2056                 (1 << ND_CMD_ARS_CAP) |
2057                 (1 << ND_CMD_ARS_START) |
2058                 (1 << ND_CMD_ARS_STATUS) |
2059                 (1 << ND_CMD_CLEAR_ERROR) |
2060                 (1 << NFIT_CMD_TRANSLATE_SPA) |
2061                 (1 << NFIT_CMD_ARS_INJECT_SET) |
2062                 (1 << NFIT_CMD_ARS_INJECT_CLEAR) |
2063                 (1 << NFIT_CMD_ARS_INJECT_GET);
2064         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
2065                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2066                         set_bit(i, &nd_desc->bus_dsm_mask);
2067 }
2068
2069 static ssize_t range_index_show(struct device *dev,
2070                 struct device_attribute *attr, char *buf)
2071 {
2072         struct nd_region *nd_region = to_nd_region(dev);
2073         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
2074
2075         return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
2076 }
2077 static DEVICE_ATTR_RO(range_index);
2078
2079 static struct attribute *acpi_nfit_region_attributes[] = {
2080         &dev_attr_range_index.attr,
2081         NULL,
2082 };
2083
2084 static const struct attribute_group acpi_nfit_region_attribute_group = {
2085         .name = "nfit",
2086         .attrs = acpi_nfit_region_attributes,
2087 };
2088
2089 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
2090         &nd_region_attribute_group,
2091         &nd_mapping_attribute_group,
2092         &nd_device_attribute_group,
2093         &nd_numa_attribute_group,
2094         &acpi_nfit_region_attribute_group,
2095         NULL,
2096 };
2097
2098 /* enough info to uniquely specify an interleave set */
2099 struct nfit_set_info {
2100         struct nfit_set_info_map {
2101                 u64 region_offset;
2102                 u32 serial_number;
2103                 u32 pad;
2104         } mapping[0];
2105 };
2106
2107 struct nfit_set_info2 {
2108         struct nfit_set_info_map2 {
2109                 u64 region_offset;
2110                 u32 serial_number;
2111                 u16 vendor_id;
2112                 u16 manufacturing_date;
2113                 u8  manufacturing_location;
2114                 u8  reserved[31];
2115         } mapping[0];
2116 };
2117
2118 static size_t sizeof_nfit_set_info(int num_mappings)
2119 {
2120         return sizeof(struct nfit_set_info)
2121                 + num_mappings * sizeof(struct nfit_set_info_map);
2122 }
2123
2124 static size_t sizeof_nfit_set_info2(int num_mappings)
2125 {
2126         return sizeof(struct nfit_set_info2)
2127                 + num_mappings * sizeof(struct nfit_set_info_map2);
2128 }
2129
2130 static int cmp_map_compat(const void *m0, const void *m1)
2131 {
2132         const struct nfit_set_info_map *map0 = m0;
2133         const struct nfit_set_info_map *map1 = m1;
2134
2135         return memcmp(&map0->region_offset, &map1->region_offset,
2136                         sizeof(u64));
2137 }
2138
2139 static int cmp_map(const void *m0, const void *m1)
2140 {
2141         const struct nfit_set_info_map *map0 = m0;
2142         const struct nfit_set_info_map *map1 = m1;
2143
2144         if (map0->region_offset < map1->region_offset)
2145                 return -1;
2146         else if (map0->region_offset > map1->region_offset)
2147                 return 1;
2148         return 0;
2149 }
2150
2151 static int cmp_map2(const void *m0, const void *m1)
2152 {
2153         const struct nfit_set_info_map2 *map0 = m0;
2154         const struct nfit_set_info_map2 *map1 = m1;
2155
2156         if (map0->region_offset < map1->region_offset)
2157                 return -1;
2158         else if (map0->region_offset > map1->region_offset)
2159                 return 1;
2160         return 0;
2161 }
2162
2163 /* Retrieve the nth entry referencing this spa */
2164 static struct acpi_nfit_memory_map *memdev_from_spa(
2165                 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
2166 {
2167         struct nfit_memdev *nfit_memdev;
2168
2169         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
2170                 if (nfit_memdev->memdev->range_index == range_index)
2171                         if (n-- == 0)
2172                                 return nfit_memdev->memdev;
2173         return NULL;
2174 }
2175
2176 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
2177                 struct nd_region_desc *ndr_desc,
2178                 struct acpi_nfit_system_address *spa)
2179 {
2180         struct device *dev = acpi_desc->dev;
2181         struct nd_interleave_set *nd_set;
2182         u16 nr = ndr_desc->num_mappings;
2183         struct nfit_set_info2 *info2;
2184         struct nfit_set_info *info;
2185         int i;
2186
2187         nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
2188         if (!nd_set)
2189                 return -ENOMEM;
2190         ndr_desc->nd_set = nd_set;
2191         guid_copy(&nd_set->type_guid, (guid_t *) spa->range_guid);
2192
2193         info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
2194         if (!info)
2195                 return -ENOMEM;
2196
2197         info2 = devm_kzalloc(dev, sizeof_nfit_set_info2(nr), GFP_KERNEL);
2198         if (!info2)
2199                 return -ENOMEM;
2200
2201         for (i = 0; i < nr; i++) {
2202                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
2203                 struct nfit_set_info_map *map = &info->mapping[i];
2204                 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2205                 struct nvdimm *nvdimm = mapping->nvdimm;
2206                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2207                 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
2208                                 spa->range_index, i);
2209                 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2210
2211                 if (!memdev || !nfit_mem->dcr) {
2212                         dev_err(dev, "%s: failed to find DCR\n", __func__);
2213                         return -ENODEV;
2214                 }
2215
2216                 map->region_offset = memdev->region_offset;
2217                 map->serial_number = dcr->serial_number;
2218
2219                 map2->region_offset = memdev->region_offset;
2220                 map2->serial_number = dcr->serial_number;
2221                 map2->vendor_id = dcr->vendor_id;
2222                 map2->manufacturing_date = dcr->manufacturing_date;
2223                 map2->manufacturing_location = dcr->manufacturing_location;
2224         }
2225
2226         /* v1.1 namespaces */
2227         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2228                         cmp_map, NULL);
2229         nd_set->cookie1 = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2230
2231         /* v1.2 namespaces */
2232         sort(&info2->mapping[0], nr, sizeof(struct nfit_set_info_map2),
2233                         cmp_map2, NULL);
2234         nd_set->cookie2 = nd_fletcher64(info2, sizeof_nfit_set_info2(nr), 0);
2235
2236         /* support v1.1 namespaces created with the wrong sort order */
2237         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2238                         cmp_map_compat, NULL);
2239         nd_set->altcookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2240
2241         /* record the result of the sort for the mapping position */
2242         for (i = 0; i < nr; i++) {
2243                 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2244                 int j;
2245
2246                 for (j = 0; j < nr; j++) {
2247                         struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
2248                         struct nvdimm *nvdimm = mapping->nvdimm;
2249                         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2250                         struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2251
2252                         if (map2->serial_number == dcr->serial_number &&
2253                             map2->vendor_id == dcr->vendor_id &&
2254                             map2->manufacturing_date == dcr->manufacturing_date &&
2255                             map2->manufacturing_location
2256                                     == dcr->manufacturing_location) {
2257                                 mapping->position = i;
2258                                 break;
2259                         }
2260                 }
2261         }
2262
2263         ndr_desc->nd_set = nd_set;
2264         devm_kfree(dev, info);
2265         devm_kfree(dev, info2);
2266
2267         return 0;
2268 }
2269
2270 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
2271 {
2272         struct acpi_nfit_interleave *idt = mmio->idt;
2273         u32 sub_line_offset, line_index, line_offset;
2274         u64 line_no, table_skip_count, table_offset;
2275
2276         line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
2277         table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
2278         line_offset = idt->line_offset[line_index]
2279                 * mmio->line_size;
2280         table_offset = table_skip_count * mmio->table_size;
2281
2282         return mmio->base_offset + line_offset + table_offset + sub_line_offset;
2283 }
2284
2285 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
2286 {
2287         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2288         u64 offset = nfit_blk->stat_offset + mmio->size * bw;
2289         const u32 STATUS_MASK = 0x80000037;
2290
2291         if (mmio->num_lines)
2292                 offset = to_interleave_offset(offset, mmio);
2293
2294         return readl(mmio->addr.base + offset) & STATUS_MASK;
2295 }
2296
2297 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
2298                 resource_size_t dpa, unsigned int len, unsigned int write)
2299 {
2300         u64 cmd, offset;
2301         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2302
2303         enum {
2304                 BCW_OFFSET_MASK = (1ULL << 48)-1,
2305                 BCW_LEN_SHIFT = 48,
2306                 BCW_LEN_MASK = (1ULL << 8) - 1,
2307                 BCW_CMD_SHIFT = 56,
2308         };
2309
2310         cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
2311         len = len >> L1_CACHE_SHIFT;
2312         cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
2313         cmd |= ((u64) write) << BCW_CMD_SHIFT;
2314
2315         offset = nfit_blk->cmd_offset + mmio->size * bw;
2316         if (mmio->num_lines)
2317                 offset = to_interleave_offset(offset, mmio);
2318
2319         writeq(cmd, mmio->addr.base + offset);
2320         nvdimm_flush(nfit_blk->nd_region);
2321
2322         if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
2323                 readq(mmio->addr.base + offset);
2324 }
2325
2326 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
2327                 resource_size_t dpa, void *iobuf, size_t len, int rw,
2328                 unsigned int lane)
2329 {
2330         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2331         unsigned int copied = 0;
2332         u64 base_offset;
2333         int rc;
2334
2335         base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
2336                 + lane * mmio->size;
2337         write_blk_ctl(nfit_blk, lane, dpa, len, rw);
2338         while (len) {
2339                 unsigned int c;
2340                 u64 offset;
2341
2342                 if (mmio->num_lines) {
2343                         u32 line_offset;
2344
2345                         offset = to_interleave_offset(base_offset + copied,
2346                                         mmio);
2347                         div_u64_rem(offset, mmio->line_size, &line_offset);
2348                         c = min_t(size_t, len, mmio->line_size - line_offset);
2349                 } else {
2350                         offset = base_offset + nfit_blk->bdw_offset;
2351                         c = len;
2352                 }
2353
2354                 if (rw)
2355                         memcpy_flushcache(mmio->addr.aperture + offset, iobuf + copied, c);
2356                 else {
2357                         if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
2358                                 arch_invalidate_pmem((void __force *)
2359                                         mmio->addr.aperture + offset, c);
2360
2361                         memcpy(iobuf + copied, mmio->addr.aperture + offset, c);
2362                 }
2363
2364                 copied += c;
2365                 len -= c;
2366         }
2367
2368         if (rw)
2369                 nvdimm_flush(nfit_blk->nd_region);
2370
2371         rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
2372         return rc;
2373 }
2374
2375 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
2376                 resource_size_t dpa, void *iobuf, u64 len, int rw)
2377 {
2378         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
2379         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2380         struct nd_region *nd_region = nfit_blk->nd_region;
2381         unsigned int lane, copied = 0;
2382         int rc = 0;
2383
2384         lane = nd_region_acquire_lane(nd_region);
2385         while (len) {
2386                 u64 c = min(len, mmio->size);
2387
2388                 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
2389                                 iobuf + copied, c, rw, lane);
2390                 if (rc)
2391                         break;
2392
2393                 copied += c;
2394                 len -= c;
2395         }
2396         nd_region_release_lane(nd_region, lane);
2397
2398         return rc;
2399 }
2400
2401 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
2402                 struct acpi_nfit_interleave *idt, u16 interleave_ways)
2403 {
2404         if (idt) {
2405                 mmio->num_lines = idt->line_count;
2406                 mmio->line_size = idt->line_size;
2407                 if (interleave_ways == 0)
2408                         return -ENXIO;
2409                 mmio->table_size = mmio->num_lines * interleave_ways
2410                         * mmio->line_size;
2411         }
2412
2413         return 0;
2414 }
2415
2416 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
2417                 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
2418 {
2419         struct nd_cmd_dimm_flags flags;
2420         int rc;
2421
2422         memset(&flags, 0, sizeof(flags));
2423         rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
2424                         sizeof(flags), NULL);
2425
2426         if (rc >= 0 && flags.status == 0)
2427                 nfit_blk->dimm_flags = flags.flags;
2428         else if (rc == -ENOTTY) {
2429                 /* fall back to a conservative default */
2430                 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
2431                 rc = 0;
2432         } else
2433                 rc = -ENXIO;
2434
2435         return rc;
2436 }
2437
2438 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
2439                 struct device *dev)
2440 {
2441         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
2442         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
2443         struct nfit_blk_mmio *mmio;
2444         struct nfit_blk *nfit_blk;
2445         struct nfit_mem *nfit_mem;
2446         struct nvdimm *nvdimm;
2447         int rc;
2448
2449         nvdimm = nd_blk_region_to_dimm(ndbr);
2450         nfit_mem = nvdimm_provider_data(nvdimm);
2451         if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
2452                 dev_dbg(dev, "missing%s%s%s\n",
2453                                 nfit_mem ? "" : " nfit_mem",
2454                                 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
2455                                 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
2456                 return -ENXIO;
2457         }
2458
2459         nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
2460         if (!nfit_blk)
2461                 return -ENOMEM;
2462         nd_blk_region_set_provider_data(ndbr, nfit_blk);
2463         nfit_blk->nd_region = to_nd_region(dev);
2464
2465         /* map block aperture memory */
2466         nfit_blk->bdw_offset = nfit_mem->bdw->offset;
2467         mmio = &nfit_blk->mmio[BDW];
2468         mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
2469                         nfit_mem->spa_bdw->length, nd_blk_memremap_flags(ndbr));
2470         if (!mmio->addr.base) {
2471                 dev_dbg(dev, "%s failed to map bdw\n",
2472                                 nvdimm_name(nvdimm));
2473                 return -ENOMEM;
2474         }
2475         mmio->size = nfit_mem->bdw->size;
2476         mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
2477         mmio->idt = nfit_mem->idt_bdw;
2478         mmio->spa = nfit_mem->spa_bdw;
2479         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
2480                         nfit_mem->memdev_bdw->interleave_ways);
2481         if (rc) {
2482                 dev_dbg(dev, "%s failed to init bdw interleave\n",
2483                                 nvdimm_name(nvdimm));
2484                 return rc;
2485         }
2486
2487         /* map block control memory */
2488         nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
2489         nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
2490         mmio = &nfit_blk->mmio[DCR];
2491         mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
2492                         nfit_mem->spa_dcr->length);
2493         if (!mmio->addr.base) {
2494                 dev_dbg(dev, "%s failed to map dcr\n",
2495                                 nvdimm_name(nvdimm));
2496                 return -ENOMEM;
2497         }
2498         mmio->size = nfit_mem->dcr->window_size;
2499         mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
2500         mmio->idt = nfit_mem->idt_dcr;
2501         mmio->spa = nfit_mem->spa_dcr;
2502         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
2503                         nfit_mem->memdev_dcr->interleave_ways);
2504         if (rc) {
2505                 dev_dbg(dev, "%s failed to init dcr interleave\n",
2506                                 nvdimm_name(nvdimm));
2507                 return rc;
2508         }
2509
2510         rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
2511         if (rc < 0) {
2512                 dev_dbg(dev, "%s failed get DIMM flags\n",
2513                                 nvdimm_name(nvdimm));
2514                 return rc;
2515         }
2516
2517         if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
2518                 dev_warn(dev, "unable to guarantee persistence of writes\n");
2519
2520         if (mmio->line_size == 0)
2521                 return 0;
2522
2523         if ((u32) nfit_blk->cmd_offset % mmio->line_size
2524                         + 8 > mmio->line_size) {
2525                 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
2526                 return -ENXIO;
2527         } else if ((u32) nfit_blk->stat_offset % mmio->line_size
2528                         + 8 > mmio->line_size) {
2529                 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
2530                 return -ENXIO;
2531         }
2532
2533         return 0;
2534 }
2535
2536 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2537                 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2538 {
2539         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2540         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2541         int cmd_rc, rc;
2542
2543         cmd->address = spa->address;
2544         cmd->length = spa->length;
2545         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2546                         sizeof(*cmd), &cmd_rc);
2547         if (rc < 0)
2548                 return rc;
2549         return cmd_rc;
2550 }
2551
2552 static int ars_start(struct acpi_nfit_desc *acpi_desc,
2553                 struct nfit_spa *nfit_spa, enum nfit_ars_state req_type)
2554 {
2555         int rc;
2556         int cmd_rc;
2557         struct nd_cmd_ars_start ars_start;
2558         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2559         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2560
2561         memset(&ars_start, 0, sizeof(ars_start));
2562         ars_start.address = spa->address;
2563         ars_start.length = spa->length;
2564         if (req_type == ARS_REQ_SHORT)
2565                 ars_start.flags = ND_ARS_RETURN_PREV_DATA;
2566         if (nfit_spa_type(spa) == NFIT_SPA_PM)
2567                 ars_start.type = ND_ARS_PERSISTENT;
2568         else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2569                 ars_start.type = ND_ARS_VOLATILE;
2570         else
2571                 return -ENOTTY;
2572
2573         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2574                         sizeof(ars_start), &cmd_rc);
2575
2576         if (rc < 0)
2577                 return rc;
2578         return cmd_rc;
2579 }
2580
2581 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2582 {
2583         int rc, cmd_rc;
2584         struct nd_cmd_ars_start ars_start;
2585         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2586         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2587
2588         memset(&ars_start, 0, sizeof(ars_start));
2589         ars_start.address = ars_status->restart_address;
2590         ars_start.length = ars_status->restart_length;
2591         ars_start.type = ars_status->type;
2592         ars_start.flags = acpi_desc->ars_start_flags;
2593         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2594                         sizeof(ars_start), &cmd_rc);
2595         if (rc < 0)
2596                 return rc;
2597         return cmd_rc;
2598 }
2599
2600 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2601 {
2602         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2603         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2604         int rc, cmd_rc;
2605
2606         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2607                         acpi_desc->max_ars, &cmd_rc);
2608         if (rc < 0)
2609                 return rc;
2610         return cmd_rc;
2611 }
2612
2613 static void ars_complete(struct acpi_nfit_desc *acpi_desc,
2614                 struct nfit_spa *nfit_spa)
2615 {
2616         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2617         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2618         struct nd_region *nd_region = nfit_spa->nd_region;
2619         struct device *dev;
2620
2621         lockdep_assert_held(&acpi_desc->init_mutex);
2622         /*
2623          * Only advance the ARS state for ARS runs initiated by the
2624          * kernel, ignore ARS results from BIOS initiated runs for scrub
2625          * completion tracking.
2626          */
2627         if (acpi_desc->scrub_spa != nfit_spa)
2628                 return;
2629
2630         if ((ars_status->address >= spa->address && ars_status->address
2631                                 < spa->address + spa->length)
2632                         || (ars_status->address < spa->address)) {
2633                 /*
2634                  * Assume that if a scrub starts at an offset from the
2635                  * start of nfit_spa that we are in the continuation
2636                  * case.
2637                  *
2638                  * Otherwise, if the scrub covers the spa range, mark
2639                  * any pending request complete.
2640                  */
2641                 if (ars_status->address + ars_status->length
2642                                 >= spa->address + spa->length)
2643                                 /* complete */;
2644                 else
2645                         return;
2646         } else
2647                 return;
2648
2649         acpi_desc->scrub_spa = NULL;
2650         if (nd_region) {
2651                 dev = nd_region_dev(nd_region);
2652                 nvdimm_region_notify(nd_region, NVDIMM_REVALIDATE_POISON);
2653         } else
2654                 dev = acpi_desc->dev;
2655         dev_dbg(dev, "ARS: range %d complete\n", spa->range_index);
2656 }
2657
2658 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc)
2659 {
2660         struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2661         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2662         int rc;
2663         u32 i;
2664
2665         /*
2666          * First record starts at 44 byte offset from the start of the
2667          * payload.
2668          */
2669         if (ars_status->out_length < 44)
2670                 return 0;
2671         for (i = 0; i < ars_status->num_records; i++) {
2672                 /* only process full records */
2673                 if (ars_status->out_length
2674                                 < 44 + sizeof(struct nd_ars_record) * (i + 1))
2675                         break;
2676                 rc = nvdimm_bus_add_badrange(nvdimm_bus,
2677                                 ars_status->records[i].err_address,
2678                                 ars_status->records[i].length);
2679                 if (rc)
2680                         return rc;
2681         }
2682         if (i < ars_status->num_records)
2683                 dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2684
2685         return 0;
2686 }
2687
2688 static void acpi_nfit_remove_resource(void *data)
2689 {
2690         struct resource *res = data;
2691
2692         remove_resource(res);
2693 }
2694
2695 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2696                 struct nd_region_desc *ndr_desc)
2697 {
2698         struct resource *res, *nd_res = ndr_desc->res;
2699         int is_pmem, ret;
2700
2701         /* No operation if the region is already registered as PMEM */
2702         is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2703                                 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2704         if (is_pmem == REGION_INTERSECTS)
2705                 return 0;
2706
2707         res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2708         if (!res)
2709                 return -ENOMEM;
2710
2711         res->name = "Persistent Memory";
2712         res->start = nd_res->start;
2713         res->end = nd_res->end;
2714         res->flags = IORESOURCE_MEM;
2715         res->desc = IORES_DESC_PERSISTENT_MEMORY;
2716
2717         ret = insert_resource(&iomem_resource, res);
2718         if (ret)
2719                 return ret;
2720
2721         ret = devm_add_action_or_reset(acpi_desc->dev,
2722                                         acpi_nfit_remove_resource,
2723                                         res);
2724         if (ret)
2725                 return ret;
2726
2727         return 0;
2728 }
2729
2730 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2731                 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2732                 struct acpi_nfit_memory_map *memdev,
2733                 struct nfit_spa *nfit_spa)
2734 {
2735         struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2736                         memdev->device_handle);
2737         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2738         struct nd_blk_region_desc *ndbr_desc;
2739         struct nfit_mem *nfit_mem;
2740         int rc;
2741
2742         if (!nvdimm) {
2743                 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2744                                 spa->range_index, memdev->device_handle);
2745                 return -ENODEV;
2746         }
2747
2748         mapping->nvdimm = nvdimm;
2749         switch (nfit_spa_type(spa)) {
2750         case NFIT_SPA_PM:
2751         case NFIT_SPA_VOLATILE:
2752                 mapping->start = memdev->address;
2753                 mapping->size = memdev->region_size;
2754                 break;
2755         case NFIT_SPA_DCR:
2756                 nfit_mem = nvdimm_provider_data(nvdimm);
2757                 if (!nfit_mem || !nfit_mem->bdw) {
2758                         dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2759                                         spa->range_index, nvdimm_name(nvdimm));
2760                         break;
2761                 }
2762
2763                 mapping->size = nfit_mem->bdw->capacity;
2764                 mapping->start = nfit_mem->bdw->start_address;
2765                 ndr_desc->num_lanes = nfit_mem->bdw->windows;
2766                 ndr_desc->mapping = mapping;
2767                 ndr_desc->num_mappings = 1;
2768                 ndbr_desc = to_blk_region_desc(ndr_desc);
2769                 ndbr_desc->enable = acpi_nfit_blk_region_enable;
2770                 ndbr_desc->do_io = acpi_desc->blk_do_io;
2771                 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2772                 if (rc)
2773                         return rc;
2774                 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2775                                 ndr_desc);
2776                 if (!nfit_spa->nd_region)
2777                         return -ENOMEM;
2778                 break;
2779         }
2780
2781         return 0;
2782 }
2783
2784 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2785 {
2786         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2787                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2788                 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2789                 nfit_spa_type(spa) == NFIT_SPA_PCD);
2790 }
2791
2792 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
2793 {
2794         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2795                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2796                 nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
2797 }
2798
2799 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2800                 struct nfit_spa *nfit_spa)
2801 {
2802         static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2803         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2804         struct nd_blk_region_desc ndbr_desc;
2805         struct nd_region_desc *ndr_desc;
2806         struct nfit_memdev *nfit_memdev;
2807         struct nvdimm_bus *nvdimm_bus;
2808         struct resource res;
2809         int count = 0, rc;
2810
2811         if (nfit_spa->nd_region)
2812                 return 0;
2813
2814         if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2815                 dev_dbg(acpi_desc->dev, "detected invalid spa index\n");
2816                 return 0;
2817         }
2818
2819         memset(&res, 0, sizeof(res));
2820         memset(&mappings, 0, sizeof(mappings));
2821         memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2822         res.start = spa->address;
2823         res.end = res.start + spa->length - 1;
2824         ndr_desc = &ndbr_desc.ndr_desc;
2825         ndr_desc->res = &res;
2826         ndr_desc->provider_data = nfit_spa;
2827         ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2828         if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
2829                 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2830                                                 spa->proximity_domain);
2831         else
2832                 ndr_desc->numa_node = NUMA_NO_NODE;
2833
2834         /*
2835          * Persistence domain bits are hierarchical, if
2836          * ACPI_NFIT_CAPABILITY_CACHE_FLUSH is set then
2837          * ACPI_NFIT_CAPABILITY_MEM_FLUSH is implied.
2838          */
2839         if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_CACHE_FLUSH)
2840                 set_bit(ND_REGION_PERSIST_CACHE, &ndr_desc->flags);
2841         else if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_MEM_FLUSH)
2842                 set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc->flags);
2843
2844         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2845                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2846                 struct nd_mapping_desc *mapping;
2847
2848                 if (memdev->range_index != spa->range_index)
2849                         continue;
2850                 if (count >= ND_MAX_MAPPINGS) {
2851                         dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2852                                         spa->range_index, ND_MAX_MAPPINGS);
2853                         return -ENXIO;
2854                 }
2855                 mapping = &mappings[count++];
2856                 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
2857                                 memdev, nfit_spa);
2858                 if (rc)
2859                         goto out;
2860         }
2861
2862         ndr_desc->mapping = mappings;
2863         ndr_desc->num_mappings = count;
2864         rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2865         if (rc)
2866                 goto out;
2867
2868         nvdimm_bus = acpi_desc->nvdimm_bus;
2869         if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2870                 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2871                 if (rc) {
2872                         dev_warn(acpi_desc->dev,
2873                                 "failed to insert pmem resource to iomem: %d\n",
2874                                 rc);
2875                         goto out;
2876                 }
2877
2878                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2879                                 ndr_desc);
2880                 if (!nfit_spa->nd_region)
2881                         rc = -ENOMEM;
2882         } else if (nfit_spa_is_volatile(spa)) {
2883                 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
2884                                 ndr_desc);
2885                 if (!nfit_spa->nd_region)
2886                         rc = -ENOMEM;
2887         } else if (nfit_spa_is_virtual(spa)) {
2888                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2889                                 ndr_desc);
2890                 if (!nfit_spa->nd_region)
2891                         rc = -ENOMEM;
2892         }
2893
2894  out:
2895         if (rc)
2896                 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
2897                                 nfit_spa->spa->range_index);
2898         return rc;
2899 }
2900
2901 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc)
2902 {
2903         struct device *dev = acpi_desc->dev;
2904         struct nd_cmd_ars_status *ars_status;
2905
2906         if (acpi_desc->ars_status) {
2907                 memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
2908                 return 0;
2909         }
2910
2911         ars_status = devm_kzalloc(dev, acpi_desc->max_ars, GFP_KERNEL);
2912         if (!ars_status)
2913                 return -ENOMEM;
2914         acpi_desc->ars_status = ars_status;
2915         return 0;
2916 }
2917
2918 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc)
2919 {
2920         int rc;
2921
2922         if (ars_status_alloc(acpi_desc))
2923                 return -ENOMEM;
2924
2925         rc = ars_get_status(acpi_desc);
2926
2927         if (rc < 0 && rc != -ENOSPC)
2928                 return rc;
2929
2930         if (ars_status_process_records(acpi_desc))
2931                 return -ENOMEM;
2932
2933         return 0;
2934 }
2935
2936 static int ars_register(struct acpi_nfit_desc *acpi_desc,
2937                 struct nfit_spa *nfit_spa)
2938 {
2939         int rc;
2940
2941         if (no_init_ars || test_bit(ARS_FAILED, &nfit_spa->ars_state))
2942                 return acpi_nfit_register_region(acpi_desc, nfit_spa);
2943
2944         set_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
2945         set_bit(ARS_REQ_LONG, &nfit_spa->ars_state);
2946
2947         switch (acpi_nfit_query_poison(acpi_desc)) {
2948         case 0:
2949         case -EAGAIN:
2950                 rc = ars_start(acpi_desc, nfit_spa, ARS_REQ_SHORT);
2951                 /* shouldn't happen, try again later */
2952                 if (rc == -EBUSY)
2953                         break;
2954                 if (rc) {
2955                         set_bit(ARS_FAILED, &nfit_spa->ars_state);
2956                         break;
2957                 }
2958                 clear_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
2959                 rc = acpi_nfit_query_poison(acpi_desc);
2960                 if (rc)
2961                         break;
2962                 acpi_desc->scrub_spa = nfit_spa;
2963                 ars_complete(acpi_desc, nfit_spa);
2964                 /*
2965                  * If ars_complete() says we didn't complete the
2966                  * short scrub, we'll try again with a long
2967                  * request.
2968                  */
2969                 acpi_desc->scrub_spa = NULL;
2970                 break;
2971         case -EBUSY:
2972         case -ENOMEM:
2973         case -ENOSPC:
2974                 /*
2975                  * BIOS was using ARS, wait for it to complete (or
2976                  * resources to become available) and then perform our
2977                  * own scrubs.
2978                  */
2979                 break;
2980         default:
2981                 set_bit(ARS_FAILED, &nfit_spa->ars_state);
2982                 break;
2983         }
2984
2985         return acpi_nfit_register_region(acpi_desc, nfit_spa);
2986 }
2987
2988 static void ars_complete_all(struct acpi_nfit_desc *acpi_desc)
2989 {
2990         struct nfit_spa *nfit_spa;
2991
2992         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2993                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
2994                         continue;
2995                 ars_complete(acpi_desc, nfit_spa);
2996         }
2997 }
2998
2999 static unsigned int __acpi_nfit_scrub(struct acpi_nfit_desc *acpi_desc,
3000                 int query_rc)
3001 {
3002         unsigned int tmo = acpi_desc->scrub_tmo;
3003         struct device *dev = acpi_desc->dev;
3004         struct nfit_spa *nfit_spa;
3005
3006         lockdep_assert_held(&acpi_desc->init_mutex);
3007
3008         if (acpi_desc->cancel)
3009                 return 0;
3010
3011         if (query_rc == -EBUSY) {
3012                 dev_dbg(dev, "ARS: ARS busy\n");
3013                 return min(30U * 60U, tmo * 2);
3014         }
3015         if (query_rc == -ENOSPC) {
3016                 dev_dbg(dev, "ARS: ARS continue\n");
3017                 ars_continue(acpi_desc);
3018                 return 1;
3019         }
3020         if (query_rc && query_rc != -EAGAIN) {
3021                 unsigned long long addr, end;
3022
3023                 addr = acpi_desc->ars_status->address;
3024                 end = addr + acpi_desc->ars_status->length;
3025                 dev_dbg(dev, "ARS: %llx-%llx failed (%d)\n", addr, end,
3026                                 query_rc);
3027         }
3028
3029         ars_complete_all(acpi_desc);
3030         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3031                 enum nfit_ars_state req_type;
3032                 int rc;
3033
3034                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3035                         continue;
3036
3037                 /* prefer short ARS requests first */
3038                 if (test_bit(ARS_REQ_SHORT, &nfit_spa->ars_state))
3039                         req_type = ARS_REQ_SHORT;
3040                 else if (test_bit(ARS_REQ_LONG, &nfit_spa->ars_state))
3041                         req_type = ARS_REQ_LONG;
3042                 else
3043                         continue;
3044                 rc = ars_start(acpi_desc, nfit_spa, req_type);
3045
3046                 dev = nd_region_dev(nfit_spa->nd_region);
3047                 dev_dbg(dev, "ARS: range %d ARS start %s (%d)\n",
3048                                 nfit_spa->spa->range_index,
3049                                 req_type == ARS_REQ_SHORT ? "short" : "long",
3050                                 rc);
3051                 /*
3052                  * Hmm, we raced someone else starting ARS? Try again in
3053                  * a bit.
3054                  */
3055                 if (rc == -EBUSY)
3056                         return 1;
3057                 if (rc == 0) {
3058                         dev_WARN_ONCE(dev, acpi_desc->scrub_spa,
3059                                         "scrub start while range %d active\n",
3060                                         acpi_desc->scrub_spa->spa->range_index);
3061                         clear_bit(req_type, &nfit_spa->ars_state);
3062                         acpi_desc->scrub_spa = nfit_spa;
3063                         /*
3064                          * Consider this spa last for future scrub
3065                          * requests
3066                          */
3067                         list_move_tail(&nfit_spa->list, &acpi_desc->spas);
3068                         return 1;
3069                 }
3070
3071                 dev_err(dev, "ARS: range %d ARS failed (%d)\n",
3072                                 nfit_spa->spa->range_index, rc);
3073                 set_bit(ARS_FAILED, &nfit_spa->ars_state);
3074         }
3075         return 0;
3076 }
3077
3078 static void __sched_ars(struct acpi_nfit_desc *acpi_desc, unsigned int tmo)
3079 {
3080         lockdep_assert_held(&acpi_desc->init_mutex);
3081
3082         acpi_desc->scrub_busy = 1;
3083         /* note this should only be set from within the workqueue */
3084         if (tmo)
3085                 acpi_desc->scrub_tmo = tmo;
3086         queue_delayed_work(nfit_wq, &acpi_desc->dwork, tmo * HZ);
3087 }
3088
3089 static void sched_ars(struct acpi_nfit_desc *acpi_desc)
3090 {
3091         __sched_ars(acpi_desc, 0);
3092 }
3093
3094 static void notify_ars_done(struct acpi_nfit_desc *acpi_desc)
3095 {
3096         lockdep_assert_held(&acpi_desc->init_mutex);
3097
3098         acpi_desc->scrub_busy = 0;
3099         acpi_desc->scrub_count++;
3100         if (acpi_desc->scrub_count_state)
3101                 sysfs_notify_dirent(acpi_desc->scrub_count_state);
3102 }
3103
3104 static void acpi_nfit_scrub(struct work_struct *work)
3105 {
3106         struct acpi_nfit_desc *acpi_desc;
3107         unsigned int tmo;
3108         int query_rc;
3109
3110         acpi_desc = container_of(work, typeof(*acpi_desc), dwork.work);
3111         mutex_lock(&acpi_desc->init_mutex);
3112         query_rc = acpi_nfit_query_poison(acpi_desc);
3113         tmo = __acpi_nfit_scrub(acpi_desc, query_rc);
3114         if (tmo)
3115                 __sched_ars(acpi_desc, tmo);
3116         else
3117                 notify_ars_done(acpi_desc);
3118         memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
3119         mutex_unlock(&acpi_desc->init_mutex);
3120 }
3121
3122 static void acpi_nfit_init_ars(struct acpi_nfit_desc *acpi_desc,
3123                 struct nfit_spa *nfit_spa)
3124 {
3125         int type = nfit_spa_type(nfit_spa->spa);
3126         struct nd_cmd_ars_cap ars_cap;
3127         int rc;
3128
3129         set_bit(ARS_FAILED, &nfit_spa->ars_state);
3130         memset(&ars_cap, 0, sizeof(ars_cap));
3131         rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
3132         if (rc < 0)
3133                 return;
3134         /* check that the supported scrub types match the spa type */
3135         if (type == NFIT_SPA_VOLATILE && ((ars_cap.status >> 16)
3136                                 & ND_ARS_VOLATILE) == 0)
3137                 return;
3138         if (type == NFIT_SPA_PM && ((ars_cap.status >> 16)
3139                                 & ND_ARS_PERSISTENT) == 0)
3140                 return;
3141
3142         nfit_spa->max_ars = ars_cap.max_ars_out;
3143         nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
3144         acpi_desc->max_ars = max(nfit_spa->max_ars, acpi_desc->max_ars);
3145         clear_bit(ARS_FAILED, &nfit_spa->ars_state);
3146 }
3147
3148 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
3149 {
3150         struct nfit_spa *nfit_spa;
3151         int rc;
3152
3153         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3154                 switch (nfit_spa_type(nfit_spa->spa)) {
3155                 case NFIT_SPA_VOLATILE:
3156                 case NFIT_SPA_PM:
3157                         acpi_nfit_init_ars(acpi_desc, nfit_spa);
3158                         break;
3159                 }
3160         }
3161
3162         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
3163                 switch (nfit_spa_type(nfit_spa->spa)) {
3164                 case NFIT_SPA_VOLATILE:
3165                 case NFIT_SPA_PM:
3166                         /* register regions and kick off initial ARS run */
3167                         rc = ars_register(acpi_desc, nfit_spa);
3168                         if (rc)
3169                                 return rc;
3170                         break;
3171                 case NFIT_SPA_BDW:
3172                         /* nothing to register */
3173                         break;
3174                 case NFIT_SPA_DCR:
3175                 case NFIT_SPA_VDISK:
3176                 case NFIT_SPA_VCD:
3177                 case NFIT_SPA_PDISK:
3178                 case NFIT_SPA_PCD:
3179                         /* register known regions that don't support ARS */
3180                         rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
3181                         if (rc)
3182                                 return rc;
3183                         break;
3184                 default:
3185                         /* don't register unknown regions */
3186                         break;
3187                 }
3188
3189         sched_ars(acpi_desc);
3190         return 0;
3191 }
3192
3193 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
3194                 struct nfit_table_prev *prev)
3195 {
3196         struct device *dev = acpi_desc->dev;
3197
3198         if (!list_empty(&prev->spas) ||
3199                         !list_empty(&prev->memdevs) ||
3200                         !list_empty(&prev->dcrs) ||
3201                         !list_empty(&prev->bdws) ||
3202                         !list_empty(&prev->idts) ||
3203                         !list_empty(&prev->flushes)) {
3204                 dev_err(dev, "new nfit deletes entries (unsupported)\n");
3205                 return -ENXIO;
3206         }
3207         return 0;
3208 }
3209
3210 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
3211 {
3212         struct device *dev = acpi_desc->dev;
3213         struct kernfs_node *nfit;
3214         struct device *bus_dev;
3215
3216         if (!ars_supported(acpi_desc->nvdimm_bus))
3217                 return 0;
3218
3219         bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3220         nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
3221         if (!nfit) {
3222                 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
3223                 return -ENODEV;
3224         }
3225         acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
3226         sysfs_put(nfit);
3227         if (!acpi_desc->scrub_count_state) {
3228                 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
3229                 return -ENODEV;
3230         }
3231
3232         return 0;
3233 }
3234
3235 static void acpi_nfit_unregister(void *data)
3236 {
3237         struct acpi_nfit_desc *acpi_desc = data;
3238
3239         nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
3240 }
3241
3242 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
3243 {
3244         struct device *dev = acpi_desc->dev;
3245         struct nfit_table_prev prev;
3246         const void *end;
3247         int rc;
3248
3249         if (!acpi_desc->nvdimm_bus) {
3250                 acpi_nfit_init_dsms(acpi_desc);
3251
3252                 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
3253                                 &acpi_desc->nd_desc);
3254                 if (!acpi_desc->nvdimm_bus)
3255                         return -ENOMEM;
3256
3257                 rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
3258                                 acpi_desc);
3259                 if (rc)
3260                         return rc;
3261
3262                 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
3263                 if (rc)
3264                         return rc;
3265
3266                 /* register this acpi_desc for mce notifications */
3267                 mutex_lock(&acpi_desc_lock);
3268                 list_add_tail(&acpi_desc->list, &acpi_descs);
3269                 mutex_unlock(&acpi_desc_lock);
3270         }
3271
3272         mutex_lock(&acpi_desc->init_mutex);
3273
3274         INIT_LIST_HEAD(&prev.spas);
3275         INIT_LIST_HEAD(&prev.memdevs);
3276         INIT_LIST_HEAD(&prev.dcrs);
3277         INIT_LIST_HEAD(&prev.bdws);
3278         INIT_LIST_HEAD(&prev.idts);
3279         INIT_LIST_HEAD(&prev.flushes);
3280
3281         list_cut_position(&prev.spas, &acpi_desc->spas,
3282                                 acpi_desc->spas.prev);
3283         list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
3284                                 acpi_desc->memdevs.prev);
3285         list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
3286                                 acpi_desc->dcrs.prev);
3287         list_cut_position(&prev.bdws, &acpi_desc->bdws,
3288                                 acpi_desc->bdws.prev);
3289         list_cut_position(&prev.idts, &acpi_desc->idts,
3290                                 acpi_desc->idts.prev);
3291         list_cut_position(&prev.flushes, &acpi_desc->flushes,
3292                                 acpi_desc->flushes.prev);
3293
3294         end = data + sz;
3295         while (!IS_ERR_OR_NULL(data))
3296                 data = add_table(acpi_desc, &prev, data, end);
3297
3298         if (IS_ERR(data)) {
3299                 dev_dbg(dev, "nfit table parsing error: %ld\n", PTR_ERR(data));
3300                 rc = PTR_ERR(data);
3301                 goto out_unlock;
3302         }
3303
3304         rc = acpi_nfit_check_deletions(acpi_desc, &prev);
3305         if (rc)
3306                 goto out_unlock;
3307
3308         rc = nfit_mem_init(acpi_desc);
3309         if (rc)
3310                 goto out_unlock;
3311
3312         rc = acpi_nfit_register_dimms(acpi_desc);
3313         if (rc)
3314                 goto out_unlock;
3315
3316         rc = acpi_nfit_register_regions(acpi_desc);
3317
3318  out_unlock:
3319         mutex_unlock(&acpi_desc->init_mutex);
3320         return rc;
3321 }
3322 EXPORT_SYMBOL_GPL(acpi_nfit_init);
3323
3324 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
3325 {
3326         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
3327         struct device *dev = acpi_desc->dev;
3328
3329         /* Bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
3330         device_lock(dev);
3331         device_unlock(dev);
3332
3333         /* Bounce the init_mutex to complete initial registration */
3334         mutex_lock(&acpi_desc->init_mutex);
3335         mutex_unlock(&acpi_desc->init_mutex);
3336
3337         return 0;
3338 }
3339
3340 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3341                 struct nvdimm *nvdimm, unsigned int cmd)
3342 {
3343         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
3344         struct nfit_spa *nfit_spa;
3345         int rc = 0;
3346
3347         if (nvdimm)
3348                 return 0;
3349         if (cmd != ND_CMD_ARS_START)
3350                 return 0;
3351
3352         /*
3353          * The kernel and userspace may race to initiate a scrub, but
3354          * the scrub thread is prepared to lose that initial race.  It
3355          * just needs guarantees that any ARS it initiates are not
3356          * interrupted by any intervening start requests from userspace.
3357          */
3358         mutex_lock(&acpi_desc->init_mutex);
3359         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
3360                 if (acpi_desc->scrub_spa
3361                                 || test_bit(ARS_REQ_SHORT, &nfit_spa->ars_state)
3362                                 || test_bit(ARS_REQ_LONG, &nfit_spa->ars_state)) {
3363                         rc = -EBUSY;
3364                         break;
3365                 }
3366         mutex_unlock(&acpi_desc->init_mutex);
3367
3368         return rc;
3369 }
3370
3371 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc,
3372                 enum nfit_ars_state req_type)
3373 {
3374         struct device *dev = acpi_desc->dev;
3375         int scheduled = 0, busy = 0;
3376         struct nfit_spa *nfit_spa;
3377
3378         mutex_lock(&acpi_desc->init_mutex);
3379         if (acpi_desc->cancel) {
3380                 mutex_unlock(&acpi_desc->init_mutex);
3381                 return 0;
3382         }
3383
3384         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3385                 int type = nfit_spa_type(nfit_spa->spa);
3386
3387                 if (type != NFIT_SPA_PM && type != NFIT_SPA_VOLATILE)
3388                         continue;
3389                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3390                         continue;
3391
3392                 if (test_and_set_bit(req_type, &nfit_spa->ars_state))
3393                         busy++;
3394                 else
3395                         scheduled++;
3396         }
3397         if (scheduled) {
3398                 sched_ars(acpi_desc);
3399                 dev_dbg(dev, "ars_scan triggered\n");
3400         }
3401         mutex_unlock(&acpi_desc->init_mutex);
3402
3403         if (scheduled)
3404                 return 0;
3405         if (busy)
3406                 return -EBUSY;
3407         return -ENOTTY;
3408 }
3409
3410 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
3411 {
3412         struct nvdimm_bus_descriptor *nd_desc;
3413
3414         dev_set_drvdata(dev, acpi_desc);
3415         acpi_desc->dev = dev;
3416         acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
3417         nd_desc = &acpi_desc->nd_desc;
3418         nd_desc->provider_name = "ACPI.NFIT";
3419         nd_desc->module = THIS_MODULE;
3420         nd_desc->ndctl = acpi_nfit_ctl;
3421         nd_desc->flush_probe = acpi_nfit_flush_probe;
3422         nd_desc->clear_to_send = acpi_nfit_clear_to_send;
3423         nd_desc->attr_groups = acpi_nfit_attribute_groups;
3424
3425         INIT_LIST_HEAD(&acpi_desc->spas);
3426         INIT_LIST_HEAD(&acpi_desc->dcrs);
3427         INIT_LIST_HEAD(&acpi_desc->bdws);
3428         INIT_LIST_HEAD(&acpi_desc->idts);
3429         INIT_LIST_HEAD(&acpi_desc->flushes);
3430         INIT_LIST_HEAD(&acpi_desc->memdevs);
3431         INIT_LIST_HEAD(&acpi_desc->dimms);
3432         INIT_LIST_HEAD(&acpi_desc->list);
3433         mutex_init(&acpi_desc->init_mutex);
3434         acpi_desc->scrub_tmo = 1;
3435         INIT_DELAYED_WORK(&acpi_desc->dwork, acpi_nfit_scrub);
3436 }
3437 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
3438
3439 static void acpi_nfit_put_table(void *table)
3440 {
3441         acpi_put_table(table);
3442 }
3443
3444 void acpi_nfit_shutdown(void *data)
3445 {
3446         struct acpi_nfit_desc *acpi_desc = data;
3447         struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3448
3449         /*
3450          * Destruct under acpi_desc_lock so that nfit_handle_mce does not
3451          * race teardown
3452          */
3453         mutex_lock(&acpi_desc_lock);
3454         list_del(&acpi_desc->list);
3455         mutex_unlock(&acpi_desc_lock);
3456
3457         mutex_lock(&acpi_desc->init_mutex);
3458         acpi_desc->cancel = 1;
3459         cancel_delayed_work_sync(&acpi_desc->dwork);
3460         mutex_unlock(&acpi_desc->init_mutex);
3461
3462         /*
3463          * Bounce the nvdimm bus lock to make sure any in-flight
3464          * acpi_nfit_ars_rescan() submissions have had a chance to
3465          * either submit or see ->cancel set.
3466          */
3467         device_lock(bus_dev);
3468         device_unlock(bus_dev);
3469
3470         flush_workqueue(nfit_wq);
3471 }
3472 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown);
3473
3474 static int acpi_nfit_add(struct acpi_device *adev)
3475 {
3476         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3477         struct acpi_nfit_desc *acpi_desc;
3478         struct device *dev = &adev->dev;
3479         struct acpi_table_header *tbl;
3480         acpi_status status = AE_OK;
3481         acpi_size sz;
3482         int rc = 0;
3483
3484         status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
3485         if (ACPI_FAILURE(status)) {
3486                 /* This is ok, we could have an nvdimm hotplugged later */
3487                 dev_dbg(dev, "failed to find NFIT at startup\n");
3488                 return 0;
3489         }
3490
3491         rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl);
3492         if (rc)
3493                 return rc;
3494         sz = tbl->length;
3495
3496         acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3497         if (!acpi_desc)
3498                 return -ENOMEM;
3499         acpi_nfit_desc_init(acpi_desc, &adev->dev);
3500
3501         /* Save the acpi header for exporting the revision via sysfs */
3502         acpi_desc->acpi_header = *tbl;
3503
3504         /* Evaluate _FIT and override with that if present */
3505         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
3506         if (ACPI_SUCCESS(status) && buf.length > 0) {
3507                 union acpi_object *obj = buf.pointer;
3508
3509                 if (obj->type == ACPI_TYPE_BUFFER)
3510                         rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3511                                         obj->buffer.length);
3512                 else
3513                         dev_dbg(dev, "invalid type %d, ignoring _FIT\n",
3514                                 (int) obj->type);
3515                 kfree(buf.pointer);
3516         } else
3517                 /* skip over the lead-in header table */
3518                 rc = acpi_nfit_init(acpi_desc, (void *) tbl
3519                                 + sizeof(struct acpi_table_nfit),
3520                                 sz - sizeof(struct acpi_table_nfit));
3521
3522         if (rc)
3523                 return rc;
3524         return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc);
3525 }
3526
3527 static int acpi_nfit_remove(struct acpi_device *adev)
3528 {
3529         /* see acpi_nfit_unregister */
3530         return 0;
3531 }
3532
3533 static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle)
3534 {
3535         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3536         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3537         union acpi_object *obj;
3538         acpi_status status;
3539         int ret;
3540
3541         if (!dev->driver) {
3542                 /* dev->driver may be null if we're being removed */
3543                 dev_dbg(dev, "no driver found for dev\n");
3544                 return;
3545         }
3546
3547         if (!acpi_desc) {
3548                 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3549                 if (!acpi_desc)
3550                         return;
3551                 acpi_nfit_desc_init(acpi_desc, dev);
3552         } else {
3553                 /*
3554                  * Finish previous registration before considering new
3555                  * regions.
3556                  */
3557                 flush_workqueue(nfit_wq);
3558         }
3559
3560         /* Evaluate _FIT */
3561         status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
3562         if (ACPI_FAILURE(status)) {
3563                 dev_err(dev, "failed to evaluate _FIT\n");
3564                 return;
3565         }
3566
3567         obj = buf.pointer;
3568         if (obj->type == ACPI_TYPE_BUFFER) {
3569                 ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3570                                 obj->buffer.length);
3571                 if (ret)
3572                         dev_err(dev, "failed to merge updated NFIT\n");
3573         } else
3574                 dev_err(dev, "Invalid _FIT\n");
3575         kfree(buf.pointer);
3576 }
3577
3578 static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle)
3579 {
3580         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3581
3582         if (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON)
3583                 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
3584         else
3585                 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_SHORT);
3586 }
3587
3588 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
3589 {
3590         dev_dbg(dev, "event: 0x%x\n", event);
3591
3592         switch (event) {
3593         case NFIT_NOTIFY_UPDATE:
3594                 return acpi_nfit_update_notify(dev, handle);
3595         case NFIT_NOTIFY_UC_MEMORY_ERROR:
3596                 return acpi_nfit_uc_error_notify(dev, handle);
3597         default:
3598                 return;
3599         }
3600 }
3601 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
3602
3603 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
3604 {
3605         device_lock(&adev->dev);
3606         __acpi_nfit_notify(&adev->dev, adev->handle, event);
3607         device_unlock(&adev->dev);
3608 }
3609
3610 static const struct acpi_device_id acpi_nfit_ids[] = {
3611         { "ACPI0012", 0 },
3612         { "", 0 },
3613 };
3614 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
3615
3616 static struct acpi_driver acpi_nfit_driver = {
3617         .name = KBUILD_MODNAME,
3618         .ids = acpi_nfit_ids,
3619         .ops = {
3620                 .add = acpi_nfit_add,
3621                 .remove = acpi_nfit_remove,
3622                 .notify = acpi_nfit_notify,
3623         },
3624 };
3625
3626 static __init int nfit_init(void)
3627 {
3628         int ret;
3629
3630         BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
3631         BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
3632         BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
3633         BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
3634         BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
3635         BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
3636         BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
3637         BUILD_BUG_ON(sizeof(struct acpi_nfit_capabilities) != 16);
3638
3639         guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]);
3640         guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]);
3641         guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]);
3642         guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]);
3643         guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]);
3644         guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]);
3645         guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]);
3646         guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]);
3647         guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]);
3648         guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]);
3649         guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
3650         guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
3651         guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
3652
3653         nfit_wq = create_singlethread_workqueue("nfit");
3654         if (!nfit_wq)
3655                 return -ENOMEM;
3656
3657         nfit_mce_register();
3658         ret = acpi_bus_register_driver(&acpi_nfit_driver);
3659         if (ret) {
3660                 nfit_mce_unregister();
3661                 destroy_workqueue(nfit_wq);
3662         }
3663
3664         return ret;
3665
3666 }
3667
3668 static __exit void nfit_exit(void)
3669 {
3670         nfit_mce_unregister();
3671         acpi_bus_unregister_driver(&acpi_nfit_driver);
3672         destroy_workqueue(nfit_wq);
3673         WARN_ON(!list_empty(&acpi_descs));
3674 }
3675
3676 module_init(nfit_init);
3677 module_exit(nfit_exit);
3678 MODULE_LICENSE("GPL v2");
3679 MODULE_AUTHOR("Intel Corporation");